id
stringlengths 27
33
| source
stringclasses 1
value | format
stringclasses 1
value | text
stringlengths 13
1.81M
|
|---|---|---|---|
warning/0506/hep-th0506123.html
|
ar5iv
|
text
|
# Contents
## 1 Introduction
The AdS/CFT duality relates gravity in asymptotically AdS spacetimes to a quantum field theory on its conformal boundary. One of the main features of the duality is that the boundary fields parametrizing the boundary conditions of bulk fields are identified with QFT sources that couple to gauge invariant operators. In particular, the boundary metric $`g_{(0)}`$ is considered as a source of the boundary energy momentum tensor and at the same time is the metric of the spacetime on which the dual field theory is defined.
In quantum field theory the sources are unconstrained, so that one can functionally differentiate w.r.t. them to obtain correlation functions. Thus the duality requires the existence of spacetimes associated with general Dirichlet boundary conditions for the metric. Such general boundary conditions go beyond what has been considered in the GR literature where asymptotically AdS spacetimes (AAdS) were defined to have a specific asymptotic conformal structure, namely that of the exact AdS solution , but they have been considered in the mathematics literature . The asymptotic structure of these more general spacetimes is only locally that of AdS; we will call them asymptotically locally AdS (AlAdS) spacetimes.
An integral part of the correspondence is how conserved charges are mapped from one side to the other. This is in fact dictated by the basic AdS/CFT dictionary. On the field theory side, conserved charges are generated by conserved currents. In particular, the energy can be computed from the energy momentum tensor. Applying the AdS/CFT dictionary, we find that one should be able to compute the gravitational energy from the energy momentum tensor obtained by varying the on-shell gravitational action w.r.t. the boundary metric. Such a definition of conserved charges is available in the literature , but the naive implementation of this idea gives infinite answers, essentially due to the infinite volume of spacetime. In the GR literature such infinities are (explicitly or implicitly) dealt with by background subtraction. The AdS/CFT correspondence, however, suggests a new approach: one subtracts the infinities by means of boundary counterterms -, as done in QFT in the process of renormalization. This procedure, called holographic renormalization, is by now a well studied method. We will call the charges defined using the holographic energy momentum tensor “holographic charges”. Notice that these charges are defined intrinsically rather than relative to some other spacetime. This is a definite advance over the background subtraction method, since a suitable reference spacetime does not exist in general.
The holographic charges agree<sup>1</sup><sup>1</sup>1The apparent difference between the holographic mass for odd dimensional AAdS and other definitions such as the one in is now understood to be due to the fact that these other approaches effectively compute masses relative to that of $`AdS_{2k+1}`$, and the holographic mass of $`AdS_{2k+1}`$ is nonzero, see for a detailed discussion. with previous definitions of conserved charges when the latter are applicable, i.e. when the spacetime approaches that of the exact AdS solution and one considers the energy relative to AdS, see for a detailed comparison between different definitions of conserved charges. The new definition on the other hand extends to arbitrary asymptotically locally AdS spacetimes. Moreover it was proved in that these charges arise as Noether charges associated to asymptotic symmetries of such spacetimes and also shown to agree with the charges defined in the covariant phase space approach of Wald et al .
AAdS spacetimes are known to have positive mass relative to the AdS solution, which saturates the bound in a positive mass theorem . The proof in generalizes Witten’s spinorial positive energy theorem for asymptotically flat spacetimes. A natural question to ask is whether the holographic energy of $`AlAdS`$ spacetimes is subjected to a positivity theorem. Such a generalization is far from obvious and is known to be false for asymptotically locally flat spacetimes . A new positivity theorem for a specific class of AlAdS spacetimes has been conjectured in . In this reference an $`AlAdS`$ solution with negative mass (relative to AdS with periodic identification) was found but it was conjectured to be the lowest energy solution among all solutions with the same asymptotics.
Notice that the positivity of the gravitational energy implies via the AdS/CFT correspondence the positivity of the quantum QFT Hamiltonian at strong coupling. This is a very strong conclusion since for a general AlAdS the dual QFT resides on a curved manifold, and in general even the very definition of a QFT on a curved manifold is subtle. Therefore, in general we expect that only a subclass of AlAdS spacetimes is subject to a positivity theorem. One might in fact turn things around and view our discussions as giving a criterion for the selection of good boundary conditions.
This paper is organized as follows. In the next section we discuss the definition of asymptotically locally AdS spacetimes and in section 3 the definition of energy for such spacetimes. The spinorial energy of Witten and Nester is reviewed in section 4. In section 5, we construct asymptotic solutions of the Witten equation and use them in section 6 to compute a regulated version of the Witten-Nester energy for AlAdS spacetimes. This leads to a number of necessary conditions for the existence of such an energy. In section 7 we compare the finite part of the Witten-Nester energy with the holographic energy. In section 8 we specialize to AAdS spacetimes and in section 9 we illustrate subtleties related to some global issues by discussing two examples, the extremal BTZ black hole and the AdS soliton. We conclude with a discussion of our results in section 10. In order to keep the line of argument clear, we have moved most of the technical details to a series of appendices.
## 2 Asymptotically locally AdS spacetimes
We discuss in this section the definition of asymptotically locally anti-de Sitter (AlAdS) spacetimes. More details can be found in and the mathematics reviews . In this paper, we restrict our attention to the case of pure gravity but the method can be generalized to include matter.
The most general asymptotic solution of Einstein’s equations with negative cosmological constant takes the form
$`ds^2=G_{\mu \nu }dx^\mu dx^\nu ={\displaystyle \frac{dz^2}{z^2}}+{\displaystyle \frac{1}{z^2}}g_{ij}(x,z)dx^idx^j,`$
$`g(x,z)=g_{(0)}+zg_{(1)}\mathrm{}+z^dg_{(d)}+h_{(d)}z^d\mathrm{log}z^2+\mathrm{}`$ (2.1)
In these coordinates $`z=0`$ is the location of the conformal boundary of spacetime and $`g_{(0)}`$ is an arbitrary non-degenerate metric (which represents the conformal structure of the boundary). Einstein equations determine uniquely all coefficients in (2) except for the transverse traceless part of $`g_{(d)}`$ (see appendix A of for concrete expressions). A short computation reveals that the Riemann tensor of the metric (2) is asymptotically equal to
$$R_{\mu \nu \kappa \lambda }=(G_{\mu \lambda }G_{\nu \kappa }G_{\kappa \mu }G_{\nu \lambda })(1+𝒪(z))$$
(2.2)
where the cosmological constant is normalized as $`\mathrm{\Lambda }=d(d1)/2`$ (i.e. we set the AdS radius equal to one). Thus the leading form of the Riemann tensor is exactly the same as the Riemann tensor of the $`AdS_{d+1}`$ spacetime. We will call solutions with this property “asymptotically locally AdS” (AlAdS) spacetimes. All solutions of pure gravity with negative cosmological constant are of this form. Notice that we do not require the conformal structure of (2) to be that of $`AdS_{d+1}`$. Spacetimes with this conformal structure are called “asymptotically AdS” .
Recall that $`AdS_{d+1}`$ is conformally flat and this implies that $`g_{(0)}`$ is also conformally flat and the expansion (2) terminates at order $`z^4`$,
$$g(x,z)=\left(1+\frac{z^2}{2}g_{(2)}g_{(0)}^1\right)g_{(0)}\left(1+\frac{z^2}{2}g_{(0)}^1g_{(2)}\right)$$
(2.3)
with
$`d=2:g_{(2)ij}`$ $`=`$ $`{\displaystyle \frac{1}{2}}(Rg_{(0)ij}+t_{ij}),^it_{ij}=0,t_i^i=R,`$ (2.4)
$`d2:g_{(2)ij}`$ $`=`$ $`{\displaystyle \frac{1}{d2}}(R_{ij}{\displaystyle \frac{1}{2(d1)}}Rg_{(0)ij}),`$
where $`R_{ij}`$ is the Ricci tensor of $`g_{(0)}`$ and the transverse traceless part of $`t_{ij}`$ is not determined by the asymptotic analysis. $`g_{(0)}`$ may be chosen to be the standard metric $`𝐑\times S^{d1}`$. By definition, $`AAdS_{d+1}`$ spacetimes have the same boundary conformal structure as $`AdS_{d+1}`$. This implies that all coefficients up to $`g_{(d)}`$ are the same as those for $`AdS_{d+1}`$, but $`g_{(d)}`$ is different. For $`AAdS_{d+1}`$ spacetimes the logarithmic term in (2) is absent.
AlAdS spacetimes have an arbitrary conformal structure $`[g_{(0)}]`$ and a general $`g_{(d)}`$, the logarithmic term is in general non-vanishing, and there is no a priori restriction on the topology of the conformal boundary. The mathematical structure of these spacetimes (or their Euclidean counterparts) is currently under investigation in the mathematics community, see and references therein. For instance, it has not yet been established how many, if any, global solutions exist given a conformal structure, although given (sufficiently regular) $`g_{(0)}`$ and $`g_{(d)}`$ a unique solution exists in a thickening $`B\times [0,ϵ)`$ of the boundary $`B`$. On the other hand, interesting examples of such spacetimes have appeared in the literature, see for a collection of examples. One of the motivations for the current work is to derive physically motivated conditions on the possible conformal structures $`[g_{(0)}]`$.
A very useful reformulation of the asymptotic analysis can be achieved by observing that for AlAdS spacetimes the radial derivative is to leading order equal to the dilatation operator . That is to say, if we write the metric in the form
$$ds^2=dr^2+\gamma _{ij}(x,r)dx^idx^j$$
(2.5)
which is related to (2) by the coordinate transformation $`z=\mathrm{exp}(r)`$, then
$$_r=\delta _D+𝒪(e^r)$$
(2.6)
where $`\delta _D`$ is the dilatation operator. For pure gravity
$$\delta _D=d^dx2\gamma _{ij}\frac{\delta }{\delta \gamma _{ij}}.$$
(2.7)
When matter fields are present $`\delta _D`$ contains additional terms according to the Weyl transformation of the corresponding boundary fields (see for examples). The asymptotic analysis can now be very effectively performed by expanding all objects in eigenfunctions of the dilatation operator and organizing the terms in the field equations according to their dilatation weight.
For the case of pure gravity, the main object is the extrinsic curvature $`K_{ij}`$ of constant-$`r`$ slices. In the coordinates where the metric is given by (2.5), the extrinsic curvature is equal to
$$K_{ij}=\frac{1}{2}\dot{\gamma }_{ij},$$
(2.8)
where the dot indicates derivative w.r.t. $`r`$. It admits the following expansion in terms of eigenfunctions of the dilatation operator,
$$K_j^i[\gamma ]=\delta _j^i+K_{(2)}{}_{j}{}^{i}+K_{(4)}{}_{j}{}^{i}+\mathrm{}+K_{(d)}{}_{j}{}^{i}+\stackrel{~}{K}_{(d)}{}_{j}{}^{i}(2r)+\mathrm{}$$
(2.9)
where all terms but $`K_{(d)}_j^i`$ transform homogeneously with the weight indicated by their subscript,
$$\delta _DK_{(n)}{}_{j}{}^{i}=nK_{(n)}{}_{j}{}^{i},n<d,\delta _D\stackrel{~}{K}_{(d)}{}_{j}{}^{i}=d\stackrel{~}{K}_{(d)}_j^i$$
(2.10)
and $`K_{(d)}_j^i`$ transforms anomalously,
$$\delta _DK_{(d)}{}_{j}{}^{i}=dK_{(d)}{}_{j}{}^{i}2\stackrel{~}{K}_{(d)}{}_{j}{}^{i}.$$
(2.11)
Notice that $`\stackrel{~}{K}_{(2k+1)}{}_{j}{}^{i}=0`$. The radial derivative admits a similar expansion:
$`_r`$ $`=`$ $`\delta _D+_{(2)}+\mathrm{}`$ (2.12)
$`=`$ $`{\displaystyle d^dx\dot{\gamma }_{ij}\frac{\delta }{\delta \gamma _{ij}}}=2{\displaystyle d^dxK_{ij}\frac{\delta }{\delta \gamma _{ij}}}=\delta _D+{\displaystyle d^dxK_{(2)ij}\frac{\delta }{\delta \gamma _{ij}}}+\mathrm{}`$
where we used the chain rule and (2.8)-(2.9).
Inserting these expansions in Einstein’s equations and grouping terms with the same weight together leads to a number of recursion relations that can be solved to uniquely determine all coefficients except for the traceless divergenceless part of $`K_{(d)}_j^i`$ .
The coefficients $`K_{(n)ij},\stackrel{~}{K}_{(d)ij}`$ determine the coefficients $`g_{(n)},h_{(d)ij}`$ in (2) and vice versa. The precise relations have been worked out in and we list them here for up to $`n=4`$,
$`\gamma _{ij}[g_{(0)}]`$ $`=`$ $`{\displaystyle \frac{1}{z^2}}(g_{(0)ij}+g_{(2)ij}z^2+\mathrm{}+z^dg_{(d)ij}+h_{(d)ij}z^d\mathrm{log}z^2+\mathrm{})`$
$`K_{(2)ij}[g_{(0)}]`$ $`=`$ $`g_{(2)ij}`$ (2.13)
$`K_{(3)ij}[g_{(0)}]`$ $`=`$ $`{\displaystyle \frac{3}{2}}g_{(3)ij},\mathrm{for}d=3`$
$`K_{(4)ij}[g_{(0)}]`$ $`=`$ $`2g_{(4)ij}3h_{(4)ij}+(g_{(2)}^2)_{ij}{\displaystyle \frac{1}{12}}g_{(0)ij}(Trg_{(2)}^2(Trg_{(2)})^2)`$
$`{\displaystyle \frac{1}{2}}g_{(2)ij}Trg_{(2)},\mathrm{for}d=4`$
$`\stackrel{~}{K}_{(d)ij}[g_{(0)}]`$ $`=`$ $`{\displaystyle \frac{d}{2}}h_{(d)ij}.`$
Explicit expressions for $`g_{(n)},h_{(n)}`$ (for low enough $`d`$) can be found in appendix A of and expressions for $`K_{ij}[\gamma ]`$ in .
Since the dilatation operator is equal to the radial derivative to leading order, the leading radial dependence of a dilatation eigenfunction $`f_{(k)}`$ of weight $`k`$ is equal to $`\mathrm{exp}(kr)`$. It will be useful to introduce the following “hat” notation for the leading coefficient:
$$f_{(k)}=e^{kr}\widehat{f}_{(k)}(x)(1+𝒪(e^r))$$
(2.14)
For instance, $`\widehat{\gamma }_{(2)ij}(x)`$ denotes the boundary metric $`g_{(0)ij}(x)`$ and $`\widehat{K}_{(n)ij}=K_{(n)ij}[g_{(0)}]`$.
## 3 Energy of Asymptotically locally AdS spacetimes
In gravitational theories energy is usually measured with respect to a reference spacetime, but such a reference spacetime may not exist for general AlAdS spacetimes. In AlAdS spacetimes that possess an asymptotic timelike Killing vector, however, one can do better: one can assign a mass in a way that is intrinsic to the spacetime, as we review in this section.
We first note that all AlAdS spacetimes possess a covariantly conserved energy momentum tensor constructed from the metric coefficients (in general there are contributions from matter , but we only discuss the pure gravity case in this paper),
$$T_{ij}=\frac{1}{\kappa ^2}(K_{(d)ij}K_{(d)}\gamma _{ij})$$
(3.15)
where $`K_{(d)}=K_{(d)}_i^i`$ and $`\kappa ^2=8\pi G`$. This energy momentum tensor is equal to the variation of the gravitational on-shell action supplemented by appropriate boundary counterterms w.r.t. the boundary metric . One can also derive (3.15) as a Noether current associated with asymptotic (global) symmetries of the bulk spacetime . When the bulk equations of motion hold, it satisfies,
$$^iT_{ij}=0,T_i^i=A,$$
(3.16)
where $`A`$ is the holographic anomaly ($`A`$ is non-vanishing only for even $`d`$ for the pure gravity case but when matter is present there may be additional conformal anomalies for all $`d`$ ).
Let us consider an AlAdS spacetime that possesses a vector that asymptotically approaches a conformal Killing vectors $`\xi ^i`$ of the boundary metric $`g_{(0)}`$ (see appendix B of for the precise fall off conditions). Conserved charges are now obtained as,
$$Q_h=_{C_tM}𝑑S_iT_j^i\xi ^j$$
(3.17)
where $`C_t`$ is an initial value hypersurface of the bulk manifold. If the anomaly vanishes one can construct conserved charges for all conformal Killing vectors of the boundary metric. In particular, the energy is associated with a timelike Killing vector.
One can compute the value of the energy with following steps (see also section 6 of ):
1. Bring the bulk metric to the form (2) by changing coordinates near $`z=0`$, and read off the coefficients $`g_{(n)}`$. From these coefficients, compute the $`K_{(d)ij}`$ coefficients using (2).
2. Compute the stress energy tensor $`T_{ij}`$ by substituting $`K_{(d)ij}`$ in (3.15).
3. Plug in $`T_{ij}`$ and the timelike Killing vector $`\xi ^i`$ of $`g_{(0)}`$ in (3.17).
Let us illustrate this procedure by computing the mass of $`AdS_5`$. We already reported the result for step 1 in (2.3). Substituting in (3.15) we obtain the stress energy tensor
$$T_{ij}=\frac{1}{64\pi G_N}(4\delta _{i,0}\delta _{j,0}+g_{(0)ij})$$
(3.18)
The boundary metric is in this case the standard metric on $`R\times S^3`$, so the timelike Killing vector is $`\xi =/t`$. Substituting in (3.17) we get
$$M_{AdS_5}=d^3x\sqrt{g}T_{00}=\frac{3\pi }{32G_N}.$$
(3.19)
In previous approaches one could only measure the energy of spacetimes relative to $`AdS_5`$. Here we see that we can compute the mass for $`AdS_5`$ itself. The fact that the mass is non-zero is due to the presence of the conformal anomaly (which is related to IR divergences of the on-shell action). Its value is exactly equal to the Casimir energy of $`N=4`$ SYM on $`S^3`$ .
The purpose of this work is to analyze under which conditions the energy defined holographically is bounded from below. To answer this questions we will connect the holographic energy to the spinorial energy of Witten and Nester that is manifestly positive definite.
## 4 Positivity of energy
Witten’s positive energy theorem is motivated by the fact that in supersymmetric theories the Hamiltonian is the square of supercharges. This implies that there is an expression for the energy in terms of spinors and that the energy is positive definite. The construction below imitates the supersymmetric argument but does not require supersymmetry.
Given an antisymmetric tensor $`E^{\mu \nu }`$, one can always obtain an identically covariantly conserved current (i.e. the conservation does not require use of field equations)
$$j^\mu =𝒟_\nu E^{\nu \mu }𝒟_\mu j^\mu =2R_{\mu \nu }E^{\mu \nu }=0,$$
(4.20)
where $`𝒟_\mu `$ is the covariant derivative associated with the bulk metric $`G`$. Integrating the time component of this current over a spacelike hypersurface $`C_t`$, we obtain a conserved charge
$$Q=_{C_t}𝑑\mathrm{\Sigma }_\mu j^\mu =_{C_t}d^dx\sqrt{{}_{}{}^{t}G}n_\mu 𝒟_\nu E^{\nu \mu },$$
(4.21)
where $`{}_{}{}^{t}G`$ is the induced metric on the hypersurface $`C_t`$ and $`n_\mu `$ is the unit normal of $`C_t`$. Using Stokes’ theorem<sup>2</sup><sup>2</sup>2Notice that $`n_\mu 𝒟_\nu E^{\nu \mu }={}_{}{}^{t}𝒟_{\nu }^{}(n_\mu E^{\nu \mu })`$, where $`{}_{}{}^{t}𝒟`$ is the covariant derivative on $`C_t`$. and assuming that the spacetime has a single boundary, we obtain a formula for the charges as an integral at infinity
$$Q=_{C_tM}𝑑\mathrm{\Sigma }_{\mu \nu }E^{\nu \mu }=_{C_tM}d^{d1}x\sqrt{{}_{}{}^{t}g_{(0)}^{}}n_\mu l_\nu E^{\nu \mu },$$
(4.22)
where $`{}_{}{}^{t}g_{(0)}^{}`$ is the induced metric on $`C_tM`$ and $`l^\mu `$ is the outward pointing unit normal of the boundary $`M`$.
The Witten-Nester spinorial energy $`E_{WN}`$ is derived using the following antisymmetric tensor constructed from a spinor fields $`ϵ`$,
$$E^{\mu \nu }=\frac{1}{\kappa ^2}(\overline{ϵ}\mathrm{\Gamma }^{\mu \nu \rho }_\rho ϵ+c.c.)$$
(4.23)
where
$$_\mu =𝒟_\mu +\frac{1}{2}\mathrm{\Gamma }_\mu ,$$
(4.24)
is the AdS covariant derivative (as noted before, we set the AdS scale $`l=1`$ throughout this paper). A standard computation (see, for instance, for details) that uses the bulk equations of motion<sup>3</sup><sup>3</sup>3As mentioned earlier, we consider the case of pure gravity in this paper. The positivity of the spinorial energy continues to hold for gravity coupled to matter with a stress energy tensor that satisfies the dominant energy condition.
$$R_{\mu \nu }\frac{1}{2}(R2\mathrm{\Lambda })G_{\mu \nu }=0,$$
(4.25)
yields
$$n_\mu 𝒟_\nu E^{\nu \mu }=2\left((_{\widehat{\alpha }}ϵ)^{}\eta ^{\widehat{\alpha }\widehat{\beta }}(_{\widehat{\beta }}ϵ)(\mathrm{\Gamma }^{\widehat{\alpha }}_{\widehat{\alpha }}ϵ)^{}(\mathrm{\Gamma }^{\widehat{\beta }}_{\widehat{\beta }}ϵ)\right),$$
(4.26)
where the indices $`\alpha ,\beta `$ run through all values except time and the hat indicates a flat index, e.g. $`_{\widehat{\alpha }}=E_{\widehat{\alpha }}^\mu _\mu `$ with $`E_{\widehat{\alpha }}^\mu `$ being the inverse vielbein, see appendix A for our conventions. It follows that if there exists a regular spinor $`ϵ`$ on $`C_t`$ satisfying the Witten equation,
$$\mathrm{\Gamma }^{\widehat{\alpha }}_{\widehat{\alpha }}ϵ=0,$$
(4.27)
the Witten-Nester energy is positive definite,
$$E_{WN}0.$$
(4.28)
Furthermore, the equality holds iff the Witten spinor is covariantly constant w.r.t. to the AdS connection,
$$E_{WN}=0_{\widehat{\alpha }}ϵ=0.$$
(4.29)
On the other hand, the value of $`E_{WN}`$ depends only on the asymptotics of the Witten spinor as follows from (4.22). We would like to compute this energy for general AlAdS spacetime. To regulate potential IR divergences we introduce a regulating surface $`\mathrm{\Sigma }_r`$. The regulated energy is now given by
$$E_{WN}[r]=_{C_t\mathrm{\Sigma }_r}d^{d1}x\sqrt{{}_{}{}^{t}\gamma }n_\mu l_\nu E^{\nu \mu }=\frac{1}{\kappa ^2}_{C_t\mathrm{\Sigma }_r}d^{d1}x\sqrt{{}_{}{}^{t}\gamma }(ϵ^{}\mathrm{\Gamma }^{\widehat{r}}\mathrm{\Gamma }^{\widehat{a}}_{\widehat{a}}ϵ+c.c.)$$
(4.30)
where we used that in our case $`n_\mu =E_\mu ^{\widehat{t}}`$ and $`l_\mu =E_\mu ^{\widehat{r}}`$, see appendix B. To compute this expression we need to know asymptotic solutions of the Witten equation.
## 5 Asymptotic solutions of the Witten equation
We would like to obtain asymptotic solutions of the Witten equation,
$$/ϵ(\mathrm{\Gamma }^{\widehat{r}}_{\widehat{r}}+\mathrm{\Gamma }^{\widehat{a}}_{\widehat{a}})ϵ=0.$$
(5.31)
This is obtained by expanding all quantities in terms of dilatation eigenfunctions, as in the asymptotic analysis of the bulk equations of motion reviewed in section 2. We present the details in appendix C. In particular, we find the Witten operator $`/`$ admits the following expansion,
$$/=/_{(0)}+/_{(1)}+\underset{k=1}{\overset{[\frac{d1}{2}]}{}}/_{(2k)}+/_{(d)}+(2r)\stackrel{~}{/}_{(d)}+\mathrm{},$$
(5.32)
where the explicit expressions can be found in appendix C ($`[k]`$ denotes the integer part of $`k`$). We only quote here the first two terms
$$/_{(0)}=(\delta _D+\frac{d1}{2})\mathrm{\Gamma }^{\widehat{r}}+\frac{d}{2},/_{(1)}=\mathrm{\Gamma }^{\widehat{a}}D_{\widehat{a}}$$
(5.33)
and note that $`\stackrel{~}{/}_{(d)}`$ is zero when $`d`$ is odd.
Let us now consider a spinor with the asymptotic expansion
$$ϵ=ϵ_{(m)}+ϵ_{(m+1)}+ϵ_{(m+2)}+\mathrm{}+ϵ_{(m+d)}+(2r)\stackrel{~}{ϵ}_{(m+d)}+\mathrm{},$$
(5.34)
where the coefficients transform as their subscript indicates,
$$\delta _Dϵ_{(n)}=nϵ_{(n)},\delta _D\stackrel{~}{ϵ}_{(m+d)}=(m+d)\stackrel{~}{ϵ}_{(m+d)}$$
(5.35)
except for $`ϵ_{(m+d)}`$ which transforms anomalously,
$$\delta _Dϵ_{(m+d)}=(m+d)ϵ_{(m+d)}2\stackrel{~}{ϵ}_{(m+d)}.$$
(5.36)
Inserting (5.34) in the Witten equation and collecting terms of the same weight, we get a series of equations. The equation for the lowest order component reads
$$/_{(0)}ϵ_{(m)}=0.$$
(5.37)
This implies that either
$$m=\frac{1}{2},ϵ_{(\frac{1}{2})}=P^{}ϵ_{(\frac{1}{2})}$$
(5.38)
or
$$m=d\frac{1}{2},ϵ_{(d\frac{1}{2})}=P^+ϵ_{(d\frac{1}{2})},$$
(5.39)
where $`P^\pm =\frac{1}{2}(1\pm \mathrm{\Gamma }^{\widehat{r}})`$ are projection operators. The Witten spinors with leading behavior as in (5.39) fall off too fast at infinity to contribute to $`E_{WN}`$, and therefore we consider only the solution with leading behavior as in (5.38) from now on. Notice, however, that a Witten spinor which is regular in the interior may require a linear combination of the two asymptotic solutions.
The remaining equations read
$`/_{(0)}ϵ_{(\frac{1}{2}+k)}`$ $`=`$ $`(/_{(1)}ϵ_{(k\frac{3}{2})}+{\displaystyle \underset{l=1}{\overset{[\frac{k}{2}]}{}}}/_{(2l)}ϵ_{(\frac{1}{2}+k2l)}),k=1,2,\mathrm{},d1`$ (5.40)
$`/_{(0)}\stackrel{~}{ϵ}_{(\frac{1}{2}+d)}`$ $`=`$ $`\stackrel{~}{/}_{(d)}ϵ_{(\frac{1}{2})},`$ (5.41)
$`/_{(0)}ϵ_{(\frac{1}{2}+d)}`$ $`=`$ $`(/_{(1)}ϵ_{(d\frac{3}{2})}+{\displaystyle \underset{l=1}{\overset{[\frac{d1}{2}]}{}}}/_{(2l)}ϵ_{(\frac{1}{2}+d2l)}+/_{(d)}ϵ_{(\frac{1}{2})}).`$ (5.42)
Using the commutation relations between $`/_{(n)}`$ and $`P^\pm `$ listed in (C.16) we conclude
$$\mathrm{\Gamma }^{\widehat{r}}ϵ_{(\frac{1}{2}+n)}=(1)^{n+1}ϵ_{(\frac{1}{2}+n)},\mathrm{\hspace{0.33em}0}n<d.$$
(5.43)
Equations (5.40) can be solved iteratively to determine locally all coefficients in terms of $`ϵ_{(\frac{1}{2})}`$ provided $`/_{(0)}`$ is invertible. The zero modes of $`/_{(0)}`$ are given in (5.38) and (5.39), so starting from $`ϵ_{(\frac{1}{2})}`$ one can determine all coefficients except for $`P^+ϵ_{(\frac{1}{2}+d)}`$ which is left undetermined. The result is<sup>4</sup><sup>4</sup>4 Use $`\left((\delta _D+\frac{d1}{2})\mathrm{\Gamma }^{\widehat{r}}\frac{d}{2}\right)/_{(0)}ϵ_{(\frac{1}{2}+k)}=k(dk)ϵ_{(\frac{1}{2}+k)}`$.
$`ϵ_{(\frac{1}{2}+k)}=c(k)(/_{(1)}ϵ_{(k\frac{3}{2})}+{\displaystyle \underset{l=1}{\overset{[\frac{k}{2}]}{}}}/_{(2l)}ϵ_{(\frac{1}{2}+k2l)}),`$ (5.44)
$`\mathrm{with}c(2l)={\displaystyle \frac{1}{2l}},c(2l+1)={\displaystyle \frac{1}{d(2l+1)}},k=1,2,\mathrm{},d1.`$
Later on we will need the explicit form for $`k=1`$:
$$ϵ_{(\frac{1}{2})}=\frac{1}{d1}\mathrm{\Gamma }^{\widehat{a}}D_{\widehat{a}}ϵ_{(\frac{1}{2})},$$
(5.45)
The solution of (5.41) and (5.42) depends on whether $`d`$ is even or odd:
$`d`$ even
$`P^{}\stackrel{~}{ϵ}_{(\frac{1}{2}+d)}`$ $`=`$ $`{\displaystyle \frac{1}{d}}\stackrel{~}{/}_{(d)}ϵ_{(\frac{1}{2})}`$
$`P^+\stackrel{~}{ϵ}_{(\frac{1}{2}+d)}`$ $`=`$ $`0`$ (5.46)
$`P^{}ϵ_{(\frac{1}{2}+d)}`$ $`=`$ $`{\displaystyle \frac{1}{d}}\left(\mathrm{\Gamma }^{\widehat{a}}D_{\widehat{a}}ϵ_{(d\frac{3}{2})}+2\stackrel{~}{ϵ}_{(\frac{1}{2}+d)}+{\displaystyle \underset{k=1}{\overset{d/2}{}}}/_{(2k)}ϵ_{(\frac{1}{2}+d2k)}\right)`$
$`P^+ϵ_{(\frac{1}{2}+d)}`$ $`\mathrm{undetermined}`$
$`d`$ odd
$`P^{}\stackrel{~}{ϵ}_{(\frac{1}{2}+d)}`$ $`=`$ $`0`$
$`P^+\stackrel{~}{ϵ}_{(\frac{1}{2}+d)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}(/_{(1)}ϵ_{(d\frac{3}{2})}+{\displaystyle \underset{k=1}{\overset{(d1)/2}{}}}/_{(2k)}ϵ_{(\frac{1}{2}+d2k)})`$
$`P^{}ϵ_{(\frac{1}{2}+d)}`$ $`=`$ $`{\displaystyle \frac{1}{d}}/_{(d)}ϵ_{(\frac{1}{2})}`$ (5.47)
$`P^+ϵ_{(\frac{1}{2}+d)}`$ $`\mathrm{undetermined}.`$
Having determined the most general asymptotic solution of the Witten equation, we next turn to the computation of the Witten-Nester energy.
## 6 Witten-Nester energy
We are now in the position to compute the Witten-Nester energy. Recall that the regulated expression is given by
$$E_{WN}[r]=\frac{1}{\kappa ^2}_{C_t\mathrm{\Sigma }_r}d^{d1}x\sqrt{{}_{}{}^{t}\gamma }(ϵ^{}\mathrm{\Gamma }^{\widehat{r}}\mathrm{\Gamma }^{\widehat{a}}_{\widehat{a}}ϵ+c.c.)$$
(6.48)
where $`r`$, the position of the radial slice, is the regulator. Using the asymptotic expansion derived in the previous section we obtain
$`q`$ $``$ $`ϵ^{}\mathrm{\Gamma }^{\widehat{r}}\mathrm{\Gamma }^{\widehat{a}}_{\widehat{a}}ϵ`$ (6.49)
$`=`$ $`q_{(1)}+q_{(0)}+q_{(1)}+\mathrm{}+q_{(d1)}+(2r)\stackrel{~}{q}_{(d1)}+\mathrm{}.`$
All terms up to $`q_{(d1)}`$ give divergent contributions in (6.48) as $`r\mathrm{}`$. Therefore, for $`E_{WN}`$ to be well-defined, these terms should vanish. Similar divergences were found in the on-shell action in and there they were canceled by means of boundary counterterms. In the present context, however, we want to maintain the manifest positivity of $`E_{WN}`$ so instead of adding counterterms we view the vanishing of the divergent terms as conditions imposed on the asymptotic data. In other words, our results show that only for a subset of AlAdS spacetimes the Witten-Nester energy is well-defined. We should add here that our discussions do not exclude the possibility that a modified Witten-Nester energy exists that is manifestly positive and is well defined for a wider class of AlAdS spacetimes.
The explicit form of $`q_{(n)}`$ is most easily obtained by using (C.17). Using the alternating chirality of the spinors $`ϵ_{(k)}`$ (5.43) we conclude
$$q_{(1)}=q_{(2l)}=0\mathrm{for}l=0,1,\mathrm{}2ld1,$$
(6.50)
The odd powers however are generically non-zero,
$`q_{(2n1)}`$ $`=`$ $`(d1){\displaystyle \underset{k=0}{\overset{n1}{}}}ϵ_{(2n2k\frac{3}{2})}^{}ϵ_{(\frac{1}{2}+2k)}+{\displaystyle \underset{k=0}{\overset{2n1}{}}}(1)^{k+1}ϵ_{(2nk\frac{3}{2})}^{}\mathrm{\Gamma }^{\widehat{a}}D_{\widehat{a}}ϵ_{(\frac{1}{2}+k)}`$ (6.51)
$`{\displaystyle \frac{1}{2}}{\displaystyle \underset{k=1}{\overset{n}{}}}{\displaystyle \underset{l=0}{\overset{2n2k}{}}}K_{(2k)}{}_{\widehat{a}}{}^{\widehat{j}}ϵ_{(2n2kl\frac{1}{2})}^{}\mathrm{\Gamma }^{\widehat{a}}\mathrm{\Gamma }_{\widehat{j}}ϵ_{(l\frac{1}{2})}`$
for $`n=1,2,\mathrm{},[\frac{d1}{2}]`$.
The result for the terms of order $`(d1)`$ depends on whether $`d`$ is even or odd,
$`d`$ odd
$`q_{(d1)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}K_{(d)}{}_{\widehat{a}}{}^{\widehat{j}}ϵ_{(\frac{1}{2})}^{}\mathrm{\Gamma }^{\widehat{a}}\mathrm{\Gamma }_{\widehat{j}}ϵ_{(\frac{1}{2})}`$ (6.52)
$`\stackrel{~}{q}_{(d1)}`$ $`=`$ $`0,`$
$`d`$ even
$`q_{(d1)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}K_{(d)}{}_{\widehat{a}}{}^{\widehat{j}}ϵ_{(\frac{1}{2})}^{}\mathrm{\Gamma }^{\widehat{a}}\mathrm{\Gamma }_{\widehat{j}}ϵ_{(\frac{1}{2})}+{\displaystyle \frac{1}{2}}A_{(d1)}`$ (6.53)
$`\stackrel{~}{q}_{(d1)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\stackrel{~}{K}_{\widehat{a}(d)}^{\widehat{j}}ϵ_{(\frac{1}{2})}^{}\mathrm{\Gamma }^{\widehat{a}}\mathrm{\Gamma }_{\widehat{j}}ϵ_{(\frac{1}{2})}`$
where we separated out in $`q_{(d1)}`$ the term that depends on the coefficient $`K_{(d)ij}`$ which is not determined by the asymptotic analysis. The remaining terms are given by
$`{\displaystyle \frac{1}{2}}A_{(d1)}`$ $`=`$ $`(d1){\displaystyle \underset{k=0}{\overset{d/21}{}}}ϵ_{(d2k\frac{3}{2})}^{}ϵ_{(\frac{1}{2}+2k)}+{\displaystyle \underset{k=0}{\overset{d1}{}}}(1)^{k+1}ϵ_{(dk\frac{3}{2})}^{}\mathrm{\Gamma }^{\widehat{a}}D_{\widehat{a}}ϵ_{(\frac{1}{2}+k)}`$ (6.54)
$`{\displaystyle \frac{1}{2}}{\displaystyle \underset{k=1}{\overset{d/21}{}}}{\displaystyle \underset{l=0}{\overset{d2k}{}}}K_{(2k)}{}_{\widehat{a}}{}^{\widehat{j}}ϵ_{(d2kl\frac{1}{2})}^{}\mathrm{\Gamma }^{\widehat{a}}\mathrm{\Gamma }_{\widehat{j}}ϵ_{(l\frac{1}{2})}`$
To summarize, we have the following result
$`d\mathrm{odd}:`$ $`q={\displaystyle \underset{l=1}{\overset{\frac{d1}{2}}{}}}q_{(2l1)}+q_{(d1)}+\mathrm{}`$ (6.55)
$`d\mathrm{even}:`$ $`q={\displaystyle \underset{l=1}{\overset{\frac{d}{2}1}{}}}q_{(2l1)}+(2r)\stackrel{~}{q}_{(d1)}+q_{(d1)}+\mathrm{}`$ (6.56)
where the various coefficients are given in (6.51), (6.52) and (6.53).
In order for the Witten energy to be well defined we need the integral of the divergent coefficients be zero. Recall that $`\stackrel{~}{K}_{(d)ij}`$ is the metric variation of the conformal anomaly and vanishes when the boundary metric is conformally Einstein , i.e. when there exists a representative of the boundary conformal structure $`g_{(0)}`$ that satisfies Einstein’s equations (with or without cosmological constant). So we conclude that a sufficient condition for the vanishing of the “logarithmic” divergence<sup>5</sup><sup>5</sup>5Recall that $`2r=\mathrm{log}z^2`$ and $`\mathrm{exp}(kr)=z^k`$ so $`(2r)\stackrel{~}{q}_{d1}`$ and $`q_{(2n1)}`$ are analogous to the logarithmic and power-law divergences in the on-shell action. (which is present only in even dimensions) is that the boundary metric is conformally Einstein.
The “power-law” divergences $`q_{(2n1)}`$ impose further conditions on the asymptotic data, namely the boundary geometry should be such that spinors $`ϵ_{(\frac{1}{2})}`$ satisfying specific differential equations exist. In $`d=2`$ there is no such divergence. For $`d=3,4`$ the only divergent term is $`q_{(1)}`$. This results in the following condition<sup>6</sup><sup>6</sup>6$`q_{(1)}`$ is equal to $`ϵ_{(\frac{1}{2})}^{}`$ times the l.h.s. of (6.57).
$$\left(\frac{1}{(d1)}(\mathrm{\Gamma }^{\widehat{a}}D_{\widehat{a}})^2+\frac{1}{2}K_{(2)}{}_{\widehat{a}}{}^{\widehat{j}}\mathrm{\Gamma }_{}^{\widehat{a}}\mathrm{\Gamma }_{\widehat{j}}\right)ϵ_{(\frac{1}{2})}=0$$
(6.57)
where
$$K_{(2)ij}=\frac{1}{(d2)}\left(R_{ij}\frac{1}{2(d1)}R\gamma _{ij}\right).$$
(6.58)
This condition is not Weyl covariant but one can understand this as a consequence of the invariance of the Witten-Nester energy under diffeomorphisms, as we discuss in appendix D. We are not aware of a classification of manifolds that admit such spinors, but we will discuss examples below where this condition is satisfied. The conditions $`q_{(2n1)}`$ for $`n>1`$ will only be discussed for AAdS spacetimes.
## 7 Holographic energy vs Witten-Nester energy
In the previous section, we discussed necessary conditions for the Witten-Nester energy to be well defined. We assume now that these conditions hold and we discuss how the finite part compares with the holographic energy.
Using (6.48)-(6.49)-(6.52)-(6.53), we get
$$E_{WN}=\frac{1}{2\kappa ^2}_{C_t\mathrm{\Sigma }_r}d^{d1}x\sqrt{{}_{}{}^{t}\gamma }K_{(d)}{}_{\widehat{a}}{}^{\widehat{j}}ϵ_{(\frac{1}{2})}^{}\mathrm{\Gamma }^{\widehat{a}}\mathrm{\Gamma }_{\widehat{j}}ϵ_{(\frac{1}{2})}\frac{1}{2\kappa ^2}_{C_t\mathrm{\Sigma }_r}d^{d1}x\sqrt{{}_{}{}^{t}\gamma }A_{(d1)}+c.c.$$
(7.59)
where $`A_{(d1)}`$ is non-zero only for even $`d`$. A simple algebra shows that
$$K_{(d)}{}_{\widehat{a}}{}^{\widehat{j}}ϵ_{(\frac{1}{2})}^{}\mathrm{\Gamma }^{\widehat{a}}\mathrm{\Gamma }_{\widehat{j}}ϵ_{(\frac{1}{2})}=\kappa ^2T^{\widehat{t}}{}_{\widehat{i}}{}^{}\overline{ϵ}_{(\frac{1}{2})}^{}\mathrm{\Gamma }^{\widehat{i}}ϵ_{(\frac{1}{2})},$$
(7.60)
where $`T_{ij}`$ is the holographic stress energy tensor (3.15). It follows
$`d\mathrm{odd}:E_{WN}=E_h,`$ (7.61)
$`d\mathrm{even}:E_{WN}=E_hE_0,`$ (7.62)
provided $`\widehat{ϵ}_{(\frac{1}{2})}`$ is chosen such that
$$\xi ^i=\overline{\widehat{ϵ}}_{(\frac{1}{2})}\mathrm{\Gamma }^i\widehat{ϵ}_{(\frac{1}{2})}$$
(7.63)
is a timelike Killing vector of the boundary metric $`g_{(0)}`$. (The hat notation explained in (2.14)) Notice that $`g_{(0)}`$ must have a timelike Killing vector in order to define energy. We show in appendix E that if the Witten spinor is asymptotically a Killing spinor then (7.63) is automatically a timelike or null conformal Killing vector of the boundary metric. In the more general case we discuss here Killing spinors may not exists even asymptotically, but $`g_{(0)}`$ can have a timelike Killing vector. In this case (7.63) is viewed as an additional condition on $`\widehat{ϵ}_{(\frac{1}{2})}`$.
The additional term for even $`d`$, i.e. for odd dimensional bulk spacetimes, is equal to
$$E_0=\frac{1}{\kappa ^2}_{C_tM}\mathrm{Re}(\widehat{A}_{(d1)}),$$
(7.64)
where $`\widehat{A}_{(d1)}`$ is given in (6.54). It depends only on asymptotic data and is a bounded quantity. For general $`AlAdS_3`$ and $`AlAdS_5`$ spacetimes $`E_0`$ is given by
$`d=2{\displaystyle \frac{1}{2}}\widehat{A}_{(1)}=|_\varphi \widehat{ϵ}_{(\frac{1}{2})}|^2`$ (7.65)
$`d=4{\displaystyle \frac{1}{2}}\widehat{A}_{(3)}={\displaystyle \frac{1}{12}}|\widehat{K}_{\widehat{a}(2)}^{\widehat{j}}\mathrm{\Gamma }^{\widehat{a}}\mathrm{\Gamma }_{\widehat{j}}\widehat{ϵ}_{(\frac{1}{2})}|^2`$ (7.66)
where in deriving (7.66) we used the finiteness condition (6.57), $`\varphi `$ stands for the spatial boundary coordinate of $`AlAdS_3`$ and $`K_{(2)ij}`$ is given in (6.58). In the next section we will derive $`E_0`$ for $`AAdS_{d+1}`$ spacetimes.
This leads us to the main result of this paper. Consider $`AlAdS`$ spacetimes where in addition to the boundary conformal structure<sup>7</sup><sup>7</sup>7 As discussed in detail in when the conformal anomaly does not vanish identically one needs to pick a specific representative $`g_{(0)}`$ in order to define the theory. $`[g_{(0)}]`$ we also specify a boundary spinor $`\widehat{ϵ}_{(\frac{1}{2})}(x)`$. We require that $`(g_{(0)},\widehat{ϵ}_{(\frac{1}{2})})`$ are such that (i) the no-divergence conditions derived in the previous section are satisfied, (ii) $`\xi ^i`$ in (7.63) is a timelike Killing vector of $`g_{(0)}`$, (iii) a regular Witten spinor approaching $`ϵ_{(\frac{1}{2})}`$ asymptotically exists and (iv) the bulk spacetime has a single boundary or if there are more than one boundary the other boundaries should give vanishing contribution to $`E_{WN}`$.
The holographic energy of $`AlAdS`$ spacetimes with such an asymptotic structure is bounded from below
$`AlAdS_{2k}:`$ $`E_h0`$ (7.67)
$`AlAdS_{2k+1}:`$ $`E_hE_0`$ (7.68)
Spacetimes saturating the bound may be considered as “the ground state” among all spacetimes with the same asymptotic data.
Notice that $`E_0`$ depends on $`\widehat{ϵ}_{(\frac{1}{2})}`$ so if $`\widehat{ϵ}_{(\frac{1}{2})}`$ is not fixed uniquely by our requirements, $`E_0`$ in (7.68) should be understood to be the maximum among all choices. The fact that the bound in odd dimensions is non-zero is related to the fact that the Witten-Nester energy vanishes for supersymmetric solutions, but the holographic energy may not be zero, essentially because of the presence of the conformal anomaly. In fact $`E_0`$ for AAdS spacetimes is related via AdS/CFT to the Casimir energy of the dual QFT. We discuss this further in the next section.
We finish this section with a few remarks. If the boundary metric has additional (conformal) isometries the Witten-Nester construction can be generalized to include all conserved charges. This is discussed for AAdS spacetimes in (see also the recent discussion in ). In such cases we expect exact agreement between the Witten-Nester charges and the holographic charges. We also expect that one is able to relax the last requirement, namely that all contributions to $`E_{WN}`$ come from a single boundary. The case of spacetimes with horizons is discussed in . Thus the main two requirements on the asymptotic structure are the no-divergence conditions and the global existence of Witten spinors.
## 8 AAdS spacetimes
In this section we restrict our attention to $`AAdS_{d+1}`$ spacetimes. This case has been discussed previously in . These spacetimes possess asymptotic Killing spinors and we take the the Witten spinor to approach such a spinor,
$$ϵ_W(x,r)=ϵ_K(x,r)(1+𝒪(e^{dr})),$$
(8.69)
where $`e_K`$ is the AdS Killing spinor given in (F.4). Properties of AdS Killing spinors are discussed in appendix F.
Recall that the asymptotics of $`AAdS`$ start differing from $`AdS`$ at the normalizable mode order and the Witten-Nester energy is zero for $`AdS_{d+1}`$. It follows that all divergent terms in $`E_{WN}[AAdS]`$ are zero. We will shortly demonstrate this for up to $`d=6`$. Furthermore, since $`E_0`$ depends only on boundary data, it is universal among all solutions with the same asymptotics. So to evaluate it, it is sufficient to consider the case of $`AdS_{d+1}`$. We thus obtain (using $`E_{WN}[AdS]=0`$)
$$E_0=E_h[AdS].$$
(8.70)
The energy of $`AdS_{2p+1}`$ (with boundary $`R\times S^{2p1}`$) can be evaluated using the results in appendix G for any $`p`$ (for even dimensions $`E_h[AdS_{2p}]=0`$). For up to $`d=6`$ one can actually compute the energy of $`AdS_{d+1}`$ with boundary metric any conformally flat metric $`g_{(0)}`$ using the following formulae derived in (the formula for $`d=6`$ corrects typos in (3.21) of ),
$`d=2:`$ $`T_{ij}={\displaystyle \frac{1}{\kappa ^2}}[g_{(2)}g_{(0)}Trg_{(2)}]_{ij}`$ (8.71)
$`d=4:`$ $`T_{ij}={\displaystyle \frac{1}{2\kappa ^2}}[g_{(2)}^2+g_{(2)}Trg_{(2)}{\displaystyle \frac{1}{2}}g_{(0)}((Trg_{(2)})^2Trg_{(2)}^2)]_{ij}`$
$`d=6:`$ $`T_{ij}={\displaystyle \frac{1}{4\kappa ^2}}[g_{(2)}^3g_{(2)}^2Trg_{(2)}+{\displaystyle \frac{1}{2}}g_{(2)}((Trg_{(2)})^2Trg_{(2)}^2)`$
$`+g_{(0)}({\displaystyle \frac{1}{2}}Trg_{(2)}Trg_{(2)}^2{\displaystyle \frac{1}{3}}Trg_{(2)}^3{\displaystyle \frac{1}{6}}(Trg_{(2)})^3)]_{ij}`$
Specializing these results to $`g_{(0)}`$ being the metric on $`R\times S^{2p1}`$ or using the results from appendix G one obtains,
$$E_0(d=2)=\frac{\pi }{\kappa ^2},E_0(d=4)=\frac{3\pi ^2}{4\kappa ^2},E_0(d=6)=\frac{5\pi ^3}{16\kappa ^2}$$
(8.72)
In the previous section we provided a formula for $`E_0`$ in terms of $`A_{(d1)}`$, see (7.64). It is a nice check on our computations that both computations give the same answer and we demonstrate this for up to $`d=6`$.
To explicitly check the cancellation of divergences and compute $`E_0`$ we need to know the coefficients $`K_{(2n)ij}`$ and $`ϵ_{(m)}`$. These are computed in appendix G and we give here only the relevant results for the computation up to $`d=6`$,
$`K_{(2n)at}[\gamma ]=0,\stackrel{~}{K}_{ij}[\gamma ]=0,K_{ab}[\gamma ]=\gamma _{ab}{\displaystyle \underset{n=0}{}}\widehat{k}_{(2n)}\gamma ^n,`$
$`\widehat{k}_{(0)}=1,\widehat{k}_{(2)}={\displaystyle \frac{1}{2}},\widehat{k}_{(4)}={\displaystyle \frac{1}{8}},\widehat{k}_{(6)}={\displaystyle \frac{1}{16}},`$ (8.73)
where
$$\gamma _{ab}=\gamma g_{(0)ab},\gamma =e^{2r}\left(1\frac{e^{2r}}{4}\right)^2,$$
(8.74)
and $`g_{(0)ab}`$ is the standard metric of $`S^{2p1}`$. For the expansion of the Killing spinor we get,
$`e_K(x,r)={\displaystyle \underset{m=0}{}}\widehat{ϵ}_{(\frac{1}{2}+m)}\gamma ^{\frac{1}{2}(\frac{1}{2}+m)},`$
$`\widehat{ϵ}_{(\frac{3}{2})}={\displaystyle \frac{1}{8}}\widehat{ϵ}_{(\frac{1}{2})},\widehat{ϵ}_{(\frac{5}{2})}={\displaystyle \frac{1}{8}}\widehat{ϵ}_{(\frac{1}{2})},\widehat{ϵ}_{(\frac{7}{2})}={\displaystyle \frac{5}{128}}\widehat{ϵ}_{(\frac{1}{2})},\widehat{ϵ}_{(\frac{9}{2})}={\displaystyle \frac{7}{128}}\widehat{ϵ}_{(\frac{1}{2})},`$ (8.75)
where $`\widehat{ϵ}_{(\pm \frac{1}{2})}`$ are given in (F.5). Using (F.9) one easily obtains that they satisfy,
$$(\mathrm{\Gamma }^{\widehat{a}}\widehat{D}_{\widehat{a}})^2\widehat{ϵ}_{(\frac{1}{2})}=\frac{1}{4}(d1)^2\widehat{ϵ}_{(\frac{1}{2})},\widehat{ϵ}_{(\frac{1}{2})}^{}\widehat{ϵ}_{(\frac{1}{2})}=\frac{1}{4}\widehat{ϵ}_{(\frac{1}{2})}^{}\widehat{ϵ}_{(\frac{1}{2})}+\mathrm{total}\mathrm{derivative}$$
(8.76)
and we normalize as $`\widehat{ϵ}_{(\frac{1}{2})}^{}\widehat{ϵ}_{(\frac{1}{2})}=1`$.
Using these results one can explicitly evaluate $`q_{(1)}`$ and $`q_{(3)}`$ and find that they are equal to zero. Furthermore, $`\stackrel{~}{q}_{(d1)}=0`$ for AAdS since the boundary metric is conformally flat. This explicitly demonstrates that the Witten-Nester energy is well-defined for up to $`d=6`$. Furthermore, one can also easily evaluate $`A_{(d1)}`$ with result,
$$A_{(d1)}=(d1)\widehat{k}_{(d)}.$$
(8.77)
This implies that
$$E_{(0)}=E_h[AdS]$$
(8.78)
since the right hand side of (8.77) is equal to $`\kappa ^2T_{\widehat{j}}^{\widehat{t}}\xi ^{\widehat{j}}`$, where $`T_j^i`$ is the holographic stress energy tensor for $`AdS_{2p+1}`$ and $`\xi ^{\widehat{i}}`$ is the standard timelike Killing vector of $`AdS_{2p+1}`$ (i.e. $`\xi ^{\widehat{t}}=1,\xi ^{\widehat{a}}=0`$).
The ground state energy $`E_0`$ is also related to the Casimir energy of a conformal field theory on $`R\times S^{d1}`$. To see this notice that the $`R\times S^{d1}`$ is conformally related to Minkowski space. One can thus obtain the vacuum energy on $`R\times S^{d1}`$ by starting from Minkowski space where the expectation value of the energy momentum vanishes and apply the conformal transformation that maps it to $`R\times S^{d1}`$. This would lead to a zero vacuum energy if the transformations were non-anomalous, but because in even dimensions there is a conformal anomaly one gets a non-zero result. We refer to for a discussion of the $`d=2`$ and $`d=4`$ case. The fact that energy of $`AdS_5`$ is equal to the Casimir energy of $`N=4`$ SYM was first discussed in .
## 9 Other examples and global issues
So far we have derived necessary conditions for the Witten-Nester energy to be well defined. Our discussion however was local in nature and thus our conditions are certainly not sufficient. In order to complete the analysis one has to address global issues as well and establish the existence of Witten spinors with the asymptotics we discuss here. In this section we illustrate some of the subtleties by means of two examples.
We assume in this paper that the boundary admits at least one spin structure that extends in the bulk. In general, however, the boundary manifold can admit many spin structures and only a subset of those may extend to the bulk<sup>8</sup><sup>8</sup>8 A spin structure exists iff the second Steifel-Whitney class vanishes, $`0=w_2H^2(M,Z_2)`$, and the number of distinct spin structures is equal to the dimension of $`H^1(M,Z_2)`$. In particular, if $`M`$ is simply connected there is a unique spin structure.. An elementary example that exemplifies the situation is the circle $`S^1`$. It admits two spin structures: spinors can be periodic or anti-periodic around $`S^1`$. If a boundary $`S^1`$ is contractible in the interior then only the anti-periodic spinors extend, but if $`S^1`$ is not contractible both spin structures extend. An example where such issues arise is in three dimensions with boundary of topology $`R\times S^1`$. $`AdS_3`$ and the BTZ black hole have a boundary of such topology, but in $`AdS_3`$ the circle is contractible in the interior whereas in the BTZ black hole not. This is the first example we discuss below. A related discussion for more general supersymmetric spacetimes in $`2+1`$ AdS supergravity can be found in .
Another related issue is the question of regularity of the Witten spinor. One may successfully satisfy the local conditions that ensure finiteness of the Witten-Nester energy by an appropriate choice of $`\widehat{ϵ}_{(\frac{1}{2})}`$, but there may not exist a globally valid regular Witten spinor satisfying these boundary conditions. We illustrate this issue with our second example, the AdS soliton.
### 9.1 Extremal BTZ Black Hole
We discuss in the subsection the extremal BTZ black hole . The metric is given by
$$ds^2=N^2(\rho )dt^2+N^2(\rho )d\rho ^2+\rho ^2\left(d\varphi \frac{\rho _0^2}{\rho ^2}dt\right)^2$$
(9.79)
where
$$N(\rho )=\frac{1}{\rho }(\rho ^2\rho _0^2).$$
(9.80)
The spacetime has an extremal horizon at $`\rho =\rho _0`$ and a conformal boundary at $`\rho \mathrm{}`$. Introducing a new radial coordinate
$$\rho =\sqrt{e^{2r}+\rho _0^2}$$
(9.81)
we bring the metric in the form used in this paper
$$ds^2=dr^2+e^{2r}(dt^2+d\varphi ^2)+\rho _0^2(dtd\varphi )^2.$$
(9.82)
The horizon is now pushed to $`r=\mathrm{}`$ and the boundary is at $`r=\mathrm{}`$. This metric is of the general form (2.3)-(2.4) with $`g_{(0)}`$ the standard metric on $`R\times S^1`$.
The holographic stress energy tensor associated with this solution can be computed using (3.15),
$$T_{tt}=T_{\varphi \varphi }=T_{t\varphi }=\frac{\rho _0^2}{\kappa ^2},$$
(9.83)
where we used $`K_{(2)ij}=g_{(2)ij}`$ and read off $`g_{(2)}`$ from (9.82). The boundary metric has the timelike Killing vector $`\zeta _{(t)}=\zeta _{(t)}^i_i=/t`$ and the spacelike Killing vector $`\zeta _{(\varphi )}=\zeta _{(\varphi )}^i_i=/\varphi `$ and we can use them to obtain the mass and angular momentum of the solution,
$`M`$ $`=`$ $`{\displaystyle _0^{2\pi }}𝑑\varphi T_i^t\zeta _{(t)}^i={\displaystyle _0^{2\pi }}𝑑\varphi T_{tt}={\displaystyle \frac{\rho _0^2}{4G}},`$ (9.84)
$`J`$ $`=`$ $`{\displaystyle _0^{2\pi }}𝑑\varphi T_i^t\zeta _{(\varphi )}^i={\displaystyle _0^{2\pi }}𝑑\varphi T_{t\varphi }={\displaystyle \frac{\rho _0^2}{4G}},`$ (9.85)
where $`G`$ is Newton’s constant, so the metric is the extremal solution with $`M=J`$. The extremal solution with $`M=J`$ is given by the same metric but with $`G_{t\varphi }G_{t\varphi }`$. Setting $`\rho _0^2=0`$ yields the massless solution.
We now want to compute the Witten-Nester energy for the this solution. The extremal BTZ black hole admits one Killing spinor , and one could consider using it as a Witten spinor, as in our discussion of AAdS spacetimes. We therefore need the explicit form of the Killing spinor. The vielbein and spin connection of the metric (9.82) are given by
$`E^{\widehat{t}}=N(r)dt,E^{\widehat{r}}=dr,E^{\widehat{\varphi }}=\rho (r)d\varphi {\displaystyle \frac{\rho _0^2}{\rho (r)}}dt,`$ (9.86)
$`\omega ^{\widehat{r}}{}_{\widehat{\varphi }}{}^{}=N(r)d\varphi ,\omega ^{\widehat{\varphi }}{}_{\widehat{t}}{}^{}={\displaystyle \frac{\rho _0^2}{\rho (r)^2}}dr,\omega ^{\widehat{t}}{}_{\widehat{r}}{}^{}={\displaystyle \frac{\rho _0^2}{\rho (r)}}d\varphi +\rho (r)dt`$
where in these formulas $`N`$ and $`\rho `$ are understood to be functions of $`r`$ (cf (9.80) and (9.81)). A straightforward computation shows that the Killing spinor is given by
$$ϵ=\sqrt{N(r)}\widehat{ϵ}_{(\frac{1}{2})},$$
(9.87)
where $`\widehat{ϵ}_{(\frac{1}{2})}`$ is a constant spinor satisfying the following conditions
$$P^+\widehat{ϵ}_{(\frac{1}{2})}=P_{\varphi t}^{}\widehat{ϵ}_{(\frac{1}{2})}=0$$
(9.88)
where $`P^+=\frac{1}{2}(1+\mathrm{\Gamma }^{\widehat{r}})`$ and $`P_{\varphi t}^{}=\frac{1}{2}(1\mathrm{\Gamma }^{\widehat{\varphi }\widehat{t}})`$. In (9.88) we impose two projections on a two dimensional spinor, so one might think that that there are no non-trivial solutions. In three dimensions however there are two inequivalent representations of the gamma matrices: (i) $`\mathrm{\Gamma }^{\widehat{t}}=i\sigma ^2,\mathrm{\Gamma }^{\widehat{\varphi }}=\sigma ^1,\mathrm{\Gamma }^{\widehat{r}}=\sigma ^3`$, where $`\sigma ^k`$ are the Pauli matrices, and (ii) $`\mathrm{\Gamma }^{\widehat{i}}=\mathrm{\Gamma }^{\widehat{i}}`$. In representation (i) we find that $`\mathrm{\Gamma }^{\widehat{r}}=\mathrm{\Gamma }^{\widehat{\varphi }t}`$ and therefore $`P^+=P_{\varphi t}^{}`$, so (9.88) admits a non-trivial solution. Notice that the Killing spinor $`ϵ`$ is periodic in $`\varphi `$, actually it is constant in $`\varphi `$, where the corresponding AdS Killing spinor (F.4) is anti-periodic.
We now choose as a Witten spinor the Killing spinor (9.87). The projection in (9.88) implies $`\mathrm{\Gamma }^{\widehat{t}}\widehat{ϵ}_{(\frac{1}{2})}=\mathrm{\Gamma }^{\widehat{\varphi }}\widehat{ϵ}_{(\frac{1}{2})}`$ and this in turn implies that the boundary Killing vector,
$$\xi ^{\widehat{i}}=\overline{\widehat{ϵ}}_{(\frac{1}{2})}\mathrm{\Gamma }^{\widehat{i}}\widehat{ϵ}_{(\frac{1}{2})}$$
(9.89)
is a null Killing vector (since $`\xi ^{\widehat{t}}=\xi ^{\widehat{\varphi }}`$). Choosing $`|\widehat{ϵ}_{(\frac{1}{2})}|^2=1`$ we have
$$\xi =\zeta _{(t)}+\zeta _{(\varphi )}.$$
(9.90)
Let us now compute the corresponding Witten-Nester conserved charge. First we compute the ground state “energy”,
$$\frac{1}{2}\widehat{A}_{(1)}=\widehat{ϵ}_{(\frac{1}{2})}^{}_\varphi ^2\widehat{ϵ}_{(\frac{1}{2})}=0,E_0=0,$$
(9.91)
since the spinor $`\widehat{ϵ}_{(\frac{1}{2})}`$ is constant. One should contrast this with the case of $`AdS_3`$, where $`E_0=\pi /\kappa ^2`$. We thus obtain,
$$E_{WN}=_0^{2\pi }𝑑\varphi T_i^t\xi ^i=M+J=0,$$
(9.92)
as expected since the Witten-Nester energy is by construction equal to zero for Witten spinors that are equal to Killing spinors. In other words, the Witten-Nester conserved charge is a linear combination of the mass and angular momentum.
The Witten spinor, however, need not be equal to a Killing spinor. To obtain a Witten-Nester expression for the mass we now consider the following spinor,
$$ϵ^{}=\sqrt{N(r)}\widehat{ϵ}_{(\frac{1}{2})}^{},$$
(9.93)
where
$$P^+\widehat{ϵ}_{(\frac{1}{2})}^{}=P_{\varphi t}^+\widehat{ϵ}_{(\frac{1}{2})}^{}=0$$
(9.94)
In order for this expression to admit a non-trivial solution we must work with the irreducible representation (ii) where $`P^+=P_{\varphi t}^+`$.
To show that this is a Witten spinor we compute,
$$_{\widehat{r}}ϵ^{}=\frac{\rho _0^2}{\rho ^2}ϵ^{},_{\widehat{\varphi }}ϵ^{}=\frac{\rho _0^2}{\rho ^2}\mathrm{\Gamma }_{\widehat{\varphi }}ϵ^{}$$
(9.95)
from which we obtain
$$(\mathrm{\Gamma }^{\widehat{r}}_{\widehat{r}}+\mathrm{\Gamma }^{\widehat{\varphi }}_{\widehat{\varphi }})ϵ^{}=0.$$
(9.96)
This Witten spinor is associated with the null Killing vector field $`\xi ^{\widehat{i}}=\overline{\widehat{ϵ}^{}}_{(\frac{1}{2})}\mathrm{\Gamma }^{\widehat{i}}\widehat{ϵ}_{(\frac{1}{2})}^{}`$, where we normalize $`|\widehat{ϵ}_{(\frac{1}{2})}^{}|^2=\frac{1}{2}`$,
$$\xi ^i_i=\overline{\widehat{ϵ}^{}}_{(\frac{1}{2})}\mathrm{\Gamma }^i\widehat{ϵ}_{(\frac{1}{2})}^{}_i=\frac{1}{2}(_{\widehat{t}}_{\widehat{\varphi }})=\frac{1}{2}(\zeta _{(t)}\zeta _{(\varphi )}).$$
(9.97)
Let us now compute the Witten-Nester energy. The ground state energy $`E_0`$ is zero because $`\widehat{ϵ}_{(\frac{1}{2})}^{}`$ is constant and
$$E_{WN}^{}=_0^{2\pi }𝑑\varphi T_i^t\xi ^i=\frac{1}{2}(MJ)=M.$$
(9.98)
Notice that the Witten spinor is regular for $`\rho ^2\rho _0^2`$ or equivalently $`r>\mathrm{}`$. Furthermore, a possible contribution to the Witten-Nester energy from the horizon vanishes since the Witten spinor vanishes at the horizon.
This example illustrates a number of points. Firstly, we see explicitly the dependence of the Witten-Nester construction on the spin structure and on the choice of Witten spinor. For $`AdS_3`$ one must choose anti-periodic boundary conditions for the Witten spinor and the dependence of $`\widehat{ϵ}_{(\frac{1}{2})}`$ on the $`S^1`$ coordinate $`\varphi `$ gives rise to the ground state energy $`E_0`$. In the BTZ case however the circle is not contractible and periodic spinors are allowed. In fact one must choose periodic spinors if one wants to preserve supersymmetry. With this choice the ground state energy vanishes. Another point that is illustrated by this example is that one may have to consider Witten spinors that do not approach a Killing spinor asymptotically in order to obtain all conserved charges.
### 9.2 AdS Soliton
In this subsection we discuss the AdS soliton . This solution has a toroidal boundary and negative energy but it has been conjectured that it is the lowest energy solution within its asymptotic class. This was checked for small perturbations in - and additional support for this conjecture was presented in , . The negative energy was shown in to be (proportional to) the Casimir energy of $`N=4`$ SYM on $`R\times T^3`$. In our general discussion we found that the Witten-Nester energy is equal to the holographic energy up to a ground state energy, which is present in odd dimensions. This ground state energy for AAdS had the interpretation of Casimir energy for the dual CFT on $`R\times S^{d1}`$. So one could have hoped that similar discussions would prove the positive energy conjecture of . However, inspection of the results in the literature and our results in section 7 shows that this cannot be the case. The AdS soliton has negative mass in all dimensions and this is incompatible with the bounds in section 7. In particular, the mass of even dimensional AlAdS spacetimes is bounded by zero and of $`AAdS_5`$ by a positive quantity. It will be instructive however to understand why our considerations do not apply in this case.
The metric for the five-dimensional AdS soliton is given by
$$ds^2=\frac{1}{N(\rho )^2}\frac{d\rho ^2}{\rho ^2}+\rho ^2\left(N(\rho )^2d\tau ^2dt^2+dx^2+dy^2\right),$$
(9.99)
where
$$N(\rho )=\sqrt{1\frac{a^4}{\rho ^4}}.$$
(9.100)
Regularity requires that $`\tau `$ is identified with period $`\frac{\pi }{a}`$, and we take $`x,y`$ to be periodic with periods $`R_x,R_y`$, respectively.
A change of the radial coordinate,
$$e^r=\frac{a^2}{\sqrt{2}\rho \sqrt{1N(\rho )}},\mathrm{or}\rho ^2=e^{2r}+\frac{a^2}{4}e^{2r},$$
(9.101)
brings the metric in the form used in this paper,
$`ds^2`$ $`=`$ $`dr^2+\left(e^{2r}+{\displaystyle \frac{a^2}{4}}e^{2r}\right)(dt^2+dx^2+dy^2)+{\displaystyle \frac{(e^{2r}\frac{a^2}{4}e^{2r})^2}{(e^{2r}+\frac{a^2}{4}e^{2r})}}d\tau ^2`$
$`=`$ $`dr^2+e^{2r}\eta _{ij}dx^idx^j+e^{2r}{\displaystyle \frac{a^4}{4}}(dt^2+dx^2+dy^23d\tau ^2)+𝒪(e^{4r})d\tau ^2.`$
From this metric we can read off the coefficients $`g_{(n)}`$ and obtain the coefficient $`\widehat{K}_{(n)}`$ by using (2). Up to $`n=4`$ the only non-zero coefficients are
$$\widehat{K}_{(4)}{}_{\widehat{t}}{}^{\widehat{t}}=\widehat{K}_{(4)}{}_{\widehat{x}}{}^{\widehat{x}}=\widehat{K}_{(4)}{}_{\widehat{y}}{}^{\widehat{y}}=\frac{1}{2}a^4\widehat{K}_{(4)}{}_{\widehat{\tau }}{}^{\widehat{\tau }}=\frac{3}{2}a^4$$
(9.103)
This boundary metric has a timelike Killing vector $`\xi =/t`$, and we can use (3.17) to compute the mass of the soliton,
$$E_h=d^3xT_{tt}=\frac{R_xR_ya^3}{16G}$$
(9.104)
in agreement with .
We now turn to the discussion of the Witten-Nester energy for this solution. Let us assume for the moment that the no-divergence condition in (6.57) holds. The ground state energy $`E_0`$ is zero in this case since
$$\frac{1}{2}\widehat{A}_{(3)}=\frac{1}{12}|\widehat{K}_{(2)}{}_{\widehat{a}}{}^{\widehat{j}}\mathrm{\Gamma }_{}^{\widehat{a}}\mathrm{\Gamma }_{\widehat{j}}\widehat{ϵ}_{(\frac{1}{2})}|^2=0$$
(9.105)
because $`K_{(2)ij}=0`$. To obtain the Witten-Nester energy, we compute
$$\widehat{q}_{(3)}=\frac{1}{2}\widehat{ϵ}_{(\frac{1}{2})}^{}\widehat{K}_{\widehat{a}(4)}^{\widehat{j}}\mathrm{\Gamma }^{\widehat{a}}\mathrm{\Gamma }_{\widehat{j}}\widehat{ϵ}_{(\frac{1}{2})}=\frac{a^4}{4}|\widehat{ϵ}_{(\frac{1}{2})}|^2<0,$$
(9.106)
so provided we can normalize $`|\widehat{ϵ}_{(\frac{1}{2})}|^2=1`$ we obtain
$$E_{WN}=E_h<0$$
(9.107)
which contradicts the positivity property of the Witten-Nester energy.
Let us now discuss the no-divergence condition (6.57) which for the solution at hand reads,
$$\widehat{ϵ}_{(\frac{1}{2})}^{}(_x^2+_y^2+_\tau ^2)\widehat{ϵ}_{(\frac{1}{2})}+c.c.=0.$$
(9.108)
This condition is solved by a constant $`\widehat{ϵ}_{(\frac{1}{2})}`$ (which may be normalized to one) so our asymptotic conditions are satisfied. Integrating the Witten equation with this boundary condition leads to
$$ϵ=\frac{a^2}{2}\frac{1}{\rho ^{3/2}}\frac{1}{\sqrt{N(\rho )}}\sqrt{\frac{1+N(\rho )}{1N(\rho )}}\widehat{ϵ}_{(\frac{1}{2})},$$
(9.109)
which is singular at $`\rho =a`$. It follows that the step from (4.21) to (4.22) relating the manifestly positive bulk integral to a surface integral does not go through in this case.
One might have anticipated problems with regularity of the Witten spinor since the circle corresponding to $`\tau `$ is contractible in the interior, so the Witten spinor and in particular $`\widehat{ϵ}_{(\frac{1}{2})}`$ should be anti-periodic in $`\tau `$. However, our $`\widehat{ϵ}_{(\frac{1}{2})}`$, and thus the Witten spinor in (9.109), is periodic. If we demand that $`\widehat{ϵ}_{(\frac{1}{2})}`$ is antiperiodic in $`\tau `$, the condition (9.108) cannot be satisfied (with $`\widehat{ϵ}_{(\frac{1}{2})}`$ periodic or anti-periodic in $`x,y`$), and the Witten-Nester energy is not well-defined.
## 10 Conclusions
We derived in this paper conditions on the asymptotic structure of asymptotically locally AdS spacetimes such that their mass is bounded from below. This was done by computing a regulated version of the manifestly positive spinorial Witten-Nester energy and analyzing the condition for this energy to be finite.
The spinorial energy $`E_{WN}`$ is constructed from Witten spinors, i.e. spinor fields satisfying a Dirac-like equation on the initial-value hypersurface. It can be written either as a bulk integral or as a surface integral at infinity. The former is manifestly positive and the latter provides the connection with the conserved charges. The two expressions are equivalent, provided the Witten spinors are regular. For $`AlAdS`$ spacetimes, the surface integral is not automatically finite, thus we introduce a cut-off $`r`$ in the radial direction to regulate the theory, as in previous work on holographic renormalization. The regulated $`E_{WN}[r]`$ can now be computed for general AlAdS spacetimes, provided that we know asymptotic solutions of the Witten equation. We computed the most general asymptotic solutions of the Witten equation using methods similar to the ones in . As a technical remark, we note that the use of the formalism of (instead of the near boundary expansion of ) was instrumental in allowing us to carry out this computation. The coefficients of the asymptotic Witten spinors are determined locally (up to a specific order) from the (still arbitrary at this stage) boundary value of the Witten spinor $`ϵ_{(\frac{1}{2})}`$ and the boundary vielbein.
Having solved the Witten equation asymptotically, we then computed the regulated Witten-Nester energy. The expression involves a number of local power-law divergences and a logarithmic divergence in odd dimensions. This means that not all AlAdS spacetimes possess a finite positive Witten-Nester energy. The ones that do, have asymptotic data such that all divergences vanish identically. Thus the vanishing of the divergences provides necessary conditions on the asymptotic data for the spacetime to possess a finite Witten-Nester energy. The vanishing of the logarithmic divergence in odd dimensions implies that the even dimensional conformal boundary should be a conformally Einstein manifold. The number of power law divergences depend on the spacetime dimension. In dimension three there are no power law divergences and in dimensions 4 and 5 there is one such divergence. In these dimensions, the vanishing of the divergence implies that the boundary manifold should admit a spinor satisfying a particular differential equation. It would be interesting to classify the four dimensional conformally Einstein spaces that admit such spinors. Such a list would provide curved backgrounds for which $`𝒩=4`$ SYM is expected to be well defined. Higher dimensions were only analyzed for AAdS spacetimes, i.e. for spacetimes that asymptotically approach the exact AdS solution. In this case, all no-divergence conditions are satisfied if we take the Witten spinor to approach asymptotically an AdS Killing spinor.
Having established the condition for finiteness we compared the finite part of the Witten-Nester energy with the expression of the holographic energy, $`E_h`$. In even dimensions the two agree exactly and in odd dimensions they differ by a bounded quantity which only depends on the asymptotic data. We give an explicit expression of the bound for $`AlAdS_3`$, $`AlAdS_5`$ and discuss it for all AAdS spacetimes. A general feature is that it is negative in $`4k1`$ dimensions and positive in $`4k+1`$ dimensions ($`k=1,2,\mathrm{}`$). This difference between $`E_{WN}`$ and $`E_h`$ in odd dimensions is due to the fact that $`E_{WN}`$ is by construction equal to zero for supersymmetric solutions, while the holographic energy may not be zero because of the conformal anomaly.
In this paper we only analyzed local properties that follow from the asymptotic analysis. In order to rigorously establish the bounds, one has to show existence of Witten spinors with the asymptotics we discuss. It is clear from examples that such a discussion will depend sensitively on global properties. For example, one would have to understand the dependence of the construction on spin structures. To illustrate such subtleties we discussed two examples, the extremal BTZ black hole and the AdS soliton. The extremal BTZ and $`AdS_3`$ spacetimes have the same conformal boundary, but the Killing spinors are periodic (along the compact boundary direction) in the supersymmetric BTZ case and antiperiodic in the case of $`AdS_3`$. The energy bound in these cases thus depends on the spin structure. The example of the BTZ black hole also illustrates the fact that, in order to construct all conserved charges, it may be necessary in some cases to consider Witten spinors that do not approach asymptotically bulk Killing spinors. The AdS soliton gives an example where all local requirements can be satisfied, but a global regular Witten spinor with these boundary conditions does not exist.
In this paper we have restricted our attention to the case of pure gravity, but the discussion can be generalized to include matter. This is interesting both intrinsically and also from the point of view of the AdS/CFT correspondence. A particularly interesting case is that of domain wall backgrounds since they are dual to holographic RG flows. A stability analysis for a class of such spacetimes was presented in . An extension of our analysis in this direction will lead to a systematic search for stable backgrounds supported by matter fields.
## Acknowledgments
We would like to thank D. Freedman, C. Núñez and M. Schnabl for discussions. MC would like to thank A. Strominger and the Center for Mathematical Sciences in Zhejiang university, where part of this work was completed, for the hospitality. KS is supported by NWO and MC by FOM.
## Appendix A Conventions and Notations
Our index conventions are as follows
$$\{\mu \}=\{r,\{i\}\};\{i\}=\{t,\{a\}\};\{a\}=\{1,2,\mathrm{}.,d1\};\{\alpha \}=\{r,\{a\}\},$$
(A.1)
and hatted indices stand for flat indices. We use mostly plus signature. Our spinor conventions and covariant derivatives are given by
$`\overline{ϵ}=ϵ^{}\mathrm{\Gamma }_{\widehat{t}}`$ (A.2)
$`\{\mathrm{\Gamma }^\mu ,\mathrm{\Gamma }^\nu \}=2G^{\mu \nu }`$ (A.3)
$`(\mathrm{\Gamma }^{\widehat{a}})^{}=\mathrm{\Gamma }^{\widehat{a}},(\mathrm{\Gamma }^{\widehat{r}})^{}=\mathrm{\Gamma }^{\widehat{r}},(\mathrm{\Gamma }^{\widehat{t}})^{}=\mathrm{\Gamma }^{\widehat{t}}`$ (A.4)
$`𝒟_\mu ϵ=(_\mu +{\displaystyle \frac{1}{4}}\mathrm{\Omega }_\mu ^{\widehat{\nu }\widehat{\rho }}\mathrm{\Gamma }_{\widehat{\nu }\widehat{\rho }})ϵ`$ (A.5)
$`𝒟_{[\mu }𝒟_{\nu ]}ϵ={\displaystyle \frac{1}{8}}R_{\mu \nu \rho \delta }\mathrm{\Gamma }^{\rho \delta }ϵ`$ (A.6)
$`_\mu ϵ(𝒟_\mu +{\displaystyle \frac{1}{2}}\mathrm{\Gamma }_\mu )ϵ.`$ (A.7)
## Appendix B The radial and the time slices
We list here the various slices used in the main text. We consider an AlAdS spacetime $`M`$ with conformal boundary $`M`$. As discussed in the main text, we can always choose coordinates near the boundary where the metric looks like
$`ds^2`$ $`=`$ $`G_{\mu \nu }dx^\mu dx^\nu =dr^2+\gamma _{ij}(x,r)dx^idx^j`$ (B.1)
$`=`$ $`E^{\widehat{r}}E^{\widehat{r}}+{\displaystyle \underset{\widehat{i},\widehat{j}}{}}\eta _{\widehat{i}\widehat{j}}E^{\widehat{i}}E^{\widehat{j}},`$
where $`\eta _{ij}`$ is the Minkowski metric in $`d`$ dimensions and we introduce the vielbein 1-forms
$$E^{\widehat{\mu }}=E_\mu ^{\widehat{\mu }}dx^\mu .$$
(B.2)
The choice of coordinates in (B.1) implies that we can choose
$$E_r^{\widehat{r}}=1+𝒪(e^{(d1)r/2}),E_{\widehat{r}}^r=1+𝒪(e^{(d1)r/2}),E_r^{\widehat{i}}=𝒪(e^{(d1)r/2}),E_{\widehat{r}}^i=𝒪(e^{(d1)r/2}),$$
(B.3)
which implies
$$E^{\widehat{r}}=dr,\stackrel{}{}_{\widehat{r}}E_{\widehat{r}}^\mu _\mu =_r.$$
(B.4)
up to the order indicated above.
### The radial slice $`\mathrm{\Sigma }_r`$
The radial slice is defined by its normal $`E^{\widehat{r}}`$. With the coordinate choice in (B.1), this is the $`r=const`$ slice. The induced metric is given by
$$ds_r^2=\gamma _{ij}(r,x)dx^idx^j=\underset{\widehat{i},\widehat{j}}{}\eta _{\widehat{i}\widehat{j}}E^{\widehat{i}}E^{\widehat{j}}$$
(B.5)
As $`r\mathrm{}`$, $`\mathrm{\Sigma }_r`$ approaches the conformal boundary $`M`$. In this limit the induced metric $`\gamma _{ij}(x,r)`$ blows up and only a conformal structure is well-defined. One can pick a specific representative $`g_{(0)}`$ by a specific choice of defining function $`\rho `$ (a defining function is a positive function that has a single zero at the boundary). Choosing as defining function $`\rho =e^r`$ we get as a boundary metric
$$ds_{\mathrm{}}^2=g_{(0)ij}dx^idx^j=\underset{\widehat{i},\widehat{j}}{}\eta _{\widehat{i}\widehat{j}}e^{\widehat{i}}e^{\widehat{j}}$$
(B.6)
where $`e^{\widehat{i}}=lim_r\mathrm{}e^rE^{\widehat{i}}`$.
### The time slice $`C_t`$
We consider the time slice $`C_t`$ defined by its normal $`E^{\widehat{t}}`$. The induced metric on $`C_t`$ is
$`ds_t^2`$ $``$ $`{}_{}{}^{t}G_{\mu \nu }^{}dx^\mu dx^\nu =\left(G_{\mu \nu }+E_\mu ^{\widehat{t}}E_\nu ^{\widehat{t}}\right)dx^\mu dx^\nu `$ (B.7)
$`=`$ $`{\displaystyle \underset{\alpha }{}}E^{\widehat{\alpha }}E^{\widehat{\alpha }}=E^{\widehat{r}}E^{\widehat{r}}+{\displaystyle \underset{a}{}}E^{\widehat{a}}E^{\widehat{a}}.`$
### The boundary of the time slice $`C_t\mathrm{\Sigma }_r`$
The induced metric on the intersection of the two slices is
$$ds_{(rt)}^2{}_{}{}^{t}\gamma _{ij}^{}dx^idx^j=\left(\gamma _{ij}+E_i^{\widehat{t}}E_j^{\widehat{t}}\right)dx^idx^j=\underset{a}{}E^{\widehat{a}}E^{\widehat{a}}$$
As $`r\mathrm{}`$, and with the same defining function as before, we get for the metric on $`C_tM`$
$$ds_{\mathrm{},t}^2{}_{}{}^{t}g_{(0)ij}^{}dx^idx^j=(g_{(0)ij}+e_i^{\widehat{t}}e_j^{\widehat{t}})dx^idx^j,$$
(B.8)
where $`e_i^{\widehat{j}}=lim_r\mathrm{}(e^rE_i^{\widehat{j}})`$ is the vielbein of the boundary metric $`g_{(0)ij}`$.
## Appendix C Asymptotic expansions
In this appendix we present some of the technical details needed in order to obtain the asymptotic solution of the Witten spinor.
The dilatation operator (2.7) is given in terms of the vielbein by
$$\delta _D=d^dxE_i^{\widehat{j}}\frac{\delta }{\delta E_i^{\widehat{j}}},$$
(C.9)
and the second fundamental form and radial derivative admits the expansions
$`K_j^i[\gamma ]`$ $`=`$ $`\delta _j^i+{\displaystyle \underset{k=1}{\overset{[\frac{d1}{2}]}{}}}K_{(2k)}{}_{j}{}^{i}+K_{(d)}{}_{j}{}^{i}+\stackrel{~}{K}_{(d)}{}_{j}{}^{i}(2r)+\mathrm{}`$ (C.10)
$`_r`$ $`=`$ $`\delta _D+{\displaystyle \underset{k=1}{\overset{[\frac{d1}{2}]}{}}}_{r(2k)}+_{r(d)}(2r)\stackrel{~}{}_{r(d)}+\mathrm{}`$
$`_{r(n)}`$ $`=`$ $`{\displaystyle d^dxK_{(n)ij}\frac{\delta }{\delta \gamma _{ij}}},\stackrel{~}{}_{r(d)}={\displaystyle d^dx\stackrel{~}{K}_{(d)ij}\frac{\delta }{\delta \gamma _{ij}}},`$ (C.11)
where $`[\frac{d1}{2}]`$ denotes the integer part of $`\frac{d1}{2}`$ and $`\stackrel{~}{K}_{(d)ij}`$ is zero when $`d`$ is odd. Furthermore, since $`K_{ij}=\dot{E}_{(i}^{\widehat{i}}E_{j)}^{\widehat{j}}\eta _{\widehat{i}\widehat{j}}`$, $`\dot{E}_i^{\widehat{i}}`$ admits an expansion of the form
$$_rE_i^{\widehat{j}}=E_i^{\widehat{j}}+\underset{k=0}{\overset{d1}{}}\dot{E}_{(k)}{}_{i}{}^{\widehat{j}}+(2r)\dot{\stackrel{~}{E}}_{(d1)}{}_{i}{}^{\widehat{j}}+\mathrm{}$$
(C.12)
In our coordinate system, the spin connections are given by
$$\mathrm{\Omega }_r^{\widehat{i}\widehat{j}}=E^{k[\widehat{i}}_rE_k^{\widehat{j}]},\mathrm{\Omega }_i^{\widehat{j}\widehat{r}}=E^{k\widehat{j}}K_{ik},\mathrm{\Omega }_i^{\widehat{i}\widehat{j}},$$
(C.13)
and the covariant derivatives take the form,
$`_rϵ`$ $`=`$ $`(_r+{\displaystyle \frac{1}{4}}E^{k[\widehat{i}}_rE_k^{\widehat{j}]}\mathrm{\Gamma }_{\widehat{i}\widehat{j}}+{\displaystyle \frac{1}{2}}\mathrm{\Gamma }_r)ϵ,`$
$`_iϵ`$ $`=`$ $`(D_i+{\displaystyle \frac{1}{2}}E^{k\widehat{j}}K_{ik}\mathrm{\Gamma }_{\widehat{j}\widehat{r}}+{\displaystyle \frac{1}{2}}\mathrm{\Gamma }_i)ϵ.`$ (C.14)
where $`D_i`$ is the covariant derivative of the induced metric $`\gamma _{ij}`$.
Using the results above one can work out the asymptotic expansion of the covariant derivatives and the operator $`/\mathrm{\Gamma }^{\widehat{r}}_{\widehat{r}}+\mathrm{\Gamma }^{\widehat{a}}_{\widehat{a}}`$ that appear in the Witten equation,
$`/_{(0)}`$ $`=`$ $`(\delta _D+{\displaystyle \frac{d1}{2}})\mathrm{\Gamma }^{\widehat{r}}+{\displaystyle \frac{d}{2}}`$
$`/_{(1)}`$ $`=`$ $`\mathrm{\Gamma }^{\widehat{a}}D_{\widehat{a}}`$ (C.15)
$`/_{(2n)}`$ $`=`$ $`\left(2{\displaystyle d^dx\gamma _{ik}K_{(2n)}{}_{j}{}^{i}\frac{\delta }{\delta \gamma _{jk}}}+{\displaystyle \frac{1}{4}}E^{k[\widehat{i}}\dot{E}_{(2n1)}{}_{k}{}^{\widehat{j}]}\mathrm{\Gamma }_{\widehat{i}\widehat{j}}^{}+{\displaystyle \frac{1}{2}}K_{(2n)}{}_{\widehat{a}}{}^{\widehat{j}}\mathrm{\Gamma }_{}^{\widehat{a}}\mathrm{\Gamma }_{\widehat{j}}\right)\mathrm{\Gamma }^{\widehat{r}}`$
$`/_{(d)}`$ $`=`$ $`\left(2{\displaystyle d^dx\gamma _{ik}K_{(d)}{}_{j}{}^{i}\frac{\delta }{\delta \gamma _{jk}}}+{\displaystyle \frac{1}{4}}E^{k[\widehat{i}}\dot{E}_{(d1)}{}_{k}{}^{\widehat{j}]}\mathrm{\Gamma }_{\widehat{i}\widehat{j}}^{}+{\displaystyle \frac{1}{2}}K_{(d)}{}_{\widehat{a}}{}^{\widehat{j}}\mathrm{\Gamma }_{}^{\widehat{a}}\mathrm{\Gamma }_{\widehat{j}}\right)\mathrm{\Gamma }^{\widehat{r}}`$
$`\stackrel{~}{/}_{(d)}`$ $`=`$ $`\left(2{\displaystyle d^dx\gamma _{ik}\stackrel{~}{K}_{(d)}{}_{j}{}^{i}\frac{\delta }{\delta \gamma _{jk}}}+{\displaystyle \frac{1}{4}}E^{k[\widehat{i}}\dot{\stackrel{~}{E}}_{(d1)}{}_{k}{}^{\widehat{j}]}\mathrm{\Gamma }_{\widehat{i}\widehat{j}}^{}+{\displaystyle \frac{1}{2}}\stackrel{~}{K}_{(d)}{}_{\widehat{a}}{}^{\widehat{j}}\mathrm{\Gamma }_{}^{\widehat{a}}\mathrm{\Gamma }_{\widehat{j}}\right)\mathrm{\Gamma }^{\widehat{r}}.`$
where $`n=1,2,\mathrm{}.,[\frac{d1}{2}]`$. Note also that $`\delta _DE^{k\widehat{i}}=E^{k\widehat{i}}`$, so $`E^{k\widehat{i}}E_{(1)}^{k\widehat{i}}`$ and that $`\stackrel{~}{/}_{(2k+1)}=0`$.
Observe that
$`[\stackrel{~}{/}_{(d)},P^\pm ]=0,[/_{(k)},P^\pm ]=0(k1),/_{(1)}P^\pm =P^{}/_{(1)}`$
$`/_{(0)}P^{}=(\delta _D+{\displaystyle \frac{1}{2}})P^{},/_{(0)}P^+=(\delta _D+d{\displaystyle \frac{1}{2}})P^+`$ (C.16)
We will also use in the main text the asymptotic expansion of the operator $`\mathrm{\Gamma }^{\widehat{a}}_{\widehat{a}}`$. It is given by
$`\mathrm{\Gamma }^{\widehat{a}}_{\widehat{a}(0)}`$ $`=`$ $`{\displaystyle \frac{d1}{2}}\mathrm{\Gamma }^{\widehat{r}}+{\displaystyle \frac{d1}{2}}=(d1)P^+`$
$`\mathrm{\Gamma }^{\widehat{a}}_{\widehat{a}(1)}`$ $`=`$ $`\mathrm{\Gamma }^{\widehat{a}}D_{\widehat{a}}`$ (C.17)
$`\mathrm{\Gamma }^{\widehat{a}}_{\widehat{a}(2n)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}K_{\widehat{a}(2n)}^{\widehat{j}}\mathrm{\Gamma }^{\widehat{a}}\mathrm{\Gamma }_{\widehat{j}}\mathrm{\Gamma }^{\widehat{r}},n=1,2,\mathrm{}.,[{\displaystyle \frac{d1}{2}}]`$
$`\mathrm{\Gamma }^{\widehat{a}}_{\widehat{a}(d)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}K_{\widehat{a}(d)}^{\widehat{j}}\mathrm{\Gamma }^{\widehat{a}}\mathrm{\Gamma }_{\widehat{j}}\mathrm{\Gamma }^{\widehat{r}}`$
$`\mathrm{\Gamma }^{\widehat{a}}\stackrel{~}{}_{\widehat{a}(d)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\stackrel{~}{K}_{\widehat{a}(d)}^{\widehat{j}}\mathrm{\Gamma }^{\widehat{a}}\mathrm{\Gamma }_{\widehat{j}}\mathrm{\Gamma }^{\widehat{r}}.`$
## Appendix D Properties under Weyl transformations
We discuss in this appendix the Weyl transformation properties of the conditions of the absence of divergences. We focus on the Weyl transformation properties of (the integral of) $`q_{(1)}`$ but the discussion can be extend to the other coefficients.
We wish to know how the leading divergence
$`{\displaystyle _{C_t\mathrm{\Sigma }_r}}\sqrt{{}_{}{}^{t}\gamma }q_{(1)}={\displaystyle d^{d1}x\sqrt{{}_{}{}^{t}g_{(0)}^{}}e^{(d2)r}\widehat{q}_{(1)}(1+𝒪(e^{2r}))}`$ (D.18)
$`={\displaystyle d^{d1}x\sqrt{{}_{}{}^{t}g_{(0)}^{}}e^{(d2)r}\widehat{ϵ}_{(\frac{1}{2})}^{}\left(\frac{1}{(d1)}(\mathrm{\Gamma }^{\widehat{a}}\widehat{D}_{\widehat{a}})^2+\frac{1}{2}\widehat{K}_{(2)}{}_{\widehat{a}}{}^{\widehat{j}}\mathrm{\Gamma }_{}^{\widehat{a}}\mathrm{\Gamma }_{\widehat{j}}\right)\widehat{ϵ}_{(\frac{1}{2})}\left(1+𝒪(e^{2r})\right)}`$
transforms under a local Weyl transformation,
$$\overline{g}_{(0)ij}=e^{2\sigma (x)}g_{(0)ij}.$$
(D.19)
(The hat notation is defined in (2.14)). In order $`\overline{g}_{(0)ij}`$ to admit a timelike Killing vector we need to impose the following condition on $`\sigma (x)`$:
$$\sigma _{,\widehat{t}}e_{\widehat{t}}^i_i\sigma =0,$$
(D.20)
The Weyl transformation of (D.18) can be worked out using
$`\overline{\widehat{D}}_{\widehat{a}}`$ $`=`$ $`e^\sigma \left(\widehat{D}_{\widehat{a}}+{\displaystyle \frac{1}{2}}\sigma _{,\widehat{b}}\mathrm{\Gamma }_{\widehat{a}}^{\widehat{b}}\right)`$ (D.21)
$`\overline{\widehat{K}}_{(2)ij}`$ $`=`$ $`\widehat{K}_{(2)ij}\widehat{D}_i\widehat{D}_j\sigma +\widehat{D}_i\sigma \widehat{D}_j\sigma {\displaystyle \frac{1}{2}}(\widehat{D}\sigma )^2g_{(0)ij}`$
$`\overline{\widehat{ϵ}}_{(\frac{1}{2})}`$ $`=`$ $`e^{\sigma /2}\widehat{ϵ}_{(\frac{1}{2})}.`$
Using these results we derive,
$$d^{d1}x\sqrt{{}_{}{}^{t}g_{(0)}^{}}e^{(d2)r}\widehat{q}_{(1)}=d^{d1}x\sqrt{{}_{}{}^{t}\overline{g}_{(0)}^{}}e^{(d2)(r\sigma )}\left(\overline{\widehat{q}}_{(1)}\overline{\widehat{ϵ}}_{(\frac{1}{2})}^{}\sigma _{,\widehat{b}}\mathrm{\Gamma }^{\widehat{b}\widehat{a}}\overline{\widehat{D}}_{\widehat{a}}\overline{\widehat{ϵ}}_{(\frac{1}{2})}\right)$$
(D.22)
The overall factor is due to the dilatation transformation property of $`q_{(1)}`$. We will now show that the additive term is required by the invariance of the Witten-Nester energy under diffeomorphisms.
Weyl transformations on the boundary are induced by special bulk diffeomorphisms,
$$e^{\overline{r}+\sigma (\overline{x})}=e^r(\mathrm{\hspace{0.17em}1}+𝒪(e^{2r})),\overline{x}^i=x^i(\mathrm{\hspace{0.17em}1}+𝒪(e^{2r})).$$
(D.23)
The regulated Witten-Nester energy is invariant under this transformation provided we also transform the cut-off,
$$E_{NW}[r]=\overline{E}_{NW}[\overline{r}+\sigma (\overline{x})].$$
(D.24)
The normal to the surface $`\overline{r}+\sigma (\overline{x})`$ is given by $`l_\mu =𝒟_\mu (\overline{r}+\sigma (\overline{x}))`$. Inserting this in the definition of the Witten-Nester energy and considering the leading term in the limit $`\overline{r}\mathrm{}`$ we get
$$\overline{E}_{NW}[\overline{r}+\sigma (\overline{x})]d^{d1}x\sqrt{{}_{}{}^{t}\overline{g}_{(0)}^{}}e^{(d2)\overline{r}}(\overline{\widehat{q}}_{(1)}\overline{\widehat{ϵ}}_{(\frac{1}{2})}^{}\sigma _{,\widehat{b}}\mathrm{\Gamma }^{\widehat{b}\widehat{a}}\overline{\widehat{D}}_{\widehat{a}}\overline{\widehat{ϵ}}_{(\frac{1}{2})}+c.c.)$$
(D.25)
which agrees with the rhs of (D.22). The additive term is due to $`\sigma `$ dependence of the normal vector $`l_\mu `$.
## Appendix E Bulk Killing spinors and Witten spinors
We show in this appendix that for an AlAdS spacetime, the bulk Killing spinor $`ϵ`$ admits the asymptotic expansion
$$ϵ=ϵ_{(\frac{1}{2})}+ϵ_{(\frac{1}{2})}+\mathrm{}$$
(E.26)
with
$$\widehat{ϵ}_{(\frac{1}{2})}=P^{}\widehat{ϵ}_{(\frac{1}{2})},\widehat{ϵ}_{(\frac{1}{2})}=\mathrm{\Gamma }^i\widehat{D}_i\widehat{ϵ}_{(\frac{1}{2})}i,\mathrm{no}\mathrm{sum}\mathrm{over}i$$
(E.27)
Furthermore,
$$\xi ^i=\overline{\widehat{ϵ}}_{(\frac{1}{2})}\mathrm{\Gamma }^i\widehat{ϵ}_{(\frac{1}{2})}$$
(E.28)
is a timelike or null boundary conformal Killing vector.
Proof:
The asymptotic expansion of the covariant derivatives can be obtained from the results in appendix C. Starting from (C.14) we obtain
$`_rϵ`$ $`=`$ $`\left((\delta _D+{\displaystyle \frac{1}{2}}\mathrm{\Gamma }_{\widehat{r}})+_{(2)}+\mathrm{}\right)ϵ`$ (E.29)
$`_iϵ`$ $`=`$ $`(E_i^{\widehat{j}}\mathrm{\Gamma }_{\widehat{j}}P^++D_i+_{(2)i}+\mathrm{})ϵ.`$ (E.30)
We can now solve asymptotically the Killing spinor equations
$$_\mu ϵ=0.$$
(E.31)
The radial equation, $`_rϵ=0`$, implies that
$`ϵ`$ $`=`$ $`e^{\frac{1}{2}r}\widehat{ϵ}_{(\frac{1}{2})}+e^{\frac{1}{2}r}\widehat{ϵ}_{(\frac{1}{2})}+\mathrm{}`$ (E.32)
$`\widehat{ϵ}_{(\frac{1}{2})}`$ $`=`$ $`P^{}\widehat{ϵ}_{(\frac{1}{2})};\widehat{ϵ}_{(\frac{1}{2})}=P^+\widehat{ϵ}_{(\frac{1}{2})}`$ (E.33)
Inserting this in the spatial equations, $`_iϵ=0`$, we obtain
$$\widehat{D}_i\widehat{ϵ}_{(\frac{1}{2})}+\mathrm{\Gamma }_i\widehat{ϵ}_{(\frac{1}{2})}=0,\widehat{ϵ}_{(\frac{1}{2})}=\mathrm{\Gamma }^i\widehat{D}_i\widehat{ϵ}_{(\frac{1}{2})}i(\mathrm{no}\mathrm{sum})$$
(E.34)
where $`\widehat{D}_i`$ and $`\mathrm{\Gamma }_ie_i^{\widehat{j}}\mathrm{\Gamma }_{\widehat{j}}`$ are defined with respect to the boundary metric $`g_{(0)ij}`$.
Notice that an asymptotic Killing spinor is in particular a Witten spinor, but not vice versa. For instance, the sub-leading asymptotic coefficient of a Witten spinor is given by (5.45),
$$\widehat{ϵ}_{(\frac{1}{2})}=\frac{1}{d1}\mathrm{\Gamma }^{\widehat{a}}\widehat{D}_{\widehat{a}}\widehat{ϵ}_{(\frac{1}{2})},\mathrm{Witten}\mathrm{spinor}$$
(E.35)
Unless $`\mathrm{\Gamma }^{\widehat{a}}\widehat{D}_{\widehat{a}}\widehat{ϵ}_{(\frac{1}{2})}=\mathrm{\Gamma }^{\widehat{b}}\widehat{D}_{\widehat{b}}\widehat{ϵ}_{(\frac{1}{2})}`$, for all $`a`$ and $`b`$, the Witten spinor will not asymptote to a Killing spinor.
The fact that $`\xi ^i=\overline{\widehat{ϵ}}_{(\frac{1}{2})}\mathrm{\Gamma }^i\widehat{ϵ}_{(\frac{1}{2})}`$ is a conformal Killing vector follows by direct computation using the asymptotics of the Killing spinor,
$`\widehat{D}_{(i}\xi _{j)}`$ $`=`$ $`\widehat{D}_{(i}\left(\overline{\widehat{ϵ}}_{(\frac{1}{2})}\mathrm{\Gamma }_{j)}\widehat{ϵ}_{(\frac{1}{2})}\right)`$ (E.36)
$`=`$ $`\overline{\widehat{ϵ}}_{(\frac{1}{2})}\mathrm{\Gamma }_{(i}\mathrm{\Gamma }_{j)}\widehat{ϵ}_{(\frac{1}{2})}\overline{\widehat{ϵ}}_{(\frac{1}{2})}\mathrm{\Gamma }_{(j}\mathrm{\Gamma }_{i)}\widehat{ϵ}_{(\frac{1}{2})}`$
$`=`$ $`g_{(0)ij}(\overline{\widehat{ϵ}}_{(\frac{1}{2})}\widehat{ϵ}_{(\frac{1}{2})}\overline{\widehat{ϵ}}_{(\frac{1}{2})}\widehat{ϵ}_{(\frac{1}{2})})`$
$`=`$ $`{\displaystyle \frac{1}{d}}g_{(0)ij}\widehat{D}_k(\overline{\widehat{ϵ}}_{(\frac{1}{2})}\mathrm{\Gamma }^k\widehat{ϵ}_{(\frac{1}{2})})`$
$`=`$ $`{\displaystyle \frac{1}{d}}g_{(0)ij}\widehat{D}_k\xi ^k.`$
We now prove that $`\xi ^{\widehat{i}}=\overline{\widehat{ϵ}}_{(\frac{1}{2})}\mathrm{\Gamma }^{\widehat{i}}\widehat{ϵ}_{(\frac{1}{2})}`$ is timelike or null. Let us introduce the hermitian matrix
$$A=v_{\widehat{a}}\mathrm{\Gamma }_{\widehat{t}}\mathrm{\Gamma }^{\widehat{a}},v^2\eta ^{\widehat{a}\widehat{b}}v_{\widehat{a}}v_{\widehat{b}}=1,i=\{t,\{a\}\}.$$
(E.37)
and consider $`\widehat{ϵ}_{(\frac{1}{2})}`$ that is an eigenvector of $`A`$,
$$A\widehat{ϵ}_{(\frac{1}{2})}=a\widehat{ϵ}_{(\frac{1}{2})}.$$
(E.38)
Since $`A^2=1,a^2=1`$ too. Multiplying (E.38) by $`\widehat{ϵ}_{(\frac{1}{2})}^{}`$ and squaring we get
$$(\xi ^{\widehat{0}})^2=\left(\underset{\widehat{a}}{}v_{\widehat{a}}\xi ^{\widehat{a}}\right)^2.$$
(E.39)
Elementary algebra shows that
$$\underset{\widehat{a}<\widehat{b}}{}(\xi ^{\widehat{a}}v^{\widehat{b}}\xi ^{\widehat{b}}v^{\widehat{a}})^2=\underset{\widehat{a}}{}(\xi ^{\widehat{a}})^2\left(\underset{\widehat{a}}{}v_{\widehat{a}}\xi ^{\widehat{a}}\right)^20,$$
(E.40)
which implies
$$|\xi |^2(\xi ^{\widehat{0}})^2+\underset{\widehat{a}}{}(\xi ^{\widehat{a}})^20.$$
(E.41)
## Appendix F Killing Spinors of AdS<sub>d+1</sub> in global coordinates
We discuss in this appendix the structure of the Killing spinors for AdS<sub>d+1</sub> spacetimes in coordinates
$$ds^2=dr^2N_+^2(r)dt^2+N_{}^2(r)d\mathrm{\Omega }_{d1}^2,$$
(F.1)
where
$$N_\pm (r)=e^r\pm \frac{1}{4}e^r$$
(F.2)
and $`d\mathrm{\Omega }_{d1}^2`$ is the standard metric on $`S^{d1}`$,
$$d\mathrm{\Omega }_n^2=d\theta _n^2+\mathrm{sin}^2\theta _nd\mathrm{\Omega }_{n1}^2;d\mathrm{\Omega }_1^2=d\theta _1^2,$$
(F.3)
The radial coordinate $`\rho `$ usually used in the standard global coordinates is given by $`\rho =N_{}`$.
We find that the Killing spinors can be written in the following compact form
$$ϵ=e^{\frac{r}{2}}\widehat{ϵ}_{(\frac{1}{2})}+e^{\frac{r}{2}}\widehat{ϵ}_{(\frac{1}{2})},$$
(F.4)
where
$`\widehat{ϵ}_{(\frac{1}{2})}`$ $`=`$ $`P^{}𝒪_{d1}^+𝒪_{d2}\mathrm{}𝒪_1𝒪_t\eta `$ (F.5)
$`\widehat{ϵ}_{(\frac{1}{2})}`$ $`=`$ $`{\displaystyle \frac{1}{2}}P^+𝒪_{d1}^{}𝒪_{d2}\mathrm{}𝒪_1𝒪_t\eta ,`$
with $`\eta `$ a constant spinor and
$`𝒪_t`$ $`=`$ $`e^{\frac{t}{2}\mathrm{\Gamma }^{\widehat{t}}}=\mathrm{cos}{\displaystyle \frac{t}{2}}\mathrm{sin}{\displaystyle \frac{t}{2}}\mathrm{\Gamma }^{\widehat{t}}`$ (F.6)
$`𝒪_j`$ $`=`$ $`e^{\frac{\theta _j}{2}\mathrm{\Gamma }^{\widehat{j+1},\widehat{j}}}=\mathrm{cos}{\displaystyle \frac{\theta _j}{2}}+\mathrm{sin}{\displaystyle \frac{\theta _j}{2}}\mathrm{\Gamma }^{\widehat{j+1},\widehat{j}}j=1,..,d2`$
$`𝒪_{d1}^\pm `$ $`=`$ $`\mathrm{cos}{\displaystyle \frac{\theta _{d1}}{2}}\pm \mathrm{sin}{\displaystyle \frac{\theta _{d1}}{2}}\mathrm{\Gamma }^{\widehat{d1}}.`$
Proof
The covariant derivatives are given by,
$`_{\widehat{r}}`$ $`=`$ $`_r+{\displaystyle \frac{1}{2}}\mathrm{\Gamma }_{\widehat{r}}`$ (F.7)
$`_{\widehat{t}}`$ $`=`$ $`{\displaystyle \frac{1}{N_+}}(_t+e^r\mathrm{\Gamma }_{\widehat{t}}P^++{\displaystyle \frac{1}{4}}e^r\mathrm{\Gamma }_{\widehat{t}}P^{})`$
$`_{\widehat{\theta }_k}`$ $`=`$ $`{\displaystyle \frac{1}{N_{}}}(\widehat{D}_{\widehat{k}}+e^r\mathrm{\Gamma }_{\widehat{k}}P^+{\displaystyle \frac{1}{4}}e^r\mathrm{\Gamma }_{\widehat{k}}P^{}),k=1,\mathrm{},d1`$
where $`\widehat{D}_{\widehat{k}}=e_{\widehat{k}}^k\widehat{D}_k`$ denotes the covariant derivatives on the unit sphere. Explicitly,
$`\widehat{D}_k`$ $`=`$ $`{\displaystyle \frac{}{\theta _k}}+{\displaystyle \frac{1}{4}}\omega _i^{kl}\mathrm{\Gamma }_{kl}`$ (F.8)
$`\omega _i^{kl}`$ $`=`$ $`\delta _i^k\mathrm{cos}\theta _l{\displaystyle \underset{m=i+1}{\overset{l1}{}}}\mathrm{sin}\theta _m(k<l),`$
$`e_k^{\widehat{k}}`$ $`=`$ $`{\displaystyle \underset{m=k+1}{\overset{d1}{}}}\mathrm{sin}\theta _m.`$
Using these expressions, one can easily verify that (F.4) satisfy $`_{\widehat{r}}ϵ=_{\widehat{t}}ϵ=0`$, so we concentrate on the spherical part. From (F.7) we see that $`_{\widehat{\theta _k}}ϵ=0`$ is equivalent to the following equations,
$`\widehat{D}_{\widehat{k}}\widehat{ϵ}_{(\frac{1}{2})}+\mathrm{\Gamma }_{\widehat{k}}\widehat{ϵ}_{(\frac{1}{2})}`$ $`=`$ $`0,`$ (F.9)
$`\widehat{D}_{\widehat{k}}\widehat{ϵ}_{(\frac{1}{2})}{\displaystyle \frac{1}{4}}\mathrm{\Gamma }_{\widehat{k}}\widehat{ϵ}_{(\frac{1}{2})}`$ $`=`$ $`0.`$
These equations are easily shown to hold for $`k=d1`$, so in the following we discuss the cases $`k=1,2,..,d2`$. Our proof is similar in spirit with the discussion in .
We begin with the fact<sup>9</sup><sup>9</sup>9To avoid cumbersome notation we drop the ”hats” from the indices of the gamma matrices in the rest of this appendix.
$$_j𝒪_k=\delta _{jk}\frac{1}{2}\mathrm{\Gamma }^{j+1,j}𝒪_j,$$
(F.10)
which can be used to rewrite (F.9) as
$$𝒪_{d1}^\pm U_j^{d2}+\omega _j^{jl}\mathrm{\Gamma }_{jl}𝒪_{d1}^\pm e_j^{\widehat{j}}\mathrm{\Gamma }_j𝒪_{d1}^{}=0,$$
(F.11)
where $`U_j^k(kj)`$ is defined by
$$U_j^k=𝒪_kU_j^{k1}𝒪_k^1;U_j^j\mathrm{\Gamma }_{j+1,j}.$$
(F.12)
Equations (F.11) can further be rewritten as
$$𝒪_{d1}^\pm U_j^{d2}𝒪_{d1}^{}+\mathrm{cos}\theta _{d1}\omega _j^{jl}\mathrm{\Gamma }_{jl}e_j^{\widehat{j}}\mathrm{\Gamma }_j(1\mathrm{sin}\theta _{d1}\mathrm{\Gamma }_{d1})=0,$$
(F.13)
where we have used the relations
$`𝒪_{d1}^+𝒪_{d1}^{}`$ $`=`$ $`𝒪_{d1}^{}𝒪_{d1}^+=\mathrm{cos}\theta _{d1}`$ (F.14)
$`(𝒪_{d1}^\pm )^2`$ $`=`$ $`1\pm \mathrm{sin}\theta _{d1}\mathrm{\Gamma }_{d1}.`$
Using (F.8) and the relation
$$𝒪_{d1}^\pm \mathrm{\Gamma }_{j,d1}𝒪_{d1}^{}=\mathrm{\Gamma }_{j,d1}\mathrm{sin}\theta _{d1}\mathrm{\Gamma }_j,$$
(F.15)
one can prove (F.13) provided $`U_j^k`$ is given by
$$U_j^k=\underset{l>j}{\overset{k}{}}\omega _j^{jl}\mathrm{\Gamma }_{jl}\mathrm{sec}\theta _{k+1}\omega _j^{j,k+1}\mathrm{\Gamma }_{j,k+1}.$$
(F.16)
We now prove this relation by induction. First observe that $`U_j^j=\mathrm{\Gamma }_{j+1,j}`$ satisfies (F.16). Suppose now that (F.16) is satisfied for $`U_j^k`$ for some $`k<d2`$. Using
$$𝒪_{k+1}\mathrm{\Gamma }_{j,k+1}𝒪_{k+1}^1=\mathrm{cos}\theta _{k+1}\mathrm{\Gamma }_{j,k+1}+\mathrm{sin}\theta _{k+1}\mathrm{\Gamma }_{j,k+2},$$
(F.17)
one finds that $`U_j^{k+1}`$ satisfies (F.16) too. This finishes the proof of (F.16) and thus the proof that (F.4) is the Killing spinor of AdS<sub>d+1</sub>.
## Appendix G Asymptotics of AAdS spacetimes
We obtain in this appendix the coefficients $`K_{(2n)ij}[g_{(0)}],n<d`$ for AAdS. As mentioned in the main text, it is sufficient to compute them for the exact $`AdS`$ solution.
Consider $`AdS_{d+1}`$ with boundary metric $`g_{(0)}`$ the standard metric on $`R\times S^{d1}`$. Then from (2.3) we obtain,
$$g_{(2)ab}=\frac{1}{2}g_{(0)ab},g_{(2)tt}=\frac{1}{2}g_{(0)tt},g_{(4)ij}=\frac{1}{16}g_{(0)ij}.$$
(G.18)
The induced metric and second fundamental form are given by
$`\gamma _{ij}`$ $`=`$ $`e^{2r}(g_{(0)ij}+e^{2r}g_{(2)ij}+e^{4r}g_{(4)ij})`$
$`K_{ij}[\gamma ]`$ $`=`$ $`{\displaystyle \frac{1}{2}}\dot{\gamma }_{ij}=e^{2r}g_{(0)ij}e^{2r}g_{(4)ij}K_{(0)ij}[\gamma ]+\mathrm{}+K_{(2n)ij}[\gamma ]+\mathrm{}`$ (G.19)
Recall that the coefficient $`K_{(2n)ij}`$ are local polynomials of dimension $`n`$ of (covariant derivatives of the) curvature tensor of the induced metric. For the case at hand, this implies that $`K_{(2n)ij}`$ is proportional to $`R^n`$, where
$$R=(d1)(d2)\gamma ^1,\gamma =e^{2r}\left(1\frac{e^{2r}}{4}\right)^2$$
(G.20)
is the curvature scalar of $`\gamma _{ij}`$. Thus, the expansion in eigenfunctions of the dilation operator is equivalent to an expansion in $`\gamma ^1`$,
$$K_{ab}[\gamma ]=\gamma _{ab}\underset{n=0}{}\widehat{k}_{(2n)}\gamma ^{2n}$$
(G.21)
Inserting this expression in (G.19) yields,
$$\underset{n=0}{}\widehat{k}_{(2n)}e^{2nr}\left(1\frac{e^{2r}}{4}\right)^{2n}=(1+\frac{e^{2r}}{4})(1\frac{e^{2r}}{4})^1.$$
(G.22)
Expanding both sides around $`r\mathrm{}`$ and matching powers of $`e^{2r}`$ determines the coefficients $`\widehat{k}_{(2n)}`$. The first few are given in (8.73).
We next turn to the expansion of the Witten spinor. As mentioned in the main text, we take the Witten spinor to be a Killing spinor up to sufficiently high order, and the Killing spinor is given by
$$ϵ_K(x,r)=e^{\frac{r}{2}}\widehat{ϵ}_{(\frac{1}{2})}(x)+e^{\frac{r}{2}}\widehat{ϵ}_{(\frac{1}{2})}(x).$$
(G.23)
To obtain the eigenfunctions of the dilatation operator we should express $`ϵ_K(x,r)`$ as a series in $`\gamma ^1`$,
$$e_K(x,r)=\underset{m=0}{}\widehat{ϵ}_{(\frac{1}{2}+m)}(x)\gamma ^{\frac{1}{2}(\frac{1}{2}+m)}.$$
(G.24)
Comparing (G.23) and (G.24) determines $`\widehat{ϵ}_{(\frac{1}{2}+m)}(x)`$. The first few are given in (8.75).
|
warning/0506/math0506359.html
|
ar5iv
|
text
|
# Seidel’s Mirror Map for the Torus
## 1. Introduction
Paul Seidel had the following idea for recovering the mirror map purely from the Fukaya category.<sup>1</sup><sup>1</sup>1The idea described was told in a private communication to the author. This may have been implicit in the works of Fukaya and/or in the minds of others in the field. Start with a symplectic Calabi-Yau $`X`$ and its family of complex structures, and assume it has a projective mirror manifold $`Y`$ with a family of symplectic structures, and that Kontsevich’s conjecture holds: $`DFuk(X)D(Y),`$ where $`DFuk(X)`$ is the Fukaya category of $`X`$ (i.e. the bounded derived category constructed from the Fukaya $`A_{\mathrm{}}`$ category) and $`D(Y)`$ is the bounded derived category of coherent sheaves on $`Y`$. Then the homogeneous coordinate ring on $`Y`$ is given by
$$=\underset{k=0}{\overset{\mathrm{}}{}}\mathrm{\Gamma }(𝒪_Y(k))=\underset{k=0}{\overset{\mathrm{}}{}}\mathrm{Hom}_{DFuk(X)}(\psi (𝒪),\psi (𝒪(k))),$$
where $`\psi `$ is the equivalence of categories. The term on the right can be evaluated solely in $`DFuk(X)`$, and thus the complex projective variety $`Y`$ can be recovered. The dependence of this construction on the symplectic structure of $`X`$ defines the mirror map.<sup>2</sup><sup>2</sup>2The case of Fano varieties is being considered in .
Let $`S\psi (𝒪)`$ be the object dual to the structure sheaf of $`Y,`$ conjecturally the Lagrangian section of the Lagrangian torus fibration (cf. ). We will often equate a geometric Lagrangian submanifold with the object in $`DFuk`$ which it defines, including, if necessary, additional data such as grading and local system. Recall that on the complex structure moduli space of $`X,`$ monodromies act by symplectomorphisms, which define autoequivalences of $`DFuk(X)`$ (we use the same notation for a symplectomorphism and the autoequivalence it induces) and that the monodromy $`\rho `$ around the large complex structure limit point is mirror to the autoequivalence of $`D(Y)`$ defined by $`𝒪(1).`$ We define $`L_k`$ by $`L_k\rho ^kS.`$ Note $`S=L_0`$ and $`LL_1`$ is dual to $`𝒪(1).`$ In fact, $`L_k=\psi (𝒪(k)),`$ so we wish to compute $`_i\mathrm{Hom}_{DFuk(X)}(S,L_k).`$ In order to interpret this as a ring, we must identify $`\mathrm{Hom}(L_k,L_{k+l}))`$ with $`\mathrm{Hom}(S,L_l)`$ (we hereafter drop the $`DFuk(X)`$ subscript), and to do so we use the symplectomorphism $`\rho ^k.`$
In this note we will compute $``$ in the case where $`X`$ is a symplectic two-torus and derive the mirror map.<sup>3</sup><sup>3</sup>3The result is guaranteed to be correct here, since Kontsevich’s conjecture has been proven in this example . Without knowing the mirror map, we can still say that $`Y`$ is some elliptic curve, and thus has a projective embedding as a cubic curve. Then $`𝒪(1)`$ is a line bundle of degree three on $`Y,`$ so its mirror must have intersection three with $`S.`$ Taking the base section $`S`$ to be the $`x`$-axis in the universal cover $`^2,`$ we have that $`L`$ is a line of slope three. So we put $`\rho =\gamma ^3,`$ where $`\gamma `$ is a minimal Dehn twist, and note that $`\rho `$ is maximally unipotent. For simplicity we take $`S`$ (and therefore $`L`$) to have trivial local systems and to pass through lattice vectors, but our results do not depend on this choice. The data of $`S`$ and $`\rho `$ now allows us to calculate $`.`$
###### Acknowledgements.
I would like to thank Paul Seidel for communicating his ideas freely. Thanks to The Fields Institute for hosting me during this project. This work was supported in part by a Clay Senior Scholars fellowship and by NSF grant DMS–0405859.
## 2. Computation
We define $`X=^2/^2`$ with $`\omega =\tau dxdy,`$ $`\tau ,`$ $`Im(\tau )>0.`$ The category constructed from Fukaya’s $`A_{\mathrm{}}`$ category in this case was described explicitly in , and we refer the reader to those papers for details. As discussed above, we have $`L_k=\{(t,3kt)mod^2:t\},`$ and we define its grading $`\alpha =\mathrm{tan}^1(k)[0,\pi /2).`$ We define $`X_i=(i/3,0)\mathrm{Hom}(S,L),`$ $`Y_i=(i/6,0)\mathrm{Hom}(S,L_2),`$ and $`Z_i=(i/9,0)\mathrm{Hom}(S,L_3),`$ where $`i`$ is taken mod $`3,`$ $`6,`$ and $`9,`$ respectively. In the sequel, when we write an equation like $`X_1X_2=\mathrm{},`$ the $`X_2`$ is understood to live in $`\mathrm{Hom}(L_1,L_2)`$ through $`\rho .`$ Explicitly, $`\rho (x,y)=(x,y+3x);`$ indeed $`\rho ^{}\omega =\omega .`$
Let us compute the products $`X_iX_j.`$ The Fukaya category for this example was discussed in . The basic computation is $`X_0X_1.`$ The minimal triangle (holomorphic map) appearing in the product connects the points $`X_0=(0,0),`$ $`\rho (X_1)=X_1=(1/3,1),`$ and $`Y_1=(1/6,0)`$ and has symplectic area $`(1/2)(1/6)(1)\tau .`$ Multiples and translates of this triangle are relevant to other products. Multiples by $`6n`$ have the same endpoints and contribute to the same product, with area $`(1/2)(n+1/6)(6n+1)\tau .`$ The coefficient of $`Y_1`$ in $`X_0X_1`$ is thus $`A_1_n\mathrm{exp}[i\pi 6\tau (n+1/6)^2]=\theta [1/6,0](6\tau ,0).`$<sup>4</sup><sup>4</sup>4We recall the definition $`\theta [a,b](\tau ,z)=_n\mathrm{exp}[i\pi \tau (n+a)^2+2\pi i(n+a)(z+b)].`$ Defining $`A_k:=\theta [k/6,0](6\tau ,0),k/6,`$ and noting $`A_k=A_{6k},`$ we get the following relations:
(1)
$$X_iX_j=\underset{k=0}{\overset{1}{}}A_{ij+3k}Y_{i+j+3k}.$$
The right hand side of this equation makes sense with $`i,`$ $`j`$ defined mod $`3`$. Commutativity is easily shown to follow from the relations among the $`A_k.`$
Next we compute $`Y_iX_j.`$ Starting with $`Y_1X_1,`$ the minimal triangle has vertices $`Y_1=(1/6,0),`$ $`\rho ^2(X_1)=X_1=(1/3,1),`$ and $`Z_2=(2/9,0),`$ with area $`(1/2)(1/18)(1)\tau .`$ Odd multiples (with left endpoint fixed) and translates of this triangle are relevant to $`Y_iX_j`$ with $`i`$ odd; even multiples and translates to $`i`$ even. Multiples by $`18n`$ have the same endpoints. Therefore $`Y_1X_1=B_1Z_2+B_7Z_5+B_{13}Z_8,`$ where $`B_k=_n\mathrm{exp}[i\pi 18\tau (n+k/18)^2]=\theta [k/18,0](18\tau ,0).`$ Note $`B_k=B_{18k}`$ and $`k`$ is defined mod $`18.`$ As an example of another product, the third multiple of the minimal triangle has endpoints $`Y_1=(1/6,0),`$ $`X_2=(2/3,3),`$ $`Z_3=(1/3,0),`$ thus $`Y_1X_2=B_3Z_3+\mathrm{}.`$ Collecting results, we find
(2)
$$Y_iX_j=\underset{k=0}{\overset{2}{}}B_{2ji+6k}Z_{i+j+3k}.$$
## 3. Commutativity and Associativity
Associativity in the (derived or cohomological) Fukaya category follows from general grounds, and in the case of the torus amounts to an equality obtained from expressing the area of a non-convex quadrangle by splitting it into triangles in two different ways. (This was noted, for example, in Section 2 of .) It also amounts to relations among the $`A_k`$ and $`B_k,`$ which we describe presently.
As for commutativity, this follows from the existence of a robust family of anti-symplectomorphisms. For example, in considering the products $`X_0Y_k,`$ one must count (among other things) triangles with vertices $`X_0,`$ $`\rho (Y_k),`$ and $`Z_k`$ arranged in clockwise orientation and with sides of appropriate slope. Now consider the map $`\phi :`$
$$(x,y)(\frac{1}{2}x\frac{7}{18}y+\frac{1}{9}k,2y).$$
We note $`\phi (X_0)=Z_k,`$ $`\phi (\rho (Y_k))=X_0=\rho ^2(X_0),`$ and $`\phi (Z_k)=Y_k.`$ Further, since $`\phi `$ is an anti-symplectomorphism, i.e. $`\phi ^{}\omega =\omega ,`$ it preserves areas and reverses the orientation and thus changes the order in which the vertices appear on the outside of the triangle. Thus $`Y_k,\rho ^2(X_0),Z_k`$ are oriented clockwise in the image triangle, which has the same area as the original. This proves commutativity among products $`X_0Y_k.`$ Translations of $`\phi `$ suffice for proving commutativity for $`X_jY_k.`$ Products $`X_iX_j`$ were already seen to be commutative, and this is all that we will require for our purposes. In short, commutativity follows from anti-symplectomorphisms mapping vertices $`(X,\rho ^nY,Z)`$ to $`(Z,\rho ^mX,Y)`$ in holomorphic triangles. It is not clear (to the author) why commutativity should hold in a general symplectic manifold.
We now return to an explicit description of the associativity constraint. We will make use of the following identity, which follows from the addition formula II.6.4 of :
(3) $`\theta [{\displaystyle \frac{a}{n}},0](n\tau ,0)\theta [{\displaystyle \frac{b}{nk}},0](nk\tau ,0)=`$
$`{\displaystyle \underset{ϵ=0}{\overset{k}{}}}\theta [{\displaystyle \frac{bka+knϵ}{k(k+1)n}},0](k(k+1)n\tau ,0)\theta [{\displaystyle \frac{a+b+knϵ}{(k+1)n}},0]((k+1)n\tau ,0).`$
When $`n=6`$ and $`k=3`$ this gives us formulas for $`A_aB_b.`$ Defining $`C_c=\theta [c/24,0](24\tau )`$ and $`D_d=\theta [d/72,0](72\tau ),`$ we have
(4)
$$A_aB_b=\underset{ϵ=0}{\overset{3}{}}C_{a+b+18ϵ}D_{b3a+18ϵ}.$$
This formula suffices for proving some of the equivalences necessary for showing associativity. Others follow from further application of (3).
For example, one wants to show that $`(X_0^2)X_1=X_0(X_0X_1).`$ This amounts to $`(A_0Y_0+A_3Y_3)X_1=X_0(A_1Y_1+A_2Y_4).`$ Using commutativity and the products (1), then equating coefficients on $`Z_k,`$ gives the conditions
$`A_0B_2+A_3B_7`$ $`=`$ $`A_1B_1+A_2B_8,`$
$`A_0B_8+A_3B_1`$ $`=`$ $`A_1B_5+A_2B_4,`$
$`A_0B_4+A_3B_5`$ $`=`$ $`A_1B_7+A_2B_2.`$
The first and third relations follow immediately from (4). The second equation is most easily seen by rewriting the right hand side as $`A_1B_5+A_2B_4.`$ Proceeding in this manner, one can prove well-definedness of $`X_iX_jX_k.`$
Again, associativity follows from quadrilateral dissection, or on general grounds for the Fukaya category, and our philosophy here should be to think of these identities as following from the associativity constraints. In either case, we will use the explicit expressions derived here.
## 4. Relations
One finds that the number of degree two polynomials in the three variables $`X_i`$ equals exactly the number of $`Y_k,`$ and in fact since $`A_0A_1A_2A_30`$ one finds that the $`Y_k`$ can be written in terms of products $`X_iX_j,`$ and vice versa, so there are no relations in $``$ at this degree. At the next level, we have ten independent polynomials and nine $`Z_k,`$ so we expect a single relation. Let us search for this relation.
Let
$$\{X_0^3,X_1^3,X_2^3,X_0^2X_1,X_1^2X_2,X_2^2X_0,X_0^2X_2,X_1^2X_0,X_2^2X_1,X_0X_1X_2\}$$
be a basis, with $`e^I`$ the $`I`$-th entry, $`I=0\mathrm{}9.`$ Using the product, we can write $`e^I=_kM_k{}_{}{}^{I}Z_{k}^{}.`$ A relation $`a`$ has the form $`_Ia_Ie^I=0,`$ or $`_k\left(_I(M_k{}_{}{}^{I}a_{I}^{})\right)Z_k=0.`$ Since the $`Z_k`$ are linearly independent generators of $`\mathrm{Hom}(S,L_3)`$ we have, in matrix form $`Ma=0,`$ or $`a\mathrm{Ker}(M).`$ $`M`$ is a $`9\times 10`$ matrix, so the kernel should be one-dimensional, and we can take $`a_I=c(1)^I\mathrm{det}(M_I),`$ where $`M_I`$ is $`M`$ with the $`I`$-th column removed and $`c0`$ is any constant.
Using the products found in Section 2, one finds
$$M=\left(\begin{array}{cccccccccc}p& q& q& 0& 0& 0& 0& 0& 0& u\\ 0& 0& 0& r& t& s& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& t& r& s& 0\\ q& p& q& 0& 0& 0& 0& 0& 0& v\\ 0& 0& 0& s& r& t& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& s& t& r& 0\\ q& q& p& 0& 0& 0& 0& 0& 0& v\\ 0& 0& 0& t& s& r& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& r& s& t& 0\end{array}\right),$$
where
$$\begin{array}{ccc}p=A_0B_0+A_3B_9\hfill & r=A_0B_2+A_3B_7\hfill & u=A_2B_0+A_1B_9\hfill \\ q=A_0B_6+A_3B_3\hfill & s=A_0B_8+A_3B_1\hfill & v=A_2B_6+A_1B_3\hfill \\ & t=A_0B_4+A_3B_5.\hfill & \end{array}$$
Up to a common multiple, one finds $`a((p+q)u2qv,pvqu,pvqu,0,0,0,0,0,0,2q^2pqp^2).`$ In fact, $`u=v`$, which follows from associativity, or equivalently the relations (3), so we can remove the common (nonzero) factor of $`pq`$ and take
$$a=(u,u,u,0,0,0,0,0,0,2qp).$$
If there are no other relations in the ring $`,`$ then this single relation defines a cubic curve in the Hesse family as
$$a_0X_0^3+a_1X_1^3+a_2X_2^3+a_9X_0X_1X_2=0.$$
The modular invariant is easily calculated in terms of $`z=(1/3)a_9(a_0a_1a_2)^{1/3}=\frac{2q+p}{3u}.`$ Explicitly,
(5)
$$j(\tau )=27z^3(z^3+8)^3(1z^3)^3.$$
This equation, which should define the $`j`$-function of the mirror curve, is written in terms of the symplectic parameter $`\tau `$ on the torus. It therefore defines the mirror map, which in this example is known to send the symplectic parameter $`\tau `$ to the modular parameter $`\tau `$ in the upper halfplane. So (5) amounts to an identity in terms of the variable $`\tau ,`$ or more conveniently for us, $`x=e^{i\pi \tau /18},`$ and it remains to verify this relation.<sup>5</sup><sup>5</sup>5We ignore the possibility of further relations in $`.`$ This assumption is justified using the mirror equivalence, but would be difficult to show working purely from the Fukaya side.
The following identities follow directly from the definitions:
$`A_k`$ $`=`$ $`x^{3k^2}+{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}x^{3(6n+k)^2}+x^{3(6nk)^2},`$
$`B_k`$ $`=`$ $`x^{k^2}+{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}x^{(18n+k)^2}+x^{(18nk)^2}.`$
Recall that the $`j`$-invariant has the expansion
$$j(x)=x^{36}+744+196884x^{36}+21493760x^{72}+864299970x^{108}+\mathrm{}.$$
These coefficients and more can be corroborated order by order in the series expansion of the right hand side of (5). A more general proof may be found in . Of course, this had to be true, by the equivalence of categories already proven in , but our intent was to find this result working only from the Fukaya category.<sup>6</sup><sup>6</sup>6Perhaps one could invert this philosophy and derive information about the Fukaya category from the known mirror maps, in cases where computing products is formidable. We find the computation a pleasant realization of Seidel’s idea.
Eric Zaslow, Department of Mathematics, Northwestern University, Evanston, IL 60208.
(zaslow@math.northwestern.edu)
|
warning/0506/cond-mat0506632.html
|
ar5iv
|
text
|
# First-principle Wannier functions and effective lattice fermion models for narrow-band compounds
## I Introduction
Many successes of modern solid-state physics and chemistry are related with the development of the Hohenberg-Kohn-Sham density-functional theory (DFT),HohenbergKohn ; KohnSham which is designed for the ground state and based on the minimization of the total energy $`E[\rho ]`$ with respect to the electron density $`\rho `$. For practical applications, DFT resorts to iterative solution of single-particle Kohn-Sham (KS) equations,
$$\left(\frac{\mathrm{}^2}{2m}^2+V_\mathrm{H}+V_{\mathrm{XC}}+V_{\mathrm{ext}}\right)\psi _i=\epsilon _i\psi _i,$$
(1)
together with the equation for the electron density:
$$\rho =\underset{i}{}n_i|\psi _i|^2,$$
(2)
defined in terms of eigenfunctions ($`\psi _i`$), eigenvalues ($`\epsilon _i`$), and the occupation numbers ($`n_i`$) of KS quasiparticles. Different terms in Eq.(1) are correspondingly the kinetic-energy operator, the Hartree potential, the exchange-correlation potential, and the external potential. In the following we will also reserve the notation $`H_{\mathrm{KS}}`$$`=`$$`(\mathrm{}^2/2m)^2`$$`+`$$`V_\mathrm{H}`$$`+`$$`V_{\mathrm{XC}}`$$`+`$$`V_{\mathrm{ext}}`$ for the total KS Hamiltonian in the real ($`𝐫`$) space.
The exchange-correlation potential is typically treated in the local-density approximation (LDA). It employs an analytical expression for $`V_{\mathrm{XC}}[\rho ]`$ borrowed from the theory of homogeneous electron gas in which the density of the electron gas is replaced by the local density of the real system. LDA is far from being perfect and there are many examples of so-called strongly-correlated materials for which the conventional LDA appears to be insufficient both for the excited-state and ground-state properties.AZA ; LDAUreview ; IFT
A typical situation realized in transition-metal (TM) oxides is shown in Fig. 1.
We would like to emphasize two points.
(i) The common feature of many TM oxides is the existence of the well isolated narrow band (or the group of bands) located near the Fermi level and well isolated from the rest of electronic stares. For compounds shown in Fig. 1, this is the TM $`t_{2g}`$ band, which is sandwiched between O($`2p`$) band (from below) and a group of bands (from above), which have an appreciable weight of the TM $`e_g`$ states (the meaning of many notations will become clear in Sec. VI, where we will discuss details of the crystal and electronic structure for the considered oxide compounds). Electronic and magnetic properties of these compounds are predetermined mainly by the behavior of this $`t_{2g}`$ band. The effect of other bands can be included indirectly, through the renormalization of interaction parameters in the $`t_{2g}`$ band.
(ii) The LDA description appears to be especially bad for the $`t_{2g}`$ states located near the Fermi level. It often fails to reproduce the insulating behavior of these compounds, as well as the correct magnetic ground state, which is directly related with the existence of the band gap.TRN The source of the problem is know to be the on-site Coulomb correlations, whose form is greatly oversimplified in the model of homogeneous electron gas. Therefore, the basic strategy which was intensively pursued already for more than decade was to incorporate the physics of on-site Coulomb correlations in LDA and to solve this problem using modern many-body techniques. This way of thinking gave rise to such directions as LDA$`+`$$`U`$ (e.g., Refs. AZA, , LDAUreview, , and PRB94, ) and LDA$`+`$DMFT (dynamical mean-field theory, Ref. LSDADMFT, ).
Taking into account two above arguments, we believe that the most logical way to approach the problem of Coulomb correlations in narrow-band compounds is to divide it in two part:
(i) mapping of conventional electronic structure calculations onto the multi-orbital Hubbard model, and derivation of the parameters of this model from the first principles, for example starting from the simplest electronic structure in LDA;
(ii) solution of this multi-orbital Hubbard model using modern many-body methods.Sr2VO4preprint
In this paper we will discuss the first part of this project and show how results of conventional LDA calculations for the $`t_{2g}`$ bands can be mapped onto the multi-orbital Hubbard model:
$$\widehat{}=\underset{\mathrm{𝐑𝐑}^{}}{}\underset{\alpha \beta }{}h_{\mathrm{𝐑𝐑}^{}}^{\alpha \beta }\widehat{c}_{𝐑\alpha }^{}\widehat{c}_{𝐑^{}\beta }^{}+\frac{1}{2}\underset{𝐑}{}\underset{\alpha \beta \gamma \delta }{}U_{\alpha \beta \gamma \delta }\widehat{c}_{𝐑\alpha }^{}\widehat{c}_{𝐑\gamma }^{}\widehat{c}_{𝐑\beta }^{}\widehat{c}_{𝐑\delta }^{},$$
(3)
where $`\widehat{c}_{𝐑\alpha }^{}`$ ($`\widehat{c}_{𝐑\alpha }`$) creates (annihilates) an electron in the Wannier orbital $`\stackrel{~}{W}_{𝐑\alpha }`$ of the site $`𝐑`$, and $`\alpha `$ is a joint index, incorporating all remaining (spin and orbital) degrees of freedom. The matrix $`h_{\mathrm{𝐑𝐑}^{}}^{\alpha \beta }`$ parameterizes the kinetic energy of electrons. The matrix elements $`h_{\mathrm{𝐑𝐑}^{}}^{\alpha \beta }`$ have the following meaning: the site-diagonal part ($`𝐑`$$`=`$$`𝐑^{}`$) describes the local level-splitting, caused by the crystal field and (or) the spin-orbit interaction, while the off-diagonal part ($`𝐑`$$``$$`𝐑^{}`$) stands for the transfer integrals (or the transfer interactions). $`U_{\alpha \beta \gamma \delta }`$$`=`$$`d𝐫d𝐫^{}\stackrel{~}{W}_\alpha ^{}(𝐫)\stackrel{~}{W}_\beta (𝐫)v_{\mathrm{scr}}(𝐫`$$``$$`𝐫^{})\stackrel{~}{W}_\gamma ^{}(𝐫^{})\stackrel{~}{W}_\delta (𝐫^{})`$ are the matrix elements of screened Coulomb interaction $`v_{\mathrm{scr}}(𝐫`$$``$$`𝐫^{})`$, which are supposed to be diagonal with respect to the site indices. In principle, the off-diagonal elements can be also included into the model. However, we do not consider them in the present work. In Sec. VIII we will discuss several open questions related with the definition of the intersite Coulomb interactions in LDA.
The first part of this paper will be devoted to derivation of the parameters of the kinetic energy. Then, we will explain how to construct the Wannier functions (WFs), which generate these parameters after applying to the KS Hamiltonian in the real space. The next part will be devoted to calculations of screened Coulomb interactions, using the WF formalism.
## II The LMTO Method
In this section we will briefly review the main ideas of linear-muffin-tin-orbital (LMTO) method, as they will be widely used in the subsequent sections for the construction of transfer interactions and the Wannier orbitals. The method is designed for the solution of KS equations in LDA. For the details and recent developments, the reader is referred to the activity of the O. K. Andersen group at the Max-Plank Institute in Stuttgart.LMTO
Majority of modern electronic structure methods use some basis. The basis functions of the LMTO method, $`\{|\chi \}`$ (the so-called muffin-tin orbitals – MTOs) have many similarities with orthogonalized atomic orbitals. As we will see below, the LMTO method is very convenient for constructing the WFs, and for certain applications, the basis function of the LMTO method, from the very beginning, can be chosen as a Wannier function.
The conventional LMTO approach employs the atomic-spheres-approximation (ASA), which assumes that the whole space of the crystal can be filled by overlapping atomic spheres (Fig. 2), so that the overlap between the spheres as well as the empty spaces, which are not encircled by any spheres, can be neglected. The MTOs are constructed from solutions of KS equations inside atomic spheres (the partial waves), calculated at some energies $`E_{\nu L}`$ (typically, the center of gravity of the occupied band or of the entire band), $`\varphi _{𝐑L}`$, and their energy derivatives $`\dot{\varphi }_{𝐑L}`$. In each atomic sphere, the KS potential is spherically averaged. Therefore, the solutions are proportional to the angular harmonics, which are specified by the indices $`L`$$``$$`(\mathrm{},m)`$ (correspondingly, orbital and azimuthal quantum numbers) . At the atomic sphere boundaries, $`\{\varphi _{𝐑L}\}`$ and $`\{\dot{\varphi }_{𝐑L}\}`$ match continuously and differenciably onto certain envelop functions. The latter are typically constructed from irregular solutions of Laplace equation, which rapidly decay in the real space (Fig. 2).
It is easy to verify that the functions $`\varphi _{𝐑L}`$ and $`\dot{\varphi }_{𝐑L}`$ obey the following “LMTO algebra”:
$$\varphi _{𝐑L}|\varphi _{𝐑L}=1,$$
(4)
$$\dot{\varphi }_{𝐑L}|\varphi _{𝐑L}=\varphi _{𝐑L}|\dot{\varphi }_{𝐑L}=0,$$
(5)
$$\dot{\varphi }_{𝐑L}|\dot{\varphi }_{𝐑L}=p_{𝐑L},$$
(6)
$$\left(H_{\mathrm{KS}}E_{\nu L}\right)|\varphi _{𝐑L}=0,$$
(7)
and
$$\left(H_{\mathrm{KS}}E_{\nu L}\right)|\dot{\varphi }_{𝐑L}=|\varphi _{𝐑L},$$
(8)
where $`H_{\mathrm{KS}}`$$`=`$$`H_{\mathrm{KS}}(𝐫)`$ is the Kohn-Sham Hamiltonian in the real space. Then, one possible choice of the muffin-tin orbitals is
$$|\chi =|\varphi +|\dot{\varphi }(\widehat{}\widehat{E}_\nu ).$$
(9)
In this paper we use the shorthanded notations by Andersen et al.. For instance, Eq. (9) should be read as follows:
$$|\chi _{𝐑L}=|\varphi _{𝐑L}+\underset{𝐑^{}L^{}}{}|\dot{\varphi }_{𝐑^{}L^{}}(_{𝐑^{}𝐑}^{L^{}L}\delta _{𝐑^{}𝐑}\delta _{L^{}L}E_{\nu L}).$$
The first and second terms in the right-hand side of this equation are sometimes called, correspondingly, the “head” and the “tail” of MTO. The angular character of MTO at the central site $`𝐑`$ is specified by that of partial wave $`\varphi _{𝐑L}`$. The matrix elements of $`H_{\mathrm{KS}}`$ in the MTOs basis (9) can be immediately derived by using the properties (4)-(8) of $`\varphi _{𝐑L}`$ and $`\dot{\varphi }_{𝐑L}`$:
$$\chi |H_{\mathrm{KS}}|\chi =\widehat{}+(\widehat{}\widehat{E}_\nu )\widehat{E}_\nu \widehat{p}(\widehat{}\widehat{E}_\nu ),$$
where $`\widehat{E}_\nu `$ and $`\widehat{p}`$ are the diagonal matrices constructed from $`\{\widehat{E}_{\nu L}\}`$ and $`\{p_{𝐑L}\}`$, and $`\chi |H_{\mathrm{KS}}|\chi `$ is the shorthanded notation for the matrix $`\chi _{𝐑L}|H_{\mathrm{KS}}|\chi _{𝐑^{}L^{}}`$. The corresponding overlap matrix, $`\chi _{𝐑L}|\chi _{𝐑^{}L^{}}`$, is
$$\chi |\chi =\widehat{1}+(\widehat{}\widehat{E}_\nu )\widehat{p}(\widehat{}\widehat{E}_\nu ).$$
Since the second term in the right-hand side of $`\chi |\chi `$ is typically small, the basis functions (9) are said to form a nearly orthogonal representation of the LMTO method, and $`\widehat{}`$ is the LMTO Hamiltonian in the second order of $`(\widehat{}`$$``$$`\widehat{E}_\nu )`$. The basis (9) can be orthonormalized numerically, by applying the transformation
$$|\chi |\stackrel{~}{\chi }=|\chi \chi |\chi ^{1/2}.$$
The corresponding LMTO Hamiltonian, $`\widehat{H}`$$`=`$$`\stackrel{~}{\chi }|H_{\mathrm{KS}}|\stackrel{~}{\chi }`$, which is formally valid in all orders of $`(\widehat{}`$$``$$`\widehat{E}_\nu )`$, is given by
$$\widehat{H}=\chi |\chi ^{1/2}\chi |H_{\mathrm{KS}}|\chi \chi |\chi ^{1/2}.$$
This Hamiltonian will be used as the starting point in the next section, for the definition of transfer interactions between certain Wannier orbitals. We start with the formal description of the downfolding method. The construction of the Wannier basis functions, underlying this approach, will be considered in Sec. IV, where we will use again all merits of the LMTO method and show that proper WFs can be constructed by retaining the “heads” of MTOs and attaching to them different “tails”.
For periodic crystals, it is convenient to work in the reciprocal ($`𝐤`$) space. Therefore, if it is not specified otherwise, we assume that the LMTO Hamiltonian is already constructed in the reciprocal space, after the Fourier transformation of MTOs:
$$|\chi _{𝐤L}=\frac{1}{\sqrt{N}}\underset{𝐑}{}e^{i\mathrm{𝐤𝐑}}|\chi _{𝐑L},$$
where $`N`$ is the number of sites.
For all considered compounds, results of our ASA-LMTO calculations are in a good agreement the ones obtained using more accurate full-potential methods. In our definition of the crystal-field splitting we go beyond the conventional ASA and take into account nonsphericity of the electron-ion interactions (see Sec. V).
## III Downfolding Method for the Kinetic-Energy part
Parameters of the kinetic energy are obtained using the downfolding method, starting from the electronic structure in LDA. In order to describe properly the electronic structure of the TM oxides in the valent part of the spectrum using the LMTO method, it is typically required several tens or even hundreds basis functions (including the ones associated with empty spheres, which are added in order to improve the atomic spheres approximation for loosely packed atomic structures). Several examples of such bases will given in Secs. VI.1-VI.4.
What we want to do next is to describe some part of this electronic structure by certain tight-binding (TB) Hamiltonian $`\widehat{h}`$, which, contrary $`\widehat{H}`$, is formulated in the basis of a very limited number of orthogonal atomic-like orbitals. For example, in order to reproduce the $`t_{2g}`$ bands located near the Fermi level, one would like to use only three $`t_{2g}`$ orbitals centered at each TM site. These orbitals have a meaning of Wannier orbitals, which will be considered in Sec. IV.
We start with the identity by noting that any eigenstate of the LMTO Hamiltonian $`\widehat{H}`$ can be presented as the sum $`|\psi `$$`=`$$`|\psi _t`$$`+`$$`|\psi _r`$, where $`|\psi _t`$ is expanded over the LMTO basis function of the $`t_{2g}`$-type, $`\{|\stackrel{~}{\chi }_t\}`$ (here, the character of the basis function is specified by its “head”), and $`|\psi _r`$ is expanded over the rest of the basis functions $`\{|\stackrel{~}{\chi }_r\}`$. Then, the matrix equations for LMTO eigenstates can be rearranged identically as
$`(\widehat{H}^{tt}\omega )|\psi _t+\widehat{H}^{tr}|\psi _r`$ $`=`$ $`0,`$ (10)
$`\widehat{H}^{rt}|\psi _t+(\widehat{H}^{rr}\omega )|\psi _r`$ $`=`$ $`0.`$ (11)
By eliminating $`|\psi _r`$ from Eq. (11) one obtains the effective $`\omega `$-dependent Hamiltonian in the basis of $`t_{2g}`$-states
$$\widehat{H}_{\mathrm{eff}}^{tt}(\omega )=\widehat{H}^{tt}\widehat{H}^{tr}(\widehat{H}^{rr}\omega )^1\widehat{H}^{rt}$$
and the “overlap” matrix
$$\widehat{S}(\omega )=1+\widehat{H}^{tr}(\widehat{H}^{rr}\omega )^2\widehat{H}^{rt},$$
satisfying the condition $`\psi _t|\widehat{S}|\psi _t`$$`=`$$`1`$. Then, the required TB Hamiltonian, $`\widehat{h}`$, is obtained after the orthonormalization of the vectors $`|\psi _t`$$``$$`|\stackrel{~}{\psi }_t`$$`=`$$`\widehat{S}^{1/2}|\psi _t`$ and fixing the energy $`\omega `$ in the center of gravity of the $`t_{2g}`$ band ($`\omega _0`$):
$$\widehat{h}=\widehat{S}^{1/2}(\omega _0)\widehat{H}_{\mathrm{eff}}^{tt}(\omega _0)\widehat{S}^{1/2}(\omega _0).$$
(12)
Typically, the downfolding is performed in the reciprocal space, and the parameter $`\omega _0`$ may also depend on $`𝐤`$. The Hamiltonian $`\widehat{h}_𝐤`$ can be Fourier transformed back to the real space:
$$\widehat{h}_{\mathrm{𝐑𝐑}^{}}=\underset{𝐤}{}e^{i𝐤(𝐑𝐑^{})}\widehat{h}_𝐤.$$
The site-diagonal part of $`\widehat{h}_{\mathrm{𝐑𝐑}^{}}`$ shall describe the crystal-field (CF) splitting caused by the lattice distortion and associated with the transfer interactions between $`t`$\- and $`r`$-orbitals, which are eliminated in the downfolding method, while the off-diagonal elements have a meaning of transfer interactions. The first application of this approach has been considered in Ref. PRB04, . In Sec. VI we will illustrate abilities of this method for several types of $`t_{2g}`$ compounds.
The crystal-field splitting may have another origin, which is related with nonsphericity of the electron-ion interactions.MochizukiImada This contribution will be considered in Sec. V.
## IV Wannier functions
In the previous section we have shown that there is a TB Hamiltonian, $`\widehat{h}`$, formulated in some basis of Wannier orbitals $`\{\stackrel{~}{W}\}`$. It allows to generate the electronic structure of isolated $`t_{2g}`$ bands, which is practically identical to the electronic structure obtained after the diagonalization of the total LMTO Hamiltonian $`\widehat{H}`$.
In this section we will solve an inverse problem and construct the basis $`\{\stackrel{~}{W}\}`$, which after applying to the original KS Hamiltonian, generates the matrix $`\widehat{h}`$:
$$\widehat{h}=\stackrel{~}{W}|H_{\mathrm{KS}}|\stackrel{~}{W}.$$
(13)
In a close analogy with the LMTO method, we will first introduce the orbitals $`\{W\}`$, which are related with $`\{\stackrel{~}{W}\}`$ by the orthonormalization transformation
$$|\stackrel{~}{W}=|WW|W^{1/2},$$
(14)
and search $`|W`$ in the form:
$$|W=|W_t+\underset{r=1}{\overset{N_r}{}}\mathrm{\Gamma }_r|W_r,$$
(15)
where $`W_t`$ is constructed entirely from the $`t_{2g}`$-type solutions of KS equations inside atomic spheres and their energy-derivatives $`\{\varphi _t,\dot{\varphi }_t\}`$: i.e. both $`\varphi _t`$ and $`\dot{\varphi }_t`$ belong to the TM sites, and ‘t’ stands for the $`3d`$-$`t_{2g}`$ partial waves. Each $`W_r`$ is constructed from the rest of the partial waves $`\{\varphi _r,\dot{\varphi }_r\}`$. Then, $`W_t`$ and $`W_r`$ can be found from the following conditions:
1. We request only the $`t`$-part of $`|W`$ to contribute to the matrix elements of the KS Hamiltonian, and search it in the form of MTO:
$$|W_t=|\varphi _t+|\dot{\varphi }_t(\widehat{𝔥}\widehat{E}_{\nu t}).$$
(16)
In this definition, $`|W_t`$ is a function of (yet unknown) matrix $`\widehat{𝔥}`$, which will be found later. The matrix elements of $`H_{\mathrm{KS}}`$ in the basis of these Wannier orbitals are given by:
$$W|H_{\mathrm{KS}}|W=\widehat{𝔥}+(\widehat{𝔥}\widehat{E}_{\nu t})\widehat{E}_{\nu t}\widehat{p}_t(\widehat{𝔥}\widehat{E}_{\nu t}).$$
(17)
2. The $`r`$-parts of the WF, $`\{W_r\}`$, do not contribute to the matrix elements (17). They are introduced only in order to make the WFs (15) orthogonal to the rest of the eigenstates of the Hamiltonian $`\widehat{H}`$. Therefore, we search $`|W_r`$ in the form:
$$|W_r=|\varphi _r+\alpha _r|\dot{\varphi }_r,$$
where
$$\alpha _r=\frac{2E_{\nu r}}{1+\sqrt{14E_{\nu r}^2p_r}}$$
is obtained from the condition $`W_r|H_{\mathrm{KS}}|W_r`$$`=`$$`0`$.
3. The coefficients $`\mathrm{\Gamma }_r`$ are found from the orthogonality condition of $`|W`$ to $`N_r`$ eigenstates $`\{|\psi _i\}`$ of the original LMTO Hamiltonian $`\widehat{H}`$:
$$\underset{r=1}{\overset{N_r}{}}\psi _i|W_r\mathrm{\Gamma }_r=\psi _i|W_t,i=1,\mathrm{},N_r.$$
(18)
This allows to include the $`r`$-components of the WFs in a systematic way. For example, by taking into consideration the $`2p`$-partial waves inside oxygen spheres ($`N_r`$$`=`$$`9`$ wavefunctions for cubic perovskites), the WFs can be orthogonalized to $`N_r`$$`=`$$`9`$ O($`2p`$) bands, etc. For a given $`\widehat{𝔥}`$, the problem is reduced to the solution of the system of linear equations (18).
Since $`|W_t`$ contributes to Eq. (18), the coefficients $`\{\mathrm{\Gamma }_r\}`$ will also depend on $`\widehat{𝔥}`$. Therefore, the total WF is an implicit function of the matrix $`\widehat{𝔥}`$: $`|W`$$``$$`|W(\widehat{𝔥})`$.
4. The last step is the orthonormalization (14), which after substitution into Eq. (13) yields the following equation for the matrix $`\widehat{𝔥}`$:
$$\widehat{𝔥}=W(\widehat{𝔥})|W(\widehat{𝔥})^{1/2}\widehat{h}W(\widehat{𝔥})|W(\widehat{𝔥})^{1/2}(\widehat{𝔥}\widehat{E}_{\nu t})\widehat{E}_{\nu t}\widehat{p}_t(\widehat{𝔥}\widehat{E}_{\nu t}).$$
This equation is solved iteratively with respect to $`\widehat{𝔥}`$.
### IV.1 Spacial extension of Wannier functions
The choice of the WFs as well as their extension in the real space is not uniquely defined. For many practical applications one would like to have “maximally localized” orbitals,MarzariVanderbilt although in the context of the WFs, the term “maximally localized” itself bears certain arbitrariness and is merely a mathematical construction, because depending on the considered physical property one can introduce different criteria of the “maximal localization”.
Although we do not explicitly employ here any procedure which would pick up the “most localized” representation for the WFs, our method well suits this general strategy and the obtained WFs are expected to be well localized around the central TM sites.
There are several ways of controlling the spacial extension of the WFs in the LMTO method.
1. By using different envelop functions one can, in principle, change the spacial extension of MTOs (Fig. 2), which controls the decay of the original LMTO Hamiltonian in the real space. For example, instead of irregular solutions of Laplace equation, one can use Hankel functions of the complex argument. However, any choice should satisfy certain criteria of the completeness of the basis set. From this point of view, the use of the Hankel functions is not well justified as it typically deteriorates the accuracy of LMTO calculations. Therefore, in the present work we leave this problem as it is and fix the LMTO basis set.
2. Once the LMTO basis is fixed, the relative weight of the TM $`d`$-states and other atomic states which contribute to the $`t_{2g}`$ band cannot be changed (see Fig. 1). For example, the contribution of the oxygen $`2p`$-states cannot be replaced by muffin-tin orbitals centered at the TM sites and vise versa. The same proportion of atomic orbitals should be preserved in the WFs, constructed for this $`t_{2g}`$ band. Then, the only parameter which can be controlled is how many WFs, centered at different sites of the lattice, contribute to the density of $`d`$-states at the given TM site. Then, the definition “localized orbital” mean that it is mainly centered around given TM site. Conversely, the “delocalized orbital” may have a long tail spreading over other TM sites. Then, it is easy to see that our procedure corresponds to the former choice. Indeed, in the first order of $`(\widehat{𝔥}`$$``$$`\widehat{E}_{\nu t})`$ and neglecting for a while the nonorthogonality to the rest of the electronic states, the norm of the WF can be obtained from Eq. (16) as $`\stackrel{~}{W}|\stackrel{~}{W}`$$`=`$$`\varphi _t|\varphi _t`$, meaning that the WF is fully localized at the central TM site. Then, it holds $`\widehat{𝔥}`$$`=`$$`\widehat{h}`$, which is valid in the second order of $`(\widehat{𝔥}`$$``$$`\widehat{E}_{\nu t})`$.LMTO Therefore, the leading correction to the above approximation, which define the actual weight of the WF at the neighboring TM sites is controlled by the parameters of the kinetic energy $`\widehat{h}`$, and is of the order of $`\widehat{h}\widehat{p}_t\widehat{h}`$. As we will see below, the latter is small. The conclusion is rather generic and well anticipated for the strongly-correlated systems for which the kinetic-energy terms is generally small.
3. The angular character of the WF at the central TM site should be consistent with the one extracted from the local density of states in the region of $`t_{2g}`$ bands (in the other words, the local density of states at the TM sites should be well represented by atomic orbitals $`\{|\stackrel{~}{\chi }_t\}`$ used in the downfolding method). Therefore, we choose $`\{|\stackrel{~}{\chi }_t\}`$ as the set of atomic orbitals which mainly contribute to the local density of states in the region of $`t_{2g}`$ bands. For these purposes, at each TM site we calculate the density matrix in the basis of five $`d`$ orbitals $`\{|\stackrel{~}{\chi }_d\}`$:
$$\widehat{𝒩}=\underset{it_{2g}}{}\stackrel{~}{\chi }_d|\psi _i\psi _i|\stackrel{~}{\chi }_d,$$
(19)
and sum up the contributions of all $`t_{2g}`$ bands (here, $`i`$ is an joint index, which incorporates the band index and the coordinates of the $`𝐤`$-point in the first Brillouin zone). Then, we diagonalize $`\widehat{𝒩}`$, and assign three most populated orbitals, obtained after the diagonalization to $`\{|\stackrel{~}{\chi }_t\}`$.
## V Crystal-field splitting caused by nonsphericity of electron-ion Interactions
The contribution of Coulomb interactions to the crystal-field splitting is a tricky issue. Despite an apparent simplicity of the problem, one should clearly distinguish different contribution and not to include them twice, in the kinetic and Coulomb parts of the model Hamiltonian (3). The use of full-potential techniques does not automatically guarantee the right answer. However, the atomic-spheres-approximation, which typically supplements the LMTO method, will also require additional corrections for the crystal-field splitting. In this section we would like to make two comments on this problem.
1. The nonsphericity of on-site Coulomb interactions is already included in the second part of the model Hamiltonian (3). The problem will be discussed in details in Sec. VII. Therefore, in order to avoid the double counting, the corresponding contribution to the kinetic-energy part should be subtracted. From this point of view the use of the spherically averaged KS potential in ASA is well justified.
The same is true for the intersite Coulomb interactions, if they are explicitly included to the model Hamiltonian (3).
2. All remaining interactions should generally contribute to the crystal-field splitting. In ASA, the proper correction at the site $`𝐑`$ can be found by considering the matrix elements of the Coulomb potential produced by all other atomic spheres (or ions) at the site $`𝐑`$,
$$\mathrm{\Delta }\widehat{h}_{\mathrm{𝐑𝐑}}=\underset{𝐑^{}𝐑}{}\stackrel{~}{W}_𝐑|\frac{Z_𝐑^{}^{}e^2}{|𝐑+𝐫𝐑^{}|}|\stackrel{~}{W}_𝐑,$$
where $`\stackrel{~}{W}_𝐑`$$``$$`\stackrel{~}{W}_𝐑(𝐫)`$ is the WF centered at the site $`𝐑`$, and $`Z_𝐑^{}^{}`$ is the total charge associated with the sphere $`𝐑^{}`$: namely, the nuclear charge minus the electronic charge encircled by the atomic sphere. The nonspherical part of this integral can be easily calculated in the real space by using the multipole expansion for $`|𝐑+𝐫𝐑^{}|^1`$.
In all forthcoming discussions, unless it is specified otherwise, the matrix elements of the crystal-field splitting will incorporate the correction $`\mathrm{\Delta }\widehat{h}_{\mathrm{𝐑𝐑}}`$ associated with nonsphericity of the electron-ion interactions.
## VI Applications to transition-metal oxides
### VI.1 Cubic Perovskites: SrVO<sub>3</sub>
SrVO<sub>3</sub> is a rare example of perovskite compounds, which crystallizes in the ideal cubic structure. It attracted a considerable attention in the connection with the bandwidth control of the metal-insulator transition.IFT
For the cubic compounds, the separation of the LMTO basis functions into $`\{|\stackrel{~}{\chi }_t\}`$ and $`\{|\stackrel{~}{\chi }_r\}`$ used in the downfolding method is rather straightforward: three $`t_{2g}`$ orbitals centered at each V site form the subspace of $`\{|\stackrel{~}{\chi }_t\}`$ orbitals, and the rest of the basis functions are associated with $`\{|\stackrel{~}{\chi }_r\}`$.
Parameters of LMTO calculations for SrVO<sub>3</sub> are given in Table 1.
The corresponding electronic structure is shown in Fig. 3.
The downfolding procedure is nearly perfect and well reproduces the behavior of three $`t_{2g}`$ bands. As expected for cubic perovskite compounds,SlaterKoster the transfer interactions between different $`t_{2g}`$ orbitals are small. The dispersion of $`t_{2g}`$ bands is well described in terms of three interaction parameters $`t_1`$, $`t_1^{}`$, and $`t_2`$:
$$\epsilon _{xy}(𝐤)=2t_1(\mathrm{cos}ak_x+\mathrm{cos}ak_y)+2t_1^{}\mathrm{cos}ak_z+4t_2\mathrm{cos}ak_x\mathrm{cos}ak_y$$
($`a`$ being the cubic lattice parameter; similar expressions for the $`yz`$ and $`zx`$ bands are obtained by cyclic permutations of the indices $`x`$, $`y`$, and $`z`$). The parameters $`t_1`$, $`t_1^{}`$, and $`t_2`$, obtained after the Fourier transformation, are listed in Table 2.
As expected, the nearest-neighbor (NN) $`dd\pi `$-interaction $`t_1`$ mediated by the oxygen $`2p`$-states is the strongest. For the $`xy`$-orbitals, it operates in the $`x`$ and $`y`$ directions. However, there is also an appreciable $`dd\delta `$-interaction $`t_1^{}`$ operating in the “forbidden” direction (for example, the direction $`z`$ in the case of $`xy`$ orbitals). These interactions are mediated by the Sr($`4d`$) states and strongly depend on the proximity of the latter to the Fermi level. Therefore, it is not quite right to say that the transfer interactions between $`t_{2g}`$ orbitals are strictly two-dimensional in the cubic lattice.HarrisPRL03 Since the La($`5d`$) states are located even lower in energy than the Sr($`4d`$) ones, the interaction $`t_1^{}`$ is expected to be even stronger in LaTiO<sub>3</sub>. However, in the case of LaTiO<sub>3</sub> we have an additional complication associated with the orthorhombic distortion. As we will see below, it changes the conventional form of transfer interactions expected for the simplified cubic perovskite structure dramatically.
The corresponding WF is shown in Fig. 4.
In this case, the $`t`$-part of the WF was constructed from the $`3d`$-$`t_{2g}`$ partial waves inside V spheres. The partial waves of the Sr($`4d5s`$), V($`3d`$-$`e_g`$), and O($`2p`$) types were included into the $`r`$-part, in order to enforce the orthogonality of the WF to the bands of the aforementioned type.
Since the $`t_{2g}`$ band is an antibonding combination of the atomic V($`3d`$-$`t_{2g}`$) and O($`2p`$) orbitals, the WF has nodes located between V and O sites. Fig. 5 illustrates the spacial extension of the WF.
It shows the electronic charge accumulated around the central V site after adding every new sphere of the neighboring atomic sites. Since the WF is normalized, the total charge should be equal to one. In the case of SrVO<sub>3</sub>, 77% of the this charge belongs to the central V site, 16% is distributed over four neighboring oxygen sites, about 5% belongs to the next eight Sr sites, and 1% – to the eight oxygen sites located in the fourth coordination sphere. Other contributions are small.
It is also instructive to calculate the expectation value of the square of the position operator, $`𝐫^2`$$`=`$$`W|𝐫^2|W`$, which characterizes the spread of the WF in the method of Marzari and Vanderbilt.MarzariVanderbilt They proposed to define the “maximally localized” Wannier orbitals as the ones which minimize $`𝐫^2`$. Using the WF shown in Fig. 4, we obtain $`𝐫^2`$$`=`$$`2.37`$ Å<sup>2</sup>. Unfortunately, at present all applications of the method by Marzari and Vanderbilt to the TM oxides are limited by MnO.Posternak Therefore, we can make only indirect comparison between two different compounds. The values of $`𝐫^2`$ reported in Ref. Posternak, for individual WFs centered at the Mn and O sites were of the order of 0.6-0.8 Å<sup>2</sup>, that is considerably smaller than 2.37 Å<sup>2</sup> obtained in our work for SrVO<sub>3</sub>. However, such a difference is not surprising.
1. Our scheme of constructing the WFs is not based on the minimization of $`𝐫^2`$. Therefore, our values of $`𝐫^2`$ should be generally larger.
2. More importantly, the spacial extension of the WFs depends on the dimensionality of the Hilbert space, which is used in the construction of the Hubbard model. For example, we will show in Sec. VI.4 that by treating explicitly the $`e_g^\sigma `$ states in V<sub>2</sub>O<sub>3</sub> one can easily find more compact representation for the Wannier orbitals. This characteristic can be further improved by including the O($`2p`$) states explicitly into the Wannier basis,comment.2 like in Ref. Posternak, for MnO. However, there is a very high price to pay for this extra localization. This is the dimensionality of the Hilbert space, which becomes crucial in the many-body methods for the numerical solution of the model Hamiltonian (3).
Thus, we believe that our WFs constructed for isolated $`t_{2g}`$ band are indeed well localized.
For cubic perovskites, there are several ways of extracting parameters of transfer interactions from the first-principles electronic structure calculations. For example, one can simply fit the LDA band structure in terms of a small number of Slater-Koster interactions.SlaterKoster However, the situation becomes increasingly complicated in materials with the lower crystal symmetry, like the orthorhombically distorted perovskite oxides, corundum, or pyrochlore compounds. First, the number of possible Slater-Koster interactions increases dramatically. Second, the form of these interactions becomes very complicated and differs substantially from the cubic perovskite compounds (one example is the mixing of $`t_{2g}`$ and $`e_g`$ orbitals by the orthorhombic distortion, which does not occur in cubic perovskites). Therefore, it seems that for complex compounds the only way to proceed is to use the downfolding method. In the next sections we will consider several examples along this line.
### VI.2 Orthorhombically Distorted Perovskites: YTiO<sub>3</sub>
YTiO<sub>3</sub> is a ferromagnetic insulator. The resent interest to this compound has been spurred by the behavior of orbital polarization, which is closely related with the origin of the ferromagnetic ground state. YTiO<sub>3</sub> is typically considered in combination with LaTiO<sub>3</sub>, which is an antiferromagnetic insulator. The magnetic behavior of these two, formally isoelectronic materials, is not fully understood.PRB04
Contrary to SrVO<sub>3</sub>, both YTiO<sub>3</sub> and LaTiO<sub>3</sub> crystallize in the strongly distorted orthorhombic structure (shown in Fig. 6 for YTiO<sub>3</sub>, the space group No. 62 in the International Tables; the Schönflies notation is $`D_{2h}^{16}`$). In this section we will illustrate abilities of the downfolding method for distorted perovskite compounds, using YTiO<sub>3</sub> as an example. Parameters of LMTO calculations for YTiO<sub>3</sub> are given in Table 3.
A new problem we have to address here is how to separate the basis functions of the LMTO method onto the $`\{|\stackrel{~}{\chi }_t\}`$ and $`\{|\stackrel{~}{\chi }_r\}`$ orbitals. Note that although the $`t_{2g}`$ band is well separated from the rest of the electronic structure also in the case of YTiO<sub>3</sub>, the atomic $`t_{2g}`$ and $`e_g`$ orbitals are strongly mixed by the crystal-field effects and the transfer interactions in the distorted perovskite structure. Therefore, the conventional separation into atomic $`t_{2g}`$ orbitals and the rest of the basis functions does not apply here, and in order to generate $`\{|\stackrel{~}{\chi }_t\}`$ we use eigenvectors of the density matrix (see Sec. IV.1).
For the site 1, shown in Fig. 6, these three “$`t_{2g}`$ orbitals” have the following form (in the basis of $`|xy`$, $`|yz`$, $`|z^2`$, $`|zx`$, and $`|x^2`$$``$$`y^2`$ orbitals, in the orthorhombic coordinate frame):
$`|\stackrel{~}{\chi }_1`$ $`=`$ $`(0.13,0.60,0.24,0.34,0.67),`$
$`|\stackrel{~}{\chi }_2`$ $`=`$ $`(0.17,0.50,0.35,0.77,0.11),`$ (20)
$`|\stackrel{~}{\chi }_3`$ $`=`$ $`(0.43,0.54,0.29,0.22,0.62).`$
At the sites 2, 3, and 4 similar orbitals can be generated from the ones at the site 1 using the symmetry properties of the $`D_{2h}^{16}`$ group and applying the 180 rotations around the orthorhombic axes $`𝐚`$, $`𝐜`$, and $`𝐛`$, respectively. These orbitals define the local basis (or the local coordinate frame) around each Ti site.
Then, the rest of the basis functions $`\{\stackrel{~}{\chi }_r\}`$ can be eliminated using the downfolding method. The corresponding electronic structure for the $`t_{2g}`$ bands is shown in Fig. 6, which reveals an excellent agreement between results of LMTO calculations and their tight-binding parametrization using the down-folding method.
Parameters obtained after the transformation to the real space are listed in Table 4, in the local coordinate frame.
We note a substantial crystal-field splitting associated with the orthorhombic distortion. After the diagonalization of the site-diagonal part of the TB Hamiltonian $`\widehat{h}`$, we obtain the following (“one-down, two-up”) splitting of $`t_{2g}`$ levels: $``$$`0.076`$, $`0.032`$, and $`0.046`$ eV. Some implications of the crystal-field splitting to the orbital polarization and the magnetic ground state of YTiO<sub>3</sub> and LaTiO<sub>3</sub> have been discussed in Refs. PRB04, and MochizukiImada, .comment.1 The form of transfer interactions becomes extremely complicated, and differs dramatically from many naive expectations based on the analogy with the cubic perovskites. Generally, the transfer interactions are three-dimensional and operate between different $`t_{2g}`$ orbitals.
The WFs are shown in Fig. 7, and their spacial extension in the real space is illustrated in Fig. 8. In these calculations, the WFs have been orthogonalized to the O($`2p`$), Ti($`3d`$-$`e_g`$), and Y($`4d`$) bands.
The orbitals appear to be more localized than in SrVO<sub>3</sub>: 82-87% of the total charge belongs the central Ti site, and only 6 to 10% is distributed over neighboring oxygen sites. This is because of the large orthorhombic distortion, which suppresses all interatomic interactions mediated by the oxygen states. Another reason is the larger energy distance between O($`2p`$) and $`t_{2g}`$ bands in YTiO<sub>3</sub> ($`3.2`$ eV against $`0.3`$ eV in SrVO<sub>3</sub> – see Fig. 1), which explains smaller weight of the atomic oxygen states in the $`t_{2g}`$ band and the WFs of YTiO<sub>3</sub>.
Another interesting feature is that the degree of localization is pretty different for three orbitals. For example, we obtain $`𝐫^2`$$`=`$ 2.28, 1.90, and 2.05 Å<sup>2</sup>, correspondingly for $`W_1`$, $`W_2`$, and $`W_3`$ shown in Fig. 7. One can paraphrase it in a different way: the degree of hybridization can be different for different $`t_{2g}`$ orbitals, unless they are related with each other by symmetry operations.
### VI.3 Pyrochlores: Y<sub>2</sub>Mo<sub>2</sub>O<sub>7</sub>
The pyrochlore compounds exhibit a variety of interesting properties. Many of them are not fully understood.
Y<sub>2</sub>Mo<sub>2</sub>O<sub>7</sub> is a canonical example of geometrically frustrated systems. In this compound, the magnetic atoms form the networks of corner-sharing tetrahedra. Then, the antiferromagnetic coupling between NN Mo spins leads to the frustration. The origin of this antiferromagnetic coupling can be understood on the basis of semi-empirical Hartree-Fock calculations.PRB03 A remaining question, which is not fully understood, is the origin of the spin-glass state realized in Y<sub>2</sub>Mo<sub>2</sub>O<sub>7</sub> below 20 K.Y2Mo2O7exp For comparison, Nd<sub>2</sub>Mo<sub>2</sub>O<sub>7</sub> is a ferromagnet, revealing a large anomalous Hall effect.PRB03 ; Taguchi
Another interesting group is superconducting $`\beta `$-pyrochlores with the chemical formula $`A`$Os<sub>2</sub>O<sub>6</sub> ($`A`$$`=`$ K, Rb, and Cs).Hiroi
In this section we will derive parameters of the kinetic energy for the $`t_{2g}`$ band of Y<sub>2</sub>Mo<sub>2</sub>O<sub>7</sub>. Very similar strategy can be applied for other pyrochlores. Parameters of LMTO calculations for Y<sub>2</sub>Mo<sub>2</sub>O<sub>7</sub> are listed in Table 5.
In the pyrochlore lattice, each Mo site is located in the trigonal environment (Fig. 9).
Therefore, the atomic Mo($`t_{2g}`$) levels will be split into one-dimensional $`a_{1g}`$ and two-dimensional $`e_g^\pi `$ representations. The latter states can mix with the Mo($`e_g^\sigma `$) states, which belong to the same representation. Therefore, the basis functions $`\{|\stackrel{~}{\chi }_t\}`$ can be constructed in the same manner as for YTiO<sub>3</sub>, by diagonalizing the site-diagonal part of the density matrix for the $`t_{2g}`$ bands. For the site 1 shown in Fig. 9, this yields the following atomic orbitals (in the basis of $`|xy`$, $`|yz`$, $`|z^2`$, $`|zx`$, and $`|x^2`$$``$$`y^2`$ orbitals):
$`|\stackrel{~}{\chi }_1`$ $`=`$ $`(0.58,0.58,0,0.58,0),`$
$`|\stackrel{~}{\chi }_2`$ $`=`$ $`(0.06,0.18,0.29,0.13,0.93),`$ (21)
$`|\stackrel{~}{\chi }_3`$ $`=`$ $`(0.18,0.04,0.93,0.14,0.29).`$
In these notations, the first orbital correspond to the $`a_{1g}`$ representation, and two other – to the $`e_g`$ representation. Similar orbitals at the sites 2, 3, and 4 can be generated from the ones at the site 1 using the symmetry operations of the $`O_h^7`$ group (No. 227 in the International Tables): namely, the 180 rotations around the cubic axes $`𝐱`$, $`𝐲`$, and $`𝐳`$, respectively. The rest of the basis functions form the subspace $`\{|\stackrel{~}{\chi }_r\}`$.
The electronic structure obtained after the elimination of the $`\{|\stackrel{~}{\chi }_r\}`$ orbitals is shown in Fig. 9. Again, we note an excellent agreement with the results of the original LMTO calculations. The parameters of the kinetic energy in the real space are listed in Table 6.
We note an appreciable ($``$$`0.25`$ eV) crystal-field splitting between the $`a_{1g}`$ and $`e_g`$ orbitals.PRB03 The NN transfer interactions in the bonds other than 1-4 can be obtained using the symmetry operations of the $`O_h^7`$ group. The transfer interactions beyond the nearest neighbors are considerably smaller.
The corresponding WFs are shown in Fig. 10, and their spacial extension is depicted in Fig. 11.
The WFs have been orthogonalized to the neighboring Y($`4d`$-$`e_g`$) and O($`2p`$) bands. Since the $`4d`$-wavefunctions are typically more extended in comparison with the $`3d`$ ones, the WFs are less localized. In the case of Y<sub>2</sub>Mo<sub>2</sub>O<sub>7</sub>, 75-80% of the total charge is located at the central site, and about 20% is distributed over neighboring oxygen sites. The degree of localization also depends on the symmetry of WFs. So, the $`a_{1g}`$ orbital is well localized within the MoO<sub>6</sub> cluster, whereas the $`e_g`$ orbitals have a noticeable weight ($``$2.5% of the total charge) at the Y and Mo sites belonging to the next coordination sphere.
### VI.4 Corundum-type V<sub>2</sub>O<sub>3</sub>
V<sub>2</sub>O<sub>3</sub> is regarded as the canonical Mott-Hubbard system, where the Coulomb interaction between conduction electrons leads to a breakdown of the conventional one-electron band theory.McWhanRiceRemeika It was and continues to be the subject of vast research activity, which has been summarized in many review articles (for instance, Ref. IFT, ).
V<sub>2</sub>O<sub>3</sub> crystallizes in the corundum structure with two formula units per rhombohedral cell (the space group is $`D_{3d}^6`$, No. 167 in the International Tables). The local environment of the V sites is trigonal (Fig. 12), in which the $`t_{2g}`$ levels are split into one-dimensional $`a_{1g}`$ and two dimensional $`e_g^\pi `$ representations.
Parameters of LMTO calculations for V<sub>2</sub>O<sub>3</sub> are given in Table 7.
In the present work, our main interest in V<sub>2</sub>O<sub>3</sub> will be purely academic. As we can see in Fig. 1, V<sub>2</sub>O<sub>3</sub> has two well separated bands, which are mainly formed by the V($`3d`$) states. One is the $`t_{2g}`$ band, which in LDA is crossed by the Fermi level. Another one is the $`e_g`$ band, which is located around 3 eV, and composed mainly of the $`e_g^\sigma `$ states. The latter can mix with the $`e_g^\pi `$ ones. Therefore, for V<sub>2</sub>O<sub>3</sub> (and related corundum-type oxides) one can introduce two different models.
The first one is more general and explicitly treats all V($`3d`$) bands. In the following we will call it as the “five-orbital” model, according to the number of basis orbitals $`\{|\stackrel{~}{\chi }_t\}`$ per one V site. The oxygen degrees of freedom will be eliminated using the downfolding method. The corresponding TB Hamiltonian will be denoted as $`\widehat{h}^{(5)}`$.
The second one is the minimal model, which can be derived from the previous one by eliminating the $`e_g^\sigma `$ states. We will call it the “three-orbital model”. The corresponding TB Hamiltonian will be denoted as $`\widehat{h}^{(3)}`$.
The basic difference between these two models is that the first one treats the $`e_g^\sigma `$ states explicitly, while in the second case the effect of these states is included implicitly, through the renormalization of interaction parameters between $`a_{1g}`$ and $`e_g^\pi `$ orbitals.
For both models, the local orbitals at each V site were obtained from the diagonalization of the density matrix, which sums up the contributions over twelve $`t_{2g}`$ bands. For the three-orbital model, such choice of the basis functions is very important, as it controls the accuracy of the downfolding method. For the five-orbital model, one can use any unitary transformation of the five $`3d`$ orbitals. Obviously, the final result will not depend on this transformation. However, for a better comparison between two models, we use the same basis in both cases. In principle, the $`e_g^\pi `$ and $`e_g^\sigma `$ states will be mixed in the density matrix, as they belong to the same representation. However, we will continue to call the lower- and upper-lying $`e_g`$ states as $`e_g^\pi `$ and $`e_g^\sigma `$, despite the fact that each of them may have an admixture of another type.
Then, the basis orbitals at the sites 1 and 2 (see Fig. 12) have the following form (in the basis of atomic $`|xy`$, $`|yz`$, $`|z^2`$, $`|zx`$, and $`|x^2`$$``$$`y^2`$ orbitals):
$`|\stackrel{~}{\chi }_1`$ $`=`$ $`(0.77,0.29,0,0.48,0.30),`$
$`|\stackrel{~}{\chi }_2`$ $`=`$ $`(0.30,0.48,0,0.29,0.77),`$
$`|\stackrel{~}{\chi }_3`$ $`=`$ $`(0,0,1,0,0),`$ (22)
$`|\stackrel{~}{\chi }_4`$ $`=`$ $`(0.49,0.51,0,0.65,0.27),`$
$`|\stackrel{~}{\chi }_5`$ $`=`$ $`(0.27,0.65,0,0.51,0.49).`$
In these notations, $`|\stackrel{~}{\chi }_1`$ and $`|\stackrel{~}{\chi }_2`$ are the $`e_g^\pi `$ orbitals, $`|\stackrel{~}{\chi }_3`$ is the $`a_{1g}`$ orbital, and $`|\stackrel{~}{\chi }_4`$ and $`|\stackrel{~}{\chi }_5`$ are the $`e_g^\sigma `$ orbitals. The basis orbitals at the sites 3, 4, and 5 are generated by the mirror-reflection $`𝐲`$$``$$``$$`𝐲`$ in Eq. (22). In the three-orbital model, first three orbitals constitute the subspace $`\{|\stackrel{~}{\chi }_t\}`$, while two remaining orbitals are included in $`\{|\stackrel{~}{\chi }_r\}`$. In the five-orbital model, all five orbitals are included in $`\{|\stackrel{~}{\chi }_t\}`$.
The electronic structure obtained after the downfolding is shown in Fig. 12. The three-orbital model well reproduces the behavior of twelve $`t_{2g}`$ bands, while the five-orbital model allows to reproduce both $`t_{2g}`$ and $`e_g`$ bands.
The corresponding parameters in the real space are given in Table 8. In the five-orbital model, one can see a noticeable hybridization between $`t_{2g}`$ and $`e_g^\sigma `$ states. Therefore, the elimination of the $`e_g^\sigma `$ states in the three-orbital model should lead to an additional renormalization of the parameters of the crystal-field splitting and the transfer interactions. Generally, the matrix elements of the kinetic-energy part in the subspace of $`t_{2g}`$ orbitals are not the same for two considered models.
As an example, we shown in Table 9 the crystal-filed splitting between $`e_g^\pi `$ and $`a_{1g}`$ levels for the series of corundum-type oxides, obtained after the diagonalization of the matrices $`\widehat{h}_{11}^{(3)}`$ and $`\widehat{h}_{11}^{(5)}`$. The splitting turns out to be very different in two different models. Since many properties of TM oxides are controlled by this crystal-field splitting,CFcorundum such a model-dependence may be viewed as somewhat unphysical. However, the crystal-field splitting cannot be considered independently from other model parameters, such as the Coulomb and transfer interactions, which should be defined on the same footing and for the same type of model. For the Coulomb interactions, it is important to follow the concept of WFs, which we will consider in the next section.
We also note an appreciable contribution coming from nonsphericity of the electron-ion interactions. This contribution, which is ignored in conventional ASA, acts against the crystal-field splitting originating from the transfer interactions and tends to stabilize the $`a_{1g}`$ level. This may revise certain conclusions obtained in the framework of ASA-LMTO method.CFcorundum
Corresponding WFs are shown in Fig. 13, and their spacial extension – in Fig. 14.
All functions have been orthogonalized to the O($`2p`$) band. In the three-orbital model, the WFs have been additionally orthogonalized to the V($`e_g`$) band. Similar to Y<sub>2</sub>Mo<sub>2</sub>O<sub>7</sub>, the spacial extension of the WFs strongly depends on their symmetry. Generally, the $`a_{1g}`$ and $`e_g^\pi `$ orbitals are more localized, while the $`e_g^\sigma `$ orbitals have a considerable weight (more than 20% of the total charge) at the neighboring oxygen sites. Furthermore, the spacial extension of the WFs depends on the model for which they are constructed. Generally, the five-orbital model allows to construct more compact WFs rather than the three-orbital one. For example, in the three-orbital model we have $`𝐫^2`$$`=`$ 1.75 and 2.38 Å<sup>2</sup>, correspondingly for the $`a_{1g}`$ and $`e_g^\pi `$ orbitals. For comparison, the five-orbital yields $`𝐫^2`$$`=`$ 1.04, 1.04, and 1.41 Å<sup>2</sup> for the $`a_{1g}`$, $`e_g^\pi `$, and $`e_g^\sigma `$ orbitals, respectively.
This is not surprising.
1. The transfer interactions in the three-orbital model are longer-ranged, as they contain additional contributions mediated by the $`e_g^\sigma `$ orbitals.
2. For the three-orbital model, the WFs should be additionally orthogonalized to the V($`e_g`$) band. At the central site, this condition can be easily satisfied by choosing proper atomic $`t_{2g}`$ and $`e_g`$ orbitals, which diagonalize the density matrix. However, the WF has a tail spreading to the neighboring sites, which should be additionally orthogonalized to the V($`e_g`$) band by including partial waves of the $`e_g^\sigma `$ type into the $`r`$-part of the WF.
Thus, there is certain compromise with the choice of the suitable model for compounds like V<sub>2</sub>O<sub>3</sub>, where smaller dimensionality of the Hilbert space in the three-orbital model is counterbalanced by necessity to deal with more extended WFs.
## VII The Effective Coulomb Interaction
The calculation of effective Coulomb interactions for the first principles is an extremely complicated problem because they are subjected to different mechanisms of screening which should be taken into consideration in the process of these calculations. So far, the solution of this problem has not been fully accomplished by any of the research groups, despite a vast activity in this direction.Dederichs ; Norman ; McMahan ; Gunnarsson ; Hybertsen ; GunnarssonPostnikov ; AnisimovGunnarsson ; PRB94.2 ; NormanBrooks ; PRB96 ; Pickett ; Springer ; Kotani ; Ferdi04 ; PRB05 It would be probably fair to say from the very beginning that we were not able to solve this problem either, without additional approximations, which will be considered in Sec. VII.1. However, we hope to present certain systematics on different points of view, which can be found in the literature. We will also summarize several open questions and unresolved problems.
It is convenient to start with the basic definition of the effective Coulomb interaction $`U`$, as it was discussed in many details by Herring.Herring According to this definition, (the spherically averaged part of) $`U`$ is nothing but the energy cost for the reaction $`2(d^n)`$$``$$`d^{n+1}`$$`+`$$`d^{n1}`$, i.e. for moving a $`d`$-electron between two atoms, located at $`𝐑`$ and $`𝐑^{}`$, and initially populated by $`n_𝐑`$$`=`$$`n_𝐑^{}`$$``$$`n`$ electrons:
$$U=E[n_𝐑+1,n_𝐑^{}1]E[n_𝐑,n_𝐑^{}].$$
(23)
This $`U`$ may depend on $`𝐑`$ and $`𝐑^{}`$, and using several combinations of $`𝐑`$ and $`𝐑^{}`$ one can extract the values of both on-site and intersite interactions. A typical example for SrVO<sub>3</sub> will be considered in Sec. VII.2. However, here we drop these atomic indices and consider more general aspects of calculations of the effective Coulomb interactions.
It is implied that the electron is transferred between two Wannier orbitals, and $`n_𝐑`$ and $`n_𝐑^{}`$ are the populations of these Wannier orbitals. A special precaution should be taken in order to avoid the double counting of the kinetic energy term. Indeed, since the kinetic-energy term is included explicitly into the Hubbard model (3), it should not contribute to the total energy difference (23). This point was emphasized by Gunnarsson and co-workers, in the series of publications.Gunnarsson ; GunnarssonPostnikov ; AnisimovGunnarsson They proposed to derive $`U`$ from constraint-LDA (c-LDA) calculations,Dederichs and suppress all matrix elements of hybridization involving the atomic $`d`$-states. Such a procedure can be easily implemented in the LMTO method. For the $`3d`$-compounds, this method typically yields $`U`$$``$5-12 eV,AZA ; PRB96 , which is too large (if correct, results of this approach would imply that all nature surrounding us would be “strongly correlated”). Therefore, although the basic strategy is correct, there is an important piece of physics, which is missing in the method of Gunnarsson et al.
Similar strategy can be pursued in our WF method. Our basic idea is to switch off the kinetic-energy term during the construction of the WFs, and to use these functions in calculations of the effective interaction $`U`$. Therefore, instead of regular WFs $`\{\stackrel{~}{W}(\widehat{h})\}`$, which after applying to the KS Hamiltonian generate the matrix $`\widehat{h}\stackrel{~}{W}(\widehat{h})|H_{\mathrm{KS}}|\stackrel{~}{W}(\widehat{h})`$, we introduce the set of auxiliary Wannier functions $`\{\overline{W}(\widehat{c})\}`$, satisfying the condition $`\overline{W}(\widehat{c})|H_{\mathrm{KS}}|\overline{W}(\widehat{c})=\widehat{c}`$. In the ground-state configuration ($`n_𝐑`$$`=`$$`n_𝐑^{}`$$``$$`n`$), $`\widehat{c}`$ is a constant, which can be dropped.comment.3 In the excited state ($`n_𝐑`$$``$$`n_𝐑^{}`$), $`\widehat{c}`$ is a diagonal matrix with respect to the site indices, $`\widehat{c}c_𝐑\delta _{\mathrm{𝐑𝐑}^{}}`$, where each matrix element $`c_𝐑`$ may depend on occupation numbers $`\{n_𝐑\}`$. Since such auxiliary WFs do not interact with each other through the kinetic-energy term, they can be used as the basis functions for the effective Coulomb interaction $`U`$.
The auxiliary WFs can be easily constructed using the method proposed in Sec. IV after the substitution $`\widehat{h}`$$`=`$$`\widehat{c}`$ in all equations. Meanwhile, the orthogonality condition to other LDA bands is strictly observed by including proper solutions of KS equations inside atomic spheres and their energy derivatives into the $`r`$-part of auxiliary WFs. This allows to retain the hybridization between TM $`d`$\- and oxygen $`p`$-states, which is an important feature of TM oxides. As we will see below, the change of this hybridization, induced by the reaction $`2(d^n)`$$``$$`d^{n+1}`$$`+`$$`d^{n1}`$, represents a very important channel of screening, which substantially reduces $`U`$ and explains many details of its behavior in solids. This channel of screening has been overlooked by Gunnarsson et al.
A similar idea, although formulated in the very different way, has been recently proposed by Aryasetiawan et al.Ferdi04 They proposed to extract the parameter $`U`$ from the GW method,Hedin ; FerdiGunnarsson and suppressed all contributions to the GW polarization function associated with the transitions between Hubbard (in our case – $`t_{2g}`$) bands, in order to avoid the double counting of these effects in the process of solution of the Hubbard model. Clearly, since the polarization function in the GW method will vanish without the kinetic-energy term, this procedure appears to be similar to the setting $`\widehat{h}`$$`=`$$`\widehat{c}`$ for the WFs, which are used as the basis functions for the effective Coulomb interaction $`U`$.
A characteristic example of the auxiliary WFs is shown in Fig. 15 for SrVO<sub>3</sub>.
We note only a minor difference between auxiliary WFs and the regular ones shown in Fig. 4, meaning that the main details of the WFs for the $`t_{2g}`$ bands are predetermined by orthogonality condition to other bands. For example, in the case of auxiliary WFs, 79% of the total charge are accumulated at the central V site (instead of 77% for the regular WFs). The values of $`𝐫^2`$ obtained for the auxiliary and regular WFs are 2.27 and 2.37 Å, respectively. Thus, the auxiliary WFs appear to be more localized. However, the difference is small. The result is well anticipated for strongly correlated systems, for which the kinetic-energy term $`\widehat{h}`$ is expected to be small.
Since the KS Hamiltonian is diagonal in the basis of auxiliary WFs, the latter can be regarded as eigestates of this Hamiltonian corresponding to certain boundary condition. The corresponding set of KS eigenvalues will be denoted as $`\{\epsilon _𝐑\}`$. Then, the occupation numbers $`\{n_𝐑\}`$ become well defined and one can use the standard properties of the density-functional theory. Namely, by using Janak’s theorem for the KS eigenvalues
$$\epsilon _𝐑=\frac{\delta E}{\delta n_𝐑}$$
and Slater’s transition-state arguments, Eq. (23) can be further rearranged as
$$U=\epsilon _𝐑[n_𝐑+\frac{1}{2},n_𝐑^{}\frac{1}{2}]\epsilon _𝐑[n_𝐑\frac{1}{2},n_𝐑^{}+\frac{1}{2}].$$
The final expression for the parameter $`U`$ is obtained by considering the deviations $`\pm `$$`1/2`$ from $`n_𝐑`$ and $`n_𝐑^{}`$ as a weak perturbation and employing the Taylor expansion. Then, in the first order of $`\pm `$$`1/2`$ one obtains:
$$U=\frac{\delta \epsilon _𝐑}{\delta n_𝐑},$$
(24)
where the energy derivative is calculated under the following condition:
$$n_𝐑+n_𝐑^{}=\mathrm{const},$$
(25)
which guarantees the conservation of the total number of particles. Strictly speaking, the definition (24) corresponds to the infinitesimal change of the occupation numbers $`2(d^n)`$$``$$`d^{n+\delta n}`$$`+`$$`d^{n\delta n}`$, which is different from original Herring’s definition (23). However, for practical purposes, these definitions can be regarded as equivalent as they yield very similar values for the parameter $`U`$.PRB94.2
Then, it is convenient to use the Hellman-Feinman theorem, which allows to relate $`U`$ with the change of the Hartree potential (the change of the exchange-correlation potential in LDA is typically small and can be neglected):PRB94.2
$$U=\overline{W}|\frac{\delta V_\mathrm{H}}{\delta n_𝐑}|\overline{W}.$$
Taking into account that $`V_\mathrm{H}(𝐫)`$$`=`$$`e^2𝑑𝐫^{}\rho (𝐫^{})/|𝐫𝐫^{}|`$, and using Eq. (2) for the electron density, the above expression can be rearranged as
$$U=e^2𝑑𝐫𝑑𝐫^{}\frac{|\overline{W}(𝐫^{})|^2}{|𝐫𝐫^{}|}\frac{\delta \rho (𝐫)}{\delta n_𝐑},$$
where
$$\frac{\delta \rho (𝐫)}{\delta n_𝐑}=\underset{i}{}\left\{\frac{\delta n_i}{\delta n_𝐑}|\psi _i(𝐫)|^2+n_i\frac{\delta }{\delta n_𝐑}|\psi _i(𝐫)|^2\right\}.$$
(26)
The last expression points out at the existence of two additive channels of screening.
(i) The first one comes from the change of occupation numbers. Due to the constraint (25) imposed on the occupation numbers, this channel involves two Wannier orbitals, centered at different TM sites, and describes the screening of on-site Coulomb interactions by intersite interactions. Other states can be affected by this term only through the change of wavefunctions in the process of iterative solution of the KS equations. We also would like to note that this channel of screening is absent in the GW methods, which may lead to an error for metallic compounds.PRB05
(ii) The second channel describes the relaxation of the wavefunctions. It affects both the auxiliary WFs and the electronic states belonging to the rest of the spectrum.
### VII.1 Approximations and Simplifications
The usual way in calculating the parameter $`U`$ is the c-LDA approach, that is to solve iteratively Eqs. (1) and (2) for a fixed set of occupation numbers $`\{n_i\}`$, which does not necessarily follow the Fermi-Dirac distribution for the ground state.Dederichs ; Norman ; McMahan ; Hybertsen ; Gunnarsson ; GunnarssonPostnikov ; AnisimovGunnarsson ; PRB94.2 In practical calculations, these occupation numbers are controlled by an external potential $`\delta V_{\mathrm{ext}}(𝐫)`$, playing a role of Lagrange multipliers in the constrained density functional theory. In spite of many limitations for the strongly correlated systems, LDA is formulated as the ground-state theory. Therefore, there is a general belief that it should provide a good estimate for the total energy difference given by Eq. (23) and all other expressions which can be derived from Eq. (23) using usual arguments of DFT.
From the practical point of view, the basic difficulty of combining c-LDA with the WF method is the necessity to deal with relaxation of these WFs. This means that the auxiliary WFs should be recalculated on each iteration, for every new value of the electron density and the KS potential. Taking into account an arbitrariness with the choice of the WFs, this procedure cannot be easily implemented in the standard c-LDA calculations. Instead, we will employ a hybrid method, which starts with c-LDA and then takes into account the effects of relaxation of the WF in an analytical form, using the well-known expressions for the screened Coulomb interaction in the random-phase approximation (RPA).Springer ; Kotani ; Ferdi04 ; Hedin ; FerdiGunnarsson
In c-LDA calculations, the change of the occupation numbers $`\{\delta n_i\}`$ is associated with some change of the total potential $`\delta V`$$`=`$$`\delta V_{\mathrm{ext}}`$$`+`$$`\delta V_\mathrm{H}`$$`+`$$`\delta V_{\mathrm{XC}}`$. Then, the change of the electron density in Eq. (26) can be identically expressed in terms of the polarization function as
$$\delta \rho (𝐫)=𝑑𝐫^{}P(𝐫,𝐫^{})\delta V(𝐫^{}).$$
(27)
Using Eq. (26), one can identify three main contributions to the polarization function associated with the following processes (correspondingly $`P^I`$, $`P^{II}`$, and $`P^{III}`$):
1. the change of the occupation numbers of the auxiliary WFs;
2. the relaxation of the auxiliary WFs;
3. the relaxation of the rest of the electronic states.
Then, each $`\overline{W}`$ can be expressed in terms of the basis functions (or partial waves) $`\{\varphi \}`$ and $`\{\dot{\varphi }\}`$, using Eq. (15). Therefore, the change of $`\overline{W}`$ includes the relaxation of these basis functions as well as the change of hybridization of the TM $`t_{2g}`$ states with (mainly) the oxygen states. The latter is given by the change of coefficients $`\{\mathrm{\Gamma }_r\}`$ in the right-hand side of Eq. (15). The corresponding contributions to the polarization function are denoted as $`P^{IIB}`$ and $`P^{IIH}`$, which stand for the change of basis functions and hybridization, respectively. The same arguments are applied to relaxation of the rest of the electronic states. The corresponding polarization function can be divided accordingly in $`P^{IIIB}`$ and $`P^{IIIH}`$. We also introduce combined notations: $`P^B`$$`=`$$`P^{IIB}`$$`+`$$`P^{IIIB}`$ and $`P^H`$$`=`$$`P^{IIH}`$$`+`$$`P^{IIIH}`$.
We would like to point out here that, conceptually, the RPA approach for treating the relaxation effects is similar to c-LDA. The main difference is that RPA is based on an analytical expression for the change of the wavefunctions, formulated in terms of the perturbation theory expansion, while c-LDA treats the same effects numerically.PRB05 Therefore, we use a hybrid c-LDA+RPA scheme, which was originally considered in Ref. PRB05, . It consists of two steps.
(i) First, we take into account the screening associated with $`P^I`$, and $`P^B`$ in the framework of conventional c-LDA method, and neglect all kinds of hybridization effects involving the TM $`d`$-orbitals. An example of such a model electronic structure is shown in Fig. 16. This part is totally equivalent to the method of Gunnarsson and co-workers.Gunnarsson ; GunnarssonPostnikov ; AnisimovGunnarsson It allows to calculate the Coulomb repulsion $`u`$ and the intra-atomic exchange (Hund’s rule) coupling $`j`$$`=`$$``$$`2\delta ^2E/\delta 𝐦^2`$ in the atomic limit ($`𝐦`$ being the spin magnetization). By using these $`u`$ and $`j`$ one can construct the full $`5`$$`\times `$$`5`$$`\times `$$`5`$$`\times `$$`5`$ matrix $`\widehat{u}`$ of Coulomb interactions between atomic $`d`$ electrons, as it is typically done in the LDA$`+`$$`U`$ method.PRB94 ; comment.4 This matrix will be used as the starting point in RPA calculations.
c-LDA is supplemented with additional approximations, such as the atomic-spheres approximation. It also disregards some hybridization effects. However, in Sec. VII.3 we will see that at least for the static Coulomb interaction, the RPA results are close to the strong-coupling regime. In such a situation, the precise value of the parameter $`u`$, which is used as the starting point for these calculations, appears to be less important, and it is sufficient to have an “order of magnitude” estimate, which can be obtained from c-LDA.
(ii) We turn on the hybridization, and evaluate the screening associated with the last portion of the polarization function, $`P^H`$, in RPA:
$$\widehat{U}=\left[1\widehat{u}\widehat{P}^H\right]^1\widehat{u},$$
(28)
where $`\widehat{P}^H`$ is the $`5`$$`\times `$$`5`$$`\times `$$`5`$$`\times `$$`5`$ matrix $`\widehat{P}^H`$$``$$`P_{\alpha \beta \gamma \delta }^H`$, which will be specified below.
Since total $`P`$ is an additive function of $`P^I`$, $`P^B`$, and $`P^H`$, this procedure can be justified within RPA, where each new contribution to the polarization function ($`P^H`$) can be included consequently by starting with the Coulomb interaction $`\widehat{u}`$, which already incorporates the effects of other terms ($`P^I`$ and $`P^B`$).Ferdi04
The physical meaning of processes associated with the change of the hybridization and their role in the screening of local Coulomb interactions is illustrated schematically in Fig. 17.
Since the creation and the annihilation of an electron in RPA are treated as two independent processes, the screening of Coulomb interactions will be generally different from that associated with the true reaction $`2(d^n)`$$``$$`d^{n+1}`$$`+`$$`d^{n1}`$. To some extent, the true screening can be simulated by imposing certain constraints on the form of RPA polarization function $`P^H`$. Namely, one can expect certain cancellation of contributions coming from Wannier orbitals centered at different TM sites at the intermediate (i.e. oxygen) sites (see Fig. 18).
Therefore, we believe that it is more physical to take into account only those contributions to the polarization function which are associated with the TM sites, and to suppress contributions associated with intermediate sites. This makes some difference from the conventional RPA,Springer ; Kotani ; Ferdi04 which constitutes the basis of the GW method.Hedin ; FerdiGunnarsson We expect this scheme to work well for the on-site Coulomb interactions. However, the effect of hybridization on the intersite Coulomb interactions remain an open and so far unresolved problem. Some estimates of these effects will be given in Sec. VII.4, using the c-LDA method. The calculations suggest that the intersite interactions are screened very efficiently by the change of hybridization. Therefore we speculate that for the considered compounds, the effective Coulomb interactions between different TM sites are small and can be neglected.
The analytical expression for $`P^H`$ can be obtained from Eq. (27) by considering the perturbation-theory expansion for the wavefunctions with the fixed occupation numbers.PRB05 The time-dependent perturbation theory, corresponding to the external perturbation $`\delta V_{\mathrm{ext}}e^{i\omega t}`$, yields in the first order:
$$P_{\alpha \beta \gamma \delta }^H(\omega )=\underset{ij}{}\frac{(n_in_j)d_{\alpha j}^{}d_{\beta i}d_{\gamma i}^{}d_{\delta j}}{\omega \epsilon _j+\epsilon _i+i\delta (n_in_j)},$$
(29)
where $`d_{\gamma i}`$$`=`$$`\varphi _\gamma |\psi _i`$ is the projection of LDA eigenstate $`\psi _i`$ onto one of partial $`d`$-waves $`\varphi _\gamma `$, belonging to the TM site, and $`i`$ and $`j`$ are the joint index, incorporating spin and band indices as well as the position of $`𝐤`$-point in the first Brillouin zone. In this notations, the matrix multiplication in Eq. 28 implies the convolution over two indices. Namely, the matrix element of the product $`\widehat{u}\widehat{P}^H`$ is given by $`(\widehat{u}\widehat{P}^H)_{\alpha \beta \gamma \delta }`$$`=`$$`_{\mu \nu }u_{\alpha \beta \mu \nu }P_{\mu \nu \gamma \delta }^H`$. Note that all transitions in Eq. (29) are allowed only between occupied and empty bands.
In the next Sections, we will discuss different contributions to the screening of Coulomb interactions more in details.
### VII.2 Results of Constraint-LDA Calculations
The results of conventional constraint-LDA calculations for TM oxides have been widely discussed in the literature.AZA ; Norman ; McMahan ; Gunnarsson ; GunnarssonPostnikov Here we only illustrate the main idea and show some basic results using SrVO<sub>3</sub> as an example.
The calculations are performed in the supercell geometry, in which the number of atomic $`3d`$ electrons is modulated around the “ground-state” value $`n`$$`=`$$`1`$ according to the formula:
$$n_𝐑=n+\delta n\mathrm{cos}(\mathrm{𝐤𝐑}).$$
Corresponding values of interaction parameter $`u_𝐤`$, calculated in several different points of the Brillouin zone, are listed in Table 10.
Then, we map $`u_𝐤`$ onto the model
$$u_𝐤=uv\underset{𝐑}{}\mathrm{cos}(\mathrm{𝐤𝐑}),$$
and extract parameters of on-site ($`u`$) and NN ($`v`$) interactions after integration over the Brillouin zone. We note that in the present context the $`\mathrm{\Gamma }`$-point result has no physical meaning, as it corresponds to the transfer of an electron to the same atomic site. Therefore, we exclude it in the process of integration, and recalculate the weights of other $`𝐤`$-points using the symmetry arguments. The new weights are shown in Table 10. This yields the following parameters: $`u`$$`=`$$`10.1`$ eV and $`v`$$`=`$$`1.2`$ eV. Similar calculations for intra-atomic exchange coupling yield $`j`$$`=`$$`1.0`$ eV.
For comparison, the values of bare Coulomb and exchange integrals, calculated on the atomic V($`3d`$) wavefunctions are 21.7 and 1.2 eV, respectively. Thus, in the c-LDA scheme the effective Coulomb interaction is reduced by factor two. The intra-atomic exchange interaction is reduced by 20%. As we will see in the next section, the Coulomb interaction will be further reduced by relaxation of the WFs due the change of hybridization.
### VII.3 The Role of Hybridization
Because of hybridization, the $`d`$-states of the TM sites may have a significant weight in other bands. A typical situation for the series of TM oxides is shown in Fig. 1, where besides the $`t_{2g}`$-band, the $`d`$-states contribute to the TM $`e_g`$ as well as the O($`2p`$) bands. On both sides of the reaction $`2(d^n)`$$``$$`d^{n+1}`$$`+`$$`d^{n1}`$, the WFs constructed for the $`t_{2g}`$ bands should be orthogonal to other bands. As it was already pointed out in Sec. VII.1, this mechanism is responsible for an additional channel of screening of the on-site Coulomb interaction associated with the change of this hybridization. The corresponding contribution can be evaluated using the Dyson equation (28) and taking the matrix of Coulomb interactions $`\widehat{u}`$ obtained in c-LDA as the starting interaction. Then, the relevant expression for the polarization matrix is given by Eq. (29).
According to the electronic structure of the TM oxides, one can identify three main contributions to the polarization matrix $`\widehat{P}^H`$, associated with the following inter-band transitions: O($`2p`$)$``$TM($`e_g`$), O($`2p`$)$``$TM($`t_{2g}`$), and TM($`t_{2g}`$)$``$TM($`e_g`$). Meanwhile, all transitions between $`t_{2g}`$ bands should be switched off, in order to avoid the double counting of these effects in the process of solution of the Hubbard model.Ferdi04 As it was already pointed out in Sec. VII, this procedure is similar to the setting $`\widehat{h}`$$`=`$$`\widehat{c}`$ for the auxiliary WFs.
Details of static screening, corresponding to $`\omega `$$`=`$$`0`$, are explained in Fig. 19 for SrVO<sub>3</sub>. It is convenient to introduce three Kanamori parameters:Kanamori the intra-orbital Coulomb interaction $`𝒰`$$`=`$$`U_{xyxyxyxy}`$, the inter-orbital interaction $`𝒰^{}`$$`=`$$`U_{xyxyyzyz}`$, and the off-diagonal (exchange-type) interaction $`𝒥`$$`=`$$`U_{xyyzxyyz}`$.
In addition to the total value of $`U`$, we calculate intermediate interactions corresponding to each type of transitions in the polarization matrix (and neglecting the other two). The screening caused by the change of the hybridization appears to be very efficient. So, by going from c-LDA to RPA the intra-orbital interaction $`𝒰`$ is reduced from 11.2 to 2.5 eV (i.e., by factor four and even more). The main contribution to this screening comes from the O($`2p`$)$``$V($`e_g`$) and O$`(2p`$)$``$V($`t_{2g}`$) transitions in the polarization functions. In the cubic perovskites, the direct interaction between V($`t_{2g}`$) and V($`e_g`$) bands plays a minor role and can be neglected.
Matrix elements of the (total) polarization function are displayed in Fig. 20.
The largest contribution comes from the site-diagonal elements of the type $`P(\omega )`$$``$$`P_{\alpha \alpha \alpha \alpha }^H(\omega )`$$`=`$$`P_{\alpha \beta \beta \alpha }^H(\omega )`$. Other contributions are considerably smaller. The static polarization $`P`$$``$$`P(0)`$ is about $``$$`0.12`$ eV<sup>-1</sup>. This is the large value because the renormalization of the on-site Coulomb interaction in the multi-orbital systems is controlled by the parameter $`MP`$, rather than $`P`$ (see Appendix). The prefactor $`M`$ stands for the total number of orbitals per one TM site ($`M`$$`=`$$`3`$ for $`t_{2g}`$ systems). Therefore, $`uMP`$ can be estimates as $`3.6`$, and the situation appears to be close to the strong-coupling regime. Then, the effective interaction is not sensitive to the exact value of the parameter $`u`$, which is used as the starting point in RPA calculations. For example, had we started with the bare Coulomb interaction, which exceed the c-LDA value by factor two and even more, we would have obtained $`𝒰`$$`=`$$`2.7`$ eV, which is close to $`2.5`$ eV derived by starting with c-LDA. This justifies some approximations discussed in Sec. VII.1, particularly the use of fast but not extremely accurate c-LDA for some channels of screening.
### VII.4 RPA versus Constraint-LDA for $`t_{2g}`$ Electrons
In this section we briefly return to the problem considered in Ref. PRB96, and reinterpret some results obtained in that work in the light of present RPA approach. The basic idea of Ref. PRB96, was to evaluate the effective Coulomb interaction for the series of TM perovskite oxides in the framework of c-LDA, which would incorporate the screening by itinerant TM($`e_g`$) electrons. Since for the considered type of screening, RPA has many similarities with c-LDA, the present section can be also regarded as a test for these two approaches.
However, it is important to remember that several basic assumptions of Ref. PRB96, were different from the present work. So, the TM($`t_{2g}`$) states in Ref. PRB96, were totally decoupled from the rest of the electronic states by switching off the matrix elements of hybridization in the LMTO method, and only the TM($`e_g`$) states were allowed to hybridize. The corresponding electronic structure of SrVO<sub>3</sub> is shown in Fig. 21.
In terms of RPA polarization function, this means that the only allowed contributions to the screening in Ref. PRB96, were due to the O($`2p`$)$``$V($`e_g`$) inter-band transitions (type 1 in Fig. 19).
Results of such c-LDA calculations for SrVO<sub>3</sub>, which have been performed along the same line as in Sec. VII.2, are summarized in Table 11.
After the Fourier transformation to the real space, we obtain the following parameters of on-site and NN interactions: $`U`$$`=`$$`3.4`$ eV and $`V`$$`=`$$`0.3`$ eV. This value of $`U`$ appears to be in a reasonable agreement with the final $`U`$$`=`$$`2.5`$ eV, extracted from RPA (Fig. 19). However, c-LDA employs an additional atomic-spheres approximation. Therefore, for a proper comparison with RPA, one should use the same level of approximation and suppress all nonspherical interactions in the matrix of Coulomb interactions $`\widehat{u}`$, which is used as the starting point in RPA. In this approximation, and considering only the O($`2p`$)$``$V($`e_g`$) transitions in the polarization function, we obtain $`U`$$`=`$$`3.6`$ eV, which is close to the c-LDA value obtained using the method proposed in Ref. PRB96, . The small difference is caused by different approximations used for treating the intersite Coulomb interactions, which were neglected in RPA and taken into account in c-LDA. For comparison, the total value of $`U`$ obtained in RPA after neglecting the nonsphericity effects is only $`1.6`$ eV.
In summarizing this section, there is a reasonable agreement between results of RPA calculations and the c-LDA approach proposed in Ref. PRB96, . However, the agreement is somewhat fortuitous because this c-LDA takes into account only one part of the total screening, corresponding to the O($`2p`$)$``$TM($`e_g`$) transitions in the polarization function. The error caused by this approximation is partially compensated by the atomic-spheres approximation supplementing the c-LDA scheme.
The c-LDA calculations give some idea about the effect of hybridization on the screening of intersite Coulomb interactions. The screening appears to be very efficient. So, by taking into account only the O($`2p`$)$``$V($`e_g`$) transitions, the NN interaction is reduced from 1.2 to 0.3 eV. We expect this value to be further reduced by including other types of transitions in the polarization function. Thus, for the considered compounds, the effective Coulomb interaction between different TM sites seems to be small and can be neglected.
### VII.5 Doping-Dependence and Kanamori Rules for Cubic Compounds
In this section we discuss the effects of electron/hole doping on the static Coulomb interactions in SrVO<sub>3</sub> using the rigid-band approximation. Results of such c-LDA+RPA calculations are shown in Fig. 22 versus the total number of electrons in the TM($`t_{2g}`$) band, $`n_{t_{2g}}`$.
We monitor the behavior of three Kanamori parameters:Kanamori $`𝒰`$, $`𝒰^{}`$, and $`𝒥`$.
The Coulomb interactions reveal a monotonic behavior as the function of doping. The screening is the most efficient when the $`t_{2g}`$ band is empty ($`n_{t_{2g}}`$$`=`$$`0`$). The situation corresponds to SrTiO<sub>3</sub>. In this case all O($`2p`$)$``$TM($`t_{2g}`$) transitions contribute to the screening in RPA (see Fig. 19). This channel of screening vanishes when the $`t_{2g}`$ band becomes occupied ($`n_{t_{2g}}`$$`=`$$`6`$). Then, the only possible screening is associated with the O($`2p`$)$``$TM($`e_g`$) transitions in the polarization function, and the effective Coulomb interaction becomes large.
The screening of off-diagonal matrix element $`𝒥`$ practically does not depend on doping. Therefore, the well know Kanamori rule, $`𝒰`$$`=`$$`𝒰^{}`$$`+`$$`2𝒥`$, which was originally established for atoms, works well in the cubic compounds, even after the screening of $`t_{2g}`$ interactions by other electrons.
The present result also supports an old empirical rule suggesting that only the Coulomb integral $`U`$ is sensitive to the crystal environment in solids. The nonspherical interactions, which are responsible for Hund’s first and second rules, appears to be much closer to their atomic values.NormanBrooks ; MarelSawatzky
### VII.6 Frequency-Dependence
We have shown that the change of hybridization plays a very important role and strongly reduces the static value of $`U`$. However, this is only one part of the story because the same effect implies the strong frequency-dependence of the effective interaction, as it immediately follows from the Kramers-Kronig transformation in RPA:FerdiGunnarsson
$$\mathrm{Re}\widehat{U}(\omega )=\widehat{u}\frac{2}{\pi }𝒫_0^{\mathrm{}}𝑑\omega ^{}\frac{\omega ^{}|\mathrm{Im}\widehat{U}(\omega ^{})|}{\omega ^2\omega ^2}.$$
(30)
Indeed, the difference $`[\widehat{u}`$$``$$`\mathrm{Re}\widehat{U}(\omega )]`$ at $`\omega `$$`=`$$`0`$, which for the diagonal matrix elements of SrVO<sub>3</sub> is about 8.7 eV, should be related with the existence of the finite spectral weight of $`|\mathrm{Im}\widehat{U}(\omega )|`$ at finite $`\omega `$. These dependencies are shown in Fig. 23 for SrVO<sub>3</sub>.
The high-frequency part of $`\widehat{U}`$ can also contribute to the low-energy part of the spectrum through the self-energy effect.Ferdi04 The latter can be evaluated in the GW approximation,Hedin ; FerdiGunnarsson where the self-energy is given by the convolution of $`\widehat{U}(\omega )`$ with the one-particle Green function $`\widehat{G}(\omega )`$ for the $`t_{2g}`$ band:
$$\widehat{\mathrm{\Sigma }}(\omega )=\frac{i}{2\pi }𝑑\omega ^{}\widehat{G}(\omega +\omega ^{})\widehat{U}(\omega ^{}).$$
For SrVO<sub>3</sub>, the diagonal matrix element of $`\widehat{\mathrm{\Sigma }}(\omega )`$ is also shown in Fig. 23. The low-frequency part of $`\mathrm{Im}\widehat{\mathrm{\Sigma }}`$ is small and can be neglected, while $`\mathrm{Re}\widehat{\mathrm{\Sigma }}`$ mainly contributes to the renormalization factor:
$$Z_{\alpha \beta }=\left[1\mathrm{Re}\mathrm{\Sigma }_{\alpha \beta }/\omega |_{\omega =0}\right]^1.$$
The latter is estimated as $`0.8`$, for the diagonal matrix elements.
### VII.7 Lattice Distortion, Formal Valency and Screening
In Sec. VI.2 we already pointed out that the degree of hybridization between atomic TM($`t_{2g}`$) and O($`2p`$) states can differ substantially for different TM oxides. A typical example is two isoelectronic perovskites: SrVO<sub>3</sub> and YTiO<sub>3</sub>. The TM($`t_{2g}`$)-O($`2p`$) hybridization is stronger in SrVO<sub>3</sub>, because of two reasons:
(i) A direct proximity of O($`2p`$) and V($`t_{2g}`$) bands in SrVO<sub>3</sub>, which is expected for tetra-valent compounds;
(ii) A strong orthorhombic distortion observed in YTiO<sub>3</sub>, which generally deteriorates the Ti($`t_{2g}`$)-O($`2p`$) hybridization.
Therefore, it is reasonable to expect a very different screening of on-site Coulomb interactions in these two compounds. This idea is nicely supported by results of RPA calculations shown in Fig. 24.
The static $`U`$ is larger YTiO<sub>3</sub>. For example, the diagonal matrix element of $`\widehat{U}`$ is about 3.4 eV, against 2.5 eV in SrVO<sub>3</sub>. This is despite the fact that c-LDA has an opposite tendency. Generally, the value of $`u`$ in c-LDA is expected to be larger for SrVO<sub>3</sub> rather than for YTiO<sub>3</sub>, due to different shape of the atomic $`3d`$-wavefunctions in the three- and tetra-valent compounds.PRB94.2 So, c-LDA yields $`u`$$`=`$ 8.9 and 10.1 eV, correspondingly for YTiO<sub>3</sub> and SrVO<sub>3</sub>. These parameters have been used as the starting point in RPA, which results in an opposite trend for the static $`U`$.
Therefore, the RPA screening is more efficient in SrVO<sub>3</sub>, which is consistent with stronger TM($`t_{2g}`$)-O($`2p`$) hybridization in this compound. On the other hand, the frequency-dependence of $`\widehat{U}`$ is weaker in YTiO<sub>3</sub>, as it immediately follows from the Kramers-Kronig transformation (30).
Finally, because of different hybridization of the $`t_{2g}`$ orbitals in YTiO<sub>3</sub>, the diagonal matrix elements of $`\widehat{U}`$ are also different (see Fig. 24). In this case, there is some deviation from the Kanamori rules. The difference is small (about 0.07 eV for diagonal matrix elements of $`\widehat{U}`$ at $`\omega `$$`=`$$`0`$). However, it may play some role in more delicate applications, such as the orbital magnetism in solids, for example.NormanBrooks
### VII.8 Two Models for V<sub>2</sub>O<sub>3</sub>
In Sec. VI.4 we introduced two possible models for the kinetic-energy part of V<sub>2</sub>O<sub>3</sub>: the “five-orbital model” treats all V($`3d`$) bands on an equal footing, while the “three-orbital model” is limited by twelve V($`t_{2g}`$) bands, close to the Fermi level. We argued that parameters of the kinetic energy can be different for these two models. The same is true for the effective Coulomb interactions. The RPA provides a very transparent explanation for this difference, which is based on the following arguments. Recall that RPA incorporates the screening of on-site Coulomb interactions caused by relaxation of the wavefunctions. This relaxation is treated analytically, using regular perturbation theory for the wavefunctions, which results in Eq. (29) for the polarization function.PRB05
Since the V($`e_g`$) band is eliminated in the three-orbital model, the effective $`\widehat{U}`$ should include the screening caused by that change of the wavefunctions, which is formulated in terms of transitions between V($`t_{2g}`$) and V($`e_g`$) bands in the perturbation-theory expansion. In the five-orbital model, the V($`e_g`$) band is included explicitly. Therefore, the relaxation caused by possible interactions between V($`t_{2g}`$) and V($`e_g`$) bands will be automatically taken into account in the process of solution of the five-orbital model, and we should get rid of this parasitic screening at the stage the construction of the model Hamiltonian. Thus, the tree-orbital model will include an additional screening of on-site interactions, caused by the V($`t_{2g}`$)$``$V($`e_g`$) transitions in the polarization function, which does not appear in the five-orbital model.
This screening appears to be rather efficient (unlike in cubic perovskites considered in Sec VII.3). So, the diagonal matrix element of static $`t_{2g}`$ interactions is about 3.2 and 3.9 eV, for the three- and five-orbital model, respectively (Fig. 25). On the other hand, according to the Kramers-Kronig transformation (30), the frequency-dependence of the effective interaction is more important in the three-orbital model.
Because of different hybridization of $`t_{2g}`$ and $`e_g^\sigma `$ states in the five-orbital model, the $`e_g^\sigma `$ interactions are screened more efficiently (the static interaction between $`e_g^\sigma `$ orbitals is about 3.7 eV).
## VIII Summary, open questions, and comparison with other methods
The ultimate goal of this work was to make a bridge between first-principle electronic structure calculations and the universe of Hubbard parameters for strongly-correlated systems. We have presented a comprehensive analysis of the problem, by starting with the brief description of the ASA-LMTO method for electronic structure calculations and ending up with realistic parameters of the kinetic-energy and the Coulomb interactions for the series of TM oxides obtained on the basis of this LMTO method. A particular attention has been paid to the analysis of microscopic processes responsible for the screening of on-site Coulomb interactions in oxide compounds.
Our strategy consists of three steps:
(i) Derivation of the kinetic-energy part of the Hubbard model from the single-particle electronic structure in LDA, using the downfolding method. We have also considered corrections to the crystal-field splitting caused by nonsphericity of electron-ion interactions, beyond the conventional atomic-spheres-approximation.
(ii) Construction of the Wannier functions using results of the downfolding method. At this stage we closely follow the idea of LMTO method, and construct the WFs as the LMTO basis functions, which after applying to the Kohn-Sham Hamiltonian in the real space generate the matrix elements of the kinetic energy obtained in the downfolding method.
(iii) Calculation of screened Coulomb interactions using the concept of auxiliary WFs. The latter are defined as the Wannier orbitals for which the kinetic-energy part is set to be zero. This construction allows to avoid the double counting of the kinetic-energy term, which is included explicitly in the Hubbard model. The screened Coulomb interactions are calculated on the basis of a hybrid approach, combining the conventional constraint-LDA with the random-phase approximation for treating the hybridization effects between atomic TM($`3d`$) and O($`2p`$) orbitals. The latter play a very important role and yields a strong renormalization of the effective Coulomb interaction for isolated $`t_{2g}`$ band. It also explains a strong material-dependence of this interaction, which is sensitive to the crystal environment in solids, the number of $`t_{2g}`$ electrons, the valent state of the TM ions, etc.
Taking into account a wide interest to the construction of effective lattice fermion models from the first principles, we would like to make a brief comparison with other works on a similar subject.
Majority of methods start with the construction of the WFs, which are then used as the basis for calculations of the parameters of the kinetic energy and the Coulomb interactions. This is different from our approach, where we start with the kinetic-energy part, and only after that construct the WFs for a given set of parameters of the kinetic energy. We believe that such an order is extremely important, as it allows us to control the contributions of the kinetic energy to the WFs and the Coulomb interactions.
Among recent works, a considerable attention is paid to the method of Marzari and Vanderbilt, because it allows to control the spacial extension of the WFs. Very recently, Schnell et al. applied this method to calculations of the parameters of the Hubbard Hamiltonian for the series of $`3d`$ transition metals.Schnell In each $`𝐤`$-point of the Brillouin zone, they constructed the WFs from all 16 bands of the LDA Hamiltonian, corresponding to the $`4s4p3d4f`$ LMTO basis. Therefore, the total number of Wannier orbitals was also 16. The coefficients of this expansion has been chosen so to minimize the square of the position operator, $`𝐫^2`$$`=`$$`W|𝐫^2|W`$. The obtained WFs were indeed well localized, and the parameter $`U`$ estimated for the $`3d`$ bands was very close to the atomic value (about 25 eV). However, the corresponding Wannier basis set is too large, that does not make a big difference from the original LMTO basis set, for which one can also introduce a localized (tight-binding) representation.LMTO From the view point of numerical solution of the Hubbard model using modern many-body techniques, it is still hardly feasible to work in the basis of 16 Wannier orbitals per one TM site, while the simplest Hartree-Fock approximation is definitely not sufficient for the transition metals.Schnell It would be interesting to see how this method will work for the TM oxides, considered in the present work, where the physical basis set is limited by three Wannier orbitals per one TM site. For example, is it possible to construct the localized Wannier orbitals for isolated $`t_{2g}`$ bands in the TM oxides, which would be as good as the Wannier orbitals derived for all bands? The problem is that when the number of bands decreases, the number of variational parameters for the optimization of the Wannier orbitals will also decreases. Therefore, the Wannier orbitals will generally become less localized, as it was demonstrated in our work for V<sub>2</sub>O<sub>3</sub>. Also, when the problem is formulated in a reduced Hilbert space of bands closest to the Fermi level, it is very important to consider the screening of Coulomb interactions by other bands, which comes from relaxation of the wavefunctions. These effects are beyond the scopes of the work of Schnell et al., who only considered the bare Coulomb interactions.
In order to construct the WFs, Ku et al. employed the projector-operator scheme,WeiKu which is basically the initial step of the method of Marzari and Vanderbilt, prior the optimization.MarzariVanderbilt In this scheme, each Wannier orbital is generated by projecting a trial wavefunction, $`|g`$, onto a chosen subset of bands (in our case, $`t_{2g}`$ bands):
$$|W=\underset{it_{2g}}{}|\psi _i\psi _i|g.$$
Since such orbitals are not orthonormal, the procedure is followed by the numerical orthonormalization, similar to the one described in Sec. II. These WFs have been used as the basis for the construction of the low-energy Hamiltonian for the series of cuprates, like La<sub>4</sub>Ba<sub>2</sub>Cu<sub>2</sub>O<sub>10</sub>, where the Wannier basis consisted of a single orbital centered around each Cu site. A weak point of this approach is that it is difficult to assess the spacial extension of the WF, which crucially depends on the trial wavefunction, and can be affected by the orthonormalization. In some sense, the result strongly depends on authors’ intuition on how they choose the trial wavefunction. For example, the bare Coulomb interaction obtained in Ref. WeiKu, in the basis of their WFs was only 7.5 eV, which is much smaller than the atomic Coulomb integral for the Cu($`3d`$) orbitals. This means that the WFs are not well localized. It is not clear at present, whether this is a result of the bad choice of the trial wavefunction, or there is a more fundamental problem related with the fact that a more compact representation for the WFs simply may not exist in this case. Note, that apart from the Berry phase, there is no further parameters available for the optimization of the WFs in the single-orbital case.
The delocalization of the WFs gives rise to appreciable direct exchange interactions operating between different Cu sites.WeiKu This result, however, rises additional questions. Note, that the kinetic part of the Hubbard model and the WFs are evaluated in LDA, where the exchange-correlation potential is set to be local. In principle, non-local effects can be already incorporated in LDA, through the renormalization of parameters of the kinetic energy and the local interactions.PRB99 Therefore, it is not clear whether the nonlocal exchange interactions should be regarded as independent parameters of the Hubbard Hamiltonian or not.
Finally, Ku et al. calculated only bare Coulomb interactions. They did not consider the screening of these interactions caused by relaxation effects, which are extremely important.
Anisimov et. al. employed a similar approach for the analysis of spectroscopic properties of TM oxides.Anisimov2005 They extracted only the kinetic-energy part of the Hubbard Hamiltonian, using the WFs constructed in the LMTO basis, and treated the Coulomb interaction $`U`$ as a parameter.
A completely different strategy has been proposed by Andersen et. al., on the basis of their order-$`N`$ muffin-tin orbital (NMTO) method, which is an extension of the LMTO method.AndersenDasgupta From the very beginning, they construct the NMTO basis functions in certain energy interval as the WFs of the original KS Hamiltonian. Pavarini et al. applied this method to the series of $`d^1`$ perovskites.Pavarini At present, it is not clear how the nonuniqueness of the WFs is reflected in the construction of the NMTO basis set. Obviously, such a basis set is also not unique, and there is some freedom left for the localization of the Wannier orbitals, which does not seem to be well controlled. Generally, our transfer integrals for the $`d^1`$ perovskites seems to be more localized and our crystal-field splitting is smaller. For example, had we relaxed the constraint condition for the construction of the “heads” of the WFs, based on the diagonalization of the density matrix (19), our conclusion would have been also different: the crystal-field splitting would increase, but the transfer integrals would become less localized.
Pavarini et al. did not calculate the Coulomb interactions. Instead, they used $`U`$$``$$`5`$ eV as a parameter, with the reference to the photoemission data.MizokawaFujimori However, the photoemission data are typically interpreted in the cluster model, which treats the O($`2p`$) band explicitly. For the isolated $`t_{2g}`$ band, the effective interaction should include an additional renormalization coming from the relaxation of the O($`2p`$) band, which is eliminated in the $`t_{2g}`$-model. Therefore, the value of the effective $`U`$ should be smaller.
Finally, due to unknown for us reason, there is a substantial difference of the parameters of $`t_{2g}`$ bandwidth ($`W_{t_2g}`$) between our work and Ref. Pavarini, , even for cubic SrVO<sub>3</sub>. The parameters reported by Pavarini et al. are generally overestimated by about 30%.comment.5 Our conclusions about magnetic properties of YTiO<sub>3</sub> and LaTiO<sub>3</sub> are also different.PRB04 Details will be presented in a separate paper.
###### Acknowledgements.
I thank Professor Masatoshi Imada for valuable suggestions, especially for drawing my attention to the problem of Wannier functions, nonsphericity of electron-ion interactions and its effect on the crystal-field splitting,MochizukiImada and the random-phase approximation for calculating the effective interaction parameters in the Hubbard model.Ferdi04
*
## Appendix A Orbital degeneracy and screening in RPA
In this Appendix we consider the screening of on-site Coulomb interactions in RPA for an $`M`$-orbital system. All orbitals are supposed to be equivalent. For simplicity we neglect small nonsphericity of bare Coulomb interactions. Then, the nonvanishing matrix elements of the Coulomb interactions are $`u_{\alpha \alpha \beta \beta }u`$, where $`\alpha `$ ($`\beta `$)$`=`$$`1,\mathrm{},M`$. They can be presented in the form $`\widehat{u}`$$`=`$$`u\widehat{I}`$, where $`\widehat{I}`$ is the $`M`$$`\times `$$`M`$ matrix, consisting of only the units:
$$\widehat{I}=\left(\begin{array}{cccc}1& 1& \mathrm{}& 1\\ 1& 1& & \mathrm{}\\ \mathrm{}& & \mathrm{}& \mathrm{}\\ 1& \mathrm{}& \mathrm{}& 1\end{array}\right).$$
The part of the polarization polarization matrix (29), which can interact with the matrix $`\widehat{u}`$, is assumed to be diagonal: $`P_{\alpha \alpha \beta \beta }P\delta _{\alpha \beta }`$. The assumption is justified for cubic perovskites, where different $`t_{2g}`$ orbitals belong to different bands. For other compounds it can be regarded as an approximation, which does not change our qualitative conclusion. Hence, for the screened Coulomb interaction (28) we have:
$$\widehat{U}=[1\widehat{u}\widehat{P}]^1\widehat{u}=\underset{n=0}{\overset{\mathrm{}}{}}(uP)^n\widehat{I}^nu\widehat{I}.$$
(31)
Since $`\widehat{I}^{n+1}`$$`=`$$`M^n\widehat{I}`$, Eq. (31) can be converted to
$$\widehat{U}=\underset{n=0}{\overset{\mathrm{}}{}}(uMP)^nu\widehat{I}=\frac{u}{1uMP}\widehat{I}.$$
This means that in the multi-orbital systems, the renormalization of the Coulomb repulsion is more efficient as it is controlled by the quantity $`(MP)`$, where the prefactor $`M`$ stands for the number of orbitals.
In the strong coupling limit, $`uMP`$$``$$`1`$, the effective interaction is $`\widehat{U}`$$`=`$$``$$`(MP)^1\widehat{I}`$, which does not depend on the value of bare Coulomb interaction $`u`$.
|
warning/0506/math0506507.html
|
ar5iv
|
text
|
# On a class of algebras associated to directed graphs
## 0. Introduction
Factorizations of noncommutative polynomials play an important role in many areas of mathematics such as operator theory, integrable systems, and Yang-Baxter equations (see, for example, \[O, V\]). In this paper we use directed graphs and algebras associated with those graphs as a natural framework for studying such factorizations.
Let $`R`$ be an associative ring with unit and $`P(\tau )R[\tau ]`$ a polynomial over $`R`$, where $`\tau `$ is a central variable. Assume that $`P(\tau )`$ is a monic polynomial (i.e. the leading coefficient in $`P(\tau )`$ equals $`1`$). When $`R`$ is a (commutative) field there exists at most one (up to rearrangement of the factors) factorization of a monic polynomial $`P(\tau )`$ of degree $`n`$ into a product of linear polynomials:
$$P(\tau )=(\tau y_n)(\tau y_{n1})\mathrm{}(\tau y_1)$$
$`0.1`$
When the ring $`R`$ is not commutative, there may exist many factorizations of type (0.1).
In \[GGRSW\] the elements $`y_1,y_2,\mathrm{},y_n`$ in formula (0.1) where called the pseudoroots of the polynomial $`P(\tau )`$. The element $`y_1`$ is a right root of $`P(\tau )`$ and the element $`y_n`$ is a left root of $`P(\tau )`$.
Let $`X=\{x_1,x_2,\mathrm{},x_n\}R`$ be a generic set of right roots of $`P(\tau )`$ (meaning that if $`k2`$ and $`\{x_{i_1},x_{i_2},\mathrm{},x_{i_k}\}X`$, the corresponding Vandermonde matrix is invertible). It was shown in \[GR1\] (see also \[GR2, GGRW\]) that the right roots $`x_1,x_2,\mathrm{},x_n`$ define a set of pseudoroots $`x_{A,i}`$ of $`P(\tau )`$. Here $`A\{1,2,\mathrm{},n\}`$, $`i\{1,2,\mathrm{},n\}A`$, and $`x_{\mathrm{},j}=x_j`$ for all $`j`$.
According to \[GR1\] (see also \[GR2, GGRW\]) for any ordering $`i_1,i_2,\mathrm{},i_n`$ of $`1,2,\mathrm{},n`$ the pseudoroots $`y_k=x_{\{i_1,\mathrm{}i_{k1}\},i_k}`$, $`k=1,2,\mathrm{},n`$ define a factorization (0.1) .
The pseudoroots $`x_{A,i}`$ satisfy the following identities for any $`i,jA`$
$$x_{A\{i\},j}+x_{A,i}=x_{A\{j\},i}+x_{A,j}$$
$`0.2a`$
$$x_{A\{i\},j}x_{A,i}=x_{A\{j\},i}x_{A,j}.$$
$`0.2b`$
The paper \[GRW\] introduces and studies the algebra $`Q_n`$, called the universal algebra of pseudoroots, generated by elements $`x_{A,i}`$ satisfying the identities (0.2).
The study of $`Q_n`$ arose from the theory of quasideterminants and, specifically, from the noncommutative version of the Viète Theorem and its relations to the theory of noncommutative symmetric functions (see, for example, \[GGRW\]). In particular, the algebra $`Q_n`$ is quadratic, Koszul, and its dual algebra $`Q_n^!`$ has finite dimension (see \[GGRSW, SW\]).
A natural description of the algebra $`Q_n`$ (and, therefore, factorizations of noncommutative polynomials) can be given by using directed graphs. Let $`\mathrm{\Gamma }_n`$ be the Hasse graph corresponding to the lattice of subsets of the set $`\{1,2,\mathrm{},n\}`$. Vertices of $`\mathrm{\Gamma }_n`$ are subsets $`A\{1,2,\mathrm{},n\}`$. Edges of $`\mathrm{\Gamma }_n`$ are defined by pairs $`(A,i)`$ where $`i\{1,2,\mathrm{},n\}A`$. Such edges go from the vertex $`A\{i\}`$ to the vertex $`A`$.
The graph $`\mathrm{\Gamma }_n`$ possesses several properties. It is a hypercube with $`2^n`$ vertices. It is a layered graph: each vertex $`A`$ has a level $`|A|=\text{card}A`$. There is only one vertex $`\{1,2,\mathrm{},n\}`$ of level $`n`$ and only one vertex $``$ of level $`0`$ and any edge $`a`$ goes from a vertex of level $`r`$ to a vertex of level $`r1`$.
One can describe the relations in the algebra $`Q_n`$ by using geometric properties of the directed graph $`\mathrm{\Gamma }_n`$. Fix a field $`F`$. Let $`T`$ be the free associative algebra over $`F`$ generated by the edges of $`\mathrm{\Gamma }_n`$. Then the algebra $`Q_n`$ is a quotient algebra of $`T`$ modulo relations (0.2).
To every directed path $`\pi =(a_1,a_2,\mathrm{},a_m)`$ in $`\mathrm{\Gamma }_n`$ (where $`a_1,\mathrm{},a_m`$ are edges of $`\mathrm{\Gamma }_n`$) there is a corresponding polynomial $`P_\pi (\tau )T[\tau ]`$, $`P_\pi (\tau )=(\tau a_1)(\tau a_2)\mathrm{}(\tau a_m)`$. Denote the image of $`P_\pi (\tau )`$ in $`Q_n[\tau ]`$ by $`\stackrel{~}{P}_\pi (\tau )`$. If two paths $`\pi _1`$ and $`\pi _2`$ both go from a vertex $`v`$ to a vertex $`v^{}`$ then
$$\stackrel{~}{P}_{\pi _1}(\tau )=\stackrel{~}{P}_{\pi _2}(\tau )$$
$`0.3`$
in $`Q_n[\tau ]`$.
The identity (0.3) defines relations in $`Q_n`$. Note that the defining relations (0.2) arise from those pairs of paths forming a diamond, i. e. a path $`\pi _1`$ consisting of edges going from $`A\{i,j\}`$ to $`A\{i\}`$ and from $`A\{i\}`$ to $`A`$, and a path $`\pi _2`$ consisting of edges going from $`A\{i,j\}`$ to $`A\{j\}`$ and from $`A\{j\}`$ to $`A`$.
In this paper we introduce and study a class of algebras associated to certain directed graphs as a natural generalization of the geometric definition of the algebra $`Q_n`$. It is more convenient for us to use formulas similar to (0.3) by introducing a new variable $`t=\tau ^1`$.
Let $`\mathrm{\Gamma }=(V,E)`$ be a directed graph with vertices $`V`$ and edges $`E`$. Assume that $`\mathrm{\Gamma }`$ is layered and that the maximum level is $`n`$. Fix a field $`F`$ and denote by $`T(E)`$ the free associative algebra over $`F`$ generated by all edges. To any path $`\pi =(e_1,e_2,\mathrm{},e_k)`$ in $`\mathrm{\Gamma }`$ corresponds a polynomial
$$P_\pi (t)=(1te_1)(1te_2)\mathrm{}(1te_k)T(E)[t]/(t^{n+1}).$$
We define the algebra $`A(\mathrm{\Gamma })`$ to be the quotient algebra of $`T(E)`$ modulo relations defined by the equalities implied by
$$P_{\pi _1}(t)=P_{\pi _2}(t)$$
$`0.4`$
where the paths $`\pi _1`$ and $`\pi _2`$ both go from a vertex $`v`$ to a vertex $`v^{}`$.
The algebra $`A(\mathrm{\Gamma }_n)`$ coincides with the algebra $`Q_n`$.
Our main examples of directed graphs are the Hasse graphs associated with lattices of subsets and subspaces, abstract polytopes, complexes, and partitions.
We have already mentioned that the algebra $`Q_n`$, its subalgebras, quotient algebras, and dual algebras possess many interesting properties (see \[GRW, GGR, GGRSW, SW, Pi\]). We believe that the same is true for algebras associated to other directed graphs. In particular, we are planning to relate the structure of the algebra $`A(\mathrm{\Gamma })`$ with the geometry of the graph.
As a first step in these directions, we construct in this paper a linear basis for the algebra $`A(\mathrm{\Gamma })`$ for a large class of directed graphs $`\mathrm{\Gamma }`$. This basis has a “geometric” nature. In particular, our result simplifies the construction of the basis for $`Q_n`$ given in \[GRW\].
We now briefly describe this basis. For each vertex $`vV`$ with $`\text{level}|v|>0`$, we choose (arbitrarily) an edge $`e_v`$ beginning at the vertex $`v`$. This defines a unique path $`\pi _v`$ from $`v`$ to $``$. In $`A(\mathrm{\Gamma })`$ write $`\stackrel{~}{e}`$ for the image of $`eE`$ and
$$P_{\pi _v}(t)=\underset{k=0}{\overset{|v|}{}}\stackrel{~}{e}(v,k)t^k.$$
We say that two pairs $`(v,k),(w,m)V\times _0`$ can be composed (and write $`(v,k)(w,m)`$) if there is a path of length $`km`$ from $`v`$ to $`w`$. Let $`𝔹(\mathrm{\Gamma })`$ be the set of all sequences
$$𝕓=((b_1,m_1),(b_2,m_2),\mathrm{},(b_k,m_k))$$
where $`k0`$, $`b_1,b_2,\mathrm{},b_kV`$, $`1m_i|b_i|`$ for $`1ik`$, and $`(b_i,m_i)\vDash ̸(b_{i+1},m_{i+1})`$ for $`1i<k`$.
For $`𝕓=((b_1,m_1),\mathrm{},(b_k,m_k))𝔹(\mathrm{\Gamma })`$ set
$$\stackrel{~}{e}(𝕓)=\stackrel{~}{e}(b_1,m_1)\mathrm{}\stackrel{~}{e}(b_k,m_k).$$
###### Theorem 4.3
Let $`\mathrm{\Gamma }=(V,E)`$ be a layered graph, $`V=_{i=0}^nV_i,`$ and $`V_0=\{\}`$ where $``$ is the unique minimal vertex of $`\mathrm{\Gamma }`$. Then $`\{\stackrel{~}{e}(𝕓)|𝕓𝔹(\mathrm{\Gamma })\}`$ is a basis for $`A(\mathrm{\Gamma })`$.
An equivalent formulation of this theorem may be obtained by replacing each “generating function coefficient” $`\stackrel{~}{e}(v,k)`$ by the “monomial” $`\stackrel{~}{e}_{v^{(0)}}\stackrel{~}{e}_{v^{(1)}}\mathrm{}\stackrel{~}{e}_{v^{(k1)}}`$ where $`v^{(0)}`$, $`v^{(1)}`$, $`\mathrm{}`$, $`v^{(l)}=`$ are the vertices of the path $`\pi _v`$. (This monomial is the leading term of $`\stackrel{~}{e}(v,k)`$ in an appropriate filtration.) This “monomial” formulation of our basis theorem, when specialized to the graph $`\mathrm{\Gamma }_n`$, gives the basis theorem of \[GRW\] for $`Q_n`$.
## 1. The directed graph $`\mathrm{\Gamma }=(V,E)`$
Let $`\mathrm{\Gamma }=(V,E)`$ be a directed graph. That is, $`V`$ is a set (of vertices), $`E`$ is a set (of edges), and $`𝕥:EV`$ and $`𝕙:EV`$ are functions. ($`𝕥(e)`$ is the tail of $`e`$ and $`𝕙(e)`$ is the head of $`e`$.) We say that $`\mathrm{\Gamma }`$ is layered if $`V=_{i=0}^nV_i`$, $`E=_{i=1}^nE_i`$, $`𝕥:E_iV_i`$, $`𝕙:E_iV_{i1}`$. We will assume throughout the remainder of the paper that $`\mathrm{\Gamma }=(V,E)`$ is a layered graph with $`V=_{i=0}^nV_i`$, and $`V_0=\{\}`$ where $``$ is the unique minimal vertex of $`\mathrm{\Gamma }`$ (i.e., for every $`vV,v`$, there exists $`eE`$ with $`𝕥(e)=v`$).
For each $`v_{i=1}^nV_i`$ we will fix, arbitrarily, some $`e_vE`$, with $`𝕥(e)=v`$. If $`vV_i`$ we write $`|v|=i`$ and say that $`v`$ has level $`i`$. Similarly, if $`eE_i`$ we write $`|e|=i`$ and say that $`e`$ has level $`i`$.
If $`v,wV`$, a path from $`v`$ to $`w`$ is a sequence of edges $`\pi =\{e_1,e_2,\mathrm{},e_k\}`$ with $`𝕥(e_1)=v`$, $`𝕙(e_k)=w`$ and $`𝕥(e_{i+1})=𝕙(e_i)`$ for $`1i<k`$. We write $`v=𝕥(\pi )`$, $`w=𝕙(\pi )`$. We also write $`v>w`$ if there is a path from $`v`$ to $`w`$.
Let $`l(\pi )`$, the length of $`\pi `$, denote $`k`$, and let $`|\pi |`$, the level of $`\pi `$, denote $`|e_1|+\mathrm{}+|e_k|`$.
If $`\pi _1=\{e_1,\mathrm{},e_k\}`$,$`\pi _2=\{f_1,\mathrm{},f_l\}`$ are paths with $`𝕙(\pi _1)=𝕥(\pi _2)`$ then $`\{e_1,\mathrm{},e_k,f_1,\mathrm{},f_l\}`$ is a path; we denote it by $`\pi _1\pi _2`$.
For $`vV`$, write $`v^{(0)}=v`$ and define $`v^{(i+1)}=𝕙(e_{v^{(i)}})`$ for $`0i<|v|`$. Then $`v^{(|v|)}=`$ and $`\pi _v=\{e_{v^{(0)}},\mathrm{},e_{v^{(|v|1)}}\}`$ is a path from $`v`$ to $``$.
## 2. The filtered algebra $`T(E)`$
Let $`T(E)`$ denote the free associative algebra on $`E`$ over a field $`F`$. Define
$$T(E)_i=span\{e_1\mathrm{}e_r|r0,|e_1|+\mathrm{}+|e_r|i\}.$$
If $`aT(E)_i`$, $`aT(E)_{i1}`$, write $`|a|=i`$. For a path $`\pi =\{e_1,e_2,\mathrm{},e_k\}`$ define
$$P_\pi (t)=(1te_1)\mathrm{}(1te_k)T(E)[t]/(t^{n+1}).$$
Note that $`P_{\pi _1\pi _2}(t)=P_{\pi _1}(t)P_{\pi _2}(t)`$ if $`𝕙(\pi _1)=𝕥(\pi _2)`$. Write
$$P_\pi (t)=\underset{k=0}{\overset{l(\pi )}{}}(1)^ke(\pi ,k)t^k.$$
Set $`e(\pi ,k)=0`$ if $`k>l(\pi ).`$ For $`v_{i=1}^nV_i`$, set $`P_v(t)=P_{\pi _v}(t)`$ and $`e(v,k)=e(\pi _v,k).`$ Also, set $`P_{}(t)=1`$ and $`e(,k)=0`$ if $`k>0.`$
Definition 2.1 Let $`R`$ be the ideal in $`T(E)`$ generated by
$$\{e(\pi _1,k)e(\pi _2,k)|𝕥(\pi _1)=𝕥(\pi _2),𝕙(\pi _1)=𝕙(\pi _2),1kl(\pi _1)\}.$$
Note that this implies
$$P_{\pi _1}(t)P_{\pi _2}(t)modR[t].$$
Now assume $`v>u`$, so there is a path $`\pi `$ from $`v`$ to $`u`$. Then
$$𝕥(\pi \pi _u)=v=𝕥(\pi _v),$$
$$𝕙(\pi \pi _u)==𝕙(\pi _v)$$
and so
$$P_{\pi \pi _u}(t)P_{\pi _v}(t)modR[t].$$
But
$$P_{\pi \pi _u}(t)=P_\pi (t)P_{\pi _u}(t)=P_\pi (t)P_u(t),$$
$$P_{\pi _v}(t)=P_v(t)$$
so
$$P_\pi (t)P_v(t)P_u(t)^1modR[t].$$
Noting that $`P_\pi (t)`$ is a polynomial of degree $`l(\pi )=|v||u|`$ and writing $`(1a)^1=1+a+a^2\mathrm{}`$ for a nilpotent element $`a`$, we obtain
$$P_\pi (t)\underset{\genfrac{}{}{0pt}{}{r0,i_00,}{i_1,\mathrm{},i_r1}}{}(1)^{i_0+\mathrm{}+i_r+r}e(v,i_0)e(u,i_1)\mathrm{}e(u,i_r)t^{i_0+\mathrm{}+i_r}$$
$$\underset{j=0}{\overset{l(\pi )}{}}(\underset{\genfrac{}{}{0pt}{}{\genfrac{}{}{0pt}{}{r0,i_00,}{i_1,\mathrm{},i_r1,}}{i_0+\mathrm{}+i_r=j}}{}(1)^{j+r}e(v,i_0)e(u,i_1)\mathrm{}e(u,i_r))t^jmodR[t].$$
Let
$$H(v,u,j)=\underset{\genfrac{}{}{0pt}{}{\genfrac{}{}{0pt}{}{r0,i_00,}{i_1,\mathrm{},i_r1,}}{i_0+\mathrm{}+i_r=j}}{}(1)^{j+r}e(v,i_0)e(u,i_1)\mathrm{}e(u,i_r)$$
so that
$$P_\pi (t)\underset{j=0}{\overset{l(\pi )}{}}H(v,u,j)t^jmodR[t].$$
It follows that, modulo $`R[t]`$,
$$P_v(t)P_\pi (t)P_u(t)$$
$$(\underset{j=0}{\overset{l(\pi )}{}}H(v,u,j)t^j)(\underset{i_{r+1}=0}{\overset{|u|}{}}(1)^{i_{r+1}}e(u,i_{r+1})t^{i_{r+1}})$$
$$\underset{\genfrac{}{}{0pt}{}{\genfrac{}{}{0pt}{}{r0,i_0,i_{r+1}0,}{i_1,\mathrm{},i_r1,}}{\genfrac{}{}{0pt}{}{i_0+\mathrm{}+i_r|v||u|,}{i_0+\mathrm{}+i_{r+1}|v|}}}{}(1)^{i_0+\mathrm{}+i_{r+1}+r}e(v,i_0)e(u,i_1)\mathrm{}e(u,i_{r+1})t^{i_0+\mathrm{}+i_{r+1}}modR[t].$$
Setting $`k=|v||u|`$ and comparing coefficients of $`t^{k+l}`$ gives
$$e(v,k+l)e(v,k)e(u,l)+$$
$$\underset{\genfrac{}{}{0pt}{}{\genfrac{}{}{0pt}{}{r0,i_0,i_{r+1}0,}{i_1,\mathrm{},i_r1,}}{\genfrac{}{}{0pt}{}{i_0<k,i_0+\mathrm{}+i_rk}{i_0+\mathrm{}+i_{r+1}=k+l}}}{}(1)^re(v,i_0)e(u,i_1)\mathrm{}e(u,i_{r+1})modR.$$
$`2.1`$
Writing
$$E(v,u,k,l)=\underset{\genfrac{}{}{0pt}{}{\genfrac{}{}{0pt}{}{r0,i_0,i_{r+1}0,}{i_1,\mathrm{},i_r1,}}{\genfrac{}{}{0pt}{}{i_0<k,i_0+\mathrm{}+i_rk}{i_0+\mathrm{}+i_{r+1}=k+l}}}{}(1)^re(v,i_0)e(u,i_1)\mathrm{}e(u,i_{r+1})$$
we obtain
$$e(v,k+l)e(v,k)e(u,l)E(v,u,k,l)modR$$
when $`v>u`$, $`|v||u|=k.`$
We also have
###### Lemma 2.1
$`E(v,u,1,l)=e(u,1)e(u,l)+e(u,l+1)`$.
Proof: Setting $`k=1`$ in the sum defining $`E(v,u,k,l)`$ gives $`i_0=0,r=0`$ (hence $`i_1=l+1`$) or $`i_0=0,r=1`$ (hence $`i_1=l,i_2=l`$). These two choices give the two terms on the right-hand side.
Note that $`|e(v,k)|=k|v|k(k1)/2`$, and so, if $`k=|v||u|`$,
$$|e(v,k)e(u,l)|=|e(v,k)|+|e(u,l)|$$
$$=k|v|k(k1)/2+l|u|l(l1)/2$$
$$=(k+l)|v|(k+l)(k+l1)/2$$
$$=|e(v,k+l)|.$$
We also have
###### Lemma 2.2
If $`|v||u|=k`$, $`|E(v,u,k,l)|<|e(v,k+l)|`$.
Proof: It is sufficient to show that each summand in the expression for $`E(v,u,k,l)`$ belongs to $`T(E)_{|e(v,k+l)|1}.`$ Thus we must show
$$i_0|v|+(k+li_0)|u|i_0(i_01)/2\mathrm{}i_{r+1}(i_{r+1}1)/2<(k+l)|v|(k+l)(k+l1)/2,$$
where $`r0,i_0,i_{r+1}0`$, $`i_1,\mathrm{},i_r1`$, $`i_0+\mathrm{}+i_rk`$, $`i_0<k`$ and $`i_0+\mathrm{}+i_{r+1}=k+l`$. Since $`|u|=|v|k`$ this is equivalent to
$$(k+li_0)k(i_0^2+\mathrm{}+i_{r+1}^2kl)/2<(k+l)(k+l1)/2$$
which may be simplified to $`(ki_0)^2+\mathrm{}+i_{r+1}^2>l^2.`$ This holds since $`i_{r+1}l`$ and $`i_0<k.`$
###### Lemma 2.3
a) If $`fE`$ then $`fe(𝕥(f),1)+e(𝕙(f),1)R`$.
b) If $`v>u`$, $`|u|=|v|1`$ then $`e(v,1)e(u,k)e(v,k+1)+e(u,k+1)e(u,1)e(u,k)R`$.
Proof: a) $`P_{𝕥(f)}(t)P_{f\pi _{𝕙(f)}}(t)=(1tf)P_{𝕙(f)}(t)modR[t],`$ so $`e(𝕥(f),1)f+e(𝕙(f),1)modR`$.
b) Since $`v>u`$ and $`|u|=|v|1`$ there is $`fE`$ with $`𝕥(f)=v`$, $`𝕙(f)=u`$. Then
$$P_v(t)P_{f\pi _{𝕙(f)}}(t)=(1tf)P_u(t)$$
$$(1+te(v,1)te(u,1))P_u(t)modR[t].$$
Thus
$$e(v,k+1)e(u,k+1)e(v,1)e(u,k)+e(u,1)e(u,k)modR.$$
###### Lemma 2.4
$`e(v,1)E(u,w,k,l)e(u,1)E(u,w,k,l)E(v,w,k+1,l)E(u,w,k+1,l)`$ $`modR.`$
Proof: By the definition, the left-hand side (LHS) is
$$(1)^re(v,1)e(u,i_0)e(w,i_1)\mathrm{}e(w,i_{r+1})(1)^re(u,1)e(u,i_0)e(w,i_1)\mathrm{}e(w,i_{r+1}),$$
where the sums are over $`r0`$, $`i_0,i_{r+1}0`$, $`i_1,\mathrm{},i_r1`$, $`i_0<k`$, $`i_0+\mathrm{}+i_rk`$ and $`i_0+\mathrm{}+i_{r+1}=k+l.`$
By Lemma 2.3, $`e(v,1)e(u,i_0)e(u,1)e(u,i_0)e(v,i_0+1)e(u,i_0+1)modR`$. Thus the LHS is congruent to
$$(1)^re(v,i_0)e(w,i_1)\mathrm{}e(w,i_{r+1})(1)^re(u,i_0)e(w,i_1)\mathrm{}e(w,i_{r+1}),$$
where the summation is over all $`r0`$, $`i_01`$, $`i_{r+1}0`$, $`i_1,\mathrm{},i_r1`$, $`i_0<k+1`$, $`i_0+\mathrm{}+i_rk+1`$ and $`i_0+\mathrm{}+i_{r+1}=k+l+1.`$
But this is equal to the same expression where the sum is over all $`r0`$, $`i_00`$, $`i_{r+1}0`$, $`i_1,\mathrm{},i_r1`$, $`i_0<k+1`$, $`i_0+\mathrm{}+i_rk+1`$ and $`i_0+\mathrm{}+i_{r+1}=k+l+1`$ (since $`e(v,0)=e(u,0)=1.`$) This is the right hand side of the asserted congruence.
###### Lemma 2.5
$`R`$ is generated by all $`e(\pi _1,k)e(\pi _2,k)`$, $`𝕥(\pi _1)=𝕥(\pi _2)`$, $`𝕙(\pi _1)=𝕙(\pi _2)=`$.
Proof: Let $`S`$ be the ideal generated by all such elements. Thus, for such $`\pi _1`$, $`\pi _2`$,
$$P_{\pi _1}(t)P_{\pi _2}(t)modS[t].$$
Let $`P_{\pi _3}(t),P_{\pi _4}(t)T(E)[t]/(t^{n+1})`$ satisfying $`𝕥(\pi _3)=𝕥(\pi _4)`$, $`𝕙(\pi _3)=𝕙(\pi _4)=w`$. Then $`𝕥(\pi _3\pi _w)=𝕥(\pi _3)=𝕥(\pi _4)=𝕥(\pi _4\pi _w)`$, $`𝕙(\pi _3\pi _w)=𝕙(\pi _w)==𝕙(\pi _4\pi _w)`$ and
$$P_{\pi _3}(t)P_{\pi _w}(t)=P_{\pi _3\pi _w}(t)P_{\pi _4\pi _w}(t)=P_{\pi _4}(t)P_{\pi _w}(t)modS[t].$$
Then since $`P_{\pi _w}(t)`$ is invertible, $`P_{\pi _3}(t)P_{\pi _4}(t)modS[t]`$. Since the coefficients of all $`P_{\pi _3}(t)P_{\pi _4}(t)`$ generate $`R`$, we have the result.
###### Lemma 2.6
Let $`S_1=\{fe(𝕥(f),1)+e(𝕙(f),1)|fE\}`$ and
$$S_2=\{e(v,1)e(u,k)e(v,k+1)+e(u,k+1)+e(u,1)e(u,k)|u,vV,v>u,|u|=|v|1>0\}.$$
Then $`S_1S_2`$ generates $`R`$.
Proof: By Lemma 2.3, $`S_1S_2R.`$ Let $`vV,|v|>0`$ and let $`\pi =\{e_1,e_2,\mathrm{},e_{|v|}\}`$ be a path from $`v`$ to $``$. By lemma 2.5, it is sufficient to show that $`P_\pi (t)P_v(t)mod(S_1S_2)[t].`$ We proceed by induction on $`|v|.`$ If $`|v|=1`$, then $`\pi =\{e_1\}`$ where $`𝕥(e_1)=v,𝕙(e_1)=.`$ Then $`P_\pi (t)=1te_11+e(v,1)e(,1)1te_vP_v(t)modS_1[t].`$
Now assume $`|v|>1`$ and write $`\pi =e_1\pi ^{}`$ with $`𝕥(e_1)=v,𝕙(e_1)=u.`$ Then $`|u|<|v|`$ and so by induction we have $`P_u(t)P_\pi ^{}(t)mod(S_1S_2)[t].`$ Since
$$P_v(t)(1te(v,1)+te(u,1))P_u(t)modS_2[t]$$
and
$$1te(v,1)+te(u,1)1te_1modS_1[t]$$
we have
$$P_v(t)(1te_1)P_u(t)(1te_1)P_\pi ^{}(t)P_\pi (t)mod(S_1S_2)[t],$$
as required.
## 3. The algebra $`A(\mathrm{\Gamma })`$
Let
$$A(\mathrm{\Gamma })=T(E)/R$$
and
$$A(\mathrm{\Gamma })_i=(T(E)_i+R)/R.$$
This gives $`A(\mathrm{\Gamma })`$ the structure of a filtered algebra. Let $`\stackrel{~}{}`$ denote the canonical homomorphism $`\stackrel{~}{}:T(E)T(E)/R=A(\mathrm{\Gamma })`$ and write $`\stackrel{~}{e}(v,k`$), $`\stackrel{~}{E}(v,u,k,l)`$, etc., for the images of the elements $`e(v,k)`$, $`E(v,u,k,l)`$, etc. If $`aA(\mathrm{\Gamma })_i,aA(\mathrm{\Gamma })_{i1},`$ write $`|a|=i.`$
Note that
$$\{\stackrel{~}{e}(v,k)|v_{i=1}^nV_i,k|v|\}$$
generates $`A(\mathrm{\Gamma })`$, since by Lemma 2.3a), $`\stackrel{~}{f}=\stackrel{~}{e}(𝕥(f),1)\stackrel{~}{e}(𝕙(f),1)`$ for any $`fE`$.
We now develop some notation for products of the $`\stackrel{~}{e}(v,k)`$.
We say that a pair $`(v,k)`$, $`vV`$, $`0k|v|`$, can be composed with the pair $`(u,l)`$, $`uV`$, $`0l|u|`$, if $`v>u`$ and $`|u|=|v|k`$. If $`(v,k)`$ can be composed with $`(u,l)`$ we write $`(v,k)(u,l)`$.
Let $`𝔹_1(\mathrm{\Gamma })`$ be the set of all sequences
$$𝕓=((b_1,m_1),(b_2,m_2),\mathrm{},(b_k,m_k))$$
where $`k0`$, $`b_1,b_2,\mathrm{},b_kV`$, $`0m_i|b_i|`$ for $`1ik`$. Let $`\mathrm{}`$ denote the empty sequence. Define $`|𝕓|,`$ the level of $`𝕓`$, by
$$|𝕓|=\underset{i=1}{\overset{k}{}}\{m_i|b_i|\frac{m_i(m_i1)}{2}\}.$$
If $`1sk`$ write $`𝕓^s=((b_s,m_s),\mathrm{},(b_k,m_k))`$. Write $`𝕓^{k+1}=\mathrm{}.`$
If $`𝕓=((b_1,m_1),(b_2,m_2),\mathrm{},(b_k,m_k))`$ and $`𝕔=((c_1,n_1),(c_2,n_2),\mathrm{},(c_s,n_s))𝔹_1(\mathrm{\Gamma })`$ define
$$𝕓𝕔=((b_1,m_1),(b_2,m_2),\mathrm{},(b_k,m_k),(c_1,n_1),(c_2,n_2),\mathrm{},(c_s,n_s)).$$
Let
$$\genfrac{}{}{0pt}{}{𝔹(\mathrm{\Gamma })=\{𝕓=((b_1,m_1),(b_2,m_2),\mathrm{},(b_k,m_k))𝔹_1(\mathrm{\Gamma })|}{(b_i,m_i)\vDash ̸(b_{i+1},m_{i+1}),1i<k\}.}$$
For
$$𝕓=((b_1,m_1),(b_2,m_2),\mathrm{},(b_k,m_k))𝔹_1(\mathrm{\Gamma })$$
set
$$\stackrel{~}{e}(𝕓)=\stackrel{~}{e}(b_1,m_1)\mathrm{}\stackrel{~}{e}(b_k,m_k).$$
Then $`|\stackrel{~}{e}(𝕓)|=|𝕓|`$ and $`\stackrel{~}{e}(𝕓𝕔)=\stackrel{~}{e}(𝕓)\stackrel{~}{e}(𝕔).`$ Clearly $`\{\stackrel{~}{e}(𝕓)|𝕓𝔹_1(\mathrm{\Gamma })\}`$ spans $`A(\mathrm{\Gamma })`$. The following lemma is immediate from (2.1) and the definition of $`E(v,u,k,l).`$
###### Lemma 3.1
If $`(v,k)(u,l)`$ then $`\stackrel{~}{e}(v,k)\stackrel{~}{e}(u,l)=\stackrel{~}{e}(v,k+l)\stackrel{~}{E}(v,u,k,l)`$.
For $`𝕓=((b_1,m_1),(b_2,m_2),\mathrm{},(b_k,m_k))𝔹(\mathrm{\Gamma })`$ define
$$z(v,k,𝕓)=min\{\{j|(v,k+m_1+\mathrm{}+m_{j1})\vDash ̸(b_j,m_j)\}\{k+1\}\}.$$
###### Lemma 3.2
Let $`𝕓𝔹(\mathrm{\Gamma })`$. Then $`\stackrel{~}{e}(v,k)\stackrel{~}{e}(𝕓)=`$
$$\stackrel{~}{e}((v,k+m_1+\mathrm{}+m_{z(v,k,𝕓)1)})𝕓^{z(v,k,𝕓)})$$
$$\underset{j=1}{\overset{z(v,k,𝕓)1}{}}\stackrel{~}{E}(v,b_j,k+m_1+\mathrm{}+m_{j1},m_j)\stackrel{~}{e}(𝕓^{j+1}).$$
Proof: For $`1iz(v,k,𝕓)1`$ , we have $`(v,k+m_1+\mathrm{}+m_{i1})(b_i,m_i)`$ and so, by Lemma 3.1, $`\stackrel{~}{e}(v,k+m_1+\mathrm{}+m_{i1})\stackrel{~}{e}(𝕓^i)=\stackrel{~}{e}((v,k+m_1+\mathrm{}+m_{i1}+m_i)𝕓^{i+1})E(v,b_i,k+\mathrm{}+m_{i1},m_i).`$ It follows, by induction on $`i`$, that for $`1iz(v,k,𝕓)1`$
$$\stackrel{~}{e}(v,k)\stackrel{~}{e}(𝕓)=$$
$$\stackrel{~}{e}((v,k+m_1+\mathrm{}+m_{i1}+m_i)𝕓^{i+1})$$
$$\underset{j=1}{\overset{i}{}}\stackrel{~}{E}(v,b_j,k+m_1+\mathrm{}+m_{j1},m_j)\stackrel{~}{e}(𝕓^{j+1}).$$
Taking $`i=z(v,k,𝕓)1`$ gives the lemma.
###### Corollary 3.3
$`S=\{\stackrel{~}{e}(𝕓)|𝕓𝔹(\mathrm{\Gamma })\}`$ spans $`A(\mathrm{\Gamma }).`$
Proof: Since $`1S`$, it is sufficient to show that $`spanS`$ is invariant under multiplication by $`\stackrel{~}{e}(v,k)`$, hence sufficient to show that $`\stackrel{~}{e}(v,k)\stackrel{~}{e}(𝕓)spanS`$, if $`𝕓𝔹(\mathrm{\Gamma })`$. We prove this by induction on $`|\stackrel{~}{e}(v,k)\stackrel{~}{e}(𝕓)|`$. If $`|\stackrel{~}{e}(v,k)\stackrel{~}{e}(𝕓)|=0`$, then $`\stackrel{~}{e}(v,k)\stackrel{~}{e}(𝕓)F1spanS`$. Now assume that $`\stackrel{~}{e}(w,l)\stackrel{~}{e}(𝕔)spanS`$ whenever $`𝕔𝔹(\mathrm{\Gamma })`$ and $`|\stackrel{~}{e}(w,l)\stackrel{~}{e}(𝕔)|<|\stackrel{~}{e}(v,k)\stackrel{~}{e}(𝕓)|.`$ In view of Lemma 2.2, this implies $`\stackrel{~}{E}(v,b_1,k+m_1+\mathrm{}+m_{j1},m_j)\stackrel{~}{e}(𝕓^{j+1})S`$ for $`1jz(v,k,𝕓)1`$. But by Lemma 3.2,
$$\stackrel{~}{e}(v,k)\stackrel{~}{e}(𝕓)=$$
$$\stackrel{~}{e}((v,k+m_1+\mathrm{}+m_{z(v,k,𝕓)1})𝕓^{z(v,k,𝕓)})$$
$$\underset{j=1}{\overset{z(v,k,𝕓)1}{}}\stackrel{~}{E}(v,b_j,k+m_1+\mathrm{}+m_{j1},m_j)\stackrel{~}{e}(𝕓^{j+1}).$$
Since $`(v,k+m_1+\mathrm{}+m_{z(v,k,𝕓)1)}𝕓^{z(v,k,𝕓)}𝔹(\mathrm{\Gamma }),`$ every summand on the right-hand side of this expression belongs to $`S`$, so $`\stackrel{~}{e}(v,k)\stackrel{~}{e}(𝕓)S`$, as required.
## 4. Independence Theorem
Define $`B`$ to be the vector space over $`F`$ with basis $`𝔹(\mathrm{\Gamma })`$. Let
$$B_i=span\{𝕓||𝕓|i\}.$$
We will define a linear transformation
$$\mu :T(E)BB$$
giving $`B`$ the structure of a $`T(E)`$ module.
First define
$$\mu _1:E\times 𝔹(\mathrm{\Gamma })B$$
by
$$\mu _1:(f,\mathrm{})e(𝕥(f),1)$$
for $`fE_1`$ (where $`\mathrm{}`$ denotes the empty sequence) and
$$\mu _1:E_i\times B_h0$$
if $`i+h>1.`$ Then, as $`T(E)`$ is the free algebra on the set $`E`$, $`\mu _1`$ extends to a linear transformation, again denoted $`\mu _1`$ from $`T(E)B`$ to $`B`$ giving $`B`$ the structure of a $`T(E)`$\- module.
Now assume $`s>1`$ and that we have defined maps
$$\mu _j:E\times 𝔹(\mathrm{\Gamma })B$$
for $`1js1`$ such that
$$\mu _j:E_i\times 𝔹(\mathrm{\Gamma })_hB_{i+h}$$
for all $`1in`$ and $`h0`$. Assume also that
$$\mu _j:E_i\times 𝔹(\mathrm{\Gamma })_h0$$
whenever $`i+hj`$ and that
$$\mu _j^{}|_{E_i\times 𝔹(\mathrm{\Gamma })_h}=\mu _{j^{\prime \prime }}|_{E_i\times 𝔹(\mathrm{\Gamma })_h}$$
whenever $`j^{}j^{\prime \prime }i+h.`$ As in the case of $`\mu _1`$, each $`\mu _j`$ extends to a linear transformation, again denoted $`\mu _j`$ from $`T(E)B`$ to $`B`$ giving $`B`$ the structure of a $`T(E)`$-module.
We now define
$$\mu _s:E\times 𝔹(\mathrm{\Gamma })B$$
by
$$\mu _s|_{E_i\times 𝔹(\mathrm{\Gamma })_h}=\mu _{s1}|_{E_i\times 𝔹(\mathrm{\Gamma })_h}$$
for $`i+hs`$, and
$$\mu _s:(f,𝕓)(v,1+m_1+\mathrm{}+m_{z(v,1,𝕓)1})𝕓^{z(v,1,𝕓)}$$
$$\underset{j=1}{\overset{z(v,1,𝕓)1}{}}E(v,b_j,1+m_1+\mathrm{}+m_{j1},m_j)𝕓^{j+1}$$
$`4.1`$
$$(u,1+m_1+\mathrm{}+m_{z(u,1,𝕓)1})𝕓^{z(u,1,𝕓)}$$
$$+\underset{j=1}{\overset{z(u,1,𝕓)1}{}}E(u,b_j,k+m_1+\mathrm{}+m_{j1},m_j)𝕓^{j+1}$$
for $`𝕓=((b_1,m_1),\mathrm{},(b_k,m_k)),𝕥(f)=v,𝕙(f)=u,|f|+|𝕓|=s.`$ (Note that, as $`e(,j)=0`$ for $`j>0`$, the last two summands vanish when $`|f|=1`$.)
Thus we have inductively defined $`\mu _j`$ for all $`j`$. Define
$$\mu :T(E)BB$$
by
$$\mu |_{T(E)_iB_h}=\mu _{i+h}|_{T(e)_iB_h}.$$
To simplify the notation we write $`f𝕓`$ for $`\mu (f𝕓).`$
###### Lemma 4.1
(a) $`e(v,k)𝕓=(v,k+m_1+\mathrm{}+m_{z(v,k,𝕓)1})𝕓^{z(v,k,𝕓)}`$
$$\underset{j=1}{\overset{z(v,k,𝕓)1}{}}E(v,b_j,k+m_1+\mathrm{}+m_{j1},m_j)𝕓^{j+1}.$$
(b) $`RB=(0)`$
Proof: Note that (a) is clear if $`k=0`$ (for $`e(v,0)=1`$ and $`z(v,0,𝕓)=1`$). Also (4.1) shows that (a) holds when $`k=1`$ (as $`e(v,1)=e_{v^{(0)}}+e_{v^{(1)}}+\mathrm{}+e_{v^{(|v|1)}}`$ and so the expresion given by (4.1) for $`e(v,1)𝕓`$ is a telescoping series). Thus (a) holds whenever $`|e(v,k)|+|𝕓|1.`$
Let $`R_i=RA(\mathrm{\Gamma })_i`$ and note that $`R_0=(0)`$ and, by Lemma 2.6, $`R_1`$ is spanned by $`\{fe(𝕥(f),1)|fE_1\}.`$ Hence $`R_1B_0`$ is spanned by $`\{f\mathrm{}e(𝕥(f),1)\mathrm{}|fE_1\}=\{0\},`$ so $`R_iB_h=(0)`$ whenever $`i+h1.`$
Now assume $`s>1`$ and that
$$e(w,l)𝕔=(w,k+n_1+\mathrm{}+n_{z(w,l,𝕔)1})𝕔^{z(w,l,𝕔)}$$
$$\underset{j=1}{\overset{z(w,l,𝕔)1}{}}E(w,c_j,l+n_1+\mathrm{}+n_{j1},n_j)𝕔^{j+1}$$
whenever $`𝕔=((c_1,n_1)\mathrm{}(c_p,c_p))𝔹(\mathrm{\Gamma })`$. Assume also that $`|e(w,l)|+|𝕔|<s`$ and $`R_iB_h=(0)`$ whenever $`i+h<s`$. We will show that if $`|e(v,k)|+|𝕓|=s`$, then (a) holds and that if $`i+h=s`$ then $`R_iB_h=(0)`$, thus proving the lemma by induction.
By the expression obtained for $`e(v,1)𝕓`$, we have
$$(fe(𝕥(f),1)+e(𝕙(f),1))𝕓=0$$
for all $`fE,𝕓𝔹(\mathrm{\Gamma }).`$ Thus, by Lemma 2.6, to show $`R_iB_h=(0)`$ whenever $`i+h=s`$ it is sufficient to show that
$$(e(v,1)e(u,k)e(v,k+1)+e(u,k+1)+e(u,1)e(u,k))𝕓=(0)$$
whenever $`|e(v,k)|+|𝕓|=s.`$
Now assume that $`|e(v,k)|+|𝕓|=s`$. Since $`|e(u,k1)|<|e(v,k)|`$ and $`|e(u,k)|<|e(v,k)|`$, the induction assumption gives values for $`e(u,k1)𝕓`$ and $`e(u,k)𝕓`$. Thus
$$e(v,1)e(u,k1)𝕓+e(u,k)𝕓e(u,1)e(u,k1)𝕓$$
$$=e(v,1)(u,k1+m_1+\mathrm{}+m_{z(u,k1,𝕓)1})𝕓^{z(u,k1,𝕓)}$$
$$\underset{j=1}{\overset{z(u,k1,𝕓)1}{}}e(v,1)E(u,b_j,k1+m_1+\mathrm{}+m_{j1},m_j)𝕓^{j+1}$$
$$+(u,k+m_1+\mathrm{}+m_{z(u,k1,𝕓)1})𝕓^{z(u,k,𝕓)}$$
$$\underset{j=1}{\overset{z(u,k1,𝕓)1}{}}e(v,1)E(u,b_j,k+m_1+\mathrm{}+m_{j1},m_j)𝕓^{j+1}$$
$$e(u,1)(u,k1+m_1+\mathrm{}+m_{z(u,k1,𝕓)1})𝕓^{z(u,k1,𝕓)}$$
$$+\underset{j=1}{\overset{z(u,k1,𝕓)1}{}}e(u,1)E(u,b_j,k1+m_1+\mathrm{}+m_{j1},m_j)𝕓^{j+1}$$
By the definition of the actions of $`e(v,1)`$ and $`e(u,1)`$ this becomes
$$(v,k+m_1+\mathrm{}+m_{z(v,k,𝕓)1})𝕓^{z(u,k1,𝕓)}$$
$$E(v,u,1,k1+m_1+\mathrm{}+m_{z(v,k,𝕓)1})𝕓^{z(u,k1,𝕓)}$$
$$\underset{j=z(u,k1,𝕓)1}{\overset{z(v,k,𝕓)1}{}}E(v,b_j,k+m_1+\mathrm{}+m_{j1},m_j)𝕓^{j+1}$$
$$\underset{j=1}{\overset{z(u,k1,𝕓)1}{}}e(v,1)E(u,b_j,k1+m_1+\mathrm{}+m_{j1},m_j)𝕓^{j+1}$$
$$+(u,k+m_1+\mathrm{}+m_{z(u,k,𝕓)1})𝕓^{z(u,k,𝕓)}$$
$$\underset{j=1}{\overset{z(u,k,𝕓)1}{}}E(u,b_j,k+m_1+\mathrm{}+m_j)𝕓^{j1}$$
$$(u,1)(u,k1+m_1+\mathrm{}+m_{z(u,k1,𝕓)1})𝕓^{z(u,k1,𝕓)}$$
$$+\underset{j=1}{\overset{z(u,k1,𝕓)1}{}}e(u,1)E(u,b_j,k1+m_1+\mathrm{}+m_{j1},m_j)𝕓^{j+1}.$$
Set
$$G=e(v,1)E(u,b_j,k1+m_1+\mathrm{}+m_{j1},m_j)$$
$$+e(u,1)E(u,b_j,k1+m_1+\mathrm{}+m_{j1},m_j)$$
$$+E(v,b_j,k+m_1+\mathrm{}+m_{j1},m_j)$$
$$E(u,b_j,k+m_1+\mathrm{}+m_j).$$
By Lemma 2.6, $`GR`$ and by Lemma 2.2 $`|G𝕓^{j+1}|<s`$. Hence by the induction assumption, $`G𝕓^{j+1}=0`$. Thus we may replace
$$e(v,1)E(u,b_j,k1+m_1+\mathrm{}+m_{j1},m_j)𝕓^{j+1}$$
$$e(u,1)E(u,b_j,k1+m_1+\mathrm{}+m_{j1},m_j)𝕓^{j+1}$$
by
$$E(v,b_j,k+m_1+\mathrm{}+m_{j1},m_j)𝕓^{j+1}$$
$$E(u,b_j,k+m_1+\mathrm{}+m_j)𝕓^{j+1}.$$
Thus our expression becomes
$$(v,k+m_1+\mathrm{}+m_{z(v,k,𝕓)1})𝕓^{z(u,k1,𝕓)}$$
$$E(v,u,1,k1+m_1+\mathrm{}+m_{z(v,k,𝕓)1})𝕓^{z(u,k1,𝕓)}$$
$$\underset{j=1}{\overset{z(v,k,𝕓)1}{}}E(v,b_j,k+m_1+\mathrm{}+m_{j1},m_j)𝕓^{j+1}$$
$$+(u,k+m_1+\mathrm{}+m_{z(u,k,𝕓)1})𝕓^{z(u,k,𝕓)}$$
$$(u,1)(u,k1+m_1+\mathrm{}+m_{z(u,k1,𝕓)1})𝕓^{z(u,k1,𝕓)}$$
But by Lemma 2.1
$$E(v,u,1,k1+m_1+\mathrm{}+m_{z(v,k,𝕓)1})$$
$$=e(u,1)e(u,k1+m_1+\mathrm{}+m_{z(v,k,𝕓)1})+e(u,k+m_1+\mathrm{}+m_{z(v,k,𝕓)1}).$$
Since
$$|E(v,u,1,k1+m_1+\mathrm{}+m_{z(v,k,𝕓)1})|+|𝕓^{z(u,k1,𝕓)}|<s,$$
by Lemma 2.2, and (as $`|u|<|v|`$)
$$|e(u,1)e(u,k1+m_1+\mathrm{}+m_{z(v,k,𝕓)1})|+|𝕓^{z(u,k1,𝕓)}|<s,$$
and
$$|e(u,k+m_1+\mathrm{}+m_{z(v,k,𝕓)1})|+|𝕓^{z(u,k1,𝕓)}|<s,$$
the induction assumption allows us to replace
$$E(v,u,1,k1+m_1+\mathrm{}+m_{z(v,k,𝕓)1})𝕓^{z(u,k1,𝕓)}$$
by
$$(u,1)(u,k1+m_1+\mathrm{}+m_{z(u,k1,𝕓)1})𝕓^{z(u,k1,𝕓)}$$
$$+(u,k+m_1+\mathrm{}+m_{z(u,k,𝕓)1})𝕓^{z(u,k,𝕓)}.$$
Making this substitution gives
$$e(v,1)e(u,k1)𝕓+e(u,k)𝕓e(u,1)e(u,k1)𝕓$$
$$=(v,k+m_1+\mathrm{}+m_{z(v,k,𝕓)1})𝕓^{z(v,k,𝕓)}$$
$`4.2`$
$$\underset{j=1}{\overset{z(v,k,𝕓)1}{}}E(v,b_j,k+m_1+\mathrm{}+m_{j1},m_j)𝕓^{j+1}.$$
Since $`P_v(t)=(1+te(v,1)te(v^{(1)},1))P_{v^{(1)}}(t)`$ we have
$$e(v,k)=e(v,1)e(v^{(1)},k1)+e(v^{(1)},k)e(v^{(1)},1)e(v^{(1)},k1).$$
Thus setting $`u=v^{(1)}`$ in (4.2) gives (a). Since the right-hand side of (4.2) is independent of $`u`$ we also obtain
$$e(v,k)𝕓=e(v,1)e(u,k1)𝕓+e(u,k)𝕓e(u,1)e(u,k1)𝕓$$
for all $`u`$ with $`v>u,|u|=|v|1.`$ As noted above, this completes the proof of (b).
###### Corollary 4.2
There is an action of $`A(\mathrm{\Gamma })`$ on $`B`$ satisfying
$$\stackrel{~}{e}(v,k)𝕓=(v,k+m_1+\mathrm{}+m_{z(v,k,𝕓)1})𝕓^{z(v,k,𝕓)}$$
$$\underset{j=1}{\overset{z(v,k,𝕓)1}{}}\stackrel{~}{E}(v,b_j,k+m_1+\mathrm{}+m_{j1},m_j)𝕓^{j+1}.$$
Consequently, for $`𝕓𝔹(\mathrm{\Gamma })`$, $`e(𝕓)1=𝕓`$. Since $`𝔹(\mathrm{\Gamma })B`$ is linearly independent, $`\{\stackrel{~}{e}(𝕓)|𝕓𝔹(\mathrm{\Gamma })\}`$ is linearly independent. Therefore, we have proved the following theorem:
###### Theorem 4.3
Let $`\mathrm{\Gamma }=(V,E)`$ be a layered graph, $`V=_{i=0}^nV_i,`$ and $`V_0=\{\}`$ where $``$ is the unique minimal vertex of $`\mathrm{\Gamma }`$. Then $`\{\stackrel{~}{e}(𝕓)|𝕓𝔹(\mathrm{\Gamma })\}`$ is a basis for $`A(\mathrm{\Gamma })`$.
Note that
$$\{\stackrel{~}{e}(𝕓)||\stackrel{~}{e}(𝕓)|i\}$$
is a basis for $`A(\mathrm{\Gamma })_i`$. Therefore, writing $`\overline{e}(𝕓)+A(\mathrm{\Gamma })_{i1}grA(\mathrm{\Gamma })`$ where $`|\stackrel{~}{e}(𝕓)|=i`$ we have:
###### Corollary 4.4
Let $`\mathrm{\Gamma }=(V,E)`$ be a layered graph, $`V=_{i=0}^nV_i,`$ and $`V_0=\{\}`$ where $``$ is the unique minimal vertex of $`\mathrm{\Gamma }`$. Then $`\{\overline{e}(𝕓)|𝕓𝔹(\mathrm{\Gamma })\}`$ is a basis for $`grA(\mathrm{\Gamma }).`$
Also, if $`|\stackrel{~}{e}(v,k)|=i,`$ we have
$$\stackrel{~}{e}(v,k)+A(\mathrm{\Gamma })_{i1}=\stackrel{~}{e}_{v^{(0)}}\stackrel{~}{e}_{v^{(1)}}\mathrm{}\stackrel{~}{e}_{v^{(k1)}}+A(\mathrm{\Gamma })_{i1}.$$
Write
$$\widehat{e}(v,k)=\widehat{e}_{v^{(0)}}\widehat{e}_{v^{(1)}}\mathrm{}\widehat{e}_{v^{(k1)}}$$
and, for $`𝕓=((v_1,k_1),\mathrm{},(v_s,k_s))𝔹(\mathrm{\Gamma }),`$ write
$$\widehat{e}(𝕓)=\widehat{e}(v_1,k_1)\mathrm{}\widehat{e}(v_s,k_s).$$
Then, if $`|\stackrel{~}{e}(𝕓)|=i`$, it follows that
$$\stackrel{~}{e}(𝕓)+A(\mathrm{\Gamma })_{i1}=\widehat{e}(𝕓)+A(\mathrm{\Gamma })_{i1}$$
and so we have:
###### Corollary 4.5
Let $`\mathrm{\Gamma }=(V,E)`$ be a layered graph, $`V=_{i=0}^nV_i,`$ and $`V_0=\{\}`$ where $``$ is the unique minimal vertex of $`\mathrm{\Gamma }`$. Then $`\{\widehat{e}(𝕓)|𝕓𝔹(\mathrm{\Gamma })\}`$ is a basis for $`A(\mathrm{\Gamma }).`$
Note that, if $`\mathrm{\Gamma }=\mathrm{\Gamma }_n`$, the basis for $`Q_n=A(\mathrm{\Gamma }_n)`$ given by Corollary 4.5 is the basis constructed in \[GRW\].
|
warning/0506/cond-mat0506727.html
|
ar5iv
|
text
|
# Conductivity magnetooscillations in 2D electron-impurity system under microwave irradiation: role of magnetoplasmons
## Abstract
It is developed a many-electron approach to explain the recently observed conductivity magnetooscillations in very high mobility 2D electron systems under microwave irradiation. For the first time a theory takes into account the microwave-induced renormalization of the screened impurity potential. As a result this potential has singular, dynamic and non-linear in electric field nature. That changes the picture of scattering of electrons at impurities in a “clean” 2D system essentially: for appearence of the rectified dissipative current responsible are excitations of 2D magnetoplasmons rather than one-electron transitions between Landau levels. In a “dirty” 2D system the role of electron-electron interaction diminishes, so the collective excitations cease to exist, and our results turn into the well-known ones, which were obtained in the one-electron approach.
For a high quality 2D electron system in structures GaAs/AlGaAs subjected to microwave (MW) field with frequency $`\mathrm{\Omega }`$ it was found that the magnetoresistance experienced oscillations governed with the ratio $`\mathrm{\Omega }/\omega _c`$ , where $`\omega _c=eB/(m^{}c)`$ is the cyclotron frequency. The states with zero resistance were observed with an increase of MW field intensity . These observations have been confirmed by other researchers, see the review . That brought about an avalanche of theoretical works.
There exist two mainstream theoretical scenarios of the effect, both being one-electron. The first one is based on the mechanism of electron displacement against strong external DC field as a result of MW absorption and impurity scattering, and this was shown to be capable of leading to the absolute negative DC conductivity . Being then unstable, the system breaks into domains, and one just registers zero resistance . The second scenario is based on the wave-induced inversion of electron population on higher Landau levels (LL) . Indisputable explanation of the main experimental data has not been achieved yet.
In this work we consider the effect of electron-electron (e-e) interaction on the impurity scattering confining ourselves to the first scenario. Here we analyse the case of an unbounded high-quality 2D electron system with very weak impurity scattering. The main obvious consequence of e-e interaction is the screening of the impurity potential with 2D electrons. At first glance it seems that e-e interaction is not able to induce a qualitative change in the results of Ref. . But it is shown to be a delusion. With that, for the very appearence of the dissipative direct current responsible are not the usually considered single-particle transitions of electrons between LL, but rather 2D magnetoplasmons. Below in the framework of the random phase approximation (RPA) it is developed a systematic theory of the non-liner dissipative conductivity applicable to the experimental conditions .
Dealing with our system, let us change the reference frame to the one connected with the external homogenious electric field. In such a reference frame electrons do not experience the external electric field if no impurity is available in the system. Being presented and so transformed, the bare potential of the impurity system becomes time-dependent:
$$V_{\mathrm{imp}}\left(𝐫\right)V_{\mathrm{imp}}\left(𝐫𝐫_0\left(t\right)\right),$$
(1)
where $`𝐫_0\left(t\right)`$ is the radius-vector describing movement of the center of the classical oscillator in external electric field. To screen the transformed potential, which is the right-hand side of the transform (1), one should use the dynamic dielectric function. The space-time Fourier transform of the screened potential is
$$V_{\mathrm{imp}}^{(\mathrm{scr})}(𝐪,\omega )=\frac{V_{\mathrm{imp}}(𝐪,\omega )}{\epsilon (q,\omega )},$$
(2)
where $`V_{\mathrm{imp}}(𝐪,\omega )`$ is the Fourier image of the right-hand side of the transform (1), and $`\epsilon (q,\omega )`$ is the dielectric function. Being obtained in RPA, it has the form:
$`\epsilon (q,\omega )=1+{\displaystyle \frac{V_{\mathrm{ee}}(q)}{\pi \mathrm{}\lambda ^2}}{\displaystyle \underset{M,M^{}}{}}{\displaystyle \frac{\left(f_Mf_M^{}\right)I_{M,M^{}}\left(q\right)}{\omega _c\left(M^{}M\right)+\omega +i0}},`$
where $`V_{\mathrm{ee}}(q)`$ is the Fourier transform of potential of e-e interaction, for 2D electron gas in a medium with the constant lattice dielectric permeability $`\kappa `$ we have $`V_{\mathrm{ee}}(q)=2\pi e^2/(\kappa q)`$, $`\lambda =\sqrt{\mathrm{}c/\left(eB\right)}`$ is the magnetic length, $`f_M`$ is the Fermi distribution function, $`M`$ and $`M^{}`$ are LL indices. And $`I_{M,M^{}}\left(q\right)`$ is the square of the absolute value of the overlap integral of the Landau functions with the oscillator centers shifted by $`q\lambda ^2`$. At the magnetoplasmon frequency $`\omega =\omega _{\mathrm{MP}}(q)`$ the denominator in Eq. (2) turns to zero, so the screening in the strong field does not ordinarily soften the impurity potental, but rather sharply strengthen it. Let the external homogeneous electric field $`𝐅`$ be a sum of AC field of the wave with the amplitude $`W`$ and DC dragging field $`F_{\mathrm{DC}}`$, both having only one ($`x`$) component for simplicity:
$`F_x=F_{\mathrm{DC}}+W\mathrm{sin}\mathrm{\Omega }t.`$
The current density in the system is $`𝐣=en_\mathrm{s}\mathrm{Tr}(\rho 𝐯)`$, where $`n_\mathrm{s}`$ is 2D electron concentration, $`𝐯`$ is the velocity operator, $`\rho `$ is the density matrix that meets the quantum kinetic equation. Following Ref. we solve the kinetic equation at low order in scattering of electrons at the screened impurities. After taking an average of all chaotic impurity configurations, asuming only one type of impurity with 2D concentration $`n_{\mathrm{imp}}`$ and the bare single impurity potential $`V_{\mathrm{imp}}^{(0)}\left(q\right)`$, we have for the time average of the dissipative current density:
$`j_x`$ $`=`$ $`{\displaystyle \frac{en_{\mathrm{imp}}}{\left(2\pi \right)^2m^{}\omega _c}}{\displaystyle \mathrm{d}^2q\frac{V_{\mathrm{imp}}^{(0)}(q)^2}{V_{\mathrm{ee}}(q)}q_y}`$ (3)
$`\times {\displaystyle \underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}}J_n^2(Q)\mathrm{Im}\epsilon ^1(q,q_yv_\mathrm{H}+n\mathrm{\Omega }),`$
where $`J_n`$ is the Bessel function, $`Q=\left(Q_x^2+Q_y^2\right)^{1/2}`$,
$`Q_x={\displaystyle \frac{q_xeW}{m^{}(\omega _c^2\mathrm{\Omega }^2)}},Q_y={\displaystyle \frac{q_yeW\omega _c}{m^{}\mathrm{\Omega }(\omega _c^2\mathrm{\Omega }^2)}},`$
$`v_\mathrm{H}=cF_{\mathrm{DC}}/B`$ is the Hall velocity. If we neglect the collision-induced LL broading,
$$\mathrm{Im}\frac{1}{\epsilon (q,q_yv_\mathrm{H}+n\mathrm{\Omega })}=\pi \underset{p}{}\frac{\delta \left(q_yv_\mathrm{H}+n\mathrm{\Omega }\omega _p\right)}{\epsilon _\omega ^{}(q,\omega _p)},$$
(4)
where $`\omega _p=\omega _p\left(q\right)`$, index $`p=\pm 1,\pm 2,\mathrm{}`$ enumerates all solutions to the dispersion equation $`\epsilon (q,\omega _p)=0`$, so that $`\omega _pp\omega _c`$ as $`q\mathrm{}`$, and $`\epsilon _\omega ^{}(q,\omega _p)=\mathrm{d}\epsilon (q,\omega )/\mathrm{d}\omega |_{\omega =\omega _p}`$.
When no MW field is given, $`W=0`$, only the term with $`n=0`$ survives in the sum of (3), $`J_0(0)=1`$. In such a form our result, which generalizes the one of Ref. obtained with leaving e-e interaction out, is applicable to explanation of the experiment .
With the help of polar coordinates in Eq. (3): $`q_x=q\mathrm{cos}\varphi `$, $`q_y=q\mathrm{sin}\varphi `$, using Eq. (4) and integrating by $`\mathrm{d}\varphi `$ we obtain the expression that allows graphical analysis:
$`j_x`$ $`=`$ $`{\displaystyle \frac{en_{\mathrm{imp}}}{2\pi m^{}\omega _c}}{\displaystyle _0^+\mathrm{}}dq{\displaystyle \frac{V_{\mathrm{imp}}^{(0)}(q)^2}{V_{\mathrm{ee}}(q)}}{\displaystyle \underset{p}{}}{\displaystyle \frac{q^2}{\epsilon _\omega ^{}(q,\omega _p)}}`$
$`\times {\displaystyle \underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}}J_n^2(\overline{Q}_p){\displaystyle \frac{\omega _pn\mathrm{\Omega }}{qv_\mathrm{H}}}{\displaystyle \frac{1}{\sqrt{q^2v_\mathrm{H}^2(\omega _pn\mathrm{\Omega })^2}}}`$
$`\times \left(\mathrm{\Theta }\left({\displaystyle \frac{\omega _pn\mathrm{\Omega }}{qv_\mathrm{H}}}+1\right)\mathrm{\Theta }\left({\displaystyle \frac{\omega _pn\mathrm{\Omega }}{qv_\mathrm{H}}}1\right)\right).`$
Here $`\mathrm{\Theta }`$ is the Heaviside step-function, and
$`\overline{Q}_p={\displaystyle \frac{qeW}{m^{}(\omega _c^2\mathrm{\Omega }^2)}}\sqrt{1+\left({\displaystyle \frac{\omega _c^2}{\mathrm{\Omega }^2}}1\right)\left({\displaystyle \frac{\omega _pn\mathrm{\Omega }}{qv_\mathrm{H}}}\right)^2},`$
In Fig. 1 shown are the spectrum of the principal magnetoplasmon (at $`p=1`$ and $`\omega _p(q)>0`$), two lines ($`\omega =n\mathrm{\Omega }\pm qv_\mathrm{H}`$) forming a region that confines all values of $`\omega _p`$ and so $`q`$ contributing to the integral (see the last line of Eq. (Conductivity magnetooscillations in 2D electron-impurity system under microwave irradiation: role of magnetoplasmons)), and the bisector ($`\omega =n\mathrm{\Omega }`$) parting that region onto two ones contributing purely positive or negative.
Let $`F_{\mathrm{DC}}>0`$ be weak enough (say $`v_\mathrm{H}<\omega _c/2k_\mathrm{F}`$, where $`\mathrm{}k_\mathrm{F}`$ is the Fermi momentum), so that the term with $`n=0`$ in the sum of Eq. (3) does not play a role with its always positive contribution. And let us consider one-photon processes only: $`n=\pm 1`$ in the sum of Eq. (3). Then $`j_x>0`$ if $`\mathrm{\Omega }<\omega _c`$, that meets the positive magnetoresistance. Other case, if $`\omega <\mathrm{\Omega }<2\omega _c`$, a bunch of magnetoplasmon modes may fall into the region of negative contribution. That gives the absolute negative conductivity. Similar picture holds for higher values $`\mathrm{\Omega }`$, and the higher magnetoplasmon modes $`|p|>1`$ take part in the play. In contrast to theories omitting the screening , the regions of positive and negative conductivities are finite even in an ideal case of no LL broading (4), and anyhow small $`F_{\mathrm{DC}}`$ be.
In a “dirty” 2D system the role of electron-electron interaction diminishes. It is somewhat equivalent to $`V_{\mathrm{ee}}0`$. Then in the vicinity of $`\omega =\omega _p(q)p\omega _c`$ we may use the approximate expression for $`\epsilon (q,\omega )`$:
$`\epsilon _p(q,\omega )=1+{\displaystyle \frac{2m^{}V_{\mathrm{ee}}(q)}{\pi \mathrm{}^2}}{\displaystyle \frac{p^2\omega _c^2}{p^2\omega _c^2\omega ^2i0\mathrm{s}\mathrm{i}\mathrm{g}\mathrm{n}\omega }}\overline{I}_p\left(q\right),`$
where
$`\overline{I}_p\left(q\right)=p^1{\displaystyle \underset{M=0}{\overset{\mathrm{}}{}}}\left(f_Mf_{M+p}\right)I_{M,M+p}\left(q\right).`$
Then $`V_{\mathrm{ee}}`$ falls out, and with condition (4), Eq. (3) transforms to
$`j_x`$ $`=`$ $`{\displaystyle \frac{en_{\mathrm{imp}}}{\left(2\pi \mathrm{}\right)^2}}{\displaystyle \mathrm{d}^2qV_{\mathrm{imp}}^{(0)}(q)^2q_y\underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}J_n^2(Q)}`$
$`\times {\displaystyle \underset{p}{}}p\overline{I}_p\left(q\right)\delta (q_yv_\mathrm{H}+n\mathrm{\Omega }p\omega _c),`$
which is the result of Ref. for ideal infinitely narrow LL, and the result of Ref. for the case of DC field only.
The work was supported by RFBR and RAS programme.
|
warning/0506/physics0506144.html
|
ar5iv
|
text
|
# Liénard–Wiechert solution revisited
## 1 Introduction
### 1.1 Preliminaries
In this paper we consider Maxwellian electromagnetic fields in the flat Minkowski spacetime with the metric tensor $`g_{\mu \nu }=\text{diag}(+1,1,1,1)`$, thus taking Cartesian coordinates (algebraic relations used or deduced here, frequently remain unaltered also in the framework of general relativity). Greek indices are four- and Latin, three-dimensional. However we more frequently use as three-dimensional quantities four-dimensional vectors (tensors) orthogonal to the timelike unit vector describing the reference frame (the monad). Different frames may be simultaneously applied (the test-object property which is essential in treatment of reference frames in non-quantum theory). In general, we do not mutually relate reference frames and systems of coordinates. A comma ($`_,`$) followed by an index is used to denote partial differentiation with respect to the corresponding coordinate. We also use natural units in which the velocity of light in a vacuum is $`c=1`$. Round brackets mean symmetrization and square brackets, antisymmetrization in the indices contained in them (the so-called Bach brackets). In concrete calculations no approximations are assumed.
This material is essentially the final chapter of my unpublished one-semester course “Relativistic Physics” given to undergraduate (Licenciatura) students at the Physics Department of the University of Guadalajara during the last seven years. The students first have to attend another course on tensor calculus which also includes the formalism of Cartan forms with some applications in physics. The subsection 4.3 was included in my course only in the last semester.
### 1.2 Preview of the paper
More than one hundred years ago, A. Liénard and E. Wiechert discovered an exact solution of Maxwell’s equations describing electromagnetic field of a pointlike electric charge in an arbitrary motion. A frequently used treatment of this solution can be found in , and its more general deduction, including the use of an arbitrary mixture of retarded and advanced potentials, in . In section 2 we consider a simple and direct deduction of the Liénard–Wiechert (below abbreviated as LW) solution with the use of the light cone concept which involves a supposition of lightlike propagation of information from this pointlike source. Some important general properties of the LW solution are discussed in section 3. Here a general classification of electromagnetic fields is outlined, and it is found that the LW field belongs to the pure electric type, thus its magnetic part can be transformed away when one passes to certain non-inertial reference frames. It is well known that in a vacuum electromagnetic waves propagate with the fundamental velocity $`c`$ ($`=1`$). However, as it is shown in section 4, a mixture of non-radiative and radiative electromagnetic fields has another propagation velocity ($`<1`$). For this reason, when we speak above and in sections 2 and 5 about ‘propagation of information,’ we do not speak strictly about propagation of electromagnetic field in the general sense. In subsection 4.3 the general method of finding reference frames co-moving with electromagnetic fields is formulated (mostly for the case of pure subtypes of electric or magnetic types of fields via transformation away of the magnetic or electric field, respectively; however also in the impure subtypes, though there it is impossible to transform away one or — asymptotically — both fields E and B, one always may make these fields mutually parallel, thus transforming away the Poynting vector in the respective frame). In frames co-moving with the electromagnetic field, the Poynting vector automatically vanishes. This method is then applied to the LW field. Relative motion of different reference frames is considered in subsection 4.4, first in general and then for the LW solution. In section 5 some results obtained in the paper are discussed. In two appendices, A and B, a short review of the Ehlers–Zel’manov covariant theory of reference frames (its algebraic part) is given together with applications to the description of electric and magnetic fields.
## 2 A systematic deduction of the LW solution
Let us consider a pointlike charge $`Q`$ in a motion along a worldline $`L`$ parametrically described as
$$\text{r}^{}=\text{r}^{}(t^{}),\mathrm{equivalently},x^i=x^i(t^{}),$$
(2.1)
$`t^{}=x^0`$, $`i=1,2,3`$. We shall determine at an arbitrary, but fixed spacetime point $`P`$ with coordinates $`x^\mu `$ (not on $`L`$), the electromagnetic field created by the charge $`Q`$ being at another point $`P^{}`$ on $`L`$; the coordinates are chosen to be Cartesian. It is obvious that the electromagnetic field created by a pointlike charge should have a singularity on $`L`$, this is why we exclude here the case of coincidence of the points $`P`$ and $`P^{}`$. Note that the coordinates of $`P`$ represent four independent scalar variables $`x^\mu `$, and those of $`P^{}`$ merely are scalar functions of some parameter (this may be $`s^{}`$, but we shall use the retarded time $`t^{}`$) along the worldline $`L`$, $`x^\mu (t^{})`$ (there are three equations, the fourth being simply an identity, $`x^0=t^{}`$). To mutually relate the spacetime points $`P`$ and $`P^{}`$, we use a hypothesis that the information about position and state of motion of the charge propagates with the fundamental velocity (that of light) in an accordance with the relativistic causality law. If the point $`P`$ and worldline $`L`$ are given, the point $`P^{}`$ can be determined as that of intersection of the past light cone with a vertex at $`P`$ and the line $`L`$ (this simultaneously means that $`P`$ is on the future light cone with a vertex at $`P^{}`$). This constructive definition is important in the subsequent calculations, but fortunately the concrete relation between the position of $`P`$ and the corresponding retarded time $`t^{}`$ at $`P^{}`$ turns out to be of no importance. Thus $`t^{}`$ is a function of all four coordinates of $`P`$ — we write it as $`t^{}(x)`$; we shall easily calculate the explicit form of derivatives of $`t^{}`$ with respect to the coordinates $`x^\mu `$ without an explicit knowledge of $`t^{}(x)`$.
We take the Minkowski metric as $`g_{\mu \nu }=g^{\mu \nu }=\text{diag}(+1,1,1,1)`$ (in fact, this is the definition of Cartesian coordinates), thus the tangent vector to $`L`$, $`u^\mu =dx^\mu /ds^{}`$ (the four-velocity of the charge) taken at the retarded point $`P^{}`$, is timelike and unitary ($`u^{}u^{}u^\mu u_\mu ^{}=1`$), its timelike property being manifested by the relation $`ds^2>0`$ along $`L`$. Locally, $`u^{}`$ determines the direction of growth of the proper time $`s^{}`$, being simultaneously the projector onto the (retarded) physical time direction of the (retarded) reference frame (retardedly) co-moving with the charge. Another projector, now a tensor, can be constructed as (A.2), here
$$b_{\mu \nu }=g_{\mu \nu }u_\mu ^{}u_\nu ^{}.$$
(2.2)
It is (a) symmetric ($`b_{\mu \nu }=b_{\nu \mu }`$), (b) orthogonal to $`u^{}`$, thus realizing projection onto the subspace $`u^{}`$ (the physical three-space of the just mentioned inertial reference frame at $`P^{}`$); (c) it possesses the property of idempotent ($`b_\lambda ^\mu b_\nu ^\lambda =b_\nu ^\mu `$ with det $`b_\lambda ^\mu =0`$), and (d) plays the rôle of the three-dimensional metric in the mentioned subspace, with the signature $`(0,,,)`$ (zero is inserted in the four-dimensional sense). Thus $`g_\lambda ^\lambda \delta _\lambda ^\lambda =4`$ and $`b_\lambda ^\lambda =3`$ give dimensionalities of the space-time and subspace under consideration.
Let us introduce a vector connecting the four-points (events) $`P^{}`$ and $`P`$,
$$R^\mu =x^\mu x^\mu (t^{}).$$
(2.3)
Of course, this is not a vector under more general transformations than the Lorentz ones (like the Euclidean ‘radius vector’ is a vector only in Cartesian systems). Since $`R^\mu `$ lies on the light cone,
$$R^\mu R_\mu =0,$$
(2.4)
this vector is null. Its projection onto $`u^{}`$ is denoted as $`D`$, and onto the retarded three-space, as $`\text{D}^\mu `$:
$$D:=u^\mu R_\mu u^{}R,\text{D}^\mu =R^\nu b_\nu ^\mu =R^\mu Du^\mu ,\text{D}u^{}.$$
(2.5)
Due to (2.2), $`\delta _\nu ^\mu =b_\nu ^\mu +u^\mu u_\nu ^{}`$, and the null property (2.4),
$$\text{D}^\mu \text{D}_\mu =D^2,D=\sqrt{\text{D}^\mu \text{D}_\mu },$$
(2.6)
thus we call $`\text{D}^\mu `$ the ‘retarded spatially projected vector between $`P^{}`$ and $`P`$.’ Similarly, $`D`$ is interpreted as the retarded three-dimensional distance between $`P^{}`$ and $`P`$. Recall also that
$$u^\mu =\frac{dx^\mu }{ds^{}}=\frac{dx^0}{ds^{}}\frac{dx^\mu }{dx^0}=u^0(1,v^i).$$
(2.7)
Now we are ready to calculate all necessary derivatives (of $`t^{}`$, $`R^\mu `$, $`D`$, $`u^\mu `$, and more) with respect to $`x^\mu `$. The first step is to write
$$R_{}^{\mu }{}_{,\alpha }{}^{}=\frac{x^\mu }{x^\alpha }\frac{x^\mu }{x^\alpha }=\delta _\alpha ^\mu \frac{dx^\mu }{ds^{}}\frac{ds^{}}{dt^{}}\frac{t^{}}{x^\alpha },$$
that is,
$$R_{}^{\mu }{}_{,\alpha }{}^{}=\delta _\alpha ^\mu \frac{u^\mu }{u^0}t_{,\alpha }^{}.$$
(2.8)
Differentiation of (2.4) yields
$$R_\mu R_{}^{\mu }{}_{,\alpha }{}^{}\frac{1}{2}\left(R_\mu R^\mu \right)_{,\alpha }=0,$$
thus
$$t_{,\alpha }^{}=\frac{u^0R_\alpha }{D},$$
(2.9)
and its substitution into (2.8) yields
$$R_{}^{\mu }{}_{,\alpha }{}^{}=\delta _\alpha ^\mu \frac{u^\mu R_\alpha }{D}.$$
(2.10)
Now,
$$u_{}^{\mu }{}_{,\alpha }{}^{}=\frac{du^\mu }{dt^{}}t_{,\alpha }^{}=\frac{du^\mu }{ds^{}}\frac{ds^{}}{dt^{}}t_{,\alpha }^{}=\frac{a^\mu R_\alpha }{D}$$
(2.11)
(similar derivatives of all primed objects are proportional to $`R`$ with the differentiation subindex), where
$$a^\mu =\frac{du^\mu }{ds^{}}$$
(2.12)
is the acceleration four-vector (at $`P^{}`$) obviously possessing the property of four-orthogonality to $`u^{}`$:
$$u^\mu a_\mu ^{}0.$$
(2.13)
This use of the acceleration four-vector is more economic than of the respective three-vector, though their mutual relation is somewhat indirect; the reader, beginning with (A.8), may easily reconstruct the corresponding formulae and apply them to interpretation of the results and to make a comparison with the treatment of LW problem in . The final step in this part of calculations is to differentiate $`D`$:
$$D_{,\alpha }=(u^{}R)_{,\alpha }=u_{\mu ,\alpha }^{}R^\mu +u_\mu ^{}R_{}^{\mu }{}_{,\alpha }{}^{}=u_\alpha ^{}\frac{R_\alpha }{D}(1a^{}R)$$
(2.14)
where, of course, $`a^{}R:=a_\mu ^{}R^\mu a^{}\text{D}`$. Let us also take into account that
$$R_{}^{\nu }{}_{,\nu }{}^{}=3\text{ and }a_{}^{\mu }{}_{,\nu }{}^{}=\frac{da^\mu }{ds^{}}\frac{R_\nu }{D}$$
(2.15)
\[see a comment to (2.11)\].
The second, and last, preparatory part of our calculations is to write down Maxwell’s equations. Outside the sources, their four-dimensional form is
$$F_{}^{\mu \nu }{}_{,\nu }{}^{}=0$$
(2.16)
where
$$F_{\mu \nu }=A_{\nu ,\mu }A_{\mu ,\nu }$$
(2.17)
is the field tensor written in terms of the four-potential $`A_\mu `$, thus $`F_{}^{\mu \nu }{}_{,\nu }{}^{}=\mathrm{}A^\mu +\left(A_{}^{\nu }{}_{,\nu }{}^{}\right)^{,\mu }=0`$, the d’Alembertian operator being $`\mathrm{}=\mathrm{\Delta }^2/t^2`$. The $`A_{}^{\nu }{}_{,\nu }{}^{}`$-term can be eliminated if we use the Lorenz condition<sup>1</sup><sup>1</sup>1This condition is due not to H.A. Lorentz as admits the majority of physicists, but to L.V. Lorenz (born in Elsinore, Denmark, in 1829), see the footnote related to formula (5.1.47) in , p. 321.
$$A_{}^{\nu }{}_{,\nu }{}^{}=0$$
(2.18)
which only fixes global gauge of the four-potential without any other restrictions. The alternative form of Maxwell’s equations should then include the Lorenz condition, thus in the form of a system
$$\mathrm{}A^\mu =0\text{and}A_{}^{\nu }{}_{,\nu }{}^{}=0.$$
(2.19)
The well-known Coulomb potential in a vacuum in electrostatics can be written as $`A^\mu =\frac{Q}{r}\delta _0^\mu `$ for a pointlike charge $`Q`$ located at the spatial origin. One notices that the four-velocity of the charge at rest is $`u^\mu =u^\mu =\delta _0^\mu `$. This potential exactly satisfies both equations of (2.19) when $`\text{r}0`$. We shall now show that a simple generalization of the Coulomb potential is also an exact solution of Maxwell’s equations, and this is precisely that of Liénard–Wiechert.
The generalization is simply
$$A^\mu =\frac{Qu^\mu }{D}.$$
(2.20)
The proof that this is the exact solution is quite short for the Lorenz condition:
$$A_{}^{\nu }{}_{,\nu }{}^{}=\frac{Q}{D}\left(u_{}^{\nu }{}_{,\nu }{}^{}\frac{u^\nu D_{,\nu }}{D}\right)=\frac{Q}{D^2}\left[a^{}R1+\frac{R_\nu u^\nu }{D}\left(1a^{}R\right)\right]0,$$
and for the d’Alembert equation \[the first in (2.19)\], a little tedious. First, we calculate
$$A_{\mu ,\nu }=\frac{Q}{D^2}\left[a_\mu ^{}R_\nu u_\mu ^{}\left(u_\nu ^{}R_\nu \frac{1a^{}R}{D}\right)\right].$$
(2.21)
Turning now to the rest of (2.19), we see that it is necessary to consider $`\mathrm{}A_\mu =A_{\mu ,\nu }^{}{}_{}{}^{,\nu }`$, taking into account (2.15) and the already known derivatives of $`u^{}`$, $`a^{}`$, $`R^\alpha `$, and $`D`$. The reader can verify after performing differentiation that for $`D0`$ all terms identically cancel:
$$\left\{\frac{Q}{D^2}\left[a_\mu ^{}R^\nu u_\mu ^{}\left(u^\nu R^\nu \frac{1a^{}R}{D}\right)\right]\right\}_{,\nu }0.$$
This completes the proof.
Since we shall need the full expression of $`F_{\mu \nu }`$ in the subsequent calculations, let us now antisymmetrize the expression (2.21) (the first term in round brackets is immediately cancelled):
$$F_{\mu \nu }=\frac{Q}{D^2}\left[R_\mu \left(a_\nu ^{}+u_\nu ^{}\frac{1a^{}R}{D}\right)R_\nu \left(a_\mu ^{}+u_\mu ^{}\frac{1a^{}R}{D}\right)\right].$$
(2.22)
This is a specific type of skew-symmetric tensor sometimes called simple bivector since it represents an antisymmetrization of only two vectors, $`R^\mu `$ (2.3) and $`U^\mu =\frac{Q}{D^2}\left(a^\mu +u^\mu \frac{1a^{}R}{D}\right)`$:
$$F_{\mu \nu }=R_\mu U_\nu U_\mu R_\nu $$
(2.23)
which can be written as a 2-form $`F=RU`$, $`R=R_\mu dx^\mu `$ and $`U=U_\mu dx^\mu `$.
## 3 General properties of the LW field
First it is worth mentioning the obvious fact that the Coulomb field is a special case of the LW solution: one simply has to consider a pointlike charge at rest, that is $`u^\mu =\delta _0^\mu `$ for any $`P^{}`$, thus $`a^\mu =0`$. This is the reason why the LW solution has to be interpreted as the electromagnetic field of an arbitrarily moving pointlike charge (of course, the Gauss theorem is here also applicable, for example, in an inertial frame instantaneously co-moving with the central charge at $`P^{}`$).
### 3.1 Classification of electromagnetic fields and its application to the LW solution
The classification of electromagnetic fields is based on existence of only two invariants built with the field tensor $`F_{\mu \nu }`$, while all other invariants are merely algebraic functions of these two (if not vanish identically). The first invariant is $`I_1=F_{\mu \nu }F^{\mu \nu }=2(\text{B}^2\text{E}^2)`$, and the second, $`I_2=F\stackrel{}{\mu \nu }F^{\mu \nu }=4\text{E}\text{B},`$ cf. (B.2) and (B.3); the definition of $`I_2`$ contains dual conjugation of $`F_{\mu \nu }`$,
$$F\stackrel{}{\mu \nu }:=\frac{1}{2}ϵ_{\mu \nu \alpha \beta }F^{\alpha \beta },F\stackrel{\mu \nu }{}:=\frac{1}{2}ϵ_{\mu \nu \alpha \beta }F_{\alpha \beta }.$$
(3.1)
Here $`ϵ_{\mu \nu \alpha \beta }`$ is the completely skew-symmetric object (not exactly a tensor) with $`ϵ_{0123}=+1`$, known as the Levi-Cività symbol. In fact, only the squared $`I_2`$ is really invariant, and $`I_2`$ itself is a pseudo-invariant which acquires the factor $`J/|J|`$ by a general transformation of coordinates, $`J`$ being the Jacobian of the transformation, thus the concrete sign of $`I_2`$ does not matter. In terms of $`I_1`$ the invariant classification suggests three types of fields: $`I_1<0`$ is the electric type (the electric field dominates), $`I_1>0`$ gives the magnetic type, and to $`I_1=0`$ corresponds the null type. On the pseudo-invariant $`I_2`$ the further working out in detail of the classification is based: the additional subtypes are impure ($`I_20`$) and pure ($`I_2=0`$). It is important that the pure electric case permits (at least, locally, if one considers only inertial frames) to completely eliminate the magnetic field, and similarly, the pure magnetic field permits to completely eliminate the electric field, while the pure null electromagnetic field in a vacuum permits to find a coordinate system (reference frame) in which the electric and magnetic field intensities would take any desired finite (nonzero and non-infinite) and equal values, but, of course, the field will continue to pertain to the same pure null type (in this case, both fields E and B will be ever equal in their absolute values and mutually orthogonal, as can be seen from the structure of both invariants). This last property is closely related to the Doppler effect (not only in the sense of the frequency, but — and more profoundly — also of the field intensity), in particular, a complete elimination of the pure null type field is ‘possible’ only asymptotically (in less rich-in-content terms, this means ‘impossible’), since there cannot exist any reference frame moving with the speed of light with respect to an arbitrary permissible reference frame. The impure electric, magnetic, and null types obviously do not permit such manipulations with the three-dimensional parts E and B of the electromagnetic field (in the impure electric and magnetic cases it is impossible to transform away the counterparts of these respective fields).
Let us now apply this classification to the LW electromagnetic field. Since $`I_2=\frac{1}{2}ϵ_{\mu \nu \alpha \beta }F^{\mu \nu }F^{\alpha \beta }0`$ for any simple bivector (2.23), even with arbitrary $`R`$ and $`U`$, the field is pure. Then it is pure electric since
$$I_1=\frac{2Q^2}{D^4}<0$$
(3.2)
(remarkably, the structure of $`I_1`$ is exactly Coulombian). This means that at any point of the spacetime (any finite value of the distance $`D`$, i.e. not asymptotically) it is possible to transform away the magnetic part of the field; moreover, it is possible to find such a global reference frame in which only electric part of the field will be present. This possibility can be globally realized for any concrete choice of the motion of the pointlike charge. In these specific reference frames which are in general non-inertial, but naturally admissible in special relativity (like those to which we are accustomed in non-relativistic physics, the area much more restricted than special relativity), the Poynting vector of the LW field will vanish globally. This fact will be discussed in more concrete details below. Its physical meaning is that at any finite point of the spacetime the electromagnetic LW field propagates with sub-luminal velocity.
## 4 Propagation of the LW electromagnetic field
### 4.1 Viewpoint of an inertial observer
This is the least interesting case of the reference frame application to LW solution while the approach reduces to use of a monad adapted to Cartesian coordinates. Let the inertial observer at $`P`$ measure electric and magnetic fields E and B as well as electromagnetic energy density $`w`$ and Poynting vector S which are two of the three decomposition parts of (B.7) (we shall not consider the stress tensor) with respect to this observer’s monad $`\tau ^\mu =\delta _0^\mu `$ (the observer is at rest with respect to the Cartesian coordinates) and to the corresponding orthogonal projector $`b_\nu ^\mu =\delta _\nu ^\mu \delta _0^\mu \delta _\nu ^0\delta _i^\mu \delta _\nu ^j\delta _j^i(0,\delta _j^i)`$ \[see (2.2)\],
$$wT_{\text{em}}^{}{}_{0}{}^{0}=\frac{1}{4\pi }\left(\frac{1}{4}F_{\alpha \beta }F^{\alpha \beta }F_{0\alpha }F^{0\alpha }\right)=\frac{1}{8\pi }\left(\text{E}^2+\text{B}^2\right),$$
(4.1)
$$\text{S}^i=T_{\text{em}}^{}{}_{0}{}^{i}=\frac{1}{4\pi }F_{i\alpha }F^{0\alpha }=\frac{1}{4\pi }\left(\text{E}\times \text{B}\right)^i,$$
(4.2)
cf. (B.8). Here
$$\text{E}^i=\frac{Q}{D^3}\left[u_0^{}\left(R^iR_0v^i\right)(1a^{}R)+D\left(a_0^{}R^iR_0a^i\right)\right]$$
(4.3)
and $`\text{B}^i=(dtRU)(\text{n}\times \text{E})^i`$ where (for the inertial frame) $`\text{n}=\text{R}/R_0`$ and the electromagnetic field 2-form $`F=RU`$ where $`R`$ and $`U`$ are 1-forms built of the respective covectors found in (2.23); see also the definitions (B.5) and (B.3). Taking into account (A.8) and relations $`D=u_0^{}\left(R_0R^iv^i\right)`$ and $`R^i\left(R^iR_0v^i\right)=R_0\left(R_0R^iv^i\right)`$, it is easy to verify that (4.3) coincides with the expression given by Landau and Lifshitz (, (63,8)) — in our notations,
$$\text{E}^i=\frac{Q}{\left(R_0R^iv^i\right)^3}\left\{\frac{1}{u_{0}^{}{}_{}{}^{2}}\left(R^iR_0v^i\right)+\left[\text{R}\times \left((\text{R}R_0\text{v}^{})\times \dot{\text{v}}^{}\right)\right]\right\}.$$
(4.4)
However, since the Poynting vector expression is nonlinear in characteristics of the electromagnetic field (due to multiplication of electric and magnetic vectors), we prefer our consideration given in the next subsection to that which splits (4.4) in two parts one of which should describe the outgoing radiation; this reasoning works only asymptotically, and the expression (4.3) is in this case more transparent than (4.4) due to the factor $`D`$ in the corresponding term in square brackets in (4.3).
Finally, it is worth mentioning that the differential characteristics of any inertial frame (its acceleration, rotation, and rate-of-strain tensor), including the frame considered above, identically vanish, thus of course simplifying the considerations given in , though at the cost of omission of some important details.
### 4.2 The retarded reference frame co-moving with the charge
The retarded reference frame at $`P`$ co-moving with the charge at $`P^{}`$ is determined by the monad $`\tau ^\mu =u^\mu `$. Thus for electric and magnetic fields we have
$$\text{E}=\frac{Q}{D^2}\left[(1a^{}R)\text{n}D\text{a}^{}\right],\text{B}=\frac{Q}{D}\text{a}^{}\times \text{n}.$$
(4.5)
However it is more direct to use projections considered in appendix B which result in the non-inertial reference frame where the Poynting vector is
$$S^\mu :=T_\lambda ^\nu u^\lambda b_\nu ^\mu ,$$
and projections have to be applied to $`F_{\lambda \alpha }F^{\nu \alpha }`$ using (2.23) and the relation $`R_\mu U^\mu =Q/D^2`$ obvious from $`U^\mu `$ given just before that expression for $`F_{\mu \nu }`$. Then
$$F_{\lambda \alpha }F^{\nu \alpha }u^\lambda b_\nu ^\mu =\frac{Q^2}{D^4}\left[D\text{D}^\mu a^{}a^{}Da^\mu \frac{\text{D}^\mu }{D}a^{}\text{D}(1a^{}\text{D})\right].$$
In order to find a more concise form of the last expression, let us introduce the unit radial vector $`n`$ perpendicular to the monad:
$$\text{n}^\sigma :=\frac{\text{D}^\sigma }{D},\text{n}\text{n}=1.$$
(4.6)
Then
$$S^\mu =\frac{Q^2}{4\pi D^2}\left[\text{n}^\mu \left(a^{}a^{}+(\text{n}a^{})^2\right)\frac{1}{D}\left(a^\mu +\text{n}^\mu (\text{n}a^{})\right)\right].$$
This expression however takes more transparent form if we also use the projector onto the two-dimensional surface simultaneously orthogonal to both $`u^{}`$ and $`n`$. This will be a spherical surface of radius $`D`$ not in a hyperplane perpendicular to $`u^{}`$, but on the future light cone with its vertex at $`P^{}`$ (a sphere corresponding to the retarded time in analogy with determination of the LW field). Thus we introduce the projector
$$c^{\sigma \tau }:=b^{\sigma \tau }+n^\sigma n^\tau \eta ^{\sigma \tau }u^\sigma u^\tau +n^\sigma n^\tau ,$$
(4.7)
$$c^{\sigma \tau }c_{\rho \tau }=c_\rho ^\sigma ,c^{\sigma \tau }n_\sigma =0,c_{}^{\sigma }{}_{\sigma }{}^{}=2,$$
and relations similar to $`a^ϵb^{ϵ\sigma }a_\sigma `$ should be also taken into account. This new projection tensor plays the rôle of metric tensor on the two-dimensional sphere with the signature $`(0,0,,)`$ involving two zeros, one with respect to direction of the proper time from the viewpoint of all four dimensions, and the second, in the sense of the radial direction ($`n`$) which corresponds to the sphere. Finally, the Poynting vector takes the form
$$S^\mu =\frac{Q^2}{4\pi }\left(\frac{1}{D^3}c^{\mu \tau }a_\tau ^{}\frac{1}{D^2}n^\mu c^{\sigma \tau }a_\sigma ^{}a_\tau ^{}\right).$$
(4.8)
This remarkably simple expression of the LW energy flux suggests the following two conclusions. First, the part proportional to $`1/D^3`$ and linear in the retarded four-acceleration $`a^{}`$, is perpendicular to the radial direction n (i.e., it is restricted to the corresponding two-sphere on the future light cone with its vertex at $`P^{}`$). Thus it describes a redistribution of energy at the fixed retarded distance $`D`$ from the field source. The integral redistribution flux becomes smaller with more distant location of the observer and asymptotically ($`D\mathrm{}`$) tends to zero due to multiplication of (4.8) by the two-dimensional surface element of the sphere ($`D^2`$), while the integration is performed only in the sense of angular coordinates on the sphere. Of course, the very surface (if taken not on the light cone), as well as the reference frame’s three-space, is non-holonom since in general this frame possesses rotation
$$\omega =(u^{}du^{})=(a^{}u^{}n)=\text{a}^{}\times \text{n},$$
(4.9)
see (A.9), (A.5), (2.11), and the final remarks in appendix A. Second, the part proportional to $`1/D^2`$ has positive radial direction (take into account that it gives exactly this contribution since four-dimensional square of the spacelike vector $`a^{}`$ is negative due to the space-time signature). Thus it describes an energy flux from the charge to spatial infinity. Moreover, all this part of energy really goes to infinity without being accumulated or rarefied at any values of $`D`$. Hence this term really describes radiation of energy by the accelerated charge. The non-holonomicity remark is here also relevant, and in this situation one has to take certain caution; this is why we mentioned a roundabout approach involving the light cone which always exists and represents a real hypersurface, though its normal vector is null, thus at the same time it is on the light cone itself. This problem goes beyond the bounds of our paper, and we only mention here that it was successfully treated in last few decades in general relativity. After all, we are living and working in the rotating reference frame of our planet, therefore our three-dimensional physical space certainly is non-holonom, but this does not prevent us to do physics and to apply it quite well.
In the retarded co-moving reference frame of the pointlike charge the LW electromagnetic energy flux has no other constituent parts. Since the problem does not take into account the sources of acceleration of the charge (the lack of a strict auto-consistency of the problem), the energy flux does not result here in any change of the state of motion of charge. One may say that there is implicitly some kind of engine which prescribes the exact world line of the charged particle (the LW problem does not involve any information about the particle’s mass and energy), thus this “engine” automatically “takes into account” the particle’s energy loss due to radiation (which at finite distances is not ligtlike, see below). Other details follow from the further consideration of a new reference frame in which the magnetic field of the LW solution simply vanishes.
### 4.3 LW solution in the reference frame co-moving with electromagnetic field, but not with the charge
In a reference frame which is co-moving with electromagnetic field, the Poynting vector should vanish. This can occur for two alternative reasons (to be realized in this frame): either electric and magnetic vectors are mutually parallel (this is the impure classification subcase), or one of them is equal to zero (the pure subcase). The first case was considered by Wheeler toward other ends. The second case pertains naturally to the LW field since this is a pure electric one (thus Wheeler’s approach is not applicable, and the magnetic part can be transformed away via a proper choice of the reference frame). In fact, this possibility is scarcely encountered in literature (I even don’t know any references), and it would be interesting to investigate it in more detail. We shall see that this task is much simpler than one could expect.
Remember the general form of the LW field tensor, (2.23): $`F_{\mu \nu }=R_\mu U_\nu U_\mu R_\nu `$. Let us (algebraically) regauge the vector $`UV=U+kR`$ where $`k`$ is a scalar function. This does not change the field tensor,
$$F_{\mu \nu }=R_\mu V_\nu V_\mu R_\nu .$$
(4.10)
Applying now the 1-form definition of the magnetic vector in a $`\tau `$-frame (B.3) and taking the monad as $`\tau =NV`$ where the scalar normalization factor is $`N=(VV)^{1/2}`$, we obviously come to B$`=0`$ in this frame. The problem is thus reduced to a proper choice of $`k`$ such that $`V`$ will be a suitable real timelike vector with $`VV>0`$. This method should work in our case (for a pure magnetic field, a similar technique can be applied, though requiring automatic representation of $`F`$ as a simple bivector).
We see that
$$V^\mu =\frac{Q}{D^2}\left(a^\mu +\frac{1a^{}R}{D}u^\mu +kR^\mu \right),$$
(4.11)
thus it was natural to include before $`k`$ the scalar coefficient $`Q/D^2`$. Then
$$VV=\left(\frac{Q}{D^2}\right)^2\left[a^{}a^{}+\frac{(1a^{}R)^2}{D^2}+2k\right].$$
(4.12)
In fact, $`k`$ still remains arbitrary. Let it be
$$k=\frac{1}{2}\left[\frac{1}{D^2}a^{}a^{}\frac{(1a^{}R)^2}{D^2}\right]$$
(4.13)
(the first term in the square brackets, $`1/D^2`$, got its denominator to fit the dimensional considerations). Finally,
$$VV=\left(\frac{Q}{D^3}\right)^2>0$$
(4.14)
and
$$\widehat{\tau }^\mu =Da^\mu +\left(1a^{}R\right)u^\mu +\frac{1}{2D}\left[1D^2a^{}a^{}\left(1a^{}R\right)^2\right]R^\mu $$
(4.15)
(it is clear that $`\widehat{\tau }\widehat{\tau }=+1`$). By its definition, the monad $`\widehat{\tau }`$ describes the reference frame co-moving with the LW electromagnetic field: in this frame the Poynting vector of the field vanishes, and the electromagnetic energy flux ceases to exist due to the absence of magnetic part $`\widehat{\text{B}}`$ of the field in this frame (applicable at any finite distance $`D`$, not asymptotically). Really, (4.10) now can be rewritten as
$$F_{\mu \nu }=\frac{Q}{D^3}\left(R_\mu \widehat{\tau }_\nu \widehat{\tau }_\mu R_\nu \right),$$
thus the expression of $`\widehat{\text{B}}`$ (B.3) contains $`\widehat{\tau }R\widehat{\tau }0`$.
Let us now calculate the electric vector $`\widehat{\text{E}}`$ in the frame $`\widehat{\tau }`$. A combination of (4.15), (4.11), and (4.10) gives
$$F=RV=\frac{Q}{D^3}R\widehat{\tau },$$
(4.16)
see also (B.1). Then the expression (B.2) yields
$$\widehat{\text{E}}=(\widehat{\tau }F)=\frac{Q}{D^3}[\widehat{\tau }(R\widehat{\tau })]=\frac{Q}{D^2}\widehat{\text{n}}$$
(4.17)
which is, up to an understandable reinterpretation of notations, exactly the form known as the Coulomb field vector. Here $`\widehat{\text{n}}^\mu =\widehat{\text{D}}^\mu /D`$ ($`\widehat{\tau }`$) where $`R^\mu u_\mu ^{}=:D\widehat{D}:=R^\mu \widehat{\tau }_\mu `$ and $`\widehat{\text{D}}^\mu =\widehat{b}_\nu ^\mu R^\nu `$ with $`\widehat{b}_\nu ^\mu =\delta _\nu ^\mu \widehat{\tau }^\mu \widehat{\tau }_\nu `$, hence
$$\widehat{\text{D}}^\mu =D^2a^\mu D\left(1a^{}R\right)u^\mu +\frac{1}{2}\left[1+D^2a^{}a^{}+\left(1a^{}R\right)^2\right]R^\mu ,$$
(4.18)
$`\widehat{\text{D}}^\mu \text{D}^\mu `$; note that $`\widehat{\text{D}}^\mu \widehat{\text{D}}_\mu =D^2`$, as this was the case for $`\text{D}^\mu `$ in (2.6). It is clear that $`\widehat{\text{D}}^\mu +D\widehat{\tau }^\mu =R^\mu `$.
### 4.4 Relative three-velocities of reference frames
Let us now simultaneously consider three distinct reference frames and denote them as A, B, and C. Between such frames there can be established quite a few algebraic relations having a clear and important physical meaning, and it is interesting that these relations hold equally in general and special relativity. One defines the relative three-velocity of frame B with respect to frame A (and measured in A) as a (co)vector $`\text{v}_{\text{BA}}`$ perpendicular to the monad $`\tau _\text{A}`$. According to (A.6),
$$\tau _\text{B}=(\tau _\text{A}+\text{v}_{\text{BA}})(\tau _\text{A}\tau _\text{B})\text{ and }\text{v}_{\text{BA}}^\mu =\frac{\tau _\text{B}^\nu b_{\text{A}\nu }^\mu }{\tau _\text{A}\tau _\text{B}}$$
(4.19)
(here the relation $`\tau _\text{B}^\mu \tau _\text{A}^\mu (\tau _\text{A}\tau _\text{B})\tau _\text{B}^\nu b_{\text{A}\nu }^\mu `$ was used); hence,
$$\tau _\text{A}\tau _\text{B}=\frac{1}{\sqrt{1+\text{v}_{\text{BA}}\text{v}_{\text{BA}}}}\frac{1}{\sqrt{1\text{v}_{\text{BA}}\text{v}_{\text{BA}}}}=\frac{1}{\sqrt{1\text{v}_{\text{BA}}^2}}.$$
(4.20)
It is clear that similar relations exist for any pair of reference frames whatever when the respective monads are introduced. We see that there is a symmetry for squared three-velocities between any pair of frames, in particular, $`\text{v}_{\text{BA}}^2=\text{v}_{\text{AB}}^2`$. Since these three-velocities are described as four-vectors perpendicular to the respective monads (of the frames corresponding to the frame subindex of $`\tau `$ and of $`b`$), they belong to different (local) three-spatial sections of spacetime and in general cannot be directly compared by measurements ones with others without further projections onto alternative subspaces. The inevitability of such a situation is quite obvious. Even in the generally used special-relativistic composition-of-velocities formula for globally inertial frames in motion along “same spatial direction,” this is in fact also the case which is tacitly assumed, but frequently not properly understood. Its strict formulation when these velocities are not mutually “parallel,” is however more laborious.
Another useful step in our calculations is to apply same procedure as in (4.19), but taken with respect to the frames C and A, then to C and B, and further applying it to the free $`\tau _\text{B}`$, thus $`\tau _\text{C}=(\tau _\text{A}+\text{v}_{\text{CA}})(\tau _\text{A}\tau _\text{C})=(\tau _\text{B}+\text{v}_{\text{CB}})(\tau _\text{B}\tau _\text{C})=[(\tau _\text{A}+\text{v}_{\text{BA}})(\tau _\text{A}\tau _\text{B})+\text{v}_{\text{CB}}](\tau _\text{B}\tau _\text{C})`$. When this expression is multiplied by $`b_\text{A}`$ under a contraction with the lower (component) index of this factor, we come to
$$\text{v}_{\text{CA}}^\nu =\left[\text{v}_{\text{BA}}^\nu (\tau _\text{A}\tau _\text{B})+\text{v}_{\text{CB}}^\mu b_{\text{A}\mu }^\nu \right]\frac{\tau _\text{B}\tau _\text{C}}{\tau _\text{A}\tau _\text{C}}.$$
(4.21)
In fact, this is the local velocities composition formula $`\text{A}\text{B}\text{C}`$ for general (not only inertial) frames in both relativities, special as well as general one. Here, of course, one has to take into account the relation (4.20). In this paper we do not consider further details of the usual composition formula.
Other relations which are worth being mentioned, are the following ones: those with projections onto the alternative monads,
$$\text{v}_{\text{BA}}^\nu b_{\text{B}\nu }^\mu =(\tau _\text{A}\tau _\text{B})\text{v}_{\text{AB}}^\mu \text{ and }\text{v}_{\text{AB}}^\nu b_{\text{A}\nu }^\mu =(\tau _\text{A}\tau _\text{B})\text{v}_{\text{BA}}^\mu ;$$
(4.22)
further, due to (4.19) and (4.22),
$$\text{v}_{\text{AB}}\text{v}_{\text{BA}}=\tau _\text{A}\text{v}_{\text{AB}}=(\tau _\text{A}\tau _\text{B})\text{v}_{\text{BA}}^2=(\tau _\text{A}\text{v}_{\text{AB}})^2/\text{v}_{\text{AB}}^2$$
(4.23)
(here the obvious symmetry $`\tau _\text{A}\text{v}_{\text{AB}}=\tau _\text{B}\text{v}_{\text{BA}}`$ was taken into account); finally,
$$\text{v}_{\text{AB}}=(\tau _\text{A}\tau _\text{B})\text{v}_{\text{BA}}+(\text{v}_{\text{AB}}\tau _\text{A})\tau _\text{A}$$
(4.24)
(decomposition with respect to the frame A). Note that $`\text{v}_{\text{AB}}^2:=\text{v}_{\text{AB}}\text{v}_{\text{AB}}=\text{v}_{\text{AB}}\text{v}_{\text{AB}}>0`$.
Let us globally (at any $`P`$) denote in the LW problem the reference frame of inertial observer as A, $`\tau _\text{A}^\mu =\delta _0^\mu `$, the retarded frame co-moving with the charge as B, $`\tau _\text{B}^\mu =u^\mu `$, and the frame co-moving with the field and introduced in subsection 4.3, as C ($`\tau _\text{C}^\mu =\widehat{\tau }^\mu `$). Then, on the one hand,
$$(\tau _\text{B}\tau _\text{C})=(u^{}\widehat{\tau })=1\frac{1}{2}\left[D^2a^{}a^{}+\left(a^{}R\right)^2\right].$$
(4.25)
On the other hand,
$$\text{v}_{\text{CB}}=\frac{\widehat{\tau }}{(u^{}\widehat{\tau })}u^{}.$$
(4.26)
Rotation of the frame C takes the (not quite easily deducible) form
$$\widehat{\omega }=\frac{1D(\dot{a}^{}R)}{1a^{}R}\text{a}^{}\times \widehat{\text{n}}+D\dot{\text{a}}^{}\times \widehat{\text{n}}$$
(4.27)
where 1-form $`\dot{a}^{}=(da_\mu ^{}/ds^{})dx^\mu `$ describes the retarded third proper-time derivative of position of the charge in its motion along the worldline $`L`$. It is worth giving some hints for the deduction of (4.27): The exterior product of any odd-rank forms $`\alpha `$ and $`\beta `$ is skew-symmetric, thus $`\alpha \alpha 0`$. The vector product (A.5) is applicable to a pair of arbitrary vectors, thus it automatically projects each of them onto the three-dimensional subspace orthogonal to the monad. One now has to apply the definition of rotation (A.9) to the monad $`\widehat{\tau }`$. Some simplifications follow immediately. Then to complete the simplification one has to take into account a relation following from the form (not directly from the general definition) of $`\widehat{\tau }`$ (4.15) and $`\widehat{\text{D}}`$ (4.18):
$$\widehat{\text{D}}_\mu =D\widehat{\tau }_\mu 2D^2a_\mu ^{}2D(1a^{}R)u_\mu ^{}+\left[D^2(a^{}a^{})+(1a^{}R)^2\right]R_\mu $$
(at each subsequent step only very few terms survive). The final result is (4.27) which should be compared with (4.9).
## 5 Concluding remarks
We tried to give in this paper a self-sufficient consideration of the LW solution, from its heuristic deduction to an analysis of important properties of the obtained field. One of these properties is that of field’s motion with respect to a given reference frame. In fact, one can relate this motion to the monad describing the frame in which the electromagnetic field does not propagate (its Poynting vector, the electromagnetic energy flux density, vanishes in this frame). It is possible to find such a frame in all cases with the exception of pure null electromagnetic fields: in this latter case both electromagnetic invariants are equal to zero, consequently there remains only an asymptotic possibility to transform away the field’s motion, but then it is transformed away always together with the field itself (this is precisely the asymptotic limit of the Doppler effect). This asymptotic situation does not belong to any admissible reference frame or system of coordinates since such a frame (or, of one wishes, a system) is a degenerate one and thus excluded from consideration (whose region of application is an open one, and the ‘boundary’ is excluded from it, though we can approach it as ‘near’ as we wish, making the non-zero field as weak as we choose it to become). In this pure null case (the definition see in section 3.1) the field by itself exercises lightlike (null) motion, that with the velocity of light. But then there cannot exist a co-moving (with this field) reference frame since its four-velocity should coincide with the monad of the co-moving frame, and the monad vector is timelike by its definition. (More physical reasons are related to the fact that the continuous swarm of observers forming, together with their measuring equipment, a reference frame, and thus being co-moving with it, should always possess non-zero rest masses, though, of course, these masses have to be infinitesimal ones to guarantee the test property of a classical frame of reference. The non-zero rest mass means a timelike worldline of the corresponding object, thus the lightlike motion of any reference frame is physically impossible.) In all other cases concerning electromagnetic fields’ types a co-moving frame is easily realizable (in this paper we discussed the pure electric and pure magnetic types, and all impure subcases should be dealt with according to the method used by Rainich and Wheeler, see ).
Another property is also related to propagation, however not of the field but of the information about its sources, thus this property belongs to the deduction of the LW field. This is a rare case when we encounter in a classical physical context the concept of information usually alien to it. And here information propagates with the velocity of light in a vacuum.
## Appendices
## Appendix A Description of reference frames
In this paper we use notations and definitions from , see also references therein. A reference frame is understood as the splitting of general four-dimensional physical quantities into parts referred to observer’s local time direction and the corresponding local three-dimensional physical subspace orthogonal to it, however the latter (or both parts) are written as four-dimensional tensor quantities (of naturally determined ranks) being orthogonal (or also, if we would wish to emphasize this geometrically, parallel) to observer’s time direction. This direction is expressed via the unit vector (or covector, the distinction should be understandable from the context, frequently mathematical) $`\tau `$, the monad, tangent to the observer’s world line, thus interpreted as the observer’s four-velocity at the event (four-dimensional point) where is located the quantity (object) under consideration. Thus we speak about a continuous swarm of observers, a congruence of their world lines without singularities (the lines do not intersect, and through any event goes one and only one such line). The monad and the metric tensor at each event are necessary and sufficient for a complete description of a reference frame. Of course, this presence of a swarm of observers, with all their equipment necessary for measuring of all physical quantities at any event, should not disturb both usual physical fields and (in general relativity) the spacetime geometry (the gravitational field). Here we consider such arbitrary reference frames only in the framework of special relativity, thus the simplest choice of coordinates is Cartesian which we use in this paper. In our treatment reference frames are generally not related to systems of coordinates, and in one and the same system of coordinates any choice of a reference frame (or different choices simultaneously) may be used.
To split spacetime tensors into their above-mentioned parts, two typical projectors are used. A projector is an idempotent, which means that its repeated action automatically reduces to a single action of it, and it differs from the metric tensor possessing a similar (just mentioned) property by the fact that an application of a projector leads to certain partial loss of information. If we describe a projector as a $`4\times 4`$ matrix (really, a rank two tensor), its determinant should be equal to zero. In more concrete terms, the matrix rank of a projector should be equal to one when we speak about a projector onto a single direction (here, $`\tau `$), or three when we perform a projection onto the local three-dimensional physical space orthogonal to $`\tau `$. Thus in the first case we can use the projector
$$\pi _\nu ^\mu =\tau ^\mu \tau _\nu $$
(A.1)
and in the second case,
$$b_\nu ^\mu =g_\nu ^\mu \tau ^\mu \tau _\nu ,$$
(A.2)
hence
$$\pi _\lambda ^\mu \pi _\nu ^\lambda =\pi _\nu ^\mu ,b_\lambda ^\mu b_\nu ^\lambda =b_\nu ^\mu ,b_\nu ^\mu \pi _\lambda ^\nu =0,b_\nu ^\mu \tau ^\nu =0.$$
(A.3)
However in the first case we frequently use a mere interior multiplication (that is, with a contraction) by $`\tau `$ since this leads to a four-dimensionally well defined quantity. It is also clear that $`b_\nu ^\mu +\pi _\nu ^\mu =g_\nu ^\mu `$. It is worth being repeated that the matrices corresponding to (A.1) and (A.2) are respectively of ranks one and three.
Traditionally, in the literature one usually finds an implicit identification of a four-dimensional Cartesian system of coordinates and the corresponding (“co-moving”) reference frame. This does not pose any ambiguities, only if different reference frames are not considered simultaneously on the background of same system of coordinates, or a non-inertial reference frame is involved. However it is better to take into account that this traditional approach represents a tacit admission that the monad coincides with the unit (timelike) vector along the $`t`$-axis and any orthonormal transformation is accompanied with a corresponding change of the monad. There is also a widespread prejudice that non-inertial frames cannot be used in or they contradict to the special theory of relativity, but this is nothing more than a prejudice. In this paper we consider such frames of non-inertial observers in two concrete cases, and the monad approach works perfectly in description of physical situation in these non-inertial frames. We also use another projector (of rank-two matrix, that is, realizing projection onto a two-dimensional subspace) when it simplifies description of the situation, and there should exist a naturally determined spatial direction which enables this description.
It is convenient, in the sense of both calculations and adequate work of physical intuition, to use the vector symbolics of scalar and vector products denoted as $``$ and $`\times `$. In fact, these operations are coincident with those of the three-dimensional vector algebra, though the objects to which they are applied are four-dimensional vectors restricted to the three-dimensional subspace orthogonal to the monad (not always to the global subspace corresponding in particular to an inertial frame, but, in rotating frames, changing to the more general local non-holonomic case: see in the end of this appendix comments related to the three-dimensional subspaces then having such a local meaning only). These products are defined as
$$\text{p}\text{q}:=b_{\mu \nu }p^\mu q^\nu [(\tau p)(\tau q)]$$
(A.4)
and
$$\text{p}\times \text{q}=(p\tau q).$$
(A.5)
We use here the Cartan exterior forms notations such as the wedge product $``$, the Hodge star operation $``$ (the dual conjugation of a $`p`$-form, not necessarily of a 2-form = skew-symmetric rank-two tensor), and, later, the exterior differential $`d`$, see for details and references .
In Cartesian coordinates, due to the spacetime signature $`(+,,,)`$, the monad of the frame co-moving with these coordinates is $`\tau ^\mu =\delta _0^\mu ,\tau _\mu =\delta _\mu ^0`$. Thus (A.5) becomes $`(\text{p}\times \text{q})^i=ϵ_{ijk}p^jq^k`$. The (co)vectors lying in the three-dimensional subspace of a reference frame are usually written as four-dimensional ones, but in some important cases we put them in boldface printing (as E and B for electric and magnetic vectors). Then E$`{}_{}{}^{2}\text{E}\text{E}=b_{\mu \nu }\text{E}^\mu \text{E}^\nu `$, etc.
The three-dimensional velocity v (described as a four-vector $`\tau `$) of a pointlike particle from the viewpoint of reference frame corresponding to the monad $`\tau `$, is determined via the splitting of its four-velocity $`u^\mu =dx^\mu /ds`$,
$$u=(\tau u)(\tau +\text{v}),\text{ or equivalently }\text{v}^\mu =b_\nu ^\mu \frac{dx^\nu }{\tau _\alpha dx^\alpha }$$
(A.6)
where $`\tau _\alpha dx^\alpha /ds=(1v^2)^{1/2}`$; cf. also (4.20) and the corresponding remarks. This is, of course, an exclusion in the general method of projecting vector and tensor quantities. Another exclusion is the relation between the four-dimensional acceleration and its usual three-dimensional counterpart which is applied in making an easier comparison with the Landau–Lifshitz treatment of the LW field . It is now convenient to write the corresponding relations in the (local) three-dimensional subspace notations. The relativistic acceleration four-vector then is
$$a^\mu =\frac{du^\mu }{ds^{}}=\frac{1}{\sqrt{1\text{v}^2}}\left[\frac{d}{dt^{}}\left(\frac{1}{\sqrt{1\text{v}^2}}\right)\right](1,\text{v}^{})+\frac{1}{1\text{v}^2}(0,\dot{\text{v}^{}}),$$
and the orthogonality of $`a^{}`$ and $`u^{}`$,
$$u^{}a^{}=\frac{d}{dt^{}}\left(\frac{1}{\sqrt{1\text{v}^2}}\right)\frac{1}{\left(1\text{v}^2\right)^{3/2}}\text{v}^{}\dot{\text{v}^{}}=0,$$
(A.7)
finally yields a simpler relation between the four- and three-acceleration
$$a^\mu =\frac{\text{v}^{}\dot{\text{v}^{}}}{\left(1\text{v}^2\right)^2}(1,\text{v}^{})+\frac{1}{1\text{v}^2}(0,\dot{\text{v}^{}}).$$
(A.8)
Rotation of a reference frame is defined as
$$\omega =(\tau d\tau )2(\tau A),A=\frac{1}{2}A_{\mu \nu }dx^\mu dx^\nu ,$$
(A.9)
while in Cartesian coordinates and with $`\tau `$ describing a non-inertial frame, $`A`$ (not the electromagnetic four-potential 1-form, but the rotation 2-form) is the skew term in the natural decomposition of gradient of the monad,
$$\tau _{\mu ,\nu }=\tau _\nu G_\mu +A_{\nu \mu }+D_{\nu \mu },A_{\mu \nu }=A_{[\mu \nu ]},D_{\mu \nu }=D_{(\mu \nu )},$$
(A.10)
$`G`$ being acceleration of the reference frame and $`D`$, the frame’s symmetric rate-of-strain tensor; $`G`$, $`A`$, and $`D`$ belong to the above-mentioned three-dimensional (local) subspace. Of course, all these quantities become equal to zero in any inertial frame globally. When $`A0`$ (equivalent to $`\omega 0`$), the three-dimensional subspace orthogonal to $`\tau `$ is non-holonom, that is, there only exists an overall distribution of elements of the corresponding (now non-holonom) hypersurface, but these elements do not fit together to form a global spatial hypersurface in the proper (holonom) sense, see , the fact well known in geometry of congruences (here we are dealing with the $`\tau `$-congruence).
## Appendix B Electromagnetic fields in arbitrary reference frames
Let us now apply the definitions given in appendix A to the electromagnetic field and related quantities. The field tensor $`F_{\alpha \beta }`$ which also can be written as a 2-form
$$F=\frac{1}{2}F_{\mu \nu }dx^\mu dx^\nu ,$$
(B.1)
splits into two four-dimensional vectors, electric
$$\text{E}_\mu =F_{\mu \nu }\tau ^\nu \text{E}=(\tau F)$$
(B.2)
and magnetic
$$\text{B}_\mu =F\stackrel{}{\mu \nu }\tau ^\nu \text{B}=(\tau F),$$
(B.3)
both $`\tau `$, see also (3.1); 2-form $`F:=\frac{1}{2}F_{\mu \nu }dx^\mu dx^\nu `$. This splitting follows from an observation that the Lorentz force can be expressed as
$$(\text{E}+\text{v}\times \text{B})_\alpha =F_{\mu \nu }\left(\tau ^\nu +\text{v}^\nu \right)b_\alpha ^\mu .$$
(B.4)
In Cartesian coordinates (and with the corresponding inertial monad) we have the same relations as for usual contravariant three-vectors:
$$\text{E}^i=F_{i0}=F^{i0},\text{B}^i=\frac{1}{2}ϵ_{ijk}F_{jk}=\frac{1}{2}ϵ_{ijk}F^{jk},$$
(B.5)
thus
$$F_{ij}=F^{ij}=ϵ_{ijk}\text{B}^k.$$
(B.6)
The electromagnetic stress-energy tensor is
$$T_{\text{em}}^{}{}_{\mu }{}^{\nu }=\frac{1}{4\pi }\left(\frac{1}{4}F_{\kappa \lambda }F^{\kappa \lambda }\delta _\mu ^\nu F_{\mu \lambda }F^{\nu \lambda }\right)$$
(B.7)
(in Gaussian units). Its deduction is most simple when one considers Maxwell’s equations in tensor form in a vacuum and without sources. Its (single) contraction with arbitrary monad includes the Poynting vector in that frame,
$$T_{\text{em}}^{}{}_{\mu }{}^{\nu }\tau _\nu =\frac{1}{8\pi }\left[\left(\text{E}^2+\text{B}^2\right)\tau _\mu +2(\text{E}\times \text{B})_\mu \right],$$
(B.8)
and the squared expression is
$$T_{\text{em}}^{}{}_{\mu }{}^{\nu }T_{\text{em}}^{}{}_{\xi }{}^{\mu }\tau _\nu \tau ^\xi =\frac{1}{(8\pi )^2}\left[\left(\text{E}^2+\text{B}^2\right)^24(\text{E}\times \text{B})^2\right]$$
$$\frac{1}{(8\pi )^2}\left[\left(\text{B}^2\text{E}^2\right)^2+4(\text{E}\text{B})^2\right]=\frac{1}{(16\pi )^2}\left(I_{1}^{}{}_{}{}^{2}+I_{2}^{}{}_{}{}^{2}\right)$$
(B.9)
(it is interesting that this expression is not only a scalar under transformations of coordinates, but it is also independent of the choice of reference frame: the right-hand side does not involve any mention of the monad at all). For the LW field \[due to (3.2)\] this takes a very concise form,
$$T_{\text{em}}^{}{}_{\mu }{}^{\nu }T_{\text{em}}^{}{}_{\xi }{}^{\mu }\tau _\nu \tau ^\xi =\left(\frac{Q^2}{8\pi D^4}\right)^2.$$
(B.10)
|
warning/0506/hep-ph0506150.html
|
ar5iv
|
text
|
# 1 Relative contributions of direct and fragmentation photons to the cross section at NLO and for the NLL resummed case, for 𝑝𝑝 collisions at √𝑆=31.5 GeV. We have chosen all scales as 𝜇=𝑝_𝑇.
Introduction. Prompt-photon production at high transverse momentum , $`pp,p\overline{p},pN\gamma X`$, has been a classic tool for constraining the nucleon’s gluon density, because at leading order a photon can be produced in the Compton reaction $`qg\gamma q`$. The “point-like” coupling of the photon to the quark provides a potentially clean electromagnetic probe of QCD hard scattering. However, a pattern of disagreement between theoretical next-to-leading order (NLO) predictions and experimental data for prompt photon production has been observed in recent years, not globally curable by “fine-tuning” the gluon density . The most serious problems relate to the fixed-target regime, where NLO theory shows a dramatic shortfall when compared to some of the data sets . We note that the mutual consistency of the data sets has been questioned . Nevertheless, for the related single-inclusive neutral-pion production, $`pp\pi ^0X`$, comparisons between NLO calculations and data from mostly the same experiments have also shown a systematic disagreement .
In a recent paper , we have shown that a drastic improvement of the theoretical description of single-inclusive pion production in the fixed-target regime is found when certain large perturbative contributions to the partonic hard-scattering cross sections are taken into account to all orders in perturbation theory. These terms, known as threshold logarithms, arise near partonic threshold, when the initial partons have just enough energy to produce a high-transverse momentum parton (which subsequently fragments into the observed hadron) and a massless recoiling jet. In this case, the phase space available for gluon radiation vanishes, resulting in large logarithmic corrections to the partonic cross section. For the cross section integrated over all rapidities, the most important (“leading”) logarithms are of the form $`\alpha _s^k\mathrm{ln}^{2k}\left(1\widehat{x}_T^2\right)`$ at the $`k`$th order in perturbation theory, where $`\alpha _s`$ is the strong coupling and $`\widehat{x}_T2p_T/\sqrt{\widehat{s}}`$, with $`p_T`$ the parton transverse momentum and $`\sqrt{\widehat{s}}`$ the partonic center-of-mass (cms) energy. Sufficiently close to threshold, NLO, which captures only the term for $`k=1`$, will not be adequate anymore; instead, all logarithmic terms will become relevant and thus need to be taken into account. This is achieved by threshold resummation .
The improvement of the comparison between data and theory due to threshold resummation in pion production has motivated us to revisit prompt-photon production. Here, too, large logarithmic corrections arise near partonic threshold. There is an extensive earlier literature on QCD resummations for the “direct” partonic processes $`qg\gamma q`$ and $`q\overline{q}\gamma g`$. The corresponding phenomenological studies for threshold resummation have found only a relatively small enhancement of the theoretical prediction by threshold resummation, not generally sufficient to provide satisfactory agreement with the fixed-target prompt-photon data. In the present paper, we will extend the previous studies of threshold resummation effects in prompt-photon production by including also the resummation for the “fragmentation” component in the cross section, to which we turn now.
As is well-known (for discussion and references, see also Ref. ), high-$`p_T`$ photons are not only produced by the “direct” contributions from the partonic hard processes $`ab\gamma c`$, but also in jet fragmentation, when a parton $`f`$ emerging from the hard-scattering process fragments into a photon plus a number of hadrons. The need for introducing a fragmentation contribution is physically motivated from the fact that a fragmentation process may produce, for example, a $`\rho `$ meson that converts into a photon, leading to the same signal. In addition, at higher orders, the perturbative direct component contains divergencies from configurations where a final-state quark becomes collinear to the photon. These long-distance contributions naturally introduce the need for non-perturbative fragmentation functions $`D^{f\gamma }`$ into which they can be absorbed. The fragmentation component is of the same perturbative order as the direct one, $`𝒪(\alpha _{em}\alpha _s)`$, since the underlying lowest-order (LO) partonic processes are the $`𝒪(\alpha _s^2)`$ QCD scatterings $`abfc`$, and the fragmentation functions $`D^{f\gamma }`$ are of order $`\alpha _{em}/\alpha _s`$ in QCD. There is some knowledge about the photon fragmentation functions from the LEP experiments . Theoretical model predictions for the photon fragmentation functions are compatible with these data. Using these sets of $`D^{f\gamma }`$, one can then estimate the fragmentation contribution to the prompt photon cross section. NLO calculations in the fixed-target regime show that fragmentation photons contribute about $`1030\%`$ to the prompt-photon photon cross section. Here, the precise value depends both on the photon transverse momentum, but also on the type of hadron beams used. Generally, because of the additional fragmentation function and because of the different underlying hard-scattering processes, the fragmentation component is suppressed and also expected to fall off more rapidly in $`p_T`$ than the direct one. On the other hand, in $`pp`$ or $`pN`$ (as opposed to $`p\overline{p}`$) collisions, the direct channels $`qg\gamma q`$ and $`q\overline{q}\gamma g`$ always involve either a sea quark or gluon distribution in the initial state, which both decrease rapidly towards larger momentum fractions, leading to a rapid decrease of the cross section at high $`p_T`$. In contrast, the fragmentation piece has contributions from $`qq`$ scattering , involving two valence densities. As a result, for $`pp`$ or $`pN`$ collisions, the fragmentation component may continue to be sizable relative to the direct part out to quite high transverse momenta.
Despite the fact that according to the NLO calculation the fragmentation contribution is only a subdominant part of the cross section, in the light of the results of Ref. it deserves a closer investigation. There, as we mentioned above, very large enhancements were found for $`pp\pi ^0X`$ in the fixed-target regime. In the theoretical calculation, the only difference between $`pp\pi ^0X`$ and the fragmentation component to $`pp\gamma X`$ is the use of different fragmentation functions. One therefore expects that also for the fragmentation component to prompt photon production there could be a large increase from resummation. Since it is known from the previous studies that the direct part receives only moderate resummation effects, it is likely that the relative importance of the fragmentation contribution in the fixed-target regime is actually much larger than previously estimated on the basis of the NLO calculations. The precise details will of course depend on the photon fragmentation functions. The $`D^{f\gamma }`$ are much more peaked at large momentum fractions $`z`$ than pion fragmentation functions, due to the perturbative (“point-like”) piece in the evolution . On the other hand, the gluon fragmentation function will be relatively much less important than in the pion case, meaning that some partonic channels with large resummation effects, such as $`gggg`$, are less important. In the present paper, we present a brief phenomenological study of the resummation effects for the fragmentation part of the prompt photon cross section, and their implications for the comparison with the fixed-target data. Irrespective of how well theory and fixed-target data sets agree after the resummation of the fragmentation part is included, the latter is an important ingredient of the theoretical calculation of the cross section.
Resummed cross section. The cross section for $`H_1H_2\gamma X`$ may be written as
$`{\displaystyle \frac{p_T^3d\sigma (x_T)}{dp_T}}={\displaystyle \underset{a,b,f}{}}`$ $`{\displaystyle _0^1}𝑑x_1f_{a/H_1}(x_1,\mu ^2){\displaystyle _0^1}𝑑x_2f_{b/H_2}(x_2,\mu ^2){\displaystyle _0^1}𝑑zz^2D^{f\gamma }(z,\mu ^2)`$
$`{\displaystyle _0^1}𝑑\widehat{x}_T\delta \left(\widehat{x}_T{\displaystyle \frac{x_T}{z\sqrt{x_1x_2}}}\right){\displaystyle _{\widehat{\eta }_{}}^{\widehat{\eta }_+}}𝑑\widehat{\eta }{\displaystyle \frac{\widehat{x}_T^4\widehat{s}}{2}}{\displaystyle \frac{d\widehat{\sigma }_{abfX}(\widehat{x}_T^2,\widehat{\eta },\mu )}{d\widehat{x}_T^2d\widehat{\eta }}}.`$ (1)
We have integrated over all pseudorapidities $`\eta `$ of the produced photon. $`\widehat{\eta }`$ is the pseudorapidity at parton level, with $`\widehat{\eta }_+=\widehat{\eta }_{}=\mathrm{ln}\left[(1+\sqrt{1\widehat{x}_T^2})/\widehat{x}_T\right]`$. The sum in Eq. (S0.Ex1) runs over all partonic subprocesses $`abfX`$, with partonic cross sections $`d\widehat{\sigma }_{abfX}`$, parton distribution functions $`f_{a/H_1}`$ and $`f_{b/H_2}`$, and parton-to-photon fragmentation functions $`D^{f\gamma }`$. The direct contributions are included and are obtained by setting $`f=\gamma `$ and $`D^{f\gamma }=\delta (1z)`$. $`\mu `$ denotes the factorization/renormalization scales, which we have chosen to be equal for simplicity.
The partonic cross sections are computed in QCD perturbation theory. Their expansions begin at $`𝒪(\alpha _s\alpha _{em})`$ for the direct part, and at $`𝒪(\alpha _s^2)`$ for the fragmentation part. Defining
$$\mathrm{\Sigma }_{abfX}(\widehat{x}_T^2,\mu )_{\widehat{\eta }_{}}^{\widehat{\eta }_+}𝑑\widehat{\eta }\frac{\widehat{x}_T^4\widehat{s}}{2}\frac{d\widehat{\sigma }_{abfX}(\widehat{x}_T^2,\widehat{\eta },\mu )}{d\widehat{x}_T^2d\widehat{\eta }},$$
(2)
one finds at NLO the structure
$`\mathrm{\Sigma }_{abfX}(\widehat{x}_T^2,\mu )=\mathrm{\Sigma }_{abfX}^{(\mathrm{Born})}(\widehat{x}_T^2)\left[1+\alpha _s(\mu )\left\{A\mathrm{ln}^2(1\widehat{x}_T^2)+B\mathrm{ln}(1\widehat{x}_T^2)+C+\mathrm{}\right\}\right],`$ (3)
where $`\mathrm{\Sigma }_{abfX}^{(\mathrm{Born})}`$ is the Born cross section for the process $`abfX`$, and $`A,B,C`$ are coefficients that depend on the partonic process. Finally, the ellipses denote terms that vanish at $`\widehat{x}_T=1`$. The logarithmic terms are the leading and next-to-leading logarithms (LL, NLL) at this order. At higher orders, the logarithmic contributions are enhanced by terms proportional to $`\alpha _s^k\mathrm{ln}^m(1\widehat{x}_T^2)`$, with $`m2k`$, at the $`k`$th order of $`\mathrm{\Sigma }_{abfX}`$. As we discussed earlier, these logarithmic terms are due to soft-gluon radiation and may be resummed to all orders in $`\alpha _s`$. The resummation discussed in this work deals with the “towers” for $`m=2k,2k1,2k2`$.
As follows from Eq. (S0.Ex1), since the observed $`x_T=2p_T/\sqrt{S}`$ is fixed, $`\widehat{x}_T`$ assumes particularly large values when the partonic momentum fractions approach the lower ends of their ranges. Since the parton distributions rise steeply towards small argument, this generally increases the relevance of the threshold regime, and the soft-gluon effects are relevant even for situations where the the hadronic center-of-mass energy is much larger than the transverse momentum of the final state hadrons. This effect, valid in general in hadronic collisions, is even enhanced in the fragmentation contribution since only a fraction $`z`$ of the available energy is actually used to produce the final-state photon.
The resummation of the soft-gluon contributions is carried out in Mellin-$`N`$ moment space, where the convolutions in Eq. (S0.Ex1) between parton distributions, fragmentation functions, and subprocess cross sections factorize into ordinary products. We take Mellin moments in the scaling variable $`x_T^2`$ as
$`\sigma (N){\displaystyle _0^1}𝑑x_T^2\left(x_T^2\right)^{N1}{\displaystyle \frac{p_T^3d\sigma (x_T)}{dp_T}}.`$ (4)
In $`N`$-space Eq.(S0.Ex1) becomes
$`\sigma (N)={\displaystyle \underset{a,b,f}{}}f_{a/H_1}(N+1,\mu ^2)f_{b/H_2}(N+1,\mu ^2)D^{f\gamma }(2N+3,\mu ^2)\mathrm{\Sigma }_{abfX}(N),`$ (5)
with the usual Mellin moments of the parton distribution functions and fragmentation functions. As before, for the direct contributions, one has $`D^{f\gamma }=\delta (1z)`$ and therefore $`D^{f\gamma }(2N+3,\mu ^2)=1`$. In addition,
$`\mathrm{\Sigma }_{abfX}(N){\displaystyle _0^1}𝑑\widehat{x}_T^2\left(\widehat{x}_T^2\right)^{N1}\mathrm{\Sigma }_{abfX}(\widehat{x}_T^2).`$ (6)
Here, the threshold limit $`\widehat{x}_T^21`$ corresponds to $`N\mathrm{}`$, and the leading soft-gluon corrections arise as terms $`\alpha _s^k\mathrm{ln}^{2k}N`$.
In Mellin-moment space, threshold resummation results in exponentiation of the soft-gluon corrections. In case of a single-inclusive cross section, the structure of the resummed result reads for a given partonic channel
$`\mathrm{\Sigma }_{abcd}^{(\mathrm{res})}(N1)=C_{abcd}\mathrm{\Delta }_N^a\mathrm{\Delta }_N^b\mathrm{\Delta }_N^cJ_N^d\left[{\displaystyle \underset{I}{}}G_{abcd}^I\mathrm{\Delta }_{IN}^{(\mathrm{int})abcd}\right]\mathrm{\Sigma }_{abcd}^{(\mathrm{Born})}(N1).`$ (7)
Each of the functions $`\mathrm{\Delta }_N^{a,b,c}`$, $`J_N^d`$, $`\mathrm{\Delta }_{IN}^{(\mathrm{int})abcd}`$ is an exponential. The $`\mathrm{\Delta }_N^{a,b,c}`$ represent the effects of soft-gluon radiation collinear to initial partons $`a,b`$ or the “observed” final-state parton $`c`$. The function $`J_N^d`$ embodies collinear, soft or hard, emission by the non-observed parton $`d`$. Large-angle soft-gluon emission is accounted for by the factors $`\mathrm{\Delta }_{IN}^{(\mathrm{int})abcd}`$, which depend on the color configuration $`I`$ of the participating partons. The sum runs over all possible color configurations $`I`$, with $`G_{abcd}^I`$ representing a weight for each color configuration, such that $`_IG_{abcd}^I=1`$. Finally, the coefficient $`C_{abcd}`$ contains $`N`$-independent hard contributions arising from one-loop virtual corrections.
The explicit NLL expressions for all the factors in Eq. (7) may be found in Refs. . The factors $`\mathrm{\Delta }_N^{a,b,c}`$ and $`J_N^d`$ contain the leading logarithms and are universal in the sense that they only depend on the color charge of the parton they represent. Eq. (7) applies to the direct as well as to the fragmentation component. In the former, the “observed” parton is the photon, and thus $`\mathrm{\Delta }_N^c=1`$. Also, in this case there is only one color structure of the hard scattering, so that the sum in Eq. (7) contains only one term. In contrast, several color channels contribute to each of the $`22`$ QCD subprocesses relevant for the fragmentation part. As a result, there are color interferences and correlations in large-angle soft-gluon emission at NLL, and the resummed cross section for each subprocess becomes a sum of exponentials, rather than a single one. The complete expressions for the $`\mathrm{\Delta }_{IN}^{(\mathrm{int})abcd}`$, $`G_{abcd}^I`$ and $`C_{abcd}`$ are also given in Ref. for the direct case, and in for the fragmentation part.
In the resummed exponent, the large logarithms in $`N`$ occur only as single logarithms, of the form $`\alpha _s^k\mathrm{ln}^{k+1}(N)`$ for the leading terms. Subleading terms are down by one or more powers of $`\mathrm{ln}(N)`$. Soft-gluon effects are partly already contained in the ($`\overline{\mathrm{MS}}`$-defined) parton distribution functions and fragmentation functions. As a result, it turns out that they enhance the cross section . We also note that the factors $`\mathrm{\Delta }_N^i`$ depend on the factorization scale in such a way that they will compensate the scale dependence (evolution) of the parton distribution and fragmentation functions. One therefore expects a decrease in scale dependence of the predicted cross section .
We finally note that from the large Mellin-$`N`$ point of view the fragmentation component is at first sight suppressed by $`1/N`$ since the photon fragmentation functions always involve a “quark-to-photon” splitting function $`P_{\gamma q}`$ which in moment space is $`1/N`$. However, as was pointed out in , this suppression may be compensated in particular for $`pp`$ or $`pN`$ collisions by the fact that the fragmentation component involves quark-quark scattering, whereas the direct piece proceeds through quark-antiquark or quark-gluon scattering (see above). At large $`N`$, the quark channels with their valence component dominate. In any case, the resummed corrections for the fragmentation component constitute by themselves a well-defined set of higher-order corrections which has much phenomenological relevance as we will see below. That said, we emphasize that a more detailed analysis of $`1/N`$-suppressed contributions also in the direct part would be desirable for future work.
Phenomenological results. We will now present some phenomenological results for the prompt photon cross section, taking into account the resummation for both the direct and the fragmentation parts. This is not meant to be an exhaustive study of the available data for direct-photon production; rather we should like to investigate the overall size and relevance of the resummation effects and in particular the question in how far they change the relative importance of direct and fragmentation contributions. We therefore select only a few representative data sets to compare to: the E706 data for prompt-photon production in $`pBe`$ scattering at $`\sqrt{S}=31.5`$ GeV, the $`pp`$ data from UA6 ($`\sqrt{S}=24.3`$ GeV), and the data from R806 taken in $`pp`$ collisions at the ISR at $`\sqrt{S}=63`$ GeV.
In order to obtain a resummed cross section in $`x_T^2`$ space, one needs an inverse Mellin transform. As previous studies we will use the “Minimal Prescription” developed in Ref. , for which one chooses a Mellin contour in complex-$`N`$ space that lies to the left of the poles at $`\lambda =1/2`$ and $`\lambda =1`$ in the resummed Mellin integrand, where $`\lambda =\alpha _s(\mu ^2)b_0\mathrm{ln}(N)`$ with $`b_0=(332N_f)/12\pi `$, but to the right of all other poles.
When performing the resummation, one of course wants to make full use of the available fixed-order cross section, which in our case is NLO ($`𝒪(\alpha _{em}\alpha _s^2)`$). Therefore, a matching to this cross section is appropriate, which may be achieved by expanding the resummed cross section to NLO, subtracting the expanded result from the resummed one, and adding the “exact” NLO cross section :
$`{\displaystyle \frac{p_T^3d\sigma ^{(\mathrm{match})}(x_T)}{dp_T}}`$ $`={\displaystyle \underset{a,b,f}{}}{\displaystyle _𝒞}{\displaystyle \frac{dN}{2\pi i}}\left(x_T^2\right)^Nf_{a/h_1}(N+1,\mu ^2)f_{b/h_2}(N+1,\mu ^2)D^{f\gamma }(2N+3,\mu ^2)`$
$`\times \left[\mathrm{\Sigma }_{abfd}^{(\mathrm{res})}(N)\mathrm{\Sigma }_{abfd}^{(\mathrm{res})}(N)|_{\mathrm{NLO}}\right]+{\displaystyle \frac{p_T^3d\sigma ^{(\mathrm{NLO})}(x_T)}{dp_T}},`$ (8)
where $`\mathrm{\Sigma }_{abcd}^{(\mathrm{res})}(N)`$ is the resummed cross section for the partonic channel $`abcd`$ as given in Eq. (7). In this way, NLO is taken into account in full, and the soft-gluon contributions beyond NLO are resummed to NLL. Any double-counting of perturbative orders is avoided. Note that, as before, this cross section is the sum of both direct and fragmentation contributions.
As we have discussed earlier, we perform the resummation for the fully rapidity-integrated cross section. In experiment always only a certain limited range of rapidity is covered. In order to be able to compare to data, we therefore approximate the cross section in the experimentally accessible rapidity region by
$$\frac{p_T^3d\sigma ^{(\mathrm{match})}}{dp_T}(\eta \mathrm{in}\mathrm{exp}.\mathrm{range})=\frac{d\sigma ^{(\mathrm{match})}(\mathrm{all}\eta )}{d\sigma ^{(\mathrm{NLO})}(\mathrm{all}\eta )}\frac{p_T^3d\sigma ^{(\mathrm{NLO})}}{dp_T}(\eta \mathrm{in}\mathrm{exp}.\mathrm{range}).$$
(9)
In other words, we rescale the matched resummed result by the ratio of NLO cross sections integrated over the experimentally relevant rapidity region or over all $`\eta `$, respectively.
Our choice for the parton distribution functions will be the CTEQ6 set . For the photon fragmentation functions we use those of . We note that other sets have been proposed for the latter.
We start by comparing the relative importance of the photon fragmentation contribution at NLO and after NLL resummation of the threshold logarithms. Figure 1 shows the corresponding ratios
$$\frac{\mathrm{direct}}{\mathrm{direct}+\mathrm{fragmentation}},\frac{\mathrm{fragmentation}}{\mathrm{direct}+\mathrm{fragmentation}},$$
as functions of the photon transverse momentum $`p_T`$, for $`\sqrt{S}=31.5`$ GeV, corresponding to a typical fixed-target energy. Here we have considered $`pp`$ collisions, and we have chosen the factorization/renormalization scales as $`\mu =p_T`$. One can see that the NLO fragmentation component contributes about $`40\%`$ of the cross section at the lowest $`p_T`$ shown and then rapidly decreases, becoming lower than $`10\%`$ at $`p_T11`$ GeV. As we anticipated earlier, threshold resummation affects the fragmentation component much more strongly than the direct part. After resummation, the fragmentation contribution is relatively much more important, as shown in Fig. 1, yielding almost $`60\%`$ of the cross section at smaller $`p_T`$ and still more than $`20\%`$ at $`p_T=11`$ GeV.
Similar conclusions are reached when one analyzes the additional enhancement that NLL resummation gives over NLO. In Fig. 2 we show the “$`K`$-factors”
$$K\frac{d\sigma ^{(\mathrm{match})}}{d\sigma ^{(\mathrm{NLO})}}$$
(10)
for the case where only the direct contribution is resummed (and the fragmentation one taken into account at NLO), and for the case when both contributions, direct and fragmentation, are resummed. We have chosen the same energy and other parameters as in the previous figure. In agreement with earlier studies , resummation of the direct contribution alone is fairly unimportant at lower $`p_T`$, yielding a “$`K`$-factor” close to unity. In contrast to this, taking into account the NLL resummation of the fragmentation component as well leads to a much bigger “$`K`$-factor”, roughly a $`50\%`$ enhancement over NLO at the lower $`p_T`$, and even a factor 2.5$``$3 at the highest $`p_T`$ considered. The insert in the figure shows the individual “$`K`$-factors” for the direct and the fragmentation components. The one for the fragmentation piece is very large, albeit not as large as what was found for the case of $`\pi ^0`$ production in our previous study . This finding is explained by the fact that gluonic channels receive much larger resummation effects than quark ones, but that the such channels are relatively suppressed in the photon production case since the gluon-to-photon fragmentation function is much smaller than the gluon-to-pion one.
From Fig. 2 we may conclude that NLL resummation of the fragmentation component leads to a significant enhancement of the theoretical prediction and will have some relevance for comparisons of data and theory. Such comparisons are shown in Figs. 3-5. In Fig. 3 we show the data for $`pBe\gamma X`$ from the E706 experiment , along with our theoretical calculations at NLO and for the NLL resummed case. The energy is $`\sqrt{S}=31.5`$ GeV, as used for the previous figures, and the data cover $`|\eta |0.75`$. We give results for three different choices of scales, $`\mu =\zeta p_T`$, where $`\zeta =1/2,1,2`$. It is first of all evident from the figure that the NLO result falls far short of the data. As we shall see below, this shortfall is particularly pronounced for the E706 data. Furthermore, there is a very large scale dependence at NLO. When the NLL resummation is taken into account, the scale dependence is drastically reduced. This observation was already made in the previous phenomenological studies of the resummed prompt-photon cross section , in which however only resummation for the direct component was implemented. As can also be seen from Fig. 3, at the lower $`p_T`$ the full resummed result is roughly at the upper end of the “band” generated by the scale uncertainty at NLO, whereas at the higher $`p_T`$ it is considerably higher. Overall, as we saw in Fig. 2, there is a further significant enhancement over previous NLL resummed results . This additional enhancement leads to a moderate improvement of the comparison between theory and the E706 data. Clearly, even with NLL resummation of the fragmentation component the calculated cross section remains far below the E706 data, except for $`p_T8`$ GeV.
Figure 4 shows similar comparisons with the data for $`pp\gamma X`$ from UA6 at $`\sqrt{S}=24.3`$ GeV. Here, the resummed calculation, which again shows a very small scale dependence, is in very good agreement with the data. As before, resummation of the fragmentation component leads to a non-negligible enhancement of the cross section, pushing the theoretical NLL results to or slightly beyond the upper end of the NLO scale uncertainty band. Finally, in Fig. 5 we show R806 results for $`pp\gamma X`$ from the ISR at $`\sqrt{S}=63`$ GeV. Similar features as before are observed. Note that we are further away from threshold here, due to the higher energy of the ISR. It is likely that the NLL resummation is not completely accurate here, but that terms subleading in $`N`$ could have some relevance. We reserve the closer investigation of this issue to a future study.
Conclusions and outlook. We have studied the NLL all-order resummation of threshold logarithms in the partonic cross sections relevant for high-$`p_T`$ prompt-photon production. The novel feature of our study is that we have also taken into account the NLL resummation of the photon fragmentation component. Here we were motivated by the rather large enhancements that we had found in a previous study of threshold resummation for the process pp$`\pi ^0X`$. The theoretical description for this process is the same as that for the fragmentation component to the prompt photon cross section; the only difference arises in the use of pion vs. photon fragmentation functions.
We have found that indeed the fragmentation component is subject to much larger resummation effects than the direct one. This implies that probably a substantially larger fraction of observed photons than previously estimated are produced in jet fragmentation. On the other hand, we also found that the enhancement of the fragmentation component due to the threshold logarithms is smaller than the enhancement previously observed for $`\pi ^0`$ production, mostly as a result of the smallness of the photon-to-gluon fragmentation function as compared to the gluon-to-pion one. We note, however, that fairly little is known about the function $`D^{g\gamma }`$. It is probably not ruled out that this functions is much bigger than expected in the set of photon fragmentation functions that we have used, in which case resummation effects would become yet more substantial.
The fully resummed prompt-photon cross section shows a much reduced scale dependence. We find that the comparison of the NLL resummed cross section with experimental data shows varied success, with the theoretical calculations still lying much lower than the E706 data, but consistent with the UA6 and R806 $`pp`$ data. In the light of this, further studies and more detailed comparisons are desirable. We note that generally any residual shortfall of the resummed theoretical results would likely need to be attributed to non-perturbative contributions that are suppressed by inverse powers of the photon transverse momentum. These could for example be related to small “intrinsic” parton transverse momenta . Resummed perturbation theory itself may provide information on the structure of power corrections, through contributions to the resummed expressions in which the running coupling constant is probed at very small momentum scales. A recent study addressed this issue in the case of the prompt-photon cross section at large $`x_T`$ and estimated power corrections to be not very sizable.
Our study improves the theoretical description and thus is a step towards a better understanding of the prompt-photon cross section in the fixed-target regime.
Acknowledgments. The work of D.dF has been partially supported by Conicet, Fundación Antorchas, UBACyT and ANPCyT. W.V. is grateful to RIKEN, Brookhaven National Laboratory and the U.S. Department of Energy (contract number DE-AC02-98CH10886) for providing the facilities essential for the completion of his work.
|
warning/0506/physics0506107.html
|
ar5iv
|
text
|
# Activation entropy of electron transfer reactions
## I Introduction
Beginning with work of Marcus on electron transfer (ET) between ions dissolved in polar solvents Marcus (1993), the understanding of the dynamics and thermodynamics of the nuclear polarization coupled to the transferred electron has been viewed as a key component of ET theories. The concept of polarization fluctuations as a major mechanism driving ET has been extended over the several decades of research from simple molecular solvents to a diversity of condensed-phase media of varying complexity. A significant part of the present experimental and theoretical effort is directed toward the understanding of ET in biology, where this process is a key component of energy transport chains Marcus and Sutin (1985); Winkler and Gray (1992); McLendon and Hake (1992); Warshel (2002). Biological systems pose a major challenge to theoretical and computational chemistry from at least two viewpoints. First, the solvent, including bulk and bound water Gregory (1995), membranes, and parts of the polar and polarizable matrix of the biopolymer, is highly anisotropic and heterogeneous. Second, the geometry of what can be separated as a solute is often very complex, including concave regions of molecular scale occupied by the solvent and regions of the biopolymer with a significant mobility of polar and ionizable residues.
Dielectric continuum models accommodate the complex solute shape by numerical algorithms solving the Poisson equation with the boundary conditions defined by a dielectric cavity Rocchia et al. (2001). The heterogeneous nature of the solvent in the vicinity of a redox site can in principle be included by assigning different dielectric constants to its heterogeneous parts Siriwong et al. (2003). Two fundamental problems inevitably arise in this algorithm. The first has been well recognized over the years of its application and is related to the ambiguity of defining the dielectric cavity for molecular solutes. This problem is often resolved by proper parameterization of the radii of atomic and molecular groups of the solute. The second problem is much less studied. It is related to the fact that collective polarization fluctuations of molecular dielectrics possess a finite correlation length which may be comparable to the length of concave regions of the solute or some other characteristic dimensions significant for solvation thermodynamics. The definition of the dielectric constant for polar regions of molecular length is very ambiguous and, in addition, once the dielectric constant is defined, it is not clear if the dielectric response can fully develop on the molecular length scale.
In addition to the problems in implementing the continuum formalism for molecular solutes there are some fundamental limitations of the continuum approximation itself that may limit its applicability to solvation and electron transfer thermodynamics. On the basic level, the definition of the molecular cavity should be re-done for each particular thermodynamic state of the solvent Roux et al. (1990); Lynden-Bell (1999) and/or electronic state of the solute Rick and Berne (1994). This precludes the use of continuum theories with a given cavity parametrization to describe derivatives of the solvation free energy, e.g. entropy and volume of solvation Vath et al. (1999). In addition, the calculation of the free energy of ET activation requires a proper separation of nuclear solvation from the overall solvent response. This problem, actively studied by formal theories in the past Lee and Hynes (1988); Kuznetsov (1992); Gehlen et al. (1992); Zhu and Cukier (1995), has been recently addressed by computer simulations Bader and Berne (1996); Ando (2001); Gupta and Matyushov (2004). Computer simulations have indicated that continuum recipes for the separation of nuclear and electronic polarization are unreliable, resulting in too strong a dependence of the solvent reorganization (free) energy on solvent refractive index. All these limitations call for an extension of traditional approaches to solvation and ET thermodynamics that would include microscopic length-scales of solvent polarization.
Microscopic theories of solvation are not yet sufficiently developed to compete efficiently with continuum models in application to solvation of biopolymers. Computer simulations provide a very detailed picture of the local solvation structure, but their application to solvation of large solutes requires very lengthy computations and often includes approximations that are hard to control. In particular, the dielectric response is very slowly converging in simulations and is potentially affected by approximations made to describe the long-range electrostatic forces. Several simulation protocols in which polarization response is (partially) integrated out by analytical techniques have been proposed Marchi et al. (2001); Leontyev et al. (2003). Integral equation theories have been successfully applied to small solutes Raineri and Friedman (1999), but examples of their application to solvation and reactivity of large solutes are just a few Beglov and Roux (1996). The formulation of the solvation problem in terms of molecular response functions holds significant promise, as it combines the molecular length scale of the polarization response with a possibility to accommodate an arbitrary shape of the solute Kornyshev (1985); Kornyshev and Ulstrup (1986); Fried and Mukamel (1990); Bagchi and Chandra (1991); Chandler (1993); Matyushov (1993); Song et al. (1996); Kornyshev and Sutmann (1996); Song and Chandler (1998); Lang et al. (1999); Ramirez et al. (2002). A recent re-formulation of the Gaussian model Chandler (1993) for solvation in polar solvents Matyushov (2004a, b) shows a good agreement with simulations of model systems and an ability to conform with experiment when applied to ET in biomolecules LeBard et al. (2003) and charge-transfer complexes Milischuk and Matyushov (2005a) and to solvation dynamics Matyushov (2005). Testing the algorithm, referred to as the non-local response function theory (NRFT), on model systems for which both computer simulations and experiment exist is critical for future applications to more complex systems. This is the aim of the present contribution.
Testing microscopic solvation theories requires comparison to computer simulations on model, yet realistic, systems. The current experimental database does not provide sufficient accuracy to test various approximations entering theoretical algorithms. On the other hand, computer experiment offers essentially exact (within the accuracy of simulation protocols) integration of the same Hamiltonian as the one used in the analytical theory. Therefore, the present calculations of the ET thermodynamics are compared to recent very extensive Molecular Dynamics (MD) simulations Ungar et al. (1999) of a donor-spacer-acceptor (DSA) complex consisting of transition-metal donor (D) and acceptor (A) sites linked by a polyproline peptide spacer (S) (Fig. 1):
$$(\mathrm{bpy})_2\mathrm{Ru}^{2+}(\mathrm{bpy}^{})(\mathrm{pro})_4\mathrm{O}^{}\mathrm{Co}^{3+}(\mathrm{NH}_3)_5,$$
where in the donor bpy=2,2-bipyridine and bpy=4-methyl-2,2-bipyridyl. The spacer is a polyproline chain whose first member (the N-terminus) is connected to the bpy carbonyl, and whose fourth member is terminated by a carboxylate moiety bound to the $`\mathrm{Co}^{3+}(\mathrm{NH}_3)_5`$ acceptor. This system, modeling ET in redox proteins, is a representative member of a homologous series of DSA complexes for which ET rates as a function of temperature have been reported Ogawa et al. (1993). This complex will be referred to as complex 1 in the text.
The analytical NRFT model is shown to agree exceptionally well with MD simulations for complex 1 (Figure 1) in TIP3P water. In order to provide a rigorous comparison between simulations and analytical theory, the set of solute charges employed in the simulations was also used in the analytical calculations. In addition, the polarization structure factors of TIP3P water were obtained from separate MD simulations to be used as input in the analytical theory. Once the accuracy and robustness of the analytical procedures are tested on MD simulations, the next step is to see if the model is capable of reflecting the behavior of real systems. To this end, we have developed a parameterization scheme for polarization structure factors applicable to polarizable polar solvents. Once this is done, the theory can be extended to calculations at varying thermodynamic conditions of the solvent (e.g., temperature) and should generate a set of predictions which can be tested experimentally.
We use the polypeptide DSA to focus on two problematic areas of dielectric continuum models: dependence of the reorganization energy on the solvent polarizability Bader and Berne (1996); Gupta and Matyushov (2004) and the entropy of nuclear solvation Matyushov (1993); Vath et al. (1999). For both areas there is a fundamental, both quantitative and qualitative, disagreement between microscopic models and continuum calculations. Unfortunately, no experimental evidence on the dependence of the solvent reorganization energy on solvent refractive index is available in the literature. There is, on the other hand, a limited number of experimental Grampp and Jaenicke (1984); Liang et al. (1989); Dong and Hupp (1992); Elliott et al. (1998); Nelsen et al. (1999); Derr and Elliott (1999); Vath et al. (1999); Vath and Zimmt (2000); Zhao et al. (2001); Coropceanu et al. (2003); Mertz (2005) and simulation Leontiev and Basilevskii (2005) studies on the entropy of reorganization. Most of the available experimental (laboratory and simulation) evidence points to a positive reorganization entropy (i.e., a negative slope of the reorganization energy vs temperature) in polar solvents, in agreement with the prediction of microscopic theory Matyushov (1993) and in disagreement with negative entropies from continuum calculations Kumar et al. (1998); Vath et al. (1999). We are aware, however, of a few measurements performed on charged donor-acceptor complexes indicating either zero or negative reorganization entropies Dong and Hupp (1992); Coropceanu et al. (2003); Mertz (2005). Our current calculations on complex 1 (Fig. 1) give absolute values of the reorganization entropy much higher than continuum calculations. This great discrepancy calls for additional tests of the theory against experimental data, which will be a subject of future work.
## II Golden Rule Rate Constant
The Golden Rule rate constant of ET is Kubo and Toyozawa (1955)
$$k_{\text{ET}}=\frac{2\pi V_{12}^2}{\mathrm{}^2}\mathrm{FCWD}(0),$$
(1)
where FCWD stands for the density-of-states weighted Franck-Condon (FC) factor
$$\mathrm{FCWD}(\omega )=\frac{dt}{2\pi }e^{iH_2t/\mathrm{}}e^{iH_1t/\mathrm{}}_ne^{i\omega t}.$$
(2)
Here, $`\mathrm{}_n`$ is an ensemble average over the nuclear degrees of freedom of the system (denoted by subscript “n”), which include the manifold of $`N`$ normal vibrational modes of the donor-acceptor complex $`Q=\{𝐪_1,\mathrm{}𝐪_N\}`$ and the nuclear component of the dipolar polarization of the solvent $`𝐏_n`$. The ensemble average is carried out over the configurations in equilibrium with the initial state. Further, $`H_i`$ ($`i=1,2`$) are the diagonal matrix elements of the unperturbed system Hamiltonian $`H`$ taken on the two-state electronic basis $`\{\mathrm{\Psi }_1,\mathrm{\Psi }_2\}`$: $`H_i=\mathrm{\Psi }_i|H|\mathrm{\Psi }_i`$ ($`i=1`$ and $`i=2`$ stand for the initial and final electronic states, respectively). The sum of $`H`$ and the perturbation $`V`$ makes the whole system Hamiltonian, $`H^{}=H+V`$, and $`V_{12}=\mathrm{\Psi }_1|V|\mathrm{\Psi }_2`$ is the off-diagonal matrix element in the Golden Rule expression.
The system Hamiltonian of a donor-acceptor complex in a condensed-phase solvent can be separated into the gas-phase component, $`H_g`$, the solute-solvent interaction, $`H_{0s}`$ (“0” stands for the solute, “s” stands for the solvent), and the bath Hamiltonian, $`H_B`$, describing thermal fluctuations of the solvent:
$$H=H_g+H_{0s}+H_B.$$
(3)
The gas-phase Hamiltonian is the sum of the kinetic energy of the electrons, kinetic energy of the nuclei, and the full electron-nuclear Coulomb energy. The solute-solvent Hamiltonian for ET in dipolar solvents is commonly given by the coupling of the operator of the solute electric field $`\widehat{𝐄}_0`$ to the dipolar polarization of the solvent $`𝐏`$
$$H_{0s}=\widehat{𝐄}_0𝐏.$$
(4)
The bath Hamiltonian represents Gaussian statistics of the collective mode $`𝐏`$ with the linear response function $`𝝌(𝐫,𝐫^{})`$
$$H_B=\frac{1}{2}𝐏𝝌^1𝐏.$$
(5)
The asterisk between the bold capital letters denotes tensor contraction (scalar product for vectors) and space integration over the volume $`\mathrm{\Omega }`$ occupied by the solvent
$$𝐄𝐏=_\mathrm{\Omega }𝐄𝐏𝑑𝐫.$$
(6)
Assuming that the intramolecular vibrations are decoupled from solvent nuclear modes allows one to cast the FCWD as a convolution of the vibrational, $`G_v(\omega )`$, and solvent, $`G_s(\omega )`$, FC densities Bixon and Jortner (1999):
$$\mathrm{FCWD}(\omega )=_{\mathrm{}}^{\mathrm{}}𝑑\omega ^{}G_v(\omega ^{})G_s(\omega \omega ^{}\mathrm{\Delta }G/\mathrm{}),$$
(7)
where the diabatic equilibrium free energy gap is the sum of the gas-phase component $`\mathrm{\Delta }G_g`$ and difference in solvation energies $`\mathrm{\Delta }G_s`$
$$\mathrm{\Delta }G=\mathrm{\Delta }G_g+\mathrm{\Delta }G_s.$$
(8)
In the absence of vibrational frequency change, the former component is equal to the 0-0 transition energy in the gas phase.
The FC density for each nuclear mode is given in terms of a broadening function $`g_n(t)`$ Ovchinnikov and Ovchinnikova (1969); Mukamel (1995)
$$G_n(\omega ^{})=_{\mathrm{}}^{\mathrm{}}\frac{dt}{2\pi }\mathrm{exp}\left[i(\lambda _n/\mathrm{}\omega ^{})tg_n(t)\right],$$
(9)
where
$$\begin{array}{cc}\hfill g_n(t)=& \frac{1}{\pi }_0^{\mathrm{}}\frac{dz}{z^2}(1\mathrm{cos}zt)\chi _n^{\prime \prime }(z)\mathrm{coth}\frac{\mathrm{}z}{2k_\text{B}T}\hfill \\ & +\frac{i}{\pi }_0^{\mathrm{}}\frac{dz}{z^2}(zt\mathrm{sin}zt)\chi _n^{\prime \prime }(z).\hfill \end{array}$$
(10)
In Eq. (9), $`\lambda _n`$ is the nuclear reorganization energy
$$\lambda _n=\frac{\mathrm{}}{\pi }_0^{\mathrm{}}\frac{dz}{z}\chi _n^{\prime \prime }(z)$$
(11)
and $`\chi _n^{\prime \prime }(z)`$ is the imaginary part of the frequency-dependent linear response function (spectral density) corresponding to the nuclear mode $`n`$ (in general, many such modes contribute to the solvent (s) and vibrational (v) FC densities).
For a set of vibrational normal modes with frequencies $`\omega _q`$ and reorganization energies $`\lambda _q`$, the spectral density is Mukamel (1995)
$$\chi _v^{\prime \prime }(z)=\pi \underset{q}{}S_q\omega _q^2\left[\delta (z\omega _q)\delta (z+\omega _q)\right],$$
(12)
where $`S_q=\lambda _q/\mathrm{}\omega _q`$ is the Huang-Rhys factor. When all nuclear modes are classical, $`g_n(t)=k_\text{B}T\lambda _nt^2/\mathrm{}^2`$ and one reaches the classical, high temperature limit of the Marcus theory
$$\begin{array}{cc}\hfill \mathrm{FCWD}_1(\omega )=& \left[4\pi (\lambda _s+\lambda _v)k_\text{B}T\right]^{1/2}\hfill \\ & \mathrm{exp}\left[\frac{(\mathrm{\Delta }G+\lambda _s+\lambda _v\mathrm{}\omega )^2}{4k_\text{B}T(\lambda _s+\lambda _v)}\right],\hfill \end{array}$$
(13)
where $`\lambda _v=_q\lambda _q`$ is the total vibrational reorganization energy.
When the solvent mode is classical and the vibrations are quantized, one can use the small $`t`$ expansion in Eq. (9), valid in the limit when $`\omega _q`$ is much smaller than the vertical energy gap \[$`|\lambda _v\omega ^{}|`$ in Eq. (9)\]. With $`\mathrm{}\omega _q/k_\text{B}T1`$, one gets
$$g_v(t)(t^2/2\mathrm{})\omega _v\lambda _v,$$
(14)
where
$$\omega _v=(\lambda _v)^1\underset{q}{}\omega _q\lambda _q$$
(15)
is the effective vibrational frequency. With the vibrational broadening function in the form of Eq. (14) the FCWD becomes Holstein (1959); Hopfield (1974); Siders and Marcus (1981); Marcus (1989)
$$\begin{array}{cc}\hfill \mathrm{FCWD}_1(\omega )=& \left[\pi (4\lambda _sk_\text{B}T+2\mathrm{}\omega _v\lambda _v)\right]^{1/2}\hfill \\ & \mathrm{exp}\left[\frac{(\mathrm{\Delta }G+\lambda _s+\lambda _v\mathrm{}\omega )^2}{4k_\text{B}T\lambda _s+2\mathrm{}\omega _v\lambda _v}\right].\hfill \end{array}$$
(16)
The above equation, present in some early papers on ET Hopfield (1974); Siders and Marcus (1981), is not very accurate as was pointed out by Jortner Jortner (1976). The set of equations given below, which can be found in work by Lax Lax (1952), Davydov Davydov (1953), and Kubo and Toyozawa Kubo and Toyozawa (1955), provides a better description of the vibronic envelope.
When the normal mode vibrations are represented by a single effective vibration with frequency defined by Eq. (15), the vibrational FCWD is a weighted sum of resonant vibrational transitions
$$G_v(\omega )=\underset{m=\mathrm{}}{\overset{\mathrm{}}{}}A_m\delta (\omega m\omega _v),$$
(17)
where
$$A_m=e^{S\mathrm{coth}\chi _v+m\chi _v}I_m\left(\frac{S}{\mathrm{sinh}\chi _v}\right),$$
(18)
$`S=\lambda _v/\mathrm{}\omega _v`$, $`\chi _v=\mathrm{}\omega _v/2k_\text{B}T`$, and $`I_m(x)`$ is the modified Bessel function of order $`m`$.
The FCWD for the classical nuclear modes of the solvent is given by the expression
$$G_s(\omega \mathrm{\Delta }G/\mathrm{})=\mathrm{}\delta (\mathrm{\Delta }E(𝐏_n)\mathrm{}\omega ),$$
(19)
where
$$\mathrm{\Delta }E(𝐏_n)=\mathrm{\Delta }G+\lambda _s\mathrm{\Delta }𝐄_0\delta 𝐏_n$$
(20)
and $`\mathrm{}_1`$ denotes an ensemble average over the fluctuations of the nuclear solvent polarization $`𝐏_n`$ coupled to the difference in initial and final state electric fields of the donor-acceptor complex, $`\mathrm{\Delta }𝐄_0=𝐄_{02}𝐄_{01}`$. In Eq. (20), $`\lambda _s`$ stands for the solvent reorganization energy (see below), and $`\delta 𝐏_n`$ is the fluctuation of the nuclear polarization with respect to its equilibrium value. With the Gaussian Hamiltonian for polarization fluctuations \[Eq. (5)\], $`G_s(\omega \mathrm{\Delta }G/\mathrm{})`$ is a Gaussian function leading to a total FCWD in the form of a weighted sum of Gaussians
$$\begin{array}{cc}\hfill \mathrm{FCWD}_i(\omega )& =\left[4\pi \lambda _sk_\text{B}T\right]^{1/2}\underset{m=\mathrm{}}{\overset{\mathrm{}}{}}A_m\hfill \\ & \mathrm{exp}\left(\frac{\left(\mathrm{\Delta }G+\lambda _s+m\mathrm{}\omega _v\mathrm{}\omega \right)^2}{4\lambda _sk_\text{B}T}\right).\hfill \end{array}$$
(21)
When the energy of vibrational excitations is much greater than $`k_\text{B}T`$ \[$`\chi _v1`$ in Eq. (18)\] the FC envelope turns into a sum of Gaussians with weights given by the Poisson distribution Bixon and Jortner (1999)
$$A_m=e^S\frac{S^m}{m!},m>0.$$
(22)
## III Solvation Thermodynamics
Inserting a solute into a molecular solvent results in solvent perturbation that can roughly be split into two components with drastically different length scales. The first component is due to repulsion of the solvent from the solute core caused by short-range, but strong repulsive forces. This perturbation creates a local density profile in the solvent around the solute which may or may not induce a polarization field acting on the solute charges. The electric field of solute charges creates yet another perturbation. The solute electric field is sufficiently long-ranged to induce the dipolar polarization $`𝐏(𝐫)`$ in a quasi-macroscopic region of the solvent around the solute. Gradients of the solute field couple to the higher-order (quadrupolar, etc.) polarization, but this interaction is more short-ranged Perng et al. (1996a, b); Matyushov and Voth (1999); Milischuk and Matyushov (2005a).
The dipolar polarization is caused by alignment of the permanent and induced solvent dipoles along the solute field. This alignment occurs on two quite different time scales: $`10^{15}`$ s for induced dipoles and $`10^{11}10^{12}`$ s for permanent dipoles. Accordingly, the polarization field splits into a fast relaxing electronic polarization (induced dipoles, $`𝐏_e`$) and a much slower nuclear polarization (permanent dipoles, $`𝐏_n`$) com (a). The electronic solvent polarization is always in equilibrium with the changing distribution of the electronic density in the donor-acceptor complex. The energy conservation condition of the Golden Rule formula is thus imposed on the energies with equilibrated electronic polarization. Therefore, before being used in the Golden Rule expression, the Hamiltonian matrix should be averaged over the fast electronic component of the dipolar polarization Gehlen et al. (1992); Matyushov and Ladanyi (1998). For the energy $`E_i`$ depending on the instantaneous configuration of the nuclear subsystem one gets
$$e^{\beta E_i}=\mathrm{Tr}_{\text{el}}\left[e^{H_i/k_\text{B}T}\right],$$
(23)
where Tr$`_{\text{el}}`$ denotes the statistical average over the electronic degrees of freedom of the solvent. Before going into the details of separate calculations for electronic and nuclear components of the polarization, we outline the general formalism of polarization response to an external electric field.
### III.1 Formalism
In the linear response approximation (LRA), the solvent polarization $`𝐏(𝐫)`$ is a linear functional of the perturbing electric field $`𝐄_0`$ (vacuum electric field of the solute for solvation):
$$𝐏(𝐫)=𝝌𝐄_0=𝝌(𝐫,𝐫^{})𝐄_0(𝐫^{})𝑑𝐫^{}.$$
(24)
Here, $`𝝌(𝐫,𝐫^{})`$ is a two-rank tensor describing the non-local linear response of the solvent to the solute electric field, dot denotes tensor contraction over the common Cartesian projections. This function is different from dielectric susceptibility appearing in Maxwell’s equations in two respects. First, $`𝝌(𝐫,𝐫^{})`$ describes the polarization response to the field of external charges and not to the total electric field $`𝐄=𝐄_0+𝐄_P`$ combining the external field with the electric field $`𝐄_P`$ created by the solvent polarization ($`𝝌`$ corresponds to $`𝝌^0`$ of Madden and Kivelson Madden and Kivelson (1984)). Second, $`𝝌(𝐫,𝐫^{})`$ is affected by the presence of the solute and thus $`𝝌(𝐫_1,𝐫^{})`$ is generally not equal to $`𝝌(𝐫_2,𝐫^{\prime \prime })`$ even if $`𝐫_1𝐫^{}=𝐫_2𝐫^{\prime \prime }`$.
Equation (24) determines the response function in terms of an external electrostatic perturbation and polarization induced by it (Fig. 2a). An alternative view of the response function is through the fluctuation-dissipation theorem which relates the response to the correlation function of polarization fluctuations in the solute vicinity
$$𝝌(𝐫,𝐫^{})=(k_\text{B}T)^1\delta 𝐏(𝐫)\delta 𝐏(𝐫^{}).$$
(25)
An important result of the LRA is that this correlation function does not depend on the long-range electrostatic field of the solute. The ensemble average $`\mathrm{}`$ in the presence of the real solute with its charge distribution is equivalent to the ensemble average $`\mathrm{}_0`$ in the presence of a fictitious solute which has the geometry of the real solute (and therefore the complete repulsion potential) but no partial charges. This notion provides a convenient route to the calculations of the response function for complex solutes. Instead of calculating the polarization in response to a non-trivial field $`𝐄_0(𝐫)`$, one can calculate the correlation of polarization fluctuations in the presence of a fictitious solute with only the hard repulsive core of the real solute retained. The correlation function is then calculated with the requirement of zero polarization within the solute (Fig. 2b)
$$𝝌(𝐫,𝐫^{})=(k_\text{B}T)^1\delta 𝐏(𝐫)\delta 𝐏(𝐫^{})_0.$$
(26)
This is the essence of the approach adopted in the present formalism, making the response function solely determined by the molecular structure inherent to the pure solvent and the short-range perturbation produced by the repulsive core of the solute.
The applicability of the LRA to solvation of large donor-acceptor complexes common for ET research in molecular solvents is well supported by existing evidence from computer simulations Hwang and Warshel (1987); Kuharski et al. (1988); Marchi et al. (1993); Yelle and Ichiye (1997); Hartnig and Koper (2001). The direct consequence of the LRA are the following relations for the moments of the solute-solvent interaction potential $`v_{0s}`$:
$$k_\text{B}Tv_{0s}=\left(\delta v_{0s}\right)^2=\left(\delta v_{0s}\right)^2_0.$$
(27)
When the solute-solvent interaction is limited to the coupling of the solute charges to the solvent dipolar polarization, $`v_{0s}=\mathrm{\Psi }|H_{0s}|\mathrm{\Psi }`$ in Eq. (4).
The independence of the response function with respect to the solute charge is propagated into equality of the variance of $`v_{0s}`$ in equilibrium with fully charged solute, $`\mathrm{}`$, and in equilibrium with uncharged solute, $`\mathrm{}_0`$. Figure 3 shows the results of simulations from Ref. Matyushov, 2004b for a model diatomic donor-acceptor complex D–A in a dense solvent of hard sphere point dipoles. The system is designed to mimic the charge separation, D–A $``$ D<sup>+</sup>–A<sup>-</sup>, and charge recombination, D<sup>+</sup>–A<sup>-</sup> $``$ D–A, reactions. The reorganization energies for charge separation, $`\lambda _1=\left(\delta v_{0s}\right)^2_0/2k_\text{B}T`$, and for charge recombination, $`\lambda _2=\left(\delta v_{0s}\right)^2/2k_\text{B}T`$, turn out to be very similar over a broad range of solvent polarities monitored by the dipolar density parameter $`y=(4\pi /9)m^2\rho /k_\text{B}T`$; $`\rho `$ is the solvent number density, $`m`$ is the solvent molecule permanent dipole moment.
The inhomogeneous character of the response functions is retained after transformation to $`𝐤`$-space. The function $`𝝌(𝐤,𝐤^{})`$ then depends on two wavevectors in contrast to the dependence on a single wave-vector for the homogeneous dielectric response. The calculation of $`𝝌(𝐤,𝐤^{})`$ is still a major challenge for microscopic theories of polar solvation. Despite some very active research in this area for the last 80 years since the formulation of the Born model for solvation of spherical ions Born (1920), no microscopic solution applicable to solutes of arbitrary shape has been presented so far. A promising strategy, adopted already in the Born Born (1920) and Onsager Onsager (1936) models, is to calculate the response functions in terms of properties of the pure solvent. This connection can be achieved by considering the polarization correlation function in the presence of the repulsive core of the solute \[Eq. (26)\].
The exclusion of the polarization field from the solute volume is provided by the Li-Kardar-Chandler approach Li and Kardar (1992); Chandler (1993), in which the trajectories defining the response function in its path integral representation are restricted from entering the solute. The result of this procedure is an integral equation relating $`𝝌(𝐤,𝐤^{})`$ to the non-local susceptibility of the pure solvent $`𝝌_s(𝐤)`$ (with a single $`𝐤`$-vector for the homogeneous response) and the shape of the solute. The equation for the response function is then equivalent to the Ornstein-Zernike equation for the solute-solvent correlation function with the Percus-Yevick closure for the solute-solvent direct correlation function Chandler (1993).
No general solution for $`𝝌(𝐤,𝐤^{})`$ in the Li-Kardar-Chandler integral equation has been obtained so far. One can, however, employ analytical properties of the response functions to obtain the solvation chemical potential Matyushov (2004a)
$$\mu _{0s}=\frac{1}{2}\frac{d𝐤d𝐤^{}}{(2\pi )^6}\stackrel{~}{𝐄}_0(𝐤)𝝌(𝐤,𝐤^{})\stackrel{~}{𝐄}_0(𝐤^{}).$$
(28)
The closed-form result for $`\mu _{0s}`$ exists when the Fourier transform of the electric field $`\stackrel{~}{𝐄}_0(𝐤)`$ is known in analytical functional form. This is not the case for many real problems, when the distribution of molecular charge is given from force fields or quantum calculations and the Fourier transform of the field is calculated numerically. Unfortunately, the analytical solution is given by the difference of two large numbers almost canceling each other. It therefore becomes not very practical in strongly polar solvents because of accumulation of numerical errors. To facilitate numerical applications, a mean-field solution for $`𝝌(𝐤,𝐤^{})`$ was offered in Ref. Matyushov, 2004b. This solution eliminates the inhomogeneous character of the response function by a non-local renormalization of its transverse component:
$$𝝌(𝐤,𝐤^{})=(2\pi )^3\delta (𝐤𝐤^{})\left[\chi ^L(𝐤)𝐉^L+\chi ^T(𝐤)𝐉^T\right],$$
(29)
where $`𝐉^L=\widehat{𝐤}\widehat{𝐤}`$ and $`𝐉^T=\mathrm{𝟏}\widehat{𝐤}\widehat{𝐤}`$ are, respectively, the longitudinal and transverse projections of a 2-rank tensor with the axial symmetry established by the direction of the wavevector, $`\widehat{𝐤}=𝐤/k`$. The 6D integral of Eq. (28) is then reduced to the computationally tractable 3D integral.
The transverse, $`\chi ^T(𝐤)`$, and longitudinal, $`\chi ^L(𝐤)`$, projections in Eq. (29) are related to corresponding components of the susceptibility of the pure polar solvent
$$\chi ^T(𝐤)=\chi _s^T(k)\frac{\chi _s^L(0)}{\chi _{\text{tr}}}f_s\chi _s^L(k)\frac{𝐅_0𝐉^L\stackrel{~}{𝐄}_0(𝐤)}{𝐅_0𝐉^T\stackrel{~}{𝐄}_0(𝐤)}$$
(30)
and
$$\chi ^L(𝐤)=\chi _s^L(k).$$
(31)
In Eq. (30), $`\chi _{\text{tr}}=(1/3)\mathrm{Tr}[𝝌_s(0)]`$ and
$$f_s=\frac{2[\chi _s^T(0)\chi _s^L(0)]}{3\chi _{\text{tr}}}.$$
(32)
Further, $`\stackrel{~}{𝐄}_0(𝐤)`$ denotes the Fourier transform of the electric field of the solute calculated on the volume of the solvent $`\mathrm{\Omega }`$ obtained by excluding the hard repulsive core of the solute from the solvent
$$\stackrel{~}{𝐄}_0(𝐤)=_\mathrm{\Omega }𝐄_0(𝐫)e^{i𝐤𝐫}𝑑𝐫.$$
(33)
The mean-field approximation adopted in deriving Eqs. (29)–(33) consists of replacing a generally non-uniform field of the solvent within the solute by its spatial average $`𝐅_0`$ \[Eq. (30)\]. The neglect of the gradients of the field induced by the solvent within the solute amounts to taking the dipolar projection of the solute field according to the following relation:
$$𝐅_0=_\mathrm{\Omega }𝐄_0(𝐫)𝐃_𝐫\frac{d𝐫}{r^3},$$
(34)
where
$$𝐃_𝐫=3\widehat{𝐫}\widehat{𝐫}\mathrm{𝟏}.$$
(35)
is the dipolar tensor. The electric field $`𝐅_0`$ is a generalization of the Onsager reaction field for the case of non-spherical solutes with non-dipolar charge distribution. $`𝐅_0`$ reduces to the Onsager field for spherical dipolar solutes.
The mean-field renormalization of the transverse component of the response function in Eq. (30) resolves the fundamental difficulty of microscopic solvation theories arising from the fact that the short-range repulsive perturbation caused by the solute produces a major change in the polarization response functions compared to those of the pure solvent. For instance, a direct replacement of $`𝝌(𝐤)`$ with $`𝝌_s(𝐤)`$ in the homogeneous approximation (see Ref. Raineri et al., 1994 for discussion) results in divergent behavior of $`\lambda _s`$ with increasing solvent dipole moment Matyushov (1996). The divergence arises from the transverse component of the response (“transverse catastrophe”) which has to be included once the dielectric cavity does not coincide with an equipotential surface of the solute charge distribution Kharkats et al. (1976). In continuum calculations, the divergent behavior is eliminated by imposing boundary conditions at the dielectric cavity on the solution of the Poisson equation.
Although the problem with the transverse response has long been recognized in the literature Kharkats et al. (1976); Matyushov (1996); Kuznetsov and Medvedev (1996), many microscopic formulations of solvation thermodynamics and dynamics have avoided the problem by neglecting the transverse response Chandra and Bagchi (1989); Bagchi and Chandra (1989); Fried and Mukamel (1990) which is also neglected in some continuum calculations, e.g. the Generalized Born approximation Schaefer and Karplus (1996). Equations (29)–(33) provide a general solution of the problem which agrees well with available simulations of polar solvation Matyushov (2004a, b) and experiment on solvation dynamics Matyushov (2005). The formalism is based on the homogeneous solvent susceptibility as input and, once the susceptibility is defined from computer experiment or liquid-state theories, can be applied to solvation in an arbitrary isotropic dielectric.
### III.2 Polarization structure factors
The dipole moment at a given molecule $`j`$ in a polar-polarizable solvent is a sum of the permanent dipole $`𝐦_j`$ and the induced dipole $`𝐩_j`$
$$𝝁_j=𝐦_j+𝐩_j.$$
(36)
The total induced dipole then splits into $`𝐩_j^0`$ created by the external electric field $`𝐄_0(𝐫_j)`$ and $`𝐩_j^R`$ induced by the reaction field (superscript “R”) caused by the dipole $`𝐦_j`$ itself (Fig. 4):
$$𝐩_j=𝐩_j^0+𝐩_j^R$$
(37)
The reaction field caused by the dipole $`𝐦_j`$ relaxes on the time-scale of translational-rotational motion of molecule $`j`$. Therefore, the induced dipole $`𝐩_j^R`$, which follows adiabatically the reaction field, should be attributed com (b) to the slow nuclear polarization of the solvent $`𝐏_n`$. In contrast, the component $`𝐩_j^0`$, following adiabatically the external field, is attributed to the the fast solvent polarization $`𝐏_e`$. The sum of the permanent dipole $`𝐦_j`$ and the induced dipole $`𝐩_j^R`$ makes the effective condensed-phase dipole Stell et al. (1981)
$$𝐦_j^{}=𝐦_j+𝐩_j^R=m^{}\widehat{𝐞}_j,$$
(38)
where $`\widehat{𝐞}_j`$ is the unit vector along the direction of $`𝐦_j`$. The dipole moment $`m^{}`$ in principle depends on the instantaneous configuration of the liquid. However, we will not consider fluctuations of $`m^{}`$ here and, following self-consistent models of polarizable liquids Stell et al. (1981), will replace $`m^{}`$ with its statistical average value.
The attribution of the electronic polarization in equilibrium with the electric field of the permanent dipoles to the nuclear (slow) polarization of the solvent is an essential part of the Pekar partitioning of the solvent polarization into fast and slow components Pekar (1946, 1963). Other partitioning schemes have been proposed Brady and Carr (1985), but they all lead to the same value of the solvation energy when correctly implemented Aguilar (2001). Computer simulation protocols in which the induced polarization is self-consistently adjusted to the instantaneous nuclear configuration provide direct access to the slow polarization in Pekar’s definition Milischuk and Matyushov (2005b). Self-consistent simulations of polarizable solvents are used here to test the analytical procedure employed for the response functions of the nuclear polarization (Sec. IV.1).
The total dipolar response function of the homogeneous solvent is a 2-rank tensor describing correlations of dipole moments $`𝝁_j`$:
$$𝝌_s(𝐤)=(\beta /\mathrm{\Omega })\underset{j,k}{}𝝁_j𝝁_ke^{i𝐤𝐫_{jk}},$$
(39)
where $`𝐫_{jk}=𝐫_j𝐫_k`$ and brackets refer to an ensemble average. Because of the isotropic symmetry of the solvent, $`𝝌_s(𝐤)`$ splits into longitudinal and transverse components Madden and Kivelson (1984)
$$𝝌_s(𝐤)=\chi _s^L(k)𝐉^L+\chi _s^T(k)𝐉^T.$$
(40)
It is convenient to factor the response function into the effective density of dipoles $`y_{\text{eff}}`$, which is mostly affected by the magnitude of the solvent dipole, and the structure factor, which reflects dipolar correlations and can be expressed through angular projections of the pair distribution function Matyushov (2004b)
$$𝝌_s(𝐤)=\frac{3y_{\text{eff}}}{4\pi }\left[S^L(k)𝐉^L+S^T(k)𝐉^T\right].$$
(41)
The structure factors $`S^{L,T}(k)`$ (Fig. 5) are defined based on the unit vectors $`\widehat{𝐮}_j=𝝁_j/\mu _j`$ in the direction of the respective total dipole moments
$$\begin{array}{cc}\hfill S^L(k)=& \frac{3}{N}\underset{i,j}{}(\widehat{𝐮}_i\widehat{𝐤})(\widehat{𝐤}\widehat{𝐮}_j)e^{i𝐤𝐫_{ij}},\hfill \\ \hfill S^T(k)=& \frac{3}{2N}\underset{i,j}{}\left[(\widehat{𝐮}_i\widehat{𝐮}_j)(\widehat{𝐮}_i\widehat{𝐤})(\widehat{𝐤}\widehat{𝐮}_j)\right]e^{i𝐤𝐫_{ij}}.\hfill \end{array}$$
(42)
The effective dipole density in Eq. (41) is
$$y_{\text{eff}}=y_p+(4\pi /3)\rho \alpha ,y_p=(4\pi /9)\rho (m^{})^2/k_\text{B}T,$$
(43)
where $`\alpha `$ is the dipolar polarizability. Only the permanent dipole moment is renormalized by the mean field of the solvent in the above equation, which corresponds to Wertheim’s 1-RPT theory Wertheim (1979) (2-RPT theory renormalizes the polarizability $`\alpha `$ to $`\alpha ^{}`$, but the 1-RPT version of the theory is in better agreement with simulations Gupta and Matyushov (2004)).
The nuclear response function reflects correlated orientations and positions of dipoles $`𝐦_j^{}`$:
$$𝝌_n(𝐤)=(\beta /\mathrm{\Omega })\underset{j,k}{}𝐦_j^{}𝐦_k^{}e^{i𝐤𝐫_{jk}}.$$
(44)
Similarly to Eq. (41), $`𝝌_n(𝐤)`$ can be separated into the longitudinal and transverse components
$$𝝌_n(𝐤)=\frac{3y_p}{4\pi }\left[S_n^L(k)𝐉^L+S_n^T(k)𝐉^T\right].$$
(45)
The nuclear structure factors are defined by Eq. (42), in which the unit vectors $`\widehat{𝐮}_j`$ are replaced by the unit vectors $`\widehat{𝐞}_j`$ \[Eq. (38)\].
The $`k=0`$ values of the structure factors are related to the macroscopic dielectric properties of the solvent. The total polarization response is defined through the static dielectric constant $`ϵ_s`$
$$\begin{array}{cc}\hfill S^L(0)& =\frac{ϵ_s1}{3ϵ_sy_{\text{eff}}},\hfill \\ \hfill S^T(0)& =\frac{ϵ_s1}{3y_{\text{eff}}}.\hfill \end{array}$$
(46)
The nuclear structure factors depend, in addition, on the high-frequency dielectric constant $`ϵ_{\mathrm{}}`$ Milischuk and Matyushov (2005b)
$$\begin{array}{cc}\hfill S_n^L(0)& =\frac{c_0}{3y_p},\hfill \\ \hfill S_n^T(0)& =\frac{ϵ_sϵ_{\mathrm{}}}{3y_p},\hfill \end{array}$$
(47)
where
$$c_0=1/ϵ_{\mathrm{}}1/ϵ_s$$
(48)
is the Pekar factor.
Both $`S_n^{L,T}(k)`$ and $`S^{L,T}(k)`$ tend to unity at $`k\mathrm{}`$. This limit is the result of the point multipole approximation for the intramolecular charge distribution within the solvent molecules. In contrast, charge-charge structure factors defined on interaction-site models of liquids decay to zero at $`k\mathrm{}`$ (Refs. Perng et al., 1996a; Raineri and Friedman, 1999; Perng and Ladanyi, 1999). The region of $`k`$-values where this distinction becomes important is, however, insignificant for the calculation of solvation thermodynamics (see below). The nuclear and the total structure factors differ in the range of small $`k`$-values and around the longitudinal peak as a result of the influence of the high-frequency dielectric constant of the solvent (Fig. 5). The effect of $`ϵ_{\mathrm{}}`$ on the longitudinal peak is insignificant for the calculation of the reorganization energy. Therefore, it is the range of small $`k`$-values and, in addition, the dependence of the liquid-state dipole moment $`m^{}`$ on the solvent polarizability, that ultimately determine the variation of the solvent reorganization energy with the solvent high-frequency dielectric constant $`ϵ_{\mathrm{}}`$ (see below).
### III.3 ET thermodynamics
The solvation thermodynamics of ET is determined by the solvent reorganization energy and the solvent component of the free energy gap. They are defined in terms of the nuclear and total response functions by the following relations
$$\lambda _s=\frac{1}{2}\frac{d𝐤d𝐤^{}}{(2\pi )^6}\mathrm{\Delta }\stackrel{~}{𝐄}_0(𝐤)𝝌_n(𝐤,𝐤^{})\mathrm{\Delta }\stackrel{~}{𝐄}_0(𝐤^{})$$
(49)
and
$$\mathrm{\Delta }G_s=\frac{d𝐤d𝐤^{}}{(2\pi )^6}\mathrm{\Delta }\stackrel{~}{𝐄}_0(𝐤)𝝌(𝐤,𝐤^{})\overline{𝐄}_0(𝐤^{}).$$
(50)
In Eqs. (49) and (50), $`\mathrm{\Delta }\stackrel{~}{𝐄}_0(𝐤)=\stackrel{~}{𝐄}_{02}(𝐤)\stackrel{~}{𝐄}_{01}(𝐤)`$ and $`\overline{𝐄}_0(𝐤)=(\stackrel{~}{𝐄}_{02}(𝐤)+\stackrel{~}{𝐄}_{01}(𝐤))/2`$; $`\stackrel{~}{𝐄}_{0i}(𝐤)`$ are the Fourier transforms of the solute electric field in the initial ($`i=1`$) and final ($`i=2`$) ET states taken over the volume $`\mathrm{\Omega }`$ occupied by the solvent \[Eq. (33)\].
The mean-field solution for the response functions \[Eq. (29)\] splits both the solvent reorganization energy and the free energy gap into their corresponding longitudinal and transverse components:
$$\lambda _s=\lambda _s^L+\lambda _s^T$$
(51)
and
$$\mathrm{\Delta }G_s=\mathrm{\Delta }G_s^L+\mathrm{\Delta }G_s^T.$$
(52)
Each projection is obtained as a $`𝐤`$-integral with the corresponding polarization structure factor. For the “T” projections one gets
$$\lambda _s^T=\frac{3y_p}{8\pi }\frac{S_n^L(0)}{g_{Kn}}\frac{d𝐤}{(2\pi )^3}\left|\mathrm{\Delta }\stackrel{~}{E}_0^T(𝐤)\right|^2S_n^T(k)$$
(53)
and
$$\begin{array}{cc}\hfill \mathrm{\Delta }G_s^T& =\frac{3y_{\text{eff}}}{8\pi }\frac{S^L(0)}{g_K}\frac{d𝐤}{(2\pi )^3}\hfill \\ & \left[|\stackrel{~}{E}_{02}^T(𝐤)|^2|\stackrel{~}{E}_{01}^T(𝐤)|^2\right]S^T(k).\hfill \end{array}$$
(54)
In Eqs. (53) and (54),
$$g_{Kn}=(1/3)\left[S_n^L(0)+2S_n^T(0)\right]$$
(55)
and
$$g_{Kn}=(1/3)\left[S^L(0)+2S^T(0)\right]$$
(56)
are the nuclear and total Kirkwood factors, respectively.
The longitudinal components of free energies, $`\lambda _s^L`$ and $`\mathrm{\Delta }G_s^L`$, include both the longitudinal and transverse projections of the solute field:
$$\lambda _s^L=\frac{3y_p}{8\pi }\frac{d𝐤}{(2\pi )^3}_\mathrm{\Delta }^{\text{eff}}(𝐤)S_n^L(k)$$
(57)
and
$$\mathrm{\Delta }G_s^L=\frac{3y_{\text{eff}}}{8\pi }\frac{d𝐤}{(2\pi )^3}\left(_2^{\text{eff}}(𝐤)_1^{\text{eff}}(𝐤)\right)S^L(k).$$
(58)
In Eqs. (57) and (58),
$$_\mathrm{\Delta }^{\text{eff}}(𝐤)=|\mathrm{\Delta }\stackrel{~}{E}_0^L(𝐤)|^2f_n|\mathrm{\Delta }\stackrel{~}{E}_0^T(𝐤)|^2\frac{\mathrm{\Delta }𝐅_0𝐉^L\mathrm{\Delta }\stackrel{~}{𝐄}_\mathrm{𝟎}(𝐤)}{\mathrm{\Delta }𝐅_0𝐉^T\mathrm{\Delta }\stackrel{~}{𝐄}_0(𝐤)}$$
(59)
and
$$_i^{\text{eff}}(𝐤)=|\stackrel{~}{E}_{0i}^L(𝐤)|^2f_s|\stackrel{~}{E}_{0i}^T(𝐤)|^2\frac{𝐅_{0i}𝐉^L\stackrel{~}{𝐄}_{0i}(𝐤)}{𝐅_{0i}𝐉^T\stackrel{~}{𝐄}_{0i}(𝐤)}.$$
(60)
The longitudinal and transverse components of the electrostatic energy density in Eqs. (53)–(58) are defined as
$$\begin{array}{cc}\hfill |\mathrm{\Delta }E_0^{L,T}(𝐤)|^2& =\mathrm{\Delta }\stackrel{~}{𝐄}_0(𝐤)𝐉^{L,T}\mathrm{\Delta }\stackrel{~}{𝐄}_0(𝐤),\hfill \\ \hfill |E_{0i}^{L,T}(𝐤)|^2& =\stackrel{~}{𝐄}_{0i}(𝐤)𝐉^{L,T}\stackrel{~}{𝐄}_{0i}(𝐤).\hfill \end{array}$$
(61)
The effective fields $`_\mathrm{\Delta }^{\text{eff}}(𝐤)`$ and $`_i^{\text{eff}}(𝐤)`$ depend on the symmetry of the charge distribution within the solute analogously to the result of imposing the boundary conditions on the solution of the Poisson equation in continuum electrostatics.
The electric field $`𝐅_{0i}`$ in Eq. (60) is a generalization of the Onsager reaction cavity field Onsager (1936) to the case of solutes of non-spherical shape and non-point-dipole charge distribution. This field is obtained by summing up a continuous distribution of dipolar electric fields induced by the solute in the solvent volume:
$$𝐅_{0i}=_\mathrm{\Omega }𝐄_{0i}(𝐫)𝐃_𝐫\frac{d𝐫}{r^3},$$
(62)
where $`𝐃_𝐫`$ is given by Eq. (35). Also, $`\mathrm{\Delta }𝐅_0`$ in Eq. (59) is $`\mathrm{\Delta }𝐅_0=𝐅_{02}𝐅_{01}`$. $`𝐅_{0i}`$ becomes the standard Onsager reaction field for a point dipole at the center of a spherical cavity. Finally, in Eqs. (59) and (60),
$$f_s=\frac{2(ϵ_s1)}{2ϵ_s+1}$$
(63)
is the usual Onsager polarity parameter Onsager (1936) and the corresponding polarity parameter for the nuclear polarization is
$$f_n=\frac{2(ϵ_{\mathrm{}}ϵ_s1)}{2ϵ_{\mathrm{}}ϵ_s+1}.$$
(64)
## IV Calculation procedure
The formalism outlined above is realized in a computational algorithm sketched in Figure 6. It includes two branches, one is for the solvent part of the calculation and another is for the solute part. The two parts are combined together in the integration over the inverted space, which yields the reorganization energy ($`\lambda _s`$) and the total free energy of nuclear plus electronic solvation ($`\mathrm{\Delta }G_s`$). We start with describing the solvent branch followed by the outline of the solute part.
### IV.1 Solvent
The calculation of the structure factors in the solvent branch in Fig. 6 requires a set of experimental input parameters: $`m`$ (gas-phase dipole moment), $`\alpha `$ (gas-phase dipolar polarizability), $`ϵ_{\mathrm{}}`$ (high-frequency dielectric constant), $`ϵ_s`$ (static dielectric constant), and $`\sigma `$ (effective hard sphere diameter of the solvent molecules). The hard sphere diameter is obtained from the experimental compressibility of the solvent by fitting it to the compressibility found from the generalized van der Waals (vdW) equation of state Schmid and Matyushov (1995). Based on these parameters, an analytical procedure has been recently proposed to calculate $`S^{L,T}(k)`$ Matyushov (2004b). This parameterization, called parametrized polarization structure factors (PPSF), makes use of the analytical solution of the mean-spherical approximation (MSA) for dipolar fluids Wertheim (1971). The MSA solution gives $`S^{L,T}(k)`$ in terms of the Baxter function $`Q(k\sigma ,\eta )`$ appearing as solution of Percus-Yevick integral equations for hard sphere fluids Gray and Gubbins (1984)
$$S(k\sigma ,\eta )=|Q(k\sigma ,\eta )|^2,$$
(65)
where
$$\begin{array}{cc}\hfill Q(k\sigma ,\eta )=& 112\eta _0^1e^{ik\sigma t}\hfill \\ & \left[a(\eta )(t^21)/2b(\eta )(t1)\right]dt\hfill \end{array}$$
(66)
and $`a(\eta )=(1+2\eta )/(1\eta )^2`$, $`b(\eta )=3\eta /2(1\eta )^2`$. For a fluid of hard sphere molecules, $`\eta =(\pi /6)\rho \sigma ^3`$ is the packing density, equal to the ratio of the volume of the solvent molecules to the volume of the liquid. In the MSA, the $`S^{L,T}(k)`$ are obtained by setting $`\eta =2\xi `$ for $`S^L(k)`$ and $`\eta =\xi `$ for $`S^T(k)`$ in Eq. (65). Here, $`\xi `$ is the MSA polarity parameter which can be related either to the dipolar density $`y_{\text{eff}}`$ or to the static dielectric constant $`ϵ_s`$ Wertheim (1971).
Two problems arise when dealing with the reorganization energy calculations using the polarization structure factors from the MSA. First, one needs a general procedure which would provide the nuclear structure factors $`S_n^{L,T}(k)`$ in polarizable solvents in contrast to total structure factors $`S^{L,T}(k)`$ given by the MSA solution. Such a formalism should thus exclude (quantum) fluctuations of the induced solvent dipoles $`𝐩_j^0`$ which are not included in the nuclear polarization field (Fig. 4). Second, the MSA does not give a consistent description of the dielectric properties of polar solvents, i.e. the polarity parameters $`\xi `$ calculated from $`y_{\text{eff}}`$ and $`ϵ_s`$ are quite different. The PPSF procedure goes around the second problem by considering $`y_{\text{eff}}`$ and $`ϵ_s`$ as two independent input parameters used to calculate $`S^{L,T}(k)`$. A convenient way to introduce the two-parameter scheme is to specify two separate polarity parameters which are obtained from the longitudinal and transverse structure factors at $`k=0`$:
$$\begin{array}{cc}\hfill \frac{(12\xi ^L)^4}{(1+4\xi ^L)^2}=& S^L(0),\hfill \\ \hfill \frac{(1+\xi ^T)^4}{(12\xi ^T)^2}=& S^T(0).\hfill \end{array}$$
(67)
Separate definitions of $`\xi ^L`$ and $`\xi ^T`$ in terms of $`S^L(0)`$ and $`S^T(0)`$ \[Eq. (46)\] allows us to incorporate contributions to macroscopic dielectric properties which are not present in the model of dipolar HS fluids. Specifically, the magnitude of parameter $`y_{\text{eff}}`$, calculated according to Wertheim’s 1-RPT algorithm Wertheim (1979), defines the solvent dipolar strength which strongly affects the dielectric constant. However, $`ϵ_s`$ also depends on such factors as solvent quadrupolar moment Stell et al. (1981), solvent non-sphericity, etc. The influence of these factors is incorporated into $`S^{L,T}(0)`$ through the dielectric constant. Similarly, the polarity parameters $`\xi _n^L`$ and $`\xi _n^T`$ are calculated from Eq. (67) with $`S^{L,T}(0)`$ replaced by $`S_n^{L,T}(0)`$ taken from Eq. (47).
Dipolar projections of the structure factors of molecular liquids modeled by site-site interaction potentials have been studied previously Fonseca and Ladanyi (1990); Raineri and Friedman (1993); Skaf and Ladanyi (1995); Perng and Ladanyi (1999). The PPSF procedure has also been tested against MC simulations of dipolar hard sphere fluids Matyushov (2004b). However, the structure factors arising from the nuclear polarization as well as the applicability of the PPSF to non-spherical molecules with site-site potentials have not been previously tested. This is the aim of the Monte Carlo (MC) and MD simulations carried out in this study. The details of the simulation protocol are given in Appendix A and here we focus only on the results.
Figure 7 shows the comparison of the transverse and longitudinal components of the nuclear structure factors calculated from the PPSF and from MC simulations. The MC simulations (dashed lines in Fig. 7) have been performed on a fluid of 1372 polarizable dipolar hard spheres characterized by dipole moment $`m`$, diameter $`\sigma `$, and isotropic polarizability $`\alpha `$ ($`(m^{})^2=\beta m^2/\sigma ^3=1.0`$, $`\alpha ^{}=\alpha /\sigma ^3=0.06`$, Appendix A). Since the simulation protocol generates the induced polarization in equilibrium with the nuclear configuration of the solvent Gupta and Matyushov (2004), the generated ensamble yields the nuclear polarization in the Pekar partitioning Pekar (1963).
The PPSF nuclear structure factors are calculated by the relations:
$$S_n^T(k)=|Q(k\sigma ,\xi _n^T)|^2$$
(68)
and
$$S_n^L(k)=|Q(\kappa k\sigma ,2\xi _n^L)|^2.$$
(69)
In Eq. (69), $`\kappa =0.95`$ is an empirical parameter introduced for a better agreement between the PPSF and MC simulations of non-polarizable dipolar fluids Milischuk and Matyushov (2005b). The simulations and the PPSF agree well in the entire range of solvent polarizabilities $`\alpha ^{}=\alpha /\sigma ^3=0.010.08`$ studied by simulations Milischuk and Matyushov (2005b).
The MSA solution in Eq. (65) was derived for a model liquid of dipolar hard spheres. The parameterization introduced by the PPSF suggests to use the experimental $`ϵ_s`$ to accommodate empirically the features which are not included in the MSA solution. Two factors, often present in real polar solvents, molecular quadrupoles and non-sphericity, are expected to affect significantly the form of the structure factors. Therefore, we have performed MD simulations for two solvents with well-developed force fields, water Jorgensen et al. (1983) and acetonitrile Edwards et al. (1984). Water is a relatively symmetric molecule with a very large quadrupole moment $`Q`$ com (c) ($`(Q^{})^2=\beta Q^2/\sigma ^5=1.1`$) among commonly used molecular solvents. On the other hand, acetonitrile has a small quadrupole moment ($`(Q^{})^2=0.13`$), but the molecule is very non-spherical with the aspect ratio $`3`$. Therefore, these two extreme cases may provide a good test of the ability of the PPSF to incorporate the complications related to molecular specifics of the solvents in terms of their macroscopic dielectric constants.
Figure 8 (lower panel) shows the comparison of the simulation results for TIP3P water to the PPSF. A slightly wrong positioning of the longitudinal peak may be related to a different hard sphere diameter of TIP3P water (see Table 5 in Appendix A) compared to the hard sphere diameter of water at ambient conditions used in scaling wavevectors in Figure 8. A downward scaling of $`\sigma `$ by just 5% results in a very good match between calculated and simulated structure factors. As expected, the steric effects of packing the solvent molecules in dense liquids is the main factor determining the position of the longitudinal peak. This indeed is seen in Fig. 9 for simulations of acetonitrile. The effective hard sphere diameter obtained from solvent compressibility does not accommodate the fact that linear dipoles tend to pack side-to-side pointing in opposite directions. The longitudinal thus peak effectively reflects a lower molecular diameter. The preferential opposite orientation of the dipoles leads to a low Kirkwood factor and the dielectric constant much lower than one would expect for a dipolar solvent with such large dipole moment ($`4.12`$ D for the force field by Edwards, Madden, and McDonald Edwards et al. (1984)). As a result, the transverse structure factor does not change with $`k`$ as much as it does for hard sphere dipolar liquids (cf. Figs. 7 and 8 to Fig. 9). As is seen, the PPSF with $`ϵ_s`$ from MD simulations accommodates this feature of the solvent quite well.
Figure 8 compares on the common scale the $`k`$-dependence of the longitudinal and transverse components of the electrostatic energy density of complex 1, $`k^2_\mathrm{\Delta }^{\text{eff}}(𝐤)_{\widehat{𝐤}}`$ and $`k^2|\mathrm{\Delta }E_0^T(𝐤)|^2_{\widehat{𝐤}}`$, with the longitudinal and transverse components of the polarization structure factors ($`\mathrm{}_{\widehat{𝐤}}`$ refers to the average over the orientations of the wavevector $`𝐤`$). This comparison shows that details of the molecular structure of the polar solvent affecting the range of $`k`$-values beyond the limit of $`k\pi /\sigma `$ are insignificant for the calculation of the reorganization energy and the free energy gap. Therefore, the discrepancies in the position of the longitudinal peak between the simulations and the PPSF do not noticeably affect the results of calculations. This statement also applies to the range of $`k`$-values ($`k>2\pi /l_s`$, where $`l_s`$ is the characteristic distance between partial charges within the solvent molecule) at which the multipolar approximation for the charge distribution within the solvent molecules breaks down. The charge-charge structure factors calculated on site-site interaction potentials Bopp et al. (1996); Perng et al. (1996a); Skaf (1997); Bopp et al. (1998); Omelyan (1999); Raineri and Friedman (1999); Perng and Ladanyi (1999) then decay to zero instead of approaching the unity limit ($`S^{L,T}(k)1`$ at $`k\mathrm{}`$) of multipolar approximations Fonseca and Ladanyi (1990); Skaf and Ladanyi (1995); Bopp et al. (1998). The range of $`k`$-values where the inaccuracy of the multipolar approximation becomes significant lays beyond the range of small $`k`$-values affecting the calculation of thermodynamic properties unless the solute is much smaller than the solvent.
### IV.2 Solute
The solute branch of the calculation algorithm (Fig. 6) consists of the numerical calculation of the Fourier transform of the electric field outside the solute placed in the vacuum. The direct-space electric fields in the initial and final states of the solute are given by a superposition of electric fields produced by partial charges $`q_{0k}^i`$
$$𝐄_{0i}(𝐫)=\underset{k=1}{\overset{M_0}{}}q_{0k}^i\frac{𝐫𝐫_{0k}}{|𝐫𝐫_{0k}|^3},$$
(70)
where the sum runs over $`M_0`$ partial charges localized on solute atoms. The field $`𝐄_{0i}(𝐫)`$ is Fourier transformed in the region $`\mathrm{\Omega }`$ accessible to the solvent molecules \[Eq. (33)\]. The region $`\mathrm{\Omega }`$ is generated by assigning vdW radii to all atoms of the solute and then adding the hard sphere radius $`\sigma /2`$ of the solvent ($`\sigma =2.87`$ Å for water and 4.14 Å for acetonitrile). This creates the solvent-accessible surface (SAS). The definition of the solute field thus requires atomic coordinates and vdW radii of $`N_0`$ atoms of the solute and $`M_0`$ partial charges $`q_{0k}`$ to be used in Eq. (70) (indicated as $`x_{0k}`$, $`y_{0k}`$, $`z_{0k}`$, $`q_{0k}`$ in Fig. 6).
The infinite-space Fourier transform of the Coulomb electric field \[Eq. (33)\] is numerically divergent Matyushov (2004b). This numerical problem is obviated by splitting the region of integration into the inner part between the SAS and a cutoff sphere and the region outside the cutoff sphere. The Fourier transform within the sphere is calculated numerically by the Fast Fourier Transform (FFT) technique Press et al. (1996) on a cube with the center at the geometrical center of the DSA complex
$$𝐫_c=N_0^1\underset{k=1}{\overset{N_0}{}}𝐫_{0k}.$$
(71)
The length of the cube is chosen by multiplying the maximum extension of the molecule measured from $`𝐫_c`$ by a factor of 9. This choice yields a sufficiently small increment of the $`𝐤`$-grid necessary for the inverted-space integration and, at the same time, avoids numerical errors arising from artificial periodicity imposed by a finite-size numerical FFT technique. The FFT calculation was done on a grid of dimension $`256\times 256\times 256`$ and the step of 0.5 Å. Calculations on complex 1 involved 143 atoms holding partial charges $`q_{0k}^i`$. The charge shifts ($`\mathrm{\Delta }q_k=q_{0k}^2q_{0k}^1`$) and coordinates used in the solvent reorganization and free energy calculations are the same as those reported in Ref. Ungar et al., 1999. The individual (i.e., initial and final state) charges used in the reaction free energy calculations are also taken from Ref. Ungar et al., 1999. The field $`\stackrel{~}{𝐄}_{0i}(𝐤)`$ obtained by combining the numerical and analytical parts is used to calculate the longitudinal and transverse components of the electrostatic energy density in Eqs. (60) and (61). These components are then used in the $`k`$-integrals with the polarization structure factors (Eqs. (53)–(58); also see Fig. 6).
## V Results and comparison to experiment
### V.1 Solvent reorganization energy
The solvent reorganization energy of complex 1 was previously obtained from MD simulations of this complex in TIP3P water Ungar et al. (1999). The permanent dipole moment in this force field is enhanced from the vacuum dipole of 1.87 D to 2.35 D to account for water polarizability. This results in a dielectric constant of $`ϵ_s=97.5`$ from our simulations, which agrees well with $`ϵ_s=97.0`$ found in the literature Guillot (2002). Table 1 lists the results of calculations of the reorganization energy with structure factors from the PPSF (column 5) and from MD simulations (column 7). The density of the solvent in the $`NVT`$ simulations was adjusted at each temperature in order to reproduce the expansivity $`\alpha _p=2.96\times 10^4`$ K<sup>-1</sup> of TIP3P water Paschek (2004). The temperature derivative of the reorganization energy thus gives the constant-pressure reorganization entropy corresponding to conditions normally employed in experiment,
$$S_\lambda =(\lambda _s/T)_P.$$
(72)
Overall, there is an exceptionally good agreement between the reorganization energies calculated by using the structure factors from PPSF and MD simulations. This is not surprising in view of the very good agreement between the two sets of structure factors shown in Fig. 8.
The PPSF result at 298 K, $`\lambda _s=64.11`$ kcal/mol, also compares well with the direct calculation of the reorganization energy from MD simulations, where the value of 60.9 kcal/mol was reported Ungar et al. (1999). The electrostatic forces in those simulations were cut off at distances greater than 10.1 Å. The cutoff is expected to lower the reorganization energy compared to that of an infinite system. In order to estimate the effect of the interaction cutoff, we have calculated the reorganization energy for a fictitious solute with the distance 10.1 Å added to the radius of each atom exposed to the solvent. This contribution amounts to 7.1 kcal/mol. Column 6 in Table 1 shows the results of calculations when the $`k`$-dependent polarization structure factors are replaced by their $`k=0`$ values. The gap in $`\lambda _s`$ values between columns 5 and 6 thus quantifies the contribution of the non-local part of solvent response to the reorganization energy. The last (10) column in Table 1 shows the PPSF calculations using parameters of ambient water. In these calculations, the gas phase dipole moment $`m=1.87`$ D is renormalized by the polarizability effect to give $`m^{}=2.43`$ D (Wertheim’s 1-RPT formalism Wertheim (1979); Gupta and Matyushov (2004)). Despite this renormalization, $`\lambda _s`$ in this calculation is substantially ($``$ 30 %) smaller than in the calculations using parameters of TIP3P water. TIP3P water thus appears to produce stronger solvation than ambient water.
Table 1 also presents two components of the solvent reorganization energy produced by solvent quadrupoles: $`\lambda _q`$ is the second cumulant of the coupling of the solute electric field gradient to solvent quadrupole moment Matyushov and Voth (1999); Milischuk and Matyushov (2005a) whereas $`\lambda _{pq}`$ is a mixed term arising from correlated fluctuations of dipoles and quadrupoles positioned at different solvent molecules Matyushov and Voth (1999); Milischuk and Matyushov (2005c). The resulting solvent reorganization energy is the sum of the dipolar component $`\lambda _p`$ and two quadrupolar components:
$$\lambda _s=\lambda _p+\lambda _{pq}+\lambda _q.$$
(73)
The problem of quadrupolar solvent reorganization has recently attracted much attention Perng et al. (1996a, b); Matyushov and Voth (1999); Jeon and Kim (2001) in connection with new experimental data showing appreciable solvent reorganization in non-dipolar solvents Britt et al. (1995); Reynolds et al. (1996); Kulinowski et al. (1995); Khajehpour and Kauffman (2000); Read et al. (2000). However, the components $`\lambda _q`$ and $`\lambda _{pq}`$ constitute only a small fraction of the overall reorganization energy despite a relatively high reduced quadrupole of water, $`\beta Q^2/\sigma ^5=1.1`$ (cf. to $`(Q^{})^2=0.13`$ of acetonitrile). For the rest of the paper we will therefore assume
$$\lambda _s\lambda _p.$$
(74)
We note that the value $`\lambda _s=69.7`$ kcal/mol calculated for TIP3P water with the account of water quadrupoles is in remarkable agreement with $`\lambda _s=68`$ kcal/mol obtained by correcting the simulated values Ungar et al. (1999) by the finite-size cutoff effects.
The dependence of $`\lambda _s`$ and the reorganization entropy on $`ϵ_{\mathrm{}}`$ are given in Table 2. In these calculations, the vacuum dipole moment of water, 1.83 D, was held constant along with the total dielectric constant $`ϵ_s=78.0`$. The change in $`ϵ_{\mathrm{}}`$ was achieved by varying the polarizability $`\alpha `$ according to the Clausius-Mossotti equation
$$\frac{ϵ_{\mathrm{}}1}{ϵ_{\mathrm{}}+2}=8\eta \alpha /\sigma ^3,$$
(75)
where $`\eta =(\pi /6)\rho \sigma ^3`$ is the solvent packing fraction.
Two drastically different predictions for the effect of solvent polarizability on $`\lambda _s`$ can be found in the literature. The classical Marcus two-sphere model Marcus (1993) predicts a drop of $`\lambda _s`$ by about a factor of 0.6 when going from $`ϵ_{\mathrm{}}=1.0`$ to $`ϵ_{\mathrm{}}=1.8`$. On the other hand, simulations using non-polarizable and polarizable versions of the water force field predict almost no dependence of $`\lambda _s`$ on solvent polarizability Bader and Berne (1996); com (d). The actual situation is in between of the two extremes. The reorganization energy does drop with increasing $`ϵ_{\mathrm{}}`$, but not as much as is predicted by continuum models Gupta and Matyushov (2004). On the other hand, the change is sufficient to make simulations based on non-polarizable solvent models unreliable.
The situation for the dependence of $`\lambda _s`$ on $`ϵ_{\mathrm{}}`$ is illustrated in Fig. 10, where continuum results for complex 1 obtained with the DelPhi Poisson-Boltzmann solver Rocchia et al. (2002) are compared to the calculations within the NRFT. The dielectric calculations with the vdW dielectric cavity (denoted “cont./vdW” in Fig. 10) show a substantial drop of $`\lambda _s`$ with $`ϵ_{\mathrm{}}`$. The dependence on $`ϵ_{\mathrm{}}`$ is much weaker in the NRFT (see also Table 2). The weak dependence of $`\lambda _s`$ on $`ϵ_{\mathrm{}}`$ is the result of the cancellation of two competing factors: the decrease of the longitudinal structure factor in the range of small $`k`$-values with increasing $`ϵ_{\mathrm{}}`$ (Fig. 5) compensated by an increase in $`y_p`$ due to higher solvent dipole $`m^{}`$ in more polarizable solvents. We note that this cancellation is strongly affected by the $`k`$-dependence of the polarization structure factors in the range of small $`k`$-values contributing to the $`k`$-integral and cannot be reduced to the cancellation of the $`y_p`$ factor in $`\lambda _s`$ \[Eqs. (53) and (57)\] with $`y_p`$ in the denominator in Eq. (47), resulting in the Pekar factor of continuum electrostatics.
The continuum limit of the NRFT is obtained when the dependence on the wavevector $`k`$ is neglected in the solvent structure factors and one assumes $`S^{L,T}(k)S^{L,T}(0)`$ and $`S_n^{L,T}(k)S_n^{L,T}(0)`$. When this assumption is incorporated in the microscopic calculations (marked NRFT/$`S(0)`$ in Fig. 10), the resultant reorganization energy gains the strong dependence on $`ϵ_{\mathrm{}}`$ characteristic of continuum theories. The continuum limit of the microscopic theory corresponds, however, to the dielectric cavity coinciding with the SAS. The corresponding DelPhi calculation (marked cont./SAS in Fig. 10) indeed goes parallel with the continuum limit of the NRFT. The distinction between these two results arises from the mean-field approximation used in the NRFT formulation and different handling of the polarizability effects in the two formulations (additive in the continuum and non-additive in the microscopic formulation Milischuk and Matyushov (2005b)). Note that the mean-field approximation is more accurate, when compared to the exact solution of the Li-Kardar-Chandler equation, in the full microscopic formulation than in its continuum limit Matyushov (2004b). The exact formulation of the theory, which does not involve the mean-field approximation, gives the solution of the Poisson equation in its continuum limit.
The numerical values for the reorganization energies shown in Fig. 10 are given in Table 2. The comparison between the microscopic and continuum calculations is instructive. At $`ϵ_{\mathrm{}}=1`$, $`\lambda _s`$ from the vdW continuum is much higher than the microscopic calculation, while $`\lambda _s`$ from the SAS continuum is close to the microscopic result. With increasing $`ϵ_{\mathrm{}}`$, on the other hand, $`\lambda _s`$ from the vdW continuum falls down almost to the microscopic value. The continuum calculation with the vdW cavity may thus appear in a reasonable accord with microscopic calculations or experiment due to the mutual cancellation of errors.
Along with reorganization energies, Table 2 lists reorganization entropies $`S_\lambda `$ \[Eq. (72)\]. Note that $`S_\lambda `$ obtained from the PPSF calibrated on TIP3P water is in a reasonable agreement with the corresponding value obtained with the structure factors from MD simulations: 84.1 e.u. and 69.9 e.u., respectively. The dielectric continuum calculation gives the wrong sign for the entropy in accord with previous reports Matyushov (1993); Vath et al. (1999). Also the magnitude of $`S_\lambda `$ is substantially higher in the microscopic theory than in the continuum calculation (cf. columns 4 and 6 in Table 2).
A similar trend is seen for the reaction free energy gap (Table 3) for which the reaction entropy is defined as
$$\mathrm{\Delta }S_s=\left(\mathrm{\Delta }G_s/T\right)_P.$$
(76)
Although the sign of $`\mathrm{\Delta }S_s`$ is correct in the continuum calculations, the entropy magnitude is much lower than in the NRFT, similar to a previous report for a different ET system Vath et al. (1999), where $`\mathrm{\Delta }S_s`$ was experimentally obtained from temperature dependent absorption and emission charge-transfer bands. Since the analytical theory seems to be consistent with the computer experiment, one needs a test against experimental data. Unfortunately, experimental evidence on the solvent entropic effects on ET reactions is very limited (see Ref. Zimmt and Waldeck, 2003 for a recent review).
### V.2 ET rate constant
The calculations of the temperature dependent reorganization energy and equilibrium energy gap can be compared to experimental Arrhenius law measurements Ogawa et al. (1993) for complex 1. Transition metal-based charge-transfer complexes are commonly characterized by metal-ligand vibrational frequencies Ungar et al. (1999) in the range $`\omega _v300500`$ cm<sup>-1</sup>, substantially lower than frequencies $`\omega _v11001500`$ cm<sup>-1</sup> normally assigned to C$``$C skeletal vibrations of organic donor-acceptor complexes. Therefore, Eqs. (1), (2), and (21) with the full quantum-mechanical description of vibrations and temperature-induced populations of vibrational states should be used for the ET rate in complex 1. Unfortunately, our calculations provide only the solvent component of the free energy gap. Its gas-phase component is unknown and the electronic coupling entering the Golden Rule ET rate in Eq. (1) is known with uncertainty Ungar et al. (1999). These two parameters ($`\mathrm{\Delta }G_g`$ and $`V_{12}`$) were varied in fitting the experimental activation enthalpy $`\mathrm{\Delta }H^{}=9.5`$ kcal/mol and the experimental activation entropy $`\mathrm{\Delta }S^{}/k_\text{B}=5.6`$ e.u. Ogawa et al. (1993). Note that the experimental quantity is an effective entropy, including contributions due to the electronic coupling element as well as solvation and inner-sphere vibrational modes Ungar et al. (1999). The Arrhenius analysis was performed by the linear regression of $`\mathrm{ln}(k_{ET}/T)`$ vs $`1/T`$ based on the transition-state expression
$$k_{\text{ET}}=\frac{k_\text{B}T}{h}e^{\mathrm{\Delta }G^{}(T)/k_\text{B}T}.$$
(77)
The rate constant was calculated based on Eqs. (1) and (21), with $`\lambda _s`$ and $`\mathrm{\Delta }G_s`$ varied linearly with temperature using the calculated entropies (Tables 2 and 3). The results of calculations are listed in Table 4. The fitted electronic coupling $`V_{12}`$ falls in the range of values given by electronic structure calculations Ungar et al. (1999) using the semiempirical INDO/s model by Zerner and co-workers Zerner et al. (1980). The equilibrium gap obtained from the fit is appreciably more negative than $`\mathrm{\Delta }G25.4`$ kcal/mol estimated from the redox potentials of separate donor and acceptor sites, based on the high spin ground state of the Co<sup>2+</sup> product (it has been argued Ungar et al. (1999) that the less exothermic low spin Co<sup>2+</sup> product may be the relevant one in the experimentally observed process). Neglecting the vibrational excitations in the analysis (0-0 transition only) results in a much lower activation enthalpy and a substantially more negative activation entropy (second row in Table 4).
The relatively low frequency of metal-ligand vibrations in transition metal complexes results in a dense manifold of vibrational levels (Fig. 11) which are partially populated at room temperature. The change of the vibrational populations with temperature may result in a contribution to the overall activation entropy Brunschwig et al. (1980). This, however, does not happen for complex 1 when $`\lambda _s`$ and $`\mathrm{\Delta }G_s`$ are fixed at their 298 K values. The dashed lines in Fig. 12 show the enthalpy and entropy of activation as a function of the vibrational frequency at constant temperature and $`\lambda _v`$. Increasing the vibrational frequency makes vibrational excitations less accessible, but this is seen to have little effect on the activation entropy and enthalpy.
This situation changes when the temperature dependence of $`\lambda _s`$ and $`\mathrm{\Delta }G_s`$ is included in the calculations of the Arrhenius activation parameters. In this case, the temperature dependence of the ET energy gap results in a change of the vibrational quantum numbers corresponding to the maximum vibronic contribution. The splitting of the activation barrier into the entropic and enthalpic contribution then becomes sensitive to the choice of $`\omega _v`$ (Fig. 12, solid lines). This sensitivity may be important for the interpretation of experimental data since the correct definition of the effective vibrational frequency \[Eq. (15)\] increases in importance once the temperature dependence of the solvation parameters is introduced into the analysis of reaction rates. The classical Marcus-Hush equation with $`\lambda _v=0`$ replaces the sum over all possible vibronic transitions with a single 0-0 transition. The result is a significantly lower enthalpy and more negative entropy of activation (Table 4).
## VI Discussion
The most relevant question in comparing microscopic solvation theories with the dielectric continuum approximation is why the latter has allowed to describe so many systems after proper parameterization of dielectric cavities, despite drastic approximations involved. The microscopic NRFT formulation contains dielectric continuum as its limit, allowing us to address this question. The continuum limit is obtained by neglecting the spatial correlations between solvent dipoles, i.e. by neglecting the $`k`$-dependence in the polarization response functions. This implies that polarization structure factors are replaced by their $`k=0`$ values (Fig. 13). This replacement is not a good approximation for the transverse structure factor, which changes quite sharply even for small $`k`$-values, but may be a reasonable approximation for the longitudinal structure factor, which is relatively flat in the range of $`k`$-values significant for solvation thermodynamics. However, for most charge configurations, even for the point dipole Matyushov (2004a), the contribution of transverse polarization to the solvation free energy is relatively small Matyushov (2004b) ($`10`$% in our calculations for complex 1 in water). Therefore, the inaccurate continuum approximation for the transverse structure factor does not significantly affect the results of calculations.
The continuum estimates for the polarization structure factors result in the following inequalities between the continuum and microscopic longitudinal and transverse components of the reorganization energy
$$\lambda _s^{L,\text{cont}}<\lambda _s^L,\lambda _s^{T,\text{cont}}>\lambda _s^T.$$
(78)
The sharp change of the transverse structure factor at small $`k`$-values is responsible for a substantial overestimate of the transverse component of solvation by continuum models Matyushov (2004a, b). This overestimate manifests itself in solvation dynamics. The transverse polarization dynamics is much slower than the longitudinal polarization dynamics Bagchi and Chandra (1991). Therefore, continuum models predict biphasic solvation dynamics with an appreciable slow component due to transverse polarization relaxation. This slow component is not observed in computer simulations of solvation dynamics Kumar and Maroncelli (1995) and it does not show up in the microscopic calculations reported in Ref. Matyushov, 2005.
The relatively flat form of the longitudinal structure factors at low $`k`$-values does not mean that replacing $`S^{L,T}(k)`$ by $`S^{L,T}(0)`$ gives accurate numbers for the solvation free energy and/or the reorganization energy. A moderate increase of $`S^L(k)`$ in the range of wavevectors contributing to the $`k`$-integral substantially affects the calculated values of solvation free energies (cf. columns 5 and 6 in Table 1). Moreover, the gap between the microscopic and continuum values changes with the solvent dielectric parameters (see, e.g., Fig. 10). This observation practically means that there is fundamentally no unique scheme for defining the dielectric cavity applicable to all solvent polarities.
The dominance of longitudinal polarization fluctuations in solvation thermodynamics is also responsible for experimentally observed linear trends of the reorganization energy with the Pekar factor Powers and Meyer (1980); Grampp and Jaenicke (1984); Hupp et al. (1993) \[Eq. (48)\]. Even at the continuum level, the polarization response function for a solute of complex shape is not represented by the Pekar factor appearing in the longitudinal projection of the solvent response function Brunschwig et al. (1986). However, large separation of charges is responsible for the predominantly longitudinal response of the solvent, and continuum reorganization energies calculated for complex 1 in polar solvents correlate well with the Pekar factor (Fig. 14a). If fact, an equally good correlation is seen in respect to the Lippert-Mataga polarity parameter commonly used for solvation of dipoles (Fig. 14b):
$$f_n^{\text{LM}}=\frac{ϵ_s1}{2ϵ_s+1}\frac{ϵ_{\mathrm{}}1}{2ϵ_{\mathrm{}}+1}.$$
(79)
The use of a particular parameter does not therefore tell much about the nature of the solute charge distribution and, obviously, reflects a linear relation between $`c_0`$ and $`f_n^{\text{LM}}`$ for common solvents.
The results of the current microscopic calculations are shown by triangles in Fig. 14. These numbers do not exhibit a linear dependence, although the extent of scatter is not uncommon for ET experiment. The comparison of the continuum and microscopic dependence on the solvent polarity does not permit a clear distinction between the two formulations. Where the distinction becomes clear is for the reorganization entropy in strongly polar solvents. Figure 15 shows reorganization entropies $`S_\lambda `$ calculated in continuum (DelPhi Rocchia et al. (2001) Poisson-Bolzmann solver) and microscopic (NRFT) theories. The continuum calculation reflects the temperature variation of the Pekar factor $`c_0`$:
$$(c_0/T)_P=ϵ_{\mathrm{}}^2(ϵ_{\mathrm{}}/T)_Pϵ_s^2(ϵ_s/T)_P$$
(80)
In low-polarity solvents, $`c_0`$ is mostly influenced by the static dielectric constant, which has a negative temperature derivative. The continuum reorganization entropy (closed circles in Fig. 15) is positive and is close to the microscopic result (open squares in Fig. 15). The continuum estimate of the temperature variation of $`\lambda _s`$ in low-polarity solvents thus gives a semi-quantitative account of the experimental observations Liang et al. (1989). In strongly polar solvents, the temperature derivative of $`c_0`$ is mostly influenced by the high-frequency dielectric constant, and continuum $`S_\lambda `$ is nagative. In this case, the predictions of the continuum model significantly depart from both the microscopic calculations and many experimental measurements Grampp and Jaenicke (1984); Elliott et al. (1998); Nelsen et al. (1999); Derr and Elliott (1999); Vath et al. (1999); Vath and Zimmt (2000); Zhao et al. (2001), showing positive reorganization entropies.
The microscopic calculations presented here show a relatively weak dependence of the reorganization energy on the solvent high-frequency dielectric constant, in qualitative accord with available computer simulation data Bader and Berne (1996); Ando (2001); Gupta and Matyushov (2004). Testing this theoretical prediction experimentally may become problematic because of the narrow range of $`ϵ_{\mathrm{}}`$ values available for common polar solvents. We note, however, that the problem of the weak dependence of the reorganization energy on $`ϵ_{\mathrm{}}`$ is related to the problem of correct sign of the reorganization entropy. The strong dependence of the continuum reorganization energy on $`ϵ_{\mathrm{}}`$ is one of major factors shifting the continuum reorganization entropy to the range of positive values.
The calculations of the quadrupolar component of the solvent reorganization energy presented here confirm the conclusion previously reached for Stokes shifts in coumarin-153 optical dye Matyushov and Newton (2001): quadrupolar solvation is insignificant in most commonly used polar solvents, and the dipolar approximation for the solvent charge distribution is sufficient for most practical calculations.
###### Acknowledgements.
D.V.M. thanks the Donors of The Petroleum Research Fund, administered by the American Chemical Society (39539-AC6), for support of this research. M.D.N. was supported by DE-AC02-98CH10886 at Brookhaven National Laboratory. The authors are grateful to Prof. G. A. Voth for sharing the structural data on the polypeptide-linked donor-acceptor complex. This is publication #596 from the ASU Photosynthesis Center.
## Appendix A Simulation and analysis.
The MC simulations of dipolar-polarizable hard sphere solvents shown in Fig. 7 were done as described in Ref. Gupta and Matyushov, 2004. Simulations of $`6\times 10^5`$ cycles long were run for 1372 polarizable molecules with periodic boundary conditions and the reaction field cutoff of dipole-dipole interactions. The MD simulations were carried out with the force field of 3-site acetonitrile (ACN3) by Edwards, Madden, and McDonald Edwards et al. (1984) and the 3-site model of water (TIP3P) by Jorgensen et al. Jorgensen et al. (1983) (Table 5). The site-site interaction potential is given by the sum of the Lennard-Jones (LJ) and Coulomb interaction potentials:
$$E_{\alpha \beta }=4\epsilon _{\alpha \beta }\left[\left(\frac{\sigma _{\alpha \beta }}{r_{\alpha \beta }}\right)^{12}\left(\frac{\sigma _{\alpha \beta }}{r_{\alpha \beta }}\right)^6\right]+\frac{q_\alpha q_\beta }{r_{\alpha \beta }},$$
(81)
where the LJ parameters are taken according to the Lorentz-Bertholet rules: $`\epsilon _{\alpha \beta }=\sqrt{\epsilon _\alpha \epsilon _\beta }`$ and $`\sigma _{\alpha \beta }=(\sigma _\alpha +\sigma _\beta )/2`$. All simulations were done with the DL\_POLY molecular dynamics package Smith and Forester (1996). We run two sets of MD simulations in the temperature range from 288 K to 308 K with a 5 K step. The timestep in each simulation is 5 fs. All MD simulation are 20 ns long.
We used the Nosé-Hoover thermostat Hoover (1985) for the ACN3 simulations with the relaxation parameter of 0.5 fs. This value ensures good stabilization of the total system energy. The energy drift for ACN3 is only about 0.1%. The simulation box was constructed to include 256 ACN3 molecules in a cube with the side length $`L=28.2025`$ Å at T=298 K to reproduce the experimental mass density of acetonitrile, $`\rho _M`$=0.777 g/cm<sup>3</sup>. The side length is adjusted at each temperature to account for temperature expansion with the experimental volume expansion coefficient $`\alpha _p=1.38\times 10^3`$ K<sup>-1</sup>.
In simulations of TIP3P water, 256 molecules reside in a cube with the side length of $`L=19.7744`$ Å at 298 K. The system is coupled to the Berendsen Berendsen et al. (1984) thermostat with the relaxation time of 0.1 fs. The drift in total energy of about 0.1 % is observed. The liquid mass density $`\rho _M=0.9896`$ g/cm<sup>3</sup> and the volume expansion coefficient $`\alpha _p=2.96\times 10^3`$ K<sup>-1</sup> are taken from Ref. Paschek, 2004. The latter value is close to the experimental expansion coefficient of ambient water, $`\alpha _p=2.6\times 10^3`$ K<sup>-1</sup>.
The cutoff for short-range LJ interaction is 13 Å for ACN3 and 9 Å for TIP3P. For long-range Coulomb interactions, Ewald summation from DL\_POLY Allen and Tildesley (1996) is used for ACN3 and smoothed particle mesh (SPME) Essmann et al. (1995) Ewald is adopted for TIP3P. Ewald summation parameters are the convergence parameter $`\alpha `$ and the maximum wavenumber $`k_{x,y,z}^{max}`$. The parameter sets $`\alpha =0.24`$ Å<sup>-1</sup>, $`k_{x,y,z}^{max}=7`$ Å<sup>-1</sup>, and $`\alpha `$ =0.35 Å<sup>-1</sup>, $`k_{x,y,z}^{max}=8`$ Å<sup>-1</sup> were used for ACN3 and TIP3P respectively.
The structure factors have been calculated as the variance of longitudinal and transverse projections of the $`𝐤`$-space solvent polarization
$$𝐌(k)=(1/m)\underset{i=1}{\overset{N}{}}𝐦_ie^{i𝐤𝐫_i},$$
(82)
where $`𝐦_i=_aq_a𝐫_i^a`$ is a dipole moment of the $`i`$th molecule and the sum runs over the $`N`$ molecules in the simulation box. The static dielectric constant is given in terms of the $`k=0`$ variance as follows Neumann (1986)
$$\epsilon _s=1+3y𝐌(\mathrm{𝟎})^2/N,$$
(83)
where $`y=(4\pi /9)\rho m^2/k_\text{B}T`$.
|
warning/0506/cond-mat0506615.html
|
ar5iv
|
text
|
# Comment on a Phys. Rev. Lett. paper: 94 (2005) 146402: Orbital symmetry and Electron Correlation in NaxCoO2
## I Appendix A
Description of the trigonal distortion within the crystal-field theory 20 . After our paper Phys. Rev. B 63 (2001) 172404 13 .
The 25 levels, originated from the <sup>5</sup>D term, and their eigenfunctions have been calculated by the direct diagonalization of the Hamiltonian (1) within the $`|LSL_zS_z`$ base. It takes a form:
$$H_d=H_{cub}+\lambda LS+B_0^2O_0^2+\mu _B(L+g_sS)B$$
()
The separation of the crystal-electric-field (CEF) Hamiltonian into the cubic and off-cubic part is made for the illustration reason as the cubic crystal field is usually very predominant. In the crystallographic structure of FeBr<sub>2</sub> the Fe ion is surrounded by 6 Br ions. Despite of the hexagonal elementary cell Br ions form the almost octahedral surrounding - it justifies the dominancy of the octahedral crystal field interactions. Moreover, this octahedral surrounding in the hexagonal unit cell can be easily distorted along the local cube diagonal - in the hexagonal unit cell this local cube diagonal lies along the hexagonal c axis. The related distortion can be described as the trigonal distortion of the local octahedron. The cubic CEF Hamiltonian takes, for the z axis along the cube diagonal, the form
$$H_{cub}=\frac{2}{3}B_4(O_4^020\sqrt{2}O_4^3)$$
(1)
where $`O_m^n`$ are the Stevens operators. The last term in Eq. (1) allows studies of the influence of the magnetic field 13 .
For remembering, the octahedral CEF Hamiltonian with the $`z`$ axis along the cube edge takes a form:
$$H_d=B_4(O_4^0+5O_4^4)$$
(2)
dedicated to Hans Bethe, Kramers and John H. Van Vleck, pioneers of the crystal-field theory, to the 75<sup>th</sup> anniversary of the crystal-field theory, and to the Pope John Paul II, a man of freedom and honesty in life and in Science.
|
warning/0506/quant-ph0506007.html
|
ar5iv
|
text
|
# Quantum damped oscillator I: dissipation and resonances
## 1 Introduction
The damped harmonic oscillator is one of the simplest quantum systems displaying the dissipation of energy. Moreover, it is of great physical importance and has found many applications especially in quantum optics. For example it plays a central role in the quantum theory of lasers and masers .
As is well known there is no room for the dissipative phenomena in the standard Hilbert space formulation of Quantum Mechanics. The Schrödinger equation defines one-parameter unitary group and hence the quantum dynamics is perfectly time-reversible. The usual approach to include dissipation is the quantum theory of open systems . In this approach the dynamics of a quantum system is no longer unitary but it is defined by a semigroup of completely positive maps in the space of density operators (for recent reviews see e.g. ).
There is, however, another way to describe dissipative quantum systems based on the old idea of Bateman . Bateman has shown that to apply the standard canonical formalism of classical mechanics to dissipative and non-Hamiltonian systems, one can double the numbers of degrees of freedom, so as to deal with an effective isolated classical Hamiltonian system. The new degrees of freedom may be assumed to represent a reservoir. Applying this idea to damped harmonic oscillator one obtains a pair of damped oscillators (so called Bateman’s dual system): a primary one and its time reversed image. The Bateman dual Hamiltonian has been rediscovered by Morse and Feshbach and Bopp and the detailed quantum mechanical analysis was performed by Feshbach and Tikochinski . The quantum Bateman system was then analyzed by many authors (see the detailed historical review with almost 600 references!).
Surprisingly, this system is still worth to study and it shows its new interesting features. Recently it was analyzed in in connection with quantum field theory and quantum groups (see also ). Different approach based on the Chern-Simons theory was applied in . In a recent paper a damped oscillator was quantized by using Feynman path integral formulation (see also ). Moreover, the corresponding geometric phase was calculated and found to be directly related to the ground-state energy of the standard one-dimensional linear harmonic oscillator. Bateman’s system has been also studied as a toy model for the recent proposal by ’t Hooft about deterministic quantum mechanics .
In the present paper we propose a slightly different approach to the Bateman system. The unusual feature of the Bateman Hamiltonian is that being a self-adjoint operator it displays a family of complex eigenvalues. We show that these eigenvalues correspond to the poles of energy eigenvectors and the corresponding resolvent operator when continued to the complex energy plane. The similar analysis for the toy model of a quantum damped system was performed in . Eigenvectors corresponding to the poles of the resolvent are well known in the scattering theory as resonant states . It shows that the appearance of resonances is responsible for the dissipation in the Bateman system. Obviously, the time evolution is perfectly reversible when considered on the corresponding system Hilbert space $`=L^2(^2)`$. It is given by the 1-parameter group of unitary transformations $`U(t)=e^{i\widehat{H}t}`$. It turns out that there are two natural subspaces $`𝒮_\pm `$ such that $`U(t)`$ restricted to $`𝒮_\pm `$ defines only two semigroups: $`U(t0)`$ on $`𝒮_{}`$, and $`U(t0)`$ on $`𝒮_+`$. These two semigroups are related by the time reversal operator $`𝒯`$ (see Section 6).
Our analysis is based on a new representation of the Bateman Hamiltonian, cf. Section 4. This representation is directly related to the old observation of Pontriagin (see Section 3 for review) that any non-Hamiltonian system of the form
$$\dot{x}_k=X_k(x_1,\mathrm{},x_N),k=1,2,\mathrm{},N,$$
(1.1)
may be treated as a Hamiltonian one in the extended phase-space $`(x_1,\mathrm{},x_N,p_1,\mathrm{},p_N)`$ with the Hamiltonian
$$H(x_1,\mathrm{},x_N,p_1,\mathrm{},p_N)=\underset{k=1}{\overset{N}{}}p_kX_k(x_1,\mathrm{},x_N).$$
(1.2)
Note, that the above Hamiltonian has exactly the form considered by ’t Hooft .
From the mathematical point of view the natural language to analyze the Bateman system is the so called rigged Hilbert space approach to quantum mechanics . There are two natural rigged Hilbert spaces, or Gel’fand triplets, corresponding to subspaces $`𝒮_\pm `$. We shall comment on that in Section 8.
## 2 Bateman Hamiltonian
The classical equation of motion for one-dimensional damped oscillator with unit mass reads
$$\ddot{x}+2\gamma \dot{x}+\kappa x=\mathrm{\hspace{0.33em}0},$$
(2.1)
where $`\gamma >0`$ denotes the damping constant. Introducing Bateman’s dual system
$$\ddot{y}2\gamma \dot{y}+\kappa y=\mathrm{\hspace{0.33em}0},$$
(2.2)
one may derive booth equations from the following Lagrangian
$$L(x,\dot{x},y,\dot{y})=\dot{x}\dot{y}\kappa xy+\gamma (x\dot{y}\dot{x}y).$$
(2.3)
Introducing canonical momenta
$$p_x=\dot{y}\gamma y,p_y=\dot{x}+\gamma x,$$
(2.4)
one easily finds the corresponding Hamiltonian
$$H(x,y,p_x,p_y)=p_xp_y\gamma (xp_xyp_y)+\omega ^2xy,$$
(2.5)
where
$$\omega =\sqrt{\kappa \gamma ^2}.$$
(2.6)
Throughout the paper we shall consider the underdamped case, i.e. $`\kappa >\gamma ^2`$.
Now, assuming symmetric Weyl ordering the canonical quantization is straightforward and leads to the following self-adjoint operator in the Hilbert space $`L^2(^2,dxdy)`$:
$$\widehat{H}=\widehat{H}_0+\widehat{H}_I,$$
(2.7)
where
$$\widehat{H}_0=\widehat{p}_x\widehat{p}_y+\omega ^2\widehat{x}\widehat{y},$$
(2.8)
and
$$\widehat{H}_I=\frac{\gamma }{2}\left((\widehat{x}\widehat{p}_x+\widehat{p}_x\widehat{x})(\widehat{y}\widehat{p}_y+\widehat{p}_y\widehat{y})\right).$$
(2.9)
Note, that
$$[\widehat{H}_0,\widehat{H}_I]=0.$$
(2.10)
Following Feshbach and Tichochinsky one introduces annihilation and creation operators
$`\widehat{A}`$ $`=`$ $`{\displaystyle \frac{1}{2\sqrt{\mathrm{}\omega }}}\left[(\widehat{p}_x+\widehat{p}_y)i\omega (\widehat{x}+\widehat{y})\right],`$ (2.11)
$`\widehat{B}`$ $`=`$ $`{\displaystyle \frac{1}{2\sqrt{\mathrm{}\omega }}}\left[(\widehat{p}_x\widehat{p}_y)i\omega (\widehat{x}\widehat{y})\right].`$ (2.12)
They satisfy the standard CCRs
$$[\widehat{A},\widehat{A}^{}]=[\widehat{B},\widehat{B}^{}]=1,$$
(2.13)
and all other commutators vanish. It turns out that the transformed Hamiltonian is given by (2.7) with
$$\widehat{H}_0=\mathrm{}\omega (\widehat{A}^{}\widehat{A}\widehat{B}^{}\widehat{B}),\widehat{H}_I=i\mathrm{}\gamma (\widehat{A}^{}\widehat{B}^{}\widehat{A}\widehat{B}).$$
(2.14)
It is easy to see that the dynamical symmetry associated with the Bateman’s Hamiltonian is that of $`SU(1,1)`$. Indeed, constructing the following generators:
$`\widehat{J}_1`$ $`=`$ $`{\displaystyle \frac{1}{2}}(\widehat{A}^{}\widehat{B}^{}+\widehat{A}\widehat{B}),`$ (2.15)
$`\widehat{J}_2`$ $`=`$ $`{\displaystyle \frac{i}{2}}(\widehat{A}^{}\widehat{B}^{}\widehat{A}\widehat{B}),`$ (2.16)
$`\widehat{J}_3`$ $`=`$ $`{\displaystyle \frac{1}{2}}(\widehat{A}^{}\widehat{A}+\widehat{B}\widehat{B}^{}),`$ (2.17)
one easily shows that they satisfy $`su(1,1)`$ commutation relations:
$$[\widehat{J}_1,\widehat{J}_2]=i\widehat{J}_3,[\widehat{J}_3,\widehat{J}_2]=i\widehat{J}_1,[\widehat{J}_1,\widehat{J}_3]=i\widehat{J}_2.$$
(2.18)
Moreover, the following operator
$$\widehat{J}_0=\frac{1}{2}(\widehat{A}^{}\widehat{A}\widehat{B}^{}\widehat{B}),$$
(2.19)
defines the corresponding $`su(1,1)`$ Casimir operator. One easily shows that
$$\widehat{J}_0^2=\frac{1}{4}+\widehat{J}_3^2\widehat{J}_1^2\widehat{J}_2^2.$$
(2.20)
It is therefore clear that the Hamiltonian (2.14) can be rewritten in terms of $`su(1,1)`$ generators as
$$\widehat{H}_0=2\mathrm{}\omega \widehat{J}_0,\widehat{H}_I=2\mathrm{}\gamma \widehat{J}_2.$$
(2.21)
The algebraic structure arising in this approach enables one to solve the corresponding eigenvalue problem. Let us define two mode eigenvectors $`|n_A,n_B`$:<sup>1</sup><sup>1</sup>1Mathematically oriented reader would prefer
$$(\widehat{A}^{}\widehat{A}1\mathrm{l}_B)|n_A,n_B=n_A|n_A,n_B,(1\mathrm{l}_A\widehat{B}^{}\widehat{B})|n_A,n_B=n_B|n_A,n_B,$$
where $`1\mathrm{l}_A`$ ($`1\mathrm{l}_B`$) denotes the identity operator in “$`A`$-sector” (“$`B`$-sector”).
$$\widehat{A}^{}\widehat{A}|n_A,n_B=n_A|n_A,n_B,\widehat{B}^{}\widehat{B}|n_A,n_B=n_B|n_A,n_B.$$
(2.22)
It is convenient to introduce
$$j=\frac{1}{2}(n_An_B),m=\frac{1}{2}(n_A+n_B),$$
(2.23)
and to label the corresponding eigenvectors of $`\widehat{J}_0`$ and $`\widehat{J}_3`$ by $`|j,m`$ rather than $`|n_A,n_B`$:
$`\widehat{J}_0|j,m`$ $`=`$ $`j|j,m,`$ (2.24)
$`\widehat{J}_3|j,m`$ $`=`$ $`\left(m+{\displaystyle \frac{1}{2}}\right)|j,m.`$ (2.25)
Clearly,
$$j=0,\pm \frac{1}{2},\pm 1,\pm \frac{3}{2},\mathrm{},m=|j|,|j|+1,|j|+2,\mathrm{}.$$
(2.26)
Finally, defining
$$|\psi _{jm}^\pm =\mathrm{exp}\left(\frac{\pi }{2}\widehat{J}_1\right)|jm,$$
(2.27)
one obtains
$$\widehat{H}|\psi _{jm}^\pm =E_{jm}^\pm |\psi _{jm}^\pm ,$$
(2.28)
with
$$E_{jm}^\pm =2\mathrm{}\omega j\pm i\mathrm{}\gamma (2m+1).$$
(2.29)
Let us emphasize that the eigenvectors corresponding to energies (2.29) cannot be normalized and should be considered as generalized eigenvectors not belonging to the Hilbert space of the problem.
## 3 Canonical quantization of non-Hamiltonian systems
As is well known any dynamical system may be regarded as a part of a larger Hamiltonian system. Bateman’s approach is based on adding to the primary system a time reversed (dual) copy. Together they define a Hamiltonian system. There exists, however, a general approach to canonical quantization of non-Hamiltonian systems based on an old observation of Pontriagin . Suppose we are given an arbitrary non-Hamiltonian system described by
$$\dot{𝐱}=𝐗(𝐱),$$
(3.1)
where $`𝐗`$ is a vector field on some configuration space $`Q`$. For simplicity assume that $`Q^N`$, that is, the system has $`N`$ degrees of freedom. This system may be lifted to the Hamiltonian system on the phase space $`𝒫=Q\times ^N`$ as follows: one defines the Hamiltonian $`H:𝒫`$ by
$$H(𝐱,𝐩)=𝐩𝐗(𝐱)=\underset{l=1}{\overset{N}{}}p_lX_l(𝐱),$$
(3.2)
where $`(𝐱,𝐩)=(x_1,\mathrm{},x_N,p_1,\mathrm{},p_N)`$ denote canonical coordinates on $`𝒫`$. The corresponding Hamilton equations read as follows:
$`\dot{x}_k`$ $`=`$ $`\{x_k,H\}=X_k(x),`$ (3.3)
$`\dot{p}_k`$ $`=`$ $`\{p_k,H\}={\displaystyle \underset{l=1}{\overset{N}{}}}p_l{\displaystyle \frac{X_l(x)}{x_k}},`$ (3.4)
for $`k=1,\mathrm{},N`$. In the above formulae $`\{,\}`$ denotes the standard Poisson bracket on $`𝒫`$
$$\{F,G\}=\underset{k=1}{\overset{N}{}}\left(\frac{F}{x_k}\frac{G}{p_k}\frac{G}{x_k}\frac{F}{p_k}\right).$$
(3.5)
Clearly, the formulae (3.3) reproduce our initial dynamical system (3.1) on $`Q`$. The canonical quantization is now straightforward. Assuming the symmetric Weyl ordering one obtains the following formula for the quantum Hamiltonian
$$\widehat{H}_{\mathrm{quantum}}=\mathrm{W}\left(\underset{l=1}{\overset{N}{}}p_lX_l(𝐱)\right),$$
(3.6)
where $`\mathrm{W}(f)`$ denotes the Wigner-Weyl transform of a space-phase function $`f=f(𝐱,𝐩)`$. Recall, that the Wigner-Weyl transform of $`f`$ is defined as follows
$$\widehat{f}=\mathrm{W}(f)=𝑑𝝈𝑑𝝉\stackrel{~}{f}(𝝈,𝝉)\mathrm{exp}\left\{i\underset{k=1}{\overset{N}{}}\left(\sigma _k\widehat{x}_k+\tau _k\widehat{p}_k\right)\right\},$$
(3.7)
where $`\stackrel{~}{f}(𝝈,𝝉)`$ denotes the Fourier transform of $`f(𝐱,𝐩)`$. Clearly, $`\widehat{H}_{\mathrm{quantum}}`$ defines a Schrödinger system in $`L^2(^N,d𝐱)`$.
Consider now a damped harmonic oscillator described by
$$\ddot{x}+2\gamma \dot{x}+\kappa x=0.$$
The above 2nd order equation may be rewritten as a dynamical system on $`^2`$
$`\dot{x}_1`$ $`=`$ $`\gamma x_1+\omega x_2,`$ (3.8)
$`\dot{x}_2`$ $`=`$ $`\gamma x_2\omega x_1,`$ (3.9)
with $`\omega `$ defined in (2.6). Clearly this system is not Hamiltonian if $`\gamma 0`$. However, applying the above Pontriagin procedure one arrives at the Hamiltonian system on $`^4`$ defined by the following damped harmonic oscillator Hamiltonian:
$$H(𝐱,𝐩)=\omega (p_1x_2p_2x_1)\gamma (p_1x_1+p_2x_2).$$
(3.10)
The corresponding Hamilton equations of motion read
$$\dot{𝐱}=\widehat{F}𝐱,\dot{𝐩}=\widehat{F}^\mathrm{T}𝐩,$$
(3.11)
where
$$\widehat{F}=\left(\begin{array}{cc}\gamma & \omega \\ \omega & \gamma \end{array}\right),$$
(3.12)
and $`\widehat{F}^\mathrm{T}`$ denotes the transposition of $`\widehat{F}`$. One may ask what is the relation between Bateman’s Hamiltonian (2.5) and that obtained via Pontriagin procedure (3.10). Surprisingly they are related by the following simple canonical transformation $`(x,y,p_x,p_y)(x_1,x_2,p_1,p_2)`$:
$`x_1`$ $`=`$ $`{\displaystyle \frac{p_y}{\sqrt{\omega }}},p_1=\sqrt{\omega }y`$ (3.13)
$`x_2`$ $`=`$ $`\sqrt{\omega }x,p_2={\displaystyle \frac{p_x}{\sqrt{\omega }}}.`$ (3.14)
Assuming the symmetric Weyl ordering one obtains the following representation of the quantum Bateman’s Hamiltonian (2.7) with
$$\widehat{H}_0=\omega (\widehat{p}_1\widehat{x}_2\widehat{p}_2\widehat{x}_1),$$
(3.15)
and
$$\widehat{H}_I=\frac{\gamma }{2}(\widehat{p}_1\widehat{x}_1+\widehat{x}_1\widehat{p}_1+\widehat{p}_2\widehat{x}_2+\widehat{x}_2\widehat{p}_2).$$
(3.16)
## 4 Spectral properties of the Hamiltonian
### 4.1 Polar representation
The formula (3.10) for $`H`$ considerably simplifies in polar coordinates:
$$x_1+ix_2=re^{i\phi }.$$
Defining the corresponding conjugate momenta
$$p_\phi =L_3,p_r=\frac{\mathrm{𝐱𝐩}}{r},$$
(4.1)
with $`L_3`$ denoting 3rd component of $`𝐋=𝐱\times 𝐩`$ in $`^3`$, one finds
$$H=\omega p_\phi \gamma rp_r.$$
(4.2)
The Hamilton equations in polar representation have the following simple form:
$$\dot{\phi }=\omega ,\dot{p}_\phi =0,$$
(4.3)
and
$$\dot{r}=\gamma r,\dot{p}_r=\gamma p_r.$$
(4.4)
The polar representation nicely shows that the Hamiltonian dynamics consists in pure oscillation in $`\phi `$–sector and dissipation (pumping) in $`r`$–sector ($`p`$–sector). In our opinion it is the most convenient representation to deal with .
The quantization of (4.2) leads to (2.7) with
$$\widehat{H}_0=\omega \widehat{p}_\phi =i\omega \mathrm{}\frac{}{\phi },$$
(4.5)
and
$$\widehat{H}_I=i\gamma \mathrm{}\left(r\frac{}{r}+1\right)=\gamma \left(r\widehat{p}_r\frac{i\mathrm{}}{2}\right),$$
(4.6)
where the radial momentum $`\widehat{p}_r`$ is defined by
$$\widehat{p}_r=i\mathrm{}\left(\frac{}{r}+\frac{1}{2r}\right).$$
(4.7)
One easily finds the polar representation of the $`su(1,1)`$ generators:
$`\widehat{J}_1`$ $`=`$ $`{\displaystyle \frac{\mathrm{}}{4}}\left({\displaystyle \frac{^2}{r^2}}+{\displaystyle \frac{1}{r^2}}{\displaystyle \frac{^2}{\varphi ^2}}\right){\displaystyle \frac{1}{4\mathrm{}}}r^2,`$ (4.8)
$`\widehat{J}_2`$ $`=`$ $`{\displaystyle \frac{i}{2}}\left(r{\displaystyle \frac{}{r}}+1\right),`$ (4.9)
$`\widehat{J}_3`$ $`=`$ $`{\displaystyle \frac{1}{4\mathrm{}}}r^2+{\displaystyle \frac{i}{2}}{\displaystyle \frac{}{\varphi }}{\displaystyle \frac{\mathrm{}}{4}}\left({\displaystyle \frac{^2}{r^2}}+{\displaystyle \frac{1}{r^2}}{\displaystyle \frac{^2}{\varphi ^2}}\right).`$ (4.10)
together with the Casimir operator
$$\widehat{J}_0=\frac{i}{2}\frac{}{\varphi }.$$
(4.11)
Note, that unitary evolution generated by $`\widehat{H}`$ is given by
$$\widehat{U}(t)=e^{i\widehat{H}t/\mathrm{}}=e^{i\widehat{H}_0t/\mathrm{}}e^{i\widehat{H}_It/\mathrm{}}=e^{\gamma t}\mathrm{exp}\left(\omega t\frac{}{\phi }\right)\mathrm{exp}\left(\gamma tr\frac{}{r}\right),$$
(4.12)
and hence
$$(\widehat{U}(t)\psi )(r,\phi )=e^{\gamma t}\psi (e^{\gamma t}r,\phi +\omega t).$$
(4.13)
### 4.2 Complete set of eigenvectors
It is evident that $`\widehat{H}`$ defines an unbounded operator in $`=L^2(^2,dx_1dx_2)`$. It has continuous spectrum $`\sigma (\widehat{H})=(\mathrm{},\mathrm{})`$. To find the corresponding generalized eigenvectors let us note that in polar representation the Hilbert space $``$ of square integrable functions in $`^2`$ factorizes as follows:
$$L^2(^2,dx_1dx_2)=L^2([0,2\pi ),d\phi )L^2(_+,rdr).$$
(4.14)
Therefore, the spectral problem splits into two separate problems in $`L^2([0,2\pi ),d\phi )`$ and $`L^2(_+,rdr)`$. One easily finds
$$\widehat{H}\mathrm{\Psi }_{l\lambda }=E_{l\lambda }\mathrm{\Psi }_{l\lambda },$$
(4.15)
with
$$E_{l\lambda }=\mathrm{}(l\omega +\lambda \gamma ).$$
(4.16)
The corresponding eigenvectors $`\mathrm{\Psi }_{l\lambda }`$ are defined by
$$\mathrm{\Psi }_{l\lambda }(r,\phi )=\mathrm{\Phi }_l(\phi )R_\lambda (r),$$
(4.17)
where
$$\mathrm{\Phi }_l(\phi ):=\frac{e^{il\phi }}{\sqrt{2\pi }},l=0,\pm 1,\pm 2,\mathrm{},$$
(4.18)
and
$$R_\lambda (r)=\frac{r^{(i\lambda +1)}}{\sqrt{2\pi }},\lambda .$$
(4.19)
Note, that $`\mathrm{\Phi }_lL^2([0,2\pi ),d\phi )`$ whereas $`R_\lambda `$ does not belong to $`L^2(_+,rdr)`$.
One easily shows that the family $`\mathrm{\Psi }_{l\lambda }`$ satisfies
$$_0^{2\pi }_0^{\mathrm{}}\overline{\mathrm{\Psi }_{l\lambda }}(r,\phi )\mathrm{\Psi }_{l^{}\lambda ^{}}(r,\phi )r𝑑r𝑑\phi =\delta _{ll^{}}\delta (\lambda \lambda ^{}),$$
(4.20)
and
$$\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}_{\mathrm{}}^{\mathrm{}}\overline{\mathrm{\Psi }_{l\lambda }}(r,\phi )\mathrm{\Psi }_{l\lambda }(r^{},\phi ^{})𝑑\lambda =\frac{1}{r}\delta (rr^{})\delta (\phi \phi ^{}).$$
(4.21)
They imply the following resolution of identity
$$1\mathrm{l}=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}_{\mathrm{}}^{\mathrm{}}𝑑\lambda |\mathrm{\Psi }_{l\lambda }\mathrm{\Psi }_{l\lambda }|,$$
(4.22)
and the spectral resolution of Hamiltonian
$$\widehat{H}=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}_{\mathrm{}}^{\mathrm{}}𝑑\lambda E_{l\lambda }|\mathrm{\Psi }_{l\lambda }\mathrm{\Psi }_{l\lambda }|,$$
(4.23)
### 4.3 Feynman propagator
Let us calculate the corresponding Feynman propagator
$$K(𝐱,t|𝐱^{},t^{})=𝐱|\widehat{U}(tt^{})|𝐱^{},$$
(4.24)
where $`\widehat{U}(\tau )=\mathrm{exp}(i\widehat{H}\tau /\mathrm{})`$. Using polar representation one finds
$$K(r,\phi ,t|r^{},\phi ^{},t^{})=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}_{\mathrm{}}^{\mathrm{}}e^{iE_{l\lambda }\tau /\mathrm{}}\mathrm{\Psi }_{l\lambda }(r,\phi )\overline{\mathrm{\Psi }_{l\lambda }}(r^{},\phi ^{})𝑑\lambda ,$$
(4.25)
with $`\tau =tt^{}`$. Now, using (4.17) one obtains
$$K(r,\phi ,t|r^{},\phi ^{},t^{})=K_1(r,t|r^{},t^{})K_2(\phi ,t|\phi ^{},t^{}),$$
(4.26)
where the radial and azimuthal propagators are given by
$$K_1(r,t|r^{},t^{})=_{\mathrm{}}^{\mathrm{}}e^{i\lambda \gamma \tau }R_\lambda (r)\overline{R_\lambda }(r^{})𝑑\lambda ,$$
(4.27)
and
$$K_2(\phi ,t|\phi ^{},t^{})=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}e^{i\omega l\tau }\mathrm{\Phi }_l(\phi )\overline{\mathrm{\Phi }_l}(\phi ^{}),$$
(4.28)
respectively. Finally, formulae (4.18) and (4.19) imply
$$K_2(\phi ,t|\phi ^{},t^{})=\delta (\phi ^{}\phi \omega \tau ),$$
(4.29)
and
$`K_1(r,t|r^{},t^{})`$ $`=`$ $`{\displaystyle \frac{1}{2\pi }}{\displaystyle \frac{1}{rr^{}}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}e^{i\lambda (\mathrm{ln}r^{}\mathrm{ln}r\gamma \tau )}𝑑\lambda `$ (4.30)
$`=`$ $`{\displaystyle \frac{1}{rr^{}}}\delta (\mathrm{ln}r^{}\mathrm{ln}r\gamma \tau )=e^{\gamma \tau }{\displaystyle \frac{\delta (r^{}re^{\gamma \tau })}{r^{}}}.`$
Therefore, the time evolution is given by
$`\psi _t(r,\phi )`$ $`=`$ $`{\displaystyle _0^{2\pi }}{\displaystyle _0^{\mathrm{}}}K(r,\phi ,t|r^{},\phi ^{},t^{}=0)\psi _0(r^{},\phi ^{})r^{}𝑑r^{}𝑑\phi ^{}`$ (4.31)
$`=`$ $`e^{\gamma t}\psi _0(e^{\gamma t}r,\phi +\omega t),`$
which perfectly agrees with (4.13).
## 5 Analyticity and complex eigenvalues
Now we are going to relate the energy eigenvectors $`\mathrm{\Psi }_{n\lambda }`$ corresponding to the real spectrum $`E_{n\lambda }`$ with the family of discrete complex eigenvalues of the Bateman’s Hamiltonian. Let us consider the distribution $`\mathrm{\Psi }_{n\lambda }`$ with $`\lambda `$, i.e. for any test function $`\varphi (r,\phi )`$
$$\mathrm{\Psi }_{l\lambda }(\varphi )=\varphi |\mathrm{\Psi }_{l\lambda }=_0^{\mathrm{}}r^{i\lambda }\overline{\varphi _l}(r)𝑑r,$$
(5.1)
where
$$\varphi _l(r)=\frac{1}{2\pi }_0^{2\pi }e^{il\phi }\varphi (r,\phi )𝑑\phi .$$
(5.2)
Now, the analytical properties of $`\mathrm{\Psi }_{l\lambda }`$ depend upon the behavior of $`\varphi _l(r)`$ at $`r=0`$. A distribution $`r^\alpha `$ acting on the space of smooth functions $`S(_+)`$
$$S(_+)f_0^{\mathrm{}}r^\alpha \overline{f}(r)𝑑r,$$
(5.3)
is well defined for all $`\alpha `$ except the discrete family of points where it may have simple poles (see e.g. ). The location of poles depends upon the behavior of a test function $`f`$ at $`r=0`$. Assuming the most general expansion of $`f(r)`$
$$f(r)=f_0+f_1r+f_2r^2+\mathrm{},$$
(5.4)
the poles are located at $`\alpha =1,2,3,\mathrm{}.`$ However, $`\varphi _l(r)`$ defined in (5.2) is much more regular. It can be observed (see Appendix B.) that $`\varphi _l(r)`$ may be expanded at $`r=0`$ as follows:
$$\varphi _l(r)=a_lr^{|l|}+a_{l+2}r^{|l|+2}+a_{l+4}r^{|l|+4}+\mathrm{}.$$
(5.5)
Therefore, the poles that remain are located at
$$\lambda _{nl}=i(|l|+2n+1),n=0,1,2,\mathrm{}.$$
(5.6)
Moreover, the corresponding residues of $`\mathrm{\Psi }_{l\lambda }`$ are given by
$$\mathrm{Res}\mathrm{\Psi }_{l\lambda }|_{\lambda =\lambda _{nl}}=\frac{1}{\sqrt{(|l|+2n)!}}\frac{𝔣_{nl}^{}}{\sqrt{2\pi }},$$
(5.7)
where
$$𝔣_{nl}^{}(r,\phi )=\mathrm{\Phi }_l(\phi )\frac{i(1)^{|l|+2n}}{\sqrt{(|l|+2n)!}}\frac{\delta ^{(|l|+2n)}(r)}{r}.$$
(5.8)
On the other hand
$$\overline{\mathrm{\Psi }_{l\lambda }}|_{\lambda =\lambda _{nl}}=\sqrt{(|l|+2n)!}\frac{\overline{𝔣_{nl}^+}}{\sqrt{2\pi }},$$
(5.9)
where
$$𝔣_{nl}^+(r,\phi )=\mathrm{\Phi }_l(\phi )\frac{r^{|l|+2n}}{\sqrt{(|l|+2n)!}}.$$
(5.10)
Now, the crucial observation is that $`𝔣_{nl}^\pm `$ satisfy
$$\widehat{J}_0|𝔣_{nl}^\pm =\frac{l}{2}|𝔣_{nl}^\pm ,$$
(5.11)
and
$$\widehat{J}_2|𝔣_{nl}^\pm =\pm \frac{i}{2}(|l|+2n+1)|𝔣_{nl}^\pm ,$$
(5.12)
which proves that they define eigenvectors of $`\widehat{H}`$
$$\widehat{H}|𝔣_{nl}^\pm =E_{nl}^\pm |𝔣_{nl}^\pm ,$$
(5.13)
corresponding to complex eigenvalues
$$E_{nl}^\pm =\mathrm{}\omega l\pm i\mathrm{}\gamma (|l|+2n+1).$$
(5.14)
The above formula is equivalent to the Bateman’s spectrum (2.29) after the following identification
$$j=\frac{l}{2},$$
(5.15)
and
$$m=\frac{1}{2}(|l|+2n)=|j|+n,$$
(5.16)
which reproduces condition (2.26). In terms of $`(n_A,n_B)`$ one has
$`n_A`$ $`=`$ $`{\displaystyle \frac{1}{2}}(|l|+l)+n,`$ (5.17)
$`n_B`$ $`=`$ $`{\displaystyle \frac{1}{2}}(|l|l)+n.`$ (5.18)
We have therefore the following relation between $`|\psi _{jm}^\pm `$ and $`|𝔣_{nl}^\pm `$:
$$|\psi _{jm}^\pm =|𝔣_{2j,m|j|}^\pm ,$$
(5.19)
that is, $`|𝔣_{nl}^\pm `$ defined in (5.8) and (5.10) may be regarded as a particular representation of $`|\psi _{jm}^\pm `$.
Let us introduce two important classes of functions : consider the space of complex functions $`f:`$. A smooth function $`f=f(\lambda )`$ is in the Hardy class from above $`_+^2`$ (from below $`_{}^2`$) if $`f(\lambda )`$ is a boundary value of an analytic function in the upper, i.e. $`\text{Im}\lambda 0`$ (lower, i.e. $`\text{Im}\lambda 0`$) half complex $`\lambda `$-plane vanishing faster than any power of $`\lambda `$ at the upper (lower) semi-circle $`|\lambda |\mathrm{}`$. Now, define
$`𝒮_{}=\left\{\varphi 𝒮\right|\mathrm{\Psi }_{l\lambda }|\varphi _{}^2\},`$ (5.20)
that is, $`\varphi 𝒮_{}`$ iff the complex function
$$\lambda \mathrm{\Psi }_{l\lambda }|\varphi ,$$
is in the Hardy class from below $`_{}^2`$. Equipped with this mathematical notion let us consider an arbitrary test function $`\varphi 𝒮_{}`$. The resolution of identity (4.22) implies
$`\varphi (r,\phi )={\displaystyle \underset{l=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑\lambda \mathrm{\Psi }_{l\lambda }(r,\phi )\mathrm{\Psi }_{l\lambda }|\varphi .`$ (5.21)
Now, since $`\mathrm{\Psi }_{l\lambda }|\varphi _{}^2`$, we may close the integration contour along the lower semi-circle $`|\lambda |\mathrm{}`$ (see Figure 1).
Hence, due to the residue theorem one obtains
$`\varphi (r,\phi )=2\pi i{\displaystyle \underset{l=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\text{Res}\mathrm{\Psi }_{l\lambda }(r,\phi )|_{\lambda =\lambda _{nl}}\mathrm{\Psi }_{l\lambda }|\varphi |_{\lambda =\lambda _{nl}}.`$ (5.22)
Finally, using (5.7) and (5.9) one gets
$$\varphi (r,\phi )=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\underset{n=0}{\overset{\mathrm{}}{}}𝔣_{nl}^{}(r,\phi )𝔣_{nl}^+|\varphi .$$
(5.23)
We have proved, therefore, that the subspace $`𝒮_{}𝒮`$ gives rise to the following resolution of identity
$$1\mathrm{l}_{}1\mathrm{l}|_𝒮_{}=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\underset{n=0}{\overset{\mathrm{}}{}}|𝔣_{nl}^{}𝔣_{nl}^+|.$$
(5.24)
The same arguments lead us to the following spectral resolution of $`\widehat{H}`$ restricted to $`𝒮_{}`$:
$$\widehat{H}_{}\widehat{H}|_𝒮_{}=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\underset{n=0}{\overset{\mathrm{}}{}}E_{nl}^{}|𝔣_{nl}^{}𝔣_{nl}^+|,$$
(5.25)
with $`E_{nl}^{}`$ defined in (5.14). Introducing the following family of operators
$$\widehat{P}_{nl}^{}=|𝔣_{nl}^{}𝔣_{nl}^+|,$$
(5.26)
the spectral decompositions (5.24) and (5.25) may be rewritten as follows
$$1\mathrm{l}_{}=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\underset{n=0}{\overset{\mathrm{}}{}}\widehat{P}_{nl}^{},$$
(5.27)
and
$$\widehat{H}_{}=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\underset{n=0}{\overset{\mathrm{}}{}}E_{nl}^{}\widehat{P}_{nl}^{}.$$
(5.28)
Note, that
$$\widehat{P}_{nl}^{}\widehat{P}_{n^{}l^{}}^{}=\delta _{nl}\delta _{n^{}l^{}}\widehat{P}_{nl}^{},$$
(5.29)
that is, the family $`\widehat{P}_{nl}^{}`$ seems to play the role of the family of orthogonal projectors. Note, however, that $`\widehat{P}_{nl}^{}`$ are not hermitian.
## 6 Time reversal
It was shown in that Bateman’s Hamiltonian is time reversal invariant
$$𝒯^{}\widehat{H}𝒯=\widehat{H},$$
(6.1)
where $`𝒯`$ denote the anti-unitary time reversal operator. Moreover, it turns out that both $`\widehat{J}_0`$ and $`\widehat{J}_2`$ satisfy
$$𝒯^{}\widehat{J}_0𝒯=\widehat{J}_0,𝒯^{}\widehat{J}_2𝒯=\widehat{J}_2.$$
(6.2)
Let us define
$$\mathrm{\Xi }_{l\lambda }=𝒯\mathrm{\Psi }_{l\lambda }.$$
(6.3)
In analogy with (4.22) and (4.23) one has the following resolution of identity
$$1\mathrm{l}=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}_{\mathrm{}}^{\mathrm{}}𝑑\lambda |\mathrm{\Xi }_{l\lambda }\mathrm{\Xi }_{l\lambda }|,$$
(6.4)
and spectral resolution of the Hamiltonian
$$\widehat{H}=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}_{\mathrm{}}^{\mathrm{}}𝑑\lambda E_{l\lambda }|\mathrm{\Xi }_{l\lambda }\mathrm{\Xi }_{l\lambda }|.$$
(6.5)
Now, let us introduce another subspace $`𝒮_+`$ in the space of test functions
$`𝒮_+=\left\{\varphi 𝒮\right|\mathrm{\Xi }_{l\lambda }|\varphi _+^2\},`$ (6.6)
that is, $`\varphi 𝒮_+`$ iff the complex function
$$\lambda \mathrm{\Xi }_{l\lambda }|\varphi ,$$
is in the Hardy class from above $`_+^2`$. It is easy to show that
$$𝒮_+=𝒯(𝒮_{}),$$
(6.7)
and vice versa
$$𝒮_{}=𝒯(𝒮_+).$$
(6.8)
Indeed, if $`\varphi 𝒮_{}`$ then $`\mathrm{\Psi }_{l\lambda }|\varphi _{}^2`$. One has therefore
$$\mathrm{\Xi }_{l\lambda }|𝒯\varphi =\varphi |𝒯^{}\mathrm{\Xi }_{l\lambda }=\overline{\mathrm{\Psi }_{l\lambda }|\varphi }_+^2,$$
(6.9)
which implies that $`𝒯\varphi 𝒮_+`$.<sup>2</sup><sup>2</sup>2In the above formulae we have used
$$\psi |𝒜\varphi =\varphi |𝒜^{}\psi ,$$
which holds for any anti-linear operator $`𝒜`$. Moreover
$$𝒮_{}𝒮_+=\{\mathrm{}\}.$$
(6.10)
To prove this property let us assume that $`\varphi 𝒮_{}𝒮_+`$. Since $`\varphi 𝒮_+`$, one has $`\mathrm{\Xi }_{l\lambda }|\varphi _+^2`$. However
$$\mathrm{\Xi }_{l\lambda }|\varphi =\overline{\varphi |𝒯\mathrm{\Psi }_{l\lambda }}=\mathrm{\Psi }_{l\lambda }|𝒯^{}\varphi _+^2.$$
(6.11)
On the other hand $`𝒯^{}𝒮_{}`$ and hence $`\mathrm{\Psi }_{l\lambda }|𝒯^{}\varphi _{}^2`$. Therefore, $`\mathrm{\Psi }_{l\lambda }|𝒯^{}\varphi _{}^2_+^2`$ which means that $`\mathrm{\Psi }_{l\lambda }|𝒯^{}\varphi `$ is en entire function vanishing on the circle $`|\lambda |\mathrm{}`$. However, any entire function is necessarily bounded and hence such $`\varphi `$ which belongs both to $`𝒮_{}`$ and $`𝒮_+`$ does not exist.
Now, take any test function $`\varphi 𝒮_+`$. Formula (6.4) implies
$`\varphi (r,\phi )={\displaystyle \underset{l=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑\lambda \mathrm{\Xi }_{l\lambda }(r,\phi )\mathrm{\Xi }_{l\lambda }|\varphi .`$ (6.12)
Let us continue the eigenvectors $`\mathrm{\Xi }_{l\lambda }`$ for the complex $`\lambda `$ plane. They have simple poles at $`\lambda =\lambda _{nl}`$ with $`\lambda _{nl}`$ defined in (5.6). The corresponding residues of $`\mathrm{\Xi }_{l\lambda }`$ follows from (5.7)
$$\mathrm{Res}\mathrm{\Xi }_{l\lambda }|_{\lambda =\lambda _{nl}}=\frac{1}{\sqrt{(|l|+2n)!}}\frac{𝒯𝔣_{nl}^{}}{\sqrt{2\pi }}.$$
(6.13)
Moreover,
$$\overline{\mathrm{\Xi }_{l\lambda }}|_{\lambda =\lambda _{nl}}=\sqrt{(|l|+2n)!}\frac{𝒯\overline{𝔣_{nl}^+}}{\sqrt{2\pi }}.$$
(6.14)
Now, since $`\mathrm{\Xi }_{n\lambda }|\varphi _+^2`$, we may close the integration contour in (6.12) along the upper semi-circle $`|\lambda |\mathrm{}`$. The residue theorem implies
$`\varphi (r,\phi )=2\pi i{\displaystyle \underset{l=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\text{Res}\mathrm{\Xi }_{l\lambda }(r,\phi )|_{\lambda =\lambda _{nl}}\mathrm{\Xi }_{l\lambda }|\varphi |_{\lambda =\lambda _{nl}}.`$ (6.15)
Finally, using (6.13) and (6.14) one gets
$$\varphi (r,\phi )=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\underset{n=0}{\overset{\mathrm{}}{}}𝒯𝔣_{nl}^{}(r,\phi )\overline{\varphi |𝒯𝔣_{nl}^+},$$
(6.16)
and hence it implies the following resolution of identity on $`𝒮_+`$:
$$1\mathrm{l}|_{𝒮_+}=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\underset{n=0}{\overset{\mathrm{}}{}}𝒯|𝔣_{nl}^{}𝔣_{nl}^+|𝒯^{}.$$
(6.17)
Now, the formula (5.12) together with (6.2) gives
$$\widehat{J}_2𝒯|𝔣_{nl}^\pm =\frac{i}{2}(|l|+2n+1)𝒯|𝔣_{nl}^\pm ,$$
(6.18)
and hence one deduces the following relations between $`|𝔣_{nl}^\pm `$ and time reversed $`𝒯|𝔣_{nl}^\pm `$
$$𝒯|𝔣_{nl}^+=e^{i\alpha _{nl}}|𝔣_{nl}^{},𝒯|𝔣_{nl}^{}=e^{i\alpha _{nl}}|𝔣_{nl}^+,$$
(6.19)
where $`\alpha _{nl}`$ are arbitrary $`(n,l)`$-depended phases. It should be stressed that these phases are physically irrelevant. Actually, one may redefine $`|𝔣_{nl}^+`$ in (5.8) and (5.10) such that these additional phase factors disappear from (6.19). Let us observe that
$$𝒯^2|𝔣_{nl}^\pm =|𝔣_{nl}^\pm ,$$
(6.20)
irrespective of $`\alpha _{nl}`$. Taking into account (6.19) one obtains from (6.17)
$$1\mathrm{l}_+1\mathrm{l}|_{𝒮_+}=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\underset{n=0}{\overset{\mathrm{}}{}}|𝔣_{nl}^{}𝔣_{nl}^+|.$$
(6.21)
The same arguments lead us to the following spectral resolution of $`\widehat{H}`$
$$\widehat{H}_+\widehat{H}|_{𝒮_+}=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\underset{n=0}{\overset{\mathrm{}}{}}E_{nl}^+|𝔣_{nl}^+𝔣_{nl}^{}|,$$
(6.22)
with $`E_{nl}^+`$ defined in (5.14). Finally, introducing
$$\widehat{P}_{nl}^+=|𝔣_{nl}^+𝔣_{nl}^{}|=(\widehat{P}_{nl}^{})^{},$$
(6.23)
with $`\widehat{P}_{nl}^{}`$ defined in (5.26), the spectral decompositions (6.21) and (6.22) may be rewritten as follows
$$1\mathrm{l}_+=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\underset{n=0}{\overset{\mathrm{}}{}}\widehat{P}_{nl}^+,$$
(6.24)
and
$$\widehat{H}_+=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\underset{n=0}{\overset{\mathrm{}}{}}E_{nl}^+\widehat{P}_{nl}^+.$$
(6.25)
## 7 Resonances and dissipation
What is the physical meaning of the complex eigenvalues $`E_{nl}^\pm `$? To answer this question let us consider the resolvent operator of the Bateman’s Hamiltonian
$$\widehat{\mathrm{R}}(\widehat{H},z)=(\widehat{H}z)^1.$$
(7.1)
Using the family of eigenfunctions $`|\mathrm{\Psi }_{l\lambda }`$ one has
$$\widehat{\mathrm{R}}(\widehat{H},z)=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}_{\mathrm{}}^{\mathrm{}}\frac{d\lambda }{E_{l\lambda }z}|\mathrm{\Psi }_{l\lambda }\mathrm{\Psi }_{l\lambda }|,$$
(7.2)
with $`E_{l\lambda }`$ defined in (4.16). Now, using the same technique as in Section 5 one easily finds
$$\widehat{\mathrm{R}}_{}(z)\widehat{\mathrm{R}}(\widehat{H},z)|_S_{}=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\underset{n=0}{\overset{\mathrm{}}{}}\frac{1}{E_{nl}^{}z}\widehat{P}_{nl}^{},$$
(7.3)
with $`P_{nl}^{}`$ defined in (5.26). This shows that $`E_{nl}^{}`$ constitute poles of the resolvent operator on $`𝒮_{}`$. In the same way using the family $`|\mathrm{\Xi }_{l\lambda }`$
$$\widehat{\mathrm{R}}(\widehat{H},z)=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}_{\mathrm{}}^{\mathrm{}}\frac{d\lambda }{E_{l\lambda }z}|\mathrm{\Xi }_{l\lambda }\mathrm{\Xi }_{l\lambda }|,$$
(7.4)
one finds
$$\widehat{\mathrm{R}}_+(z)\widehat{\mathrm{R}}(\widehat{H},z)|_{S_+}=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\underset{n=0}{\overset{\mathrm{}}{}}\frac{1}{E_{nl}^+z}\widehat{P}_{nl}^+,$$
(7.5)
which shows that $`E_{nl}^+`$ constitute poles of the resolvent operator on $`𝒮_+`$. As is well known the poles of the resolvent operator correspond to resonant states. Hence, the complex eigenvalues $`E_{nl}^\pm `$ may be interpreted as resonances of the Bateman’s Hamiltonian. Note that due to the Cauchy theorem operators $`\widehat{P}_{nl}^\pm `$ may be represented by the following integrals
$$\widehat{P}_{nl}^\pm =\frac{1}{2\pi i}_{\gamma _{nl}^\pm }\widehat{\mathrm{R}}_\pm (z)𝑑z,$$
(7.6)
where $`\gamma _{nl}^\pm `$ is any (clockwise) closed curve which encircles a single pole $`z=E_{nl}^\pm `$ (see Figure 2).
Finally, let us turn to the evolution generated by the Bateman’s Hamiltonian. Clearly,
$$t\widehat{U}(t)=\mathrm{exp}(i\widehat{H}t/\mathrm{}),$$
defines a group of unitary operators on the Hilbert space $`L^2(^2)`$. Now, it is easy to see that if $`\psi _{}𝒮_{}`$, then $`\widehat{U}(t)\psi _{}`$ belongs to $`𝒮_{}`$ only if $`t0`$. Similarly, if $`\psi _+𝒮_+`$, then $`\widehat{U}(t)\psi _+`$ belongs to $`𝒮_+`$ only if $`t0`$. Therefore, we have two natural semigroups
$$\widehat{U}_{}(t):𝒮_{}𝒮_{},\mathrm{for}t0,$$
(7.7)
and
$$\widehat{U}_+(t):𝒮_+𝒮_+,\mathrm{for}t0,$$
(7.8)
where
$$\widehat{U}_{}(t)=\widehat{U}(t)|_𝒮_{},\mathrm{and}\widehat{U}_+(t)=\widehat{U}(t)|_{𝒮_+}.$$
(7.9)
One has
$$\psi _{}(t)=\widehat{U}_{}(t)\psi _{}=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}e^{i\omega lt}\underset{n=0}{\overset{\mathrm{}}{}}e^{\gamma (|l|+n+1)t}\widehat{P}_{nl}^{}\psi _{},$$
(7.10)
for $`t0`$, and
$$\psi _+(t)=\widehat{U}_+(t)\psi _+=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}e^{i\omega lt}\underset{n=0}{\overset{\mathrm{}}{}}e^{\gamma (|l|+n+1)t}\widehat{P}_{nl}^+\psi _+,$$
(7.11)
for $`t0`$. It should be clear that these two semigroups are related by the time reversal operator $`𝒯`$: indeed formulae (6.19) imply
$$𝒯\widehat{P}_{nl}^{}𝒯^{}=\widehat{P}_{nl}^+\mathrm{and}𝒯\widehat{P}_{nl}^+𝒯^{}=\widehat{P}_{nl}^{},$$
(7.12)
and hence
$`𝒯\widehat{U}_{}(t)𝒯^{}`$ $`=`$ $`{\displaystyle \underset{l=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}𝒯\left(e^{iE_{nl}^{}t/\mathrm{}}\widehat{P}_{nl}^{}\right)𝒯^{}`$ (7.13)
$`=`$ $`{\displaystyle \underset{l=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}e^{iE_{nl}^+(t)/\mathrm{}}\widehat{P}_{nl}^+=\widehat{U}_+(t),`$
for $`t0`$. Similarly, one finds
$$𝒯\widehat{U}_+(t)𝒯^{}=\widehat{U}_{}(t),$$
(7.14)
for $`t0`$. We have shown that perfectly reversible quantum dynamics $`\widehat{U}(t)`$ on the full Hilbert space $`L^2(^2)`$ is no longer reversible when restricted to the subspaces $`𝒮_{}`$ and $`𝒮_+`$. This effective irreversibility is caused by the presence of resonant states $`|𝔣_{nl}^\pm `$ corresponding to complex eigenvalues $`E_{nl}^\pm `$.
## 8 Conclusions
In this paper we have studied the spectral properties of the Bateman Hamiltonian. It was shown that the complex eigenvalues $`E_{jm}^\pm `$ given by (2.29) corresponds to the poles of the resolvent operator $`\widehat{\mathrm{R}}(\widehat{H},z)=(\widehat{H}z)^1`$. Therefore, the corresponding generalized eigenvectors may be interpreted as resonant states of the Bateman dual system. It proves that dissipation and irreversibility is caused by the presence of resonances.
From the mathematical point of view the Bateman system gives rise to the so called Gel’fand triplet or rigged Hilbert space (see also ). A Gel’fand triplet (rigged Hilbert space) is a collection of spaces
$$\mathrm{\Phi }\mathrm{\Phi }^{},$$
(8.1)
where $``$ is a Hilbert space, $`\mathrm{\Phi }`$ its dense subspace and $`\mathrm{\Phi }^{}`$ is the dual space of continuous linear functionals on $`\mathrm{\Phi }`$. Note, that elements from $`\mathrm{\Phi }^{}`$ do not belong to $``$. This is a typical situation when one deals with the continuum spectrum. The corresponding generalized eigenvectors are no longer elements from the system Hilbert space. They are elements from the dual space $`\mathrm{\Phi }^{}`$, i.e. distributions acting on $`\mathrm{\Phi }`$ . In our case we have two natural Gel’fand triplets:
$$𝒮_{}L^2(^2)𝒮_{}^{},$$
(8.2)
and
$$𝒮_+L^2(^2)𝒮_+^{}.$$
(8.3)
The first triplet corresponds to the forward dynamics $`\widehat{U}_{}`$ and the second one corresponds to the backward semigroup $`\widehat{U}_+`$. A similar analysis based on rigged Hilbert space approach was performed in for a toy model damped system defned by $`\dot{x}=\gamma x`$.
## Appendix A
Let us briefly sketch calculations leading to (5.7) and (5.9). We introduce a distribution $`\mathrm{\Psi }_{l\lambda }`$ acting on a test function $`\varphi (r,\phi )`$ as an antilinear functional defined by the integral
$$\mathrm{\Psi }_{l\lambda }(\varphi )=\varphi |\mathrm{\Psi }_{l\lambda }=\frac{1}{2\pi }_^2e^{il\phi }r^{i\lambda 1}\overline{\varphi }(r,\phi )𝑑S=_0^{\mathrm{}}r^{i\lambda }\overline{\varphi }_l(r)𝑑r,$$
(A.4)
where $`\lambda `$, $`dS=rdrd\phi `$, and $`\varphi _l(r)`$ is given by (5.2). Expanding $`\varphi _l(r)`$ in the power series and rewriting the last integral as
$`{\displaystyle _0^{\mathrm{}}}r^{i\lambda }\overline{\varphi }_l(r)𝑑r`$ $`=`$ $`{\displaystyle _0^1}r^{i\lambda }\left[\overline{\varphi }_l(r)\overline{\varphi }_l(0)r\overline{\varphi }_l^{}(0)\mathrm{}{\displaystyle \frac{r^{l1}}{(l1)!}}\overline{\varphi }_l^{(l1)}(0)\right]𝑑r`$
$`+`$ $`{\displaystyle _1^{\mathrm{}}}r^{i\lambda }\overline{\varphi }_l(r)𝑑r`$
$`+`$ $`{\displaystyle _0^1}r^{i\lambda }\left[\overline{\varphi }_l(0)+r\overline{\varphi }_l^{}(0)+\mathrm{}+{\displaystyle \frac{r^{l1}}{(l1)!}}\overline{\varphi }_l^{(l1)}(0)\right]𝑑r,`$
one can observe that the first two summands are regular for all $`\lambda `$. The last integral, however, equals to
$`{\displaystyle \underset{k=0}{\overset{l1}{}}}{\displaystyle \frac{\overline{\varphi }_l^{(k)}(0)}{k!}}{\displaystyle _0^1}r^{i\lambda }r^k𝑑r={\displaystyle \underset{k=0}{\overset{l1}{}}}{\displaystyle \frac{\overline{\varphi }_l^{(k)}(0)}{k!}}{\displaystyle \frac{1}{ki\lambda +1}}`$ (A.6)
and has simple poles in $`\lambda _k=i(k+1)`$, $`k=0,1,\mathrm{},l1`$. Moreover, one can read from (A.6) that
$$\mathrm{Res}\varphi |\mathrm{\Psi }_{l\lambda }|_{\lambda =i(k+1)}=\frac{\overline{\varphi }_l^{(k)}(0)}{k!}.$$
(A.7)
Finally, using
$$\overline{\varphi }_l^{(k)}(0)=\frac{1}{2\pi }_0^{2\pi }e^{il\phi }\overline{\varphi }^{(k)}(0,\phi )𝑑\phi $$
(A.8)
and
$$\overline{\varphi }^{(k)}(0,\phi )=(1)^k_0^{\mathrm{}}\frac{\delta ^{(k)}(r)}{r}\overline{\varphi }(r,\phi )r𝑑r,$$
(A.9)
we get
$$\overline{\varphi }_l^{(k)}(0)=\frac{(1)^k}{2\pi }_^2e^{il\phi }\frac{\delta ^{(k)}(r)}{r}\overline{\varphi }(r,\phi )𝑑S.$$
(A.10)
But due to (5.5) in the case investigated here $`k=|l|+2n`$, hence the poles are located at $`\lambda _{nl}=i(|l|+2n+1)`$ and
$$\mathrm{Res}\varphi |\mathrm{\Psi }_{l\lambda }|_{\lambda =\lambda _{nl}}=\frac{1}{\sqrt{(|l|+2n)!}}\frac{\varphi |𝔣_{nl}^{}}{\sqrt{2\pi }},$$
(A.11)
where $`𝔣_{nl}^{}`$ is a distribution given by (5.8).
The conjugate distribution $`\overline{\mathrm{\Psi }_{l\lambda }}`$ is defined as
$$\overline{\mathrm{\Psi }_{l\lambda }}(\varphi )=\varphi |\overline{\mathrm{\Psi }_{l\lambda }}=\frac{1}{2\pi }_^2e^{il\phi }r^{i\lambda 1}\overline{\varphi }(r,\phi )𝑑S=_0^{\mathrm{}}r^{i\lambda }\overline{\varphi }_l(r)𝑑r$$
(A.12)
and it is regular in $`\lambda =\lambda _{nl}`$. The poles of $`\varphi |\overline{\mathrm{\Psi }}_{l\lambda }`$ are located at $`\lambda =\overline{\lambda }_{nl}`$. Hence
$$\varphi |\overline{\mathrm{\Psi }_{l\lambda }}|_{\lambda =\lambda _{nl}}=\sqrt{(|l|+2n)!}\frac{\varphi |\overline{𝔣_{nl}^+}}{\sqrt{2\pi }},$$
(A.13)
where $`𝔣_{nl}^+`$ is a distribution given by (5.10).
## Appendix B
Let us briefly proof that $`\varphi _l(r)`$ given by (5.2) has a power series expansion (5.5) starting from $`r^{|l|}`$. Supposing that $`\varphi (x_1,x_2)`$ is an analytic function
$$\varphi (x_1,x_2)=\varphi (0,0)+\underset{k_1,k_2}{}\frac{^{k_1+k_2}\varphi (0,0)}{x_1^{k_1}x_2^{k_2}}x_1^{k_1}x_2^{k_2}$$
(B.1)
in cartesian coordinates $`(x_1,x_2)`$ it is obvious that in polar $`(r,\phi )`$-coordinates one obtains the following expansion for $`\varphi _l(r)`$:
$`\varphi _l(r)`$ $`=`$ $`{\displaystyle \frac{1}{2\pi }}{\displaystyle _0^{2\pi }}e^{il\phi }\varphi (r\mathrm{cos}\phi ,r\mathrm{sin}\phi )𝑑\phi `$ (B.2)
$`=`$ $`{\displaystyle \frac{1}{2\pi }}{\displaystyle _0^{2\pi }}e^{il\phi }\left[\varphi (0,0)+{\displaystyle \underset{k_1,k_2}{}}A_{k_1,k_2}r^{k_1+k_2}(\mathrm{cos}\phi )^{k_1}(\mathrm{sin}\phi )^{k_2}\right]𝑑\phi `$
$`=`$ $`{\displaystyle \frac{1}{2\pi }}{\displaystyle \underset{k_1,k_2}{}}A_{k_1,k_2}r^{k_1+k_2}{\displaystyle _0^{2\pi }}e^{il\phi }(\mathrm{cos}\phi )^{k_1}(\mathrm{sin}\phi )^{k_2}𝑑\phi ,`$
where
$$A_{k_1,k_2}=\frac{^{k_1+k_2}\varphi (0,0)}{x_1^{k_1}x_2^{k_2}},$$
stand for derivatives of $`\varphi (x_1,x_2)`$ in $`(0,0)`$. Now, the question is: for which values of $`kk_1+k_2`$ the sum in (B.2) does not vanish? Clearly, it should be
$$_0^{2\pi }e^{il\phi }(\mathrm{cos}\phi )^{k_1}(\mathrm{sin}\phi )^{k_2}𝑑\phi 0.$$
(B.3)
Using the Newton expansions
$`(\mathrm{cos}\phi )^{k_1}`$ $`=`$ $`\left({\displaystyle \frac{1}{2}}\right)^{k_1}\left(e^{i\phi }+e^{i\phi }\right)^{k_1}=\left({\displaystyle \frac{1}{2}}\right)^{k_1}{\displaystyle \underset{m_1=0}{\overset{k_1}{}}}\left({\displaystyle \genfrac{}{}{0pt}{}{k_1}{m_1}}\right)e^{im_1\phi }e^{i(k_1m_1)\phi },`$
$`(\mathrm{sin}\phi )^{k_2}`$ $`=`$ $`\left({\displaystyle \frac{1}{2i}}\right)^{k_2}\left(e^{i\phi }e^{i\phi }\right)^{k_1}=\left({\displaystyle \frac{1}{2i}}\right)^{k_2}{\displaystyle \underset{m_2=0}{\overset{k_2}{}}}\left({\displaystyle \genfrac{}{}{0pt}{}{k_2}{m_2}}\right)(1)^{k_2m_2}e^{im_2\phi }e^{i(k_2m_2)\phi },`$
one can rewrite (B.3) as
$$\left(\frac{1}{2}\right)^{k_1}\left(\frac{1}{2i}\right)^{k_2}\underset{m_1=0}{\overset{k_1}{}}\underset{m_2=0}{\overset{k_2}{}}\left(\genfrac{}{}{0pt}{}{k_1}{m_1}\right)\left(\genfrac{}{}{0pt}{}{k_2}{m_2}\right)(1)^{k_2m_2}_0^{2\pi }e^{i(l+k2(m_1+m_2))\phi }𝑑\phi 0,$$
(B.4)
hence (B.4) will not vanish iff
$$l+k2(m_1+m_2)=0.$$
(B.5)
Clearly,
$$0m_1k_1,0m_2k_2.$$
(B.6)
Now, let $`l<0`$, so $`l=|l|`$ and
$$k=|l|+2(m_1+m_2)=|l|+2n,$$
(B.7)
where $`n=m_1+m_20`$. Due to (B.6), in order to satisfy (B.7) it should be $`k|l|`$.
On the other hand, if $`l>0`$, then
$$k=l+2(m_1+m_2)l+2k,$$
(B.8)
because of (B.6) and finally $`kl=|l|`$. Note that in this case $`k=l+2(m_1+m_2l)`$, where $`m_1+m_2ln0`$.
As a result we obtained that for a given $`n`$ the lowest power of $`r`$ in expansion (B.2) is $`k_1+k_2=|l|`$. Moreover, in both cases
$$k_1+k_2=k=|l|+2n,n=0,1,2,\mathrm{}$$
(B.9)
holds.
## Acknowledgments
This work was partially supported by the Polish State Committee for Scientific Research Grant Informatyka i inżynieria kwantowa No PBZ-Min-008/P03/03.
|
warning/0506/cond-mat0506224.html
|
ar5iv
|
text
|
# Ground-states of the three-dimensional Falicov-Kimball model
## 1 Introduction
The Falicov-Kimball model (FKM) has become, since its introduction in 1969, one of the most popular examples of a system of interacting electrons with short-range interactions. It has been used in the literature to study a great variety of many-body effects in rare-earth compounds, of which metal-insulator transitions, mixed-valence phenomena, and charge-density waves are the most common examples . The model is based on the coexistence of two different types of electronic states in a given material: localized, highly correlated ionic-like states and extended, uncorrelated, Bloch-like states. It is generally accepted that the above mentioned cooperative phenomena result from a change in the occupation numbers of these electronic states, which remain themselves basically unchanged in their character. Taking into account only the intra-atomic Coulomb interaction between the two types of states, the Hamiltonian of the spinless FKM can be written as the sum of three terms:
$$H=\underset{ij}{}t_{ij}d_i^+d_j+U\underset{i}{}f_i^+f_id_i^+d_i+E_f\underset{i}{}f_i^+f_i,$$
(1)
where $`f_i^+`$, $`f_i`$ are the creation and annihilation operators for an electron in the localized state at lattice site $`i`$ with binding energy $`E_f`$ and $`d_i^+`$, $`d_i`$ are the creation and annihilation operators of the itinerant spinless electrons in the $`d`$-band Wannier state at site $`i`$.
The first term of (1) is the kinetic energy corresponding to quantum-mechanical hopping of the itinerant $`d`$ electrons between sites $`i`$ and $`j`$. These intersite hopping transitions are described by the matrix elements $`t_{ij}`$, which are $`t`$ if $`i`$ and $`j`$ are the nearest neighbours and zero otherwise (in the following all parameters are measured in units of $`t`$). The second term represents the on-site Coulomb interaction between the $`d`$-band electrons with density $`n_d=N_d/L=\frac{1}{L}_id_i^+d_i`$ and the localized $`f`$ electrons with density $`n_f=N_f/L=\frac{1}{L}_if_i^+f_i`$, where $`L`$ is the number of lattice sites. The third term stands for the localized $`f`$ electrons whose sharp energy level is $`E_f`$.
Since in this spinless version of the FKM without hybridization the $`f`$-electron occupation number $`f_i^+f_i`$ of each site $`i`$ commutes with the Hamiltonian (1), the $`f`$-electron occupation number is a good quantum number, taking only two values: $`w_i=1`$ or 0, according to whether or not the site $`i`$ is occupied by the localized $`f`$ electron.
Now the Hamiltonian (1) can be written as
$$H=\underset{ij}{}h_{ij}d_i^+d_j+E_f\underset{i}{}w_i,$$
(2)
where $`h_{ij}(w)=t_{ij}+Uw_i\delta _{ij}`$.
Thus for a given $`f`$-electron configuration $`w=\{w_1,w_2\mathrm{}w_L\}`$ defined on the three-dimensional lattice with periodic boundary conditions, the Hamiltonian (2) is the second-quantized version of the single-particle Hamiltonian $`h(w)=T+UW`$, so the investigation of the model (2) is reduced to the investigation of the spectrum of $`h`$ for different configurations of $`f`$ electrons.
Despite its relative simplicity and an impressive research activity in the past, the properties of this model remained unclear for a long time. The crucial break in this direction has been done recently by exact analytical and numerical calculations. These calculations showed that the spinless FKM can describe (at least qualitatively) such important phenomena observed experimentally in some rare-earth and transition metal compounds like the discontinuous valence and metal insulator transitions, phase separation, charge ordering, stripes formation, etc. In addition, it was found that at non-zero temperatures the model is able to provide the qualitative explanation for the anomalous large values of the specific heat coefficient and for the extremely large changes of the electrical conductivity found in some intermediate valence compounds (e.g., in $`SmB_6`$). These results indicate that the spinless FKM, in spite of its simplicity, could be a convenient microscopic model for a description of ground-state, thermodynamic and transport properties of real materials. However, real materials are usually three dimensional while the most of above mentioned results have been obtained for the limiting cases of $`D=1,D=2`$ and $`D=\mathrm{}`$. Thus one can ask if these results, or at least some of them hold also in three dimensions. This is the question that we would like to answer in this paper. Here we focus our attention on the ground-state properties of model. The special attention is devoted to examine the three dimensional analogs of phase segregation, charge ordering, stripes formation and metal-insulator transitions observed in $`D=1`$ and $`D=2`$. From this point of view the paper represents the first attempt to describe systematically the ground-state properties of the FKM in three dimensions. To attain this goal we use a well-controlled numerical method that we have elaborated recently . The method is based on the simple modification of the exact diagonalization method on finite clusters and consists of following steps. (i) Chose a trial configuration $`w=\{w_1,w_2\mathrm{}w_L\}`$. (ii) Having $`w`$, $`U`$ and $`E_f`$ fixed, find all eigenvalues $`\lambda _k`$ of $`h(w)=T+UW`$. (iii) For a given $`N_f=_iw_i`$ determine the ground-state energy $`E(w)=_{k=1}^{LN_f}\lambda _k+E_fN_f`$ of a particular $`f`$-electron configuration $`w`$ by filling in the lowest $`N_d=LN_f`$ one-electron levels (here we consider only the case $`N_f+N_d=L`$, which is the point of the special interest for valence and metal-insulator transitions caused by promotion of electrons from localized $`f`$ orbitals $`(f^nf^{n1})`$ to the conduction band states). (iv) Generate a new configuration $`w^{}`$ by moving a randomly chosen electron to a new position which is chosen also as random. (v) Calculate the ground-state energy $`E(w^{})`$. If $`E(w^{})<E(w)`$ the new configuration is accepted, otherwise $`w^{}`$ is rejected. Then the steps (ii)-(v) are repeated until the convergence (for given $`U`$ and $`E_f`$ ) is reached. Of course, one can move instead of one electron (in step (iv)) simultaneously two or more electrons, thereby the convergence of method is improved. Indeed, tests that we have performed for a wide range of the model parameters showed that the latter implementation of the method, in which $`1<p<p_{max}`$ electrons ($`p`$ should be chosen at random) are moved to new positions overcomes better the local minima of the ground state energy. In this paper we perform calculations with $`p_{max}=N_f`$. The main advantage of this implementation is that in any iteration step the system has a chance to lower its energy (even if it is in a local minimum), thereby the problem of local minima is strongly reduced (in principle, the method becomes exact if the number of iteration steps goes to infinity). On the other hand a disadvantage of this selection is that the method converges slower than for $`p_{max}=2`$ and $`p_{max}=3`$. To speed up the convergence of the method (for $`p_{max}=N_f`$) and still to hold its advantage we generate instead the random number $`p`$ (in step (iv)) the pseudo-random number $`p`$ that probability of choosing decreases (according to the power law) with increasing $`p`$. Such a modification improves considerably the convergence of the method. Repeating this procedure for different values of $`E_f`$ and $`U`$ one can immediately study the dependence of the $`f`$-electron occupation number $`N_f=_iw_i^{min}`$ on the $`f`$-level position $`E_f`$ (valence transitions) or the phase diagram of the model in the $`n_fU`$ plane. This method was first used in our recent paper to study the ground-state properties of the one and two-dimensional FKM. It was found that for small and intermediate clusters, where the exact numerical solution is possible ($`L30`$), the method is able to reproduce exactly the ground states of the spinless FKM, even after relative small number of iterations (typically 10000 per site).
## 2 Results and discussion
To examine ground-state properties of the spinless FKM in three dimensions we have performed an exhaustive numerical study of the model for weak ($`U=1`$), intermediate ($`U=2`$) and strong ($`U=8`$) interactions. For each selected value of $`U`$ and $`N_f`$ ($`N_f=0,1,..,L`$) the ground-state configuration $`w^{min}`$ is determined by the above described method (we remember that the total filling is fixed at 1). To reveal the finite-size effects numerical calculations were done on two different clusters of $`4\times 4\times 4`$ and $`6\times 6\times 6`$ sites. A direct comparison of numerical results obtained on $`4\times 4\times 4`$ and $`6\times 6\times 6`$ clusters showed that the ground-state configurations fall into several different categories which stability regions are practically independent of $`L`$. Let us start a discussion of our results with a description of these configuration types for different values of $`U`$ and $`N_f`$ (in the remainder of the paper the values of $`N_f`$ always correspond to $`6\times 6\times 6`$ cluster).
The largest number of configuration types is observed in the weak-coupling limit. Going with $`N_f`$ from zero to half-filling ($`N_f=L/2`$) we have observed the following configuration types for $`U=1`$. At low $`f`$-electron concentrations the ground-states are the phase segregated configurations ($`f`$-electrons clump together while remaining part of lattice is free of $`f`$-electrons) listed in Fig. 1a for two selected values of $`N_f`$. Since the ground-states corresponding to the segregated configurations are metallic we arrive at an important conclusion, and namely, that the metallic domain that exists in the one and two dimensional FKM persists also in three dimensions. In the one dimensional case the region of stability of this metallic domain was restricted to low $`f`$-electron concentrations $`n_f<1/4`$ and small Coulomb interactions $`U1`$ . The numerical calculations performed in two dimensions revealed that with increasing dimension the stability region of this metallic domain shifts to higher values of $`U`$ ($`U3`$). From this point of view it is interesting to examine if this trend holds also for three dimensions. To verify this conjecture we have determined the ground-state configurations for increasing $`U`$ at low $`f`$-electron concentrations on $`4\times 4\times 4`$, $`6\times 6\times 6`$ and $`8\times 8\times 8`$ clusters. We have found that the metallic region in $`D=3`$ extends up to $`U5`$, what confirms the trend conjectured from two dimensional calculations (in addition, in accordance with two-dimensional results we have found that the critical value of $`U_c`$ decreases with increasing $`n_f`$). It should be noted that this result is crucial for description of insulator-metal transitions in real materials (like rare-earth and transition metal compounds). In these materials the values of the interaction constant $`U`$ are much larger than the values of hopping integrals $`t_{ij}`$ , and thus for the correct description of valence and metal-insulator transitions in these compounds one has to take the limit $`U>t`$ and not $`U<t`$. On the other hand it should be mentioned that in the Falicov-Kimball picture it is possible to get the metal-insulator transition much easier, for example by including spins. Indeed, numerical calculations performed for the spin-one-half FKM showed that the metallic domain is stable in this model for a wide range of model parameters, including large values of $`U`$ and $`n_f`$. Above the region of phase segregation we have observed the region of stripes formation ($`N_f=10,..,20`$). In this region the $`f`$-electrons form the one-dimensional charge lines (stripes) that can be perpendicular or parallel (see Fig. 1b). This result shows that the crucial mechanism leading to the stripes formation in strongly correlated systems should be the competition between the kinetic and short-range Coulomb interaction. Going with $`N_f`$ to higher values of $`N_f`$ the stripes vanish and again appear at $`N_f=26`$, however in a fully different distribution (see Fig. 2a). While at smaller values of $`N_f`$ the stripes have been distributed inhomogeneously (only over one half of lattice) the stripes in the region $`N_f=26,..,31`$ are distributed regularly. Above this region a new type of configurations (see Fig. 2b) starts to develop. We call them the diagonal charge planes with incomplete chessboard structure, since the $`f`$-electrons prefer to occupy the diagonal planes with slope 1 and within these planes they form the chessboard structure. Of course, there is a considerable freedom in categorization of ground state configurations according to some common features and the case of diagonal planes used by us is only one of possible ways. The region of diagonal charge planes is relatively broad and extends up to $`N_f50`$. Then follows the region in which the chessboard structure starts to develop. As illustrated in Fig. 3a the $`f`$-electrons begin to occupy preferably the sites of sublattice A, leaving the sublattice B free of $`f`$-electrons. In addition, the configurations that can be considered as mixtures of previous configuration types are also observed in this region (see Fig. 3b). However, with increasing $`N_f`$ the configurations of chessboard type become dominant. Analysing these configurations we have found that the transition to the purely chessboard configuration realizes through several steps. The first step, the formation of the chessboard structure has been illustrated in Fig. 3a. The second step is shown in Fig. 4a. It is seen that the chessboard structure is fully developed in some regions (planes) that are separated by planes with incomplete developed chessboard structure. Such a type of distribution is replaced for larger values of $`N_f`$ by a new type of distributions (step three), where both regions with complete and incomplete chessboard structure have the three-dimensional character (see Fig. 4b).
The same picture we have observed also for intermediate values of Coulomb interactions ($`U=2`$). The larger values of $`U`$ only slightly modify the stability regions of some phases, but no new configuration types appear. In particular, the domain of phase segregation, as well as the domain of stripes formation are reduced while the domain of diagonal planes with chessboard structure increases. This trend is observed also for larger values of $`U`$. In the strong coupling limit ($`U=8`$) the phase segregated and striped phases absent and the region of stability the diagonal planes extends to relatively small values of $`N_f20`$. Below this value a homogeneous distribution of $`f`$-electrons is observed. Thus we can conclude that all fundamental results found in one and two-dimensional solutions of the FKM (the phase segregation, the stripes formation, the phase separation, etc.) holds also in three dimensions, thereby the FKM becomes interesting for a description of ground-state properties (e.g., valence and metal-insulator transitions induced by doping and pressure) of real (three dimensional) systems . The work in this direction is currently in progress.
In summary, the ground-state properties of the three-dimensional FKM were examined by a well-controlled numerical method. The results obtained were used to categorize the ground-state configurations according to common features for weak ($`U=1`$), intermediate ($`U=2`$) and strong interactions ($`U=8`$). It was shown that only a few configuration types form the basic structure of the phase diagram in the $`n_fU`$ plane. In particular, the largest regions of stability correspond to phase segregated configurations, striped configurations and configurations in which electrons are distributed in diagonal planes with incomplete chessboard structure. Near half-filling, mixtures of two phases with complete and incomplete chessboard structure were determined.
This work was supported by the Slovak Grant Agency VEGA under grant No. 2/4060/04 and the Science and Technology Assistance Agency under Grant APVT-20-021602. Numerical results were obtained using computational resources of the Computing Centre of the Slovak Academy of Sciences.
|
warning/0506/cond-mat0506247.html
|
ar5iv
|
text
|
# Chiral phase from three spin interactions in an optical lattice
## I Introduction
The proposal Jaks98 and subsequent realization Raithel ; Mandel1 ; Mandel3 of optical lattices for the manipulation of ultra-cold atoms has attracted a great deal of research towards the implementation of quantum computation Deutsch ; Jaksch ; Ripoll and the simulation of condensed matter systems Kuklov ; Belen ; Kukl ; Jask03 ; Duan . The main advantages of optical lattices are their long decoherence times and high degrees of controllability. Hence, one is able to probe phenomena that manifest themselves at higher orders in perturbation theory such as many body interactions Pachos . This gives the possibility to engineer and control exotic interactions in many body systems and subsequently realize novel phases of matter. Examples include cases where frustration effects are present or competing phenomena coexist giving rise to large degeneracy structures Sachdev ; Nielsen . Towards this direction we consider a semi-one dimensional lattice comprising a triangular ladder. The dynamics of two boson species mounting the optical lattice in the limit of strong collisional interactions can be effectively described by a chain of spin-$`1/2`$ interacting particles. Here we consider systems with two and three spin interactions given by $`ZZ=_i\sigma _i^z\sigma _{i+1}^z`$ and $`ZZZ=_i\sigma _i^z\sigma _{i+1}^z\sigma _{i+2}^z`$ respectively. We shall see explicitly that it is possible to make $`ZZZ`$ dominant by appropriate tuning of the collisional and tunnelling couplings. As simple as these interactions may seem, when combined, they give rise to a rich variety of phase transitions, the study of which is the subject of this article.
In particular, we consider a Hamiltonian that includes the two and three spin interactions with couplings $`\lambda _1`$ and $`\lambda _2`$ respectively, in the presence of a transverse field with unit amplitude. We observe that for $`\lambda _2=0`$ the model is exactly solvable, as it reduces to the Ising interaction with critical points at $`\lambda _1=\pm 1`$. For $`\lambda _1=0`$ we obtain a Hamiltonian that is not analytically diagonalizable. Nevertheless, a numerical study reveals a tricritical point at $`\lambda _2=1`$ that is in the same universality class as the three-state Potts model Wu . When the two spin interaction is dominant the ground state has a spin-order of period 2 while the dominance of the higher order interaction brings about a spin-order of period 3. Though these phases are gapped, there are values of relative couplings where these orders compete, giving rise to a high degree of degeneracy. This characteristic allows for the presence of a gapless incommensurate chiral phase that extends to a wide range of parameters. In Figure 1 the fidelity of the actual ground state of the Hamiltonian is plotted against the ground state of each individual term comprising it. While we see that these agree for a large range of couplings there is an area between spin-order 2 and 3 where these states fail to accurately describe the nature of the system. We shall identify this as the gapless incommensurate chiral phase. A similar region has been presented by Fendley, Sengupta and Sachdev Sachdev in a 1-dimensional hard boson system.
The article is organized as follows. In Section II we present the realization of the two and three spin interactions by ultra-cold atoms superposed by optical lattices. Section III is a study of the tricritical point present at $`\lambda _1=0,\lambda _2=1`$. In Section IV, we present the properties of the full Hamiltonian. For that we employ perturbation theory, a bosonization of the model and a study of an effective field theory that eventually reveal the incommensurate chiral phase. Finally, in Section V, we conclude and present future directions.
## II Optical Lattices And Three Spin Interactions
Consider the physical setup where an ultra-cold atomic cloud of two different species, $`a`$ and $`b`$, is superposed with a three dimensional optical lattice. The atoms can tunnel through the potential barriers of the lattice from one site to the next with a coupling $`J`$. When two or more atoms are present in the same site, they collide with coupling $`U`$. For sufficiently large intensities of the laser radiation, where $`JU`$, the system is in the Mott insulator regime with a regular number of atoms per site. In particular, we can arrange the density of the atomic cloud to be low enough so that only one atom can exist at each site of the lattice, i.e. $`n_a+n_b1`$. In this way each lattice site is a simple two state system, that can be viewed as a spin-1/2 particle. Within this representation, the Hamiltonian can be written in terms of Pauli spin operators.
Indeed, for the triangular ladder seen in Figure 2 we obtain the Hamiltonian of the form Pachos ; Pachos1
$$H=B\sigma _i^x+\lambda _1\sigma _i^z\sigma _{i+1}^z+\lambda _2\sigma _i^z\sigma _{i+1}^z\sigma _{i+2}^z,$$
(1)
where $`B`$, $`\lambda _1`$ and $`\lambda _2`$ are all functions of the initial tunnelling and collisional couplings, $`J`$ and $`U`$. The values of $`\lambda _1`$ and $`\lambda _2`$ can be controlled independently by varying $`J`$ and $`U`$. In particular, it is possible to make $`\lambda _2`$ large compared to $`\lambda _1`$ as can be seen in Figure 3. A careful consideration of the system would reveal that terms of the form $`\sigma _i^z\sigma _{i+2}^z`$ also appear, due to the triangular configuration. This can be remedied by employing superlattices that do not alter the other terms.
## III The $`ZZZ`$ Interaction
Ultimately, we would like to study the entire $`(\lambda _1,\lambda _2)`$ plane, identify the regions of criticality behavior and understand the global properties of the ground state. Before turning to the general problem let us consider the two special limiting cases. Initially, when $`\lambda _2=0`$, the Hamiltonian reduces to the Ising interaction between neighboring spins in the presence of a transverse magnetic field. This well-studied model exhibits criticality behavior for $`\lambda _1=\pm 1`$.
Another interesting model Alcaraz can be obtained for $`\lambda _1=0`$. For simplicity we can take $`\lambda _2=1`$ and consider the Hamiltonian to be a function of the transverse field with amplitude $`B`$ Penson ; Igloi . As we vary $`B`$ we observe that there are two distinctive regions. For $`B1`$, the ground state has all the spins oriented towards the $`x`$ direction, and for $`B1`$ there is a degeneracy between the states $`|`$, its translations (the $`Z_3`$ symmetry) and $`|`$. It is of interest to study the behavior of the system between these two limiting cases.
Due to the symmetry of the Hamiltonian we can pinpoint where a possible critical point can lie. For that, one can define, $`\mu _i^x\sigma _i^z\sigma _{i+1}^z\sigma _{i+2}^z`$ and $`\mu _i^z_{n=0}^{\mathrm{}}\sigma _{i3n}^x\sigma _{i3n1}^x`$. It is easily verified Turban that these operators obey the Pauli algebra commutation relations. Moreover, under this transformation the Hamiltonian becomes $`BH(B^1)`$ which has exactly the same spectrum as $`H(B)`$. Hence, if there exists a single critical point it has to be at $`B=1`$.
We can verify this numerically and identify the critical exponents, by observing the minimum in the energy gap between the ground and first excited state on a finite chain of spins with periodic boundary conditions. In the thermodynamical limit this minimum, if it corresponds to a critical point, will become zero. Near this region the energy gap, $`\mathrm{\Delta }`$, is expected to scale as follows Goldenfield ,
$$\mathrm{\Delta }=N^z\mathrm{\Phi }\left(N^{1/\nu }(BB_c)\right).$$
(2)
Here $`\mathrm{\Phi }`$ is a universal scaling function, $`z`$ is the dynamic critical exponent, and $`\nu `$ is the correlation length exponent. Figure 4(a) gives a plot of $`N^{1.00263}\mathrm{\Delta }`$ versus $`B`$, which shows that the energy minimum progresses towards $`B=1`$ for increasing $`N`$. We may employ Eqn. (2) to estimate $`z`$ and $`\nu `$. Figure 4(b) shows $`N^{1.0002623}\mathrm{\Delta }`$ against $`N^{1/0.757868}(B_xB_x^c)`$, where the data for systems of different sizes come together into a single curve. From Eqn. (2) we deduce the value of the critical exponents to be $`z1`$ and $`\nu 0.76`$.
We can also determine numerically the central charge of the critical theory and see if it corresponds to the same universality class. The central charge can be obtained by studying the scaling behavior of the entropy of entanglement Latorre . The latter is given by the von Neumann entropy, $`S_L`$, of the reduced density matrix, $`\rho _L`$. It can be calculated from the density matrix of the original system, being in a pure state, where all but $`L`$ contiguous spins are traced out. This indicates quantitatively the degree of entanglement of the $`L`$ spins with the rest of the chain. Indeed, we have
$$\begin{array}{ccc}\hfill \rho _L& & \text{tr}_{NL}|\psi \psi |,\hfill \\ \hfill S_L& & \text{tr}(\rho _L\mathrm{log}\rho _L),\hfill \end{array}$$
(3)
where $`|\psi `$ is the ground state of the system. For a critical configuration and for large L we expect $`𝒮_L\frac{c+\overline{c}}{6}\mathrm{log}L`$, where $`c`$ is the central charge of the corresponding conformal field theory and $`\overline{c}`$ is its complex conjugate. The central charge uniquely corresponds to the critical exponents of the energy and of the correlation length, $`z`$ and $`\nu `$, respectively. We know that for non-critical chains $`S_L`$ should be saturated for large enough values of $`L`$. This behavior is observed from the simulations when $`\lambda _21`$. On the other hand, when we are at $`\lambda _2=B=1`$ we obtain Figure 5 that shows the expected logarithmic progression. By a logarithmic fitting we can deduce that $`c4/5`$. Hence, our model is in the same universality class as the three-state Potts model and it corresponds to the critical exponents $`z=1`$ and $`\nu =3/4`$ in agreement with the earlier findings.
## IV The Full Hamiltonian
We can now turn to the full Hamiltonian for arbitrary values of the coupling parameters. For convenience we set $`B=1`$ and vary only the interaction couplings $`\lambda _1`$ and $`\lambda _2`$. Without loss of generality we can restrict on the $`\lambda _2>0`$ half-plane as the case $`\lambda _2<0`$ is automatically obtained by exchanging $`\mathbf{}\mathbf{}\mathbf{}\mathbf{}`$.
For large values of $`\lambda _1`$ or $`\lambda _2`$ it is possible to neglect the transverse field $`X=_i\sigma _i^x`$ and treat the system classically. We find two asymptotes along which the ground state should undergo a first order phase transition. By introducing $`X`$ as a small perturbation one can predict how these curves behave when we approach the origin $`\lambda _1=\lambda _2=0`$. We find that along the asymptote $`\lambda _1/\lambda _2=3/2`$ the classical ground state is highly degenerate: it is in fact infinitely degenerate in the thermodynamic limit. In the quantum regime, this gives rise to an intermediate region between the two phases with spin-order 2 and 3, the size and nature of which will be studied in this section.
### IV.1 Perturbation Theory
To understand the behavior of Hamiltonian (1) we shall employ perturbation theory. Surprisingly, the second order perturbation will give a very good approximation to the numerical findings shown in Figure 1.
#### IV.1.1 Classical Regime
As a first step we take $`\lambda _1,\lambda _21`$, thus neglecting the $`X`$ term the Hamiltonian reduces to
$$H\lambda _1\underset{i}{}\sigma _i^z\sigma _{i+1}^z+\lambda _2\underset{i}{}\sigma _i^z\sigma _{i+1}^z\sigma _{i+2}^z.$$
(4)
This Hamiltonian has a classical behavior, so it can be solved exactly by examining eigenstates of $`\sigma _z^N`$. There are three regions on the $`(\lambda _1,\lambda _2)`$ plane of distinct ground state behavior. Consider first $`\lambda _1<0`$, $`\lambda _2>0`$. The energy of the system is minimized uniquely by the state
$$|A=|\mathrm{}.$$
(5)
From the expectation values, $`A|_i\sigma _i^z\sigma _{i+1}^z|A/N=1`$ and $`A|_i\sigma _i^z\sigma _{i+1}^z\sigma _{i+2}^z|A/N=1`$, one can calculate the energy per site to be $`E_A=\lambda _1\lambda _2`$. Similarly we take the cases of $`\lambda _1>0`$, $`\lambda _2\lambda _1`$ and $`\lambda _1>0`$, $`\lambda _2\lambda _1`$, resulting in the following ground states
$$|B=|\mathrm{},$$
(6)
with $`E_B=\frac{1}{3}\lambda _1\lambda _2`$ and
$$|C=|\mathrm{},$$
(7)
with $`E_C=\lambda _1`$. Observe that these states are $`3`$ and $`2`$ periodic respectively giving rise to $`Z_3`$ and $`Z_2`$ symmetries. To evaluate the position of the phase transitions we simply equate energies in neighboring regions. In this way we obtain the asymptotes $`\lambda _1=0`$ and $`\lambda _1=\frac{3}{2}\lambda _2`$ that separate the states of the system $`|A`$, $`|B`$ and $`|C`$ as seen in Figure 6.
Another classical region is obtained at $`\lambda _1,\lambda _21`$, where the transverse field, $`X`$, dominates. In this region the ground state is given by
$$|D=|^N,$$
(8)
with $`|(||)/\sqrt{2}`$ and an energy per site of $`E_D=1`$.
#### IV.1.2 Quantum Regime
We now consider the transverse field as a perturbation to the theory, so we rewrite our Hamiltonian as
$$H=\lambda _2\underset{i}{}(\lambda \sigma _i^x+\mu \sigma _i^z\sigma _{i+1}^z+\sigma _i^z\sigma _{i+1}^z\sigma _{i+2}^z),$$
(9)
where $`\lambda 1/\lambda _2`$ and $`\mu \lambda _1/\lambda _2`$. Cyclic permutations of states $`|B`$ and $`|C`$ are also ground states of the classical theory due to the translational invariance of the Hamiltonian. Nevertheless, we need not employ degenerate perturbation theory at this point since the permutations behave identically under the transverse field, $`X`$. For this Hamiltonian the energies per site are
$$\stackrel{~}{E}_A=\lambda _1\lambda _2\frac{1}{\lambda _2}\frac{1}{64\mu },$$
(10)
$$\stackrel{~}{E}_B=\left(\lambda _1+\frac{\lambda _2}{3}\right)\lambda _2\left[\left(\frac{1}{9}\right)+\left(\frac{1}{3}\right)\frac{1}{6+4\mu }\right].$$
(11)
By inspection we see that the two energies are equal for $`\mu =0`$, which coincides exactly with the classical solution. Hence, at $`𝒪(\lambda ^2)`$, the boundary between $`|A`$ and $`|B`$ remains unchanged, as seen in Figure 6.
Let us turn to the perturbation treatment of the boundary between the $`|D`$ and the $`|A`$ states. In fact, to the second order in the couplings, $`\lambda _1`$ and $`\lambda _2`$, we find
$$\stackrel{~}{E}_D=1\frac{\lambda _1^2}{4}\frac{\lambda _2^2}{6}.$$
(12)
Equating $`\stackrel{~}{E}_D=\stackrel{~}{E}_A`$ we obtain the corresponding boundary to be a correction of the classical one given by $`\lambda _1=\lambda _21`$. The computational solution of $`\stackrel{~}{E}_D=\stackrel{~}{E}_A`$ is given in Figure 6. It provides a line of second order phase transition joining the Potts critical point and the Ising critical point that fits firmly with the fidelity plot in Figure 1.
#### IV.1.3 Competing Ground States
We saw above that for $`\mu =\frac{3}{2}`$ the period 2 and period 3 phases have the same ground state energy. Interestingly, on this asymptote the two spin-orders can mix, giving rise in the thermodynamic limit to an infinitely degenerate ground state. A perturbation theory around this area is non-trivial as care has to be taken in the way the degeneracy is lifted. Using a bosonization procedure similar to that presented in Sachdev , we will calculate the spectrum of the theory up to second order in the transverse field. We will find a finite intermediate phase separating the period 2 and period 3 gapped phases.
Let us present in detail the bosonization procedure necessary to develop the perturbation theory. Consider two new types of bosons – the $`|3`$, created by the $`t^{}`$ operator, and $`|2`$, created by $`p^{}`$. In the spin formalism, $`|3`$ corresponds to $`|`$, while $`|2`$ is equivalent to $`|`$. Thus,
$$\begin{array}{ccc}|B=|33\mathrm{}3\hfill & \text{and}& \hfill |C=|22\mathrm{}2.\end{array}$$
(13)
All states that are composed of $`2`$s and $`3`$s are degenerate along the line $`\mu =\frac{3}{2}`$. Therefore, the region of the phase diagram that we want to study is given by
$$0<|\mu 3/2||\lambda _1|,|\lambda _2|.$$
(14)
Since there are no first order contributions in the perturbation theory, the region we want to study can be parameterized by
$$\mu =\frac{3}{2}+\sigma \lambda ^2,$$
(15)
for a dimensionless parameter $`\sigma `$.
The approach is to create a theory with $`t`$ and $`p`$ bosons that is equivalent to the present one. Let us first work with $`t`$ bosons. The vacuum state has the form $`|22\mathrm{}2`$, with no $`3`$s. To construct the Hamiltonian we need the self energy of a $`t`$ boson, the interaction energy and the hopping amplitude. For that, we calculate the energies per site needed to add $`22`$, $`33`$ and $`23`$ bonds, given by $`E_{22}`$, $`E_{33}`$ and $`E_{23}`$, respectively. In particular,
$$|\underset{m1}{\underset{}{22\mathrm{}2}}|\underset{m}{\underset{}{22\mathrm{}22}}$$
(16)
gives rise to the energy gap $`E_{22}`$, where $`m`$ is an integer. A similar transformation gives $`E_{33}`$. Furthermore,
$$|\underset{2(m1)}{\underset{}{23\mathrm{}23}}|\underset{(2m)}{\underset{}{23\mathrm{}2323}}$$
(17)
induces the energy gap $`2E_{23}`$. The energy of each of these states can be calculated perturbatively. Up to order $`𝒪(\lambda ^2)`$, we obtain
$$\begin{array}{ccc}\hfill E_{22}& =& 2\mu +\frac{\lambda ^2}{2(12\mu )},\hfill \\ \hfill E_{33}& =& (3+\mu )\frac{\lambda ^2}{3},\hfill \\ \hfill E_{23}& =& \frac{3}{2}(1+\mu )\frac{\lambda ^2}{6}.\hfill \end{array}$$
(18)
Consider now the energy, $`E_t`$, of creating a single $`t`$ boson in a background of $`2`$s, and the interaction energy, $`E_{t,int}`$, generated when two $`t`$ bosons are brought to adjacent sites. The former can be evaluated by considering the energy gap of the transformation
$$|\underset{5m}{\underset{}{22\mathrm{}222}}|\underset{4m}{\underset{}{23\mathrm{}23}},$$
(19)
which creates $`2m`$ bosons. The states are chosen so that the lengths of the underlying spin chains remain the same. For the interaction energy, the relevant energy gap arises from
$$|\underset{4m}{\underset{}{23\mathrm{}23}}|\underset{2m}{\underset{}{2\mathrm{}2}}\underset{2m}{\underset{}{3\mathrm{}3}}.$$
(20)
Hence these energies per boson are given by
$$\begin{array}{ccc}E_t\hfill & =& 2E_{23}\frac{5}{2}E_{22}=\lambda ^2\left(\frac{7}{24}+2\sigma \right),\hfill \\ E_{t,int}\hfill & =& 2E_{23}E_{22}E_{33}=\frac{\lambda }{4}^2.\hfill \end{array}$$
(21)
Finally, we calculate the hopping amplitude, which is given by the activation energy needed to perform the exchange
$$|23|32.$$
This has a value of $`\frac{\lambda ^2}{6}`$. We are now in a position to rewrite the Hamiltonian, which takes the following form
$$\begin{array}{ccc}\hfill H& =& \frac{N}{2}\left(2\mu +\frac{\lambda ^2}{2(12\mu )}\right)\hfill \\ & & \lambda ^2\underset{l}{}[\frac{1}{6}(t_{l+2}^{}t_l+t_l^{}t_{l+2})\hfill \\ & & t_l^{}t_l(\frac{7}{24}+2\sigma )+\frac{1}{4}t_{l+3}^{}t_l^{}t_{l+3}t_l].\hfill \end{array}$$
(22)
The label $`l`$ is over the original indices, with a $`t`$ boson being centered in the middle of the $`3`$ deformation, and the integer $`N`$ is the total number of sites in the original spin formalism. We consider $`\sigma 1`$, which corresponds to the spin-order 2 phase. The ground state is the vacuum state $`|22\mathrm{}2`$, which is in agreement with original analysis. By employing the Fourier transform of $`t_l`$ we obtain that the lowest excited energy above the vacuum state is $`\lambda ^2\left(2\sigma \frac{1}{24}\right)`$. Thus at $`\sigma =\frac{1}{48}`$ a phase transition occurs from period 2 to the intermediate region.
Following the same procedure we can find the boundary of the period 3 gapped phase. Now the vacuum state is $`|33\mathrm{}3`$ and the relevant energies are with respect to the $`p`$ boson. We have
$$\begin{array}{ccc}E_p\hfill & =& 2E_{23}\frac{5}{3}E_{33}=\lambda ^2\left(\frac{4}{3}\sigma +\frac{2}{9}\right),\hfill \\ E_{p,int}\hfill & =& 2E_{23}E_{22}E_{33}=\frac{\lambda }{4}^2,\hfill \end{array}$$
(23)
while the hopping amplitude is the same. In terms of these variables the Hamiltonian takes the form
$$\begin{array}{ccc}\hfill H& =& \frac{N}{3}\left(3+\mu +\frac{\lambda }{3}^2\right)\hfill \\ & & \lambda ^2\underset{l}{}[\frac{1}{6}(p_{l+3}^{}p_l+p_l^{}p_{l+3})\hfill \\ & & +p_l^{}p_l(\frac{4}{3}\sigma \frac{2}{9})+\frac{1}{4}p_{l+2}^{}p_l^{}p_{l+2}p_l].\hfill \end{array}$$
(24)
An identical analysis is now necessary, by simply considering $`\sigma 1`$. The $`p`$ bosons cost large positive energy, making the ground state the vacuum, as one would expect. The first excited state has an energy of $`\lambda ^2\left(\frac{4}{3}\sigma +\frac{7}{9}\right)`$, so the boundary is at $`\sigma =\frac{7}{12}`$.
We therefore have a prediction of the width of the intermediate phase that is illustrated in Figure 6. This result is in perfect agreement with the fidelity plot in Figure 1.
### IV.2 Criticality Behavior
At this point we can discuss the criticality behavior of our system. Apart from the Ising critical points, P<sub>1</sub> and P<sub>2</sub>, and the three-state Potts model critical point, P<sub>3</sub>, we can identify the critical behavior of the lines, L<sub>1</sub>, L<sub>2</sub>, L<sub>3</sub> and L<sub>4</sub>, seen in Figure 6, where our results of the perturbation treatment are summarized.
Specifically, we obtain first order critical behavior for $`\lambda _1=0`$ and $`\lambda _2>1`$ and second order behaviors for the rest of the curves. The perturbation theory shows that the line, L<sub>3</sub>, acquires no second order contribution indicating its first order nature. This is also supported by a numerical study of the energy gap for a chain of 18 spins. Indeed, we find that its minima lie along lines similar to the ones in Figure 6. Furthermore, there is an actual energy crossing along the first order line while all the other critical points remain gapped.
Studying the critical exponents along the line L<sub>4</sub> we find that in moving from P<sub>3</sub> to P<sub>2</sub> along the line of criticality, the exponent $`z`$ decreases to zero and becomes discontinuous at the point P<sub>2</sub>. Moving in the other direction along L<sub>2</sub> yields ever increasing values of $`z`$. Hence, in view of Eqn. (2), we confirm that as the transverse field becomes insignificant, the phase transitions return to first order behavior with an actual energy level crossing. Along the upper asymptote, L<sub>1</sub>, we observe similar behavior. The critical exponent is large in configurations far away from the origin and decreases continuously down to $`z=1/2`$ at the Ising critical point P<sub>1</sub>.
These results present a rich variety of phase transition behavior. From Figures 1 and 6 we see that the second order perturbation theory gives a good picture of the ground state of the full Hamiltonian in most of the parametric region. The only part of the $`(\lambda _1,\lambda _2)`$ plane that is not yet understood is the strip bounded by the lines, $`L_1`$ and $`L_2`$. Probing the physics of this region is the subject of the next subsection.
### IV.3 Incommensurate Phase
It becomes apparent from the previous that perturbation theory cannot give reliable information about the physics between $`L_1`$ and $`L_2`$. To study that region we resort to the Landau-Ginzburg approximation QPT . As we shall see, the resulting theory corresponds to the chiral clock model Ostlund , which predicts non-zero chirality for our model. This also indicates the presence of a gapless incommensurate phase between $`L_1`$ and $`L_2`$. For a finite chain we find strong evidence for the presence of this phase using exact numerical diagonalization.
To gain a further insight into the nature of the ground states of our model we introduce the order parameter Sachdev
$$\mathrm{\Psi }_p=\underset{j}{}e^{2\pi ij/p}|_j|,$$
(25)
where the operator $`|_j|`$ is the projector onto the down state of the $`j^{\text{th}}`$ site. The periodicity of a state is revealed by the expectation value of $`\mathrm{\Psi }_p`$: it is maximal for $`p`$-periodic states, and minimal (or zero) otherwise. In particular, the wave order parameter, $`\mathrm{\Psi }_3`$, is a complex operator that detects the period 3 states. We wish to use $`\mathrm{\Psi }_3`$ as an order parameter to probe the behavior of our model between the boundaries $`L_1`$ and $`L_2`$. With this in mind we construct a continuum quantum field theory with an action that has the same symmetries as $`\mathrm{\Psi }_3`$ does in the $`Z_3`$ region.
It is easy to verify that $`\mathrm{\Psi }_3`$ is invariant under translations $`\mathrm{\Psi }_3e^{2\pi il/3}\mathrm{\Psi }_3`$, for integer $`l`$. A more intriguing symmetry is given when simultaneous spatial reflections and complex conjugation are performed, $`xx,\mathrm{\Psi }_3\mathrm{\Psi }_3^{}`$ Huse . Finally, $`\mathrm{\Psi }_3`$ is invariant under time reversal, $`\tau \tau `$. According to these symmetries, we can write down the corresponding effective field theory, given by,
$$\begin{array}{ccc}\hfill 𝒮_3& =& dxd\tau [i\alpha \mathrm{\Psi }_3^{}_x\mathrm{\Psi }_3+\text{c.c.}+|_\tau \mathrm{\Psi }_3|^2\hfill \\ & +& v^2|_x\mathrm{\Psi }_3|^2+r|\mathrm{\Psi }_3|^2+v\mathrm{\Psi }_3^3+\text{c.c.}+\mathrm{}].\hfill \end{array}$$
(26)
The parameters $`\alpha ,v`$ and $`r`$ are non-universal functions of the original $`\lambda _1`$ and $`\lambda _2`$. It has been shown Baxter that this model has a critical point at $`\alpha =r=0`$ that belongs in the same universality class as the point P<sub>3</sub>. We may therefore identify them.
The term $`\alpha \mathrm{\Psi }_3^{}_x\mathrm{\Psi }_3`$ breaks chiral symmetry when $`\mathrm{\Psi }_3`$ is complex. The resulting theory has been studied by Ostlund Ostlund and it has been shown that it belongs in the same universality class as the chiral clock model. Hence, it suggests the presence of a chiral phase for non-zero values of $`\alpha `$. Furthermore this model predicts that this phase is gapless and incommensurate Huse ; Cardy . By analogy we conjecture that our model has such a phase away from $`P_3`$ and we shall investigate it numerically.
The gapless and incommensurate characteristics of this phase indicate that $`\mathrm{\Psi }_p`$ can build a non-zero value by applying the appropriate perturbation. Hence, we add to the Hamiltonian a small periodic potential of the form
$$V=10^4\underset{j}{}e^{2\pi ij/p}|_j|$$
(27)
and we plot $`\mathrm{\Psi }_p`$ for several $`p`$ as a function of $`\lambda _1`$ and $`\lambda _2`$. Indeed, for $`p<3`$, we find that $`\mathrm{\Psi }_p0`$ between $`L_1`$ and $`L_2`$ and deduce that $`\alpha 0`$ there (see Figure 7). A naive analysis of our field theory in which we treat $`\mathrm{\Psi }_p`$ as a fluctuation of the uniform solution indicates that $`\alpha `$ may be proportional to $`2\pi /p2\pi /3`$. This suggests a region of $`\alpha 0`$, with opposite sign corresponding to $`p>3`$, in which a second chiral phase exists on the other side of $`P_3`$. We discover numerical evidence for this in Figure 8 when $`\lambda _1<0`$ and $`\lambda _2<1`$, but its effect is of significantly lower order than in Figure 7.
Along with the gapless incommensurate properties of this phase we would like to investigate its chiral nature. Note that, on the one hand, the chirality operator, given by $`\chi _i\stackrel{}{\sigma }_i\stackrel{}{\sigma }_{i+1}\times \stackrel{}{\sigma }_{i+2}`$, is an imaginary hermitian operator. On the other hand the Hamiltonian (1) is real. Thus, if the Hamiltonian possesses a ground state with non-zero expectation value $`\chi `$, that ground state should be multidegenerate. If this is the case in the thermodynamic limit we would like to see if it is also approximated in the finite case. By numerical diagonalization of the finite-size Hamiltonian, we can obtain the lowest energy gap, $`\mathrm{\Delta }E`$, for values of the couplings $`\lambda _1`$ and $`\lambda _2`$ between the $`L_1`$ and $`L_2`$ lines. In Figure 9 we can clearly observe an exponential damping of $`\mathrm{\Delta }E`$ as a function of the length of the spin chain. This allows us to assume that in the thermodynamic limit the spectrum will eventually become degenerate.
## V Conclusion
The present spin model approximates the behavior of the Mott insulator of two species of atoms when a triangular ladder geometry is realized. The spin interpretation, which makes the study of the model simpler and more intuitive, is valid when there is only one atom per lattice site, i.e. at the regime of strong collisional couplings. With this condition satisfied the criticality behavior of the spin model straightforwardly represents the behavior of the Mott insulator. We can argue that non-zero chirality in the spin model corresponds to a ground state with persistent currents Wang . This suggests a counterflow of the two different atomic species that could be measured by using atomic spatial correlations Grondal .
To summarize, in this article, we have generalized the one-dimensional Ising model, $`ZZ`$, to include an additional triple $`\sigma ^z`$ interaction, $`ZZZ`$, in the presence of a transverse magnetic field. A rich criticality behavior has been revealed when the coupling of $`ZZ`$ and $`ZZZ`$ are varied. The phases of the ground state are identified according to their periodic structure. In particular, a complex order parameter related to the states with period three has revealed a chiral, gapped, incommensurate phase. In Sachdev a thorough study of this phase in a one-dimensional hard boson model exhibiting similar behavior was performed. An in depth study of the incommensurate phase will be the subject of a future publication.
###### Acknowledgements.
This work was supported by the Royal Society.
|
warning/0506/hep-th0506031.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Hawking’s semiclassical analysis for the black hole radiation suggests that most of information in initial states is shield behind the event horizon and is never back to the asymptotic region far from the evaporating black hole. This means that the unitarity is violated by an evaporating black hole. However, this conclusion has been debated for decades. The information loss paradox is closely related to the question of whether the formation and subsequent evaporation of a black hole is unitary. One of the most urgent problems in the black hole physics is to resolve the unitarity issue.
Recently, Maldacena proposed that the unitarity can be restored if one takes into account the topological diversity of gravitational instantons with the same AdS boundary in (1+2)-dimensional gravity. Actually, (1+2)-dimensional gravity is not directly related to the information loss problem because there is no physical degrees of freedom. If this gravity could be considered to be part of string theory, the AdS/CFT correspondence requires that the black hole formation and evaporating process be unitary because its boundary can be described by a unitary CFT. On later, Hawking has withdrawn his argument on the information loss and suggested that the unitarity can be restored by extending Maldacena’s proposal to (1+3)-dimensional gravity system. The topological diversity is credited with the restoration of the S-matrix unitarity in the formation and evaporation of a black hole. In this approach of the Euclidean path integral, the pure AdS space plays an important role in restoring unitarity. However, the proposal which is to resolve the information loss paradox by summing over bulk topologies seems to be failed even in the (1+2)-dimensional model.
Since the (1+2)-dimensional gravity and its boundary CFT can provide a prototype to compute quasinormal modes and boundary correlators exactly, they play a crucial role in investigating the unitarity issue. Solodukhin has tried to find an alternative view to resolving the unitarity problem by introducing a non-classical deformation of the BTZ black hole which resembles the geometry of wormhole. As a result of disappearing the event horizon, real and discrete modes are found, which means that the considering system turns out to be unitary.
In this work we focus on the study of the (1+2)-dimensional wave equation for a scalar field which include a massless scalar as well as a dilaton derived from string theory. The pure AdS spacetime provides a unitary evolution, but it is topologically trivial. It is important to find a topologically non-trivial spacetime which provides a unitary evolution. This work may put a further step to understand the unitarity in a semiclassical way because we propose the two extremal BTZ black holes (MBTZ, EBTZ) for the unitarity systems. Our method consists of two steps: the potential analysis using the Schrödinger-equation and obtaining its eigenvalue $`E=\omega ^2`$ by solving the wave equation exactly. Actually, we translate the black hole-unitary problem into the boundary-value problem in the Schrödinger-like equation. If the Schrödinger operator $`=d^2/dr_{}^2+V(r_{})`$ is self-adjoint ($`^{}=`$), its eigenvalue is real upon imposing appropriate boundary conditions. In this case there is no information loss and the unitarity is preserved. If one finds quasinormal modes, the information is lost during evolution and thus the system is not unitaryWe just study the wave propagation to test how an object (the black hole) responds to an external perturbation. In the case of black hole physics, it is impossible to investigate the interior region of the event horizon using the Klein-Gordon equation for a scalar. Hence the radial part of the Klein-Gordon equation leads to the Schrödinger-like equation but not the exact Schrödinger equation because the Klein-Gordon equation belongs to the relativistic wave equation. The use of quantum mechanical terminology is here an analogy to understand the external perturbation intuitively. The system under consideration is not an exact quantum system. Therefore we don’t need to do a further work, the self-adjoint extension of the Schrödinger operator, even if the quasinormal mode is found.. In asymptotically AdS spacetime, quasinormal modes are defined as the solutions which are purely ingoing wave at the event horizon and those which vanish at infinity because the potential is growing at infinity. The last condition means that any leakage of the energy (information) is not allowed through the boundary at infinity. More precisely, we use the flux boundary condition: the ingoing flux at the horizon and the vanishing flux at infinity.
In a (1+2)-dimensional spacetime, the Einstein equation with a negative cosmological constant $`\mathrm{\Lambda }=1/\mathrm{}^2`$ and $`8G_3=1`$ provides the BTZ solution:
$$ds_{BTZ}^2=\left(M+\frac{r^2}{l^2}+\frac{J^2}{4r^2}\right)dt^2+\left(M+\frac{r^2}{l^2}+\frac{J^2}{4r^2}\right)^1dr^2+r^2\left(d\varphi \frac{J}{2r^2}dt\right)^2,$$
(1)
where $`M`$ and $`J`$ turn out to be the mass and angular momentum, respectively. The above metric allows the two horizons of $`r_\pm ^2=Ml^2(1\pm \mathrm{\Delta })/2`$ with $`\mathrm{\Delta }=(1J^2/M^2l^2)^{1/2}`$. Here the conditions of $`|J|Ml`$ and $`M0`$ are required to have the black hole spacetime. Its thermodynamic quantities of energy, Hawking temperature, Bekenstein-Hawking entropy, and heat capacity are given by $`E=M,T_H=M\mathrm{\Delta }/2\pi r_+,S_{BH}=4\pi r_+,C_J=4\pi r_+\mathrm{\Delta }/(2\mathrm{\Delta })`$ . We note that the heat capacity of BTZ black hole is always positive. Hence the BTZ black hole can be thermally in equilibrium with any size of heat reservoir. This explains thermodynamically why the BTZ black hole belongs to an eternal black hole. In this work we consider four interesting cases. i) The non-rotating BTZ black hole (NBTZ) with $`J=0`$: $`r_+^2=l^2M,T_H=\frac{r_+}{2\pi l^2},C_J=4\pi r_+=S_{BH}`$. ii) The pure AdS spacetime (PADS) with $`M=1,J=0`$: $`T_H=C_J=S_{BH}=0`$. This case corresponds to the spacetime picture of the NS-NS vacuum state . iii) The massless BTZ black hole (MBTZ) with $`M=J=0`$: $`T_H=C_J=S_{BH}=0`$. This is called the spacetime picture of the R-R vacuum state. iv) The extremal BTZ black hole (EBTZ) with $`|J|=lM`$: $`r_+^2=r_{}^2=l^2M/2,T_H=C_J=0,S_{BH}=4\pi r_+`$.
## 2 NBTZ VS PADS
We start with the wave propagation for a massive scalar field with mass $`m`$
$$(^2m^2)\mathrm{\Phi }=0$$
(2)
in the background of the non-rotating BTZ black hole. Its line element is given by $`ds_{NBTZ}^2=(M+r^2/l^2)dt^2+(M+r^2/l^2)^1dr^2+r^2d\varphi ^2`$. On the other hand, the pure AdS spacetime is defined by $`ds_{PADS}^2=(1+r^2/l^2)dt^2+(1+r^2/l^2)^1dr^2+r^2d\varphi ^2`$. Here we set $`M=l=1`$ for simplicity, unless otherwise stated. Assuming a mode solution<sup>§</sup><sup>§</sup>§There is no a globally defined time-like Killing vector in the AdS black hole spacetime. But a time-like Killing vector of $`/t`$ is future directed at the region I, the Fig. 3 in Ref.. This means that an appropriate time evolution is allowed if $`_t\mathrm{\Phi }=i\omega \mathrm{\Phi }`$ with $`\omega >0`$ in this work.
$$\mathrm{\Phi }(r,t,\varphi )=f(r)e^{i\omega t}e^{i\mathrm{}\varphi },\mathrm{}𝐙$$
(3)
we find the radial equation
$$(r^21)f^{\prime \prime }(r)+\left(3r\frac{1}{r}\right)f^{}(r)+\left[\frac{\omega ^2}{r^21}\frac{\mathrm{}^2}{r^2}m^2\right]f(r)=0,$$
(4)
where the prime () denotes the derivative with respect to its argument. The upper (lower) signs denote the NBTZ (PADS) cases. Introducing $`f(r)=\stackrel{~}{f}(r)/\sqrt{r}`$, the above equation reduces to
$$(r^21)\stackrel{~}{f}^{\prime \prime }(r)+2r\stackrel{~}{f}^{}(r)+\left[\frac{\omega ^2}{r^21}\frac{3}{4}\frac{1}{4r^2}\frac{\mathrm{}^2}{r^2}m^2\right]\stackrel{~}{f}(r)=0$$
(5)
which is suitable for the potential analysis.
First of all, it is important to see how a scalar wave propagates in the exterior of the NBTZ. For this purpose, we introduce a tortoise coordinate $`2r_{}=\mathrm{ln}[(r1)/(r+1)](r=\mathrm{coth}[r_{}])`$. We have $`r_{}\mathrm{}(rr_+)`$ and $`r_{}0(r\mathrm{})`$. We transform Eq.(5) into the Schrödinger-like equation with the Schrödinger operator $`_{NBTZ}`$ and energy $`E=\omega ^2`$
$$\frac{d^2}{dr_{}^2}\stackrel{~}{f}+V_{NBTZ}(r_{})\stackrel{~}{f}_{NBTZ}\stackrel{~}{f}=E\stackrel{~}{f}$$
(6)
where the NBTZ potential is given by
$$V_{NBTZ}(r_{})=\left[\left(\frac{3}{4}+m^2\right)\mathrm{coth}^2[r_{}]m^2+\mathrm{}^2\frac{1}{2}(\mathrm{}^2+\frac{1}{4})\mathrm{tanh}^2[r_{}]\right].$$
(7)
We observe that the potential decreases exponentially to zero ($`V_{NBTZ}e^{2r_{}}`$) as one approaches the event horizon ($`r_{}\mathrm{})`$, while it goes infinity ($`V_{NBTZ}1/r_{}^2`$) as one approaches infinity ($`r_{}0`$). $`V_{NBTZ}(r_{})`$ looks like the right-half of $``$. A plane wave appears near the event horizon, whereas a genuine travelling wave does not appear at infinity. In order to obtain the solution which is valid for whole region outside the black hole, we solve equation (4) directly by transforming it into a hypergeometric equation. With $`z=(r^21)/r^2`$, our working region is between $`z=0`$ and $`z=1`$, covering the exterior of the NBTZ. Eq.(4) takes a form
$$z(1z)f^{\prime \prime }(z)+(1z)f^{}(z)+\frac{1}{4}\left[\frac{\omega ^2}{z}\frac{\mathrm{}^2}{1z}m^2\right]f(z)=0.$$
(8)
In order to obtain quasinormal modes , we use the flux boundary conditionActually, the radial flux of $`\mathrm{\Phi }`$ is expressed in terms of $`f(z)`$ as $`(z=z_0)2\frac{2\pi }{i}[f^{}z_zffz_zf^{}]|_{z=z_0}`$. As in quantum mechanics, this measures the particle current (flow of energy or information). Hence this quantity is usually used to calculate the black hole greybody factor and quasinormal modes. : the ingoing flux $`(_{in}(z=0)<0)`$ at the horizon and the vanishing flux ($`(z=1)=0`$) at infinity. Then we find two types of quasinormal modes with AdS curvature radius $`l`$
$$\omega _{1/2}=\pm \frac{\mathrm{}}{l}i\frac{2}{l}\left(n+s_+\right)$$
(9)
which means that the operator $`_{NBTZ}`$ is not self-adjoint. Here we have $`2s_+=1+(1+m^2l^2)^{1/2}`$. The discreteness comes from the fact that the NBTZ is a compact (finite) system. Decomposing $`\omega _{1/2}=\pm \omega _Ri\omega _I`$, $`w_I`$ should be positive because the corresponding mode decays into the horizon. This bulk perturbation decays, as does in the linear response of conformal field theory. The presence of quasinormal modes is a mathematically precise formalism of the lack of unitarity in the semiclassical approach to the bulk system.
To find a unitary system, we study the pure AdS spacetime which does not contain any topologically distinct object. First we transform the wave equation (5) with lower signs into the Schrödinger-like equation . We introduce a coordinate $`r_{}=\mathrm{tan}^1[r](r=\mathrm{tan}[r_{}])`$ to transform Eq.(5) into the Schrödinger-like equation (6) with the energy $`E=\omega ^2`$ and PADS potential
$$V_{PADS}(r_{})=\left[\left(\mathrm{}^2\frac{1}{4}\right)\mathrm{cot}^2[r_{}]+m^2+\mathrm{}^2+\frac{1}{2}+\left(\frac{3}{4}+m^2\right)\mathrm{tan}^2[r_{}]\right]$$
(10)
which is defined on a box between $`r_{}=0`$ and $`r_{}=\pi /2`$. It is observed that for $`m^2>0`$, $`V_{PADS}(r_{})`$ increases to infinity as one approaches $`r_{}0(r0`$), while it also goes infinity as one approaches $`r_{}\pi /2(r\mathrm{}`$). $`V_{PADS}(r_{})`$ looks like $``$. Using the WKB prescription, we expect to have oscillating modes between two turning points for $`\omega `$ which satisfies $`\omega ^2>V_{PADS}^{min}=2(m^2+\mathrm{}^2)+1`$ and $`\omega ^2=V_{PADS}(r_{})`$.
In order to obtain an explicit form for the frequency $`\omega `$, we solve equation (4) with lower signs directly by transforming it into a hypergeometric equation. With $`z=r^2/(1+r^2)`$, our working region is also between $`z=0`$ and $`z=1`$. Eq.(4) takes a form
$$z(1z)f^{\prime \prime }(z)+(1z)f^{}(z)+\frac{1}{4}\left[\omega ^2\frac{\mathrm{}^2}{z}\frac{m^2}{1z}\right]f(z)=0.$$
(11)
We require that the wave function be zero at infinity because the potential diverges at infinity. This means that the wave function is normalizable and its flux is zero: $`(z=1)=0`$. In addition, requiring a regular solution at $`z=0(r=0)`$ lead to real and discrete modes with AdS curvature radius $`l`$
$$\omega _{1/2}=\pm \frac{\mathrm{}}{l}\pm \frac{2}{l}\left(n+s_+\right)$$
(12)
which means that $`_{PADS}`$ is self-adjoint. The discreteness comes from the finite PADS system. These normal modes are consistent with those found in the AdS approach . Also this oscillating behavior of a bulk perturbation is mirrored by the oscillating behavior of the CFT-boundary approach . It is obvious that we cannot find any complex mode because there is no the event horizon (dissipative object).
## 3 MBTZ
Although the PADS provides a unitary evolution, it is a topologically trivial spacetime. It is important to find a topologically non-trivial spacetime which provides a unitary evolution. One candidate is the massless BTZ black hole. We start with the wave equation (2) in the background of the massless BTZ black hole spacetime: $`ds_{MBTZ}^2=(r/l)^2dt^2+(l/r)^2dr^2+r^2d\varphi ^2`$. Assuming a mode solution (3) with $`l=1`$, the wave equation for $`f(r)`$ is given by
$$r^2f^{\prime \prime }(r)+3rf^{}(r)+\left[\frac{\omega ^2\mathrm{}^2}{r^2}m^2\right]f(r)=0.$$
(13)
Because this equation is so simple, it is not easy to make a potential analysis. Introducing $`r_{}=1/r^2`$, Eq.(13) can be rewritten as the Schrödinger-like equation (6) with the zero energy $`E=0`$ and the potential $`V_{MBTZ}`$
$$V_{MBTZ}(r_{})=\frac{k_1}{r_{}}+\frac{k_2}{r_{}^2},k_1=\frac{\omega ^2\mathrm{}^2}{4},k_2=\frac{m^2}{4}.$$
(14)
It seems that the MBTZ case could not be translated into the boundary value problem because its eigenvalue is determined as $`E=0`$ initially. Near infinity at $`r_{}=0(r=\mathrm{})`$, one finds an approximate equation of $`d^2f_0(r_{})/dr_{}^2(k_2/r_{}^2)f_0(r_{})=0`$ whose solution is given by
$$f_0(r_{})=A_{MBTZ}r_{}^{s_+}+B_{MBTZ}r_{}^s_{}\left(f_{\mathrm{}}(r)=A_{MBTZ}r^{2s_+}+B_{MBTZ}r^{2s_{}}\right)$$
(15)
with $`s_\pm =(1\pm \sqrt{1+m^2})/2`$. Here the first is a normalizable mode and the second is a nonnormalizable mode. However, it is not easy to obtain approximate solution near the event horizon at $`r_{}=\mathrm{}(r=0)`$ since the potential $`V_{MBTZ}`$ contains a long-range interaction term like $`k_1/r_{}`$.
Introducing $`z=1/r`$, the working region is extended from $`z=0`$ to $`z=\mathrm{}`$. Eq.(13) leads to
$$z^2f^{\prime \prime }(z)zf^{}(z)+\left[(\omega ^2\mathrm{}^2)z^2m^2\right]f(z)=0.$$
(16)
In order to solve this equation, one first transforms it into the Bessel’s equation with $`f(z)=z\stackrel{~}{f}(z)`$. Using $`\eta =\sqrt{\omega ^2\mathrm{}^2}z`$, one finds the Bessel’s equation
$$\eta ^2\stackrel{~}{f}^{\prime \prime }(\eta )+\eta \stackrel{~}{f}^{}(\eta )+\left[\eta ^2\nu ^2\right]\stackrel{~}{f}(z)=0$$
(17)
with $`\nu =\sqrt{1+m^2}=2s_+1`$. For massless (dilatonic) scalar, it is given by $`\nu =1(3)`$. Then we find the waveform which is valid for whole region between $`z=0`$ and $`z=\mathrm{}`$
$$f(z)=C_1zJ_\nu \left(\sqrt{\omega ^2\mathrm{}^2}z\right)+C_2zY_\nu \left(\sqrt{\omega ^2\mathrm{}^2}z\right).$$
(18)
In the limit of $`z0(r\mathrm{})`$, one has
$$f(z)f_0(z)=A_{MBTZ}z^{2s_+}+B_{MBTZ}z^{2s_{}}=A_{MBTZ}r^{2s_+}+B_{MBTZ}r^{2s_{}},$$
(19)
where $`f_0(z)`$ is consistent with the approximate solution $`f_{\mathrm{}}(r)`$ in Eq.(15). Hence we choose $`B_{MBTZ}=0(C_2=0)`$ by imposing the boundary condition at infinity of $`r=\mathrm{}`$.
Near the event horizon at $`z=\mathrm{}(r=0)`$, one has
$`f(z)f_{\mathrm{}}(z)=C_1\sqrt{z}\mathrm{cos}\left(\sqrt{\omega ^2\mathrm{}^2}z\pi s_+\right)`$ (20)
$`=C_1{\displaystyle \frac{\sqrt{z}}{2}}\left[e^{i\left(\sqrt{\omega ^2\mathrm{}^2}z\pi s_+\right)}+e^{i\left(\sqrt{\omega ^2\mathrm{}^2}z\pi s_+\right)}\right]f_{MBTZ}^{in}+f_{MBTZ}^{out},`$ (21)
where the first term is an ingoing mode and the last is an outgoing mode. In this case we have $`f_{MBTZ}^{out}=[f_{MBTZ}^{in}]^{}`$ and thus $`f_{\mathrm{}}(z)`$ is real. Then it means that the total flux near the event horizon is zero. However, there exist an ingoing flux and an outgoing flux such that $`^{in}(z=\mathrm{})+^{out}(z=\mathrm{})=0`$. Therefore, we cannot obtain the wanted case: $`^{in}(z=\mathrm{})0,^{out}(z=\mathrm{})=0`$ because the event horizon is a degenerate point and is located at $`r=0`$. In other words, there is no room to determine the frequency $`\omega `$ since the spectrum of $`E`$ is set to be zero initially. This implies that $`_{MBTZ}`$ is self-adjoint and the MBTZ is unitary during evolution. At this stage we have to distinguish between the eigenvalue $`E`$ of $`_{MBTZ}`$ and the own frequency $`\omega `$ for $`\mathrm{\Phi }`$. The continuous frequency reflects that the MBTZ is an infinite (non-compact) system. We conclude that its frequency remains real and continuous. In order to study the extremal black hole further, we need to introduce the other extremal BTZ black hole in the next section.
## 4 EBTZ
We study the wave propagation for a massive scalar field in the background of the extremal BTZ black hole given by $`ds_{EBTZ}^2=[(r/l)^22(r_+/l)^2]dt^2+[r^2l^2/(r^2r_+^2)^2]dr^2[2r_+^2/l]dtd\varphi +r^2d\varphi ^2`$ . In this case $`g^{rr}`$ is also degenerate at the event horizon of $`r=r_+=l\sqrt{M/2}`$. Assuming a mode solution in Eq.(3), the radial equation for $`f(r)`$ is
$`{\displaystyle \frac{(r^2r_+^2)^4}{r^2l^4}}f^{\prime \prime }(r)+{\displaystyle \frac{(r^2r_+^2)^3(3r^2+r_+^2)}{r^3l^4}}f^{}(r)`$ (22)
$`+\left[{\displaystyle \frac{1}{l^2}}(\omega l+\mathrm{})\left((\omega l\mathrm{})r^2+2\mathrm{}r_+^2\right){\displaystyle \frac{m^2}{l^2}}(r^2r_+^2)^2\right]f(r)=0.`$
Choosing a good coordinate $`z=r_+^2/(r^2r_+^2)`$, the above equation reduces to the Schrödinger-like equation (6). Here the potential $`V_{EBTZ}`$ and its energy $`E=k_0^2`$ are given by
$$V_{EBTZ}(z)=\frac{k_1}{z}+\frac{k_2}{z^2},k_0^2=\mathrm{\Omega }_+^2,k_1=\mathrm{\Omega }_+\mathrm{\Omega }_{},k_2=\frac{m^2l^2}{4}$$
(23)
with $`\mathrm{\Omega }_\pm =(\omega l\pm \mathrm{})/\sqrt{2M}`$. We comment that the Schrödinger-like equation for the MBTZ can be obtained from the EBTZ-equation by substituting $`z`$ and $`E=k_0^2`$ into $`r_{}`$ and $`E=0`$. Thus we include the previous MBTZ as the special case of the EBTZ with $`E=0`$.
First we may consider a naively approximate equation of $`d^2f_{\mathrm{}}(z)/dz^2+k_0^2f_{\mathrm{}}(z)=0`$ near the horizon at $`z\mathrm{}(r=r_+)`$ whose solution is given by a plane wave
$$f_{\mathrm{}}=C_{EBTZ}e^{i\mathrm{\Omega }_+z}+D_{EBTZ}e^{i\mathrm{\Omega }_+z},$$
(24)
where the first term corresponds to an ingoing mode and the last is an outgoing one. On the other hand, near infinity at $`z0(r=\mathrm{})`$ one obtains an approximate equation $`d^2f_0(z)/dz^2(k_2/z^2)f_0(z)=0`$ which gives us a solution of $`f_0(z)=A_{EBTZ}z^{s_+}+B_{EBTZ}z^s_{}`$. Here the first term is a normalizable mode and the second is a nonnormalizable mode.
Up to now we obtain approximate solutions near $`z=\mathrm{},0`$. However, we don’t know whether these are true solutions because of the long-range potential $`V_{EBTZ}`$. In order to obtain the solution which is valid for whole region outside the EBTZ, we have to solve equation (22) explicitly. Plugging $`f(z)=f_{\mathrm{}}(z)f_0(z)\stackrel{~}{f}(z)`$ with $`D_{EBTZ}=B_{EBTZ}=0`$ into Eq.(22), it takes the form with $`\xi =2i\mathrm{\Omega }_+z`$
$$\xi \stackrel{~}{f}^{\prime \prime }(\xi )+(2s_+\xi )\stackrel{~}{f}^{}(\xi )\left(s_+\frac{i\mathrm{\Omega }_{}}{2}\right)\stackrel{~}{f}(\xi )=0.$$
(25)
This corresponds to the confluent hypergeometric equation and its solution is given by
$$\stackrel{~}{f}(z)=F[s_+\frac{i\mathrm{\Omega }_{}}{2},2s_+;2i\mathrm{\Omega }_+z].$$
(26)
Considering the Kummer’s transformation of $`F[a,c;\xi ]=e^\xi F[ca,c;\xi ]`$ with $`a=s_+\frac{i\mathrm{\Omega }_{}}{2}`$ and $`c=2s_+`$, it is easy to show that the mode solution $`f(z)`$ is real: $`[f(z)]^{}=f(z)`$.
We choose an ingoing mode near $`z=\mathrm{}`$ and a normalizable solution at $`z=0`$ as the solution which is valid for whole region outside the horizon
$$f(z)e^{i\mathrm{\Omega }_+z}z^{s_+}F[a,c;2i\mathrm{\Omega }_+z].$$
(27)
First we calculate the flux at $`z=0(r=\mathrm{})`$. In this case it confirms that $`f_0(z)z^{s_+}`$ is a real function because $`F[a,c;2i\mathrm{\Omega }_+z]1,e^{i\mathrm{\Omega }_+z}1`$ as $`z0`$. Then it is obvious that the corresponding flux disappears as
$$(z=0)=2\frac{2\pi }{i}[f^{}z_zffz_zf^{}]|_{z=0}=0.$$
(28)
In order to obtain the flux near the event horizon at $`z=\mathrm{}(r=r_+)`$, we use simply the reality condition of $`[f(z)]^{}=f(z)`$. Also we find
$$(z=\mathrm{})=0.$$
(29)
It is very curious from Eqs.(24) and(29) that even though an ingoing mode of $`f_{\mathrm{}}e^{i\mathrm{\Omega }_+z}`$ is present near the horizon, its flux is zero. This implies that we need a further investigation on the wave propagation in the EBTZ background. Also it suggests that there is no restriction on the frequency of $`\omega `$. Thus its mode is real and continuous.
Now let us derive an explicit waveform near the event horizon at $`z=\mathrm{}(r=r_+)`$. For this purpose we introduce the asymptotic expansion of the confluent hypergeometric function for purely imaginary argument $`\xi =2i\mathrm{\Omega }_+`$ and large $`|\xi |`$
$$F[a,c;\xi ]\frac{\mathrm{\Gamma }(c)}{\mathrm{\Gamma }(ca)}|\xi |^ae^{\pm i\pi a/2}+\frac{\mathrm{\Gamma }(c)}{\mathrm{\Gamma }(a)}|\xi |^{ac}e^{\pm i\pi (ac)/2}$$
(30)
where the upper sign being taken if $`\pi /2<\mathrm{arg}(\xi )<3\pi /2`$ and the lower one if $`3\pi /2<\mathrm{arg}(\xi )\pi /2`$. Using the above formula, we can easily prove that the Kummer’s transformation of $`F[a,c;\xi ]=e^\xi F[ca,c;\xi ]`$ is also valid for large $`|\xi |`$. The approximate wave function is given by
$`f(z)=e^{i\mathrm{\Omega }_+z}z^{s_+}F[a,c;2i\mathrm{\Omega }_+z]`$ (31)
$`f_{\mathrm{}}(z)={\displaystyle \frac{\mathrm{\Gamma }(2s_+)}{\mathrm{\Gamma }(s_++i\mathrm{\Omega }_{}/2)}}(2\mathrm{\Omega }_+)^{s_+}e^{\frac{\pi \mathrm{\Omega }_{}}{4}}e^{i\left[\mathrm{\Omega }_+z+\frac{\mathrm{\Omega }_{}}{2}\mathrm{ln}|2\mathrm{\Omega }_+z|\frac{\pi s_+}{2}\right]}`$
$`+{\displaystyle \frac{\mathrm{\Gamma }(2s_+)}{\mathrm{\Gamma }(s_+i\mathrm{\Omega }_{}/2)}}(2\mathrm{\Omega }_+)^{s_+}e^{\frac{\pi \mathrm{\Omega }_{}}{4}}e^{i\left[\mathrm{\Omega }_+z+\frac{\mathrm{\Omega }_{}}{2}\mathrm{ln}|2\mathrm{\Omega }_+z|\frac{\pi s_+}{2}\right]}f_{EBTZ}^{in}+f_{EBTZ}^{out}.`$
Comparing the above with Eq.(24) leads to the fact that the first term corresponds the ingoing mode and the last one is the outgoing mode. We observe here that the presence of $`k_1`$-term in Eq.(23)(like Coulomb potential) prevents the ingoing waveform a plane wave in Eq.(24). Also we note that the contribution from $`k_1`$-term to the phases is a logarithmic function of $`z`$. Even starting with an ugly form of $`f_0(z)z^{s_+}`$, a nearly travelling waveform near the event horizon is developed after transformation. Further it is important to confirm that the wave function is real ($`f_{\mathrm{}}(z)=[f_{\mathrm{}}(z)]^{}`$) near the event horizon because of $`f_{EBTZ}^{out}=[f_{EBTZ}^{in}]^{}`$.
In order to obtain quasinormal modes, it requires that the wave function be a purely ingoing mode near the event horizon and $`f(z=0)=0`$ at infinity. Here we obtain a condition of $`s_+i\mathrm{\Omega }_{}/2=n,n𝐍`$ from $`f_{EBTZ}^{out}=0`$. Then we may find the complex and discrete modes with the AdS curvature radius $`l`$ and $`M=1`$ as
$$\omega =\frac{\mathrm{}}{l}i\frac{2\sqrt{2}}{l}\left(n+s_+\right).$$
(32)
At the first glance there may exist quasinormal modes for a massive scalar propagation on the EBTZ background. However, this condition leads in turn to the zero ingoing flux because the flux expression
$$_{in}(z=\mathrm{})\frac{\mathrm{\Gamma }(2s_+)}{\mathrm{\Gamma }(s_++i\mathrm{\Omega }_{}/2)}\frac{\mathrm{\Gamma }(2s_+)}{\mathrm{\Gamma }(s_+i\mathrm{\Omega }_{}/2)}$$
(33)
leads to zero exactly when choosing $`s_+i\mathrm{\Omega }_{}/2=n`$. This implies that there is no room to accommodate quasinormal modes of a massive scalar in the background of the EBTZ. Therefore, we show that there is no restriction on the frequency $`\omega `$ and thus it remains real and continuous. This is an enhanced situation when comparing it with the MBTZ case. A complete analysis is possible for the EBTZ, because the size of its event horizon is finite and it is located at $`r_+0`$ even it corresponds to a degenerate event horizon. The absence of quasinormal modes in the EBTZ is consistent with the picture of a stable event horizon with thermodynamic properties $`T_H=C_J=0,S_{BH}=4\pi r_+`$. This is so because the presence of quasinormal modes implies that a massive scalar wave is losing its energy continuously into the extremal event horizon. Here we mention that the absence of quasinormal modes for the EBTZ is very similar to the case of the de Sitter cosmological horizon.
The Schrödinger operator $`_{EBTZ}`$ is self-adjoint because its spectrum is real and continuous. Its continuous spectrum reflects the fact that the EBTZ is an infinite system. Consequently, the EBTZ is unitary during evolution without loss of information.
## 5 Summary
We study the wave equation for a massive scalar in three-dimensional AdS-black hole spacetimes to understand the unitarity issues in a semiclassical way. Here we introduce four interesting spacetimes: the non-rotating BTZ black hole (NBTZ), pure AdS spacetime (PADS), massless BTZ black hole (MBTZ), and extremal BTZ black hole (EBTZ). In the NBTZ case, one finds quasinormal modes, whereas one finds real and discrete modes for the PADS case. The presence of quasinormal modes means that it shows a leakage of information into the event horizon (dissipative object) and thus it signals a breakdown of the unitarity. We can easily achieve the unitarity for the PADS. This is not a dissipative system because the perturbations never disappears completely and always can be restored within the Poincaré recurrence time $`t_P`$ as in the motion of oscillation.
On the other hand, we find real and continuous modes for the MBTZ and EBTZ cases. These are unitary systems. The reasons are as follows. Firstly, the Schrödinger operator becomes self-adjoint upon imposing the Dirichlet condition at infinity, as the same condition at the origin of radial coordinate in the Coulomb scattering in quantum mechanics. Secondly, the corresponding wave functions are real in whole region outside the event horizon, especially for near the event horizon and infinity. This means that there is no leakage of information into the two boundaries: event horizon and infinity. This confirms from the fact that their frequencies are real. Thirdly, we find that the radial flux is identically zero outside the event horizon, even though their wave functions are non-zero. Actually, we obtain the ingoing flux as well as the outgoing flux, but summing over these gives us the zero flux near the event horizon exactly. This means that there is no leakage of information into the event horizon. Hence we argue that the two extremal BTZ black holes are unitary systems. In this case we cannot obtain discrete spectra because the two belong to the non-compact system.
Consequently, we propose two additional systems MBTZ and EBTZ for the unitarity system.
A recent work of Hawking does not explain where the semiclassical analysis of the black hole breaks down. In the Euclidean path integral approach, the contribution from the topologically trivial sector (pure AdS space), which he had previously neglected, is sufficient to restore the unitarity. However, his arguments are schematic and thus requires more detailed computations. This proposal seems to be incorrect even in the (1+2)-dimensional AdS spacetimes. In the higher-dimensional AdS spacetimes, there exists the Hawking-Page transition between the AdS black hole and pure AdS space. This is a first-order phase transition. This supports partly that Hawking’s arguments is correct. In the (1+2)-dimensional AdS spacetimes, there exists a second-order phase transition between NBTZ and MBTZ (not PADS). This may explain why we use the MBTZ instead of the PADS, even both are unitary systems. At this stage, however, we don’t know how the EBTZ plays a role in resolving the non-unitarity issue of the non-extremal black hole, the NBTZ.
## Acknowledgement
Y. Myung was supported by the Korea Research Foundation Grant (KRF-2005-013-C00018). H. Lee was in part supported by KOSEF, Astrophysical Research Center for the Structure and Evolution of the Cosmos.
|
warning/0506/astro-ph0506016.html
|
ar5iv
|
text
|
# On the stratorotational instability in the quasi-hydrostatic semi-geostrophic limit Research supported by the Israel Science Foundation, the Helen and Robert Asher Fund and the Technion Fund for the Promotion of Research
## 1 Equations and Boundary Conditions
### 1.1 The Large Shearing Box and Boussinesq Equations
There are a few formal asymptotic derivations for the set of equations appropriate to the dynamics of a localized section of a rotationally supported disc (e.g. Goldreich & Lynden-Bell,1965). For this discussion we shall begin with the Large-Shearing Box (LSB) equations as they appear in Umurhan & Regev (2004),
$`(_tq\mathrm{\Omega }_0x_y)\rho +(\rho _b+\rho )𝐮^{}=\mathrm{𝟎},`$ (1)
$`(_tq\mathrm{\Omega }_0x_y)u^{}+𝐮^{}u^{}2\mathrm{\Omega }_0v^{}={\displaystyle \frac{_xp}{\rho _b+\rho }}`$ (2)
$`(_tq\mathrm{\Omega }_0x_y)v^{}+𝐮^{}v^{}+(2q)\mathrm{\Omega }_0u^{}={\displaystyle \frac{_yp}{\rho _b+\rho }},`$ (3)
$`(_tq\mathrm{\Omega }_0x_y)w^{}+𝐮^{}w^{}={\displaystyle \frac{_zp}{\rho _b+\rho }}{\displaystyle \frac{\rho g(z)}{\rho _b+\rho }},`$ (4)
$`(_tq\mathrm{\Omega }_0x_y)\mathrm{\Sigma }+𝐮^{}\mathrm{\Sigma }=0`$ (5)
in which the total entropy is defined by
$$\mathrm{\Sigma }C_V\mathrm{ln}\frac{p_b+p}{\left(\rho _b+\rho \right)^\gamma }$$
where $`C_V`$ is the specific heat at constant volume and where $`\gamma `$ is the usual ratio of the specific heats at constant pressure and volumes. <sup>4</sup><sup>4</sup>4 In Umurhan & Regev (2004), the entropy (heat) equation was written explicitly in terms of the pressure and densities. There is, then, no substantive difference between these two expressions of the same conservation law. These equations represent the dynamics taking place in a ”large-shearing box” section near the midplane of a Keplerian disc rotating with radius $`R_0`$ about the central star with the local rotation vector, $`\mathrm{\Omega }(R_0)\widehat{𝐳}`$.
The above equations have already been non-dimensionalized. Time is scaled by the local rotation time of the box. All lengths have been scaled according to a length scale $`L`$ which is comparable to the disc thickness. Pressures are scaled according to the local scale of the soundspeed, which is in turn based on some fiducial characteristic temperature scale. For further details see Umurhan & Regev (2004). The above equations, in which the vertical component of gravity is constant, are identical to the equations considered by Tevzadze et al. (2003).
In the language of this paper, $`x`$ corresponds to the shearwise (radial) coordinate which is a small section located around the disc radius $`R_0`$, the azimuthal coordinate $`y`$ is streamwise and $`z`$ is the vertical coordinate corresponding to the normal direction of the original disc midplane. The velocity disturbances in the radial, azimuthal and vertical directions expressed by the variables $`𝐮^{}=\{u^{},v^{},w^{}\}`$. These velocities represent deviations over the steady Keplerian flow.
$`\mathrm{\Omega }_0`$, sometimes also referred to as the Coriolis parameter, is $`1`$ in these nondimensionalized units, meaning to say because time has been scaled according to the dimensional value of the rotation rate at $`R_0`$, i.e. $`\mathrm{\Omega }(R_0)`$, the local Coriolis parameter formally is equal to one. We retain this symbol in order to flag the Coriolis effects in this calculation. The local shear gradient is defined to be
$$q\left[\frac{R}{\mathrm{\Omega }}\left(\frac{\mathrm{\Omega }}{R}\right)\right]_{R_0},$$
(6)
in which $`\mathrm{\Omega }(R)`$ is the full disc rotation rate. For Keplerian discs the value of $`q`$ is $`3/2`$. The local Keplerian flow is represented here by a linear shear in the azimuthal direction, i.e. $`q\mathrm{\Omega }_0x\widehat{𝐲}`$.
The derivation of the LSB equations from the full-scale disc equations begins from the observation that the disc is in fact rotationally supported to leading order. This means that in steady state the disc radial force balance is, to leading order, between rotation and the radial component of gravity as emanating from the central compact object. The radial steady state pressure gradients provide corrections which are on the order of $`ϵ^2=H^2/R_0^2`$, where $`H`$ ia the characteristic height of the disk. Cold discs are those in which $`ϵ`$ is formally assumed to be a small quantity. Given the scaling and expansion steps leading to the LSB this rotational support translates to saying that a radial buoyancy term in a rotationally supported disc is asymptotically smaller than the Coriolis effect by $`𝒪\left(ϵ^2\right)`$. It is for this reason that there is no buoyancy term in (2). However, this rotational balance implies that the vertical structure of the disc is primarily characterized by the usual hydrostatic balance since, by the symmetries inherent to the geometry of the problem, there is no rotational support in that direction. It is for this reason there is an associated buoyancy term in (4). It should be noted that in the context of the LSB, the steady state pressure and density functions are only functions of the $`z`$ coordinate (see below). The vertically oriented gravitational field emanating from the primary compact object is linearly proportional to the distance from the midplane in the LSB system (again see next paragraph). This is a consequence of the asymptotic expansions leading to the LSB in which the central object’s gravitational potential is expanded in a Taylor series about the disc’s symmetry plane (i.e. $`z=0`$). Again, for details of this procedure see Umurhan & Regev (2004).
The steady state density and pressure functions are given by $`p_b,\rho _b`$. These relate to each other according to the aforementioned local hydrostatic equilibrium relationship,
$$_zp_b=\rho _bg(z),$$
with $`g(z)=\mathrm{\Omega }_0^2z`$. The detailed solution for the steady states are then determined once something has been said relating the steady quantities. For simple theoretical investigations this comes in the form of a barotropic equation of state, namely, the statement that $`p_b=p_b(\rho _b)`$. Note that in this paper, especially in Section 4, we explicitly assume that these steady quantities depend only on $`z`$ and not on $`x`$. The dynamic pressures and densities, $`p`$ and $`\rho `$, represent deviations about their corresponding steady state quantities.
We want to gain some insight as to the effects that gravity and gradients of state quantities have on the local dynamics. We take an incremental approach towards this goal by considering a more simplified version of these equations. In more concrete terms, the linear theory (1-5) will show the presence of three ”types” of temporal modes, (i) a pair of acoustic modes, (ii) a pair of inertial-gravity modes and (iii) an entropy mode (Tevzadze et al., 2003). Although the acoustic modes are interesting, we chose to consider the dynamics of a system in which the acoustic modes are effectively filtered out.
To do this we invoke the Boussinesq approximation which, in this sense, we replace the continuity equation (1) with the statement of incompressibility and we replace the entropy conservation equation (5) with an evolution equation for the temperature fluctuation, $`\theta `$. All density fluctuations are set to zero accept the one associated with the buoyancy term in (1) in which it is related to the temperature fluctuation via
$$\rho ^{}=\alpha _p\theta ;\alpha _p\left(\frac{\rho }{T}\right)_p.$$
In other words, $`\alpha _p`$ is the coefficient of thermal expansion at constant pressure. This is the typical formulation of the Boussinesq approximation (Spiegel and Veronis, 1960). We furthermore posit that in this limit the basic state density profile is a constant and in order to distinguish this from a spatially varying density profile we denote the former with $`\overline{\rho }_b`$. The resulting model set of equations are similar to those assumed in the studies by Yavneh et al. (2001), Dubrulle, et al. (2005) (these being viscous studies) and Rudiger et al. (2005) (a cylindrical Taylor-Couette analysis). We have then,
$`𝐮^{}=\mathrm{𝟎},`$ (7)
$`(_tq\mathrm{\Omega }_0x_y)u^{}+𝐮^{}u^{}2\mathrm{\Omega }_0v^{}={\displaystyle \frac{_xp}{\overline{\rho }_b}},`$ (8)
$`(_tq\mathrm{\Omega }_0x_y)v^{}+𝐮^{}v^{}+(2q)\mathrm{\Omega }_0u^{}={\displaystyle \frac{_yp}{\overline{\rho }_b}},`$ (9)
$`(_tq\mathrm{\Omega }_0x_y)w^{}+𝐮^{}w^{}={\displaystyle \frac{_zp}{\overline{\rho }_b}}+{\displaystyle \frac{\theta g(z)\alpha _p}{\overline{\rho }_b}},`$ (10)
$`(_tq\mathrm{\Omega }_0x_y)\theta ^{}+𝐮^{}\theta +w_zT_b=0.`$ (11)
These are otherwise known as the Boussinesq equations (BE) in plane-Couette shear. The term $`T_b`$ is the way in which the basic state temperature profile varies with height in this model formulation of the disc system. Such a term would vary according to $`_zT_b=\overline{T}_{zz}z`$, in other words, the gradient of the basic state temperature has a linear dependence with respect to the disc height in which $`\overline{T}_{zz}`$ is the parameter that controls the slope of this variation. For situations in which $`\overline{T}_{zz}`$ is negative, the atmosphere can be thought of as being classically buoyantly unstable which could lead to Rayleigh-Benard convection (see for instance, Cabot, 1996).
The BE equations here are mathematically equivalent to the inviscid limit of the equations studied in D05. The only difference here is in interpretation. Whereas we follow a temperature perturbation, $`\theta `$, and steady temperature profile, $`T_b`$, they follow an entropy perturbation, denoted by $`h`$, and steady entropy distribution $`H`$. The two disturbance quantities are related to each other via
$$h=\frac{\gamma \alpha _pC__V}{\overline{\rho }_b}\theta .$$
(Also see Appendix D). $`\gamma `$ is the ratio of specific heats and $`C__V`$ is the specific heat at constant volume. The equations (7-11) are the basis of the discussion in Section 3. The steady configuration of these equations which will be perturbed is
$$u^{}=v^{}=w^{}=\theta =0,p=\overline{p}=\mathrm{constant}.$$
(12)
The set (1-5), including their corresponding analogous boundary conditions will be considered in Section 4.
### 1.2 Boundary conditions
There is no obvious choice of boundary conditions for this reduced set of inviscid flow equations. We consider all disturbances to be periodic in the $`y`$ direction. Because these are equations meant to model what happens near the midplane of a circumstellar disc, we can consider periodic conditions in the vertical direction only under special circumstances (i.e. the constant Brunt-Vaisaila frequency approximation of the BE in Section 3, see below). In a more general sense we will distinguish, instead, between either varicose or sinuous modes. By varicose modes we mean to say disturbances which have even symmetry with respect to the $`z=0`$ plane in all disturbances except the vertical velocities, which have odd symmetry with respect to the $`z=0`$ plane. Sinuous modes have the reverse symmetry of the varicose modes. In situations in which a boundary condition needs to be specified on vertical boundaries of the atmosphere, we assume that there is no normal flow.
The more troublesome of the boundary conditions has to do with what to say about the flow variables in the radial direction. This is because an injudicious choice of a boundary condition might incite disturbances of the fluid into instability by drawing energy across the boundaries. It is our interest here to minimize this potential as much as possible. Of the myriad of possible choices, we see the following three sets of boundary conditions as ones that achieve this objective (and motivated further in Appendix A): (a) that the flow be confined between channel walls lying at $`x=0,1`$, which means in practice that the radial velocities are set to zero there, i.e., no normal-flow conditions; (b) the flow has zero Lagrangian pressure fluctuations (defined below) on both of the moving radial boundaries and; (c) a mixture of these two conditions, for example, by requiring there to be no normal-flow on the inner boundary while there is no Lagrangian pressure fluctuation at the outer boundary.
The vanishing of the Lagrangian pressure fluctuation on the undulating radial bounding surface $`𝐒_𝐫`$ in linear theory translates to requiring
$$p^{}+\xi _x_x\overline{p}=0,$$
where $`p^{}`$ is the pressure fluctuation about the steady state. The position of any particular radial surface, initially at rest at coordinate $`x`$, is denoted by $`\xi _x(y,z)`$, and evolves according to its Lagrangian equation of motion (Drazin & Reid, 1984)
$$u^{}(x,y,z,t)=\frac{d\xi _x}{dt}=(_tq\mathrm{\Omega }_0x_y)\xi _x+v^{}_y\xi _x+w_z\xi _x.$$
(13)
Because the steady state pressure configurations of both the BE ($`\overline{p}`$) and the LSB ($`p_b`$) are constant with respect to $`x`$, the condition simplifies to requiring
$$p^{}=0,$$
(14)
at $`x=0,1`$.
## 2 Linear Dynamics of the Boussinesq Equations
In this inviscid limit there emerges a natural timescale defined as the Brunt-Vaisaila frequency, $`N`$. This time scale is defined through the product of the vertical temperature gradient and vertical gravity via
$$N^2g\frac{1}{\overline{\rho }_b}\alpha _p_zT_b$$
(15)
which is, in general, a function of the vertical coordinate $`z`$. Throughout this study $`N`$ is taken to be real (buoyantly stable). Linearization of (7-11) reduces to,
$`\left(_tq\mathrm{\Omega }_0x_y\right)u2\mathrm{\Omega }_0v`$ $`=`$ $`\frac{1}{\overline{\rho }_b}_xP,`$ (16)
$`\left(_tq\mathrm{\Omega }_0x_y\right)v+\mathrm{\Omega }_0(2q)u`$ $`=`$ $`\frac{1}{\overline{\rho }_b}_yP,`$ (17)
$`\left(_tq\mathrm{\Omega }_0x_y\right)w`$ $`=`$ $`\frac{1}{\overline{\rho }_b}_zP+\mathrm{\Theta },`$ (18)
$`\left(_tq\mathrm{\Omega }_0x_y\right)\mathrm{\Theta }`$ $`=`$ $`N^2w,`$ (19)
$`_xu+_yv+_zw`$ $`=`$ $`0,`$ (20)
where the temperature variable has been slightly redefined as $`\mathrm{\Theta }g\alpha _p\theta /\overline{\rho }_b`$. $`\overline{\rho }_b`$ is set to $`1`$ from here on out. It is now to be understood that unprimed velocity expressions (i.e. $`u,v,w`$) represent linearized disturbances.
### 2.1 A conserved quantity for linearized flow
There exists a conserved quantity in these equations. Operating on (16) by $`_y`$ followed by operating (17) by $`_x`$ and subtracting the result reveals
$$\left(_tq\mathrm{\Omega }_0x_y\right)\left(_xv_yu\right)=\mathrm{\Omega }_0(2q)_zw,$$
(21)
where the incompressibility condition has been used. The term on the LHS of this expression is the vertical vorticity, i.e. $`\zeta _xv_yu`$. With a similar tack one can multiply (19) by $`\mathrm{\Omega }_0(2q)/N^2`$ and then operate on the result with $`_z`$ to get
$$\left(_tq\mathrm{\Omega }_0x_y\right)\left(\frac{}{z}\frac{\mathrm{\Omega }_0(2q)}{N^2}\mathrm{\Theta }\right)=\mathrm{\Omega }_0(2q)_zw.$$
(22)
Adding the results together yields a general conserved quantity of linearized flow of this type:
$$\left(_tq\mathrm{\Omega }_0x_x\right)\mathrm{\Xi }=0,$$
(23)
where
$$\mathrm{\Xi }=\zeta +\frac{}{z}\frac{\mathrm{\Omega }_0(2q)}{N^2}\mathrm{\Theta }.$$
(24)
The quantity $`\mathrm{\Xi }`$ can be thought of as a generalized potential vorticity for this type of flow whose analogous quantity is discussed in Tevzadze et al. (2003). The conservation of $`\mathrm{\Xi }`$ immediately implies that there always exists a continuous spectrum (see Schmid & Henningson, 2001, Tevzadze et al., 2003) for this type of physical system. Note also that the system of linearized equations (i.e. 16-20) is third order in time. One may suppose that there are three independent normal modes for any given set of quantum numbers of the system (see below), however, given that there exists a conserved quantity, together with its associated continuous spectrum, it means that there are at most only two normal modes for any given quantum number set.
Inspection of $`\mathrm{\Xi }`$ shows that disturbances behave in a quasi two-dimensional fashion in some limits. One of these is when $`q=2`$, that is at the critical Rayleigh condition (Drazin & Reid, 1984): it follows that the vertical vorticity is conserved by the flow. The second of these is to notice that if the temperature fluctuation remains an order one quantity as $`N^2`$ gets large then the flow again exhibits quasi two-dimensionality with the vertical vorticity being conserved. We reflect upon the consequences of this conserved quantity some more in the Discussion.
### 2.2 Constant $`N^2`$
Since part of the purpose of this work is to further develop some amount of intuition about the dynamics of such disc environments primarily through analytical means, it will be more tractable for us to first treat the vertical gravity component $`g(z)`$ to be
$$g(z)=g_0\mathrm{sgn}(z).$$
(25)
The non-dimensional constant $`g_0`$ is technically arbitrary. In a similar vein we approximate the steady state temperature gradient by saying
$$_zT_b=\overline{T}_z\mathrm{sgn}(z),$$
(26)
in which $`\overline{T}_z`$ is another non-dimensional parameter. The consequence of this is that $`N^2`$ is a constant for $`z0`$, and is zero at $`z=0`$. At this stage, these assumptions are qualitatively no different than what has be done in Yavneh et al. (2001), D05 and Shalybkov & Rudiger (2005), although we take a more realistic interpretation of a constant $`N^2`$ (and see below). From here on out we set $`\overline{\rho }_b=1`$. Additionally, we restrict analysis of the dynamics to $`z>0`$ and keeping in mind that modes are considered to have either sinuous or varicose spatial character in the vertical.
We write general solutions into the form
$$\left(\begin{array}{c}u\\ v\\ P\end{array}\right)=\left(\begin{array}{c}u_{_{\alpha \beta }}\\ v_{_{\alpha \beta }}\\ P_{_{\alpha \beta }}\end{array}\right)\left\{\begin{array}{c}\mathrm{cos}\beta z\\ \mathrm{sin}\beta z\end{array}\right\}e^{i\omega t+i\alpha y}+c.c.,$$
(27)
while for the other variables
$$\left(\begin{array}{c}w\\ \mathrm{\Theta }\end{array}\right)=\left(\begin{array}{c}w_{_{\alpha \beta }}\\ \mathrm{\Theta }_{_{\alpha \beta }}\end{array}\right)\left\{\begin{array}{c}\mathrm{sin}\beta z\\ \mathrm{cos}\beta z\end{array}\right\}e^{i\omega t+i\alpha y}+c.c.,$$
(28)
The terms above in the curly brackets represent varicose disturbances while the terms below are the sinuous disturbances. In this sense the ”quantum numbers” of the system are given by $`\alpha `$, $`\beta `$ (varicose or sinuous) and a radial overtone number (if there are more than one) subject to solution of the normal mode boundary value problem below.<sup>5</sup><sup>5</sup>5The use of quantum numbers should be considered only in terms of conventional nomenclature. There is no real quantization in the horizontal and vertical directions per se since we allow these quantities to take on any value from the continuum of real numbers. Note that because this is a single Fourier expansion we restrict our considerations to $`0<\alpha <\mathrm{}`$ together with $`0<\beta <\mathrm{}`$. Insertion of (27-28) into the governing linear equations gives,
$`i\left(\omega q\mathrm{\Omega }_0x\alpha \right)u_{_{\alpha \beta }}2\mathrm{\Omega }_0v_{_{\alpha \beta }}`$ $`=`$ $`_xP_{_{\alpha \beta }},`$ (29)
$`i\left(\omega q\mathrm{\Omega }_0x\alpha \right)v_{_{\alpha \beta }}+\mathrm{\Omega }_0(2q)u_{_{\alpha \beta }}`$ $`=`$ $`i\alpha P_{_{\alpha \beta }},`$ (30)
$`i\left(\omega q\mathrm{\Omega }_0x\alpha \right)w_{_{\alpha \beta }}`$ $`=`$ $`\beta P_{_{\alpha \beta }}\mathrm{\Theta }_{_{\alpha \beta }},`$ (31)
$`i\left(\omega q\mathrm{\Omega }_0x\alpha \right)\mathrm{\Theta }_{_{\alpha \beta }}`$ $`=`$ $`N^2w_{_{\alpha \beta }},`$ (32)
$`_xu_{_{\alpha \beta }}+i\alpha v_{_{\alpha \beta }}\beta w_{_{\alpha \beta }}`$ $`=`$ $`0,`$ (33)
It turns out that it is much more tractable to consider the linearized normal-mode behavior in terms of equations describing the pressure fluctuation and radial velocities. This is entirely analogous to what was done in D05 and the following equations should be compared to the ones quoted in D05 as Equations (21-22).
$`\left({\displaystyle \frac{\beta ^2\sigma ^2}{N^2\sigma ^2}}\alpha ^2\right)iP_{_{\alpha \beta }}`$ $`=`$ $`\sigma _xu_{_{\alpha \beta }}+\mathrm{\Omega }_0(2q)\alpha u_{_{\alpha \beta }}`$
$`(\sigma ^2\omega _ϵ^2)u_{_{\alpha \beta }}`$ $`=`$ $`(\sigma _xiP_{_{\alpha \beta }}+2\mathrm{\Omega }_0\alpha iP_{_{\alpha \beta }}),`$ (34)
where, for the sake of compact notation we use the expression $`\sigma \omega q\mathrm{\Omega }_0\alpha x`$. The epicyclic frequency $`\omega _ϵ^2`$ is equivalent to the expression $`2(2q)\mathrm{\Omega }_0^2`$.
We are mainly interested in analytically expressible solutions to the above set of equations. To achieve this in an asymptotically rigorous manner the following scalings seem natural: when the horizontal wavenumber is small, it follows that the frequency scales similarly. Using $`ϵ`$ to measure this smallness it follows,
$$\alpha =ϵ\alpha _1,\omega =ϵ\omega _1+\mathrm{}.$$
To lowest order it also follows that $`\sigma =ϵ\sigma _1+\mathrm{}`$. The pressure and velocities are consequently expanded by
$`P_{_{\alpha \beta }}`$ $`=`$ $`P_0+ϵ^2P_2+\mathrm{}`$
$`u_{_{\alpha \beta }}`$ $`=`$ $`ϵu_1+ϵ^3u_3+\mathrm{}`$
Implementing these expansions into the governing equations (34) yields at lowest order in $`ϵ`$ a single equation for the pressure perturbation,
$$(\omega _1q\mathrm{\Omega }_0\alpha _1x)\left(_x^2F_e^2\beta ^2\right)P_0=0.$$
(35)
Normal-mode type solutions to (35) are,
$$P_0=A\mathrm{cosh}k__Fx+B\mathrm{sinh}k__Fx,$$
(36)
where the Froude-wavenumber, $`k__F`$, is defined as $`k__F^2\omega _e^2\beta ^2/N^2=\beta ^2F_e^2`$. The epicyclic-Froude number is denoted by $`F_e`$. This mathematical structure of (35) is identical to the operator describing the evolution of plane-Couette disturbances in a channel (e.g. Case, 1960).
We compare the analytical solutions generated here with numerical solutions generated for the boundary value problem defined by the un-approximated full linearized equation set (34). A second order correct (in the $`x`$ direction derivatives) Newton-Raphson-Kantorovitch (NRK) relaxation scheme on a grid of approximately 1000 to 2000 points is used for the verification. Relative convergence was checked by doubling the size of the domain. Eigenvalues are determined with errors that were no more than $`𝒪\left(5\times 10^6\right)`$. As such the eigenvalues generated asymptotically are considered to be valid in all cases where normal modes exist.
Because this is a relaxation method reasonably good initial guesses are required, both in the eigenfunction and the eigenvalue, in order to accurately obtain an answer. When the initial guesses were far off from the actual solution the scheme admitted solutions belonging to the continuous spectrum. This occurs for all sets of boundary condition but is the only solution possible for the case of fixed pressure conditions (see below). Therefore we discuss the general features of the continuous spectrum in Section 2.2.2. Nonetheless, we find that it helps to avoid the continuous spectrum if the initial eigenvalue guess is set so that $`\mathrm{Im}(\omega )0`$. The possibility of the solution jumping onto a random continuous mode solution is satisfactorily bypassed in this way (also see below).
In what follows we consider the discussion for each of the three boundary conditions. We finally note that the resulting mathematical structure resulting from this asymptotic limit is closely similar to the WKB analysis done in D05. Whereas in this study the small parameter is the horizontal wavenumber, in D05 the small parameter is the shear term $`q\mathrm{\Omega }_0`$ (denoted by ’$`S`$’ in their equation 3). In this sense, their asymptotic form is valid for small values of the shear while ours is valid for small values of $`\alpha `$.
#### 2.2.1 No normal-flow conditions
The wall conditions, i.e. that $`u_{_{\alpha \beta }}=0`$ at $`x=0,1`$, becomes in terms of the pressure the requirement
$$\omega _e^2u_1=i(\sigma _1_xP_0+2\mathrm{\Omega }_0\alpha _1P_0)=0,\mathrm{at}x=0,1,$$
(37)
at lowest order. Implementing these conditions and a little algebra gives a dispersion relation for $`\omega _1`$
$`\omega _1`$ $`=`$ $`\alpha _1q\mathrm{\Omega }_0\left({\displaystyle \frac{1}{2}}\pm {\displaystyle \frac{1}{2k__Fq}}\mathrm{\Delta }__F^{1/2}\right),`$
$`\mathrm{\Delta }__F`$ $`=`$ $`16+k__F^2q^2{\displaystyle \frac{8k__fq}{\mathrm{tanh}k__F}}`$ (38)
As the dispersion relation clearly indicates, if $`\mathrm{\Delta }__F<0`$ then there appears a pair of complex modes, one which grows and one which decays. When $`\mathrm{\Delta }__F>0`$ there are two propagating modes oscillating with no overall growth in amplitude. The character of the stability is dictated only by the two parameters $`q`$ and $`k__F`$. The limit where $`N^20`$ (i.e. $`k__F\mathrm{}`$) reveals that $`\omega _10,q\alpha _1\mathrm{\Omega }_0.`$
The striking feature of this general solution is that there exists a band of vertical wavenumbers for which a stable/unstable solution exists. In Figure 1 we plot this dispersion curve for the case $`q=3/2`$. The plot shows a band in Froude-wavenumber within which the stable/unstable pair exists. Recall that the Froude-wavenumber is really the vertical wavenumber scaled by $`F_e`$. The bifurcation into the stable/unstable pair occurs when the frequencies of the two modes become the same. In the case depicted in the figure ($`q=3/2`$), when the frequencies merge the instability emerges near $`k_F2.1`$. Until about $`k_F3`$, where the instability vanishes, the frequencies remain the same.
The boundaries of the unstable band for general values of $`q`$ may be inferred from the expression for the growth rate: this means determining the function $`q_\pm (k_F)`$ that satisfies $`\mathrm{\Delta }_f(q,k_F)=0`$ as defined in (38). The two functions are $`q_{}=4\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{h}(\mathrm{k}_\mathrm{F}/2)/\mathrm{k}_\mathrm{F}`$ and $`q_+=4\mathrm{c}\mathrm{o}\mathrm{t}\mathrm{h}(\mathrm{k}_\mathrm{F}/2)/\mathrm{k}_\mathrm{F}`$. We see that the band structure for the instability range persists until $`q=2`$, which happens to also correspond to the Rayleigh instability line. Beyond $`q=2`$ the instability range is bounded from below by zero vertical wavenumber but it is still bounded from above by a finite $`\beta `$. As the shear becomes weak, the band of unstable modes gets correspondingly thinner. Note that this instability disappears in the non-shearing limit, that is when $`q0`$. Figure 2 graphically summarizes these results.
#### 2.2.2 Fixed pressure conditions
The boundary conditions on the lowest order pressure conditions becomes
$$P_0=0,$$
(39)
at both boundaries $`x=0,1`$. Given the form of the underlying equations it turns out that there is no normal mode solution possible. The numerical procedure admits solutions, however, these are always modes associated with the continuous spectrum of the linearized system. They are not true normal modes in the usual sense because they exhibit a discontinuity in some quantity: here being in the horizontal velocity, $`v`$, and manifesting explicitly as a step in the quantity $`_xP`$. Discontinuities of this sort, referred to sometimes as singular eigenfunctions, are typical features of modes associated with a continuous spectrum (Case, 1960, Balmforth & Morrison, 1999). In all cases, the location of the discontinuity is at some value of $`x=x_c`$ which is the location of the critical layer, in other words, the place where the quantity $`\omega _1q\mathrm{\Omega }_0\alpha _1x`$ is zero. This means (and this is verified numerically) for given values of $`\alpha `$,$`q`$ and $`\mathrm{\Omega }_0`$ there will be a continuum frequencies, $`\omega _c(\alpha ,q,\mathrm{\Omega }_0)`$, existing between $`0`$ and $`q\mathrm{\Omega }_0\alpha `$.
Figure 7 displays examples of this continuum mode together with an analytic representation of the continuum mode (75) developed in Appendix B.
#### 2.2.3 Mixed conditions
We define terms by identifying mixed-A boundary conditions with zero radial velocities at $`x=0`$ and zero pressure perturbation at $`x=1`$, while mixed-B boundary conditions indicate zero pressure fluctuations at $`x=0`$ with zero radial velocity perturbations at $`x=1`$. Both boundary conditions yield single normal mode solutions Consequently for mixed-A conditions the frequency response is
$$\omega _1=\mathrm{\Omega }_0q\alpha _1\frac{2\mathrm{\Omega }_0\alpha _1}{k__f}\mathrm{tanh}k__F,$$
(40)
while for mixed-B conditions we have
$$\omega _1=\frac{2\mathrm{\Omega }_0\alpha _1}{k__F}\mathrm{tanh}k__F.$$
(41)
In Figure 3 we plot sample eigenfunctions for all boundary conditions we considered in these sections, as well as comparisons between the analytic and numerical solutions obtained.
It is important to mention that when discrete normal modes have frequencies which sit in the continuous sea, i.e. when $`0<\omega <q\mathrm{\Omega }_0\alpha `$ it becomes challenging for the numerical method to not mistake it with a mode belonging to the continuous spectra. To circumvent this possible ambiguity (and to properly numerically verify this limit) we follow modes in $`\beta `$ starting initially with values of the discrete normal mode frequency which is beyond the continuum sea. In this way, discrete normal mode solutions are easily found and, using these as a starting point, one may incrementally move into the realm where normal modes exist within the continuous sea. This is depicted in Figure 4. In all boundary condition cases investigated, we successfully trace the discrete mode spectrum.
### 2.3 The QHSG approximation: vertically varying $`N^2`$
The asymptotic results of the last section hints toward a tractable approach in evaluating the generality of the SRI under a variety of conditions. The limit where the horizontal wavenumbers (i.e. $`\alpha `$) are small suggests that there exists well-posed reduction of the governing equations of motion (8-11). With the azimuthal scales of disturbances scaling as $`𝒪\left(1/\alpha \right)`$ it followed that the temporal disturbances scale as $`𝒪\left(\alpha \right)`$ and, as such, it implied that the radial and vertical velocities similarly scale as $`𝒪\left(\alpha \right)`$ while the pressure, the temperature fluctuation and the azimuthal velocities scale as $`𝒪\left(1\right)`$. We therefore propose that when the horizontal scales are large compared to the corresponding vertical and radial ones the following scalings hold, in general, with respect to quantities and operators of the system:
$$_t,_y,u,w𝒪\left(\alpha \right),_x,_z,p,v,\theta 𝒪\left(1\right).$$
(42)
This means that in the limit where the azimuthal scales are long, we have the following effective reduction of the nonlinear equations of motion (7-11),
$`_xu+_yv+_zw`$ $`=`$ $`0,`$ (43)
$`0+𝒪\left(\alpha ^2\right)`$ $`=`$ $`2\mathrm{\Omega }_0v{\displaystyle \frac{_xp}{\overline{\rho }_b}},`$ (44)
$`(_tq\mathrm{\Omega }_0x_y)v+𝐮^{}v`$ $`=`$ $`(2q)\mathrm{\Omega }_0u{\displaystyle \frac{_yp}{\overline{\rho }_b}},`$ (45)
$`0+𝒪\left(\alpha ^2\right)`$ $`=`$ $`{\displaystyle \frac{_zp}{\overline{\rho }_b}}+{\displaystyle \frac{\theta g(z)\alpha _p}{\overline{\rho }_b}}`$ (46)
$`(_tq\mathrm{\Omega }_0x_y)\theta ^{}+𝐮^{}\theta `$ $`=`$ $`w_zT_b.`$ (47)
The above set is similar to the quasi-geostrophic, quasi-hydrostatic approximation used in the study of atmospheric flows (e.g. Pedlosky, 1987, Salmon, 2002). Whereas in the terrestrial analog full quasi-geostrophy involves retaining the inertial terms in the radial momentum equation, here they are absent (cf. 44), and it is for this reason we consider the above set of equations to be a sort of semi-geostrophic limit. The semi-geostrophic nature of this set shares some similarities with the so-called elongated-vortices equations derived in Barranco et al. (2000). The set presented here differs from that work in that the elongated-vortex equations do not make the hydrostatic approximation as is a natural and necessary consequence here.
The power in this reduced set of equations, aside from exactly reproducing the asymptotic limit explored in the previous section, is that it allows one to investigate the effects that a position dependent function of gravity and background state temperature gradient, i.e. $`g(z)`$ and $`_zT_b`$, have on the SRI. In this sense, unlike the approximation utilized in Section 2.2, we relax the condition that $`g`$ and $`_zT_b`$ are constants and let them be general functions of $`z`$. It therefore means that the Brunt-Vaisaila frequency is now $`z`$-dependent.
When specific forms are considered here we assume that these quantities are simply proportional to $`z`$, that is to say,
$$g(z)=\mathrm{\Omega }_0^2z;_zT_b=\stackrel{~}{T}_{zz}z.$$
(48)
The constant $`\stackrel{~}{T}_{zz}`$ sets the severity of the background temperature gradient (see Section 1.1).
We linearize the set (43-47) about the quiet state $`u=v=w=\theta =0`$, $`p=`$ constant. Disturbances are denoted with primes. A little algebra shows that these equations may be simplified into a single one for the pressure perturbation:
$$(_t\mathrm{\Omega }qx_y)\left[\frac{}{z}\frac{\omega _ϵ^2}{N^2(z)}\frac{p^{}}{z}+\frac{^2p^{}}{x^2}\right]=0,$$
(49)
where the $`z`$-dependent Brunt-Vaisaila frequency is given as
$$N^2(z)=\frac{1}{\overline{\rho }_b}\alpha _pg_zT_b=\frac{1}{\overline{\rho }_b}\mathrm{\Omega }_0^2\stackrel{~}{T}_{zz}\alpha _pz^2.$$
This all also means that we can consider a $`z`$-dependent Froude number according to
$$F_ϵ^2=\frac{\omega _e^2}{N^2(z)}=\stackrel{~}{F}_e^2\frac{1}{z^2},$$
(50)
because all variables and quantities have been non-dimensionalized, the Froude-number scale $`\stackrel{~}{F}_e`$ should be considered in parallel to the Froude-number $`F_e`$ treated in Section 2.2.
The expression inside the square brackets of (49) looks analogous to the potential-vorticity of atmospheric flows. When $`N^2(z)`$ is a constant then the expression within the square brackets exactly recovers the asymptotically valid governing equation for the pressure perturbation in (35).
Separable solutions are assumed of the form
$$p^{}=\mathrm{\Pi }(z)P_0(x)e^{i\omega t+i\alpha y}+\mathrm{c}.\mathrm{c}.$$
(51)
and applied to the governing equation (49). This results in two ODE’s, one for the vertical structure function and one for the radial structure function,
$`{\displaystyle \frac{}{z}}F_ϵ^2{\displaystyle \frac{\mathrm{\Pi }}{z}}`$ $`=`$ $`k__F^2\mathrm{\Pi },`$ (52)
$`{\displaystyle \frac{^2P_0}{x^2}}`$ $`=`$ $`k__F^2P_0`$ (53)
The separation constant for this procedure is $`k__F`$. We notice immediately that the equation for the radial structure function is mathematically the equivalent to (35). Since the boundary conditions and associated relationships are identical in this asymptotic limit, it follows then the same stability properties that was determined in Section 2.2 carry over here to this particular example of a $`z`$-dependent function of $`N^2`$ if the allowed values of $`k__F`$ are real. In remaining consistent with the terminology introduced in D05, modes for which $`k__F`$ are real are referred hereafter as exponential, or as e-modes, since this describes the quality of the radial structure function that results from solving (53). By contrast, modes in which the $`k__F`$ are imaginary are referred to as oscillating or as o-modes, (again see D05) and *these in principle will have different stability properties* than those determined for the e-modes.
Thus the task that remains is to determine the allowed values of the separation constant. The analysis below indicates that that the only types of disturbances permitted for $`N^2(z)z^2`$ are *e-modes on a finite domain*.
#### 2.3.1 Exponential modes
Though these may be artificial, for the sake of simplicity and comparison we consider only the vanishing of the normal velocities at the boundaries $`z=\pm 1`$ (e.g. Barranco & Marcus, 2005). When assuming the specific forms for $`g`$ and $`_zT_b`$ as in (48) we find two possible solution forms to (52)
$$\mathrm{\Pi }(z)=\left\{\begin{array}{c}\mathrm{\Pi }_{(varicose)}\\ \mathrm{\Pi }_{(sinuous)}\end{array}\right\}=\left\{\begin{array}{c}z^{\frac{3}{2}}𝒥_{_{\frac{3}{4}}}\left(\frac{1}{2}\frac{k__F}{\stackrel{~}{F}_e}z^2\right)\\ z^{\frac{3}{2}}𝒥_{_{\frac{3}{4}}}\left(\frac{1}{2}\frac{k__F}{\stackrel{~}{F}_e}z^2\right)\end{array}\right\},$$
(54)
where the symbol $`𝒥`$ denotes the Bessel function of the first kind. The expression $`k__F/\stackrel{~}{F}_e`$ can be considered to be parallel to the vertical wavenumber $`\beta `$ introduced and treated in Section 2.2. A Taylor Series expansion of these solutions near $`z=0`$ verifies the even (odd) symmetry of the varicose (sinuous) solutions. Given these inherent symmetries we are left with the task of setting to zero $`w^{}`$ at $`z=1`$ which, given the relationships between (46-47), is equivalent to setting $`_zp^{}=0`$ there. Given the general properties of Bessel Functions (Abramowitz & Stegun, 1972) the set of $`k__F`$ values that satisfy the boundary conditions are always real. In Figure 5 we display the functions $`\mathrm{\Pi }(z)`$ for the first three values of the separation constant $`k__F`$. It should be noted that the asymptotic expansion of the solution forms presented above are characterized by an amplitude function which grows as $`\sqrt{z}`$ for $`|z|\mathrm{}`$.
#### 2.3.2 Oscillating modes
Consideration of o-modes starts by redefining the separation constant $`k__F`$ according to
$$k__F=i\kappa __F,$$
where $`\kappa __F`$ is real. Therefore, as in the previous section there are two independent solutions to (52) with $`k__F`$ so defined given by
$$z^{\frac{3}{2}}I_{_{\frac{3}{4}}}\left(\frac{1}{2}\frac{\kappa __F}{\stackrel{~}{F}_e}z^2\right),z^{\frac{3}{2}}I_{_{\frac{3}{4}}}\left(\frac{1}{2}\frac{\kappa __F}{\stackrel{~}{F}_e}z^2\right),$$
respectively corresponding to sinuous and varicose disturbances. However, Modified Bessel Functions as these do not have zeros for real values of the ratio $`\kappa __F/\stackrel{~}{F}_e`$. It means that it is not possible to satisfy the same sort of boundary conditions considered in the previous section (e.g. $`w=0`$ at $`z=\pm 1`$) which, in turn, means that o-modes are not allowed solutions on a finite vertical domain with no-vertical flow boundary conditions.
However, a consideration of the problem of (52) on an infinite domain (i.e. $`z\pm \mathrm{}`$) where, instead, we require boundedness and asymptotic decay of all quantities as $`|z|\mathrm{}`$, suggests that the solution for $`\mathrm{\Pi }(z)`$ could be the following,
$$\mathrm{\Pi }(z)=z^{\frac{3}{2}}𝒦_{_{\frac{3}{4}}}\left(\frac{1}{2}\frac{\kappa __F}{\stackrel{~}{F}_e}z^2\right),$$
(55)
where $`𝒦__\nu \left(x\right)`$ is the Modified Bessel Function of the Second Kind (Abramowitz & Stegun, 1972). Indeed an asymptotic expansion of the leading order behavior of $`𝒦_{_{3/4}}\left(z^2\right)`$ in the large argument limit shows that it behaves like a Gaussian, i.e. like $`\mathrm{exp}\frac{1}{2}\frac{\kappa __F}{\stackrel{~}{F}_e}z^2`$, for large $`z^2`$. On its surface this behaviour seems to satisfy the requirements on the functions as $`z\pm \mathrm{}`$. Although the above conclusion is correct for $`z\mathrm{}`$, to extend this conclusion as $`z\mathrm{}`$ based on the above representation would be wrong. In fact, given the functional form (55), it becomes a matter of subtlety as to how one must cross the point $`z=0`$. In Appendix C it is shown that the behaviour of (55) *for $`z<0`$ is*
$$(z)^{\frac{3}{2}}\left[𝒦_{_{\frac{3}{4}}}\left(\frac{1}{2}\frac{\kappa __F}{\stackrel{~}{F}_e}\left(z\right)^2\right)2_{_{\frac{3}{4}}}\left(\frac{1}{2}\frac{\kappa __F}{\stackrel{~}{F}_e}\left(z\right)^2\right)\right].$$
(56)
Inspection reveals exponential divergence as $`z\mathrm{}`$ since $`_\nu (x)`$ behaves exponentially as $`x\mathrm{}`$. It appears that this analysis indicates that there are no bounded solutions possible for $`\mathrm{\Pi }(z)`$ and, consequently, it implies there are no o-modes permitted when $`N^2(z)z^2`$ in the context of the BE model.
## 3 Large shearing box equations: the QHSG approximation and linearized dynamics
Given the clues revealed by using the QHSG for the BE, we consider the same scalings expressed in (42) and apply them to the full LSB equations (1-5). The major departure here, of course, is that disturbances are now not incompressible. One scaling relationship is to say that the density and pressure variables are of comparable scale, i.e. $`𝒪\left(\rho \right)=𝒪\left(p\right)1`$. Therefore the QHSG reduction of the nonlinear LSB becomes
$`(_tq\mathrm{\Omega }_0x_y)\rho +(\rho _b+\rho )𝐮^{}=\mathrm{𝟎},`$ (57)
$`0=2\mathrm{\Omega }_0v{\displaystyle \frac{_xp}{\rho _b+\rho }}+𝒪\left(\alpha ^2\right),`$ (58)
$`(_tq\mathrm{\Omega }_0x_y)v^{}+𝐮^{}u+(2q)\mathrm{\Omega }_0u={\displaystyle \frac{_yp}{\rho _b+\rho }},`$ (59)
$`0=_zp\rho g(z)+𝒪\left(\alpha ^2\right),`$ (60)
$`(_tq\mathrm{\Omega }_0x_y)\mathrm{\Sigma }+𝐮^{}(\mathrm{\Sigma }_b+\mathrm{\Sigma })=0,`$ (61)
where we have introduced the basic state entropy $`\mathrm{\Sigma }_b`$ and its dynamically varying counterpart $`\mathrm{\Sigma }`$ which are defined by
$$\mathrm{\Sigma }_b=\mathrm{ln}\frac{p_b}{\rho _b^\gamma },\mathrm{\Sigma }=\mathrm{ln}\frac{1+\frac{p}{p_b}}{\left(1+\frac{\rho }{\rho _b}\right)^\gamma },$$
(62)
where $`\gamma `$ is the the usual thermodynamic ratio of specific heats. Linearizing (57-61) and sorting through the algebra (see Appendix D) leaves us with a single master equation for the pressure perturbation
$$(_tq\mathrm{\Omega }_0x_y)\left[\frac{\omega _ϵ^2}{g}_zp+_z\frac{\omega _ϵ^2}{N__\mathrm{\Sigma }^2}\left(\frac{g}{c^2}p+_zp\right)+_x^2p\right]=0.$$
(63)
The generalized Brunt-Vaisaila frequency is defined by
$$N__\mathrm{\Sigma }^2\frac{g}{\gamma }_z\mathrm{ln}\frac{p_b}{\rho _b^\gamma },$$
(64)
while the (nondimensional) adiabatic soundspeed $`c`$ is defined by
$$c^2\frac{\gamma p_b}{\rho _b}.$$
(65)
(63) is the LSB equivalent, in this QHSG limit, of a potential vorticity for a local section of a circumstellar disc. Comparing this equation for the potential-vorticity with the analogous one for the BE in (49) reveals some differences between them being, namely,
$$\frac{\omega _ϵ^2}{g}\frac{p}{z},\frac{}{z}\frac{\omega _ϵ^2}{N__\mathrm{\Sigma }^2}\frac{g}{c^2}p.$$
The first of these is associated with the time rate of change of the density fluctuation in the continuity equation. This is explicitly absent in the BE due to the assumption of incompressibility. The second of these is associated with the generalized entropy fluctuation and is inversely proportional to the soundspeed. This term is absent in the Boussinesq Equations because the assumption of incompressibility is equivalent to the interpretation that the soundspeed is infinite.
Having $`N__\mathrm{\Sigma }^2<0`$ is equivalent to the Schwarzschild condition for buoyant instability (Tassoul, 2000). As before, we assume that the atmosphere is stable to buoyant oscillations ($`N__\mathrm{\Sigma }^2>0`$). However we must also say something about the soundspeed $`c`$: for the sake of this discussion we will assume that it is a constant with respect to $`z`$, that is, we assume the atmosphere is isothermal. We proceed toward determining normal mode solutions of the expression inside the square brackets of (63),
$$\frac{\omega _ϵ^2}{g}\frac{p}{z}+\frac{}{z}\frac{\omega _ϵ^2}{N__\mathrm{\Sigma }^2}\left(\frac{g}{c^2}p+\frac{p}{z}\right)+\frac{^2p}{x^2}=0$$
(66)
Assuming separable solutions of the form (51) we find, once again, the following two problems to solve:
$`{\displaystyle \frac{\omega _ϵ^2}{g}}{\displaystyle \frac{\mathrm{\Pi }}{z}}+{\displaystyle \frac{}{z}}{\displaystyle \frac{\omega _ϵ^2}{N__\mathrm{\Sigma }^2}}\left({\displaystyle \frac{g}{c^2}}\mathrm{\Pi }+{\displaystyle \frac{\mathrm{\Pi }}{z}}\right)`$ $`=`$ $`k__F^2\mathrm{\Pi },`$ (67)
$`{\displaystyle \frac{^2P_0}{x^2}}`$ $`=`$ $`k__F^2P_0`$ (68)
The separation constant $`k__F`$ is the same as before.
Because the equation for $`P_0`$ is the same as in the BE model, cf. (53), it immediately follows that the same stability properties that was determined for the BE apply here too if the set of $`k__F`$ values are all real (i.e. e-modes). For this study we restrict our attention to finite vertical domains (see below). It means, then, that the task that remains is to determine the eigenvalues of $`k__F`$ by seeking solutions of (67) subject to the boundary condition that the vertical velocity vanishes at $`z=\pm 1`$. In the LSB, this condition amounts to setting
$$\frac{1}{N_\mathrm{\Sigma }^2}\left(\frac{g}{c^2}\mathrm{\Pi }+\frac{\mathrm{\Pi }}{z}\right)=0,$$
(69)
at $`z=\pm 1`$. (see Appendix D). Also, as before, we consider solutions to $`\mathrm{\Pi }`$ that are either sinuous or varicose.
We emphasize that we are restricting our attention here to solutions on a finite $`z`$ domain. This is because attention needed to treat such problems on an infinite domain is challenging and it is, thus, outside the scope of this current work (see Discussion).
Aside from very special values of the parameters, there are no simple or analytically tractable solutions to the ordinary differential equation posed by (67) <sup>6</sup><sup>6</sup>6General solutions of this equation are linear combinations of hypergeometric functions which require numerical evaluation anyway.. Therefore we numerically solve for this equation and $`k__F`$ using a fourth-order variant of the NRK scheme discussed in the previous section. We use a grid of 300 points which lets us determine solutions up to machine accuracy (i.e. an error of less than $`10^{11})`$. The solutions were all normalized by setting $`\mathrm{\Pi }=1`$ at $`z=1`$. We verify the robustness of the numerical scheme by using it to solve the simpler equation (49) and comparing the numerically generated results against the exact solutions (54).
There are two parameters that govern the solutions. The first of these is the scale measure, $`\overline{F}__\mathrm{\Sigma }`$, of the height-dependent Froude-number
$$F__\mathrm{\Sigma }^2=\frac{\omega _ϵ^2}{N__\mathrm{\Sigma }^2}=\overline{F}__\mathrm{\Sigma }^2z^2.$$
Given that this atmosphere is isothermal this Froude-number scale is measured by the parameter
$$\overline{F}__\mathrm{\Sigma }^2=\frac{\omega _ϵ^2\gamma }{\mathrm{\Omega }_0^2(\gamma 1)}.$$
For a medium dominated by molecular hydrogen $`\gamma 7/5`$. It means that in a Keplerian flow $`\overline{F}__\mathrm{\Sigma }^27/2`$. The second parameter is the relative measure of the vertical scale height of the atmosphere defined by $`H`$ and given to be
$$H^2\frac{c^2}{\mathrm{\Omega }_0^2}=\frac{\gamma _\mu \overline{T}}{\mathrm{\Omega }_0^2},$$
in which $`_\mu `$ is the non-dimensionalized gas constant for the given composition, $`\overline{T}`$ is the non-dimensionalized temperature of the atmosphere. This quantity is essentially the same as the classic $`ϵ`$-parameter governing thin-disc theory (Shakura and Sunyaev, 1973, Lynden-Bell and Pringle, 1974) When $`H`$ is small, the atmosphere is very shallow and, consequently, cold.
The solutions that we scan all show that the $`k__F`$ values are always real indicating these are e-modes (cf. Section 2.3). We were unable to find purely imaginary solutions for $`k__F`$ on this finite domain. Thus it implies that the stability properties determined for e-modes (i.e. Section 2) carry over to here too and that this system does not support o-modes on this finite-domain.
## 4 Summary and Discussion
### 4.1 The QHSG and the persistence of the SRI with height dependence in $`N^2`$
We have achieved here an extension of the SRI to models which takes into consideration the vertical structure of the physical environment. One of the departures taken here from previous work is to explicitly include the effects of a vertically varying Brunt-Vaisaila frequency. The results of the previous sections shows that the SRI, under channel-wall boundary conditions, persists unaltered irrespective of the model equations considered (i.e. either the BE or the LSB) or type of mode (i.e. e-mode or o-mode) so long as one is in the inviscid-QHSG asymptotic limit. This is not to say that if one relaxes the restriction of long horizontal length scale disturbances (i.e. small $`\alpha `$) then this instability will not continue - we merely mean to say that its existence under those conditions remains open. It does seem likely, however, given the pattern of the presence of the SRI in D05, that it will do so also in the $`𝒪\left(\alpha \right)1`$ case too.
The advantage of the inviscid-QHSG approximation employed here is that an analytical analysis of normal-modes is possible through a separation of variables. In general cases where both $`g`$ and the other state variables like $`\rho _b`$ and $`p_b`$ are functions of the vertical coordinate $`z`$, the resulting linear equations and mode structure are generally non-separable. This makes for assessing the normal-mode behaviour of disturbances to be challenging (at best) although not impossible, as has been shown here. In this sense it appears that this is a reasonable peek into situations where complicating background structure in both the radial direction (here the shear) and vertical direction (here gravity and other state variables) can be taken into account together.
It is also worth noting that in the inviscid QHSG limit of both the BE and LSB the radial structure of the eigenmodes (be they either exponential or oscillatory) are unaffected by the vertical stratification: in other words, the radial eigenfunctions are always the same. As a result, the stability criterion turns out to be insensitive to vertical stratification in the state variables of the system.
The QHSG limiting process is achieved by looking at disturbances with horizontal length scales that are large compared to the other dimensions which, therefore, implies that the associated time scale of disturbances can scale with proportional smallness. From the framework of the the LSB equations, the implication is that one recovers the part of the dynamics of inertio-gravity waves modified, to some extent, by the effect of weak compressibility (dilatation) and a finite sound speed. The inclusion of these effects is not to say that some facet of acoustic disturbances are recovered in this limit: this is precisely because the time scales associated with acoustics are much shorter than the time scales explored here. In this sense this asymptotic limit naturally filters out direct acoustic effects and preserves the dynamics of disturbances that would usually be associated with Rossby waves in geophysical flows (Pedlosky, 1987). This is also not meant to imply that acoustic effects are unimportant (an issue that is far from settled), it merely means that this limit is effective at isolating the dynamics of these waves.
### 4.2 On the nature of e-modes and o-modes with height dependent $`N^2`$
D05 showed that there are two types of SRI modes which are characterized by the quality of the mode’s radial structure function: either exponential or oscillatory. Because both the vertical background temperature gradient and component of gravity are constant D05 showed that both e-modes and o-modes are supported for such a model atmosphere.
For example, e-modes, which have vertical structure functions which are sinusoidal with respect to the vertical coordinate, are permitted in D05 even if the atmosphere extends mathematically to infinity. Because $`N^2`$ is constant with respect to $`z`$, the corresponding vertical structure functions have no envelope growth or decay with respect to the vertical coordinate. As such, such solutions satisfy reasonable expectations of boundedness as the atmosphere extends indefinitely.
By contrast the analysis of the QHSG limit of the BE equations, in which $`N^2`$ depends on $`z`$ quadratically, shows that the resulting vertical structure functions for e-modes to have envelope structures which grow ($`\sqrt{|z|}`$). This fact makes it impossible to meaningfully impose boundary conditions or boundedness conditions on solutions as $`|z|\mathrm{}`$. The analysis performed here seems to indicate that e-modes are restricted to BE systems involving finite vertical domains with height dependent $`N^2`$.
O-modes in model atmospheres with quadratic dependence of $`N^2`$ with respect to $`z`$ are not allowed. In atmospheres with constant $`N^2`$ o-modes are admitted on account of the exponential decay of the vertical structure function. For $`N^2(z)z^2`$ (cf. 2.3.2) it appears there is no way to construct bounded solutions in the directions $`z\pm \mathrm{}`$ simultaneously. It was also demonstrated that o-modes are ruled out on finite domains.
The conclusions regarding the existence of o-modes is mainly based on the analysis of the BE with $`N^2(z)z^2`$ in the QHSG limit. We showed also in Section 3 that similar conclusions seem to hold for SRI e-modes and o-modes in the isothermal-QHSG limit of the LSB model set when considered only on a finite domain. It is not entirely clear how the existence properties of SRI modes are affected when: (i) the domain mathematically extends to infinity, (ii) the soundspeed varies with height. These questions are reasonable points of departure for further investigation.
### 4.3 The absence of the SRI for non-reflecting boundaries
The troubling aspect of this investigation is that when one considers boundary conditions other than no-flow conditions on the radial boundaries, the instability appears to vanish in the asymptotic limit considered. When the Lagrangian pressure is zero on both radial boundaries the analysis predicts that there are no normal-modes with the time scales assumed and, furthermore, there are only modes associated with the continuous spectrum. It is probably safe to conclude that for this type of boundary condition, that there are no normal-modes whose frequencies have magnitudes on the order of or greater than the small scaling parameter ($`\alpha `$, the horizontal wavenumber) of the limit explored here. In situations where there is a mixture of no-flow conditions on one boundary and no Lagrangian perturbations on the other, only one normal-mode is admitted by the system which propagates with no growth or decay.
The circumstance encountered here shares a number of similarities with the Eady problem of baroclinic instability in geophysical shear flows (Eady 1949, Criminale & Drazin, 1990). Drazin and Reid (1994) show that the Eady problem is essentially equivalent to the stability of inviscid plane Couette flow (pCf) subject to boundary conditions where the pressure perturbations are fixed on the channel walls, instead of the usual condition in which the normal velocities are set to zero. As Case (1960) showed, inviscid pCf flows with no-normal flow boundary conditions admit only continuous spectrum modes and no discrete modes. By contrast, Criminale & Drazin (1990) showed that the Eady problem has, in addition to the continuous spectrum, a number of discrete modes present (which are possibly unstable under suitable conditions of the disturbances) when disturbances in inviscid pCf flow have fixed pressures at the boundaries (Criminale et al. 2003).
For the SRI investigated here, an entirely analogous situation occurs: the linear operator governing the system is mathematically equivalent to the one characterizing inviscid pCf, but, the variables and stability characteristics are interchanged. Whereas in inviscid pCf the operator operates on the radial velocity (wall-normal) in the SRI case here it operates on the pressure perturbation. Thus the inviscid pCf problem admits normal modes (no normal modes) for fixed pressure (no normal flow) boundary conditions while the SRI problem admits normal modes (no normal modes) for no normal flow (fixed pressure) conditions. Of course, both scenarios reveal the presence of a continuous spectrum irrespective of the boundary conditions employed. It seems as though the inviscid pCf and SRI problems have properties and stability characteristics that are interchanged.
### 4.4 Questions and a conjecture
From a more physically motivated standpoint, we have experimented with these set of boundary conditions because they allow one to exert some comparative control between conditions. It is shown in Appendix A that disturbances in the BE, subject to these boundary conditions, have a total disturbance energy, $`E`$, which evolves according to the exchange of energy that takes place between the (Keplerian) shear and disturbance modes via the Orr-Mechanism and measured by the Reynolds Stress term, i.e. the RHS of (72). Conditions other than these would cause there to be some net work (positive or negative) to be performed on the layer during the ensuing course of the disturbances (Schmid & Henningson, 2001).
It is a puzzle, then, that in this QHSG limit there is an instability in the case of no-normal flow conditions and none otherwise. Although this is merely a conjecture, is it possible that the SRI occurs because of the double reflecting boundary conditions? The instability shares many of the same properties of the acoustic instability uncovered by Papaloizou & Pringle (1984,1985), otherwise known as the PP instability (Li et al. 2000). It is an instability of an acoustic disturbance in a domain like this with reflecting inner and outer walls in which there exists a critical layer, sometimes referred to as a corotation radius (Li et al., 2000). The waves grow in a resonant fashion because the reflecting walls, either one or both, allows for repeated passages of the wave across the critical layer which allows it to draw energy from the shear (Drury, 1985). It was found that the PP instability vanishes when the amplifying agent, usually the second reflecting boundary, is removed (Narayan, Goldreich & Goodman, 1987).
Like the PP instability, the SRI as determined in this work are waves existing in a domain containing a critical point along with reflecting boundaries. When one of the boundaries no longer reflects, there is no instability. Although these are neither compressible modes nor two-dimensional is it possible that the SRI arises in an analogous way due to the pathology that afflicts the PP instability? This is an open question which should be clarified in future work.
A clue towards this end might be found in the observation that there exists a second energy integral, as developed in Appendix D, involving a total energy expression $``$ which says something interesting. The domain integral of the quantity $``$ is conserved under no-normal radial flow conditions whereas it is not for the others. Perhaps the instability is related to this constraint placed on the dynamics of the system?
### 4.5 Flow two-dimensionalization and another conjecture
We demonstrated in Section 2.1 that there exists a conserved quantity ($`\mathrm{\Xi }`$) of the general linearized system of the BE that is advected by the local basic shear. This quantity, which looks like a potential vorticity, is conserved independent of the QHSG asymptotic limit explored. Its analog is implied to exist in the for the LSB model equations as discussed by Tevzadze et al. (2003). According to its definition, (24), $`\mathrm{\Xi }`$ is composed of the vertical vorticity and a quantity representing buoyancy motions driven by density fluctuations. We also noted that in the limit where the buoyancy oscillations become very strong the term associated with it in $`\mathrm{\Xi }`$ may become less important. In this circumstance it implies that the flow will take on a nearly two-dimensional character.
In particular if the quantity $`N^2`$ becomes large then, according to (19) one possible scaling between quantities in an initial value problem is to have $`\mathrm{\Theta }`$ remain an order 1 quantity while the vertical velocity, $`w`$, be $`𝒪\left(N^2\right)`$. If all other quantities remain correspondingly order 1, that is to say if $`u,v,P,_x,_y,_z,_t𝒪\left(1\right)`$, then to lowest order it would imply that the disturbances are dynamically in hydrostatic equilibrium and it would imply that the flow is nearly two dimensional conserving its vertical component of vorticity (cf. 23-24).
Barranco and Marcus (2005) demonstrated, in their shearing sheet simulations of a stratified fluid, the appearance of coherent vortical structures with vorticity vector pointing in the vertical direction. When they manifested themselves, the anticyclonic vortices appear near the vertical boundaries of the system, in other words, in that part of the atmosphere where the vertical component of gravity is greatest in magnitude. They also demonstrated the robustness and persistence of these anticyclones by artificially removing the vortex structure (after having developed) and replacing the flow field with noise. They show that the noisy spectrum quickly redeveloped into coherent anticyclone(s) much as it is known to do so in two dimensional shear flows (e.g. Umurhan & Regev, 2004). This fact is consistent with the implications suggested by the advected conservation of the linear quantity $`\mathrm{\Xi }`$. Of course, only a nonlinear reformulation and reexamination of $`\mathrm{\Xi }`$ can offer a more solid basis to any connection that there may exist here.
Is it possible that it is a generic feature of stratified flow with a Couette shear profile and a vertically dependent Brunt-Vaisaila frequency (e.g. appropriate for a local representation of a circumstellar disk as here) to behave two-dimensionally in substantial parts of the atmosphere significantly away from the disk midplane, i.e. those regions dominated by a large Brunt-Vaisaila frequency?
## 5 Acknowledgements
I thank the anonymous referee who made a critical suggestion pertaining to the existence of oscillatory modes in this study. I also thank Oded Regev for his suggestions and support for this work. I am also deeply indebted to discussions with Professors Hugh Davies and Eyal Heifetz who helped me to cast the asymptotic limits developed here in terms of a quasi-geostrophic formalism.
## Appendix A An Integral Statement for the Boussinesq Equations
It is instructive to consider integrals of the system since they can help guide one into deciding which boundary conditions to be used. We begin with by noticing that for the situations considered in here, the functional forms relating $`g`$ and $`_zT_b`$ are always constant multiplicative factors of each other (see Sections 2.1 and 2.2). Therefore we take the ratio of these two quantities to be always a constant, that is,
$$_zT_b/g=\mathrm{constant},$$
over the full spatial domain under consideration. With this assumption in hand one may (i) take the scalar product of (8-10) and $`\rho _b𝐮^{}`$ , (ii) multiply (11) by $`\theta g\alpha _p/_zT_b`$ and (iii) adding the results of (i) and (ii) together and making use of the incompressibility condition (7) to reveal
$$\left(_tq\mathrm{\Omega }_0x_y\right)+𝐯\left(+p\right)=0,$$
(70)
where
$$\frac{\overline{\rho }_b𝐮_{}^{}{}_{}{}^{2}}{2}+\frac{g\alpha _p}{_zT_b}\frac{\theta ^2}{2}.$$
Using condition (7) once more, we may integrate (70) over the full spatial domain to find,
$$\frac{dE}{dt}=_𝐒(+p)𝐮^{}\widehat{𝐧}𝑑S,_𝐕qu^{}v^{}𝑑V$$
(71)
with
$$E_𝐕\left(\frac{\overline{\rho }_b𝐮_{}^{}{}_{}{}^{2}}{2}+\frac{g\alpha _p}{_zT_b}\frac{\theta ^2}{2}\right)𝑑V,$$
in which $`𝐕`$ and $`𝐒`$ is the volume and surface-boundary of the domain in which $`\widehat{𝐧}`$ is the unit normal of the surface. The above result is true in general for both linear and nonlinear perturbations. We interpret the quantities in $``$ in the following way: the term $`\overline{\rho }_b𝐮_{}^{}{}_{}{}^{2}/2`$ represents the kinetic energy in the disturbances while the term $`g\alpha _p\theta ^2/2_zT_b`$ represents the energy in thermal processes. By definition $``$ is zero in steady state, while the steady pressure is constant, denoted by $`\overline{p}`$.
The global integral $`E`$, which we refer to as the total energy in disturbances, can change due to the influx of $``$ across the boundaries, through work done upon the system externally as represented by the boundary flux term $`_𝐒p𝐮^{}\widehat{𝐧}𝑑S`$, and finally due to the interaction with the background shear via the Reynolds stress term $`_𝐕qu^{}v^{}𝑑V`$ (for a discussion of this see Schmid & Henningson, 2002).
The general evaluation of (71) may proceed once boundary conditions are specified. As we have stated earlier, we will consider disturbances to be periodic in the $`y`$ and $`z`$ directions (sinuous or varicose for the latter). The radial boundary conditions and the motivation for their choices deserve some additional reflection. We remind the reader that one of the goals of this study is to assess whether or not the SRI is an intrinsic instability of the fluid and not some artifact of boundary conditions. One reasonable control is to require that there is neither work done on the system from outside nor there be any flux of energy across the bounding walls. This requirement requires that either the normal velocities are zero on either of the two walls or that (for linear disturbances only) the Lagrangian pressure perturbations are zero at the two bounding surfaces. A mixture of these can also affect the same outcome. That is to say, for example, one could require that the normal velocity at one bounding surface is zero while the Lagrangian pressure perturbation is zero at the other surface. Imposing these conditions therefore implies that $`E`$ can change only due to the interactions of the perturbations directly with the shear, in other words, all such disturbances behave according to
$$\frac{dE}{dt}=q_𝐕u^{}v^{}𝑑V.$$
(72)
In this sense, then, these solutions share some common property that allows for some comparison.
## Appendix B Development of the continuous spectrum mode
The discussion here largely follows the tack taken by Case (1960). Beginning with (35) we can say that
$$\left(_x^2F_e^2\beta ^2\right)P_0=0,$$
(73)
is true for $`xx_c`$ where
$$x_c=\frac{\omega _1}{q\mathrm{\Omega }_0\alpha _1}.$$
Taking $`P_0^{}`$ to denote the solution of (73) to the left and right (respectively) of $`x_c`$, and, assuming zero pressure conditions at $`x=0,1`$ we have that
$$P_0^{}=A^{}\mathrm{sinh}\left[k__Fx\right],P_0^+=A^+\mathrm{sinh}\left[k__F(1x)\right]$$
(74)
Then, to enforce continuity of the pressure at $`x=x_c`$ we see that
$$A^{}=A^+\frac{\mathrm{sinh}\left[k__F(1x_c)\right]}{\mathrm{sinh}\left[k__Fx_c\right]}.$$
Once some normalization is specified, that is, a value of $`A^+`$ is assumed, the solution is complete. For numerically generated solutions, we set $`P(x=0.99)=0.01`$. In summary, then, we have the continuous mode pressure eigenfunction, $`P_0^{(c)}`$, is
$$P_0^{(c)}=A^+\{\begin{array}{cc}\frac{\mathrm{sinh}\left[k__F(1x_c)\right]}{\mathrm{sinh}\left[k__Fx_c\right]}\mathrm{sinh}\left[k__Fx\right]\hfill & \hfill 0<x<x_c,\\ \mathrm{sinh}\left[k__F(1x)\right]\hfill & \hfill x_c<x<1.\end{array},$$
(75)
in which $`A^+=0.01/\mathrm{sinh}[0.99\times k__F]`$. Note that it means that any value of $`\omega _1`$ which satisfies the requirement $`0<x_c<1`$ is an allowed solution. This is the nature of the continuous spectrum (Schmid & Henningson, 2001). Also, the mode associated with the continuous spectrum is not technically defined at $`x=x_c`$.
## Appendix C Subtlety in the $`K_\nu `$ oscillating mode solution
A closer analysis of (55) reveals that it is everywhere analytic and, therefore, an entire function along the real $`z`$ axis, despite the presence of a branch point at $`z=0`$ for the $`K_\nu `$ Bessel function. It means, therefore, that in order to express the nature of this function as you one passes through $`z=0`$ one must be very careful about the relative phases that are incurred by crossing the $`z=0`$ point. To be more specific, let us define
$$\zeta =\left|\frac{\kappa __F}{\stackrel{~}{F}_e}\right|^{1/2}z$$
and rewrite the solution (55) according to its composition of Modified Bessel Functions of the first kind,
$$\mathrm{\Pi }(z)\zeta ^{\frac{3}{2}}[_{_{\frac{3}{4}}}\left(\frac{1}{2}\zeta ^2\right)_{_{\frac{3}{4}}}\left(\frac{1}{2}\zeta ^2\right),]$$
which, according to the series representation of $`I__\nu `$ would be
$$=\zeta ^{\frac{3}{2}}\left[\zeta ^{\frac{3}{2}}\mathrm{{\rm Y}}(3/4,\zeta )\zeta ^{\frac{3}{2}}\mathrm{{\rm Y}}(3/4,\zeta )\right]$$
where
$$\mathrm{{\rm Y}}(\nu ,\zeta )=\left(\frac{1}{4}\right)^\nu \underset{k=0}{\overset{\mathrm{}}{}}\frac{\left(\frac{1}{16}\zeta ^4\right)^k}{k!\mathrm{\Gamma }(\nu +k+1)},$$
(Abramowitz & Stegun, 1972). The function $`\mathrm{{\rm Y}}`$ is symmetric with respect to the reflections $`\zeta \zeta `$. However, the pre-factors appearing in the above expressions imply that one must be very careful in interpreting the function as $`\zeta `$ crosses over zero. This is best illustrated by considering the behaviour of (55) for $`\zeta <0`$. Expressed initially without restriction of $`\zeta `$ (55) is,
$`=`$ $`\zeta ^3\mathrm{{\rm Y}}(3/4,\zeta )\mathrm{{\rm Y}}(3/4,\zeta )`$
Then if we restrict attention to $`\zeta <0`$ by defining the variable $`s=\zeta `$ and restricting attention to $`s>0`$ we see that the above becomes
$`=`$ $`s^3\mathrm{{\rm Y}}(3/4,s)\mathrm{{\rm Y}}(3/4,s)`$
$`=`$ $`s^{\frac{3}{2}}\left[s^{\frac{3}{2}}\mathrm{{\rm Y}}(3/4,s)s^{\frac{3}{2}}\mathrm{{\rm Y}}(3/4,s)\right]`$
$`=`$ $`s^{\frac{3}{2}}\left[_{_{\frac{3}{4}}}\left(\frac{1}{2}s^2\right)_{_{\frac{3}{4}}}\left(\frac{1}{2}s^2\right)\right]`$
$`=`$ $`s^{\frac{3}{2}}\left[2_{_{\frac{3}{4}}}\left(\frac{1}{2}s^2\right)+𝒦_{_{\frac{3}{4}}}\left(\frac{1}{2}s^2\right)\right]`$
rewriting the above we see that the solution $`\mathrm{\Pi }(z)`$ for $`z<0`$ becomes as it is expressed in the text (56).
## Appendix D QHSG Linearization of the LSB
We linearize (57-61). It now means that $`\rho ^{}`$ and $`p^{}`$ are the linearized density and pressure fluctuations. The resulting equations become
$`(_tq\mathrm{\Omega }_0x_y)\rho ^{}+_xm_u+_ym_v+_zm_w=0,`$ (76)
$`0=2\mathrm{\Omega }_0m_v_xp^{},`$ (77)
$`(_tq\mathrm{\Omega }_0x_y)m_v+(2q)\mathrm{\Omega }_0m_u=_yp^{},`$ (78)
$`0=_zp^{}\rho ^{}g(z),`$ (79)
$`(_tq\mathrm{\Omega }_0x_y)\rho _b\mathrm{\Sigma }^{}+m_w_zS_b=0,`$ (80)
where for the sake of compact notation we have introduced the perturbed momentum fluxes
$$m_u\rho _bu^{},m_v\rho _bv^{},m_w\rho _bw^{},$$
and the perturbed entropy
$$\mathrm{\Sigma }^{}=\frac{\gamma }{\rho _b}\left(\frac{p^{}}{c^2}\rho ^{}\right).$$
Operating on (78) with $`_x`$ and then making use of (76) reveals
$$(_tq\mathrm{\Omega }_0x_y)\left[_xm_v(2q)\mathrm{\Omega }_0\rho ^{}\right]=(2q)\mathrm{\Omega }_0_zm_w.$$
This is followed by multiplying (80) by $`\mathrm{\Omega }_0(2q)/_zS_b`$ and then operating on the result with $`_z`$. This gives
$$(_tq\mathrm{\Omega }_0x_y)\frac{}{z}\mathrm{\Omega }_0(2q)\frac{\rho _b\mathrm{\Sigma }^{}}{_zS_b}=(2q)\mathrm{\Omega }_0_zm_w.$$
Adding these two equations gives
$$(_tq\mathrm{\Omega }_0x_y)\left[_xm_v(2q)\mathrm{\Omega }_0\rho ^{}+\frac{}{z}\mathrm{\Omega }_0(2q)\frac{\rho _b\mathrm{\Sigma }^{}}{_zS_b}\right]=0.$$
Given the definition of $`\mathrm{\Sigma }^{}`$, the hydrostatic relationship (79), and the radial geostrophic balance (77) recovers (63).
## Appendix E A Second Integral Statement of the Boussinesq Equations
Following the steps in Section A one may generate a second energy integral. The dynamical equations (7-11) describe the evolution of the both the disturbance velocities, i.e. the velocity fluctuations over and above the steady state Keplerian velocity, and the temperature fluctuations. We denote the total velocity of disturbances in the frame of the shearing box as $`𝐔`$ and given to be
$$𝐔q\mathrm{\Omega }_0x\widehat{𝐲}+𝐮^{}=\{u^{},v^{}q\mathrm{\Omega }_0x,w^{}\}$$
(81)
As such the governing equations of motion (8-11) are more concisely written in vector form as
$`_t𝐔+𝐔𝐔`$ $`=`$ $`\frac{1}{\overline{\rho }_b}p`$ (82)
$`+`$ $`2\mathrm{\Omega }_0\widehat{𝐳}\times (𝐔+q\mathrm{\Omega }_0x\widehat{𝐲})+\frac{1}{\overline{\rho }_b}g\alpha _p\theta \widehat{𝐳}`$
$`_t\theta +𝐔\theta `$ $`=`$ $`w_zT_b`$ (83)
$`𝐔`$ $`=`$ $`0.`$ (84)
As we have posited $`_zT_b`$ and $`g`$ multiplicative factors of each other over the full spatial domain under consideration. With this assumption in hand one may (i) multiply (82) by $`\rho _b𝐯`$ , (ii) multiply (83) by $`\theta g\alpha _p/_zT_b`$ and (iii) adding the results of (i) and (ii) together to yield
$$_t+𝐔\left(+p\right)=0,$$
(85)
where
$$\frac{\overline{\rho }_b𝐔^2}{2}+\frac{g\alpha _p}{_zT_b}\frac{\theta ^2}{2}q\mathrm{\Omega }_0^2x^2.$$
With use of the incompressibility condition (84) we may integrate (85) over the full spatial domain to find,
$$\frac{d\mathrm{\Phi }}{dt}=_𝐒(+p)𝐔\widehat{𝐧}𝑑S,$$
(86)
with
$$\mathrm{\Phi }_𝐕𝑑V=_𝐕\left(\frac{\overline{\rho }_b𝐔^2}{2}+\frac{g\alpha _p}{_zT_b}\frac{\theta ^2}{2}q\mathrm{\Omega }_0^2x^2\right)𝑑V,$$
in which $`𝐕`$ and $`𝐒`$ are as they were defined before. We interpret the quantities in $``$ in the following way: the term $`\overline{\rho }_b𝐔^2/2`$ represents the kinetic energy, the term $`q\mathrm{\Omega }_0^2x^2`$ is like a potential energy and $`g\alpha _p\theta ^2/2_zT_b`$ represents the energy in thermal processes. The global integral $`\mathrm{\Phi }`$ can change due to the influx of $``$ across the dynamically undulating boundaries as well as through the work done upon the system externally as represented by the boundary flux term $`_𝐒p𝐔\widehat{𝐧}𝑑S`$.
The point of this exercise is to note that only for no-normal flow boundary conditions does $`\mathrm{\Phi }`$ remain fixed for disturbances. The other conditions, like fixing the Lagrangian pressure, can cause $`\mathrm{\Phi }`$ to vary over the course of its evolution. This is because although $`\overline{p}`$ may be constant in steady state, the total quantity $``$ is not constant in steady state where a simple inspection of its definition clearly reveals. By fixing only the Lagrangian pressure fluctuation, the otherwise moving boundary can allow $``$ to seep in and out of the domain.
Perhaps, then, the reason for the existence of the instability under no-normal flow boundary conditions arises because of this preserved property of the disturbances. The reflection property of the boundaries perhaps traps energy in a way that causes growth to be encouraged. In this case, the energy of the disturbances must come from the energy contained in the background shear state and because there is an overall trapping of the energy, a runaway extraction processes takes place - ironically, leaving the total energy budget , $`\mathrm{\Phi }`$, fixed over the course of the evolution.
|
warning/0506/cond-mat0506451.html
|
ar5iv
|
text
|
# Magnetic density of states at low energy in geometrically frustrated systems
## Abstract
Using muon-spin-relaxation measurements we show that the pyrochlore compound Gd<sub>2</sub>Ti<sub>2</sub>O<sub>7</sub>, in its magnetically orderered phase below $`1`$ K, displays persistent spin dynamics down to temperatures as low as 20 mK. The characteristics of the induced muon relaxation can be accounted for by a scattering process involving two magnetic excitations, with a density of states characterized by an upturn at low energy and a small gap depending linearly on the temperature. We propose that such a density of states is a generic feature of geometrically frustrated magnetic materials.
The study of geometrically frustrated materials (GFMs) is a subject at the forefront of research in condensed matter physics not only because of their own interest but also because the concept of the frustration of the interactions plays a role for understanding the physics of e.g. ice, cholesteric crystals and metallic glasses; see for instance Ref. Ramirez (2001) for a discussion.
Magnetic materials based on lattices with triangular motifs and nearest-neighbor antiferromagnetic exchange interaction belong to the family of GFMs. They are, in the absence of further terms in the expression of their energy, believed to remain disordered and fluctuating down to zero temperature Villain (1979). The absence of magnetic order stems from their highly degenerate ground state. It is only upon inclusion of perturbations such as exchange interactions extending beyond nearest-neighbor magnetic ions or dipole coupling that magnetic ordering may appear. Experiments on Kagomé, garnet and pyrochlore structure compounds support these general predictions Ramirez (2001).
A fingerprint for the geometrical frustration of a material is the shift towards low energy of the spectral weight of excitations Ramirez (2001). The first convincing experimental proof has been obtained in a Kagomé-like magnetic material, where the low temperature specific heat Ramirez et al. (2000) is found to be dominated by singlet excitations arising from correlated spins, rather than from individual spins. This enhanced density of spectral weight at low energy could be linked with the persistence of spin dynamics observed at low temperature in many systems, as magnetic excitations are continuously available from zero energy. This behavior has been evidenced in the Kagomé compounds SrCr<sub>9p</sub>Ga<sub>12-9p</sub>O<sub>19</sub> with $`0.39p0.89`$ Uemura et al. (1994); Keren et al. (2000), for which a spin-glass transition is detected at $`T_\mathrm{g}=4p`$ K, in Kagomé-like systems Fukaya et al. (2003); Bono et al. (2004), in the garnet Gd<sub>3</sub>Ga<sub>5</sub>O<sub>12</sub> Dunsiger et al. (2000); Marshall et al. (2002); Bonville et al. (2004a) with $`T_\mathrm{g}=0.15`$ K and in the pyrochlore compounds Y<sub>2</sub>Mo<sub>2</sub>O<sub>7</sub> ($`T_\mathrm{g}=22`$ K) Dunsiger et al. (1996), Tb<sub>2</sub>Mo<sub>2</sub>O<sub>7</sub> ($`T_\mathrm{g}=25`$ K) Dunsiger et al. (1996). It was also observed in the spin liquid Tb<sub>2</sub>Ti<sub>2</sub>O<sub>7</sub> Gardner et al. (1999, 2003); Keren et al. (2004), for which there is no evidence of a transition down to 0.07 K Gardner et al. (2001). The spin dynamics becomes approximately temperature independent below about $`T_\mathrm{g}`$ for the spin-glass systems, except for the two molybdates for which it occurs in the temperature range $`T/T_\mathrm{g}<`$ 0.05, and for Tb<sub>2</sub>Ti<sub>2</sub>O<sub>7</sub> which displays a temperature independent relaxation below about 1 K. Appreciable spin dynamics is also found in Yb<sub>2</sub>Ti<sub>2</sub>O<sub>7</sub> below the temperature at which the specific heat presents a sharp anomaly Hodges et al. (2002); Yaouanc et al. (2003); Gardner et al. (2004a). In compounds where the fluctuations have been studied by the muon spin relaxation ($`\mu `$SR) technique, the muon spin-lattice relaxation function is usually found to be a stretched exponential, i.e. $`P_Z^{\mathrm{sl}}(t)=\mathrm{exp}[(\lambda _Zt)^\alpha ]`$ with $`\alpha 1`$.
More unconventional is the spin dynamics recently observed by Mössbauer spectroscopy, down to 30 mK, in Gd<sub>2</sub>Sn<sub>2</sub>O<sub>7</sub> which shows long-range ordering below 1 K Bertin et al. (2002), and which was confirmed by $`\mu `$SR measurements down to 20 mK Bonville et al. (2004b); Dalmas de Réotier et al. (2004). This anomalous spin dynamics cannot arise from conventional magnons, since their population vanishes at low temperature. In order to further investigate these zero temperature fluctuations, we performed $`\mu `$SR measurements in the parent compound Gd<sub>2</sub>Ti<sub>2</sub>O<sub>7</sub>, for which single crystals are available. This compound undergoes a first magnetic transition at $`T_{\mathrm{c1}}`$ 1 K, followed by a second one at $`T_{\mathrm{c2}}`$ 0.75 K Ramirez et al. (2002). We report the results of specific heat and $`\mu `$SR measurements in this material, and propose to account for the observed persistent spin dynamics in GFMs in terms of an unconventional density of states at low energy.
A Gd<sub>2</sub>Ti<sub>2</sub>O<sub>7</sub> polycrystalline rod was prepared by mixing, heating and compacting the constituent oxides Gd<sub>2</sub>O<sub>3</sub> and TiO<sub>2</sub> of respective purity 5N and 4N5. A single crystal was then grown by the traveling solvent floating zone technique using a Crystal System Inc. optical furnace with a velocity of 8 mm per hour. Oriented platelets were cut from the crystal and subsequently annealed under oxygen pressure to ensure optimized physical properties. The top panel of Fig. 2 shows the result of specific heat measurements performed using a dynamic adiabatic technique. The two phase transitions are observed at $`T_{\mathrm{c1}}`$ = 1.02 K and $`T_{\mathrm{c2}}`$ = 0.74 K respectively in agreement with recent single crystal measurements Petrenko et al. (2004). The peak at $`T_{\mathrm{c2}}`$ is found much sharper than previously reported Ramirez et al. (2002); Bonville et al. (2003); Petrenko et al. (2004). Its shape suggests that the lower phase transition is first order. Although we do not have any experimental evidence, it is possible that at this lower magnetic transition corresponds a structural transition which could be induced through magneto-elastic coupling. Anyhow, the magnetic transitions do not relieve the frustration since in the following we do report the observation of persistent spin dynamics, a fingerprint of geometrical magnetic frustration. At low temperature, up to $`0.55`$ K, the specific heat divided by temperature $`C_\mathrm{p}/T`$ is proportional to $`T`$. Such a behavior was also found for Gd<sub>2</sub>Sn<sub>2</sub>O<sub>7</sub> Bonville et al. (2003) and Kagomé-like compounds Ramirez et al. (2000), but it is not a general rule for GFMs since a $`T^2`$ dependence is reported for Er<sub>2</sub>Ti<sub>2</sub>O<sub>7</sub> Champion et al. (2003). The entropy release reaches $``$ 90 % of $`R\mathrm{ln}8`$ only near 5 K.
The $`\mu `$SR measurements (see Ref. Dalmas de Réotier and Yaouanc (1997) for an introduction to this technique) were performed at the Low Temperature Facility ($`\pi `$M3 beamline) of the Swiss Muon Source (Paul Scherrer Institute, Villigen, Switzerland). As expected, the $`\mu `$SR spectra in the paramagnetic phase are well described by a single exponential relaxation function, i.e. they are proportional to $`P_Z^{\mathrm{exp}}(t)=\mathrm{exp}(\lambda _Zt)`$ where $`\lambda _Z`$ is the spin-lattice relaxation rate. $`Z`$ labels the direction of the initial muon beam polarization. Below $`T_{\mathrm{c1}}`$, the observed oscillations, due to spontaneous precession of the muon spin (see insert of Fig. 1), are a signature of the long-range order of the magnetic structure, consistent with the observation of magnetic Bragg reflections by neutron diffraction Champion et al. (2001); Stewart et al. (2004). Surprisingly, even at 20 mK not (a), a stretched exponential decay of the spin-lattice relaxation channel is observed, superposed on the damped wiggles (Fig. 1). This implies that excitations of the spin system are present, inducing spin-lattice relaxation of the muon levels. In Fig. 2 are displayed the thermal variations of the two parameters accounting for this channel, $`\lambda _Z`$ and the stretched exponent $`\alpha `$.
Approaching $`T_{\mathrm{c1}}`$ from above, $`\lambda _Z`$ increases, reflecting the slowing down of the paramagnetic fluctuations. As expected, it drops when crossing $`T_{\mathrm{c1}}`$, but it remains roughly temperature independent below 0.5 K with a value $``$ 1 $`\mu `$s<sup>-1</sup>. So the persistent spin dynamics known to exist in GFMs with no long-range magnetic order is also present in a magnetically ordered compound. Interestingly, $`\alpha `$ is not far from 1/2 for $`T<T_{\mathrm{c2}}`$ and jumps to $`3/4`$ for $`T_{\mathrm{c2}}<T<T_{\mathrm{c1}}`$. It is expected to be equal to 1 for a homogeneous spin system, as found in the paramagnetic state. A value $`\alpha =1/2`$ suggests that relaxation stems from only a small fraction $`c`$ of the spins McHenry et al. (1972), fluctuating with a correlation time $`\tau _c`$ and distributed at random in the lattice. These spins create a field distribution at the muon site which is known to have a squared Lorentzian shape Walker and Walstedt (1980), with a width $`\mathrm{\Delta }_{\mathrm{Lor}}`$. A recent neutron diffraction study of Gd<sub>2</sub>Ti<sub>2</sub>O<sub>7</sub> Stewart et al. (2004) has shown that $`1/4`$ of the Gd magnetic moments are only partially ordered in the low temperature phase, and we tentatively attribute the relaxation channel with $`\alpha `$ = 1/2 to a fraction of these spins. Our measurements at 0.1 K with a longitudinal field (not shown) yield $`\tau _c=0.7(2)`$ ns and the relationship $`\lambda _Z`$ = $`4\mathrm{\Delta }_{\mathrm{Lor}}^2\tau _c`$ yields a width $`\mathrm{\Delta }_{\mathrm{Lor}}=18(5)`$ mT. According to Uemura et al. Uemura et al. (1985), $`\mathrm{\Delta }_{\mathrm{Lor}}=\sqrt{\pi /2}c\mathrm{\Delta }_{\mathrm{max}}`$, where $`\mathrm{\Delta }_{\mathrm{max}}`$ is the field width if all the Gd moments were contributing to the muon relaxation. Using the scaling between $`\mathrm{\Delta }_{\mathrm{max}}`$ and the rare earth moment which was successfully used for Tb<sub>2</sub>Ti<sub>2</sub>O<sub>7</sub> Gardner et al. (1999) and Yb<sub>2</sub>Ti<sub>2</sub>O<sub>7</sub> Yaouanc et al. (2003) we deduce $`c`$ 10%. We note that a significant distribution in the spin fluctuation times has been observed, by the neutron spin echo technique, in Gd<sub>2</sub>Ti<sub>2</sub>O<sub>7</sub> Gardner et al. (2004b) and in Tb<sub>2</sub>Ti<sub>2</sub>O<sub>7</sub> Gardner et al. (2003). This backs our observation that only a fraction of the Gd<sup>3+</sup> ions relax the muon spin, whereas the remaindering part does not because it is characterized by fluctuation times either too short or too long.
Our specific heat and $`\mu `$SR measurements show therefore that, in the long range order phase of Gd<sub>2</sub>Ti<sub>2</sub>O<sub>7</sub>, magnetic excitations with a non-vanishing density at low energy are present. These excitations lead to a $`T^2`$ behaviour for the specific heat and to a muon spin lattice relaxation rate which is quasi-independent of temperature. In fact below 0.5 K a fit of $`\lambda _Z(T)`$ to a power law gives $`\lambda _Z(T)T^\beta `$ with $`\beta 1/3`$. This is a negligible temperature dependence relative to power laws observed for usual ferromagnets ($`\beta =2`$) and conventional antiferromagnets ($`\beta =5`$) Dalmas de Réotier et al. (2004). In the following, we aim at determining the density of magnetic excitations responsible for the observed thermal behaviours of $`C_p`$ and $`\lambda _Z`$.
First of all, assuming these excitations to obey Bose-Einstein statistics like conventional magnons, it is easy to derive that, if their density of states per volume unit $`g_m(ϵ)`$ is proportional to $`ϵ^q`$, then $`C_pT^{q+1}`$. Hence $`g_m(ϵ)ϵ`$ accounts for the $`T^2`$ behaviour of $`C_p`$. Second, for the muon relaxation rate, if one considers a direct process with a single excitation, then energy conservation with $`ϵ=ϵ_\mu 0`$ ($`ϵ_\mu =\mathrm{}\omega _\mu `$, $`\omega _\mu `$ being the muon angular frequency) leads to $`\lambda _ZT`$, which is in disagreement with the experimental data for Gd<sub>2</sub>Ti<sub>2</sub>O<sub>7</sub> and other frustrated systems mentioned in the introduction. The relaxation process to consider next involves a two-excitation scattering (Raman process). Within the harmonic approximation, energy conservation implies that only the component of the spin-spin correlation tensor parallel to the magnetization, $`\mathrm{\Lambda }^{}(𝐪,\omega _\mu =0)`$, is probed Dalmas de Réotier and Yaouanc (1995). Thus, $`\lambda _Z𝒞(𝐪)\mathrm{\Lambda }^{}(𝐪,\omega _\mu =0)\mathrm{d}^3𝐪`$, where $`\omega _\mu `$ is set to zero because it is vanishingly small, $`𝒞(𝐪)`$ accounts for the interaction of the muon spin with the lattice spins and the integral extends over the Brillouin zone. For antiferromagnets with two collinear sublattices, the muon relaxation rate due to a Raman magnon process has been derived assuming for simplicity no orientation dependence for $`ϵ(q)`$ and $`𝒞(q)`$:
$`\lambda _Z`$ $`=`$ $`{\displaystyle \frac{8(2\pi )^3𝒟\mathrm{}}{15}}{\displaystyle \frac{(B_e+B_a)^2}{(B_e+B_a)^2B_e^2}}`$ (1)
$`\times `$ $`{\displaystyle _\mathrm{\Delta }^{\mathrm{}}}n\left(ϵ/k_\mathrm{B}T\right)\left[n\left(ϵ/k_\mathrm{B}T\right)+1\right]g_m^2(ϵ)dϵ,`$ (2)
where $`𝒟`$ = $`(\mu _0/4\pi )^2\gamma _\mu ^2g^2\mu _B^2`$, $`n(ϵ/k_\mathrm{B}T)`$ is the Bose-Einstein occupation factor, and $`\mathrm{\Delta }`$ the energy gap of the excitations at zero energy. $`B_e`$ and $`B_a`$ are respectively the exchange and anisotropy fields; a mean-field estimate for $`B_e`$ is 10 T and, with reference to another Gd compound Yaouanc et al. (1996), we take $`B_a`$ = 0.2 T. Equation 2 contains two population factors standing for the creation and annihilation of an excitation, a density of states being associated with each of them. The expression given by Eq. 2 assumes a muon site of high symmetry, e.g. the octahedral or tetrahedral interstitial sites of a face centered cubic lattice. For a muon site of lower symmetry, as expected in the case of the pyrochlores, Eq. 2 is essentially modified by a multiplicative factor $`\eta `$ which depends on the actual site and is in the range between 1 and $``$ 10. We took $`\eta `$ = 7 in the calculation below.
As a first step, let us assume $`\lambda _Z`$ to be temperature independent. The density of states must be $`g_m(ϵ)=b_\mu ϵ^{1/2}`$ not (c). It must also be assumed that $`\mathrm{\Delta }=ak_\mathrm{B}T`$, i.e. the gap at zero energy is proportional to temperature. If $`a`$ is of order 1 or lower, it can be shown that $`\lambda _Zb_\mu ^2/a^2`$ and that $`g_m(ϵ)`$ is essentially probed for $`\mathrm{\Delta }ϵ3\mathrm{\Delta }`$. Therefore the $`ϵ^{1/2}`$ dependence for $`g_m(ϵ)`$ holds only in a very restricted energy interval. As stated above, the $`T^2`$ dependence of $`C_\mathrm{p}`$ implies that $`g_m(ϵ)`$ is linear with energy, and this must prevail for $`ϵ3\mathrm{\Delta }`$. Combining the two regimes, we obtain: $`g_m(ϵ)=b_\mu ϵ^{1/2}+b_{\mathrm{sh}}ϵ`$. The numerical calculations show that $`a0.1`$ is compatible with the $`C_\mathrm{p}/T`$ and $`\lambda _Z`$ data. The predictions of the model with $`a`$ = 0.1 are shown in Fig. 2. As expected the model predicts $`\lambda _Z`$ to be independent of $`T`$ at low temperature. Note that a further refinement of the model, namely a slight change in the value of the exponent in the former term of $`g_m(ϵ)`$ ($`0.50.4`$), allows to improve the fit of $`\lambda _Z(T)`$ at low $`T`$ for the specific case of Gd<sub>2</sub>Ti<sub>2</sub>O<sub>7</sub>. This change affects the details in the shape of $`g_m(ϵ)`$, but not its main features: the upturn at small energy and the existence of a gap proportionnal to the temperature.
The calculated $`C_\mathrm{p}/T`$ shown in Fig. 2 is $`T`$-linear as expected, but shows an upturn below about 40 mK, caused by the $`ϵ^{1/2}`$ dependence of the density of states. This signature could be searched for in very low temperature specific heat measurements performed on a sample enriched with Gd isotopes having zero nuclear moment, thus showing no nuclear Schottky anomaly. The computed curve for $`\lambda _Z`$ above $`0.4`$ K overestimates the measured values. This could indicate that a fraction of the density of states, in the region where the dependence $`g_m(ϵ)ϵ`$ prevails, arises from singlet-like states Ramirez et al. (2000) which do not contribute to the relaxation.
In summary we have shown that, in GFMs, a non-vanishing muon spin-lattice relaxation at low $`T`$, with a weak temperature dependence, if any, is the signature of a low energy upturn in the density of magnetic states. This density is characterized by a gap varying linearly with temperature, leading to an accumulation of states at low energy (see the upper insert of Fig. 2). This is a rare feature, which is also observed in BCS superconductors. We propose that the gap is due to the dipole interaction between magnetic moments, which breaks rotational invariance. However, its increase with $`T`$ is unexpected. It is usually temperature independent or, for superconductors, decreases as temperature is increased Anderson (1984). A linear thermal increase strongly suggests that the thermal energy exceeds the energy involved in the excitation scattering process. Indeed, an analytical classical calculation performed for a triangular planar model with nearest-neighbor antiferromagnetic interactions and dipole interactions predicts a gap proportional to $`T`$ with $`a=0.2`$ for one of the phases Rastelli et al. (2003).
As pointed out at the beginning of this Letter, persistent and weakly temperature spin dynamics has been observed for a large number of GFM, included magnetically ordered powder samples of Gd<sub>2</sub>Ti<sub>2</sub>O<sub>7</sub> Yaouanc et al. (2003), Gd<sub>2</sub>Sn<sub>2</sub>O<sub>7</sub> Bonville et al. (2004a); Dalmas de Réotier et al. (2004) and Er<sub>2</sub>Ti<sub>2</sub>O<sub>7</sub> Lago et al. (2005). A combined analysis of $`C_p(T)`$ and $`\lambda _Z(T)`$ is always possible, resulting in $`g_m`$ with a gap linear in temperature.
The possibility for a GFM to order magnetically at finite temperature, although it is predicted to remain disordered down to zero temperature, was discovered theoretically a long time ago Villain et al. (1980), as “order from thermal disorder”. The gap we infer here is a consequence of the same mechanism. It appears if rotational invariance is broken. This naturally occurs because of the presence of the dipole interaction.
We thank B. Canals, G. Jackeli, S.V. Maleyev for enlightning discussions. Part of this work was performed at the S$`\mu `$S, Paul Scherrer Institute, Villigen, Switzerland.
|
warning/0506/math0506332.html
|
ar5iv
|
text
|
# Profinite Etale Cobordism
## 1 Introduction
### 1.1 Motivation and main results
The aim of this paper is the construction of a new cohomology theory for smooth schemes over a field, called profinite étale cobordism.
Since the 1990s two approaches to the theory of algebraic cobordism for smooth schemes have been made. On the one hand, Morel and Voevodsky have developed $`𝔸^1`$-homotopy theory for schemes. Voevodsky used it for the construction of cohomology theories on schemes. In particular, he showed the existence of an algebraic cobordism theory $`MGL^,`$ in analogy to Thom’s homotopical definition of complex cobordism in topology.
On the other hand, Levine and Morel used Quillen’s insight for a geometric construction of complex cobordism for proving that there is another possible definition of algebraic cobordism via a purely geometric construction. They proved that this $`\mathrm{\Omega }^{}`$ is the universal object for oriented cohomology theories on smooth schemes over a field. It is conjectured that $`\mathrm{\Omega }^{}`$ is in fact a geometrical description of the $`MGL^{2,}`$-part of Voevodsky’s theory. Hopkins and Morel recently proved that the canonical map $`\mathrm{\Omega }^{}MGL^{2,}`$ is surjective over a field of characteristic zero.
Both approaches have been used to prove famous results. Voevodsky used his construction for the proof of the Milnor Conjecture. Levine and Morel proved Rost’s Degree Formula. Nevertheless, both theories are still hard to compute and most of their features have only been proved over fields of characteristic zero.
The purpose of this paper is the construction of an étale topological version of cobordism for smooth schemes that is easier to compute since it is closely related to the étale cohomology. The idea is due to Eric Friedlander who constructed in \[Fr1\] a first version of an étale topological K-theory for schemes that turned out to be a powerful tool for the study of algebraic K-theory with finite coefficients. In particular, Thomason proved in his famous paper \[Th\] that algebraic K-theory with finite coefficients agrees with étale K-theory after inverting a Bott element. The aim of this paper is also to construct a candidate for an analogous statement for algebraic cobordism, see Conjecture 1.4.
At the end of the 1970s, in \[Sn\] Victor P. Snaith has already constructed a $`p`$-adic cobordism theory for schemes. His approach is close to the definition algebraic K-theory by Quillen. He defines for every scheme $`V`$ over $`𝔽_q`$, $`q=p^n`$, a topological cobordism spectrum $`\underset{¯}{A𝔽}_{q,V}`$. The homotopy groups of this spectrum are the $`p`$-adic cobordism groups of $`V`$. He has calculated these groups for projective bundles, Severi-Brauer schemes and other examples.
In this paper, we do not follow Snaith’s construction, but we provide Friedlander’s idea with a general setting. We consider the profinite completion $`\widehat{\mathrm{Et}}`$ of Friedlander’s étale topological type functor of \[Fr2\] with values in the category of simplicial profinite sets $`\widehat{𝒮}`$. We construct a stable homotopy category $`\widehat{𝒮}`$ for $`\widehat{𝒮}`$ and define general cohomology theories for profinite spaces. We apply these cohomology theories to $`\widehat{\mathrm{Et}}X`$ for a scheme $`X`$ over a field $`k`$. This yields a general foundation for different étale topological cohomology theories.
When we apply this construction to the cohomology theory represented by the profinitely completed $`\widehat{MU}`$-spectrum of complex cobordism, we get an étale topological cobordism theory $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}`$ for schemes of finite type over a field, which we call profinite étale cobordism. Note that this theory depends on a fixed prime number $`\mathrm{}`$, which must be different from the characteristic of the base field $`k`$.
The main feature of $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}`$ is the existence of an Atiyah-Hirzebruch spectral sequence, see Proposition 7.15 and Theorem 7.16:
###### Theorem 1.1
1. Let $`k`$ be a separably closed field. There are isomorphisms
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k)MU^{}_{}_{\mathrm{}}\mathrm{and}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k;/\mathrm{}^\nu )MU^{}_{}/\mathrm{}^\nu .$$
2. Let $`k`$ be a field. For every smooth scheme $`X`$ over k, there is a convergent spectral sequence $`\{E_r^{p,q}\}`$ with
$$E_2^{p,q}=H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^p(X;/\mathrm{}^\nu MU^q)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{p+q}(X;/\mathrm{}^\nu ).$$
These properties gives rise to several applications.
As a first corollary, we deduce from comparison results for étale cohomology that over $`k=`$ profinite étale and complex cobordism with finite coefficients are naturally isomorphic, see Theorem 7.18.
On the one hand, based on the results of Panin \[Pa\], they enable us to prove, see Theorem 7.29:
###### Theorem 1.2
Let $`k`$ be a separably closed field. Profinite étale cobordism $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}()`$, resp. $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(;/\mathrm{}^\nu )`$, is an oriented cohomology theory on $`\mathrm{Sm}/k`$
This implies that we have a unique natural morphism $`\theta _{\widehat{MU}}:\mathrm{\Omega }^{}(X)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(X)`$ for every $`X`$ in $`\mathrm{Sm}/k`$. On the other hand, via a stable étale realization functor of motivic spectra, we construct a natural map $`\varphi :MGL^,(X)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X)`$, see Section 8.2, Theorem 8.10 and Corollary LABEL:MGLsurjection:
###### Theorem 1.3
This natural map $`\varphi `$ fits into a commutative triangle for every $`X`$ in $`\mathrm{Sm}/k`$
$$\begin{array}{ccc}\mathrm{\Omega }^{}(X)& \stackrel{\theta _{MGL}}{}& MGL^{2,}(X)\\ \theta _{\widehat{MU}}& & \varphi \\ & \widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(X).& \end{array}$$
There is a similar triangle for $`/\mathrm{}^\nu `$-coefficients.
For a field $`k`$, we start the study of the absolute Galois group $`G_k=\mathrm{Gal}(k_s/k)`$ on étale cobordism, where $`k_s`$ denotes a separable closure of $`k`$. We deduce from the previous results that the action of $`G_k`$ on $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k_s)`$, resp. $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k_s;/\mathrm{}^\nu )`$, is trivial, see Theorem 9.3. Together with the Atiyah-Hirzebruch spectral sequence applied to Galois cohomology, this enables us to determine the étale cobordism of a finite field $`k=𝔽_q`$, $`\mathrm{}q`$, see Theorem LABEL:hMUofFq.
Finally, the Atiyah-Hirzebruch spectral sequence together with the results of Levine \[Le2\] on motivic cohomology with inverted Bott element yield good reasons for the following conjecture, see Section 8.4 and Conjecture 8.12:
###### Conjecture 1.4
Let $`X`$ be a smooth scheme of finite type over a separably closed field $`k`$ of characteristic different from $`\mathrm{}`$. Suppose that $`\mathrm{}`$ is odd or that $`\mathrm{}^\nu 4`$. Let $`\beta MGL^{0,1}(k;/\mathrm{}^\nu )`$ be the Bott element. The induced morphism
$$\varphi :MGL^,(X;/\mathrm{}^\nu )[\beta ^1]\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )$$
is an isomorphism.
At the end of Section 8, we will explain a strategy to prove this conjecture, which is close to Levine’s new proof of Thomason’s K-theory theorem in \[Le1\]. In particular, we would like to use the Atiyah-Hirzebruch spectral sequence of Hopkins and Morel for algebraic cobordism in \[HM\].
### 1.2 Survey of the paper
The outline of the paper is as follows. Towards the construction of profinite étale cobordism, there are several technical problems to solve. The first main result of this paper is that it is possible to establish a stable model structure on profinite spectra. It is based on Morel’s $`/\mathrm{}`$-cohomological model structure on $`\widehat{𝒮}`$ of \[Mo2\], see Theorem 3.2:
###### Theorem 1.5
There is a stable model structure on profinite spectra $`\mathrm{Sp}(\widehat{𝒮}_{})`$ such that the suspension functor $`S^1`$ is an equivalence on the corresponding homotopy category $`\widehat{𝒮}`$.
This result cannot be deduced in the same way as the stable structure on simplicial spectra, since $`\widehat{𝒮}`$ is not proper. We use a generalized theorem on left Bousfield localization of model categories similar to a result of Hirschhorn \[Hi\] and a generalized result on stable model structure on spectra similar to the work of Hovey in \[Ho2\]. We have discussed these theorems in the appendix.
The next step is the study of generalized cohomology theories on profinite spaces. Our knowledge on profinite completion of spectra and the results of Bousfield-Kan on $`\mathrm{}`$-completion of spaces enables us to determine the coefficients of $`\widehat{MU}`$. We also construct an Atiyah-Hirzebruch spectral sequence for profinite spaces. The proof of the existence of this spectral sequence is mainly due to Dehon in \[De\]. Profinitely completed cobordism has already been used for other purposes. In particular, Francois-Xavier Dehon uses the completed cobordism theory $`\widehat{MU}^{}()`$ for the study of Lannes’ $`T`$-functor in \[De\]. Many of his ideas were fruitful for this paper.
With the stable category on $`\widehat{𝒮}`$ we give this approach a general setting. Hence the discussions of the first sections may already be interesting on their own.
In order to clarify the argument and the relevance of the discussions of Sections 2-4 for the reader, we mention the following point.
Since our definitions are all compatible with étale cohomology in a certain sense, see Remark 5.3, we can deduce from these facts about profinite cohomology theories the existence of the mentioned Atiyah-Hirzebruch spectral sequence starting from étale cohomology and converging to profinite étale cobordism. Furthermore, we can determine the coefficients $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k)`$ for a separably closed field $`k`$.
In Sections 5 and 6, we dedicate our attention to the study of the completed étale topological type functor $`\widehat{\mathrm{Et}}`$ on schemes of finite type over a field with values in $`\widehat{𝒮}`$. This functor is due to Artin-Mazur and Friedlander. It has been extended to the $`𝔸^1`$-homotopy category of schemes independently by Daniel Isaksen \[I3\] and Alexander Schmidt \[Sch\].
We use this functor to construct a stable étale realization of motivic spectra, see Section 6 and Theorem 6.5:
###### Theorem 1.6
The functor $`\widehat{\mathrm{Et}}`$ yields an étale realization of the stable motivic homotopy category of $`^1`$-spectra:
$$\mathrm{L}\widehat{\mathrm{Et}}:𝒮^^1(k)\widehat{𝒮}.$$
This is another main result of the paper and may be interesting on its own. Although we do not know if $`\widehat{\mathrm{Et}}MGL`$ is isomorphic to $`\widehat{MU}`$ in $`\widehat{𝒮}`$ over an arbitrary base field, this realization is the key ingredient in the construction of the map $`\varphi :MGL^,(X)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X)`$ from algebraic to profinite étale cobordism.
In Sections 7 and 8 we start the study of étale cohomology theories and the comparison with algebraic cobordism. These two sections contain the study of profinite étale cobordism and give the proofs for the main results that we mentioned above.
The reader could even start with Section 7 if one accepts first the existence of a stable category of profinite spectra, second the existence of the functor $`\widehat{\mathrm{Et}}`$ on schemes and motivic spectra and third the existence of an Atiyah-Hirzebruch spectral sequence for profinite generalized cohomology theories and the isomorphism $`\widehat{MU}^{}MU^{}_{}_{\mathrm{}}`$. For the last fact, we even give another proof in Section 7.
The purpose of the appendix is to prove a general stabilization theorem for left proper fibrantly generated model categories, see Theorem B.11:
###### Theorem 1.7
Let $`𝒞`$ be a left proper fibrantly generated simplicial model category with all small limits and let $`T`$ be a left adjoint endofunctor on $`𝒞`$. There is a model structure on the category of spectra $`\mathrm{Sp}(𝒞,T)`$ such that $`T`$ becomes a Quillen equivalence on $`\mathrm{Sp}(𝒞,T)`$.
This model structure is again a left proper fibrantly generated simplicial model structure.
We apply this theorem in Section 3 to the left proper fibrantly generated model category of profinite spaces. This stabilization is the only purpose of the appendix. But since we gave the proofs in a very general setting, we have postponed them to the end of the paper.
In Appendix 1, we prove first a theorem on left Bousfield localization of fibrantly generated model categories, based on \[Hi\].
In Appendix 2, this theorem is applied to an intermediate model structure on spectra in order to get a stable category, based on \[Ho2\]. The ideas for the proofs are due to the papers \[Ho2\] of Hovey and \[Hi\] of Hirschhorn plus an idea of Christensen-Isaksen in \[CI\]. Nevertheless, our results are slightly more general and might be of interest for other situations.
I would like to use this opportunity to thank first of all Prof. Dr. Christopher Deninger for the suggestion of the topic of this thesis which satisfied all my interests and gave me the opportunity to learn a lot. He helped me with important suggestions and gave me the motivating impression to work on something that captured his interest. I would like to thank Dr. Christian Serpé for many very helpful discussions and suggestions for further studies. I am grateful to Prof. Marc Levine, Prof. Fabien Morel and Prof. Dr. Alexander Schmidt, who answered all my questions and explained to me both the great picture and the little details. I thank Francois-Xavier Dehon for an answer to a specific problem on Künneth isomorphisms for cobordism with coefficients.
## 2 Profinite spaces
We introduce the category of (pointed) profinite spaces $`\widehat{𝒮}`$ (resp. $`\widehat{𝒮}_{}`$) and its model structure defined by Morel \[Mo2\] for a fixed prime number $`\mathrm{}`$. An important construction is the Bousfield-Kan-$`\mathrm{}`$-completion of profinite spaces, which is an explicit fibrant replacement functor in this model structure.
Via this fibrant replacement functor, we introduce homotopy groups of profinite spaces and study their behavior under the profinite completion functor $`𝒮\widehat{𝒮}`$. In particular, we show that these homotopy groups are always pro-$`\mathrm{}`$-groups.
Then we define a completion functor $`\mathrm{pro}𝒮\widehat{𝒮}`$ from the category of pro-simplicial sets to profinite spaces and show that the respective $`/\mathrm{}`$-homotopy categories are equivalent. This compares our approach to the one of \[I4\].
We finish this section with a technical statement. The model structure of $`\widehat{𝒮}`$ is fibrantly generated. This is necessary in order to apply the localization and stabilization theorems of the appendix to $`\widehat{𝒮}`$. At this point, the category $`\widehat{𝒮}`$ is less well behaved than $`𝒮`$ and $`\mathrm{pro}𝒮`$, but it has properties that are good enough for our purposes.
### 2.1 The profinite $`\mathrm{}`$-completion of Bousfield-Kan
Let $``$ denote the category of sets and let $``$ be the full subcategory of $``$ whose objects are finite sets. Let $`\widehat{}`$ be the category of compact and totally disconnected topological spaces. We may identify $``$ with a full subcategory of $`\widehat{}`$ in the obvious way. The limit functor $`lim:\mathrm{pro}\widehat{}`$, which sends a pro-object $`X`$ of $``$ to the limit in $`\widehat{}`$ of the diagram corresponding to $`X`$, is an equivalence of categories.
We denote by $`𝒮`$ (resp. $`s`$, resp. $`\widehat{𝒮}`$) the category of simplicial sets (resp. simplicial finite sets, resp. simplicial profinite sets). The objects of $`𝒮`$ (resp. $`\widehat{𝒮}`$) will be called spaces (resp. profinite spaces). There is a subtle distinction between the two categories pro-$`s`$ and $`\widehat{𝒮}`$. The obvious functor between them is not an equivalence. See \[I2\] for more details and a counter example.
For a profinite space $`X`$ we define the ordered set $`𝒬(X)`$ of simplicial open equivalence relations on $`X`$. For every element $`Q`$ of $`𝒬(X)`$ the quotient $`X/Q`$ is a simplicial finite set and the map $`XX/Q`$ is a map of profinite spaces. In fact, when we consider the limit $`lim_{Q𝒬(X)}X/Q`$ in $`\widehat{𝒮}`$, the map $`Xlim_{Q𝒬(X)}X/Q`$ is an isomorphism, cf. \[Qu2\] , Lemma 2.3.
The forgetful functor $`\widehat{}`$ admits a left adjoint $`\widehat{()}:\widehat{}`$. It induces dimensionwisely a functor $`\widehat{()}:𝒮\widehat{𝒮}`$, which we call profinite completion. It is left adjoint to the forgetful functor $`||:\widehat{𝒮}𝒮`$ which sends a profinite space $`X`$ to its underlying simplicial set $`|X|`$. By adjunction, the profinite completion of a simplicial set $`Z`$ may be identified with the filtered colimit in $`\widehat{𝒮}`$ of the simplicial finite subsets of $`Z`$.
Let $`X`$ be a profinite space. The continuous cohomology $`H^{}(X;\pi )`$ of $`X`$ with coefficients in the profinite abelian group $`\pi `$ is defined as the cohomology of the complex $`C^{}(X;\pi )`$ of continuous cochains of $`X`$ with values in $`\pi `$, i.e. $`C^n(X;\pi )`$ denotes the set $`\mathrm{Hom}_\widehat{}(X_n,\pi )`$ of continuous maps $`X_n\pi `$.
###### Remark 2.1
(\[Mo2\], §1.2.)
1. Let $`X`$ be a simplicial profinite set. If we denote by $`C_{}(X)`$ the chain complex defined in degree $`n`$ to be the free profinite abelian group on the profinite set $`X_n`$, then the universal property of the free profinite group yields for every profinite abelian group $`\pi `$ an isomorphism $`C^n(X)=\mathrm{Hom}_\widehat{}(X_n,\pi )\mathrm{Hom}(C_n(X),\pi )`$ for each $`n`$ where the last $`\mathrm{Hom}`$ denotes the morphisms of profinite abelian groups. This shows the relation to the classical definition of the cohomology of simplicial sets, cf. \[Ma1\] p. 5.
2. Let $`\pi `$ be a finite abelian group. There is a natural isomorphism between the continuous cohomology with coefficients in $`\pi `$ of the profinite completion $`\widehat{Y}`$ of a simplicial set $`Y`$ and its ordinary cohomology with coefficients in $`\pi `$, i.e.
$$H^{}(Y;\pi )H^{}(\widehat{Y};\pi ).$$
This isomorphism exists already on the level of complexes, since there is a natural bijection $`\mathrm{Hom}_\widehat{}(\widehat{X}_n,\pi )\mathrm{Hom}_{}(X_n,\pi )`$. We will explain later how to generalize this result to an arbitrary abelian pro-$`\mathrm{}`$-group $`\pi `$.
3. Let $`\pi `$ be a finite abelian group and let $`\{X_s\}`$ be a cofiltered diagram of profinite spaces. The canonical map
$$\mathrm{colim}_sH^{}(X_s;\pi )\stackrel{}{}H^{}(\underset{s}{lim}X_s;\pi )$$
is an isomorphism, cf. \[De\], Lemme 1.1.1.
4. Let $`\pi `$ be a finite abelian group, the complex $`C^{}(X;\pi )`$ is naturally identified with the filtered colimit of complexes $`\mathrm{colim}_QC^{}(X/Q;\pi )`$, with $`Q`$ running through $`𝒬(X)`$, and the cohomology $`H^{}(X;\pi )`$ is naturally isomorphic to the colimit of the cohomologies $`H^{}(X/Q;\pi )`$.
5. There is a Künneth formula, i.e. for two profinite spaces $`X`$ and $`Y`$ the map $`H^{}(X;/\mathrm{})H^{}(Y;/\mathrm{})H^{}(X\times Y;/\mathrm{})`$ is an isomorphism, cf. \[Mo2\].
Let $`\mathrm{}`$ be a fixed prime number. Fabien Morel has shown in \[Mo2\] that the category $`\widehat{𝒮}`$ can be given the structure of a closed model category in the sense of \[Qu1\]. The weak equivalences are the maps inducing isomorphisms in continuous cohomology with coefficients $`/\mathrm{}`$; the cofibrations are the degreewise monomorphisms and the fibrations are the maps that have the right lifting property with respect to the cofibrations that are also weak equivalences. One should note that there are functorial factorizations for any map in $`\widehat{𝒮}`$ into a trivial cofibration followed by a fibration, respectively into a cofibration followed by a trivial fibration. For example, we get a functorial factorization by fixing the construction of the proof of Proposition 1 on page 355 of \[Mo2\] on the existence of factorizations. All the constructions done there are functorial. Furthermore, all small limits and products exist in $`\widehat{𝒮}`$.
Morel has also given an explicit construction of fibrant replacements in $`\widehat{𝒮}`$, cf. \[Mo2\], 2.1. It is based on the $`/\mathrm{}`$-completion functor of \[BK\]. Let $`Y`$ be a simplicial set. We denote by $`\mathrm{Res}^{}Y`$ its cosimplicial $`/\mathrm{}`$-resolution, by $`\mathrm{Tot}_s(\mathrm{Res}^{}Y)`$ its $`s`$-th total space and by $`P^t\mathrm{Tot}_s\mathrm{Res}^{}Y`$ its $`t`$-th Postnikov decomposition, cf. \[BK\], part I. If $`Y`$ is a finite set in each degree, then $`\mathrm{Tot}_s\mathrm{Res}^{}Y`$ is also finite in each degree for all $`s0`$. Hence the total space $`\mathrm{TotRes}^{}Y`$, i.e. the $`\mathrm{}`$-completion of Bousfield-Kan of $`Y`$, which is the limit of the tower $`\mathrm{Tot}_s\mathrm{Res}^{}Y`$, has a natural structure of a profinite space, which we denote by $`\widehat{Y}^{\mathrm{}}`$. Now let $`X`$ be a profinite space. We denote by $`\widehat{X}^{\mathrm{}}`$ the limit in $`\widehat{𝒮}`$ of the $`\widehat{X/Q}^{\mathrm{}}`$, $`Q`$ running through $`𝒬(X)`$. We call this space the $`\mathrm{}`$-completion of Bousfield-Kan of the profinite space $`X`$. For a simplicial set $`Y`$ we denote by $`\widehat{Y}^{\mathrm{}}\widehat{𝒮}`$ its profinite $`\mathrm{}`$-completion defined as the limit $`lim_{Q,s,t}P^t\mathrm{Tot}_s\mathrm{Res}^{}(X/Q)`$ in $`\widehat{𝒮}`$, where $`X/Q`$ are the simplicial finite quotients of $`X`$. Its underlying simplicial set is isomorphic in $`\mathrm{Ho}(𝒮)`$ to the $`\mathrm{}`$-completion in $`𝒮`$ of Sullivan \[Su\], which is the limit of the pro-Artin-Mazur-$`\mathrm{}`$-completion, cf. \[Mo1\].
Morel shows that the natural map $`\theta _X:X\widehat{X}^{\mathrm{}}`$ defines a functorial fibrant replacement in $`\widehat{𝒮}`$, \[Mo2\], 2.1, Prop. 2. The spaces $`P^t\mathrm{Tot}_s(\mathrm{Res}^{}X/Q)`$ are in fact $`\mathrm{}`$-finite spaces, i.e. fibrant simplicial finite sets whose homotopy groups for every choice of basepoint are finite $`\mathrm{}`$-groups, trivial except for a finite number of them. The profinite space $`\widehat{X}^{\mathrm{}}`$ may be identified with the filtered limit in $`Q`$, $`s`$ and $`t`$ of the $`\mathrm{}`$-spaces $`P^t\mathrm{Tot}_s(\mathrm{Res}^{}X/Q)`$. Hence $`\widehat{X}^{\mathrm{}}`$ is a pro-$`\mathrm{}`$-space in the terminology of \[Mo2\].
For the study of generalized cohomology theories on profinite spaces it will be crucial to check the compatibility of the $`\mathrm{}`$-completion with other constructions in $`\widehat{𝒮}`$, resp. $`\widehat{𝒮}_{}`$, such as products, smash products and function spaces.
### 2.2 Pointed profinite spaces and the simplicial structure
The category $`\widehat{𝒮}`$ has a pointed analogue $`\widehat{𝒮}_{}`$ whose objects are maps $`X`$ in $`\widehat{𝒮}`$, where $``$ denotes the constant simplicial set equal to a point. Its morphisms are maps in $`\widehat{𝒮}`$ that respect the basepoints. The forgetful functor $`\widehat{𝒮}_{}\widehat{𝒮}`$ has a left adjoint, which consists in adding a disjoint basepoint $`XX_+`$. The $`\mathrm{}`$-completion of a pointed profinite space is naturally pointed and $`XX_+`$ is compatible with $`\mathrm{}`$-completion. $`\widehat{𝒮}_{}`$ has the obvious induced closed model category structure.
The product for two pointed profinite spaces $`X`$ and $`Y`$ in $`\widehat{𝒮}_{}`$ is the smash product $`XY\widehat{𝒮}_{}`$, defined in the usual way as the quotient $`(X\times Y)/(XY)`$ in $`\widehat{𝒮}`$. For a pointed profinite space $`X`$ we define its reduced cohomology with coefficients in the profinite abelian group $`\pi `$, denoted $`\stackrel{~}{H}^{}(X;\pi )`$, to be the kernel of the induced morphism $`H^{}(X;\pi )H^{}(pt;\pi )`$. It is clear that $`\stackrel{~}{H}^{}(X_+;\pi )H^{}(X;\pi )`$. We also have a Künneth formula.
###### Lemma 2.2
For two pointed profinite spaces $`X`$ and $`Y`$, the canonical map $`\stackrel{~}{H}^{}(X;/\mathrm{})\stackrel{~}{H}^{}(Y;/\mathrm{})\stackrel{~}{H}^{}(XY;/\mathrm{})`$ is an isomorphism.
Proof The canonical map is an isomorphism in the case that $`X`$ and $`Y`$ are simplicial finite sets. Hence it is an isomorphism in general since the tensor product commutes with filtered colimits. $`\mathrm{}`$
###### Proposition 2.3
Let $`X`$ and $`Y`$ be simplicial sets whose $`/\mathrm{}`$-cohomology is finite in each degree. Then the map $`\widehat{X\times Y}^{\mathrm{}}\widehat{X}^{\mathrm{}}\times \widehat{Y}^{\mathrm{}}`$ is a weak equivalence in $`\widehat{𝒮}`$.
Proof We have $`\widehat{X\times Y}^{\mathrm{}}=lim_Q\widehat{(X\times Y)/Q}^{\mathrm{}}=lim_{Q_1,Q_2}\widehat{(X/Q_1\times Y/Q_2)}^{\mathrm{}}`$, which is due to the fact that the finite quotients $`(X\times Y)/Q`$ are in 1-1-correspondence with finite quotients $`X/Q_1\times Y/Q_2`$; and we have $`\widehat{X}^{\mathrm{}}\times \widehat{Y}^{\mathrm{}}=lim_{Q_1}\widehat{X/Q_1}^{\mathrm{}}\times lim_{Q_1}\widehat{Y/Q_2}^{\mathrm{}}`$. The hypothesis implies that $`X`$ and $`Y`$ are nilpotent spaces in the sense of \[BK\]. Thus the same is true for the finite quotients $`X/Q_1`$ and $`Y/Q_2`$. Then \[BK\] VI, §6 Proposition 6.5, says that
$$H^{}(\widehat{(X/Q_1\times Y/Q_2)}^{\mathrm{}};/\mathrm{})H^{}(\widehat{X/Q_1}^{\mathrm{}}\times \widehat{Y/Q_2}^{\mathrm{}};/\mathrm{})$$
is an isomorphism. We conclude by taking colimits and by Remark 2.1. $`\mathrm{}`$
###### Proposition 2.4
Let $`X`$ and $`Y`$ be pointed simplicial sets whose $`/\mathrm{}`$-cohomology is finite in each degree. Then the map $`\widehat{XY}^{\mathrm{}}\widehat{X}^{\mathrm{}}\widehat{Y}^{\mathrm{}}`$ is a weak equivalence in $`\widehat{𝒮}_{}`$.
Proof This follows as the previous proposition from \[BK\] VI, §6, Proposition 6.6. $`\mathrm{}`$
The categories $`\widehat{𝒮}`$ and $`\widehat{𝒮}_{}`$ have natural simplicial structures. There is a simplicial set $`\mathrm{Map}_{\widehat{𝒮}_{}}(X,Y)=\mathrm{Map}_{}(X,Y)`$ for all profinite spaces $`X`$ and $`Y`$. It is naturally pointed by the map $`XY`$. It is characterized by the bijection
$$\mathrm{Hom}_𝒮_{}(Z,\mathrm{Map}_{}(X,Y))\mathrm{Hom}_{\widehat{𝒮}_{}}(XZ,Y)$$
for every pointed simplicial finite set $`Z`$. The space $`\mathrm{Map}_{}(X,Y)`$ is given in the $`n`$-th dimension by the set $`\mathrm{Hom}_{\widehat{𝒮}_{}}(X\mathrm{\Delta }[n]_+,Y)`$, where $`\mathrm{\Delta }[n]_+`$ is the well-known pointed simplicial finite set. For a pointed simplicial finite set $`Z`$ the space $`\mathrm{Map}_{}(Z,Y)`$ has a natural structure of a profinite space considered as the filtered limit in $`\widehat{𝒮}_{}`$ of the simplicial finite sets $`\mathrm{Map}_{}(\mathrm{sk}_nZ,Y/Q)`$. We denote this profinite space by $`\mathrm{hom}_{}(Z,Y)\widehat{𝒮}_{}`$. For an arbitrary simplicial set $`Z`$ we define the profinite space $`\mathrm{hom}_{}(Z,Y)`$ by taking the filtered limit in $`\widehat{𝒮}_{}`$ of the profinite spaces $`\mathrm{hom}_{}(Z_s,Y)`$ over all simplicial finite subsets $`Z_s`$ of $`Z`$. This defines a right adjoint to the functor $`XZX`$, $`\widehat{𝒮}_{}\widehat{𝒮}_{}`$. We define the functor $`X:=X:𝒮_{}\widehat{𝒮}_{}`$ for a profinite space $`X`$ by taking the filtered limit in $`\widehat{𝒮}_{}`$ of the profinite spaces $`X/QZ_s`$ for all simplicial finite subsets $`Z_s`$ of the simplicial set $`Z`$ and all finite quotients $`X/Q`$ of $`X`$. These functors define a simplicial structure on $`\widehat{𝒮}_{}`$ and similarly on $`\widehat{𝒮}`$.
Let $`ZXY`$ be a diagram of profinite spaces and let $`XY`$ be a cofibration. We denote its colimit by $`Z_XY`$. On the level of continuous cochains we get that $`C^{}(X,/\mathrm{})`$ is the colimit of the induced diagram $`C^{}(Z,/\mathrm{})C^{}(Z_XY,/\mathrm{})C^{}(Y,/\mathrm{})`$. This yields a Mayer-Vietoris long exact sequence in cohomology
$$\mathrm{}H^n(Z_XY;/\mathrm{})H^nYH^nZH^nXH^{n+1}(Z_XY;/\mathrm{})\mathrm{}$$
Using this sequence Dehon shows that $`\widehat{𝒮}`$ and $`\widehat{𝒮}_{}`$ carry in fact a simplicial model structure, cf. \[De\] §1.4.
As an application, we define for $`X\widehat{𝒮}_{}`$ the profinite space $`\widehat{\mathrm{\Omega }}X\widehat{𝒮}_{}`$. Let $`S^1`$ be the simplicial finite set $`\mathrm{\Delta }[1]/\mathrm{\Delta }[1]`$. We set $`\widehat{\mathrm{\Omega }}X:=\mathrm{hom}_{}(S^1,X)`$. There is a natural isomorphism $`\mathrm{Hom}_{\widehat{𝒮}_{}}(S^1X,Y)\mathrm{Hom}_{\widehat{𝒮}_{}}(X,\widehat{\mathrm{\Omega }}Y))`$. In addition, $`\widehat{\mathrm{\Omega }}X`$ is fibrant if $`X`$ is fibrant.
This adjunction may be extended to the homotopy category $`\mathrm{Ho}(\widehat{𝒮}_{})`$ of $`\widehat{𝒮}_{}`$. For every profinite space $`X`$ and every fibrant profinite space $`Y`$, we have a natural bijection $`\mathrm{Hom}_{\mathrm{Ho}(\widehat{𝒮}_{})}(S^1X,Y)\mathrm{Hom}_{\mathrm{Ho}(\widehat{𝒮}_{})}(X,\widehat{\mathrm{\Omega }}Y))`$. Since $`S^1X`$ is a cogroup object and $`\widehat{\mathrm{\Omega }}X`$ a group object in $`\mathrm{Ho}(\widehat{𝒮}_{})`$, we conclude that the previous bijection is in fact an isomorphism of groups.
The following proposition has been proved by Francois-Xavier Dehon in \[De\] using its Eilenberg-Moore spectral sequence for profinite spaces. We give an alternative proof using the results of Bousfield and Kan \[BK\].
###### Proposition 2.5
Let $`X`$ be a simply connected simplicial set whose $`/\mathrm{}`$-cohomology is finite in every degree. Then the map $`\widehat{\mathrm{\Omega }X}^{\mathrm{}}\widehat{\mathrm{\Omega }}(\widehat{X}^{\mathrm{}})`$ is a weak equivalence in $`\widehat{𝒮}_{}`$. This map is in fact a homotopy equivalence.
Proof Since $`S^1`$ is a simplicial finite set, the finite quotients $`\widehat{X}/Q`$ of $`\widehat{X}`$ and $`\mathrm{hom}_{}(S^1,X)/Q`$ of $`\mathrm{hom}_{}(S^1,X)`$ are in 1-1 correspondence. Hence we may write
$$\widehat{\mathrm{\Omega }}(\widehat{X}^{\mathrm{}})=\mathrm{hom}_{}(S^1,\underset{Q}{lim}\widehat{X/Q}^{\mathrm{}})=\underset{Q}{lim}\mathrm{hom}_{}(S^1,\widehat{X/Q}^{\mathrm{}})=\underset{Q}{lim}\widehat{\mathrm{\Omega }}(\widehat{X/Q}^{\mathrm{}});$$
and
$$\widehat{\mathrm{\Omega }X}^{\mathrm{}}=\underset{Q}{lim}\widehat{(\mathrm{hom}_{}(S^1,X)/Q)}^{\mathrm{}}=\underset{Q}{lim}\widehat{(\widehat{\mathrm{\Omega }}(X/Q))}.$$
By our hypothesis $`X`$ is a nilpotent space in the sense of \[BK\] and hence $`X/Q`$ is nilpotent, too; by \[BK\] VI, §6, Prop. 6.5, we know that
$$H^{}(\widehat{(\widehat{\mathrm{\Omega }}(X/Q))}^{\mathrm{}};/\mathrm{})H^{}(\widehat{\mathrm{\Omega }}(\widehat{X/Q}^{\mathrm{}});/\mathrm{}).$$
Hence we get
$$\begin{array}{ccc}\hfill H^{}(\widehat{(\widehat{\mathrm{\Omega }}(X))}^{\mathrm{}};/\mathrm{})& & \mathrm{colim}_QH^{}(\widehat{(\widehat{\mathrm{\Omega }}(X/Q))}^{\mathrm{}});/\mathrm{})\hfill \\ & & \mathrm{colim}_QH^{}(\widehat{\mathrm{\Omega }}(\widehat{X/Q}^{\mathrm{}});/\mathrm{})H^{}(\widehat{\mathrm{\Omega }}(\widehat{X}^{\mathrm{}});/\mathrm{}),\hfill \end{array}$$
which is what we had to prove.
The last statement follows from the fact that the spaces in question are cofibrant and fibrant and the fact that weak equivalences between such objects are homotopy equivalences. $`\mathrm{}`$
### 2.3 Homotopy groups of profinite spaces
###### Definition 2.6
Let $`X\widehat{𝒮}`$ be a pointed profinite space and let $`\widehat{X}^{\mathrm{}}`$ be its $`\mathrm{}`$-completion. We denote by $`\pi _0X`$ the coequalizer in $`\widehat{}`$ of the diagram $`d_0,d_1:X_1\stackrel{}{}X_0`$. We define $`\pi _nX`$ for $`n1`$ to be the group $`\pi _n|\widehat{X}^{\mathrm{}}|`$ where $`|\widehat{X}^{\mathrm{}}|`$ is the underlying fibrant simplicial set of $`\widehat{X}^{\mathrm{}}`$. We call $`\pi _nX`$ the $`n`$-th homotopy group of $`X`$.
###### Proposition 2.7
A map $`f:XY`$ in $`\widehat{𝒮}_{}`$ is a weak equivalence if and only if $`\pi _{}(f)`$ is an isomorphism.
Proof The maps $`X\widehat{X}^{\mathrm{}}`$ and $`Y\widehat{Y}^{\mathrm{}}`$ are weak equivalences. Hence $`f`$ is a weak equivalence if and only if $`\widehat{f}^{\mathrm{}}`$ is a weak equivalence. Since all objects in $`\widehat{𝒮}_{}`$ are cofibrant, the map $`\widehat{f}^{\mathrm{}}`$ is a weak equivalence if and only if $`\widehat{f}^{\mathrm{}}`$ is a homotopy equivalence. This finishes the proof. $`\mathrm{}`$
We can already conclude from the definition $`\pi _nX=\mathrm{Hom}_{\mathrm{Ho}(𝒮_{})}(S^n,|\widehat{X}^{\mathrm{}}|)`$ that $`\pi _nX`$ has a natural structure of a profinite group. The following proposition shows that our definition coincides with the one given by Dehon in \[De\].
###### Proposition 2.8
1. For every profinite space $`X`$, the natural map
$$\mathrm{Hom}_{\mathrm{Ho}(\widehat{𝒮})}(,X)\pi _0X$$
is an isomorphism.
2. For every pointed profinite space $`X`$ and every $`n0`$ we have
(1)
$$\pi _n(\widehat{\mathrm{\Omega }}X)\pi _{n+1}(X).$$
In particular, we have $`\pi _nX\pi _0(\widehat{\mathrm{\Omega }}^nX)`$.
Proof The first statement is Proposition 1.3.4 a) of \[De\].
The second statement follows from the fact that $`S^n`$ is a simplicial finite set for all $`n1`$. For $`n=0`$, we use the first statement to conclude by adjunction
$$\begin{array}{ccc}\hfill \pi _0(\widehat{\mathrm{\Omega }}X)& & \mathrm{Hom}_{\mathrm{Ho}(\widehat{𝒮}_{})}(,\widehat{\mathrm{\Omega }}X)\mathrm{Hom}_{\mathrm{Ho}(\widehat{𝒮}_{})}(,\widehat{\mathrm{\Omega }}\widehat{X}^{\mathrm{}})\hfill \\ & & \mathrm{Hom}_{\mathrm{Ho}(\widehat{𝒮}_{})}(S^1,\widehat{X}^{\mathrm{}})\mathrm{Hom}_{\mathrm{Ho}(𝒮_{})}(S^1,|\widehat{X}^{\mathrm{}}|)\hfill \\ & =& \pi _1|\widehat{X}^{\mathrm{}}|=\pi _1X.\hfill \end{array}$$
For $`n1`$, we have the following sequence of isomorphisms using Proposition 2.5 and adjunction
$$\begin{array}{ccc}\hfill \pi _n(\widehat{\mathrm{\Omega }}X)& =& \pi _n(|\widehat{\widehat{\mathrm{\Omega }}X}^{\mathrm{}}|)=\mathrm{Hom}_{\mathrm{Ho}(𝒮_{})}(S^n,|\widehat{\widehat{\mathrm{\Omega }}X}^{\mathrm{}}|)\mathrm{Hom}_{\mathrm{Ho}(\widehat{𝒮}_{})}(\widehat{S^n},\widehat{\widehat{\mathrm{\Omega }}X}^{\mathrm{}})\hfill \\ & =& \mathrm{Hom}_{\mathrm{Ho}(\widehat{𝒮}_{})}(S^n,\widehat{\widehat{\mathrm{\Omega }}X}^{\mathrm{}})\mathrm{Hom}_{\mathrm{Ho}(\widehat{𝒮}_{})}(S^n,\widehat{\mathrm{\Omega }}(\widehat{X}^{\mathrm{}}))\hfill \\ & & \mathrm{Hom}_{\mathrm{Ho}(\widehat{𝒮}_{})}(S^{n+1},\widehat{X}^{\mathrm{}})\mathrm{Hom}_{\mathrm{Ho}(\widehat{𝒮}_{})}(S^{n+1},\widehat{X}^{\mathrm{}})\hfill \\ & & \mathrm{Hom}_{\mathrm{Ho}(𝒮_{})}(S^{n+1},|\widehat{X}^{\mathrm{}}|)\hfill \\ & =& \pi _{n+1}(X).\hfill \end{array}$$
$`\mathrm{}`$
As a corollary, we see that $`\pi _nX`$ has the natural structure of a pro-$`\mathrm{}`$-group for every $`n1`$, see also \[Mo1\], Cor. 1.4.1.3.
###### Proposition 2.9
For every pointed profinite space $`X`$ and every $`n0`$, we have an isomorphism $`\pi _nXlim_{Q,s,t}\pi _n(P^t\mathrm{Tot}_s(\mathrm{Res}^{}X/Q))`$.
In particular, for every $`n1`$ the group $`\pi _nX`$ has the structure of a pro-$`\mathrm{}`$-group.
Proof We know that there is an isomorphism $`\widehat{X}^{\mathrm{}}lim_{Q,s,t}P^t\mathrm{Tot}_s\mathrm{Res}^{}(X/Q)`$. Since $`\pi _0`$ commutes with limits and sends weak equivalences to homeomorphisms in $`\widehat{}`$, we conclude that we have an isomorphism
$$\pi _nX\underset{Q,s,t}{lim}\pi _n(P^t\mathrm{Tot}_s\mathrm{Res}^{}(X/Q))$$
for every $`n0`$. The description of the homotopy groups of the simplicial finite sets $`P^t\mathrm{Tot}_s\mathrm{Res}^{}(X/Q)`$ finishes the proof. $`\mathrm{}`$
We conclude this subsection with a collection of results on the homotopy groups of the $`\mathrm{}`$-completion of a simplicial set, see also \[BK\] VI, §5.
###### Proposition 2.10
Let $`X`$ be a pointed connected simplicial set.
1. We have an isomorphism
$$\pi _1(\widehat{X}^{\mathrm{}})\widehat{\pi _1(X)}^{\mathrm{}}.$$
In particular, if $`\pi _1X`$ is a finitely generated abelian group, then $`\pi _1(\widehat{X}^{\mathrm{}})_{\mathrm{}}_{}\pi _1X`$.
2. If $`X`$ is in addition simply connected, and its higher homotopy groups are finitely generated, then we have for all $`n2`$
$$\pi _n\widehat{X}^{\mathrm{}}\widehat{\pi _nX}^{\mathrm{}}_{\mathrm{}}_{}\pi _nX,$$
where $`_{\mathrm{}}`$ denotes the $`\mathrm{}`$-adic integers.
Proof We deduce this proposition from the methods of \[BK\]. The hypothesis on $`X`$ implies that $`X`$ is a nilpotent space in the sense of \[BK\], i.e. $`\pi _1X`$ acts nilpotently on all $`\pi _iX`$. We denote by $`/\mathrm{}_{\mathrm{}}X`$ the Bousfield-Kan $`/\mathrm{}`$-completion in $`𝒮`$ of $`X`$. The assumptions imply that $`X`$ is a $`/\mathrm{}`$-good space, i.e. $`H^{}(X;/\mathrm{})H^{}(/\mathrm{}_{\mathrm{}}X;/\mathrm{})`$, cf. \[BK\] VI, §5, Proposition 5.3. Considering the isomorphism $`H^{}(X;/\mathrm{})H^{}(\widehat{X};/\mathrm{})H^{}(|\widehat{X}^{\mathrm{}}|;/\mathrm{})`$, this implies that $`|\widehat{X}^{\mathrm{}}|`$ is $`/\mathrm{}`$-complete by \[BK\] V, §3, Proposition 3.3. Then Proposition 5.4 of \[BK\], VI, §5, says that the two $`\mathrm{}`$-completions $`|\widehat{X}^{\mathrm{}}|`$ and $`/\mathrm{}_{\mathrm{}}X`$ are weakly equivalent. Then \[BK\] VI, §5, Proposition 5.2 yields the result.
Alternatively, one could deduce the result from \[Su\], Theorem 3.1, and the fact that $`|\widehat{X}^{\mathrm{}}|`$ is isomorphic in $`\mathrm{Ho}(𝒮)`$ to the $`\mathrm{}`$-completion of Sullivan. $`\mathrm{}`$
###### Example 2.11
In the special case of the simplicial circle $`S^1`$, viewed as a profinite space, one has
$$\pi _1(S^1)=\pi _1(\widehat{S^1}^{\mathrm{}})\widehat{}^{\mathrm{}}=_{\mathrm{}}.$$
The last proposition is crucial for our studies of cohomology theories on $`\widehat{𝒮}`$, since it will allow us to compute the coefficients for some cohomology theories, e.g. for profinite cobordism $`\widehat{MU}`$.
### 2.4 Comparison with the category of pro-simplicial sets
Let $`\mathrm{pro}𝒮`$ be the category of pro-objects in $`𝒮`$. Its objects are cofiltered diagrams $`X():I𝒮`$ and its morphisms are defined by
$$\mathrm{Hom}_{\mathrm{pro}𝒮}(X(),Y()):=\underset{tJ}{lim}\mathrm{colim}_{sI}\mathrm{Hom}_𝒮(X(s),Y(t)),$$
see \[AM\] for more details. The cohomology with $`/\mathrm{}`$-coefficients is defined to be
$$H^{}(X(),/\mathrm{}):=\mathrm{colim}_sH^{}(X(s),/\mathrm{}).$$
Isaksen has constructed several model structures on $`\mathrm{pro}𝒮`$, cf. \[I1\] and \[I4\]. We are interested in the $`/\mathrm{}`$-cohomological model structure of \[I4\] in which the weak equivalences are morphisms inducing isomorphisms in $`/\mathrm{}`$-cohomology and the cofibrations are levelwise monomorphisms.
We want to compare $`\mathrm{pro}𝒮`$ with $`\widehat{𝒮}`$. These two categories are not equivalent, but we will see that completion induces an equivalence on the level of homotopy categories. We define a completion functor $`\widehat{()}:\mathrm{pro}𝒮\widehat{𝒮}`$ as the composite of two functors. First we apply $`\widehat{()}:𝒮\widehat{𝒮}`$ levelwise to get a functor $`\mathrm{pro}𝒮\mathrm{pro}\widehat{𝒮}`$. Then we take the limit in $`\widehat{𝒮}`$ of the underlying diagram. The next lemma shows that the functor $`\widehat{()}`$ has good properties with respect to the model structures on $`\mathrm{pro}𝒮`$ and $`\widehat{𝒮}`$.
###### Lemma 2.12
1. Let $`X\mathrm{pro}𝒮`$ be a pro-simplicial set. Then we have a natural isomorphism of cohomology groups
$$H^{}(X;/\mathrm{})\stackrel{}{}H^{}(\widehat{X};/\mathrm{}).$$
2. A morphism $`f:XY`$ of pro-simplicial sets induces an isomorphism in $`/\mathrm{}`$-cohomology if and only if the morphism $`\widehat{f}:\widehat{X}\widehat{Y}`$ in $`\widehat{𝒮}`$ induces an isomorphism in continuous $`/\mathrm{}`$-cohomology.
3. The functor $`\widehat{()}:\mathrm{pro}𝒮\widehat{𝒮}`$ preserves monomorphisms.
4. The functor $`\widehat{()}:\mathrm{pro}𝒮\widehat{𝒮}`$ preserves finite limits and finite colimits.
Proof 1. This follows from the definition of $`\widehat{()}`$ and Remark 2.1.
2. Suppose $`f:XY`$ in $`\mathrm{pro}𝒮`$ induces an isomorphism in $`/\mathrm{}`$-cohomology. We have the natural sequence of commutative squares
$$\begin{array}{ccc}H^{}(X;/\mathrm{})& \stackrel{f^{}}{}& H^{}(Y;/\mathrm{})\\ =& & =\\ \mathrm{colim}_sH^{}(X_s;/\mathrm{})& & \mathrm{colim}_tH^{}(Y_t;/\mathrm{})\\ & & \\ \mathrm{colim}_sH^{}(\widehat{X_s};/\mathrm{})& & \mathrm{colim}_tH^{}(\widehat{Y_t};/\mathrm{})\\ & & \\ H^{}(\widehat{X};/\mathrm{})& \stackrel{\widehat{f}^{}}{}& H^{}(\widehat{Y};/\mathrm{})\end{array}$$
which proves the second assertion.
3. This is clear.
4. Since $`\widehat{()}:𝒮\widehat{𝒮}`$ is a left adjoint functor it preserves all colimits and it commutes with finite limits. The functor $`lim:\mathrm{pro}\widehat{𝒮}\widehat{𝒮}`$ commutes with all limits and finite colimits. Hence the last statement holds, too. $`\mathrm{}`$
###### Corollary 2.13
The functor $`\widehat{()}:\mathrm{pro}𝒮\widehat{𝒮}`$ induces an equivalence of homotopy categories
$$\mathrm{Ho}(\mathrm{pro}𝒮)\stackrel{}{}\mathrm{Ho}(\widehat{𝒮}),$$
where the left hand side is the $`/\mathrm{}`$-cohomological homotopy category of \[I4\].
Isaksen has considered a different functor $`\mathrm{pro}𝒮\widehat{𝒮}`$ in \[I4\] and has shown that it induces an equivalence on the level of homotopy categories, too. Our approach fits better in the later picture for the comparison of generalized cohomology theories.
### 2.5 The model structure on $`\widehat{𝒮}`$ is fibrantly generated
For our studies of generalized cohomology theories and a stable structure on profinite spaces, we have to show the technical point that the model structure on $`\widehat{𝒮}`$, hence also on $`\widehat{𝒮}_{}`$, is fibrantly generated. The necessity of this point becomes clear from the localization and stabilization results of the appendix, cf. Theorem A.3.
We recall some notations and constructions from \[Mo2\]. Let $`n0`$ be a non-negative integer and $`S`$ be a profinite set. The functor $`\widehat{𝒮}^{\mathrm{op}}`$, $`X\mathrm{Hom}_\widehat{}(X_n,S)`$ is representable by a simplicial profinite set, which we denote by $`L(S,n)`$. It is given by the formula
$$L(S,n):\mathrm{\Delta }^{\mathrm{op}}\widehat{},[k]S^{\mathrm{Hom}_\mathrm{\Delta }([n],[k])}.$$
If $`M`$ is a profinite abelian group, then $`L(M,n)`$ has a natural structure of a simplicial profinite abelian group and the abelian group $`\mathrm{Hom}_{\widehat{𝒮}}(X,L(M,n))`$ can be identified with the group $`C^n(X;M)`$ of continuous $`n`$-cochains of $`X`$ with values in $`M`$. Furthermore, for every $`k`$ $`L(M,)([k])`$ may be considered in the usual way as an abelian cochain complex. For a profinite abelian group $`M`$, let $`Z^n(X;M)`$ denote the abelian group of $`n`$-cocycles of the complex $`C^{}(X;M)`$. The functor $`\widehat{𝒮}^{\mathrm{op}}`$, $`XZ^n(X;M)`$ is also representable by a simplicial profinite abelian group, which we denote by $`K(M,n)`$, called the profinite Eilenberg-MacLane space of type $`(M,n)`$. We may define $`K(M,n)([k])`$ to be the subgroup of all cocycles of $`L(M,n)([k])`$; see also \[Ma1\] §23, for these constructions.
The natural homomorphism $`C^n(X;M)Z^{n+1}(X;M)`$ given by the differential defines a natural morphism of simplicial profinite abelian groups $`L(M,n)K(M,n+1)`$.
Consider the two sets of morphisms
$$P:=\{L(M,n)K(M,n+1),K(M,n)|M\mathrm{abelian}\mathrm{pro}\mathrm{}\mathrm{group},n0\}$$
and
$$Q:=\{L(M,n)|M\mathrm{abelian}\mathrm{pro}\mathrm{}\mathrm{group},n0\}$$
of Lemme 2 of \[Mo2\].
###### Theorem 2.14
The simplicial model structure on $`\widehat{𝒮}`$, in which the weak equivalences are the $`/\mathrm{}`$-cohomological isomorphisms and the cofibrations are the dimensionwise monomorphisms, is left proper and fibrantly generated with $`P`$ as the set of generating fibrations and $`Q`$ as the set of generating trivial fibrations.
Proof The left properness follows from the fact that all objects in $`\widehat{𝒮}`$ are cofibrant, cf. Corollary 13.1.3 of \[Hi\].
We write Fib and Cof for the classes of fibrations and cofibrations of the model structure on $`\widehat{𝒮}`$ of \[Mo2\]. Note that the subcategory $`Q`$-cocell of relative $`Q`$-cocell complexes consists here of limits of pullbacks of elements of $`Q`$ and look at Definition 2.1.7 of \[Ho1\] for the notations $`P\mathrm{proj}`$ and $`Q\mathrm{fib}`$ used below. In order to prove the second statement we check the six conditions of the dual of Theorem 2.1.19 of \[Ho1\]. We only assume that the category has all small limits but only finite colimits. But since we use cosmall instead of small objects, the theorem holds for this kind of fibrantly generated model categories as well. We check now the six conditions of \[Ho1\], Theorem 2.1.19.
1. It is clear that the weak equivalences satisfy the two-out-of-three axioms.
2. and 3. We have to show that the codomains of the maps in $`P`$ and $`Q`$ are cosmall relative to $`P`$-cocell and $`Q`$-cocell, respectively. Note that the terminal object $``$ is cosmall relative to all maps. Hence it remains to show that the objects $`K(M,n)`$ are cosmall relative to $`Q`$-cocell. It suffices to show that they are cosmall relative to a filtered sequence of maps $`\mathrm{}Y_1Y_0`$ of maps in $`\widehat{𝒮}`$. Consider the canonical map $`f:\mathrm{colim}_\alpha \mathrm{Hom}_{\widehat{𝒮}}(Y_\alpha ,K(M,n))\mathrm{Hom}_{\widehat{𝒮}}(lim_\alpha Y_\alpha ,K(M,n)).`$ We have to show that it is an isomorphism. By the definition of the spaces $`K(M,n)`$ this map is equal to the map
$$\mathrm{colim}_\alpha Z^n(Y_\alpha ,M)Z^n(\underset{\alpha }{lim}Y_\alpha ,M).$$
But this map is already an isomorphism on the level of complexes $`\mathrm{colim}_\alpha C^n(Y_\alpha ,M)C^n(lim_\alpha Y_\alpha ,M)`$ as in\[Se\], Proposition 8; hence $`f`$ is an isomorphism.
4. We know that $`L(M,n)`$ is a trivial fibration. Since trivial fibrations are preserved under pullbacks and limits, we get $`Q\mathrm{cocell}W\mathrm{Fib}WP\mathrm{fib}`$ which is what we had to show.
5. Given a diagram
(2)
$$\begin{array}{ccc}A& & L(M,n)\\ & & \\ B& & \end{array}$$
we have to show that there is a lift if the map $`f:AB`$ is in $`P\mathrm{proj}`$, i.e. has the left lifting property with respect to $`P`$. But if $`F`$ is in $`P\mathrm{proj}`$, then we get a lifting $`BK(M,n+1)`$ for any map $`AK(M,n+1)`$. Hence the diagram (2) yields a diagram
(3)
$$\begin{array}{ccc}A& & L(M,n)\\ & & \\ B& & K(M,n+1)\\ & & \\ B& & \end{array}$$
and we know that there is a lifting $`BL(M,n)`$ in the upper rectangle which is also the lift of the diagram (2) above. Hence $`fQ\mathrm{proj}`$.
It remains to show that $`P\mathrm{proj}W`$. Let $`f:AB`$ be a map in $`P\mathrm{proj}`$. By definition of the spaces $`K(/\mathrm{},n)`$ and the definition of $`P\mathrm{proj}`$ we get that $`f^{}:Z^n(B,/\mathrm{})Z^n(A,/\mathrm{})`$ is surjective for all $`n0`$. Hence it is enough to show that $`f^{}(\mathrm{Im}(C^nAZ^{n+1}A))\mathrm{Im}(C^nBZ^{n+1}B).`$ The other lifting property of maps in $`P\mathrm{proj}`$ is equivalent to the surjectivity of the map $`C^nBC^nA\times _{Z^{n+1}A}Z^{n+1}B`$ from which we get the desired result that $`f^{}:H^n(B,/\mathrm{})H^n(A,/\mathrm{})`$ is an isomorphism for all $`n0`$.
6. We show that $`WQ\mathrm{proj}P\mathrm{proj}`$. Let $`f:AB`$ be a map that belongs to $`W`$ and $`Q\mathrm{proj}`$. Since $`f^{}`$ is an isomorphism on $`/\mathrm{}`$-cohomology, it can be shown that $`f`$ induces an isomorphism on cohomology with coefficients in any abelian pro-$`\mathrm{}`$-group $`M`$. This implies already the surjectivity of the map $`f^{}:Z^n(B,M)Z^n(A,M)`$ for all $`n0`$. The lifting property of maps in $`Q\mathrm{proj}`$ implies the surjectivity of the induced map $`f^{}:C^n(B,M)C^n(A,M)`$ for all $`n0`$. Using the isomorphism on cohomology it follows that $`C^nBC^nA\times _{Z^{n+1}A}Z^{n+1}B`$ is surjective which is equivalent to the other desired lifting property of maps in $`P\mathrm{proj}`$. Hence $`fP\mathrm{proj}`$.
Now we have proved using theorem 2.1.19 of \[Ho1\] that the quadruple $`(\widehat{𝒮},W,Q\mathrm{proj},P\mathrm{fib})`$ is a fibrantly generated model category. It remains to show that it coincides with the given structure ($`\widehat{𝒮}`$, $`W`$, Cof, Fib). Since $`L(M,n)`$ is a trivial fibration, we know already Cof $`Q\mathrm{proj}`$. It remains to show that every map in $`Q\mathrm{proj}`$ is a monomorphism in each dimension. Then the cofibrations and weak equivalences of the two structures coincide and hence the fibrations coincide as well.
Let $`f:XY`$ be a map in $`Q\mathrm{proj}`$. Hence every map $`XL(M,n)`$ can be lifted to a map $`YL(M,n)`$. This means by using the definition of the spaces $`L(M,n)`$ that the map $`C^n(Y,M)C^n(X,M)`$ induced by $`f`$ is surjective for all $`n0`$. This is equivalent to the surjectivity of the maps $`\mathrm{Hom}_\widehat{}(Y_n,M)\mathrm{Hom}_\widehat{}(X_n,M)`$ for all $`n0`$ and all abelian pro-$`\mathrm{}`$-groups $`M`$. If we choose $`M=F(X_n)`$ to be the free abelian pro-$`\mathrm{}`$-group on the set $`X_n`$ defined in \[Se\], then we see that the map $`X_nF(X_n)`$, sending $`x`$ to its class in $`F(X_n)`$, is in the image of $`\mathrm{Hom}_\widehat{}(Y_n,M)\mathrm{Hom}_\widehat{}(X_n,M)`$ only if $`X_nY_n`$ is a monomorphism for each $`n`$. Hence $`f`$ is also a cofibration. $`\mathrm{}`$
## 3 Profinite spectra
We introduce the usual notion of spectra on $`\widehat{𝒮}`$. The results of the appendix show that there is a stable model structure on $`\mathrm{Sp}(\widehat{𝒮}_{})`$. This is the main technical result for the construction of generalized cohomology theories on schemes via the profinite étale topological type functor.
We study the behavior of the profinite completion functor $`\widehat{()}:\mathrm{Sp}(𝒮)\mathrm{Sp}(\widehat{𝒮}_{})`$ and show that it factorizes through stable equivalences and is in fact a left Quillen functor with right adjoint the forgetful functor.
The following subsection is dedicated to stable homotopy groups and their relation to stable equivalences. We show that these homotopy groups behave well under completion of spaces. In particular, we study them for profinite Eilenberg-MacLane spectra and the completed complex cobordism spectrum $`\widehat{MU}`$. In this last case, we get $`\pi _{}(\widehat{MU})_{\mathrm{}}\pi _{}(MU)`$.
Furthermore, we give an explicit construction of a fibrant replacement functor for spectra satisfying some conditions among which is $`MU`$. These results are similar to those of \[De\].
At the end of this section we deduce the existence of Postnikov-decompositions for profinite spectra from general model category theory.
### 3.1 The stable structure of profinite spectra
In order to stabilize the category $`\widehat{𝒮}`$ of profinite spaces we begin with the usual definition of spectra.
###### Definition 3.1
A spectrum $`X`$ of simplicial profinite sets consists of a sequence $`X_n\widehat{𝒮}_{}`$ of pointed profinite spaces for $`n0`$ and maps $`\sigma _n:S^1X_nX_{n+1}`$ in $`\widehat{𝒮}_{}`$, where $`S^1=\mathrm{\Delta }^1/\mathrm{\Delta }^1\widehat{𝒮}_{}`$.
A map $`f:XY`$ of spectra consists of maps $`f_n:X_nY_n`$ in $`\widehat{𝒮}_{}`$ for $`n0`$ such that $`\sigma _n(1f_n)=f_{n+1}\sigma _n`$.
We denote by $`\mathrm{Sp}(\widehat{𝒮}_{})`$ the corresponding category and call it the category of profinite spectra.
We apply the results of Appendix B to the category $`\widehat{𝒮}_{}`$. By Theorem 2.14 its model structure is simplicial, left proper and fibrantly generated. Note that $`S^1`$ is a left Quillen endofunctor since it takes monomorphisms to monomorphisms and preserves cohomological equivalences by Lemma 2.2.
###### Theorem 3.2
There is a stable model structure on $`\mathrm{Sp}(\widehat{𝒮}_{})`$ for which the prolongation $`S^1:\mathrm{Sp}(\widehat{𝒮}_{})\mathrm{Sp}(\widehat{𝒮}_{})`$ is a Quillen equivalence.
In particular, the stable equivalences are the maps that induce an isomorphism on all generalized cohomology theories, represented by profinite $`\widehat{\mathrm{\Omega }}`$-spectra; the stable cofibrations are the maps $`i:AB`$ such that $`i_0`$ and the induced maps $`j_n:A_n_{S^1A_{n1}}S^1B_{n1}B_n`$ are monomorphisms; the stable fibrations are the maps with the right lifting property with respect to all maps that are both stable equivalences and stable cofibrations.
Proof By Theorem 2.14, we know that the model structure on $`\widehat{𝒮}_{}`$ is fibrantly generated, left proper and simplicial. Hence we can apply Theorem B.11 to $`\widehat{𝒮}_{}`$, $`S^1`$. The fact that $`S^1`$ is a Quillen equivalence for this model structure is implied by Theorem B.13 for $`T=S^1`$. $`\mathrm{}`$
###### Remark 3.3
The stable fibrant objects are exactly the $`\widehat{\mathrm{\Omega }}`$-spectra, i.e. spectra $`E`$ such that each $`E_n`$ is a fibrant object and the adjoint structure maps $`E_n\widehat{\mathrm{\Omega }}E_{n+1}`$ are weak equivalences in $`\widehat{𝒮}_{}`$ for all $`n0`$.
Define the functor $`\widehat{\mathrm{\Omega }}:\mathrm{Sp}(\widehat{𝒮}_{})\mathrm{Sp}(\widehat{𝒮}_{}),E\widehat{\mathrm{\Omega }}(E)`$, where $`\widehat{\mathrm{\Omega }}(E)_n:=\widehat{\mathrm{\Omega }}(E_n)`$. As usual one can prove the following lemma.
###### Lemma 3.4
1. The functor
$$\widehat{𝒮}\widehat{𝒮},ES^1E$$
is an equivalence of categories.
2. The homotopy category $`\widehat{𝒮}`$ of profinite spectra is an additive category.
Proof Choose a functorial stable replacement
$$EE^f.$$
Then $`E\widehat{\mathrm{\Omega }}(E^f)`$ is an inverse to $`ES^1E`$ using the two stable equivalences
$$E\widehat{\mathrm{\Omega }}((S^1E)^f)$$
and
$$S^1\widehat{\mathrm{\Omega }}(E^f)E^f$$
which are induced by the adjunction morphisms.
As $`S^1`$ is a cogroup object in $`\mathrm{Ho}(\widehat{𝒮}_{})`$, its codiagonal $`\varphi :S^1S^1S^1`$ induces, via first assertion of the lemma and $`\varphi \mathrm{id}_E:S^1ES^1ES^1E`$, for any spectrum $`E`$ a morphism $`EEE`$. This yields a natural structure of a cogroup object on $`E`$. Being natural in $`E`$, this structure has to be abelian. Thus the category $`\widehat{𝒮}`$ is additive. $`\mathrm{}`$
###### Remark 3.5
Since $`\mathrm{Sp}(\widehat{𝒮}_{})`$ is a pointed model category we may conclude by the general theory of \[Qu1\] I, §2 and §3, that $`\mathrm{Sp}(\widehat{𝒮}_{})`$ has fiber and cofiber sequences, which satisfy the axioms of a triangulation. Since the above theorem tells us that $`\mathrm{\Sigma }`$ is an equivalence on $`\widehat{𝒮}`$, we may call $`\mathrm{Sp}(\widehat{𝒮}_{})`$ a stable category.
### 3.2 Profinite completion of spectra
###### Lemma 3.6
The functor $`\widehat{()}:𝒮_{}\widehat{𝒮}_{}`$ commutes with smash products.
Furthermore, the two functors $`\mathrm{\Omega }`$ and $`\widehat{\mathrm{\Omega }}`$ agree after applying the forgetful functor $`||:\widehat{𝒮}_{}𝒮_{}`$, i.e. $`\mathrm{\Omega }(|Z|)|\widehat{\mathrm{\Omega }}(Z)|`$ for every pointed profinite space $`Z\widehat{𝒮}_{}`$.
Proof Since $`\widehat{()}`$ is left adjoint to the forgetful functor it preserves colimits. Hence it remains to check that it commutes with finite products. But this follows from the fact that the finite quotients of a product $`X\times Y`$ of two spaces are in bijective correspondence with pairs of finite quotients of $`X`$ and $`Y`$ respectively. Since $`\widehat{()}`$ is defined by applying a limit we see immediately that it commutes with finite products.
For every $`X𝒮_{}`$ we have natural bijections coming from adjunction
$$\begin{array}{ccccc}\hfill \mathrm{Hom}_𝒮(S^1X,|\widehat{\mathrm{\Omega }}Z|)& & \mathrm{Hom}_{\widehat{𝒮}}(\widehat{X},\widehat{\mathrm{\Omega }}Z)& & \mathrm{Hom}_{\widehat{𝒮}}(S^1\widehat{X},Z)\hfill \\ & & \mathrm{Hom}_{\widehat{𝒮}}(\widehat{S^1X},Z)& & \mathrm{Hom}_𝒮(S^1X,|Z|)\hfill \\ & & \mathrm{Hom}_𝒮(X,\mathrm{\Omega }|Z|).& & \end{array}$$
Hence the two Hom-functors defined by $`\mathrm{\Omega }(|Z|)`$ and $`|\widehat{\mathrm{\Omega }}(Z)|`$ agree on $`𝒮`$ and hence by Yoneda the two spaces $`\mathrm{\Omega }(|Z|)`$ and $`|\widehat{\mathrm{\Omega }}(Z)|`$ are isomorphic for every profinite space $`Z`$. $`\mathrm{}`$
Let $`\mathrm{Sp}(𝒮)`$ be the stable model structure of simplicial spectra defined in \[BF\]. We may extend the profinite completion to spectra. Let $`\widehat{()}:\mathrm{Sp}(𝒮)\mathrm{Sp}(\widehat{𝒮}_{})`$ be the profinite completion applied levelwise that takes the spectrum $`X`$ to the profinite spectrum $`\widehat{X}`$ whose structure maps are given, using Lemma 3.6, by
$$S^1\widehat{X}_n\widehat{S^1X_n}\stackrel{\widehat{\sigma }}{}\widehat{X}_{n+1}.$$
Since the two functors $`\widehat{()}`$ and $`||`$ on $`\widehat{𝒮}_{}`$ commute with smash products we get that commutative diagrams
$$\begin{array}{ccc}X_{n+1}& & |Y_{n+1}|\\ & & \\ S^1X_n& & S^1|Y_n|\end{array}$$
are in bijective correspondence with commutative diagrams
$$\begin{array}{ccc}\widehat{X}_{n+1}& & Y_{n+1}\\ & & \\ S^1\widehat{X}_n& & S^1Y_n\end{array}$$
for every $`X\mathrm{Sp}(𝒮)`$ and $`Y\mathrm{Sp}(\widehat{𝒮}_{})`$. Let $`||:\mathrm{Sp}(\widehat{𝒮}_{})\mathrm{Sp}(𝒮)`$ be the levelwise applied forgetful functor. Hence $`\widehat{()}`$ and $`||`$ form an adjoint pair of functors.
###### Proposition 3.7
The functor $`\widehat{()}:\mathrm{Sp}(𝒮)\mathrm{Sp}(\widehat{𝒮}_{})`$ preserves weak equivalences and cofibrations.
The functor $`||:\mathrm{Sp}(\widehat{𝒮}_{})\mathrm{Sp}(𝒮)`$ preserves fibrations and weak equivalences between fibrant objects.
In particular, $`\widehat{()}`$ induces a functor on the homotopy categories and the adjoint pair $`(\widehat{()},||)`$ is a Quillen pair of adjoint functors.
Proof Let $`i:AB`$ be a cofibration in $`\mathrm{Sp}(𝒮)`$. Since $`\widehat{()}:𝒮_{}\widehat{𝒮}_{}`$ preserves cofibrations and pushouts as a left Quillen functor, the maps $`\widehat{i_0}`$ and $`\widehat{j_n}`$ are cofibrations in $`\widehat{𝒮}_{}`$. Hence $`\widehat{i}`$ is a cofibration in $`\mathrm{Sp}(\widehat{𝒮}_{})`$.
Now let $`f:XY`$ be a weak equivalence in $`\mathrm{Sp}(𝒮)`$. We want to show that $`\widehat{f}`$ is a weak equivalence in $`\mathrm{Sp}(\widehat{𝒮}_{})`$. By \[Hi\], Theorem 9.7.4, this is equivalent to the statement that $`\mathrm{Map}(\stackrel{~}{\widehat{f}},E):\mathrm{Map}(\stackrel{~}{\widehat{Y}},E)\mathrm{Map}(\stackrel{~}{\widehat{X}},E)`$ is a weak equivalence of simplicial sets for every fibrant object $`E`$ and some cofibrant approximation $`\stackrel{~}{\widehat{f}}`$ of $`\widehat{f}`$ in $`\mathrm{Sp}(\widehat{𝒮}_{})`$. Since $`\widehat{()}`$ and $`X`$ are left adjoints, we have natural isomorphisms extending the adjunction of $`\widehat{()}`$ and $`||`$ for every $`U\mathrm{Sp}(𝒮)`$ and $`V\mathrm{Sp}(\widehat{𝒮}_{})`$
(4)
$$\mathrm{Map}_{\mathrm{Sp}(𝒮)}(U,|V|)\mathrm{Map}_{\mathrm{Sp}(\widehat{𝒮}_{})}(\widehat{U},V).$$
Hence $`\mathrm{Map}_{\mathrm{Sp}(\widehat{𝒮}_{})}(\widehat{f},E)`$ is a weak equivalence of simplicial sets if and only if $`\mathrm{Map}_{\mathrm{Sp}(𝒮)}(f,|E|)`$ is a weak equivalence. Since $`\widehat{()}`$ preserves cofibrations, the completion of $`\stackrel{~}{f}`$ is a cofibrant approximation of $`\widehat{f}`$. We conclude the first part by \[Hi\], Theorem 9.7.4, and the following lemma, reminding the fact that the fibrant objects of $`\mathrm{Sp}(𝒮)`$ are exactly the $`\mathrm{\Omega }`$-spectra.
###### Lemma 3.8
If $`E`$ is an $`\widehat{\mathrm{\Omega }}`$-spectrum, then $`|E|`$ is an $`\mathrm{\Omega }`$-spectrum.
This is the analogue of the fact that the underlying simplicial set $`|X|`$ of a fibrant profinite space is a Kan simplicial set, cf. \[Mo2\], §2.1, Proposition 1.
Proof Let $`E`$ be an $`\widehat{\mathrm{\Omega }}`$-spectrum. This implies that each $`E_n`$ is fibrant and $`E_n\widehat{\mathrm{\Omega }}E_{n+1}`$ is a weak equivalence for each $`n0`$. By \[Mo2\], $`\mathrm{\S }`$ 2, Proposition 1, this implies that $`|E_n|`$ is fibrant and $`|E_n|\mathrm{\Omega }|E_{n+1}|`$ is a weak equivalence for each $`n`$. Hence $`|E|`$ is an $`\mathrm{\Omega }`$-spectrum. $`\mathrm{}`$
We continue the proof of the second statement of the proposition. The fact that $`||`$ preserves fibrations follows now from adjunction since $`\widehat{()}`$ preserves trivial cofibrations. Now let $`f:EF`$ be a weak equivalence between fibrant objects in $`\mathrm{Sp}(\widehat{𝒮}_{})`$. Again by \[Hi\], Theorem 9.7.4, we have to show that $`\mathrm{Map}_{\mathrm{Sp}(𝒮)}(W,|f|)`$ is a weak equivalence for every cofibrant object $`W`$ of $`\mathrm{Sp}(𝒮)`$. By the isomorphism (4) this is equivalent to that $`\mathrm{Map}_{\mathrm{Sp}(\widehat{𝒮}_{})}(\widehat{W},f)`$ is a weak equivalence for every such $`W`$. But since $`\widehat{W}`$ is also cofibrant in $`\mathrm{Sp}(\widehat{𝒮}_{})`$ and since $`f`$ is a weak equivalence in $`\mathrm{Sp}(\widehat{𝒮}_{})`$, we get that $`\mathrm{Map}_{\mathrm{Sp}(\widehat{𝒮}_{})}(\widehat{W},f)`$ is a weak equivalence of simplicial sets.
The last statement follows from general model category theory. $`\mathrm{}`$
### 3.3 Stable homotopy groups and fibrant replacements
For a level fibrant replacement functor of a profinite spectrum $`E`$ we may use the explicit construction of the $`\mathrm{}`$-completion functor $`X\widehat{X}^{\mathrm{}}`$ in $`\widehat{𝒮}_{}`$. Since all construction in \[BK\] and \[Mo2\] are compatible with products there are natural maps
$$X\times \widehat{Y}^{\mathrm{}}\widehat{X}^{\mathrm{}}\times \widehat{Y}^{\mathrm{}}\stackrel{}{}\widehat{X\times Y}^{\mathrm{}}$$
and similarly maps
$$X\widehat{Y}^{\mathrm{}}\widehat{X}^{\mathrm{}}\widehat{Y}^{\mathrm{}}\stackrel{}{}\widehat{XY}^{\mathrm{}}.$$
For a simplicial spectrum $`E`$ this yields structure maps
$$S^1\widehat{E_n}^{\mathrm{}}(\widehat{S^1E_n})^{\mathrm{}}\stackrel{\widehat{\sigma _n}^{\mathrm{}}}{}\widehat{E_{n+1}}^{\mathrm{}},$$
which ensure that we may associate to every profinite spectrum $`E`$ a level fibrant spectrum $`E^{lf}`$ such that $`EE^{lf}`$ is a level equivalence.
We may also define homotopy groups of profinite spectra. For $`n`$ we set
(5)
$$\pi _n(E):=\pi _n(E^{lf})=\mathrm{colim}_k\pi _{n+k}(E_k^{lf}).$$
For morphisms of spectra we set $`\pi _n(g):=\pi _n(g^{lf})`$.
###### Remark 3.9
In order to calculate the stable homotopy groups of a spectrum $`E`$, we may replace it by a fibrant spectrum $`E^f`$. Then its homotopy groups are the homotopy groups of the infinite loop space $`E_0^f`$. By Proposition 2.9 we know that the homotopy groups of a profinite space have a natural structure of pro-$`\mathrm{}`$-groups. Hence we conclude that the stable homotopy groups of any profinite spectrum have the structure of pro-$`\mathrm{}`$-groups.
We may describe stable equivalences via stable homotopy groups.
###### Proposition 3.10
A map of profinite spectra $`g:EF`$ is a stable equivalence if and only if $`\pi _n(g)`$ is an isomorphism for all $`n`$.
Proof Since we are interested in the homotopy type of the map, we may consider a stable fibrant replacement $`g^f:E^fF^f`$ of $`g`$. The map $`g`$ is a stable equivalence if and only if $`g^f`$ is a stable equivalence. But a map between $`\widehat{\mathrm{\Omega }}`$-spectra is a stable equivalence if and only if it is a level equivalence, e.g. see \[Ho2\]. But this means that all maps $`\pi _n(g_k^f)`$ are isomorphisms for all $`n`$ and $`k`$, which finishes the proof by Proposition 2.7. $`\mathrm{}`$
###### Corollary 3.11
Let $`f:XY`$ be a map in $`\mathrm{Sp}(\widehat{𝒮}_{})`$ such that $`|f|:|X||Y|`$ is a stable equivalence of $`\mathrm{Sp}(𝒮)`$. Then $`f`$ is a stable equivalence of $`\mathrm{Sp}(\widehat{𝒮}_{})`$.
Proof This follows immediately from the proposition above. $`\mathrm{}`$
From model category theory we know that fibrant replacements with respect to the stable model structure exist for every $`E\mathrm{Sp}(\widehat{𝒮}_{})`$. We would like to construct explicit stable fibrant replacements in $`\mathrm{Sp}(\widehat{𝒮}_{})`$. We will employ the Bousfield-Kan-$`\mathrm{}`$-completion functor defined on profinite spaces. Since the $`\mathrm{}`$-completion behaves well only under certain conditions, cf. \[BK\], we give a construction only for profinite spectra satisfying the following hypotheses.
The idea for the following construction is inspired by Dehon \[De\]. Let $`E`$ be a $`(1)`$-connected profinite spectrum, i.e. each profinite space $`E_n`$ is $`(n1)`$-connected for all $`n0`$ and let $`|E|\mathrm{Sp}(𝒮)`$ be its underlying simplicial spectrum. We may consider an $`\mathrm{\Omega }`$-spectrum $`R(|E|)`$ stable equivalent to $`|E|`$, i.e. a stable fibrant replacement of $`|E|`$ in $`\mathrm{Sp}(𝒮)`$. After the application of the levelwise $`\mathrm{}`$-completion we get maps $`\widehat{(R|E|)_n}^{\mathrm{}}\widehat{\mathrm{\Omega }(R|E|)_{n+1}}^{\mathrm{}}`$ that are weak equivalences in $`\widehat{𝒮}_{}`$ for all $`n`$. Then Proposition 2.5 implies that all the composite maps
$$\widehat{(R|E|)_n}^{\mathrm{}}\widehat{(\mathrm{\Omega }R|E|_{n+1})}^{\mathrm{}}\widehat{\mathrm{\Omega }}\widehat{(R|E|)_{n+1}}^{\mathrm{}}$$
are weak equivalences in $`\widehat{𝒮}_{}`$ for all $`n1`$. It is clear that the resulting profinite spectrum is also level fibrant. From Corollary 3.11, we deduce that it is also stable equivalent to $`E`$. Hence we have constructed an explicit stable fibrant replacement of $`E`$ in $`\mathrm{Sp}(\widehat{𝒮}_{})`$.
###### Definition and Proposition 3.12
Let $`E`$ be a $`(1)`$-connected profinite spectrum. We suppose that the $`/\mathrm{}`$-cohomology of each $`E_n`$, $`n1`$, is finite dimensional in each degree. We consider the profinite spectrum $`\widehat{E}^{\mathrm{}}\mathrm{Sp}(\widehat{𝒮}_{})`$ whose spaces are defined by $`\widehat{E}_0^{\mathrm{}}:=\widehat{\mathrm{\Omega }}\widehat{(R|E|)_1}^{\mathrm{}}`$ for $`n=0`$ and by $`\widehat{(R|E|)_n}^{\mathrm{}}`$ for all $`n1`$ with structure maps adjoint to the above defined maps $`\widehat{(R|E|)_n}^{\mathrm{}}\widehat{\mathrm{\Omega }}\widehat{(R|E|)_{n+1}}^{\mathrm{}}`$ for $`n1`$ and the obvious map for $`n=0`$. By the preceding discussion, the profinite spectrum $`\widehat{E}^{\mathrm{}}`$ is a stable fibrant replacement of $`E`$ in $`\mathrm{Sp}(\widehat{𝒮}_{})`$ and we call it the $`\mathrm{}`$-completion of $`E`$.
The reason why we use this definition for $`\widehat{E}_0^{\mathrm{}}`$ is that if $`E_0`$ has infinitely many connected components, the natural structure map will not be a weak equivalence any more.
This discussion also leads to an $`\mathrm{}`$-completion of spectra $`E\mathrm{Sp}(𝒮)`$ which is inspired by \[De\], §1.6.
###### Definition 3.13
Let $`E`$ be a $`(1)`$-connected spectrum in $`\mathrm{Sp}(𝒮)`$. We suppose that the $`/\mathrm{}`$-cohomology of each $`E_n`$, $`n1`$, is finite dimensional in each degree. Let $`RE`$ be an equivalent $`\mathrm{\Omega }`$-spectrum. Then we define the profinite $`\mathrm{}`$-completion $`\widehat{E}^{\mathrm{}}`$ to be the profinite spectrum whose spaces are defined by $`\widehat{E}_0^{\mathrm{}}:=\widehat{\mathrm{\Omega }}\widehat{(R(E)_1)}^{\mathrm{}}`$ for all $`n=0`$ and by $`\widehat{(R(E)_n)}^{\mathrm{}}`$ for all $`n1`$ with structure maps adjoint to the above defined maps $`\widehat{R(E)_n}^{\mathrm{}}\widehat{\mathrm{\Omega }}\widehat{R(E)_n}^{\mathrm{}}`$ for $`n1`$ and the obvious map for $`n=0`$.
###### Remark 3.14
As above, for $`E\mathrm{Sp}(𝒮)`$, it is clear that $`\widehat{E}^{\mathrm{}}`$ is stable equivalent to $`\widehat{E}`$ in $`\mathrm{Sp}(\widehat{𝒮}_{})`$. From the homotopy theoretic point of view we may consider either $`\widehat{E}^{\mathrm{}}`$ or $`\widehat{E}`$. The point is that we want to apply the comparison results of Proposition 2.10 to the situation of spectra. The previous definition gives us the form of the $`\widehat{E}`$ that fits well with the results in Proposition 2.10.
Similar versions of the following facts have also been proved by Dehon \[De\].
###### Proposition 3.15
Let $`E`$ be a $`(1)`$-connected spectrum and suppose that the $`/\mathrm{}`$-cohomology of each $`E_n`$, $`n1`$, is finite dimensional in each degree. Then the stable homotopy groups of the profinite $`\mathrm{}`$-completion $`\widehat{E}^{\mathrm{}}`$ are given by the following isomorphism for all $`n`$
$$\pi _n\widehat{E}^{\mathrm{}}\widehat{\pi _nE}^{\mathrm{}}_{\mathrm{}}_{}\pi _nE.$$
Proof Since $`\widehat{E}^{\mathrm{}}`$ is an $`\widehat{\mathrm{\Omega }}`$-spectrum, the stable homotopy group $`\pi _n\widehat{E}^{\mathrm{}}`$ is equal to the homotopy group $`\pi _{n+1}((\widehat{E}^{\mathrm{}})_1)`$ for all $`n`$. The hypothesis on $`E`$ and the definition of $`\widehat{E}^{\mathrm{}}`$ imply that the spaces $`\widehat{E}_n^{\mathrm{}}`$ satisfy the conditions of Proposition 2.10. This completes the proof. $`\mathrm{}`$
###### Corollary 3.16
Let $`MU`$ be the simplicial spectrum representing complex cobordism. For the profinite $`\mathrm{}`$-completion $`\widehat{MU}^{\mathrm{}}`$ of $`MU`$ there is an isomorphism
$$\pi _{}\widehat{MU}=\pi _{}\widehat{MU}^{\mathrm{}}_{\mathrm{}}_{}\pi _{}MU_{\mathrm{}}_{}𝕃_{}$$
where $`𝕃`$ denotes the Lazard ring, cf. \[Ad\].
Proof Since $`MU`$ is $`(1)`$-connected and its cohomology groups are finitely generated in each degree, cf. \[Ad\], $`MU`$ satisfies the hypothesis of Proposition 3.15 above. $`\mathrm{}`$
We finally consider the profinite completion $`\widehat{KU}`$ of the $`\mathrm{\Omega }`$-spectrum representing complex $`K`$-theory, see e.g \[Ma2\]: $`\widehat{KU}_{2i}=\widehat{BU\times }`$ and $`\widehat{KU}_{2i+1}=\widehat{U}`$ for all $`i0`$. By Lemma 2.8 and Proposition 2.10 we get
###### Corollary 3.17
$`\pi _{2i}(\widehat{KU})_{\mathrm{}}\mathrm{and}\pi _{2i+1}(\widehat{KU})=0\mathrm{for}\mathrm{all}i.`$
## 4 Generalized cohomology theories on profinite spaces
We define generalized cohomology theories on $`\widehat{𝒮}`$ via the stable category of profinite spectra in the classical way. Such profinitely completed cohomology theories have already been studied by Dehon in \[De\].
It is the main advantage of the stable category of profinite spectra that it gives a canonical and general setting for cohomology theories on profinite spaces and the profinite completion of cohomology theories on simplicial sets. In fact, our main objective are the cohomology theories represented by the completion of simplicial spectra such as $`\widehat{MU}`$. The results of the previous section enable us to calculate the coefficients of this theory.
An important tool for the calculation of cohomology groups is the profinite Atiyah-Hirzebruch spectral sequence. The construction of this spectral sequence is the analog of the topological Atiyah-Hirzebruch spectral sequence. It has already been constructed by Dehon in a more restricted setting. We slightly generalize his proof.
At the end we will prove a Künneth formula for $`\widehat{MU}`$ with $`/\mathrm{}^\nu `$-coefficients, similar to and inspired by the results of \[De\].
### 4.1 Generalized cohomology theories
In analogy to the stable homotopy theory of simplicial spectra the above construction enables us to consider generalized cohomology theories for profinite spaces. We define them to be the functors represented by profinite spectra. Let $`E`$ be a spectrum in $`\mathrm{Sp}(\widehat{𝒮}_{})`$, we set
(6)
$$E^n(X):=\mathrm{Hom}_{\widehat{𝒮}}(X,E[n]),$$
where $`\mathrm{Hom}_{\widehat{𝒮}}(X,E[n])`$ denotes the set of maps that lower the dimension by $`n`$, and call this the $`n`$-th cohomology group of $`X`$ with values in $`E`$. We set $`E^{}(X):=_nE^n(X)`$. For a pointed profinite space $`X`$ we define its $`n`$-th cohomology groups for the spectrum $`E`$ by
(7)
$$E^n(X):=\mathrm{Hom}_{\widehat{𝒮}}(\mathrm{\Sigma }^{\mathrm{}}(X),ES^n)$$
For a profinite space $`X`$ without a chosen basepoint, let $`X_+`$ be the profinite space $`Xpt`$ with additional basepoint. We define the cohomology of $`X`$ to be the one of $`X_+`$.
For a pair $`(X,A)`$ of profinite spaces we define the relative cohomology by
(8)
$$E^n(X,A):=E^n(X/A).$$
We have the identity $`E^{}(X,pt)=E^{}(X)`$. Just as for simplicial spectra \[Swi\], 8.21, one can prove the following isomorphism
(9)
$$\mathrm{Hom}_{\widehat{𝒮}}(\mathrm{\Sigma }^{\mathrm{}}(X),ES^n)\mathrm{colim}_k\mathrm{Hom}_{Ho(\widehat{𝒮}_{})}(\mathrm{\Sigma }^k(X),E_{n+k}).$$
In particular, for an $`\widehat{\mathrm{\Omega }}`$-spectrum $`E`$, we get:
(10)
$$E^n(X)\mathrm{Hom}_{Ho(\widehat{𝒮}_{})}(X,E_n).$$
### 4.2 Continuous cohomology theories
###### Definition 4.1
Let $`E`$ be a spectrum in $`\mathrm{Sp}(𝒮)`$. We define the continuous cohomology represented by $`E`$ to be the theory $`\widehat{E}^{}()`$ on $`\widehat{𝒮}_{}`$.
###### Remark 4.2
1. Note that since profinite completion commutes with smash-products by Lemma 3.6, the completion of a multiplicative cohomology theory is multiplicative, too.
2. Let $`E`$ be a $`(1)`$-connected spectrum. We suppose that the $`/\mathrm{}`$-cohomology of each $`E_n`$, $`n1`$, is finite dimensional in each degree. Since the spectra $`\widehat{E}`$ and $`\widehat{E}^{\mathrm{}}`$ are equivalent in $`\mathrm{Sp}(\widehat{𝒮}_{})`$, the continuous $`\mathrm{}`$-completed cohomology theory on $`\widehat{𝒮}_{}`$ represented by $`\widehat{E}^{\mathrm{}}`$ is isomorphic to $`\widehat{E}`$, i.e. for all $`X\widehat{𝒮}_{}`$ we have $`\widehat{E}^{\mathrm{}}(X)=\widehat{E}^{}(X)`$.
###### Example 4.3
By abuse of notations we write $`MU`$ for the simplicial spectrum $`\mathrm{Sing}(MU)`$ and we will write $`KU`$ for $`\mathrm{Sing}(KU)`$. There are the obvious examples for profinite spectra representing generalized cohomology theories:
1. The Eilenberg-MacLane spectra $`H\pi `$ given by $`H\pi _n=K(\pi ,n)`$ for an abelian pro-$`\mathrm{}`$-group $`\pi `$.
2. The profinite completion $`\widehat{KU}`$ of the simplicial spectrum representing complex K-theory.
3. The profinite completion $`\widehat{MU}`$ of the simplicial spectrum representing complex cobordism.
###### Proposition 4.4
Let $`E`$ be a multiplicative $`(1)`$-connected spectrum and suppose that the $`/\mathrm{}`$-cohomology of each $`E_n`$, $`n1`$, is finite dimensional in each degree. Then the coefficient ring $`\widehat{E}^{}:=\widehat{E}^{}(pt)`$ of the profinite completion satisfies the following isomorphism
$$\widehat{E}^{}(pt)\widehat{E^{}(pt)}^{\mathrm{}}_{\mathrm{}}_{}E^{}(pt).$$
For the homotopy groups of $`\widehat{E}`$ we get
$$\pi _{}(\widehat{E})\widehat{\pi _{}(E)}^{\mathrm{}}_{\mathrm{}}_{}\pi _{}(E).$$
Proof This is a corollary of Proposition 3.15, since $`\widehat{E}`$ and $`\widehat{E}^{\mathrm{}}`$ are equivalent objects in $`\mathrm{Sp}(\widehat{𝒮}_{})`$. $`\mathrm{}`$
Since the spectrum $`MU`$ satisfies the hypotheses of the previous proposition, we get the following result.
###### Corollary 4.5
For the homotopy groups of $`\widehat{MU}`$ there is an isomorphism
$$\pi _{}(\widehat{MU})\widehat{\pi _{}(MU)}^{\mathrm{}}_{\mathrm{}}_{}\pi _{}(MU).$$
For the coefficient ring of the profinite cobordism $`\widehat{MU}^{}`$ we have the following isomorphism
$$\widehat{MU}^{}=\widehat{MU}^{\mathrm{}}_{\mathrm{}}_{}MU^{}.$$
Let $`G`$ be a finite group. There is a simplicial Moore spectrum $`MG`$. For a simplicial spectrum $`E`$ we define $`EG:=EMG`$ to be the spectrum with coefficients in $`G`$ corresponding to $`E`$, cf. \[Ad\], Chapter III.
###### Definition 4.6
Let $`E`$ be a simplicial spectrum and $`G`$ a finite abelian group. We define the continuous cohomology theory with $`G`$-coefficients to be
$$\widehat{E}^{}(;G):=\widehat{EG}^{}().$$
By \[Ad\] III, Prop. 6.6, there are exact sequences
(11)
$$0\pi _n(E)G\pi _n(EG)\mathrm{Tor}_1^{}(\pi _{n1}(E),G)0$$
(12)
$$0E^n(X)G(EG)^n(X)\mathrm{Tor}_1^{}(E^{n+1}(X),G)0$$
for all spectra $`E`$ and all spaces $`X`$.
###### Corollary 4.7
For the homotopy groups of the profinite cobordism with $`/\mathrm{}^\nu `$-coefficients $`\widehat{MU}/\mathrm{}^\nu `$ there is an isomorphism
$$\pi _{}(\widehat{MU}/\mathrm{}^\nu )\widehat{\pi _{}(MU/\mathrm{}^\nu )}^{\mathrm{}}/\mathrm{}^\nu _{}\pi _{}(MU).$$
For the coefficient ring of the profinite cobordism $`\widehat{MU}^{}`$ we have the following isomorphism
$$(\widehat{MU}/\mathrm{}^\nu )^{}=(\widehat{MU/\mathrm{}^\nu }^{\mathrm{}})^{}/\mathrm{}^\nu _{}MU^{}.$$
Proof Since $`\pi _{}(MU)`$ and $`MU^{}`$ have no torsion, the assertions follow from the previous exact sequences and Proposition 4.4. $`\mathrm{}`$
For the Eilenberg-MacLane spectra recall the construction of the spaces $`K(\pi ,n)`$ in the previous section. We have the following results.
###### Proposition 4.8
Let $`\pi `$ be an abelian pro-$`\mathrm{}`$-group.
1. The Eilenberg-MacLane spaces $`K(\pi ,n)`$ represent continuous cohomology in $`\widehat{}`$, i.e. for every profinite space $`X`$ and every $`n0`$ there is an isomorphism
$$H^n(X;\pi )\mathrm{Hom}_\widehat{}(X,K(\pi ,n)).$$
2. The Eilenberg-MacLane spectrum $`H\pi =H(\pi )`$ represents continuous cohomology with coefficients in $`\pi `$, i.e. $`[\mathrm{\Sigma }^{\mathrm{}}(X),H\pi [n]]H^n(X;\pi )`$ for every $`n`$ and every $`X\widehat{𝒮}`$.
3. Let $`Y`$ be a simplicial set. We denote by $`|\pi |`$ the underlying abstract group. There is a natural isomorphism
$$H^{}(Y;|\pi |)H^{}(\widehat{Y};\pi ).$$
In particular, if $`G`$ is a finitely generated abelian group, there are isomorphisms
(13)
$$H^{}(Y;_{\mathrm{}}_{}G)H^{}(\widehat{Y};\widehat{G}^{\mathrm{}})$$
and
(14)
$$H^{}(Y;/\mathrm{}^\nu _{}G)H^{}(\widehat{Y};/\mathrm{}^\nu _{}G)$$
for every $`\nu `$.
Proof 1. Since every space in $`\widehat{𝒮}`$ is cofibrant and since the spaces $`K(\pi ,n)`$ are fibrant if $`\pi `$ is an abelian pro-$`\mathrm{}`$-group, cf. \[Mo2\], §1.4, Lemme 2, we have
$$\mathrm{Hom}_{\mathrm{Ho}(\widehat{𝒮})}(X,K(\pi ,n))\mathrm{Hom}_{\widehat{𝒮}}(X,K(\pi ,n))/$$
where $``$ denotes simplicial homotopy. Then the proof of the analogue result \[Ma1\], Theorem 24.4, works here as well.
2. We know that $`K(\pi ,n)\widehat{\mathrm{\Omega }}K(\pi ,n+1)`$ is a homotopy equivalence, hence it is a $`/\mathrm{}`$-equivalence and $`H\pi `$ is an $`\widehat{\mathrm{\Omega }}`$-spectrum. From (10) and $`H^n(X;\pi )\mathrm{Hom}_\widehat{}(X,K(\pi ,n))`$ the assertion follows.
3. Since $`\pi `$ is an abelian pro-$`\mathrm{}`$-group, $`K(\pi ,n)`$ is a fibrant profinite space for each $`n`$. Hence by adjunction we have natural isomorphisms $`\mathrm{Hom}_{}(Y,|K(\pi ,n)|)\mathrm{Hom}_\widehat{}(\widehat{Y},K(\pi ,n))`$. But by the construction of Eilenberg-MacLane spaces we have $`|K(\pi ,n)|K(|\pi |,n)`$ in $`𝒮`$ and we deduce the desired result from the first point above. $`\mathrm{}`$
###### Remark 4.9
1. The third statement (13) of the proposition is a useful generalization of Remark 2.1. The case (14) is in fact a special case of (13), but also of Remark 2.1 with $`\pi =/\mathrm{}^\nu _{}G`$, since this is a finite abelian group if $`G`$ is finitely generated abelian.
2. Since the model structures on $`\widehat{𝒮}`$ and $`\mathrm{Sp}(\widehat{𝒮}_{})`$ depend on the chosen prime number $`\mathrm{}`$, it is clear that one cannot represent continuous cohomology with arbitrary coeffiecients in $`\widehat{𝒮}`$. The homotopy groups of profinite spaces are pro-$`\mathrm{}`$-groups and hence only cohomology with pro-$`\mathrm{}`$-coefficients is representable in $`\widehat{𝒮}`$.
3. It follows from the proposition that the mod $`\mathrm{}`$-Steenrod algebra of continuous cohomology operations is identical to the usual one.
### 4.3 Postnikov decomposition
We show the existence of a Postnikov decomposition for every profinite spectrum. Although some books use the existence of arbitrary colimits in the usual category of spectra for this purpose, there is another way to do so without referring to colimits. This enables us to use this construction in our profinite setting.
First we note that by general nonsense on pointed model categories we get the existence of fiber and cofiber sequences in $`\mathrm{Sp}(\widehat{𝒮}_{})`$. We can construct for every profinite spectrum $`E`$ a connective covering $`E_0E`$ which induces an isomorphism $`\pi _k(E_n)=\pi _k(E)`$ for all $`k0`$ and with $`\pi _k(E_0)=0`$ for all $`k<0`$. We may construct it by considering the diagram
(15)
$$\begin{array}{ccc}& & P(H\pi _0(E)[1])\\ & & \\ E& & H\pi _0(E)[1]\end{array}$$
where $`H\pi `$ denotes the Eilenberg-MacLane spectrum for a profinite abelian group $`\pi `$ and $`P(H(\pi _0(E)[1])`$ is the path object of $`H\pi _0(E)[1]`$. We define $`E_0`$ to be the fiber product of this diagram. Then we obtain inductively spectra $`E_n`$ and maps $`EE_n`$ for every $`n`$ and we call $`E_n`$ the Postnikov $`n`$-truncation of $`E`$. It has the property $`\pi _k(E_n)=\pi _k(E)`$ for $`kn`$ and $`\pi _k(E_n)=0`$ for $`k>n`$. We get in fact a Postnikov tower $`\mathrm{}E_nE_{n1}\mathrm{}`$ for every $`E`$. The cofibre of each morphism $`E_nE_{n1}`$ is identified with the Eilenberg-MacLane spectrum $`H(\pi _n(E))[n+1]`$.
Again by general nonsense \[Qu1\] I,§3, we know that the morphism $`EE_{n1}`$ fits into a fiber sequence $`E_nEE_{n1}`$ and we call the spectrum $`E_n`$ the $`n`$-connective covering of $`E`$ which has the property $`\pi _k(E_n)=\pi _k(E)`$ for $`kn`$ and $`\pi _k(E_n)=0`$ for $`k<n`$.
### 4.4 Atiyah-Hirzebruch spectral sequence
Let $`X`$ be a profinite space. For every integer $`p`$ we denote by $`\mathrm{sk}_pX`$ the profinite subspace of $`X`$ which is generated by the simplices of degree less or equal $`p`$. We call $`\mathrm{sk}_pX`$ the $`p`$-skeleton of $`X`$. For every $`k`$ the set of $`k`$-simplices of $`\mathrm{sk}_pX`$ is closed in $`X_k`$ such that $`\mathrm{sk}_pX`$ is simplicial profinite subset of $`X`$, cf. \[De\]. In addition, $`\mathrm{sk}_pX`$ is equal to the colimit of the diagram
(16)
$$\begin{array}{ccc}X_p\times \mathrm{sk}_{n1}\mathrm{\Delta }[p]& & \mathrm{sk}_{p1}X\\ & & \\ X_p\times \mathrm{\Delta }[p]& & \mathrm{sk}_pX\end{array}$$
which is induced by the obvious map $`X_p\times \mathrm{\Delta }[p]X`$, where we consider $`X_p`$ as a constant simplicial profinite set. This description implies that we get morphisms $`\mathrm{sk}_{p1}X\mathrm{sk}_pX`$, which are cofibrations; and it implies that $`X`$ is equal to the colimit in $`\widehat{𝒮}`$ of the sequence $`\mathrm{}\mathrm{sk}_{p1}X\mathrm{sk}_pX\mathrm{}`$ One should note that if $`X/`$ denotes the diagram of the simplicial finite quotients of $`X`$, then the limit of the diagram $`\mathrm{sk}_p(X/)`$ is equal to $`\mathrm{sk}_pX`$. With this knowledge in hands we may continue to construct in the classical way the Atiyah-Hirzebruch spectral sequence for profinite cohomology theories. In the following we will write $`X^p`$ for $`\mathrm{sk}_pX`$.
Let $`\pi `$ be an abelian profinite group. We define the cochain complex
$$D^p(X,\pi ):=\stackrel{~}{H}^p(X^p/X^{p1};\pi )$$
with differential $`d:D^p(X;\pi )D^{p+1}(X;\pi )`$ defined as the composite
$$\stackrel{~}{H}^p(X^p/X^{p1};\pi )H^{p+1}(X^{p+1};\pi )\stackrel{~}{H}^{p+1}(X^{p+1}/X^p;\pi ).$$
One verifies easily that this defines a cochain complex. For the construction of the profinite Atiyah-Hirzebruch spectral sequence the following result will be important in order to identify the $`E_2^{p,q}`$-term.
###### Proposition 4.10
For every profinite space $`X`$, the cohomology of $`X`$ is given by the above cochain complex, i.e.
$$H(D^{}(X;\pi ),d)H^{}(X;\pi ).$$
Proof The proof is essentially the same as for CW-complexes. We consider the exact couple arising from the long exact sequence for the pair $`(X^p,X^{p1})`$:
(17)
$$\begin{array}{ccc}H^{p+q}(X^{p1};\pi )& & H^{p+q}(X^p;\pi )\\ & & \\ & \stackrel{~}{H}^{p+q}(X^p/X^{p1};\pi ).& \end{array}$$
Via Diagram (16) we see that $`X^p/X^{p1}`$ is isomorphic to a limit of joints of $`p`$-spheres and that hence $`\stackrel{~}{H}^{p+q}(X^p/X^{p1};\pi )=0`$ for $`q>0`$. Since $`d_1:E_1^{p,0}E_1^{p+1,0}`$ is equal to $`d:D^p(X;\pi )D^{p+1}(X;\pi )`$ we have
$$E_2^{,0}=E_{\mathrm{}}^{p,0}=H(D^{}(X;\pi ),d).$$
From the definition of continuous cohomology for profinite spaces and from $`(\mathrm{sk}_{p+r}X)_p=X_{p+r}`$ for $`r1`$, it is clear that
$$H^p(X^{p+r};\pi )=H^p(X;\pi )\mathrm{for}r1.$$
By induction on $`p`$, we show in addition that
$$H^{p+r}(X^p;\pi )=0\mathrm{for}r1.$$
This is clear for $`p=0`$. Using (16) we get the exact sequence
$$0=\stackrel{~}{H}^{p+r}(X^p/X^{p1};\pi )\stackrel{~}{H}^{p+r}(X^p;\pi )\stackrel{~}{H}^{p+r}(X^{p1};\pi )\stackrel{~}{H}^{p+r+1}(X^p/X^{p1};\pi )=0.$$
By induction, $`H^{p+r}(X^{p1};\pi )=0`$ for $`r1`$ and, by exactness, $`H^{p+r}(X^p;\pi )=0`$ for $`r1`$, too. One continues just as in topology to conclude the desired isomorphism $`E_2^{,0}H^{}(X;\pi )`$, see e.g. \[Mc\], Theorem 4.11. $`\mathrm{}`$
Let $`E`$ be a profinite spectrum. We want to construct an Atiyah-Hirzebruch spectral sequence for $`E`$. We consider the skeletal filtration $`\mathrm{sk}_0X\mathrm{}\mathrm{sk}_pX\mathrm{sk}_{p+1}X\mathrm{}X`$. This filtration yields a filtration on $`E^{}X`$ defined by $`F^pE^{}X:=\mathrm{Ker}(E^{}XE^{}X^{p1})`$.
The proof of the following theorem is essentially the one given by Adams in \[Ad\]. We use the special consideration of the profinite setting by Dehon in \[De\], Proposition 2.1.9. But we are able to prove a slightly more general statement than Dehon for any profinite spectrum.
By Remark 3.9 we know that the homotopy groups $`\pi _qE`$ have a natural profinite structure and hence also the coefficient groups $`E^q`$. Thus we may consider continuous cohomology with coefficients in $`E^q`$.
###### Theorem 4.11
For any profinite spectrum $`E`$ and for any profinite space $`X`$ there is a spectral sequence $`\{E_r^{p,q}\}`$ with $`E_2^{p,q}H^p(X;E^q)E^{p+q}(X)`$. The spectral sequence converges strongly, in the sense of \[Ad\] III, §8.2, to the graded term $`e_{\mathrm{}}^{p,q}=F^pE^{}X/F^{p+1}E^{}X`$ of the filtration on $`E^{}X`$ if $`lim_{r}^{}{}_{}{}^{1}E_r^{p,q}=0`$. In particular, the spectral sequence converges if $`H^p(X;E^q)=0`$ for $`p`$ sufficiently large or if $`H^p(X;E^q)`$ is finite for all $`p`$. We call it the profinite Atiyah-Hirzebruch spectral sequence.
Proof The construction of the spectral sequence is a standard argument. The cofibre sequence $`X_+^{p1}X_+^pX_+^p/X_+^{p1}`$ induces an exact couple in cohomology
(18)
$$\begin{array}{ccc}\underset{p}{}E^{}(X^{p1})& & \underset{p}{}E^{}(X^p)\\ & & \\ & \underset{p}{}E^{}(X^p/X^{p1})& \end{array}$$
which leads to a spectral sequence with $`E_1^{p,q}=E^{p+q}(X^p/X^{p1})E^{p+q}(X)`$, the differentials of which are of bidegree $`(r,1r)`$. The convergence properties are described in and follow from \[Ad\] III, §8.2 or \[Fr1\], Proposition A.2. In particular, if $`H^p(X;E^q)=0`$ for $`pn+1`$, the spectral sequence degenerates at the $`E_{n+1}`$-term and $`_pE_{n+1}^{p,}`$ is isomorphic to the quotient $`F^pE^{}X/F^{p+1}E^{}X`$. Hence it remains to identify the $`E_2`$-term.
Claim: We have a natural isomorphism
$$E_1^{p,q}=E^{p+q}(X^p/X^{p1})\stackrel{~}{H}^p(X^p/X^{p1};\pi _pE_{p+q}).$$
If we can prove this claim, we get, together with Proposition 4.10 and the fact $`\pi _pE_{p+q}=E^q`$, that $`E_2^{p,q}H^p(X;E^q)`$ and we are done.
Hence it remains to prove the claim. In the following we suppose that $`E`$ is a fibrant profinite spectrum, i.e. an $`\widehat{\mathrm{\Omega }}`$-spectrum. This identification is due to \[De\]. We follow his argumentation. By the characterizing diagram (16) of the $`p`$th-skeleton $`X^p`$ of $`X`$ we see that $`X^p/X^{p1}lim_{F𝒬(X_p)}(_FS^p)`$ is a limit in $`\widehat{𝒮}_{}`$ of joints of $`p`$-spheres. The limit comes from the profinite structure of $`X_p`$; $`F`$ runs through the finite sets equal to quotients of $`X_p`$. Since $`E`$ is fibrant we can identify $`E_{p+q}`$ with the limit $`lim_tE_{p+q}(t)`$ where $`E_{p+q}(t)`$ is a finite $`\mathrm{}`$-space for every $`t`$, i.e. $`\pi _iE_{p+q}(t)`$ is a finite $`\mathrm{}`$-group for every $`i`$ and $`t`$ and $`\pi _iE_{p+q}(t)0`$ only for a finite number of $`i`$. Since $`_FS^p`$ is a simplicial finite set and since $`\pi _iE_{p+q}(t)`$ is finite, we conclude that we have an isomorphism
$$\underset{t}{lim}\mathrm{colim}_{F𝒬(X_p)}\stackrel{~}{H}^p(_FS^p;\pi _pE_{p+q}(t))\stackrel{~}{H}^p(X^p/X^{p1};\pi _pE_{p+q}).$$
By \[Swi\], Proposition 6.39, we have
$$\begin{array}{ccc}\hfill \stackrel{~}{H}^p(_FS^p;\pi _pE_{p+q}(t))& & \mathrm{Hom}_\widehat{}_{}(_FS^p,K(\pi _pE_{p+q}(t),p))\hfill \\ & & \mathrm{Hom}_{\mathrm{Ab}}(\pi _p(_FS^p),\pi _pE_{p+q}(t)).\hfill \end{array}$$
Hence we get an isomorphism
$$\stackrel{~}{H}^p(X^p/X^{p1};\pi _pE_{p+q})\mathrm{Hom}_{\mathrm{pro}\mathrm{Ab}}(\pi _p(_{}S^p),\pi _pE_{p+q}()).$$
On the other hand we have the natural map
$$\begin{array}{ccc}\hfill E^{p+q}(X^p/X^{p1})& =& \mathrm{Hom}_\widehat{}_{}(X^p/X^{p1},E_{p+q})\hfill \\ & & \mathrm{Hom}_{\mathrm{pro}_{}}(X^p/X^{p1},E_{p+q}())\hfill \\ & & \mathrm{Hom}_{\mathrm{pro}\mathrm{Ab}}(\pi _p(_{}S^p),\pi _pE_{p+q}()).\hfill \end{array}$$
Hence in order to prove the claim, it suffices to show the following lemma.
###### Lemma 4.12
Cf. Lemma 2.1.10 of \[De\]. Let $`S()`$ be a filtered diagram of finite joints of $`p`$-spheres with limit $`S`$ in $`\widehat{𝒮}_{}`$ and let $`E`$ be a fibrant profinite spectrum. Then the composite
$$\mathrm{Hom}_\widehat{}_{}(S,E_{p+q})\mathrm{Hom}_{\mathrm{pro}_{}}(S(),E_{p+q}())\mathrm{Hom}_{\mathrm{pro}\mathrm{Ab}}(\pi _p(_{}S^p),\pi _pE_{p+q}())$$
is an isomorphism of abelian groups.
Proof The proof is the same as for Lemma 2.1.10 of \[De\]. For the sake of completeness, we reformulate it here in more detail. For every pair of indices $`F`$ and $`t`$ we have a sequence of isomorphisms using the facts that $`\pi _pS^p`$, $`S^p`$ considered in $`𝒮_{}`$, and that $`E_{p+q}(t)`$ is fibrant:
$$\begin{array}{ccc}\hfill \mathrm{Hom}_{_{}}(_FS^p,E_{p+q}(t))& & _F\mathrm{Hom}_{_{}}(S^p,E_{p+q}(t))\hfill \\ & & _F\pi _p(E_{p+q}(t))_F\mathrm{Hom}_{\mathrm{Ab}}(\pi _p(S^p),\pi _p(E_{p+q}(t)))\hfill \\ & & \mathrm{Hom}_{\mathrm{Ab}}(\pi _p(_FS^p),\pi _p(E_{p+q}(t)))\hfill \end{array}$$
from which we deduce that the map
$$\mathrm{Hom}_{\mathrm{pro}_{}}(_{}S^p,E_{p+q}())\mathrm{Hom}_{\mathrm{pro}\mathrm{Ab}}(\pi _p(_{}S^p),\pi _p(E_{p+q}()))$$
is an isomorphism. Hence it remains to show that the left hand side is isomorphic to $`\mathrm{Hom}_\widehat{}_{}(lim_{F𝒬(X_p)}(_FS^p),E_{p+q})`$. Since $`E_{p+q}(t)`$ is a finite $`\mathrm{}`$-space for every $`t`$ and since $`_FS^p`$ is a simplicial finite set, Proposition 1.3.2 of \[De\] implies that
$$\mathrm{colim}_{F𝒬(X_p)}\mathrm{Hom}_{_{}}(_FS^p,E_{p+q}(t))\stackrel{}{}\mathrm{Hom}_\widehat{}_{}(\underset{F𝒬(X_p)}{lim}_FS^p,E_{p+q}(t))$$
is an isomorphism for every $`t`$. Furthermore, Corollary 1.4.3 of \[De\] implies that
$$\mathrm{Hom}_\widehat{}_{}(\underset{F𝒬(X_p)}{lim}_FS^p,\underset{t}{lim}E_{p+q}(t))\underset{t}{lim}\mathrm{Hom}_\widehat{}_{}(\underset{F𝒬(X_p)}{lim}_FS^p,E_{p+q}(t))$$
is an isomorphism if $`lim_{t}^{}{}_{}{}^{1}\pi _1(\mathrm{hom}_{}(lim_{F𝒬(X_p)}_FS^p,E_{p+q}(t)))`$ vanishes. If we could show that $`lim_{t}^{}{}_{}{}^{1}\pi _1(\mathrm{hom}_{}(lim_{F𝒬(X_p)}_FS^p,E_{p+q}(t)))=0`$ we would get
$$\underset{t}{lim}\mathrm{colim}_{F𝒬(X_p)}\mathrm{Hom}_{_{}}(_FS^p,E_{p+q}(t))\mathrm{Hom}_\widehat{}_{}(\underset{F𝒬(X_p)}{lim}_FS^p,E_{p+q}(t)).$$
Hence we have to show that $`lim_{t}^{}{}_{}{}^{1}\pi _1(\mathrm{hom}_{}(lim_{F𝒬(X_p)}_FS^p,E_{p+q}(t)))`$ vanishes. Again since $`_FS^p`$ is a simplicial finite set and since $`E_{p+q}(t)`$ is fibrant we get for every $`t`$
$$\begin{array}{ccc}\hfill \pi _1(\mathrm{hom}_{}(lim_{F𝒬(X_p)}_FS^p,E_{p+q}(t)))& & \mathrm{colim}_{F𝒬(X_p)}\pi _1(\mathrm{hom}_{}(_FS^p,E_{p+q}(t))\hfill \\ & & \mathrm{colim}_{F𝒬(X_p)}\pi _1(_F\mathrm{\Omega }^p(E_{p+q}(t)))\hfill \\ & & \mathrm{colim}_{F𝒬(X_p)}_F\pi _{p+1}(E_{p+q}(t)).\hfill \end{array}$$
Since $`\pi _{p+1}E_{p+q}(t)`$ is a finite group, the pro-abelian group $`\{\mathrm{colim}_{F𝒬(X_p)}_F\pi _{p+1}(E_{p+q}(t))\}_t`$ is pro-isomorphic to a tower of surjections. Hence the $`lim^1`$-term of the tower $`\{\pi _1(\mathrm{hom}_{}(lim_{F𝒬(X_p)}_FS^p,E_{p+q}(t))\}_t`$ vanishes. $`\mathrm{}`$ $`\mathrm{}`$
### 4.5 A Künneth formula for profinite cobordism
We finish this section with a discussion of a Künneth isomorphism in $`\widehat{MU}/\mathrm{}^\nu `$ for the product $`X\times Y`$ in $`\widehat{𝒮}`$ for profinite spaces $`X`$ and $`Y`$ satisfying certain conditions. The whole argumentation is inspired by the work of Dehon \[De\], where he proves the corresponding result for $`\widehat{MU}`$. We say that $`Y\widehat{𝒮}`$ is without $`\mathrm{}`$-torsion if the reduction map $`/\mathrm{}^n/\mathrm{}`$ induces a surjection $`H^{}(Y;/\mathrm{}^n)H^{}(Y;/\mathrm{})`$ for every positive $`n`$, cf. \[De\], Definition 2.1.5. This is equivalent to the fact that $`H^{}(Y;_{\mathrm{}})`$ has no torsion. For the application to étale cohomology theories in later sections, we only need the following version of a Künneth isomorphism for finite dimensional profinite spaces.
###### Theorem 4.13
Let $`X`$ and $`Y`$ be finite dimensional profinite spaces and let $`X`$ be without $`\mathrm{}`$-torsion. Then we have for every $`\nu `$ Künneth isomorphisms
$$\widehat{MU}^{}(X)_{\widehat{MU}^{}}\widehat{MU}^{}(Y)\stackrel{}{}\widehat{MU}^{}(X\times Y)$$
and
$$\widehat{MU}^{}(X;/\mathrm{}^\nu )_{(\widehat{MU}/\mathrm{}^\nu )^{}}\widehat{MU}^{}(Y;/\mathrm{}^\nu )\stackrel{}{}\widehat{MU}^{}(X\times Y;/\mathrm{}^\nu ).$$
Proof The first assertion is a restricted version of \[De\], Proposition 4.3.2.
The part for $`/\mathrm{}^\nu `$-coefficients follows in a similar way. I am very grateful to Francois-Xavier Dehon for an explanation of this point. We will prove the assertion following his argument.
We suppose first that $`X`$ and $`Y`$ are finite dimensional without $`\mathrm{}`$-torsion. The point is that the morphism of coefficients $`(\widehat{MU}/\mathrm{}^\nu )^{}(HMU/\mathrm{}^\nu )^{}`$ is injective and, since $`X`$ has no $`\mathrm{}`$-torsion, the map $`E_2^,(X;MU/\mathrm{}^\nu )E_2^,(X;HMU/\mathrm{}^\nu )`$ is also injective. Since the spectral sequence for $`HMU/\mathrm{}^\nu `$ degenerates at the $`E_2`$-stage, the same is true for the spectral sequence for $`MU/\mathrm{}^\nu `$. One can deduce from this fact that we have an isomorphism $`\widehat{MU}^{}(X;/\mathrm{}^\nu )/f^1H^{}(X;/\mathrm{})`$ as described in the proof of \[De\], Proposition 2.1.8, where $`f^{}`$ is an $`\mathrm{}`$-adic filtration on $`MU^{}`$-modules defined in \[De\], §2.1.
Since $`Y`$ and $`X\times Y`$ have no $`\mathrm{}`$-torsion, the same argument is valid for them as well. By Corollaire 2.1.4 of \[De\], it suffices to check that the map
$$\widehat{MU}^{}(X;/\mathrm{}^\nu )/f^1_{(\widehat{MU}/\mathrm{}^\nu )^{}}\widehat{MU}^{}(Y;/\mathrm{}^\nu )/f^1\widehat{MU}^{}(X\times Y;/\mathrm{}^\nu )/f^1$$
is an isomorphism. But this follows from the previous discussion and the Künneth isomorphism for $`/\mathrm{}`$-cohomology. This finishes the case that both $`X`$ and $`Y`$ have no $`\mathrm{}`$-torsion.
For the case that $`Y`$ has $`\mathrm{}`$-torsion, we use induction on the skeletal filtration of $`Y`$. Let $`Y^n`$ be the $`n`$-th skeleton of $`Y`$. If $`\widehat{MU}^{}(X;/\mathrm{}^\nu )_{(\widehat{MU}/\mathrm{}^\nu )^{}}\widehat{MU}^{}(Z;/\mathrm{}^\nu )\widehat{MU}^{}(X\times Z;/\mathrm{}^\nu )`$ is an isomorphism for $`Z=Y^n`$ and for $`Z=Y^{n+1}/Y^n`$, then it is also an isomorphism for $`Z=Y^{n+1}`$. This follows from the long exact sequence associated to the cofibre sequence $`Y^nY^{n+1}Y^n/Y^{n+1}`$ and the exactness of the tensor product with $`\widehat{MU}^{}(X;/\mathrm{}^\nu )`$. The exactness of $`\widehat{MU}^{}(X;/\mathrm{}^\nu )`$ is due to the fact that $`X`$ has no $`\mathrm{}`$-torsion. We deduce the assertion of the theorem by induction on $`n`$ from the proof of Proposition 4.3.2 of \[De\] using the point that we consider only the $`n`$-th skeleton of $`Y`$ instead of the whole space. $`\mathrm{}`$
###### Remark 4.14
There is a version of the previous theorem with the assumption that $`Y`$ is an arbitrary profinite space and $`X`$ is a profinite space without $`\mathrm{}`$-torsion whose $`/\mathrm{}`$-cohomology is finite in each degree.
One proves the $`/\mathrm{}^\nu `$-coefficient case by taking the limit over the skeletal filtration.
### 4.6 Comparison with generalized cohomology theories of pro-spectra
Isaksen constructs in \[I5\] a stable model structure on the category of pro-spectra and defines generalized cohomology theories of pro-spectra. If $`E`$ is a pro-spectrum then the $`r`$-th cohomology $`E_{\mathrm{pro}}^r(X)`$ of a pro-spectrum $`X`$ with coefficients in $`E`$ is the set $`[X,E]_{\mathrm{pro}}^r`$ maps that lower the degree by $`r`$ in the stable homotopy category of pro-spectra. In addition, for two pro-spectra $`X`$ and $`E`$ he shows the existence of an Atiyah-Hirzebruch spectral sequence $`E_2^{p,q}=H^p(X;\pi _qE)[X,E]_{\mathrm{pro}}^{p+q}`$ which is in particular convergent if $`E`$ is a constant pro-spectrum and $`X`$ is the suspension spectrum of a finite dimensional pro-space. We show that this definition of cohomology of pro-spectra gives in the cases of interest the same groups as our definition via profinite spectra.
First we construct a functor $`\widehat{()}:\mathrm{pro}\mathrm{Sp}(𝒮)\mathrm{Sp}(\widehat{𝒮}_{})`$ that induces morphisms on cohomology theories. As for pro-spaces we define $`\widehat{()}`$ to be the composition of the profinite completion on spectra followed by taking the homotopy limit in $`\mathrm{Sp}(\widehat{𝒮}_{})`$ of the corresponding diagram in $`\mathrm{Sp}(\widehat{𝒮}_{})`$. Note that the homotopy limit exists by \[Hi\], Chapter 18, keeping in mind that all limits exist, see Convention A.1 on limits. Since the completion commutes with suspension this functor obviously agrees with our previous definition for the case of the suspension spectrum of a space. A map $`f:EF`$ in $`\mathrm{pro}\mathrm{Sp}(𝒮)`$ is a weak equivalence if $`f`$ is a levelwise $`n`$-equivalence for all $`n`$. This implies that $`f`$ induces an isomorphism $`\pi _kf`$ on pro-homotopy groups, cf. \[I5\], Theorem 8.4. We want to show that $`\widehat{()}`$ sends a weak equivalence $`f`$ to a stable equivalence in $`\mathrm{Sp}(\widehat{𝒮}_{})`$. For this we may suppose that $`f`$ is a level map $`\{f_s:E_sF_s\}`$ and that $`\pi _kf_s`$ is an isomorphism of stable homotopy groups for every $`k`$ and every $`s`$. We know that the completion $`\widehat{()}:\mathrm{Sp}(𝒮)\mathrm{Sp}(\widehat{𝒮}_{})`$ sends stable equivalences to stable equivalences. Since the homotopy limit is by construction well behaved with respect to levelwise weak equivalences $`\mathrm{holim}_s\widehat{f_s}`$ is also a stable equivalence. Hence the functor induces maps $`\mathrm{Hom}_{\mathrm{Ho}(\mathrm{pro}\mathrm{Sp}(𝒮))}(X,E)\mathrm{Hom}_{\widehat{𝒮}}(\widehat{X},\widehat{E})`$ and hence also maps $`E_{\mathrm{pro}}^{}(X)\widehat{E}^{}(\widehat{X})`$. These maps yield morphisms of Atiyah-Hirzebruch spectral sequences.
###### Theorem 4.15
For any $`k`$, consider the constant pro-spectrum $`MU/\mathrm{}^\nu `$. Let $`\{X_s\}`$ be a finite dimensional pro-space. Then we have an isomorphism
$$MU_{\mathrm{pro}}^{}(\{X_s\};/\mathrm{}^\nu )\widehat{MU}^{}(\widehat{X};/\mathrm{}^\nu ).$$
Proof As pointed out above the completion functor induces a map of spectral sequences. But for a constant pro-spectrum the completion is just the completion defined in Section 3 on profinite spectra.
Hence our previous results on the coefficients of $`MU/\mathrm{}^\nu `$ and $`\widehat{MU}/\mathrm{}^\nu `$ imply that it induces an isomorphism of coefficients $`(MU/\mathrm{}^\nu )^{}(\widehat{MU}/\mathrm{}^\nu )^{}`$ and hence an isomorphism of $`E_2`$-terms. Since both spectral sequences converge for the given $`X`$, the abutments are also isomorphic. $`\mathrm{}`$
## 5 Profinite étale realization on the unstable $`𝔸^1`$-homotopy category
The construction of the étale topological type functor $`\mathrm{Et}`$ from locally noetherian schemes to pro-simplicial sets is due to Artin-Mazur and Friedlander. The construction of the $`𝔸^1`$-homotopy category of schemes makes naturally arise the question if one could enlarge this functor to the category of spaces and if this functor behaves well with respect to the model structure. These questions have been answered by Isaksen and Schmidt. The first step into this direction was the construction of a model structure on $`\mathrm{pro}𝒮`$. In view of $`\mathrm{Et}`$, there are in fact at least two interesting structures. It turns out that the $`/\mathrm{}`$-cohomological model structure of $`\mathrm{pro}𝒮`$ fits better with the $`𝔸^1`$-localized homotopy category. This is due to the fact that over a field of positive characteristic $`p>0`$, the étale fundamental group of $`𝔸^1`$ is highly non-trivial. To avoid this obstruction one has to complete away from the characteristic of the ground field using étale cohomology. The projection $`X\times 𝔸^1X`$ induces an isomorphism in étale cohomology $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{})H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X\times 𝔸^1;/\mathrm{})`$ for every prime $`\mathrm{}p`$.
We will adapt the constructions of Isaksen \[I3\] and Dugger \[D\] to $`\widehat{𝒮}`$ and check that we still get an induced left derived functor on the homotopy categories. We will calculate $`\widehat{\mathrm{Et}}`$ for some examples.
### 5.1 The functor $`\widehat{\mathrm{Et}}`$
###### Definition 5.1
Definition 4.4 of \[Fr2\].
Let $`X_{}`$ be a locally noetherian simplicial scheme. The étale topological type of $`X_{}`$ is defined to be the following pro-simplicial set
$$\mathrm{Et}X=\pi \mathrm{\Delta }:HRR(X_{})𝒮$$
sending a hypercovering $`U_,`$ of $`X_{}`$ to the simplicial set of connected components of the diagonal of $`U_,`$. If $`f:X_{}Y_{}`$ is a map of locally noetherian simplicial schemes, then the strict étale topological type of $`f`$ is the strict map $`\mathrm{Et}f:\mathrm{Et}X_{}\mathrm{Et}Y_{}`$ given by the functor $`f^{}:HRR(Y_{})HRR(X_{})`$ and the natural transformation $`\mathrm{Et}X_{}\mathrm{Et}f\mathrm{Et}Y_{}`$.
We refer the reader to \[Fr2\] and \[I3\] for a detailed discussion of the category $`HRR(X_{})`$ of rigid hypercoverings of $`X_{}`$ and rigid pullbacks.
Isaksen now uses the insight of Dugger \[D\] that one can construct the $`𝔸^1`$-homotopy category in a universal way. Starting from an almost arbitrary category $`𝒞`$, Dugger constructs an enlargement of $`𝒞`$ that carries a model structure and is universal for this property. He also shows how to enlarge functors $`𝒞`$ from $`𝒞`$ to a model category $``$. In \[I3\], Isaksen takes Dugger’s model and extends the functor $`\mathrm{Et}`$ via this general method to the $`𝔸^1`$-homotopy category. The idea is that $`\mathrm{Et}X`$ should be the above $`\mathrm{Et}X`$ on a representable presheaf $`X`$ and should preserve colimits and the simplicial structure.
For our purpose, we would like to define $`\widehat{\mathrm{Et}}`$ directly in the way Dugger suggests. But the problem is that $`\widehat{()}:\mathrm{pro}𝒮\widehat{𝒮}`$ is not a left adjoint functor and does not preserve all small colimits. Therefore, we define $`\widehat{\mathrm{Et}}`$ to be the composition of $`\mathrm{Et}`$ followed by completion. To be precise we make the following definition:
###### Definition 5.2
If $`X`$ is a representable presheaf, then $`\mathrm{Et}X`$ is the étale topological type of $`X`$. If $`P`$ is a discrete presheaf, i.e. each simplicial set $`P(U)`$ is $`0`$-dimensional, i.e. $`P`$ is just a presheaf of sets, then $`P`$ can be written as a colimit $`\mathrm{colim}_iX_i`$ of representables and we define $`\mathrm{Et}P:=\mathrm{colim}_i\mathrm{Et}X_i`$. Finally, an arbitrary simplicial presheaf can be written as the coequalizer of the diagram
$$\underset{[m][n]}{}P_m\mathrm{\Delta }^n\underset{[n]}{}P_n\mathrm{\Delta }^n,$$
where each $`P_n`$ is discrete. Define $`\mathrm{Et}P`$ to be the coequalizer of the diagram
$$\underset{[m][n]}{}\mathrm{Et}P_m\mathrm{\Delta }^n\underset{[n]}{}\mathrm{Et}P_n\mathrm{\Delta }^n.$$
We define the profinite étale topological type functor $`\widehat{\mathrm{Et}}`$ to be the composition of $`\mathrm{Et}`$ and the profinite completion functor $`\mathrm{pro}𝒮\widehat{𝒮}`$:
(19)
$$\widehat{\mathrm{Et}}:=\widehat{()}\mathrm{Et}:\mathrm{\Delta }^{\mathrm{op}}\mathrm{PreShv}(\mathrm{Sm}/S)\widehat{𝒮}.$$
For computations the following remark is crucial.
###### Remark 5.3
1. By \[Fr2\], Proposition 5.9, we know that $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;M)H^{}(\mathrm{Et}X,M)`$ where $`H^{}(Z;M)`$ denotes the cohomology of a pro-simplicial set $`Z`$ with coefficients in the local coefficient system $`M`$ corresponding to the sheaf $`M`$. For a finite abelian group $`\pi `$, we have in addition a natural isomorphism by Lemma 2.12: $`H^{}(Z;\pi )H^{}(\widehat{Z};\pi )`$ for every pro-simplicial set $`Z`$. Hence we get as well
$$H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;\pi )H^{}(\widehat{\mathrm{Et}}X,\pi )$$
for every locally noetherian scheme $`X`$ and every finite abelian group $`\pi `$.
2. For a morphism $`g:ZX`$ of schemes over $`k`$, there are relative cohomology groups $`H^{}(\widehat{\mathrm{Et}}X,\widehat{\mathrm{Et}}Z;/\mathrm{})`$ fitting in a natural long exact sequence
$$\mathrm{}H^{}(\widehat{\mathrm{Et}}X,\widehat{\mathrm{Et}}Z;/\mathrm{})H^{}(\widehat{\mathrm{Et}}X;/\mathrm{})H^{}(\widehat{\mathrm{Et}}Z;g^{}/\mathrm{})\mathrm{}$$
For a closed immersion $`ZX`$ with open complement $`U`$, in \[Fr2\], Corollary 14.5, Friedlander has shown that these relative cohomology groups coincide with the étale cohomology of $`X`$ with support in $`Z`$, i.e. we have a natural isomorphism
$$H_{\stackrel{´}{\mathrm{e}}\mathrm{t},Z}^{}(X;/\mathrm{})H^{}(\widehat{\mathrm{Et}}X,\widehat{\mathrm{Et}}U;/\mathrm{}).$$
Although $`\widehat{\mathrm{Et}}`$ does not commute with products we have a weaker compatibility.
###### Proposition 5.4
Let $`X`$ and $`Y`$ be smooth schemes of finite type over a separably closed field of characteristic $`p\mathrm{}`$. Then the canonical map is a weak equivalence in $`\widehat{𝒮}`$
$$\widehat{\mathrm{Et}}(X\times Y)\stackrel{}{}\widehat{\mathrm{Et}}X\times \widehat{\mathrm{Et}}Y.$$
Proof This follows from the Künneth formula for smooth schemes over a separably closed field proved in \[SGA$`4\frac{1}{2}`$\] \[Th. finitude\], Corollaire 1.11. It provides an isomorphism
$$H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X\times Y;/\mathrm{})H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{})H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(Y;/\mathrm{}).$$
On the other hand, we have a Künneth formula for the cohomology of profinite spaces which yields an isomorphism
$$H^{}(\widehat{\mathrm{Et}}X\times \widehat{\mathrm{Et}}Y;/\mathrm{})H^{}(\widehat{\mathrm{Et}}X;/\mathrm{})H^{}(\widehat{\mathrm{Et}}Y;/\mathrm{}).$$
This implies that $`\widehat{\mathrm{Et}}(X\times Y)\widehat{\mathrm{Et}}X\times \widehat{\mathrm{Et}}Y`$is a weak equivalence in $`\widehat{𝒮}`$. $`\mathrm{}`$
In the following example, we consider objects in the category $`\mathrm{\Delta }^{\mathrm{op}}\mathrm{PreShv}(\mathrm{Sm}/k)`$ of simplicial presheaves over $`\mathrm{Sm}/k`$.
###### Example 5.5
Let $`R`$ be a strict local henselian ring, i.e. a local henselian ring with separably closed residue field. Then $`\mathrm{Spec}R`$ has no nontrivial étale covers and the étale topological type of $`\mathrm{Spec}R`$ is a contractible space, i.e. $`\widehat{\mathrm{Et}}(\mathrm{Spec}R)`$.
###### Example 5.6
Let $`k`$ be a separably closed field with $`\mathrm{char}(k)\mathrm{}`$. Let $`c:𝒮\mathrm{\Delta }^{\mathrm{op}}\mathrm{PreShv}(\mathrm{Sm}/k)`$ be the functor that sends a simplicial set $`Z`$ to the constant presheaf defined by $`Z`$. For every $`n`$, $`cZ([n])`$ is isomorphic to a disjoint union of copies of $`\mathrm{Spec}k`$. Hence, since $`\mathrm{Et}`$ commutes with coproducts and since $`\mathrm{Et}(\mathrm{Spec}k)=`$ by Example 5.5 above, its étale topological type $`\mathrm{Et}cZ`$ is just the simplicial set $`Z`$ itself.
This shows that the pro-simplicial set $`\mathrm{Et}Z`$ is in fact just a simplicial set. The completion $`\widehat{()}:𝒮\widehat{𝒮}`$ preserves all colimits. Hence the same argument holds for $`\widehat{\mathrm{Et}}`$, $`\widehat{\mathrm{Et}}(cZ)=\widehat{\mathrm{Et}cZ}`$ and $`\widehat{\mathrm{Et}}(cZ)=Z`$ for every $`Z\widehat{𝒮}`$.
In particular, if $`S^1`$ denotes the simplicial circle, then $`S^1([n])=_{S^1([n])}\mathrm{Spec}k`$ and $`(S^1𝒳)([n])=_{S^1([n])}𝒳([n])`$. This yields an isomorphism in $`\widehat{𝒮}_{}`$
(20)
$$S^1\widehat{\mathrm{Et}}(𝒳)\stackrel{}{}\widehat{\mathrm{Et}}(S^1)\widehat{\mathrm{Et}}(𝒳)\stackrel{}{}\widehat{\mathrm{Et}}(S^1𝒳).$$
Furthermore, Friedlander proved in \[Fr2\] that the étale fundamental group $`\pi _1^{\stackrel{´}{\mathrm{e}}\mathrm{t}}(X)`$ of a scheme is isomorphic as pro-group to the pro-fundamental group of $`\mathrm{Et}X`$. From Proposition 2.10 we deduce the following
###### Proposition 5.7
Let $`X`$ be a connected locally noetherian scheme. The fundamental group of $`\widehat{\mathrm{Et}}X`$ is isomorphic to the $`\mathrm{}`$-completion of the étale fundamental group of $`X`$, i.e.
$$\pi _1^{\mathrm{}}(\widehat{\mathrm{Et}}X)\widehat{\pi _1^{\stackrel{´}{\mathrm{e}}\mathrm{t}}(X)}^{\mathrm{}}.$$
###### Example 5.8
Let $`k`$ be a separably closed field with $`\mathrm{char}(k)\mathrm{}`$. Since $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(𝔸_k^1;/\mathrm{})H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k;/\mathrm{})`$ we know that $`\widehat{\mathrm{Et}}𝔸_k^1`$, i.e. that $`\widehat{\mathrm{Et}}𝔸_k^1`$ is contractible in $`\widehat{𝒮}`$ and
$$\pi _1^{\mathrm{}}(\widehat{\mathrm{Et}}𝔸_k^1)=\{1\}.$$
###### Example 5.9
Let $`k`$ be a separably closed field with $`\mathrm{char}(k)\mathrm{}`$. The space $`𝔾_m`$ is connected and its étale fundamental group is $`\widehat{}`$. Its $`\mathrm{}`$-completion is hence equal to $`_{\mathrm{}}`$ and we get $`\widehat{\mathrm{Et}}𝔾_mK(_{\mathrm{}},1)`$. Hence $`\widehat{\mathrm{Et}}𝔾_m`$ is isomorphic to $`S^1`$ in $`\widehat{𝒮}_{}`$ by (2.11).
###### Example 5.10
Let $`k`$ be a separably closed field with $`\mathrm{char}(k)\mathrm{}`$. Let $`_k^1`$ be the projective line over $`k`$. Since $`_k^1`$ is connected and $`\pi _{1,\stackrel{´}{\mathrm{e}}\mathrm{t}}^{\mathrm{}}(_k^1)=0`$ is trivial, cf. e. g. \[Mi\] I, Example 5.2 f), $`\widehat{\mathrm{Et}}_k^1`$ is a simply connected space. Apart from $`H^0`$ its only nonzero étale cohomology group is $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(_k^1,/\mathrm{})/\mathrm{}`$, see \[Mi\] VI, Example 5.6 with a chosen isomorphism $`/\mathrm{}/\mathrm{}(1)`$. Hence $`\widehat{\mathrm{Et}}_k^1`$ is isomorphic in $`\widehat{𝒮}_{}`$ to the simplicial finite set $`S^2`$.
###### Example 5.11
Let $`k`$ be a field of characteristic $`p\mathrm{}`$. The étale realization of the projective line $`_k^1`$ over $`k`$ is given by the isomorphism $`\widehat{\mathrm{Et}}_k^1\stackrel{}{}S^2\times \widehat{\mathrm{Et}}k`$ in $`\widehat{𝒮}`$. For, the projective bundle formula in étale cohomology $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^1;/\mathrm{})H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k;/\mathrm{})H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k;/\mathrm{})`$ and the Künneth formula in $`\widehat{𝒮}`$ yield the identification.
###### Example 5.12
Let $`k=𝔽_q`$ be a finite field with $`\mathrm{char}(k)=p\mathrm{}`$. The étale topological type of $`k`$ is isomorphic to $`S^1`$ in $`\widehat{𝒮}`$. For, $`\widehat{\mathrm{Et}}k`$ is a connected space by \[Fr2\] Proposition 5.2. Its $`\mathrm{}`$-completed fundamental group is the $`\mathrm{}`$-completion of the absolute Galois group of $`k`$ by Proposition 5.7, i.e. $`\pi _{1,\stackrel{´}{\mathrm{e}}\mathrm{t}}^{\mathrm{}}(k)=_{\mathrm{}}`$. Since all cohomology groups $`H^i(k;/\mathrm{})`$ vanish for $`i>1`$ by \[Se\], $`\mathrm{Et}k`$ is a space of dimension one and weakly equivalent to $`S^1`$, hence it is isomorphic to $`S^1`$ in $`\widehat{𝒮}`$.
### 5.2 Unstable profinite étale realization
We want to show that $`\widehat{\mathrm{Et}}`$ induces a functor on the homotopy category. As indicated above we choose the universal model of \[D\] for the unstable $`𝔸^1`$-homotopy category of smooth schemes over a base field $`k`$.
Dugger showed that after some localization the projective model structure on $`U(\mathrm{Sm}/k):=\mathrm{\Delta }^{\mathrm{op}}\mathrm{PreShv}(\mathrm{Sm}/k)`$ is a model for the $`𝔸^1`$-homotopy category $`_{𝔸^1}(k)`$ of \[MV\]. In this projective model structure the weak equivalences (fibrations) are objectwise weak equivalences (fibrations) of simplicial sets. The cofibrations are the maps having the left lifting property with respect to all trivial fibrations. Then one takes the left Bousfield localization of this model structure at the set $`S`$ of maps:
1. for every finite collection $`\{X^a\}`$ of schemes with disjoint union $`X`$, the map $`X^aX`$ from the coproduct of the presheaves represented by each $`X^a`$ to the presheaf represented by $`X`$;
2. every étale (Nisnevich) hypercover $`UX`$;
3. $`X\times 𝔸^1X`$ for every scheme $`X`$.
We call this the étale (Nisnevich) $`𝔸^1`$-local projective model structure according to \[I3\], and denote it by $`L_{\stackrel{´}{\mathrm{e}}\mathrm{t}}U(k)=L_{S,\stackrel{´}{\mathrm{e}}\mathrm{t}}U(\mathrm{Sm}/k)`$ (resp. $`L_{\mathrm{Nis}}U(k)=L_{S,\mathrm{Nis}}U(\mathrm{Sm}/k)`$).
Proposition 8.1 of \[D\] states that $`L_{\mathrm{Nis}}U(k)`$ is Quillen equivalent to the Nisnevich $`𝔸^1`$-localized model category $`𝒱_k`$ of \[MV\] and the analogue holds for the étale case.
The following lemma is clear from the theory of Bousfield localization.
###### Lemma 5.13
Let $`F:𝒞𝒟`$ be a functor between model categories. Let $`S`$ be a set of maps in $`𝒞`$ and let $`QC`$ denote a fibrant replacement for objects $`C`$ of $`𝒞`$. Suppose that the total left derived functor $`LF`$ of $`F`$ and that the left Bousfield localizations $`𝒞/S`$ and $`𝒟/FQ(S)`$ exist. Then $`F`$ induces a functor
$$F/S:𝒞/S𝒟/FQ(S)$$
and if $`F`$ sends the maps in $`S`$ into weak equivalences in $`𝒟`$, $`F`$ induces a total left derived functor on the localized category $`\mathrm{Ho}(𝒞/S)\mathrm{Ho}(𝒟)`$.
Since $`\mathrm{Et}`$ is a left Quillen functor on $`U(\mathrm{Sm}/k)`$ by \[D\], Proposition 2.3, in order to show that $`\mathrm{Et}`$ induces a left Quillen functor on $`L_TU(\mathrm{Sm}/k)`$, it suffices to show that $`\mathrm{Et}`$ takes the relations defined above into weak equivalences in $`\mathrm{pro}𝒮`$.
###### Theorem 5.14
Let $`\mathrm{}`$ be a prime different from the characteristic of $`k`$. With respect to the étale (Nisnevich) $`𝔸^1`$-local projective model structure on simplicial presheaves on $`\mathrm{Sm}/k`$, the functor $`\widehat{\mathrm{Et}}`$ induces a functor $`\mathrm{L}\widehat{\mathrm{Et}}`$ from the étale (Nisnevich) $`𝔸^1`$-homotopy category to the $`/\mathrm{}`$-cohomological homotopy category of $`\widehat{𝒮}`$. In particular, $`\mathrm{L}\widehat{\mathrm{Et}}X`$ is just $`\widehat{\mathrm{Et}}X`$ for every scheme in $`\mathrm{Sm}/k`$, and hence $`\widehat{\mathrm{Et}}`$ preserves $`𝔸^1`$-weak equivalences between smooth schemes over $`k`$.
Proof Since every Nisnevich hypercover is also an étale hypercover it suffices to check this for the étale case. The first part of the proof is the one of \[I3\], Theorem 2.6. The functor $`\mathrm{Et}`$ above is exactly the functor $`\mathrm{Re}:U(\mathrm{Sm}/k)=\mathrm{\Delta }^{\mathrm{op}}\mathrm{Pre}(\mathrm{Sm}/k)\mathrm{pro}𝒮`$ of \[D\], Proposition 2.3, hence it is a left Quillen functor. In order to see that $`\mathrm{Et}`$ is in fact a left Quillen functor on $`L_S(U(\mathrm{Sm}/k))`$, by Theorem 3.1.6 in \[Hi\] it remains to show that it takes cofibrant approximations of maps in $`S`$ to weak equivalences of $`\mathrm{pro}𝒮`$. This is done in \[I3\].
Note that the condition on $`\mathrm{}`$ is needed to ensure that the projection $`X\times 𝔸^1X`$ induces an isomorphism on étale cohomology $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{})\stackrel{}{}H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X\times 𝔸^1;/\mathrm{})`$. The functor $`\mathrm{Et}`$ takes weak equivalences between cofibrant objects to weak equivalences.
By Lemma 2.12 the completion functor $`\widehat{()}`$ preserves weak equivalences and cofibrations. Hence the composition $`\widehat{\mathrm{Et}}`$ sends weak equivalences between cofibrant objects into weak equivalences and the total left derived functor
$$\mathrm{L}\widehat{\mathrm{Et}}:_{𝔸^1}^{\stackrel{´}{\mathrm{e}}\mathrm{t}}(k)\widehat{}$$
exists.
The last statement of the theorem follows from the definition of the total left derived functor and the fact that all representable presheaves are cofibrant in $`L_S(U(\mathrm{Sm}/k))`$, cf. \[D\]. $`\mathrm{}`$
###### Remark 5.15
Since every Nisnevich hypercover is also an étale hypercover, and since $`\mathrm{Et}`$ and hence also $`\widehat{\mathrm{Et}}`$ send all étale hypercovers to weak equivalences, the functor from the Nisnevich $`𝔸^1`$-homotopy category factors through the étale $`𝔸^1`$-homotopy category:
(21)
$$\mathrm{L}\widehat{\mathrm{Et}}:_{𝔸^1}^{\mathrm{Nis}}(k)_{𝔸^1}^{\stackrel{´}{\mathrm{e}}\mathrm{t}}(k)\stackrel{\mathrm{L}\widehat{\mathrm{Et}}}{}\widehat{}$$
where the first functor corresponds to sending a presheaf to its associated sheaf in the étale topology.
We conclude this section with a result of \[I3\] on distinguished square. We will deduce from this result that we get a Mayer-Vietoris sequence for profinite étale cohomology theories.
We recall the definition of an elementary distinguished square of \[MV\]. It is a diagram
(22)
$$\begin{array}{ccc}U\times _XV& & V\\ & & p\\ U& \stackrel{i}{}& X\end{array}$$
of smooth schemes in which $`i`$ is an open inclusion, $`p`$ is étale and $`p:p^1(XU)XU`$ is an isomorphism, where $`XU`$ and $`p^1(XU)`$ are given the reduced structure. In particular, the maps $`i`$ and $`p`$ form a Nisnevich cover of $`X`$. As in \[I3\], Theorem 2.10, we can prove the following excision theorem for elementary distinguished squares. A similar result has already been proved by Friedlander in \[Fr2\], Lemma 14.10.
###### Theorem 5.16
Given an elementary distinguished square of smooth schemes over $`k`$. Then the square
(23)
$$\begin{array}{ccc}\widehat{\mathrm{Et}}(U\times _XV)& & \widehat{\mathrm{Et}}V\\ & & \\ \widehat{\mathrm{Et}}U& & \widehat{\mathrm{Et}}X\end{array}$$
is a homotopy pushout square of profinite spaces.
Proof The proof is the exact analog of the proof of Theorem 2.10 of \[I3\] with $`\mathrm{LEt}`$ replaced by $`\mathrm{L}\widehat{\mathrm{Et}}`$. $`\mathrm{}`$
## 6 Profinite étale realization on the stable $`𝔸^1`$-homotopy category
We extend the results of the previous section to the stable $`𝔸^1`$-homotopy category. One of the reasons why we consider the model $`\widehat{𝒮}`$ of $`\mathrm{Ho}(\widehat{𝒮})`$, instead of $`\mathrm{pro}𝒮`$, is that it seems to be easier to extend the functor $`\widehat{\mathrm{Et}}`$ to the category of profinite spectra rather than pro-spectra.
We consider $`S^1`$-spectra and motivic $`^1`$-spectra separately. The first point for both will be to choose the correct model for the stable $`𝔸^1`$-homotopy theory. Then we deduce the existence of an étale realization as a left derived functor on the stable homotopy category from \[Ho2\].
The problem for the motivic version is that $`\mathrm{Et}`$ and hence also $`\widehat{\mathrm{Et}}`$ do not commute with products in general. Fortunately, $`\widehat{\mathrm{Et}}`$ commutes with the smash product by the simplicial circle $`S^1`$. But for $`^1`$ we have to construct an intermediate category and then we will show by a zig-zag of functors that we get a functor on the homotopy level.
This étale realization of the stable motivic category is the technical key point for the construction of a transformation from algebraic cobordism given by the $`MGL`$-spectrum to the profinite étale cobordism of the next section.
### 6.1 Etale realization of motivic spectra
Let $`k`$ be the base field and let $`\overline{k}`$ be its separable closure. For the category of motivic $`_k^1`$-spectra over $`k`$, we have to consider presheaves $`𝒳`$ pointed by a morphism $`\mathrm{Spec}k𝒳`$. In particular, $`\mathrm{Spec}k`$ is the initial and terminal object. If we want to construct a stable étale realization, this forces us to consider the category $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$ of pointed profinite spaces relative over $`\widehat{\mathrm{Et}}k`$. Its objects $`(X,p,s)`$ are pointed profinite spaces $`X`$ together with a projection morphism $`p:X\widehat{\mathrm{Et}}k`$ and a section morphism $`s:\widehat{\mathrm{Et}}kX`$. The morphisms in this category are commutative diagrams in the obvious sense. The étale realization of a pointed presheaf $`𝒳`$ is naturally an object of $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$ via the images of the projection and section morphisms of $`𝒳`$. They are pointed by the composite $`=\widehat{\mathrm{Et}}\overline{k}\widehat{\mathrm{Et}}k𝒳`$.
Via the canonical maps $`X=\widehat{\mathrm{Et}}\overline{k}\widehat{\mathrm{Et}}k`$ and $`\widehat{\mathrm{Et}}kX`$ we may view every space $`X`$ in $`\widehat{𝒮}_{}`$ as an object in $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$. In particular, the pointed space $`S^2`$ is naturally an object in $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$.
We consider the usual model structure on $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$ where weak equivalences (resp. cofibrations, fibrations) are those maps which are $`/\mathrm{}`$-weak equivalences (resp. cofibrations, fibrations) in $`\widehat{𝒮}_{}`$ after forgetting the projection and section maps. This model structure on $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$ has the same properties as the one on $`\widehat{𝒮}_{}`$. In particular, it is left proper and fibrantly generated and we construct a stable model structure on the category $`\mathrm{Sp}(\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k,S^2)`$ of profinite $`S^2`$-spectra over $`\widehat{\mathrm{Et}}k`$ exactly in the same way. Its homotopy category will be denoted by $`\widehat{𝒮}_2/\widehat{\mathrm{Et}}k`$.
###### Remark 6.1
The smash product $`X_{\widehat{\mathrm{Et}}k}Z`$ in $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$ of a space $`X\widehat{𝒮}_{}`$ considered in $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$ and an object $`Z\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$ is canonically isomorphic to the smash product $`XZ`$ in $`\widehat{𝒮}_{}`$. For, if maps $`TX`$ and $`TZ`$ commute over $`\widehat{\mathrm{Et}}k`$, they commute over $``$, since the projection map of $`X`$ factors through $`=\widehat{\mathrm{Et}}\overline{k}\widehat{\mathrm{Et}}k`$. Hence both products are canonically isomorphic via their universal property.
When we want to extend $`\widehat{\mathrm{Et}}`$ to $`^1`$-spectra, we have to take into account the problem that the étale topological type functor does not commute with products in general. However the projections to each factor induce a canonical map
(24)
$$\widehat{\mathrm{Et}}(_k^1_kX)\widehat{\mathrm{Et}}(_k^1)_{\widehat{\mathrm{Et}}k}\widehat{\mathrm{Et}}(X).$$
###### Lemma 6.2
For every pointed presheaf $`𝒳`$ on $`\mathrm{Sm}/k`$ the sequence of canonical maps in $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$
(25)
$$S^2\widehat{\mathrm{Et}}𝒳\stackrel{}{}\widehat{\mathrm{Et}}(_k^1)_{\widehat{\mathrm{Et}}k}\widehat{\mathrm{Et}}(𝒳)\stackrel{}{}\widehat{\mathrm{Et}}(_k^1_k𝒳)$$
is a sequence of $`/\mathrm{}`$-weak equivalences.
Proof The projective bundle formula for étale cohomology implies that the étale cohomology $`H^{}(\widehat{\mathrm{Et}}_k^1;/\mathrm{})`$ is a free module of rank $`2`$ over $`H^{}(k;/\mathrm{})`$. We deduce from this fact on the one hand that we have a relative Künneth isomorphism for $`\widehat{\mathrm{Et}}_k^1`$ in $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$, see e.g. \[Sm\] for a discussion of relative Künneth theorems:
$$H^{}(\widehat{\mathrm{Et}}_k^1_{\widehat{\mathrm{Et}}k}\widehat{\mathrm{Et}}X;/\mathrm{})H^{}(\widehat{\mathrm{Et}}_k^1;/\mathrm{})_{H^{}(\widehat{\mathrm{Et}}k;/\mathrm{})}H^{}(\widehat{\mathrm{Et}}X;/\mathrm{}).$$
This implies that the canonical map (24) is a $`/\mathrm{}`$-weak equivalence in $`\widehat{𝒮}_{}`$ for everey $`X\mathrm{Sm}/k`$. On the other hand we deduce that the base extension map $`\widehat{\mathrm{Et}}_{\overline{k}}^1\widehat{\mathrm{Et}}X\widehat{\mathrm{Et}}_k^1_{\widehat{\mathrm{Et}}k}\widehat{\mathrm{Et}}X`$ is a $`/\mathrm{}`$-weak equivalence in $`\widehat{𝒮}_{}`$. Moreover, we have an isomorphism $`\widehat{\mathrm{Et}}_{\overline{k}}^1S^2`$ in $`\widehat{𝒮}_{}`$ by Example 5.10, which implies the assertion of the lemma for $`X\mathrm{Sm}/k`$.
If $`𝒳`$ denotes a presheaf on $`\mathrm{Sm}/k`$, we can replace $`X`$ by $`𝒳`$ and get the same results. For, $`𝒳`$ is isomorphic to the colimit of representable presheaves $`𝒳=\mathrm{colim}_sX_s`$. Since $`\mathrm{Et}`$ commutes with colimits, we get $`H^{}(\mathrm{Et}𝒳;/\mathrm{})lim_sH^{}(\mathrm{Et}X_s;/\mathrm{})`$. Since each $`X_s`$ is a smooth scheme over $`k`$, the étale cohomology groups $`H^i(\mathrm{Et}X_s;/\mathrm{})=H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^i(X_s;/\mathrm{})`$ are finite $`/\mathrm{}`$-vector spaces in each degree. Hence the limit over all $`s`$ of $`H^{}(\mathrm{Et}X_s;/\mathrm{})`$ commutes with the functor$`H^{}(\mathrm{Et}_k^1;/\mathrm{})_{H^{}(\mathrm{Et}k;/\mathrm{})}`$. $`\mathrm{}`$
Hence if $`\sigma _n:_k^1_kE_nE_{n+1}`$ is the structure map of a motivic $`_k^1`$-spectrum, then $`\widehat{\mathrm{Et}}`$ yields a sequence of maps
(26)
$$S^2\widehat{\mathrm{Et}}E_n\stackrel{}{}\widehat{\mathrm{Et}}(_k^1)_{\widehat{\mathrm{Et}}k}\widehat{\mathrm{Et}}(E_n)\stackrel{}{}\widehat{\mathrm{Et}}(_k^1_kE_n)\stackrel{\widehat{\mathrm{Et}}\sigma _n}{}E_{n+1}$$
where the first two maps are weak equivalences in $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$. Unfortunately, the map in the middle points to the wrong direction. This leads to the following construction.
Since there is no natural inverse map $`\widehat{\mathrm{Et}}(_k^1)_{\widehat{\mathrm{Et}}k}\widehat{\mathrm{Et}}(𝒳)\widehat{\mathrm{Et}}(_k^1_k𝒳)`$ in $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$, we may only construct an étale realization on the level of the homotopy categories. Therefore, we consider an intermediate category $`𝒞`$ and deduce from a zig-zag of functors
$$\mathrm{Sp}^^1(k)\stackrel{\stackrel{~}{\widehat{\mathrm{Et}}}}{}𝒞\stackrel{i}{}\mathrm{Sp}(\widehat{𝒮},S^2)$$
the existence of a stable realization functor $`𝒮^^1(k)\widehat{𝒮}_2`$.
In view of Lemma 6.2, it is natural to consider the following definition. The objects of the category $`𝒞/\widehat{\mathrm{Et}}k`$ are sequences
$$\{F_n,F_n^{},F_n^{\prime \prime };S^2F_n\stackrel{p_n}{}F_n^{}\stackrel{q_n}{}F_n^{\prime \prime }\stackrel{r_n}{}F_{n+1}\}_n$$
where $`F_n`$, $`F_n^{}`$, $`F_n^{\prime \prime }`$ are pointed profinite spaces over $`\widehat{\mathrm{Et}}k`$ and $`p_n`$, $`q_n`$ and $`r_n`$ are maps in $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$; furthermore the maps $`p_n`$ and $`q_n`$ are weak equivalences in $`\widehat{𝒮}_{}`$.
The morphisms of $`𝒞/\widehat{\mathrm{Et}}k`$ are levelwise morphisms of $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$ which make the obvious diagrams commute, where the map $`S^2E_nS^2F_n`$ is the map induced by $`E_nF_n`$. The functor
$$\mathrm{Sp}(\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k,S^2)\stackrel{i}{}𝒞/\widehat{\mathrm{Et}}k$$
denotes the full embedding which sends $`\{F_n,S^2F_n\stackrel{\sigma _n}{}F_{n+1}\}`$ to $`\{F_n,S^2F_n,S^2F_n;S^2F_n\stackrel{\mathrm{id}}{}S^2F_n\stackrel{\mathrm{id}}{}S^2F_n\stackrel{\sigma _n}{}F_{n+1}\}`$.
When we apply $`\widehat{\mathrm{Et}}`$ levelwise, we get a functor
$$\mathrm{Sp}^^1(k)\stackrel{\stackrel{~}{\widehat{\mathrm{Et}}}}{}𝒞/\widehat{\mathrm{Et}}k.$$
We define a class $`W`$ of maps in $`𝒞/\widehat{\mathrm{Et}}k`$ as the image of the stable equivalences of $`\mathrm{Sp}(\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k,S^2)`$ under the embedding $`i`$. Since the maps in $`W`$ are the images of weak equivalences in a model structure and since $`i`$ is a full embedding, it is clear that $`W`$ admits a calculus of fractions. This ensures that we may form the localized category $`\mathrm{Ho}(𝒞/\widehat{\mathrm{Et}}k):=𝒞[W^1]`$ in which the maps in $`W`$ become isomorphisms. We call the maps in $`W`$ weak equivalences or stable equivalences, by abuse of notation. We will call a map in $`W`$ a level equivalence if it is in the image of the level equivalences of $`\mathrm{Sp}(\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k,S^2)`$ under $`i`$.
Furthermore, $`i`$ sends stable equivalences into weak equivalences by definition and hence it induces a functor on the homotopy categories, which we also denote by $`i`$.
###### Proposition 6.3
The induced functor $`i:\widehat{𝒮}_2/\widehat{\mathrm{Et}}k\mathrm{Ho}(𝒞/\widehat{\mathrm{Et}}k)`$ is an equivalence of categories. In particular, there is an inverse equivalence $`j:\mathrm{Ho}(𝒞/\widehat{\mathrm{Et}}k)\widehat{𝒮}_2/\widehat{\mathrm{Et}}k`$.
Proof By \[MacL\], IV, 4, Theorem 1, since $`i`$ is a full embedding, it suffices to show that for every $`F\mathrm{Ho}(𝒞/\widehat{\mathrm{Et}}k)`$ there is a spectrum $`E\widehat{𝒮}_2/\widehat{\mathrm{Et}}k`$ such that $`i(E)F`$ in $`\mathrm{Ho}(𝒞/\widehat{\mathrm{Et}}k)`$. This implies already that $`i`$ is an equivalence and that there is an inverse functor $`j`$.
The crucial point is to construct the structure map of a spectrum from the given data of $`F`$. For the convenience of notations, we will omit in the following proof the structure maps of spaces over $`\widehat{\mathrm{Et}}k`$.
Let $`R`$ be a fixed fibrant replacement functor in $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$. We consider the category $`\mathrm{Sp}(\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k,RS^2)`$ as a Quillen equivalent model for $`\widehat{𝒮}_2/\widehat{\mathrm{Et}}k`$. We apply $`R`$ on each level. Since $`R`$ commutes with products, we get the following sequence
(27)
$$RS^2RF_n\stackrel{Rp_n}{}RF_n^{}\stackrel{Rq_n}{}RF_n^{\prime \prime }\stackrel{Rr_n}{}RF_{n+1}.$$
In addition, the functor $`R`$ can be chosen such that the map $`Rq_n`$ is a trivial fibration between fibrant and cofibrant objects, see Proposition 8.1.23 of \[Hi\]. The sequence we get is still isomorphic in $`\mathrm{Ho}(𝒞/\widehat{\mathrm{Et}}k)`$ to the initial one since they are level equivalent. By Proposition 9.6.4 of \[Hi\], there is a right inverse $`s_n`$ of $`Rq_n`$ in $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$ such that $`Rq_ns_n=\mathrm{id}_{RF_n^{}}`$ and a homotopy $`s_nRq_n\mathrm{id}_{RF_n^{\prime \prime }}`$. We denote by $`E`$ the resulting spectrum with structure maps $`\sigma _n:=Rr_ns_nRp_n`$.
In order to check that $`i(E)`$ is isomorphic to $`F`$ in $`\mathrm{Ho}(𝒞/\widehat{\mathrm{Et}}k)`$, we consider the following diagram
(28)
$$\begin{array}{ccc}RS^2RF_n& \stackrel{\mathrm{id}}{}& RS^2RF_n\\ \mathrm{id}& & Rp_n\\ RS^2RF_n& \stackrel{Rp_n}{}& RF_n^{}\\ \mathrm{id}& & Rq_n\\ RS^2RF_n& \stackrel{s_nRp_n}{}& RF_n^{\prime \prime }\\ \sigma _n& & Rr_n\\ RF_{n+1}& \stackrel{\mathrm{id}}{}& RF_{n+1}\end{array}$$
representing the $`n`$-th level of the canonical map $`i(E)F`$ in $`\mathrm{Ho}(𝒞/\widehat{\mathrm{Et}}k)`$. The upper square commutes obviously. The lower square commutes by the definition of $`\sigma _n:=Rr_ns_nRp_n`$. The middle square commutes by the construction of $`s_n`$ such that $`Rq_ns_n=\mathrm{id}_{RF_n^{}}`$. Hence this is in fact a morphism in $`𝒞/\widehat{\mathrm{Et}}k`$. Since all horizontal maps are weak equivalences, the morphism is a level equivalence in $`𝒞/\widehat{\mathrm{Et}}k`$ and hence it is an isomorphism in $`\mathrm{Ho}(𝒞/\widehat{\mathrm{Et}}k)`$. $`\mathrm{}`$
We have to show that $`\stackrel{~}{\widehat{\mathrm{Et}}}_{\mathrm{Sp}}:\mathrm{Sp}^^1(k)𝒞/\widehat{\mathrm{Et}}k`$ has a total left derived functor. Therefor, we have to choose the right model for the stable motivic category $`𝒮^^1(k)`$. By \[D\], we know that $`LU(k)`$ is a left proper cellular simplicial model category which allows us to apply the methods of \[Ho2\].
###### Proposition 6.4
The canonical functors
$$\mathrm{Sp}^^1(k)\mathrm{Sp}(𝒱_k,^1)\mathrm{Sp}(LU(k),^1)$$
are Quillen equivalences and we get
$$𝒮^^1(k)\mathrm{Ho}^{\mathrm{stable}}(\mathrm{Sp}(𝒱_k,^1))\mathrm{Ho}^{\mathrm{stable}}(\mathrm{Sp}(LU(k),^1)).$$
Proof The first equivalence follows from \[Ho2\], Corollary 3.5, and the obvious fact that the cofibrations are mapped to cofibrations and the fibrant objects correspond to each other in both categories.
The second equivalence follows from \[Ho2\], Theorem 5.7, taking into account that $`LU(k)`$ is Quillen equivalent to $`𝒱_k`$ and that all objects in $`𝒱_k`$ are cofibrant. $`\mathrm{}`$
We use $`\mathrm{Sp}(LU(k),^1)`$ as a model for $`𝒮^^1(k)`$.
###### Theorem 6.5
The functor $`\widehat{\mathrm{Et}}`$ induces an étale realization of the stable motivic homotopy category of $`^1`$-spectra:
$$\mathrm{L}\widehat{\mathrm{Et}}:𝒮^^1(k)\widehat{𝒮}_2/\widehat{\mathrm{Et}}k$$
defined to be the composite $`\mathrm{L}\widehat{\mathrm{Et}}:𝒮(k)\stackrel{\mathrm{L}\stackrel{~}{\widehat{\mathrm{Et}}}}{}\mathrm{Ho}(𝒞/\widehat{\mathrm{Et}}k)\stackrel{j}{}\widehat{𝒮}_2/\widehat{\mathrm{Et}}k`$.
Proof We have to show that stable equivalences in $`\mathrm{Sp}(LU(k),^1)`$ are sent to isomorphisms in $`\mathrm{Ho}(𝒞/\widehat{\mathrm{Et}}k)`$.
We know that $`\stackrel{~}{\widehat{\mathrm{Et}}}_{\mathrm{Sp}}`$ sends level weak equivalences between cofibrant objects in $`LU(k)`$ to weak equivalences in $`𝒞/\widehat{\mathrm{Et}}k`$, since $`\widehat{\mathrm{Et}}`$ sends weak equivalences between cofibrant objects to weak equivalences in $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$. Hence it induces a total left derived functor on the projective model structure of $`\mathrm{Sp}(LU(k),^1)`$.
We use the notation $`\mathrm{\Sigma }_n^^1:LU(k)\mathrm{Sp}(LU(k),^1)`$ for the left adjoint to the $`n`$-th evaluation functor. It is given by $`(\mathrm{\Sigma }_n^^1𝒳)_m=(^1)^{mn}𝒳`$ if $`mn`$ and $`(\mathrm{\Sigma }_n^^1𝒳)_m=\mathrm{Spec}k`$ otherwise.
We denote by $`F_n:\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k𝒞/\widehat{\mathrm{Et}}k`$ the composition of the corresponding functor $`\widehat{\mathrm{\Sigma }}_n:\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k\mathrm{Sp}(\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k,S^2)`$ followed by the embedding $`i:\mathrm{Sp}(\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k,S^2)𝒞/\widehat{\mathrm{Et}}k`$.
In order to show that there exists a derived functor on the stable structure it suffices to show that $`\widehat{\mathrm{Et}}_{\mathrm{Sp}}(\zeta _n^𝒳)`$ is a stable equivalence for maps $`\zeta _n^𝒳:\mathrm{\Sigma }_{n+1}^1𝒳\mathrm{\Sigma }_n𝒳`$ in $`\mathrm{Sp}(LU(k),^1)`$ for all cofibrant presheaves $`𝒳LU(k)`$, since these are the maps at which we localize for the stabilization.
We consider the commutative diagram in $`\mathrm{Ho}(𝒞/\widehat{\mathrm{Et}}k)`$
$$\begin{array}{cccc}\zeta _n^{\widehat{\mathrm{Et}}𝒳}:& F_{n+1}(S^2\widehat{\mathrm{Et}}𝒳)& \stackrel{}{}& F_n(\widehat{\mathrm{Et}}𝒳)\\ & & & \\ & F_{n+1}(\widehat{\mathrm{Et}}_k^1_{\widehat{\mathrm{Et}}k}\widehat{\mathrm{Et}}𝒳))& & F_n(\widehat{\mathrm{Et}}𝒳)\\ & & & \\ & F_{n+1}(\widehat{\mathrm{Et}}(^1𝒳))& & F_n(\widehat{\mathrm{Et}}𝒳)\\ & & & \\ \widehat{\mathrm{Et}}_{\mathrm{Sp}}\zeta _n^𝒳:& \widehat{\mathrm{Et}}_{\mathrm{Sp}}(\mathrm{\Sigma }_{n+1}^^1(^1𝒳))& & \widehat{\mathrm{Et}}_{\mathrm{Sp}}(\mathrm{\Sigma }_n^^1𝒳).\end{array}$$
The upper and middle vertical isomorphism on the left hand side are given by the obvious level equivalences given by the canonical sequence of weak equivalences as in Lemma 6.2. The lower vertical isomorphisms are given by the following composition:
We discuss the isomorphism $`F_n(\widehat{\mathrm{Et}}𝒳)\widehat{\mathrm{Et}}_{\mathrm{Sp}}(\mathrm{\Sigma }_n^^1𝒳)`$. The lower left hand isomorphism is constructed in the same way. We define an intermediate object $`E^n𝒞/\widehat{\mathrm{Et}}k`$ given in degree $`mn`$ by
$$E_m^n=(\widehat{\mathrm{Et}}_k^1)^{mn}_{\widehat{\mathrm{Et}}k}\widehat{\mathrm{Et}}𝒳,E_m^n^{}=(\widehat{\mathrm{Et}}_k^1)^{m+1n}_{\widehat{\mathrm{Et}}k}\widehat{\mathrm{Et}}𝒳,E_m^{n^{\prime \prime }}=E_m^n^{}$$
with the obvious strcuture maps induced by $`S^2\widehat{\mathrm{Et}}_k^1`$ respectively the identity; in degree $`m<n`$ it is defined by $`E_m^n=E_m^n^{}=E_m^{n^{\prime \prime }}=\widehat{\mathrm{Et}}k`$ with identity maps and the map to the terminal object $`\widehat{\mathrm{Et}}k`$ of $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$.
The object $`E^n`$ is defined such that there are canonical maps
$$F_n(\widehat{\mathrm{Et}}𝒳)\stackrel{\alpha }{}E^n\stackrel{\beta }{}\widehat{\mathrm{Et}}_{\mathrm{Sp}}(\mathrm{\Sigma }_n^^1𝒳)$$
induced in degree $`mn`$ by the maps
$$(S^2)^{mn}\widehat{\mathrm{Et}}𝒳(\widehat{\mathrm{Et}}_k^1)^{mn}_{\widehat{\mathrm{Et}}k}\widehat{\mathrm{Et}}𝒳\widehat{\mathrm{Et}}((_k^1)^{mn}_k\widehat{\mathrm{Et}}𝒳).$$
One checks easily using the canonical weak equivalences of Lemma 6.2 that the maps $`\alpha `$ and $`\beta `$ are both level equivalences in $`𝒞/\widehat{\mathrm{Et}}k`$. Hence $`\alpha `$ and $`\beta `$ are isomorphisms in $`\mathrm{Ho}(𝒞/\widehat{\mathrm{Et}}k)`$. Their composition is the isomorphism $`F_n(\widehat{\mathrm{Et}}𝒳)\widehat{\mathrm{Et}}_{\mathrm{Sp}}(\mathrm{\Sigma }_n^^1𝒳)`$.
We deduce from the diagram that $`\zeta _n^{\widehat{\mathrm{Et}}𝒳}`$ and $`\widehat{\mathrm{Et}}_{\mathrm{Sp}}\zeta _n^𝒳`$ differ only by an isomorphism in $`\mathrm{Ho}(𝒞/\widehat{\mathrm{Et}}k)`$ given by level equivalences. Since $`\zeta _n^{\widehat{\mathrm{Et}}𝒳}`$ is a stable equivalence in $`𝒞/\widehat{\mathrm{Et}}k`$ by definition, the map $`\widehat{\mathrm{Et}}_{\mathrm{Sp}}\zeta _n^𝒳`$ is in fact an isomorphism in $`\mathrm{Ho}(𝒞/\widehat{\mathrm{Et}}k)`$.
The last point is to show that $`\widehat{\mathrm{Et}}_{\mathrm{Sp}}`$ factors through $`𝔸^1`$-weak equivalences. By Lemma 5.13 and \[Mo3\] it suffices to show that $`\widehat{\mathrm{Et}}_{\mathrm{Sp}}`$ sends the maps $`\mathrm{\Sigma }_^1^{\mathrm{}}((X\times 𝔸^1)_+)[n]\mathrm{\Sigma }_^1^{\mathrm{}}(X_+)[n]`$ to isomorphisms for all $`X\mathrm{Sm}/k`$ and all $`n`$.
We know that $`\widehat{\mathrm{Et}}(X\times 𝔸^1)\widehat{\mathrm{Et}}(X)`$ is a weak equivalence in $`\widehat{𝒮}`$ for $`X\mathrm{Sm}/k`$. Hence $`\widehat{\mathrm{Et}}_{\mathrm{Sp}}\mathrm{\Sigma }_^1^{\mathrm{}}((X\times 𝔸^1)_+)[n]\widehat{\mathrm{Et}}_{\mathrm{Sp}}\mathrm{\Sigma }_^1^{\mathrm{}}(X_+)[n]`$ is a level weak equivalence in $`𝒞/\widehat{\mathrm{Et}}k`$. This shows that $`\widehat{\mathrm{Et}}_{\mathrm{Sp}}`$ induces a total derived functor on the category $`𝒮^^1(k)=𝒮_{𝔸^1}^^1(k)`$ of $`𝔸^1`$-localized $`^1`$-spectra. $`\mathrm{}`$
As in Theorem 5.14, since the suspension spectrum of a smooth scheme is a stable cofibrant object, we have $`\mathrm{L}\widehat{\mathrm{Et}}=\widehat{\mathrm{Et}}`$ on such spectra and $`\widehat{\mathrm{Et}}`$ sends stable $`𝔸^1`$-equivalences to the image in $`𝒞/\widehat{\mathrm{Et}}k`$ of stable equivalences in $`\widehat{𝒮}_2/\widehat{\mathrm{Et}}k`$. We summarize this discussion in the following
### 6.2 Etale realization of $`S^1`$-spectra
The étale realization of $`S^1`$-spectra is essentially simpler. The structure maps $`\sigma _n:S_k^1_kE_nE_{n+1}`$ induce by base extension canonical maps $`S_{\overline{k}}^1E_nS_k^1_kE_n\stackrel{\sigma _n}{}E_{n+1}`$ in $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$. By Example 5.6, we have a canonical isomorphism $`\widehat{\mathrm{Et}}S_{\overline{k}}^1=S^1`$ in $`\widehat{𝒮}_{}`$ and by Formula (20) we have an isomorphism $`S^1\widehat{\mathrm{Et}}E_n\stackrel{}{}\widehat{\mathrm{Et}}(S_{\overline{k}}^1E_n)`$ in $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$. This implies that the functor $`\widehat{\mathrm{Et}}`$ on presheaves induces a functor on $`S^1`$-spectra
$$\widehat{\mathrm{Et}}:\mathrm{Sp}^{S^1}(k)\mathrm{Sp}(\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k)$$
by sending a spectrum $`E`$ to the profinite spectrum $`\widehat{\mathrm{Et}}E`$ given in degree $`n`$ by $`(\widehat{\mathrm{Et}}E)_n:=\widehat{\mathrm{Et}}(E_n)`$ with structure maps
$$S^1\widehat{\mathrm{Et}}E_n=\widehat{\mathrm{Et}}(S_{\overline{k}}^1E_n)\stackrel{}{}\widehat{\mathrm{Et}}(S_k^1_kE_n)\stackrel{\widehat{\mathrm{Et}}\sigma _n}{}\widehat{\mathrm{Et}}E_{n+1}$$
where $`\sigma `$ is the structure map of the given spectrum $`E`$ and the map in the middle is obviously a $`/\mathrm{}`$-weak equivalence in $`\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k`$.
The proof of the following theorem is essentially the same as for the previous one on $`^1`$-spectra with the obvious simplifications.
###### Theorem 6.6
The functor $`\widehat{\mathrm{Et}}_{\mathrm{Sp}}:\mathrm{Sp}^{S^1}(k)\mathrm{Sp}(\widehat{𝒮}_{}/\widehat{\mathrm{Et}}k)`$ admits a left derived functor $`\mathrm{L}\widehat{\mathrm{Et}}_{\mathrm{Sp}}:𝒮^{S^1}(k)\widehat{𝒮}/\widehat{\mathrm{Et}}k`$ on the stable $`𝔸^1`$-homotopy category of $`S^1`$-spectra.
### 6.3 Examples
Let $`MGL`$ denote the motivic Thom spectrum defined in \[Vo2\] representing algebraic cobordism.
###### Theorem 6.7
Let $`k`$ be a separably closed field. There is an isomorphism in $`\widehat{𝒮}_2`$
$$\mathrm{L}\widehat{\mathrm{Et}}(MGL)\widehat{MU}.$$
Proof Let $`G_k(n,N)`$ be the Grassmannian over $`k`$ and $`G_k(n,N)`$ be the colimit over $`N`$. By the Thom isomorphisms in étale and singular cohomology, it suffices to show that there is a weak equivalence between $`\widehat{\mathrm{Et}}(G_k(n,N))`$ and $`\widehat{G}_{}(n,N)`$, the profinite completion of the corresponding simplicial set of the complex Grassmannian manifold.
If $`\mathrm{char}k>0`$ let $`R`$ be the ring of Witt vectors of $`k`$, otherwise $`R=k`$. By choosing a common embedding of $`R`$ and $``$ into an algebarically closed field, Friedlander proved in \[Fr1\], 3.2.2, that there is a natural sequence of $`/\mathrm{}`$-weak equivalences in $`\mathrm{pro}𝒮_{}`$
$$\mathrm{Sing}(G_{}(n,N))\mathrm{Et}(G_{}(n,N))\mathrm{Et}(G_R(n,N))\mathrm{Et}(G_k(n,N)),$$
where one should recall that weak equivalences in $`\mathrm{pro}_0`$ in the sense of \[Fr1\] correspond to $`/\mathrm{}`$-weak equivalences in $`\mathrm{pro}𝒮_{}`$ in the sense of \[I4\].
By taking colimits with respect to $`N`$ we get a sequence of $`/\mathrm{}`$-weak equivalences in $`\mathrm{pro}𝒮_{}`$
$$\mathrm{Sing}(G_{}(n))\mathrm{Et}(G_{}(n))\mathrm{Et}(G_R(n))\mathrm{Et}(G_k(n)).$$
Since $`/\mathrm{}`$-weak equivalences are preserved under completion, this shows that there is an isomorphism $`\widehat{G}_{}(n)\widehat{\mathrm{Et}}(G_k(n))`$ in $`\widehat{}_{}`$. Together with the Thom isomorphism, this proves that there is an isomorphism of spectra $`\mathrm{L}\widehat{\mathrm{Et}}(MGL)\widehat{MU}`$ in $`\widehat{𝒮}_2`$. $`\mathrm{}`$
## 7 Profinite étale cobordism
The study of profinite étale cohomology theories is the main purpose of this paper. The idea is to apply generalized profinite cohomology theories represented by a profinite spectrum to the functor $`\widehat{\mathrm{Et}}`$.
We show that every such profinite étale cohomology theory is in fact a cohomology theory on $`\mathrm{Sm}/k`$ in the sense of \[Pa\]. One should note that in order to get the $`𝔸^1`$-invariance, we have to complete away from characteristic of $`k`$ and cannot use a $`\pi _{}`$-model structure, since $`\pi _1^{\stackrel{´}{\mathrm{e}}\mathrm{t}}(𝔸_k^1)`$ is non-trivial over a field of positive characteristic, see \[Sch\].
We show that the naturally arising profinite étale K-theory represented by $`\widehat{KU}`$ for schemes over a separably closed field is isomorphic to the étale K-theory of Friedlander \[Fr1\].
The second theory we study is profinite étale cobordism represented by $`\widehat{MU}`$. The coefficients of this theory are determined by our knowledge about profinite cohomology theories. We deduce an Atiyah-Hirzebruch spectral sequence from Theorem 4.11 for étale cobordism with finite coefficients $`\widehat{MU}/\mathrm{}^\nu `$ starting with étale cohomology.
We show in the last section that $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$ is in fact an oriented cohomology theory on $`\mathrm{Sm}/k`$ if $`k`$ is separably closed. In order to prove this, we have to use non-trivial results on $`\widehat{MU}`$ to be able to reduce the problems to questions in étale cohomology.
The ideas for the proof that $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$ is an oriented cohomology theory are not new. For our purpose we use the techniques of \[Pa\]. We define an orientation of $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$ with respect to smooth schemes over $`k`$. We use this orientation and a projective bundle formula to construct a Chern structure on $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$. The key ingredient for the proof of the projective bundle formula is the Atiyah-Hirzebruch spectral sequence of $`\widehat{MU}`$. From facts about étale cohomology and the results of \[Pa\], we deduce that étale cobordism is an oriented cohomology theory on smooth schemes over a separably closed field $`k`$ in the sense of \[Pa\] and hence in the sense of \[LM\].
This result will enable us to compare profinite étale cobordism with algebraic cobordism in the next section.
For the convenience of the reader we recall some notations and definitions: we denote by $`\widehat{𝒮}`$, resp. $`\widehat{𝒮}_{}`$, the category of simplicial profinite sets, resp. pointed simplicial profinite sets; $`\widehat{}`$, resp. $`\widehat{}_{}`$, is the corresponding homotopy category; $`\mathrm{Sp}(\widehat{𝒮}_{})`$ is the category of profinite spectra and $`\widehat{𝒮}`$ denotes the stable homotopy category of profinite spectra; we write $`\mathrm{\Sigma }^{\mathrm{}}(Y)`$ for the suspension spectrum of a profinite space $`Y`$, see Sections 2 and 3.
Every profinite spectrum $`E`$ determines a generalized cohomology theory $`E^{}()`$ on profinite spaces, e.g. $`\widehat{MU}`$ and $`\widehat{MU}/\mathrm{}^\nu `$ yield profinite cobordism theories $`\widehat{MU}^{}()`$ and $`\widehat{MU}^{}(;/\mathrm{}^\nu )`$, see Section 4.
The functor $`\widehat{\mathrm{Et}}:\mathrm{Sch}/k\widehat{𝒮}`$ is the profinitely completed étale topological type functor, see Section 5; we denote by $`\mathrm{L}\widehat{\mathrm{Et}}:𝒮^^1(k)\widehat{𝒮}`$ the stable étale realization functor, see Section 6 and Theorem 6.5.
### 7.1 Profinite étale cohomology theories
Let $`k`$ be a fixed base field and let $`\mathrm{}`$ be a fixed prime different from the characteristic of $`k`$. We consider the category $`\mathrm{Sch}/k`$ of schemes of finite type over $`\mathrm{Spec}k`$. This implies in particular that all schemes are noetherian and we may apply the étale topological type functor.
###### Definition 7.1
Let $`E\mathrm{Sp}(\widehat{𝒮}_{})`$ be a profinite spectrum and let $`X\mathrm{Sch}/k`$.
1. We define the profinite étale cohomology of $`X`$ in $`E`$ to be the profinite cohomology theory represented by $`E`$ applied to the profinite space $`\widehat{\mathrm{Et}}X`$, i.e.
$$E_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^n(X):=E^n(\widehat{\mathrm{Et}}X)=\mathrm{Hom}_{\widehat{𝒮}}(\mathrm{\Sigma }^{\mathrm{}}(\widehat{\mathrm{Et}}X),E[n]),$$
where we add a base-point if $`X`$ is not already pointed.
2. We define relative étale cohomology groups $`E_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^n(X,U)`$ for an open subscheme $`UX`$ by
$$E_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^n(X,U):=E^n(\widehat{\mathrm{Et}}(X)/\widehat{\mathrm{Et}}(U)).$$
3. We define the étale cohomology group
$$E_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^n(_k^{\mathrm{}}):=\underset{V}{lim}E_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^n((V))$$
to be the limit over all finite dimensional vector spaces $`Vk^{\mathrm{}}`$ of the just defined groups $`E_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^n((V))`$.
###### Remark 7.2
Since $`\widehat{\mathrm{Et}}`$ is also a functor on spaces, we could have defined $`E_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^n(_k^{\mathrm{}})`$ just as for schemes to be the group $`E^n(\widehat{\mathrm{Et}}_k^{\mathrm{}})`$. But we will use results that are based on the first definition. Since $`\widehat{\mathrm{Et}}`$ does not commute with arbitrary colimits, the two definitions differ in general. But there is a canonical map $`E^{}(\widehat{\mathrm{Et}}^{\mathrm{}})E_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(^{\mathrm{}})`$.
###### Proposition 7.3
If $`\pi `$ is a finite abelian group, then the étale cohomology $`H^{}(X;\pi )`$ with constant coefficients $`\pi `$ is equal to the profinite étale cohomology theory represented by the Eilenberg-MacLane spectrum $`H\pi `$.
Proof This follows directly from Remark 5.3 and Proposition 4.8. $`\mathrm{}`$
###### Proposition 7.4
Let $`U\stackrel{i}{}X\stackrel{p}{}V`$ induce an elementary distinguished square, cf. (22). Let $`E`$ be a profinite spectrum. Then there is a Mayer-Vietoris long exact sequence of graded groups
$$\mathrm{}E_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^n(X)E_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^n(U)E_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^n(V)E_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^n(U\times _XV)E_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{n+1}(X)\mathrm{}.$$
Proof This follows directly from Theorem 5.16. $`\mathrm{}`$
###### Proposition 7.5
Let $`E`$ be a profinite spectrum. Let $`UX`$ be an open subscheme of $`X`$. We get a long exact sequence of cohomology groups
$$\begin{array}{ccc}\hfill E^{}(\widehat{\mathrm{Et}}X)\stackrel{j^{}}{}& E^{}(\widehat{\mathrm{Et}}U)\stackrel{}{}& E^{}(\widehat{\mathrm{Et}}(X)/\widehat{\mathrm{Et}}(U))\stackrel{i^{}}{}\hfill \\ \hfill \stackrel{i^{}}{}E^{+1}(\widehat{\mathrm{Et}}X)& \stackrel{j^{}}{}E^{+1}(\widehat{\mathrm{Et}}U)& \end{array}$$
where $`j:\widehat{\mathrm{Et}}U\widehat{\mathrm{Et}}X`$ and $`i:(\widehat{\mathrm{Et}}X,)(\widehat{\mathrm{Et}}X,\widehat{\mathrm{Et}}U)`$ denote the natural induced inclusions.
Proof Since $`\widehat{\mathrm{Et}}U\widehat{\mathrm{Et}}X\widehat{\mathrm{Et}}(X)/\widehat{\mathrm{Et}}(U)`$ is isomorphic to a cofiber sequence in $`\widehat{𝒮}_{}`$, this is just the usual long exact sequence of $`\mathrm{Hom}`$-groups induced by a cofiber sequence in a simplicial model category, see \[Qu1\]. $`\mathrm{}`$
On the category $`\mathrm{Sm}/k`$ of smooth schemes of finite type over $`k`$, every étale cohomology theory satisfies homotopy invariance.
###### Proposition 7.6
Let $`E`$ be a profinite spectrum. Let $`X\mathrm{Sm}/k`$. The projection $`p:X\times 𝔸^1X`$ induces an isomorphism $`p^{}:E_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X)\stackrel{}{}E_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X\times 𝔸^1)`$.
Proof This is clear since $`p`$ induces a weak equivalence in $`\widehat{𝒮}`$, see Theorem 5.14, and hence induces isomorphisms on cohomology theories. $`\mathrm{}`$
More generally, this implies
###### Corollary 7.7
Let $`E`$ be a profinite spectrum. Let $`VX`$ be an $`𝔸^n`$-bundle over $`X`$ in $`\mathrm{Sm}/k`$. Then $`p^{}:E_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X)\stackrel{}{}E_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(V)`$ is an isomorphism.
Proof This follows immediately from the Mayer-Vietoris sequence of Proposition 7.4 and the previous proposition. $`\mathrm{}`$
In \[Pa\], Definition 2.0.1, Panin gives the definition of a cohomology theory on $`\mathrm{Sm}/k`$. The following theorem states that we have constructed a way to define such cohomology theories starting with a profinite spectrum and applying its associated cohomology theory to $`\widehat{\mathrm{Et}}`$ on $`\mathrm{Sm}/k`$.
###### Theorem 7.8
Let $`E`$ be a profinite spectrum. The étale cohomology theory $`E_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}()`$ represented by $`E`$ satisfies the axioms of a cohomology theory on $`\mathrm{Sm}/k`$ of \[Pa\].
Proof We check the axioms of a cohomology theory in the sense of \[Pa\], Definition 2.0.1.
1. Localization: This is clear from Remark 7.5.
2. Excision: Let $`e:(X^{},U^{})(X,U)`$ be a morphism of pairs of schemes in $`\mathrm{Sm}/k`$ such that $`e`$ is étale and for $`Z=XU`$, $`Z^{}=X^{}U^{}`$ one has $`e^1(Z)=Z^{}`$ and $`e:Z^{}Z`$ is an isomorphism. By \[Mi\] III, Proposition 1.27, we know that the morphism $`e`$ induces an isomorphism in étale cohomology $`H^{}(\widehat{\mathrm{Et}}(X)/\widehat{\mathrm{Et}}(U);/\mathrm{})H^{}(\widehat{\mathrm{Et}}(X^{})/\widehat{\mathrm{Et}}(U^{});/\mathrm{})`$. Hence $`\widehat{\mathrm{Et}}(X^{})/\widehat{\mathrm{Et}}(U^{})\widehat{\mathrm{Et}}(X)/\widehat{\mathrm{Et}}(U)`$ is an isomorphism in $`\widehat{}_{}`$. Therefore, it induces the desired isomorphism
$$E^{}(\widehat{\mathrm{Et}}(X)/\widehat{\mathrm{Et}}(U))E^{}(\widehat{\mathrm{Et}}(X^{})/\widehat{\mathrm{Et}}(U^{}))$$
for every étale cohomology theory.
3. Homotopy invariance: This is the content of Proposition 7.6. $`\mathrm{}`$
Later we will prove that profinite étale cobordism is in fact an oriented cohomology theory in the sense of \[Pa\] and of \[LM\].
### 7.2 Profinite étale K-theory
As a first application, we consider a comparison statement for profinite K-theory and the pro-K-theory of \[Fr1\], and hence for profinite étale K-theory and for Friedlander’s étale K-theory. Let $`\{X_s\}\mathrm{pro}𝒮`$ be a pro-object in $`𝒮`$ and let $`BU`$ be the simplicial set representing complex K-theory. Friedlander defines the K-theory of $`\{X_s\}`$ for $`ϵ=0,1`$ and $`k>0`$ by
$$K^ϵ(\{X_s\};/\mathrm{}^\nu )=\mathrm{Hom}_{\mathrm{pro}}(\{\mathrm{\Sigma }^ϵX_s\},\{P^nBU/\mathrm{}^\nu \})$$
where $`\{P^nBU\}`$ denotes the Postnikov tower of $`BU`$ considered as a pro-object in $`𝒮`$. For a locally noetherian scheme $`X`$, Friedlander defines the étale K-theory by
$$K_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^ϵ(X;/\mathrm{}^\nu ):=K^ϵ(\mathrm{Et}X;/\mathrm{}^\nu ).$$
Friedlander also establishes an Atiyah-Hirzebruch spectral sequence, \[Fr1\], Proposition 1.4,
$$E_2^p(/\mathrm{}^\nu )=H^n(\{X_s\};/\mathrm{}^\nu )K^{}(\{X_s\};/\mathrm{}^\nu )$$
which is convergent provided that $`E_2^p(/\mathrm{}^\nu )=0`$ for $`n`$ sufficiently large or that $`E_2^p(/\mathrm{}^\nu )`$ is finite for each $`p`$.
Following Definition 7.1 we set
###### Definition 7.9
We define the profinite K-theory of a profinite space $`X`$ to be the cohomology theory represented by the profinitely completed spectrum $`\widehat{KU}`$ and define profinite K-theory with $`/\mathrm{}^\nu `$-coefficients to be represented by $`\widehat{KU}/\mathrm{}^\nu `$. The profinite étale K-theory of a scheme $`X`$ over $`k`$, is hence defined by
$$\widehat{KU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^i(X):=\widehat{KU}^i(\widehat{\mathrm{Et}}X)$$
and profinite étale K-theory with $`/\mathrm{}^\nu `$-coefficients by
$$\widehat{KU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^i(X;/\mathrm{}^\nu ):=\widehat{KU}^i(\widehat{\mathrm{Et}}X;/\mathrm{}^\nu ).$$
###### Theorem 7.10
Let $`k`$ be a field. For any smooth scheme $`X`$ over $`k`$ and any $`\nu `$ there is a strongly convergent spectral sequence $`\{E_r^{p,q}\}`$:
$$E_2^{p,q}=H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^p(X;/\mathrm{}^\nu )\widehat{KU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{p+q}(X;/\mathrm{}^\nu ).$$
Proof This is the spectral sequence of Theorem 4.11 where we use the isomorphism
$$H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )H^{}(\widehat{\mathrm{Et}}X;/\mathrm{}^\nu )$$
of Remark 5.3 for the groups $`(\widehat{KU}/\mathrm{}_0^\nu )^q`$ which are isomorphic to $`/\mathrm{}^\nu `$, where $`\widehat{KU}/\mathrm{}_0^\nu `$ denotes a connective covering. Since the étale cohomology groups $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^i(X;/\mathrm{}^\nu )`$ vanish for large $`i`$, see e.g. \[Mi\] VI, Theorem 1.1, the convergence also follows from Theorem 4.11. $`\mathrm{}`$
Since $`\widehat{KU}`$ is an $`\widehat{\mathrm{\Omega }}`$-spectrum, we have for every profinite space $`X`$
$$\widehat{KU}^i(X;/\mathrm{}^\nu )\mathrm{Hom}_\widehat{}_{}(X_+,\widehat{KU}_i/\mathrm{}^\nu ).$$
Hence the functor $`\widehat{()}:\mathrm{pro}𝒮\widehat{𝒮}`$ yields natural maps
$$K^ϵ(\{X_s\};/\mathrm{}^\nu )\widehat{KU}^ϵ(\widehat{X};/\mathrm{}^\nu )$$
for every $`\{X_s\}\mathrm{pro}𝒮`$. Furthermore, by Corollary 3.17 and (11) we have $`\pi _q(\widehat{KU}/\mathrm{}^k)/\mathrm{}^\nu `$ for $`q`$ even and $`0`$ for $`q`$ odd. Now Proposition 2.12 implies that we have, on the one hand, for every $`\{X_s\}\mathrm{pro}𝒮`$ an isomorphism of the $`E_2`$-terms of the corresponding Atiyah-Hirzebruch spectral sequences and, on the other hand, a natural map between the abutments. Both maps are compatible. Hence we have the diagram
$$\begin{array}{ccccc}\hfill E_2^p(/\mathrm{}^\nu )& =& H^p(\{X_s\};/\mathrm{}^\nu )& & K^{}(\{X_s\};/\mathrm{}^\nu )\\ & & & & \\ \hfill E_2^p(/\mathrm{}^\nu )& =& H^p(X;/\mathrm{}^\nu )& & \widehat{KU}^{}(\widehat{X};/\mathrm{}^\nu ).\end{array}$$
The two sequences are in particular both convergent if $`H^p(\{X_s\};/\mathrm{}^\nu )H^p(\widehat{X};/\mathrm{}^\nu )`$ either equals $`0`$ for $`p`$ sufficiently large or is finite for all $`p`$. Since the étale types of schemes of finite type over a separably closed field satisfy the convergence conditions of the theorem, we get the following
###### Theorem 7.11
For every scheme $`X`$ of finite type over a separably closed field $`k`$ the étale $`K`$-theory groups $`K_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )`$ of \[Fr1\] are isomorphic to the profinite étale $`K`$-theory groups $`\widehat{KU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )`$ defined above. $`\mathrm{}`$
### 7.3 Profinite étale cobordism
Following Definition 7.1 we set
###### Definition 7.12
1. We define the profinite étale cobordism of a scheme $`X`$ of finite type over $`k`$, to be the cohomology theory represented by the profinitely completed spectrum $`\widehat{MU}`$ applied to $`\widehat{\mathrm{Et}}X`$, i.e.
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X):=\widehat{MU}^{}(\widehat{\mathrm{Et}}X),$$
and profinite étale cobordism with $`/\mathrm{}^\nu `$-coefficients to be the cohomology theory represented by the profinitely completed spectrum $`\widehat{MU}/\mathrm{}^\nu `$ applied to $`\widehat{\mathrm{Et}}`$, i.e.
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu ):=(\widehat{MU}/\mathrm{}^\nu )^{}(\widehat{\mathrm{Et}}X).$$
2. We also have relative étale cobordism groups $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X,U)`$, resp. $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X,U;/\mathrm{}^\nu )`$, for an open subscheme $`UX`$ defined by
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X,U):=\widehat{MU}^{}(\widehat{\mathrm{Et}}(X)/\widehat{\mathrm{Et}}(U))$$
and
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X,U;/\mathrm{}^\nu ):=\widehat{MU}^{}(\widehat{\mathrm{Et}}(X)/\widehat{\mathrm{Et}}(U);/\mathrm{}^\nu ).$$
3. We define the étale cobordism group
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^{\mathrm{}}):=\underset{V}{lim}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}((V))$$
and similarly for $`\widehat{MU}/\mathrm{}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^\nu `$ as in Definition 7.1.
4. We define $`\mathrm{}`$-adic étale cobordism to be the limit
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;_{\mathrm{}}):=\underset{\nu }{lim}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu ).$$
###### Remark 7.13
1. The exterior product $`MU_n\times MU_mMU_{m+n}`$ on $`MU`$ together with the diagonal map $`\mathrm{\Delta }_{\widehat{\mathrm{Et}}X}:\widehat{\mathrm{Et}}X\widehat{\mathrm{Et}}X\times \widehat{\mathrm{Et}}X`$ yield a commutative ring structure on $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X)`$ and $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )`$.
2. We have introduced the notion of $`\mathrm{}`$-adic cobordism, since it is the étale cohomology theory to which $`\mathrm{}`$-adic étale cohomology converges, see Theorem 7.16. We will not give any further comment on this theory, since most of its properties follow from those of $`\widehat{MU}/\mathrm{}^\nu `$.
One should note that it is not true in general that $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;_{\mathrm{}})=\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X)`$.
The morphism of profinite spectra $`\widehat{MU}H/\mathrm{}^\nu `$ induced by the orientation also yields maps of profinite étale cohomology theories:
###### Proposition 7.14
Let $`k`$ be a field. For every $`X`$ in $`\mathrm{Sm}/k`$, there are unique induced morphisms of cohomology theories
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X)H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )$$
and
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu ).\mathrm{}$$
###### Proposition 7.15
Let $`R`$ be a strict local henselian ring, i.e. a local henselian ring with separably closed residue field. Then
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(\mathrm{Spec}R)MU^{}_{}_{\mathrm{}}.$$
In particular, for a separably closed field $`k`$ we get
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k)MU^{}_{}_{\mathrm{}}\mathrm{and}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k;/\mathrm{}^\nu )MU^{}_{}/\mathrm{}^\nu .$$
Proof This follows from Corollary 4.5 and Example 5.5. The last assertion follows from the exact sequence (12), since $`MU^{}`$ has no torsion. $`\mathrm{}`$
###### Theorem 7.16
Let $`k`$ be a field. For every scheme $`X`$ in $`\mathrm{Sm}/k`$, there is a convergent spectral sequence $`\{E_r^{p,q}\}`$ with
$$E_2^{p,q}=H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^p(X;/\mathrm{}^\nu MU^q)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{p+q}(X;/\mathrm{}^\nu ).$$
In addition, we have also a convergent spectral sequence for $`\mathrm{}`$-adic cobordism
$$E_2^{p,q}=\underset{\nu }{lim}H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^p(X;/\mathrm{}^\nu MU^q)=H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^p(X;_{\mathrm{}}MU^q)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{p+q}(X;_{\mathrm{}}).$$
Proof This is the spectral sequence of Theorem 4.11 together with the isomorphisms
$$H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu MU^q)H^{}(\widehat{\mathrm{Et}}X;/\mathrm{}^\nu MU^q)$$
of Remark 5.3 for the finite group $`M=/\mathrm{}^\nu MU^q=/\mathrm{}^\nu \widehat{MU}^q`$. The condition on the cohomology of $`X`$ for strong convergence in Theorem 4.11 is satisfied since $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^p(X;/\mathrm{}^\nu MU^q)`$ are finite groups for all $`p`$ by \[Mi\] VI, Corollary 5.5.
For the second sequence, the fact that $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^p(X;/\mathrm{}^\nu MU^q)`$ are finite groups for all $`p`$ implies that exact couples remain exact after taking limits and hence that $`E_2^{p,q}=lim_\nu E_2^{p,q}(/\mathrm{}^\nu )=H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^p(X;_{\mathrm{}}MU^q)`$ converges to $`lim_\nu \widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{p+q}(X;/\mathrm{}^\nu )=\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{p+q}(X;_{\mathrm{}})`$. $`\mathrm{}`$
###### Remark 7.17
1. Since $`/\mathrm{}^\nu MU^t`$ is a finitely generated free $`/\mathrm{}^\nu `$-module, i.e. isomorphic to a direct sum of finitely many copies of $`/\mathrm{}^\nu `$, and since étale cohomology is compatible with direct limits, the group $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^s(X;/\mathrm{}^\nu MU^t)`$ is isomorphic to $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^s(X;/\mathrm{}^\nu )MU^t`$. Hence in order to calculate the groups $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^s(X;/\mathrm{}^\nu MU^t)`$, we may reduce the problem to determine the groups $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^s(X;/\mathrm{}^\nu )`$.
2. By Theorem 4.11, we get a spectral sequence
$$E_2^{p,q}=H^p(\widehat{\mathrm{Et}}X;\widehat{MU}^q)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{p+q}(X).$$
But the problem is that, in general, we do not have an isomorphism $`H^p(\widehat{\mathrm{Et}}X;\widehat{MU}^q)H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^p(X;\widehat{MU}^q)`$, since $`\widehat{MU}^q`$ is infinite. In addition, the étale cohomology with $`_{\mathrm{}}`$-coefficients does not have good properties.
###### Theorem 7.18
Let $`X`$ be an algebraic variety over $``$. Let $`X()`$ be the topological space of complex points. For every $`\nu `$, there is an isomorphism
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )MU^{}(X();/\mathrm{}^\nu ).$$
Proof By Corollary 3.16 we know that
$$(\widehat{MU}/\mathrm{}^\nu )^q(MU/\mathrm{}^\nu )^qMU^q/\mathrm{}^\nu .$$
Since $`\pi _q(MU)`$ is a finitely generated abelian group, we get that $`MU^q/\mathrm{}^k`$ is a finite group. We denote by $`\mathrm{Sing}(X())`$ the singular simplicial set of the topological space $`X()`$. By Theorem 8.4 and Corollary 8.5 of \[Fr2\], by Remark 5.3 and Corollary 3.16 we get a map of spectral sequences and an isomorphism
$$H^p(\mathrm{Sing}(X());(MU/\mathrm{}^\nu )^q)H^p(\widehat{\mathrm{Et}}X;(\widehat{MU}/\mathrm{}^\nu )^q)$$
between singular and profinite étale cohomology for all $`\nu `$, $`p`$ and $`q`$. Since $`X`$ is finite dimensional, the corresponding complex, respectively étale Atiyah-Hirzebruch spectral sequences are convergent with isomorphic $`E_2`$-terms. $`\mathrm{}`$
Also the following statements are easy consequences of the existence of the spectral sequence and results on étale cohomology.
###### Proposition 7.19
Let $`A`$ be a strict local henselian ring and $`S=\mathrm{Spec}A`$. Let $`f:XS`$ be a proper morphism and let $`X_0`$ be the closed fibre of $`f`$. Then the induced map
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )\stackrel{}{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X_0;/\mathrm{}^\nu )$$
is an isomorphism on étale cobordism with $`/\mathrm{}^\nu `$-coefficients.
Proof The morphism $`X_0X`$ induces a morphism of Atiyah-Hirzebruch spectral sequences. By Theorem 1.2 of Chapter IV in \[SGA$`4\frac{1}{2}`$\] Arcata, it induces an isomorphism on étale cohomology with $`/\mathrm{}^\nu `$-coefficients and hence, by Remark 7.17 an isomorphism on étale cobordism with $`/\mathrm{}^\nu `$-coefficients. $`\mathrm{}`$
###### Proposition 7.20
Let $`K/k`$ be an extension of separably closed fields with characteristic different from $`\mathrm{}`$. Let $`X`$ be a $`k`$-scheme. The induced map
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )\stackrel{}{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X_K;/\mathrm{}^\nu )$$
is an isomorphism on étale cobordism with $`/\mathrm{}^\nu `$-coefficients.
Proof Again the natural morphism $`X_KX`$ induces a map of spectral sequences and the assertion then follows from the isomorphism on étale cohomology with $`/\mathrm{}^\nu `$-coefficients of Corollaire 3.3 of Chapter V in \[SGA$`4\frac{1}{2}`$\] Arcata and Remark 7.17. $`\mathrm{}`$
### 7.4 Profinite étale cobordism is an oriented cohomology theory
We prove that $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2()`$ and $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(;/\mathrm{}^\nu )`$ are oriented cohomology theories in the sense of \[LM\] on the category $`\mathrm{Sm}/k`$ of smooth quasi-projective schemes of finite type over a separably closed field $`k`$ with char $`k\mathrm{}`$.
We have to check the axioms given in \[LM\]. The axioms of a cohomology theory have already been shown for every étale cohomology theory. In order to check the axioms of an oriented cohomology theory for $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$, our strategy is as follows. We define an orientation class for $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$ and use it to define the first Chern class of a line bundle. Using the Chern class of the canonical line bundle on the projective $`n`$-space, we construct an explicit isomorphism for the cobordism of $`_k^n`$. The method to prove this isomorphism relies on the Atiyah-Hirzebruch spectral sequence for profinite cobordism. We use the spectral sequence to reduce the computation of the cobordism of $`^n`$ to the computation of the étale cohomology of $`^n`$ with $`/\mathrm{}`$-coefficients. This reduction uses non-trivial results on profinite cobordism and is the key point for the proof that $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}`$, resp. $`(\widehat{MU}/\mathrm{}^\nu )_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}`$, yields an oriented cohomology theory on $`\mathrm{Sm}/k`$.
This computation and the Künneth theorem for $`\widehat{MU}`$, resp. $`\widehat{MU}/\mathrm{}^\nu `$, yield a projective bundle formula. Then we use Grothendieck’s method to define Chern classes for all vector bundles.
The general theory of cohomology theories with Chern classes given in \[Pa\] then implies all the remaining properties.
But one should note that the fact that $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$ is an oriented cohomology theory on $`\mathrm{Sm}/k`$ is a nontrivial result and does not just follow from the fact that $`MU`$ is an oriented theory on topological spaces. This is due to the fact that we may not use an embedding of $`k`$ in $``$. We need an orientation with respect to smooth schemes over any separably closed field.
To deal with the twists in étale cohomology we choose a primitive $`\mathrm{}^\nu `$-th-root of unity which yields an isomorphism $`/\mathrm{}^\nu /\mathrm{}^\nu (1)`$, see Remarks 7.34 and 8.11.
###### Definition 7.21
Let $`_k^{\mathrm{}}`$ be the infinite projective space in the category $`LU(k)`$ of spaces over $`k`$ and let $`E`$ be a commutative ring spectrum in $`\mathrm{Sp}(\widehat{𝒮}_{})`$. Since we have an isomorphism $`\widehat{\mathrm{Et}}_k^1S^2`$ in $`\widehat{}`$ by Example 5.10, we have a canonical map
$$E^2(\widehat{\mathrm{Et}}_k^{\mathrm{}})E^2(\widehat{\mathrm{Et}}_k^1)\stackrel{}{}E^2(S^2)\stackrel{}{}\pi _0(E).$$
An orientation for $`E`$ with respect to $`\mathrm{Sm}/k`$ is a class $`x_E\stackrel{~}{E}^2(\widehat{\mathrm{Et}}^{\mathrm{}})`$ that maps to $`1`$ under the above map.
###### Example 7.22
1. For $`\widehat{MU}`$, resp. $`\widehat{MU}/\mathrm{}^\nu `$, we choose the orientation $`x_{\widehat{MU}}\widehat{MU}^2(\widehat{\mathrm{Et}}_k^{\mathrm{}})`$, resp. $`x_{\widehat{MU}/\mathrm{}^\nu }\widehat{MU}^2(\widehat{\mathrm{Et}}_k^{\mathrm{}};/\mathrm{}^\nu )`$, as the image of the orientation $`x_{MGL}`$ under $`\widehat{\mathrm{Et}}`$. This means that we define the map $`x_{\widehat{MU}}:\mathrm{L}\widehat{\mathrm{Et}}(^{\mathrm{}})\widehat{MU}S^2`$ in the preimage of $`1\pi _0(\widehat{MU})`$ such that the following diagram
$$\begin{array}{ccc}\hfill \widehat{\mathrm{Et}}(^{\mathrm{}})& \stackrel{\stackrel{~}{x}_{\widehat{MU}}}{}& \widehat{MU}S^2\\ \hfill \mathrm{L}\widehat{\mathrm{Et}}(x_{MGL})& \varphi & \\ \hfill \mathrm{L}\widehat{\mathrm{Et}}(MGL)^1& & \end{array}$$
is commutative, where $`\varphi `$ is the morphism defined in the next section using an étale Thom class.
Note that $`x_{\widehat{MU}}\widehat{MU}^2(\widehat{\mathrm{Et}}_k^{\mathrm{}})`$ also defines an element $`\stackrel{~}{x}_{\widehat{MU}}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(_k^{\mathrm{}})`$ via the canonical map $`\widehat{MU}^{}(\widehat{\mathrm{Et}}^{\mathrm{}})\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(^{\mathrm{}})`$, recall Definition 7.1 and Remark 7.2.
If $`k=`$, this orientation $`x_{\widehat{MU}}`$ is mapped to the classical complex orientation of $`MU`$ via the isomorphism of Proposition 7.18. This is due to the fact that $`x_{MGL}`$ corresponds to the complex orientation of $`MU`$ under the complex realization functor, \[Vo1\].
2. For $`H/\mathrm{}^\nu `$ an orientation $`x_{H/\mathrm{}^\nu }`$ is given by a generator of $`H^2(\widehat{\mathrm{Et}}_k^{\mathrm{}};/\mathrm{}^\nu )/\mathrm{}^\nu `$. This last isomorphism is deduced from
$$H^2(\widehat{\mathrm{Et}}_k^{\mathrm{}};/\mathrm{}^\nu )H^2(\mathrm{Et}_k^{\mathrm{}};/\mathrm{}^\nu )\underset{n}{lim}H^2(\mathrm{Et}_k^n;/\mathrm{}^\nu )/\mathrm{}^\nu ,$$
where we use the fact that $`\mathrm{Et}`$ commutes with colimits.
The importance of an orientation of $`\widehat{MU}`$, resp. $`\widehat{MU}/\mathrm{}^\nu `$, is that it enables us to define the first Chern class of a line bundle.
A line bundle $`L/X`$ over a scheme $`X`$ of dimension $`n`$ over $`k`$ corresponds to a morphism $`X_k^{n+1}`$ unique up to $`𝔸^1`$-weak equivalences. Since $`\mathrm{L}\widehat{\mathrm{Et}}`$ agrees with $`\widehat{\mathrm{Et}}`$ on smooth schemes by Theorem 5.14, this also defines a unique map $`\lambda :\widehat{\mathrm{Et}}X\widehat{\mathrm{Et}}_k^{n+1}`$ in $`\widehat{}`$.
It is clear from the definition that the orientation $`x_{\widehat{MU}}`$ restricted to $`\widehat{\mathrm{Et}}_k^{n+1}`$ corresponds to a map $`x_{\widehat{MU}}:\widehat{\mathrm{\Sigma }}^{\mathrm{}}(\widehat{\mathrm{Et}}_k^{n+1})\widehat{MU}S^2`$. The composition of these two maps is the first Chern class of $`L/X`$.
###### Definition 7.23
A line bundle defines via composition a morphism in $`\widehat{𝒮}`$
$$c_1(L):\widehat{\mathrm{\Sigma }}^{\mathrm{}}(\widehat{\mathrm{Et}}X)\stackrel{\lambda }{}\widehat{\mathrm{\Sigma }}^{\mathrm{}}(\widehat{\mathrm{Et}}_k^{n+1})\stackrel{x_{\widehat{MU}}}{}\widehat{MU}S^2,$$
i.e. an element $`c_1(L)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(X)`$ which we call the first Chern class of $`L`$.
Similarly, we define the first Chern class of a line bundle in $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(X;/\mathrm{}^\nu )`$.
Now we have to show that this Chern class for line bundles yields a Chern structure on étale cobordism. Therefore, we have to calculate the étale cobordism of the projective $`n`$-space $`_k^n`$ over $`k`$. The method we use is indirectly based on the Atiyah-Hirzebruch spectral sequence for profinite cobordism. We use the Atiyah-Hirzebruch spectral sequence to reduce the computation of $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(^n)`$ to the well-known case of étale cohomology $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(^n;/\mathrm{})`$ with $`/\mathrm{}`$\- coefficients. A crucial point is the isomorphism
(29)
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X)/(\mathrm{}\widehat{MU}^0)H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{})$$
for every $`X`$ in $`\mathrm{Sm}/k`$.
###### Proposition 7.24
Let $`k`$ be a separably closed field. Let $`_k^n`$ be the projective space of dimension $`n`$ over $`k`$. Let $`𝒪(1)_k^n`$ be the canonical quotient line bundle. 1. Set $`\xi :=c_1(𝒪(1))\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(_k^n)`$. The étale cobordism of $`_k^n`$ is given by the isomorphism
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n)\widehat{MU}^{}[u]/(u^{n+1}),\xi u.$$
2. Set $`\xi :=c_1(𝒪(1))\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(_k^n;/\mathrm{}^\nu )`$. The étale cobordism with $`/\mathrm{}^\nu `$-coefficients of $`_k^n`$ is given by the isomorphism
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n;/\mathrm{}^\nu )(\widehat{MU}/\mathrm{}^\nu )^{})[u]/(u^{n+1}),\xi u.$$
Proof We deduce both results from general facts about filtered $`MU^{}`$-modules in \[De\]. The crucial point is that $`H^{}(\widehat{\mathrm{Et}}_k^n;/\mathrm{}^m)H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n;/\mathrm{}^m)`$ surjects onto $`H^{}(\widehat{\mathrm{Et}}_k^n;/\mathrm{})`$ for every $`m`$, cf. \[Mi\] VI, 5.6. Hence $`\widehat{\mathrm{Et}}_k^n`$ is a profinite space of finite dimension without $`\mathrm{}`$-torsion, see the discussion of Theorem 4.13.
By Proposition 2.1.8 of \[De\] respectively by definition, both $`MU^{}`$-modules $`M:=\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n)`$ and $`N:=\widehat{MU}^{}[u]/(u^{n+1})`$ are objects of the category $`\widehat{}`$ of free $`MU^{}`$-modules that are complete with respect to the $`\mathrm{}`$-adic filtration $`f^{}`$ of \[De\], §2.1.
By Corollaire 2.1.4 of \[De\], it suffices to show that $`M/f^1N/f^1`$ as $`/\mathrm{}`$-vector spaces. But since $`\widehat{\mathrm{Et}}_k^n`$ is a finite dimensional profinite space, the isomorphism $`M/f^1N/f^1`$ follows from the isomorphism (29), proved in Proposition 2.1.8 of \[De\] together with Remark 5.3, and the isomorphism of étale cohomology $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n;/\mathrm{})/\mathrm{}[u]/(u^{n+1})`$. This proves the assertion for $`\widehat{MU}`$. One should remark that the proof of Proposition 2.1.8 of \[De\] is based on the Atiyah-Hirzebruch spectral sequence for $`\widehat{MU}`$.
The second assertion follows from the first one using the fact that for a space $`X`$ without $`\mathrm{}`$-torsion, we have $`\widehat{MU}^{}(X;/\mathrm{}^\nu )\widehat{MU}^{}(X)/\mathrm{}^\nu `$, see \[Ad\] III, Proposition 6.6.
We could deduce the second assertion also directly from the Atiyah-Hirzebruch spectral sequence for $`\widehat{MU}/\mathrm{}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^\nu `$ as in \[Ad\] I, Lemma 2.5, using the determination of $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n;/\mathrm{}^\nu )`$, since all differentials vanish. $`\mathrm{}`$
###### Proposition 7.25
For the projective space $`_k^n`$ over a separably closed field $`k`$ and every $`X\mathrm{Sm}/k`$, there are Künneth isomorphisms for the étale cobordism
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X)_{\widehat{MU}^{}}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n)\stackrel{}{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X\times _k^n)$$
and
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )_{(\widehat{MU}/\mathrm{}^\nu )^{}}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n;/\mathrm{}^\nu )\stackrel{}{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X\times _k^n;/\mathrm{}^\nu ).$$
Proof Again, the crucial point is that $`H^{}(\widehat{\mathrm{Et}}_k^n;/\mathrm{}^m)H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n;/\mathrm{}^m)`$ surjects onto $`H^{}(\widehat{\mathrm{Et}}_k^n;/\mathrm{})`$ for every $`m`$. Hence $`\widehat{\mathrm{Et}}_k^n`$ is a profinite space without $`\mathrm{}`$-torsion. Over the separably closed field $`k`$, $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^i(X;/\mathrm{}^\nu )`$ and $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^i(_k^n;/\mathrm{})`$ vanish for large $`i`$ and hence the profinite spaces $`\widehat{\mathrm{Et}}X`$ and $`\widehat{\mathrm{Et}}_k^n`$ are finite dimensional spaces in $`\widehat{𝒮}`$. This enables us to deduce the proposition from Theorem 4.13. $`\mathrm{}`$
###### Theorem 7.26
Projective Bundle Formula
Let $`EX`$ be a rank $`n`$ vector bundle over $`X`$ in $`\mathrm{Sm}/k`$, $`𝒪(1)(E)`$ the canonical quotient bundle with zero section $`s:(E)𝒪(1)`$. Set $`\xi :=c_1(𝒪(1))\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2((E))`$. Then $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}((E))`$ is a free $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X)`$-module with basis $`(1,\xi ,\mathrm{},\xi ^{n1}).`$
Similarly, $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}((E);/\mathrm{}^\nu )`$ is a free $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )`$-module with basis $`(1,\xi ,\mathrm{},\xi ^{n1}),`$ where $`\xi :=c_1(𝒪(1))\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2((E);/\mathrm{}^\nu )`$.
Proof We consider $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$. We prove this theorem first for the case of a trivial bundle on $`X`$. We have to show that the map $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X\times _k^n)\stackrel{}{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X)[u]/(u^n)`$, sending $`\xi `$ to $`u`$, is an isomorphism. By Proposition 5.4, we know that the canonical map $`\widehat{\mathrm{Et}}(X)\times \widehat{\mathrm{Et}}(_k^n)\widehat{\mathrm{Et}}(X\times _k^n)`$ is a weak equivalence in $`\widehat{𝒮}`$. Proposition 7.24 yields an isomorphism
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X)[u]/(u^n)\stackrel{}{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X)_{\widehat{MU}^{}}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n).$$
Hence it suffices to prove that the projections induce an isomorphism
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X)_{\widehat{MU}^{}}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n)\stackrel{}{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X\times _k^n).$$
This follows from the previous proposition and finishes the proof of the theorem in the case that $`E`$ is a trivial bundle on $`X`$.
A similar argument using the second assertion of Proposition 7.24 proves the case of a trivial bundle for $`\widehat{MU}/\mathrm{}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^\nu `$.
For the general case, since $`E`$ is by definition locally trivial for the Zariski topology on $`X`$, it suffices to show that the theorem holds for $`X`$ if it holds for open subsets $`X_0`$, $`X_1`$ and $`X_0X_1`$, with $`X=X_0X_1`$. This point now follows from comparing the two Mayer-Vietoris-sequences
$$\begin{array}{c}\mathrm{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^i((E_0))\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^i((E_1))\\ \widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^i((E_{01}))\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{i+1}((E))\mathrm{}\end{array}$$
$$\begin{array}{c}\widehat{MU}^{}[u]/(u^n)(\mathrm{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^i(X_0)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^i(X_1)\\ \widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^i(X_{01})\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{i+1}(X)\mathrm{})\end{array}$$
where we have written $`X_{01}`$ for $`X_0X_1`$, $`E_0`$, $`E_1`$ and $`E_{01}`$ for the restrictions of $`E`$ to $`X_0`$, $`X_1`$ and $`X_{01}`$.
Again, a similar argument shows the general case for $`\widehat{MU}/\mathrm{}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^\nu `$. $`\mathrm{}`$
We use Grothendieck’s idea to introduce higher Chern classes for vector bundles.
###### Definition 7.27
Let $`E`$ be a vector bundle of rank $`n`$ on $`X`$ in $`\mathrm{Sm}/k`$ and let $`\xi :=c_1(𝒪(1))`$ be as above. Then we define the $`i`$th Chern class of $`E`$ to be the element $`c_i(E)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{2i}(X)`$ such that $`c_0(E)=1`$, $`c_i(E)=0`$ for $`i>n`$ and $`\mathrm{\Sigma }_{i=0}^n(1)^ic_i(E)\xi ^{ni}=0`$. By the projective bundle formula of Theorem 7.26 these elements are unique.
Similarly, we define $`c_i(E)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{2i}(X;/\mathrm{}^\nu )`$ by Theorem 7.26.
###### Lemma 7.28
1. For each line bundle over a smooth $`X`$, the class $`c_1(L)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(X)`$, resp. $`c_1(L)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(X;/\mathrm{}^\nu )`$, is nilpotent.
2. The assignment of Chern classes is functorial with respect to the pullback, i.e. $`f^{}(c_i(E))=c_i(f^{}E)`$ for a bundle $`E`$ on $`X`$ and a morphism $`f:YX`$ in $`\mathrm{Sm}/k`$.
Proof 1. This may be proved exactly as in \[Pa\], Lemma 3.6.4.
2. This follows immediately from Theorem 7.26. $`\mathrm{}`$
###### Theorem 7.29
The étale cobordism theories satisfy the axioms of an oriented ring cohomology theory on $`\mathrm{Sm}/k`$ of \[Pa\], Definition 3.1.1. This implies that for every projective morphism $`f:YX`$ of codimension $`d`$ there are natural transfer maps
$$f_{}:\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(Y)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{+2d}(X)$$
and
$$f_{}:\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(Y;/\mathrm{}^\nu )\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{+2d}(X;/\mathrm{}^\nu )$$
which satisfy all the axioms of a push-forward in an oriented cohomology theory of \[LM\].
Proof We have proved in Theorem 7.8 that $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}()`$ and $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}(;/\mathrm{}^\nu )`$ satisfy the axioms of a cohomology theory in the sense of \[Pa\], Definition 2.0.1. By Remark 7.13, we know that $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}()`$, resp. $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}(;/\mathrm{}^\nu )`$, is also a ring cohomology theory.
Secondly, the orientation on $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}()`$, resp. $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(;/\mathrm{}^\nu )`$, and Theorem 7.26 show that we have in fact an oriented theory with Chern classes in the sense of \[Pa\], Definition 3.1.1, using Theorem 3.2.4 of \[Pa\].
Now there is a general procedure, described in 4.2 and 4.3 of \[Pa\], to define natural transfer maps for every projective morphism in $`\mathrm{Sm}/k`$. Note that since we require transfer maps only for projective maps, we only have to construct transfers for closed immersions of smooth schemes and the projection of the projective space $`_X^nX`$ over $`X`$. In Theorem 4.7.1 of \[Pa\] it is proved that the push-forward is independent of the decomposition of $`f`$.
Furthermore, also the compatibility of pullbacks and push-forwards for transversal morphisms is proved in Theorem 4.7.1 of \[Pa\].
That $`f_{}`$ shifts the degree by $`2d`$ comes from the fact that Chern classes live in even degrees and the fact that the construction of $`f_{}`$ depends on Chern classes.
This finishes the proof of the theorem. $`\mathrm{}`$
We state a general consequence of this extra structure on étale cobordism.
###### Corollary 7.30
For every closed embedding of smooth varieties $`i:ZX`$ with open complement $`j:UX`$ there are exact sequences
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(Z)\stackrel{i_{}}{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X)\stackrel{j^{}}{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(U)$$
and
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(Z;/\mathrm{}^\nu )\stackrel{i_{}}{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )\stackrel{j^{}}{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(U;/\mathrm{}^\nu ).$$
We call them the Gysin sequences for $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}()`$, resp. $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(;/\mathrm{}^\nu )`$.
Proof This is proved in \[Pa\] 4.4.2. $`\mathrm{}`$
As another corollary of the last theorem we get
###### Corollary 7.31
Let $`k`$ be a separably closed field and $`\mathrm{}`$ a prime number different from the characteristic of $`k`$. Profinite étale cobordism $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2()`$, resp. $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(;/\mathrm{}^\nu )`$, is an oriented cohomology theory on $`\mathrm{Sm}/k`$ in the sense of \[LM\]. $`\mathrm{}`$
Using Theorem 7.26 we can identify $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(^n)`$ with the ring $`\widehat{MU}^{}[u]/(u^{n+1})`$, identifying $`c_1(𝒪(1))`$ with $`u`$, and in the same way
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(^n\times ^n)=\widehat{MU}^{}[u,v]/(u^{n+1},v^{n+1}).$$
As Quillen pointed out in \[Qu3\], Proposition 2.7, we can associate to the theory $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$, resp. $`\widehat{MU}/\mathrm{}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^\nu `$, a formal group law, i.e. a power series $`F(u,v)\widehat{MU}^{}[[u,v]]`$, $`F(u,v)=\mathrm{\Sigma }_{i,j0}a_{ij}u^iv^j`$ satisfying some axioms with $`a_{ij}\widehat{MU}^{22i2j}`$, to an oriented cohomology theory such that
$$c_1(LM)=F(c_1(L),c_1(M))$$
for any two line bundles $`L`$, $`M`$ over $`X`$.
Let $`𝕃_{}`$ be the Lazard ring. There is Quillen’s famous theorem that $`𝕃_{}\pi _2(MU)`$ is an isomorphism.
We have already determined the coefficients of $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$ over a separably closed field to be $`\widehat{MU}^{}MU^{}_{}_{\mathrm{}}`$, resp. $`(\widehat{MU}/\mathrm{}^\nu )^{}MU^{}_{}/\mathrm{}^\nu `$, see Proposition 7.15. But we have to check that the natural morphism $`\mathrm{\Phi }:𝕃^{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}`$ induced by the formal group law on $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$ agrees with this isomorphism.
###### Proposition 7.32
Let $`k`$ be a separably closed field. The map $`\mathrm{\Phi }_{\mathrm{}}:𝕃^{}_{\mathrm{}}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(k)`$ induced by the natural map of formal group laws is an isomorphism.
Similarly, the induced map $`\mathrm{\Phi }/\mathrm{}^\nu :𝕃^{}/\mathrm{}^\nu \widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(k;/\mathrm{}^\nu )`$ is an isomorphism.
Proof We consider the induced map $`\mathrm{\Phi }_{\mathrm{}}:𝕃^{}_{\mathrm{}}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(k)`$. It is clear that this map $`\mathrm{\Phi }_{\mathrm{}}`$ is a morphism in the category $`\widehat{}`$ of \[De\], §2.1, that we already used in the proof of Proposition 7.24. Furthermore, it is clear that this map induces an isomorphism $`/\mathrm{}=𝕃^{}_{\mathrm{}}/f^1\stackrel{}{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(k)/f^1/\mathrm{}`$ modulo the filtration $`f^{}`$ of \[De\], §2.1. By \[De\], Corollaire 2.1.4, this implies that the $`\mathrm{\Phi }_{\mathrm{}}`$ is an isomorphism.
The second assertion follows from the first one by the coefficient exact sequence. $`\mathrm{}`$
###### Remark 7.33
The proof of the last proposition is similar to the proof of $`\mathrm{\Phi }:𝕃^{}MGL^{2,}(k)`$ by Hopkins and Morel \[HM\]. The idea is to construct a spectrum $`MGL/(x_1,x_2,\mathrm{})`$, where the $`x_i`$’s are the generators of $`𝕃^{}[x_1,x_2,\mathrm{}]`$. The crucial point for the proof of the surjectivity of the map $`\mathrm{\Phi }:𝕃^{}MGL^{2,}(k)`$ is to show that the map of spectra
$$MGL/(x_1,x_2,\mathrm{})H$$
is an isomorphism, where $`H`$ denotes the motivic Eilenberg-MacLane spectrum. In the proof of Proposition 7.32, this corresponds to the isomorphism $`/\mathrm{}=𝕃^{}_{\mathrm{}}/f^1\stackrel{}{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(k)/f^1H^{}(\widehat{\mathrm{Et}}k;/\mathrm{})=/\mathrm{}`$.
Then one shows by induction on $`n`$ that $`𝕃^nMGL^{2n,n}(k)`$ is surjective.
The injectivity of $`\mathrm{\Phi }`$ can be shown as for $`𝕃^{}\mathrm{\Omega }^{}(k)`$ in \[LM\].
###### Remark 7.34
The proof of Theorem 7.29 is based on two results. On the one hand, we used Panin’s discussion and results on oriented cohomology theories to deduce the missing structures from the existence of an orientation and the projective bundle bundle formula. On the other hand, we had to prove the projective bundle formula. For this purpose, we used the Atiyah-Hirzebruch spectral sequence to reduce the problem to the projective bundle formula for étale cohomology.
Although we worked over a separably closed base field, étale cohomology $`_qH_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(;\mu _\mathrm{}^\nu ^q)`$ is an oriented theory over any field $`k`$ with char $`k\mathrm{}`$. The problem for our argument arises with the twist of the coefficients. Nevertheless, it seems to be very likely that the above proof also works over any field with characteristic prime to $`\mathrm{}`$. Maybe one needs to modify the definition of étale cobordism for this purpose.
One could start with a naive version of twisted coefficients for $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$ as follows. For any field $`k`$, let $`\mu _\mathrm{}^\nu `$ be the group of $`\mathrm{}^\nu `$th roots of unity in $`k`$. Since we can form a Moore spectrum $`MG`$ for every finite abelian group, we may consider the twisted Moore spectrum $`M(\mu _\mathrm{}^\nu ^q)=M/\mathrm{}^\nu (q)`$. Since $`\mu _\mathrm{}^\nu `$ is finite, we can consider the profinite completion $`\widehat{MU}/\mathrm{}^\nu (q)`$ of $`\mathrm{MU}M/\mathrm{}^\nu (q)`$. We could define profinite étale cobordism with twisted coefficients $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(;/\mathrm{}^\nu ())`$ to be the profinite cohomology represented by the profinite spectrum $`\widehat{MU}/\mathrm{}^\nu ()`$ applied to the functor $`\widehat{\mathrm{Et}}`$ on schemes over $`k`$.
If $`k`$ is a separably closed field and if $`X\mathrm{Sm}/k`$, we have a canonical isomorphism
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu (r))\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )/\mathrm{}^\nu (r)$$
for every $`r`$.
For, we have a canonical isomorphism for étale cohomology for every $`r`$
$$E_2^{p,q}=H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^p(X;MU^q/\mathrm{}^\nu (r))H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^p(X;MU^q/\mathrm{}^\nu )/\mathrm{}^\nu (r)=:\stackrel{~}{E}_2^{p,q}.$$
The two converging Atiyah-Hirzebruch spectral sequences imply the result.
But the geometry of $`\widehat{𝒮}`$ does not seem to reflect this extra structure. It is more likely, that one has to find another way to define twisted étale cobordism to handle this problem.
### 7.5 Etale Cobordism over finite fields
In the case of a finite base field $`k=𝔽_q`$ of characteristic $`p\mathrm{}`$ and with $`q=p^r`$ a power of $`p`$, we can also show that étale cobordism is an oriented cohomology theory on $`\mathrm{Sm}/k`$. This is a remarkable feature of this étale topological theory.
Again, we have to prove that $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}()`$ satisfies a projective bundle formula. We proceed in an analgue way as before.
###### Proposition 7.35
1. The profinite space $`\widehat{\mathrm{Et}}k`$ is a space without $`\mathrm{}`$-torsion in $`\widehat{𝒮}`$.
2. The étale cobordism of the finite field $`k`$ is given by the following isomorphism
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^n(k)=\{\begin{array}{cc}\hfill \widehat{MU}^n:& n\mathrm{even}\hfill \\ \hfill \widehat{MU}^{n1}:& n\mathrm{odd}.\hfill \end{array}$$
We get an analogue result for $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k;/\mathrm{})`$.
Proof 1. The cohomology $`H^i(\widehat{\mathrm{Et}}k;/\mathrm{}^m)=H^i(G_k;/\mathrm{}^m)`$ surjects onto $`H^i(\widehat{\mathrm{Et}}k;/\mathrm{})`$ for every $`m`$, since the Galois cohomology $`H^i(G_k;/\mathrm{}^m)`$ of $`k`$ with coefficients in the trivial $`G_k`$-module $`/\mathrm{}^m`$ equals $`/\mathrm{}^m`$ for $`i=0,1`$ and vanishes for $`i>1`$ by \[Se\].
2. Let us denote the right hand side of the equation by $`M^{}`$. The isomorphism follows from the fact that $`\widehat{\mathrm{Et}}k`$ has no $`\mathrm{}`$-torsion and hence $`\widehat{MU}^{}(\widehat{\mathrm{Et}}k)/f^1H^{}(\widehat{\mathrm{Et}}k;/\mathrm{})`$ is an isomorphism by Proposition2.1.8 of \[De\]. Now $`H^i(\widehat{\mathrm{Et}}k;/\mathrm{})`$ equals $`/\mathrm{}`$ for $`i=0,1`$ and vanishes otherwise. Hence we have also an isomorphism $`M^{}/f^1H^{}(\widehat{\mathrm{Et}}k;/\mathrm{})`$ of free $`MU^{}`$-algebras in $`\widehat{}`$. By Corollaire 2.1.4 of \[De\], this implies the second assertion. $`\mathrm{}`$
###### Proposition 7.36
The profinite space $`\widehat{\mathrm{Et}}_k^n`$ is a space without $`\mathrm{}`$-torsion in $`\widehat{𝒮}`$ for every $`n`$.
Proof By the projective bundle formula for étale cohomology, the cohomology $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n;/\mathrm{}^m)`$ of $`_k^n`$ is a free module over $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k;/\mathrm{}^m)`$ for every $`m`$. Since $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k;/\mathrm{}^m)`$ surjects onto $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k;/\mathrm{})`$ for every $`m`$, we deduce that the map $`H^{}(\widehat{\mathrm{Et}}_k^n;/\mathrm{}^m)H^{}(\widehat{\mathrm{Et}}_k^n;/\mathrm{})`$ is surjective for every $`m`$, too. $`\mathrm{}`$
###### Proposition 7.37
Let $`k`$ be as above and let $`_k^n`$ be the projective space of dimension $`n`$ over $`k`$.
1. The étale cobordism of $`_k^n`$ is given by the isomorphism
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k)[u]/(u^{n+1}).$$
2. The étale cobordism with $`/\mathrm{}`$-coefficients of $`_k^n`$ is given by the isomorphism
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n;/\mathrm{})\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k;/\mathrm{})[u]/(u^{n+1}).$$
Proof By Proposition 7.36 and Remark 7.35 above, $`\widehat{\mathrm{Et}}_k^n`$ and $`\widehat{\mathrm{Et}}k`$ are profinite spaces without $`\mathrm{}`$-torsion. Now one can continue as above. $`\mathrm{}`$
###### Proposition 7.38
For every $`X\mathrm{Sm}/k`$ the canonical map $`\widehat{\mathrm{Et}}(_k^n\times _kX)\widehat{\mathrm{Et}}_k^n\times _{\widehat{\mathrm{Et}}k}\widehat{\mathrm{Et}}X`$ is a weak equivalence in $`\widehat{𝒮}`$.
Proof There is a relative Künneth spectral sequence for the category $`\widehat{𝒮}/Z`$ of objects over $`Z`$
$$E_2=\mathrm{Tor}_{H^{}(Z;/\mathrm{})}(H^{}(Y;/\mathrm{}),H^{}(X;/\mathrm{}))$$
converging to $`H^{}(Y\times _ZX;/\mathrm{})`$. One can prove this result using the corresponding Künneth spectral sequence for simplicial finite sets and using the fact that colimits respect tensor products, see e.g. \[McC\] Theorem 8.34.
Since $`H^{}(\widehat{\mathrm{Et}}_k^n;/\mathrm{})`$ is a free $`H^{}(\widehat{\mathrm{Et}}k;/\mathrm{})`$-algebra, the Tor-term in the spectral sequence for $`Y=\widehat{\mathrm{Et}}_k^n`$ and $`Z=\widehat{\mathrm{Et}}k`$ is trivial. Hence in this special case, we deduce an isomorphism
$$H^{}(\widehat{\mathrm{Et}}_k^n;/\mathrm{})_{H^{}(\widehat{\mathrm{Et}}k;/\mathrm{})}H^{}(\widehat{\mathrm{Et}}X;/\mathrm{})\stackrel{}{}H^{}(\widehat{\mathrm{Et}}_k^n\times _{\widehat{\mathrm{Et}}k}\widehat{\mathrm{Et}}X;/\mathrm{})$$
for every $`X\mathrm{Sm}/k`$. Since the left hand side is isomorphic to $`H^{}(\widehat{\mathrm{Et}}(_k^n\times _kX);/\mathrm{})`$ by the projective bundle formula for étale cohomology, we have proved the assertion of the proposition. $`\mathrm{}`$
The next step is a relative Künneth formula for profinite cobordism.
###### Proposition 7.39
Let $`Z\widehat{𝒮}`$ and let $`X`$ and $`Y`$ be objects in the category of profinite spaces over $`Z`$. We suppose that $`Y`$ and $`Z`$ are finite dimensional spaces without $`\mathrm{}`$-torsion such that both $`H^{}(Y;/\mathrm{})`$ is a free $`H^{}(Z;/\mathrm{})`$-algebra and $`\widehat{MU}^{}(Y)`$ is a free $`\widehat{MU}^{}(Z)`$-algebra, respectively $`\widehat{MU}^{}(Y;/\mathrm{})`$ is a free $`\widehat{MU}^{}(Z;/\mathrm{})`$-algebra. Furthermore, we suppose that the $`/\mathrm{}`$-cohomology of $`X`$ is finite in all degrees.
Then the canonical maps
$$\widehat{MU}^{}(Y)_{\widehat{MU}^{}(Z)}\widehat{MU}^{}(X)\stackrel{}{}\widehat{MU}^{}(Y_ZX)$$
and
$$\widehat{MU}^{}(Y;/\mathrm{})_{\widehat{MU}^{}(Z;/\mathrm{})}\widehat{MU}^{}(X;/\mathrm{})\stackrel{}{}\widehat{MU}^{}(Y_ZX;/\mathrm{})$$
are isomorphisms.
Proof I am very grateful to Francois-Xavier Dehon for an explanation of this point. We will prove the assertion following his argument.
We suppose first that both $`X`$ and $`Y`$ are finite dimensional without $`\mathrm{}`$-torsion. As described in the proof of Proposition 2.1.8 in \[De\], the spectral sequence for $`MU`$ degenerates at $`E_2`$. Since $`X`$, $`Y`$ and $`Z`$ have no $`\mathrm{}`$-torsion, the same is true for $`X_ZY`$ as one shows inductively via the long exact sequence associated to the exact sequence of coefficients $`0/\mathrm{}/\mathrm{}^{n+1}/\mathrm{}^n0`$. Hence the classical relative Künneth formula for $`H^{}(X_ZY;\widehat{MU}^{})`$, see e.g. \[Sm\] using the fact that $`H^{}(Y;/\mathrm{})`$ is a free $`H^{}(Z;/\mathrm{})`$-algebra, impies the relative Künneth formula for $`\widehat{MU}`$.
For the case that $`X`$ has $`\mathrm{}`$-torsion, we use induction on the skeletal filtration of $`X`$. Let $`X^n`$ be the $`n`$-th skeleton of $`X`$. If $`\widehat{MU}^{}(Y)_{\widehat{MU}^{}(Z)}\widehat{MU}^{}(W)\widehat{MU}^{}(Y_ZW)`$ is an isomorphism for $`W=X^n`$ and for $`W=X^{n+1}/X^n`$, then it is also an isomorphism for $`W=X^{n+1}`$. This follows from the long exact sequence associated to the cofibre sequence $`X^nX^{n+1}X^n/X^{n+1}`$ and the exactness of the tensor product with $`\widehat{MU}^{}(Y)`$ over $`\widehat{MU}^{}(Z)`$. The exactness of $`\widehat{MU}^{}(Y)_{\widehat{MU}^{}(Z)}`$ is due to the fact that $`Y`$ and $`Z`$ have no $`\mathrm{}`$-torsion and $`\widehat{MU}^{}(Y)`$ is a free $`\widehat{MU}^{}(Z)`$-algebra.
Finally, if $`X`$ has arbitrary dimension, we argue as above and take the limit over the skeletons of $`X`$. Then we deduce the result by taking into account that the cohomology groups of $`X`$ are finite dimensional $`/\mathrm{}`$-vector spaces by hypotheses.
Exactly the same argument holds for the case of $`\widehat{MU}^{}(;/\mathrm{})`$ instead of $`\widehat{MU}^{}()`$. $`\mathrm{}`$
###### Corollary 7.40
For the projective $`n`$-space and every smooth scheme $`X`$ over $`k`$ the canonical maps for étale cobordism
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n)_{\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k)}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X)\stackrel{}{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n\times _kX)$$
and
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n;/\mathrm{})_{\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k;/\mathrm{})}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{})\stackrel{}{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n\times _kX;/\mathrm{})$$
are isomorphisms.
Proof This follows immediately, since $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n)`$ is a free $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k)`$-algebra by Proposition 7.37. $`\mathrm{}`$
Let $`𝒪(1)_k^n`$ be the canonical quotient line bundle. In fact, the isomorphism (LABEL:hMUmodfinite) is proved in \[De\] by the following argument. Since $`\widehat{\mathrm{Et}}_k^n`$ has no $`\mathrm{}`$-torsion the Atiyah-Hirzebruch spectral sequence implies that the orientation map $`\widehat{MU}^{}(\widehat{\mathrm{Et}}_k^n)H^{}(\widehat{\mathrm{Et}}k;/\mathrm{})`$ factors through the isomorphism
$$\widehat{MU}^{}(\widehat{\mathrm{Et}}_k^n;/\mathrm{})_sH^s(\widehat{\mathrm{Et}}_k^n;/\mathrm{})\widehat{MU}^s,$$
see the proof of Proposition 2.1.9 of \[De\]. This implies that the element $`\xi _H=c_1(𝒪(1))H^2(\widehat{\mathrm{Et}}_k^n;/\mathrm{})`$ induces an element $`\xi _{\widehat{MU}}\widehat{MU}^2(\widehat{\mathrm{Et}}_k^n)`$, and $`\xi _{\widehat{MU}}\widehat{MU}^2(\widehat{\mathrm{Et}}_k^n;/\mathrm{})`$ respectively, and that $`\xi _{\widehat{MU}}`$ is the image of $`u`$ under the isomorphism of Proposition 7.37 above.
Hence we can reformulate the above assertion as
(30)
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(_k^n)_{i=0}^n\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k)\xi ^i$$
and similarly with $`/\mathrm{}`$-coefficients.
The element $`\xi _{\widehat{MU}}\widehat{MU}^2(\widehat{\mathrm{Et}}_k^n)`$ is given by a morphism in $`\widehat{𝒮}`$
$$\xi _{\widehat{MU}}:\widehat{\mathrm{\Sigma }}^{\mathrm{}}(\widehat{\mathrm{Et}}_k^n)\widehat{MU}S^2.$$
Now let $`EX`$ be a vector bundle over $`X`$ in $`\mathrm{Sm}/k`$. Let $`(E)`$ be the associated projective bundle and let $`𝒪(1)`$ be the canonical quotient line bundle. This line bundle determines a morphism $`(E)_k^N`$ for some sufficiently large $`N`$. Together with the morphism $`\xi _{\widehat{MU}}`$ we get an element $`\xi \widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2((E))`$.
Now one can continue as above to show that cobordism with $`/\mathrm{}`$-coefficients over a finite field is an oriented cohomology theory.
## 8 Algebraic versus Profinite Etale Cobordism
Let $`\mathrm{}`$ be a fixed prime number different from the characteristic of the base field $`k`$. Using the results of the previous sections, we compare étale cobordism with the algebraic cobordism theories of Levine/Morel and Voevodsky over a separably closed field.
Furthermore, we construct a map from the $`MGL`$-theory to profinite étale cobordism via the results of the previous section on the stable étale realization functor. The construction is not as straightforward as one would like, since we do not know if $`\widehat{\mathrm{Et}}MGL`$ is isomorphic to $`\widehat{MU}`$ in $`\widehat{𝒮}`$. Nevertheless, the stable étale realization of Theorem 6.5 is the main technical advantage of the use of $`\widehat{\mathrm{Et}}`$ instead of $`\mathrm{Et}`$ and is a necessary ingredient for the construction of the map from algebraic to profinite étale cobordism. The other key point is an étale Thom class for vector bundles. This Thom class yields the map $`\widehat{\mathrm{Et}}MGL\widehat{MU}`$.
At the end, we conjecture that our map from algebraic cobordism to profinite étale cobordism is an isomorphism for $`/\mathrm{}^\nu `$-coefficients after inverting a Bott element. We explain a strategy to prove this conjecture and give arguments for the truth of the conjecture.
### 8.1 Comparison with $`\mathrm{\Omega }^{}`$
We consider the algebraic cobordism theories $`\mathrm{\Omega }^{}()`$ and $`\mathrm{\Omega }^{}(;/\mathrm{}^\nu )`$. As a corollary of Theorem 7.29 we get
###### Theorem 8.1
By universality of $`\mathrm{\Omega }^{}`$, the structure of an oriented cohomology theory on $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$ and $`\widehat{MU}/\mathrm{}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^\nu `$ yields canonical morphisms
$$\theta :\mathrm{\Omega }^{}(X)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(X)$$
and
$$\theta :\mathrm{\Omega }^{}(X;/\mathrm{}^\nu )\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(X;/\mathrm{}^\nu )$$
defined by sending a generator $`[f:YX]\mathrm{\Omega }^{}(X)`$, $`f:YX`$ a projective morphism between smooth schemes, to the element $`f_{}(1_Y)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(X)`$; similarly with $`/\mathrm{}^\nu `$-coefficients.
If $`k`$ is a field of characteristic zero, Theorem 4 of \[LM\] states that $`\mathrm{\Omega }^{}(k)𝕃^{}`$. But it is conjectured that this isomorphism holds for every field, see Conjecture 2 of \[LM\]. We refer to this conjecture as
###### Assumption 8.2
For the field $`k`$, the canonical morphism $`𝕃^{}\stackrel{}{}\mathrm{\Omega }^{}(k)`$ is an isomorphism.
If we suppose that this isomorphism holds for $`k`$, we may conclude
###### Proposition 8.3
For every separably closed field $`k`$, the morphisms
$$\theta :\mathrm{\Omega }^{}(k;/\mathrm{}^\nu )\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(k;/\mathrm{}^\nu )$$
and
$$\theta _{\mathrm{}}:\mathrm{\Omega }^{}(k)_{}_{\mathrm{}}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(k)$$
are surjective.
If we suppose in addition that Assumption 8.2 is true for $`k`$, then $`\theta `$ becomes an isomorphism on coefficient rings
$$\theta :\mathrm{\Omega }^{}(k;/\mathrm{}^\nu )\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(k;/\mathrm{}^\nu )$$
and
$$\theta :\mathrm{\Omega }^{}(k)_{}_{\mathrm{}}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(k).$$
Proof For the surjectivity, consider the canonical map $`\mathrm{\Phi }:𝕃^{}/\mathrm{}^\nu \mathrm{\Omega }^{}(k;/\mathrm{}^\nu )`$. Levine and Morel have shown that it is injective for every field, see \[LM\], Corollary 12.3. When we compose this map with $`\theta :\mathrm{\Omega }^{}(k;/\mathrm{}^\nu )\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(k;/\mathrm{}^\nu )`$, we get the canonical map
$$𝕃^{}/\mathrm{}^\nu \stackrel{\mathrm{\Phi }}{}\mathrm{\Omega }^{}(k;/\mathrm{}^\nu )\stackrel{\theta }{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(k;/\mathrm{}^\nu ).$$
Since this map is unique, it must be the canonical isomorphism. Hence $`\theta :\mathrm{\Omega }^{}(k;/\mathrm{}^\nu )\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(k;/\mathrm{}^\nu )`$ is surjective.
A similar argument works for $`\theta :\mathrm{\Omega }^{}(k)_{\mathrm{}}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(k)`$.
Note that for the surjectivity assertion we did not need to suppose that Assumption 8.2 is true.
The last part follows from the assumption and Proposition 7.15.
$`\mathrm{}`$
### 8.2 Comparison with $`MGL^,`$
Now we dedicate our attention to the comparison with the algebraic cobordism theory represented by the $`MGL`$-spectrum in the stable motivic $`𝔸^1`$-homotopy category. The technical advantage of $`\widehat{\mathrm{Et}}`$ compared to $`\mathrm{Et}`$ now becomes apparent since we could construct a stable realization of $`^1`$-spectra in the previous section. We will use this functor to define a map from $`MGL`$-theory to profinite étale cobordism. First we define Thom classes which will enable us to define canonical maps of oriented cohomology theories. We will do this in essentially the same way in which one shows that $`MGL`$ is the universal oriented $`^1`$-spectrum.
Let $`k`$ be a separably closed field. Let $`V`$ be a vector bundle of rank $`d`$ over $`X`$ in $`\mathrm{Sm}/k`$. We recall that the Thom space $`\mathrm{Th}(V)LU(k)`$ of $`V`$ is defined to be the quotient
$$\mathrm{Th}(V)=\mathrm{Th}(V/X)=V/(Vi(X))$$
where $`i:XV`$ denotes the zero section of $`V`$. We reformulate a lemma from $`𝔸^1`$-homotopy theory.
###### Proposition 8.4
Let $`V`$ be a vector bundle over $`X`$ and $`(V)(V𝒪)`$ be the closed embedding at infinity. Then the canonical morphism of pointed sheaves: $`(V𝒪)/(V)\mathrm{Th}(V)`$ induces a weak equivalence in $`\widehat{𝒮}_{}`$ via $`\widehat{\mathrm{Et}}`$.
Proof This is the same proof as for Proposition 3.2.17 of \[MV\] where we use the fact that $`\widehat{\mathrm{Et}}`$ preserves $`𝔸^1`$-weak equivalences between smooth schemes, see Theorem 5.14, and commutes with quotients, i.e $`\widehat{\mathrm{Et}}(X/A)=\widehat{\mathrm{Et}}X/\widehat{\mathrm{Et}}A`$. $`\mathrm{}`$
###### Corollary 8.5
Let $`T=𝔸^1/(𝔸^10)`$ in $`LU(k)`$. Then we have an isomorphism in $`\widehat{}_{}`$ of pointed profinite spaces $`\widehat{\mathrm{Et}}(T)S^2`$.
Proof This follows from the previous proposition using the fact that $`T=\mathrm{Th}(𝒪)`$ and $`\widehat{\mathrm{Et}}^1S^2`$ in $`\widehat{}`$ by Example 5.10. $`\mathrm{}`$
Now we define the Thom class of $`V`$ in $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{2d}(\mathrm{Th}(V))`$. From the isomorphism $`(V𝒪)/(V)\mathrm{Th}(V)`$ of the previous proposition we deduce an exact sequence induced by the cofiber sequence
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(\mathrm{Th}(V))\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}((V𝒪))\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}((V)).$$
Using the projective bundle formula of Theorem 7.26 this sequence is isomorphic to the exact sequence
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(\mathrm{Th}(V))\widehat{MU}^{}[1,u,\mathrm{},u^d]\widehat{MU}^{}[1,u,\mathrm{},u^{d1}].$$
The element $`u^dc_1(V)u^{d1}+\mathrm{}+(1)^dc_d(V)\widehat{MU}^{}[1,u,\mathrm{},u^d]`$ is sent to $`u^dc_1(V)u^{d1}+\mathrm{}+(1)^dc_d(V)\widehat{MU}^{}[1,u,\mathrm{},u^{d1}]`$ which is $`0`$ by the definition of Chern classes via the projective bundle formula. By exactness, there is an element $`\mathrm{th}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}(V)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{2d}(\mathrm{Th}(V))`$ that is sent to the above element $`u^d+\mathrm{}+(1)^dc_d(V)\widehat{MU}^{}[1,u,\mathrm{},u^d]`$ which is homogeneous of degree $`2d`$ since $`u`$ has degree $`2`$. We call $`\mathrm{th}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}(V)`$ an étale Thom class of $`V`$. Hence the Thom class of $`V`$ is a morphism
$$\mathrm{th}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}(V):\widehat{\mathrm{\Sigma }}^{\mathrm{}}(\widehat{\mathrm{Et}}\mathrm{Th}(V))\widehat{MU}S^{2d}$$
in $`\widehat{𝒮}`$.
In the same way, we define an étale Thom class for $`\widehat{MU}/\mathrm{}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^\nu `$.
We apply this argument to the tautological $`n`$-bundle $`\gamma _n`$ over the infinite Grassmannian. Since $`MGL_n=\mathrm{Th}(\gamma _n)`$ we get via $`\widehat{\mathrm{Et}}`$ a map of profinite spaces
$$\widehat{\mathrm{Et}}(MGL_n)\widehat{MU}_{2n}.$$
In the proof of Theorem 6.5 we have seen that $`\widehat{\mathrm{Et}}`$ yields a map of spectra on the level of homotopy categories. Hence the above map yields a morphism in $`\widehat{𝒮}_2`$
(31)
$$\varphi :\mathrm{L}\widehat{\mathrm{Et}}(MGL)\widehat{MU}$$
where $`\mathrm{L}\widehat{\mathrm{Et}}`$ denotes the total left derived functor of $`\widehat{\mathrm{Et}}`$ on $`^1`$-spectra.
Since smooth schemes $`X\mathrm{Sm}/k`$ are cofibrant objects, $`\mathrm{L}\widehat{\mathrm{Et}}`$ and $`\widehat{\mathrm{Et}}`$ agree on $`\mathrm{Sm}/k`$ by Theorem 5.14. Hence this morphism in $`\widehat{𝒮}_2`$ yields a natural morphism of cohomology theories
$$(\mathrm{L}\widehat{\mathrm{Et}}MGL)^{}(\widehat{\mathrm{Et}}X)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X)$$
for every scheme $`X`$ in $`\mathrm{Sm}/k`$. It is clear that this construction is natural in $`X`$.
Recall the definition of algebraic cobordism for a scheme $`X`$ in $`\mathrm{Sm}/k`$
$$MGL^{p,q}(X):=\mathrm{Hom}_{𝒮^^1(k)}(\mathrm{\Sigma }_^1^{\mathrm{}}(X),MGL[p2q](^1)^q),$$
where $`E[n]:=E(S_s^1)^n`$, $`S_s^1`$ denoting the simplicial circle. We have seen in Section 6 that $`S_s^1`$ commutes with $`\widehat{\mathrm{Et}}`$. Hence the elements in degree $`p,q`$ are sent to elements in degree $`p`$ via $`\varphi `$. We summarize this discussion in the following
###### Theorem 8.6
The obvious induced map of Theorem 6.5 yields a natural map for every $`X`$ in $`\mathrm{Sm}/k`$
(32)
$$\varphi :MGL^,(X)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X),$$
which is $`\varphi :MGL^{p,q}(X)\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^p(X)`$ on the $`p,q`$-level.
In the same way we get a map for $`MGL`$ with $`/\mathrm{}^\nu `$-coefficients
(33)
$$\varphi :MGL^,(X;/\mathrm{}^\nu )\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu ).$$
###### Remark 8.7
In the same way as above we get a morphism of profinite spectra $`\widehat{\mathrm{Et}}MGLH/\mathrm{}^\nu `$ which, for every $`X`$ in $`\mathrm{Sm}/k`$, gives rise to maps from algebraic cobordism to étale cohomology
$$MGL^,(X)H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )$$
and
$$MGL^,(X;/\mathrm{}^\nu )H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu ).$$
Recall that a morphism of oriented cohomology theories is a natural transformation that commutes with Chern classes and transfer maps. In fact, Theorem 4.1.4 of \[Pa\] states that we have to check only the compatibility with one of these structures. It is clear from the construction that we get the following
###### Theorem 8.8
The natural transformations $`\varphi :MGL^,()\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}()`$ and $`\varphi :MGL^,(;/\mathrm{}^\nu )\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(;/\mathrm{}^\nu )`$ are morphisms of oriented cohomology theories on $`\mathrm{Sm}/k`$.
Proof We have to check that $`\varphi `$ respects Chern classes in both theories. Then Theorem 4.1.4 of \[Pa\] implies that the transfer maps in both theories agree. But the construction of Chern classes in $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$ is the exact analog of the construction of Chern classes in $`MGL`$ only involving the functor $`\widehat{\mathrm{Et}}`$ at the appropriate places, i.e. the orientation $`x_{\widehat{MU}}`$ of $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$ is just $`\widehat{\mathrm{Et}}(x_{MGL})`$, i.e. the image of the orientation $`x_{MGL}`$ of $`MGL`$ under $`\widehat{\mathrm{Et}}`$. Hence we have $`\varphi (c_1^{MGL}(L))=c_1^{\widehat{MU}}(L)`$ in $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(X)`$. Then Theorem 4.1.4 of \[Pa\] implies that $`\varphi `$ respects the oriented structures and pushforwards, i.e. $`\varphi `$ is a morphism of oriented cohomology theories.
The same argument holds for $`/\mathrm{}^\nu `$-coefficients. $`\mathrm{}`$
###### Remark 8.9
We could have deduced from this theorem that $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$ carries in fact the formal group law induced by $`𝕃^{}𝕃^{}_{\mathrm{}}`$. Since Hopkins and Morel proved in \[HM\] that $`\mathrm{\Phi }:𝕃^{}MGL^{2,}(k)`$ is an isomorphism, the above theorem implies that the formal group law on $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$ is the one induced by $`MGL^{2,}`$.
As a corollary we get the following
###### Theorem 8.10
Let $`k`$ be a separably closed field. The canonical map of oriented cohomology theories $`\theta :\mathrm{\Omega }^{}()\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2()`$ factors through $`MGL^{2,}()`$, i.e. the following diagram
$$\begin{array}{ccc}\mathrm{\Omega }^{}(X)& \stackrel{\theta _{MGL}}{}& MGL^{2,}(X)\\ \theta _{\widehat{MU}}& & \varphi \\ & \widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(X)& \end{array}$$
is commutative for every $`X`$ in $`\mathrm{Sm}/k`$.
Furthermore, the diagram
$$\begin{array}{ccc}\mathrm{\Omega }^{}(X;/\mathrm{}^\nu )& \stackrel{\theta _{MGL}}{}& MGL^{2,}(X;/\mathrm{}^\nu )\\ \theta _{\widehat{MU}}& & \varphi \\ & \widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(X;/\mathrm{}^\nu )& \end{array}$$
is commutative for every $`X`$ in $`\mathrm{Sm}/k`$.
Proof By universality of $`\mathrm{\Omega }^{}`$, $`\theta _{\widehat{MU}}`$ is the unique morphism of oriented cohomology theories on $`\mathrm{Sm}/k`$ from $`\mathrm{\Omega }^{}()`$ to $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2()`$. Since the composition of $`\theta _{MGL}:\mathrm{\Omega }^{}()MGL^{2,}()`$ with $`\varphi `$ is also a morphism of oriented cohomology theories by Theorem 8.8, the triangle commutes.
The same argument holds for $`/\mathrm{}^\nu `$-coefficients. $`\mathrm{}`$
###### Remark 8.11
1. Hopkins and Morel have proved that $`\mathrm{\Omega }^{}(X)MGL^{2,}(X)`$ is surjective for every $`X`$ in $`\mathrm{Sm}/k`$ over every field of characteristic zero in \[HM\].
2. Let $`MGL_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$ be the $`^1`$-Thom-spectrum of algebraic cobordism in $`𝒮^^1(k)_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$, i.e. instead of using the Nisnevich topology, we consider $`MGL`$ and smooth schemes as sheaves for the étale topology.
The map $`\varphi `$ actually factors through $`MGL_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^,(X)`$, since $`\widehat{\mathrm{Et}}`$ factors through the étale version of $`𝔸^1`$-homotopy theory, see Remark 5.15.
3. Suppose that one has constructed an Atiyah-Hirzebruch spectral sequence from étale cohomology with coefficients in $`/\mathrm{}^\nu MU^{}`$ to $`MGL_{\stackrel{´}{\mathrm{e}}\mathrm{t}}(;/\mathrm{}^\nu )`$. It is clear from the construction of $`\varphi `$ and the isomorphism $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu MU^{})\stackrel{}{}H^{}(\widehat{\mathrm{Et}}X;/\mathrm{}^\nu MU^{})`$ that this spectral sequence would imply an isomorphism
$$\varphi _{\stackrel{´}{\mathrm{e}}\mathrm{t}}:MGL_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^,(X;/\mathrm{}^\nu )\stackrel{}{}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )$$
whenever both spectral sequences converge.
Such an Atiyah-Hirzebruch spectral sequence for motivic cohomology and Nisnevich $`MGL`$ has been constructed by Hopkins and Morel \[HM\]. We will discuss an application of it in the following section.
4. We have seen that this definition of profinite étale cobordism is not good enough to reflect questions of the $`q`$-twist of the $`MGL`$-theory, see also Remark 7.34. This point should be clarified in future work.
### 8.3 A conjecture
At the end of this section we would like to explain and give arguments for a conjecture on the relation of algebraic and profinite étale cobordism after inverting a Bott element.
First we recall a theorem of Levine for cohomology. Let $`H^p(X;/n(q))`$ denote the motivic cohomology of a smooth scheme $`X`$ over a field $`k`$. For $`\mathrm{Spec}k`$ there is an isomorphism $`H^0(\mathrm{Spec}k;/n(1))\mu _n(k)`$ with the group of $`n`$-th roots of unity in $`k`$. We suppose that $`k`$ contains an $`n`$-th root of unity $`\zeta `$, we have a corresponding motivic Bott element $`\beta _nH^0(\mathrm{Spec}k;/n(1))`$.
Levine has shown in \[Le2\] that motivic $`/n`$-cohomology of a smooth scheme over $`k`$ agrees with étale $`/n`$-cohomology after inverting the Bott element. More precisely, we form the bigraded ring
$$H^{}(X;/n())[\beta _n^1]:=(_{p,q}H^p(X;/n(q)))[\beta _n^1].$$
There is a cycle class map $`cl^,:H^{}(X;/n())H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;\mu _n^{})`$ with $`cl^{0,1}(\beta _n)=\zeta `$. Since $`k`$ contains $`\zeta `$, the cup product with $`\zeta `$ is an isomorphism on étale cohomology. Hence we get an induced map
(34)
$$cl^,:H^{}(X;/n())[\beta _n^1]H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;\mu _n^{}).$$
This map is an isomorphism for every smooth scheme $`X`$ over $`k`$ if $`n`$ is odd or if $`4|n`$ or if $`k`$ contains a square root of $`1`$, cf. Theorem 1.1 of \[Le2\].
Furthermore, Hopkins and Morel \[HM\] have constructed an Atiyah-Hirzebruch spectral sequence for algebraic cobordism
(35)
$$E_2^{,,}=H^;(X,MU^{})MGL^,(X),$$
where $`H^,(X;MU^{})`$ denotes motivic cohomology with coefficients in the ring $`MU^{}`$. We still suppose that $`k`$ contains an $`n`$-th root of unity. We consider the spectral sequence for $`\mathrm{Spec}k`$. Since $`E_2^{p,1,q}(\mathrm{Spec}k)`$ is concentrated in degrees $`p=0`$ and $`p=1`$ we deduce an isomorphism $`MGL^{0,1}(k)k^\times `$. For $`/n`$-coefficients, the exact sequence for coefficients implies that we get an isomorphism
$$MGL^{0,1}(k;/n)\mu _n(k)$$
and, via the spectral sequence, the motivic Bott element defined above, more precisely $`\beta _n1_{MU/n}`$, is sent to an induced Bott element $`\beta _nMGL^{0,1}(k;/n)`$.
Let us now suppose that $`k`$ is separably closed, $`n=\mathrm{}^\nu `$, and that the characteristic of $`k`$ is not equal to $`\mathrm{}`$. In particular, $`k`$ contains an $`\mathrm{}^\nu `$-th root of unity $`\zeta `$. It defines an element $`\zeta H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^0(\mathrm{Spec}k;\mu _\mathrm{}^\nu )`$. The element $`\zeta 1_{MU/\mathrm{}^\nu }`$ induces via the Atiyah-Hirzebruch spectral sequence an element $`\zeta _{\widehat{MU}}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^0(\mathrm{Spec}k;/\mathrm{}^\nu )`$ or equivalently a unit element in $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^0(\mathrm{Spec}k;/\mathrm{}^\nu )/\mathrm{}^\nu `$. The multiplication with $`\zeta `$ yields an isomorphism of spectral sequences and hence multiplication with $`\zeta _{\widehat{MU}}`$ is an isomorphism on étale cobordism. Furthermore, the map $`\varphi `$ of (32) maps the element $`\beta _\mathrm{}^\nu `$ to the element $`\zeta _{\widehat{MU}}`$. This implies that $`\varphi `$ also yields a map $`\varphi :MGL^,(X;/\mathrm{}^\nu )[\beta ^1]\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )`$.
###### Conjecture 8.12
Let $`X`$ be a smooth scheme of finite type over a separably closed field $`k`$ of characteristic different from $`\mathrm{}`$. Suppose that $`\mathrm{}`$ is odd or that $`\mathrm{}^\nu 4`$. Let $`\beta MGL^{0,1}(k;/\mathrm{}^\nu )`$ be the Bott element defined above. We conjecture that the morphism
$$\varphi :MGL^,(X;/\mathrm{}^\nu )[\beta ^1]\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )$$
is an isomorphism.
In fact, one should conjecture that $`\varphi `$ can be defined for every field and yields an isomorphism over every field that satisfies the hypothesis. For our construction of $`\varphi `$ in the previous section, we have used that $`k`$ was separably closed. But it seems to be very likely, that $`\varphi `$ does not depend on that fact. On the other hand, we have argued in the previous section, see Remarks 7.34 and 8.11, that $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}`$ may not be the best theory over a non-separably closed field. We intend to ameliorate the theory for this purpose.
Nevertheless, the proof of the conjecture over a separably closed field would already be a big step towards the proof of a more general conjecture. Thomason \[Th\] and Levine \[Le2\] proved that algebraic K-theory, respectively motivic cohomology, satisfies étale descent after inverting the Bott element. Then they showed that the statement holds over an algebraically closed field and reduced the general case via descent to this special case.
Hence for algebraic cobordism one could try the same, i.e. one could try to prove that $`MGL^,(;/\mathrm{}^\nu )[\beta ^1]`$ satisfies étale descent and then reduce to the special case of a separably closed field.
## 9 Calculations and Applications
Most of the following calculations of profinite étale cobordism are due to the Atiyah-Hirzebruch spectral sequence. We consider some examples of schemes whose étale cohomology groups are known and deduce from this the étale cobordism groups with $`/\mathrm{}^\nu `$-coefficients. But we start with the naturally arrising question how the absolute Galois group of a field acts on étale cobordism.
### 9.1 The Galois action on étale cobordism
The comparison with $`\mathrm{\Omega }^{}(;/\mathrm{}^\nu )`$ enables us to study the action of the Galois group of a field on $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}`$.
Let $`k`$ be a field of characteristic $`p\mathrm{}`$ and let $`\overline{k}`$ be a separable closure of $`k`$. If $`X`$ is a scheme over $`k`$, there is a natural action on $`X_{\overline{k}}=X_k\overline{k}`$ of the Galois group $`G_k:=\mathrm{Gal}(\overline{k}/k)`$ of $`k`$. The definition of $`\widehat{\mathrm{Et}}`$ immediately implies an action of $`G_k`$ on $`\widehat{\mathrm{Et}}X_{\overline{k}}`$.
###### Definition 9.1
Let $`X`$ be defined over a field $`k`$. We define the action of the Galois group $`G_k:=\mathrm{Gal}(\overline{k}/k)`$ on $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X_{\overline{k}})`$, resp. $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X_{\overline{k}};/\mathrm{}^\nu )`$, to be the one induced by the action of $`G_k`$ on $`\widehat{\mathrm{Et}}X_{\overline{k}}`$.
In order to study the Galois action on $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(\overline{k};/\mathrm{}^\nu )`$, we use Proposition 8.3. We deduce the following result from \[LM\], §3.2. Let $`H_{m,n}^n\times ^m`$ denote a smooth closed subscheme defined by the vanishing of a transverse section of the line bundle $`p_1^{}𝒪(1)p_2^{}𝒪(1)`$. The smooth projective $`\overline{k}`$-schemes $`H_{m,n}`$ are known as Milnor hypersurfaces.
From the definition of the map $`\mathrm{\Omega }^{}(\overline{k})\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^2(\overline{k};/\mathrm{}^\nu )`$, sending the class $`[Y\mathrm{Spec}\overline{k}]`$ to $`\pi _{}(1_Y)`$, we deduce the following
###### Proposition 9.2
Let $`\overline{k}`$ be a separably closed field. Then $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(\overline{k})`$, resp. $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(\overline{k};/\mathrm{}^\nu )`$, is generated by the classes of projective $`n`$-spaces and the classes of smooth Milnor hypersurfaces, i.e by the elements $`\pi _{}(1_{_{\overline{k}}^n})`$ for the projections $`\pi :_{\overline{k}}^n\mathrm{Spec}\overline{k}`$ and the elements $`\pi _{}(1_{H_{m,n}})`$ with projection $`\pi :H_{m,n}\mathrm{Spec}\overline{k}`$.
Proof This follows from Proposition 8.3 and the fact that $`\mathrm{\Omega }^{}(\overline{k})`$ is generated by the classes $`[_{\overline{k}}^n]`$ and $`[H_{m,n}]`$ for all $`m`$, $`n`$, see \[LM\], Remark 3.7. $`\mathrm{}`$
Hence, in order to determine the action of the absolute Galois group on the group $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}(\overline{k})`$, resp. $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}(\overline{k};/\mathrm{}^\nu )`$, it suffices to know the action of the Galois group on the projective space $`\pi :_{\overline{k}}^n\overline{k}`$ and the hypersurface $`H_{m,n}\overline{k}`$.
By \[Mi\] VI, 5.6, the étale cohomology of $`Y`$, for $`Y=_{\overline{k}}^n`$ or $`Y=H_{m,n}`$, is given by $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^i(Y;/\mathrm{})/\mathrm{}(i/2)`$ if $`i`$ is even, $`0i2n`$, resp. $`0i2(m+n1)`$, and vanishes otherwise, where the twist is given by the cyclotomic character. Consider $`\sigma G_k`$. Since $`k_s`$ is separably closed, this implies that the induced morphism $`\sigma :YY^\sigma `$ induces an isomorphism on étale cohomology with $`/\mathrm{}`$-coefficients, where $`Y^\sigma `$ is the twisted $`\overline{k}`$-scheme with structure morphism $`\sigma \pi `$. Hence $`\sigma `$ induces an isomorphism in $`\widehat{}`$:
$$\sigma :\widehat{\mathrm{Et}}Y_{\overline{k}}^n\stackrel{}{}\widehat{\mathrm{Et}}Y^\sigma $$
for $`Y=_{\overline{k}}^n`$ or $`Y=H_{m,n}`$. This implies the following
###### Theorem 9.3
The action of $`G_k`$ on $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(\overline{k};/\mathrm{}^\nu )`$ is trivial.
###### Remark 9.4
It is clear from the construction of the Atiyah-Hirzebruch spectral sequence in Theorem 4.11 that the corresponding spectral sequence for $`\widehat{MU}/\mathrm{}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^\nu `$ is equivariant under the Galois action. This can be seen as follows. By definition, the simplicial structure of $`\widehat{\mathrm{Et}}X`$ is given by the connected components $`\pi (U_i)`$ of the étale hypercoverings $`\mathrm{}U_1U_0X`$. Hence the action of $`G_k`$ on $`\widehat{\mathrm{Et}}X`$ is in fact an action on every simplicial dimension. In particular, this implies that the action of $`G_k`$ preserves the skeletal filtration on $`\widehat{\mathrm{Et}}X_{\overline{k}}`$ and induces an action on every exact couple. Since both actions on $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X;/\mathrm{}^\nu )`$ and on $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(X_{\overline{k}};/\mathrm{}^\nu )`$ are induced by the one on $`\widehat{\mathrm{Et}}X_{\overline{k}}`$, the assertion follows.
### 9.2 Etale Cobordism for local fields
Next, we calculate the profinite étale cobordism groups of a local field.
###### Theorem 9.5
Let $`k`$ be a local field, i.e. either a finite extension of the field $`_p`$ or a finite extension of the field of formal power series $`𝔽((t)))`$ over a finite field of characteristic $`p`$. We assume $`p\mathrm{}`$ and we denote by $`q=p^f`$ the number of elements in the residue class field of $`k`$. Let $`\nu _0\nu `$ be such that $`\mathrm{}^{\nu _0}=(q1,\mathrm{}^\nu )`$ is the greatest common divisor of $`q1`$ and $`\mathrm{}^\nu `$. For all $`n`$, the profinite étale cobordism groups with $`/\mathrm{}^\nu `$-coefficients of $`k`$ are given by
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^n(k;/\mathrm{}^\nu )=\{\begin{array}{cc}\hfill (/\mathrm{}^{\nu _0}MU^{n2})(/\mathrm{}^\nu MU^n):& n\mathrm{even}\hfill \\ \hfill (/\mathrm{}^\nu /\mathrm{}^{\nu _0})MU^{n1}:& n\mathrm{odd}.\hfill \end{array}$$
Proof We prove this theorem again via the Atiyah-Hirzebruch spectral sequence for étale cobordism using the identification of Galois and étale cohomology. Since $`/\mathrm{}^\nu MU^t`$ is a finitely generated free $`/\mathrm{}^\nu `$-module with a trivial $`G_k`$-operation, we may identify the Galois cohomology groups $`H^i(k;/\mathrm{}^\nu MU^t)`$ with $`H^i(k;/\mathrm{}^\nu )MU^t`$. Hence we only have to consider the cohomology groups $`H^i(k;/\mathrm{}^\nu )`$. These groups vanish for $`i>2`$, see e.g. \[NSW\], Chapter VII §1.
Since the $`G_k`$-operation on $`/\mathrm{}^\nu `$ is trivial, we have
$$H^0(k;/\mathrm{}^\nu )=/\mathrm{}^\nu .$$
By Theorem 7.2.9 of Chapter VII, §2 in \[NSW\], there is a duality for the finite $`G_k`$-modules $`A:=/\mathrm{}^\nu `$ and $`A^{}:=\mathrm{Hom}(A,\mu _\mathrm{}^\nu )`$ between the finite groups $`H^i(k;A)`$ and $`H^{2i}(k;A^{})`$ for $`0i2`$ given by the cup-product pairing
(36)
$$H^i(k;A)\times H^{2i}(k;A^{})H^2(k;\mu _\mathrm{}^\nu )=/\mathrm{}^\nu .$$
For $`i=1`$, by Theorem 7.1.8 in \[NSW\], Chapter VII §1, we know $`H^1(k;\mu _\mathrm{}^\nu )=k^\times /k^{\times \mathrm{}^\nu }`$. This yields by (36) an isomorphism $`H^1(k;/\mathrm{}^\nu )\mathrm{Hom}(H^1(k;\mu _\mathrm{}^\nu ),/\mathrm{}^\nu )=\mathrm{Hom}(k^\times /k^{\times \mathrm{}^\nu },/\mathrm{}^\nu )`$. By local class field theory, \[N\], Chapter II §5, Proposition 5.7, using the fact that $`p\mathrm{}`$, there is an isomorphism $`k^\times /k^{\times \mathrm{}^\nu }/\mathrm{}^\nu /\mathrm{}^{\nu _0}`$, where $`\mathrm{}^{\nu _0}`$ with $`\nu _0\nu `$ is the greatest common divisor $`(q1,\mathrm{}^\nu )`$. Hence we get an isomorphism
$$H^1(k;/\mathrm{}^\nu )/\mathrm{}^\nu /\mathrm{}^{\nu _0}.$$
For $`i=2`$, by Theorem 7.2.9 in \[NSW\], Chapter VII §1, we know $`H^2(k;\mu _\mathrm{}^\nu )=/\mathrm{}^\nu `$. Pairing (36) yields an isomorphism $`H^2(k;/\mathrm{}^\nu )\mathrm{Hom}(H^0(k;\mu _\mathrm{}^\nu ),/\mathrm{}^\nu )`$. But $`H^0(k;\mu _\mathrm{}^\nu )`$ is equal to the group $`\mu (k)`$of $`\mathrm{}^\nu `$th roots of unity in $`k`$, which is isomorphic to $`/\mathrm{}^{\nu _0}`$, with $`\nu _0=(q1,\mathrm{}^\nu )`$, by \[N\], Chapter II §5, Proposition 5.3. Hence we get an isomorphism
$$H^2(k;/\mathrm{}^\nu )/\mathrm{}^{\nu _0}.$$
Hence the associated Atiyah-Hirzebruch spectral sequence from the Galois cohomology of $`k`$ to $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^{}(k;/\mathrm{}^\nu )`$ is still easy to describe. It collapses at the $`E_2`$-stage since $`E_2^{s,t}0`$ if and only if $`s=0,1`$ or $`2`$ and $`t`$ even, and we know the occuring terms by the above arguments.
If $`n`$ is odd, $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^n(k;/\mathrm{}^\nu )`$ is isomorphic to $`E_2^{1,n1}=H^1(G_k;(\widehat{MU}/\mathrm{}^\nu )^{n1})(/\mathrm{}^\nu /\mathrm{}^{\nu _0})MU^{n1}`$.
If $`n`$ is even, there are two terms occurring: $`E_2^{0,n}=H^0(G_k;(\widehat{MU}/\mathrm{}^\nu )^n)/\mathrm{}^\nu MU^n`$ and $`E_2^{2,n2}=H^2(G_k;(\widehat{MU}/\mathrm{}^\nu )^{n2})/\mathrm{}^{\nu _0}MU^{n2}`$. Since there only two non-vanishing terms, these two groups are related by an exact sequence of $`/\mathrm{}^\nu `$-modules
$$0E_2^{2,n2}\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^n(k;/\mathrm{}^\nu )E_2^{0,n}0.$$
But since $`E_2^{0,n}/\mathrm{}^\nu MU^n`$ is a free $`/\mathrm{}^\nu `$-module, this sequence splits. Hence, if $`n`$ is even, we get an isomorphism
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^n(k;/\mathrm{}^\nu )/\mathrm{}^{\nu _0}MU^{n2}/\mathrm{}^\nu MU^n.$$
$`\mathrm{}`$
### 9.3 Etale Cobordism for smooth curves
The Atiyah-Hirzebruch spectral sequence and the knowledge on the étale cohomology allows us to determine the étale cobordism of a smooth curve.
###### Theorem 9.6
Let $`k`$ be a separably closed field of characteristic $`p\mathrm{}`$ and let $`X`$ be a connected projective smooth curve of genus $`g`$ over $`k`$. Then the profinite étale cobordism of $`X`$ with $`/\mathrm{}^\nu `$-coeffiecients is given by
$$\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^n(X;/\mathrm{}^\nu )\{\begin{array}{cc}\hfill (/\mathrm{}^\nu MU^n)(/\mathrm{}^\nu MU^{n2}):& n\mathrm{even}\hfill \\ \hfill _{i=1}^{i=2g}(/\mathrm{}^\nu MU^{n1}):& n\mathrm{odd}.\hfill \end{array}$$
Proof By \[SGA$`4\frac{1}{2}`$\], Arcata, III, Corollaire 3.5, we know that the cohomology groups $`H_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^s(X;/\mathrm{}^\nu )`$ vanish for $`s>2`$ and are free over $`/\mathrm{}^\nu `$ of rank $`1,2g,1`$ for $`s=0,1,2`$. The $`E_2^{s,t}`$-terms of the AHSS do not vanish only for $`s=0,1,2`$ and $`t0`$, $`t`$ even. Hence the spectral sequence collapses at the $`E_2`$-stage and since the occuring terms are all free $`/\mathrm{}^\nu `$-modules, the $`n`$th étale cobordism group $`\widehat{MU}_{\stackrel{´}{\mathrm{e}}\mathrm{t}}^n(X;/\mathrm{}^\nu )`$ is the direct sum of the $`E_2^{s,t}`$-terms with $`s+t=n`$. $`\mathrm{}`$
## Appendix A Existence of left Bousfield localization of fibrantly generated model categories
The aim of the first part of this appendix is the proof of the localization Theorem A.3 for the special situation of a left proper fibrantly generated model category $``$, e.g. $`\widehat{𝒮}`$. The dual version of this theorem is the left localization theorem of \[Hi\], but our result is more general. We only assume that the given model structure is fibrantly generated whereas in \[Hi\] it is used that the model structure is cellular, which is stronger than cofibrantly generated. The outline and most of the lemmas used to prove the theorem are just dual version of the ones in \[Hi\]. We will give the proof only of those statements that are not exact dual assertions. In particular, we point out at which places we use left properness and fibrantly generated and how we avoid the dual of cellularity.
We use this theorem in the next appendix to prove that there is a stable modle structure on the category of spectra for every fibrantly generated left proper model category. This enables us to stabilize the category of profinite spaces.
We assume the reader to be familiar with the notions of model categories, as explained e.g. in \[Hi\] or \[Ho1\].
###### Convention A.1
We make the convention that a model category has by definition all finite limits and finite colimits as in \[Qu1\]. If they need to have more than finite limits or finite colimits we will indicate this. In our applications the categories have all small limits and all results we need only use the existence of small limits.
In this subsection we prove an analogue of Theorem 4.1.1 of \[Hi\]. We recall the definition of local objects and local equivalences.
###### Definition A.2
Let $``$ be a simplicial model category. Let $`K`$ be a set of objects in $``$.
1. A map $`f:AB`$ is called a $`K`$-local equivalence if for every element $`X`$ of $`K`$ the induced map of simplicial mapping spaces $`f^{}:\mathrm{Map}(B,X)\mathrm{Map}(A,X)`$ is a weak equivalence of simplicial sets.
2. Let $`𝒞`$ denote the class of $`K`$-local equivalences. An object $`X`$ is called $`K`$-local if it is $`𝒞`$-local, i.e. if $`X`$ is fibrant and for every $`K`$-local equivalence $`f:AB`$ the induced map $`f^{}:\mathrm{Map}(B,X)\mathrm{Map}(A,X)`$ is a weak equivalence.
###### Theorem A.3
Let $``$ be a left proper fibrantly generated simplicial model category with all small limits. Let $`K`$ be a set of fibrant objects in $``$ and let $`𝒞`$ be the class of $`K`$-local equivalences.
1. The left Bousfield localization of $``$ with respect to $`𝒞`$ exists, i.e. there is a model category structure $`L_𝒞`$ on the underlying category $``$ in which
1. the class of weak equivalences of $`L_𝒞`$ equals the class of $`K`$-local equivalences of $``$,
2. the class of cofibrations of $`L_𝒞`$ equals the class of cofibrations of $``$,
3. the class of fibrations of $`L_𝒞`$ is the class of maps with the right lifting property with respect to those maps that are both cofibrations and $`K`$-local equivalences.
2. The fibrant objects of $`L_𝒞`$ are the $`K`$-local objects of $``$.
3. $`L_𝒞`$ is left proper. It is fibrantly generated if every object of $``$ is cofibrant.
4. The simplicial structure of $``$ gives $`L_𝒞`$ the structure of a simplicial model category.
###### Remark A.4
1. Although fibrantly generated is the dual notion of cofibrantly generated the proof of theorem A.3 is not the dual of Theorem 4.1.1 of \[Hi\] for left Bousfield localization. But we can use a dual proof of Theorem 5.1.1 of \[Hi\] for right Bousfield localization where the cofibrations and fibrations change their role plus an idea of \[CI\] in order to avoid the dual of Proposition 5.2.3 the proof of which uses the cellularization. The fact that we can construct a left localization is due to the left properness.
2. The theorem does not depend on the simplicial structure of $``$. In fact, it is also true for a non-simplicial $``$ but we would have to use the machinery of homotopy function complexes instead of just simplicial mapping spaces.
3. Note that no result that we need below makes use of the fact that more than finite colimits exist in $``$. Only in order to verify the factorization model axiom we need the existence of small limits since we use a variant of the cosmall object argument.
We split the proof into some lemmas that we collect from \[Hi\] and \[CI\]. Let $`P`$ be the set of generating fibrations in $``$ and let $`Q`$ denote the set of generating trivial fibrations. Let $`K`$ be a fixed set of fibrant objects in $``$. We define the two sets of maps
$$\mathrm{\Lambda }(K):=\{f:X^{\mathrm{\Delta }^n}X^{\mathrm{\Delta }^n},XK,n0\}$$
and
$$\overline{\mathrm{\Lambda }(K)}:=\mathrm{\Lambda }(K)Q.$$
###### Lemma A.5
Let $``$ be fibrantly generated model category and let $`K`$ be a set of objects of $``$. If $`B`$ is a cofibrant object of $``$ then a map $`g:AB`$ is $`\overline{\mathrm{\Lambda }(K)}`$-projective if and only if it is both a cofibration and a $`K`$-local equivalence.
Proof The proof is dual to the one of Proposition 5.2.4 of \[Hi\]. $`\mathrm{}`$
###### Lemma A.6
Let $``$ be fibrantly generated model category and let $`K`$ be a set of objects of $``$. Then a map that is a transfinite tower of pullbacks of maps in $`\overline{\mathrm{\Lambda }(K)}`$ is a $`K`$-local fibration.
Proof The proof is dual to the one of Proposition 5.2.5 of \[Hi\]. Note that it does not make use of the cellularity of $``$. $`\mathrm{}`$
###### Lemma A.7
Let $``$ be a model category and $`K`$ a set of objects. A map $`g:XY`$ is both a $`K`$-local fibration and a $`K`$-local weak equivalence if and only if it is a trivial fibration.
Proof This is a dual of Lemma 5.3.2 of \[Hi\] whose proof does not depend on the cellularity or right properness of $``$. Hence the dual proof of Lemma 5.3.2 gives a proof for $``$ left proper and fibrantly generated. $`\mathrm{}`$
###### Lemma A.8
Let $``$ be a model category and $`K`$ a set of objects. Then there is a functorial factorization for every map $`g:XY`$ in $``$ as $`X\stackrel{q}{}W\stackrel{p}{}Y`$ in which $`q`$ is a $`K`$-local cofibration and $`p`$ is both a $`K`$-local weak equivalence and a $`K`$-local fibration.
Proof This follows from Lemma A.7 and the functorial factorization into a cofibration followed by a trivial fibration. $`\mathrm{}`$
###### Lemma A.9
Let $``$ be a left proper fibrantly generated model category and $`K`$ a set of objects. If $`g:XY`$ is a weak equivalence, $`h:YZ`$ is a fibration and the composition $`hg:XZ`$ is a $`K`$-local fibration, then $`h`$ is a $`K`$-local fibration.
Proof This is the dual proof to Lemma 5.3.4 in \[Hi\], but we point out where we use left properness and how the diagram looks like.
If $`i:AB`$ is a map in $`KWK\mathrm{Cof}=W\mathrm{Cof}`$ (last equality holds by Lemma A.7), then one can choose by Proposition 8.1.23 of \[Hi\] a cofibrant approximation $`\stackrel{~}{i}:\stackrel{~}{A}\stackrel{~}{B}`$ to $`i`$ such that $`\stackrel{~}{i}`$ Cof. By Propositions 3.1.5 and 3.2.3 of \[Hi\] $`\stackrel{~}{i}`$ is a $`K`$-local equivalence. Since $``$ is left proper, Proposition 13.2.1 of \[Hi\] implies that it is sufficient to show that $`h`$ has the right lifting property with respect to $`\stackrel{~}{i}`$.
Suppose we have a diagram
$$\begin{array}{ccc}& & X\hfill \\ & & g\hfill \\ \hfill \stackrel{~}{A}& \stackrel{s}{}& Y\hfill \\ \hfill \stackrel{~}{i}& & h\hfill \\ \hfill \stackrel{~}{B}& \stackrel{t}{}& Z.\hfill \end{array}$$
Now we continue as in the proof of Lemma 5.3.4 in \[Hi\] by constructing lifts $`j:\stackrel{~}{A}X`$ and $`k:\stackrel{~}{B}X`$, consider the composition $`u:=gk:\stackrel{~}{B}Y`$ which has the property $`uis`$ in the category of objects over $`Z`$ and conclude by the homotopy extension property of cofibrations in this case (Proposition 7.3.10 of \[Hi\]) that there is a map $`v:\stackrel{~}{B}Y`$ such that $`vu`$, $`v\stackrel{~}{i}=s`$ and hence $`hv=t`$. $`\mathrm{}`$
In \[Hi\] the cellularity of the given model category is used for the existence of factorizations. We have to avoid this point. In order to do so, we need a dual version of Lemma 2.5 of \[CI\].
###### Lemma A.10
Let $``$ be a left proper fibrantly generated simplicial model category and let $`K`$ be a set of fibrant objects. Then every map $`g:AB`$ has a factorization into a cofibration $`s:AW`$ that has the left lifting property with respect to all maps $`X^{\mathrm{\Delta }^n}X^{\mathrm{\Delta }^n}`$, $`XK`$, $`n0`$ followed by a $`K`$-fibration.
Proof This is a variation of the small object argument, dual to the proof of Lemma 2.5 of \[CI\]. We prove this lemma by transfinite induction of length $`\kappa `$ where $`\kappa `$ is a regular cardinal such that the codomains of the generating trivial fibrations $`Q`$ and the objects of $`K`$ are $`\kappa `$-cosmall relative to the fibrations.
Let $`J_o`$ be the set of all squares
$$\begin{array}{ccc}\hfill A& & X^{\mathrm{\Delta }^n}\\ \hfill g& & \\ \hfill B& & X^{\mathrm{\Delta }^n}\end{array}$$
for all objects $`XK`$. Define $`Z_0`$ to be the pullback $`_{J_0}X^{\mathrm{\Delta }^n}\times _{_{J_0}X^{\mathrm{\Delta }^n}}B`$ and let $`j_0:AZ_0`$, $`q_0:Z_0B`$ be the obvious maps. Now we factor $`j_0`$ into a cofibration $`i_0:AW_0`$ followed by a trivial fibration $`p_0:W_0Z_0`$. This is the first step of the induction.
If $`\beta `$ is a limit ordinal, set $`W_\beta :=lim_{\alpha <\beta }W_\alpha `$ and set $`i_\beta :=\mathrm{colim}_{\alpha <\beta }i_\alpha `$.
If $`\beta `$ is a successor ordinal, define $`J_\beta `$ to be the set of all squares
$$\begin{array}{ccc}\hfill A& & X^{\mathrm{\Delta }^n}\\ \hfill g& & \\ \hfill W_{\beta 1}& & X^{\mathrm{\Delta }^n}\end{array}$$
for all $`XK`$. Define $`Z_\beta `$ to be the pullback $`_{J_\beta }X^{\mathrm{\Delta }^n}\times _{_{J_\beta }X^{\mathrm{\Delta }^n}}W_{\beta 1}`$ and let $`j_\beta :AZ_\beta `$, $`q_\beta :Z_\beta Z_\beta `$ be the obvious maps. Now we factor $`j_\beta `$ into a cofibration $`i_\beta :AW_\beta `$ followed by a trivial fibration $`p_\beta :W_\beta Z_\beta `$.
By transfinite induction we get a $`\kappa `$-tower
$$\mathrm{}W_\beta \mathrm{}W_1W_0B.$$
Note that there are compatible maps $`i_\beta :AW_\beta `$. Since filtered limits exist in $``$ we may continue with the following definition. Let $`W`$ be $`lim_\beta W_\beta `$ and let $`i:AW`$ be $`lim_\beta i_\beta `$. By construction, each $`i_\beta `$ is a cofibration for every successor ordinal $`\beta `$. Since $``$ is fibrantly generated and $`\kappa `$ is such that the codomains of $`Q`$ and the objects of $`K`$ are $`\kappa `$-cosmall relative to the fibrations we get that $`i`$ is a cofibration. Furthermore one can show that $`i`$ has the left lifting property with respect to all maps $`X^{\mathrm{\Delta }^n}X^{\mathrm{\Delta }^n}`$, $`XK`$, $`n0`$ by the cosmall object argument as follows:
Suppose we have a diagram
$$\begin{array}{ccc}\hfill A& \stackrel{k}{}& X^{\mathrm{\Delta }^n}\hfill \\ \hfill i& & f\hfill \\ \hfill W& \stackrel{h}{}& X^{\mathrm{\Delta }^n}\hfill \end{array}$$
for some $`XK`$ and $`n0`$. By the theory of simplicial model categories if $`X`$ is a fibrant object then the map $`f:X^{\mathrm{\Delta }^n}X^{\mathrm{\Delta }^n}`$ is a fibration. Since the objects in $`K`$ are $`\kappa `$-cosmall relative to the fibrations, there is a $`\beta <\kappa `$ such that $`h`$ is the composite $`WW_\beta \stackrel{h_\beta }{}X^{\mathrm{\Delta }^n}`$. By construction, there is a map $`k_\beta :W_{\beta +1}X^{\mathrm{\Delta }^n}`$ such that $`fk_\beta =h_\beta j`$ and $`k=k_\beta i_{\beta +1}`$ where $`j`$ is the map $`W_{\beta +1}W_\beta `$. The composition $`WW_{\beta +1}\stackrel{k_\beta }{}X^{\mathrm{\Delta }^n}`$ is the required lift in the diagram.
It remains to show that $`p:WY`$ is a $`K`$-fibration. Since $`K`$-fib is defined by a right lifting property and $`p`$ is a transfinite tower of maps $`W_{\beta +1}W_\beta `$ it suffices to show that each $`W_{\beta +1}W_\beta `$ is a $`K`$-fibration. The map $`W_{\beta +1}W_\beta `$ is the composition $`W_{\beta +1}Z_{\beta +1}W_\beta `$ where the first map is a trivial fibration and hence also a map in $`KWK\mathrm{fib}`$. The second map is a product of maps of the form $`X^{\mathrm{\Delta }^n}X^{\mathrm{\Delta }^n}`$ each of which is a $`K`$-fibration by Lemma A.5. Hence $`Z_{\beta +1}W_\beta `$ is a $`K`$-fibration. $`\mathrm{}`$
###### Lemma A.11
Let $``$ be a left proper fibrantly generated model category and let $`K`$ be a set of objects. Then there is functorial factorization for every map $`g:XY`$ in $``$ as $`X\stackrel{q}{}W\stackrel{p}{}Y`$ in which $`q`$ is both a $`K`$-local weak equivalence and a $`K`$-local cofibration and $`p`$ is a $`K`$-local fibration.
Proof Again we dualize the proof of Proposition 5.3.5 of \[Hi\] point out how the diagrams must look like and where we use the left properness. Let $`j:\stackrel{~}{X}X`$ be a fibrant cofibrant approximation to $`X`$. Lemma A.10 implies that there is a functorial factorization of the composition $`gj`$ as $`\stackrel{~}{X}\stackrel{s}{}\stackrel{~}{W}\stackrel{r}{}X`$ such that $`s`$ is a cofibration and in $`\overline{\mathrm{\Lambda }(K)}`$-projective and $`r`$ is a $`K`$-local fibration. Let $`Z:=X_{\stackrel{~}{X}}\stackrel{~}{W}`$. We factor the induced map $`ZY`$ into $`Z\stackrel{u}{}W\stackrel{p}{}Y`$ where $`u`$ is a trivial cofibration and $`p`$ is a fibration. This yields the following diagram
Let $`q:=uv`$. Now since $`j`$ is a weak equivalence and $`s`$ is a cofibration the left properness of $``$ implies that $`t`$ is also a weak equivalence. Hence $`ut`$ is a weak equivalence and $`s`$ is a cofibrant approximation to $`q`$. Lemma A.5 implies that $`s`$ is a $`K`$-local equivalence.
By Lemma A.5, since $`s`$ is in $`\overline{\mathrm{\Lambda }(K)}`$-proj it is in particular a $`K`$-local equivalence. Hence $`qKW`$ since $`jKW`$ and $`qj=tsKW`$ (two out of three holds for $`KW`$). Now since $`q=uv`$ is the composition of two cofibrations and the $`K`$-local cofibrations equal the cofibrations in $``$, we have $`qKWKCof`$. By Lemma A.6 this q is a weak equivalence and a cofibration.
Since $`u`$, $`t`$ and hence also $`ut`$ are weak equivalences and $`r=put`$ is by construction a $`K`$-local fibration, we conclude by Lemma A.9 that $`p`$ is a $`K`$-local fibration. $`\mathrm{}`$
Proof of Theorem A.3 1. Axiom M1 is satisfied since it is satisfied in $``$. Axiom M2 follows from Proposition 3.2.3 of \[Hi\]. Axiom M3 follows from Proposition 3.2.4 and Lemma 7.2.8 of \[Hi\]. The right lifting property of $`K`$-local fibrations is given by definition. The left lifting property of $`K`$-local cofibrations follows from Lemma A.7. The factorization of a map into a $`K`$-local cofibration followed by a trivial $`K`$-local fibration follows from Lemma A.8 and the factorization into a trivial $`K`$-local cofibration followed by a $`K`$-local fibration follows from Lemma A.11.
2. This follows from Proposition 3.4.1 of \[Hi\].
3. The left properness follows from Proposition 3.4.4 of \[Hi\]. If every object in $``$ is cofibrant, then Lemma A.10 and lemma A.6 imply that $`L_𝒞`$ is fibrantly generated with generating fibrations $`\overline{\mathrm{\Lambda }(K)}`$ and generating trivial fibrations $`Q`$.
4. This is the analogue proof as for part 4 of Theorem 5.1.1 in \[Hi\]. $`\mathrm{}`$
## Appendix B Stable model structure on spectra
The well known stabilization of the category of simplicial spectra of \[BF\] uses the fact that $`𝒮`$ is proper in an essential way in order to construct functorial factorizations. Since the category $`\widehat{𝒮}`$ is only left but not right proper, we may not use their localization method.
Hovey \[Ho2\] has pointed out a general way to stabilize a left proper cellular model category with respect to a left Quillen endofunctor $`T`$. Since we do have left properness but not the property of being cofibrantly generated we have to modify this construction.
The aim is to construct a stable structure on $`𝒞`$ with respect to an endofunctor via the well known construction of spectra on $`𝒞`$. We reformulate the results of \[Ho2\] for the situation of a left proper fibrantly generated model category $`𝒞`$, e.g. $`\widehat{𝒮}`$. The projective and stable model structure is the same as in \[Ho2\], but we have to give different proofs for the key points of the construction. We also have to choose different generating fibrations. But the idea for the construction and the proof that the structure in fact becomes stable are due to \[Ho2\].
In the first part, we construct a stable structure on the usual classical notion of spectra in analogy to the one of \[BF\].
In the second part, we show, following \[Ho2\], that this works also for symmetric spectra on a symmetric monoidal model structure that is left proper and fibrantly generated.
### B.1 Spectra
In this section we assume always that $`𝒞`$ is a left proper fibrantly generated simplicial model category with all small limits. Let $`T:𝒞𝒞`$ be a left Quillen endofunctor on $`𝒞`$. Let $`U:𝒞𝒞`$ be its right adjoint.
###### Definition B.1
A spectrum $`X`$ is a sequence $`(X_n)_{n0}`$ of objects of $`𝒞`$ together with structure maps $`\sigma :TX_nX_{n+1}`$ for all $`n0`$. A map of spectra $`f:XY`$ is a collection of maps $`f_n:X_nY_n`$ commuting with the structure maps. We denote the category of spectra by $`\mathrm{Sp}(𝒞,T)`$.
We begin by defining an intermediate strict structure, see \[BF\] or \[Ho2\]. But note that we are in a dual situation of a fibrantly generated model structure. Consider the following functors:
###### Definition B.2
Given $`n0`$, the evaluation functor $`\mathrm{Ev}_n:\mathrm{Sp}(𝒞,T)𝒞`$ takes $`X`$ to $`X_n`$. It has a left adjoint $`F_n:𝒞\mathrm{Sp}(𝒞,T)`$ defined by $`(F_nA)_m=T^{mn}A`$ if $`mn`$ and $`(F_nA)_m=`$ otherwise. The structure maps are the obvious ones.
The evaluation functor also has a right adjoint $`R_n:𝒞\mathrm{Sp}(𝒞,T)`$ defined by $`(R_nA)_i=U^{ni}A`$ if $`in`$ and $`(R_nA)_i=`$ otherwise. The structure map $`TU^{ni}AU^{ni}A`$ is adjoint to the identity map of $`U^{ni}`$ when $`i<n`$.
Note that $`Ev_n`$ commutes with limits and colimits, where limits and colimits are constructed levelwise in $`\mathrm{Sp}(𝒞,T)`$. If $`G:I\mathrm{Sp}(𝒞,T)`$ is a functor from a small category to $`\mathrm{Sp}(𝒞,T)`$, then, for limits, the structure maps are defined as the composites
$$T(lim\mathrm{Ev}_nG)lim(T\mathrm{Ev}_nG)\stackrel{lim(\sigma G)}{}lim\mathrm{Ev}_{n+1}G.$$
For colimits one uses the fact that $`T`$ is left adjoint and hence preserves colimits to get the obvious structure maps.
###### Definition B.3
A map $`f`$ in $`\mathrm{Sp}(𝒞,T)`$ is a projective weak equivalence (resp. projective fibration) if each map $`f_n`$ is a weak equivalence (resp. fibration). A map $`i`$ is a projective cofibration if it has the left lifting property with respect to all projective trivial fibrations.
We will show that the projective structure is in fact a fibrantly generated model structure. We denote the set of generating fibrations of $`𝒞`$ by $`P`$ and the set of generating trivial fibrations by $`Q`$. Inspired by Proposition B.5, we set
$$\stackrel{~}{P}:=\{g_n:R_nRR_nS\times _{R_{n1}US}R_{n1}UR,\mathrm{for}\mathrm{all}f:RS\mathrm{in}P\}$$
and
$$\stackrel{~}{Q}:=\{g_n:R_nRR_nS\times _{R_{n1}US}R_{n1}UR,\mathrm{for}\mathrm{all}f:RS\mathrm{in}Q\}$$
where $`R_n`$ is the right adjoint of the evaluation functor $`Ev_n`$ and $`g_n`$ is the map induced by the commutative diagram
(37)
$$\begin{array}{ccc}R_nR& & R_nS\\ & & \\ R_{n1}UR& & R_{n1}US.\end{array}$$
We will show that the maps in $`\stackrel{~}{P}`$ are the generating fibrations and the maps in $`\stackrel{~}{Q}`$ are the generating trivial fibrations for the projective model structure on $`\mathrm{Sp}(𝒞,T)`$.
###### Lemma B.4
If $`f:XY`$ is a fibration (resp. trivial fibration) in $`𝒞`$, then the maps $`g_n:R_nXR_nY\times _{R_{n1}UY}R_{n1}UX`$ are fibrations (resp. trivial fibrations) in $`\mathrm{Sp}(𝒞,T)`$.
Proof Since $`R_n`$ is defined via the right Quillen functor $`U`$ which preserves fibrations and trivial fibrations, it is clear that the map $`(R_nf)_i=U^{ni}(f)`$ on the $`i`$-th level is a fibration (resp. trivial fibration). In diagram (37), the maps on the $`i`$-th level are either identities or equal to $`U^{ni}(f)`$. Hence the induced map $`g_n`$ is also a fibration (resp. trivial fibration). $`\mathrm{}`$
The definition of $`\stackrel{~}{P}`$ and $`\stackrel{~}{Q}`$ becomes clear after the following
###### Proposition B.5
A map $`i:AB`$ in $`\mathrm{Sp}(𝒞,T)`$ is a projective (trivial) cofibration if and only if $`i_0:A_0B_0`$ and the induced maps $`j_n:A_n_{TA_{n1}}TB_{n1}`$ for $`n1`$ are (trivial) cofibrations in $`𝒞`$.
Proof We prove the cofibration case, the case of trivial cofibrations is similar. We use the standard idea of \[Ho2\], Proposition 1.14. Let $`i:AB`$ be a map with the left lifting property with respect to projective trivial fibrations. Since projective fibrations are defined levelwise, it is clear that $`i_0:A_0B_0`$ is a cofibration. Let $`f:XY`$ be a trivial fibration in $`𝒞`$. We have to show that there is a lift $`B_nX`$ for any commutative diagram
$$\begin{array}{ccc}A_n_{TA_{n1}}TB_{n1}& & X\\ j_n& & f\\ B_n& & Y.\end{array}$$
By adjunction, this diagram has a lift if and only if the induced diagram
$$\begin{array}{ccc}A& & R_nX\\ i& & g_n\\ B& & R_nY\times _{R_{n1}UY}R_{n1}UX\end{array}$$
has a lift. Now $`R_n`$ is defined via the right Quillen functor $`U`$, see Definition B.2. This implies that all maps in Diagram (37) are projective trivial fibrations and hence the induced map is one, too. So $`g_n`$ is a trivial fibration if $`f`$ is one, see Lemma B.4. Hence there is a lift in the last diagram.
Now suppose that $`i_0`$ and $`j_n`$ are cofibrations in $`𝒞`$ for all $`n>0`$. Let
$$\begin{array}{ccc}A& \stackrel{a}{}& X\\ i& & f\\ B& \stackrel{b}{}& Y\end{array}$$
be a commutative diagram in $`\mathrm{Sp}(𝒞,T)`$ and let $`f`$ be a trivial fibration. We construct a lift $`h_n:B_nX_n`$, compatible with the structure maps, by induction on $`n`$. There is a lift $`h_0`$ since $`i_0`$ is a cofibration and $`f_0`$ is a trivial fibration in $`𝒞`$. Suppose $`h_j`$ is constructed for all $`j<n`$. Then we can consider the diagram
$$\begin{array}{ccc}A_n_{TA_{n1}}TB_{n1}& \stackrel{(f_n,\sigma Th_{n1})}{}& X\\ j_n& & f_n\\ B_n& \stackrel{b_n}{}& Y_n.\end{array}$$
The lift $`h_n:B_nX_n`$ of this diagram is the desired map. $`\mathrm{}`$
###### Theorem B.6
Let $`𝒞`$ be a left proper fibrantly generated simplicial model category with all small limits. The projective weak equivalences, projective fibrations and projective cofibrations define a left proper fibrantly generated simplicial model structure on $`\mathrm{Sp}(𝒞,T)`$ with set of generating fibrations $`\stackrel{~}{P}`$ and set of generating trivial fibrations $`\stackrel{~}{Q}`$.
We will prove this theorem in a series of lemmas. The existence of finite colimits and limits is clear since they are defined levelwise. The two-out-of-three- and the retract-axiom are immediate as well. The crucial point is the factorization axiom for which we use the cosmall object argument. For the notations $`\stackrel{~}{P}\mathrm{proj}`$, $`\stackrel{~}{P}\mathrm{fib}`$, $`\stackrel{~}{Q}\mathrm{proj}`$, $`\stackrel{~}{Q}\mathrm{fib}`$ and the cosmall object argument, which is the exact dual of the small object argument, we refer the reader to the book of Hovey \[Ho1\]: in $`\stackrel{~}{P}\mathrm{proj}`$ are all the maps having the left lifting property with respect to all maps in $`\stackrel{~}{P}`$; in $`\stackrel{~}{P}\mathrm{fib}`$ are all the maps having the right lifting property with respect to maps in $`\stackrel{~}{P}\mathrm{proj}`$.
###### Proposition B.7
The projective cofibrations are exactly the maps which have the left lifting property with respect to all maps in $`\stackrel{~}{Q}`$. The projective trivial cofibrations are exactly the maps which have the left lifting property with respect to all maps in $`\stackrel{~}{P}`$.
Proof By Proposition B.5, $`i:AB`$ is a projective cofibration if and only if $`i_0`$ and $`j_n`$ are cofibrations in $`𝒞`$ for all $`n>0`$. But since $`𝒞`$ is a fibrantly generated model category with $`Q`$ the set of generating trivial fibrations, $`i_0`$ and $`j_n`$ are cofibrations if and only if they have the left lifting property with respect to all maps in $`Q`$, i.e. if and only if $`i_0`$ and $`j_n`$ are in $`Q\mathrm{proj}`$. We have seen that a lift in a diagram
$$\begin{array}{ccc}A_n_{TA_{n1}}TB_{n1}& & R\\ j_n& & f\\ B_n& & S\end{array}$$
is by adjunction equivalent to a lift in the diagram
$$\begin{array}{ccc}A& & R_nR\\ i& & g_n\\ B& & R_nS\times _{R_{n1}US}R_{n1}UR.\end{array}$$
Hence $`i_0`$ and $`j_n`$ are in $`Q\mathrm{proj}`$ if and only if the map $`i`$ is an element of $`\stackrel{~}{Q}\mathrm{proj}`$. This completes the proof in the cofibration case. The trivial cofibration case is similar. $`\mathrm{}`$
###### Corollary B.8
The projective fibrations are exactly the maps in $`\stackrel{~}{P}\mathrm{fib}`$. The projective trivial cofibrations are exactly the maps in $`\stackrel{~}{Q}\mathrm{fib}`$.
###### Lemma B.9
If $`Z`$ is a cosmall object in $`𝒞`$ relative to the fibrations, then $`R_nZ`$, $`n0`$, is cosmall relative to the level fibrations in $`\mathrm{Sp}(𝒞,T)`$. Similarly, if $`Z`$ is a cosmall object in $`𝒞`$ relative to the trivial fibrations, then $`R_nZ`$, $`n0`$, is cosmall relative to the level trivial fibrations in $`\mathrm{Sp}(𝒞,T)`$. In addition, the codomains of the maps in $`\stackrel{~}{P}`$ (resp. $`\stackrel{~}{Q}`$) are cosmall relative to the level fibrations (resp. trivial fibrations) in $`\mathrm{Sp}(𝒞,T)`$.
Proof This is clear since $`Ev_n`$ commutes with all limits. $`\mathrm{}`$
We can now finish the proof of Theorem B.6. It remains to show the factorization axiom. Note that by hypothesis, $`𝒞`$ and hence $`\mathrm{Sp}(𝒞,T)`$ have all small limits. This enables us to apply the cosmall object argument. By Lemma B.9 and since we have shown in Corollary B.8 that the class of projective fibrations (resp. trivial fibrations) is equal to $`\stackrel{~}{P}\mathrm{fib}`$ (resp. $`\stackrel{~}{Q}\mathrm{fib}`$), we can apply the cosmall object argument to the class $`\stackrel{~}{P}`$ (resp. $`\stackrel{~}{Q}`$). In the first case, this yields a factorization into a map in $`\stackrel{~}{P}\mathrm{proj}`$ followed by a map in $`\stackrel{~}{P}\mathrm{cocell}`$, where $`\stackrel{~}{P}\mathrm{cocell}`$ consists of maps which are transfinite compositions of pullbacks of maps in $`\stackrel{~}{P}`$. But we have shown that $`\stackrel{~}{P}\mathrm{proj}`$ is equal to the class of projective trivial cofibrations and $`\stackrel{~}{P}\mathrm{cocell}\stackrel{~}{P}\mathrm{fib}`$, \[Ho1\] dual to Lemma 2.1.10. Since we have shown that $`\stackrel{~}{P}\mathrm{fib}`$ is equal to the class of projective fibrations, this is the first desired factorization.
The cosmall object argument applied to $`\stackrel{~}{Q}`$ yields a factorization into a map in $`\stackrel{~}{Q}\mathrm{proj}`$ followed by a map in $`\stackrel{~}{Q}\mathrm{cocell}`$. Since we know that $`\stackrel{~}{Q}\mathrm{proj}`$ is equal to the class of projective cofibrations and $`\stackrel{~}{Q}\mathrm{cocell}\stackrel{~}{Q}\mathrm{fib}`$ and since $`\stackrel{~}{Q}\mathrm{fib}`$ is equal to the class of projective trivial fibrations, we have the second factorization.
This proves that the projective structure is in fact a model structure on $`\mathrm{Sp}(𝒞,T)`$. But we have also shown that this structure satisfies the axioms of a fibrantly generated model category, see \[Ho1\], Definition 2.1.17.
The left properness is clear since coproducts are defined levelwise in $`\mathrm{Sp}(𝒞,T)`$. This completes the proof of Theorem B.6. $`\mathrm{}`$
It remains to modify this structure in order to get a stable structure, i.e. one in which the prolongation of $`T`$ is a Quillen equivalence. We will do this by applying the localization Theorem A.3 to the projective model structure on spectra. We want the stable weak equivalences to be the maps that induce isomorphisms on all generalized cohomology theories. A generalized cohomology theory is represented by the analogue of an $`\mathrm{\Omega }`$-spectrum. Since the $`K`$-local objects are the fibrant objects in the localized model structure, we have to choose the set $`K`$ to consist exactly of these analogues of $`\mathrm{\Omega }`$-spectra.
###### Definition B.10
A spectrum $`E\mathrm{Sp}(𝒞,T)`$ is defined to be an $`\mathrm{\Omega }`$-spectrum if each $`E_n`$ is fibrant and the adjoint structure maps $`E_nUE_{n+1}`$ are weak equivalences for all $`n0`$.
By Corollary 9.7.5 in \[Hi\], since each $`E_n`$ is fibrant and since the right Quillen functor $`U`$ preserves fibrations, we have that $`E_nUE_{n+1}`$ is a weak equivalence in $`𝒞`$ if and only if the induced map $`\mathrm{Map}(A,E_n)\mathrm{Map}(A,UE_{n+1})`$ is a weak equivalence in $`S`$ for every cofibrant object $`A`$. By adjunction this is equivalent to $`\mathrm{Map}(F_nA,E)\mathrm{Map}(F_{n+1}TA,E)`$ being a weak equivalence in $`S`$ for every cofibrant object $`A`$. Note that by \[Hi\], Corollary 9.7.5, this is what the weak equivalences have to be in a model structure.
So in order to get the $`\mathrm{\Omega }`$-spectra as stable fibrant objects and the maps inducing isomorphisms on cohomology theories as stable equivalences we have to choose the maps $`F_{n+1}TAF_nA`$ adjoint to the identity map of $`TA`$ to be the $`K`$-local equivalences. Hence we define the set $`K`$ to consist of all $`\mathrm{\Omega }`$-spectra.
Using the fact that the projective model structure on $`\mathrm{Sp}(𝒞,T)`$ is fibrantly generated, left proper and simplicial, we may deduce the following
###### Theorem B.11
Let $`𝒞`$ be a left proper fibrantly generated simplicial model category with all small limits. Let $`K`$ be the set of $`\mathrm{\Omega }`$-spectra. There is a stable model structure on $`\mathrm{Sp}(𝒞,T)`$ which is defined to be the $`K`$-localized model structure $`L_S\mathrm{Sp}(𝒞,T)`$ of Theorem A.3 where $`S`$ is the class of all $`K`$-local equivalences.
###### Remark B.12
In particular, the stable equivalences are the $`K`$-local equivalences; the stable cofibrations are the projective cofibrations; the stable fibrations are the maps that have the right lifting property with respect to all stable trivial cofibrations; the stable fibrant objects are the $`\mathrm{\Omega }`$-spectra.
Define the prolongation of $`T`$ to be the functor $`T:\mathrm{Sp}(𝒞,T)\mathrm{Sp}(𝒞,T)`$, defined by $`(TX)_n=TX_n`$ with structure maps $`T(TX_n)\stackrel{T\sigma }{}TX_{n+1}`$ where $`\sigma `$ is the structure map of $`X`$. The adjoint $`\mathrm{\Omega }`$ is defined in the analog way. Just as in \[Ho2\] one can prove that $`T`$ is a Quillen equivalence on the stable structure.
###### Theorem B.13
Let $`𝒞`$ be a left proper fibrantly generated simplicial model category with a left Quillen endofunctor $`T`$. Then the prolongation $`T:\mathrm{Sp}(𝒞,T)\mathrm{Sp}(𝒞,T)`$ is a Quillen equivalence with respect to the stable structure.
Proof This is the proof of Theorem 3.9 in \[Ho2\]. $`\mathrm{}`$
### B.2 Symmetric spectra
We can transfer these results also to symmetric spectra on a left proper fibrantly generated simplicial model category $`𝒞`$ with a symmetric monoidal structure. The ideas and definitions are due to \[Ho2\]. We extend his results to our unusual and slightly more general situation. For the definition of these notions, we refer the reader to \[Ho1\].
For the whole section, let $`𝒞`$ be a left proper fibrantly generated simplicial closed symmetric monoidal model category with all small limits. For simplicity, we suppose that its unit object $`S`$ is cofibrant. Since any endofunctor $`T:𝒞𝒞`$ is of the form $`T(L)=LK`$ for $`K=T(S)`$, we will consider in this section the functor $`K`$ instead of $`T`$. The right adjoint of $`T`$ is then given by $`LL^K`$. All the definitions and ideas for the construction of symmetric spectra are taken from \[Ho2\].
###### Definition B.14
A symmetric sequence in $`𝒞`$ is a sequence $`X_0,X_1,\mathrm{},X_n,\mathrm{}`$ of objects in $`𝒞`$ with an action of $`\mathrm{\Sigma }_n`$ on $`X_n`$, where $`\mathrm{\Sigma }_n`$ is the symmetric group of $`n`$ letters. A map of symmetric sequences is a sequence of $`\mathrm{\Sigma }_n`$-equivariant maps $`X_nY_n`$. We denote the resulting category by $`𝒞^\mathrm{\Sigma }`$.
We define the object $`\mathrm{Sym}(K)`$ in $`𝒞^\mathrm{\Sigma }`$ to be the symmetric sequence $`(S,K,KK,\mathrm{},K^n,\mathrm{})`$ where $`\mathrm{\Sigma }_n`$ acts on $`K^n`$ by permutation, using the commutativity and associativity isomorphisms.
###### Definition B.15
A symmetric spectrum $`X`$ is a sequence of $`\mathrm{\Sigma }_n`$-objects $`X_n`$ in $`𝒞`$ with $`\mathrm{\Sigma }_n`$-equivariant structure maps $`X_nKX_{n+1}`$, such that the composite
$$X_nK^pX_{n+1}K^{p1}\mathrm{}X_{n+p}$$
is $`\mathrm{\Sigma }_n\times \mathrm{\Sigma }_p`$-equivariant for all $`n,p0`$. A map of symmetric spectra is a collection of $`\mathrm{\Sigma }_n`$-equivariant maps $`X_nY_n`$ compatible with the structure maps. We denote the category of symmetric spectra on $`𝒞`$ by $`\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$.
We define the symmetric spectrum $`\overline{\mathrm{Sym}(K)}`$ in $`\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$ to be the initial object $`0`$ in degree $`0`$ and $`K^n`$ in degree $`n`$ for $`n>0`$ with the obvious structure maps.
One can show as in \[HSS\] that $`\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$ is again a closed symmetric monoidal category with $`\mathrm{Sym}(K)`$ as the unit. We denote the monoidal structure by $`XY=X_{\mathrm{Sym}(K)}Y`$, and the closed structure by $`\mathrm{Hom}_{\mathrm{Sym}(K)}(X,Y)`$, where we consider $`X`$ and $`Y`$ as $`\mathrm{Sym}(K)`$-modules.
###### Definition B.16
1. Given $`n0`$, the evaluation functor $`\mathrm{Ev}_n:\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)𝒞`$ takes $`X`$ to $`X_n`$.
2. The evaluation functor has a left adjoint $`F_n:𝒞\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$, defined by $`F_nA=\stackrel{~}{F}_nA\mathrm{Sym}(K)`$, where $`\stackrel{~}{F}_nA`$ is the symmetric sequence $`(0,\mathrm{},0,\mathrm{\Sigma }_n\times A,0,\mathrm{})`$.
3. The evaluation functor has a right adjoint $`R_n:𝒞\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$, defined by $`R_nA=\mathrm{Hom}(\mathrm{Sym}(K),\stackrel{~}{R}_nA)`$, where $`\stackrel{~}{R}_nA`$ is the symmetric sequence that is the terminal object in dimensions other than $`n`$ and is the cofree $`\mathrm{\Sigma }_n`$-object $`\mathrm{Hom}_𝒞(\mathrm{\Sigma }_n,A)`$, i.e. $`n!`$-product of $`A`$ in $`𝒞`$ together with the $`\mathrm{\Sigma }_n`$-action defined by $`(\rho f)(\rho ^{})=f(\rho ^{}\rho )`$ for $`\rho \mathrm{\Sigma }_n`$ and $`f\mathrm{Hom}_𝒞(\mathrm{\Sigma }_n,A)`$.
###### Definition B.17
A map $`f`$ in $`\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$ is a projective weak equivalence (resp. projective fibration) if each map $`f_n`$ is a weak equivalence (resp. fibration). A map $`i`$ is a projective cofibration if it has the left lifting property with respect to all projective trivial fibrations.
Again we will show that the projective structure is in fact a fibrantly generated model structure. We recall that the set of generating fibrations of $`𝒞`$ is denoted by $`P`$ and the set of generating trivial fibrations by $`Q`$. Inspired by the forthcoming proposition we set
$`\stackrel{~}{P}`$ $`:=`$ $`\{q_n:R_nXR_nY\times _{R_n\mathrm{Hom}_K(K^n,Y)}R_n\mathrm{Hom}_K(K^n,X),`$
$`\mathrm{for}\mathrm{all}p:XY\mathrm{in}P\}`$
and
$`\stackrel{~}{Q}`$ $`:=`$ $`\{q_n:R_nXR_nY\times _{R_n\mathrm{Hom}_K(K^n,Y)}R_n\mathrm{Hom}_K(K^n,X),`$
$`\mathrm{for}\mathrm{all}p:XY\mathrm{in}Q\}`$
where $`R_n`$ is the right adjoint of the evaluation functor $`Ev_n`$ and $`q_n`$ is the induced map in the commutative diagram
(38)
$$\begin{array}{ccc}R_nX& & R_nY\\ & & \\ R_n(\mathrm{Hom}_K(K^n,X))& & R_n(\mathrm{Hom}_K(K^n,Y)).\end{array}$$
We will show that the maps in $`\stackrel{~}{P}`$ are the generating fibrations and the maps in $`\stackrel{~}{Q}`$ are the generating trivial fibrations for the projective model structure on $`\mathrm{Sp}(𝒞,T)`$. First we describe the projective cofibrations in $`\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$. This is slightly more complicated than in the case of usual spectra. The reason for this comes from the fact that we have to ensure that all maps are equivariant with respect to the actions of the groups $`\mathrm{\Sigma }_n`$. The following definition and the idea how to solve this problem is taken from \[Ho2\], Definition 8.4.
The following proposition is well-known, see \[Ho2\], Proposition 8.5, or \[HSS\], Proposition 5.2.2. But we will give a different proof similar to the one of Proposition B.5, which enables us to understand the definition of the generating fibrations and trivial fibrations in the projective model structure on $`\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$. It avoids the assumption that we had already shown that the projective structure is in fact a model structure. Furthermore, we do not have to assume that $`𝒞`$ is cofibrantly generated, which is not the case in our situation.
###### Definition B.18
We define the $`n`$-th latching space of $`A\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$ by $`L_nA:=\mathrm{Ev}_n(A\overline{\mathrm{Sym}(K)})`$. The obvious map $`\overline{\mathrm{Sym}(K)}\mathrm{Sym}(K)`$ induces a map of spectra $`i:A\overline{\mathrm{Sym}(K)}A\mathrm{Sym}(K)`$ and a $`\mathrm{\Sigma }_n`$-equivariant map $`\mathrm{Ev}_n(i):L_nAA_n`$.
Note that the latching space is a $`\mathrm{\Sigma }_n`$-object in $`𝒞`$. There is a model structure on the category of $`\mathrm{\Sigma }_n`$-objects of $`𝒞`$, where the fibrations and weak equivalences are the underlying ones.
We recall that for given maps $`f:AB`$ and $`g:CD`$ in $`𝒞`$ one defines the pushout product $`f\mathrm{}g`$ of $`f`$ and $`g`$ to be the map
$$f\mathrm{}g:(AD)_{AC}(BC)BD$$
induced by the obvious commutative diagram.
###### Proposition B.19
A map $`f:AB`$ in $`\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$ is a projective cofibration if and only if the induced map $`\mathrm{Ev}_n(f\mathrm{}i):A_n_{L_nA}L_nBB_n`$ is a $`\mathrm{\Sigma }_n`$-cofibration in $`𝒞`$ for all $`n`$. Similarly, $`f`$ is a projective trivial cofibration if and only if $`\mathrm{Ev}_n(f\mathrm{}i)`$ is a trivial $`\mathrm{\Sigma }_n`$-cofibration in $`𝒞`$ for all $`n`$.
Proof As usual we prove only the cofibration case, since the case of trivial cofibrations is similar. Suppose first that each map $`\mathrm{Ev}_n(f\mathrm{}i)`$ is a $`\mathrm{\Sigma }_n`$-cofibration in $`𝒞`$. The following argument is exactly the one in the proof of Proposition 8.5 in \[Ho2\].
Suppose that $`f`$ is a projective cofibration of symmetric spectra. We have to show that each map $`\mathrm{Ev}_n(f\mathrm{}i)`$ is a $`\mathrm{\Sigma }_n`$-cofibration in $`𝒞`$, i.e. each map $`\mathrm{Ev}_n(f\mathrm{}i)`$ has the left lifting property with respect to every trivial $`\mathrm{\Sigma }_n`$-fibration $`p:XY`$ in $`𝒞`$. We recall that this just means that $`p`$ is an underlying trivial fibration in $`𝒞`$. Suppose we have a commutative diagram
(39)
$$\begin{array}{ccc}\hfill A_n_{L_nA}L_nB& & X\\ \hfill \mathrm{Ev}_n(f\mathrm{}i)& & p\\ \hfill B_n& & Y.\end{array}$$
By adjunction, this diagram has a lift if and only if the induced diagram
$$\begin{array}{ccc}A& & R_nX\\ f& & \\ B& & R_nY\times _{\mathrm{Hom}_{\mathrm{Sym}(K)}(\overline{\mathrm{Sym}(K)},R_nY)}\mathrm{Hom}_{\mathrm{Sym}(K)}(\overline{\mathrm{Sym}(K)},R_nX).\end{array}$$
Using the fact, that $`\mathrm{Ev}_n`$ and $`R_n`$ is a pair of adjoint closed monoidal functors, we get an isomorphism of objects in $`\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$
(40)
$$\mathrm{Hom}_{\mathrm{Sym}(K)}(A,R_nZ)R_n\mathrm{Hom}_K(\mathrm{Ev}_nA,Z)$$
where $`\mathrm{Hom}_K`$ and $`\mathrm{Hom}_{\mathrm{Sym}(K)}`$ denote the closed structures of $`𝒞`$ and $`\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$, respectively. When we apply this isomorphism to the above diagram, we deduce that there is a lift in diagram(39) if and only if there is a lift in the following diagram
(41)
$$\begin{array}{ccc}A& & R_nX\\ f& & q_n\\ B& & R_nY\times _{R_n\mathrm{Hom}_K(K^n,Y)}R_n\mathrm{Hom}_K(K^n,X).\end{array}$$
Since $`K`$ is a cofibrant object in $`𝒞`$ and $`𝒞`$ being a closed monoidal model category, the functors $`R_n`$, $`\mathrm{Hom}_K(K^n,)`$ and pullbacks preserve fibrations and trivial fibrations, the map $`q_n`$ is a projective trivial fibration in $`\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$. Hence there is a lift in the last diagram, when $`f`$ is a projective cofibration.
Now, if each map $`\mathrm{Ev}_n(f\mathrm{}i)`$ is a $`\mathrm{\Sigma }_n`$-cofibration in $`𝒞`$, one can construct by induction on $`n`$ a lift $`BX`$ in the commutative diagram
$$\begin{array}{ccc}A& \stackrel{a}{}& X\\ f& & p\\ B& \stackrel{b}{}& Y\end{array}$$
where $`p`$ is a level trivial fibration. The argument is exactly the one given in the first part of the proof of Proposition 8.5 in \[Ho2\]. $`\mathrm{}`$
###### Proposition B.20
The projective cofibrations are exactly the maps which have the left lifting property with respect to all maps in $`\stackrel{~}{Q}`$. The projective trivial cofibrations are exactly the maps which have the left lifting property with respect to all maps in $`\stackrel{~}{P}`$.
Proof By Proposition B.19 $`f:AB`$ is a projective cofibration if and only if $`\mathrm{Ev}_n(f\mathrm{}i)`$ are $`\mathrm{\Sigma }_n`$-cofibrations in $`𝒞`$ for all $`n`$. Since $`𝒞`$ is a fibrantly generated model category with $`Q`$ the set of generating trivial fibrations, $`\mathrm{Ev}_n(f\mathrm{}i)`$ are $`\mathrm{\Sigma }_n`$-cofibrations if and only if they have the left lifting property with respect to all maps in $`Q`$, i.e. if and only if $`\mathrm{Ev}_n(f\mathrm{}i)`$ are in $`Q\mathrm{proj}`$. We have seen that a lift in a diagram
$$\begin{array}{ccc}\hfill A_n_{L_nA}L_nB& & X\\ \hfill \mathrm{Ev}_n(f\mathrm{}i)& & p\\ \hfill B_n& & Y\end{array}$$
is by adjunction equivalent to a lift in the diagram
$$\begin{array}{ccc}A& & R_nX\\ f& & q_n\\ B& & R_nY\times _{R_n\mathrm{Hom}_K(K^n,Y)}R_n\mathrm{Hom}_K(K^n,X).\end{array}$$
Hence the maps $`\mathrm{Ev}_n(f\mathrm{}i)`$ are in $`Q\mathrm{proj}`$ if and only if the map $`f`$ is an element of $`\stackrel{~}{Q}\mathrm{proj}`$. This completes the proof in the case of a cofibration. The case of a trivial cofibration is similar. $`\mathrm{}`$
###### Corollary B.21
The projective fibrations are exactly the maps in $`\stackrel{~}{P}\mathrm{fib}`$. The projective trivial cofibrations are exactly the maps in $`\stackrel{~}{Q}\mathrm{fib}`$.
As in the case of ordinary spectra we deduce the following
###### Theorem B.22
Let $`𝒞`$ be a left proper fibrantly generated simplicial closed symmetric monoidal model category with all small limits. The projective weak equivalences, projective fibrations and projective cofibrations define a left proper fibrantly generated simplicial closed symmetric monoidal model structure on $`\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$ with set of generating fibrations $`\stackrel{~}{P}`$ and set of generating trivial fibrations $`\stackrel{~}{Q}`$ and unit object $`\mathrm{Sym}(K)`$.
Proof The model structure part may be deduced exactly as in Theorem B.6. The only thing we have to explain is the closed symmetric monoidal structure on $`\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$. We have to show that given a projective cofibration $`f:AB`$ and a projectivfe fibration $`g:WZ`$ the induced map $`\mathrm{Hom}_{\mathrm{Sym}(K),\mathrm{}}(f,g)`$
$$\mathrm{Hom}_{\mathrm{Sym}(K)}(B,W)\mathrm{Hom}_{\mathrm{Sym}(K)}(B,Z)\times _{\mathrm{Hom}_{\mathrm{Sym}(K)}(A,Z)}\mathrm{Hom}_{\mathrm{Sym}(K)}(A,W)$$
is a fibration which is trivial if $`f`$ or $`g`$ is trivial. By the adjoint statement of Corollary 4.2.5 of \[Ho1\], it suffices to show this if $`g`$ is a generating fibration or generating trivial fibration, respectively. So we suppose $`g=q_n:R_nXR_nY\times _{R_n\mathrm{Hom}_K(K^n,Y)}R_n\mathrm{Hom}_K(K^n,X)`$ in $`\stackrel{~}{P}`$ or $`\stackrel{~}{Q}`$ as above. Then the map $`\mathrm{Hom}_{\mathrm{Sym}(K),\mathrm{}}(f,q_n)`$ becomes
$$\mathrm{Hom}_{\mathrm{Sym}(K)}(B,R_nX)\mathrm{Hom}_{\mathrm{Sym}(K)}(B,R_n(X,Y))\times _{\mathrm{Hom}_{\mathrm{Sym}(K)}(A,R_n(X,Y))}\mathrm{Hom}_{\mathrm{Sym}(K)}(A,R_nX)$$
where we denote $`R_nY\times _{R_n\mathrm{Hom}_K(K^n,Y)}R_n\mathrm{Hom}_K(K^n,X)`$ by $`R_n(X,Y)`$. Using adjuntion (40) and the fact that the diagrams (39) and (41) are adjoint to each other, we see that the map $`\mathrm{Hom}_{\mathrm{Sym}(K),\mathrm{}}(f,q_n)`$ is in fact equal to the map $`R_n\mathrm{Hom}_{K,\mathrm{}}(\mathrm{Ev}_n(f\mathrm{}i),p)`$
$$R_n\mathrm{Hom}_K(B_n,X)R_n\mathrm{Hom}_K(B_n,Y)\times _{R_n\mathrm{Hom}_K(A_n_{L_nA}L_nB,Y)}R_n\mathrm{Hom}_K(A_n_{L_nA}L_nB,X)$$
where $`p:XY`$ is the map in $`𝒞`$ corresponding to $`q_n`$, as above. Since $`R_n`$ is a right Quillen functor and since $`𝒞`$ is a closed symmetric monoidal model category, we get that the map $`R_n\mathrm{Hom}_{K,\mathrm{}}(\mathrm{Ev}_n(f\mathrm{}i),p)`$ is a fibration which is trivial if $`\mathrm{Ev}_n(f\mathrm{}i)`$ or $`p`$ is trivial. By Proposition B.19, the map $`f`$ is a (trivial) projective cofibration if and only if the maps $`\mathrm{Ev}_n(f\mathrm{}i)`$ are (trivial) cofibrations for all $`n`$. Since $`q_n`$ is a (trivial) fibration if and only if $`p`$ is a (trivial) fibration, we are done. $`\mathrm{}`$
In order to obtain a stable model structure on $`\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$ we have to localize the projective structure at an appropriate set of fibrant objects.
###### Definition B.23
A symmetric spectrum $`X\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$ is a symmetric $`\mathrm{\Omega }`$-spectrum if $`X`$ is level fibrant and the adjoint $`X_nX_{n+1}^K`$ of the structure map of $`X`$ is a weak equivalence for all $`n`$.
A symmetric Spectrum $`E`$ is called injective if it has the extension property with respect to every monomorphism $`f`$ of symmetric spectra that is also a level equivalence, i.e. for every diagram in $`\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$
$$\begin{array}{ccc}\hfill X& \stackrel{g}{}& E\\ \hfill f& & \\ \hfill Y& & \end{array}$$
where $`f`$ is a level equivalence and a monomorphism there is a map $`h:YE`$ such that $`g=hf`$.
We want the symmetric $`\mathrm{\Omega }`$-spectra to be the stable fibrant objects in $`\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$. As in the previous section we have to choose the set maps
$$𝒮:=\{\zeta _n^A:F_{n+1}(AK)F_nA;A\mathrm{cofibrant}\mathrm{in}𝒞\}$$
to be the $`S`$-local equivalences, where the map $`\zeta _n^A`$ is adjoint to the map
$$AK\mathrm{Ev}_{n+1}F_nA=\mathrm{\Sigma }_{n+1}\times (AK)$$
corresponding to the identity of $`\mathrm{\Sigma }_{n+1}`$. With the set $`S`$ consisting of all symmetric $`\mathrm{\Omega }`$-spectra we obtain the desired set $`𝒮`$ of $`S`$-local equivalences.
###### Definition B.24
Let $`𝒞`$ be a left proper fibrantly generated simplicial model category with all small limits. Let $`S`$ be the set of all symmetric $`\mathrm{\Omega }`$-spectra. We define the stable model structure on $`\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$ to be the $`S`$-localized model structure $`L_𝒮\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$ of Theorem A.3 where $`𝒮`$ is the class of all $`S`$-local equivalences.
###### Remark B.25
In particular, the stable equivalences are the $`S`$-local equivalences; the stable cofibrations are the projective cofibrations; the stable fibrations are the maps that have the right lifting property with respect to all stable trivial cofibrations; the stable fibrant objects are the symmetric $`\mathrm{\Omega }`$-spectra.
###### Theorem B.26
The functor $`K`$ is a Quillen equivalence with respect to the stable structure of $`\mathrm{Sp}^\mathrm{\Sigma }(𝒞,K)`$.
Proof The proof is given by \[Ho2\], Theorem 8.10.$`\mathrm{}`$
Now we specialize to the case of $`𝒞=\widehat{𝒮}`$, which is a simplicial fibrantly generated closed symmetric monoidal model category. We can prove the following theeorem only for this special case. In order to prove the following theorem we use the ideas of the proof of Theorem 5.3.7 of \[HSS\]. in particular, we use the fact that we can build the cofiber $`Ci`$ of a map of spectra $`i:XY`$. In the following we denote the internal Hom-object in $`\mathrm{Sp}^\mathrm{\Sigma }(\widehat{𝒮},S^1)`$ by $`\mathrm{Hom}_S(X,Y)`$.
###### Proposition B.27
Let $`f:XY`$ be a map of symmetric profinite spectra. The following conditions are equivalent:
a) $`E^0f`$ is an isomorphism for every injective $`\mathrm{\Omega }`$-spectrum $`E`$.
b) $`\mathrm{Map}_{\mathrm{Sp}^\mathrm{\Sigma }(\widehat{𝒮},S^1)}(f,E)`$ is a weak equivalence for every injective $`\mathrm{\Omega }`$-spectrum $`E`$.
c) $`\mathrm{Hom}_S(f,E)`$ is a level equivalence for every injective $`\mathrm{\Omega }`$-spectrum $`E`$.
Proof The proof is the one of Proposition 3.1.4 in \[HSS\]. CHECK!! $`\mathrm{}`$
###### Lemma B.28
Let $`f:XY`$ be a map of symmetric profinite spectra.
1. If $`E\mathrm{Sp}^\mathrm{\Sigma }(\widehat{𝒮},S^1)`$ is an injective spectrum and $`f`$ is a level equivalence, then $`E^0f`$ is an isomorphism of sets.
2. If $`f:XY`$ is a map of injective spectra, $`f`$ is a level equivalence if and only if $`f`$ is a simplicial homotopy equivalence.
Proof The proof is the one of Lemma 3.1.6 in \[HSS\]. CHECK!! $`\mathrm{}`$
###### Theorem B.29
The stable model structure on $`\mathrm{Sp}^\mathrm{\Sigma }(\widehat{𝒮},S^1)`$ is monoidal.
Proof Since we have already proved that the projective model structure is monoidal and the stable cofibrations are exactly the projective cofibrations, we only have to show that for two stabel cofibrations $`f`$ and $`g`$ the map $`f\mathrm{}g`$ is also a stable equivalence if $`f`$ or $`g`$ is a stably trivial cofibration.
Suppose that $`g`$ is a stable equivalence. The other case is of course similar. A level cofibration $`i:XY`$ is a stabel equivalence if and only if its cofiber $`C_i=Y/X`$ is stably trivial. We have already seen that $`f\mathrm{}g`$ is a projective and hence in particular a level cofibration. By commuting colimits, the cofiber of $`f\mathrm{}g`$ is the smash product $`CfCg`$ of the cofiber $`Cf`$ of $`f`$ and the cofiber $`Cg`$ of $`g`$. Let $`E`$ be a symmetric $`\mathrm{\Omega }`$-spectrum. We show that $`\mathrm{Hom}_S(CfCg,E`$ is a level trivial spectrum, and thus $`CfCg`$ is stably trivial.
By adjointness, we have $`\mathrm{Hom}_S(CfCg,E)\mathrm{Hom}_S(Cg,\mathrm{Hom}_S(Cf,E))`$. We will prove in the following lemma that $`D:=\mathrm{Hom}_S(Cf,E)`$ is an injective $`\mathrm{\Omega }`$-spectrum. Together with the above adjuntion and the fact that $`Cg`$ is stably trivial by hypothesis, this proves that $`CfCg`$ is stably trivial.
###### Lemma B.30
Let $`E`$ be an injective spectrum and let $`h`$ be a projective cofibration. Then $`\mathrm{Hom}_S(h,E)`$ has the right lifting property with respect to all level monomorphisms which are also level equivalences.
In particular, if $`A`$ is a projective cofibrant spectrum, then $`\mathrm{Hom}_S(A,E)`$ is an injective spectrum.
Proof Let $`g`$ and $`h`$ be level monomorphisms and let $`g`$ be a level equivalence. The $`g\mathrm{}h`$ is defined as a coequalizer of colimits and pushouts of the maps $`g_p\mathrm{}h_q`$. Since $`\widehat{𝒮}_{}`$ is a monoidal model category, these maps are monomorphisms and weak equivalences. Since colimits and pushouts preserve trivial cofibrations when all objects are cofibrant, $`g\mathrm{}h`$ is still a level monomorphism and level equivalence. This implies that if $`E`$ is an injective spectrum, then the pair $`(g\mathrm{}h,E)`$ has the lifting property. By adjointness, we have an isomoprhism $`\mathrm{Hom}_{\mathrm{Sp}^\mathrm{\Sigma }(\widehat{𝒮},S^1)}(g,\mathrm{Hom}_S(h,E))\mathrm{Hom}_{\mathrm{Sp}^\mathrm{\Sigma }(\widehat{𝒮},S^1)}(g\mathrm{}h,E)`$. This proves that $`\mathrm{Hom}_S(h,E)`$ has the right lifting property with respect to all level trivial cofibrations. $`\mathrm{}`$
###### Lemma B.31
Let $`E`$ be an injective $`\mathrm{\Omega }`$-spectrum and let $`A`$ be a projective cofibrant spectrum. Then $`\mathrm{Hom}_S(A,E)`$ is an $`\mathrm{\Omega }`$-spectrum, too.
Proof We write $`D:=\mathrm{Hom}_S(A,E)`$. By adjointness, we have an isomorphism
$$\mathrm{Ev}_nD\mathrm{Map}_{\mathrm{Sp}^\mathrm{\Sigma }(\widehat{𝒮},S^1)}(AF_nS^0,E)\mathrm{Map}_{\mathrm{Sp}^\mathrm{\Sigma }(\widehat{𝒮},S^1)}(A,\mathrm{Hom}_S(F_nS^0,E)).$$
Since $`E`$ is an $`\mathrm{\Omega }`$-spectrum, $`(F_nS^0\lambda )^{}:\mathrm{Hom}_S(F_nS^0,E)\mathrm{Hom}_S(F_{n+1}S^1,E)`$ is a level equivalence, where $`\lambda :F_1S^1F_0S^0`$ is the adjoint to the identity map $`S^1\mathrm{Ev}_1S=S^0`$. Since $`E`$ is injective, both the source and thee target are also injective, and so this map is a simplicial homotopy equivalence by Lemma B.28. Hence $`\mathrm{Ev}_nD(\mathrm{Ev}_{n+1}D)^{S^1}`$ is still a level equivalence, so $`D=\mathrm{Hom}_S(A,E)`$ is an injective $`\mathrm{\Omega }`$-spectrum. $`\mathrm{}`$ $`\mathrm{}`$
###### Corollary B.32
The profinite completion is a monoidal left Quillen functor from symmetric spectra to profinite symmetric spectra.
Mathematisches Institut, WWU Münster, Einsteinstr. 62, 48149 Münster, Germany
E-mail address: gquick@math.uni-muenster.de
|
warning/0506/hep-th0506007.html
|
ar5iv
|
text
|
# Contents
## 1 Introduction and overview
Time evolution between two metastable configurations in string theory was related by A. Sen to a renormalisation group flow between the underlying worldsheet theories. The original proposal concerned dynamic processes of D-branes in a given closed string background, driven by the better understood condensation of open string tachyons. In recent years, however, the condensation of closed string tachyons has been addressed ingeniously in a simple class of toy models, namely non-supersymmetric orbifolds of flat space $`^2/_{N,p}`$ .
These toy models are examples of theories with extended worldsheet supersymmetry which nevertheless have space-time tachyons (and hence break space-time supersymmetry). The key idea is that this extended supersymmetry is preserved all along the renormalisation group flows driven by target space tachyons which are BPS states of the worldsheet theory. This property allows to describe some properties of the endpoint of the flow; thus, the flows studied for $`^2/_{N,p}`$ were shown to describe a gradual opening of the orbifold singularity, eventually leading in the infrared to flat space.
One of the original difficulties of the topic – that closed string tachyon condensation would lead, according to the Zamolodchikov $`c`$-theorem, to a decrease of the central charge – has been turned into the so-called $`g_{cl}`$-conjecture : that localised closed string tachyons in a non-compact space contribute at subleading order (as compared to untwisted states) to the free energy of the theory, and that it is this contribution $`g_{cl}`$ which decreases with their condensation. In the toy models above, the tachyons are localised at the orbifold singularity and everywhere else the theory is free.
The purpose of this note is to show how this strategy can be applied to study closed string tachyons in certain curved, smooth backgrounds. These include orbifolds of the $`SU(2)`$ WZW model, the generalised lens spaces $`L_{N,p}=SU(2)/_{N,p}`$ studied in .
The note is organised as follows: we first describe embeddings of generalised lens spaces in string theory in section 2, using NS5-branes. In section 3, we briefly describe the corresponding CFT constructions. We show how space-time supersymmetry is broken and then focus on the chiral ring of the $`𝒩=2`$ worldsheet superconformal algebra. It turns out there exists a level-independent subring which is defined by the pair $`(N,p)`$ and coincides with the chiral ring of the $`^2/_{N,p}`$ theories. Since the methods used in to analyse RG flows rely exclusively on the structure of the chiral ring, such tachyons in our backgrounds drive processes $`(N,p)(N^{},p^{})`$ just like those described for $`^2/_{N,p}`$. The level-dependent subring, on the other hand, is identical to that of $`/_k`$. We save the conclusions and outlook for section 4.
## 2 Smooth, curved, tachyonic backgrounds
Consider a stack of $`k`$ coincident NS5-branes in $`^{1,9}`$. It is well known that the 4-dimensional space transverse to the NS5-branes becomes curved due to backreaction and in the near horizon limit develops a throat of section $`S^3`$ whose radius is parametrized by the dilaton: the corresponding CFT is $`SU(2)_k\times _{dil}`$ . If instead we put the NS5-branes along the $`_{1,5}`$ directions of the flat space orbifold $`_{1,5}\times ^2/_{N,p}`$, the transverse $`SU(2)`$ becomes a generalized lens space $`L_{N,p}`$. To see this, let us consider the $`_N`$ actions involved.
If we parametrise $`^2`$ by the polar coordinates $`(r_1e^{i\varphi },r_2e^{i\psi })`$, then the $`_N`$ acts as $`\varphi \varphi +2\pi /N`$ and $`\psi \psi +2\pi ip/N`$. Adding the NS5-branes then curves the transverse space to $`SU(2)\times `$, where $`SU(2)`$ is spanned by $`(\varphi ,\psi ,\mathrm{arctan}(r_1/r_2))`$, up to these identifications. If we now switch to the Euler coordinates $`g=e^{i\frac{\psi +\varphi }{2}\sigma _3}e^{i\frac{\theta }{2}\sigma _1}e^{i\frac{\psi \varphi }{2}\sigma _3}`$ for $`SU(2)`$, these identifications become
$$g\omega ^{\frac{p+1}{2}}g\omega ^{\frac{p1}{2}}$$
(1)
where we chose $`\omega =e^{i\sigma _3/2}`$ to be the $`_N`$ generator. The quotient by (1) defines the lens space $`L_{N,p}`$. Notice that $`p`$ takes values in $`\{1,\mathrm{},N\}`$. This action is free as long as $`p`$ and $`N`$ are mutually prime, which we will assume in the following. The particular case $`p=1`$ corresponds to the more familiar left quotient of $`SU(2)`$ (furthermore, $`L_{N,p}`$ and $`L_{N,p}`$ are diffeomorphic, being related by a conjugation by $`\omega `$). The various lens spaces with a given $`N`$ are topologically very similar (they have the same cohomologies, but can be distinguished by certain knot invariants ).
Thus the near horizon limit of our configuration is $`L_{N,p}\times _{dil}`$ as advertised. Notice in particular that adding the NS5-branes smoothens out the orbifold singularity by replacing it with the infinite throat. On the other hand, the $`_N`$ quotient in general breaks all spacetime supersymmetry, since it acts differently on the right and on the left. In fact, analysing the spectrum in the next section, we will see that theories with $`p1`$ have spacetime tachyons.
## 3 CFT description and chiral rings
Let us be more precise in the conformal field theory description of our configuration of NS5-branes. Adding $`k`$ NS5 branes at the origin of flat space yields $`SU(2)_k\times _{dil}`$. The $`SU(2)`$ transverse to the NS5-branes can now only be orbifolded by an action whose order $`N`$ divides the level $`k`$, such that the exponential of the Wess-Zumino action is well defined . Thus, we are in fact considering a system of $`k=NN^{}`$ NS5-branes spread out in a circle at a finite distance of the orbifold singularity. This singularity, which would describe the strong string coupling region where the CFT description breaks down, is thus naturally cut off from the theory. According to the analysis of , the chiral algebra describing the theory near the NS5-branes is then $`SL(2,)/U(1)\times SU(2)/U(1)`$.
To construct the superconformal field theory appearing in this near horizon limit, let us recall the bosonic CFT describing string propagation in lens spaces $`L_{N,p}`$ . As explained in the introduction, for our study of tachyon condensation, we need an $`𝒩=2`$ supersymmetric and non-compact version of that. The simplest extension is to add one non-compact direction to the three-dimensional $`L_{N,p}`$.<sup>1</sup><sup>1</sup>1Recall that $`𝒩=2`$ worldsheet supersymmetry implies a target space with complex structure, and thus with even dimension. For that purpose, recall that the relevant conformal field theory has as chiral algebra $`SU(2)/U(1)_k\times U(1)_k`$, and that its spectrum is given by fields of the form $`\mathrm{\Phi }_{j,m}^{pf}\mathrm{\Phi }_n^{u(1)}\overline{\mathrm{\Phi }}_{j^{},m^{}}^{pf}\overline{\mathrm{\Phi }}_n^{}^{u(1)}`$ (where the $`pf`$ stands for the parafermions $`SU(2)/U(1)`$) with selection rules
$`mn=(p+1){\displaystyle \frac{k}{N}}r,m+n=(p1)\left({\displaystyle \frac{k}{N}}r+s\right)\mathrm{mod}2N`$ (2)
$`m^{}=m+2s,n^{}=n+2s+2{\displaystyle \frac{k}{N}}r,j^{}=j`$ (3)
with $`r=0,\mathrm{},N1`$ and $`s=0,\mathrm{},k1`$ and $`p\{1,ldots,N\}`$ (recall that $`N`$ divides $`k`$). In CFT terms the orbifold group is $`_k\times _N`$, where the $`_N`$ factor acts only on the $`u(1)`$ chiral algebra, so there are $`Nk1`$ twisted (left-right asymmetric) sectors, corresponding to non-zero values of $`r`$ and $`s`$. Notice that different choices of $`p`$ only affect the set of asymmetric fields in the spectrum; in (rational) conformal field theory this choice is referred to as discrete torsion .
In moving from the bosonic $`L_{N,p}`$ towards an $`𝒩=2`$ superconformal field theory, it suffices to replace the $`U(1)`$ factor, since the parafermions $`SU(2)_k/U(1)`$ are naturally an $`𝒩=2`$ minimal model. Given the analysis above, the most natural choice is to start from the chiral algebra
$$𝒜=\frac{SL(2,)_{k+2}}{U(1)}\times \frac{SU(2)_{k2}}{U(1)}$$
(4)
The level of the $`𝒩=2`$ “cigar CFT” $`SL(2,)/U(1)`$ has been chosen such that the chiral algebra has (bosonic) central charge $`c=4`$. As the name ”cigar” suggests, the target space of the full $`SL(2,)/U(1)`$ CFT is a semi-infinite cylinder whose compact direction shrinks to zero size at the origin. The natural interpretation of a CFT with symmetry $`𝒜_L\times 𝒜_R`$ and spectrum (2,3), would be that of a lens space whose radius grows along the non-compact cigar dimension. Indeed, different orbifolds of $`𝒜`$ have already been used to describe $`AdS_3\times SU(2)`$, namely
$$\frac{\frac{SL(2,)}{U(1)}\times U(1)}{}\times \frac{\frac{SU(2)}{U(1)}\times U(1)}{_k}$$
(5)
and also to describe $`_\varphi \times SU(2)_k`$, namely
$$\frac{SL(2,)/U(1)\times SU(2)/U(1)}{_k}$$
(6)
In both cases, the orbifold acts at the level of the chiral algebra, ie. it is a symmetric orbifold. In contrast, generalising the $`SU(2)`$ factor in these models to lens spaces involves imposing the asymmetric orbifold constraints (2,3) on the spectrum.
While the proof of modular invariance of general simple current partition functions is complicated , that for the $`L_{N,p}`$ case relies only in the modular properties of the $`U(1)`$ characters, much like Vafa’s original analysis of discrete torsion in the 2-torus . Nevertheless, this analysis does not carry straight away to our case here, because (unlike for the parafermions) the modular transformations of $`SL(2,)/U(1)`$ extended characters do not factorize into $`SL(2,)`$ and $`U(1)`$. Fortunately, in the following the explicit expression of the partition function is not essential, so we will just assume that our theory is defined by the triplet $`(k,N,p)`$ such that its bosonic sector is given by (2,3), as justified by the geometric arguments of the previous section.
We can now nevertheless see how space-time supersymmetry is broken. As mentioned in , the left-moving supercharges are mutually local with the fields of the theory if the latter obey
$$\frac{mn}{k}$$
since that is the conformal weight appearing in the most singular term of the OPE of the supercharges with a given state (and similarly for the right movers). This means that only for $`(N,p)=(1,1)`$ or $`(k,1)`$ are all the left- and right-moving supercharges conserved (actually, the configurations $`(N,1)`$ and $`(k/N,1)`$ are T-dual to each other ). In this case there is a GSO projection which leads to a consistent type II string theory . For non-trivial quotients (both $`N`$ and $`N^{}`$ greater than $`1`$) with $`p=1`$, only the left-moving supercharges are conserved . Again there exists a GSO projection (the restriction of the one above to a chiral half of the Hilbert space) such that these supercharges still generate a consistent type II superstring theory<sup>2</sup><sup>2</sup>2However, in , only the untwisted part of the spectrum of this theory was studied since the purpose was to match it to the supergravity description.. Finally, when $`p1`$, none of the supercharges is conserved and there is no space-time supersymmetry.
### 3.1 The chiral ring
Chiral fields $`\mathrm{\Phi }`$ of the $`𝒩=2`$ superconformal algebra saturate the BPS bound
$$\mathrm{\Delta }_\mathrm{\Phi }=\frac{1}{2}Q_\mathrm{\Phi }$$
(7)
where $`\mathrm{\Delta }_\mathrm{\Phi }`$ is the conformal weight of $`\mathrm{\Phi }`$ and $`Q`$ is its R-charge.<sup>3</sup><sup>3</sup>3The worldsheet $`𝒩=2`$ symmetry factorises naturally in the algebra $`𝒜`$ (4). Here we choose fields which are chiral for both the cigar and the parafermions, and also for both the left and the right-moving algebras. Other choices, involving antichiral fields with $`\mathrm{\Delta }_\mathrm{\Phi }=\frac{1}{2}Q_\mathrm{\Phi }`$ would lead to similar results. Since $`Q`$ is additive, the set of chiral fields forms a ring, called the chiral ring . Following , we are interested in the chiral ring as a way to characterize our theories $`(k,N,p)`$ which is manageable under RG flows. Indeed, one can see in a lagrangian formalism that this BPS property is preserved along the renormalization group flow driven by a chiral relevant perturbation. The chiral ring is deformed under such a flow, but can provide information about the infrared fixed point.
Chiral fields are always in the NS sector. The NS representations of our chiral algebra are specified by 4 indices, eg. the left moving NS states are $`|j,m^{cg}|l,n^{pf}`$. The parafermion indices are restricted to $`l=0,\frac{1}{2},\mathrm{},k/21`$ and $`n`$ is an integer defined modulo $`2k`$ such that $`2l+n=0`$ mod $`2`$. The cigar representations (normalizable and unitary, with integer level $`k+2`$) are similarly labelled by two parameters $`(j,m)`$ and fall into three classes: the positive discrete series, with $`j=0,\frac{1}{2},\mathrm{}`$ and $`m=j+t`$ where $`t`$; the negative discrete series, with $`j=0,\frac{1}{2},\mathrm{}`$ and $`mjt`$ where $`t`$; and the continuous representations, with $`j=\frac{1}{2}+i\lambda `$, where $`\lambda `$, and $`m`$. Their weights and charges are
$`\mathrm{\Delta }_{jm}^{cg}={\displaystyle \frac{m^2j(j1)}{k}},Q_{jm}^{cg}={\displaystyle \frac{2m}{k}}\mathrm{mod}2`$ (8)
$`\mathrm{\Delta }_{ln}^{pf}={\displaystyle \frac{n^2}{4k}}+{\displaystyle \frac{l(l+1)}{k}},Q_{ln}^{pf}={\displaystyle \frac{n}{2k}}\mathrm{mod}2`$ (9)
The chiral fields are therefore of the form
$$|j,j_L^{cg}|l,l_L^{pf}|j^{},j^{}_R^{cg}|l^{},l^{}_R^{pf}$$
(10)
Selecting the chiral fields (10) which obey the selection rule (2,3) will then yield the chiral fields in the theory with target space $`L_{N,p}\times _\varphi `$, see below.
Before we go on to study the structure of the chiral ring, notice that only the discrete representations of $`SL(2,)/U(1)`$ appear in the chiral fields. Since these states are localised at the tip of the cigar, the chiral states of (4) will be localised at the origin of the non-compact direction. In particular, and even though the target space is smooth, the chiral tachyons (those chiral fields with $`Q<1`$, ie. chiral relevant operators) are localised in the $`L_{N,p}`$ submanifold. Rolling down the potential of such tachyons will therefore not change the total central charge, which is dominated by large volume contributions .
Let us now combine the $`𝒩=2`$ chirality condition (7) with the lens space selection rules (2,3). The analysis can be made at the CFT chiral level; left-moving chiral fields satisfying (2) can be divided into three branches, specified by their cigar and parafermionic $`U(1)`$ charges (previously called $`m,n`$):
$`\mathrm{I}:\mathrm{fields}\mathrm{with}r0,s=0,`$ $`\mathrm{of}\mathrm{the}\mathrm{form}`$ $`W_r=|rp{\displaystyle \frac{k}{N}}^{cg}|r{\displaystyle \frac{k}{N}}^{pf}`$
$`\mathrm{II}:\mathrm{fields}\mathrm{with}r=0,s0,`$ $`\mathrm{of}\mathrm{the}\mathrm{form}`$ $`V_s=|(p1)s^{cg}|(p1)s^{pf}`$ (11)
$`\mathrm{III}:\mathrm{fields}\mathrm{with}r=s=0,`$ $`\mathrm{of}\mathrm{the}\mathrm{form}`$ $`T_c=|Nc^{cg}|Nc^{pf}`$ (12)
where in branches I and II $`s=1,N1`$ and $`r=1,k1`$ and in branch III $`c=1,\mathrm{},2k/N`$. Recall that fields with $`r,s0`$ have different left and right $`U(1)`$ charges, via the selection rules (3). In particular, since for lens spaces the parafermionic number $`l`$ must be the same in the left-movers and the right-movers, $`l=l^{}`$, the chiral left-movers with a non-zero $`s`$ value (those in branch II) are coupled to non-chiral right movers and vice-versa. Similarly, in branch I the variable $`r`$ shifts the $`U(1)`$ charge $`n^{}`$ associated to the right-moving $`SL(2,)/U(1)`$. So branch I and II do not give rise to chiral fields of the full conformal field theory.
Finally, the fields in branch III give rise to left-right symmetric (ie. independent of $`p`$) chiral fields of the full CFT. This branch includes tachyonic fields for any $`N>1`$, even in the supersymmetric cases reviewed in section 2 when $`p=1`$. We are led to conclude that these fields are projected out in the supersymmetric $`p=1`$ case, and perhaps also for general $`p`$. In any case, these fields are in the untwisted sector, and so we will not consider them in the following.
### 3.2 Twist fields
So we have no twisted chiral states in our spectrum. Nevertheless, the twisted sectors are just as well described by the twist fields which permute them (even though they are not in the partition function, being non-local). Thus to branches I and II above correspond two branches of twisted fields. An analysis of the boundary conditions of the fields in branch I, for instance, is exactly similar to the one for $`^2/_{N,p}`$ (modulo the change of $`U(1)`$ variables mentioned before equation 1). The branch I twist fields actually form a chiral ring isomorphic to that of $`^2/_{N,p}`$, and is thus independent of $`k`$. In fact, these twist fields are characterized by the same (left-moving) R-charges as the fields in branch I, which are independent of $`k`$. This means that the ring doesn’t change as we increase $`k`$, when the target space $`L_{N,p}\times _\varphi `$ approaches its large volume limit – the flat space orbifold $`^2/_{N,p}`$. It is not surprising, therefore, that branch I of our chiral ring in fact coincides with the chiral fields of $`^2/_{N,p}`$. In fact, if the branch I fields are volume independent they should depend only on the asymptotic structure of the target space; the background curvature is then effectively replaced with a singularity in the same class of ALE spaces.
For branch II, which has only one generator, the analysis is similar to that for $`/_k`$ . These twist fields are similarly characterized by the branch II (left-moving) R-charges, which do depend on $`k`$. These fields thus probe the high-curvature region of $`L_{N,p}\times _{dil}`$, if only the preserved diagonal $`U(1)`$ symmetry of $`SL(2,)/U(1)\times SU(2)/U(1)`$.
To clarify the structure of the complete chiral ring, the diagram of the corresponding R-charges, for the theory with level $`k=20`$, orbifold order $`N=10`$ and discrete torsion parameter $`p=3`$, is presented in Figure 1.
Here we have used the fact that the R-charges are defined modulo 1. The red line indicates the marginal fields, with $`Q_{tot}=Q_{Pf}+Q_{cg}=1`$. The fields below this line are tachyons. The diagram is in units of $`1/k`$, and from it we can extract the structure of the ring – in particular its generators, as pictured. In the general case, the analysis then follows directly from . For completeness, we briefly review it here: for the branch I subring, there are $`[N,p]`$ generators, where $`[N,p]`$ is the number of entries in the continued fraction
$$\frac{N}{p}=a_1\frac{1}{a_2\frac{1}{\mathrm{}\frac{1}{a_{[N,p]}}}}[a_1,\mathrm{},a_{[N,p]}]$$
(13)
and the $`a_i`$ encode the ring structure of branch I. In terms of the twisted sectors (11) this is
$$W_r^{a_r}=W_{(r+1)}W_{(r1)}$$
(14)
In the example of Figure 1, equation (13) becomes $`10/3=[4,2,2]`$ and in particular there are three generators.
On the other hand, branch II has only one generator, corresponding to the twisted sector $`V_1`$, of order $`k`$ and R-charge $`(p1)/k`$. This branch always includes tachyons, except in the supersymmetric case $`p=1`$ where the whole branch collapses to the identity.
### 3.3 Tachyon condensation
Having perturbed the theory with a chiral tachyon, one can study by a variety of methods the chiral ring of the endpoint of the RG flow. For ALE spaces in the class of $`^2/_{N,p}`$, such as $`_{dil}\times L_{N,p}`$, the analysis in is very efficient, so we may use it to study RG flows under our level-independent tachyons $`W_r`$. It was shown there that the R-charge diagram provides an algebraic description of the ALE structure of our space in terms of the pair $`(N,p)`$, and that tachyon condensation here can be thought of as blowing up particular curves of that space. This method thus identifies the ALE structure of the target space of the endpoint of condensation under our tachyon $`W_r`$
$$(N,p)_{W_r}(N_{r_1},p_{r_1})\mathrm{}(N_{r_t},p_{r_t})$$
where the $`r_i`$ and $`t`$ depend on $`W_r`$. Even though the ALE structure on the rhs (ie. the chiral ring of the infrared theory) does not specify the worldsheet theory entirely, it is reasonable to assume that the endpoint of the flow is in fact the (set of) lens space with the corresponding ALE structure. In particular, all the condensations determined in in this way verified the $`g_{cl}`$ conjecture.
The branch II chiral subring (11) coincides with the chiral ring of the flat space orbifold $`/_k`$ . Naively then (if one forgets about the branch II subring), one may be tempted to assume that condensation of a $`V_n`$ tachyon will similarly drive the theory to smaller deficit angles from the point of view of the twisted sectors:
$$/_{2k}_{V_n}/_{k_1(n)}\mathrm{}/_{k_t(n)}$$
where in particular $`_ik_i=k`$. This would describe the gradual disengagement of the $`SL(2,)/U(1)`$ and $`SU(2)/U(1)`$ factors, eventually leading to the direct product $`SL(2,)/U(1)\times SU(2)/U(1)`$. However, a more detailed analysis would be needed to investigate how such flows affect the entire chiral ring.
## 4 Conclusion and Outlook
We have embedded generalized lens spaces $`L_{N,p}=SU(2)/_{N,p}`$ in string theory by adding NS5-branes to the flat space orbifold $`^2/_{N,p}`$ of . This preserves the $`𝒩=2`$ superconformal worldsheet symmetry but leads to spacetime tachyons. The worldsheet chiral ring can be divided into two parts: a level-independent part identical to that of $`^2/_{N,p}`$, and another part identical to that of $`/_k`$. Thus, chiral fields which are spacetime tachyons drive RG flows similar to those analysed in for these two cases.
It would be important to explicitely write the partition functions of our backgrounds, to explore how modular invariant partition functions of non-rational theories exist in families parametrised by discrete parameters (such as the discrete torsion $`p`$) corresponding to different choices of twisted sectors.
It would also be interesting to have an effective description of the tachyons in terms of NS5-branes interactions. For instance, our branch I tachyons act only at the level of the orbifold, so here the question is shifted to how this orbifold is actually implemented in string theory. To circumvent this particularity of our construction, one could try to embed a generalized lens space in string theory by modifying a configuration which already includes a usual lens space $`L_{N,1}`$. Examples include the appearance of Taub-NUT spaces as the 4d transverse space to KK-monopoles . For $`N`$ monopoles distributed at positions $`(\psi _i,\stackrel{}{x}_i)`$ in transverse space, we get the multicentered Taub-NUT metric
$$ds_{}^2=V(x)^1\left(d\psi +\omega x\right)^2+V(x)d\stackrel{}{x}\stackrel{}{x}$$
(15)
where $`\omega `$ is a one-form determined by $`V`$ and we take the ansatz $`V(x)=ϵ+_{i=1}^N|\stackrel{}{x}\stackrel{}{x}_i|`$. Writing $`\stackrel{}{x}`$ in polar coordinates $`(r,\theta ,\varphi )`$ around an $`\stackrel{}{x}_i`$, we can take the near horizon limit $`r0`$. The space spanned by $`(\psi ,\theta ,\varphi )`$ is then a lens space $`L_{M_i,1}`$ where $`M_i`$ is the number of monopoles stacked at $`\stackrel{}{x}_i`$ (the non-trivial $`S^1`$ is along $`\psi `$, whose periodicity assures the space is smooth, turning the singularity at $`x=x_i`$ into a coordinate singularity). It would be very interesting to find a modification of this configuration such that the near horizon limit would include a generalized lens space $`L_{N,p}`$ instead.
#### Acknowledgements:
It is a pleasure to acknowledge very useful conversations with G. d’Appollonio, A. Recknagel, S. Ribault and also with N. Lambert and D. Israel. The author is supported by the grant SFRH/BPD/18872/2004 provided by the Fundação para a Ciência e Tecnologia, Portugal.
|
warning/0506/quant-ph0506099.html
|
ar5iv
|
text
|
# 1 INTRODUCTION
## 1 INTRODUCTION
Entanglement is a distinctively quantum phenomenon whereby a pure state of a composite quantum system may no longer be determined by the states of its constituent subsystems (Schrödinger, 1935). Entangled pure states are those that have mixed subsystem states. To determine an entangled state requires knowledge of the correlations between the subsystems. As no pure state of a classical system can be correlated, such correlations are intrinsically non-classical, as strikingly manifested by the possibility of violating local realism and Bell’s inequalities Bell (1993). In the science of quantum information processing (QIP), entanglement is regarded as the defining resource for quantum communication, as well as an essential feature needed for unlocking the power of quantum computation.
The standard definition of quantum entanglement requires a preferred partition of the overall system into subsystems— that is, mathematically, a factorization of the Hilbert space as a tensor product. Even within quantum mechanics, there are motivations for going beyond such subsystem-based notions of entanglement. Whenever indistinguishable particles are sufficiently close to each other, quantum statistics forces the accessible state space to be a proper subspace of the full tensor product space, and exchange correlations arise that are not a usable resource in the conventional QIP sense. Thus, the natural identification of particles with preferred subsystems becomes problematic. Even if a distinguishable-subsystem structure may be associated to degrees of freedom different from the original particles (such as a set of position or momentum modes, as in Zanardi (2002)), inequivalent factorizations may occur on the same footing. Entanglement-like notions not tied to modes have been proposed for bosons and fermions (Eckert *et al.*, 2002). However, the introduction of quasiparticles, or the purposeful transformation of the algebraic language used to analyze the system (Batista and Ortiz, 2001; Batista *et al.*, 2002), may further complicate the choice of preferred subsystems.
In this paper, we review and further develop *generalized entanglement* (GE) introduced in Barnum *et al.* (2003b), which incorporates the entanglement settings introduced to date in a unifying framework. In quantum-mechanical settings, the key idea behind GE is that entanglement is an *observer-dependent concept*, whose properties are determined by the expectations of a *distinguished subspace of observables* of the system of interest, without reference to a preferred subsystem decomposition. Distinguished observables may represent, for instance, a limited means of manipulating and measuring the system. Standard entanglement is recovered when these means are restricted to arbitrary *local* observables acting on subsystems. The central idea is to generalize the observation that standard entangled pure states are precisely those that look mixed to local observers.
The most fundamental aspects of this notion of GE make use only of the convex structures of the spaces of quantum states and observables. Therfore it is also applicable in contexts much broader than that of quantum systems with distinguished subspaces of observables. It may be formulated within general convex frameworks, based on ordered linear spaces or the closely related notion of convex effect algebras, suitable for investigating the foundations of quantum mechanics and related physical theories (cfr. Beltrametti and Bugajski (1997) and references therein). While commenting on physically motivated special cases, we will concentrate on this general setting in the present paper. Though we make no major advances over Barnum *et al.* (2003b) and Barnum *et al.* (2003a), new material here includes Theorem 3.4 which gives another characterization of the convex cones framework, in terms of restriction to a subspace of observables, and more detailed investigation of the distinguished quantum observables subspace. This includes the introduction of the unique preimage property (Def. 3) and the relationship between the quadratic purity measure, generalized entanglement, and the UPIP in this context, notably Problems 3, 4, and 4, and Propositions 4 and 4.
Two sets of articles contain related ideas. The first originated in the context of $`C^{}`$ and von Neumann algebras, for example in Connes *et al.* (1987), where the dynamical entropy of automorphisms of algebras, intended to generalize the Kolmogorov-Sinai dynamical entropy, is defined — using a notion of entropy of a state’s restriction to a subalgebra introduced in Narnhofer and Thirring (1985). These ideas were further developed with special attention to finding optimal decompositions for the convex roof construction of entropy relative to a subalgebra, and applied to quantum information concepts such as quantum parameter estimation and the entanglement of formation. See e.g. Benatti (1996); Uhlmann (1998); Benatti *et al.* (1996); Benatti and Narnhofer (1998); Benatti *et al.* (2003). The association of subsystems, whether physical or “virtual”/”encoded,” of a quantum system with associative subalgebras appeared in in a second set of articles Knill *et al.* (2000); Filippo (2000); Viola *et al.* (2001); Zanardi (2001); this association was recently revisited, and examples collected, in Zanardi *et al.* (2004). Note, however, that these latter articles were not directly concerned with the extremality properties of reduced states which form the basis of our GE notion. Also, in both sets of articles, the context of subalgebras, whether $`C^{}`$, von Neumann, or associative, is considerably more restrictive than the general context we work in here, except for the fact that Benatti, Connes, Narnhofer, Thirring, and Uhlmann often include and are sometimes primarily interested in infinite-dimensional algebras, whereas we focus here exclusively on the finite-dimensional setting.
## 2 MATHEMATICAL BACKGROUND
For background on cones and convexity, we highly recommend the text by Barvinok (2002), or the short introductory portion of Hilger *et al.* (1989); however, the summary we give here should suffice for what follows.
2.1 Definition A positive cone is a proper subset $`K`$ of a real vector space $`V`$ closed under multiplication by nonnegative scalars. It is called regular if it is (a) convex (equivalently, closed under addition: $`K+K=K`$), (b) generating ($`KK=V`$, equivalently $`K`$ linearly generates $`V`$,) (c) pointed ($`KK=\{0\}`$, so that it contains no non-null subspace of $`V`$), and (d) topologically closed (in the Euclidean metric topology, for finite dimension). In the remainder of this paper, “vector space” and “linear space” will mean finite-dimensional vector space, “cone” will mean a regular cone in a finite-dimensional vector space, unless otherwise stated.
A cone $`K`$ induces a partial order $`_K`$ on $`V`$, defined by $`x_Ky:=xyK`$. It is “linear-compatible”: inequalities can be added, and multiplied by positive scalars. If one removes the requirement that the cones be generating, cones are in one-to-one correspondence with linear-compatible partial orderings. A pair $`V,`$ of a linear space and a distinguished such ordering is called an ordered linear space. The categories of real linear spaces with distinguished cones and partially ordered real linear spaces are equivalent.
Note that the intersection of the interior of a generating cone with a subspace is (if not equal to $`\{0\}`$) a (non-closed but otherwise regular) cone that generates the subspace. When a cone or other set is said to generate a linear space, it does so via linear combination. When a set is said to generate a cone, it does so via positive linear combination. We will use the notation $`\dot{C}`$, for the set $`C\{0\}`$.
By an extremal state in a convex set of states, we mean the usual convex-set notion that a point $`x`$ is extremal in a convex set $`S`$ if (and only if) it cannot be written as a nontrivial convex combination $`x=\lambda _1x_1+\lambda _2x_2`$ of points $`x_1,x_2`$ in $`S`$. (Convex combination means $`\lambda _i0,\lambda _1+\lambda _2=1`$, and nontrivial means $`\lambda _i0,x_1x_2`$). We sometimes use the physics term pure state for an extremal point in a convex set of states, but for clarification we emphasize that when this convex set is the set of all quantum states on some Hilbert space, the term “pure state” in the present paper refers to a projector $`\pi :=|\psi \psi |`$, and not to a vector $`|\psi `$ in the underlying Hilbert space. We write $`\mathrm{Extr}\mathrm{S}`$ for the set of extremal points of a convex set $`S`$.
A ray belonging to a cone $`K`$ is a set $`R`$ such that there exists an $`xK`$ for which $`R=\{\lambda x:\lambda 0\}`$, i.e. it is the set of all nonnegative scalar multiples of some element of the cone. An extreme ray in $`K`$ is a ray $`R`$ such that no $`yR`$ can be written as a convex (or equivalently, positive) combination of elements of $`K`$ that are not in $`R`$. The topological closure condition guarantees, through an easy but not trivial argument using the Krein-Milman theorem, that a (regular) cone is convexly (equivalently, positively) generated by its extreme rays. We’ll say a point is extremal in a cone if it belongs to an extreme ray of the cone; note that such points are not usually extremal in the convex set sense, although the cone is a convex set; the only point in a cone extremal in the convex set sense is zero.
The dual vector space $`V^{}`$ for real $`V`$ is the space of all linear functionals from $`V`$ to $``$; the dual cone $`C^{}V^{}`$ of the cone $`CV`$ is the set of such linear functionals which are nonnegative on $`C`$. $`\lambda V^{}`$ is said to separate $`C`$ from $`C`$ if $`\lambda (x)0`$ for all nonzero $`xC`$. For $`\alpha V^{}`$, $`xV`$, we write the value of $`\alpha `$ on $`x`$ as $`\alpha [x]`$, rather than $`\alpha (x)`$. The adjoint $`\varphi ^{}:V_2^{}V_1^{}`$ of a linear map $`\varphi :V_1V_2`$ is defined by
$`\varphi ^{}(\alpha )[x]=\alpha [\varphi (x)],`$ (1)
for all $`\alpha V_2^{},xV_1`$. The following proposition is easily (but instructively) verified.
2.2 Proposition Let $`C_i`$ be a cone in $`V_i`$ for $`i=1,2,`$ and let $`\varphi (C_1)C_2`$. Then $`\varphi ^{}(C_2^{})C_1^{}`$.
We will also use the following:
2.3 Proposition Let $`C_i`$ be a cone in $`V_i`$ for $`i=1,2,`$ and let $`\varphi (C_1)=C_2`$. Then $`\varphi ^{}(C_2^{})C_1^{}`$ and $`\varphi ^{}`$ is one-to-one.
Proof: Let $`\eta _1,\eta _2C_2^{}`$, and $`\eta _1eta_2`$. $`\eta _1\eta _2`$ is equivalent to the existence of $`y`$ in $`C_2`$ such that $`\eta _1[y]\eta _2[y]`$. By the assumption that $`\varphi `$ maps $`C_1`$ onto $`C_2`$, there is an $`xC_1`$ such that $`\varphi (x)=y`$; thus $`\eta _1[\varphi (x)]\eta _2[\varphi (x)]`$. By the definition of $`\varphi ^{}`$, this implies that $`\varphi ^{}(\eta _1)[x]\varphi ^{}(\eta _2)[x]`$, which implies that $`\varphi ^{}(\eta _1)\varphi ^{}(\eta _2)`$. $`\mathrm{}`$
## 3 GENERALIZED ENTANGLEMENT
We now introduce GE of states in a convex set of states given by the intersection $`\widehat{C}`$ of an affine “normalization” plane $`\{x:\lambda (x)=\alpha \}`$ (for a fixed $`\alpha `$, which we’ll take to be one) with a cone $`C`$ of “unnormalized states.” This GE is a relative notion: states are entangled or unentangled relative to another such state-set $`\widehat{D}`$, and a choice of normalization-preserving map of the first state-set onto the second, which generalizes the notion of computing the reduced density matrices of a bipartite system. To fix ideas, note that in the case where $`C`$ is supposed to represent states on a finite dimensional quantum system whose Hilbert space has dimension $`d`$, $`C`$ is isomorphic to the set of $`d\times d`$ positive semidefinite matrices, whose normalized (i.e. unit-trace) members form the convex set of density matrices for the system, while the ambient linear space $`V`$ is the space of $`d\times d`$ Hermitian matrices. We shall often use the abbreviation “PSD” for “positive semidefinite.”
3.4 Definition Let $`V,W`$ be finite-dimensional real linear spaces equipped with cones $`CV`$, $`DW`$, and distinguished linear functionals $`\lambda C^{}`$, $`\stackrel{~}{\lambda }D^{}`$ that separate $`C,D`$ from $`C,D`$ respectively. Let $`\pi :VW`$ be a linear map that takes $`C`$ onto $`D`$ (that is, $`\pi (C)=D`$), and maps the affine plane $`L_\lambda :=\{xV:\lambda (x)=1\}`$ onto the plane $`M_{\stackrel{~}{\lambda }}:=\{yW:\stackrel{~}{\lambda }(y)=1\}`$. An element (“state”) in $`\widehat{C}:=L_\lambda C`$ is called generalized unentangled (GUE) relative to $`D`$ if it is in the closure of the convex hull of the set of extreme points $`x`$ of $`\widehat{C}`$ whose images $`\pi (x)`$ are extreme in $`\widehat{D}:=DM_{\stackrel{~}{\lambda }}`$.
3.5 Definition We will call a pair of linear spaces $`V,W`$ equipped with distinguished cones $`C,D`$, functionals $`\lambda ,\stackrel{~}{\lambda }`$, and a map $`\pi `$, satisfying the conditions in the above definition, a cone-pair. As noted above, we write $`\widehat{C},\widehat{D}`$ for the normalized subsets of $`C,D`$, i.e. for $`\{xC:\lambda (x)=1\}`$ and $`\{xD:\stackrel{~}{\lambda }(x)=1\}`$. We will also sometimes call $`\lambda ,\stackrel{~}{\lambda }`$ the traces on their respective cones, so that the condition on $`\pi `$ above may be called trace-preservation. That is, with the usual physics terminology that extremal states are “pure” and nonextremal ones “mixed,” unentangled pure states of $`\widehat{C}`$ are those whose “reduced” states (images under $`\pi `$) are pure, and the notion extends to mixed states as in standard entanglement theory: unentangled mixed states in $`\widehat{C}`$ are those expressible as convex combinations of unentangled pure states (or limits of such combinations, though the latter is unnecessary in finite dimension).
It is easy to see that the motivating example of ordinary bipartite entanglement is a special case of this definition. Here, $`C`$ is the cone of PSD operators on some tensor product $`AB`$ of finite-dimensional Hilbert spaces, while $`D`$ is the direct product of the cones of PSD operators on $`A`$ and on $`B`$ (intuitively, it is just the cone of all ordered pairs whose first member is a positive operator on $`A`$ and whose second is one on $`B`$). $`\lambda `$ is the trace. $`\pi `$ is the map that takes an operator on $`AB`$ to the ordered pair of its “marginal” or “reduced” operators (“partial traces”) on $`A`$ and $`B`$. Similarly, standard multipartite entanglement is a special case of GE. So we may view the GUE definition (in particular condition (a) of Definition 3.3 below) as based on extending the long-standing observation that for ordinary multipartite finite-dimensional quantum systems, a pure state is entangled if and only if at least one of its reduced density matrices is mixed.
It is perhaps mathematically more natural to define the unnormalized unentangled states of $`C`$ relative to $`D`$, omitting all mention of $`\lambda ,\stackrel{~}{\lambda }`$, and the normalization-preservation requirement on $`\pi `$. That is:
3.6 Definition Let $`C,D`$ be cones in finite-dimensional real linear spaces $`V,W`$ respectively, and let $`\pi :VW`$ map $`C`$ onto $`D`$. $`xC`$ is generalized unentangled (relative to $`D,\pi `$) if either (a) $`x`$ belongs to an extreme ray of $`C`$, and $`\pi (x)`$ belongs to an extreme ray of $`D`$, or (b) $`x`$ is a positive linear combination of states satisfying (a), or a limit of such combinations.
It is easy to verify that the unnormalized GUE states are a (possibly non-generating, but otherwise regular) cone in $`V`$. If one introduces the notion of normalization in $`C`$ via a functional $`\lambda `$, it is also easily verified that the normalized GUE states of Definition 3 are precisely the intersection of this cone with the normalization plane. (It is straightforward to introduce a normalization plane, and associated functional $`\stackrel{~}{\lambda }`$ on $`W`$, if desired, as the image of $`L_\lambda `$ under $`\pi `$.)
Barnum *et al.* (2003b), and especially Barnum *et al.* (2003a), stressed applications in which the reduced state-set is obtained by selecting a distinguished subspace of the observables (Hermitian operators) on some quantum system. The reduced state-set is then the set of linear functionals (equivalently, consistent lists of expectation values for the distinguished observables) on this subspace of the space of all observables, that are induced by normalized quantum states<sup>2</sup><sup>2</sup>2It is worth noting that beyond the setting of standard quantum entanglement this is not in general a vacuous requirement: there can be normalized linear functionals on the reduced state set that are not obtainable by restriction from a quantum state on the set of all observables. Although all normalized functionals on the distinguished observables can be extended in many ways to normalized functionals on the full set, in some cases not all can be extended to positive functionals.. We dub this class of cone-pairs the distinguished quantum observables setting. Even in the more general cones setting, there is a natural notion of observables, and Definition 3 can be interpreted as restriction of the states to a subspace of the observables. To show this we employ a formalism of states, measurements, and observables that, in many variants, is frequently used as a touchstone of “operational” approaches to theories in the abstract <sup>3</sup><sup>3</sup>3By an “operational theory,” we mean one in which a theory describes various measurements or operations one can perform on systems of the type described by the theory, and specifies a set of possible “states,” each of which determines the probabilities for the outcomes of all possible measurements, when the system is in that state..
We view $`V^{}`$ as a space of real-valued observables. For $`xV^{}`$ and $`\eta \widehat{C}`$, we interpret $`x[\eta ]`$ as the expectation value of observable $`x`$ in state $`\eta `$. We view $`V`$ as the dual of $`V^{}`$ in such a way that $`x[\eta ]=\eta [x]`$ for all $`xV^{},\eta V`$. But what guarantee do we have that these expectation values behave in a reasonable way, as observables in an operational theory should? That is, can we view the expectation value $`\eta (x)`$ of an observable $`x`$ in a state $`\eta `$ as the expected value of some quantity being measured? By this we mean that $`x`$ is associated with a quantity that takes different values depending on the outcome of the measurement, and the state determines the expectation value by determining probabilities for the different outcomes of the measurement, such that the value $`\eta (x)`$ is indeed the expectation value of the outcome-dependent quantity, calculated according to the probabilities assigned to the outcomes by the state.
We will only sketch the answer to this question; more details may be found in many places (though accompanied by additional concepts and formalism), notably Beltrametti and Bugajski (1997). In the structure we have described, of state-space and dual observable space, we are able to find a special class of observables, the “decision effects,” whose expectation value may be viewed as the probability of a measurement outcome. These “effects” are the elements of the initial interval $`:=[0,\lambda ]C^{}`$, i.e. the set of $`xC^{}`$ satisfying $`\lambda _C^{}x`$. A (finite) resolution of $`\lambda `$ is a set of effects $`x_i`$ such that $`_ix_i=\lambda `$. For normalized states $`\omega `$, it follows that $`\omega (x_i)0`$ and $`_i\omega (x_i)=1`$, so the values $`\omega (x_i)`$ may be viewed as probabilities of measurement outcomes, with a resolution of $`\lambda `$ representing the mutually exclusive and exhaustive outcomes of some measurement. Then it can be shown that for any observable $`AV^{}`$, a resolution $``$ of $`\lambda `$ and an assignment of real values $`v(x_i)`$ to the outcomes $`x_i`$ in $``$ can be found, such that for all normalized states $`\omega `$, $`\omega (A)=_i\omega (x_i)v(x_i)`$. For example, this is a consequence of (i) of Theorem 1 in Beltrametti and Bugajski (1997). In general the converse does not hold, giving rise to a generalization of observables sometimes known as stochastic observables for which not only does the analogous statement (which is (i) of Theorem 1 of Beltrametti and Bugajski (1997) where stochastic observables are just called observables) hold, but so does the converse of this analogue. The relation between the convex and the effect-algebras approach has been treated in various places (and aspects of it appear in some contexts, e.g. Ludwig (1983), even earlier than the formal notion of effect algebra). Some references are Gudder and Pulmannová (1998), Gudder *et. al.* (1999), Gudder (1999), and the book DallaChiara (2004) (especially Ch. 6). Barnum (2003a) explores the relation between probabilistic operational theories and “weak effect algebras,” as well as related more dynamical objects termed operation algebras, but without explicit consideration of observables. Bennett and Foulis (1997), Foulis*et al.* (1998), and Foulis (2000) address very closely related representational issues but without the constraint of convexity. The relations between convex and general effect-algebras and their representations are discussed in Foulis2005a.
We now show that our formalism of maps $`\pi `$ onto cones $`D`$ is equivalent to restriction to a subspace of observables.
3.7 Theorem
I) (“Observable restriction implies cone-pair”). Let $`C`$ be a cone in $`V`$, and let $`\lambda V^{}`$ separate $`V`$ from $`V`$ (as in Def. 3), and let $`W^{}`$ be a subspace of $`V^{}`$, containing $`\lambda `$. For $`\eta V`$, define $`\eta :W^{}`$ as the restriction of $`\eta `$ to the subspace $`W^{}`$, i.e. $`\eta (x)=\eta (x)`$ for $`xW^{}`$ and otherwise $`\eta (x)`$ is undefined. Thus $`\eta (W^{})^{}=:W`$. Define $`D=\{\eta :\eta C\}`$, $`M_\lambda =\{yW:\lambda (y)=1\}`$. Define $`\pi `$ as the restriction map $`\pi :=:VW,\eta \eta `$. Then $`V,W,C,D,\lambda ,\stackrel{~}{\lambda }(:=\lambda ),\pi `$ form a cone-pair in the sense of Definition 3. That is, $`D`$ is a cone in $`W`$, $`\pi (C)=D`$, and the image under $``$ of the plane $`L_\lambda \{\eta V:\lambda (\eta )=1\}`$ is a translation of a plane separating $`D`$ from $`D`$.
II) (“Cone-pair implies observable restriction”). Let $`V,W,C,D,\lambda ,\stackrel{~}{\lambda },\pi `$ be a cone-pair. Then there exists an injection (one-to-one map) $`\tau :W^{}V^{}`$, taking $`\stackrel{~}{\lambda }`$ to $`\lambda `$, such that $`\pi `$ is the map from $`V`$ to $`W`$ that takes $`x`$ to the function $`x_W^{}`$. Here $`x_W^{}`$ defined as the linear functional on $`W^{}`$ whose value on $`aW^{}`$ is the value of $`x`$’s restriction to $`\tau (W^{})`$ on $`\tau (a)`$.
Remark concerning I: The restriction that the subspace $`W^{}`$ contain $`\lambda `$ is hardly objectionable from an operational point of view. $`\lambda `$’s expectation value is just the normalization constant, and is independent of which normalized state has been prepared. Therefore it can be measured without any resources, and there is no point in claiming that omitting it could represent a physically significant restriction on the means available to observe or manipulate a system.
Remark concerning II: The definition of $``$ in part I of the theorem involved a subspace $`W^{}`$ of $`V^{}`$; in part II we have defined $`W^{}`$ abstractly rather than as a subspace of $`V^{}`$, so it is $`\tau (W^{})`$, which is isomorphic to $`W^{}`$ but is a subspace of $`V^{}`$, to which we restrict states in defining $``$. (Of course, $`W^{}`$ itself is a subspace of $`V^{}`$ according to the category-theoretic definition of subspace.)
Proof: To prove part I, we must show that $`D`$ is a cone in $`W`$, and $`\lambda `$ separates it from $`D`$. It is easy to verify linearity of $`\pi `$ from the definition, and in finite dimensions, it is also easy to verify that linear maps from one vector space onto another (such as $``$) take cones to cones. For all $`x\dot{C}=:C\{0\}`$, $`\lambda [x]>0`$. But $`\lambda [x]=x[\lambda ]`$ by duality, and by the definition of $``$ and the fact that $`\lambda W^{}`$, $`x[\lambda ]=x[\lambda ]\lambda [x]`$, so $`\lambda [x]>0`$ for all $`x\dot{C}`$, i.e. (since $``$ maps $`\dot{C}`$ onto $`\dot{D}`$), $`\lambda [y]>0`$ for all $`y\dot{D}`$. That is, $`\lambda `$ separates $`D`$ from $`D`$.
To prove part II, let $`\tau `$ be $`\pi ^{}`$. That is, for all $`xW^{}`$, $`\eta V,\tau (x)[\eta ]=x[\pi (\eta )]`$. By duality, this gives $`\eta [\tau (x)]=\pi (\eta )[x]`$. Since, by Proposition 2, $`\tau `$ is an injection, this last equation determines $`\pi (\eta )`$ to be essentially $`\eta _{\tau (W^{})}`$, as desired. The “essentially” refers to the fact that $`\pi (\eta )`$ is actually the pullback along $`\tau `$ of this restriction; the two are the same function only if one identifies $`W^{}`$ with its image under $`\tau `$. $`\tau `$, in other words, tells us how $`W^{}`$ can be identified with a subspace of the full space $`V^{}`$ of observables, in such a way that $`\pi (\eta )`$ becomes identified with the restriction of $`\eta `$ to $`W^{}`$. $`\mathrm{}`$
3.8 Proposition In a cone-pair, $`\pi `$ has the property that for $`x\mathrm{Extr}\widehat{\mathrm{D}}`$, the set $`\pi ^1(x)`$ is convex, compact, and closed, and its extremal elements are extremal in $`\widehat{C}`$.
Proof: Convexity is immediate: if $`y_1,y_2C`$, $`\pi (y_1)=x`$ and $`\pi (y_2)=x`$, then $`\lambda y_1+(1\lambda )y_2C`$ by convexity of $`C`$, and by linearity of $`\pi `$, $`\pi (\lambda y_1+(1\lambda )y_2)=\lambda (\pi (y_1))+(1\lambda )\pi (y_2)=x`$. Closedness of $`\pi ^1(x)`$ in the Euclidean metric topology follows from the fact that $`\pi `$, being a function from a finite-dimensional inner product space to a finite-dimensional normed space, is continuous (cf. e.g. Young, N., (1988), Exercise 7.3), and the preimage of a closed set under a continuous function is closed (cf. e.g. Kripke (1968), Corollary IV.C.4). Since finite intersections of closed sets are closed, $`C\pi ^1(x)`$ is closed as well. Compactness follows from the fact that $`\widehat{C}`$ is compact (cf. e.g. Barvinok (2002)) hence a compact metric space, and a closed subset of a compact metric space is compact (Kripke (1968), Corollary VII.A.11).
Now let $`x\mathrm{Extr}\widehat{\mathrm{D}}`$, and let $`y\pi ^1(x)C`$ not be extremal in $`\widehat{C}`$. We need to show that such a $`y`$ is not extremal in $`\pi ^1(x)`$ either. $`y\mathrm{Extr}\widehat{\mathrm{C}}`$ means there are $`y_1,y_2\widehat{C}`$ with $`y_1y_2`$, $`y=\lambda y_1+(1\lambda )y_2`$. By linearity of $`\pi `$, $`x\pi (y)=\lambda \pi (y_1)+(1\lambda )\pi (y_2)`$; since $`x\mathrm{Extr}(\widehat{\mathrm{D}})`$, $`\pi (y_1)=\pi (y_2)=x`$. Hence $`y_1,y_2\pi ^1(x)`$, so $`y\mathrm{Extr}(\pi ^1(\mathrm{x})\mathrm{C}`$. $`\mathrm{}`$
In important classes of examples, a stronger property holds:
3.9 Definition A cone-pair including $`C,D,\lambda ,\pi `$ is said to have the unique preimage property (UPIP) if $`x\mathrm{Extr}\widehat{\mathrm{D}}`$ implies that $`\pi ^1(x)`$ consists of a single element (which must therefore be extremal).
Equivalently (because of Prop. 3), extremal reduced states have only extremal preimages.
3.10 Problem Find nontrivial necessary and/or sufficient conditions (some are given below, but others almost certainly exist) for cone-pairs $`C,D,\pi `$ to have the UPIP.
Finally, note that the converse of the UPIP follows from Proposition 3: If $`\pi ^1(x)`$ is unique, then it must be extremal, and $`x`$ must be extremal as well.
## 4 GENERALIZED ENTANGLEMENT IN SPECIAL CLASSES OF CONES
We now formally define several “settings” in which to study GE; these are special classes of cone-pairs, physically and/or mathematically motivated.
4.11 Definition
* Distinguished quantum observables setting, defined above. An equivalent formulation is the Hermitian-closed (aka $``$-closed) operator subspace setting, in which the distinguished observable subspace is the Hermitian operators belonging to a $``$-closed subspace, containing the identity operator, of the complex vector space of all linear operators on a quantum system.
* Lie-algebraic setting. Here, $`C`$ is the cone of positive Hermitian operators on a (finite-dimensional) Hilbert space carrying a Hermitian-closed Lie algebra $`𝔤`$ (playing the role of $`W^{}`$) of Hermitian operators (with Lie bracket $`[X,Y]:=i(XYYX)`$, and containing the identity operator) and $`D`$ the cone (in $`(W^{})^{}=:W`$) of linear functionals on $`𝔤`$ induced from positive Hermitian elements of $`C`$ by restriction to $`W^{}`$.
* Associative algebraic setting. Here, the distinguished observables are the Hermitian elements of some associative subalgebra of the associative algebra of all operators on a quantum system.
By construction, the Lie-algebraic and associative algebraic settings are special cases of the distinguished quantum observables case. As noted in Barnum *et al.* (2003b), since the Lie-algebraic setting was defined to involve finite-dimensional $``$-closed matrix representations, the Lie algebras involved are necessarily reductive i.e., the direct product<sup>4</sup><sup>4</sup>4As Lie algebras; the induced direct product of the algebras considered as vector spaces (i.e. without their Lie bracket structure) is also a vector space direct sum. of a semisimple and an Abelian part.
A distinction that can be nontrivially made within all the settings in the above list is between those in which the distinguished observables act irreducibly, and those in which there is a nontrivial subspace invariant under the action of all observables.
4.12 Proposition In the $``$-closed operator subspace setting, the distinguished subspace has a basis of Hermitian operators that is orthonormal in the trace inner product $`A,B=\mathrm{tr}AB`$.
Because of this proposition, we may construct an orthogonal projection operator (some would call it a superoperator) $`\mathrm{\Pi }_S`$, acting on the space of Hermitian operators by projecting into the subspace of distinguished observables. We can also use such a basis to define a measure of entanglement for pure states, the relative purity (although the name may be slightly misleading, for reasons we will explain).
4.13 Definition Let $`\omega `$ be a state on a $``$-closed set $`S`$ of quantum observables. The purity $`P(\omega )`$ of a state $`\omega `$ is defined by letting $`X_\alpha `$ be an orthonormal (in trace inner product) basis of $`S`$. Then
$`P(\omega ):={\displaystyle \underset{\alpha }{}}(\omega (X_\alpha ))^2.`$ (2)
Note that any state $`\omega `$ on the full operator space corresponds uniquely to a density operator $`\rho _\omega `$, defined by the condition $`\mathrm{tr}(\rho _\omega X)=\omega (X)`$ for all observables $`X`$.
Closely related to the above purity is the relative purity of a pure state $`|\psi `$ of the overall quantum system; this is defined equal to the purity of the state it induces on $`S`$, or equivalently, with $`X_\alpha `$ as above,
$`P_S(|\psi ):={\displaystyle \underset{\alpha S}{}}|\psi |X_\alpha |\psi |^2.`$ (3)
In fact, this definition could be straightforwardly extended to mixed states $`\omega `$ on the full Hilbert space, as
$$P_S(\omega ):=\underset{\alpha S}{}|\mathrm{tr}\omega X_\alpha |^2.$$
(4)
However, a requirement for entanglement measures is convexity (Vidal, 2000), and the above extension lacks this as well as other desirable properties. We will generally extend pure-state entanglement measures $`\mu `$ to mixed states via the convex hull (often called convex roof) construction (cf. e.g. Uhlmann (1998); Bennett *et al.* (1996a)) standard in ordinary entanglement theory: the value of the measure $`\mu `$ on a mixed state $`\omega `$ is the infimum, over convex decompositions $`\omega =_ip_i\pi _i`$ of $`\omega `$ into pure states $`\pi _i`$, of the average value of the pure-state measure, that is, of $`_ip_i\mu (\pi _i)`$. This is convex by construction.
Defining $`\mathrm{\Pi }_S`$ as the projection superoperator onto the operator subspace $`S`$, it is easily verified that
$`P_S(\omega ):={\displaystyle \underset{\alpha }{}}|\mathrm{tr}\mathrm{\Pi }_S(\rho _\omega )X_\alpha |^2\mathrm{tr}[\mathrm{\Pi }_S(\rho _\omega )^2].`$ (5)
For any density operator $`\rho `$, we call $`\mathrm{\Pi }_S(\rho )`$ the associated reduced density operator; note that it need not be a positive operator on the full state space (although it is in the standard multipartite case). This is not problematic because for any PSD element $`R`$ of the distinguished observable space, $`\mathrm{tr}\mathrm{\Pi }_S(\rho )R0`$, of course.
The following proposition is immediate from Theorem 14 of Barnum *et al.* (2003b).
4.14 Proposition In the irreducible Lie-algebraic setting, pure states with maximal relative purity are generalized unentangled.
The converse is not true in general. Also, the analogue of Prop. 4 for the general Lie-algebraic setting (allowing reducible representations) can be shown by example to be false.
Another situation in which maximal relative purity implies generalized unentanglement is embodied in the following.
4.15 Proposition In the $``$-closed operator subspace setting states with unit relative purity have unique preimages, and are therefore generalized unentangled.
Proof: A necessary and sufficient condition for a normalized state $`\omega `$ on the space of all observables to be pure is $`\mathrm{tr}(\rho _\omega ^2)=1`$. (Henceforth we suppress the $`\omega `$-dependence of $`\rho `$.) Letting $`X_\alpha `$ be an orthonormal basis for the space of all observables such that a subset (denoted by the letter $`\beta `$ for the index) indexes the distinguished subspace $`S`$, with another subset (indexed by $`\gamma `$) indexing $`S^{}`$, and writing $`X_\alpha `$ for $`\mathrm{tr}\rho X_\alpha `$, we have $`\rho =_\alpha X_\alpha X_\alpha `$. From this and orthonormality of the $`X_\alpha `$ it is easy to see that $`\mathrm{tr}(\rho ^2)=_\alpha X_\alpha ^2`$. $`P_S(\rho )_{\beta S}X_\beta ^2`$; since extremal overall states have $`\mathrm{tr}(\rho ^2)=1`$, $`P_S(\rho )`$ for a pure state $`\rho `$ can never be greater than $`1`$, since it is a sum of a subset of the positive quantities $`X_\alpha ^2`$ which sum to $`1`$. Let $`X`$ have unit relative purity, i.e. $`_{\beta S}X_\beta ^2=1`$. This implies $`_{\gamma S^{}}Y_\gamma ^2=0,`$ which requires $`X_\gamma =0`$ for all $`\gamma S^{}`$. Thus, $`P_S(\rho )`$ has a unique preimage, namely itself, so $`\rho =P_S(\rho )`$. If $`P_S(\rho )`$ did not induce an extremal state in the convex set of reduced states, it would be a convex combination of distinct operators $`\rho _1`$ and $`\rho _2`$ inducing distinct reduced states; these would have distinct preimages, but the convex combination of these preimages be $`P_S(\rho )\rho `$, violating the assumption that $`\omega `$ is pure. $`\mathrm{}`$
What about states whose relative purity is maximal among all states, even when this maximal value is not unity? When the maximum is not unity, no pure state has an unchanged reduced density matrix: all pure state density matrices project to reduced “density matrices” that are either mixed, or not even PSD. Thus we cannot immediately conclude that $`_\beta X_\beta ^2=1`$, so we do not have $`X_\gamma =0`$ for all $`X_\gamma S^{}`$. If there is nevertheless a unique preimage, i.e. the $`X_\gamma `$ are uniquely determined by the $`X_\beta `$ (for $`\beta `$ indexing $`S`$), it must be a consequence of positive semidefiniteness of the initial state, since linear algebra alone gives no restrictions on the $`X_\gamma `$. However, because relative purity is a strictly convex function of the reduced density matrix, a state’s having maximal, even if not unit, relative purity, implies generalized unentanglement in the $``$-closed operator subspace framework. It does not, however, imply the other part of Proposition 4, that the reduced state has a unique preimage. Formally:
4.16 Proposition Let $`x\widehat{C}`$ be such that the relative purity of $`x`$ is no less than that of every other element of $`\widehat{C}`$. Then $`x`$ is generalized unentangled.
Proof: The relative purity of $`\omega `$ is just the Euclidean norm of $`\mathrm{\Pi }_S(\rho _\omega )`$ (with respect to the trace inner product). Suppose $`\omega `$ has maximal relative purity, i.e. $`|\mathrm{\Pi }_S(\rho _\sigma )||||\mathrm{\Pi }_S(\rho _\omega )||`$ for all $`\sigma \widehat{C}`$. Suppose there are $`\alpha ,\beta \widehat{D},\alpha \beta ,`$ such that $`\mathrm{\Pi }_S(\rho _\omega )=\mu \rho _\alpha +(1\mu )\rho _\omega `$. Then by the triangle inequality $`\mathrm{\Pi }_S(\rho _\omega )\mu \rho _\alpha +(1\mu )\rho _\beta =\mu \rho _\alpha +(1\mu )\rho _\beta `$. Since neither $`\rho _\alpha `$ nor $`\rho _\beta `$ is greater than $`\mathrm{\Pi }_S(\rho _\omega )`$, we must have $`\rho _\alpha =\rho _\omega =\mathrm{\Pi }_S(\rho _\omega )`$, so there is equality in the triangle inequality. That requires $`\mu \rho _\alpha `$ to be proportional to $`(1\mu )\rho _\beta `$, however, so that $`\rho _\alpha =\rho _\beta =\mathrm{\Pi }_s(\rho _\omega )`$. This shows that $`\mathrm{\Pi }_S(\rho _\omega )`$ is extremal in the set of reduced density operators corresponding to states in $`\widehat{D}`$. In other words, $`\omega `$ is generalized unentangled. $`\mathrm{}`$
It follows from the representation theory of associative algebras that the UPIP holds for the irreducible associative algebraic setting. The other case in which we know it holds is the irreducible semisimple Lie algebraic setting. In this setting, the observables consist of the Hermitian part (itself a real Lie algebra) of a complex Lie algebra represented faithfully and irreducibly by matrices acting on a finite-dimensional complex Hilbert space, and including the identity matrix $`I`$. Such Hermitian parts of irreducible matrix Lie algebras are precisely the real semisimple algebras possibly extended by the identity. The identity is relatively unimportant since all normalized states have the same value on it: the normalization condition is the affine plane $`\omega (I)=1`$, so the convex structure of the state space is entirely determined by the expectation values of the traceless operators. We introduce a bit more notation in order to state a result, proved in Barnum *et al.* (2003b), that includes this and other important facts about the irreducible Lie-algebraic case.
A *real* Lie algebra of Hermitian operators may be thought of as a distinguished family of Hamiltonians, which generate (via $`he^{ih}`$) a Lie group of unitary operators, describing a distinguished class of reversible quantum dynamics. More generally, we might want Lie-algebraically distinguished completely positive (CP) maps, $`\rho _iA_i\rho A_i^{}`$ describing open-system quantum dynamics. We call the operators $`A_i`$ the “Hellwig-Kraus” or “HK” operators, since they appear to have been introduced in Hellwig and Kraus (1969, 1970) (see also Kraus (1983); Choi (1975)). The HK operators for a given CP map are not unique, but this does not lead to nonuniqueness of any of the objects we define in terms of them below. A natural Lie-algebraic class of CP-maps has HK operators $`A_i`$ in the topological closure $`\overline{e^{𝔥_c1\mathrm{l}}}`$ of the Lie group generated by the *complex* Lie algebra $`𝔥_c1\mathrm{l}`$ <sup>5</sup><sup>5</sup>5 $`𝔥_c`$ is constructed by taking the complex linear span of a basis for $`𝔥`$. $`𝔥_c\text{1}\text{1}`$ guarantees inclusion of the identity operator 11.. Having HK operators in a group ensures closure under composition. Using $`𝔥_c1\mathrm{l}`$ allows non-unitary HK operators. Topological closure introduces singular operators such as projectors.
Define an $`𝔥`$-state to be a linear functional on a complex matrix Lie algebra $`𝔥`$ belonging to the convex set of such states induced by normalized quantum states on the full representation space. Complex-linearity ensures that the convex structure of such a state space is the same as that of the states induced by taking as the distinguished observables only the Hermitian elements (a real Lie algebra we denote $`\mathrm{Re}(𝔥)`$), which is the definition we used above for the Lie-algebraic setting.
4.17 Theorem Let $`𝔥`$ be a complex irreducible matrix Lie algebra, with $`𝔥_{}`$ its traceless (semisimple) part and $`\mathrm{Re}(𝔥)`$ its Hermitian part. The following are equivalent for a density matrix $`\rho `$ inducing the $`𝔥`$-state $`\lambda `$:
* $`\lambda `$ is a pure $`𝔥`$-state, that is, it is extremal in the convex set of normalized linear functionals on $`𝔥`$.
* $`\rho =|\psi \psi |`$ with $`|\psi `$ the unique ground state of some $`H`$ in $`\mathrm{Re}(𝔥)`$.
* $`\rho =|\psi \psi |`$ with $`|\psi `$ a minimum-weight vector (for some simple root system of some Cartan subalgebra) of $`𝔥_{}`$.
* $`\lambda `$ has maximum purity relative to the subspace $`\mathrm{Re}(𝔥)`$ of observables.
* $`\rho `$ is a one-dimensional projector in the topological closure of $`\overline{e^𝔥}`$.
4.18 Problem Does the implication from GUE to maximal relative purity, hold in other natural situations?
As already noted it is fairly easy to show by example that in the Lie-algebraic setting but without the assumption of irreducibility, the UPIP need not hold. A more general question suggests itself:
4.19 Problem In the $``$-closed operator subspace setting, does the UPIP hold whenever the distinguished operators act irreducibly?
## 5 ANALOGUES OF LOCAL MAPS
Our work on GE raises many questions arising from the closely related problems of finding natural generalizations or analogues of the notions of LOCC (Local Operations and Classical Communication) and of monotone entanglement measures (or entanglement monotones (Vidal, 2000)). The relation comes from requiring that a reasonable entanglement measure be nonincreasing under LOCC operations; if one found a natural generalization of this notion of LOCC to our more general settings, it would also be natural to look for measures of GE monotone under this generalization. Here, we briefly present some ideas from Barnum *et al.* (2003b) (with a few minor extensions) on how to generalize LOCC; that paper contains more on this topic and on GE measures. Some of the most fundamental questions remain open, so we will concentrate on sketching the situation in hopes of stimulating further work.
The semigroup of LOCC maps, introduced in Bennett *et al.* (1996b), and the preordering it induces on states according to whether or not a given state can be transformed to another by an LOCC operation are at the core of entanglement theory. LOCC maps are precisely those implementable by using CP quantum maps on the local subsystems, and classical communication, e.g. of “measurement results,” between systems. We now formalize this notion, beginning with the notion of explicitly decomposed map which, however, can apply to the general case, not just the quantum one. An explicitly decomposed trace-preserving map $`\{M_k\}_{kK}`$ is a set of maps $`M_k`$ that sum to a trace-preserving one $`M`$. The conditional composition of an explicitly decomposed map $`\{M_k\}_{kK}`$ with a set of explicitly decomposed maps $`N_k:=\{N_{nk}\}_{nN_k}`$ is the explicitly decomposed map $`\{N_{nk}M_k\}_{kK,nN_k}`$. We can view each $`M_k`$ as being associated with measurement outcome $`k`$, obtained (given a state $`\omega `$) with probability $`\mathrm{tr}M_k(\omega )`$, and leading to the state $`M_k(\omega )`$ when outcome $`k`$ is obtained. The conditional composition of $`\{M_k\}_{kK}`$ and $`\{N_{nk}\}_{nN_k}`$ can be implemented by first applying $`M`$ and then, given measurement outcome $`k`$, applying $`N_k`$. There are analogous definitions of explicitly decomposed maps and conditional composition without the trace-preservation condition.
In the usual quantum case, closing the set of one-party (aka unilocal) maps (for all parties) under conditional composition gives the LOCC maps. The semigroup generated by composition of unilocal explicitly decomposed maps having a single HK operator in their decomposition, is often known as SLOCC (for stochastic LOCC). SLOCC involves local quantum measurements and classical communication conditional on a single sequence of local measurement results, when each local measurement is performed in a manner that preserves all pure states (i.e., with a single HK operator for each outcome). Its mathematical structure is relatively simple, as the part generated by nonsingular HK operators is the trace-nonincreasing part of a representation of a product of various GL$`(d_i)`$, with the factors acting on local systems of dimension $`d_i`$ <sup>6</sup><sup>6</sup>6We are not certain if the full LOCC semigroup is the trace-nonincreasing part of the topological closure of this representation, but it seems a reasonable possibility..
When the distinguished observables form a semisimple Lie algebra $`𝔥`$, a natural multipartite structure can be exploited to generalize LOCC, as well as the larger, more tractable class of separable maps; see Barnum *et al.* (2003a, b). In generalizing LOCC to the convex setting, two aspects of LOCC must be considered: first, that it constrains maps to be completely positive; second, that it also constrains them to have certain locality properties.
A *positive* map of $`D`$ is a linear map $`A:VV`$ such that $`A(D)D`$. The map $`A`$ is *trace preserving* if $`\text{tr}(x)=\text{tr}(A(x))`$ for all $`x`$. This definition corresponds to positive, but not necessarily CP, maps in quantum settings. Without additional algebraic structure, it is not possible to define a unique “tensor product” of cones, as would be required to distinguish between positive and CP maps (Namioka and Phelps (1969); Wittstock (1974) (cited in Wilce (1992))). In a continuum of possible products of cones, there are two natural possibilities that are in a sense the two extremes. The first is the convex closure of the set of tensor-products of the cones’ vectors, which for the case of the product of two quantum systems’ unnormalized state spaces gives the separable states of the bipartite system. The second is to use the dual cone of the cone obtained by applying the first construction to the duals of the cones; in the quantum case, it gives the set of (unnormalized) states that are positive on product effects (this is isomorphic to the cone of positive but not CP operators between the state spaces, by the “Choi-Jamiolkowski” isomorphism between $`VV`$ and $`(V)`$). It is not clear how to pick out a natural case between these extremes in general without adding algebraic structure, except perhaps if the cones are self-dual with respect to non-degenerate inner products on the real vector spaces. In that case, one could pick a self-dual cone between the two constructions (which would give the usual state space of a bipartite system in the quantum case).
The family of positive maps of $`C`$ is closed under positive combinations and hence forms a cone. In the Lie-algebraic, or even the bipartite setting, the extreme points of this cone are not easy to characterize (see, for example, Wilce (1992), p. 1927, Gurvits (2002)). We seek generalizations of the notion of complete positivity to the cones setting. We might explicitly introduce a cone representing the “tensor product” extension of $`D`$ and require extendibility or “liftability” of the map to $`D`$. Another, perhaps more uniquely determined, approach might begin from the observation that the extreme points of the cone of completely positive maps are extremality preserving: for all extremal (belonging to an extreme ray) $`xD`$, $`A(x)`$ is extremal. However there are extremality preserving positive, not CP, maps. An example is partial transposition for density operators of qubits. In Barnum *et al.* (2003b) we explore how one might rule these out. There is also the question of why extremality preservation would be a natural physical or operational, as opposed to mathematical, requirement.
To try to generalize the notion of locality, we introduce the idea of liftability. We say that a positive map $`A`$ on $`D`$ can be lifted to $`C`$ if $`A`$ preserves the nullspace of $`\pi `$, or, equivalently, if there exists a positive map $`A^{}`$ on $`C`$ such that $`\pi (A(x))=A^{}(\pi (x))`$. In this case, we say that $`A^{}`$ is the lifting of $`A`$ to $`C`$.
In standard multipartite quantum entanglement, unilocal maps (ones that act nontrivially only on one factor) are liftable to the cone of local observables; they have a well-defined action there. But so are tensor product maps $`𝒜\mathrm{}𝒵`$, and in the case when some of the subsystems are of the same dimension, so are maps performing permutations among the isodimensional factors. To get LOCC we would need to rule out the latter two cases, leaving the unilocal maps; then one can generate a semigroup from the unilocal maps by conditional composition of explicitly decomposed trace-preserving maps. On the other hand, in the standard quantum case the semigroup of maps generated by conditional composition of maps liftable to the distinguished subcone might enjoy many of the same properties of the usual LOCC maps, so it may be worth study in the general setting.
5.20 Problem Is the semigroup generated by completely positive unilocal quantum maps and pairwise exchanges of isodimensional systems the full semigroup generated by conditional composition of liftable-to-local-observables explicitly decomposed maps?
Note that using liftability to define locality may be of some help in ruling out local non-completely positive maps, since all maps must be positive on the overall cone. It is especially helpful if the answer to Problem 5 is “yes.” When no subsystem has dimension greater than the square root of the overall dimension, it is then fully effective in imposing complete positivity, because for any local map $`M`$, complete positivity of $`M`$ is equivalent to positivity of the unilocal map $`\text{id}M`$ where the identity map id acts on a Hilbert space at least as large as the one $`M`$ acts on.
In the standard multipartite quantum case, the high degeneracy of unilocal operators can also be used to help distinguish them in a way not so directly dependent on explicit introduction of cones to represent individual systems—and similarly one can use spectral information about HK operators to characterize ones that act on the same single system, thereby characterizing LOCC in terms of conditional composition of explicitly decomposed maps whose HK operators together satisfy certain spectral conditions (Barnum *et al.*, 2003b). However, it is not clear how to abstract this to general cones. Perhaps something can be done with the facial structure of the cone $`D`$, or of the cone of positive maps on $`D`$ (or of other subcones of maps chosen as abstractions capturing aspects of complete positivity). A more in-depth investigation of dynamics generalizing LOCC thus remains as a challenging and many-faceted area for research, as does the investigation of measures of GE nonincreasing under such maps.
Acknowledgements We thank Manny Knill for valuable discussions and for collaboration on earlier work summarized and built upon in the present paper. Work at Los Alamos was supported by the US DOE through Los Alamos National Laboratory’s Laboratory Directed Research and Development (LDRD) program.
|
warning/0506/hep-th0506084.html
|
ar5iv
|
text
|
# References
The $`AdS`$ particle
Subir Ghosh
Physics and Applied Mathematics Unit,
Indian Statistical Institute,
203 B. T. Road, Calcutta 700108, India
Abstract:
In this note we have considered a relativistic Nambu-Goto model for a particle in $`AdS`$ metric. With appropriate gauge choice to fix the reparameterization invariance, we recover the previously discussed ”Exotic Oscillator”. The Snyder algebra and subsequently the $`\kappa `$-Minkowski spacetime are also derived. Lastly we comment on the impossibility of constructing a noncommutative spacetime in the context of open string where only a curved target space is introduced.
Introduction: It is now accepted in the High Energy Physics community that nonlocality in quantum field theory, or in a more fundamental way the fuzziness (or Non-Commutativity (NC)) in space(time), will be an integral part of present-day theories. Intuitive arguments that are used in avoiding the paradoxes one faces in trying to localize a spacetime point within the Planck length lead to a lower-bound in spacetime interval. This feature is also favored in the modifications of the Heisenberg uncertainty principle that one obtains in string scattering results (see for example ). It was first demonstrated by Snyder that Lorentz invariance and discretization requires an NC spacetime.
The NC spacetime has been revived by the seminal work of Seiberg and Witten who explicitly demonstrated the emergence of NC manifold in certain low energy limit of open strings moving in the background of a two form gauge field. In this instance, the NC spacetime is expressed by the Poisson bracket algebra (to be interpreted as commutators in the quantum analogue),
$$\{x^\mu ,x^\nu \}=\theta ^{\mu \nu },$$
(1)
where $`\theta ^{\mu \nu }`$ is a $`c`$-number constant. However, quantum field theories built on this spacetime do not enjoy Poincare invariance . On the other hand, this type of pathology can be avoided if one works with NC spacetime of the Snyder form or Lie algebraic form . In these examples the NC is operatorial in nature and thus it does not jeopardize the Lorentz invariance in relativistic models. The Lie algebra form of NC spacetime is typically given by,
$$\{x^\mu ,x^\nu \}=C_\lambda ^{\mu \nu }x^\lambda ,$$
(2)
where the structure constants $`C_\lambda ^{\mu \nu }`$ are constants.
In the present work, we will encounter both the Snyder and Lie algebraic forms of NC. In particular, we will concentrate on a restricted class of Lie algebra valued spacetime known as $`\kappa `$-Minkowski spacetime (or $`\kappa `$-spacetime in short), that is described by the basic Poisson structure,
$$\{x_i,t\}=kx_i,\{x_i,x_j\}=\{t,t\}=0.$$
(3)
In the above, $`x_i`$ and $`t`$ denote the space and time operators respectively. The present work is in continuation of our recent paper .
Some of the important works in $`\kappa `$-spacetime that discusses, among other things, construction of a quantum field theory in $`\kappa `$-spacetime, are provided in . Amelino-Camelia has pioneered an alternative approach to quantum gravity - ”the doubly special relativity” - in which two observer independent parameters, (the velocity of light and Planck’s constant), are present. It has been shown that $`\kappa `$-spacetime is a realization of the above. Furthermore, the mapping between $`\kappa `$-spacetime and Snyder spacetime , (the first example of an NC spacetime), shows the inter-relation between these models and ”two-time physics” , since the Snyder spacetime can be derived from two-time spaces in a particular gauge choice .
In we have proposed a physically motivated realization of the $`\kappa `$-spacetime in a quantum mechanical model. This is quite in tune with the connection between the noncommutativity arising in the Landau problem and that in the open string boundary with a background field . It is quite well known that for the planar, non-relativistic motion of a charged particle in a magnetic field (in the perpendicular direction), the particle configuration space becomes effectively noncommutative, if the dynamics is projected to the lowest Landau level. This is the celebrated Peirls substitution . Physically this is applicable in the limit of strong magnetic field . However, it has gained significance in recent times because of its (qualitative) analogy with the noncommutativity in open string boundary manifolds ($`D`$-branes), in the presence of a background two form gauge field . Unfortunately, a similar prototype of a simple physical system, picturizing the $`\kappa `$-spacetime was lacking. In our previous paper we have shed some light on this area. Specifically, in , we have put forward a non-relativistic quantum mechanical model that has an underlying phase space algebra, isomorphic to the $`\kappa `$-Minkowski one (3). In we have provided a Lagrangian of the model. As was mentioned in , (this point was noted in as well), the action has an uncanny similarity with the structure of the d$`S`$ or $`AdS`$ metric.
Let us put the present work in its proper perspective. The $`\kappa `$-spacetime requires the time to be operatorial in nature since it bears a non-trivial commutation relation with the space variables as given in (3). However, our model in was non-relativistic with conventional definition of time. To incorporate the operatorial behavior of time, we had to convert our model to a generalized one with reparameterization invariance and then exploit this symmetry to (gauge)fix time accordingly so that the $`\kappa `$-spacetime algebra emerged. This somewhat roundabout mechanism of has led us to the present work where we extend the non-relativistic particle model of to a relativistic, reparameterization invariant (Nambu-Goto) one. This allows us to fix the form of the time operator directly in the model. It is interesting to note that a similar type of time operator as in reduces the present model to the one considered in . We also recover a generalized form of the Snyder algebra , first given in . But more importantly, now the $`AdS`$ spacetime comes in to play directly and hence its connection to the $`\kappa `$-spacetime, via the Snyder algebra and exotic oscillator becomes clear. The advantage of working in a gauge invariant framework is that other convenient gauge choices, besides the one mentioned above, are indeed possible.
In an interesting alternative approach, it might be possible to obtain the $`\kappa `$-spacetime directly from quantum (or noncommutative) $`AdS`$ spacetime <sup>1</sup><sup>1</sup>1I thank Professor H. Steinacker for pointing this out. One can obtain a broad indication of this connection from the fact that the classical $`AdS`$-space can be embedded in a higher dimensional space with two-time metric and the $`\kappa `$-spacetime is directly related to the latter . At a more explicit level, since the $`\kappa `$-Poincare group can be obtained from the quantum $`AdS`$ group by contraction , it is possible the corresponding spaces are related as well.
(Non-relativistic) Mechanical model for $`\kappa `$-spacetime: It will be worthwhile to recapitulate briefly the model proposed in . We posited the Lagrangian,
$$L=\frac{m}{2}\stackrel{}{\dot{X}}^22mkc\eta (\stackrel{}{X}.\stackrel{}{\dot{X}})+c\eta ^2+2mk^2c^2\eta ^2\stackrel{}{X}^2,$$
(4)
where $`m`$ denotes the mass of the non-relativistic particle and $`k`$ and $`c`$ are constant parameters, and as shown below, $`\kappa `$ and $`c`$ induce noncommutativity in phase space related to $`\kappa `$-spacetime.
In the Hamiltonian constraint analysis, as formulated by Dirac , with the canonical phase space,
$$\{X_i,P_j\}=\delta _{ij},\{\eta ,\pi \}=1,$$
(5)
(where the sets $`(X_i,P_j)`$ and $`(\eta ,\pi )`$ are decoupled), there are two Second Class Constraints (SCC) <sup>2</sup><sup>2</sup>2In the Dirac terminology , First Class Constraints (FCC) commute with other constraints and generate gauge invariance. We will come across them in the present work later.
$$\chi _1\pi ,\chi _2\eta k(\stackrel{}{P}.\stackrel{}{X}).$$
(6)
Time independence of $`\chi _1`$ reproduces $`\chi _2`$, $`(\chi _2=\{\chi _1,H\})`$, with $`H`$ representing the Hamiltonian. SCCs require the use of Dirac Brackets (DB) defined by,
$$\{A,B\}_{DB}=\{A,B\}\{A,\chi _i\}\{\chi _i,\chi _j\}^1\{\chi _j,B\},$$
(7)
such that DB between an SCC and any operator vanishes. Note that $`\{\chi _i,\chi _j\}^1`$ indicates inverse of the Poisson bracket matrix $`\{\chi _i,\chi _j\}`$. A brief computation reveals the following non-canonical Dirac bracket algebra,
$$\{X_i,\eta \}=kX_i,\{P_i,\eta \}=kP_i,\{X_i,P_j\}=\delta _{ij}$$
(8)
Since we will always deal with DBs the subscript DB is dropped. Clearly $`\eta `$ behaves as time should in $`\kappa `$-spacetime, but a direct identification of $`\eta `$ with time is obviously not possible. This was done in by extending the model to a generally covariant. Incidentally, this way of exploiting a non-standard gauge condition to induce NC coordinates has been used in in case of constant spacetime noncommutativity.
One can eliminate $`\eta `$ and via an inverse Legendre transformation, obtain the following Lagrangian:
$$L=P_i\dot{X}_iH=\frac{m}{2}[(\dot{X}_i)^2(2m\kappa ^2c)\frac{(X_i\dot{X}_i)^2}{1+(2m\kappa ^2c)X_i^2}].$$
(9)
Depending on the sign of $`c`$, in the context of relativistic point particle to be demonstrated below, the above expression is generalized to d$`S`$ or $`AdS`$ spacetime.
Relativistic model for the $`AdS`$ particle: The form of the non-relativistic action in (9) in some sense forces up on us its following relativistic counterpart:
$$L=m[(\dot{X}.\dot{X})\kappa \frac{(X.\dot{X})^2}{1+\kappa (X.X)}]^{\frac{1}{2}}mA,$$
(10)
where $`(X.\dot{X})=X^\mu \dot{X}_\mu `$ etc.. Here we have considered a generic form with a single parameter $`\kappa `$ and $`\dot{X}^\mu =\frac{dX^\mu }{d\tau }`$. The above Nambu-Goto action clearly has the built-in $`AdS`$ metric since the action is
$$𝒜=𝑑\tau \sqrt{g_{\mu \nu }\dot{X}_\mu \dot{X}_\nu },g_{\mu \nu }=\eta _{\mu \nu }\frac{\kappa }{1+\kappa X^\lambda X_\lambda }X^\mu X^\nu .$$
(11)
The momentum is defined in the usual way,
$$P_\mu \frac{\delta L}{\delta \dot{X}^\mu }=\frac{m}{A}[\dot{X}_\mu \kappa \frac{(X.\dot{X})}{1+\kappa X^2}X_\mu ].$$
(12)
We directly obtain a modified mass-shell condition
$$(P.P)=m^2\kappa (P.X)^2,$$
(13)
which reduces to the conventional one for $`\kappa =0`$. The action has a $`\tau `$-parameterization symmetry that generates a zero Hamiltonian:
$$H=(P.\dot{X})L=0.$$
(14)
Note that the mass shell constraint in (13) represents the FCC We reexpress (4) in the form,
$$P_0=\frac{1}{1+\kappa X_0^2}[\kappa (\stackrel{}{P}.\stackrel{}{X})X_0\pm m(1+\kappa X_0^2)^{\frac{1}{2}}\{1+\frac{\stackrel{}{P}^2}{m^2}\frac{\kappa (\stackrel{}{P}.\stackrel{}{X})^2}{m^2(1+\kappa X_0^2)}\}^{\frac{1}{2}}].$$
(15)
We first demonstrate how the present system reduces to the non-relativistic model of . Let us consider the large $`m`$ or equivalently the non-relativistic limit,
$$P_0\frac{1}{1+\kappa X_0^2}[\kappa (\stackrel{}{P}.\stackrel{}{X})X_0\pm m(1+\kappa X_0^2)^{\frac{1}{2}}\{1+\frac{\stackrel{}{P}^2}{2m^2}\frac{\kappa (\stackrel{}{P}.\stackrel{}{X})^2}{2m^2(1+\kappa X_0^2)}\}].$$
(16)
Keeping terms up to $`O(\kappa )`$ we rewrite $`P_0`$ in the following suggestive way,
$$P_0m+\frac{\stackrel{}{P}^2}{2m}\frac{\kappa }{2m}(\stackrel{}{P}.\stackrel{}{X})^2+\kappa X_0[(\stackrel{}{P}.\stackrel{}{X})\frac{m}{2}X_0(1+\frac{\stackrel{}{P}^2}{2m^2})].$$
(17)
Thus, modulo the last term, we have obtained the expression for the Hamiltonian derived in . We can now exploit the reparameterization symmetry to introduce the gauge condition,
$$X_0=\frac{2(\stackrel{}{P}.\stackrel{}{X})}{m}(1+\frac{\stackrel{}{P}^2}{2m^2})^1.$$
(18)
Clearly the gauge fixed Hamiltonian reduces to that of .
However, the gauge constraint has rendered the FCC system to an SCC one with the SCC pair,
$$\varphi _1P_0\frac{\stackrel{}{P}^2}{2m}+\frac{\kappa }{2m}(\stackrel{}{P}.\stackrel{}{X})^2+O(\frac{1}{m^3}),\varphi _2X_0\frac{2(\stackrel{}{P}.\stackrel{}{X})}{m}+O(\frac{1}{m^3}).$$
(19)
They satisfy a non-zero Poisson bracket:
$$\{\varphi _1,\varphi _2\}=(1\frac{2\stackrel{}{P}^2}{m^2})\alpha .$$
(20)
The canonical phase space with $`\{P_\mu ,X^\nu \}=\eta _\mu ^\nu `$ gets modified to the Dirac brackets,
$$\{X_i,X_j\}=\frac{2}{\alpha m^2}(X_iP_jX_jP_i),\{P_i,P_j\}=0$$
$$\{X_i,P_j\}=\delta _{ij}+\frac{2}{\alpha m^2}P_iP_j.$$
(21)
The Dirac brackets with time operator $`X^0`$ turn out to be,
$$\{X_i,X_0\}=\frac{2X_i}{\alpha m},\{P_i,X_0\}=\frac{2P_i}{\alpha m}.$$
(22)
Time evolution is given by the Heisenberg equations of motion,
$$\dot{X}_i=\{X_i,P_0\}=\frac{1}{m}(P_i\kappa (\stackrel{}{P}.\stackrel{}{X})X_i),\dot{P}_i=\{P_i,P_0\}=\frac{\kappa }{m\alpha }(\stackrel{}{P}.\stackrel{}{X})P_i.$$
(23)
A further iteration reveals the dynamics:
$$\ddot{X}_i=\frac{2\kappa }{m}P_0X_i.$$
(24)
The other equation for $`\ddot{P}_i`$ is given below,
$$\ddot{P}_i=\frac{2\kappa }{m}(\frac{\stackrel{}{P}^2}{2m}+\frac{\kappa }{2m}(\stackrel{}{P}.\stackrel{}{X})^2)P_i.$$
(25)
Thus we have recovered the ”Exotic Oscillator” dynamics of . A redefinition of the variables, as given in , will lead to the $`\kappa `$-spacetime. The generalized form of the Snyder algebra, first given , is also recovered here in (21). Notice that in the approximations that we have considered, the NC algebra is $`\kappa `$-independent but $`\kappa `$ appears in the dynamics because otherwise we will have a free particle system.
As we are interested in the $`\kappa `$-spacetime, quite obviously the choice of time $`(X_0)`$ that is obtained from the form of gauge fixing is not canonical. Hence it might be interesting to compare the dynamics with this choice of time and the conventional ($`c`$-number parameter) one $`X_0=\tau `$ by considering an alternative choice of gauge gauge $`\varphi _2X_0\tau `$. In this case, the SCC system is,
$$\varphi _1P_0[\frac{\stackrel{}{P}^2}{2m}\frac{\kappa }{2m}(\stackrel{}{P}.\stackrel{}{X})^2+\kappa \tau [(\stackrel{}{P}.\stackrel{}{X})\frac{m}{2}\tau (1+\frac{\stackrel{}{P}^2}{2m^2})]],$$
$$\varphi _2X_0\tau ,$$
(26)
where $`\varphi _2`$ has been used in $`\varphi _1`$. Since now the Hamiltonian $`P_0`$ depends explicitly on time $`\tau `$, one has to consider the generalized form of Heisenberg equation for a generic operator $`A`$,
$$\frac{dA}{d\tau }=\frac{A}{\tau }+\{A,P_0\}.$$
(27)
It is clear that the canonical structure $`(\{X_i,P_j\}=\delta _{ij})`$ of phase space is not altered by this gauge choice. Thus in case of conventional time, the dynamics is governed by,
$$\ddot{X}_i=\frac{2\kappa }{m}P_0X_i+\kappa (X_i\frac{\tau P_i}{m}).$$
(28)
We find that the basic characteristics of the dynamics of the ”Exotic Oscillator” obtained in (24) remains intact, since vanishing of the last term defines the constant non-relativistic momentum. Perhaps this feature is not so surprising if we recall that in the ”Exotic Oscillator” dynamics was reproduced in conventional time with canonical phase space brackets.
Open String in curved background: The next step in generalization aught to be the de Sitter string that is string moving in a de Sitter background. However, instead of considering de Sitter metric in particular, we will consider a generic form of $`X^\mu `$-dependent metric $`G_{\mu \nu }(X)`$. In a previous work we have shown how the boundary conditions affect the Poisson bracket structures, considering the specific case of spacetime noncommutativity arising from the non-trivial boundary conditions occurring in the interacting system of open string and two-form background gauge field. As a concrete example, in we have shown the noncommutativity appearing in the open string boundary manifolds ($`D`$-branes) in the presence of a two-form background field can be rigorously obtained once the boundary conditions are properly taken in to account. Here we will show that a curved metric indeed modifies the boundary conditions but it does not induce noncommutativity.
The canonical phase space algebra
$$\{X^\mu (\sigma ),X^\nu (\sigma ^{})\}=\{\mathrm{\Pi }_\mu (\sigma ),\mathrm{\Pi }_\nu (\sigma ^{})\}=0,\{\mathrm{\Pi }_\mu (\sigma ),X^\nu (\sigma ^{})\}=g_\mu ^\nu \delta (\sigma \sigma ^{}),$$
is incompatible with the boundary conditions that one obtains for free open strings at the boundary and a modified form of $`\delta `$-function $`(\mathrm{\Delta }(\sigma \sigma ^{}))`$ appears, whose $`\sigma `$-derivative vanished at the string boundary . On the other hand, for open strings moving in the presence of two-form background field, the modified boundary conditions require a non-vanishing $`\{X_\mu (\sigma ),X_\nu (\sigma ^{})\}`$, indicating noncommutativity . This point is explained at the end.
The Polyakov action for the motion of an open string in a curved background $`G_{\mu \nu }(X)`$ is,
$$𝒮=\frac{1}{2}𝑑\sigma 𝑑\tau \sqrt{g}g^{ab}_aX^\mu _bX^\nu G_{\mu \nu },$$
(29)
where $`G_{\mu \nu }`$ is the curved target space metric and $`g_{ab}`$ is the induced metric. The momentum is defined below:
$$\mathrm{\Pi }_\mu =\frac{\delta 𝒮}{\delta _0X^\mu }=\sqrt{g}G_{\mu \nu }g^{0a}_aX^\nu =\sqrt{g}G_{\mu \nu }^0X^\nu $$
(30)
Variation of the induced metric $`g^{ab}`$ determines the energy-momentum tensor
$$T_{ab}=(_aX^\mu _bX^\nu +\frac{1}{2}g_{ab}g^{cd}_cX^\mu _dX^\nu )G_{\mu \nu }.$$
(31)
Vanishing of the above,
$$T_{ab}=0$$
(32)
provides the constraints of the theory that confirms reparameterization invariance. From the Hamiltonian constraint analysis point of view, the following combinations of constraints are useful:
$$T_{11}=\frac{1}{2}(g^0X^\mu ^0X^\nu _1X^\mu _1X^\nu )G_{\mu \nu },$$
$$\sqrt{g}T_{}^{0}{}_{1}{}^{}=\sqrt{g}^0X^\mu _1X^\nu G_{\mu \nu }.$$
(33)
We can reexpress the constraints in terms of phase space variables,
$$T_{11}\chi _1=\frac{1}{2}(\mathrm{\Pi }_\mu \mathrm{\Pi }_\nu G^{\mu \nu }+_1X^\mu _1X^\nu G_{\mu \nu }),$$
$$\sqrt{g}T_{}^{0}{}_{1}{}^{}\chi _2=\mathrm{\Pi }_\mu _1X^\mu .$$
(34)
The constraints $`\chi _1`$ and $`\chi _2`$ are FCC and satisfy the normal diffeomorphism algebra:
$$\{\psi _1(\sigma ),\psi _1(\sigma ^{})\}=4(\psi _2(\sigma )+\psi _2(\sigma ^{}))_\sigma \delta (\sigma \sigma ^{}),$$
$$\{\psi _2(\sigma ),\psi _1(\sigma ^{})\}=(\psi _1(\sigma )+\psi _1(\sigma ^{}))_\sigma \delta (\sigma \sigma ^{}),$$
$$\{\psi _2(\sigma ),\psi _2(\sigma ^{})\}=(\psi _2(\sigma )+\psi _2(\sigma ^{}))_\sigma \delta (\sigma \sigma ^{}).$$
(35)
Let us now return to the Lagrangian framework. The equation of motion and boundary condition arising from the action (29) are respectively,
$$_b[\sqrt{g}g^{ab}_aX^\mu G_{\mu \nu }]\frac{1}{2}\sqrt{g}g^{ab}_aX^\mu _bX^\lambda \frac{\delta G_{\mu \lambda }}{\delta X^\nu }=0,$$
(36)
$$\sqrt{g}g^{1a}(_aX^\mu )G_{\mu \nu }_{\sigma =0,\pi }=0,$$
(37)
where $`\sigma =0,\pi `$ are the string extremities. The boundary condition, expressed in terms of phase space variables, becomes,
$$_1X^\mu +\sqrt{g}g^{10}\mathrm{\Pi }^\mu _{\sigma =0,\pi }=0,$$
(38)
where some unimportant factors have been dropped. Notice that the diffeomorphism algebra (35) ensures that we can choose a gauge, in particular the conformal gauge, in which case $`g^{10}=0`$, and we are left with $`_1X^\mu _{\sigma =0,\pi }=0`$ as the boundary condition. This boundary condition is compatible with commutative spacetime. Comparing with our earlier work we establish that spacetime noncommutativity is not induced by only considering a curved spacetime.
Let us briefly elaborate on the last comment regarding and its connection to the present conclusion. The importance of obtaining the purported noncommutativity from different (in particular Hamiltonian) formalisms was stressed in the original work of Seiberg and Witten , since the concept of noncommutative spacetime was quite alien to the physics community. The first works in this connection , tried to establish that the noncommutative spacetime algebra should be interpreted as Dirac brackets provided one treats the boundary conditions as Second Class Constraints . However, these works contained various assumptions and computational steps that were ambiguous from the perspective of conventional Dirac analysis , of constrained systems. Subsequently it was realized , that the problem lies at the basic premises of : The boundary conditions are not to be treated as (field theoretic) constraints since the former apply only at the boundaries whereas the latter are valid for the whole region of phase space. This led us to our analysis where we generalized the earlier works . It was demonstrated in that for the case of open strings, basic phase space Poisson brackets are to be modified in order to be consistent with boundary conditions. In we explicitly showed that the boundary conditions for the interacting system of open string in an external two-form gauge field are consistent only with a noncommutative spacetime algebra. The counter intuitive idea of interpreting boundary conditions as constraints as in need not be introduced at all. This explains our conclusion that in the present case that the boundary conditions do not require a noncommutative spacetime.
|
warning/0506/hep-th0506026.html
|
ar5iv
|
text
|
# Metastable de Sitter vacua from critical scalar theory
## I Introduction
Recently, we showed that the fluctuations of the scalar field around the classical trajectory of massless $`\varphi ^4`$ model in four dimensional flat Euclidean spacetime is governed by a conformally coupled scalar field theory in four dimensional de Sitter background Solitons . This result is interesting due to its uniqueness. In four dimensions, in principle, one can consider two classes of critical (classically scale-free ) scalar field theories i.e. massless $`\varphi ^4`$ models on Euclidean (Minkowski) spacetime with $`g`$, the coupling constant, either positive or negative (we assume the potential $`V(\varphi )=\frac{g}{4}\varphi ^4`$). Although scalar theory with $`g>0`$ seems to be not physical as energy is not bounded from below but as is shown in Solitons in this case, the Euler-Lagrange equation of motion of the scalar theory on Euclidean space has an interesting classical solution say $`\varphi _0`$ with finite action $`S[\varphi _0]g^1`$. Interestingly, as far as we are considering real scalar field theories the model with the physical potential i.e. the case of $`g<0`$ has not such a solution, see Eq.(1). We should clarify that from our previous viewpoint Solitons , for $`g<0`$, one can still consider a solution like $`\varphi _0`$ obtained by an analytic continuation from $`g>0`$ to $`g<0`$ region. But such a solution is singular on the surface of a sphere which radius is proportional to $`g`$. Consequently the action $`S[\varphi _0]`$ is infinite and $`\varphi _0`$ can not be considered as a classical trajectory. If one ignores this problem and follows the calculations one obtains a conformally coupled scalar field on $`\text{AdS}_4`$ background. The reason why we do not follow our previous point of view turns back to our new machinery for constructing $`\varphi _0`$ from the first principles explained in section II. Of course similar singularities appear even for $`g>0`$, when one switches to Minkowski spacetime by Wick rotation $`x^0ix^0`$. But in this case ($`g>0`$) one can argue that the singularity is beyond the horizon of observers living in the corresponding de Sitter spacetime and consequently is safe. We do not study the case of Minkowski spacetime in this paper.
As is shown in Solitons the information geometry of the moduli of $`\varphi _0`$ is $`\text{AdS}_5`$ ($`\text{dS}_5`$) if $`g>0`$ ($`g<0`$) which resembles the information geometry of $`k=1`$ $`SU(2)`$ instantons Blau <sup>1</sup><sup>1</sup>1$`V(\varphi _0)`$ can be shown to be proportional to the SU(2) one-instanton density Blau .. The moduli here are the location of the center of $`\varphi _0`$ and its size and are consequences of the invariance of the action under rescaling and translation. The similarity between this solution and SU(2) instantons can be explained in terms of the ’t Hooft $`\varphi ^4`$ ansatz for instantons Thooft .
In this paper focusing on the case $`g>0`$ we show that $`\varphi _0`$ can be responsible for viewing a metastable de Sitter background. We first show that $`\varphi _0`$ is a metastable local minima of the action. Then we show that the weak coupling limit $`g{}_{}{}^{+}0`$ is equivalent to the classical ($`\mathrm{}0`$) limit therefore in that limit the partition function picks up contribution only around the classical trajectory $`\varphi _0`$. As we said before, fluctuations around $`\varphi _0`$ is governed by a scalar theory on a de Sitter background with a reversed Mexican potential. The hight and width of the barrier is proportional to $`R`$ and $`g^1`$. Here $`R`$ is the scalar curvature of the universe which can not be determined in our model but one can show that the lifetime of this local minima is proportional to $`e^{g^1}`$.
Summarizing these results we verify that the critical scalar theory can not be analytically continued (at least through $`g=0`$) from $`g<0`$ region to $`g>0`$ region. These results are strictly important from perturbation theory point of view. In the $`g{}_{}{}^{}0`$ limit, valid perturbations are around $`\varphi =0`$ in a flat Euclidean background but in the $`g{}_{}{}^{+}0`$ limit perturbations are around $`\varphi =0`$ in an Euclidean de Sitter background. More explicitly at $`g=0^{}`$, the theory is simply a free massless scalar theory on flat Euclidean space but at $`g=0^+`$ due to $`\mathrm{}`$ considerations the theory is a (free and stable) conformally coupled scalar theory on an Euclidean de Sitter background. See sections II and III.
But why we are interested in de Sitter background. In fact WMAP results WMAP combined with earlier cosmological observations shows that we are living in an accelerating universe. Constructing four dimensional de Sitter vacuum as a string theory (M-theory) solution has been a long standing challenge. In KKLT KKLT constructed a metastable de Sitter vacua of type IIB string theory by adding $`\overline{\text{D3}}`$-branes to the GKP GKP model of highly warped IIB compactifications with nontrivial NS and RR three-form fluxes. The mean lifetime of KKLT solution is $`10^{10^{120}}`$ years. KKLMMT constructed a model of inflation by adding mobile D3 branes to the KKLT solution KKLMMT . In such models the inflaton (the position of D3 brane deep inside the warped throat geometry) is a conformally coupled scalar in the effective four-dimensional geometry. Therefore the mass of such a scalar $`m_\varphi ^2`$ is close to $`2H^2=\xi R`$ where $`H^2`$ is the Hubble parameter, $`\xi =\frac{1}{6}`$ (in four dimension) is the conformal coupling constant and $`R`$ is the curvature of de Sitter space. As is uncovered in KKLMMT this does not meet the observational requirement $`m_\varphi ^210^2H^2`$ <sup>2</sup><sup>2</sup>2In D3/D7 model of inflation one does not encounter the $`m_\varphi ^2H^2`$ problem. But here in contrast to the model of D3$`/\overline{\text{D3}}`$ inflation in the highly warped throat, there is no natural mechanism for suppressing the contribution of cosmic strings formed at the end of inflation to the CMB anisotropy. See DHKLZ for a solution to this problem.. Similar considerations show that our model can not be a successful model for inflation since in the $`g{}_{}{}^{+}0`$ limit we also obtain a conformally coupled scalar theory on de Sitter background.
Another problem in our model is the existence of a continuum of de Sitter bubbles, given by the location of the center of $`\varphi _0`$ and its size. Naively this number is proportional to the volume of the $`\text{AdS}_5`$ moduli space. If $`\varphi _0`$ is responsible to observe, say, by optical methods a de Sitter geometry (see section IV), a natural question is to ask which bubble we live in. Studying the variation of action around the $`\varphi _0`$ solution, we have verified, by numerical calculations that smaller bubbles are more stable than larger ones. Consequently there is a transition: larger bubbles decay to smaller ones and probably finally there remains a gas of zero size bubbles. The mechanism of such a decays is not clear to us yet but its phenomenology, might be similar to that of the discretuum of possible de Sitter vacua in KKLT models fall .
As we said before our theory can not do any prediction about the size of our de Sitter universe or the nature of the scalar field but it predicts that the lifetime of this metastable vacua is $`\tau e^{g^1}`$. The model is interesting due to its uniqueness and symmetries which brings hopes to be constructible from a fundamental theory like string theory.
The organization of the paper is as follows. In section II, we study the critical scalar theory in $`D=4`$ Euclidean space and $`\varphi _0`$, the exact solution to the corresponding Euler-Lagrange equation of motion. Considering the weak coupling limit we show that the partition function only picks up contribution around $`\varphi _0`$. In section III, we show that the critical scalar theory considered as the action for (not essentially small) fluctuations around $`\varphi _0`$ is a scalar theory on $`D=4`$ de Sitter background. Consequently we verify that in the weak coupling limit the critical scalar theory of section IIis in fact a metastable scalar theory on de Sitter background with finite lifetime increasing exponentially as the coupling constant decreases. In section IV we generalize our model to scalar theory coupled to U(1) gauge field. Such a generalization is essential as it shows how by optical observations people living in a flat Euclidean space can view a de Sitter geometry for their universe.
## II Critical scalar theory in $`D=4`$ Euclidean space
In this section we study scalar field theories in four dimensional Euclidean space invariant under rescaling transformation. By rescaling we mean a coordinate transformation $`xx^{}=\lambda x`$, $`\lambda >0`$. Requiring the kinetic term of a scalar theory to be invariant under rescaling, one verifies that the scalar field should be transformed as $`\varphi (x)\varphi ^{}(x^{})=\lambda ^1\varphi (x)`$. A scale free theory by definition is a theory given by an action $`S`$ invariant under rescaling. In addition to the Kinetic term which variation is a total derivative, we search for polynomials $`V(\varphi )`$ in $`\varphi `$ as the potential term such that $`\delta V=0`$, up to total derivatives. Such polynomials exist only in three, four and six dimensions. In the case of our interest i.e. $`D=4`$, $`V(\varphi )=\frac{g}{4}\varphi ^4`$. Here $`g`$, the coupling constant is some arbitrary real constant which is by construction invariant under rescaling. Such a scalar model is called critical as it is scale-free and its correlation length is infinite. The corresponding Euler-Lagrange equation of motion is a nonlinear Laplace equation $`^2\varphi +g\varphi ^3=0`$. One can easily show that for $`g>0`$, a solution of the non-linear Laplace equation is Solitons ; Fubini ,
$$\varphi _0(x;\beta ,a^\mu )=\sqrt{\frac{8}{g}}\frac{\beta }{\beta ^2+(xa)^2},$$
(1)
where $`(xa)^2=\delta _{\mu \nu }(xa)^\mu (xa)^\nu .`$ $`\beta `$ and $`a^\mu `$ are undetermined parameters describing the the size and location of $`\varphi _0`$. These moduli are consequences of symmetries of the action i.e. invariance under rescaling and translation. The information geometry of the moduli space, given by Hitchin formula Hit
$$𝒢_{IJ}=\frac{1}{N}d^4x_0_I\left(\mathrm{log}_0\right)_J\left(\mathrm{log}_0\right),$$
(2)
is an Euclidean $`\text{AdS}_5`$ space Solitons ,
$$𝒢_{IJ}d\theta ^Id\theta ^J=\frac{1}{\beta ^2}\left(d\beta ^2+da^2\right).$$
(3)
$`I=1,\mathrm{},5`$ counts space directions of moduli space $`\theta ^I\{\beta ,a^\mu \}`$, $`N=\frac{4^3}{5}d^4x_0`$ is a normalization constant and $`_0=\frac{g}{4}\varphi _0^4`$ is the Lagrangian density calculated at $`\varphi =\varphi _0`$. The moduli $`a^\mu `$ are present since the action is invariant under transformation. The existence of $`\beta `$ is the result of invariance under rescaling. To see this let us consider scale-free fields i.e. those fields that satisfy the relation $`\delta _ϵ\varphi =0`$. Here $`\delta _ϵ\varphi (x)=\varphi ^{}(x)\varphi (x)`$ is the infinitesimal scale transformation given by $`\lambda =1+ϵ`$ for some infinitesimal $`ϵ`$. To this aim we first note that rescaling leaves the origin ($`x=0`$) invariant. Consequently $`\varphi (0)`$ is distinguished from the values of the field at the other points since $`\varphi (0)\varphi ^{}(0)=\lambda ^1\varphi (0)`$. Therefore, it is plausible to make the dependence of scalar fields on their values at the origin explicit and represent the rescaling transformation by $`\varphi (x;\varphi (0))\varphi ^{}(x;\varphi (0))=\lambda ^1\varphi (\lambda ^1x;\lambda \varphi (0))`$. Defining $`\beta ^1=\varphi (0)`$, one can show that $`\delta _ϵ\varphi (x)=ϵ(1+x^I_I)\varphi (x)`$, where $`x^I\{x^\mu ,\beta \}.`$ The $`SO(4)`$ invariant solutions of equation $`\delta _ϵ\varphi =0`$ satisfying the condition $`\varphi (0;\varphi (0))=\varphi (0)`$ are $`\varphi _k=c\beta ^1\left(\frac{\beta }{\sqrt{\beta ^2+x^2}}\right)^{k+2}`$ where $`c`$ is some arbitrary constant. It is easy to see that for $`c=\sqrt{\frac{8}{g}}`$, $`\varphi _0`$, among the others, is the solution of classical equation of motion.
Now it is time to show that $`\varphi _0`$ is a metastable local minima of the action. Since $`\varphi _0`$ is a solution of Euler-Lagrange equation of motion $`\delta S=0`$ it is a local extremum of the action<sup>3</sup><sup>3</sup>3Of course from $`\delta S=0`$ we can only conclude that $`\varphi _0`$ is a stationary point and not necessarily a local extremum. We continue by assuming that $`\varphi _0`$ is a local extremum. This assumption can be proved following the results of section III.. So it is enough to show that there are field variations $`\varphi _0\varphi _\eta =\varphi _0+ϵ\eta `$ for $`𝒞^1`$ functions $`\eta `$ vanishing as $`x\mathrm{}`$ such that $`\delta S=c_\eta ϵ^2+𝒪(ϵ^3)`$ for some real positive constant $`c_\eta `$. For simplicity one can assume $`\eta =\left(\frac{1}{1+x^2}\right)^n`$, $`g=8`$, $`b=1`$ and calculate $`\delta S=S[\varphi _\eta ]S[\varphi _0]`$ for some integers $`n`$. One recognizes that $`c_n>0`$ for $`n>5`$, though it is negative for $`0<n5`$. A good sign for metastability of the action at $`\varphi _0`$. Section III provides an exact proof for this claim. Another interesting observation is that bubbles with larger size are less stable than those with smaller size. This can be checked noting that the size of a bubble is proportional to $`\beta ^1`$. By repeating the above calculations one easily verifies that for example for $`b=3`$ $`c_n>0`$ even for $`n=3`$. Unfortunately without analytic data all these observations make only a rough picture of the phenomena which can not be used to make an exact statement.
### The weak coupling limit
To study the weak coupling limit of the theory one can scale $`g\frac{g}{\lambda ^2}`$ and consider the $`\lambda \mathrm{}`$ limit, while keeping fields $`\varphi `$ undistorted. To this aim we do the following transformations,
$$\begin{array}{c}x\lambda x,\hfill \\ \varphi \lambda ^1\varphi ,\hfill \\ gg,\hfill \end{array}\text{and}\begin{array}{c}xx,\hfill \\ \varphi \lambda \varphi ,\hfill \\ g\lambda ^2g,\hfill \end{array}$$
(4)
which results in the desired transformation:
$$\begin{array}{c}x\lambda x,\hfill \\ \varphi \varphi ,\hfill \\ g\lambda ^2g,\hfill \\ S\lambda ^2S,\hfill \end{array}$$
(5)
More explicitly due to invariance under rescaling we have,
$$D[\varphi ]e^{\frac{1}{\mathrm{}}S[\varphi ;\lambda ^2g]}=D[\varphi ]e^{\frac{\lambda ^2}{\mathrm{}}S[\varphi ,g]}.$$
(6)
To calculate the partition function one can instead of the transformation $`S\lambda ^2S`$ assume that $`SS`$ but $`\mathrm{}\lambda ^2\mathrm{}`$.
Consequently the $`g{}_{}{}^{+}0`$ limit is equivalent to $`\mathrm{}{}_{}{}^{+}0`$ limit <sup>4</sup><sup>4</sup>4Of course we are also blowing our universe as $`x\lambda x`$. Since the flat Euclidean space we considered is not compact it does not seems to cause any problem at this level. In the case $`n`$-point functions one should note that non-coincident points go to infinity with respect to each other. Therefore in the weak coupling limit the partition function picks up contribution only from trajectories close to $`\varphi _0`$. This key observation when we study coupling to $`U(1)`$ gauge field proves why people living in a flat Euclidean space with a critical scalar field at $`g=0^+`$ observe a de Sitter universe.
## III $`\varphi _0`$ as a de Sitter background
In this section we show that fluctuations around $`\varphi _0`$ are governed by a scalar theory on de Sitter background. This section is a review of the calculations made in Solitons . In order to study the fluctuations around $`\varphi _0`$ one should rewrite the action
$$S[\varphi ]=d^4x\left(\frac{1}{2}\delta ^{\mu \nu }_\mu \varphi _\nu \varphi \frac{g}{4}\varphi ^4\right),$$
(7)
in terms of new fields $`\stackrel{~}{\varphi }=\varphi \varphi _0`$. In this way one obtains a new action,
$$S[\varphi ]=S[\varphi _0]+S_{\text{free}}[\stackrel{~}{\varphi }]+S_{\text{int}}[\stackrel{~}{\varphi }],$$
(8)
where $`S[\varphi _0]=d^4x_0=\frac{8\pi ^2}{3g}`$, and
$$S_{\text{free}}[\stackrel{~}{\varphi }]=d^4x\left(\frac{1}{2}\delta ^{\mu \nu }_\mu \stackrel{~}{\varphi }_\nu \stackrel{~}{\varphi }+\frac{1}{2}M^2(x)\stackrel{~}{\varphi }^2\right),$$
(9)
in which,
$$M^2(x)=3g\varphi _0^2=24\frac{\beta ^2}{\left(\beta ^2+(xa)^2\right)^2}.$$
(10)
The mass dependent term can be interpreted as interaction with $`\varphi _0`$ background. Now recall that in general, by inserting $`\stackrel{~}{\varphi }=\sqrt{\mathrm{\Omega }}\overline{\varphi }`$ and $`\delta _{\mu \nu }=\mathrm{\Omega }^1g_{\mu \nu }`$ in the action $`S[\stackrel{~}{\varphi }]=d^4x\frac{1}{2}\delta ^{\mu \nu }_\mu \stackrel{~}{\varphi }_\nu \stackrel{~}{\varphi }`$, one obtains,
$$S[\stackrel{~}{\varphi }]=d^4x\sqrt{g}\left(\frac{1}{2}g^{\mu \nu }_\mu \overline{\varphi }_\nu \overline{\varphi }+\frac{1}{2}\xi R\overline{\varphi }^2\right),$$
(11)
i.e. a scalar theory on conformally flat background given by the metric $`g_{\mu \nu }`$ in which $`\mathrm{\Omega }>0`$ is an arbitrary $`𝒞^{\mathrm{}}`$ function. $`R`$ is the scalar curvature of the background and $`\xi =\frac{1}{6}`$ is the conformal coupling constant. For details see Ted or appendix C of Solitons . Thus, defining $`\overline{\varphi }=\mathrm{\Omega }^{\frac{1}{2}}\stackrel{~}{\varphi }`$, one can show that $`S_{\text{free}}[\stackrel{~}{\varphi }]`$ given in Eq.(9) is the action of the scalar field $`\overline{\varphi }`$ on some conformally flat background with metric $`g_{\mu \nu }=\mathrm{\Omega }\delta _{\mu \nu }`$:
$$S_{\text{free}}[\stackrel{~}{\varphi }]=d^4x\sqrt{\left|g\right|}\left(\frac{1}{2}g^{\mu \nu }_\mu \overline{\varphi }_\nu \overline{\varphi }+\frac{1}{2}(\xi R+m^2)\overline{\varphi }^2\right).$$
(12)
Here, $`m^2\mathrm{\Omega }=M^2(x)`$, where $`m^2`$ is the mass of $`\overline{\varphi }`$ (undetermined) and $`M^2(x)`$ is given in Eq.(10). This result is surprising as one can show that the Ricci tensor $`R_{\mu \nu }=\mathrm{\Lambda }g_{\mu \nu }`$, where $`\mathrm{\Lambda }=\frac{m^2}{2}>0`$ as far as $`\mathrm{\Omega }>0`$. Consequently $`\overline{\varphi }`$ lives in a four dimensional de Sitter space which scalar curvature $`R=2m^2`$. The interacting part of the action, $`S_{\text{int}}[\stackrel{~}{\varphi }]=d^4x\sqrt{\left|g_{\mu \nu }\right|}_{\text{int}}`$ is well-defined in terms of $`\overline{\varphi }`$ on the corresponding $`\text{dS}_4`$:
$$_{\text{int}}=g\sqrt{\frac{m^2}{3g}}\overline{\varphi }^3\frac{g}{4}\overline{\varphi }^4.$$
(13)
Interestingly after a shift of the scalar field $`\overline{\varphi }\overline{\varphi }\sqrt{\frac{m^2}{3g}}`$ the action (8) can be written in the $`\text{dS}_4`$ as follows:
$$S[\overline{\varphi }]=d^Dx\sqrt{\left|g\right|}\left(\frac{1}{2}g^{\mu \nu }_\mu \overline{\varphi }_\nu \overline{\varphi }+\frac{1}{2}(\xi R)\overline{\varphi }^2\frac{g}{4}\overline{\varphi }^4\right).$$
(14)
This is a scalar theory in a de Sitter background with reversed Mexican hat potential. $`\overline{\varphi }=0`$ corresponds to the local minima of the potential which distance to the center of the hill (the location of $`\varphi _0`$) is $`\sqrt{\frac{\xi R}{g}}`$. The hight of the hill is $`\frac{\xi ^2R^2}{4g}`$. The lifetime of the metastable vacua can be estimated using the WKB method: the transition rate $`\mathrm{\Gamma }`$ is
$$\mathrm{log}\mathrm{\Gamma }\mathrm{\Delta }S,\mathrm{\Delta }SV_4\frac{(\xi R)^2}{g}$$
(15)
in which $`V_4R^2`$ is the volume of the $`𝒮^4`$, the Euclidean de Sitter space. Consequently the lifetime $`\tau e^{g^1}`$.
## IV Critical scalar theory coupled to $`U(1)`$ gauge field
In this section we study complex critical scalar theory on Euclidean space coupled to $`U(1)`$ gauge field $`A_\mu `$,
$$S=d^4x\left(\left|D_\mu \varphi \right|^2\frac{g}{2}\left|\varphi \right|^4\right)+S_A,$$
(16)
where $`D_\mu =_\mu +ieA_\mu `$ is the covariant derivative and $`S_A=\frac{1}{4}F_{\mu \nu }F_{\rho \sigma }\delta ^{\rho \mu }\delta ^{\sigma \nu }`$. It is easy to verify that $`A_\mu =0`$ and $`\varphi =\varphi _0`$ is a solution of the Euler-Lagrange equation of motion. Inserting $`\stackrel{~}{\varphi }=\varphi \varphi _0`$ in the Eq.(16), one obtains
$$S=S[\varphi _0]+\stackrel{~}{S}[\stackrel{~}{\varphi }]+S_{int}+S_A,$$
(17)
where
$`S_{int}`$ $`=`$ $`{\displaystyle }d^4x\delta ^{\mu \nu }[(ieA_\mu (\stackrel{~}{\varphi }+\varphi _0)_\nu (\stackrel{~}{\varphi }^{}+\varphi _0)+c.c.)`$
$`+e^2A_\mu A_\nu |\stackrel{~}{\varphi }+\varphi _0|^2],`$
$`\stackrel{~}{S}[\stackrel{~}{\varphi }]`$ $`=`$ $`{\displaystyle d^4x\left(\frac{1}{2}\left|_\mu (\stackrel{~}{\varphi }+\varphi _0)\right|^2\frac{g}{4}\left|\stackrel{~}{\varphi }+\varphi _0\right|^4\right)}.`$ (18)
Inserting $`\overline{\varphi }=\mathrm{\Omega }^{\frac{1}{2}}\stackrel{~}{\varphi }`$ where $`\mathrm{\Omega }`$ is defined as before by the relation $`\varphi _0=\sqrt{\frac{m^2}{3g}\mathrm{\Omega }}`$ one obtains,
$`S_{int}`$ $`=`$ $`{\displaystyle }d^4x{\displaystyle \frac{1}{2}}\mathrm{\Omega }\delta ^{\mu \nu }[ieA_\mu (\overline{\varphi }+\sqrt{{\displaystyle \frac{m^2}{3g}}})_\nu \overline{\varphi }^{}+c.c.`$ (19)
$`+e^2A_\mu A_\nu |\overline{\varphi }+\sqrt{{\displaystyle \frac{m^2}{3g}}}|^2],`$
and
$`\stackrel{~}{S}[\stackrel{~}{\varphi }]`$ $`=`$ $`{\displaystyle }d^4x\left({\displaystyle \frac{1}{2}}\mathrm{\Omega }\right|_\mu \overline{\varphi }|^2{\displaystyle \frac{g}{4}}|\overline{\varphi }+\sqrt{{\displaystyle \frac{m^2}{3g}}}|^4`$ (20)
$`{\displaystyle \frac{1}{2}}\sqrt{\mathrm{\Omega }}^2\sqrt{\mathrm{\Omega }}|\overline{\varphi }+\sqrt{{\displaystyle \frac{m^2}{3g}}}|^2),`$
where $`^2=\delta ^{\mu \nu }_\mu _\nu `$. Defining $`g_{\mu \nu }=\mathrm{\Omega }\delta _{\mu \nu }`$ and noting that $`\sqrt{\mathrm{\Omega }}^2\sqrt{\mathrm{\Omega }}=\sqrt{g}\xi R`$, after an obvious shift $`\overline{\varphi }\overline{\varphi }\sqrt{\frac{m^2}{3g}}`$, one obtains,
$`S`$ $`=`$ $`{\displaystyle d^4x\sqrt{g}\left(\frac{1}{2}g^{\mu \nu }D_\mu \overline{\varphi }D_\nu \overline{\varphi }^{}+\frac{1}{2}\xi R\left|\overline{\varphi }\right|^2\frac{g}{4}\left|\overline{\varphi }\right|^4\right)}`$ (21)
$`+`$ $`S_A,`$
It is known that in $`D=4`$, $`S_A`$ is invariant under conformal transformation $`\delta _{\mu \nu }\mathrm{\Omega }\delta _{\mu \nu }`$ and $`A_\mu A_\mu `$ thus one can write the action $`S_A`$ equivalently as follows,
$$S_A=\frac{1}{4}d^4x\sqrt{g}g^{\mu \rho }g^{\nu \sigma }F_{\mu \nu }F_{\rho \sigma },$$
(22)
where $`F_{\mu \nu }=_\mu A_\nu _\nu A_\mu `$ is the field strength in the de Sitter background. Consequently at $`g=0^+`$ the theory given by the action (16) is a conformally coupled scalar theory minimally coupled to $`U(1)`$ gauge field on de Sitter background. Therefore at $`g=0^+`$ using optical instruments people living on flat Euclidean space observe an accelerating universe.
## Acknowledgement
The financial support of Isfahan University of Technology (IUT) is acknowledged.
|
warning/0506/quant-ph0506113.html
|
ar5iv
|
text
|
# Entanglement in an expanding spacetime
## Abstract
We show that a dynamical spacetime generates entanglement between modes of a quantum field. Conversely, the entanglement encodes information concerning the underlying spacetime structure, which hints at the prospect of applications of this observation to cosmology. Here we illustrate this point by way of an analytically exactly soluble example, that of a scalar quantum field on a two-dimensional asymptotically flat Robertson-Walker expanding spacetime. We explicitly calculate the entanglement in the far future, for a quantum field residing in the vacuum state in the distant past. In this toy universe, it is possible to fully reconstruct the parameters of the cosmic history from the entanglement entropy.
The phenomenon of entanglement has attracted much attention in recent years. Its central importance in the exciting discipline of quantum information science is undisputable: it has emerged as a fundamental resource in quantum communication quantcomm , quantum cryptography quantcrypto , quantum teleportation quanttele and quantum computation quantcomp . Recent effort has begun to translate some of the aforementioned concepts to the special relativistic setting special1 ; special2 ; special3 ; special4 ; special5 ; special6 and recently progress has been made in examining teleportation teleport and entanglement between modes of a quantum field ivette when one of the observers is uniformly accelerated.
However, quantum information remains a discipline that commonly avoids the conceptual challenges implied by one of the most fundamental insights of modern physics, namely that spacetime is dynamical and curved. When first attempting to understand the basic principles of quantum information, the simplifying assumption of a flat, or even non-relativistic, spacetime is justifiable. However, since our ultimate goal is to properly understand the nature of the universe on a more complete level, and not merely a restricted, simplified aspect of it, we must necessarily extend our analysis to the more general curved spacetime domain in which we live. Moreover, while for many physical systems of interest to quantum information theory a non-relativistic approximation is sufficient, the latter entirely fails for massless particles, such as photons, or in the presence of strong gravitational fields. The advent of precision measurements in cosmology cosmo , for instance, produces a host of data, whose interpretation will be further enriched by translation of concepts from quantum information theory to curved spaces.
In this brief note we consider the effect that an expanding universe has on the entanglement shared between scalar particles residing in that spacetime, demonstrating that such a dynamic background structure actually creates entanglement. Furthermore, we explicitly demonstrate the fascinating possibility to deduce cosmological parameters of the underlying spacetime from the entanglement shared between certain modes of a quantum field.
The observation that entanglement is affected by the underlying spacetime structure may appear rather startling to those unfamiliar with quantum field theory on curved spacetime, but is an immediate consequence of the latter Birrelldavies . Indeed, many of the constructions of quantum theory in flat spacetime, such as the notion of a particle, only possess limited validity in the general setting. Despite such conceptual challenges, we show that the utility of entanglement can be fruitfully extended beyond its usual domain of non-relativistic quantum information.
We present the required theory in fair generality, but for simplicity discuss as an explicit example a toy universe in two spacetime dimensions. More realistic examples differ not in principle, but only in their analytical complexity.
On a curved $`d`$-dimensional spacetime $`M`$ with metric $`g`$, we consider a complex valued scalar field $`\varphi `$, whose dynamics is given by variation of the action
$$S=_Md^dx\sqrt{g}[g^{\mu \nu }(x)_\mu \varphi _\nu \varphi m^2\varphi ^2],$$
(1)
where $`g=(1)^{d+1}detg_{\mu \nu }`$ and $`m`$ is a real positive parameter. The corresponding equation of motion is
$$(\mathrm{}+m^2)\varphi (x)=0,$$
(2)
where $`\mathrm{}\varphi =_\mu (\sqrt{g}g^{\mu \nu }_\nu \varphi )/\sqrt{g}`$. Due to the linearity in $`\varphi `$, the space of solutions constitutes a vector space, which is made into a Hilbert space $``$ by equipping it with the time-independent inner product
$$(\varphi ,\psi )=i_\mathrm{\Sigma }𝑑\mathrm{\Sigma }^\mu [\psi ^{}_\mu \varphi \varphi _\mu \psi ^{}],$$
(3)
where $`\mathrm{\Sigma }`$ is a spacelike Cauchy surface. As in any complex Hilbert space, we can find an orthonormal basis of the form $`\{u_p,u_p^{}\}`$ such that
$$(u_p,u_q)=(u_p^{},u_q^{})=\delta _{pq}\text{ and }(u_p,u_q^{})=0,$$
(4)
where $`p,q`$ are in some (possibly continuous) index set $`𝒫`$. In general, no particular splitting of the basis into $`u_p`$ and $`u_p^{}`$ is distinguished. In a spacetime with time translation symmetry, however, we have a timelike Killing vector field $`K`$, satisfying $`_Kg=0`$, where $``$ denotes the Lie derivative. The corresponding energy conservation allows one to meaningfully classify solutions of (2) into positive and negative frequency solutions, if they are eigenfunctions of $`i_K`$ with positive or negative eigenvalues, respectively.
Canonical quantization of the theory consists of promoting the field $`\varphi `$ to an operator field
$$\widehat{\varphi }(x)=_𝒫[a_p^{}u(p)+a_p^+u^{}(p)],$$
(5)
where $`a_p^{}`$ and $`a_p^+`$ act on the bosonic Fock space
$$=()()\mathrm{}$$
(6)
by linear extension of their action on $`^n`$,
$$\begin{array}{cc}\hfill a_p^{}|q_1\mathrm{}.|q_n& =\underset{i=1}{\overset{n}{}}|q_1\mathrm{}.p|q_i\mathrm{}.|q_n^{(n1)},\hfill \\ \hfill a_p^+|q_1\mathrm{}.|q_n& =|p|q_1\mathrm{}.|q_n^{(n+1)},\hfill \end{array}$$
(7)
where the order of the kets is irrelevant due to the symmetric tensor product $``$. The Fock space $``$ inherits the inner product (3) from $``$, from which it immediately follows that $`a_p^+=(a_p^{})^{}`$, where $``$ denotes the adjoint with respect to the inner product on $``$. We will from now on denote $`a_p^{}`$ simply as $`a_p`$. From the above construction it follows that
$$[a_p^{},a_q]=\delta _{pq}\text{ and }[a_p,a_q]=[a_p^{},a_q^{}]=0.$$
(8)
These commutation relations are often stated as the essence of the canonical quantization procedure for bosons. If the $`u_p`$ are positive frequency solutions with respect to some timelike Killing vector field $`K`$, the operators $`a_p^{}`$ and $`a_p`$ may be meaningfully interpreted as creation and annihilation operators for a particle excitation (of mass $`m`$, momentum $`p`$, and energy $`\sqrt{p^2+m^2}`$), if the vacuum state $`|01_{}0_{}\mathrm{}`$ is interpreted as the no-particle state. Note that from this definition it immediately follows that $`a_p|0=0`$ for all $`p𝒫`$.
On a generic curved spacetime, there exists no global timelike Killing vector field $`K`$, so that in general no meaningful particle interpretation can be attached to a state of the quantum field. However, if in some region of the spacetime under consideration, there exists a local Killing vector field, we will exploit the fact that a particle interpretation exists in that region. In the absence of a global Killing vector field, the same quantum state then generically has particle interpretations varying between those regions that feature their own individual local Killing vector fields, and thus different positive and negative frequency modes. The corresponding change from one set of positive and negative frequency modes $`\{u_p,u_p^{}\}`$ to another set $`\{\overline{u}_p,\overline{u}_p^{}\}`$ is simply a change of the Hilbert space basis. As creation and annihilation operators are defined with respect to a specific mode decomposition, changing the latter induces the Bogolubov transformations
$$\overline{a}_p=_{q𝒫}[\alpha _{pq}^{}a_q\beta _{pq}^{}a_q^{}],$$
(9)
parameterized by the projection coefficients $`\alpha _{pq}=(\overline{u}_p,u_q)`$ and $`\beta _{pq}=(\overline{u}_p,u_q^{})`$ of the basis change in $``$. In the case of non-vanishing $`\beta _{pq}`$, it follows that the vacua $`|0`$ and $`|\overline{0}`$, defined with respect to the different mode decompositions, are inequivalent. As a consequence, the particle concept, so familiar and widely used in discussions of conventional quantum information, is a more intricate one on curved spacetime.
The majority of explicit calculations in quantum field theory on curved spacetime are notoriously difficult. However, there exist certain specific models of the spacetime structure that are exactly analytically soluble. We shall consider one such simple model that is asymptotically flat in the distant past and far future, as this allows us to easily illustrate the main principles and consequences of examining entanglement in curved spacetime without having to resort to approximations or numerical solutions.
Specifically, we consider a two-dimensional Robertson-Walker expanding spacetime with line element
$$ds^2=C(\tau )(d\tau ^2dx^2),$$
(10)
where $`\tau `$ is the conformal time and the conformal scale factor is given as
$$C(\tau )=1+ϵ(1+\mathrm{tanh}\sigma \tau ),$$
(11)
with positive real parameters $`ϵ`$ and $`\sigma `$, controlling the total volume and rapidity of the expansion. This describes a toy universe undergoing a period of smooth expansion. In the distant past and far future, the spacetime becomes Minkowskian since $`C(\tau )`$ tends to $`1+2ϵ`$ and $`1`$ as $`\tau \pm \mathrm{}`$, respectively. As a consequence, the vector field $`K=/\tau `$ has the Killing property in both the asymptotic in-region ($`\tau \mathrm{}`$) and out-region ($`\tau \mathrm{}`$), but not for finite $`\tau `$. In the asymptotic regions it is possible to sensibly discuss the particle content of a scalar field; in the intermediate region, however, the concept of a particle breaks down.
In order to find the solutions to the Klein-Gordon equation Eq. (2) on this spacetime, we note that $`C(\tau )`$ is independent of $`x`$. We exploit the resulting spatial translational invariance and separate the solutions into
$$\varphi _p(\tau ,x)=(2\pi )^{1/2}e^{ipx}\chi _p(\tau ),$$
(12)
so that $`𝒫=`$ in this example. Inserting this into the Klein-Gordon equation, we obtain a simple differential equation for $`\chi _p(\tau )`$ which may be solved in terms of hypergeometric functions Birrelldavies . We then apply the Lie derivative $`_K`$ to the solutions $`\varphi _p`$, in order to identify the normalized modes $`\overline{u}_p`$ which behave like positive frequency modes in the remote past, and the positive frequency modes $`u_p`$ in the far future, respectively:
$`\overline{u}_p(\tau ,x)`$ $`\stackrel{}{_\tau \mathrm{}}`$ $`(4\pi \omega _{\text{in}})^{1/2}e^{i(px\omega _{\text{in}}\tau )},`$
$`u_p(\tau ,x)`$ $`\stackrel{}{_{\tau +\mathrm{}}}`$ $`(4\pi \omega _{\text{out}})^{1/2}e^{i(px\omega _{\text{out}}\tau )},`$ (13)
where the angular frequencies have the form
$`\omega _{\text{in}}`$ $`=`$ $`[p^2+m^2]^{1/2},`$
$`\omega _{\text{out}}`$ $`=`$ $`[p^2+m^2(1+2ϵ)]^{1/2},`$
$`\omega _\pm `$ $`=`$ $`{\displaystyle \frac{1}{2}}(\omega _{\text{out}}\pm \omega _{\text{in}}).`$ (14)
We shall henceforth denote all quantities related to the in-region using a bar, and those referring to the out-region without a bar. A consequence of the linear transformation properties of hypergeometric functions is that the Bogolubov transformations associated with the transformation from $`\{\overline{u}_p,\overline{u}_p^{}\}`$ to $`\{u_p,u_p^{}\}`$ take the simple form
$$\overline{a}_s=\alpha _s^{}a_s\beta _s^{}a_s^{},$$
(15)
so that mixing occurs only between states labelled by $`s`$ and $`s`$.
Now consider the case where the ingoing scalar field is in a vacuum state $`_{s^+}|\overline{0}_s|\overline{0}_s`$, with no excitations present in any of the modes (from the point of view of an inertial observer in the in-region). Because of the simple mixing properties of the Bogolubov transformations, we may focus solely on a component $`|\overline{0}_k|\overline{0}_k`$ of the input state (disregarding all other modes, by tracing the total density matrix over them, yields an overall factor of unity), and express this component in terms of the modes in the out-region. As a pure state of a bi-partite system, this can be written as a Schmidt decomposition
$$|\overline{0}_k|\overline{0}_k=\underset{n=0}{\overset{\mathrm{}}{}}c_n|n_k|n_k,$$
(16)
where $`n`$ labels the number of excitations in the field mode $`k`$ (as seen by an inertial observer in the out-region) and the coefficients $`c_n`$ are real. An explicit expression for the Schmidt coefficients $`c_n`$ can be obtained by applying Eq. (15) to Eq. (16):
$$0=\overline{a}_k|\overline{0}_k|\overline{0}_k=\left(\alpha _k^{}a_k\beta _k^{}a_k^{}\right)\underset{n=0}{\overset{\mathrm{}}{}}c_n|n_k|n_k.$$
(17)
A simple relabelling of the mode excitation number then allows one to deduce that
$$c_n=\left(\frac{\beta _k^{}}{\alpha _k^{}}\right)^nc_0,$$
(18)
whilst taking the inner product of Eq. (16) with its hermitian conjugate yields the following value for the first Schmidt coefficient:
$$c_0=\sqrt{1\left|\frac{\beta _k}{\alpha _k}\right|^2}.$$
(19)
Thus Eq. (16) shows that a state which is interpreted as a vacuum in the in-region appears as a state with particle excitations in the out-region. We must interpret this fact as the creation of particles as a direct result of the cosmic expansion. Recall however that in the interim region, when our toy universe is undergoing expansion, no sensible notion of a particle exists.
We are now in a position to apply familiar methods of quantum information theory to extract information about the entanglement of the state Eq. (16). It is a simple matter to construct the asymptotic output state bipartite density matrix $`\varrho =|\overline{0}_k|\overline{0}_k_k\overline{0}|{}_{k}{}^{}\overline{0}|`$, describing excitations in the scalar field modes $`k`$ and $`k`$. Because the Schmidt coefficients in Eq. (16) are non-zero, the in-vacuum is entangled from the point of view of an observer in the out-region. Since the density matrix $`\varrho `$ describes a pure state, the von Neumann entropy $`S`$ of the reduced density matrix
$$\varrho _k=\underset{m=0}{\overset{\mathrm{}}{}}{}_{k}{}^{}m|\varrho |m_{k}^{}$$
(20)
presents a well-defined measure for this entanglement of the modes $`k`$ with the modes $`k`$. One finds, after some algebra, the entanglement
$$S=\text{Tr}(\varrho _k\mathrm{log}_2\varrho _k)=\mathrm{log}_2\frac{\gamma ^{\gamma /(\gamma 1)}}{1\gamma },$$
(21)
where
$$\gamma =\left|\frac{\beta _k}{\alpha _k}\right|^2=\frac{\mathrm{sinh}^2(\pi \omega _{}/\sigma )}{\mathrm{sinh}^2(\pi \omega _+/\sigma )}$$
(22)
depends on the cosmological parameters $`ϵ`$ and $`\sigma `$, and the momentum $`k`$ of the selected modes. This means that the expansion of the universe creates entanglement between massive modes of opposite momenta. Modes of a massless quantum field do not get entangled, due to the conformal flatness of the model studied here. Although this total decoupling of massless modes is an artefact of the specific example studied here, it illustrates that the massive and massless case are generically qualitatively different, and non-relativistic intuition fails entirely. Since $`0\gamma <1`$, the entanglement is monotonically increasing in $`\gamma `$, and we may invert Eq. (21) to obtain $`\gamma (S)`$.
Indeed, the entanglement between the field modes encodes the entire information about the underlying spacetime. For light particles, a direct relation between the cosmological parameters and the degree of entanglement can be obtained. To see this explicitly, assume that a universe of the type discussed above, with cosmological parameters $`\sigma `$ and $`ϵ`$, is populated by a particle species of mass $`m2\sigma ϵ^{1/2}`$. Then we can consider quanta of energy $`E_p=\sqrt{p^2+m^2}`$ such that $`m\sqrt{ϵ}E_p2\sigma `$, and we obtain the frequencies
$`\omega _+`$ $``$ $`E_p\sigma ,`$ (23)
$`\omega _{}`$ $``$ $`\omega _+{\displaystyle \frac{m^2}{2E_p^2}}ϵ\sigma ,`$ (24)
such that the cosmological expansion parameter $`ϵ`$ can be determined, by expanding Eq. (22) to leading order in $`ϵ`$, as a monotonically increasing function of the entanglement $`S`$:
$$ϵ\frac{2E_p^2}{m^2}\sqrt{\gamma (S)}.$$
(25)
The cosmological parameter $`\sigma `$ can be determined from the respective entanglement of two modes of slightly different energy. More precisely, one finds by differentiating Eq. (22) with respect to the particle energy $`E`$ that
$$\sigma \frac{\pi }{2}\left(\frac{1+\gamma (S)}{\frac{E}{4}\frac{d}{dE}\mathrm{ln}\gamma (S)1}\right)^{\frac{1}{2}}E,$$
(26)
using the above approximations. Thus, in our simple toy universe, the entanglement of massive states carries the complete information about the cosmological parameters.
In a more realistic four-dimensional setting, neutrinos present a natural candidate for such very light neutrinomass and approximately only gravitationally interacting particles. The required Bogolubov transformations for a four-dimensional Friedmann-Robertson-Walker universe have been calculated stefan , and for spacetime manifolds admitting a spin structure, the Dirac equation for spin-$`1/2`$ fermions can be written down and quantized. Although this more realistic setting is analytically considerably more involved, the qualitative features are the same. In particular, DeWitt dewitt gives the general vacuum to multi-particle production and annihilation amplitudes needed in order to generalize our simple example. The degree of entanglement of neutrinos therefore likely carries information concerning our cosmic history, a fascinating speculation.
Our analysis assumed that the quantum field is in a vacuum state from the point of view of an inertial observer in the distant past. While without further assumptions this presents a natural departure point, one could instead calculate the fate of a non-vacuum state, or even an initially entangled state. The formalism employed above is directly applicable also to those cases. We have neglected, however, any back reaction of the quantum field on the spacetime through the Einstein equations. If the dynamics of the universe are not significantly driven by the quantum field under consideration, this provides a reasonable approximation.
In conclusion, by extending our understanding of entanglement to a curved spacetime background, we have learned that the latter has a significant effect. We showed that an expanding spacetime generates entanglement between certain modes of an only gravitationally interacting scalar field. While this is not an unexpected result within quantum field theory on curved spacetime, we made the interesting observation that, conversely, information on the underlying spacetime can be recovered from the entanglement of very light particles. We exemplified this by pointing out how all cosmological parameters of a toy expanding universe can be extracted from quantum correlations. We will show elsewhere in detail how our treatment can be applied to other spacetimes that are either more realistic or otherwise of interest. Far from being of mere interest to quantum information theory, a further exploration of entanglement in a curved spacetime therefore promises to contribute to a cross-fertilization between gravity and quantum information. Recent progress on generally covariant quantum field theory supplies valuable new tools for such investigations verch .
###### Acknowledgements.
The authors would like to thank Stefan Hofmann for useful discussions. JLB acknowledges financial support from Keble College in Oxford and EPSRC, and thanks Perimeter Institute for its generous hospitality.
|
warning/0506/hep-th0506189.html
|
ar5iv
|
text
|
# Gravitons as super-strong interacting particles, and low-energy quantum gravity
## 1 Introduction
By a full coincidence of the forms of Coulomb’s and Newton’s laws, which describe an interaction of electric charges and a gravitational interaction of bodies, we see a dramatic difference in developing the pictures of these interactions on a quantum level. Constructed with pillars on multiple experiments QED is one of the most exact physical theories and an archetype for imitation by creation of new models. While the attempts to quantize the remarkable in its logical beauty theory of general relativity, which describes gravitation on a classical level so fully and delicately, (see review ) not only have not a hit until today but gave a specific side psychological effect - there exists conceptualization that quantum gravity may be described only by some sophisticated theory. An opinion is commonly accepted, too, that quantum gravity should manifest itself only on the Planck scale of energies, i.e. it is a high-energy phenomenon. The value of the Planck energy $`10^{19}`$ GeV has been got from dimensional reasonings. Still one wide-spread opinion is that we know a mechanism of gravity (bodies are exchanging with gravitons of spin 2) but cannot correctly describe it. If an apple from the legend about Newton’s afflatus can imagine all these complications, it would hesitate to fall to the ground so artlessly as it is accepted among the apples.
Perhaps, physicists would be able to refuse easier the preconceived stereotypes which balk - as it seems to me - to go ahead in understanding quantum gravity if experiments or observations would give more essential meat for reasonings. But in this area, at least up to recent years, there was observed nothing that may serve if not Ariadne’s clew but such a simple physical contradiction that an aspiration to overcome it would advantage introduction of new ideas and revision of the ”inviolable”.
In a few last years, the situation has been abruptly changed. I enumerate those discoveries and observations which may force, in my opinion, the ice to break up.
1. In 1998, Anderson’s team reported about the discovery of anomalous acceleration of NASA’s probes Pioneer 10/11 ; this effect is not embedded in a frame of the general relativity, and its magnitude is somehow equal to $`Hc`$, where $`H`$ is the Hubble constant, $`c`$ is the light velocity.
2. In the same 1998, two teams of astrophysicists, which were collecting supernovae 1a data with the aim to specificate parameters of cosmological expansion, reported about dimming remote supernovae ; the one would be explained on a basis of the Doppler effect if at present epoch the universe expands with acceleration. This explanation needs an introduction of some ”dark energy” which is unknown from any laboratory experiments.
3. In January 2002, Nesvizhevsky’s team reported about discovery of quantum states of ultra-cold neutrons in the Earth’s gravitational field . Observed energies of levels (it means that and their differences too) in full agreement with quantum-mechanical calculations turned out to be equal to $`10^{12}`$ eV. The formula for energy levels had been found still by Bohr and Sommerfeld. If transitions between these levels are accompanied with irradiation of gravitons then energies of irradiated gravitons should have the same order - but it is of 40 orders lesser than the Planck energy by which one waits quantum manifestations of gravity.
The first of these discoveries obliges to muse about the borders of applicability of the general relativity, the third - about that quantum gravity would be a high-energy phenomenon. It seems that the second discovery is far from quantum gravity but it obliges us to look at the traditional interpretation of the nature of cosmological redshift critically. An introduction into consideration of an alternative model of redshifts which is based on a conjecture about an existence of the graviton background gives us odds to see in the effect of dimming supernovae an additional manifestation of low-energy quantum gravity. Under the definite conditions, an effective temperature of the background may be the same one as a temperature of the cosmic microwave background, with an average graviton energy of the order $`10^3`$ eV.
In this chapter, the main results of author’s research in this approach are described. Starting from a statistical model of the graviton background with low temperature, it is shown - under the very important condition that gravitons are super-strong interacting particles - that if a redshift would be a quantum gravitational effect then one can get from its magnitude an estimate of a new dimensional constant characterizing a single act of interaction in this model. It is possible to calculate theoretically a dependence of a light flux relaxation on a redshift value, and this dependence fits supernova observational data very well at least for $`z<0.5`$. Further it is possible to find a pressure of single gravitons of the background which acts on any pair of bodies due to screening the graviton background with the bodies . It turns out that the pressure is huge (a corresponding force is $`1000`$ times stronger than the Newtonian attraction) but it is compensated with a pressure of gravitons which are re-scattered by the bodies. The Newtonian attraction arises if a part of gravitons of the background forms pairs which are destructed by interaction with bodies. It is interesting that both Hubble’s and Newton’s constants may be computed in this approach with the ones being connected between themselves. It allows us to get a theoretical estimate of the Hubble constant. An unexpected feature of this mechanism of gravity is a necessity of ”an atomic structure” of matter - the mechanism doesn’t work without the one.
Collisions with gravitons should also call forth a deceleration of massive bodies of order $`Hc`$ \- namely the same as of NASA’s probes. But at present stage it turns out unclear why such the deceleration has not been observed for planets. The situation reminds by something of the one that took place in physics before the creation of quantum mechanics when a motion of electrons should, as it seemed by canons of classical physics, lead to their fall to a nucleus.
Because the very unexpected hypothesis founded into the basis of this approach is a super-strong character of gravitational interaction on a quantum level, I would like to explain my motivation which conduced namely to such the choice. Learning symmetries of the quantum two-component composite system , I have found that its discrete symmetries in an 8-space may be interpreted by an observer from a 4-dimensional world as the exact global symmetries of the standard model of particle physics if internal coordinates of the system (the composite fermion) are rigidly fixed. This conclusion was hard for me and took much enough time. But to ensure almost full fixation of components of the system, an interaction connecting them should be very strong. Because of it, when a choice arise - an amount of gravitons or an intensity of the interaction, - I have remembered this overpassed earlier barrier and chose namely the super-strong interaction. Without this property, the graviton background would not be in the thermodynamical equilibrium with the cosmic microwave background that could entail big difficulties in the model.
So, in this approach we deal with the following small quantum effects of low-energy gravity: redshifts, its analog - a deceleration of massive bodies, and an additional relaxation of any light flux. The Newtonian attraction turns out to be the main statistical effect, with bodies themselves being not sources of gravitons - only correlational properties of in and out fluxes of gravitons in their neighbourhood are changed due to an interaction with bodies. There does still not exist a full and closed theory in this approach, but even the initial researches in this direction show that in this case quantum gravity cannot be described separately of other interactions, and also manifest the boundaries of applicability of a geometrical language in gravity.
## 2 Passing photons through the graviton background
Let us introduce the hypothesis, which is considered in this approach as independent from the standard cosmological model: there exists the isotropic graviton background. Photon scattering is possible on gravitons $`\gamma +h\gamma +h,`$ where $`\gamma `$ is a photon and $`h`$ is a graviton, if one of the gravitons is virtual. The energy-momentum conservation law prohibits energy transfer to free gravitons. Due to forehead collisions with gravitons, an energy of any photon should decrease when it passes through the sea of gravitons.
From another side, none-forehead collisions of photons with gravitons of the background will lead to an additional relaxation of a photon flux, caused by transmission of a momentum transversal component to some photons. It will lead to an additional dimming of any remote objects, and may be connected with supernova dimming.
We deal here with the uniform non-expanding universe with the Euclidean space, and there are not any cosmological kinematic effects in this model.
### 2.1 Forehead collisions with gravitons: an alternative explanations of the redshift nature
We shall take into account that a gravitational ”charge” of a photon must be proportional to $`E`$ (it gives the factor $`E^2`$ in a cross-section) and a normalization of a photon wave function gives the factor $`E^1`$ in the cross-section. Also we assume here that a photon average energy loss $`\overline{ϵ}`$ in one act of interaction is relatively small to a photon energy $`E.`$ Then average energy losses of a photon with an energy $`E`$ on a way $`dr`$ will be equal to :
$$dE=aEdr,$$
(1)
where $`a`$ is a constant. If a whole redshift magnitude is caused by this effect, we must identify $`a=H/c,`$ where $`c`$ is the light velocity, to have the Hubble law for small distances .
A photon energy $`E`$ should depend on a distance from a source $`r`$ as
$$E(r)=E_0\mathrm{exp}(ar),$$
(2)
where $`E_0`$ is an initial value of energy.
The expression (2) is just only so far as the condition $`\overline{ϵ}<<E(r)`$ takes place. Photons with a very small energy may lose or acquire an energy changing their direction of propagation after scattering. Early or late such photons should turn out in the thermodynamic equilibrium with the graviton background, flowing into their own background. Decay of virtual gravitons should give photon pairs for this background, too. Perhaps, the last one is the cosmic microwave background .
It follows from the expression (2) that an exact dependence $`r(z)`$ is the following one:
$$r(z)=ln(1+z)/a,$$
(3)
if an interaction with the graviton background is the only cause of redshifts. It is very important, that this redshift does not depend on a light frequency. For small $`z,`$ the dependence $`r(z)`$ will be linear.
The expressions (1) - (3) are the same that appear in other tired-light models (compare with ). In this approach, the ones follow from a possible existence of the isotropic graviton background, from quantum electrodynamics, and from the fact that a gravitational ”charge” of a photon must be proportional to $`E.`$
### 2.2 Non-forehead collisions with gravitons: an additional dimming of any light flux
Photon flux’s average energy losses on a way $`dr`$ due to non-forehead collisions with gravitons should be proportional to $`badr,`$ where $`b`$ is a new constant of the order $`1.`$ These losses are connected with a rejection of a part of photons from a source-observer direction. Such the relaxation together with the redshift will give a connection between visible object’s diameter and its luminosity (i.e. the ratio of an object visible angular diameter to a square root of visible luminosity), distinguishing from the one of the standard cosmological model.
Let us consider that in a case of a non-forehead collision of a graviton with a photon, the latter leaves a photon flux detected by a remote observer (an assumption of a narrow beam of rays). The details of calculation of the theoretical value of relaxation factor $`b`$ which was used in author’s paper were given later in the preprint . So as both particles have velocities $`c,`$ a cross-section of interaction, which is ”visible” under an angle $`\theta `$ (see Fig. 1), will be equal to $`\sigma _0|\mathrm{cos}\theta |`$ if $`\sigma _0`$ is a cross-section by forehead collisions. The function $`|\mathrm{cos}\theta |`$ allows to take into account both front and back hemispheres for riding gravitons. Additionally, a graviton flux, which falls on a picked out area (cross-section), depends on the angle $`\theta .`$ We have for the ratio of fluxes:
$$\mathrm{\Phi }(\theta )/\mathrm{\Phi }_0=S_s/\sigma _0,$$
where $`\mathrm{\Phi }(\theta )`$ and $`\mathrm{\Phi }_0`$ are the fluxes which fall on $`\sigma _0`$ under the angle $`\theta `$ and normally, $`S_s`$ is a square of side surface of a truncated cone with a base $`\sigma _0`$ (see Fig. 1).
Finally, we get for the factor $`b:`$
$$b=2_0^{\pi /2}\mathrm{cos}\theta (S_s/\sigma _0)\frac{d\theta }{\pi /2}.$$
(4)
By $`0<\theta <\pi /4,`$ a formed cone contains self-intersections, and it is $`S_s=2\sigma _0\mathrm{cos}\theta `$. By $`\pi /4\theta \pi /2,`$ we have $`S_s=4\sigma _0\mathrm{sin}^2\theta \mathrm{cos}\theta `$.
After computation of simple integrals, we get:
$$b=\frac{4}{\pi }(_0^{\pi /4}2\mathrm{cos}^2\theta d\theta +_{\pi /4}^{\pi /2}\mathrm{sin}^22\theta d\theta )=\frac{3}{2}+\frac{2}{\pi }2.137.$$
(5)
In the considered simplest case of the uniform non-expanding universe with the Euclidean space, we shall have the quantity
$$(1+z)^{(1+b)/2}(1+z)^{1.57}$$
in a visible object diameter-luminosity connection if a whole redshift magnitude would caused by such an interaction with the background (instead of $`(1+z)^2`$ for the expanding uniform universe). For near sources, the estimate of the factor $`b`$ will be some increased one.
The luminosity distance (see ) is a convenient quantity for astrophysical observations. Both redshifts and the additional relaxation of any photonic flux due to non-forehead collisions of gravitons with photons lead in our model to the following luminosity distance $`D_L:`$
$$D_L=a^1\mathrm{ln}(1+z)(1+z)^{(1+b)/2}a^1f_1(z),$$
(6)
where $`f_1(z)\mathrm{ln}(1+z)(1+z)^{(1+b)/2}`$.
### 2.3 Comparison of the theoretical predictions with supernova data
To compare a form of this predicted dependence $`D_L(z)`$ by unknown, but constant $`H`$, with the latest observational supernova data by Riess et al. , one can introduce distance moduli $`\mu _0=5\mathrm{log}D_L+25=5\mathrm{log}f_1+c_1`$, where $`c_1`$ is an unknown constant (it is a single free parameter to fit the data); $`f_1`$ is the luminosity distance in units of $`c/H`$.
In Figure 2, the Hubble diagram $`\mu _0(z)`$ is shown with $`c_1=43`$ to fit observations for low redshifts; observational data (82 points) are taken from Table 5 of . The predictions fit observations very well for roughly $`z<0.5`$. It excludes a need of any dark energy to explain supernovae dimming.
Discrepancies between predicted and observed values of $`\mu _0(z)`$ are obvious for higher $`z`$: we see that observations show brighter SNe that the theory allows, and a difference increases with $`z`$. It is better seen on Figure 3 with a linear scale for $`f_1`$; observations are transformed as $`\mu _010^{(\mu _0c_1)/5}`$ with the same $`c_1=43`$.<sup>1</sup><sup>1</sup>1A spread of observations raises with $`z`$; it might be partially caused by quickly raising contribution of a dispersion of measured flux: it should be proportional to $`f_1^6(z)`$.
It would be explained in the model as a result of specific deformation of SN spectra due to a discrete character of photon energy losses. Today, a theory of this effect does not exist, and I explain its origin only qualitatively . For very small redshifts $`z,`$ only a small part of photons transmits its energy to the background (see below Fig. 8 in Section 6). Therefore any red-shifted narrow spectral strip will be a superposition of two strips. One of them has a form which is identical with an initial one, its space is proportional to $`1n(r)`$ where $`n(r)`$ is an average number of interactions of a single photon with the background, and its center’s shift is negligible (for a narrow strip). Another part is expand, its space is proportional to $`n(r),`$ and its center’s shift is equal to $`\overline{ϵ}_g/h`$ where $`\overline{ϵ}_g`$ is an average energy loss in one act of interaction. An amplitude of the red-shifted step should linear raise with a redshift. For big $`z,`$ spectra of remote objects of the universe would be deformed. A deformation would appear because of multifold interactions of a initially-red-shifted part of photons with the graviton background. It means that the observed flux within a given passband would depend on a form of spectrum: the flux may be larger than an expected one without this effect if an initial flux within a next-blue neighbour band is big enough - due to a superposition of red-shifted parts of spectrum. Some other evidences of this effect would be an apparent variance of the fine structure constant or of the CMB temperature with epochs. In both cases, a ratio of red-shifted spectral line’s intensities may be sensitive to the effect.
This comparison with observations is very important; to see some additional details, we can compute and graph the ratio $`f_{1obs}(z)/f_1(z)`$, where $`f_{1obs}(z)`$ is an observed analog of $`f_1(z)`$ (see Fig. 4) . An expected value of the ratio should be equal to 1 for any $`z`$.
### 2.4 Computation of the Hubble constant
Let us consider that a full redshift magnitude is caused by an interaction with single gravitons. If $`\sigma (E,ϵ)`$ is a cross-section of interaction by forehead collisions of a photon with an energy $`E`$ with a graviton, having an energy $`ϵ,`$ we consider really (see (1)), that
$$\frac{d\sigma (E,ϵ)}{Ed\mathrm{\Omega }}=const(E),$$
where $`d\mathrm{\Omega }`$ is a space angle element, and the function $`const(x)`$ has a constant value for any $`x`$. If $`f(\omega ,T)d\mathrm{\Omega }/2\pi `$ is a spectral density of graviton flux in the limits of $`d\mathrm{\Omega }`$ in some direction ($`\omega `$ is a graviton frequency, $`ϵ=\mathrm{}\omega `$), i.e. an intensity of a graviton flux is equal to the integral $`(d\mathrm{\Omega }/2\pi )_0^{\mathrm{}}f(\omega ,T)𝑑\omega ,`$ $`T`$ is an equivalent temperature of the graviton background, we can write for the Hubble constant $`H=ac,`$ introduced in the expression (1):
$$H=\frac{1}{2\pi }_0^{\mathrm{}}\frac{\sigma (E,ϵ)}{E}f(\omega ,T)𝑑\omega .$$
If $`f(\omega ,T)`$ can be described by the Planck formula for equilibrium radiation, then
$$_0^{\mathrm{}}f(\omega ,T)𝑑\omega =\sigma T^4,$$
where $`\sigma `$ is the Stephan- Boltzmann constant. As carriers of a gravitational ”charge” (without consideration of spin properties), gravitons should be described in the same manner as photons (compare with ), i.e. one can write for them:
$$\frac{d\sigma (E,ϵ)}{ϵd\mathrm{\Omega }}=const(ϵ).$$
Now let us introduce a new dimensional constant $`D`$, so that for forehead collisions:
$$\sigma (E,ϵ)=DEϵ.$$
(7)
Then
$$H=\frac{1}{2\pi }D\overline{ϵ}(\sigma T^4),$$
(8)
where $`\overline{ϵ}`$ is an average graviton energy. Assuming $`T3K,\overline{ϵ}10^4eV,`$ and $`H=1.610^{18}s^1,`$ we get the following rough estimate for $`D:`$
$$D10^{27}m^2/eV^2,$$
(see below Section 4.3 for more exact estimate of $`D`$ and for a theoretical estimate of $`H`$) that gives us the phenomenological estimate of cross-section by the same and equal $`E`$ and $`\overline{ϵ}`$:
$$\sigma (E,\overline{ϵ})10^{35}m^2.$$
### 2.5 Some new constants from dimensional reasonings
We can introduce the following new constants (see ): $`G_0,l_0,E_0,`$ which are analogues, on this new scale, of classical constants: the Newton constant $`G,`$ the Planck length $`l_{Pl},`$ and the Planck energy $`E_{Pl}`$ correspondingly. Let us accept from dimensional reasonings that
$$D(l_0/E_0)^2=(G_0/c^4)^2,$$
where $`l_0=\sqrt{G_0\mathrm{}/c^3},E_0=\sqrt{\mathrm{}c^5/G_0}.`$ Then we have for these new constants:
$$G_01.610^{39}m^3/kgs^2,l_02.410^{12}m,E_01.6keV.$$
If one would replace $`G`$ with $`G_0,`$ then an electrostatic force, acting between two protons, will be only $`210^{13}`$ times smaller than a gravitational one by the same distance (a theoretical finding of the Newton constant $`G`$ is given below in Section 4.3).
Using $`E_0`$ instead of $`E_{Pl},`$ we can evaluate the new non-dimensional ”constant” (a bilinear function of $`E`$ and $`ϵ`$), $`k,`$ which would characterize one act of interaction: $`kEϵ/E_0^2.`$ We must remember here, that a universality of gravitational interaction allows to expect that this floating coupling ”constant” $`k`$ should characterize interactions of any particles with an energy $`E,`$ including gravitons, with single gravitons. For $`E1eV`$ and $`ϵ10^4eV,`$ we have $`k410^9.`$ But for $`E25MeV`$ and $`ϵ10^3eV,`$ we shall have $`k10^2,`$ i.e. $`k`$ will be comparable with QED’s constant $`\alpha .`$ Already by $`Eϵ5keV,`$ such an interaction would have the same intensity as the strong interaction ($`k10`$).
## 3 Deceleration of massive bodies: an analog of redshifts
As it was reported by Anderson’s team , NASA deep-space probes (Pioneer 10/11, Galileo, and Ulysses) experience a small additional constant acceleration, directed towards the Sun (the Pioneer anomaly). Today, a possible origin of the effect is unknown. It must be noted here that the reported direction of additional acceleration may be a result of the simplest conjecture, which was accepted by the authors to provide a good fit for all probes. One should compare different conjectures to choose the one giving the best fit.
We consider here a deceleration of massive bodies, which would give a similar deformation of cosmic probes’ trajectories . The one would be a result of interaction of a massive body with the graviton background, but such an additional acceleration will be directed against a body velocity.
It follows from a universality of gravitational interaction, that not only photons, but all other objects, moving relative to the background, should lose their energy, too, due to such a quantum interaction with gravitons. If $`a=H/c,`$ it turns out that massive bodies must feel a constant deceleration of the same order of magnitude as a small additional acceleration of cosmic probes.
Let us now denote as $`E`$ a full energy of a moving body which has a velocity $`v`$ relative to the background. Then energy losses of the body by an interaction with the graviton background (due to forehead collisions with gravitons) on the way $`dr`$ must be expressed by the same formula (1):
$$dE=aEdr,$$
where $`a=H/c.`$ If $`dr=vdt,`$ where $`t`$ is a time, and $`E=mc^2/\sqrt{1v^2/c^2},`$ then we get for the body acceleration $`wdv/dt`$ by a non-zero velocity:
$$w=ac^2(1v^2/c^2).$$
(9)
We assume here, that non-forehead collisions with gravitons give only stochastic deviations of a massive body’s velocity direction, which are negligible. For small velocities:
$$wHc.$$
(10)
If the Hubble constant $`H`$ is equal to $`2.1410^{18}s^1`$ (it is the theoretical estimate of $`H`$ in this approach, see below Section 4.3), a modulus of the acceleration will be equal to
$$|w|Hc=6.41910^{10}m/s^2,$$
(11)
that has the same order of magnitude as a value of the observed additional acceleration $`(8.74\pm 1.33)10^{10}m/s^2`$ for NASA probes.
We must emphasize here that the acceleration $`w`$ is directed against a body velocity only in a special frame of reference (in which the graviton background is isotropic). In other frames, we may find its direction, using transformation formulae for an acceleration (see ). We can assume that the graviton background and the microwave one are isotropic in one frame (the Earth velocity relative to the microwave background was determined in ).
To verify my conjecture about the origin of probes’ additional acceleration, one could re-analyze radio Doppler data for the probes. One should find a velocity of the special frame of reference and a constant probe acceleration $`w`$ in this frame which must be negative, as it is described above. These two parameters must provide the best fit for all probes, if the conjecture is true. In such a case, one can get an independent estimate of the Hubble constant, based on the measured value of probe’s additional acceleration: $`H=w/c.`$ I would like to note that a deep-space mission to test the discovered anomaly is planned now at NASA by the authors of this very important discovery .
Under influence of such a small additional acceleration $`w`$, a probe must move on a deformed trajectory. Its view will be determined by small seeming deviations from exact conservation laws for energy and angular momentum of a not-fully reserved body system which one has in a case of neglecting with the graviton background. For example, Ulysses should go some nearer to the Sun when the one rounds it. It may be interpreted as an additional acceleration, directed towards the Sun, if we shall think that one deals with a reserved body system.
It is very important to understand, why such an acceleration has not been observed for planets. This acceleration will have different directions by motion of a body on a closed orbit, and one must take into account a solar system motion, too. As a result, an orbit should be deformed. The observed value of anomalous acceleration of Pioneer 10 should represent the vector difference of the two accelerations : an acceleration of Pioneer 10 relative to the graviton background, and an acceleration of the Earth relative to the background. Possibly, the last is displayed as an annual periodic term in the residuals of Pioneer 10 . If the solar system moves with a noticeable velocity relative to the background, the Earth’s anomalous acceleration projection on the direction of this velocity will be smaller than for the Sun - because of the Earth’s orbital motion. It means that in a frame of reference, connected with the Sun, the Earth should move with an anomalous acceleration having non-zero projections as well on the orbital velocity direction as on the direction of solar system motion relative to the background. Under some conditions, the Earth’s anomalous acceleration in this frame of reference may be periodic. The axis of Earth’s orbit should feel an annual precession by it. This question needs a further consideration.
## 4 Gravity as the screening effect
It was shown by the author that screening the background of super-strong interacting gravitons creates for any pair of bodies both attraction and repulsion forces due to pressure of gravitons. For single gravitons, these forces are approximately balanced, but each of them is much bigger than a force of Newtonian attraction. If single gravitons are pairing, an attraction force due to pressure of such graviton pairs is twice exceeding a corresponding repulsion force if graviton pairs are destructed by collisions with a body. In such the model, the Newton constant is connected with the Hubble constant that gives a possibility to obtain a theoretical estimate of the last. We deal here with a flat non-expanding universe fulfilled with super-strong interacting gravitons; it changes the meaning of the Hubble constant which describes magnitudes of three small effects of quantum gravity but not any expansion or an age of the universe.
### 4.1 Pressure force of single gravitons
If gravitons of the background run against a pair of bodies with masses $`m_1`$ and $`m_2`$ (and energies $`E_1`$ and $`E_2`$) from infinity, then a part of gravitons is screened. Let $`\sigma (E_1,ϵ)`$ is a cross-section of interaction of body $`1`$ with a graviton with an energy $`ϵ=\mathrm{}\omega ,`$ where $`\omega `$ is a graviton frequency, $`\sigma (E_2,ϵ)`$ is the same cross-section for body $`2.`$ In absence of body $`2,`$ a whole modulus of a gravitonic pressure force acting on body $`1`$ would be equal to:
$$4\sigma (E_1,<ϵ>)\frac{1}{3}\frac{4f(\omega ,T)}{c},$$
(12)
where $`f(\omega ,T)`$ is a graviton spectrum with a temperature $`T`$ (assuming to be Planckian), the factor $`4`$ in front of $`\sigma (E_1,<ϵ>)`$ is introduced to allow all possible directions of graviton running, $`<ϵ>`$ is another average energy of running gravitons with a frequency $`\omega `$ taking into account a probability of that in a realization of flat wave a number of gravitons may be equal to zero, and that not all of gravitons ride at a body.
Body $`2,`$ placed on a distance $`r`$ from body $`1,`$ will screen a portion of running against body $`1`$ gravitons which is equal for big distances between the bodies (i.e. by $`\sigma (E_2,<ϵ>)4\pi r^2`$) to:
$$\frac{\sigma (E_2,<ϵ>)}{4\pi r^2.}$$
(13)
Taking into account all frequencies $`\omega ,`$ the following attractive force will act between bodies $`1`$ and $`2:`$
$$F_1=_0^{\mathrm{}}\frac{\sigma (E_2,<ϵ>)}{4\pi r^2}4\sigma (E_1,<ϵ>)\frac{1}{3}\frac{4f(\omega ,T)}{c}𝑑\omega .$$
(14)
Let $`f(\omega ,T)`$ is described with the Planck formula:
$$f(\omega ,T)=\frac{\omega ^2}{4\pi ^2c^2}\frac{\mathrm{}\omega }{\mathrm{exp}(\mathrm{}\omega /kT)1}.$$
(15)
Let $`x\mathrm{}\omega /kT,`$ and $`\overline{n}1/(\mathrm{exp}(x)1)`$ is an average number of gravitons in a flat wave with a frequency $`\omega `$ (on one mode of two distinguishing with a projection of particle spin). Let $`P(n,x)`$ is a probability of that in a realization of flat wave a number of gravitons is equal to $`n,`$ for example $`P(0,x)=\mathrm{exp}(\overline{n}).`$
A quantity $`<ϵ>`$ must contain the factor $`(1P(0,x)),`$ i.e. it should be:
$$<ϵ>\mathrm{}\omega (1P(0,x)),$$
(16)
that let us to reject flat wave realizations with zero number of gravitons.
But attempting to define other factors in $`<ϵ>,`$ we find the difficult place in our reasoning. On this stage, it is necessary to introduce some new assumption to find the factors. Perhaps, this assumption will be well-founded in a future theory - or would be rejected. If a flat wave realization, running against a finite size body from infinity, contains one graviton, then one cannot consider that it must stringent ride at a body to interact with some probability with the one. It would break the uncertainty principle by W. Heisenberg. We should admit that we know a graviton trajectory. The same is pertaining to gravitons scattered by one of bodies by big distances between bodies. What is a probability that a single graviton will ride namely at the body? If one denotes this probability as $`P_1,`$ then for a wave with $`n`$ gravitons their chances to ride at the body must be equal to $`nP_1.`$ Taking into account the probabilities of values of $`n`$ for the Poisson flux of events, an additional factor in $`<ϵ>`$ should be equal to $`\overline{n}P_1.`$ I have admitted in that
$$P_1=P(1,x),$$
(17)
where $`P(1,x)=\overline{n}\mathrm{exp}(\overline{n});`$ (below it is admitted for pairing gravitons: $`P_1=P(1,2x)`$ \- see Section 4.3).
In such the case, we have for $`<ϵ>`$ the following expression:
$$<ϵ>=\mathrm{}\omega (1P(0,x))\overline{n}^2\mathrm{exp}(\overline{n}).$$
(18)
Then we get for an attractive force $`F_1:`$
$$F_1=\frac{4}{3}\frac{D^2E_1E_2}{\pi r^2c}_0^{\mathrm{}}\frac{\mathrm{}^3\omega ^5}{4\pi ^2c^2}(1P(0,x))^2\overline{n}^5\mathrm{exp}(2\overline{n})𝑑\omega =$$
(19)
$$\frac{1}{3}\frac{D^2c(kT)^6m_1m_2}{\pi ^3\mathrm{}^3r^2}I_1,$$
where
$$I_1_0^{\mathrm{}}x^5(1\mathrm{exp}((\mathrm{exp}(x)1)^1))^2(\mathrm{exp}(x)1)^5\mathrm{exp}(2(\mathrm{exp}(x)1)^1)𝑑x=$$
(20)
$$5.63610^3.$$
This and all other integrals were found with the MathCad software.
If $`F_1G_1m_1m_2/r^2,`$ then the constant $`G_1`$ is equal to:
$$G_1\frac{1}{3}\frac{D^2c(kT)^6}{\pi ^3\mathrm{}^3}I_1.$$
(21)
By $`T=2.7K:`$
$$G_1=1215.4G,$$
(22)
that is three order greater than the Newton constant, $`G.`$
But if single gravitons are elastically scattered with body $`1,`$ then our reasoning may be reversed: the same portion (13) of scattered gravitons will create a repulsive force $`F_1^{^{}}`$ acting on body $`2`$ and equal to
$$F_1^{^{}}=F_1,$$
(23)
if one neglects with small allowances which are proportional to $`D^3/r^4.`$
So, for bodies which elastically scatter gravitons, screening a flux of single gravitons does not ensure Newtonian attraction. But for gravitonic black holes which absorb any particles and do not re-emit them (by the meaning of a concept, the ones are usual black holes; I introduce a redundant adjective only from a caution), we will have $`F_1^{^{}}=0.`$ It means that such the object would attract other bodies with a force which is proportional to $`G_1`$ but not to $`G,`$ i.e. Einstein’s equivalence principle would be violated for them. This conclusion, as we shall see below, stays in force for the case of graviton pairing, too. The conclusion cannot be changed with taking into account of Hawking’s quantum effect of evaporation of black holes .
### 4.2 Graviton pairing
To ensure an attractive force which is not equal to a repulsive one, particle correlations should differ for in and out flux. For example, single gravitons of running flux may associate in pairs . If such pairs are destructed by collision with a body, then quantities $`<ϵ>`$ will be distinguished for running and scattered particles. Graviton pairing may be caused with graviton’s own gravitational attraction or gravitonic spin-spin interaction. Left an analysis of the nature of graviton pairing for the future; let us see that gives such the pairing.
To find an average number of pairs $`\overline{n}_2`$ in a wave with a frequency $`\omega `$ for the state of thermodynamic equilibrium, one may replace $`\mathrm{}2\mathrm{}`$ by deducing the Planck formula. Then an average number of pairs will be equal to:
$$\overline{n}_2=\frac{1}{\mathrm{exp}(2x)1},$$
(24)
and an energy of one pair will be equal to $`2\mathrm{}\omega .`$ It is important that graviton pairing does not change a number of stationary waves, so as pairs nucleate from existing gravitons. The question arises: how many different modes, i.e. spin projections, may graviton pairs have? We consider that the background of initial gravitons consists of two modes. For massless transverse bosons, it takes place as by spin $`1`$ as by spin $`2.`$ If graviton pairs have maximum spin $`2,`$ then single gravitons should have spin $`1.`$ But from such particles one may constitute four combinations: $`,`$ (with total spin $`2`$), and $`,`$ (with total spin $`0).`$ All these four combinations will be equiprobable if spin projections $``$ and $``$ are equiprobable in a flat wave (without taking into account a probable spin-spin interaction).
But it is happened that, if expression (24) is true, it follows from the energy conservation law that composite gravitons should be distributed only in two modes. So as
$$\underset{x0}{lim}\frac{\overline{n}_2}{\overline{n}}=1/2,$$
(25)
then by $`x0`$ we have $`2\overline{n}_2=\overline{n},`$ i.e. all of gravitons are pairing by low frequencies. An average energy on every mode of pairing gravitons is equal to $`2\mathrm{}\omega \overline{n}_2,`$ the one on every mode of single gravitons - to $`\mathrm{}\omega \overline{n}.`$ These energies are equal by $`x0,`$ because of that, the numbers of modes are equal, too, if the background is in the thermodynamic equilibrium with surrounding bodies.
The above reasoning does not allow to choose a spin value $`2`$ or $`0`$ for composite gravitons. A choice of namely spin $`2`$ would ensure the following proposition: all of gravitons in one realization of flat wave have the same spin projections. From another side, a spin-spin interaction would cause it.
The spectrum of composite gravitons is also the Planckian one, but with a smaller temperature; it has the view:
$$f_2(2\omega ,T)d\omega =\frac{\omega ^2}{4\pi ^2c^2}\frac{2\mathrm{}\omega }{\mathrm{exp}(2x)1}d\omega \frac{(2\omega )^2}{32\pi ^2c^2}\frac{2\mathrm{}\omega }{\mathrm{exp}(2x)1}d(2\omega ).$$
(26)
It means that an absolute luminosity for the sub-system of composite gravitons is equal to:
$$_0^{\mathrm{}}f_2(2\omega ,T)d(2\omega )=\frac{1}{8}\sigma T^4,$$
(27)
where $`\sigma `$ is the Stephan-Boltzmann constant; i.e. an equivalent temperature of this sub-system is
$$T_2(1/8)^{1/4}T=\frac{2^{1/4}}{2}T=0.5946T.$$
(28)
The portion of pairing gravitons, $`2\overline{n}_2/\overline{n},`$ a spectrum of single gravitons, $`f(x),`$ and a spectrum of subsystem of pairing gravitons, $`f_2(2x),`$ are shown on Fig. 5 as functions of the dimensionless parameter $`x\mathrm{}\omega /kT`$.
It is important that the graviton pairing effect does not change computed values of the Hubble constant and of anomalous deceleration of massive bodies: twice decreasing of a sub-system particle number due to the pairing effect is compensated with twice increasing the cross-section of interaction of a photon or any body with such the composite gravitons. Non-pairing gravitons with spin $`1`$ give also its contribution in values of redshifts, an additional relaxation of light intensity due to non-forehead collisions with gravitons, and anomalous deceleration of massive bodies moving relative to the background.
### 4.3 Computation of the Newton constant, and a connection between the two fundamental constants, $`G`$ and $`H`$
If running graviton pairs ensure for two bodies an attractive force $`F_2,`$ then a repulsive force due to re-emission of gravitons of a pair alone will be equal to $`F_2^{^{}}=F_2/2.`$ It follows from that the cross-section for single additional scattered gravitons of destructed pairs will be twice smaller than for pairs themselves (the leading factor $`2\mathrm{}\omega `$ for pairs should be replaced with $`\mathrm{}\omega `$ for single gravitons). For pairs, we introduce here the cross-section $`\sigma (E_2,<ϵ_2>),`$ where $`<ϵ_2>`$ is an average pair energy with taking into account a probability of that in a realization of flat wave a number of graviton pairs may be equal to zero, and that not all of graviton pairs ride at a body ($`<ϵ_2>`$ is an analog of $`<ϵ>`$). This equality is true in neglecting with small allowances which are proportional to $`D^3/r^4`$ (see Section 4.4). Replacing $`\overline{n}\overline{n}_2,\mathrm{}\omega 2\mathrm{}\omega ,`$ and $`P(n,x)P(n,2x),`$ where $`P(0,2x)=\mathrm{exp}(\overline{n}_2),`$ we get for graviton pairs:
$$<ϵ_2>2\mathrm{}\omega (1P(0,2x))\overline{n}_2^2\mathrm{exp}(\overline{n}_2).$$
(29)
This expression does not take into account only that beside pairs there may be single gravitons in a realization of flat wave. To reject cases when, instead of a pair, a single graviton runs against a body (a contribution of such gravitons in attraction and repulsion is the same), we add the factor $`P(0,x)`$ into $`<ϵ_2>:`$
$$<ϵ_2>=2\mathrm{}\omega (1P(0,2x))\overline{n}_2^2\mathrm{exp}(\overline{n}_2)P(0,x).$$
(30)
Then a force of attraction of two bodies due to pressure of graviton pairs, $`F_2`$, - in the full analogy with (19) - will be equal to <sup>2</sup><sup>2</sup>2In initial version of this paper, factor 2 was lost in the right part of Eq. (31), and the theoretical values of $`D`$ and $`H`$ were overestimated of $`\sqrt{2}`$ times:
$$F_2=_0^{\mathrm{}}\frac{\sigma (E_2,<ϵ_2>)}{4\pi r^2}4\sigma (E_1,<ϵ_2>)\frac{1}{3}\frac{4f_2(2\omega ,T)}{c}𝑑\omega =$$
(31)
$$\frac{8}{3}\frac{D^2c(kT)^6m_1m_2}{\pi ^3\mathrm{}^3r^2}I_2,$$
where
$$I_2_0^{\mathrm{}}\frac{x^5(1\mathrm{exp}((\mathrm{exp}(2x)1)^1))^2(\mathrm{exp}(2x)1)^5}{\mathrm{exp}(2(\mathrm{exp}(2x)1)^1)\mathrm{exp}(2(\mathrm{exp}(x)1)^1)}𝑑x=$$
(32)
$$2.318410^6.$$
The difference $`F`$ between attractive and repulsive forces will be equal to:
$$FF_2F_2^{^{}}=\frac{1}{2}F_2G_2\frac{m_1m_2}{r^2},$$
(33)
where the constant $`G_2`$ is equal to:
$$G_2\frac{4}{3}\frac{D^2c(kT)^6}{\pi ^3\mathrm{}^3}I_2.$$
(34)
Both $`G_1`$ and $`G_2`$ are proportional to $`T^6`$ (and $`HT^5,`$ so as $`\overline{ϵ}T`$).
If one assumes that $`G_2=G,`$ then it follows from (34) that by $`T=2.7K`$ the constant $`D`$ should have the value:
$$D=0.79510^{27}m^2/eV^2.$$
(35)
An average graviton energy of the background is equal to:
$$\overline{ϵ}_0^{\mathrm{}}\mathrm{}\omega \frac{f(\omega ,T)}{\sigma T^4}𝑑\omega =\frac{15}{\pi ^4}I_4kT,$$
(36)
where
$$I_4_0^{\mathrm{}}\frac{x^4dx}{\mathrm{exp}(x)1}=24.866$$
(it is $`\overline{ϵ}=8.9810^4eV`$ by $`T=2.7K`$).
We can use (8) and (34) to establish a connection between the two fundamental constants, $`G`$ and $`H`$, under the condition that $`G_2=G.`$ We have for $`D:`$
$$D=\frac{2\pi H}{\overline{ϵ}\sigma T^4}=\frac{2\pi ^5H}{15k\sigma T^5I_4};$$
(37)
then
$$G=G_2=\frac{4}{3}\frac{D^2c(kT)^6}{\pi ^3\mathrm{}^3}I_2=\frac{64\pi ^5}{45}\frac{H^2c^3I_2}{\sigma T^4I_4^2}.$$
(38)
So as the value of $`G`$ is known much better than the value of $`H,`$ let us express $`H`$ via $`G:`$
$$H=(G\frac{45}{64\pi ^5}\frac{\sigma T^4I_4^2}{c^3I_2})^{1/2}=2.1410^{18}s^1,$$
(39)
or in the units which are more familiar for many of us: $`H=66.875kms^1Mpc^1.`$
This value of $`H`$ is in the good accordance with the majority of present astrophysical estimations (for example, the estimate $`(72\pm 8)`$ km/s/Mpc has been got from SN1a cosmological distance determinations in ), but it is lesser than some of them and than it follows from the observed value of anomalous acceleration of Pioneer 10 .
### 4.4 Restrictions on a geometrical language in gravity
The described quantum mechanism of classical gravity gives Newton’s law with the constant $`G_2`$ value (34) and the connection (38) for the constants $`G_2`$ and $`H.`$ We have obtained the rational value of $`H`$ (39) by $`G_2=G,`$ if the condition of big distances is fulfilled:
$$\sigma (E_2,<ϵ>)4\pi r^2.$$
(40)
Because it is known from experience that for big bodies of the solar system, Newton’s law is a very good approximation, one would expect that the condition (40) is fulfilled, for example, for the pair Sun-Earth. But assuming $`r=1AU`$ and $`E_2=m_{}c^2,`$ we obtain assuming for rough estimation $`<ϵ>\overline{ϵ}:`$
$$\frac{\sigma (E_2,<ϵ>)}{4\pi r^2}410^{12}.$$
It means that in the case of interaction of gravitons or graviton pairs with the Sun in the aggregate, the considered quantum mechanism of classical gravity could not lead to Newton’s law as a good approximation. This ”contradiction” with experience is eliminated if one assumes that gravitons interact with ”small particles” of matter - for example, with atoms. If the Sun contains of $`N`$ atoms, then $`\sigma (E_2,<ϵ>)=N\sigma (E_a,<ϵ>),`$ where $`E_a`$ is an average energy of one atom. For rough estimation we assume here that $`E_a=E_p,`$ where $`E_p`$ is a proton rest energy; then it is $`N10^{57},`$ i.e. $`\sigma (E_a,<ϵ>)/4\pi r^210^{45}1.`$
This necessity of ”atomic structure” of matter for working the described quantum mechanism is natural relative to usual bodies. But would one expect that black holes have a similar structure? If any radiation cannot be emitted with a black hole, a black hole should interact with gravitons as an aggregated object, i.e. the condition (40) for a black hole of sun mass has not been fulfilled even at distances $`10^6AU.`$
For bodies without an atomic structure, the allowances, which are proportional to $`D^3/r^4`$ and are caused by decreasing a gravitonic flux due to the screening effect, will have a factor $`m_1^2m_2`$ or $`m_1m_2^2.`$ These allowances break the equivalence principle for such the bodies.
For bodies with an atomic structure, a force of interaction is added up from small forces of interaction of their ”atoms”:
$$FN_1N_2m_a^2/r^2=m_1m_2/r^2,$$
where $`N_1`$ and $`N_2`$ are numbers of atoms for bodies $`1`$ and $`2`$. The allowances to full forces due to the screening effect will be proportional to the quantity: $`N_1N_2m_a^3/r^4,`$ which can be expressed via the full masses of bodies as $`m_1^2m_2/r^4N_1`$ or $`m_1m_2^2/r^4N_2.`$ By big numbers $`N_1`$ and $`N_2`$ the allowances will be small. The allowance to the force $`F,`$ acting on body $`2,`$ will be equal to:
$$\mathrm{\Delta }F=\frac{1}{2N_2}_0^{\mathrm{}}\frac{\sigma ^2(E_2,<ϵ_2>)}{(4\pi r^2)^2}4\sigma (E_1,<ϵ_2>)\frac{1}{3}\frac{4f_2(2\omega ,T)}{c}𝑑\omega =$$
(41)
$$\frac{2}{3N_2}\frac{D^3c^3(kT)^7m_1m_2^2}{\pi ^4\mathrm{}^3r^4}I_3,$$
(for body $`1`$ we shall have the similar expression if replace $`N_2N_1,m_1m_2^2m_1^2m_2`$), where
$$I_3_0^{\mathrm{}}\frac{x^6(1\mathrm{exp}((\mathrm{exp}(2x)1)^1))^3(\mathrm{exp}(2x)1)^7}{\mathrm{exp}(3(\mathrm{exp}(x)1)^1)}𝑑x=1.098810^7.$$
Let us find the ratio:
$$\frac{\mathrm{\Delta }F}{F}=\frac{DE_2kT}{N_22\pi r^2}\frac{I_3}{I_2}.$$
(42)
Using this formula, we can find by $`E_2=E_{},r=1AU:`$
$$\frac{\mathrm{\Delta }F}{F}10^{46}.$$
(43)
An analogical allowance to the force $`F_1`$ has by the same conditions the order $`10^{48}F_1,`$ or $`10^{45}F.`$ One can replace $`E_p`$ with a rest energy of very big atom - the geometrical approach will left a very good language to describe the solar system. We see that for bodies with an atomic structure the considered mechanism leads to very small deviations from Einstein’s equivalence principle, if the condition (40) is fulfilled for microparticles, which prompt interact with gravitons.
For small distances we shall have:
$$\sigma (E_2,<ϵ>)4\pi r^2.$$
(44)
It takes place by $`E_a=E_p,<ϵ>10^3eV`$ for $`r10^{11}m.`$ This quantity is many orders larger than the Planck length. The equivalence principle should be broken at such distances.
Under the condition (44), big digressions from Newton’s law will be caused with two factors: 1) a screening portion of a running flux of gravitons is not small and it should be taken into account by computation of the repulsive force; 2) a value of this portion cannot be defined by the expression (13).
Instead of (13), one might describe this portion at small distances with an expression of the kind:
$$\frac{1}{2}(1+\sigma (E_a,<ϵ>)/\pi r^2(1+\sigma (E_a,<ϵ>)/\pi r^2)^{1/2})$$
(45)
(the formula for a spheric segment area is used here ). Formally, by $`\sigma (E_a,<ϵ>)/\pi r^2\mathrm{}`$ we shall have for the portion (45):
$$\frac{1}{2}(\sigma (E_a,<ϵ>)/\pi r^2(\sigma (E_a,<ϵ>)/\pi )^{1/2}/r),$$
where the second term shows that the interaction should be weaker at small distances. We might expect that a screening portion may tend to a fixing value at super-short distances, and it will be something similar to asymptotic freedom of strong interactions. But, of course, at such distances the interaction will be super-strong and our naive approach would be not valid.
## 5 Some cosmological consequences of the model
If the described model of redshifts is true, what is a picture of the universe? It is interesting that in a frame of this model, every observer has two own spheres of observability in the universe (two different cosmological horizons exist for any observer) . One of them is defined by maximum existing temperatures of remote sources - by big enough distances, all of them will be masked with the CMB radiation. Another, and much smaller, sphere depends on their maximum luminosity - the luminosity distance increases with a redshift much quickly than the geometrical one. The ratio of the luminosity distance to the geometrical one is the quickly increasing function of $`z:`$
$$D_L(z)/r(z)=(1+z)^{(1+b)/2},$$
(46)
which does not depend on the Hubble constant. An outer part of the universe will drown in a darkness.
By the found theoretical value of the Hubble constant: $`H=2.1410^{18}s^1`$ (then a natural light unit of distances is equal to $`1/H14.85`$ light GYR), plots of two theoretical functions of $`z`$ in this model - the geometrical distance $`r(z)`$ and the luminosity distance $`D_L(z)`$ \- are shown on Fig. 6 . As one can see, for objects with $`z10`$, which are observable now, we should anticipate geometrical distances of the order $`35`$ light GYR and luminosity distances of the order $`1555`$ light GYR in a frame of this model. An estimate of distances to objects with given $`z`$ is changed, too: for example, the quasar with $`z=5.8`$ should be in a distance approximately of 2.8 times bigger than the one expected in the model based on the Doppler effect.
We can assume that the graviton background and the cosmic microwave one are in a state of thermodynamical equilibrium, and have the same temperatures. CMB itself may arise as a result of cooling any light radiation up to reaching this equilibrium. Then it needs $`z1000`$ to get through the very edge of our cosmic ”ecumene” (see Fig. 7).
Some other possible cosmological consequences of an existence of the graviton background were described in . Observations of last years give us strong evidences for supermassive and compact objects (named now supermassive black holes) in active and normal galactic nuclei . Massive nuclear ”black holes” of $`10^610^9`$ solar masses may be responsible for the energy production in quasars and active galaxies . In a frame of this model, an existence of black holes contradicts to the equivalence principle. It means that these objects should have another nature; one must remember that we know only that these objects are supermassive and compact.
There should be two opposite processes of heating and cooling the graviton background which may have a big impact on cosmology. Unlike models of expanding universe, in any tired light model one has a problem of utilization of energy, lost by radiation of remote objects. In the considered model, a virtual graviton forms under collision of a photon with a graviton of the graviton background. It should be massive if an initial graviton transfers its total momentum to a photon; it follows from the energy conservation law that its energy $`ϵ^{^{}}`$ must be equal to $`2ϵ`$ if $`ϵ`$ is an initial graviton energy. In force of the uncertainty relation, one has for a virtual graviton lifetime $`\tau :`$ $`\tau \mathrm{}/ϵ^{^{}},`$ i.e. for $`ϵ^{^{}}10^4eV`$ it is $`\tau 10^{11}s.`$ In force of conservation laws for energy, momentum and angular momentum, a virtual graviton may decay into no less than three real gravitons. In a case of decay into three gravitons, its energies should be equal to $`ϵ,ϵ^{^{\prime \prime }},ϵ{}_{}{}^{\prime \prime \prime },`$ with $`ϵ^{^{\prime \prime }}+ϵ{}_{}{}^{\prime \prime \prime }=ϵ.`$ So, after this decay, two new gravitons with $`ϵ^{^{\prime \prime }},ϵ{}_{}{}^{\prime \prime \prime }<ϵ`$ inflow into the graviton background. It is a source of adjunction of the graviton background.
From another side, an interaction of gravitons of the background between themselves should lead to the formation of virtual massive gravitons, too, with energies less than $`ϵ_{min}`$ where $`ϵ_{min}`$ is a minimal energy of one graviton of an initial interacting pair. If gravitons with energies $`ϵ^{^{\prime \prime }},ϵ^{\prime \prime \prime }`$ wear out a file of collisions with gravitons of the background, its lifetime increases. In every such a collision-decay cycle, an average energy of ”redundant” gravitons will double decrease, and its lifetime will double increase. Only for $`93`$ cycles, a lifetime will increase from $`10^{11}s`$ to $`10`$ Gyr. Such virtual massive gravitons, with a lifetime increasing from one collision to another, would duly serve dark matter particles. Having a zero (or near to zero) initial velocity relative to the graviton background, the ones will not interact with matter in any manner excepting usual gravitation. An ultra-cold gas of such gravitons will condense under influence of gravitational attraction into ”black holes” or other massive objects. Additionally to it, even in absence of initial heterogeneity, the one will easy arise in such the gas that would lead to arising of super compact massive objects, which will be able to turn out ”germs” of ”black holes”. It is a method ”to cool” the graviton background.
So, the graviton background may turn up ”a perpetual engine” of the universe, pumping energy from any radiation to massive objects. An equilibrium state of the background will be ensured by such a temperature $`T,`$ for which an energy profit of the background due to an influx of energy from radiation will be equal to a loss of its energy due to a catch of virtual massive gravitons with ”black holes” or other massive objects. In such the picture, the chances are that ”black holes” would turn out ”germs” of galaxies. After accumulation of a big enough energy by a ”black hole” (to be more exact, by a super-compact massive object) by means of a catch of virtual massive gravitons, the one would be absolved from an energy excess in via ejection of matter, from which stars of galaxy should form. It awaits to understand else in such the approach how usual matter particles form from virtual massive gravitons.
There is a very interesting but non-researched possibility: due to relative decreasing of an intensity of graviton pair flux in an internal area of galaxies (pairs are destructed under collisions with matter particles), the effective Newton constant may turn out to be running on galactic scales. It might lead to something like to the modified Newtonian dynamics (MOND) by Mordehai Milgrom (about MOND, for example, see ). But to evaluate this effect, one should take into account a relaxation process for pairs, about which we know nothing today. It is obvious only that gravity should be stronger on a galactic periphery. The renormalization group approach to gravity leads to modifications of the theory of general relativity on galactic scales , and a growth of Newton’s constant at large distances takes place, too. Kepler’s third law receives quantum corrections that may explain the flat rotation curves of the galaxies.
## 6 How to verify the main conjecture of this approach in a laser experiment on the Earth
I would like to show here (see ) a full realizability at present time of verifying my basic conjecture about the quantum gravitational nature of redshifts in a ground-based laser experiment. Of course, many details of this precision experiment will be in full authority of experimentalists.
It was not clear in 1995 how big is a temperature of the graviton background, and my proposal to verify the conjecture about the described local quantum character of redshifts turned out to be very rigid: a laser with instability of $`10^{17}`$ hasn’t appeared after 9 years. But if $`T=2.7K`$, the satellite of main laser line of frequency $`\nu `$ after passing the delay line will be red-shifted at $`10^3`$ eV/h and its position will be fixed (see Fig. 8). It will be
caused by the fact that on a very small way in the delay line only a small part of photons may collide with gravitons of the background. The rest of them will have unchanged energies. The center-of-mass of laser radiation spectrum should be shifted proportionally to a photon path. Then due to the quantum nature of shifting process, the ratio of satellite’s intensity to main line’s intensity should have the order:
$$\frac{h\nu }{\overline{ϵ}}\frac{H}{c}l,$$
where $`l`$ is a path of laser photons in a vacuum tube of delay line. It gives us a possibility to plan a laser-based experiment to verify the basic conjecture of this approach with much softer demands to the equipment. An instability of a laser of a power $`P`$ must be only $`10^3`$ if a photon energy is of $`1eV`$. It will be necessary to compare intensities of the red-shifted satellite at the very beginning of the path $`l`$ and after it. Given a very low signal-to-noise ratio, one could use a single photon counter to measure the intensities. When $`q`$ is a quantum output of a cathode of the used photomultiplier (a number of photoelectrons is $`q`$ times smaller than a number of photons falling to the cathode), $`N_n`$ is a frequency of its noise pulses, and $`n`$ is a desired ratio of a signal to noise’s standard deviation, then an evaluated time duration $`t`$ of data acquisition would have the order:
$$t=\frac{\overline{ϵ}^2c^2}{H^2}\frac{n^2N_n}{q^2P^2l^2}.$$
(47)
Assuming $`n=10,N_n=10^3s^1,q=0.3,P=100mW,l=100m,`$ we would have the estimate: $`t=200,000`$ years, that is unacceptable. But given $`P=300W`$, we get: $`t8`$ days, that is acceptable for the experiment of such the potential importance. Of course, one will rather choose a bigger value of $`l`$ by a small laser power forcing a laser beam to whipsaw many times between mirrors in a delay line - it is a challenge for experimentalists.
## 7 Gravity in a frame of non-linear and non-local QED? - the question only to the Nature
From thermodynamic reasons, it is assumed here that the graviton background has the same temperature as the microwave background. Also it follows from the condition of detail equilibrium, that both backgrounds should have the Planckian spectra. Composite gravitons will have spin $`2`$, if single gravitons have the same spin as photons. The question arise, of course: how are gravitons and photons connected? Has the conjecture by Adler et al. (that a graviton with spin $`2`$ is composed with two photons) chances to be true? Intuitive demur calls forth a huge self-action, photons should be endued with which if one unifies the main conjecture of this approach with the one by Adler et al. - but one may get a unified theory on this way.
To verify this combined conjecture in experiment, one would search for transitions in interstellar gas molecules caused by the microwave background, with an angular momentum change corresponding to absorption of spin $`2`$ particles (photon pairs). A frequency of such the transitions should correspond to an equivalent temperature of the sub-system of these composite particles $`T_2=0.5946T,`$ if $`T`$ is a temperature of the microwave background.
From another side, one might check this conjecture in a laser experiment, too. Taking two lasers with photon energies $`h\nu _1`$ and $`h\nu _2`$, one may force laser beams to collide on a way $`L`$ (see Fig. 9). If photons are self-interacting particles, we might wait that photons with energies $`h\nu _1h\nu _2`$, if $`h\nu _1>h\nu _2`$, would arise after collisions of initial photons.
If we assume (only here) that single gravitons are identical to photons, it will be necessary to take into account the following circumstances to calculate an analog of the Hubble constant for this experiment: an average graviton energy should be replaced with $`h\nu _2`$, the factor $`1/2\pi `$ in (8) should be replaced with $`1/\phi `$, where $`\phi `$ is a divergence of laser beam 2, and one must use a quantity $`P/S`$ instead of $`\sigma T^4`$ in (8), where $`P`$ is a laser 2 power and $`S`$ is a cross-section of its beam. Together all it means that we should replace the Hubble constant with its analog for a laser beam collision, $`H_{laser}`$:
$$HH_{laser}=\frac{1}{\phi }Dh\nu _2\frac{P}{S}.$$
(48)
Taken $`\phi =10^4`$, $`h\nu _21eV`$, $`P10mW`$, and $`P/S10^3W/m^2`$, that is characterizing a He-Ne laser, we get the estimate: $`H_{laser}0.06s^1`$. Then photons with energies $`h\nu _1h\nu _2`$ would fall to a photoreceiver with a frequency which should linearly rise with $`L`$ (proportionally to $`\frac{H_{laser}}{c}L`$), and it would be of $`10^7s^1`$ if both lasers have equal powers $`10mW`$, and $`L1m`$. It is a big enough frequency to give us a possibility to detect easy a flux of these expected photons in IR band.
I think there is not any sense to try to analyze theoretically consequences of this conjecture - it will be easier to verify it experimentally. The Nature may answer the question if we ask correctly. All that was said in the above sections doesn’t depend on the answer, but it would be very important for our understanding of known interactions. If this tentative non-linear vacuum effect exists, it would lead us far beyond standard quantum electrodynamics to take into account new non-linearities (which are not connected with electron-positron pair creation) and an essential impact of such a non-locally born object as the graviton background.
## 8 Conclusion
It follows from the above consideration that the geometrical description of gravity should be a good idealization for any pair of bodies at a big distance by the condition of an ”atomic structure” of matter. This condition cannot be accepted only for black holes which must interact with gravitons as aggregated objects. In addition, the equivalence principle is roughly broken for black holes, if the described quantum mechanism of classical gravity is realized in the nature. Because attracting bodies are not initial sources of gravitons, a future theory must be non-local in this sense to describe gravitons running from infinity. Non-local models were considered by G.V. Efimov in his book . The Le Sage’s idea to describe gravity as caused by running ab extra particles was criticized by the great physicist Richard Feynman in his public lectures at Cornell University , but the Pioneer 10 anomaly , perhaps, is a good contra argument pro this idea.
The described quantum mechanism of classical gravity is obviously asymmetric relative to the time inversion. By the time inversion, single gravitons would run against bodies to form pairs after collisions with bodies. It would lead to replacing a body attraction with a repulsion. But such the change will do impossible the graviton pairing. Cosmological models with the inversion of the time arrow were considered by Sakharov . Penrose has noted that a hidden physical law may determine the time arrow direction ; it will be very interesting if namely realization in the nature of Newton’s law determines this direction.
A future theory dealing with gravitons as usual particles should have a number of features which are not characterizing any existing model to image the considered here features of the possible quantum mechanism of gravity. If this mechanism is realized in the nature, both the general relativity and quantum mechanics should be modified. Any divergencies, perhaps, would be not possible in such the model because of natural smooth cut-offs of the graviton spectrum from both sides. Gravity at short distances, which are much bigger than the Planck length, needs to be described only in some unified manner.
|
warning/0506/physics0506022.html
|
ar5iv
|
text
|
# Two-dimensional atomic lithography by sub-micron focusing of atomic beams
## I Introduction
Controlling the motion of neutral atoms using light fields has been an important topic in atomic physics for several decades. Focusing of atoms from a source onto a planar substrate can be used for lithography where the writing is done by an atomic beam instead of a light field. This technique is potentially useful for fabrication of structures with sub-micron resolution. Atomic focusing can be achieved with magnetic or optical fields. In atomic lithography with light fields, an optical profile creates a spatially dependent dipole force that alters the trajectories of neutral atoms. One and two dimensional standing wave light patterns have been used to create periodic atomic patternsTimp et al. (1992); McClelland et al. (1993); Schulze et al. (2000); Bradley et al. (1999); Petra et al. (2004). Imaging of an atomic beam is also possible using a magnetic lens as was demonstrated by Kaenders et al. Kaenders et al. (1995), and a wide range of atomic guiding and imaging tasks have been demonstrated using magnetic fieldsHinds and Hughes (1999).
While there has been a great deal of work in atomic lithography (recent reviews can be found in Refs. Oberthaler and Pfau, 2003; Meschede and Metcalf, 2003; McClelland et al., 2004), only spatially periodic or quasiperiodicJurdik et al. (2004); Schulze et al. (2000) patterns have been demonstrated. As this limits the range of applications and usefulness of the technique there is considerable interest in devising approaches that will allow spatially complex structures to be created. One approach to creating non-periodic patterns is to use a more complex optical field, as in Refs. Mützel et al., 2002, 2003. An alternative serial writing approach is to focus the atomic beam to a very small spot and then move the spot to draw an arbitrary two-dimensional structure. Spot motion can be achieved either by scanning the spot over a stationary substrate, or by moving the substrate. An example of an optically scanned atomic beam focused to a size of about $`200\mu \mathrm{m}`$ can be found in Ref. Oberst et al., 2003. To obtain higher resolution tightly focused atomic beams are necessary which can be created by propagation in hollow core fibersRenn et al. (1995) which have the drawback of low atomic flux, or using Bessel beamsBjorkholm et al. (1978); Balykin and Letokhov (1987); Dubetsky and Berman (1998); Okamoto et al. (2001). A drawback of the Bessel beam approach is the existence of secondary maxima that can lead to atom localization in rings surrounding the central peak.
In this paper we analyze a new approach to focusing an atomic beam to a spot with a characteristic size of about $`100\mathrm{nm}`$. The optical profile used for the atomic focusing is created with a spatial light modulator (SLM) that allows the spot to be scanned with no mechanical motion. By controlling both the phase and intensity profile of the incident beam we create an optical “funnel” that focuses a high percentage of the atomic flux into a single spot. The proposed approach, as shown in Fig. 1, uses a magneto-optical trap (MOT) as a source of cold atoms. A continuous flow of atoms is pushed out of the MOTLu et al. (1996); Mandonnet et al. (2000); Meschede and Metcalf (2003) and collimated using a magnetic guide followed by an optical focusing region. The magnetic waveguide creates a micron sized atomic beam, with the final focusing down to a full width at half maximum atomic spot size of $`w_{a,\mathrm{FWHM}}=110\mathrm{nm}`$ provided by a far detuned optical profile together with near resonant cooling beams. It is then possible to move the optical profile and write a two-dimensional pattern by changing the phases of the laser beams with the SLM. We study the feasibility of this approach using numerical simulations of the atomic trajectories including propagation and cooling in the magnetic guide and the optical profile.
In Sec. II we summarize the main features of optical focusing of atoms, and describe the creation of a Bessel profile using side illumination with a SLM. The optical funnel for reduction of atomic trapping in secondary maxima is described in Sec. III. The design of a cold atom setup coupled to a magnetic waveguide and then followed by the optical funnel is described in Sec. IV. Numerical results showing the feasibility of writing a two-dimensional structure are given in Sec. V.
## II Optical potential
The goal of optically mediated atomic lithography is to control the trajectories of atoms by means of light fields. A collimated atomic beam is passed through a region of spatially varying optical intensity that modifies the atomic trajectories such that a desired atomic pattern is deposited on a substrate. The conservative optical potential for a two-level atom including the effects of saturation isMcClelland (1995)
$$U=\frac{\mathrm{}\mathrm{\Delta }}{2}\mathrm{ln}\left[1+\frac{I}{I_s}\frac{1}{(1+4\mathrm{\Delta }^2/\gamma ^2)}\right],$$
(1)
where $`I_s`$ is the saturation intensity, $`\gamma `$ is the natural linewidth, $`\mathrm{\Delta }=\omega \omega _a`$ is the detuning from resonance, $`\omega `$ is the optical frequency, and $`\omega _a`$ is the atomic transition frequency. We write the intensity as $`I=I(x,y)g(z)`$ where $`I(x,y)`$ gives the dependence in the $`x,y`$ plane and $`g(z)`$ is an envelope function which describes the intensity profile along the $`z`$ axis, which we will take to be the propagation direction of the atomic beam. Atoms propagating through a region of spatially varying intensity experience a dipole force $`𝐅=U`$ which alters their trajectories, and can be used to focus the atoms into a desired pattern. When $`\mathrm{\Delta }<0`$ (red detuning) we get an attractive potential that concentrates the atoms where the intensity is highest, while for $`\mathrm{\Delta }>0`$ (blue detuning) the potential is repulsive. For potentials of interest we calculate the atomic trajectories numerically using the classical equations of motion for the atomic center of mass. It is also assumed that the atoms do not collide and only interact with the given potential. Therefore, each atom trajectory can be treated individually and a large number of single atom trajectories resulting from a statistical distribution of initial conditions can be combined to determine an output distribution.
The optical potential can be constructed by combining several laser beams. When only a few beams are used the potential has a periodic structure, e.g. a one-dimensional standing wave created by two counter propagating beams, or a checkerboard pattern created by four beams. To focus all of the atoms to a single spot, the periodic structure must be removed. This can be done by adding more laser fields. Consider a two-dimensional field formed using $`N`$ laser beams all propagating in the same plane and arranged to cross at a common point. The shape of the resulting intensity profile is determined by the angles between the beams as well as the magnitude and phase of the fields. The simplest possibility is to cross laser beams, which all have equal electric field phases and magnitudes, with equal angular spacing. As the number of beams goes to infinity the intensity profile tends to $`J_0^2(k\sqrt{x^2+y^2}),`$ the square of the zeroth order Bessel function which has a ring structure whose scale is dependent only on the wavelength of the light through $`k=2\pi /\lambda .`$ An axicon can be used to create a Bessel beam for this purposeDubetsky and Berman (1998), and it is possible to create higher order Bessel profiles such as a $`J_1`$ profile as proposed by Okamoto et al Okamoto et al. (2001), by altering the phase profile of the beam.
An alternative approach to creating a Bessel beam, as well as higher order beams, is to use a SLM. There has been substantial recent interest in using SLM technology in atom optics McGloin et al. (2003) as well as an experimental demonstration of manipulation of atoms in microscopic optical trapsBergamini et al. (2004). Superpositions of Bessel functions may also be useful for addressing individual atoms in optical latticesSaffman (2004). Referring to the geometry of Fig. 2, when the number of laser beams $`N>32`$ the pattern is periodic on scales much longer than the size of the central Bessel peak, so we obtain a well isolated Bessel profile. After the optical profile is constructed, serial writing of a pattern can be accomplished by translating the central spot where atoms are focused. The proposed method to accomplish this is to change the phase of each of the individual beams to construct the profile at a new point on the plane. Each beam with index $`j=(1,N)`$ can be represented as an electric field $`E_j=E_0\mathrm{exp}\{ik[\mathrm{cos}(\theta _j)x+\mathrm{sin}(\theta _j)y]i\varphi _j\}`$ where $`E_0`$ is the amplitude, $`k`$ is the wavenumber, $`\varphi _j`$ is a phase, and $`\theta _j`$ is the angle of the beam propagation direction with respect to the $`x`$ axis. We assume all beams are polarized along $`\widehat{z}`$ so we can neglect vectorial effects. To create a Bessel profile centered at $`x=y=0`$ we choose all $`\varphi _j=0.`$ To move the profile to be centered at coordinates $`(x_0,y_0)`$ we put
$$\varphi _j=k[\mathrm{cos}(\theta _j)x_0+\mathrm{sin}(\theta _j)y_0].$$
(2)
This enables translation of the pattern as shown in Fig. 3. A spatial light modulator can readily be used to create these phase shifts in a side illumination geometry which is compatible with deposition of an atomic beam, as shown in Fig. 2.
The results of a numerical simulation of atom focusing using a $`J_0^2`$ Bessel profile can be seen in Fig. 4. In order to reduce heating that occurs when the atoms enter the focusing region two-dimensional near resonant molasses beams with the same axial profile as the focusing potential are included. Details of the numerical method, including parameters of the cooling beams, are given in Sec. V. A serious problem with this approach is that the atoms are not focused to a single spot. It is difficult to obtain a large atomic flux in a beam that is narrow enough to prevent focusing into the surrounding ring structure. To correct this, the rings of the Bessel function need to be removed. One possibility is to superimpose a red detuned $`J_0`$ profile with a blue detuned and repulsive $`J_1`$ (or higher order) profile. The wavelengths and amplitudes of the two Bessel beams can be chosen to suppress the ring structure. While the first ring can be suppressed, higher order rings are still present and the blue detuned Bessel profile needs to have a very large detuning. This is because the first maximum of higher order Bessel profiles is not at the same radius as the first ring of the zeroth order Bessel. For example, the third order Bessel has its first maximum at $`0.668\lambda `$ while the first ring of the zeroth order Bessel is at $`0.61\lambda `$. Therefore to get the first maximum to line up with the first ring, we must change the wavelength to $`776\mathrm{nm}`$, as compared to the $`852\mathrm{nm}`$ used for $`J_0`$ focusing with Cs atoms. Since the potential is proportional to $`1/\mathrm{\Delta }`$ for large detuning, the power required to obtain the correct well depth to cancel the first order ring is extremely large.
## III Optical Funnel
In this section we discuss an alternative profile that uses traveling waves to avoid the ring structure that mars the applicability of the Bessel profile. We call this structure an optical funnel. The central spot of the Bessel squared profile has a FWHM diameter $`w_{\mathrm{FWHM}}=0.359\lambda `$. This small width may be relaxed in exchange for a profile that has a less troublesome ring structure. One possible optical profile is the funnel shown in Fig. 5. The funnel is a traveling wave field that subtends an angular range of $`\pi `$ and has amplitudes that decrease linearly on either side of the maximum. The radius of the rim of the funnel depends on how many beams are used to create the profile, as shown in Fig. 6. All atoms that enter the profile inside the rim will be funneled towards the point of lowest energy at the center. The result is a high percentage of the atomic flux being directed into the central spot. The width of the atomic spot that is written then scales as $`w_{a,\mathrm{FWHM}}w_{\mathrm{FWHM}}\sqrt{k_BT_a/U_0}`$, where $`U_0`$ is the maximum well depth of the funnel, and $`T_a`$ is the temperature of the transverse atomic motion. The FWHM of the funnel intensity profile is $`0.48\lambda `$ in $`x`$ and $`1.38\lambda `$ in $`y`$. We can create an approximately circular potential by combining two noninteracting funnels (they have a relative detuning that is large compared to $`\gamma `$, yet small compared to $`\mathrm{\Delta }`$), to get $`w_{\mathrm{FWHM}}=0.72\lambda `$ in both the $`x`$ and $`y`$ directions, as shown in Fig. 7.
Even though the width of the symmetrized funnel profile is approximately twice that of the Bessel profile, this optical potential is preferred since it is possible to focus a large percentage of the atomic flux in a single spot with no rings. Figure 8 shows focusing to a single spot using the same atomic beam parameters as in Fig. 4. For the simulation two noninteracting funnel profiles are placed on top of each other to create a symmetric potential, as described above. The atomic spot in the funnel has $`w_{a,\mathrm{FWHM}}=110\mathrm{nm}`$. Note that the ring structure which is very pronounced when using a Bessel beam has been essentially eliminated. Both optical profiles have depths of approximately $`21\mathrm{mK}`$. For both profiles cooling beams were used together with the optical focusing as discussed in Sec. V. The results of the simulation show that the funnel captures $`94\%`$ of the atoms into the $`2\mu \mathrm{m}\times 2\mu \mathrm{m}`$ square, the size of the figure, compared to only $`45.3\%`$ for the Bessel profile with only 6% captured in the central spot.
It is interesting to compare the localized potential created with the funnel to simply using tightly focused Gaussian beams. The use of Gaussian beams would completely eliminate the background ring problem, but comes at the expense of larger spot size. Experiments with a high numerical aperture lens systemSchlosser et al. (2001) have demonstrated focusing of a single Gaussian beam to a spot diameter of $`w_{\mathrm{FWHM}}=0.86\lambda .`$ A Gaussian beam is highly elongated so we can superpose two incoherent Gaussians to create a symmetric optical potential. Because there is a large degree of elongation, even for such a tightly focused Gaussian, the resulting symmetrized intensity profile has $`w_{\mathrm{FWHM}}=1.4\lambda `$ which is almost twice as big as we obtain for the funnel. Comparing Bessel beams, the optical funnel, and Gaussian beams, we see that the funnel combines a relatively small spot size with large radius rings. In Sec. V we demonstrate that two dimensional structures can be written by translating the funnel profile.
## IV Atomic source and magnetic precollimation
Most experimental demonstrations of atomic lithography have used an oven as a source of thermal atoms. The atomic beam is then collimated using mechanical apertures and/or transverse laser cooling to create a beam suitable for lithography experiments. At least two experiments Fujita et al. (1996); Engels et al. (1999) have also used cold or axially cooled atom sources for lithography experiments. A detailed discussion of the relative merits and requirements of different types of atomic sources for lithography experiments can be found in Ref. Meschede and Metcalf, 2003. Generally speaking oven based sources provide higher flux and therefore faster writing speeds than cold atom sources. One advantage of cold sources is the low longitudinal velocity which minimizes surface damage and sputtering from atom impact on the deposition substrate. We are interested here in a technique that is suitable for writing feature sizes as small as 100 nm. The requirement for high flux, and fast writing over a large area may therefore be less important than minimization of surface effects.
The above considerations motivate us to consider the suitability of a cold atom source using the geometry shown schematically in Fig. 1. In order to match a MOT source with a transverse size of $`0.11.0\mathrm{mm}`$ to the micron sized funnel profile so that secondary rings are eliminated it is necessary to precollimate the atomic beam. We note that this would also be necessary with an oven based source. We propose to do so using a magnetic guide and transverse laser cooling.
The potential $`U(x,y)`$ due to a magnetic field $`B(x,y)`$ in a quadrupole magnetic wave guide is
$`U(x,y)`$ $`=`$ $`\mu _BgmB(x,y)`$ (3)
$`B(x,y)`$ $`=`$ $`b^{}\sqrt{x^2+y^2}`$ (4)
where $`b^{}=2\mu _0J/(\pi a^2),`$ $`m`$ is the magnetic quantum number, $`g`$ is the Landé factor, $`\mu _B`$ is the Bohr magneton, $`\mu _0`$ is the magnetic permeability, $`J`$ is the current, and $`a`$ is the distance from the center of the guide to each wire. As indicated in Fig. 1 two-dimensional molasses beams are used to cool the atoms in the magnetic guide. The resulting size of the atomic beam after cooling is given by the virial theorem to be
$$<r>=\frac{k_BT_a}{gm\mu _Bb^{}}.$$
(5)
Using $`T_a=22.5\mu \mathrm{K}`$, $`J=500\mathrm{A}`$ and $`a=4\mathrm{mm}`$, the average radial distance of the atoms from the center of the guide is $`1.34\mu \mathrm{m}`$.
The magnetic potential is attractive for atoms that are prepared in states with positive $`m.`$ Since the atoms pass close to the trap axis, we must add a bias field along the longitudinal axis of the guide to minimize Majorana spin flips. The result of this bias field will be a slightly larger atomic beam. The simulations in the next section were done with a bias field $`B_0`$ to give a total magnetic field $`B(x,y)=\sqrt{B_0^2+b^^2(x^2+y^2)}`$. Simulations show that adding a bias field of $`B_0=0.1\mathrm{G}`$ in the axial direction results in a slightly larger average radial distance of $`1.42\mu \mathrm{m}`$ for the above parameters.
## V Numerical Results
Atomic focusing was simulated using a 4th order Runge-Kutta code to trace the trajectories of individual atoms from the injection into the magnetic wave guide through the optical guide. The atomic parameters were chosen to correspond to the Cs D2 line ($`6^2S_{1/2}6^2P_{3/2}`$) transition with decay rate $`\gamma =2\pi \times 5.22\mathrm{MHz}`$. Parameter values consistent with magnetic waveguide experiments were chosen Mandonnet et al. (2000): $`a=4\mathrm{mm}`$ and $`J=500\mathrm{A}`$. A small bias field of $`B_0=0.1\mathrm{G}`$ was included in the simulation as discussed above. The atomic beam from the MOT at the entrance to the $`20\mathrm{cm}`$ long magnetic guide was taken to be a Gaussian distribution with $`1/e^2`$ radius of $`100\mu \mathrm{m}`$, transverse temperature of $`20\mu \mathrm{K},`$ and mean longitudinal velocity of $`14\mathrm{m}/\mathrm{s}.`$ Since compression in the magnetic guide heats the atoms we added two dimensional molasses beams to the simulation. The intensity of the cooling beams was set to $`2.9\mathrm{W}/\mathrm{m}^2`$, or $`I/Is=.26`$, and $`\mathrm{\Delta }_m=\gamma /2`$. The cooling was simulated by randomly changing the momenta of each atom by either $`2\mathrm{}k_m/m_a`$,$`2\mathrm{}k_m/m_a`$, or 0 once per scattering time with probabilities of $`25\%`$, $`25\%`$, and $`50\%`$ respectively while constantly damping the atoms with a forceMetcalf and van der Straten (1999) $`\beta \mathrm{v}`$ . Here, $`k_m`$ is the wavenumber of the molasses beams, $`m_a`$ is the mass of the atom, and $`\beta `$ is the damping coefficient. The velocity kick was added in both transverse directions to independently cool along each axis. The linewidth $`\gamma `$, was decreased by a factor of 8 and the recoil velocity was decreased to 0.23 cm/s to simulate sub-Doppler cooling to approximately $`22.5\mu \mathrm{K}`$ in the transverse plane. The resulting FWHM of the atomic beam after cooling but still in the magnetic wave guide is $`w_{a,\mathrm{FWHM}}1.68\mu \mathrm{m}`$.
At the end of the magnetic wave guide we feed the atoms into the optical funnel. The funnel was comprised of two noninteracting funnel profiles, each consisting of 32 laser beams. The two funnels are overlapped to form a symmetric potential, as in Fig 7. The axial profile of the beams was $`g(z)=\mathrm{exp}(2z^2/w_z^2)`$, with $`w_z=0.6\mathrm{mm}`$ and their cross section was assumed to be circular in the focusing region. A peak intensity of $`7\times 10^6\mathrm{W}/\mathrm{m}^2`$ and $`\mathrm{\Delta }/2\pi =10\mathrm{GHz}`$ was chosen for one funnel and $`6.3\times 10^6\mathrm{W}/\mathrm{m}^2`$ and $`\mathrm{\Delta }/2\pi =9\mathrm{GHz}`$ for the other. This choice of parameters gives the same well depth for both funnels, but detuned such that they do not interact. This results in a total well depth of $`U_0/k_B=21\mathrm{mK}`$ with a laser power requirement of approximately 124 and 111 mW respectively.
Since the atoms are heated as they travel into the optical funnel as shown in Fig. 9, we add cooling beams to the optical profile. For the cooling beams, the detuning was set to half the line width and the intensity of each beam was 40 % of the saturation intensity. Two pairs of orthogonal cooling beams were used with the same axial spatial mode as the beams used to create the funnel. The cooling beams help to limit the temperature, but do not interact strongly enough with the atoms to cool them to the Doppler temperature before the atoms hit the substrate. The substrate is placed at the center of the Gaussian profile as has been done in some experimentsLison et al. (1997). The transverse temperature of the atoms just before the substrate is approximately 2.5 mK. Since the atoms are heated to well above the Doppler temperature when entering the optical profile, cooling was simulated solely by a damping force, ignoring photon kicks. The resulting number density in the optical profile has $`w_{a,\mathrm{FWHM}}=110\mathrm{nm}`$ with 82% of the atoms falling within a radius of 500 nm and 20% falling within the FWHM of the beam. The trajectories of 100 atoms are shown in the inset of Fig. 10.
Due to the large intensity of the funnel beams additional heating due to photon scattering may also be of concern. The peak scattering rate at the center of the many beam optical funnel is Metcalf and van der Straten (1999)
$$r\frac{\gamma }{2}\frac{\frac{I_0}{I_s}}{1+\frac{4\mathrm{\Delta }^2}{\gamma ^2}+\frac{I_0}{I_s}}$$
(6)
For the funnel parameters given above this results in a maximum scattering rate of $`r=1.3\times 10^6\mathrm{s}^1.`$ The atom-light interaction time is approximately 0.1 ms and results in scattering of about 150 photons. The scattering of these photons adds $`\frac{m_a}{k_B}(\frac{\sqrt{150}\mathrm{}k}{m_a})^2=30\mu \mathrm{K}`$ to the atomic temperature which can be neglected since the temperature of the atoms at the center of the profile is approximately $`2.5\mathrm{mK}.`$
To simulate writing of a two-dimensional pattern the optical funnel is reconstructed at varying distances from the atomic beam axis using the phases given by Eq. (2). A simulation of a W is shown in Fig. 11. The W is made by positioning the funnel at 101 different spots and depositing a total of 24240 atoms. The FWHM of the atomic beam at the substrate increases the further the optical funnel is from the center of the magnetic guide. This results in a small spreading of the ends of the W, visible in Fig. 11. Figure 12 shows that the FWHM of the atomic beam on the substrate increases by about a factor of two when the funnel is moved by a distance of $`3\mu \mathrm{m}.`$ This is because the atoms that enter the potential near the rim obtain a larger radial velocity while traveling to the potential minimum. As a result, the atoms are hotter when they arrive at the substrate and thus have a larger $`w_{a,\mathrm{FWHM}}.`$
In order for this type of lithography to be practical the writing speed must not be too slow. Coverage of a surface with one monolayer of Cs corresponds to a surface density of about Lison et al. (1997) $`4\times 10^{15}\mathrm{atoms}/\mathrm{cm}^2`$. It has been shownBerggren et al. (1997) that between 3 and 7 monolayers of Cs are needed to create enough damage for exposure of organic self-assembled monolayer (SAM) coatings. Given a spot of $`w_{a,\mathrm{FWHM}}=110\mathrm{nm}`$, which defines a pixel with area $`\pi 55^2=9500\mathrm{nm}^2`$, the exposure per pixel for one monolayer of Cs is $`3.8\times 10^5`$ atoms/pixel. Given a flux of $`5\times 10^8\mathrm{atoms}/\mathrm{s}`$ and 20% of the atoms falling within an area of diameter $`w_{a,\mathrm{FWHM}}`$, it will take about $`3.8\mathrm{ms}`$ to deposit one monolayer of Cs. The time needed to successfully write 3 to 7 monolayers of cesium to a pixel is then between $`11`$ and $`27\mathrm{ms}`$. The W has lengths which total $`4.8\mu \mathrm{m}`$ which corresponds to at least $`4.8\mu \mathrm{m}/110\mathrm{nm}44\mathrm{spots}`$. Therefore, the writing time for the W would be approximately 1 second.
## VI Discussion
We have described an atomic lithography system that uses magnetic and optical fields to focus atoms from a MOT onto a sub-micron spot. Doing so requires combining confining potentials with optical molasses to maintain low atomic temperatures. Using numerical simulations of experimentally realistic parameters we produce pixels with $`w_{a,\mathrm{FWHM}}=110\mathrm{nm}`$, and a writing speed of order $`20\mathrm{ms}`$ per pixel. The position of the pixel can be scanned to write arbitrary planar structures using phase shifts created by a SLM. The optical funnel that produces the final focusing has an acceptance region with diameter of about $`8\mu \mathrm{m}.`$ However, the funnel cannot be scanned that far since when the funnel position moves a distance comparable to the width of the beam leaving the magnetic guide the spot size created by the funnel starts to increase. We can therefore say that for the parameters investigated it appears possible to write no more than several hundred independent pixels with $`w_{a,\mathrm{FWHM}}110\mathrm{nm}`$. Alternatively the system could be optimized to write a single small, but stationary spot, and the substrate scanned mechanically. This would in principle allow an arbitrarily large number of pixels to be written.
A long term goal of atomic lithography is to produce a large scale lithographic process. To do so a number of challenges will have to be overcome. The optical funnel will only capture atoms into a small spot if they enter the funnel close enough to the center. Theoretically, we could increase the axial thickness of the funnel, or increase the laser intensity to increase the range at which the funnel will capture atoms. Experimentally, this is limited by available laser power. Another experimental difficulty will be the sub-Doppler cooling that is required in the magnetic wave guide. Others have shown theoretically that sub-Doppler cooling is possible with small spot sizesBalykin and Minogin (2003), but experiments have not yet demonstrated this level of cooling of a traveling beam in a quadrupole wave guide. Requirements on cooling efficiency could be traded off against stronger compression due to larger magnetic fields. This is limited by the ability to run large currents through small wires. Approaches based on lithographically patterned wiresHinds and Hughes (1999) which will enable waveguides with smaller dimensions may enable even tighter confinement.
In conclusion we expect that solutions to these issues, as well as further optimization of performance by refinement of parameters will be possible. Full evaluation of the suitability of the atom-optical approach for writing complex structures described here will ultimately rely on experimental tests.
The authors thank Deniz Yavuz for helpful discussions. This work was supported by NSF grant PHY-0210357, and an Advanced Opportunity Fellowship from the University of Wisconsin graduate school.
|
warning/0506/cond-mat0506692.html
|
ar5iv
|
text
|
# Influence of dipole interaction on lattice dynamics of crystalline ice
## 1. INTRODUCTION
The vibrational studies of solid ice have gained attraction for scientific investigation over the decades. Experimentally, the vibrational spectra of the different phases of ice have been investigated using infra-red (IR) and Raman techniques, but due to the proton disordering in most such structures, the normal selection rules governing the interaction of radiation with these lattices are broken and hence any analysis of the spectra is difficult. On the other hand, the IR and Raman spectra are very sensitive to the intramolecular modes involving O-H stretching and bending, and less sensitive to the intermolecular modes involving the vibrations where the whole water molecules are moving against each other. Therefore, in normal circumstances, only limited information (the acoustic frequencies in particular) can be obtained in the translational region. Inelastic incoherent neutron scattering (IINS) is only more reliable because its spectrum is directly proportional to the phonon density of states weighed by mean square amplitude associated with each mode and also the selection rule is not involved as all modes are measured simultaneously . Despite the detail information that can be obtained from the present days methods of coherent and incoherent scattering on the vibrational motions of the atoms or molecules of most systems based on the peak positions, intensities and their width, there are still many problems about the ice systems. These problems as mentioned, are usually associated with the the proton disordering. The phonon spectra measurements of Renker on Ice-Ih (D<sub>2</sub>O) made by time of flight using a chopper spectrometer were extensive but not complete in that information above 20 meV in the (001) direction and above 30 meV in other directions is missing, presumably because of a lack of scattering intensity and of the nature of the disordering of the protons in the ice structure.
First-principles calculation of crystal structure and crystal properties is becoming standard technique, and the progress in the methods, algorithms, and computer capabilities allows to study larger systems of solids crystals in which crystalline ice cannot be left out. The method has recently gained ground not only because of its reliability in the study of static and dynamics properties of ice but also some important features such as modeling ordered periodic ice structure , and also to probe the nature of hydrogen bond in different geometries . Its theoretical counterpart such as the classical modeled potential through empirical method had some success in describing some important dynamical features, but to date, there is none capable of describing the ice dynamics and related properties across its whole spectral range and describing certain key spectral features.
There are two techniques currently in use in the first-principles method in the study of lattice dynamics of crystals: The linear response method and the direct method. In the linear response method the dynamical matrix is obtained from the modification of the electron density, via the inverse dielectric matrix. The dielectric matrix is calculated from the eigenfunctions and energy levels of the unperturbed system . It can be determined at any wave vector in the Brillouin zone with the computational effort required comparable to that of a ground state optimization. Only linear effects, such as harmonic phonons, are accessible to this technique. On the other hand, the direct-method is based on the solution of the Kohn-Sham equation and it allows one to study both linear and non-linear effects. The calculations deal with a supercell, which allows explicit account to be taken of any perturbation. This method is rather straightforward computationally and there are a few standard software packages. Within the direct method the phonon frequencies are calculated from Hellmann-Feyman forces generated by the small atomic displacements, one at a time. Hence using the information of the crystal symmetry space group the force constants are derived, and the dynamical matrix is built and diagonalized, and its eigenvalues arranged into phonon dispersion relations. In this way, phonon frequencies at selected high frequencies at high-symmetry points of the Brillouin zone can be calculated . However, when the interaction range ceases to be within the supercell, phonons at all wave vectors are determined exactly. The above statement has to be modified for polar crystals for which the macroscopic electric field splits off the infrared-active optic modes. The long-range part of the Coulomb interaction corresponds to the macroscopic electric field arising from ionic displacements.
Ice is a tetrahedrally covalently bonded polar system whose dipole-dipole interactions give rise to the electric field when they are disturbed. The origin of the splitting is therefore the electrostatic field created by long wavelength modes of vibrations in such crystals. Usually a microscopic electric field influences only the LO modes while TO modes remain unaltered. The field therefore breaks the Born-von Kármán conditions, as a consequence with a direct method only finite wave vector $`𝐤0`$ calculations are possible. The LO/TO splitting has therefore been found by calculating the effective Born charge tensor and electronic dielectric constant introduced into the dynamical matrix in the form of a non-analytical term or by calculating LO modes from elongated supercells as will be discussed below.
In our previous work, we have applied with success the direct method to the calculation of phonon dispersion of ice and its corresponding vibrational density of states . The method reproduces the important features in the translational mode, librational mode, bending as well as the stretching region in comparison to the experimental results. The only ingredient that was missing in the previous work has been explained above, i.e., a macroscopic electric field arising from dipolar interaction which is not taken into account. Despite the huge computational demand of this problem, we did the additional calculation of Born-effective charges which is supplied as ingredients to the direct method to identify the long-range part of interatomic force constants and makes the interpolation of phonon frequencies tractable. Our overall aim is to help resolve the discrepancies in the reported phonon frequencies especially the puzzle behind the LO/TO splitting of some optical modes at $`𝐤=0`$ and provide first principles Born-effective charges and dielectric tensors for direct method phonon calculations.
## 2. METHOD OF CALCULATION
The calculation of phonon dispersion relations were performed with the direct method. The direct method uses the Hellmann-Feymann (HF) forces calculated for the optimized supercell with one atom displaced from equilibrium position, derived from the force constants using the symmetry elements of the space group of the crystal, and calculates phonon frequencies by diagonalizing the dynamical matrix.
In this work, the ice crystal structure optimization and calculation of HF forces have been performed with the density functional theory using the PAW method within the generalized gradient approximation (GGA), as implemented in the Vienna Ab Initio Simulation Package (VASP) software. A unit cell of ice crystal was prepared in a cubic box according to Fig. 1 with 8 molecules of water. All the atomic degrees of freedom were relaxed with high precision. The optimum Monkhorst Pack of $`4\times 4\times 4`$ $`k`$-point was used in addition to the GGA of Perdew-Wang to describe the exchange-correlation and the hydrogen bonding of water. We used a energy cut-off of 500 eV because the 2p valence electrons in oxygen require a large plane wave basis set to span the high energy states described by the wavefunction close to the oxygen nucleus, and also the hydrogen atoms require a larger number of planes waves in order to describe localization of their charges in real space. The relaxed geometry for the unit cell from the initial configurations containing 8 molecules is shown in Fig. 1. The starting geometry of the molecules in the cubic simulation box shown is such that no hydrogen bonds were present but the positions of oxygen atoms follow the tetrahedral geometry. After the relaxation, all the protons perfectly point to the right direction of oxygen atoms and make the required hydrogen bonds necessary as indicated by the dotted lines in Fig. 1 to preserve the tetrahedral orientation of the ice structure. This final structure is in accordance to Bernal Fowler’s rules which are based on the statistical model of ice.
The relaxed geometry is tetragonal with calculated lattice parameters $`a`$ = 6.1568 Å, $`b`$ = 6.1565 Å, $`c`$ = 6.0816Å i.e. with $`c/a`$ ratio $``$ 0.988. The experimental lattice constant reported by Blackman et. al. is 6.35012652 Å for the cubic geometry. The calculations of force constants was carried out by considering a $`3\times 1\times 1`$ supercell containing 24 molecules of water which is obtained by matching 3 tetragonal unit cells. At the first step of the calculation, the PHONON software is used to define the appropriate crystal supercell for use of the direct method. The phonon frequencies $`\omega (𝐤,j)`$ are calculated as square roots of eigenvalues of the supercell dynamical matrix:
$`𝐃^{\mathrm{𝐒𝐂}}(𝐤)e(𝐤,j)`$ $`=`$ $`\omega ^2(𝐤,j)e(𝐤,j),`$ (1)
where the $`e(𝐤,j)`$ are the polarization vectors. The supercell dynamical matrix is defined as
$`𝐃^{\mathrm{𝐒𝐂}}(𝐤,\mu \nu )`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{M_\mu M_\nu }}}{\displaystyle \underset{mSC}{}}𝚽^{\mathrm{𝐒𝐂}}(0,\mu ;m,\nu )`$
$`\times `$ $`exp(2\pi \mathrm{𝐢𝐤}.[𝐑(0,\mu )𝐑(m,\nu )]),`$
where the summation over $`m`$ runs over all atoms of the supercell; $`M_\mu `$, $`M_\nu `$ and $`𝐑(0,\mu )`$, $`𝐑(m,\nu )`$ are atomic masses and equilibrium vectors, respectively; and $`𝐤`$ is the wavevector. The cummulant force constants $`𝚽_{ij}^{\mathrm{𝐒𝐂}}`$ are the sums of terms containing the second derivatives of the ground-state energy with respect to the position vectors of interacting atoms $`i`$ and $`j`$. The HF forces in the direct method are derived using
$`𝐅_i(n,\nu )`$ $`=`$ $`{\displaystyle \underset{m,\nu ,j}{}}𝚽_{ij}^{\mathrm{𝐒𝐂}}(n,\nu ;m,\mu )u_j(m,\mu ),`$ (3)
where $`𝐮_j(m,\mu )`$ is an amplitude of displacement of an atom in the supercell specially shifted from the equilibrium position.
The symmetry of the supercell and the site symmetry of the non-equivalent atoms usually considerably reduce the number of displacements needed for reconstruction of $`𝚽_{ij}^{\mathrm{𝐒𝐂}}`$. We are unfortunate in our case because of the hydrogen bonding fluctuations which makes the crystal structure complicated. Therefore, we have to consider the whole 24 atoms in the supercell as independent displacements: In the positive and negative non-coplanar, $`x`$, $`y`$ and $`z`$ directions. As done for the primitive unit cell, all the internal coordinates were relaxed until the atomic forces were less than $`10^4`$ eV/Å. Complete information of the values of force constants were obtained by displacing every atom of the primitive unit cells by 0.02 Å in both positive and negative $`x`$, $`y`$ and $`z`$ directions. Therefore, minimization of the anharmonic effects and systematic errors are achieved by calculating $`𝚽_{ij}^{\mathrm{𝐒𝐂}}`$ with Eq. 3 using forces arising from both positive and negative displacements $`u_j`$. As mentioned above, we use a $`3\times 1\times 1`$ supercell, which implies that 3 points in the direction are treated exactly according to the direct method. The points are \[$`\zeta `$00\], with $`\zeta `$ = 1, 1/3, 2/3. We calculate forces induced on all atoms of the supercell when a single atom is displaced from its equilibrium position, to obtain the force constant matrix, and hence the dynamical matrix. This is then followed by diagonalization of the dynamical matrix which leads to a set of eigenvalues for the phonon frequencies and the corresponding normal-mode eigenvectors. The vibrational density of states (VDOS) is obtained by integrating over $`𝐤`$-dependent phonon frequencies from the force-constant matrix in supercells derived from the primitive molecule unit cells.
For the ionic crystals the macroscopic electric field is taken into account by adding to Eq. (1) the non-analytic term of the dynamical matrix at the wave vector $`𝐤=0`$ . However, since one knows the phonon frequencies only at discrete wave vectors, it is justified to extend the non-analytical term to the $`𝐤0`$ region, through multiplying it by the Gaussian damping factor. Therefore we replace Eq. (1) by the following expression:
$`𝐃_{\alpha ,\beta }^M(𝐤;\mu \nu )`$ $`=`$ $`𝐃_{}^{\mathrm{𝐒𝐂}}{}_{\alpha ,\beta }{}^{}(𝐤;\mu \nu )`$ (4)
$`+`$ $`{\displaystyle \frac{4\pi e^2}{Vϵ_{\mathrm{}}\sqrt{M_\mu M_\nu }}}{\displaystyle \frac{[𝐤𝐙^{}(\mu )]_\alpha [𝐤𝐙^{}(\nu )]_\beta }{|𝐤|^2}}`$
$`\times `$ $`\mathrm{exp}[2\pi i𝐠(𝐫(\mu )𝐫(\nu ))]`$
$`\times `$ $`d(𝐪)\mathrm{exp}\left\{\pi ^2\left[\left({\displaystyle \frac{k_x}{\rho _x}}\right)^2+\left({\displaystyle \frac{k_y}{\rho _y}}\right)^2+\left({\displaystyle \frac{k_z}{\rho _z}}\right)^2\right]\right\},`$
where $`𝐤`$ is the wave vector within the Brillouin zone with its centre at the reciprocal-lattice vector $`𝐠`$, $`V`$ stands for the volume of the primitive unit cell, and $`M_\mu `$, $`𝐫_\mu `$ are atomic masses and internal positions, respectively. The $`𝐙^{}(\mu )`$ are the tensors of the Born-effective charges. $`ϵ_{\mathrm{}}`$ is the electronic part of the dielectric constant and $`\rho `$ $`(x`$, $`y`$ and $`z`$) are damping factors; then the non-analytical term vanishes close to the zone boundary. Consideration of the effective charges leads to the LO/TO splitting of the optical parts of the phonon modes of of ice at the $`\mathrm{\Gamma }`$-point as discussed in the next Section. This observation has long been been speculated both from theory and experiment.
By definition , the Born effective charge tensor $`𝐙_{}^{}{}_{i,\alpha \beta }{}^{}`$ quantifies to linear order the polarization per unit cell ($`𝐏`$) generated by zone-center k=0, created along the direction $`\beta `$ when the atoms of sublattice $`i`$ are displaced in the direction $`\alpha `$ under the condition of zero electric field. It is calculated according to the equation:
$`𝐙_{}^{}{}_{i,\alpha \beta }{}^{}=𝐙_i+\mathrm{\Omega }{\displaystyle \frac{𝐏_\alpha }{𝐮_{𝐢,\beta }}}.`$ (5)
The macroscopic dielectric constant is found via the relation
$`ϵ_{\mathrm{}}=1+{\displaystyle \frac{4\pi 𝐏}{𝐄}},`$ (6)
where $`𝐄=𝐄_{ext}4\pi 𝐏`$ is the total macroscopic electric field.
## 3. RESULTS AND DISCUSSION
Table I shows the dielectric constants of the $`ϵ_{\mathrm{}}`$ tensor and the Born-effective charges calculated according to Eq. (6) and (5) implemented in PWSCF<sub>2.1</sub> code . The dielectric constant tensor is symmetric with non-zero off-diagonal terms but with negligibly contribution in comparison to the diagonal element. The diagonal term $``$ 1.88 is in a very good range for the high frequency limit (Thz) of the dielectric constant of ice . Under an applied field the individual molecules are polarized by the field. This involves the displacements of the electrons relative to the nuclei and small distortions of the molecules under the restoring forces. The response to a change in field is very rapid, so that the effects are independent of frequency up to microwave frequencies. The polarization in ice in general is due to the reorientation of molecules or bonds, that is, the energies of some of the proton configurations, that are compatible with the ice rules , are lowered relative to others, so that in thermal equilibrium there is net polarization of ice. The achievement of this equilibrium state is a comparatively slow process that requires thermal activation and local violations of the ice rules.
Our results for the dynamical effective charges are shown in Table I for 24 non-equivalent atoms. The complication of the system due to the protons re-orientation and the hydrogen bonding make the symmetry consideration difficult. The charge neutrality condition requires that the acoustic mode frequencies vanish for $`𝐤=0`$ such that
$`𝐙_{i,\alpha \beta }^{}=0.`$ (7)
Therefore, this condition is satisfied to at least order of $`10^4`$ electron which is accurate enough for any reliable calculation. The unequal effective charge tensor components $`Z_\mathrm{H}^{}`$ and $`Z_\mathrm{O}^{}`$ of each hydrogen and oxygen atoms $`Z_{xx}^{}`$ $``$ $`Z_{yy}^{}`$ $``$ $`Z_{zz}^{}`$ are due to the broken symmetry arising from the lattice distortion. The values alternate among the component atoms of H and also for O in order to preserve the overall neutrality. For instance, the $`Z_{zz}^{}`$ of the hydrogen atoms is 0.624($`\pm `$0.001) electron, while the other two elements $`Z_{yy}^{}`$ and $`Z_{zz}^{}`$ alternate within 0.666($`\pm 10^4`$) electron. The off-diagonal elements have the values which ranges within $`\pm 0.408`$, $`\pm 0.407`$, $`\pm 0.402`$, $`\pm 0.377`$ and $`\pm 0.382`$ with deviation $`\pm 10^4`$ electron. Similar features are observed for the oxygen atoms with $`Z_{zz}^{}`$ -1.070 $`\pm 10^4`$ electrons while $`Z_{xx}^{}`$ and $`Z_{yy}^{}`$ alternate between -1.084 and -0.961 with deviation ($`\pm 10^4`$) electrons. Some of the off-diagonal contributions are too small. The observed anisotropic features can be attributed to the complexity of the hydrogen bond during the electron transfer process and also due to the dipole interaction of the water molecules in Eq. 4.
Figure 2 shows the dispersion relation obtained by supplying the calculated effective charges and the corresponding highest frequency dielectric constants, shown in Table I, as the correction from the analytical term which was added to the dynamical matrix as explained in Section 2. We also compare the dispersion obtained in the absence of these charges to see the magnitude of splitting in the optical mode. The calculated VDOS for both dispersions do not appreciably change much because the states are not complete as it requires summation over all points in the first Brillouin zone. The LO modes that was formally degenerate in the absence of $`Z^{}`$ at 27.1 meV in the translational region is now shifted to a higher value 30.2 meV as shown in Fig. 3. Except for the isotopic effect due to the different masses of hydrogen and deuterium atoms in ice (see the experimental dispersion on the right of Fig. 3). This observation correctly shows the splitting in the optical modes due to the dipole interaction of the water molecules which correspondingly induce a dipole moment in the optical mode. Apart from the translational region, as shown in Fig. 4, other splitting of LO modes occur at 111.0 meV in the librational region with a small shift to 112.5 meV. Also, there is a small shift of 0.1 meV from 204.6 meV in the bending region. The large shift of about 12.0 meV is observed from 362 meV in the stretching region because the strength of dipole interaction is large when there is symmetric stretching ($`\nu _1`$) of O-H bond, while a tiny effect observed in the antisymmetric stretching is due to the compensation arising from simultaneous bond lengthening and shortening of the O-H covalent bond.
## 4. SUMMARY
In summary, we have performed the analysis of the lattice dynamics in crystalline tetrahedrally coordinated ice and found a very strong influence of the dipole interaction on the phonon spectra in the optical regions of ice. Ab initio calculations clearly show the splitting in the region between 27.1 and 30.2 meV of the translational region . This observation can be correlated with the experimental observation using high resolution inelastic neutron measurements of the phonon density of states, in which two separate molecular optical bands at about 28 and 37 meV for ice Ih and ice Ic have been observed . Other splitting of LO modes in the librational, bending and stretching region were also predicted due to the dipole interactions. The calculation also shows the extent to which the direct method can be used to calculate the phonon spectra of the dipole system. To our knowledge, a strong influence of the dipole interaction on the lattice dynamics of ice from effective charges calculation was not yet reported.
## Acknowledgments
We acknowledge the support by the Deutsche Forschungsgemeinschaft (Graduate College 277 “Structure and Dynamics of Heterogeneous Systems”).
TABLES
* Calculated dielectric constants and Born effective charges of cubic ice using linear response in PWSCF<sub>2.1</sub> code .
FIGURE CAPTIONS
* Initial and the relaxed geometry of the unit cell of ice. The ice structure was initially packed in a cubic unit cell with initial lattice constant taken from the literature to be 6.35 Å. There are no hydrogen bonds in the initial prepared structure shown on the left but were perfectly formed after the relaxation according to the GGA calculation. The relaxed geometry has the values of $`abc`$ which implies that the relaxed structure is tetragonal with $`c/a`$ ratio $``$ 0.988
* Calculated dispersion curves for ice (a) with no effective charges $`Z^{}`$ taken into account and (b) with $`Z^{}`$ taken into account. The curve on the left is the integrated phonon density of states. The frequencies $`\nu _1`$, $`\nu _2`$, are $`\nu _3`$ are respectively bending, symmetric and anti-symmetric stretching analogous to the vibrational mode of an isolated water molecule . Note that the vibrational phonon density of states is not complete as it requires summation over all points in the first Brillouin zone, nevertheless the calculated G($`\omega `$) for both with and without $`Z^{}`$ do not differ.
* Phonon dispersion in the transitional region showing the splitting of LO mode in the translational region. The dispersions are compared for the cases of both with and without the dipole interaction through the calculation of effective charges $`Z^{}`$. The case with $`Z^{}`$ is marked by a. The experimental dispersion ( taken from ) is shown on the right
* Phonon dispersion in the librational, bending and stretching region showing the splitting of LO modes. The case with $`Z^{}`$ is marked by a as in Fig. 3
Table I: Adeagbo et. al
```
Dielectric constant in cartesian axis
( 1.881163770 -0.000048586 -0.000019997 )
( -0.000048586 1.881154759 0.000112163 )
( -0.000019996 0.000112161 1.883382267 )
Effective charges E-U in cartesian axis
Water molecule (1) (hydrogen atom 1)
( 0.66721 0.37702 0.40166 )
( 0.40921 0.53464 0.38200 )
( 0.40785 0.36304 0.62434 )
Water molecule (1) (hydrogen atom 2)
( 0.66611 -0.37720 0.40208 )
( -0.40909 0.53493 -0.38272 )
( 0.40784 -0.36353 0.62525 )
Water molecule (2) (hydrogen atom 1)
( 0.66619 0.37729 -0.40199 )
( 0.40909 0.53496 -0.38262 )
( -0.40785 -0.36360 0.62507 )
Water molecule (2) (hydrogen atom 2)
( 0.66694 -0.37691 -0.40154 )
( -0.40907 0.53471 0.38197 )
( -0.40768 0.36296 0.62440 )
Water molecule (3) (hydrogen atom 1)
( 0.53461 0.40937 -0.38251 )
( 0.37708 0.66692 -0.40207 )
( -0.36332 -0.40814 0.62483 )
Water molecule (3) (hydrogen atom 2)
( 0.53459 -0.40938 0.38266 )
( -0.37714 0.66694 -0.40229 )
( 0.36338 -0.40822 0.62507 )
Water molecule (4) (hydrogen atom 1)
( 0.53516 -0.40898 -0.38225 )
( -0.37710 0.66634 0.40159 )
( -0.36333 0.40759 0.62478 )
Water molecule (4) (hydrogen atom 2)
( 0.53515 0.40892 0.38210 )
( 0.37715 0.66626 0.40141 )
( 0.36335 0.40747 0.62434 )
Water molecule (5) (hydrogen atom 1)
( 0.66666 -0.37682 -0.40144 )
( -0.40902 0.53480 0.38199 )
( -0.40761 0.36297 0.62446 )
Water molecule (5) (hydrogen atom 2)
( 0.66648 0.37748 -0.40192 )
( 0.40929 0.53519 -0.38256 )
( -0.40784 -0.36361 0.62469 )
Water molecule (6) (hydrogen atom 1)
( 0.66694 0.37695 0.40158 )
( 0.40919 0.53476 0.38206 )
( 0.40781 0.36307 0.62443 )
Water molecule (6) (hydrogen atom 2)
( 0.66639 -0.37738 0.40199 )
( -0.40926 0.53512 -0.38263 )
( 0.40780 -0.36352 0.62484 )
Water molecule (7) (hydrogen atom 1)
( 0.53456 0.40948 -0.38257 )
( 0.37720 0.66723 -0.40225 )
( -0.36340 -0.40832 0.62487 )
Water molecule (7) (hydrogen atom 2)
( 0.53464 -0.40928 0.38261 )
( -0.37704 0.66666 -0.40213 )
( 0.36332 -0.40807 0.62503 )
Water molecule (8) (hydrogen atom 1)
( 0.53514 0.40901 0.38225 )
( 0.37720 0.66637 0.40161 )
( 0.36345 0.40765 0.62469 )
Water molecule (8) (hydrogen atom 2)
( 0.53515 -0.40887 -0.38209 )
( -0.37703 0.66620 0.40138 )
( -0.36322 0.40739 0.62441 )
Water molecule (1) (oxygen atom 1)
( -1.08499 -0.00011 -0.07462 )
( 0.00002 -0.96152 0.00068 )
( -0.09759 0.00062 -1.07021 )
Water molecule (2) (oxygen atom 2)
( -1.08486 -0.00006 0.07446 )
( -0.00023 -0.96140 0.00063 )
( 0.09737 0.00068 -1.07026 )
Water molecule (3) (oxygen atom 3)
( -0.96205 0.00002 -0.00017 )
( -0.00008 -1.08534 0.07521 )
( -0.00013 0.09795 -1.06997 )
Water molecule (4) (oxygen atom 4)
( -0.96115 0.00000 -0.00001 )
( -0.00008 -1.08473 -0.07392 )
( 0.00006 -0.09639 -1.07014 )
Water molecule (5) (oxygen atom 5)
( -1.08488 -0.00031 0.07430 )
( -0.00036 -0.96143 0.00038 )
( 0.09733 0.00058 -1.07024 )
Water molecule (6) (oxygen atom 6)
( -1.08501 0.00014 -0.07445 )
( 0.00014 -0.96156 0.00039 )
( -0.09755 0.00049 -1.07018 )
Water molecule (7) (oxygen atom 7)
( -0.96206 -0.00013 -0.00004 )
( -0.00004 -1.08536 0.07520 )
( 0.00010 0.09795 -1.06999 )
Water molecule (8) (oxygen atom 8)
( -0.96113 0.00011 -0.00001 )
( 0.00005 -1.08471 -0.07392 )
( -0.00004 -0.09639 -1.07012 )
```
|
warning/0506/cond-mat0506031.html
|
ar5iv
|
text
|
# Spin polarization induced tenfold magneto-resistivity of highly metallic 2D holes in a narrow GaAs quantum well
## Abstract
We observe that an in-plane magnetic field ($`B_{||}`$) can induce an order of magnitude enhancement in the low temperature ($`T`$) resistivity ($`\rho `$) of metallic 2D holes in a narrow (10nm) GaAs quantum well. Moreover, we show the first observation of saturating behavior of $`\rho (B_{||})`$ at high $`B_{||}`$ in GaAs system, which suggests our large positive $`\rho (B_{||})`$ is due to the spin polarization effect alone. We find that this tenfold increase in $`\rho (B_{||})`$ even persists deeply into the 2D metallic state with the high $`B_{||}`$ saturating values of $`\rho `$ lower than 0.1h/e<sup>2</sup>. The dramatic effect of $`B_{||}`$ we observe on the highly conductive 2D holes (with $`B`$=0 conductivity as high as 75e<sup>2</sup>/h) sets strong constraint on models for the spin dependent transport in dilute metallic 2D systems.
The metallic behavior and metal-insulator transition (MIT) in dilute electrons or holes in two dimensional (2D) semiconductor structures have received much recent interestmitreview ; review2004 . In these low density 2D systems, when the carrier density is above the critical density, the system exhibits a significant resistivity drop at low temperature, setting a challenge for conventional localization theory. While novel properties (e.g. the dramatic change in compressibility at MITjiangcompress , the anomalous thermopower thermopower and enhanced phonon coupling phononcoupling effects) are continuing to be discovered in this 2D metallic state, many critical issues still remain unresolved. Outstanding questions include: Does the Fermi liquid (FL) phenomenology still hold for the 2D metallic state where $`r_s`$1? Is this MIT a true quantum phase transition or simply a crossover at finite temperature? And most importantly, what is the mechanism for the resistivity drop?
The spin degeneracy is believed to be essential for inducing the metallic resistivity, as it was found that an in-plane magnetic field $`B_{||}`$ suppresses the metallicity and in some cases drives the system insulatingsimonian ; yoon ; papadakis ; Gao ; tutucSDHBp . Recent experiments on GaAs quantum well (QW) further revealed an intriguing $`B_{||}`$ insensitivity of the energy scale of the 2D metal as well as the FL-like logarithmically diverging $`\rho (T)`$ of 2D holes in strong $`B_{||}`$Gao . Many theoretical models were proposed to explain the $`B_{||}`$ destruction of the 2D metallic transport, such as the superconductivity scenarioPhilips , the FL-Wigner solid coexisting microemulsion modelspivakkivelson , or screening model based on conventional FL wisdom goldJETP ; dassarmaparaB . It was even noticed that positive $`\rho (B_{||})`$ can be induced by the magneto-orbital effect of $`B_{||}`$ due to the finite thickness of the sample, without involving any spin effectdassarmaB|| .
In this paper, we present a study of the in-plane magnetic field induced magneto-transport of a low density 2D hole system (2DHS) in a narrow (10nm wide) GaAs QW down to as low as $`T`$=20mK. We show that the resistivity of our 2DHS can increase by nearly an order of magnitude followed by a saturation as $`B_{||}`$ increases, similar to the case for Si-MOSFET’s. In contrast to previous experiments on GaAs heterostructures or wider QWs yoon ; papadakis ; tutucSDHBp ; zhunGaAschi , our result clearly disentangles the spin effect from the orbital effectdassarmaB|| in the $`B_{||}`$-dependent transport studies of the 2D metallic state. Moreover, it is striking that this spin polarization induced tenfold magneto-resistivity even persists deeply into the metallic state where the conductivity $`\sigma `$ is as high as 75e<sup>2</sup>/h at $`B`$=0. In-plane magneto-transport has been extensively calculated for low density 2D systems within the screening theory for FL. For weak disorder, semi-classical calculations based on $`T`$\- and $`B`$-dependent screeningdassarmaparaB showed good agreement with highly conductive Si-MOSFET’s, in which a factor 3 to 4 increase in $`\rho (B_{||})`$ and a weak $`T`$-dependent $`\rho (T)`$ in the spin polarized state were observedtsui ; shashkin05 . Refined screening models including exchange and correlation effects may produce a larger increase in $`\rho (B_{||})`$, but only when disorder is sufficiently strong and carrier density sufficiently low to be in the vicinity of the MITgoldJETPdisorder . Our observation of such large $`\rho (B_{||})`$ for metallic 2DHS with $`\sigma e^2/h`$(or $`k_Fl`$1) calls for further theoretical understanding of spin-dependent transport in dilute metallic 2D systems with strong correlations and weak disorder.
Our experiments were performed on a high mobility low-density 2DHS in a 10nm wide GaAs QW similar to refGao ; GaoHall ; phononcoupling . The sample was grown on a (311)A GaAs wafer using Al<sub>.1</sub>Ga<sub>.9</sub>As barrier. Delta-doping layers of Si dopants were symmetrically placed above and below the pure GaAs QW. Diffused In(1$`\%`$Zn) was used as contacts. The hole density $`p`$ was tuned by a backgate voltage. The ungated sample has a low temperature hole mobility, $`\mu 5\times 10^5`$cm<sup>2</sup>/Vs, and a density $``$1.6$`\times `$10<sup>10</sup>cm<sup>-2</sup> from doping. The sample was prepared in the form of Hall bar, with an approximate total sample area 0.2cm<sup>2</sup>. With the relatively large sample area and the measuring current induced heating power at the level of fWatts/cm<sup>2</sup>, the low density 2DHS can be reliably cooled down to 20mK with negligible self-heatingphononcoupling . All the data in this paper were taken with the current along the \[$`\underset{¯}{2}`$33\] high mobility direction. $`B_{||}`$ was also applied along the \[$`\underset{¯}{2}`$33\] direction, where the effective g-factor$``$0.6 Gao . During the experiments, the sample was immersed in the <sup>3</sup>He/<sup>4</sup>He mixture in a top-loading dilution refrigerator.
Figure 1 shows the $`T`$=20mK $`\rho `$ vs. $`B_{||}`$ and the zero magnetic field $`\rho `$ vs. $`T`$ data of our 2DHS with $`p`$=1.35$`\times `$10<sup>10</sup>cm<sup>-2</sup> in the metallic phase of MIT. At $`B`$=0, $`\rho `$ shows a factor of three drop below 0.4K. The $`\rho (B_{||})`$ curve shows a very large magneto-resistivity below 1.5T and a nearly constant $`\rho `$ at higher $`B_{||}`$. This behavior is rather similar to the $`\rho (B_{||})`$ data in Si-MOSFET’s and the magnetic field $`B_P`$ at which $`\rho `$ starts saturating was identified to be the field when the system obtains full spin polarizationokamotoSDHBp ; vitkalov ; kravchenkoBp . We mention that all previous $`\rho (B_{||})`$ data on GaAs 2D electron/hole systems show somewhat different behavior: the resistivity continuously increases with a reflection point around $`B_P`$ upon applying $`B_{||}`$ yoon ; papadakis ; tutucSDHBp ; zhunGaAschi . We believe that the saturating behavior of our $`\rho (B_{||})`$ here at $`B_{||}>B_P`$ is due to the smaller thickness of our QW. The constant $`\rho (B_{||})`$ above $`B_P`$ of our QW also suggests that the magneto-orbital effect related scatteringdassarmaB|| is small in our case. For GaAs heterostructures or wider QW’s (and low carrier concentration), the magnetic length at several Tesla becomes comparable or smaller than the width of the 2D electron/hole wavefunction in the $`z`$-direction and the magneto-orbital effect can induce a continuous positive magneto-resistivity as discussed by Das Sarma and HwangdassarmaB|| . Note that for experiments on Si-MOSFET’s the confinement in the $`z`$-direction is also narrow and a saturation in $`\rho `$ is often observed after an increasing $`\rho `$ at low $`B_{||}`$okamotoSDHBp ; vitkalov ; kravchenkoBp . Thus our data suggest that the thickness effect is certainly able to explain most of the differences in $`\rho (B_{||})`$ behavior between GaAs and Si-MOSFET systems, although the valley degeneracy may play some additional role.
Now we discuss how the temperature affects the magneto-transport. In Fig.2a we plot the $`\sigma (B_{||})`$ for $`p`$=1.35$`\times `$10<sup>10</sup>cm<sup>-2</sup> at 20mK, 0.15K, 0.26K and 0.40K. All the iso-thermal $`\sigma (B_{||})`$ curves cross around 1.2T, indicating the ‘$`B_{||}`$ induced MIT’yoon ; papadakis ; Gao . As suggested by Vitkalov et al.vitkalovdisorder , we can determine the magnetic field $`B_P`$ for the onset of full spin polarization of $`delocalized`$ holes by the intersection of linear extrapolations of $`\sigma (B_{||})`$ at low and high field regions. Nonetheless, we obtain a $`B_P`$ only 10$`\%`$ higher if the extrapolating process is applied to $`\rho (B_{||})`$, suggesting most holes are delocalized. We find that $`B_P`$ is strongly temperature dependent. As one can see in Fig.2b where $`B_P(T)`$ is plotted for this density, $`B_P`$ at $`T`$=20mK is only 40$`\%`$ of its value at 0.4K. Since the $`B_P`$ is generally regarded as the magnetic field required to fully polarize the spins of delocalized carriersokamotoSDHBp ; vitkalov , one natural interpretation of the $`T`$-dependent $`B_P`$ is that the spin susceptibility $`\chi `$ is largely enhanced as $`T`$ is reduced. This strong $`T`$-dependent $`B_P`$ has implications in other models as well. For instance, in the ‘microemulsion’ model it would mean that it requires much less Zeeman energy to solidify the FL phase at lower temperaturesspivakkivelson .
Figure2c shows the density dependence of $`B_P`$ at 20mK, 0.15K, 0.26K and 0.4K. Previously, extrapolating $`B_P(p)`$ to $`B_P`$=0 was used as a way to determine if a ferromagnetic instability exists in the systemreview2004 ; zhunGaAschi ; kravchenkoBp . If $`B_P(p)`$ extrapolates to zero at a finite density, then such density corresponds to the ferromagnetic instability. It can be seen in our Fig.2c that our $`B_P(p)`$ data taken at different $`T`$ extrapolate to zero at different densities. Only at low temperatures $`B_P(p)`$ linearly extrapolates to zero at a finite density. At $`T`$=20mK, $`B_P(p)`$ extrapolates to zero at a density very close to the critical density of the $`B=0`$ MITfootnote . Our $`T`$-dependent study of $`B_P(p)`$ is consistent with Si-MOSFET’ssarachikchiT . Although the meaning of a diminishing $`B_P`$ at finite $`p`$ is controversial and may actually be associated with other physics (e.g. the instability to crystallization instead of ferromagnetism)spivakkivelson ; spivakkivelsonBp , our experiment on p-GaAs corroborates the universal existence of this behavior.
Figure1 has shown that the narrow p-GaAs QW responds to $`B_{||}`$ quite similarly to Si-MOSFET’s, where $`\rho (B_{||})`$ shows large increase at low $`B_{||}`$ and a saturation at high field. It is striking that for our p-GaAs, this order of magnitude positive $`\rho (B_{||})`$ even persists deeply into the metallic phase where $`\rho `$h/e<sup>2</sup> for both the $`B_{||}`$=0 low spin polarization phase and the high $`B_{||}`$($`>B_P`$) spin-polarized phasespinnote . Fig.3 shows $`\rho (B_{||})`$ at 20mK for seven densities up to 2.1$`\times `$10<sup>10</sup>cm<sup>-2</sup>. The large enhancement and saturation in $`\rho (B_{||})`$ are observed for all these densities. Note that for $`p`$=2.1 (the lowest curve) the high field ($`B>B_P`$) value of $`\rho `$ is clearly below 0.1h/e<sup>2</sup>. For Si-MOSFET system with comparable resistivity, $`\rho `$ usually shows only a factor of 3-4 increase below $`B_P`$kravchenkoBp ; vitkalov ; goldJETPdisorder , in agreement with the screening modelgoldJETP ; dassarmaparaB ; goldJETPdisorder . Note that the screening model predicts at most a factor of four increase in $`\rho `$ due to reduced screening from the lifted spin degeneracy for $`\rho h/e^2`$ goldJETP ; dassarmaparaB . Only very near the critical density of the MIT can the Si-MOSFET show $`\rho (B_{||})/\rho (0)>`$4, resulting from many-body and strong disorder effects in the screening modelgoldJETPdisorder . Moreover, the original publication of screening theory predicts a weak metallic like $`\sigma (T)`$ at $`B_{||}>B_P`$dassarmaparaB , in disagreement with our data in Fig.4 below. It is possible that exchange (Fock) term of the electron-electron interactionZNA could account for the difference; however, to date, the only FL theory including both Hartree and Fock interactionsZNA , is perturbative, valid only at $`TT_F`$ and not applicable to our experimental regime. More sophisticated non-perturbative Fermi liquid calculations are needed for further comparison with our data.
The dramatic effect of spin polarization induced by $`B_{||}`$ on our dilute 2DHS also exhibits in the temperature dependence of the conductivity. In Fig.4 we plot $`\sigma (T)`$ at various $`B_{||}`$ for $`p`$=2.1$`\times `$10<sup>10</sup>cm<sup>-2</sup>. At $`B`$=0,the 2DHS shows a factor of three increase in the conductivity below 0.8K and the low $`T`$ conductivity is as high as 75e<sup>2</sup>/h. With the application of $`B_{||}`$, the metallic conductivity enhancement becomes smaller and eventually $`\sigma (T)`$ turns into insulating-like(d$`\sigma `$/d$`T>`$0) above 2T. In the inset of Fig.4 we plot d$`\sigma `$/d$`T`$, the slope of $`\sigma (T)`$, as a function of $`B_{||}`$ to demonstrate this strong effect of $`B_{||}`$ on the 2D metallic transport. It can be seen that the absolute values of the slope of $`\sigma (T)`$ differ by about a factor of ten between the zero and high field regimes. A similar effect was also seen in Si-MOSFET’stsui .
The $`B_{||}`$ suppression of 2D metallic transport was attributed to the FL interaction correction effects in the ballistic regimeZNA in various recent experimental papersproskuzna . Here we do not attempt to fit our data to extract the FL parameter $`F_0^\sigma `$ since we believe that the $`perturbative`$ FL calculation should not be taken as a quantitative theory for our order of magnitude increase in $`\rho (B_{||})`$. Recent Hall coefficient measurements on similar samples also provide experimental evidence against the interaction correction interpretation for the metallic $`\sigma (T)`$ at $`B`$=0GaoHall , further reflecting the fact that the $`TT_F`$ theory is inapplicable to our datadassarmahall . A non-perturbative FL calculation including both the Hartree and Fock interaction terms and extending to temperatures $`T>T_F`$ would be required to make a direct comparison with our data. Another possible explanation for our large magnetoresistivity effect comes from a non-perturbative non-FL approach: it has been theoretically argued that intermediate phases (‘microemulsions’) exist in clean 2D systems between the FL phase and the Wigner solid phasespivakkivelson ; spivakkivelsonBp . In such a scenario, the dramatic suppression of the slope of $`\sigma (T)`$ by an in-plane magnetic field $`B_{||}`$ would be analogous to the magnetic field effect on the Pomaranchuk effect in <sup>3</sup>Hespivakkivelson . It will be of interest to develop more quantitative calculations based on such model for a direct comparison with our experimental data.
The authors are indebted to S.A. Kivelson and B. Spivak for discussions and encouragement which stimulated the present paper. We also thank S. Das Sarma for bringing ref.dassarmaparaB to our attention. The NHMFL is supported by the NSF and the State of Florida.
|
warning/0506/astro-ph0506247.html
|
ar5iv
|
text
|
# Discovery of pulsations in the X-ray transient 4U 1901+03
## 1. Introduction
More than half of the known X-ray pulsars are transient sources which were discovered during bright outbursts. Most of these are Be-star/X-ray binaries, which are believed to be progenitors of double neutron star binaries (e.g. Bhattacharya & van den Heuvel, 1991). In contrast to binaries with low-mass ($`1M_{}`$) companions, the neutron stars in these high-mass X-ray binaries (HMXBs) typically accrete through the companion’s stellar wind. The mass transfer rate $`\dot{M}`$ — as well as its variability — depends in a sensitive manner on the wind properties, as well as on the neutron star spin period and magnetic field strength. A smaller class of pulsars accrete from supergiant companions, which may fill their Roche lobes and thus accrete persistently. Tidal forces in these binaries act to circularize the orbits on a time scale much shorter than the active lifetime, while the wider, wind-accreting binaries typically have moderate to high eccentricities (e.g. Bildsten et al., 1997). In recent years, a third class of binaries has emerged, with wide $`20`$ d orbits but low eccentricities $`e0.1`$. These sources present difficulties for the commonly accepted formation scenario in which the natal supernova event imparts a “kick” to the neutron star, leading to an initially eccentric orbit. The tidal forces which act so efficiently in the Roche-lobe filling systems cannot circularize wider orbits within the source lifetime, suggesting that the initial kicks in these wide, circular binaries may be unusually small Pfahl et al. (2002).
For some transients the interval between outbursts can be as long as 20 yr. Little is generally known about sources with such long duty cycles, due to the dearth of observations (in particular with large-area modern instruments with good timing capabilities). For some candidate X-ray pulsars no pulsations have even been detected, and the classification comes from a hard X-ray spectrum, typical in confirmed pulsars. 4U 1901+03 ($`l=37\stackrel{}{\mathrm{.}}16`$, $`b=1\stackrel{}{\mathrm{.}}25`$) is such a source, previously detected just once before in outburst by Uhuru and Vela 5B in 1970–1 Forman et al. (1976); Priedhorsky & Terrell (1984). Due to the hard spectrum measured during those observations, the source was tentatively identified as an HXMB. Consequently, we selected this source as one of a group of hard transients with positions known to $`10\mathrm{}`$ or better, as candidates for target-of-opportunity observations by the Rossi X-ray Timing Explorer (RXTE). In 2003 February a new outburst of 4U 1901+03 was detected by the All-Sky Monitor (ASM) aboard RXTE Galloway et al. (2003b). The source was also detected with the IBIS and JEM-X hard X-ray instruments aboard INTEGRAL between 2003 March 10 and April 13 2003 Molkov et al. (2003). An RXTE Proportional Counter Array (PCA) scan across the Uhuru position led to more precise coordinates of $`R.A.=19^\mathrm{h}03^\mathrm{m}37\stackrel{\mathrm{s}}{\mathrm{.}}1`$, decl. = $`+3\mathrm{°}11\mathrm{}31\mathrm{}`$ (J2000.0), with an estimated 90% confidence uncertainty of $`1^{}`$ Galloway et al. (2003a). Followup pointed RXTE observations detected coherent pulsations with a period of 2.763 s. Variations in the observed pulse frequency were also observed, suggesting an orbital period of around 25 d.
Here we present timing and spectral analyses of RXTE observations of 4U 1901+03 throughout the 2003 outburst, as well as results from a search for the optical/IR counterpart.
## 2. Observations
We made observations of 4U 1901+03 with the Proportional Counter Array (PCA; Jahoda et al., 1996) and the High-Energy X-ray Timing Experiment (HEXTE; Gruber et al., 1996) instruments aboard RXTE. The PCA consists of 5 Proportional Counter Units (PCUs) each with a collecting area of $`1400\mathrm{cm}^2`$ and a $`1\mathrm{°}`$ field of view, that are sensitive to X-ray photons with energies in the range 2.5–90 keV. Photon arrival times are measured to $`1\mu \mathrm{s}`$, while spectra are accumulated in up to 256 energy channels. The HEXTE comprises two clusters, each with 4 scintillation detectors sensitive to photons in the range 15–250 keV, collimated to view a common $`1\mathrm{°}`$ field. The detectors in the two clusters provide a total collecting area of $`1600\mathrm{cm}^2`$.
Short ($`3`$ ks) observations were scheduled every 3–4 days between 2003 February 10 and July 16 (MJD 52,680 and 52,837) in order to adequately sample the 25 d candidate orbital period. In addition, several longer observations were scheduled near the peak of the outburst as part of a separate proposal to search for cyclotron resonance features (PI: Heindl). Data were analysed using lheasoft release 5.3 (2003 November 17). We extracted PCA and HEXTE spectra from intervals within each observation during which the center of the field-of-view was within $`0.02\mathrm{°}`$ of the position of 4U 1901+03, and for which the limb of the Earth was more than $`10\mathrm{°}`$ from the source direction. Spectra were extracted from standard observing mode data for each instrument (“Standard-2”, with 129 channels between 2–60 keV for the PCA, and “Archive”, with 64 channels between 15–250 keV for the HEXTE). PCA spectra were accumulated separately for each PCU, and instrument response matrices were generated for each PCU and each observation using pcarsp v.10.1. We estimated background count spectra for the PCA using “bright” source models (suitable for when the count rate exceeds $`40\mathrm{counts}\mathrm{s}^1\mathrm{PCU}^1`$) developed for gain epoch 5 (2000 May 13 onwards) with pcabackest. We measured the mean flux for each observation by fitting spectra from individual PCUs separately (using the model described in §3.1) between 2.5–25 keV<sup>1</sup><sup>1</sup>1The Crab flux in this energy range is $`3.3\times 10^8\mathrm{ergs}\mathrm{cm}^2\mathrm{s}^1`$. We corrected the integrated 2.5–25 keV flux (except for PCU 2) by dividing by the mean ratio of the fluxes for each PCU relative to PCU 2, and adopted the residual standard deviation on the rescaled fluxes as the $`1\sigma `$ uncertainty. For our timing analysis we used full-range PCA lightcurves with 4 ms time resolution, with time bins corrected to the solar system barycenter.
## 3. Results
### 3.1. Flux evolution and X-ray spectrum
The X-ray flux peaked at almost $`8\times 10^9\mathrm{ergs}\mathrm{cm}^2\mathrm{s}^1`$ (2.5–25 keV; approximately 240 mCrab) around 2003 February 20 (MJD 52,690) and decreased linearly down to $`10^{10}\mathrm{ergs}\mathrm{cm}^2\mathrm{s}^1`$ by 2003 July 15 (Fig. 3.1). Although bright outbursts in transient pulsars are sometimes followed by extended periods of lower-level activity (e.g. KS 1947+300; Galloway et al., 2004), the ASM did not detect 4U 1901+03 at a significant
level between the end of the 2003 outburst through 2005 June. The modest PCA energy resolution at low energies meant that it was not possible to constrain the column density $`n_H`$ for neutral absorption in the spectral fits. Thus, for all our spectral fits we froze $`n_H`$ at $`1.2\times 10^{22}\mathrm{cm}^2`$ (at the lower end of the expected range inferred from the $`A_V`$ estimates; see §3.4). Our broadband (absorbed) flux measurements were not sensitive to the assumed value of $`n_H`$. Fits with commonly-used pulsar X-ray spectral models including power law, cutoff power law and a combination of blackbody and power law gave $`\chi ^2`$ values indicating statistically unacceptable fits. We found the best agreement with the data for a model consisting of a Comptonisation component (comptt in xspec; Titarchuk, 1994) and a Gaussian component centered around 6.4 keV to represent fluorescent Fe line emission, both attenuated by neutral absorption with column density at the survey value. We assumed a systematic error of 1% in order to achieve a reduced $`\chi ^2`$ (averaged over the fits to spectra from each PCU) of $`1`$. A systematic error of this magnitude is typically required for PCA spectral fits to bright sources, for example the Crab pulsar (R. Remillard, pers. comm.). Near the end of the outburst (between MJD 52,780 and 52,820) the reduced-$`\chi ^2`$ was somewhat larger, between 2 and 3.5. The main factor contributing to the poor $`\chi ^2`$ value was a deficit of photons (compared to the model) between 8 and 10 keV; this deficit was observed irrespective of the choice of continuum components tested. In low signal-to-noise spectra (e.g. from short observations), the residuals could be removed by including a blackbody component with $`kT_{\mathrm{bb}}1`$ keV, but for longer observations the residuals were more complex.
Although the residuals indicate that the adopted spectral model does not completely describe the source spectrum, the derived parameters present a qualitative description that is adequate for flux measurements as well as a general characterisation of the spectral shape and its
variation. The fitted optical depth $`\tau `$ decreased slightly over the course of the outburst, from 6 at the peak to 4 near the end of the outburst. The temperature of the scattering electrons $`kT_e`$ increased over the same period from 4 to 7 keV. A narrow emission feature around 6.4 keV was present throughout, with equivalent width between 60 and 150 eV.
We also made combined fits to PCA and HEXTE spectra for selected observations near the peak and end of the outburst. The spectral parameters for these broadband fits were similar to the fits to the PCA data only. The spectrum was rather soft for a HMXB pulsar, with little emission detected above 80 keV. As indicated by the evolution of the Comptonisation model parameters, the spectrum hardened considerably over the course of the outburst (Fig. 3.1). Although the best-fit residuals still indicated systematic deviations at energies $`10`$ keV from the best-fit model spectrum, we found no evidence for cyclotron resonance features in the spectrum. Using the broadband fits, we estimated the bolometric correction as the ratio of fluxes integrated over an idealised response matrix spanning 0.1–200 keV, and the flux in the range 2.5–25 keV, as 1.12.
### 3.2. Pulse timing
We estimated the pulse frequency for each observation by first folding the 4-ms lightcurve on a trial period, to obtain an observation-averaged pulse profile with (typically) 32 phase bins. We then folded individually 256-s segments on the same period, and cross-correlated the resulting pulse profiles with the observation-averaged profile to obtain the phase delay for each segment. We then adjusted the period and repeated the procedure until the phase delay exhibited no net trend with time throughout the observation. The error was estimated from the uncertainty on the first-order term of a linear fit to the phase delays. The resulting frequency history shows approximately sinusoidal variations indicative of Doppler shifts from binary orbital motion, superimposed on a significant (non-linear) spin-up trend over the course of the outburst (Fig. 3.2).
We fit the frequency measurements with a linear model comprising the spin-up due to accretion torques in addition to the apparent changes due to orbital Doppler shifts:
$`f(t)`$ $`=`$ $`f_{spin}(t){\displaystyle \frac{2\pi f_0a_X\mathrm{sin}i}{P_{\mathrm{orb}}}}`$ (1)
$`\times (\mathrm{cos}l+g\mathrm{sin}2l+h\mathrm{cos}2l)`$
where $`f_{spin}(t)`$ is the time-dependent neutron-star spin frequency, $`f_0`$ is a constant approximating $`f_{spin}(t)`$, $`a_X\mathrm{sin}i`$ is the projected orbital semimajor axis in units of light travel time, and $`P_{\mathrm{orb}}`$ is the orbital period. The coefficients $`g(=e\mathrm{sin}\omega )`$ and $`h(=e\mathrm{cos}\omega )`$ are functions of the eccentricity $`e`$ and the longitude of periastron $`\omega `$. Finally, $`l=2\pi (tT_{\pi /2})/P_{\mathrm{orb}}+\pi /2`$ is the mean longitude, with $`T_{\pi /2}`$ the epoch at which the mean longitude $`l=\pi /2`$. For a circular orbit, $`T_{\pi /2}`$ is the epoch of superior conjunction (when the neutron star is behind the companion). The right-most term in Eqn. 2 represents the orbital Doppler shifts to first order in $`e`$; given the magnitudes of the uncertainties of our measurements, this should be an adequate approximation as long as $`e0.2`$.
We described the intrinsic spin frequency evolution with both $`\dot{f}`$ and $`\ddot{f}`$ terms:
$$f_{spin}(t)=f_0+\dot{f}(tt_0)+\ddot{f}(tt_0)^2$$
(2)
where $`f_0`$ and $`t_0`$ are the frequency and time, respectively, of the first frequency measurement and the $`\dot{f}`$, $`\ddot{f}`$ are constant over the outburst duration.
We fit the frequency model to the measurements using a nonlinear gradient-expansion algorithm (curvefit in IDL). We achieved an acceptable fit ($`\chi ^2=77.22`$ for 53 degrees of freedom) with orbital parameters $`P_{\mathrm{orb}}=22.58`$ d, $`a_X\mathrm{sin}i=106.9`$ lt-sec and $`e=0.035`$. We used this preliminary orbital model to perform a fit of the accumulated phase delay measurements. We first integrated equations 1 and 2 to obtain an expression for the pulse phase evolution with time:
$`\varphi (t)`$ $`=`$ $`\varphi _0+f_0(tt_0)+\frac{\dot{f}}{2}(tt_0)^2+\frac{\ddot{f}}{3}(tt_0)^3`$ (3)
$`f_0a_X\mathrm{sin}i\left(\mathrm{sin}l\frac{g}{2}\mathrm{cos}2l\frac{h}{2}\mathrm{sin}2l\right)`$
where $`\varphi _0`$ is an arbitrary reference phase, corresponding in this case to the peak of the fundamental of the first pulse observed in the first observation. We then fit this model to the measured pulse arrival times, defined as the peak of the fundamental Fourier component.
The frequency of observations during the first full orbital cycle of the outburst (between MJD 52,680 and 52,700) was such that we were able to unambiguously track the pulse phase over the entire cycle. Even so, we found significant residual phase delays which varied systematically on a timescale of a few days, with an rms amplitude of $`0.1`$ cycles. Although variations in the pulse profile over the entire outburst (see §3.3) may contribute to the residuals to the phase fit, residuals were also present during intervals when the pulse profile shape was relatively consistent. Thus, there is significant intrinsic timing noise present, perhaps arising from small changes in the instantaneous accretion rate.
Beyond MJD 52700 the 2–3 d gaps between the RXTE observations introduced the possibility of pulse count ambiguities, although only of magnitude $`\pm 1`$ cycle in general. We repeatedly computed the phase fit after adding or subtracting a cycle within the data gaps, in order to minimise the total $`\chi ^2`$ until no further improvement was possible. Because of the timing noise, the resulting $`\chi ^2`$ calculated using the errors on the individual phase measurements was much larger than the number of degrees of freedom. In order to estimate the confidence limits for the orbital parameters, we re-scaled the pulse arrival time errors so that the resulting $`\chi ^2`$ was 1 per degree of freedom. We then varied each parameter in turn, fitting with all other parameters free to vary, to determine the parameter range for which the rescaled $`\chi ^2\chi _{\mathrm{min}}^2(1+1/n)`$, where $`n`$ is the number of degrees of freedom (1473 for the full set of arrival time measurements). The resulting orbital parameters and uncertainties are listed in Table 1; the predicted frequency, intrinsic spin frequency for the neutron star and the residuals from the model are shown in Fig. 3.2. The pulse arrival time delays with respect to the intrinsic spin evolution model are shown in Fig. 3.2.
Our best-fit parameters describing the intrinsic spin evolution indicate that the spin-up rate was initially around $`3\times 10^{11}\mathrm{Hz}\mathrm{s}^1`$, similar to the maximum measured for other transient pulsars (e.g. Bildsten et al., 1997). According to the intrinsic spin model, the spin-up decreased throughout the outburst, falling to zero just before the end of the outburst, around MJD 52,824. The
2.5–25 keV flux by this time had dropped to around $`5.5\times 10^{10}\mathrm{ergs}\mathrm{cm}^2\mathrm{s}^1`$.
### 3.3. Pulse profile variability
The pulse profile was consistently non-sinusoidal, and exhibited several intervals of stability punctuated by relatively rapid change. The fractional pulse amplitude also varied over the outburst, and was generally between 4 and 22% (rms). From the beginning of the outburst until 2003 March 3 the profile was consistently similar to the example shown from February 19 (MJD 52,689; Fig. 3.2). During that observation the pulse amplitude was 15% rms. Between March 3 and 13 the profile switched to the characteristic double-peaked shape of March 19; also on March 13 the rms amplitude dropped to 7%. The profile continued to evolve throughout March and April, with gradually increasing pulse fraction and small drifts of the harmonic components until another abrupt shift between 2003 June 14–18. The pulse amplitude had risen to 13% rms on June 14 before falling abruptly to 9% rms on June 18, and then recovering to an overall maximum for the outburst of 22% rms on June 28. After this final peak, the profile remained similar in shape but with steadily decreasing amplitude towards the end of the outburst. Pulsations became undetectable ($`<1`$% rms) after 2003 July 13 (MJD 52,833), by which time the 2.5–25 keV flux had dropped to below $`3\times 10^{11}\mathrm{ergs}\mathrm{cm}^2\mathrm{s}^1`$.
### 3.4. A search for the optical/IR counterpart
Following the improved X-ray position obtained from the PCA scan Galloway et al. (2003a), we examined the field of 4U 1901+03 in Digital Sky Survey and 2MASS images to identify the optical counterpart. The mass function implies a minimum companion mass (assuming a $`1.4M_{}`$ neutron star) of $`4.5M_{}`$. The most probable companion mass (for an inclination of $`60^{}`$) is $`6.0M_{}`$. The donor stars in long-period binary pulsars are typically either Be stars or (sometimes Roche-lobe filling) OB supergiants. The column density interpolated from Hi survey observations towards the source is between (1.1–$`1.24)\times 10^{22}\mathrm{cm}^2`$ Dickey & Lockman (1990); Stark et al. (1992), which translates to an $`A_V=6.1`$–6.9 (assuming a standard dust to gas ratio; Predehl & Schmitt, 1995). Alternatively, the reddening estimated from dust IR emission is higher at $`A_V=10.2`$ Schlegel et al. (1998). The USNO A2.0 astrometric catalog Monet et al. (1998) contains just two stars within the $`1^{}`$ error circle of 4U 1901+03 having colors consistent with early-type stars suffering extinction with $`A_V>6`$ (i.e. $`BR3`$; Fig. 3.4, upper panel). We obtained low-resolution spectra of these two candidates using the Low Dispersion Survey Spectrograph (LDSS-2) on the 6.5 m Clay (Magellan II) telescope at Las Campanas, Chile. We accumulated two 600 s spectra of each candidate on 2003 August 10, using the medium red grism with a $`1\mathrm{}`$ long slit, covering the range 4500–9000 Å. However, the overall spectral features suggest that these candidates are instead low-mass K0–5 stars (Fig. 3.4, lower panel), which are ruled out as the counterpart on the basis of the X-ray mass function. Clearly, it is not possible to distinguish between early-type stars with high extinction and nearby late-type stars from optical photometry alone.
Thus, we also examined the $`J`$, $`H`$ and $`K`$ magnitudes of candidates within $`1^{}`$ of the X-ray position from the 2MASS point source catalogue<sup>2</sup><sup>2</sup>2http://www.ipac.caltech.edu/2mass/releases/allsky/doc/explsup.html. The expected range of colors for a B-star for the estimated extinction range are $`JH=0.5`$–1.1 and $`HK=0.3`$–0.6 Cox (2000). The colors expected for a supergiant counterpart fall within similar ranges, slightly larger in $`JH`$ and smaller in $`HK`$. We found approximately ten stars in the $`1^{}`$ RXTE error circle with colors within these ranges, including the optically identified candidates A and B. Star B was the brightest in the IR bands, with $`J=11.5`$; this is the only star consistent with the expected brightness of a supergiant counterpart at $`10`$ kpc, and can be ruled out as the counterpart on the basis of our spectroscopic observations. Two other stars had $`J`$ magnitudes similar to that of star A, at $`J13`$; we expect these candidates are also nearby low-mass stars, like star A. The remaining candidates all had $`J>14`$, much fainter than the limit of $`J12`$ expected for a supergiant companion at $`d10`$ kpc. Several of these stars were not detected in the DSS image, suggesting that $`B20`$ perhaps consistent with the upper end of the estimated $`A_V`$ range towards the source.
## 4. Discussion
4U 1901+03 is one of a small, but growing class of low-eccentricity high-mass X-ray binaries. Although we were unable to identify the optical counterpart, the limit of $`J13`$ for stars in the $`1^{}`$ RXTE error circle rules out a supergiant companion, unless the distance to the source is $`>10`$ kpc.
Furthermore, the source is located on the Corbet diagram Corbet (1986) with confirmed Be transients such as 4U 0115+63 (e.g. Bildsten et al., 1997). Like that source, 4U 1901+03 exhibits infrequent outbursts that can span multiple orbital periods, a behavior quite unlike the typically persistent activity of partially Roche-lobe filling supergiants. If the mass donor is not filling it’s Roche lobe, the efficacy of tidal forces in circularizing such a wide orbit is negligible, which makes the present low eccentricity of $`0.036`$ difficult to understand (given the initially eccentric orbit expected to arise as a result of the natal supernova kick). As suggested by Pfahl et al. (2002), the formation events for low-eccentricity O-B transients like 4U 1901+03 may be dynamically distinct from the more common high-eccentricity binaries due to a much lower initial kick to the neutron star.
The lack of eclipses indicates the inclination is $`85\mathrm{°}`$. The 95% upper limit on the companion mass for an a priori isotropic distribution of inclination angles is $`88M_{}`$. The estimated peak bolometric luminosity (for a distance of 10 kpc) was $`1.1\times 10^{38}\mathrm{ergs}\mathrm{s}^1`$, and the integrated luminosity over the course of the outburst was $`7.4\times 10^{44}(d/10\mathrm{kpc})^2\mathrm{ergs}`$ (for a neutron star with $`R=10`$ km and $`M=1.4M_{}`$). The maximum intensity observed by Uhuru during the 1970–1 outburst was $`87\pm 11\mathrm{count}\mathrm{s}^1`$ at epoch 1971.0, corresponding to $`1.5\times 10^9\mathrm{ergs}\mathrm{cm}^2\mathrm{s}^1`$ in the range 2–6 keV Forman et al. (1976). Our broadband RXTE spectra suggest that 20–30% of the source flux is emitted in this energy range, so that the estimated maximum bolometric luminosity for the Uhuru observations was $`8\times 10^{37}\mathrm{ergs}\mathrm{s}^1`$ (for $`d=10`$ kpc), consistent with the peak RXTE value to within the error. Vela 5B measured a peak of $`8\mathrm{count}\mathrm{s}^1`$ around epoch 1970.9 Priedhorsky & Terrell (1984). The maximum observed flux was thus $`3.6\times 10^9\mathrm{ergs}\mathrm{cm}^2\mathrm{s}^1`$ in the range 3–12 keV, in which range 60% of the flux observed by RXTE is emitted; thus, the estimated maximum bolometric luminosity measured by Vela 5B was $`7\times 10^{37}\mathrm{ergs}\mathrm{cm}^2\mathrm{s}^1`$, roughly consistent with both the Uhuru measurements of the 1970–1 outburst and the the peak measured by RXTE for the 2003 outburst. We also note that the estimated duration of the 1970–1 outburst (excluding the peak around epoch 1970.6) was at least 120 d, similar to the 150 d duration of the 2003 outburst. No emission prior to 2003 February was detected by the RXTE/ASM, although there is the possibility that one or more intervening outbursts occurred sometime between 1971 and the launch of RXTE in 1995. A similarly bright outburst before 1980 would probably have been detected by Vela 5B or the ASM onboard Ariel 5 Holt (1976). An outburst between 1987 and 1996 would likely have been detected with the Ginga ASM (operational until 1991 November; Tsunemi et al., 1989) or the BATSE experiment onboard CGRO (1991 April to 2000 June; Zhang et al., 1995). Assuming no intermediate outbursts occurred, we derive a time-averaged accretion rate for a 32.2 yr recurrence time of $`8.1\times 10^{11}(d/10\mathrm{kpc})^2M_{}\mathrm{yr}^1`$. We note that the 32.2 yr interval between outbursts in 4U 1901+03 may be the longest presently known for any X-ray transient.
Whilst no cyclotron absorption features were detected in the X-ray spectrum, the observed range of source flux over which pulsations were detected allows a rough estimate of the dipole magnetic field strength of the neutron star. Assuming that a disk is present, it must be truncated above the surface of the neutron star in order to allow the magnetic field to channel the accreting material and produce observable pulsations. The truncation radius $`r_M`$ is inversely proportional to the mass accretion rate (e.g. Frank et al., 1992), so that the requirement for $`r_M>R_{}`$ (where $`R_{}10`$ km is the neutron star radius) even at the peak of the outburst implies a lower limit on the magnetic field strength (although for 4U 1901+03 this limit is several orders of magnitude below the canonical field strength for long-period pulsars of $`10^{12}`$ G). For accretion to be dynamically feasible also requires that the inner disk radius is within the corotation radius $`r_{\mathrm{co}}`$, i.e. the radius at which the Keplerian orbital frequency equals the neutron star spin frequency. For the lowest flux at which pulsations were detected, this implies that $`B<0.5\times 10^{12}(d/10\mathrm{k}\mathrm{p}\mathrm{c})^2`$ G, giving a fundamental cyclotron frequency of $`\nu _{\mathrm{cyc}}4(d/10\mathrm{k}\mathrm{p}\mathrm{c})^2`$ keV. This is consistent with the absence of detectable cyclotron features in the broadband spectrum, although it is possible that the residuals frequently present at $`10`$ keV may arise from higher harmonics of a low-energy cyclotron absorption line. We can also estimate the magnetic moment by assuming that the long-term mass accretion has left the pulsar close to spin equilibrium, i.e. $`r_Mr_{\mathrm{co}}`$ (e.g. Bildsten et al., 1997). In that case, we find $`B0.3\times 10^{12}(d/10\mathrm{k}\mathrm{p}\mathrm{c})^{6/7}`$ G, consistent with the above estimate. We note that the inferred limit on $`\nu _{\mathrm{cyc}}`$ and the low values of $`kT_e`$ are qualitatively consistent with the observed correlation in other pulsars between the cyclotron energy and the spectral cutoff (e.g. Coburn et al., 2002).
Identification of the counterpart to 4U 1901+03 is essential to confirm the nature of the mass donor suggested by these observations, which may prove difficult unless the source becomes active again. If the outbursts occur regularly every $`30`$ yr, significantly more advanced X-ray instruments may be available for the next outburst, perhaps sufficient to measure the position more precisely and identify the optical counterpart, as well as resolve the residuals below 10 keV and measure the properties of the low-energy cyclotron lines, thus allowing direct measurement of the field strength.
This research has made use of data obtained through the High Energy Astrophysics Science Archive Research Center Online Service, provided by the NASA/Goddard Space Flight Center. The Second Palomar Observatory Sky Survey (POSS-II) was made by the California Institute of Technology with funds from the National Science Foundation, the National Geographic Society, the Sloan Foundation, the Samuel Oschin Foundation, and the Eastman Kodak Corporation. 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. This work was supported in part by the NASA Long Term Space Astrophysics program under grant NAG 5-9184.
|
warning/0506/gr-qc0506076.html
|
ar5iv
|
text
|
# Q-stars in scalar-tensor gravitational theories
## 1 Introduction
Interesting alternative gravitational theories are the scalar-tensor gravitational theories, which appeared at the original paper of Brans and Dicke , where the Newtonian constant $`G`$ was replaced by a scalar field $`\varphi _{\text{BD}}`$, and the total action contained kinetic terms for the new field times an $`\omega _{\text{BD}}`$ quantity. $`\omega _{\text{BD}}`$ was regarded as a constant in the original paper. The theory was generalized in a series of papers, , mainly in the direction of replacing the constant $`\omega _{\text{BD}}`$ with a function of the Brans-Dicke (BD) scalar field.
Boson stars appeared in the literature as stable field configurations of massive scalar matter with a global $`U(1)`$ symmetry, coupled to gravity , now known as “mini” boson stars due to their small relative magnitude. Other works took into account self interactions or the case of local symmetry . Their common feature is that gravity plays the role of the non-linear interaction that stabilizes the star against decay into free particles.
When the scalar potential is of a special type, admitting stable non-topological soliton solutions in the absence of gravity, the soliton stars appear as relativistic generalizations of the above solitons. The so called “large” soliton stars, with radius of order of lightyears, discussed analytically in a series of papers by Friedberg, Lee and Pang, . Another class of soliton stars appeared as a generalization of q-balls. Q-balls are non-topological solitons in Lagrangians with a global $`U(1)`$ symmetry, , or a local one, , or a global $`SU(3)`$ or $`SO(3)`$ symmetry, . Q-balls are supposed to appear in the flat directions of the superpotential in supersymmetric extensions of Standard Model, , and play a special role in the baryogenesis, . Q-stars are relativistic extensions of q-balls, with one or two scalar fields and a global, , or local, , $`U(1)`$ symmetry, non-abelian symmetry, , or with fermions and a scalar field, in asymptotically flat or anti de Sitter spacetime, . Any type of the bosonic stars may offer a solution to the problem of Dark Matter, when the q-stars have the additional feature to be of the same order of magnitude as neutron stars and, generally, as other stellar objects.
Within the BD gravitational framework, Gunderson and Jensen investigated the coupling of a scalar field with quartic self-interactions with the metric and the BD scalar field, $`\varphi _{\text{BD}}`$, . The properties of boson stars within this framework have been extensively studied in a series of papers . Their results generalized in scalar-tensor gravitational theories, where $`\omega _{\text{BD}}`$ is no more a constant, but a function of the BD field . The case of charged boson-stars in a scalar-tensor gravitational theory has been analyzed in .
In the present article we follow the work of -. Our aim is to study the formation of non-topological soliton stars in the context of BD or general scalar-tensor gravitational theory, their stability with respect to fission into free particles and to gravitational collapse, and the influence of $`\omega _{\text{BD}}`$ in the star parameters. We also compare our results with those obtained in the framework of General Relativity.
## 2 Q-stars with one scalar field
We consider a static, spherically symmetric metric:
$$ds^2=e^\nu dt^2+e^\lambda d\rho ^2+\rho ^2d\mathrm{\Omega }^2,$$
(1)
with $`g_{tt}=e^\nu `$. In order to realize such a spacetime, we regard both the matter and the BD fields as spherically symmetric and the former with an harmonic time dependence, assuring minimum energy for the matter field, and the latter time independent. If $`\varphi `$ is the matter field and $`\varphi _{\text{BD}}`$ the BD field, we write the action for the BD theory:
$`S={\displaystyle \frac{1}{16\pi }}`$ $`{\displaystyle d^4x\sqrt{g}\left(\varphi _{\text{BD}}R\omega _{\text{BD}}g^{\mu \nu }\frac{_\mu \varphi _{\text{BD}}_\nu \varphi _{\text{BD}}}{\varphi _{\text{BD}}}\right)}`$
$`+`$ $`{\displaystyle d^4x\sqrt{g}_{\text{matter}}},`$ (2)
with:
$$_{\text{matter}}=(_\mu \varphi )^{}(^\nu \varphi )U,$$
(3)
and $`\omega _{\text{BD}}`$ a constant in BD gravity and a certain function of the $`\varphi _{\text{BD}}`$ field in a generalized scalar-tensor gravitational theory, which we will discuss later.
Varying the action with respect to the metric and scalar fields we obtain the Einstein and Lagrange equations respectively, as follows:
$`G_{\mu \nu }={\displaystyle \frac{8\pi }{\varphi _{\text{BD}}}}T_{\mu \nu }+{\displaystyle \frac{1}{\varphi _{\text{BD}}}}(\varphi _{\text{BD}}^{}{}_{,\mu ;\nu }{}^{}g_{\mu \nu }\varphi _{\text{BD}}^{}{}_{;\lambda }{}^{;\lambda })`$
$`+{\displaystyle \frac{\omega _{\text{BD}}}{\varphi _{\text{BD}}^2}}\left(_\mu \varphi _{\text{BD}}_\nu \varphi _{\text{BD}}{\displaystyle \frac{1}{2}}g_{\mu \nu }_\lambda \varphi _{\text{BD}}^\lambda \varphi _{\text{BD}}\right),`$ (4)
$$\varphi _{;\lambda }^{;\lambda }\frac{dU}{d|\varphi |^2}\varphi =0,$$
(5)
$$\frac{2\omega _{\text{BD}}}{\varphi _{\text{BD}}}\varphi _{\text{BD}}^{}{}_{;\lambda }{}^{;\lambda }\omega _{\text{BD}}\frac{^\lambda \varphi _{\text{BD}}_\lambda \varphi _{\text{BD}}}{\varphi _{\text{BD}}^2}+R=0.$$
(6)
$`G_{\mu \nu }`$ is the Einstein tensor, $`T_{\mu \nu }`$ the energy momentum tensor for the matter field given by:
$$T_{\mu \nu }=(_\mu \varphi )^{}(_\nu \varphi )+(_\mu \varphi )(_\nu \varphi )^{}g_{\mu \nu }[g^{\alpha \beta }(_\alpha \varphi )^{}(_\beta \varphi )]g_{\mu \nu }U$$
(7)
and $`R`$ is the scalar curvature. Tracing eq. 2 we take:
$$\frac{8\pi }{\varphi _{\text{BD}}}T\frac{\omega _{\text{BD}}}{\varphi _{\text{BD}}^2}^\lambda \varphi _{\text{BD}}_\lambda \varphi _{\text{BD}}+\frac{3}{\varphi _{\text{BD}}}\varphi _{\text{BD}}^{}{}_{;\lambda }{}^{;\lambda }=R.$$
with $`T`$ the trace of the energy-momentum tensor. Substituting in eq. 6 we find:
$$\varphi _{\text{BD}}^{}{}_{;\lambda }{}^{;\lambda }=\frac{8\pi }{2\omega _{\text{BD}}+3}T.$$
(8)
The above results hold true for every case of bosonic, spherically symmetric, static field configurations coupled to BD gravity.
We will now insert the q-soliton ansatz writing:
$$\varphi (\stackrel{}{\rho },t)=\sigma (\rho )e^{ı\omega t}.$$
(9)
with $`\omega `$ the frequency with which the q-soliton rotates within its internal $`U(1)`$ space. The Lagrange equation for the $`\varphi `$ field is:
$$\sigma ^{\prime \prime }+[2/\rho +(1/2)(\nu ^{}\lambda ^{})]\sigma ^{}+e^\lambda \omega ^2e^\nu \sigma e^\lambda \frac{dU}{d\sigma ^2}\sigma =0.$$
(10)
We define:
$$A=e^\lambda ,B=e^\nu ,$$
(11)
$$\begin{array}{c}\hfill We^\nu \left(\frac{\varphi }{t}\right)^{}\left(\frac{\varphi }{t}\right)=e^\nu \omega ^2\sigma ^2,\\ \hfill Ve^\lambda \left(\frac{\varphi }{\rho }\right)^{}\left(\frac{\varphi }{\rho }\right)=e^\lambda \sigma _{}^{}{}_{}{}^{2}\end{array}$$
(12)
and rescale:
$$\begin{array}{cc}\hfill \stackrel{~}{\rho }=\rho m,\stackrel{~}{\omega }& =\omega /m,\stackrel{~}{\varphi }=\varphi /m,\hfill \\ \hfill \stackrel{~}{U}=U/m^4,\stackrel{~}{W}& =W/m^4,\stackrel{~}{V}=V/m^4.\hfill \end{array}$$
(13)
In roughly approximation, gravity becomes important when $`RG(R)`$, where $`R`$ is defined as the radius, within which the matter field differs from zero, and $`(R)`$ is the mass trapped within this area. For the case of q-solitons the eigen-frequency is of the same order of magnitude as the mass and the absolute value of the scalar field: $`\omega m\sigma `$. If $``$ is the energy density, then: $`Um^4`$. Remembering that $`(R)_0^Rd^3\rho `$, we find that for a q-star:
$$Rϵ^1,ϵ\sqrt{8\pi Gm^2},$$
so if we redefine:
$$\stackrel{~}{r}=ϵ\stackrel{~}{\rho },$$
(14)
we expect $`\stackrel{~}{r}1`$. We also use a suitable rescaled potential, admitting q-ball type solutions in the absence of gravity, namely:
$$\stackrel{~}{U}=|\stackrel{~}{\varphi }|^2\left(1|\stackrel{~}{\varphi }|^2+\frac{1}{3}|\stackrel{~}{\varphi }|^4\right)=\stackrel{~}{\sigma }^2\left(1\stackrel{~}{\sigma }^2+\frac{1}{3}\stackrel{~}{\sigma }^4\right).$$
(15)
Dropping form now on the tildes, we will use the Lagrange equation to find an analytical solution for the matter field. $`ϵ`$ is a very small quantity for $`m`$ of the order of some (hundreds) $`GeV`$, so ignoring the $`O(ϵ)`$ terms from the Lagrange equation, we find:
$$\sigma ^2=1+\omega B^{1/2},U=\frac{1}{3}(1+\omega ^3B^{3/2}).$$
(16)
The surface is determined by the star radius. The radius of the solitonic configuration is defined as the radius within which the *matter* Lagrangian differs from zero. Outside this radius the matter Lagrangian is zero, when the BD Lagrangian *may not necessarily be*. The surface width is of order of $`m^1`$. Within this, the matter field varies very rapidly from a $`\sigma `$ value at the inner edge of the surface, to zero at the outer, but the metric fields vary very slowly. So, dropping from the Lagrange equation the $`O(ϵ)`$ terms we take:
$$\frac{\delta (WUV)}{\delta \sigma }=0.$$
(17)
The above equation can be straightforward integrated and, because all energy quantities are zero at the outer edge of the surface, the result gives the following equation, holding true only within the surface:
$$V+WU=0.$$
(18)
At the inner edge of the surface $`\sigma ^{}`$ is zero in order to match the interior with the surface solution. So, at the inner edge of the surface the equality $`W=U`$ together with eq. 16 gives:
$$\omega =\frac{A_{\text{sur}}^{1/2}}{2}=\frac{B_{\text{sur}}^{1/2}}{2}.$$
(19)
Eq. 19 is the eigenvalue equation for the frequency of the q-star, revealing the relation between a feature of the star and the spacetime curvature.
Redefining:
$$\mathrm{\Phi }_{\text{BD}}=\frac{2\omega _{\text{BD}}+3}{2\omega _{\text{BD}}+4}G\varphi _{\text{BD}},$$
(20)
and dropping the $`O(ϵ)`$ terms, the Lagrange equation for the BD field and the Einstein equations take the form respectively:
$$A\left[\frac{d^2\mathrm{\Phi }_{\text{BD}}}{dr^2}+\left(\frac{2}{r}+\frac{1}{2A}\frac{dA}{dr}\frac{1}{2B}\frac{dB}{dr}\right)\frac{d\mathrm{\Phi }_{\text{BD}}}{dr}\right]=\frac{2W4U}{2\omega _{\text{BD}}+4},$$
(21)
$`{\displaystyle \frac{A1}{r^2}}+{\displaystyle \frac{1}{r}}{\displaystyle \frac{dA}{dr}}={\displaystyle \frac{2\omega _{\text{BD}}+3}{(2\omega _{\text{BD}}+4)\mathrm{\Phi }_{\text{BD}}}}\left(WU{\displaystyle \frac{2W4U}{2\omega _{\text{BD}}+3}}\right)`$
$`{\displaystyle \frac{\omega _{\text{BD}}A}{2\mathrm{\Phi }_{\text{BD}}^2}}\left({\displaystyle \frac{d\mathrm{\Phi }_{\text{BD}}}{dr}}\right)^2{\displaystyle \frac{A}{2\mathrm{\Phi }_{\text{BD}}B}}{\displaystyle \frac{dB}{dr}}{\displaystyle \frac{d\mathrm{\Phi }_{\text{BD}}}{dr}},`$ (22)
$`{\displaystyle \frac{A1}{r^2}}{\displaystyle \frac{A}{B}}{\displaystyle \frac{1}{r}}{\displaystyle \frac{dB}{dr}}={\displaystyle \frac{2\omega _{\text{BD}}+3}{(2\omega _{\text{BD}}+4)\mathrm{\Phi }_{\text{BD}}}}\left(WU{\displaystyle \frac{2W4U}{2\omega _{\text{BD}}+3}}\right)`$
$`+{\displaystyle \frac{\omega _{\text{BD}}A}{2\mathrm{\Phi }_{\text{BD}}^2}}\left({\displaystyle \frac{d\mathrm{\Phi }_{\text{BD}}}{dr}}\right)^2+{\displaystyle \frac{A}{\mathrm{\Phi }_{\text{BD}}}}\left({\displaystyle \frac{d^2\mathrm{\Phi }_{\text{BD}}}{dr^2}}+{\displaystyle \frac{1}{2A}}{\displaystyle \frac{dA}{dr}}{\displaystyle \frac{d\mathrm{\Phi }_{\text{BD}}}{dr}}\right),`$ (23)
with boundary conditions:
$$A(0)=1,A(\mathrm{})=1/B(\mathrm{})=1,\mathrm{\Phi }_{\text{BD}}^{}=0,\mathrm{\Phi }_{\text{BD}}(\mathrm{})=1,$$
(24)
where the first condition reflects the freedom to define the $`g_{\rho \rho }`$ metric at least locally when the second arises from the flatness of the spacetime. One may alternatively, instead of eq. 2, use the relation resulting from the Schwarzschild formula: $`A(\rho )=1\frac{2G(\rho )}{\rho }`$, which can be written with our rescalings as:
$$A=1\frac{(r)}{4\pi r}.$$
(25)
With the new variable eq. 2 takes the form:
$`{\displaystyle \frac{1}{4\pi r^2}}{\displaystyle \frac{d}{dr}}={\displaystyle \frac{2\omega _{\text{BD}}+3}{(2\omega _{\text{BD}}+4)\mathrm{\Phi }_{\text{BD}}}}\left(WU{\displaystyle \frac{2W4U}{2\omega _{\text{BD}}+3}}\right)`$
$`{\displaystyle \frac{\omega _{\text{BD}}\left(1\frac{}{4\pi r}\right)}{2\mathrm{\Phi }_{\text{BD}}^2}}\left({\displaystyle \frac{d\mathrm{\Phi }_{\text{BD}}}{dr}}\right)^2{\displaystyle \frac{1\frac{}{4\pi r}}{2\mathrm{\Phi }_{\text{BD}}B}}{\displaystyle \frac{dB}{dr}}{\displaystyle \frac{d\mathrm{\Phi }_{\text{BD}}}{dr}},`$ (26)
with $`(0)=0`$ which reflects the absence of anomalies at the center of the star.
The stability of the star results from a conserved Noether charge. There is a Noether current due to the global $`U(1)`$ symmetry defined as:
$$j^\mu =\sqrt{g}g^{\mu \nu }ı(\varphi ^{}_\nu \varphi \varphi _\nu \varphi ^{})$$
(27)
and a conserved Noether charge defined as:
$$Q=d^3xj^0=8\pi 𝑑rr^2\omega \sigma ^2\sqrt{B/A}.$$
(28)
In our figures $`R`$ is in $`(8\pi Gm^4)^{1/2}`$ units, the total mass in $`(8\pi G)^3m^2`$ units and the charge in $`(8\pi Gm)^3`$ units. The total charge equals to the particle number if every single particle is assigned with a unity “baryon” number. The particle number also equals to the total energy of the free particles as their mass is taken to be unity. So, when the particle number exceeds the total mass, the star decays into free particles as the energetically favorable case. All the field configurations depicted in our figures are stable. An experimental lower limit for $`\omega _{\text{BD}}`$ is $`500`$, . The results obtained in the BD context coincide with general relativity when $`\omega _{\text{BD}}\mathrm{}`$. We investigate the phase space of the star with $`\omega _{\text{BD}}`$ varying between $`5`$ and $`1000`$, following the works of -, so as to explore thoroughly the influence of $`\omega _{\text{BD}}`$ in the features of the star. As a general result we find that the star parameters, mass, particle number, radius and absolute value of the scalar field at the center of the star, increase when $`\omega _{\text{BD}}`$ decreases.
## 3 General scalar-tensor theory
In the original BD gravitational theory $`\omega _{\text{BD}}`$ is a constant. In a more general theory it may be regarded as a function, usually of the BD field. We will use one of the forms that Barrow and Parsons investigated in a cosmological framework namely:
$$2\omega _{\text{BD}}+3=\omega _0\varphi _{\text{BD}}^n,$$
(29)
with $`\omega _0`$ and $`n`$ constants. This form for $`\omega _{\text{BD}}`$ gives an analytical solution, , for the metrics within the above mentioned cosmological framework. The Lagrange equation for the BD field is:
$$\varphi _{\text{BD}}^{}{}_{;\lambda }{}^{;\lambda }=\frac{1}{2\omega _{\text{BD}}+3}\left(8\pi T\frac{d\omega _{\text{BD}}}{d\varphi _{\text{BD}}}\varphi _{\text{BD}}^{}{}_{}{}^{,\rho }\varphi _{\text{BD}}^{}{}_{,\rho }{}^{}\right),$$
(30)
If we rescale:
$$\stackrel{~}{\omega }_0=\left(\frac{2\omega _{\text{BD}}+3}{2\omega _{\text{BD}}+4}\right)^nG^n\omega _0,$$
(31)
and the other quantities as in 13-14 eqs. and drop the tildes and the $`O(ϵ)`$ quantities we take for the Einstein and the Lagrange equation for the BD field:
$`G_t^t={\displaystyle \frac{\omega _0}{\omega _0\mathrm{\Phi }_{\text{BD}}+1}}[WU{\displaystyle \frac{1}{\omega _0\mathrm{\Phi }_{\text{BD}}}}\times `$
$`(2W4U{\displaystyle \frac{A\mathrm{\Phi }_{\text{BD}}^2}{2}}{\displaystyle \frac{\omega _0\mathrm{\Phi }_{\text{BD}}+1}{\mathrm{\Phi }_{\text{BD}}}})]`$
$`{\displaystyle \frac{\omega _0\mathrm{\Phi }_{\text{BD}}3}{2}}{\displaystyle \frac{A\mathrm{\Phi }_{\text{BD}}^2}{2\mathrm{\Phi }_{\text{BD}}^2}}{\displaystyle \frac{AB^{}\mathrm{\Phi }_{\text{BD}}^{}}{2\mathrm{\Phi }_{\text{BD}}B}},`$ (32)
$`G_r^r={\displaystyle \frac{\omega _0}{\omega _0\mathrm{\Phi }_{\text{BD}}+1}}[WU{\displaystyle \frac{1}{\omega _0\mathrm{\Phi }_{\text{BD}}}}\times `$
$`(2W4U{\displaystyle \frac{A\mathrm{\Phi }_{\text{BD}}^2}{2}}{\displaystyle \frac{\omega _0\mathrm{\Phi }_{\text{BD}}+1}{\mathrm{\Phi }_{\text{BD}}}})]`$
$`+{\displaystyle \frac{\omega _0\mathrm{\Phi }_{\text{BD}}3}{2}}{\displaystyle \frac{A\mathrm{\Phi }_{\text{BD}}^2}{2\mathrm{\Phi }_{\text{BD}}^2}}+{\displaystyle \frac{A\mathrm{\Phi }_{\text{BD}}^{\prime \prime }}{\mathrm{\Phi }_{\text{BD}}}}+{\displaystyle \frac{A^{}\mathrm{\Phi }_{\text{BD}}^{}}{2\mathrm{\Phi }_{\text{BD}}}},`$ (33)
$`A\left[\mathrm{\Phi }_{\text{BD}}^{\prime \prime }+\left({\displaystyle \frac{2}{r}}+{\displaystyle \frac{A^{}}{2A}}{\displaystyle \frac{B^{}}{2B}}\right)\mathrm{\Phi }_{\text{BD}}^{}\right]=`$
$`{\displaystyle \frac{1}{\omega _0\mathrm{\Phi }_{\text{BD}}+1}}\left[2W4U{\displaystyle \frac{A\mathrm{\Phi }_{\text{BD}}^2}{2}}{\displaystyle \frac{\omega _0\mathrm{\Phi }_{\text{BD}}+1}{\mathrm{\Phi }_{\text{BD}}}}\right].`$ (34)
We solved the coupled Einstein and and Lagrange equations for several integral or half-integral values values of $`n`$ and found that the star parameters are rather constant. This owes to the $`\mathrm{\Phi }_{\text{BD}}1`$ relation, because $`\omega _0`$ and $`\omega _{\text{BD}}`$ are not small enough so as to deviate considerably from the results of General Relativity. When $`\omega _0`$ decreases the star parameters are larger than in the case of General Relativity, and when $`\omega _0\mathrm{}`$ its results are reproduced.
## 4 Concluding remarks
We investigated q-stars in a BD gravitational theory. We also studied the case of q-stars in the framework of generalized scalar tensor theories, with $`\omega _{\text{BD}}`$ a simple polynomial function of the BD scalar. All the field configurations discussed here are stable with respect to fission into free particles as the ratio of their energy to the energy of the free particles, equal to the mass of the free particles times the particle number, is smaller than unity. We investigated their properties, particle number, mass, radius of the matter field configuration and the value of the matter scalar field at the center of the star. The free parameters of their phase space are mainly the eigenfrequency, straightforwardly connected with the surface gravity, and the value of the $`\omega _{\text{BD}}`$ or the $`\omega _0`$ for the generalized scalar-tensor theory.
We found that the star parameters, mass, particle number, radius and absolute value at the center, are in generally larger when $`\omega _{\text{BD}}`$ or $`\omega _0`$ is small and coincide with the results of general relativity when $`\omega _{\text{BD}},\omega _0\mathrm{}`$.
ACKNOWLEDGMENTS
I wish to thank N. D. Tracas and E. Papantonopoulos for helpful discussions.
|
warning/0506/quant-ph0506156.html
|
ar5iv
|
text
|
# Quantum State Transfer Characterized by Mode Entanglement
## I Introduction
Quantum entanglement is a fascinating feature of quantum theory of many body systems chung . The concurrence Wootters:98 , as a widely used measure of pairwise entanglement defined for the spin-1/2 systems, has been intensively investigated. Through various concurrences defined by different authors, people have explored the relations between entanglement and some physical observables such as energy and momentum etc. entangled rings ; Qian , as well as the relations between entanglement and some physical phenomena, such as quantum correlation Wang and quantum phase transitions etc. Osterloh:02 ; Lin ; Chen .
On the other hand, people have proposed many protocols for the quantum state transfer (QST) recently Bose ; Christandal ; Shi ; Li ; kaba . In these schemes based on quantum spin systems, almost without any spatial or dynamical control over the interactions among qubits, the quantum state can be transferred with high fidelity through a quantum channel, or quantum data bus, which is necessary for scalable quantum computations based on realistic silicon devices. The physical process of QST through a quantum spin system can be understood as a dynamical permutation or translation preserving the initial shape of a quantum state, which can be realized as a specific evolution of the total quantum spin system from an initial wave function localized around a single site of the lattice to a distant one. The basic feature of QST is characterized by fidelity, which is usually the overlap of the identical image of an initial state with its transferred counterpart.
This paper will be devoted to understand the intrinsic relation between quantum entanglement and QST for the engineered quantum spin chains, or quantitatively, between concurrence and fidelity. Some rigorous results are obtained to reveal the essential relationship between these two fascinating issues for the tight-bonding Bloch electrons. Actually, the QST from one location to another can be considered as perfect if the fidelity can reach its maximum value one at some instants. Literally, the perfect QST is a dynamic process starting from an initially factorized state (product state) to a finally factorized state through a middle process with the superposition of factorized states. Since a superposition of single particle states of Bloch electrons can be understood as a mode entanglement x wang , the studies of QST can be naturally referred to the various phenomena of quantum entanglement.
Motivated by arguments about the entanglement concurrence and the quantum correlations Wootters:98 ; x wang , we first define the mirror mode concurrence (MMC) $`C(t)`$ to characterize the mode entanglement of a wave packet in Bloch electron systems with mirror symmetry. It will be proved that the MMC is no less than the overlap of the wave packet at time $`t`$ with its mirror image. By quantitatively comparing the MMC with the time dependent fidelity $`F(t)`$ of QST, a novel complementary relation is discovered as the increase of $`F(t)`$ is accompanied by a decrease of $`C(t)`$ (vice versa). Especially, at the instant $`\tau /2,`$ where $`\tau `$ is the characteristic time to accomplish a perfect QST with $`F(\tau )=1`$, the MMC can reach its maximum
$$C(\tau /2)=\mathrm{max}(C(t))=1.$$
(1)
An engineered Bloch electron model with a certain spectrum structure, which admits perfect QST, is discovered and used to demonstrate this complementary relation through numerical simulations.
## II One-dimensional Bloch electron system with mirror symmetry
We consider a one-dimensional Bloch electron system in an engineered crystal lattice of $`N`$ sites with mirror symmetry with respect to the center of the lattice. The model Hamiltonian with tight-bonding approximation is written as
$$H=\underset{j=1}{\overset{N1}{}}J_ja_j^{}a_{j+1}+h.c.$$
(2)
in terms of the fermion creation (annihilation) operator $`a_j^{}`$ ($`a_j`$), where the site-dependent coupling constants $`J_j`$ are real. The single-particle space is spanned by $`N`$ basis vectors
$`|1=|1\text{}0\text{}0\text{}\mathrm{}\text{}0\text{}0,`$
$`|2=|0\text{}1\text{}0\text{}\mathrm{}\text{}0\text{}0,`$
$`\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}..,`$ (3)
$`|N=|0\text{}0\text{}0\text{}\mathrm{}\text{}0\text{}1`$
where $`|n_1\text{}n_2\text{}\mathrm{}\text{}n_N`$ ($`n_j=0`$, $`1`$) denotes the Fock state of fermion systems. Then the reflection operator $`R`$ is defined as $`R|j=|N+1j`$. Obviously, the mirror symmetry by $`[R,H]=0`$ means that $`J_j=J_{Nj}`$ .
We describe the localized electron state around the $`l`$-th site as a superposition $`|\psi _l=_jc_j|j`$ with the summation over the small domain containing the site $`l`$. This assumption means that $`|\psi _l`$ is a wave packet around the site $`l`$. If $`l^{}`$ denotes another site far from the site $`l`$, we can approximately assume the vanishing overlap $`\psi _l|\psi _l^{}0`$ for two wave packets $`|\psi _l`$ and $`|\psi _l^{}`$. With this assumption the perfect QST is described as the dynamic process that the initial state $`|\psi _l`$ can evolve exactly into its mirror image. Mathematically, the time evolution operator $`U(t)=\mathrm{exp}(iHt)`$ becomes the reflection operator $`R`$ at the instant $`\tau `$, i.e., $`U(\tau )=R`$. We define the fidelity as
$$F_j(t)=\left|R\psi _j\left|U(t)\right|\psi _j\right|=\left|\psi _j\left|R^{}U(t)\right|\psi _j\right|.$$
(4)
A perfect QST can be depicted by the maximized fidelity $`F_j(\tau )=1`$.
Now we can intuitionally recognize that QST phenomenon is associated with the mode entanglement. In the terminology of mode entanglement, the single electron state
$`|E`$ $`=`$ $`\alpha |1+\beta |N`$
$``$ $`\alpha |1,0,\mathrm{},0+\beta |0,0,\mathrm{},1`$
can be regarded as an entangled state if the single fermion at the $`1`$-th site and $`N`$-th site can be probed in principle x wang . In this sense $`|\psi _l`$ can be viewed as an $`N`$-component entanglement. The perfect QST from $`|1`$ to $`|N`$ through the middle state $`|\psi (t)=U(t)|1`$ can be understood as a dynamic process starts from a localized (unentangled) state $`|1`$ to another localized state $`|N`$ through the entangled state $`|\psi (t)`$.
## III Mirror mode concurrence as the fingerprints of perfect QST
Actually a QST is a process, during which mode entanglement is generated first and then destroyed. To quantitatively characterize this dynamic feature, we define the mirror mode concurrence (MMC)
$$C(t)=\underset{j=1}{\overset{N/2}{}}C_{j,N+1j}$$
(6)
with respect to a pure state, evolved from a localized initial wave packet, $`|\psi (t)=U(t)|\psi (0)`$. Here each term $`C_{j,N+1j}`$ in the summation concerns two separated sites, the site $`j`$ and its mirror imagine $`l=N+1j,`$ and is defined by the pairwise mode concurrence x wang
$$C_{jl}=2\mathrm{max}\{0,\text{ }\left|Z_{jl}\right|\sqrt{X_{jl}^+X_{jl}^{}}\},$$
(7)
constructed in terms of the correlation functions
$$Z_{jl}=a_j^{}a_l,X_{jl}^+=\widehat{n}_j\widehat{n}_l,$$
(8)
and
$$X_{jl}^{}=(1\widehat{n}_j)(1\widehat{n}_l).$$
(9)
where the average $``$ is defined with respect to the pure state $`|\psi (t).`$
The physical significance will be in two folds that explicitly reveals the close relationship between the mode entanglement and the dispersion of the wave packet in time evolution. Firstly we notice that $`Z_{jl}`$ and $`X_{jl}^{}`$ are the non-zero elements of the two mode reduced density matrixx wang
$`\rho _{jl}`$ $`=`$ $`tr_{N2}(|\psi (t)\psi (t)|)`$ (14)
$`=`$ $`\left(\begin{array}{cccc}X_{jl}^+& & & \\ & Y_{jl}^+& Z_{jl}^{}& \\ & Z_{jl}& Y_{jl}^{}& \\ & & & X_{jl}^{}\end{array}\right)`$
for a system conserving total particle number, where $`tr_{N2}`$ means tracing over the variables besides the two on the sites $`j`$ and $`l=N+1j`$. In the single-particle subspace we have $`X_{jl}^+=0`$ and thus
$$C(t)=\underset{j=1}{\overset{N}{}}\left|Z_{j,N+1j}\right|.$$
(15)
It is obvious that MMC is a generalization of the usual entanglement measure–concurrence and thus characterize the quantum entanglement in some sense.
Secondly the MMC defined above have a geometric interpretation for the dynamic dispersion of the wave packet. We rewrite the MMC as
$$C(t)=\underset{j=1}{\overset{N}{}}\left|\psi (j,t)\right|\left|\psi (N+1j,t)\right|$$
(16)
where $`\psi (j,t)=j|\psi (t)`$. It is easy to show that
$`C(t)`$ $``$ $`|{\displaystyle \underset{j=1}{\overset{N}{}}}\psi (t)|jj\left|R\right|\psi (t)|`$ (17)
$`=`$ $`\left|\psi (t)|R|\psi (t)\right|`$
where we have used $`RR^{}=R^{}R=1`$. The above equation clearly implies that $`C(t)`$ is no less than the overlap integral of the state $`|\psi (t)`$ with its mirror image. Especially, for a large class of states $`|\psi (t)=_{j=1}^Nc_j|j`$ listed in two situations as follows, $`C(t)`$ is exactly equal to the overlap integral: (i) The electronic wave function are completely localized in a finite domain $`D=[1,N/2]`$ with no overlap with its mirror image $`[N/2,N]`$. In this case, the MMC vanishes exactly. (ii) The the coefficients of each pair of mirror symmetric non-zero components in $`|\psi (t)`$ have the same sign or opposite sign.
For a perfect QST accomplished at the instant $`t=\tau `$, the evolution operator $`U(\tau )`$ becomes the reflection operator $`R`$ and $`|\psi (0)`$ evolves exactly into its mirror image $`R|\psi (0)`$. Since the initial wave packet $`|\psi (0)`$ is usually a very localized wave function, the wave function $`|\psi (\tau )=R|\psi (0)`$ and its mirror image $`|\psi (0)`$ almost do not overlap with each other (see the illustration in Fig. 1). Thus we have
$$C(\tau )=C(0)\left|\psi (0)|R|\psi (0)\right|0.$$
(18)
Therefore, at $`t=\tau `$, the MMC $`C(\tau )`$ almost vanishes when the fidelity $`F(t)`$ reaches its maximum ($`F(\tau )=1`$).
From the above argument we see that there exists a quite interesting relationship between entanglement and fidelity. We provide a model more universal than the QST model in Ref. . Their mode is a mapping to the collective spin system with $`SU(2)`$ dynamic symmetry (by $`J_j=J_0\sqrt{j\left(Nj\right)}`$ more concretely), but our model only requires a much smaller mirror symmetry (by $`J_j=J_{Nj}`$ more generally ) and thus have much wider applications. In fact we have shown many examples in Ref. Li as well as in the following discussions. The further arguments about the complementarity relationship between entanglement and fidelity will also be presented in such a general framework.
## IV Maximal mode entanglement
The above analysis has confirmed our intuition about the complementary relation between the fidelity of QST and the MMC of mode entanglement. As for the other feature of this complementary relation, we need to consider when the MMC can reach its maximum.
Obviously there exists the inequality
$$C(t)\frac{1}{2}\underset{j=1}{\overset{N}{}}(\left|\psi (j,t)\right|^2+\left|\psi (N+1j,t)\right|^2)=1,$$
(19)
which takes the equal sign only when the wave function evolves into its mirror imagine, i.e.,
$$\left|\psi (j,t)\right|=\left|\psi (N+1j,t)\right|$$
(20)
at some instants $`t`$. This means that $`C(t)`$ will reach its maximum $`\mathrm{max}(C(t))=1`$ at the instants when Eq. (20) holds.
In order to determine the time when $`C(t)`$ reaches its maximum one, we need to solve the equation (20) about time $`t`$. To this end we use a time-independent real symmetric matrix $`W`$ to diagonalize the Hamiltonian $`H`$ or the evolution operator $`U(t)`$ as $`WU^{}(t)W^T=A(t)`$, where $`A(t)`$ is a diagonal matrix. With these notations, the above equation (20) can be transformed into
$$\left|\psi _W\left|A(t)\right|W_j\right|=\left|\psi _W\left|Q(t)\right|W_j\right|,$$
(21)
where
$`|\psi _W`$ $`=`$ $`W|\psi (0),`$
$`|W_j`$ $`=`$ $`W|j,`$ (22)
$`Q(t)`$ $`=`$ $`WU^{}(t)U(\tau )W^T,`$
We notice that, in general, $`|W_j=W|j`$. $`|\psi _W`$ and $`|W_j`$ are real for a real initial state $`|\psi (0)`$. Then the solutions to the equation (21) are sufficiently given by $`Q(t)=A^{}(t)`$ or $`Q(t)=A(t)`$, of which the non-trivial one is just $`t=\tau /2`$. Indeed, since $`\alpha \left|A\right|\beta =\beta \left|A\right|\alpha `$ for any two real vectors $`|\alpha `$ and $`|\beta `$, we have
$`\left|\psi _W\left|Q(t)\right|W_j\right|`$ $`=`$ $`\left|\psi _W\left|A^{}(t)\right|W_j\right|`$
$`=`$ $`\left|W_j\left|A^{}(t)\right|\psi _W\right|=\left|\psi _W\left|A(t)\right|W_j\right|.`$
Therefore, the solution $`t=\tau /2`$ is obviously given by $`Q(t)=A^{}(t)`$ or
$$U^{}(t)U(\tau )=U(t).$$
(24)
We summarize the above argument as a proposition: If $`F(t)`$ reach its maximum $`1`$ at the instant $`t=\tau `$, then at time $`t=\tau /2`$$`C(\tau /2)=1`$. In appendix A, we will prove its inverse proposition: if $`C(t)`$ reach its maximum $`1`$ at the instant $`t=\tau /2`$, then at time $`t=\tau `$$`F(\tau )=1`$. Furthermore, we can generalize these conclusion for the more general situation even with a higher dimensional Hamiltonian (also see the appendix A)
The solution $`t=\tau /2`$ to the equation (21) indicates that the time required to form the maximal mode entanglement is just half of the time needed to implement the perfect QST. Furthermore we can prove that, for a real vector $`|\psi (0)`$, the MMC $`C(t)`$ is symmetric with respect to both $`t=\tau /2`$ and $`t=\tau `$, namely,
$`C({\displaystyle \frac{\tau }{2}}t)`$ $`=`$ $`C({\displaystyle \frac{\tau }{2}}+t)\text{,}`$ (25)
$`C(\tau t)`$ $`=`$ $`C(\tau +t).`$
Actually, for the second equation in Eqs. (25) we have
$$C(\tau \pm t)=\underset{j=1}{\overset{N}{}}\left|\psi (0)\left|U_\pm (t)R^{}\right|j\right|\left|\psi (0)\left|U_\pm (t)\right|j\right|,$$
(26)
where $`U(t)_+=U^{}(t)`$ and $`U(t)_{}=U(t)`$. Obviously the second equation in Eqs. (25) holds since we have
$$\left|\psi (0)\left|U(t)V\right|j\right|=\left|\psi (0)\left|U^{}(t)V\right|j\right|$$
(27)
for $`V=1,`$ $`R^{}`$. Also, the first equation in Eqs. (25) will give a similar proof.
Numerical methods are now employed to give a demonstration of the above analytical results. We concern a class of schemes that admit perfect QST, which are presented in Ref. Shi . The couplings of the Hamiltonian $`H`$ are given that
$$J_j=J_0\sqrt{(j+\theta _jk)\left(Nj+\theta _jk\right)}$$
(28)
where $`\theta _j=1(1)^j`$, $`k=0`$, $`1`$, $`2`$, $`\mathrm{}`$ and $`J_0`$ is a constant. This model possesses a commensurate structure of energy spectrum that is matched with the corresponding parity. We demonstrate the exact numerical results of the models with $`N=4`$ and $`k=0`$, $`4`$ in Figs. 2(a) and 2(b). Actually, when $`k=0`$ (Fig. 2(a)), the model is just the one proposed in Ref. Christandal . We have used the localized initial wave packet as $`|\psi _{1,2}(0)=c_1|1+c_2|2`$, where $`c_1=5/6`$, $`c_2=\sqrt{11/36}`$. From Figs. 2(a) and 2(b) we can observe that $`C(0)=C(\tau )=0`$, $`C(\tau /2)=1`$, and $`C(t)`$ is symmetric with respect to $`t=\tau `$, $`\tau /2`$. These results are in agreement with our analytical results. It also implies the complementary relation between MMC and fidelity, for inside the range from $`t=\tau /2`$ to $`3\tau /2`$, the increase of $`F(t)`$ is accompanied by a decrease of $`C(t)`$ (vice versa).
It is pointed out that our results about MMC $`C(t)`$ at $`t=\tau /2`$ are based on the condition that the initial wave packet $`|\psi (0)`$ is real except for a global phase. One may be interested in the situation when $`c_1`$ and $`c_2`$ are not real for $`|\psi _{1,2}(0)`$. For this situation, e.g., $`c_1=(1+i)/2`$, $`c_2=1/5+i\sqrt{23/50}`$, the numerical calculation shows that $`C(t)`$ is not just symmetric with respect to $`t=\tau /2`$, $`C(\tau /2)1`$, $`\mathrm{max}\left(C(t)\right)`$ is very close, yet not equal to one and $`C(\tau /2)\mathrm{max}(C(t))`$ (see Figs. 2(c) and 2(d)).
## V Perfect QST of Bloch electrons in an engineered lattice
Based on the above recognitions about the relation between a perfect QST and mode entanglement, we can construct various lattice models with mirror symmetry to achieve perfect QST. Furthermore we can characterize these QSTs with the MMC. Actually, a large class of models for QST have been discovered by us most recently Shi by generalizing the spin model in Ref. Christandal .
Now we further generalize the perfect QST model to a much larger class. The Hamiltonian is given in Eq. (2) with the engineered coupling constants
$$J_j=J_0\sqrt{(j+\xi _j)\left(Nj+\xi _j\right)},$$
(29)
where
$$\xi _j=[1(1)^j]l/(2m+1)$$
(30)
for the given $`m,`$ $`l0,`$ $`1,`$ $`2,`$ $`3`$, $`\mathrm{}`$. We notice that it will return to the previous models in Refs. Christandal ; Shi when $`m=0`$.
Numerical analysis shows that the above Hamiltonian possesses a commensurate structure of energy spectrum by an experiential formula
$$\epsilon _n=N_nE_0(N+1)J_0,$$
(31)
where the energy unit is
$$E_0=\frac{2J_0}{2m+1},$$
(32)
$`N_n=n(2m+1)l`$ for $`n=1`$, $`2`$, $`\mathrm{}`$, $`N/2,`$ and $`N_n=n(2m+1)+l`$ for $`n=N/2+1`$, $`\mathrm{}`$, $`N.`$ Numerical results show that the above experiential formula (31) still holds when $`N=3000`$. It can be checked that the energy spectrum is matched with the corresponding parity (the eigen-value of $`R`$) as
$$p_n=(1)^{N_n}\mathrm{exp}\{i[(m+\frac{1}{2})N+1]\pi \}.$$
(33)
The corresponding eigen-states $`|\phi _n=_{j=1}^Nc_j(n)|j`$ can be determined by the matrix equation $`H|\phi _n=\epsilon _n|\phi _n.`$
According to Refs. Shi ; Li , the characteristic time to perform a perfect QST is $`\tau =\pi /E_0`$, provided that $`l/(2m+1)`$ is an irreducible fraction. Now we can show that, at $`t=\tau `$, the time evolution operator
$$U(t)=\underset{n}{}\mathrm{exp}(i\epsilon _nt)|\phi _n\phi _n|$$
(34)
is just the mirror reflection operator $`R`$ by neglecting a global phase, namely,
$$U(\tau )=\underset{n}{}(1)^{N_n}|\phi _n\phi _n|=(1)^lR.$$
(35)
Thus, the present model admits perfect QSTs when
$$\xi _j=[1(1)^j]l/(2m+1).$$
(36)
In order to verify the prediction about the relation between the MMC and fidelity, a numerical analysis is carried out for the present QST model. We investigate the 4-site case with $`m=1`$ and $`l=2`$. The real initial wave packet is also $`|\psi _{1,2}(0)`$. Detail behaviors of the MMC and fidelity between the instants $`t=0`$ and $`t=2\tau `$ are shown in Fig. 3. We notice, in Fig. 3, that $`C(t)=0,`$ $`1,`$ $`0,`$ for $`t=0,`$ $`\tau /2,`$ $`\tau `$ respectively and $`C(t)`$ is symmetric with respect to $`t=\tau ,`$ $`\tau /2`$. Obviously, it is in agreement with our prediction.
## VI Summary
In summary we have defined the mirror mode concurrence (MMC) to describe how a perfect quantum state transfer (QST) can be achieved for a large class of lattice model of fermion systems with mirror symmetry. By investigating the property of MMC of these perfect QST models, a novel complementary relation between the MMC and fidelity is revealed. Actually our definition of MMC is just a part of total concurrence Qian ; yang . However, when the symmetry of our systems is taken into consideration, MMC is a better measurement in characterizing the process of a perfect QST. A new class of QST models are discovered to support our observations. Therefore, a perfect QST can now be understood as a process of establishing an entanglement and then destroying it at the correlated instants. Finally we remark that our main results are valid in other perfect QST models with general symmetries such as translation, rotation and etc. It is very interesting to further investigate the QST vs entanglement relation based on solid state systems with the symmetries described by point groups or the crystallographic space groups.
## Appendix A A general proof for complementarity $`F(\tau )=1C(\tau /2)=1`$
We have proved that the MMC $`C(t)`$ will reach its maximum $`1`$ at the instant $`t=\tau /2`$ where $`\tau `$ is the instant, at which the fidelity $`F(t)`$ reaches its maximum $`F(\tau )=1`$. We now prove the inverse proposition: If $`C(t)`$ reaches its maximum $`1`$ at the instant $`t=\tau /2`$, then at time $`t=\tau ,`$ $`F(\tau )=1`$. Namely, we have a theorem in a sufficient and necessary statement
$$F(\tau )=1C(\tau /2).$$
(37)
In this appendix, we will prove the above theorem for a general model even for higher dimensional fermion systems with a Hamiltonian,
$$H=\underset{ij}{\overset{N}{}}J_{ij}a_i^{}a_j\text{,}$$
(38)
on the one particle Fock space spanned by $`N`$ basis vectors $`\{|j\}`$, $`j=1`$, $`2`$, $`3`$, $`\mathrm{}`$, $`N`$. Suppose the Hamiltonian has a symmetry $`S`$ and $`[S,H]=0`$, and the basis vectors can be decomposed into two subspaces $`\{|n_j`$ $`|j=1`$, $`2`$, $`3`$, $`\mathrm{}`$, $`N/2\}`$ and $`\{|m_j`$ $`|j=1`$, $`2`$, $`3`$, $`\mathrm{}`$, $`N/2\}`$ such that
$$S|n_j=|m_j\text{}S|m_j=|n_j,$$
(39)
then perfect QST requires that at a certain instant $`t=\tau `$, $`U(\tau )=\mathrm{exp}(iH\tau )=S`$. The case by Eq. (38) is just a generalization of the situation of mirror symmetry Hamiltonian.
Through the definition of total concurrence
$$C(t)=\underset{j}{\overset{N/2}{}}C_{n_j,m_j}=\underset{j}{\overset{N/2}{}}2\left|\psi (t)\left|a_{n_j}^{}a_{m_j}\right|\psi (t)\right|,$$
(40)
we can first prove the proposition from $`F(\tau )=1`$ to $`C(\tau /2)=1`$.
As for an initial state $`|\psi (0)`$, the fidelity of a state $`|\psi (t)=U(t)|\psi (0)`$ reads as
$$F(t)=\left|S\psi (0)\left|U(t)\right|\psi (0)\right|=\left|\psi (0)\left|S^+U(t)\right|\psi (0)\right|,$$
(41)
and a perfect QST at $`t=\tau `$ can be depicted by the maximized fidelity $`F(\tau )=1`$ when
$$U(\tau /2)U(\tau /2)=U(\tau )=S$$
satisfies Eq. (38). We calculate the total concurrence of $`|\psi (t)`$ as
$`C(t)`$ $`=`$ $`{\displaystyle \underset{j}{\overset{N/2}{}}}2\left|\psi (0)\left|U^+(t)a_{n_j}^{}a_{m_j}U(t)\right|\psi (0)\right|`$ (42)
$`=`$ $`{\displaystyle \underset{j=1}{\overset{N/2}{}}}2\left|\psi (0)\left|U^+(t)\right|n_jm_j\left|SS^+U(t)\right|\psi (0)\right|`$
$`=`$ $`{\displaystyle \underset{j=1}{\overset{N/2}{}}}2\left|\psi (0)\left|U^+(t)\right|n_j\right|\left|n_j\left|U^+(t)\right|\psi (0)\right|.`$
Then at the instant $`t=\tau /2`$
$$C(\tau /2)=\underset{j=1}{\overset{N/2}{}}2\left|\psi (0)\left|U^+(\tau /2)\right|n_j\right|\left|n_j\left|U^+(\tau /2)\right|\psi (0)\right|.$$
(43)
For real $`|\psi (0)`$ we have $`\left|\psi (0)\left|U^+(\tau /2)\right|n_j\right|=\left|n_j\left|U^+(\tau /2)\right|\psi (0)\right|`$ and then
$$C(\tau /2)=\underset{j=1}{\overset{N/2}{}}2\left|\psi (0)\left|U^+(\tau /2)\right|n_j\right|^2=1.$$
(44)
Thus we have
$$\mathrm{max}(C(t))=C(\tau /2)=1.$$
(45)
Now we prove the proposition from $`C(\tau /2)=1`$ to $`F\left(\tau \right)=1`$. According to Eq. (41), if we require $`C(\tau /2)=\mathrm{max}(C(t))=1`$ at some instant $`t=\tau /2`$, then
$$\left|\psi (0)\left|U^+(\tau /2)\right|n_j\right|=\left|m_j\left|U(\tau /2)\right|\psi (0)\right|.$$
(46)
Therefore we have
$$\left|\psi (0)\left|U^+(\tau /2)\right|n_j\right|=\left|m_j\left|U(\tau /2)\right|\psi (0)\right|,$$
or
$$\left|\psi (0)\left|U^+(\tau /2)\right|n_j\right|=\left|\psi (0)\left|U^+(\tau /2)\right|m_j\right|,$$
or
$$\left|\psi (0)\left|U^+(\tau /2)\right|n_j\right|=\left|\psi (0)\left|U^+(\tau /2)S\right|n_j\right|$$
This means $`U^+(\tau /2)S=U^+(\tau /2)`$ or $`U^+(\tau /2)S=U(\tau /2)`$. It has a trivial solution $`S=1`$ and an approved non-trivial solution $`S=U(\tau )`$. With the non-trivial solution $`S=U(\tau )`$, there will be a perfect QST, i.e.,
$`F(\tau )`$ $`=`$ $`\left|S\psi (0)\left|U(\tau )\right|\psi (0)\right|`$
$`=`$ $`\left|\psi (0)\left|S^+U(\tau )\right|\psi (0)\right|=1.`$
As is stands, we have verified the theorem $`F(\tau )=1C(\tau /2)`$ in a general situation.
We acknowledge the support of the NSFC (grant No. 90203018, 10474104, 10447133), the Knowledge Innovation Program (KIP) of Chinese Academy of Sciences, the National Fundamental Research Program of China (No. 2001CB309310).
|
warning/0506/hep-ph0506277.html
|
ar5iv
|
text
|
# Search for isotensor exotic meson and twist 4 contribution to 𝛾^∗𝛾→𝜌𝜌
## I I. Introduction
Exclusive reactions $`\gamma ^{}\gamma A+B`$ which may be accessed in $`e^+e^{}`$ collisions have been shown DGPT to have a partonic interpretation in the kinematical region of large virtuality of one photon and of small center of mass energy. The scattering amplitude factorizes in a long distance dominated object – the generalized distribution amplitude (GDA) – and a short distance perturbatively calculable scattering matrix. A phenomenological analysis of the $`\pi \pi `$ channel DGP has shown that precise experimental data could be collected at intense $`e^+e^{}`$ collider experiments such as BABAR and BELLE. Meanwhile, first data on the $`\rho ^0\rho ^0`$ channel at LEP have been published L3Coll1 and analyzed APT , showing the compatibility of the QCD leading order analysis with experiment at quite modest values of $`Q^2`$.
In this paper, we focus on the comparison of processes $`\gamma ^{}\gamma \rho ^0\rho ^0`$ and $`\gamma ^{}\gamma \rho ^+\rho ^{}`$ in the context of searching an exotic isospin $`2`$ resonance decaying in two $`\rho `$ mesons; such channels have recently been studied at LEP by the L3 collaboration L3Coll1 ; L3Coll2 . A related study for photoproduction Rosner raised the problem of $`\rho ^0\rho ^0`$ enhancement with respect to $`\rho ^+\rho ^{}`$ at low energies. One of the solutions of this problem was based on the prediction Achasov0 and further exploration Achasov of the possible existence of isotensor state, whose interference with the isoscalar state is constructive for neutral mesons and destructive for charged ones. This option was also independently considered in Liu . The crucial property of such an exotic state is the absence of $`\overline{q}q`$ wave function at any momentum resolution. In other words, quark-antiquark component is absent both in its non-relativistic description and at the level of the light-cone distribution amplitude. This is by no means common: for instance, the $`1^+`$ state which is a quark-gluon hybrid at the non-relativistic level is described by a leading twist quark-antiquark distribution amplitude AnHyb . Contrary to that, an isotensor state on the light cone corresponds to the twist $`4`$ or higher and its contribution is thus power suppressed at large $`Q^2`$. This is supported by the mentioned L3 data, where the high $`Q^2`$ ratio two of the cross sections of charged and neutral mesons production points out an isoscalar state.
We studied both perturbative and non-perturbative ingredients of QCD factorization for the description of an isotensor state. Namely, we calculated the twist $`4`$ coefficient function and extracted the non-perturbative matrix elements from L3 data. Our analysis is compatible with the existence of an isotensor exotic meson with a mass around $`1.5`$ GeV.
## II II. Amplitude of $`\gamma ^{}\gamma \rho \rho `$ process
The reaction which we study here is $`e(k)+e(l)e(k^{})+e(l^{})+\rho (p_1)+\rho (p_2)`$, where $`\rho `$ stands for the triplet $`\rho `$ mesons; the initial electron $`e(k)`$ radiates a hard virtual photon with momentum $`q=kk^{}`$, with $`q^2=Q^2`$ quite large. This means that the scattered electron $`e(k^{})`$ is tagged. To describe the given reaction, it is useful to consider the sub-process $`e(k)+\gamma (q^{})e(k^{})+\rho (p_1)+\rho (p_2)`$. Regarding the other photon momentum $`q^{}=ll^{}`$, we assume that, firstly, its momentum is almost collinear to the electron momentum $`l`$ and, secondly, that $`q^{\mathrm{\hspace{0.17em}2}}`$ is approximately equal to zero, which is a usual approximation when the second lepton is untagged.
In two $`\rho `$ meson production, we are interested in the channel where the resonance corresponds to the exotic isospin, i.e $`I=2`$, and usual $`J^{PC}`$ quantum numbers. The $`J^{PC}`$ quantum numbers are not essential for our study. Because the isospin $`2`$ has only a projection on the four quark correlators, the study of mesons with the isospin $`2`$ can help to throw light upon the four quark states. We thus, together with the mentioned reactions, study the following processes: $`e(k)+e(l)e(k^{})+e(l^{})+R(p)`$ and $`e(k)+\gamma (q^{})e(k^{})+R(p)`$, where meson $`R(p)`$ possesses isospin $`I=2`$.
Considering the amplitude of the $`\gamma ^{}\gamma `$ subprocess, we write
$`𝒜_{(i,j)}(\gamma \gamma ^{}\rho \rho )=\epsilon _\mu ^{(i)}\epsilon _\nu ^{(j)}{\displaystyle d^4z_1d^4z_2e^{iq^{}z_1iqz_2}\rho (p_1)\rho (p_2)|T\left[J_\mu (z_1)J_\nu (z_2)\right]|0},`$ (1)
where $`J_\mu `$ denotes the quark electromagnetic current $`J_\mu =\overline{\psi }𝒬\gamma _\mu \psi `$ with the charge matrix $`𝒬`$ belonging to $`SU_F(2)`$ group. The photon polarization vectors read
$`\epsilon _\mu ^{(\pm )}=(0,{\displaystyle \frac{1}{\sqrt{2}}},{\displaystyle \frac{+i}{\sqrt{2}}},0),\epsilon _\mu ^{(\pm )}=(0,{\displaystyle \frac{1}{\sqrt{2}}},{\displaystyle \frac{i}{\sqrt{2}}},0),\epsilon _\mu ^{(0)}=({\displaystyle \frac{|q|}{\sqrt{Q^2}}},0,0,{\displaystyle \frac{q_0}{\sqrt{Q^2}}}),`$ (2)
for the real and virtual photons, respectively. The coefficient functions of twist $`2`$ operators to Operator Product Expansion of currents product in (1) were discussed in detail in APT , while the contributions of new twist $`4`$ operators are described by coefficient functions calculated long ago in JS when considering the problem of twist $`4`$ corrections to Deep Inelastic Scattering.
Let us now turn on the flavour or isospin structure of the corresponding amplitudes. The $`\rho \rho `$ state with $`I=0`$ can be projected on both the two and four quark operators, while the state with $`I=2`$ on the four quark operator only. Indeed, let us start from the consideration of the vacuum–to–$`\rho \rho `$ matrix element in (1)
$`\rho ^a\rho ^b|\overline{\psi }_f(0)\mathrm{\Gamma }\psi _g(z)|0=\delta ^{ab}I_{fg}\mathrm{\Phi }^{I=0}+i\epsilon ^{abc}\tau _{fg}^c\mathrm{\Phi }^{I=1},`$ (3)
where the quark fields are shown with free flavour indices and $`\mathrm{\Gamma }`$ stands for the corresponding $`\gamma `$-matrix. The isoscalar and isovector GDA’s $`\mathrm{\Phi }^I`$ in (3) are well-known, see for instance Diehlrep . Note that, in (3), the correspondence between triplets $`\{\rho ^1,\rho ^2,\rho ^3\}`$ and $`\{\rho ^+,\rho ^{},\rho ^0\}`$ is given by the standard way.
Moreover, for the coefficient function at higher order in the strong coupling constant, the corresponding matrix element gives us
$`\rho ^a\rho ^b|[\overline{\psi }_{f_1}(0)\mathrm{\Gamma }_1\psi _{g_1}(\eta )][\overline{\psi }_{f_2}(z)\mathrm{\Gamma }_2\psi _{g_2}(\xi )]|0.`$ (4)
Using the Clebsch-Gordan decomposition, we obtain
$`\left([\overline{\psi }_{f_1}\psi _{g_1}][\overline{\psi }_{f_2}\psi _{g_2}]\right)^{I=0,I_z=0}`$ $``$ $`{\displaystyle \frac{1}{\sqrt{3}}}\left[{\displaystyle \frac{1}{2}}\tau _{f_1g_1}^0\tau _{f_2g_2}^0+\tau _{f_1g_1}^+\tau _{f_2g_2}^{}+\tau _{f_1g_1}^{}\tau _{f_2g_2}^+\right]\stackrel{~}{\mathrm{\Phi }}^{I=0,I_z=0}`$ (5)
for the isospin $`0`$ and $`I_z=0`$ projection of the four quark operator in (4), and
$`\left([\overline{\psi }_{f_1}\psi _{g_1}][\overline{\psi }_{f_2}\psi _{g_2}]\right)^{I=2,I_z=0}`$ $``$ $`{\displaystyle \frac{1}{\sqrt{6}}}\left[\tau _{f_1g_1}^0\tau _{f_2g_2}^0\tau _{f_1g_1}^+\tau _{f_2g_2}^{}\tau _{f_1g_1}^{}\tau _{f_2g_2}^+\right]\stackrel{~}{\mathrm{\Phi }}^{I=2,I_z=0}`$ (6)
for the isospin $`2`$ and $`I_z=0`$ projection of the four quark operator in (4). The four quark GDA’s $`\stackrel{~}{\mathrm{\Phi }}^{I,I_z=0}`$ can be defined in an analogous manner as the two quark GDA’s. Hence, one can see that the amplitudes (1) for $`\rho ^0\rho ^0`$ and $`\rho ^+\rho ^{}`$ productions can be written in the form of the decomposition:
$`𝒜_{(+,+)}=𝒜_{(+,+)\mathrm{\hspace{0.17em}2}}^{I=0,I_z=0}+𝒜_{(+,+)\mathrm{\hspace{0.17em}4}}^{I=0,I_z=0}+𝒜_{(+,+)\mathrm{\hspace{0.17em}4}}^{I=2,I_z=0},`$ (7)
where the subscripts $`2`$ and $`4`$ in the amplitudes imply that the given amplitudes are associated with the two and four quark correlators, respectively. The amplitudes corresponding to $`\rho ^+\rho ^{}`$ production are not independent and can be expressed through the corresponding amplitudes of $`\rho ^0\rho ^0`$ production. Indeed, one can derive the following relations:
$`𝒜_{(+,+)k}^{I=0,I_z=0}(\gamma \gamma ^{}\rho ^+\rho ^{})=𝒜_{(+,+)k}^{I=0,I_z=0}(\gamma \gamma ^{}\rho ^0\rho ^0)\mathrm{for}k=2,\mathrm{\hspace{0.17em}4}`$
$`𝒜_{(+,+)\mathrm{\hspace{0.17em}4}}^{I=2,I_z=0}(\gamma \gamma ^{}\rho ^+\rho ^{})={\displaystyle \frac{1}{2}}𝒜_{(+,+)\mathrm{\hspace{0.17em}4}}^{I=2,I_z=0}(\gamma \gamma ^{}\rho ^0\rho ^0).`$ (8)
The amplitude of two $`\rho `$ meson production in two photon collision can be also presented through a resonant intermediate state. The vacuum to $`\rho \rho `$ matrix element in the r.h.s. of (1) can be traded for
$`{\displaystyle \underset{I=0,1,2}{}}\rho (p_1)\rho (p_2)|R^I(p){\displaystyle \frac{1}{M_{R^I}^2p^2i\mathrm{\Gamma }_{R^I}M_{R^I}}}R^I(p)|T\left[J_\mu (0)J_\nu (z)\right]|0.`$ (9)
where $`R^I(p)`$ is the resonance with three possible isospin $`I=0,\mathrm{\hspace{0.17em}1},\mathrm{\hspace{0.17em}2}`$. Note that, in our case, only isospin $`0`$ and $`2`$ cases are relevant due to the positive $`C`$-parity of the initial and final states. The matrix element $`\rho \rho |R^I`$ defines the corresponding coupling constant of meson and $`R^I|T\left[J_\mu (0)J_\nu (z)\right]|0`$ is considered up to the second order of strong coupling constant $`\alpha _S`$, i.e this matrix element is written as a sum of two- and four-quark correlators.
## III III. Differential cross sections
Previously, the theoretical description of the experimental data collected for the $`\rho ^0\rho ^0`$ production has been performed in APT . Now, the subject of our study is the differential cross section corresponding to both the $`\rho ^0\rho ^0`$ and $`\rho ^+\rho ^{}`$ productions in the electron–positron collision.
Using the equivalent photon approximation Budnev we find the expression for the corresponding cross section :
$`{\displaystyle \frac{d\sigma _{eeee\rho \rho }}{dQ^2dW^2}}={\displaystyle }..{\displaystyle }d\mathrm{cos}\theta d\varphi dx_2{\displaystyle \frac{\alpha }{\pi }}F_{WW}(x_2){\displaystyle \frac{d\sigma _{e\gamma e\rho \rho }}{dQ^2dW^2d\mathrm{cos}\theta d\varphi }},`$ (10)
where the usual Weizsacker-Williams function $`F_{WW}`$ is used. In (10), the cross section for the subprocess reads
$`{\displaystyle \frac{d\sigma _{e\gamma e\rho \rho }}{dQ^2dW^2d\mathrm{cos}\theta d\varphi }}={\displaystyle \frac{\alpha ^3}{16\pi }}{\displaystyle \frac{\beta }{S_{e\gamma }^2}}{\displaystyle \frac{1}{Q^2}}\left(1{\displaystyle \frac{2S_{e\gamma }(Q^2+W^2S_{e\gamma })}{(Q^2+W^2)^2}}\right)\left|A_{(+,+)}\right|^2`$ (11)
where the amplitude $`A_{(+,+)}`$ is defined by (7). For the case of $`\rho ^0\rho ^0`$ production, the cross section (10) takes the form:
$`{\displaystyle \frac{d\sigma _{eeee\rho ^0\rho ^0}}{dQ^2dW^2}}={\displaystyle \frac{100\alpha ^4}{9}}G(S_{ee},Q^2,W^2)\beta `$
$`({\displaystyle \frac{\mathrm{\Gamma }_{R^0}M_{R^0}}{\beta _0((M_{R^0}^2W^2)^2+\mathrm{\Gamma }_{R^0}^2M_{R^0}^2)}}[𝐒_2^{I=0,I_3=0}+{\displaystyle \frac{\alpha _S(Q^2)M_{R^0}^2}{Q^2}}𝐒_4^{I=0,I_3=0}]^2+.`$
. $`{\displaystyle \frac{\mathrm{\Gamma }_{R^2}M_{R^2}}{\beta _2((M_{R^2}^2W^2)^2+\mathrm{\Gamma }_{R^2}^2M_{R^2}^2)}}\left[{\displaystyle \frac{\alpha _S(Q^2)M_{R^2}^2}{Q^2}}𝐒_4^{I=2,I_3=0}\right]^2+.`$
. $`2\sqrt{{\displaystyle \frac{\mathrm{\Gamma }_{R^0}\mathrm{\Gamma }_{R^2}M_{R^0}M_{R^2}}{\beta _0\beta _2}}}{\displaystyle \frac{(M_{R^0}^2W^2)(M_{R^2}^2W^2)+(\mathrm{\Gamma }_{R^0}M_{R^0})(\mathrm{\Gamma }_{R^2}M_{R^2})}{((M_{R^0}^2W^2)^2+\mathrm{\Gamma }_{R^0}^2M_{R^0}^2)((M_{R^2}^2W^2)^2+\mathrm{\Gamma }_{R^2}^2M_{R^2}^2)}}\times .`$
. $`[𝐒_2^{I=0,I_3=0}+{\displaystyle \frac{\alpha _S(Q^2)M_{R^0}^2}{Q^2}}𝐒_4^{I=0,I_3=0}]{\displaystyle \frac{\alpha _S(Q^2)M_{R^2}^2}{Q^2}}𝐒_4^{I=2,I_3=0}),`$
where $`\mathrm{\Gamma }_{R^I}`$ stand for the total widths. The dimensionful structure constants $`𝐒_4^{I,I_3=0}`$ and $`𝐒_2^{I=0,I_3=0}`$ are related to the nonperturbative vacuum–to–meson matrix elements. The $`\beta `$–functions are also defines in the standard ways: $`\beta =\sqrt{14m_\rho ^2/W^2}`$ and $`\beta _I=\sqrt{14m_\rho ^2/M_{R^I}^2}`$. The function $`G`$ in (III) is equal to
$`G(S_{ee},Q^2,W^2)={\displaystyle \underset{0}{\overset{1}{}}}𝑑x_2F_{WW}(x_2)\left[{\displaystyle \frac{1}{x_2^2S_{ee}^2Q^2}}{\displaystyle \frac{2}{x_2S_{ee}Q^2(Q^2+W^2)}}+{\displaystyle \frac{2}{Q^2(Q^2+W^2)^2}}\right].`$ (13)
The differential cross section corresponding to $`\rho ^+\rho ^{}`$ production can be obtained using (II), we have
$`{\displaystyle \frac{d\sigma _{eeee\rho ^+\rho ^{}}}{dQ^2dW^2}}={\displaystyle \frac{200\alpha ^4}{9}}G(S_{ee},Q^2,W^2)\beta `$
$`({\displaystyle \frac{\mathrm{\Gamma }_{R^0}M_{R^0}}{\beta _0((M_{R^0}^2W^2)^2+\mathrm{\Gamma }_{R^0}^2M_{R^0}^2)}}[𝐒_2^{I=0,I_3=0}+{\displaystyle \frac{\alpha _S(Q^2)M_{R^0}^2}{Q^2}}𝐒_4^{I=0,I_3=0}]^2+.`$
. $`{\displaystyle \frac{1}{4}}{\displaystyle \frac{\mathrm{\Gamma }_{R^2}M_{R^2}}{\beta _2((M_{R^2}^2W^2)^2+\mathrm{\Gamma }_{R^2}^2M_{R^2}^2)}}\left[{\displaystyle \frac{\alpha _S(Q^2)M_{R^2}^2}{Q^2}}𝐒_4^{I=2,I_3=0}\right]^2.`$
. $`\sqrt{{\displaystyle \frac{\mathrm{\Gamma }_{R^0}\mathrm{\Gamma }_{R^2}M_{R^0}M_{R^2}}{\beta _0\beta _2}}}{\displaystyle \frac{(M_{R^0}^2W^2)(M_{R^2}^2W^2)+(\mathrm{\Gamma }_{R^0}M_{R^0})(\mathrm{\Gamma }_{R^2}M_{R^2})}{((M_{R^0}^2W^2)^2+\mathrm{\Gamma }_{R^0}^2M_{R^0}^2)((M_{R^2}^2W^2)^2+\mathrm{\Gamma }_{R^2}^2M_{R^2}^2)}}\times .`$
. $`[𝐒_2^{I=0,I_3=0}+{\displaystyle \frac{\alpha _S(Q^2)M_{R^0}^2}{Q^2}}𝐒_4^{I=0,I_3=0}]{\displaystyle \frac{\alpha _S(Q^2)M_{R^2}^2}{Q^2}}𝐒_4^{I=2,I_3=0}),`$
Note that we have explicitly separated out, in (III) and (III), the running coupling constant $`\alpha _S(Q^2)`$ which appears in the twist $`4`$ terms. Because of we will study the $`Q^2`$ dependence of the corresponding cross sections at rather small values of $`Q^2`$, we use the Shirkov and Solovtsov’s analytical approach Shirkov to determine the running coupling constant in the region of small $`Q^2`$. Detailed discussion on different aspects of using the analytical running coupling constant may be found in Bakulev ; AnHyb and references therein.
## IV IV. LEP data fitting
In the previous section we derived the differential cross sections $`d\sigma _{eeee\rho \rho }/dQ^2dW^2`$ for both the $`\rho ^0\rho ^0`$ and $`\rho ^+\rho ^{}`$ channels, based on the QCD analysis. These expressions contain a number of unknown phenomenological parameters, which are intrinsically related to non perturbative quantities encoded in the generalized distribution amplitudes. One should now make a fit of these phenomenological parameters in order to get a good description of experimental data. The best values of the parameters can be found by the method of least squares, $`\chi ^2`$-method, which flows from the maximum likelihood theorem, but we postpone a comprehensive $`\chi ^2`$-analysis to a forthcoming more detailed paper. Here, we implement a naive fitting analysis to get an acceptable agreement with the experimental data. Thus, we have the following set of parameters for fitting:
$`𝐏=\{M_{R^0},\mathrm{\Gamma }_{R^0},M_{R^2},\mathrm{\Gamma }_{R^2},𝐒_2^{I=0,I_3=0},𝐒_4^{I=0,I_3=0},𝐒_4^{I=2,I_3=0}\}.`$ (15)
We start with the study of the $`W`$ dependence of the cross sections. For this goal, following the papers L3Coll1 ; L3Coll2 , we determine the cross section of process $`eeee\rho \rho `$ normalized by the integrated luminosity function:
$`\sigma _{\gamma \gamma ^{}}(W)={\displaystyle \frac{𝑑Q^2(Q^2,W)\sigma _{\gamma \gamma ^{}}(Q^2,W)}{𝑑Q^2(Q^2,W)}},`$ (16)
where the definition of the luminosity function $``$ is taken from JF . The value $`W`$ corresponds to the center of each bin, see L3Coll1 ; L3Coll2 . Focussing first on the region of larger $`Q^2`$ we fit the parameters associated with the dominant contribution which comes from the twist $`2`$ term amplitude, which is associated with the non-exotic resonance (or background) with isospin $`I=0`$. Generally speaking, there are many isoscalar resonances with masses in the region of $`13`$ GeV. To include their total effect we introduce a mass and width for an ”effective” isoscalar resonance. We then determine the values of the mass and width by fitting the data for the region of larger $`Q^2`$ (i.e., when $`Q^2`$ is in the interval $`1.2<Q^2<8.5`$ GeV<sup>2</sup>). We thus can fix the parameters $`𝐒_2^{I=0,I_3=0}`$, $`M_{R^0}`$ and $`\mathrm{\Gamma }_{R^0}`$. Good agreement can be achieved with $`M_{R^0}=1.8\mathrm{GeV}`$, $`\mathrm{\Gamma }_{R^0}=1.00\mathrm{GeV}`$ and $`𝐒_2^{I=0,I_3=0}`$ within the interval $`(0.12,\mathrm{\hspace{0.17em}0.16})`$. As can be expected the width of the effective isoscalar ”resonance” is fairly large. It means that we actually deal with a non-resonant background.
Next, we fit the $`W`$dependence of the cross section for small values of $`Q^2`$, i.e. $`0.2<Q^2<0.85`$ GeV<sup>2</sup>. In this region all twist contributions may be important. We find that the experimental data can be described by the following choice of the parameters: $`M_{R^2}=1.5\mathrm{GeV}`$, $`\mathrm{\Gamma }_{R^2}=0.4\mathrm{GeV}`$ while the parameters $`𝐒_4^{I=0,I_3=0}`$ and $`𝐒_4^{I=2,I_3=0}`$ are in the intervals $`(0.002,\mathrm{\hspace{0.17em}0.006})`$ and $`(0.012,\mathrm{\hspace{0.17em}0.018})`$, respectively.
Further, we include in our analysis the $`Q^2`$ dependence of $`\rho ^0\rho ^0`$ and $`\rho ^+\rho ^{}`$ production cross sections, i.e. $`d\sigma _{eeee\rho \rho }/dQ^2`$, which should fix the remaining arbitrariness of the parameters. We finally find that the best description of both $`W`$ and $`Q^2`$ dependence is reached at
$`M_{R^2}=1.5\mathrm{GeV},\mathrm{\Gamma }_{R^2}=0.4\mathrm{GeV},`$
$`𝐒_2^{I=0,I_3=0}=0.12\mathrm{GeV},𝐒_4^{I=0,I_3=0}=0.006\mathrm{GeV},𝐒_4^{I=2,I_3=0}=0.018\mathrm{GeV}.`$ (17)
Note that these rather small values of twist $`4`$ structure constants $`𝐒_4`$ compared to the twist $`2`$ structure constant $`𝐒_2`$ indicate that leading twist contribution dominate for the values $`Q^2`$ around or greater than $`1\mathrm{GeV}^2`$. This should be compared with what was obtained in a particular renormalon model in AGP .
Our theoretical description of the LEP experimental data are presented on Figs. 15. The plots depicted on Figs. 14 have the following notations: the short-dashed line corresponds to the contribution coming from the leading twist term of (III); the dash-dotted line – to the contribution from the twist $`4`$ term of (III); the middle-dashed line – to the contributions from the interference between twist $`2`$ and twist $`4`$ terms of (III) and (III); the long-dashed line – to the contribution from the interference between isoscalar and isotensor terms. Finally the solid line corresponds to the sum of all contributions. On Fig. 5, we present the LEP data and our theoretical curves for both the $`\rho ^0\rho ^0`$ and $`\rho ^+\rho ^{}`$ production differential cross sections as functions of $`Q^2`$. The solid line on Fig. 5 corresponds to the $`\rho ^0\rho ^0`$ differential cross section while the dashed one – to the $`\rho ^+\rho ^{}`$ differential cross section.
## V V. Discussions and Conclusions
The fitting of LEP data based on the QCD factorization of the amplitude into a hard subprocess and a generalized distribution amplitude thus allows us to claim evidence of the existence of an isospin $`I=2`$ exotic meson Achasov0 ; Achasov ; Maiani with a mass in the vicinity of $`1.5\mathrm{GeV}`$ and a width around $`0.4\mathrm{GeV}`$. The contributions of such an exotic meson in the two $`\rho `$ meson production cross sections (see, (III) and (III)) are directly associated with some twist $`4`$ terms that we have identified. At large $`Q^2`$, these twist $`4`$ contributions become negligible and the behaviours of the $`\rho ^0\rho ^0`$ and $`\rho ^+\rho ^{}`$ cross sections are controlled by the leading twist $`2`$ contributions, see Fig. 1 and 2. Figs. 3 and 4 show the increasing role of higher twist contributions when decreasing $`Q^2`$. Namely, the interference between twist $`2`$ and $`4`$ amplitudes gives the dominant contributions to $`\rho ^0\rho ^0`$ production in the lower $`Q^2`$ interval, and is thus responsible of the $`W`$ dependence of the cross section in these kinematics. In particular, in this interference term the main contribution arises from the interference between isoscalar and isotensor structures, see the long-dashed lines on Fig. 3 and 4.
Analysing the $`Q^2`$ dependence, we can see that due to the presence of a twist $`4`$ amplitude and its interference with the leading twist $`2`$ component, the $`\rho ^0\rho ^0`$ cross section at small $`Q^2`$ is a few times higher than the $`\rho ^+\rho ^{}`$ cross section, see Fig. 5. While for the region of large $`Q^2`$ where any higher twist effects are negligible the $`\rho ^0\rho ^0`$ cross section is less than the $`\rho ^+\rho ^{}`$ cross section by the factor $`2`$, which is typical from an isosinglet channel (see also (III) and (III)).
The reaction $`\gamma ^{}\gamma \rho \rho `$ and its QCD analysis in the framework of Ref. DGPT thus proves its efficiency to reveal facts on hadronic physics which would remain quite difficult to explain in a quantitative way otherwise. The leading twist dominance is seen to persist down to values of $`Q^2`$ around $`1\mathrm{GeV}^2`$. Other aspects of QCD may be revealed in different kinematical regimes through the same reaction other . Its detailed experimental analysis at intense electron colliders within the BABAR and BELLE experiments is thus extremely promising. Data at higher energies in a future linear collider should also be foreseen.
Note that the non-perturbative calculations of the relevant $`I=2`$ twist $`4`$ matrix elements also deserve special interest. In particular, one may follow the ideas developed for pion distribution max which allowed to relate higher and lower twists in multicolour QCD. The generalization for the case of $`\rho `$ mesons, anticipated by the authors of max , and use of crossing relations between various kinematical domains provided by Radon transform technique radon may allow to apply these result in the case under consideration.
In conclusion, let us stress that the L3 data allows to estimate the contribution of higher twist four quark light cone distribution to the production amplitude of vector meson pairs. Our numerical analysis leads to a rather small width for the corresponding resonant state, which is nothing else as an exotic four-quark isotensor meson. At the same time, a more elaborate experimental, theoretical and numerical analysis is required to confirm, with better accuracy, the smallness of the width and the existence of an exotic meson.
## VI Acknowledgements
We are grateful to N.N. Achasov, A. Donnachie, J. Field, K. Freudenreich, M. Kienzle, N. Kivel, K.F. Liu, M.V. Polyakov and I. Vorobiev for useful discussions and correspondence. O.V.T. is indebted to Theory Division of CERN and CPHT, École Polytechnique, for warm hospitality. I.V.A. expresses gratitude to Theory Division of CERN and University of Geneva for financial support of his visit. This work has been supported in part by RFFI Grant 03-02-16816. I.V.A. thanks NATO for a grant.
|
warning/0506/math0506446.html
|
ar5iv
|
text
|
# Structure Relations in Special 𝐴_∞-bialgebras
## 1. Introduction
A general $`A_{\mathrm{}}`$-infinity bialgebra is a DG module $`(H,d)`$ equipped with a family of structurally compatible operations $`\omega _{j,i}:H^iH^j,`$where $`i,j1`$ and $`i+j3`$ (see ). In *special* $`A_{\mathrm{}}`$-bialgebras, $`\omega _{j,i}=0`$ whenever $`i,j2,`$ and the remaining operations $`m_i=\omega _{1,i}`$ and $`\mathrm{\Delta }_j=\omega _{j,1}`$ define the underlying $`A_{\mathrm{}}`$-(co)algebra substructure. Thus special $`A_{\mathrm{}}`$-bialgebras have the form $`(H,d,m_i,\mathrm{\Delta }_j)_{i,j2}`$ subject to the appropriate structure relations involving $`d,`$ the $`m_i`$’s and $`\mathrm{\Delta }_j`$’s. These relations are much easier to describe than those in the general case, which require the S-U diagonal $`\mathrm{\Delta }_P`$ on permutahedra. Instead, the S-U diagonal $`\mathrm{\Delta }_K`$ on Stasheff’s associahedra $`K=K_n`$ is required here (see ).
$`A_{\mathrm{}}`$-bialgebras are fundamentally important structures in algebra and topology. In general, the homology of every $`A_{\mathrm{}}`$-bialgebra inherits an $`A_{\mathrm{}}`$-bialgebra structure ; in particular, this holds for the integral homology of a loop space. In fact, over a field, the $`A_{\mathrm{}}`$-bialgebra structure on the homology of a loop space specializes to the $`A_{\mathrm{}}`$-(co)algebra structures observed by Gugenheim and Kadeishvili .
The main result of this paper is the following simple formulation of the structure relations in special $`A_{\mathrm{}}`$-bialgebras that do not involve $`d`$: Let $`TH`$ denote the tensor module of $`H`$ and let $`e^{n2}`$ denote the top dimensional face of $`K_n.`$ There is a “fraction product” on $`M=End\left(TH\right)`$ (denoted here by “$``$”) and certain cellular cochains $`\xi ,\zeta C^{}(K;M)`$ such that for each $`i,j2,`$
$$\mathrm{\Delta }_jm_i=\xi ^j\left(e^{i2}\right)\zeta ^i\left(e^{j2}\right),$$
where the exponents indicate certain $`\mathrm{\Delta }_K`$-cup powers.
I must acknowledge the fact that many of the ideas in this paper germinated during conservations with Samson Saneblidze, whose openness and encouragement led to this paper. For this I express sincere thanks.
## 2. Matrix Considerations
We begin with a brief review of the algebraic machinery we need; for a detailed exposition see . Let $`M=_{m,n1}M_{n,m}`$ be a bigraded module over a commutative ring $`R`$ with identity $`1_R`$ and consider the module $`TTM`$ of tensors on $`TM.`$ Given matrices $`X=\left[x_{ij}\right]`$ and $`Y=\left[y_{ij}\right]^{q\times p},`$ $`p,q1,`$ consider the submodule
$`M_{Y,X}`$ $`=\left(M_{y_{11},x_{11}}\mathrm{}M_{y_{1p},x_{1p}}\right)\mathrm{}\left(M_{y_{q1},x_{q1}}\mathrm{}M_{y_{qp},x_{qp}}\right)`$
$`\left(M^p\right)^qTTM.`$
Represent a monomial $`A=\left(\theta _{y_{11},x_{11}}\mathrm{}\theta _{y_{1p},x_{1p}}\right)\mathrm{}\left(\theta _{y_{q1},x_{q1}}\mathrm{}\theta _{y_{qp},x_{qp}}\right)`$$`M_{Y,X}`$ as the $`q\times p`$ matrix $`\left[A\right]=\left[a_{ij}\right]`$ with $`a_{ij}=\theta _{y_{ij},x_{ij}}.`$ Then $`A`$ is the $`q`$-fold tensor product of the rows of $`[A]`$ thought of as elements of $`M^p;`$ we refer to $`A`$ as a $`q\times p`$ monomial and often write $`A`$ when we mean $`\left[A\right].`$ The *matrix submodule of* $`TTM`$ is the sum
$$\overline{𝐌}=\underset{\begin{array}{c}X,Y^{q\times p}\\ p,q1\end{array}}{}M_{Y,X}=\underset{p,q1}{}(M^p)^q.$$
Given $`𝐱\times 𝐲=(x_1,\mathrm{},x_p)\times (y_1,\mathrm{},y_q)^p\times ^q,`$ set $`X=\left[x_{ij}=x_j\right]_{1iq},`$ $`Y=\left[y_{ij}=y_i\right]_{1jp}`$ and denote $`𝐌_𝐱^𝐲=M_{Y,X}.`$ The *essential submodule of* $`TTM`$ is
$$𝐌=\underset{\begin{array}{c}𝐱\times 𝐲^p\times ^q\\ p,q1\end{array}}{}𝐌_𝐱^𝐲$$
and a $`q\times p`$ monomial $`A𝐌`$ has the form
$$A=\left[\begin{array}{ccc}\theta _{y_1,x_1}\hfill & \mathrm{}\hfill & \theta _{y_1,x_p}\hfill \\ \mathrm{}& & \mathrm{}\\ \theta _{y_q,x_1}\hfill & \mathrm{}\hfill & \theta _{y_q,x_p}\hfill \end{array}\right].$$
Graphically represent $`A=\left[\theta _{y_j,x_i}\right]𝐌_𝐱^𝐲`$ two ways: (1) as a matrix of “double corollas” in which $`\theta _{y_j,x_i}`$ is pictured as two corollas joined at the root–one opening downward with $`x_i`$ leaves and the other opening upward with $`y_j`$ leaves–and (2) as an arrow in the positive integer lattice $`^2`$ from $`(\left|𝐱\right|,q)`$ to $`(p,\left|𝐲\right|),`$ where $`\left|𝐮\right|=u_1+\mathrm{}+u_k`$ (see Figure 1).
Figure 1. Graphical representations of a typical monomial.
Each pairing $`\gamma :_{r,s1}M^rM^sM`$ induces an *upsilon product* $`\mathrm{{\rm Y}}:\overline{𝐌}\overline{𝐌}\overline{𝐌}`$ supported on “block transverse pairs,” which we now describe.
###### Definition 1.
A monomial pair $`A^{q\times s}B^{t\times p}=\left[\theta _{y_k\mathrm{},v_k\mathrm{}}\right]\left[\eta _{u_{ij},x_{ij}}\right]\overline{𝐌}\overline{𝐌}`$ is a
1. Transverse Pair (TP) if $`s=t=1,`$ $`u_{1,j}=q`$ and $`v_{k,1}=p`$ for all $`j,k,`$ i.e., setting $`x_j=x_{1,j}`$ and $`y_k=y_{k,1}`$ gives
$$AB=\left[\begin{array}{c}\theta _{y_1,p}\\ \mathrm{}\\ \theta _{y_q,p}\end{array}\right]\left[\begin{array}{ccc}\eta _{q,x_1}\hfill & \mathrm{}\hfill & \eta _{q,x_p}\hfill \end{array}\right]𝐌_p^𝐲𝐌_𝐱^q.$$
2. Block Transverse Pair (BTP) if there exist $`t\times s`$ block decompositions $`A=\left[A_k^{}\mathrm{}^{}\right]`$ and $`B=\left[B_{ij^{}}^{}\right]`$ such that $`A_i\mathrm{}^{}B_i\mathrm{}^{}`$ is a TP for all $`i,\mathrm{}`$.
Unlike the blocks in a standard block matrix, the blocks $`A_i\mathrm{}^{}`$ (or $`B_i\mathrm{}^{}`$) in a general BTP may vary in length within a given row (or column). However, when $`AB𝐌_𝐯^𝐲𝐌_𝐱^𝐮`$ is a BTP with $`𝐮=(q_1,\mathrm{}q_t),`$ $`𝐯=(p_1,\mathrm{},p_s),`$ $`𝐱=(𝐱_1,\mathrm{},𝐱_s)`$ and $`𝐲=(𝐲_1,\mathrm{},𝐲_t)`$, the TP $`A_i\mathrm{}^{}B_i\mathrm{}^{}𝐌_p_{\mathrm{}}^{𝐲_i}𝐌_𝐱_{\mathrm{}}^{q_i}`$ so that for a fixed $`i`$(or $`\mathrm{}`$) the blocks $`A_i\mathrm{}^{}`$ (or $`B_i\mathrm{}^{}`$) have constant length $`q_i`$ (or $`p_{\mathrm{}}`$); furthermore, $`AB`$ is a BTP if and only if $`𝐲^{|𝐮|}`$ and $`𝐱^{|𝐯|}`$ if and only if the initial point of arrow $`A`$ coincides with the terminal point of arrow $`B`$. Note that BTP block decomposition is unique.
###### Example 1.
A pairing of monomials $`A^{4\times 2}B^{2\times 3}𝐌_{2,1}^{1,5,4,3}𝐌_{1,2,3}^{3,1}`$ is a $`2\times 2`$ BTP per the block decompositions
Given a pairing $`\gamma =_{𝐱\times 𝐲}\gamma _𝐱^𝐲:𝐌_p^𝐲𝐌_𝐱^q𝐌_{\left|𝐱\right|}^{\left|𝐲\right|},`$ extend $`\gamma `$ to an *upsilon product* $`\mathrm{{\rm Y}}:\overline{𝐌}\overline{𝐌}\overline{𝐌}`$ via
(2.1)
$$\mathrm{{\rm Y}}\left(AB\right)_i\mathrm{}=\{\begin{array}{cc}\gamma \left(A_i\mathrm{}^{}B_i\mathrm{}^{}\right),\hfill & \text{if}AB\text{is a}\text{BTP}\hfill \\ & \\ 0,\hfill & \text{otherwise.}\hfill \end{array}$$
Then $`\mathrm{{\rm Y}}`$ sends a BTP $`A^{q\times s}B^{t\times p}𝐌_𝐯^𝐲𝐌_𝐱^𝐮`$ with $`A_i\mathrm{}^{}B_i\mathrm{}^{}𝐌_p_{\mathrm{}}^{𝐲_i}𝐌_𝐱_{\mathrm{}}^{q_i}`$ to a $`t\times s`$ monomial in $`𝐌_{\left|𝐱_1\right|,\mathrm{},\left|𝐱_s\right|}^{\left|𝐲_1\right|,\mathrm{},\left|𝐲_t\right|}.`$ We denote $`AB=\mathrm{{\rm Y}}(AB);`$ when $`\left[\theta _j\right]\left[\eta _i\right]`$ is a TP we denote $`\gamma (\theta _1,\mathrm{},\theta _q;\eta _1,\mathrm{},\eta _p)=\left(\theta _1\mathrm{}\theta _q\right)\left(\eta _1\mathrm{}\eta _p\right)`$. As an arrow, $`AB`$ runs from the initial point of $`B`$ to the terminal point of $`A.`$ Note that $`𝐌𝐌𝐌`$ so that $`\mathrm{{\rm Y}}`$ restricts to an upsilon product on $`𝐌.`$
###### Example 2.
Continuing Example 1, the action of $`\mathrm{{\rm Y}}`$ on $`A^{4\times 2}B^{2\times 3}𝐌_{2,1}^{1,5,4,3}𝐌_{1,2,3}^{3,1}`$ produces a $`2\times 2`$ monomial in $`𝐌_{3,3}^{10,3}:`$
In the target, $`(\left|𝐱_1\right|,\left|𝐱_2\right|)=(1+2,3)`$ since $`(p_1,p_2)=(2,1);`$ and $`(\left|𝐲_1\right|,\left|𝐲_2\right|)=(1+5+4,3)`$ since $`(q_1,q_2)=(3,1).`$ As an arrow, $`AB`$ initializes at $`(6,2)`$ and terminates at $`(2,13).`$
The applications below relate to the following special case: Let $`H`$ be a graded module over a commutative ring with unity and view $`M=End(TH)`$ as a bigraded module via $`M_{n,m}=Hom(H^m,H^n).`$ Then a $`q\times p`$ monomial $`A𝐌_𝐱^𝐲`$ admits a representation as an operator on $`^2`$ via
$$\left(H^{\left|𝐱\right|}\right)^q\left(H^{x_1}\mathrm{}H^{x_p}\right)^q\stackrel{𝐴}{}\left(H^{y_1}\right)^p\mathrm{}\left(H^{y_q}\right)^p$$
$$\stackrel{\sigma _{y_1,p}\mathrm{}\sigma _{y_q,p}}{}\left(H^p\right)^{y_1}\mathrm{}\left(H^p\right)^{y_q}\left(H^p\right)^{\left|𝐲\right|},$$
where $`(s,t)^2`$ is identified with $`\left(H^s\right)^t`$ and $`\sigma _{s,t}:\left(H^s\right)^t\stackrel{}{}\left(H^t\right)^s`$ is the canonical permutation of tensor factors $`\sigma _{q,p}:\left(\left(a_{11}\mathrm{}a_{q1}\right)\mathrm{}\left(a_{1p}\mathrm{}a_{qp}\right)\right)`$ $`\left(\left(a_{11}\mathrm{}a_{1p}\right)\mathrm{}\left(a_{q1}\mathrm{}a_{qp}\right)\right)`$. The canonical structure map is
(2.2)
$$\gamma =\gamma _𝐱^𝐲:𝐌_p^𝐲𝐌_𝐱^q\stackrel{\iota _p\iota _q}{}𝐌_{pq}^{|𝐲|}𝐌_{|𝐱|}^{qp}\stackrel{\text{id}\sigma _{q,p}^{}}{}𝐌_{pq}^{|𝐲|}𝐌_{|𝐱|}^{pq}\stackrel{}{}𝐌_{|𝐱|}^{|𝐲|},$$
where $`\iota _p`$ and $`\iota _q`$ are the canonical isomorphisms and $`\sigma _{q,p}^{}`$ is induced by $`\sigma _{q,p}`$ (c.f. , ), induces a canonical *associative* $`\mathrm{{\rm Y}}`$ product on $`𝐌`$ whose action on matrices of double corollas typically produces a matrix of non-planar graphs (see Figure 2).
Figure 2. The $`\gamma `$-product as a non-planar graph.
In this setting, $`\gamma `$ agrees with the composition product on the universal preCROC .
## 3. Cup products
The two pairs of dual cup products defined in this section play an essential role in the theory of structure relations. Let $`(H,d)`$ be a DG module over a commutative ring with unity. For each $`i,j2,`$ choose operations $`m_i:H^iH`$ and $`\mathrm{\Delta }_j:HH^j`$ thought of as elements of $`M=End\left(TH\right).`$ Recall that planar rooted trees (PRT’s) parametrize the faces of Stasheff’s associahedra $`K=\underset{n2}{}K_n`$and provide module generators for cellular chains $`C_{}\left(K\right)`$ . Whereas top dimensional faces correspond with corollas, lower dimensional faces correspond with more general PRT’s. Now given a face $`aK,`$ consider the class of all planar rooted trees with levels (PLT’s) representing $`a`$ and choose a representative with exactly one node in each level. In this way, we obtain a particularly nice set of module generators for $`C_{}\left(K\right),`$ denoted by $`𝒦`$. Note that the elements of a class of PLT’s represent the same function obtained by composing in various ways. The results obtained here are independent of choice since they depend only on the function.
Let $`G`$ be a DGA concentrated in degree zero and consider the cellular cochains on $`K`$ with coefficients in $`G`$:
$$C^p(K;G)=Hom^p(C_p\left(K\right);G).$$
A diagonal $`\mathrm{\Delta }`$ on $`C_{}\left(K\right)`$ induces a cup product $``$ on $`C^{}(K;G)`$ via
$$fg=(fg)\mathrm{\Delta },$$
where “$``$” denotes multiplication in $`G.`$
The essential submodule $`𝐌,`$ which serves as our coefficient module, is canonically endowed with dual associative *wedge* and *Čech cross products* defined on a monomial pair $`AB𝐌_𝐯^𝐲𝐌_𝐱^𝐮`$ by
$$A\stackrel{}{\times }B=\{\begin{array}{cc}AB,\hfill & \text{if }𝐯=𝐱,\hfill \\ 0,\hfill & \text{otherwise,}\hfill \end{array}\text{ and }A\stackrel{}{\times }B=\{\begin{array}{cc}AB,\hfill & \text{if }𝐮=𝐲,\hfill \\ 0,\hfill & \text{otherwise.}\hfill \end{array}$$
Denote $`\stackrel{}{𝐌}=(𝐌,\stackrel{}{\times })`$ and $`\stackrel{}{𝐌}=(𝐌,\stackrel{}{\times })`$ and note that $`𝐌_𝐱^𝐲\stackrel{}{\times }𝐌_𝐱^𝐮𝐌_𝐱^{𝐲,𝐮}`$ and $`𝐌_𝐯^𝐲\stackrel{}{\times }𝐌_𝐱^𝐲𝐌_{𝐯,𝐱}^𝐲.`$ Thus non-zero cross products concatenate matrices:
$$A\stackrel{}{\times }B=\left[\genfrac{}{}{0.0pt}{}{A}{B}\right]\text{ and }A\stackrel{}{\times }B=\left[A\text{ }B\right].$$
As arrows, $`A\stackrel{}{\times }B`$ runs from vertical $`x=\left|𝐱\right|`$ to vertical $`x=p,`$ whereas $`A\stackrel{}{\times }B`$ runs from horizontal $`y=q`$ to horizontal $`y=\left|𝐲\right|.`$ In particular, if $`A𝐌_a^b`$, then $`A^{\stackrel{}{\times }n}𝐌_a^{b\mathrm{}b}`$ is an arrow from $`(a,n)`$ to $`(1,nb)`$ and $`A^{\stackrel{}{\times }n}𝐌_{a\mathrm{}a}^b`$ is an arrow from $`(na,1)`$ to $`(n,b).`$ These cross products together with the S-U diagonal $`\mathrm{\Delta }_K`$ induce wedge and Čech cup products $``$ and $``$ in $`C^{}(K;\stackrel{}{𝐌})`$ and $`C^{}(K;\stackrel{}{𝐌}),`$ respectively.
The modules $`C^{}(K;\stackrel{}{𝐌})`$ and $`C^{}(K;\stackrel{}{𝐌})`$ are equipped with second cup products $`_{\mathrm{}}`$ and $`_{\mathrm{}}`$ arising from the $`\mathrm{{\rm Y}}`$-product on $`𝐌`$ together with the “leaf coproduct” $`\mathrm{\Delta }_{\mathrm{}}:C_{}\left(K\right)C_{}\left(K\right)C_{}\left(K\right),`$ which we now define. Let $`T=T^1𝒦`$ be a $`k`$-level PLT. Prune $`T`$ immediately below the first (top) level, trimming off a single corolla with $`n_1`$ leaves and $`r_11`$ stalks. Numbering from left-to-right, let $`i_1`$ be the position of the corolla. The (first) *leaf sequence* of $`T`$ is the $`r_1`$-tuple $`𝐱_{i_1}\left(n_1\right)=\left(1\mathrm{}n_1\mathrm{}1\right)`$ with $`n_1`$ in position $`i_1`$ and $`1`$’s elsewhere. Label the pruned tree $`T^2;`$ inductively, the $`\text{j}^{th}`$ *leaf sequence* of $`T`$ is the leaf sequence of $`T^j.`$ The induction terminates when $`j=k,`$ in which case $`i_k=r_k=1`$ and $`𝐱_{i_k}\left(n_k\right)=n_k.`$ The *descent sequence of* $`T`$ is the $`k`$-tuple $`(𝐱_{i_1}\left(n_1\right),\mathrm{},𝐱_{i_k}\left(n_k\right)).`$
###### Definition 2.
Let $`T𝒦`$ and identify $`T`$ with its descent sequence $`𝐧=(𝐧_1,\mathrm{},𝐧_k).`$ The leaf coproduct of $`T`$ is given by
$$\mathrm{\Delta }_{\mathrm{}}\left(T\right)=\{\begin{array}{cc}\underset{2ik}{}(𝐧_1,\mathrm{},\left|𝐧_i\right|)(𝐧_i,𝐧_{i+1},\mathrm{},𝐧_k),& k>1\\ 0,& k=1.\end{array}$$
Define the *leaf cup products* $`_{\mathrm{}}`$ and $`_{\mathrm{}}`$ on $`C^{}(K;\stackrel{}{𝐌)}`$ and $`C^{}(K;\stackrel{}{𝐌)}`$ by
$$f_{\mathrm{}}g=(fg)\tau \mathrm{\Delta }_{\mathrm{}}\text{ and }f_{\mathrm{}}g=(fg)\mathrm{\Delta }_{\mathrm{}},$$
where $`\tau `$ interchanges tensor factors and $``$ denotes the $`\mathrm{{\rm Y}}`$-product.
Note that all cup products defined in this section are non-associative and non-commutative. Unless explicitly indicated otherwise, iterated cup products are parenthesized on the extreme left, e.g., $`fgh=\left(fg\right)h.`$
## 4. Special $`A_{\mathrm{}}`$-bialgebras
Structural compatibility of $`d,`$ the $`m_i`$’s and $`\mathrm{\Delta }_j`$’s is expressed in terms of the (restricted) biderivative $`d_\omega `$ and the “fraction product” $``$ by the equation $`d_\omega d_\omega =0.`$ We begin with a construction of the biderivative in our restricted setting. Let $`\phi C^{}(K;\stackrel{}{𝐌})`$ and $`\psi C^{}(K;\stackrel{}{𝐌)}`$ be the cochains with top dimensional support such that
$$\phi \left(e^{i2}\right)=m_i\text{ and }\psi \left(e^{j2}\right)=\mathrm{\Delta }_j.$$
We think of $`\phi `$ and $`\psi `$ as acting on uprooted and downrooted trees, respectively (see Figure 3).
Figure 3: The actions of $`\phi `$ and $`\psi .`$
Let $`T^cH`$ denote the tensor coalgebra of $`H.`$ The *coderivation cochain of* $`\phi `$ is the cochain $`\phi ^cC^{}(K;\stackrel{}{𝐌})`$ that extends $`\phi `$ to cells of $`K`$ in codim $`1`$ such that
$$\underset{\text{codim}\text{ }e\text{ }=\text{ }0,1}{}\phi ^c\left(e\right)Coder\left(T^cH\right)$$
is the cofree linear coextension of $`\phi \left(K\right)=_{i2}\phi \left(e^{i2}\right)`$ as a coderivation. Thus if $`T𝒦`$ is an uprooted $`2`$-level tree with $`n+k`$ leaves and leaf sequence $`𝐱_i\left(k\right)`$,
$$\phi ^c\left(T\right)=1^{i1}m_k1^{ni+1}=\left[1\text{ }\mathrm{}\text{ }m_k\text{ }\mathrm{}\text{ }1\right]𝐌_{𝐱_i\left(k\right)}^1$$
and is represented by the arrow from $`(n+k,1)`$ to $`(n+1,1)`$ on the horizontal axis in $`^2.`$ Dually, let $`T^a\left(H\right)`$ denote the tensor algebra of $`H.`$ The *derivation cochain of* $`\psi `$ is the cochain $`\psi ^aC^{}(K;\stackrel{}{𝐌)}`$ that extends $`\psi `$ to cells of $`K`$ in codim $`1`$ such that
$$\underset{\text{codim}\text{ }e\text{ }=\text{ }0,1}{}\psi ^a\left(e\right)Der\left(T^aH\right)$$
is the free linear extension of $`\psi \left(K\right)=_{i2}\psi \left(e^{i2}\right)`$ as a derivation. Thus if $`T𝒦`$ is an downrooted $`2`$-level tree with $`n+k`$ leaves and leaf sequence $`𝐲_i\left(k\right)`$,
$$\psi ^a\left(T\right)=1^{i1}\mathrm{\Delta }_k1^{ni+1}=\left[1\text{ }\mathrm{}\text{ }\mathrm{\Delta }_k\text{ }\mathrm{}\text{ }1\right]^T𝐌_1^{𝐲_i\left(k\right)}$$
and is represented by the arrow from $`(1,n+1)`$ to $`(1,n+k)`$ on the vertical axis.
Evaluating leaf cup powers of $`\phi ^c`$ (respt. $`\psi ^a`$) generates a representative of each class of compositions involving the $`m_i`$’s (respt. $`\mathrm{\Delta }_j`$’s). So let
$$\xi =\phi ^c+\phi ^c_{\mathrm{}}\phi ^c+\mathrm{}+\left(\phi ^c\right)^_{\mathrm{}}k+\mathrm{}\text{ }$$
$$\zeta =\psi ^a+\psi ^a_{\mathrm{}}\psi ^a+\mathrm{}+\left(\psi ^a\right)^_{\mathrm{}}k+\mathrm{}$$
and note that if $`e`$ is a cell of $`K,`$ each non-zero component of $`\xi \left(e\right)`$ (respt. $`\zeta \left(e\right)`$) is represented by a left-oriented horizontal (respt. upward-oriented vertical) arrow.
Furthermore, evaluating wedge and Čech cup powers of $`\xi `$ (respt. $`\zeta `$) generates the components of the cofree coextension of $`\xi \left(K\right)`$ as a $`\mathrm{\Delta }_K`$-coderivation (respt. free extension of $`\zeta \left(K\right)`$ as a $`\mathrm{\Delta }_K`$-derivation). So let
$$\stackrel{}{\phi }=\xi +\xi \xi +\mathrm{}+\xi ^k+\mathrm{}$$
$$\stackrel{}{\psi }=\zeta +\zeta \zeta +\mathrm{}+\zeta ^k+\mathrm{}$$
and note that the component $`\xi ^k\left(e^{i2}\right):\left(H^i\right)^k\left(H^1\right)^k`$ is represented by a left-oriented horizontal arrow from $`(i,k)`$ to $`(1,k)`$ while the component $`\zeta ^k\left(e^{i2}\right):\left(H^1\right)^k\left(H^i\right)^k`$ is represented by a upward-oriented vertical arrow from $`(k,1)`$ to $`(k,i).`$
Let $`M_0=M_{1,1}`$. For reasons soon to become clear, the only structure relations involving the differential $`d`$ are the classical quadratic relations in an $`A_{\mathrm{}}`$-(co)algebra. Note that $`dM_0`$ and let $`\mathrm{𝟏}^s=(1,\mathrm{},1)^s.`$ Given $`\theta M_0`$ and $`p,q1,`$ consider the monomials $`\theta _i^{q\times 1}𝐌_\mathrm{𝟏}^{\mathrm{𝟏}^q}`$ and $`\theta _j^{1\times p}𝐌_{\mathrm{𝟏}^p}^1`$ all of whose entries are the identity except the $`i^{th}`$ in $`\theta _i^{q\times 1}`$ and the $`j^{th}`$ in $`\theta _j^{1\times p},`$ both of which are $`\theta .`$ Define $`Bd_0:M_0𝐌`$ by
$$Bd_0(\theta )=\underset{\begin{array}{c}1iq,\text{ }1jp\\ p,q1\end{array}}{}\theta _i^{q\times 1}+\theta _j^{1\times p}.$$
Then $`Bd_0\left(\theta \right)`$ is the (co)free linear (co)extension of $`\theta `$ as a (co)derivation. Note that each component of $`Bd_0\left(\theta \right)`$ is represented by an arrow of “length” zero.
Let $`M_1=\left(M{}_{1,}{}^{}M_{,1}\right)/M_{1,1}`$ and define $`Bd_1:M_1𝐌`$ by
(4.1)
$$Bd_1\left(\theta \right)=\underset{\begin{array}{c}e\text{ }\text{ }K\\ \text{codim }e\text{ }=\text{ }0\end{array}}{}(\stackrel{}{\phi }+\stackrel{}{\psi })\left(e\right)+\underset{\begin{array}{c}e\text{ }\text{ }K\\ \text{codim}\text{ }e\text{ }=\text{ }1\end{array}}{}\left(\phi ^c+\psi ^a\right)\left(e\right).$$
Note that the components of $`Bd_1\left(\theta \right)`$ are represented by upward-oriented vertical arrows and left-oriented horizontal arrows; the right-hand component of (4.1) is given by Gerstenhaber’s $``$-(co)operation.
Let $`\rho _0:𝐌𝐌_0`$ and $`\rho _1:𝐌𝐌_1`$ denote the canonical projections.
###### Definition 3.
The restricted biderivative is the (non-linear) map $`d_\underset{¯}{}:𝐌𝐌`$ given by
$$d_\underset{¯}{}=Bd_0\rho _0+Bd_1\rho _1.$$
The symbol $`d_\theta `$ denotes the restricted biderivative of $`\theta .`$
Finally, the composition
$$:𝐌\times 𝐌\stackrel{d_\underset{¯}{}d_\underset{¯}{}}{}𝐌\times 𝐌\stackrel{{\rm Y}}{}𝐌$$
defines the *fraction product*. Special $`A_{\mathrm{}}`$-bialgebras are defined in terms of the fraction product as follows:
###### Definition 4.
Let $`\omega =d+_{i,j2}\left(m_i+\mathrm{\Delta }_j\right)M_0M_1.`$ Then $`(H,d,m_i,\mathrm{\Delta }_j)_{i,j2}`$ is a special $`A_{\mathrm{}}`$-bialgebra provided
$$d_\omega d_\omega =0.$$
Note that one recovers the classical quadratic relations in an $`A_{\mathrm{}}`$-algebra when $`\omega =d+_{i2}m_i.`$
## 5. Structure Relations
The structure relations in a special $`A_{\mathrm{}}`$-bialgebra $`(H,d,m_i,\mathrm{\Delta }_j)_{i,j2}`$ follow easily from the following two observations:
1. If $`\theta ,\eta 𝐌,`$ then $`\theta \eta =0`$ whenever the projection of $`\theta `$ or $`\eta `$ to $`M_0M_1`$ is zero.
2. Each non-zero component in the projections of $`\theta `$ and $`\eta `$ is represented by a horizontal, vertical or zero length arrow.
By (1), each component of $`d_\omega d_\omega `$ is a “transgression” represented by a “2-step” path of arrows from the horizontal axis $`M_{1,}`$ to the vertical axis $`M_{,1};`$ and by (2), each such 2-step path follows the edges of a (possibly degenerate) rectangle positioned with one of its vertices at $`(1,1)`$.
Now relations involving $`d`$ arise from degenerate rectangles since arrows of length zero represent components in the (co)extensions of $`d`$. Hence $`d`$ interacts with the $`m_i`$’s or the $`\mathrm{\Delta }_j`$’s exclusively and the relations involving $`d`$ are exactly the classical quadratic relations in an $`A_{\mathrm{}}`$-(co)algebra.
On the other hand, relations involving the $`m_i`$’s and $`\mathrm{\Delta }_j`$’s arise from non-degenerate rectangles since $`m_i`$ and $`\mathrm{\Delta }_j`$ are represented by the arrows $`(i,1)(1,1)`$ and $`(1,1)(1,j)`$. While the two-step path $`(i,1)(1,1)(1,j)`$ represents the (usual) composition $`\mathrm{\Delta }_jm_i,`$ the two-step path $`(i,1)(i,j)(1,j)`$ represents $`\xi ^j\left(e^{i2}\right)\zeta ^i\left(e^{j2}\right).`$ Thus we obtain the relation
$$\mathrm{\Delta }_jm_i=\xi ^j\left(e^{i2}\right)\zeta ^i\left(e^{j2}\right).$$
For example, by setting $`i=j=2`$ we obtain the classical bialgebra relation
$$\mathrm{\Delta }_2m_2=\left[\begin{array}{c}m_2\\ m_2\end{array}\right]\left[\mathrm{\Delta }_2\text{ }\mathrm{\Delta }_2\right].$$
And with $`(i,j)=(3,2)`$ we obtain
$$\mathrm{\Delta }_2m_3=\left\{\left[\begin{array}{c}m_3\\ m_2\left(1m_2\right)\end{array}\right]+\left[\begin{array}{c}m_2\left(m_21\right)\\ m_3\end{array}\right]\right\}\left[\mathrm{\Delta }_2\text{ }\mathrm{\Delta }_2\text{ }\mathrm{\Delta }_2\right]$$
(see Figure 4).
Figure 4: Some low order arrows in $`𝐌.`$
We summarize the discussion above in our main theorem:
###### Theorem 1.
$`(H,d,m_i,\mathrm{\Delta }_j)_{i,j2}`$ is a special $`A_{\mathrm{}}`$-bialgebra if $`(H,d,m_i)_{i2}`$ is an $`A_{\mathrm{}}`$-algebra, $`(H,d,\mathrm{\Delta }_j)_{j2}`$ is an $`A_{\mathrm{}}`$-coalgebra and for all $`i,j2,`$
$$\mathrm{\Delta }_jm_i=\xi ^j\left(e^{i2}\right)\zeta ^i\left(e^{j2}\right).$$
|
warning/0506/math0506622.html
|
ar5iv
|
text
|
# Stable base loci, movable curves, and small modifications, for toric varieties
## 1. Introduction
Let $`X`$ be a smooth $`n`$-dimensional complex projective variety. Boucksom, Demailly, Paun, and Peternell have recently shown that the dual of the cone of numerical classes of effective divisors on $`X`$ is the closure of the cone of numerical classes of movable curves \[BDPP, Theorem 0.2\]; see also \[PAG, 11.4.C\]. Their theorem is analogous to a theorem of Kleiman that says that the dual of the cone of numerical classes of ample divisors is the closure of the cone of numerical classes of effective curves \[PAG, 1.4.23\]. Debarre and Lazarsfeld asked whether these results might generalize in the following way: for $`1<k<n`$, is the dual of the cone of numerical classes of divisors on $`X`$ whose stable base locus has dimension less than $`k`$ generated by the numerical class of a natural collection of curves, such as curves moving in a family that sweeps out a subvariety of dimension $`k`$? This paper gives an affirmative answer to their question in the toric case, with a slight twist—the curves that one must consider include not only curves on $`X`$, but also curves on small modifications of $`X`$.
Recall that a small modification $`f:XX^{}`$ is a birational map that is an isomorphism in codimension 1. If $`V`$ is a subvariety of $`X`$, we say that $`f`$ maps $`V`$ birationally to $`f(V)`$ if $`f`$ is defined on an open set meeting $`V`$ and maps a dense open subset of $`V`$ isomorphically onto a dense open subset of $`f(V)`$. If $`D=d_iD_i`$ is a divisor on $`X`$, then we write $`f(D)`$ for the divisor $`d_if(D_i)`$ on $`X^{}`$.
###### Theorem 1.
Let $`X`$ be a complete $``$-factorial toric variety with dense torus $`T`$, and let $`D`$ be a divisor on $`X`$. The following are equivalent:
1. The stable base locus $`B(D)`$ has dimension less than $`k`$.
2. For every $`T`$-invariant subvariety $`VX`$ of dimension $`k`$, and for every small modification $`f:XX^{}`$ that maps $`V`$ birationally to $`f(V)`$, with $`X^{}`$ projective and $``$-factorial, and for every irreducible curve $`C`$ on $`X^{}`$ moving in a family sweeping out $`f(V)`$, $`(f(D)C)0.`$
Recall that the stable base locus $`B(D)`$ of a $``$-divisor $`D`$ is the set-theoretic intersection of the base loci of the complete linear systems $`|mD|`$ for all positive integers $`m`$ such that $`mD`$ is integral. Let $`N^1(X)_{}`$ be the space of $``$-divisors on $`X`$, modulo numerical equivalence. Since the stable base locus of the sum of two divisors is contained in the union of their stable base loci, the set of classes in $`N^1(X)_{}`$ that are represented by divisors whose stable base locus has codimension greater than $`k`$ is a convex cone (for a general numerical class, the stable base locus is independent of the choice of representative \[PAG, 2.1 and 10.3\]). We write $`\overline{\mathrm{Amp}}^k(X)`$ for the closure of this cone in $`N^1(X)_{}:=N^1(X)_{}`$. This notation is suggestive of the fact that $`\overline{\mathrm{Amp}}^k(X)`$ is also the closure of the cone of numerical classes of divisors that are ample in codimension $`k`$, that is, the cone of numerical classes of divisors $`D`$ such that there is an open set $`U`$, with $`𝒪(D)|_U`$ ample, whose complement has codimension greater than $`k`$. The cone $`\overline{\mathrm{Eff}}^1(X)`$ of pseudo-effective divisors, the closure of the cone of numerical clases of effective divisors, is filtered by the cones $`\overline{\mathrm{Amp}}^k(X)`$:
$$\overline{\mathrm{Eff}}^1(X)=\overline{\mathrm{Amp}}^0(X)\mathrm{}\overline{\mathrm{Amp}}^{n1}(X)=\overline{\mathrm{Amp}}(X).$$
We have an analogous filtration of the cone $`\overline{\mathrm{Eff}}_1(X)_{}`$ of pseudo-effective curves, as follows. Let $`N_1(X)_{}`$ be the space of 1-cycles in $`X`$ with coefficients in $``$, modulo numerical equivalence. The intersection pairing on $`X`$ induces a natural isomorphism $`N_1(X)_{}N^1(X)_{}^{}`$. Let $`\mathrm{Mov}_k(X)`$ be the the convex cone in $`N_1(X)_{}`$ generated by classes of irreducible curves moving in a family that sweeps out a $`k`$-dimensional subvariety of $`X`$. So
$$\overline{\mathrm{Mov}}(X)=\overline{\mathrm{Mov}}_n(X)\mathrm{}\overline{\mathrm{Mov}}_1(X)=\overline{\mathrm{Eff}}_1(X).$$
If a curve $`C`$ moves in a family that sweeps out a $`k`$-dimensional subvariety, and if $`B(D)`$ has dimension smaller than $`k`$, then $`C`$ can be moved off of the support of some effective divisor linearly equivalent to a multiple of $`D`$, and hence $`(DC)`$ is nonnegative. So there are natural inclusions
$$\overline{\mathrm{Mov}}_k(X)\overline{\mathrm{Amp}}^{nk}(X)^{}\text{ and }\overline{\mathrm{Amp}}^{nk}(X)\overline{\mathrm{Mov}}_k(X)^{}.$$
Kleiman’s theorem says that equality holds for $`k=1`$, and the theorem of Boucksom, Demailly, Paun, and Peternell says that equality holds for $`k=n`$. As Lazarsfeld points out, one does not expect equality for $`1<k<n`$; in light of the equality for $`k=n`$, equality for general $`k`$ would say, roughly speaking, that the restriction of a big divisor to its stable base locus is not pseudo-effective, which seems too strong (a classical theorem of Fujita says that the restriction of a divisor to its stable base locus is not ample \[Fuj, Theorem 1.19\]).
Another heuristic reason to believe that equality should not hold for general $`k`$ comes from the consideration of small modifications $`f:XX^{}`$, with $`X^{}`$ complete and $``$-factorial. Numerical equivalence on $`X`$ and on $`X^{}`$ coincide (this follows, for instance, from Kleiman’s characterization of numerically trivial line bundles \[SGA6, XIII, Theorem 4.6\]). So the map $`Df(D)`$ induces an isomorphism $`N^1(X)_{}N^1(X^{})_{}`$. The global sections of $`D`$ and of $`f(D)`$ are canonically identified, so if $`V`$ is contained in $`B(D)`$, and if $`f`$ maps $`V`$ birationally to $`f(V)`$, then $`f(V)`$ is in the stable base locus of $`f(D)`$. In particular, $`\overline{\mathrm{Amp}}^1(X)`$ is identified with $`\overline{\mathrm{Amp}}^1(X^{})`$. But there is no natural way to identify $`\overline{\mathrm{Mov}}_{n1}(X)`$ and $`\overline{\mathrm{Mov}}_{n1}(X^{})`$ compatibly with the intersection pairing.
We now introduce additional cones of numerical classes of 1-cycles which take into account $`k`$-movable curves on small modifications of $`X`$. Let $`f:XX^{}`$ be a small modification, with $`X^{}`$ complete and $``$-factorial. We define
$$\mathrm{Mov}_k(X,X^{})N_1(X)_{}$$
to be the image of the convex cone generated by numerical classes of irreducible curves $`C`$ on $`X^{}`$ moving in a family that sweeps out the birational image of a $`k`$-dimensional subvariety of $`X`$ under the identifications
$$N_1(X^{})_{}=N^1(X^{})_{}^{}N^1(X)_{}^{}=N_1(X)_{}.$$
There is an obvious inclusion
$$\overline{\mathrm{Mov}}_k(X,X^{})\overline{\mathrm{Amp}}^{nk}(X)^{}.$$
With this notation, Theorem 1 can be restated as follows.
###### Theorem 2.
Let $`X`$ be a complete $`n`$-dimensional $``$-factorial toric variety. Then
$$\overline{\mathrm{Amp}}^{nk}(X)^{}=\underset{f:XX^{}}{}\overline{\mathrm{Mov}}_k(X,X^{}),$$
where the sum is over all small modifications $`f:XX^{}`$ such that $`X^{}`$ is projective and $``$-factorial.
It seems unclear whether a statement like Theorem 2 should be true in general, because it is not clear whether a general variety $`X`$ will have enough $``$-factorial projective small modifications. By \[HK, Lemma 1.6\], this depends essentially on whether $`X`$ has enough divisors $`D`$ whose section ring
$$R(X,D)=\underset{m0}{}H^0(X,𝒪(mD))$$
is finitely generated.
Problem. Let $`X`$ be a projective $`n`$-dimensional $``$-factorial complex variety. Is the inclusion
$$\underset{f:XX^{}}{}\overline{\mathrm{Mov}}_k(X,X^{})\overline{\mathrm{Amp}}^{nk}(X)^{}$$
an equality?
The following example illustrates the necessity of considering curves on small modifications; we give an example of a threefold $`X`$ such that $`\overline{\mathrm{Mov}}_2(X)`$ is properly contained in $`\overline{\mathrm{Amp}}^1(X)^{}`$.
###### Example 1.
*Let $`Y`$ be the projectivized vector bundle*
$$Y=(𝒪(3)𝒪_^2)\stackrel{\pi }{}^2,$$
*and let $`s`$ be the section of $`\pi `$ whose image is $`(𝒪(3))`$. Let $`p_1,p_2,`$ and $`p_3`$ be noncolinear points in $`^2`$, and let $`X`$ be the blow up of $`Y`$ at $`s(p_1)`$, $`s(p_2)`$, and $`s(p_3)`$. Let $`E_i`$ be the exceptional divisor over $`s(p_i)`$. We make the following claims, which will be justified in the next section using toric methods.*
1. The divisors $`E_1,E_2,`$ and $`E_3`$, together with $`D^+`$, the strict transform of $`(𝒪(3))`$, and $`D^{}`$, the strict transform of $`(𝒪_^2)`$, give a basis for $`N^1(X)_{}`$.
2. The class $`cN_1(X)_{}`$ defined by $`(E_ic)=1`$, $`(D^+c)=0`$, and $`(D^{}c)=3`$, generates an extremal ray of $`\overline{\mathrm{Amp}}^1(X)^{}`$.
*Assume these claims are true. Choose in $`N^1(X)_{}`$ an open cone $`U`$ containing $`c`$ sufficiently small so that any class in $`U`$ pairs negatively with $`D^{}`$ and positively with the $`E_i`$. Any effective representative of a class in $`U`$ must have a component that lies in $`D^{}`$, and $`D^{}`$ is disjoint from the $`E_i`$, so $`U`$ does not contain the class of any irreducible curve on $`X`$. By perturbing a supporting hyperplane cutting out the extremal ray spanned by $`c`$, we can find a closed half space $`H`$ containing $`\overline{\mathrm{Amp}}^1(X)^{}(\overline{\mathrm{Amp}}^1(X)^{}U)`$ but not containing $`c`$. Since $`\overline{\mathrm{Mov}}_2(X)`$ is generated by classes in $`\overline{\mathrm{Amp}}^1(X)^{}(\overline{\mathrm{Amp}}^1(X)^{}U)`$, $`\overline{\mathrm{Mov}}_2(X)`$ is contained in $`H`$ and does not contain $`c`$.*
*However, consider the small modification*
$$f:XX^{}$$
*given by flopping the three curves $`\pi ^1(p_i)`$. The variety $`X^{}`$ can be realized as a projectivized vector bundle over the blow up of $`^2`$ at $`p_1`$, $`p_2`$, and $`p_3`$,*
$$X^{}=(L𝒪_{B\mathrm{}_3^2})\stackrel{\pi ^{}}{}B\mathrm{}_3^2,$$
*where $`L`$ is the line bundle given by the strict transform of a smooth cubic passing through the $`p_i`$. Let $`s^{}`$ be the section of $`\pi ^{}`$ whose image is $`(𝒪_{B\mathrm{}_3^2})`$. Then $`c`$ is represented on $`X^{}`$ by $`s^{}(C)`$, where $`C`$ is the strict transform of a smooth conic in $`^2`$ passing through the $`p_i`$, and $`s^{}(C)`$ moves in a family that sweeps out the surface $`(𝒪_{B\mathrm{}_3^2})`$ in $`X^{}`$.*
## 2. Toric preliminaries
For the remainder of this paper, we fix a complete simplicial fan $`\mathrm{\Delta }`$ in an $`n`$-dimensional latticed vector space $`N_{}=N_{}`$. Let $`X=X(\mathrm{\Delta })`$ be the corresponding toric variety over some fixed algebraically closed field, and let $`T`$ be the dense torus in $`X`$. Let $`\rho _1,\mathrm{},\rho _r`$ be the rays (or 1-dimensional cones) of $`\mathrm{\Delta }`$. Let $`v_iN`$ be the primitive generator of $`\rho _i`$, and let $`D_i`$ be the corresponding prime $`T`$-invariant divisor on $`X`$. See \[Ful\] for details on the correspondence between fans and toric varieties.
Since $`\mathrm{\Delta }`$ is complete and simplicial, $`X(\mathrm{\Delta })`$ is $``$-factorial, and numerical and rational equivalence coincide on $`X`$. We have a short exact sequence
$$0N_1(X)_{}^rN_{}0,$$
where the map from $`^r`$ takes $`(a_1,\mathrm{},a_r)`$ to $`a_1v_1+\mathrm{}+a_rv_r`$. The dual short exact sequence is
$$0N^1(X)_{}^rM_{}0,$$
where $`M`$ is the character lattice of $`T`$, and the map from $`^r`$ takes $`(d_1,\mathrm{},d_r)`$ to the numerical class of the divisor $`d_1D_1+\mathrm{}+d_rD_r`$.
If $`D=d_1D_1+\mathrm{}+d_rD_r`$, then the polytope $`P_DM_{}`$ is defined by
$$P_D=\{uM_{}:u,v_id_i\text{ for all }1ir\}.$$
If $`D`$ is integral, the characters $`\chi ^u`$, for all $`uP_DM`$, form a basis for $`H^0(X,𝒪(D))`$. The vanishing locus of $`\chi ^uH^0(X,𝒪(D))`$ is supported exactly on the union of the divisors $`D_i`$ such that $`u,v_i>d_i`$. It follows that the base locus, and hence the stable base locus, of any divisor on $`X`$ is $`T`$-invariant. See \[HKP\] for more details.
If $`V(\tau )`$ is the $`T`$-invariant subvariety corresponding to a cone $`\tau \mathrm{\Delta }`$, then $`V(\tau )`$ is not contained in the stable base locus of $`D`$ if and only if the class of $`D`$ in $`N^1(X)_{}`$ is in the cone $`\mathrm{\Gamma }_\tau `$ spanned by the classes of $`T`$-invariant divisors not containing $`V(\tau )`$. In other words,
$$\mathrm{\Gamma }_\tau =D_j:\rho _j\tau .$$
Since base loci are $`T`$-invariant, and the codimension of $`V(\tau )`$ is the dimension of $`\tau `$,
$$\overline{\mathrm{Amp}}^k(X)=\underset{dim\tau =k}{}\mathrm{\Gamma }_\tau .$$
Since each $`\mathrm{\Gamma }_\tau `$ is a rational polyhedral cone, it follows that $`\overline{\mathrm{Amp}}^k(X)`$ is also a rational polyhedral cone. Furthermore, the dual of $`\overline{\mathrm{Amp}}^k(X)`$ is given by
$$\overline{\mathrm{Amp}}^k(X)^{}=\underset{dim\tau =k}{}\mathrm{\Gamma }_\tau ^{}.$$
From the description of $`\mathrm{\Gamma }_\tau `$ and the exact sequences above, we have
$$\mathrm{\Gamma }_\tau ^{}=\{(a_1,\mathrm{},a_r)^r:a_1v_1+\mathrm{}+a_rv_r=0\text{ and }a_i0\text{ for }v_i\tau \}.$$
The cones $`\{\mathrm{\Gamma }_\tau \}`$, for maximal cones $`\tau `$, appear prominently in the “bunches of cones” in \[BH\].
We now prove the claims made in Example 1.
###### Example 2.
*Suppose $`N=^3`$, $`r=8`$, and*
$$\begin{array}{cccc}v_1=(1,1,1),\hfill & v_2=(1,0,1),\hfill & v_3=(0,1,1),\hfill & v_4=(1,0,1),\hfill \\ v_5=(0,1,1),\hfill & v_6=(1,1,1),\hfill & v_7=(0,0,1),\hfill & v_8=(0,0,1),\hfill \end{array}$$
*with $`\mathrm{\Delta }`$ being the fan whose maximal cones are*
$$\begin{array}{cccc}v_1,v_4,v_8,\hfill & v_1,v_5,v_8,\hfill & v_2,v_5,v_8,\hfill & v_2,v_6,v_8,\hfill \\ v_3,v_6,v_8,\hfill & v_3,v_4,v_8,\hfill & v_1,v_4,v_5,\hfill & v_2,v_5,v_6,\hfill \\ v_3,v_4,v_6,\hfill & v_4,v_5,v_7,\hfill & v_5,v_6,v_7,\hfill & v_4,v_6,v_7.\hfill \end{array}$$
*Then $`X=X(\mathrm{\Delta })`$ is the variety considered in Example 1, $`D_i=E_i`$ for $`i=1,2,`$ and $`3`$, $`D_7=D^{}`$ is the strict transform of $`(𝒪_^2)`$, and $`D_8=D^+`$ is the strict transform of $`(𝒪(3))`$. See \[Oda, pp.58–59\] for the construction of the fan corresponding to a projectivized split vector bundle on a toric variety. The following diagram illustrates the fan $`\mathrm{\Delta }^{}`$ corresponding to $`X^{}`$, as well as the fan $`\mathrm{\Delta }`$; the intersection of each fan with the hyperplane $`\{(x,y,z)^3:z=1\}`$ is shown. See \[Rei\] for the changes in the fan corresponding to a flop.*
$`\mathrm{\Delta }^{}(z=1)`$ $`\mathrm{\Delta }(z=1)`$
*We claimed that the classes of $`D_1,D_2,D_3,D_7`$, and $`D_8`$ give a basis for $`N^1(X)_{}`$. This now follows immediately from the short exact sequence*
$$0M_{}^8N^1(X)_{}0,$$
*and the fact that $`\{v_4,v_5,v_6\}`$ is a basis for $`N_{}=M_{}^{}`$. It remains to show that the class $`cN_1(X)_{}`$ given by $`(D_1c)=(D_2c)=(D_3c)=1`$, $`(D_7c)=3`$, and $`(D_8c)=0`$ spans an extremal in $`\overline{\mathrm{Amp}}^1(X)^{}`$. Recall that*
$$\overline{\mathrm{Amp}}^1(X)^{}=\underset{i=1}{\overset{8}{}}\mathrm{\Gamma }_{\rho _i}^{}.$$
*For $`i7`$, $`D_7`$ is in $`\mathrm{\Gamma }_{\rho _i}`$, and hence $`c`$ is not in $`\mathrm{\Gamma }_{\rho _i}^{}`$. Therefore, it suffices to show that $`c`$ spans an extremal ray in $`\mathrm{\Gamma }_{\rho _7}^{}`$, which is clear: $`c`$ spans the inward normal to the face of $`\mathrm{\Gamma }_{\rho _7}`$ spanned by the classes of $`D_4,D_5,D_6`$, and $`D_8`$.*
## 3. Constructing curves on toric varieties
By the exact sequence
$$0N_1(X)_{}^rN_{}0,$$
a numerical class $`cN_1(X)_{}`$ is represented uniquely by an $`r`$-tuple of real numbers $`(a_1,\mathrm{},a_r)`$, such that $`(D_ic)=a_i`$, and these numbers satisfy the relation
$$a_1v_1+\mathrm{}+a_rv_r=0.$$
Conversely, any such $`r`$-tuple $`(a_i)`$ corresponds to some numerical class in $`N_1(X)_{}`$.
Given a cone $`\tau \mathrm{\Delta }`$, we say that a ray $`\rho _i`$ is adjacent to $`\tau `$ if there is a cone $`\sigma \mathrm{\Delta }`$ containing both $`\tau `$ and $`\rho _i`$.
###### Proposition 1.
Let $`C`$ be an irreducible curve in $`X`$, with $`(D_iC)=a_i`$. Then there is a cone $`\tau \mathrm{\Delta }`$ such that
1. $`C`$ moves in a family sweeping out $`V(\tau )`$,
2. $`a_i=0`$ unless $`\rho _i`$ is adjacent to $`\tau `$,
3. $`a_i0`$ unless $`\rho _i`$ is in $`\tau `$.
###### Proof.
Let $`\tau `$ be the unique cone such that the orbit $`O_\tau `$ contains an open dense subset of $`C`$. The action of $`T`$ moves $`C`$ in a family sweeping out $`V(\tau )`$. Furthermore, if $`\rho _i`$ is not adjacent to $`\tau `$, then $`D_i`$ is disjoint from $`V(\tau )`$, so $`D_iC=0`$. Similarly, if $`\rho _i`$ is not in $`\tau `$, then $`D_i`$ does not contain $`O_\tau `$, and hence does not contain $`C`$, so $`(D_iC)0`$. ∎
Proposition 1 restricts the numerical classes of 1-cycles that can be represented by a positive scalar multiple of an irreducible curve. We now show, by explicitly constructing curves with given numerical properties, that these are the only nontrivial restrictions.
###### Proposition 2.
Let $`\tau `$ be a cone in $`\mathrm{\Delta }`$, and let $`a_1,\mathrm{},a_r`$ be integers such that
1. $`a_1v_1+\mathrm{}+a_rv_r=0`$,
2. $`a_i=0`$ unless $`\rho _i`$ is adjacent to $`\tau `$,
3. $`a_i0`$ unless $`\rho _i`$ is in $`\tau `$.
Then there is an irreducible curve $`C`$ that moves in a family sweeping out $`V(\tau )`$ such that $`(D_iC)=a_i`$ for all $`1ir`$.
###### Proof.
We first consider the case $`\tau =0`$. So we are given nonnegative integers $`a_1,\mathrm{},a_r`$ such that $`a_1v_1+\mathrm{}+a_rv_r=0`$, and must construct a curve $`C`$ sweeping out $`X`$ such that $`(D_iC)=a_i`$ for all $`1ir`$. Choose distinct elements $`\lambda _1,\mathrm{},\lambda _r`$ in the base field, let $`\phi _i:𝔾_mT`$ be the one-parameter subgroup corresponding to $`v_i`$, and consider the rational map $`\varphi :𝔸^1T`$ given by
$$\varphi (z)=\underset{i=1}{\overset{r}{}}\phi _i(z\lambda _i)^{a_i},$$
where the product is given by the group law on $`T`$. Since $`X`$ is complete, $`\varphi `$ extends to a regular morphism $`^1X`$. It is straightforward to check, using local coordinates, that if $`a_i=0`$ then $`\varphi (^1)`$ is disjoint from $`D_i`$, and if $`a_i`$ is positive then
$$D_i\varphi (^1)=\varphi (\lambda _i).$$
Furthermore, in the latter case, $`\varphi (\lambda _i)`$ is a point in the dense orbit $`O_{\rho _i}D_i`$, along which $`D_i`$ is locally principal, and the local intersection multiplicity is
$$(D_i\varphi (^1))_{\varphi (\lambda _i)}=a_i.$$
The action of $`T`$ moves $`\varphi (^1)`$ in a family sweeping out $`X`$.
We now consider the case of a nonzero cone $`\tau `$ in $`\mathrm{\Delta }`$. Let $`N_\tau `$ be the quotient lattice
$$N_\tau =N/(N\mathrm{Span}\tau ),$$
and let $`\mathrm{\Delta }_\tau `$ be the fan in $`(N_\tau )_{}`$ whose cones are the projections of cones of $`\mathrm{\Delta }`$ containing $`\tau `$, so $`V(\tau )`$ is the toric variety corresponding to $`\mathrm{\Delta }_\tau `$ \[Ful, 3.1\]. The rays of $`\mathrm{\Delta }_\tau `$ are the projections of the rays of $`\mathrm{\Delta }`$ adjacent to $`\tau `$, but not contained in $`\tau `$; after renumbering, we may assume that these rays are $`\rho _1,\mathrm{},\rho _s`$. Let $`\overline{\rho }_i`$ be the image of $`\rho _i`$ under the projection $`\pi :N_{}(N_\tau )_{}`$, let $`w_i`$ be the primitive generator of $`\overline{\rho }_i`$ in $`N_\tau `$, let $`m_i`$ be the positive integer such that $`\pi (v_i)=m_iw_i`$, and let $`\overline{D}_i`$ be the divisor in $`V(\tau )`$ corresponding to $`\overline{\rho }_i`$, for $`1is`$. By hypothesis,
$$a_1m_1w_1+\mathrm{}+a_rm_rw_r=0$$
in $`N_\tau `$. Hence there is a curve $`C`$ sweeping out $`V(\tau )`$ such that $`(\overline{D}_iC)=a_im_i`$ for $`1is`$. We claim that $`C`$ is the required curve. Let $`\iota :V(\tau )X`$ be the natural inclusion, so $`(D_iC)=(\iota ^{}D_iC)`$. By the basic properties of toric intersection theory, as developed in \[Ful, 5.1\], $`\iota ^{}D_i=\overline{D}_i/m_i`$, so
$$(D_iC)=(\overline{D}_iC)/m_i=a_i,$$
for $`1is`$. For $`i>s`$, if $`\rho _i`$ is not adjacent to $`\tau `$, then $`D_i`$ is disjoint from $`V(\tau )`$ and hence $`D_iC=0`$. And if $`\rho _i`$ is in $`\tau `$, then $`(D_iC)`$ is uniquely determined by the condition $`(D_iC)v_i=0`$. Therefore $`(D_iC)=a_i`$, for all $`i`$, as required. ∎
## 4. Constructing small modifications of toric varieties
In order to prove the conjecture in the toric case, we need to know that we have enough small modifications to work with. Although it would suffice to construct enough small modifications torically, it is perhaps helpful to observe that every small modification $`f:XX^{}`$, with $`X^{}`$ projective, is toric, in the sense that $`X^{}`$ has the structure of a toric variety with dense torus $`T`$ such that $`f`$ is $`T`$-equivariant. This fact follows from \[HK, Proposition 1.11\]; it can also be seen directly by choosing a very ample divisor $`A^{}`$ on $`X^{}`$. Then $`A^{}`$ is the birational transform of some divisor $`A`$ on $`X`$, which is linearly equivalent to a $`T`$-invariant divisor $`A^{}`$. The natural map
$$f^{}:X(H^0(X,𝒪(A^{})))$$
is $`T`$-equivariant, and agrees with the map to $`X^{}`$, embedded by $`A^{}`$, up to a projective linear change of coordinates.
Roughly speaking, the following proposition says that, given a cone $`\tau `$ and a collection of rays “close to $`\tau `$” and “surrounding $`\tau `$” in $`\mathrm{\Delta }`$, there is a complete simplicial fan $`\mathrm{\Delta }^{}`$ containing $`\tau `$, whose rays are exactly the rays of $`\mathrm{\Delta }`$, such that each of the rays in the given collection is adjacent to $`\tau `$ in $`\mathrm{\Delta }^{}`$. Such a fan corresponds to a complete $``$-factorial small modification of $`X`$.
###### Proposition 3.
Let $`v_1,\mathrm{},v_k`$ span a cone $`\tau `$ in $`\mathrm{\Delta }`$. Assume there exists $`s>k+1`$ such that
1. $`\{v_1,\mathrm{},v_{s1}\}`$ is linearly independent,
2. there is a linear relation
$$a_1v_1+\mathrm{}+a_kv_k=a_{k+1}v_{k+1}+\mathrm{}+a_sv_s,$$
with $`a_i>0`$ for all $`i`$,
3. $`v_j`$ is not contained in the convex cone $`v_1,\mathrm{},v_s`$, for $`j>s`$.
Then there is a complete simplicial fan $`\mathrm{\Delta }^{}`$ containing $`\tau `$, whose rays are exactly the rays of $`\mathrm{\Delta }`$, such that $`\rho _i`$ is adjacent to $`\tau `$ in $`\mathrm{\Delta }^{}`$ for $`k+1is`$. Furthermore, $`\mathrm{\Delta }^{}`$ can be chosen such that $`X(\mathrm{\Delta }^{})`$ is projective.
###### Proof.
Choose large positive numbers $`pq0`$, and small positive numbers $`ϵ_{s+1},\mathrm{},ϵ_r`$. Let $`Q`$ be the polytope in $`N_{}`$
$$Q=\mathrm{conv}\{pv_1,\mathrm{},pv_k,qv_{k+1},\mathrm{},qv_s,ϵ_{s+1}v_{s+1},\mathrm{},ϵ_rv_r\},$$
and let $`\mathrm{\Delta }_Q`$ be the fan whose nonzero cones are the cones over the faces of $`Q`$. It is straightforward to check that
$$\mathrm{conv}\{pv_1,\mathrm{},pv_k,qv_{k+1},\mathrm{},\widehat{qv_i},\mathrm{},qv_s\}$$
is a face of $`Q`$, and so $`v_1,\mathrm{},\widehat{v_i},\mathrm{},v_s`$ is a cone in $`\mathrm{\Delta }_Q`$, for $`k+1is`$.
By construction, the rays of $`\mathrm{\Delta }_Q`$ are a subset of the rays of $`\mathrm{\Delta }`$, and, provided that the $`ϵ_j`$ are sufficiently general, $`\mathrm{\Delta }_Q`$ is simplicial. Let $`\mathrm{\Delta }^{}`$ be constructed by successive star subdivisions of $`\mathrm{\Delta }_Q`$ with respect to the rays of $`\mathrm{\Delta }`$ that are not in $`\mathrm{\Delta }_Q`$. So $`\mathrm{\Delta }^{}`$ is also simplicial, and since, by hypothesis, none of the rays that are added lie in the cone spanned by $`\{v_1,\mathrm{},v_s\}`$, the cones $`v_1,\mathrm{},\widehat{v_i},\mathrm{},v_s`$ in $`\mathrm{\Delta }_Q`$ remain unchanged in $`\mathrm{\Delta }^{}`$. In particular, $`\tau `$ is a cone in $`\mathrm{\Delta }^{}`$, and $`\rho _i`$ is adjacent to $`\tau `$ in $`\mathrm{\Delta }^{}`$ for $`k+1is`$.
It remains to check that $`X(\mathrm{\Delta }^{})`$ is projective. Since $`Q`$ is a convex polytope, $`X(\mathrm{\Delta }_Q)`$ is projective, and since $`\mathrm{\Delta }^{}`$ is constructed from $`\mathrm{\Delta }_Q`$ by a sequence of star subdivisions, $`X(\mathrm{\Delta }^{})`$ is constructed from $`X(\mathrm{\Delta }_Q)`$, which is projective, by a sequence of blow ups. So $`X(\mathrm{\Delta }^{})`$ is also projective. ∎
## 5. Proof of Theorems 1 and 2
###### Proof.
Let $`cN_1(X)_{}`$ be a numerical class spanning an extremal ray of $`\overline{\mathrm{Amp}}^{\mathrm{}}(X)^{}`$. Since
$$\overline{\mathrm{Amp}}^{\mathrm{}}(X)^{}=\underset{dim\sigma =\mathrm{}}{}\mathrm{\Gamma }_\sigma ^{},$$
$`c`$ spans an extremal ray of $`\mathrm{\Gamma }_\sigma ^{}`$ for some $`\mathrm{}`$-dimensional cone $`\sigma \mathrm{\Delta }`$. We will show that there is a face $`\tau \sigma `$, a small modification $`f:XX^{}`$ that is birational on $`V=V(\tau )`$, and an irreducible curve $`C`$ sweeping out the birational transform $`f(V)`$ on $`X^{}`$, such that $`C`$ represents the numerical class $`c`$.
Since $`c`$ is in $`\mathrm{\Gamma }_\sigma ^{}`$, $`c`$ is given by $`(D_ic)=a_i`$, with $`a_i0`$ for $`v_i\sigma `$. After renumbering, we may assume that
$$(D_ic)\{\begin{array}{cc}<0\hfill & \text{ for }1ik.\hfill \\ >0\hfill & \text{ for }k+1is.\hfill \\ =0\hfill & \text{ for }i>s.\hfill \end{array}$$
Let $`\tau `$ be the face of $`\sigma `$ spanned by $`\{v_1,\mathrm{},v_k\}`$. Now we have
$$a_1v_1\mathrm{}a_kv_k=a_{k+1}v_{k+1}+\mathrm{}+a_sv_s.$$
If $`k=0`$ then, by Proposition 2, there is a curve $`C`$ moving in a family that sweeps out $`X`$ such that $`(DC)=a_i`$ for all $`i`$. So we may assume $`k1`$. Since $`v_j`$ is not in $`\tau `$ for $`j>k`$, we must also have $`s>k+1`$. By Propositions 2 and 3, it will therefore suffice to show that $`\{v_1,\mathrm{},v_{s1}\}`$ is linearly independent, and that $`v_j`$ is not contained in the convex cone $`v_1,\mathrm{},v_s`$ for $`j>s`$.
Suppose there is a linear relation $`b_1v_1+\mathrm{}+b_{s1}v_{s1}=0`$. Then we have a class $`bN_1(X)_{}`$ given by
$$(D_ib)=\{\begin{array}{cc}b_i\hfill & \text{ for }1i<s.\hfill \\ 0\hfill & \text{ for }is.\hfill \end{array}$$
For small $`ϵ`$, the classes $`c+ϵb`$ and $`cϵb`$ lie in $`\mathrm{\Gamma }_\sigma ^{}`$, but not in the ray spanned by $`c`$, and $`2c`$ can be written as
$$2c=(cϵb)+(c+ϵb),$$
contradicting the assumption that $`c`$ spans an extremal ray of $`\mathrm{\Gamma }_\sigma ^{}`$.
Similarly, if $`v_j`$ is in the cone spanned by $`\{v_1,\mathrm{},v_s\}`$ for some $`j>s`$, then we have a linear relation $`v_j=b_1v_1+\mathrm{}+b_sv_s`$, with all of the $`b_i0`$. So there is a class $`bN_1(X)_{}`$ given by
$$(D_ib)=\{\begin{array}{cc}b_i\hfill & \text{ for }1is.\hfill \\ 1\hfill & \text{ for }i=j.\hfill \\ 0\hfill & \text{ for }i>s\text{ and }ij.\hfill \end{array}$$
Now $`b`$ is contained in $`\mathrm{\Gamma }_j^{}\overline{\mathrm{Amp}}^1(X)^{}`$, which is contained in $`\overline{\mathrm{Amp}}^{\mathrm{}}(X)^{}`$, since $`\mathrm{}k`$ and we have assumed $`k1`$. For small positive $`ϵ`$, $`cϵb`$ is also contained in $`\mathrm{\Gamma }_\tau ^{}\overline{\mathrm{Amp}}^{\mathrm{}}(X)^{}`$. Then $`c`$ can be written as
$$c=(cϵb)+ϵb,$$
contradicting the assumption that $`c`$ spans an extremal ray. ∎
Acknowledgments. I thank M. Hering, A. Küronya, and M. Mustaţǎ for helpful conversations related to this work, and O. Debarre for useful comments on an earlier draft of this paper. I am especially grateful to R. Lazarsfeld for suggesting this question, and for valuable advice.
|
warning/0506/cond-mat0506237.html
|
ar5iv
|
text
|
# Downward Shift of Infrared Conductivity Spectral Weight at the DDW Transition: Role of Anisotropy
## I Introduction
It has been proposed DDW that the anomalous behaviors observed in the pseudogap phase of the underdoped cuprates can be explained by assuming the existence of a new long-range ordered density-wave with $`d_{x^2y^2}`$ symmetry (DDW) Chetan1 which competes with superconducting order. As a result of this order, the excitation spectrum acquires a partial gap with the same symmetry and structure as the gap observed in the pseudogap state. The DDW theory has been applied to explain several observed anomalies in the pseudogap phase of the cuprates DDWARPES ; DDWHallAngle ; DDWSTM ; DDWSuperfluid ; DDWHallNumber ; Benfatto1 ; Benfatto2 . Attempts to directly observe it in neutron scattering are encouraging Mook .
One of the arguments against the DDW theory is that it is inconsistent with in-plane optical conductivity measurements. The assumption has been that as the temperature is reduced below the pseudogap transition temperature $`T^{}`$, and a (partial) gap opens, some of the low-energy excitations are lost and this should cause a reduction in the low-frequency optical conductivity, which is compensated by an increase in the spectral weight at higher frequencies. Such behavior is seen in some charge density-wave systems CDW , but is not observed in the cuprates PuchkovJPC ; TimuskRPP ; BontempsPRL , which might be taken as an indication that the pseudogap is associated with superconducting fluctuations, which would cause a downward motion of spectral weight (which would eventually coalesce into a zero-frequency $`\delta `$-function in the superconducting state).
These experiments find no changes in the in-plane spectral weight within their error bars PuchkovJPC ; TimuskRPP ; BontempsPRL : spectral weight lost at the low-energy end of the measured frequency range appears (perhaps unexpectedly) at very low frequencies, and hence accompanies a narrowing of the low-energy optical conductivity peak. This phenomenon was interpreted in ref. BontempsPRL, as the lack of any in-plane optical evidence for the pseudogap.
In this paper we show how the DDW theory of the pseudogap can be consistent with this observed effect. The key to this phenomenon, we believe, (as conjectured in BontempsPRL, ) is the strong anisotropy in the quasiparticle spectrum and scattering rate in the cuprates. ARPES experiments clearly show that quasiparticles in the antinodal region of the Brillouin zone ($`(k_x,k_y)(0,\pm \pi ),(\pm \pi ,0)`$), where the gap opens in the pseudogap phase, scatter much more strongly than they do in the nodal region ($`(k_x,k_y)(\pm \pi ,\pm \pi )`$), where the gap is zero. (While the transport lifetime is not the same as the quasiparticle lifetime observed in ARPES, we assume that it has a similar anisotropy, as in “hot spot” and “cold spot” theories hotspots ; coldspots .) The lost carriers, therefore, give a relatively small contribution to the conductivity. The loss of these carriers can be cancelled by a general reduction of all the scattering rates, as the temperature is reduced. As pointed out in ref. Basov02, , this may also explain the observed behavior in several classes of spin-density-wave (SDW) and CDW systems.
These ideas are borne out by a model calculation with the mean-field DDW Hamiltonian and an ansatz for the quasiparticle lifetime. We show that when the lifetime has an anisotropic form motivated by ARPES measurements, the conductivity spectral weight shifts downward, as in the cuprates. However, for an isotropic lifetime, the spectral weight shifts upward.
## II Mean-Field Hamiltonian
The mean-field Hamiltonian for the DDW state is
$$H_{\mathrm{DDW}}=\underset{𝐤,\sigma }{}[(ϵ_𝐤\mu )c_{𝐤\sigma }^{}c_{𝐤\sigma }+(i\mathrm{\Delta }_𝐤c_{𝐤\sigma }^{}c_{𝐤+𝐐\sigma }+\text{ h.c.})]$$
(1)
where $`c_𝐤`$ is the annihilation operator of an electron in a state with momentum $`𝐤`$ and spin $`\sigma `$, $`\mu `$ is the chemical potential, $`\mathrm{\Delta }_𝐤=c_{𝐤\sigma }^{}c_{𝐤+𝐐\sigma }=\mathrm{\Delta }_0(\mathrm{cos}k_x\mathrm{cos}k_y)/2`$ is the DDW order parameter and $`𝐐=(\pi ,\pi )`$ is the DDW ordering wave vector, with lattice spacing set to unity. The tight-binding band structure is given by $`ϵ_𝐤=ϵ_{1𝐤}+ϵ_{2𝐤}`$, with
$$ϵ_{1𝐤}=2t(\mathrm{cos}k_x+\mathrm{cos}k_y),ϵ_{2𝐤}=4t^{}\mathrm{cos}k_x\mathrm{cos}k_y.$$
(2)
where $`t`$ and $`t^{}`$ are nearest-neighbor and next-neighbor hopping parameters.
Introducing the two-component field operator $`\chi _{𝐤\sigma }^{}=\left(\begin{array}{cc}c_{𝐤\sigma }^{}& ic_{𝐤+𝐐\sigma }^{}\end{array}\right)`$ the mean field Hamiltonian can be written as
$$H_{DDW}=\underset{𝐤,\sigma }{}\chi _{𝐤\sigma }^{}B_𝐤\chi _{𝐤\sigma }$$
(3)
with
$$B_𝐤=\left(\begin{array}{ccc}ϵ_𝐤\mu & \mathrm{\Delta }_𝐤& \\ \mathrm{\Delta }_𝐤& ϵ_{𝐤+𝐐}\mu & \end{array}\right)$$
(4)
or
$$B_𝐤=(ϵ_{2𝐤}\mu )+ϵ_{1𝐤}\sigma ^3+\mathrm{\Delta }_𝐤\sigma ^1$$
(5)
where $`\sigma ^i`$ are the Pauli matrices and the sum is over half the original Brillouin zone (reduced Brillouin zone - RBZ), i.e. $`|k_x|+|k_y|\pi `$, .
Diagonalizing the Hamiltonian will then give the DDW quasiparticle energy bands
$$E_{1,2}=(ϵ_{2𝐤}\mu )\pm \sqrt{ϵ_{1𝐤}^2+\mathrm{\Delta }_𝐤^2}$$
(6)
as depicted in figure 1.
## III Green Function
The ”non-interacting” Nambu Green’s function can then be obtained by inverting the matrix $`B_𝐤`$ . (For now, the only relevant interactions are the electron-electron interactions which generate the DDW coupling. All other interactions including quasiparticle-quasiparticle and quasiparticle-impurity interactions will be taken into account later by assuming a non-zero self energy).
The $`2\times 2`$ Nambu Green function is defined by
$$\begin{array}{c}G_0(𝐤,t)=T\chi _𝐤(t)\chi _𝐤^{}(0)\hfill \\ \hfill =\left(\begin{array}{ccc}Tc_𝐤(t)c_𝐤^{}(0)& iTc_𝐤(t)c_{𝐤+𝐐}^{}(0)& \\ iTc_{𝐤+𝐐}(t)c_𝐤^{}(0)& Tc_{𝐤+𝐐}(t)c_{𝐤+𝐐}^{}(0)& \end{array}\right)\end{array}$$
(7)
The Fourier transform of the Green function matrix is then
$$\begin{array}{c}G_0(𝐤,\omega )=𝑑te^{i\omega t}G_0(𝐤,t)\hfill \\ \hfill =\frac{1}{(\omega +i\delta )(ϵ_{2𝐤}\mu )ϵ_{1𝐤}\sigma ^3\mathrm{\Delta }_𝐤\sigma ^1}\\ \hfill =\frac{(\omega (ϵ_{2𝐤}\mu )+ϵ_{1𝐤}\sigma ^3+\mathrm{\Delta }_𝐤\sigma ^1}{\omega (ϵ_{2𝐤}\mu )+i\delta )^2ϵ_{1𝐤}^2+\mathrm{\Delta }_𝐤^2}\end{array}$$
(8)
We now consider the effects of impurity scattering and ‘residual’ electron-electron interactions. Here, we will capture their combined effect by introducing a non-zero single-particle self-energy $`\mathrm{\Sigma }(𝐤,\omega )=\mathrm{\Sigma }_1(𝐤,\omega )+i\mathrm{\Sigma }_2(𝐤,\omega )`$, where the real and imaginary parts give the energy renormalization and quasiparticle lifetime, respectively. Neglecting the shift in the excitation energy due to the real part $`\mathrm{\Sigma }_1(𝐤,\omega )`$, the self-energy can be written in terms of the quasiparticle lifetime as
$$\mathrm{\Sigma }(𝐤,\omega )=\frac{i}{\tau (𝐤,\omega )}$$
(9)
where $`1/\tau (𝐤,\omega )`$ is the quasiparicle scattering rate (inverse lifetime). Including the self-energy, the full Green’s function $`G(𝐤,\omega )`$ will then be given according to Dyson’s equation by
$$G(𝐤,\omega )=\left(G_0(𝐤,\omega )^1\mathrm{\Sigma }(𝐤,\omega )\right)^1$$
(10)
The spectral function $`A(𝐤,\omega )`$ can then be calculated by taking the imaginary part of the Green function
$$A(𝐤,\omega )=\frac{1}{\pi }\text{Im }G(𝐤,\omega )=\frac{1}{\pi \tau }\frac{(\omega ϵ_{2𝐤}+\mu )^2+(ϵ_{1𝐤}^2+\mathrm{\Delta }_𝐤^2)+(1/\tau )^2+2(\omega ϵ_{2𝐤}+\mu )(ϵ_{1𝐤}\sigma ^3+\mathrm{\Delta }_𝐤\sigma ^1)}{[(\omega E_{1𝐤})(\omega E_{2𝐤})(1/\tau )^2]^2+(2(\omega ϵ_{2𝐤}+\mu )/\tau )^2}$$
(11)
In Figure 2, the diagonal elements of the Nambu spectral function matrix $`A(\omega ,k_x,k_y=0)`$ are plotted against $`k_x`$ and $`\omega `$. For demonstration purposes, the scattering rate here is assumed to be a constant $`1/\tau =0.02`$ eV (independent of momentum and frequency). The diagonal entries $`A_{11}`$ and $`A_{22}`$ have peaks centered mostly at energies corresponding to the upper and lower energy bands ($`E_1`$ and $`E_2`$) respectively.
## IV Optical conductivity
Having found the spectral function $`A(\omega ,𝐤)`$, the real part of the AC conductivity can now be calculated by using the Kubo formula
$$\text{Re }\sigma _{xx}(\omega )=\frac{1}{\omega }\text{Im }\mathrm{\Pi }_{xx}(i\omega _n\omega +i\delta )$$
(12)
where $`\mathrm{\Pi }(i\omega _n)`$ is the Fourier tansform of the current-current correlation function in Matsubara formalism
$$\mathrm{\Pi }_{xx}(i\omega _n)=_0^\beta 𝑑\tau e^{i\omega _n\tau }\mathrm{\Pi }_{xx}(\tau )$$
(13)
with
$$\mathrm{\Pi }_{xx}(\tau )=T_\tau j_x(\tau )j_x(0)$$
(14)
The current operator for the DDW quasiparticles can be obtained by minimally-coupling the mean-field Hamiltonian (1) to the electromagnetic field, $`𝐀`$, and differentiating once with respect to $`𝐀`$:
$$\begin{array}{c}𝐣=\underset{RBZ}{}[𝐯_{F2}(𝐤)\left(\chi _𝐤^{}\chi _𝐤\right)+𝐯_{F1}(𝐤)\left(\chi _𝐤^{}\sigma ^3\chi _𝐤\right)\hfill \\ \hfill +𝐯_\mathrm{\Delta }(𝐤)\left(\chi _𝐤^{}\sigma ^1\chi _𝐤\right)]\end{array}$$
(15)
where $`𝐯_{F1}(𝐤)=_𝐤ϵ_1(𝐤)`$ ,$`𝐯_{F2}(𝐤)=_𝐤ϵ_2(𝐤)`$ and $`𝐯_\mathrm{\Delta }(𝐤)=_𝐤\mathrm{\Delta }(𝐤)`$.
Using this form of the current operator, the current-current correlation function can be written in terms of the elements of the imaginary-time Nambu Green’s function $`𝒢_{ij}`$ (the non-interacting Green’s function $`G_{ij}`$ is now promoted to the interacting one $`𝒢_{ij}`$, to take the effect of the residual interactions into account). We will have
$$\begin{array}{c}T_\tau j(\tau )j(0)=\underset{RBZ}{}([𝐯_{F2}(𝐤)]^2\mathrm{tr}(𝒢(\tau )𝒢(\tau ))\hfill \\ \hfill +[𝐯_{F1}(𝐤)]^2\mathrm{tr}(\sigma ^3𝒢(\tau )\sigma ^3𝒢(\tau ))\\ \hfill +[𝐯_\mathrm{\Delta }(𝐤)]^2\mathrm{tr}\left(\sigma ^1𝒢(\tau )\sigma ^1𝒢(\tau )\right))\\ \hfill +(𝐯_{F1}.𝐯_{F2})\mathrm{tr}\left((\sigma ^3𝒢(\tau )+𝒢(\tau )\sigma ^3)𝒢(\tau )\right))\\ \hfill +(𝐯_{F2}.𝐯_\mathrm{\Delta })\mathrm{tr}\left((\sigma ^1𝒢(\tau )+𝒢(\tau )\sigma ^1)𝒢(\tau )\right))\\ \hfill +(𝐯_{F1}.𝐯_\mathrm{\Delta })\mathrm{tr}\left((\sigma ^1𝒢(\tau )\sigma ^3+\sigma ^3𝒢(\tau )\sigma ^1)𝒢(\tau )\right))\end{array}$$
(16)
In this equation, we have ignored vertex corrections. These are important when the scattering rate is strongly angle-dependent, as they distinguish the transport and quasiparticle lifetimes (for instance, through a $`(1\mathrm{cos}\theta )`$ factor) and also distinguishing umklapp scattering from momentum-conserving scattering. In what follows, we will assume that the replacement $`\tau \tau _{\mathrm{tr}}`$ is made (i.e. ignore vertex corrections). Referencevertex has considered the vertex correction for DDW conductivity.
Writing $`𝒢`$ in terms of the spectral function $`A(𝐤,\omega )`$, evaluating the Matsubara sum, and doing the analytic continuation $`(i\omega _n\omega +i\delta )`$, we find the optical conductivity to be Mahan
$$\begin{array}{c}\sigma (\omega )\frac{1}{\omega }\underset{RBZ}{}_{\mathrm{}}^{\mathrm{}}\frac{d\epsilon }{2\pi }[n_F(\epsilon )n_F(\epsilon +\omega )]\times \hfill \\ \hfill \{\left[𝐯_{F2}(𝐤)\right]^2[A_{11}(k,\epsilon )A_{11}(k,\epsilon +\omega )+2A_{12}(k,\epsilon )A_{12}(k,\epsilon +\omega )+A_{22}(k,\epsilon )A_{22}(k,\epsilon +\omega )]\\ \hfill +\left[𝐯_{F1}(𝐤)\right]^2\left[A_{11}(k,\epsilon )A_{11}(k,\epsilon +\omega )2A_{12}(k,\epsilon )A_{12}(k,\epsilon +\omega )+A_{22}(k,\epsilon )A_{22}(k,\epsilon +\omega )\right]\\ \hfill +\left[𝐯_\mathrm{\Delta }(𝐤)\right]^2\left[A_{22}(k,\epsilon )A_{11}(k,\epsilon +\omega )+2A_{12}(k,\epsilon )A_{12}(k,\epsilon +\omega )+A_{11}(k,\epsilon )A_{22}(k,\epsilon +\omega )\right]\\ \hfill +2𝐯_{F1}(𝐤).𝐯_{F2}(𝐤)\left[A_{11}(k,\epsilon )A_{11}(k,\epsilon +\omega )A_{22}(k,\epsilon )A_{22}(k,\epsilon +\omega )\right]\\ \hfill +2𝐯_{F2}(𝐤).𝐯_\mathrm{\Delta }(𝐤)\left[A_{12}(k,\epsilon )(A_{11}(k,\epsilon +\omega )+A_{22}(k,\epsilon +\omega ))+(A_{11}(k,\epsilon )+A_{22}(k,\epsilon ))A_{12}(k,\epsilon +\omega )\right]\\ \hfill +2𝐯_{F1}(𝐤).𝐯_\mathrm{\Delta }(𝐤)[A_{12}(k,\epsilon )(A_{11}(k,\epsilon +\omega )A_{22}(k,\epsilon +\omega ))+(A_{11}(k,\epsilon )A_{22}(k,\epsilon ))A_{12}(k,\epsilon +\omega )]\}\end{array}$$
(17)
where $`n_F(ϵ)`$ is the Fermi distribution function. For demonstration purposes, in fig. 3 the real part of the optical conductivity has been plotted against $`\omega `$ for two different temperatures (one above $`T^{}`$, depicted with a solid line, one below $`T^{}`$, depicted with a dashed-dotted line), assuming that the quasiparticle lifetime is a constant (temperature and momentum independent) and the gap is unrealistically big ($`W_0=0.25`$ eV). As expected an upward shift of the SW occurs when the gap opens. Similar calculations have also been done in Valenzuela .
## V Quasiparticle lifetime
Applying the formulas of the preceding section to the underdoped cuprates presupposes that the quasiparticle picture makes sense there. This is questionable, particularly in the anti-nodal regions, where even lowest-order perturbation theory around the DDW mean-field Hamiltonian DDWARPES predicts short lifetimes which may indicate a breakdown of quasiparticles. However, we will compute the conductivity in the quasiparticle approximation to show that, even at this level, an upward shift of spectral weight is not expected. If there are no quasiparticles at the anti-nodes, then the situation may be even better.
In order to proceed with this strategy, we need one final ingredient, an ansatz for the quasiparticle scattering rate $`1/\tau (𝐤,\omega ;T)=\gamma (𝐤,\omega ;T)`$, as a function of momentum, frequency, and temperature in the underdoped cuprates. A number of angle-resolved photoemission experiments have measured the inverse lifetime (imaginary part of the self energy), as a function of these different parameters. In these experiments, the width of the quasiparticle peaks in the energy distribution curves (EDC’s) or momentum distribution curve (MDC’s) are measured as functions of momentum, energy and temperature VallaPRL00 ; VallaScience99 ; Kaminski0404385 ; KaminskiPRL00 . While the transport lifetimes are not necessarily identical to the quasiparticle lifetimes (or, equivalently, the vertex corrections are not necessarily small), we expect that they will have a similar anisotropic behavior. For the purposes of this model calculation, we will simply take them to be the same. We emphasize that the lifetimes which we adopt below are for illustrative purposes since our main goal is to show that an upward movement of spectral weight is not a necessary concomitant of the emergence of DDW order at finite temperature. We are not making any claims here about the correctness of these lifetimes.
We take the imaginary part of the self-energy (quasiparticle scattering rate) to have the form
$$\mathrm{\Sigma }_2(𝐤,\omega ;T)=\mathrm{\Sigma }_2(\omega ;T)+\mathrm{\Gamma }(𝐤).$$
(18)
where $`\mathrm{\Sigma }_2(\omega ;T)`$ is temperature and frequency dependent with no momentum dependence and $`\mathrm{\Gamma }(𝐤)`$ is strongly momentum dependent.
Such a form has been motivated by marginal Fermi liquid phenomenology MFL ; PNAS (MFL). In this way of analyzing the data, it is assumed that the quasiparticle lifetime which comes from electron-electron scattering is independent of temperature and linear in energy for small temperatures and linear in temperature, independent of the binding energy for higher temperatures:
$$\mathrm{\Sigma }_2^{MFL}(\omega ;T)=\lambda \text{Max}(|\omega |,T).$$
(19)
In this ansatz, all of the angular dependence comes from elastic electron-electron scattering. This is a convenient form, but we will show that our results hold even for some others.
For instance, we repeat our calculations with the standard Fermi liquid (FL) quasiparticle lifetime:
$$\mathrm{\Sigma }_2^{FL}(\omega ;T)=\lambda \text{Max}(\omega ^2,T^2).$$
(20)
again assuming that the angular dependence comes from $`\mathrm{\Gamma }(𝐤)`$. In order to show that these particular forms of $`\omega `$-dependence are not playing a big role in shifts of SW, we have also tried $`\omega `$-independent forms of these scattering rates: $`\mathrm{\Sigma }_2^{TLinear}(\omega ;T)=\lambda T`$ and $`\mathrm{\Sigma }_2^{T^2}(\omega ;T)=\lambda T^2`$.
While this form of the quasiparticle lifetime does not give the correct DC conductivity, it is still useful as a check because our goal is to emphasize the role of anisotropy and show that it can lead to a downward shift of spectral weight, regardless of its detailed frequency and temperature dependence.
The quasiparticle lifetime is strongly anisotropic Kaminski0404385 : excitations in the antinodal region are more strongly scattered than the ones in the nodal region by up to a factor of 5. Hence, the assumed form (18) necessitates that $`\mathrm{\Gamma }(𝐤)`$ be a strongly anisotropic function of $`𝐤`$. We take the form:
$$\mathrm{\Gamma }_{\mathrm{Aniso}}(𝐤)=\gamma _0(1+(\mathrm{cos}k_x\mathrm{cos}k_y)^2),$$
(21)
where $`\gamma _0`$ is the scattering strength in the nodal region.
In order to check the role of anisotropy in the eventual shifting of the SW, we also check the case in which $`\mathrm{\Gamma }(𝐤)`$ is (unrealistically) isotropic:
$$\mathrm{\Gamma }_{\mathrm{Iso}}(𝐤)=\gamma _0.$$
(22)
## VI Results
In Figures 4 to 7, we have plotted the real part of the optical conductivity vs $`\omega `$, when the momentum-independent part of the lifetime is given by the four different forms listed above. In each figure the isotropic case is compared with the anisotropic one with the same set of parameters (listed in the figure captions) except for $`\gamma _0`$, which is smaller in the anisotropic case because of its extra (larger than 1) prefactor. In all cases we have $`t=0.3`$, $`t^{}=0.09`$, $`\mu =0.3`$ and $`T^{}=0.03`$.
It is clear that in figures 4a, 5a, 6a, 7a, in which the scattering rate is isotropic, there is an upward movement of spectral weight (though it is small in some cases). However, in figures 4b, 5b, 6b, 7b, in which the scattering rate is anisotropic, there is a clear downward movement of spectral weight. (Broadening the Drude peak has also been seen in Valenzuela and vertex , in which scattering has been isotropic.)
## VII Conclusions
Recent data by Santander-Syro et al BontempsPRL , has questioned the previous belief that the opening of the pseudogap in the underdoped cuprates can be detected by in-plane optical conductivity measurements. The unexpected result that as the temperature is reduced (and although a gap opens in the antinodal region of the Brillouin zone), the spectral weight is still transferred to lower frequencies, was interpreted as lack of any pseudogap signature in the optical data. In this paper we showed that this effect is consistent with the DDW theory of the pseudogap state of the underdoped cuprates.
In the four sets of graphs of the previous section it can clearly be seen that the key factor in deciding which way the SW is transferred is isotropy or anisotropy of the scattering rate. It is shown that regardless of the form of the temperature and frequency dependence of the quasiparticle lifetime (e.g. Fermi liquid or non-Fermi liquid) the SW is shifted upward for the isotropic scattering rate while it is shifted downward for the anisotropic case. (We have not tried to find a form for the quasiparticle lifetime which correctly reproduces all of the transport data, but have focussed on the role of anisotropy and have shown that similar results are obtained for several different forms of frequency and temperature dependence.)
We believe the explanation to be as follows: the pseudogap opens in the antinodal region where the carriers are more strongly scattered than the ones in the nodal region, where there is no gap. Therefore, we have lost those excitations which already gave relatively little contribution to the transport properties of the normal state. Furthermore, by reducing the temperature, the scattering rate of all the excitations is reduced. Our results clearly show that in the (more realistic) anisotropic case, the effect of lost excitations due to the gap opening can be more than canceled by a temperature-dependent reduction of the scattering rate for the rest of the excitations and hence a downward shift of the optical spectral weight. It is obvious that anisotropy is the key here, since for similar parameters for isotropic scattering rate, an upward transfer is observed. We note that similar considerations should apply to the case of 2H-TaSe<sub>2</sub>Forro .
###### Acknowledgements.
We would like to thank Sudip Chakravarty and Dmitri Basov for discussions. This work has been supported by the NSF under Grant No. DMR-0411800.
|
warning/0506/astro-ph0506161.html
|
ar5iv
|
text
|
# Uncertainties and systematics in stellar evolution models
## 1. Introduction
During the second half of last century, stellar evolution theory has allowed us to understand the Color Magnitude Diagram (CMD) of both galactic globular clusters (GGCs) and open clusters, so that now we can explain the distribution of stars in the observed CMDs in terms of the nuclear evolution of stellar structures and, thus, in terms of cluster age and chemical composition. In recent years, however, the impressive improvements achieved for both photometric and spectroscopic observations, has allowed us to collect data of an unprecedent accuracy, which provide at the same time a stringent test and a challenge for the accuracy of the models.
On the theoretical side, significant improvements have been achieved in the determination of the Equation of State (EOS) of the stellar matter, opacities, nuclear cross sections, neutrino emission rates, that are, the physical inputs needed in order to solve the equations of stellar structure. At the same time, models computed with this updated physics have been extensively tested against the latest observations.
The capability of current stellar models to account for all the evolutionary phases observed in stellar clusters is undoubtedly an exciting achievement which crowns with success the development of stellar evolutionary theories as pursued all along the second half of the last century. Following such a success, one is often tempted to use evolutionary results in an uncritical way, i.e., taking these results at their face values without accounting for theoretical uncertainties. However, theoretical uncertainties do exist, as it is clearly shown by the not negligible differences still existing among evolutionary results provided by different theoretical groups.
The discussion of these theoretical uncertainties was early addressed by Chaboyer (1995) in a pioneering paper investigating the reliability of theoretical predictions concerning H-burning structures presently evolving in GGCs (i.e. low-mass, metal-poor stars) and, in turn, on the accuracy of current predictions about GC ages. More recently, such an investigation has been extended to later phases of stellar evolution by Cassisi et al. (1998, 1999), and Castellani & Degl’Innocenti (1999). Recently, Cassisi (2004) has reviewed the issue of the main uncertainties affecting the evolutionary properties of intermediate-mass stars.
In the next sections, we will discuss in some detail the main ’ingredients’ necessary for computing stellar models and show how the residual uncertainties on these inputs affect theoretical predictions of the evolutionary properties of low-mass stars. In particular we will devote a significant attention to the analysis of the evolutionary phases corresponding to the core H-burning stage with special emphasis on the late phases of this burning process, to the shell H-burning and to both the central and shell He-burning stages (see fig. 1). For the various evolutionary phases, we will discuss what are the main inputs, adopted in the evolutionary computations, which have the largest impact on the theoretical predictions.
## 2. Stellar evolution: the ingredients
The mathematical equations describing the physical behaviour of any stellar structure are well known since long time, and a clear description of the physical meaning of each one of them can be found in several books (as, for instance, Cox & Giuli 1968 and Kippenhan & Weigart 1990).
The (accurate) numerical solution of these differential equations is no more a problem and it can be easily and quickly achieved when using modern numerical solution schemes and current generation of powerful computers. So, from the point of view of introducing a certain amount of uncertainty in the computations of stellar models, the solution of the differential equations constraining the stellar structure is not a real concern.
This notwithstanding, in order to solve these equations, boundary conditions have to be provided: the boundary condition at the stellar centre are trivial (see the discussion in Kippenhan & Weigart 1990 and Salaris & Cassisi 2005); however the same does not apply for those at the stellar surface. Let us briefly remember that ‘to provide the outer boundary conditions’ means to provide the values of temperature and pressure at the base of the stellar atmosphere: this requirement can be accomplished either by adopting an empirical relation for the thermal stratification like that provided by Krishna-Swamy (1966) or a theoretical approximation as the so-called Eddington approximation.
A more rigorous procedure is to use results from model atmosphere computations to obtain the outer boundary conditions (see Morel et al. 1994). In general model atmospheres are computed considering a plane-parallel geometry, and solving the hydrostatic equilibrium equation together with the frequency dependent (no diffusion approximation is allowed in these low-density layers) equation of radiative transport and convective transport when necessary, plus the appropriate equation of state.
In the following, we will discuss the impact of different choices about the outer boundary conditions on various theoretical predictions.
### 2.1. The physical inputs
In order to compute a stellar structure, it is fundamental to have an accurate description of the physical behaviour of the matter in the thermal conditions characteristics of the stellar interiors and atmospheres. This means that we need to know as much accurately as possible several physical inputs as:
* the opacity: the ‘radiative’ opacity is related to the mean free path of photons inside the stars and it plays a pivotal role in determining the efficiency of heat transfer via radiative processes. When the stellar matter is under conditions of partial or full electron degeneracy, electrons are able to transport energy with a large efficiency since they have a longer mean free path than in case of non degenerate electrons. In this case, the energy transport by conduction becomes quite important and the value of the conductive opacity has to be properly evaluated.
* the equation of state: the EOS of the stellar matter is another key input for the model computations; it connects pressure, density, temperature and chemical composition at each point within the star, determines the value of the adiabatic gradient (which is the temperature gradient in most of the convective region), the value of the specific heat (which appears in the expression of the gravitational energy term), and plays a crucial role in the evaluation of the extension of the convective regions.
* the nuclear cross-sections: the evaluation of the cross-sections for the various nuclear burning processes occurring in the stellar interiors is quite important in order to properly establish the energy balance in the star. Thanks to laboratory experiments, many nuclear cross-sections are nowadays known with a high accuracy. However, there are still some important nuclear processes for both the H- and the He-burning, whose nuclear rate is poorly known.
* the neutrino energy losses: a precise determination of the energy losses due to neutrino emission is also important when the star is characterized by high density and low temperature as it occurs in the interiors of Red Giant stars.
It exists a quite rich literature describing the improvements which have been achieved in this last decade concerning our knowledge of the physical inputs required for computing stellar models. Therefore, in the following, unless quite relevant for our discussion, we will not discuss in detail this issue and refer the interested reader to the exhaustive reference lists reported by Chaboyer (1995), Catelan et al. (1996), Cassisi et al. (1998, 1999, 2001), Salaris, Cassisi & Weiss (2002) and references therein.
### 2.2. The microscopic mechanisms
When computing a stellar model, some important assumptions have to be done concerning the efficiency of some microscopic mechanisms. With the expression ‘microscopic mechanisms’ we refer to all those mechanisms which, working selectively on the different chemical species, can modify the chemical stratification in the stellar interiors and/or atmosphere. These mechanisms are: atomic diffusion and radiative levitation.
* atomic diffusion: atomic diffusion is a basic physical transport mechanism driven by collisions of gas particles. Pressure, temperature and chemical abundance gradients are the driving forces behind atomic diffusion. A pressure gradient and a temperature gradient tend to push the heavier elements in the direction of increasing pressure and increasing temperature, whereas the resulting concentration gradient tends to oppose the above processes. The speed of the diffusive flow depends on the collisions with the surrounding particles. The efficiency of the different mechanisms involved in the atomic diffusion process is given in terms of atomic diffusion coefficients which have to be estimated on the basis of laboratory measurements.
* radiative levitation: it is an additional transport mechanism caused by the interaction of photons with the gas particles, which acts selectively on different atoms and ions. Since within the star a net energy flux is directed towards the surface, photons provide an upward ‘push’ to the gas particle with which they interact, effectively reducing the gravitational acceleration. Since, at the basis of this process there are the interactions of photons with gas particles, it is clear that the efficiency of radiative levitation is related to the opacity of the stellar matter, in particular to the monochromatic opacity, and increases for increasing temperature (let us remember that radiation pressure $`P_{rad}T^4`$). The evaluation of the radiative accelerations is really a thorny problems due to the need of accounting for different interaction processes between photons and chemical elements and how the momentum of photons is distributed among ions and free electrons.
Until few years ago, all these non-canonical processes were usually ignored in stellar models computations. However, helioseismology has clearly shown how important is to include atomic diffusion in the computation of the so-called Standard Solar Model (SSM), in order to obtain a good agreement between the observed and the predicted frequencies of the non-radial p-modes (see for instance Christensen-Dalsgaard, Proffitt & Thompson 1993). In the meantime, quite recent spectroscopical measurements of the iron content in low-mass, metal-poor stars in galactic globular clusters strongly point out the importance of including radiative levitation in stellar computations in order to put in better agreement empirical estimates with the predictions provided by diffusive models.
### 2.3. The macroscopic mechanisms
When computing a stellar model, one has unavoidably to account for the occurrence of mixing in the real stars. Due to current poor knowledge of how to manage the mixing processes in a stellar evolutionary code, the efficiency of convection is commonly treated by adopting some approximate theory. In this context, it has to be noticed that when treating a region where convection is stable, one has to face with two problems:
* What is the ‘right’ temperature gradient in such region?
* What is the ‘real’ extension of the convective region?
The first question is really important only when considering the outer convective regions such as the convective envelopes of cool stars. This occurrence is due to the evidence that, in the stellar interiors as a consequence of the high densities and, in turn, of the high capability of convective bubbles to transport energy, the ‘real’ temperature gradient has to be equal to the adiabatic one. This consideration does not apply when considering the outer, low-density, stellar regions, where the correct temperature gradient has to be larger than the adiabatic one: the so-called superadiabatic gradient. One of the main problem we still have in computing star models is related to the correct estimate of this superadiabatic gradient. It is important to notice that this is not an academic question since the radius and, in turn, the effective temperature of cool stars (let us say: stars with $`T_{eff}<8000K`$) is drastically affected by the choice of the superadiabatic gradient.
Almost all evolutionary computations available in literature rely on the mixing length theory (MLT; Böhm-Vitense 1958). It contains a number of free parameters, whose numerical values affect the model $`T_{\mathrm{eff}}`$; one of them is $`\alpha _{\mathrm{MLT}}`$, the ratio of the mixing length to the pressure scale height, which provides the scale length of the convective motions (increasing $`\alpha _{\mathrm{MLT}}`$ increases the model $`T_{\mathrm{eff}}`$). There exist different versions of the MLT, each one assuming different values for these parameters; however, as demonstrated by Pedersen, Vandenberg & Irwin (1990), the $`T_{\mathrm{eff}}`$ values obtained from the different formalisms can be made consistent, provided that a suitable value of $`\alpha _{\mathrm{MLT}}`$ is selected. Therefore, at least for the evaluation of $`T_{\mathrm{eff}}`$, the MLT is basically a one-parameter theory.
The value of $`\alpha _{\mathrm{MLT}}`$ is usually calibrated by reproducing the solar $`T_{\mathrm{eff}}`$, and this solar-calibrated value is then used for computing models of stars very different from the Sun (e.g. metal poor Red Giant Branch (RGB) and Main Sequence (MS) stars of various masses). We will come back on this issue in the following.
It is worth recalling that there exists also an alternative formalism for the computation of the superadiabatic gradient, which in principle does not require the calibration of any free parameter. It is the so-called Full-Spectrum-Turbulence theory (FST, see, e.g., Canuto & Mazzitelli 1991, Canuto, Goldman & Mazzitelli 1996), a MLT-like formalism with a more sophisticated expression for the convective flux, and the scale-length of the convective motion fixed a priori (at each point in a convective region, it is equal to the harmonic average between the distances from the top and the bottom convective boundaries). From a practical point of view, the FST theory contains also a free parameter which has to be fixed, even if it seems to have a physical meaning larger than that of $`\alpha _{\mathrm{MLT}}`$.
The problem of the real extension of a convective region really affects both convective core and envelope. In the canonical framework it is assumed that the border of a convective region is fixed by the condition - according to the classical Schwarzschild criterion - that the radiative gradient is equal to the adiabatic one. However, it is clear that this condition marks the point where the acceleration of the convective cells is equal to zero, so it is realistic to predict that the convective elements can move beyond, entering and, in turn, mixing the region surrounding the classical convective boundary. This process is commonly referred to as convective overshoot.
Convective core overshoot is not at all a problem for low-mass stars during the H-burning phase since the burning process occurs in a radiative region. However, the approach used for treating convection at the border of the classical convective core is important during the following core He-burning phase as discussed in the next sections. Convective envelope overshoot could be important for low-mass stars, since these structures have large convective envelope during the shell H-burning phase and the brightness of the bump along the RGB could be significantly affected by envelope overshoot (this topic will be addressed in more detail in section 4.2.).
In low-mass stars, during the core He-burning phase, the occurrence of convection-induced mixing is a relevant problem when computing stellar models along this evolutionary stage. We discuss in more detail this issue in section 5..
Near the end of the core He-burning phase, there is another process associated with mixing, that could potentially largely affect the evolutionary properties of the models: a sort of pulsating instability of convection, the so-called breathing pulse (Castellani et al. 1985), can occur, driving fresh helium into the core and so, affecting the core He-burning lifetime as well as the carbon/oxygen ratio at the center of the star. It is still under debate if this mixing instability occurs in real stars or, if it is a fictitious process occurring as a consequence of our poor treatment of convection, as for instance, of the commonly adopted assumption of instantaneous mixing.
## 3. H-burning structures: the Turn-Off
The brightness of the bluest point along the MS, the so-called Turn Off (TO) (see fig. 1), is the most important clock marking the age of the stellar clusters. It is well known that in order to use this observational feature of the Color-Magnitude diagram for estimating the cluster age, one needs to know the distance to the cluster and the chemical composition of the stars belonging to the stellar system. The impact of current uncertainties in both stellar cluster distances and chemical composition measurements on the age estimates has been extensively discussed in literature so it will not be repeated here and we refer to the interesting work by Renzini (1991). So, from now on, we will assume to know ‘perfectly’ both the distance and the chemical composition of the stellar clusters and will concentrate our discussion on the reliability and accuracy of the age - luminosity (of the TO) calibration provided by evolutionary stellar models.
It is clear that the check of the accuracy of the evolutionary models should correspondingly become a cornerstone in our attempt of obtaining accurate ages for globular clusters as well as robust results concerning the star formation history of composite stellar populations.
The main ’ingredients’ adopted in stellar models computations which affect the observational properties of stellar models at the TO and, in turn, the age - luminosity calibration are the following :
* EOS $``$ luminosity, effective temperature
* Radiative opacity $``$ luminosity, effective temperature
* Nuclear reaction rates $``$ luminosity
* Superadiabatic convection $``$ effective temperature
* Chemical abundances $``$ luminosity, effective temperature
* Atomic diffusion $``$ luminosity, effective temperature
* Treatment of the boundary conditions $``$ effective temperature
For each ingredient, we have also listed the observational property of the TO structure which is affected by a change of the corresponding ingredient. Therefore, it appears evident that some ‘inputs’ affect directly the age - luminosity relation because they modify the bolometric magnitude of the TO for a fixed age; some other inputs really can modify also (or only) the effective temperature of the TO models, so they affect the age - luminosity relation indirectly through the change induced in the bolometric correction adopted for transferring the theoretical predictions from the H-R diagram to the various observational planes. Now we discuss in some details the effect of current uncertainties in these inputs in the calibration of the age - luminosity relation.
#### The EOS:
the importance of an accurate EOS when computing SSM has been largely emphasized by all helioseismological analysis (see for instance Degl’Innocenti et al. 1997 and references therein). However, Chaboyer & Kim (1995) were the first to strongly point out the relevance of an accurate EOS for computing H-burning stellar models of low-mass, metal-poor stars due to the huge impact on the age - luminosity calibration and, in turn, on the dating of GGCs. More in detail, they have clearly shown how the proper treatment of non-ideal effects such as Coulomb interactions significantly affects the thermal properties of low-mass stars and then their core H-burning lifetime.
Chaboyer & Kim (1995) showed how the use of the OPAL EOS (Rogers 1994) - the most updated EOS available at that time - would imply a reduction of the GC age of about 1Gyr (i.e. of about 7% when compared with the ages derived by using models based on less accurate EOSs).
The OPAL EOS has been largely updated along this decade (see Rogers & Nayfonov 2002). However, the results for H-burning structures do not change significantly with respect the predictions obtained at the time of the first EOS release. It is worth emphasizing that the OPAL EOS results have been recently confirmed by independent analysis such that performed by A. Irwin (Irwin 2005, Cassisi, Salaris & Irwin 2003). One can also notice that almost all more recent library of stellar models are based on updated EOS. Therefore, we think that current residual uncertainties on the EOS for low-mass stars do not significantly affect the reliability of current age - luminosity calibrations.
#### The radiative opacity:
it is one of the most essential ingredients of the model input physics. As a general rule, increasing the radiative opacity makes dimmer stars (roughly speaking $`L1/\kappa `$), which then take longer to burn their central hydrogen. So for a given stellar mass, the TO luminosity is decreased, and the time needed to reach it is increased. So the two effects tend to balance each other, and the age - luminosity calibration is less affected. However, larger opacities favor the envelope expansion, and therefore the MS TO is anticipated. In this last decade, a big effort has been devoted to a better determination of both high- and low-temperature ($`T<10000K`$) opacities. Concerning the high-temperature opacity the largest contribution has been provided by the OPAL group (Iglesias & Rogers 1996) whose evaluations represent a sizeable improvement with respect the classical Los Alamos opacity. So the question is: how much accurate are current evaluations of radiative opacity in the high-T regime?
Recently this issue has been investigated by two independent analysis for thermal conditions and chemical compositions appropriate in the Sun: Rose (2001) considered several opacity tabulations and found that for temperatures typical of the solar core there is a standard deviation of about 5% around the average; Neuforge-Verheecke et al. (2001) performed an accurate comparison between the OPAL opacity and that provided by Magee et al. (1995) and disclosed that the mean difference between the two opacity set is of about 5%, being the OPAL opacity larger than the Magee et al.’s on almost the whole temperature range.
When considering metal-poor stars, due to the lack of heavy elements, one can expect that the opacity evaluation is simpler than for metal-rich stars and, then the estimates should be more robust. In fact, as verified by Chaboyer & Krauss (2002) the difference between the OPAL and the LEDCOP opacities in the metal-poor regime ranges from $`1`$% at the star centre to about 4% at the base of the convective envelope.
However, the existence of a good agreement between independent estimates does not represent an evidence that the predicted opacity is equal to the ‘true’ one: there is a general consensus that, at least, for conditions appropriate for the core of metal-poor stars, current uncertainty should not be larger than about 5%. For temperatures of the order of $`10^6K`$, a larger uncertainty seems to be possible: quite recently Seaton & Badnell (2004) have shown that, for temperature of this order of magnitude, a difference of the order pf $`13`$% does exist between the monochromatic opacities provided by the OPAL group with respect those provided by the Opacity Project.
As verified by Chaboyer & Krauss (2002) on the basis of an extended set of Monte Carlo simulations, an increase of about 2% in the high-T opacity would imply an increase of about 3% in the age determination based on a theoretical calibration of the relation between the cluster age and the mass of the star evolving along the Sub-giant. Since the sensitivity of the TO brightness to change in the adopted radiative opacity is lower than the age indicator considered by Chaboyer & Krauss (2002), we expect that a change of 2% in the high-T opacity leaves almost unaffected the age - luminosity calibration (see also Chaboyer 1995 and Cassisi et al. 1998).
As far as it concerns low-temperature opacities, since they affect mostly cool stars like RGB ones \- TO stars have effective temperatures large enough for not being significantly affected by different choices about low-T opacities - we postpone a discussion about the impact of current uncertainty on stellar models, to the section devoted to RGB stellar models. Here it is suffice to note that current errors on this ingredient have a negligible effect (of the order of 1%) on age estimates as verified by Chaboyer & Kim (1995).
#### Nuclear reaction rates:
the reliability of theoretical predictions about evolutionary lifetimes critically depends on the accuracy of the nuclear reaction rates since nuclear burning provides the bulk of the stellar luminosity during the main evolutionary phases. In these last years, a large effort has been devoted to increase the measurement accuracy at energies as close as possible to the Gamow peak, i.e. at the energies at which the nuclear reactions occur in the stars.
The effect on the age - luminosity calibration of current uncertainties on the rates of the nuclear reactions involved in the p-p chain has been extensively investigated by several authors (Chaboyer 1995, Chaboyer et al. 1998, Brocato, Castellani & Villante 1998). The main result was that, for a realistic estimate of the possible errors on these rates, the effect on the derived ages was almost negligible (lower than $`2`$%). The explanation of this result is simply that the nuclear processes involved in the p-p chain are really well understood so the associated uncertainty is quite small.
However, near the end of core H-burning stage, due to the lack of H, the energy supplied by the H-burning becomes insufficient and the star reacts contracting its core in order to produce the requested energy via gravitation. As a consequence, both the central temperature and density increase and, when the temperature attains a value of the order of $`1315\times 10^6K`$, the H-burning process is really governed by the more efficient CNO cycle, whose efficiency is critically depending on the reaction rate for the nuclear process $`{}_{}{}^{14}N(p,\gamma )^{15}O`$, since this is the slowest reaction in the CNO cycle.
The TO luminosity depends on the rate of this nuclear process: the larger the rate, the fainter the TO is (roughly speaking $`\mathrm{\Delta }\mathrm{log}(L_{TO}/L_{})0.015\delta CNO`$ (Brocato et al. 1998)). On the contrary, the core H-burning lifetime is marginally affected by the rate of this process, being mainly controlled by the efficiency of the p-p chain. For an exhaustive discussion on this issue, we refer to Weiss et al. (2005).
Until a couple of year ago, the rate for the $`{}_{}{}^{14}N(p,\gamma )^{15}O`$ reaction was uncertain, at least, at the level of a factor of 5. In fact, all available laboratory measurements were performed at energies well above the range of interest for astrophysical purpose and, therefore, a crude extrapolation was required (Caughlan & Fowler 1988, Angulo et al. 1999). Due to the presence of a complex resonance in the nuclear cross section at the relevant low energies, this extrapolation was really unsafe (see Angulo et al. 1999, while for a discussion of the impact of the estimated uncertainty on the age-luminosity relation we address the reader to the paper by Chaboyer et al. 1998).
Luckily enough, recently the LUNA experiment (Formicola et al. 2003) has significantly improved the low energy measurements of this reaction rate, obtaining an estimate which is about a factor of 2 lower than previous determinations. The effect on H-burning stellar models and, in turn, on the age - luminosity relation has been investigated by Imbriani et al. (2004) and Weiss et al. (2005).
The lower rate for the $`{}_{}{}^{14}N(p,\gamma )^{15}O`$ reaction leads to a brighter and hotter TO for a fixed age. The impact of this new rate on the age - luminosity relation is the following: for a fixed TO brightness the new calibration predicts systematically older cluster ages, being the difference with respect the ‘old’ calibration of the order of 0.8-0.9Gyr on average.
#### Superadiabatic convection:
as already stated, the convection in the outer layers is commonly managed by adopting the mixing length formalism in which a free parameter is present: the mixing length. Its value is usually calibrated on the Sun<sup>1</sup><sup>1</sup>1Since atomic diffusion (see the following discussion) modifies the envelope chemical stratification and, in turn, the envelope opacity, the value obtained for the solar calibrated mixing length does depend on if atomic diffusion is taken into account when computing the SSM. Needless to say that the most correct approach is to calibrate the mixing length on a diffusive SSM. (see for instance, Salaris & Cassisi 1996, and Pietrinferni et al. 2004). However, since there is no compelling reason according to which the mixing length should be the same for the Sun and metal-poor stars or constant for different evolutionary phases, it is worthwhile to investigate the impact of different choice about the mixing length calibration (see also below).
One has to bear in mind that a change in the mixing length, i.e. a change in the superadiabatic convection efficiency, alters only the stellar radius and, in turn, the effective temperature, leaving unchanged the surface luminosity. This is shown in fig. 2, where we plot two isochrones computed adopting two different values for the mixing length.
Since, the effect on the stellar radius due to a mixing length variation depends on the extension of the superadiabatic region - that is larger for stars in the mass range: $`0.7M_{}1.4M_{}`$ (less massive stars being more dense objects are almost completely adiabatic, while in more massive stars the superadiabatic region is extremely thin) and, in turn, on the total star mass, any change in the efficiency of outer convection alters the shape of the theoretical isochrones in the region around the TO (see fig. 2).
Therefore, from the point of view of the age - luminosity relation, the uncertainty in the superadiabatic convection efficiency introduces a certain amount of indetermination as a consequence of the induced change in the $`T_{eff}`$ and, then, in the adopted bolometric correction. According to Chaboyer (1995), the maximum uncertainty related to the treatment of convection in stellar models is of the order of 10%. However, one has to note that this estimate was obtained by changing the mixing length value in the range from 1 to 3. Really, a so huge variation of the mixing length seems not to be requested by current physical framework: almost all independent set of stellar models have been computed by using similar mixing length values (the spread is of the order of 0.2-0.3); in addition within a given theoretical framework there is no need to change the mixing length of almost a factor of 2 (see for instance the analysis of the mixing length calibration performed by Salaris & Cassisi 1996, in a wide metallicity range).
This issue has been recently revised by Chaboyer et al. (1998): they found (see their fig. 5) that a change of 0.1 in the mixing length causes a variation of about 1% in the globular cluster age; since they assume a realistic uncertainty of about 0.25 in the convection efficiency, this translates in an uncertainty of $`3`$% in the cluster age.
#### Diffusive processes:
since, at least, a decade, helioseismological constraints have brought to light the evidence that diffusion of helium and heavy elements must be at work in the Sun. So, it is immediate to assume that this physical process is also efficient<sup>2</sup><sup>2</sup>2Really, on theoretical grounds, one expects that atomic diffusion is more efficient in metal-poor MS stars, because in such structures the extension of the convective envelope is lower than in metal-rich objects, and it is well known that convection drastically reduces the efficiency of any diffusive process. in more metal-poor stars like those currently evolving in galactic GCs.
This notwithstanding, the evaluation of the atomic diffusion coefficient is not an easy task as a consequence of the complex physics one has to manage when analizing the various diffusive mechanisms, and moreover the range of efficiency allowed by helioseismology is still relatively large. For the Sun, Fiorentini et al. (1999) estimated an uncertainty of about 30% in the atomic diffusion coefficient. Perhaps the uncertainty is also larger for metal-poor stars due to the lack of any asteroseismological constraint. On this ground, it is not unrealistic to estimate an uncertainty of about a factor of two in the atomic diffusion efficiency. From an evolutionary point of view, the larger the atomic diffusion efficiency, the lower the cluster age estimate is (see Castellani et al. 1997 and references therein). The impact of this source of uncertainty on stellar models has been extensively investigated by Castellani & Degl’Innocenti (1999): for ages of the order of 10Gyr, to change the efficiency of diffusion within the quoted range modifies the TO brightness of about 0.16 mag, which corresponds to a variation of the cluster age of about $`0.7/+0.5`$Gyr; however for larger ages the situation is worst and for an average age of 15Gyr the error is equal to $`1.7/+1`$Gyr. So it appears evident that current uncertainty in the atomic diffusion coefficients is one of the largest source of error in the theoretical calibration of the age - luminosity relation.
However, we are now faced with an additional and, perhaps, more important problem concerning atomic diffusion: as already discussed, helioseismology strongly support SSMs accounting for atomic diffusion, but recent spectroscopical measurements (Castlho et al. 2000, Gratton et al. 2001, Ramirez et al. 2001) of the iron content are in severe disagreement with the predictions provided by diffusive models: the measured iron content does not appear to be significantly reduced with respect the abundance estimated for giant stars in the same cluster as one has to expect as a consequence of diffusion being at work. So the question is: how to reconcile these two independent evidence?
This can be achieved only accounting for two important pieces of evidence:
* according to Turcotte et al. (1998), radiative acceleration in the Sun can amount to about the 40% of gravitational acceleration, and one can expect that its value is larger in more metal-poor, MS stars (see above);
* there are some evidence according to which a slow mixing process (turbulence?) below the solar convective envelope could help in explaining better the observed Be and Li abundances (Richard et al. 1996) and improve the agreement between the predicted sound speed profile and that derived from helioseismological data (Brun et al. 1999).
Both these evidence, coupled with helioseismological analysis, clearly support the computations of MS stellar models accounting simultaneously for atomic diffusion, radiative levitation and some sort of extra-mixing. This new generation of models has been recently provided by Richard et al. (2002) and Vandenberg et al. (2002): the main outcome is that these models are able to reconcile helioseismology with the recent spectroscopical measurements of the iron in GGCs. In fact, these models predict that, at odds, with predictions provided by models accounting only for atomic diffusion, the surface abundance of iron (and also of other heavy elements) is depleted with a quite lower efficiency and it can also become overabundant with respect the initial value as a consequence of radiative levitation which pushes iron from the interior toward the stellar surface (see fig. 8 in the paper of Richard et al. 2002).
Concerning the age - luminosity calibration, it is worth noticing that it is not significantly affected by the inclusion of radiative levitation in stellar models computations: models accounting for both atomic diffusion and radiative levitation lead to a reduction of the order of 10% in the GGC age at a given TO brightness, i.e. more or less the same reduction which is obtained when accounting only for a (standard) efficiency of microscopic diffusion.
#### The treatment of boundary conditions:
the effect of adopting different choices about the outer boundary conditions on the age - luminosity calibration has been extensively investigated by Chaboyer (1995) and Chaboyer et al. (1998). The main result was that the calibration is only marginally affected by the adopted outer boundary conditions.
#### The chemical abundances:
the evolutionary properties of stars strongly depend on the initial chemical abundances, i.e. on the initial He content (Y = abundance by mass of helium) and heavy elements abundance ($`Z`$ = metallicity = abundance by mass of all elements heavier than helium; in the spectroscopical notation it is indicated as $`[M/H]`$). So, the age - luminosity relation depends on both $`Y`$ and $`[M/H]`$: the typical dependences (Renzini 1997) are $`\mathrm{log}t_9/Y0.4`$ and $`\mathrm{log}t_9/[M/H]0.1`$, where $`t_9`$ is the cluster age in billion of years.
The initial He content of the old, metal-poor galactic GCs is well known (see Cassisi, Salaris & Irwin 2003 and references therein) and it has to be in the range Y=0.23 - 0.25. So, assuming an uncertainty of about 0.02 in the initial He abundance, the previous relation indicates that this uncertainty gives a negligible 2% error in age. The metal content of the best studied clusters is uncertain by perhaps 0.2-0.3 dex - most of it being systematic, which translates into a $`9`$% uncertainty in age. However, when discussing the uncertainty associated to the adopted metallicity, one has also to pay attention to the ‘composition’ of the metallicity, i.e. to the distribution of the various heavy elements in the mixture. In particular, there are clear indications that $`\alpha `$elements (such as O, Ne, Mg, Si, S, Ar, Ca and Ti) are enhanced in metal-poor stellar systems with respect to the Sun (i.e. $`[\alpha /Fe]>0`$).
The effect of an $`\alpha `$enhanced mixture on the evolutionary properties of stellar models has been extensively investigated in literature and we refer to Salaris, Chieffi & Straniero (1993) and Vandenberg et al. (2000) and references therein: as a general rule, at a given iron content, an increase of the $`\alpha `$elements abundance makes the evolutionary tracks fainter and cooler. As a consequence, for a fixed TO brightness, the cluster age decreases when $`[\alpha /Fe]`$ increases: a more accurate statement is that, at a fixed TO luminosity, the cluster age is reduced by $``$7% ($`1`$Gyr) for each 0.3 dex increase in the $`[\alpha /Fe]`$ value (see fig. 4 in Vandenberg, Bolte & Stetson 1996). An accurate analysis of the stellar models discloses that almost the 60% of this variation in the age is due to the change in the radiative opacity associated with the modification of the heavy elements mixture, while the remaining difference is provided by the change in the efficiency of the CNO cycle related to the increased abundance of O in the $`\alpha `$enhanced mixture.
These considerations clearly suggest that if we would know the exact $`\alpha `$enhanced mixture of a stellar system we could compute stellar models for that mixture once the appropriate $`\alpha `$enhanced radiative opacities are provided and the burning network is correspondingly updated. On practice, this is almost impossible: 1) it is not possible to compute extended set of stellar models for any specified heavy elements distribution, 2) radiative opacity tabulations for any $`\alpha `$enhanced distribution are difficult to be provided (mostly in the low-temperature regime). However, there is the possibility to overcome this problem. In fact, it has been shown by Salaris et al. (1993) that isochrones for enhanced $`\alpha `$element abundances are well mimicked by those for a scaled-solar mixture, simply by requiring the total abundance of heavy elements to be the same: this is the so-called ‘rescaling’ approximation. This topic has been recently reanalyzed by Vandenberg et al. (2000, but see also Vandenberg & Irwin 1997) which have demonstrated that the rescaling approach is quite reliable for metallicity of the order of $`[Fe/H]0.8`$ ($`Z0.002`$) or lower. For larger metallicity, it is no longer correct to rely on this assumption and $`\alpha `$enhanced stellar models have to be used when comparing theory with observations.
Concerning the heavy elements distribution, there is an additional possible source of uncertainty: recent analyses of spectroscopical data based on 3-D hydrodynamic atmospheric models (Asplund et al. 2004) suggest that the heavy elements distribution in the Sun is significantly different with respect previous estimates (Grevesse & Sauval 1998). More in detail, the abundance of oxygen and of other heavy elements has been drastically reduced by these new measurements. As a consequence, the metal over hydrogen ratio $`(Z/X)_{}`$ has been significantly changed from $`(Z/X)_{}=0.0230`$ to 0.0165, so the Sun’s metallicity has been drastically reduced by a factor of $`1.4`$. In our belief, these new measurements have to be confirmed by other independent and accurate analyses<sup>3</sup><sup>3</sup>3It is important to note that the new estimates of the solar metallicity put the SSMs in severe disagreement with helioseismological constraints (Basu & Antia 2004, Bahcall, Serenelli & Pinsonneault 2004). However, a possible solution for this problem has been suggested by Seaton & Badnell 2004, through an increase of the radiative opacity at the boundary of the solar convective envelope.. However, it is interesting to analyze what is their impact on the evolutionary scenario.
In fig. 3, we show the evolutionary track of a $`0.8M_{}`$ computed adopting different assumptions about the distribution of heavy elements in the mixture: the effect of adopting the new Asplund et al’s mixture is quite negligible<sup>4</sup><sup>4</sup>4One has to notice that the effect would be slightly larger - but always very small - if we would consider a more massive stars and/or a higher global metallicity.; so one can expect that also the effect on the age - luminosity calibration is irrelevant. This has been demonstrated via accurate evolutionary computations by Degl’Innocenti et al. (2005).
Although, the effect of the new solar heavy elements distribution on the theoretical age - luminosity relations is quite negligible, the new estimate of the solar metallicity could potentially have a huge impact on the GGC age scale: let us assume as a first order approximation that the spectroscopical measurement of the metallicity of the stellar systems is not affected by the use of these updated set of 3-D model atmospheres (but see the preliminary analysis of Asplund 2004). If so, the value of $`[M/H]`$ for a cluster remains unchanged. Since the relation connecting the abundance by mass of heavy elements ($`Z`$) to the global metallicity in the spectroscopical notation ($`[M/H]`$) implies the use of the solar metallicity $`Z_{}`$: $`ZZ_{}\times 10^{[M/H]}`$. Simply due to the change in the value of $`Z_{}`$, now when comparing the theoretical framework with the cluster observations we must use a metallicity $`Z_{new}`$ that is equal to $`0.65Z_{old}`$ (being $`Z_{old}`$ the metallicity adopted when accounting for the ‘old’ solar heavy elements distribution). This occurrence implies that for a fixed TO brightness, the cluster age is increased of about 0.7Gyr.
This notwithstanding, we think that the real problem is another one: if the use of these new generation of 3-D model atmospheres has to drastically modified our knowledge of the solar chemical composition, one should expect that their use affects also the determination of the metallicity of metal-poor clusters. So, the main question is: what is - nowadays - the correct metallicity scale for stellar clusters?
### 3.1. The age - Turn-Off brightness calibration: the state of art
In the previous sections, we have discussed the main sources of uncertainty in the theoretical calibration of the age - TO brightness relation. However, in order to have an idea of the level of confidence in this important age indicator, we show in fig. 4 the TO brightness - age calibrations provided by the most updated set of evolutionary models presently available. It is worth noticing that all the theoretical predictions, but the one provided by the Yale group<sup>5</sup><sup>5</sup>5It could be possible that the mismatch between the Yale results and the other ones is simply due to the evidence that in the original files where the isochrones are listed, only a few number of lines are reported and this makes a problem to exactly define the TO location. This occurrence is more evident when considering metal-poor isochrones, whose morphology in the TO region is very much ‘vertical’., are in very good agreement. In fact, at a given TO brightness, the difference in the estimated age is of the order of 1 Gyr or lower. It is also comforting that this difference among the various calibrations can be almost completely explained by accounting for the different choices about the initial He content and the physical inputs.
## 4. H-burning structures: the Red Giant Branch
The RGB is one of the most prominent and well populated features in the CMD of stellar populations older than about $`1.52`$ Gyr.
Since RGB stars are cool, reach high luminosities during their evolution, and their evolutionary timescales are relatively long, they provide a major contribution to the integrated bolometric magnitude and to integrated colors and spectra at wavelengths larger than about 900 nm of old distant, unresolved stellar populations (e.g. Renzini & Fusi-Pecci 1988; Worthey 1994). A correct theoretical prediction of the RGB spectral properties and colors is thus of paramount importance for interpreting observations of distant stellar clusters and galaxies using population synthesis methods, but also for determining the ages of resolved stellar systems by means of isochron fitting techniques.
Both the RGB location and slope in the CMD are strongly sensitive to the metallicity, and for this reason, they are widely used as metallicity indicators.
The $`I`$-band brightness of the tip of the RGB (TRGB) provides a robust standard candle, largely independent of the stellar age and initial chemical composition, which can allow to obtain reliable distances out to about 10 Mpc using $`HST`$ observations (e.g., Lee, Freedman & Madore 1993). Due to the lingering uncertainties on the empirical determination of the TRGB brightness zero point, RGB models provide an independent calibration of this important standard candle (Salaris & Cassisi 1997, 1998). Theoretical predictions about the structural properties of RGB stars at the Tip of the RGB play a fundamental role in determining the main evolutionary properties of their progeny: the core He-burning stars during the Horizontal Branch (HB) evolutionary phase. In particular, HB luminosities (like the TRGB ones) are mostly determined by the value of the electron degenerate He-core mass ($`M_{core}^{He}`$) at the end of the RGB evolution.
Predicted evolutionary timescales along the RGB phase play also a fundamental role in the determination of the initial He abundance of globular cluster stars through the R parameter (number ratio between HB stars and RGB stars brighter than the HB at the RR Lyrae instability strip level; see, e.g. Iben 1968a, Salaris et al. 2004 and references therein), while an accurate modeling of the mixing mechanisms efficient in the RGB stars is necessary to correctly interpret spectroscopic observations of their surface chemical abundance patterns.
The possibility to apply RGB stellar models to fundamental astrophysical problems crucially rely on the ability of theory to predict correctly:
– the CMD location (in $`T_{\mathrm{eff}}`$ and color) and extension (in brightness) of the RGB as a function of the initial chemical composition and age;
– the evolutionary timescales (hence the relative numbers of stars at different luminosities) all along the RGB;
– the physical and chemical structure of RGB stars, as well as their evolution with time.
A detailed analysis of the existing uncertainties in theoretical RGB models, and of the level of confidence in their predictions has been performed by Salaris, Cassisi & Weiss (2002). In the following, we will briefly review the main observational properties of the RGB such as its location and slope, the bump of the luminosity function (LF) and the brightness of the Tip; discussing in some detail the main sources of uncertainty in the corresponding theoretical predictions as well as the level of agreement currently existing between theory and observations.
### 4.1. The location and the slope of the RGB
The main parameters affecting the RGB location and slope are: the EOS, the low-temperature opacity, the efficiency of superadiabatic convection, the choice about the outer boundary conditions and the chemical abundances.
#### The EOS:
Until a couple of years ago, the best available EOS was probably the OPAL one (see previous discussion). However, its range of validity does not cover the electron degenerate cores of RGB stars and their cooler, most external layers, below 5000 K. RGB models computed with the OPAL EOS must employ some other EOS to cover the most external and internal stellar regions. As a consequence, it was a common procedure to ‘mix’ together EOSs provided by different authors in order to have EOS tables suitable for the whole range of thermal conditions encountered by low-mass stars from the H-burning stage to the more advanced evolutionary phase. However, there are some notable exception about this as in the case of the stellar models computed by Vandenberg et al. (2000), Cassisi et al. (2003) and Pietrinferni et al. (2004). In particular, in the case of the models presented by Cassisi et al. (2003) and Pietrinferni et al. (2004), we take advantage by the use of the updated EOS computed by A. Irwin which consistently allows the computation of stellar models in both the H- and He-burning phases.
Till now, no detailed study exists highlighting the effect of the various EOS choices on the evolution and properties of RGB stars. In fig. 5, the RGB of a $`1M_{}`$ stellar structure computed adopting different assumptions about the EOS is shown. One can notice that the models based on the OPAL EOS and the EOS by A. Irwin are in very good agreement (this comparison is meaningful only for $`T_{eff}`$ larger than about 4500K for the reason discussed before); there is a significant change in the RGB slope with respect the model based on the Straniero (1988) EOS supplemented at the lower temperature by a Saha EOS. On average, there is a difference of about 100K between RGB models based of the two different EOSs.
#### The low-temperature opacity:
as shown by Salaris et al (1993), it is the low-$`T`$ opacities which mainly determine the $`T_{\mathrm{eff}}`$ location of theoretical RGB models, while the high-$`T`$ ones - in particular those for temperature around $`10^6K`$ \- enter in the determination of the mass extension of the convective envelope.
Current generations of stellar models employ mainly the low-$`T`$ opacity calculations by Alexander & Ferguson (1994) – and in some cases the Kurucz (1992) ones – which are the most up-to-date computations suitable for stellar modelling, spanning a large range of initial chemical compositions. The main difference between these two sets of data is the treatment of molecular absorption, most notably the fact that Alexander & Ferguson (1994) include the effect of the $`\mathrm{H}_2\mathrm{O}`$ molecule. This last set of opacity accounts also for the presence of grains. These low-$`T`$ radiative opacity tabulations represent a remarkable improvement with respect the old evaluations provided by Cox & Stewart (1970) as far as it concerns the treatment of molecules and grains. Although significant improvements are still possible as a consequence of a better treatment of the different molecular opacity sources, we do not expect dramatic changes in the temperature regime where the contribution of atoms and molecules dominate. Huge variation can be foreseen in the regime ($`T<2000K`$) where grains dominates the interaction between radiation and matter. These considerations appear fully supported by the recent reanalysis of the low-$`T`$ opacities performed by Ferguson et al. (2005).
Salaris & Cassisi (1996) have compared, at different initial metallicities, stellar models produced with these two sets of opacities (as well as with the less used Neuforge 1993 ones, which provide results almost undistinguishable from models computed with Kurucz 1992 data), showing that a very good agreement exists when $`T_{\mathrm{eff}}`$ is larger than $``$4000 K as shown in fig. 6. As soon as the RGB $`T_{\mathrm{eff}}`$ goes below this limit (when the models approach the TRGB and/or their initial metallicity is increased), Alexander & Ferguson (1994) opacities produce progressively cooler models (differences reaching values of the order of 100 K or more), due to the effect of the $`\mathrm{H}_2\mathrm{O}`$ molecule which contributes substantially to the opacity in this temperature range (see the right panel in fig. 6).
#### The outer boundary conditions:
the procedure commonly used in the current generation of stellar models is the integration of the atmosphere by using a functional (semi-empirical or theoretical) relation between the temperature and the optical depth ($`T(\tau )`$). Recent studies of the effect of using boundary conditions from model atmospheres are in V00 and Montalban et al. (2001). In fig. 7 it is shown the effects on RGB stellar models of different $`T(\tau )`$ relations, namely, the Krishna-Swamy (1966) solar T$`(\tau )`$ relationship, and the gray one. One notices that RGBs computed with a gray T$`(\tau )`$ are systematically hotter by $``$100 K. In the same Fig. 7, we show also a RGB computed using boundary conditions from the Kurucz (1992) model atmospheres, taken at $`\tau `$=10. The three displayed RGBs, for consistency, have been computed by employing the same low-T opacities, namely the ones provided by Kurucz (1992), in order to be homogeneous with the model atmospheres. The model atmosphere RGB shows a slightly different slope, crossing over the evolutionary track of the models computed with the Krishna-Swamy (1966) solar T$`(\tau )`$, but the difference with respect to the latter stays always within $`\pm `$50 K.
Even if it is, in principle, more rigorous the use of boundary conditions provided by model atmospheres, one has also to bear in mind that the convection treatment in the adopted model atmospheres (Montalban et al. 2001) is usually not the same as in the underlying stellar models (i.e., a different mixing length formalism and a different value for the scale height of the convective motion is used).
#### The chemical composition:
as far as it concerns the helium abundance, the evolutionary properties of RGB stars, at least concerning their effective temperatures, are not strongly affected by different assumptions on the initial He content. This occurrence is due to the combination of two different reasons: 1) stellar matter opacity does not strongly depend on the He abundance, 2) the initial He abundance for old, stellar systems such as GGCs is well constrained and variations larger that $`0.020.03`$ are unrealistic. On the contrary, the abundance of heavy elements is one of the parameters which most affects the RGB morphology: any increase of $`Z`$ produces a larger envelope opacity and, in turn, a more extended envelope convection zone and a cooler RGB. The strong dependence of the RGB effective temperature on the metallicity makes the RGB one of the most important metallicity indicators for stellar systems. An important issue is the dependence of the shape and location of the RGB on the distribution of the metals: different heavy elements have different ionization potentials, and provide different contribution to the envelope opacity. The abundance of low ionization potential elements such as Mg, Si, S, Ca, Ti and Fe strongly influences the RGB effective temperature, through their direct contribution to the opacity due to the formation of molecules such as TiO which strongly affects the stellar spectra at effective temperatures lower than $`50006000K`$, and through the electrons released when ionization occurs, which affect the envelope opacity via the formation of the H<sup>-</sup> ion $``$ one of the most important opacity sources in RGB structures. As an example, a change of the heavy elements mixture from a scaled solar one to an $`\alpha `$-element enhanced distribution with the same iron content, produces a larger envelope opacity and the RGB becomes cooler and less steep: the change in the slope being due to the increasing contribution of molecules to the envelope opacity when the stellar effective temperature decreases along the RGB.
#### The treatment of superadiabatic convection:
in section 3. we already noticed that the value of $`\alpha _{\mathrm{MLT}}`$ is usually calibrated by reproducing the solar $`T_{\mathrm{eff}}`$, and this solar-calibrated value is then used for stellar models of different masses and along different evolutionary phases, including the RGB one. The adopted procedure guarantees that the models always predict correctly the $`T_{\mathrm{eff}}`$ of at least solar type stars. However, the RGB location is much more sensitive to the value of $`\alpha _{\mathrm{MLT}}`$ than the main sequence. This is due to the evidence that along the RGB the extension (in radius) of the superadiabatic layers - as a consequence of the much more expanded configuration achieved by the star - is quite larger when compared with the MS evolutionary phase. Therefore, it is important to verify that a solar $`\alpha _{\mathrm{MLT}}`$ is always suitable also for RGB stars of various metallicities.
An independent way of calibrating $`\alpha _{\mathrm{MLT}}`$ for RGB stars is to compare empirically determined RGB $`T_{\mathrm{eff}}`$ values for galactic GCs with theoretical models of the appropriate chemical composition (see also Salaris & Cassisi 1996, Vandenberg, Stetson & Bolte 1996 and references therein). In fig. 8, as an example taken from the literature, we show a comparison between the $`T_{\mathrm{eff}}`$ from Frogel, Persson & Cohen (1983) for a sample of GCs and the $`\alpha `$-enhanced models by SW98. For a detailed discussion of how current empirical uncertainties on the GGCs distance scale, metallicity scale and RGB temperature scale affects the comparison shown in fig. 8, we refer to Salaris et al. (2002). The results shown in this figure (recently confirmed also by Vandenberg et al. 2000, by using their own updated set of RGB stellar models) seem to suggest that the solar $`\alpha _{\mathrm{MLT}}`$ value is a priori adequate also for RGB stars (but see also the discussion in Salaris et al. 2002).
This notwithstanding, a source of concern about an a priori assumption of a solar $`\alpha _{\mathrm{MLT}}`$ for RGB computations comes from the fact that recent models from various authors, all using a suitably calibrated solar value of $`\alpha _{\mathrm{MLT}}`$, do not show the same RGB temperatures. This means that – for a fixed RGB temperature scale – the calibration of $`\alpha _{\mathrm{MLT}}`$ on the empirical $`T_{\mathrm{eff}}`$ values would not provide always the solar value. Figure 9 displays several isochrones produced by different groups (see labels and figure caption), all computed with the same initial chemical composition, same opacities, and the appropriate solar calibrated values of $`\alpha _{\mathrm{MLT}}`$: the Vandenberg et al. (2000) and Salaris & Weiss (1998) models are identical, the Padua ones (Girardi et al. 2000) are systematically hotter by $``$200 K, while the $`Y^2`$ ones (Yi et al. 2001) have a different shape. This comparison shows clearly that if one set of MLT solar calibrated RGBs can reproduce a set of empirical RGB temperatures, the others cannot, and therefore in some case a solar calibrated $`\alpha _{\mathrm{MLT}}`$ value may not be adequate. The reason for these discrepancies must be due to some difference in the input physics, like the EOS and/or the boundary conditions, which is not compensated by the solar recalibration of $`\alpha _{\mathrm{MLT}}`$.
To illustrate this point in more detail, we show in fig. 10 two evolutionary tracks for a $`1M_{}`$ stellar model with solar chemical composition. The only difference between them is the treatment of the boundary conditions. Two different T($`\tau `$) relationships, namely gray and Krishna-Swamy (1966), have been adopted. The value of $`\alpha _{\mathrm{MLT}}`$ for the two models has been calibrated in each case, in order to reproduce the Sun, and in fact the two tracks completely overlap along the main sequence, but the RGBs show a difference of the order of 100 K.
This occurrence clearly points out the fact that one cannot expect the same RGB $`T_{\mathrm{eff}}`$ from solar calibrated models not employing exactly the same input physics. The obvious conclusion is that it is always necessary to compare RGB models with observations to ensure the proper calibration of $`\alpha _{\mathrm{MLT}}`$ for RGB stars.
### 4.2. The bump of the RGB luminosity function
The RGB luminosity function (LF), i.e. the number of stars per brightness bin among the RGB as a function of the brightness itself, of GGCs is an important tool to test the chemical stratification inside the stellar envelopes (Renzini & Fusi Pecci 1988). The most interesting feature of the RGB LF is the occurrence of a local maximum in the luminosity distribution of RGB stars, which appears as a bump in the differential LF, and as a change in the slope of the cumulative LF. This feature is caused by the sudden increase of H-abundance left over by the surface convection upon reaching its maximum inward extension at the base of the RGB (*first dredge up*) (see Thomas 1967 and Iben 1968b). When the advancing H-burning shell encounters this discontinuity, its efficiency is affected (sudden increase of the available fuel), causing a temporary drop of the surface luminosity. After some time the thermal equilibrium is restored and the surface luminosity starts to increase again. As a consequence, the stars cross the same luminosity interval three times, and this occurrence shows up as a characteristic peak in the differential LF of RGB stars. Moreover, since the H-profile before and after the discontinuity is different, the rate of advance of the H-burning shell changes when the discontinuity is crossed, thus causing a change in the slope of the cumulative LF.
The brightness of the RGB bump is therefore related to the location of this H-abundance discontinuity, in the sense that the deeper the chemical discontinuity is located, the fainter is the bump luminosity. As a consequence, any physical inputs and/or numerical assumption adopted in the computations, which affects the maximum extension of the convective envelope at the *first dredge up*, strongly affects the bump brightness. A detailed analysis of the impact of different physical inputs on the predicted RGB bump luminosity can be found in Cassisi & Salaris (1997) and Cassisi, Salaris & Degl’Innocenti (1997) and it will not be repeated. However, it is worth noting that a comparison between the predicted bump luminosity and the observations allows a direct check of how well theoretical models for RGB stars predict the extension of convective regions in the stellar envelope and, then provide a plain evidence of the reliability of current evolutionary framework (Valenti, Ferraro & Origlia 2004).
In this context it is worth noting that, for each fixed global metallicity, the theoretical predictions about the Bump luminosity provided by Bergbush & Vandenberg (2001) are in fine agreement - within $`0.05`$ mag -, with the values given by Cassisi & Salaris (1997). This is a plain evidence of the fact that current, updated canonical stellar models do agree to a significant level about this relevant evolutionary feature.
However, when comparing theory with observations, one needs a preliminary estimate of both the cluster metallicity and distance. Current uncertainty in the GGC metallicity scale strongly reduces our capability to constrain the plausibility of the theoretical framework (see the discussion in Bergbusch & Vandenberg 2001), and for such reason, it has became a common procedure to use simultaneously all available metallicity scales (see Riello et al. 2003). Another critical issue is related to the need of knowing the cluster distance, whose accuracy could strongly hamper the possibility of a meaningful comparison between theory and observations. Following the early prescription provided by Fusi Pecci et al. (1990), the observed magnitude difference between the RGB bump and the HB at the RR Lyrae instability strip ($`\mathrm{\Delta }V_{\mathrm{HB}}^{\mathrm{bump}}`$) is usually adopted in order to test the theoretical predictions for the bump brightness. This quantity presents several advantages from the observational point of view (see Fusi Pecci et al. 1990, and Salaris et al. 2002) and it is empirically well-defined because it does not depend on a previous knowledge of the cluster distance and reddening. However, on the theoretical side, one should keep in mind that such comparison requires the use of a theoretical prediction about the Horizontal Branch brightness which is a parameter still affected by a significant uncertainty (see section 5.). Nevertheless, empirical estimates about the $`\mathrm{\Delta }V_{\mathrm{HB}}^{\mathrm{bump}}`$parameter have been extensively compared with theoretical predictions (Riello et al. 2003 and references therein). Figure 11 shows the results of the comparison performed by Riello et al. (2003): even though a qualitative agreement between theory and observations of $`\mathrm{\Delta }V_{\mathrm{HB}}^{\mathrm{bump}}`$ does exist, a more definitive assessment of the confidence level appears clearly hampered by the not negligible uncertainties still affecting both the cluster \[Fe/H\] and \[$`\alpha `$/Fe\] estimates. In conclusion, it is realistic to consider that due to lingering uncertainties on the (theoretically determined!) HB brightness and the GGCs metallicity scale, there is the possibility of a discrepancy between theory and observation about the $`\mathrm{\Delta }V_{\mathrm{HB}}^{\mathrm{bump}}`$parameter at the level of $`0.20`$ mag.
Before concluding this section, we wish to notice that the RGB LF bump provides other important constraints besides brightness for checking the accuracy of theoretical RGB models. More in detail, both the shape and the location of the bump along the RGB LF can be used for investigating on how ‘steep’ is the H-discontinuity left over by envelope convection at the *first dredge up*. So these features appear, potentially, a useful tool for investigating on the efficiency of non-canonical mixing at the border of the convective envelope (Cassisi, Salaris & Bono 2002) able to partially smooth the chemical discontinuity. In addition, since the evolutionary rate along the RGB is strongly affected by any change in the chemical profile, it is clear that the star counts in the bump region can provide reliable information about the size of the jump in the H profile left over by envelope convection after the *first dredge up*. This issue,as well as the level of agreement between theory and observations, has been investigated by Bono et al. (2001) and Riello et al. (2003).
### 4.3. Star counts along the RGB
The number of stars in any given bin of the RGB LF is determined by the local evolutionary rate so the comparison between empirical and theoretical RGB LF represents a key test for the accuracy of the predicted RGB timescales (see Renzini & Fusi Pecci 1988). In addition, there are many more reasons for which to investigate the RGB star counts is quite important, for instance: i) being the RGB stars among the brightest objects in a galaxy, their number has a strong influence on the integrated properties of the galactic stellar population; ii) the number ratio between RGB and stars along the Asymptotic Giant Branch (AGB) can be used to constrain the Star Formation History of a galaxy (Greggio 2000).
A recent, investigation of the accuracy of theoretical RGB LFs has been performed by Zoccali & Piotto (2000) by adopting a large database of GGC RGB LFs. The main outcome of their analysis was the evidence of, on average, a good agreement, on the whole explored metallicity range, between observations and the theoretical predictions, available at that time. However, more recently, Gallart, Zoccali & Aparicio (2005) have reanalyzed this issue and have noticed that for a fixed number of MS stars, the number of RGB stars as predicted by different sets of evolutionary models are not in good agreement: in the smooth part of the RGB LF the different models show differences as large as 0.15 dex, that correspond to a factor of $`1.4`$ of difference in the number counts. This result is shown in fig. 12, where LFs from different authors and for various metallicities are plotted.
### 4.4. The Tip of the RGB
As a consequence of the H-burning occurring in the shell, the mass size of the He core ($`M_{core}^{He}`$) grows. When $`M_{core}^{He}`$ reaches about 0.50 $`M_{}`$ (the precise value depends weakly on the total mass of the star for structures less massive than about $`1.2M_{}`$, i.e. older than about 4-5Gyr, being more sensitive to the initial chemical composition), He-ignition occurs in the electron degenerate core. This process is the so called He-flash, that terminates the RGB phase by removing the electron degeneracy in the core and, driving the star onto its Zero Age Horizontal Branch (ZAHB) location, that marks the start of quiescent central He-burning plus shell H-burning. The brightest point along the RGB, that marks the He ignition through the He flash is the so-called Tip of the RGB (TRGB).
The observational and evolutionary properties of RGB stars at the TRGB play a pivotal role in current stellar astrophysical research. The reasons are manifold: i) the mass size of the He core at the He flash fixes not only the TRGB brightness but also the luminosity of the Horizontal Branch, ii) the TRGB brightness is one of the most important primary distance indicators.
More in detail, the reasons which make the TRGB brightness a quite suitable standard candle are the following: the TRGB luminosity that is a strong function of the He core mass at the He flash, is weakly dependent on the stellar mass, and therefore on the cluster age over a wide age interval. This is due to the already mentioned evidence that the value of $`M_{core}^{He}`$ at the He-flash is fairly constant over large part of the low-mass star range. However, $`M_{core}^{He}`$ decreases for increasing metallicity, while the TRGB brightness increases due to the increased efficiency of the H-shell, which compensates for the reduced core mass<sup>6</sup><sup>6</sup>6We recall that the brightness of the subsequent ZAHB phase follows the behaviour of $`M_{core}^{He}`$, decreasing for increasing metallicity..
Luckily enough, the value of $`\mathrm{M}_\mathrm{I}^{\mathrm{TRGB}}`$ \- the I-cousin band magnitude of the TRGB - appears to be very weakly sensitive to the heavy element abundance (Lee, Freedman & Madore 1993 and Salaris & Cassisi 1997): for $`[M/H]`$ ranging between $``$2.0 and $``$0.6, $`\mathrm{M}_\mathrm{I}^{\mathrm{TRGB}}`$ changes by less than 0.1 mag. This lucky occurrence stands from the evidence that $`\mathrm{M}_{\mathrm{bol}}^{\mathrm{TRGB}}`$ is proportional to $``$0.18\[M/H\], while $`\mathrm{BC}_\mathrm{I}`$ is proportional to $`0.14[M/H]`$. Therefore, the slope of the $`\mathrm{BC}_\mathrm{I}[\mathrm{M}/\mathrm{H}]`$ relationship is quite similar to the slope of the $`\mathrm{M}_{\mathrm{bol}}^{\mathrm{TRGB}}[\mathrm{M}/\mathrm{H}]`$ relationship, and since $`\mathrm{M}_\mathrm{I}^{\mathrm{TRGB}}`$=$`\mathrm{M}_{\mathrm{bol}}^{\mathrm{TRGB}}\mathrm{BC}_\mathrm{I}`$, it results that $`\mathrm{M}_\mathrm{I}^{\mathrm{TRGB}}`$ is almost independent of the stellar metal content.
As far as it concerns the uncertainties affecting theoretical predictions about the TRGB brightness, it is clear that, being the TRGB brightness fixed by the He core mass, any uncertainty affecting the predictions about the value of $`M_{core}^{He}`$ immediately translates in an error on $`\mathrm{M}_{\mathrm{bol}}^{\mathrm{TRGB}}`$. An exhaustive analysis of the physical parameters which affects the estimate of $`M_{core}^{He}`$ provided by stellar models has been provided by Salaris et al. (2002, but see also Castellani & Degl’Innocenti 1999): the main result of these analyses was that the physical inputs which have the largest impact in the estimate of $`M_{core}^{He}`$ are the efficiency of atomic diffusion and the conductive opacity.
#### Atomic diffusion:
Castellani & Degl’Innocenti (1999) have clearly shown that to change by a factor of 2 the efficiency of microscopic diffusion (a realistic estimate of current uncertainty affecting the efficiency of this process) causes a change of about $`(0.002/+0.004)M_{}`$ in the value of $`M_{core}^{He}`$ (the He core mass increasing when the atomic diffusion efficiency is increased).
#### Conductive opacities:
since the conductive transport efficiency regulates the thermal state of the electron degenerate He core, a reliable estimate of the conductive opacities is fundamental for deriving the correct value of the He-core mass at the He-flash. As a general rule, higher conductive opacities cause a less efficient cooling of the He-core and an earlier He-ignition (i.e., at a lower core mass).
Until few years ago, only two choices were available, neither of which is totally satisfactory: the analytical relation provided by Itoh et al. (1983, I83), or the old Hubbard & Lampe (1969, HL) tabulation. As pointed out by Catelan, de Freitas Pacheco & Horvath (1996), the most recent results by I83 are an improvement over the older HL ones, but their range of validity does not cover the He-cores of RGB stars. When using the I83 conductive opacity, Castellani & Degl’Innocenti (1999) found an increase - with respect to the models based on the HL opacity - by 0.005$`M_{}`$ of $`M_{core}^{He}`$ core at the He-flash for a 0.8$`M_{}`$ model with initial metallicity Z=0.0002, while in case of a 1.5$`M_{}`$ star with solar chemical composition the increase amounts to 0.008$`M_{}`$ (Castellani et al. 2000). Quite recently, new estimates for the conductive opacity has been provided by Potekhin (1999). This new set represents a significant improvement (both in the accuracy and in the range of validity) with respect to previous estimates. It is worth noting that RGB stellar models based on these new conductive opacities provided He core masses at the He-ignition whose values are intermediate between those provided by the previous conductive opacity estimates (although more similar to the determinations based on the Itoh et al. (1983) ones). However, as noticed by Potekhin (1999) and further emphasized by Catelan (2005), not even these new conductive opacity fully covers the thermal conditions characteristic of electron degenerate cores in low-mass, metal-poor stars. So it is evident that additional work in this direction is strongly encouraged.
As far as it concerns the EOS, a preliminary investigation on the effect of different EOS choices has been performed by Vandenberg & Irwin (1997) and more recently by Cassisi et al. (2003): it has been noticed that, when the adopted EOS accounts for all the different physical processes at work in the dense core of RGB stars, the residual uncertainty on the value of $`M_{core}^{He}`$ can be small. The impact of current uncertainties on the relevant nuclear reaction rates as the one corresponding to the $`3\alpha `$ process has been recently investigated by Weiss et al. (2005, but see also Brocato et al. 1998) with the result that present uncertainty on the relevant rate has no significant influence on theoretical predictions about the TRGB.
When, one considers as ‘standard’ a model accounting for standard atomic diffusion, current uncertainties in diffusion efficiency and conductive opacity can globally contribute to an uncertainty on the He core mass of the order of $`0.01M_{}`$. It can be useful to briefly remember that, since $`\mathrm{log}(L_{TRGB})/M_{core}^{He}4.7`$, this uncertainty immediately translates in an error of about $`0.10`$mag in the bolometric magnitude of the TRGB. In our belief, this is a realistic estimate of current uncertainty affecting theoretical predictions on this relevant feature.
We show in fig. 13 the comparison of the most recent results concerning the TRGB bolometric magnitude and $`M_{core}^{He}`$ at the He-flash; the displayed quantities refer to a 0.8$`M_{}`$ model and various initial metallicities (scaled solar metal distribution).
There exists fair agreement among the various predictions of the $`\mathrm{M}_{\mathrm{bol}}^{\mathrm{TRGB}}`$ metallicity dependence, and all the $`\mathrm{M}_{\mathrm{bol}}^{\mathrm{TRGB}}`$ values at a given metallicity are in agreement within $`0.10`$ mag, with the exception of the Padua models (Girardi et al. 2000) and the Yale ones (Yi et al. 2001), which appear to be underluminous with respect to the others. As for the Padua models this difference follows from their smaller $`M_{core}^{He}`$ values; it is worth noticing that the recent models by Salasnich et al. (2000), which are an update of the Padua ones, provide brighter $`\mathrm{M}_{\mathrm{bol}}^{\mathrm{TRGB}}`$, similar to the results by Vandenberg et al. (2000). In case of the Yale models, the result is surprising since the fainter TRGB luminosity cannot be explained by much smaller $`M_{core}^{He}`$ values, since this quantity is very similar to, for instance, the results given by Vandenberg et al. (2000). When neglecting the Padua and Yale models, the 0.1 mag spread among the different TRGB brightness estimates can be interpreted in terms of differences in the adopted physical inputs such as for instance the electron conduction opacities.
Due to the large relevance of the TRGB as standard candle, it is worthwhile showing a comparison between (some) theoretical predictions about the I-cousin band magnitude of the TRGB and empirical calibrations. This comparison is displayed in fig. 14, where we show also the recent empirical calibration provided by Bellazzini et al. (2001) based on the GGC $`\omega `$ Cen. In this plot, we have shown different calibrations of $`M_I^{TRGB}`$ as a function of the metallicity based on our own stellar models. These calibration are about $`0.150.20`$ mag brighter than the most recent, empirical ones. When considering also various theoretical calibrations as those displayed in fig. 14, one notices that these updated calibrations of $`M_I^{TRGB}`$ are within $`1.5\sigma `$ of the calibration provided by Bellazzini et al. (2001).
In this context, it can be useful to remember that in order to derive this calibration a bolometric correction scale for the $`I`$band has to be used (see Salaris & Cassisi 1998) that, as it is well known, can be affected by large uncertainty. Therefore, it appears quite difficult at this time to disentangle the contribution to the global discrepancy between theoretical and empirical calibration, due to current uncertainty in the $`BC_I`$ scale from that associated to present uncertainties in stellar RGB models. In order to illustrate better this point, we show in fig. 15 a comparison between our theoretical calibration in different photometric planes and the corresponding empirical ones provided by Bellazzini et al. (2004): the evidence that the same theoretical calibration does not work properly for the I-cousin band while providing a very good match in the near-infrared bands strongly shed light on the importance of an accurate and critical analysis on the uncertainties affecting the $`BC`$ scales for the various photometric bands.
## 5. The He-burning structures
The Horizontal Branch is one of the most important evolutionary sequences in the CMD. The reasons for this relevance are manifold: 1) the brightness of the RR Lyrae stars and more, in general, the brightness of the HB, is the traditional distance indicator for metal-poor stellar populations; 2) the number of stars observed along this branch (a quantity tightly related with the core He-burning lifetimes) enters in the R parameter definition (Buzzoni et al. 1983), the most important He indicator for old stellar systems; 3) the HB morphology is related to the long-standing, and still unsettled, problem of the ‘second parameter’ in the galactic GC system.
From a theoretical point of view, although we know well for a long time the structural and evolutionary properties of HB stars, we can not yet be fully confident in the theoretical predictions concerning this evolutionary phases, at least as far as it concerns the luminosity and the evolutionary lifetime. This is simply due to the evidence that the evolutionary properties of HB stars strongly depend on all the physical processes at work during the early RGB phase. Therefore, the uncertainties affecting the physical scenario used for computing H-burning structures appear, in some sense, amplified when considering HB stellar models which are, in addition, affected by other sources of uncertainty as the rates of the He-burning processes and the efficiency of mixing at the border of the convective core.
This topic has been accurately analyzed by Castellani & Degl’Innocenti (1999), which have investigated the sensitivity of HB luminosities to the uncertainties affecting the various physical inputs. The errors affecting the HB evolutionary lifetimes have been extensively reviewed by Brocato et al. (1998) and Cassisi et al. (1998, 2003).
### 5.1. The luminosity of the Horizontal Branch
It is well known that the bolometric luminosity of a ZAHB structure is governed by two parameters: mostly the He core mass and, to a minor extent, the chemical stratification of the envelope. On the basis of this consideration, one can easily realize that the whole set of uncertainties affecting the value of $`M_{core}^{He}`$ at the TRGB directly affect also the HB brightness. When considering only as sources of error: the atomic diffusion efficiency and the different choices about the conductive opacity; one derives that the visual absolute magnitude of the HB is uncertain at the level of $`0.10`$ mag, corresponding to a $`10`$% uncertainty on this relevant feature. A more detailed analysis accounting for all error sources shows that the stellar evolution theory predicts the ZAHB luminosity with an uncertainty on $`M_V`$ of $`0.06/+0.11`$ mag<sup>7</sup><sup>7</sup>7This estimate does not account for any contribution to the error budget coming from the adopted $`BC_V`$ scale..
In fig. 16, a comparison among different, updated, theoretical predictions is shown. For the sake of comparison, we display in the same plot the semi-empirical ZAHB brightness estimates provided by De Santis & Cassisi (1999).
### 5.2. The core He-burning lifetime
As for any evolutionary phase, the lifetime of the core He-burning phase - for a given total mass - does depend on the ‘speed’ at which the nuclear processes occur, i.e., on the nuclear reaction rates, and on the amount of available fuel, i.e., in the case of burning occurring in a convective core, on the location of the outer convective boundary. Really, the uncertainty on the HB evolutionary lifetimes is dominated by the uncertainty on nuclear reaction rates and by, to a minor extent (see below) from the not-well known, efficiency of convective processes.
Concerning the reaction rates, it is worth emphasizing that the $`{}_{}{}^{12}\mathrm{C}(\alpha ,\gamma ){}_{}{}^{16}\mathrm{O}`$ reaction is, together with the triple$`\alpha `$ process, the most important among those involved in the He-burning. This occurrence being due to the evidence that: i) its nuclear cross-section strongly affects the C/O ratio in the core of carbon-oxygen white dwarfs and, in turn, their cooling times; ii) when the abundance of He inside the convective core is significantly reduced, the $`{}_{}{}^{12}\mathrm{C}(\alpha ,\gamma ){}_{}{}^{16}\mathrm{O}`$ reaction becomes strongly competitive with the $`3\alpha `$ reactions (which need three $`\alpha `$ particles) in supplying the nuclear energy budget. This means that the cross-section of this nuclear process has a huge impact on the core He-burning lifetime as well as on the chemical stratification in the core at the central He exhaustion.
Unfortunately, this reaction has a resonance and a very low cross-section at low energies, and so the nuclear parameters are difficult to measure experimentally or to calculate by theoretical analysis. As a consequence, an uncertainty of a factor of 2 is reasonable for this nuclear reaction rate (Caughlan & Fowler 1988, but see also the recent analysis by Kunz et al 2002).
The uncertainty on the $`{}_{}{}^{12}\mathrm{C}(\alpha ,\gamma ){}_{}{}^{16}\mathrm{O}`$ reaction rate strongly affects the HB lifetime: according to Brocato et al. (1998) the HB lifetime change correlates with a variation of the rate as $`\mathrm{\Delta }t_{HB}/t_{HB}0.10\mathrm{\Delta }\sigma _{{}_{}{}^{12}C}/\sigma _{{}_{}{}^{12}C}`$ (see also Zoccali et al. 1999).
As already noted, the HB lifetime does strongly depend also on the efficiency of convection-induced mixing at the boundary of the convective core. More in detail, all along the HB evolutionary phase, the treatment of mixing at the boundary of the convective core, is really a relevant problem. In fact, as a consequence of the burning processes, He is transformed into carbon and oxygen whose associate opacity is larger with respect that of an He rich mixture. This change in opacitive properties of the stellar matter in the core, strongly modifies the behaviour of the radiative gradient, producing an increasing of the mass size of the convective core.
Unfortunately, in spite of the many theoretical works published over the last three decades, the physics that determines the extent of this convective region is still poorly known. The theoretical calculations available so far leave various scenarios open. Classical models, those based on a bare Schwarzschild criterion (Iben & Rood 1970), are still calculated and widely used in many studies (e.g., Althaus et al. 2002). However, models that include some algorithm to handle the discontinuity of the opacity that forms at the external border of the convective core as a consequence of the conversion of He into C and O should be considered as more reliable (Castellani, Giannone, & Renzini 1971a, 1971b; Demarque & Mengel 1972; Sweigart & Gross 1976; Castellani et al. 1985; Dorman & Rood 1993 and references therein). According to this mixing scheme, the change in the opacitive properties of the core naturally leads to the growth of the convective core (the so-called induced overshoot) and to the formation of a semiconvective layer outside the fully convective region.
In an alternative approach, it is assumed that a mechanical overshoot takes place at the boundary of the convective region (Saslaw & Schwarzschild 1965; Girardi et al. 2000). Although, the real occurrence of this phenomenon is out of debate, a quantitative estimate of the overshoot efficiency is still a unsettled issue. However, detailed evolutionary computations (Straniero et al. 2003) show that a moderate efficiency of mechanical overshoot mimics the effect of induced overshoot, whereas a large efficiency would produce a convective core so large to include the semiconvective region, so causing large changes in the structural and evolutionary properties.
The approach adopted for managing the convection-induced mixing in HB structures does affect largely the evolutionary lifetimes as a consequence of the change in available amount of fuel, but in addition it largely affects also the C/O ratio in the CO core at the central He exhaustion. Therefore the effects are quite similar to different assumptions about the rate for the nuclear process $`{}_{}{}^{12}\mathrm{C}(\alpha ,\gamma ){}_{}{}^{16}\mathrm{O}`$. Therefore, there exists a sort of degeneracy between this reaction rate and the efficiency of mixing during the core He-burning phase. The question is how we can break this degeneracy.
The answer is positive: this can be obtained by using different independent empirical constraints whose comparison with theoretical predictions allows to disentangle the evolutionary and structural effects associated with nuclear reaction rates and mixing processes. In this context, it is useful to remember that in these last years a large effort has been devoted to the calibration of the R parameter (Buzzoni et al. 1983) in order to estimate the primordial He abundance of the GGC system. This parameter is defined as the number ratio between HB and RGB stars brighter than the HB. So its theoretical calibration is strongly affected by model predictions about the HB lifetime. The recent analysis performed by Cassisi et al. (2003) and Salaris et al. (2004) have shown that the new generation of HB models based on the more recent evaluation of the $`{}_{}{}^{12}\mathrm{C}(\alpha ,\gamma ){}_{}{}^{16}\mathrm{O}`$ rate and on the semiconvective mixing scheme<sup>8</sup><sup>8</sup>8These models neglect also the occurrence of breathing pulses in the late phase of the core He-burning stage. For a detailed discussion about the reasons for which the occurrence of this process in real stars is considered implausible we refer to Caputo et al. (1989). are able to provide an estimate of the initial He abundance in very good agreement with the measurements obtained through the analysis of the Cosmic Microwave Background anisotropies and primordial nuclesynthesis models. They also predict a value for the parameter $`R_2`$ (i.e. the ratio between the number of AGB stars and that of HB objects - Caputo, Castellani & Wood 1978) in fair agreement with observations. It is worth noting that the $`R_2`$ parameter is strongly affected by the adopted mixing approach, since the larger the mixing during the core He-burning, the less the amount of He will be available for the subsequent AGB evolutionary phase.
In addition, the recent analysis of the non-radial pulsations of White Dwarfs (Metcalfe et al. 2000, 2001) can provide important clues about the C/O ratio within the CO core as well as on the ratio between the CO core mass and the total WD mass. All these empirical constraints when analyzed within a self-consistent, and updated evolutionary framework can be of extreme relevance in order to improve our knowledge on the physical processes at work in He-burning, low-mass stars.
## 6. The clump of the Asymptotic Giant Branch
Stellar evolutionary models (Castellani, Chieffi, & Pulone 1990) consistently predict that after the central He-exhaustion, the He burning rapidly moves from the core to the shell surrounding the CO core whose extension is fixed by the mass size of the convective core during the previous HB phase. Thus, the beginning of the AGB is characterized by a rapid increase in luminosity. When the shell He burning stabilizes, a slowing down of the evolutionary rate is expected. These theoretical predictions are well confirmed by empirical evidence in GGCs showing that the transition between the central and the shell He burning is marked by a clear gap (where few stars are observed), and that a well-defined clump of stars (at least in the more populous - or the well sampled - clusters) is found at the base of the AGB.
From a theoretical point of view (Pulone 1992; Bono et al. 1995) it is well known that the luminosity level of the AGB clump is almost independent of the chemical composition, i.e. it does not depend significantly on the initial He abundance and metallicity (see fig. 17). As a consequence, it was suggested by Pulone (1992) the use of this observational feature as standard candle.
When checking the reliability of theoretical predictions about the luminosity of the AGB clump by comparison with empirical evidence, in order to overcome the problem related with the quite uncertain cluster distance scale, it is common to consider, in analogy with the RGB bump, the brightness (for instance in the V band) difference between the AGB clump and the HB, i.e. the parameter $`\mathrm{\Delta }V_{HB}^{AGB}`$.
A detailed analysis of the effects on the AGB clump brightness associated to the current uncertainties in the adopted physical inputs as well as in the approach adopted for treating the mixing processes during the core He-burning phase has been performed by Cassisi et al. (2001). They found that the $`\mathrm{\Delta }V_{HB}^{AGB}`$ parameter is not affected at all by current uncertainty in the atomic diffusion coefficients.
However, this quantity is quite sensitive to the approach used for the treatment of the breathing pulses phenomenon at the end of the core He-burning stage. More in detail, we note that the empirical evidence of the AGB clump in GGCs seems to clearly rule out the occurrence of this phenomenon in real stars. In fact, numerical simulations show that, when breathing pulses are allowed to occur, the drop in luminosity associated to the AGB clump is almost vanishing. As a consequence, population synthesis models, based on stellar models accounting for the occurrence of breathing pulses, do not show any evident increase in the star count in the region of the CMD where the AGB clump is really observed.
In addition, Cassisi et al. (2001) have shown that, at odds with what occur for the HB and AGB evolutionary lifetimes (see the discussion in section 5.2.), the $`\mathrm{\Delta }V_{HB}^{AGB}`$ parameter is not largely affected by present uncertainties in the physical inputs adopted for computing the stellar models.
Recently Ferraro et al. (1999) have investigated the dependence of the AGB clump brightness on the HB morphology. The existence of a clear correlation between the brightness of the clump and the HB type, in the sense that old stellar systems with bluer HBs are expected to show an AGB clump becoming bluer and bluer and less clumpy (and, in turn, less observable) has to be accounted for before using this feature as a distance indicator.
## 7. Final remarks
In this paper (but see also the rich, quoted, literature), we have shown that theoretical predictions on stellar models are affected by sizeable uncertainties, a clear proof being the occurrence of not-negligible differences between results provided by different theoretical groups.
From the point of view of stellar models users, the best approach to be used for properly accounting for these uncertainties, is to not use evolutionary results with an uncritical approach and, also to adopt as many as possible independent theoretical predictions in order to have an idea of the uncertainty existing in the match between theory and observations. It would be also worthwhile to pay attention to the improvements adopted both in the physical inputs as well as in the physical assumptions, by people computing the evolutionary models.
On the other hand, stellar model makers should continue their effort of continuously updating their models in order to account for the ‘best’ physics available at any time, and consider the various empirical constraints as a benchmark of their stellar models. This represents a fundamental step for obtaining as much as possible accurate and reliable stellar models.
In the previous sections, we have mentioned that a quite important source of uncertainty in the comparison between theory and observations derives from the errors still affecting both theoretical and empirical color - effective temperature relations and the bolometric correction scales for the different photometric bands. It is evident that these uncertainties do strongly hamper the possibility of a sound comparison between stellar models and empirical evidence and, of course, make extremely problematic to assess the level of accuracy of present evolutionary scenario. In our belief, a big effort should be devoted in the near future in order to improve the accuracy and reliability of the transformations adopted for transferring stellar models from the H-R diagram to the various observational planes.
It has also been emphasized the huge impact of both GGCs metallicity and distance scale uncertainties on the possibility to realize a meaningful comparison between theory and observations. Although, large improvements have been achieved in these fields, current errors are still too large for offering the opportunity of a plain assessment of residual uncertainties in the present generation of stellar models.
These considerations make clear that a sizeable improvement in the stellar evolution framework could be achieved, in the near future, only if scientists, working in different fields of astrophysical research, will provide their own contribution to reduce the still existing uncertainties affecting both the theoretical framework as well as the observational scenario.
##### Acknowledgments.
We warmly acknowledge G. Bono, V. Castellani and A. Irwin for their pertinent suggestions and positive comments on an early draft of this manuscript. We are also grateful to them for many enlightening discussions. We also wish to thank M. Bellazzini, S. Degl’Innocenti and A. Pietrinferni for providing the data shown in some figures. We warmly thank the LOC and the SOC for organizing this interesting meeting. It is a real pleasure to thank David Valls-Gabaud for all the help provided and the pleasant discussions. This research has made use of NASA’s Astrophysics Data System Bibliographic Services.
|
warning/0506/math-ph0506058.html
|
ar5iv
|
text
|
# Quantum bundles and quantum interactions
## Introduction
Quantisation, intended as the construction of a quantum theory by applying suitable rules to classical systems, is perhaps the most common approach to the study of the foundations of quantum physics; indeed, this philosophy has produced an immense physical and mathematical literature. There is, however, a widespread opinion that the true relation between classical and quantum theories should rather go in the opposite sense: at least in principle, classical physics should derive from quantum physics, thought to be more fundamental.
As a first step in that direction, one could try and build a stand-alone mathematical model, not derived from a quantisation procedure, which should reproduce (at least) the basic observed facts of elementary particle physics. The present article is a proposal in this sense, based on two main ingredients: *free states*, and *interaction*. A further interesting feature of the model is its freedom from the requirement of spacetime flatness.
The fundamental mathematical tool of my exploration is the geometry of *distributional bundles*, that is bundles over classical (finite-dimensional Hausdorff) manifolds whose fibres are distributional spaces. These arise naturally from a class of finite-dimensional 2-fibred bundles, which turns out to contain the most relevant physical cases. The basics of their geometry have been exposed in two previous papers \[C00a, C04a\] along the line of thought stemming from Frölicher’s notion of smoothness \[Fr82, FK88, KM97, MK98, CK95\].
While I do not quantise classical fields, at this stage I do consider certain finite-dimensional geometric structures which are related to classical field theories.<sup>1</sup><sup>1</sup>1 In particular gravitation, here, is a fixed background. From these one can naturally build 2-fibred bundles and, eventually, *quantum bundles*: distributional bundles whose fibres are spaces of one-particle states, and the related *Fock bundles*. It turns out that the underlying, finite-dimensional geometric structure determines a distinguished connection on a quantum bundle; this connection is related to the description of *free-particle states*.
The basic idea about quantum interactions is that they should be described by a new connection on the Fock bundle, obtained by adding an interaction morphism to the free-particle connection. This approach requires the notion of a *detector*, defined to be a timelike 1-dimensional submanifold of the spacetime manifold. Then a natural interaction morphism indeed exists in the fibres of the restricted Fock bundle. It turns out that a detector carries a quantum “formalism” which can be seen as a kind of complicated clock; in the flat, inertial case this turns out to reproduce the basic results of the usual quantum field theory,<sup>2</sup><sup>2</sup>2 The usual quantum fields can be recovered \[C04b\] as certain natural geometric structures of the quantum bundles, but they only play a marginal role in this approach. while in general it could be seen as a local, “linearized” description of the actual physics.
The paper’s plan is as follows. In the two first sections I will summarize the basic ideas about distributional bundles and quantum bundles, the latter being defined as certain bundles of generalized half-densities on classical momentum bundles; then I will introduce generalized frames for quantum bundles and the notion of a detector. In section 5 I will illustrate the construction of the quantum interaction from a general (and necessarily sketchy) point of view. In section 6 these ideas will be implemented in the simplest case, a theory of two scalar particles; in sections 7, 8, 9 and 10 I will show how to treat QED in the above said setting; in the flat inertial case one then recovers the basic known results. Here, the role of 2-fibred bundles turns out to be specially meaningful.
## 1 Distributional bundles
For details about the ideas reviewed in this section, see \[C00a, C04a\].
Let $`𝗉:𝒀\underset{¯}{𝒀}`$ be a real or complex classical vector bundle, namely a finite-dimensional vector bundle over the Hausdorff paracompact smooth real manifold $`\underset{¯}{𝒀}`$ . Moreover assume that $`\underset{¯}{𝒀}`$ is oriented, let $`n:=dim\underset{¯}{𝒀}`$ , and consider the positive component $`𝕍^{}\underset{¯}{𝒀}:=(^n\mathrm{T}^{}\underset{¯}{𝒀})^+\underset{¯}{𝒀}`$ , called the bundle of *positive densities* on $`\underset{¯}{𝒀}`$ .
Let $`𝓨_{}𝓓_{}(\underset{¯}{𝒀},𝕍^{}\underset{¯}{𝒀}\underset{\underset{¯}{𝒀}}{}𝒀^{})`$ be the vector space of all ‘test sections’, namely smooth sections $`\underset{¯}{𝒀}𝕍^{}\underset{¯}{𝒀}\underset{\underset{¯}{𝒀}}{}𝒀^{}`$ which have compact support. A topology on this space can be introduced by a standard procedure \[Sc66\]; its topological dual will be denoted as $`𝓨𝓓(\underset{¯}{𝒀},𝒀)`$ and called the space of *generalized sections*, or *distribution-sections* of the given classical bundle. Some particular cases of generalized sections are that of *$`r`$-currents* ($`𝒀^r\mathrm{T}^{}\underset{¯}{𝒀}`$ , $`r`$) and that of *half-densities* ($`𝒀(𝕍^{}\underset{¯}{𝒀})^{1/2}𝕍^{1/2}\underset{¯}{𝒀}`$).
A curve $`\alpha :𝓨`$ is said to be *F-smooth* if the map
$$\alpha ,u::t\alpha (t),u$$
is smooth for every $`u𝓨_{}`$ . Accordingly, a function $`\varphi :𝓨`$ is called F-smooth if $`\varphi \alpha :`$ is smooth for all F-smooth curve $`\alpha `$ . The general notion of F-smoothness, for any mapping involving distributional spaces, is introduced in terms of the standard smoothness of all maps, between finite-dimensional manifolds, which can be defined through compositions with F-smooth curves and functions. Moreover, it can be proved that a function $`f:𝑴`$ , where $`𝑴`$ is a classical manifold, is smooth (in the standard sense) iff the composition $`fc`$ is a smooth function of one variable for any smooth curve $`c:𝑴`$ . Thus one has a unique notion of smoothness based on smooth curves, including both classical manifolds and distributional spaces.
In the basic classical geometric setting underlying distributional bundles one considers a classical $`2`$-fibred bundle
$$\begin{array}{ccccc}𝑽& \stackrel{𝗊}{}& 𝑬& \stackrel{\underset{¯}{𝗊}}{}& 𝑩,\end{array}$$
where $`𝗊:𝑽𝑬`$ is a vector bundle, and the fibres of the bundle $`𝑬𝑩`$ are smoothly oriented. Moreover, one assumes that $`𝗊\underset{¯}{𝗊}:𝑽𝑩`$ is also a bundle, and that for any sufficiently small open subset $`𝑿𝑩`$ there are bundle trivializations
$$(\underset{¯}{𝗊},\underset{¯}{𝗒}):𝑬_𝑿𝑿\times \underset{¯}{𝒀},(𝗊\underset{¯}{𝗊},𝗒):𝑽_𝑿𝑿\times 𝒀$$
with the following projectability property: there exists a surjective submersion $`𝗉:𝒀\underset{¯}{𝒀}`$ such that the diagram
$$\begin{array}{ccc}𝑽_𝑿& \stackrel{(𝗊\underset{¯}{𝗊},𝗒)}{}& 𝑿\times 𝒀\\ 𝗊& & 11{}_{𝑿}{}^{}\times 𝗉& & \\ 𝑬_𝑿& \underset{(\underset{¯}{𝗊},\underset{¯}{𝗒})}{}& 𝑿\times \underset{¯}{𝒀}\end{array}$$
commutes; this implies that $`𝒀\underset{¯}{𝒀}`$ is a vector bundle, which is not trivial in general.
The above conditions are easily checked to hold in many cases which are relevant for physical applications, and in particular when $`𝑽=𝑬\underset{𝑩}{\times }𝑾`$ where $`𝑾𝑩`$ is a vector bundle, when $`𝑽=\mathrm{V}𝑬`$ (the vertical bundle of $`𝑬𝑩`$) and when $`𝑽`$ is any component of the tensor algebra of $`\mathrm{V}𝑬𝑬`$ .
For each $`x𝑩`$ one considers the distributional space $`𝓥_x:=𝓓(𝑬_x,𝑽_x)`$, and obtains the fibred set
$$\mathrm{}:𝓥𝓓_𝑩(𝑬,𝑽):=\underset{x𝑩}{}𝓥_x𝑩.$$
An isomorphism of vector bundles yields an isomorphism of the corresponding spaces of generalized sections; hence, a local trivialization of the underlying classical 2-fibred bundle, as above, yields a local bundle trivialization
$$(\mathrm{},𝖸):𝓥_𝑿𝑿\times 𝓨,𝓨𝓓(\underset{¯}{𝒀},𝒀)$$
of $`𝓥𝑩`$ . Moreover, a smooth atlas of 2-bundle trivializations determines a linear F-smooth bundle atlas on $`𝓥𝑩`$ , which is said to be an *F-smooth distributional bundle*. In general, the F-smoothness of any map from or to $`𝓥`$ is equivalent to the F-smoothness of its local trivialized expression.
One defines the *tangent space* of any F-smooth space through equivalence classes of F-smooth curves; tangent prolongations of any F-smooth mappings can also be shown to exist. Thus one gets, in particular, the tangent space $`\mathrm{T}𝓥`$, which has local trivializations as $`\mathrm{T}𝑿\times \mathrm{T}𝓨`$, its *vertical subspace* and the *first jet bundle* $`\mathrm{J}𝓥𝓥`$. A *connection* is defined to be an F-smooth section $`𝔊:𝓥\mathrm{J}𝓥`$.
With some care, many of the usual chart expressions of finite-dimensional differential geometry can be extended to the distributional case. In particular, let $`\sigma :𝑩𝓥`$ be an F-smooth section and $`\sigma ^𝖸:=𝖸\sigma :𝑩𝓨`$ its ‘chart expression’. Then its covariant derivative has the chart expression
$$(\sigma )^𝖸=\dot{𝗑}^a(_a\sigma ^𝖸𝔊_{𝖸a}\sigma ^𝖸),$$
where $`(𝗑^a)`$ is a chart on $`𝑿𝑩`$ and $`𝔊_{𝖸a}:𝑿\mathrm{End}(𝓨)`$ , $`a=1,\mathrm{},dim𝑩`$ .
The notions of *curvature* and of *adjoint connection* can also be introduced. Furthermore, it can be shown that any projectable connection on the underlying classical 2-bundle determines a distributional connection; however, not all distributional connections arise from classical ones.
## 2 Quantum bundles
Let $`𝕃`$ be the semi-vector space of *length units* (see \[CJM95, C00b\] for a review of unit spaces) and $`(𝑴,g)`$ a spacetime. The spacetime metric $`g`$ has ‘conformal weight’ $`𝕃^2𝕃𝕃`$ , namely it is a bilinear map $`\mathrm{T}𝑴\underset{𝑴}{\times }\mathrm{T}𝑴𝕃^2`$, while its inverse $`g^\mathrm{\#}`$ has conformal weight $`𝕃^2𝕃^{}𝕃^{}`$.
For $`m𝕃^1𝕃^{}`$ let $`𝑷_m𝑲_m^+\mathrm{T}^{}𝑴`$ be the subbundle over $`𝑴`$ of all future-pointing $`p\mathrm{T}^{}𝑴`$ such that<sup>3</sup><sup>3</sup>3Throughout this paper, the signature of the metric is $`(1,3)`$ ; moreover $`\mathrm{}=c=1`$ , so that $`𝕃`$ is the unique unit space involved. $`g^\mathrm{\#}(p,p)=m^2`$ . Then $`𝑷_m`$ is the classical *phase bundle* for a particle of mass $`m`$ ; the case $`m=0`$ can be also considered. Furthermore, consider the 2-fibred bundle
$$\left[(^3\mathrm{T}^{}𝑷_m)^+\right]{}_{}{}^{1/2}𝕍^{1/2}𝑷_m𝑷_m𝑴$$
whose upper fibres are the spaces of half-densities on the fibres of $`𝑷_m𝑴`$. There is a distinguished section
$$\sqrt{\omega _m}:𝑷_m𝕃^1𝕍^{1/2}𝑷_m;$$
here, $`\omega _m`$ is the *Leray form* of the hyperboloids (the fibres of $`𝑷_m𝑴`$), usually indicated as $`\delta (g^{}m^2)`$ where $`g^{}`$ is the contravariant quadratic form associated with the metric. If $`(𝗉_\lambda )=(𝗉_0,𝗉_i)`$ are $`𝕃^1`$-scaled orthonormal coordinates on the fibres of $`\mathrm{T}^{}𝑴𝑴`$ , then one finds the coordinate expression
$$\sqrt{\omega _m}=\frac{\sqrt{\mathrm{d}^3𝗉_{}}}{\sqrt{2\mathrm{E}_m}},$$
where $`\mathrm{E}_m:=\sqrt{m^2+|𝗉_{}^2|}=\sqrt{m^2+\delta ^{ij}𝗉_i𝗉_j}`$ is indicated simply as $`𝗉_0`$ if no confusion arises. Note that $`\mathrm{d}^3𝗉_{}\mathrm{d}𝗉_1\mathrm{d}𝗉_2\mathrm{d}𝗉_3`$ is the “spatial” (scaled) volume form determined by the “observer” associated with the coordinates, ans can be seen as a volume form on the fibres of $`𝑷_m`$ via orthogonal projection.
It can be seen \[C04a\] that the spacetime connection $`\mathrm{\Gamma }`$ determines a connection $`\mathrm{\Gamma }_m`$ of $`𝑷_m𝑴`$ , as well as a linear connection of the 2-fibred bundle $`\mathrm{V}^{}𝑷_m𝑷_m𝑴`$ which is projectable on $`\mathrm{\Gamma }_m`$ ; on turn this determines a linear projectable connection $`\widehat{\mathrm{\Gamma }}_m`$ of $`𝕍^{1/2}𝑷_m𝑷_m𝑴`$ , with the coordinate expression
$$(\mathrm{\Gamma }_m)_{ai}=\mathrm{\Gamma }{}_{ai}{}^{0}𝗉_{0}^{}\mathrm{\Gamma }{}_{ai}{}^{j}𝗉_{j}^{},(\widehat{\mathrm{\Gamma }}_m)_a=\frac{\mathrm{\Gamma }{}_{ai}{}^{0}g_{}^{ij}𝗉_j}{𝗉_0}+\frac{1}{2}\mathrm{\Gamma }_{ai}^i$$
(here $`a`$ is an index for coordinates on $`𝑴`$ and the spacelike coordinates $`(𝗉_j)`$ play the role of fibre coordinates on $`𝑷_m𝑴`$).
Next consider the distributional bundle
$$𝓟_m:=\text{ /}𝓓_𝑴(𝑷_m)𝓓_𝑴(𝑷_m,𝕍^{1/2}𝑷_m)𝑴,$$
whose fibre over each $`x𝑴`$ is the vector space of all (complex-valued) generalized half-densities on $`(𝑷_m)_x`$ . The connection $`\widehat{\mathrm{\Gamma }}_m`$ determines a smooth (in Frölicher’s sense) connection $`𝓟_m𝑴`$ which can be characterized in various ways \[C04a\], the most simple being the following: let $`c:𝑴`$ be any local curve and $`p:𝑴𝑷_m`$ a local section which is parallely transported along $`c`$ ; then the local section
$$\delta _p(\omega _m)^{1/2}:𝑴𝕃𝓟_m:x\delta _{p(x)}(\omega _m)^{1/2},x𝑴$$
is parallely transported along $`c`$ , where $`\delta _{p(x)}`$ denotes the Dirac density on $`(𝑷_m)_x`$ whose support is the point $`p(x)`$ .
Let now $`𝑽𝑷_m𝑴`$ be a (real or complex) 2-fibred vector bundle, and consider the distributional bundle
$$𝓥^1:=\text{ /}𝓓_𝑴(𝑷_m,𝑽)𝓓_𝑴(𝑷_m,𝕍^{1/2}𝑷_m\underset{𝑷_m}{}𝑽)𝑴,$$
whose fibre over each $`x𝑴`$ is the vector space of all $`𝑽`$-valued generalized half-densities on $`(𝑷_m)_x`$ . In practice, this $`𝑽`$ will be related to the bundle whose sections are the fields of the classical theory which, in the usual approach, correspond to the quantum theory under consideration. One could think that it suffices to deal with a “semi-trivial” 2-fibred bundle $`𝑷_m\underset{𝑴}{\times }𝑽`$ where $`𝑽𝑴`$ is a vector bundle, however it will be seen (§10) that the general setting is actually needed.
Remark. If a Hermitian metric on the fibres of $`𝑽`$ is given, then one can define a *Hilbert bundle* $`𝓗𝑴`$ , and has inclusions $`𝓥_{}^1𝓗𝓥^1`$ (where $`𝓥_{}^1𝑴`$ is the subbundle whose fibres are constituted by test sections); namely one has a bundle of ‘rigged Hilbert spaces’ \[BLT75\].
A *Fock bundle* can be constructed as
$$𝓥:=\underset{j=0}{\overset{\mathrm{}}{}}𝓥^j,$$
where
$$\text{either}𝓥^j:=^j𝓥^1\text{or}𝓥^j:=^j𝓥^1$$
(antisymmetrized and symmetrized tensor products).
If a connection $`\gamma `$ of $`𝑽𝑷_m𝑴`$ linear projectable over $`\mathrm{\Gamma }_m`$ is given (which is the case in most physical situations), then one also gets a connection of $`𝓥^1𝑴`$ ; this can be naturally extended to a connection on $`𝓥𝑴`$, which will be called the *free particle connection*. For any local section $`\sigma :𝑴𝓥^1`$ one has the coordinate expression
$$_a^A(\sigma )=\gamma {}_{aB}{}^{A}\sigma _{}^{B}(\mathrm{\Gamma }_m)_{ai}^i\sigma ^A.$$
## 3 Generalized frames
For each $`p(𝑷_m)_x`$ , $`x𝑴`$, let $`\delta [p]`$ denote the Dirac generalized density on the fibre $`(𝑷_m)_x`$ with support $`\{p\}`$ ; namely, if $`f:(𝑷_m)_x`$ is a test function then one has $`\delta [p],f=f(p)`$ . It can be written as<sup>4</sup><sup>4</sup>4 While $`p`$ denotes an element of $`𝑷_m`$ , the sans-serif symbol $`𝗉`$ is used for the fibre coordinates.
$$\delta [p]=\stackrel{˘}{\delta }[p]\mathrm{d}^3𝗉_{}$$
where $`\stackrel{˘}{\delta }[p]`$ is the $`𝕃^3`$-scaled *generalized function*, usually denoted as $`\stackrel{˘}{\delta }[p](q)\delta (qp)`$ , acting on test densities $`\varphi =\stackrel{˘}{\varphi }\mathrm{d}^3𝗉_{}`$ as
$$\stackrel{˘}{\delta }[p],\varphi =\delta [p],\stackrel{˘}{\varphi }=\stackrel{˘}{\varphi }(p).$$
Actually any generalized density can be expressed in this way as a generalized function times a given volume form; moreover, note that the spacelike volume form $`\mathrm{d}^3𝗉_{}`$ , as well as the induced volume form on the fibres of $`𝑷_m𝑴`$ denoted in the same way, only depends on the choice of an ‘observer’ (i.e. a timelike future-pointing unit vector field) and not on the particular frame of $`\mathrm{T}^{}𝑴`$ adapted to it.
Let now $`l𝕃`$ be an arbitrarily fixed length unit, and consider the unscaled generalized half-density
$$𝖡_p:=l^{3/2}\stackrel{˘}{\delta }[p]\sqrt{\mathrm{d}^3𝗉_{}}=\frac{1}{\sqrt{2l^3𝗉_0}}\delta [p]\omega _m^{1/2},$$
acting on test half-densities $`\theta =\stackrel{˘}{\theta }\sqrt{\mathrm{d}^3𝗉_{}}`$ as
$$𝖡_p,\theta =l^{3/2}\delta [p],\stackrel{˘}{\theta }=l^{3/2}\stackrel{˘}{\theta }(p);$$
for each $`x𝑴`$ the set $`\{𝖡_p\}`$ , $`p(𝑷_m)_x`$ , can be seen as a *generalized frame* of the distributional bundle $`𝓟_m`$ at $`x`$ . Let moreover $`\{𝖻_A\}`$ be a frame of the classical vector bundle $`𝑽𝑷_m`$ , $`A=1,\mathrm{},n`$ ; then
$$\{𝖡_{pA}\}:=\{𝖡_p𝖻_A\}$$
is a generalized frame of $`𝓥^1𝑴`$ . In fact, any $`\psi =\psi ^A𝖻_A𝓥^1`$ can be written as $`\psi ^A(𝗉)𝖡_{𝗉A}`$ , which is to be intended in the generalized sense
$$\psi ,\theta =\stackrel{˘}{\psi }^A(𝗉)\stackrel{˘}{\theta }_A(𝗉)\mathrm{d}^3𝗉_{}$$
where $`\theta 𝓥_{}^1`$ is a test half density in the same fibre as $`\psi `$ .
Let $`𝓐`$ be a set (index set); a *generalized multi-index* is defined to be a map
$$I:𝓐\{0\}$$
vanishing outside some finite subset $`𝓐_I𝓘`$ ; it can be represented through its graphic
$$\{(\alpha _1,I_1),(\alpha _2,I_2),\mathrm{},(\alpha _r,I_r)\},𝓐_I=\{\alpha _1,\mathrm{},\alpha _r\}$$
for any (arbitrary and inessential) ordering of $`𝓐_I`$ . Now one extends the generalized frame $`\{𝖡_{pA}\}`$ to a generalized frame of the Fock bundle $`𝓥𝑴`$ by letting $`𝓐_x=(𝑷_m)_x\times \{1,\mathrm{},n\}`$ for each $`x𝑴`$ , and setting
$`𝖡_I:={\displaystyle \frac{(𝖡_{\alpha _1})^{I_1}(𝖡_{\alpha _2})^{I_2}\mathrm{}(𝖡_{\alpha _r})^{I_r}}{\sqrt{I_1!I_2!\mathrm{}I_r!}}}`$ $`\text{(bosons)},`$
$`𝖡_I:=𝖡_{\alpha _1}𝖡_{\alpha _2}\mathrm{}𝖡_{\alpha _r}`$ $`\text{(fermions)},`$
where
$$(𝖡_\alpha )^k:=\underset{k\text{times}}{\underset{}{𝖡_\alpha \mathrm{}𝖡_\alpha }}.$$
In a more detailed way one writes $`\alpha _i=(p_i,A_i)`$ and
$`𝖡_I:={\displaystyle \frac{(𝖡_{p_1A_1})^{I_1}(𝖡_{p_2A_2})^{I_2}\mathrm{}(𝖡_{p_rA_r})^{I_r}}{\sqrt{I_1!I_2!\mathrm{}I_r!}}}`$ $`\text{(bosons)},`$
$`𝖡_I:=𝖡_{p_1A_1}𝖡_{p_2A_2}\mathrm{}𝖡_{p_rA_r}`$ $`\text{(fermions)}.`$
If one has a Hermitian structure in the fibres of $`𝑽𝑷_m`$ and $`\{𝖻_A\}`$ is an orthonormal classical frame, then one gets an ‘orthonormality’ relation $`𝖡_I,𝖡_J=\delta _{IJ}`$ , to be interpreted in a generalized (i.e. distributional) sense.
## 4 Detectors
By a ‘detector’ I mean a 1-dimensional time-like submanifold $`𝑻𝑴`$ . Locally this determines, via the exponentiation map, a time$`+`$space splitting, which in a sense relates the momentum-space based approach presented here to a position-space approach, though the relation is precise only if the induced splitting is global.
Consider restrictions of the quantum bundles previously introduced to bundles over $`𝑻`$, so write
$$𝓟_m𝑻,𝓥^1𝑻,𝓥𝑻,$$
and the like. Clearly, the free particle connection determines connections of these bundles; it actually turns out that one gets (possibly local) splittings of them. So one writes, for example
$$𝓥𝑻\times 𝓥_{t_0}$$
where $`t_0𝑻`$ is some arbitrarily chosen point. Note that the free particle connection, by construction, preserves “particle number”, namely is reducible to a connection of each of the subbundles $`𝓥^j`$, $`j\{0\}`$ .
The unit future-pointing vector field $`\mathrm{\Theta }_0:𝑻𝕃(\mathrm{T}^{}𝑴)_𝑻`$ tangent to $`𝑻`$ determines an orthogonal splitting $`(\mathrm{T}^{}𝑴)_𝑻=\mathrm{T}^{}𝑻\underset{𝑻}{\times }(\mathrm{T}^{}𝑴)_𝑻^{}`$, and a diffeomorphism $`(𝑷_m)_𝑻(\mathrm{T}^{}𝑴)_𝑻^{}`$ ; the ‘spacelike’ volume form on the fibres of $`𝕃(\mathrm{T}^{}𝑴)_𝑻^{}`$ then yields a scaled volume form on the fibres of $`(𝑷_m)_𝑻`$ ; with the choice of a length unit one obtains a generalized frame $`\{𝖡_{pA}\}`$ of $`𝓥^1𝑻`$ ; in practice, this is defined in the same way as the generalized frame of $`𝓥^1𝑴`$ introduced in §3, where now the orthonormal coordinates $`(𝗉_\lambda )(𝗉_0,𝗉_i)`$ are *adapted* to the above said splitting.
Let now $`p:𝑻(𝑷_m)_𝑻`$ be a covariantly constant section and $`\left(𝖻_A(p)\right)`$ a frame of $`𝑽𝑷_m`$ covariantly constant over $`p`$ . If $`𝑻𝑴`$ is a geodesic submanifold, then $`𝖡_{pA}`$ is covariantly constant along $`𝑻`$ relatively to the free-particle connection. Thus the generalized orthonormal set $`\{𝖡_{pA}\}`$ indexed by covariantly constant sections $`p:𝑻(𝑷_m)_𝑻`$ and by the classical index $`A`$ is constant in the same sense. If $`𝑻`$ is not geodesic then one can either construct the generalized frame at some chosen $`t_0𝑻`$ and then parallely propagate it along $`𝑻`$, or modify the definition of the free-particle connection of $`𝓥𝑻`$ by relating it to *Fermi transport* rather than parallel transport along $`𝑻`$. From the physical point of view one may expect different interpretations of these two settings, which however give rise essentially to the same formalism.
## 5 Quantum interaction
The general idea of quantum interaction is the following. Consider a Fock bundle $`𝓥=𝓥^{}\underset{𝑻}{}𝓥^{\prime \prime }\underset{𝑻}{}𝓥^{\prime \prime \prime }\underset{𝑻}{}\mathrm{}`$ (each factor being, on turn, a Fock bundle accounting for a given particle type) endowed with a free-particle connection $`𝔊`$ . Suppose that there exists a distinguished section $`:𝑻𝕃^1\mathrm{End}(𝓥)`$ ; by considering the unit future-oriented section $`\mathrm{d}𝗍:𝑻𝕃\mathrm{T}^{}𝑻`$ determined via the spacetime metric, one can introduce a new connection $`𝔊\mathrm{i}\mathrm{d}𝗍`$ (possibly mixing particle numbers and types). A section $`𝑻𝓥`$ which is constant relatively to $`𝔊\mathrm{i}\mathrm{d}𝗍`$ describes the evolution of a particle system, or rather the evolution of a quantum *clock* (in a broad sense) of the detector. This evolution can be compared to that determined by $`𝔊`$ alone, namely it can be read in a fixed Fock *space* $`𝓨𝓥_{t_0}`$ , $`t_0𝑻`$, such that $`𝓥𝑻\times 𝓨`$ is the splitting determined by $`𝔊`$ .<sup>5</sup><sup>5</sup>5Related ideas, describing the evolution of a quantum system in terms of a connection on a functional bundle in a Galileian setting, have been introduced in \[JM02, CJM95\].
Now the evolution operator $`𝒰_{t_0}:𝑻\mathrm{End}(𝓨)`$ can be written as the formal series
$$𝒰_{t_0}(t)=11+\underset{N=1}{\overset{\mathrm{}}{}}\frac{(\mathrm{i})^N}{N!}_{t_0}^td𝗍_1_{t_0}^td𝗍_2\mathrm{}_{t_0}^td𝗍_N(𝗍_1)(𝗍_2)\mathrm{}(𝗍_N),$$
where $``$ denotes the *time-ordered product*; the *scattering operator* is defined to be $`𝒮:=𝒰_{\mathrm{}}(+\mathrm{})\mathrm{End}(𝓨)`$ . However, besides any convergence questions, the basic problem is the existence of $``$ ; actually I’m going to show that there is a natural way of introducing it, and a way which is consistent with the results of the standard theory, but only as a morphism $`𝓥_{}𝓥`$ (where $`𝓥_{}𝓥`$ is the subbundle of test elements). This implies that many *single terms* of the above series are not defined. Nevertheless, parts of it give considerable information which turns out to be physically true, at least in the standard, flat spacetime situation. Furthermore, in some way $`𝒮`$ turns out to be well-defined in renormalizable theories.
In the rest of this section I will expose the basic ideas for the construction of $``$ .
For each $`m\{0\}𝕃^1`$ the spacetime geometry yields $`𝕃^3`$-scaled volume forms $`\eta _m=\mathrm{d}^3𝗉_{}`$ on the fibres of $`(\mathrm{T}^{}𝑴)^{}𝑻`$ , giving rise to equally scaled volume forms, denoted by the same symbols, on the fibres of $`𝑷_m𝑻`$.
Now for $`m^{},m^{\prime \prime },m^{\prime \prime \prime }\{0\}𝕃^1`$ consider the bundle ‘of three momenta’
$$𝑷_{\mathrm{}}:=𝑷_m^{}\underset{𝑴}{\times }𝑷_{m^{\prime \prime }}\underset{𝑴}{\times }𝑷_{m^{\prime \prime \prime }}𝑴,$$
and the section of *scaled* densities
$$\delta _{\mathrm{}}:𝑻𝕃^6𝓓(𝑷_{\mathrm{}})=𝕃^6𝓓(𝑷_m^{})\underset{𝑻}{}𝓓(𝑷_{m^{\prime \prime }})\underset{𝑻}{}𝓓(𝑷_{m^{\prime \prime \prime }}),$$
given by
$`\delta _{\mathrm{}},f`$ $`:={\displaystyle f(𝗉_{}^{},𝗉_{}^{\prime \prime },𝗉_{}^{}𝗉_{}^{\prime \prime })\eta _m^{}}\eta _{m^{\prime \prime }}={\displaystyle f(𝗉_{}^{},𝗉_{}^{\prime \prime },𝗉_{}^{}𝗉_{}^{\prime \prime })\mathrm{d}^3𝗉^{}\mathrm{d}^3𝗉^{\prime \prime }}`$
$`={\displaystyle f(𝗉_{}^{},𝗉_{}^{\prime \prime },𝗉_{}^{\prime \prime \prime })\delta (𝗉_{}^{}+𝗉_{}^{\prime \prime }+𝗉_{}^{\prime \prime \prime })\mathrm{d}^3𝗉^{}\mathrm{d}^3𝗉^{\prime \prime }\mathrm{d}^3𝗉^{\prime \prime \prime }},`$
where $`f𝓓_{}(𝑷_{\mathrm{}})`$ . It can be writen in the form
$$\delta _{\mathrm{}}=\stackrel{˘}{\delta }\eta _m^{}\eta _{m^{\prime \prime }}\eta _{m^{\prime \prime \prime }}=\stackrel{˘}{\delta }_{\mathrm{}}\mathrm{d}^3𝗉^{}\mathrm{d}^3𝗉^{\prime \prime }\mathrm{d}^3𝗉^{\prime \prime \prime },$$
where $`\stackrel{˘}{\delta }_{\mathrm{}}\delta (𝗉_{}^{}+𝗉_{}^{\prime \prime }+𝗉_{}^{\prime \prime \prime })`$ is an $`𝕃^3`$-valued generalized function.
Now one introduces the true (unscaled) generalized half-density
$$\underset{¯}{\mathrm{\Lambda }}:=\stackrel{˘}{\delta }_{\mathrm{}}\sqrt{\omega _m^{}}\sqrt{\omega _{m^{\prime \prime }}}\sqrt{\omega _{m^{\prime \prime \prime }}}:𝑻\text{ /}𝓓(𝑷_{\mathrm{}})=\text{ /}𝓓(𝑷_m^{})\underset{𝑻}{}\text{ /}𝓓(𝑷_{m^{\prime \prime }})\underset{𝑻}{}\text{ /}𝓓(𝑷_{m^{\prime \prime \prime }}),$$
which has the coordinate expression
$$\underset{¯}{\mathrm{\Lambda }}=\frac{\delta (𝗉_{}^{}+𝗉_{}^{\prime \prime }+𝗉_{}^{\prime \prime \prime })}{\sqrt{8𝗉_0^{}𝗉_0^{\prime \prime }𝗉_0^{\prime \prime \prime }}}\sqrt{\mathrm{d}^3𝗉^{}}\sqrt{\mathrm{d}^3𝗉^{\prime \prime }}\sqrt{\mathrm{d}^3𝗉^{\prime \prime \prime }}.$$
The fact that $`\underset{¯}{\mathrm{\Lambda }}`$ is unscaled, independently of the choice of a length unit, will turn out to be essential for its role in the quantum interaction; here it will describe the interaction of three particles, but clearly it can be readily generalized for describing the interaction of any number of particles. The different particle types are characterized by different complex 2-fibred bundles $`𝑽^{}𝑷_m^{}𝑴`$ and the like, and one must have a ‘classical interaction Lagrangian’ that is a scalar-valued 3-linear contraction among the fibres; this is a section
$$\mathrm{}{}_{\mathrm{int}}{}^{}:𝑷_{\mathrm{}}𝑽^{}{}_{}{}^{\mathrm{}}\underset{𝑷_{\mathrm{}}}{}𝑽^{\prime \prime }{}_{}{}^{\mathrm{}}\underset{𝑷_{\mathrm{}}}{}𝑽^{\prime \prime \prime }{}_{}{}^{\mathrm{}},$$
which can be seen as 3-linear fibred contraction. The structure of these bundles must allow for ‘index raising and lowering’, thus yielding a number of objects related to $`\mathrm{}_{\mathrm{int}}`$ and distinguished by various combinations of index types. Of course these arise in the easiest way when one has fibred Hermitian structures of the considered bundles (the fundamental case of electrodynamics, however, will be seen \[§10\] to be somewhat more involved). In particular, $`\mathrm{}{}_{\mathrm{int}}{}^{}:𝑷_{\mathrm{}}𝑽^{}\underset{𝑷_{\mathrm{}}}{}𝑽^{\prime \prime }\underset{𝑷_{\mathrm{}}}{}𝑽^{\prime \prime \prime }`$ .
Now one gets a section
$$\mathrm{\Lambda }:=\underset{¯}{\mathrm{\Lambda }}\mathrm{}{}_{\mathrm{int}}{}^{}:𝑻\text{ /}𝓓_𝑻(𝑷_{\mathrm{}},𝑽^{}\underset{𝑷_{\mathrm{}}}{}𝑽^{\prime \prime }\underset{𝑷_{\mathrm{}}}{}𝑽^{\prime \prime \prime })𝓥^{}{}_{}{}^{1}\underset{𝑻}{}𝓥^{\prime \prime }{}_{}{}^{1}\underset{𝑻}{}𝓥^{\prime \prime \prime }{}_{}{}^{1},$$
where $`𝓥^{}{}_{}{}^{1}:=\text{ /}𝓓_𝑻(𝑷_{\mathrm{}},𝑽^{})`$ and the like. The essential idea of the quantum interaction is then the following: make $`\mathrm{\Lambda }`$ act in the fibres of the Fock bundle $`𝓥𝓥^{}\underset{𝑻}{}𝓥^{\prime \prime }\underset{𝑻}{}𝓥^{\prime \prime \prime }𝑻`$ by using each one of its tensor factors either as ‘absorption’ (contraction) or as ‘creation’ (tensor product). However, a fundamental issue is immediately apparent (and will be furtherly discussed later on): in general, this action is only well-defined on the subbundle $`𝓥_{}𝓥`$ of test elements, so actually it gives rise to a morphism $`𝓥_{}𝓥`$ whose extendibility will have to be carefully examined.
The various ‘index types’ of $`\mathrm{}_{\mathrm{int}}`$ correspond to the various actions performed by the corresponding tensor factors: a covariant index determines a particle absorption, a contravariant index determines a particle creation. Furthermore one considers different types of $`\underset{¯}{\mathrm{\Lambda }}`$ , each one to be coupled to a corresponding type of $`\mathrm{}_{\mathrm{int}}`$ and obtained by changing the *sign* of the momenta in the $`\delta `$ generalized function. So, for example, the type of $`\mathrm{}_{\mathrm{int}}`$ which is a section $`𝑷_{\mathrm{}}𝑽^{}{}_{}{}^{\mathrm{}}\underset{𝑷_{\mathrm{}}}{}𝑽^{\prime \prime }\underset{𝑷_{\mathrm{}}}{}𝑽^{\prime \prime \prime }`$ (the first factor is an absorption factor, the second and third are creation factors) is tensorialized by the ‘version’ of $`\underset{¯}{\mathrm{\Lambda }}`$ which has $`\delta (𝗉_{}^{}+𝗉_{}^{\prime \prime }+𝗉_{}^{\prime \prime \prime })`$ in its coordinate expression. In practice, I find it convenient using a ‘generalized index’ notation in which generalized indices are either high or low, and repeated momentum indices are to interpreted as integration indices (just as repeated ordinary indices are interpreted as ordinary summation indices). So, for $`f=\stackrel{˘}{f}\sqrt{\mathrm{d}^3𝗉^{}}\sqrt{\mathrm{d}^3𝗉^{\prime \prime }}\sqrt{\mathrm{d}^3𝗉^{\prime \prime \prime }}𝓥_{}`$ , I’ll write
$`f=f_{𝗉^{}𝗉^{\prime \prime }𝗉^{\prime \prime \prime }}𝖡^𝗉^{}𝖡^{𝗉^{\prime \prime }}𝖡^{𝗉^{\prime \prime \prime }},`$ $`f_{𝗉^{}𝗉^{\prime \prime }𝗉^{\prime \prime \prime }}=l^{9/2}\stackrel{˘}{f}(𝗉^{},𝗉^{\prime \prime },𝗉^{\prime \prime \prime }),`$
$`\underset{¯}{\mathrm{\Lambda }}=\underset{¯}{\mathrm{\Lambda }}^{𝗉^{}𝗉^{\prime \prime }𝗉^{\prime \prime \prime }}𝖡_𝗉^{}𝖡_{𝗉^{\prime \prime }}𝖡_{𝗉^{\prime \prime \prime }},`$ $`\underset{¯}{\mathrm{\Lambda }}^{𝗉^{}𝗉^{\prime \prime }𝗉^{\prime \prime \prime }}={\displaystyle \frac{\delta (𝗉_{}^{}+𝗉_{}^{\prime \prime }+𝗉_{}^{\prime \prime \prime })}{\sqrt{8l^9𝗉_0^{}𝗉_0^{\prime \prime }𝗉_0^{\prime \prime \prime }}}}`$
(where $`𝖡^𝗉𝖡_𝗉`$), to be intended as
$$\underset{¯}{\mathrm{\Lambda }},f=\underset{¯}{\mathrm{\Lambda }}^{𝗉^{}𝗉^{\prime \prime }𝗉^{\prime \prime \prime }}f_{𝗉^{}𝗉^{\prime \prime }𝗉^{\prime \prime \prime }}=\frac{\delta (𝗉_{}^{}+𝗉_{}^{\prime \prime }+𝗉_{}^{\prime \prime \prime })}{\sqrt{8𝗉_0^{}𝗉_0^{\prime \prime }𝗉_0^{\prime \prime \prime }}}\stackrel{˘}{f}(𝗉^{},𝗉^{\prime \prime },𝗉^{\prime \prime \prime })\mathrm{d}^3𝗉^{}\mathrm{d}^3𝗉^{\prime \prime }\mathrm{d}^3𝗉^{\prime \prime \prime }$$
(remark: the generalized index summation is performed via the unscaled volume form $`l^3\mathrm{d}^3𝗉_{}`$).
Analogously, the various index types of $`\underset{¯}{\mathrm{\Lambda }}`$ (there are 8 of them) can be written as
$$\underset{¯}{\mathrm{\Lambda }}_𝗉^{}{}_{}{}^{𝗉^{\prime \prime }𝗉^{\prime \prime \prime }}𝖡_{}^{𝗉^{}}𝖡_{𝗉^{\prime \prime }}𝖡_{𝗉^{\prime \prime \prime }},\underset{¯}{\mathrm{\Lambda }}_{𝗉^{}𝗉^{\prime \prime }}{}_{}{}^{𝗉^{\prime \prime \prime }}𝖡_{}^{𝗉^{}}𝖡^{𝗉^{\prime \prime }}𝖡_{𝗉^{\prime \prime \prime }},\mathrm{}\text{etcetera},$$
where
$$\underset{¯}{\mathrm{\Lambda }}_𝗉^{}{}_{}{}^{𝗉^{\prime \prime }𝗉^{\prime \prime \prime }}=\frac{\delta (𝗉_{}^{}+𝗉_{}^{\prime \prime }+𝗉_{}^{\prime \prime \prime })}{\sqrt{8𝗉_0^{}𝗉_0^{\prime \prime }𝗉_0^{\prime \prime \prime }}},\underset{¯}{\mathrm{\Lambda }}_{𝗉^{}𝗉^{\prime \prime }}{}_{}{}^{𝗉^{\prime \prime \prime }}=\frac{\delta (𝗉_{}^{}𝗉_{}^{\prime \prime }+𝗉_{}^{\prime \prime \prime })}{\sqrt{8𝗉_0^{}𝗉_0^{\prime \prime }𝗉_0^{\prime \prime \prime }}},$$
and so on. Correspondingly, the various types of $`\mathrm{\Lambda }`$ can be written as
$`\mathrm{\Lambda }^{𝗉^{}A^{},𝗉^{\prime \prime }A^{\prime \prime },𝗉^{\prime \prime \prime }A^{\prime \prime \prime }}𝖡_{𝗉^{}A^{}}𝖡_{𝗉^{\prime \prime }A^{\prime \prime }}𝖡_{𝗉^{\prime \prime \prime }A^{\prime \prime \prime }},`$ $`\mathrm{\Lambda }^{𝗉^{}A^{},𝗉^{\prime \prime }A^{\prime \prime },𝗉^{\prime \prime \prime }A^{\prime \prime \prime }}=\mathrm{\Lambda }^{𝗉^{}𝗉^{\prime \prime }𝗉^{\prime \prime \prime }}(\mathrm{}{}_{\mathrm{int}}{}^{})^{A^{}A^{\prime \prime }A^{\prime \prime \prime }},`$
$`\mathrm{\Lambda }_{𝗉^{}A^{}}{}_{}{}^{𝗉^{\prime \prime }A^{\prime \prime },𝗉^{\prime \prime \prime }A^{\prime \prime \prime }}𝖡_{}^{𝗉^{}A^{}}𝖡_{𝗉^{\prime \prime }A^{\prime \prime }}𝖡_{𝗉^{\prime \prime \prime }A^{\prime \prime \prime }},`$ $`\mathrm{\Lambda }_{𝗉^{}A^{}}{}_{}{}^{𝗉^{\prime \prime }A^{\prime \prime },𝗉^{\prime \prime \prime }A^{\prime \prime \prime }}=\mathrm{\Lambda }_𝗉^{}{}_{}{}^{𝗉^{\prime \prime }𝗉^{\prime \prime \prime }}(\mathrm{}{}_{\mathrm{int}}{}^{})_{A^{}}^{}{}_{}{}^{A^{\prime \prime }A^{\prime \prime \prime }},`$
$`\mathrm{}\mathrm{}`$ $`\mathrm{}\mathrm{}`$
$`\mathrm{\Lambda }_{𝗉^{}A^{},𝗉^{\prime \prime }A^{\prime \prime },𝗉^{\prime \prime \prime }A^{\prime \prime \prime }}𝖡^{𝗉^{}A^{}}𝖡^{𝗉^{\prime \prime }A^{\prime \prime }}𝖡^{𝗉^{\prime \prime \prime }A^{\prime \prime \prime }},`$ $`\mathrm{\Lambda }_{𝗉^{}A^{},𝗉^{\prime \prime }A^{\prime \prime },𝗉^{\prime \prime \prime }A^{\prime \prime \prime }}=\mathrm{\Lambda }_{𝗉^{}𝗉^{\prime \prime }𝗉^{\prime \prime \prime }}(\mathrm{}{}_{\mathrm{int}}{}^{})_{A^{}A^{\prime \prime }A^{\prime \prime \prime }}.`$
The morphism $`:𝓥_{}𝕃^1𝓥`$ is essentially a sum whose terms are the various types of $`\mathrm{\Lambda }`$ , with a further ingredient: each term also has a factor
$$\lambda \mathrm{e}^{\mathrm{i}(\pm 𝗉_0^{}\pm 𝗉_0^{\prime \prime }\pm 𝗉_0^{\prime \prime \prime })𝗍},$$
where $`\lambda 𝕃^1`$ is a constant; the signs in the exponential match those of the corresponding spatial momenta. Then $`\mathrm{i}\mathrm{d}𝗍:𝑻𝓥_{}𝓥\mathrm{T}^{}𝑻`$ is the interaction term which modifies the free-particle connection.
The reader will note that, according to the setting above sketched, the elements $`𝖡_{𝗉A}`$ and $`𝖡^{𝗉A}`$ in the various generalized frames can be thought of, essentially, as the usual creation and absorption operators. Moreover one could obtain further types of $`\mathrm{\Lambda }`$ by exchanging tensor factors; this only make a difference if particles of the same type are involved, and is settled by considering only those terms in which the creation operators stand on the right.
## 6 Scalar particles
Let’s see how, in practice, the somewhat sketchy ideas exposed in §5 can be implemented in the simplest case. Many of the arguments used in this section are more or less standard, the point is to show how they arise from a not-so-standard approach.
Consider a model of two types of scalar particles, one of mass $`m`$ and one massless, with one-particle state bundles $`𝓥^{}{}_{}{}^{1}𝓟_m`$ and $`𝓥^{\prime \prime }{}_{}{}^{1}𝓟__0`$ and generalized frames $`\{𝖠_p\}`$ , $`p𝑷_m`$ , and $`\{𝖡_k\}`$ , $`k𝑷__0`$ . The classical interaction is assumed to be just a constant $`\mathrm{}𝕃^1`$, and it incorporates the $`\lambda `$ introduced above.
At first-order, the formal series expression of the scattering operator is $`𝒮=11+𝒮_1`$ with $`𝒮_1=\mathrm{i}_{\mathrm{}}^+\mathrm{}(𝗍)d𝗍`$ . In terms of generalized index notation, one says that $`𝒮_1`$ has a ‘matrix element’
$$(𝒮_1)_{\mathrm{𝗉𝗊}}^𝗄=\frac{\mathrm{i}\mathrm{}\delta (𝗉_{}+𝗄_{}𝗊_{})}{\sqrt{8l^9𝗄_0𝗉_0𝗊_0}}\underset{\mathrm{}}{\overset{+\mathrm{}}{}}\mathrm{e}^{\mathrm{i}(𝗄_0𝗉_0𝗊_0)}d𝗍=\frac{2\pi \mathrm{i}\mathrm{}\delta (𝗄𝗉𝗊)}{\sqrt{8l^9𝗄_0𝗉_0𝗊_0}}.$$
There are eight matrix elements of this kind, each one describing *one-point interaction* and labelled by an elementary Feynman graph (time running *upwards*):
$`(𝒮_1)_{\mathrm{𝗉𝗊𝗄}}={\displaystyle \frac{2\pi \mathrm{i}\mathrm{}\delta (𝗉𝗄𝗊)}{\sqrt{8l^9𝗉_0𝗊_0𝗄_0}}},`$ $`(𝒮_1)_{\mathrm{𝗊𝗄}}^𝗉={\displaystyle \frac{2\pi \mathrm{i}\mathrm{}\delta (𝗉𝗊𝗄)}{\sqrt{8l^9𝗉_0𝗊_0𝗄_0}}},`$
$`(𝒮_1)_{\mathrm{𝗉𝗊}}^𝗄={\displaystyle \frac{2\pi \mathrm{i}\mathrm{}\delta (𝗄𝗉𝗊)}{\sqrt{8l^9𝗄_0𝗉_0𝗊_0}}},`$ $`(𝒮_1)_{\mathrm{𝗉𝗄}}^𝗊={\displaystyle \frac{2\pi \mathrm{i}\mathrm{}\delta (𝗊𝗉𝗄)}{\sqrt{8l^9𝗄_0𝗉_0𝗊_0}}},`$
$`(𝒮_1)_𝗊^{\mathrm{𝗄𝗉}}={\displaystyle \frac{2\pi \mathrm{i}\mathrm{}\delta (𝗄+𝗉𝗊)}{\sqrt{8l^9𝗄_0𝗉_0𝗊_0}}},`$ $`(𝒮_1)_𝗄^{\mathrm{𝗉𝗊}}={\displaystyle \frac{2\pi \mathrm{i}\mathrm{}\delta (𝗉+𝗊𝗄)}{\sqrt{8l^9𝗄_0𝗉_0𝗊_0}}},`$
$`(𝒮_1)_𝗉^{\mathrm{𝗄𝗊}}={\displaystyle \frac{2\pi \mathrm{i}\mathrm{}\delta (𝗄+𝗊𝗉)}{\sqrt{8l^9𝗄_0𝗉_0𝗊_0}}},`$ $`(𝒮_1)^{\mathrm{𝗉𝗊𝗄}}={\displaystyle \frac{2\pi \mathrm{i}\mathrm{}\delta (𝗉+𝗊+𝗄)}{\sqrt{8l^9𝗄_0𝗉_0𝗊_0}}}.`$
Propagators arise when one consideris second-order matrix elements, representing processes described, for example, by the diagrams
Here one has two types of second order processes whose initial and final states contain two massive particles. These types are labelled as (I) and (II), and each of them comes in two subtypes, distinguished by the time order of the two interactions involved and respectively labelled as (I’) and (I”), (II’) and (II”). By considering the form of the interaction, one sees that the first diagram yields a contribution
$`(𝒮_\mathrm{I}^{})_{\mathrm{𝗉𝗊}}^{𝗉^{}𝗊^{}}`$ $`={\displaystyle \frac{\mathrm{}^2}{l^6}}{\displaystyle \underset{\mathrm{}}{\overset{+\mathrm{}}{}}}\mathrm{d}𝗍_2{\displaystyle \underset{\mathrm{}}{\overset{+\mathrm{}}{}}}\mathrm{d}𝗍_1{\displaystyle }\mathrm{d}^3𝗄\mathrm{H}(𝗍_2𝗍_1){\displaystyle \frac{\delta (𝗉_{}^{}+𝗊_{}^{}𝗄_{})\delta (𝗄_{}𝗉_{}𝗊_{})}{\sqrt{16𝗉_0^{}𝗊_0^{}𝗉_0𝗊_0}\mathrm{\hspace{0.25em}2}𝗄_0}}`$
$`\mathrm{e}^{\mathrm{i}(𝗉_0+𝗄_0𝗊_0)𝗍_1}\mathrm{e}^{\mathrm{i}(𝗉_0^{}𝗄_0+𝗊_0^{})𝗍_2},`$
where $`\mathrm{H}`$ is the Heaviside function, arising from the explicit expression of the time-ordered product (which also yields two identical terms, so the $`1/2!`$ factor in the $`𝒮`$ series cancels out).
Now one proceeds essentially in a more or less standard way. First one uses a technical result: if $`\phi `$ is a test function on any fibre of $`𝑷_m𝑻`$, then
$`{\displaystyle \mathrm{d}^3𝗄_{}\mathrm{H}(t)\mathrm{e}^{\pm \mathrm{i}𝗍\mathrm{E}_m(𝗄_{})}\phi (𝗄_{})}`$ $`={\displaystyle \frac{1}{2\pi \mathrm{i}}}\underset{\epsilon 0^+}{lim}{\displaystyle \mathrm{d}^4𝗄\frac{\mathrm{e}^{\mathrm{i}𝗍𝗄_0}}{𝗄_0\mathrm{E}_m(𝗄_{})\mathrm{i}\epsilon }\phi (𝗄_{})}=`$
$`={\displaystyle \frac{1}{2\pi \mathrm{i}}}\underset{\epsilon 0^+}{lim}{\displaystyle \mathrm{d}^4𝗄\frac{\mathrm{e}^{\mathrm{i}𝗍𝗄_0}}{𝗄_0\mathrm{E}_m(𝗄_{})\mathrm{i}\epsilon }\phi (𝗄_{})}`$
(the proof uses the integral representation $`2\pi \mathrm{i}\mathrm{H}(𝗍)=lim_{\epsilon 0^+}\underset{\mathrm{}}{\overset{+\mathrm{}}{}}\frac{\mathrm{e}^{\mathrm{i}𝗍\tau }}{\tau \mathrm{i}\epsilon }d\tau `$ and some integration variable changes). Eventually
$$(𝒮_\mathrm{I}^{})_{\mathrm{𝗉𝗊}}^{𝗉^{}𝗊^{}}=\frac{2\pi \mathrm{i}\mathrm{}^2}{l^6\sqrt{16𝗉_0^{}𝗊_0^{}𝗉_0𝗊_0}}\mathrm{d}^4𝗄\frac{\delta (𝗉^{}+𝗊^{}𝗄)\delta (𝗄𝗉𝗊)}{2|𝗄_{}|(𝗄_0+|𝗄_{}|\mathrm{i}\epsilon )}$$
(the limit for $`\epsilon 0^+`$ is intended). So, by a technical trick, an integral over $`𝑻\times (𝑷_m)_{t_0}`$ , $`t_0𝑻`$ , was transformed into an integral over a whole space $`\mathrm{T}_{t_0}^{}𝑴`$ of 4-momenta (the momentum of the intermediate particle is ‘off-shell’).
The calculation relative to the diagram $`\mathrm{I}^{\prime \prime }`$ is similar but one has a few sign differences, getting
$$(𝒮_{\mathrm{I}^{\prime \prime }})_{\mathrm{𝗉𝗊}}^{𝗉^{}𝗊^{}}=\frac{2\pi \mathrm{i}\mathrm{}^2}{l^6\sqrt{16𝗉_0^{}𝗊_0^{}𝗉_0𝗊_0}}\mathrm{d}^4𝗄\frac{\delta (𝗉^{}+𝗊^{}𝗄)\delta (𝗄𝗉𝗊)}{2|𝗄_{}|(𝗄_0+|𝗄_{}|\mathrm{i}\epsilon )}.$$
Finally,
$$(𝒮_\mathrm{I})_{\mathrm{𝗉𝗊}}^{𝗉^{}𝗊^{}}=(𝒮_\mathrm{I}^{})_{\mathrm{𝗉𝗊}}^{𝗉^{}𝗊^{}}+(𝒮_{\mathrm{I}^{\prime \prime }})_{\mathrm{𝗉𝗊}}^{𝗉^{}𝗊^{}}=\frac{2\pi \mathrm{i}\mathrm{}^2}{l^6\sqrt{16𝗉_0^{}𝗊_0^{}𝗉_0𝗊_0}}\mathrm{d}^4𝗄\frac{\delta (𝗉^{}+𝗊^{}𝗄)\delta (𝗄𝗉𝗊)}{g(𝗄,𝗄)+\mathrm{i}\epsilon }.$$
In case II one finds exactly the same result.
When the intermediate particle is massive one gets a similar expression, with $`g(𝗄,𝗄)`$ in the propagator’s expression replaced by $`g(𝗉,𝗉)m^2`$ .
A word is due about the ‘infinities’ arising when one considers diagrams containing loops, as for example
etc.
Doing the calculation in the first instance (say) one has a contribution to the scattering matrix which, apart from constant factors, is given by the integral
$$\mathrm{d}^4𝗄\frac{\delta (𝗉+𝗊+𝗄)\delta (𝗊𝗄+𝗉^{})}{4\mathrm{E}_m(𝗊_{})|𝗄_{}|(𝗄_0+|𝗄_{}|\mathrm{i}\epsilon )}.$$
Now if this were a well-defined distribution $`\varphi `$ in two variables, then $`u\varphi v`$ should be a (finite) number, but it is immediate to check that it is not. Similar results are found in the other cases.
## 7 Electron and positron free states
The 4-spinor bundle is a complex vector bundle $`𝑾𝑴`$ with 4-dimensional fibres, endowed with a *scaled Clifford morphism* (*Dirac map*)
$$\gamma :\mathrm{T}𝑴𝕃\mathrm{End}(𝑾):v\gamma [v]$$
over $`𝑴`$ and a Hermitian metric $`\mathrm{k}`$ on the fibres fulfilling
$$\mathrm{k}(\gamma [v]\varphi ,\psi )=\mathrm{k}(\varphi ,\gamma [v]\psi ),v\mathrm{T}_x𝑴,\varphi ,\psi 𝑾_x,x𝑴.$$
Then $`\mathrm{k}`$ (which yields the *Dirac adjoint* anti-isomorphism $`\psi \mathrm{k}^{\mathrm{}}(\psi )`$ , usually denoted as $`\psi \overline{\psi }`$) turns out to have the signature $`(+,+,,)`$. If $`p:𝑴𝑷_m`$ then
$$𝑾=𝑾_p^+\underset{𝑴}{}𝑾_p^{},𝑾_p^\pm :=\mathrm{Ker}(\gamma [p^\mathrm{\#}]m),$$
where $`p^\mathrm{\#}g^\mathrm{\#}(p):𝑴𝕃^2\mathrm{T}𝑴`$ is the contravariant form of $`p`$ . The restrictions of $`\mathrm{k}`$ to these two subbundles turn out to have the signatures $`(+,+)`$ and $`(,)`$ , respectively.
Now for each $`m\{0\}𝕃^1`$ one is led to consider the 2-fibred bundles $`𝑾_m^\pm 𝑷_m𝑴`$ defined by
$$𝑾_m^\pm :=\underset{p𝑷_m}{}𝑾_p^\pm 𝑷_m\underset{𝑴}{\times }𝑾.$$
The 4-spinor bundle is also endowed<sup>6</sup><sup>6</sup>6See \[CJ97a, C00b\] for a review of the geometry 4-spinors and 2-spinors for electrodynamics and other field theories. with a *spinor connection* $`\text{ ̵}\mathrm{\Gamma }`$, strictly related to the spacetime connection, such that $`\gamma `$ and $`\mathrm{k}`$ are covariantly constant. It is easy to see<sup>7</sup><sup>7</sup>7 If $`p:𝑴𝑷_m`$ and $`\psi :𝑴𝑾`$ are parallely transported along some curve in $`𝑴`$, then $`\gamma [p^\mathrm{\#}]\psi m\psi `$ is also parallely transported along the same curve; so it vanishes along the curve if it vanishes at any one point of the curve. that $`\mathrm{\Gamma }_m`$ and $`\text{ ̵}\mathrm{\Gamma }`$ determine projectable connections of $`𝑾_m^\pm 𝑷_m𝑴`$.
Now if $`𝑻𝑴`$ is a detector, then a free one-electron state is defined to be a covariantly constant section $`𝑻𝑾_m^+`$ . Namely, a free one-electron state is determined by a covariantly constant section $`p:𝑻𝑷_m`$ and by a covariantly constant section $`\psi :𝑻𝑾`$ such that $`\psi (t)𝑾_{p(t)}^+`$ for each $`t𝑻`$. On the other hand, a free one-positron state will be represented as a covariantly constant section $`𝑻\overline{𝑾}_m^{}`$ .<sup>8</sup><sup>8</sup>8 If $`𝑽`$ is a finite-dimensional complex vector space then its *conjugate space* can be defined as $`\overline{𝑽}:=𝑽^\mathrm{}\overline{\mathrm{}}𝑽^\overline{\mathrm{}}\mathrm{}`$, where $`𝑽^{\mathrm{}}`$ and $`𝑽^\overline{\mathrm{}}`$ are, respectively, the $``$-dual and antidual spaces, that is the spaces of linear and antilinear maps $`𝑽`$ . There is an anti-isomorphism $`𝑽\overline{𝑽}:v\overline{v}`$ . The indices relative to a conjugate basis are distinguished by a dot. For brevity, these 2-fibred bundles and the related 1-particle state quantum bundles (of vector-valued generalized half-densities) are denoted as
$`𝑭𝑾_m^+𝑷_m𝑻,`$ $`\stackrel{~}{𝑭}\overline{𝑾}_m^{}𝑷_m𝑻,`$
$`𝓕^1:=\text{ /}𝓓_𝑻(𝑷_m,𝑭)\text{ /}𝓓_𝑻(𝑷_m,𝑾_m^+),`$ $`\stackrel{~}{𝓕}{}_{}{}^{1}:=\text{ /}𝓓_𝑻(𝑷_m,\stackrel{~}{𝑭})\text{ /}𝓓_𝑻(𝑷_m,\overline{𝑾}_m^{}).`$
In order to introduce appropriate generalized frames for free electron and positron states, one needs, for each $`p(𝑷_m)_𝑻`$ a frame
$$(𝗎_A(p),𝗏_A(p)),A=1,2$$
of $`𝑾_p`$ which is adapted to the splitting $`𝑾_p=𝑾_p^+𝑾_p^{}`$ . A consistent choice can be made by extending usual procedure of the flat inertial case.<sup>9</sup><sup>9</sup>9 At some point in $`𝑻`$ one fixes a spinor frame adapted to the splitting determined by the unit vector $`\tau _0`$ , and Fermi transports it along $`𝑻`$ ; then, in each fibre, one takes the unique boost sending $`\tau _0`$ to $`p^\mathrm{\#}/m`$ ; up to sign (which can be fixed by continuity) this boost transforms the given spinor frame to the desired one. Now one gets the generalized frames
$`𝖠_{pA}:=𝖠_p𝗎_A(p):𝑻𝓕^1\text{ /}𝓓{}_{𝑻}{}^{}(𝑷_m,𝑾_m^+),`$
$`𝖢_{pA\dot{}}:=𝖠_p\overline{𝗏}_{A\dot{}}(p):𝑻\stackrel{~}{𝓕}{}_{}{}^{1}\text{ /}𝓓{}_{𝑻}{}^{}(𝑷_m,\overline{𝑾}_m^{}),`$
respectively for electrons and positrons.
An important technical result, which is proved by elementary linear algebra, is
$`𝗎_A(p)𝗎^A(p)=\frac{1}{2}(11+\gamma [p^\mathrm{\#}/m]):𝑾𝑾_p^+,`$
$`𝗏_A(p)𝗏^A(p)=\frac{1}{2}(11\gamma [p^\mathrm{\#}/m]):𝑾𝑾_p^{}.`$
## 8 Photon free states
For brevity, henceforth I will use the shorthand $`𝑯𝕃^1\mathrm{T}𝑴`$, so that $`𝑯^{}𝕃\mathrm{T}^{}𝑴`$ and the spacetime metric $`g`$ is an *unscaled* (i.e. ‘confomally invariant’) Lorentz metric in the fibres of $`𝑯𝑴`$ .
Remember that $`𝑷__0\mathrm{T}^{}𝑴`$ denotes the subbbundle over $`𝑴`$ of future null half-cones in the fibres of $`\mathrm{T}^{}𝑴`$. Consider the 2-fibred bundle $`𝑯_0𝑷__0𝑴`$ whose fibre over any $`k(𝑷__0)_x`$ , $`x𝑴`$, is the 3-dimensional real vector space
$$(𝑯_0)_k:=\{\alpha 𝑯^{}:g^\mathrm{\#}(\alpha ,k)=0\}.$$
Then $`𝑯_0𝑷__0\underset{𝑴}{\times }𝑯^{}`$ (but note that $`𝑯_0`$ itself is *not* a ‘semi-trivial’ bundle of the type $`𝑷__0\underset{𝑴}{\times }𝒁`$ ). Next, consider the real vector bundle $`𝑩_{}^{}𝑷__0`$ whose fibre over any $`k𝑷__0`$ is the (2-dimensional) quotient space
$$(𝑩_{})_k^{}:=(𝑯_0)_k/kk^{}/k,$$
where $`k`$ denotes the vector space generated by $`k`$ . Moreover, $`𝑩_{}𝑩_{}^{}`$ can be equivalently introduced by a similar contravariant construction.
It turns out that the spacetime metric ‘passes to the quotient’, so it naturally determines a negative metric $`g_𝑩`$ in the fibres of $`𝑩_{}𝑷__0`$ , as well as a ‘Hodge’ isomorphism $`_𝑩`$ which can be characterized through the rule<sup>10</sup><sup>10</sup>10 $`k\beta `$ is well defined because $`\beta `$ is an equivalence class of covectors differing for a term proportional to $`k`$ .
$$(k\beta )=k(_𝑩\beta ).$$
Now define the *optical bundle*<sup>11</sup><sup>11</sup>11 In the literature this term is often used in a somewhat different (but related) sense, denoting a vector bundle over $`𝑴`$ associated with the choice of a congruence of null lines \[Nu96\]. to be the 2-fibred bundle
$$𝑩:=𝑩_{}𝑷__0𝑴.$$
This has the canonical splitting
$$𝑩=𝑩^+\underset{𝑷__0}{}𝑩^{},$$
where the fibres of the subbundles $`𝑩^\pm 𝑷__0`$ are defined to be the eigenspaces of $`\mathrm{i}_𝑩`$ with eigenvalues $`\pm 1`$ (and turn out to be 1-dimensional complex $`g_𝑩`$-null subspaces).
Let $`u:𝑴\mathrm{T}𝑴`$ be any given ‘observer’, i.e. a unit timelike vector field on $`𝑴`$ ; through $`u`$ one can identify $`𝑩_{}𝑷__0`$ with $`𝑯_0u^{}𝑷__0`$ (‘radiation gauge’). Take any $`k(𝑷__0)_x`$ , $`x𝑴`$, and let $`(𝖾_\lambda )`$ , $`\lambda =0,1,2,3`$, be an orthonormal basis of $`𝑯_x`$ such that $`𝖾_0u(x)`$ and $`k^\mathrm{\#}𝖾_0+𝖾_3`$ ; then the basis
$$(𝖻_1,𝖻_2)(𝖻_+,𝖻_{}):=(\frac{1}{\sqrt{2}}(𝖾_1+\mathrm{i}𝖾_2),\frac{1}{\sqrt{2}}(𝖾_1\mathrm{i}𝖾_2))(𝑯_0u^{})$$
is *adapted* to the splitting of $`𝑩_k`$ , i.e. $`𝖻_\pm 𝑩_k^\pm `$ . In this way, locally one can construct smooth frames of $`𝑯\underset{𝑴}{\times }𝑷__0`$ , and smooth frames of $`𝑩𝑷__0`$ which are adapted to its splitting.
Let now $`\{𝖡_k\}`$ be the generalized frame of the quantum bundle $`𝓟_0𝑴`$ defined as usual; then one gets a generalized frame
$$\{𝖡_{\kappa Q}\}:=\{𝖡_k𝖻_Q(k)\},Q=1,2,$$
of the quantum bundle
$$𝓑^1:=\text{ /}𝓓_𝑴(𝑷__0,𝑩)𝑴.$$
In particular, all the above bundles and constructions can be restricted to a detector $`𝑻𝑴`$. In that case, $`𝖾_0`$ will be chosen to be the unit future-pointing vector tangent to $`𝑻`$.
Free asymptotic 1-photon states will be described as covariantly constant sections $`𝑻𝓑^1`$ (they only possess *transversal polarization modes*). Virtual photons, on the other hand, span a larger bundle; they are described as covariantly constant sections
$$𝑻\stackrel{~}{𝓑}^1\text{ /}𝓓{}_{𝑻}{}^{}(𝑷__0,𝑯),$$
where one uses the generalized frame
$$\{𝖡_{\kappa \lambda }\}:=\{𝖡_k𝖾_\lambda \},\lambda =0,1,2,3.$$
## 9 Electromagnetic interaction
The ‘classical electromagnetic interaction’ is the 3-linear morphism
$`\mathrm{}_{\mathrm{int}}`$ $`:\overline{𝑾}\underset{𝑴}{\times }𝑯\underset{𝑴}{\times }𝑾`$
$`:(\varphi ,b,\psi )e\mathrm{k}(\varphi ,\gamma [b]\psi )e\mathrm{k}^{\mathrm{}}\varphi ,\gamma [b]\psi ,`$
where $`e^+`$ is the positron’s charge (a pure number in natural units).
As sketched in §5, this geometric structure of the underlying classical bundles, together with the generalized half-density $`\underset{¯}{\mathrm{\Lambda }}`$ , determines the quantum interaction $`\mathrm{i}`$ . A short discussion is needed in order to see how the index types of the various terms in $``$ arise.
First, $`\mathrm{}_{\mathrm{int}}`$ is extended to $`\overline{𝑾}\underset{𝑴}{\times }𝑯_{}\underset{𝑴}{\times }𝑾`$, where $`𝑯_{}𝑯`$. In the fibres of the complex vector bundle $`𝑯_{}𝑴`$ indices are raised and lowered through the obvious extension of the spacetime metric $`g`$ , while in the fibres of $`𝑩𝑷__0`$ one uses $`g_𝑩`$ ; when an observer is chosen and one works in the radiation gauge, the latter operation can be viewed essentially as a restriction of the former.
Now $`\mathrm{}_{\mathrm{int}}`$ can be seen as a $``$-linear function on the fibres of
$$\overline{𝑾}\underset{𝑷_{\mathrm{}}}{}𝑯_{}\underset{𝑷_{\mathrm{}}}{}𝑾𝑷_{\mathrm{}}𝑷_m\underset{𝑴}{\times }𝑷__0\underset{𝑴}{\times }𝑷_m.$$
Note that the isomorphism $`\mathrm{k}^{\mathrm{}}:\overline{𝑾}𝑾^{\mathrm{}}`$ induced by the Hermitian metric $`\mathrm{k}`$ preserves the splitting $`𝑾\underset{𝑴}{\times }𝑷_m=𝑾_m^+\underset{𝑷_m}{}𝑾_m^{}`$ , namely $`\mathrm{k}^{\mathrm{}}:\overline{𝑾}_m^\pm (𝑾_m^\pm )^{\mathrm{}}`$ . Then
$`𝑷_m\underset{𝑴}{\times }\overline{𝑾}=\overline{𝑾}_m^+\underset{𝑷_m}{}\overline{𝑾}_m^{}(𝑾_m^+)^{\mathrm{}}\underset{𝑷_m}{}\overline{𝑾}_m^{},`$
$`𝑷_m\underset{𝑴}{\times }𝑾=𝑾_m^+\underset{𝑷_m}{}𝑾_m^{}𝑾_m^+\underset{𝑷_m}{}(\overline{𝑾}_m^{})^{\mathrm{}}.`$
When $`𝑷_m\underset{𝑴}{\times }𝑾`$ and $`𝑷_m\underset{𝑴}{\times }\overline{𝑾}`$ are written in this way, the coordinate expression of
$$\mathrm{}{}_{\mathrm{int}}{}^{}:\left((𝑾_m^+)^{\mathrm{}}\underset{𝑷_m}{}\overline{𝑾}_m^{}\right)\underset{𝑷_{\mathrm{}}}{}𝑯_{}\underset{𝑷_{\mathrm{}}}{}\left(𝑾_m^+\underset{𝑷_m}{}(\overline{𝑾}_m^{})^{\mathrm{}}\right)$$
contains four terms with different index types; all dotted (i.e. ‘conjugated’) indices, either high or low, refer to the positron bundle $`\overline{𝑾}_m^{}`$ or to its dual, while undotted indices refer to the electron bundle or to its dual. Finally, a further extension through $`g`$ gives
$$\mathrm{}{}_{\mathrm{int}}{}^{}:\left((𝑾_m^+)^{\mathrm{}}\underset{𝑷_m}{}\overline{𝑾}_m^{}\right)\underset{𝑷_{\mathrm{}}}{}\left(𝑯_{}\underset{𝑷__0}{}𝑯_{}^{\mathrm{}}\right)\underset{𝑷_{\mathrm{}}}{}\left(𝑾_m^+\underset{𝑷_m}{}(\overline{𝑾}_m^{})^{\mathrm{}}\right),$$
which is the sum of *eight* terms of different index types. Explicitely, if $`\beta 𝑯_{}`$ then one replaces $`\gamma [b]`$ in the expression of $`\mathrm{}_{\mathrm{int}}`$ with
$$\gamma ^\mathrm{\#}[\beta ]:=\gamma [\beta \mathrm{\#}]\gamma [g^\mathrm{\#}\beta ].$$
Further objects can be obtained by exchanging tensor factors in $`\mathrm{}_{\mathrm{int}}`$ . However, objects only distinguished for a different order of indices referring to different particle types are regarded as equivalent, while in the different ordering of indices referring to the same particle type only those terms are retained which have the covariant indices *on the right* of the contravariant ones.
Let now $`𝑻𝑴`$ be a detector, consider the restrictions to $`𝑻`$ of the various quantum bundles and the Fock bundle $`\stackrel{~}{𝓕}\underset{𝑻}{}\stackrel{~}{𝓑}\underset{𝑻}{}𝓕`$ . At this point, one has all the ingredients needed to write down the interaction morphism over $`𝑻`$
$$:\stackrel{~}{𝓕}_{}\underset{𝑻}{}\stackrel{~}{𝓑}_{}\underset{𝑻}{}𝓕_{}𝕃^1\stackrel{~}{𝓕}\underset{𝑻}{}\stackrel{~}{𝓑}\underset{𝑻}{}𝓕,$$
where the subscript circles indicate the subbundles of test elements. The constant $`\lambda 𝕃^1`$ introduced at the end of §5 is, here, the electron’s mass $`m`$ . Thus one finds
$`{\displaystyle \frac{1}{m}}`$ $`=\mathrm{e}^{\mathrm{i}(𝗉_0𝗄_0𝗊_0)𝗍}\mathrm{\Lambda }_{𝗉A\dot{}𝗄\lambda 𝗊B}𝖢^{𝗉A\dot{}}𝖡^{𝗄\lambda }𝖠^{𝗊B}+`$
$`+\mathrm{e}^{\mathrm{i}(𝗉_0𝗄_0𝗊_0)𝗍}\mathrm{\Lambda }{}_{𝗄\lambda 𝗊B}{}^{𝗉A}𝖠_{𝗉A}^{}𝖡^{𝗄\lambda }𝖠^{𝗊B}+`$
$`+\mathrm{e}^{\mathrm{i}(𝗉_0+𝗄_0𝗊_0)𝗍}\mathrm{\Lambda }{}_{𝗉A\dot{}𝗊B}{}^{𝗄\lambda }𝖢_{}^{𝗉A\dot{}}𝖡_{𝗄\lambda }𝖠^{𝗊B}+`$
$`+\mathrm{e}^{\mathrm{i}(𝗉_0𝗄_0𝗊_0)𝗍}\mathrm{\Lambda }{}_{𝗄\lambda 𝗊B\dot{}}{}^{𝗉A\dot{}}𝖢_{𝗉A\dot{}}^{}𝖡^{𝗄\lambda }𝖢^{𝗊B\dot{}}+`$
$`+\mathrm{e}^{\mathrm{i}(𝗉_0+𝗄_0𝗊_0)𝗍}\mathrm{\Lambda }{}_{𝗊B}{}^{𝗉A𝗄\lambda }𝖠_{𝗉A}^{}𝖡_{𝗄\lambda }𝖠^{𝗊B}+`$
$`+\mathrm{e}^{\mathrm{i}(𝗉_0𝗄_0+𝗊_0)𝗍}\mathrm{\Lambda }{}_{𝗄\lambda }{}^{𝗉A𝗊B\dot{}}𝖠_{𝗉A}^{}𝖡^{𝗄\lambda }𝖢_{𝗊B\dot{}}+`$
$`+\mathrm{e}^{\mathrm{i}(𝗉_0+𝗄_0𝗊_0)𝗍}\mathrm{\Lambda }{}_{𝗊B\dot{}}{}^{𝗉A\dot{}𝗄\lambda }𝖢_{𝗉A\dot{}}^{}𝖡_{𝗄\lambda }𝖢^{𝗊B\dot{}}+`$
$`+\mathrm{e}^{\mathrm{i}(𝗉_0+𝗄_0+𝗊_0)𝗍}\mathrm{\Lambda }^{𝗉A𝗄\lambda 𝗊B\dot{}}𝖠_{𝗉A}𝖡_{𝗄\lambda }𝖢_{𝗊B\dot{}}.`$
where
$$\mathrm{\Lambda }_{𝗉A\dot{}𝗄\lambda 𝗊B}=\mathrm{}_{𝗉A\dot{}𝗄\lambda 𝗊B}\underset{¯}{\mathrm{\Lambda }}_{𝗉\mathrm{𝗄𝗊}}=\mathrm{}_{𝗉A\dot{}𝗄\lambda 𝗊B}\frac{\delta (𝗉_{}+𝗄_{}+𝗊_{})}{\sqrt{8l^9𝗉_0𝗄_0𝗊_0}}$$
and the like. Explicitely, the $`\mathrm{}`$-factors are given by
$`\mathrm{}_{𝗉A\dot{}𝗄\lambda 𝗊B}=e\overline{𝗏}_{𝗉A\dot{}}\gamma _{𝗄\lambda }𝗎_{𝗊B},`$ $`\mathrm{}{}_{𝗄\lambda 𝗊B}{}^{𝗉A}=e𝗎^{𝗉A}\gamma _{𝗄\lambda }𝗎_{𝗊B},`$
$`\mathrm{}{}_{𝗉A\dot{}𝗊B}{}^{𝗄\lambda }=e\overline{𝗏}_{𝗉A\dot{}}\gamma ^{𝗄\lambda }𝗎_{𝗊B},`$ $`\mathrm{}{}_{𝗄\lambda 𝗊B\dot{}}{}^{𝗉A\dot{}}=e\overline{𝗏}_{𝗊B\dot{}}\gamma _{𝗄\lambda }\overline{𝗏}^{𝗉A\dot{}},`$
$`\mathrm{}{}_{𝗊B}{}^{𝗉A𝗄\lambda }=e𝗎^{𝗉A}\gamma ^{𝗄\lambda }𝗎_{𝗊B},`$ $`\mathrm{}{}_{𝗄\lambda }{}^{𝗉A𝗊B\dot{}}=e𝗎^{𝗉A}\gamma _{𝗄\lambda }\overline{𝗏}^{𝗊B\dot{}},`$
$`\mathrm{}{}_{𝗊B\dot{}}{}^{𝗉A\dot{}𝗄\lambda }=e\overline{𝗏}_{𝗊B\dot{}}\gamma ^{𝗄\lambda }\overline{𝗏}^{𝗉A\dot{}},`$ $`\mathrm{}^{𝗉A𝗄\lambda 𝗊B\dot{}}=e𝗎^{𝗉A}\gamma ^{𝗄\lambda }\overline{𝗏}^{𝗊B\dot{}},`$
where $`𝖾_{𝗄\lambda }𝖾_\lambda (𝗄)`$ , $`𝖾^{𝗄\lambda }𝖾^\lambda (𝗄)`$ denotes its dual frame of $`𝑯_{}^{\mathrm{}}`$ . Morever, $`\gamma _{𝗄\lambda }\gamma [𝖾_{𝗄\lambda }]`$ and $`\gamma ^{𝗄\lambda }\gamma ^\mathrm{\#}[𝖾^{𝗄\lambda }]`$ .
In the above elementary diagrams time runs upwards; so, lines entering the vertex from below represent absorbed particles, lines entering from above represent created particles; electron lines are labelled by up arrows, positron lines are labelled by down arrows, and photon lines are wavy.
## 10 QED
In this section I will show how two-point interactions<sup>12</sup><sup>12</sup>12 One-point interactions in QED are nearly obvious at this stage. give rise to scattering matrix contributions which, at least formally, have the same expressions as in standard treatments; these expressions are the so-called *propagators* of the particles in momentum space. In the flat inertial case one recovers standard results.
Consider a second order process in which the initial and final states both contain one electron and one photon. One has two types of diagrams, and for each type on turn two subtypes can be distinguished, according to the time order of the vertices:
Here, external photon lines are labelled by an index $`Q=1,2`$ referring to the classical frame $`(𝖻_{𝗄Q})`$ of the bundle $`𝑩𝑷__0`$ of transversal polarization modes; 4-momenta are indicated by letters $`p`$, $`q`$, $`k`$ etcetera.
First, consider the diagram labelled as (I’). Following the usual procedure one finds
$`(𝒮_\mathrm{I}^{}){}_{𝗉A𝗄Q}{}^{𝗉^{}A^{}𝗄^{}Q^{}}=m^2{\displaystyle \underset{\mathrm{}}{\overset{+\mathrm{}}{}}}\mathrm{d}𝗍_2{\displaystyle \underset{\mathrm{}}{\overset{+\mathrm{}}{}}}\mathrm{d}𝗍_1{\displaystyle }\mathrm{d}^3𝗊_{}\mathrm{H}(𝗍_2𝗍_1)({\displaystyle \underset{B=1}{\overset{2}{}}}\mathrm{}{}_{𝗊B}{}^{𝗉^{}A^{}𝗄^{}Q^{}}\mathrm{}{}_{𝗉A𝗄Q}{}^{𝗊B})`$
$`{\displaystyle \frac{\delta (𝗉_{}^{}𝗊_{}+𝗄_{}^{})\delta (𝗉_{}+𝗊_{}𝗄_{})}{l^6\sqrt{16𝗉_0^{}𝗄_0^{}𝗉_0𝗄_0}\mathrm{\hspace{0.25em}2}𝗊_0}}\mathrm{e}^{\mathrm{i}(𝗉_0+𝗊_0𝗄_0)𝗍_1}\mathrm{e}^{\mathrm{i}(𝗉_0^{}𝗊_0+𝗄_0^{})𝗍_2}.`$
Note that the summation over $`B`$ , here, is to be performed before all other operations: it must be performed before integration over $`𝗊`$ because the index $`B`$ ‘resides’ over $`𝗊`$ , and before the transformation of the integral into a 4-dimensional one because the index $`B`$ cannot reside over an off-shell momentum. Hence one considers, for *fixed* $`𝗊`$ ,
$`{\displaystyle \underset{B=1}{\overset{2}{}}}\mathrm{}{}_{𝗊B}{}^{𝗉^{}A^{}𝗄^{}Q^{}}\mathrm{}{}_{𝗉A𝗄Q}{}^{𝗊B}=e^2{\displaystyle \underset{B=1}{\overset{2}{}}}(𝗎^{𝗉^{}A^{}}\gamma ^{𝗄^{}Q^{}}𝗎_{𝗊B})(𝗎^{𝗊B}\gamma _{𝗄Q}𝗎_{𝗉A})=`$
$`=e^2{\displaystyle \underset{B=1}{\overset{2}{}}}𝗎^{𝗉^{}A^{}}\gamma ^{𝗄^{}Q^{}}(𝗎_{𝗊B}𝗎^{𝗊B})\gamma _{𝗄Q}𝗎_{𝗉A}=`$
$`=e^2𝗎^{𝗉^{}A^{}}\gamma ^{𝗄^{}Q^{}}(11+\frac{1}{m}\gamma ^\mathrm{\#}[𝗊])\gamma [𝖻_{𝗄Q}]𝗎_{𝗉A},`$
since $`_{B=1}^2𝗎_{𝗊B}𝗎^{𝗊B}`$ , for fixed $`𝗊`$ , is just the projection $`11+\frac{1}{m}\gamma ^\mathrm{\#}[𝗊]:𝑾𝑾_𝗊^+`$ . Furthermore, observe that for $`𝗊𝑷_m`$ one has
$$11+\frac{1}{m}\gamma ^\mathrm{\#}[𝗊]=\frac{1}{m}(m+\mathrm{E}_m(𝗊_{})\gamma ^0+\gamma ^\mathrm{\#}[𝗊_{}]),\mathrm{E}_m(q_{}):=\sqrt{m^2+|𝗊_{}|^2}.$$
Now when one performs the usual trick for transforming the integral into a 4-dimensional one, the above factor remains unchanged, so that
$`(𝒮_\mathrm{I}^{}){}_{𝗉A𝗄Q}{}^{𝗉^{}A^{}𝗄^{}Q^{}}={\displaystyle \frac{2\pi \mathrm{i}me^2}{l^6\sqrt{16𝗉_0^{}𝗄_0^{}𝗉_0𝗄_0}}}𝗎^{𝗉^{}A^{}}\gamma ^{𝗄^{}Q^{}}`$
$`\left({\displaystyle \mathrm{d}^4𝗊\frac{\delta (𝗉𝗄+𝗊)\delta (𝗊+𝗉^{}+𝗄^{})}{2\mathrm{E}_m(𝗊_{})(𝗊_0+\mathrm{E}_m(𝗊_{})\mathrm{i}\epsilon )}\left(m+\mathrm{E}_m(𝗊_{})\gamma ^0+\gamma ^\mathrm{\#}[𝗊_{}]\right)}\right)\gamma _{𝗄Q}𝗎_{𝗉A}.`$
Next, consider the diagram labelled as (I”). Like in the scalar case, the different time order of the vertices yields different signs in the arguments of the Dirac deltas. The classical Lagrangian yields now a further difference, since
$$\underset{B\dot{}=1}{\overset{2}{}}\mathrm{}^{𝗉^{}A^{}𝗄^{}Q^{}𝗊B\dot{}}\mathrm{}_{𝗉A𝗄Q𝗊B\dot{}}=e^2𝗎^{𝗉^{}A^{}}\gamma ^{𝗄^{}Q^{}}(11\frac{1}{m}\gamma ^\mathrm{\#}[𝗊])\gamma _{𝗄Q}𝗎_{𝗉A}.$$
Then one finds
$`(𝒮_{\mathrm{I}^{\prime \prime }}){}_{𝗉A𝗄Q}{}^{𝗉^{}A^{}𝗄^{}Q^{}}={\displaystyle \frac{2\pi \mathrm{i}me^2}{l^6\sqrt{16𝗉_0^{}𝗄_0^{}𝗉_0𝗄_0}}}𝗎^{𝗉^{}A^{}}\gamma ^{𝗄^{}Q^{}}`$
$`({\displaystyle }\mathrm{d}^4𝗊{\displaystyle \frac{\delta (𝗉_{}𝗄_{}𝗊_{})\delta (𝗊_{}+𝗉_{}^{}+𝗄_{}^{})}{2\mathrm{E}_m(𝗊_{})(𝗊_0+\mathrm{E}_m(𝗊_{})\mathrm{i}\epsilon )}}\delta (𝗉_0𝗄_0+𝗊_0)\delta (𝗊_0+𝗉_0^{}+𝗄_0^{})`$
$`(m\mathrm{E}_m(𝗊_{})\gamma ^0\gamma ^\mathrm{\#}[𝗊_{}]))\gamma _{𝗄Q}𝗎_{𝗉A}.`$
In order to simplify $`(𝒮_\mathrm{I}){}_{𝗉A𝗄Q}{}^{𝗉^{}A^{}𝗄^{}Q^{}}=(𝒮_\mathrm{I}^{}+𝒮_{\mathrm{I}^{\prime \prime }})_{𝗉A𝗄Q}^{𝗉^{}A^{}𝗄^{}Q^{}}`$ one has to make the integration variable change $`𝗊_{}𝗊_{}`$ in the second contribution, so that the $`\delta `$-factors are the same. Eventually,
$`(𝒮_\mathrm{I}){}_{𝗉A𝗄Q}{}^{𝗉^{}A^{}𝗄^{}Q^{}}={\displaystyle \frac{2\pi \mathrm{i}me^2}{l^6\sqrt{16𝗉_0^{}𝗄_0^{}𝗉_0𝗄_0}}}𝗎^{𝗉^{}A^{}}\gamma ^{𝗄^{}Q^{}}`$
$`\left({\displaystyle \mathrm{d}^4𝗊\delta (𝗉𝗄+𝗊)\delta (𝗊+𝗉^{}+𝗄^{})\frac{m+\gamma ^\mathrm{\#}[𝗊]}{g(𝗊,𝗊)m^2+\mathrm{i}\epsilon }}\right)\gamma _{𝗄Q}𝗎_{𝗉A},`$
which contains the *electron propagator*, namely the distribution
$$\underset{\epsilon 0^+}{lim}\frac{m\gamma ^\mathrm{\#}[𝗊]}{g(𝗊,𝗊)m^2+\mathrm{i}\epsilon }.$$
The *positron propagator*
$$\underset{\epsilon 0^+}{lim}\frac{m+\gamma ^\mathrm{\#}[𝗊]}{g(𝗊,𝗊)m^2+\mathrm{i}\epsilon }.$$
is found by a similar procedure.
Next, consider the diagrams
From case (II) one can obtain two further similar cases by inverting one or both fermion paths. In all cases the calculation is essentially the same; case (II) is somewhat simpler notationally since it has no dotted indices. Diagram (II’) yields the summation
$$\underset{\lambda =0}{\overset{3}{}}\mathrm{}{}_{𝗉A}{}^{𝗉^{}A^{}𝗄\lambda }\mathrm{}{}_{𝗄\lambda 𝗊B}{}^{𝗊^{}B^{}}=e^2\underset{\lambda =0}{\overset{3}{}}(𝗎^{𝗉^{}A^{}}\gamma ^{𝗄\lambda }𝗎_{𝗉A})(𝗎^{𝗊^{}B^{}}\gamma _{𝗄\lambda }𝗎_{𝗊B})$$
over the internal polarization degrees of freedon of the photon; the generalized index $`𝗄`$ is kept fixed (no summation on it). Here $`\gamma _{𝗄\lambda }\gamma [𝖾_{𝗄\lambda }]`$ and $`\gamma ^{𝗄\lambda }\gamma ^\mathrm{\#}[𝖾^{𝗄\lambda }]`$ .
In order to handle the above expression conveniently, look at the Dirac map $`\gamma `$ as a linear morphism $`𝑯𝑾𝑾^{\mathrm{}}`$ , so that
$$\gamma [y]\gamma [y](𝑾𝑾{}_{}{}^{\mathrm{}})(𝑾𝑾{}_{}{}^{\mathrm{}})=\mathrm{Lin}(𝑾{}_{}{}^{\mathrm{}}𝑾,𝑾𝑾{}_{}{}^{\mathrm{}}),y𝑯.$$
One then finds
$$\gamma ^{𝗄\lambda }\gamma _{𝗄\lambda }=g^{\lambda \mu }\gamma _{𝗄\lambda }\gamma _{𝗄\mu }=g_{\lambda \mu }\gamma ^{𝗄\lambda }\gamma ^{𝗄\mu }:𝑾{}_{}{}^{\mathrm{}}𝑾𝑾𝑾{}_{}{}^{\mathrm{}}.$$
Moreover, mote that the generalized index $`𝗄`$ in the above expression *can be dropped*, since the described object is independent of the frame in which it is written, namely
$$\gamma ^{𝗄\lambda }\gamma _{𝗄\lambda }=\gamma ^{𝗄^{}\lambda ^{}}\gamma _{𝗄^{}\lambda ^{}}\gamma ^\lambda \gamma _\lambda .$$
Now the previously considered summation over the virtual photon’s polarization states can be rewritten as
$$\underset{\lambda =0}{\overset{3}{}}\mathrm{}{}_{𝗉A}{}^{𝗉^{}A^{}𝗄\lambda }\mathrm{}{}_{𝗄\lambda 𝗊B}{}^{𝗊^{}B^{}}=e^2g_{\lambda \mu }(𝗎^{𝗉^{}A^{}}𝗎_{𝗉A})(\gamma ^\lambda \gamma ^\mu )(𝗎^{𝗊^{}B^{}}𝗎_{𝗊B}).$$
From diagram (II’) one gets, similarly,
$$\underset{\lambda =0}{\overset{3}{}}\mathrm{}{}_{𝗄\lambda 𝗉A}{}^{𝗉^{}A^{}}\mathrm{}{}_{𝗊B}{}^{𝗊^{}B^{}𝗄\lambda }=e^2g_{\lambda \mu }(𝗎^{𝗉^{}A^{}}𝗎_{𝗉A})(\gamma ^\lambda \gamma ^\mu )(𝗎^{𝗊^{}B^{}}𝗎_{𝗊B}).$$
Thus, eventually, the contraction over the internal states of the virtual photon gives the same result in the two subcases (II’) and (II”) differing for the time ordering of the interaction. The fact that the generalized index $`𝗄`$ disappears in this operation implies that the photon propagator is simply the scalar massless one tensorialized by the spacetime metric, that is
$`(𝒮_{\mathrm{II}}){}_{𝗉A𝗊B}{}^{𝗉^{}A^{}𝗊^{}B^{}}={\displaystyle \frac{2\pi \mathrm{i}m^2e^2}{l^6\sqrt{16𝗉_0^{}𝗊_0^{}𝗉_0𝗊_0}}}(𝗎^{𝗉^{}A^{}}𝗎_{𝗉A})\gamma ^\lambda `$
$`\left({\displaystyle \mathrm{d}^4𝗄\delta (𝗉𝗄+𝗊)\delta (𝗊+𝗉^{}+𝗄^{})\frac{g_{\lambda \mu }}{g(𝗄,𝗄)+\mathrm{i}\epsilon }}\right)\gamma ^\mu (𝗎^{𝗊^{}B^{}}𝗎_{𝗊B}).`$
|
warning/0506/physics0506074.html
|
ar5iv
|
text
|
# Statistical analysis of 22 public transport networks in Poland
## I Introduction
Since the explosion of the complex network science that has taken place after works of Watts and Strogatz watts\_nat as well as Barabási and Albert barabasi\_sci ; bamf a lot of real-world networks have been examined. The examples are technological networks (Internet, phone calls network), biological systems (food webs, metabolic systems) or social networks (co-authorship, citation networks) barabasi\_rmp ; newman\_siam ; mendes\_book ; satorras\_book . Despite this, at the beginning little attention has been paid to transportation networks \- mediums as much important and also sharing as much complex structure as those previously listed. However, during the last few years several public transport systems (PTS) have been investigated using various concepts of statistical physics of complex networks amaral\_pnas ; strogatz\_nat ; albert\_pre ; crucitti\_physa ; marichiori\_physa ; latora\_prl ; latora\_physa ; sen\_pre ; seaton\_physa ; guimera\_arxiv ; guimera\_epjb ; barrat\_pnas ; li\_pre ; bagler\_arxiv .
Chronogically the first works regarding transportation networks have dealt with power grids barabasi\_sci ; amaral\_pnas ; watts\_nat ; strogatz\_nat . One can argue that transformators and transmission lines have little in common with PTS (i.e. underground, buses and tramways), but they definitely share at least one common feature: embedding in a two-dimensional space. Research done on the electrical grid in United States - for Southern California barabasi\_sci ; amaral\_pnas ; watts\_nat ; strogatz\_nat and for the whole country albert\_pre as well as on the GRTN Italian power network crucitti\_physa revealed a single-scale degree distributions ($`p(k)\mathrm{exp}(\alpha k)`$ with $`\alpha 0.5`$), a small average connectivity values and relatively large average path lengths.
All railway and underground systems appear to share well known small-world properties watts\_nat . Moreover this kind of networks possesses several other characteristic features. In fact Latora and Marichiori have studied in details a network formed by the Boston subway marichiori\_physa ; latora\_prl ; latora\_physa . They have calculated a network efficiency defined as a mean value of inverse distances between network nodes. Although the global efficiency is quite large $`E_{glob}=0.63`$ the local efficiency calculated in the subgraphs of neighbors is low $`E_{local}=0.03`$ what indicates a large vulnerability of this network against accidental damages. However, the last parameter increases to $`E_{local}^{^{}}=0.46`$ if the subway network is extended by the existing bus routes network. Taking into account geographical distances between different metro stations one can consider the network as a weighted graph and one is able to introduce a measure of a network cost. The estimated relative cost of the Boston subway is around 0.2 % of the total cost of fully connected network.
Sen et al. sen\_pre have introduced a new topology describing the system as a set of train lines, not stops, and they have discovered a clear exponential degree distribution in Indian railway network. This system has shown a small negative value of assortativity coefficient. Seaton and Hackett seaton\_physa have compared real data from underground systems of Boston (first presented in latora\_physa ) and Vienna with the prediction of bipartite graph theory (here: graph of lines and graph of stops) using generation function formalism. They have found a good correspondence regarding value of average degree, however other properties like clustering coefficient or network size have shown differences of 30 to 50 percent.
In works of Amaral, Barrat, Guimerà et al. amaral\_pnas ; guimera\_arxiv ; guimera\_epjb ; barrat\_pnas a survey on the World-Wide Airport Network has been presented. The authors have proposed truncated power-law cumulative degree distribution $`P(k)k^\alpha f(k/k_x)`$ with the exponent $`\alpha =1.0`$ and a model of preferential attachment where a new node (flight) is introduced with a probability given by a power-law or an exponential function of physical distance between connected nodes. However, only an introduction of geo-political constrains barrat\_pnas (i.e. only large cities are allowed to establish international connections) explained the behavior of betweenness as a function of node degree.
Other works on airport networks in India li\_pre and China bagler\_arxiv have stressed small-world properties of those systems, characterized by small average path lengths ($`l2`$) and large clustering coefficients ($`c>0.6`$) with comparison to random graph values. Degree distributions have followed either a power-law (India) or a truncated power-law (China). In both cases an evidence of strong disassortative degree-degree correlation has been discovered and it also appears that Airport Network of India has a hierarchical structure expressed by a power-law decay of clustering coefficient with an exponent equal to $`1`$.
In the present paper we have studied a part of data for PTS in $`22`$ Polish cities and we have analyzed their nodes degrees, path lengths, clustering coefficients, assortativity and betweenness. Despite large differences in sizes of considered networks (number of nodes ranges from $`N=152`$ to $`N=2881`$) they share several universal features such as degree and path length distributions, logarithmic dependence of distances on nodes degrees or a power law decay of clustering coefficients for large nodes degrees. As far as we know, our results are the first comparative survey of several public transport systems in the same country using universal tools of complex networks.
## II The idea of space L and P
To analyze various properties of PTS one should start with a definition of a proper network topology. The idea of the space L and P, proposed in a general form in sen\_pre and used also in seaton\_physa is presented at Fig. 1. The first topology (space L) consists of nodes representing bus, tramway or underground stops and a link between two nodes exists if they are consecutive stops on the route. The node degree $`k`$ in this topology is just the number of directions (it is usually twice the number of all PTS routes) one can take from a given node while the distance $`l`$ equals to the total number of stops on the path from one node to another.
Although nodes in the space P are the same as in the previous topology, here an edge between two nodes means that there is a direct bus, tramay or underground route that links them. In other words, if a route $`A`$ consists of nodes $`a_i`$, i.e. $`A=\{a_1,a_2,\mathrm{},a_n\}`$, then in the space P the nearest neighbors of the node $`a_1`$ are $`a_2,a_3,\mathrm{},a_n`$. Consequently the node degree $`k`$ in this topology is the total number of nodes reachable using a single route and the distance can be interpreted as a number of transfers (plus one) one has to take to get from one stop to another.
Another idea of mapping a structure embedded in two-dimensional space into another, dimensionless topology has recently been used by Rosvall et al. in rosvall\_prl where a plan of the city roads has been mapped into an ”information city network”. In the last topology a road represents a node and an intersection between roads - an edge, so the network shows information handling that has to be performed to get oriented in the city.
We need to stress that the spaces L and P do not take into account Euclidean distance between nodes. Such an approach is similar to the one used for description of several other types of network systems: Internet barabasi\_sci , power grids albert\_pre ; crucitti\_physa , railway sen\_pre or airport networks li\_pre ; bagler\_arxiv .
## III Explored systems
We have analyzed PTS (bus and tramways systems) in $`22`$ Polish cities, located in various state districts as it is depicted at Fig. 2. Table 1 gathers fundamental parameters of considered cities and data on average path lengths, average degrees, clustering coefficients as well as assortativity coefficients for corresponding networks.
Numbers of nodes in different networks (i.e. in different cities) range from $`N=152`$ to $`N=2811`$ and they are roughly proportional to populations $`I`$ and surfaces $`S`$ of corresponding cities (see Fig. 3). One should notice that other surveys exploring the properties of transportation networks have usually dealt with smaller numbers of vertices, such as $`N=76`$ for U-Bahn network in Vienna seaton\_physa , $`N=79`$ for Airport Network of India (ANI) bagler\_arxiv , $`N=124`$ in Boston Underground Transportation System (MBTA) latora\_physa or $`N=128`$ in Airport Network of China (ANC) li\_pre . Only in the case of the Indian Railway Network (IRN) sen\_pre where $`N=579`$ and World-Wide Airport Network (WAN) barrat\_pnas with $`3880`$ nodes sizes of networks have been similar or larger than for PTS in Poland. Very recently, von Ferber et al. ferber\_cmp have presented a paper on three large PTS: Düsseldorf with $`N=1615`$, Berlin with $`N=2952`$ and Paris where $`N=4003`$.
## IV Degree distributions
### IV.1 Degree distribution in the space L
Fig. 4 shows typical plots for degree distribution in the space L. One can see that there is a slightly better fit to the linear behavior in the log-log description as compared to semi-logarithmic plots. Points $`k=1`$ are very peculiar since they correspond to routes’ ends. Remaining parts of degree distributions can be approximately described by a power law
$$p(k)k^\gamma $$
(1)
although the scaling cannot be seen very clearly and it is limited to less than one decade. Pearson correlation coefficients of the fit to Eq. (1) range from 0.95 to 0.99. Observed characteristic exponents $`\gamma `$ are between $`2.4`$ and $`4.1`$ (see Table 2), with the majority (15 out of 22) $`\gamma >3`$. Values of exponents $`\gamma `$ are significantly different from the value $`\gamma =3`$ which is characteristic for Barabási-Albert model of evolving networks with preferential attachment bamf and one can suppose that a corresponding model for transport network evolution should include several other effects. In fact various models taking into account effects of fitness, atractiveness, accelerated growth and aging of vertices mendes\_adv or deactivation of nodes vazquez\_pre ; klemm\_pre lead to $`\gamma `$ from a wide range of values $`\gamma 2,\mathrm{})`$. One should also notice that networks with a characteristic exponent $`\gamma >4`$ are considered topologically close to random graphs havlin\_prl \- the degree distribution is very narrow - and a difference between power-law and exponential behavior is very subtle (see the Southern California power grid distribution in barabasi\_sci presented as a power-law with $`\gamma 4`$ and in strogatz\_nat depicted as a single-scale cumulative distribution).
Degree distributions obtained for airport networks are also power-law (ANC, ANI) or power-law with an exponential cutoff (in the case of WAN). For all those systems exponent $`\gamma `$ is in the range of $`2.02.2`$, which differs significantly from considered PTS in Poland, however one has to notice, that airport networks are much less dependent on the two-dimensional space as it is in the case of PTS. This effect is also seen when analyzing average connectivity ($`k=5.77`$ for ANI, $`k=9.7`$ for WAN and $`k=1214`$ for ANC depending on the day of the week the data have been collected).
Let us notice that the number of nodes of degree $`k=1`$ is smaller as compared to the number of nodes of degree $`k=2`$ since $`k=1`$ nodes are ends of transport routes. The maximal probability observed for nodes with degree $`k=2`$ means that a typical stop is directly connected to two other stops. Still some nodes (hubs) can have a relatively high degree value (in some cases above 10) but the number of such vertices is very small.
### IV.2 Degree distribution in the space P
In our opinion, the key structure for the analysis of PTS are routes and not single bus/tramway stops. Therefore we especially take under consideration the degree distribution in the space P.
To smooth large fluctuations, we use here the cumulative distribution $`P(k)`$ newman\_siam according to the formula
$$P(k)=_k^{k_{max}}p(k)𝑑k$$
(2)
The cumulative distributions in the space P for eight chosen cities are shown at Fig 5. Using the semi-log scale we observe an exponential character of such distributions:
$$P(k)=Ae^{\alpha k}$$
(3)
As it is well known bamf the exponential distribution (3) can occur for evolving networks when nodes are attached completely randomly. This suggests that a corresponding evolution of public transport in the space P possesses an accidental character that can appear because of large number of factors responsible for urban development. However in the next sections we show that other network’s parameters such as clustering coefficients or degree-degree correlations calculated for PTS are much larger as compared to corresponding values of randomly evolving networks analyzed in bamf .
In the case of IRN sen\_pre degree distribution in the space P has also maintained the single-scale character $`P(k)e^{\alpha k}`$ with the characteristic exponent $`\alpha =0.0085`$. Values of average connectivity in the studies of MBTA ($`k=27.60`$) and U-Bahn in Vienna ($`k=20.66`$) are smaller than for considered systems in Poland, however one should notice that sizes of networks in MBTA and Vienna are also smaller.
### IV.3 Average degree and average square degree
Taking into account the normalization condition $`P(k_{min})=1`$ we get the following equations for the average degree and the average square degree:
$$k=\frac{k_{min}e^{\alpha k_{min}}k_{max}e^{\alpha k_{max}}}{e^{\alpha k_{min}}e^{\alpha k_{max}}}+\frac{1}{\alpha }$$
(4)
$`k^2={\displaystyle \frac{k_{min}^2e^{\alpha k_{min}}k_{max}^2e^{\alpha k_{max}}}{e^{\alpha k_{min}}e^{\alpha k_{max}}}}+`$ (5)
$`+{\displaystyle \frac{2(k_{min}e^{\alpha k_{min}}k_{max}e^{\alpha k_{max}})}{\alpha (e^{\alpha k_{min}}e^{\alpha k_{max}})}}+{\displaystyle \frac{2}{\alpha ^2}}`$
Dropping all terms proportional to $`e^{\alpha k_{max}}`$ we receive simplified equations for $`k`$ i $`k^2`$:
$$kk_{min}+\frac{1}{\alpha }$$
(6)
$$k^2k_{min}^2+\frac{2k_{min}}{\alpha }+\frac{2}{\alpha ^2}$$
(7)
Since values of $`k_{min}`$ range between $`3`$ and $`16`$ and they are independent from network sizes $`N`$ as well as observed exponents $`\alpha `$ we have approximated $`k_{min}`$ in Eqs. (6) - (7) by an average value (mean arithmetical value) for considered networks, $`k_{min}8.5`$. At Figs. 6 and 7 we present a comparison between the real data and values calculated directly form Eqs. (6) and (7).
## V Path length’s properties
### V.1 Path length’s distributions
Plots presenting path length distributions $`p(l)`$ in spaces L and P are shown at Figs. 8 and 9 respectively. The data well fit to asymmetric, unimodal functions. In fact for all systems a fitting by Lavenberg - Marquardt method has been made using the following trial function:
$$p(l)=Ale^{Bl^2+Cl}$$
(8)
where $`A,B`$ and $`C`$ are fitting coefficients.
Inserts at Figs. 8 and 9 present a comparison between experimental results of $`l`$ and corresponding mean values obtained from Eq. (8). One can observe a very good agreement between averages from Eq. (8) and experimental data.
The agreement is not surprising in the case of Fig. 9 since the number of fitted data points to curve (8) is quite small, but it is more prominent for Fig. 8.
Ranges of distances in the space L are much broader as compared to corresponding ranges in the space P what is a natural effect of topology differences. It follows that the average distance in the space P is much smaller ($`l<3`$) than in the space L. The characteristic length 3 in the space P means that in order to travel between two different points one needs in average no more than two transfers. Other PTS also share this property, depending on the system size the following results have been obtained: $`l=1.81`$ (MBTA), $`l=1.86`$ (Vienna), $`l=2.16`$ (IRN). In the case of the space L the network MBTA with its average shortest path length $`l=15.55`$ is placing itself among the values acquired for PTS in Poland. Average path length in airport networks is very small: $`l=2.07`$ for ANC, $`l=2.26`$ for ANI and $`l=4.37`$ for WAN. However, because flights are usually direct (i.e. there are no stops between two cities) one sees immediately that the idea of the space L does not apply to airport networks - they already have an intrinsic topology similar to the space P. Average shortest path lengths $`l`$ in those systems should be relevant to values obtained for other networks after a transformation to the space P.
The shape of path length distribution can be explained in the following way: because transport networks tend to have an inhomogeneous structure, it is obvious that distances between nodes lying on the suburban routes are quite large and such a behavior gives the effect of observed long tails in the distribution. On the other hand shortest distances between stops not belonging to suburban routes are more random and they follow the Gaussian distribution. A combined distribution has an asymmetric shape with a long tail for large paths.
We need to stress that inter-node distances calculated in the space L are much smaller as compared to the number of network nodes (see Table I). Simultaneously clustering coefficients $`c_L`$ are in the range $`0.03,0.15`$. Such a behavior is typical for small-world networks watts\_nat and the effect has been also observed in other transport networks amaral\_pnas ; latora\_physa ; sen\_pre ; seaton\_physa ; li\_pre ; bagler\_arxiv . The small world property is even more visible in the space P where average distances are between $`1.80,2.90`$ and the clustering coefficient $`c_P`$ ranges from 0.682 to 0.847 which is similar to MBTA ($`c=0.93`$), Vienna ($`c=0.95`$) or IRN ($`c=0.69`$).
### V.2 Path length as function of product $`k_ik_j`$
In agatac an analytical estimation of average path length $`l`$ in random graphs has been found. It has been shown that $`l`$ can be expressed as a function of the degree distribution. In fact the mean value for shortest path length between $`i`$ and $`j`$ can be written as agatac :
$$l_{ij}(k_i,k_j)=\frac{\mathrm{ln}k_ik_j+\mathrm{ln}\left(k^2k\right)+\mathrm{ln}N\gamma }{\mathrm{ln}\left(k^2/k1\right)}+\frac{1}{2}$$
(9)
where $`\gamma =0.5772`$ is Euler constant.
Since PTS are not random graphs and large degree-degree correlation in such networks exist we have assumed that Eq. (9) is only partially valid and we have written it a more general form nasz\_physa ; nasz\_prl ; nasz\_aip ; nasz\_app :
$$l_{ij}=AB\mathrm{log}k_ik_j.$$
(10)
To check the validity of Eq. (10) we have calculated values of average path length between $`l_{ij}`$ as a function of their degree product $`k_ik_j`$ for all systems in the space L . The results are shown at Fig. 10, which confirms the conjunction (10). A similar agreement has been received for the majority of investigated PTS. Eq. (10) can be justified using a simple model of random graphs and a generating function formalism motter or a branching tree approach nasz\_physa ; nasz\_prl ; nasz\_aip ; nasz\_app . In fact the scaling relation (10) can be also observed for several other real world networks nasz\_physa ; nasz\_prl ; nasz\_aip ; nasz\_app .
It is useless to examine the relation (10) in the space P because corresponding sets $`l_{ij}`$ consist usually of 3 points only.
## VI Clustering coefficient
We have studied clustering coefficients $`c_i`$ defined as a probability that two randomly chosen neighbors of node $`i`$ possess a common link.
The clustering coefficient of the whole network seems to depend weakly on parameters of the space L and of the space P. In the first case its behavior with regard to network size can be treated as fluctuations, when in the second one it is possible to observe a small decrease of $`c`$ along with the networks size (see Table 1). We shall discuss only properties of the clustering coefficients in the space P since the data in the space L are meaningless.
It has been shown in sen\_pre that clustering coefficient in IRN in the space P decays linearly with the logarithm of degree for large $`k`$ and is almost constant (and close to unity) for small $`k`$. In the considered PTS we have found that this dependency can be described by a power law (see Fig. 11):
$$c(k)k^\beta $$
(11)
Such a behavior has been observed in many real systems with hierarchical structures ravasz\_pre ; ravasz\_sci . In fact, one can expect that PTS should consist of densely connected modules linked by longer paths.
Observed values of exponents $`\beta `$ are in the range $`\beta 0.54,0.93`$. This can be explained using a simple example of a star network: suppose that the city transport network is a star consisting of $`n`$ routes with $`L`$ stops each. Node $`i`$, at which all $`n`$ routes cross is a vertex that has the highest degree in the network. We do not allow any other crossings among those $`n`$ routes in the whole system. It follows that the degree of node $`i`$ is $`k_i=n(L1)`$ and the total number of links among the nearest neighbors of $`i`$ is $`E_i=n(L1)(L2)/2`$. In other words the value of the clustering coefficient for the node with the maximum degree is:
$$c(k_{max})=\frac{2E_i}{k_i(k_i1)}=\frac{L2}{n(L1)1}$$
(12)
where $`k_{max}=n(L1)`$. It is obvious that the minimal degree in the network is $`k_{min}=L1`$ and this correspondences to the value $`c(k_{min})=1`$. Using these two points and assuming that we have a power-law behavior we can express $`\beta `$ as:
$$\beta =\frac{\mathrm{ln}c(k_{max})\mathrm{ln}c(k_{min})}{\mathrm{ln}k_{max}\mathrm{ln}k_{min}}=\frac{\mathrm{ln}\frac{L2}{n(L1)1}}{\mathrm{ln}n}$$
(13)
Because $`n(L1)1`$ and $`L1L2`$ we have $`\beta 1`$.
In real systems the value of clustering coefficient of the highest degree node is larger than in Eq. (12) due to multiple crossings of routes in the whole network what leads to a decrease of the exponent $`\beta `$ (see Fig. 11). This decrease is also connected to the presence of degree-degree correlations (see the next Section).
## VII Degree-degree correlations
To analyze degree-degree correlations in PTS we have used the assortativity coefficient $`r`$, proposed by Newman new2 that corresponds to the Pearson correlation coefficient new4 of the nodes degrees at the end-points of link:
$$r=\frac{_ij_ik_i\frac{1}{M}_ij_i_ik_i}{\sqrt{_ij_i^2\frac{1}{M}(_ij_i)^2}\sqrt{_ik_i^2\frac{1}{M}(_ik_i)^2}}$$
(14)
where $`M`$ \- number of pairs of nodes (twice the number of edges), $`j_i,k_i`$ \- degrees of vertices at both ends of $`i`$-th pair and index $`i`$ goes over all pairs of nodes in the network.
Values of the assortativity coefficient $`r`$ in the space L are independent of the network size and are always positive (see Table 1), what can be explained in the following way: there is a little number of nodes characterized by high values of degrees $`k`$ and they are usually linked among themselves. The majority of remaining links connect nodes of degree $`k=2`$ or $`k=1`$, because $`k=2`$ is an overwhelming degree in networks.
Similar calculations performed for the space P lead to completely different results (Fig. 12). For small networks the correlation parameter $`r`$ is negative and it grows with $`N`$, becoming positive for $`N500`$. The dependence can be explained as follows: small towns are described by star structures and there are only a few doubled routes, so in this space a lot of links between vertices of small and large $`k`$ exist. Using the previous example of a star network and taking into account that the degree of the central node is equal to $`k_c=n(L1)`$, the degree of any other node is $`k_o=L1`$, after some algebra we receive the following expression for the assortativity coefficient of such a star network:
$$r=\frac{1}{L1}$$
(15)
Let us notice that the coefficient $`r`$ is independent from the number of crossing routes and is always a negative number.
On the contrary, in the large cities there are lots of connections between nodes characterized by large $`k`$ (transport hubs) as well as there is a large number of routes crossing in more than one point (see Fig. 13). It follows that the coefficient $`r`$ can be positive for such networks. A strange behavior for the largest network (GOP) can be explained as an effect of its peculiar structure: the system is rather a conglomerate of many towns than a single city. Thus, the value of $`r`$ is lowered by single links between the subsets of this network.
At Fig. 14 we show coefficients $`\beta `$ as a function of $`r`$ in the space P. One can see that in general positive values of the assortativity coefficient correspond to lower values of $`\beta `$, being an effect of existence of several links between hubs in the networks.
Reported values of assortativity coefficients in other transport networks have been negative ($`r=0.402`$ for ANI li\_pre and $`r=0.033`$ for IRN sen\_pre ) and since these systems are of the size $`N<600`$ thus it is in agreement with our results.
## VIII Betweenness
The last property of PTS examined in this work is betweenness soc which is the quantity describing the ”importance” of a specific node according to equation bar1 :
$$g(i)=\underset{jk}{}\frac{\sigma _{jk}(i)}{\sigma _{jk}}$$
(16)
where, $`\sigma _{jk}`$ is a number of the shortest paths between nodes $`j`$ and $`k`$, while $`\sigma _{jk}(i)`$ is a number of these paths that go through the node $`i`$.
### VIII.1 Betweenness in the space L
Fig. 15 shows dependence of the average betweenness $`g`$ on node degree calculated using the algorithm proposed in new3 (see also brandes ). Data at Fig. 15 fit well to the scaling relation:
$$gk^\eta $$
(17)
observed in Internet Autonomous Systems vaz , co-authorship networks goh and BA model or Erdős-Rényi random graphs bar1 .
The coefficient $`\eta `$ is plotted at Fig. 16 as a function of network size. One can see, that $`\eta `$ is getting closer to 2 for large networks. Since it has been shown that there is $`\eta =2`$ for random graphs bar1 with Poisson degree distribution thus it can suggest that large PTS are more random than small ones. Such an interpretation can be also received from the Table 2 where larger values of the exponent $`\gamma `$ are observed for large cities.
### VIII.2 Betweenness in the space P
The betweenness as a function of node degree $`k`$ in the space P is shown at Fig. 17. One can see large differences between Fig. 15 and 17. In the space P there is a saturation of $`g`$ for small $`k`$ what is a result of existence of the suburban routes while the scale-free behavior occurs only for larger $`k`$. The saturation value observed in the limit of small $`k`$ is given by $`g(k_{min})=2(N1)`$ and the length of the saturation line increases with the mean value of a single route’s length observed in a city.
## IX Conclusions
In this study we have collected and analyzed data for public transport networks in 22 cities that make over 25 % of population in Poland. Sizes of these networks range from $`N=152`$ to $`N=2881`$. Using the concept of different network topologies we show that in the space L, where distances are measured in numbers of passed bus/tramway stops, the degree distributions are approximately given by a power laws with $`\gamma =2.44.1`$ while in the space P, where distances are measured in numbers of transfers, the degree distribution is exponential with characteristic exponents $`\alpha =0.0130.050`$. Distributions of paths in both topologies are approximately given by a function $`p(l)=Ale^{Bl^2+Cl}`$. Small world behavior is observed in both topologies but it is much more pronounced in space P where the hierarchical structure of network is also deduced from the behavior of $`c(k)`$. The assortativity coefficient measured in the space L remains positive for the whole range of $`N`$ while in the space P it changes from negative values for small networks to positive values for large systems. In the space L distances between two stops are linear functions of the logarithm of their degree products.
Many of our results are similar to features observed in other works regarding transportation networks: underground, railway or airline systems amaral\_pnas ; marichiori\_physa ; latora\_prl ; latora\_physa ; sen\_pre ; seaton\_physa ; guimera\_arxiv ; guimera\_epjb ; barrat\_pnas ; li\_pre ; bagler\_arxiv ; ferber\_cmp . All such networks tend to share small-world properties and show strong degree-degree correlations that reveal complex nature of those structures.
###### Acknowledgements.
The work was supported by the EU Grant Measuring and Modelling Complex Networks Across Domains - MMCOMNET (Grant No. FP6-2003-NEST-Path-012999), by the State Committee for Scientific Research in Poland (Grant No. 1P03B04727) and by a special Grant of Warsaw University of Technology.
|
warning/0506/math0506580.html
|
ar5iv
|
text
|
# Untitled Document
thanks: 05-01-00939
, , . , $`M`$ ,
$$ds^2=du^2+2\mathrm{cos}\theta (u,v)dudv+dv^2.$$
, , $`\theta (u,v)`$ — .
( ., , \[BVK, . 413\] , , $`|K𝑑S|>2\pi `$, . $`K`$ — , $`S`$ — .
, . 2- (‘‘ ’’).
2- ,
, . \[AZ\] \[Resh\].
. . . . \[Bak\], , ,
$`\frac{\pi }{2}`$. , .
, $`C^{\mathrm{}}`$\- , . . \[SD\]:
$`C^{\mathrm{}}`$\- . — . ,
; \[SD\] \[LSh\].
, . .
###### 1.
$`M`$ — , $`\omega ^+(M)<2\pi ,\omega ^{}(M)<2\pi `$. $`M`$ .
$`\omega ^+(M)<2\pi ϵ,\omega ^{}(M)<2\pi ϵ`$, $`ϵ>0`$, , 0 $`\pi `$ ( , $`ϵ/4`$).
$`\omega ^+`$ $`\omega ^{}`$$`M`$, ;
$`\omega ^+(M)=_MK^+𝑑S,\omega ^{}(M)=_MK^{}𝑑S`$,
$`a^+=\mathrm{max}\{a,0\},a^{}=\mathrm{max}\{a,0\}`$.
, .
$`M`$ , . .
.
. , . , $`Q`$, $`\gamma _1,\gamma _2`$ $`p=\gamma _1(0)=\gamma _2(0)`$ . $`\tau `$ $`Q`$, . ( ); $`\tau ^+,\tau ^{}`$ $`\tau `$. ( $`\gamma :(0,a)M`$ 2- $`M`$$`\tau ^\pm (E)=_Ek_g^\pm 𝑑s`$, $`k_g`$$`Q`$, ( ) $`E(0,a)`$.) , $`\stackrel{~}{\omega }^\pm (M)=\omega ^\pm (M)+\tau ^\pm (\gamma _1)+\tau ^\pm (\gamma _2).`$
###### 2.
( . ) $`Q`$ :
$$\stackrel{~}{\omega }^+(Q)<\alpha ,\stackrel{~}{\omega }^{}(Q)<\pi \alpha ,$$
$`\alpha `$$`Q`$ $`p`$, $`Q`$ , $`\gamma _1,\gamma _2`$ .
0 $`\pi `$ $`\alpha \stackrel{~}{\omega }^+(Q)`$ $`\pi \alpha \stackrel{~}{\omega }^{}(Q)`$.
, 1 $`M`$ , 2.
, .
, , . \[BB, B\]. , , . . $`\mathrm{min}\{inf\theta ,inf(\pi \theta )\}`$.
$`\omega ^{}(M)>2\pi `$ $`M`$ , ‘‘ ’’ , , . , , , . . \[BL\] , $`\omega ^+(M)<2\pi ϵ<2\pi ,\omega ^{}(M)<C<\mathrm{}`$ $`M`$ $`ϵ^{\frac{1}{2}}(2\pi +C)^{\frac{1}{2}}`$. ‘‘ ’’ ( . . ) – .
. ; , , ; . 2,
, . $`^2`$ , .
, , . , , ( ) .
$`M`$ — 2- . $`M`$,
$`\gamma _i`$,
(i) $`\gamma _1:(\mathrm{},\mathrm{})M`$, $`\gamma _i:(0,\mathrm{})M`$ $`i=2,3`$;
(ii) $`\gamma _1`$ $`M`$ ( );
(iii) $`\gamma _2`$ $`\gamma _3`$ $`\gamma _1`$ $`\gamma _1`$, .
$`M`$ $`Q_i`$. $`\alpha _i`$ $`Q_i`$ $`O_i`$.
, , .
###### 3.
M - ,
$$\omega ^+<2\pi 4ϵ,\omega ^{}<2\pi 4ϵ$$
(1)
$`ϵ>0`$. $`M`$ , $`i=1,2,3,4`$$`\omega ^+(Q_i)\alpha _iϵ`$, $`\omega ^{}(Q_i)\pi \alpha _iϵ`$ $`\gamma _i(t),\gamma _i(t^{})`$, $`i=1,2,3`$,
$$|tt^{}|\mathrm{sin}ϵd_M(\gamma _i(t),\gamma _i(t^{})).$$
(2)
, .
###### 1.
$`M`$ (1). $`M`$ , , 3, $`\omega ^\pm `$ $`\stackrel{~}{\omega }^\pm `$, .
, 2 1.
. 1 , .
\[SD\]
, . .
3 , 6.1 \[BL\].
3.
1. – , $`\gamma _1`$, (2) $`M`$ $`M_k`$, $`k=1,2`$, . ,
$$\omega ^+(M_k)<\pi 2ϵ,\omega ^{}(M_k)<\pi 2ϵ.$$
(3)
$`M_1`$. $`M_1`$ $`\gamma _1`$, $`M_1`$ . , (2) (3) , . \[AZ\], 2 IX.
\[BL\], , $`\stackrel{~}{D}\text{int}M_1`$, , ,
$`\gamma _1`$. , , $`M_1`$ ; , , $`M_2`$ $`M_1`$ $`M_1`$ $`M_2`$.
2. $`S\stackrel{~}{D}`$ $`\stackrel{~}{D}`$. ( — $`\stackrel{~}{D}`$ $`S^1`$.) $`\sigma `$ $`S\stackrel{~}{D}`$ ( ) $`M_1`$. $`v=(p,\theta )S\stackrel{~}{D}`$, $`\sigma (v)`$ \- , $`p`$ $`v`$, , $`\sigma (v)`$ $`p`$ $`v`$ $`M_1`$.
– , $`\sigma (v)`$
$`M_1`$ , . $`U(v)`$ , $`v`$. $`S\stackrel{~}{D}`$ $`\alpha (v)`$ : $`U(v)`$ — , $`\alpha (v)=0`$; $`U(v)`$ — , $`\sigma (v)`$, $`\alpha (v)=\pi `$; $`U(v)`$ — , $`\alpha (v)`$ — ; $`U(v)`$$`\alpha _1`$ $`\alpha _2`$ — , $`\alpha (v)={\displaystyle \frac{\alpha _1^2+\alpha _2^2}{\alpha _1+\alpha _2}}`$; ( , $`\alpha _1`$) ,
$`\alpha (v)={\displaystyle \frac{\pi ^2/4+\alpha _2^2}{\pi /2+\alpha _2}}+\alpha _1{\displaystyle \frac{\pi }{2}}`$; , $`U(v)`$$`M_1`$ , $`\alpha (v)=\pi \alpha (v)`$. , $`\alpha `$ $`S\stackrel{~}{D}`$. , – , .
, $`\alpha `$
$$\alpha (v)+\alpha (v)=\pi .$$
$`\omega ^+`$ $`\omega ^{}`$, :
$$W^\pm (A)=(\pi 2ϵ)\frac{\omega ^\pm (A)}{\omega ^\pm (\stackrel{~}{D})}.$$
$`\varphi :S\stackrel{~}{D}R^2`$, $`vS\stackrel{~}{D}`$
$$(W^+(U(v))\alpha +ϵ,W^{}(U(v))(\pi \alpha )+ϵ).$$
, $`\varphi (v)=\varphi (v)`$, , $`\varphi (v)=(0,0)`$, $`\varphi (v)=(0,0)`$.
$`v_0`$, $`\varphi (v_0)=(0,0)`$.
$`O`$ $`\stackrel{~}{D}`$. $`S\stackrel{~}{D}`$ , $`\gamma _1`$ $`\gamma _2`$, $`O`$.
$`\gamma _1`$ $`v`$ $`O`$, , $`\stackrel{~}{S}`$ $`\stackrel{~}{D}`$ .
$`\gamma _2`$ $`\stackrel{~}{S}`$ ( $`O`$) , $`v`$, $`\stackrel{~}{S}`$ $`\stackrel{~}{D}`$.
. , $`g`$ $`\stackrel{~}{D}`$ $`D`$, $`dg`$ $`\stackrel{~}{D}`$, , $`O^{}=g(O)`$ \- , $`dg`$\- $`tD`$, $`0t1`$, , .
$`\varphi `$ $`\gamma _1`$, $`\gamma _2`$ $`\text{Im}\varphi `$; , $`\text{Im}\varphi `$.
$`\gamma _2`$ $`x+y=\pi 2ϵ`$, $`\text{Im}\varphi `$.
, $`\alpha `$, $`\gamma _1`$ $`\varphi (v)+\varphi (v)=(0,0)`$. , $`(0,0)`$ $`\gamma _1`$ $`^2(0,0)`$, $`\text{Im}\varphi `$. , , $`(0,0)`$ $`\gamma _1`$, $`(2k+1)\pi `$.
, - $`\text{Im}\varphi `$ ( ), $`\gamma _1`$ , . , .
, $`v`$ , $`\varphi (v)=(0,0)`$. , $`W^+(U(v))=\alpha ϵ,W^{}(U(v)=\pi \alpha ϵ`$. , $`U(v)`$ — . - , , $`ϵ\alpha \pi ϵ`$, , , $`U(v)`$ — . , $`U(v)`$$`\alpha _1`$ $`\alpha _2`$. , $`\alpha _1\alpha _2`$. $`\alpha (v)=\alpha _0`$. $`\alpha `$ , $`\alpha _0\alpha _1`$. – $`\alpha _1+\alpha _2=\omega ^+(U(v))\omega ^{}(U(v))`$. $`\alpha _1+\alpha _2\omega ^+(U(v))\alpha _0ϵ\alpha _1ϵ<\alpha _1+\alpha _2`$, . , $`U(v)`$ — , , , $`\varphi (v)=\varphi (v)=(0,0)`$, , ,
$`U(v)`$, ,, .
, , . .
|
warning/0506/hep-ph0506083.html
|
ar5iv
|
text
|
# Global analysis of three-flavor neutrino masses and mixings
## 1 Introduction
Neutrinos provide, on a macroscopic scale, the realization of two key concepts of quantum mechanics: linear superposition of states and noncommuting operators. In fact, there is compelling experimental evidence that the three known neutrino states with definite flavor ($`\nu _e`$, $`\nu _\mu `$ and $`\nu _\tau `$) are linear combinations of states with definite mass $`\nu _i`$ ($`i=1,2,3`$), and that the Hamiltonian of neutrino propagation in vacuum and matter does not commute with flavor. The effects of flavor nonconservation (“oscillations”) take place on macroscopic distances, for typical ultrarelativistic neutrinos. The evidence for such effects comes from a series of experiments performed during about four decades of research with very different neutrino beams and detection techniques: the solar neutrino experiments Homestake , Kamiokande , SAGE , GALLEX-GNO , Super-Kamiokande (SK) and Sudbury Neutrino Observatory (SNO) ; the long-baseline reactor neutrino experiment KamLAND ; the atmospheric neutrino experiments Kamiokande , Super-Kamiokande , MACRO , and Soudan-2 ; and the long-baseline accelerator neutrino experiment KEK-to-Kamioka (K2K) .
Together with the null results from the CHOOZ (and Palo Verde ) short-baseline reactor experiments, the above oscillation data provide stringent constraints on the basic parameters governing the quantum aspects of neutrino propagation, namely, the superposition coefficients between flavor and mass states (i.e., the neutrino mixing matrix), the energy levels of the Hamiltonian in vacuum (i.e., the splittings between squared neutrino masses), and the analogous levels in matter (i.e., the neutrino interaction energies). The energy levels at rest (i.e., the absolute neutrino masses) are being probed by different, non-oscillation searches: beta decay experiments , neutrinoless double beta decay searches ($`0\nu 2\beta `$) , and precision cosmology . Current non-oscillation data provide only upper limits on neutrino masses, except for the claim by part of the Heidelberg-Moscow experimental collaboration , whose possible $`0\nu 2\beta `$ signal would imply a lower bound on neutrino masses.
A highly nontrivial result emerging from these different neutrino data sample is their consistency, at a very detailed level, with the simplest extension of the standard electroweak model needed to accommodate nonzero neutrino masses and mixings, namely, with a scenario where the three known flavor states $`\nu _{e,\mu ,\tau }`$ are mixed with only three mass states $`\nu _{1,2,3}`$, no other states or new neutrino interactions being needed. This “standard three-neutrino framework” (as recently reviewed, e.g., in ) appears thus as a new paradigm of particle and astroparticle physics, which will be tested, refined, and possibly challenged by a series of new, more sensitive experiments planned for the next few years or even for the next decades . The first challenge might actually come very soon from the running MiniBooNE experiment , which is probing the only piece of data at variance with the standard three-neutrino framework, namely, the controversial result of the Liquid Scintillator Neutrino Experiment (LSND) .
In this review we focus on the current status of the standard three-neutrino framework and on the neutrino mass and mixing parameters which characterize it, as derived from a comprehensive, state-of-the-art analysis of a large amount of oscillation and nonoscillation neutrino data (as available in August 2005). All the results and figures shown in the review are either new or updated or improved in various ways, with respect to our previous publications in the field of neutrino phenomenology. In this sense we have tried to be as complete as possible, so as to present a self-consistent overview of the current status of the three-neutrino mass-mixing parameters. While we aimed at obtaining technically accurate and complete results, we have not aimed at being bibliographically complete; we refer the reader to for an incomplete list of excellent reviews with rich bibliographies of old and recent neutrino papers.
## 2 Notation
While for quark mixing a standard notation and parametrization has emerged , this is not (yet) the case for (some) neutrino mass and mixing parameters. In this section we define and motivate the conventions used hereafter.
### 2.1 Mixing angles and CP-violating phase
At the lagrangian level, the left-handed neutrino fields with definite flavors $`\nu _{\alpha L}`$ $`(\alpha =e,\mu ,\tau )`$ are assumed to be linear superpositions of the neutrino fields with definite masses $`\nu _{iL}`$ $`(i=1,2,3)`$, through a unitary complex matrix $`U_{\alpha i}`$:
$$\nu _{\alpha L}=\underset{i=1}{\overset{3}{}}U_{\alpha i}\nu _{iL}.$$
(1)
This convention implies that one-particle neutrino states $`|\nu `$ are instead related by $`U^{}`$ (see, e.g., ),
$$|\nu _\alpha =\underset{i=1}{\overset{3}{}}U_{\alpha i}^{}|\nu _i.$$
(2)
A common parameterization for the matrix $`U`$ is:
$$U=O_{23}\mathrm{\Gamma }_\delta O_{13}\mathrm{\Gamma }_\delta ^{}O_{12},$$
(3)
where the $`O_{ij}`$’s are real Euler rotations with angles $`\theta _{ij}[0,\pi /2]`$ , while $`\mathrm{\Gamma }_\delta `$ embeds a CP-violating phase $`\delta [0,2\pi ]`$,
$$\mathrm{\Gamma }_\delta =\mathrm{diag}(1,1,e^{+i\delta }).$$
(4)
Notice that the above definitions imply $`\mathrm{det}(U)=+1`$, which may be a useful property in some theoretical contexts .
By considering $`\mathrm{\Gamma }_\delta O_{13}\mathrm{\Gamma }_\delta ^{}`$ as a single (complex) rotation, this parametrization coincides with the one recommended (together with Eq. (2)) in the Review of Particle Properties ,
$$U=\left(\begin{array}{ccc}1& 0& 0\\ 0& c_{23}& s_{23}\\ 0& s_{23}& c_{23}\end{array}\right)\left(\begin{array}{ccc}c_{13}& 0& s_{13}e^{i\delta }\\ 0& 1& 0\\ s_{13}e^{i\delta }& 0& c_{13}\end{array}\right)\left(\begin{array}{ccc}c_{12}& s_{12}& 0\\ s_{12}& c_{12}& 0\\ 0& 0& 1\end{array}\right)$$
(5)
where $`c_{ij}=\mathrm{cos}\theta _{ij}`$ and $`s_{ij}=\mathrm{sin}\theta _{ij}`$.
Other conventions sometimes used in the literature involve $`U`$ instead of $`U^{}`$ in Eq. (2), or $`\delta `$ instead of $`+\delta `$ in Eq. (4), or only one CP-violating factor (either $`\mathrm{\Gamma }_\delta `$ or $`\mathrm{\Gamma }_\delta ^{}`$, not both) in Eq. (3), or a combination of the above. In our opinion such alternatives, although legitimate, do not bring particular advantages over the above convention.
For the sake of simplicity, the phase $`\delta `$ will not be considered in full generality in this work. Numerical examples will refer, when needed, only to the two inequivalent CP-conserving cases, namely, $`e^{i\delta }=\pm 1`$. In these two cases, the mixing matrix takes a real form $`U_{\mathrm{CP}}`$,
$$U_{\mathrm{CP}}=\left(\begin{array}{ccc}c_{13}c_{12}& s_{12}c_{13}& \pm s_{13}\\ s_{12}c_{23}s_{23}s_{13}c_{12}& c_{23}c_{12}s_{23}s_{13}s_{12}& s_{23}c_{13}\\ s_{23}s_{12}s_{13}c_{23}c_{12}& s_{23}c_{12}s_{13}s_{12}c_{23}& c_{23}c_{13}\end{array}\right),$$
(6)
where the upper (lower) sign refers to $`\delta =0`$ ($`\delta =\pi `$). The two cases are formally related by the replacement $`s_{13}s_{13}`$. In any case, CP violation effects do not affect at all solar and reactor oscillation searches, where the indistinguishability of $`\nu _\mu `$ and $`\nu _\tau `$ in the final state allows to rotate away both the angle $`\theta _{23}`$ and the CP phase $`\delta `$ from the parameter space, even in the presence of matter effects (see, e.g., ).
### 2.2 Masses, splittings and hierarchies
The current neutrino phenomenology implies that the three-neutrino mass spectrum $`\{m_i\}_{i=1,2,3}`$ is formed by a “doublet” of relatively close states and by a third “lone” neutrino state, which may be either heavier than the doublet (“normal hierarchy,” NH) or lighter (“inverted hierarchy,” IH).<sup>1</sup><sup>1</sup>1In this context, “hierarchy” does not refers to neutrino masses but only to mass differences. In particular, it is not excluded that such differences can be much smaller than the masses themselves—a scenario often indicated as “degenerate mass spectrum.” In the most frequently adopted labeling of such states, the lightest (heaviest) neutrino in the doublet is called $`\nu _1`$ ($`\nu _2`$), so that their squared mass difference is
$$\delta m^2=m_2^2m_1^2>0$$
(7)
by convention. The lone state is then labeled as $`\nu _3`$, and the physical sign of $`m_3^2m_{1,2}^2`$ distinguishes NH from IH.<sup>2</sup><sup>2</sup>2Another convention, sometimes used in the literature, labels the states so that $`m_1<m_2<m_3`$ in both NH and IH. In this case, however, the mixing angles $`\theta _{ij}`$ have a different meaning in NH and IH.
Very often, the second independent squared mass difference $`\mathrm{\Delta }m^2`$ is taken to be either $`m_3^2m_1^2`$ or $`m_3^2m_2^2`$. However, these two definitions may not be completely satisfactory in both hierarchies. In fact, in passing from NH to IH, the difference $`m_3^2m_1^2`$ not only changes its sign, but also changes from being the largest squared mass gap to being the next-to-largest gap (while the opposite happens for $`m_3^2m_2^2`$). Whenever terms of O($`\delta m^2/\mathrm{\Delta }m^2`$) are relevant, this fact makes somewhat tricky the comparison of results obtained in different hierarchies. For such reason, we prefer to define $`\mathrm{\Delta }m^2`$ as
$$\mathrm{\Delta }m^2=\left|m_3^2\frac{m_1^2+m_2^2}{2}\right|,$$
(8)
so that the two hierarchies are simply related by the transformation $`+\mathrm{\Delta }m^2\mathrm{\Delta }m^2`$. The largest and next-to-largest squared mass gaps are given $`\mathrm{\Delta }m^2\pm \delta m^2/2`$ in both cases.
More precisely, the squared mass matrix
$$M^2=\mathrm{diag}(m_1^2,m_2^2,m_3^2)$$
(9)
reads, in our conventions,
$$M^2=\frac{m_1^2+m_2^2}{2}\mathbf{\hspace{0.17em}1}+\mathrm{diag}(\frac{\delta m^2}{2},+\frac{\delta m^2}{2},\pm \mathrm{\Delta }m^2),$$
(10)
where the upper (lower) sign refers to normal (inverted) hierarchy.
In the previous equation, the term proportional to the unit matrix $`\mathrm{𝟏}`$ is irrelevant in neutrino oscillations, while it matters in observables sensitive to the absolute neutrino mass scale, such as in $`\beta `$-decay and precision cosmology. In particular, we remind that $`\beta `$-decay experiments are sensitive to the so-called effective electron neutrino mass $`m_\beta `$,
$$m_\beta =\left[\underset{i}{}|U_{ei}|^2m_i^2\right]^{\frac{1}{2}}=\left[c_{13}^2c_{12}^2m_1^2+c_{13}^2s_{12}^2m_2^2+s_{13}^2m_3^2\right]^{\frac{1}{2}},$$
(11)
as far as the single $`\nu _i`$ mass states are not experimentally resolvable . On the other hand, precision cosmology is sensitive, to a good approximation (up to small hierarchy-dependent effects which may become important in next-generation precision measurements ) to the sum of neutrino masses $`\mathrm{\Sigma }`$ ,
$$\mathrm{\Sigma }=m_1+m_2+m_3.$$
(12)
### 2.3 Majorana phases
If neutrinos are indistinguishable from their antiparticles (i.e., if they are Majorana rather than Dirac neutrinos), the mixing matrix $`U`$ acquires a (diagonal) extra factor
$$UUU_M,$$
(13)
which is parametrized in various ways in the literature. In particular, within the Review of Particle Properties, two different conventions are used . We adopt the one in , which—after a slight change in notation—reads:
$$U_M=\mathrm{diag}(1,\mathrm{e}^{\frac{\mathrm{i}}{2}\varphi _2},\mathrm{e}^{\frac{\mathrm{i}}{2}(\varphi _3+2\delta )}),$$
(14)
$`\varphi _2`$ and $`\varphi _3`$ being unknown Majorana phases. The “advantage” of this convention is that, in the expression of the effective Majorana mass $`m_{\beta \beta }`$ probed in neutrinoless double beta decay $`(0\nu 2\beta )`$ experiments , the CP-violating phase $`\delta `$ is formally absent:
$$m_{\beta \beta }=\left|\underset{i}{}U_{ei}^2m_i\right|=\left|c_{13}^2c_{12}^2m_1+c_{13}^2s_{12}^2m_2e^{i\varphi _2}+s_{13}^2m_3e^{i\varphi _3}\right|.$$
(15)
### 2.4 Matter effects
In the flavor basis, the hamiltonian of ultrarelativistic ($`m_ip`$) neutrino propagation in matter reads
$$H=\frac{1}{2E}UM^2U^{}+V_{\mathrm{MSW}},$$
(16)
up to an irrelevant momentum term $`p\mathrm{𝟏}`$ which, acting as a zero-point energy, produces only an unobservable overall phase in flavor oscillation phenomena. In the above equation, $`V_{\mathrm{MSW}}=\mathrm{diag}(V,0,0)`$ is the Mikheyev-Smirnov-Wolfenstein (MSW) term embedding the interaction energy difference (or “neutrino potential”),
$$V(x)=\sqrt{2}G_FN_e(x),$$
(17)
$`E`$ being the neutrino energy, and $`N_e`$ the electron density at the position $`x`$. For antineutrinos, one has to replace $`UU^{}`$ and $`VV`$. We shall also use an auxiliary variable with the dimensions of a squared mass ,
$$A(x)=2EV=2\sqrt{2}G_FN_e(x)E.$$
(18)
Matter effects are definitely important when one squared mass difference (either $`\delta m^2`$ or $`\mathrm{\Delta }m^2`$) is of the same order of magnitude as $`A(x)`$.
When needed, the eigenvalues of $`H`$ in matter will be denoted as $`\stackrel{~}{m}_i^2/2E`$, and the diagonalizing matrix as $`\stackrel{~}{U}`$ (with rotation angles $`\stackrel{~}{\theta }_{ij}`$):
$$H=\frac{1}{2E}\stackrel{~}{U}\stackrel{~}{M}^2\stackrel{~}{U}^{}.$$
(19)
The eigenvalue labeling is fixed by the condition $`\stackrel{~}{m}_i^2m_i^2`$ for $`A(x)0`$. The parameters $`\stackrel{~}{m}_i`$ and $`\stackrel{~}{\theta }_{ij}`$ are often called “effective” neutrino masses and mixing angles in matter.
We remind that, in the absence of matter effect, and within the two CP-conserving cases $`(e^{i\delta }=\pm 1U=U^{})`$, the (vacuum) flavor oscillation probability $`P_{\alpha \beta }=P(\nu _\alpha \nu _\beta )`$ takes the form
$$P_{\alpha \beta }^{\mathrm{vac}}=\delta _{\alpha \beta }4\underset{i<j}{}U_{\alpha i}U_{\alpha j}U_{\beta i}U_{\beta j}\mathrm{sin}^2\left(\frac{m_i^2m_j^2}{4E}L\right),$$
(20)
where $`L`$ is the neutrino pathlength. The same functional form is retained in matter with constant density, but with mass-mixing parameters $`(\theta _{ij},m_i^2m_j^2)`$ replaced by their effective values in matter $`(\stackrel{~}{\theta }_{ij},\stackrel{~}{m}_i^2\stackrel{~}{m}_j^2)`$ :
$$P_{\alpha \beta }^{\mathrm{mat}}=\delta _{\alpha \beta }4\underset{i<j}{}\stackrel{~}{U}_{\alpha i}\stackrel{~}{U}_{\alpha j}\stackrel{~}{U}_{\beta i}\stackrel{~}{U}_{\beta j}\mathrm{sin}^2\left(\frac{\stackrel{~}{m}_i^2\stackrel{~}{m}_j^2}{4E}L\right).$$
(21)
For non-constant matter density, $`P_{\alpha \beta }`$ cannot be generally cast in compact form and may require numerical evaluation, although a number of analytical approximations can be found in the literature for specific classes of density profiles.
### 2.5 Conventions on confidence level contours
In this work, the constraints on the neutrino oscillation parameters have been obtained by fitting accurate theoretical predictions to a large set of experimental data, through either least-square or maximum-likelihood methods. In both cases, parameter estimations reduce to finding the minimum of a $`\chi ^2`$ function (see the Appendix) and to tracing iso-$`\mathrm{\Delta }\chi ^2`$ contours around it.
Hereafter, we adopt the convention used in and call “region allowed at $`n\sigma `$” the subset of the parameter space obeying the inequality
$$\mathrm{\Delta }\chi ^2n^2.$$
(22)
The projection of such allowed region onto each single parameter provides the $`n\sigma `$ bound on such parameter. In particular, we shall also directly use the relation $`\sqrt{\mathrm{\Delta }\chi ^2}=n`$ to derive allowed parameter ranges at $`n`$ standard deviations.
## 3 Solar neutrinos and KamLAND
In this section we present an updated analysis of the constraints on the mass-mixing parameters placed by oscillation searches with solar neutrino detectors and long-baseline reactors (KamLAND) in the parameter space $`(\delta m^2,\mathrm{sin}^2\theta _{12},\mathrm{sin}^2\theta _{13})`$. We start with the limiting case $`\theta _{13}0`$, and discuss in detail the bounds on $`(\delta m^2,\theta _{12})`$. We also discuss the current evidence for the occurrence of matter effects in the Sun, and then describe some details of the statistical analysis. We conclude the section by discussing the more general case with $`\theta _{13}`$ unconstrained.
Some technical remarks are in order. The latest KamLAND results are analyzed through a maximum-likelihood approach including the event-by-event energy spectrum . Here we do not include the additional time information available in which, as discussed in , does not improve significantly the bounds on the oscillation parameters. Solar neutrino data are analyzed through the pull method discussed in . With respect to , Chlorine and Super-Kamiokande data are unchanged, while Gallium results have been updated . In addition to the SNO-I (no salt) results already discussed in , we include in this work the complete SNO-II data (with salt) , namely, day and night charged-current (CC) spectra (17+17 bins), and global day and night neutral-current (NC) and elastic-scattering (ES) event rates (2+2 bins), together with 16 new sources of correlated systematic errors affecting the theoretical predictions . Correlations of statistical errors (treated as in ) in SNO-II data are also included. Some of the SNO systematics are highly asymmetrical and even one-sided , and their statistical treatment is not obvious. We have chosen to apply the prescription proposed in to deal with combinations of asymmetric errors: for each $`i`$-th pair of asymmetric errors $`(\sigma _i^+,\sigma _i^{})`$ affecting a theoretical quantity $`R`$, we apply the pull method to the shifted theoretical quantity $`R+\mathrm{\Delta }R_i`$ with symmetric errors $`\pm \sigma _i`$, where $`2\mathrm{\Delta }R_i=\sigma _i^+\sigma _i^{}`$ and $`2\sigma _i=\sigma _i^++\sigma _i^{}`$ . Care must be taken to account for relative bin-to-bin error signs. We understand that the SNO approach to asymmetric errors (not explicitly described in ) is different from ours ; this fact might account for some differences in our allowed regions, which appear to be somewhat more conservative at high-$`\delta m^2`$ values, as compared with those in . Finally, the input standard solar model (SSM) used in this work is the one developed by Bahcall and Serenelli (BS) in by using a new input (Opacity Project, OP) for the opacity tables and older heavy-element abundances consistent with helioseismology . In this SSM (denoted as “BS05 (OP)” in ), the “metallicity” systematics, previously lumped into a single uncertainty, are now split into 9 element components. In total, our solar neutrino data analysis accounts for 119 observables \[1 Chlorine + 2 Gallium (total rate and winter-summer asymmetry) + 44 SK + 34 SNO-I + 38 SNO-II\] and 55 (partly correlated)<sup>3</sup><sup>3</sup>3All the SSM sources of uncertainties are independent, with the exception of some of those concerning the “new” SSM metallicities, whose correlations are taken as recommended in . systematic error sources. Further technical details can be found in the Appendix.
### 3.1 Solar and KamLAND constraints ($`\theta _{13}=0`$)
For $`\theta _{13}=0`$, electron neutrinos are a mixture of $`\nu _1`$ and $`\nu _2`$ only. So, the parameter space relevant for solar $`\nu _e`$’s and KamLAND $`\overline{\nu }_e`$’s reduces to the two variables governing the $`(\nu _1,\nu _2)`$ oscillations, namely, $`\delta m^2`$ and $`\theta _{12}`$. Trigonometric functions useful to plot $`\theta _{12}`$ are either $`\mathrm{tan}^2\theta _{12}`$ in logarithmic scale or $`\mathrm{sin}^2\theta _{12}`$ in linear scale; these choices graphically preserve octant symmetry ($`\theta _{12}\pi /2\theta _{12}`$) when applicable (e.g., in the limit of vacuum oscillations).
Figure 1 shows the current solar neutrino constraints from separate data sets (Chlorine, Gallium, SK, SNO) at the $`2\sigma `$ level, using the BS05 (OP) SSM input . In each panel, we also superpose the small region allowed at $`2\sigma `$ around $`\delta m^2\mathrm{few}\times 10^5`$ eV<sup>2</sup> and $`\mathrm{tan}^2\theta _{12}\mathrm{few}\times 10^1`$, which provides the solution to the solar neutrino problem at large mixing angle (LMA). In Fig. 1 one can appreciate that the global LMA solution completely overlaps with each of the regions separately allowed by the different experimental data (at $`2\sigma `$), i.e., there is a strong consistency between different observations. The shape of the global solar LMA solution appears to be dominated by SNO and (to a lesser extent) by the SK experiment. Since both SK and SNO are sensitive to the high-energy tail of the solar neutrino spectrum (i.e., to the <sup>8</sup>B neutrino flux ), and since the SNO determination of the <sup>8</sup>B flux is already a factor of two more accurate than the corresponding prediction in the BS05 (OP) standard solar model (see next Sec. 3.2), the shape of the global LMA solution in Fig. 1 is rather robust with respect to possible variations in the standard solar model input (including those related to the recent chemical controversy about the solar photospheric metallicity ).
Notice that current solar neutrino data, by themselves, identify a unique (LMA) solution in Fig. 1; this was not the case only a few years ago (see, e.g., ), when at least another region at low $`\delta m^2`$ (“LOW” solution) was allowed . From a test of hypothesis, we get that the current probability of the LOW solution is only $`P_{\mathrm{LOW}}=1.2\times 10^3`$. Former solutions in the vacuum oscillation regime (VAC) or at small mixing angle (SMA) (with acronyms taken, e.g., from ) are now characterized by exceedingly low probabilities ($`P_{\mathrm{VAC}}=4.8\times 10^6`$ and $`P_{\mathrm{SMA}}=4.0\times 10^8`$ from current solar neutrino data).
The LMA solution is heavily affected by solar matter (MSW) effects (see, e.g., for a recent review of the LMA-MSW properties). Figure 2 shows the neutrino potential $`V(x)`$ as a function of the normalized Sun radius $`x=R/R_{}`$, together with typical solar $`\nu _e`$ production regions (in arbitrary vertical scale), as taken from the BS05 (OP) model . From this figure one can easily derive that, for $`\delta m^2`$ values in the LMA region, matter effects are definitely important ($`\delta m^2A(x)`$) for neutrinos with $`E\mathrm{few}`$ MeV. More precisely, the LMA solar $`\nu _e`$ survival probability at the Earth \[$`P_{ee}=P(\nu _e\nu _e)`$\] reads
$$P_{ee}=\frac{1}{2}+\frac{1}{2}\mathrm{cos}2\stackrel{~}{\theta }_{12}(x)\mathrm{cos}2\theta _{12},$$
(23)
where
$$\mathrm{cos}2\stackrel{~}{\theta }_{12}=\frac{\mathrm{cos}2\theta _{12}A(x)/\delta m^2}{\sqrt{(\mathrm{cos}2\theta _{12}A(x)/\delta m^2)^2+\mathrm{sin}^22\theta _{12}}},$$
(24)
with $`\mathrm{cos}2\stackrel{~}{\theta }_{12}`$ slowly changing from its vacuum value $`(\mathrm{cos}2\theta _{12})`$ to its matter-dominated values (close to $`1`$) as $`E`$ increases from sub-MeV to multi-MeV values.
Figure 3 shows the energy profile of $`P_{ee}`$, averaged over the production regions relevant to pp, <sup>7</sup>Be, and <sup>8</sup>B solar neutrinos, for representative LMA oscillation parameters. Also shown are the energy profiles of corresponding solar $`\nu _e`$ fluxes (in arbitrary vertical scale). The value of $`P_{ee}`$ decreases from its vacuum value ($`10.5\mathrm{sin}^22\theta _{12}`$) to its matter-dominated value $`(\mathrm{sin}^2\theta _{12})`$ as the energy increases. The vacuum-matter transition is faster for neutrinos produced in the inner regions of the Sun. In Fig. 3 we also show the small difference between day (D) and night (N) curves, due to matter effects in the Earth<sup>4</sup><sup>4</sup>4The treatment of Earth matter effects in the present work is the same as in but with eight density shells . (calculated, for definiteness, at the SNO latitude). The vacuum-matter transition is slightly slower during the night, due to the Earth regeneration effect (see and references therein). Within current energy thresholds and experimental uncertainties, the vacuum-matter transition and the Earth regeneration effects have not been yet observed in the SK and SNO time-energy spectra. Nevertheless, as we shall see later, matter effects in the Sun must definitely occur to explain the data.
Let us consider now the impact of KamLAND data. For typical LMA parameters, reactor $`\overline{\nu }_e`$ are expected to have a relatively large oscillation amplitude ($`\mathrm{sin}^22\theta _{12}`$), as well as a sizable oscillation phase \[$`\delta m^2L/4EO(1)`$\] over long baselines ($`LO(10^2)\mathrm{km}`$). The $`\overline{\nu }_e`$ disappearance signal observed in KamLAND has not only confirmed the solar LMA solution but has greatly reduced its $`\delta m^2`$ range , by observing a strong distortion in the energy spectrum . Figure 4 shows the mass-mixing parameter regions separately allowed by the KamLAND total rate, by the energy spectrum shape, and by their combination, at the 1, 2, and $`3\sigma `$ level, as obtained by our unbinned maximum-likelihood analysis of the latest energy spectrum data . The overall reactor neutrino disappearance (rate information) and its energy distribution (shape information) are highly consistent, the latter being dominant in the combination. At the $`2\sigma `$ level, both the shape-only and the rate+shape analyses identify a single solution at $`\delta m^28\times 10^5`$ eV<sup>2</sup> and large mixing; only at the $`3\sigma `$ level two disconnected solutions appear at higher and lower values of $`\delta m^2`$. Notice the linear scale on both axes, and the reduction of the parameter space, as compared with Fig. 1.
Figure 5 shows the regions separately allowed by all solar neutrino data and by KamLAND, both separately and in combination, at the 1, 2, and 3$`\sigma `$ level. The current solar LMA solution, as compared with results prior to complete SNO-II data (see, e.g., , is slightly shifted toward larger values of $`\mathrm{sin}^2\theta _{12}`$ and allows higher values of $`\delta m^2`$.<sup>5</sup><sup>5</sup>5Our current best-fit point for solar data only is at $`\delta m^2=6.3\times 10^5\mathrm{eV}^2`$ and $`\mathrm{sin}^2\theta _{12}=0.314.`$ This trend is substantially due to the larger value of the CC/NC ratio measured in the complete SNO II phase (0.34 ) with respect to the previous central value (0.31 ). We also find that the SNO-II CC spectral data contribute to allow slightly higher values of $`\delta m^2`$ with respect to older results. The consistency of solar and reactor allowed regions is impressive, with a large overlap even at the $`1\sigma `$ level, and with very close best-fit points. The solar+KamLAND combination eliminates the extra (KamLAND-only) solutions at high and low $`\delta m^2`$, and identifies a single allowed region characterized by the following $`2\sigma `$ ranges:
$$\delta m^2=7.92\times 10^5\mathrm{eV}^2(1\pm 0.09)\mathrm{at}\pm 2\sigma ,$$
(25)
$$\mathrm{sin}^2\theta _{12}=0.314(1_{0.15}^{+0.18})\mathrm{at}\pm 2\sigma .$$
(26)
The determination of these two parameters at $`O(10\%)`$ level represents one of the most remarkable successes of the last few years in neutrino physics.
The $`\delta m^2`$ uncertainty is currently dominated by the KamLAND observation of half-period of oscillations and can be improved with higher statistics . The $`\mathrm{sin}^2\theta _{12}`$ uncertainty is instead dominated by the SNO ratio of CC to NC events, which is a direct measurement of $`P_{ee}`$ at high energy: $`R_{\mathrm{CC}}/R_{\mathrm{NC}}P_{ee}\mathrm{sin}^2\theta _{12}`$. Figure 6 shows isolines of this ratio in the mass-mixing parameter space, which can be used as a guidance to understand the effect of prospective SNO measurements on the $`\mathrm{sin}^2\theta _{12}`$ range. In the same figure we show isolines of the day-night asymmetry ($`A_{\mathrm{DN}}`$) of CC events in SNO, whose measurement could, in principle, help to reduce the $`\delta m^2`$ uncertainty ; however, it is unlikely that the SNO errors can be reduced enough ($`<1\%`$) to clearly observe a day-night effect (see, e.g., ).
### 3.2 Evidence for matter effects in the Sun
As shown in Fig. 3, solar matter effects make $`P_{ee}`$ decrease from its vacuum value ($`>1/2`$) to a matter-dominated value ($`<1/2`$), for typical LMA parameters around the current best fit. Model-independent tests of the presence of matter effects are derived in the following, first by showing that SK and SNO data consistently indicate that $`P_{ee}<1/2`$ in their energy range, and secondly by showing that all solar+reactor data consistently indicate that the neutrino potential $`V(x)`$ must be nonzero.
As discussed in , the normalized energy spectra of neutrinos which do interact in SK and in SNO (i.e., the SK and SNO “response functions” to the incoming <sup>8</sup>B neutrinos) can be equalized, to a good approximation, by choosing a proper SK energy threshold, for any given SNO threshold. The current best “equalization” of SK and SNO response functions is shown in Fig. 7. In this case, both SK and SNO are sensitive to the same energy-averaged survival probability, $`P_{ee}`$. Moreover, the CC and NC event rates in SNO, together with the ES event rate in SK, overconstrain $`P_{ee}`$ and the unoscillated <sup>8</sup>B solar neutrino flux $`\mathrm{\Phi }_B`$ in a completely model-independent way<sup>6</sup><sup>6</sup>6For purely active (no sterile) neutrino flavor transitions. (i.e., independently of the mass-mixing parameters and of the standard solar model) through the equations
$`\mathrm{\Phi }_{\mathrm{ES}}^{\mathrm{SK}}`$ $`=`$ $`\mathrm{\Phi }_B[P_{ee}+r_\sigma (1P_{ee})],`$ (27)
$`\mathrm{\Phi }_{\mathrm{CC}}^{\mathrm{SNO}}`$ $`=`$ $`\mathrm{\Phi }_BP_{ee},`$ (28)
$`\mathrm{\Phi }_{\mathrm{NC}}^{\mathrm{SNO}}`$ $`=`$ $`\mathrm{\Phi }_B,`$ (29)
where $`r_\sigma 0.154`$ is the ratio of the energy-averaged ES cross-sections of $`\nu _{\mu ,\tau }`$ and $`\nu _e`$ in SK.
Figure 8 shows the current bounds at $`2\sigma `$ on $`\mathrm{\Phi }_B`$ and $`P_{ee}`$, as obtained by using the latest SNO CC and NC and SK ES event rates, both separately (bands) and in combination ($`2\sigma `$ elliptical regions). The dotted ellipse represent the combination of SNO NC and CC data; the addition of SK ES data—which are consistent with SNO NC and CC data—slightly increases the preferred value of $`\mathrm{\Phi }_B`$ (solid ellipse). In particular, the SNO+SK combination (dominated by SNO) provides the following ranges:
$$\mathrm{\Phi }_B=5.2_{0.8}^{+0.7}\times 10^6\mathrm{cm}^2\mathrm{s}^1(\pm 2\sigma ),$$
(30)
$$P_{ee}=0.34_{0.06}^{+0.08}(\pm 2\sigma ).$$
(31)
The above SNO+SK range for $`\mathrm{\Phi }_B`$ is consistent with the $`\pm 2\sigma `$ prediction of the BS05(OP) standard solar model , $`\mathrm{\Phi }_B^{\mathrm{SSM}}=5.7(1\pm 0.32)\times 10^6\mathrm{cm}^2\mathrm{s}^1`$, the difference in the central values ($`10\%`$) being not statistically significant, as also evident in Fig. 8. Notice that the SK+SNO data determine $`\mathrm{\Phi }_B`$ with an error a factor of 2 smaller than the SSM prediction. At the same time, the SK+SNO data constrain $`P_{ee}`$ to be definitely less than $`1/2`$ , and in particular close to $`1/3`$, as predicted for high-energy <sup>8</sup>B neutrinos and LMA parameters (see Fig. 3). The model-independent SNO+SK analysis is thus fully consistent with the LMA-MSW expectations; removal of the MSW effect in the LMA region would give a prediction $`P_{ee}=10.5\mathrm{sin}^22\theta _{12}>1/2`$, inconsistently with the results in Fig. 8.<sup>7</sup><sup>7</sup>7For $`\theta _{13}>0`$, the no-MSW prediction would be slightly modified as $`P_{ee}>0.5c_{13}^4+s_{13}^4=0.46`$ (using the $`3\sigma `$ upper limit $`s_{13}^2<0.047`$ discussed later), still inconsistently with Fig. 8.
One can perform, however, a more powerful test of the presence of the neutrino potential $`V(x)`$, by artificially altering its magnitude through a free parameter $`a_{\mathrm{MSW}}`$ ,
$$V(x)a_{\mathrm{MSW}}V(x),$$
(32)
both in the Sun (relevant for solar neutrino oscillations) and in the Earth (relevant for both solar and reactor neutrino oscillations), and by renalyzing solar and KamLAND data with $`a_{\mathrm{MSW}}`$ free. Testing matter effects amounts then to reject the case $`a_{\mathrm{MSW}}=0`$ (no effect) and to prove that $`a_{\mathrm{MSW}}=1`$ (standard effect) is favored. In the analysis, we add CHOOZ reactor data, which help to exclude the appearance of spurious high-$`\delta m^2`$ solutions for $`a_{\mathrm{MSW}}1`$ . Figure 9 shows the results of a fit to all the current solar and reactor data in the parameter space $`(\delta m^2,\mathrm{sin}^2\theta _{12},a_{\mathrm{MSW}})`$, marginalized with respect to the first two parameters, in terms of the function $`(\mathrm{\Delta }\chi ^2)^{1/2}=n\sigma `$. The preference for standard matter effects $`(a_{\mathrm{MSW}}=1)`$ is really impressive, and is currently even more pronounced then with previous data . The hypothetical case of no matter effects $`(a_{\mathrm{MSW}}=0)`$ is rejected at $`>5\sigma `$. Since $`VG_F`$, the results in Fig. 9 can not only be seen as a confirmation of matter effects, but can also be interpreted as an alternative “measurement” of the Fermi constant $`G_F`$ through neutrino oscillations in matter, within a factor of $`2`$ uncertainty at $`2\sigma `$.
### 3.3 Statistical checks
We have seen that, globally, solar neutrino experiments agree with each other and with the KamLAND observation of reactor neutrino disappearance, that solar+KamLAND data identify a restricted range of LMA mass-mixing parameters, and that there is solid evidence for the associated matter effects in such range. However, it makes sense to look at the statistical consistency of the LMA best-fit solution in more detail, for at least two reasons: (1) the analysis involves a large number of observables and of systematics, some of which might deviate from the predictions without really altering the global fit; (2) the preferred shifts of some quantities might reveal something interesting.
We remind that the solar neutrino analysis is performed through the so-called pull approach , namely, by allowing shifts of each $`n`$-th theoretical prediction $`R_n`$ through independent systematic uncertainties $`c_{nk}`$,
$$R_n\overline{R}_n=R_n+\underset{k}{}\xi _kc_{nk},$$
(33)
whose amplitudes $`\xi _k`$ are constrained through a quadratic penalty term. The shifted predictions $`\overline{R}_n`$’s are then compared to the experimental values $`R_n^{\mathrm{exp}}`$ via the uncorrelated (mainly statistical) error components. The method can be generalized to include correlation of statistical and systematic errors. The global $`\chi ^2`$ function is then given by two terms, $`\chi ^2=\chi _{\mathrm{obs}}^2`$ \+ $`\chi _{\mathrm{sys}}^2`$, embedding the quadratic pulls of the observables (i.e., the deviations of theory vs experiment) and of the systematics (i.e., their offset with respect to zero). This method allow a detailed check of possible pathological deviations (pulls) of some quantities. (See also the Appendix.)
Figure 10 shows the pulls of the 119 observables at the global (solar+KamLAND) best-fit point. From top to bottom, the observables include the Chlorine rate, the Gallium rate and its winter-summer asymmetry , the SK distribution in energy and zenith angle (44 bins) , the SNO-I (no-salt) CC spectrum in 17+17 day-night bins , the SNO-II (with salt added) CC spectrum in 17+17 day-night bins and the day-night values of the NC and ES rate . The SNO-II data and their correlations are treated as prescribed in ; for all the other observables we refer the reader to . None of the pulls in Fig. 10 exceeds $`3\sigma `$, and their distribution, which is roughly gaussian, reveals nothing pathological. We conclude that none of the solar neutrino observables shows an anomalous or suspect deviation from the LMA best-fit predictions.
Figure 11 shows the pulls of the 55 systematic errors which enter in the analysis. From top to bottom, they include 11 “old” standard solar systematics as in , 9 “new” SSM metallicity systematics , the <sup>8</sup>B spectrum shape uncertainty , 11 SK and 7 SNO-I systematics as in , and 16 “new” SNO-II systematics . All the offsets are small $`(<1\sigma )`$, indicating that the allowance to shift the theoretical predictions $`R_n`$ through systematic uncertainties is only moderately exploited in the fit; in other words, there is no need to stretch the systematics beyond their stated $`1\sigma `$ range to achieve a good fit.
Finally, Fig. 12 shows a by-product of the pull approach, namely, the preferred shifts of the solar neutrino fluxes with respect to their central SSM values . The $`10\%`$ downward shift of $`\mathrm{\Phi }_B`$ is consistent with the results in Fig. 8. The global fit also prefers a $`10\%`$ reduction of beryllium (Be) and CNO solar neutrino fluxes with respect to the BS05 (OP) prediction—an indication which may be of interest for future experiments directly sensitive to such fluxes . Such preferred reductions are well within SSM uncertainties .
In conclusion, the detailed analysis of the LMA best-fit solution reveals a very good agreement between all single pieces of experimental and theoretical information in the solar neutrino analysis. No statistically alarming deviation is found.
Concerning the statistical analysis of KamLAND data, the adopted maximum-likelihood approach involves only three systematic uncertainties, namely, two free background normalization factors $`\alpha ^{}`$ and $`\alpha ^{\prime \prime }`$ plus one constrained pull $`\alpha `$ for the energy scale offset (the overall rate normalization error being incorporated in the likelihood rate factor, see the Appendix and ). Therefore, our pull analysis for KamLAND involves a single parameter ($`\alpha `$), which we find to be very small ($`\alpha 0.15\sigma _\alpha `$) at the LMA best fit. In addition, as discussed in , the statistical analysis of KamLAND data shows no hints for anomalous effects beyond the standard scenario involving known reactor sources and neutrino flavor disappearance.
### 3.4 Solar and KamLAND constraints ($`\theta _{13}`$ free)
For $`\theta _{13}>0`$, electron neutrino mixing includes also $`\nu _3`$ (besides $`\nu _1`$ and $`\nu _2`$), and $`\mathrm{\Delta }m^2`$-driven oscillations can take place, with amplitude governed by $`\theta _{13}`$. Therefore, the $`\nu _e`$ survival probability for $`\theta _{13}>0`$ ($`P_{3\nu }`$) generally differs from the one for $`\theta _{13}=0`$ ($`P_{2\nu }`$).
In KamLAND, $`\mathrm{\Delta }m^2`$-driven oscillations are so fast to be smeared away by the finite energy resolution, leaving only the average ($`\theta _{13}`$ mixing) effect,
$$P_{3\nu }=c_{13}^4P_{2\nu }+s_{13}^4.$$
(34)
In first approximation, a similar formula holds for solar neutrinos, provided that the neutrino potential $`V`$ is multiplied everywhere by $`\mathrm{cos}^2\theta _{13}`$ (see and refs. therein):
$$P_{3\nu }c_{13}^4P_{2\nu }^{}+s_{13}^4,$$
(35)
$$P_{2\nu }P_{2\nu }^{}=P_{2\nu }|_{Vc_{13}^2V}.$$
(36)
This replacement generates a mild energy-dependence of the correction, which is absent in Eq. (34).
In second approximation, solar neutrinos develop a subleading dependence of $`P_{3\nu }`$ on $`\mathrm{\Delta }m^2`$ and on its sign (i.e., on the hierarchy, see ; such dependence disappears for $`\mathrm{\Delta }m^2\mathrm{}`$, where one recovers the above equations. Accurate analytic expressions for the subleading $`\mathrm{\Delta }m^2`$ effects on $`P_{3\nu }`$ as a function of energy can be found in .
Figure 13 shows the size of leading ($`\theta _{13}`$-driven) and subleading ($`\pm \mathrm{\Delta }m^2`$-driven) effects, through the fractional difference between $`P_{2\nu }`$ and $`P_{3\nu }`$, calculated for the representative value $`s_{13}^2=0.04`$ and for best-fit LMA parameters (and averaged over the <sup>8</sup>B solar neutrino production region, for definiteness). The solid curve is calculated for $`\mathrm{\Delta }m^2=\mathrm{}`$, i.e., no subleading effect; the leading effect (about $`7\%`$) is almost entirely due to the factor $`c_{13}^4`$ in front of $`P_{2\nu }`$, plus a mild energy dependence. The dashed and dot-dashed curves are instead calculated for $`\mathrm{\Delta }m^2=+2.4`$ and $`2.4`$ ($`\times 10^3`$ eV<sup>2</sup>), respectively; their difference from the solid curve quantifies the size of $`\mathrm{\Delta }m^2`$ subleading effects. Although the dependence of $`P_{3\nu }`$ on $`\mathrm{\Delta }m^2`$ and on the hierarchy is theoretically interesting (see, e.g., ), such subleading effect is an order of magnitude smaller than the “leading” $`\theta _{13}`$-effect in Fig. 13, and its inclusion would not change in any appreciable way the analysis of solar neutrino data (as we have explicitly checked). Therefore, in the following, we can safely assume the approximations in Eqs. (35) and (36), i.e., neglect the effect of $`\mathrm{\Delta }m^2`$ and its sign in the solar(+KamLAND) data analysis, as it was the case for older data .
Figure 14 shows the results of our analysis of solar and KamLAND data (both separately and in combination) for unconstrained values of $`\theta _{13}`$, in terms of the $`2\sigma `$ projections of the $`(\delta m^2,\mathrm{sin}^2\theta _{12},\mathrm{sin}^2\theta _{13})`$ allowed region onto each of the three coordinate planes. There is no statistically significant preference for $`\theta _{13}0`$, and upper bounds are placed by both solar and KamLAND data separately.
Concerning KamLAND data only, there is a slight anticorrelation between $`\mathrm{sin}^2\theta _{13}`$ and $`\mathrm{sin}^2\theta _{12}`$ (upper left panel in Fig. 14), since the total rate information constrains both parameters , and a higher $`\mathrm{sin}^2\theta _{13}`$ can be traded for a lower $`\mathrm{sin}^2\theta _{12}`$. However, $`\mathrm{sin}^2\theta _{12}`$ cannot decrease indefinitely—since it would suppress the amplitude of the observed shape distortions —and thus an upper bound on $`\mathrm{sin}^2\theta _{13}`$ emerges in KamLAND.<sup>8</sup><sup>8</sup>8The KamLAND analysis in this work includes event-by-event energy information but not the event time information . We have explicitly checked that the time information, which does not significantly alter the bounds on $`(\delta m^2,\mathrm{sin}^2\theta _{12})`$ , has also negligible effects on the the bounds on $`\mathrm{sin}^2\theta _{13}`$.
We remind that the solar $`\nu `$ sensitivity to $`\mathrm{sin}^2\theta _{13}`$ (Fig. 14) comes from the combination of all solar neutrino experiments, in contrast with the bounds on $`(\delta m^2,\mathrm{sin}^2\theta _{12})`$, which are dominated by the “high energy” <sup>8</sup>B neutrino experiments (SNO and SK). As discussed, e.g., in , for increasing values of $`\theta _{13}`$ a tension arises among different data sets and, in particular, between SNO and Gallium data. Such two experiments, probing respectively the high and low energy part of the solar neutrino spectrum, exhibit different correlation properties between the two mixing parameters $`\theta _{12}`$ and $`\theta _{13}`$. In particular, for increasing values of $`\theta _{13}`$, the SNO and Gallium experiments tend to prefer higher and lower values of $`\mathrm{sin}^2\theta _{12}`$, respectively , worsening the good agreement currently reached at $`\theta _{13}0`$. Therefore, a “collective” effect of different experiments is responsible for the solar neutrino constraints on $`\mathrm{sin}^2\theta _{13}`$. (See also for a discussion of bounds on $`\theta _{13}`$ with earlier data.)
Very interestingly, the combination of solar and KamLAND data in Fig. 14 is now powerful enough to place a combined upper bound on $`\mathrm{sin}^2\theta _{13}`$ at the 5% level at $`2\sigma `$, not much weaker than the bound coming from the CHOOZ plus atmospheric data discussed below in Sec. 4.3 (see also for an earlier discussion of solar+KamLAND constraints on $`\theta _{13}`$). Notice also that, in the combined (solar+KamLAND) regions of Fig. 14, there are negligible correlations among the three parameters; this fact implies that the bounds in Eqs. (25) and (26), derived for $`\theta _{13}=0`$, hold without significant changes also for $`\theta _{13}`$ unconstrained. It also justifies (a posteriori) our choice to discuss in detail the case $`\theta _{13}=0`$, which embeds most of the relevant information on the leading parameters $`(\delta m^2,\mathrm{sin}^2\theta _{12})`$.
## 4 SK atmospheric neutrinos, K2K, and CHOOZ
In this Section we discuss the constraints on the mass-mixing parameters $`(\mathrm{\Delta }m^2,\theta _{23},\theta _{13})`$ coming from the SK atmospheric neutrino detector , from the K2K long-baseline accelerator neutrino experiment , and from the short-baseline CHOOZ reactor neutrino experiment .
Our SK atmospheric neutrino analysis is performed by using the same event classification (binning) and systematic error treatment as in . In particular, we consider (in order of increasing average energy) the zenith angle distributions of the so-called Sub-GeV (SG) electron and muon samples (SG$`e`$ and SG$`\mu `$) in 10 zenith angle $`(\theta _z)`$ bins; Multi-GeV (MG) electron and muon samples (MG$`e`$ and MG$`\mu `$) in 10 zenith angle $`(\theta _z)`$ bins; Upward Stopping muons (US$`\mu `$) in 5 bins; and Upward Through-going muons (UT$`\mu `$) in 10 bins, for a total of 55 accurately computed observables. We include 11 sources of systematic errors with the pull method which allows a better understanding of systematic shifts.<sup>9</sup><sup>9</sup>9The SK Collaboration has used a finer classification of events and systematics in , as well as an alternative $`L/E`$ binning in . Such refined analyses cannot be performed outside the Collaboration. With respect to our previous SK analysis , we use updated results and—unless otherwise stated—atmospheric neutrino input fluxes from the three-dimensional (3D) simulation of (see also for other 3D results).
Concerning the K2K experiment, we use the latest spectrum data from , but regrouped in the same 6 bins as in (by using information from ); this choice is motivated by the fact that information about K2K correlated systematics has been made publicly available only for 6 bins (see and references therein). Finally, the CHOOZ spectral data are analyzed as in . Further technical details are given in the Appendix.
In the following, we discuss first the impact of the SK+K2K data on the neutrino parameters $`(\mathrm{\Delta }m^2,\mathrm{sin}^2\theta _{23})`$ for $`\theta _{13}=0`$. This allows to appreciate the subleading effect induced by nonzero values of $`(\delta m^2,\mathrm{sin}^2\theta _{12})`$, especially on atmospheric neutrinos. Then we consider the more general case $`\theta _{13}0`$, and discuss in some detail the related subleading effects in SK, as well as the constraints from the SK+K2K+CHOOZ analysis.
A final remark is in order. The MACRO and Soudan-2 atmospheric neutrino experiments provide $`(\mathrm{\Delta }m^2,\mathrm{sin}^2\theta _{23})`$ constraints which are consistent with those from SK , but are also affected by larger uncertainties (due to the lower statistics and narrower $`L/E`$ range); they are not included in this work. Similarly, the negative results of the Palo Verde reactor experiment and of the K2K searches in $`\nu _\mu \nu _e`$ appearance mode (consistent with, but less constraining than CHOOZ ) are not included here. Future improved global analyses might take into account these additional data, finer SK and K2K spectral binning, and the covariance of the SK and K2K common systematics (interaction cross section, detector fiducial volume, and event reconstruction errors).
### 4.1 SK and K2K constraints for $`\theta _{13}=0`$ and statistical checks
While for $`\theta _{13}=0`$ the solar+KamLAND parameter space reduces to $`(\delta m^2,\mathrm{sin}^2\theta _{12})`$ exactly, the atmospheric+K2K parameter space reduces to $`(\mathrm{\Delta }m^2,\mathrm{sin}^2\theta _{23})`$ only to a first approximation. Indeed, while the assumption $`\theta _{13}=0`$ forbids solar and reactor $`\nu _e\nu _{\mu ,\tau }`$ transitions involving $`\nu _3`$ and its associated parameters ($`\mathrm{\Delta }m^2,\mathrm{sin}^2\theta _{23}`$), it does not forbid, e.g., atmospheric $`\nu _\mu \nu _\tau `$ transitions involving the pair $`(\nu _1,\nu _2)`$, which depend on the $`\delta m^2`$ parameter. This is most easily seen in the vacuum case, where Eq. (20) implies that, for $`i,j=1,2`$ and even for $`\theta _{13}=0`$, the $`P_{\mu \tau }`$ transition probability contains the following nonzero $`(\nu _1,\nu _2)`$-induced term,
$$4U_{\mu 1}U_{\mu 2}U_{\tau 1}U_{\tau 2}\mathrm{sin}^2\left(\frac{m_2^2m_1^2}{4E}L\right)\stackrel{\theta _{13}=0}{=}4s_{12}^2c_{12}^2s_{23}^2c_{23}^2\mathrm{sin}^2\left(\frac{\delta m^2}{4E}L\right).$$
(37)
The small effect of nonzero (LMA) values of $`(\delta m^2,\mathrm{sin}^2\theta _{12})`$ in the atmospheric neutrino data analysis, phenomenologically noted in for any $`\theta _{13}`$, has often been legitimately neglected (except occasionally, see the bibliography in ), being basically hidden by large statistical and systematic uncertainties. The full implementation of such effect is nontrivial (it requires a numerical $`3\nu `$ evolution in the Earth matter layers), and its main theoretical aspects have been elucidated only recently , in connection with the progressive confirmation and determination of the LMA parameters by solar and KamLAND data, and with the increasing accuracy of atmospheric neutrino data. Although still small, the effect is definitely not smaller than others which are usually taken care of, and deserves to be included in state-of-the-art analyses .
For instance, Fig. 15 shows the results of our analysis of the latest SK data in the plane $`(\mathrm{\Delta }m^2,\mathrm{sin}^2\theta _{23})`$ at $`\theta _{13}=0`$, for three increasingly accurate inputs: Atmospheric neutrino fluxes from one-dimensional (1D) simulations and $`\delta m^2=0`$ (top panel); atmospheric neutrino fluxes from full three-dimensional (3D) simulations and $`\delta m^2=0`$ (middle panel); and finally, 3D fluxes and LMA best-fit values for $`(\delta m^2,\mathrm{sin}^2\theta _{12})`$. In all panels, the three curves refers to 1, 2, and $`3\sigma `$ contours, and the best-fit point is marked by horizontal and vertical lines to guide the eye. One can appreciate that the (now customarily included) 3D flux input shifts $`\mathrm{\Delta }m^2`$ downward by $`0.5\sigma `$ with respect to the 1D flux input; on the other hand, the inclusion of subleading LMA effects shifts $`\mathrm{sin}^2\theta _{23}`$ by $`0.5\sigma `$ with respect to the hypothetical case $`\delta m^2=0`$. As expected, both effects are small, but there is no reason to keep the first and to neglect the second. Moreover, the LMA effect intriguingly breaks the $`\theta _{23}`$ octant sysmmetry, which is in principle an important indication for model building (see, e.g., ). Hereafter, the analysis of the SK atmospheric data will explicitly include nonzero values of $`(\delta m^2,\mathrm{sin}^2\theta _{12})`$, fixed at their best-fit values in Eqs. (2526) but with no uncertainty (whose effect is really negligible). For the sake of completeness, LMA-induced and matter effects will also be included in the calculation of the K2K oscillation probabilities, where, however, such effects are even smaller than in SK, as discussed in Sec. 4.2.
Figure 16 shows, for $`\theta _{13}=0`$, the results of our analysis of SK and K2K data, both separately and in combination. Notice that the top panel in Fig. 16 is the same as the bottom panel in Fig. 15. The K2K constraints are octant-symmetric and relatively weak in $`\mathrm{sin}^2\theta _{23}`$, while they contribute appreciably to reduce the overall $`\mathrm{\Delta }m^2`$ uncertainty. Therefore, not only K2K confirms the neutrino oscillation solution to the atmospheric neutrino anomaly with accelerator neutrinos , but it also helps in reducing the oscillation parameter space. Moreover, there is still room for improvements in K2K. Figure 17 shows our both unoscillated and oscillated K2K spectrum of events (at the SK+K2K best-fit in Fig. 16) in terms of the reconstructed neutrino energy, as used in this work. The oscillated spectrum is shown both with and without the systematic shifts in our pull approach; such shifts are modest as compared with the large statistical errors. Therefore, one can reasonably expect that, with higher statistics, the final K2K data sample can further contribute to reduce the $`\mathrm{\Delta }m^2`$ uncertainty.
Systematic effects are instead quite important in the SK atmospheric neutrino analysis. Fig. 18 shows the ratio of experimental data and of best-fit theoretical predictions (with and without systematic pulls) with respect to no oscillations, as a function of the zenith angle of the scattered lepton ($`e`$ or $`\mu `$), for the five samples used in the analysis. In particular, in terms of Eq. (33), the dashed histograms represent the unshifted theoretical predictions (central values $`R_n`$), while the dashed histograms represent the systematically shifted predictions ($`\overline{R}_n`$) for the given mass-mixing parameters (which correspond to the SK+K2K best-fit point in Fig. 16). Vertical error bars represent the $`1\sigma `$ statistical uncertainties of the data. The electron data (SG$`e`$ and MG$`e`$) show some excess with respect to the unshifted predictions (dashed lines), which tends to be reduced when systematic shifts are allowed (solid lines). Notice that the dashed lines slightly differ from unity for upward $`(\mathrm{cos}\theta _z1)`$ events in the SG$`e`$ and MG$`e`$ sample, as a result of subleading LMA effects ($`\delta m^20`$). A systematic, upward shift of the predictions is also preferred in the high-energy muon samples, US$`\mu `$ and UT$`\mu `$, and especially in the latter, where it amounts to $`20\%`$. The pull analysis of the observables in Fig. 19 tells us that such shifts are not necessarily alarming from a statistical viewpoint, since they are all smaller than two standard deviations. However, their distribution is definitely not random: within each of the six data samples, most of the pulls in Fig. 19 are one-sided, indicating that there seems to be some normalization offset. This is confirmed by the pull analysis of the systematics in Fig. 20, where the two largest pulls ($`1.5\sigma `$) refer to normalization parameters ($`\rho `$ and $`\rho _t`$) which govern the relative normalization of muon samples with increasing energy (fully-contained, partially-contained, and upward-stopping muons) . Also the sub-GeV muon-to-electron flavor ratio error ($`\beta _s`$) is stretched beyond $`1\sigma `$ in Fig. 20. Although there is no alarming “$`3\sigma `$” offset anywhere, it is clear that a better understanding and reduction of the systematic error sources (i.e., atmospheric neutrino fluxes, interaction cross sections, detector uncertainties) is needed if one wants to observe in the future small subleading effects, as those induced by $`\delta m^20`$ and $`\theta _{13}0`$ and discussed in more detail in the next section.
### 4.2 Discussion of subleading effects
Our calculations of atmospheric neutrino oscillations are based on a full three-flavor numerical evolution of the Hamiltonian along the neutrino path in the atmosphere and (below horizon) in the known Earth layers . Semianalytical approximations to the full numerical evolution (although not used in the final results) can, however, be useful to understand the behavior of the oscillation probability and of some atmospheric neutrino observables. A particularly important observable is the excess of expected electron events ($`N_e`$) as compared to no oscillations ($`N_e^0`$):
$$\frac{N_e}{N_e^0}1=(P_{ee}1)+rP_{e\mu },$$
(38)
where $`P_{\alpha \beta }=P(\nu _\alpha \nu _\beta )`$, and $`r`$ is the ratio of atmospheric $`\nu _\mu `$ and $`\nu _e`$ fluxes ($`r2`$ and $`3.5`$ at sub-GeV and multi-GeV energies, respectively). In fact, this quantity is zero when both $`\theta _{13}=0`$ and $`\delta m^2=0`$, and is thus well suited to study the associated subleading effects (which may carry a dependence on the matter density) in cases when $`\delta m^2`$ and $`\theta _{13}`$ are different from zero .
We remind that matter effects are governed by $`A(x)=2\sqrt{2}G_FN_e(x)E`$, with $`N_e2`$ mol/cm<sup>3</sup> in the Earth mantle and $`5`$ mol/cm<sup>3</sup> in the core. It can be easily derived that
$$\frac{A}{\mathrm{\Delta }m^2}1.3\left(\frac{2.4\times 10^3\mathrm{eV}^2}{\mathrm{\Delta }m^2}\right)\left(\frac{E}{10\mathrm{GeV}}\right)\left(\frac{N_e}{2\mathrm{mol}/\mathrm{cm}^3}\right),$$
(39)
implying that Earth matter can substantially affect $`\mathrm{\Delta }m^2`$-driven oscillations \[i.e., $`A/\mathrm{\Delta }m^2O(1)`$\] for $`EO(10)`$ GeV, i.e., in multi-GeV and upward-stopping events. Similarly,
$$\frac{A}{\delta m^2}3.8\left(\frac{8\times 10^5\mathrm{eV}^2}{\delta m^2}\right)\left(\frac{E}{1\mathrm{GeV}}\right)\left(\frac{N_e}{2\mathrm{mol}/\mathrm{cm}^3}\right),$$
(40)
implying that $`A/\delta m^2O(1)`$ for sub-GeV SK events (and, in principle, for accelerator K2K neutrinos as well). In the constant-density approximation $`A(x)=\mathrm{const}`$ (i.e., by neglecting mantle-core interference effects to keep the following discussion simple), the oscillation probabilities $`P_{ee}`$ and $`P_{e\mu }`$ can be evaluated through Eq. (21), in terms of the effective mass-mixing parameters in matter $`(\stackrel{~}{\theta }_{ij},\stackrel{~}{m}_i^2\stackrel{~}{m}_j^2)`$.
Suitable approximations for such parameters have been reported in many papers. If we use, e.g., those reported in the classic review , after some algebra we get from Eqs. (38) and (21) that the electron excess at sub- or multi-GeV energies can be written as a sum of three terms,
$$\frac{N_e}{N_e^0}1\mathrm{\Delta }_1+\mathrm{\Delta }_2+\mathrm{\Delta }_3,$$
(41)
where
$`\mathrm{\Delta }_1`$ $``$ $`\mathrm{sin}^22\stackrel{~}{\theta }_{13}\mathrm{sin}^2\left(\mathrm{\Delta }m^2{\displaystyle \frac{\mathrm{sin}2\theta _{13}}{\mathrm{sin}2\stackrel{~}{\theta }_{13}}}{\displaystyle \frac{L}{4E}}\right)(rs_{23}^21)`$ (42)
$`\mathrm{\Delta }_2`$ $``$ $`\mathrm{sin}^22\stackrel{~}{\theta }_{12}\mathrm{sin}^2\left(\delta m^2{\displaystyle \frac{\mathrm{sin}2\theta _{12}}{\mathrm{sin}2\stackrel{~}{\theta }_{12}}}{\displaystyle \frac{L}{4E}}\right)(rc_{23}^21)`$ (43)
$`\mathrm{\Delta }_3`$ $``$ $`\mathrm{sin}^22\stackrel{~}{\theta }_{12}\mathrm{sin}^2\left(\delta m^2{\displaystyle \frac{\mathrm{sin}2\theta _{12}}{\mathrm{sin}2\stackrel{~}{\theta }_{12}}}{\displaystyle \frac{L}{4E}}\right)rs_{13}c_{13}^2\mathrm{sin}2\theta _{23}(\mathrm{tan}2\stackrel{~}{\theta }_{12})^1`$ (44)
with
$`{\displaystyle \frac{\mathrm{sin}2\theta _{13}}{\mathrm{sin}2\stackrel{~}{\theta }_{13}}}`$ $``$ $`\sqrt{\left({\displaystyle \frac{A}{\mathrm{\Delta }m^2+\frac{\delta m^2}{2}\mathrm{cos}2\theta _{12}}}\mathrm{cos}2\theta _{13}\right)^2+\mathrm{sin}^22\theta _{13}},`$ (45)
$`{\displaystyle \frac{\mathrm{sin}2\theta _{12}}{\mathrm{sin}2\stackrel{~}{\theta }_{12}}}`$ $``$ $`\sqrt{\left({\displaystyle \frac{Ac_{13}^2}{\delta m^2}}\mathrm{cos}2\theta _{12}\right)^2+\mathrm{sin}^22\theta _{12}}.`$ (46)
The above expressions for $`\mathrm{\Delta }_i`$, which hold for neutrinos with normal hierarchy and $`\delta =0`$, coincide with those reported in (up to higher-order terms or CP-violating terms, not included here). The corresponding expressions for antineutrinos, for inverted hierarchy, and for $`\delta =\pi `$, can be obtained, respectively, through the replacements:
$`+AA`$ $`(\mathrm{swaps}(\mathrm{anti})\mathrm{neutrinos}),`$ (47)
$`+\mathrm{\Delta }m^2\mathrm{\Delta }m^2`$ $`(\mathrm{swaps}\mathrm{hierarchy}),`$ (48)
$`+s_{13}s_{13}`$ $`(\mathrm{swaps}\mathrm{CP}\mathrm{parity}),`$ (49)
where by “CP parity” we mean $`\mathrm{cos}\delta =\pm 1`$. Under such transformations, the terms $`\mathrm{\Delta }_i`$ behave as follows: (1) all $`\mathrm{\Delta }_i`$’s are affected by $`AA`$ through $`\stackrel{~}{\theta }_{12}`$ or $`\stackrel{~}{\theta }_{13}`$; (2) only $`\mathrm{\Delta }_1`$ is sensitive to $`\mathrm{\Delta }m^2\mathrm{\Delta }m^2`$; (3) only $`\mathrm{\Delta }_3`$ is sensitive to $`+s_{13}s_{13}`$.
Concerning the dependence on the oscillation parameters, one has that: (1) all $`\mathrm{\Delta }_i`$’s depend on $`\theta _{23}`$; (2) $`\mathrm{\Delta }_1`$ arises for $`\theta _{13}>0`$, and is independent of $`\delta m^2`$; (3) $`\mathrm{\Delta }_2`$ arises for $`\delta m^2>0`$, and is independent of $`\theta _{13}`$; only $`\mathrm{\Delta }_3`$ (“interference term” ) depends on both $`\theta _{13}`$ and $`\delta m^2`$.
Concerning the dependence on energy, in the sub-GeV range one has that: (1) $`\stackrel{~}{\theta }_{13}\theta _{13}`$, so that for large $`L`$ the first term is simply $`\mathrm{\Delta }_12s_{13}^2c_{13}^2(rs_{23}^21)`$; (2) since $`r2`$, the term $`\mathrm{\Delta }_1`$ flips sign as $`s_{23}^2`$ crosses the maximal mixing value 1/2 , and similarly for $`\mathrm{\Delta }_2`$ (with opposite sign) ; (3) for neutrinos, which give the largest contribution to atmospheric events, it turns out that $`\mathrm{tan}2\stackrel{~}{\theta }_{12}<0`$, and thus typically $`\mathrm{\Delta }_3<0`$ for $`\delta =0`$ ($`\mathrm{\Delta }_3>0`$ for $`\delta =\pi `$). In the multi-GeV range one has that $`\stackrel{~}{\theta }_{12}\pi /2`$, so that only $`\mathrm{\Delta }_1`$ dominates, with typically positive values (being $`r3.5`$ and $`s_{23}^2`$ not too different from $`1/2`$).
Figure 21 shows exact numerical examples (extracted from our SK data analysis) where, from top to bottom, the dominant term is $`\mathrm{\Delta }_1`$, $`\mathrm{\Delta }_2`$, and $`\mathrm{\Delta }_3`$. Here, as in Fig. 18, the dashed histograms represent the unshifted theoretical predictions, while the solid histograms represent the systematically shifted predictions, i.e., $`R_n`$ and $`\overline{R}_n`$ respectively \[in terms of Eq. (33)\]. Let us focus on subleading effects in the dashed histograms of Fig. 21, which refer to the sub-GeV (left) and multi-GeV (right) electron samples. In the figure, we have taken $`\mathrm{\Delta }m^2=+2.4\times \times 10^3`$ eV<sup>2</sup> (normal hierarchy); other relevant parameters are indicated at the right of each panel. In the upper panel, we have set $`\delta m^2=0`$, so as to switch off $`\mathrm{\Delta }_2`$ and $`\mathrm{\Delta }_3`$. We have also taken $`s_{23}^2=0.4<0.5`$, so that $`\mathrm{\Delta }_1<0`$ in the sub-GeV sample; it is instead $`\mathrm{\Delta }_1>0`$ in the multi-GeV sample. In the middle panel, we have set $`(\delta m^2,\mathrm{sin}^2\theta _{12})`$ at their best-fit LMA values, but have taken $`\mathrm{sin}^2\theta _{13}=0`$, so that only $`\mathrm{\Delta }_2`$ survives. In particular, while there is no observable effect of $`\mathrm{\Delta }_2`$ in the multi-GeV sample (where the energy is relatively high and $`\mathrm{sin}2\stackrel{~}{\theta }_{12}0`$), the effect is positive for sub-GeV neutrinos, where $`s_{23}^2=0.4<1/2`$. Notice that the upper and middle panel results are insensitive to $`\delta =0`$ or $`\pi `$, since $`\mathrm{\Delta }_30`$ in both cases. Finally, in the bottom plot we have taken $`s_{23}^2=1/2`$, so as to suppress $`\mathrm{\Delta }_1`$ and $`\mathrm{\Delta }_2`$ is the sub-GeV sample, where $`\mathrm{\Delta }_3>0`$ for our choice $`\delta =\pi `$. In the multi-GeV sample, however, $`\mathrm{\Delta }_1`$ is still operative.
The subleading dependence of atmospheric electron neutrino events on the hierarchy, $`\delta m^2`$, $`\theta _{13}`$, and CP-parity is intriguing and is thus attracting increasing interest . However, Fig. 21 clearly shows that such dependence is currently well hidden, not only by statistical uncertainties (vertical error bars) but, more dangerously, by allowed systematic shifts of the theoretical predictions (solid histograms). For instance, in the upper panel, systematics can “undo” the negative effect of $`\mathrm{\Delta }_1`$ in the SG$`e`$ sample and make it appear positive. In all cases, they tend to magnify the zenith spectrum distortion; this is particularly evident in the right middle panel, where the unshifted theoretical prediction is flat.
We think it useful to quantify at which level one has to reduce systematic uncertainties, in order to appreciate subleading effects in future, larger SK-like atmospheric neutrino experiments such as those proposed in (see also ). Since normalization systematics are large (as discussed in the previous section) and a significant reduction may be difficult, we prefer to focus on a normalization-independent quantity, namely, the fractional deviation of the up-down asymmetry of electron events from their no-oscillation value,
$$A_e=\frac{U/D}{U_0/D_0}1,$$
(50)
where “up” ($`U`$) and “down” ($`D`$) refer to the zenith angle ranges $`\mathrm{cos}\theta _z[1,0.4]`$ and $`[0.4,1]`$, respectively. We perform a full numerical calculation of this quantity for both SG$`e`$ and MG$`e`$ events, assuming the SK experimental setting for definiteness. Notice that the up-down asymmetry involves the first and last three bins of the SG$`e`$ and MG$`e`$ samples in Fig. 18.
Fig. 22 shows isolines of $`100\times A_e`$ for the SG$`e`$ sample, plotted in the $`(\mathrm{sin}^2\theta _{23},\mathrm{sin}^2\theta _{13})`$ plane at fixed $`\mathrm{\Delta }m^2=2.4\times 10^3`$ eV<sup>2</sup>, for both normal hierarchy ($`+\mathrm{\Delta }m^2`$, left panels) and inverse hierarchy ($`\mathrm{\Delta }m^2`$, right panels). In both hierarchies, we consider first the “academic” case $`\delta m^2=0`$ (top panels), then we switch on the LMA parameters $`(\delta m^2,\mathrm{sin}^2\theta _{12})`$ at their best-fit values, for the the two CP-conserving cases $`\delta =0`$ (middle panels) and $`\delta =\pi `$ (bottom panels). The isolines in the upper panels reflect the behavior of the $`\mathrm{\Delta }_1`$ term, which is positive (negative) for $`s_{23}^2>1/r`$ ($`<1/r`$), and vanishes for $`\theta _{13}0`$, with a weak dependence on the hierarchy through $`\stackrel{~}{\theta }_{13}`$. In the middle panel, subleading LMA effects are operative through $`\mathrm{\Delta }_2`$ and $`\mathrm{\Delta }_3`$. The variation of $`A_e`$ in sign and magnitude is now more modest as $`s_{23}^2`$ increases, since the variation of the $`\mathrm{\Delta }_1`$ term is now partially compensated by the opposite variation of the $`\mathrm{\Delta }_2`$ term. In particular, the term $`\mathrm{\Delta }_2`$ is responsible for nonzero values of $`A_e`$ at $`\theta _{13}=0`$, which break the $`\theta _{23}`$ octant symmetry , as also phenomenologically observed in Fig. 15. The difference between the middle and bottom panels is due to the interference term $`\mathrm{\Delta }_3`$, which is typically negative (positive) for $`\delta =0`$ ($`\delta =\pi `$), and thus either adds or subtracts to $`\mathrm{\Delta }_1`$ and $`\mathrm{\Delta }_2`$ in the two cases. However, for $`\theta _{13}=0`$ the values of $`A_e`$ basically coincide in all middle and bottom panels, since the only surviving term ($`\mathrm{\Delta }_2`$) carries no dependence on the hierarchy or the CP parity. From the results in Fig. 22 we learn that: (1) subleading $`\delta m^2`$-induced effects are of the same size of $`\theta _{13}`$-induced effects in the SG sample, so none can be neglected in a precise $`3\nu `$ oscillation analysis; (2) for nonzero values of both $`\delta m^2`$ and $`\theta _{13}`$, the sub-GeV electron asymmetry is typically more pronounced (and positive) for $`\delta =\pi `$, as compared with the case $`\delta =0`$; one can thus expect the latter case to be slightly disfavored in a global fit (since the SG$`e`$ data show a slight asymmetry, see Fig. 18); (3) in any case, the electron asymmetry is typically at the percent or sub-percent level for $`\mathrm{sin}^2\theta _{13}<\mathrm{few}\%`$; therefore, statistical and systematic uncertainties need to be reduced at this extraordinary small level in order to really “observe” the effects in future atmospheric neutrino experiments .
Fig. 23 shows our numerical calculation of the up-down electron asymmetry for the SK multi-GeV sample. The six panels refer to the same cases as in Fig. 22. In the MG$`e`$ sample, the terms $`\mathrm{\Delta }_2`$ and $`\mathrm{\Delta }_3`$ are small, and there is little dependence on $`\delta m^2`$ and on the CP parity (top, middle and bottom panels being quite similar). The dominant $`\mathrm{\Delta }_1`$ term makes the asymmetry generally positive, and with significant dependence on the hierarchy (left vs right panels) through $`\stackrel{~}{\theta }_{13}`$. The MG$`e`$ asymmetry can be of $`O(10\%)`$ and thus relatively large; with some luck, such asymmetry might be seen in future large Cherenkov detectors if $`\theta _{13}`$ is not too small (see, e.g., and refs. therein). In any case, one can expect some dependence of the current SK fit on the hierarchy through multi-GeV events (see also for older data); it is difficult, however, to “predict” which of the two hierarchies (normal or inverted) is currently preferred, since large statistical fluctuations make the zenith-angle pattern of SK MG$`e`$ data somewhat erratical (see Fig. 18).
We conclude this Section with a brief note on subleading effects in K2K (which have been numerically included throughout this work). For $`\theta _{13}=0`$, it is easy to derive that, in vacuum,
$$P_{\mu \mu }^{\mathrm{K2K}}1\mathrm{sin}^22\theta _{23}\mathrm{sin}^2\left(\frac{\mathrm{\Delta }m^2\frac{\delta m^2}{2}\mathrm{cos}\theta _{12}}{4E}L\right),$$
(51)
at first order in $`\delta m^2/\mathrm{\Delta }m^2`$. It is also not difficult to check that this formula is not significantly affected by matter effects, as far as $`A\delta m^2`$, which is true for most of the K2K event spectrum. The $`\theta _{23}`$ octant-symmetry of the above equation is responsible for the appearance of two degenerate best-fits in the K2K analysis of Fig. 16. The above equation is not invariant under a change of hierarchy ($`+\mathrm{\Delta }m^2\mathrm{\Delta }m^2`$), which leads to a (really tiny) relative change of the oscillation phase equal to $`\pm (\delta m^2/2)\mathrm{cos}\theta _{12}/\mathrm{\Delta }m^20.6\%`$; this change does not produce graphically observable effects in Fig. 16. Although very small, these and other subleading K2K effects (e.g., those arising for both $`\theta _{13}`$ and $`\delta m^2`$ nonzero) have been kept in the analysis, in order to be consistent with the atmospheric neutrino data analysis (where such effects have also been included, as already discussed), and in order to show explicitly their impact on the global SK+K2K+CHOOZ analysis presented in the next section.
### 4.3 SK, K2K and CHOOZ constraints ($`\theta _{13}`$ free)
In this section we present the results of our analysis of SK+K2K+CHOOZ data for unconstrained values of $`(\mathrm{\Delta }m^2,\mathrm{sin}^2\theta _{23},\mathrm{sin}^2\theta _{13})`$, and for fixed values $`(\delta m^2,\mathrm{sin}^2\theta _{12})=(8\times 10^5\mathrm{eV}^2,0.314)`$ in all the three data samples. There are four discrete subcases in our analysis, corresponding to a change in hierarchy or CP parity:
$$[\mathrm{sign}(\mathrm{\Delta }m^2)=\pm 1][\mathrm{cos}\delta =\pm 1].$$
(52)
In particular we remind that, in this work, we do not consider generic values of $`\delta `$, but only the two inequivalent CP-conserving cases ($`\delta =0`$ and $`\delta =\pi `$). They are related by the transformation $`+s_{13}s_{13}`$ which, of course, does not mean that $`s_{13}`$ can be negative, but just that $`\mathrm{cos}\delta s_{13}`$ can change sign. Since the two cases smoothly merge for $`s_{13}0`$, we think it useful to show the results of our analysis also in terms of the variable $`\mathrm{cos}\delta \mathrm{sin}\theta _{13}`$, i.e., of $`\pm \mathrm{sin}\theta _{13}`$, for both normal and inverted hierarchy.
Figure 24 shows the $`\chi ^2`$ function from the SK+K2K+CHOOZ fit, in terms of $`\mathrm{cos}\delta \mathrm{sin}\theta _{13}`$, for marginalized $`(\mathrm{\Delta }m^2,\mathrm{sin}^2\theta _{23})`$ parameters.<sup>10</sup><sup>10</sup>10This representation is inspired by Ref. where, however, $`\mathrm{cos}\delta \mathrm{sin}^2\theta _{13}`$ was used. The use of $`\mathrm{cos}\delta \mathrm{sin}\theta _{13}`$ makes the $`\chi ^2`$ curves smooth (no cusp) across $`\theta _{13}=0`$. The solid and dashed curves correspond to normal and inverted hierarchy, respectively, while their left and right parts correspond to $`\delta =\pi `$ and $`\delta =0`$, respectively. Notice that the solid and dashed curves do not exactly coincide at $`\theta _{13}=0`$, since for $`\delta m^2>0`$ there is a very weak dependence on the hierarchy even at $`\theta _{13}=0`$ in reactor , accelerator , and atmospheric neutrino oscillations. The difference is, however, really tiny within the current global analysis ($`\mathrm{\Delta }\chi ^20.2`$ at $`s_{13}=0`$). The absolute $`\chi ^2`$ minimum is reached in the left half of the figure ($`\delta =\pi `$) for $`\mathrm{sin}\theta _{13}0.1`$; the minimum in the right half ($`\delta =0`$), which is reached for $`\theta _{13}=0`$, is only slight higher $`(\mathrm{\Delta }\chi ^2<1)`$. The slight difference between these two CP-conserving cases is mainly due to sub-GeV SK events, as discussed in the comments to Fig. 22. Finally, normal and inverted hierarchies give basically the same results for small values of $`s_{13}`$ (say, $`<0.1`$), while the latter hierarchy is slightly preferred for higher values of $`s_{13}`$. The fit becomes rapidly worse for $`s_{13}0.2`$ or higher. Figure 24 nicely summarizes our current (unfortunately weak) sensitivity to the neutrino mass hierarchy and to the extremal (CP-conserving) cases $`\delta =0`$ and $`\delta =\pi `$.
Figure 25 shows the parameter space orthogonal to the one in Fig. 24, i.e., the bounds on $`(\mathrm{\Delta }m^2,\mathrm{sin}^2\theta _{23})`$ for marginalized $`\theta _{13}`$, in each of the four cases in Eq. (52). The differences between such cases are very small. Figure 24 and 25 confirm that our current sensitivity to the subleading effects—which distinguish the four subcases in Eq. (52)—is not statistically appreciable yet. Therefore, it makes sense to make a further marginalization over the four subcases, by minimizing the SK+K2K+CHOOZ function with respect to hierarchy and CP parity. The results are shown in Fig. 26, in terms of the projections of the $`(\mathrm{\Delta }m^2,\mathrm{sin}^2\theta _{23},\mathrm{sin}^2\theta _{13})`$ region allowed at 1, 2, and $`3\sigma `$ onto each of the coordinate planes. The best fit is reached for nonzero $`\theta _{13}`$ (as expected from Fig. 24), but $`\theta _{13}=0`$ is allowed within less than $`1\sigma `$. The preferred value of $`\mathrm{sin}^2\theta _{23}`$ remains slightly below maximal mixing. The best-fit value of $`\mathrm{\Delta }m^2`$ is $`2.4\times 10^3`$ eV<sup>2</sup>. Notice that the correlations among the three parameters in Fig. 26 are very weak.
## 5 Global analysis of oscillation data
The results of the global analysis of solar and KamLAND data (Sec. 3.4) and of SK+K2K+CHOOZ data (Sec. 4.3) can now be merged to provide our best estimates of the five parameters $`(\delta m^2,\mathrm{\Delta }m^2,\theta _{12},\theta _{13},\theta _{23})`$, marginalized over the four cases in Eq. (52). The bounds will be directly shown in terms of the “number of sigmas”, corresponding to the function $`(\mathrm{\Delta }\chi ^2)^{1/2}`$ for each parameter.
Figure 27 shows our global bounds on $`\mathrm{sin}^2\theta _{13}`$, as coming from all data (solid line) and from the following partial data sets: KamLAND (dotted), solar (dot-dashed), solar+KamLAND (short-dashed) and SK+K2K+CHOOZ (long-dashed). Only the latter set, as observed before, gives a weak indication for nonzero $`\theta _{13}`$. Interestingly, solar+KamLAND data are now sufficiently accurate to provide bounds which are not much weaker than the dominant SK+K2K+CHOOZ ones, also because the latter slightly prefer $`\theta _{13}>0`$ as best fit, while the former do not.
Figure 28 shows our global bounds on the four mass-mixing parameters which present both upper and lower limits with high statistical significance. Notice that the accuracy of the parameter estimate is already good enough to lead to almost “linear” errors, especially for $`\delta m^2`$ and $`\mathrm{sin}^2\theta _{12}`$. For $`\mathrm{\Delta }m^2`$ and $`\mathrm{sin}^2\theta _{23}`$, such “linearity” is somewhat worse in the region close to the best fit (say, within $`\pm 1\sigma `$), and thus $`2\sigma `$ (or $`3\sigma `$) errors should be taken as reference.
We summarize our results through the following $`\pm 2\sigma `$ ranges (95% C.L.) for each parameter:
$`\mathrm{sin}^2\theta _{13}`$ $`=`$ $`0.9_{0.9}^{+2.3}\times 10^2,`$ (53)
$`\delta m^2`$ $`=`$ $`7.92(1\pm 0.09)\times 10^5\mathrm{eV}^2,`$ (54)
$`\mathrm{sin}^2\theta _{12}`$ $`=`$ $`0.314(1_{0.15}^{+0.18}),`$ (55)
$`\mathrm{\Delta }m^2`$ $`=`$ $`2.4(1_{0.26}^{+0.21})\times 10^3\mathrm{eV}^2,`$ (56)
$`\mathrm{sin}^2\theta _{23}`$ $`=`$ $`0.44(1_{0.22}^{+0.41}).`$ (57)
Notice that the lower uncertainty on $`\mathrm{sin}^2\theta _{13}`$ is purely formal, corresponding to the positivity constraint $`\mathrm{sin}^2\theta _{13}0`$. Correlations among parameters are not quoted, being currently small (as already observed).
The above bounds have been obtained from a global analysis of oscillation data (for $`U=U^{}`$). They have, however, an impact also on non-oscillation observables. In particular, the smallness of the squared mass splittings induces significant correlations on the three parameters $`(m_\beta ,m_{\beta \beta },\mathrm{\Sigma })`$ which are sensitive to absolute neutrino masses (see and references therein). Figure 29 shows the updated $`2\sigma `$ allowed bands in each of the three corresponding coordinate planes, for both normal and inverted hierarchy. There is an evident positive correlation, especially between $`m_\beta `$ and $`\mathrm{\Sigma }`$; the correlation is less pronounced when it involves $`m_{\beta \beta }`$, due to our ignorance of the Majorana phases $`\varphi _2`$ and $`\varphi _3`$ (that we take as free parameters). The two hierarchies split up only at very low values of the observables, where mass splittings start to be of the order of the absolute masses (non-degenerate cases). Non-oscillation data on $`m_\beta `$, $`m_{\beta \beta }`$ and $`\mathrm{\Sigma }`$ can reduce the allowed parameter space in Fig. 29, hopefully leading to a single solution and thus to the determination of the absolute neutrino masses. Such data are discussed in the next Section.
## 6 Global analysis of oscillation and non-oscillation data
In this Section we discuss first non-oscillation data on the three observables $`(m_\beta ,m_{\beta \beta },\mathrm{\Sigma })`$, and then show how these data further constrain and reduce the allowed regions in Fig. 29. As we shall see, when all the data are taken at face value, no combination is possible: a strong tension arises, indicating that either some experimental information or their theoretical interpretation is wrong or biased. In particular, it appears difficult to reconcile the claimed $`0\nu 2\beta `$ signal and the most recent upper bounds on $`\mathrm{\Sigma }`$ from precision cosmology . However, relaxing one of either pieces of data reduces the tension and allows a global combination, which can be valuable for prospective studies .
Needless to say, the relations between the variables $`(m_\beta ,m_{\beta \beta },\mathrm{\Sigma })`$ have been subject to intensive studies, which form a large specialized literature on absolute neutrino mass observables. We refer the reader to the review papers for extensive bibliographies, and to the articles for recent up-to-date discussions.
### 6.1 Bounds on $`m_\beta `$
Experimental constraints on the effective electron neutrino mass $`m_\beta `$ have been recently presented for the Mainz and Troitsk tritium $`\beta `$-decay experiments. The experimental values are consistent with zero within errors. Their combined upper bound at $`2\sigma `$ has been estimated in as:
$$m_\beta <1.8\mathrm{eV}(\mathrm{Mainz}+\mathrm{Troitsk}),$$
(58)
which is less conservative than the $`3`$ eV upper limit recommended in . It should be mentioned that the Troitsk results are to be taken with some caution, being affected by an unexplained anomaly (namely, a fluctuating excess of counts near the endpoint) . However, as we will see, upper limits on $`m_\beta `$ in the 2–3 eV range are, in any case, too weak to contribute significantly to the current global fit in the $`(m_\beta ,m_{\beta \beta },\mathrm{\Sigma })`$ parameter space, so that “conservativeness” is not (yet) an issue in this context.
### 6.2 Bounds on $`m_{\beta \beta }`$
Neutrinoless double beta decay processes of the kind $`(Z,A)(Z+2,A)+2e^{}`$ have been searched in many experiments with different isotopes, yielding negative results (see for reviews). Recently, members of the Heidelberg-Moscow experiment have claimed the detection of a $`0\nu 2\beta `$ signal from the <sup>76</sup>Ge isotope . If this signal is entirely due to light Majorana neutrino masses, the $`0\nu 2\beta `$ half-life $`T`$ is related to the $`m_{\beta \beta }`$ parameter by the relation
$$m_{\beta \beta }^2=\frac{m_e^2}{C_{mm}T},$$
(59)
where $`m_e`$ is the electron mass and $`C_{mm}`$ is the nuclear matrix element for the considered isotope .
Unfortunately, theoretical uncertainties on $`C_{mm}`$ are rather large (see e.g. ), and their—somewhat arbitrary—estimate is matter of debate (see and refs. therein). In we adopted a naive but very conservative estimate, by defining the range spanned by “extremal” published values of $`C_{mm}`$ as an “effective $`3\sigma `$ range,” thus obtaining $`\mathrm{log}_{10}(C_{mm}/\mathrm{y}^1)=13.36\pm 0.97`$ (at $`\pm 3\sigma `$). Here we prefer to adopt the results of a recent detailed discussion of the nuclear model uncertainties for $`C_{mm}`$, performed within the (Renormalized) Quasiparticle Random Phase Approximation, and calibrated to known $`2\nu \beta \beta `$ decay rates . For our purposes, we cast the results of such promising approach in the form $`\mathrm{log}_{10}(C_{mm}/\mathrm{y}^1)=13.36\pm 0.15`$ (at $`\pm 3\sigma `$), where systematic coupling constant uncertainties ($`g_A=1`$–1.25, see ) have been included. This “new” range for $`C_{mm}`$ has (accidentally) the same central value as before, but with significantly reduced errors. Under the assumption of a positive $`0\nu 2\beta `$ signal , we then derive that
$$\mathrm{log}_{10}(m_{\beta \beta }/\mathrm{eV})=0.23\pm 0.14(2\sigma ),$$
(60)
i.e., $`0.43<m_{\beta \beta }<0.81`$ (at $`2\sigma `$, in eV). See also for our previous (more conservative) estimated range.
The claim in has been subject to strong criticism, especially after the first publication (see and refs. therein). Therefore, we will also consider the possibility that $`T=\mathrm{}`$ is allowed (i.e., that there is no $`0\nu 2\beta `$ signal), in which case the experimental lower bound on $`m_{\beta \beta }`$ disappears, and only the upper bound remains. In conclusion, we adopt the following two possible $`0\nu 2\beta `$ inputs for our global analysis:
$`\mathrm{log}_{10}(m_{\beta \beta }/\mathrm{eV})`$ $`=`$ $`0.23\pm 0.14(0\nu 2\beta \mathrm{signal}\mathrm{assumed}),`$ (61)
$`\mathrm{log}_{10}(m_{\beta \beta }/\mathrm{eV})`$ $`=`$ $`0.23_{\mathrm{}}^{+0.14}(0\nu 2\beta \mathrm{signal}\mathrm{not}\mathrm{assumed}),`$ (62)
where errors are at $`2\sigma `$ level. Concerning the unknown Majorana phases $`\varphi _2`$ and $`\varphi _3`$ in Eq. (15), we simply assume that they are independent and uniformly distributed in the range $`[0,\pi ]`$, which covers all physically different cases in $`m_{\beta \beta }`$.
### 6.3 Bounds on $`\mathrm{\Sigma }`$
The neutrino contribution to the overall energy density of the universe can play a relevant role in large scale structure formation, leaving key signatures in several cosmological data sets. More specifically, neutrinos suppress the growth of fluctuations on scales below the horizon when they become non relativistic. Massive neutrinos of a fraction of eV would therefore produce a significant suppression in the clustering on small cosmological scales. Data on large scale structures, combined with Cosmic Microwave Background (CMB) and other precision astrophysical data, can thus constrain the sum of neutrino masses $`\mathrm{\Sigma }`$ (see for recent reviews).<sup>11</sup><sup>11</sup>11Future cosmological data might become slightly sensitive to finer details (e.g., the neutrino mass hierarchy) through subleading effects not included in this work.
In this work we use the bounds on $`\mathrm{\Sigma }`$ previously obtained in collaboration with other authors in , to which we refer the reader for technical details. We briefly remind that the experimental input used in included CMB data from the Wilkinson Microwave Anisotropy Probe (WMAP) , large scale structure data from the 2 degrees Fiels (2dF) Galaxy Redshift Survey and, optionally, constraints on mall scales from the recent Lyman $`\alpha `$ (Ly$`\alpha `$) forest data of the Sloan Digital Sky Survey (SDSS) . The latter data have a strong impact on the current upper bounds on $`\mathrm{\Sigma }`$ , but are also affected by large systematics, which deserve further study . As in , we conservatively quote (and use) upper bounds on $`\mathrm{\Sigma }`$ both with and without such Ly$`\alpha `$ forest data; in particular, the $`2\sigma `$ upper bounds from read:
$`\mathrm{\Sigma }`$ $`<`$ $`0.5\mathrm{eV}(\mathrm{with}\mathrm{Ly}\alpha \mathrm{data}),`$ (63)
$`\mathrm{\Sigma }`$ $`<`$ $`1.4\mathrm{eV}(\mathrm{without}\mathrm{Ly}\alpha \mathrm{data}).`$ (64)
### 6.4 Impact of non-oscillation observables
The experimental limits on the non-oscillation observables $`(m_\beta ,m_{\beta \beta },\mathrm{\Sigma })`$, previously reported in terms of $`2\sigma `$ ranges, are appropriately combined with oscillation data through $`\mathrm{\Delta }\chi ^2`$ functions . Although such combination can provide allowed regions at any confidence level, in the following we shall continue to show only $`2\sigma `$ bounds for simplicity.
Figure 30 shows the impact of all the available non-oscillation data, taken at face value, in the parameter space $`(m_{\beta \beta },\mathrm{\Sigma })`$. Bounds on the third observable $`m_\beta `$ are projected away, being too weak to alter the discussion of the results in this figure. The horizontal band is allowed by the positive $`0\nu 2\beta `$ experimental claim equipped with the nuclear uncertainties of through Eq. (61). The slanted bands (for normal and inverted hierarchy) are allowed by all other neutrino data, i.e., by the combination of neutrino oscillation constraints (as in Fig. 29) and of astrophysical CMB+2dF+Ly$`\alpha `$ constraints through Eq. (63). The tight cosmological upper bound on $`\mathrm{\Sigma }`$ prevents the overlap between the slanted and horizontal bands at $`2\sigma `$, indicating that no global combination of oscillation and non-oscillation data is possible in the sub-eV range. The “discrepancy” is even stronger than it was found in , due to the adoption of smaller $`0\nu 2\beta `$ nuclear uncertainties . It is premature, however, to derive any definite conclusion as to which piece of the data or of the $`3\nu `$ scenario is “wrong” in this conflicting picture. Further experimental and theoretical research is needed to clarify absolute neutrino observables in the sub-eV range.
It is tempting, however, to see if the removal of some pieces of data can relax the tension in Fig. 30. The effect of removing only the lower bound on $`m_{\beta \beta }`$ through Eq. (62) is shown in Fig. 31. Of the three remaining upper bounds on $`m_\beta `$, on $`m_{\beta \beta }`$, and on $`\mathrm{\Sigma }`$, the latter is definitely dominant, and implies that future beta and double-beta decay searches should push their sensitivity below 0.2 eV, irrespective of the hierarchy. Conversely, the effect of removing only the Ly$`\alpha `$ forest data through Eq. (64) is shown in Fig. 32. In this case, the combination of the claimed $`0\nu 2\beta `$ signal with oscillation data dominates the global fit, and “predicts” the observation of $`\mathrm{\Sigma }1.5`$ eV and $`m_\beta 0.5`$ eV within formally small uncertainties (about $`\pm 20\%`$ at $`2\sigma `$). These predictions would really be “around the corner” from the observational viewpoint, both for $`\mathrm{\Sigma }`$ and for $`m_\beta `$ . Future searches are expected to clarify the—currently controversial—situation about absolute mass observables in the sub-eV range, as depicted in Figs. 30–32.
## 7 Conclusions
We have performed a comprehensive phenomenological analysis of a vast amount of data from neutrino flavor oscillation and non-oscillation searches, including solar, atmospheric, reactor, accelerator, beta-decay, double-beta decay, and precision astrophysical observations. In the analysis, performed within the standard scenario with three massive and mixed neutrinos (for both mass hierarchies and for the two inequivalent CP-conserving cases), we have paid particular attention to implement subleading oscillation effects in numerical calculations, and to carefully include all known sources of uncertainties in the statistical comparison of theoretical predictions and experimental data. We have discussed the impact of solar and reactor data on the parameters $`(\delta m^2,\mathrm{sin}^2\theta _{12},\mathrm{sin}^2\theta _{13})`$, as well as the impact of atmospheric and reactor data on $`(\mathrm{\Delta }m^2,\mathrm{sin}^2\theta _{23},\mathrm{sin}^2\theta _{13})`$. The bounds from the global analysis of oscillation data have been summarized, and several subleading effects have been discussed. Finally, we have analyzed the interplay between the oscillation parameters $`(\delta m^2,\mathrm{\Delta }m^2,\mathrm{sin}^2\theta _{12},\mathrm{sin}^2\theta _{23},\mathrm{sin}^2\theta _{13})`$ and the non-oscillation observables sensitive to absolute neutrino masses ($`m_\beta ,m_{\beta \beta },\mathrm{\Sigma }`$), both with and without controversial data, which may or may not allow a reasonable global combination of all data. The detailed results discussed in this review article represent a state-of-the-art, accurate and up-to-date (as of August 2005) overview of the neutrino mass and mixing parameters within the standard three-generation framework.
## Acknowledgments
This work is supported by the Italian Ministero dell’Istruzione, Università e Ricerca (MIUR) and Istituto Nazionale di Fisica Nucleare (INFN) through the “Astroparticle Physics” research project. The authors have greatly benefited of earlier collaborations (on various topics or papers quoted in this review) with several researchers, including J.N. Bahcall, B. Faïd, G. Fiorentini, P. Krastev, A. Melchiorri, A. Mirizzi, D. Montanino, S.T. Petcov, A.M. Rotunno, G. Scioscia, S. Sarkar, P. Serra, J. Silk, F. Villante.
## Appendix
In this Appendix we report—for more expert readers—additional technical information about the $`\chi ^2`$ analysis of each data sample, which has been used to derive partial and global parameter bounds in this work. In particular, global constraints have been obtained by adding up all $`\chi ^2`$ contributions and by scanning the (CP-conserving) $`3\nu `$ mass-mixing parameter space
$$𝐩=\{\pm \mathrm{\Delta }m^2,\delta m^2,s_{23}^2,s_{13}^2,s_{12}^2,\mathrm{cos}\delta =\pm 1\},$$
(65)
with allowed regions being derived through $`\mathrm{\Delta }\chi ^2`$ cuts with respect to the best-fit point. As discussed in the preceding Sections of this review, some data samples are actually not sensitive to all of the above parameters; the relevant variables in $`𝐩`$ will be explicitly emphasized in the following subsections.
### 7.1 CHOOZ
The CHOOZ experiment has measured the positron energy spectra induced by $`\overline{\nu }_e`$’s produced by two nuclear reactors located at $`L_1=1114.6`$ km and $`L_2=997.9`$ km from the detector. Each of the two spectra is divided in 7 energy bins, for a total of 14 event rate bins. In our analysis, such data are included through the following $`\chi ^2`$ function:
$$\chi _{\mathrm{CHOOZ}}^2(𝐩)=\underset{\alpha }{\mathrm{min}}\left\{\underset{i,j=1}{\overset{14}{}}[R_i^{\mathrm{expt}}\alpha R_i^{\mathrm{theo}}(𝐩)][\sigma _{ij}^2]^1[R_j^{\mathrm{expt}}\alpha R_j^{\mathrm{theo}}(𝐩)]+\left(\frac{\alpha 1}{\sigma _\alpha }\right)^2\right\},$$
(66)
where $`R_i^{\mathrm{expt}}`$ and $`R_i^{\mathrm{theo}}`$ are the observed and predicted rates in each bin, respectively, and $`\alpha `$ is an overall normalization factor with uncertainty $`\sigma _\alpha =2.7\times 10^2`$. The squared error matrix is defined as :
$$\sigma _{ij}^2=\delta _{ij}(s_i^2+u_i^2)+(\delta _{i,j7}+\delta _{i,j+7})c_i^2,$$
(67)
where $`s_i`$ and $`u_i`$ represent statistical errors and uncorrelated systematic errors, respectively, while the $`c_i`$’s represent fully correlated systematic errors between equal-energy bins in the two reactor spectra. The theoretical rate in each bin is estimated as
$$R_i^{\mathrm{theo}}=R_i^0P_{ee}(𝐩)_i,$$
(68)
where $`R_i^0`$ is the no-oscillation rate, and $`P_{ee}(𝐩)_i`$ is the $`\overline{\nu }_e`$ survival probability, averaged over the appropriate energy range for the i-$`th`$ bin, taking into account the detector energy resolution and the reactor core size. Numerical values for the rates $`R_i^0`$ and the errors $`s_i`$, $`u_i`$, and $`c_i`$, can be found in .
We have checked that, through the above $`\chi ^2`$ function, we can accurately reproduce the published CHOOZ limits (analysis A of ) in the parameter space $`𝐩=\{\mathrm{\Delta }m^2,s_{13}^2\}`$. For the sake of precision, in this work we have used the most general $`3\nu `$ parameter space for CHOOZ
$$𝐩=\{\pm \mathrm{\Delta }m^2,\delta m^2,s_{13}^2,s_{12}^2\}.$$
(69)
The effect of the subdominant parameters $`[\delta m^2,s_{12}^2,\mathrm{sign}(\pm \mathrm{\Delta }m^2)]`$ is, however, rather small in the current data analysis.
### 7.2 KamLAND
The published KamLAND energy spectrum contains 258 events (background+signal), which we analyze through a maximum-likelihood approach , described in detail in . In particular, the KamLAND (KL) $`\chi ^2`$ function is defined as
$$\chi _{\mathrm{KL}}^2(𝐩)=2\mathrm{ln}\underset{(\alpha ,\alpha ^{},\alpha ^{\prime \prime })}{\mathrm{max}}_{\mathrm{KL}}(𝐩,\alpha ,\alpha ^{},\alpha ^{\prime \prime }),$$
(70)
where $`\alpha `$ parametrizes a systematic energy-scale offset, while $`\alpha ^{}`$ and $`\alpha ^{\prime \prime }`$ represent free normalization factors for two (poorly constrained) background components . The above likelihood function is factorized as
$$_{\mathrm{KL}}(𝐩,\alpha ,\alpha ^{},\alpha ^{\prime \prime })=_{\mathrm{rate}}(𝐩,\alpha ,\alpha ^{},\alpha ^{\prime \prime })\times _{\mathrm{shape}}(𝐩,\alpha ,\alpha ^{},\alpha ^{\prime \prime })\times _{\mathrm{syst}}(\alpha ),$$
(71)
where the first, second, and third term embed the probability distribution for the total rate, for the spectrum shape, and for the systematic offset $`\alpha `$, respectively; explicit expressions are reported in . In particular, the spectrum shape term is further factorized into the probability distributions $`D(E_i)`$ for finding the 258 KamLAND events with observed energies $`\{E_i\}_{i=1,\mathrm{},258}`$:
$$_{\mathrm{shape}}=\underset{i=1}{\overset{258}{}}D(E_i).$$
(72)
A final remark is in order. In the KamLAND analysis was performed in the $`2\nu `$ parameter space $`𝐩=\{\delta m^2,s_{12}\}`$, where the published bounds have been accurately reproduced. In this work we have instead used the full $`3\nu `$ parameter space relevant for KamLAND,
$$𝐩=\{\delta m^2,s_{12}^2,s_{13}^2\}.$$
(73)
We have checked, for a number of representative points in the $`3\nu `$ parameter space, that the addition of KamLAND time-variation information does not alter in any appreciable way the KamLAND constraints in such $`3\nu `$ space.
### 7.3 SK atmospheric data
Our SK data analysis includes the zenith angle distributions of leptons induced by atmospheric neutrinos, for a total of 55 bins, as discussed in . With respect to , we have updated the experimental event rates and their statistical errors $`\{R_n^{\mathrm{expt}}\pm \sigma _n^{\mathrm{stat}}\}`$ from . We also use three-dimensional neutrino fluxes for the calculation of the theoretical rates $`R_n^{\mathrm{theo}}(𝐩)`$. For convenience, we normalize both the experimental and theoretical rates to their no-oscillation value in each bin, as shown in Fig. 18. We consider eleven sources of systematic errors, which can produce a shift of the theoretical predictions through a set of “pulls” $`\{\xi _k\}_{k=1,\mathrm{},11}`$ ,
$$\overline{R}_n^{\mathrm{theo}}(𝐩)=R_n^{\mathrm{theo}}(𝐩)+\underset{k=1}{\overset{11}{}}\xi _kc_n^k,$$
(74)
where the response $`c_n^k`$ of the $`n`$-th bin to the $`k`$-th systematic source is numerically given in . The $`\chi ^2`$ function is then obtained by minimization over the $`\xi _k`$’s (which are partly correlated through a matrix $`\rho _{hk}^{\mathrm{syst}}`$ ),
$$\chi _{\mathrm{SK}}^2(𝐩)=\underset{\{\xi _k\}}{\mathrm{min}}\left[\underset{n=1}{\overset{55}{}}\left(\frac{\overline{R}_n^{\mathrm{theo}}(𝐩)R_n^{\mathrm{expt}}}{\sigma _n^{\mathrm{stat}}}\right)^2+\underset{k,h=1}{\overset{11}{}}\xi _k[\rho _{\mathrm{syst}}^1]_{hk}\xi _h\right].$$
(75)
Minimization leads to a set of linear equations in the $`\xi _k`$’s, which are solved numerically. The solution $`\overline{\xi }_k`$ can provide useful statistical information about the preferred systematic offsets and theoretical rate shifts, as discussed in Sec. 4.1.
Finally, the $`3\nu `$ parameter space used for the SK data analysis in this work is
$$𝐩=\{\pm \mathrm{\Delta }m^2,s_{23}^2,s_{13}^2,\mathrm{cos}\delta =\pm 1,\delta m^2,s_{12}^2\},$$
(76)
where $`\delta m^2`$ and $`s_{12}^2`$ are the (fixed) best-fit values from the solar+KamLAND data analysis in Sec. 3.1. We have verified, in a number of representative points, that variations of these two parameters within their $`\pm 2\sigma `$ limits do not alter the SK atmospheric data analysis in any appreciable way.
### 7.4 K2K
In this work, the K2K analysis is based on a 6-bin energy spectrum as in , but including updated data as shown in Fig. 17. We cannot perform a K2K spectral analysis with finer binning (as the official one ) for lack of published information about bin-to-bin systematic errors and their correlations in the last data release. The $`\chi ^2`$ definition is based on a pull approach (with 7 systematic error sources ), but the small number $`N`$ of events in each bin requires a Poisson statistics, implemented through
$$\chi _{\mathrm{K2K}}^2(𝐩)=\underset{\{\xi _k\}}{\mathrm{min}}\left[2\underset{n=1}{\overset{6}{}}\left(\overline{N}_n^{\mathrm{theo}}N_n^{\mathrm{expt}}N_n^{\mathrm{expt}}\mathrm{ln}\frac{\overline{N}_n^{\mathrm{theo}}}{N_n^{\mathrm{expt}}}\right)+\underset{k,h=1}{\overset{7}{}}\xi _k[\rho _{\mathrm{syst}}^1]_{hk}\xi _h\right],$$
(77)
with shifted predictions
$$\overline{N}_n^{\mathrm{theo}}(𝐩)=N_n^{\mathrm{theo}}(𝐩)+\underset{k=1}{\overset{7}{}}\xi _kc_n^k.$$
(78)
Numerical values for the K2K response functions $`c_n^k`$ and for the correlation matrix $`\rho _{hk}^{\mathrm{syst}}`$ are given in .
For the sake of precision, and for consistency in the SK+K2K(+CHOOZ) combination, we have used for K2K the same $`3\nu `$ parameter as for SK
$$𝐩=\{\pm \mathrm{\Delta }m^2,s_{23}^2,s_{13}^2,\mathrm{cos}\delta =\pm 1,\delta m^2,s_{12}^2\},$$
(79)
although the subleading effects of the last two parameters are rather small in the K2K data analysis.
### 7.5 Solar neutrinos
The definition of the solar neutrino $`\chi ^2`$ is rather complex, both because it includes 119 observables and 55 sources of systematics, and because it currently involves also correlations of statistical errors in the SNO salt data and of systematic error sources in the BS05 (OP) solar model . Here we will mainly highlight the differences of the new $`\chi ^2`$ inputs, with respect to our previous discussion in .<sup>12</sup><sup>12</sup>12Such new input includes, for all solar neutrino observables, also the electron density and neutrino production profiles from the BS05 (OP) SSM , which are relevant for calculating solar matter effects on neutrino flavor evolution.
The formal $`\chi ^2`$ definition is based on a pull approach,
$$\chi _{\mathrm{sol}}^2(𝐩)=\underset{\{\xi _k\}}{\mathrm{min}}\left[\underset{n,m}{\overset{119}{}}x_n[\rho ^1]_{nm}x_m+\underset{h,k}{\overset{55}{}}\xi _h[\rho _{\mathrm{syst}}^1]_{hk}\xi _k\right],$$
(80)
where
$$x_n=\frac{\overline{R}_n^{\mathrm{theo}}(𝐩)R_n^{\mathrm{expt}}}{\sigma _n},$$
(81)
and
$$\overline{R}_n^{\mathrm{theo}}=R_n^{\mathrm{theo}}+\underset{k=1}{\overset{55}{}}\xi _kc_n^k.$$
(82)
The presence of statistical error correlations ($`\rho _{nm}\delta _{nm}`$) in the latest SNO data does not spoil the advantages of the pull approach discussed in (where $`\rho _{nm}=\delta _{nm}`$). We remind that the parameter space used in the solar neutrino data analysis is
$$𝐩=\{\delta m^2,s_{12}^2,s_{13}^2\},$$
(83)
except for Fig. 13 and related comments, where the very weak sensitivity to $`\pm \mathrm{\Delta }m^2`$ has been discussed.
In the following we briefly describe, in ascending order, the $`n=1\mathrm{}119`$ observables and the $`k=1\mathrm{}55`$ systematic error sources of the solar neutrino analysis.
#### 7.5.1 Observables
$`n=1`$ (Chlorine rate). The Chlorine rate input is $`R_1^{\mathrm{expt}}\pm \sigma _1^{}=2.56\pm 0.23`$ SNU. A $`𝐩`$-dependent cross-section error is added in quadrature to $`\sigma _1^{}`$ (as described in ) to obtain the total uncorrelated error $`\sigma _1`$.
$`n=2,3`$ (Gallium total rate and winter-summer asymmetry). The Gallium (GALLEX/GNO+SAGE) input for the total rate is $`R_2^{\mathrm{expt}}\pm \sigma _2^{}=68.1\pm 3.75`$ SNU . A $`𝐩`$-dependent cross-section error is added in quadrature to $`\sigma _2^{}`$ (as described in ) to obtain the total uncorrelated error $`\sigma _2`$. The combined (GALLEX/GNO+SAGE) input for the winter-summer asymmetry (corrected for geometrical eccentricity effects ) is $`R_3\pm \sigma _3=0.6\pm 7`$ SNU .
$`n=4,\mathrm{},47`$ (SK solar neutrino spectrum in energy and nadir angle). The analysis of the SK 44-bin spectrum uses the same experimental input as described in appendix C of , but all the theoretically computed quantities have been updated (in each point of the parameter space $`p`$) to account for the new hep and <sup>8</sup>B solar neutrino input in the BS05 (OP) model .
$`n=48,\mathrm{},81`$ (SNO spectrum, pure D<sub>2</sub>O phase). In this data set , events from NC, CC and ES scattering (and from various backgrounds) are not separated, and the global NC+CC+ES energy spectrum is analyzed. The spectrum information includes 34 bins (17 day + 17 night ) and is treated as described in appendix D of . As for the above (SK) data set, the only update concerns theoretical calculations, in order to account for BS05 (OP) solar model hep and <sup>8</sup>B input.
$`n=82,\mathrm{},119`$ (SNO CC spectrum and ES+NC rates, salt phase). In this recent SNO data set , the addition of salt has allowed a statistical separation of CC, NC and ES events. The CC spectrum includes 34 bins (17 day and 17 night, $`n=82,\mathrm{}115`$). Four more data points concern the total NC rate (day and night, $`n=116,117`$) and ES rate (day and night, $`n=118,119`$). The experimental values for the rates and their statistical errors $`(R_n^{\mathrm{expt}}\pm \sigma _n)`$ are taken from Table XXX (CC) and XXIV (NC+ES) of . These are the only data where the (statistical) errors are correlated, namely, $`\rho _{n,m>82}\delta _{nm}`$. The two ($`19\times 19`$) statistical correlation matrices for day and night data are taken from Tables XXXII and XXXIII of , respectively.
#### 7.5.2 Systematics
$`k=1,\mathrm{},20`$ (Standard Solar Model systematics). In 12 sources $`X_k`$ of systematic errors were considered in the input solar model, individually denoted as $`S_{11}`$, $`S_{33}`$, $`S_{34}`$, $`S_{1,14}`$, $`S_{17}`$, Lum, Opa, Diff, $`C_{\mathrm{Be}}`$, $`S_{\mathrm{hep}}`$, and $`Z/X`$, each source being affected by a relative error $`\mathrm{\Delta }X_k/X_k`$ (see also and refs. therein). With respect to , the first 11 systematic uncertainties are unchanged, except for the updated value of $`\mathrm{\Delta }X/X`$ for $`S_{17}`$, currently set to 0.038 (it was 0.106 in ). The former systematic error source $`Z/X`$ (solar metallicity) is now separated into 9 elemental uncertainties related to C, N, O, Ne, Mg, Si, S, Ar, and Fe, whose $`\mathrm{\Delta }X_k/X_k`$ values are taken from the conservative estimate in Table 4 of . Three among these new error sources (O, Ne, Ar) are fully correlated , and make the matrix $`\rho _{hk}^{\mathrm{syst}}`$ different from unity for (and only for) $`h,k=14,15,19`$. The 11+9=20 SSM systematic error sources $`X_k`$ affect the neutrino fluxes through log-derivatives $`\alpha _{ik}`$ (see ). For $`k=1,\mathrm{},11`$, such derivatives are unchanged with respect to . For the new metallicity uncertainties ($`k=12,\mathrm{},20`$), we take the log-derivatives from Table 1 of . Finally, all SSM systematic uncertainties are propagated to the final $`c_n^k`$ values in each point of the parameter space, as described in .
$`k=21`$ (<sup>8</sup>B spectrum shape uncertainty). The treatment of this systematic error (which affect all solar neutrino observables) is unchanged from .
$`k=22,\mathrm{},32`$ (SK spectrum) and $`k=33,\mathrm{},39`$ (SNO no-salt spectrum). The SK spectrum data $`\{R_n\}_{n=4\mathrm{},47}`$ are affected by 11 systematics $`k=22,\mathrm{},32`$. Analogously, the SNO (pure D<sub>2</sub>O phase, $`\{R_n\}_{n=48,\mathrm{},81}`$) data are affected by 7 systematics. They are treated as described in .
$`k=40,\mathrm{},55`$ (SNO salt-phase systematics). The recent SNO data in the salt phase $`\{R_n\}_{n=82,\mathrm{},119}`$ are affected (in equal way during day and night) by 16 sources of systematic errors . The response function $`c_n^k`$ is numerically given in Table XXXIV of , in terms of (generally asymmetrical) upper and lower values $`c_n^{k+}`$ and $`c_n^k`$. Such values cannot be incorporated at face value in a $`\chi ^2`$ function, which postulates symmetrical uncertainties. As mentioned in Sec. 3, in the pull approach we solve this problem through the prescription suggested in , namely, by shifting the central values of the observables (through the error half-difference) and by attaching symmetrized (half-sum) systematic errors as
$$\overline{R}_n=R_n+\underset{k=40}{\overset{55}{}}\frac{c_n^{k+}c_n^k}{2}+\underset{k=40}{\overset{55}{}}\xi _k\frac{c_n^{k+}+c_n^k}{2}(n=82,\mathrm{},119).$$
(84)
### 7.6 Observables sensitive to absolute neutrino masses
For the $`\chi ^2`$ functions related to $`m_\beta `$, $`m_{\beta \beta }`$, and $`\mathrm{\Sigma }`$, we refer the reader to the thorough discussion given in . With respect to , in this work we have only reduced the theoretical nuclear uncertainty affecting $`m_{\beta \beta }`$, according to the recent results reported in (see Sec. 6.2).
|
warning/0506/math0506462.html
|
ar5iv
|
text
|
# The Low Lying Zeros of a GL(4) and a GL(6) family of 𝐿-functions
## 1. Introduction
Assuming GRH, the non-trivial zeros of any $`L`$-function lie on its critical line, and therefore it is possible to investigate the statistics of its normalized zeros. The general philosophy, born out of many examples and proven cases in function fields \[CFKRS, KS1, KS2, KeSn, ILS\], is that the statistical behavior of eigenvalues of random matrices (resp., random matrix ensembles) is similar to that of the critical zeros of $`L`$-functions (resp., families of $`L`$-functions).
The global $`n`$-level correlations of high zeros of primitive automorphic cuspidal $`L`$-functions, assuming a certain technical restriction, have been found to agree with the corresponding statistics of the eigenvalues of complex hermitian matrices (the Gaussian Unitary Ensemble, or GUE) \[Mon, Hej, RS\]. If the technical restriction mentioned above were to be removed, the results on $`n`$-level correlations would imply that the distributions of the normalized neighbor spacings between consecutive critical zeros of an $`L`$-function and between GUE eigenvalues coincide, as has been numerically observed \[Od, Ru1\]. The same correlations describe the global statistical behavior of the eigenvalues of other matrix ensembles, most notably of the classical compact groups (orthogonal, unitary, symplectic). Being insensitive to the effect of finitely many zeros, these correlations miss the behavior of the low lying zeros, the zeros near the central point $`s=1/2`$.
Katz and Sarnak \[KS1, KS2\] showed that there is another statistic that can distinguish between the classical compact groups. It is the $`n`$-level density, and it depends only on eigenvalues near $`1`$. In a number of cases \[FI, Gü, HM, HR2, ILS, Mil2, Ro, Ru2, Yo\], the behavior of the low lying zeros of families of $`L`$-functions is found to be in agreement with that of the eigenvalues near $`1`$ for random matrices in one of the classical compact groups: unitary, symplectic, and orthogonal (which is further split into $`\mathrm{SO}(\mathrm{even})`$ and $`\mathrm{SO}(\mathrm{odd})`$). This correspondence allows us, at least conjecturally, to assign a definite symmetry type to each family of $`L`$-functions.
Let $`\varphi `$ be a fixed even Hecke-Maass cusp form and $`H_k`$ a Hecke eigenbasis for the space of holomorphic cusp forms of (even) weight $`k`$ for the full modular group. Iwaniec-Luo-Sarnak \[ILS\] proved that as $`k\mathrm{}`$, the family $`\{f:fH_k\}`$ has $`\mathrm{SO}(\mathrm{even})`$ or $`\mathrm{SO}(\mathrm{odd})`$ symmetry (depending on whether $`k/2`$ is even or odd), and the family $`\{\mathrm{sym}^2f:fH_k\}`$ has symplectic symmetry. We consider the twisted families $`_{\varphi \times H_k}`$ $`=`$ $`\{\varphi \times f:fH_k\}`$ and $`_{\varphi \times \mathrm{sym}^2H_k}`$ $`=`$ $`\{\varphi \times \mathrm{sym}^2f:fH_k\}`$; the family $`\{\varphi \times \mathrm{sym}^2f\}`$ arose in the work of Luo-Sarnak \[LS\], where it is shown to be intimately connected with the relation between the quantum and classical fluctuations of observables on the modular surface. In both families, all functional equations are even. We show that the first family has symplectic symmetry, and the second $`\mathrm{SO}(\mathrm{even})`$. Explicitly, our main results are
###### Theorem 1.1.
Let $`\varphi `$ be a fixed even Hecke-Maass cusp form. As $`k\mathrm{}`$, for test functions whose Fourier transform has small but computable support, the $`1`$-level density of the family $`_{\varphi \times H_k}`$ only agrees with symplectic matrices, suggesting that the underlying symmetry of this family is symplectic (and uniquely so).
###### Theorem 1.2.
Let $`\varphi `$ be a fixed even Hecke-Maass cusp form. As $`k\mathrm{}`$, for test functions $`g`$ with $`\mathrm{supp}(\widehat{g})(\frac{5}{24},\frac{5}{24})`$, the $`1`$-level density of the family $`_{\varphi \times \mathrm{sym}^2H_k}`$ only agrees with $`\mathrm{SO}(\mathrm{even})`$, $`\mathrm{O}`$ and $`\mathrm{SO}(\mathrm{odd})`$ matrices. For small but computable support, the $`2`$-level density only agrees with $`\mathrm{SO}(\mathrm{even})`$ matrices, suggesting that the underlying symmetry of this family is $`\mathrm{SO}(\mathrm{even})`$ (and uniquely so).
For families where the signs of the functional equations are all even and there is no corresponding family with odd functional equations, a “folklore” conjecture (for example, see page 2877 of \[KeSn\]) states that the symmetry is symplectic, presumably based on the observation that $`\mathrm{SO}(\mathrm{even})`$ and $`\mathrm{SO}(\mathrm{odd})`$ symmetries in the examples known to date arise from splitting orthogonal families according to the sign of the functional equations. *A priori* the symmetry type of a family with all functional equations even is either symplectic or SO(even). All $`L`$-functions from elements of $`_{\varphi \times H_k}`$ and $`_{\varphi \times \mathrm{sym}^2H_k}`$ have even functional equations, and neither family seems to naturally arise from splitting sign within a full orthogonal family. By calculating the 1-level density we quickly see the symmetry of the first is symplectic (as predicted); however, the second family has orthogonal symmetry (we cannot distinguish between $`\mathrm{SO}(\mathrm{even}),\mathrm{O}`$ and $`\mathrm{SO}(\mathrm{odd})`$ due to the small-support restriction on the allowable test functions). By calculating the 2-level density for the second family, we can discard $`\mathrm{O}`$ and $`\mathrm{SO}(\mathrm{odd})`$. Thus our calculations are only consistent with the symmetry being $`\mathrm{SO}(\mathrm{even})`$. As our purpose is to show that the theory of low lying zeros is more than just a theory of signs of functional equations, we do not concern ourselves with obtaining optimal bounds in terms of support, instead simplifying the arguments but still distinguishing the various classical compact group candidates.
In studying the symmetry groups of $`_{\varphi \times H_k}`$ and $`_{\varphi \times \mathrm{sym}^2H_k}`$, we see that twisting a family with orthogonal (respectively, symplectic) symmetry by a fixed $`\mathrm{GL}(2)`$ form flips the symmetry to symplectic (respectively, orthogonal). The effect on the symmetry group by $`\mathrm{GL}(n)`$ twisting (by a fixed form, or by a second family) in some cases will be described in a subsequent paper (\[DM\]). The main result is that, for any family $``$ satisfying certain technical conditions, we can attach a symmetry constant $`c_{}`$, with $`c_{}=0`$ $`(1,1)`$ if the family is unitary (symplectic, orthogonal). For such families $``$ and $`𝒢`$, the family $`\times 𝒢`$ (Rankin-Selberg convolution) has symmetry constant $`c_{\times 𝒢}=c_{}c_𝒢`$ (compare with \[KS1\]). In other words, the symmetry of a product of families is the product of the family symmetries. This is consistent with earlier results and should be compared, for instance, with Rubinstein’s work \[Ru2\] on twisting the symplectic family of quadratic Dirichlet characters by a fixed $`\mathrm{GL}(n)`$ form.
We assume the Generalized Riemann Hypothesis for all $`L`$-functions encountered. Mostly GRH is used for interpretation purposes (i.e., if GRH is true than the non-trivial zeros lie on the critical line, and we may interpret the $`n`$-level correlations and densities as spacing statistics between ordered zeros), though in a few places GRH is assumed to simplify the derivation of needed bounds (though these bounds can be derived unconditionally at the cost of a more careful analysis). In §2 we review the necessary preliminaries. We concentrate on the more difficult $`\mathrm{GL}(6)`$ family in §3, and merely sketch the changes needed to handle the $`\mathrm{GL}(4)`$ family in §4; for completeness the details of the calculation of the gamma factors and signs of the functional equations are given in Appendix A. In §5 we analyze our results for these two families. The evidence suggests that the theory of low lying zeros is not just a theory of signs of functional equations, but rather more about the second moment of the Satake parameters. In this regard it is similar to the universality Rudnick and Sarnak \[RS\] found for the $`n`$-level correlations of high zeros of a primitive $`L`$-function $`L(s,\pi )`$ ($`\pi `$ a cuspidal automorphic representation); their results are a consequence of the universality of the second moments of the Satake parameters $`a_\pi (p)`$.
## 2. Preliminaries
### 2.1. $`1`$\- and $`2`$-Level Densities
Let $`g`$ be an even Schwartz test function on $``$ whose Fourier transform
$$\widehat{g}(y)=_{\mathrm{}}^{\mathrm{}}g(x)e^{2\pi ixy}dx$$
(2.1)
has compact support. Let $``$ be a finite family, all of whose $`L`$-functions satisfy GRH. We define the $`1`$-level density associated to $``$ by
$$D_{1,}(g)=\frac{1}{||}\underset{f}{}\underset{j}{}g\left(\frac{\mathrm{log}c_f}{2\pi }\gamma _f^{(j)}\right),$$
(2.2)
where $`\frac{1}{2}+i\gamma _f^{(j)}`$ runs through the non-trivial zeros of $`L(s,f)`$. Here $`c_f`$ is the analytic conductor of $`f`$, and gives the natural scale for the low zeros. Since $`g`$ is Schwartz, only low lying zeros (i.e., zeros within a distance $`\frac{1}{\mathrm{log}c_f}`$ of the central point) contribute significantly. Thus the $`1`$-level density is a local statistic which can potentially help identify the symmetry type of the family.
###### Remark 2.1.
For technical convenience, as in \[ILS, Ro\] we will modify (2.2) by weighting each $`f`$ by a factor $`w_f`$ which varies slowly with $`f`$. These factors simplify applying the Petersson formula, and (see \[ILS\]) can be removed at the cost of additional book-keeping.
Based in part on the function-field analysis where $`G()`$ is the monodromy group associated to the family $``$, it is conjectured that for each reasonable irreducible family of $`L`$-functions there is an associated symmetry group $`G()`$ (typically one of the following five subgroups of unitary matrices: unitary $`U`$, symplectic $`\mathrm{USp}`$, orthogonal $`\mathrm{O}`$, $`\mathrm{SO}(\mathrm{even})`$, $`\mathrm{SO}(\mathrm{odd})`$), and that the distribution of critical zeros near $`\frac{1}{2}`$ mirrors the distribution of eigenvalues near $`1`$. The five groups have distinguishable $`1`$-level densities.
To evaluate (2.2), one applies the explicit formula, converting sums over zeros to sums over primes. Unfortunately, these prime sums can often only be evaluated for small support. If one allows test functions with $`\mathrm{supp}(\widehat{g})(\delta ,\delta )`$, then for any $`\delta >0`$ the orthogonal, symplectic and unitary symmetries can be mutually distinguished via their $`1`$-level density. However, if $`\delta 1`$ then the $`1`$-level densities of the three orthogonal types O, SO(even) and SO(odd) cannot be distinguished from one another.
In order to distinguish between the three orthogonal symmetry types we study the $`2`$-level density of the family, defined as follows. Let $`g(x)=g_1(x_1)g_2(x_2)`$, each $`\widehat{g_i}`$ of compact support. Then
$$D_{2,}(g)=\frac{1}{||}\underset{f}{}\underset{j_1\pm j_2}{}g_1\left(\frac{\mathrm{log}c_f}{2\pi }\gamma _f^{(j_1)}\right)g_2\left(\frac{\mathrm{log}c_f}{2\pi }\gamma _f^{(j_2)}\right),$$
(2.3)
Miller \[Mil1\] observed that an advantage of studying the $`2`$-level density is that, even for arbitrarily small support, the three orthogonal types of symmetry are mutually distinguishable (see \[Mil2\] where it is used to discern the symmetry group of families of elliptic curves). An analogous definition holds for the $`n`$-level density; as the signs of our families are constant, our arguments can easily be extended to determining the $`n`$-level density (though the support decreases with $`n`$).
By \[KS1\], the $`n`$-level densities for the classical compact groups are
$`\begin{array}{ccc}W_{n,\mathrm{SO}(\mathrm{even})}(x)\hfill & =& det(K_1(x_i,x_j))_{i,jn}\hfill \\ W_{n,\mathrm{SO}(\mathrm{odd})}(x)\hfill & =& det(K_1(x_i,x_j))_{i,jn}+_{k=1}^n\delta (x_k)det(K_1(x_i,x_j))_{i,jk}\hfill \\ W_{n,\mathrm{O}}(x)\hfill & =& \frac{1}{2}W_{n,\mathrm{SO}(\mathrm{even})}(x)+\frac{1}{2}W_{n,\mathrm{SO}(\mathrm{odd})}(x)\hfill \\ W_{n,\mathrm{U}}(x)\hfill & =& det(K_0(x_i,x_j))_{i,jn}\hfill \\ W_{n,\mathrm{USp}}(x)\hfill & =& det(K_1(x_i,x_j))_{i,jn},\hfill \end{array}`$ (2.9)
where $`K(y)=\frac{\mathrm{sin}\pi y}{\pi y}`$, $`K_ϵ(x,y)=K(xy)+ϵK(x+y)`$ for $`ϵ=0,\pm 1`$ and $`\delta (x)`$ is the Dirac delta functional; see \[HM\] for a more tractable formula for the $`n`$<sup>th</sup> centered moments for test functions whose Fourier transforms have support suitably restricted. It is often more convenient to work with the Fourier transforms of the densities. For the 1-level densities we have
$`\begin{array}{ccc}\widehat{W}_{1,\mathrm{SO}(\mathrm{even})}(u)\hfill & =& \delta (u)+\frac{1}{2}I(u)\hfill \\ \widehat{W}_{1,\mathrm{SO}(\mathrm{odd})}(u)\hfill & =& \delta (u)\frac{1}{2}I(u)+1\hfill \\ \widehat{W}_{1,\mathrm{O}}(u)\hfill & =& \delta (u)+\frac{1}{2}\hfill \\ \widehat{W}_{1,U}(u)\hfill & =& \delta (u)\hfill \\ \widehat{W}_{1,\mathrm{USp}}(u)\hfill & =& \delta (u)\frac{1}{2}I(u),\hfill \end{array}`$ (2.15)
where $`I(u)`$ is the characteristic function of $`[1,1]`$. The three orthogonal densities are indistinguishable for test functions of small support. Explicitly, for test functions $`g`$ such that $`\mathrm{supp}(\widehat{g})(1,1)`$, we have
$`\begin{array}{ccc}\widehat{g}(u)\widehat{W}_{1,\mathrm{SO}(\mathrm{even})}(u)du\hfill & =& \widehat{g}(u)+\frac{1}{2}g(0)\hfill \\ \widehat{g}(u)\widehat{W}_{1,\mathrm{SO}(\mathrm{odd})}(u)du\hfill & =& \widehat{g}(u)+\frac{1}{2}g(0)\hfill \\ \widehat{g}(u)\widehat{W}_{1,\mathrm{O}}(u)du\hfill & =& \widehat{g}(u)+\frac{1}{2}g(0)\hfill \\ \widehat{g}(u)\widehat{W}_{1,\mathrm{U}}(u)du\hfill & =& \widehat{g}(u)\hfill \\ \widehat{g}(u)\widehat{W}_{1,\mathrm{USp}}(u)du\hfill & =& \widehat{g}(u)\frac{1}{2}g(0).\hfill \end{array}`$ (2.21)
We record the effect of the Fourier transform of the 2-level density kernel on our test functions. Let $`c(𝒢)=0`$ (respectively $`\frac{1}{2}`$, $`1`$) for $`𝒢=\mathrm{SO}(\mathrm{even})`$ (respectively $`\mathrm{O}`$, $`\mathrm{SO}(\mathrm{odd})`$). For even functions $`\widehat{g}_1(u_1)\widehat{g}_2(u_2)`$ supported in $`|u_1|+|u_2|<1`$,
$`\begin{array}{ccc}\widehat{g_1}(u_1)\widehat{g_2}(u_2)\widehat{W_{2,𝒢}}(u)du_1du_2\hfill & =& \left[\widehat{g}_1(0)+\frac{1}{2}g_1(0)\right]\left[\widehat{g}_2(0)+\frac{1}{2}g_2(0)\right]\hfill \\ & & +2|u|\widehat{g}_1(u)\widehat{g}_2(u)du\hfill \\ & & 2\widehat{g_1g_2}(0)g_1(0)g_2(0)\hfill \\ & & +c(𝒢)g_1(0)g_2(0).\hfill \end{array}`$ (2.23)
Thus, for arbitrarily small support, the 2-level density distinguishes the three orthogonal groups (see \[Mil1\] for the calculation).
### 2.2. Cusp Forms
We quickly review some facts about cusp forms; see \[Iw2, ILS\] for details. Let $`S_k`$ be the space of holomorphic cusp forms of weight $`k`$ (an even positive integer) and level $`1`$ (that is, for the full modular group $`\mathrm{\Gamma }=SL(2,)`$). Let $`H_k`$ be a basis of Hecke eigenforms. Then
$$dimS_k=|H_k|=\frac{k}{12}+O(1).$$
(2.24)
Any $`fH_k`$ has a Fourier expansion
$$f(z)=\underset{n=1}{\overset{\mathrm{}}{}}a_f(n)e(nz),$$
(2.25)
and we shall henceforth assume $`f`$ is normalized so that $`a_f(1)=1`$. Two other useful normalizations for the coefficients are
$`\lambda _f(n)`$ $`=a_f(n)n^{\frac{k1}{2}}`$ (2.26)
$`\psi _f(n)`$ $`=\sqrt{{\displaystyle \frac{\mathrm{\Gamma }(k1)}{(4\pi n)^{k1}}}}{\displaystyle \frac{1}{f}}a_f(n),`$ (2.27)
with $`f`$ the Petersson $`L^2`$-norm of $`f`$. As mentioned in Remark 2.1, $`\psi _f(n)`$ will lead to a weighted sum which simplifies the application of the Petersson formula. Essential in our investigations will be the multiplicativity properties of the Fourier coefficients.
###### Lemma 2.2.
Let $`f`$ be a cuspidal Hecke eigenform of level $`1`$. Then
$$\lambda _f(m)\lambda _f(n)=\underset{d|(m,n)}{}\lambda _f\left(\frac{mn}{d^2}\right).$$
(2.28)
In particular, we have
###### Corollary 2.3.
Let $`(m,n)=1`$, and $`p`$ be a prime. Then
$`\lambda _f(m)\lambda _f(n)`$ $`=`$ $`\lambda _f(mn)`$
$`\lambda _f(p)^2`$ $`=`$ $`\lambda _f(p^2)+1.`$ (2.29)
### 2.3. Summation Formulas
We recall some standard formulas for summing Fourier coefficients over our families and test functions over primes.
###### Definition 2.4 (Diagonal Symbol).
$`\mathrm{\Delta }_k(m,n)`$ $`=`$ $`{\displaystyle \underset{fH_k}{}}\psi _f(m)\overline{\psi }_f(n)`$
$`\delta (m,n)`$ $`=`$ $`\{\begin{array}{cc}1\hfill & \text{if }m=n\hfill \\ 0\hfill & \mathrm{otherwise}.\hfill \end{array}`$ (2.30)
We rephrase the results from \[ILS\] in our language. By their equations 2.8, 2.52-2.54, and recalling that $`|H_k|=\frac{k}{12}+O(1)`$, we find
$`\mathrm{\Delta }_k(m,n)`$ $`=`$ $`{\displaystyle \frac{\zeta (2)}{|H_k|+O(1)}}{\displaystyle \underset{fH_k}{}}{\displaystyle \frac{\lambda _f(m)\lambda _f(n)}{L(1,\mathrm{sym}^2f)}}.`$ (2.31)
###### Lemma 2.5 (Petersson Formula).
For $`(m,n)=1`$, $`m`$ and $`n`$ of at most $`b`$ factors,
$$\mathrm{\Delta }_k(m,n)=\delta (m,n)+O_b\left(\frac{m^{\frac{1}{4}}n^{\frac{1}{4}}\mathrm{log}mn}{k^{\frac{5}{6}}}\right).$$
(2.32)
For $`m,n`$ as above and $`12\pi \sqrt{mn}k`$,
$$\mathrm{\Delta }_k(m,n)=\delta (m,n)+O\left(\frac{\sqrt{mn}}{2^k}\right).$$
(2.33)
Note (2.32) and (2.33) are Corollaries 2.2 and 2.3 of \[ILS\].
The Petersson formula allows us to easily evaluate certain weighted sums of the Fourier coefficients. As $`f`$ is related to $`L(1,\mathrm{sym}^2f)`$, the natural weights are the harmonic weights $`\omega _f=\zeta (2)/L(1,\varphi \times \mathrm{sym}^2f)`$. These weights are almost constant (see (3.34)), and following \[ILS\] we may remove these weights in the applications below. See §3.2.1 for more details. We call terms with $`m=n`$ *diagonal* terms; the remaining terms are called *non-diagonal*. For small support, the non-diagonal terms will not contribute.
###### Remark 2.6.
There exist explicit formulas, involving Bessel functions and Kloosterman sums, for the error terms in the expansion of $`\mathrm{\Delta }_k(m,n)`$. For the families studied in \[ILS\], by analyzing these terms’ contributions they are able to work with test functions with support greater than $`[1,1]`$, and hence distinguish $`\mathrm{SO}(\mathrm{even})`$ from $`\mathrm{SO}(\mathrm{odd})`$; see also \[HM\] where these terms are handled for the $`n`$-level densities. Increasing the support has applications to non-vanishing results at the central point. We will not be able to obtain such large support for our families; however, by studying the $`2`$-level density, we can still distinguish $`\mathrm{SO}(\mathrm{even})`$ from $`\mathrm{SO}(\mathrm{odd})`$.
The following are immediate applications of the Prime Number Theorem:
###### Lemma 2.7.
Let $`\widehat{F}`$ be an even Schwartz function of compact support. Then for any positive integer $`a`$,
$`{\displaystyle \underset{p}{}}\widehat{F}\left(a{\displaystyle \frac{\mathrm{log}p}{\mathrm{log}R}}\right){\displaystyle \frac{\mathrm{log}p}{\mathrm{log}R}}{\displaystyle \frac{1}{p}}`$ $`={\displaystyle \frac{1}{2a}}F(0)+O\left({\displaystyle \frac{1}{\mathrm{log}R}}\right)`$ (2.34)
$`{\displaystyle \underset{p}{}}\widehat{F}\left({\displaystyle \frac{\mathrm{log}p}{\mathrm{log}R}}\right){\displaystyle \frac{4\mathrm{log}^2p}{\mathrm{log}^2R}}{\displaystyle \frac{1}{p}}`$ $`=2{\displaystyle _{\mathrm{}}^{\mathrm{}}}|u|\widehat{F}(u)du+O\left({\displaystyle \frac{1}{\mathrm{log}R}}\right).`$ (2.35)
## 3. $`_{\varphi \times \mathrm{sym}^2H_k}=\{\varphi \times \mathrm{sym}^2f:fH_k\}`$
We provide evidence that the underlying symmetry of the family $`_{\varphi \times \mathrm{sym}^2H_k}=\{\varphi \times \mathrm{sym}^2f:fH_k\}`$ is $`\mathrm{SO}(\mathrm{even})`$. In §3.1 we calculate the needed quantities to investigate the distribution of the low lying zeros. In §3.2 we calculate the one-level density for test functions whose Fourier transform has small support, proving the first half of Theorem 1.2. Although the support is contained in $`(1,1)`$, the evidence is enough to discard the possibility of symplectic (or unitary) symmetry; however the $`1`$-level density in this range cannot distinguish between $`\mathrm{O},\mathrm{SO}(\mathrm{even})`$ and $`\mathrm{SO}(\mathrm{odd})`$ (even though the even functional equations suggest, of course, that $`\mathrm{SO}(\mathrm{even})`$ is the type). To rectify this deficiency, in §3.3 we calculate the 2-level density for small support; this suffices to eliminate $`\mathrm{O}`$ and $`\mathrm{SO}(\mathrm{odd})`$ and will complete the proof of Theorem 1.2.
### 3.1. Definition, Gamma Factors, Functional Equation
We use the notation of §2.2 for the holomorphic Hecke eigenform $`f`$ and its Hecke eigenvalues $`\lambda _f(n)`$. Let $`\varphi `$ be a fixed even Hecke-Maass cuspidal eigenform with Laplacian eigenvalue $`\lambda _\varphi =\frac{1}{4}+t_\varphi ^2`$ for the full modular group $`\mathrm{\Gamma }=\text{SL}(2,)`$. We normalize $`\varphi `$ so that $`a_\varphi (1)=1`$, and denote by $`\lambda _\varphi (n)`$ the corresponding Hecke eigenvalues.
For any unramified prime $`p`$, the Satake parameters (of the principal-series representation of $`GL_2(_p)`$) associated to $`f`$ are two complex numbers $`\alpha _p,\stackrel{~}{\alpha }_p=\alpha _p^1`$ satisfying
$$\lambda _f(p^\nu )=\underset{\mathrm{}=1}{\overset{\nu }{}}\alpha _p^{\mathrm{}}\stackrel{~}{\alpha }_{p}^{}{}_{}{}^{\nu \mathrm{}}.$$
(3.1)
Since $`f`$ is of level $`1`$, every prime is unramified. By the work of Deligne, $`|\alpha _p|=1`$ —the local representation is tempered— so that in fact $`\stackrel{~}{\alpha }_p`$ is the complex conjugate $`\overline{\alpha }_p`$ of $`\alpha _p`$. Thus $`\lambda _f(p)=\alpha _p+\alpha _p^1`$ alone determines $`\alpha _p,\alpha _p^1`$. By (3.1), all of the $`\lambda _f(p^\nu )`$ are algebraically expressible in terms of $`\lambda _f(p)`$ (formula (3.1) is indeed equivalent to the multiplicativity of the Fourier coefficients).
The Maass form $`\varphi `$ has Satake parameters $`\beta _p,\stackrel{~}{\beta }_p=\beta _p^1`$. The Ramanujan conjecture states that $`|\beta _p|=1`$; while this is still open for Maass forms, powerful bounds towards Ramanujan are available. Kim and Shahidi \[KiSh\] proved the crucial (for us) bound $`|\beta _p|,|\stackrel{~}{\beta }_p|p^{\frac{5}{34}}`$. Observe that $`\frac{5}{34}<\frac{1}{6}`$, which has many important consequences (see Section 8 of \[KiSh\]), and is perhaps not coincidentally all we need below. The exponent has been recently improved by Kim and Sarnak to $`\frac{7}{64}`$ (see Appendix Two of \[K\]).
Denote by $`\mathrm{sym}^2f`$ be the Gelbart-Jacquet (symmetric-square) lift to (an automorphic cuspidal representation of) $`\mathrm{GL}(3)`$ of the cusp form $`f`$ \[GeJa\]. Its Fourier coefficients are \[Bu1, Bu2\]
$$a_{\mathrm{sym}^2f}(m_1,m_2)=\underset{d|(m_1,m_2)}{}\lambda _{\mathrm{sym}^2f}(\frac{m_1}{d},1)\lambda _{\mathrm{sym}^2f}(\frac{m_2}{d},1)\mu (d),$$
(3.2)
where $`\mu `$ is the Möbius function and
$$\lambda _{\mathrm{sym}^2f}(r,1)=\underset{s^2t=r}{}\lambda _f(t^2).$$
(3.3)
The symmetric-square $`L`$-function of $`f`$ is then
$$L(s,\mathrm{sym}^2f)=\underset{m=1}{\overset{\mathrm{}}{}}\lambda _{\mathrm{sym}^2f}(m,1)m^s.$$
(3.4)
If, as before, $`\alpha _p,\alpha _p^1`$ are the Satake parameters of $`f`$, then the parameters $`\sigma _p(j)`$ ($`j=1,2,3`$) of $`\mathrm{sym}^2f`$ at any prime $`p`$ are the numbers $`\alpha _p^2,1,\alpha _p^2`$.
Denoting by $`\lambda _\varphi (r)`$ the $`r`$<sup>th</sup> Hecke eigenvalue of $`\varphi `$, the Rankin-Selberg convolution $`L(s,\varphi \times \mathrm{sym}^2f)`$ is the Dirichlet series
$`L(s,\varphi \times \mathrm{sym}^2f)`$ $`={\displaystyle \underset{m_1,m_21}{}}\lambda _\varphi (m_1)\lambda _f(m_2)a_F(m_1,m_2)(m_1m_2^2)^s`$
$`={\displaystyle \underset{m}{}}\lambda _{\varphi ,\mathrm{sym}^2f}(m)m^s,`$ (3.5)
where
$$\lambda _{\varphi \times \mathrm{sym}^2f}(m)=\underset{m_1m_2^2=m}{}\lambda _\varphi (m_1)\lambda _f(m_2)a_F(m_1,m_2).$$
(3.6)
In fact, also by the work of Kim-Shahidi \[KiSh\] (and the appendix by Bushnell-Henniart), $`L(s,\varphi \times \mathrm{sym}^2f)`$ is an automorphic $`L`$-function $`L(s,\pi )`$ (for a suitable an automorphic representation $`\pi `$ of $`\mathrm{GL}(6)`$.) This ensures the standard properties (entire of order one, bounded in vertical strips, and functional equation) for $`L(s,\varphi \times \mathrm{sym}^2f)`$. In particular, $`L(s,\varphi \times \mathrm{sym}^2f)`$ conjecturally satisfies the Riemann Hypothesis in the usual sense: $`L(s,\varphi \times \mathrm{sym}^2f)=0`$ and $`0\mathrm{}s1`$ implies $`\mathrm{}s=\frac{1}{2}`$.
The Satake parameters $`\delta _p(j)`$ ($`j=1,\mathrm{}6`$) of $`\pi _p`$ are the six numbers $`\alpha _p^{\pm 2}\beta _p^{\pm 1}`$ and $`\beta _p^{\pm 1}`$. Furthermore, each $`L(s,\varphi \times \mathrm{sym}^2f)`$ has an even functional equation. The proof of this assertion is given in Appendix A.
For $`\mathrm{}s`$ large, the logarithmic derivative of $`L(s,\varphi \times \mathrm{sym}^2f)`$ is given by the Dirichlet series
$$\frac{L^{}}{L}(s,\varphi \times \mathrm{sym}^2f)=\underset{m=0}{\overset{\mathrm{}}{}}\mathrm{\Lambda }(m)a_{\varphi \times \mathrm{sym}^2f}(m)m^s,$$
(3.7)
where $`\mathrm{\Lambda }(m)`$ is von Mangoldt’s function and
$$a_{\varphi \times \mathrm{sym}^2f}(p^\nu )=\underset{j=1}{\overset{6}{}}\delta _p(j)^\nu .$$
(3.8)
Define now the archimedean (gamma) factor
$$\begin{array}{c}L_{\mathrm{}}(s,\varphi ,\mathrm{sym}^2f):=\mathrm{\Gamma }_{}(s+k1+it_\varphi )\mathrm{\Gamma }_{}(s+k1it_\varphi )\times \hfill \\ \hfill \times \mathrm{\Gamma }_{}(s+k+it_\varphi )\mathrm{\Gamma }_{}(s+kit_\varphi )\mathrm{\Gamma }_{}(s+1+it_\varphi )\mathrm{\Gamma }_{}(s+1it_\varphi ),\end{array}$$
(3.9)
where
$$\mathrm{\Gamma }_{}(s):=\pi ^{\frac{s}{2}}\mathrm{\Gamma }\left(\frac{s}{2}\right).$$
(3.10)
The completed $`L`$-function
$$\mathrm{\Lambda }(s,\varphi \times \mathrm{sym}^2f):=L_{\mathrm{}}(s,\varphi ,\mathrm{sym}^2f)L(s,\varphi \times \mathrm{sym}^2f)$$
(3.11)
for (3.1) satisfies the functional equation
$$\mathrm{\Lambda }(s,\varphi \times \mathrm{sym}^2f)=\mathrm{\Lambda }(1s,\varphi \times \mathrm{sym}^2f).$$
(3.12)
As the functional equation is even, we expect to observe either $`\mathrm{SO}(\mathrm{even})`$ or symplectic symmetry.
Following Rudnick and Sarnak \[RS\], we define the six archimedean parameters $`\mu _j`$ ($`j=1,\mathrm{},6`$) by the requirement that $`\frac{1}{2}+\mu _j`$ is one of
$$k\pm \frac{1}{2}\pm it_\varphi \text{or}\frac{3}{2}\pm it_\varphi .$$
(3.13)
#### 3.1.1. Explicit Formula
A smooth form of the explicit formula for $`L(s,\varphi \times \mathrm{sym}^2f)`$ is as follows (see \[RS\] for a proof). Let $`gC_c^{\mathrm{}}()`$ be an even Schwartz function whose Fourier transform
$$\widehat{g}(y)=_{\mathrm{}}^{\mathrm{}}g(x)e^{2\pi ixy}dx$$
(3.14)
is compactly supported. Let $`R>0`$ and write the non-trivial zeros of $`L(s,\varphi \times \mathrm{sym}^2f)`$ as $`\rho _j=\frac{1}{2}+i\gamma _j`$; we have $`j\{0\}`$ as the functional equation is even. Note $`\gamma _j`$ is equivalent to GRH. Then
$$\underset{j}{}g\left(\frac{\gamma _j}{2\pi }\mathrm{log}R\right)=\frac{A}{\mathrm{log}R}2\underset{p}{}\underset{\nu =1}{\overset{\mathrm{}}{}}\widehat{g}\left(\frac{\nu \mathrm{log}p}{\mathrm{log}R}\right)\frac{a_{\varphi \times \mathrm{sym}^2f}(p^\nu )\mathrm{log}p}{p^{\nu /2}\mathrm{log}R},$$
(3.15)
where
$$A=_{\mathrm{}}^{\mathrm{}}\underset{j=1}{\overset{6}{}}\left(\frac{\mathrm{\Gamma }_{}^{}}{\mathrm{\Gamma }_{}}\left(\mu _j+\frac{1}{2}+\frac{2\pi ix}{\mathrm{log}R}\right)+\frac{\mathrm{\Gamma }_{}^{}}{\mathrm{\Gamma }_{}}\left(\overline{\mu _j}+\frac{1}{2}+\frac{2\pi ix}{\mathrm{log}R}\right)\right)g(x)\mathrm{d}x.$$
(3.16)
#### 3.1.2. Gamma Factor Contribution
Recall $`\mathrm{\Gamma }_{}(s)=\pi ^{s/2}\mathrm{\Gamma }(s/2)`$. Thus
$$\frac{\mathrm{\Gamma }_{}^{}(s)}{\mathrm{\Gamma }_{}(s)}=\frac{\mathrm{log}\pi }{2}+\frac{1}{2}\frac{\mathrm{\Gamma }^{}(\frac{s}{2})}{\mathrm{\Gamma }(\frac{s}{2})}.$$
(3.17)
Let $`r=2\pi x/\mathrm{log}R`$. Then the sum in (3.16) equals
$$6\mathrm{log}\pi +\frac{1}{2}\underset{j=1}{\overset{6}{}}\left[\frac{\mathrm{\Gamma }^{}}{\mathrm{\Gamma }}\left(\frac{1}{4}+\frac{\mu _j}{2}+\frac{ir}{2}\right)+\frac{\mathrm{\Gamma }^{}}{\mathrm{\Gamma }}\left(\frac{1}{4}+\frac{\overline{\mu _j}}{2}\frac{ir}{2}\right)\right],$$
(3.18)
where $`\mu _j=k\pm \frac{1}{2}\pm it_\varphi `$ (for four values) and $`\frac{3}{2}\pm it_\varphi `$ (for the other two). We use (see \[ILS\] or \[GR\] 8.363.3) that for $`a,b`$, $`a>0`$,
$$\frac{\mathrm{\Gamma }^{}}{\mathrm{\Gamma }}\left(a+bi\right)+\frac{\mathrm{\Gamma }^{}}{\mathrm{\Gamma }}\left(abi\right)=2\frac{\mathrm{\Gamma }^{}}{\mathrm{\Gamma }}\left(a\right)+O(a^2b^2),$$
(3.19)
and for $`\alpha \frac{1}{4}`$,
$$\frac{\mathrm{\Gamma }^{}}{\mathrm{\Gamma }}\left(\alpha +\frac{1}{4}\right)=\mathrm{log}\alpha +O(1).$$
(3.20)
Thus, in the $`\mathrm{\Gamma }^{}/\mathrm{\Gamma }`$ factors, the $`\mu _j=\frac{3}{2}\pm it_\varphi `$ terms are $`O(1)`$ with respect to $`k`$. Set $`a_+=\frac{1}{2}`$ and $`a_{}=0`$. Matched in complex-conjugate pairs, the other eight terms give
$`{\displaystyle \frac{\mathrm{\Gamma }^{}}{\mathrm{\Gamma }}}\left({\displaystyle \frac{k}{2}}+a_\pm +i\left(\pm {\displaystyle \frac{t_\varphi }{2}}+r\right)\right)+{\displaystyle \frac{\mathrm{\Gamma }^{}}{\mathrm{\Gamma }}}\left({\displaystyle \frac{k}{2}}+a_\pm i\left(\pm {\displaystyle \frac{t_\varphi }{2}}+r\right)\right)`$
$`=2{\displaystyle \frac{\mathrm{\Gamma }^{}}{\mathrm{\Gamma }}}\left({\displaystyle \frac{k}{2}}+a_\pm \right)+O\left({\displaystyle \frac{|t_\varphi |^2+r^2}{k^2}}\right)`$ (3.21)
and
$`{\displaystyle \frac{\mathrm{\Gamma }^{}}{\mathrm{\Gamma }}}\left({\displaystyle \frac{k}{2}}+a_\pm \right)=\mathrm{log}\left({\displaystyle \frac{k}{2}}+a_\pm {\displaystyle \frac{1}{4}}\right)+O(1)=\mathrm{log}k+O(1).`$ (3.22)
Note for $`k2`$, the condition of having the argument greater than $`\frac{1}{4}`$ is trivially met. The main term in the sum in (3.16) is simply $`\frac{1}{2}42\mathrm{log}k=4\mathrm{log}k`$. The main contribution of the term $`\frac{A}{\mathrm{log}R}`$ in (3.15) is
$$\frac{4\mathrm{log}k}{\mathrm{log}R}_{\mathrm{}}^{\mathrm{}}g(x)𝑑x=\frac{\mathrm{log}k^4}{\mathrm{log}R}\widehat{g}(0).$$
(3.23)
In the $`k`$-aspect, $`\varphi \times \mathrm{sym}^2f`$ looks like a $`\mathrm{GL}(4)`$ object. A natural choice for the analytic conductors is therefore $`k^4`$. With this scaling of the zeros, the test function on the the left-hand side of (3.23) is evaluated at points which have mean average spacing one near the central point (Riemann’s classical critical zero-counting formula). As the quotient depends only on the logarithm of the conductor, as $`k\mathrm{}`$ the choice of any fixed constant multiple of $`k^4`$ for the conductor will give the same answer (see \[ILS\]). We have proved
###### Lemma 3.1.
For $`L(s,\varphi \times \mathrm{sym}^2f)`$, up to lower order terms the contribution from the $`\mathrm{\Gamma }`$-factors in the explicit formula equals $`\widehat{g}(0)`$, and the analytic conductor equals $`k^4`$. Assuming GRH, the non-trivial zeros of $`L(s,\varphi \times \mathrm{sym}^2f)`$ are $`\frac{1}{2}+i\gamma _{\varphi \times \mathrm{sym}^2f}^{(j)}`$ with $`\gamma _{\varphi \times \mathrm{sym}^2f}^{(j)}`$. Taking $`R=k^4`$, the explicit formula (3.15) becomes
$$\begin{array}{c}\underset{j}{}g\left(\frac{\gamma _{\varphi \times \mathrm{sym}^2f}^{(j)}}{2\pi }\mathrm{log}R\right)=\hfill \\ \hfill \widehat{g}\left(0\right)2\underset{p}{}\underset{\nu =1}{\overset{\mathrm{}}{}}\widehat{g}\left(\frac{\nu \mathrm{log}p}{\mathrm{log}R}\right)\frac{a_{\varphi \times \mathrm{sym}^2f}(p^\nu )\mathrm{log}p}{p^{\nu /2}\mathrm{log}R}+O\left(\frac{1}{\mathrm{log}R}\right).\end{array}$$
(3.24)
The appearance of the term $`\widehat{g}\left(0\right)=g(x)dx`$ on the right-hand side of (3.24) naturally corresponds to the (expected) term $`\widehat{g}\left(\xi \right)\delta (\xi )d\xi `$ due to the delta mass at the origin in the Fourier transform of the $`1`$-level density (see (2.21)). The second term (double sum) above will eventually be matched to $`\widehat{g}\left(\xi \right)\frac{1}{2}\eta (\xi )d\xi `$, and this will exclude symplectic as a possibility.
###### Remark 3.2.
It is fortunate for us that the analytic conductors of $`L(s,\varphi \times \mathrm{sym}^2f)`$ depend weakly on $`f`$. Specifically, as the only dependence on $`f`$ is through its weight $`k`$, one scaling works for all elements of our family. Oscillating conductors in a family can sometimes be handled (one recourse is to use the average log conductor as in \[Si, Yo\]; another approach is a more careful analysis and sieving, as in \[Mil2\] where the conductors are monotone).
#### 3.1.3. Relation of $`a_{\varphi \times \mathrm{sym}^2f}`$ to $`\lambda _f`$ and $`\lambda _\varphi `$
To evaluate the double sum in (3.24), we express $`a_{\varphi \times \mathrm{sym}^2f}(p^\nu )`$ in terms of $`\lambda _f,\lambda _\varphi `$. Note
$`a_{\varphi \times \mathrm{sym}^2f}(p^\nu )`$ $`=`$ $`\alpha _p^{2\nu }\beta _p^\nu +\alpha _p^{2\nu }\beta _p^\nu +\alpha _p^{2\nu }\beta _p^\nu +\alpha _p^{2\nu }\beta _p^\nu +\beta _p^\nu +\beta _p^\nu `$ (3.25)
$`=`$ $`(\alpha _p^{2\nu }+1+\alpha _p^{2\nu })(\beta _p^\nu +\beta _p^\nu ).`$
Case $`\nu =1`$: We have
$$\begin{array}{cc}\hfill a_{\varphi \times \mathrm{sym}^2f}(p)& =(\alpha _p^2+1+\alpha _p^2)(\beta _p+\beta _p^1)=\lambda _f(p^2)\lambda _\varphi (p).\hfill \end{array}$$
(3.26)
Case $`\nu 2`$: We have
$$\begin{array}{c}\alpha _p^{2\nu }+1+\alpha _p^{2\nu }=(\alpha _p^{2\nu }+\alpha _p^{2(\nu 1)}+\mathrm{}+\alpha _p^{2(\nu 1)}+\alpha _p^{2\nu })\hfill \\ \hfill (\alpha _p^{2(\nu 1)}+\mathrm{}+\alpha _p^{2(\nu 1)})+1,\end{array}$$
(3.27)
and
$$\begin{array}{c}\beta _p^\nu +\beta _p^\nu =\beta _p^\nu +\beta _p^{\nu 2}+\mathrm{}+\beta _p^{(\nu 2)}+\beta _p^\nu \hfill \\ \hfill (\beta _p^{\nu 2}+\mathrm{}+\beta _p^{(\nu 2)}),\end{array}$$
(3.28)
yielding
$$\begin{array}{cc}\hfill a_{\varphi \times \mathrm{sym}^2f}(p^\nu )& =(\alpha _p^{2\nu }+1+\alpha _p^{2\nu })(\beta _p^\nu +\beta _p^\nu )\hfill \\ & =(\lambda _f(p^{2\nu })\lambda _f(p^{2(\nu 1)})+1)(\lambda _\varphi (p^\nu )\lambda _\varphi (p^{\nu 2})).\hfill \end{array}$$
(3.29)
Of course $`\lambda _\varphi (p^{\nu 2})=1`$ when $`\nu =2`$.
#### 3.1.4. Summary
We have shown
###### Lemma 3.3.
$`a_{\varphi \times \mathrm{sym}^2f}(p)`$ $`=\lambda _\varphi (p)\lambda _f(p^2)`$ (3.30)
$`a_{\varphi \times \mathrm{sym}^2f}(p^2)`$ $`=(\lambda _\varphi (p^2)1)(\lambda _f(p^4)\lambda _f(p^2)+1).`$ (3.31)
As we shall see below, the single term ‘$`1`$’ in the first factor of (3.31) is responsible for flipping the symmetry from symplectic (for the $`\{\mathrm{sym}^2f\}`$ family of \[ILS\] which had $`a_{\mathrm{sym}^2f}=\lambda _f(p^4)\lambda _f(p^2)+1`$) to $`\mathrm{SO}(\mathrm{even})`$ (for the $`\{\varphi \times \mathrm{sym}^2f\}`$ family we are considering). This behavior is described in more detail in \[DM\].
Using the results from Kim-Sarnak \[K\], we have $`|\beta _p^{\pm 1}|p^{\frac{7}{64}}`$. Since $`|\alpha _p|1`$, equation (3.8) yields
$$a_{\varphi \times \mathrm{sym}^2f}(p^\nu )=(\beta _p^\nu +\beta _p^\nu )(\alpha _p^{2\nu }+\alpha _p^{2\nu }+1)p^{\frac{7\nu }{64}}.$$
(3.32)
Therefore
$$\frac{a_{\varphi \times \mathrm{sym}^2f}(p^\nu )}{p^{\nu /2}}p^{\frac{25\nu }{64}}.$$
(3.33)
This immediately implies
###### Lemma 3.4.
The contribution from terms with $`\nu 3`$ in (3.24) can be absorbed into the error term.
###### Remark 3.5.
We do not need the full strength of $`|\beta _p^{\pm 1}|p^{\frac{7}{64}}`$; any exponent less than $`\frac{1}{6}`$ suffices. Without such a bound, we would later need to obtain cancellation when averaging the Fourier coefficients over the family (a result of this nature is significantly weaker than proving bounds towards Ramanujan, and follows from the Petersson formula).
### 3.2. $`1`$-Level Density
As $`_{\varphi \times \mathrm{sym}^2H_k}=\{\varphi \times \mathrm{sym}^2f,fH_k\}`$, we have $`|_{\varphi \times \mathrm{sym}^2H_k}|=|H_k|`$. For each $`L`$-function from $`_{\varphi \times \mathrm{sym}^2H_k}`$ we calculate the $`1`$-level density for its low lying zeros via the explicit formula; we then average over the family $`_{\varphi \times \mathrm{sym}^2H_k}`$.
#### 3.2.1. Preliminaries
Let $`g`$ be an even Schwartz function with $`\mathrm{supp}(\widehat{g})(\sigma ,\sigma )`$. Following Iwaniec-Luo-Sarnak \[ILS\] or Royer \[Ro\], we consider a weighted average over the family $`_{\varphi \times \mathrm{sym}^2H_k}`$ of the expressions (3.24). The weight factors that we use are $`\omega _f=\zeta (2)/L(1,\varphi \times \mathrm{sym}^2f)`$. These are positive (by GRH for $`L(s,\mathrm{sym}^2f)`$), slowly varying, and satisfy
$$k^ϵ_ϵ\frac{1}{L(1,\mathrm{sym}^2f)}_ϵk^ϵ$$
(3.34)
and
$$\frac{1}{|H_k|}\underset{fH_k}{}\frac{\zeta (2)}{L(1,\mathrm{sym}^2f)}=1+O\left(\frac{1}{k}\right).$$
(3.35)
To simplify the application of the Petersson Formula, we have introduced the slowly varying weights $`\zeta (2)/L(1,\mathrm{sym}^2f)`$; arguing along the lines of \[ILS\] allows one to remove these weights at no cost. We have chosen to leave in the weights in order to emphasize the features of this $`\mathrm{GL}(6)`$ family.
By GRH, we may denote the non-trivial zeros of $`L(s,\varphi \times \mathrm{sym}^2f)`$ by $`\frac{1}{2}+i\gamma _{\varphi \times \mathrm{sym}^2f}^{(j)}`$ with $`\gamma _{\varphi \times \mathrm{sym}^2f}^{(j)}`$. Let $`R=k^4`$. All $`L(s,\varphi \times \mathrm{sym}^2f)`$ have the same analytic conductor, which up to lower order terms is $`k^4`$. Averaging (3.24) by incorporating the weights and using Lemma 3.4 to absorb the $`\nu 3`$ terms into the error shows that the $`1`$-level density for the family $`_{\varphi \times \mathrm{sym}^2H_k}`$ is
$`D_{1,_{\varphi \times \mathrm{sym}^2H_k}}(g)`$ (3.36)
$`=`$ $`{\displaystyle \frac{1}{|H_k|}}{\displaystyle \underset{fH_k}{}}{\displaystyle \frac{\zeta (2)}{L(1,\mathrm{sym}^2f)}}{\displaystyle \underset{j}{}}g\left(\gamma _{\varphi \times \mathrm{sym}^2f}^{(j)}{\displaystyle \frac{\mathrm{log}c_{f_\varphi }}{2\pi }}\right)`$
$`=`$ $`\widehat{g}\left(0\right){\displaystyle \frac{2}{|H_k|}}{\displaystyle \underset{fH_k}{}}{\displaystyle \frac{1}{L(1,\mathrm{sym}^2f)}}{\displaystyle \underset{\nu =1}{\overset{2}{}}}{\displaystyle \underset{p=2}{\overset{R^\sigma }{}}}{\displaystyle \frac{a_{\varphi \times \mathrm{sym}^2f}(p^\nu )\mathrm{log}p}{p^{\nu /2}\mathrm{log}R}}\widehat{g}\left(\nu {\displaystyle \frac{\mathrm{log}p}{\mathrm{log}R}}\right)`$
$`+O\left({\displaystyle \frac{1}{\mathrm{log}R}}\right).`$
We are left with analyzing the contribution from the $`\nu =1,2`$ terms, and comparing this to (2.21). For small support, we will show there is no contribution from the $`\nu =1`$ term, and the $`\nu =2`$ term contributes $`\frac{1}{2}g(0)`$. This proves the symmetry group is neither unitary nor symplectic. We cannot discard the $`\mathrm{O}`$ and $`\mathrm{SO}(\mathrm{odd})`$ symmetries; however, we will be able to eliminate them later by studying the $`2`$-level density.
###### Remark 3.6.
Since Iwaniec-Luo-Sarnak exclusively use $`1`$-level density arguments, they must use extra averaging to extend their support past $`[1,1]`$. By studying the $`2`$-level density we provide compelling evidence for the underlying symmetry being $`\mathrm{SO}(\mathrm{even})`$ without extra averaging.
#### 3.2.2. Contribution from $`\nu =1`$
We must evaluate
$$T_1=\frac{1}{|H_k|}\underset{fH_k}{}\frac{\zeta (2)}{L(1,\mathrm{sym}^2f)}\underset{p=2}{\overset{R^\sigma }{}}\frac{a_{\varphi \times \mathrm{sym}^2f}(p)\mathrm{log}p}{\sqrt{p}\mathrm{log}R}\widehat{g}\left(\frac{\mathrm{log}p}{\mathrm{log}R}\right).$$
(3.37)
By Lemma 3.3, $`a_{\varphi \times \mathrm{sym}^2f}(p)=\lambda _\varphi (p)\lambda _f(p^2)`$. Since $`\lambda _f(1)=1`$,
$$T_1=\underset{p=2}{\overset{R^\sigma }{}}\frac{\lambda _\varphi (p)\mathrm{log}p}{\sqrt{p}\mathrm{log}R}\widehat{g}\left(\frac{\mathrm{log}p}{\mathrm{log}R}\right)\frac{\zeta (2)}{|H_k|}\underset{fH_k}{}\frac{\lambda _f(1)\lambda _f(p^2)}{L(1,\mathrm{sym}^2f)}.$$
(3.38)
We are led to studying
$$\frac{\zeta (2)}{|H_k|}\underset{fH_k}{}\frac{\lambda _f(1)\lambda _f(p^2)}{L(1,\mathrm{sym}^2f)}.$$
(3.39)
We have a non-diagonal term since $`p^21`$. By (2.33) of Lemma 2.5, $`\delta (1,p^2)=0`$; for $`pk`$ these terms are $`\frac{p}{2^k}`$. Substituting into the expansion for $`T_1`$, we see there is no contribution for $`R^\sigma <k`$. Since $`R=k^4`$, this implies there is no contribution for $`\sigma <\frac{1}{4}`$. Note that in executing the prime sum, any polynomial bound on $`\lambda _\varphi (p)`$ suffices, since the decay in $`k`$ is exponential.
For primes $`p>R^{\frac{1}{4}}`$, we cannot use (2.33); instead we use (2.32), which gives $`\frac{\sqrt{p}\mathrm{log}p}{k^{5/6}}`$. This yields a $`p`$-sum of
$$\frac{1}{k^{5/6}}\underset{p}{\overset{k^{4\sigma }}{}}\frac{\mathrm{log}^2p}{\mathrm{log}R}\frac{\lambda _\varphi (p)\sqrt{p}}{\sqrt{p}}.$$
(3.40)
Let $`\delta `$ be the the best bound towards Ramanujan for $`\lambda _\varphi (p)`$; namely, $`\lambda _\varphi (p)p^\delta `$ (the Ramanujan conjecture is $`\delta =0`$). We find this sum is $`k^{4\sigma (1+\delta )\frac{5}{6}}`$; thus, $`\sigma <\frac{5}{24(1+\delta )}`$. Even assuming Ramanujan does not help —this bound is worse than the previous one. Thus (2.33) is better, and we obtain that there is no contribution for support up to $`\frac{1}{4}`$.
###### Remark 3.7.
The reason there is no contribution for small support is that we have a non-diagonal term in the the Petersson formula.
#### 3.2.3. $`\nu =2`$
We must evaluate
$$T_2=\frac{2}{|H_k|}\underset{fH_k}{}\frac{\zeta (2)}{L(1,\mathrm{sym}^2f)}\underset{p=2}{\overset{R^\sigma }{}}\frac{a_{\varphi \times \mathrm{sym}^2f}(p^2)\mathrm{log}p}{p\mathrm{log}R}\widehat{g}\left(2\frac{\mathrm{log}p}{\mathrm{log}R}\right).$$
(3.41)
By Lemma 3.3, $`a_{\varphi \times \mathrm{sym}^2f}(p^2)=(\lambda _\varphi (p^2)1)(\lambda _f(p^4)\lambda _f(p^2)+1)`$. Almost all of the terms are non-diagonal. Using $`\lambda _f(1)=1`$ we have the following terms: from $`\lambda _\varphi (p^2)`$, we get
$`\lambda _\varphi (p^2)\lambda _f(1)\lambda _f(p^4),\lambda _\varphi (p^2)\lambda _f(1)\lambda _f(p^2),\lambda _\varphi (p^2)\lambda _f(1)\lambda _f(1).`$
The first two terms are non-diagonal; the Petersson formula yields no contribution for small support. The third term *is* diagonal. For small support, up to lower order terms it yields $`+1`$ by (3.35):
$$\frac{\zeta (2)}{|H_k|}\underset{fH_k}{}\frac{\lambda _f(1)\lambda _f(1)}{L(1,\mathrm{sym}^2f)}=1+O\left(\frac{1}{k}\right).$$
(3.43)
This gives
$$2\underset{p=2}{\overset{R^\sigma }{}}\frac{\lambda _\varphi (p^2)\mathrm{log}p}{p\mathrm{log}R}\widehat{g}\left(2\frac{\mathrm{log}p}{\mathrm{log}R}\right).$$
(3.44)
By GRH for $`L(s,\mathrm{sym}^2\varphi )`$, this sum is $`O(\frac{1}{\mathrm{log}R})`$ (see Section 4 of \[ILS\]).
We now handle the three terms from the $`1`$ in the first factor of $`a_{\varphi \times \mathrm{sym}^2f}(p^2)`$; these are
$$\lambda _f(1)\lambda _f(p^4),\lambda _f(1)\lambda _f(p^2),\lambda _f(1)\lambda _f(1).$$
(3.45)
The first two are non-diagonal, and by the Petersson formula do not contribute for small support. The third term, however, *is* a diagonal term; up to lower order corrections, from the Petersson formula its contribution is $`1`$, and we are left with
$$2\underset{p}{}\frac{\mathrm{log}p}{p\mathrm{log}R}\widehat{g}\left(2\frac{\mathrm{log}p}{\mathrm{log}R}\right).$$
(3.46)
By Lemma 2.7, the above sum (up to lower order terms) is $`\frac{1}{2}g(0)`$.
Therefore, for small support, the $`\nu =2`$ piece contributes $`\frac{1}{2}g(0)+o(1)`$, with the main term arising from the sixth term in the expansion of $`a_{\varphi \times \mathrm{sym}^2f}(p^2)`$. At this point we have enough evidence to discard the unitary and symplectic symmetries. Since the functional equations in (3.12) are even, this certainly points to the underlying symmetry being $`\mathrm{SO}(\mathrm{even})`$, but we cannot yet discard the full orthogonal or $`\mathrm{SO}(\mathrm{odd})`$ symmetries. This will be done in §3.3.
We now determine how large we may take the support. Of the six pieces which do not contribute to the main term, the worst error term is from $`\lambda _\varphi (p^2)\lambda _f(1)\lambda _f(p^4)`$. By (2.33), if $`1p^4k^2`$, the sum over $`fH_k`$ is $`\frac{p^2}{2^k}`$. Again, any polynomial bound for $`\lambda _\varphi (p^2)`$ yields the sum over primes $`pk^{\frac{1}{2}}`$ is a lower order term. Since $`R=k^4`$, this yields no contribution for $`\sigma <\frac{1}{8}`$. Therefore, up to lower order terms the contribution is $`\frac{1}{2}g(0)`$ for $`\sigma <\frac{1}{8}`$.
The reason for the sharp decrease in support (relative to the $`\nu =1`$ term) is because we have a $`\lambda _f(1)\lambda _f(p^4)`$. Another possibility is to use (2.32), which gives for the $`\lambda _\varphi (p^2)\lambda _f(1)\lambda _f(p^4)`$ term
$$\frac{1}{k^{5/6}}\underset{p}{\overset{k^{4\sigma }}{}}\frac{\mathrm{log}^2p}{\mathrm{log}R}\frac{\lambda _\varphi (p^2)p}{p}.$$
(3.47)
From (2.3), $`\lambda _\varphi (p^2)=\lambda _\varphi (p)^21`$. Substituting into (3.47), the -1 does not contribute for $`\sigma <\frac{5}{24}`$, and we are left with bounding
$$\frac{1}{k^{5/6}}\underset{p=2}{\overset{k^{4\sigma }}{}}\frac{\mathrm{log}^2p}{\mathrm{log}R}\lambda _\varphi (p)^2\frac{\mathrm{log}k^4}{k^{5/6}}\underset{n=1}{\overset{k^{4\sigma }}{}}|\lambda _\varphi (n)|^2.$$
(3.48)
One could use bounds towards Ramanujan; however, all we need is that Ramanujan holds on average, namely the sum of $`|\lambda _\varphi (n)|^2`$ (see \[Iw1\], equation 8.7) is
$$\underset{n=1}{\overset{X}{}}|\lambda _\varphi (n)|^2_\varphi X.$$
(3.49)
This yields the sum in (3.48) does not contribute for $`\sigma <\frac{5}{24}`$.
A similar argument applied to the other terms show none of them contribute as well for such support; one must check that the error term for the $`\lambda _f(1)\lambda _f(1)`$ term (which is the diagonal piece responsible for $`\frac{1}{2}g(0)`$) does not contribute in this range. This completes the proof of the first part of Theorem 1.2.
As $`\text{supp}(\widehat{\varphi })(1,1)`$, while every $`\mathrm{\Lambda }(s,\varphi \times \mathrm{sym}^2f)`$ has even functional equation, we cannot conclude the symmetry is $`\mathrm{SO}(\mathrm{even})`$ and not $`\mathrm{O}`$ or $`\mathrm{SO}(\mathrm{odd})`$. There are two natural ways to try and increase the support. The first is to average over even Maass forms $`\varphi `$; unfortunately, we would have to let $`t_j`$ grow to a power of $`k`$, which would change the conductor arguments. Another approach is to average over the weight $`k`$ (as in the investigation of $`\mathrm{sym}^2f`$ in \[ILS\]). By averaging over weight, they triple the support; however, as we start with support less than $`\frac{1}{3}`$, such methods are insufficient to break $`(1,1)`$. We therefore study the 2-level density, which even for arbitrarily small support can distinguish the three orthogonal groups (see \[Mil1, Mil2\]).
### 3.3. $`2`$-Level Density
We complete the proof of Theorem 1.2. As in §3.2, for convenience in applying the Petersson formula we study a weighted 2-level density. Thus (2.3) becomes
$`D_{2,_{\varphi \times \mathrm{sym}^2H_k}}(g)=`$
$`{\displaystyle \frac{1}{|H_k|}}{\displaystyle \underset{fH_k}{}}{\displaystyle \frac{\zeta (2)}{L(1,\mathrm{sym}^2f)}}{\displaystyle \underset{\genfrac{}{}{0pt}{}{j_1,j_2}{j_1\pm j_2}}{}}g_1\left({\displaystyle \frac{\mathrm{log}R}{2\pi }}\gamma _f^{(j_1)}\right)g_2\left({\displaystyle \frac{\mathrm{log}R}{2\pi }}\gamma _f^{(j_2)}\right).`$ (3.50)
The sum is over zeros $`j_1\pm j_2`$. As all functional equations are even and we are assuming GRH, the zeros occur in complex conjugate pairs; this is the first time the sign of the functional equation enters our arguments. We may rewrite the $`2`$-level expression as a sum over all pairs of zeros, minus twice the sum over all zeros, yielding
$`D_{2,_{\varphi \times \mathrm{sym}^2H_k}}(g)=`$
$`{\displaystyle \frac{1}{|H_k|}}{\displaystyle \underset{fH_k}{}}{\displaystyle \frac{\zeta (2)}{L(1,\mathrm{sym}^2f)}}{\displaystyle \underset{j_1,j_2}{}}g_1\left({\displaystyle \frac{\mathrm{log}R}{2\pi }}\gamma _f^{(j_1)}\right)g_2\left({\displaystyle \frac{\mathrm{log}R}{2\pi }}\gamma _f^{(j_2)}\right)`$
$`2D_{1,_{\varphi \times \mathrm{sym}^2H_k}}(g_1g_2).`$ (3.51)
In the above, the second term is a $`1`$-level density with test function $`(g_1g_2)(x)`$. For small support, we have shown this term is just $`\widehat{g_1g_2}(0)+\frac{1}{2}(g_1g_2)(0)`$. Crucial in the above expansion is that each $`\mathrm{\Lambda }(s,\varphi \times \mathrm{sym}^2f)`$ has even sign. This allows us to pair off the zeros, $`\gamma _f^{(j)}`$ and $`\gamma _f^{(j)}`$. Thus summing over distinct zeros is the same as subtracting off twice a 1-level sum over all zeros. This would be false if the functional equation were odd. In that case we would have to add back $`g_1(0)g_2(0)`$ for the extra zero at the central point, and in fact the presence or absence of this additional term is the cause of the differences in the 2-level densities of the three orthogonal groups.
For the first term in (3.3), since we are summing over all zeros, we may use the explicit formula for the sum over each $`j_i`$. Let
$`b_{\varphi \times \mathrm{sym}^2f}(p)`$ $`=`$ $`a_{\varphi \times \mathrm{sym}^2f}(p)`$
$`b_{\varphi \times \mathrm{sym}^2f}(p^2)`$ $`=`$ $`a_{\varphi \times \mathrm{sym}^2f}(p^2)+1.`$ (3.52)
In the expansions with the explicit formula, we isolate the contribution from the part of the $`\nu =2`$ term which contributes $`\frac{1}{2}g_i(0)`$ for small support; this is the $`+1`$ term in $`b_{\varphi \times \mathrm{sym}^2f}(p^2)`$. We have also removed the error terms arising from $`\nu 3`$; the argument is standard (see \[Ru2, RS\]). We are left with considering the weighted average of
$$\underset{i=1}{\overset{2}{}}\left[\left(\widehat{g_i}\left(0\right)+\frac{1}{2}g_i(0)\right)2\underset{\nu _i=1}{\overset{2}{}}\underset{p_i}{}\frac{b_{\varphi \times \mathrm{sym}^2f}(p^{\nu _i})\mathrm{log}p}{p^{\nu _i/2}\mathrm{log}R}\widehat{g_i}\left(\nu _i\frac{\mathrm{log}p_i}{\mathrm{log}R}\right)\right].$$
(3.53)
There are three terms for each $`i`$. For small support, we have shown the $`\nu _i`$-sums by themselves do not contribute, though we will see that there are contributions when a $`\nu _1`$-sum hits a $`\nu _2`$-sum. Thus when a $`\widehat{g_i}\left(0\right)+\frac{1}{2}g_i(0)`$ hits a $`\nu `$-sum, there is no contribution. We have $`[\widehat{g_1}\left(0\right)+\frac{1}{2}g_1(0)][\widehat{g_2}\left(0\right)+\frac{1}{2}g_2(0)]`$, plus the weighted average of the mixed sums
$$4\underset{p_1}{}\underset{p_2}{}\frac{b_{\varphi \times \mathrm{sym}^2f}(p_1^{\nu _1})b_{\varphi \times \mathrm{sym}^2f}(p_2^{\nu _2})\mathrm{log}p_1\mathrm{log}p_2}{p_1^{\nu _1/2}p_2^{\nu _2/2}\mathrm{log}^2R}\widehat{g_1}\left(\nu _1\frac{\mathrm{log}p_1}{\mathrm{log}R}\right)\widehat{g_2}\left(\nu _2\frac{\mathrm{log}p_2}{\mathrm{log}R}\right)$$
(3.54)
for $`(\nu _1,\nu _2)\{(1,1),(2,1),(2,1),(2,2)\}`$. For each pair there are two cases, when $`p_1=p_2`$ and $`p_1p_2`$. Since we are only interested in the $`2`$-level density for arbitrarily small support as a means to distinguish $`\mathrm{SO}(\mathrm{even})`$ from orthogonal and $`\mathrm{SO}(\mathrm{odd})`$ symmetry, we do not record how large we may take the support.
#### 3.3.1. $`(1,1)`$ Terms
We have
$`{\displaystyle \frac{4}{|H_k|}}{\displaystyle \underset{fH_k}{}}{\displaystyle \frac{\zeta (2)}{L(1,\mathrm{sym}^2f)}}{\displaystyle \underset{p_1}{}}{\displaystyle \underset{p_2}{}}{\displaystyle \frac{a_{\varphi \times \mathrm{sym}^2f}(p_1)a_{\varphi \times \mathrm{sym}^2f}(p_2)\mathrm{log}p_1\mathrm{log}p_2}{\sqrt{p_1p_2}\mathrm{log}^2R}}`$
$`\times \widehat{g_1}\left({\displaystyle \frac{\mathrm{log}p_1}{\mathrm{log}R}}\right)\widehat{g_2}\left({\displaystyle \frac{\mathrm{log}p_2}{\mathrm{log}R}}\right).`$ (3.55)
As $`a_{\varphi \times \mathrm{sym}^2f}(p)=\lambda _\varphi (p)\lambda _f(p^2)`$, we have
$$\begin{array}{c}\frac{4}{|H_k|}\underset{fH_k}{}\frac{\zeta (2)}{L(1,\mathrm{sym}^2f)}\underset{p_1}{}\underset{p_2}{}\lambda _\varphi (p_1)\lambda _\varphi (p_2)\frac{\lambda _f(p_1^2)\lambda _f(p_2^2)\mathrm{log}p_1\mathrm{log}p_2}{\sqrt{p_1p_2}\mathrm{log}^2R}\hfill \\ \hfill \times \widehat{g_1}\left(\frac{\mathrm{log}p_1}{\mathrm{log}R}\right)\widehat{g_2}\left(\frac{\mathrm{log}p_2}{\mathrm{log}R}\right).\end{array}$$
(3.56)
If $`p_1p_2`$, when we use the Petersson formula there is no contribution for small support, since it is a non-diagonal term. Note $`p_1=p_2`$ is a diagonal term, giving $`\frac{1}{|H_k|}_f\frac{\zeta (2)}{L(1,\mathrm{sym}^2f)}\lambda _f(p^2)\lambda _f(p^2)=1+O(\sqrt{p^4}/2^k)`$. For sufficiently small support the error term is negligible, and thus the diagonal term is
$$4\underset{p}{}\lambda _\varphi (p)\lambda _\varphi (p)\frac{\mathrm{log}^2p}{p\mathrm{log}^2R}\widehat{g_1}\left(\frac{\mathrm{log}p_1}{\mathrm{log}R}\right)\widehat{g_2}\left(\frac{\mathrm{log}p_2}{\mathrm{log}R}\right)+o(1).$$
(3.57)
We use $`\lambda _\varphi (p)\lambda _\varphi (p)=1+\lambda _\varphi (p^2)`$; we saw in §3.2.3 that the $`\lambda _\varphi (p^2)`$ term will not contribute by GRH for $`L(s,\mathrm{sym}^2\varphi )`$. The $`+1`$ will contribute, with test function $`\widehat{g_1}\widehat{g_2}`$. By Lemma 2.7, we have
$$4\underset{p}{}\frac{\mathrm{log}^2p}{p\mathrm{log}^2R}\widehat{g_1}\widehat{g_2}\left(\frac{\mathrm{log}p}{\mathrm{log}R}\right)=2|u|\widehat{g_1}\widehat{g_2}(u)du+O\left(\frac{1}{\mathrm{log}R}\right).$$
(3.58)
#### 3.3.2. $`(1,2)`$ and $`(2,1)`$ Terms
As these terms are handled identically, we confine ourselves to $`(1,2)`$. We have weighted averages of
$$4\underset{p_1}{}\underset{p_2}{}\frac{a_{\varphi \times \mathrm{sym}^2f}(p_1)b_{\varphi \times \mathrm{sym}^2f}(p_2^2)\mathrm{log}p_1\mathrm{log}p_2}{p_1^{1/2}p_2\mathrm{log}^2R}\widehat{g_1}\left(\frac{\mathrm{log}p_1}{\mathrm{log}R}\right)\widehat{g_2}\left(2\frac{\mathrm{log}p_2}{\mathrm{log}R}\right).$$
(3.59)
By Lemma 3.3 and (3.3)
$`a_{\varphi \times \mathrm{sym}^2f}(p)`$ $`=\lambda _\varphi (p)\lambda _f(p^2)`$ (3.60)
$`b_{\varphi \times \mathrm{sym}^2f}(p^2)`$ $`=\lambda _\varphi (p^2)(\lambda _f(p^4)\lambda _f(p^2)+1)(\lambda _f(p^4)\lambda _f(p^2)).`$ (3.61)
If $`p_1p_2`$, all terms are non-diagonal, and the Petersson formula yields no contribution for small support.
If $`p_1=p_2`$, only two terms give diagonal terms: $`\lambda _\varphi (p)\lambda _\varphi (p^2)\lambda _f(p^2)\lambda _f(p^2)`$ and $`\lambda _\varphi (p)\lambda _f(p^2)\lambda _f(p^2)`$. However, while the Petersson formula will give a $`\pm 1`$ for each of these diagonal terms, this is immaterial since we are dividing by $`p^{\frac{3}{2}}`$. Using the Kim-Sarnak bound of $`p^{\frac{7}{64}}`$ for Maass forms is enough to show there is no contribution, for any support, from these terms. We are dividing by $`p^{\frac{3}{2}}`$, and we have at most $`p^{\frac{73}{64}}\mathrm{log}p`$ in the numerator. This gives $`p^{1\frac{11}{64}}\mathrm{log}p`$, which yields a $`O(\frac{1}{\mathrm{log}R})`$ contribution. Arguing as in (3.47) to (3.49), one may replace the Kim-Sarnak bound with any non-trivial bound towards Ramanujan.
#### 3.3.3. $`(2,2)`$ Term
We now consider the $`(2,2)`$ term. We have a weighted average of
$$4\underset{p_1}{}\underset{p_2}{}\frac{b_{\varphi \times \mathrm{sym}^2f}(p_1^2)b_{\varphi \times \mathrm{sym}^2f}(p_2^2)\mathrm{log}p_1\mathrm{log}p_2}{p_1p_2\mathrm{log}^2R}\widehat{g_1}\left(2\frac{\mathrm{log}p_1}{\mathrm{log}R}\right)\widehat{g_2}\left(2\frac{\mathrm{log}p_2}{\mathrm{log}R}\right),$$
(3.62)
where by Lemma 3.3 and (3.3)
$`b_{\varphi \times \mathrm{sym}^2f}(p^2)`$ $`=`$ $`\lambda _\varphi (p^2)(\lambda _f(p^4)\lambda _f(p^2)+1)`$ (3.63)
$`1(\lambda _f(p^4)\lambda _f(p^2)).`$
When $`p_1=p_2`$, as $`\lambda _\varphi (p)=O(p^\delta )`$ for some $`\delta [0,\frac{1}{2}]`$, $`\frac{b_{\varphi \times \mathrm{sym}^2f}(p^2)^2}{p^2}=O(p^{4\delta 2})`$, and these terms will not contribute for any $`\delta <\frac{1}{4}`$. We do not need the full strength of the Kim-Sarnak bound; the $`\frac{5}{28}`$ of \[BDHI\] suffices.
We are left with the case $`p_1p_2`$. The only diagonal term will be $`\lambda _\varphi (p_1^2)\lambda _\varphi (p_2^2)\lambda _f(1)\lambda _f(1)`$. For small support, the other terms will not contribute, and by Petersson’s formula we have
$$4\underset{i=1}{\overset{2}{}}\underset{p_i}{}\frac{\lambda _\varphi (p_i^2)\mathrm{log}p_i}{p_i\mathrm{log}R}\widehat{g_i}\left(2\frac{\mathrm{log}p_i}{\mathrm{log}R}\right).$$
(3.64)
As in §3.2.3, by GRH for $`L(s,\mathrm{sym}^2\varphi )`$ each prime sum is $`O(\frac{1}{\mathrm{log}R})`$. Thus there is no contribution from the $`(2,2)`$ terms.
#### 3.3.4. Summary
We have shown
$$\begin{array}{c}D_{2,_{\varphi \times \mathrm{sym}^2H_k}}(g)=\left[\widehat{g_1}\left(0\right)+\frac{1}{2}g_1(0)\right]\left[\widehat{g_2}\left(0\right)+\frac{1}{2}g_2(0)\right]\hfill \\ \hfill +2_u|u|\widehat{g_1}\widehat{g_2}(u)du2\widehat{g_1g_2}(0)g_1(0)g_2(0),\end{array}$$
(3.65)
and (3.65) agrees only with the $`2`$-level density for $`\mathrm{SO}(\mathrm{even})`$ (see (2.23)), completing the proof of Theorem 1.2.
## 4. $`_{\varphi \times H_k}=\{\varphi \times f:fH_k\}`$
In this section we prove Theorem 1.1, namely that the symmetry type of the family $`_{\varphi \times H_k}=\{\varphi \times f:fH_k\}`$ ($`k\mathrm{}`$) agrees only with symplectic, where $`\varphi `$ is a fixed even cuspidal Hecke-Maass form with eigenvalue $`\lambda _\varphi =\frac{1}{4}+t_\varphi ^2`$ and $`fH_k`$ is a Hecke holomorphic modular form of weight $`k`$. These $`L`$-functions have associated Euler products of degree 4 and are indeed associated to automorphic representations of $`\mathrm{GL}(4)`$ \[Ra\]. As in §3.1 we first derive the explicit formula and find the analytic conductors for the family. In order to compute the $`1`$-level density we analyze the local parameters and find evidence for symplectic symmetry (for small support). Since the symplectic $`1`$-level density is distinguishable from the other classical compact groups for arbitrarily small support, there is no need to investigate the $`2`$-level density. As the arguments are similar to those for $`\varphi \times \mathrm{sym}^2f`$, we merely sketch the calculations below. See also Appendix A for details of the determination of the gamma factors and signs of the functional equation.
### 4.1. Logarithmic Derivative, Gamma Factors, Functional Equation
In terms of the Fourier coefficients $`\{\lambda _f(n)\},\{\lambda _\varphi (n)\}`$ and the Maass eigenvalue $`t_\varphi `$ ($`\frac{1}{4}+t_\varphi ^2`$ is the Laplacian eigenvalue), we have
$$\begin{array}{cc}\hfill L(s,\varphi \times f)& =\zeta (2s)\underset{m}{}\lambda _\varphi (m)\lambda _f(m)m^s=\underset{m}{}\lambda _{\varphi \times f}(m)m^s,\hfill \end{array}$$
(4.1)
where
$$\lambda _{\varphi \times f}(m)=\underset{m_1^2m_2=m}{}\lambda _\varphi (m_2)\lambda _f(m_2).$$
(4.2)
The logarithmic derivative of $`L(s,\varphi \times f)`$ is
$$\frac{L^{}}{L}(s,\varphi \times f)=\underset{m}{}\mathrm{\Lambda }(m)a_{\varphi \times f}(m)m^s,$$
(4.3)
with
$$a_{\varphi \times f}(p^\nu )=\underset{j=1}{\overset{4}{}}\tau _p(j)^\nu .$$
(4.4)
The archimedean (gamma) factor is
$`L_{\mathrm{}}(s,\varphi \times f)`$ $`:=`$ $`\mathrm{\Gamma }_{}(s+it_\varphi +\frac{k1}{2})\mathrm{\Gamma }_{}(sit_\varphi +\frac{k1}{2})`$ (4.5)
$`\times \mathrm{\Gamma }_{}(s+it_\varphi +\frac{k+1}{2})\mathrm{\Gamma }_{}(sit_\varphi +\frac{k+1}{2}),`$
and the completed $`L`$-function
$$\mathrm{\Lambda }(s,\varphi \times f):=L_{\mathrm{}}(s,\varphi \times f)L(s,\varphi \times f)$$
(4.6)
satisfies the functional equation
$$\mathrm{\Lambda }(s,\varphi \times f)=\mathrm{\Lambda }(1s,\varphi \times f).$$
(4.7)
Note the functional equation is even. We define the archimedean parameters $`\mu _j`$, $`1j4`$, to be the numbers
$$\frac{k\pm 1}{2}\pm it_\varphi .$$
(4.8)
### 4.2. Explicit Formula
As in §3.1.1, let $`R>0`$ be a parameter; later we take $`R=k^4`$. Assuming GRH we may write the non-trivial zeros of $`L(s,\varphi \times f)`$ as $`\rho _j=\frac{1}{2}+i\gamma _j`$, $`j\{0\}`$ (since all signs are even). Then
$$\underset{j}{}g\left(\frac{\gamma _j}{2\pi }\mathrm{log}R\right)=\frac{A}{\mathrm{log}R}2\underset{p}{}\underset{\nu =1}{\overset{\mathrm{}}{}}\widehat{g}\left(\frac{\nu \mathrm{log}p}{\mathrm{log}R}\right)\frac{a_{\varphi \times f}(p^\nu )\mathrm{log}p}{p^{\nu /2}\mathrm{log}R},$$
(4.9)
where
$$A=_{\mathrm{}}^{\mathrm{}}\underset{j=1}{\overset{4}{}}\left(\frac{\mathrm{\Gamma }_{}^{}}{\mathrm{\Gamma }_{}}\left(\mu _j+\frac{1}{2}+\frac{2\pi ix}{\mathrm{log}R}\right)+\frac{\mathrm{\Gamma }_{}^{}}{\mathrm{\Gamma }_{}}\left(\overline{\mu _j}+\frac{1}{2}+\frac{2\pi ix}{\mathrm{log}R}\right)\right)g(x)\mathrm{d}x.$$
(4.10)
An analogous calculation as in §3.1.2 gives, up to lower order terms, that the conductor is $`\left(k^2/4\right)^2`$. As we only care about the logarithm of the conductor, we take $`R=k^4`$. The contribution to the $`1`$-level density will be $`\widehat{g}(0)`$ plus lower order terms.
### 4.3. Relation of $`a_{\varphi \times f}`$ to $`\lambda _\varphi `$ and $`\lambda _f`$.
We consider the local parameters $`\alpha _p^{\pm 1}`$ at any prime $`p`$ for $`f`$, as well as $`\beta _p^{\pm 1}`$ for $`\varphi `$. The local parameters $`\tau _p(j)`$ ($`j=1,2,3,4`$) for the automorphic representation associated to $`L(s,\varphi \times f)`$ are the four numbers $`\alpha _p^{\pm 1}\beta _p^{\pm 1}`$. A calculation similar to but simpler than that in §3.1.3 gives
$`a_{\varphi \times f}(p)`$ $`=\lambda _\varphi (p)\lambda _f(p)`$ (4.11)
$`a_{\varphi \times f}(p^\nu )`$ $`=(\lambda _\varphi (p^\nu )\lambda _\varphi (p^{\nu 2})(\lambda _f(p^\nu )\lambda _f(p^{\nu 2})),\nu 2.`$ (4.12)
In particular
$$a_{\varphi \times f}(p^2)=(\lambda _\varphi (p^2)1)(\lambda _f(p^2)1).$$
(4.13)
#### 4.3.1. $`\nu 3`$ Terms
We show there is no contribution to the $`1`$-level density from terms with $`\nu 3`$ in (4.9). The Satake parameters are the four numbers $`\alpha _p^{\pm 1}\beta _p^{\pm 1}`$, each of which is bounded by $`p^\delta `$ with $`\delta <\frac{1}{6}`$ by Kim-Sarnak \[K\]. Thus
$$\frac{a_{\varphi ,f}(p^\nu )}{p^{\frac{\nu }{2}}}p^{\nu (\delta \frac{1}{2})},$$
(4.14)
and as $`\delta <\frac{1}{6}`$, summing over $`p`$ and $`\nu 3`$ is $`O(1)`$. Dividing by $`\mathrm{log}R=\mathrm{log}k^4`$, we see there is no contribution.
#### 4.3.2. $`\nu =1`$ Terms
As $`a_{\varphi \times f}=\lambda _\varphi (p)\lambda _f(p)`$ and $`\mathrm{log}R=\mathrm{log}k^4`$, we have
$`2{\displaystyle \underset{p}{}}\widehat{g}\left({\displaystyle \frac{\mathrm{log}p}{\mathrm{log}R}}\right){\displaystyle \frac{\lambda _\varphi (p)\lambda _f(p)\mathrm{log}p}{\sqrt{p}\mathrm{log}R}}.`$ (4.15)
As $`1=\lambda _f(1)`$, summing over $`fH_k`$ yields no contribution for small support, as $`\lambda _f(p)\lambda _f(1)`$ is a non-diagonal term.
#### 4.3.3. $`\nu =2`$
Terms
As $`a_{\varphi \times f}(p^2)=(\lambda _\varphi (p^2)1)(\lambda _f(p^2)1)`$, we have
$$2\underset{p}{}\widehat{g}\left(\frac{2\mathrm{log}p}{\mathrm{log}R}\right)\frac{(\lambda _\varphi (p^2)1)(\lambda _f(p^2)1)\mathrm{log}p}{p\mathrm{log}R}.$$
(4.16)
There are four types of terms:
$$\lambda _\varphi (p^2)\lambda _f(p^2)\lambda _f(1),\lambda _\varphi (p^2),\lambda _f(p^2)\lambda _f(1),(1)(1).$$
(4.17)
The first and third are non-diagonal, and by the Petersson formula will not contribute for small support. The second is diagonal; however, by GRH for $`L(s,\mathrm{sym}^2\varphi )`$ (see (3.44)), this term is $`O(\frac{1}{\mathrm{log}R})`$. We are left with the fourth piece,
$$(1)^22\underset{p}{}\widehat{g}\left(\frac{2\mathrm{log}p}{\mathrm{log}R}\right)\frac{\mathrm{log}p}{p\mathrm{log}R}.$$
(4.18)
By Lemma 2.7, the $`p`$-sum is $`\frac{g(0)}{4}`$, thus, the $`\nu =2`$ terms contribute, for small support, $`\frac{1}{2}g(0)`$.
#### 4.3.4. Summary
The previous subsections proved Theorem 1.1, that as $`k\mathrm{}`$ the $`1`$-level density of $`_{\varphi \times H_k}`$ agrees only with symplectic. We took two orthogonal families (when $`k2mod4`$ then $`H_k`$ has $`\mathrm{SO}(\mathrm{odd})`$ symmetry, and when $`k0mod4`$ then $`H_k`$ has $`\mathrm{SO}(\mathrm{even})`$ symmetry), and showed that their twists by a fixed even, full level Maass form give a symplectic family. This should be compared to Theorem 1.2, where we twisted a symplectic family and obtained an $`\mathrm{SO}(\mathrm{even})`$ family.
###### Remark 4.1.
The reason for the symmetry flipping can be found in (4.13). For the $`\nu =2`$ terms, the Maass form introduces an extra factor of $`1`$ in the diagonal contribution. This changes the sign of the contribution from the $`\nu =2`$ terms, and switches us from symplectic to orthogonal symmetries (if we have orthogonal symmetries, we need to evaluate the 2-level density to determine which one as our supports are too small to distinguish $`\mathrm{SO}(\mathrm{even})`$, $`\mathrm{O}`$ and $`\mathrm{SO}(\mathrm{odd})`$).
## 5. Conclusion
We investigated the distribution of low lying zeros for two families. The first is a $`\mathrm{GL}(6)`$ family, $`\{\varphi \times \mathrm{sym}^2f:fH_k\}`$; here $`\varphi `$ is a fixed Hecke-Maass cusp form and $`k\mathrm{}`$. Though this is a $`\mathrm{GL}(6)`$ family, only four of the six Gamma factors depend on $`k`$, and the analytic conductor is $`k^4`$. Since all elements of this family have even functional equation and there is no natural complementary family with odd sign, a folklore conjecture predicted that the underlying group symmetry should be symplectic. However, the symmetry type is $`\mathrm{SO}(\mathrm{even})`$, proving that low lying zeros is more than just a theory of signs of functional equations.
We calculated the $`1`$-level density for test functions $`g`$ such that $`\mathrm{supp}(\widehat{g})`$ is small ($`\mathrm{supp}(\widehat{g})(\frac{5}{24},\frac{5}{24}`$)). For such small support, only the diagonal terms in the Petersson formulas contribute. Thus we can eliminate two of the five classical compact groups, namely symplectic and unitary. Unfortunately, since the support is significantly less than $`(1,1)`$, all three orthogonal candidates are still possible; however, as all members of the family have even functional equation, we do not expect the underlying symmetry to be either $`\mathrm{O}`$ or $`\mathrm{SO}(\mathrm{odd})`$. Observe that the reason for the flipping of symmetry from symplectic to (some flavor of) orthogonal is that the contribution from the squares of primes (which is what is responsible in any case for the term $`\pm \frac{1}{2}g(0)`$) changes sign.
More precisely, in \[ILS\] for the family $`\{\mathrm{sym}^2f\}`$, there is a contribution of $`\frac{1}{2}\widehat{g}\left(0\right)`$, arising from the diagonal term $`+1`$ in $`\lambda _f(p^4)\lambda _f(p^2)+1`$. In our family, this term is multiplied by the factor $`\lambda _\varphi (p^2)1`$ (see (3.29)), and the $`1`$ results in a diagonal contribution of opposite sign, hence the symmetry flipping.
To discard the $`\mathrm{O}`$ and $`\mathrm{SO}(\mathrm{odd})`$ symmetries, we calculated the $`2`$-level density. It is shown in \[Mil1, Mil2\] that for arbitrarily small support the $`2`$-level densities of the three orthogonal groups are distinguishable. We see that our answer agrees only with $`\mathrm{SO}(\mathrm{even})`$, further supporting the claim that the symmetry group of this family is $`\mathrm{SO}(\mathrm{even})`$.
Our second example is a $`\mathrm{GL}(4)`$ family, $`\{\varphi \times f:fH_k\}`$. Here, for the same reason as before, twisting flips the symmetry (this time from orthogonal to symplectic). In a subsequent paper, we will describe the interplay between twisting by a fixed $`\mathrm{GL}(n)`$ form (or family) and the symmetry type (in certain cases). The arguments of this paper can be generalized to families satisfying certain natural technical conditions. It can be shown that a natural “family constant” $`c`$ can be attached to a family in such a manner that $`c_{\times 𝒢}=c_{}c_𝒢`$, where $`c=0`$ $`(1,1)`$ for unitary (symplectic, orthogonal) symmetry. Here $`\times 𝒢`$ is the family obtained by Rankin-Selberg convolution of the $`L`$-functions in the families $``$ and $`𝒢`$. These results are similar in spirit to the universality found by Rudnick and Sarnak \[RS\] in the $`n`$-level correlations of high zeros, and will be described in further detail in \[DM\]. Specifically, it again seems that the second moment of the Satake parameters determines the answer. For the $`\mathrm{GL}(6)`$ family, the main term from averaging the second moment of the Satake parameters (see (3.8)) is $`1`$ and leads to $`\mathrm{SO}(\mathrm{even})`$ symmetry, while in the $`\mathrm{GL}(4)`$ family the main term (see (4.13)) is $`+1`$ and leads to symplectic symmetry.
## Appendix A Gamma factors and signs of functional equations
In this appendix we derive the precise forms (equations (3.9) and (4.5)) of the gamma factors for the completed $`L`$-functions $`L(s,\varphi \times \mathrm{sym}^2f)`$ and $`L(s,\varphi \times f)`$, as well as their functional equations (equations (3.12) and (4.7)). In particular, we show that both functional equations are even.
Being Hecke eigenforms of level 1, $`f`$ and $`\varphi `$ can be identified with automorphic cuspidal representations $`F`$ and $`\mathrm{\Phi }`$ of $`\mathrm{GL}_2(𝔸_{})`$ with trivial central character \[Gel\]. The latter are isomorphic to restricted tensor products
$$F\underset{v}{^{}}F_v\mathrm{\Phi }\underset{v}{^{}}\mathrm{\Phi }_v$$
(A.1)
of representations of $`GL_2(_v)`$ for each place $`v`$ of $``$ (here $`v=p`$ for $`p`$ prime, or $`v=\mathrm{}`$, in which case $`_{\mathrm{}}=`$). In the case at hand, as every (finite) prime $`p`$ is unramified, the corresponding principal series representations of $`\mathrm{GL}_2(_p)`$ have associated $`\mathrm{SL}_2()`$-conjugacy classes<sup>1</sup><sup>1</sup>1$`A^{\mathrm{}}`$ denotes the conjugacy class of $`A`$.
$$F_p\left(\begin{array}{cc}\alpha _p& 0\\ 0& \alpha _p^1\end{array}\right)^{\mathrm{}},\mathrm{\Phi }_p\left(\begin{array}{cc}\beta _p& 0\\ 0& \beta _p^1\end{array}\right)^{\mathrm{}}.$$
(A.2)
Denote by $`M(\alpha _p),M(\beta _p)`$ the matrices in (A.2).
Taking the symmetric square of the standard representation of $`\mathrm{GL}_2()`$, one obtains the conjugacy class $`\mathrm{sym}^2M(\alpha _p)^{\mathrm{}}=\mathrm{diag}(\alpha _p^2,1,\alpha _p^2)^{\mathrm{}}`$, whence the Satake parameters of $`\mathrm{sym}^2f`$. Similarly, the conjugacy classes of $`M(\alpha _p)M(\beta _p)`$ in $`\mathrm{GL}_4()`$ and $`\mathrm{sym}^2M(\alpha _p)M(\beta _p)`$ in $`\mathrm{GL}_6()`$ define the Satake parameters of $`\varphi \times f`$ and $`\varphi \times \mathrm{sym}^2f`$. The local $`L`$-factors are defined in terms of these Satake parameters in the usual manner, and the product of all local factors defines the (incomplete) $`L`$-functions (3.1) and (4.1).
The representations $`F_{\mathrm{}}`$ and $`\mathrm{\Phi }_{\mathrm{}}`$ are the discrete series representation of weight $`k`$, and the representation $`I(||^{it},||^{it})`$ of $`\mathrm{GL}_2()`$, respectively (recall that $`\frac{1}{4}+t^2`$ is the Laplacian eigenvalue of $`\varphi `$.)<sup>2</sup><sup>2</sup>2$`I(||^{it},||^{it})`$ is the unitary induction from the group $`Q`$ of upper-triangular matrices to $`\mathrm{GL}_2()`$ of the representation $`\left(\begin{array}{cc}a& \\ & d\end{array}\right)|a|^{it}|d|^{it}`$. Selberg proved that $`t`$ is real for Maass forms of level $`1`$ (and conjectured that $`t`$ is still real for forms of any weight). The first published proof is due to Roelcke \[Roe\]; see also \[Iw1\].
Proofs of equations (3.9) and (4.5) involve parametrizing the representations $`F_{\mathrm{}}`$ and $`\mathrm{\Phi }_{\mathrm{}}`$, through the Langlands correspondence, by semisimple representations of the Weil group $`W_{}`$ (see \[Kn\] and \[CM\]). We number the discrete series as in \[Kn\], (see the note at the top of page 1588 of \[CM\]), so replacing $`k`$ by $`\mathrm{}+1`$ in what follows would make our notation agree with that of \[CM\]. Then (cf., equations (3.2) and (3.3) of \[Kn\])
$`F_{\mathrm{}}`$ $`\rho _{(k1,0)},`$ (A.3)
$`\mathrm{\Phi }_{\mathrm{}}`$ $`\rho _{(+,it)}\rho _{(+,it)}.`$ (A.4)
We have denoted by $`\rho _{(a,b)}`$ the semisimple representation of $`W_{}`$ with parameters $`(a,b)`$. The known cases of functoriality \[GeJa, Ra, KiSh\] imply the existence of automorphic (cuspidal) representations $`\mathrm{sym}^2F`$, $`\mathrm{\Phi }\times \mathrm{sym}^2F`$, and $`\mathrm{\Phi }\times F`$ such that, by the archimedean Langlands correspondence,
$`(\mathrm{sym}^2F)_{\mathrm{}}`$ $`\mathrm{sym}^2\rho _{(k1,0)}`$ (A.5)
$`(\mathrm{\Phi }\times \mathrm{sym}^2F)_{\mathrm{}}`$ $`\left(\rho _{(+,it)}\rho _{(+,it)}\right)\mathrm{sym}^2\rho _{(k1,0)}`$ (A.6)
$`(\mathrm{\Phi }\times F)_{\mathrm{}}`$ $`\left(\rho _{(+,it)}\rho _{(+,it)}\right)\rho _{(k1,0)}.`$ (A.7)
By Proposition 3.1 of \[CM\],<sup>3</sup><sup>3</sup>3Note that the weight $`k`$ of our $`f`$ is always even. Thus, the appearance of $`(,0)`$ in (A.8) is due to the fact that $`(1)^{k1}=1`$.
$$\mathrm{sym}^2\rho _{(k1,0)}\rho _{(,0)}\rho _{(2k,0)}.$$
(A.8)
Moreover, it is easily checked that
$`\rho _{(+,\pm it)}`$ $`\rho _{(,0)}\rho _{(,\pm it)}`$ (A.9)
$`\rho _{(+,\pm it)}`$ $`\rho _{(\mathrm{},0)}\rho _{(\mathrm{},\pm it)}.`$ (A.10)
Hence,
$`(\mathrm{\Phi }\times \mathrm{sym}^2F)_{\mathrm{}}`$ $`\rho _{(,it)}\rho _{(,it)}\rho _{(2k2,it)}\rho _{(2k2,it)}`$ (A.11)
$`(\mathrm{\Phi }\times F)_{\mathrm{}}`$ $`\rho _{(k1,it)}\rho _{(k1,it)}.`$ (A.12)
The archimedean (gamma) factors can be found using these decompositions and the local Langlands correspondence. In terms of irreducible semisimple representations of $`W_{}`$, we have
$$L(s,\rho )=\{\begin{array}{cc}\mathrm{\Gamma }_{}(s\pm it)\hfill & \rho =\rho _{(+,\pm it)},\hfill \\ \mathrm{\Gamma }_{}(s\pm it+1)\hfill & \rho =\rho _{(,\pm it)},\hfill \\ \mathrm{\Gamma }_{}(s\pm it+\frac{\mathrm{}}{2})\mathrm{\Gamma }_{}(s\pm it+\frac{\mathrm{}}{2}+1)\hfill & \rho =\rho _{(\mathrm{},\pm it)}.\hfill \end{array}$$
(A.13)
These local factors are multiplicative under direct sums of representations of $`W_{}`$, so definitions (3.9) and (4.5) are consistent with (A.11) and (A.12) via (A.13).
Since all automorphic representations under discussion are self-contragredient, the functional equations relate each $`L`$-function to itself as $`s1s`$. In general, the root number $`\epsilon (s,\pi )`$ associated to an automorphic cuspidal representation $`\pi `$ is a product
$$\epsilon (s,\pi )=\underset{v}{}\epsilon (s,\pi _v)$$
of local root numbers.<sup>4</sup><sup>4</sup>4We have omitted the dependence of the local root numbers on the choice of additive character $`\psi `$ of $`𝔸_{}`$. For self-contragredient representations $`\pi =\stackrel{~}{\pi }`$ the $`\epsilon `$-factor agrees with the sign of the functional equation (up to a factor $`Q^s`$ which is not present in the level-$`1`$ case that concerns us). Moreover, local root numbers can be computed via the local Langlands correspondence as root numbers associated to Weil group representations. Additionally, $`\epsilon (s,\pi _p)=1`$ at any prime $`p`$ such that $`\pi _p`$ is unramified.<sup>5</sup><sup>5</sup>5 We assume $`\psi _p`$ is unramified at all primes $`p`$. The local root number of an irreducible representation $`\rho `$ of $`W_{}`$ is
$`\epsilon (s,\rho )=\{\begin{array}{cc}1\hfill & \rho =\rho _{(+,\pm it)},\hfill \\ i\hfill & \rho =\rho _{(,\pm it)},\hfill \\ i^{\mathrm{}+1}\hfill & \rho =\rho _{(\mathrm{},\pm it)}.\hfill \end{array}`$ (A.14)
From (A.14) and (A.11), (A.12) we obtain
$`\epsilon ((\mathrm{\Phi }\times \mathrm{sym}^2F)_{\mathrm{}})`$ $`=iii^{2k1}i^{2k1}=+1,`$ (A.15)
and, since $`k`$ is even,
$`\epsilon ((\mathrm{\Phi }\times F)_{\mathrm{}})`$ $`=i^ki^k=+1.`$ (A.16)
Since all (finite) primes $`p`$ are unramified for $`\mathrm{\Phi }\times F`$ and $`\mathrm{\Phi }\times \mathrm{sym}^2F`$, we conclude
$$\epsilon (s,\mathrm{\Phi }\times \mathrm{sym}^2F)=\epsilon (s,\mathrm{\Phi }\times F)=+1,$$
(A.17)
so the global functional equations have even sign, proving (3.12) and (4.7).
|
warning/0506/quant-ph0506109.html
|
ar5iv
|
text
|
# Hyperspheres and control of spin chains
## 1 Introduction
It is convenient to use sphere to represent state of one spin-$`1/2`$ system. It is so called Bloch or Poincaré sphere. The state of space of $`n`$ spin-$`1/2`$ systems is complex Hilbert space with dimension $`2^n`$, but due to normalization condition there are $`2^n1`$ complex or $`2(2^n1)`$ real parameters. So for $`n=1`$ there are $`2=2(2^11)`$ real parameters, e.g. two Euler angles describing point on surface of sphere. Any unitary transformation of spin-$`1/2`$ system corresponds to rotation of Bloch sphere in agreement with $`21`$ isomorphism of groups SU$`(2)`$ and SO$`(3)`$.
For $`n>1`$ there is no such convenient visualization of space of all states with higher dimensional spheres<sup>1</sup><sup>1</sup>1But see Note on page 2., but exist some interesting subspaces with such property. The subspaces are important for theory of quantum computations and control, because for some physical systems they may correspond to simpler accessible set of physical states.
## 2 Rotations and Spin groups
Let us consider for chain with $`n`$ qubits set with $`2n`$ Hermitian matrices
$$\begin{array}{cc}\hfill 𝒆_{2k}& =\underset{k}{\underset{}{𝝈_z\mathrm{}𝝈_z}}𝝈_x\underset{nk1}{\underset{}{\mathrm{𝟏}\mathrm{}\mathrm{𝟏}}},\hfill \\ \hfill 𝒆_{2k+1}& =\underset{k}{\underset{}{𝝈_z\mathrm{}𝝈_z}}𝝈_y\underset{nk1}{\underset{}{\mathrm{𝟏}\mathrm{}\mathrm{𝟏}}}.\hfill \end{array}$$
(1)
The set is well known in quantum mechanics due to Jordan, Wigner and Weyl works , because operators
$$𝒂_k=\frac{𝒆_{2k}+i𝒆_{2k+1}}{2},𝒂_k^{}=\frac{𝒆_{2k}i𝒆_{2k+1}}{2}$$
(2)
(i.e., $`2^n\times 2^n`$ complex matrices) provide representation of canonical anticommuting relations (CAR)
$$\{𝒂_k,𝒂_j\}=\{𝒂_k^{},𝒂_j^{}\}=0,\{𝒂_k,𝒂_j^{}\}=\delta _{kj}.$$
(3)
On the other hand, Eq. (1) may be used for construction Spin$`(2n)`$ and Spin$`(2n+1)`$ groups . The Spin$`(2n+1)`$ groups has $`21`$ isomorphism with group of rotations SO$`(2n+1)`$ and so for $`n=1`$ we have usual model with Bloch sphere and group SO$`(3)`$ of three-dimensional rotations of the sphere.
For $`n>1`$ it is also possible to consider groups Spin$`(2n+1)`$ and SO$`(2n+1)`$, but they may not describe all possible transformations of system with $`n`$ qubits. Such transformations may be described by huge group SU$`(2^n)`$ with dimension $`4^n1`$, but group SO$`(2n+1)`$ has dimension $`(2n+1)n`$ and only for $`n=1`$ both numbers coinside:
$$\begin{array}{ccccccc}& & & & & & \\ n& \hfill 1& \hfill 2& \hfill 3& \hfill 4& \hfill 5& \hfill 10\\ & & & & & & \\ & & & & & & \\ (2n+1)n& \hfill 3& \hfill 10& \hfill 21& \hfill 36& \hfill 55& \hfill 210\\ & & & & & & \\ 4^n1& \hfill 3& \hfill 15& \hfill 63& \hfill 255& \hfill 1023& \hfill 1048575\end{array}$$
The Eq. (1) are Hermitian matrices and may be considered as set of $`2n`$ Hamiltonians for system with $`n`$ qubits. If to use only these Hamiltonians for control of system, then unitary evolution belong only some subgroup of $`S_o\mathrm{SU}(2^n)`$, i.e., the control is not universal. Despite the subgroup $`S_o`$ belongs to such exponentially big space, it is ismorphic to Spin$`(2n+1)`$ and due to usual relation of Spin groups with rotations of $`2n+1`$-dimensional hypersphere may be considered as higher dimensional analogue model of Bloch sphere rotations.
It should be mentioned, that together with $`S_o\mathrm{SU}(2^n)`$, $`S_o\mathrm{Spin}(2n+1)`$, it is also useful to consider (maybe more familiar) even subgroup $`S_eS_o`$, $`S_e\mathrm{Spin}(2n)`$ . The subgroup is generated by even elements $`𝒅_k=i𝒆_k𝒆_{k+1}`$
$`𝒅_{2k}`$ $`=\underset{k}{\underset{}{\mathrm{𝟏}\mathrm{}\mathrm{𝟏}}}𝝈_z\underset{nk1}{\underset{}{\mathrm{𝟏}\mathrm{}\mathrm{𝟏}}},`$ (4a)
$`𝒅_{2k+1}`$ $`=\underset{k}{\underset{}{\mathrm{𝟏}\mathrm{}\mathrm{𝟏}}}𝝈_x𝝈_x\underset{nk2}{\underset{}{\mathrm{𝟏}\mathrm{}\mathrm{𝟏}}}.`$ (4b)
Now it is possible to add any operator $`𝒆_k`$, say
$$𝒆_0=𝝈_x\underset{n1}{\underset{}{\mathrm{𝟏}\mathrm{}\mathrm{𝟏}}},$$
(5)
to provide control on $`S_o`$ and it was shown in , that it is enough to add also any third (or fourth) order operator like $`𝒆_k𝒆_l𝒆_m`$, say
$$𝒆_0𝒆_1𝒆_3=\mathrm{𝟏}𝝈_y\underset{n2}{\underset{}{\mathrm{𝟏}\mathrm{}\mathrm{𝟏}}},$$
(6)
to provide universal control, whole group SU$`(2^n)`$.
The subgroup $`S_e`$ also may be generated by Hermitian bilinear combinations of fermionic annihilation and creation operators Eq. (2), i.e.
$$𝒂_j𝒂_k^{}+𝒂_k𝒂_j^{},𝒂_j𝒂_k+𝒂_k^{}𝒂_j^{}.$$
(7)
It should be mentioned, that Eq. (2) and Eq. (7) here should be considered rather from point of view of simulation of quantum control and computations with fermionic systems , because most methods used above may be applied to arbitrary system of $`n`$ qubits and are not related directly with fermionic statistic of particles in spin chain.
The note about simulation may be quite essential, say in usual fermionic systems as well as in linear optics KLM model of quantum computing (close related with the fermionic operators ) appearance of the group $`S_e`$ generated by bilinear combinations produces specific difficulty. Really, universal control suggest exponentially big space of parameters ($`4^n`$), but group $`S_e`$ has dimension only quadratic with respect to number of systems ($`dimS_e=2n^2n<dimS_o=2n^2+ndim\mathrm{SU}(2^n)=4^n1`$).
For spin chain considered here the problem with non-universality is not so essential, because it is enough to use Hamiltonians Eq. (5) and Eq. (6) to extend group $`S_e`$ to exponentially big group of universal control. And the two extra Hamiltonian are simply one-qubit rotations $`𝝈_x`$ and $`𝝈_y`$ with first and second qubit and so complexity of realization for such operations may not exceed analogous operations $`𝒅_{2k}`$ with Hamiltonian $`𝝈_z`$ Eq. (4a).
On the other hand, in fermionic computations and linear optics, analogues of operators with higher order like Eq. (6) may not be realized so simply because need physical processes with very law amplitude, like nonlinear $`\gamma +\gamma `$ interactions. So spin chain not only may simulate some fermionic or optic computations and control available for modern state of technologies, but also some currently inaccessible, very weak processes.
Using modern jargon whole universal set of gates considered above may be denoted
$$\begin{array}{ccc}\hfill 𝐈)& 𝒁(k+1)\hfill & =𝒅_{2k},\hfill \\ \hfill 𝐈𝐈)& 𝑿(k+1)𝑿(k+2)\hfill & =𝒅_{2k+1}\hfill \\ & 𝑿(1)\hfill & =𝒆_0\hfill \\ \hfill 𝐈𝐈𝐈)& 𝒀(2)\hfill & =𝒆_0𝒆_1𝒆_3,\hfill \end{array}$$
(8)
(see Fig. 2), where gate $`𝑿(1)`$ for simplicity considered as part of $`𝑿(k+1)𝑿(k+2)`$-bus. In such a case buses I and II with $`2n`$ gates generate subgroup $`S_o`$ and may be associated with control of rotations in dimension $`D=2n+1`$.
Such method let us not only model some processes in linear optics and fermionic systems, but also decomposes difficult task of control with exponentially big group SU$`(2^n)`$ on simpler task of control with groups SO$`(2n)`$ or SO$`(2n+1)`$ and SO$`(2)`$.
Let us consider structure of group $`S_o`$ with more details. Unitary matrices from groups $`S_o`$ and $`S_e`$ may be represented as
$$\left\{US_e\right|U=\mathrm{exp}\left(\underset{j=0}{\overset{2n1}{}}\underset{k=0}{\overset{j1}{}}b_{kj}𝒆_k𝒆_j\right)\},$$
(9)
$$\left\{US_o\right|U=\mathrm{exp}\left(i\underset{j=0}{\overset{2n1}{}}b_j𝒆_j+\underset{\begin{array}{c}j,k=0\\ k<j\end{array}}{\overset{2n1}{}}b_{kj}𝒆_k𝒆_j\right)\}.$$
(10)
The number of parameters is in agreement with dimensions mentioned above $`dimS_e=2n^2n`$, $`dimS_o=2n^2+n`$.
On the other hand, due to general defininion of group for any $`U_1,U_2S_o`$ product $`U_1U_2S_o`$, but $`i𝒆_k=\mathrm{exp}(i\frac{\pi }{2}𝒆_k)S_o`$ and so $`2n`$ elements $`𝒆_k`$ and all $`2^{2n}`$ possible products of the elements (up to insignificant multiplier $`i`$) belong to $`S_o`$.
The $`2^{2n}`$ products of $`𝒆_k`$ are simply $`4^n`$ possible tensor products with $`n`$ terms. Each term is either Pauli martix or $`2\times 2`$ unit matrix. It is discrete Pauli group, widely used in theory of quantum computations .
So the discrete Pauli group with $`4^n`$ elements is subgroup of continuous group $`S_o`$. The property show, that $`S_o`$ has rather nontrivial structure, because matrices from Pauli group are basis in $`4^n`$ dimensional space of matrices and so convex hull of $`S_o`$ is also $`4^n`$-dimensional.
Anyway, $`S_o`$ may be described as subspace of SU$`(2^n)`$ isomorphic to SO$`(2n+1)`$ and so quantum control with exponentially big space of parameters may be considered as alternating of control over “winding” subspace $`S_o`$ isomorphic to SO$`(2n+1)`$ and one-parametric rotations $`e^{i\alpha 𝒀(2)}`$. The control over $`S_o`$ represented on Fig. 2 by two buses and one-parametric rotations as one “exceptional” gate.
### Note (June 2005)
It should be mentioned a specific case, corresponding to two qubits and not discussed in presented paper. Due to ‘sporadic’ isomorphism
$$\mathrm{SU}(4)\mathrm{Spin}(6)$$
transformations of two qubits may be associated with rotations of 6D sphere and so it is also relevant to considered theme. It may be described using so-called Klein correspondence and Plücker coordinates (introduced first in few papers written between 1865 and 1870 by F. Klein and J. Plücker, see R. Penrose and W. Rindler, Spinors and Space-Time, vol.2, Spinor and Twistor Methods in Space-Time Geometry, Cambridge Univ. Press 1986), but it should be discussed elsewhere.
|
warning/0506/gr-qc0506130.html
|
ar5iv
|
text
|
# On the black hole limit of rotating fluid bodies in equilibrium1footnote 11footnote 1Fondly dedicated to Gernot Neugebauer on the occasion of his 65th birthday.
## I Introduction
It was shown in Meinel (2004) that a continuous sequence of stationary and axisymmetric, uniformly rotating perfect fluid bodies with a “cold matter” equation of state and finite baryonic mass can reach a (Kerr) black hole limit only if the relation
$$M=2\mathrm{\Omega }J$$
(1)
between the gravitational mass $`M`$, the angular velocity $`\mathrm{\Omega }`$ and the angular momentum $`J`$ is satisfied in the limit.<sup>2</sup><sup>2</sup>2We use units in which the speed of light as well as Newton’s gravitational constant are equal to 1. Strictly speaking, $`\mathrm{\Omega }`$ and $`J`$ are the components of angular velocity and angular momentum with respect to the axis of symmetry and can have either sign, corresponding to the two possible directions of rotation. This result implies the impossibility of black hole limits of non-rotating equilibrium configurations (cf. “Buchdahl’s inequality”), and moreover, since $`\mathrm{\Omega }`$ must become equal to the “angular velocity of the horizon”
$$\mathrm{\Omega }^H=\frac{J}{2M^2\left[M+\sqrt{M^2(J/M)^2}\right]},$$
(2)
the relation
$$J=\pm M^2,$$
(3)
characteristic of an extreme Kerr black hole, must hold in the limit. The possibility of such a limit was first demonstrated for infinitesimally thin disks – numerically by Bardeen and Wagoner Bardeen and Wagoner (1971) and analytically by Neugebauer and Meinel Neugebauer and Meinel (1995), see also Meinel (2002). Further numerical examples, for genuine fluid bodies, were provided by the “relativistic Dyson rings” Ansorg et al. (2003) and their generalizations Fischer et al. .
The aim of the present paper is to show that condition (1) is not only necessary, but also sufficient for reaching the black hole limit. An equivalent condition will turn out to be the statement that a zero angular momentum photon emitted from the fluid’s surface suffers an infinite redshift.
## II Basic relations
In the following, some basic relations for rotating fluids in equilibrium are provided, for more details see, for example, Hartle and Sharp (1967); Bardeen and Wagoner (1971); Thorne ; Lindblom (1992); Friedman and Ipser (1992); Neugebauer and Meinel (2003). The four-velocity of the fluid must point in the direction of a linear combination of the two commuting Killing vectors $`\xi =/t`$ and $`\eta =/\varphi `$ corresponding to stationarity and axisymmetry:
$$u^i=e^V(\xi ^i+\mathrm{\Omega }\eta ^i),\mathrm{\Omega }=\text{constant.}$$
(4)
The Killing vector $`\xi `$ is fixed by the normalization $`\xi ^i\xi _i1`$ at spatial infinity (we assume asymptotic flatness).<sup>3</sup><sup>3</sup>3The spacetime signature is chosen to be ($`+++`$). The orbits of the spacelike Killing vector $`\eta `$ are closed and $`\eta `$ is zero on the axis of symmetry. The constant of eq. (4), $`\mathrm{\Omega }=u^\varphi /u^t`$, is the angular velocity of the fluid body with respect to infinity. Using $`u^iu_i=1`$, the factor $`e^V=u^t`$ is given by
$$(\xi ^i+\mathrm{\Omega }\eta ^i)(\xi _i+\mathrm{\Omega }\eta _i)=e^{2V}.$$
(5)
The energy-momentum tensor is
$$T_{ik}=(ϵ+p)u_iu_k+pg_{ik},$$
(6)
where the mass-energy density $`ϵ`$ and the pressure $`p`$ are related by a “cold” equation of state, $`ϵ=ϵ(p)`$, following from
$$p=p(\rho ,T),ϵ=ϵ(\rho ,T)$$
(7)
for $`T=0`$, where $`\rho `$ is the baryonic mass-density and $`T`$ the temperature. The specific enthalpy
$$h=\frac{ϵ+p}{\rho }$$
(8)
can be calculated from $`ϵ(p)`$ via the thermodynamic relation
$$dh=\frac{1}{\rho }dp(T=0)$$
(9)
leading to
$$\frac{dh}{h}=\frac{dp}{ϵ+p}h(p)=h(0)\mathrm{exp}\left[\underset{0}{\overset{p}{}}\frac{dp^{}}{ϵ(p^{})+p^{}}\right].$$
(10)
Note that $`h(0)=1`$ in most cases. For our purposes, however, it is sufficient to assume $`0<h(0)<\mathrm{}`$. From $`T_{}^{ik}{}_{;k}{}^{}=0`$ we obtain
$$h(p)e^V=h(0)e^{V_0}=\mathrm{constant},$$
(11)
where $`V_0`$, the constant surface value (corresponding to $`p=0`$) of the function $`V`$ defined in (5), is related to the relative redshift $`z`$ of zero angular momentum photons<sup>4</sup><sup>4</sup>4Zero angular momentum means $`\eta _ip^i=0`$ ($`p^i`$: four-momentum of the photon). emitted from the surface of the fluid and received at infinity:
$$z=e^{V_0}1.$$
(12)
Equilibrium models, for a given equation of state, are fixed by two parameters, for example $`\mathrm{\Omega }`$ and $`V_0`$. (When we discuss a “sequence” of solutions, what is meant is a curve in the two-dimensional parameter space.) The gravitational mass and the angular momentum can be calculated by
$$M=2\underset{\mathrm{\Sigma }}{}(T_{ik}\frac{1}{2}Tg_{ik})n^i\xi ^k𝑑𝒱,J=\underset{\mathrm{\Sigma }}{}T_{ik}n^i\eta ^k𝑑𝒱,$$
(13)
where $`\mathrm{\Sigma }`$ is a spacelike hypersurface ($`t=\mathrm{constant}`$) with the volume element $`d𝒱=\sqrt{{}_{}{}^{(3)}g}d^3x`$ and the future pointing unit normal $`n^i`$, see for example Wald (1984). The baryonic mass $`M_0`$ corresponding to the local conservation law $`(\rho u^i)_{;i}=0`$ is
$$M_0=\underset{\mathrm{\Sigma }}{}\rho u_in^i𝑑𝒱.$$
(14)
Note that nearby equilibrium configurations with the same equation of state are related by Hartle and Sharp (1967); Bardeen and Wagoner (1971)
$$\delta M=\mathrm{\Omega }\delta J+\mu \delta M_0,\mu =h(0)e^{V_0}.$$
(15)
The parameter $`\mu `$ (“chemical potential”) represents the specific injection energy of zero angular momentum baryons.
## III Conditions for a black hole limit
A combination of (13) and (14) leads to the formula
$$M=2\mathrm{\Omega }J+\frac{ϵ+3p}{\rho }e^V𝑑M_0,$$
(16)
cf. equation (II.28) in Bardeen and Wagoner (1971). With (8) and (11) we get
$$M=2\mathrm{\Omega }J+h(0)e^{V_0}\frac{ϵ+3p}{ϵ+p}𝑑M_0.$$
(17)
Since $`1(ϵ+3p)/(ϵ+p)3`$ (we assume $`ϵ`$ and $`p`$ to be non-negative), condition (1) is equivalent to<sup>5</sup><sup>5</sup>5We assume $`0<M_0<\mathrm{}`$.
$$V_0\mathrm{}(z\mathrm{}).$$
(18)
We now want to show that this condition is not only necessary (as discussed in Meinel (2004)), but also sufficient for approaching a black hole limit.
Because of (5) and (11), the surface of the fluid is characterized in general by
$$\chi ^i\chi _i=e^{2V_0},\chi ^i\xi ^i+\mathrm{\Omega }\eta ^i.$$
(19)
The Killing vector $`\chi ^i`$ is tangential to the hypersurface $``$ generated by the timelike world lines of the fluid elements of the surface of the body with four-velocity $`u^i=e^{V_0}\chi ^i`$, see (4). Each of the Killing vectors $`\xi ^i`$ and $`\eta ^i`$ must itself be tangential to $``$ because of the symmetries of the spacetime. In the limit $`V_0\mathrm{}`$, we approach a situation in which $`\chi ^i`$ becomes null on $``$:
$$\chi ^i\chi _i0.$$
(20)
Moreover, with the reasonable assumption<sup>6</sup><sup>6</sup>6The condition $`\xi ^iu_i1`$ ensures that a particle resting on the surface of the fluid is (at least marginally) bound, i.e. cannot escape to infinity on a geodesic; $`\xi ^iu_i0`$ follows from $`\xi ^iu_i=(\chi ^i\mathrm{\Omega }\eta ^i)u_i=e^{V_0}+\mathrm{\Omega }\eta ^iu_i`$, since $`\eta ^iu_i`$ will always have the same sign as $`\mathrm{\Omega }`$ ($`\eta ^iu_i=0`$ on the axis, of course).
$$0\xi ^iu_i1,$$
(21)
we find that $`\chi ^i`$ also becomes orthogonal to $`\xi ^i`$ (and thus to $`\eta ^i`$) on $``$ in the limit:
$$\chi ^i\xi _i0,\chi ^i\eta _i0.$$
(22)
Together with the orthogonal transitivity<sup>7</sup><sup>7</sup>7The conditions of the theorem by Kundt and Trümper Kundt and Trümper (1966) are satisfied. of the spacetime, $`\chi ^i`$ therefore becomes orthogonal to three linearly independent tangent vectors at each point of $``$, i.e. normal to $``$. Because of (20), we thus approach a situation in which $``$ is a null hypersurface and satisfies all defining conditions for a horizon of a stationary (and axisymmetric) black hole with $`\mathrm{\Omega }`$ being the angular velocity of the horizon, see Carter (1973). According to the black hole uniqueness theorems (see Hawking and Ellis (1973); Heusler (1996) and also Neugebauer and Meinel (2003)) we conclude that (outside the horizon) the Kerr metric with $`|J|M^2`$ results<sup>8</sup><sup>8</sup>8Note that the black hole uniqueness proof by construction given in Neugebauer and Meinel (2003) can be extended to the case in which the horizon is degenerate, leading to $`|J|=M^2`$. This will be shown in a future publication.. Then, with (1), we are necessarily led to the case $`|J|=M^2`$. Therefore, the metric of an extreme Kerr black hole (outside the horizon) results, whenever a sequence of fluid bodies admits a limit $`V_0\mathrm{}`$.
## IV Discussion
The special properties of the extreme Kerr metric with its degenerate horizon and the infinitely long “throat region” allow for the existence of a black hole limit independent of the fluid body’s topology. Indeed, such a limit was found numerically for bodies of toroidal topology Ansorg et al. (2003). Strictly speaking, there is not yet a horizon in the limit. Instead, a separation of an “inner” and an “outer” world occurs. The “inner world” contains the fluid body and is not asymptotically flat, but approaches the “extreme Kerr throat geometry” Bardeen and Horowitz (1999) at spatial infinity. The “outer world” is given by the $`r>M`$ part of the extreme Kerr metric, where $`r`$ is the radial Boyer-Lindquist coordinate. Note that the horizon as well as the “throat region” are characterized by $`r=M`$. Here, the whole “inner world” corresponds to $`r=M`$. It should be mentioned in this connection that the conditions (20) and (22) are also satisfied inside the fluid as $`V_0\mathrm{}`$, cf. (10) and (11). More details can be found in Bardeen and Wagoner (1971); Meinel (2002); Ansorg et al. (2003).
It is interesting to note that a similar separation of spacetimes has been observed for some limiting solutions of the static, spherically symmetric Einstein-Yang-Mills-Higgs equations Bartnik and McKinnon (1988); Breitenlohner et al. (1995) leading to the extreme Reissner-Nordström metric in the “outer world”.
As discussed in Bardeen (1973), the slightest dynamical perturbation will lead to a genuine black hole. Therefore, it is tempting to continue a sequence of fluid bodies beyond the black hole limit as a sequence of Kerr black holes and to discuss the transition from the “normal matter state” to the “black hole state” as a (one-way) phase transition Neugebauer and Meinel (1993). From the exterior point of view, this transition is continuous, i.e. all gravitational multipole moments change continuously. It is interesting to compare the mass formula (17) as well as the differential relation (15) with the black hole formulas Smarr (1973); Bardeen et al. (1973)
$$M=2\mathrm{\Omega }^HJ+\frac{\kappa }{4\pi }A,$$
(23)
$$\delta M=\mathrm{\Omega }^H\delta J+\frac{\kappa }{8\pi }\delta A,$$
(24)
where $`\kappa `$ denotes the “surface gravity” and $`A`$ the area of the horizon. In the two-dimensional parameter space, the transition line is characterized by $`M=2\mathrm{\Omega }J=2\mathrm{\Omega }^HJ`$. On the “fluid side” of this line, the parameter $`\mu =h(0)e^{V_0}`$ of the chemical potential vanishes ($`V_0\mathrm{}`$), whereas $`\kappa `$ (related to the temperature in black hole thermodynamics) vanishes on the “black hole side” of the transition line ($`\kappa =0`$ for extreme Kerr black holes). The quantities $`M_0`$ (baryonic mass of the fluid body) and $`A`$ (related to the black hole entropy) are defined in the corresponding regions of the parameter space only. This is consistent with $`\mu 0`$ in the black hole region and $`T0`$ in the fluid region, i.e. $`\mu `$ and $`T`$ are continuous across the transition line. (An alternative interpretation was given in Neugebauer (1998).)
The parametric transitions from fluid bodies to black holes discussed here may be used as a starting-point for dynamical collapse investigations far from the spherically symmetric case.
###### Acknowledgements.
I would like to thank G. Neugebauer, D. Petroff, A. Kleinwächter and H. Labranche for valuable discussions. This work was supported in part by the Deutsche Forschungsgemeinschaft (DFG project SFB/TR7-B1).
|
warning/0506/astro-ph0506625.html
|
ar5iv
|
text
|
# A model for the infrared dust emission from forming galaxies
## 1 Introduction
In spite of the recent vast progress both in observational and theoretical studies, our understanding of the physics of galaxy formation and evolution is still far from sufficient.
The cosmic star formation (SF) history, introduced by Tinsley & Danly (1980) and developed by subsequent studies (e.g., Lilly et al., 1996; Madau et al., 1996), has always drawn much attention. Now the observations reach up to $`6\text{}8`$ (e.g, , Stanway et al.2003; Bouwens et al., 2004a, b). In such studies, the role of dust has been increasingly recognized when we try to understand the evolution of galaxies in the context of cosmic star formation history, because dust grains absorb stellar light and re-emit it in the far infrared (FIR). Even a small amount of dust can lead to a significant underestimation of the star formation rate (SFR) (Steidel et al., 1999; Adelberger & Steidel, 2000). Indeed, there is another extreme category of high-$`z`$ galaxies which have large amount of dust and are extremely luminous in the FIR and submillimetre (submm) wavelengths (e.g., Hughes et al., 1998; Eales et al., 2003). Heavily hidden SF is suggested in these galaxies (e.g., Takeuchi et al., 2001a, b; Totani & Takeuchi, 2002).
Further, from physical point of view, dust grains are one of the fundamental ingredients in the activity of galaxies. Since they are formed in a variety of environments ranging from explosive ejecta of novae and supernovae to the outflowing gas of evolved low-mass stars, the dust formation is closely related to the SF activity (Dwek, 1998). Moreover, the existence of dust is crucial in the physical process of galaxy formation and evolution through the formation of molecular hydrogen (e.g., Hirashita & Ferrara, 2002). The dust itself plays a leading part in the physics of SF activity in galaxies in the whole period of the cosmic history.
Then, how about very young galaxies in the early universe at $`z5`$? It is often assumed, without deliberation, that the effect of dust is negligible for such young galaxies, because of their low metallicities. There is, however, a good counter-intuitive example in the local Universe: a local dwarf star-forming galaxy SBS 0335$``$052 has a very young stellar age ($`10^7\text{yr}`$) and low metallicity ($`1/41Z_{}`$), but has a heavily embedded active star formation and strong continuum radiation in N–MIR wavelength regime (Dale et al., 2001; , Hunt et al.2001). In addition, there are increasing number of observations suggesting the existence of dust in high-$`z`$ systems such as Lyman $`\alpha `$ systems (e.g., , Ledoux et al.2002, Ledoux et al.2003) or QSOs (e.g., Bertoldi et al., 2003).
To produce dust effectively in such a young system, the dust enrichment should have occurred primarily in the ejecta of supernovae (SNe), especially Type II supernova (SN II) explosions<sup>1</sup><sup>1</sup>1 More specifically, core-collapse supernovae., because the lifetime of the progenitor is short enough ($`10^6\text{yr}`$) (e.g., Dwek & Scalo, 1980; , Kozasa et al.1989). Recent theoretical studies claim the possibility of very massive stars for the first generation stars, which end their lives as pair-instability supernovae (PISNe) (Heger & Woosley, 2002; Umeda & Nomoto, 2002). We should also take into account for the PISNe to study the very early evolution of dust. Recently, the dust formation in SNe ejecta has been observationally supported (e.g., Douvion et al., 2001; Morgan et al., 2003; Dunne et al., 2003).
For investigating the properties of dust in details, theoretical predictions for the amount and composition of dust are required. So far, some theoretical models of dust production by SNe have been developed. Todini & Ferrara (2001, hereafter TF01) showed that the dust mass produced by a SN II is 0.1–0.4 $`M_{}`$ applying the theory of nucleation and grain growth by Kozasa & Hasegawa (1987). In the calculations, they assumed an adiabatic cooling in the ejecta with uniform gas density as well as elemental composition within He-core. They also found that SNe form amorphous carbon with size around 300 Å and silicate grains around 10–20 Å. Schneider, Ferrara & Salvaterra (2004) extended the progenitor mass range to the regime of PISNe (140–260 $`M_{}`$) and found that 10–60 $`M_{}`$ of dust forms per PISN. The grain radius depends on the species and is distributed from 0.001 to 0.3 $`\mu `$m.
Recently, Nozawa et al. (2003) (hereafter N03) carefully took into account the radial density profile and the temperature evolution in the calculation of the dust mass in the ejecta of SNe II and PISNe. N03 showed that the produced dust species depends strongly on the mixing within SNe. In particular, carbon dust is not produced in the mixed case, because the carbon and oxygen are mixed and combined to form CO molecules. On the contrary, it forms in unmixed SN, since there is a carbon-rich region at a certain location in the ejecta of SNe. In addition, N03 predicts a dust mass larger than that of TF01 for SNe II.
The amount of dust produced in SNe is thus still a matter of debate. Morgan & Edmunds (2003) adopted a dust production smaller than those of TF01 or N03 in their theoretical calculation, in line with IR observations at the time. On the other hand, various observational efforts have been devoted to constrain the dust production efficiency of SNe (e.g., Arendt, Dwek, & Moseley, 1999; Dunne et al., 2003; Morgan et al., 2003; , Green et al.2004; Hines et al., 2004; Krause et al., 2004; Wilson & Batrla, 2005). These results range from $`0.003M_{}`$ to $`1.0M_{}`$ per SN, and N03 dust production seems consistent with the upper range of these observational constraints. However, in these observational evaluations, a large uncertainty still remains due to the uncertainties in the estimation of dust mass in Cas A caused by the foreground contamination (Krause et al., 2004), although the evidence for dust in some SN remnants (SN1987A and Kepler) is still pointing towards SN origin. Thus, further close interaction between theoretical and observational works is required to overcome the uncertainties and obtain a reasonable picture of dust production in SNe.
In order to examine the dust properties of such high-$`z`$ galaxies, the most direct observable is their spectral energy distribution (SED), especially at the FIR wavelengths. (Hirashita et al.2002) modeled the evolution of FIR luminosity and dust temperature in a young starburst on the basis of SNe II grain formation model of TF01. Takeuchi et al. (2003) (T03) subsequently constructed a model of infrared (IR) spectral energy distribution (SED) of galaxies starting from the model of (Hirashita et al.2002). T03, for the first time, properly consider the dust size distribution peculiar to the very early stage of galaxy evolution in the model of the IR SED of very young galaxies, and successfully reproduced the peculiar MIR SED of SBS 0335$``$052 (though their result seems to have overestimated the FIR continuum). Takeuchi & Ishii (2004) (T04) applied the T03 model to the Lyman-break galaxies. They found that the suggested hot dust in the Lyman-break galaxies (e.g., Ouchi et al., 1999; Chapman et al., 2000; Sawicki, 2001) can be naturally explained by the T03 model.
Since these works are based on TF01 conjecture, the dust formation model used is based on the classical nucleation theory (Feder et al., 1966). Recently, however, the importance of nonequilibrium (non-steady state) effects on the dust grain formation has been recognized in various astrophysical contexts (e.g., , Gail et al.1984, Tanaka et al.2002, N03). In addition, N03 found that the radial density profile of the SN progenitor and the temperature evolution of the ejecta also affect the dust grain formation. Hence, now is the time to take into account these concepts to further investigate the dust emission from young galaxies. In this work, we construct a new model of IR SED of extremely young galaxies based on N03 SN dust formation model. Starting from the size distribution and the amount of dust predicted by N03, we calculate the dust emission model by extending the T03 model to treat multiple dust species.
The paper is organized as follows: In §2 we explain the framework of our SED model. We present the basic result, the evolution of the SED of extremely young galaxies, in §3. Related discussions on local star-forming dwarf galaxies and high-$`z`$ galaxies will be in §4. §5 is devoted to our conclusions. Throughout this paper, we use a cosmological parameter set of $`(h,\mathrm{\Omega }_0,\lambda _0)=(0.7,0.3,0.7)`$, where $`hH_0/100[\text{km\hspace{0.17em}s}^1\text{Mpc}^1]`$.
## 2 SED Model for Forming Galaxies
### 2.1 Species and size distribution of dust grains produced by SNe II
#### 2.1.1 Dust production model of Nozawa et al. (2003) (N03)
N03 investigated the formation of dust grains in the ejecta of Population III SNe (SNe II and PISNe, whose progenitors are initially metal-free). As we mentioned above, they treat some aspects which TF01 have not taken into account: (i) the time evolution of gas temperature is calculated by solving the radiative transfer equation including the energy deposition of radioactive elements. (ii) the radial density profile of various metals is properly considered, and (iii) unmixed and uniformly mixed cases in the He core are considered. In the unmixed case, the original onion-like structure of elements is preserved, and in the mixed case, all the elements are uniformly mixed in the helium core.
It should be mentioned here again that TF01 assumed an adiabatic cooling in the ejecta and adjusted the adiabatic index $`\gamma `$ to 1.25, referring to the formation episode of dust grains observed in SN 1987A. As pointed out by (Kozasa et al.1989), the condensation time as well as the resulting average size of dust grains strongly depend on the value of $`\gamma `$. SN 1987A is somehow a peculiar SN in the sense that the progenitor is not a red supergiant but blue supergiant: see Arnett et al. (1989) for the details. Using SN 1987A as a template may not be appropriate in comparisons to supernovae as a whole. Therefore in this paper, we use the result of N03 as a standard model.
We should note that N03 also assume the complete formation of CO and SiO molecules, neglecting the destruction of those molecules, i.e., no carbon-bearing grain condenses in the region of $`\mathrm{C}/\mathrm{O}<1`$ and no Si-bearing grain, except for oxide grains, condenses in the region of $`\mathrm{Si}/\mathrm{O}<1`$. The formation of CO and SiO may be incomplete because of the destruction by energetic electron impact within SNe. TF01 treat both formation and destruction of CO and SiO, finding that both are mostly destroyed. The decrease of CO leads to the formation of carbon grains, which could finally be oxidised with available oxygen. The destruction of SiO could decrease the formation of grains composed of SiO<sub>2</sub>, MgSiO<sub>3</sub>, and Mg<sub>2</sub>SiO<sub>4</sub>, and increase other oxidised grains and Si grains. Observationally, it is still a matter of debate if CO and SiO are efficiently destroyed or not. For the detailed discussions on this issue, see Appendix B of N03.
#### 2.1.2 Dust grain species produced by N03 model
In the unmixed ejecta, a variety of grain species (Si, Fe, Mg<sub>2</sub>SiO<sub>4</sub>, MgSiO<sub>3</sub>, MgO, Al<sub>2</sub>O<sub>3</sub>, SiO<sub>2</sub>, FeS, and C) condense, and in the mixed ejecta, in contrast, only oxide grains (SiO<sub>2</sub>, MgSiO<sub>3</sub>, Mg<sub>2</sub>SiO<sub>4</sub>, Al<sub>2</sub>O<sub>3</sub>, and Fe<sub>3</sub>O<sub>4</sub>) form. This is because carbon atoms are consumed to form CO molecules. We summarize the species formed in SNe in Table 1, where the species marked with a circle are relevant for unmixed and mixed SNe. The size of the grains spans a range of three orders of magnitude, depending on the grain species. The size spectrum summed up over all the grain species has a very broad distribution, and very roughly speaking, it might be approximated by a power law. This size distribution is different from that of the SN II calculation of TF01, which has a typical sizes of 300 Å for amorphous carbon and 10–20 Å for oxide grains.
In this work, we adopt the representative progenitor mass of SNe II as $`20M_{}`$. N03 have shown that the size distribution of each grain species is almost independent of the progenitor mass, if the SN type is fixed (i.e., SN II or PISN). We examine the unmixed and mixed cases. The size distributions of dust grains in the mixed and unmixed cases calculated by N03 are shown in Figure 1. We used these size distributions after binning with a bin width of 0.2 dex for our calculations. Throughout this work, we assume a uniform and spherical grain. It should be mentioned that a different shape of dust grains in SN is suggested (e.g., Dwek, 2004).
### 2.2 Star formation, chemical evolution, and dust production
For constructing the chemical evolution model of a young galaxy, we adopt the following assumptions:
1. We use a closed-box model, i.e., we neglect an infall and outflow of gas in the scale of a star-forming region.
2. For the initial mass function (IMF), we adopt the Salpeter IMF (Salpeter, 1955)
$`\varphi (m)m^{2.35}`$ (1)
with mass range of $`(m_\mathrm{l},m_\mathrm{u})=(0.1M_{},100M_{})`$.
3. We neglect the contribution of SNe Ia and winds from low-mass evolved stars to the formation of dust, because we consider the timescale younger than $`10^9\text{yr}`$.
4. The interstellar medium is treated as one zone, and the growth of dust grains by accretion is neglected. Within the short timescale considered here, it can be assumed safely (see, e.g., Whittet, 1992, pp.223–224).
5. We also neglect the destruction of dust grains within the young age considered (see e.g., , Jones et al.1996).
6. We assumed a constant SFR for simplicity.
using these assumptions, we calculate the chemical evolution. Details of the formulation are presented in Appendix A.
The evolution of the total dust amount for $`\text{SFR}=1M_{}\text{yr}^1`$ is shown in Figure 2. The dust mass fraction for unmixed and mixed cases are summarized in Tables 2 and 3, respectively. Comparing the evolution of dust mass given by T03, dust mass starts to accumulate later than the case of T03, and gradually approaches the T03 result toward the age of about $`10^8`$Gyr. This difference is caused by the different formulae we adopted in the calculations of stellar lifetime: the formula of Schaerer (2002) gives a longer lifetime for the same stellar mass than that of Inoue, Hirashita, & Kamaya (2000) which is used in T03. At $`10^8`$Gyr, T03 and present work yield the same dust mass.
### 2.3 SED construction
In this subsection, we present the construction of the SED from dust. Since most of the formulations are in parallel to those in T03 and T04, here we list the essence of our calculation.
1. Stochastic heating of very small grains
Very small grains cannot establish thermal equilibrium with the ambient radiation field, which is called stochastic heating (e.g., Krügel, 2003). To treat this effect, we applied the Debye model to the specific heat of the grain species as discussed in T03 and T04. We adopt a multidimensional Debye model (e.g., Draine & Li, 2001)(DL01) for carbon (C) and silicate (SiO<sub>2</sub>, MgSiO<sub>3</sub>, and Mg<sub>2</sub>SiO<sub>4</sub>) grains. For other species, we adopt the classical three-dimensional Debye model with a single Debye temperature. The specific heat model is summarized in Table 4.
2. Emission
The emission from dust is calculated in the same way as T03/T04, basically according to Draine & Anderson (1985). Total dust emission is obtained as a superposition of the emission from each grain species. The total mass of each grain component is given by N03. With this value and material density of each species (Table 1), we can determine the normalization of the dust size distribution (see Appendix B).
We constructed $`Q(a,\lambda )`$ of each grain species from available experimental data via Mie theory (e.g., Krügel, 2003, Chap. 2). The extinction efficiencies are presented in Figure 3. T03 adopted Draine & Lee (1984) for the optical properties of these species, being different from the present work. On the contrary, we treat detailed species in the present work. Hence, the mass fraction distributed to carbon and silicate are significantly reduced comparing to the case of T03. In the IR, $`Q(a,\lambda )`$ of these grains is much larger than other grains at the whole range of grain size. As a result, the sum of the contributions of all grains are less than that of the case where only two species are considered as T03.
3. Extinction
Self-absorption in the MIR for a very optically thick case is treated by a thin shell approximation, in the same manner as T04.
## 3 Results
### 3.1 Evolution of infrared SED
We first show the evolution of the IR SED of forming galaxies based on our baseline model. For these calculation we adopted the star formation rate $`\mathrm{SFR}=1M_{}\text{yr}^1`$. We adopt $`r_{\mathrm{SF}}=30`$ pc and 100 pc, the same as those used in T03. These values are relevant when describing ‘dwarf-like’ young galaxies. The former successfully reproduce various properties of active star-forming dwarf galaxies like SBS 0335$``$052, whereas the latter is representative of the size of quiescent dwarf galaxies like I Zw 18 (Hirashita & Hunt, 2004). We will revisit these representative dwarf starburst galaxies in Section 4.
The results are presented in Figures 4 and 5. Figure 4 is the SED of a galaxy with $`r_{\mathrm{SF}}=30\text{pc}`$, while Figure 5 is the one for $`r_{\mathrm{SF}}=100\text{pc}`$. We calculated the evolution of the SED in the age range of $`10^{6.5}\text{}10^8`$ yr. In a very young phase ($`\text{age}=10^{6.75}\text{}10^{7.25}`$ yr), unmixed-case SED has an enhanced N–MIR continuum. After $`10^{7.25}`$ yr, the N–MIR continuum is extinguished by the self-absorption in the case of $`r_{\mathrm{SF}}=30`$ pc. In contrast, the self-absorption is not significant for $`r_{\mathrm{SF}}=100`$ pc.
In both cases, the SEDs have their peaks at a wavelength $`\lambda 20\text{}30\mu `$m, which is much shorter than those of dusty giant galaxies at $`z=1\text{}3`$ detected by SCUBA. At submillimetre wavelengths, the shape of the continuum is very similar to each other for mixed and unmixed cases, though the unmixed case predicts stronger fluxes by a factor of two. This is explained by the grain size distributions of the unmixed and mixed cases: carefully examining Figure 1, we see that the largest-size dust grains are more abundant for the unmixed-case size distribution. It means that larger grains contribute more significantly than smaller ones. Since larger grains radiate their energy at longer wavelengths, the total SED has a stronger FIR-submm continuum.
As a whole, our model SEDs are less luminous than those of T03 by a factor of 2–3. This is due to the difference in considered grain species, and more importantly, also due to the improvement of the treatment of $`Q(a,\lambda )`$ at the UV; $`Q(a,\lambda )`$ was approximated to be unity in T03. Also T03 treated the size dependence of $`Q(a,\lambda )`$ by a simple scaling law, which was not very accurate. As seen in Figure 3, the size dependence of $`Q(a,\lambda )`$ cannot be described by a simple scaling law, especially at the MIR (see, e.g., Si, Fe, and FeS). We, in this work, used the exact value of $`Q(a,\lambda )`$ in all the wavelengths. Since some elements have a very small value for $`Q(a,\lambda )`$ at the UV regime, the total absorption probability of UV photons is smaller than that of T03. A strong N–MIR continuum at a very young phase and the dependence on $`r_{\mathrm{SF}}`$ are qualitatively consistent with the previous result of T03. However, the dust grains expected from N03 model predict a peak of the SED at shorter wavelengths than those of T03. In addition, a relatively small amount of small dust grains also makes the N–MIR SED weaker than the T03 one.
### 3.2 Contribution of each species to the SED
The dust species treated here are much more detailed than those of T03. In order to see the detailed contributors to the total SEDs, we show the individual SED of each species in Figure 6. We show the SED with the age of burst $`10^{6.75}`$ yr, where the effect of self-absorption is negligible.
Both in unmixed and mixed cases, the smallest dust grain species is Al<sub>2</sub>O<sub>3</sub> (Figure 1). However, because of its small fraction in mass (Figure 2), Al<sub>2</sub>O<sub>3</sub> contributes to the total SED very little. The main contributor to the (unextinguished) N–MIR continuum is Si for unmixed case and Fe<sub>3</sub>O<sub>4</sub> in mixed case, respectively. These species are the second smallest grains. For the unmixed case, since the emissivity of Si grains with the size of $`10^7\text{cm}`$ is much smaller than that of Fe<sub>3</sub>O<sub>4</sub> at $`\lambda 1\mu `$m, the grain temperature becomes much higher than that of Fe<sub>3</sub>O<sub>4</sub>, which makes the very strong N–MIR continuum emission. On the contrary, since $`Q(a,\lambda )`$ of Fe<sub>3</sub>O<sub>4</sub> is large, the resultant SED for the mixed case does not have a strong N–MIR continuum.
At $`\lambda =10\text{}20\mu `$m, several species contribute equally to the SED in unmixed case, while SiO<sub>2</sub> play a dominant role for mixed case. At longer wavelengths, the main contributors to the SED are Mg<sub>2</sub>SiO<sub>4</sub> and amorphous carbon for the unmixed case, and MgSiO<sub>3</sub> and Mg<sub>2</sub>SiO<sub>4</sub> in mixed case. In addition, especially at $`\lambda 200\mu `$m, FeS (unmixed) and Fe<sub>3</sub>O<sub>4</sub> (mixed) are also important because of their shallow slope of the extinction efficiency at longer wavelength regime. Note that, for the mixed case, Fe<sub>3</sub>O<sub>4</sub> contribute to both NIR and FIR.
### 3.3 Opacity and its evolution
The extinction curve $`A_\lambda `$ is obtained as
$`A_\lambda =1.086\tau _{\mathrm{dust}}(\lambda ).`$ (2)
In this work we concentrate on the extinction curve in the IR. Detailed discussions of the extinction curves in the optical wavelength regime are given in Hirashita et al. (2005).<sup>5</sup><sup>5</sup>5 Since we assume a very simple geometry, the extinction curve is directly related to the ‘attenuation curve’, in which the geometrical effect is also included (for a thorough discussion on this subject, see, e.g., Inoue, 2005).
Figure 7 presents the evolution of the IR opacity for a young galaxy with the age of $`10^{6.75}\text{}10^8`$yr (from the bottom to the top). For the case of $`r_{\mathrm{SF}}=30`$ pc, the extinction becomes almost unity at $`10\mu `$m when the age is $`10^{7.25}`$ yr. Hence, after this age, the N–MIR regime of the SED becomes optically thick, and this makes the continuum extinguished in Figure 4. On the other hand, for $`r_{\mathrm{SF}}=100`$ pc, it remains optically thin even at the age of $`10^8`$ yr at $`10\mu `$m.
In Figure 7, we find an interesting difference between the extinction curves of the unmixed and mixed cases. Though their behavior is similar to each other at wavelengths $`\lambda 10\mu `$m, there is no dip at $`\lambda 10\mu `$m in the curve of the unmixed case. This makes the extinction stronger for the unmixed-case galaxies at N–MIR. In addition, the MIR bumps at 10 and $`20\mu `$m are less prominent for unmixed case.
Comparing these IR extinction curves with T03 model, we find that the amount of the extinction is smaller than that of T03 by a factor of 2–10, depending on the wavelength and unmixed/mixed production. This is explained by the discussion presented in Subsection 2.3: T03 supposed silicates and carbon grains as the constituents of dust grains from SNe. Since their $`Q(a,\lambda )`$ is larger than the other species predicted by N03, the total extinction becomes smaller in this work.
As for the shape of the curve, the extinction curve for the mixed model resembles that of T03 extinction curves, while the curve for the unmixed model is qualitatively different at MIR, because of the weak MIR bump features and the lack of the MIR dip mentioned above.
Figure 8 demonstrates the contribution of each dust grain species to the total IR extinction curves. The adopted age is $`10^7`$ yr as a representative, but the contribution is constant with time.
For the extinction curve of the unmixed case, bump features at MIR are relatively weak (Figure 7). Similar to those of the Galactic IR extinction curve, they are due to silicate grains. In the unmixed-model extinction, the bumps are dominated by Mg<sub>2</sub>SiO<sub>4</sub>, and the contribution from MgSiO<sub>3</sub> is not important. We also find a strong contribution from amorphous carbon grains which have very smooth dependence of $`Q(a,\lambda )`$ on wavelength. They makes the silicate features rather weak, because of their comparable contribution with Mg<sub>2</sub>SiO<sub>4</sub>. We note that the amorphous carbon does not have any dip at $`\lambda 5\text{}10\mu `$m. Further, large grains of Si have a large $`Q(a,\lambda )`$ at these wavelength regime. We find that both the amorphous carbon and Si grains equally contribute to the total extinction at N–MIR, and these species plug up the MIR dip of Mg<sub>2</sub>SiO<sub>4</sub>. At NIR wavelengths close to $`1\mu `$m, FeS also contributes to the total extinction.
We next see the mixed case. In this case, MgSiO<sub>3</sub> and Mg<sub>2</sub>SiO<sub>4</sub> equally contribute to the total extinction around the MIR bumps. In contrast to the unmixed case, there is no species which fill up the dip of silicates, hence we see clear silicate bump features and dip in the IR extinction curve. In the N–MIR, main contributors to the opacity are Fe<sub>3</sub>O<sub>4</sub> and SiO<sub>2</sub>. The contribution from Al<sub>2</sub>O<sub>3</sub> is not important at any wavelength.
## 4 Discussion
### 4.1 Nearby forming dwarf galaxies
It is still a difficult task to observe galaxies in their very first phase of the SF, especially to detect their dust emission directly. Along the line of studies made by (Hirashita et al.2002), T03 considered two local star-forming dwarf galaxies, SBS 0335$``$052 and I Zw 18, for investigating the dust emission from very young small galaxies. Here we revisit these representative star-forming dwarfs with our framework. Since a recent observation of SBS 0335$``$052 by Spitzer has been reported (Houck et al., 2004), it is timely to reconsider these ‘textbook objects’ with the new data. In addition, understanding the SEDs of these objects will shed light to the physics of interstellar matter and radiation of high-$`z`$ galaxies also via empirical studies (e.g., Takeuchi, Yoshikawa, & Ishii, 2003; Takeuchi et al., 2005).<sup>6</sup><sup>6</sup>6We should keep in mind that the following discussion is limited to a particular class of dwarf galaxies dominated by newly formed stars. For the modeling of the SED of normal star-forming dwarfs, see, e.g., Galliano et al. (2003, 2005).
#### 4.1.1 SBS 0335$``$052
SBS 0335$``$052 is a local galaxy ($`54\text{Mpc}`$) with $`\text{SFR}=1.7M_{}\text{yr}^1`$ (, Hunt et al.2001) and extremely low metallicity $`Z=1/41Z_{}`$. This galaxy is known to have an unusual IR SED and strong flux at N–MIR. It has a very young starburst ($`\text{age}5\text{Myr}`$) without significant underlying old stellar population (Vanzi et al., 2000). T03 have modeled the SED of SBS 0335$``$052 and reported a good agreement with the available observations at that time.
However, Houck et al. (2004) presented new data of the MIR SED by Spitzer, and reported a deviation of the model by a factor of two or three. Their observation indicated that SBS 0335$``$052 has even more FIR-deficient SED than ever thought. Hence, it is interesting to examine whether our present model can reproduce the extreme SED of this galaxy. We are also interested in the possibility to determine which of the two pictures, unmixed or mixed, is plausible, by a direct measurement of the SED.
We show the model SEDs for SBS 0335$``$052 in Figure 9. We have calculated the SED for $`r_{\mathrm{SF}}=10`$, 20, and 30 pc both for unmixed and mixed cases. The SFR is fixed to be $`1.7M_{}\text{yr}^1`$, and the age is $`10^{6.5}`$ yr. Solid squares are the observed data from ISO, while open squares represent the upper limits obtained from IRAS and ISO observations. Open triangles are the expected contribution of dust emission calculated by the recipe of Joy & Lester (1988). Filled triangles depict the measured SED of SBS 0335$``$052 by Spitzer IRS taken from Houck et al. (2004) in the MIR. The filled triangle at FIR is the estimate from the radio observation of Hunt et al. (2004), which is used by Houck et al. (2004). Details of the other observational data are found in Section 4 of T03.
In the FIR regime, our model SEDs are consistent with the strong constraint given by Houck et al. (2004), both for the unmixed and mixed cases. This is because our present model predicts a peak of the SED at shorter wavelengths than that of T03. At MIR, though we cannot give an excellent fit to the observed data, the model SEDs roughly agree with them, and the unmixed-case SEDs with $`r_{\mathrm{SF}}=10\text{}20`$ pc give a better fit. The very strong N–MIR continuum of SBS 0335$``$052 is well reproduced by the SED of the unmixed dust production picture. On the contrary, the SEDs of the mixed case seriously underpredicts the observationally suggested N–MIR continuum. Since we can also calculate the extinction value $`A_\lambda `$, we can distinguish between a mixed-case SED and heavily extinguished unmixed SED without ambiguity.
In summary, the SED for the unmixed dust production with $`r_{\mathrm{SF}}=10\text{}20`$ pc yields a reasonable fit to the latest observations by Spitzer. This result suggests that we may determine the dust production (unmixed or mixed) of SNe through the observation of the N–MIR SEDs of forming galaxies. As for SBS 0335$``$052, the unmixed dust production is suggested to take place. In a previous work, we have modeled a high-$`z`$ ($`z=6.2`$) extinction curve (Maiolino et al., 2004) and found that it is well represented by dust produced in unmixed SNe (Hirashita et al., 2005). This conclusion is strengthened by our result above.
The dust mass calculated by our model at this age of SBS 0335$``$052 is $`1\text{}2\times 10^3M_{}`$, consistent with the observationally estimated value by Dale et al. (2001). We note that the mass estimation is strongly dependent on the assumed dust species and their emissivities, and grain size distribution. As we discussed in Section 3, the continuum radiation in the N–MIR is dominated by stochastically heated dust emission, which is completely different from modified blackbody. Therefore, when we try to estimate the dust mass, we must take care to determine the corresponding grain properties, i.e., radiative processes and grain which are related to the observed SED of galaxies.
#### 4.1.2 I Zw 18
I Zw 18 is the most metal-deficient star-forming galaxy in the local universe ($`Z=1/50Z_{}`$). We adopt its distance of 12.6 Mpc (Östlin, 2000), and the SFR of $`0.04M_{}\text{yr}^1`$. The existence of the underlying old stellar population is suggested for this galaxy, but their contribution is not dominant (, Hunt et al.2003). Thus, it is not unreasonable to adopt our present model to predict the IR SED of I Zw 18.
Cannon et al. (2002) estimated the total dust mass $`M_{\mathrm{dust}}=2\text{}3\times 10^3M_{}`$ from HST WFPC2 narrow-band imaging. In our model, the dust mass reaches this value at burst age of $`10^{7.25}\text{}10^{7.5}`$ yr. This age is consistent with the observationally suggested age of its major SF (e.g., , Hunt et al.2003). Cannon et al. (2002) obtained the extinction of $`A_V=0.5`$ mag for some patches in this galaxy, which is reproduced by our model with the above age.
We show our model prediction for the SED of I Zw 18 in Figure 10. We present the SED for the age $`10^{7.25}`$ yr and $`10^{7.5}`$ yr. The NIR flux measurements are taken from (Hunt et al.2003). I Zw 18 is not detected at all the four IRAS bands, hence we calculated the upper limits in the M–FIR from the sensitivity limits of IRAS observation. At $`850\mu `$m, the upper limit is converted from the data reported by (Hunt et al.2005).
The IR luminosity is slightly smaller than the previous prediction of T03. We also find that the FIR peak wavelength is located at $`\lambda 60\mu `$m, shorter than that of the T03 model SED. In addition, the peak intensity is larger for the mixed model, since the mixed model produce about 30 % larger amount of dust than the unmixed model does (see Fig. 2). Since the system still remains optically thin at this age because of the large $`r_{\mathrm{SF}}`$, this difference is directly reflected to the difference of the peak intensity.
The low opacity of I Zw 18 also leads to a low flux at N–MIR in the model SED in Figure 10 both for unmixed and mixed cases compared with the T03 model SED. As mentioned by T03, the NIR continuum of I Zw 18 is expected to be dominated by stellar and nebular continuum radiation, and the contribution of dust may be negligible.
We expect that these features will be found commonly in young galaxies if they are optically thin in the IR. Hence, it may be useful to obtain some suggestions for the observational strategy of young galaxies at high-$`z`$. We will discuss a well known real sample of such population of galaxies, Lyman-break galaxies, in the next subsection.
### 4.2 Lyman-break galaxies
Lyman-break galaxies (LBGs) are one of the most well-studied categories of high-$`z`$ star-forming galaxies (e.g., Steidel et al., 1999, 2003). Even in LBGs, there is clear evidence that they contain non-negligible amount of dust (e.g., Adelberger & Steidel, 2000; Calzetti, 2001). A high dust temperature ($`70`$K) is suggested by subsequent studies (Ouchi et al., 1999; Chapman et al., 2000; Calzetti, 2001; Sawicki, 2001).
In order to make a consistent picture of the dust emission from LBGs, T04 applied the T03 model and made various predictions for these galaxies. They also considered a power-law dust size distribution to examine its effect on the resulting SEDs. The T04 model reproduced the known observational properties of the dust emission from LBGs.
In this paper, we investigate the expected appearance of the LBGs with an improved dust grain formation of N03.
#### 4.2.1 Evolution of the SED of LBGs
We set the input parameters of the SED model for LBGs as follows. The SFR of LBGs spreads over the range of $`\text{SFR}1\text{}300M_{}\text{yr}^1`$ with a median of $`\text{SFR}20M_{}\text{yr}^1`$ (e.g., Erb et al., 2003). A constant SFR up to the age of $`10^9`$ yr is suggested to be a good approximation (e.g., Baker et al., 2004). Thus, the basic framework of the T03 model is also valid for LBGs. In this work, we consider the moderate case of $`\text{SFR}=30M_{}\text{yr}^1`$ over the age of $`10^{6.5}\text{}10^8`$ yr.
The most important information to calculate the IR SED is the effective size of the star forming region, but it is the most uncertain quantity (see discussion of T04). Since the mean half-light radius of LBGs is estimated to be $`1.6\text{kpc}`$ from HST observations (Erb et al., 2003), we use the galaxy radius as the radius of a star-forming region, and set $`r_{\mathrm{SF}}=2\text{kpc}`$ according to T04.
The SED evolution of LBGs is shown in Figure 11. Note that the scale in the ordinate of Figure 11 is different from those in Figures 4 and 5. Aside from luminosity, the behavior of the SED evolution of LBG is qualitatively very similar to that of a dwarf-like forming galaxy with $`r_{\mathrm{SF}}=100`$ pc (i.e., a weak NIR continuum and a strong M–FIR emission is commonly seen in both). This is because both the UV photon number density and dust number density are in the same order for an LBG and dwarf galaxy ($`r_{\mathrm{SF}}=100`$ pc). Since the IR luminosity depends on the dust column density, the total luminosity is different, but the shape of the SED is determined by the balance between the densities of UV photons and dust grains, they are similar with each other as for the shape.
The difference between the SEDs of the unmixed and mixed cases is small, and they give almost the same result especially in the M–FIR. This is explained as follows: the most prominent difference between the two is the N–MIR continuum, which is produced mainly by stochastic heating of grains. If the UV photon density is low, they are almost always in the lowest temperature state because the incidence of a UV photon is less frequent. In this case the N–MIR continuum from stochastically heated dust does not contribute the continuum significantly, and the global SED shape is determined by larger grains in the equilibrium with the ambient UV radiation field. Under such condition, they are similar in shape with each other. Thus, we see that it will be difficult to determine which scenario of dust production is correct from the observation of LBGs.
#### 4.2.2 Observability of the dust emission from LBG
The flux density of a source at observed frequency, $`\nu _{\mathrm{obs}}`$ is obtained by
$`S_{\nu _{\mathrm{obs}}}={\displaystyle \frac{(1+z)L_{(1+z)\nu _{\mathrm{obs}}}}{4\pi d_\mathrm{L}(z)^2}}={\displaystyle \frac{(1+z)L_{\nu _{\mathrm{em}}}}{4\pi d_\mathrm{L}(z)^2}},`$ (3)
where $`d_\mathrm{L}(z)`$ is the luminosity distance corresponding to a redshift $`z`$, and $`\nu _{\mathrm{obs}}`$ and $`\nu _{\mathrm{em}}`$ are observed and emitted frequency, respectively.
We show the observed IR/submm SEDs of LBGs at $`z=2`$, 3, and 4 in Figure 12. For simplicity we only show the SED with the age of $`10^8`$ yr. The thick black short horizontal lines indicate the 3-$`\sigma `$ detection limits for 8-hour observation by ALMA (Atacama Large Millimeter Array).<sup>7</sup><sup>7</sup>7 URL: http://www.alma.info/. Here we assumed 64 antennas and three wavelength bands, 450, 850, and $`1080\mu `$m. We also show the 3-$`\sigma `$ source confusion limit of Herschel<sup>8</sup><sup>8</sup>8 URL: http://www.rssd.esa.int/herschel/. at 75, 160, 250, and 350$`\mu `$m bands by thick gray horizontal lines. These limits are based on ‘the photometric criterion’ of (, Lagache et al.2003). See also (Ishii et al.2002); Takeuchi & Ishii (2004a).
As discussed in T04, the detectability of LBGs is not strongly dependent on their redshifts. The predicted IR luminosity is factor of two smaller than that of T04, hence it is more difficult to detect LBGs than the discussion in T04: we need roughly four times longer integration time than that derived from T04, if we fix all the other conditions. Detection at $`350\mu `$m seems impossible for moderate-SFR LBGs. However at longer wavelengths, if the age $`10^8\mathrm{yr}`$ and $`\text{SFR}30M_{}\mathrm{yr}^1`$, LBGs can be detected at a wide range of redshifts in the submm by ALMA deep survey. In the FIR, Herschel will detect the dust emission from LBGs at $`z2`$, but difficult at higher-$`z`$. Since the SED does not differ for the unmixed or mixed dust formation, our evaluation of the observability of LBGs holds both for the mixed and unmixed dust formation.
### 4.3 Toward higher redshifts
Based on the above discussions, we give a brief consideration on the observation of very high-$`z`$ galaxies ($`z5`$) here. Direct observation of such galaxies are of vital importance to explore the physics of galaxy formation.
First, assume a LBG at these redshifts. Figure 13 shows such a situation. We assumed the same physical parameters for LBGs. From Figure 13, we find that LBG-like objects can be detectable even at $`z10`$ if $`\text{SFR}30M_{}\text{yr}^1`$. At $`z20`$, it may be more difficult, but might be possible to detect LBGs with very high SFR, by an ALMA ultra-deep survey with much longer integration time than eight hour.
However, it may not be reasonable to assume a galaxy like LBGs at $`z10`$ because they are rather massive system in general ($`M_{\mathrm{star}}10^{10}M_{}`$), and, from the modern cosmological viewpoint, such a massive system is very rare at such a high-$`z`$, young universe.
Here we discuss an object less massive than LBGs. In modern hierarchical structure formation scenarios, it would be more reasonable to assume a small, subgalactic clump as a first forming galaxy. Consider a dark halo of mass $`10^9M_{}`$, then it is expected to contain a gas with mass $`10^8M_{}`$. For this purpose, we calculate the SEDs for a dwarf star-forming galaxy. If gas collapses on the free-fall timescale with an efficiency of $`ϵ_{\mathrm{SF}}`$ (we assume $`ϵ_{\mathrm{SF}}=0.1`$), we obtain the following evaluation of the SFR Hirashita & Hunt (2004):
$`\text{SFR}0.1\left({\displaystyle \frac{ϵ_{\mathrm{SF}}}{0.1}}\right)\left({\displaystyle \frac{M_{\mathrm{gas}}}{10^7M_{}}}\right)\left({\displaystyle \frac{n_\mathrm{H}}{100\text{cm}^3}}\right)^{1/2}[M_{}\text{yr}^1],`$ (4)
and
$`n_\mathrm{H}100\left({\displaystyle \frac{r_{\mathrm{SF}}}{100\text{pc}}}\right)^3\left({\displaystyle \frac{M_{\mathrm{gas}}}{10^7M_{}}}\right)[\text{cm}^3],`$ (5)
where $`n_\mathrm{H}`$ is the hydrogen number density. Then we have
$`\text{SFR}0.1\left({\displaystyle \frac{ϵ_{\mathrm{SF}}}{0.1}}\right)\left({\displaystyle \frac{M_{\mathrm{gas}}}{10^7M_{}}}\right)^{3/2}\left({\displaystyle \frac{r_{\mathrm{SF}}}{100\text{pc}}}\right)^{3/2}[M_{}\text{yr}^1].`$ (6)
If we consider $`M_{\mathrm{gas}}10^8M_{}`$, we have $`\text{SFR}3(r_{\mathrm{SF}}/100\text{pc})^{3/2}M_{}\text{yr}^1`$. In addition, as we see below, an extremely high-$`z`$ galaxy observed by HST has a very compact morphology (Kneib et al., 2004). We also mention that, from a theoretical side, high-$`z`$ galaxies are suggested to be dense and compact compared to nearby galaxies (Norman & Spaans, 1997; Hirashita & Ferrara, 2002). Hence, it may be reasonable to assume a same type of galaxy as local dwarfs discussed in Section 4.2.1. Thus, we consider a dwarf galaxy with $`\text{SFR}=10M_{}\text{yr}^1`$ as an example, and we adopt $`r_{\mathrm{SF}}=30`$ pc and 100 pc. The age is set to be $`10^7`$ yr. If the age is older, they will become easier to detect if a constant SFR takes place. We show the expected SEDs for such galaxies at $`z=5`$, 10, and 20 in Figures 14 and 15. As expected, it seems almost impossible to detect such objects by Herschel or ALMA.
There is, however, a hope to observe such a small forming galactic clump directly: gravitational lensing works very well as a natural huge telescope. For example, MS 1512$``$cB58 is an ideal case of a lensed LBG (Nakanishi et al., 1997; Baker et al., 2004), and more recently, a small star-forming galaxy has been detected by HST and Spitzer observations (Kneib et al., 2004; Egami et al., 2005). Hence, we can expect a forming galaxy which is intrinsically small and faint but magnified by a gravitational potential of a cluster of galaxies. If we assume a lens magnification factor of 40, such a small galaxy becomes detectable. This is depicted by the thin lines in Figures 14 and 15. Since the expected SED of such a compact dwarf galaxy has a strong MIR continuum at their rest frame, it can be feasible to detect at the FIR in the observed frame. Indeed, at $`z=5`$ they appear above the confusion limits of Herschel. Since the confusion limit of Herschel may be, in fact, very difficult to reach in actual observations, a cooled FIR space telescope is more suitable for such observation, and this will be a strong scientific motivation for a future project like SPICA.<sup>9</sup><sup>9</sup>9URL: http://www.ir.isas.jaxa.jp/SPICA/index.html. In the FIR wavelength regime ($`50\text{}200\mu \text{m}`$) , the SEDs of unmixed and mixed cases might be distinguishable. If a FIR spectroscopy is performed, which picture of dust production is plausible might be examined. At higher-$`z`$, they can be detectable by ALMA survey. Even at $`z20`$, they can be detected by a standard 8-hour survey of ALMA, if a lensing takes place. It is interesting to note that, at $`z10`$, the peak of their SED locates at the wavelength range where the detection limits of both Herschel and ALMA are not very good. A new effective (probably space) facility should be developed to overcome this difficulty in observation.
For this line of study, we must estimate how frequently such lensing events occur for high-$`z`$ objects. Suppose a cluster of galaxies at $`z_\mathrm{l}0.1\text{}0.2`$ whose dynamical mass $`M_{\mathrm{dyn}}`$ is $`5\times 10^{14}M_{}`$ and whose mass distribution obeys the singular isothermal sphere. We denote the strong lensing cross section, i.e., the area of the region in the source plane for which the resulting magnification by a cluster is larger than $`\mu `$, as $`\sigma (>\mu )`$. Perrotta et al. (2002) presented $`\sigma (>\mu )`$ as a function of $`M_{\mathrm{dyn}}`$ for $`z_{\mathrm{lens}}=1.0`$. Since $`\sigma (>\mu )D_{\mathrm{ls}}^{}{}_{}{}^{2}`$ \[$`D_{\mathrm{ls}}`$ is the angular-diameter distance between the lens and the source (e.g., Asada, 1998)\] (Covone, Sereno, & de Ritis, 2005), we can convert their result to our condition and obtain $`\sigma (>10)30\text{arcsec}^2`$ on the source plane.<sup>10</sup><sup>10</sup>10On the image plane, the area corresponds to $`\mu \sigma (>\mu )`$. This result is almost independent of the source redshifts. Setting the limiting flux density $`S_\nu =1\mu `$Jy and using the number counts of Hirashita & Ferrara (2002) for galaxies at $`z>5`$, we have an expected number of galaxies suffering a strong lensing to be $`1\text{}3`$. Thus, we expect at least a few strongly lensed IR galaxies to this survey depth.
To have a concrete idea for the observational strategy, we must consider an important aspect of the lensing. It is known that the gravitational lensing affects the galaxy number counts in two opposite directions: i) it magnifies the flux density of galaxies, and ii) it stretches the sky area and decreases their number density on the sky. This is called the ‘magnification bias’ (Fort, Mellier, & Dantel-Fort, 1997; Bézecourt, Pelló, & Soucail, 1998). The unlensed number count $`N_0(>S_\nu )`$ is generally approximated by a power law
$`N_0(>S_\nu )S_{\nu }^{}{}_{}{}^{\beta }.`$ (7)
Then, by denoting the magnification factor as $`\mu `$, we obtain the lensed number counts $`N_{\mathrm{lens}}(>S_\nu )`$ as
$`N_{\mathrm{lens}}(>S_\nu )=N_0(>S_\nu )\mu ^{(\beta +1)}`$ (8)
(e.g., Fort, Mellier, & Dantel-Fort, 1997; , 2000). Equation (8) shows that whether the number counts increases or decreases depends on the slope of the counts, $`\beta `$: if $`\beta <1`$ lensing works as an enhancement of the counts. Since the slope of the number counts of such galaxies is expected to be steeper than $`1`$ (e.g., Hirashita & Ferrara, 2002), the source density on the image plane will increase (see also Perrotta et al., 2002). Thus, we are confident that the lensing will work as a really useful tool to detect forming small galaxies with dust emission.
## 5 Conclusion
Dust plays various important roles even in the very early phase of galaxy evolution (e.g., Hirashita & Ferrara, 2002). In such a young phase with a typical age less than $`10^9`$ yr, dust is predominantly supplied by SNe. With the aid of a new physical model of dust production by SNe developed by N03, we constructed a model of dust emission from a very young galaxies according to T03.
N03 carefully took into account the radial density profile and the temperature evolution in the calculation of the dust formation in the ejecta of SNe II and PISNe. They also showed that the produced dust species depends strongly on the mixing within SNe. We treated both unmixed and mixed cases and calculated the IR SED of young galaxies for both cases.
The SEDs constructed from N03 dust production are less luminous than those by T03 model by a factor of 2–3. This difference is due to the improvement in the treatment of $`Q(a,\lambda )`$ at UV and the considered grain species. The SED for the unmixed case is found to have a strong N–MIR continuum radiation in its early phase of the evolution ($`\text{age}10^{7.25}`$yr) compared with that for the mixed case. The N–MIR continuum is due to the emission from Si grains, which only exist in the species of the unmixed dust production.
We also calculated the IR extinction curves for young galaxies. N03 dust gives a weaker extinction than that of T03 model because of the small relative number of very small dust grains. This is also the cause of the smaller IR luminosity of the present model. For the unmixed case, NIR extinction is dominated by large grain of Si and amorphous carbon, and silicate features are less prominent compared to the curve given by T03. On the contrary, the extinction curve of the mixed case has a similar shape with that of T03.
The SED of a local starbursting dwarf galaxy, SBS 0335$``$052, was calculated. Recent Spitzer observations (Houck et al., 2004) have implied a hotter dust than previously thought. Our present model SED naturally reproduces the strong N–MIR continuum and the lack of cold FIR emission of SBS 0335$``$052. We found that only the SED of unmixed case can reproduce the N–MIR continuum of this galaxy. Hence, as for SBS 0335$``$052, the unmixed dust production is preferred. It will be interesting to proceed this line of study for higher-$`z`$ galaxies.
A prediction for the SED of another typical nearby star-forming dwarf galaxy, I Zw 18, was then made. Since I Zw 18 is dominated by very young star formation, we may adopt the model to this galaxy. Using the dust species from N03 with this model, we find a weaker FIR emission than that of T03. The N–MIR continuum is also expected to be much weaker than that of T03 SED.
We also calculated the evolution of the SED of LBGs. For the parameters of LBGs, the unmixed and mixed picture does not affect the appearance of the SED. Hence, if we combine our model with the knowledge obtained from optical observations, the observability of LBGs at submillimetre wavelengths are robust conclusion independent of the details of the SNe dust production theory.
Finally, we considered the observations of forming galaxies at $`z5`$. If there exist LBG-like galaxies at these redshifts, they can be detected at $`z10`$ by ALMA 8-hour survey if $`\text{SFR}30M_{}\text{yr}^1`$. For small forming galaxies or subgalactic clumps, it is almost impossible to detect their intrinsic flux by ALMA or by Herschel. However, the gravitational lensing is found to be a very effective tool to detect such small star-forming galaxies at $`z5`$. If we consider a compact dwarf galaxy with $`\text{SFR}10M_{}\text{yr}^1`$ and $`r_{\mathrm{SF}}30\text{pc}`$, we can expect a few strongly magnified galaxies behind a typical cluster of galaxies at $`z0.1\text{}0.2`$.
## Acknowledgments
First we are grateful to the anonymous referee for many insightful comments which have improved the clarity of this paper much. We deeply thank Akio K. Inoue, Jean-Paul Kneib and Giovanni Covone for illuminating discussions. We made extensive use of the NASA Astrophysics Data System. TTT, TTI, and HH have been supported by the Japan Society of the Promotion of Science. TK is supported by a Grant–in–Aid for Scientific Research from JSPS (16340051).
## Appendix A Details of the chemical evolution
Here we present a detailed description of the chemical evolution we used in this work. Full treatment of the present chemical evolution model including the feedback, dust destruction, and more complex SF history is given elsewhere (Nozawa et al., in preparation). Within the framework of the closed-box model, the total baryonic mass of a galaxy, $`M_\mathrm{T}`$, is conserved as
$`M_\mathrm{T}=M_{\mathrm{gas}}(0)=M_{\mathrm{gas}}(t)+M_{\mathrm{dust}}(t)+M_{\mathrm{star}}(t)+M_{\mathrm{rem}}(t),`$ (9)
where $`M_{\mathrm{gas}}(t)`$ is the gas mass, $`M_{\mathrm{dust}}(t)`$ is the dust mass, $`M_{\mathrm{star}}(t)`$ is the stellar mass, and $`M_{\mathrm{rem}}(t)`$ is the mass of stellar remnants in the galaxy at an age of $`t`$. We start the calculation from homogeneous pristine gas, i.e., $`M_{\mathrm{dust}}(0)=M_{\mathrm{star}}(0)=M_{\mathrm{rem}}(0)=0`$. The time evolution of the mass of interstellar medium, $`M_{\mathrm{ISM}}(t)M_{\mathrm{gas}}(t)+M_{\mathrm{dust}}(t)`$ is given by
$`{\displaystyle \frac{dM_{\mathrm{ISM}}}{dt}}|_t=\psi (t)+{\displaystyle _{m_\mathrm{l}}^{m_\mathrm{u}}}\left[mm_{\mathrm{rem}}(m)\right]\psi (t\tau _m)\varphi (m)𝑑m,`$ (10)
where $`\psi (t)`$ is the SFR at an age $`t`$, $`\varphi (m)`$ is the IMF normalized to unity in the mass interval $`[m_\mathrm{l},m_\mathrm{u}]`$ (subscripts l and u mean lower and upper mass limit, respectively), $`m_{\mathrm{rem}}(m)`$ is the mass of a single stellar remnant resulting from a progenitor star with mass $`m`$, and $`\tau _m`$ is the lifetime of a star with mass $`m`$.
For the IMF, as we mentioned in the main text, we adopt the Salpeter IMF with mass range of $`(m_\mathrm{l},m_\mathrm{u})=(0.1M_{},100M_{})`$. The lifetime $`\tau _m`$ is evaluated by a fitting formula of zero-metallicity tracks with no mass loss, given by Schaerer (2002). For SNe II (progenitor mass $`8\text{}40M_{}`$), we approximate the remnant mass $`m_{\mathrm{rem}}`$ by the linear fit based on the numerical result of SN explosions given by Umeda & Nomoto (2002):
$`m_{\mathrm{rem}}(m)=0.06m+0.93(8M_{}<m<40M_{}).`$ (11)
For other progenitor mass ranges, we set $`m_{\mathrm{rem}}(m)=0`$ at $`m<8M_{}`$ (no remnants), and $`m_{\mathrm{rem}}(m)=m`$ at $`40M_{}<m<100M_{}`$ (all the mass is swallowed by a black hole). In this work, we do not treat PISNe in the calculation. By Equation (11), the time evolution of $`M_{\mathrm{rem}}(t)`$ is expressed by
$`{\displaystyle \frac{dM_{\mathrm{rem}}}{dt}}|_t={\displaystyle _{m_\mathrm{l}}^{m_\mathrm{u}}}m_{\mathrm{rem}}(m)\psi (t\tau _m)\varphi (m)𝑑m.`$ (12)
The time evolution of $`M_{\mathrm{dust}}(t)`$ is, then, obtained as
$`{\displaystyle \frac{dM_{\mathrm{dust}}}{dt}}|_t={\displaystyle \frac{M_{\mathrm{dust}}(t)}{M_{\mathrm{ISM}}(t)}}\psi (t)+{\displaystyle _{m_\mathrm{l}}^{m_\mathrm{u}}}m_{\mathrm{dust}}(m)\psi (t\tau _m)\varphi (m)𝑑m`$ (13)
where $`m_{\mathrm{dust}}(m)`$ denotes the dust mass produced by a progenitor with mass $`m`$. We apply the dust formation model of N03 for $`m_{\mathrm{dust}}`$. As we mentioned in §§2.1, we used the result of progenitor mass $`20M_{}`$ for the fraction and the size distribution of dust grain species, hence the mass fraction and size distribution are constant in time along with the evolution.
## Appendix B Normalization of the number density of grain species
In this Appendix, we show how to obtain the normalization of the dust size distribution $`dN_i/da_i`$,
$`{\displaystyle \frac{dN_i}{da}}daAf_i(a)da`$ (14)
where subscript $`i`$ denotes the species of dust, and $`f_i`$ is the mass fraction of dust of species $`i`$, given by N03, and $`A`$ is the normalization constant. We denote the material density of $`i`$-species dust by $`\rho _i`$, and the SF region radius as $`r_{\mathrm{SF}}`$. The total mass of dust species $`i`$, $`M_i`$, is then written as
$`M_i`$ $`=`$ $`\rho _i{\displaystyle \frac{4\pi a^3}{3}\frac{dN_i}{da}𝑑a}=\rho _i{\displaystyle \frac{4\pi a^3}{3}Af_i(a)𝑑a}`$ (15)
$`=`$ $`{\displaystyle \frac{4\pi A}{3}}\rho _i{\displaystyle a^3f_i(a)𝑑a}.`$
The total dust mass of all the species, $`M`$, is
$`M={\displaystyle \underset{i}{}}{\displaystyle \frac{4\pi A}{3}}\rho _i{\displaystyle f_i(a)a^3𝑑a}={\displaystyle \frac{4\pi A}{3}}{\displaystyle \underset{i}{}}\rho _i{\displaystyle f_i(a)a^3𝑑a}.`$ (16)
Hence
$`A={\displaystyle \frac{3M}{4\pi {\displaystyle \underset{i}{}}\rho _i{\displaystyle f_i(a)a^3𝑑a}}}.`$ (17)
Thus we can determine $`A`$ from the available data. Note that only $`M`$ depends on time $`t`$, hence the time dependence of $`A`$ is given only through $`M`$. The total number of grains, $`N_i`$, is expressed as
$`N_i={\displaystyle \frac{dN_i}{da_i}𝑑a_i}.`$ (18)
|
warning/0506/cs0506052.html
|
ar5iv
|
text
|
# Comments on ‘Bit Interleaved Coded Modulation’
## I Introduction
A comprehensive study of BICM is presented in . There, in addition to an information theoretic analysis of BICM, a detailed analysis of the probability of error is presented. In the error analysis of BICM in , various upper bounds and approximations to the probability of error are derived, notable among which is the expurgated bound. In the first half of this paper, counter examples are given for the two theorems in leading to the expurgated bound. Consequently, the validity of the expurgated bound in is questionable. The second half of this paper focuses on the important and practical case of square QAM constellations with Gray labeling. For such cases, an alternate upper bound is presented. Numerical results are given for $`16`$-QAM and $`64`$-QAM and a rate-$`\frac{1}{2}`$ convolutional code. For these examples, the new bound is nearly equal to, and slightly tighter than, the expurgated bound of . The reader is referred to \[1, Sections $`2`$ and $`4`$\] for notation.
## II Two counter examples
Counter example to \[1, Theorem 1\]: Consider the constellation in Figure 1. It is similar to QPSK, except the point $`z^{(1)}`$ which is on the unit circle but closer to one of its neighbors. The labeling $`\mu `$ is chosen to be Gray labeling. For simplicity, assume $`d=1`$. For concreteness, assume that the message to be sent is $`b=0`$, and that it is to be sent over the first label position, i.e. $`S=1`$. Further, let $`U`$ be $`0`$ so that the signal labels are not complemented. Then, an element of $`𝒳_0^1=\{00,01\}`$ is transmitted. Suppose $`𝐱=00`$ is transmitted. Let $`𝐳=11`$. As in , let $`\mathrm{\Gamma }_{\underset{¯}{𝐱},\underset{¯}{𝐳}}=\{\underset{¯}{𝐲}:\underset{¯}{𝐲}\underset{¯}{𝐱}_2^2\underset{¯}{𝐲}\underset{¯}{𝐳}_2^2\}`$. Clearly $`\mathrm{\Gamma }_{𝐱,𝐳}\mathrm{\Gamma }_{𝐱,𝐳^{(1)}}\mathrm{\Gamma }_{𝐱,𝐳^{(2)}}`$ where $`𝐳^{(1)}=01`$, $`𝐳^{(2)}=10`$. So, by \[1, Theorem 1\], $`\mathrm{\Gamma }_{𝐱,𝐳}`$ can be neglected in the union bound. This leaves only $`\mathrm{\Gamma }_{𝐱,𝐳^{(2)}}`$ in the union bound, but that fails to cover some part of the pairwise error region. In particular, referring to Figure 1, we see that the darkly shaded region is left out of the bound on $`P(\text{decoder error}|𝐱=00)`$ though it is part of the pairwise error region. Similarly, the lightly shaded region is left out of the bound on $`P(\text{decoder error}|𝐱=01)`$. Thus, neglecting the term $`P(𝐱𝐳)`$ in the union bound alters the inequality. This disproves \[1, Theorem 1\]. The theorem may hold in the presence of stronger conditions on the constellation such as symmetry. We are unable to identify such sufficient conditions though.
The problem with the proof of \[1, Theorem 1\] lies in eliminating the sub-region $`\{\mathrm{\Gamma }_{\underset{¯}{\overset{^}{𝐱}},\underset{¯}{\overset{^}{𝐳}}}\mathrm{\Gamma }_{\underset{¯}{\overset{^}{𝐱}},\underset{¯}{𝐳}^{(i)}}\}`$ where $`\underset{¯}{𝐳}^{(i)}𝒳_{\underset{¯}{𝐜}}^{\underset{¯}{𝐒}}`$. Suppose $`\underset{¯}{𝐲}`$ is received. It is reasoned in that, if $`|\underset{¯}{𝐳}^{(i)}\underset{¯}{𝐲}|<|\underset{¯}{𝐱}\underset{¯}{𝐲}|`$, then $`\underset{¯}{𝐲}`$ should not be counted in the pairwise error region of $`(\underset{¯}{𝐱}\underset{¯}{𝐳})`$ since the decoder either chooses $`\underset{¯}{𝐳}^{(i)}`$ (and thus does not make an error) or the decoder makes an error which is already included at least once, in the term $`\underset{¯}{𝐳}^{(i)}\underset{¯}{\overset{^}{𝐳}}`$.
The right hand side of \[1, (30)\] can be viewed as an average over the transmitted signal sequence $`\underset{¯}{𝐱}`$, and a sum, or union bound, over $`\underset{¯}{𝐳}`$ in the bad signal subset $`𝒳_{\underset{¯}{\overset{^}{𝐜}}}^{\underset{¯}{𝐒}}`$. The decoder error region, for a given $`\underset{¯}{𝐜}`$, $`\underset{¯}{𝐒}`$ and $`\underset{¯}{𝐔}`$, is the same for all transmitted $`\underset{¯}{𝐱}`$, and can be denoted by:
$$\mathrm{\Gamma }_{\underset{¯}{𝐜},\underset{¯}{\overset{^}{𝐜}}}=\{\underset{¯}{𝐲}:\underset{\underset{¯}{𝐱}𝒳_{\underset{¯}{𝐜}}^{\underset{¯}{𝐒}}}{\mathrm{min}}\underset{¯}{𝐲}\underset{¯}{𝐱}^2\underset{\underset{¯}{𝐳}𝒳_{\underset{¯}{\overset{^}{𝐜}}}^{\underset{¯}{𝐒}}}{\mathrm{min}}\underset{¯}{𝐲}\underset{¯}{𝐳}^2\}$$
When the union bound in \[1, (30)\] is expurgated, it is not known whether the inequality remains valid unless all points in $`\mathrm{\Gamma }_{\underset{¯}{𝐜},\underset{¯}{\overset{^}{𝐜}}}`$ are covered in the expurgated union bound for each transmitted $`\underset{¯}{𝐱}`$. So, the region $`\mathrm{\Gamma }_{\underset{¯}{𝐜},\underset{¯}{\overset{^}{𝐜}}}`$ should be counted $`|𝒳_{\underset{¯}{𝐜}}^{\underset{¯}{𝐒}}|=2^{d(m1)}`$ times, each time being weighted by a probability measure depending on which $`\underset{¯}{𝐱}𝒳_{\underset{¯}{𝐜}}^{\underset{¯}{𝐒}}`$ was transmitted.
Counter example to \[1, Theorem 2\]: Consider 16QAM constellation with Gray labeling, as given in Figure 2. Let $`d=1`$, $`U=0`$ and $`b=0`$, as before. Let the information be transmitted in $`S=2`$. So, some $`\widehat{𝐱}𝒳_0^2`$ is transmitted. Suppose $`\widehat{𝐱}=1011`$ is transmitted. Let $`\widehat{𝐳}=0111`$. Setting $`𝐳^{}=0010`$ and $`𝐳^{\prime \prime }=0001`$, it can be shown that the conditions of \[1, Theorem 2\] are satisfied. It follows from the theorem that $`P(\widehat{𝐱}\widehat{𝐳})`$ can be neglected in \[1, (30)\] without altering the inequality. It can be similarly concluded that all $`\widehat{𝐳}𝒳_1^2`$ except $`1111`$ can be neglected. Referring to Figure 2, it follows that the darkly shaded region is left out of the bound on $`P(\text{decoder error}|\widehat{𝐱}=1011)`$, though it is part of the pairwise error region. This occurs for all $`\widehat{𝐱}`$ of the form $`10`$ (where $``$ can be $`0`$ or $`1`$). Similarly, the lightly shaded region is left out of the bound on $`P(\text{decoder error}|\widehat{𝐱}=00)`$. Thus, the inequality does not hold and \[1, Theorem 2\] is disproved.
The extension to fading channels of the above theorem, namely, \[1, §IV.C Corollary 1\], uses stronger conditions than the above theorem. However, it is easy to see how the above example works as a counter example to the corollary as well.
The proof that $`f_{ex}(d,\mu ,𝒳)`$ is greater than or equal to $`f(d,\mu ,𝒳)`$ is hence not valid for the case of square QAM signal sets with Gray labeling. It is to be kept in mind that $`f_{ex}(d,\mu ,𝒳)`$ is presented as an upper bound to $`f(d,\mu ,𝒳)`$ in only for this case - for any other choice of modulation, it is presented as only an approximation.
## III Revised expurgated bound
For the case of square QAM signal sets with Gray labeling, a revised expurgated bound, denoted by $`f_{ex,new}`$, is derived. Two variants of $`f_{ex,new}`$, denoted by $`f_{ex,new}^I`$ and $`f_{ex,new}^{II}`$, are presented. First, consider the case when $`d=1`$. Fix a point $`𝐱`$ in the signal set, and a bit position $`i`$ in its label. Since Gray labeling is used, the bit value of the corresponding position remains the same across either the row or the column containing $`𝐱`$. So, if $`𝐱𝒳_b^i`$, then either the entire row or the entire column in the signal set belongs to $`𝒳_b^i`$. This reduces the problem of identifying regions contributing to the probability of error to a single dimensional problem. As in any such single dimensional problem, the regions contributing to the probability of error can be covered by choosing two neighbors, one on each side, and constructing a union bound with two PEP terms. This is in contrast to the original expurgated bound where only one neighbor is considered.
Suppose $`𝐱𝒳_b^i`$ is transmitted. Then, the original expurgated bound considers only the unique nearest neighbor in $`𝒳_b^i`$. An easy fix is to consider the nearest neighbors in $`𝒳_{1b}^i`$ on both sides of $`𝐱`$. This can be improved further by choosing the neighbors such that the PEP decision boundaries coincide with the actual decision boundaries. As an illustration, consider the constellation in Figure 2. As in the counter example to \[1, Theorem 2\], suppose $`d=1`$, $`U=0`$, $`S=2`$ and $`b=0`$. Some $`𝐱𝒳_0^2`$ is transmitted – let $`1011`$ be transmitted. Since all the points in the column of $`𝐱`$ belong to $`𝒳_0^2`$, the decision boundaries are vertical. The original expurgated bound includes only the unique nearest neighbor in $`𝒳_1^2`$, namely $`1111`$. The easy fix is to include the nearest neighbor in $`𝒳_1^2`$ on the other side of $`𝐱`$ also - namely $`0111`$. This variant of $`f_{ex,new}`$ shall be referred to as $`f_{ex,new}^I`$. Alternatively, the signal set can be extended in a lattice fashion, as shown by triangles in Figure 2), and two points in the (extended) signal set can be identified such that the PEP decision boundaries coincide with the actual decision boundaries. For this example, the two points then are $`1111`$ (as before) and $`A`$ (instead of $`0111`$). This variant of $`f_{ex,new}`$ shall be referred to as $`f_{ex,new}^{II}`$.
The above two methods have their relative merits and demerits. While both of the above methods yield upper bounds to $`f(1,\mu ,𝒳)`$ (this follows from a union bound argument), the bounds obtained using the second method will clearly be tighter than the first. However, specifying the two points is more straightforward in the first method.
In either way, there are two points in $`𝒳_{1b}^i`$ corresponding to each $`𝐱𝒳_b^i`$. Let them be denoted by $`𝐳_1(𝐱)`$ and $`𝐳_2(𝐱)`$ when referring to $`f_{ex,new}`$, by $`𝐳_1^I(𝐱)`$ and $`𝐳_2^I(𝐱)`$ when referring to the variant $`f_{ex,new}^I`$, and similarly by $`𝐳_1^{II}(𝐱)`$ and $`𝐳_2^{II}(𝐱)`$ when referring to $`f_{ex,new}^{II}`$. Then, $`f_{ex,new}`$ can be defined by:
$`f_{ex,new}(1,\mu ,𝒳)=`$
$`{\displaystyle \frac{1}{m2^m}}{\displaystyle \underset{S,U}{}}{\displaystyle \underset{𝐱𝒳_c^S}{}}(P(𝐱𝐳_1(𝐱))+P(𝐱𝐳_2(𝐱)))`$ (1)
Similar definitions hold for $`f_{ex,new}^I`$ and $`f_{ex,new}^{II}`$. In some cases, when a point in $`𝒳_b^i`$ is transmitted, there may be no points belonging to $`𝒳_{1b}^i`$ on one side. For instance, suppose some point in $`𝒳_0^1`$ such as $`0001`$ is transmitted. Then, all points in $`𝒳_1^1`$ are to one side of the transmitted point. In such a case, $`𝐳_1(.)`$ is set in the usual manner by choosing from these points, and $`𝐳_2(.)`$ is set to a special symbol $`\mathrm{}`$, with the understanding that the PEP $`P(𝐱\mathrm{})=0`$ for all $`𝐱`$ in the constellation. Here, $`\mathrm{}`$ has the interpretation of a point in the extended constellation at infinite distance from the regular points of the constellation.
The above methods are now extended to the case $`d>1`$ as follows. Suppose $`\underset{¯}{𝐱}𝒳_{\underset{¯}{𝐜}}^{\underset{¯}{𝐒}}`$ where $`\underset{¯}{𝐜}=(c_1,\mathrm{},c_d)`$, $`\underset{¯}{𝐒}=(i_1,\mathrm{},i_d)`$ and $`\underset{¯}{𝐱}=(𝐱_1,\mathrm{},𝐱_d)`$. Each $`𝐱_l𝒳_{c_l}^{i_l}`$ for $`1ld`$. Let the set $`Z_{\underset{¯}{𝐱}}`$ consist of sequences of length $`d`$, where the $`l^{\text{th}}`$ element is either $`𝐳_1(𝐱_l)`$ or $`𝐳_2(𝐱_l)`$. Clearly, $`|Z_{\underset{¯}{𝐱}}|=2^d`$. For any two signal sequences, $`\underset{¯}{𝐱}`$ and $`\underset{¯}{𝐳}`$, the PEP $`P(\underset{¯}{𝐱}\underset{¯}{𝐳})`$ is set to zero if any element in $`\underset{¯}{𝐳}`$ is $`\mathrm{}`$. Define
$$f_{ex,new}(d,\mu ,𝒳)=\frac{1}{m^d2^{md}}\underset{\underset{¯}{𝐒},\underset{¯}{𝐔}}{}\underset{\underset{¯}{𝐱}𝒳_{\underset{¯}{𝐜}}^{\underset{¯}{𝐒}}}{}\underset{\underset{¯}{𝐳}Z_{\underset{¯}{𝐱}}}{}P\left(\underset{¯}{𝐱}\underset{¯}{𝐳}\right)$$
(2)
Since $`𝒳_{\underset{¯}{𝐜}}^{\underset{¯}{𝐒}}`$ is a product set, the union bound arguments developed for the case when $`d=1`$ readily extend to the case when $`d>1`$ to yield the following upper bound.
$$f(d,\mu ,𝒳)f_{ex,new}(d,\mu ,𝒳)$$
(3)
A computationally efficient form of $`f_{ex}`$ is derived in \[1, (48)\]. The revised expurgated bound $`f_{ex,new}`$ can be expressed in a similar form, with $`\psi _{ex}(s)`$ replaced by $`\psi _{ex,new}(s)`$ given by
$$\frac{1}{m2^m}\underset{i=1}{\overset{m}{}}\underset{b=0}{\overset{1}{}}\underset{𝐱𝒳_b^i}{}\left\{\varphi _{\mathrm{\Delta }(𝐱,𝐳_1(𝐱))}(s)+\varphi _{\mathrm{\Delta }(𝐱,𝐳_2(𝐱))}(s)\right\}$$
The asymptotic behavior of $`f_{ex}`$ at large SNR ($`\sigma <<1`$) in the presence of fading is given in \[1, (62)\]. A similar expression can be derived for $`f_{ex,new}`$ with $`d_h`$ replaced by $`d_{h_c}`$, where $`d_{h_c}^2`$ is given by
$$\frac{1}{m2^m}\underset{i=1}{\overset{m}{}}\underset{b=0}{\overset{1}{}}\underset{𝐱𝒳_b^i}{}\left\{\frac{1}{|𝐱𝐳_1(𝐱)|^2}+\frac{1}{|𝐱𝐳_2(𝐱)|^2}\right\}$$
In the design guidelines listed in \[1, §V\], the harmonic mean square distance $`d_h^2`$ should be substituted with $`d_{h_c}^2`$. As an illustration, Table I of is given here with the revised harmonic mean square distances. Here, $`d_{h_c}^{I}{}_{}{}^{2}`$ corresponds to $`f_{ex,new}^I`$, and $`d_{h_c}^{II}{}_{}{}^{2}`$ corresponds to $`f_{ex,new}^{II}`$.
## IV Numerical results
In this section, the original expurgated bound is compared numerically with the two versions of the revised expurgated bound for a Rayleigh fading channel ($`K=0`$) with full CSI at the receiver. The modulation schemes considered are 16QAM and 64QAM with Gray labeling. The binary code used is the standard rate-$`1/2`$, $`64`$-state binary convolutional code with generators (o$`133`$, o$`171`$) used in (also given in \[2, pp. 507\] as the (o$`634`$, o$`564`$) code). This code has a minimum distance $`d_2=10`$. The revised versions of the expurgated bound are numerically evaluated on the same lines as the original expurgated bound.
In Figure 3, the bounds on BER are graphed along with simulation results, for 16QAM and 64QAM with Gray labeling. Curves marked by ‘EX orig’ denote the original BICM expurgated bound, ‘EX new1’ the bound on BER corresponding to $`f_{ex,new}^I`$, ‘EX new2’ the bound corresponding to $`f_{ex,new}^{II}`$ and SIM computer simulation. The simulation results are obtained by using the suboptimal branch metric \[1, (9)\]. The following observations can be made from the figure. The ‘EX new1’ upper bound is greater than ‘EX orig’ by a factor of about $`2`$ for 16QAM and about $`3`$ for 64QAM. For both 16QAM and 64QAM, the ‘EX new2’ upper bound is nearly indistinguishable from (but is tighter than) the original expurgated bound (EX orig) for moderate to high SNR. This is related to $`d_{h_c}^{II}`$ nearly coinciding with $`d_h`$ for the square QAM constellations with Gray labeling listed in Table I. While this suggests that the original expurgated bound may be a valid upper bound for square QAM with Gray labeling, we do not have a proof to support such a claim.
|
warning/0506/math0506537.html
|
ar5iv
|
text
|
# The Strong Lefschetz Property and simple extensions
## Introduction
Let $`K`$ be a field, $`A`$ be a standard graded Artinian $`K`$-algebra and $`aA`$ a homogeneous form of degree $`k`$. The element $`a`$ is called a Lefschetz element if for all integers $`i`$ the $`K`$-linear map $`a:A_iA_{i+k}`$ (induced by multiplication with $`a`$) has maximal rank. One says that $`A`$ has the weak Lefschetz property if there exists a Lefschetz element $`aA`$ of degree 1. An element $`aA_1`$ for which all powers $`a^r`$ are Lefschetz is called a strong Lefschetz element, and $`A`$ is said to have the strong Lefschetz property if $`A`$ admits a strong Lefschetz element. Note that the set of Lefschetz elements $`aA_1`$ form a Zariski open subset of $`A_1`$. The same holds true for the set of strong Lefschetz elements.
Assuming that the characteristic of $`K`$ is zero and the defining ideal of $`A`$ is generated by generic forms, it is conjectured that $`A`$ has the strong Lefschetz property. Thus in particular, $`A=K[x_1,\mathrm{},x_n]/(f_1,\mathrm{},f_n)`$ should have the strong Lefschetz property for generic forms $`f_1,\mathrm{},f_n`$. Note that such an algebra is an Artinian complete intersection. It is expected that any standard graded Artinian complete intersection over a base field of characteristic 0 has the strong Lefschetz property. Stanley and later J. Watanabe proved this in case $`A`$ is a monomial complete intersection. Stanley used the Hard Lefschetz Theorem to prove this result, while Watanabe used the representation theory of the Lie algebra $`sl(2)`$.
As a main result of this paper we prove the following
Theorem: Let $`K`$ be a field of characteristic $`0`$, $`A`$ be a standard graded Artinian Gorenstein $`K`$-algebra having the strong Lefschetz property, and let $`fA[x]`$ be a monic homogeneous polynomial. Then the algebra $`B=A[x]/(f)`$ has the strong Lefschetz property.
The proof only uses techniques from linear algebra. The result implies in particular Stanley’s theorem. More generally it implies that a complete intersection $`K[x_1,\mathrm{},x_n]/(f_1,\mathrm{},f_n)`$ with $`f_iK[x_1,\mathrm{},x_i]`$ for $`i=1,\mathrm{},n`$ has the strong Lefschetz property.
## 1. The proof of the main theorem
Let $`A`$ be a standard graded $`K`$-algebra and $`IA`$ a graded ideal. For convenience we will say that $`aA`$ is Lefschetz for $`A/I`$ if the residue class $`a+I`$ is a Lefschetz element of $`A/I`$.
In the proof of the main theorem we shall use the following two lemmata.
###### Lemma 1.1.
Let $`A`$ be a standard graded $`K`$-algebra, $`f,gA`$ homogeneous elements which are nonzero divisors on $`A`$. Then $`f`$ is Lefschetz for $`A/(g)`$ if and only if $`g`$ is Lefschetz for $`A/(f)`$.
###### Proof.
Consider the long exact sequence for Koszul homology (see \[1, Corollary 1.6.13\])
$$\begin{array}{ccc}\mathrm{}H_1(g;A)H_1(f,g;A)H_0(g;A)& \stackrel{f}{}& H_0(g;A)H_0(f,g;A)0.\end{array}$$
Since $`g`$ is a non-zerodivisor on $`A`$ this yields the exact sequence
$$\begin{array}{ccc}0H_1(f,g;A)A/(g)& \stackrel{f}{}& A/(g)H_0(f,g;A)0.\end{array}$$
Similarly we obtain an exact sequence
$$\begin{array}{ccc}0H_1(f,g;A)A/(f)& \stackrel{g}{}& A/(f)H_0(f,g;A)0.\end{array}$$
Comparing this two exact sequences, the assertion follows.
###### Lemma 1.2.
Let $`K`$ be field of characteristic 0, $`A`$ a standard graded Artinian $`K`$-algebra with strong Lefschetz property and $`fA[y]`$ a monic homogeneous polynomial. Then for any strong Lefschetz element $`aA_1`$ there exists a non-zero element $`cK`$ such that $`f(a/c)`$ is a Lefschetz element of $`A`$.
###### Proof.
Let $`f=y^d+a_1y^{d1}+\mathrm{}+a_d`$, and $`s=\mathrm{max}\{i:A_i0\}`$. We may assume that $`ds`$ because otherwise the statement is trivial. For $`cK`$ we set $`f_c=y^d+_{i=1}^dc^ia_iy^{di}`$. Let $`aA_1`$ be a strong Lefschetz element. Then $`a^d`$ is a Lefschetz element, that is, for all $`i`$ the multiplication map $`f_0(a):A_iA_{i+d}`$ has maximal rank.
Fix $`isd`$ and $`K`$-bases of the nonzero $`K`$-vector spaces $`A_i`$ and $`A_{i+d}`$, and let $`D_c`$ be the matrix describing the $`K`$-linear map $`f_c(a):A_iA_{i+d}`$. Note that the entries of $`D_c`$ are polynomial expressions in $`c`$ with coefficients in $`K`$. Now $`P_c(a)`$ has maximal rank if and only if one maximal minor $`M_j(D_c)`$ of $`D_c`$ does not vanish. In particular, $`M_{j_0}(D_0)0`$ for some $`j_0`$. Since $`M_{j_0}(D_c)`$ is a (non-zero) polynomial expression in $`c`$ with coefficients in $`K`$, there exist only finitely many $`cK`$ such that $`M_{j_0}(D_c)=0`$. Thus, since $`K`$ is infinite, we have $`M_{j_0}(D_c)0`$ for infinitely many $`cK`$, and so $`f_c(a):A_iA_{i+d}`$ has maximal rank for infinitely many $`cK`$. Since $`A`$ has only finitely many non-zero components, we can therefore find $`cK`$, $`c0`$ such that $`f_c(a)`$ has maximal rank for all $`i`$. Then $`a/cA_1`$ has the desired property, since $`f(a/c)=f_c(a)/c^d`$.
Now we are ready to begin with the proof of the main theorem. Let $`A`$ be a standard graded Artinian Gorenstein $`K`$-algebra having the strong Lefschetz property.
In a first step we will prove: suppose $`C=A[x]/(x^r)`$ has the strong Lefschetz property for all $`r>1`$, then $`B=A[x]/(f)`$ has the strong Lefschetz property for any monic homogeneous polynomial $`fA[x]`$.
Let $`U_rB_1`$ be the Zariski open set of elements $`bB_1`$ for which $`b^r`$ is a Lefschetz element. If $`U_r\mathrm{}`$ for all $`r1`$, then the finite intersection $`U=_rU_r`$ is non-empty, as well, and any $`bU`$ is then a strong Lefschetz element. Thus it suffices to show that for each $`r1`$ there exists an element $`b_rB_1`$ such that $`b_r^r`$ is a Lefschetz element.
By Lemma 1.2 we may choose an element $`aA_1`$ such that $`f(a)`$ is a Lefschetz element of $`A`$. It follows that $`f(x)`$ is Lefschetz for $`A[x]/(ax)`$. Thus by Lemma 1.1, the element $`b_1=ax`$ is Lefschetz for $`B`$.
In case $`r>1`$, we may view $`f(y)`$ as a polynomial in $`C[y]`$ where $`C=A[x]/(x^r)`$. By our assumption $`C`$ has a strong Lefschetz element. Now Lemma 1.2 implies that we can find a strong Lefschetz element $`cC_1`$ such that $`f(c)`$ is a Lefschetz element of $`C`$. Since the strong Lefschetz elements form a nonempty Zariski open set in $`C_1`$, we may assume that $`c=a+\lambda x`$ with $`aA_1`$ and $`\lambda K`$, $`\lambda 0`$. Applying the substitution $`xb_r=\lambda ^1(xa)`$ it follows that $`f(x)`$ is Lefschetz for $`A[x]/b_r^r`$. Thus by Lemma 1.2, the element $`b_r^r`$ is Lefschetz for $`A[x]/(f)`$.
In order to complete the proof of the theorem it remains to be shown that if $`A`$ is a standard graded Artinian Gorenstein $`K`$-algebra having the strong Lefschetz property, then $`A[x]/(x^q)`$ has the strong Lefschetz property. We use Lemma 1.1 and show instead that if $`aA_1`$ is a strong Lefschetz element, then for all $`k`$ the element $`x^q`$ is Lefschetz for $`B=A[x]/(a+x)^k`$.
In $`B`$ we have
$$x^k=\underset{j=0}{\overset{k1}{}}\left(\genfrac{}{}{0pt}{}{k}{j}\right)a^{kj}x^j.$$
By induction on $`r`$ it follows that
$$x^r=(1)^{rk1}\underset{j=0}{\overset{k1}{}}\left(\genfrac{}{}{0pt}{}{rj1}{rk}\right)\left(\genfrac{}{}{0pt}{}{r}{j}\right)a^{rj}x^j\text{for}rk.$$
Note that
$$\left(\genfrac{}{}{0pt}{}{rj1}{rk}\right)\left(\genfrac{}{}{0pt}{}{r}{j}\right)=\frac{k}{rj}\left(\genfrac{}{}{0pt}{}{r}{k}\right)\left(\genfrac{}{}{0pt}{}{k1}{j}\right),$$
so that
$$x^r=(1)^{rk1}\underset{j=0}{\overset{k1}{}}\left(\genfrac{}{}{0pt}{}{r}{k}\right)\left(\genfrac{}{}{0pt}{}{k1}{j}\right)\frac{k}{rj}a^{rj}x^j$$
for all $`rk`$. Thus for all $`r0`$ we have
$$x^r=\underset{j=0}{\overset{k1}{}}c_{rj}a^{rj}x^j$$
with
$$c_{rj}=\{\begin{array}{ccc}\delta _{rj},\hfill & \text{if}\hfill & rk1\hfill \\ (1)^{rk1}\left(\genfrac{}{}{0pt}{}{r}{k}\right)\left(\genfrac{}{}{0pt}{}{k1}{j}\right)\frac{k}{rj},\hfill & \text{if}\hfill & rk,\hfill \end{array}$$
where $`\delta _{rj}`$ denotes the Kronecker symbol.
Now we show that the map $`\beta _t^q:B_tB_{t+q}`$ given by multiplication with $`x^q`$ has maximal rank.
We denote by $`\alpha _i^j:A_iA_{i+j}`$ the $`K`$-linear map given by multiplication with $`a^j`$. For each element $`ux^iA_{ti}x^i`$ we have
$$x^q(ux^i)=\underset{j=0}{\overset{k1}{}}c_{q+i,j}a^{q+ij}ux^j=\underset{j=0}{\overset{k1}{}}c_{q+i,j}\alpha _{ti}^{q+ij}(u)x^j.$$
Since for each $`j`$ the $`K`$-vectorspace $`B_j`$ has the direct sum decomposition
$$B_j=\underset{i=0}{\overset{k1}{}}A_{ti}x^i,$$
the linear map $`\beta _t^q`$ can be described by the following block matrix
$$M=\left(\begin{array}{cccc}c_{q,0}\alpha _t^q& c_{q+1,0}\alpha _{t1}^{q+1}& \mathrm{}& c_{q+k1,0}\alpha _{tk+1}^{q+k1}\\ c_{q,1}\alpha _t^{q1}& c_{q+1,1}\alpha _{t1}^q& \mathrm{}& \mathrm{}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ c_{q,k1}\alpha _t^{qk+1}& c_{q+1,k1}\alpha _{t1}^{qk+2}& \mathrm{}& c_{q+k1,k1}\alpha _{tk+1}^q\end{array}\right).$$
Our aim is to show that $`M`$ has maximal rank. Assume first that $`q<k`$, then
$$M=\left(\begin{array}{cc}0& N\\ \mathrm{id}& \end{array}\right),$$
where $`N=(c_{q+j,i}\alpha _{tj}^{q+ji})_{\genfrac{}{}{0pt}{}{i=0,\mathrm{},q1}{j=kq,\mathrm{},k1}}`$. It follows that $`M`$ has maximal rank if and only if $`N`$ has maximal rank. Thus the general case is treated if we can prove that for all $`q`$ the matrix
$$N=(c_{q+j,i}\alpha _{tj}^{q+ji})_{\genfrac{}{}{0pt}{}{i=0,\mathrm{},s1}{j=rq,\mathrm{},k1}}\text{with}s=\mathrm{min}\{q,k\}\text{and}r=\mathrm{max}\{q,k\}$$
has maximal rank.
We show this by applying certain block row and block column operations in order to simplify the matrix without changing its rank. The kind of operations we will apply are the following:
1. multiplication of a block row or a block column of $`N`$ with a non-zero rational number;
2. for $`d`$ , $`d0`$ and $`j<i`$ compose $`d\alpha _{ti}^{ij}`$ with each block $`c_{jl}\alpha _{tj}^{jl}`$ of the $`j`$th block column of $`N`$ to obtain a $`j`$th block column whose blocks are $`dc_{jl}\alpha _{tj}^{jl}\alpha _{ti}^{ij}=dc_{jl}\alpha _{ti}^{il}`$, and subtract this new block column from the $`i`$th block column to obtain the new $`i`$th block column whose blocks are
$$(c_{il}dc_{jl})\alpha _{ti}^{il},l=0,\mathrm{},s1.$$
These operations only change the coefficients $`c_{ji}`$ of the block entries, that is, the matrix $`N=(c_{q+j,i}\alpha _{tj}^{q+ji})_{\genfrac{}{}{0pt}{}{i=0,\mathrm{},s1}{j=rq,\mathrm{},k1}}`$ will be transformed into a matrix of the form $`N^{}=(c_{q+j,i}^{}\alpha _{tj}^{q+ji})_{\genfrac{}{}{0pt}{}{i=0,\mathrm{},s1}{j=rq,\mathrm{},k1}}`$ with certain new coefficients $`c_{q+j,i}^{}`$.
Consider the “coefficient matrix” $`L=(c_{q+j,i})_{\genfrac{}{}{0pt}{}{i=0,\mathrm{},s1}{j=rq,\mathrm{},k1}}`$ of $`N`$. Then the coefficient matrix $`L^{}`$ of $`N^{}`$ is obtained from $`L`$ by the following row and column operations:
1. multiplication or division of a row or a column with a non-zero rational number;
2. subtraction of a multiple of the $`j`$th column from the $`i`$th column where $`j<i`$.
Next we intend to show that by these operations $`L`$ can be transformed into a matrix $`L^{}`$ such that all entries of $`L^{}`$ on the anti-diagonal are non-zero, while the entries below the anti-diagonal are all zero.
We first simplify $`L`$ by dividing each $`j`$th column by $`(1)^{jk1}\left(\genfrac{}{}{0pt}{}{j}{k}\right)k`$ and each $`i`$th row by $`\left(\genfrac{}{}{0pt}{}{k1}{i}\right)`$. The result of these operations is the matrix
$$\left(\begin{array}{cccc}1/r& 1/(r+1)& \mathrm{}& 1/(r+s1)\\ 1/(r1)& 1/r& \mathrm{}& 1/(r+s2)\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ 1/(rs+1)& 1/(rs+2)& \mathrm{}& 1/r\end{array}\right),$$
which we again denote by $`L`$.
We will use the following simple fact from linear algebra: suppose $`F=(f_{ij})_{i,j=1,\mathrm{},n}`$ is an $`n\times n`$-matrix with coefficients in a field $`K`$. Then the following conditions are equivalent:
1. the matrix $`F`$ can be transformed by operations of type (ii) into a matrix $`F^{}`$ with $`f_{ij}^{}0`$ for $`i+j=n+1`$, and $`f_{ij}^{}=0`$ for $`i+j>n+1`$;
2. $`det(F_i)0`$ for $`i=1,\mathrm{},n`$ where $`F_i=(f_{kl})_{\genfrac{}{}{0pt}{}{k=i,\mathrm{},n}{l=1,\mathrm{}i}}`$.
Indeed, it is clear that (a)$``$(b). Conversely, assuming (b) we have $`det(F_n)=f_{n1}0`$. Thus by subtracting suitable multiples of the first column from the other columns we obtain a matrix $`G=(g_{ij})`$ with $`g_{ni}=0`$ for $`i=2,\mathrm{},n`$, and such that $`det(F_i)=det(G_i)`$ for $`i=1,\mathrm{},n`$, where $`G_i=(g_{kl})_{\genfrac{}{}{0pt}{}{k=i,\mathrm{},n}{l=1,\mathrm{}i}}`$. Applying the induction hypothesis to the matrix $`G^{}=(g_{kl})_{\genfrac{}{}{0pt}{}{k=1,\mathrm{},n1}{l=2,\mathrm{},n}}`$, the assertion follows.
Applying this result from linear algebra, we see that $`L`$ can be transformed by operations of type (ii) into the matrix $`L^{}`$ of the desired form if for all integers $`0t<s`$ the matrices of the shape
$$S=(1/(ri+j))_{i,j=0,\mathrm{},t}$$
are non-singular. It is an easy exercise in linear algebra to show that this is indeed the case.
After all these operations our matrix $`N`$ is transformed into the matrix $`N^{}`$ whose anti-diagonal has the block entries
$$c_1^{}\alpha _{t+qr}^{rs+1},c_2^{}\alpha _{t+qr1}^{rs+3},\mathrm{},c_{s1}^{}\alpha _{tk+1}^{q+k1}$$
with non-zero rational coefficients $`c_i^{}`$, and whose block entries below the anti-diagonal are all zero.
We will show that for $`i=0,\mathrm{},s1`$ either all $`\alpha _{t+qri}^{rs+2i+1}`$ are injective maps, or else all $`\alpha _{t+qri}^{rs+2i+1}`$ are surjective maps. Then clearly $`N^{}`$ has maximal rank, and consequently $`N`$ has maximal rank.
For all integers $`i`$ and $`j`$ with $`0i<j`$ the maps
$$\alpha _i^{ji}:A_iA_j$$
have maximal rank, by assumption. In particular, $`\alpha _i^{ji}`$ is injective if $`dimA_idimA_j`$ and surjective if $`dimA_idimA_j`$.
Let $`\sigma =\mathrm{max}\{i:A_i0\}`$. Then, since $`A`$ is Gorenstein, the Hilbert function of $`A`$ is symmetric (see e.g. \[1, Corollary 4.4.6, Remark 4.4.7\]), that is,
$$dimA_i=dimA_{\sigma i}\text{for all}i,$$
and since $`A`$ has the weak Lefschetz property ($`A`$ even has the strong Lefschetz property), the Hilbert function of $`A`$ is unimodal (see e.g. \[2, Remark 3.3\]). It then follows that
$$dimA_idimA_j\text{if and only if}i\sigma j.$$
Thus we conclude that
$$\alpha _i^{ji}\text{is}\{\begin{array}{ccc}\text{injective}\hfill & \text{if}\hfill & i\sigma j,\hfill \\ \text{surjective}\hfill & \text{if}\hfill & i\sigma j.\hfill \end{array}$$
Thus in case of the maps $`\alpha _{t+qri}^{rs+2i+1}`$, we have to compare the size of the numbers $`t+qri`$ and $`\sigma [(rs+2i+1)+(t+qri)]=\sigma tq+si1`$. Since it does not depend on $`i`$ which of the two numbers is less than or equal to other, it follows $`\alpha _{t+qri}^{rs+2i+1}`$ is injective for all $`i`$, or $`\alpha _{t+qri}^{rs+2i+1}`$ is surjective for all $`i`$, as desired.
## 2. Some comments
As an immediate consequence of our main theorem we obtain
###### Corollary 2.1.
Let $`K`$ be field of characteristic 0, and $`A`$ be an Artinian Gorenstein $`K`$-algebra. For $`i=1,\mathrm{},n`$ let $`f_iA[x_1,\mathrm{},x_i]`$ be a homogeneous polynomial which is monic in $`x_i`$. Then the $`K`$-algebra
$$A[x_1,\mathrm{},x_n]/(f_1,\mathrm{},f_n)$$
has the strong Lefschetz property.
The result implies in particular that $`K[x_1,\mathrm{},x_n]/(f_1,\mathrm{},f_n)`$ has the strong Lefschetz property, if for $`i=1,\mathrm{},n`$, $`f_iK[x_1,\mathrm{},x_i]`$ is a homogeneous and monic polynomial in $`x_i`$. In the special case that $`f_i=x_i^{a_i}`$ for $`i=1,\mathrm{},n`$, we obtain the theorem of Stanley . The slightly more general result with the $`f_i`$ as described before, can also be deduced directly from Stanley’s theorem using the following result of Wiebe \[6, Proposition 2.9\]: let $`IK[x_1,\mathrm{},x_n]`$ be a graded ideal, and assume that $`K[x_1,\mathrm{},x_n]/\mathrm{in}(I)`$ has the strong Lefschetz property, where $`\mathrm{in}(I)`$ is the initial ideal with respect to some term order. Then $`K[x_1,\mathrm{},x_n]/I`$ has the strong Lefschetz property.
In the above situation we have $`\mathrm{in}(f_i)=x_i^{\mathrm{deg}f_i}`$ for $`i=1,\mathrm{},n`$, if we choose the lexicographical order induced by $`x_n>x_{n1}>\mathrm{}>x_1`$. Since the initial terms of the generators form a regular sequence it follows that
$$\mathrm{in}(I)=(\mathrm{in}(f_1),\mathrm{},\mathrm{in}(f_n))=(x_1^{a_1},\mathrm{},x_n^{a_n}).$$
In case the $`K`$-algebra $`A`$ is not Gorenstein, the proof of our main theorem yields the following weaker result.
###### Proposition 2.2.
Let $`K`$ be a field, $`A`$ a standard graded Artinian $`K`$-algebra having the strong Lefschetz property, and let $`fA[x]`$ be a monic homogeneous polynomial. Then the algebra $`B=A[x]/(f)`$ has the weak Lefschetz property.
###### Proof.
Recall the following step in the proof of the main theorem: by Lemma 1.2 we may choose an element $`aA_1`$ such that $`f(a)`$ is a Lefschetz element of $`A`$. It follows that $`f(x)`$ is Lefschetz for $`A[x]/(ax)`$. Thus by Lemma 1.1, $`b=ax`$ is Lefschetz for $`B`$.
Analyzing the arguments in the proof of our main theorem we see that all results remain valid if the characteristic of the base field is large enough. More precisely we have
###### Corollary 2.3.
Let $`K`$ be a field and $`A`$ an Artinian Gorenstein $`K`$-algebra with socle degree $`\sigma =\mathrm{max}\{t:A_t0\}`$ and multiplicity $`e(A)=_{t=0}^\sigma dim_KA_t`$. Let $`fA[x]`$ be a homogeneous monic polynomial of degree $`q`$. Then $`B=A[x]/(f)`$ has the strong Lefschetz property if
$$\mathrm{char}K\{\begin{array}{ccc}2q+\sigma 1,\hfill & \text{and }f=x^q\text{,}\hfill & \\ \mathrm{max}\{e(A),2q+\sigma 1\},\hfill & \text{otherwise}.\hfill & \end{array}$$
###### Proof.
In case $`f=x^q`$ we must make sure that all the binomials in the expression $`x^r=(1)^{rk1}_{j=0}^{k1}\left(\genfrac{}{}{0pt}{}{rj1}{rk}\right)\left(\genfrac{}{}{0pt}{}{r}{j}\right)a^{rj}x^j`$ are units in the field $`K`$, and this must be satisfied for all $`r=q+k`$ where are less than or equal the socle degree of $`A[x]/(x^q)`$. Since the socle degree of $`A[x]/(x^q))`$ is equal to $`q+\sigma 1`$, we therefore need that $`\mathrm{char}K`$ does not divide any prime number $`2q+\sigma 1`$.
In the general case we had to apply Lemma 1.2. For the proof of this lemma it was necessary that the field $`K`$ has enough elements, so that for all the polynomials in $`c`$ defined by the maximal minors considered in the proof we find a common element $`cK`$ for which these polynomials do not vanish. This is possible if $`\mathrm{char}K>e(A)`$.
We conclude this note with the following
###### Example 2.4.
If $`A`$ has the strong Lefschetz property and $`fA`$ is a generic form. One expect that $`A/(f)`$ has again the strong Lefschetz property. However, in general this is no the case. Indeed, let $`A=K[x_1,\mathrm{},x_5]/(x_1^4,x_2^4,x_3^4,x_4^4,x_5^2)`$. Then $`A`$ has the strong Lefschetz property. Let $`fA`$ be a generic form of degree 8 and set $`B=A/(f)`$. (We use the “Randomized” command of CoCoA to produce generic forms.)
The Hilbert series of $`B`$ is given by
$$\text{Hilb}_B(t)=1+5t+14t^2+30t^3+51t^4+71t^5+84t^6+84t^7+70t^8+46t^9+16t^{10}.$$
Let $`bB`$ be a generic linear form, and set $`C=B/(b^9)`$. Then
$$\text{Hilb}_C(t)=1+5t+14t^2+30t^3+51t^4+71t^5+84t^6+84t^7+70t^8+45t^9+12t^{10}.$$
It follows that the map $`B_1\stackrel{b^9}{}B_{10}`$ is not surjective but also not injective because $`dim_KB_1+dim_KC_{10}=5+12>16=dim_KB_{10}`$. Thus $`B`$ does not have the strong Lefschetz property.
On the other hand it can be checked that $`B`$ has the maximal rank property, that is, any generic form in $`B`$ has maximal rank. Such an example seems to be new, see .
|
warning/0506/hep-th0506266.html
|
ar5iv
|
text
|
# I Introduction
## I Introduction
There has been much progress recently in understanding the mechanisms by which the compactification moduli and the dilaton of string theory are stabilized<sup>1</sup><sup>1</sup>1For recent reviews with references to the original literature see Balasubramanian (2004) Silverstein (2004).. In particular Giddings et al. Giddings et al. (2002) discussed type IIB compactification on a Calabi-Yau orientifold $`X`$, and showed that turning on fluxes of NSNS and RR three and five form fields can generate a potential for the complex structure moduli of $`X`$ and the dilaton-axion field, which as one would expect is of the supergravity form. However the Kaehler moduli (which includes the overall scale modulus usually denoted by $`T`$ in the string-phenomenology literature) cannot be fixed by the fluxes. A suggestion for fixing these was subsequently made by Kachru et al. Kachru et al. (2003a) (KKLT).
The KKLT proposal was to argue that at least for certain choices of fluxes the dilaton-axion ($`S`$) and the complex structure moduli ($`z^i`$) would have masses that are close to the string scale and could be integrated out classically to get a theory for the light Kaehler moduli. But at the classical minimum, the potential being of the no-scale type, is zero and does not fix $`T`$ (for simplicity we will just consider the case of one Kaehler modulus). In order to get a potential for $`T`$ these authors proposed that a contribution coming from certain non-perturbative effects be included. Concretely the argument is that the fluxes which fix $`S`$ and $`z^i`$ give a constant $`W_0`$ in the superpotential, to which the exponential contribution coming from the non-perturbative (NP) effects should be added, resulting in a total superpotential
$$W=W_0+Ce^{aT},$$
for the theory of the light modulus $`T`$. Here the pre-factor of the NP term is taken to be a constant since we are ignoring perturbative corrections as in KKLT. The Kaehler potential of the theory is taken to be its classical value
$$K=3\mathrm{ln}(T+\overline{T}).$$
With this prescription it is easy to see Kachru et al. (2003a) that the modulus $`T`$ is fixed at a supersymmetric Anti-deSitter (AdS) point. KKLT go on to lift this minimum by adding a term coming from $`\overline{D}`$ branes.
In this note we will examine the consistency of the assumptions that lead to the KKLT theory for the light modulus. These include the expectation that the low energy supergravity action is a good starting point for finding clssical string vacua and also that one can find flux configurations such that the compolex stucture moduli and the dilaton are heavy. As KKLT observed, in order for the procedure to make sense, the minimum of the potential should be at a large value of $`T`$, so that the size of $`X`$ is large on the string scale justifying the ten-dimensional SUGRA starting point, and the superpotential itself is valid only in the region $`aT_R>>1`$, so that the NP term can be regarded as a small correction to the classical theory. KKLT argued that although generically the fluxes would give a value of $`W_0`$ that is of order one (thus violating the first requirement) there would be (at least for CY manifolds $`X`$with large $`h_{21}`$) flux configurations which would give small values of this constant.
Now of course there are obvious corrections to this theory coming from perturbative effects which, even though they leave the superpotential unchanged, will affect the Kaehler potential. There is also a non-perturbative correction to the Kaehler potential (see for example Kaplunovsky and Louis (1994)Burgess et al. (1996)). Thus one would expect a corrected Kaehler potential of the form
$$K=3\mathrm{ln}(T+\overline{T}+f+ke^{a(T+\overline{T})})$$
(1)
where $`f`$ is a constant and in principle the coefficient $`k`$ could be of the same order as the prefactor $`C`$ in the superpotential. We will ignore these corrections in most of this paper and will touch on their effects at the end of our discussion. What we are going to investigate is just the procedure of first ignoring the non-perturbative term in the superpotential in order to integrate out $`S,z^i`$ to get a constant superpotential, and then including the non-perturbative term. We will find that if the non-perturbative term is included from the beginning, there are terms which are necessarily controlled by the same coefficient as the terms which are included by KKLT and therefore cannot be set to zero. Related obeservations have been made by Choi et al. Choi et al. (2004) <sup>2</sup><sup>2</sup>2See also Lust et al. (2005). Related issues in the heterotic context have been discussed in Curio et al. (2005). but we will find that there needs to be some modifications of their arguments also. In the course of this investigation we came across some issues in the procedure of integrating out heavy fields in supersymmetric theories that are of general interest, but we will reserve that discussion to a separate publication de Alwis (2005).
The two stage calculation of KKLT appears to lead only to a critical point that is AdS supersymmetric. To get a dS minimum (and broken supersymmetry) KKLT add an uplifting terms namely a contribution from a Dbar brane. An alternate suggestion is to find some sector that gives a D-term Burgess et al. (2003). However it is easy to see that such a term will not lift an AdS supersymmetric minimum. This is because of the relation
$$2\mathrm{}f^{ab}D_b=\frac{ik^{ai}D_iW}{W}$$
(with $`k^{ai}`$ a generator of a Killing symmetry of the Kahler metric and $`f`$ the gauge coupling function) between the D and F terms that is valid at generic points where the superpotential is non-zero <sup>3</sup><sup>3</sup>3This relation can be found for example in Gates et al. (1983) eqns. (8.7.7b) (8.7.8). It has been recently rediscovered in the current context in Choi et al. (2005). The right hand side of this equation can also be rewritten as $`ik^{ai}_iK+\xi \mathrm{tr}T^a`$ (with $`\xi `$a FI parameter) and would give an independent condition if $`W=0`$.. So a critical point where the F-term is zero with $`W0`$ giving an AdS minimum as in KKLT will not be lifted by adding a D term. This also means that a Dbar term as in KKLT if it is to lift the AdS minimum would have to be an explicit breaking term from the point of view of four dimensional supergravity. Of course one can lift a non supersymmetric $`(D_iW0)`$ AdS critical point (such an example can be found in Brustein and de Alwis (2004)) by a D-term.
One of the outcomes of the current investigation is that if one integrates out the heavy moduli in one stage then one has extra terms in the potential (compared to the two stage procedure). These terms enable one to find examples where the F-term potential by itself has positive local minima thus obviating the need for uplifting terms.<sup>4</sup><sup>4</sup>4It is possible that $`\alpha ^{}`$corrections also achieve the same end. See for instance Balasubramanian and Berglund (2004). In view of this it would be interesting to revisit other issues such as the question of getting a viable cosmology in the context of such models.
## II Model with S and T
Consider first a compactification with fluxes on a rigid CY manifold $`X`$, i.e. one with $`h_{21}=0`$.
The classical Kaehler potential is
$$K=\mathrm{ln}(S+\overline{S)}3\mathrm{ln}(T+\overline{T}),$$
(2)
and with the non-perturbative contribution included we have for the superpotential,
$$W=A+SB+Ce^{aT},$$
(3)
where $`A,B`$ are determined by the fluxes and $`C`$ is an $`O(1)`$ prefactor which may be determined by an instanton calculation. Now let us first solve for $`S`$ in terms of $`T`$ as in KKLT by requiring that the Kaehler derivative with respect to S of this superpotential is zero <sup>5</sup><sup>5</sup>5There are some issues involved in integrating out heavy fields in supersymmetric theories that have not been discussed in the literature. In particular it turns out that even in global supersymmetric theories the condition $`_HW=0`$ which is imposed in order to integrate out a heavy field $`H`$ is valid only if we also restrict the light field space to range over values that are less than the mass of the heavy field. A similar restriction holds in supergravity. These issues are discussed in a recent paper by the author de Alwis (2005). In particular it is shown there that solving the Kaehler derivative equated to zero for the heavy field is an acceptable method of computing the bosonic effective potential for the light fields even though to get the complete action for the light fields addtional terms involving the fermions need to be kept.. This gives,
$$D_SW=B\frac{A+SB+Ce^{aT}}{S+\overline{S}}=0,$$
(4)
implying
$$\overline{S}=(A+Ce^{aT})/B.$$
(5)
Clearly if we substitute this back into (3) we get an expression which contains both $`T`$and $`\overline{T}`$ i.e. it is not holomorphic. As far as the scalar potential goes this is not a problem - the coupling of the chiral scalars to supergravity is actually determined by one Kaehler invariant real function (see for example Wess and Bagger (1992) or Gates et al. (1983))
$$G=K(\mathrm{\Phi },\overline{\mathrm{\Phi }})+\mathrm{ln}W(\mathrm{\Phi })+\mathrm{ln}\overline{W}(\overline{\mathrm{\Phi }}).$$
(6)
After solving for $`S`$ this becomes
$`G`$ $`=`$ $`\mathrm{ln}({\displaystyle \frac{(A+Ce^{aT})}{B}}+{\displaystyle \frac{(\overline{A}+\overline{C}e^{a\overline{T}})}{\overline{B}}})3\mathrm{ln}(T+\overline{T)}`$ (7)
$`+\mathrm{ln}(A+B{\displaystyle \frac{(\overline{A}+\overline{C}e^{a\overline{T}})}{\overline{B}}}+Ce^{aT})+c.c.).`$
This may in effect be regarded as the new Kaehler potential with the superpotential being taken to be unity. The potential for $`T`$ may now be computed from the standard formula
$$V=e^G(G_iG_{\overline{j}}G^{i\overline{j}}3),$$
(8)
where $`G_i=G/\mathrm{\Phi }^i`$ and $`G_{i\overline{j}}=_i_{\overline{j}}G`$ is the Kaehler metric.
On the other hand if we had followed the prescription of KKLT we would have solved for $`S`$ in the absence of the non-perturbative term to get $`\overline{S}=A/B`$, a constant superpotential $`W=A+B\overline{A}/\overline{B}W_0`$ and apart from an irrelevant constant the Kaehler potential is $`K=3\mathrm{ln}(T+\overline{T})`$. Now the non-perturbative term is added to $`W`$ to get a Kaehler invariant function,
$$G=3\mathrm{ln}(T+\overline{T})+(\mathrm{ln}(A+B\frac{\overline{A}}{\overline{B}}+Ce^{aT})+c.c.)$$
(9)
The problem is that in this two stage process one is ignoring non-perturbative terms in (7) that are in fact controlled by the same constant $`C`$ as the terms that are being kept. *There is no approximation in which one can keep the latter and ignore the former.* In other words the procedure of first integrating out $`S`$ and then adding the non-perturbative term to $`W`$ cannot be justified. It should be noted also that this correction is of the same order as the term kept by KKLT independently of the condition $`W_0<<1`$ required by KKLT.
## III Models with complex structure moduli
The model studied in the previous section however does not give a viable theory in any case. Choi et al Choi et al. (2004) have analyzed the stability of this model (without first integrating out the dilaton-axion $`S`$). They find that the supersymmetric extremum $`D_SW=D_TW=0`$ is in fact a saddle point. Although this does not make the supersymmetric point unstable (since a saddle point or even a maximum can be a stable AdS solution) it becomes problematic when one adds a “lifting potential” as in KKLT to get a dS solution, since it is unlikely that such a corrected potential would have a stable critical point and indeed that is what Choi et al find. As they have argued, the point is that the mass of the field that is integrated out depends on the light field and thus it cannot be integrated out as suggested by KKLT. Our argument above highlights this point directly by showing that the procedure of KKLT ignores effects that simply cannot be set to zero or assumed to be small.
Choi et al go on to analyze models with complex structure moduli. However the analysis is done by assuming that the complex structure moduli can be integrated out holomorphically (resulting in a holomorphic superpotential) to get a potential in just $`S`$ and $`T`$. We will find that this procedure is not consistent and has the same problems that we highlighted before.
The Kaehler potential is now
$$K=\mathrm{ln}(S+\overline{S})3\mathrm{ln}(T+\overline{T})+k(z^i,\overline{z}^{\overline{j}}).$$
(10)
Here $`k=\mathrm{ln}\mathrm{\Omega }\overline{\mathrm{\Omega }}`$ (with $`\mathrm{\Omega }`$ being the holomorphic 3-form on the Calabi-Yau space) is the Kaehler potential on the complex structure moduli space (with complex coordinates $`z^i,i=1,\mathrm{},h_{21}`$). Also we have assumed that there is only one Kaehler structure. The superpotential is taken to be
$$W=A(z^i)+SB(z^i)+Ce^{aT}.$$
(11)
The Kaehler derivatives with respect to the chiral scalars are,
$`D_TW`$ $`=`$ $`aCe^{aT}{\displaystyle \frac{3}{T+\overline{T}}}W,`$ (12)
$`D_SW`$ $`=`$ $`B{\displaystyle \frac{W}{S+\overline{S}}},`$ (13)
$`D_iW`$ $`=`$ $`_iA+S_iB+_ikW.`$ (14)
Thus there are $`h_{12}+2`$ complex equations for as many complex variables ($`h_{12}`$ complex structure moduli, one Kaehler modulus and the dilaton-axion) so that all of them can be fixed. Choi et al assume that the equation $`D_iW=0`$ can be solved holomorphically, giving <sup>6</sup><sup>6</sup>6Choi et al ignore the $`T`$ dependence but we keep it here, in any case the relevant issue is holomorphy. an effective theory for $`S`$ and $`T`$ with a superpotential
$$W=W_{eff}+Ce^{aT}$$
where
$$W_{eff}=A(z^i(S,T))+B(z^i(S,T))S.$$
and a Kaehler potential
$$K=\mathrm{ln}(S+\overline{S})3\mathrm{ln}(T+\overline{T})+k(z(S,T),\overline{z}(\overline{S},\overline{T}))$$
(15)
The SUSY conditions in the effective theory are,
$`D_SW`$ $`=`$ $`D_SW|_{z^i}+{\displaystyle \frac{z^i}{S}}D_iW=0`$
$`F_T`$ $`=`$ $`D_TW|_{z^i}+{\displaystyle \frac{z^i}{T}}D_iW=0`$
which are of course implied by the equations of the original theory $`D_SW=D_TW=D_iW=0`$, with the chiral fields being all independent variables. However this equivalence is guaranteed only if we do not ignore the last term in (15). In Choi et al however the effective Kaehler potential is taken to be just the first two terms of (15). This is not really consistent and it is in fact the dependence of $`k`$ on $`z^i`$ as well as $`\overline{z^{\overline{i}}}`$ that makes it impossible to find an holomorphic solution for $`S`$and $`T`$ in terms of $`z^i`$ and hence a holomorphic $`W_{eff}`$. To see this consider the equation that needs to be solved,
$$D_iW=_iA(z^i)+S_iB(z^i)+W(S,T,z^i)k_i=0.$$
(16)
This is supposed to have solutions $`z^i=z^i(S,T)`$ such that $`_{\overline{S}}z^i=_{\overline{T}}z^i=0`$ for some range of values of $`S`$ and $`T`$. So differentiating (16) with respect to $`\overline{S}`$ we get from the assumed holomorphicity,
$$W(S,T,z^i(S,T))k_{i,\overline{j}}\frac{\overline{z}^{\overline{j}}}{\overline{S}}=0.$$
But the superpotential should not vanish (except at particular points) and $`k_{i\overline{j}}`$ is the Kaehler metric on the complex structure moduli space which is non-degenerate. Hence the above equation implies $`\frac{z^i}{S}=0`$ and similarly $`\frac{z^i}{T}=0`$! Clearly what is at fault is the assumption of holomorphicity. In other words the solution of (16) must be of the form $`z^i=z^i(S,T,\overline{S},\overline{T})`$. As we saw explicitly in the case without complex structure moduli where $`S`$ was integrated out, in supergravity, fields cannot be integrated out in a holomorphic fashion. As in that case, we expect that the effective supergravity theory is one with a superpotential that is unity and a Kaehler potential
$$G=K=\mathrm{ln}(S+\overline{S})3\mathrm{ln}(T+\overline{T})+k(z(S,T,\overline{S},\overline{T}),\overline{z}(\overline{S},\overline{T},S,T))+\mathrm{ln}|W(S,T,z(S,T,\overline{S},\overline{T}))|^2.$$
Clearly similar remarks would apply to an effective theory that is obtained by integrating out both the complex structure moduli as well as the dilaton-axion to get an effective theory for $`T`$.
## IV Effective potential for T
Let us now try to find the effective potential for the modulus $`T`$ (assumed light) after integrating out the complex structure moduli $`z_i`$ and the dilaton-axion $`S`$ which we assume to be heavy. Such an effective potential is useful for cosmological applications. Hitherto it has been derived using the two stage process of KKLT, but as we have already seen in section 3 even in the absence of the $`z_i`$ there are terms in the effective potential that are as large as the terms that are kept in the two stage argument but were ignored there.
The potential below the string and Kaluza-Klein scale is of the standard $`N=1`$ SUGRA form with the Kaehler and superpotentials being given respectively by eqns (10) (11). To classically integrate out the $`z_i`$and $`S`$ we need to solve eqns (12) (14) for these variables in terms of $`T`$ and then plug those solutions into the expression for the potential given in (8) and (6). However the equations to be solved are non-linear in the $`z_i`$ so the best we can do is to write the general form of the solution in a power series expansion in $`Ce^{aT}`$ for $`aT>>1`$. So we write,
$`S`$ $`=`$ $`\alpha +\beta Ce^{aT}+\gamma \overline{C}e^{a\overline{T}}+\mathrm{}`$
$`z^i`$ $`=`$ $`\alpha ^i+\beta ^iCe^{aT}+\gamma ^i\overline{C}e^{a\overline{T}}+\mathrm{}`$ (17)
where the ellipses denote higher order terms in $`Ce^{aT}`$. The coefficients $`\alpha ,\beta ,\gamma `$ are functions of the (integer) fluxes. To compare with the KKLT two stage calculation we actually need to keep terms up to second order. If we plug this expansion into the expression for $`G`$ we get
$$G=\mathrm{ln}(v+bCe^{aT}+\overline{b}\overline{C}e^{a\overline{T}}+cC^2e^{2aT}+\overline{c}\overline{C}^2e^{2a\overline{T}}+d|C|^2e^{a(T+\overline{T})}+\mathrm{})3\mathrm{ln}(T+\overline{T})$$
(18)
with the new constants (note that $`v,d`$ are real) being functions of the ones in (17) and hence of the flux integers. Calculating the potential using (8) then gives
$$V=\frac{1}{(T+\overline{T})^2}[a(bCe^{aT}+2cC^2e^{2aT}+c.c.)+a|C|^2((4\frac{a|b|^2}{v}3ad)\frac{T+\overline{T}}{3}+2d)e^{a(T+\overline{T})}]$$
(19)
Note that the terms $`cC^2e^{2aT}+c.c.`$ would not have been present if we had done the calculation in two stages as in Kachru et al. (2003a). The expression is also different even in the real direction of the potential since now we have more parameters. Since the functional dependence of the parameters in the potential on the flux integers is hard to evaluate explicitly and in any case is model dependent, we believe that the only real test of the implications of the potential coming from type IIB flux compactifications is to confront this general form of $`V`$ with experiment/observation.
In fact one of the immediate consequences of the above form of the potential is that it is possible to find metastable deSitter minima. A simple example (with just one condensate) is illustrated in the figures below. This example has the following parameters for (19): $`a=\frac{2\pi }{320},v=0.22941751641574312,b=1,c=1.4097828718993035,d=15.786002156414208,`$and $`C=1`$. The minimum is at $`\mathrm{}T_{min}=117.138,\mathrm{}T=0,V_{min}=10^{15}`$ with $`M_p=1`$.
Such potentials may also have a supersymmetric AdS minimum at $`G_T=0`$ though in this particular example this seems to be absent. Also as is the case for all moduli potentials in string theory, for large $`\mathrm{}T`$ the potential goes to zero and the positive minimum is only meta-stable.
A comment about the parameters chosen for our example is in order here. The parameters are chosen to give a positive minimum at a reasonable large value of ($`\mathrm{}T10^2`$) such that also the expenential factor $`e^{a\mathrm{}T}0.1`$ is small thus justifying the perturbation expansion of adding instanton corrections to the classical superpotential. Of course to get a critical point from such an expansion obviously two or more of such an expansion have to be of the same order. Thus here the term $`de^{a(T+\overline{T})}0.16`$ is of the same order as constant $`v.2`$ and $`be^{a\mathrm{}T}0.1`$. This doesn’t necessarily mean that the perturbation expansion is violated. It just means that some of these coefficients need to be fine tuned in order to get a critical point though the generic coefficients would be expected to be of $`O(1)`$. This is inevitable in any such calculation of the KKLT type (in the original calculation $`W_0Ce^{aT}`$at the critical point and is anomalously small) since the existence of a critical point requires that the classical terms be balanced by the non-perturbative correction terms. For perturbation theory to be violated one would need the coefficients of the higher order terms to continue to grow like $`e^{a\mathrm{}T}`$and this is extremely unlikely.<sup>7</sup><sup>7</sup>7This is reminiscent of a well known argument in large N guage theory \- the so-called Banks-Zachs fixed point - which is obtained by cancelling different orders in perturbation theory, but justified on the grounds that the resulting fixed point is still at small coupling. The argument being that one of the coefficients of the perturbation expansion is anomalously large thus giving the cancellation required to get a fixed point, but that the higher order terms could still be expected to have coefficients that did not grow with the power of the coupling constant.
So far we have worked with just one condensate. If we have several (so that $`WC_ie^{a_iT}`$) then one would need to make the following replacements
$`abCe^{aT}`$ $``$ $`{\displaystyle \underset{i}{}}a_ib_iC_ie^{a_iT}`$
$`acC^2e^{2aT}`$ $``$ $`{\displaystyle \underset{ij}{}}(a_i+a_j)c_{ij}C_iC_je^{(a_i+a_j)T}`$
$`a^2|b|^2|C|^2e^{a(T+\overline{T})}`$ $``$ $`{\displaystyle \underset{ij}{}}a_ia_jb_i\overline{b}_jC\overline{C}_je^{(a_iT+a_j\overline{T})}`$
$`a^2d|C|^2e^{a(T+\overline{T})}`$ $``$ $`{\displaystyle \underset{ij}{}}a_ia_jd_{ij}C_i\overline{C}_je^{(a_iT+a_j\overline{T})}`$
$`2ad|C|^2e^{a(T+\overline{T})}`$ $``$ $`{\displaystyle \underset{ij}{}}(a_i+a_j)d_{ij}C_i\overline{C}_je^{(a_iT+a_j\overline{T})}`$
Finally we note that if one includes the leading perturbative correction and non-perturbative corrections to the Kaehler potential then one would need to replace $`3\mathrm{ln}(T+\overline{T})3\mathrm{ln}(T+\overline{T}+f+ke^{a(T+\overline{T})})`$ with $`f,k`$ constants. Such an addition will clearly not change the qualitative features of the potential.
## V Conclusions
In this note we examined the validity of the KKLT procedure of first classically integrating out the dilaton-axion and the complex structure moduli, to obtain an effective theory for the Kaehler moduli, and then adding a non-perturbative term to the superpotential to obtain a potential that stabilizes the Kaehler moduli. We find that there is no approximation scheme in which the procedure of first integrating out the $`S`$ and $`z^i`$ fields classically and then adding a $`T`$-dependent non-perturbative term to the superpotential is justified. The latter term needs to be included from the beginning and gives additional contributions to the potential that cannot be ignored. Also we find that the procedure cannot be done holomorphically, i.e. the effective theory has to be defined entirely in terms of a Kaehler potential (in effect the Kaehler invariant function $`G`$) and a superpotential that is just unity. These considerations of course do not affect the result of KKLT that the Kaehler modulus can be stabilized by non-perturbative effects. But it does change the form of the potential for the Kaehler modulus so that the physical effects (in particular the cosmological considerations) emerging from the theory need to be reconsidered. For instance in Brustein and de Alwis (2004) it was shown that if one follows the KKLT procedure, then there is no way of getting a broken supersymmetric minimum with a positive or zero cosmological constant with just one light modulus $`T`$, (even with an arbitrary number of non-perturbative terms) without adding an uplifting term. However this is no longer the case if one correctly integrates out the heavy moduli, and we showed in an explicit example that it is possible to obtain a positive local minimum with just the F-term potential. Also the cosmological considerations based on KKLT such as Kachru et al. (2003b)Blanco-Pillado et al. (2004), need to be revisited in light of the present results. In addition to the effects considered here, perturbative corrections to the Kaehler potential also need to taken into account. A complete treatment of the physics of such models, the possibility of getting small supersymmetry breaking, a small cosmological constant, sufficient inflation etc. after including these corrections, will be discussed in forthcoming work <sup>8</sup><sup>8</sup>8R. Brustein, S.P. de Alwis and P. Martens - work in progress..
## VI Acknowledgments
I wish to thank Ramy Brustein, Marc Grisaru, Hans-Peter Nilles and especially Martin Rocek, for useful discussions, and Paul Martens for the mathematica plots. This research is supported in part by the United States Department of Energy under grant DE-FG02-91-ER-40672.
|
warning/0506/cond-mat0506563.html
|
ar5iv
|
text
|
# Extended electron states and magneto-transport in a 3-simplex fractal
## I Introduction
The physics of noninteracting electrons on lattices without any translational invariance has been the subject of intense theoretical activity since many years. Beginning with the work of Anderson pw58 describing the absence of diffusion in randomly disordered lattices, a wealth of knowledge has now accumulated in this field. Over the last couple of decades the problem of electron localization in low dimensional quasicrystalline and fractal lattice models have enriched this field further jbs85 -yg81 .
Regular fractal networks have already been appreciated to be linked to certain disordered structures, such as the percolation backbone clusters bm82 ; yg81 . Such systems, by the way they are generated, are self similar and exhibit the absence of any long ranged translational order. Yet, they are not random. This gives the fractal networks the status of being intermediate between perfectly periodic structures and completely disordered ones. Naturally, they became the objects of detailed theoretical study over all these years.
A regular fractal lattice with finite ramification (meaning that a large part can be detached by removing a finite number of bonds) is generally solvable, and is found to possess exotic spectral features edom83 -sch91 . The properties are similar for electrons and other excitations such as phonons or magnons. For example, it is found edom83 that, contrary to the absolutely continuous electronic spectrum of a perfectly crystalline sample, or the pure point spectrum of a randomly disordered lattice, the electronic spectrum of a deterministic fractal is typically a Cantor set of measure zero. The changing local environment around each lattice point leads to localization of the single particle eigenstates, but in a way not found common in conventional disordered samples. The conductance in such fractalline systems is also found to show up power-law scaling with the size of the system. The scaling exponent may be distributed in a multifractal way sch91 , as the system approaches its thermodynamic limit.
It is now known, that even a deterministic fractal may support a countable infinity of extended eigenstates xr95 -bb96 . This observation is interesting, as fractal usually does not exhibit any short ranged ‘positional correlation’ between the lattice points, as in the cases of certain random or quasiperiodic lattice models, known to support extended single particle states dun90 -ac94 . No local clusters of sites can really be identified which lead to resonance causing an extendedness in the basic character of the eigenfunction. The delocalization of an infinite number of extended single particle states in a fractal can thus be attributed to the structure of the lattice as a whole.
In this communication we report a real space renormalization group (RSRG) analysis of the spectral features of a $`3`$-simplex fractal dd77 in the presence of a constant magnetic field. The growth of a $`3`$-simplex fractal is described in Fig. 1. We choose the magnetic flux to penetrate the small triangular plaquettes at each generation, in a direction perpendicular to the plane of the fractal. This results in a break in the time reversal symmetry as the electron hops along the edges of a basic triangle. The hopping along the ‘bond’ joining the two neighboring triangle remains unaffected by the field. The magnetic field is already known to have a non-trivial effect on the Cantor set energy spectrum usually supported by such fractals as has been demonstrated in the case of a Sierpinski gasket network jr85 . The degeneracy of the zero field solution is found to be broken, and the energy spectrum broadens up, though the isolated character of the eigenstates is preserved. Later, Wang xr96 re-examined the Sierpinski gasket in the presence of a magnetic field using an RSRG method, but allowed only a subset of the full parameter space to evolve under RSRG transformations. The suppression of the evolution of the full parameter space, to our mind, sometimes may lead to incomplete information about the eigenstates.
We take up the investigation of the spectral properties of a $`3`$-simplex lattice mainly motivated by the following ideas:
First, apart from being representatives of certain percolation clusters, deterministic fractal geometries have also been realized experimentally. Modern nano-fabrication techniques have made it possible to experimentally investigate model systems, such as the Josephson junction arrays and superconducting networks developed following a Sierpinski gasket fractal jmg86 ; sek95 where both the nature, and the amount of disorder can be accurately controlled. The level of frustration in such case can be tuned by an external magnetic field. Additionally, recent experimental measurements of persistent current in an array of mesoscopic rings rab01 suggest that, one can think of a deterministic fractal network with multiple loops at all scales of length and observe the interplay of magnetic field and the fractal geometry in the transport and related issues in such networks. Detailed study in this regard has received relatively little attention so far ac97 , and deserves more analysis.
Second, we wish to have a deeper look at whether the external magnetic field, if it generates extended eigenfunctions for the fractal, is linked to any cyclic invariance of the Hamiltonian. If such cycles at all exist, it is important to understand whether they have any correlation with the values of the external flux responsible for them. In that case it would be possible to tune the flux, which is an external agent to control the transport in finite fractal networks, as well as a classification between different extended states could be achieved. This aspect which has remained, to the best of our knowledge, really un-explored so far, is a major focus of the present work.
We find several interesting features. In the absence of any field, and with the introduction of anisotropy in the amplitude of electron hopping, the corner - to - corner transmission of a $`3`$-simplex fractal turns out to be better than its isotropic counterpart. Clusters of high transmittivity pack the transmission spectrum in the former case, whereas, the system turns out to be poorly transmitting in the isotropic situation. This remains the general feature for both the isotropic and the anisotropic cases when a magnetic field is turned on, with the transmission coefficient exhibiting interesting Aharonov-Bohm (AB) oscillations.
The magnetic field is also found to have a dramatic effect on the local density of states at a corner site of an infinite (or, semi-infinite) $`3`$-simplex fractal. An ‘apparently’ continuous distribution of eigenvalues appear around the centre of the spectrum, suggesting the formation of a ‘band’ of extended states. We have not been able to prove conclusively the existence of a ‘band’ of extended states. However, extensive numerical investigation of the local density of states and the behavior of the nearest neighbor hopping integrals under renormalization group iterations are suggestive of the fact.
Finally, we have particularly focused on a specific value of the energy of the electron which supports an extended eigenstate only in the presence of a magnetic flux, and have carried out careful numerical investigations on the flow of the parameters of the Hamiltonian under repeated renormalization of the system, as the flux is changed systematically. We find that, for this specific extended eigenstate, values of the magnetic flux show up a strong correlation with the fixed point cycles of the parameter space, that can be predicted successfully. Such observations may provide an idea of classifying the extended states in regards of the fixed points of the Hamiltonian brought about by the external magnetic field.
In what follows, we report our results. In section II, we introduce the model and the RSRG scheme. Section III contains a discussion on the nature of the energy spectrum both in the absence and the presence of a field. Section IV is devoted to the investigation of magneto-transport, while in section V we discuss how the flux values can be correlated to the cycles of the fixed point. In section VI we draw our conclusions.
## II The model and the RSRG scheme
We begin by referring to Fig. 1. The three basic triangular plaquettes (Fig. 1(a)) are placed as shown in the figure to generate a second-generation fractal (Fig. 1(b)). The process continues. A magnetic field penetrates each small triangle. We work within a tight binding formalism in which the Hamiltonian for the electron in a basic triangular plaquette is written as,
$$H=\underset{i}{}ϵ_ic_i^{}c_i+\underset{<ij>}{}t_{ij}e^{i\theta }c_i^{}c_j$$
(1)
In the above, $`ϵ_i`$ is the on-site potential which, in the most general anisotropic model, can assume two values, $`ϵ_\alpha `$ (at the corner sites at the horizontal base of each elementary triangle) and $`ϵ_\mu `$ at the remaining vertices. The nearest neighbor hopping integrals $`t_{ij}`$ are assigned amplitudes $`t_x`$ and $`t_y`$ for hopping across the horizontal and the angular bonds within each elementary triangle respectively. The inter-triangle connection is given by $`t_{ij}=T_x`$ in the horizontal direction, and $`t_{ij}=T_y`$ otherwise, as depicted in Fig. 1(b), and these are free from any associated flux. It is to be appreciated that the status of a vertex ($`\alpha `$ or $`\mu `$) is governed only by the bonds $`t_x`$ or $`t_y`$ attached to it. $`T_x`$ and $`T_y`$ remain un-decimated on renormalization, and does not play a part in fixing the status of a vertex. It may also be noted at this point that, this $`3`$-simplex network differs, in the presence of a magnetic field penetrating its elementary plaquettes, from its closest look-alike the Sierpinski gasket xr96 in the fact that in the $`3`$-simplex network, the time reversal symmetry of the electron hopping is broken only partially, i.e., when the electron hops along the edges of an elementary triangle, compared to a Sierpinski gasket, where it is broken uniformly. It will be interesting to see the consequence of this at various scales of length.
The magnetic flux threading each small triangle enters the Hamiltonian only through the hopping integrals along the sides of the elementary triangles jr85 . We set $`\theta =2\pi \mathrm{\Phi }/(3\mathrm{\Phi }_0)`$. Here, $`\mathrm{\Phi }`$ is the flux threading each small triangle, and $`\mathrm{\Phi }_0`$ is the flux quantum. Following Banavar, Kadanoff and Pruisken jr85 , we select the gauge in such a way that the a factor of $`\mathrm{exp}(i\theta )`$ is associated with either $`t_x`$ or $`t_y`$ when the electron hops in the direction given by the arrow. The phase is opposite when it hops back. Thus, the ‘forward’ or the ‘backward’ hopping integrals in the presence of the field will be written as, $`t_{x(y)}^{F(B)}=t_{x(y)}e^{\pm i\theta }`$. The model, as it is presented, allows one to study a perfectly general anisotropic case ($`t_xt_y`$), including a hierarchical distribution of the bonds $`T_x`$ or $`T_y`$, which is known to exhibit a very interesting ‘restoration of isotropy’ in a $`3`$-simplex network of classical resistors lin96 , although, we do not pursue this topic here. The isotropic limit is easily retrieved as well.
The sites marked with decorated circles are now decimated to yield the following recursion relations for the on-site potentials and the hopping integrals.
$`ϵ_{\alpha ,n+1}`$ $`=`$ $`ϵ_{\alpha ,n}+C_{1,n}t_{y,n}^F+D_{1,n}t_{x,n}^B`$
$`ϵ_{\mu ,n+1}`$ $`=`$ $`ϵ_{\mu ,n}+C_{4,n}t_{y,n}^B+D_{4,n}t_{y,n}^B`$
$`t_{x,n+1}^F`$ $`=`$ $`C_{3,n}^{}t_{y,n}^B+D_{3,n}^{}t_{x,n}^F`$
$`t_{y,n+1}^F`$ $`=`$ $`C_{2,n}t_{y,n}^F+D_{2,n}t_{x,n}^B`$
$`t_{x,n+1}^B`$ $`=`$ $`t_{x,n+1}^{F}{}_{}{}^{}`$
$`t_{y,n+1}^B`$ $`=`$ $`t_{y,n+1}^{F}{}_{}{}^{}`$ (2)
In the above, $`n`$ and $`n+1`$ in the subscript denote the stages of renormalization , $`t_{}^{F(B)}{}_{j,n}{}^{}=t_{j,n}e^{\pm i\theta _n}`$ with $`j=x`$ or $`y`$ representing the $`x`$\- or $`y`$-hopping along the edge of a triangle, and, $`\theta _n`$ is the ‘re-normalized’ flux at the $`n`$th stage, which of course, we do not have to calculate separately. Here,
$`C_{1,n}`$ $`=`$ $`{\displaystyle \frac{A_{1,n}+A_{2,n}B_{1,n}}{1A_{2,n}B_{2,n}}}`$
$`C_{2,n}`$ $`=`$ $`{\displaystyle \frac{A_{3,n}+A_{2,n}B_{3,n}}{1A_{2,n}B_{2,n}}}`$
$`C_{3,n}`$ $`=`$ $`{\displaystyle \frac{A_{4,n}+A_{2,n}B_{4,n}}{1A_{2,n}B_{2,n}}}`$
$`C_{4,n}`$ $`=`$ $`{\displaystyle \frac{B_{5,n}+A_{5,n}B_{6,n}}{1A_{6,n}B_{6,n}}}`$
$`D_{1,n}`$ $`=`$ $`{\displaystyle \frac{B_{1,n}+B_{2,n}A_{1,n}}{1A_{2,n}B_{2,n}}}`$
$`D_{2,n}`$ $`=`$ $`{\displaystyle \frac{B_{3,n}+B_{2,n}A_{3,n}}{1A_{2,n}B_{2,n}}}`$
$`D_{3,n}`$ $`=`$ $`{\displaystyle \frac{B_{4,n}+B_{2,n}A_{4,n}}{1A_{2,n}B_{2,n}}}`$
$`D_{4,n}`$ $`=`$ $`{\displaystyle \frac{A_{5,n}+B_{5,n}A_{6,n}}{1A_{6,n}B_{6,n}}}`$ (3)
At any stage $`n`$ of renormalization, we have defined (suppressing the subscript $`n`$),
$`A_1=rt_y^B/s`$, $`A_2=(rt_y^F+T_xT_y^2t_x^Bt_y^B)/s`$, $`A_3=T_y(qt_y^F+pt_y^Bt_x^B)/s`$, $`A_4=T_y^2t_x^B[(Eϵ_\alpha )t_y^F+t_x^Bt_y^B]/s`$, $`A_5=t_y^Bz_1/[z_1(Eϵ_\alpha )T_y^2]`$, and, $`A_6=(z_1t_x^F+v_1T_yt_y^B)/[z_1(Eϵ_\alpha )T_y^2]`$.
$`B_1=t_x^F/z`$, $`B_2=(rt_y^B+T_xT_y^2t_x^Ft_y^F)/(rz)`$, $`B_3=uT_xT_yt_y^F/(pz)`$, $`B_4=T_x[wT_yt_y^F+t_x^B(Eϵ_\mu )+(t_y^F)^2]/(pz)`$, $`B_5=t_y^F/z_2`$, and, $`B_6=(T_yw_2+t_x^B)/z_2`$.
The quantities $`u`$, $`w`$, $`z`$, $`v_1`$, $`z_1`$, $`w_2`$ and $`z_2`$ are defined as,
$`u`$ $`=`$ $`{\displaystyle \frac{p\left[t_x^F(qt_y^F+pt_y^Bt_x^B)+rt_y^B\right]}{qr}}`$
$`w`$ $`=`$ $`{\displaystyle \frac{T_y[t_x^Bt_y^B+(Eϵ_\alpha )t_y^F]}{qr}}(pt_x^Ft_x^B+r)`$
$`z`$ $`=`$ $`(Eϵ_\alpha ){\displaystyle \frac{T_x[(Eϵ_\mu )T_x+t_y^FT_yv]}{p}}`$ (4)
with, $`v=[pT_xT_yt_x^Ft_x^Bt_y^B+rT_xT_yt_y^B]/(qr)`$,
$`v_1`$ $`=`$ $`{\displaystyle \frac{T_xT_yt_y^B}{p(Eϵ_\alpha )T_x^2(Eϵ_\mu )}}`$
$`z_1`$ $`=`$ $`(Eϵ_\mu )w_1t_y^B`$
with, $`w_1=t_y^Fp/[p(Eϵ_\alpha )T_x^2(Eϵ_\mu )]`$, and,
$`w_2`$ $`=`$ $`{\displaystyle \frac{(t_y^F)^2T_xT_y}{(Eϵ_\mu )[p(Eϵ_\alpha )T_x^2(Eϵ_\mu )]pt_y^Ft_y^B}}`$
$`z_2`$ $`=`$ $`(Eϵ_\alpha )T_yv_2`$ (6)
with, $`v_2=\frac{T_y[p(Eϵ_\alpha )T_x^2(Eϵ_\mu )]}{(Eϵ_\mu )[p(Eϵ_\alpha )T_x^2(Eϵ_\mu )]pt_y^Ft_y^B}`$.
Finally, the remaining factors are,
$`p`$ $`=`$ $`(Eϵ_\alpha )(Eϵ_\mu )t_y^Ft_y^B`$
$`q`$ $`=`$ $`(pT_y^2)(Eϵ_\alpha )`$
$`r`$ $`=`$ $`q(Eϵ_\alpha )pt_x^Ft_x^B`$
$`s`$ $`=`$ $`r(Eϵ_\mu )qT_y^2`$
The inter-triangle hopping integrals $`T_x`$ and $`T_y`$ remain un-affected as a result of renormalization, i.e,. $`T_{j,n+1}=T_{j,n}`$ at any stage $`n`$, $`j`$ representing $`x`$ or $`y`$.
The set of recursion relations given by Eq. (2) is a highly non-linear one, and can be reduced to a simple form only under simple isotropic model in the absence of any field. However, they are not so difficult to deal with numerically, and yield quite a few interesting results which we shall now discuss.
## III Energy spectrum and the nature of eigenstates
### III.1 The zero field case
To begin with, we have evaluated the local density of states (LDOS) at a corner site of a $`3`$-simplex gasket in the isotropic limit, and in the absence of any magnetic field. We set $`ϵ_\alpha =ϵ_\mu =ϵ`$, $`t_x=t_y=t`$, and, $`T_x=T_y=\tau `$. The recursion relations given by Eq. (2) now get reduced to,
$`ϵ_{n+1}`$ $`=`$ $`ϵ_n+{\displaystyle \frac{P_n}{Q_nR_n}}`$
$`t_{n+1}`$ $`=`$ $`{\displaystyle \frac{U_n}{Q_nR_n}}`$ (8)
with, $`P_n=2t_{n}^{}{}_{}{}^{2}[(Eϵ_n)^2\tau (Eϵ_n)t_n^2]`$, $`Q_n=(\tau +t_n+ϵ_nE)`$, $`R_n=\tau ^2(Eϵ_n)^2+t_n^2\tau t_n`$ and, $`U_n=\tau t_n^2(Eϵ_n+t_n\tau )`$. $`n`$ again stands for the stage of renormalization. In the absence of any field, it is a simple task to evaluate the LDOS at a corner site using the standard Green’s function technique bw83 . We have checked that the recursion relations produce the correct LDOS at the corner site of a one dimensional chain ad97 in the limit $`t_y,T_y0`$. With a small imaginary part added to the energy $`E`$, the hopping integral $`t`$ flows to zero after certain steps of RSRG. This implies that the lattice, at that scale of length, breaks up into an assembly of diatomic molecules, with the hopping $`\tau `$ connecting the ‘atoms’ remaining unchanged. Each such molecule will be decoupled from its neighbors, and the transmission across the lattice will be zero. The on-site potential at the corner site reaches its fixed point value $`\stackrel{~}{ϵ}`$, and, the LDOS (meaningful only at an extreme corner site, which is truly decoupled from the rest of the lattice) is then evaluated as,
$$\rho (E)=\frac{1}{\pi }G_{00}(E+i\eta )$$
(9)
where, the diagonal Green’s function $`G_{00}=1/(E\stackrel{~}{ϵ})`$. A plot of the LDOS at a corner site is shown in Fig. 2 (top figure). The LDOS is exhibited is checked, as permitted by the limit of accuracy, to be stable against a decreasing value of the imaginary part of the energy $`E`$, and has been displayed within a value unity to give prominence to the smaller peaks compared to the much bigger ones. The fragmented, scanty appearance is consistent with the usual Cantor set spectrum common to such systems.
Before introducing the magnetic field, it is pertinent to comment on the existence of extended eigenstates in this simplified model. If the recursion relations Eq. (8) are iterated for any arbitrary energy with no imaginary part added to it, the hopping integral $`t`$ flows to zero after certain steps of RSRG. This means that the corresponding energy should lie in a gap of the spectrum of the infinite system, or, corresponds to an exponentially localized eigenstate, though it is not possible to distinguish between these two cases by simply looking at the flow of the hopping integral. It should be remembered that a zero LDOS is not a conclusive proof for an energy not to be in the spectrum of the infinite lattice. On the other hand, if, for certain energy, the hopping integral remains non-zero under an indefinite number of iterations, then we have definitely hit upon an extended eigenstate. It is not difficult, using Eq.s (8) to fix an energy for which $`ϵ_1=ϵ`$. The energy, evaluated in this fashion can then be inserted into the second equation in (8), and one can select $`T`$ in such a manner that $`t_1`$ under RSRG remains equal to $`t`$. As the inter-triangle hopping is always $`T`$, we thus have a fixed point of the parameter space, viz, $`(ϵ_n,t_n,T_n)(ϵ_{n1},t_{n1},T_{n1})`$ for $`n1`$. The corresponding state will be extended in nature. This of course demands a definite relationship between $`ϵ`$, $`t`$ and $`T`$ , that is, we are talking of a specific model. For example, with $`ϵ=0`$, and $`t=1`$, if we set $`E=0.5t`$, then the fixed point behavior sets in from the first RSRG step onwards if $`T`$ is assigned a value equal to $`3/2`$ (in units of $`t`$). Much more involved relationship between the parameters of the system is capable of leading to a fixed point behavior starting at deeper scales of renormalization. However, in each case, a different set of parameters essentially means that we are dealing with a different system. In principle, we will observe extended eigenstates in each of those cases which provide a meaningful relationship between the parameters, but only at certain discrete energy values.
### III.2 A magnetic field is turned ‘on’
A magnetic field, however, is capable of bringing dramatic changes into the spectrum even when we deal with a particular system with a pre-defined set of parameters. In the lower one in Fig. 2 the LDOS at the corner site is shown for a flux $`\mathrm{\Phi }=\mathrm{\Phi }_0/4`$ for the simplest isotropic model with $`ϵ_\alpha =ϵ_\mu =0`$, and $`t_x=t_y=T_x=T_y=1`$. The spectrum shows very closely spaced zones of finite density of states. Of particular interest is the narrow energy interval around the ‘centre’ $`E=0`$, where, the appearance of a smooth region of almost constant density of states exits. It has not been possible to prove the existence of a continuous band of states, though we have made very fine scan around $`E=0`$, each time reducing the energy interval to be scanned and diminishing the imaginary part to be added to the energy $`E`$. The LDOS is found to be stable under a variation in the imaginary part $`\eta `$ from $`10^3`$ to $`10^{10}`$ (in unit of $`t`$), and within the limits of machine accuracy, it is tempting to conjecture the existence of a continuous zone of eigenstates around $`E=0`$. The occurrence of dense clusters of non-zero density of states has been tested with other values of the magnetic field. The general conclusion is the same (except for $`\mathrm{\Phi }=m\mathrm{\Phi }_0/2`$, $`m`$ being an integer) , and the qualitative features do not essentially change even when we deal with the anisotropic gasket.
## IV Transmission across finite 3-simplex fractals
To calculate the quantum mechanical transmission across a finite $`3`$-simplex network of any size, we adopt the very well known formalism proposed by Douglas Stone et al . stone81 , and consequently used by others as well bb96 . The essential method consists in placing the desired fractal network between perfectly ordered semi-infinite leads (shown by dashed lines in Fig.1(b)) connected to the two extreme $`\alpha `$-sites at the base. The leads may be described by a uniform on-site potential $`ϵ_0`$, and constant nearest neighbor hopping integral $`t_0`$. A network at the $`n`$th generation is then renormalized $`n1`$ times to reduce it to a simple triangle, and finally to a diatomic ‘molecule’ bb96 , still clamped between the leads, with an effective on-site potential $`ϵ_{eff}`$, and an effective hopping integral $`t_{eff}^{F(B)}`$. The transmission coefficient is then easily obtained in terms of the quantities $`ϵ_{eff}`$, $`t_{eff}^{F(B)}`$, $`ϵ_0`$, $`t_0`$ and the electron energy $`E`$. The method is so well known that we skip the detailed mathematical expressions to save space, and present the results only.
### IV.1 Zero flux situation
Let us start with the isotropic case, that is, $`ϵ_\alpha =ϵ_\mu `$, and, $`t_x=t_y=T_x=T_y`$. Consider no flux, i.e., $`\mathrm{\Phi }=0`$. The transmission spectrum consists of the expected isolated peaks, consistent with the LDOS spectrum. With increasing size the fractal turns out to be even poorly transmitting. Interesting changes however start showing up with the introduction of anisotropy. Quite arbitrarily, we set , $`t_x=T_x=1`$ and start reducing the values of $`t_y(=T_y)`$ from one towards zero (the limit when the fractal reduces to a linear chain clamped between the leads). The on-site potentials $`ϵ_\alpha `$ and $`ϵ_\mu `$ are chosen to be equal, and have been set equal to zero. With the introduction of anisotropy, regions of finite transmission increase in number. For example, putting $`t_y=T_y=0.9`$ (that is, a small departure from isotropy), the spectrum is still very much like the isotropic situation, with new clusters of appreciable transmittance (sometimes unity as well) appearing in many places. With gradual decrease in the values of the $`y`$-hopping, the newly generated small spiky zones increase in number, join ‘hand in hand’ and the shape of the entire spectrum start drifting towards what it should be in the case of a periodic chain clamped between the leads. The spectrum tends to be restricted within $`E=ϵ_0\pm 2t_0`$, which is the allowed band of the ordered lead, and resembles the spectrum of a $`1`$-d chain, as $`t_y=T_y`$ becomes vanishingly small. In Fig. 3 we show the transmission spectrum of a $`7`$-th generation $`3`$-simplex structure, both for the isotropic situation (top), and the anisotropic cases (the middle and bottom ones) in support of the above remarks.
### IV.2 Influence of the magnetic field
We have investigated the general features of the transmission coefficient as the flux through each elementary plaquette is varied. The features are of course sensitive to the energy of the electron that enters the system through the lead. For calculation, we have chosen $`ϵ_\alpha =ϵ_\mu =0`$, and $`t_x=T_x=t_y=T_y=1`$. The lead parameters in this case are chosen to be $`ϵ_0=0`$, and $`t_0=2`$ to encompass the full spectrum of the fractal network. In the case of an isotropic $`3`$-simplex network, the magnetic field is found to generate clusters of resonant transmission throughout the spectrum, a particularly noticeable broadening taking place at and around $`E=0`$.
In Fig. 4 we display the transmission spectrum of a sixth generation fractal with energy $`E`$ with a flux $`\mathrm{\Phi }=0.25\mathrm{\Phi }_0`$ threading each elementary triangle. The spectrum is notable for a thick population of high transmission values at and around $`E=0`$. Several other mini bands of resonant transmission also mark the spectrum. This feature is in sharp contrast to that in the absence of any field (Fig. 3, the top figure for example). We have observed the change in the width of the cluster of high transmission zone around $`E=0`$ by carefully scanning the energy interval. It is seen that the width of the transmission window around $`E=0`$, for any generation, changes continuously with the introduction of the flux , from zero at $`\mathrm{\Phi }=0`$ to a maximum at $`\mathrm{\Phi }=\frac{\mathrm{\Phi }_0}{4}`$, and then shrinks back to zero again at $`\mathrm{\Phi }=\frac{\mathrm{\Phi }_0}{2}`$. The change in width of this central transmitting window shows a periodic variation with flux. The period is equal to a half flux quantum. Similar observations as above are made with anisotropy in the hopping integrals, but with no real new qualitative features. The portion of the $`T`$ \- $`\mathrm{\Phi }/\mathrm{\Phi }_0`$ graph immediately around $`E=0`$ remains densely packed with increasing size of the network.
The second half of the study of transmission spectrum consists of an examination of the Aharonov-Bohm (AB) oscillations in the transmission coefficient at a fixed energy of the electron. We have displayed results for $`E=0`$. The period of oscillations is found to be equal to one flux quantum. The detailed features of the spectrum are of course sensitive to the chosen energy of the electron. In Fig. 5 we show the AB oscillations within one period for a third and a sixth generation gasket. Once again, the transmission window between zero and one-half flux quantum shows multiple resonance peaks in the sixth generation fractal compared to a fairly broad and structureless shape observed in the third generation. Interestingly, with increasing generation, the spectrum is enriched by the appearance of multiple peaks with transmission equal to one (or, very close to one), but, the transmittivity really doesn’t fall to zero for any appreciable value of the flux between $`0<\mathrm{\Phi }<\mathrm{\Phi }_0/2`$, and, $`\mathrm{\Phi }_0/2<\mathrm{\Phi }<\mathrm{\Phi }_0`$.
The sensitivity of the spectral features on the parameters of the system are easily revealed when we look at the variation of the transmission coefficient against changing magnetic flux in the case of an anisotropic simplex lattice. While the electron with $`E=0`$ doesn’t distinguish between an isotropic and an anisotropic fractal, other energy values may lead to gross changes in the fine structure of the spectrum. We point out an interesting phenomenon. It is possible to fix up the electron energy in such a way, that, with large anisotropy (that is, for low enough values of $`t_y=T_y`$) , the amplitudes of the AB oscillations start decreasing as the transmission coefficient assumes values very close to one. At one stage, the transmittance $`T`$ becomes practically indistinguishable from unity, as $`t_y=T_y`$ is brought below certain value. By magnifying the scale of observation, it is still possible to see how the system tries to preserve the AB oscillations, which soon smoothes out if the $`Y`$-hopping is diminished further. In such cases, with increasing generation, the hopping integral $`t_{y,n}`$ flows to zero, while $`t_{x,n}`$ does not. It means, as the system grows in size, the intricate geometry of the fractal starts ‘disappearing’ to the incoming electron. It essentially feels an ordered chain, and if the energy chosen happens to lie in the allowed band of that ordered chain, we get a ballistic transport. In Fig. 6 we display one such example where, $`E=0.5`$ in unit of $`t_x`$.
## V Field induced extended states
### V.1 General remarks
The existence of extended eigenstates in systems without any translational order has always been an intriguing feature in the study of disordered systems. However, the complex forms of the recursion relations make an analytical attempt rather difficult for a $`3`$-simplex network. We have therefore relied on a careful numerical study. We specially observe the flow of the hopping integrals under successive RSRG steps. The non-zero values of the hopping integrals under RSRG stand out to be a definite signature of the state being extended. In all the discussion that follows, we confine ourselves to the isotropic case only.
It is now evident that, the magnetic field generates a dense set of eigenvalues, almost resembling a continuum, around $`E=0`$. Quite arbitrarily we have selected a portion $`0.1E0.1`$, in unit of $`t_x`$ in a model with $`ϵ_\alpha =ϵ_\mu =0`$, $`t_x=T_x=t_y=T_y=1`$, and have chosen the value of the magnetic flux $`\mathrm{\Phi }=\mathrm{\Phi }_0/4`$. The LDOS in this portion is found to be very stable as the imaginary part added to the energy is decreased from $`10^3`$ to $`10^9`$. A fine scan over the points in this interval reveals that, any energy we hit upon quite randomly in this range, corresponds to a non-zero value of the nearest neighbor hopping integral under successive renormalization. This indicates the presence of extended eigenstates. Of course, this is not a conclusive proof of the existence of a band of extended states, but, the smaller and smaller widths of the energy interval chosen, remaining close to $`E=0`$, lead to similar behavior. This is suggestive of the fact that the presence of a band of extended eigenstates may not be a remote possibility in this case. We have carried out the numerical investigation for other values of the flux and other energy intervals, and in many occasions similar observation has been made. The observations are in accord with the transmission spectrum displayed in Fig. 5.
### V.2 Values of Flux and cycles of the fixed point: an interesting correlation
Let us now turn to a different aspect of the problem. We restrict the discussion to the isotropic model with $`ϵ_\alpha =ϵ_\mu =ϵ`$, and $`t_x=T_x=t_y=T_y=t`$. The phase associated with the hopping, at any $`n`$th stage of RSRG will be denoted by $`\theta _n`$. We draw the attention of the reader to Fig. 7 which presents the transmission as a function of the applied flux for a seventh generation $`3`$-simplex fractal (solid line), together with the LDOS (dashed line) at $`E=0`$, plotted by varying the flux from zero to a single flux quantum. It is interesting to see how the energy $`E=0`$ is periodically brought in and out of the spectrum of the infinite system by the magnetic field, the period being equal to $`\mathrm{\Phi }_0/2`$. Once again, we have tested the robustness of the LDOS spectrum by diminishing the imaginary part added to energy from bigger $`(10^3)`$ to much smaller $`(10^9)`$ values, so that it is not unjustified to conclude that we definitely have a state at $`E=0`$. The spectrum also gives us an indication that we have a continuous LDOS as flux changes from zero to half flux quantum. We fix our energy of interest at $`E=0`$. Looking at the evolution of the hopping integral under successive RSRG steps as we gradually increase the flux threading an elementary triangular plaquette from zero to $`\mathrm{\Phi }_0/2`$ and beyond, we make two interesting observations. First, for $`0<\mathrm{\Phi }<\mathrm{\Phi }_0/2`$, and, $`\mathrm{\Phi }_0/2<\mathrm{\Phi }<\mathrm{\Phi }_0`$, the hopping integrals do not flow to zero under iteration, bringing out the fact that $`E=0`$ corresponds to a perfectly extended eigenstate for what appears to be a continuous distribution of flux values.
Second, different flux values chosen for the above observation unravel the existence of fixed points with multiple cycles. Most interestingly, we have found that the values of the magnetic flux leading to these cyclic fixed points group into a definite pattern, the formation of which can be predicted from the results of our numerical scan of the range of the flux values. The same pattern (of the number of cycles) is repeated as we change the flux at certain specially chosen equal intervals along the flux line from $`\mathrm{\Phi }=0`$ to $`\mathrm{\Phi }=\mathrm{\Phi }_0/2`$. We clarify the meaning of the above statements below by citing specific results.
Let us consider the interval $`\mathrm{\Phi }=0`$ to $`\mathrm{\Phi }=\mathrm{\Phi }_0/2`$, and divide this interval into $`2^m`$ subintervals with $`m`$ being a positive integer, and $`m2`$. If we exclude the two extreme points at $`\mathrm{\Phi }=0`$ and $`\mathrm{\Phi }=\mathrm{\Phi }_0/2`$ on this ‘flux line’, then the remaining values of the magnetic flux making this division are distributed at the locations (coordinates) $`\mathrm{\Phi }_j=\frac{j}{2^{m+1}}\mathrm{\Phi }_0`$, $`j=1,2,3,\mathrm{}.,2^m1`$. Let us pick up the simplest choice, viz, $`m=2`$. The interval $`0<\mathrm{\Phi }<\mathrm{\Phi }_0/2`$ is then split into four equal sub-intervals. The values of the flux inside the line (that is, excluding the values at the boundary) are located at $`\mathrm{\Phi }_1=\mathrm{\Phi }_0/8`$, $`\mathrm{\Phi }_2=\mathrm{\Phi }_0/4`$ and $`\mathrm{\Phi }_3=3\mathrm{\Phi }_0/8`$, sequentially from the left. We set $`\mathrm{\Phi }=\mathrm{\Phi }_1=\mathrm{\Phi }_0/8`$. It is immediately observed that, the entire parameter space defined by the trio $`(ϵ_n,t_n,\theta _n)`$ gets locked into a $`2`$-cycle fixed point beginning at $`n=3`$, where $`n`$ represents the RSRG step. That is, we have for $`\mathrm{\Phi }=\mathrm{\Phi }_0/8`$,
$$(ϵ_n,t_n,\theta _n)=(ϵ_{n+2},t_{n+2},\theta _{n+2}),n3$$
With $`\mathrm{\Phi }=\mathrm{\Phi }_2=\mathrm{\Phi }_0/4`$, we again get a $`2`$-cycle fixed point of the parameter space $`(ϵ_n,t_n,\theta _n)`$. But now, the fixed point behavior is observed for $`n2`$. Setting the external flux equal to the remaining value in this interval, viz, $`\mathrm{\Phi }=\mathrm{\Phi }_3=3\mathrm{\Phi }_0/8`$, we get a $`1`$-cycle fixed point of the same parameter space beginning at $`n=1`$, that is, at the first stage of RSRG onwards. It should be appreciated that, the values of the flux $`\mathrm{\Phi }=\mathrm{\Phi }_0/8`$ and $`\mathrm{\Phi }=\mathrm{\Phi }_0/4`$ can be termed ‘equivalent’ only in the number of cycles of the fixed point they generate. The extended wavefunctions they represent, are characteristically different as the invariant cycles of the parameter space set in at different stages of renormalization.
We now increase the value of $`m`$ in steps. The behavior of the parameter space defined by $`(ϵ_n,t_n,\theta _n)`$ is observed under successive RSRG iteration as the magnetic flux is made to assume sequential values given by $`\mathrm{\Phi }=\frac{j}{2^{m+1}}\mathrm{\Phi }_0`$ from the left. Our ‘experiment’ leads to the following very interesting observations:
(i) If we denote the flux leading to an $`n`$-cycle fixed point by $`\mathrm{\Phi }[n]`$, then the values of the flux between zero and the half flux quantum group themselves into a series of triplets, viz, $`(\mathrm{\Phi }[2],\mathrm{\Phi }[2],\mathrm{\Phi }[1])`$.
(ii) This triplet repeats itself periodically as we sweep through the points $`\mathrm{\Phi }_j=j/2^{m+1}\mathrm{\Phi }_0`$ along the flux axis between zero and half flux quantum. For any given value of $`m`$ which fixes the number of intervals, the cyclic invariance of the full parameter space will start showing up at a specific step of renormalization. This ‘step’ $`n`$ is given by the ‘power’ $`m`$ of $`2`$ in the denominator in the expression for the flux $`\mathrm{\Phi }`$, whenever $`\mathrm{\Phi }=\frac{(2l+1)}{2^m}\mathrm{\Phi }_0`$, $`l=0,1,2,\mathrm{}`$.
Thus, for an eight sub-interval splitting of the range $`0\mathrm{\Phi }\mathrm{\Phi }_0/2`$, the first value (after zero) at $`\mathrm{\Phi }=\frac{1}{2^4}\mathrm{\Phi }_0`$ exhibits a $`2`$-cycle fixed point starting at $`n=4`$. The second flux $`\mathrm{\Phi }=\frac{2}{2^4}\mathrm{\Phi }_0=\frac{1}{2^3}\mathrm{\Phi }_0`$ leads to a $`2`$-cycle fixed point beginning at $`n=3`$. For $`\mathrm{\Phi }=\frac{3}{2^4}\mathrm{\Phi }_0`$ we have a $`1`$-cycle fixed point for $`n4`$. This sequence of cycles repeats periodically, but the stage at which the invariance starts showing up is not the same for every flux. However, for values of $`m>2`$, a subset of the fixed points ($`1`$-cycle, or $`2`$-cycle) are truly equivalent in the sense that they start showing up at the same stage of RSRG. In this sense, the fixed points arising out of flux values $`\mathrm{\Phi }=\mathrm{\Phi }_0/2^m`$, $`\mathrm{\Phi }=5\mathrm{\Phi }_0/2^m`$, $`\mathrm{\Phi }=7\mathrm{\Phi }_0/2^m`$ are truly equivalent.
It should be appreciated that the sequence of flux values responsible for a certain cyclic behavior has a deterministic feature. If we select $`\mathrm{\Phi }=\frac{1}{2^m}\mathrm{\Phi }_0`$, with $`m=2`$, $`3`$, $`4`$,$`\mathrm{}`$ sequentially, then its easy to check that for all such values of $`\mathrm{\Phi }`$ we have a $`2`$-cycle fixed point beginning at $`n=m`$. Thus, at the left-most value of the flux, the cyclic invariance starts revealing at the deepest scale of length. As we move along the flux line, the fixed point character begins to show up earlier, and at $`\mathrm{\Phi }=\varphi _0/4`$, we see it immediately from $`n=2`$ onwards. Similarly, whenever $`\mathrm{\Phi }=\frac{3p}{2^m}\mathrm{\Phi }_0`$, with $`p=1`$, $`2`$, $`3`$,$`\mathrm{}`$, and $`m2`$, we come across a $`1`$-cycle fixed point for $`nm`$. For example, by selecting $`m=5`$, the values of the flux at $`\mathrm{\Phi }/\mathrm{\Phi }_0=3/2^5`$, $`6/2^5=3/2^4`$, $`9/2^5`$, $`12/2^5=3/2^3`$, and $`15/2^5`$ exhibit $`1`$-cycle fixed point beginning at $`n=5`$, $`4`$, $`5`$, $`3`$ and $`5`$ respectively. Similar deterministic feature has also been possible to locate for other sub-divisions.
It is obvious that the separation between the one cycle, and the two-cycle values of the flux becomes exponentially smaller as we split the flux interval between zero and one-half flux quantum more and more by increasing $`m`$. The behavior of ($`\mathrm{\Phi }[2],\mathrm{\Phi }[2],\mathrm{\Phi }[1])`$ is consistent with our expectation, so far as we have tested. This encourages us to conclude that, speaking just in terms of the one and two-cycle fixed points, there will be a ‘quasi-continuous’ cross-over in the character of the extended eigenstates as one sweeps over the specific flux values at $`\mathrm{\Phi }_j=\frac{j}{2^{m+1}}\mathrm{\Phi }_0`$ along the flux line between zero and half flux quantum.
What happens in the range $`\mathrm{\Phi }_0/2\mathrm{\Phi }\mathrm{\Phi }_0`$ ?
Similar correlations between the flux values and the cycles of the fixed points may also be looked for the flux ranging from $`\mathrm{\Phi }=\mathrm{\Phi }_0/2`$ to one flux quantum. The pattern however, may be different. For example, if we split the full range of flux, from $`\mathrm{\Phi }=0`$ to $`\mathrm{\Phi }=\mathrm{\Phi }_0`$, in $`2^5=32`$ equal intervals, then it is seen that from $`\mathrm{\Phi }=j\mathrm{\Phi }_0/32`$, with $`j=1,2,\mathrm{},15`$, the pattern $`(\mathrm{\Phi }[2],\mathrm{\Phi }[2],\mathrm{\Phi }[1])`$ is periodically repeated. The mid-point $`\mathrm{\Phi }=\mathrm{\Phi }_0`$ has to be omitted as we do not get any state there. In the rest of the flux interval, that is, from $`\mathrm{\Phi }=17\mathrm{\Phi }_0/32`$ and upto $`\mathrm{\Phi }=31\mathrm{\Phi }_0/32`$ in interval of $`\mathrm{\Phi }_0/32`$, we now get a triplet $`(\mathrm{\Phi }[2],\mathrm{\Phi }[1],\mathrm{\Phi }[2])`$ repeating periodically. One can proceed in this way for other values of $`m`$. For any given $`m`$, the entire pattern observed between $`\mathrm{\Phi }=0`$ and $`\mathrm{\Phi }=\mathrm{\Phi }_0`$ obviously repeats beyond one flux quantum. We have tested these observations, by working out the patterns for a few values of $`m`$ to begin with, and then speculated the pattern for larger values of $`m`$. This test has been successful and gives us confidence to predict the correlation as it has been described above.
Before we end this section, it is good to note that, we have chosen to speak in terms of the one and the two cycle fixed points only. There are, other flux values as well for which one gets different multiple cyclic behavior, even a completely chaotic behavior of the parameter space. All these point towards the existence of extended eigenstates in a $`3`$-simplex network. However, we have tried to focus on a definite correlation between cycles of invariance of the parameter space, and a given set of values of the magnetic flux by citing the above example.
## VI Conclusion
We have examined the spectral properties of a $`3`$-simplex fractal network in the presence of a magnetic field penetrating a subspace of this fractal space. Both the isotropic and the anisotropic limits of the model have been discussed with special emphasis on the flux dependent electronic transmission and the existence of extended electronic states. Based on a numerical study of the exact renormalization group recursion relations we show that there is a subtle correlation between the value of the magnetic flux and the fixed point behavior of the Hamiltonian, and propose that such an observation may lead to a method of classification of the extended eigenstates for a given value of the energy of the electron.
Acknowledgment
The author is grateful to the Max Planck Institute für Physik Complexer Systeme, in Dresden, for their kind hospitality and for supporting the present work. Special thanks must be given to Magnus Johansson for a critical reading of the manuscript and for valuable comments.
|
warning/0506/physics0506063.html
|
ar5iv
|
text
|
# 1 Introduction and Background
## 1 Introduction and Background
The use of fractal analysis methods to study structure in art and music is not a new field. Recently, the question of perceptability of such fractal structure has been addressed. The authors of References pose the question of whether or not humans are “attuned” to the perception of fractal-like optical and auditory stimuli. Similarly, suggests that there is a fractal-like signature in memory processes which can be detected in the statistical variance of averaged repeated actions (such as repeated drawing lines of specific lengths or shapes; the statistical variations in the lengths are shown to be not purely random noise, but fractally ordered “$`1/f`$” noise).
In the visual arts, there have been several contributions made by the authors of , in which the paintings of Jackson Pollock play a prominent role. Of their many interesting conclusions, the most striking is that Pollock’s drip paintings almost uniformly possess a fractal dimension around 1.7. This was confirmed by the authors of , who also extended the study to other artists of the abstract expressionist school (notably the Québec-based group Les Automatistes). It was discovered that many of the paintings from these other artists possess a similar fractal dimension.
Similar research on computer-generated cosmological models suggests that the fractal (or box) dimension is a vague statistic for identifying structural differences in point sets, and that the full multifractal spectrum can yield deeper information as to the nature of the distributions . However, it has been concluded that the utility of the method seems limited to identifying classes of distributions formed by different mechanisms, and not differences between individual members of the same class.
In the following paper, this method will be applied to two-dimensional, non-representational images, to ascertain whether or not similar statements may be made about the analysis. Summary results of this study have been reported in Reference , but this paper will greatly expand upon the data and present technical details of the analysis in diverse ways. An overview of fractal and multifractal theory is first presented. Following this, the fractal (box) dimension for several abstract expressionist paintings by different artists is performed, and this is contrasted with the information dimension for the same works. The box method is tested for robustness in Section 5.
As a control, these abstract expressionist are contrasted with the deterministic “Artonomy” paintings by Tsion Avital . Like the three dimensional distributions, the two former works can be interpreted to be formed by one type of mechanism (although the comparison is perhaps more ambiguous), while the latter a decidedly different mechanism (this differentiation will be further discussed in Section 8).
The analysis will then be extended to include the full multifractal spectrum for each work. A comprehensive analysis is performed on color patterns for the artists in question, beginning in Section 7. Furthermore, issues addressing potential image reconstruction from the multifractal spectrum is discussed in Section 9. Finally, as a toy model for visual discrimination based on the notion that human perception may be influenced by contrast edges instead of colors, the multifractal spectra of contrast patterns in the paintings are analyzed in Section 10. Some limitations of the method are discussed in Section 11, and concluding remarks are summarized in Section 12.
## 2 Theory of Multifractals
The similarity in form and function of the classic fractal (or box) dimension, Shannon’s information dimension, and the statistical correlation dimension is not a coincidence (see References for general details on these dimensions). In fact, these quantities are but three members of an infinite set of dimensions which characterize a fractal set. Since first being introduced as a method of describing or quantifying the behavior of strange attractor sets and turbulent flows, multifractal analysis has gained steady momentum in physics and fields abroad . Comprehensive reviews of multifractals and their applications in the physical sciences are available in the aforementioned references, as well as such works as , in addition to many of the references cited hereafter.
Classic geometric “monofractals” such as the Koch snowflake or Sierpinski carpet are defined by a single scale invariant behavior, which of course is the fractal dimension, but in many cases a single such power law fails to characterize completely the distribution in question. Multifractals, on the other hand, may be regarded as an intricate weave of an infinite number of fractals, all of which are characterized by different (local) scaling dimensions. That is, each subset forms a “sub-fractal” describing a distinct sub-structure of the whole. It is much more reasonable and realistic to expect natural objects to exhibit this behavior.
Multifractal dimensions are generalizations of the Hausdorff measure . The partition function for an $`ϵ`$-covering (i.e. balls of radius $`ϵ_i`$) is defined as
$$\mathrm{\Gamma }(q,\tau )=\underset{i}{}\frac{p_i^q}{ϵ_i^\tau },$$
(1)
with $`p_i`$ a measure of the set or pattern density in the ball. For given $`q,\tau (q)`$, take the supremum (or infimum, depending on whether $`q`$ is positive or negative, respectively) in the limit $`ϵ_i0`$ (thus find the minimal covering set for the generalized measure). Then, there exists a critical value of $`\tau \tau (q)`$ for which $`\mathrm{\Gamma }(q,\tau )`$ goes from convergence to 0, to divergence to $`\mathrm{}`$. At this transitionary value, the sum converges as $`\mathrm{\Gamma }(q,\tau )=\mathrm{const}`$. The minimal covering may be generalized to balls of equal radii $`ϵ_i=ϵ`$, whence it follows that
$$\mathrm{\Gamma }(q,\tau )=ϵ^{\tau (q)}\underset{i}{}p_i^q1,$$
(2)
for a suitable renormalization of the measure. Hence,
$$\underset{i}{}p_i^qϵ^{\tau (q)},$$
(3)
and thus in the limit $`ϵ0`$, it can be shown that
$$\tau (q)=\underset{ϵ0}{lim}\frac{\mathrm{log}[_ip_i^q]}{\mathrm{log}ϵ}.$$
(4)
Let the measure partition function over $`N(ϵ)`$ balls of radius $`i`$ be
$$Z(q,ϵ)=\underset{i=1}{\overset{N(ϵ)}{}}[p_i(ϵ)]^q,$$
(5)
and define the general scaling relation $`Z(q,ϵ)ϵ^{(q1)D_q}`$ , which ensures recovery of the Hausdorff dimension for $`q=0`$, as well as normalization of (5) for $`q=1`$. From (4), it can be concluded that
$$D_q=\frac{\tau (q)}{q1}.$$
(6)
In the limit $`q0`$, (1) reduces to the usual box dimension. Furthermore, the information and correlation dimensions are recovered in the limits $`q1,2`$.
In practical applications, the box counting method can be generalized to obtain the values of $`\tau (q)`$ for given $`q`$. That is, modify the partition function $`Z(q,ϵ)`$ over $`N(ϵ)`$ to sets of covering boxes (instead of balls) of equal side $`ϵ`$, where as before $`p_i(ϵ)`$ is the relative density of the set in box $`i`$. The scaling information of each moment is obtained by taking the logarithmic derivative of (5) with respect to box size, where
$$\tau (q)=\frac{d\mathrm{log}[Z(q,ϵ)]}{d\mathrm{log}(ϵ)}.$$
(7)
The moment parameter $`q`$ can be thought of as a filter, which identifies only the singularity characteristics of the distribution at a particular “degree” of clustering. Increasing values of $`q>0`$ emphasize the stronger local clustering nature of the pattern, while decreasing values of $`q<0`$ the less singular regions. That is, higher values of $`q`$ serve to “eliminate” the smaller values of $`p_i`$, yielding a subset of the overall distribution whose scaling behavior is more condensed (and vice versa for negative $`q`$). Likewise, the other $`D_q`$ provide a measure of the number of q-tuples whose mutual separation is contained within a covering box (ball) of size $`ϵ`$. Thus, a quantitative measure of the $`D_q`$ spectrum yields an understanding of all order of correlations amongst clusters of varying densities.
Certain key values of $`D_q`$ are extremely useful in characterizing the physical clustering characteristics of a set. In addition to the values $`D_{0,1,2}`$ mentioned before, the generalized dimensions for the limits $`q\pm \mathrm{}`$ yield valuable information about the maximal and minimal density regions of the set. $`D_{\mathrm{}}`$ is a measure of the scaling behavior for the densest clustering regions of the multifractal, while $`D_{\mathrm{}}`$ corresponds to the equivalent for the least dense or “rarefied” regions.
In essence, the multifractal measures give an indication of “how fractal is the fractal”. A measure of the difference between the box dimension and any successive $`D_q`$ value
$$\mathrm{\Delta }=D_qD_0.$$
(8)
provides an estimate of the “degree” of inhomogeneity of the associated probability distribution . In particular, it seems reasonable to evaluate $`\mathrm{\Delta }|_q\mathrm{}`$ as an overall gauge of the “depth” of inhomogeneity. Clearly, it follows that $`\mathrm{\Delta }=0`$ for single-scaling Euclidean normal or monofractal sets, so the larger the value of $`\mathrm{\Delta }`$, the greater the “multifractality”.
## 3 Fractal Expressionism
In the last 1990s, the application of fractal analysis to the study of abstract expressionist art began to gain momentum, the first of which was reported in . It was concluded that Jackson Pollock’s work did indeed present certain fractal characteristics. Coined “Fractal Expressionism”, the authors in question proposed that Pollock’s drip paintings were constructed by processes not unlike those which help to forge the myriad of similarly fractal natural phenomena. In fact, they further suggested that by painting with such “automatism”, Pollock succeeded in capturing the very essence of nature within his works.
In particular, it was noted that most of the paintings studied contain at least two distinct scaling behaviors at different levels, much the same as the debated transitions to homogeneity in galaxy clustering. The first of these occurred at scales on the order of 1 mm to about 5 cm, beyond which point a second scaling is observed up to scales of several meters . After a review of Pollock’s painting methods and techniques, it was determined that these two dimensions were the result of two distinct physical processes. The larger-scale patterns resulted from Pollock’s “Levy flights” across the canvas (a Levy Flight is a combination of discrete, random jumps coupled with local fractal Brownian motion ). Likewise, the small-scale structure was attributed to his infamous “drip” technique, which was largely dependent on the physical characteristics of the paint (viscosity, the height from which it was dripped, absorption into the canvas, etc…).
These two dimensions are coined $`D_L`$ (Levy) and $`D_D`$ (Drip), and in general it was found that $`D_L>D_D`$ (in fact, the authors of claim that $`D_L`$ tended to values close to 2, indicative of the “space-filling” behavior of Pollock’s Levy Flights). Furthermore, it was claimed that $`D_L`$ tended to increase from 1 to about 1.7 between the early 1940s to the late 40s / early 50s (around the time Pollock perfected his drip technique ).
Their analysis focused exclusively on the works of Jackson Pollock, and these dimensions are attribute to his own artistic style. In the spirit of the aforementioned conclusions of , however, the question should be asked as to whether or not such an analysis truly pinpoints anything “unique” about the artist in question, or whether the resulting statistics are shared by a common set of images and patterns formed by similar methods.
### 3.1 Image specifics
The majority of the images considered in this study are digital scans of Pollock’s works from the references . Images by Les Automatistes have been scanned from . The resolution of the scans was chosen as 300 dpi, creating images roughly 1000 pixels (px) in length (longest side) and files 20 Mb in size. The analysis has been performed on approximately 25 Jackson Pollock paintings, revealing similar trends for each. However, this discussion will be restricted to a small sample set of six. The images herein are listed in Table 1. At the specified resolution, each pixel corresponds to approximately 0.1-0.4 cm, although this will depend on the actual reduction scale from the base image.
The covering boxes range in size from $`d=`$1024 px to $`d=`$4 px, or length scales of roughly $`1.52.5`$m to a few millimeters. Hence, the analysis covers about 3 orders of magnitude. Higher resolutions could allow for greater range of scales, but would correspond to much larger images and longer run-times / higher memory requirements for the code. It was verified that the quality of the fits did not change appreciably for a lower limit of $`d=2`$, and the estimated dimensions were statistically equal to within the associated error.
### 3.2 Color Variance Filter Process
Accurate definitions of colors and color differences are very difficult to obtain. Any investigation which relies on color matching must do so with care. The following procedure is a rough example of how like colors might be extracted from an image, based on their Euclidean separation in the three dimensional RGB color space.
To trace or filter the pattern of a given pigment, the variation in shading is accounted for via the color-variance filter process. The images studied herein are 24-bit color maps, hence each separate channel may assume 256 possible values. An RGB triplet is chosen as the target color, each pixel (channel) intensity in the image is then compared to the initial triplets R<sub>0</sub>, G<sub>0</sub>, B<sub>0</sub> (hereafter RGB<sub>0</sub>), and the Euclidean distance (or color radius) is calculated,
$$R_{RGB}=\sqrt{(R_0R_{pix})^2+(G_0G_{pix})^2+(B_0B_{pix})^2},$$
(9)
Figure 1 shows the filtered pattern for $`\beta _{RGB}=20`$ for image P02. Patterns are isolated by including pixels for which $`R_{RGB}\beta `$, a cutoff whose value is determined by examination of the RGB histogram for the color in question. Figure 2 shows the R, G, and B pixel intensity histogram for the “black” pigment of image P04, which is generally of the same form for all images considered herein. The peaks of each correspond roughly to the values $`(\mathrm{R},\mathrm{G},\mathrm{B})=(21,17,21)`$, which is taken as the target color RGB<sub>0</sub>. Note the smooth drop-off for increasing (and decreasing) values of the pixel intensities.
For the paintings considered, it was found that most RGB histogram spreads tend to extend no more than 5-20 pixel intensities from the central peak. Hence, it seems reasonable to assume that the cutoff $`\beta `$ should be between $`\beta (10\sqrt{3}20\sqrt{3})(10,40)`$.
The pseudo-normal nature of the distribution in Figure 2 suggests that a Gaussian filter, which weights colors according to their distance from the “target”, would be more appropriate that the cut-off filter considered presently. This type of filter is inappropriate for calculation of the box Dimension, for which any box is counted in which there exists a point in the allowable range (i.e. this would result in a severe over-count of boxes). However, a “weighted” information dimension is certainly feasible, in which one assigns the color match a value of $`\mathrm{exp}(\mathrm{R}_{\mathrm{RGB}}^2/\mathrm{a}^2)`$, with $`a^2`$ the FWHM corresponding to the average histogram spread. This filter would be better exploited in the multifractal analysis of 7. However, preliminary calculations suggest that the results will not vary significantly from those of the cut-off method described herein. Since the very notion of color distance discrimination itself is somewhat of a fuzzy area (see e.g. or similar references), it is best not to “over-complicate” the procedure at this given stage of development. Thus, only the cut-off will be used in this study.
## 4 The Box and Information Dimension of Jackson Pollock’s Work
As previously mentioned, the information dimension can be considered a better statistic for the study of recursive patterns. That is, the box dimension can sometimes provide an overestimate of the scaling behavior, since it does not account for the relative density of points within the box. Although these specific results have been reported in , a slightly different analysis of the findings was given in that Reference. What follows will be a more technical discussion of the results.
Table 2 presents the corresponding dimension estimates for each painting. The Box Dimensions calculated by a least-squares regression on the data points seems to provide a good agreement to the results cited in , who found e.g. $`D_0=1.67`$ for P02 and 1.72 for P01. This suggests that a value of $`\beta [20,30]`$ would be in rough agreement with their analysis.
However, closer inspection of the results of Table 2 reveals that in certain cases the estimate of the dimension is critically dependent on a correct choice of $`\beta `$: the darker colors appear more stable, while the lighter ones show wider variation. In order for this analysis technique to be useful, these selection criteria must be extremely well defined. Otherwise, the results risk becoming meaningless. Ideally, some kind of variance in choice of $`\beta `$ should be incorporated into the overall error estimate. The issue of color space selection is discussed in Reference .
As mentioned in the previous section, the authors noted an apparent break in the slope of the log-log plot, and assumed that this represented different scaling behavior of two different mechanisms. The shallower slope was taken to be representative of Pollock’s painting technique. In fact, in reference , the author discusses the association of two distinct dimensions based on the topological morphology of the fractal (for higher length scales), as well as its texture (lower scales). These two dimensions are appropriately labeled as those of the structural fractal and textural fractal, respectively. Relating to the work of , it is not unreasonable to interpret their two dimensions accordingly, i.e. the overall “structure” of the painting at higher length scales, and the fine-grained refinement at lower scales.
Note that in the analysis presented herein, the two-slope hypothesis of is not supported when one considers the magnitudes of the associated errors in the least-squares procedure. The confidence level curves suggest that the “shallow” slope at lower length scales could be explained as statistical variation in the fit. The fits in Figure 3 show box and information sample plots for P02 with 95% confidence level curves from the least-squares fit. The information Dimension $`D_1`$ is shown as a “refinement” of $`D_0`$, which demonstrates even less bi-scale behavior, suggesting that the two-slope hypothesis may be an artifact of the box-counting method. The data provides a very clean linear fit in both cases, generally better for the information dimension $`D_1`$, albeit not significantly ($`r^2=99.9\%vs99.8\%`$). Similar behavior is observed for the other images.
In fact, the lower-scale measurement process is somewhat dependent on the resolution of the image. Some of the fits suggest a shallower slope at smaller scale lengths, but it is not necessarily justified to assume that this behavior is an artifact of the pattern, and not the resolution of the image, or limitations of the counting/analysis method.
Although the methods herein and those of reference differ interpretationally, roughly the same end result is obtained. That is, one can still associate an effective fractal dimension in the range $`D_01.61.8`$ with the patterns on the paintings, simply by considering the slope of the entire fit. A changing slope from box counting does not immediately imply multifractal behavior.
While it may be that the slope tends to be shallower at lower scales, this may not be an artifact of the data set so much as a manifestation of estimates and assumptions about the data. There may also be lower-level resolution limitations due to the finite size of the pixels. Surely, in the mapping from a 5 metre painting to a 30 cm page – or to a 1000 px binary image – there must be some significant level of information loss at the lower scales of resolution (in both the photograph and the data scanning process).
There does not seem to be significant variation in dimensions between lighter and darker colors, although in certain cases it is observed that the lighter pigment patterns tend to exhibit lower fractal dimensions. This could be due to a different deposition mechanism than simple dripping, as well. It is perhaps a sweeping generalization to assume that all the pigments were applied in exactly the same fashion.
So, it becomes somewhat unclear how one can define the “dimension” of the entire image. This suggests an application of the Fractal Union Theorem (see e.g. ). Since the fractal dimension of the union of fractals $`F_i`$ has dimension $`\mathrm{max}\{D_i\}`$, then the fractal dimension of the entire image will correspond to that of the most complex pattern. Thus, isolation of the pattern with the highest dimension can be interpreted to characterize the fractal nature of the entire image. This is consistent with the notion of the “anchor layer” discussed in references (i.e. the pattern which seems to strongly influence the dimension of the whole image). However, note that these authors mention that the overall dimension increases as more patterns are considered, driving the overall dimension to $`D2`$. It is unclear what is meant by this statement, but from a mathematical approach, it seems contradictory to the associated theorems.
## 5 Robustness of Analysis Method
The exact determination of the fractal dimensions depends on the cutoff for the colors under consideration. Thus, there is a certain amount of variability in the estimation. To test for further variability (and hence potential limitations of the box counting method applied to such images), P01, Reflections of the Big Dipper, and Number One 1949 were each rotated by 90, and the corresponding Box and information Dimensions were calculated for a color radius of $`R_{\mathrm{RGB}}=20`$ pixels:
* Blue Poles: $`D_0=1.68\pm 0.03`$; $`D_1=1.65\pm 0.02`$
* Reflections …: $`D_B=1.77\pm 0.04`$; $`D_I=1.72\pm 0.03`$
* Number One 1949: $`D_B=1.73\pm 0.05`$; $`D_I=1.70\pm 0.04`$
These are quite commensurate with the values obtained in Table 2, subject to the cited error, confirming the rotational invariance of the result.
Pixels are randomly displaced by 5, 10, and 20 positions from their original location, and the appropriate dimensions are again calculated for the same paintings. Table 3 shows the results for the same paintings as above.
## 6 Fractals in Gestural Expressionism
If the patterns which appear in these paintings truly are the product of physical processes, rather than pure artistic expressionism, than such structure should be visible in similar works by other artists. Based on the a similar analysis to that of Reference , it seems reasonable that other images formed by similar processes should be classifiable by similar statistics.
Roughly contemporaneous with Jackson Pollock, the Québec School Les Automatistes also produced non–representational art not unlike the drip paintings studied above. The group was spearheaded by Jean-Paul Riopelle and Marcel Barbeau who collectively produced their works over the 35 year period spanning 1945-1980.
Figure 4 shows a section of a drip painting from Les Automatistes, as well as the filtered black pigment pattern. Table 4 lists the calculated Box and information Dimensions for select works by Les Automatistes, subject to the same selection criteria as before.
As with Pollock’s drip works, the dimensions of the patterns fall roughly between $`1.61.8`$. The box and information dimensions do not explicitly differentiate between Pollock’s work and that of Les Automatistes. In fact, the difference in the average fractal dimensions for each artists was shown to be statistically insignificant using a two-way ANOVA in Reference . The lighter colors display mildly lower dimensions than the darker pigments, although this may be due to cutoff limitations of the filtering process. Similar behavior was observed in the images by Pollock, so whether or not this is an actual artifact of the pattern or a numerical effect is a subject for future investigations.
While this was somewhat the case with Pollock’s works, there are perhaps sufficient discrepancies to suggest that such a measure could be indicative of different uses of colors and techniques between these artists. This includes using lighter colors for balance in an image, versus their use for adding contrasting depth.
In any event, the general equivalence of the dimensions of Pollock’s works and those of Les Automatistes suggests that the utility of this technique as a “fingerprinting” mechanism for individual artwork/artist association may be in vain. As with the galaxy clustering models, one could assert that the technique can isolate only construction method, and not structural variation within the method. Those who are dissuaded by the effective reductionist implications of the analysis may find comfort herein. In order to further address this point, the multifractal analysis will be addressed in Section 7.
## 7 Multifractal Spectrum of Non-Representational Images
Figure 5 shows the range of generalized dimensions $`D_q`$ for these patterns. Note that the overall depth of the generalized dimension spectrum is not excessive, suggestive that if these patterns can be described by multifractal statistics, their overall structure is not that extensive. Furthermore, note that for the majority of the cases considered, there is no appreciable difference in the range or shape of the spectrum. The errors from the linear fits are generally of the order 0.05 or less, but these may be underestimates since no error is introduced for variation in the color. The limiting values of $`D_{\mathrm{}}`$ give less intuitive insight into the densest clustering regions, unlike in the case of the three-dimensional sets considered earlier.
Table 5 shows inhomogeneity measure for several Pollock and Automatistes works, defined by Equation (8). In general, the results suggest that Pollock’s works tend to be “deeper” than those by the Automatistes (i.e. greater degree of inhomogeneity), perhaps a result of painting styles and refinement techniques. This could hint at a potential method of distinction for the sets of similar classes, but one must be extremely cautious of the selection criteria for the pattern in question. It is more likely that these measurements are simply too “noisy” for any useful approximation.
## 8 Comparison of Construction Method: Gestural Expressionism versus Artonomy
The utility of the analysis methods contained herein seem limited in the context of analysis of differing works of the abstract expressionist class. For the cases considered, the variance in the data seem too small to be of any particular import for specific identification. However, when applied to other images, certain differences do arise, enabling one to make distinctions at least on some level.
In particular, the artwork of Tsion Avital will be considered. In his seminal work on the subject , Avital introduces the concept of Artonomy, the focal blend of artistic expressionism with scientific order. A complete description of the intricacies of the method will not be discussed here, and the interested reader is referred to the aforementioned citation for further details. The crucial point is that the construction “philosophy” for these images is strictly different than those of the gestural expressionist class considered previously.
Avital notes that the concept of Artonomy is based on certain principles of of “isotropy” in the creation process. There are no preferred sets of colors, and the use and applications of each color are deemed “equal” in value to every other. Colors (or elements) are combined into a variety of rigorously-defined mathematical sets (dubbed “moments”), and the final paintings are constructed from combinations of these moments subject to the appropriate rules. Paints are applied in a simple manner (e.g. controlled brush, or “toothbrush spray”) and as with the color selection, there is no preferred method.
The moments are methodically positioned on the canvas in a recursive fashion quite reminiscent of the basic structure of multifractals (such as, for example, the framework outlined in Figure 6). Of particular interest is Avital’s “type $`\gamma `$” moment construction rule , which operates on the basis of information density on the canvas. Here, he defines the density as low when like colors or hues are assembled (homogeneous elements), and high with the neighboring placement of contrasting elements (heterogeneous elements). Avital defines an abstract field as one which is comprised of low density regions, and a concrete field as one composed of high density regions. Abstract and concrete fields may be inter-mixed to form heterogeneous fields. Images AV01-03 represent “homogeneous” constructs, while AV04-06 are “heterogeneous”.
So, in a sense, comparison of gestural expressionist “structures” with those of Avital constitutes a contrast in construction methodologies – random versus algorithmic – and thus Avital’s works can be taken to be a control or model comparison.
Table 6 shows measured generalized multifractal dimensions for various color distributions in Avital’s works. It is somewhat difficult to define an exact base color in the homogeneous images (AV01-03), since the resulting pattern is due to integrated “aerosol” deposition. In any event, note that unlike the Pollock and Automatistes images, Avital’s works show no significant multiscaling behavior. In many cases, the calculated $`D_{\mathrm{}}`$ is higher than $`D_0`$, yielding a negative $`\mathrm{\Delta }_{0,\mathrm{}}`$. It should be noted that similar behavior was observed for some monofractals and simple geometric shapes (i.e. objects for which there is a single scaling dimension), where the $`D_q`$ for small $`q`$ tend to underestimate the actual dimension. Also, if one considers the associated statistical error, then these negative values are easily accounted for. Thus, it can be concluded that Avital’s systemic blobs are devoid of the “rich” structure with which the gestural expressionists endow their works, due perhaps in part to the very algorithmic (less random) nature of the construction.
Furthermore, Avital’s homogeneous works (e.g. AV01) were constructed from the spray of a paint from a toothbrush. Thus, the resulting structure is probably similar to the deposition from an aerosol source. The dimensionality most likely reflects this mechanism, to a certain extent. Avital’s heterogeneous works (AV11, AV12) were constructed with controlled paint brush strokes. So, these could actually be considered two separate sub-construction mechanisms.
## 9 Reconstructing Images From the Multifractal Spectra
Accurate determination of the multifractal spectra of singularities for a dynamical process can yield important information about its construction processes and associated constraints. As previously mentioned, the $`D_q`$ provide important information about the n-tuple “pair-wise” clustering behavior of the set, and provide a unique characterization of the object under investigation.
The quantities thus obtained can be used as physical constraints to be used in development of any model, and can perhaps yield interesting information about the dynamics of the pattern generator during the construction phase. Since the multifractal analysis herein seems only to have the ability to discern one class of structure from another, one must ask whether or not there is a useful tool to distinguish between like sets. A short analysis is performed herein on the like image arrays of the abstract expressionist class, in order to address this problem.
By definition, a multifractal is an inhomogeneous recursive scaling (a multifractal lattice). Suppose a square (or box, to be consistent with the current nomenclature) is divided into four sub-units of equal area. Then, one can describe the relative portion of the pattern contained in each box by the probabilities $`r_1,r_2,r_3`$, and $`r_4`$ respectively (see Figure 6).
At the next level of recursion, the weights $`r_i`$ are randomly reassigned to each sub-box of the previous layer, and the process repeats (cascades) down to any level of recursion desired.
Recall that the generalized dimensions are calculated from the partition function (5), and furthermore $`(q1)D_q=\tau (q)`$. From (7), one can estimate the difference between two successive cut scales $`\delta `$ and $`\delta /2`$ (c.f. Figure 6) as
$$D_q(q1)=\tau (q)=\frac{\mathrm{\Delta }\mathrm{log}[Z(q,\delta )]}{\mathrm{\Delta }\mathrm{log}[\delta ]},$$
(10)
In terms of the probability $`r_i`$ for each box, this becomes
$$\frac{\mathrm{\Delta }\mathrm{log}[Z(q,\delta )]}{\mathrm{\Delta }\mathrm{log}[\delta ]}=\frac{\mathrm{log}[_ir_i^q]\mathrm{log}[_jr_j^q_kr_k^q]}{\mathrm{log}[\delta ]\mathrm{log}[\delta /2]},$$
(11)
which reduces to
$$D_q(1q)=\frac{\mathrm{log}[Z(q,\delta )]}{\mathrm{log}(2)},$$
(12)
So, one can substitute $`Z(q,\delta )=r_1^q+r_2^q+r_3^q+r_4^q`$ to obtain
$$r_1^q+r_2^q+r_3^q+r_4^q=2^{D_q(1q)},$$
(13)
and the distribution probabilities $`r_i`$ may be obtained from a system of four equations. Note that this expression may be simplified, by noting the constraint $`r_1+r_2+r_3+r_4=1`$. Furthermore, the $`q=2`$ version of Equation 13 represents the equation of a 4-sphere, whose roots may be easily obtained.
Hence, the system of four equations may be reduced to a system of two unknowns, in this case $`r_3`$ and $`r_4`$. The values of the four possible $`r_i`$ may be isolated by optimizing the possible values of $`r_3,r_4`$ which fit the measured $`D_q`$ spectrum of generalized dimensions. This is achieved by finding sets of $`r_i`$ for which the individual separations $`\mathrm{\Delta }D_i=D_{\mathrm{calculated}}D_{\mathrm{measured}}<ϵ`$, for $`ϵ0.0001`$ on average.
In Table 7, the $`r_i`$ values for various shapes of known monofractal dimension are presented. Note that for a figure of topological dimension $`D_T=1`$ (e.g. the line), the weighting factors suggest that for the appropriate cut of the plane in Figure 6, the shape will only have a nonzero probability of being in any two of the 4 sub-boxes, a result which certainly makes sense. Similarly, a figure of dimension $`D_T=2`$, which “fills the plane”, will have equal probability of being in every box. The negative component for the Koch Curve (Island) is most likely the result of numerical uncertainty, since negative probabilities would not make sense. The calculated $`r_i`$s for the Koch Curve and Sierpinski Gasket can also be interpreted to reflect the construction algorithms and symmetries for each figure.
Table 8 lists the calculated values of $`r_i`$ for the “anchor layer” pigment shapes in several of the images considered previously. The values are relatively consistent for each painting, although this is not a particularly surprising result, since the $`D_q`$ spectra themselves are not significantly different.
Each set is characterized by a rather even distribution amongst three of the boxes, and a fourth which is smaller by an order of magnitude. The almost homogeneous distribution is no doubt reflective of the fact that the generalized dimensions are close to 2. It can be shown that for a distribution with $`r_i=0.25`$ for all $`i`$, the generalized dimensions all collapse to 2 (or vice versa).
It may be somewhat discouraging to note that these values are rather close to one another, and are seemingly indistinguishable. However, it should be noted that the formalism outlined above is not a singular representation of a multifractal scaling process. Four quadrants have been used to show recursive scaling in part for computational efficiency. This could be a significant source of error if the scaling behavior is radically different than this model requires. Additionally, this may again be a fundamental problem with the resolution limitations of the method.
The parameters herein can conceivably be used in the formulation of a physical model which could reproduce the associated images, at least on a statistical level. Furthermore, the authors of have studied video recordings of Jackson Pollock in his creative process, and have found that the “fractality” of the overall work took less than a minute to define. Surely, this provides an additional constraint on a such a cascading model.
On a subjective level, one wonders whether or not the smaller fourth quadrant could conceivably be interpreted as an imprint of the presence of the “source” of the image pattern (in this case, the physical presence of the artist). That is, at any point during the construction of the painting, the artist has free choice to paint in three of the four “quadrants” (the last being occupied by himself). Thus, this could be nested in the recursion, and detectable by such an analysis. If this explanation were to accurately represent the evolution of the pattern, it could be used to distinguish between patterns constructed by humans, and those created by machines or other natural processes.
## 10 Visual Multifractals
The analysis in the previous Sections relied predominantly on a color filtering process dependent on the distance in RGB space of pixel color to its target “match”. However, many reports suggest that the hierarchical clustering of the images has some variety of psychological effect on the viewer. While using RGB primaries as the filtering criteria isolated the physical structure of the blob, it may not be an effective measure of the perceived structure. Taylor et al. have recently studied physiological responses to fractal viewing, and have concluded that observers do exhibit definite responses when presented with certain fractal patterns .
The problem of structure identification and discrimination is not a new one in psychological circles, nor is it by any means a solved one. Implicitly related to this topic, the authors of reference discuss the perceptability of hierarchical structures in abstract or non-representational constructs (whose subject matter is used in a comparative study in Section 8). In fact, rapid object recognition and categorization via boundary isolation versus “blob” identification is a subject of growing scientific interest (see and related references therein).
A complete understanding of the nature of color perception is still lacking. Thus, the notion of a visual fractal is introduced in contrast to those fractals previously considered. Instead of direct observation of colors, the focus is instead shifted to edge structures. This is effectively an analysis of luminance gradients within the image, and not directly on the RGB color field distribution (although the luminance values are determined by R, G, and B mixes). In fact, after completing this research, the work of references was discovered. Therein, the authors discuss the potential uses of measuring the multifractal spectrum of luminance gradients in natural color images, to determine whether or not it conveys relevant information about the image. The analysis presented herein is quite similar in these respects, and thus is not performed without physical justification.
### 10.1 Luminance Edges as Visual Fractals
While ripe with theory, the actual dynamics of human color visual processing are poorly understood, yet it is clear that one does not require a wealth of color information to visualize a scene. A subject of ongoing interest (see e.g. and similar references) is whether or not object/pattern recognition occurs on the level of “blob” or “edge” identification. Studies of eye moments in subjects viewing artistic scenes seem to support the notion that human fixate on particular aspects of an image, supporting the notion that “blobs” are viewed but it is perhaps unclear as to how these objects are distinguished.
Similarly, the images formed by one’s brain may not be fully representative of the scene which one views. Both chromatic and achromatic information received from stimulation of the photopigment receptors in the rods and cones, are “preprocessed” before being sent to the visual cortex via the optic nerve.
In a similar vein, it is useful to find a “one-parameter” method of analysis for such color images, as an attempt to find a suitable way to discriminate between them. The results of the previous Sections suggest that different choice of colors yield somewhat differing dimensions, so it would be helpful to find an element common to all images which is independent of any particular color. Thus, one can consider analyzing luminance properties of the image.
Edge detection in the visual system occurs on several different levels, although it is not necessarily know which one is “dominant”. One such mechanism is known as lateral inhibition (LI). In short, this process measures the relative excitatory signal output from one photoreceptor with inhibitory signals from adjacent neighbors, effectively producing a difference output signal which is sent to the visual cortex. The result is that the strongest excitatory signals will be sent from those retinal neurons which detect luminance changes across the field .
Coupled to the visual system’s ability to interpolate information in a field from missing stimuli (e.g. as with the blind spot), LI can create artificial luminance and brightness variation effects which are not physically present in the original scene . For example, a black and white checkerboard will seem to have greyscale variations across the pattern. The intensity (luminance) of the central squares is the same in each case, but the square surrounded by white appears to be darker than the other (see Figure 7). This exemplifies the eye’s ability to create artificial variations in scenes which are otherwise not physically present. Hence, this provides a rather simple example of how visual interpretation of an object may not be complete commensurate with the actual physical characteristics.
Lateral inhibition is, however, only one of several mechanisms responsible for the detection of contrast edges in a visual field. While LI mechanisms operate in the eye, such detection is known to occur in the visual cortex itself. Hubel and Wiesel were responsible for the discovery of “orientation columns” within the visual cortex, cells responsible for the identification of specific edges or boundaries orientations. The aforementioned researchers share the 1981 Nobel Prize in Medicine for their research efforts. The interested reader is directed to reference for an expository account of their work. Thus, there is sufficient physiological and psychological motivation to consider possible structural differences in contrast edges.
The transformation from RGB primaries is of the form $`Y=0.299R+0.587G+0.114B`$ (note that the color coordinates must be normalized), which implies pure white coordinates $`(R,G,B)=(1.0,1.0,1.0)`$. Note the relatively higher weighting of R and G primaries to B. This is reflective of the eye’s sensitivity to similar wavelength intensities. In fact, these roughly correspond to the three basic types of cone cells with similar thresholds, denoted as L, M, and S (for long, medium, and short wavelengths). This is actually one component of a separate CIE color system known as YIQ (the channels I and Q are encode chromacy information, hue and saturation). The luminance channel is what one generally associates with greyscale images, an in fact is that information which is transmitted in black and white television signals . Edge detection is performed by generally-available image manipulation tools, which measure the vector sum of two perpendicular Sobel gradient operators (see e.g. for more information).
These are perhaps crude approximations to the actual physiological processes at hand, so implicit limitations in the estimates should be accordingly recognized. Certainly, the method does not purport to be a realistic model of the visual system. It should, however, provide a decent first-pass approximation to any inherent structures and effects therein.
### 10.2 Pollock vs Les Automatistes
Figure 8 shows a sample edge-detect transform for a images of Pollock and Les Automatistes, with the associated $`D_q`$ spectra in Figure 9. For the color-filter process, the target color in this case is pure white, and the color radius is taken to be the linear distance away from the point. Thus, for a small radius, the images with the highest gradients will have the largest dimensions. Table 9 and 10 give dimensions for both $`\beta =1`$ and $`\beta =30`$, which give an indication of the “value” of the strongest gradients. It should be noted that since neuronal firings are triggered by threshold-breaking stimuli, a discrete cutoff is more realistic than a Gaussian drop-off.
The calculated box dimensions for the edge-detected Pollock images tend to be higher than for the individual blobs, generally $`D_0>1.8`$. This can be interpreted as implying that the luminance edges form a much more complex visual field, and that the edge lines are more “space filling”, and providing a “busier” or “fuller” visual experience.
When the method is applied to the gestural expressionist works of Les Automatistes, differences become more apparent (see Figure 9). The box dimensions for Les Automatistes is generally lower (albeit not much) than those of Pollock’s. Similar results are obtained for other images (see Table 9). Note that the measured dimensions do not increase significantly from $`\beta =1`$ to $`\beta =30`$.
This suggests that there is a potential visual difference between images by these different artists. While the final products may resemble each other at first glance, the intricacies of the two images from a luminance gradient / visual standpoint appear quite different. Again, in Reference the difference in average fractal dimension of the “edge” patterns was determined to be statistically significant.
Based on the work of previous authors and their own survey on preferential response to drip patterns, the authors of conclude that patterns possessing a fractal dimension of roughly 1.8 are inherently aesthetically-pleasing to the observer. A follow-up study suggests that “creative individuals” have a preference for high values of $`D`$ . One could imagine that this type of perceived structural difference could contribute to an observer’s “appreciation” of one image or style over another.
### 10.3 Comparison With Avital
Avital’s definitions of homogeneous and heterogeneous fields (not to be confused with homogeneous fractal distributions), along with the concept of information content, are a natural extension of the notions of luminance gradient structures proposed in Section 10.1. In fact, the very notion of information content is at the heart of the multifractal formalism. Thus, a luminance-gradient analysis of Avital’s images should reveal certain properties about the formulaic construction of the pieces, or at the very least lend contrast to the more psychological algorithms used by the Abstract Expressionist artists (or perhaps any other artist).
Table 10 shows the effective edge-detection dimensions of various works by Avital , as well as a rough definition of the type of image. Figure 9 shows the associated spectra, in comparison to the previous images. Since reproduction of every image in this work is not warranted, The previous color panels demonstrate the general qualities of each type of image (labeled homogeneous and heterogeneous), while Figure 10 shows the resulting luminance gradients. It should be noted that the images classified as “homogeneous” all conform to Avital’s “S/D/$`\delta `$” construction algorithm (made from the spray of a toothbrush) , which indeed embodies an inherently smooth transition to complexity vis-a-vis color selection and application. In contrast, the images noted as “heterogeneous” are from Avital’s “S/C/$`\gamma `$” algorithm, which allows for a counterbalance between abstractness and concreteness. These are simply painted with controlled brush strokes.
The lower dimensions and higher error for the homogeneous images in Table 10 demonstrate the low color contrast nature of the images, and hence the shallow depth of luminance variation across the canvas. In particular, note that while image AV01 is physically a mix of bright pigments, there is virtually no strong luminance gradient across the canvas (hence to effective dimension of 0). These dimensions imply there is little “luminance information” in the fields. As the images begin to approach heterogeneity, the background field is contrasted with patches of color, whose overall boundaries are quite regular. This is indicative by the relatively low range of $`D_0=1.11.3`$, in contrast to the exceedingly high dimensionality of the gestural expressionist paintings, as seen in Table 9. The latter is indicative that the luminance gradients densely fill the canvas for this particular school or movement, while Avital’s shapes are more concentrated and simple.
Thus, Avital’s homogeneous work is less “interesting” from an edge detect view than the Pollock or Les Automatistes images considered (there is less edge information conveyed about the scene), while the heterogeneous work brings focus to particular objects via these edges (although still with much lower dimension that the gestural expressionist images).
These results suggest that this analysis method could distinguish between sources, as well as construction mechanisms of the images. Further investigation would be required. However, whether or not this type of distinction is possible, identification of such structural signatures could have applications in external fields. For example, albeit beyond the scope of physics, visual detection of fractal structure in luminance gradients could have profound consequences for the fields of aesthetics and visual appreciation of complex scenes (e.g. what qualities makes an image interesting to us?).
It is interesting to note that Avital himself classified works such as Pollock’s as “moment type $`\omega `$”, whose paradigm rests on the notion of “arbitrariness”, in which the combinations of elements (moments) are scattered at personal will about the canvas (he further notes these to require “minimal capacity of inventiveness”, and casts Pollock’s art as containing nothing meaningful or interesting ). Avital is careful to note the distinction between “arbitrariness” (whose choice of elements is human) and “randomness” (whose source is instead probabilistic). Unfortunately, Avital presents no simulations of type-$`\omega `$, so it is not possible to compare these with the works of the gestural expressionists considered previously.
## 11 Potential Limitations of the Method
Of course, the method described herein is not without limitations, and is only designed to be a “first-order” attack of the issues at hand. As discussed previously, digital image analysis techniques provide a statistical description of the entire physical image, with no regard for perceptual interpretations by observers. The calculated dimensions assumed equal weighting for all portions of the canvas, when in fact (depending on the distance from which they view the scene) observers will not register all portions of the field equally. Both rod and cone cells are unevenly distributed about the retina, with a disproportionately large number of cones clustered in the fovea centralis . This cone clustering is crucial for perception of color and fine visual detail via fixation, and is the primary reason for the drop in acuity in peripheral vision.
So, if the image of interest fills the visual field, only the central-most regions will convey the largest amount of information. However, this should not necessarily affect the overall “visual estimation” of the fractal nature of the piece, although edges may become more blurred (resulting in potential shifts in a “visual multifractal spectrum”). Furthermore, the method does not account for other biasing effects such as color blindness, or any visual acuity drops (e.g. myopia or other focal abnormalities). The robustness tests of Section 5 suggest that the dimensionality of patterns will increase for dispersive patterns (as they should, approaching homogeneity), which could replicate such vision problems.
Finally, the methods outlined herein do not all complete correspond to actual physiological processes which occur in the eye. Reference provides several alternative color space transformations which are perhaps more appropriate for the actual analysis of cone/photoreceptor excitations from lightness/luminance and chromatic stimuli. A full investigation and implementation of these methods is discussed in Reference .
## 12 Concluding Remarks and Future Directions
The use of fractal and multifractal analysis as a discriminator or fingerprint method for classifying abstract expressionist art is a budding field. However, the available results are indicative that the method may well yield promising results. The fractal signatures obtained from paint blobs are not significantly different from one another, implying that this method is not useful for “authenticating” works by any one particular artist within a movement. It apparently does differentiate between the movements themselves. This is similar to the behavior observed in , where the multifractal spectrum was shown to differentiate between galaxy cluster formation mechanisms, but could not discriminate between instances within the same model. The “edge multifractal” does yield differentiable results, curiously, which based on aspects of visual processing lends to the interpretation that this could represent some type of “aesthetic preference”.
One of the motivational questions which inspired the fractal analysis of gestural expressionist art is: “Does there exist an inherent structure within the painted patterns which one perceives, and hence yields an unconscious psychological effect on the observer?”. Rephrased, on can pose the question: does the brain possess a mechanism whereby the observer can gain information from a scene previously unknown to them?
This question certainly addresses the very heart of recognition and learning methodologies, but unfortunately the exact neural mechanisms which lead to cognition are not well understood. Recent discoveries in Neuroscience have paved the way for a potential revolution in this field, however.
Recent studies have revealed striking neural activity in several species of primates which respond not only to physical imitation of observed movements by others, but also passive observation of such actions. That is, such neural firings are indicative that the individual need not repeat the action in order to cognitively process its meaning – quite literally, a case of “monkey-see, monkey-do”. Based on these imitation characteristics, such cells have been dubbed mirror neurons. For a basic introduction, see and references therein.
Additional studies suggest that mirror neurons may be present in higher species of primates (and in particular may be central to the development of language skills in humans ). If observation of action can trigger their firings and initiate comprehension of its meaning, then it may not be unreasonable to expect that observation of the trace of an action can also prompt similar neurophysiological responses.
The authors of note that many natural patterns possess multifractal scaling behavior, but these are not “art” per se. What is the underlying differentiator, then, that ascribes to these statistically-similar patterns the label of “art”? Thus, by observing a complex but statistically-ordered scene such as Pollock’s art, mirror neurons could help to bridge the gap between the initial visual processing and associative comprehension and appreciation of the actions required to form the work . Further study into these hypotheses are currently underway.
Acknowledgments
We gratefully thank Tsion Avital for permission to reproduce his art. This work made possible by grants from the Natural Sciences and Engineering Council of Canada (NSERC) and by financial support from the Walter C. Sumner foundation.
|
warning/0506/quant-ph0506248.html
|
ar5iv
|
text
|
# Quantum-classical correspondence in the oracle model of computation
## Abstract
The oracle model of computation is believed to allow a rigorous proof of quantum over classical computational superiority. Since quantum and classical oracles are essentially different, a correspondence principle is commonly implicitly used as a platform for comparison of oracle complexity. Here, we question the grounds on which this correspondence is based. Obviously, results on quantum speed-up depend on the chosen correspondence. So, we introduce the notion of genuine quantum speed-up which can serve as a tool for reliable comparison of quantum vs classical complexity, independently of the chosen correspondence principle.
Quantum mechanics offers a new, promising perspective for computer science. Quantum computers are believed to hold a computational advantage over classical ones. One of the most spectacular examples of this quantum speed-up is Shor’s famous algorithm shor for factorisation of large numbers. The executing time of Shor’s algorithm scales polynomially with the size of the problem, whereas the best known classical algorithm - General Number Field Sieve scales subexponentially. Of course the notion of quantum speed-up is not absolute unless it is judged by comparing optimal quantum and classical algorithms. However, finding lower bounds for NP problems is not easy in general. Thus, in order to prove the advantage of quantum computers in a rigorous way a special model of computation, namely the oracle model of computation (OMC) was introduced. In the OMC, algorithmic complexity is identified with query complexity, i.e. the number of oracle calls required for solving a problem. Within this model, quantum and classical bounds for many problems have been obtained. The cases when quantum complexity is lower than classical complexity are usually claimed to be rigorous proofs of quantum speed-up. Obviously, algorithms which are to be compared within the OMC should call the same oracle. Quantum and classical oracles are essentially different. Thus, strictly speaking, reliable comparison of quantum and classical algorithms is not possible. To overcome this problem the notion of correspondence between quantum and classical oracles is commonly used.
In this Letter, we show that this correspondence can not in general be unique. As a consequence, we propose a modified procedure for reliable comparison of quantum and classical algorithms within the OMC. Within this framework it turns out that quantum speed-up offered by some algorithms is just an artefact of the ambiguity of the previously used correspondence. Our arguments also shed some light on the role of entanglement in quantum speed-up.
Let us consider the question of “quantizing” a given classical operation. As an example to clear notions, suppose the classical operation is the (one bit) NOT gate which converts a bit ($`a`$) into its compliment ($`1a`$), ($`a=0,1`$). It seems natural to choose as a quantum counterpart of this gate the $`\sigma _x`$ Pauli operator
$$\sigma _x=\left(\begin{array}{cc}0\hfill & \hfill 1\\ 1\hfill & \hfill 0\end{array}\right).$$
(3)
Indeed, the transformation invoked by this matrix on choosing the quantum bit in the form $`\chi _a=|aa|`$, where $`|a=\left(\begin{array}{c}1a\\ a\end{array}\right)`$ is exactly
$$\chi _a\stackrel{\sigma _x}{}\chi _{1a}.$$
(4)
On the other hand, the $`\sigma _z`$ Pauli operation
$$\sigma _z=\left(\begin{array}{cc}1\hfill & \hfill 0\\ 0\hfill & \hfill 1\end{array}\right)$$
(7)
also implements the NOT gate
$$\eta _a\stackrel{\sigma _z}{}\eta _{1a}$$
(8)
provided that the computational basis states are chosen in a different way, namely $`\eta _a=|a^{\prime \prime }a|`$, where $`|a^{}=\frac{1}{\sqrt{2}}\left(\begin{array}{c}1\\ (1)^a\end{array}\right)`$. This simple example illustrates the dependence of the quantum-classical correspondence on the choice of computational basis states. Furthermore, even after choosing a given computational basis there remains a different ambiguity described below. Suppose we choose the bit $`\chi _a`$, then the operation $`\stackrel{~}{\sigma }_x`$
$$\stackrel{~}{\sigma }_x=\left(\begin{array}{cc}0\hfill & \hfill e^{i\theta }\\ e^{i\varphi }\hfill & \hfill 0\end{array}\right)$$
(11)
implements the bit compliment
$$\chi _a\stackrel{\stackrel{~}{\sigma }_x}{}\chi _{1a}$$
(12)
for any $`\theta ,\varphi `$. Let us emphasize that the common convention of choosing $`\theta =\varphi =0`$ is arbitrary and can not be justified simply by correspondence. This is because physical states are represented by rays not vectors in Hilbert space.
The argument given above is bidirectional, i.e. to a given quantum operation there may correspond different classical operations. For example, the $`\sigma _z`$ operation corresponds to the classical identity
$$\chi _a\stackrel{\sigma _z}{}\chi _a$$
(13)
as well as NOT operation
$$\eta _a\stackrel{\sigma _z}{}\eta _{1a}.$$
(14)
In general, we say that a classical reversible operation $`O`$ transforming m-bits into m-bits according to the rule
$$\stackrel{}{a}\stackrel{𝑂}{}\stackrel{}{b},$$
(15)
where $`\stackrel{}{z}=(z_1,z_2,\mathrm{},z_m),\stackrel{}{z}=\stackrel{}{a},\stackrel{}{b}`$, corresponds to a quantum operation $`U`$ transforming m-qubits into m-qubits iff
$$\rho _\stackrel{}{a}\stackrel{𝑈}{}\rho _\stackrel{}{b}$$
(16)
for $`\rho _\stackrel{}{z}=\rho _{z_1}^{(1)}\rho _{z_2}^{(2)}\mathrm{}\rho _{z_m}^{(m)}`$. This correspondence is based on the formal identification of the action of the classical and quantum operations on their respective computational states. Note that each of the single qubit quantum computational states may be chosen arbitrarily and may be different for each qubit as long as they are pure and satisfy the orthogonality condition $`\rho _{z_j}^{(j)}.\rho _{1z_j}^{(j)}=0`$.
Let us now consider the so-called standard oracle which is a $`n+1`$-qubit unitary operation defined below
$$|\stackrel{}{x}|y\stackrel{U_S^f}{}|\stackrel{}{x}|yf(\stackrel{}{x}),$$
(17)
where $`|\stackrel{}{x}=_j|x_j^{(j)}`$ denotes an $`n`$-bit quantum register ($`x_j,y=0,1`$) and $`f:\{0,1\}^n\{0,1\}`$. On choosing the computational states of the form $`\rho _{x_j}^{(j)}=\chi _{x_j},\rho _y=\chi _y`$, the action of the oracle is
$$\rho _\stackrel{}{x}\rho _y\stackrel{U_S^f}{}\rho _\stackrel{}{x}\rho _{yf(\stackrel{}{x})},$$
(18)
where $`\rho _\stackrel{}{x}=_{j=1}^n\rho _{x_j}`$. Thus, a natural corresponding classical counterpart $`O_S^f`$ of this oracle transforms $`n+1`$ bits as follows:
$$(\stackrel{}{x}),(y)\stackrel{O_S^f}{}(\stackrel{}{x}),(yf(\stackrel{}{x})),$$
(19)
where $`\stackrel{}{x}=(x_1,x_2,\mathrm{},x_n)`$.
As explained earlier, this classical oracle need not be a unique counterpart of $`U_S^f`$. Indeed, by choosing a different set of computational states of the form $`\rho _{x_1}^{(1)}=\eta _{x_1}`$, $`\rho _{x_j}^{(j)}=\chi _{x_j}`$ ($`j1`$), $`\rho _y=\eta _y`$ the quantum oracle implements the transformation
$$\rho _\stackrel{}{x}\rho _y\stackrel{U_S^f}{}\rho _{\stackrel{}{x}\stackrel{}{c}}\rho _y,$$
(20)
where $`\stackrel{}{c}=(c,0,\mathrm{},0)`$ and $`c=f(0,x_2,\mathrm{},x_n)f(1,x_2,\mathrm{},x_n)`$. The classical counterpart corresponding to this transformation of computational states is now $`O_A^f`$
$$(\stackrel{}{x}),(y)\stackrel{O_A^f}{}(\stackrel{}{x}\stackrel{}{c}),(y).$$
(21)
So, there are at least two mainfestly different classical oracles corresponding to the standard quantum oracle. It is interesting to see what effect this ambiguity in the quantum classical correspondence has on the quantum speed-up of oracle problems. Let us focus on the well known PARITY problem parity which generalizes the original Deutsch problem deutsch . In the oracle setting, this problem requires deciding whether $`_\stackrel{}{x}f(\stackrel{}{x})`$ is even or odd. The optimal classical algorithm to the standard classical oracle $`O_S^f`$ requires $`N=2^n`$ queries to solve this problem, whereas it suffices to query the quantum oracle $`U_S^f`$ just $`N/2`$ times. Hence, the quantum speed-up exhibited by $`U_S^f`$ as compared to $`O_S^f`$ is simply by a constant factor. On the other hand, the classical oracle $`O_A^f`$ requires exactly the same number of queries $`N/2`$ as the quantum oracle. Thus, there is no quantum speed-up at all when the oracles $`U_S^f`$ and $`O_A^f`$ are compared.
As another example, consider the slightly modified Bernstein-Vazirani problem. The oracle function $`f:\{0,1\}^n\{0,1\}`$ is promised to be of the form $`f(\stackrel{}{x})=k_0+\stackrel{}{k}\stackrel{}{x}`$, where $`\stackrel{}{k}=(k_1,k_2,\mathrm{},k_n)`$ is the $`n`$-bit string to be identified. Notice that both the Deutsch problem and the original Bernstein-Vazirani (BV) problem bv are special cases of the stated problem when $`n=1`$ and $`k_0=0`$ respectively. Comparison of the standard classical oracle $`O_S^f`$ which optimally requires $`n+1`$ calls with the quantum oracle $`U_S^f`$ which requires only a single call yields linear quantum speed-up. Now, let us choose a different set of computational states for the quantum oracle of the form $`\rho _{x_j}^{(j)}=\eta _{x_j}`$ (for all $`j`$) and $`\rho _y=\eta _y`$. The action of the quantum oracle on these states is
$$\rho _\stackrel{}{x}\rho _y\stackrel{U_S^f}{}\rho _{\stackrel{}{x}+\stackrel{}{k}}\rho _y.$$
(22)
Hence, the classical oracle corresponding to this transformation is $`O_B^f`$
$$(\stackrel{}{x}),(y)\stackrel{O_B^f}{}(\stackrel{}{x}\stackrel{}{k}),(y).$$
(23)
Obviously a single call to the classical oracle $`O_B^f`$ suffices to solve the promise problem. Again, we conclude that there exists a classical oracle corresonding to the quantum oracle which is just as efficient.
Let us now comment on the interpretation of the above simple examples. In the usual scenario, one starts with the standard classical oracle which is then replaced by its quantum counterpart. Our results do not question the fact that the quantum oracle may provide a more efficient solution to a formulated oracular problem than the standard classical oracle. However, it is the algorithms not that the oracles that should be compared. As mentioned in the introduction, since quantum and classical oracles are completely different, strict comparison of the algorithms that call these oracles is meaningless. If indeed this comparison is made, then the source of the advantage of the better “quantum” solution could be hidden within the quantum oracle itself, although it may seem to be manifested in the quantum algorithm. Indeed providing the quantum oracle may be equivalent to providing different (non-standard) classical oracles.
To clarify further the notion of quantum speed-up in the OMC, suppose Alice and Bob (who is constrained to use only classical operations on logical bits) compete with each other to get a quicker solution to a given oracular problem. Suppose both Alice and Bob are given the same classical device (oracle). In this case, Alice cannot use quantum mechanical operations to her advantage since one cannot construct a quantum oracle given a closed classical black-box. Now suppose both Alice and Bob are given the same quantum oracle. Quantum speed-up occurs when Alice can provide a more efficient solution than Bob (this is indeed the case, e.g. in Grover’s search algorithm). In the BV problem however, Alice will manage a quicker solution only if Bob is additionally forced to use a particular (inefficient) encoding of logical states. Suppose the quantum oracle is implemented by an optical system closed in a black box whose input and output ports consist of optical fibres. Assume the logical bits to be encoded in the polarization of light. The classical nature of Bob’s state implies that he can use only two orthogonal polarization states, e.g. vertical and horizontal. Notice that the number of steps Bob needs to solve the problem ($`n`$ or $`1`$) depends just on the orientation of the device ($`0^o`$ or $`45^o`$ respectively).
Thus, we pose the question whether the speed-up in oracle problems is genuine quantum speed-up or just the result of the interplay between two classical oracles. Answering this question requires a refined procedure of comparing quantum and classical oracles. Here, we postulate the detection of genuine quantum speed-up by comparing the quantum oracle to its best possible corresponding classical counterpart.
Our considerations also resolve the apparent puzzle of “infinite” quantum speed-up in BV algorithms. From Eqs.(22) and (23), notice that the query bit is not transformed at all. Therefore, especially in experimental realizations exp , the query bit is completely excluded and the BV circuit is implemented as a controlled-$`f`$ phase shift oracle,
$$|\stackrel{}{x}\stackrel{U_f}{}(1)^{\stackrel{}{x}\stackrel{}{k}}|\stackrel{}{x}$$
(24)
The standard classical counterpart of this oracle does not allow the extraction of any information about $`\stackrel{}{k}`$, since it is simply the Identity oracle:
$$(\stackrel{}{x})\stackrel{O_S^f}{}(\stackrel{}{x}).$$
(25)
Since the quantum oracle recovers the value of $`\stackrel{}{k}`$ in a single query, it would seem that there is “infinite” quantum speed-up for this oracle setting. However, notice that there exists a different classical counterpart
$$(\stackrel{}{x})\stackrel{O_{\stackrel{~}{B}}^f}{}(\stackrel{}{x}\stackrel{}{k})$$
(26)
which also recovers $`\stackrel{}{k}`$ in a single call and thus resolves the puzzle.
Below, we sketch the general problem of finding all possible classical counterparts of an arbitrary unitary and solve it for the simplest case of 2-qubit unitaries. A general reversible classical oracle acting on $`m`$-bits is a permutation $`O`$ of all $`2^m`$ possible input strings. On the other hand, a quantum oracle is of course a general $`m`$-qubit unitary operation $`U`$. When does a classical oracle $`O`$ correspond to a quantum oracle $`U`$? As discussed in the introductory part of this Letter (see Eq.(11)), for characterizing classical counterparts, one must consider generalized permutation unitaries $`P`$ whose non-zero entries are unit modulus complex numbers. We say that a unitary matrix $`U`$ has a classical counterpart $`O`$ (in accordance with Eqs.(15) and (16)) iff $`U`$ is locally equivalent to $`P=DO`$, i.e.
$$(_iL_i^{(1)})U(_iL_i^{(2)})=P,$$
(27)
where all $`L_i^{(1)},L_i^{(2)}`$ are single qubit operations and $`D`$ is a diagonal unitary matrix. The problem of finding all possible classical counterparts is a particular subset of the general problem of local equivalence of unitary operations. Unfortunately, no general solution to this problem has been obtained so far.
In the simplest case of $`2`$-qubit unitaries $`U`$, three real parameters completely characterize local equivalence makhlin ; cirac ; zhang . A computationally appealing choice of these parameters are given by Makhlin makhlin
$$\alpha =\text{Re }\frac{\text{Tr }^2V}{16\text{det }U},$$
(28)
$$\beta =\text{Im }\frac{\text{Tr }^2V}{16\text{det }U},$$
(29)
$$\gamma =\frac{\text{Tr }^2V(\text{Tr}V)^2}{4\text{det }U},$$
(30)
where $`V=W^\text{T}W`$, $`W=Q^{}UQ`$ and
$$Q=\frac{1}{\sqrt{2}}\left(\begin{array}{cccc}1& 0& 0& i\\ 0& i& 1& 0\\ 0& i& 1& 0\\ 1& 0& 0& i\end{array}\right).$$
(35)
In particular, these parameters uniquely classify equivalence classes of all 2-qubit generalized permutations and thus the the set CC$`(U)`$ of all classical counterparts of a given unitary $`U`$. For compactness, it is convinient to first divide the the group of permutations $`S_4`$ into $`6`$ cosets with respect to the subgroup of local permutations $`S_2S_2=\{\sigma _x^j\sigma _x^k|j,k=0,1\}`$. These cosets may then be identified with their respective representatives chosen as follows: $`\mathrm{I}`$, $`\mathrm{SWAP}`$, $`\mathrm{CNOT}_{12}`$, $`\mathrm{CNOT}_{21}`$, $`\mathrm{SWAT}_{12}`$, $`\mathrm{SWAT}_{21}`$, where $`\mathrm{SWAT}\mathrm{SWAP}\mathrm{CNOT}`$. The classes of CC$`(U)`$ are identified in Table 1. There are four non-trivial classes and one empty class.
Finally, let us turn to the important question of the source of quantum speed-up. Although quantum entanglement is believed to be the key to quantum speed-up, there is no proof that this is indeed the case. For example, BV problem is a commonly mentioned case where quantum speed-up seems to be obtained without entanglement meyer . The PARITY problem solution also doesnot use any entanglement. We believe, that our notion of genuine quantum speed-up may help clarify the role of entanglement as a necessary constituent of quantum over classical algorithmic superiority. In the examples mentioned above we have been able to show that there is actually no genuine quantum speed-up where entanglement is absent. Moreover, in examples such as Grover’s problem and the Deutsch-Jozsa problem where entanglement is crucial, we have not been able to report finding corresponding classical oracles that diminish the quantum speed-up. Of course, in order to prove the link between genuine quantum speed-up and entanglement the non-trivial task of finding all the classical counterparts of an arbitrary multi-qubit quantum unitary operation must be solved.
Summarizing, we have shown that the common procedure for comparing quantum and classical oracles is ambiguous. This has led us to introduce the notion of genuine quantum speed-up which allows reliable comparison of quantum and classical oracles. As an example, we have shown that the Bernstein-Vazirani and PARITY problems do not exhibit genuine quantum speed-up.
###### Acknowledgements.
A. W. would like to thank the State Commision for Scientific Research for financial support under grant no. 0 T00A 003 23.
|
warning/0506/hep-ph0506025.html
|
ar5iv
|
text
|
# Intrinsic CPT violation and decoherence for entangled neutral mesons
## I Introduction
To date, the Special Theory of Relativity, the established theory of flat space-time physics based on Lorentz symmetry, is very well tested. In fact, during the current year, it completes a century of enormous success, having passed very stringent and diverse experimental tests. On the other hand, a quantum theory of Gravity, that is, a consistent quantized version of Einstein’s General Relativity, still eludes us. This may be partially attributed to the lack of any concrete observational evidence on the structure of space-time at the characteristic scale of quantum gravity (QG), the Planck mass-scale $`M_P10^{19}`$ GeV. One would hope that, like any other successful physical theory, a physically relevant quantum theory of gravity should lead to experimental predictions testable in the foreseeable future. In the QG case, however, such predictions may be both numerically suppressed and experimentally difficult to isolate. This is mainly due to the extremely weak nature of the gravitational interaction as compared with the rest of the known forces in nature. Specifically, the dimension-full coupling constant of gravity, the Newton constant $`G_N=1/M_P^2`$, appears as a very strong suppression factor of any physical observable that could be associated with predictions of quantum gravity. If the feebleness of the gravitational interaction were to be combined with the exact conservation of all symmetries and properties in particle physics, one would arrive at the inescapable conclusion that a true “phenomenology of QG” should be considered wishful thinking.
In recent years, however, physicists have ventured into the idea that laws, such as Lorentz invariance and unitarity, which are characteristic of a flat space-time quantum field theory, *may not* be valid in a full *quantum* theory of curved space-time. For instance, the invariance of the laws of Nature under the CPT symmetry cpt (i.e. the combined action of charge conjugation (C), parity-reflection (P), and time-reversal (T)) may not be exact in the presence of QG effects. Indeed, the possibility of a violation of CPT invariance (CPTV) by QG has been raised in a number of theoretical models, that go beyond conventional local quantum field theoretic treatments of gravity hawking ; ehns ; lopez ; peskin ; kostel ; bl ; bmp .
CPT violating scenarios may be classified into two major categories, according to the specific way the symmetry is violated. The first category contains models in which CPTV is associated with (spontaneous) breaking of (linear) Lorentz symmetry kostel and/or locality of the interactions bl . In these cases the generator of the CPT symmetry is a well-defined quantum mechanical operator, which however *does not commute* with the (effective) Hamiltonian of the system. In the second major category, the CPT operator, due to a variety of reasons, is *ill-defined*. The most notable example of such a type of CPTV is due to the entanglement of the matter theory with an “environment” of quantum gravitational degrees of freedom, leading to decoherence of matter ehns ; lopez ; peskin ; bmp . This decoherence, in turn, implies, by means of a powerful mathematical theorem due to Wald wald , that the quantum mechanical operator generating CPT transformations cannot be consistently defined. Wald refers to this phenomenon as “microscopic time irreversibility”, which captures the essence of the effect. We shall refer to such a failure of CPT through decoherence as “intrinsic”, in contradistinction to “extrinsic” violations, related to the non-commutativity of the matter Hamiltonian with a well-defined CPT generator kostel ; bl .
As emphasized recently in bmp , this intrinsic CPTV leads to modifications to the concept of an “antiparticle”. The resulting loss of particle-antiparticle identity in the neutral-meson system induces a breaking of the Einstein-Podolsky-Rosen (EPR) correlation imposed by Bose statistics. Specifically, in the CPT invariant case, the antiparticle of a given particle is defined as the state of equal mass (and life-time) and opposite values for all charges. If the CPT operator is well defined, such a state is obtained by the action of this operator on the corresponding particle state. If, however, the operator is ill-defined, the particle and antiparticle spaces should be thought of as independent subspaces of matter states. In such a case, the usual operating assumption that the electrically neutral meson states (Kaons or $`B`$ mesons) are *indistinguishable* from (“identical” to) their respective antiparticles ($`\overline{K}^0`$ or $`\overline{B}^0`$) is relaxed. This, in turn, modifies the symmetry properties in the description of (neutral) meson entangled states, and may bring about deviations to their EPR correlations bmp ; bmp2 . We emphasize, though, that the associated intrinsic CPTV effects are treated as *small perturbations*, being attributed to quantum-gravitational physics. From now on, we will refer to this effect as the “$`\omega `$ effect”, due to the complex parameter $`\omega `$ used in its parametrization.
It should be stressed at this point that the $`\omega `$ effect is distinct, both in its origin as well as the physical consequences, from analogous effects generated by the evolution of the system in the decoherence-inducing medium of QG, discussed in peskin . As already mentioned in bmp , and will be discussed in detail in the next section, the $`\omega `$ effect appears immediately after the $`\varphi `$-decay (at time $`t=0`$) due to the violation of the symmetry properties of the entangled Kaon state wave-function under permutations, assuming conservation of angular momentum. In other words, the $`\omega `$ effect provides an answer to the question: “What is the (symmetry of the) initial state of the $`K^0\overline{K}^0`$ system in the presence of a QG-decoherence inducing background ?” This is to be contrasted to the effects of peskin , stemming from the decoherent evolution of the system in the QG medium, which have been interpreted as a violation of angular momentum conservation. Our working hypothesis is that the QG medium conserves angular momentum, at least as far as the $`\omega `$ effect is concerned, extrapolating from no-hair theorems of classical macroscopic black holes.
In bmp it was assumed for simplicity that the intrinsic CPTV manifested itself only at the level of the wave-function describing the entangled two-meson state immediately after the decay of the initial resonance, whereas the subsequent time evolution of the system was taken to be due to an effective Hamiltonian. However, given that the origin of this intrinsic CPTV is usually attributed to the decoherent nature of QG, in the spirit discussed originally in wald , a complete description would necessitate the introduction of such “medium” effects in the time evolution as well. To accomplish such a description, at least for the purpose of providing a phenomenological framework for decoherent QG, no detailed knowledge of the underlying microscopic quantum gravity theory is necessary. This can be achieved following the so-called Lindblad or mathematical semi-groups approach to decoherence lindblad , which is a very efficient way of studying open systems in quantum mechanics footnote1 . The time irreversibility in the evolution of such semigroups, which is linked to decoherence, is inherent in the mathematical property of the lack of an inverse in the semigroup. This approach has been followed for the study of quantum-gravity decoherence in the case of neutral kaons ehns ; lopez , as well as other probes, such as neutrinos neutrinos ; footnote2 . The Lindblad approach to decoherence does not require any detailed knowledge of the environment, apart from energy conservation, entropy increase and complete positivity of the (reduced) density matrix $`\rho (t)`$ of the subsystem under consideration. The basic evolution equation for the (reduced) density matrix of the subsystem in the Lindblad approach is linear in $`\rho (t)`$. We notice that the Lindblad part, expressing interaction with the “environment” cannot be written as a commutator (of a Hamiltonian function) with $`\rho `$. Environmental contributions that can be cast in Hamiltonian evolution (commutator form) are absorbed in an “effective” hamiltonian.
The purpose of this article is to extend the analysis of bmp , where the intrinsic CPTV manifested itself only in the initial entangled meson state, to include QG environmental entanglement, by resorting to the above-mentioned Lindblad evolution formalism. Specifically, we shall compute various observables relevant for precision decoherence tests in a $`\varphi `$-factory, placing emphasis on the possibility of disentangling the decoherent evolution parameters $`\alpha `$, $`\beta `$, and $`\gamma `$ of ehns from the $`\omega `$ parameter of bmp , expressing the loss of particle-antiparticle identity mentioned above. The structure of the article is as follows: in section 2 we present our formalism, including notations and conventions, which agree with those used in peskin . In section 3 we describe the various observables of a $`\varphi `$ factory that constitute sensitive probes of our $`\omega `$ and decoherence $`\alpha ,\beta ,\gamma `$ effects, and explain how the various effects (or bounds thereof) can be disentangled experimentally. In this section we restrict ourselves only to leading order corrections in the CPTV and decoherence effects, and ignore $`ϵ^{}`$ corrections. Such corrections do not affect the functional form of the CPTV and decoherent evolution terms of the various observables, and their inclusion is discussed in an Appendix. Conclusions and discussion are presented in section 4. Particular emphasis in the discussion is given in the contamination of the observable giving traditionally (i.e. ignoring decoherent CPTV evolution) the $`ϵ^{}`$ effects, by terms depending on the decoherence coefficients. Finally, as already mentioned, in an Appendix we give the complete formulae for the observables of section 3, including $`ϵ^{}`$ corrections.
## II Formalism
In this section we shall follow the notation and conventions of dafne ; peskin . The CP violating parameters are defined as $`ϵ_S=ϵ_M+\mathrm{\Delta }`$, $`ϵ_L=ϵ_M\mathrm{\Delta }`$, according to which the physical mass-eigenstates are expressed as
$`|K_S`$ $`=`$ $`{\displaystyle \frac{N_S}{\sqrt{2}}}\left((1+ϵ_S)|K_0+(1ϵ_S)|\overline{K_0}\right)`$
$`|K_L`$ $`=`$ $`{\displaystyle \frac{N_L}{\sqrt{2}}}\left((1+ϵ_L)|K_0(1ϵ_L)|\overline{K_0}\right),`$ (1)
where $`N_{S,L}`$ are positive normalization constants, $`ϵ_M`$ is odd under CP, but even under CPT, and $`\mathrm{\Delta }`$ is odd under both CP and CPT. Furthermore,
$`ϵ_i^\pm `$ $`=`$ $`ϵ_i\pm {\displaystyle \frac{\beta }{d}}`$
$`ϵ_i^\pm `$ $`=`$ $`ϵ_i^{}\pm {\displaystyle \frac{\beta }{d^{}}},i=L,S`$ (2)
with
$`d`$ $`=`$ $`\mathrm{\Delta }m+{\displaystyle \frac{i}{2}}\mathrm{\Delta }\mathrm{\Gamma }=|d|e^{i(\pi /2\varphi _{SW})}`$ (3)
$`=`$ $`\left((3.483\pm 0.006)+i(3.668\pm 0.003)\right)\times 10^{15}\mathrm{GeV}`$
and the super-weak angle Eidelman:2004wy $`\varphi _{SW}=(43.5\pm 0.7)^\mathrm{o}`$.
To find the correct expressions for the observables in the density matrix formalism of ehns ; lopez ; peskin , matching the standard phenomenology in the CPT-conserving quantum mechanical case, we use the following parametrization for the amplitudes of the decay of $`|K_1,|K_2|\pi ^+\pi ^{}`$ peskin :
$`(K_1\pi ^+\pi ^{})=\sqrt{2}A_0e^{i\delta _0}+\mathrm{Re}A_2e^{i\delta _2}`$
$`(K_2\pi ^+\pi ^{})=i\mathrm{Im}A_2e^{i\delta _2}`$ (4)
where $`A_0`$ is real, and $`ϵ^{}=\frac{i}{\sqrt{2}}\frac{\mathrm{Im}A_2}{A_0}e^{i\delta }`$, $`\delta \delta _2\delta _0`$, and $`|\eta _+|^2|\frac{(K_L\pi ^+\pi ^{})}{(K_S\pi ^+\pi ^{})}|^2=|ϵ_L+\frac{1}{\sqrt{2}}\frac{i\mathrm{Im}A_2}{A_0}e^{i\delta }|^2`$, $`|\eta _{00}|^2|\frac{(K_L\pi ^0\pi ^0)}{(K_S\pi ^0\pi ^0)}|^2=|ϵ_L\sqrt{2}\frac{i\mathrm{Im}A_2}{A_0}e^{i\delta }|^2`$. We also introduce the standard parametrization $`\eta _+=\pi ^+\pi ^{}|K_L/\pi ^+\pi ^{}|K_S=|\eta _+|e^{i\varphi _+}`$, where Eidelman:2004wy $`|\eta _+|=(2.288\pm 0.014)\times 10^3`$, $`\varphi _+=(43.4\pm 0.7)^\mathrm{o}`$. Similar parametrizations can be used for the corresponding rations representing Kaon decays to two neutral as well as three pions. The corresponding phases are denoted $`\varphi _{00}`$ and $`\varphi _{3\pi }`$. We remind the reader that within quantum mechanics $`\eta _+=ϵ_L+ϵ^{}`$, $`\eta _{00}=ϵ_L2ϵ^{}`$.
All the above numbers are extracted without assuming CPT invariance. In the presence of CPTV decoherence parameters ($`\beta `$) this relation is modified, given that now $`ϵ_{L,S}`$ are replaced by $`ϵ_{L,S}^\pm `$ (2). In that case peskin , $`ϵ_L^{}`$ is related to $`\overline{\eta }_+`$, which replaces the quantum-mechanical $`\eta _+`$ defined above, through $`|\overline{\eta }_+|e^{i\varphi _+}=ϵ_L^{}+Y_+`$, where $`Y_+`$ includes the $`ϵ^{}`$ effects, $`Y_+=\pi ^+\pi ^{}|K_2/\pi ^+\pi ^{}|K_1`$, with $`K_1=\frac{1}{\sqrt{2}}\left(|K^0+|\overline{K}^0\right)`$ and $`K_2=\frac{1}{\sqrt{2}}\left(|K^0|\overline{K}^0\right)`$ the standard CP eigenstates. We will also use the abbreviations $`\mathrm{\Delta }ϵ_{SL}=ϵ_Sϵ_L`$ and $`ϵ_{SL}=ϵ_S+ϵ_L`$, and $`\overline{\mathrm{\Gamma }}=\frac{1}{2}(\mathrm{\Gamma }_L+\mathrm{\Gamma }_S)`$, as well as the definition peskin
$`R_L`$ $`=`$ $`|ϵ_L^{}|^2+{\displaystyle \frac{\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}+4{\displaystyle \frac{\beta }{\mathrm{\Delta }\mathrm{\Gamma }}}\mathrm{Im}\left({\displaystyle \frac{ϵ_L^{}d}{d^{}}}\right),`$ (5)
which we shall make use of in the present article. We also note the current experimental values $`\sqrt{R_L}=(2.30\pm 0.035)\times 10^3`$, $`2\mathrm{R}\mathrm{e}ϵ_L^+=(3.27\pm 0.12)\times 10^3`$. Finally, the relevant density matrices read $`\rho _L=|K_LK_L|`$, $`\rho _S=|K_SK_S|`$, $`\rho _I=|K_SK_L|`$, $`\rho _{\overline{I}}=|K_LK_S|`$. In the present article we shall use values for the decoherence parameters $`\alpha `$, $`\beta `$, and $`\gamma `$, which saturate the experimental bounds obtained by the CPLEAR experiment cplear :
$$\alpha <4.0\times 10^{17}\mathrm{GeV},|\beta |<2.3\times 10^{19}\mathrm{GeV},\gamma <3.7\times 10^{21}\mathrm{GeV}$$
(6)
We note at this stage that in certain models benatti , where complete positivity of the density matrix of the entangled kaon state has been invoked, there is only one non-vanishing decoherence parameter ($`\alpha =\gamma `$) in the formalism of ehns , which we adopt here. However, as the analysis of benatti has demonstrated, in such completely positive entangled models there are other parametrizations, with more decoherence parameters. Our intrinsic CPT-violation $`\omega `$-effects bmp can be easily incorporated in those formalisms, which however we shall not follow here. For our purposes, complete positivity is achieved as the special case of only one decoherence parameter ($`\gamma =\alpha `$) in the formulae that follow. We emphasize, however, that the issue of complete positivity is not entirely clear in a quantum-gravity context, where the relevant master equation may even be non-linear emnnl .
In conventional formulations of entangled meson states dunietz ; botella ; bernabeu one imposes the requirement of Bose statistics for the state $`K^0\overline{K}^0`$ (or $`B^0\overline{B}^0`$), which implies that the physical neutral meson-antimeson state must be symmetric under the combined operation $`C𝒫`$, with $`C`$ the charge conjugation and $`𝒫`$ the operator that permutes the spatial coordinates. Specifically, assuming conservation of angular momentum, and a proper existence of the antiparticle state (denoted by a bar), one observes that, for $`K^0\overline{K}^0`$ states which are $`C`$-conjugates with $`C=(1)^{\mathrm{}}`$ (with $`\mathrm{}`$ the angular momentum quantum number), the system has to be an eigenstate of $`𝒫`$ with eigenvalue $`(1)^{\mathrm{}}`$. Hence, for $`\mathrm{}=1`$, we have that $`C=`$, implying $`𝒫=`$. As a consequence of Bose statistics this ensures that for $`\mathrm{}=1`$ the state of two identical bosons is forbidden dunietz . However, these assumptions may not be valid if CPT symmetry is intrinsically violated, in the sense of the loss of particle-antiparticle identity. In such a case $`\overline{K}^0`$ cannot be considered as identical to $`K^0`$, and thus the requirement of $`C𝒫=+`$, imposed by Bose-statistics, is relaxed. Therefore, the initial state after the $`\varphi `$ decay, when expressed in terms of mass-eigenstates contains, in addition to the standard $`K_LK_S`$ terms, otherwise forbidden terms of the type $`K_LK_L`$, $`K_SK_S`$:
$`|i`$ $`=`$ $`C[(|K_S(\stackrel{}{k}),K_L(\stackrel{}{k})|K_L(\stackrel{}{k}),K_S(\stackrel{}{k}))`$ (7)
$`+`$ $`\omega (|K_S(\stackrel{}{k}),K_S(\stackrel{}{k})|K_L(\stackrel{}{k}),K_L(\stackrel{}{k}))]`$
where C is to be computed, and $`\omega `$ is a complex parameter. We hasten to emphasize at this stage that the genuine, quantum gravity induced $`\omega `$-effect, due to the loss of particle-antiparticle identity, should not be confused with the ordinary C-even-background effects that have been studied in benatti . In fact the $`\omega `$-effect can be easily disentangled from background effects, as has been discussed in bmp .
Under the non-unitary decoherent Lindblad evolution lindblad , appropriately tailor to the present problem lopez ; peskin
$$_t\rho (t)=i\rho (t)HiH^{}\rho (t)+\widehat{\delta H}\rho (t)$$
(8)
where $`H`$ denotes the (non-hermitian) Hamiltonian of the system (taking decay into account) and $`\widehat{\delta H}`$ contains the decoherent effects, the initial state evolves to a mixed state, which assumes the general structure peskin
$$\rho =\underset{i,j}{}A_{ij}\rho _i\rho _j,i,j=S,L,$$
(9)
where the constants $`A_{ij}`$ depend on the decoherence parameters.
The relevant observables are computed by means of double-decay rates with the following generic structure
$`𝒫(f_1,\tau _1;f_2,\tau _2)={\displaystyle \underset{ij}{}}A_{ij}tr[\rho _i𝒪_{f_1}]tr[\rho _j𝒪_{f_2}]e^{\lambda _i\tau _1\lambda _j\tau _2},`$ (10)
where the constants $`\lambda _i`$, $`\lambda _j`$ also depend on the decoherence parameters.
The double decay rate $`𝒫(f_1,\tau ;f_2,\tau )`$ interpolated at *equal times* $`\tau _1=\tau _2=\tau `$ is a quantity that pronounces the unusual time dependences of the decoherent evolution, which will also be calculated in this work.
Finally, of particular experimental interest are also integrated distributions at fixed time intervals $`\mathrm{\Delta }\tau =\tau _2\tau _1>0`$ (which we can assume for our purposes here):
$`\overline{𝒫}(f_1;f_2;\mathrm{\Delta }\tau >0)={\displaystyle \frac{1}{2}}{\displaystyle _{\mathrm{\Delta }\tau }^{\mathrm{}}}d(\tau _1+\tau _2)𝒫(f_1,\tau _1;f_2,\tau _2).`$ (11)
where the factor $`\frac{1}{2}`$ originates from the Jacobian when changing variables $`(\tau _1,\tau _2)(\tau _1+\tau _2,\mathrm{\Delta }\tau )`$.
In the particular case of a $`\varphi `$ factory that we analyze here, the pertinent density matrix describing the decay of one kaon of momentum $`\stackrel{}{k}`$ at time $`\tau _1`$ to a final state $`f_1`$ and the other of momentum $`\stackrel{}{k}`$ at time $`\tau _2`$ to a final state $`f_2`$ is given by
$`\rho `$ $`=`$ $`\rho _S\rho _L\left(1+\omega \mathrm{\Delta }ϵ_{SL}+\omega ^{}\mathrm{\Delta }ϵ_{SL}^{}\right)e^{\mathrm{\Gamma }_L\tau _2\mathrm{\Gamma }_S\tau _1}+\rho _L\rho _S\left(1\omega \mathrm{\Delta }ϵ_{SL}\omega ^{}\mathrm{\Delta }ϵ_{SL}^{}\right)e^{\mathrm{\Gamma }_L\tau _1\mathrm{\Gamma }_S\tau _2}`$
$`(\rho _I\rho _{\overline{I}}(1+\omega \mathrm{\Delta }ϵ_{SL}\omega ^{}{\displaystyle }ϵ_{SL}^{})e^{i\mathrm{\Delta }m(\tau _1\tau _2)}`$
$`+\rho _{\overline{I}}\rho _I(1\omega \mathrm{\Delta }ϵ_{SL}+\omega ^{}{\displaystyle }ϵ_{SL}^{})e^{+i\mathrm{\Delta }m(\tau _1\tau _2)})e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )(\tau _1+\tau _2)}`$
$`\left(\left(|\omega |^2{\displaystyle \frac{i\alpha }{\mathrm{\Delta }m}}\right)\rho _I\rho _Ie^{i\mathrm{\Delta }m(\tau _1+\tau _2)}+\left(|\omega |^2+{\displaystyle \frac{i\alpha }{\mathrm{\Delta }m}}\right)\rho _{\overline{I}}\rho _{\overline{I}}e^{+i\mathrm{\Delta }m(\tau _1+\tau _2)}\right)e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )(\tau _1+\tau _2)}`$
$`+\left(|\omega |^2{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}\right)\rho _S\rho _Se^{\mathrm{\Gamma }_S(\tau _1+\tau _2)}+\left(|\omega |^2+{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}\right)\rho _L\rho _Le^{\mathrm{\Gamma }_L(\tau _1+\tau _2)}`$
$`+\rho _I\rho _Se^{i\mathrm{\Delta }m\tau _1\mathrm{\Gamma }_S\tau _2(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _1}\left(\omega {\displaystyle \frac{2\beta }{d}}\right)\rho _S\rho _Ie^{i\mathrm{\Delta }m\tau _2\mathrm{\Gamma }_S\tau _1(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _2}\left(\omega +{\displaystyle \frac{2\beta }{d}}\right)`$
$`+\rho _{\overline{I}}\rho _Se^{+i\mathrm{\Delta }m\tau _1\mathrm{\Gamma }_S\tau _2(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _1}\left(\omega ^{}{\displaystyle \frac{2\beta }{d^{}}}\right)\rho _S\rho _{\overline{I}}e^{+i\mathrm{\Delta }m\tau _2\mathrm{\Gamma }_S\tau _1(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _2}\left(\omega ^{}+{\displaystyle \frac{2\beta }{d^{}}}\right)`$
$`+\rho _I\rho _Le^{i\mathrm{\Delta }m\tau _1\mathrm{\Gamma }_L\tau _2(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _1}\left(\omega ^{}+{\displaystyle \frac{2\beta }{d^{}}}\right)\rho _L\rho _Ie^{i\mathrm{\Delta }m\tau _2\mathrm{\Gamma }_L\tau _1(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _2}\left(\omega ^{}{\displaystyle \frac{2\beta }{d^{}}}\right)`$
$`+\rho _{\overline{I}}\rho _Le^{+i\mathrm{\Delta }m\tau _1\mathrm{\Gamma }_L\tau _2(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _1}\left(\omega +{\displaystyle \frac{2\beta }{d}}\right)\rho _L\rho _{\overline{I}}e^{+i\mathrm{\Delta }m\tau _2\mathrm{\Gamma }_L\tau _1(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _2}\left(\omega {\displaystyle \frac{2\beta }{d}}\right)`$
from the which the expression for the various constants in Eq.(10) may be gleaned. An important remark is now in order. Notice that above we have treated $`|\omega |^2`$ effects as being of comparable order to decoherence $`\alpha ,\gamma `$, and have kept only terms at most linear in the decoherence parameters. This is a very simplifying assumption, which, due to a lack of a microscopic theory of QG, cannot be made rigorous. However, as we shall discuss below, a reasonable *a posteriori* justification of this approximation may come from the fact that such terms appear on an equal footing as medium-generated corrections to certain physically important terms of (LABEL:initden) to be analyzed below.
In Eq.(LABEL:initden) we notice that, starting from the third line, terms of entirely novel type make their appearance. Such terms are due to both the intrinsic CPTV loss of particle-antiparticle identity ($`\omega `$-related) and decoherent evolution ($`\alpha `$,$`\beta `$,$`\gamma `$,-related). Such terms persist in the limit $`\tau _1=\tau _2=0`$, due to a specific choice of boundary conditions in the solution of the appropriate density-matrix evolution equation (8), expressing the omnipresence of the QG space-time foam effects. In particular, terms linear in $`\omega `$ are always accompanied by the parameter $`\beta `$, whereas terms quadratic in $`\omega `$ always combine with the $`\alpha `$ or $`\gamma `$ parameters. Of particular physical importance are the otherwise forbidden terms $`\rho _L\rho _L`$ and $`\rho _S\rho _S`$. At first sight, it is tempting to interpret peskin such terms as signaling a subtle breakdown of angular momentum conservation, in addition to the CPTV introduced by the “medium”. That seems to be an inescapable conclusion, if one were to begin the time evolution from a purely antisymmetric wave function (i.e. with $`\omega =0`$ in (7)), as assumed in peskin , which is an eigenstate of the orbital angular momentum $`L`$ with eigenvalue $`\mathrm{}=1`$. Since, then, the medium generates in the final state forbidden terms violating this last property, one interprets this as non-conservation of $`L`$. However, if one accepts bmp that the intrinsic CPTV affects not only the evolution but also the concept of the *antiparticle*, such terms are present already in the initial state. In such a case, there is no issue of non-conservation of angular momentum; the only effect of the time evolution is to simply modify the relative weights between $`\rho _L\rho _L`$ and $`\rho _S\rho _S`$ terms.
One can therefore say that QG may behave as a CPTV medium differentiating between particles and antiparticles. From a physical point of view one may even draw a vague analogy between QG media and “regenerators”, in the following sense: exactly as the regenerator differentiates between $`K^0`$ and $`\overline{K}^0`$, by means of different relative interactions, in a similar way QG acts differently on particles and antiparticles; however, unlike the regenerator case, where the experimentalist can control its position, due to their universal nature, QG effects are present even in the initial state after the $`\varphi `$ decay.
## III Observables at $`\varphi `$ factories
In this section we present a detailed study of a variety of observables measurable at $`\varphi `$ factories, and determine their dependence on both intrinsic CPTV ($`\omega `$) and decoherence ($`\alpha `$, $`\beta `$, $`\gamma `$) parameters, in an attempt to disentangle and separately constrain the two possible effects. Specifically, we will derive expressions for the double-decay rates of Eq.(10) and their integrated counter-parts of Eq.(11), for the cases of identical as well as general (non-identical) final states. We will restrict ourselves to linear effects in $`\alpha `$, $`\beta `$, $`\gamma `$, while keeping terms linear and quadratic in $`\omega `$, for the reasons explained in the previous section. For our purposes here we shall be interested in the decays of $`K_{L,S}`$ to final states consisting of two pions, $`\pi ^+\pi ^{}`$ or $`\pi ^0`$, three pions, as well as semileptonic decays to $`\mathrm{}^+\pi ^{}\nu `$ or $`\mathrm{}^{}\pi ^+\overline{\nu }`$. In this section we shall also ignore explicit $`ϵ^{}`$ effects. Such effects appear in the branching ratios $`\eta `$ of various pion channels, and their inclusion affects the form of the associated observables. We present complete formulae for the relevant double-decay time distributions, including such effects, in an Appendix. As we note there, the inclusion of such effects does not affect significantly the functional form of the decoherent/CPTV evolution.
In the density-matrix formalism of ehns ; peskin the relevant observables are given by
$`𝒪_+=\left(\begin{array}{cc}𝒳_+𝒴_+& \\ 𝒴_+^{}|\mathrm{Im}A_2e^{i\delta }|^2& \end{array}\right)`$ (15)
$`𝒪_{00}=\left(\begin{array}{cc}𝒳_{00}𝒴_{00}& \\ 𝒴_{00}^{}|2i\mathrm{Im}A_2e^{i\delta }|^2& \end{array}\right)`$ (18)
$`𝒪_\mathrm{}^+={\displaystyle \frac{|a|^2}{2}}\left(\begin{array}{cc}11& \\ 11& \end{array}\right),`$ (21)
$`𝒪_{\mathrm{}^{}}={\displaystyle \frac{|a|^2}{2}}\left(\begin{array}{cc}11& \\ 11& \end{array}\right),`$ (24)
$`𝒪_{3\pi }=|X_{3\pi }|^2\left(\begin{array}{cc}0& 0\\ 0& 1\end{array}\right).`$ (27)
where $`𝒳_+=|\sqrt{2}A_0+\mathrm{Re}A_2e^{i\delta }|^2`$, $`𝒴_+=(\sqrt{2}A_0+\mathrm{Re}A_2e^{i\delta })(i\mathrm{Im}A_2e^{i\delta })`$, $`𝒳_{00}=|\sqrt{2}A_02\mathrm{R}\mathrm{e}A_2e^{i\delta }|^2`$ and $`𝒴_{00}=(\sqrt{2}A_02\mathrm{R}\mathrm{e}A_2e^{i\delta })(2i\mathrm{Im}A_2e^{i\delta })`$, and $`X_{3\pi }=3\pi |K_2`$, in an obvious shorthand notation $`O_f`$, where $`f`$ denotes the observable associated with a Kaon decay leading to $`f`$ final state, for instance $`f=+`$ denotes decay to $`\pi ^+\pi ^{}`$, $`f=00`$ denotes decay to two $`\pi ^0`$, $`f=\mathrm{}^+`$ denotes semileptonic decay $`\mathrm{}^+\pi ^{}\nu `$ etc. Notice that if $`ϵ^{}`$ effects are ignored, which we shall do here for brevity, then we do not distinguish between the observables for the two cases of the two-pion final states.
Note that above we ignored CPT violating effects in the decay amplitudes to the final states. If such effects are included then the observables acquire the form peskin :
$`𝒪_+=|X_+|^2\left(\begin{array}{cc}1Y_+& \\ Y_+^{}|Y_+|^2& \end{array}\right)`$ (30)
$`𝒪_{00}=|X_{00}|^2\left(\begin{array}{cc}1Y_{00}& \\ Y_{00}^{}|Y_{00}|^2& \end{array}\right)`$ (33)
$`𝒪_\mathrm{}^+={\displaystyle \frac{|a+b|^2}{2}}\left(\begin{array}{cc}11& \\ 11& \end{array}\right),`$ (36)
$`𝒪_{\mathrm{}^{}}={\displaystyle \frac{|ab|^2}{2}}\left(\begin{array}{cc}11& \\ 11& \end{array}\right),`$ (39)
$`𝒪_{3\pi }=|X_{3\pi }|^2\left(\begin{array}{cc}|Y_{3\pi }|^2& Y_{3\pi }^{}\\ Y_{3\pi }& 1\end{array}\right).`$ (42)
where $`X=<\pi \pi |K_+>`$, $`Y=\frac{<\pi \pi |K_{}>}{<\pi \pi |K_+>}`$, and more specifically $`Y_+=(\frac{\mathrm{Re}(B_0)}{A_0})+ϵ^{}`$, $`Y_+=(\frac{\mathrm{Re}B_0}{A_0})2ϵ^{}`$, $`Y_{3\pi }=\frac{3\pi |K_1}{3\pi |K_2}`$, and $`\mathrm{Re}(a/b)`$ and $`\mathrm{Re}(B_0)/A_0`$ parametrize CPT violation in the decay amplitudes to the appropriate final states.
### III.1 Identical final states
Consider first the case where the kaons decay to $`\pi ^+\pi ^{}`$. Ignoring CPTV in the decays and $`ϵ^{}`$ effects, the relevant observable, $`𝒪_+`$, is obtained from the first of Eq.(27) by setting $`A_2=0`$, namely
$`𝒪_+=\left(\begin{array}{cc}2A_0^2& 0\\ 0& 0\end{array}\right)`$
The corresponding double-decay rate is then
$`𝒫`$ $`(\pi ^+\pi ^{},\tau _1;\pi ^+\pi ^{},\tau _2)=2A_0^4[R_L(e^{\mathrm{\Gamma }_L\tau _2\mathrm{\Gamma }_S\tau _1}+e^{\mathrm{\Gamma }_L\tau _1\mathrm{\Gamma }_S\tau _2})`$
$`2|\overline{\eta }_+|^2\mathrm{cos}(\mathrm{\Delta }m(\tau _1\tau _2))e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )(\tau _1+\tau _2)}+\left(|\omega |^2{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}{\displaystyle \frac{4\beta }{|d|}}|\overline{\eta }_+|{\displaystyle \frac{\mathrm{sin}\varphi _+}{\mathrm{cos}\varphi _{SW}}}\right)e^{\mathrm{\Gamma }_S(\tau _1+\tau _2)}`$
$`+{\displaystyle \frac{4\beta |\overline{\eta }_+|}{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _1+\varphi _+\varphi _{SW})e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _1\mathrm{\Gamma }_S\tau _2}`$
$`+{\displaystyle \frac{4\beta |\overline{\eta }_+|}{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _2+\varphi _+\varphi _{SW})e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _2\mathrm{\Gamma }_S\tau _1}`$
$`+2|\omega ||\overline{\eta }_+|(\mathrm{cos}(\tau _1+\varphi _+2\mathrm{\Omega })e^{\mathrm{\Gamma }_S(\tau _2)(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _1}\mathrm{cos}(\tau _2+\varphi _+\mathrm{\Omega })e^{\mathrm{\Gamma }_S\tau _1(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _2})]`$
The integrated observable $`\overline{𝒫}(f_1,f_2,\mathrm{\Delta }\tau )`$ is obtained by integrating over $`(\tau _1+\tau _2)`$ keeping $`\mathrm{\Delta }\tau `$ fixed. The result is:
$`\overline{𝒫}`$ $`(\pi ^+\pi ^{};\pi ^+\pi ^{};\mathrm{\Delta }\tau )=A_0^4[R_L{\displaystyle \frac{e^{\mathrm{\Delta }\tau \mathrm{\Gamma }_L}+e^{\mathrm{\Delta }\tau \mathrm{\Gamma }_S}}{\mathrm{\Gamma }_L+\mathrm{\Gamma }_S}}`$
$`|\overline{\eta }_+|^2\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau ){\displaystyle \frac{e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }}{(\overline{\mathrm{\Gamma }}+\alpha \gamma )}}+\left(|\omega |^2{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}{\displaystyle \frac{4\beta }{|d|}}|\overline{\eta }_+|{\displaystyle \frac{\mathrm{sin}\varphi _+}{\mathrm{cos}\varphi _{SW}}}\right){\displaystyle \frac{e^{\mathrm{\Gamma }_S\mathrm{\Delta }\tau }}{2\mathrm{\Gamma }_S}}`$
$`+{\displaystyle \frac{1}{\mathrm{\Delta }m^2+(\overline{\mathrm{\Gamma }}+\alpha \gamma )^2+2(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Gamma }_S+\mathrm{\Gamma }_S^2}}`$
$`\times [{\displaystyle \frac{4\beta |\overline{\eta }_+|}{|d|}}((\mathrm{\Delta }m\mathrm{cos}(\varphi _+\varphi _{SW})+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{sin}(\varphi _+\varphi _{SW}))e^{\mathrm{\Delta }\tau \mathrm{\Gamma }_S}`$
$`+e^{\mathrm{\Delta }\tau (\overline{\mathrm{\Gamma }}+\alpha \gamma )}(\mathrm{\Delta }m\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\varphi _+\varphi _{SW})`$
$`+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\varphi _+\varphi _{SW})))`$
$`+2|\omega ||\overline{\eta }_+|e^{\mathrm{\Delta }\tau \mathrm{\Gamma }_S}((\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{cos}(\varphi _+\mathrm{\Omega })\mathrm{\Delta }m\mathrm{sin}(\varphi _+\mathrm{\Omega }))`$
$`2|\omega ||\overline{\eta }_+|e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }((\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\varphi _+\mathrm{\Omega })`$
$`\mathrm{\Delta }m\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\varphi _+\mathrm{\Omega }))]]`$
which we plot in Figs. 1, 2. In the figures we assume rather large vales of $`|\omega |`$, of order $`|\overline{\eta }_+|`$. As evidenced by looking at the curves for relatively large values of $`\mathrm{\Delta }\tau `$, this assumption, together with the assumed values of the decoherence parameters (6), allows for a rather clear disentanglement of the decoherence plus $`\omega `$ situation from that with only $`\omega `$ effects present (i.e. unitary evolution). Indeed, in the latter case, the result for the pertinent time-integrated double-decay rate quickly converges to the quantum-mechanical situation, in contrast to the decoherent case. Moreover, the non-zero value of the time-integrated asymmetries at $`\mathrm{\Delta }\tau =0`$, is another clear deviation from the quantum mechanical case, which in the (unitary evolution) case with only $`\omega 0`$ is pronounced, due to the presence of $`|\overline{\eta }_+|`$ factors bmp .
We notice at this stage, though, that this may not be the case in an actual quantum gravity foam situation. For instance, it has been argued benatti that if *complete positivity* is imposed for the density matrix of entangled states, then the only non-trivial decoherence parameter would be $`\alpha =\gamma >0`$, whilst $`\beta =0`$. It is not inconceivable that $`|\omega |^2`$ are of the same order as $`2\gamma /\mathrm{\Delta }\mathrm{\Gamma }`$, in which case, for all practical purposes the integrated curve for the two-pion double decay rate (LABEL:twopionint) would pass through zero for $`\mathrm{\Delta }\tau =0`$. In such a situation, one should *also* look at the linear in $`\omega `$ interference terms for $`\mathrm{\Delta }\tau 0`$, in order to disentangle the $`\omega `$ from $`\gamma `$ effects. Lacking a fundamental microscopic theory underlying these phenomenological considerations, however, which would make definite predictions on the relative order of the various CPTV effects, we are unable to make more definite statements at present.
We next look at the semileptonic decays of Kaons. First we consider the case where both Kaons decay to $`\mathrm{}^+`$. As discussed in the beginning of the section, (27) the relevant observable is $`𝒪_{l+}`$,
$`𝒪_{l+}={\displaystyle \frac{|a|^2}{2}}\left(\begin{array}{cc}1& 1\\ 1& 1\end{array}\right)`$
Recalling that $`ϵ_L=ϵ_M\mathrm{\Delta }`$ and $`ϵ_S=ϵ_M+\mathrm{\Delta }`$ so $`(ϵ_{SL}+ϵ_{SL}^{})=4\mathrm{R}\mathrm{e}ϵ_M`$ we have for the double-decay rate:
$`𝒫(l^\pm ,\tau _1;l^\pm ,\tau _2)`$ $`=`$ $`{\displaystyle \frac{a^4}{8}}[(1\pm 4\mathrm{R}\mathrm{e}(ϵ_M))(e^{\mathrm{\Gamma }_L\tau _2\mathrm{\Gamma }_S\tau _1}+e^{\mathrm{\Gamma }_L\tau _1\mathrm{\Gamma }_S\tau _2})`$
$`2(1\pm 4\mathrm{R}\mathrm{e}(ϵ_M))\mathrm{cos}(\mathrm{\Delta }m(\tau _1\tau _2))e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )(\tau _1+\tau _2)}`$
$`2|\omega |^2\mathrm{cos}(\mathrm{\Delta }m(\tau _1+\tau _2))e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )(\tau _1+\tau _2)}+|\omega |^2\left(e^{\mathrm{\Gamma }_S(\tau _1+\tau _2)}+e^{\mathrm{\Gamma }_L(\tau _1+\tau _2)}\right)`$
$`+{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}\left(e^{\mathrm{\Gamma }_L(\tau _1+\tau _2)}e^{\mathrm{\Gamma }_S(\tau _1+\tau _2)}\right)+{\displaystyle \frac{2\alpha }{\mathrm{\Delta }m}}\mathrm{sin}(\mathrm{\Delta }m(\tau _1+\tau _2))e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )(\tau _1+\tau _2)}`$
$`\pm \left({\displaystyle \frac{4\beta }{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _1\varphi _{SW})+2|\omega |\mathrm{cos}(\mathrm{\Delta }m\tau _1\mathrm{\Omega })\right)e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _1}e^{\mathrm{\Gamma }_S\tau _2}`$
$`\pm \left({\displaystyle \frac{4\beta }{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _2\varphi _{SW})2|\omega |\mathrm{cos}(\mathrm{\Delta }m\tau _2\mathrm{\Omega })\right)e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _2}e^{\mathrm{\Gamma }_S\tau _1}`$
$`\pm \left({\displaystyle \frac{4\beta }{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _1+\varphi _{SW})+2|\omega |\mathrm{cos}(\mathrm{\Delta }m\tau _1+\mathrm{\Omega })\right)e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _1}e^{\mathrm{\Gamma }_L\tau _2}`$
$`\pm ({\displaystyle \frac{4\beta }{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _2+\varphi _{SW})2|\omega |\mathrm{cos}(\mathrm{\Delta }m\tau _2+\mathrm{\Omega }))e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _2}e^{\mathrm{\Gamma }_L\tau _1}]`$
and the integrated quantity reads:
$`\overline{𝒫}(`$ $`l^\pm `$ $`;l^\pm ,\mathrm{\Delta }\tau )={\displaystyle \frac{a^4}{16}}[2(1\pm 4\mathrm{R}\mathrm{e}ϵ_M){\displaystyle \frac{e^{\mathrm{\Gamma }_L\mathrm{\Delta }\tau }+e^{\mathrm{\Gamma }_S\mathrm{\Delta }\tau }}{\mathrm{\Gamma }_S+\mathrm{\Gamma }_L}}`$
$`(1\pm 4\mathrm{R}\mathrm{e}ϵ_M)2\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau ){\displaystyle \frac{e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }}{(\overline{\mathrm{\Gamma }}+\alpha \gamma )}}`$
$`2|\omega |^2e^{\mathrm{\Delta }\tau (\overline{\mathrm{\Gamma }}+\alpha \gamma )}{\displaystyle \frac{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau )\mathrm{\Delta }m\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau )}{(\overline{\mathrm{\Gamma }}+\alpha \gamma )^2+\mathrm{\Delta }m^2}}`$
$`+|\omega |^2\left({\displaystyle \frac{e^{\mathrm{\Gamma }_S\mathrm{\Delta }\tau }}{\mathrm{\Gamma }_S}}+{\displaystyle \frac{e^{\mathrm{\Gamma }_L\mathrm{\Delta }\tau }}{\mathrm{\Gamma }_L}}\right)+{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}\left({\displaystyle \frac{e^{\mathrm{\Gamma }_L\mathrm{\Delta }\tau }}{\mathrm{\Gamma }_L}}{\displaystyle \frac{e^{\mathrm{\Gamma }_S\mathrm{\Delta }\tau }}{\mathrm{\Gamma }_S}}\right)`$
$`+{\displaystyle \frac{2\alpha }{\mathrm{\Delta }m}}e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }{\displaystyle \frac{((\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau )+\mathrm{\Delta }m\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau ))}{(\overline{\mathrm{\Gamma }}+\alpha \gamma )^2+\mathrm{\Delta }m^2}}`$
$`\pm {\displaystyle \frac{4}{\mathrm{\Delta }m^2+(\overline{\mathrm{\Gamma }}+\alpha \gamma )^2+2(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Gamma }_S+\mathrm{\Gamma }_S^2}}`$
$`\times ({\displaystyle \frac{2\beta }{|d|}}e^{\mathrm{\Delta }\tau \mathrm{\Gamma }_S}(\mathrm{\Delta }m\mathrm{cos}(\varphi _{SW})(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{sin}(\varphi _{SW})))`$
$`+|\omega |e^{\mathrm{\Delta }\tau \mathrm{\Gamma }_S}(\mathrm{\Delta }m\mathrm{sin}(\mathrm{\Omega })+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{cos}(\mathrm{\Omega }))`$
$`+{\displaystyle \frac{2\beta }{|d|}}e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }(\mathrm{\Delta }m\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau \varphi _{SW})+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau \varphi _{SW}))`$
$`+|\omega |e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }(\mathrm{\Delta }m\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau \mathrm{\Omega })+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau \mathrm{\Omega })))`$
$`\pm {\displaystyle \frac{4}{\mathrm{\Delta }m^2+(\overline{\mathrm{\Gamma }}+\alpha \gamma )^2+2(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Gamma }_L+\mathrm{\Gamma }_L^2}}`$
$`\times ({\displaystyle \frac{2\beta }{|d|}}e^{\mathrm{\Delta }\tau \mathrm{\Gamma }_L}(\mathrm{\Delta }m\mathrm{cos}(\varphi _{SW})+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_L)\mathrm{sin}(\varphi _{SW}))`$
$`+|\omega |e^{\mathrm{\Delta }\tau \mathrm{\Gamma }_L}(\mathrm{\Delta }m\mathrm{sin}(\mathrm{\Omega })+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_L)\mathrm{cos}(\mathrm{\Omega }))`$
$`+{\displaystyle \frac{2\beta }{|d|}}e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }(\mathrm{\Delta }m\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau \varphi _{SW})+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_L)\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau \varphi _{SW}))`$
$`+|\omega |e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }(\mathrm{\Delta }m\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\mathrm{\Omega })+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_L)\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\mathrm{\Omega })))]`$
We plot the corresponding integrated double-decay rate for $`\mathrm{}^+\mathrm{}^+`$ in Fig. 3.
Next we consider the case in which one of the Kaons decays to $`\mathrm{}^+`$ and the other to $`\mathrm{}^{}`$. On noting that the relevant observable corresponding to the $`\mathrm{}^{}`$ case is associated with the operator $`𝒪_l`$ of (27):
$`𝒪_l={\displaystyle \frac{|a|^2}{2}}\left(\begin{array}{cc}1& 1\\ 1& 1\end{array}\right)`$
the relevant double-decay rate reads:
$`𝒫(l^+,\tau _1;l^{},\tau _2)`$ $`=`$ $`{\displaystyle \frac{a^4}{8}}[(1+4\mathrm{R}\mathrm{e}(\mathrm{\Delta }\beta /d))e^{\mathrm{\Gamma }_L\tau _2\mathrm{\Gamma }_S\tau _1}+(1+4\mathrm{R}\mathrm{e}(\beta /d\mathrm{\Delta }))e^{\mathrm{\Gamma }_L\tau _1\mathrm{\Gamma }_S\tau _2}`$
$`+2\left(\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau )+4\mathrm{I}\mathrm{m}(\mathrm{\Delta }+\beta /d)\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau )\right)e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )(\tau _1+\tau _2)}`$
$`+\left(2|\omega |^2\mathrm{cos}(\mathrm{\Delta }m(\tau _1+\tau _2)){\displaystyle \frac{2\alpha }{\mathrm{\Delta }m}}\mathrm{sin}(\mathrm{\Delta }m(\tau _1+\tau _2))\right)e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )(\tau _1+\tau _2)}`$
$`+\left(\left(|\omega |^2{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}\right)e^{\mathrm{\Gamma }_S(\tau _1+\tau _2)}+\left(|\omega |^2+{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}\right)e^{\mathrm{\Gamma }_L(\tau _1+\tau _2)}\right)`$
$`+\left({\displaystyle \frac{4\beta }{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _1\varphi _{SW})+2|\omega |\mathrm{cos}(\mathrm{\Delta }m\tau _1\mathrm{\Omega })\right)e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _1}e^{\mathrm{\Gamma }_S\tau _2}`$
$`\left({\displaystyle \frac{4\beta }{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _2\varphi _{SW})2|\omega |\mathrm{cos}(\mathrm{\Delta }m\tau _2\mathrm{\Omega })\right)e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _2}e^{\mathrm{\Gamma }_S\tau _1}`$
$`+\left({\displaystyle \frac{4\beta }{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _1+\varphi _{SW})+2|\omega |\mathrm{cos}(\mathrm{\Delta }m\tau _1+\mathrm{\Omega })\right)e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _1}e^{\mathrm{\Gamma }_L\tau _2}`$
$`({\displaystyle \frac{4\beta }{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _2+\varphi _{SW})2|\omega |\mathrm{cos}(\mathrm{\Delta }m\tau _2+\mathrm{\Omega }))e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _2}e^{\mathrm{\Gamma }_L\tau _1}]`$
and the integrated over time distribution is:
$`\overline{𝒫}`$ $`(l^+;l^{},\mathrm{\Delta }\tau )={\displaystyle \frac{a^4}{16}}[(1+4\mathrm{R}\mathrm{e}(\mathrm{\Delta }\beta /d)){\displaystyle \frac{2e^{\mathrm{\Gamma }_L\mathrm{\Delta }\tau }}{\mathrm{\Gamma }_S+\mathrm{\Gamma }_L}}+(1+4\mathrm{R}\mathrm{e}(\beta /d\mathrm{\Delta })){\displaystyle \frac{2e^{\mathrm{\Gamma }_S\mathrm{\Delta }\tau }}{\mathrm{\Gamma }_S+\mathrm{\Gamma }_L}}`$
$`+2\left(\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau )+4\mathrm{I}\mathrm{m}(\mathrm{\Delta }+\beta /d)\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau )\right){\displaystyle \frac{e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }}{(\overline{\mathrm{\Gamma }}+\alpha \gamma )}}`$
$`+{\displaystyle \frac{e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }}{(\overline{\mathrm{\Gamma }}+\alpha \gamma )+\mathrm{\Delta }m^2}}(2|\omega |^2((\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau )\mathrm{\Delta }m\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau ))`$
$`{\displaystyle \frac{2\alpha }{\mathrm{\Delta }m}}(\mathrm{\Delta }m\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau )+(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau )))`$
$`+\left(\left(|\omega |^2{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}\right){\displaystyle \frac{e^{\mathrm{\Gamma }_S\mathrm{\Delta }\tau }}{\mathrm{\Gamma }_S}}+\left(|\omega |^2+{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}\right){\displaystyle \frac{e^{\mathrm{\Gamma }_L\mathrm{\Delta }\tau }}{\mathrm{\Gamma }_L}}\right)`$
$`+{\displaystyle \frac{4}{\mathrm{\Delta }m^2+(\overline{\mathrm{\Gamma }}+\alpha \gamma )^2+2(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Gamma }_S+\mathrm{\Gamma }_S^2}}`$
$`\times ({\displaystyle \frac{2\beta }{|d|}}e^{\mathrm{\Delta }\tau \mathrm{\Gamma }_S}(\mathrm{\Delta }m\mathrm{cos}(\varphi _{SW})(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{sin}(\varphi _{SW}))`$
$`+|\omega |e^{\mathrm{\Delta }\tau \mathrm{\Gamma }_S}(\mathrm{\Delta }m\mathrm{sin}(\mathrm{\Omega })+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{cos}(\mathrm{\Omega }))`$
$`{\displaystyle \frac{2\beta }{|d|}}e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }(\mathrm{\Delta }m\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau \varphi _{SW})+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau \varphi _{SW}))`$
$`|\omega |e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }(\mathrm{\Delta }m\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau \mathrm{\Omega })+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau \mathrm{\Omega })))`$
$`+{\displaystyle \frac{4}{\mathrm{\Delta }m^2+(\overline{\mathrm{\Gamma }}+\alpha \gamma )^2+2(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Gamma }_L+\mathrm{\Gamma }_L^2}}`$
$`\times ({\displaystyle \frac{2\beta }{|d|}}e^{\mathrm{\Delta }\tau \mathrm{\Gamma }_L}(\mathrm{\Delta }m\mathrm{cos}(\varphi _{SW})+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_L)\mathrm{sin}(\varphi _{SW}))`$
$`+|\omega |e^{\mathrm{\Delta }\tau \mathrm{\Gamma }_L}(\mathrm{\Delta }m\mathrm{sin}(\mathrm{\Omega })+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_L)\mathrm{cos}(\mathrm{\Omega }))`$
$`{\displaystyle \frac{2\beta }{|d|}}e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }(\mathrm{\Delta }m\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau \varphi _{SW})+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_L)\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau \varphi _{SW}))`$
$`+|\omega |e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }(\mathrm{\Delta }m\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\mathrm{\Omega })+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_L)\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\mathrm{\Omega })))]`$
We next give the three pion decay channels using the relevant observable (27),
$$𝒪_{3\pi }=|X_{3\pi }|^2\left(\begin{array}{cc}|Y_{3\pi }|^2& Y_{3\pi }^{}\\ Y_{3\pi }& 1\end{array}\right).$$
(52)
Here we take $`|Y_{3\pi }|=0`$, since we ignore the associated CP and CPT violation in the decay and obtain:
$$𝒪_{3\pi }=|X_{3\pi }|^2\left(\begin{array}{cc}0& 0\\ 0& 1\end{array}\right).$$
(53)
This leads to the following expression for the associated double-time distribution
$`𝒫`$ $`(3\pi ,\tau _1;3\pi ,\tau _2)={\displaystyle \frac{|X_{3\pi }|^4}{2}}[R_Se^{\mathrm{\Gamma }_L\tau _2\mathrm{\Gamma }_S\tau _1}+R_Se^{\mathrm{\Gamma }_L\tau _1\mathrm{\Gamma }_S\tau _2}`$
$`2|\overline{\eta }_{3\pi }|^2\mathrm{cos}(\mathrm{\Delta }m(\tau _1\tau _2))e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )(\tau _1+\tau _2)}+\left(|\omega |^2+{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}+{\displaystyle \frac{4\beta }{|d|}}|\overline{\eta }_{3\pi }|{\displaystyle \frac{\mathrm{sin}\varphi _{3\pi }}{\mathrm{cos}\varphi _{SW}}}\right)e^{\mathrm{\Gamma }_L(\tau _1+\tau _2)}`$
$`+|\overline{\eta }_{3\pi }|e^{\mathrm{\Gamma }_L\tau _1(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _2}\left(2|\omega |\mathrm{cos}(\mathrm{\Delta }m\tau _2\varphi _{3\pi }+\mathrm{\Omega })+{\displaystyle \frac{4\beta }{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _2\varphi _{3\pi }+\varphi _{SW})\right)`$
$`+|\overline{\eta }_{3\pi }|e^{\mathrm{\Gamma }_L\tau _2(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _1}(2|\omega |\mathrm{cos}(\mathrm{\Delta }m\tau _1\varphi _{3\pi }+\mathrm{\Omega })+{\displaystyle \frac{4\beta }{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _1\varphi _{3\pi }+\varphi _{SW}))],`$
which integrated over time gives
$`\overline{𝒫}(3\pi ,\tau _1`$ ; $`3\pi ,\tau _2)={\displaystyle \frac{|X_{3\pi }|^4}{4}}[{\displaystyle \frac{2R_S(e^{\mathrm{\Gamma }_L\mathrm{\Delta }\tau }+e^{\mathrm{\Gamma }_S\mathrm{\Delta }\tau })}{\mathrm{\Gamma }_S+\mathrm{\Gamma }_L}}`$
$`2|\overline{\eta }_{3\pi }|^2\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau ){\displaystyle \frac{e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }}{\overline{\mathrm{\Gamma }}+\alpha \gamma }}+\left(|\omega |^2+{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}+{\displaystyle \frac{4\beta }{|d|}}|\overline{\eta }_{3\pi }|{\displaystyle \frac{\mathrm{sin}\varphi _{3\pi }}{\mathrm{cos}\varphi _{SW}}}\right){\displaystyle \frac{e^{\mathrm{\Gamma }_L\mathrm{\Delta }\tau }}{\mathrm{\Gamma }_L}}`$
$`+{\displaystyle \frac{4|\overline{\eta }_{3\pi }|e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }}{\mathrm{\Delta }m^2+(\overline{\mathrm{\Gamma }}+\alpha \gamma )^2+2(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Gamma }_L+\mathrm{\Gamma }_L^2}}`$
$`\times (|\omega |e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }((\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_L)\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau \varphi _{3\pi }+\mathrm{\Omega })`$
$`+\mathrm{\Delta }m\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau \varphi _{3\pi }+\mathrm{\Omega }))`$
$`+{\displaystyle \frac{2\beta }{|d|}}e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }((\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_L)\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau \varphi _{3\pi }+\varphi _{SW})`$
$`+\mathrm{\Delta }m\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau \varphi _{3\pi }+\varphi _{SW}))`$
$`+|\omega |e^{\mathrm{\Gamma }_L\mathrm{\Delta }\tau }((\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_L)\mathrm{cos}(\mathrm{\Omega }\varphi _{3\pi })\mathrm{\Delta }m\mathrm{sin}(\mathrm{\Omega }\varphi _{3\pi }))`$
$`+{\displaystyle \frac{2\beta }{|d|}}e^{\mathrm{\Gamma }_L\mathrm{\Delta }\tau }((\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_L)\mathrm{sin}(\varphi _{SW}\varphi _{3\pi })+\mathrm{\Delta }m\mathrm{cos}(\varphi _{SW}\varphi _{3\pi })))].`$
We plot this function vs. $`\mathrm{\Delta }\tau `$ in Fig. 6. We use units of $`\tau _L`$ for convenience. The quantum-mechanical case with $`\omega =\alpha =\beta =\gamma =0`$ coincides with the horizontal axis. As we observe from the graph, the CPTV and decoherence effects are pronounced near $`\mathrm{\Delta }\tau =0`$. Thus, in principle this is the cleanest way of bounding such effects, but unfortunately in practice this is a very difficult channel to measure experimentally.
This completes the analysis of identical-final-states observables.
### III.2 General Final States
We consider in this subsection observables for the case where the final states are different.
We first consider the case in which one Kaon decays to $`\pi ^+\pi ^{}`$ and the other to $`\pi ^0\pi ^0`$. For the double-decay rate we find
$`𝒫`$ $`(\pi ^+\pi ^{},\tau _1;\pi ^0\pi ^0,\tau _2)=2A_0^4[R_L^{00}e^{\mathrm{\Gamma }_L\tau _2\mathrm{\Gamma }_S\tau _1}+R_L^+e^{\mathrm{\Gamma }_L\tau _1\mathrm{\Gamma }_S\tau _2}`$
$`2|\overline{\eta }_+||\overline{\eta }_{00}|\mathrm{cos}(\mathrm{\Delta }m(\tau _1\tau _2)+\varphi _+\varphi _{00})e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )(\tau _1+\tau _2)}`$
$`+\left(|\omega |^2{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}{\displaystyle \frac{2\beta }{|d|}}{\displaystyle \frac{(|\overline{\eta }_+|\mathrm{sin}\varphi _++|\overline{\eta }_{00}|\mathrm{sin}\varphi _{00})}{\mathrm{cos}\varphi _{SW}}}\right)e^{\mathrm{\Gamma }_S(\tau _1+\tau _2)}`$
$`+|\overline{\eta }_+|e^{\mathrm{\Gamma }_S\tau _2(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _1}\left(2|\omega |\mathrm{cos}(\mathrm{\Delta }m\tau _1+\varphi _+\mathrm{\Omega })+{\displaystyle \frac{4\beta }{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _1+\varphi _+\varphi _{SW})\right)`$
$`+|\overline{\eta }_{00}|e^{\mathrm{\Gamma }_S\tau _1(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _2}(2|\omega |\mathrm{cos}(\mathrm{\Delta }m\tau _2+\varphi _{00}\mathrm{\Omega })+{\displaystyle \frac{4\beta }{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _2+\varphi _{00}\varphi _{SW}))].`$
The integrated-over-time distribution is given by
$`\overline{𝒫}`$ $`(\pi ^+\pi ^{};\pi ^0\pi ^0;\mathrm{\Delta }\tau )=2A_0^4[R_L^{00}{\displaystyle \frac{e^{\mathrm{\Gamma }_L\mathrm{\Delta }\tau }}{\mathrm{\Gamma }_L+\mathrm{\Gamma }_S}}+R_L^+{\displaystyle \frac{e^{\mathrm{\Gamma }_S\mathrm{\Delta }\tau }}{\mathrm{\Gamma }_L+\mathrm{\Gamma }_S}}`$
$`|\overline{\eta }_+||\overline{\eta }_{00}|\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\varphi _+\varphi _{00}){\displaystyle \frac{e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }}{\overline{\mathrm{\Gamma }}+\alpha \gamma }}`$
$`+\left(|\omega |^2{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}{\displaystyle \frac{2\beta }{|d|}}{\displaystyle \frac{(|\overline{\eta }_+|\mathrm{sin}\varphi _++|\overline{\eta }_{00}|\mathrm{sin}\varphi _{00})}{\mathrm{cos}\varphi _{SW}}}\right){\displaystyle \frac{e^{\mathrm{\Gamma }_S\mathrm{\Delta }\tau }}{2\mathrm{\Gamma }_S}}`$
$`+{\displaystyle \frac{2}{\mathrm{\Delta }m^2+(\overline{\mathrm{\Gamma }}+\alpha \gamma )^2+2(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Gamma }_S+\mathrm{\Gamma }_S^2}}`$
$`\times [|\overline{\eta }_+|e^{\mathrm{\Gamma }_S\mathrm{\Delta }\tau }(|\omega |((\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{cos}(\varphi _+\mathrm{\Omega })\mathrm{\Delta }m\mathrm{sin}(\varphi _+\mathrm{\Omega }))`$
$`+{\displaystyle \frac{2\beta }{|d|}}((\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{sin}(\varphi _+\varphi _{SW})+\mathrm{\Delta }m\mathrm{cos}(\varphi _+\varphi _{SW})))`$
$`+|\overline{\eta }_{00}|e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }(|\omega |((\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\varphi _{00}\mathrm{\Omega })+\mathrm{\Delta }m\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\varphi _{00}\mathrm{\Omega }))`$
$`+{\displaystyle \frac{2\beta }{|d|}}(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)(\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\varphi _{00}\varphi _{SW})+\mathrm{\Delta }m\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\varphi _{00}\varphi _{SW})))]].`$
Finally we consider the case in which one of the final states consists of two pions, and the other is the result of a kaon semileptonic decay. The relevant double-decay rate is given by:
$`𝒫`$ $`(\pi ^\pm ,\tau _1;l^\pm ,\tau _2)={\displaystyle \frac{A_0^2a^2}{4}}[(1\pm \delta _L+{\displaystyle \frac{\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}})e^{\mathrm{\Gamma }_L\tau _2\mathrm{\Gamma }_S\tau _1}+R_Le^{\mathrm{\Gamma }_L\tau _1\mathrm{\Gamma }_S\tau _2}`$
$`2|\overline{\eta }_+|\mathrm{cos}(\mathrm{\Delta }m(\tau _1\tau _2)+\varphi _+)e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )(\tau _1+\tau _2)}+\left(|\omega |^2{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}{\displaystyle \frac{4\beta |\overline{\eta }_+|}{|d|}}{\displaystyle \frac{\mathrm{sin}\varphi _+}{\mathrm{cos}\varphi _{SW}}}\right)e^{\mathrm{\Gamma }_S(\tau _1+\tau _2)}`$
$`\pm ({\displaystyle \frac{4\beta }{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _2\varphi _{SW})2|\omega |\mathrm{cos}(\mathrm{\Delta }m\tau _2\mathrm{\Omega }))e^{\mathrm{\Gamma }_S\tau _1(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _2}]`$
where we used the definition $`\delta _L=2\mathrm{R}\mathrm{e}ϵ_L^+`$.
The integrated double-decay rate reads:
$`\overline{𝒫}`$ $`(\pi ^\pm ;l^\pm ,\mathrm{\Delta }\tau )={\displaystyle \frac{A_0^2a^2}{4}}[(1\pm \delta _L+{\displaystyle \frac{\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}){\displaystyle \frac{e^{\mathrm{\Gamma }_L\mathrm{\Delta }\tau }}{\mathrm{\Gamma }_L+\mathrm{\Gamma }_S}}+R_L{\displaystyle \frac{e^{\mathrm{\Gamma }_S\mathrm{\Delta }\tau }}{\mathrm{\Gamma }_L+\mathrm{\Gamma }_S}}`$
$`|\overline{\eta }_+|\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\varphi _+){\displaystyle \frac{e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }}{(\overline{\mathrm{\Gamma }}+\alpha \gamma )}}+\left(|\omega |^2{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}{\displaystyle \frac{4\beta |\overline{\eta }_+|}{|d|}}{\displaystyle \frac{\mathrm{sin}\varphi _+}{\mathrm{cos}\varphi _{SW}}}\right){\displaystyle \frac{e^{\mathrm{\Gamma }_S\mathrm{\Delta }\tau }}{2\mathrm{\Gamma }_S}}`$
$`\pm {\displaystyle \frac{2}{\mathrm{\Delta }m^2+(\overline{\mathrm{\Gamma }}+\alpha \gamma )^2+2(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Gamma }_S+\mathrm{\Gamma }_S^2}}`$
$`+({\displaystyle \frac{2\beta }{|d|}}e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }(\mathrm{\Delta }m\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau \varphi _{SW})(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau \varphi _{SW}))`$
$`|\omega |e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }(\mathrm{\Delta }m\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau \mathrm{\Omega })+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau \mathrm{\Omega })))`$
This completes our analysis of observables in a $`\varphi `$ factory that can be used as sensitive probes for tests of possible CPTV and quantum decoherence, to leading order in the small parameters parametrizing the effects.
## IV Discussion and Conclusions
In this work, we have embarked into a combined treatment of decoherent and intrinsic CPTV effects in a $`\varphi `$ factory. By studying a variety of observables, involving identical as well as general final states, we have derived analytical expressions for double-decay rates and their integrated counterparts (over time $`\tau _1+\tau _2`$, keeping $`\mathrm{\Delta }\tau =\tau _1\tau _2`$ fixed) , which we plotted as functions of $`\mathrm{\Delta }\tau `$. Our analysis included $`\omega `$, as well as decoherence $`\alpha ,\beta ,\gamma `$ effects. We presented the results to leading order in these small parameters.
Although the pertinent formulae are algebraically rather complex, nevertheless from our general study it became evident that one may disentangle the $`\omega `$ from the decoherent evolution effects by looking simultaneously at different regimes of $`\mathrm{\Delta }\tau `$, namely, (i) $`\mathrm{\Delta }\tau =0`$, (ii) intermediate values of $`\mathrm{\Delta }\tau `$ (as compared with $`\tau _S`$, which sets a convenient characteristic time scale in the problem), with emphasis on interference terms with sinusoidal time dependence, and (iii) rather large values of $`\mathrm{\Delta }\tau `$, with emphasis on the exponential damping of the relevant quantities, which is affected (slowed down) by the presence of decoherence terms $`\alpha \gamma `$, in an $`\omega `$-independent way.
In addition to these results in a $`\varphi `$ factory, one can obtain valuable information on the decoherence parameters $`\alpha ,\beta `$,and $`\gamma `$ by looking at experiments involving single neutral Kaon beams, such as CPLEAR cplear , which is achieved through flavour tagging of the neutral Kaon by means of the electric charge of the pion on the complementary side. As discussed in lopez , the decoherence parameters $`\alpha ,\beta ,\gamma `$ can be separately disentangled by combined studies of several kaon asymmetries, which are again time-dependent functions. Notice though that, if complete positivity of entangled state density matrices is imposed benatti , then only one decoherence parameter $`\gamma >0`$ survives in the model of ehns , which facilitates the situation enormously.
We would like at this point to make a clarification concerning the important differences between proposed experiments here, using entangled Kaon states, and some existing proposals, notably by the CPLEAR collaboration cplearepr , to measure EPR correlations. The reader should notice that the accuracy for measuring the appropriate observables in cplearepr appears to be of order $`10^1`$; however, this accuracy pertains only to an asymmetry built from $`\overline{p}pK^0\overline{K}^0`$ in CPLEAR, looking for the dilepton decay channel, between “like” ($`K^0K^0`$ or $`\overline{K}^0\overline{K}^0`$) and “unlike” ($`K^0\overline{K}^0`$ or $`\overline{K}^0K^0`$) decays after time evolution. As shown in our paper, the most interesting observable for the $`\omega `$-effect comes from identical decay channels, i.e., in this case the “intra” asymmetry between $`K^0K^0`$ and $`\overline{K}^0\overline{K}^0`$, which was not discussed in ref. cplearepr . Furthermore, our observation bmp that the channel ($`\pi ^+\pi ^{}`$) ($`\pi ^+\pi ^{}`$) is automatically enhanced by three orders of magnitude, due to the relative $`(\omega /\eta _+)`$ amplitude between the “wrong” and the “right” symmetry states suggests that even a $`10^1`$ experimental effect would represent a $`10^4`$ effect in $`\omega `$. In fact, the expectation for the channel $`e^+e^{}K^0\overline{K}^0`$ in an upgraded $`DA\mathrm{\Phi }NE`$ is definitely much better, anticipating an experimental effect of the order of $`10^410^5`$.
In our analysis we saturated the bounds for the decoherence parameters $`\alpha ,\beta ,\gamma `$ obtained from CPLEAR cplear , to disentangle the $`\omega `$ effect by looking at decoherent evolution of observables in a $`\varphi `$ factory. As mentioned above, the most sensitive probe appears at first sight to be the two-charged-pion channels, as a result of enhancement factors of $`\overline{\eta }_+`$. In fact the clearest test of deviation from quantum mechanics is to look at the behaviour of the pertinent time-integrated decay rate (LABEL:twopionint) near the $`\mathrm{\Delta }\tau =0`$ regime, where the novel CPTV effects would yield a non-zero result for the integrated asymmetry.
However, in the presence of decoherence there is a reduction of the value of the asymmetry as compared to the pure $`\omega 0`$ unitary case of bmp , by shifting the value of the asymmetry at $`\mathrm{\Delta }\tau =0`$ from $`|\omega ^2|`$ to $`|\omega ^2|2\gamma /\mathrm{\Delta }\mathrm{\Gamma }`$ (in appropriate units). This reduction may be significant in the case where the $`|\omega |`$ effects are of order $`\sqrt{\gamma /\mathrm{\Delta }\mathrm{\Gamma }}`$. Of course, for a reliable order-of-magnitude estimate of the $`\omega `$-parameter, one needs to resort to detailed microscopic models of QG space-time foam, a task we hope to undertake in the near future. The fact that the interference terms for $`\mathrm{\Delta }\tau 0`$ of the asymmetry (LABEL:twopionint), which depend sinusoidally on $`\mathrm{\Delta }\tau `$, are proportional to $`|\omega |`$, and they do not depend (apart from damping factors) on decoherence parameters, allows in principle for a disentanglement of $`\omega `$ from decoherence ($`\gamma `$, …) effects. Notice also that another clear (in principle) probe of such effects would be the time-integrated observable (LABEL:3pion), associated with three-pion decay channels, near $`\mathrm{\Delta }\tau =0`$. Unfortunately, however, the current experimental sensitivity for this observable is low.
Before closing we would like to make a few comments on the contamination of the actual observable for the measurement of $`ϵ^{}`$ by decoherence peskin and intrinsic CPTV effects. The traditional observable for the measurement of $`ϵ^{}`$ is the asymmetry based on charged-pion/neutral-pion (LABEL:pipmpi0)-type observables:
$`𝒜(\pi ^+,\pi ^{};\pi ^0,\pi ^0;\delta \tau )={\displaystyle \frac{\overline{𝒫}(\pi ^+\pi ^{};\pi ^0\pi ^0;\mathrm{\Delta }\tau )\overline{𝒫}(\pi ^0\pi ^0;\pi ^+\pi ^{};\mathrm{\Delta }\tau )}{\overline{𝒫}(\pi ^+\pi ^{};\pi ^0\pi ^0;\mathrm{\Delta }\tau )+\overline{𝒫}(\pi ^0\pi ^0;\pi ^+\pi ^{};\mathrm{\Delta }\tau )}}`$
$`=3\mathrm{R}\mathrm{e}{\displaystyle \frac{ϵ^{}}{ϵ}}_13\mathrm{I}\mathrm{m}{\displaystyle \frac{ϵ^{}}{ϵ}}_2`$ (60)
In particular we find
$`_1`$ $`=`$ $`{\displaystyle \frac{1}{𝒟}}[e^{\mathrm{\Gamma }_L\mathrm{\Delta }\tau }(1+2{\displaystyle \frac{\beta }{|d||\overline{\eta }_+|}}\mathrm{sin}(\varphi _{SW}\varphi _+))`$
$`e^{\mathrm{\Gamma }_S\mathrm{\Delta }\tau }(1+2{\displaystyle \frac{\beta }{|d||\overline{\eta }_+|}}(\mathrm{sin}(\varphi _{SW}\varphi _+)|z|\mathrm{sin}(\varphi _{SW}\varphi _++\varphi _z))`$
$`+{\displaystyle \frac{|\omega |}{|\overline{\eta }_+|}}|z|\mathrm{cos}(\mathrm{\Omega }\varphi _++\varphi _z))`$
$`+e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }(2{\displaystyle \frac{\beta }{|d||\overline{\eta }_+|}}|z|\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\varphi _+\varphi _{SW}\varphi _z)`$
$`{\displaystyle \frac{|\omega |}{|\overline{\eta }_+|}}|z|\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau \mathrm{\Omega }+\varphi _+\varphi _z))],`$
$`_2`$ $`=`$ $`{\displaystyle \frac{1}{𝒟}}[e^{\mathrm{\Gamma }_L\mathrm{\Delta }\tau }\left(2{\displaystyle \frac{\beta }{|d||\overline{\eta }_+|}}\mathrm{cos}(\varphi _{SW}\varphi _+)\right)`$
$`e^{\mathrm{\Gamma }_S\mathrm{\Delta }\tau }(2{\displaystyle \frac{\beta }{|d||\overline{\eta }_+|}}(\mathrm{cos}(\varphi _{SW}\varphi _+)|z|\mathrm{cos}(\varphi _{SW}\varphi _++\varphi _z))`$
$`+{\displaystyle \frac{|\omega |}{|\overline{\eta }_+|}}|z|\mathrm{sin}(\mathrm{\Omega }\varphi _++\varphi _z))`$
$`+e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }(2\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau )2{\displaystyle \frac{\beta }{|d||\overline{\eta }_+|}}|z|\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\varphi _+\varphi _{SW}\varphi _z)`$
$`{\displaystyle \frac{|\omega |}{|\overline{\eta }_+|}}|z|\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\mathrm{\Omega }\varphi _{SW}\varphi _z))],`$
and
$`𝒟`$ $`=`$ $`e^{\mathrm{\Gamma }_L\mathrm{\Delta }\tau }\left(1+{\displaystyle \frac{\gamma }{\mathrm{\Delta }\mathrm{\Gamma }|\overline{\eta }_+|^2}}+2{\displaystyle \frac{\beta }{|d||\overline{\eta }_+|}}{\displaystyle \frac{\mathrm{sin}(2\varphi _{SW}\varphi _+)}{\mathrm{cos}(\varphi _{SW})}}\right)`$
$`+e^{\mathrm{\Gamma }_S\mathrm{\Delta }\tau }(1+{\displaystyle \frac{\gamma }{\mathrm{\Delta }\mathrm{\Gamma }|\overline{\eta }_+|^2}}{\displaystyle \frac{\mathrm{\Gamma }_L}{\mathrm{\Gamma }_S}}+2{\displaystyle \frac{\beta }{|d||\overline{\eta }_+|}}({\displaystyle \frac{\mathrm{sin}(2\varphi _{SW}\varphi _+)}{\mathrm{cos}(\varphi _{SW})}}2|z|\mathrm{sin}(\varphi _{SW}+\varphi _z\varphi _+))`$
$`+{\displaystyle \frac{|\omega |^2\overline{\mathrm{\Gamma }}}{|\overline{\eta }_+|^2\mathrm{\Gamma }_S}}2{\displaystyle \frac{|\omega |}{|\overline{\eta }_+|}}|z|\mathrm{cos}(\mathrm{\Omega }+\varphi _z\varphi _+))`$
$`e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )}(2\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau )4{\displaystyle \frac{\beta }{|d||\overline{\eta }_+|}}|z|\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\varphi _+\varphi _{SW}\varphi _z)`$
$`+{\displaystyle \frac{2|\omega |}{|\overline{\eta }_+|}}|z|\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\varphi _+\mathrm{\Omega }\varphi _z))`$
where we have defined peskin : $`|z|e^{i\varphi _z}=\frac{2\overline{\mathrm{\Gamma }}}{\mathrm{\Gamma }_S+\overline{\mathrm{\Gamma }}+i\mathrm{\Delta }m}`$.
It therefore becomes clear that, although in conventional situations, where the foam and intrinsic CPTV $`\omega `$-effects are absent, the quantities $`\mathrm{Re}(ϵ^{}/ϵ)`$ and $`\mathrm{Im}(ϵ^{}/ϵ)`$ can be extracted from a measurement of $`𝒜(\pi ^+,\pi ^{};\pi ^0,\pi ^0;\mathrm{\Delta }\tau )`$ by a simple two-parameter fit, in the presence of the quantum-gravity effects this is no longer true. The coefficients $`_i,i=1,2`$ are modified in such a case by decoherence peskin and $`\omega `$-dependent terms.
It is worthy of pointing out at this stage that the $`\omega `$-dependent terms in the expression for the above asymmetry are always accompanied by factors $`e^{\mathrm{\Gamma }_S\mathrm{\Delta }\tau }`$. Therefore, in the limit $`\mathrm{\Gamma }_S\mathrm{\Delta }\tau 1`$ such terms are suppressed. We remind the reader that, in this limit, in conventional situations, one has simply that $`𝒜(\pi ^+,\pi ^{};\pi ^0,\pi ^0;\mathrm{\Delta }\tau )3\mathrm{R}\mathrm{e}\frac{ϵ^{}}{ϵ}`$. In contrast, in the CPTV foamy situation the same limit contains, to leading order, terms proportional to *both* real and imaginary parts of $`ϵ^{}/ϵ`$, with coefficients dependent on the decoherence parameters, $`\beta ,\gamma `$, but independent of $`\omega `$. It is also worthy of noting the form of $`_2`$ in this limit: $`_2=𝒪(\beta )`$ to leading order in the small parameters. In complete positivity models benatti , therefore, which require $`\beta =0`$, such $`_2`$ terms are absent from the right-hand-side of the observable (60), and the (limiting) result for the asymmetry involves, in such a case, only the factor $`3\mathrm{R}\mathrm{e}\frac{ϵ^{}}{ϵ}\left(1|𝒪(\gamma /\mathrm{\Delta }\mathrm{\Gamma }|\overline{\eta }_+|^2)|\right)`$ to leading order.
The above issues are important to bear in mind in experimental derivations of the $`ϵ^{}`$ parameter, which in the general case require disentanglement of the (possible) decoherence and $`\omega `$ effects by means of a combined study of the $`\varphi `$-factory observables described in this article and in peskin .
## Acknowledgments
This work has been partially supported by the Grant CICYT FPA2002-00612. N.E.M. wishes to thank the University of Valencia, Department of Theoretical Physics and IFIC for the hospitality during the final stages of the collaboration. A.W.-L. also thanks the University of Valencia, Department of Theoretical Physics for the hospitality and support during the early stages of this work.
## Appendix: Complete Formulae with $`ϵ^{}`$ corrections
In this Appendix we give the $`ϵ^{}`$ corrections to the pertinent double-decay rates. The inclusion of such effects is accounted for by the $`Y_+`$ parts of the observables for the Kaon decays to two and three pions in (27).
For the two-charged-pion decay we have
$`𝒪_+=|X_+|^2\left(\begin{array}{cc}1& Y_+\\ Y_+^{}& |Y_+|^2\end{array}\right)`$
with $`X_+\pi \pi |K_2`$ $`Y_{\pi \pi }=\frac{\pi \pi |K_1}{\pi \pi |K_2}`$ and we remind the reader that the quantum mechanical amplitudes $`\eta _+=ϵ+ϵ^{}`$, $`\eta _{00}=ϵ2ϵ^{}`$, should be replaced now by their barred counterparts $`\overline{\eta }`$ that include decoherent contributions as well. In particular, above we have used the relations peskin $`R_L=\frac{\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}+|\overline{\eta }_+|^2+4\frac{\beta }{\mathrm{\Delta }\mathrm{\Gamma }}\mathrm{Im}[\overline{\eta }_+d/d^{}Y_+]`$ and $`|\overline{\eta }_+|e^{i\varphi _+}=ϵ_L^{}+Y^+`$ (the reader should recall that $`ϵ_L^{}=ϵ_L\frac{\beta }{d}`$).
The relevant double-decay time distribution reads then:
$`𝒫`$ $`(\pi ^+\pi ^{},\tau _1;\pi ^+\pi ^{},\tau _2)={\displaystyle \frac{|X_+|^4}{2}}[R_L(e^{\mathrm{\Gamma }_L\tau _2\mathrm{\Gamma }_S\tau _1}+e^{\mathrm{\Gamma }_L\tau _1\mathrm{\Gamma }_S\tau _2})`$
$`2|\overline{\eta }_+|^2\mathrm{cos}(\mathrm{\Delta }m(\tau _1\tau _2)e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )(\tau _1+\tau _2)}`$
$`+{\displaystyle \frac{4\beta |\overline{\eta }_+|}{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _1+\varphi _+\varphi _{SW})e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _1\mathrm{\Gamma }_S\tau _2}`$
$`+{\displaystyle \frac{4\beta |\overline{\eta }_+|}{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _2+\varphi _+\varphi _{SW})e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _2\mathrm{\Gamma }_S\tau _1}`$
$`+2|\omega ||\overline{\eta }_+|(\mathrm{cos}(\mathrm{\Delta }m\tau _1+\varphi _+\mathrm{\Omega })e^{\mathrm{\Gamma }_S(\tau _2)(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _1}\mathrm{cos}(\mathrm{\Delta }m\tau _2+\varphi _+\mathrm{\Omega })e^{\mathrm{\Gamma }_S\tau _1(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _2})`$
$`+(|\omega |^2{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}8{\displaystyle \frac{\beta }{\mathrm{\Delta }\mathrm{\Gamma }}}\mathrm{Im}[\overline{\eta }_+Y_+])e^{\mathrm{\Gamma }_S(\tau _1+\tau _2)}]`$
The time Integrated distribution reads
$`\overline{𝒫}`$ $`(\pi ^+\pi ^{};\pi ^+\pi ^{};\mathrm{\Delta }\tau )={\displaystyle \frac{|X_+|^4}{2}}[R_L{\displaystyle \frac{e^{\mathrm{\Delta }\tau \mathrm{\Gamma }_L}+e^{\mathrm{\Delta }\tau \mathrm{\Gamma }_S}}{\mathrm{\Gamma }_L+\mathrm{\Gamma }_S}}`$
$`|\overline{\eta }_+|^2\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau ){\displaystyle \frac{e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }}{(\overline{\mathrm{\Gamma }}+\alpha \gamma )}}`$
$`+{\displaystyle \frac{1}{\mathrm{\Delta }m^2+(\overline{\mathrm{\Gamma }}+\alpha \gamma )^2+2(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Gamma }_S+\mathrm{\Gamma }_S^2}}`$
$`\times [{\displaystyle \frac{4\beta |\overline{\eta }_+|}{|d|}}((\mathrm{\Delta }m\mathrm{cos}(\varphi _+\varphi _{SW})+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{sin}(\varphi _+\varphi _{SW}))e^{\mathrm{\Delta }\tau \mathrm{\Gamma }_S}`$
$`+e^{\mathrm{\Delta }\tau (\overline{\mathrm{\Gamma }}+\alpha \gamma )}(\mathrm{\Delta }m\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\varphi _+\varphi _{SW})`$
$`+(\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\varphi _+\varphi _{SW})))`$
$`+2|\omega ||\overline{\eta }_+|e^{\mathrm{\Delta }\tau \mathrm{\Gamma }_S}((\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{cos}(\varphi _+\mathrm{\Omega })\mathrm{\Delta }m\mathrm{sin}(\varphi _+\mathrm{\Omega }))`$
$`2|\omega ||\overline{\eta }_+|e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )\mathrm{\Delta }\tau }((\overline{\mathrm{\Gamma }}+\alpha \gamma +\mathrm{\Gamma }_S)\mathrm{cos}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\varphi _+\mathrm{\Omega })`$
$`\mathrm{\Delta }m\mathrm{sin}(\mathrm{\Delta }m\mathrm{\Delta }\tau +\varphi _+\mathrm{\Omega }))+(|\omega |^2{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}8{\displaystyle \frac{\beta }{\mathrm{\Delta }\mathrm{\Gamma }}}\mathrm{Im}[\overline{\eta }_+Y_+]){\displaystyle \frac{e^{\mathrm{\Gamma }_S\mathrm{\Delta }\tau }}{2\mathrm{\Gamma }_S}}]]`$
For the neutral-pion decay channel we use the observable:
$`𝒪_{00}=|X_{00}|^2\left(\begin{array}{cc}1& Y_{00}\\ Y_{00}^{}& |Y_{00}|^2\end{array}\right)`$
and the associated double-decay rate involving charged- and neutral-pion decay channels has the form:
$`𝒫`$ $`(\pi ^+\pi ^{},\tau _1;\pi ^0\pi ^0,\tau _2)={\displaystyle \frac{|X_+|^2|X_{00}|^2}{2}}[R_L^{00}e^{\mathrm{\Gamma }_L\tau _2\mathrm{\Gamma }_S\tau _1}+R_L^+e^{\mathrm{\Gamma }_L\tau _1\mathrm{\Gamma }_S\tau _2}`$
$`2|\overline{\eta }_+||\overline{\eta }_{00}|\mathrm{cos}(\mathrm{\Delta }m(\tau _1\tau _2)+\varphi _+\varphi _{00})e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )(\tau _1+\tau _2)}`$
$`+|\overline{\eta }_+|e^{\mathrm{\Gamma }_S\tau _2(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _1}\left(2|\omega |\mathrm{cos}(\mathrm{\Delta }m\tau _1+\varphi _+\mathrm{\Omega })+{\displaystyle \frac{4\beta }{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _1+\varphi _+\varphi _{SW})\right)`$
$`+|\overline{\eta }_{00}|e^{\mathrm{\Gamma }_S\tau _1(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _2}\left(2|\omega |\mathrm{cos}(\mathrm{\Delta }m\tau _2+\varphi _{00}\mathrm{\Omega })+{\displaystyle \frac{4\beta }{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _2+\varphi _{00}\varphi _{SW})\right)`$
$`+(|\omega |^2{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}4{\displaystyle \frac{\beta }{\mathrm{\Delta }\mathrm{\Gamma }}}\mathrm{Im}[\overline{\eta }_+Y_++\overline{\eta }_{00}Y_{00}])e^{\mathrm{\Gamma }_S(\tau _1+\tau _2)}].`$
The double-decay time distribution involving charged-pion and dilepton channels is given by:
$`𝒫(\pi ^\pm ,\tau _1`$ ; $`l^\pm ,\tau _2)={\displaystyle \frac{|X_+|^2|a|^2}{4}}[(1\pm \delta _L+{\displaystyle \frac{\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}})e^{\mathrm{\Gamma }_L\tau _2\mathrm{\Gamma }_S\tau _1}+R_Le^{\mathrm{\Gamma }_L\tau _1\mathrm{\Gamma }_S\tau _2}`$
$`2|\overline{\eta }_+|\mathrm{cos}(\mathrm{\Delta }m(\tau _1\tau _2)+\varphi _+)e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )(\tau _1+\tau _2)}`$
$`\pm \left({\displaystyle \frac{4\beta }{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _2\varphi _{SW})2|\omega |\mathrm{cos}(\mathrm{\Delta }m\tau _2\mathrm{\Omega })\right)e^{\mathrm{\Gamma }_S\tau _1(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _2}`$
$`+(|\omega |^2{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}{\displaystyle \frac{8\beta }{\mathrm{\Delta }\mathrm{\Gamma }}}\mathrm{Im}[\overline{\eta }_+])e^{\mathrm{\Gamma }_S(\tau _1+\tau _2)}]`$
Finally for the three-pion decay channel the relevant observable is:
$`𝒪_{3\pi }=|X_{3\pi }|^2\left(\begin{array}{cc}|Y_{3\pi }|^2& Y_{3\pi }^{}\\ Y_{3\pi }& 1\end{array}\right)`$
with $`3\pi |K_2`$ $`Y_{3\pi }=\frac{3\pi |K_1}{3\pi |K_2}`$
In the following we use $`R_S=\frac{\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}+|\overline{\eta }_{3\pi }|^24\frac{\beta }{\mathrm{\Delta }\mathrm{\Gamma }}\mathrm{Im}[\overline{\eta }_{3\pi }d/d^{}Y_{3\pi }]`$ where $`|\overline{\eta }_{3\pi }|e^{i\varphi _{3\pi }}=ϵ_S^++Y_{3\pi }`$.
The pertinent three-pion double-decay time distribution is:
$`𝒫(3\pi ,\tau _1`$ ; $`3\pi ,\tau _2)={\displaystyle \frac{|X_{3\pi }|^4}{2}}[R_Se^{\mathrm{\Gamma }_L\tau _2\mathrm{\Gamma }_S\tau _1}+R_Se^{\mathrm{\Gamma }_L\tau _1\mathrm{\Gamma }_S\tau _2}`$
$`2|\overline{\eta }_{3\pi }|^2\mathrm{cos}(\mathrm{\Delta }m(\tau _1\tau _2))e^{(\overline{\mathrm{\Gamma }}+\alpha \gamma )(\tau _1+\tau _2)}`$
$`+|\overline{\eta }_{3\pi }|e^{\mathrm{\Gamma }_L\tau _1(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _2}\left(2|\omega |\mathrm{cos}(\mathrm{\Delta }m\tau _2\varphi _{3\pi }+\mathrm{\Omega })+{\displaystyle \frac{4\beta }{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _2\varphi _{3\pi }+\varphi _{SW})\right)`$
$`+|\overline{\eta }_{3\pi }|e^{\mathrm{\Gamma }_L\tau _2(\overline{\mathrm{\Gamma }}+\alpha \gamma )\tau _1}(2|\omega |\mathrm{cos}(\mathrm{\Delta }m\tau _1\varphi _{3\pi }+\mathrm{\Omega })+{\displaystyle \frac{4\beta }{|d|}}\mathrm{sin}(\mathrm{\Delta }m\tau _1\varphi _{3\pi }+\varphi _{SW}))],`$
$`+\left(|\omega |^2+{\displaystyle \frac{2\gamma }{\mathrm{\Delta }\mathrm{\Gamma }}}+{\displaystyle \frac{8\beta }{\mathrm{\Delta }\mathrm{\Gamma }}}\mathrm{Im}[\overline{\eta }_{3\pi }Y_{3\pi }]\right)e^{\mathrm{\Gamma }_L(\tau _1+\tau _2)}`$
This completes our analysis. We observe that the inclusion of the $`ϵ^{}`$ corrections does not affect the functional form of the decoherence and CPTV effects.
|
warning/0506/astro-ph0506690.html
|
ar5iv
|
text
|
# The VVDS-VLA Deep Field II.
## 1 Introduction
Deep 1.4-GHz counts show an upturn below a few millijansky (mJy), corresponding to a rapid increase in the number of faint sources. Photometric and spectroscopic studies suggest that the faint excess at 1.4 GHz is composed predominantly of star-forming galaxies, with a contribution also from low-power AGNs and early type galaxies. However, despite many dedicated efforts (see for example Benn et al. 1993, Hammer et al. 1995, Gruppioni et al. 1999, Richards et al. 1999, Prandoni et al. 2001) the relative fraction of the various populations responsible of the sub-mJy radio counts (AGN, starburst, late and early type galaxies), are not well established yet. In fact, the photometric and spectroscopic work needed in the optical identification process is very demanding in terms of telescope time, since a significant fraction of faint radio sources have also very faint optical counterparts. It is therefore clear that in order to investigate the nature and evolution of the sub-mJy population it is absolutely necessary to couple deep radio and optical (both imaging and spectroscopic) observations over a reasonably large area of the sky.
The VVDS-VLA Deep survey is one of the best available surveys to investigate the nature of the sub-mJy population. A deep radio survey has been obtained in 1 deg<sup>2</sup> with the VLA down to a 5$`\sigma `$ flux limit of $``$80 $`\mu `$Jy (Bondi et al. 2003). A deep multi-color BVRI photometric survey of the whole area is available (Le Fèvre et al., 2004a, McCracken et al. 2003), together with U band (Radovich et al. 2004) and J and K band (Iovino et al. 2005) data. Furthermore, a deep spectroscopic survey is being performed with the VIMOS spectrograph at the VLT (Le Fèvre et al. 2004b).
In this paper we present the optical identification of the VIMOS radio sources obtained with the photometric data in the $`U,B,V,R,I`$ and $`K`$ bands. In Sect. 2 we give a description of the available radio, optical and near infrared data, while the description of the technique used for the optical identification of the radio sources is presented in Sect. 3. Finally, Sect. 4 is a discussion of our results, while Sect. 5 summarizes our conclusion.
## 2 The available data
### 2.1 Radio data
The radio data were obtained with the Very Large Array (VLA) in B configuration. A 1 square degree mosaic map with an approximately uniform noise of 17.5$`\mu `$Jy (1$`\sigma `$) and with a 6$`\times `$6 arcsec FWHM gaussian resolution beam has been obtained. This map (centered at RA(J2000)=02:26:00 DEC(J2000)=$``$04:30:00) has been used to extract a complete catalogue of 1054 radio sources, 19 of which are considered as multiple, i.e. fitted with at least two separate component. A detailed description of the radio observations, data reduction, sources extraction and radio source counts is reported in Bondi et al. (2003).
### 2.2 The optical and near infrared data in the $`U,B,V,R,I`$ and $`K`$ bands
Almost the whole square degree of the VVDS-VLA field has been observed in the $`B,V,R`$ and $`I`$ bands with the CFH12K wide-field mosaic camera during the CFH12K-VIMOS deep imaging survey (Le Fèvre et al. 2004a). These observations reach limiting magnitudes (50% completeness for point sources) of $`B_{AB}`$26.5, $`V_{AB}`$26.2, $`R_{AB}`$25.9 and $`I_{AB}`$25.0. A detailed description of these optical data is given in McCracken et al. (2003). Moreover, a $`U`$ band survey has been carried out with the wide field imaging (WFI) mosaic camera mounted on the ESO MPI 2.2 meter telescope at La Silla, Chile. The total effective area covered so far by the $`U`$ band survey is $``$0.71 deg<sup>2</sup>. The limiting magnitude of the catalogue obtained with these observations is 25.4 in $`U_{AB}`$ (50% completeness). A detailed description of the VIMOS U band imaging survey is reported in Radovich et al. 2004. Finally, a fraction of the VVDS-VLA field has been observed in the $`J`$ and $`K`$ bands with the SOFI instrument mounted on the ESO NTT telescope. The total area covered by the $`J`$ and $`K`$ bands survey is $``$ 165 arcmin<sup>2</sup>, down to limiting magnitudes of $`J_{AB}`$24.2 and $`K_{AB}`$23.9 (50% completeness for point sources). A detailed description of the VIMOS J and K band imaging survey is reported in Iovino et al. 2005.
In order to obtain a $`B_{AB},V_{AB},R_{AB},I_{AB}`$ catalogue, all the images were combined to obtain a unique detection image using the $`\chi ^2`$ technique (Szalay et al. 1999). Subsequently, the photometry in each bandpass has been performed at the position defined in the $`\chi ^2`$ image. The primary advantage in using this technique is to simplify the generation of multi-band catalogues and to reduce the number of spurious detections. For a detailed description of the $`B_{AB},V_{AB},R_{AB},I_{AB}`$ catalogue see McCracken et al. (2003). Subsequently, the $`\chi ^2`$ image has been updated with the $`U_{AB}`$ , $`J_{AB}`$ and $`K_{AB}`$ data (in the areas covered in these bands) using the same technique described by McCracken et al. (2003).
## 3 Optical identification
### 3.1 Radio-optical off-set
The relative off-set between the radio and optical catalogue has been estimated using the radio position of the 160 unresolved radio sources with a peak flux density greater than 0.17 mJy ($`i.e.`$ detected with peak flux $``$ 10$`\sigma `$) identified with point like optical counterparts. Their $`\mathrm{\Delta }`$RA and $`\mathrm{\Delta }`$DEC offset are shown in Fig. 1. The mean offsets, with their standard errors ($`\sigma /\sqrt{N}`$), are $`<\mathrm{\Delta }`$RA$`>`$ = +0.13$`\pm `$0.03<sup>′′</sup> and $`<\mathrm{\Delta }`$DEC$`>`$ = $``$0.32$`\pm `$0.03<sup>′′</sup>. These offsets should be removed (subtracting 0.13<sup>′′</sup> in RA and adding 0.32<sup>′′</sup> in DEC at the radio position) to obtain the radio position in the same reference frame as the optical CCD.
### 3.2 Optical identification of the radio sources
Of the 1054 radio sources in the complete radio sample, 57 are outside the presently available catalogues based on our CCD data. The optical identification of these 57 sources with the available public catalogues is described in the next Section. Moreover, 24 radio sources are within the masked regions in the $`UBVRIK`$ catalogue, since they are spatially coincident with bright stars or their diffraction spikes. The total number of radio sources for which optical data are available in the VIMOS CCD catalogue is therefore 973.
For the optical identification of these 973 radio sources, we used the likelihood ratio technique, first used in this context by Richter (1975) and in modified form by de Ruiter, Willis & Arp (1977), Prestage & Peacock (1983), Sutherland & Saunders (1992) and Ciliegi et al. (2003). The mean off-set between the radio and optical positions estimated in the previous section has been removed from the radio positions to compute the positional offset.
The likelihood ratio LR is the ratio between the probability that a given source at the observed position and with the measured magnitude is the true optical counterpart, and the probability that the same source is a chance background object :
$$LR=\frac{q(m)f(r)}{n(m)}$$
(1)
where q(m) is the expected distribution as a function of magnitude of the optical counterparts, f(r) is the probability distribution function of the positional errors, while n(m) is the surface density of background objects with magnitude m (see Ciliegi et al. 2003 for a detailed discussion on the procedure to calculate q(m), f(r) and n(m)). For each source we adopted an elliptical Gaussian distribution for the positional errors with the errors in RA and DEC on the radio position reported in the radio catalogue (Bondi et al. 2003) and assuming a value of 0.3 arcsec as optical position uncertainty (McCracken et al. 2003).
The presence or absence of more than one optical candidates for the same radio source provides additional information to that contained in LR. The reliability $`Rel_j`$ for object j being the correct identification is (Sutherland & Saunders, 1992) :
$$Rel_j=\frac{(LR)_j}{\mathrm{\Sigma }_i(LR)_i+(1Q)}$$
(2)
where the sum is over the set of all candidates for this particular source and Q is the probability that the optical counterpart of the source is brighter than the magnitude limit of the optical catalogue ($`Q=^{m_{lim}}q(m)𝑑m`$).
The adopted value for $`Q`$ is 0.65. This value has been estimated by the comparison between the expected number of identifications (658) derived from the integral of the $`q(m)`$ distribution and the number of the radio sources (973) that we used in the Likelihood Ratio technique (see Ciliegi et al. 2003 for more details). However, to check how this assumption could affect our results, we repeated the likelihood ratio analysis using different values of $`Q`$ in the range 0.6–0.8. No substantial difference in the final number of identifications and in the associated reliability has been found.
Once q(m), f(r) and n(m) were obtained, we computed the $`LR`$ value for all the optical sources within a distance of 5 arcsec from the radio position. Having determined the $`LR`$ for all the optical candidates, one has to choose the best threshold value for $`LR`$ ($`L_{\mathrm{th}}`$) to discriminate between spurious and real identifications. As $`LR`$ threshold we adopted $`L_{\mathrm{th}}`$=0.35. With this value, according to Eq. (2) and considering that our estimate for Q is 0.65, all the optical counterparts of radio sources with only one identification (the majority in our sample) and $`LR>LR_{\mathrm{th}}`$ have a reliability greater than 0.5. This choice also approximately maximizes the sum of sample reliability and completeness. With this threshold value we find 718 radio sources with a likely identification, 14 of which have two optical candidates with $`L_{\mathrm{th}}>`$0.35 for a total of 732 optical candidates with $`L_{\mathrm{th}}>`$0.35. The maximum distance between the radio and optical position is 2.98 arcsec. The number of expected real identifications (obtained by summing the reliability of all the objects with $`L_{\mathrm{th}}>`$0.35) is about 683, $`i.e`$ we expect that about 35 of the 718 proposed radio-optical associations may be spurious positional coincidences. In the 14 cases in which more than one optical object with $`LR>LR_{\mathrm{th}}`$ has been found associated to the same radio source, we assumed the object with the highest Likelihood Ratio value as the counterpart of this radio source. Among the 255 unidentified radio sources, 17 are empty fields ($`i.e.`$ they have no optical source within 5 arcsec from their position), while the other 238 sources have at least one optical source within 5 arcsec, but all with $`LR<LR_{\mathrm{th}}`$.
### 3.3 Optical identification of the radio sources outside the VIMOS CCD catalogue
As explained above, 57 radio sources lie outside the currently available VIMOS CCD catalogue. For the optical identification of these sources we used the Guide Sky Catalogue 2.2 <sup>1</sup><sup>1</sup>1http://www-gsss.stsci.edu/gsc/gsc2/GSC2home.htm, which provides a two bandpass (J$``$B and F$``$R) catalogue with a magnitude limit of J$``$ 19.5 mag and F$``$ 18.5 mag. Using a maximum distance of 3 arcsec between the radio and optical positions (2.98 arcsec is the maximum radio-optical offset for the radio sources identified with the likelihood ratio technique, see previous section), we identified 7 radio sources, only one of which has a radio-optical offset greater than 2 arcsec. The optical properties of these 7 radio sources are given in Table 1. For each source we report the name, the 20 cm total radio flux, the distance between the optical and radio position, the J and F magnitude with relative errors and the optical classification $`𝚌`$ as reported in the GSC2.2 catalogue (0 for stellar and 3 for non stellar). Given the small number of sources (7 in comparison to the 718 identified within the CCD area) and the difference in the magnitude systems (J and F in the GSC2 catalogue vs. B<sub>AB</sub> and R<sub>AB</sub> in the CCD catalogue), these 7 sources have not been considered in the analysis and discussion reported in the next sections and are not included in the identification summary reported in Table 2.
### 3.4 Near-Infrared identification
Only 65 of the 1054 radio sources lie within the $``$ 165 arcmin<sup>2</sup> covered by our deep K band survey: 21 are unidentified both in the optical and K bands, 1 has a reliable optical counterpart but is unidentified in the K band, while 43 have a reliable counterpart both in the optical and K bands. To fully explore the near-infrared properties of our radio sample, we searched also the Two Micron All Sky Survey (2MASS <sup>2</sup><sup>2</sup>2http://www.ipac.caltech.edu/2mass/, Cutri+ 2003 ) database for near-IR ($`J,H`$ and $`K`$ bands) counterparts of our 1054 radio sources. The 3$`\sigma `$ limits of the 2MASS survey are 17.1, 16.4 and 15.3 mag respectively in the three bandpasses $`J,H`$ and $`K`$. The image pixel scale of the 2MASS detectors is 2.0<sup>′′</sup>, and the positional uncertainties are $``$0.5<sup>′′</sup>. Using a maximum distance of 3<sup>′′</sup> between the radio and 2MASS positions, we found 105 reliable matches, with 105 detections in $`J`$, 104 in $`H`$, and 93 in $`K`$, including the 7 sources outside the optical CCD area reported in Table 1. Owing to the relatively shallow flux limits of the 2MASS survey, the surface density of background sources is low enough that with a maximum off-set of 3<sup>′′</sup> between the radio and 2MASS positions, chance associations with radio positions are very unlikely. All the 2MASS sources have $`I_{AB}`$19.0. The $`J,H`$ and $`K`$ magnitudes have been converted to the AB system using J<sub>AB</sub> = J + 0.90, H<sub>AB</sub> = H + 1.37 and K<sub>AB</sub> = K + 1.84.
### 3.5 Identification summary
A summary of the optical identifications is reported in Table 2. For each band we report the number of sources that lie outside the available CCD area, the number of sources within the masked regions ($`i.e.`$ sources coincident with bright stars or with their diffraction spikes for which a reliable identification is impossible), the number of sources for which reliable data are available (1054 $``$ outside CCD $``$ within masked regions) and the number of sources with a reliable identification. Finally in the last line we report the number of reliable identifications with an $`I_{AB}`$ magnitude brighter than 24, $`i.e.`$ corresponding to the sources that are potential targets for the VIMOS spectroscopic survey (Le Fèvre et al 2004b). The 7 radio sources identified in the GSC 2.2 catalogue are not included in Table 2.
In Figure 2 we report the radio peak flux distributions for the whole radio sample, for the 718 radio sources with a reliable optical identification and for the 105 radio sources with a near infrared counterpart in the 2MASS survey. As clearly shown in the figure, while the peak flux distribution of radio sources with a 2MASS counterpart is strongly biased towards high radio flux (as expected due to the bright limits of the 2MASS survey), the flux peak distribution of the 718 radio sources with an optical identification does not show any statistically significant difference with respect to the distribution of the whole radio sample. The final catalogue of all 718 identified radio sources is available on the web at http://virmos.bo.astro.it/radio/catalogue.html. A sample of the catalogue is shown in Table 3.
## 4 DISCUSSION
### 4.1 Optical magnitude distributions
In absence of spectroscopic data, the magnitude and color distributions of the optical counterparts can be used to derive some informations on the nature of faint radio sources. In Figure 3 we show the magnitude distributions in the $`U_{AB},B_{AB},R_{AB}`$, and $`I_{AB}`$ bands of the optical counterparts of the radio sources as filled histograms. The empty histograms show the magnitude distributions of the whole optical data set.
This figure clearly shows that the magnitude distributions of the optical counterparts of the radio sources are significantly flatter than those of the global optical catalogues, reaching a maximum at magnitudes well above our optical limiting magnitudes. This turnover in the magnitude distribution of faint radio sources was initially hinted in the Leiden-Barkeley Deep Survey (LBDS) radio sample (Windhorst et al. 1984) and more recently confirmed in the LBDS Hercules subsample (Waddington et al. 2000) and in the identification of the radio sources in the Hubble Deep Field region (Richards et al. 1999).
In Figure 4 we show the 20 cm radio flux versus the $`I_{AB}`$ band magnitude for all the radio sources with an optical identification. Superimposed are the lines corresponding to constant values for the observed radio-to-optical ratios $`R`$ defined as $`R`$ = S $`\times `$ 10<sup>0.4(mag-12.5)</sup>, where S and $`mag`$ are the radio flux in mJy and the apparent magnitude of the optical counterparts respectively. In Figure 5 we show the $`I_{AB}`$ and $`(VI)_{AB}`$ distributions in different radio flux bins. The shaded histograms in the $`I_{AB}`$ magnitude distributions show the number of unidentified radio sources, arbitrarily plotted in four equal bins between $`I_{AB}`$=25.0 and $`I_{AB}`$=27.0.
In Table 4, for each radio flux bin, we report the mean, median and standard deviations for the $`I_{AB}^{med}`$ and $`(VI)_{AB}^{med}`$ color distributions of all the identified radio sources, the number of radio sources with an optical identification, the total number of radio sources (excluding the sources outside the CCD area and the sources in the masked regions, see Section 3), the percentage of radio sources with a reliable optical identifications and the median $`I_{AB}^{med+unid}`$ magnitude calculated considering also the unidentified sources, all assumed fainter than $`I_{AB}`$=25.0.
Figure 4 shows an extremely large scatter between radio flux and magnitude, while the analysis of the optical properties in different radio flux bins (see Table 4) shows a mean (and median) optical magnitude that becomes fainter as fainter radio flux bins are considered, although this trend is not statistically significant due the large spread of the distributions (see column 4 (standard deviations) in Table 4)
However, while about 35% of the radio sources are optically unidentified in the first radio flux bin (see Figure 5 and Table 4), the percentage of unidentified sources decreases to about 25% in the faintest two radio bins. Because of this decrease of unidentified sources, although within the Poisson errors, the median $`I_{AB}`$ magnitude for the total sample of radio sources, i.e. including also the unidentified ones, is actually brighter of $``$0.6 mag in the faintest radio bin than in the bin with higher radio flux (see last column in Table 4). This result shows that the faintest radio sources are not in general the faintest sources at optical wavelengths and would suggest that most of the faintest radio sources are likely to be associated to relatively low redshift star forming objects with a low radio luminosity, rather than radio-powerful, AGN type objects at high redshift. Their median photometric redshift z<sub>phot</sub> (see next Section for more details on z<sub>phot</sub>) is 0.67, with $``$ 90% of the sources in the photometric redshift range 0.1$``$z$`{}_{phot}{}^{}`$1.5.
This result is consistent with the expectations from previous work in the optical identification of the radio sources. Many authors (Windhorst et al. 1995, Richards et al. 1998, Richards et al. 1999, Roche et al. 2002) have shown, in fact, that the majority of the optical identifications of the $`\mu `$Jy radio sources are with luminous ($`L>L_{}`$) galaxies at relatively modest redshifts (0$``$z$``$1), many of which with evidence for recent star formation.
Finally, the analysis of the optical color $`(VI)_{AB}`$ in different radio flux bins (see Figure 5 and Table 4 ) shows that there is a small reddening of the median color as fainter radio fluxes are considered.
### 4.2 Unidentified Radio Sources
As described in section 3, we have 255 radio sources (26% of the radio sample) without optical counterpart in our optical and near infrared images. The presence of a significant fraction ($``$25-30%) of radio sources with an optical magnitude fainter than $``$25 was already noted by several authors in radio surveys of similar depth. In the identification of the microjansky radio sources in the HDF, only 84 (out of 111, $``$76%) sources have been identified to $`I`$=25 mag, with the bulk of the sample identified with relatively bright ($`I`$22) galaxies (Richards et al. 1999), while in the Phoenix Deep Survey (PDS) about 79% of the sources (659 out of 839) have been identified using optical $`UBVRI`$ images to $`R`$24.5 (Sullivan et al. 2004)
The 255 unidentified VIMOS radio sources have a median radio flux of 0.15 mJy, equal to that of identified sources (see also Figure 5 and Table 4 for their radio flux distribution). Given the very faint optical counterpart, these unidentified radio sources might contain a significant fraction of obscured and/or high redshift galaxies. Deep surveys performed at wave-bands free from dust absorption ($`i.e`$ far-infrared and X-ray) may provide the most powerful tool to identify these objects. The VIRMOS VVDS-02h field is a selected target for far-infrared and X-ray observations with the Spitzer and XMM satellites in the framework of the Spitzer Legacy Programme SWIRE (Lonsdale et al. 2003) and the XMM Large Scale Structure Survey (Pierre et al. 2004) respectively.
### 4.3 Photometric Redshifts
The optical data in the $`B_{AB}`$, $`V_{AB}`$ $`R_{AB}`$ and $`I_{AB}`$ bands, plus $`U_{AB}`$, $`J_{AB}`$ and $`K_{AB}`$ data (when available) allow us to estimate photometric redshifts for all the 718 radio sources with an optical counterpart. A detailed description and discussion of the method used to estimate the photometric redshift in the VIMOS survey is reported in Bolzonella et al. in preparation.
Photometric redshifts have been computed with two codes developed by the authors: Hyperz, by Bolzonella et al. (2000) <sup>3</sup><sup>3</sup>3http://webast.ast.obs-mip.fr/hyperz/, using the Bruzual & Charlot (2003) library, and the code Le Phare by Arnouts & Ilbert <sup>4</sup><sup>4</sup>4http://www.lam.oamp.fr/arnouts/LE\_PHARE.html using 72 CWW (Coleman et al. 1980) extended templates. Both of them have been modified allowing a training of photometric redshifts using the spectroscopic data.
The first step of the method consists in training the photometry: observations are affected by possible uncertainties on photometric calibration and zeropoints. At the same time, the template SEDs may be not fully representative of the observed galaxy population and it may be difficult to reproduce precisely the response functions of filters. Therefore we calibrated photometry and templates with the spectroscopic sample in order to obtain a good agreement between the observed colour-redshift relation and the one derived from templates.
Moreover, to avoid spurious solutions at high redshift, frequently found when the U magnitude is not available, we imposed a prior in the redshift distribution. To this aim we used the formalism described in Benítez (2000), without considering the mix of spectrophotometric types. We applied a prior roughly reproducing the $`N(z)`$ of the spectroscopic sample, but with a non null probability of being at high redshift. By imposing this prior we do not significantly affect the redshift distribution and we considerably improve the agreement between photometric and spectroscopic redshifts, in particular when only the optical magnitudes are available. The values of photometric redshifts used in this paper have been obtained with the second of the two codes, Le Phare, although the two codes produced very similar results.
So far, only a small fraction (54/718) of the optical counterparts of the radio sources has a spectroscopically measured redshift. For these sources the spectroscopic and photometric redshifts are in good agreement (90% of the objects have $`\mathrm{\Delta }z/(1+z_{\mathrm{spec}})<0.07`$). The analysis of the sub-sample of spectroscopically identified radio-sources is in progress and will be presented elsewere.
Figure 6 shows the photometric redshift histogram for all the radio sources with a reliable optical counterpart. About 80% of the sources are estimated to be at z $``$ 1, with a small high redshift tail extending up to z $``$ 3, although among the 157 radio sources with z$`{}_{phot}{}^{}>`$1.0 we have the $`J`$ and $`K`$ bands data (and then a more reliable z<sub>phot</sub>) for only 11 sources. This redshift distribution is in good agreement with that obtained by Sullivan et al. 2004 in their analysis of the optical and near infrared counterparts of the radio sources in the Phoenix Deep Survey.
### 4.4 Color-color diagrams and comparison with the whole optical data set
In this section we compare the optical colors properties of the radio sources with those of the whole optical data set. A detailed discussion of the color properties of the whole optical data set with a comparison to those from other deep surveys is reported in McCracken et al. (2003).
In Figure 7 we show the $`(BV)_{AB}`$ and $`(VI)_{AB}`$ colors for the whole data set (small dots) and for the optical counterparts of the radio sources (open squares). The sources classified as point-like in the optical band ($`I_{AB}<`$21.5) are plotted in the top-left panel, while the sources classified as extended are plotted in the other three panels in three different magnitude ranges (18$`<I_{AB}<`$20, 20$`<I_{AB}<`$22 and 22$`<I_{AB}<`$24, respectively). Solid, dashed and dashed dot-dot-dot lines show the path in redshift of early-type, Sab and Sbc galaxies. The evolutionary tracks were computed using the “2000” revision of the GISSEL libraries (Bruzual & Charlot, 1993). The median value of $`(BV)_{AB}`$, $`(VI)_{AB}`$, z<sub>phot</sub> and $`I_{AB}`$ for the sources classified as extended are reported in Table 5, where the same values are given also for the whole optical sample in the same magnitude ranges. The errors on the median values reported in Table 5 have been calculated using 1.2533$`\sigma `$/$`\sqrt{N}`$, where $`\sigma `$ is the standard deviation of the distribution and N is the number of objects (Akin & Colton 1970).
Moreover, for each data set ($`(BV)_{AB}`$, $`(VI)_{AB}`$, z<sub>phot</sub> and $`I_{AB}`$) in each magnitude bin, the Kolmogorov-Smirnov test (KS) has been used to test the hypothesis that the two distributions (that from the whole optical sample and that from the radio sources) are drawn from the same distribution. The results of the KS test are reported in Table 5 (last row in each magnitude bin). In particular, for each KS test performed, we report the significance level of the KS statistic. Small values show that the two distributions are significantly different.
The color-color plot for point-like sources (top-left panel of Figure 7) shows that, as expected, the majority of the radio sources with a point-like optical counterpart have different colors with respect to the bulk of the whole optical data set. In fact, while the majority of the point-like sources are expected to be stars, those associated with radio sources are, with very high probability, active galactic nuclei (AGN) or actively star forming compact galaxies.
The top-right panel of Figure 7 shows the bright magnitude slice at 18$`<I_{AB}<`$20 for extended sources. As shown in the Figure and in Table 5 the color properties of the radio sources in this bright magnitude slice are not significantly different from those of the whole optical data set. The bulk of the optical and radio sources are consistent with model tracks of a galaxy population around z$``$ 0.2-0.3 but the median photometric redshift of the radio sources is higher than that of the whole optical sample (see Table 5).
The lower-left panel of Figure 7 shows the intermediate magnitude slice at 20$`<I_{AB}<`$22. As already noted by McCracken et al. (2003), the whole galaxy population (predominantly late-type galaxies) occupies two distinct loci, and in this color-color space there is a reasonably well defined separation between high (z$`>`$0.4) and low (z$`<`$0.4) redshift galaxies. However, as clearly shown in the figure, the radio sources essentially populate only the locus of the high redshift (z$``$0.4-0.5) galaxies. This is confirmed by the photometric redshift analysis, with a median redshift of the whole data sample (0.56) significantly lower than the median redshift of the radio sample (0.82; see Table 5). The median $`(VI)_{AB}`$ color of the radio sources ($`(VI)_{AB}^{med}`$=1.77, see Table 5) is also significantly redder than that of the whole optical data set ($`(VI)_{AB}^{med}`$=1.16). The KS test shows that the $`(VI)_{AB}`$, z<sub>phot</sub> and $`I_{AB}`$ distributions of the whole optical sample are very significantly different from that of the radio sources (see Table 5).
Finally, in the fainter magnitude slice (22$`<I_{AB}<`$24, lower-right panel of Figure 7) the colors are bluer than in the brighter magnitude slice both for the radio sources and the whole optical data set. Even if less evident from the figure, also in this magnitude slice there is a statistically significant difference between the color, photometric redshift and magnitude distributions of the radio sources and the radio quiet galaxies (see results of KS test in Table 5).
The fact that, in each magnitude bin, the median photometric redshift of the radio sample is higher than that of the whole optical sample shows that radio detection is preferentially selecting galaxies with higher intrinsic optical luminosity. This conclusion is strengthened by the fact that, because of the different magnitude distributions (see Figure 3), the median magnitude $`I_{AB}^{med}`$ of the radio sample is brighter than that of the optical sample in each magnitude bin (see column 5 in Table 5).
### 4.5 Color - redshift diagram for the radio sample
In Figure 8 we show the optical color $`(BI)_{AB}`$ versus the photometric redshift for all the radio sources with an optical counterpart, together with predicted loci for early type galaxies (solid line), Sa galaxies (dashed line) and Sbc galaxies (dashed dot-dot-dot line). Sources above the Sa track are expected to be mainly early type galaxies and most of them (222/263) have z$`{}_{phot}{}^{}`$0.5, while the sources below the Sa track are expected to be mainly late-type, star-forming galaxies and they more uniformly occupy the entire redshift range, including the low redshift one.
In order to study, at least statistically, the properties of the two classes of objects, we have adopted in the following the Sa track as separation between early and late type galaxies. We are fully aware that such a classification is only a first approximation, which does not work on an object-by-object basis, and can be significantly uncertain especially at redshift greater than $``$1. In fact, above z$``$ 1 the photometric redshifts are more uncertain. This, coupled with the rapidly decreasing expected $`(BI)_{AB}`$ color of the Sa track up to z$``$2, makes the separation between early and late type galaxies only indicative in this redshift range. A somewhat similar separation has been already used by Ciliegi et al. 2003 in their analysis of the 6cm radio sources in the Lockman Hole. They used the color V-K=5.2 to separate the high redshift (z$``$0.5) early-type galaxies from the late-type, star forming galaxies at all redshifts. Due to the lack of the K-band covering for the whole VIMOS radio sample, we cannot use the V-K color to separate the two classes. However, as clearly shown in their Figure 13, the two methods are largely equivalents: only two sources (out of 63 objects) have a V-K color consistent with early-type galaxies (V-K$`>`$5.2) but lie well below the Sa track.
We then used these two radio sub-samples selected on the basis of the Sa track to search for possible differences in the radio-to-optical ratio R (defined as in Section 4.1) between early and late type galaxies. Previous works on the optical identifications of radio sources (Kron et al. 1985, Gruppioni et al. 1999, Richards et al. 1999, Geogakakis et al. 1999, Prandoni et al. 2001, Ciliegi et al. 2003) have suggested, in fact, that the majority of the sources with low radio-to-optical ratio are associated with star forming galaxies, characterized by moderately weak intrinsic radio power. Viceversa, early type galaxies, in which radio luminosity is likely connected to nuclear activity, cover a much larger range in radio power, and hence in $`R`$, becoming the dominant population at high $`R`$. In Figure 9 we present the distribution of the radio-to-optical ratio $`R`$ for the two samples of early type galaxies (shaded histogram) and the sample of late type galaxies (empty histogram). The two distribution are very significantly different (at more than 7$`\sigma `$ level, on the basis of KS test): late-type, star forming galaxies are the dominant population at low radio-to-optical ratio $`R`$ (median Log$`R`$ = 3.08 $`\pm `$ 0.05, using the $`B_{AB}`$ magnitude to calculate $`R`$), while the majority of the galaxies in the region of high $`R`$ are early-type galaxies (median Log$`R`$= 4.06 $`\pm `$0.06). In the high R tail we still have a component of objects classified as late type galaxies (about one third of the objects with Log$`R`$ 4.6). A better understanding of the physical nature of these objects, which may include also star forming galaxies in which the high $`R`$ value is due to significant dust obscuration, requires spectroscopic data.
Finally, we estimated the relative fraction of the two populations (early and late type galaxies) in different radio flux bins. We used the four flux bins defined in Table 4. Assuming, consistently with Figure 9, that about 2/3 of the unidentified radio sources (which have, by definition, high $`R`$ values) are early-type galaxies, we find that while in the first three radio flux bins the fraction of early type galaxies is approximately constant at $``$40%, in the fainter radio flux bin this fraction decreases to $`30`$%. Therefore, although this result is still highly qualitative and based on very simple assumptions, we confirm that in the fainter radio flux limit bin our radio sample is likely to be dominated by late-type star forming galaxies as already suggested by the analysis of the optical magnitude distribution in different radio flux intervals (see Section 4.1).
### 4.6 Near Infrared magnitude properties
In Figure 10 we report the magnitude distributions in the three 2MASS bands for the 105 sources with an identification in the 2MASS catalogue and the J<sub>AB</sub> and K<sub>AB</sub> distributions for the 43 radio sources identified with the VIMOS near infrared survey. Moreover Figure 11 shows the $`(IK)_{AB}`$ color as function of the $`I_{AB}`$ magnitude. In considering these figures, it must be remembered how incomplete these data are : only 65 of the radio sources are within the area covered by the VVDS J and K bands survey, while the radio flux distribution of the radio sources with a 2MASS counterpart is strongly biased towards high radio flux due to the bright limit of the 2MASS survey. Figure 11 clearly shows that there is a monotonic trend for the optically fainter radio sources to have redder optical-infrared colors, and their color reaches $`(IK)_{AB}`$2 at $`I_{AB}`$23 mag. Similar results have been obtained for different radio surveys (Richards et al. 1999, Waddington et al. 2000, Ciliegi et al. 2003). From Figure 11 it is also interesting to note that only 7% (3/43) of the radio sources are very red sources with $`(IK)_{AB}>`$3.0. An even smaller percentage of EROS has been recently obtained by Sullivan et al. 2004 during the optical and near infrared identification of the radio sources in the Phoenix Deep Survey. None of the 91 radio sources detected in the $`K`$ band (over a total of 111 radio sources) has an $`(IK)_{AB}`$ color greater than 3.0 and only one source has an $`(IK)_{AB}>`$2.5. The fact that EROs sources are only a few percent of the total number of counterparts in radio surveys with a flux limit around $``$100$`\mu `$Jy is not surprising. Deep radio observations of optically/near-infrared selected EROS have shown that the EROs population is not a class of strong radio sources. For example, Smail et al. 2002, using very deep 1.4GHz radio data (3.5$`\mu `$Jy at 1$`\sigma `$) found only 3 EROs (over a total of 68) with a radio flux greater than 85$`\mu `$Jy, while Cimatti et al. 2003 studying a sample of 47 EROs, detected only one object at 1.4 GHz with a radio flux of $``$106 $`\mu `$Jy. However, the correlation between color and magnitude shown in Figure 11 suggests that the fraction of EROs is likely to be higher among the unidentified radio sources. This can be tested with deeper optical and infrared data.
## 5 Summary and Conclusion
In this paper we have presented the optical and near-infrared identifications of the 1054 radio sources detected in the 20cm deep radio survey obtained with the VLA in the VIMOS VLT Deep Survey VVDS-02h deep field (VVDS). Almost the whole square degree of the VVDS-VLA field has been observed in the $`B,V,R`$ and $`I`$ band down to a limiting magnitude of $`B_{AB}`$26.5, $`V_{AB}`$26.2, $`R_{AB}`$25.9 and $`I_{AB}`$25.0 (McCracken et al. 2003). Moreover, $``$0.71 deg<sup>2</sup> have been observed in the $`U`$ band down to $`U_{AB}`$25.4 (Radovich et al. 2004) and $``$165 arcmin<sup>2</sup> have been observed in the $`J`$ and $`K`$ bands down to $`J_{AB}`$24.2 and $`K_{AB}`$23.9 (Iovino et al. 2005). Using the Likelihood Ratio technique we optically identified 718 radio sources ($``$74% of the whole sample) . Sixty-five radio sources lie within the $`K`$ band area and we found a reliable counterpart for 43 of them. Among the 255 unidentified radio sources, 17 are empty fields ($`i.e.`$ they have no optical source within 5 arcsec from their position), while the other 238 sources have at least one optical source within 5 arcsec but all with $`LR<LR_{\mathrm{th}}`$.
The color properties of the optical counterparts of the radio sources have been analysed using the $`(BV)_{AB}`$ and $`(VI)_{AB}`$ colors. The optical counterparts of the radio sources classified as extended have been analysed in three different magnitude slices. While in the brightest magnitude range (18$`<I_{AB}<`$20) the optical color properties of the radio sources are not significantly different from those of the whole optical sample, at fainter magnitude the median $`(VI)_{AB}`$ color of the radio sources is redder than the median color of the whole optical sample, suggesting a higher redshift for the radio sources. This is also supported by the photometric redshift analysis which shows that, in each magnitude bin, the radio sample has a higher median photometric redshift than the whole optical sample. This suggests that radio detection is preferentially selecting galaxies with higher intrinsic optical luminosity.
Using the $`(BI)_{AB}`$ color and the photometric redshift for all the radio sources with a reliable optical counterpart, we have tentatively divided the radio sample in two sub-samples: the sources above the Sa galaxies track have been considered as early type galaxies, while the sources below the same track have been considered as late type galaxies. The analysis of the radio-to-optical ratio $`R`$ of the two sub-samples confirms with high statistical significance the results already obtained by other authors : late-type, star forming galaxies are the dominant population at low $`R`$, while in the region of high $`R`$ the majority of these objects are early-type galaxies.
From the analysis of the optical properties of the radio sources in different radio flux bins, we found that while about 35% of the radio sources are optically unidentified in the first radio flux bin, the percentage of unidentified sources decreases to about 25% in the faintest two bins (S$`<`$ 0.5 mJy). The median $`I_{AB}`$ magnitude for the total sample of radio sources, i.e. including also the unidentified ones, is brighter in the faintest two radio bins than in the bin with higher radio flux. This result shows that the faintest radio sources are not in general the faintest sources at optical wavelengths and would suggest that most of the faintest radio sources are likely to be associated to relatively lower radio luminosity objects at relatively modest redshift, rather than radio-powerful, AGN type objects at high redshift. Using the above classification in early-type and late-type galaxies we found that the majority of the radio sources below $``$0.15 mJy are indeed late-type star forming galaxies in the photometric redshift range 0.1$``$z$`{}_{phot}{}^{}`$1.5. These results are in agreement with the results obtained by several authors : the majority of the optical identification of the $`\mu `$Jy radio sources are with luminous ($`L>L_{}`$) galaxies at modest redshifts (0$``$z$``$1), many of which with evidence for recent star formation (Windhorst et al. 1995, Richards et al. 1998, Richards et al. 1999, Roche et al. 2002).
Finally, as already noted by other authors for different radio surveys, we found a monotonic trend for the optically fainter radio sources to be associated with redder galaxies. In the area covered by $`K`$ data 3 out of 43 radio sources with a likely $`K`$ band counterpart have very red $`(IK)_{AB}`$ colors.
The analysis of the sub-sample of spectroscopically identified radio-sources is in progress and will be presented elsewhere.
###### Acknowledgements.
This research has been developed within the framework of the VVDS consortium.
This work has been partially supported by the Italian Ministry (MIUR) grants COFIN2003 (num.2003020150).
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.
|
warning/0506/cond-mat0506455.html
|
ar5iv
|
text
|
# Shot noise free conductance reduction in quantum wires
(June 2005)
## Abstract
We show that a shot noise free current at conductance below $`2e^2/h`$ is possible in short interacting quantum wires without spin-polarization. Our calculation is done for two exactly solvable limits of the “Coulomb Tonks gas”, a one-dimensional gas of impenetrable electrons that can be realized in ultra-thin quantum wires. In both cases we find that charge transport through such a wire is noiseless at zero temperature while the conductance is reduced to $`e^2/h`$.
The current through short electrical conductors is generically noisy even in the absence of thermal fluctuations. This nonequilibrium noise is usually referred to as shot noise and originates from backscattering of electrons in the conductor Bla00 . The model of noninteracting electrons predicts that shot noise can only be avoided in fully ballistic wires without scatterers for electrons. The conductance of such wires is quantized in multiples of the conductance quantum $`2e^2/h`$ vWees88 ; Wha88 . This implies that the electrical current through a noninteracting wire with conductance below $`2e^2/h`$ is bound to be noisy unless one polarizes the spins of electrons. The suppression of shot noise for ballistic wires with conductance equal to a multiple of $`2e^2/h`$ has been verified experimentally for quantum point contacts Rez95 . Recently, however, experiments on similar conductors have shown that shot noise can be suppressed as well at conductance $`0.7\times 2e^2/h`$, i.e., below the ballistic conductance minimum $`2e^2/h`$ noise .
The measured suppression cannot be understood within the model of noninteracting electrons unless one assumes the breaking of a symmetry, e.g. by spin-polarization noise . Hence, one may ask whether it is possible to have noise-free conduction at a conductance below $`2e^2/h`$ as a result of electron-electron interactions, but without spin-polarization. In this letter we show that the answer is positive. For a specific theoretical model, the strongly interacting “Coulomb Tonks gas” of impenetrable electrons Fog05 ; Fiete05 , we find a noiseless charge current in a wire with conductance $`0.5\times 2e^2/h`$. It is accompanied by pronounced spin current fluctuations, which precludes a spin-polarization of electrons as the mechanism of conductance reduction. Our calculation is done for a finite-length wire in two exactly solvable limits: a high applied bias, and a fast spin-relaxation rate. The Coulomb Tonks gas can be realized in ultra-thin wires, such as carbon nanotubes Fog05 , for which there exists some experimental evidence of an anomalous conductance reduction CN .
Theoretically it has been shown by Matveev that the low-bias conductance of a quantum wire is reduced to $`e^2/h`$ at low electron density, when interactions induce the formation of a Wigner crystal, if the temperature is larger than the typical energy $`J`$ of spin excitations Mat04 . This is one of the scenarios that may explain the anomalous conductance reduction observed experimentally in quantum point contacts 07 . The Coulomb Tonks gas studied here is closely related to Matveev’s model at $`J=0`$, since impenetrable electrons have no exchange interaction.
We model a wire in the Coulomb Tonks gas regime by a Hubbard chain with infinite on-site repulsion, $`U\mathrm{}`$. For an infinite-length Hubbard chain, this model can be formulated in terms of spinless holes and a static spin background. As the only charge carriers are spinless fermions, one expects that such a chain has a reduced conductance of $`e^2/h`$ without shot noise. Nanoscale conductors studied in experiments, however, have a finite length and they are connected to bulk leads with well-screened interactions. As the example of Luttinger Liquids shows, these contacts can nullify interaction effects on the conductance LL as well as the shot noise LLnoise in infinite wires. Once coupled to noninteracting reservoirs, the spin background in the infinite-$`U`$ Hubbard chain acquires nontrivial dynamics because of spin-exchange processes of the chain with the reservoirs, making the model hard to solve. Below, we will identify two limits in which this spin dynamics can be controlled and the transport properties of the system can be evaluated exactly.
The system we consider is shown in Fig. 1. The sites of the infinite-$`U`$ Hubbard chain are labeled $`j=1,\mathrm{},n`$. The Hubbard chain is coupled to two noninteracting leads, with sites $`j0`$ and $`j>n`$. We introduce operators $`\psi _{j\sigma }`$ that annihilate an electron with spin $`\sigma `$ on site $`j`$ in the leads and $`c_j`$ that annihilate a hole at position $`j`$ in the Hubbard chain. The spin index $`\sigma `$ takes the values $`1,\mathrm{},2S+1`$. We also introduce operators $`S_{\mathrm{L}\sigma }^{}`$ and $`S_{\mathrm{R}\sigma }^{}`$ that add a spin $`\sigma `$ to the left and right of the spin configuration of the Hubbard chain, respectively. Our model Hamiltonian then reads
$`H`$ $`=`$ $`H_{\mathrm{lead}}+H_{\mathrm{chain}}+H^{},`$ (1)
with
$`H_{\mathrm{lead}}`$ $`=`$ $`\lambda {\displaystyle \underset{\sigma }{}}\left[{\displaystyle \underset{j<0}{}}\psi _{j\sigma }^{}\psi _{j+1\sigma }+{\displaystyle \underset{j>n}{}}\psi _{j\sigma }^{}\psi _{j+1\sigma }\right]+\text{h.c.}`$
$`H_{\mathrm{chain}}`$ $`=`$ $`{\displaystyle \underset{j=1}{\overset{n1}{}}}\lambda _jc_j^{}c_{j+1}+\text{h.c.}`$
$`H^{}`$ $`=`$ $`{\displaystyle \underset{\sigma }{}}\left(\lambda _0\psi _{0\sigma }^{}c_1^{}S_{\mathrm{L}\sigma }^{}+\lambda _n\psi _{n+1\sigma }^{}c_n^{}S_{\mathrm{R}\sigma }^{}\right)+\text{h.c.},`$
where we allowed for spatial variations of the hopping amplitude $`\lambda _j`$ of the Hubbard chain.
Moments of the number $`\mathrm{\Delta }N_\sigma =N_\sigma (\tau )N_\sigma (0)`$ of spin $`\sigma `$ electrons that are transported through the chain during a time $`\tau `$ can be obtained from the generating function
$$𝒵(\{\xi _\sigma \})=\mathrm{Tr}e^{i\xi N}e^{iH\tau }e^{i\xi N}\rho e^{iH\tau }.$$
(2)
Here $`\rho `$ is the initial density matrix of the conductor and $`\xi N`$ abbreviates $`_\sigma \xi _\sigma N_\sigma `$ with $`N_\sigma =_{j0}\psi _{j\sigma }^{}\psi _{j\sigma }^{}`$. For large $`\tau `$, $`𝒵`$ generates zero-frequency (spin) current correlators upon differentiation. We rewrite $`𝒵`$ as
$$𝒵(\{\xi _\sigma \})=\mathrm{Tr}e^{iH_\xi \tau }\rho e^{iH\tau },$$
(3)
where $`H_\xi =\mathrm{exp}(i\xi N)H\mathrm{exp}(i\xi N)`$. $`H_\xi `$ is obtained from $`H`$ by the substitution $`\psi _{j\sigma }e^{i\xi _\sigma }\psi _{j\sigma }`$ for $`j0`$. Since the Hamiltonian (1) is quadratic in the fermions, the lead fermions may be integrated out, so that the generating function $`𝒵`$ takes the form
$`𝒵(\{\xi _\sigma \})`$ $`=`$ $`T_\mathrm{c}\mathrm{exp}[i\lambda _0^2{\displaystyle _\mathrm{c}}dt_1dt_2{\displaystyle \underset{\sigma }{}}S_{\mathrm{L}\sigma }^{}(t_1)c_1(t_1)G_{\mathrm{L}\sigma \xi }(t_1,t_2)c_1^{}(t_2)S_{\mathrm{L}\sigma }^{}(t_2)`$ (4)
$`i\lambda _n^2{\displaystyle _\mathrm{c}}dt_1dt_2{\displaystyle \underset{\sigma }{}}S_{\mathrm{R}\sigma }^{}(t_1)c_n(t_1)G_{\mathrm{R}\sigma }(t_1,t_2)c_n^{}(t_2)S_{\mathrm{R}\sigma }^{}(t_2)]_{\mathrm{chain}},`$
where c denotes the Keldysh contour, $`G_\mathrm{L}`$ and $`G_\mathrm{R}`$ are Keldysh Green functions for the sites $`0`$ and $`n+1`$, respectively, and the averaging brackets $`\mathrm{}_{\mathrm{chain}}`$ represent an average with respect to the Hamiltonian $`H_{\mathrm{chain}}`$ of the uncoupled Hubbard chain. Using matrix notation in Keldysh space, one has $`G_{\mathrm{L}\sigma \xi }=e^{i\xi _\sigma \tau _3/2}G_{\mathrm{L}\sigma }e^{i\xi _\sigma \tau _3/2}`$, where $`\tau _3`$ is the third Pauli matrix in Keldysh space.
The first limit in which exact results can be obtained is that of reservoirs with a large bandwidth $`\lambda \mathrm{}`$ that are completely filled with electrons on the left and completely empty on the right side of the chain. In that case, the reservoir Green functions $`G_\mathrm{L}`$ and $`G_\mathrm{R}`$ take a particularly simple form,
$`G_\mathrm{L}(t,t^{})`$ $`=`$ $`{\displaystyle \frac{2i}{\lambda }}\delta _\lambda (tt^{})\left(\begin{array}{cc}\theta (t^{}t)& 1\\ 0& \theta (tt^{})\end{array}\right),`$ (7)
$`G_\mathrm{R}(t,t^{})`$ $`=`$ $`{\displaystyle \frac{2i}{\lambda }}\delta _\lambda (tt^{})\left(\begin{array}{cc}\theta (tt^{})& 0\\ 1& \theta (t^{}t)\end{array}\right),`$ (10)
where $`\delta _\lambda `$ is an approximation to a Dirac delta-function with width $`1/\lambda `$. In order to be able to take the limit $`\lambda \mathrm{}`$ while having a finite transport current we assume that the bandwidth of the chain is adiabatically reduced. We assume that $`\lambda _j=\lambda `$ for $`j`$ close to the boundaries of the Hubbard chain at $`j=1`$ and $`j=n`$, and that $`\lambda _j`$ is adiabatically reduced to a value $`\lambda _j=\lambda _\mathrm{r}\lambda `$ in the center of the chain, $`jn/2`$.
Below, we will argue that, in this limit, one may make the following substitutions in the exponent of Eq. (4),
$`{\displaystyle \underset{\sigma }{}}S_{\mathrm{L}\sigma }^{}(t_1)G_{\mathrm{L}\sigma \xi }(t_1,t_2)S_{\mathrm{L}\sigma }^{}(t_2)`$ $``$ $`{\displaystyle \underset{\sigma }{}}G_{\mathrm{L}\sigma \xi }(t_1,t_2),`$
$`{\displaystyle \underset{\sigma }{}}S_{\mathrm{R}\sigma }^{}(t_1)G_{\mathrm{R}\sigma }(t_1,t_2)S_{\mathrm{R}\sigma }^{}(t_2)`$ $``$ $`G_{\mathrm{R}\sigma }(t_1,t_2).`$ (11)
After these replacements, the generating function $`𝒵`$ is readily calculated, and we find
$$\mathrm{ln}𝒵=\frac{\tau }{h}_{2\lambda _\mathrm{r}}^{2\lambda _\mathrm{r}}𝑑\omega \mathrm{ln}\left[1+\frac{T(\omega )}{2S+1}\underset{\sigma }{}(e^{i\xi _\sigma }1)\right],$$
(12)
where, for $`\lambda _n=\lambda `$,
$$T(\omega )=\frac{(2S+1)|\lambda _0|^2(4\lambda ^2\omega ^2)}{(\lambda ^2+(2S+1)|\lambda _0|^2)^2(2S+1)|\lambda _0|^2\omega ^2}.$$
(13)
(We set $`\lambda _n=\lambda `$ to simplify the expressions reported here; it is not essential for the calculation.) The function $`T(\omega )`$ reaches its maximum $`T^{\mathrm{max}}(\omega )=1`$ for $`\lambda _0=\lambda /\sqrt{2S+1}`$ arti . While the maximum current in the corresponding wire without interactions would be $`I_\mathrm{n}^{\mathrm{max}}=4(2S+1)\lambda _\mathrm{r}e/h`$, Eq. (12) implies for the interacting case
$$I^{\mathrm{max}}=\frac{1}{2S+1}I_\mathrm{n}^{\mathrm{max}}.$$
(14)
Hence, for electrons with spin $`S=1/2`$ interactions reduce the maximal charge current by a factor of $`2`$. This signals a reduction of the conductance of the wire from $`2e^2/h`$ to $`e^2/h`$, as in the infinitely long system. Every spin component carries an equal part of that total charge current independently of the chosen spin quantization direction: the electrons in the wire are not spin-polarized. This absence of spin-polarization in the wire is also reflected in fluctuations of the spin currents through the wire. At maximal transmission we find for their variance
$$\text{var}\mathrm{\Delta }N_\sigma =\frac{4\lambda _\mathrm{r}\tau }{h}\frac{2S}{(2S+1)^2}.$$
(15)
In contrast, the charge current is noiseless in this case:
$$\mathrm{var}\left(\underset{\sigma }{}\mathrm{\Delta }N_\sigma \right)=0.$$
(16)
Our model therefore indeed exemplifies a mechanism of conductance reduction through electron-electron interactions that does not introduce any shot noise. Instead of blocking the transport of electrons with certain spin directions by spin-polarizing the wire, this mechanism reduces the conductance by limiting the density of conduction electrons to the spinless case while allowing them to have arbitrary spin.
We now discuss the justification of the replacements (11). This is done by expanding the exponential of Eq. (4) in powers of $`\lambda _0`$ and $`\lambda _n`$. For simplicity we specialize to the case $`\lambda _0=\lambda _n/\sqrt{2S+1}`$. Since the spin operators $`S_{\mathrm{L}\sigma }^{}`$ and $`S_{\mathrm{L}\sigma }^{}`$ have no dynamics in the absence of coupling to the reservoirs, the time dependence of the spin operators is not important; only their order in the contour-ordered expression (4) matters. To second order in $`\lambda _n`$, the causal structure of the reservoir Green function $`G_\mathrm{L}`$ is such that $`S_{\mathrm{L}\sigma }^{}(t_2)`$ always appears to the left of $`S_{\mathrm{L}\sigma }^{}(t_1)`$. Since $`S_{\mathrm{L}\sigma }^{}S_{\mathrm{L}\sigma }^{}=1`$, the first substitution rule (11) follows. Similarly, the causal structure of the reservoir Green function $`G_\mathrm{R}`$ is such that $`S_{\mathrm{R}\sigma }^{}(t_2)`$ always appears to the right of $`S_{\mathrm{R}\sigma }^{}(t_1)`$. The operator product $`S_{\mathrm{R}\sigma }^{}S_{\mathrm{R}\sigma }^{}=1`$ if the right-most electron of the Hubbard chain has spin $`\sigma `$, and $`S_{\mathrm{R}\sigma }^{}S_{\mathrm{R}\sigma }^{}=0`$ otherwise. Hence $`_\sigma S_{\mathrm{R}\sigma }^{}S_{\mathrm{R}\sigma }^{}=1`$, and the second substitution rule of Eq. (11) follows.
We next prove the substitution rules (11) for higher orders in $`\lambda _n`$. A key ingredient in this will be the locality of the reservoir Green functions which groups spin operators into ordered pairs that act almost simultaneously. This essentially decouples different spin operator pairs and the rules (11) follow like at first order in $`\lambda _n`$. To show this we decompose the hole annihilation operators $`c_j=c_{j,e}+c_{j,t}`$ into a contribution $`c_{j,e}`$ from evanescent states of the chain at energies $`ϵ2\lambda _\mathrm{r}`$ and a contribution $`c_{j,t}`$ from transmitting states with $`ϵ<2\lambda _\mathrm{r}`$. In any term of an expansion of $`𝒵`$ in powers of $`\lambda _n`$ we first contract all hole operators $`c_{j,e}`$ into pairs using Wick’s theorem. This produces sequences of the form
$`D_{\mathrm{L};l}`$ $`=`$ $`{\displaystyle _\mathrm{c}}𝑑t_2\mathrm{}𝑑t_{l1}c_{1,t}^{}(t_1)S_\mathrm{L}^{}(t_1)G_{\mathrm{L}\xi }(t_1,t_2)`$ (17)
$`\times S_\mathrm{L}^{}(t_2)g_e(t_2,t_3)\mathrm{}S_\mathrm{L}^{}(t_l)c_{1,t}^{}(t_l),`$
consisting of spin operators and two transmitting hole operators $`c_{1,t}`$, connected by Green functions for lead and chain fermions. All evanescent hole states on the left side of the chain are unoccupied because of their proximity to the left reservoir, so that the evanescent hole Green function $`g_e`$ has the same causal structure as $`G_\mathrm{R}`$, cf. Eq. (10). This assures that after time ordering all spin operators in such sequences are grouped in pairs $`S_\mathrm{L}^{}(t_i)G_{\mathrm{L}\xi }(t_i,t_{i+1})S_\mathrm{L}^{}(t_{i+1})`$ originating from the same reservoir Green function with no other spin operator $`S_\mathrm{L}^{}`$ or $`S_\mathrm{L}^{}`$ acting in between. The substitution rules (11) follow then for such sequences as before for a single pair. The same arguments hold for the right side of the chain. It remains to contract the operators $`c_{j,t}`$ at the beginning and at the end of sequences $`D_{\mathrm{L},\mathrm{R},l}`$ of the form (17). Here, nothing prevents spin operators of one sequence $`D_{\mathrm{L},\mathrm{R};k}`$ to occur in time between operators $`S_\mathrm{L}^{}(t_i)G_{\mathrm{L}\xi }(t_i,t_{i+1})S_\mathrm{L}^{}(t_{i+1})`$ in another sequence $`D_{\mathrm{L},\mathrm{R};l}`$. Such events cause corrections to the generating function calculated with the substitution rules (11). However, because of the locality of the lead Green functions $`G`$, such corrections occur only in time intervals of length $`1/\lambda `$. This is much shorter than the range of the integrals over the times $`t_1`$ and $`t_l`$ of sequences $`D_{\mathrm{L},\mathrm{R};l}`$, which is set by the time scale $`1/\lambda _\mathrm{r}`$ on which the Green functions of transmitting hole states vary. Therefore, corrections are of order $`\lambda _\mathrm{r}/\lambda `$, and can be neglected in our limit $`\lambda _\mathrm{r}/\lambda 0`$.
To make this argument rigorous one needs to estimate the corrections to all terms in an expansion of the cumulant generating function $`\mathrm{ln}𝒵`$. Only “connected” diagrams contribute to $`\mathrm{ln}𝒵`$. Spin operators in its expansion can be connected either by fermion Green functions or by deviations from the operator ordering underlying the substitution rules (11). As we argued above, the corrections to diagrams where all spin operators are connected by Green functions are of order $`\lambda _\mathrm{r}/\lambda `$. Let us now estimate the correction arising from diagrams that are connected because of deviations from the ordering underlying the rules (11). Hereto, note that a product of two sequences $`j`$ and $`k`$ of spin operators that are connected cyclically by fermion Green functions only within themselves still contributes to $`\mathrm{ln}𝒵`$ if a spin operator of one of the sequences acts in between an operator pair $`S_\mathrm{L}^{}(t_i)G_{\mathrm{L}\xi }(t_i,t_{i+1})S_\mathrm{L}^{}(t_{i+1})`$ of the other sequence. When uncorrelated the two subsequences are of order $`\tau (\lambda _n/\lambda )^{2n_j}\lambda _\mathrm{r}`$ and $`\tau (\lambda _n/\lambda )^{2n_k}\lambda _\mathrm{r}`$, respectively, where $`n_m`$ is the number of reservoir Green functions $`G`$ contained in sequence $`m`$. Correlations between the subsequences occur for the time $`1/\lambda `$ over which spin operators interfere and the corresponding correction is therefore of order $`\tau (\lambda _n/\lambda )^{2n_j+2n_k}(\lambda _\mathrm{r}/\lambda )\lambda _\mathrm{r}`$. This is of the same order as the correction to a sequence connected entirely by Green functions at the same order in $`\lambda _n`$. This correspondence holds for diagrams with an arbitrary number of connections by spin operators. We now take into account that every one of the discussed sequences contains many pairs of spin operators that can interfere with other operators. We estimate that at order $`(\lambda _\mathrm{r}/\lambda )^k`$ the correction $`\mathrm{\Delta }_{k;l}`$ to a term of order $`2l`$ in $`\lambda _n`$ in the expansion of $`\mathrm{ln}𝒵`$ is of relative order
$$\mathrm{\Delta }_{k;l}\left(\begin{array}{cc}2l& \\ k& \end{array}\right)\left(\frac{\lambda _n}{\lambda }\right)^{2l}\left(\frac{\lambda _\mathrm{r}}{\lambda }\right)^k.$$
(18)
Summing up these contributions one finds that the total correction to $`\mathrm{ln}𝒵`$ is of relative order $`ϵ1`$ if $`\lambda _n/\lambda <\mathrm{exp}(\lambda _\mathrm{r}\mathrm{ln}ϵ/\lambda ϵ)`$. In the limit $`\lambda _\mathrm{r}/\lambda 0`$ the substitution rules Eq. (11) are therefore justified for arbitrarily good transmission $`T<1`$, where $`T1`$ in the limit $`\lambda _n\lambda `$.
The second limit in which exact results can be obtained is in the presence of additional spin relaxation processes in the Hubbard chain. To the Hamiltonian (1) we add a random time-dependent magnetic field that causes spin memory to be lost at a rate $`\gamma _S`$. In the limit $`\gamma _S|\lambda |`$ our model is again exactly solvable, now both in and out of equilibrium and without assumptions about $`\lambda _j`$. To demonstrate this we consider a wire with a constant $`\lambda _j=\lambda `$. Every operator $`S_{\mathrm{R}\sigma }^{}(t)`$ or $`S_{\mathrm{L}\sigma }^{}(t)`$ in an expansion of $`𝒵`$ is nonvanishing only if $`\sigma `$ equals the state of the last or first spin of the Hubbard chain at time $`t`$. In the limit $`\gamma _S|\lambda |`$ any memory of the state of the spin put into the chain by previous actions of the operators $`S_{\mathrm{R}\sigma }^{}`$ and $`S_{\mathrm{L}\sigma }^{}`$ is immediately erased by relaxation processes. After averaging over the random magnetic field, every term from the exponent of Eq. (4) contributes therefore only for one random spin state $`\sigma `$. This allows us again to represent $`𝒵`$ by an expression of the form of Eq. (4) without spin operators. To obtain the generating function of charge currents we set $`\xi _\sigma =e\xi `$ and evaluate Eq. (4) with the substitutions
$`{\displaystyle \underset{\sigma }{}}S_{\mu \sigma }^{}(t_1)G_{\mu \sigma }(t_1,t_2)S_{\mu \sigma }(t_2)`$ $``$ $`G_{\mu \sigma }(t_1,t_2),`$ (19)
$`\mu \{\mathrm{L},R\}`$, to find that
$`\mathrm{ln}𝒵`$ $`=`$ $`{\displaystyle \frac{\tau }{h}}{\displaystyle _{2\lambda }^{2\lambda }}d\omega \mathrm{ln}\{1+T(\omega )`$ (20)
$`[f_L(1f_R)(e^{ie\xi }1)+f_R(1f_L)(e^{ie\xi }1)]\}.`$
Here, $`T(\omega )`$ is given by Eq. (13) evaluated at $`S=0`$ and $`f_\mu `$ are the occupation numbers of reservoir states. $`T(\omega )`$ now reaches its maximum $`T^{\mathrm{max}}(\omega )=1`$ for $`\lambda _0=\lambda _n=\lambda `$. Charge transport through the chain becomes indistinguishable from that through a noninteracting single-channel conductor for spinless fermions. We again find noiseless charge current at zero temperature while the conductance is reduced to $`e^2/h`$, a factor $`2S+1`$ smaller than its noninteracting value. This shows that the discussed mechanism of noiseless conductance reduction works also in equilibrium and for a not fully occupied electronic band. In fact, the condition for the spin relaxation rate $`\gamma _S`$ may be relaxed for a wire with an almost empty conduction band. For a long wire, $`n\lambda ^2/\mathrm{\Delta }\lambda \gamma _S`$, Eq. (20) can be shown to hold also in the experimentally more relevant situation that the spin decoherence rate in the wire exceeds the range $`\mathrm{\Delta }\lambda `$ of energies for which electronic states are appreciably occupied, $`\gamma _S\mathrm{\Delta }\lambda `$ unpub .
In conclusion, by studying two exactly solvable limits of a Coulomb Tonks gas coupled to bulk leads, we have shown that strong electron-electron interactions can lead to a noiseless conductance reduction without spin-polarization. One may speculate that the same mechanism could be operative also in other parameter regimes. As such it appears to be a plausible alternative interpretation of the measurement of a suppression of shot noise at the 0.7-plateau of quantum point contacts reported in Refs. noise . Further evidence for the studied mechanism could be found in similar measurements on carbon nanotubes, to which our model applies more directly Fog05 . Such measurements could moreover put an important constraint on theories of the anomalous conductance reduction observed in some quantum wires 07 ; CN .
This work was supported by the NSF under grant no. DMR 0334499 and by the Packard Foundation.
|
warning/0506/astro-ph0506409.html
|
ar5iv
|
text
|
# Deep echelle spectrophotometry of S 311, a Galactic H ii region located outside the solar circle Based on observations collected at the European Southern Observatory, Chile, proposal number ESO 68.C-0149(A)
## 1 Introduction
S 311 —also known as NGC 2467— is an H ii region of the Sharpless catalogue (Sharpless, 1959) which is located outside the solar circle. This nebula forms part of the Puppis I association located at 4.0 kpc from the Sun (Russeil, 2003) and at a Galactocentric distance of 10.43 kpc (assuming a Galactocentric solar distance of 8.0 kpc). (Albert et al.1986) concluded that the radio morphology of S 311 is consistent with blister processes. There are few spectrophotometric studies of S 311 in the literature, most of them forming part of analisys involving several H ii regions (e.g. Hawley, 1978; Peimbert, Torres-Peimbert & Rayo, 1978; Kennicutt et al., 2000). All these works are based on the analysis of collisionally excited lines (hereinafter CELs).
We have taken long-exposure high-spectral-resolution spectra with the Very Large Telescope (VLT) UVES echelle spectrograph to obtain accurate measurements of very faint permitted lines of heavy element ions in S 311. We have determined the physical conditions and the chemical abundances of S 311 with high accuracy. An important improvement in this work is the derivation of C<sup>++</sup> and O<sup>++</sup> abundances from several pure recombination lines (hereinafter RLs) of C ii and O ii, making use of high spectral resolution and avoiding the problems of line blending.
Traditionally, the abundance studies for H ii regions have been based on determinations from CELs, whose emissivities are strongly dependent on the temperature variations over the observed volume of the nebula. Alternatively, the emissivities of RLs are almost independent of such variations and are, in principle, more precise indicators of the true chemical abundances of the nebula. The use of high resolution echelle spectrographs has permitted our group to obtain deep high resolution spectra of bright Galactic H ii regions (e.g. Esteban et al., 1998, 1999a; Esteban et al., 1999b, 2004; García-Rojas et al., 2004), and extragalactic H ii regions (e.g. Esteban et al., 2002; Peimbert, 2003); all these works have found that abundance determinations from RLs are systematically larger than those obtained using CELs (the so called *abundance discrepancy* problem). One of the most probable causes of this abundance discrepancy is the presence of spatial variations or fluctuations in the temperature structure of the nebulae (Torres-Peimbert, Peimbert, & Daltabuit, 1980). Both phenomena can be related due to the different functional dependence of the line emissivities of CELs and RLs on the electron temperature, which is stronger —exponential— in the case of CELs. The temperature fluctuations have been parametrized traditionally by *$`t^2`$*, the mean square temperature fluctuation of the gas (Peimbert, 1967). We have computed *$`t^2`$* values from the comparison of abundances derived using CELs or RLs, and from the temperatures derived from CELs and recombination processes.
The main aims of this work are to present the high-quality spectrophotometric data for S 311 obtained with the ESO Very Large Telescope (VLT), to assess the old-fashioned abundance analysis of S 311 in the literature, calculate C<sup>++</sup> and O<sup>++</sup> abundances from recombination lines, and report the detection and measurement of weak deuterium Balmer lines.
In §§ 2 and 3 we describe the observations, the data reduction procedure and the measurement and identification of the emission lines. In § 4 we obtain temperatures and densities using several diagnostic ratios. In § 5 ionic abundances are determined based on CELs, as well as on RLs. In § 6 we discuss the *$`t^2`$* results. Total abundances are determined in § 7. In § 8 we discuss the detection of deuterium Balmer lines. Finally, in §§ 9 and 10 we present the discussion and the conclusions, respectively.
## 2 Observations and Data Reduction
The observations were made on 2003 March 30 with the Ultraviolet Visual Echelle Spectrograph, UVES (D’Odorico et al., 2000), at the VLT Kueyen Telescope in Cerro Paranal Observatory (Chile). We used the standard settings in both the red and blue arms of the spectrograph, covering the region from 3100 to 10400 Å . The log of the observations is presented in Table 1.
The wavelength regions 5783–5830 Å and 8540–8650 Å were not observed due to a gap between the two CCDs used in the red arm. There are also five small gaps that were not observed, 9608–9612 Å, 9761–9767 Å, 9918–9927 Å, 10080–10093 Å and 10249–10264 Å, because the five redmost orders did not fit completely within the CCD. We took long and short exposure spectra to check for possible saturation effects.
The slit was oriented east-west and the atmospheric dispersion corrector (ADC) was used to keep the same observed region within the slit regardless of the air mass value. The slit width was set to 3.0$`\mathrm{}`$ and the slit length was set to 10$`\mathrm{}`$ in the blue arm and to 12$`\mathrm{}`$ in the red arm; the slit width was chosen to maximize the S/N ratio of the emission lines and to maintain the required resolution to separate most of the weak lines needed for this project. The effective resolution at a given wavelength is approximately $`\mathrm{\Delta }\lambda \lambda /8800`$. The centre of the slit was placed 126$`\mathrm{}`$ south of the main ionizing star HD 64315 (O6e), covering the brightest region of S 311 (see Figure 1). The reductions were made for an area of 3$`\mathrm{}`$$`\times `$8.5$`\mathrm{}`$.
The spectra were reduced using the IRAF<sup>1</sup><sup>1</sup>1IRAF is distributed by NOAO, which is operated by AURA, under cooperative agreement with NSF. echelle reduction package, following the standard procedure of bias subtraction, aperture extraction, flatfielding, wavelength calibration and flux calibration. The standard stars EG 247, C-32d9927 (Hamuy et al., 1992, 1994) and HD 49798 were observed for flux calibration.
## 3 Line Intensities and Reddening Correction
Line intensities were measured integrating all the flux in the line between two given limits and over a local continuum estimated by eye. In the cases of line blending, a multiple Gaussian profile fit procedure was applied to obtain the line flux of each individual line. Most of these measurements were made with the SPLOT routine of the IRAF package. In some cases of very tight blends or blends with very bright telluric lines the analysis was performed via Gaussian fitting making use of the Starlink DIPSO software (Howard & Murray, 1990).
Table 2 presents the emission line intensities of S 311. The first and fourth columns list the adopted laboratory wavelength, $`\lambda _0`$, and the observed wavelength in the heliocentric framework, $`\lambda `$. The second and third columns give the ion and the multiplet number, or series for each line. The fifth and sixth columns give the observed flux relative to H$`\beta `$, $`F(\lambda `$), and the flux corrected for reddening relative to H$`\beta `$, $`I(\lambda `$). The seventh column gives the fractional error (1$`\sigma `$) in the line intensities (see García-Rojas et al., 2004, for details in error analysis).
A total of 263 emission lines were measured; of them 178 are permitted, 65 are forbidden and 19 are semiforbidden (see Table 2). Two H i Paschen lines and one C ii line are blended with telluric lines, making impossible their measurement. Several other lines were strongly affected by atmospheric features in absorption, by internal reflections or charge transfer in the CCD, rendering their intensities unreliable. Also, 23 lines are dubious identifications and one emission line could not be identified in any of the available references. Those lines are indicated in Table 2.
The identification and adopted laboratory wavelengths of the lines were obtained following several previous identifications in the literature (see García-Rojas et al., 2004; Esteban et al., 2004, and references therein)
We have assumed the standard extinction for the Milky Way (R<sub>v</sub>=3.1) parametrized by Seaton (1979). We have derived a logarithmic interstellar extinction coefficient of $`c(H\beta )`$=0.64 $`\pm `$ 0.04 dex was determined by fitting the observed $`I`$(H Balmer lines)/$`I`$(H$`\beta `$) ratios (from H16 to H$`\beta `$) and $`I`$(H Paschen lines)/$`I`$(H$`\beta `$) (from P22 to P7), to the theoretical ones computed by Storey & Hummer (1995) for $`T_\mathrm{e}`$ = 10000 K and $`n_\mathrm{e}`$ = 1000 cm<sup>-3</sup> (see below). H i lines affected by blends or atmospheric absorption were not considered. The derived value of $`c(H\beta )`$ is in very good agreement with previous determinations in the same nebula: Hawley (1978) derived $`c(H\beta )`$=0.61 and 0.67 for two slit positions with offsets of 96$`\mathrm{}`$ north and 96$`\mathrm{}`$ north, 35$`\mathrm{}`$ west respectively from our slit position. Furthermore, Peimbert, Torres-Peimbert & Rayo (1978) derived $`c(H\beta )`$=0.6 and 0.7 for slit positions 33$`\mathrm{}`$ north, 12$`\mathrm{}`$ west and 33$`\mathrm{}`$ north, 106$`\mathrm{}`$ west, respectively. These last authors used the Whitford (1958) extinction law, which is almost coincident with the one we have adopted. Moreover, Kennicutt et al. (2000), derived a value of $`c(H\beta )`$=0.67 using the average interstellar reddening curve from Cardelli, Clayton & Mathis (1989). Shaver et al. (1983) observed two positions in S 311, one of them (position 2), almost coincident with our slit position. They derived $`c(H\beta )`$=0.5 and 0.7 for their positions 1 and 2 respectively. We can conclude then that apparently there are no significant variations of the extinction inside S 311.
## 4 Physical Conditions
The large number of emission lines identified and measured in the spectra allows us the derivation of physical conditions using different emission line ratios. The temperatures and densities are presented in Table 3. Most of the determinations were carried out with the IRAF task TEMDEN of the package NEBULAR (Shaw & Dufour, 1995).
The methodology followed for the derivation of $`n_\mathrm{e}`$ and $`T_\mathrm{e}`$ has been described in a previous paper (i.e. García-Rojas et al., 2004). In the case of electron densities, ratios of CELs of several ions have been used. The lastest version of NEBULAR (February 2004) uses the transition probabilities recommended by Wiese et al. (1996) and the collision strengths of McLaughlin & Bell (1993) for the \[O ii\] $`\lambda `$3729/$`\lambda `$3726 doublet ratio. These atomic data yield electron densities systematically lower than those deduced from the \[S ii\] $`\lambda `$6716/$`\lambda `$6731 doublet ratio (see Esteban et al., 2004; García-Rojas et al., 2004). Following the arguments of Copetti & Writzl (2002) and Wang et al. (2004) we have adopted the transition probabilities from Zeippen (1982) and the collision strengths from Pradhan (1976) for the \[O ii\] $`\lambda `$3729/$`\lambda `$3726 doublet ratio, which give electron densities that are in good agreement with the other density indicators. We have derived the \[Fe iii\] density from the intensity of the 6 brightest lines lines, which have errors less than 30 % and seem not to be affected by line blending, together with the computations of Rodríguez (2002). All the computed values of $`n_\mathrm{e}`$ are consistent within the errors (see Table 3).
A weighted mean of $`n_\mathrm{e}`$(O ii), $`n_\mathrm{e}`$(Fe iii), $`n_\mathrm{e}`$(Cl iii) and $`n_\mathrm{e}`$(S ii) has been used to derive $`T_\mathrm{e}`$(N ii), $`T_\mathrm{e}`$(O iii), $`T_\mathrm{e}`$(Ar iii) and $`T_\mathrm{e}`$(S iii), and iterated until convergence. So, for all the species we have adopted $`n_\mathrm{e}`$=310 $`\pm `$ 80 cm<sup>-3</sup>. We have excluded $`n_\mathrm{e}`$(N i) from the average because this ion is representative of the very outer part of the nebula, and probably does not coexist with most of the other ions.
Electron temperatures have been derived from the ratio of CELs of several ions and making use of NEBULAR routines. In the case of $`T_\mathrm{e}`$(S iii) instead of using the collision strengths of Galavís, Mendoza & Zeippen (1995), included by default in NEBULAR, we have used the ones by Tayal & Gupta (1999). The last set of collision strengths gives $`T_\mathrm{e}`$(S iii) values more consistent with the rest of temperature determinations. Table 4 shows the atomic data that we have changed in the last version of NEBULAR.
To obtain $`T_\mathrm{e}`$(O ii) it is necessary to subtract the contribution to $`\lambda \lambda `$7320+7330 due to recombination; Liu et al. (2000) find that the contribution to the intensities of the \[O ii\] $`\lambda \lambda `$ 7319, 7320, 7331, and 7332 lines due to recombination can be fitted in the range 0.5$``$T/10<sup>4</sup>$``$1.0 by:
$$\frac{I_R(7320+7330)}{I(\mathrm{H}\beta )}=9.36\times (T_4)^{0.44}\times \frac{\mathrm{O}^{++}}{\mathrm{H}^+},$$
(1)
where $`T_4`$=$`T`$/10<sup>4</sup>. With this equation we estimate a contribution of approximately 2% to the observed line intensities.
Liu et al. (2000) also determined that the contribution to the intensity of the $`\lambda `$ 5755 \[N ii\] line due to recombination can be estimated from:
$$\frac{I_R(5755)}{I(\mathrm{H}\beta )}=3.19\times (T_4)^{0.30}\times \frac{\mathrm{N}^{++}}{\mathrm{H}^+},$$
(2)
in the range 0.5$``$ $`T`$/10<sup>4</sup>$``$2.0. We have obtained a contribution of recombination of about 0.5%, that does not affect significantly the temperature determination.
Figure 2 shows the spectral regions near the Balmer and the Paschen limits. The discontinuities can be easily appreciated. They are defined as $`I_c(Bac)=I_c(\lambda 3646^{})I_c(\lambda 3646^+)`$ and $`I_c(Pac)=I_c(\lambda 8203^{})I_c(\lambda 8203^+)`$ respectively. The high spectral resolution of the spectra permits to measure the continuum emission in zones very near de discontinuity, minimizing the possible contamination of other continuum contributions. We have obtained power-law fits to the relation between $`I_c(Bac)/I(Hn)`$ or $`I_c(Pac)/I(Pn)`$ and $`T_\mathrm{e}`$ for different $`n`$ corresponding to different observed lines of both series. The emissivities as a function of electron temperature for the nebular continuum and the H i Balmer and Paschen lines have been taken from Brown & Mathews (1970) and Storey & Hummer (1995) respectively. The $`T_e(Bac)`$ adopted is the average of the values using the lines from $`H\alpha `$ to H 10 (the brightest ones). In the case of $`T_e(Pac)`$, the adopted value is the average of the individual temperatures obtained using the lines from P 7 to P 13 (the brightest lines of the series), excluding P 8 and P 10 because their intensity seems to be affected by sky absorption.
Peimbert, Peimbert & Luridiana (2002) developed a method to derive the helium temperature, $`T_\mathrm{e}`$(He i), in the presence of temperature fluctuations. Assuming a 2-zone ionization scheme and the formulation of Peimbert, Peimbert & Luridiana (2002) we have derived $`T_\mathrm{e}`$(He i)=8750 $`\pm `$ 500 K, which is highly consistent with $`T_\mathrm{e}`$(H i) assumed above.
We have assumed a 2-zone ionization scheme for the derivation of ionic abundances (see § 5). We have adopted the average of electron temperatures obtained from \[N ii\] and \[O ii\] lines as representative for the low ionization zone, and the average of the values obtained from \[O iii\], \[Ar iii\] and \[S iii\] lines for the high ionization zone (see Table 3).
## 5 Ionic Abundances
### 5.1 He<sup>+</sup> abundance
We have measured 50 He i emission lines identified in our spectra. These lines arise mainly from recombination but they can be affected by collisional excitation and self-absorption effects. We have determined the He<sup>+</sup>/H<sup>+</sup> ratio using the effective recombination coefficients of Storey & Hummer (1995) for H i and those of Smits (1996) and Benjamin, Skillman & Smits (1999) for He i. The collisional contribution was estimated from Sawey & Berrington (1993) and Kingdon & Ferland (1995), and the optical depths in the triplet lines were derived from the computations by Benjamin, Skillman & Smits (2002). From a maximum likelihood method (e.g. Peimbert, Peimbert & Ruiz, 2000), using $`n_\mathrm{e}`$=310 $`\pm `$ 80 cm<sup>-3</sup> and $`T`$(O ii+iii)=9600 $`\pm `$ 450 K (see § 6), we have obtained He<sup>+</sup>/H<sup>+</sup>=0.0795 $`\pm `$ 0.0009, $`\tau _{3889}`$=2.52 $`\pm `$ 0.44, and *$`t^2`$*=0.034 $`\pm `$ 0.010. In Table 5 we have included the He<sup>+</sup>/H<sup>+</sup> ratios we have obtained for the individual He i lines not affected by line blending and with the highest signal-to-noise ratio. We have excluded He i $`\lambda `$5015 for the same reasons outlined by Esteban et al. (2004). We have done a $`\chi ^2`$ optimization of the values in the table, and we have obtained a $`\chi ^2`$ parameter of 8.15, which indicates a reasonable goodness of the fit for a system with nine degrees of freedom.
### 5.2 Ionic Abundances from CELs
Ionic abundances of N<sup>+</sup>, O<sup>+</sup>, O<sup>++</sup>, Ne<sup>++</sup>, S<sup>+</sup>, S<sup>++</sup>, Cl<sup>+</sup>, Cl<sup>++</sup>, Ar<sup>++</sup> and Ar<sup>3+</sup> have been determined from CELs, using the IRAF package NEBULAR (except for Cl<sup>+</sup>, see García-Rojas et al., 2004). Additionally, we have determined the ionic abundances of Fe<sup>++</sup> following the methods and data discussed in García-Rojas et al. (2004). Ionic abundances are listed in Table 6 and correspond to the mean value of the abundances derived from all the individual lines of each ion observed (weighted by their relative intensities).
To derive the abundances for $`t^2`$ = 0.038 (see § 6) we used the abundances for $`t^2`$=0.00 and the formulation of Peimbert (1967) and Peimbert & Costero (1969) for $`t^2>`$0.00. To derive abundances for other $`t^2`$ values it is possible to interpolate or extrapolate the values presented in Table 6.
Many \[Fe ii\] lines have been identified in our spectra, but all of them are severely affected by fluorescence effects (Rodríguez, 1999; Verner et al., 2000). The only \[Fe ii\] line in the spectral range 3100 $`\mathrm{\AA }`$ to 10400 $`\mathrm{\AA }`$ which is not affected by fluorescence effects is the \[Fe ii\] $`\lambda 8617`$ $`\mathrm{\AA }`$ line, but unfortunately it is in one of our observational gaps. We have measured \[Fe ii\] $`\lambda 7155`$, a line which is not much affected by fluorescence effects (Verner et al., 2000), but it has a high observational error. Therefore, it was no possible to derive a reliable value of the Fe<sup>+</sup>/H<sup>+</sup> ratio. The calculations for Fe<sup>++</sup> have been done with a 34 level model-atom that uses the collision strengths of Zhang (1996) and the transition probabilities of Quinet (1996). We have used 6 \[Fe iii\] lines that do not seem to be affected by line-blending and with errors less than 30 %. We find a mean value and a standard deviation of Fe<sup>++</sup>/H<sup>+</sup>=(1.115 $`\pm `$ 0.153)$`\times `$10<sup>-7</sup>. Adding errors in $`T_\mathrm{e}`$ and $`n_\mathrm{e}`$ we finally obtain 12+log(Fe<sup>++</sup>/H<sup>+</sup>)=5.05 $`\pm `$ 0.06. The value of the Fe<sup>++</sup> abundance for t$`{}_{}{}^{2}>`$0.00 is also shown in Table 6.
### 5.3 Ionic Abundances from Recombination Lines
We have measured a large number of permitted lines of heavy element ions such as O i, O ii, C i, C ii, S ii, N i, N ii, Ar i, Si i, Si ii, and Fe i most of them detected for the first time in S 311. Those permitted lines produced by recombination can give accurate determinations of ionic abundances because their intensities depend weakly on electron temperature and density. Unfortunately most of the permitted lines are affected by fluorescence effects or are blended with telluric emission lines making their intensities unreliable; also Ruiz et al. (2003) have shown that, to determine the abundances, it is important to measure all the lines of a multiplet, because for low densities there could be an anomalous distribition of the line intensities within the multiplet. Detailed discussions on the mechanism of formation of the permitted lines are in Esteban et al. (1998, and references therein); Esteban et al. (2004, and references therein).
We have been able of measuring ionic abundance ratios of O<sup>++</sup>/H<sup>+</sup> and C$`{}_{}{}^{++}/`$H<sup>+</sup> from pure recombination O ii and C ii lines respectively, from multiplet 1 for O ii (see Figure 3) and from multiplet 6 for C ii \[see Figure 1 of Esteban et al. (2005)\]. We have computed the abundances for $`T_\mathrm{e}`$(High)=9050 K and $`n_\mathrm{e}`$=310 cm<sup>-3</sup>. Atomic data and methodology for the C abundance are the same as in García-Rojas et al. (2004). For the Multiplet 1 of O II it has been shown that for densities $`n_e`$ $`<`$ 10000 cm<sup>-3</sup> the upper levels of the transition are not in LTE, and if one uses only one line to determine the abundances one can have errors as large as a factor of 4 (Ruiz et al., 2003); instead we used the prescription presented by Peimbert, Peimbert & Ruiz (2005) to calculate those populations; these abundances show very good agreement among themselves and with the abundance determined using the sum of all the lines, “Sum”, that is not expected to be affected by this effect. Tables 7 and 8 show the abundance ratios.
## 6 Temperature variations
It is well known that under the assumption of a constant temperature, RLs of heavy element ions yield higher abundance values relative to hydrogen than CELs in H ii regions (e.g. Peimbert, Storey & Torres-Peimbert, 1993; Esteban et al., 1998; Esteban, 2002; Esteban et al., 2004; García-Rojas et al., 2004; Tsamis et al., 2003, and references therein). Torres-Peimbert, Peimbert, & Daltabuit (1980) proposed the presence of spatial temperature fluctuations (parametrized by *$`t^2`$*) as the cause of this discrepancy, because CELs and RLs emissivities have different dependences on the electron temperature. On the other hand, Peimbert (1971) proposed that there is a dichotomy between $`T_\mathrm{e}`$ derived from the \[O iii\] lines and from the hydrogen recombination continuum discontinuities, which is strongly correlated with the discrepancy between CEL and RL abundances, so the comparison between electron temperatures obtained from both methods is an additional indicator of *$`t^2`$*. A complete formulation of temperature fluctuations has been developed by Peimbert (1967), Peimbert & Costero (1969) and Peimbert (1971). We have assumed a two-zone ionization scheme, and we have followed the re-formulation of Peimbert, Peimbert & Ruiz (2000) and Peimbert, Peimbert & Luridiana (2002) to derive the value of *$`t^2`$* comparing $`T_\mathrm{e}`$($`Bac`$) and $`T_\mathrm{e}`$($`Pac`$) with the combination of $`T_\mathrm{e}`$(\[O ii\]) and $`T_\mathrm{e}`$(\[O iii\]), *$`t^2`$*($`BacFL`$) and *$`t^2`$*($`PacFL`$), respectively, using equation (A1) of Peimbert, Peimbert & Luridiana (2002). Table 9 shows the different *$`t^2`$* values obtained as well as the adopted value, *$`t^2`$*=0.038 $`\pm `$ 0.007 which is the weighted average of O<sup>++</sup> and He<sup>+</sup> values which are rather consistent and show the lowest uncertainties.
On the other hand, as it can be seen from Table 6, *$`t^2`$*($`BacFL`$) and *$`t^2`$*($`PacFL`$) are significantly lower than the other values. One possible explanation is that the nebular continuum could be affected by dust scattered light. To explore that possibility we have derived the atomic continua contribution, which includes the free-free and free-bound continua of the hydrogen and helium atoms and the two-quantum continuum, from the computations of Brown & Mathews (1970) for $`T_\mathrm{e}`$=7620 K, $`n_\mathrm{e}`$=310 cm<sup>-3</sup> and He<sup>+</sup>/H<sup>+</sup>=0.0795 (see section 5.1). The temperature of 7620 K we have assumed is that which implies *$`t^2`$*($`BacFL`$) and *$`t^2`$*($`PacFL`$) equal to the final adopted value. Table 10 shows the observed and the expected atomic continua, and the derived scattered light contribution. From these data it is easy to derive the contribution of dust scattered light to the continuum near the Balmer lines and the Balmer and Paschen jumps. Including observational uncertainties, we have derived that a contribution between 10% and 30% to the Balmer jump of the scattered continuum integrated light is enough to explain the high $`T_\mathrm{e}`$($`Bac`$) obtained from the discontinuity and, therefore, the lower *$`t^2`$*($`BacFL`$). The shape of the spectral distribution of the scattered light is then similar to a B3 V star. Assuming that the ionising star of S 311 –HD 64315– is a main sequence O6e, we have modelled the optical flux distribution using FASTWIND, an spherically symmetric, NLTE model atmosphere code (Santolaya-Rey, Puls, & Herrero, 1997; Puls et al., 2005) and the corresponding $`T_{\mathrm{eff}}`$ and $`logg`$ derived from Martins, Schaerer & Hillier (2005) calibrations. From this model, we have obtained that the stellar Balmer jump is about 10% of the stellar continuum. Anyway, our slit position is far from HD 64315 (see Figure 1) and other late O and early B stars may contribute to the continuum scattered light, so the contribution of the discontinuities would be even higher, and therefore, the nebular temperature determination would decrease. In fact, Feinstein, & Vázquez (1989) list a B8.5 star near our slit position (their star labelled as NGC 2467–12), whose position is also indicated in Figure 1. It can be seen that, as expected, the continuum scattered light increases monotonically towards the blue. A more detailed study of the properties of the dust in S311 (albedo, reddening and geometrical distribution) is needed to solve this problem in an appropiate way, but this is outside the scope of this paper.
Recently, Zhang et al. (2005) have computed $`T_\mathrm{e}`$(He i) for 48 planetary nebulae from He i recombination line ratios (He i$`\lambda `$7281/$`\lambda `$6678). These authors have found that temperature fluctuations do not predict the general behaviour of the $`T_\mathrm{e}`$(He i) vs. $`T_\mathrm{e}`$(H i) diagram. Using the expression given by Zhang et al. (2005) we have derived $`T_\mathrm{e}`$(He i)=7960 $`\pm `$ 1000 K, for S 311, which is lower, but consistent within the errors with our $`T_\mathrm{e}`$(He i) and $`T_\mathrm{e}`$(H i). In Figure 4 we have compared results from Zhang et al. (2005) for planetary nebulae (PNe) with our results for H ii regions, in which we have included S311 (this paper), NGC 3576 (García-Rojas et al., 2004) and still unpublished echelle VLT spectra of M8, M17, M20, M16 and NGC 3603. It can be seen that most of the H ii regions, except for the Orion nebula (Esteban et al., 2004) and 30 Dor (Peimbert, 2003), do not follow the behaviour of the bulk of PNe and do not contradict the temperature fluctuations paradigm. Zhang et al. (2005) solve the problem of the anomalous position of PNe in the $`T_e`$(He i) vs. $`T_e`$(H i) diagram proposing the presence of a small amount of H- deficient material for most of the sample nebulae, in the context of the scenario of H- deficient inclusions constructed by (Pequignot et al.2003). That scenario seems to explain successfully the large abundance discrepancies reported in some PNe. However, Figure 4 suggests that the behaviour of H ii regions is qualitatively different to that of PNe and that the position of most H ii regions in the diagram is consistent with the classical *$`t^2`$* paradigm. This result is in agreement with the suggestions outlined by Esteban (2002) who propose that the processes producing abundance discrepancies in H ii regions and PNe -at least the extreme cases of PNe- could be different. In the case of H ii regions the observational evidences are still consistent with the *$`t^2`$* scheme, while in the case of the extreme PNe this scheme seems to fail.
## 7 Total abundances
We have adopted a set of ionization correction factors (ICFs) to correct for the unseen ionization stages and then derive the total gaseous abundances of the elements we have studied. We have adopted the ICF scheme used by García-Rojas et al. (2004) for all the elements except for carbon, neon, chlorine and iron.
The absence of He ii lines in our spectra indicates that He<sup>++</sup>/H<sup>+</sup> is negligible. However, the total helium abundance has to be corrected for the presence of neutral helium. Based on the ICF(He<sup>0</sup>) given by Peimbert, Torres-Peimbert & Ruiz (1992), and our data ICF(He<sup>0</sup>) amounts to 1.22 $`\pm `$ 0.05 for *$`t^2`$* = 0.00 and 1.16 $`\pm `$ 0.04 for *$`t^2`$* $`>`$ 0.00.
We have derived the O/H ratio both from CELs and from the combination of O<sup>++</sup>/H<sup>+</sup> ratio from RLs and O<sup>+</sup>/H<sup>+</sup> ratio from CELs and the assumed *$`t^2`$*.
For carbon we have adopted the ICF derived from photoionization models by Garnett et al. (1999).
For neon the ICF proposed by Peimbert & Costero (1969) given by:
$$\frac{N(\mathrm{Ne})}{N(\mathrm{H})}=\left(\frac{N(\mathrm{O}^+)+N(\mathrm{O}^{++})}{N(\mathrm{O}^{++})}\right)\frac{N(\mathrm{Ne}^{++})}{N(\mathrm{H}^+)}.$$
(3)
has been generally used. Nevertheless this ICF underestimates the Ne/H abundance for nebulae of low degree of ionization because a considerable fraction of Ne<sup>+</sup> coexists with O<sup>++</sup> (see , Torres-Peimbert, & Peimbert1977; Peimbert, Torres-Peimbert & Ruiz, 1992). For S 311 based on the O<sup>+</sup>/O ratio and the data by (Torres-Peimbert, & Peimbert1977), we estimate that the ICF(Ne) should be 0.4 $`\pm `$ 0.1 dex higher than that provided by the previous equation.
We have measured lines of two ionization stages of chlorine: Cl<sup>+</sup> and Cl<sup>++</sup>. The Cl abundance has been assumed to be equal to the sum of these ionic abundances without taking into account Cl<sup>3+</sup> fraction. This assumption seems reasonable taking into account the small Cl<sup>3+</sup>/Cl<sup>++</sup> ratio found for M17 ($``$0.03, see Esteban et al., 1999a), for the Orion nebula ($``$0.04, see Esteban et al., 2004), and for NGC3576 ($``$0.02, see García-Rojas et al., 2004), and the lower ionization degree of S 311 with respect to those nebulae.
We have measured lines of two stages of ionization of iron: Fe<sup>+</sup> and Fe<sup>++</sup>, but in § 5.2 we have shown that Fe<sup>+</sup>/H<sup>+</sup> ratio is not reliable. Recently, Rodríguez & Rubin (2005) have derived an ICF from a least-squares fit to the results of a set of models in which it is represented $`\chi `$(O<sup>+</sup>)/$`\chi `$(Fe<sup>++</sup>), the ratio of ionization fractions of O<sup>+</sup> and Fe<sup>++</sup>, respectively, as a function of the degree of ionization given by O<sup>+</sup>/O<sup>++</sup>. The ionization correction scheme they have derived is as follows:
$$\frac{N(\mathrm{Fe})}{N(\mathrm{H})}=0.9\left[\frac{N(\mathrm{O}^+)}{N(\mathrm{O}^{++})}\right]^{0.08}\times \frac{N(\mathrm{Fe})^{++}}{N(\mathrm{O})^+}\times \frac{N(\mathrm{O})}{N(\mathrm{H})}.$$
(4)
In Table 11 we show the total abundances obtained for our slit position in S 311 for *$`t^2`$*=0.00 and *$`t^2`$*=0.038 $`\pm `$ 0.007.
## 8 Deuterium Balmer lines
We have detected, for the first time in an H ii region outside the solar circle, the four brightest deuterium Balmer lines, D$`\alpha `$, D$`\beta `$, D$`\gamma `$ and D$`\delta `$ as very weak lines in the blue wings of the corresponding H i Balmer lines (see Figure 5). The apparent shift in radial velocity of these weak lines with respect to the hydrogen ones is $``$82.7 km s<sup>-1</sup> (see Table 12), which is in excellent agreement with the isotopic shift of deuterium, $``$81.6 km s<sup>-1</sup>. We have discarded these weak features as high velocity components of hydrogen for the following reasons:
* We have not found blue-shifted features in the wings of the brightest \[N ii\], \[O ii\] and \[O iii\] lines, indicating that the faint features in the blue wings of H i lines can not come from emission of high-velocity ionized material.
* They are narrower (FWHM $`10`$ km s<sup>-1</sup>) than hydrogen lines (FWHM $`20`$ km s<sup>-1</sup>). FWHM has been derived from gaussian fits, after correcting from the underlying blue wing of the corresponding Balmer line, and after quadratic subtraction of the instrumental point-spread function. Although the relatively low velocity resolution of our spectra, which is not enough to derive precisely the value of the thermal width of the deuterium Balmer lines (colons in the FWHM indicates high uncertainties, and the low values of the FWHM of the deuterium lines are because deuterium line widths are of the order of the instrumental one), it is sufficient to compare qualitatively with the value of the width of hydrogen Balmer lines (see Table 12). This result supports the idea that deuterium lines arise from a cold material with smaller thermal velocity, probably the photon dominated region or PDR (Hébrard et al., 2000a).
The detection and identification of the deuterium Balmer lines in an H ii region were first reported by Hébrard et al. (2000a) from VLT/UVES data; they confirmed the detection and identification of deuterium Balmer lines (up to D$`\eta `$) in the Orion nebula. Subsequently, Hébrard et al. (2000b) published the detection and identification of at least D$`\alpha `$ and D$`\beta `$ in four additional H ii regions (M8, M16, M20 and DEM S 103 in SMC), and confirmed fluorescence as the main excitation mechanism of D i, recombination being negligible. The D i/H i ratios presented in Table 12 correspond to the intensity ratios and are upper limits to the abundance ratio because in addition to recombination the D i line intensities include the fluorescence contribution that is considerably larger than the recombination one.
On the other hand, O’Dell, Ferland & Henney (2001) presented observations and a model for the emission of the deuterium lines in Orion, and concluded that they are produced by fluorescent excitation of the upper energy states by the far-UV radiation of the ionising star. In S 311 the D$`\alpha `$/H$`\alpha `$ and D$`\beta `$/H$`\beta `$ ratios are somewhat larger than in the case of the Orion nebula. In the light of the model outlined by O’Dell, Ferland & Henney (2001), and taking into account that the spectral types of the ionizing stars of both nebulae are similar, this can be due to an additional contribution of UV radiation from other nearby cooler stars or to a lower UV grain extinction in S 311. On the other hand, the comparison of the Balmer decrements of the hydrogen and deuterium lines observed in our spectra follow closely the standard fluorescence model by O’Dell, Ferland & Henney (2001, see their Figure 13) for the Orion nebula.
## 9 Discussion
Few optical spectrophotometric studies of S 311 have been published in the literature. The most complete are those presented by Hawley (1978) and Peimbert, Torres-Peimbert & Rayo (1978). Unfortunately, these papers do not study the same slit position as ourselves (see § 3), and the ionic abundances they have obtained are also very different. In fact, the ratio O<sup>+</sup>/O<sup>++</sup>, which is an indication of the ionization degree of the nebula is very similar in these works ranging from 0.89 to 1.00; on the contrary, our O<sup>+</sup>/O<sup>++</sup>=2.8 (for *$`t^2`$*=0.00) is much larger, pointing out that our slit position is nearer the ionization front of the nebula, where the O<sup>+</sup>/O<sup>++</sup> ratio increases significantly. In fact, as it can be seen in Figure 1, our slit position is on the brightest part of the nebula, coinciding with a filament or an ionization front.
On the other hand, we can make comparisons with spectrophotometric data of Shaver et al. (1983). Their slit position 2 is closer to ours. These authors only derived ionic abundances for 5 species: He<sup>+</sup>, O<sup>++</sup>, N<sup>+</sup>, S<sup>+</sup> and S<sup>++</sup>. They assumed temperatures much lower than ours, about 800 K lower for the high ionization species, and more than 1500 K lower for low ionization ones. This implies, in general, an overestimation of the abundances with respect to us, except for O<sup>++</sup>, which would be underestimated. We have derived the abundances from Shaver et al. (1983) line intensities using the atomic data listed in Table 4 and we have obtained a good agreement between them and our results, except for S<sup>++</sup>/H<sup>+</sup> which is 0.21 dex lower than our derived abundance. Although Shaver et al. (1983) did not quote errors in the line intensities, this fact is possibly due to uncertainties in the measured flux of the \[S iii\]$`\lambda `$6312 line. Taking into account the faintness of this line in the spectrum of S 311 (1% of H$`\beta `$), we have estimated their error in the flux measurement of \[S iii\]$`\lambda `$6312 in about 50%, which is enough to explain the difference between the derived abundances. For the He<sup>+</sup>/H<sup>+</sup> ratio the highest temperature assumed and to consider temperature fluctuations make us to derive a value 0.13 dex higher than the one derived by Shaver et al. (1983).
Obtaining deep good-quality spectra of H ii regions located outside the solar circle is of great importance for deriving radial abundance gradients in the Galactic disk. H ii regions in this part of the Galaxy are scarce and usually very faint (Russeil, 2003). In fact, part of the abundance data presented in this paper (C, N and O abundances) have been included in recent works devoted to the calculation and modelling of abundance gradients (Esteban et al., 2005; , Carigi et al.2005b).
In Table 11 we compare S 311 and Orion nebula gas abundances and the solar values. For the Sun: He comes from Christensen-Dalsgaard (1998) and the rest of elements from Asplund, Grevesse & Sauval (2005). As expected, the total abundances of S 311 are somewhat lower than those of the Orion nebula and the Sun because of the existence of Galactic radial abundance gradients and the larger Galactocentric distance of S 311. Interestingly, the Fe abundance is very different in both nebulae, this could be due to their different dust-depletion factors.
It is important to appreciate that nebulae are 3-D objects and the spectrum really corresponds to the integral of the emission contained in the column of gas covered by the slit area. This implies that the emission comes from a range of densities, degrees of ionization, temperatures and even extinctions within the column. However, our observations are limited to a small area covering a bright rim, probably coincident with a filament or an ionization front, precisely the brightest part of the nebula, where we expect the 3-D effects should be minimum. Nevertheless, it would be interesting to make realistic 3-D (Ercolano et al., 2003) or pseudo 3-D models (Morisset, Stasińska & Peña, 2005) for Galactic H ii regions, combined with medium-high resolution long-slit spectroscopic data and narrow-band imaging in different emission line filters. On the other hand, 3-D effects should be much more severe in the case of extragalactic HII regions, where a small slit area covers a enormous volume of gas (several orders of magnitude larger than in Galactic H ii regions). In a very recent paper, Pilyugin (2005) has found that our data for S 311 do not fit his strong-line diagnostic diagrams for the empirical derivation of chemical abundances. This deviation is clearly due to the fact that the line fluxes accepted by the slit are not representative of the nebula as a whole. This effect has to be taken into account when spatially resolved observations of small zones of a nebula are used to apply empirical methods for the derivation of abundances.
## 10 Conclusions
We present echelle spectroscopy in the 3100–10400 $`\mathrm{\AA }`$ range of the brightest zone of the Galactic H ii region S 311. We have measured the intensity of 263 emission lines. This is the deepest spectra ever taken for a Galactic H ii region located outside the solar circle.
We have derived the physical conditions of S 311 making use of several line intensities and continuum ratios. The chemical abundances have been derived using CELs for a large number of ions. We have determined also, for the first time in S 311, the C<sup>++</sup> and O<sup>++</sup> abundances from RLs. We have obtained an average *$`t^2`$*=0.038 $`\pm `$ 0.007 both by comparing the O<sup>++</sup> ionic abundance derived from RLs to those derived from CELs and by applying a chi-squared method which minimizes the dispersion of He<sup>+</sup>/H<sup>+</sup> ratio from individual lines. The adopted average value has been used to derive the abundances determined from CELs.
We have compared $`T_\mathrm{e}`$(He i) vs. $`T_\mathrm{e}`$(H i) for S 311 and other Galactic H ii regions finding that the *$`t^2`$* paradigm is consistent with the observed behaviour of most of the objects, in contrast with what has been found for PNe. This result suggests that the process or processes that produce the abundance discrepancies in both kinds of objects could be different.
We have detected four deuterium Balmer emission lines in S 311. These are the first detections of these lines in this object. Comparison with previous observations and with models lead us to support that fluorescence seems to be the most probable excitation mechanism of these lines.
## Acknowledgments
This work is based on observations collected at the European Southern Observatory, Chile, proposal number ESO 68.C-0149(A). We thank the annonymous referee for useful comments. We would like to thank A. R. López-Sánchez for providing us the H$`\alpha `$ image of S 311. JGR would like to thank S. Simón-Díaz for providing us the results of stellar modelling for HD 64315, and E. Pérez-Montero for fruitful discussions and excellent suggestions. JGR and CE would like to thank the members of the Instituto de Astronomía, UNAM, for their always warm hospitality. This work has been partially funded by the Spanish Ministerio de Ciencia y Tecnología (MCyT) under projects AYA2001-0436 and AYA2004-07466. MP received partial support from DGAPA UNAM (grant IN 114601). MTR received partial support from FONDAP(15010003) and Fondecyt(1010404). MR acknowledges support from Mexican CONACYT project J37680-E.
|
warning/0506/nucl-th0506043.html
|
ar5iv
|
text
|
# Random-phase approximation based on relativistic point-coupling models
## I Introduction
The covariant self-consistent mean-field approach to the nuclear many-body problem has reached a mature stage BHR.03 ; VALR.05 . Models based on the relativistic mean-field approximation are successfully employed in the description of structure phenomena not only in medium-heavy and heavy stable nuclei, but also in regions of exotic nuclei far from the line of $`\beta `$-stability and close to the nucleon drip-lines.
Most applications have been based on the finite-range meson-exchange representation of the relativistic mean-field (RMF) theory, in which the nucleus is described as a system of Dirac nucleons coupled to exchange mesons and the electromagnetic field through an effective Lagrangian. A medium dependence can be introduced either by including non-linear meson self-interaction terms in the Lagrangian BB.77 , or by assuming an explicit density dependence for the meson-nucleon couplings. The former approach has been adopted in the construction of several very successful phenomenological RMF interactions, for instance, the popular NL3 LKR.97 parameterization of the effective Lagrangian. In the latter case, the density dependence of the meson-nucleon vertex functions can be determined either from microscopic Dirac-Brueckner calculations in nuclear matter FLW.95 ; JL.98 , or it can be completely phenomenological TW.99 ; NVF.02 , with parameters adjusted to data on finite nuclei and empirical properties of symmetric and asymmetric nuclear matter. A series of recent studies has shown that, when compared with standard non-linear meson-exchange effective Lagrangians, effective interactions with an explicit density dependence of the meson-nucleon couplings are more flexible and provide an improved description of asymmetric nuclear matter, neutron matter and finite nuclei far from stability.
An alternative representation of the self-consistent relativistic mean-field approach to nuclear structure is formulated in terms of point-coupling (contact) nucleon-nucleon interactions. When applied in the description of finite nuclei, the framework of relativistic mean-field point-coupling (RMF-PC) models MM.89 ; MNH.92 ; Hoch.94 ; FML.96 ; RF.97 ; BMM.02 produces results that are comparable to those obtained in the meson-exchange picture. In principle, the point-coupling approach is more general and the interaction terms are not restricted by the constraints imposed by the finite range of meson exchange. Of course, also in the case of contact interactions, medium effects can be taken into account by the inclusion of higher order interaction terms, for instance, six-nucleon vertices $`(\overline{\psi }\psi )^3`$, and eight-nucleon vertices $`(\overline{\psi }\psi )^4`$ and $`[(\overline{\psi }\gamma _\mu \psi )(\overline{\psi }\gamma ^\mu \psi )]^2`$, or it can be encoded in the effective couplings, i.e. in the strength parameters of the interaction in the isoscalar and isovector channels. Several studies of the RMF-PC framework have been reported over the last ten years, but it is only recently that reliable and accurate phenomenological parameterizations have been adjusted and applied in the description of ground state properties of finite nuclei on a quantitative level. In particular, based on an extensive multiparameter $`\chi ^2`$ minimization procedure, in Ref. BMM.02 Bürvenich et al. have adjusted the PC-F1 set of coupling constants for an effective point-coupling Lagrangian with higher order interaction terms. The PC-F1 interaction has been tested in the calculation of ground state properties of a large number of spherical and deformed nuclei, and the results are on the level of the best meson-exchange effective interactions.
Concepts of effective field theory and density functional theory have been used to derive a microscopic relativistic point-coupling model of nuclear many-body dynamics constrained by in-medium QCD sum rules and chiral symmetry FKV.03 ; FKV.04 ; VW.04 . The effective Lagrangian is characterized by density-dependent coupling strengths, determined by chiral one- and two-pion exchange and by QCD sum rule constraints for the large isoscalar nucleon self-energies that arise through changes of the quark condensate and the quark density at finite baryon density. This approach has been tested in the analysis of the equations of state for symmetric and asymmetric nuclear matter, and of bulk and single-nucleon properties of finite nuclei. In comparison with purely phenomenological mean-field approaches, the built-in QCD constraints and the explicit treatment of pion exchange restrict the freedom in adjusting parameters and functional forms of density-dependent couplings.
A number of studies have shown that, both for finite-range meson-exchange and for point-coupling mean-field models, the empirical data set of ground-state properties of finite nuclei can only determine six or seven parameters in the general expansion of the effective Lagrangian in powers of the fields and their derivatives FS.00b . The influence of the adjustment procedure and of the choice of ground-state data on the properties and predictive power of the relativistic mean-field model with point-couplings has recently been investigated in Ref. BMR.04 . While virtually all phenomenological relativistic effective interactions have been adjusted to empirical properties of symmetric and asymmetric nuclear matter, and to ground-state properties of a set of spherical nuclei, in Ref. NVR.02 it has been shown that a comparison of relativistic RPA results on multipole giant resonances with experimental excitation energies can provide additional constrains on the parameters that characterize the isoscalar and isovector channels of the effective interactions. Data on giant resonances have been taken into account in the recent adjustment of a new improved relativistic mean-field effective interaction with explicit density dependence of the meson-nucleon couplings LNV.05 .
Relativistic RPA calculations have been performed since the early eighties, but it is only more recently that non-linear meson self-interaction terms or density-dependent meson-nucleon couplings have been included in the RRPA framework Ma.97 ; VWR.00 ; NVR.02 . As in the case of ground-state properties, the inclusion of a medium dependence in the residual interaction is necessary for a quantitative description of collective excited states. Another essential feature of the RRPA is the fully consistent treatment of the Dirac sea of negative energy states. In addition to the usual particle-hole pairs, the RRPA configuration space must also include pair-configurations built from positive-energy states occupied in the ground-state solution, and empty negative-energy states in the Dirac sea. These configurations ensure not only current conservation and the decoupling of the spurious states DF.90 , but also a quantitative comparison with the experimental excitation energies of giant resonances Rin.01 . Collective excitations in open-shell nuclei can be analyzed with the relativistic quasiparticle random-phase approximation (RQRPA), which in Ref. Paar.03 has been formulated in the canonical single-nucleon basis of the relativistic Hartree-Bogoliubov (RHB) model.
Some of the recent applications of the RRPA include studies of nuclear compression modes VWR.00 ; Ma.01 ; Pie.01 , of multipole giant resonances and low-lying collective states in spherical nuclei Ma.02 , of the evolution of the low-lying isovector dipole response in nuclei with a large neutron excess VPRLa.01 ; VPRLb.01 , and of the toroidal dipole response VPNR.02 . The RHB+RQRPA approach has been employed in the investigation of the multipole response of weakly bound neutron-rich nuclei, and of spin-isospin excitations in finite nuclei Paar.03 ; Paar.04 .
In this work we introduce the RPA based on the relativistic mean-field framework with point-coupling interactions. Illustrative calculations of excitation energies of giant resonances in spherical nuclei will test the PC-F1 effective interaction BMM.02 . The RRPA matrix equations are derived in Sec. II. Isoscalar and isovector giant resonances in spherical nuclei are analyzed in Sec. III. The results are summarized in Sec. IV.
## II Random-phase approximation based on the point-coupling relativistic mean-field model
The relativistic point-coupling Lagrangian is built from basic densities and currents bilinear in the Dirac spinor field $`\psi `$ of the nucleon:
$$\overline{\psi }𝒪_\tau \mathrm{\Gamma }\psi ,𝒪_\tau \{1,\tau _i\},\mathrm{\Gamma }\{1,\gamma _\mu ,\gamma _5,\gamma _5\gamma _\mu ,\sigma _{\mu \nu }\}.$$
(1)
Here $`\tau _i`$ are the isospin Pauli matrices and $`\mathrm{\Gamma }`$ generically denotes the Dirac matrices. The interaction terms of the Lagrangian are products of these bilinears. Although a general effective Lagrangian can be written as a power series in the currents $`\overline{\psi }𝒪_\tau \mathrm{\Gamma }\psi `$ and their derivatives, it is well known from numerous applications of relativistic mean-field models that properties of symmetric and asymmetric nuclear matter, as well as empirical ground state properties of finite nuclei, constrain only the isoscalar-scalar (S), the isoscalar-vector (V), the isovector-vector (TV), and to a certain extent the isovector-scalar (TS) channels. Here we consider a model with four-, six-, and eight-fermion point couplings (contact interactions) BMM.02 , defined by the Lagrangian density:
$$\begin{array}{ccc}\hfill & =& ^{\mathrm{free}}+^{4\mathrm{f}}+^{\mathrm{hot}}+^{\mathrm{der}}+^{\mathrm{em}},\hfill \\ ^{\mathrm{free}}\hfill & =& \overline{\psi }(\mathrm{i}\gamma _\mu ^\mu m)\psi ,\hfill \\ ^{4\mathrm{f}}\hfill & =& \frac{1}{2}\alpha _\mathrm{S}(\overline{\psi }\psi )(\overline{\psi }\psi )\frac{1}{2}\alpha _\mathrm{V}(\overline{\psi }\gamma _\mu \psi )(\overline{\psi }\gamma ^\mu \psi )\hfill \\ & & \frac{1}{2}\alpha _{\mathrm{TS}}(\overline{\psi }\stackrel{}{\tau }\psi )(\overline{\psi }\stackrel{}{\tau }\psi )\frac{1}{2}\alpha _{\mathrm{TV}}(\overline{\psi }\stackrel{}{\tau }\gamma _\mu \psi )(\overline{\psi }\stackrel{}{\tau }\gamma ^\mu \psi ),\hfill \\ ^{\mathrm{hot}}\hfill & =& \frac{1}{3}\beta _\mathrm{S}(\overline{\psi }\psi )^3\frac{1}{4}\gamma _\mathrm{S}(\overline{\psi }\psi )^4\frac{1}{4}\gamma _\mathrm{V}[(\overline{\psi }\gamma _\mu \psi )(\overline{\psi }\gamma ^\mu \psi )]^2,\hfill \\ ^{\mathrm{der}}\hfill & =& \frac{1}{2}\delta _\mathrm{S}(_\nu \overline{\psi }\psi )(^\nu \overline{\psi }\psi )\frac{1}{2}\delta _\mathrm{V}(_\nu \overline{\psi }\gamma _\mu \psi )(^\nu \overline{\psi }\gamma ^\mu \psi )\hfill \\ & & \frac{1}{2}\delta _{\mathrm{TS}}(_\nu \overline{\psi }\stackrel{}{\tau }\psi )(^\nu \overline{\psi }\stackrel{}{\tau }\psi )\frac{1}{2}\delta _{\mathrm{TV}}(_\nu \overline{\psi }\stackrel{}{\tau }\gamma _\mu \psi )(^\nu \overline{\psi }\stackrel{}{\tau }\gamma ^\mu \psi ),\hfill \\ ^{\mathrm{em}}\hfill & =& eA_\mu \overline{\psi }[(1\tau _3)/2]\gamma ^\mu \psi \frac{1}{4}F_{\mu \nu }F^{\mu \nu }.\hfill \end{array}$$
(2)
Vectors in isospin space are denoted by arrows, and bold-faced symbols will indicate vectors in ordinary three-dimensional space. In addition to the free nucleon Lagrangian $`_{\mathrm{free}}`$, the four-fermion interaction terms contained in $`_{4\mathrm{f}}`$, and higher order terms in $`_{\mathrm{hot}}`$, when applied to finite nuclei the model must include the coupling $`_{\mathrm{em}}`$ of the protons to the electromagnetic field $`A^\mu `$, and derivative terms contained in $`_{\mathrm{der}}`$. In the terms $`_\nu (\overline{\psi }\mathrm{\Gamma }\psi )`$ the derivative is understood to act on both $`\overline{\psi }`$ and $`\psi `$. One could, of course, construct many more higher order interaction terms, or derivative terms of higher order, but in practice only a relatively small set of free parameters can be adjusted from the data set of ground state nuclear properties.
The Lagrangian is understood to be used in the mean-field approximation. The single-nucleon Dirac equation is derived by the variation of the Lagrangian (2) with respect to $`\overline{\psi }`$
$$i_t\psi _i=\left\{𝜶[i\mathbf{}𝑽(r,t\mathbf{)}]+V(𝒓,t)+𝜷\left(m+S(𝒓,t)\right)\right\}\psi _i.$$
(3)
The Dirac hamiltonian contains the scalar and vector potentials
$$S(𝒓,t)=\mathrm{\Sigma }_S(𝒓,t)+\stackrel{}{\tau }\stackrel{}{\mathrm{\Sigma }}_{TS}(𝒓,t),$$
(4)
$$V_\mu (𝒓,t)=\mathrm{\Sigma }^\mu (𝒓,t)+\stackrel{}{\tau }\stackrel{}{\mathrm{\Sigma }}_T^\mu (𝒓,t),$$
(5)
with the nucleon isoscalar-scalar, isovector-scalar, isoscalar-vector and isovector-vector self-energies defined by the following relations:
$`\mathrm{\Sigma }_S`$ $`=`$ $`\alpha _S(\overline{\psi }\psi )+\beta _S(\overline{\psi }\psi )^2+\gamma _S(\overline{\psi }\psi )^3\delta _S\mathrm{}(\overline{\psi }\psi ),`$ (6)
$`\stackrel{}{\mathrm{\Sigma }}_{TS}`$ $`=`$ $`\alpha _{TS}(\overline{\psi }\stackrel{}{\tau }\psi )\delta _S\mathrm{}(\overline{\psi }\stackrel{}{\tau }\psi ),`$ (7)
$`\mathrm{\Sigma }^\mu `$ $`=`$ $`\alpha _V(\overline{\psi }\gamma ^\mu \psi )+\gamma _V(\overline{\psi }\gamma ^\alpha \psi )(\overline{\psi }\gamma _\alpha \psi )(\overline{\psi }\gamma ^\mu \psi )\delta _V\mathrm{}(\overline{\psi }\gamma ^\mu \psi )eA^\mu {\displaystyle \frac{1\tau _3}{2}},`$ (8)
$`\stackrel{}{\mathrm{\Sigma }}_T^\mu `$ $`=`$ $`\alpha _{TV}(\overline{\psi }\stackrel{}{\tau }\gamma ^\mu \psi )\delta _{TV}\mathrm{}(\overline{\psi }\stackrel{}{\tau }\gamma ^\mu \psi ),`$ (9)
respectively. The self-energies are determined by the corresponding local densities and currents calculated in the no-sea approximation
$`\rho _S(𝒓,t)`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{A}{}}}\overline{\psi }_i(𝒓,t)\psi _i(𝒓,t),`$
$`\stackrel{}{\rho }_{TS}(𝒓,t)`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{A}{}}}\overline{\psi }_i(𝒓,t)\stackrel{}{\tau }\psi _i(𝒓,t),`$
$`j_\mu (𝒓,t)`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{A}{}}}\overline{\psi }_i(𝒓,t)\gamma _\mu \psi _i(𝒓,t),`$
$`\stackrel{}{j}_\mu (𝒓,t)`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{A}{}}}\overline{\psi }_i(𝒓,t)\stackrel{}{\tau }\gamma _\mu \psi _i(𝒓,t).`$ (10)
The summation runs over all A occupied states in the Fermi sea, i.e. only occupied single-nucleon states with positive energy explicitly contribute to the nucleon self-energies. Even though the stationary solutions for the negative-energy states do not contribute to the densities in the no-sea approximation, their contribution is implicitly included in the time-evolution of the nuclear system Rin.01 ; VBR.95 .
In an effective theory with the parameters of the Lagrangian determined from a set of ground-state data, a large part of vacuum polarization effects is already taken into account in adjusting the parameters to experiment. The stationary solutions of the relativistic mean-field equations correspond to the ground-state of a nucleus. The Dirac spinors which determine the ground-state densities (i.e. positive-energy states) can be expanded, for instance, in terms of vacuum solutions, which form a complete set of plane wave functions in spinor space. This set is only complete, however, if in addition to the positive-energy states, it also contains the states with negative energy, in this case the Dirac sea of the vacuum. Positive-energy solutions of the RMF equations in a finite nucleus automatically contain vacuum components with negative energy. In the same way, solutions that describe excited states, contain negative-energy components which correspond to the ground-state solution.
This is also true, in particular, for the solutions of the time-dependent problem. In the time-evolution of A nucleons in the effective mean-field potential, at each time $`t`$ the single-nucleon spinors $`\psi _i(t)`$ can be expanded in terms of the complete set of solutions of the stationary Dirac equation $`\psi _k^{(0)}`$. This means that at each time $`t`$ one finds a local Fermi sea of $`A`$ time-dependent spinors which, of course, contain components of negative-energy solutions of the stationary Dirac equation. The states which form the local Dirac sea are orthogonal to the local Fermi sea at each time. This is the meaning of the no-sea approximation in the time-dependent problem.
The relativistic random-phase approximation (RRPA) equations can be derived from the response of the density matrix to an external field that oscillates with a small amplitude (for details see Refs. Rin.01 ; NVR.02 ). The matrix form of these equations reads
$$\left(\begin{array}{cc}A& B\\ B^{}& A^{}\end{array}\right)\left(\begin{array}{c}X_\nu \\ Y_\nu \end{array}\right)=E_\nu \left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)\left(\begin{array}{c}X_\nu \\ Y_\nu \end{array}\right),$$
(11)
where $`E_\nu `$ denotes the eigenfrequency, and $`X_\nu `$ and $`Y_\nu `$ are the corresponding RPA amplitudes. The RRPA matrices A and B read
$$A=\left(\begin{array}{cc}(ϵ_pϵ_h)\delta _{pp^{}}\delta _{hh^{}}& \\ & (ϵ_\alpha ϵ_h)\delta _{\alpha \alpha ^{}}\delta _{hh^{}}\end{array}\right)+\left(\begin{array}{cc}V_{ph^{}hp^{}}& V_{ph^{}h\alpha ^{}}\\ V_{\alpha h^{}hp^{}}& V_{\alpha h^{}h\alpha ^{}}\end{array}\right)$$
(12)
and
$$B=\left(\begin{array}{cc}V_{pp^{}hh^{}}& V_{p\alpha ^{}hh^{}}\\ V_{\alpha p^{}hh^{}}& V_{\alpha \alpha ^{}hh^{}}\end{array}\right).$$
(13)
The matrix elements of the residual interaction are derived from the Dirac hamiltonian of Eq. (3):
$$V_{abcd}=\frac{h_{ac}}{\rho _{db}},$$
(14)
where the generic indices ($`a,b,c,d,\mathrm{}`$) denote quantum numbers that specify the single-nucleon states $`\{\psi _a\}`$. These belong to three distinct sets: the index $`p`$ (particle) denotes unoccupied states above the Fermi sea, the index $`h`$ (hole) is for occupied states in the Fermi sea, and with $`\alpha `$ we denote the unoccupied negative-energy states in the Dirac sea.
Since the RRPA is derived in the small amplitude limit, the currents and densities can be expanded around their ground-state values
$`\rho _S(𝒓,t)`$ $`=`$ $`\rho _S^{gs}(𝒓)+\delta \rho _S(𝒓,t)`$
$`\stackrel{}{\rho }_{TS}(𝒓,t)`$ $`=`$ $`\rho _{TS}^{gs}(𝒓)+\delta \stackrel{}{\rho }_{TS}(𝒓,t)`$
$`j_\mu (𝒓,t)`$ $`=`$ $`\rho _V^{gs}(𝒓)+\delta j_\mu (𝒓,t),`$
$`\stackrel{}{j}_\mu (𝒓,t)`$ $`=`$ $`\rho _{TV}^{gs}(𝒓)+\delta \stackrel{}{j}_\mu (𝒓,t).`$ (15)
In this work we only consider spherical even-even nuclei. Because of time-reversal invariance, the spatial components of the currents vanish in the nuclear ground state. Furthermore, charge conservation implies that only the 3-component of the isovector scalar and vector densities contributes in the ground state. The individual contribution of each field to $`V_{abcd}`$ can now be obtained by inserting the expansions Eq. (15) in the matrix element of the Dirac Hamiltonian Eq. (3):
$`V_{abcd}^S`$ $`=`$ $`{\displaystyle \psi _a^{}\beta \psi _c\left(\alpha _S+2\beta _S\rho _S^{gs}+3\gamma _S\rho _{S}^{gs}{}_{}{}^{2}+\delta _S\mathrm{\Delta }\right)\psi _b^{}\beta \psi _dd^3r},`$
$`V_{abcd}^{TS}`$ $`=`$ $`{\displaystyle \psi _a^{}\beta \stackrel{}{\tau }\psi _c\left(\alpha _{TS}+\delta _{TS}\mathrm{\Delta }\right)\psi _b^{}\beta \stackrel{}{\tau }\psi _dd^3r},`$
$`V_{abcd}^V`$ $`=`$ $`{\displaystyle \psi _a^{}\psi _c\left(\alpha _V+3\gamma _V\rho _{V}^{gs}{}_{}{}^{2}+\delta _V\mathrm{\Delta }\right)\psi _b^{}\psi _dd^3r}`$
$``$ $`{\displaystyle \psi _a^{}𝜶\psi _c\left(\alpha _V+\gamma _V\rho _{V}^{gs}{}_{}{}^{2}+\delta _V\mathrm{\Delta }\right)\psi _b^{}𝜶\psi _dd^3r},`$
$`V_{abcd}^{TV}`$ $`=`$ $`{\displaystyle \psi _a^{}\beta \stackrel{}{\tau }\gamma _\mu \psi _c\left(\alpha _{TV}+\delta _{TV}\mathrm{\Delta }\right)\psi _b^{}\beta \stackrel{}{\tau }\gamma ^\mu \psi _dd^3r}.`$ (16)
For open-shell nuclei calculations are performed in the framework of the fully self-consistent RHB plus relativistic QRPA model. The RHB represents a relativistic extension of the Hartree-Fock-Bogoliubov model, and it provides a unified description of particle-hole ($`ph`$) and particle-particle ($`pp`$) correlations. In most applications of the RHB model VALR.05 the pairing part of the well known and very successful Gogny force BGG.84 has be employed in the $`pp`$ channel.
$$V^{pp}(1,2)=\underset{i=1,2}{}e^{((𝐫_1𝐫_2)/\mu _i)^2}(W_i+B_iP^\sigma H_iP^\tau M_iP^\sigma P^\tau ),$$
(17)
with the set D1S BGG.91 for the parameters $`\mu _i`$, $`W_i`$, $`B_i`$, $`H_i`$, and $`M_i`$ $`(i=1,2)`$. This force has been very carefully adjusted to the pairing properties of finite nuclei all over the periodic table. In particular, the basic advantage of the Gogny force is the finite range, which automatically guarantees a proper cut-off in momentum space.
In Ref. Paar.03 the RQRPA has been formulated in the canonical single-nucleon basis of the RHB model. By definition, the canonical basis diagonalizes the density matrix and it is always localized. It describes both the bound states and the positive-energy single-particle continuum. This particular representation of the RQRPA is very convenient because, in order to describe transitions to low-lying excited states in weakly-bound nuclei, the two-quasiparticle configuration space must include states with both nucleons in the discrete bound levels, states with one nucleon in a bound level and one nucleon in the continuum, and also states with both nucleons in the continuum. In addition, the full RQRPA equations are rather complicated and it is considerably simpler to solve these equations in the canonical basis where the RHB wave functions take a simple BCS-form. In this case one needs only the matrix elements of the interactions in the $`ph`$ and $`pp`$ channels, and certain combinations of the occupation factors of canonical states. The relativistic QRPA of Ref. Paar.03 is fully self-consistent. For the interaction in the particle-hole channel effective Lagrangians with nonlinear meson self-interactions or nucleon point couplings are used, and pairing correlations are described by the pairing part of the finite range Gogny interaction. Both in the $`ph`$ and $`pp`$ channels, the same interactions are used in the RHB equations that determine the canonical quasiparticle basis, and in the matrix equations of the RQRPA. The RQRPA configuration space includes also the Dirac sea of negative energy states.
In the next section we will present illustrative relativistic RPA/QRPA calculations of the multipole response of spherical nuclei. For the multipole operator $`\widehat{Q}_{\lambda \mu }`$, the response function $`R(E)`$ is defined
$$R(E)=\underset{f}{}B(0_i\lambda _f)\frac{\mathrm{\Gamma }/2\pi }{(EE_f)^2+(\mathrm{\Gamma }/2)^2},$$
(18)
where $`\mathrm{\Gamma }`$ is the width of the Lorentzian distribution, and
$$B(0_i\lambda _f)=|\lambda _f||\widehat{Q}_\lambda ||0_i|^2.$$
(19)
For all calculations in this work the discrete spectrum of the RRPA states is folded with the Lorentzian of Eq. (18), with the width $`\mathrm{\Gamma }=1`$ MeV.
## III Multipole giant resonances
We have performed fully consistent relativistic RPA/QRPA calculations of isoscalar monopole, isovector dipole, and isoscalar quadrupole giant resonances in spherical nuclei. The interaction in the particle-hole channel is determined by the effective point-coupling Lagrangian Eq. (2), and pairing correlations are described by the pairing part of the finite range Gogny interaction D1S BGG.91 . The R(Q)RPA configuration space includes the Dirac sea of negative energy states. Both in the particle-hole and particle-particle channels, the same interactions are used in the calculation of the ground state and in the matrix equations of the R(Q)RPA.
The point-coupling Lagrangian Eq. (2) contains 11 adjustable coupling constants. The PC-F1 effective interaction, adjusted in Ref. BMM.02 , corresponds to a restricted set of 9 coupling parameters and does not include the isovector-scalar channel. The parameters have been determined in a $`\chi ^2`$-minimization procedure, adjusted to ground-state observables (binding energy, charge radius, diffraction radius and surface thickness) of 17 spherical nuclei. The resulting parameters of the PC-F1 effective interaction are displayed in Tab. 1. This interaction has been tested in the analysis of the equation of state of symmetric nuclear matter and neutron matter, binding energies and form-factor- and shell-structure-related ground-state properties of several isotopic and isotonic chains, including superheavy nuclei with known experimental masses, and of the fission barrier in <sup>240</sup>Pu. The comparison with data has shown that the RMF-PC model with the PC-F1 interaction can compete with the best phenomenological finite-range meson-exchange interactions. It should be noted, however, that PC-F1 exhibits a relatively large volume asymmetry at saturation $`a_438`$ MeV, resulting in a very stiff equation of state for neutron matter, and too large values for the neutron skin in finite nuclei. The most recent meson-exchange RMF effective forces, on the other hand, include an explicit medium dependence both in the isoscalar and isovector channels TW.99 ; NVF.02 , and thus provide an improved description of asymmetric nuclear matter and neutron matter, and realistic values of the neutron skin.
In the following examples we will test the PC-F1 interaction in the calculation of excitation energies of giant resonances in spherical nuclei. We will also try to determine whether R(Q)RPA calculations of excited states can be used to discriminate between different point-coupling interactions, or even place additional constraints on the parameters of the effective interactions NVR.02 .
### III.1 The isoscalar giant monopole resonance
The isoscalar giant monopole resonance (ISGMR) represents the simplest mode of collective oscillations in finite nuclei (the breathing mode), and provides valuable information on the nuclear matter incompressibility. The range of values of the nuclear matter compression modulus $`K_{\mathrm{}}`$ can be best determined by comparing results of fully consistent microscopic calculation of both ground state properties and the ISGMR excitation energies in spherical nuclei with data. Moreover, since $`K_{\mathrm{}}`$ determines bulk properties of nuclei and, on the other hand, the GMR excitation energies depend also on the surface compressibility, measurements and microscopic calculations of GMR in heavy spherical nuclei should, in principle, provide a more reliable estimate of the nuclear matter incompressibility Bla.80 ; BBDG.95 . A recent relativistic mean-field plus R(Q)RPA analysis of the ISGMR, based on effective Lagrangians with density-dependent meson-nucleon vertex functions, has shown that the nuclear matter compression modulus of effective interactions based on the relativistic mean-field approximation should be restricted to a rather narrow interval $`K_{\mathrm{}}250270`$ MeV VNR.03 .
Although the point-coupling PC-F1 interaction has been adjusted to ground-state data only, its compression modulus $`K_{\mathrm{}}=270`$ MeV is within the range predicted by the meson-exchange models. The latter, however, tend to underestimate the surface thickness of finite nuclei BMM.02 . Since the excitation energies of the ISGMR generally depend also on the surface compressibility, point-coupling and meson-exchange effective interactions with nearly identical values of the nuclear matter compression modulus, do not necessarily predict identical GMR excitation energies, especially in lighter nuclei in which surface effects are more pronounced.
In Fig. 1 we display the isoscalar monopole strength distribution in <sup>208</sup>Pb, calculated in the relativistic RPA with the PC-F1 effective interaction. For the ISGMR peak at $`E=14.16`$ MeV excitation energy, in the right panel we plot the corresponding proton, neutron and total isoscalar transition densities. The node at the surface is characteristic for the the breathing mode of oscillations. The calculated peak energy is in excellent agreement with the newest data on the ISGMR centroid energy $`m_1/m_0=13.96\pm 0.20`$ MeV from Ref. Young2 (denoted by the arrow in Fig. 1). In Table 2 we have also compared the R(Q)RPA $`m_1/m_0`$ centroids with very recent data on medium-heavy nuclei Young1 ; Young2 ; Young3 . It appears that the excitation energies predicted by the PC-F1 effective interaction are systematically somewhat higher than the experimental ISGMR’s, indicating that the nuclear matter compression modulus of a relativistic point-coupling model should probably be closer to $`K_{\mathrm{}}250`$ MeV. This would be in agreement with the modern density-dependent meson-exchange effective interactions: DD-ME1 with $`K_{\mathrm{}}=245`$ MeV NVF.02 , and DD-ME2 with $`K_{\mathrm{}}=251`$ MeV LNV.05 .
### III.2 The isovector giant dipole resonance
The isovector giant dipole resonance (IVGDR), calculated in the R(Q)RPA, is predominantly determined by the isovector channel of the effective interaction. In particular, the position of the IVGDR is directly related to the nuclear matter asymmetry energy. This quantity can be expanded in a Taylor series in $`\rho `$ Lee.98
$$S_2(\rho )=a_4+\frac{p_0}{\rho _{sat}^2}(\rho \rho _{sat})+\frac{\mathrm{\Delta }K}{18\rho _{sat}^2}(\rho \rho _{sat})^2+\mathrm{}.$$
(20)
The value of the asymmetry energy at the saturation density (volume asymmetry) is denoted by $`a_4`$, the parameter $`p_0`$ defines the linear density dependence of the asymmetry energy, and $`\mathrm{\Delta }K`$ is the correction to the incompressibility. The asymmetry energy directly determines the difference $`r_nr_p`$ between the radii of the neutron and proton ground-state density distributions. In a recent study that has analyzed available data on neutron radii and the excitation energies of the IVGDR in the framework of the density-dependent meson-exchange RMF models, the volume asymmetry of RMF effective interactions has been constrained to the interval: $`32`$ MeV $`a_4`$ $`36`$ MeV VNR.03 . For the PC-F1 effective interaction the volume asymmetry is somewhat larger: $`a_4=37.8`$ MeV. This is also the case for older meson-exchange RMF forces with non-linear meson self-interaction terms, and is due to the fact that the isovector channel of these interactions is basically parameterized by a single constant: $`\alpha _{\mathrm{TV}}`$ in the point-coupling version, or the $`\rho `$-meson coupling $`g_\rho `$ in the meson-exchange models. With a single parameter in the isovector channel, it is simply not possible to simultaneously lower $`a_4`$ to its empirical value and reproduce the masses of $`NZ`$ nuclei. This only becomes possible if a density dependence is included in the isovector channel, as it is done in modern density-dependent meson-exchange RMF forces TW.99 ; NVF.02 ; LNV.05 .
For the point-coupling effective Lagrangian Eq. (2), in Ref. BMM.02 it has been shown that the strength parameter $`\delta _{\mathrm{TV}}`$ of the derivative term in the isovector channel cannot be determined from ground-state properties of finite nuclei. The isovector channel of the point-coupling Lagrangian was also investigated by the inclusion of the isovector-scalar terms. Two additional interactions were generated: PC-F2 which includes only the linear isovector-scalar term with the coupling constant $`\alpha _{\mathrm{TS}}`$, and PC-F4 which contains both the linear and derivative isovector-scalar terms, with the corresponding parameters $`\alpha _{\mathrm{TS}}`$ and $`\delta _{\mathrm{TS}}`$. In comparison to the PC-F1 interaction, however, the $`\chi ^2`$ for the extended sets PC-F2 and PC-F4 was reduced by less then $`1\%`$, and it was concluded that these extensions are not well determined by the ground-state data included in the fit. Since properties of isovector collective modes could, in principle, provide additional information on the isovector channel of the effective interaction, it is interesting to compare the isovector dipole strength distributions calculated with PC-F1, PC-F2, and PC-F4. For <sup>208</sup>Pb the resulting curves, shown in the left panel of Fig. 2, are practically indistinguishable. This is simply because the corresponding asymmetry energies begin to differ only at densities high above the saturation density (see Fig. 3). On the other hand, since the IVGDR corresponds to a predominantly surface mode of oscillations, the density region which determines this resonance is located below saturation density. This is illustrated in the right panel of Fig. 2, where we plot the neutron, proton, and total isovector transition densities for the peak at $`E=13.0`$ MeV, calculated with the PC-F1 interaction. For the point-coupling Lagrangian, this means that properties of the IVGDR do not determine more precisely the couplings $`\alpha _{\mathrm{TS}}`$ and $`\delta _{\mathrm{TS}}`$.
We note that the PC-F1 interaction predicts the excitation energy of the IVGDR in <sup>208</sup>Pb at $`E=13.0`$ MeV, which is below the experimental value $`E=13.3\pm 0.1`$ MeV Rit.93 . This is also the case for the lighter nuclei: <sup>90</sup>Zr, <sup>116</sup>Sn, <sup>118</sup>Sn, <sup>120</sup>Sn, and <sup>124</sup>Sn, for which in Table 3 we compare the R(Q)RPA IVGDR excitation energies with data BF.75 . The fact that effective interactions with large volume asymmetry underestimate the energy of the IVGDR has already been demonstrated in two recent studies of the isovector dipole response performed with non-relativistic and relativistic RPA Rei.99 ; NVR.02 .
### III.3 The isoscalar giant quadrupole resonance
In nonrelativistic RPA calculations, the excitation energy of the isoscalar giant quadrupole resonance (ISGQR) can be directly related to the nucleon effective mass $`m^{}`$ associated with a given interaction. For Skyrme-type effective interactions, in particular, the excitation energy of the ISGQR exhibits a linear dependence on $`m^{}`$. The larger the effective mass, i.e., the higher the density of states around the Fermi surface, the lower is the calculated ISGQR excitation energy. Calculations of both ground-state properties and ISGQR excitation energies in spherical nuclei, constrain the effective mass for Skyrme-type interactions to the interval: $`m^{}/m=0.8\pm 0.1`$ Rei.99 .
The situation is slightly more complicated in the relativistic framework, because one finds several different quantities denoted as the ”effective mass”. The quantity which is most often used to characterize an effective interaction is the Dirac mass
$$m_D=m+S(𝐫),$$
(21)
where m is the nucleon mass and $`S(𝐫)`$ is the scalar nucleon self-energy. The concept of the effective mass in the relativistic framework has been extensively analyzed in Refs. JM.89 ; JM.90 . In particular, it has been shown that one must not identify the Dirac mass with the effective mass of the nonrelativistic mean-field models. Instead, the quantity which should be compared with the empirical effective mass derived from nonrelativistic analyses of scattering and bound state data is given by
$$m^{}=mV(𝐫),$$
(22)
where $`V(𝐫)`$ denotes the time-like component of the vector self-energy. However, both $`m_D`$ and $`m^{}`$ are essentially determined by: (i) the empirical spin-orbit splittings in finite nuclei, and (ii) the binding energy at saturation density in nuclear matter. They place the following constraints on the effective masses: $`0.55mm_D0.6m`$ and $`0.64mm^{}0.67m`$. In comparison to the nonrelativistic self-consistent mean-field models, the allowed values for the relativistic $`m^{}`$ are rather low, resulting in a smaller density of states around the Fermi surface. Moreover, the allowed interval of $`m^{}`$ values is so narrow, that there is no room for any significant enhancement of the single-nucleon level densities in the framework of the standard phenomenological RMF models VNR.02 .
These arguments are also valid for point-coupling RMF models. Specifically, for the PC-F1 effective interaction the Dirac mass $`m_D=0.61m`$, the effective mass $`m^{}=0.69m`$, and therefore one should not expect PC-F1 to accurately reproduce data on the ISGQR. In the left panel of Fig. 4 we plot the RRPA isoscalar quadrupole strength distribution in <sup>208</sup>Pb, calculated with the PC-F1 interaction. The experimental excitation energy of the ISGQR ($`10.89\pm 0.3`$ MeV Young2 ) is denoted by the arrow. Because of the low nucleon effective mass, the calculated excitation energy of the ISGQR is above the corresponding experimental value. Similar results are also obtained for lighter spherical nuclei. In Table 4 we display a comparison between the experimental excitation energies and the PC-F1 predictions for the location of the ISGQR in <sup>90</sup>Zr, <sup>116</sup>Sn, <sup>112</sup>Sn, <sup>124</sup>Sn, <sup>144</sup>Sm, and <sup>208</sup>Pb. For all these nuclei the calculated ISGQR excitation energy is more than 1 MeV above the experimental centroids $`m_1/m_0`$.
## IV Summary and outlook
During the last decade standard meson-exchange RMF models, with either non-linear meson self-interaction terms, or with density-dependent meson-nucleon vertex functions, have been very successfully applied in the description of a variety of nuclear structure phenomena. However, the explicit inclusion of the meson degrees of freedom, in particular of the fictitious $`\sigma `$-meson, places physical constraints on the model parameters, thereby reducing the predictive power of the model. The limitations of the meson-exchange representation of the RMF theory are especially pronounced in the description of surface properties of finite nuclei. Virtually all meson-exchange RMF effective interactions, which otherwise accurately reproduce data on bulk nuclear properties and giant resonances, underestimate the empirical surface thickness. This does not seem to be the case for the self-consistent point-coupling RMF models, which therefore represent an interesting alternative to the standard meson-exchange picture of the effective nuclear interaction.
Self-consistent point-coupling RMF models have recently attracted considerable interest. For the phenomenological models, in particular, it has been shown that the new PC-F1 effective interaction reproduces data with a quality comparable to that of standard meson-exchange forces. However, all calculations performed so far have only considered ground-state nuclear properties. It is, therefore, important to develop a consistent microscopic framework, based on the point-coupling RMF effective Lagrangian, in which dynamical properties and excited states can be investigated.
In this work the matrix equations of the random-phase approximation (RPA) have been derived for the point-coupling Lagrangian of the (RMF) model. Fully consistent RMF plus RPA, and RHB plus QRPA illustrative calculations of the isoscalar monopole, isovector dipole and isoscalar quadrupole response of spherical medium-heavy and heavy nuclei have been performed. A comparison with experiment has shown that the best point-coupling effective interactions, and in particular PC-F1, accurately reproduce not only ground-state properties, but also data on excitation energies of giant resonances. We have also investigated the possibility to determine the parameters of the isovector-scalar channel from R(Q)RPA calculations of the isovector dipole response. This is really not feasible because the isovector-scalar terms influence the symmetry energy only for nucleon densities well above the saturation density, whereas the density region characteristic for the IVGDR is located below the saturation density.
The R(Q)RPA based on the point-coupling RMF models presents an important addition to the theoretical tools that are employed in description of the nuclear many-body problem. Future applications will include studies of the multipole response of exotic nuclei far from the valley of $`\beta `$-stability. On a more microscopic level, the R(Q)RPA will be used to investigate dynamical properties predicted by the recently introduced relativistic point-coupling model constrained by in-medium QCD sum rules and chiral symmetry FKV.03 ; FKV.04 ; VW.04 .
ACKNOWLEDGMENTS
This work has been supported in part by the Bundesministerium für Bildung und Forschung - project 06 MT 193, by the Alexander von Humboldt Stiftung, and by the Croatian Ministry of Science - project 0119250.
|
warning/0506/astro-ph0506599.html
|
ar5iv
|
text
|
# The AAO/UKST SuperCOSMOS H𝛼 Survey (SHS)
## 1 Introduction
H$`\alpha `$ emission from HII regions is one of the most direct optical tracers of current star formation activity and is routinely used to measure star formation rates in external galaxies (e.g. Kennicutt 1992). In our own galaxy, HII regions are seen by direct UV illumination of molecular clouds from adjacent hot stars and as highly structured shells, bubbles and sheets of emission resulting from supernovae, planetary nebula, Wolf-Rayet stars and other stellar outflows. Some large-scale outflows can, in turn, be themselves a trigger of star formation, and their morphology is strongly influenced by the nature and density of the ISM into which they expand. H$`\alpha `$ imaging allows this to be studied in great detail in our immediate Galactic neighbourhood and to be detected at a great distance in external galaxies. The UV flux from hot stars also excites a more diffuse emission from the ISM, unconnected to current star formation and detectable over large areas of sky.
The perimeters of some large emission shells appear to enclose the locations of more recent star formation which may in turn generate further supernovae and stellar winds (Dopita, Matthewson & Ford 1985), while their morphology informs the processes by which star formation is propagated (e.g. Elmegreen & Lada 1977, Gerola & Seiden 1978). Because of proximity, some of these structures can present very large angular sizes such as Barnard’s loop (probably the first large-scale emission structure detected in the Galaxy) which subtends 13 (Pickering 1890) and the Gum nebula which at 36, is even larger (Gum 1952). More distant complexes or groups of HII regions, such as NGC 6334, can still be of the order 1 across and yet present fine detail on arcsecond scales (Meaburn & White 1982). Given their interaction with their external large-scale environment (Tenorio-Tagle & Palous 1987) it was clear that emission-line imaging of these structures required an efficient wide-field capability and high spatial resolution.
On smaller scales, stellar H$`\alpha `$ emission characterises the short lived, least well understood stages of stellar evolution, i.e. those of pre- and post-main sequence stars, planetary nebulae and close binary systems. Previous efforts to detect emission sources have either offered modest area coverage; e.g. the UBVI and H$`\alpha `$ photometric surveys of Sung, Chun & Bessell (2000) or Keller et al. (2001) or, where a large-area survey has been conducted, becomes incomplete at relatively bright magnitudes. An example is the objective-prism survey of Stephenson & Sanduleak (1977) which reaches only $``$ 14<sup>th</sup> magnitude. Such surveys are highly incomplete so their emission source catalogues provide only limited samples upon which to build our understanding of these rarely observed phases of stellar evolution.
From the above, the importance of Galactic H$`\alpha `$ line emission from both stars and nebulae is evident and this has encouraged many surveys for HII regions in particular, e.g. Sharpless (1953, 1959), Gum (1955), Hase & Shajn (1955), Bok, Bester & Wade (1955), Johnson (1955, 1956), Rodgers, Campbell & Whiteoak (1960), Georgelin & Georgelin (1970) and Sivan (1974). These earlier surveys were limited to relatively small targeted areas or had such wide fields of view that small-scale detail was lost due to the low angular resolution; e.g. the survey of Sivan (1974) used 60 field diameters giving a plate scale of $`6`$ mm<sup>-1</sup>. Relatively little optical emission-line survey work had been done in a way that combined wide-angle coverage with good sensitivity and high resolution. These characteristics are essential to allow thorough examination of the morphology and interaction of emission regions with their environment on arcsecond to degree scales and to detect the large variety of stellar emission sources to suitably faint levels.
Hence, in the mid-1990s, a number of the present authors suggested that the U.K. Schmidt Telescope (UKST) should be used to make a narrow band photographic H$`\alpha `$ survey of the Southern Milky Way and Magellanic Clouds. The only previous wide-area UKST H$`\alpha `$ material comes from the work by Meaburn and co-workers in the 1970s (see Davies, Elliot & Meaburn 1976; Meaburn 1980). They used a 100Å band-pass multi-element mosaic filter and fast, but coarse grained, 098-04 emulsion. It covered some limited areas close to the Galactic Plane (Meaburn & White 1982), but was mainly influential in the study of the ionised gas in the Magellanic Clouds, showing the first evidence for “supergiant-shells” and other large-scale features (Davies et al. 1976, Meaburn 1980). There are other recent wide-area H$`\alpha `$ surveys such as that by the Virginia group in the northern hemisphere (VTSS – Dennison et al. 1998) and the Mount Stromlo group in the south (Buxton et al. 1998). Recently, Gaustad et al. (2001) have released the full “Southern H$`\alpha `$ Sky Survey Atlas” (SHASSA), covering the entire southern sky. This imaging survey has rather coarse, 48-arcsecond pixels and strong artefacts from uncancelled stars in the continuum-subtracted product, but has the major benefit of being directly calibrated in Rayleighs. These surveys continue the tradition of deep, low spatial resolution studies, but use CCDs which permit low flux densities of a few tenths of a Rayleigh to be achieved.
An alternative approach, taken by the Marseille and Wisconsin Fabry-Perot groups in the southern and northern skies respectively (see Russeil et al. 1997, 1998 and Haffner et al. 2003 for the WHAM – Wisconsin H$`\alpha `$ Mapper), was to obtain high resolution spectral (i.e. velocity) data, but again with low spatial resolution (e.g. 1 pixels for WHAM).
A critical comparison between the WHAM, SHASSA and VTSS surveys was undertaken by Finkbeiner (2003) who presented a ‘whole-sky’ H$`\alpha `$ map. Significantly, none of these major surveys offer the arcsecond spatial resolution of the AAO/UKST H$`\alpha `$ survey. A summary of fundamental properties of these modern surveys is given in Table 1.
## 2 The AAO/UKST H$`\alpha `$ Survey
The AAO/UKST H$`\alpha `$ survey provides a 5 Rayleigh sensitivity narrow-band survey of Galactic emission (H$`\alpha `$ plus \[NII\] 6548, 6584Å) with arcsecond spatial resolution. Henceforth the survey will be refered to simply as the H$`\alpha `$ survey though it is understood that this includes any \[NII\] emission component that is sampled by the filter band-pass (such emission can completely dominate H$`\alpha `$ for some PN types for example). Approximately 4000 deg<sup>2</sup> of the Southern Milky Way have been covered to $`|b|1013`$ together with a separate contiguous region of 700 deg<sup>2</sup> in and around the Magellanic Clouds. Matching 3-hour H$`\alpha `$ and 15-min broad-band (5900-6900Å) short red (SR) exposures were taken over the 233 distinct but overlapping fields of the Galactic Plane and 40 fields of the Magellanic Clouds. These were done on 4 centres because of the circular aperture of the H$`\alpha `$ interference filter which has a dielectric coating diameter of about 305mm ($`5.7`$) deposited on a standard $`356\times 356`$mm red glass (RG610) substrate (refer Section 5). The overlapping 4 field centres enable full, contiguous coverage in H$`\alpha `$ despite the circular filter aperture. Because of the slightly smaller effective field, a new Southern sky-grid of 1111, 4 field centres was created (whose numbers should not be confused with the 893 standard 5 field centres of the UKST Southern Sky Surveys). A map of the survey region in a standard UKST RA/DEC plot together with the new field numbers is available on the SHS web site<sup>1</sup><sup>1</sup>1http://www-wfau.roe.ac.uk/sss/halpha/. In the on-line UKST plate catalogue these fields have a ‘h’ prefix (e.g. h123) to avoid confusion with the ESO/SERC 5 fields.
The survey began in 1997 and took six years to complete. This latest and final UKST photographic survey was the first large-scale, narrow band survey undertaken on the telescope and is the first where the sole method of dissemination to the community is via access to on-line digital data products. Preliminary survey details and results were given by Parker & Phillipps (1998, 2003). The present paper is intended as the definitive reference for the survey. We describe the key characteristics of the survey, the on-line data product, some survey limitations, a flux calibration scheme, comparisons with other surveys and a brief overview of the potential for current and future scientific exploitation.
The arcsecond resolution of the AAO/UKST H$`\alpha `$ survey makes it a particularly powerful tool, not only for investigating the detailed morphology of emission features across the widest range of angular scales, but also as a means of identifying large numbers of faint point-source H$`\alpha `$ emitters, which include cataclysmic variables, T Tauri, Be and symbiotic stars, compact Herbig-Haro objects and unresolved planetary nebulae (PNe). Given the coincidence of the broad CIV/HeII blend in late-type Wolf Rayet stars, these objects can also be detected. Most other comparative surveys (Table 1) are largely insensitive to point-source emitters as they lack spatial resolution, being optimised instead for the faintest levels of resolved and diffuse emission.
On larger scales, the detailed spatial structure of the ionized ISM component traced by the new AAO/UKST H$`\alpha `$ survey can provide key data for many studies, e.g. mapping of specific areas for detailed spectroscopic follow-up to obtain emission-line gas kinematics or for dynamical studies of star forming regions, with their implications for the energetics of the central stars. Furthermore, comparisons with other indicators of star formation from other wavebands should provide essential clues to the active mechanisms. The survey also complements the recent Galactic Plane radio maps from MOST (Green et al. 1999), the new NIR maps from 2MASS (Jarrett et al. 2000) and the mid-infrared maps from the MSX satellite (Price et al. 2001).
Figure 1 presents two panels showing the 233 survey fields (mosaiced together by M.Read) to illustate the overall survey coverage. The entire survey has been incorporated into an on-line mosaic within the freeware ‘Zoomify’ environment (see http://www.zoomify.com) which enables preliminary survey visualisation and scanning. The lowest resolution map can be zoomed-in to a level where each pixel represents about 12 arcseconds. This interactive map is available on-line.<sup>2</sup><sup>2</sup>2 http://surveys.roe.ac.uk/ssa/hablock/hafull.html This map is a factor of 18 lower in resolution than the full 0.67 arcsecond pixel survey data available on-line which should be used for serious scientific work. The success of this survey has led directly to a northern counterpart, currently underway on the 2.5m Isaac Newton Telescope on La Palma using a wide-field CCD camera and H$`\alpha `$ R and I band imaging; the Isaac Newton telescope Photometric H$`\alpha `$ Survey (IPHAS). This important survey is the subject of a separate paper (Drew et al. 2005) though a brief comparison in an overlap region is included later in Section 12.
## 3 The detector: Technical Panchromatic film-based emulsion
The survey was carried out using Kodak Technical Panchromatic (Tech-Pan) Estar based films (e.g. Kodak 1987). The superb qualities of this emulsion and its adaptation for UKST use has been described in detail by Parker & Malin (1999) so only a very brief summary is given here. The Tech-Pan emulsion has remarkably high quantum efficiency for a photographic material with hypersensitised films having a DQE approaching 10 per cent (Phillipps & Parker 1993). Due to its original development in connection with solar patrol work, it has particularly high efficiency around H$`\alpha `$. The Tech-Pan films are also extremely fine grained, with an inherent resolution of $`5\mu `$m, leading to an excellent high-resolution imaging capability and a depth for point sources that exceeded that achieved for the more widely used glass-based IIIa-F emulsion by about a magnitude for standard UKST R-band survey 1-hour exposures (e.g. Parker & Malin 1999). These factors, combined with the wide area coverage available to Schmidt photographic surveys, made Tech-Pan an ideal choice for the Southern Galactic Plane H$`\alpha `$ survey. The colour term stability of Tech-Pan compared to the IIIa- emulsions used at the UKST is given by Morgan & Parker (2005) where these terms are shown to be stable, reproducible, generally small, and similar to those previously derived for the older IIIa- emulsions. This gives confidence in the survey’s photometric integrity. Over the survey life-time, photography on a Schmidt telescope still offered several advantages over CCD images, especially low cost and very fine spatial resolution and uniformity across a large physical area ($`356\times 356`$ mm) giving a 40 deg<sup>2</sup> wide field of view. However, a key limitation is that the detector response is linear over only a narrow dynamic range so recovering and calibrating the intensity information needs careful treatment (see Section 11). In Figure 2 we present small, $`3\times 3`$ arcminute regions to demonstrate the qualitative difference between the 3-hour H$`\alpha `$ and 15-min SR Tech-Pan exposures and the standard 60-min R-band IIIa-F UKST survey data. This region includes a newly discovered planetary nebula (PHR1706-3544) found from the H$`\alpha `$ survey data as part of the Macquarie/AAO/Strasbourg H$`\alpha `$ planetary nebulae catalogue (Parker et al. 2003 and 2005 in preparation). Note the improved resolution of the Tech-Pan image, the very similar depth of the respective exposures for point sources and the tighter point-spread function (psf) for the Tech-Pan compared to the IIIa-F emulsion.
## 4 The narrow band H$`\alpha `$ band-pass filter
To take advantage of UKST’s large field of view it was necessary to obtain a physically large narrow band-pass filter to be placed as close as possible to the telescope’s focal plane. The issues involved with mounting such filters with Schmidt telescopes has been described by Meaburn (1978) and previous large interference filters were generally of the mosaic type (e.g. Meaburn 1980). Such smaller scale interference filters are easier to manufacture and can be made to higher optical quality. However, difficulties associated with their mounting often lead to problems of missing data in the filter gaps, degraded, variable resolution and lack of homogeneity over large survey areas, even when the optical quality of the elements themselves are excellent. This was the case for the Meaburn mosiac filters which did not fully deliver the anticipated performance due to an unfortunate index mis-match in optical cement between the components which resulted in reflection ghosts (which can be got rid of numerically after scanning), coupled with the practical difficulty of mounting the components in a sandwich to eliminate optical path variations (Meaburn, private communication).
Fortunately, it proved possible for the AAO to obtain a custom-made, exceptionally large, monolithic, thin-film interference filter from Barr Associates in the USA which avoids the problems that can be associated with mosaic filters. Detailed filter specifications and characteristics are given by Parker & Bland-Hawthorn (1998). The essential features are reviewed here for completeness together with some additional modeling of the filter profile in the converging beam when off-axis (see Pierce 2005 for further details). An RG610 glass substrate was cut to $`356\times 356`$mm ($`6.5`$), the standard size of UKST filters and coated with a multilayer, dielectric stack to give a 3-cavity design with a clear aperture of $``$ 305 mm diameter and with an effecive refractive index of the equivalent monolayer of 1.34. This circular aperture of layered coating constitutes the interference filter so the corners of the square glass substrate do not behave as an H$`\alpha `$ filter. Nevertheless this is probably the world’s largest astronomical, narrow band filter. At the UKST plate scale this covers an on-sky area roughly 5.7 in diameter (slightly less than the full Schmidt field). To ensure complete and contiguous survey coverage with the circular aperture interference filter it was necessary to move use 4 field centres.
The filter central wavelength was set slightly longward of rest-frame H$`\alpha `$ for two reasons, one instrumental and one astronomical. First, the UKST has a fast, f/2.48 converging beam. This leads to the interference filter ‘scanning down’ in transmitted wavelengths for off-axis beams compared to beams incident normal to the filter. Secondly we wished to survey positive velocity gas (in our own and nearby galaxies). Given a band pass (FWHM) of 70Å, we chose to centre the filter at 6590Å in collimated light compared to 6563Å for rest-frame H$`\alpha `$. The peak filter transmission is around 90 per cent. Measurements of the filter at the CSIRO National Measurement Laboratory in Sydney quantitatively confirmed the excellent conformity of the filter to our original specifications (see Parker & Bland-Hawthorn 1998). First light filter images were obtained in April 1997.
### 4.1 The filter model
Figure 3a-b shows two spectral scans of the filter, both taken near the centre using light at normal incidence. Figure 3a is the result of a high resolution scan around the H$`\alpha `$ region and shows that the bandpass is well centred on 6590Å and has $``$ 70 Å FWHM. The transmission is high across the reasonably flat top of the bandpass, reaching over 90 per cent. Figure 3b is based on a scan with an extended spectral range from 4000 to 11000Å. The CSIRO tests show that the out-of-band filter transmission is 0.01 per cent or less up to 7600Å. Figure 3b shows that the filter does transmit redward of 7600 Å at up to $``$ 85 per cent, but the survey data will be unaffected by this as the Tech-Pan film used as detector is insensitive beyond 6990 Å.
While this satisfies the intended performance of the filter in light of normal incidence, in the $`f/2.48`$ beam of the UKST, light from an object in the field centre is focused into a cone and enters the filter at a range of angles up to 11.4. The bandpass of an interference filter is blue-shifted for light entering at an angle. This was modelled by breaking down the contributions from the light cone into a series of concentric rings of size 1 covering the telescope beam over a range 0.4 to 11.4, each entering the filter at a different angle. The contribution from the central part of the cone will not be significantly blue-shifted. The spectral shift was calculated for each ring according to Equation 1 adapted from Elliott & Meaburn (1976).
$$\lambda _\theta =\lambda _0\mathrm{cos}(\mathrm{arcsin}(\mathrm{sin}(\theta )/\mu ))$$
(1)
Here $`\lambda _0`$ is the chosen central wavelength for the filter bandpass in this case 6590 Å, $`\lambda _\theta `$ is the shifted central wavelength of the filter profile based on the angle, $`\theta `$, of the incident light and $`\mu `$ is the refractive index. A higher refractive index will minimise the blue shifting of the filter transmission with incident angle of light and the filter was designed with this in mind. Tests performed by the CSIRO using light at 0, 5 and 10 incidence found the effective refractive index of all the layers combined, ie. the effective monolayer, is $`\mu `$ = 1.34. This is the value used in Equation 1 to generate the shifts for the spectral response of the H$`\alpha `$ filter in the UKST beam. These shifts are shown in Figure 4a. The solid lines are the shifting response curves with the most red response curve being applicable to light of normal incidence and the most blue response curve tracing the filter response to light entering at the most extreme angle from the telescope beam.
In order to combine these to generate a smeared filter response curve which accounts for the telescope beam, each shifted bandpass is weighted by the area of the contributing ring as a fraction of the whole cone. The weighted response curves are shown in Figure 4b and the resulting, summed bandpass is shown in Figure 4c. The FWHM of this smeared bandpass is 80 Å, centred on $``$ 6550 Å.
This models the transmission of the filter in the centre of a survey field. Towards the edges of the field the shape of the cone changes and the maximum angle of incidence is over 14 which will shift the filter response further to the blue. The smeared out filter profile is significant as it permits calculation of the contribution of the contaminant \[NII\] lines at 6548 Å and 6584 Å, to the flux recorded by the survey. Based on the smeared out filter response shown in Figure 4c the filter transmits H$`\alpha `$$`\lambda `$6563 Å at 80 per cent, \[NII\]$`\lambda `$6548 Å at 82% and \[NII\]$`\lambda `$6584 Å at 50 per cent. Given that the \[NII\]$`\lambda `$6584 Å line is quantum mechanically fixed to be three times as strong as the \[NII\]$`\lambda `$6548 Å line (Osterbrock 1989), this gives a transmission of 58 per cent for any \[NII\] emission compared with 80 per cent transmission for the H$`\alpha `$ line. This is especially important when considering planetary nebulae (PNe) because the strength of the \[NII\] lines varies with respect to the H$`\alpha `$ line from PNe to PNe and will have a very significant impact on any calibration scheme based on PNe line flux standards if not taken into account. Of course, for general diffuse H$`\alpha `$ emission, the point to point H$`\alpha `$ to \[NII\] ratio is in general unknown without independent spectroscopic information, so we assume a \[NII\]/H$`\alpha `$ of 0.3, typically used for the warm ionised medium (e.g. Bland-Hawthorn et al. 1998).
### 4.2 Survey depth and quality control
The H$`\alpha `$ films are not sky-limited after a 3-hour exposure, but this was chosen as a pragmatic limit which optimises depth, image quality and survey productivity. Field rotation and atmospheric differential refraction can adversely affect longer exposures (Watson 1984) which are also more susceptible to short-term weather and seeing variations. The associated 15-minute broad-band SR exposures were taken through the OG590 red filter. At this exposure level they are well matched to the depth of continuum point-sources on the matching H$`\alpha `$ exposure. For completeness we include in Figure 5 the effective SR bandpass as a function of wavelength obtained from a calibration spectrogram for the OG590 filter in combination with the Tech-Pan emulsion.
With photographic surveys, the magnitude limit for a given survey field is not a fixed parameter but is a function of factors such as seeing, hypersensitisation and development of the films after exposure, emulsion batch variations and the brightness of the night-sky. Nevertheless, it is clear from comparison with the generally deeper, standard UKST R-band survey data, that the approximate magnitude limit for a typical H$`\alpha `$ survey field in an equivalent R magnitude for continuum point sources is $`20.5`$ (Arrowsmith & Parker 2001). This value can be directly determined by examining the number magnitude counts from the matched H$`\alpha `$ SR and R band SuperCOSMOS Image Analysis Mode (IAM) data (see later) for a given field and determining the point where completeness breaks down. As an illustration we give magnitude limit estimates for continuum point sources in A and B grade exposures of two H$`\alpha `$ survey fields in Table 2.
Additionally, the use of the same emulsion for both H$`\alpha `$ and SR exposures ensures an excellent correspondence of their image psf’s when film pairs are taken under the same observing conditions. The intention was to take the H$`\alpha `$ and SR exposures consecutively as far possible. This greatly simplifies the inter-comparability of both types of exposure. Of the 233 survey fields, only 100 are in fact sequential pairs while most of the rest were taken a few days apart. However 45 fields had a gap of one or more years between the H$`\alpha `$ and SR survey exposures because one or other of the exposures had to be repeated to satisfy the stringent survey quality acceptance criteria.
Strict quality control has been applied to the survey pairs by M.Hartley and S.Tritton according to well established criteria before any exposure is allowed to be incorporated into the survey. This ensures that the most uniform and homogeneous data set possible is created. Each exposure grade is determined by means of a score with ‘0’ being the best and ‘3’ being the limit for an exposure to be considered an ‘A’ grade (highest quality). The image grade is recorded in the information and data sheets which accompany the survey data, together with a letter code to indicate which is the most significant contribution to the score. Long, 3-hour exposures are prone to field rotation which can cause image trailing (denoted by T in the image grade), poor weather can lead to curtailed exposure times (U for underexposed). Cosmetic defects such as emulsion faults (E), haze halos (H) and processing streaks (P) can also contribute to a poor grade. These defects can be present in either the H$`\alpha `$ or the SR image. Where possible, any survey exposure which was not rated A grade was repeated. Unfortunately, a few B-grades had to be accepted into the survey though over 90 per cent were deemed survey quality, maintaining the high standards set for all UKST surveys.
## 5 Astrometric accuracy of the SHS
Astrometric calibration of survey photographic material measured on SuperCOSMOS is discussed in Hambly et al. (2001). The calibration procedure consists of applying a six coefficient (linear) plate model to measured positions of Tycho–2 catalogue reference stars, along with a radial distortion coefficient appropriate to Schmidt optics (i.e. $`\mathrm{tan}r/r`$) and a fixed, higher order two–dimensional correction map to account for distortion induced by mechanical deformation of the photographic material when clamped in the telescope plate holder to fit the spherical focal surface. As demonstrated in Hambly et al. (2001c), this yields absolute positional accuracy of typically $`\pm 0.2`$ arcseconds for glass plates. The SHS, on the other hand, employs film media which cannot be as mechanically stable as glass on the largest scales. However, provided a sufficiently dense grid of reference stars is available, it is possible to map out the unique distortion pattern that any one film may present.
In order to achieve the best possible astrometry for the SHS, the generic SuperCOSMOS Sky Survey (SSS) astrometric reduction procedure was modified by replacing the averaged distortion map with a correction stage where the individual film distortion pattern is measured with respect to the UCAC astrometric reference catalogue (Zacharias et al. 2004). In Figure 6 we show the results of comparing first-pass SHS astrometry (i.e. without correction of any higher order systematic distortion) with the UCAC catalogue for a single SHS film. Residuals have been averaged in 1 cm boxes and smoothed and filtered using a scale length of 3 box widths. A systematic distortion pattern is clearly seen, and comparing with figure 1 of Hambly et al. (2001) there is no four-fold symmetry in the pattern, which is a characteristic of mechanical deformation of rigid glass plates. Moreover, similar plots for different films show different patterns, so a fixed correction map cannot be applied across the entire survey film set. Figure 7(a,b) shows histograms of the residuals of individual UCAC standards from which Figure 6 is derived; a robustly estimated RMS (i.e. a median of absolute deviations scaled by 1.48, to be equivalent to a Gaussian sigma) is found to be about 0.4 arcseconds. Now, if the SHS positional data are corrected during the astrometric reduction procedure using the map values displayed in Figure 6, the RMS drops to $`0.3`$ arcseconds; the new histograms of individual residuals are displayed in Figure 7(c,d). The value of $`\pm 0.3`$ arcseconds can be taken as indicative of the typical global astrometric accuracy of the SHS in either co-ordinate, and compares favourably with the figure quoted for the SuperCOSMOS Sky Surveys (SSS) of $`0.2`$ arcseconds, given the higher level of crowding of the SHS fields.
## 6 The Survey SuperCOSMOS digital data
The high speed ‘SuperCOSMOS’ measuring machine at the Royal Observatory Edinburgh (e.g. Miller et al. 1992, Hambly et al. 1998) has been used to scan the H$`\alpha `$ and SR exposure A-grade pairs at $`10\mu `$m (0.67 arcsec) resolution. The same general scanning and post-processing reduction process is employed as for the directly analogous SuperCOSMOS broad-band surveys of the Southern Sky (SSS) currently on-line and outlined in detail by Hambly et al. (2001 a,b,c). The user interface is broadly equivalent and the main features are summarised neatly in Figure 1 of Hambly et al. (2001a). However, due to the special nature of the survey, some additional processing steps and H$`\alpha `$ specific options have been added to create the on-line SuperCOSMOS H$`\alpha `$ Survey (SHS) described below.
### 6.1 Basic characteristics of the on-line ‘SHS’ H$`\alpha `$ Survey
The Wide-Field Astronomy Unit (WFAU) of the Institute for Astronomy Edinburgh is responsible for maintaining the H$`\alpha `$ survey data products. Both the H$`\alpha `$ and SR data for the 233 Southern Galactic Plane survey fields are available on-line <sup>3</sup><sup>3</sup>3http://www-wfau.roe.ac.uk/sss/halpha. Unfortunately, there are no plans for the 40-field Magellanic Cloud H$`\alpha `$ and SR survey pairs to also be put online. The data products are given as FITS files (see http://heasarc.gsfc.nasa.gov/docs/heasarc/fits.html) with comprehensive FITS header information detailing key photographic, photometric, astrometric and scanning parameters (e.g. Hambly et al. 2001b). The FITS images also have an accurate built-in World Co-ordinate System (WCS). This permits easy incorporation into other software packages such as the STARLINK GAIA environment for subsequent visualisation, investigation, manipulation and comparison with other data. The entire survey data are stored on RAID disks for fast access and a comprehensive set of web-based documentation has been provided. The pixel data map for each field is about 2 Gb. The scanned pixel data are processed through the standard SuperCOSMOS thresholded object detection and parameterisation software (e.g. Beard, MacGillivray & Thanisch 1990) to produce the associated Image Analysis Mode (IAM) data for each field. This process determines a set of 32 image-moment parameters which provide the astrometry, photometry and morphology of the detected objects. Full details of the image detection and parameterisation are given in Hambly et al. (2001b). For the SHS survey, a selection of the 32 most important IAM parameters from the merging of the H$`\alpha `$, SR and I band data for each detected image in the SHS are available and are given in Table 3.
The full resolution, $`10\mu `$m pixel data and associated IAM parameterised data for both the H$`\alpha `$ and SR scanned exposures are stored on-line on a field by field basis. On the SuperCOSMOS web-site the scanned survey data for each field has the prefix ‘HAL’ before the survey field number (so H$`\alpha `$ survey field h350 = HAL0350 for example, when referring to the on-line digital SuperCOSMOS data). The SR images have been transformed to exactly match the pixel grid of the master H$`\alpha `$ exposures which permits direct image blinking and comparison between the pixel data for each field. The general H$`\alpha `$ survey data products are accessed via a web interface that has the same look and feel as the existing broad-band SuperCOSMOS on-line ‘SSS’ surveys but with some additional functionality. The IAM data produced for each field can be downloaded separately if desired or assembled into seamless catalogues on-the-fly which can cover several adjacent fields using the ‘Get a Catalogue’ option. The combined IAM data is organised into a full listing of 53 image parameters or a more manageable subset of the most useful 32 as in Table 3. A set of ‘expert’ options are also available to further select catalogue extraction parameters. A special feature to create a difference image of each field following variable image psf matching techniques developed by Bond et al. (2001) also exists to permit large-scale resolved emission maps to be created with reduced artefacts from uncancelled stars. This can be computationally intensive and so is not generally available without prior arrangement with the Wide-Field Astronomy Unit. For most applications simple quotient imaging between the H$`\alpha `$ and SR pixel data is sufficient due to the well matched psf’s and depth.
A $`16\times `$ blocked-down version of each field is also available as both a GIF image and as a FITS file which has the WCS built in to the FITS header. These whole field maps can be studied to select smaller regions of interest for extraction at full resolution using the ‘Get an Image’ option. The full resolution pixel data access limit is currently set at $`9000`$ arcmin<sup>2</sup> with regions downloaded as FITS files (also with WCS) and both the SR and H$`\alpha `$ data for the same region can be downloaded simultaneously. Areas for extraction can be chosen via equatorial (J2000 or B1950) or galactic (l,b) co-ordinates in a variable $`m\times n`$ arcminute rectangular region format. A clickable map of the current fields on-line enables individual field details to be displayed prior to viewing the blocked full field image. A batch mode enables large numbers of thumb-nail images to be extracted around objects of interest with the option to return H$`\alpha `$ and/or SR postscript plots of the extracted images. An option to apply a ‘Flat-Field’ to the H$`\alpha `$ pixel data in intensity space is included to permit correction of the non-uniformities in the measured exposures arising from the excellent but slightly varying H$`\alpha `$ filter transmission profile. This has been shown to work effectively and is described in Section 9. A radius of 153 mm ($`2.85`$ degrees) from each survey field centre has been adopted as the region with good data ($`<15`$ per cent correction factors). The ‘good data’ radius from each scanned H$`\alpha `$ field centre has been used in creating a confidence map which is incorporated into the extracted FITS image as an additional FITS extension (extension , e.g test-image.fits). This can be used to flag areas of the extracted image that might not be quite as good as others. Currently this has values of 100 for regions extracted interior to this radius and 0 for regions outside.
### 6.2 Incorporation of the SSS ‘$`I`$’ band data
The IAM catalogue downloaded directly via the ‘Get a Catalogue’ option or as incorporated in the FITS table extension to the dowloaded pixel data via the ‘Get an Image’ option, contains information not just from the H$`\alpha `$ and the SR images but also from the SERC-I (near infrared) survey which has been carefully matched in with the SHS data. The I-band data is particularly useful when searching for point-source H$`\alpha `$ emitters as it can help to eliminate contamination from late-type stars. However there are issues that any user should be aware of when combining the I-band data with the SHS magnitudes. When calibrating UKST data, positional and magnitude-dependent systematic errors are present as a result of variations in emulsion sensitivity and vignetting towards the image corners (Hambly et al. 2001b). The I-band survey was taken on standard ESO/SERC 5 field centres, so the vignetting will have a different effect on the two sets of photometry at a given survey location. Furthermore, the I-band data are calibrated to relatively few standards. These differences are evident when looking at a plot of SR magnitude versus $`RI`$ colour derived from the SR and paired I-band photometry. Figure 8 shows two colour magnitude diagrams (CMD) for stars taken from a 10 arcminute region centred on the middle of SHS field h1109 in Monoceros where the low galactic reddening of $`E(BV)`$ = 0.24 (Schlegel, Finkbeiner & Davis 1998) should leave the $`RI`$ colour roughly constant for much of the observed magnitude range. Figure 8a shows the raw result, where a large, unphysical variation of 3.5 magnitudes is seen in the $`RI`$ colour from the brightest to the faintest objects. This is removed as a first-order correction from the survey data by selecting a master colour, in this case the SR, and correcting the I-band across all survey fields. Note however that at fainter limits one in fact expects redder $`RI`$ colours as such stars are likely to be further away and prone to be more dust reddened or intrinsically fainter and therefore more likely to be late types. Hence some modest slope is expected. Figure 8b shows the CMD for the same patch of sky after the colour correction has been applied, giving a result in better agreement with expectation. The data can be downloaded in corrected or uncorrected form via an option in the “Expert” parameters of the SHS website. It is important to ensure that the I-band correction is not applied inappropriately, i.e. in a field of intrinsically high reddening, as such a correction will remove genuine features from the data. It should be used with caution.
## 7 SHS point source photometry
A significant advantage of the SHS data over its rivals is the ability to detect point sources which have been photometrically calibrated to CCD standards (e.g. Boyle et al. 1995, Croom et al 1999). With measurements of isophotal magnitude and object classifications, it is possible to apply a photometric calibration to the H$`\alpha `$ and SR films by comparing the SuperCOSMOS raw magnitudes of stars from the Tycho-2 Catalogue (Hog et al. 2000) and the Guide Star Photometric Catalogue (Lasker et al. 1988). These in turn are checked against photometric standards derived from the CCD observations given by Croom et al. (1999) and Boyle et al. (1995). The narrow-band H$`\alpha `$ images are calibrated to an ‘R-equivalent’ scale. The 3 hour H$`\alpha `$ and 15 minute SR exposures are matched so that both reach similar depths of m<sub>R</sub>$``$ 20.5 for point sources. Where an object is detected in one band but not in the other a default value of 99.999 is given in the catalogue data for the magnitude in the missing bandpass. Positional and magnitude dependent errors are seen in the raw photometric data, created by varying transmission profile and diffraction effects through the thick (5.5 mm) H$`\alpha `$ filter, but these are corrected for in the data available through the SHS website by comparison with the SR data. Photometric consistency is achieved by using the overlap regions between fields to match zero-points across the survey. These corrected magnitudes provide a means of selecting point-source emitters. The variations in measured IAM stellar parameters as a function of field position arising from the variable psf from field rotation, vignetting etc, especially at large radii from the field centres, requires that such selection is performed over limited 1-degree areas. In this way stars with an emission line at H$`\alpha `$ will show an enhanced H$`\alpha `$ magnitude compared with the SR magnitude. At the bright end of the magnitude distribution, severe photographic and SuperCOSMOS saturation effects come in to play, limiting stellar photometry to R of about 11-12 in both the H$`\alpha `$ and SR pass-bands.
## 8 Spurious images in the SHS
Spurious images appear from time to time in all photographic images scanned by SuperCOSMOS. They have a variety of forms and causes and are present in images extracted from the SSS as well as SHS. They can sometimes be picked up by examination of the pixel images directly, though they are often missed, and can also appear as spurious detections in the IAM data. They have a variety of sizes and shapes and may be in or out of focus depending on whether the contaminating source is on the emulsion surface or on the platten used by SuperCOSMOS to sandwich the film flat for scanning. Here, we differentiate between spurious images in the emulsion itself caused by processing defects, emulsion flaws and static marks, and those caused by foreign objects on the surface of the emulsion or on the back of the film. Holes and scratches in the emulsion surface can also give rise to spurious images. Satellite trails and transient phenomenon also give rise to real developed images which may have no counterpart in other survey bands of the same region. We do not consider these here.
### 8.1 Basic causes
The SuperCOSMOS facility is situated in a class-100 clean room and each film is pressure air-cleaned prior to scanning. However, despite best efforts, particles that may already have been present on the emulsion before shipment to SuperCOSMOS manifest themselves as spurious images. The biggest cause is fine particulate dust (20–100$`\mu `$m). Unfortunately the Estar film base of the Tech-Pan emulsion is prone to static charge build-up.
### 8.2 Recognising spurious images
The SuperCOSMOS scanning system is highly specular so detritus present on the emulsion surface that is often invisible when viewed under diffuse illumination conditions (such as on a light table) is revealed in sharp relief in the SuperCOSMOS data. The number of artefacts seen in the SHS data is somewhat worse than on other glass plate based surveys of the SSS. Fortunately, having matched exposures in two bands makes identification of such artefacts more straightforward. For example, since the H$`\alpha `$ and SR exposures are registered on the same pixel grid, quotient imaging can reveal the locations of spurious images. A $`5\times 5`$ arcminute region extracted from h273, a field with a particularly high number of spurious images, is shown in Figure 9.
We can take advantage of the fact that the pixel image properties of spurious images are usually quite distinct from real astronomical images, often having a sharpness below that possible from the combination of telescope optics and seeing disk. Their shapes are often highly irregular and non-symmetrical such that they would not fit any normal psf. This makes them amenable to Fourier filtering. Objects that have no counterpart in the other band are potential spurious image candidates though variable objects, novae and the effects of de-blending complicate the issue significantly. Various IAM parameters such as the profile statistic, ellipticity etc. may also aid in identification. Furthemore, spurious IAM objects arising from de-blending overlaying contaminating fibres or hairs usually have very high ellipticities which may help in isolating likely candidates. Storkey et al. (2004) discuss techniques for recognising and eliminating spurious objects in the on-line SuperCOSMOS surveys. As yet this procedure has not been applied to the SHS data.
## 9 Flat-Fielding of the survey data
For any interference filter of the size used here, low-level non-uniformities exist which lead to residual non-physical background variations in the exposed images. In order to establish the magnitude of such effects, three flat-field exposures were taken with the filter subject to uniform illumination. The flat-field images permitted evaluation of the combined effects of filter transmission in the fast, $`f/2.48`$ Schmidt beam and telescope vignetting (see UKST Unit handbook, Tritton, 1983). The flat-field images were exposed to place them on the linear portion of the film’s characteristic curve and were averaged to give the filter/telescope transmission profile shown in Figure 10a-b.
Figure 10a shows transmission contours at 85, 90, 92, 95, 97 and 98 per cent of maximum transmission in the central region. The response is seen to be asymmetric, with the 97 per cent contour extending beyond the edge of the 5.16 field on the right, which corresponds to the west of the survey fields. In the east the transmission decreases more rapidly, reaching 85 per cent at the eastern edge. Towards the filter corners the transmission drops further. However, the 4 overlapping centres (Section 4) and the asymmetric nature of the response allows the selection of H$`\alpha `$ data requiring flat-field correction of less than 15 per cent for any given area of sky, provided that adjacent fields are available. Most data will require much smaller flat-field corrections. The effect of flat-field corrections as large as 15 per cent on pixel data is considered in the survey calibration section. Figure 10b is a histogram equalization of the actual flat-field pixel map which reveals the extremely low level artefacts present at the 0.1 per cent level which are invisible in a linear rendition. The flat-field correction has been stored as a transmission array with maximum value unity, so it is applied by dividing survey image values in intensity space by the relevant correction array elements. This is available as the default option on the SHS website. The correction breaks down towards the corners of the scanned image and in regions outside of the clear aperture because the density of the exposures at the edges of the circular aperture is too low to lie on the straight line portion of the characteristic curve leading to an over-correction and also because the H$`\alpha `$ filter transmission is becoming increasingly skewed in these extreme regions. Raw transmission or photographic density values and generically calibrated intensity values without flat-field correction can also be requested on the download form. The IAM data is obtained from the raw SuperCOSMOS scans without flat-field correction.
### 9.1 Specific filter features
Despite the superb quality of the filter, low-level, large-scale variations in transmitted flux can be seen in the H$`\alpha `$ survey images under certain exposure conditions. In particular there are two parallel bands of slightly enhanced transmission (leading to elevated photographic density) going E-W in the north and south of the filter. These bands are only 1–3 per cent higher in intensity than the surrounding regions. A series of low-level artefacts which are not obvious in the contour plot because they are at a level of $`<`$ 1 per cent can just be discerned in the filter transmission image in Figure 10a-b. They can also just be seen in the contour plot as the spike in the 98 per cent contour just right of centre. The observed shape of these artefacts mimics the series of shallow concentric grooves scored into the surface of the mandrel to enable the Tech-Pan film to be sucked under light vacuum to the curved focal surface of the plate-holder to ensure proper focus. They are thought to arise from the backscattering off these grooves of light that has passed right through the film. Again, it is gratifying that these artefacts, present at the $`<0.1`$ per cent level, are effectively removed by application of the flat-field.
### 9.2 Application of the flat-field and correction validity
Field h410, which sits away from the Galactic Plane on the extreme edge of the survey at ($`l,b`$) 330.2, +10.28 has been chosen to test the vailidty of the flat-field correction as it contains a very low-level isotropic background of Galactic line emission. In the survey image this will be moderated by the filter response. Figure 11a-b shows two 16$`\times `$ blocked down images of survey field h410. The top image Figure 11a is the raw H$`\alpha `$ data, before the application of the flat-field correction. The structure evident on the field as lighter areas is not Galactic emission but matches the filter transmission profile. The two horizontal bars are present and the image is less exposed towards the corners where the recorded intensity is lower and the star density also falls. The bright star just to the right of centre is $`ϵ`$ Lupi.
Application of the flat-field correction results in Figure 11b which should have a flat background wash of emission across it. The structure from the filter is no longer evident and the bright areas are now in the corners, where the larger flat-field correction over-corrects the SuperCOSMOS intensity counts. This will not adversely affect the majority of the pixel data available on the SHS website. Data can always be taken from the best area of filter response and no flat-field correction larger than 15 per cent is necessary for any of the survey data that overlap a neighbouring field. Data from the edge of the survey, where no adjacent field exists, have been made available and may require a correction greater than 15%. Areas affected in this way are flagged in the third extension table which accompanies the downloaded FITS image.
## 10 Geocoronal H$`\alpha `$ emission
Geocoronal H$`\alpha `$ emission is caused by fluorescence after solar Lyman $`\beta `$ excitation of atomic hydrogen in the exosphere. Because imaging surveys lack velocity resolution for the emission they record, the geocoronal contribution will be present in all of them but indistinguishable from bona fide Galactic H$`\alpha `$ emission. Fortunately, one modern H$`\alpha `$ survey, WHAM (e.g. Haffner et al. 2003), offers very good velocity resolution ($``$12 km/s) and is able to separate the atmospheric emission from the Galactic emission and measure the intensity. Nossal et al. (2001) report on H$`\alpha `$ observations carried out by the WHAM instrument in 1997, the same time as the SHS imaging was starting at the UKST. They find that the geocoronal emission intensity depends on how much the earth shades the line of sight from sunlight. Their resulting plot of geocoronal H$`\alpha `$ emission as a function of earth shadow height shows that at heights greater than 6000 km only a very low level $``$2 R wash of geocoronal H$`\alpha `$ emission is present. Based on the observational information available in the headers of the SHS images, it is possible to calculate the shadow heights for any field. For a random sample of 6 SHS fields the shadow height for the whole three hour observation and across the five-degree field of view was found to be greater than 6000 km, so low-level geocoronal emission is not problematic, as most Galactic Plane fields covered by the SHS will contain significantly stronger emission.
## 11 Application of an absolute Calibration to the H$`\alpha `$ Survey data
The AAO/UKST H$`\alpha `$ survey data needs an absolute intensity calibration if the full scientific value of its sensitivity to faint, diffuse emission is to be realized. The intensity calibration must provide a reliable means of transforming the pixel intensity values from SuperCOSMOS scans of the H$`\alpha `$ images into meaningful intensity units such as Rayleighs which is consistent from field to field. We show that continuum emission can be successfully removed from the H$`\alpha `$ images by scaling and subtracting the SR continuum image. Unlike CCD data, which enjoys a linear response over a wide range of emission strength, photographic data can be very difficult to calibrate because the response of the emulsion and SuperCOSMOS scanner is linear only over a relatively small dynamic range. Variations in sensitivity and background occur from exposure to exposure, especially when the H$`\alpha `$ and SR pairs were taken on different nights, phases of the moon etc. Despite this, we show that the survey data have been well exposed to capture Galactic emission on the linear part of the characteristic curve and can be calibrated by means of comparison with the complementary, accurately intensity calibrated SHASSA survey (Gaustad et al. 2001). This process does not form part of the current SHS release but can be undertaken by the user as required.
### 11.1 Image comparison with SHASSA
The Southern H$`\alpha `$ Sky Survey Atlas by Gaustad et al. (2001) provides wide-field narrow-band CCD H$`\alpha `$ images of the southern sky below $`\delta `$ = +15 taken with a robotic imaging camera sited at Cerro Tololo Inter-American Observatory (CTIO). The camera used a small, fast, f/1.6 Canon lens which gave a very large (13) field of view and a spatial resolution of 48 arcseconds. Each SHASSA field was imaged through a narrow-band interference filter of width 32 Å centred at 6563 Å as well as a continuum filter with two bands of 61 Å at 6440 Å and 6770 Å, on either side the H$`\alpha `$ line. The SHASSA website<sup>4</sup><sup>4</sup>4http://amundsen.swarthmore.edu/SHASSA/ makes available the raw H$`\alpha `$ and continuum images as well as 48 arcsecond and 4 arcminute resolution continuum subtracted, intensity calibrated data.
The SHS data is superior in terms of resolution while the general sensitivity of both surveys appears qualitatively similar for large-scale emission features. For example, in Figure 12 we present SHS and SHASSA images of the HII region RCW 19.
### 11.2 The SHASSA intensity calibration
The SHASSA intensity calibration was derived from the planetary nebula spectrophotometric standards of Dopita & Hua (1997) after the continuum images had been scaled and subtracted from the H$`\alpha `$ frames. Aperture photometry for eighteen of the bright PNe standards was measured from the SHASSA images and used to calculate the calibration factor for the whole survey. A difficulty in applying PNe line fluxes to H$`\alpha `$ narrow-band imaging is the proximity of the two \[NII\] $`\lambda \lambda `$6548, 6584 lines which are included in the flanks of the SHASSA H$`\alpha `$ filter bandpass. These vary in strength relative to H$`\alpha `$ between PNe and could significantly affect the result. Calculating the transmission properties of the interference filter to these lines is complicated by the blue-shifting of the bandpass with incident angle, an effect which must be treated carefully in the SHS data as the filter sits in the fast f/2.48 beam of the UKST. This problem is not as severe for the SHASSA data because in this case the filter sits in front of the camera lens, leaving only the effects of the very large field of view. These effects are considered in Section 4 of Gaustad et al. (2001) and carefully accounted for in their calibration.
To allow a more detailed comparison, aperture photometry for 87 PNe with a range of surface brightness and integrated flux and with an independent measure of H$`\alpha `$ flux, was carried out on SHASSA images. Published spectroscopic data were used to deconvolve the contribution from the \[N II\] lines passed by the SHASSA filter. The results agree with published data to $`\mathrm{\Delta }`$ F(H$`\alpha `$) = –0.01dex, $`\sigma `$ = 0.05 for SHASSA minus literature fluxes (Frew 2005, in preparation; cf. Pierce et al. 2004). Since the PNe literature fluxes have associated errors, the SHASSA calibration is better than $`\pm 10`$ per cent across the whole survey, in agreement with the nominal error supplied by Gaustad et al. (2001).
An additional uncertainty is introduced to the zero-point of the SHASSA intensity calibration from the contribution of geocoronal emission. Gaustad et al. (2001) estimate this by comparison with overlapping WHAM data points and interpolating where there are none. Our check of the intensity calibration against independent flux measures of planetary nebulae indicates that the geocoronal contribution to the SHASSA H$`\alpha `$ images has been successfully removed. Finkbeiner (2003) also showed there is no significant offset between WHAM and SHASSA data. So we conclude that the SHASSA data has been well calibrated to a zero-point consistent with independent measurements and therefore have confidence in its use as a baseline calibration for the SHS data.
### 11.3 Continuum Subtraction of the SHS
Diffuse emission recorded through the narrow-band H$`\alpha `$ filter on the Tech-Pan films will be a combination of Galactic H$`\alpha `$ line emission, continuum emission, night-sky auroral lines and geocoronal emission. Ideally all of these components would need to be disentangled to extract just the Galactic H$`\alpha `$ emission. In practice the geocoronal and auroral emission is considered as a low ($`2`$ R) level but temporally varying uniform wash which simply elevates the general background on each exposure to a slightly varying degree.
The matching SR images provide a measure of the continuum component and, properly scaled, can be used to produce continuum subtracted H$`\alpha `$ images. Although the H$`\alpha `$ and SR exposures are generally exposed to attain the same depth for continuum point sources, the nature of photography and the vagaries of the observing conditions (e.g. if the exposure pairs were not contemporaneous and taken in different moon phases or if the seeing changed) mean that the depth and image quality between the H$`\alpha `$ and SR exposures can and does vary. Hence it is necessary to determine a continuum subtraction scaling factor between H$`\alpha `$ and SR on a field by field basis. This factor must be precisely determined for the continuum subtraction to be effective. For high-dynamic range CCD exposures, the standard method for determining the appropriate scaling factor to subtract continuum from narrow-band is to compare aperture photometry for stars on both images whose exposures are normally interleaved on short timescales.
Unfortunately, this does not work well with the H$`\alpha `$ SR film exposure pairs (Pierce 2005), often leading to under or over-subtraction of the continuum. This arises due to varying backgrounds on the film exposures caused by; low-level emulsion sensitivity variations between films (especially if they come from different hypersensitised batches), inherent chemical fog variations in the emulsion, processing variations and true sky background variations arising for the reasons given above. These varying backgrounds result in the same magnitude stars saturating at different levels on different exposures as their Gaussian point spread functions are superimposed on top of any diffuse emission and elevated background which can severely truncate their peaks. The limited dynamic range of SuperCOSMOS also acts as a further low ceiling above the background, which leaves little room for these bright stellar Gaussians making it hard to effectively utilise stellar photometry to determine the correct scaling factors.
Fortunately, we are able to use the existing SHASSA data to provide a well-determined scale with an independently confirmed zero-point to compare with and calibrate the SHS images. Even exposure pairs taken years apart can be successfully continuum subtracted. A detailed investigation of the SHS calibration process has been undertaken by Pierce (2005) but the essential aspects of this process and its application are given here. For example, Figure 13 shows three images of a 30 arcminute region taken from field h350, which shows strong, varying Galactic H$`\alpha `$ emission. The top image is the H$`\alpha `$ image downloaded from the SHS website, the middle image is its SR counterpart while the bottom image has had the SR ‘continuum’ image scaled and subtracted via a comparison with SHASSA data. In general this subtraction is very good, removing most of the stellar images and the diffuse continuum. Only the stars which sit in the strongest emission have been over-subtracted and appear as white spots in the image.
For a given area of sky, pixel data from each survey can be downloaded and, after matching for spatial resolution, the SuperCOSMOS intensity counts can be compared directly with the Rayleigh values in the SHASSA data. A plot of continuum-subtracted SHASSA pixels against equivalent SHS pixels should return a linear relation with a common zero point if the reduction and intensity calibration have been properly carried out. Comparing incorrectly continuum-subtracted SHS data with SHASSA data results in an offset between the two surveys. A range of values for the scaling factor can be applied to the SHS data until the zero-point of the continuum-subtracted UKST survey images best matches the zero-point of the equivalent SHASSA data, indicating the appropriate value to use. A calibration based on the SHASSA data will provide an advantage for the SHS over the CTIO survey as it can be applied to the full resolution pixel data. This offers the chance to determine intensities for emission structures not resolved by SHASSA such as the new sample of extended PNe discovered from the SHS data (e.g. Parker et al. 2003, 2005, Pierce et al. 2004, Frew & Parker, in preparation and see Section 14.1).
Each scanned SHS survey field, at the full 0.67 arcsecond resolution, contains over 2 Gb of data so it was not practical to download and compare all the pixel data for each field. Instead, most of the emission variation on a given survey field can be sampled using carefully selected 30 arcminute regions. For fifteen fields, several 30 arcminute areas were downloaded to sample the complete dynamic range of emission present. Once the scaling reliability over a range of flux levels and central aperture locations was established, the best single, 30 arcminute region was chosen from the whole frame for the rest of the 233 survey fields to provide the base calibration for each field. The H$`\alpha `$ filter flat-field correction was applied to remove the low-level non-uniformities in transmission across the narrow-band filter.
Data requiring a flat-field correction of up to 15 per cent have been shown to be suitable for inclusion in the survey, though in most cases data returned from the SHS website require less correction than this. For the regions considered here, the pixel data only required flat-field correction $`3`$ per cent in most cases and, in general, no more than 8 per cent. For two fields, h350 and h1109 regions well away from the best area of the filter were also selected to confirm that the pixel data behaves as expected when larger flat-field corrections are required. The results are discussed below.
### 11.4 The SHS Calibration process
A detailed description of the SHS survey calibration process based on zero-pointing each SHS field to SHASSA is given in the thesis of Pierce (2005). The essentials of this scheme are described here. Each SHS field is completely covered by just one 13 SHASSA field and data from the best area of SHASSA filter response was chosen for comparison with the SHS images. For direct comparison, the H$`\alpha `$ and SR SHS data at 0.67 arcsecond was re-binned to match the 48 arcsecond SHASSA pixels using the IDL routine Hrebin<sup>5</sup><sup>5</sup>5Interactive Data Language: http://www.rsinc.com. This returns the mean value of the 72 $`\times `$ 72 full resolution SHS pixels that constitute a single SHASSA 48 arcsecond pixel, so the calibration factor determined from the comparison plot applies to the full resolution SHS data. At this coarse resolution the SR data was scaled and subtracted from the H$`\alpha `$ image. Because the correct scaling factor was not yet known, a range of scaling factors from 0.4 to 2.0 was applied so that the best value could be selected by matching the SHASSA zero-point. The equivalent area of SHASSA data was selected, aligned and trimmed to match using the IDL, Hastrom routine. These images were then compared directly, pixel by pixel, with the re-binned, continuum-subtracted SHS data to give a plot of SHASSA values in Rayleighs versus SuperCOSMOS counts per re-binned pixel. The linear portion of the resulting comparison was then fitted to determine the number of SuperCOSMOS counts per 0.67 arcsecond pixel per Rayleigh. Averaging SuperCOSMOS data in this way only works properly if all the SuperCOSMOS pixels are on the linear portion of the characteristic curve at the faint end and unsaturated at the bright end. Once the SuperCOSMOS dynamic range is exceeded pixel saturation arises and the SHASSA to SuperCOSMOS linear relation breaks down as the SuperCOSMOS flux becomes increasingly underestimated (e.g. Phillipps & Parker, 1993) as seen in Figure 17.
The adopted process was followed for the same 30 arcminute region from SHS field h350 as shown in Figure 13. The upper image in Figure 14 shows the SHS image with the continuum-subtracted at 48 arcsecond resolution. The lower image is the trimmed and aligned SHASSA data. Bright stars on the continuum subtracted SHASSA images leave significant residuals while in the SHS data any stellar residuals are barely visible and then only for the very brightest stars. The pixel-to-pixel comparison plot for this area is shown in Figure 15. Each point on this plot is the pixel value from the SuperCOSMOS scan against the Rayleigh value from the SHASSA data. Immediately, it can be seen that the SHS data follows a tight linear relation with the SHASSA values over a range of several hundred Rayleighs. The outliers from this distribution at around 4000 SuperCOSMOS counts correspond to the SHASSA stellar residuals noted from Figure 14.
A further nine 30 arcminute areas were examined from field h350 to cover the whole dynamic range of diffuse emission evident in the field. Figure 16 shows the 16$`\times `$ blocked down H$`\alpha `$ image for this field with contours of filter response overlaid and boxes indicating the areas used. The large area of strong emission in the west of this field is caused by the UV flux from OB association Ara 1A A (Mel’Nik & Efremov 1995). In all but two cases the areas were selected from the best area of the filter response, the exception being the two areas extracted from the SE corner which were included to test the effect of using pixel data from the edge of the SHS survey fields where the flat-field correction is larger.
Figure 17 shows the resulting SHASSA versus SHS comparison plot with differently shaded points belonging to different 30 arcminute areas. Almost the full range of the SHS data is shown, with the pixel values showing a good linear relation to the SHASSA data from the faintest emission on the field at $``$ 20 Rayleighs right up to $``$ 500 Rayleighs. The curve in the trend beyond 500 Rayleighs is due to saturation effects with the SHS data. These points were discarded when making the calibration fit.
The reciprocal of the slope of the linear part of the relation provides a calibration factor of 15.1 counts/pixel/Rayleigh to convert the SuperCOSMOS intensity counts to Rayleighs for this field. An estimate of the error in this calibration is possible, based on the vertical scatter of points about this trend ,as a given SuperCOSMOS value matches a Rayleigh value in this vertical distribution. In this field the 1$`\sigma `$ scatter is 21 Rayleighs.
The data taken from the two areas in the SE corner of the SHS survey field required flat-field correction of up to 20 per cent. They behave very well when compared with the data requiring less correction, neatly overlaying the main trend, with no change in slope or increase in scatter. This justifies use of the data out to the 15 per cent flat-field contour where the emulsion records strong emission.
### 11.5 SHS sensitivity limit
While the results from field h350 have shown that the survey has been well tuned to the detection of diffuse emission, as well as giving an approximate limit to the point at which the photographic emulsion saturates, the faint limit has not yet been constrained because there is no really faint emission on this field. At the faintest point of 20 Rayleighs there is no sign of either survey struggling for sensitivity so a field exhibiting fainter emission is required to probe the SHS faint sensitivity limit. According to the SHASSA data an area of the Southern Galactic plane in Monoceros harbours diffuse emission that reaches a level as faint as $``$ 2 Rayleighs, which is ideal to test the faint limit of the SHS data. Figure 18 shows the 16 times blocked down H$`\alpha `$ image of SHS field h1109, which covers this area of sky, with contours of filter response and the areas selected shown as with Figure 16. On this field, nine 30 arcminute regions have been examined as two groups: one, labelled 1–5 in Figure 18, from the area of best filter response and the other, labelled A–D, probing the combined effects of extremely low levels of emission and decreased filter transmission.
The comparison plot from the first group is shown in Figure 19. Once again, there is a clearly defined relation between the two, although on this faint field the upper saturation limit is not reached. It is immediately obvious that the SHS data can match the SHASSA data right down to the faintest level of emission, although the linear response of the SHS data is difficult to determine at this low level. The SHS data are therefore detecting emission structure as faint as 2 Rayleighs on this field, although this sensitivity is tempered by the scatter evident in the plot and in the examination of the areas from the area of poor filter transmission discussed below. The linear fit returns a value of $`7.7\pm 0.1`$ SuperCOSMOS counts/pixel/Rayleigh for this particular field and provides a reasonable calibration as the 1$`\sigma `$ scatter to the fit is just 6.2 Rayleighs. Note the factor of two difference in the slope of the calibration curve for this low emission level field h1109 (no emission measure greater than about 80 Rayleighs seen in the area considered) compared to that obtained for high emission field h350 in Figure 17, which returned 15.1 counts/pixel/Rayleigh, which is closer to that generally obtained for most fields. This serves to emphasise the need for individual field calibration due to the variation in SuperCOSMOS pixel intensities on a given field, arising primarily from variable fog-level, sky background and resulting SuperCOSMOS and emulsion saturation.
### 11.6 Fifteen Fields Studied in Depth
A total of fifteen fields from a wide variety of Galactic environments were studied using several 30 arcminute regions in each in order to build up a global picture of the survey behaviour. The comparison plots for 4 of these are shown in Figure 20 with the linear fit overlaid. In each case a clear, essentially linear relation can be seen between the SHS and SHASSA data. Generally, different 30 arcminute regions follow a single trend which indicates little variation in emulsion sensitivity across the large SHS images. Three of the fields examined, h175, h350 and h555 show evidence for saturation at the bright end and fix the bright limit at $``$ 500 to $``$ 600 Rayleighs, while none of the fields appear to reach the background fog level in the areas that were compared.
### 11.7 Calibration of the entire SHS
From the analysis of the fifteen fields considered above, it is clear that each field requires individual calibration and that one, well selected, 30 arcminute area of SHS pixel data can be compared with SHASSA to give a working estimate of the calibration factor and a satisfactory result in most cases. On this basis, the 30 arcminute areas of pixel data which covered the greatest dynamic range of emission were downloaded from each of the 233 SHS fields, re-binned and compared with the equivalent area of SHASSA data. The results for each field can be found on the SHS web site. For each field the position of the area used, the computed scaling factor for continuum subtraction, the linear fit and, where appropriate, the coefficients from a third-degree polynomial fit are given. The 1$`\sigma `$ vertical scatter about the linear fit is also quoted to offer an estimate of the error in the calibration. Of the 233 survey fields, 76 are relatively featureless and exhibit little emission. These are difficult to fit and in 9 cases the fit failed completely. Forty three of these show evidence of the low-level photographic fog. Of the remaining 157 fields, 122 are well constrained by a linear fit. For the other 35 fields the fit can be improved with a low-order polynomial relation. Where this is the case the coefficients are included in the table on the SHS website.
### 11.8 Calibration check of SHS field overlap regions
There is generous overlap between survey fields because of the circular aperture of the filter which allows field-to-field consistency check of the calibration. For six overlap regions between eight fields, a 30 arcminute area was carefully selected from the best possible compromise of filter response between two fields, never requiring flat-field correction greater than 15 per cent. The calibration factor calculated from the field centre in the best region of filter response was applied to these pixel data from the edge. For seven of the eight fields the calibration factor determined from the linear fit to the data was used. Calibrated and aligned data from the two overlapping SHS fields were plotted pixel by pixel at 0.67 arcsecond, 10 arcsecond and 48 arcsecond resolution. If the independently determined calibration applied to each field is consistent, the resulting plot of Rayleighs from one field against Rayleighs from its neighbour should yield a linear relation with slope of unity and no offset.
From the six fields examined in this way the results give good agreement. Four examples are shown in Figure 21 from comparisons at 48 arcecond resolution. Here the ordinate and abscissa values are in Rayleighs with each axis labelled with its field. The fits are quoted on the plots and in Table 4. Two comparisons, 459-392 and 459-458, agree to better than 10 per cent, three more, 391-329, 458-391 and 460-459 also return linear results but agree to just 14, 29 and 43 per cent respectively. The bottom 3 comparisons are for overlapping fields so the good match implies that one could construct seamless, large pixel mosaics. For the h350 and h349 the linear relation is not so well behaved. The slighly curved trend between these two fields probably results from the bright saturation evident in the pixel data of the overlap region used (note the higher Rayleigh limits for this comparison).
## 12 Comparison with IPHAS
The SHS and IPHAS H$`\alpha `$ surveys have areas of overlap at low declinations which permit a direct comparison to be made between the two complementary surveys. Figure 22 shows a $`3.3\times 2`$ arcminute region centred on $`18^h47^m42.6^s,+01^o33^{}04^{\prime \prime }`$ which includes the newly discovered planetary nebula PHR1847+0132, taken from a slightly shallow SHS survey field h1332 (exposure number HA18088, survey grade A2, but exposure time cut short to 168mins cf. 180 normally). The data have been carefully matched in terms of co-ordinate projection but not otherwise processed. It is clear the two surveys achieve similar depth for diffuse emission but that the IPHAS survey goes deeper for point-sources due to its better resolution. Further details of the IPHAS survey are given by Drew et al. (2005).
## 13 Colour-colour photometric plots
The SR, H$`\alpha `$ ‘R-equivalent’ and $`I`$ magnitudes can be combined to provide $`H\alpha R`$ and $`RI`$ colours for the objects detected by the IAM software. The narrowband H$`\alpha `$ photometry should be sensitive to point-source emitters, so the SHS stellar photometry is of particular interest. A colour-magnitude diagram (CMD) of SR magnitude versus $`H\alpha R`$ can be constructed from the survey data to trace the average values of $`H\alpha R`$ for normal stars by brightness and help identify emitters by highlighting objects which stand apart from this. Figure 23a shows a CMD constructed from a 1 region from the centre of survey field h135. Most stars can be seen to congregate around a stellar locus running vertically in the diagram at $`H\alpha R`$ $`0.2`$ and with an increasing spread towards fainter magnitudes. This is due to the census including a more complete population of objects at fainter magnitudes and increasing photon errors. Outliers can be seen either side of the distribution and, if present, emission line objects will sit to the left of the main stellar locus.
A colour-colour plot of $`RI`$ vs $`H\alpha R`$ provides more information about the stellar population along a given line of sight. Figure 23b presents such a plot for the same area of sky as described above and uses the $`I`$ band magnitudes which have been corrected to the SR photometry. In this plot the stellar locus is centred at $`H\alpha R`$ $`0.2`$ and $`RI`$ $`=0.15`$. The area covered suffers high reddening of up to $`E(BV)=4.7`$ mag (Schlegel, Finkbeiner, & Davis, 1998) and this is evident in the stretching of the stellar locus towards larger values of $`RI`$ and, significantly, $`H\alpha R`$ making them appear to be emission line stars. These reddened non-emitters can be identified in the colour-colour plot of $`RI`$ versus $`H\alpha R`$ and excluded from studies of potential point-source emitters (Pierce 2005). An additional complication is the potential contamination from late type stars. A TiO opacity minimum near 6536Å enables the continuum to be attained, producing the peak, compared to the TiO band heads on either side of the H$`\alpha `$ filter. Further to the red, the Tech-Pan emulsion sensitivity cuts off at 6990Å. Such objects can thus appear as apparent H$`\alpha `$ emitters when compared to the matching SR photometry unless the complementary I band photometry is included, as such late type stars will be brighter in this band than H$`\alpha `$ emitters. However, colour-colour plots created in 1-degree sub-regions, avoids the smearing effects on the photometry due to small positional shifts in the stellar locus across a H$`\alpha `$ survey field. These have proved very effective in identifying point-source emission candidates and has been successfully employed to provide targets for follow-up multi-object spectroscopy with 6dF and 2dF at the AAO (e.g. Hopewell et al. 2005).
## 14 Scientific Exploitation
The SHS on-line atlas was released in stages starting in 2002 with the complete survey made available in 2003. A variety of programmes are underway to exploit the scientific potential of this new resource. Several illustrative project examples are briefly mentioned below.
### 14.1 Planetary Nebulae
The largest project arising from the AAO/UKST H$`\alpha `$ survey has been the Macquarie/AAO/Strasbourg H$`\alpha `$ planetary nebula project (MASH; Parker et al. 2003, 2005 in preparation) which has uncovered about 1000 new Galactic planetary nebulae (PNe), nearly doubling the sample accrued from all sources over the last 100 years. Related projects concern identification of a significant new PNe population in the Galactic Bulge (e.g. Peyaud, Parker & Acker 2003) and the discovery of an important sample of Wolf-Rayet central stars of PNe (e.g. Morgan, Parker & Russeil 2001; Parker & Morgan 2003) including the detection of the only \[WN\] PN central star in the galaxy (Morgan, Parker & Cohen 2003). A possible new phase of PNe evolution has also been reported around a strongly masing OH-IR star (Cohen, Parker & Chapman 2005). A very large PN in an early stage of interaction with the ISM has also been discovered (Pierce et al. 2004) as well as two, very large bipolar PNe previously mis-identified as HII regions (Frew, Parker & Russeil 2005). A new sample of large ($`>`$4 arcminute), highly evolved, low surface brightness PNe have also been found from examination of 16 $`\times `$ blocked down FITS images of the entire 233 fields of SHS survey. These blocked-down images effectively enhance large angular size low surface brightness features (e.g. Pierce et al. 2004; Frew & Parker 2005, in preparation). All these discoveries are being investigated with follow-up spectroscopy to determine the fundamental parameters of this significant new sample.
### 14.2 HII regions and regions of star formation
This type of study is currently ripe for exploitation with little work currently undertaken. Mader et al. (1999) report the discovery of a significant new population of Herbig-Haro objects from an SHS extension field in Orion and a large wind-blown bubble with secondary star-forming regions around its periphery by Cohen et al. (2002). Numerous very faint HII regions have also been discovered as a by-product of the MASH survey (Parker et al. in preparation). Many large features are evident including bubbles which trigger star-formation and induce velocity departures of the associated HII regions. Using data from the SHS and additional kinematic information from Fabry-Perot observations (Georgelin et al. 2000), the filamentary H$`\alpha `$ counterparts and triggered HII regions for the HI shell centered at 290.1+0.2 (Rizzo and Arnal, 1998) have been revealed. The high resolution of the data permits precise description of the morphology and extent of such HII regions. This is vital information in determining the location of the exciting stars which can be inferred via orientation of the observed rims and dust “elephant trunk” with respect to the HII region as a whole. Furthermore, the H$`\alpha `$ counterpart and visible extension can be directly compared to radio HII regions. This is essential information needed to determine the distance of HII regions in the framework of the study of the large scale structure of our Galaxy (Russeil et al., 2005).
### 14.3 Supernova Remnants
Several programmes searching for the optical counterparts of supernova remnants (SNRs) in the SHS have already been undertaken. Walker, Zealey & Parker (2001) report finding new filamentary shell structures traced by H$`\alpha `$ emission that are likely associated with Galactic SNRs. A more recent project is underway to uncover SNR candidates across the entire SHS from careful scrutiny of both the blocked-down FITS images and the original survey films, with several new SNRs already confirmed (Stupar, private communication). Searches for new optical H$`\alpha `$ counterparts around the known Galactic SNR overlapping the SHS is also underway. One new Galactic SNR discovered serendipitously via the MASH programme has already been reported (Parker, Frew & Stupar 2004). A significant increase in the known population of optically detected Galactic SNRs is promised.
### 14.4 Point-source emitters
One area of more recent study is the search for point-source emitters and the subsequent follow-up spectroscopy of candidates identified from the SHS H$`\alpha `$ and SR photometry. Drew et al (2004) report the discovery, via SHS photometry, of only the 4<sup>th</sup> known massive WO star in the Milky Way Galaxy identified as part of a general programme of candidate point-source follow-up. Additionally, Hopewell et al. (2005) present five new WC9 stars discovered from the SHS data in a similar fashion. Pierce (2005) and Pierce et al. (2005, in preparation) demonstrate the power of the SHS to reveal significant new populations of H$`\alpha `$ emitters via a particular study in the Vela molecular ridge, especially when combined with $`I`$ band and 2MASS photometry. These preliminary projects have been finding, for the magnitude range explored most thoroughly ($`12<R<16.5`$), that 10-20 per cent of candidates are confirmed as emission line objects via follow-up spectroscopy. Their H$`\alpha `$ equivalent widths usually exceed 20Å. More recent work by Pierce (2005) indicates that the situation can be improved by weeding out M stars more thoroughly using the 2MASS data.
## 15 Conclusions
The AAO/UKST H$`\alpha `$ survey as scanned by SuperCOSMOS is now complete and on-line as the SHS atlas. It represents a powerful tool for the study of the ionized gas content of our galaxy on a range of spatial scales from arcsecond to tens of degrees. The distribution and structure of the ionised gas result from a wide range of astrophysically interesting phenomena. The astrometric and photometric properties have been described and shown to be well behaved and adequate for most purposes. Importantly, despite difficulties associated with photographic data and the scanning process, comparison with the independently calibrated SHASSA images has shown that the SHS survey faithfully records diffuse Galactic emission over a wide range of intensities from $``$ 5 Rayleighs to 500 Rayleighs. Emission down to $``$ 2 Rayleighs has been detected on one field, h1109. A calibration scheme for all 233 survey fields has been generated, based on comparison of a carefully selected, 30 arcminute region from each field with the equivalent area of intensity calibrated SHASSA H$`\alpha `$ image. If the limitations of the data are respected in terms of dynamic range, reliable flux estimates are possible. The survey is clearly appropriate for studies of individual H$`\alpha `$ emitting objects including point-sources as well as being suitable for the study of the ionized interstellar medium in general. A variety of projects exploiting this resource are already underway and many exciting discoveries have already been made. The community is invited to consider use of this valuable survey when undertaking any study of the Southern Galactic Plane.
## Acknowledgements
The authors gratefully acknowledge the support of the AAO board, the Wide-Field Astronomy Unit at the University of Edinburgh, the Wide-Field Astronomy Panel (UK), the Particle-Physics and Astronomy Research Council and the AAO directors Russell Cannon and Brian Boyle and UKST astronomers-in-charge Ann Savage and Fred Watson for making the SHS survey possible. This paper used comparison data from SHASSA which was produced with support from the National Science Foundation. MC thanks NASA for supporting his participation in the SHS through LTSA grant NAG5-7936 with UC Berkeley. MJP thanks PPARC for provision of a PhD studentship. We also thank the referee John Meaburn for valuable comments on this paper.
## References
Arrowsmith P., & Parker Q.A., 2001, ROE internal report
Beard S.M., MacGillivray H.T., Thanisch P.F., 1990, MNRAS, 247, 311
Bland-Hawthorn J.B., Veilleux S., Cecil G.N., Putman M.E., Gibson B.K., Maloney P.R., 1998, MNRAS, 299, 611
Bok B.J., Bester, M.J., Wade, C.M., 1955, Daedalus, 86, 9
Bond I.A. et al., 2001, MNRAS, 327, 868
Boyle B.J., Shanks T., Croom S.M., 1995, MNRAS, 276, 33
Buxton M. et al., 1998, PASA, 15, 24
Cohen M., Green A., Parker Q.A., Mader S., Cannon R.D. 2002, MNRAS, 336, 736
Cohen M., Parker, Q.A., Chapman, J., 2005, MNRAS, 357, 1189
Croom S.M., Ratcliffe A., Parker, Q.A., Shanks T., Boyle, B.J., Smith R.J., 1999, MNRAS, 306, 592
Davies R.D., Elliott K.H., Meaburn J., 1976, Mem. RAS, 81, 89
Dennison B., et al., 1998, PASA, 15, 147
Dopita M.A., Hua C.T., 1997, ApJS, 108, 515
Dopita M.A., Mathewson D.S., Ford V.L., 1985, ApJ, 297, 599
Drew J., Barlow M.J., Unruh Y.C., Parker Q.A., Wesson R., Pierce M.J., Masheder M.R.W., Phillipps S., 2004, MNRAS, 351, 206
Drew J. et al., 2005, MNRAS, submitted
Elliot K.H., Meaburn J., 1976, Ap&SS, 39, 437
Elmegreen B., Lada C.J., 1977, ApJ, 214, 725
Finkbeiner D.P., 2003, ApJS, 146, 407
Frew D.J., Parker Q.A., Russeil D., 2005, MNRAS, submitted
Gaustad J.E., McCullough P.R., Rosing W., & Van Buren D., 2001, PASP, 113, 1326
Georgelin Y.P., Georgelin Y.M., 1970, A&AS, 3, 1
Georgelin Y.M., Russeil D., Amram P. et al., 2000, A&A, 357,c308
Gerola H., Seiden P., 1978, ApJ, 223, 129
Green A.J., Cram L.E., Large M.I., Ye T., 1999, ApJS, 122, 207
Gum C.S., 1952, Observatory, 72, 151
Gum C.S., 1955, Mem.RAS, 67, 155
Haffner L.M., Reynolds R.J., Tufte S.L., Madsen G.J., Jaehnig K.P., Percival J.W., 2003, ApJS, 149, 405
Hambly N.C., Miller L., MacGillivray H.T., Herd J.T., Cormack W.A., 1998, MNRAS, 298, 897
Hambly N.C., et al., 2001a, MNRAS, 326, 1279
Hambly N.C., Irwin M.J., MacGillivray H.T., 2001b, MNRAS, 326, 1295
Hambly N.C., Davenhall A.C., Irwin M.J., MacGillivray H.T., 2001c, MNRAS, 326, 1315
Hase V.F., Shajn G.A., 1955, Isv. Krym. Astrofiz. Obs., 15, 11
Hog E., Fabricius C., Makarov V.V., Urban S., Corbin T., Wycoff G., Bastian U., Schwekendiek P., Wicenec A., 2000, A&A, 355, L27
Hopewell E. C., Barlow M. J., Drew J. E., Unruh Y. C., Pierce M. J., Parker Q. A., Knigge C., Phillipps S., Zijlstra A. A., 2005, MNRAS, submitted.
Jarrett T, Chester T., Cutri R., Schneider S., Skrutskie M., Huchra J.P., 2000, AJ, 119, 2498
Johnson H.M., 1955, ApJ, 121, 604
Johnson H.M., 1956, ApJ, 124, 90
Keller S.C., Grebel E.K., Miller G.J., Yoss K.M., 2001, AJ, 122, 248
Kennicutt R.C., 1992, ApJ, 388, 310
Kodak publication P-315, 1987, Scientific imaging with Kodak films and plates
Lasker B.M., et al., 1988, ApJS,68,1
Mader S.L., Zealey W.J., Parker Q.A., Masheder M.R.W., 1999, MNRAS, 310,331
Meaburn J., 1978, Aplied Optics, 17, 1271
Meaburn J., 1980, MNRAS, 192, 365
Meaburn J., White N.J., 1982, MNRAS, 200, 771
Mel’Nik A.M., Efremov Y.N., 1995, AstL, 21, 10
Miller L., Cormack W.A.., Paterson M.G., Beard S.M., Lawrence L., 1992, In MacGillivray H.T., Thomson E.B., eds, Digitised Optical Sky Surveys, Kluwer, Dordrecht, p.133
Morgan D.H., Parker Q.A., & Russeil D., 2001, MNRAS, 322, 877.
Morgan D.H., Parker Q.A., Cohen M., 2003, MNRAS, 346, 729
Morgan D.H., Parker Q.A., 2005, MNRAS, in press
Nossal S., et al., 2001, J.Geophys.Res. 5605
Osterbrock D.E., 1989, Astrophysics of Gaseous Nebulae, University Science Books
Parker Q.A., Bland-Hawthorn J., 1998, PASA, 15, 33
Parker Q.A., Malin D.F., 1999, PASA, 16, 288
Parker Q.A., Phillipps S., 1997, PASA, 15, 28
Parker Q.A., Phillipps S., 1998, A&G, 39, 4.10
Parker Q.A., Phillipps S., 2003, in ASP Conf.Ser. 289: Proceedings of the IAU 8th Asian-Pacific Regional Meeting, Volume I, 165
Parker Q.A., Hartley M., Russeil D., Acker A., Ochsenbein F., Morgan D.H., Beaulieu S., Morris R., Phillipps S., Cohen,M., 2003, ASP Conf.Ser. eds M.Dopita, S.Kwok and R Sutherland, p.41
Parker Q.A., Morgan D.H., 2003, MNRAS, 341, 961
Parker Q.A., Frew D.J., Stupar M., 2004, AAO Newsletter, 104, 9
Peyaud A.E.J., Parker Q.A., Acker A., 2003, in SF2A-2003: Semaine de l’Astrophysique Francaise, 311
Phillipps S., Parker Q.A., 1993, MNRAS, 265, 385
Pickering W.H., 1890, Sidereal Messenger, 9, 2
Pierce M.J., Frew D.J., Parker Q.A., Koppen J., 2004, PASA, 21, 334
Pierce, M, 2005, PhD thesis, University of Bristol
Price S.D., Egan M.P., Carey S.J., Mizuno D., Kuchar T. 2001, AJ, 121, 2819
Rizzo J.R., Arnal E.M., 1998, A&A 332, 1025
Rodgers A.W., Campbell C.T., Whiteoak J.B., 1960, MNRAS, 121, 103
Russeil D., et al., 1997, A&A, 319, 788
Russeil D., et al., 1998, PASA, 15, 9
Russeil D., Adami C., Amram P., Coarer E., Georgelin T.M., Marcelin M., Parker Q.A., 2005, A&A, 429, 497
Schlegel D.J., Finkbeiner D.P., Davis M., 1998, ApJ, 500, 525
Sharpless S., 1953, ApJ, 118, 362
Sharpless S., 1959, ApJS, 4, 257
Sivan, 1974, A&A,16, 163
Stephenson C.B., & Sanduleak N., 1977, ApJS, 33, 459
Storkey A.J., Hambly N.C., Williams C.K.I., Mann R.G., 2004, MNRAS, 347, 36
Sung H., Chun M., Bessell M.S., 2000, AJ, 120, 333
Tenorio-Tagle G., Palous J., 1987, A&A, 186, 287
Tritton S.B., 1993, UKSTU Handboook, publication of the Royal Observatory Edinburgh
Walker A., Zealey W.J., Parker Q.A., 2001, PASA, 18, 259
Watson F.G. 1984, MNRAS, 206, 661
Zacharias N., Urban S.E., Zacharias M.I., Wycoff G.L., Hall D.M., Monet D.G., Raffert T.J., 2004, AJ, 127,3043
|
warning/0506/astro-ph0506080.html
|
ar5iv
|
text
|
# RADIO QUIET AGN
## 1 Introduction
Active Galactic Nuclei (hereafter AGN) are the most powerful persistent and compact sources of radiation in the Universe. As such, they are clearly of interest from the point of view of high energy processes. AGN activity is caused by accretion of material onto the massive black hole residing at the center of a host galaxy. Therefore, in those objects we can observe the behavior of the matter in a strong gravity field.
Although the basic engine is always the same — extraction of gravitational energy of the infalling material — AGN are quite an inhomogeneous class of objects. The most important differentiating property is perhaps the radio loudness. A small fraction of those sources are strong radio emitters and display spectacular jets and/or radio extended radio lobes. The majority, however, forms a population of radio quiet sources. The transition between the two populations is rather smooth, but nevertheless it is convenient to introduce two separate classes of objects, with a focus on strong relativistic jets as the main property of radio-loud sources. The customary border is set at the 5 GHz radio to B band flux ratio $`\mathrm{log}F_{5GHz}/\mathrm{log}F_B=10`$. The properties of the radio-loud objects are discussed by R. Moderski and A. Celotti (these proceedings). Here I will concentrate on the radio-quiet population and their relevance to high energy astrophysics.
## 2 Basic facts
AGN span a very broad range of luminosities. The brightest AGN (found among quasars) have bolometric luminosities almost up to $`10^{48}`$ erg s<sup>-1</sup> $`^\mathrm{?}`$. They are mostly found among the high redshift objects. Nowadays, numerous quasars are found up to $`z6`$ $`^\mathrm{?}`$ as a result of massive surveys (e.g. SDSS $`^\mathrm{?}`$). Nearby Seyfert galaxies are a few orders of magnitude fainter. The lower limit for a nuclear activity is unspecified since the determination of the weak nuclear activity is observationally difficult. However, it is now widely believed that all regular galaxies with bulges contain a supermassive black hole and must show some level of activity. In this sense we can also count SGR A\* as an example of weak activity (see S. Nayakshin, these proceedings), with its occasional flares reaching up to the level of $`10^{35}`$ erg s<sup>-1</sup>.
Black hole mass estimates indicate much narrower range, $`10^610^{10}M_{}`$ $`^\mathrm{?}`$, than the luminosity range which means that the Eddington ratio differs considerably between the sources.
### 2.1 Classification
Radio-quiet AGN are divided into several classes, with two parameters most plausibly underlying this classification scheme: inclination angle and the Eddington ratio.
Generally, all AGN are divided into type 1 and type 2 objects. Type 1 objects have very broad emission lines while type 2 objects show only narrow lines in their optical/UV spectra. It is generally accepted that class 2 is simply an obscured version of type 1 objects so we have no direct view of the nucleus in those sources. This obscuration is due to the material predominantly located in the equatorial plane, in a form of a dusty/molecular torus. It can be noted, however, that there may be intrinsic type 2 sources among low Eddington ratio AGN $`^\mathrm{?}`$. It only means that classification based on the width of the hydrogen lines may not always represent well the characteristics of the central engine. In further text we will discuss only type 1 AGN.
Another classification, mostly historical, divides type 1 radio quiet AGN into quasars, Seyfert galaxies, Narrow Line Seyfert 1 galaxies and LINERS. The transition between these classes (connected mostly with the luminosity and the Eddington ratio) is smooth. Moreover, some quasars are also classified as NLS1 galaxies, and some LINERS perhaps are mostly starburst so the issue is confusing but we will preserve these subgroups to indicate the luminosity class of the discussed sources. Also for bright quasars the transition to the corresponding ’Narrow Line Type 1’ class happens at much higher width ($`4000`$ km s<sup>-1</sup> $`^\mathrm{?}`$) than for Seyfert galaxies ($`2000`$ km s<sup>-1</sup>). This can be easily understood if the transition happens at a fixed Eddington ratio and the difference in black hole mass is taken into account (see formulae in $`^\mathrm{?}`$).
### 2.2 Broad band spectra
The radiation spectrum of radio quiet AGN is very broad and span from radio to gamma band. The spectrum is best studied for high luminosity sources. A schematic view, representative for sources with high Eddington ratio (quasars and Narrow Line Seyfert 1 galaxies; $`L/L_{Edd}1`$) is shown in Fig. 1 $`^\mathrm{?}`$. The spectrum usually peaks in the far UV/soft X-ray range, not accessible to the observations due to the Galactic extinction. However, the observational gap can be partially filled by combining low redshift and high redshift sources into a single composite spectrum $`^\mathrm{?}`$. The extension of the spectra beyond 100 keV is not well studied and we return to this point in Sect. 3.1.
The spectra of AGN with lower Eddington ratio like Seyfert 1 galaxies ($`L/L_{Edd}0.01`$) show less pronounced far UV peak and relatively stronger X-ray emission, and the X-ray spectrum is harder. The black holes accreting at the lowest rate ($`L/L_{Edd}10^4`$) like those in LINERS or LLAGN do not show any strong UV emission while they are still relatively bright in X-rays. The broad band spectra of such objects are difficult to determine since the optical spectrum is in this case dominated by the host galaxy. Their X-ray spectra are as hard as in Seyfert 1 galaxies.
Practically all radio-quiet AGN are rather variable $`^{\mathrm{?},\mathrm{?}}`$. The shortest variability timescales (hundreds of seconds) are seen in X-ray band, in Seyfert galaxies. Optical/UV emission varies in timescales of days in those sources while in quasars monitoring over years is needed to see significant changes in optical band $`^\mathrm{?}`$.
## 3 Radio-quiet AGN as sources of high energy photons and particles
Radio-quiet AGN are not as obviously attractive from the point of view of the high energy physics as are radio-loud AGN. However, they also represent certain challenge for such studies. In order to show that we will discuss now the gamma-ray emission from these sources and the most plausible environment of its production.
### 3.1 Extension of the AGN spectra into high energies
The direct signature of the importance of high energy processes in a given object is the presence of the gamma-ray emission. Unfortunately, spectral measurements going beyond 100 keV were performed only for a few sources, and the results were uncertain. Fits to the OSSE composite spectrum $`^\mathrm{?}`$ gave the high energy cut-off $`120_{60}^{+220}`$ keV for Seyfert 1 and $`130_{50}^{+220}`$ keV for Seyfert 2 galaxies (at such high energies the obscuration by the torus is relatively unimportant). Beppo-SAX data, however, did not give such a strong constraints (cut-off energy $`231_{168}^+\mathrm{}`$ keV for Seyfert 1 galaxies, no cut-off for Seyfert 2 galaxies $`^\mathrm{?}`$). Comparison of the model to the observed X-ray background suggest that if the typical photon index is $`1.9`$ the spectrum extends up to $`300`$ keV $`^\mathrm{?}`$ while observations of high redshift quasars give results from 100 keV to 500 keV and more, depending on the object and the adopted geometry $`^{\mathrm{?},\mathrm{?}}`$.
Overall similarity of AGN and galactic black holes may suggest that sources with rather hard spectra (photon index 1.9 or less) are thermal (electron temperature of order of 100 keV). Such a spectrum is well explained as a result of the Comptonization of the soft photons by a predominantly thermal plasma with the electron temperature around $`10^9`$ K. Ions may have much higher temperature, or may not be fully thermalized, but we have no direct methods of estimating their properties. Efficient thermalization is supposed to be achieved through synchrotron self-absorption $`^\mathrm{?}`$. Sources with steep spectra (photon index 2.0 and more) are likely to be significantly non-thermal, with a population of electrons having a power law distribution of energies, as expected in case of effective acceleration and ineffective thermalization.
Direct observation constraints should come from Astro-E2 and GLAST.
### 3.2 accretion flow geometry
The broad band spectrum clearly shows that the accretion flow is (at least) a two-phase medium. The profound optical/UV/soft X-ray bump (Big Blue Bump) so characteristic for high $`L/L_{Edd}`$ sources is well modeled as a thermal emission of a Keplerian, optically thick and geometrically thin disk (see Fig. 2). The hard X-ray emission must come from a hot optically thin plasma in the vicinity of the disk. The IR emission is due to reprocessing of a part of the radiation by circumnuclear dust, usually referred to as a dusty-molecular torus. The presence of this torus is responsible for shielding the central parts for highly inclined observers, as is the case for type 2 objects. The torus is most probably clumpy, and quite possibly it is rather a kind of outflowing dusty wind instead of a structure in the hydrostatic equilibrium.
The location and geometry of the hot plasma is uncertain. A plausible possibility is shown in Fig. 2. The disk is covered by magnetic loops emerging from its interior $`^\mathrm{?}`$. The mechanism behind this is the magneto-rotational instability (MRI) operating in the disk interior. This instability is responsible for the disk viscosity, and the corresponding Shakura-Sunyaev parameter $`\alpha `$ is $`0.01`$, according to numerical simulations. Occasionally, large loops emerge high above the disk surface $`^\mathrm{?}`$, and the magnetic field reconnection results in formation of a flare (hard X-ray flash). Therefore, a corona similar to the solar corona forms above the disk. In the innermost part the cold disk can be evaporated due to the interaction with such an active corona $`^\mathrm{?}`$. A flow towards the black hole proceeds through a hot plasma phase $`^{\mathrm{?},\mathrm{?},\mathrm{?}}`$, and a (perhaps significant) fraction of the plasma can be expelled out in the process. In strongly radio-loud objects this outflow takes a form of a relativistic well-collimated jet but in radio-quiet objects the outflow is slower and uncollimated. The mechanism of the outflow is unknown.
Within this picture, we would expect the formation of non-thermal plasma in the coronal loops above the disk while in the hot phase the electrons would be predominantly thermal. The relative role of the two media is determined by the disk trucation radius which in turn is determined by $`L/L_{Edd}`$. At $`L/L_{Edd}0.10.5`$ or more the disk is supposed to extend down to the marginally stable orbit.
The geometry shown in Fig. 2 is not a unique possibility. Other accretion flow models include (i) the lamp-post model in which the disk always extends up to the marginally stable orbit and the high energy emission comes from a shock formed in the outflowing plasma and localized on the symmetry axis $`^{\mathrm{?},\mathrm{?}}`$ (ii) disk always extending down to the marginally stable orbit with magnetic flares above it $`^\mathrm{?}`$ (iii) accretion in the form of clumps of cold material embedded in a hot plasma $`^\mathrm{?}`$. In those models the spectral differences between high Eddington and low Eddington sources are less naturally explained.
## 4 X-ray spectroscopy as a tool to study the plasma motion
Given the lack of sufficient spatial resolution the key to the flow geometry lies in X-ray spectroscopy. First extremely useful results came from Ginga $`^\mathrm{?}`$ and ASCA $`^\mathrm{?}`$ data, now Chandra and XMM-Newton offer still higher quality data. A number of blueshifted absorption lines was identified which allowed to measure the outflow velocity of the material, and measured a number of emission line profiles, including the famous iron K$`\alpha `$ line.
### 4.1 outflow
Chandra and XMM-Newton observations confirmed the earlier findings that a partially ionized warm absorber exists in many Seyfert 1 and NLS1 galaxies $`^\mathrm{?}`$. Outflow velocities range from hundreds km s<sup>-1</sup> in Seyfert 1 $`^{\mathrm{?},\mathrm{?}}`$ galaxies to a fraction of light speed in NLS1 $`^{\mathrm{?},\mathrm{?}}`$. The distance of this outflowing material from the black hole is difficult to assign. Simple arguments based on the assumption of a roughly Keplerian velocity suggest that the slow outflow originates somewhere between the Broad Line Region and a Narrow Line Region, i.e. at a few parsecs from the nucleus, and the fast outflow originates closer in, at distances of order of hundreds of Schwarzschild radii. The amount of outflowing material may be quite large - fast outflows may actually carry out considerable fraction of the inflowing material $`^\mathrm{?}`$. The estimates are difficult since a fraction of the material may be completely ionized and leave no direct signature in the soft X-ray spectrum. However, this material will scatter a fraction of the nuclear emission and may lead to modification of the optical spectrum of an AGN through the irradiation of the outer disk $`^{\mathrm{?},\mathrm{?}}`$.
### 4.2 quasi-Keplerian motion
Chandra and XMM-Newton confirmed that in some objects the very broad emission lines (iron line and soft X-ray lines) are seen, with the shape well represented as the effect of the relativistic smearing due to the Keplerian motion of the emitting material located very close to the black hole $`^\mathrm{?}`$. Some observations indicate that the inner radius of the disk increases when the source becomes fainter $`^\mathrm{?}`$.
However, the observed variability of the iron line is not well understood and remains a major puzzle $`^{\mathrm{?},\mathrm{?}}`$. Models require some fine tuning in order to reproduce the observed trends.
Models which explore full observational information have even more difficulties. We attempted to reproduce both the fractional variability amplitude and point-to-point fractional variability amplitude for MCG-6-15-30. The model assumed that magnetic flares are randomly distributed above the disk surface. The flare flux, flare duration and the probability of a flare at a given radius were assumed to have a power law dependence on the radius. Black hole was assumed to rotate, disk was assumed to extend down to the marginally stable orbit appropriate for an adopted Kerr parameter. Each flare created a hot spot on the disk surface by irradiation and the reflection (including iron line and other soft lines) was calculated using the code titan/noar $`^\mathrm{?}`$ and taking into account that the incident flux depends on the distance from the flare center. Flare was assumed to be short-lasting so the vertical structure of the disk was taken from an unilluminated model and the disk expansion was neglected. Predicted spectrum was integrated in specific time bins, as in the data, all relativistic effects were included using the code ky $`^\mathrm{?}`$.
Within this scenario, we could not find a parameter range which would reproduce the shape of both fractional variability functions. Either still the model is too simple (we did not consider the dependence of the shape of the reprocessed radiation on the radius) or an important element is still missing. There is some evidence that variations in the outflowing warm absorber can contribute significantly. It was even suggested that warm absorber is mostly responsible for the observed shape of the spectrum in soft X-ray band $`^\mathrm{?}`$, and the broad iron line is partially an artefact $`^\mathrm{?}`$.
## 5 Shocks and magnetic field reconnections as a heating mechanism of the hot plasma
Although the geometry of the hot material is still under discussion, some mechanisms of plasma heating are needed to explain the observed X-ray emission.
Models that can apply to magnetic flares above the disk surface were discussed in numerous papers $`^{\mathrm{?},\mathrm{?}}`$.
In ADAF type solutions most of the gravitational energy is used to heat ions, and subsequently directed towards electrons through Coulomb interaction $`^{\mathrm{?},\mathrm{?}}`$. However, a fraction of energy will be inevitably used to heat electrons directly $`^\mathrm{?}`$ thus limiting the low radiative efficiency of the flow.
Several other aspects were also discussed, like disk-corona coupling $`^\mathrm{?}`$, disk evaporation and the ion irradiation of the disk $`^\mathrm{?}`$ but the picture is far from being complete.
Observationally, very interesting results were obtained in the context of a galactic source (microquasar) GRS 1915+105 $`^\mathrm{?}`$ and they may apply to AGN when timescales are corrected by the black hole mass ratio. Methods, like Fourier-resolved spectroscopy, successfully applied to galactic sources $`^\mathrm{?}`$, may be also helpful although the AGN data at present are hardly of the appropriate quality.
Therefore, observational determination of the gamma-ray emission from radio-quiet objects
## Acknowledgments
We thank Aneta Siemiginowska and Piotr Życki for very helpful discussions. Part of this work was supported by the grant PBZ-KBN-054/P03/2001 of the Polish State Committee for Scientific Research, and by the Laboratoire Europeén Associé Astrophysique Pologne-France.
## References
|
warning/0506/nucl-th0506063.html
|
ar5iv
|
text
|
# nucl-th/0506063 LA-UR-05-4415 New broad 8Be nuclear resonances
## 1 Introduction
What are the properties of the resonances of <sup>8</sup>Be? This question is most comprehensively answered by a global analysis of all experimental data based on the best reaction theory available, for example R-matrix theory. Resonance structure tends to be based on single experiments, most recently compiled by TUNL . In contrast, the results of a coupled channel R-matrix analysis of data from 69 experimental references are given here. This analysis does not include all experimental data, and hence is not expected to provide the best parameters for all resonances. This is particularly true of narrow resonances (with widths less than 100 keV), which can be approximated by the Breit-Wigner formula. The strength of a coupled channel R-matrix analysis becomes apparent for broad resonances, whose structure can only be determined by analyzing data over a large energy range in various channels, and for which the full force of reaction theory is needed.
The physical content of scattering can be summarized by knowledge of the S-matrix for real energies. However, a more intuitive picture is provided by resonances, which are defined as complex energy poles of the S-matrix. The real part of the pole $`\lambda `$ is defined as the excitation energy $`E_x`$, and two times the imaginary part as the width $`\mathrm{\Gamma }`$. Because these parameters can only be found for complex energies, which cannot be experimentally accessed, resonances involve a mathematical extrapolation beyond observation. Since resonances with small widths tend to have the most pronounced experimental effects, this analysis is limited to resonances fairly near to the real energy axis (the “unphysical sheet closest to the physical sheet” ). Even so, controversy centers around very broad resonances which are not observable as clear bumps in experimental cross-sections, particularly a total angular momentum, parity, isospin and excitation energy $`J^\pi T(E_x)=2^+0(16)`$ resonance found in this analysis. This resonance was previously found in an R-matrix analysis by Barker et al. . They also found a broad $`0^+`$ at about $`10`$ MeV. This analysis also discovers a previously unreported broad $`4^+0(18)`$ resonance.
## 2 Analysis technique
The analysis is performed with the EDA R-matrix code . Integrated cross-section, differential cross-section and polarization data, consisting of more than 4700 points, are fitted with a $`\chi ^2/(d.o.f.)`$ of $`7.4`$ utilizing about $`100`$ free parameters (the R-matrix level eigenenergies and reduced width amplitudes discussed in the next section). This high $`\chi ^2`$ is mostly related to contradictory data, as well as underestimates of experimental relative and normalization errors . Since the resonance structure is insensitive to exclusion of data that fit with more than three standard deviations , it is robust under inclusion of the worst fitting data points. Experimental nuclear data on the reactions $`{}_{}{}^{4}He(\alpha ,\alpha _0)`$, $`{}_{}{}^{4}He(\alpha ,p_0)`$, $`{}_{}{}^{4}He(\alpha ,d_0)`$, $`{}_{}{}^{7}Li(p,\alpha _0)`$, $`{}_{}{}^{7}Li(p,p_0)`$, $`{}_{}{}^{7}Li(p,n_0)`$, $`{}_{}{}^{7}Be(n,p_0)`$, $`{}_{}{}^{6}Li(d,\alpha _0)`$, $`{}_{}{}^{6}Li(d,p_0)`$, $`{}_{}{}^{6}Li(d,n_0)`$ and $`{}_{}{}^{6}Li(d,d_0)`$, leading to the <sup>8</sup>Be intermediate state, are included. All recoil nuclei are in the ground state. Table 1 contains a complete list of the data in the analysis. Substantial data are entered for the $`{}_{}{}^{4}He(\alpha ,\alpha _0)`$ and $`{}_{}{}^{7}Li(p,p_0)`$ reactions, and the least data are entered for the $`{}_{}{}^{4}He(\alpha ,p_0)`$, $`{}_{}{}^{4}He(\alpha ,d_0)`$ and $`{}_{}{}^{6}Li(d,d_0)`$ reactions . The maximum excitation energy above the <sup>8</sup>Be ground state is $`2526`$ MeV for all reactions except $`{}_{}{}^{4}He(\alpha ,\alpha _0)`$ and $`{}_{}{}^{7}Be(n,p_0)`$. In the $`{}_{}{}^{4}He(\alpha ,\alpha _0)`$ reaction, data above the maximum $`\alpha `$ laboratory energy for which data are entered ($`38.4`$ MeV) and below the limit of this analysis, are only available as phase shifts , and have not been incorporated. For the $`{}_{}{}^{7}Be(n,p_0)`$ reaction no data above the near-threshold data entered are found below the maximum excitation energy of this analysis. Further details of the data and cross-section fits are available .
The excitation energies of the thresholds of the various analyzed channels, with respect to the unstable <sup>8</sup>Be ground state, are $`0.09`$ ($`\alpha ^4`$He), $`17.26`$ ($`p^7`$Li), $`18.90`$ ($`n^7`$Be) and $`22.28`$ MeV ($`d^6`$Li) . The two-body channels $`p^7`$Li, $`n^7`$Be and $`d^6`$Li, involving resonances less than $`100`$ keV wide, are neglected. These could reasonably be included in an R-matrix analysis. All the channels included are strongly constrained by unitarity (via the R-matrix formalism) and, as explained in the next section, isospin symmetry (charge independence). The channel radii are fixed as follows based on earlier R-matrix analyses: $`\alpha ^4`$He (4.0 fm), $`p^7`$Li and $`n^7`$Be (3.0 fm) and $`d^6`$Li (6.5 fm). The fit is insensitive to variation in the $`d^6`$Li radius . The orbital angular momenta included between the two scattered nuclei are: $`\alpha ^4`$He (S- , D- , G- , I- and L-waves), $`p^7`$Li and $`n^7`$Be (S- , P- , D- and F-waves) and $`d^6`$Li (S- , P- and D-waves). The inclusion of the highest wave for each channel did not seem to change the qualitative features of the fit, indicating that a sufficient number of waves has been used.
## 3 Procedure
The Kapur-Peierls expression for the S-matrix at real energies $`E`$ for channels $`c^{}`$ and $`c`$ is (Eq. 28 of Ref. )
$$S_{c^{}c}=\frac{I_c(a_c,k_c)}{O_c(a_c,k_c)}\delta _{c^{}c}+i\underset{\mu }{}\frac{\rho _{\mu c^{}}\rho _{\mu c}}{_\mu (E)E}\text{where }\rho _{\mu c}=\frac{\sqrt{2k_ca_c}𝒢_{\mu c}(E)}{O_c(a_c,k_c)}.$$
(1)
Here the incoming and outgoing wave functions $`I`$ and $`O`$ are functions of $`E`$ through the wave number $`k`$. In principle the S-matrix is independent of the channel radii $`a`$. The complex functions $`_\mu (E)`$ and $`𝒢_{\mu c}(E)`$ are determined by the R-matrix fit (see below, and also Ref. ). Eq. 1 can be extended to complex $`E`$, and the S-matrix remains independent of $`a`$. The poles of the S-matrix then occurs at complex $`E_0=_\mu (E_0)`$, where $`E_xRe[E_0]`$ is the resonance excitation energy and $`\mathrm{\Gamma }2Im[E_0]`$ is the resonance total width. The partial width $`\mathrm{\Gamma }_c|\rho _{\mu c}|^2=2|k_{0c}|a_c|𝒢_{\mu c}(E_0)/O_c(a_c,k_{0c})|^2`$ is evaluated at the pole in terms of the reduced width amplitude $`g_c|𝒢_{\mu c}(E_0)|`$, and is related to the residue at the pole (see Eq. 1). The quantities $`E_x`$, $`\mathrm{\Gamma }`$ and $`\mathrm{\Gamma }_c`$ are independent of $`a`$. Contrary to physical intuition, the sum of $`\mathrm{\Gamma }_c`$ for kinematically open channels is not equal to $`\mathrm{\Gamma }`$. It should be cautioned that $`E_x,\mathrm{\Gamma }`$ and $`\mathrm{\Gamma }_c`$ all depend on how the extension to complex $`E`$ is done, and are accordingly quantities that cannot be measured experimentally. However, for narrow resonances where $`_\mu (E)`$ is almost real, $`E_x,\mathrm{\Gamma }`$ and $`\mathrm{\Gamma }_c`$ respectively collapse to the usual notions of excitation energy, width and partial width, which can be measured experimentally.
The method of calculation of the S-matrix poles and residues in terms of the R-matrix parameters is briefly summarized from the more complete discussion . To obtain the S-matrix pole positions from the real R-matrix eigenenergies $`E_\lambda `$ and the real reduced width amplitudes $`\gamma _{\lambda c}`$ for the real boundary conditions $`B_c`$ (fixed in this analysis), as defined in Ref. , a complex energy $`E_0`$ is found such that at least one eigenvalue of the complex “energy-level” matrix (p. 294 of Ref. ),
$$_{\lambda ^{}\lambda }E_\lambda \delta _{\lambda ^{}\lambda }\underset{c}{}\gamma _{\lambda ^{}c}[L_c(a_c,k_c)B_c]\gamma _{\lambda c}$$
(2)
is the same as $`E_0`$. Here the outgoing-wave logarithmic derivatives $`L`$ are defined in terms of the outgoing wave functions $`O`$ in the usual way (Eq. 4.4, p. 271 of Ref. ), and are functions of $`E`$ through the wave number $`k`$. The residue at the pole $`i\rho _{\mu c^{}}\rho _{\mu c}`$ has already been written in terms of the function $`𝒢_{\mu c}(E_0)`$ in Eq. 1. This function can be calculated from the R-matrix parameters by using Eq. 4 of Ref. . Although this function and the energy-level matrix (Eq. 2) are defined for real energies, extension to complex $`E`$ is done by simply using the functional form of these expressions when working with complex energies. In this way both the S-matrix pole $`E_0`$ and the function $`𝒢_{\mu c}(E_0)`$, needed to calculate the excitation energy, (partial) width and reduced width amplitude, are defined in terms of the R-matrix parameters.
The EDA code used to perform the R-matrix analysis implements the standard Wigner R-matrix theory without approximations, except for restricting the number of R-matrix levels for a given $`J^\pi T`$ to a finite number of levels in the energy region of interest. The analysis employs isospin symmetry in the limited sense that isospin constraints on the $`\gamma _{\lambda c}`$ are implemented as follows. The $`\alpha ^4`$He and $`d^6`$Li channels couple to an isospin 0 level, but not to an isospin 1 level. Hence the $`\gamma `$’s for an isospin 1 level coupling to these channels are set to zero. Also, a level’s $`\gamma `$’s for the $`p^7`$Li and $`n^7`$Be channels are related by isospin Clebsch-Gordon coefficients, which are different for isospin 0 and 1 levels.
Let us consider the dissociation of the compound nucleus $`A`$ into nucleus $`A^{}`$ and ejectile $`a`$. Define the channel cluster form factor $`F`$, proportional to the overlap between the internal wave function of nucleus $`A`$ and the internal wave functions of the nuclei $`A^{}`$ and $`a`$, as
$$F(r_{aA^{}})[\psi _A^{}(\xi _A^{})\psi _a(\xi _a)]^{}\psi _A(\xi _A)𝑑\xi _A^{}𝑑\xi _a.$$
(3)
Here $`r_{aA^{}}`$ is the relative coordinate between the C.M. of $`a`$ and $`A^{}`$. The symbols $`\xi _A,\xi _A^{}`$ and $`\xi _a`$ denote internal coordinates of the nuclei $`A`$, $`A^{}`$ and $`a`$, respectively; and $`\psi `$ are the corresponding internal wave functions. A full definition of $`F`$ can be found elsewhere (Eq. 7 of Ref. ). The integral of $`|F|^2`$ over $`r_{aA^{}}`$ is the widely predicted “spectroscopic factor”. The R-matrix reduced width amplitude $`\gamma _{\lambda c}`$ for the breakup of a level $`\lambda `$ of the nucleus $`A`$ into $`A^{}`$ and $`a`$ in channel $`c`$ is defined as
$$\gamma _{\lambda c}=\sqrt{\frac{\mathrm{}^2a_c}{2M_c}}F(a_c),$$
(4)
where $`M_c`$ is the reduced mass for relative motion between $`A^{}`$ and $`a`$. Comparison between theory calculations and the predictions here are possible by comparing $`F(a_c)`$ calculated from theory and $`\gamma _{\lambda c}`$ using Eq. 4. However, this is only possible when the same boundary conditions $`B_c`$ are imposed at $`a_c`$ is as standardly done in R-matrix theory. As theory calculations do not usually do this, it is more useful to compare them to $`𝒢_{\mu c}(E)`$ in Eq. 1, which is the equivalent of $`\gamma _{\lambda c}`$ for wave functions with outgoing wave (Kapur-Peierls) boundary conditions (Eq. 30 of Ref. ). Hence the R.H.S. of Eq. 4, calculated from theory (usually) for bound states, should be compared to the $`g_c`$ which will be tabulated in the next section for scattering states.
## 4 Resonance structure
The $`E_x`$, $`\mathrm{\Gamma }`$ and isospin impurity of the resonances are displayed in Table 2. All $`J^\pi `$ are allowed, so that the $`J^\pi `$ is independently established by the R-matrix analysis. Isospin 0 and 1 are allowed for all resonances, because these are the only isospins that can couple to the channels in this analysis if isospin symmetry is assumed. The resonances found in Table 2 should be compared to the “experimental” resonances believed to exist on the basis of a summary of resonances found in experimental data and other analyses . A comparison with experiment indicates substantial agreement. Disagreements partially stem from the difference between defining the energy and width from poles of the S-matrix, as is done in the R-matrix analysis, and defining them from Breit-Wigner formulae, as is often the case in experimental analyses. For example, agreement between the energy and width of the well-known narrowest resonances ($`J^\pi T(E_x)=0^+0(0)`$, $`1^+0(18)`$, $`1^+1(18)`$, $`3^+0(19)`$ and $`3^+1(19)`$) is much better than those of the well-known broadest resonances ($`2^+0(3)`$ and $`4^+0(11)`$). However, the parameters of the $`4^+0(11)`$ resonance found from $`{}_{}{}^{4}He(\alpha ,\alpha )`$ alone ($`E_x=11.5(3)`$ MeV, $`\mathrm{\Gamma }=4000(400)`$ keV) are in perfect agreement with this analysis. Since the R-matrix analysis contains more data than any known analysis, the experimental masses and widths may well be in doubt, although this is less likely for narrow experimental resonances.
Except for the two very narrow experimental resonances $`2^+(16.6;16.9)`$ that are not considered in the R-matrix fit because no data are entered in their energy region, the following experimental resonances are not found in the analysis: $`4^+0(20)`$, $`(1,2)^{}1(24)`$, and three resonances in the region $`2223`$ MeV with unknown $`J^\pi T`$ . For the latter three resonances, and $`(1,2)^{}1(24)`$, the reason is that these resonances were observed in reactions other than those analyzed here . Of the reactions studied here, the $`4^+0(20)`$ resonance is only non-negligibly observed in $`{}_{}{}^{4}He(\alpha ,\alpha _0)`$ , and data from the experimental reference are not included here.
The narrow ground state $`0^+0`$ resonance parameters in Table 2 are not an improvement on experiment, since no low-energy $`{}_{}{}^{4}He(\alpha ,\alpha _0)`$ data are included at the same excitation energy as the resonance energy. The experimental $`J^\pi T=1^{}\mathrm{?}`$ at $`19`$ MeV , and the $`4^{}\mathrm{?}`$ , are found to have isospin 0, having allowed for both isospins.
The quantum numbers of the peak at $`21.5`$ MeV in the $`{}_{}{}^{7}Li(p,n_0)`$ reaction is experimentally thought to be $`J=3`$, with the parity possibly positive . Our fits prefers the quantum numbers $`J^\pi T=3^{}0`$, having allowed for both parity and both isospin possibilities. The new data included hence updates the old experimental parity assignment based on old data . A positive parity assignment of the $`21.5`$ MeV resonance is inconsistent with theory for the following reason. The only kinematically allowed decay channels analysed here are to $`p^7`$Li and $`n^7`$Be. The NCSM predicts that the $`3^+0`$ and $`3^+1`$ resonances above the lowest-energy resonances with the same quantum numbers have weak couplings to $`p^7`$Li and $`n^7`$Be . The same is true for VMC if the $`T=1`$ <sup>8</sup>Li states are taken as a guide to the $`T=1`$ <sup>8</sup>Be states . The weak couplings to $`p^7`$Li and $`n^7`$Be are not consistent with the need for the resonance here.
Two resonances with the same quantum numbers are found at $`2223`$ MeV in Table 2. The $`2^+0(23)`$ resonance at $`22.78`$ MeV fits the peak observed around 1 MeV $`d`$ laboratory energy in the $`{}_{}{}^{6}Li(d,\alpha _0)`$, $`{}_{}{}^{6}Li(d,p_0)`$ and $`{}_{}{}^{6}Li(d,n_0)`$ reactions. On the other hand, the $`2^+0(22)`$ resonance fits the peak at around 6 MeV $`p`$ laboratory energy in the $`{}_{}{}^{7}Li(p,\alpha _0)`$, and around 45 MeV $`\alpha `$ laboratory energy in the time-inverse $`{}_{}{}^{4}He(\alpha ,p_0)`$ reactions. Although it is conceivable that all these peaks can be fitted with just one $`2^+0`$ resonance, with the $`d^6Li`$ threshold at $`22.28`$ MeV, the current fit clearly prefers two resonances. The lower mass resonance is well established . The existence of the higher mass resonance only became apparent once $`{}_{}{}^{6}Li(d,X)`$ data above $`1`$ MeV $`d`$ laboratory energy were included, and hence does not contradict an analysis of $`{}_{}{}^{6}Li(d,\alpha )`$ data below 1 MeV which only found the $`2^+0(22)`$. The existence of two $`2^+`$ resonances at $`21.5`$ MeV and $`22.5`$ MeV were previously suggested by a qualitative analysis of the $`{}_{}{}^{7}Li(p,n_1)`$ and $`{}_{}{}^{7}Li(p,p_1)`$ reactions not analyzed here, in order to explain a broad dip in the $`n_1`$ yield at the same energy as a broad bump in the $`p_1`$ yield. However, this analysis cannot be regarded as strong evidence for two $`2^+0`$ resonances. It is unclear whether two $`2^+0`$ resonances at $`2223`$ MeV is confirmed by NCSM theory calculations . This calculation does find an extra $`2^+0`$ state at $`1421`$ MeV, which is known as an “intruder” state because it does not appear in the naïve shell model. Whether this intruder should be identified with the $`2^+0(23)`$ or with the extremely broad $`2^+0(16)`$, discussed below, is unclear.
The $`23.25`$ MeV resonance found in the R-matrix analysis (Table 2) is denoted by $`2^+0(25)`$. The reason is that when the peak in $`{}_{}{}^{6}Li(d,\alpha _0)`$ at a $`d`$ laboratory energy of $`3.5`$ MeV is artificially enhanced by substantially decreasing the size of the error bars, the resonance appears at $`25.06`$ MeV, in agreement with experiment, with an unchanged width.
Most of the resonances found in the R-matrix analysis correspond to resonances known experimentally. The exceptions are the extremely broad $`2^+0(16)`$ and very broad $`4^+\mathrm{?}(18)`$ resonances (as well as the $`1^+(20)`$ discussed in the next paragraph). The $`2^+0(16)`$ has previously been reported in an R-matrix analysis of $`\alpha ^4`$He elastic scattering, $`{}_{}{}^{9}Be(p,d)`$ and $`\beta `$-delayed $`2\alpha `$ spectra from <sup>8</sup>Li and <sup>8</sup> at $`9`$ MeV . The energy, but not the existence, of this level is dependent on the channel radius used in the R-matrix fit . For example, an analysis of $`\beta `$-delayed $`2\alpha `$ spectra from <sup>8</sup>Li and <sup>8</sup>B together with $`\mathrm{}=2`$ $`\alpha ^4`$He phase shifts finds that $`2^+`$ intruder states below excitation energy $`26`$ MeV need not be introduced . Although the S-matrix (and its poles and residues) are formally independent of the chosen channel radii for infinitely many R-matrix levels, actual analyses employ a finite number of levels, which can lead to different energies for different channel radii. In addition, the energy of $`2^+0`$ varies by several MeV as new data are included, consistent with the expectation that the energy should not be particularly well constrained for a very broad resonance. A NCSM theory calculation finds the $`2^+0`$ and $`4^+0`$ intruders at $`1421`$ and $`2026`$ MeV respectively . However, a recent GFMC calculation finds no need to introduce extra $`2^+`$ or $`4^+`$ states below respectively $`22`$ and $`19`$ MeV . The disagreement between NCSM and GFMC may be due to the large widths of the intruder states (Table 2), which imply substantial variation in the energies extracted from these calculations which treat all the states as bound. Whether very broad states should be seen in calculations that treat states as bound is debatable.
The current fit has a new $`1^+1(20)`$ resonance. Although it is not listed in the standard experimental compilation , it is interesting to note that theory calculations predict such states: NCSM predicts one $`1^+0`$ resonance and two $`1^+1`$ resonances at $`2022`$ MeV , and GFMC one $`1^+0`$ at $`19`$ MeV . It is intriguing to note two coincidences between this analysis and theory. (i) The NCSM predicts large couplings of a $`20.37`$ MeV $`1^+1`$ state to $`p^7`$Li and $`n^7`$Be and not to $`d^6`$Li . The robust $`1^+1(20)`$ resonance seen in this analysis is at $`E_x=20.45`$ MeV from Table 2, with strong couplings to $`p^7`$Li and $`n^7`$Be and not to $`d^6`$Li according to Table 3. (ii) Of the three $`1^+`$ resonances predicted at $`2022`$ MeV in NCSM, only the $`20.37`$ MeV $`1^+1`$ has large couplings to $`p^7`$Li and $`n^7`$Be, which are the only kinematically open channels for decay, amongst the channels analysed here . The same is true for VMC if the $`T=1`$ <sup>8</sup>Li states are taken as a guide to the $`T=1`$ <sup>8</sup>Be states . This coincides with the finding here that only one new $`1^+`$ state is needed, and that this state has isospin 1.
The $`2^{}`$ resonance is conceptually complicated because it lies exactly at the $`n^7`$Be threshold, and hence requires sophysticated analysis. Several such analyses have been performed , typically yielding a resonance with $`E_x=18.9`$ MeV and $`\mathrm{\Gamma }100`$ keV, although there is disagreement on the width. Most strikingly, an analysis of $`{}_{}{}^{7}Li(p,n_0)`$ and $`{}_{}{}^{7}Be(n,p_0)`$ data finds $`\mathrm{\Gamma }=1634`$ keV , based on a prescription whereby the sum of the $`\mathrm{\Gamma }_c`$ equals $`\mathrm{\Gamma }`$. As previously mentioned, this is not the case in our analysis. In contrast, another multi-level R-matrix analysis defines the resonance energy and width as the properties of the pole of the S-matrix, yielding a total width much lower than the sum of the partial widths. This corresponds closely to our conventions, yielding $`\mathrm{\Gamma }=122`$ keV, $`T=0`$ and isospin impurity $`24`$. This isospin impurity is at odds with $`10`$% obtained from $`{}_{}{}^{7}Li(p,\gamma )^8Be^{}(18.9)`$ . The current analysis assigns $`T=1`$ for the $`2^{}`$ resonance (Table 2). A cautionary note should be mentioned. For all the resonances reported here except the $`2^{}`$, the parameters of the pole on the unphysical sheet closest to the physical sheet are quoted in Tables 2-3, as this is thought to be physically most relevant. However, there are poles on other sheets which are physically less relevant. The $`2^{}0(19)`$ is unique in that the resonance is very close to threshold, which blurs the usual prescription for which of the poles are most physically relevant. The parameters of the pole which has an energy exactly at the $`n^7`$Be threshold is displayed in Tables 2-3 because its $`E_x`$ and $`\mathrm{\Gamma }`$ correspond most closely to other analyses. There is another nearby pole (on the unphysical sheet closest to the physical sheet) with $`E_x=18.73`$ MeV, a much larger width $`\mathrm{\Gamma }=640`$ keV, $`T=1`$ and isospin impurity $`31`$%. This pole has the opposite pattern of coupling to the channels: it couples stronger to $`p^7`$Li and weaker to $`n^7`$Be.
The $`1^{}1(22)`$ resonance has previously only been observed in the $`{}_{}{}^{7}Li(p,\gamma _0)`$ reaction . This analysis finds a need to introduce this resonance with a strong coupling to $`p^7`$Li and $`n^7`$Be in the spin $`2`$, D-wave. The parameters of $`1^{}1(22)`$ are not strongly fixed by this analysis and are hence not displayed.
## 5 Conclusions
The <sup>8</sup>Be resonance parameters of most of the resonances up to $`26`$ MeV are determined. The isospins of the $`19`$ MeV $`J^\pi =1^{}`$ and the $`4^{}`$ resonances are determined for the first time to be 0. The $`21`$ MeV resonance which was previously assigned to possibly have positive parity is found to be $`J^\pi T=3^{}0`$. The previously known $`22`$ MeV $`2^+0`$ resonance likely splits into two resonances. A new $`1^+1`$ resonance at $`20`$ MeV is discovered. The resonance parameters enable comparison with GFMC and NCSM theory calculations. Two broad resonances are found, which may not appear in these calculations that treat the states as bound. These resonances are the extremely broad $`2^+0`$ resonance at $`16`$ MeV, whose existence is confirmed, and a very broad $`4^+0`$ resonance at $`18`$ MeV, which is discovered for the first time. The location of the $`T=1`$ resonances is relevant to sorting out the structure of <sup>8</sup>Li and <sup>8</sup>B. Incorporation of the resonance structure found here in future TUNL evaluations is advocated.
Helpful discussions with G.M. Hale are gratefully acknowledged. The RESP code for the extraction of S-matrix poles was written by G.M. Hale. Some of the data analysed was entered by others, including G.M. Hale and A.S. Johnson . P. Navrátil provided detailed calculations further to Refs. . This research is supported by the Department of Energy under contract W-7405-ENG-36.
|
warning/0506/quant-ph0506104.html
|
ar5iv
|
text
|
# Canonical quantization of nonlinear many body systems
## I Introduction
A wide class of diffusive processes in nature, known as normal diffusion, are successfully described by the linear Fokker-Planck equation. Its relation to Boltzmann-Gibbs entropy (BG-entropy) in the framework of the irreversible thermodynamics is well established deGroot ; Prigogine ; Glansdorff .
However, nonlinear Fokker-Planck equations (NFPEs) Frank1 ; Frank2 ; Chavanis1 ; Chavanis4 ; Chavanis2 and their connection in the field of the generalized thermodynamics Abe ; Kaniadakis0 ; Kaniadakis00 is nowadays an intense research area. In particular, many physical phenomena, in presence of memory effects, nonlocal effects, long-range effects or, more in general, nonlinear effects, are well understood with the help of NFPEs.
To cite a few, let us recall the problem of diffusion in polymers Ott , on liquid surfaces Bychuk , in Lévy flights Solomon and enhanced diffusion in active intracellular transport Caspi . Many anomalous diffusion systems have a quantum nature, like for instance charge transport in anomalous solids Sher , subrecoil laser cooling Bardou and the aging effect in quantum dissipative systems Mauger .
A still open question concerns the dynamics underlying the nonlinear kinetics governing the above anomalous systems. Langevin-like, Fokker-Planck-like or Boltzmann-like equations have been used by different authors to generate nonlinear terms in the Schrödinger equation with the aim of describing, in the mean field approximation, the many quantum particle interactions Kostin ; Schuch1 ; Schuch ; Kaniadakis .
It is now widely recognized that the presence of a nonlinear drift term as well as the presence of a diffusive term in a quantum particle current originates complex nonlinearities in the evolution equation for the $`\psi `$-wave function.
Different examples are known in literature of nonlinear Schrödinger equations (NSEs) originating from the study of the kinetics governing the many-body quantum system. For instance, the Doebner-Goldin family equations Doebner1 have been introduced from topological considerations as the most general class of Schrödinger equations compatible with the linear Fokker-Planck equation. In Ref. Scarfone3 the authors introduced a NSE starting from a generalized exclusion-inclusion principle (EIP) in order to describe systems of quantum particles with different statistics interpolating with continuity between the Bose-Einstein and the Fermi-Dirac ones. In Ref. Scarfone7 , in the stochastic quantization framework, starting from the most general nonlinear kinetics containing a nonlinear drift term and compatible with a linear diffusion term, a class of NSEs with a complex nonlinearity was obtained.
Recently, a kinetic interaction principle (KIP) has been proposed Kaniadakis1 to define a special collective interaction among the $`N`$-identical particles of a classical system. On the one hand, the KIP imposes the form of the generalized entropy associated with the system, while on the other hand it governs the evolution of the system toward equilibrium by fixing the expression of the nonlinear current of particles in the NFPE, thus governing the kinetics underlying the system.
The link between the generalized entropic functional and the corresponding NFPE can also be obtained starting from a maximum entropic production principle. In Refs. Chavanis1 ; Chavanis4 , taking into account a variational principle maximizing the dissipation rate of a generalized free energy, the authors obtained a NFPE in the Smoluchowski limit. The same NFPE was obtained in Ref. Chavanis2 from a stochastic process described by a generalized Langevin equation where the strength of the noise is assumed to depend on the density of the particle.
In the present paper we perform the quantization of a classical system obeying KIP, where the statistical information is supplied by a very general entropy.
Up to today, different methods have been proposed for the microscopic description of systems. Schrödinger’s wave mechanics, Heisenberg’s matrix mechanics or Feynman’s path-integral mechanics are some of the many. Another approach is given by the hydrodynamic theory of quantum mechanics originally owing to Madelung Madelung and de Broglie Broglie and successively reconsidered by Bohm Bohm in connection with his theory of hidden variables.
In the hydrodynamic formulation of quantum mechanics, the complex linear Schrödinger equation is replaced by two real nonlinear differential equations for two independent fields: the probability density and its velocity field. Basically, such equations are formally similar to the equations of continuity and the Euler equation of ordinary hydrodynamics.
This formalism is fruitful, as in the present situation, when the expression of the quantum continuity equation is inherited from the one describing the kinetics of the ancestor classical system. However, for a complete quantum mechanical description, besides the continuity equation, we need to know if and how we should generalize the Euler equation that describes the dynamics of the system. In this paper, in order to fix the nonlinear terms in the Euler equation, we require that the whole model be formulated in the canonical formalism.
We obtain a class of NSEs with complex nonlinearity describing a quantum system of interacting particles obeying the KIP in the mean field approximation. We study the case of a quantum system undergoing a constant diffusion process. The generalization to the case of a nonconstant diffusive process is also presented at the end of the paper. It is shown that the form of the entropy of the ancestor classical system fixes the nonlinearity appearing in the evolution equation. By means of a recently proposed nonlinear gauge transformation Doebner1 ; Scarfone1 ; Scarfone4 this family of evolution equations is transformed into another one describing a nondiffusive process. In particular, when the kinetics of the system is governed by a linear drift term, the new family of NSEs contains a purely real nonlinearity depending only on the density of particles $`\rho =|\psi |^2`$.
As working examples we present the quantization of some classical systems described by entropies already known in the literature: BG-entropy, Tsallis-entropy Tsallis , Kaniadakis-entropy Kaniadakis1 and the interpolating quantum statistics entropy Quarati1 .
The plan of the paper is the following. In Section II we recall the relation between a given generalized entropy and the associated NFPE describing the kinetic evolution of the classical system in the nonequilibrium thermodynamic framework. This kinetic equation is justified on the ground of KIP. In Section III, firstly first present an overall summing up of the hydrodynamic formulation of the linear Schrödinger equation, then we generalize the method to quantize the classical system obeying EIP. The Hamiltonian formulation of this model is presented and a family of NSEs with complex nonlinearity is obtained. In Section IV we study the Ehrenfest relations and discuss the conserved mean quantities. In Section V, the nonlinear gauge transformation is introduced. Some relevant examples are presented in Section VI. The final Section VII present comments and conclusions. In Appendix A we give the derivation of the Ehrenfest relations while in Appendix B we briefly discuss the generalization of the model for a quantum system whose kinetics undergoes a nonconstant diffusive process.
## II Nonlinear Fokker-Planck equation
Our starting point, according to nonlinear kinetics, is to relate the production of the entropy of a classical system to a Fokker-Planck equation. This can be accomplished by following the classical approach to diffusion deGroot ; Prigogine .
We start by assuming a very general trace-form expression for the entropy (throughout this paper, we use units with the Boltzmann constant $`k__\mathrm{B}`$ set equal to unity)
$$S(\rho )=𝑑𝒙𝑑\rho \mathrm{ln}\kappa (\rho ),$$
(2.1)
where $`\kappa (\rho )`$ is an arbitrary functional of the density particles field $`\rho =\rho (t,𝒙)`$, with $`𝒙(x__1,\mathrm{},x__n)`$ a point in the $`n`$-dimensional space.
The constraints
$$\rho 𝑑𝒙=1,$$
(2.2)
on the normalization and
$$(𝒙)\rho 𝑑𝒙=E,$$
(2.3)
total energy of the system, with $`(𝒙)=𝒑^2/2m+V(𝒙)`$ the energy for each particle, are accounted for by introducing the constrained entropic functional
$$𝒮(\rho )=𝑑𝒙𝑑\rho \mathrm{ln}\kappa (\rho )\beta (𝒙)\rho 𝑑𝒙\beta ^{}\rho 𝑑𝒙.$$
(2.4)
The two constants $`\beta `$ and $`\beta ^{}`$ are the Lagrange multipliers associated with constraints (2.2) and (2.3).
Quite generally, the evolution of the field $`\rho `$ in the configuration space is governed by the continuity equation
$$\frac{\rho }{t}+\mathbf{}𝑱=0,$$
(2.5)
with $`\mathbf{}(/x__1,\mathrm{},/x__n)`$, and assures the conservation of the constraint (2.2) in time.
We assume a nonlinear relation between the current $`𝑱`$ and the constrained thermodynamic force
$$𝓕(\rho )=\mathbf{}\left(\frac{\delta 𝒮}{\delta \rho }\right),$$
(2.6)
by posing
$$𝑱=D\gamma (\rho )𝓕(\rho ),$$
(2.7)
with $`D`$ the diffusion coefficient and $`\gamma (\rho )`$ still an arbitrary functional of $`\rho `$.
Putting Eq. (2.7) in Eq. (2.5), and taking into account the expression of $`𝒮`$ given in Eq. (2.4) we obtain the following continuity equation
$$\frac{\rho }{t}+\mathbf{}\left\{D\gamma (\rho )\mathbf{}\left[\beta (𝒙)+\beta ^{}+\mathrm{ln}\kappa (\rho )\right]\right\}=0.$$
(2.8)
Introducing drift velocity
$$𝒖_{\mathrm{drift}}=D\beta \mathbf{}(𝒙),$$
(2.9)
Eq. (2.8) takes the form of a NFPE for the field $`\rho `$
$$\frac{\rho }{t}+\mathbf{}\left[𝒖_{\mathrm{drift}}\gamma (\rho )Df(\rho )\mathbf{}\rho \right]=0,$$
(2.10)
where
$$f(\rho )=\gamma (\rho )\frac{\mathrm{ln}\kappa (\rho )}{\rho }.$$
(2.11)
Total current $`𝑱=𝑱_{\mathrm{drift}}+𝑱_{\mathrm{diff}}`$ is the sum of a nonlinear drift current $`𝑱_{\mathrm{drift}}=𝒖_{\mathrm{drift}}\gamma (\rho )`$, and a nonlinear diffusion current $`𝑱_{\mathrm{diff}}=Df(\rho )\mathbf{}\rho `$, different from Fick’s standard one $`𝑱_{\mathrm{Fick}}=D\mathbf{}\rho `$, which is recovered by posing $`\gamma (\rho )=\kappa (\rho )=\rho `$.
Eq. (2.10) describes a class of nonlinear diffusive processes varying the functionals $`\gamma (\rho )`$ and $`\kappa (\rho )`$.
We observe that for any given entropy (2.1) an infinity of associated NFPEs exists, one for any choice of $`\gamma (\rho )`$.
In Refs. Chavanis1 ; Chavanis4 , starting from a variational principle which maximizes the dissipation rate of a generalized free energy functional, substantially equivalent to Eq. (2.4), a NFPE in the position space as in Eq. (2.10) has been obtained. The same NFPE (2.10) was also obtained in Ref. Chavanis2 , starting from a stochastic process described by a generalized Langevin equation, where the strength of the noise is assumed to depend on the density of the particle. The nonlinear current, as in Eq. (2.7), is given by the gradient of the functional derivative of a generalized free energy equivalent to Eq. (2.4).
In Ref. Frank1 the problem of the NFPE derived from generalized linear nonequilibrium thermodynamics was also discussed at length.
At equilibrium, the particle current must vanish, and from Eq. (2.6) it follows
$$\mathrm{ln}\kappa (\rho _{\mathrm{eq}})+\beta (𝒙)+\beta ^{}=0,$$
(2.12)
where, without loss of generality, we posed the integration constant equal to zero (otherwise it can be included in the Lagrange multiply $`\beta ^{}`$).
We obtain the equilibrium distribution of the system
$$\rho _{\mathrm{eq}}=\kappa ^1\left(\mathrm{exp}\left(\beta (𝒙)\beta ^{}\right)\right).$$
(2.13)
In particular, with the choice $`\kappa (\rho )=e\rho `$, Eq. (2.1) reduces to standard BG-entropy and Eq. (2.13) gives the well-known Gibbs-distribution.
Let us now justify Eq. (2.10) starting from the kinetic approach introduced in Kaniadakis1 through the KIP. In accordance with the arguments presented in Ref. Kaniadakis1 , we consider the following classical Markovian process
$$\frac{\rho }{t}=\left[\pi (t,𝒚𝒙)\pi (t,𝒙𝒚)\right]𝑑𝒚,$$
(2.14)
describing the kinetics of a system of $`N`$-identical interacting particles.
For transition probability $`\pi (t,𝒙𝒚)`$ we assume a suitable expression in terms of the populations of the initial site $`𝒙`$ and the final site $`𝒚`$.
According to KIP we pose
$$\pi (t,𝒙𝒚)=r(t,𝒙,𝒙𝒚)\gamma (\rho ,\rho ^{}),$$
(2.15)
where $`\rho \rho (t,𝒙)`$ and $`\rho ^{}\rho (t,𝒚)`$ are the particle density functions in the starting site $`𝒙`$ and in the arrival site $`𝒚`$ respectively, whereas $`r(t,𝒙,𝒙𝒚)`$ is the transition rate which depends only on the starting $`𝒙`$ and arrival $`𝒚`$ sites, during particle transition $`𝒙𝒚`$.
The functional $`\gamma (\rho ,\rho ^{})`$ can be factorized in
$$\gamma (\rho ,\rho ^{})=a(\rho )b(\rho ^{})c(\rho ,\rho ^{}).$$
(2.16)
The first factor $`a(\rho )`$ is a functional of the particle population $`\rho `$ of the starting site and satisfies the boundary condition $`a(0)=0`$, since if the starting site is empty transition probability is equal to zero. The second factor $`b(\rho ^{})`$ is a functional of the particle population $`\rho ^{}`$ at the arrival site, and satisfies the condition $`b(0)=1`$, because the transition probability does not depend on the arrival site if particles are absent there. Finally, the third factor $`c(\rho ,\rho ^{})`$ takes into account that the populations of the two sites can eventually affect the transition collectively and symmetrically.
The expression of the functional $`b(\rho ^{})`$ plays a very important role in the particle kinetics because it can stimulate or inhibit the transition $`𝒙𝒚`$, allowing, in this way, interactions originating from collective effects.
With the assumptions made in Eqs. (2.15) and (2.16) for transition probability, according to the Kramers-Moyal expansion and assuming the first neighbor approximation, we can expand up to the second order the quantities $`r(t,𝒚,𝒚𝒙)\gamma (\rho (t,𝒚),\rho (t,𝒙))`$ and $`\gamma (\rho (t,𝒙),\rho (t,𝒚))`$ in Taylor series of $`𝒚=𝒙+𝒖`$ and $`𝒚=𝒙𝒖`$, respectively, in an interval around $`𝒙`$, for $`𝒖𝒙`$. We obtain
$`r(t,𝒙+𝒖,𝒖)\gamma (\rho (t,𝒙+𝒖),\rho (t,𝒙))`$
$`=\{r(t,𝒚,𝒖)\gamma (\rho (t,𝒚),\rho (t,𝒙))+{\displaystyle \frac{}{y__i}}\left[r(t,𝒚,𝒖)\gamma (\rho (t,𝒚),\rho (t,𝒙))\right]u__i`$
$`+{\displaystyle \frac{1}{2}}{\displaystyle \frac{^2}{y__iy__j}}\left[r(t,𝒚,𝒖)\gamma (\rho (t,𝒚),\rho (t,𝒙))\right]u__iu__j\}_{𝒚𝒙},`$ (2.17)
and
$`\gamma (\rho (t,𝒙),\rho (t,𝒙𝒖))`$ $`=\{\gamma (\rho (t,𝒙),\rho (t,𝒚)){\displaystyle \frac{}{y__i}}\gamma (\rho (t,𝒙),\rho (t,𝒚))u__i.`$ (2.18)
$`+{\displaystyle \frac{1}{2}}{\displaystyle \frac{^2}{y__iy__j}}\gamma (\rho (t,𝒙),\rho (t,𝒚))u__iu__j\}_{𝒚𝒙}.`$
Using Eqs. (2.17) and (2.18) in Eq. (2.15), from Eq. (2.14) it follows
$$\frac{\rho }{t}=\frac{}{x__i}\left[\left(\zeta __i+\frac{\zeta _{_{ij}}}{x__j}\right)\gamma (\rho )+\zeta _{_{ij}}\gamma (\rho )\frac{}{x__j}\mathrm{ln}\kappa (\rho )\right],$$
(2.19)
with $`i=1,\mathrm{},n`$ and summation over repeated indices is assumed.
In Eq. (2.19)
$$\gamma (\rho )\gamma (\rho ,\rho ^{})|_{\rho =\rho ^{}},$$
(2.20)
and
$$\kappa (\rho )=\frac{a(\rho )}{b(\rho )},$$
(2.21)
while the coefficients $`\zeta __i`$ and $`\zeta _{_{ij}}`$ are given by
$$\zeta __i=r(t,𝒚,𝒖)u__i𝑑𝒖,$$
(2.22)
$$\zeta _{_{ij}}=\frac{1}{2}r(t,𝒚,𝒖)u__iu__j𝑑𝒖.$$
(2.23)
Defining $`(u__i)_{\mathrm{drift}}=\zeta __i\zeta _{_{ij}}/x__j`$, the $`i`$-th component of $`𝒖_{\mathrm{drift}}`$, and assuming the independence of motion in different directions of the isotropic configuration space we can pose $`\zeta _{_{ij}}=D\delta _{_{ij}}`$. It is easy to see that Eq. (2.19) reduces to Eq. (2.10).
In conclusion we observe that Eq. (2.10) is a NFPE in the Smoluchowski limit since it describes a kinetic process in the position space rather than in the phase space. This is a suitable form for the quantum treatment of the following sections. The passage from the NFPE in the phase space to the NFPE in the position space was rigorously elaborated in Ref. Chavanis3 in the limit of strong friction, by means of a Chapman-Enskog-like expansion.
## III Canonical quantization
### III.1 Quantization in the hydrodynamic representation
In the hydrodynamic representation, the quantum mechanics formulation, can readily be obtained from the standard Schrödinger equation
$$i\mathrm{}\frac{\psi }{t}=\frac{\mathrm{}^2}{2m}\mathrm{\Delta }\psi +V(𝒙)\psi ,$$
(3.1)
where $`V(𝒙)`$ is a real external potential. The complex field $`\psi \psi (t,𝒙)`$ describing the quantum system is related to the hydrodynamic fields $`\rho (t,𝒙)`$ and $`\mathrm{\Sigma }(t,𝒙)`$ through polar decomposition Bohm ; Madelung
$$\psi (t,𝒙)=\rho ^{1/2}(t,𝒙)\mathrm{exp}\left(\frac{i}{\mathrm{}}\mathrm{\Sigma }(t,𝒙)\right).$$
(3.2)
Eq. (3.1) is separated into a couple of real equations
$`m{\displaystyle \frac{\widehat{𝒗}}{t}}+m\left(\widehat{𝒗}\mathbf{}\right)\widehat{𝒗}=\mathbf{}\left({\displaystyle \frac{\mathrm{}^2}{2m}}{\displaystyle \frac{\mathrm{\Delta }\sqrt{\rho }}{\sqrt{\rho }}}V(𝒙)\right),`$ (3.3)
$`{\displaystyle \frac{\rho }{t}}+\mathbf{}𝒋__0=0,`$ (3.4)
where quantum velocity $`\widehat{𝒗}`$, which in the linear case coincides with quantum drift velocity $`\widehat{𝒖}_{\mathrm{drift}}`$, is related to the phase $`\mathrm{\Sigma }(t,𝒙)`$ through
$$m\widehat{𝒗}=\mathbf{}\mathrm{\Sigma }(t,𝒙),$$
(3.5)
and
$$𝒋__0=\rho \widehat{𝒗},$$
(3.6)
is the same relationship between current and velocity of the standard hydrodynamic theory. We remark that the quantum current (3.6) contains only a linear drift term.
According to the orthodox interpretation of quantum mechanics the quantity $`\rho (t,𝒙)=|\psi (t,𝒙)|^2`$ represents the position probability density of the system normalized as $`\rho (t,𝒙)𝑑𝒙=1`$.
Eqs. (3.3)-(3.6) form the basis of the hydrodynamic formulation which consists of a quasi classical approach to quantum mechanics. In this picture the evolution of the system can be interpreted in terms of a flowing fluid with density $`\rho (t,𝒙)`$ associated with a local velocity field $`\widehat{𝒗}(t,𝒙)`$. The dynamics of such fluid is described by the Euler equation (3.3) and is governed by forces arising not only from the external field $`𝑭_{\mathrm{ext}}(𝒙)=\mathbf{}V(𝒙)`$ but also from an additional potential $`U_q=(\mathrm{}^2/2m)\mathrm{\Delta }\sqrt{\rho }/\sqrt{\rho }`$ known as the quantum potential Bohm . Remarkably, the expectation value for the quantum force vanishes at all times, i.e. $`\mathbf{}U_q=0`$. Finally, the continuity equation (3.4) assures the conservation of the normalization of wave function $`\psi `$ during the evolution of the system.
Let us remark that the quantum fluid has a very special property. Because $`\mathrm{\Sigma }(t,𝒙)`$ is a potential field for the quantum velocity, the quantum fluid is irrotational. As a consequence, in the linear Schrödinger theory, a non vanishing vorticity $`𝝎`$, defined by
$$𝝎=\mathbf{}\times \widehat{𝒗},$$
(3.7)
is possible only at the nodal region where neither $`\mathrm{\Sigma }(t,𝒙)`$ nor $`\mathbf{}\mathrm{\Sigma }(t,𝒙)`$ are well defined. At such a point $`\mathbf{}\times \mathbf{}\mathrm{\Sigma }(t,𝒙)`$ does not vanish in general, thus leading to the appearance of point-like vortices.
Finally, putting Eq. (3.5) into Eq. (3.3) we obtain
$$\frac{\mathrm{\Sigma }}{t}+\frac{\left(\mathbf{}\mathrm{\Sigma }\right)^2}{2m}\frac{\mathrm{}^2}{2m}\frac{\mathrm{\Delta }\sqrt{\rho }}{\sqrt{\rho }}+V(𝒙)=0.$$
(3.8)
This equation, in the classical limit $`\mathrm{}0`$, reduces to the Hamilton-Jacobi equation for the function $`\mathrm{\Sigma }`$.
Eqs. (3.4) and Eq. (3.8) can be obtained by means of the Hamiltonian equations
$`{\displaystyle \frac{\mathrm{\Sigma }}{t}}={\displaystyle \frac{\delta H}{\delta \rho }},`$ (3.9)
$`{\displaystyle \frac{\rho }{t}}={\displaystyle \frac{\delta H}{\delta \mathrm{\Sigma }}},`$ (3.10)
where the Hamiltonian
$$H=(\rho ,\mathrm{\Sigma })𝑑𝒙,$$
(3.11)
with
$$(\rho ,\mathrm{\Sigma })=\frac{(\mathbf{}\mathrm{\Sigma })^2}{2m}\rho +\frac{\mathrm{}^2}{8m}\frac{\left(\mathbf{}\rho \right)^2}{\rho }+V(𝒙)\rho .$$
(3.12)
represents the total energy of the quantum system.
### III.2 The many-body quantum system
Let us now generalize the method described above by replacing the linear continuity equation Eq. (3.4) with the more general one obtained in analogy with the continuity equation (2.10) describing the kinetics of a classical system obeying KIP. In the following we assume that the quantum system undergoes a constant diffusion process with $`D=const.`$
We begin by introducing the wave function $`\psi \psi (t,𝒙)`$ describing, in the mean field approximation, a system of quantum interacting particles. We postulate that the following NSE describes the evolution equation of the system
$$i\mathrm{}\frac{\psi }{t}=\frac{\mathrm{}^2}{2m}\mathrm{\Delta }\psi +\mathrm{\Lambda }(\psi ^{},\psi )\psi +V(𝒙)\psi ,$$
(3.13)
where $`\mathrm{\Lambda }(\psi ^{},\psi )=W(\psi ^{},\psi )+i𝒲(\psi ^{},\psi )`$ is a complex nonlinearity, with $`W(\psi ^{},\psi )`$ and $`𝒲(\psi ^{},\psi )`$ the real and the imaginary part, respectively.
Using polar decomposition (3.2), Eq. (3.13) is separated into a couple of real nonlinear evolution equations for phase and amplitude
$`{\displaystyle \frac{\mathrm{\Sigma }}{t}}+{\displaystyle \frac{\left(\mathbf{}\mathrm{\Sigma }\right)^2}{2m}}+U_q+W(\rho ,\mathrm{\Sigma })+V(𝒙)=0,`$ (3.14)
$`{\displaystyle \frac{\rho }{t}}+\mathbf{}𝒋__0{\displaystyle \frac{2}{\mathrm{}}}\rho 𝒲(\rho ,\mathrm{\Sigma })=0.`$ (3.15)
We require that both Eqs. (3.14) and (3.15) can be obtained through the Hamilton equations (3.9)-(3.10) and, to accommodate nonlinearities $`W(\rho ,\mathrm{\Sigma })`$ and $`𝒲(\rho ,\mathrm{\Sigma })`$, we introduce in the Hamiltonian density $``$ an additional real nonlinear potential $`U(\rho ,\mathrm{\Sigma })`$ which describes the collective interaction between the particles belonging to the system
$$(\rho ,\mathrm{\Sigma })=\frac{(\mathbf{}\mathrm{\Sigma })^2}{2m}\rho +\frac{\mathrm{}^2}{8m}\frac{\left(\mathbf{}\rho \right)^2}{\rho }+U(\rho ,\mathrm{\Sigma })+V(𝒙)\rho .$$
(3.16)
By means of Eqs. (3.9) and (3.10) it follows that the nonlinear functionals $`W(\rho ,\mathrm{\Sigma })`$ and $`𝒲(\rho ,\mathrm{\Sigma })`$ are related to the nonlinear potential $`U(\rho ,\mathrm{\Sigma })`$ as
$`W(\rho ,\mathrm{\Sigma })={\displaystyle \frac{\delta }{\delta \rho }}{\displaystyle U(\rho ,\mathrm{\Sigma })𝑑𝒙},`$ (3.17)
$`𝒲(\rho ,\mathrm{\Sigma })={\displaystyle \frac{\mathrm{}}{2\rho }}{\displaystyle \frac{\delta }{\delta \mathrm{\Sigma }}}{\displaystyle U(\rho ,\mathrm{\Sigma })𝑑𝒙}.`$ (3.18)
We assume that the quantum fluid satisfies a continuity equation formally equal to the classical one described by the NFPE (2.10). By matching Eq. (3.15) with Eq. (2.10) we obtain the expression $`𝒲`$ and, accounting for Eq. (3.18), we have the nonlinear potential $`U(\rho ,\mathrm{\Sigma })`$. Finally, the nonlinearity $`W(\rho ,\mathrm{\Sigma })`$, which follows from Eq. (3.17), together with the quantum potential $`U_q`$ and the external potential $`V(𝒙)`$, describes the dynamic behavior of the quantum fluid according to Eq. (3.14).
We observe that if the following equation holds
$$\frac{\delta }{\delta \mathrm{\Sigma }}U(\rho ,\mathrm{\Sigma })𝑑𝒙=\mathbf{}𝑭(\rho ,\mathrm{\Sigma }),$$
(3.19)
with $`𝑭(\rho ,\mathrm{\Sigma })`$ an arbitrary functional, taking into account Eq. (3.18), Eq. (3.15) becomes
$$\frac{\rho }{t}+\mathbf{}\left[𝒋__0𝑭(\rho ,\mathrm{\Sigma })\right]=0.$$
(3.20)
Eq. (3.19) is fulfilled if functional $`U(\rho ,\mathrm{\Sigma })`$ depends on phase $`\mathrm{\Sigma }`$ only through its spatial derivatives Scarfone1 .
Introducing the quantum drift velocity
$$\widehat{𝒖}_{\mathrm{drift}}=\frac{\mathbf{}\mathrm{\Sigma }}{m},$$
(3.21)
which in the linear case coincides with the quantum velocity $`\widehat{𝒗}`$ given in Eq. (3.5), and by comparing Eq. (3.20) with Eq. (2.10) we have
$$𝑭(\rho ,\mathrm{\Sigma })=\frac{\mathbf{}\mathrm{\Sigma }}{m}\left[\rho \gamma (\rho )\right]+Df(\rho )\mathbf{}\rho .$$
(3.22)
By integrating Eq. (3.18), the nonlinear potential assumes the expression
$$U(\rho ,\mathrm{\Sigma })=\frac{(\mathbf{}\mathrm{\Sigma })^2}{2m}\left[\gamma (\rho )\rho \right]Df(\rho )\mathbf{}\rho \mathbf{}\mathrm{\Sigma }+\stackrel{~}{U}(\rho ),$$
(3.23)
where $`\stackrel{~}{U}(\rho )`$ is an arbitrary real potential depending only on field $`\rho `$. Eqs. (3.9) and (3.10) give the following coupled nonlinear evolution equations
$`{\displaystyle \frac{\mathrm{\Sigma }}{t}}+{\displaystyle \frac{(\mathbf{}\mathrm{\Sigma })^2}{2m}}{\displaystyle \frac{\gamma (\rho )}{\rho }}{\displaystyle \frac{\mathrm{}^2}{2m}}{\displaystyle \frac{\mathrm{\Delta }\sqrt{\rho }}{\sqrt{\rho }}}+mDf(\rho )\mathbf{}\left({\displaystyle \frac{𝒋__0}{\rho }}\right)+G(\rho )+V(𝒙)=0,`$
$`{\displaystyle \frac{\rho }{t}}+\mathbf{}\left[{\displaystyle \frac{\mathbf{}\mathrm{\Sigma }}{m}}\gamma (\rho )Df(\rho )\mathbf{}\rho \right]=0,`$ (3.25)
where $`G(\rho )=\delta \stackrel{~}{U}(\rho )𝑑𝒙/\delta \rho `$.
In the classical limit $`\mathrm{}0`$ Eq. (III.2) becomes a nonlinear Hamilton-Jacobi equation for function $`\mathrm{\Sigma }`$. It differs from the classical one owing to the presence of the nonlinear term which functionally depends on both $`\rho `$ and $`\mathrm{\Sigma }`$. We recall that such a nonlinearity was introduced consistently with the requirement of a final canonical formulation of the theory.
We stress once again that in the approach described in this paper, we start from a nonlinear generalization of the continuity equation that gives us only information on the kinetics. This equation is not enough to completely determine the time evolution of the quantum system. As a consequence, we have ample freedom in the definition of nonlinear potential $`U(\rho ,\mathrm{\Sigma })`$. Such freedom is reflected in the arbitrary functional $`\stackrel{~}{U}(\rho )`$ which cannot be fixed only on the basis of the kinetic equation. There are many possible dynamic behaviors, one for any choice of $`\stackrel{~}{U}(\rho )`$, compatible with the same kinetics. The nonlinear potential $`\stackrel{~}{U}(\rho )`$ can be used to describe other possible interactions among the many particles of the system that have an origin different from the one introduced by the kinetic equation (3.25).
Actually, Eq. (3.25) is a quantum continuity equation for field $`\rho `$ with a nonlinear quantum current given by
$$𝒋=\frac{\mathbf{}\mathrm{\Sigma }}{m}\gamma (\rho )Df(\rho )\mathbf{}\rho .$$
(3.26)
We observe that, differently from the hydrodynamic formulation of the linear quantum mechanics, where the Bohm-Madelung fluid is irrotational, in nonlinear quantum theory the situation can be very different. In fact, by defining quantum velocity through Eq. (3.6), from Eq. (3.26) we have
$$m\widehat{𝒗}=\frac{\gamma (\rho )}{\rho }\mathbf{}\left[\mathrm{\Sigma }mD\mathrm{ln}\kappa (\rho )\right],$$
(3.27)
which states the relationship between quantum velocity $`\widehat{𝒗}`$ and quantum drift velocity $`\widehat{𝒖}_{\mathrm{drift}}`$ for the nonlinear case.
Expression (3.27) can be justified in terms of Clebsh potentials. In fact, as is well known, a nonvanishing vorticity can be accounted for in the Schrödinger theory by introducing three potentials $`\mu `$, $`\nu `$ and $`\lambda `$ related to quantum velocity through the relation
$$m\widehat{𝒗}=\mathbf{}\mu +\nu \mathbf{}\lambda .$$
(3.28)
Vorticity $`𝝎`$ assumes a nonvanishing expression given by
$$𝝎=\frac{1}{m}\mathbf{}\nu \times \mathbf{}\lambda .$$
(3.29)
By comparing Eq. (3.28) with Eq. (3.27) we readily obtain $`\mu =const`$, $`\nu =\gamma (\rho )/\rho `$ and $`\lambda =\mathrm{\Sigma }mD\mathrm{ln}\kappa (\rho )`$, respectively, and Eq. (3.29) becomes
$$𝝎=\frac{1}{m}\mathbf{}\left(\frac{\gamma (\rho )}{\rho }\right)\times \mathbf{}\mathrm{\Sigma },$$
(3.30)
with no any contribution from the diffusive term. The irrotational case is recovered in linear drift $`\gamma (\rho )=\rho `$.
The final expression of the NSE (3.13) is given by
$$i\mathrm{}\frac{\psi }{t}=\frac{\mathrm{}^2}{2m}\mathrm{\Delta }\psi +\left[W(\rho ,\mathrm{\Sigma })+i𝒲(\rho ,\mathrm{\Sigma })\right]\psi +V(𝒙)\psi ,$$
(3.31)
with the nonlinearities
$`W(\rho ,\mathrm{\Sigma })={\displaystyle \frac{m}{2}}\left({\displaystyle \frac{\gamma (\rho )}{\rho }}1\right)\left({\displaystyle \frac{𝒋_0}{\rho }}\right)^2+mDf(\rho )\mathbf{}\left({\displaystyle \frac{𝒋__0}{\rho }}\right)+G(\rho ),`$ (3.32)
and
$`𝒲(\rho ,\mathrm{\Sigma })={\displaystyle \frac{\mathrm{}}{2\rho }}\mathbf{}\left\{[\gamma (\rho )\rho ]\left({\displaystyle \frac{𝒋_0}{\rho }}\right)\right\}+{\displaystyle \frac{\mathrm{}D}{2\rho }}\mathbf{}\left[f(\rho )\mathbf{}\rho \right].`$ (3.33)
Eqs. (3.32)-(3.33) differ from the one obtained in Ref. Scarfone7 where a family of NSE was derived in the stochastic quantization framework starting from the most general nonlinear classical kinetics compatible with constant diffusion coefficient $`D=\mathrm{}/2m`$. In particular, the real nonlinearity $`W`$ arising in the stochastic quantization is found to depend only on field $`\rho `$, in contrast with expression (3.32), where functional $`W`$ depends on both fields $`\rho `$ and $`\mathrm{\Sigma }`$.
Remarkably, we observe that when the kinetics of the system is governed by a linear drift, with $`\gamma (\rho )=\rho `$, the expression of nonlinear terms (3.32) and (3.33) simplify to
$$W(\rho ,\mathrm{\Sigma })=mD\stackrel{~}{f}(\rho )\mathbf{}\left(\frac{𝒋__0}{\rho }\right)+G(\rho ),$$
(3.34)
and
$$𝒲(\rho ,\mathrm{\Sigma })=\frac{\mathrm{}D}{2\rho }\mathbf{}\left[\stackrel{~}{f}(\rho )\mathrm{ln}\kappa (\rho )\mathbf{}\rho \right],$$
(3.35)
where $`\stackrel{~}{f}(\rho )=\rho (/\rho )\mathrm{ln}\kappa (\rho )`$.
They are determined only through functional $`\kappa (\rho )`$ which also defines the entropy (2.1) of the ancestor classical system.
## IV Ehrenfest relations and conserved quantities
In this section we study the time evolution of the most important physical observables of the system described by the Hamiltonian density (3.16) with the nonlinear potential (3.23): mass center, linear and angular momentum and total energy. The proofs are given in Appendix A.
Let us recall that, given an Hermitian operator $`𝒪=𝒪^{}`$ associated with a physical observable, its time evolution is given by
$$\frac{d}{dt}𝒪=\frac{i}{\mathrm{}}\left(\frac{\delta H}{\delta \psi }𝒪\psi \psi ^{}𝒪\frac{\delta H}{\delta \psi ^{}}\right)𝑑𝒙+\frac{𝒪}{t},$$
(4.1)
where the mean value $`𝒪=\psi ^{}𝒪\psi 𝑑𝒙`$. The last term in Eq. (4.1) takes into account a possible explicit time dependence on the operator $`𝒪`$.
Observing that the NSE (3.31) can be written in
$$i\mathrm{}\frac{\psi }{t}=\text{H}\psi ,$$
(4.2)
where
$$\text{H}=\frac{\mathrm{}^2}{2m}\mathrm{\Delta }+W(\rho ,\mathrm{\Sigma })+i𝒲(\rho ,\mathrm{\Sigma })+V(𝒙),$$
(4.3)
Eq. (4.1) assumes the equivalent expression
$$\frac{d}{dt}𝒪=\frac{i}{\mathrm{}}[\mathrm{Re}\text{H},𝒪]+\frac{1}{\mathrm{}}\{\mathrm{Im}\text{H},𝒪\}+\frac{𝒪}{t},$$
(4.4)
where $`[,]`$ and $`\{,\}`$ indicate the commutator and the anticommutator, respectively.
By choosing $`𝒪=𝒙`$, from Eq. (4.1) we obtain the first Ehrenfest relation for the time evolution of the mass center of the system
$$𝒗_{\mathrm{mc}}\frac{d}{dt}𝒙=\frac{\gamma (\rho )}{\rho }\widehat{𝒖}_{\mathrm{drift}}.$$
(4.5)
We observe that only drift nonlinearity appears in this equation whereas the diffusion term makes no contribution. Eq. (4.5) states that, quite generally, $`𝒗_{\mathrm{mc}}`$ is not a motion constant. This fact implies that the quantum system is not Galilei invariant. The origin of the nonconservation of $`𝒗_{\mathrm{cm}}`$ can be found in the difference between quantity $`𝒑_{\mathrm{mc}}=m𝒗_{\mathrm{mc}}`$ and the expectation value of the momentum operator $`𝒑i\mathrm{}\mathbf{}=\rho \mathbf{}\mathrm{\Sigma }d𝒙`$. These two quantities are equivalent only in the linear drift case. Differently from the former, the latter is in all cases conserved during the time evolution of the system, in absence of the external potential. This can be shown by means of the second Ehrenfest relation which follows from Eq. (4.1) by posing $`𝒪=i\mathrm{}\mathbf{}`$
$$\frac{d}{dt}𝒑=𝑭_{\mathrm{ext}}(𝒙).$$
(4.6)
The time evolution of the expectation value of momentum is governed only by external potential $`V(𝒙)`$. On the average, the KIP introduce no effect on the dynamics of the system. This is a consequence of the invariance of nonlinearity $`W[\rho ,\mathrm{\Sigma }]+i𝒲[\rho ,\mathrm{\Sigma }]`$ under uniform space translation.
In the same way, accounting for the invariance of nonlinearity for uniform rotations, the third Ehrenfest relation follows
$$\frac{d}{dt}𝑳=𝑴_{\mathrm{ext}}(𝒙),$$
(4.7)
where $`𝑴_{\mathrm{ext}}(𝒙)=𝒙\times 𝑭_{\mathrm{ext}}(𝒙)`$ is the momentum of the external force field. Eq. (4.7) is obtained from Eq. (4.1) after posing $`𝒪=𝒙\times (i\mathrm{}\mathbf{})`$. Again, the nonlinear terms introduced by KIP as well as nonlinearity $`G(\rho )`$ make no contribution, on the average, to angular momentum.
Finally, the last relation concerns the total energy of the system given by the Hamiltonian $`EH`$. By posing
$$𝒪=\frac{\mathrm{}^2}{2m}\mathrm{\Delta }+\frac{1}{\rho }U(\rho ,\mathrm{\Sigma })+V(𝒙),$$
(4.8)
we have $`𝒪E`$ and from Eq. (4.1) we obtain
$$\frac{dE}{dt}=0.$$
(4.9)
In conclusion, for a constant diffusion process we have shown that in absence of the external potential the system admits three constants of motion: total linear momentum $`𝒑`$, total angular momentum $`𝑳`$ and total energy $`E`$. Such conserved quantities, according to the Noether theorem, follow as a consequence of the invariance of the system under uniform space-time translation and uniform rotation. Moreover, the system is also invariant for global U(1) transformation which implies conservation of the normalization of field $`\psi `$ throughout the evolution of the system.
In Appendix B we briefly discuss the case of a quantum system with a diffusion coefficient $`D(t,𝒙)`$ that depends on time and position. This space-time dependence destroys the invariance of the system under uniform space-time translation and space rotation. As a consequence, all quantities $`𝐩`$, $`𝐋`$ and $`E`$ are no longer conserved, even for a vanishing external potential.
It should be remarked that the results discussed here, although very general in that they are independent of the form of nonlinearities $`W`$ and $`𝒲`$, are valid only for the class of the canonical systems. In literature there are many noncanonical NSEs, obtained starting from certain physically motivated conditions, which are worthy of being taken into account. For these equations, the expression of H appearing on the right hand side of the Schrödinger equation cannot be obtained from Eqs. (3.9) and (3.10) by means of a Hamiltonian function $`H=𝑑𝒙`$.
Despite this, even for these noncanonical systems the time evolution of the mean values of the quantum operators associated with the observables can be derived through Eq. (4.4), but what is important is that these operators can assume a different definition with respect to the one given in the canonical theory. For instance, in the canonical framework the energy is supplied by the Hamiltonian $`H`$ of the system, whereas in a noncanonical theory it is identified with the operator $`i\mathrm{}/t\text{H}`$. (We remark that in the canonical framework $`H`$ and H are, in general, different quantities). Moreover, for a noncanonical theory, conservation of the energy and the momentum do not follow merely from the principle of invariance of the system under space-time translation. Their time evolution depends on the expression of the nonlinearities appearing in the Schrödinger equation. All of this clearly causes a profound difference in the resulting Ehrenfest relations.
For instance, in Ref. Schuch1 a noncanonical Schrödinger equation with complex nonlinearity was derived starting from a Fokker-Planck equation for density field $`\rho `$ by assuming some physically justified separability conditions. The resulting evolution equation has the real and the imaginary nonlinearity given by $`W(\rho ,\mathrm{\Sigma })=\gamma (\mathrm{\Sigma }\mathrm{\Sigma })`$ and $`𝒲(\rho ,\mathrm{\Sigma })=(\mathrm{}D/2)\mathrm{\Delta }\rho /\rho `$, respectively, where $`\gamma `$ is a constant related to diffusion coefficient $`D`$ and such that $`D0`$ if $`\gamma 0`$. It is easy to see that such nonlinearities cannot be obtained starting from a nonlinear potential $`U(\rho ,\mathrm{\Sigma })`$ through Eqs. (3.17) and (3.18). The system described by this NSE turns out to be dumped and dissipative, even in presence of a constant diffusive process. In fact, it can be shown that, following Ref. Schuch1 , from Eq. (4.4) it follows $`d𝒑/dt=𝑭_{\mathrm{ext}}\gamma 𝒑`$ and $`dE/dtd\text{H}/dt=(\gamma /m)𝒑^2`$, which is a very different situation with respect to the one discussed in the present paper, with the exception of the trivial case $`\gamma =0`$.
## V Gauge equivalence
We introduce a nonlinear gauge transformation of the third kind Scarfone1
$$\psi \varphi =\psi \mathrm{exp}\left(\frac{i}{\mathrm{}}mD\mathrm{ln}\kappa (\rho )\right),$$
(5.1)
which, being a unitary transformation, does not change the amplitude of wave function $`|\psi |^2=|\varphi |^2=\rho `$, and transforms the phase $`\mathrm{\Sigma }`$ of the old field $`\psi `$, into phase $`\sigma `$ of the new field $`\varphi `$ according to the equation
$$\sigma =\mathrm{\Sigma }mD\mathrm{ln}\kappa (\rho ).$$
(5.2)
Consequently, the nonlinear current (3.26) takes the expression
$$𝒋\stackrel{~}{𝒋}=\frac{\mathbf{}\sigma }{m}\gamma (\rho ).$$
(5.3)
with only a nonlinear drift term.
Let us observe that, at the classical level, the similar transformation
$$𝒖_{\mathrm{drift}}^{}=𝒖_{\mathrm{drift}}D\mathbf{}\mathrm{ln}\kappa (\rho ),$$
(5.4)
changes total current $`𝑱𝑱^{}=𝒖_{\mathrm{drift}}^{}\gamma (\rho )`$ into another one consisting only of a nonlinear drift term.
Performing the transformation (5.1), Eq. (3.31) becomes
$$i\mathrm{}\frac{\varphi }{t}=\frac{\mathrm{}^2}{2m}\mathrm{\Delta }\varphi +\left[\stackrel{~}{W}(\rho ,\sigma )+i\stackrel{~}{𝒲}(\rho ,\sigma )\right]\varphi +V(𝒙)\varphi ,$$
(5.5)
where the new nonlinearities $`\stackrel{~}{W}(\rho ,\sigma )`$ and $`\stackrel{~}{𝒲}(\rho ,\sigma )`$ are given by
$`\stackrel{~}{W}(\rho ,\sigma )={\displaystyle \frac{m}{2}}\left({\displaystyle \frac{\gamma (\rho )}{\rho }}1\right)\left({\displaystyle \frac{\stackrel{~}{𝒋}_0}{\rho }}\right)^2+mD^2\left[f_1(\rho )\mathrm{\Delta }\rho +f_2(\rho )\left(\mathbf{}\rho \right)^2\right]+G(\rho ),`$
(5.6)
with $`\stackrel{~}{𝒋}_0=\rho \mathbf{}\sigma /m`$,
$`f_1(\rho )=\gamma (\rho )\left[{\displaystyle \frac{}{\rho }}\mathrm{ln}\kappa (\rho )\right]^2,`$ (5.7)
$`f_2(\rho )={\displaystyle \frac{1}{2}}{\displaystyle \frac{f_1(\rho )}{\rho }},`$ (5.8)
and
$$\stackrel{~}{𝒲}(\rho ,\mathrm{\Sigma })=\frac{\mathrm{}}{2\rho }\mathbf{}\left\{[\gamma (\rho )\rho ]\left(\frac{\stackrel{~}{𝒋}_0}{\rho }\right)\right\}.$$
(5.9)
Eq. (5.5) is still a NSE with a complex nonlinearity due to the presence of the nonlinear drift term in the quantum current expression (5.3).
Basically, both equations (3.31) and (5.5) are different NSEs describing the same physical system. This is a consequence of the unitary structure of the transformation (5.1) which implies that the probability position density for field $`\psi `$ and field $`\varphi `$ assumes the same value at any instant of time Doebner1 .
In the case of $`\gamma (\rho )=\rho `$ expressions (5.6) and (5.9) can be simplified and the NSE (5.5) assumes the form
$`i\mathrm{}{\displaystyle \frac{\varphi }{t}}=`$ $``$ $`{\displaystyle \frac{\mathrm{}^2}{2m}}\mathrm{\Delta }\varphi +mD^2\left[\stackrel{~}{f}_1(\rho )\mathrm{\Delta }\rho +\stackrel{~}{f}_2(\rho )\left(\mathbf{}\rho \right)^2\right]\varphi +G(\rho )\varphi +V(𝒙)\varphi ,`$
with
$`\stackrel{~}{f}_1(\rho )=\rho \left[{\displaystyle \frac{}{\rho }}\mathrm{ln}\kappa (\rho )\right]^2,`$ (5.11)
$`\stackrel{~}{f}_2(\rho )={\displaystyle \frac{1}{2}}{\displaystyle \frac{\stackrel{~}{f}_1(\rho )}{\rho }},`$ (5.12)
which contains a purely real nonlinearity depending only on field $`\rho `$.
We observe that although Eq. (5.1) transforms the nonlinear current into another one without the diffusive term, NSEs (5.5) and (LABEL:schroedinger3) contain a dependence from on diffusion coefficient $`D`$.
The NSE (5.5) is still canonical. It can be obtained from the following Hamiltonian density
$$(\rho ,\sigma )=\frac{(\mathbf{}\sigma )^2}{2m}\rho +\frac{\mathrm{}^2}{8m}\frac{\left(\mathbf{}\rho \right)^2}{\rho }+\widehat{U}(\rho ,\sigma )+V(𝒙)\rho ,$$
(5.13)
with nonlinear potential
$$\widehat{U}(\rho ,\sigma )=\frac{(\mathbf{}\sigma )^2}{2m}\left[\gamma (\rho )\rho \right]\frac{mD^2}{2}f_1(\rho )\left(\mathbf{}\rho \right)^2+\stackrel{~}{U}(\rho ).$$
(5.14)
In this sense Eq. (5.1) defines a canonical transformation.
In conclusion, let us make the following observation. Eq. (5.5) admits the following continuity equation
$$\frac{\rho }{t}+\mathbf{}\left[\frac{\mathbf{}\sigma }{m}\gamma (\rho )\right]=0.$$
(5.15)
A natural question is: what kind of NSE is obtained if we quantize a classical system obeying the continuity equation $`\rho /t+\mathbf{}𝑱^{}=0`$ with the method described above?
We easily have
$`i\mathrm{}{\displaystyle \frac{\varphi }{t}}=`$ $``$ $`{\displaystyle \frac{\mathrm{}^2}{2m}}\mathrm{\Delta }\varphi {\displaystyle \frac{m}{2}}\left({\displaystyle \frac{\gamma (\rho )}{\rho }}1\right)\left({\displaystyle \frac{\stackrel{~}{𝒋}_0}{\rho }}\right)^2\varphi `$ (5.16)
$``$ $`i{\displaystyle \frac{\mathrm{}}{2\rho }}\mathbf{}\left\{[\gamma (\rho )\rho ]\left({\displaystyle \frac{\stackrel{~}{𝒋}_0}{\rho }}\right)\right\}\varphi +G(\rho )\varphi +V(𝒙)\varphi ,`$
where now $`\rho `$ and $`\sigma `$ are independent fields representing the amplitude and phase of wave function $`\varphi `$. Eq. (5.16) can be derived through the Hamiltonian density (5.13) with nonlinear potential
$$\widehat{U}__1(\rho ,\sigma )=\frac{(\mathbf{}\sigma )^2}{2m}\left[\gamma (\rho )\rho \right]+\stackrel{~}{U}(\rho ).$$
(5.17)
Potentials (5.14) and (5.17) differ for the quantity
$$\overline{U}(\rho )=\widehat{U}(\rho ,\sigma )\widehat{U}__1(\rho ,\sigma )=\frac{mD^2}{2}f_1(\rho )\left(\mathbf{}\rho \right)^2,$$
(5.18)
which depends only on field $`\rho `$. This nonlinear potential $`\overline{U}(\rho )`$ does not affect the continuity equation and thus cannot be obtained starting directly from Eq. (5.15).
## VI Some examples
To illustrate the relevance and applicability of the theory described in the previous sections, we derive and discuss some different NSEs obtained starting from kinetic equations known in literature. In the following Section, for simplicity’s sake we omit the arbitrary nonlinear potential $`\stackrel{~}{U}(\rho )`$ and focus our attention only on the effect yield through the potential introduced by the KIP.
### VI.1 Boltzmann-Gibbs-entropy
It is well known that when the many body system is governed by short-range interactions, or when interaction energy is neglecting with respect to the total energy of the system, the suitable entropic functional is given by the BG-entropy
$$S_{\mathrm{BG}}(\rho )=\rho \mathrm{ln}\left(\rho \right)𝑑𝒙.$$
(6.1)
This entropy arises from Eq. (2.1) by posing $`\kappa (\rho )=e\rho `$ with $`a(\rho )=e\rho `$ and $`b(\rho )=1`$. It is readily seen that $`\gamma (\rho )=e\rho c(\rho )`$.
Among the many NFPEs compatible with entropy (6.1) we consider the simplest case of linear drift by posing $`c(\rho )=1/e`$. Then the continuity equation (3.25) becomes the standard linear Fokker-Planck equation
$$\frac{\rho }{t}+\mathbf{}\left(𝒋__0D\mathbf{}\rho \right)=0,$$
(6.2)
whereas the evolution equation for the quantum system is given by the following NSE
$`i\mathrm{}{\displaystyle \frac{\psi }{t}}=`$ $``$ $`{\displaystyle \frac{\mathrm{}^2}{2m}}\mathrm{\Delta }\psi +mD\mathbf{}\left({\displaystyle \frac{𝒋__0}{\rho }}\right)\psi +i{\displaystyle \frac{\mathrm{}}{2}}D{\displaystyle \frac{\mathrm{\Delta }\rho }{\rho }}\psi +V(𝒙)\psi ,`$ (6.3)
which is recognized as the canonical sub-family of the class of Doebner-Goldin equations parameterized by diffusion coefficient $`D`$. We recall that Eq. (6.2) was obtained in the quantum mechanics theory starting from the study of the physical interpretation of a certain family of diffeomorphismin group Doebner1 .
By performing gauge transformation (5.1), Eq. (6.3) becomes
$$i\mathrm{}\frac{\varphi }{t}=\frac{\mathrm{}^2}{2m}\mathrm{\Delta }\varphi +mD^2\left[\frac{\mathrm{\Delta }\rho }{\rho }\frac{1}{2}\left(\frac{\mathbf{}\rho }{\rho }\right)^2\right]\varphi +V(𝒙)\varphi ,$$
(6.4)
which was studied previously in Guerra . In particular, Eq. (6.4) is equivalent to the following linear Schrödinger equation
$$ik^{}\frac{\chi }{t}=\frac{k_{}^{}{}_{}{}^{2}}{2m}\mathrm{\Delta }\chi +V(𝒙)\chi ,$$
(6.5)
with $`k^{}=\mathrm{}\sqrt{1(2mD/\mathrm{})^2}`$ and field $`\chi `$ is related to hydrodynamic fields $`\rho `$ and $`\sigma `$ through the relation $`\chi =\rho ^{1/2}\mathrm{exp}(i\sigma /k^{})`$.
This appear to be an interesting result. By quantizing a classical system described by MB-entropy the standard linear Schrödinger equation was obtained. In this equation the nonlinear terms describing the interaction between the many particles of the quantum system are absent. This is in accordance with the general statement that MB-entropy is suitable for describing systems with no (or negligible) interaction among the particles.
### VI.2 Generalized entropies
In presence of long-range interactions or memory effects persistent in time, it has been argued that MB-entropy may not be appropriate in describing such systems. For this reason, many different versions of Eq. (6.1) have been proposed in literature.
Very recently, Ref. Scarfone9 ; Scarfone8 introduced a bi-parametric deformation of the logarithmic function
$$\mathrm{ln}_{_{\{\kappa ,r\}}}(x)=\frac{x^{r+\kappa }x^{r\kappa }}{2\kappa },$$
(6.6)
which reduces, in the $`(\kappa ,r)(0,\mathrm{\hspace{0.17em}0})`$ limit, to the standard logarithm: $`\mathrm{ln}_{_{\{0,0\}}}(x)=\mathrm{ln}x`$. By replacing the logarithmic function in Eq. (6.1) with its generalized version (6.6), we obtain a bi-parametric family of generalized entropies
$$S_{_{\{\kappa ,r\}}}(\rho )=\rho \mathrm{ln}_{_{\{\kappa ,r\}}}\left(\rho \right)𝑑𝒙,$$
(6.7)
introduced, for the first time, in Refs. Mittal ; Sharma . Remarkably, this family of entropies includes, as special cases, some generalized entropies, well known in literature, used in the study of systems exhibiting distribution with asymptotic power law behavior. Among them we can cite Tsallis-entropy Tsallis which follows by posing $`r=\pm |\kappa |`$
$$S__q(\rho )=\frac{\rho ^q\rho }{1q}𝑑𝒙,$$
(6.8)
with $`q=1\pm 2|\kappa |`$ and Kaniadakis-entropy Kaniadakis1 , for $`r=0`$
$$S_{_{\{\kappa \}}}(\rho )=\frac{\rho ^{1+\kappa }\rho ^{1\kappa }}{2\kappa }𝑑𝒙.$$
(6.9)
Both these entropies, as well as other one-parameter deformed entropies, originated from Eq. (6.7) Scarfone8 , can be employed to describe generalized statistical systems such as, for instance, charge particles in electric and magnetic fields Rossani , 2d-turbulence in pure-electron plasma Boghosian , Bremsstrahlung Souza and anomalous diffusion of the correlated and Lévy type Borland ; Zanette .
In addition to the many applications where Tsallis-entropy has been employed Tsallisbiblio , Kaniadakis-entropy (6.9) has been successfully applied in the description of the energy distribution of fluxes of cosmic rays Kaniadakis1 , whereas the entropy in (6.7) with $`\kappa ^2=(r+1)^21`$ has been applied in the generalized statistical mechanical study of $`q`$-deformed oscillators in the frame-work of quantum-groups Abe2 .
Despite the topics recalled above, there is currently great interest in studying quantum systems with long-range microscopic interactions. Systems such as quantum wires, which are now possible in practice thanks to recent technological advances, require on the theoretical ground, the development of a quantum (nonlinear) theory capable of capturing the emergent facts Nazareno .
The entropy in (6.7) arises from Eq. (2.1) by posing
$$\mathrm{ln}\kappa (\rho )=\lambda \mathrm{ln}_{_{\{\kappa ,r\}}}\left(\frac{\rho }{\alpha }\right),$$
(6.10)
with $`\lambda =(1+r\kappa )^{(r+\kappa )/2\kappa }/(1+r+\kappa )^{(r\kappa )/2\kappa }`$ and $`\alpha =[(1+r\kappa )/(1+r+\kappa )]^{1/2\kappa }`$.
Among the many different possibilities, we discuss the case of linear drift with $`\gamma (\rho )=\rho `$. By taking into account Eq. (6.10) we have continuity equation (3.25) with
$$f(\rho )=a_+\rho ^{r+\kappa }a_{}\rho ^{r\kappa },$$
(6.11)
where $`a_\pm =(r\pm \kappa )(1+r\pm \kappa )/2\kappa `$ are constants.
The associated NSE assumes the expression
$`i\mathrm{}{\displaystyle \frac{\varphi }{t}}=`$ $``$ $`{\displaystyle \frac{\mathrm{}^2}{2m}}\mathrm{\Delta }\varphi +mD^2{\displaystyle \frac{f(\rho )}{\rho }}\left[f(\rho )\mathrm{\Delta }\rho +\stackrel{~}{f}(\rho )\left(\mathbf{}\rho \right)^2\right]\varphi +V(𝒙)\varphi ,`$ (6.12)
with
$$\stackrel{~}{f}(\rho )=b_+\rho ^{r+\kappa 1}b_{}\rho ^{r\kappa 1},$$
(6.13)
and $`b_\pm =a_\pm (r\pm \kappa 1/2)`$.
Eq. (6.12) contains only a purely real nonlinearity and reduces to Eq. (6.4) in the $`(\kappa ,r)(0,\mathrm{\hspace{0.17em}0})`$ limit, as well as Eq. (6.7), which reduces to the standard BG-entropy.
In particular, for Tsallis-entropy, the continuity equation (3.25), with
$$f(\rho )=q\rho ^{q1},$$
(6.14)
becomes the diffusive NFPE Compte1 while the corresponding NSE is given through Eq. (6.12) with
$`\stackrel{~}{f}(\rho )=\left(q{\displaystyle \frac{3}{2}}\right)\rho ^{q2},`$ (6.15)
and reduces to Eq. (6.4) in the $`q1`$ limit just as entropy (6.8) reduces to BG-entropy.
We observe that in Refs. Olavo1 ; Olavo the quantization of a classical system described by Tsallis-entropy has been already discussed. There, a NLS compatible with the continuity equation $`\rho ^\mu /t+\mathbf{}(\rho ^\mu \widehat{𝒖}_{_{\mathrm{drift}}})=0`$ was obtained with a different approach. The nonlinearity appearing in the NLS of Refs. Olavo1 ; Olavo reduces, for $`\mu =1`$ and $`q2q`$, to the same one reported here.
On the other hand, for Kaniadakis-entropy, the continuity equation is given in Eq. (3.25) with
$$f(\rho )=\frac{1}{2}\left[(\kappa +1)\rho ^\kappa (\kappa 1)\rho ^\kappa \right],$$
(6.16)
which coincides with that proposed in Ref. Kaniadakis1 while the associated NSE is given in Eq. (6.12) with
$`\stackrel{~}{f}(\rho )={\displaystyle \frac{1}{2\rho }}\left[(\kappa +1)\left(\kappa {\displaystyle \frac{1}{2}}\right)\rho ^\kappa +(\kappa 1)\left(\kappa +{\displaystyle \frac{1}{2}}\right)\rho ^\kappa \right],`$ (6.17)
and reduces to Eq. (6.4) in the $`\kappa 0`$ limit just as entropy (6.9) reduces to BG-entropy.
### VI.3 Interpolating bosons-fermions-entropy
In Ref. Quarati1 , on the basis of the generalized exclusion-inclusion principle the authors introduced a family of NFPEs describing the evolution of a classical system of particles whose statistical behavior interpolates between bosonic and fermionic particles. The equilibrium distribution governed by the EIP can be obtained by maximizing the following entropy
$$S_{\mathrm{EIP}}(\rho )=\left[\rho \mathrm{ln}\rho \frac{1}{\kappa }(1+\kappa \rho )\mathrm{ln}(1+\kappa \rho )\right]𝑑𝒙,$$
(6.18)
with $`1\kappa 1`$. In particular, for $`\kappa =\pm 1`$ we recognize the well-known Bose-Einstein and Fermi-Dirac entropies, whereas intermediary behavior follows for $`1<\kappa <1`$. Entropy (6.18) can be obtained from Eq. (2.1) by posing $`a(\rho )=\rho `$ and $`b(\rho )=1+\kappa \rho `$.
Some examples of real physical systems where EIP can be usefully applied are to be found in the Bose-Einsten condensation. Typically, the cubic NSE is used to describe the behavior of the condensate by simulating in this way the statistical attraction between the many bodies constituting the system. In spite of the simplest cubic interaction, other interactions like the one introduced by the EIP can be adopted to simulate an attraction among the particles.
In the opposite direction, almost-fermionic systems can be found in the study of the motion of electrons and holes in a semiconductor. In fact, while if separately considered electrons and holes are fermions, together they constitute an excited state behaving differently from a fermion or a boson. The same argument can be applied to the Cooper-pair in the superconductivity theory. Such excitation differs from a pure boson state because of the spatial delocalization of the two electrons, which are not completely overlying. Deviation from Bose statistics must be taken into account.
In the following we discuss separately two different choices for functional $`\gamma (\rho )`$.
In the linear drift case, with $`c(\rho )=1/(1+\kappa \rho )`$, the evolution equation for field $`\rho `$ assumes the expression
$$\frac{\rho }{t}+\mathbf{}\left(𝒋__0D\frac{\mathbf{}\rho }{1+\kappa \rho }\right)=0,$$
(6.19)
which was proposed in Ref. Kaniadakis . By means of Eq. (5.1), nonlinear current $`𝒋__0D\mathbf{}\rho /(1+\kappa \rho )\stackrel{~}{𝒋}__0`$ assumes the standard bilinear form and the corresponding NSE follows from Eq. (LABEL:schroedinger3)
$`i\mathrm{}{\displaystyle \frac{\varphi }{t}}=`$ $``$ $`{\displaystyle \frac{\mathrm{}^2}{2m}}\mathrm{\Delta }\varphi +{\displaystyle \frac{mD^2}{(1+\kappa \rho )^2}}\left[{\displaystyle \frac{\mathrm{\Delta }\rho }{\rho }}{\displaystyle \frac{13\kappa \rho }{2(1+\kappa \rho )}}\left({\displaystyle \frac{\mathbf{}\rho }{\rho }}\right)^2\right]\varphi +V(𝒙)\varphi .`$ (6.20)
We can observe that in Eq. (6.20) the EIP is accounting through a diffusion process and its effect vanishes in the $`D0`$ limit where it reduces to the standard linear Schrödinger equation. Eq. (6.20) has a purely real nonlinearity depending only on field $`\rho `$.
In a different way, by making the choice $`c(\rho )=1`$, the continuity equation (3.25) becomes
$$\frac{\rho }{t}+\mathbf{}\left[𝒋__0(1+\kappa \rho )D\mathbf{}\rho \right]=0.$$
(6.21)
The gauge transformation changes nonlinear current $`𝒋𝒋__0(1+\kappa \rho )D\mathbf{}\rho \stackrel{~}{𝒋}\stackrel{~}{𝒋}__0(1+\kappa \rho )`$ containing only a nonlinear drift term and Eq. (6.21) reduces to
$$\frac{\rho }{t}+\mathbf{}\left[\stackrel{~}{𝒋}__0(1+\kappa \rho )\right]=0.$$
(6.22)
This equation was introduced at the classical level in Ref. Quarati1 and subsequently reconsidered at the quantum level in Ref. Scarfone3 . The NSE associated with Eq. (6.22) is given by
$`i\mathrm{}{\displaystyle \frac{\varphi }{t}}=`$ $``$ $`{\displaystyle \frac{\mathrm{}^2}{2m}}\mathrm{\Delta }\varphi +{\displaystyle \frac{mD^2}{1+\kappa \rho }}\left[{\displaystyle \frac{\mathrm{\Delta }\rho }{\rho }}{\displaystyle \frac{1+2\kappa \rho }{2(1+\kappa \rho )}}\left({\displaystyle \frac{\mathbf{}\rho }{\rho }}\right)^2\right]\varphi `$ (6.23)
$`+`$ $`\kappa {\displaystyle \frac{m}{\rho }}\left({\displaystyle \frac{\stackrel{~}{𝒋}}{1+\kappa \rho }}\right)^2\varphi i\kappa {\displaystyle \frac{\mathrm{}}{2\rho }}\mathbf{}\left({\displaystyle \frac{\stackrel{~}{𝒋}\rho }{1+\kappa \rho }}\right)\varphi +V(𝒙)\varphi .`$
We observe that Eq. (6.23) still has a complex nonlinearity due to the nonlinear structure of quantum current $`\stackrel{~}{𝒋}`$ and both the nonlinearities $`W`$ and $`𝒲`$ depend on fields $`\rho `$ and $`\sigma `$. Moreover, in Eq. (6.23), EIP is accounted through a nonlinear drift term and survives even in absence of a diffusion process $`(D0)`$.
Factor $`(1+\kappa \rho )`$ in nonlinear current $`\stackrel{~}{𝒋}`$ takes into account the EIP in the many particle system. In fact, transition probability (2.15) from site $`𝒙`$ to $`𝒚`$ is defined as $`\pi (t,𝒙𝒚)=r(t,𝒙,𝒙𝒚)\rho (t,𝒙)[1+\kappa \rho (t,𝒚)]`$. For $`\kappa 0`$ the EIP holds and parameter $`\kappa `$ quantifies to what extent particle kinetics is affected by the particle population of the arrival site.
If $`\kappa >0`$ the $`\pi (t,𝒙𝒚)`$ contains an inclusion principle. In fact, the population density at arrival point $`𝒚`$ stimulates the particle transition and therefore transition probability increases linearly with $`\rho (t,𝒚)`$. Where $`\kappa <0`$ the $`\pi (t,𝒙𝒚)`$ takes into account the Pauli exclusion principle. If the arrival point $`𝒚`$ is empty $`\rho (t,𝒚)=0`$, the $`\pi (t,𝒙𝒚)`$ depends only on the population of the starting point. If arrival site is populated $`0<\rho (t,𝒚)\rho _{max}`$, the transition is inhibited. The range of values that parameter $`\kappa `$ can assume is limited by the condition that $`\pi (t,𝒙𝒚)`$ be real and positive as $`r(t,𝒙,𝒙𝒚)`$. We may conclude that $`\kappa 1/\rho _{max}`$.
A physical meaning of parameter $`\kappa `$ can be supplied by the following considerations. We recall that Bose-Einstein and Fermi-Dirac statistics originate from the fundamental principle of indistinguishability in quantum mechanics which is closely related to the symmetrization of the wave function. Completely symmetric wave functions are used to describe bosons while fermions are described by completely anti-symmetric wave functions. Thus, intermediate statistics arise in presence of incomplete symmetrization or anti-symmetrization of the wave function and the concept of degree of symmetrization or degree of anti-symmetrization has been introduced Quarati1 . Parameter $`\kappa `$ has the meaning of degree of indistinguishability of fermions or bosons, corresponding to the degree of symmetrization or anti-symmetrization, respectively. Value $`\kappa =1`$ corresponds to the case of fermions and denotes a complete anti-symmetric wave function whereas value $`\kappa =1`$ corresponds to the case of bosons and denotes a complete symmetric wave function. In addition, value $`\kappa =0`$ is associated with classical MB statistics and all the intermediate cases arise when $`\kappa `$ assumes all the values between $`1`$ and $`1`$.
Eq. (6.23), for $`D=0`$, was obtained previously in Ref. Scarfone3 , where the canonical quantization of the classical system obeying EIP was accounted for. As discussed in Section V, Eq. (6.23) differs from the NSE obtained in Scarfone3 for a real nonlinearity originated from nonlinear potential $`\stackrel{~}{U}(\rho )=mD^2(\mathbf{}\rho )^2/\rho (1+\kappa \rho )`$ and depending only on field $`\rho `$.
Finally, we observe that different from Eq. (6.20), Eq. (6.23) has vorticity different from zero. The Clebsh potentials corresponding to current $`\stackrel{~}{𝒋}=(\mathbf{}\sigma /m)\rho (1+\kappa \rho )`$ are given by $`\nu =1+\kappa \rho `$, $`\lambda =\sigma `$ and $`\mu =const`$ and vorticity assumes the expression
$$𝝎=\frac{\kappa }{m}\mathbf{}\rho \times \mathbf{}\sigma .$$
(6.24)
In Ref. Scarfone5 ; Scarfone6 localized, static, fermion-like vortex solutions ($`\kappa <0`$) were obtained and studied starting from Eq. (6.23) with $`D=0`$. We observe that in Scarfone5 ; Scarfone6 a different definition of the Clebsh potentials corresponding to $`\mu =\lambda =\sigma `$ and $`\nu =\kappa \rho `$ was adopted. Despite this, vorticity assumes the same expression that is given by Eq. (6.24) in both cases.
EIP vortex solutions are important on the theoretical ground and for interpretation of experimental results of several applications. For instance, they can be employed in the study of fermion-like vortices observed in <sup>3</sup>He-A superfluidity or in heavy fermion superconductors UPt<sub>3</sub> and U<sub>0.97</sub>Th<sub>0.03</sub>Be<sub>13</sub> Williams ; Matthews ; Madison .
## VII Conclusions
We have presented the quantization of a classical system of interacting particles obeying a kinetic interaction principle. The KIP both fixes the expression of the Fokker-Planck equation describing the kinetic evolution of the system and imposes the form of its entropy.
In the framework of canonical quantization, we have introduced a class of NSEs with complex nonlinearity obtained from the classical system obeying KIP. The form of nonlinearity $`\mathrm{\Lambda }(\psi ^{},\psi )`$ is determined by functional $`\kappa (\rho )`$, which also fixes the form of the entropy of ancestor classical system.
Among the many interesting solutions of the family of NSEs (3.31) we observe that for a free system with $`V(𝒙)=0`$, and posing $`G(\rho )=0`$, the planar wave
$$\psi (t,𝒙)=A\mathrm{exp}\left(\frac{i}{\mathrm{}}(\omega t𝒌𝒙)\right),$$
(7.1)
with constant amplitude $`A=const`$ is the simplest solution, where the relationship between $`\omega `$ and $`𝒌`$ is given by
$$\omega =\frac{\mathrm{}^2𝒌^2}{2m}\frac{\gamma (\rho )}{\rho }|_{\rho =A^2},$$
(7.2)
and reduces to the standard dispersion relation for $`\gamma (\rho )=\rho `$.
When the quantum system is in a stationary state such that $`\rho _\mathrm{s}/t=0`$, the relationships between distribution $`\rho _\mathrm{s}`$ and phase $`\mathrm{\Sigma }_\mathrm{s}`$ follow from Eq. (3.25)
$$\rho __\mathrm{s}=\kappa ^1\left(\mathrm{exp}\left(\frac{\mathrm{\Sigma }_\mathrm{s}(𝒙)}{mD}\beta ^{}\right)\right),$$
(7.3)
which mimics the classical equilibrium distribution (2.12), as can be seen by replacing $`\mathrm{\Sigma }_\mathrm{s}(𝒙)/mD`$ with $`\beta (𝒙)`$. Despite this, we stress that such an analogy is purely formal. The equivalence between Eqs. (2.12) and (7.3) requires that the following relation $`\mathrm{\Sigma }_\mathrm{s}(𝒙)/mD=\beta (𝒙)`$ must hold. In the general case the expression of stationary phase $`\mathrm{\Sigma }_\mathrm{s}(𝒙)`$ must be obtained from Eq. (III.2), after posing $`\mathrm{\Sigma }_\mathrm{s}/t=0`$, with $`\rho `$ given through Eq. (7.3).
Finally, another interesting class of possible solutions are solitons. It is well know that soliton solutions in NSE arise when the dispersive effects, principally due to term $`(\mathrm{}^2/2m)\mathrm{\Delta }\psi `$, is exactly balanced by the nonlinear terms. The existence of this class of solutions depends on the particular form of functionals $`\gamma (\rho )`$ and $`\kappa (\rho )`$ which fix the expression of nonlinearities $`W(\rho ,\mathrm{\Sigma })`$ and $`𝒲(\rho ,\mathrm{\Sigma })`$. A special situation, where soliton solutions are found within the NSEs derived in this paper, is given by the EIP-equation (6.23) with $`D=0`$ Scarfone3 where $`\gamma (\rho )=\rho (1+\kappa \rho )`$ and $`\kappa (\rho )=\rho /(1+\kappa \rho )`$.
The study of soliton solutions for other functional choices of $`\gamma (\rho )`$ and $`\kappa (\rho )`$, like, for instance, the ones related to the generalized entropies discussed in Section VI-B, is a very important task which deserves further research. These solutions may lead to practical applications. In fact, in recent years there has been great interest in the formulation of models where solitons can interact with a long-range force Gaididei . Typical nonlinear models supporting solitons, like the sine-Gordon model, arise from short-range forces. However, there is experimental evidence that most real transfer mechanisms have long-range interaction, as noted in condensed matter theory Scott or in spin glasses Ford .
## APPENDIX A
We present proof of the Ehrenfest equations discussed in Section IV. In the following we assume uniform boundary conditions on the fields in order to neglect the surface terms.
Let us rewrite Eq. (4.1) in a more suitable form. Accounting for the relation
$$\frac{\delta }{\delta \psi }=\psi ^{}\left(\frac{\delta }{\delta \rho }\frac{i\mathrm{}}{2\rho }\frac{\delta }{\delta \mathrm{\Sigma }}\right),$$
(A.1)
Eq. (4.1) becomes
$$\frac{d}{dt}𝒪=\frac{i}{\mathrm{}}[\frac{\delta H}{\delta \rho },𝒪]+\frac{1}{2}\{\frac{1}{\rho }\frac{\delta H}{\delta \mathrm{\Sigma }},𝒪\}+\frac{𝒪}{t}.$$
(A.2)
Eq. (4.5) can be obtained starting from Eq. (A.2) by posing $`𝒪=𝒙`$
$`{\displaystyle \frac{d}{dt}}𝒙`$ $`=`$ $`{\displaystyle \frac{i}{\mathrm{}}}{\displaystyle \left[\psi ^{}\frac{\delta H}{\delta \rho }𝒙\psi \psi ^{}𝒙\frac{\delta H}{\delta \rho }\psi \right]𝑑𝒙}+{\displaystyle \frac{1}{2}}{\displaystyle \left[\psi ^{}\frac{1}{\rho }\frac{\delta H}{\delta \mathrm{\Sigma }}𝒙\psi +\psi ^{}𝒙\frac{1}{\rho }\frac{\delta H}{\delta \mathrm{\Sigma }}\psi \right]𝑑𝒙}`$ (A.3)
$`=`$ $`{\displaystyle 𝒙\frac{\delta H}{\delta \mathrm{\Sigma }}𝑑𝒙}`$
$`=`$ $`{\displaystyle 𝒙\mathbf{}\left[\frac{\mathbf{}\mathrm{\Sigma }}{m}\gamma (\rho )Df(\rho )\mathbf{}\rho \right]𝑑𝒙}`$
$`=`$ $`{\displaystyle \left[\frac{\mathbf{}\mathrm{\Sigma }}{m}\gamma (\rho )Df(\rho )\mathbf{}\rho \right]𝑑𝒙}`$
$`=`$ $`{\displaystyle \frac{\mathbf{}\mathrm{\Sigma }}{m}\gamma (\rho )𝑑𝒙}D{\displaystyle \mathbf{}F(\rho )𝑑𝒙}`$
$`=`$ $`{\displaystyle \frac{\gamma (\rho )}{\rho }}\widehat{𝒖}_{\mathrm{drift}},`$
where
$$F(\rho )=^\rho f(\rho ^{})𝑑\rho ^{}.$$
(A.4)
To show the validity of Eq. (4.6) we pose $`𝒪=i\mathrm{}\mathbf{}`$ in Eq. (A.2) so that
$`{\displaystyle \frac{d}{dt}}𝒑`$ $`=`$ $`{\displaystyle \left[\psi ^{}\frac{\delta H}{\delta \rho }\mathbf{}\psi \psi ^{}\mathbf{}\left(\frac{\delta H}{\delta \rho }\psi \right)\right]𝑑𝒙}`$ (A.5)
$``$ $`i{\displaystyle \frac{\mathrm{}}{2}}{\displaystyle \left[\psi ^{}\frac{1}{\rho }\frac{\delta H}{\delta \mathrm{\Sigma }}\mathbf{}\psi +\psi ^{}\mathbf{}\left(\frac{1}{\rho }\frac{\delta H}{\delta \mathrm{\Sigma }}\psi \right)\right]𝑑𝒙}`$
$`=`$ $`{\displaystyle \frac{\delta H}{\delta \rho }\left(\psi ^{}\mathbf{}\psi +\psi \mathbf{}\psi ^{}\right)}i{\displaystyle \frac{\mathrm{}}{2}}{\displaystyle \frac{1}{\rho }\frac{\delta H}{\delta \mathrm{\Sigma }}\left(\psi ^{}\mathbf{}\psi \psi \mathbf{}\psi ^{}\right)𝑑𝒙}`$
$`=`$ $`{\displaystyle \left(\frac{\delta H}{\delta \rho }\mathbf{}\rho +\frac{\delta H}{\delta \mathrm{\Sigma }}\mathbf{}\mathrm{\Sigma }\right)𝑑𝒙},`$
where an integration by parts has been performed, and we have posed
$`\psi ^{}\mathbf{}\psi +\psi \mathbf{}\psi ^{}=\mathbf{}\rho ,`$ (A.6)
$`\psi ^{}\mathbf{}\psi \psi \mathbf{}\psi ^{}=i{\displaystyle \frac{2}{\mathrm{}}}\rho \mathbf{}\mathrm{\Sigma }.`$ (A.7)
Taking into account the relation
$`\mathbf{}={\displaystyle \frac{\delta H}{\delta \rho }}\mathbf{}\rho +{\displaystyle \frac{\delta H}{\delta \mathrm{\Sigma }}}\mathbf{}\mathrm{\Sigma }+\rho \mathbf{}V(𝒙),`$ (A.8)
from Eq. (A.5) it follows
$`{\displaystyle \frac{d}{dt}}𝒑`$ $`=`$ $`{\displaystyle \mathbf{}d𝒙}{\displaystyle \rho \mathbf{}V(𝒙)𝑑𝒙}`$ (A.9)
$`=`$ $`𝑭_{\mathrm{ext}}(𝒙),`$
Eq. (4.7) can easily be obtained following the same steps used in the proof of Eq. (4.6).
Finally, by posing $`𝒪=(\mathrm{}^2/2m)\mathrm{\Delta }+U(\rho ,\mathrm{\Sigma })/\rho +V(𝒙)`$ in Eq. (A.2), where $`U(\rho ,\mathrm{\Sigma })`$ is given in Eq. (3.23), we have
$`{\displaystyle \frac{dE}{dt}}`$ $`=`$ $`{\displaystyle \frac{i}{\mathrm{}}}{\displaystyle \left\{\psi ^{}\frac{\delta H}{\delta \rho }\left(\frac{\mathrm{}^2}{2m}\mathrm{\Delta }+\frac{U}{\rho }+V\right)\psi \psi ^{}\left[\left(\frac{\mathrm{}^2}{2m}\mathrm{\Delta }+\frac{U}{\rho }+V\right)\frac{\delta H}{\delta \rho }\psi \right]\right\}𝑑𝒙}`$ (A.10)
$`+`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \left\{\psi ^{}\frac{1}{\rho }\frac{\delta H}{\delta \mathrm{\Sigma }}\left(\frac{\mathrm{}^2}{2m}\mathrm{\Delta }+\frac{U}{\rho }+V\right)\psi +\psi ^{}\left[\left(\frac{\mathrm{}^2}{2m}\mathrm{\Delta }+\frac{U}{\rho }+V\right)\frac{1}{\rho }\frac{\delta H}{\delta \rho }\psi \right]\right\}𝑑𝒙}`$
$`+`$ $`{\displaystyle \rho \frac{}{t}\left(\frac{U}{\rho }+V\right)𝑑𝒙}`$
$`=`$ $`{\displaystyle \frac{i\mathrm{}}{2m}}{\displaystyle \left[\psi ^{}\frac{\delta H}{\delta \rho }\mathrm{\Delta }\psi \psi ^{}\mathrm{\Delta }\left(\frac{\delta H}{\delta \rho }\psi \right)\right]𝑑𝒙}`$
$`+`$ $`{\displaystyle \frac{i}{\mathrm{}}}{\displaystyle \left[\psi ^{}\frac{\delta H}{\delta \rho }\left(\frac{U}{\rho }+V\right)\psi \psi ^{}\left(\frac{U}{\rho }+V\right)\frac{\delta H}{\delta \rho }\psi \right]𝑑𝒙}`$
$``$ $`{\displaystyle \frac{\mathrm{}^2}{4m}}{\displaystyle \left[\psi ^{}\frac{1}{\rho }\frac{\delta H}{\delta \mathrm{\Sigma }}\mathrm{\Delta }\psi +\psi ^{}\mathrm{\Delta }\left(\frac{1}{\rho }\frac{\delta H}{\delta \mathrm{\Sigma }}\psi \right)\right]𝑑𝒙}`$
$`+`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \left[\psi ^{}\frac{1}{\rho }\frac{\delta H}{\delta \mathrm{\Sigma }}\left(\frac{U}{\rho }+V\right)\psi +\psi ^{}\left(\frac{U}{\rho }+V\right)\frac{1}{\rho }\frac{\delta H}{\delta \mathrm{\Sigma }}\psi \right]𝑑𝒙}+{\displaystyle \rho \frac{}{t}\left(\frac{U}{\rho }\right)𝑑𝒙}`$
$`=`$ $`{\displaystyle \frac{i\mathrm{}}{2m}}{\displaystyle \frac{\delta H}{\delta \rho }\left(\psi ^{}\mathrm{\Delta }\psi \psi \mathrm{\Delta }\psi ^{}\right)𝑑𝒙}{\displaystyle \frac{\mathrm{}^2}{4m}}{\displaystyle \frac{1}{\rho }\frac{\delta H}{\delta \mathrm{\Sigma }}\left(\psi ^{}\mathrm{\Delta }\psi +\psi \mathrm{\Delta }\psi ^{}\right)𝑑𝒙}`$
$`+`$ $`{\displaystyle \left[\frac{\delta H}{\delta \mathrm{\Sigma }}\left(\frac{U}{\rho }+V\right)+\frac{U}{t}\frac{U}{\rho }\frac{\rho }{t}\right]𝑑𝒙},`$
where a double integration by parts has been performed. Taking into account
$`{\displaystyle \frac{i\mathrm{}}{2m}}\left(\psi ^{}\mathrm{\Delta }\psi \psi \mathrm{\Delta }\psi ^{}\right)=\mathbf{}\left({\displaystyle \frac{\mathbf{}\mathrm{\Sigma }}{m}}\rho \right),`$ (A.11)
$`\psi ^{}\mathrm{\Delta }\psi +\psi \mathrm{\Delta }\psi ^{}=2\rho \left[{\displaystyle \frac{\mathrm{\Delta }\sqrt{\rho }}{\sqrt{\rho }}}\left({\displaystyle \frac{\mathbf{}\mathrm{\Sigma }}{\mathrm{}}}\right)^2\right],`$ (A.12)
which follow from Eq. (3.2), and the relation
$$\frac{U}{t}=\left(\frac{\delta }{\delta \rho }U𝑑𝒙\right)\frac{\rho }{t}+\left(\frac{\delta }{\delta \mathrm{\Sigma }}U𝑑𝒙\right)\frac{\mathrm{\Sigma }}{t},$$
(A.13)
Eq. (A.10) becomes
$`{\displaystyle \frac{dE}{dt}}`$ $`=`$ $`{\displaystyle \left\{\frac{\delta H}{\delta \rho }\mathbf{}\left(\frac{\mathbf{}\mathrm{\Sigma }}{m}\rho \right)\frac{\mathrm{}^2}{2m}\frac{\delta H}{\delta \mathrm{\Sigma }}\left[\frac{\mathrm{\Delta }\sqrt{\rho }}{\sqrt{\rho }}\left(\frac{\mathbf{}\mathrm{\Sigma }}{\mathrm{}}\right)^2\right]\right\}𝑑𝒙}`$
$`+`$ $`{\displaystyle \left[\frac{\delta H}{\delta \mathrm{\Sigma }}\left(\frac{U}{\rho }+V\right)+\left(\frac{\delta }{\delta \rho }U𝑑𝒙\frac{U}{\rho }\right)\frac{\rho }{t}+\left(\frac{\delta }{\delta \mathrm{\Sigma }}U𝑑𝒙\right)\frac{\mathrm{\Sigma }}{t}\right]𝑑𝒙}.`$
By using the relations
$`{\displaystyle \frac{\mathrm{}^2}{2m}}\left[{\displaystyle \frac{\mathrm{\Delta }\sqrt{\rho }}{\sqrt{\rho }}}\left({\displaystyle \frac{\mathbf{}\mathrm{\Sigma }}{\mathrm{}}}\right)^2\right]={\displaystyle \frac{\delta }{\delta \rho }}{\displaystyle U𝑑𝒙}{\displaystyle \frac{\delta H}{\delta \rho }}+V,`$ (A.15)
$`\mathbf{}\left({\displaystyle \frac{\mathbf{}\mathrm{\Sigma }}{m}}\rho \right)={\displaystyle \frac{\delta }{\delta \mathrm{\Sigma }}}{\displaystyle U𝑑𝒙}{\displaystyle \frac{\delta H}{\delta \mathrm{\Sigma }}},`$ (A.16)
which follow from Eqs. (3.11), (3.16) and (3.23), and motion equations (3.9) and (3.10), we obtain
$`{\displaystyle \frac{dE}{dt}}`$ $`=`$ $`{\displaystyle \left[\frac{\delta H}{\delta \rho }\left(\frac{\delta }{\delta \mathrm{\Sigma }}U𝑑𝒙\frac{\delta H}{\delta \mathrm{\Sigma }}\right)\frac{\delta H}{\delta \mathrm{\Sigma }}\left(\frac{\delta }{\delta \rho }U𝑑𝒙\frac{\delta H}{\delta \rho }+V\right)\right]𝑑𝒙}`$ (A.17)
$`+`$ $`{\displaystyle \left[\frac{\delta H}{\delta \mathrm{\Sigma }}\left(\frac{U}{\rho }+V\right)+\frac{\delta H}{\delta \mathrm{\Sigma }}\left(\frac{\delta }{\delta \rho }U𝑑𝒙\frac{U}{\rho }\right)\frac{\delta H}{\delta \rho }\left(\frac{\delta }{\delta \mathrm{\Sigma }}U𝑑𝒙\right)\right]𝑑𝒙}`$
$`=`$ $`0.`$
## APPENDIX B
We briefly discuss the generalization of the theory for quantum systems obeying the KIP and undergoing a diffusive process with a diffusion coefficient $`D(t,𝒙)`$ depending both on time and space position.
Given the following Hamiltonian density
$`(\rho ,\mathrm{\Sigma })={\displaystyle \frac{(\mathbf{}\mathrm{\Sigma })^2}{2m}}\gamma (\rho )+{\displaystyle \frac{\mathrm{}^2}{8m}}{\displaystyle \frac{\left(\mathbf{}\rho \right)^2}{\rho }}D(t,𝒙)\gamma (\rho )\mathbf{}\mathrm{ln}\kappa (\rho )\mathbf{}\mathrm{\Sigma }+\stackrel{~}{U}(\rho )+V(𝒙)\rho ,`$
(B.1)
from the Hamilton equations (3.9)-(3.10) we obtain the NSE
$$i\mathrm{}\frac{\psi }{t}=\frac{\mathrm{}^2}{2m}\mathrm{\Delta }\psi +\left[W(\rho ,\mathrm{\Sigma })+i𝒲(\rho ,\mathrm{\Sigma })\right]\psi +G(\rho )\psi +V(𝒙)\mathrm{\Psi },$$
(B.2)
with nonlinearities
$`W(\rho ,\mathrm{\Sigma })={\displaystyle \frac{m}{2}}\left({\displaystyle \frac{\gamma (\rho )}{\rho }}1\right)\left({\displaystyle \frac{𝒋_0}{\rho }}\right)^2+m\gamma (\rho ){\displaystyle \frac{}{\rho }}\mathrm{ln}\kappa (\rho )\mathbf{}\left(D(t,𝒙){\displaystyle \frac{𝒋__0}{\rho }}\right)+G(\rho ),`$ (B.3)
and
$`𝒲(\rho ,\mathrm{\Sigma })={\displaystyle \frac{\mathrm{}}{2m\rho }}\mathbf{}\left\{[\gamma (\rho )\rho ]\mathbf{}\mathrm{\Sigma }\right\}+{\displaystyle \frac{1}{2\rho }}\mathbf{}\left[D(t,𝒙)\gamma (\rho )\mathbf{}\mathrm{ln}\kappa (\rho )\right].`$ (B.4)
The system described by Hamiltonian (B.1) is dissipative since $`dE/dt0`$. This is a consequence of the time dependence of $`D`$ which breaks the invariance of Eq. (B.1) under uniform time translation. In the same way, linear momentum as well as angular momentum are no longer conserved, even in absence of the external potential, as a consequence of the position dependence of $`D`$ which breaks the invariance of Eq. (B.1) under uniform space translation and uniform space rotation. This can also be seen from the Ehrenfest relations
$`{\displaystyle \frac{d}{dt}}𝒙={\displaystyle \frac{\gamma (\rho )}{\rho }}\widehat{𝒖}_{\mathrm{drift}}D(t,𝒙)f(\rho )\mathbf{}\mathrm{ln}\rho ,`$ (B.5)
$`{\displaystyle \frac{d}{dt}}𝒑=mA(\rho ,\mathrm{\Sigma })\mathbf{}D(t,𝒙)+𝑭_{\mathrm{ext}}(𝒙),`$ (B.6)
$`{\displaystyle \frac{d}{dt}}𝑳=mA(\rho ,\mathrm{\Sigma })\left(𝒙\times \mathbf{}D(t,𝒙)\right)+𝑴_{\mathrm{ext}}(𝒙),`$ (B.7)
$`{\displaystyle \frac{dE}{dt}}=mA(\rho ,\mathrm{\Sigma }){\displaystyle \frac{}{t}}D(t,𝒙),`$ (B.8)
where $`A(\rho ,\mathrm{\Sigma })=f(\rho )\mathbf{}\mathrm{ln}\rho \widehat{𝒖}_{\mathrm{drift}}`$.
Finally, the gauge transformation described in Section V cannot be performed, in general, when the diffusion coefficient has spatial dependence. In fact, the transformation in (5.1) is well defined only if the following condition is fulfilled
$$\mathbf{}\times \left[D(t,𝒙)\mathbf{}\mathrm{ln}\kappa (\rho )\right]=0,$$
(B.9)
as can be seen by applying the curl operator to both sides of equation
$$\mathbf{}\sigma =\mathbf{}\mathrm{\Sigma }mD(t,𝒙)\mathbf{}\mathrm{ln}\kappa (\rho ),$$
(B.10)
which follows from Eqs. (3.26) and (5.3). We remark that if the dynamics of the system evolves in one spatial dimension, Eq. (B.9) is trivially verified and the transformation in (5.1) can in all cases be accomplished.
|
warning/0506/cond-mat0506802.html
|
ar5iv
|
text
|
# Statistics of charge transfer in a tunnel junction coupled to an oscillator
## I Introduction
In recent years it has become possible to couple charge dynamics of electrons to vibrational modes of a nanostructure, and the new field of nanoelectromechanics has emerged.Ble04 Nanoelectromechanical devices are expected to lead to new technologies, such as ultra-small mass detection techniques,IliCraKry04 ; EkiHuaRou04 as well as stimulating fundamental studies of quantum phenomena in macroscopic systems.MarSimPen03 For example, experiments have probed high-frequency nanomechanical resonators in order to reach the quantum uncertainty limit.KnoCle03 ; LahBuuCam04 ; GaiZolBad05 ; CleAldDri02 Charge transfer by mechanical motion has been studied in experiments shuttling single electrons. SheGorJon05 ; ErbWeiZwe01 ; SchBli04 Nanomechanical resonators are investigated for use in quantum information processing.ArmBleSch02 ; GelCle05
The simplest possibility of detection and control of the vibrational degree of freedom in a nanomechanical system is at present via the coupling to a quantum point contact or a tunnel junction. Recent theoretical studies of nanoelectromechanical systems have therefore considered the coupling of a harmonic oscillator to a tunnel junction. The master equation for the oscillator was originally obtained by Mozyrsky and Martin,MozMar02 using a method limiting considerations to the zero-temperature case, and the influence of the coupled oscillator on the average current was obtained. Smirnov, Mourokh and Horing examined the non-equilibrium fluctuations of an oscillator coupled to a biased tunnel junction.SmiMouHor03 Clerk and Girvin derived the master equation for the oscillator, and considered the current noise power spectrum in the shot noise regime, studying both the dc and ac cases.CleGir04 Armour, Blencowe and Zhang considered the dynamics of a classical oscillator coupled to a single electron transistor,ArmBleZha04 and Armour the induced current noise.Arm04 Related models were studied in molecular electronics.MitAleMil04
Tunnel junctions functioning as position detectors or being monitored by a vibrational mode put emphasis on a description of the current properties of a tunnel junction in terms of its charge dynamics. Recently we considered a general many-body system coupled to a quantum object and considered their joint dynamics in the charge representation.RamSheWab04 This approach, based on a previously introduced charge projection technique,SheRam03 provides a quantum description of charge dynamics based directly on the density matrix for the system, and allows us to treat the number of particles in a given piece of material as a quantum degree of freedom, establishing thereby in proper quantum mechanical context the charge representation. The evolution of the coupled systems is described in terms of the charge specific density matrix for the quantum object, $`\widehat{\rho }_n(t)`$, i.e., the dynamics conditioned on the number $`n`$ of charges in a specified spatial region of the environment. When a many-body environment is coupled to another quantum object, the method allows evaluating at any moment in time the joint probability distribution describing the quantum state of the object and the number of charges in the chosen region of the many-body system. The charge specific density matrix description of the dynamics of a quantum object is therefore an optimal tool to study transport in nanostructures since in electrical measurements any information beyond the charge distribution is irrelevant. So far we have applied the method to charge counting in a tunnel junction coupled to a discrete quantum degree of freedom, viz. that of a two-level system, and shown that the charge state of the junction can function as a meter providing a projective measurement of the quantum state of the two-level system.RamSheWab04
In this paper we shall apply the charge projection method to the case where the quantum object coupled to the junction is a continuous degree of freedom. In particular, we shall concentrate on the properties of the current through the junction due to the coupling to the quantum object. The statistical properties of the current through the junction and its correlations with the dynamics of the quantum object coupled to it, shall be expressed through the charge specific density matrix. We shall illustrate the results for the case of a harmonic oscillator coupled to the tunnel junction. It should be noticed, that current experimental setups studying nanoelectromechanical systems are operated under conditions where temperature, oscillator excitation energy, and voltage bias across the junction are comparable.EkiHuaRou04 ; KnoCle03 ; LahBuuCam04 Furthermore, nanomechanical oscillators, such as a suspended beam, are in addition to the charge dynamics of the electrons in the junction invariably coupled to a thermal environment, say the substrate upon which the oscillator is mounted. We are thus considering the situation where a quantum object in addition to interacting with a heat bath is interacting with an environment out of equilibrium. Having the additional parameter, the voltage, characterizing the environment in non-equilibrium, gives rise to features not present for an object coupled to a many-body system in equilibrium. The presented approach is applicable in a broad region of temperatures and voltages of the junction and arbitrary frequency of the oscillator and thus generalizes previous treatments.
The paper is organized as follows: In Sec. II we introduce the model Hamiltonian for a generic electromechanical nanoresonator, a harmonic oscillator coupled to a tunnel junction, and derive the Markovian master equation for the charge specific density matrix. The master equation for the charge unconditional density matrix, i.e., the charge specific density matrix traced with respect to the charge degree of freedom of the junction, is discussed in detail for the case of a harmonic oscillator coupled to a tunnel junction. In Sec. III we consider the influence of the oscillator on the current-voltage characteristic of the junction. In Sec. IV we consider the properties of the stationary state of the oscillator. We calculate the heating of the oscillator due to the nonequilibrium state of the junction, and calculate the steady state I-V characteristic of the junction. In Sec. V we consider the current noise in the junction using the charge representation, and obtain the explicit expression for the current-current correlator in the Markovian approximation. In Sec. VI, the theory is then applied to the case of an oscillator influencing the current noise of the junction. Finally, in Sec. VII we summarize and conclude. Details of calculations are presented in appendices.
## II Master equation
As a model of a nanoelectromechanical system we consider a harmonic oscillator coupled to a tunnel junction. The transparency of the tunnel barrier is assumed perturbed by the displacement, $`x`$, of the oscillator. The resulting Hamiltonian is
$$\widehat{H}=\widehat{H}_0+H_l+H_r+\widehat{H}_T$$
(1)
where $`\widehat{H}_0`$ is the Hamiltonian for the isolated harmonic oscillator with bare frequency $`\omega _B`$ and mass $`m`$. A hat marks operators acting on the oscillator degree of freedom. The Hamiltonians $`H_{l,r}`$ specify the isolated left and right electrodes of the junction
$$H_l=\underset{𝐥}{}\epsilon _𝐥c_𝐥^{}c_𝐥,H_r=\underset{𝐫}{}\epsilon _𝐫c_𝐫^{}c_𝐫$$
(2)
where $`𝐥,𝐫`$ labels the quantum numbers of the single particle energy eigenstates in the left and right electrodes, respectively, with corresponding energies $`\epsilon _{𝐥,𝐫}`$. The operator $`\widehat{H}_T`$ describes the tunnelling,
$$\widehat{H}_T=\widehat{𝒯}+\widehat{𝒯}^{},\widehat{𝒯}=\underset{𝐥,𝐫}{}\widehat{T}_{\mathrm{𝐥𝐫}}c_𝐥^{}c_𝐫$$
(3)
with the tunneling amplitudes, $`\widehat{T}_{\mathrm{𝐥𝐫}}=\widehat{T}_{\mathrm{𝐫𝐥}}^{}`$, depending on the oscillator degree of freedom. Due to the coupling, the tunnelling amplitudes and thereby the conductance of the tunnel junction depend on the state of the oscillator. In the following we assume a linear coupling between the oscillator position and the tunnel junction
$$\widehat{T}_{\mathrm{𝐥𝐫}}=v_{\mathrm{𝐥𝐫}}+w_{\mathrm{𝐥𝐫}}\widehat{x}$$
(4)
where $`v_{\mathrm{𝐥𝐫}}=v_{\mathrm{𝐫𝐥}}^{}`$ is the unperturbed tunneling amplitude and $`w_{\mathrm{𝐥𝐫}}=w_{\mathrm{𝐫𝐥}}^{}`$ its derivative with respect to the position of the oscillator. The derivation of the equation of motion for the charge specific density matrix presented in appendix A shows that the following combinations of the model parameters $`v_{\mathrm{𝐥𝐫}}`$ and $`w_{\mathrm{𝐥𝐫}}`$ enter the master equation:
$$\left\{\begin{array}{c}G_0\\ G_{xx}\\ G_x\\ g_x\end{array}\right\}=\frac{2\pi }{\mathrm{}}\underset{\mathrm{𝐥𝐫}}{}\left\{\begin{array}{c}|v_{\mathrm{𝐥𝐫}}|^2\\ |w_{\mathrm{𝐥𝐫}}|^2\\ \mathrm{}\left(v_{\mathrm{𝐥𝐫}}^{}w_{\mathrm{𝐥𝐫}}\right)\\ \mathrm{}\left(v_{\mathrm{𝐥𝐫}}^{}w_{\mathrm{𝐥𝐫}}\right)\end{array}\right\}\left(\frac{f(\epsilon _𝐥)}{\epsilon _𝐥}\right)\delta (\epsilon _𝐥\epsilon _𝐫).$$
(5)
These lumped parameters for the junction have the following physical meaning: Let $`𝖦(x)=e^2G(x)`$, $`e`$ being the electron charge, denote the conductance as a function of the oscillator coordinate $`x`$ when it is treated as a classical variable. Then, $`G_0`$ gives the conductance of the junction in the absence of coupling to the oscillator, $`G_0=\frac{1}{e^2}𝖦|_{x=0}`$, and $`G_x=\frac{1}{2e^2}\frac{d𝖦}{dx}|_{x=0}`$ and $`G_{xx}=\frac{1}{2e^2}\frac{d^2𝖦}{dx^2}|_{x=0}`$. The coupling constant $`g_x`$ cannot be expressed via $`𝖦(x)`$. Note that $`g_x`$ changes its sign upon interchange of tunneling amplitudes between the states in the two electrodes, i.e., after the substitution $`𝐥𝐫`$. Therefore, it is only finite for an asymmetric junction and is a measure of the asymmetry. As shown in section III, $`g_x`$ generates effects similar to charge pumping, as well as nontrivial features in the electric current noise as discussed in section VI. For later use it is convenient to present the coupling constants in terms of conductances by introducing the characteristic length of the oscillator
$$\stackrel{~}{G}_{xx}=G_{xx}x_0^2,\stackrel{~}{G}_x=G_xx_0,\stackrel{~}{g}_x=g_xx_0,$$
(6)
where $`x_0=(\mathrm{}/m\omega _0)^{1/2}`$, and $`\omega _0`$ is the frequency of the coupled oscillator as introduced in appendix A.
### II.1 Charge specific master equation
To study the interaction of charge dynamics in a tunnel junction with the dynamics of a quantum object, we describe the combined system, the quantum degree of freedom coupled to a tunnel junction, using the charge specific density matrix method introduced in Ref. RamSheWab04, . The approach employs charge projectors to study the dynamics of the quantum object conditioned on the charge state of the junction. The charge projection operator, $`𝒫_n`$, projects the state of the conduction electrons in the junction onto its component for which exactly $`n`$ electrons are in a given spatial region, say in the left electrode. The charge specific density matrix is then specified by
$$\widehat{\rho }_n(t)=\text{Tr}_{el}(𝒫_n\rho (t))$$
(7)
where $`\rho (t)`$ is the full density matrix for the combined system, and $`\text{Tr}_{el}`$ denotes the trace with respect to the conduction electrons in the junction. Provided the system at the initial time, $`t=0`$, is in a definite charge state, i.e., described by a charge specific density matrix of the form $`\widehat{\rho }_n(t)=\delta _{n0}\widehat{\rho }_0`$, where $`\widehat{\rho }_0`$ is the initial state of the quantum object, the charge index $`n`$ can be interpreted as the number of charges *transferred* through the junction. Thus the charge projector method provides a basis for charge counting statistics in the cases where the distribution function for *transferred* charge is relevant as discussed in Ref. SheRam03, . The charge specific density matrix allows therefore the evaluation, at any moment in time, of the joint probability of the quantum state of the object and the number of charges transferred through the junction. For example, if the charge specific density matrix is traced over the quantum object degree of freedom, the probability $`p_n(t)`$ that $`n`$ charges in time span $`t`$ are *transferred* through the low transparency tunnel junction is the expectation value of the charge projector, or expressed in terms of the charge specific density matrix
$$p_n(t)=\mathrm{Tr}(\rho _n(t))$$
(8)
where the trace is with respect to the degree of freedom of the coupled quantum object.
The Markovian master equation for the charge specific density matrix, $`\widehat{\rho }_n(t)`$, for the case of coupling of the junction to a quantum object is derived and discussed in appendix A. The Markovian approximation is valid for describing slow time variations of the density matrix; the exact conditions of the applicability are specified later once the characteristic times of the problem have been identified. To lowest order in the tunneling, the master equation for the charge specific density matrix can be generally written in terms of super-operators: a Lindblad-like term $`\mathrm{\Lambda }`$, a diffusion term $`𝒟`$ and a drift term $`𝒥`$: RamSheWab04
$$\dot{\widehat{\rho }}_n=\frac{i}{\mathrm{}}[\widehat{H}_0,\widehat{\rho }_n]+\mathrm{\Lambda }\{\widehat{\rho }_n\}+𝒟\{\widehat{\rho }_n^{\prime \prime }\}+𝒥\{\widehat{\rho }_n^{}\},$$
(9)
where $`\widehat{\rho }_n^{}`$ and $`\widehat{\rho }_n^{\prime \prime }`$ denote the discrete derivatives,
$`\widehat{\rho }_n^{}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(\widehat{\rho }_{n+1}\widehat{\rho }_{n1}\right),`$ (10)
$`\widehat{\rho }_n^{\prime \prime }`$ $`=`$ $`\widehat{\rho }_{n+1}+\widehat{\rho }_{n1}2\widehat{\rho }_n.`$ (11)
General expressions for the super-operators in Eq. (9) are presented in appendix A as well as their specific form for the case of coupling to an oscillator. The equation shall in Sec. IV be used to study the current noise in the junction due to the coupling to a quantum object, before we in Sec. V consider the explicit case of an oscillator coupled to the junction. However, first we analyze the master equation for the unconditional density matrix, i.e., the charge specific density matrix traced with respect to the charge degree of freedom of the junction.
### II.2 Unconditional Master equation
Often interest is not in the detailed information of the charge evolution of the tunnel junction contained in the charge specific density matrix. If for example interest is solely in properties of the oscillator, this information is contained in the traced charge specific density matrix. We are thus led to study the master equation for the reduced or charge unconditional density matrix, the density matrix traced with respect to the charge degree of freedom, $`\widehat{\rho }(t)=_n\widehat{\rho }_n(t)`$. Performing the charge trace on Eq. (9), the master equation for the reduced density matrix can be written in the form
$`\dot{\widehat{\rho }}(t)`$ $`=`$ $`{\displaystyle \frac{1}{i\mathrm{}}}[\widehat{H}_R,\widehat{\rho }]+{\displaystyle \frac{\gamma }{i\mathrm{}}}[\widehat{x},\{\widehat{p},\widehat{\rho }\}]{\displaystyle \frac{D}{\mathrm{}^2}}[\widehat{x},[\widehat{x},\widehat{\rho }]]`$ (12)
$`+`$ $`{\displaystyle \frac{A}{\mathrm{}^2}}[\widehat{x},[\widehat{p},\widehat{\rho }]].`$
The form of the master equation is generic to any continuous quantum degree of freedom coupled linearly to the junction, and has the well-known form for a particle coupled to a heat bath.CalLeg83 ; BrePet02 In the following we consider the model Hamiltonian for a nanoelectromechanical system introduced in section II, and encounter the renormalized oscillator Hamiltonian
$$\widehat{H}_R=\frac{\widehat{p}^2}{2m}+\frac{m\omega _0^2\widehat{x}^2}{2}$$
(13)
which in addition to having a renormalized oscillator frequency $`\omega _0`$, suffers a voltage dependent linear shift in the equilibrium position of the oscillator, which in the following is assumed absorbed into the position of the oscillator (for details see appendix A).
The second and third term on the right in Eq. (12) represent the physical influences of friction and fluctuations of the environment. For a nanomechanical object, the environment consists of several parts. The first one, which we have explicitly included in the model, is the tunnel junction. The other one is included phenomenologically by introducing $`\gamma _0`$ and $`D_0`$, the values of the friction and diffusion parameters in the absence of coupling to the junction. The physical mechanism generating the friction coefficient $`\gamma _0`$ and diffusion coefficient $`D_0`$ is, e.g., the heat exchange of the nanoscale oscillator and the bulk substrate it is mounted on. Thus the latter environment could also be modeled microscopically in the standard manner of coupling a quantum object to a heat bath. FeyVer63 ; CalLeg83 Then, with the model Hamiltonian in Eq. (1) giving the electronic environment contribution due to the coupling to the junction, the total friction and diffusion coefficients become
$$\gamma =\gamma _0+\gamma _e,D=D_0+D_e$$
(14)
where $`\gamma _e`$ is the electronic contribution to the damping coefficient
$$\gamma _e=\frac{\mathrm{}^2G_{xx}}{m}$$
(15)
proportional to the coupling strength $`G_{xx}`$, and the electronic contribution to the diffusion coefficient is
$$D_e=m\gamma _e\mathrm{\Omega }\mathrm{coth}\frac{\mathrm{\Omega }}{2T_e}$$
(16)
where $`\mathrm{\Omega }=\mathrm{}\omega _0`$ and the voltage dependent parameter, $`T_e`$, is given by the relation
$$\mathrm{coth}\frac{\mathrm{\Omega }}{2T_e}=\frac{V+\mathrm{\Omega }}{2\mathrm{\Omega }}\mathrm{coth}\frac{V+\mathrm{\Omega }}{2T}+\frac{V\mathrm{\Omega }}{2\mathrm{\Omega }}\mathrm{coth}\frac{V\mathrm{\Omega }}{2T}.$$
(17)
Here $`T`$ is the temperature of the junction and we assume a dc voltage bias, $`V=eU`$, $`U`$ being the applied voltage. We note that the r.h.s. of Eq. (17) is proportional to the well-known value of the power spectrum of current noise of the isolated junction, taken at the frequency of the oscillator.Kog96 The fact that the junction as a part of the environment is in a non-equilibrium state, is reflected in the voltage dependence of the electronic contribution to the diffusion coefficient. In section IV we show that $`T_e`$ is the effective temperature of the junction seen by the oscillator.
The phenomenological parameters $`D_0`$ and $`\gamma _0`$ are related to each other by virtue of the fluctuation-dissipation theorem. Assuming that the junction and the part of the environment responsible for $`\gamma _0`$ and $`D_0`$ have the same temperature, $`T`$, the diffusion coefficient can be generally presented in the form
$$D=m\mathrm{\Omega }\left(\gamma _0\mathrm{coth}\frac{\mathrm{\Omega }}{2T}+\gamma _e\mathrm{coth}\frac{\mathrm{\Omega }}{2T_e}\right).$$
(18)
The master equation Eq. (12) contains the term proportional to the coupling constant $`A`$. A term with this structure has been obtained in previous discussions of quantum Brownian motion.HaaRei85 ; BrePet02 The derivation of the master equation for the oscillator (see appendix A) shows that the main contribution to the coefficient $`A`$ in Eq. (12) comes from virtual tunneling processes with energy difference of initial and final states of the order of the Fermi energy, in contrast to the friction and diffusion coefficients which are controlled solely by the tunneling events in the vicinity of the Fermi surface. Besides, compared with the other terms in Eq. (12), the $`A`$-term has different symmetry relative to time reversal, i.e., the transformation $`\widehat{\rho }\left(\widehat{\rho }\right)^{}`$ and $`tt`$. The damping and diffusion terms, which are odd relative to time reversal, describe the irreversible dynamics of the oscillator, whereas the last term in Eq. (12) just like the Hamiltonian term is time reversible. These observations give the hint that the $`A`$-term is responsible for renormalization-like effects. This suggests that the $`A`$-term should be treated on a different footing than the dissipative terms. Indirectly, the $`A`$-term can be absorbed into the Hamiltonian dynamics at the price of having the time evolution of the oscillator described by a “renormalized” density matrix. Indeed, if we apply a (non-unitary) transformation to the density matrix, $`\widehat{\stackrel{~}{\rho }}=\left\{\widehat{\rho }\right\}`$, by acting on the density matrix $`\widehat{\rho }`$ with the super-operator
$$\left\{\widehat{\rho }\right\}=\widehat{\rho }+\mu [\widehat{p},[\widehat{p},\widehat{\rho }],$$
(19)
we can by proper choice of the parameter $`\mu `$,
$$\mu =\frac{A}{2m\mathrm{\Omega }^2},$$
(20)
cancel the $`A`$-term in the equation for the transformed matrix $`\widehat{\stackrel{~}{\rho }}`$. Leaving the $`\gamma `$ and $`D`$terms intact, the counter term is produced by the super-operator $``$ acting on the Hamiltonian part of the master equation Eq. (12). Applied to the original $`A`$-term, this procedure generates an additional contribution proportional to the product $`\mu AA^2`$, and it can be neglected to lowest order in the coupling, the limit which can be consistently studied.
The renormalized density matrix now obeys the master equation
$$\dot{\widehat{\rho }}(t)=𝒦\left\{\widehat{\rho }\right\}$$
(21)
where
$$𝒦\left\{\widehat{\rho }\right\}=\frac{1}{i\mathrm{}}[\widehat{H}_R,\widehat{\rho }]+\frac{\gamma }{i\mathrm{}}[\widehat{x},\{\widehat{p},\widehat{\rho }\}]\frac{D}{\mathrm{}^2}[\widehat{x},[\widehat{x},\widehat{\rho }]],$$
(22)
up to a term quadratic in the coupling constant. For compact notation, we have introduced the super-operator $`𝒦`$, and dropped the tilde for marking the renormalized density matrix: thus in the following the renormalized density matrix will also be denoted by $`\widehat{\rho }`$. The master equation being derived for the case of coupling to a tunnel junction is seen to be of the same form as for coupling linearly to a heat bath, i.e., an equilibrium state of a many-body system; the generic form of a damped quantum oscillator known from numerous investigations on quantum Brownian motion.CalLeg83 However, the diffusion term is qualitatively different from the usual case where the quantum object is coupled only to a heat bath. The non-equilibrium state of the junction, characterized by its voltage, gives rise to features not present when the coupling is simply to a many-body system in equilibrium.
We note that the super-operator, $``$, does not change the trace of the density matrix it operates on, and the renormalized density matrix is also normalized to unity. However, one has to keep in mind that the observables are to be calculated with the ”unrenormalized” density matrix. Up to first order in the coupling constant the expectation value of an observable $`O`$ are now given in terms of the renormalized density matrix according to
$$O=\text{Tr}\left(\left(\widehat{O}\frac{A}{2m\mathrm{\Omega }^2}[\widehat{p},[\widehat{p},\widehat{O}]]\right)\widehat{\rho }\right).$$
(23)
This relation transfers renormalization from the density matrix to observables. In the language of the Feynman diagram technique, Eq. (23) corresponds to a vertex correction.
In this paper, we use the Markovian approximation to describe the time evolution of the density matrix. This approximation is valid in the low frequency limit. For the dc-bias case, the characteristic frequency of time variation of the density matrix, $`\omega `$, must be small enough to meet the condition
$$\omega \omega _{max},\omega _{max}\mathrm{max}(\frac{T}{\mathrm{}},\frac{V}{\mathrm{}}).$$
(24)
From the unconditional master equation, the characteristic frequency is seen to be determined by the friction coefficient, $`\omega \gamma `$. This means that the coupling constant $`G_{xx}`$ in Eq. (15) must be small enough to meet the condition $`\gamma \omega _{max}`$.
## III Current-voltage characteristic
The average value of the current through the junction is given in terms of the probability distribution for charge transfers, i.e.,RamSheWab04
$$I(t)=e\frac{d}{dt}\underset{n}{}n\text{Tr}\widehat{\rho }_n(t)$$
(25)
where Tr denotes the trace with respect to the degree of freedom of the coupled quantum object. However, to lowest order in the tunneling, the average current turns out to be expressible through the charge unconditional density matrix, the reduced density matrix for the coupled quantum object. Indeed, the master equation for the charge specific density matrix then enables one to express the time derivative in Eq. (25) in terms of the reduced density matrix for the quantum object, the charge unconditional density matrix
$$I(t)=e\text{Tr}𝒥\{\widehat{\rho }(t)\}$$
(26)
where the drift super-operator, $`𝒥`$, is specified in Eq. (95).
### III.1 Contributions to the current under dc bias
For a dc bias $`V=eU`$, $`U`$ being the applied voltage, the drift operator $`𝒥`$ is given by Eq. (95), and the current, Eq. (26), is specified by
$$\frac{1}{e}I(t)=VG_t+I_q(V)+I_p(t)$$
(27)
and in general time dependent due to the coupling to the oscillator. The current consists of three physically distinct contributions. The first term is the Ohmic-like part of the current proportional to the conductance
$$G_t=G_0+2G_xx_t+G_{xx}x^2_t,$$
(28)
the instantaneous value of the conductance operator, Eq. (96), where $`x^n_t\text{Tr}\left(\widehat{x}^n\widehat{\rho }(t)\right)`$. We note that besides the pure Ohmic term of the isolated junction, the additional terms due to the coupling to the oscillator will in general contribute to the non-linear part of the current-voltage characteristic since the state of the oscillator will depend on the voltage. A case in question is discussed in the next section, where the stationary state of the oscillator is considered.
The second term, $`I_q`$, originates according to the derivation due to position and momentum operators are not commuting and for this reason we refer to it as the quantum correction to the current
$$I_q(V)=\frac{1}{2}\stackrel{~}{G}_{xx}\mathrm{\Delta }_V,$$
(29)
where $`\mathrm{\Delta }_V`$ is specified in Eq. (97) or equivalently
$$\mathrm{\Delta }_V=V+(\mathrm{\Omega }+V)N_{\mathrm{\Omega }+V}(\mathrm{\Omega }V)N_{\mathrm{\Omega }V}$$
(30)
and $`N_{\mathrm{\Omega }\pm V}=1/(e^{\frac{\mathrm{\Omega }\pm V}{T}}1)`$.
The last term in Eq. (27),
$$I_p(t)=e\mathrm{}g_x\dot{\widehat{x}}_t,\dot{\widehat{x}}=\frac{\widehat{p}}{m},$$
(31)
is proportional to the average velocity of the oscillator, and present only for an asymmetric junction, $`g_x0`$.
The Ohmic part of the current is calculated in section IV.2 for the stationary case. Next we turn to discuss the quantum correction and the dissipationless contribution to the current.
### III.2 Quantum correction to the current
The quantum dynamics of the oscillator leads to a suppresion of the dc current as expressed by the quantum correction to the current, $`I_q(V)`$. Unlike the other terms in the expression for the current, Eq. (27), the quantum correction, $`I_q(V)`$, does not depend on the state of the oscillator, but only on its characteristic energy and the temperature of the junction. At low voltages, $`VT`$, the quantum correction is linear in the voltage
$$I_q=V\stackrel{~}{G}_{xx}\left(\frac{1}{2}+N_\mathrm{\Omega }\frac{\mathrm{\Omega }}{T}N_\mathrm{\Omega }(N_\mathrm{\Omega }+1)\right)$$
(32)
where $`N_\mathrm{\Omega }=1/(e^{\frac{\mathrm{\Omega }}{T}}1)`$. At large voltages it reaches a constant value
$$I_q\frac{1}{2}\stackrel{~}{G}_{xx}\mathrm{\Omega },VT,\mathrm{\Omega },$$
(33)
in agreement with an earlier result obtained by a technique valid at zero-temperature.MozMar02 Our approach generalizes the expression for the current to arbitrary relations between junction voltage and temperature, and the frequency of the oscillator. The voltage dependence of the quantum correction to the conductance, $`G_q=I_q/V`$, is shown in Fig.1 for different temperatures.
### III.3 Dissipationless current
The last contribution in Eq. (27) to the current, $`I_p`$, is qualitatively different from the other terms. From Eq. (31), one sees that $`I_p`$ is proportional to the velocity of the oscillator. Therefore, the corresponding transferred charge through the junction, $`\delta Q_p=I_p\delta t`$, is controlled by the coordinate of the oscillator: $`\delta Q_p=e\mathrm{}g_x\delta x`$. Being proportional to a velocity, the current contribution $`I_p`$ is odd with respect to time reversal, and therefore a dissipationless current.
The presence of the term $`I_p`$ in the current, which does not depend explicitly on the voltage, means that a current through the junction can be induced just by the motion of the oscillator alone, i.e., by the time variation of a system parameter which in our case is the junction transparency. This effect is closely related to the well known physics of quantum pumping,Bro98 ; AltGla99 but has not, to our knowledge, been discussed in the present context. The dissipationless “pumping”-like current, $`I_p`$, is proportional to the coupling constant $`g_x`$. This contribution to the current is thus only present if the tunnel junction is asymmetric. This is in concordance with the quantum pumping effect.AleAltKam00 The global symmetry properties of the system thus crucially determine the existence and magnitude of the induced current.
A single mode oscillator driven by an external periodic force at frequency $`\omega `$ induces an $`ac`$-current, $`I_{}`$, with the same frequency and a phase of the ac current rigidly following the phase of the external force. For a given amplitude of the oscillations, $`x_{max}`$, the magnitude of the $`ac`$-current can be estimated as
$$I_{}\alpha _{as}e\omega $$
(34)
where the dimensionless parameter $`\alpha _{as}=\mathrm{}g_xx_{max}`$ characterizes the effective asymmetry of the junction. In principle, $`\alpha _{as}`$ may be comparable to unity so that $`I_{}e\omega `$, provided the amplitude of the oscillations $`x_{max}`$ is large enough and the conductance of the junction is not too small.
One can show that in the case of an oscillator with two or more modes interacting with the junction, the corresponding term generates directed pumping of charge.denis
## IV Stationary state properties
In this section, we shall study the stationary state of the reduced density matrix for the oscillator in the Markovian approximation. The question arises whether the stationary state of the oscillator is a thermal equilibrium state even though the environment is in a non-equilibrium state as the junction is biased. According to Eq. (21), the stationary renormalized density matrix of the oscillator, $`\widehat{\rho }_s`$, is determined by the equation
$$𝒦\left\{\widehat{\rho }_s\right\}=0,$$
(35)
and the solution is indeed of the form of a thermal density matrix, $`\widehat{\rho }_s\mathrm{exp}\left(\widehat{H}_R/T^{}\right)`$, where the temperature of the oscillator, $`T^{}`$, is specified by the relation
$$\mathrm{coth}\frac{\mathrm{\Omega }}{2T^{}}=\frac{D}{\gamma m\mathrm{\Omega }}.$$
Using Eq. (18), the temperature of the oscillator is related to the environment temperature and the voltage bias according to
$$\mathrm{coth}\frac{\mathrm{\Omega }}{2T^{}}=\frac{\gamma _0}{\gamma _0+\gamma _e}\mathrm{coth}\frac{\mathrm{\Omega }}{2T}+\frac{\gamma _e}{\gamma _0+\gamma _e}\mathrm{coth}\frac{\mathrm{\Omega }}{2T_e}.$$
(36)
The average occupation number for the oscillator, $`N^{}`$, is given by the Bose function, $`N^{}=1/(e^{\mathrm{\Omega }/T^{}}1)`$, and seen to be populated separately by the interaction with the two environments
$$N^{}=\frac{\gamma _0}{\gamma }N_\mathrm{\Omega }+\frac{\gamma _e}{\gamma }N_e$$
(37)
where $`N_e=1/(e^{\frac{\mathrm{\Omega }}{T_e}}1)`$.virial
We observe, that the oscillator acquires the temperature of the bath, $`T^{}T`$, if the interaction with the junction is weak and $`\gamma _0`$ is the dominant contribution to the friction, $`\gamma _0\gamma _e`$. The general case and the opposite limit where the dynamics of the junction is dominating, $`\gamma _e\gamma _0`$, we proceed to consider.
### IV.1 Oscillator heating
When the oscillator is well isolated and the interaction with the junction dominates, $`\gamma _e\gamma _0`$, the oscillator attains according to Eq. (36) the effective temperature of the junction $`T_e`$, as given by Eq. (17). As expected, in the absence of a bias voltage across the junction, $`V=0`$, the temperature of the oscillator equals that of the junction irrespective of its temperature. When the junction is biased, the oscillator is generally heated except at zero temperature and low voltages, and we first discuss the case of a junction at zero temperature.
At zero junction temperature, $`T=0`$, we must distinguish two voltage regions. If the voltage is smaller than the frequency of the oscillator, $`V<\mathrm{\Omega }`$, the temperature of the oscillator is also zero, $`T^{}=0`$, independent of the voltage as it follows from Eq. (36). In this regime, the interaction with the tunneling electrons is unable to excite the oscillator from its ground state. Heating can only take place beyond the voltage threshold given by the oscillator frequency. If instead the voltage is larger than the frequency of the oscillator, $`V>\mathrm{\Omega }`$, the temperature of the oscillator, $`T^{}`$, is determined by the following relation to the voltage
$$\mathrm{tanh}\frac{\mathrm{\Omega }}{2T^{}}=\frac{\mathrm{\Omega }}{|V|},$$
(38)
At high voltages, $`V\mathrm{\Omega }`$, the temperature of the oscillator approaches half the bias voltage, $`T^{}=V/2`$, in agreement with the result obtained in a previous study where the temperature of the junction was assumed to vanish.MozMar02
The heating of the oscillator, its excess temperature, $`\mathrm{\Delta }T=T^{}T`$, as a function of the bias, is shown in Fig. 2, both for the case where the coupling to the junction dominates and the opposite case of dominating external damping. The effect of the external damping is shown for moderate to strong external coupling, $`\gamma /\gamma _e=5,10,100`$. Increasing the coupling to the external heat bath leads to suppression of the heating of the oscillator, the additional environment acting as a heat sink. In the case where the coupling to the junction dominates, the inset shows that at low temperatures the oscillator is not excited at voltages below the frequency of the oscillator.
At high voltages, $`V\mathrm{\Omega },T`$, in the shot noise regime, the temperature of the oscillator $`T^{}`$ can be found from Eq. (38). Just as in the case of vanishing junction temperature, the oscillator temperature approaches half the bias voltage, $`T^{}=V/2`$, at large bias, and these results generalize previous studies which were limited to zero temperature and high voltages, $`V\mathrm{\Omega }`$.MozMar02 ; CleGir04
In the quest for using tunnel junctions to measure the position of a coupled object with the ultimate precision set by the uncertainty principle,KnoCle03 ; LahBuuCam04 ; GaiZolBad05 ; CleAldDri02 it is important to take into account that the measuring, involving a finite voltage, will invariably heat the oscillator. In this respect the presence of the additional heat bath, described by the coupling $`\gamma _0`$, is important. For example, envisioning the oscillator has been cooled to a temperature $`T`$ much lower than $`\mathrm{\Omega }`$ and a voltage is turned on. In order to obtain an appreciable signal the voltage must be larger than $`\mathrm{\Omega }`$. The oscillator will then in a time span of the order of $`\gamma ^1`$ be heated and attain a temperature for which the average number of quanta in the oscillator is
$$N^{}=\frac{\gamma _e}{2\gamma }\frac{|V|\mathrm{\Omega }}{\mathrm{\Omega }}.$$
(39)
Estimating the oscillator temperature, we have $`T^{}\mathrm{\Omega }/\mathrm{ln}(\gamma /\gamma _e)`$. A strong environmental coupling can thus be beneficial for retaining the oscillator in the ground state.
### IV.2 $`IV`$ characteristic
In the stationary state, the dc current $`I(V)`$, Eq. (27), is conveniently written on the form
$`{\displaystyle \frac{1}{e}}I(V)=VG_0`$ $`+`$ $`{\displaystyle \frac{1}{2}}\stackrel{~}{G}_{xx}(\text{}2VN^{}+(\mathrm{\Omega }V)N_{\mathrm{\Omega }V}`$ (40)
$`(\mathrm{\Omega }+V)N_{\mathrm{\Omega }+V}\text{}).`$
As before, $`N^{}`$ is the occupation number of the oscillator at the temperature $`T^{}`$. In the stationary state, only the coupling constant $`G_{xx}`$ is effective.
The first term in Eq. (40) is the current through the isolated junction, and the remaining term describes the influence of the oscillator on the current. It is interesting to consider the latter in the limit of vanishing environment temperature, $`T=0`$. At zero temperature, we must distinguish two voltage regimes. If the voltage is smaller than the frequency of the oscillator, $`V<\mathrm{\Omega }`$, we observed in the previous section that the oscillator remains in the ground state, $`N^{}=0`$, and the average current turns out to be equal to that of an isolated junction, $`I=eG_0V`$. One observes that the effect of zero point fluctuations present in the conductance, Eq. (28), is exactly cancelled by the quantum correction, Eq. (29). This lack of influence of zero point fluctuations is expected since the oscillator in its ground state is inert to the tunneling electrons for such low bias.
If the voltage is larger than the oscillator frequency, $`V>\mathrm{\Omega }`$, the slope of the I-V characteristic, at $`T=0`$, abruptly increases, $`I=VG_0+(V\mathrm{\Omega })\stackrel{~}{G}_{xx}/2`$.
At arbitrary temperatures, the linear conductance of the junction, $`G=I/V`$, $`V0`$, is given by
$$\frac{1}{e^2}G=G_0+\stackrel{~}{G}_{xx}\frac{\mathrm{\Omega }}{T}N_\mathrm{\Omega }(N_\mathrm{\Omega }+1).$$
(41)
To derive this formula, we recall that in the limit of vanishing bias, $`V0`$, the oscillator attains the temperature of the junction, $`T^{}=T`$.
At high temperatures, $`TV,\mathrm{\Omega }`$, the quantum correction to the current, and thereby the nonlinear quantum corrections, vanish, and we obtain the result
$$\frac{1}{e}I(V)=V(G_0+G_{xx}x^2_{}),x^2_{}=\frac{T^{}}{m\omega _0^2}$$
(42)
where $`x^2_{}`$ is the mean square of the oscillator coordinate at temperature $`T^{}`$. This result is to be expected from a classical oscillator in thermal equilibrium influencing the conductance of a tunnel junction. The $`IV`$ characteristic is in this regime nonlinear due to the voltage dependence of $`T^{}`$, Eq. (36).
## V Current noise
In this section we use the charge projection technique to develop the description of the statistical properties of the current of a tunnel junction coupled to a quantum object. The discussion will be kept quite general before we in the next section specialize to the case of a harmonic oscillator, the nano-electromechanical model of section II. We shall show how the charge dynamics, described by the master equation for the charge specific density matrix, can be used to obtain the statistical properties of the junction current, such as the noise power spectrum. The prerequisite for the success of this endeavour is that for the considered low transparency tunnel junction, the charge representation in fact provides the probability distribution for the charges transferred through the junction.SheRam03 ; RamSheWab04
### V.1 Current noise in the charge representation
The probability, $`p_n(t)`$, for $`n`$ charge-transfers in time span $`t`$ is according to Eq. (8) given by $`p_n(t)=\text{Tr}\widehat{\rho }_n(t)`$, where $`\widehat{\rho }_n(t)`$ is the charge specific density matrix, Eq. (7). The charge-transfer probability distribution specifies the stochastic process of charge transfers, $`n(t)`$. The variance of the charge fluctuations
$$n^2(t)=n^2(t)n(t)^2,$$
(43)
is defined in terms of the moments of the probability distribution of charge transfers
$$n^r(t)=\underset{n}{}n^rp_n(t),p_n(t)=\text{Tr}\left(\widehat{\rho }_n(t)\right).$$
(44)
To express the statistical properties of the current in terms of the probabilities for charge transfers, we inherit the stochastic current process, $`i(t)`$, through its relation to the charge transfer process
$$n(t)=_0^t𝑑t^{}i(t^{}).$$
(45)
The average current, given by $`i(t)=dn(t)/dt`$, is in accordance with Eq. (25). The variance of the charge fluctuations are expressed via the current fluctuations according to
$$n^2(t)=_0^t𝑑t_1_0^t𝑑t_2\delta i(t_1)\delta i(t_2).$$
(46)
where $`\delta i(t)=i(t)i`$, and $`i`$ is the average $`dc`$ current since we in the following shall consider the stationary state. Stationary current noise is characterized by the current-current correlator
$$S(\tau )=\delta i(t+\tau )\delta i(t).$$
(47)
Inserting this expression into Eq. (46) one obtains
$$n^2(t)=2_0^t𝑑\tau (t\tau )S(\tau ).$$
(48)
This expression allows one to relate the charge and current fluctuations.
Taking time derivatives of Eq. (48) gives
$$\frac{dn^2(t)}{dt}=2_0^t𝑑\tau S(\tau ),$$
(49)
and
$$S(t)=\frac{1}{2}\frac{d^2n^2(t)}{dt^2},$$
(50)
i.e., the current-current correlator equals the second derivative of the variance of charge transfers. This relation allows one to calculate the current-current correlator, $`S(t)`$, by evaluating the charge fluctuations using the master equation for the charge specific density matrix.
Eventually, interest is in the current noise power spectrum, $`S_\omega `$, given by
$$S_\omega =4_0^{\mathrm{}}𝑑t\mathrm{cos}(\omega t)S(t)$$
(51)
where $`\omega `$ is the frequency at which the noise is measured.Ric54
We observe that the zero frequency noise power, according to Eqs. (49) and (51), can be calculated from the general relation
$$S_{\omega =0}=2\frac{dn^2(t)}{dt}|_t\mathrm{}$$
(52)
i.e., as the rate of change of the charge variance at large times.
In the present approach it is convenient to calculate directly the current-current correlator, as done in the next section and appendix B. Only at the end we then transform to obtain the noise power spectrum. However, we note that the approach is equivalent to employing the widely used MacDonald formula.MacD
### V.2 Current-current correlator
In this section we show how the master equation for the charge specific density matrix, can be used to obtain the noise power spectrum of the current. A convenient feature of the method is that it allows one directly to obtain the time dependence of the current noise.
The probability distribution of charge transfers, $`p_n(t)`$, is obtained from the master equation for the charge specific density matrix given the initial condition corresponding to a state of definite initial charge
$$\widehat{\rho }_n(t=0)=\delta _{n,0}\widehat{\rho }_s$$
(53)
at the time when the charge counting starts, the initial time $`t=0`$. We are interested in the noise properties of the stationary state and, therefore, the stationary density matrix of the oscillator, the thermal state $`\widehat{\rho }_s`$, enters the initial condition.
In the following we shall treat the charge specific dynamics in the Markovian approximation. The charge specific density matrix $`\widehat{\rho }_n(t)`$, is obtained as the solution of the master equation Eq. (9). For notational convenience we write the charge specific master equation in the form
$$\frac{d\widehat{\rho }_n}{dt}=𝒦\left\{\widehat{\rho }_n\right\}+𝒟\left\{\widehat{\rho }_n^{\prime \prime }\right\}+𝒥\left\{\widehat{\rho }_n^{}\right\},$$
(54)
where $`𝒦`$ is the super-operator introduced in Eq. (21). Although the $`\widehat{\rho }_n`$’s are time dependent, the unconditional density matrix, $`\widehat{\rho }=_n\widehat{\rho }_n(t)`$, remains equal to the thermal state, $`\widehat{\rho }_s`$, by virtue of its stationarity property, $`𝒦\left\{\widehat{\rho }_s\right\}=0`$. Our goal is now to evaluate the variance of the charge transfers and thereby the current-current correlator with the help of Eq. (50).
The rate of change of the first charge moment, i.e., the dc current according to Eq. (25), becomes in the stationary state
$$\frac{1}{e}I=\text{Tr}\left(𝒥\left\{\widehat{\rho }_s\right\}\right),$$
(55)
following from Eq. (26) and the stationarity property of the density matrix for the coupled quantum object, $`\widehat{\rho }(t)=\widehat{\rho }_s`$. The dc current was calculated in section IV.2.
It readily follows Eq. (54), that the time derivative of the variance of charge transfers, $`n^2(t)`$, can be presented in the form
$$\frac{d}{dt}n^2(t)=2\text{Tr}\left(𝒟\left\{\widehat{\rho }_s\right\}\right)2\text{Tr}\left(𝒥\left\{\delta \widehat{N}(t)\right\}\right),$$
(56)
where $`\delta \widehat{N}(t)`$ denotes the traceless matrix
$$\delta \widehat{N}(t)=\underset{n}{}\left(nn(t)\right)\widehat{\rho }_n(t).$$
(57)
We observe that only the truncated density matrix, $`\delta \widehat{N}(t)`$, is needed to calculate the noise.
Comparing Eq. (56) and Eq. (49), one concludes that the current-current correlator has a $`\delta `$function like singularity at the initial time, $`t=0`$, where the charge counting starts. Indeed, the r.h.s. of Eq. (56) has a finite limit as $`t0`$, given by the first term, since the second term initially vanishes, $`\delta \widehat{N}(t=0)=0`$. For this result to be compatible with Eq. (49) and Eq. (50), the current-current correlator, $`S(t)`$, must have the following structure
$$S(t)=S_1(t)+S_2(t),$$
(58)
the sum of a singular contribution,
$$S_1(t)=2\text{Tr}\left(𝒟\left\{\widehat{\rho }_s\right\}\right)\delta (t)$$
(59)
where $`\delta (t)`$ denotes a function peaked at $`t=0`$ and normalized according to the condition $`_0^{\mathrm{}}𝑑t\delta (t)=\frac{1}{2}`$, and a regular part given by
$$S_2(t)=\text{Tr}\left(𝒥\left\{\delta \widehat{I}(t)\right\}\right),$$
(60)
where $`\delta \widehat{I}`$ denotes the matrix, $`\delta \widehat{I}=\frac{d}{dt}\delta \widehat{N}`$. The finite time correlation of the current described by the regular part $`S_2(t)`$ is solely due to the interaction with the quantum object, as follows from $`\delta \widehat{I}(t)`$ being traceless. We note here that the $`\delta `$-function singularity, which would provide noise at arbitrary high frequencies, is an artefact of the Markovian approximation.
The task of calculating the time dependent current noise is thus reduced to obtaining the time derivative of the charge-averaged density matrix, $`\delta \widehat{N}(t)`$, given in Eq. (57). From the master equation for the charge specific density matrix one obtains the following equation for $`\delta \widehat{I}(t)`$,
$$\frac{d}{dt}\delta \widehat{I}=𝒦\{\delta \widehat{I}\},$$
(61)
and the initial condition
$$\delta \widehat{I}|_{t=0}=\delta 𝒥\left\{\widehat{\rho }_s\right\}.$$
(62)
Here the super-operator $`\delta 𝒥`$ acts on its argument matrix according to
$$\delta 𝒥\left\{X\right\}=𝒥\left\{X\right\}X\left(\text{Tr}𝒥\left\{X\right\}\right).$$
(63)
We note, that acting on a matrix $`X`$ with unit trace, $`\text{Tr}X=1`$, the super-operator $`\delta 𝒥`$ returns a traceless matrix. The dynamics of the charge averaged quantity $`\delta \widehat{I}`$ is thus identical to that of the charge unconditional density matrix of the oscillator.
The formal solution to Eq. (61) can be written in terms of the time evolution super-operator for the charge unconditional density matrix of the oscillator
$$𝒰_t=e^{𝒦t}$$
(64)
as
$$\delta \widehat{I}(t)=𝒰_t\left\{\delta 𝒥\left\{\widehat{\rho }_s\right\}\right\}$$
(65)
and the regular part of the current-current correlator can be written on the form
$$S_2(t)=\text{Tr}\left(\delta 𝒥\left\{𝒰_t\left\{\delta 𝒥\left\{\widehat{\rho }_s\right\}\right\}\right\}\right).$$
(66)
Here $`𝒥`$ in Eq. (60) has been replaced for $`\delta 𝒥`$ in Eq. (60); the replacement is valid under the trace operation since $`\mathrm{Tr}\delta \widehat{I}(t)=0`$. Combined with the singular part in Eq. (58), this gives the general expression in the Markovian approximation for the current-current correlator of a tunnel junction interacting with a quantum system in its stationary state. The current noise correlator has thus conveniently been written with the help of the Markovian super-operators $`𝒦`$, $`𝒥`$ and $`𝒟`$.
In the next section we shall turn to calculating the noise properties for the case of the nanoelectromechanical device described in section II.
## VI Noise power spectrum
We now turn to calculate the current-current correlator of the tunnel junction coupled to the harmonic oscillator as described by the model of section II. Taking advantage of the general analysis in the Markovian approximation presented above in Section V.2, the current-current correlator can be written in the form:
$$S(t)=S_1(t)+S_x(t)+S_{x^2}(t).$$
(67)
Here $`S_1`$ is the singular part defined in Eq. (59) and specified in Eq. (115). The regular contribution is given by second and third terms on the right in Eq. (67), as obtained by inserting the expression for the drift super-operator $`𝒥`$, Eq. (95), into the expression for the regular contribution, Eq. (66), giving
$$S_x(t)=2VG_xx_J(t)+g_x\frac{\mathrm{}}{m}p_J(t)$$
(68)
and
$$S_{x^2}(t)=VG_{xx}x_J^2(t)$$
(69)
where the time dependent quantities are given by
$$X_J(t)=\mathrm{Tr}\left(\widehat{X}𝒰_t\left\{\delta 𝒥\left\{\widehat{\rho }_s\right\}\right\}\right),$$
(70)
with $`\widehat{X}=\widehat{x},\widehat{p},\widehat{x}^2`$, respectively. It is readily checked that the latter quantities evolve in time in accordance with their corresponding classical equations of motion for a damped oscillator with initial conditions given by
$$X_J(0)=\mathrm{Tr}\left(\widehat{X}\delta 𝒥\left\{\widehat{\rho }_s\right\}\right),\widehat{X}=\widehat{x},\widehat{p},\widehat{x}^2,\widehat{p}^2.$$
(71)
The calculation of these quantities are presented in appendix B, giving according to Eq. (68) and Eq. (69) an explicit expression for the current-current correlator $`S(t)`$ and thereby, according to Eq. (51), for the noise power $`S_\omega `$. Fourier transforming the current-current correlator gives, according to Eqs. (113, 114, 115), peaks in the noise power as well as a constant up-shift in the noise floor, the noise pedestal. The noise power spectrum is displayed in the inset in Fig. 3. As expected, there is a pronounced peak at the frequency of the oscillator, and in addition two side peaks each shifted by the frequency of the oscillator, one at zero frequency and one at twice the oscillator frequency.
Below, we analyze the noise in two frequency regions: (i) low frequency noise at frequencies $`\omega \gamma `$; and (ii) noise in the vicinity of the oscillator resonance frequency $`\omega \omega _0`$ and $`\omega 2\omega _0`$. We examine the voltage and temperature features of the noise power.
### VI.1 Low frequency noise
Let us consider low frequency noise, at frequencies of the order of the damping rate and lower, $`\omega \gamma `$. Then the noise power spectrum is given by the Fourier transform of the correlation functions $`S_1(t)`$ and $`S_{x^2}(t)`$ in Eqs. (114), and (115), respectively, giving
$$S_\omega =S^{(0)}+S^{(1)}+S_\omega ^{(2)}$$
(72)
where
$$S^{(0)}=2G_0V\mathrm{coth}\frac{V}{2T}$$
(73)
is the low frequency, $`\omega V`$, white Nyquist or Schottky noise of the isolated junction, and $`S^{(1)}`$ is the correction to the white noise due to the interaction with the oscillator
$$S^{(1)}=2\stackrel{~}{G}_{xx}\mathrm{\Omega }\left(N^{}(N_e+1)+N_e(N^{}+1)\right).$$
(74)
Together these two contributions form the noise pedestal, $`S_{\mathrm{}}=S^{(0)}+S^{(1)}`$. The frequency dependent part, $`S_\omega ^{(2)}`$, becomes at low frequencies
$$S_\omega ^{(2)}=\stackrel{~}{G}_{xx}\frac{4\gamma \gamma _e}{\omega ^2+4\gamma ^2}\frac{V}{\mathrm{\Omega }}\left(2VN^2+\left(V\mathrm{\Delta }_V\right)\left(2N^{}+1\right)\right).$$
(75)
The low frequency noise is displayed explicitly proportional to the coupling to the electronic tunnel junction environment as we have taken advantage of the relation $`\gamma _e=\stackrel{~}{G}_{xx}\mathrm{\Omega }`$. The width of the low frequency peak is twice the damping rate $`2\gamma `$. At zero bias, $`V=0`$, where $`S_\omega ^{(2)}=0`$ and the oscillator and effective junction temperatures equal the environment temperature, leaving $`N^{}=N_e=N_\mathrm{\Omega }`$, one recovers the fluctuation-dissipation relation for the noise power, $`S_{\omega =0}=4TG`$, where $`G`$ is the linear conductance of the junction in the presence of the interaction with the oscillator, i.e., given by Eq. (41).
In the following we discuss the features of the low frequency excess noise, the noise due to the coupling to the oscillator, and in particular the noise peak height at zero frequency, in the limits of temperatures high and low compared to the oscillator frequency.
#### VI.1.1 Low temperature noise
First we consider the low frequency noise at low temperatures, $`T\mathrm{\Omega }`$. As expected, no excess noise is according to Eq. (72) generated by the oscillator at zero temperature and voltages below the oscillator frequency, $`|V|<\mathrm{\Omega }`$, where the oscillator cannot be excited from its ground state. Indeed, in the region of low temperatures, $`T\mathrm{\Omega }`$, and low voltages, $`V<\mathrm{\Omega }`$, the oscillator is non-responsive and the excess noise, $`S^{(1)}`$ and $`S_\omega ^{(2)}`$, vanishes exponentially in $`\mathrm{\Omega }/T`$ below the activation energy $`\mathrm{\Omega }`$.
Close to the noise onset threshold, $`|V|\mathrm{\Omega }`$, in the narrow region, $`||V|\mathrm{\Omega }|T`$, the peak height relative to the pedestal rises linearly with temperature
$$S_{\omega =0}^{(2)}=\frac{\gamma _e}{\gamma }T\stackrel{~}{G}_{xx}.$$
(76)
At voltages much higher than threshold, $`|V|\mathrm{\Omega }`$, the peak height relative to the pedestal becomes
$$S_{\omega =0}^{(2)}=2\frac{\stackrel{~}{G}_{xx}^2}{\gamma }V^2\left(\frac{1}{2}+\frac{\gamma _e}{\gamma }\frac{|V|}{2\mathrm{\Omega }}+\left(\frac{\gamma _e}{\gamma }\frac{V}{2\mathrm{\Omega }}\right)^2\right).$$
(77)
The zero frequency noise is proportional to $`V^4`$ if the effective coupling to the electronic environment is appreciable, i.e., the ratio $`\gamma _e/\gamma `$ is not too small. In the high voltage limit, the oscillator is in the classical regime but with the oscillator temperature given by $`T^{}=|V|/2`$ as discussed in section IV.
#### VI.1.2 High temperature noise
At temperatures higher than the oscillator frequency, $`T\mathrm{\Omega }`$, we can distinguish two voltage regimes. At low voltages, $`VT`$, the peak height scales quadratically in both the temperature and voltage
$$S_{\omega =0}^{(2)}=\frac{2}{\gamma }\stackrel{~}{G}_{xx}^2V^2\left(\frac{T}{\mathrm{\Omega }}\right)^2,$$
(78)
and we recall that the oscillator temperature equals the junction temperature, $`T^{}=T`$.
At high voltages, $`VT`$, the peak height becomes
$$S_{\omega =0}^{(2)}=\frac{2}{\gamma }V^2\stackrel{~}{G}_{xx}^2\left(\frac{\gamma _0}{\gamma }\frac{T}{\mathrm{\Omega }}+\frac{\gamma _e}{\gamma }\frac{\left|V\right|}{2\mathrm{\Omega }}\right)^2,$$
(79)
At high temperatures, the oscillator is in the classical regime and the average occupation number depends linearly on the oscillator temperature. The peak height, proportional to the fluctuations in the oscillator position squared, is proportional to the square of the average occupation number, and is therefore proportional to the square of the oscillator temperature.
### VI.2 High frequency noise
Next, we investigate the properties of the peaks in the noise power spectrum occurring at finite frequencies, at the oscillator frequency, $`\omega \omega _0`$, and its harmonic, $`\omega 2\omega _0`$. The Markovian approximation allows us to consider the high frequency noise only under the condition Eq. (24), that the frequency is much smaller than the maximum value of the voltage or the temperature, and for frequencies in question, this means that $`\mathrm{max}(T,V)\mathrm{\Omega }`$ for consistency. The inset in Figure 3 shows the frequency dependence of the noise power spectrum, Eq. (51), in the case of high temperatures, $`T\mathrm{\Omega }`$. The noise power displays three peaks. The noise power spectrum, consists at $`\omega \omega _0`$ of a Lorentzian part, as given by the Fourier transform of Eq. (113), with a width given by the damping rate, $`\gamma `$, and an asymmetric part specified by Eq. (113) and present only for an asymmetric junction. At $`\omega 2\omega _0`$, the noise power spectrum is according to Eq. (114) a Lorentzian with a width given by twice the damping rate, $`2\gamma `$. We now turn to discuss these peak heights at high and low temperatures.
#### VI.2.1 High frequency noise at high temperatures
At high temperatures, $`TV\mathrm{\Omega }`$, the height of the peak at the oscillator frequency relative to the pedestal depends linearly on temperature
$$S_{\omega =\omega _0}^{(2)}=2\frac{T}{\gamma }\left[\stackrel{~}{G}_x^2\frac{4V^2}{\mathrm{\Omega }}\stackrel{~}{g}_x^2\mathrm{\Omega }\right],$$
(80)
and is determined by the conductances $`\stackrel{~}{G}_x`$ and $`\stackrel{~}{g}_x`$.
At double the oscillator frequency the peak height depends quadratically on temperature
$$S_{\omega =2\omega _0}^{(2)}=\frac{1}{\gamma }\stackrel{~}{G}_{xx}^2\frac{V^2T^2}{\mathrm{\Omega }^2}$$
(81)
and just as the peak at zero frequency determined by the conductance $`\stackrel{~}{G}_{xx}`$. We note that its height is half that of the peak at zero frequency, Eq. (78).
The expressions for the excess noise, Eq. (80) and Eq. (81), are in fact also valid at low voltage. Contrary to the peaks at $`\omega =0`$ and $`\omega =2\omega _0`$, which vanish in the absence of voltage, the excess noise power at $`\omega \omega _0`$ is therefore finite for an asymmetric junction even at zero voltage, and according to Eq. (80) in fact negative. An asymmetric junction with a $`Q`$-factor much larger than $`T/\mathrm{\Omega }`$ can thus at zero voltage lead to a suppression of the noise power below that of an isolated junction.
#### VI.2.2 High frequency noise at low temperatures
At high voltages and low temperatures, $`V\mathrm{\Omega }T`$, the peak height at the oscillator frequency becomes
$$S_{\omega =\omega _0}^{(2)}=\frac{2V}{\gamma }\left[2\stackrel{~}{G}_x^2V\left(1+\frac{\gamma _e}{\gamma }\frac{|V|}{\mathrm{\Omega }}\right)\stackrel{~}{g}_x^2\mathrm{\Omega }\left(1\frac{\gamma _e}{2\gamma }\right)\right],$$
(82)
and the peak height at twice the oscillator frequency
$$S_{\omega =2\omega _0}^{(2)}=\frac{1}{\gamma }\stackrel{~}{G}_{xx}^2V^2\frac{\gamma _e}{\gamma }\frac{|V|}{2\mathrm{\Omega }}\left(\frac{\gamma _e}{\gamma }\frac{|V|}{2\mathrm{\Omega }}+1\right).$$
(83)
The noise can be large due to the high oscillator temperature.
#### VI.2.3 Noise asymmetry
A striking feature of the finite frequency noise is the contribution proportional to $`g_xG_x`$. It is *odd* relative to the sign of the voltage and does *not* depend on the state of the oscillator (see Eq. (114)). This term, which is only present for an asymmetric junction, $`g_x0`$, does not contribute to the peak height at the oscillator frequency, but provides the asymmetry of the peak in the frequency region around the oscillator frequency, $`\omega =\omega _0`$. Separating the even and odd voltage contributions in the noise power, $`S_\omega ^\pm =\frac{1}{2}\left(S_\omega (V)\pm S_\omega (V)\right)`$, the odd contribution becomes
$`S_\omega ^{}`$ $`=`$ $`\stackrel{~}{g}_x\stackrel{~}{G}_x\left(F_{\omega _0}(\omega )F_{\omega _0}(\omega )\right)`$
$`\times `$ $`\left(V^2\mathrm{coth}{\displaystyle \frac{V}{2T}}+\mathrm{\Omega }V(2N_e+1){\displaystyle \frac{\mathrm{\Omega }}{2}}\mathrm{\Delta }_V\right),`$
with the frequency dependence given by the function
$$F_{\omega _0}(\omega )=\frac{2(\omega _0\omega )}{\gamma ^2+(\omega \omega _0)^2}.$$
The noise power spectrum is displayed by the full line in Figure 3 for the temperature $`T=0.01\mathrm{\Omega }`$, where only the peak at the oscillator frequency is appreciable. The even part in voltage of the noise power, $`S_\omega ^+`$, displayed by the dotted line, is a symmetric function of the frequency relative to the frequency of the oscillator, and the odd part in the voltage, $`S_\omega ^{}`$, displayed by the dash-dotted line, is an anti-symmetric function of the frequency relative to the oscillator frequency. If the voltage is reversed, the frequency dependence of the asymmetric part is mirrored around the frequency $`\omega =\omega _0`$.
In contrast to an isolated junction, the noise power of the coupled junction-oscillator system shows asymmetry in the voltage. This behavior is a novel feature that arises when an asymmetric junction is coupled to an additional degree of freedom.
## VII Conclusions
We have applied the charge projection technique to obtain the charge specific dynamics of a continuous quantum degree of freedom coupled to a tunnel junction. The master equation for the charge specific density matrix has been derived, describing the charge conditioned dynamics of the coupled object as well as the charge transfer statistics of the junction. The method allows evaluating at any moment in time the joint probability distribution describing the quantum state of the object and the number of charges transferred through the junction.
The approach, generally valid for any quantum object coupled to the junction, has been applied to the generic case of a nanoelectromechanical system, a harmonic oscillator coupled to the charge dynamics of a tunnel junction. In this regard it is important that the method allows inclusion of a thermal environment in addition to the electronic environment of the tunnel junction since nanoresonators are invariably coupled to a substrate. The oscillator dynamics, described by the reduced density matrix for the harmonic oscillator, the charge specific density matrix traced with respect to the charge index, has upon a renormalization been shown to satisfy a master equation of the generic form valid for coupling to a heat bath. Even though the electronic environment is in a non-equilibrium state, the master equation is of the Caldeira-Leggett type, consisting of a damping and a fluctuation term. Though the coefficients of the terms are not related by the equilibrium fluctuation-dissipation relation, the fluctuation term originating from the coupling to the junction is of the steady-state fluctuation-dissipation type, containing the current noise power spectral function of the isolated junction taken at the frequency of the oscillator. The diffusion parameter is thus determined by all energy scales of the problem including temperature, voltage and oscillator frequency. The presence of an environment in a non-equilibrium state thus leads to features which are absent when the oscillator is only coupled to a heat bath.
The Markovian master equation for the charge specific density matrix has been used to calculate the current. In general, the average junction current consists of an Ohmic term, however, with a conductance modified due to the coupling to the oscillator dynamics, a quantum correction, and a dissipationless ac current only present for an asymmetric junction and proportional to the instantaneous velocity of the oscillator. The latter term does not depend on voltage explicitly and is an example of an effect similar to quantum pumping.
The stationary state of the oscillator has been shown to be a thermal state even though the environment is in a non-equilibrium state. Therefore, the only effect of the bias is heating of the junction. Thus the stationary oscillator state is a thermal equilibrium state, though in equilibrium at a higher temperature than that of the environment if the junction is in a non-equilibrium state of finite voltage. This is a backaction effect of the measuring device, the tunnel junction, on the oscillator. At zero temperature and voltages below the oscillator frequency, the oscillator remains, to lowest order in the tunneling, in its ground state, and the dc current equals that of an isolated junction. The coupling of the oscillator to the additional heat bath, described by the coupling constant $`\gamma _0`$, is shown to be beneficial for avoiding heating of the oscillator due to a finite voltage. This is of importance for application of quantum point contacts and tunnel junctions to position measurements aiming at a precision reaching the quantum limit.
The charge projection method has been used to infer the statistical properties of the junction current from the charge probability distribution. For example, the noise power spectrum is specified in terms of the variance of the charge distribution. The master equation for the charge specific density matrix can therefore be used to obtain the current-current correlator directly, and this has been done explicitly in the Markovian approximation. The excess noise power spectrum due to the coupling to the oscillator consists of a main peak located at the oscillator frequency and two smaller peaks located at zero frequency and twice the oscillator frequency, respectively. The peaks at zero frequency and at twice the oscillator frequency have heights proportional to the coupling constant $`G_{xx}`$ squared, whereas the height of the peak at the oscillator frequency is proportional to the coupling constants $`G_x`$ and $`g_x`$ squared. The voltage and temperature dependencies of the peaks has been examined in detail.
For an asymmetric junction, the noise power spectrum contains a term with the striking feature of being an odd function of the voltage and independent of the state of the oscillator. Contrary to the case of a symmetric junction, coupling of an oscillator to an asymmetric junction with temperature higher than the oscillator frequency results even at zero voltage in a suppression of the noise power at the oscillator frequency, the excess noise power being negative. For an asymmetric junction, the noise power at $`\omega \omega _0`$ can thus be suppressed below the Nyquist level of the isolated junction.
The Markovian approximation employed to calculate the noise power can not be validated at arbitrary frequencies compared to temperature or voltage. Not surprisingly, naive attempts to extend expressions beyond the Markovian applicability range, Eq. (24), leads to unphysical results for the noise. For example, at zero temperature and voltages below the oscillator frequency, a spurious noise power arises even for the oscillator in the ground state.
###### Acknowledgements.
This work was supported by The Swedish Research Council.
## Appendix A Charge specific master equation
In a previous paper we introduced the charge representation for a general many-body system.RamSheWab04 The approach is based on the use of charge projectors previously introduced in the context of counting statistics.SheRam03 In the charge representation, the dynamics of a quantum object coupled to a many-body system is described by the charge specific density matrix
$$\widehat{\rho }_n(t)=\text{Tr}_{el}(𝒫_n\rho (t))$$
(85)
where $`\rho (t)`$ is the full density matrix for a many-body system and a quantum object coupled to it, and $`\text{Tr}_{el}`$ denotes the trace with respect to the degrees of freedom of the many-body system, in the following assumed the conduction electrons of a tunnel junction. The charge projection operators $`𝒫_n`$, which project the state of the system onto its component for which exactly $`n`$ electrons are in a specified region of space, have been discussed in detail earlier.SheRam03 ; RamSheWab04 There we discussed the circumstances under which the charge index, $`n`$, can be interpreted as the number of charges *transferred* through the junction, and the charge projector method thus provides a basis for charge counting statistics in the cases where the distribution function for transferred charge is a relevant concept. In this case, the charge specific density matrix allows the evaluation, at any moment in time, of the joint probability of the quantum state of the object and the number of charges transferred through the junction. In the previous paper, the non-Markovian master equation for the charge specific density matrix for an arbitrary quantum object coupled to a low transparency tunnel junction was derived.RamSheWab04 A non-Markovian master equation is less tractable for calculational purposes and the Markovian approximation is employed in the present paper. This is quite sufficient for calculations of average properties, such as the average current through the tunnel junction, where only the long time behavior needs to be addressed. However, when calculating the current noise of the junction, the Markovian approximation limits the description to the low frequency noise as discussed in section VI.
The Markovian charge specific master equation for a quantum object coupled to the junction was in general shown to have the formRamSheWab04
$$\dot{\widehat{\rho }}_n(t)=\frac{1}{i\mathrm{}}[\widehat{H}_0,\widehat{\rho }_n(t)]+\mathrm{\Lambda }\{\widehat{\rho }_n(t)\}+𝒟\{\widehat{\rho }_n^{\prime \prime }(t)\}+𝒥\{\widehat{\rho }_n^{}(t)\},$$
(86)
where $`\widehat{\rho }_n^{}`$ and $`\widehat{\rho }_n^{\prime \prime }`$ denote the “discrete derivatives” introduced in Eqs. (10, 11), and the Lindblad-like super operator, $`\mathrm{\Lambda }\{\widehat{\rho }_n\}`$, has the form
$`\mathrm{\Lambda }\{\widehat{\rho }\}={\displaystyle \frac{1}{\mathrm{}}}[{\displaystyle \underset{\mathrm{𝐥𝐫}}{}}f_𝐥(1f_𝐫)(\left[\widehat{T}_{\mathrm{𝐥𝐫}}^{}\right]\widehat{\rho }\widehat{T}_{\mathrm{𝐥𝐫}}\widehat{T}_{\mathrm{𝐥𝐫}}\left[\widehat{T}_{\mathrm{𝐥𝐫}}^{}\right]\widehat{\rho })`$ (87)
$`+{\displaystyle \underset{\mathrm{𝐥𝐫}}{}}f_𝐫(1f_𝐥)(\widehat{T}_{\mathrm{𝐥𝐫}}\widehat{\rho }\left[\widehat{T}_{\mathrm{𝐥𝐫}}^{}\right]\widehat{\rho }\left[\widehat{T}_{\mathrm{𝐥𝐫}}^{}\right]\widehat{T}_{\mathrm{𝐥𝐫}})]+H.c.`$
where here and in the following $`H.c.`$ represents the hermitian conjugate term with respect to the variable of the quantum object. The bracket denotes the operation
$$\left[\widehat{T}_{\mathrm{𝐥𝐫}}\right]=\frac{1}{\mathrm{}}\underset{0}{\overset{\mathrm{}}{}}𝑑\tau e^{i(\epsilon +eV)\tau /\mathrm{}}e^{i\widehat{H}_0\tau /\mathrm{}}\widehat{T}_{\mathrm{𝐥𝐫}}e^{i\widehat{H}_0\tau /\mathrm{}},$$
(88)
where $`\widehat{T}_{\mathrm{𝐥𝐫}}`$ is the oscillator perturbed tunneling amplitude, Eq. (4), and $`\epsilon =\epsilon _𝐥\epsilon _𝐫`$, and $`f_𝐥`$ and $`f_𝐫`$ are the single particle energy distribution functions for the electrodes which in the following are assumed in equilibrium described by the junction temperature $`T`$. In this paper, we restrict ourself to the case where the junction is biased by a constant voltage $`U`$, denoting $`V=eU`$, where $`e`$ is the electron charge. The dagger indicates hermitian conjugation of operators of the coupled quantum object.
The drift super-operator is
$`𝒥\{\widehat{R}\}`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{}}}{\displaystyle \underset{\mathrm{𝐥𝐫}}{}}{\displaystyle \frac{F_{\mathrm{𝐥𝐫}}^a}{2}}(\left[\widehat{T}_{\mathrm{𝐥𝐫}}^{}\right]\widehat{R}\widehat{T}_{\mathrm{𝐥𝐫}}+\widehat{T}_{\mathrm{𝐥𝐫}}\widehat{R}\left[\widehat{T}_{\mathrm{𝐥𝐫}}^{}\right]+H.c)`$ (89)
$`+`$ $`{\displaystyle \frac{F_{\mathrm{𝐥𝐫}}^s}{2}}(\left[\widehat{T}_{\mathrm{𝐥𝐫}}^{}\right]\widehat{R}\widehat{T}_{\mathrm{𝐥𝐫}}\widehat{T}_{\mathrm{𝐥𝐫}}\widehat{R}\left[\widehat{T}_{\mathrm{𝐥𝐫}}^{}\right]+H.c.),`$
and it has been written in terms of the symmetric and antisymmetric combinations of the distribution functions
$$F_{\mathrm{𝐥𝐫}}^s=f_𝐥+f_𝐫2f_𝐥f_𝐫,F_{\mathrm{𝐥𝐫}}^a=f_𝐥f_𝐫.$$
The diffusion super-operator can be obtained from the drift super-operator according to
$$𝒟\{\widehat{R}\}=\frac{1}{2}𝒥_{sa}\{\widehat{R}\},$$
(90)
where the subscript indicates that symmetric and antisymmetric combinations of the distribution functions should be interchanged, $`F_{\mathrm{𝐥𝐫}}^sF_{\mathrm{𝐥𝐫}}^a`$.
The bracket notation is specified in Eq. (88), where in general $`H_0`$ denotes the Hamiltonian for the isolated arbitrary quantum object. In the following we consider an oscillator coupled to the junction, and $`H_0`$ represents the isolated harmonic oscillator.
As expected, the coupling of the oscillator to the tunnel junction leads to a renormalization of its frequency, $`\omega _B^2\omega _0^2`$. The renormalization originates technically in the term present in the Lindblad-like operator, $`\mathrm{\Lambda }\{\widehat{\rho }\}`$, which is quadratic in the oscillator coordinate and of commutator form with the charge specific density matrix, and gives for the renormalized frequency
$$\omega _0^2=\omega _B^2\frac{2}{m}\underset{\mathrm{𝐥𝐫}}{}F_{\mathrm{𝐥𝐫}}^a(P_++P_{})|w_{\mathrm{𝐥𝐫}}|^2$$
where
$`P_\pm =\mathrm{}{\displaystyle \frac{1}{\epsilon V\mathrm{}\omega _0+i0}}.`$ (91)
For the considered interaction, the renormalization can be simply handled by changing in Eq. (88) from the evolution by the bare oscillator Hamiltonian to the oscillator Hamiltonian with the renormalized frequency, the shift being compensated by subtracting an identical counter term. The above frequency shift is then identified by the counter term having to cancel the quadratic oscillator term of commutator form generated by the $`\mathrm{\Lambda }\{\widehat{\rho }\}`$-part in Eq. (86). Substituting for the bare oscillator Hamiltonian, $`\widehat{H}_0`$, in (88) the renormalized Hamiltonian, Eq. (13), all quantities are expressed in terms of the physically observed oscillator frequency $`\omega _0`$. In particular, the bracket becomes
$$\left[\widehat{T}_{\mathrm{𝐥𝐫}}\right]=\pi \left(\delta _0v_{\mathrm{𝐥𝐫}}+(\delta _++\delta _{})\frac{w_{\mathrm{𝐥𝐫}}}{2}\widehat{x}i(\delta _+\delta _{})\frac{w_{\mathrm{𝐥𝐫}}}{2m\omega _0}\widehat{p}\right)+i\left(P_0v_{\mathrm{𝐥𝐫}}+(P_++P_{})\frac{w_{\mathrm{𝐥𝐫}}}{2}\widehat{x}i(P_+P_{})\frac{w_{\mathrm{𝐥𝐫}}}{2m\omega _0}\widehat{p}\right),$$
(92)
where
$`\delta _0=\delta (\epsilon V),\delta _\pm =\delta (\epsilon V\mathrm{}\omega _0),`$ (93)
and
$`P_0=\mathrm{}{\displaystyle \frac{1}{\epsilon V+i0}}.`$ (94)
In the following the notation $`\mathrm{\Omega }=\mathrm{}\omega _0`$ for the characteristic oscillator energy is introduced.
Evaluating the diffusion and drift operators for the case of position coupling of the oscillator to the junction, Eq. (4), we obtain for the drift super-operator
$$𝒥\{\widehat{R}\}=V𝒢\{\widehat{R}\}+\frac{\mathrm{}}{m\mathrm{\Omega }}G_{xx}\mathrm{\Delta }_V\mathrm{}(\widehat{x}\widehat{R}\widehat{p})+G_x\frac{i\mathrm{}}{2m\mathrm{\Omega }}\mathrm{\Delta }_V[\widehat{p},\widehat{R}]+g_x\frac{\mathrm{}}{2m}\{\widehat{p},\widehat{R}\}+g_x\frac{1}{2i}\left(V\mathrm{coth}\frac{V}{2T}+S_V\right)[\widehat{x},\widehat{R}],$$
(95)
where the conductance super-operator is defined as
$$𝒢\{\widehat{R}\}G_0\widehat{R}+G_x\{\widehat{x},\widehat{R}\}+G_{xx}\widehat{x}\widehat{R}\widehat{x},$$
(96)
and
$$\mathrm{\Delta }_V=\frac{V+\mathrm{\Omega }}{2}\mathrm{coth}\frac{V+\mathrm{\Omega }}{2T}\frac{V\mathrm{\Omega }}{2}\mathrm{coth}\frac{V\mathrm{\Omega }}{2T},$$
(97)
and
$$S_V=\frac{V+\mathrm{\Omega }}{2}\mathrm{coth}\frac{V+\mathrm{\Omega }}{2T}+\frac{V\mathrm{\Omega }}{2}\mathrm{coth}\frac{V\mathrm{\Omega }}{2T},$$
(98)
the latter being proportional to the current noise power spectrum at the frequency of the oscillator. For the diffusion super-operator we obtain
$`𝒟\{\widehat{R}\}`$ $`=`$ $`{\displaystyle \frac{V}{2}}\mathrm{coth}{\displaystyle \frac{V}{2T}}𝒢\{\widehat{R}\}+{\displaystyle \frac{G_{xx}}{2}}\left(B_V\widehat{x}\widehat{R}\widehat{x}+{\displaystyle \frac{\mathrm{}}{m}}\mathrm{}(\widehat{x}\widehat{R}\widehat{p})\right)+{\displaystyle \frac{\mathrm{}A}{m\mathrm{\Omega }}}\mathrm{}(\widehat{p}\widehat{R}\widehat{x})`$ (99)
$`+{\displaystyle \frac{G_x}{4}}\left(B_V\{\widehat{x},\widehat{R}\}+{\displaystyle \frac{i\mathrm{}}{m}}[\widehat{p},\widehat{R}]\right)+g_x\left({\displaystyle \frac{V}{2i}}[\widehat{x},\widehat{R}]+{\displaystyle \frac{\mathrm{}\mathrm{\Delta }_V}{4m\mathrm{\Omega }}}\{\widehat{p},\widehat{R}\}\right),`$
where
$$B_V=S_VV\mathrm{coth}\frac{V}{2T}$$
(100)
and we have introduced the notation
$$\mathrm{}(\widehat{p}\widehat{R}\widehat{x})=\frac{1}{2}\left((\widehat{p}\widehat{R}\widehat{x})+(\widehat{p}\widehat{R}\widehat{x})^{}\right),$$
(101)
and
$$\mathrm{}(\widehat{p}\widehat{R}\widehat{x})=\frac{1}{2i}\left((\widehat{p}\widehat{R}\widehat{x})(\widehat{p}\widehat{R}\widehat{x})^{}\right),$$
(102)
in (95) and (99).
The parameter $`A`$ in Eq. (99) is given by
$$A_V=\frac{\mathrm{}^2}{2m\mathrm{\Omega }}\underset{\mathrm{𝐥𝐫}}{}|w_{\mathrm{𝐥𝐫}}|^2F_{\mathrm{𝐥𝐫}}^s(P_+P_{}),$$
(103)
and was encountered and discussed in connection with the unconditional master equation, Eq. 12. Technically, it originates in our model from the principal value of integrals, i.e., from virtual processes where electronic states far from the Fermi surface are involved. Estimating its magnitude under the assumption that the couplings $`|w_{\mathrm{𝐥𝐫}}|^2`$ are constants, one obtains with logarithmic accuracy
$$A\frac{2\mathrm{}^2G_{xx}}{\pi m}\mathrm{ln}\frac{E_F}{\mathrm{max}(\mathrm{V},\mathrm{T},\mathrm{\Omega })}.$$
(104)
In the course of evaluating the diffusion and drift operators, combinations like $`\mathrm{}/\mathrm{}\left(v_{\mathrm{𝐥𝐫}}^{}w_{\mathrm{𝐥𝐫}}\right)`$ appear together with principal value terms. The phase of $`v_{\mathrm{𝐥𝐫}}^{}w_{\mathrm{𝐥𝐫}}`$ will in general be a random function of the electron reservoir quantum numbers $`𝐥`$ and $`𝐫`$. Summing over these quantum numbers, where the principal value term does not provide any restriction of the energy interval, as it happens in the case of terms proportional to delta functions, they will tend to average to zero, and we shall therefore in the following neglect such terms.fterm
## Appendix B Noise in Markovian approximation
In this section we evaluate in the Markovian approximation the current-current correlator of the tunnel junction for the case of a harmonic oscillator coupled to the junction. The task has been reduced to evaluating the expressions in Eq. (68) and Eq. (69), i.e., quantities of the form Eq. (70) where the involved super-operator is the evolution operator for the charge unconditional density matrix given in Eq. (21).
It immediately follows from the master equation, Eq. (21), that quantities like $`X(t)=\mathrm{Tr}(\widehat{X}\widehat{\rho }(t))`$, where $`\widehat{X}`$ can denote $`\widehat{x}`$, $`\widehat{p}`$, $`\widehat{x}^2`$, and $`\widehat{p}^2`$, and $`\widehat{\rho }`$ is an arbitrarily normalized solution to the master equation, satisfy the corresponding classical equations of motion for a damped oscillator. The variables entering Eq. (70) can therefore be expressed in terms of their initial values at time $`t=0`$, and restricting ourselves for simplicity to the case of weak damping, $`\gamma \omega _0`$, they have the form corresponding to that of an underdamped classical oscillator
$$x_J(t)=x_J(0)e^{\gamma t}\mathrm{cos}\omega _0t+\frac{p_J(0)}{m\omega _0}e^{\gamma t}\mathrm{sin}\omega _0t$$
(105)
and
$$p_J(t)=m\omega _0x_J(0)e^{\gamma t}\mathrm{sin}\omega _0t+p_J(0)e^{\gamma t}\mathrm{cos}\omega _0t$$
(106)
and
$`x_J^2(t)`$ $`=`$ $`e^{2\gamma t}x_J^2(0)\mathrm{cos}^2\omega _0t+{\displaystyle \frac{p_J^2(0)}{m^2\omega _0^2}}e^{2\gamma t}\mathrm{sin}^2\omega _0t`$ (107)
$`+`$ $`e^{2\gamma t}{\displaystyle \frac{\{x,p\}_J(0)}{2m\omega _0}}\mathrm{sin}2\omega _0t.`$
The initial values in these equations are found from Eq. (71) to be
$$x_J(0)=G_x\left(\frac{\mathrm{}^2}{m\mathrm{\Omega }}\right)\left(V(2N^{}+1)\frac{1}{2}\mathrm{\Delta }_V\right)$$
(108)
$$\frac{p_J(0)}{m}=g_x\frac{\mathrm{}}{2}\left(\mathrm{\Omega }(2N^{}+1)\left(V\mathrm{coth}\frac{V}{2T}+S_V\right)\right)$$
(109)
$`{\displaystyle \frac{p_J^2(0)}{2m}}+{\displaystyle \frac{m\omega _0^2x_J^2(0)}{2}}=`$
$`\stackrel{~}{G}_{xx}\left(\mathrm{\Omega }VN^2+\mathrm{\Omega }\left(V\mathrm{\Delta }_V\right)(N^{}+{\displaystyle \frac{1}{2}})\right)`$ (110)
$$\frac{m\omega _0^2x_J^2(0)}{2}\frac{p_J^2(0)}{2m}=\stackrel{~}{G}_{xx}\mathrm{\Omega }VN^{}(N^{}+1)$$
(111)
$$\{x,p\}_J(0)=0.$$
(112)
Substituting these initial values into equations Eqs. (105-107), we obtain using Eqs. (68) and (69) the expressions in Eq. (67) for the regular part of the current-current correlator
$`S_x(t)`$ $`=`$ $`\stackrel{~}{G}_x^2e^{\gamma t}\mathrm{cos}\omega _0tV\left(2V(2N^{}+1)\mathrm{\Delta }_V\right)\stackrel{~}{g}_x^2e^{\gamma t}\mathrm{cos}\omega _0t\left({\displaystyle \frac{1}{2}}\mathrm{\Omega }V\mathrm{coth}{\displaystyle \frac{V}{2T}}+\mathrm{\Omega }^2(N_eN^{})\right)`$ (113)
$`\stackrel{~}{g}_x\stackrel{~}{G}_xe^{\gamma t}\mathrm{sin}\omega _0t\left(V^2\mathrm{coth}{\displaystyle \frac{V}{2T}}+\mathrm{\Omega }V(2N_e+1){\displaystyle \frac{1}{2}}\mathrm{\Omega }\mathrm{\Delta }_V\right)`$
and
$$S_{x^2}(t)=\frac{1}{2}\stackrel{~}{G}_{xx}^2e^{2\gamma t}V\left(2VN^2+\left(V\mathrm{\Delta }_V\right)(2N^{}+1)\right)+\stackrel{~}{G}_{xx}^2e^{2\gamma t}V^2N^{}(N^{}+1)\mathrm{cos}2\omega _0t.$$
(114)
Evaluating in Eq. (59) the trace of the diffusion super-operator in the stationary state of the oscillator, $`𝒟\left\{\widehat{\rho }_s\right\}`$, we obtain according to Eq. (99) for the singular part of the noise correlator
$$S_1(t)=2\delta (t)\left(G_0\frac{V}{2}\mathrm{coth}\frac{V}{2T}+\stackrel{~}{G}_{xx}\frac{\mathrm{\Omega }}{2}\left(N^{}(N_e+1)+N_e(N^{}+1)\text{}\right)\right).$$
(115)
We observe that as to be expected, the Markovian approximation for the dynamics of the charge specific density matrix only captures the low-frequency noise, $`\omega <\mathrm{max}(\frac{T}{\mathrm{}},\frac{V}{\mathrm{}})`$, and in section VI we shall therefore only discuss this limit.nonmark
|
warning/0506/math0506497.html
|
ar5iv
|
text
|
# Plane curves in boxes and equal sums of two powers
## 1 Introduction
Let $`F[x_1,x_2,x_3]`$ be an absolutely irreducible form of degree $`d`$, producing a plane curve in $`^2`$. The central aim of this paper is to analyze the density of rational points on such curves, which are contained in boxes with unequal sides. We shall see below how such considerations may be used to obtain new paucity results for equal sums of two powers.
Suppose that $`𝐏=(P_1,P_2,P_3)`$ for fixed real numbers $`1P_1P_2P_3`$, say. Then we define
$$N(F;𝐏)=\mathrm{\#}\{𝐱^3:F(𝐱)=0,|x_i|P_i,(1i3),𝐱\text{ primitive}\},$$
where $`𝐱=(x_1,x_2,x_3)`$ is said to be primitive if $`\mathrm{h}.\mathrm{c}.\mathrm{f}.(x_1,x_2,x_3)=1`$. Our starting point is the recent work of the second author , who has shown that $`N(F;𝐏)=O(P^{2/d+\epsilon })`$ for any $`\epsilon >0`$, whenever $`P_i=P`$ for $`1i3`$. The implied constant in this bound depends at most upon the choice of $`\epsilon `$ and $`d`$, a convention that we shall follow throughout this paper. This is essentially best possible for curves of genus zero, and it is natural to ask what can be said about the quantity $`N(F;𝐏)`$ when the $`P_i`$ are of genuinely different sizes. With this in mind we define
$$T=\mathrm{max}\{P_1^{e_1}P_2^{e_2}P_3^{e_3}\},$$
(1)
where the maximum is taken over all integer triples $`(e_1,e_2,e_3)`$ for which the corresponding monomial $`x_1^{e_1}x_2^{e_2}x_3^{e_3}`$ occurs in $`F(𝐱)`$ with non-zero coefficient. Then for any $`\epsilon >0`$, the second author’s principal result for curves \[2, Theorem 3\] states that
$$N(F;𝐏)\left(\frac{P_1P_2P_3}{T^{1/d}}\right)^{1/d+\epsilon }.$$
(2)
This clearly reduces to the previous bound whenever $`P_i=P`$ for $`1i3`$. Turning to the case of unequal $`P_i`$, it is easy to construct examples for which (2) is not best possible. Indeed, let $`F=x_1^{d1}x_3x_2^d`$ and $`𝐏=(1,P,Q)`$, say. Then it follows from (2) that $`N(F;𝐏)P^{1/d}Q^{(d1)/d^2+\epsilon }`$ whenever $`P^dQ`$, whereas in fact $`N(F;𝐏)`$ has order of magnitude $`\mathrm{min}\{P,Q^{1/d}\}`$.
It transpires that in the case of unequal $`P_i`$ there is scope for improvement within the proof of (2) itself. This has been demonstrated by the second author \[3, Theorem 15\] in the special case $`P_1=1`$. For any $`\epsilon >0`$, it is shown that
$$N(F;1,P_2,P_3)P_3^\epsilon \mathrm{exp}\left(\frac{\mathrm{log}P_2\mathrm{log}P_3}{\mathrm{log}T}\right),$$
(3)
where $`T`$ is given by (1). In particular, this is always at least as sharp as (2) and is essentially best possible by our example above. Moreover, the formulation we have given obviously incorporates the corresponding problem for integral points on affine plane curves $`F(1,x_2,x_3)=0`$.
We build upon these results by returning once again to the framework provided by the proof of (2). Our aim is to establish a sharper bound for the interim case in which $`P_11`$ and the region $`|x_i|P_i`$ is sufficiently lopsided. Unfortunately the statement of the bound is somewhat complicated, and it will be convenient to introduce the quantities
$$\alpha =\frac{\mathrm{log}P_1}{\mathrm{log}P_3},\beta =\frac{\mathrm{log}P_2}{\mathrm{log}P_3},\tau =\frac{\mathrm{log}T}{d\mathrm{log}P_3},$$
for given $`1P_1P_2P_3`$ and $`T`$ as in (1). With this in mind, we have the following result.
###### Theorem 1.
Suppose that $`T(P_1P_2)^d`$, and let $`\epsilon >0`$ be given. Then we have
$$N(F;𝐏)P_3^\epsilon \mathrm{exp}\left(\frac{\alpha \beta +(\alpha +\beta \alpha \beta )(\tau \alpha \beta )}{d(\tau \alpha )(\tau \beta )}\mathrm{log}P_3\right).$$
In particular, Theorem 1 reduces to (3) in the case $`P_1=1`$. Indeed, we always have the lower bound $`TP_2^d`$ whenever $`F`$ is an absolutely irreducible form, and the bounds in Theorem 1 and (3) agree when $`\alpha =0.`$ On the face of it, one might think of the condition $`T(P_1P_2)^d`$ as being unduly restrictive. In fact a straightforward calculation shows that this is precisely the region for which Theorem 1 is sharper than (2).
Turning to our application of Theorem 1, we fix a choice of $`k4`$ and consider the diagonal equation
$$w^k+x^k=y^k+z^k.$$
(4)
For any $`X1`$, we denote by $`N_k(X)`$ the number of positive integer solutions in the region $`\mathrm{max}\{w,x,y,z\}X`$. There are $`2X^2+O(X)`$ trivial solutions in which $`y,z`$ are a permutation of $`w,x`$, and so we write $`N_k^{(0)}(X)`$ for the number of non-trivial solutions. This quantity has received a great deal of attention lately, and we mention in particular the results of Hooley and the second author \[2, Theorem 11\]. Together, they comprise the best available estimates for values of $`k`$ in the interval $`4k12`$. The first of these provides the bound
$$N_k^{(0)}(X)X^{5/3+\epsilon }$$
(5)
for any $`k4`$, whereas the second yields
$$N_k^{(0)}(X)X^{1+\epsilon }+X^{3/\sqrt{k}+2/(k1)+\epsilon }$$
(6)
for any such $`k`$, and supersedes Hooley’s bound for $`k6`$.
The aim of the second part of this paper is to improve upon the previous bounds whenever $`k=5`$ or $`6`$. This will be done via a suitable application of Theorem 1. It is unfortunate that we shall only make use of the special case (3), and not of Theorem 1 *per se*. Nonetheless, it is our belief that the bound in Theorem 1 still merits a full presentation. We shall establish the following result in Section $`3`$.
###### Theorem 2.
For any $`k4`$ and any $`\epsilon >0`$, we have
$$N_k^{(0)}(X)X^{3/2+1/(2k2)+\epsilon }.$$
We take this opportunity to remark that the proof of (6) may be modified slightly to give a sharper result. In fact it is possbile to establish the estimate
$$N_k^{(0)}(X)X^{1+\epsilon }+X^{3/\sqrt{k}+2/k+\epsilon },$$
(7)
for any $`\epsilon >0`$ and $`k4`$. At this point it is convenient to tabulate the various available bounds for $`N_k^{(0)}(X)`$, for $`k`$ in the range $`4k8`$. Let $`\epsilon >0`$. Then we may write $`N_k^{(0)}(X)=O(X^{\theta _k+\epsilon })`$, where the permissible values of $`\theta _k`$ are given in the following table. The rows in this table correspond to the estimates (5), (6), (7) and Theorem $`2`$, respectively.
Thus we see that Hooley’s bound (5) remains unbeaten only for $`k=4`$. For $`k=5`$ the exponent in Theorem $`2`$ is the sharpest known, but (7) should be used for larger values of $`k`$.
We now indicate how (7) can be established. An inspection of the proof \[2, §8\] of (6), reveals that it suffices to offer an alternative treatment of the curves of degree $`k1`$ which are contained in the non-singular projective surface (4). Our observed improvement rests upon a reformulation of Colliot-Thélène’s result \[2, Appendix\], as used in the proof of (6). This states that any non-singular surface of degree $`k`$ in $`^3`$ contains $`O(1)`$ curves of degree $`k2`$. In recent communications with the authors, Professor Colliot-Thélène has shown that any such non-singular surface actually contains $`O(1)`$ non-degenerate curves of degree $`2(k2)`$. Here, a curve in $`^3`$ is said to be non-degenerate if it is not contained in any $`^2^3`$. Since any absolutely irreducible curve of degree $`d`$ in $`^3`$ contains $`O(X^{2/d+\epsilon })`$ rational points of height at most $`X`$, by \[2, Theorem 5\], it remains to handle the plane curves of degree $`k1`$ which are contained in (4). Such curves arise as the intersection of (4) with a plane
$$aw+bx+cy+dz=0,$$
(8)
say, that contains one of the lines in (4). But we know that all of the lines contained in this surface are given by $`\{|w|,|x|\}=\{|y|,|z|\}`$ (for $`k`$ even), or by $`\{w,x\}=\{y,z\}`$ or $`w=x`$, $`y=z`$ (for $`k`$ odd). Hence it is trivial to deduce from (8) that the only available planes have $`\{|a|,|b|\}=\{|c|,|d|\}`$ (for $`k`$ even), or $`\{a,b\}=\{c,d\}`$ or $`a=b`$, $`c=d`$ (for $`k`$ odd). Thus there are relatively few planes (8) of low height that need to be considered. The proof may then be completed by counting points according to the height of the corresponding plane (8).
In the case $`k=4`$, it is worthwhile remarking that the proof of Theorem $`2`$ can readily be adapted to show that for any $`\epsilon >0`$ there are $`O(X^{5/3+\epsilon })`$ positive integer solutions to the equation
$$w^4+x^4+y^4=z^4,$$
in the region $`\mathrm{max}\{w,x,y,z\}X`$. This supersedes work of the first author , who has already obtained the exponent $`7/4+\epsilon `$.
Notation. We shall follow common practice in allowing the small positive quantity $`\epsilon `$ to take different values at different points in all that follows.
Acknowledgement. While working on this paper, the first author was supported by EPSRC Grant number GR/R93155/01.
## 2 Proof of Theorem 1
In this section we shall prove Theorem 1. If $`P_1=1`$, then Theorem 1 reduces to (3). Henceforth we assume that $`P_1>1`$. But then the condition $`T(P_1P_2)^d`$ implies that $`f_30`$, where we suppose that $`(f_1,f_2,f_3)`$ is the maximal triple taken in the definition (1) of $`T`$. We set
$$\kappa =(2f_1+f_2)/f_3+3,$$
(9)
and note that $`3\kappa 3d`$. During the course of our argument we will encounter difficulties if the values of $`\mathrm{log}P_i`$ are too close together. We therefore replace $`P_1,P_2,P_3`$ by
$$B_1=P_1P_3^\delta ,B_2=P_2P_3^{2\delta },B_3=P_3^{1+\kappa \delta }.$$
(10)
Here $`\kappa `$ is given by (9), and $`\delta `$ is defined to be
$$\delta =\frac{\epsilon }{180d^3}.$$
(11)
Writing $`B_1^{f_1}B_2^{f_2}B_3^{f_3}=T^{}`$, say, we observe that
$$\mathrm{log}T^{}=\mathrm{log}T+3d\delta \mathrm{log}P_3.$$
For any $`x0`$, define the functions
$$\alpha (x)=\frac{\alpha +x}{1+\kappa x},\beta (x)=\frac{\beta +2x}{1+\kappa x},\tau (x)=\frac{\tau +3x}{1+\kappa x},$$
(12)
where $`\alpha ,\beta ,\tau `$ are the quantities appearing in the statement of Theorem 1. Then (10) implies that
$$\alpha (\delta )=\frac{\mathrm{log}B_1}{\mathrm{log}B_3},\beta (\delta )=\frac{\mathrm{log}B_2}{\mathrm{log}B_3},\tau (\delta )=\frac{\mathrm{log}T^{}}{d\mathrm{log}B_3}.$$
It will be convenient to record that for any $`x0`$ we have $`0<\alpha (x)\beta (x)`$ and
$$\alpha (x)+\beta (x)\tau (x)1,$$
(13)
since $`P_3^dT(P_1P_2)^d`$ and $`\kappa 3`$. Finally, we define the function
$$g(x)=\frac{\alpha (x)^2}{\tau (x)\beta (x)}\frac{\beta (x)1}{\tau (x)\alpha (x)}+\frac{\alpha (x)+\beta (x)\alpha (x)\beta (x)}{\tau (x)\alpha (x)}.$$
(14)
With these notations, our task is to establish that
$$N(F;𝐏)P_3^\epsilon \mathrm{exp}\left(\frac{1}{d}g(0)\mathrm{log}P_3\right),$$
(15)
whenever $`T(P_1P_2)^d`$.
Our first step is to note that $`N(F;𝐏)N(F;𝐁)`$, and we proceed to estimate the latter. Fortunately we shall only need to make minor alterations to the second author’s proof of (2) to do so. Let $`𝒫\mathrm{log}^2(FB_3)`$, where $`F`$ denotes the maximum modulus of the coefficients of $`F`$. Then according to \[1; Lemma 4\] it suffices to consider the set of $`𝐱`$ counted by $`N(F;𝐁)`$ for which $`pF(𝐱)`$, for a fixed prime $`p`$ in the range $`𝒫p𝒫`$. For each non-singular $`𝐭=(t_1,t_2,t_3)`$ on the projective variety $`F(𝐭)0(modp)`$, we write $`S(𝐭)`$ for the set of points counted by $`N(F;𝐁)`$ for which $`𝐱\lambda 𝐭(modp)`$ for some integer $`\lambda `$. Clearly there are $`O(𝒫)`$ possible values of $`𝐭`$. Let $`\delta `$ be given by (11). We plan to show that whenever
$$𝒫B_3^\epsilon \mathrm{exp}\left(\frac{1}{d}g(\delta )\mathrm{log}B_3\right)\mathrm{log}^2F,$$
(16)
there is an auxiliary form $`G(𝐱)`$ of degree $`O(1)`$, such that $`FG`$ and $`G(𝐱)=0`$ for all $`𝐱S(𝐭)`$. It turns out that we may only do this if
$$\mathrm{log}T^{}d\mathrm{log}B_1+d\mathrm{log}B_2.$$
(17)
Under this assumption we easily deduce the estimate
$$N(F;𝐁)B_3^\epsilon \mathrm{exp}\left(\frac{1}{d}g(\delta )\mathrm{log}B_3\right),$$
(18)
via an application of Bézout’s Theorem and \[1; Theorem 4\], just as in the proof of (2). We now show how Theorem 1 can be derived from (18). We first observe that (17) holds in view of the first of the inequalities (13). Next we show how (15) follows from the corresponding estimate (18) for $`N(F;𝐁)`$. This will require the following result.
###### Lemma 1.
Let $`\delta `$ be given by (11). Then we have
$$g(\delta )g(0)+\epsilon .$$
Since $`F`$ is absolutely irreducible we must have $`TP_3`$, and hence (12) implies the lower bound $`\tau (x)1/d`$. Once combined with (13), we deduce that
$$\tau (x)\alpha (x)\mathrm{max}\{\beta (x),\frac{1}{d}\alpha (x)\}\mathrm{max}\{\alpha (x),\frac{1}{d}\alpha (x)\}\frac{1}{2d}.$$
(19)
To prove Lemma 1 we shall also use the fact that for any $`x0`$ we have the trivial inequalities
$$0<\alpha (x),\beta (x),\tau (x)1,|\alpha ^{}(x)|,|\beta ^{}(x)|,|\tau ^{}(x)|6d.$$
(20)
To see the last three inequalities we note that
$$|\tau ^{}(x)|=\left|\frac{3}{1+\kappa x}\frac{\kappa (\tau +3x)}{(1+\kappa x)^2}\right|3+\kappa 6d,$$
for example, since $`3\kappa 3d`$. If we write
$$h_1(x)=\frac{\alpha (x)+\beta (x)\alpha (x)\beta (x)}{\tau (x)\alpha (x)},$$
then there exists some $`0<\xi <\delta `$ such that $`h_1(\delta )h_1(0)=\delta h_1^{}(\xi ),`$ by the mean value theorem. Using (13) it is easy to see that $`0\alpha (x)+\beta (x)\alpha (x)\beta (x)1`$, and so (19) and (20) yield
$$|h_1^{}(x)|24d^2+48d^372d^3,$$
for any $`x0`$. Similarly, we write
$$h_2(x)=\frac{1\beta (x)}{\tau (x)\alpha (x)},$$
and deduce that $`|h_2(\delta )h_2(0)|60d^3\delta `$. Finally, we write
$$h_3(x)=\frac{\alpha (x)^2}{\tau (x)\beta (x)},$$
and consider the derivative
$$h_3^{}(x)=\frac{2\alpha (x)\alpha ^{}(x)}{\tau (x)\beta (x)}\left(\frac{\alpha (x)}{\tau (x)\beta (x)}\right)^2(\tau ^{}(x)\beta ^{}(x)).$$
But (13) implies that $`\alpha (x)\tau (x)\beta (x)`$, and so we easily obtain the bound $`|h_3(\delta )h_3(0)|24d\delta `$. Taken together with (14), it therefore follows that
$`|g(\delta )g(0)|`$ $``$ $`|h_3(\delta )h_3(0)|h_2(\delta )+|h_2(\delta )h_2(0)|h_3(0)+|h_1(\delta )h_1(0)|`$
$``$ $`2d|h_3(\delta )h_3(0)|+|h_2(\delta )h_2(0)|+|h_1(\delta )h_1(0)|`$
$``$ $`\delta \{48d^2+60d^3+72d^3\}`$
$``$ $`\epsilon ,`$
by (11). This completes the proof of Lemma 1.
We are now in a position to deduce (15) from (18) and Lemma 1. Using the same notation it is easy to deduce from (12), (13) and (14) that
$$g(0)=h_3(0)h_2(0)+h_1(0)4d.$$
Recall that $`\kappa 3d`$. Then (10), (11) and Lemma 1 yield
$`{\displaystyle \frac{1}{d}}g(\delta )\mathrm{log}B_3{\displaystyle \frac{1}{d}}g(0)\mathrm{log}P_3`$ $``$ $`{\displaystyle \frac{1}{d}}\mathrm{log}P_3\{\epsilon +\kappa \delta (g(0)+\epsilon )\}`$
$``$ $`2\epsilon \mathrm{log}P_3,`$
whence (15).
We now show how the lower bound (16) suffices for the existence of a suitable auxiliary form $`G(𝐱)`$. Let $`Dd`$ and $`A>0`$, and consider the exponent set
$$(A)=\{𝐞^3:e_i0,\underset{i=1}{\overset{3}{}}e_i=D,\underset{i=1}{\overset{3}{}}e_i\mathrm{log}B_iA,j\mathrm{s}.\mathrm{t}.e_j<f_j\}.$$
Let $`E=\mathrm{\#}(A)`$, and suppose for the moment that $`E\mathrm{\#}S(𝐭)`$. If we choose any distinct vectors $`𝐱^{(1)},\mathrm{},𝐱^{(E)}S(𝐭)`$, then it will actually suffice just to show that the determinant
$$\mathrm{\Delta }=det(𝐱^{(i)𝐞})_{1iE,𝐞(A)}$$
vanishes whenever (16) occurs. Indeed, the construction of the auxiliary polynomial
$$G(𝐱)=\underset{𝐞(A)}{}a_𝐞x_1^{e_1}x_2^{e_2}x_3^{e_3}$$
is then identical to that given in the proof of (2). We have written $`𝐰^𝐞=w_1^{e_1}w_2^{e_2}w_3^{e_3}`$ in the definition of $`\mathrm{\Delta }`$, in which rows correspond to the different vectors $`𝐱^{(i)}`$, and columns correspond to the various $`𝐞(A)`$. Furthermore, it is immediate from the proof of (2) that any such form $`G`$ cannot be divisible by $`F`$. The advantage over the previous situation is that our new exponent set $`(A)`$ allows us to choose an optimal value of $`A`$, for which better control over the size of $`|\mathrm{\Delta }|`$ is possible in certain situations.
Our proof now breaks into two parts. Firstly we must obtain an estimate for the real modulus of $`\mathrm{\Delta }`$, and then secondly show that its $`p`$-adic order is sufficiently large that $`\mathrm{\Delta }`$ must in fact vanish. We begin with the first of these, and use the fact that $`|x_j^{(i)}|B_j`$ for $`1j3`$ to deduce that the column corresponding to the exponent vector $`𝐞`$ consists of elements of modulus at most $`B_1^{e_1}B_2^{e_2}B_3^{e_3}.`$ It therefore follows that
$$|\mathrm{\Delta }|E^E\underset{𝐞(A)}{}B_1^{e_1}B_2^{e_2}B_3^{e_3}.$$
(21)
For any $`𝐞^3`$ with $`e_i0`$, we henceforth set
$$\sigma (𝐞)=e_1\mathrm{log}B_1+e_2\mathrm{log}B_2+e_3\mathrm{log}B_3.$$
For $`1i3`$, define $`_i`$ to be the subset of $`(A)`$ for which $`e_i<f_i`$. Then we have
$$\mathrm{log}\underset{𝐞(A)}{}B_1^{e_1}B_2^{e_2}B_3^{e_3}=\underset{𝐞(A)}{}\sigma (𝐞)\underset{i=1}{\overset{3}{}}\underset{𝐞_i}{}\sigma (𝐞),$$
(22)
since $`(A)=_1_2_3`$. Hence it suffices to estimate $`_{𝐞_i}\sigma (𝐞)`$, for which it will be convenient to write
$$b_i=\mathrm{log}B_i,1i3.$$
Let $`c=1/\delta +\kappa `$, where $`\kappa `$ and $`\delta `$ are given by (9) and (11), respectively. Then it follows from our initial change of variables (10) that we have the inequalities
$$0<b_1<b_2<b_3cb_1,b_3c(b_jb_i),$$
(23)
for each $`1i<j3`$.
It is convenient at this point to make the assumption that $`A`$ is contained in the interval
$$Db_2<ADb_3.$$
(24)
Clearly our definition of $`(A)`$ would be rather pointless if we allowed $`A>Db_3`$, since then the condition $`\sigma (𝐞)A`$ is automatic and we retrieve the exponent set considered in the proof of (2). Similarly, $`(A)`$ is obviously empty for any $`ADb_1`$. Our motive for omitting any treatment of the interval $`Db_1<ADb_2`$ is not so apparent. Indeed, it is possible to adjust our argument to take such values of $`A`$ into account and actually achieve something new at the end of it. We have chosen not to do so simply because we would ultimately be led to a weaker result than Theorem 1. Moreover, we are able to simplify our work considerably under the hypothesis (24).
Henceforth let $`i,j,k`$ denote distinct elements of the set $`\{1,2,3\}`$, and define
$$_{jk}=\{(m_j,m_k)^2:m_j,m_k0,m_j+m_k=D,\tau (m_j,m_k)A\},$$
where $`\tau (m_j,m_k)=m_jb_j+m_kb_k`$. We shall apply the following result to simplify our estimate for $`_{𝐞_i}\sigma (𝐞)`$.
###### Lemma 2.
For $`1i3`$ we have
$$\underset{𝐞_i}{}\sigma (𝐞)f_i\underset{(m_j,m_k)_{jk}}{}\tau (m_j,m_k)+O(A),$$
provided that (24) holds.
We prove the result for $`i=3`$, say, and consider values of $`𝐞_3`$ with a fixed component $`e_3<f_3`$. If $`m_1=e_1`$ and $`m_2=e_2+e_3`$ then $`|\tau (m_1,m_2)\sigma (𝐞)|db_3`$. In particular we either have $`(m_1,m_2)_{12}`$ or
$$A<\tau (m_1,m_2)A+db_3.$$
(25)
Let $`R_{12}`$ denote the number of non-negative $`m_1,m_2`$ such that $`m_1+m_2=D`$ and (25) holds. It follows that
$$\underset{𝐞_3}{}\sigma (𝐞)\underset{e_3<f_3}{}\left\{AR_{12}+\underset{(m_1,m_2)_{12}}{}(\tau (m_1,m_2)+db_3)\right\}.$$
Moreover, we observe that
$$A>Db_2Db_3,$$
(26)
by (23) and (24). Since $`\mathrm{\#}_{12}D+1`$ and $`Dd`$, this yields
$$\underset{𝐞_3}{}\sigma (𝐞)f_3\underset{(m_1,m_2)_{12}}{}\tau (m_1,m_2)+O(A+AR_{12}).$$
In order to estimate $`R_{12}`$ we set $`\nu =b_1D/(b_2b_1)`$. Then $`R_{12}`$ is the number of non-negative integers $`m_2D`$ for which
$$\frac{A}{b_2b_1}<m_2+\nu \frac{A}{b_2b_1}+\frac{db_3}{b_2b_1}.$$
It follows from (23) that $`db_3/(b_2b_1)1`$, whence $`R_{12}=O(1)`$. This suffices for the proof of Lemma 2.
Recall that $`\{i,j,k\}`$ is a permutation of $`\{1,2,3\}`$. Our next task is to establish the estimate
$$\underset{(m_j,m_k)_{jk}}{}\tau (m_j,m_k)=\frac{1}{2}\underset{g,h}{}^{}\frac{A^2D^2b_h^2}{b_gb_h}+O(A),$$
(27)
where $`\{g,h\}`$ runs over permutations of $`\{j,k\}`$, and $`\mathrm{\Sigma }^{}`$ denotes the condition $`b_h<A/D`$. Taking the case corresponding to $`i=3`$ first, we automatically have $`\tau (m_1,m_2)<A`$ since $`Db_2<A`$, by (24). Therefore
$`{\displaystyle \underset{(m_1,m_2)_{12}}{}}\tau (m_1,m_2)`$ $`=`$ $`{\displaystyle \underset{m_1+m_2=D}{}}\tau (m_1,m_2)`$
$`=`$ $`{\displaystyle \underset{m_2=0}{\overset{D}{}}}\{(b_2b_1)m_2+b_1D\}`$
$`=`$ $`(b_1+b_2)D(D+1)/2`$
$`=`$ $`{\displaystyle \frac{1}{2}}\left\{{\displaystyle \frac{A^2D^2b_1^2}{b_2b_1}}+{\displaystyle \frac{A^2D^2b_2^2}{b_1b_2}}\right\}+O(A).`$
In the case $`i=1`$, we find
$`{\displaystyle \underset{(m_2,m_3)_{23}}{}}\tau (m_2,m_3)`$ $`=`$ $`{\displaystyle \underset{m_3=0}{\overset{[\frac{ADb_2}{b_3b_2}]}{}}}\{(b_3b_2)m_3+b_2D\}`$
$`=`$ $`{\displaystyle \frac{A^2D^2b_2^2}{2(b_3b_2)}}+O(A),`$
on using (26). The treatment of the case $`i=2`$ is similar. This completes the proof of (27).
Upon combining (21), (22), Lemma 2 and (27), we are therefore led to the following result.
###### Lemma 3.
We have
$$\mathrm{log}|\mathrm{\Delta }|\frac{1}{2}\underset{i=1}{\overset{3}{}}f_i\underset{g,h}{}^{}\frac{A^2D^2b_h^2}{b_gb_h}+E\mathrm{log}E+O(A),$$
provided that (24) holds.
For our fixed prime $`p`$ with order of magnitude $`𝒫`$, we must now determine a lower bound for $`\nu _p(\mathrm{\Delta })`$, the $`p`$-adic order of $`\mathrm{\Delta }`$. However it suffices to compute $`E=\mathrm{\#}(A)`$, since \[1; Lemma 6\] implies that
$$\nu _p(\mathrm{\Delta })\frac{1}{2}E^2\{1+o(1)\},$$
(28)
as $`E\mathrm{}`$. In fact a lower bound for $`E`$ can easily be deduced by mimicking the previous calculation. Beginning with the analogue of (22), one obviously has
$$\left|\underset{𝐞(A)}{}1\underset{i=1}{\overset{3}{}}\underset{𝐞_i}{}1\right|\underset{i<j}{}\underset{𝐞_i_j}{}1=O(1).$$
(29)
Moreover, the corresponding version of Lemma 2 is
$$\underset{𝐞_i}{}1f_i\underset{(m_j,m_k)_{jk}}{}1+O(1),$$
(30)
under the same assumption that (24) holds. We prove this for $`i=3`$, say. Recall that $`m_1+m_2=D`$ and $`\tau (m_1,m_2)A`$ whenever $`(m_1,m_2)_{12}`$. In particular, for each integer $`0k<f_3`$ we either have $`(m_1,m_2k,k)_3`$ or $`0m_2<k`$ or $`Adb_3<\tau (m_1,m_2)A`$. Hence an argument similar to that used in the proof of Lemma 2 yields the upper bound
$$\underset{k<f_3}{}\underset{(m_1,m_2)_{12}}{}1\underset{𝐞_3}{}1+O(1),$$
which establishes (30). Combining (29) and (30), together with the corresponding version of (27), we obtain the lower bound
$$E\underset{i=1}{\overset{3}{}}f_i\underset{g,h}{}^{}\frac{ADb_h}{b_gb_h}+O(1),$$
(31)
provided that (24) holds.
Let $`A`$ be contained in the interval (24). Then as $`D\mathrm{}`$ we will have both $`E\mathrm{}`$ and $`A=o(D^2b_2)`$. It now follows from Lemma 3 that
$$\mathrm{log}|\mathrm{\Delta }|\frac{1}{2}\left\{f_1\frac{A^2D^2b_2^2}{b_3b_2}+f_2\frac{A^2D^2b_1^2}{b_3b_1}+f_3D^2(b_1+b_2)\right\}(1+o(1))+o(E^2),$$
as $`D\mathrm{}`$. Moreover from (28) and (31) we also have
$$\nu _p(\mathrm{\Delta })\frac{1}{2}\left\{f_1\frac{ADb_2}{b_3b_2}+f_2\frac{ADb_1}{b_3b_1}+f_3D\right\}^2(1+o(1)).$$
We may therefore conclude that
$$\frac{\mathrm{log}|\mathrm{\Delta }|}{\nu _p(\mathrm{\Delta })}\frac{f_1\frac{A^2D^2b_2^2}{b_3b_2}+f_2\frac{A^2D^2b_1^2}{b_3b_1}+f_3D^2(b_1+b_2)}{\left\{f_1\frac{ADb_2}{b_3b_2}+f_2\frac{ADb_1}{b_3b_1}+f_3D\right\}^2}(1+o(1)),$$
as $`D\mathrm{}`$. Define the constants
$$\lambda =\frac{db_3\mathrm{log}T^{}}{(b_3b_1)(b_3b_2)},\varphi =\frac{db_1b_2+b_3(\mathrm{log}T^{}db_1db_2)}{(b_3b_1)(b_3b_2)},$$
and
$$\gamma =\varphi (b_1+b_2)+\lambda b_1b_2.$$
Then in particular $`\lambda 0`$ and (17) implies that $`\varphi >0`$. We shall consider the behaviour of the real-valued function
$$f(A)=\frac{\lambda A^2+\gamma D^2}{(\lambda A+\varphi D)^2},$$
as $`A`$ varies over the interval (24). We recall that $`\mathrm{\Delta }`$ necessarily vanishes if $`p^{\nu _p(\mathrm{\Delta })}>|\mathrm{\Delta }|`$. Using the identities $`f_1+f_2+f_3=d`$ and $`\sigma (𝐟)=\mathrm{log}T^{}`$, a straightforward calculation reveals that $`\mathrm{\Delta }`$ vanishes if $`\mathrm{log}p>(1+o(1))f(A)`$ for any $`A`$ in the interval (24).
It remains to choose a suitable value of $`A`$ for which the function $`f(A)`$ is minimized. In fact, an easy calculation reveals that $`f`$ has a turning point at $`A=\gamma D/\varphi `$. Moreover, this value of $`A`$ is contained in the interval (24) precisely when $`\varphi b_2<\gamma \varphi b_3`$. The lower bound here always holds, whereas it is not hard to see that the upper bound is true if and only if (17) holds. Hence it suffices to take
$$p>\mathrm{exp}\left(\frac{db_1b_2b_3+(b_1b_3+b_2b_3b_1b_2)(\mathrm{log}T^{}db_1db_2)}{(\mathrm{log}T^{}db_1)(\mathrm{log}T^{}db_2)}(1+o(1))\right),$$
under this assumption. Therefore (16) is indeed satisfactory, provided that $`D`$ is chosen to be sufficiently large in terms of $`\epsilon `$ and $`d`$.
## 3 Equal sums of two powers
Let $`k4`$ and $`X1`$. We now turn to our estimate for the number $`N_k^{(0)}(X)`$ of positive non-trivial integral solutions of the Diophantine equation (4), which are contained in the region $`\mathrm{max}\{w,x,y,z\}X`$. It will suffice to count positive integers $`w,x,y,z`$ such that $`x<yz<w`$. For each such solution we define
$$v_1=zx,v_2=z+x.$$
It follows that
$$1y<wX,1v_1<v_22X.$$
(32)
Furthermore, under this transformation (4) takes the shape
$$2^{k1}\{w^ky^k\}=v_1f(v_1,v_2),$$
(33)
where
$$f(v_1,v_2)=\underset{0j<k/2}{}\left(\genfrac{}{}{0pt}{}{k}{2j+1}\right)v_1^{2j}v_2^{k2j1}$$
(34)
is a binary form of degree $`k1`$. In order to estimate the number of integers $`w,y,v_1,v_2`$ such that (32) and (33) hold, we begin by considering the contribution corresponding to a fixed choice of $`v_1`$. For this we define
$$\xi =\xi (v_1)=\underset{pv_1,p>2}{}p$$
to be the odd square-free kernel of $`v_1`$, and consider the set
$$S=\{(w,y)^2:\xi w^ky^k\}.$$
Let $`(w,y)S`$ and let $`p`$ be any prime divisor of $`\xi `$. Then we see that either $`p`$ divides $`y`$, or there exist at most $`k`$ integers $`\lambda _1,\mathrm{},\lambda _t`$, say, such that
$$w\lambda _iy(modp)$$
for some $`1it`$. Collecting these lattice conditions together via the Chinese Remainder Theorem, we therefore conclude that $`S`$ is a union of $`O(k^{\omega (\xi )})`$ lattices in $`^2`$, each of determinant $`\xi `$.
We henceforth fix our attention upon those $`w,y`$ contained in the region (32), which lie in one such lattice $`\mathsf{\Lambda }`$, say. By \[2, Lemma 1, (iii)\] there exist basis vectors $`𝐞^{(1)},𝐞^{(2)}\mathsf{\Lambda }`$ with
$$\xi |𝐞^{(1)}||𝐞^{(2)}|\xi ,$$
(35)
and such that whenever we write
$$(w,y)=u_1𝐞^{(1)}+u_2𝐞^{(2)}$$
(36)
for appropriate integers $`u_1,u_2`$, we automatically have
$$u_1X/|𝐞^{(1)}|,u_2X/|𝐞^{(2)}|.$$
(37)
We assume without loss of generality that $`|𝐞^{(1)}||𝐞^{(2)}|`$, and so
$$|𝐞^{(1)}|^i|𝐞^{(2)}|^j(|𝐞^{(1)}||𝐞^{(2)}|)^{(i+j)/2}$$
(38)
for any choice of $`ji0`$.
For each $`v_1`$, it suffices to count the number of integers $`u_1,u_2,v_2`$ lying in the region defined by (32) and (37), for which
$$g(u_1,u_2)=v_1f(v_1,v_2),$$
where $`g(u_1,u_2)`$ is obtained from $`2^{k1}\{w^ky^k\}`$ via the substitution (36). In fact we shall also fix a choice of $`u_2`$, and then count the number $`M(X;u_2,v_1)`$, say, of integers $`r,s`$ for which $`rX/|𝐞^{(1)}|`$, $`1s2X`$, and
$$p(r)=q(s),$$
(39)
where $`p(r)=g(r,u_2)`$ and $`q(s)=v_1f(v_1,s)`$.
An important issue here is whether or not the polynomial $`pq`$ is absolutely irreducible. But it is not hard to see that (39) is obtained from (4) via an appropriate affine plane section. We now distinguish the projective plane sections of (4) into three distinct types: those producing absolutely irreducible curves of degree $`k`$, those that produce a line and an absolutely irreducible curve of degree $`k1`$, and finally those that produce a union of absolutely irreducible curves each of degree $`k2`$. In the first case it is clear that the corresponding affine plane section is absolutely irreducible. In the second case we may deduce that the polynomial defining the resulting affine curve (39) is either the product of a linear polynomial and an absolutely irreducible polynomial of degree $`k1`$, or it is absolutely irreducible of degree $`k1`$. The latter possibility is clearly satisfactory, whereas the former possibility implies that $`pq`$ has degree $`k`$. But then $`\mathrm{deg}p=k`$ and $`\mathrm{deg}q=k1`$, and a result of Schmidt \[6, Theorem III.1B\] tells us that $`p(r)q(s)`$ should be absolutely irreducible.
In the final degenerate case, we may assume that the projective plane section of (39) produces at least two distinct absolutely irreducible curves. Indeed, the existence of any plane section producing precisely one line in (4) would imply the existence of a singular point on the surface. Now we know by the previously discussed result of Colliot-Thélène \[2, Appendix\] that (4) contains finitely many plane curves of degree $`k2`$. It follows that there can only be $`O(1)`$ projective plane sections which produce two distinct absolutely irreducible curves of degree $`k2`$. Recall from the introduction that any absolutely irreducible plane curve of degree $`d`$ contains $`O(X^{2/d+\epsilon })`$ rational points of height at most $`X`$. Since trivial integral solutions to (4) correspond to rational points lying on projective lines in the surface, we therefore conclude that there is a total contribution of $`O(X^{1+\epsilon })`$ to $`N_k^{(0)}(X)`$, from those affine plane sections of (4) which lead to reducible curves (39). This is clearly satisfactory for Theorem $`2`$, and we may assume henceforth that $`pq`$ is absolutely irreducible.
In order to estimate $`M(X;u_2,v_1)`$, we shall apply (3) to $`p(r)q(s)`$. In view of the shape (34) that $`f`$ takes, it is apparent that $`q(s)`$ contains the monomial $`s^{k1}`$ with non-zero coefficient. Taking $`P_2X/|𝐞^{(1)}|`$ and $`P_3=2X`$ in (3), we see that $`TX^{k1}`$ and hence
$$M(X;u_2,v_1)X^{1/(k1)+\epsilon }|𝐞^{(1)}|^{1/(k1)}.$$
Summing over the values of $`u_2X/|𝐞^{(2)}|`$ in (37) we therefore obtain the contribution
$``$ $`{\displaystyle \frac{X^{1+1/(k1)+\epsilon }}{|𝐞^{(1)}|^{1/(k1)}|𝐞^{(2)}|}}`$ (40)
$``$ $`(\xi ^{1/2}X)^{1+1/(k1)+\epsilon },`$
via (38) and (35). Let $`Y1`$, and write $`A_\epsilon =(12^\epsilon )^1`$ for a fixed choice of $`\epsilon >0`$. Then for any $`\theta 1`$ we have
$`{\displaystyle \underset{nY}{}}\xi (n)^\theta `$ $``$ $`{\displaystyle \underset{nY}{}}\xi (n)^\theta \left({\displaystyle \frac{Y}{n}}\right)^{1\theta +\epsilon }`$
$`=`$ $`Y^{1\theta +\epsilon }{\displaystyle \underset{nY}{}}\xi (n)^\theta n^{\theta 1\epsilon }`$
$``$ $`Y^{1\theta +\epsilon }{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\xi (n)^\theta n^{\theta 1\epsilon }`$
$``$ $`Y^{1\theta +\epsilon }{\displaystyle \underset{p}{}}\left\{1+p^\theta p^{\theta 1\epsilon }+p^\theta p^{2\theta 22\epsilon }+\mathrm{}\right\}`$
$`=`$ $`Y^{1\theta +\epsilon }{\displaystyle \underset{p}{}}\left\{1+p^\theta {\displaystyle \frac{p^{\theta 1\epsilon }}{1p^{\theta 1\epsilon }}}\right\}`$
$``$ $`Y^{1\theta +\epsilon }{\displaystyle \underset{p}{}}\left\{1+p^{1\epsilon }A_\epsilon \right\}`$
$`=`$ $`c_\epsilon Y^{1\theta +\epsilon },`$
say. Upon taking $`\theta =\frac{1}{2}(1+1/(k1))`$, so that $`\theta 1`$ for every $`k2`$, it therefore follows from (40) that
$`N_k^{(0)}(X)`$ $``$ $`X^{1+1/(k1)+\epsilon }{\displaystyle \underset{v_1<2X}{}}\xi (v_1)^\theta `$
$``$ $`X^{3/2+1/(2k2)+\epsilon }.`$
This completes the proof of Theorem 2.
|
warning/0506/hep-ph0506002.html
|
ar5iv
|
text
|
# Magnetism and superconductivity in quark matter
## 1 Introduction
QCD has been believed to be the basic theory of strong interaction and there are many successful consequences about the properties of hadrons and their interactions. Recently many studies have been devoted to figure out the phase diagram of QCD in temperature ($`T`$) - density ($`\rho _B`$) plane . At high temperature or high density, quarks confined inside hadrons should be liberated to form matter consisting of quarks and gluons (deconfinement transition). Such basic constituents, especially quarks exhibit interesting properties there as electrons in condensed matter through many-body dynamics; one of the interesting possibility is phase transition as temperature or density changes. When we emphasize the low $`T`$ and high $`\rho _B`$ region, the subjects are sometimes called high-density QCD. The main aims in this field should be to elucidate the new phases and their properties, and to extract their symmetry breaking pattern and low-energy excitation modes there on the basis of QCD. On the other hand, these studies have phenomenological implications on relativistic heavy-ion collisions and compact stars like neutron stars or quark stars .
Color superconductivity (CSC) should be very popular . Its mechanism is similar to the BCS theory for the electron-phonon system , in which the attractive interaction of electrons is provided by phonon exchange and causes the Cooper instability near the Fermi surface. As for quark matter, the quark-quark interaction is mediated by colored gluons, and is often approximated by some effective interactions, e.g., the one-gluon-exchange (OGE) or the instanton-induced interaction, both of which give rise to the attractive quark-quark interaction in the color anti-symmetric $`\overline{3}`$ channel. Many people believe that it is robust due to the Cooper instability even for small attractive quark-quark interaction in color $`\overline{3}`$ channel.
Here we’d like to address another interesting property of quark matter, magnetic properties of quark matter. We shall see various types of magnetic ordering may be expected in quark matter at finite density or temperature. They arise due to the quark particle-hole ($`ph`$) correlations in the pseudo-scalar or axial-vector channel.
Phenomenologically the concept of magnetism should be directly related to the origin of strong magnetic field observed in compact stars ; e.g., it amounts to $`O(10^{12}`$G) at the surface of radio pulsars. Recently a new class of pulsars called magnetars has been discovered with super strong magnetic field, $`B_s10^{1415}`$G, estimated from the $`P\dot{P}`$ curve . First observations are indirect evidences for the superstrong magnetic field, but discoveries of some absorption lines stemming from the cyclotron frequency of protons have been currently reported ; when it is confirmed that they originate from protons, they give a direct evidence for the superstrong magnetic field.
The origin of the strong magnetic field in compact stars has been a long standing problem since the first discovery of a pulsar . A simple working hypothesis is the conservation of the magnetic flux and its squeezing during the evolution from a main-sequence progenitor star to a compact star; $`BR^2`$ with $`R`$ being the radius. Taking the sun as a typical main-sequence star, we have $`B10^3`$G and $`R10^{10}`$cm. If it is squeezed to a typical radius of usual neutron stars, $`R10`$km, the conservation of the magnetic flux gives $`10^{11}`$G, which is consistent with the observations for radio pulsars. However, we find $`R100`$m to explain $`B10^{15}`$G observed for magnetars, which may lead to a contradiction since the Schwatzschild radius is $`O(1`$km) for the canonical mass of $`O(M_{})`$, which is much larger than $`R`$.
Since dense hadronic matter should be widely developed inside compact stars, it would be reasonable to inquire a microscopic origin of such strong magnetic field: ferromagnetism (FM) or spin polarization is one of the candidates to explain it. Makishima also suggested the hadronic origin of the magnetic field observed in binary X-ray pulsars or radio pulsars, since it looks no field decay in these objects.
When we consider the magnetic-interaction energy by a simple formula, $`E_{\mathrm{mag}}=\mu _iB`$ with the magnetic moment, $`\mu _i=e_i/(2m_i)`$, we can easily estimate it for $`B=O(10^{15}`$G) (see Table 2); it amounts to several MeV for electrons, while several keV for nucleons and 10 keV- 1MeV for quarks dependent on their mass.
This simple consideration may imply that strong interaction gives a feasible origin for the strong magnetic field, since its typical energy scale is MeV. The possibility of ferromagnetism in nuclear matter has been elaborately studied since the first discovery of pulsars, but negative results have been reported so far . We consider here its possibility in quark matter as an alternative in light of recent development of high-density QCD .
If FM is realized in quark matter, there should be some interplay with CSC; we examine a possibility of the coexistence of FM and CSC in quark matter, where we shall see an interplay between particle-particle and particle-hole correlations. As far as we know, interplay between the color superconducting phase and other phases characterized by the non-vanishing mean fields of the spinor bilinears has not been explored except for the case of chiral symmetry breaking .
It would be worth mentioning in this context that ferromagnetism (or spin polarization) and superconductivity are fundamental concepts in condensed matter physics, and their coexistent phase has been discussed and expected for a long time . As a recent progress, superconducting phases have been discovered in some ferromagnetic materials and many efforts have been made to understand the coexisting mechanism . In this phenomenon itinerant electrons may be responsible, while its mechanism is not fully elucidated yet. Many people believe that the electron Cooper pair should be $`P`$ wave, since this type can be compatible with spin polarization. We can easily consider the similar situation in quark matter. Since the volumes of the Fermi seas of quarks with different spins result in being different due to the net presence of magnetization, we could not construct a quark Cooper pair in a usual manner as $`J^P=0^+`$. Instead, we consider the $`J^P=0^{}`$ pairing with orbital angular momentum $`L=1`$ and total spin $`S=1`$. For the $`S=1`$ state we further consider two possibilities: spin-parallel pair or spin-anti-parallel pair. We first discuss the former case in detail, which may have a direct resemblance to the electron case. Subsequently we briefly sketch our idea about the former case. Anyway we shall see the gap functions become anisotropic in the momentum space like in <sup>3</sup>He or nuclear matter .
We discuss another magnetic aspect in quark matter at moderate densities, where the QCD interaction is still strong and some non-perturbative effects still remain. One of the most important phenomena observed there is restoration of chiral symmetry. In the vacuum chiral symmetry is spontaneously broken to give finite mass for quarks or nucleon; we may bear in mind such a picture that the vacuum is in a kind of superconducting phase with massless quark ($`q`$)-anti-quark ($`\overline{q}`$) pair condensate, and the gap opened at the top of the Dirac sea corresponds to the finite mass. As a consequence the vacuum does not possess chiral symmetry any more. At finite densities the suppression of $`\overline{q}q`$ excitations due to the existence of the Fermi sea gives rise to restoration of chiral symmetry at a certain density, and many people believe that deconfinement transition occurs at almost the same density.
There have been proposed various types of the p-h condensations at moderate densities , in which the p-h pair in scalar or tensor channel has the finite total momentum indicating standing waves (the chiral density waves). The instability for the density wave in quark matter was first discussed by Deryagin et al. at asymptotically high densities where the interaction is very weak, and they concluded that the density-wave instability prevails over the BCS one in the large $`N_c`$ (the number of colors) limit due to the dynamical suppression of colored BCS pairings.
In general, density waves are favored in 1-D (one spatial dimension) systems and have the wave number $`Q=2k_F`$ according to the Peierls instability , e.g., charge density waves (CDW) in quasi-1-D metals . The essence of its mechanism is the nesting of Fermi surfaces and the level repulsion (crossing) of single particle spectra due to the interaction for the finite wave number. Thus the low dimensionality has a essential role to produce the density-wave states. In the higher dimensional systems, however, the transitions occur provided the interaction of a corresponding (p-h) channel is strong enough. For the 3-D electron gas, it was shown by Overhauser that paramagnetic state is unstable with respect to the formation of the static spin density wave (SDW), in which spectra of up- and down-spin states are deformed to bring about the level crossing due to the Fock exchange interactions, while the wave number does not precisely coincide with $`2k_F`$ because of the incomplete nesting in higher dimension.
We shall see a kind of spin density wave develops there, in analogy with SDW mentioned above.. It occurs along with the chiral condensation and is represented by a dual standing wave in scalar and pseudo-scalar condensates (we have called it ‘dual chiral-density wave’, DCDW). DCDW has different features in comparison with the previously discussed chiral density waves . One outstanding feature concerns its magnetic aspect; DCDW induces spin density wave.
## 2 Ferromagnetism in QCD
### 2.1 A heuristic argument
Quark matter bears some resemblance to electron gas interacting with the Coulomb potential; the one gluon exchange (OGE) interaction in QCD has some resemblance to the Coulomb interaction in QED , and color neutrality of quark matter corresponds to total charge neutrality of electron gas under the background of positively charged ions.
It was Bloch who first suggested a mechanism leading to ferromagnetism of itinerant electrons within the Hartree-Fock approximation . The mechanism looks very simple but largely reflects the Fermion nature of electrons in a model-independent way. Since there works no direct interaction between charged particles as a whole, the Fock exchange interaction gives a leading contribution. Then it is immediately conceivable that a most attractive channel is the parallel spin pair, whereas the anti-parallel pair gives null contribution (see Eq. (8) below). This is nothing but a consequence of the Pauli exclusion principle: electrons with the same spin cannot closely approach to each other, which efficiently avoid the Coulomb repulsion. Thus a completely polarized state is favored by the interaction. On the other hand a polarized state should have a larger kinetic energy by rearranging the two Fermi spheres. Thus there is a trade-off between the kinetic and interaction energies, which leads to a spontaneous spin polarization (SSP) or FM at a certain density. Subsequently it has been proved that Bloch’s idea is qualitatively justified, but the critical density can not be reliably estimated without examining the higher-order correlation diagrams ; especially the ring diagrams have been known to be important in the calculation of the susceptibility of electron gas. Recently the possibility of ferromagnetism in the electron gas has been studied by the quantum Monte Carlo simulation and it has been shown that the electron gas is in ferromagnetic phase at very low electron density . Authors in ref. have confirmed it experimentally.
One of the essential points we learned here is that we need no spin-dependent interaction in the original Lagrangian to see SSP: a symmetry principle gives rise to a spin dependent interaction.
Then it might be natural to ask how about in QCD. We list here some features of QCD related to this subject. (1) the quark-gluon interaction in QCD is rather simple, compared with the nuclear force; it is a gauge interaction like in QED. (2) quark matter should be a color neutral system and only the $`\mathrm{𝐹𝑜𝑐𝑘}\mathrm{𝑒𝑥𝑐ℎ𝑎𝑛𝑔𝑒}`$ interaction is also relevant like in the electron system. (3) there is an additional flavor degree of freedom in quark matter; gluon exchange never change flavor but it becomes effective through the generalized Pauli principle. (4) quarks should be treated relativistically, different from the electron system.
The last feature requires a new definition and formulation of SSP or FM in relativistic systems since “spin” is no more a good quantum number for relativistic particles; spin couples with momentum and its direction changes during the motion. It is well known that the Pauli-Lubanski vector $`W^\mu `$ is the four vector to represent the spin degree of freedom in a covariant form; the spinor of the free Dirac equation is the eigenstate of the operator,
$$Wa=\frac{1}{2}\gamma _5a/k/,$$
(1)
where a 4-axial-vector $`a^\mu `$ is orthogonal to $`k`$ s.t.
$$𝐚=𝜻+\frac{𝐤(𝜻𝐤)}{m(E_k+m)},a^0=\frac{𝐤𝜻}{m}$$
(2)
with the axial vector $`𝜻`$. We can see that $`a^\mu `$ is reduced to a three vector $`(0,𝜻)`$ in the rest frame, where we can allocate $`𝜻=(0,0,\pm 1)`$ to spin “up” and “down” states. Thus we can still use $`𝜻`$ to specify the two intrinsic polarized states even in the general Lorentz frame. To characterize the degeneracy of the plane wave solution $`u^{(\alpha )}(k)`$ ($`\alpha =1,2`$) for a positive energy state, we can use such spinors $`u^{(\alpha )}(k)`$ that are eigenstates of the operator $`Wa/m_q`$: for the standard representation of $`u^{(\alpha )}(k)`$ ,
$$\frac{Wa}{m_q}u^{(\alpha )}(k)=\pm \frac{1}{2}u^{(\alpha )}(k).$$
(3)
Accordingly, the polarization density matrix $`\rho (k,𝜻)`$ is given by the expression,
$$\rho (k,𝜻)=\frac{1}{2m_q}(k/+m_q)P(a),P(a)=\frac{1}{2}(1+\gamma _5a/),$$
(4)
which is normalized by the condition, $`\mathrm{tr}\rho (k,𝜻)=1`$ .
Consider the spin-polarized quark liquid with the total number density of quarks $`n_q`$ <sup>1</sup><sup>1</sup>1We, hereafter, consider one flavor quark matter, since the OGE interaction never changes flavors. ; we denote the number densities of quarks with spin up and down by $`n_+`$ and $`n_{}`$, respectively , and introduce the polarization parameter $`p`$ by the equations, $`n_\pm =\frac{1}{2}n_q(1\pm p),`$ under the condition $`0p1`$. We assume as usual that three color states are occupied to be neutral for each momentum and spin state. The Fermi momenta in the spin-polarized quark matter are then $`k_F^\pm =k_F(1\pm p)^{1/3}`$ with $`k_F=(\pi ^2n_q)^{1/3}`$. The kinetic energy density is given by the standard formula,
$$ϵ_{kin}=\frac{3}{16\pi ^2}\underset{i=\pm }{}\left[k_F^iE_F^i(2k_F^{i2}+m_q^2)m_q^4\mathrm{ln}\left(\frac{E_F^i+k_F^i}{m_q}\right)\right],$$
(5)
with the Fermi energy $`E_F^i=(m_q^2+k_F^{i2})^{1/2}`$.
Let us consider the OGE interaction between two quarks with momenta, $`k`$ and $`q`$, and spin vectors, $`𝜻`$ and $`𝜻^{}`$, respectively. The color symmetric matrix element $`_{𝐤\zeta ,𝐪\zeta ^{}}^s`$ is given only by the exchange term; the direct term vanishes because the color symmetric combinations ($`\mathrm{tr}\lambda _a`$) does not couple to gluons. Thus
$`_{𝐤\zeta ,𝐪\zeta ^{}}^s`$ $`=`$ $`g^2{\displaystyle \frac{1}{9}}\mathrm{tr}(\lambda _a/2\lambda _a/2)\overline{u}^{(\zeta ^{})}(𝐪)\gamma _\mu u^{(\zeta )}(𝐩)\overline{u}^{(\zeta )}(𝐩)\gamma ^\mu u^{(\zeta ^{})}(𝐪){\displaystyle \frac{1}{(kq)^2}}`$ (6)
$`=`$ $`{\displaystyle \frac{4}{9}}g^2{\displaystyle \frac{1}{4}}\mathrm{tr}\left[\gamma _\mu \rho (k,\zeta )\gamma ^\mu \rho (q,\zeta ^{})\right]{\displaystyle \frac{1}{(kq)^2}},`$
by the use of Eq. (4). If we choose both $`𝜻`$ and $`𝜻^{}`$ in parallel along the $`z`$ axis, $`𝜻=𝜻^{}=(0,0,\pm 1)`$, we have the spin-nonflip amplitude $`_{\mathrm{𝐩𝐪}}^{s,nonflip}`$, while if we choose them in anti-parallel, $`𝜻=𝜻^{}`$, we have the spin-flip amplitude $`_{\mathrm{𝐩𝐪}}^{s,flip}`$. Each form of the spin-nonflip or spin-flip amplitude is complicated, but their average gives a simple form,
$$\overline{}_{\mathrm{𝐩𝐪}}^s=\frac{2}{9}g^2\frac{2m_q^2kq}{(kq)^2},$$
(7)
which is nothing but the matrix element for the unpolarized case . In the nonrelativistic limit, $`m_q|𝐩|,|𝐪|`$, the matrix element is reduced to the form,
$$_{𝐤\zeta ,𝐪\zeta ^{}}^s=\frac{2}{9}g^2\frac{m_q^2(1+𝜻𝜻^{})}{|𝐤𝐪|^2},$$
(8)
so that there is no correlation between quarks with different spins. On the other hand, there is some correlation included in the relativistic case.
After summing up over the color degree of freedom and performing the integrals of the color symmetric matrix element $`_{𝐤\zeta ,𝐪\zeta ^{}}^s`$ over the Fermi seas of spin up and down quarks, we have the exchange energy density $`ϵ_{ex}`$ consisting of two contributions,
$$ϵ_{ex}=ϵ_{ex}^{nonflip}+ϵ_{ex}^{flip}.$$
(9)
In the nonrelativistic case, the spin-flip contribution becomes tiny and the dominant contribution for the OGE energy density in Eq.(9) comes from the spin-nonflip contribution (see Eq. (8)),
$$ϵ_{ex}\frac{\alpha _ck_F^4}{2\pi ^3}\left\{(1+p)^{4/3}+(1p)^{4/3}\right\}.$$
(10)
The exchange energy is negative and takes a minimum at $`p=1`$. The form of the energy density (10) is exactly the same as in electron gas. It is the difference of density dependence between the contributions given in Eqs. (5) and (10) which causes a ferromagnetic instability; this mechanism was first pointed out by Bloch for electron gas .
In the relativistic case there are some different features from the nonrelativistic case. First, there is a spin-flip contribution due to the lower component of the Dirac spinor even for the Coulomb-like interaction. Secondly, the transverse (magnetic) gluons becomes important, where the spin-flip effect is prominent. Finally, the density dependence of kinetic energy as well as the exchange energy is very different . Before discussing the general case, we consider the relativistic limit, $`k_F^im_q`$; the Fock exchange-energy density looks like
$$ϵ_{ex}\frac{\alpha _c}{8\pi ^3}k_F^4\left\{(1+p)^{4/3}+(1p)^{4/3}+2(1p^2)^{2/3}\right\},$$
(11)
which is a decreasing function and takes a minimum again at $`p=1`$. This is due to the characteristic feature of the spin-flip and spin-nonflip interactions: both give a repulsive contribution in the relativistic limit and there is no spin-flip interaction in the polarized state ($`p=1`$). Thus, ferromagnetism in the relativistic limit arises by a different mechanism from that in the nonrelativistic case.
In Fig.1 a typical shape of the total energy density, $`ϵ_{tot}=ϵ_{kin}+ϵ_{ex}`$, is depicted as a function of the polarisation parameter $`p`$, e.g. for the parameter set ,$`m_q=300`$MeV of the $`s`$ quark and $`\alpha _c=2.2`$ as in the MIT bag model <sup>2</sup><sup>2</sup>2The difficulties to determine the values of these parameters have been discussed in ref. , and we must allow some range for them.. We can see that paramagnetic quark matter ($`p=0`$) becomes unastable as density decreases, and ferromagnetic phase is favored at a certain density between 0.1 and 0.2 fm<sup>-3</sup>. This phase transition is of weakly first-order and the completely polarized ($`p=1`$) state appears at the critical density.
To figure out the features of the ferromagnetic transition, we study other quantities. For small $`p1`$, the energy density behaves like
$$ϵ_{tot}ϵ_{tot}(p=0)=\chi ^1p^2+O(p^4)$$
(12)
with $`\chi ^1\chi _{kin}^1+\chi _{ex}^1`$. $`\chi `$ is proportional to the magnetic susceptibility, and its sign change indicates a ferromagnetic transition, if it is of the second order. It consists of two contributions: the kinetic energy gives $`\chi _{kin}^1=k_F^5/(3\pi ^2E_F)`$ (c.f. the Pauli paramagnetism), which changes from $`\chi _{kin}^1O(k_F^5)`$ at low densities to $`\chi _{kin}^1O(k_F^4)`$ at high densities. On the other hand, the Fock exchange energy gives
$$\chi _{ex}^1=\frac{2\alpha _ck_F^4}{9\pi ^3}\left[2\frac{k_F^2}{E_F^2}\frac{3m_q^2k_F}{E_F^3}\mathrm{ln}\left(\frac{E_F+k_F}{m_q}\right)+\frac{4m_qk_F^2}{3E_F^2(E_F+m_q)}\right]$$
(13)
, which is reduced to
$$\chi _{ex}^1\frac{4\alpha _c}{9\pi ^3}k_F^4$$
(14)
in the nonrelativistic limit, $`p_Fm_q`$. In the relativistic limit, $`p_Fm_q`$, it behaves like
$$\chi _{ex}^1\frac{\alpha _c}{9\pi ^3}k_F^4\frac{\alpha _c}{3\pi ^3}k_F^4=\frac{2\alpha _c}{9\pi ^3}k_F^4,$$
(15)
where the first term stems from the spin-nonflip contribution, while the second term from the spin-flip contribution. Then we can see that the effect of the spin-flip contribution overwhelms the one of the spin-nonflip contribution. The interaction contribution $`\chi _{ex}^1`$ is always negative , and dominant over $`\chi _{kin}^1`$ at low densities, while the kinetic contribution $`\chi _{kin}`$ is always positive. If $`\alpha _c>3\pi /2=4.7`$, $`\chi `$ becomes negative over all densities.
For a given set of $`m_q`$ and $`\alpha _c`$, $`\chi `$ changes its sign at a certain density, denoted by $`n_{c1}`$, and it is a signal for the second-order phase transition. Note that the ferromagnetic transition in our case is of the first order, so that it is not sufficient to only see the magnetic susceptibility; even above that density the ferromagnetic phase may be possible. Actually there is a range, $`n_{c1}<n_q<n_{c2}`$, where $`\chi >0`$ but $`ϵ<0`$.
Above the density $`n_{c2}`$ there is no longer the stable ferromagnetic phase. However, the metastable state is still possible up to the density $`n_{c3}`$, which is specified by the condition s.t. $`\eta ϵ_{tot}/p|_{p=1}<0`$. In Fig.2 we depict the quantities $`\chi ^1,\delta ϵ`$ and $`\eta `$ as the functions of density, e.g. for the set ,$`m_q=300`$MeV and $`\alpha _c=2.2`$. The crossing points with the horizontal axis indicate the critical deisities $`n_{c1},n_{c2}`$ and $`n_{c3}`$, respectively. We can see that the ferromagnetic instability occurs at low densities, while the metastable state can exist up to rather high densities.
Finally we show the critical lines satisfying $`\chi ^1=0,\delta ϵ=0`$ and $`\eta =0`$ in the QCD parameter ($`\alpha _c`$ and $`m_q`$) plane, which seperate the three characteristic regions for a given density. In Fig. 3 we demonstrate them at a density $`n_q=0.3`$fm<sup>-3</sup>. All the lines have the maxima around the medium quark mass, and the mechanism of ferromagnetism is different for each side of the maximum, as already discussed. If we take $`m_q=300`$MeV for the $`s`$ quark or $`m_q0`$MeV for the $`u`$ or $`d`$ quark, and $`\alpha _c=2.2`$ as in the MIT bag model again , the quark liquid can be ferromagnetic as a metastable state.
We have seen that the ferromagnetic phase is realized at low densities and the metastable state is plausible up to rather high densities for a reasonable range of the QCD parameters. Our calculation is based on the lowest-order perturbation. So we need to examine the higher-order gluon-exchange contributions to confirm the possibility. It should be interesting to refer a recent paper , where the author also found the ferromagnetic transition at low densities within the perturbative QCD calculation beyond the lowest-order diagram.
If a ferromagnetic quark liquid exists stably or metastably around or above nuclear density, it has some implications on the properties of strange quark stars and strange quark nuggets . They should be magnetized in a macroscopic scale. Considering a possibility to attribute magnetars to strange quark stars in a ferromagnetic phase, we roughly estimate the strength of the magnetic field at the surface of a strange quark star. Taking the stellar parameters of strange quark stars to be similar to those for canonical neutron stars with the typical mass around $`M_G=1.4M_{}`$, we find the total magnetic dipole moment $`M_q`$, $`M_q=\mu _q(4\pi /3r_q^3)n_q`$ for the quark sphere with the density $`n_q`$ and the radius $`r_q`$, where $`\mu _q`$ is the magnetic moment of each quark. Then the dipolar magnetic field at the star surface $`r=R10`$km takes a maximal strength at the poles,
$$B_{\mathrm{max}}=\frac{8\pi }{3}\left(\frac{r_q}{R}\right)^3\mu _qn_q=10^{15}[\mathrm{G}]\left(\frac{r_q}{R}\right)^3\left(\frac{\mu _q}{\mu _N}\right)\left(\frac{n_q}{0.1\mathrm{fm}^3}\right)$$
(16)
with nuclear magneton $`\mu _N`$, which looks enough for magnetars.
### 2.2 Self-consistent calculation
If we understand FM or magnetic properties of quark matter more deeply, we must proceeds to a self-consistent approach, like the Hartree-Fock theory, beyond the previous perturbative argument <sup>3</sup><sup>3</sup>3Simple plane wave is the solution of the Hartree-Fock equation in the nonrelativistic electron gas, while it is not in quark matter. .
We begin with an OGE action:
$`I_{int}=g^2{\displaystyle \frac{1}{2}}{\displaystyle \mathrm{d}^4x\mathrm{d}^4y\left[\overline{\psi }(x)\gamma ^\mu \frac{\lambda _a}{2}\psi (x)\right]D_{\mu \nu }(x,y)\left[\overline{\psi }(y)\gamma ^\nu \frac{\lambda _a}{2}\psi (y)\right]},`$ (17)
where $`D^{\mu \nu }`$ denotes the gluon propagator. By way of the mean-field approximation, we have
$$I_{MF}=\frac{\mathrm{d}^4p}{(2\pi )^4}\overline{\psi }(p)G_A^1(p)\psi (p).$$
(18)
The inverse quark Green function $`G_A^1(p)`$ involves various self-energy (mean-field) terms, of which we only keep the color singlet particle-hole mean-field $`V(p)`$,
$$G_A(p)^1=/pm+/\mu +V(p).$$
(19)
Taking into account the lowest diagram, we can then write down the self-consistent equations for the mean-field, $`V`$:
$`V(k)=(ig)^2{\displaystyle \frac{\mathrm{d}^4p}{i(2\pi )^4}\{iD^{\mu \nu }(kp)\}\underset{(A)}{\underset{}{\gamma _\mu \frac{\lambda _\alpha }{2}\{iG_A(p)\}\gamma _\nu \frac{\lambda _\alpha }{2}}}}.`$ (20)
Applying the Fierz transformation for the OGE action (17) we can see that there appear the color-singlet scalar, pseudo-scalar, vector and axial-vector self-energies by the Fock exchange interaction. Taking the Feynman gauge for the gluon propagator, a manupilation gives
$`(A)`$ $`=`$ $`{\displaystyle \frac{N_c^21}{4N_c^2}}{\displaystyle \frac{1}{N_f}}\left\{\mathrm{Tr}(G_A)+i\gamma _5\mathrm{Tr}(G_Ai\gamma _5){\displaystyle \frac{1}{2}}[\gamma ^\mu \mathrm{Tr}(G_A\gamma _\mu )+\gamma _5\gamma ^\mu \mathrm{Tr}(G_A\gamma _5\gamma _\mu )]\right\}`$ (21)
$`+`$ $`\{\mathrm{color}\mathrm{non}\mathrm{singlet}\mathrm{or}\mathrm{flavor}\mathrm{non}\mathrm{singlet}\mathrm{terms}\}.`$
When we restrict the ground state to be an eigenstate with respect to color and flavor, there is only left the first term which is color singlet and flavor singlet. Still we must take into account various mean-fields in $`V`$, $`V=U_s+i\gamma _5U_{ps}+\gamma _\mu U_v^\mu +\gamma _\mu \gamma _5U_{av}^\mu `$ with the mean-fields $`U_i`$. Here we only retain $`𝐔_{av}(𝐔_A)`$ for simplicity and suppose that others to be vanished;
$$V(k)=𝜸\gamma _5𝐔_A(𝐤),$$
(22)
with the static axial-vector mean-field $`U_A(𝐤)`$.
The poles of $`G_A(p)`$, $`det`$$`G_A^1`$($`p_0+\mu `$$`=`$$`ϵ_n`$)$`=`$$`0`$, give the single-particle energy spectrum:
$`ϵ_n=\pm ϵ_\pm `$ (23)
$`ϵ_\pm (𝐩)=\sqrt{𝐩^2+𝐔_A^2(𝐩)+m^2\pm 2\sqrt{m^2𝐔_A^2(𝐩)+(𝐩𝐔_A(𝐩))^2}},`$ (24)
where the subscript in $`ϵ_s(𝐩),s=\pm `$ represents spin degrees of freedom, and the dissolution of the degeneracy corresponds to the exchange splitting of different “spin” states; the spectrum is reduced to a familiar form $`ϵ_\pm m+\frac{p^2}{2m}\pm |𝐔_A|`$ in the non-relativistic limit .
There appear two Fermi seas with different volumes for a given quark number due to the exchange splitting in the energy spectrum. The appearance of the rotation symmetry breaking term, $`𝐩𝐔_A`$ in the energy spectrum implies deformation of the Fermi sea: thus rotation symmetry is violated in the momentum space as well as the coordinate space, $`O(3)O(2)`$. Accordingly the Fermi sea of majority quarks exhibits a “prolate” shape ($`F^{}`$), while that of minority quarks an “oblate” shape ($`F^+`$) as seen in Fig. 4.
Then the self-consistent equation (20) is reduced to the form,
$$U_A(𝐤)=\frac{N_c^21}{4N_c}g^2\frac{\mathrm{d}^3p}{(2\pi )^3}\underset{s=\pm }{}\frac{1}{ϵ_s(𝒑)^2|𝐤𝐩|^2}\theta (\mu ϵ_s(𝒑))\frac{U_A(𝐩)+s\beta _p}{ϵ_s(𝒑)}$$
(25)
with $`\beta _p=\sqrt{p_z^2+m^2}`$ by taking $`𝐔_A`$ along the $`z`$ axis. Here we have discarded the contribution of the Dirac seas and only taken into account that of the Fermi seas.
In the following we demonstrate some numerical results by replacing the original OGE by the “contact” (zero-range) interaction, $`D^{\mu \nu }g^{\mu \nu }/\mathrm{\Lambda }^2,`$ which may correspond to the Stoner model in the condensed matter physics . <sup>4</sup><sup>4</sup>4When we take into account the Debye screening, the time component of the gluon propagator becomes finite range due to the Debye screening. If typical momentum transfer $`Q`$ is much smaller than the screening mass $`M_D^2N_fg^2\mu ^2/(2\pi ^2)`$ , we may replace the OGE with the infinite range by the zero-range effective interaction.
We can easily see that the mean-field $`U_A`$ becomes then momentum-independent, and the expression for $`U_A`$, Eq. (25), is proportional to the simple sum of the expectation value of the spin operator over the Fermi seas;
$`\overline{s}_z={\displaystyle \frac{1}{2}}\mathrm{\Sigma }_z`$ $`=`$ $`i{\displaystyle _C}{\displaystyle \frac{d^4p}{(2\pi )^4}}\mathrm{tr}\gamma _5\gamma _3G_A(p)`$ (26)
$`=`$ $`{\displaystyle \frac{1}{2}}\left[{\displaystyle _{F^+}}{\displaystyle \frac{d^3p}{(2\pi )^3}}{\displaystyle \frac{U_A(𝐩)+\beta _p}{ϵ_+(𝐩)}}+{\displaystyle _F^{}}{\displaystyle \frac{d^3p}{(2\pi )^3}}{\displaystyle \frac{U_A(𝐩)\beta _p}{ϵ_{}(𝐩)}}\right].`$
### 2.3 Phase diagram on the temperature-density plane
We will present the phase diagram in the three-flavor case under two conditions : the chemical equilibrium condition (CEC) $`\mu _u=\mu _d=\mu _s`$ and the charge neutral condition without electrons (CNC) $`\rho _u=\rho _d=\rho _s`$, where quark masses are taken as $`m_u=m_d=5`$MeV and $`m_s=150350`$MeV, i.e., $`\mu _s=\sqrt{\mu _{u,d}^2+m_s^2m_{u,d}^2}`$ for $`T=0`$. In both conditions, since the spin polarization caused by the axial-vector mean-field is fully enhanced by the quark mass for given density or temperature, choice of the current quark mass seriously affects the results; especially, largeness of the strange quark mass has an essential effect on spin polarization. To get the phase diagram or critical line on the temperature-density plane, we use the thermodynamic potential $`\mathrm{\Omega }`$ within the mean-field approximation,
$`\mathrm{\Omega }`$ $`=`$ $`N_c{\displaystyle \underset{B=\pm 1}{}}{\displaystyle \underset{s=\pm }{}}{\displaystyle \underset{i=u,d,s}{}}{\displaystyle \frac{\mathrm{d}^3𝐤}{(2\pi )^3}T\mathrm{log}\left\{\mathrm{exp}\left[\frac{ϵ_s(𝐤,m_i,U_A)B\mu _i}{T}\right]+1\right\}}`$ (27)
$`N_c{\displaystyle \underset{s=\pm }{}}{\displaystyle \underset{i=u,d,s}{}}{\displaystyle \frac{\mathrm{d}^3𝐤}{(2\pi )^3}ϵ_s(𝐤,m_i,U_A)}+{\displaystyle \frac{U_A^2}{4\stackrel{~}{g}^2}},`$
where we have used the “contact” interaction, $`\stackrel{~}{g}^2g^2/\mathrm{\Lambda }^2`$, in place of the OGE interaction. Note that we take into accout the vacuum contribution in this formula (the second term in Eq. (27)), which should be regularized by ,e.g., the proper-time method (see §4.3). We can confirm that the thermodynamic potential reproduces the self-consistent equation for the order parameter $`U_A`$ Eq. (25) in the three-flavor case, except the vacuum contribution. The vacuum (the Dirac sea) contribution always works against spin polarization as it should do, while the contribution of the Fermi sea gives rise to spontaneous spin polarization.
Fig. 5 shows the critical temperature (the Curie temperature) as a function of baryon-number density under the two conditions mentioned above; CEC and CNC. We can see that CNC tends to facilitate the system having spin polarization than CEC. This is because CNC holds the larger strange-quark density than CEC.
Since the axial-vector mean-field arises from the Fock exchange interaction among quarks in the Fermi sea and causes a kind of particle-hole condensation, there exists a critical density for a given coupling constant $`\stackrel{~}{g}`$. We show the critical density by varying the effective coupling constant $`\stackrel{~}{g}`$ in Fig. 6.
The critical density is more lowered with the larger coupling strength, and this tendency is remarkable in the case of CNC. The result also indicates that even for the weak-coupling regime in QCD, spin polarization may appear at sufficiently large densities and low temperatures.
## 3 Color magnetic superconductivity
### 3.1 General framework
If FM is realized in quark matter, it might be in the CSC phase. In this section we discuss a possibility of the coexistence of FM and CSC, which we call Color magnetic superconductivity .
Recall the OGE action Eq. (17). By way of the mean-field approximation, we have
$$I_{MF}=\frac{1}{2}\frac{\mathrm{d}^4p}{(2\pi )^4}\left(\begin{array}{c}\overline{\psi }(p)\hfill \\ \overline{\psi }_c(p)\hfill \end{array}\right)^TG^1(p)\left(\begin{array}{c}\psi (p)\hfill \\ \psi _c(p)\hfill \end{array}\right)$$
(28)
in the Nambu-Gorkov formalism, allowing not only the particle-hole but also the particle-particle mean-field. The inverse quark Green function $`G^1(p)`$ involves various self-energy (mean-field) terms, of which we only keep the color singlet particle-hole $`V(p)`$ and color $`\overline{3}`$ particle-particle ($`\mathrm{\Delta }`$) mean-fields; the former is responsible to ferromagnetism, while the latter to superconductivity,
$`G^1(p)`$ $`=`$ $`\left(\begin{array}{cc}/pm+/\mu +V(p)& \gamma _0\mathrm{\Delta }^{}(p)\gamma _0\\ \mathrm{\Delta }(p)& /pm/\mu +\overline{V}(p)\end{array}\right),`$ (31)
$`=`$ $`\left(\begin{array}{cc}G_{11}(p)& G_{12}(p)\\ G_{21}(p)& G_{22}(p)\end{array}\right)^1`$ (34)
where
$$\psi _c(k)=C\overline{\psi }^T(k),\overline{V}CV^TC^1.$$
(35)
Taking into account the lowest diagram, we can then write down the self-consistent equations for the mean-fields, $`V`$ and $`\mathrm{\Delta }`$:
$`V(k)=(ig)^2{\displaystyle \frac{\mathrm{d}^4p}{i(2\pi )^4}\{iD^{\mu \nu }(kp)\}\gamma _\mu \frac{\lambda _\alpha }{2}\{iG_{11}(p)\}\gamma _\nu \frac{\lambda _\alpha }{2}}.`$ (36)
and
$$\mathrm{\Delta }(k)=(ig)^2\frac{\mathrm{d}^4p}{i(2\pi )^4}\{iD^{\mu \nu }(kp)\}\gamma _\mu \frac{(\lambda _\alpha )^T}{2}\{iG_{21}(p)\}\gamma _\nu \frac{\lambda _\alpha }{2},$$
(37)
(c.f. Eq. (20)).
The structure of Eq. (36) is the same as Eq. (20), and we can see there appear the color-singlet scalar, pseudoscalar, vector and axial-vector self-energies by applying the Fierz transformation . Here we retain only $`U_s,U_v^0,U_{av}^3`$ in $`V`$ and suppose that others to be vanished as before. We shall see this ansatz gives self-consistent solutions for Eq.(36) within the zero-range approximation for the OGE interaction because of axial and reflection symmetries of the Fermi seas. We furthermore discard the scalar mean-field $`U_s`$ and the time component of the vector mean-field $`U_v^0`$ for simplicity since they are irrelevant for the spin degree of freedom.
According to the above assumptions and considerations the mean-field $`V`$ in Eq.(34) renders
$$V=\gamma _3\gamma _5U_A,U_AU_{av}^3,$$
(38)
with the axial-vector mean-field $`U_A`$, as in Eq. (22). Then the diagonal component of the Green function $`G_{11}(p)`$ is written as
$$G_{11}(p)=\left[G_A^1\gamma _0\mathrm{\Delta }^{}\gamma _0\stackrel{~}{G}_A\mathrm{\Delta }\right]^1$$
(39)
with
$`G_A^1(p)`$ $`=`$ $`/pm+/\mu \gamma _5\gamma _3U_A,`$ (40)
$`\stackrel{~}{G}_A^1(p)`$ $`=`$ $`/pm/\mu \overline{\gamma _5\gamma _3}U_A,`$ (41)
where $`\overline{\gamma _5\gamma _3}=\gamma _5\gamma _3`$ and $`G_A(p)`$ is the Green function with $`U_A`$ which is determined self-consistently by way of Eq. (20).
### 3.2 $`{}_{}{}^{3}P`$ type anisotropic pairing
Before constructing the gap function $`\mathrm{\Delta }`$, we first find the single-particle spectrum and their eigenspinors in the absence of $`\mathrm{\Delta }`$, which is achieved by diagonalization of the operator $`G_A^1`$. We have already known four single-particle energies $`ϵ_\pm `$ (positive energies) and $`ϵ_\pm `$ (negative energies), which are given as
$`ϵ_\pm (𝒑)=\sqrt{𝒑^2+U_A^2+m^2\pm 2U_A\sqrt{m^2+p_z^2}},`$ (42)
and the eigenspinors $`\varphi _s,s=\pm `$ should satisfy the equation, $`G_A^1(ϵ_s,𝐩)\varphi _s=0`$.
Here we take the following ansatz for $`\mathrm{\Delta }`$:
$`\mathrm{\Delta }(𝒑)`$ $`=`$ $`{\displaystyle \underset{s=\pm }{}}\stackrel{~}{\mathrm{\Delta }}_s(𝒑)B_s(𝒑),`$
$`B_s(𝒑)`$ $`=`$ $`\gamma _0\varphi _s(𝒑)\varphi _s^{}(𝒑).`$ (43)
The structure of the gap function (43) is then inspired by a physical consideration of a quark pair as in the usual BCS theory: we consider here the quark pair on each Fermi surface with opposite momenta, $`𝐩`$ and $`𝐩`$ so that they result in a linear combination of $`J^\pi =0^{},1^{}`$ (see Fig. 9) <sup>5</sup><sup>5</sup>5Recently spin-one color superconductivity has been also studied in the normal matter . .
$`\stackrel{~}{\mathrm{\Delta }}_s`$ is still a matrix in the color-flavor space. Since the anti-symmetric nature of the fermion self-energy imposes a constraint on the gap function ,
$`C\mathrm{\Delta }(𝒑)C^1=\mathrm{\Delta }^T(𝒑).`$ (44)
$`\stackrel{~}{\mathrm{\Delta }}_n(𝒑)`$ must be a symmetric matrix in the spaces of internal degrees of freedom. Taking into account the property that the most attractive channel of the OGE interaction is the color anti-symmetric $`\overline{3}`$ state, it must be in the flavor singlet state. Thus we can choose the form of the gap function as
$$\left(\stackrel{~}{\mathrm{\Delta }}_s\right)_{\alpha \beta ;ij}=ϵ^{\alpha \beta 3}ϵ^{ij}\mathrm{\Delta }_s$$
(45)
for the two-flavor case (2SC), where $`\alpha ,\beta `$ denote the color indices and $`i,j`$ the flavor indices. Then the quasi-particle spectrum can be obtained by looking for poles of the diagonal Green function, $`G_{11}`$:
$`E_s(𝒑)`$ $`=\{\begin{array}{cc}\sqrt{(ϵ_s(𝒑)\mu )^2+|\mathrm{\Delta }_s(𝒑)|^2}\hfill & \text{for color 1, 2}\hfill \\ \sqrt{(ϵ_s(𝒑)\mu )^2}\hfill & \text{for color 3}\hfill \end{array}`$ (48)
Note that the quasi-particle energy is independent of color and flavor in this case, since we have assumed a singlet pair in flavor and color.
Gathering all these stuffs to put them in the self-consistent equations, we have the coupled gap equations for $`\mathrm{\Delta }_s`$,
$`\mathrm{\Delta }_s^{}(k,\theta _k)={\displaystyle \frac{N_c+1}{2N_c}}\stackrel{~}{g}^2{\displaystyle \frac{\mathrm{d}p\mathrm{d}\theta _p}{(2\pi )^2}p^2\mathrm{sin}\theta _p\underset{s}{}T_{s^{}s}(k,\theta _k,p,\theta _p)\frac{\mathrm{\Delta }_s(p,\theta _p)}{2E_s(p,\theta _p)}},`$ (49)
and the equation for $`U_A`$,
$$U_A=\frac{N_c^21}{4N_c^2}\stackrel{~}{g}^2\frac{\mathrm{d}^3p}{(2\pi )^3}\underset{s}{}\left[\theta (\mu ϵ_s(𝒑))+2v_s^2(𝒑)\right]\frac{U_A+s\beta _p}{ϵ_s(𝒑)},$$
(50)
within the “contact” interaction, $`\stackrel{~}{g}^2g^2/\mathrm{\Lambda }^2`$, where $`v_s^2(𝒑)`$ denotes the momentum distribution of the quasi-particles. We find that the expression for $`U_A`$, Eq. (50), is nothing but the simple sum of the expectation value of the spin operator with the weight of the occupation probability of the quasi-particles $`v_s^2`$ for two colors and the step function for remaining one color (cf. (25)).
Carefully analyzing the structure of the function $`T_{s^{}s}`$ in Eq. (49), we can easily find that the gap function $`\mathrm{\Delta }_s`$ should have the polar angle ($`\theta `$) dependence on the Fermi surface,
$$\mathrm{\Delta }_s(p_s^F,\theta )=\frac{p_s^F(\theta )\mathrm{sin}\theta }{\mu }\left[s\frac{m}{\sqrt{m^2+(p_s^F(\theta )\mathrm{cos}\theta )^2}}R+F\right],$$
(51)
with constants $`F`$ and $`R`$ to be determined (see Fig. 9).
As a characteristic feature, both the gap functions have nodes at poles ($`\theta =0,\pi `$) and take the maximal values at the vicinity of equator ($`\theta =\pi /2`$), keeping the relation, $`\mathrm{\Delta }_{}\mathrm{\Delta }_+`$. This feature is very similar to $`{}_{}{}^{3}P`$ pairing in liquid <sup>3</sup>He or nuclear matter ; actually we can see our pairing function Eq. (51) to exhibit an effective $`P`$ wave nature by a genuine relativistic effect by the Dirac spinors . Accordingly the quasi-particle distribution is diffused (see Fig. 9)
We demonstrate some self-consistent solutions here. Since we have little information to determine the values of the parameters $`\stackrel{~}{g}`$ and $`\delta `$ (there may be other more reasonable form factors than the present cut-off function), and our purpose is to figure out qualitative properties of spin polarization in the color superconducting phase, we mainly set in the following calculations them as $`\stackrel{~}{g}=0.13`$ MeV<sup>-1</sup> and $`\delta =0.1\mu `$, for example, which is not so far from the couplings in NJL-like models .
We first examine spin polarization in the absence of CSC. In Fig. 10 we show the the axial-vector mean-field $`U_A`$, with $`\mathrm{\Delta }_\pm `$ being set to be zero, as a function of baryon number density $`\rho _B(\rho _q/3)`$ relative to the normal nuclear density $`\rho _0=0.16`$ fm<sup>-3</sup> for $`m=1425`$ MeV (dashed lines). It is seen that the axial-vector mean-field (spin polarization) appears above a critical density and becomes larger as baryon number density gets higher. Moreover, the results for different values of the quark mass show that spin polarization grows more for the larger quark mass. This is because a large quark mass gives rise to much difference in the Fermi seas of two opposite “spin” states, which leads to growth of the exchange energy in the axial-vector channel.
Next we solve the coupled equations (49) and (50) with Eq. (51). Results for $`U_A`$, $`R`$ and $`F`$ are shown in Fig. 10 (solid lines)
As a consequence, we can say that FM and CSC barely interfere with each other .
### 3.3 Another possibility - Gapless type pairing
Nowadays there have been many studies about the pairing of quarks in the two Fermi spheres with different sizes, which is caused by the mass and charge differences among three-flavor quarks. It is well known that fermion pairing between two different Fermi surfaces gives rise to the LOFF phase or the gapless superconducting phase . There have been discussions about the phase separation and the mixed phase in these context .
In the presence of magnetization we have seen that there are two Fermi seas with different size and deformation, depending on the spin polarization. So we can consider another pairing than the previous one: two quarks with opposite momenta and polarizations with each other take part in the pairing . Introducing the following notation,
$$ϵ_n=\{ϵ_{},ϵ_+,ϵ_{}ϵ_+\}(n=14)$$
(52)
for the single quark energy by using Eq. (42). The pairing function can be written in the similar form to Eq. (43)
$$\mathrm{\Delta }(𝒑)=\underset{n=1}{\overset{4}{}}\stackrel{~}{\mathrm{\Delta }}_n(𝒑)B_n(𝒑)$$
(53)
with
$$B_n(𝒑)=\gamma _0\varphi _{\stackrel{~}{n}}(𝒑)\varphi _n^{}(𝒑),$$
(54)
where $`\varphi _{\stackrel{~}{n}}(𝒑)`$ is defined by $`\varphi _{\stackrel{~}{n}}(𝒑)\varphi _{1+(1)^n}(𝒑)`$. As a combination of the quark pair in the color and flavor spaces, we assume it to be anti-symmetric in both spaces,
$$\left[\mathrm{\Delta }_n(𝒑)\right]_{\alpha \beta ;ij}=ϵ_{\alpha \beta 3}ϵ_{ij}\mathrm{\Delta }_n(𝒑)$$
(55)
as before. Then the quasi-particle energy is given by
$`E_n(𝒑)_\pm `$ $`=\{\begin{array}{cc}E_n^A\pm \sqrt{(E_n^S)^2+|\mathrm{\Delta }_n(𝒑)|^2}\hfill & \text{for color 1, 2}\hfill \\ \pm \sqrt{(ϵ_n(𝒑)\mu )^2}\hfill & \text{for color 3}\hfill \end{array}`$ (58)
with
$$E_n^{S,A}=\frac{(ϵ_n\mu )(ϵ_{\stackrel{~}{n}}\mu )}{2},$$
(59)
which clearly exhibits a gapless excitation. We can also see that the gap function shows $`\mathrm{cos}\theta `$-like dependence on the Cooper surface defined by the equation, $`E_n^S(p,\theta )=0`$. These features resemble those given in ref., where the electron pairing with spin anti-parallel component of the $`S=1`$ triplet is considered in the presence of magnetization.
## 4 Dual chiral density wave
### 4.1 Chiral symmetry restoration and Instability of the directional mode
We consider here another type of magnetism in quark matter at moderate densities, which is closely connected with chiral symmetry. We shall see that the ground state in the spontaneously symmetry breaking (SSB) phase becomes unstable with respect to producing a density wave. Accordingly the quark magnetic moment spatially oscillates and a kind of spin density wave is induced. The density wave can be described as a dual standing wave in the scalar and pseudo-scalar densities , where they spatially oscillate in the phase difference of $`\pi /2`$ to each other. It is well known that chiral symmetry is spontaneously broken due to the quark ($`q`$)-anti-quark ($`\overline{q}`$) pair condensate in the vacuum and at low densities; since we take the vacuum as an eigenstate of parity operation, only the scalar density is non vanishing to generate finite mass of quarks. Geometrically both the scalar and pseudo-scalar densities always reside on the chiral sphere with the finite modulus in the SSB phase, and any chiral transformation with a constant chiral angle $`\theta _a`$ shifts each value on the sphere, leaving the QCD Lagrangian invariant. The spatially variant chiral angle $`\theta (𝐫)`$ represents the degree of freedom of the Nambu-Goldstone mode in the SSB vacuum. The dual chiral density wave (DCDW) is described by such a chiral angle $`\theta (𝐫)`$. When the chiral angle has some space-time dependence, there should appear extra terms in the effective potential as a consequence of chiral symmetry: one trivial term is the one describing the quark and DCDW interaction due to the non-commutability of $`\theta (𝐫)`$ with the kinetic (differential) operator in the Dirac operator. Another one is nontrivial and comes from the vacuum polarization effect: the energy spectrum of the quark is modified in the presence of $`\theta (𝐫)`$ and thereby the vacuum energy has an additional term, $`(\theta )^2`$ in the lowest order. This can be regarded as an appearance of the kinetic term for DCDW through the vacuum polarization . Thus, the interaction becomes strong enough to overwhelm the kinetic energy increase, the state becomes unstable to generate DCDW.
Many studies have suggested that chiral symmetry is restored at a certain density by suppression of $`q\overline{q}`$ excitation due to the presence of the Fermi sea, where none of the mean-fields is present. In usual discussion of such symmetry restoration, one implicitly discards the pseudo-scalar mean-field and is concentrated in the behavior of the scalar mean-field, while there is no compelling reason for the pseudo-scalar density to be vanished. Allowance of the degree of freedom of the chiral angle is nothing else but the appearance of DCDW. Thus we can say that instability of the ground state with respect to forming DCDW provides another path to symmetry restoration (see Fig. 11).
### 4.2 DCDW in the NJL model
Taking the Nambu-Jona-Lasinio (NJL) model as a simple but nontrivial example, we explicitly demonstrate that quark matter becomes unstable for a formation of DCDW above a certain density; the NJL model has been recently used as an effective model of QCD, embodying spontaneous breaking of chiral symmetry in terms of quark degree of freedom <sup>6</sup><sup>6</sup>6We can see that the OGE interaction gives the same form after the Fierz transformation in the zero-range limit . We shall explicitly see the DCDW state exhibits a ferromagnetic property.
We start with the NJL Lagrangian with $`N_f=2`$ flavors and $`N_c=3`$ colors,
$$_{NJL}=\overline{\psi }(i/m_c)\psi +G[(\overline{\psi }\psi )^2+(\overline{\psi }i\gamma _5𝝉\psi )^2],$$
(60)
where $`m_c`$ is the current mass, $`m_c5`$MeV. Under the Hartree approximation, we linearize Eq. (60) by partially replacing the bilinear quark fields by their expectation values with respect to the ground state.
In the usual treatment to study the restoration of chiral symmetry at finite density, authors implicitly discarded the pseudo-scalar mean-field, while this is justified only for the vacuum of a definite parity. We assume here the following mean-fields,
$`\overline{\psi }\psi `$ $`=`$ $`\mathrm{\Delta }\mathrm{cos}(𝐪𝐫)`$
$`\overline{\psi }i\gamma _5\tau _3\psi `$ $`=`$ $`\mathrm{\Delta }\mathrm{sin}(𝐪𝐫),`$ (61)
and others vanish <sup>7</sup><sup>7</sup>7It would be interesting to see that the DCDW configuration is similar to pion condensation in high-density nuclear matter within the $`\sigma `$ model, considered by Dautry and Nyman (DN), where $`\sigma `$ and $`\pi ^0`$ meson condensates take the same form as Eq. (61). The same configuration has been also assumed for non-uniform chiral phase in hadron matter by the use of the Nambu-Jona-Lasinio model . However, DCDW is by no means the pion condensation but should be directly considered as particle-hole and particle-antiparticle quark condensation in the deconfinement phase. . This configuration looks to break the translational invariance as well as rotation symmetry, but the former invariance is recovered by absorbing an additional constant by a global chiral transformation. Accordingly, we define a new quark field $`\psi _W`$ by the Weinberg transformation ,
$$\psi _W=\mathrm{exp}[i\gamma _5\tau _3𝐪𝐫/2]\psi ,$$
(62)
to separate the degrees of freedom of the amplitude and phase of DCDW in the Lagrangian. In terms of the new field the effective Lagrangian renders
$$_{MF}=\overline{\psi }_W[i/M1/2\gamma _5\tau _3q/]\psi _WG\mathrm{\Delta }^2,$$
(63)
where we put $`M2G\mathrm{\Delta }`$ and $`q^\mu =(0,𝐪)`$, taking the chiral limit ($`m_c=0`$). The form given in (63) appears to be the same as the usual one, except the axial-vector field generated by the wave vector of DCDW; the amplitude of DCDW produces the dynamical quark mass in this case. We shall see the wave vector $`𝐪`$ is related to the magnetization: the phase of DCDW induces the magnetization. With this form we can find a spatially uniform solution for the quark wave function (see Table 3), $`\psi _W=u_W(p)\mathrm{exp}(i𝐩𝐫)`$, with the eigenvalues,
$$E_p^\pm =\sqrt{E_p^2+|𝐪|^2/4\pm \sqrt{(𝐩𝐪)^2+M^2|𝐪|^2}},E_p=(M^2+|𝐩|^2)^{1/2}$$
(64)
for positive-energy (valence) quarks with different spin polarizations (c.f. (42)). <sup>8</sup><sup>8</sup>8This feature is very different from refs., where wave function is no more uniform.
### 4.3 Thermodynamic potential
The thermodynamic potential is given as
$`\mathrm{\Omega }_{\mathrm{total}}`$ $`=`$ $`\gamma {\displaystyle \frac{d^3p}{(2\pi )^3}\underset{s=\pm }{}\left[(E_p^s\mu )\theta _sE_p^s\right]}+M^2/4G`$ (65)
$``$ $`\mathrm{\Omega }_{\mathrm{val}}+\mathrm{\Omega }_{\mathrm{vac}}+M^2/4G.`$
where $`\theta _\pm =\theta (\mu E_p^\pm )`$, $`\mu `$ is the chemical potential and $`\gamma `$ the degeneracy factor $`\gamma =N_fN_c`$. The first term $`\mathrm{\Omega }_{\mathrm{val}}`$ is the contribution by the valence quarks filled up to the chemical potential, while the second term $`\mathrm{\Omega }_{\mathrm{vac}}`$ is the vacuum contribution that is apparently divergent. We shall see both contributions are indispensable in our discussion. Once $`\mathrm{\Omega }_{\mathrm{total}}`$ is properly evaluated, the equations to be solved to determine the optimal values of $`\mathrm{\Delta }`$ and $`q`$ are
$$\frac{\delta \mathrm{\Omega }_{\mathrm{total}}}{\delta \mathrm{\Delta }}=\frac{\delta \mathrm{\Omega }_{\mathrm{total}}}{\delta q}=0.$$
(66)
Since NJL model is not renormalizable, we need some regularization procedure to get a meaningful finite value for the vacuum contribution $`\mathrm{\Omega }_{\mathrm{vac}}`$, which can be recast in the form,
$$\mathrm{\Omega }_{\mathrm{vac}}=i\gamma \frac{d^4p}{(2\pi )^4}\mathrm{trln}S_W,$$
(67)
with use of the propagator $`S_W=(p/M1/2\tau _3\gamma _5q/)^1`$. There are various kinds of regularization and we must carefully choose the relevant one to the theoretical framework. Since the energy spectrum is no more rotation symmetric, we cannot apply the usual energy or momentum cut-off regularization (MCOR) scheme to regularize $`\mathrm{\Omega }_{\mathrm{vac}}`$. Moreover, the regularization should be, at least, independent of the order parameters $`\mathrm{\Delta }`$ and $`q`$. Note that this demand is essential to discuss the phase transition: improper regularizations spoil the consistency of the framework and give unphysical results for the order parameters $`\mathrm{\Delta }`$ and $`q`$ through Eq. (66). We adopt here the proper-time regularization (PTR) scheme , which is one of regularizations compatible with Eq. (66) <sup>9</sup><sup>9</sup>9The Pauli-Villars reguralization may be another candidate. . Introducing the proper-time variable $`\tau `$, we eventually find
$$\mathrm{\Omega }_{\mathrm{vac}}=\frac{\gamma }{8\pi ^{3/2}}_0^{\mathrm{}}\frac{d\tau }{\tau ^{5/2}}_{\mathrm{}}^{\mathrm{}}\frac{dp_z}{2\pi }\left[e^{(\sqrt{p_z^2+M^2}+q/2)^2\tau }+e^{(\sqrt{p_z^2+M^2}q/2)^2\tau }\right]\mathrm{\Omega }_{\mathrm{ref}},$$
(68)
which is reduced to the standard formula in the limit $`q0`$.
The integral with respect to the proper time $`\tau `$ is still divergent due to the $`\tau 0`$ contribution. Regularization proceeds by replacing the lower bound of the integration range by $`1/\mathrm{\Lambda }^2`$, which corresponds to the momentum cut-off in the MCOR scheme.
Now we examine a possible instability of quark matter with respect to formation of DCDW. In the following we first inquire the sign change of the curvature of $`\mathrm{\Omega }_{\mathrm{total}}`$ at the origin (stiffness parameter), $`\beta `$. Expanding $`\mathrm{\Omega }_{\mathrm{vac}}`$ with respect to $`q`$ up to $`O(q^2)`$, we find
$$\mathrm{\Omega }_{\mathrm{vac}}=\mathrm{\Omega }_{\mathrm{vac}}^0+\beta _{\mathrm{vac}}q^2+O(q^4)$$
(69)
where the vacuum stiffness parameter $`\beta _{\mathrm{vac}}`$ is given by
$$\beta _{\mathrm{vac}}=\frac{\gamma \mathrm{\Lambda }^2}{16\pi ^2}J(M^2/\mathrm{\Lambda }^2)$$
(70)
with a universal function, $`J(x)=x\mathrm{Ei}(x).`$ The nontrivial term originates from a vacuum polarization effect in the presence of DCDW and provides a kinetic term ($`(\theta )^2`$) for DCDW . The vacuum stiffness parameter $`\beta _{\mathrm{vac}}`$ can be also written as $`\beta _{vac}=\frac{1}{2}f_\pi ^2`$ with the pion decay constant $`f_\pi `$, and is always positive; it gives a ’repulsive’ contribution, so that the vacuum is stable against formation of DCDW. Note that it gives a null contribution in case of $`M=0`$ , irrespective of $`q`$, as it should be.
For given $`\mu ,M`$ and $`q`$ we can evaluate the contribution by the Fermi seas $`\mathrm{\Omega }_{\mathrm{val}}`$ using Eq. (64), but its general formula is very complicated . However, it may be sufficient to consider the small $`q`$ case for our present purpose. Then the thermodynamic potential can be expressed as
$`\mathrm{\Omega }_{\mathrm{val}}`$ $`=`$ $`\mathrm{\Omega }_{\mathrm{val}}^0{\displaystyle \frac{\gamma }{8\pi ^2}}M^2q^2H(\mu /M)+O(q^4)`$ (71)
$``$ $`\mathrm{\Omega }_{\mathrm{val}}^0+\mathrm{\Omega }_{\mathrm{val}}^{mag}+O(q^4)`$
up to $`O(q^2)`$, where $`H(x)=\mathrm{ln}(x+\sqrt{x^21})`$ and $`\mathrm{\Omega }_{\mathrm{val}}^0=ϵ_{\mathrm{val}}^0\mu \rho _{\mathrm{val}}^0`$ with $`\rho _{\mathrm{val}}^0=\frac{\gamma }{3\pi ^2}(\mu ^2M^2)^{3/2}`$ for normal quark matter. The valence stiffness parameter then reads
$$\beta _{val}=\frac{\gamma }{8\pi ^2}M^2H(\mu /M)$$
(72)
Since the function $`H(x)`$ is always positive and accordingly $`\beta _{val}0`$, the magnetic term $`\mathrm{\Omega }_{\mathrm{val}}^{mag}`$ always gives a negative energy and approaches to zero as $`M0`$ (triviality).
We may easily understand why the valence quarks always favor the formation of DCDW. First, consider the energy spectra for massless quarks (see Fig. 12).
As is already discussed, our theory becomes trivial in this case and we find two spectra
$$E_p^\pm =\sqrt{p_{}^2+(|p_z|\pm q/2)^2},𝐩_{}=(p_x,p_y,0),$$
(73)
which are essentially equivalent to $`E_p^\pm =|𝐩|`$ with definite chirality.
There is a level crossing at $`𝐩=\mathrm{𝟎}`$. Once the mass term is taken into account this degeneracy is resolved and the energy splitting arises there. Hence it causes an energy gain, if $`q=O(2\mu )`$; we can see that this mechanism is very similar to that of SDW by Overhauser .
Using Eqs. (65), (69), (71) we write the thermodynamic potential as
$$\mathrm{\Omega }_{\mathrm{total}}=\mathrm{\Omega }_{NJL}+\beta q^2+O(q^4)$$
(74)
with the total stiffness parameter $`\beta =\beta _{vac}+\beta _{val}`$ and the usual NJL expression without DCDW, $`\mathrm{\Omega }_{NJL}=\mathrm{\Omega }_{\mathrm{vac}}^0(M)+\mathrm{\Omega }_{\mathrm{val}}^0(M)+M^2/4G.`$ The dynamical quark mass $`M`$ is given by the equation, $`\mathrm{\Omega }_{\mathrm{total}}/M=0`$; At the order of $`q^0`$ the dynamical quark mass $`M^0`$ is determined by the equation, $`\mathrm{\Omega }_{NJL}/M|_{M^0}=0.`$ Since $`MM^0=O(q^2)`$, DCDW onsets at a certain density where the total stiffness parameter $`\beta `$ becomes negative: the critical chemical potential $`\mu ^{cr}`$ is determined by the equation,
$$\beta =\frac{1}{2}f_\pi ^2\frac{\gamma }{8\pi ^2}\left(M^0\right)^2H(\mu ^{cr}/M^0)=0.$$
(75)
Note that this is only a sufficient condition for formation of DCDW, and we can never exclude the possibility of the first-order phase transition or metamagnetism . Actually, we shall see that DCDW occurs as a first-order phase transition.
### 4.4 First-order phase transition
The values of the order parameters $`M`$ and $`q`$ are obtained from the minimum of the thermodynamic potential (65) for $`T=0`$. Fig. 13 shows the contours of $`\mathrm{\Omega }_{\mathrm{total}}`$ in the $`M`$-$`q`$ plane as the chemical potential increases, where the parameters are chosen as $`G\mathrm{\Lambda }^2=6`$ and $`\mathrm{\Lambda }=850`$ MeV, to reproduce the constituent quark mass in the vacuum ($`\mu =0`$) .
The crossed points denote the absolute minima. There are two critical chemical potential $`\mu =\mu _{c1},\mu _{c2}`$: for the lower densities (Fig. 13(a)-(b)) the absolute minimum resides at the point $`(M0,q=0)`$ indicating the SSB phase. At $`\mu =\mu _{c1}`$ (Fig. 13(c)) the potential has the two absolute minima at $`(M0,q=0)`$ and $`(M0,q0)`$, showing the first-order transition to the DCDW phase which is stable for $`\mu _{c1}<\mu <\mu _{c2}`$ (Fig. (13)d-e). At $`\mu =\mu _{c2}`$ (Fig. 13(f)) the axis of $`M=0`$ and a point $`(M0,q0)`$ become minima, the system undergoes the first-order transition again to the chiral-symmetric phase.
Fig. 15 summarizes the behaviors of the order-parameters $`M`$ and $`q`$ as functions of $`\mu `$ at $`T=0`$, where that of $`M`$ without DCDW is also shown for comparison. It is found that DCDW develops at finite range of $`\mu `$ ($`\mu _{c1}\mu \mu _{c2}`$), where the wave number $`q`$ increases with $`\mu `$ but its value is smaller than twice of the Fermi momentum $`2k_F`$($`2\mu `$ for free quarks) since the nesting of Fermi surfaces is incomplete in the present 3-D system; actually, the ratio becomes $`q/k_F=1.171.47`$ for the baryon-number densities $`\rho _b/\rho _0=3.625.30`$ where DCDW is stable (see Fig. 15).
### 4.5 Correlation functions
In this section, we consider scalar- and pseudoscalar-correlation functions, $`\mathrm{\Pi }_{\mathrm{s},\mathrm{sp}}(k)`$, in the massless limit $`M0`$, and discuss their relation with the mechanism for DCDW. In the static limit $`k_00`$, the correlation functions have a physical correspondence to the static susceptibility for the spin- or charge-density wave . We shall see that these functions have a differential singularity at $`k=2k_F`$, reflecting the sharp Fermi surface at $`T=0`$.
We explicitly evaluate the effective interactions, $`\mathrm{\Gamma }_{\mathrm{s},\mathrm{sp}}(k)`$, in the pseudo-scalar and scalar channels within the random phase approximation , which are related to the correlation functions $`\mathrm{\Pi }_{\mathrm{s},\mathrm{sp}}(k)`$, i.e., $`2G\mathrm{\Pi }_{\mathrm{s},\mathrm{sp}}(k)=\mathrm{\Gamma }_{\mathrm{s},\mathrm{sp}}(k)\mathrm{\Pi }_{\mathrm{s},\mathrm{sp}}^0(k)`$:
$`i\mathrm{\Gamma }_{\mathrm{s},\mathrm{ps}}(k)`$ $`=`$ $`{\displaystyle \frac{2Gi}{12G\mathrm{\Pi }_{\mathrm{s},\mathrm{ps}}^0(k)}},`$ (76)
where $`\mathrm{\Pi }_{\mathrm{s},\mathrm{ps}}^0(k)`$ are the polarization functions in medium,
$`\mathrm{\Pi }_\mathrm{s}^0(|𝐤|)`$ $`=`$ $`\mathrm{\Pi }_{\mathrm{ps}}^0(|𝐤|)`$ (77)
$`=`$ $`{\displaystyle \frac{N_fN_c}{4\pi ^2}}(\mathrm{\Lambda }^22k_F^2)2N_fN_ci𝐤^2I(𝐤^2)|_{M0}`$
$`+`$ $`{\displaystyle \frac{N_fN_c|𝐤|}{4\pi ^2}}[\left(k_F{\displaystyle \frac{|𝐤|}{2}}\right)\mathrm{log}\left({\displaystyle \frac{2k_F+|𝐤|}{2k_F|𝐤|}}\right)+{\displaystyle \frac{|𝐤|}{2}}\mathrm{log}({\displaystyle \frac{2k_F}{|𝐤|}}+{\displaystyle \frac{|𝐤|}{2k_F}})],`$
in the static and chiral limit . It is well known that poles of the effective interaction give the energies of scalar and pseudo-scalar mesons . Note that the inverse of the effective interaction in the massless limit also gives the coefficient of $`M^2`$ in the effective potential in the presence of DCDW,
$`\mathrm{\Omega }_{\mathrm{total}}=\mathrm{\Omega }_{\mathrm{total}}|_{M0}+{\displaystyle \frac{1}{2}}\mathrm{\Gamma }_{\mathrm{ps}}^1(q)|_{M0}M^2+O(M^4).`$ (78)
Hence, the critical density and the critical wave vector $`q_{\mathrm{crit}}`$ should be given by the equations,
$$\mathrm{\Gamma }_{\mathrm{ps}}^1(q_{\mathrm{crit}})|_{M0}=0,\mathrm{\Gamma }_{\mathrm{ps}}^1(q)|_{M0}/q|_{q_{\mathrm{crit}}}=0,$$
(79)
in the case of the second-order or weakly first-order phase transition.
Fig.16 show the function $`1/\mathrm{\Gamma }_{\mathrm{ps}}(|𝐤|)|_{M=0}`$ and we can see these conditions are almost satisfied at the terminal density, $`\mu =\mu _{c2}`$, due to a tiny jump in the dynamical mass (weakly first-order phase transition) . A numerical calculation in the chiral limit gives $`\mu _t(=k_F)=0.5320\mathrm{\Lambda }`$ and $`|𝐤_t|=1.498k_F`$, which almost coincide with the previous results given in Figs. 13 and 15, $`\mu _{c2}=0.53254\mathrm{\Lambda }`$ and $`q=1.469k_F`$ (where $`k_F\sqrt{\mu _{c2}^2M^2}`$; $`M=0.034\mathrm{\Lambda }`$). On the other hand, the phase transition is of first order at the onset density, $`\mu =\mu _{c1}`$: there is a discontinuous jump in the dynamical mass (the amplitude of DCDW) $`M`$, so that the above argument cannot be applied any more. Besides, the correlation functions or the effective interactions provide a powerful tool to analyze the DCDW phase as far as the dynamically generated mass (the amplitude of DCDW) $`M`$ is small, as in the present case. From the behavior of the function $`\mathrm{\Gamma }_{\mathrm{ps}}(|𝐤|)^1`$ shown in Fig. 16, it is found that $`\mathrm{\Gamma }_{\mathrm{ps}}(|𝐤|)^1`$ takes the lowest value at $`|𝐤|1.31.5k_F(O(2k_F))`$, reflecting the sharp Fermi surface, <sup>10</sup><sup>10</sup>10Actually the function $`\mathrm{\Gamma }_{\mathrm{ps}}(|𝐤|)^1`$ diverges at $`|𝐤|=2k_F`$ in the one-dimensional case, which means the complete nesting. and thus a finite wave number $`q`$ gives the lower potential energy in Eq. (78) than $`q=0`$ in the density range of DCDW. These values are consistent with those in Fig. 15, and we can see again that DCDW is closely related to the sharpness of the Fermi surface.
It should be noted again that the negative value of the function $`\mathrm{\Gamma }_{\mathrm{ps}}(|𝐤|)^1|`$ gives a necessary condition for formation of DCDW, but the sign change does not necessarily imply the critical condition in the case of first-order phase transitions, as in the present case; the terminal transition is weakly first-order and we can apply there. It should be also noted that its minimum point always gives an optimal value of the wave vector in the presence of DCDW. Thus we can see by the use of the correlation functions that the particle-hole pairing with finite momentum $`q=O(2k_F)`$ effectively lowers the free energy in comparison with the zero total momentum.
The above argument might also be available even for the case of a finite current-quark mass, $`m_c5`$MeV: Fig. 16 shows that the minimum of $`\mathrm{\Gamma }_{\mathrm{ps}}(|𝐤|)^1`$ has little shift from that in the chiral limit.
### 4.6 Magnetic properties
The mean-value of the spin operator is given by
$$\overline{s}_z=\frac{1}{2}u_W^{}\mathrm{\Sigma }_zu_W=\frac{1}{2}\frac{q/2\pm \beta _p}{E_p^\pm }+\mathrm{vac},$$
(80)
with $`\beta _p=\sqrt{p_z^2+m^2}`$, where ”vac” means the vacuum contribution. First note that the integral of $`\overline{s}_z`$ over the Fermi seas should be proportional to $`q`$, and the solution with $`q0`$ seems to imply FM. However, we can show that PTR gives the vacuum (the Dirac sea) contribution oppositely to cancel the total mean-value of the spin operator, which is consistent with Eq. (66). Instead we can see that the magnetization spatially oscillates,
$$M_z\overline{q}\sigma _{12}q=\gamma _0\sigma _{12}\mathrm{cos}(𝐪𝐫),$$
(81)
with
$$\gamma _0\sigma _{12}=_{F^+F^{}}\frac{d^3p}{(2\pi )^3}\frac{2M}{\sqrt{M^2+p_z^2}},$$
(82)
which means a kind of spin density wave .
### 4.7 Phase diagram in the $`T\mu `$ plane
To establish the phase diagram in the $`T\mu `$ plane, we derive the thermodynamic potential at finite temperature in the Matsubara formalism. The partition function for the mean-field Hamiltonian is given by
$`Z_\beta `$ $`=`$ $`{\displaystyle D\overline{\psi }D\psi \mathrm{exp}_0^\beta 𝑑\tau d^3r\left\{\overline{\psi }\left[i\stackrel{~}{}+M\mathrm{exp}\left(i\gamma _5𝐪𝐫\right)\gamma _0\mu \right]\psi \frac{M^2}{4G}\right\}}`$ (83)
$`=`$ $`{\displaystyle \underset{𝐤,n,s=\pm }{}}\left\{(i\omega _n+\mu )^2E_s^2(𝐤)\right\}^{N_fN_c}\times \mathrm{exp}\left\{\left({\displaystyle \frac{M^2}{4G}}\right)V\beta \right\},`$
where $`\beta =1/T`$, $`\stackrel{~}{}\gamma _0_\tau +i\gamma `$ and $`\omega _n`$ the Matsubara frequency. Thus the thermodynamic potential $`\mathrm{\Omega }_\beta `$ is obtained,
$`\mathrm{\Omega }_\beta (q,M)`$ $`=`$ $`T\mathrm{log}Z_\beta (q,M)/V`$ (84)
$`=`$ $`N_fN_c{\displaystyle \frac{d^3\mathrm{k}}{(2\pi )^3}\underset{s}{}\left\{T\mathrm{log}\left[e^{\beta \left(E_s(𝐤)\mu \right)}+1\right][e^{\beta \left(E_s(𝐤)+\mu \right)}+1]+E_s(𝐤)\right\}}`$
$`+{\displaystyle \frac{M^2}{4G}}.`$
From the absolute minima of the thermodynamic potential (84), it is found that the order parameters at $`T0`$ behave similarly to those at $`T=0`$ as a function of $`\mu `$, while the chemical-potential range of the DCDW at finite temperature, $`\mu _{c1}(T)\mu \mu _{c2}(T)`$, gets smaller as $`T`$ increases. We show the resultant phase diagram in Fig. 17, where the ordinary chiral-transition line is also given. Comparing phase diagrams with and without $`q`$, we find that the DCDW phase emerges in the area (closed area in Fig. 17) which lies just outside the boundary of the ordinary chiral transition. We thus conclude that the DCDW is induced by finite-density contributions, and has an effect to expand the chiral-condensed phase ($`M0`$) toward low temperature and high density region.
## 5 Summary and Concluding remarks
We have seen some magnetic aspects of quark matter: ferromagnetism at high densities and spin density wave at moderate densities within the zero-range approximation for the interaction vertex. These look to follow the similar development about itinerant electrons: Bloch mechanism at low densities and spin density wave at high densities by Overhauser.
By a perturbative calculation with the OGE interaction, we have seen ferromagnetism in quark matter at low densities (§2.1) . It would be worth mentioning that another study with higher-order diagrams qualitatively supports it . These studies suggest an opposite tendency to the one using the zero-range interaction (§§2.2,2.3). Note that we can also see the same situation for itinerant electrons; the Hartree-Fock calculation based on the infinite-range Coulomb interaction favors ferromagnetism at a low density region, while the Stoner model, which introduces the zero-range effective interaction instead of the Coulomb interaction, gives ferromagnetism at high densities. So we must carefully examine the possibility of ferromagnetism in quark matter by taking into account the finite-range effect.
We have examined the coexistence of spin polarization and color superconductivity by choosing a quark pair with the same polarization. We have introduced the axial-vector self-energy and the quark pair field (the gap function), whose forms are derived from the one-gluon-exchange interaction by way of the Fierz transformation under the zero-range approximation. Within the relativistic Hartree-Fock framework we have evaluated their magnitudes in a self-consistent manner by way of the coupled Schwinger-Dyson equations.
As a result of numerical calculations spontaneous spin polarization occurs at a high density for a finite quark mass in the absence of CSC, while it never appears for massless quarks as an analytical result. In the spin-polarized phase the single-particle energies corresponding to spin degrees of freedom, which are degenerate in the non-interacting system, are split by the exchange energy in the axial-vector channel. Each Fermi sea of the single-particle energy deforms in a different way, which causes an asymmetry in the two Fermi seas and then induces the axial-vector mean-field in a self-consistent manner. In the superconducting phase, however, spin polarization is slightly reduced by the pairing effect; it is caused by competition between reduction of the deformation and enhancement of the difference in the phase spaces of opposite “spin” states due to the anisotropic diffuseness in the momentum distribution.
We have also noted another possibility of the pairing: the quark pair with opposite polarization to each other. It may lead to a gapless superconductor, but we need a further study.
We have seen that dual chiral desnity wave (DCDW) appears at a certain density and develops at moderate densities (§3). It occurs as a result of the interplay between the $`\overline{q}q`$ and particle-hole correlations. The phase transition is of weakly first order, and the restoration of chiral symmetry is delayed compared with the usual scenario.
For the discussion of DCDW given in §4, we have seen the remarkable roles of the Fermi sea and the Dirac sea: the former always favors DCDW, while the latter works against it. The similar situation also appears about the magnetic property of quark matter. The mean value of the spin operator over the Fermi seas of valance quarks always gives a finite value in the presence of DCDW, which is a kind of ferromagnetism, but the vacuum contribution given by the Dirac seas completely cancels it. As a result there is no net spin polarization in this case, but we have seen magnetization spatially oscillates instead (spin density wave). This is one of the typical examples in which the nonrelativistic picture is qualitatively different from the relativistic one by the vacuum effect.
It would be interesting to recall that DCDW is similar to pion condensation within the $`\sigma `$ model, considered by Dautry and Nyman , where $`\sigma `$ and $`\pi ^0`$ meson condensates take the same form as Eq. (61). So it might be intriguing to connect pion condensation before deconfinement with DCDW after it in light of symmetry consideration. Note that this type of hadron-quark continuity has been also suggested in the context of hadron and quark superconductivities .
If ferromagnetism is realized in quark matter, it may give a microscopic origin of the magnetic field in compact stars; actually we have seen that it can give a possible explanation for the superstrong magnetic field observed in magnetars, if they are quarks stars. It would be challenging to explain other characteristic phenomena in magnetars such as a sudden braking down observed in a soft gamma-ray repeater SGR 1806-20 or SGR 1900+14 ; some global reconfiguration of of the magnetic has been suggested for these phenomena . It would be also ambitious to give a scenario based on magnetic properties of quark matter, which can explain the hierarchy of the magnetic field observed in three classes of neutron stars, magnetars, radio pulsars and recycled millisecond pulsars. Ferromagnetism may give a permanent magnetization and there is no field decay, in difference from the dynamo mechanism caused by the charged current.
The magnetic phases considered here accompany the symmetry breaking, $`SO(3)O(2)`$, so that we can expect the Nambu-Goldstone modes as lowest excitations in the ground state: spin wave in ferromagnetism and phason in DCDW. It should be interesting to study these modes. Such low excitation modes may affect the thermal evolution of compact stars . It would be also interesting to investigate how the effective interaction by exchanging such excitations between quarks affects superconductivity.
## Acknowledgments
This work is partially supported by the Grant-in-Aid for the 21st Century COE “Center for the Diversity and Universality in Physics ” from the Ministry of Education, Culture, Sports, Science and Technology of Japan. It is also partially supported by the Japanese Grant-in-Aid for Scientific Research Fund of the Ministry of Education, Culture, Sports, Science and Technology (13640282, 16540246).
|
warning/0506/hep-th0506188.html
|
ar5iv
|
text
|
# Classical solution of a sigma model in curved background
## 1 Introduction
Klimčík and Ševera in their seminal work described the conditions and procedure for transforming solutions of a $`\sigma `$–model to those of a dual one. Namely, let us assume that the $`\sigma `$–model is defined on a Lie group $`G`$ on which a covariant second order tensor field $`F`$ is given. The classical action of the $`\sigma `$–model then is
$$S_F[\varphi ]=d^2x_{}\varphi ^\mu F_{\mu \nu }(\varphi )_+\varphi ^\nu $$
(1)
where the functions $`\varphi ^\mu :𝐑^2𝐑,\mu =1,2,\mathrm{},dimG`$ are obtained by the composition $`\varphi ^\mu =y^\mu \varphi `$ of a map $`\varphi :𝐑^2G`$ and a coordinate map $`y:U_g𝐑^n,n=dimG`$ of a neighborhood of an element $`\varphi (x_+,x_{})=gG`$. If $`F`$ satisfies
$$_{v_i}(F)_{\mu \nu }=F_{\mu \kappa }v_j^\kappa \stackrel{~}{f}_i^{jk}v_k^\lambda F_{\lambda \nu },i,\mu ,\nu =1,\mathrm{},dimG$$
(2)
where $`v_i`$ form a basis of left–invariant fields on $`G`$ and $`\stackrel{~}{f}_i^{jk}`$ are structure coefficients of a Lie group $`\stackrel{~}{G},dim\stackrel{~}{G}=dimG`$, then there is a relation between solutions of the equations of motion for $`S_F`$ and $`S_{\stackrel{~}{F}}`$ where $`\stackrel{~}{F}_{\mu \nu }`$ is a second order tensor field on $`\stackrel{~}{G}`$.
The relation between the solutions $`\varphi (x_+,x_{})`$ of the model given by $`F`$ and $`\stackrel{~}{\varphi }(x_+,x_{})`$ of the model given by $`\stackrel{~}{F}`$ is given by two possible decompositions of elements $`d`$ of Drinfel’d double
$$d=g.\stackrel{~}{h}=\stackrel{~}{g}.h$$
(3)
where $`g,hG,\stackrel{~}{g},\stackrel{~}{h}\stackrel{~}{G}`$.
The map $`\stackrel{~}{h}:𝐑^2\stackrel{~}{G}`$ that we need for this duality transform satisfies the equations
$$((_+\stackrel{~}{h}).\stackrel{~}{h}^1)_j=A_{+,j}:=v_j^\lambda F_{\lambda \nu }(\varphi )_+\varphi ^\nu $$
(4)
$$((_{}\stackrel{~}{h}).\stackrel{~}{h}^1)_j=A_{,j}:=_{}\varphi ^\lambda F_{\lambda \nu }(\varphi )v_j^\nu $$
(5)
Even though the equations (35) define the so called Poisson–Lie T–duality transformation their solution is usually very complicated to use them for finding the solutions. There are three steps in performing the transformation:
1. You must know a solution $`\varphi (x_+,x_{})`$ of the $`\sigma `$–model given by $`F`$.
2. For the given $`\varphi (x_+,x_{})`$ you must find $`\stackrel{~}{h}(x_+,x_{})`$ i.e. solve the system of PDE’s (4,5).
3. For given $`d(x_+,x_{})=\varphi (x_+,x_{}).\stackrel{~}{h}(x_+,x_{})D`$ you must find the decomposition $`d(x_+,x_{})=\stackrel{~}{\varphi }(x_+,x_{}).h(x_+,x_{})`$ where $`\stackrel{~}{\varphi }(x_+,x_{})\stackrel{~}{G},h(x_+,x_{})G`$.
The goal of this paper is to present an example of a three–dimensional $`\sigma `$–model with nontrivial (i.e. curved) background for which all the three steps can be done so that the $`\sigma `$–model can be explicitly solved by this transformation. The tensor $`\stackrel{~}{F}`$ of this model is
$$\stackrel{~}{F}_{\mu \nu }(\stackrel{~}{y})=\left(\begin{array}{ccc}\frac{\stackrel{~}{y}_{1}^{}{}_{}{}^{2}}{\kappa ^3+U\kappa \stackrel{~}{y}_1}& \frac{\stackrel{~}{y}_1}{\kappa ^2+U\stackrel{~}{y}_1}& \frac{1}{\kappa }\\ \frac{\stackrel{~}{y}_1}{\kappa ^2+U\stackrel{~}{y}_1}& \frac{\kappa }{\kappa ^2+U\stackrel{~}{y}_1}& 0\\ \frac{\kappa }{\kappa ^2+U\stackrel{~}{y}_1}& \frac{U}{\kappa ^2+U\stackrel{~}{y}_1}& 0\end{array}\right).$$
(6)
where $`U`$ and $`\kappa `$ are constants. The Gauss curvature of the metric $`\stackrel{~}{G}_{\mu \nu }(\stackrel{~}{y}):=(\stackrel{~}{F}_{\mu \nu }(\stackrel{~}{y})+\stackrel{~}{F}_{\nu \mu }(\stackrel{~}{y}))/2`$ is
$$R=\frac{7U^4}{8\kappa \left(\kappa ^2+U\stackrel{~}{y}_1\right)^2}$$
(7)
so that for $`U0`$ we have a $`\sigma `$–model in a curved background (and with nontrivial torsion). The equations of motion have the form
$$_{}_+\varphi ^\mu +\mathrm{\Gamma }_{\nu \lambda }^\mu _{}\varphi ^\nu _+\varphi ^\lambda =0$$
(8)
where
$$\mathrm{\Gamma }_{\nu \lambda }^\mu :=\frac{1}{2}\stackrel{~}{G}^{\mu \rho }(\stackrel{~}{F}_{\rho \lambda ,\nu }+\stackrel{~}{F}_{\nu \rho ,\lambda }\stackrel{~}{F}_{\nu \lambda ,\rho }).$$
(9)
## 2 T-duality of the model
The reason why the above given model can be solved is that it is T-dual to a model with the flat background (Actually it was constructed in this way). It is easy to check that the tensor $`\stackrel{~}{F}`$ satisfies the equations dual to (2)
$$_{\stackrel{~}{v}_i}(\stackrel{~}{F})_{\mu \nu }=\stackrel{~}{F}_{\mu \kappa }\stackrel{~}{v}_j^\kappa f_i^{jk}\stackrel{~}{v}_k^\lambda \stackrel{~}{F}_{\lambda \nu },$$
(10)
for vector fields on $`𝐑^3`$ that are left–invariant with respect to the Abelian group structure and $`f_i^{jk}`$ being structure constants of the Lie algebra given by
$$[T^1,T^2]=0,[T^2,T^3]=T^1,[T^3,T^1]=0.$$
(11)
It means that the equations of motion (8) of the $`\sigma `$–model can be rewritten (see , ) as equations on the six–dimensional Drinfel’d double $`D`$ – connected Lie group whose Lie algebra $`𝒟`$ admits a decomposition
$$𝒟=\stackrel{~}{𝒢}+𝒢$$
into two subalgebras that are maximally isotropic with respect to a bilinear, symmetric, nondegenerate, ad–invariant form.
In this case, the subalgebras $`\stackrel{~}{𝒢}`$ and $`𝒢`$ are the three–dimensional Abelian and the second Bianchi algebra (11) so that the Poisson–Lie T–duality reduces to the nonabelian T-duality. The tensor field $`\stackrel{~}{F}`$ can be obtained as
$$\stackrel{~}{F}(\stackrel{~}{\varphi })=(E+\stackrel{~}{\pi }(\stackrel{~}{\varphi }))^1$$
(12)
where
$$E=\left(\begin{array}{ccc}0& U& \kappa \\ 0& \kappa & 0\\ \kappa & 0& 0\end{array}\right)$$
(13)
and the matrix function $`\stackrel{~}{\pi }(\stackrel{~}{\varphi })`$ follows from the adjoint representation of $`\stackrel{~}{G}`$ on $`𝒟`$ (see e.g. ). Similarly, the tensor field $`F`$ of the dual $`\sigma `$–model can be obtained as
$$F(\varphi )=e(\varphi )Ee(\varphi )^t,$$
(14)
where the matrix $`e(\varphi )`$ is the vielbein field on the group $`G`$ corresponding to the second Bianchi algebra (11) and $`e(\varphi )^t`$ is its transpose. From this formula one gets
$$F_{\mu \nu }(\varphi ^\rho )=\left(\begin{array}{ccc}0& U& \kappa \\ 0& \kappa & 0\\ \kappa & U\varphi ^2& 2\kappa \varphi ^2\end{array}\right)$$
(15)
The metric of this model is flat in the sense that its Riemann tensor vanishes.
## 3 Solution of the curved model
In the following subsections we are going to perform the above given steps of the duality transform between solutions of equations of motion for $`S_F`$ and $`S_{\stackrel{~}{F}}`$.
### 3.1 Solution of the flat model
Even though we know that the model given by the tensor (15) is on the flat background it is not easy to find the functions $`\varphi ^\mu (x_+,x_{})`$ that solve the equation of motion given by the action $`S_F[\varphi ]`$ because the Christoffel symbols are not zero in spite of the fact that the metric is flat. To solve the equation of motion we must express $`\varphi ^\mu `$ in terms of coordinates $`\xi `$ for which the metric become constant. This was done in for even more general forms of flat metrics.
Transformation of coordinates
$$\varphi ^1=\xi _12\xi _2\mathrm{\Omega }\frac{8\mathrm{\Omega }^3}{3}+\frac{U}{4\kappa }(\xi _2+2\mathrm{\Omega }^2)^2$$
$$\varphi ^2=\xi _2+2\mathrm{\Omega }^2$$
(16)
$$\varphi ^3=2\mathrm{\Omega }\frac{U}{2\kappa }(\xi _2+2\mathrm{\Omega }^2)$$
where $`\mathrm{\Omega }=\xi _3/2+\xi _2U/(4\kappa )`$ transform the metric obtained as the symmetric part of (15) to constant
$$G^{}(\xi )=\left(\begin{array}{ccc}0& U/2& \kappa \\ U/2& \kappa & 0\\ \kappa & 0& 0\end{array}\right).$$
and equations of motion transform to the wave equations so that
$$\xi _j(x_+,x_{})=W_j(x_+)+Y_j(x_{})$$
(17)
with arbitrary $`W_j`$ and $`Y_j`$. Functions $`\varphi ^\mu (x_+,x_{})`$ that solve the equations of motion for $`S_F[\varphi ]`$ then follow from (16) and (17).
This finishes the first step in obtaining the solution of the $`\sigma `$–model in the curved background by the duality transform. The second step in the duality transform requires solving the system (4,5).
### 3.2 Solution of the system (4,5)
The coordinates $`\stackrel{~}{h}_\nu `$ in the Abelian group $`\stackrel{~}{G}`$ can be chosen so that the left–hand sides of the equations (4,5) are just $`_\pm \stackrel{~}{h}_\nu `$. The right–hand sides are
$$A_+=\left(\begin{array}{c}U_+\varphi ^2+\kappa _+\varphi ^3\\ \kappa \varphi ^3_+\varphi ^3+\kappa _+\varphi ^2U\varphi ^3_+\varphi ^2\\ \kappa _+\varphi ^1+U\varphi ^2_+\varphi ^2+2\kappa \varphi ^2_+\varphi ^3\end{array}\right)$$
(18)
$$A_{}=\left(\begin{array}{c}\kappa _{}\varphi ^3\\ U_{}\varphi ^1\kappa _{}\varphi ^2U\varphi ^2_{}\varphi ^3+\kappa \varphi ^3_{}\varphi ^3\\ \kappa _{}\varphi ^12\kappa \varphi ^2_{}\varphi ^3\end{array}\right)$$
(19)
and for the solution $`\varphi ^\mu (x_+,x_{})`$ found in the previous section they become rather extensive expressions in $`W(x_+)`$ and $`Y(x_{})`$. Nevertheless, the equations (4,5) can be solved and the general solution is
$`\stackrel{~}{h}_1(x_+,x_{})`$ $`=`$ $`\kappa \left(Y_3(x_{})W_3(x_+)\right)UW_2(x_+)U\mathrm{\Omega }^2,`$
$`\stackrel{~}{h}_2(x_+,x_{})`$ $`=`$ $`\kappa \left(Y_2(x_{})W_2(x_+)\right)+UY_1(x_{})+{\displaystyle \frac{U}{2}}\beta (x_+,x_{})+`$
$`{\displaystyle \frac{U}{2}}\left(W_2(x_+)Y_3(x_{})W_3(x_+)Y_2(x_{})\right)+{\displaystyle \frac{2U}{3}}\mathrm{\Omega }^3`$
$`{\displaystyle \frac{U^2}{2\kappa }}\left({\displaystyle \frac{1}{2}}(W_2(x_+)+Y_2(x_{}))+\mathrm{\Omega }^2\right)^2,`$
$`\stackrel{~}{h}_3(x_+,x_{})`$ $`=`$ $`\kappa \left(Y_1(x_{})W_1(x_+)\right)+\kappa \left(W_2(x_+)Y_3(x_{})W_3(x_+)Y_2(x_{})\right)+`$
$`C+\kappa \beta (x_+,x_{})U\left({\displaystyle \frac{1}{2}}(W_2(x_+)+Y_2(x_{}))+\mathrm{\Omega }^2\right)^2`$
where $`C`$ is a constant,
$$\mathrm{\Omega }=\frac{1}{2}(W_3(x_+)+Y_3(x_{}))+\frac{U}{4\kappa }(W_2(x_+)+Y_2(x_{}))$$
and the function $`\beta `$ solves
$$_+\beta =W_2^{}(x_+)W_3(x_+)W_3^{}(x_+)W_2(x_+),$$
$$_{}\beta =Y_2(x_{})Y_3^{}(x_{})Y_3(x_{})Y_2^{}(x_{}).$$
### 3.3 Dual decomposition of elements of the Drinfel’d double
The final step in the dual transformation follows from the possibility of rewriting
$$d(x_+,x_{})=\varphi (x_+,x_{}).\stackrel{~}{h}(x_+,x_{})$$
as $`\stackrel{~}{\varphi }(x_+,x_{}).h(x_+,x_{})`$ where $`\varphi (x_+,x_{}),h(x_+,x_{})G,`$ and $`\stackrel{~}{\varphi }(x_+,x_{}),\stackrel{~}{h}(x_+,x_{})\stackrel{~}{G}`$. As both $`G`$ and $`\stackrel{~}{G}`$ are solvable (even nilpotent) we can write all group elements as product of elements of one–parametric subgroups and the two possible decompositions yield an equation for $`\stackrel{~}{\varphi }_\mu `$ and $`h^\nu `$ in terms of $`\stackrel{~}{h}_\lambda `$ and $`\varphi ^\rho `$
$$\mathrm{e}^{\varphi ^1T_1}\mathrm{e}^{\varphi ^2T_2}\mathrm{e}^{\varphi ^3T_3}\mathrm{e}^{\stackrel{~}{h}_1\stackrel{~}{T}^1}\mathrm{e}^{\stackrel{~}{h}_2\stackrel{~}{T}^2}\mathrm{e}^{\stackrel{~}{h}_3\stackrel{~}{T}^3}=\mathrm{e}^{\stackrel{~}{\varphi }_1\stackrel{~}{T}^1}\mathrm{e}^{\stackrel{~}{\varphi }_2\stackrel{~}{T}^2}\mathrm{e}^{\stackrel{~}{\varphi }_3\stackrel{~}{T}^3}\mathrm{e}^{h^1T_1}\mathrm{e}^{h^2T_2}\mathrm{e}^{h^3T_3}.$$
(21)
To solve it might be rather complicated in general but in this case when the only nonzero Lie products are
$$[T_2,T_3]=T_1,[T_2,\stackrel{~}{T}^1]=\stackrel{~}{T}^3,[T_3,\stackrel{~}{T}^1]=\stackrel{~}{T}^2$$
it can be easily done. We can use the Baker–Campbell–Hausdorff formula that now implies
$$\mathrm{e}^A\mathrm{e}^B=\mathrm{e}^B\mathrm{e}^A\mathrm{e}^{[A,B]}$$
(22)
and by repeated application of this formula we get
$$\stackrel{~}{\varphi }_1=\stackrel{~}{h}_1,\stackrel{~}{\varphi }_2=\stackrel{~}{h}_2+\stackrel{~}{h}_1\varphi ^3,\stackrel{~}{\varphi }_3=\stackrel{~}{h}_3\stackrel{~}{h}_1\varphi ^2$$
(23)
and $`h^\nu =\varphi ^\nu ,\nu =1,2,3`$. Inserting (16), (17) and (3.2) into (23) we get the solution of the equations of motion for the $`\sigma `$–model given by the action $`S_{\stackrel{~}{F}}`$ where $`\stackrel{~}{F}`$ is given by (6).
An example of a simple solution dependent on both $`x_+`$ and $`x_{}`$ is
$`\stackrel{~}{\varphi }_1(x_+,x_{})`$ $`=`$ $`U\mathrm{sin}t\mathrm{cos}x,`$
$`\stackrel{~}{\varphi }_2(x_+,x_{})`$ $`=`$ $`2\kappa \mathrm{cos}t\mathrm{sin}x+{\displaystyle \frac{U^2}{2\kappa }}\mathrm{sin}^2t\mathrm{cos}^2x,`$ (24)
$`\stackrel{~}{\varphi }_3(x_+,x_{})`$ $`=`$ $`U\mathrm{sin}^2t\mathrm{cos}^2x`$
where $`t=(x_++x_{})/2,x=(x_+x_{})/2`$ or more generally
$`\stackrel{~}{\varphi }_1(x_+,x_{})`$ $`=`$ $`{\displaystyle \frac{U}{2}}\left(Y_2(x_{})+W_2(x_+)\right),`$
$`\stackrel{~}{\varphi }_2(x_+,x_{})`$ $`=`$ $`\kappa \left(Y_2(x_{})W_2(x_+)\right)+{\displaystyle \frac{U^2}{8\kappa }}\left(Y_2(x_{})+W_2(x_+)\right)^2,`$ (25)
$`\stackrel{~}{\varphi }_3(x_+,x_{})`$ $`=`$ $`{\displaystyle \frac{U}{4}}\left(Y_2(x_{})+W_2(x_+)\right)^2`$
obtained from (23) for
$$Y_1=W_1=0,Y_3=\frac{U}{2\kappa }Y_2,W_3=\frac{U}{2\kappa }W_2,W_2,Y_2\mathrm{arbitrary}.$$
Another very simple solution is
$$\stackrel{~}{\varphi }_1(x_+,x_{})=0,\stackrel{~}{\varphi }_2(x_+,x_{})=UY_1(x_{}),\stackrel{~}{\varphi }_3(x_+,x_{})=\kappa \left(Y_1(x_{})W_1(x_+)\right).$$
(26)
obtained for $`Y_2=W_2=Y_3=W_3=0,W_1,Y_1\mathrm{arbitrary}`$.
## 4 Conclusions
We have explicitly solved the equations of motion of the three–dimensional $`\sigma `$–model in the curved background (6) by the Poisson–Lie T-duality transformation. The solution $`\stackrel{~}{\varphi }(x_+,x_{})`$ is given by composition of the formulas (23), (3.2), (16) and (17). Even though the transformation is known for more than ten years it is for the first time when it was used, to the best knowledge of the author, for finding an explicit solution. The reason may be that performing the three steps of the transformation mentioned in the Introduction may be rather difficult in general.
To solve the equations of motion we have used the fact that we know several $`\sigma `$–models in the curved background that can be transformed to the flat ones (see ). We were also able to find the transformation of group coordinates of the flat model to those for which the metric is constant. In the latter coordinates solution of the flat $`\sigma `$–model reduces to the solution of the wave equation. Performability of the next two steps of the Poisson–Lie T–duality transformation depends critically on the complexity of the structure of Drinfel’d double where the $`\sigma `$–models live. In our case one of the subgroup of the decomposition of the Drinfel’d double was Abelian and the other one nilpotent. Because of that the systems of equations (4) and (5) separate and the formula (22) for solution of (21) can be used. More complicated cases are under investigation now.
Let us note that in we have tried to solve the equations of motion for $`\sigma `$–models on the solvable groups with curved backgrounds by the Inverse scattering method. It turned out that for Lax pairs linear in currents $`g^1_\mu g`$ the $`\sigma `$–models solvable by the Inverse scattering method must have constant Christoffel symbols which is not the case of (6).
The author is grateful to Libor Šnobl for valuable comments.
|
warning/0506/quant-ph0506144.html
|
ar5iv
|
text
|
# Quantum storage and information transfer with superconducting qubits
thanks: E-mail: zwang@hkucc.hku.hk
## Abstract
We design theoretically a new device to realize the general quantum storage based on dcSQUID charge qubits. The distinct advantages of our scheme are analyzed in comparison with existing storage scenarios. More arrestingly, the controllable $`XY`$-model spin interaction has been realized for the first time in superconducting qubits, which may have more potential applications besides those in quantum information processing. The experimental feasibility is also elaborated.
As solid state quantum devices, Josephson junctions and superconducting quantum interference devices (SQUIDs) have manifested arresting and robust macroscopic quantum behaviors. They can be used to develop new quantum bits and logic gates in the context of quantum information science Makhlin . Since the favorable elements of good coherence, controllability, and scalability are integrated in these superconducting devices, they are very promising for the realization of quantum information processing. Recently, a series of exciting experimental progresses have been made in this field, including high quality single-qubits vion1 ; han1 ; martinis ; mooij1 , the quantum entanglement between the two qubits nec2 ; berkley , and the CNOT gate nec3 ; mcdermott realized in various superconducting devices. Besides, both experimental and theoretical efforts have also been devoted to explore new quantum information processing devices based on the coupling of superconducting qubits with other quantum modes/degrees mooij2 ; Yale2 ; Yale1 ; WangZ . Nevertheless, most interests have been focused on the design/implementation of single and multi-qubit logic gates, while few attention has been paid on quantum storage in superconducting qubits cleland .
As is well-known, memory is an indispensable part of information processing. Its quantum counterpart is even more important because of the fragility of quantum coherence. Roughly speaking, there are two kinds of quantum memory: a basic one is to temporarily store the intermediate computational results, just as the role played by the RAM (Random Access Memory) in classical computers; the other is used to store the ultimate results, playing a similar role of the classical hard disks. To fully accomplish quantum information processing, certain bus is required to transfer the information from these basic temporary memory units to other types of memory units as well as among themselves. Therefore, it is timely and significant to design basic storage units based on superconducting qubits and connect them via an appropriate bus to achieve a workable storage network. In this Letter, we design an experimentally feasible basic storage unit based on Josephson charge qubits and propose to couple them with a one-dimensional (1D) transmission line to physically realize a quantum storage network. The distinct advantages of our scheme include (i) the 1/f noise caused by background charge fluctuation may be significantly suppressed because the bias voltage for the charge qubit can be set to degeneracy point in the proposed storage process vion1 ; nec4 ; (ii) it is not necessary to adjust the magnetic flux instantaneously; (iii) in sharp contrast to dynamic quantum storage scenarios, no restriction has to be imposed on the initial state of our temporary memory units; and (iv) the relevant fabrication technique of the designed circuits are currently available. All of these enable our new scheme of quantum storage and information transfer to be more promising for the future solid state quantum computing.
A basic storage unit. A basic storage unit is designed to consist of three symmetrical dcSQUIDs as shown in Fig.1. The original Hamiltonian of the system includes Coulomb energy and Josephson coupling energy, i.e.,
$$H=H_c\underset{i=1}{\overset{3}{}}E_{Ji}\mathrm{cos}\pi \frac{\varphi _{xi}}{\mathrm{\Phi }_0}\mathrm{cos}\theta _i,$$
(1)
where $`E_{Ji}`$, $`\varphi _{xi}`$, and $`\theta _i`$ are the Josephson coupling energy, the magnetic flux, and the phase difference in the $`i`$-th SQUID, $`\mathrm{\Phi }_0=hc/2e`$ is the usual superconducting flux quantum. The Coulomb energy part $`H_c=E_{c1}(n_1n_{g1})^2+E_{c2}(n_2n_{g2})^2+4E_3(n_1n_{g1})(n_2n_{g2})`$. Here $`n_i`$ is the number of the excess Cooper pair in the $`i`$-th Cooper pair box and $`n_{gi}=C_{gi}V_{gi}/2e`$ with $`V_{gi}`$ and $`C_{gi}`$ as the corresponding gate voltage and capacitance. The coefficients $`E_\alpha `$ are derived as $`E_{c1}=2e^2C_{\mathrm{\Sigma }2}/\left(C_{\mathrm{\Sigma }1}C_{\mathrm{\Sigma }2}C_{J3}^2\right)`$, $`E_{c2}=2e^2C_{\mathrm{\Sigma }1}/\left(C_{\mathrm{\Sigma }1}C_{\mathrm{\Sigma }2}C_{J3}^2\right)`$, $`E_3=e^2C_{J3}/2\left(C_{\mathrm{\Sigma }1}C_{\mathrm{\Sigma }2}C_{J3}^2\right)`$ with $`C_{\mathrm{\Sigma }i}=C_{Ji}+C_{J3}+C_{gi}`$ as the summation of all the capacitances connected to the $`i`$-th Cooper pair box.
When $`E_{ci}E_{Ji}`$, the charging energy dominates the system and the state evolution is approximately confined in the two eigenstates of charge operator $`\{|0_i,|1_i\}`$. Then the Pauli operators can be introduced to express the dynamic variables. The reduced Hamiltonian becomes
$`H`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{2}{}}}\mathrm{\Omega }_i\sigma _{zi}+E_3\sigma _{z1}\sigma _{z2}{\displaystyle \underset{i=1}{\overset{2}{}}}E_{Ji}\mathrm{cos}\pi {\displaystyle \frac{\varphi _{xi}}{\mathrm{\Phi }_0}}\sigma _{xi}`$ (2)
$`E_{J3}\mathrm{cos}\pi {\displaystyle \frac{\varphi _{x3}}{\mathrm{\Phi }_0}}\left(\sigma _{x1}\sigma _{x2}\sigma _{y1}\sigma _{y2}\right),`$
where $`\mathrm{\Omega }_i=E_{ci}\left(n_{gi}\frac{1}{2}\right)+2E_3\left(n_{gj}\frac{1}{2}\right)`$ ($`ij`$). In the derivation of Eq.(2), we have used the constraint $`\theta _1+\theta _2+\theta _3=0`$. Here, the Pauli matrices are defined as $`\sigma _{xi}=|1_{ii}0|+|0_{ii}1|,`$ $`\sigma _{yi}=i(|1_{ii}0||0_{ii}1|)`$ and $`\sigma _{zi}=|0_{ii}0||1_{ii}1|`$ in the bases $`|1_i`$ and $`|0_i`$, which are the eigenstates of the number operator of Cooper pair on the $`i`$-th box with one and zero Cooper pair.
In this setup, the first SQUID is a computational qubit and the second one is used for storage, while the third one serves as the controllable coupling element between qubits 1 and 2. Prior to the storage process, the two qubits are set to be uncorrelated by simply letting $`\varphi _{x3}=\mathrm{\Phi }_0/2`$.
We now illustrate that the storage process begins whenever the flux in the third dcSQUID is switched away from $`\mathrm{\Phi }_0/2`$. In fact, the coupling between the two qubits is turned on for $`\varphi _{x3}\mathrm{\Phi }_0/2`$. If both of the bias voltages are set to let $`n_{g1}=n_{g2}=1/2`$ and the magnetic fluxes $`\varphi _{xi}`$ threading the first two SQUIDs equal to $`\mathrm{\Phi }_0/2`$, the first and third terms in Eq. (2) vanish. Moreover, if $`C_{\mathrm{\Sigma }i}/C_{J3}`$ ($`i=1`$ or $`2`$) is sufficiently large such that $`E_3E_{J3}`$, the third term in Eq. (2) is negligibly small (here we shall neglect it first for simplicity and address its influence on the results later). As a result, we have
$$H=E_{J3}\mathrm{cos}\pi \frac{\varphi _{x3}}{\mathrm{\Phi }_0}\left(\sigma _{x1}\sigma _{x2}\sigma _{y1}\sigma _{y2}\right)$$
(3)
Defining the Pauli operators of the second qubit in another representation $`\{|\stackrel{~}{1}_2,|\stackrel{~}{0}_2\}`$ with $`|\stackrel{~}{1}_2=|0_2,|\stackrel{~}{0}_2=|1_2`$, one has $`\sigma _{x2}=\stackrel{~}{\sigma }_{x2}`$, $`\sigma _{y2}=\stackrel{~}{\sigma }_{y2}`$, $`\sigma _{z2}=\stackrel{~}{\sigma }_{z2}`$. The corresponding Hamiltonian becomes
$$H=E_{J3}\left(\sigma _{x1}\stackrel{~}{\sigma }_{x2}+\sigma _{y1}\stackrel{~}{\sigma }_{y2}\right),$$
(4)
where we set $`\varphi _{x3}=0`$ to maximize the interaction strength between two qubits. This is a central result of the present work. It is remarkable that this controllable interaction is a typical $`XY`$-coupling of spin-1/2 systems often addressed in many-body spin physics; while to our knowledge, it is realized for the first time in sueprconducting qubits and thus is of great significance in solid state quantum information processing including quantum storage as the total effective ’spin’ is conserved with this interacting Hamiltonian. Besides, this controllable coupling may have applications in exploring in-depth spin physics.
It is straightforward to find the time evolution operator in the two qubit charge basis $`\{|00,|01,|10,|11\}`$ as
$$U\left(t\right)=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& \mathrm{cos}\xi \left(t\right)& i\mathrm{sin}\xi \left(t\right)& 0\\ 0& i\mathrm{sin}\xi \left(t\right)& \mathrm{cos}\xi \left(t\right)& 0\\ 0& 0& 0& 1\end{array}\right)$$
(5)
where $`\xi \left(t\right)=2E_{J3}t/\mathrm{}`$. We can see that at the time $`t=\pi \mathrm{}/\left(4E_{J3}\right)`$ the evolution leads to $`|00|00`$, $`|01i|10`$, $`|10i|01`$ and $`|11|11`$. That is to say, the quantum states of the two qubits are swapped (with an unimportant phase shift) makhlin2 . For example, if the density matrix of the first qubit is initially $`\rho _1\left(0\right)=_{n,m=0}^1c_{mn}|m_1n|`$ while the second qubit is prepared in $`|\stackrel{~}{0}_2`$, the final state at $`t=\pi \mathrm{}/4E_{J3}`$ is
$$\rho (t=\frac{\pi \mathrm{}}{4E_{J3}})=|0_10|\underset{n,m=0}{\overset{1}{}}c_{mn}\overline{|m}_2\overline{n|}$$
(6)
where $`\overline{|m}_2=e^{i\frac{\pi }{2}m}|\stackrel{~}{m}_2`$. Therefore the quantum information carried by the first qubit (the computational one) has been stored in the second one. In the meanwhile the first qubit is set to the ground state to prepare for the next round of computation. It is notable that this process may also be regarded as certain ”readout” process.
After the state of the computational qubit has been stored in the temporary memory, the flux threading the third SQUID is tuned back to be $`\mathrm{\Phi }_0/2`$ and the two qubits are decoupled. The first qubit can perform new computational task.
It is worth pointing out that the second qubit is not restricted to stay in its ground state. Actually our storage protocol works for any state of the second qubit even for the mixed state $`\rho _2\left(0\right)=_{n,m=0}^1a_{mn}\overline{|m}_2\overline{n|}`$. This feature is quite different from most existing dynamical storage schemes lukin1 ; cleland ; wang , in which a prerequisite is to prepare the storage qubit in the ground state. Also note that although some adiabatic quantum storage schemes sun ; liyong ; han do not have this restriction they are seriously flawed by the adiabatic condition that demands rather long time to complete the whole storage process.
Another advantage of this protocol is a comparatively loose requirement on the adjustment of the magnetic flux $`\varphi _{x3}`$ during the storage process. In most quantum computing proposals controlled by the magnetic flux, the instantaneous switch of magnetic flux is normally required. In our protocol, even if $`\varphi _{x3}`$ is dependent of $`t`$, rather than a step function, namely the Hamiltonian (3) depends on time, since $`H\left(t\right)`$ at different time commute with each other, the time dependence modifies only the definition of $`\xi \left(t\right)`$ in Eq. (5) as
$$\overline{\xi }\left(t\right)=2E_{J3}_0^t\mathrm{cos}\left(\pi \frac{\varphi _{x3}\left(t^{}\right)}{\mathrm{\Phi }_0}\right)𝑑t^{}.$$
(7)
In this case, one can adjust the storage time $`\tau `$ to satisfy $`\overline{\xi }\left(\tau \right)=\pi /2`$. As for the other external magnetic fluxes $`\varphi _{x1}`$ and $`\varphi _{x2}`$, it is obvious that they do not require the instantaneous manipulation.
An additional merit lies in that, the bias voltage is set to the degeneracy point during the whole storage process, which strongly suppresses the charge fluctuation induced 1/f noise, the most predominant resource of noise in Josephson charge qubits nec4 .
All of the above three distinct features make our protocol more arresting and fault-tolerant than most existing storage schemes. We also wish to remark that a two-qubit system similar to our setup nec2 ; nec3 and a three-junction loop circuit mooij1 have already been fabricated experimentally and illustrated to have good quantum coherence. Therefore the designed architecture of basic storage unit is likely experimentally feasible with current technology and thus is quite promising for near future experimental realization.
Information transfer between the units. Generally speaking, a computational task requires the cooperation of several (or more) qubits. The state of one qubit usually needs to be transferred to another in order to conduct further computations. Also, it is necessary to store the final results to certain physical system with longer coherence time. Therefore a storage network is indispensable in quantum information processing. One possible scenario to realize such a network is to use a common data bus with controllable coupling to all basic units. Through this data bus, the communication of any two basic units becomes feasible.
Currently, there are some alternative suggestions for possible common data buses including a microcavity, a nanomechanical resonator cleland , and a large junction etc. Another promising one is the so-called 1D transmission line Yale1 ; Yale2 , which has been illustrated to have several practical advantages including strong coupling strength, reproducibility, immunity to 1/f noise, and suppressed spontaneous emission Yale1 .
As an example, here we elaborate the transfer process with the 1D transmission line. Consider an array of identical basic units placed along a 1D transmission line (see Fig.2). The information stored in the second qubit of any unit can be transferred to another unit via the transmission line. The coupling between the transmission line and the units can be either electrical or magnetic. For concreteness, here we focus only on the magnetic coupling. Different from the 3D microcavity where the magnetic dipole interaction is usually too weak to be considered, the present interaction can be sufficiently strong to accomplish the transfer task by an appropriate design of the circuit.
For an ideal 1D transmission line with the boundary conditions $`j(0,t)=j(L,t)=0`$, the quantized magnetic field at $`x=nL/2n_0`$, where $`n_0`$ is the mode resonant with the qubits, $`n`$ is an arbitrary integer, and $`L`$ is the length of the line along the $`x`$-direction, is
$$B_y\left(x=\frac{n}{n_0}\frac{L}{2}\right)=\frac{1}{d}\sqrt{\frac{\mathrm{}l\omega _{n_0}}{L}}\left(a_{n_0}+a_{n_0}^{}\right),$$
(8)
while the electric field is zero at these points. Here $`\omega _{n0}=n_0\pi /\left(L\sqrt{lc}\right)`$, $`d`$ is the distance between the qubit and the transmission line, $`l`$ ($`c`$) the inductance (capacitance) per unit length. The flux induced by the transmission line in a dcSQUID with an enclosed area $`S`$ reads
$$\mathrm{\Phi }_x=\frac{S}{d}\sqrt{\frac{\mathrm{}l\omega _{n_0}}{L}}\left(a_{n_0}+a_{n_0}^{}\right).$$
(9)
It is a reasonable approximation to consider only the effect of the transmission line on the SQUID $`2`$ if the distance between the third (or first) SQUID is significantly longer than $`d`$ or we simply insert a magnetic shield screen (dotted line in Fig.2). Under this consideration and the Lamb-Dicke approximation ($`g1`$), the Hamiltonian for the qubit 2 in the $`k`$-th unit with $`\varphi _{x2}=\mathrm{\Phi }_0/2`$ becomes
$$H^{\left(k\right)}=\mathrm{\Omega }_2^{\left(k\right)}\sigma _{z2}gE_{J2}\left(a+a^{}\right)\sigma _{x2}^{\left(k\right)}+\mathrm{}\omega \left(a^{}a+\frac{1}{2}\right),$$
(10)
where $`g=S\sqrt{\mathrm{}l\omega }/\left(d\mathrm{\Phi }_0\sqrt{L}\right)`$ (here for simplicity we denote $`a_{n_0}`$as $`a`$ and $`\omega _{n0}`$ as $`\omega `$). During the storage process for the basic units, the second term in the above equation can be neglected because the qubit is largely detuned from the transmission line.
Under the condition $`|\mathrm{\Omega }_2^{\left(k\right)}\omega |/(\mathrm{\Omega }_2^{\left(k\right)}+\omega )1`$, the terms oscillating with the frequency $`\pm (\mathrm{\Omega }_2^{\left(k\right)}\omega )`$ are singled out under the rotating-wave-approximation, i.e.,
$$H^{\left(k\right)}=\mathrm{\Omega }_2^{\left(k\right)}\sigma _{z2}^{\left(k\right)}+\mathrm{}\omega a^{}a(gE_{J2}a\sigma _{+2}^{\left(k\right)}+h.c).$$
(11)
For each qubit, this is a typical Jaynes-Cummings model JC and there exist many two dimensional invariant subspaces. Driven by this Hamiltonian, if the qubit 2 of the $`k`$-th unit is resonant with the cavity by adjusting $`n_{g2}^{\left(k\right)}`$, any state of this qubit can be mapped onto the subspace $`\{|0_{\text{TLR}},|1_{\text{TLR}}\}`$ of the transmission line resonator wang . This information can also be retrieved by the qubit 2 of another $`k^{}`$-th unit. Consequently, the information carried by the $`k`$-th unit is transferred to the $`k^{}`$-th unit, with the whole process being detailed as below.
Prepare first the transmission line in its ground state $`|0`$. Tune $`n_{g2}^{\left(k\right)}`$ to have $`\mathrm{\Omega }_2^{\left(k\right)}=\omega `$ for a period $`\pi /2gE_{J2}`$ such that the state of the $`k`$-th unit is stored in the transmission line. Then let this qubit be largely detuned with the transmission line resonator while make the frequency of another qubit to satisfy $`\mathrm{\Omega }_2^{\left(k^{}\right)}=\omega `$ for another $`t=\pi /2gE_{J2}`$. This process can be explicitly illustrated as
$`\left(\alpha |1_2^{\left(k\right)}+\beta |0_2^{\left(k\right)}\right)|0_{\text{TLR}}|0_2^{\left(k^{}\right)}`$
$``$ $`|0_2^{\left(k\right)}\left(\beta e^{i\xi }|0_{\text{TLR}}i\alpha e^{i\xi }|1_{\text{TLR}}\right)|0_2^{\left(k^{}\right)}`$
$``$ $`|0_2^{\left(k\right)}|0_{\text{TLR}}\left(\alpha |1_2^{\left(k^{}\right)}+\beta |0_2^{\left(k^{}\right)}\right).`$
In this way the information is transferred from the $`k`$-th to the $`k^{}`$-th unit.
Discussions and remarks. To see the experimental feasibility, we now examine the used conditions and approximations based on the available/possible experimental parameters. We indeed verified that these conditions and approximations are reasonable and acceptable. For example, if we take $`C_{\mathrm{\Sigma }2}500`$aF, $`C_{J3}100`$aF, and $`C_{\mathrm{\Sigma }1}1\times 10^4`$aF, where the large capacitance of $`C_{\mathrm{\Sigma }1}`$ can be achieved by shunting an additional large capacitance (see Fig.1) and the small Josephson coupling energy of $`E_{J1}`$ may be realized by using the SQUID coupling. Then $`E_{c1}32\mu `$eV, $`E_{c2}640\mu `$eV, $`E_31.6\mu `$eV, $`E_{J2}100\mu `$eV, $`E_{J3}100\mu `$eV, and $`g0.1`$ g . With these parameters, we can see that $`E_3gE_{J2},E_{J3}`$ and the Lamb-Dicke approximation is also justified. Besides, the operation time is estimated to be $`30`$ps for one basic storage in a unit and $`1`$ns for one information transfer process, being much shorter than the coherence time for charge qubits at the degeneracy point ($`800`$ns currently). Therefore this process can be completed before the quantum decoherence happens.
Finally, we turn to address the effect of the $`E_3`$ neglected earlier. First, it is worthwhile to point out that even if $`E_3`$ is not negligible the basic unit part of our protocol still works. This is because an additional term $`E_3\sigma _{z1}\sigma _{z2}`$ commutes with Eq. (3), and thus just brings an additional phase to the storage process. Secondly, although this term represents also an unremovable correlation between the two qubits in one unit, fortunately, following the same technique used by the NEC group nec2 ; nec3 , a single qubit behavior can still be achieved in this system with an appropriate pulse, provided that $`E_3`$ is small. This setting makes the two qubits approximately independent. On the other hand, the transfer process may not be implemented successfully if $`E_3`$ is not so small. In this case, the first qubit of a unit has to be set in a certain state when the second qubit is transferring information to the transmission line, though this may reduce the efficiency of the transfer process.
We thank useful discussions with Y.X. Liu, J. You, and Y. Yu. The work was supported by the RGC grants of Hong Kong (HKU7114/02P and HKU7045/05P), the URC fund of HKU, and the NSFC grants (10429401).
|
warning/0506/quant-ph0506122.html
|
ar5iv
|
text
|
# A Quantum–Classical Bracket from 𝑝-Mechanics
## 1 Introduction
There is a strong and persistent interest over decades in *a self-consistent model for an aggregate system, which combines components with both quantum and classical behaviour* (see and references therein). There are various reasons for such an interest. Firstly, there are many questions where considerations of quantum-classical aggregates are unavoidable, e.g. measurement of a quantum system by a classical apparatus, a quantum particle in the classical gravity field, etc. Secondly, even for purely quantum conglomerates we expect that a quantum-classical approximation may be easier for investigation than the purely quantum picture. Thus it is natural that models of quantum-classical interaction became of separate theoretical interest.
The discussion is typically linked to a search of *quantum-classical bracket* which should combine properties of the quantum commutator $`[,]`$ and Poisson’s bracket $`\{,\}`$ in the corresponding sectors. Some simple algebraic combinations like
$$\frac{1}{\mathrm{i}h}[A,B]+\frac{1}{2}(\{A,B\}\{B,A\})$$
(1)
were guessed during the last twenty five years but neither of them turned to be completely satisfactory. Moreover several ‘‘no-go’’ theorems in that direction were proved over the last ten years . Thus the prevailing opinion now is that no consistent quantum-classical bracket is possible. However the explicit similarity between the Hamiltonian descriptions of quantum and classical dynamics repeatedly undermine such a believe.
This paper builds a consistent quantum-classical bracket within the framework of $`p`$-mechanics . This approach is based on the representation theory of nilpotent Lie groups (the Heisenberg group $`^n`$ in the first instance) and naturally embeds both quantum and classical descriptions. $`p`$-Mechanical observables are convolutions on a nilpotent group $`G`$ and contain both classical and quantum pictures for all values of Planck’s constants at the same time. These pictures can be separated by a restriction of $`p`$-observables to irreducible representations of $`G`$, e.g. by considering their actions on $`p`$-mechanical states .
The important step is the definition of the universal bracket between convolutions on the Heisenberg group, which are transformed by the above mentioned representations into the quantum commutator (Moyal bracket ) and the Poisson bracket correspondingly. Consequently it is sufficient to solve the dynamic equation in $`p`$-mechanics in order to obtain both quantum and classical dynamics. Since the universal bracket is based on the usual commutator of convolutions (i.e. inner derivations of the convolution algebra) they satisfy all important requirements, i.e. linearity, antisymmetry, Leibniz and Jacoby identities . Moreover due to presence of antiderivative operator (12) the universal bracket with a Hamiltonian has the dimensionality of time derivative . This approach was extended to quantum field theory in . A brief account of $`p`$-mechanics is provided in the first part of this paper.
To construct quantum-classical bracket we develop $`p`$-mechanics on the group $`𝔻^n`$, which is the product of two copies of the Heisenberg group $`^n`$. The group $`𝔻^n`$ was already used to this end in our earlier paper but the right bracket was not derived there due to several reasons: the derivation followed the notorious semiclassical limit procedure; the universal bracket was not known at that time. A correct derivation of quantum-classical bracket in the consistent $`p`$-mechanical framework is given in the second part of the present note. This bracket (26) includes as a part the Aleksandrov’s bracket (1) together with an extra term of analytical nature, which involves derivative with respect to the second Planck’s constant, see (26). This analytic term escapes all previous purely algebraic considerations and ‘‘no-go’’ theorems for the obvious reasons.
Future investigations of these new quantum-classical bracket will be given elsewhere.
## 2 The Heisenberg group and $`p`$-mechanical bracket
### 2.1 The Heisenberg group and its representations
Let $`(s,x,y)`$, where $`x`$, $`y^n`$ and $`s`$, be an element of the Heisenberg group $`^n`$ . The group law on $`^n`$ is given as follows:
$$(s,x,y)(s^{},x^{},y^{})=(s+s^{}+\frac{1}{2}\omega (x,y;x^{},y^{}),x+x^{},y+y^{}),$$
(2)
where the non-commutativity is due to $`\omega `$—the *symplectic form* $`\omega (x,y;x^{},y^{})=xy^{}x^{}y`$ on $`^{2n}`$ \[3, § 37\]. The Lie algebra $`𝔥^n`$ of $`^n`$ is spanned by left-invariant vector fields
$$S=_s,X_j=_{x_j}\frac{1}{2}y_j_s,Y_j=_{y_j}+\frac{1}{2}x_j_s$$
(3)
on $`^n`$ with the Heisenberg *commutator relations* $`[X_i,Y_j]=\delta _{i,j}S`$ and all other commutators vanishing. There is the *co-adjoint representation* \[11, § 15.1\] $`\mathrm{Ad}^{}:𝔥_n^{}𝔥_n^{}`$ of $`^n`$:
$$\mathrm{ad}^{}(s,x,y):(h,q,p)(h,q+hy,phx),$$
(4)
where $`(h,q,p)𝔥_n^{}`$ in bi-orthonormal coordinates to the exponential ones on $`𝔥^n`$. There are two types of orbits in (4) for $`\mathrm{Ad}^{}`$, i.e. Euclidean spaces $`^{2n}`$ and single points:
$`𝒪_h`$ $`=`$ $`\{(h,q,p):\text{ for }h0,(q,p)^{2n}\},𝒪_{(q,p)}=\{(0,q,p):\text{ for }(q,p)^{2n}\}.`$ (5)
All representations are *induced* \[11, § 13\] by a character $`\chi _h(s,0,0)=e^{2\pi \mathrm{i}hs}`$ of the centre of $`^n`$ generated by $`(h,0,0)𝔥_n^{}`$ and shifts (4) from the *left* on orbits (5). The explicit formula respecting *physical units* is:
$$\rho _h(s,x,y):f_h(q,p)e^{2\pi \mathrm{i}(hs+qx+py)}f_h(q\frac{h}{2}y,p+\frac{h}{2}x).$$
(6)
The Stone–von Neumann theorem \[11, § 18.4\], \[9, Chap. 1, § 5\] describes all unitary irreducible representations of $`^n`$ parametrised up to equivalence by two classes of orbits (5):
* The infinite dimensional representations by transformation $`\rho _h`$ (6) for $`h0`$ in Fock space $`F_2(𝒪_h)L_2(𝒪_h)`$ of null solutions of Cauchy–Riemann type operators .
* The one-dimensional representations which drops out from (6) for $`h=0`$:
$$\rho _{(q,p)}(s,x,y):ce^{2\pi \mathrm{i}(qx+py)}c.$$
(7)
Commutative representations (7) are oftenly forgotten, however their union naturally (see the appearance of Poisson bracket in (15)) act as the classic *phase space*: $`𝒪_0=_{(q,p)^{2n}}𝒪_{(q,p)}.`$
### 2.2 Convolutions (observables) on $`^n`$ and commutator
Using a left invariant measure $`dg`$ on $`^n`$ the linear space $`L_1(^n,dg)`$ can be upgraded to an algebra with the convolution:
$`(k_1k_2)(g)`$ $`=`$ $`{\displaystyle _^n}k_1(g_1)k_2(g_1^1g)𝑑g_1.`$ (8)
Convolutions on $`^n`$ are *observables* in $`p`$-mechanic . Inner *derivations* $`D_k`$ of the convolution algebra $`L_1(^n)`$ are given by the *commutator*:
$`D_k:f[k,f]`$ $`=`$ $`kffk{\displaystyle _^n}k(g_1)\left(f(g_1^1g)f(gg_1^1)\right)𝑑g_1.`$ (9)
A unitary representation $`\rho _h`$ of $`^n`$ extends to $`L_1(^n,dg)`$:
$$\rho _h(k)=_^nk(g)\rho _h(g)𝑑g.$$
(10)
Thus $`\rho _h(k)`$ for a fixed $`h0`$ depends only on $`\widehat{k}_s(h,x,y)=k(s,x,y)e^{2\pi \mathrm{i}hs}𝑑s`$—the partial Fourier transform $`sh`$ of $`k(s,x,y)`$. Consequently the representation of commutator (9) depends only on:
$`[k^{},k]\widehat{_s}`$ $`=`$ $`2\mathrm{i}{\displaystyle _{^{2n}}}\mathrm{sin}(\pi h(xy^{}yx^{}))\widehat{k}_s^{}(h,x^{},y^{})\widehat{k}_s(h,xx^{},yy^{})𝑑x^{}𝑑y^{},`$ (11)
which is exactly the Moyal bracket for the full Fourier transforms of $`k^{}`$ and $`k`$. Also it vanishes for $`h=0`$ as can be expected from the commutativity of representations (7).
### 2.3 $`p`$-Mechanical bracket on $`^n`$
An antiderivative $`𝒜`$ is a scalar multiple of a right inverse operator to the vector field $`S𝔥^n`$ (3):
$$S𝒜=4\pi ^2I,\text{ or }𝒜e^{2\pi \mathrm{i}hs}=\{\begin{array}{cc}\frac{2\pi }{\mathrm{i}h}e^{2\pi \mathrm{i}hs},\hfill & \text{if }h0,\hfill \\ 4\pi ^2s,\hfill & \text{if }h=0.\hfill \end{array}$$
(12)
It can be extended by the linearity to $`L_1(^n)`$. We introduce $`p`$-mechanical bracket on $`L_1(^n)`$ as a modified commutator of observables:
$$\{[k_1,k_2]\}=(k_1k_2k_2k_1)𝒜.$$
(13)
Then from (10) one gets $`\rho _h(𝒜k)=(ih)^1\rho _h(k)`$ for $`h0`$. Consequently the modification of (11) for $`h0`$ is only slightly different from the original one:
$`\{[k^{},k]\}\widehat{_s}`$ $`=`$ $`{\displaystyle _{^{2n}}}{\displaystyle \frac{2\pi }{h}}\mathrm{sin}(\pi h(xy^{}yx^{}))\widehat{k}_s^{}(h,x^{},y^{})\widehat{k}_s(h,xx^{},yy^{})𝑑x^{}𝑑y^{}.`$ (14)
However the last expression for $`h=0`$ is significantly distinct from (11), which vanishes as noted above. From the natural assignment $`\frac{4\pi }{h}\mathrm{sin}(\pi h(xy^{}yx^{}))=4\pi ^2(xy^{}yx^{})`$ for $`h=0`$ we get the Poisson bracket for the Fourier transforms of $`k^{}`$ and $`k`$ defined on $`𝒪_0`$:
$$\rho _{(q,p)}\{[k^{},k]\}=\frac{\widehat{k}^{}}{q}\frac{\widehat{k}}{p}\frac{\widehat{k}^{}}{p}\frac{\widehat{k}}{q}.$$
(15)
Furthermore the dynamical equation
$$\dot{f}=\{[H,f]\}$$
(16)
based on the bracket (13) with a Hamiltonian $`H(g)`$ for an observable $`f(g)`$ is reduced to Moyal’s and Poisson’s equations by $`\rho _h`$ with $`h0`$ and $`h=0`$ correspondingly. The same connections are true for the solutions of these three equations, see .
## 3 Mixed Quantum-Classical Bracket
### 3.1 A nilpotent group with two dimensional centre
To derive quantum-classical bracket we again use the ‘‘quantum-classical’’ group $`𝔻^n=^n^n`$ . This is a step 2 nilpotent Lie group of the (real) dimension $`4n+2`$. The group law is given by the formula:
$`(g_1;g_2)(g_1^{};g_2^{})=(g_1g_1^{};g_2g_2^{}),`$ (17)
where $`g_i^{()}=(s_i^{()},x_i^{()},y_i^{()})^n`$, $`i=1,2`$ and products $`g_ig_i^{}`$ are the same as in (2).
The group $`𝔻^n`$ has a two-dimensional centre $`=\{(s_1,0,0;s_2,0,0)s_1,s_2\}`$. The irreducible representations of a nilpotent group $`𝔻^n`$ are induced \[11, § 13.4\], \[20, § 6.2\] by the characters of the centre: $`\mu :(s_1,s_2)\mathrm{exp}(2\pi \mathrm{i}(h_1s_1+h_2s_2)).`$ For $`h_1h_20`$ the induced representation coincides with the irreducible representation of $`^{n+n}`$:
$$\rho _{(h_1;h_2)}(g_1;g_2)=\rho _{h_1}(g_1)\rho _{h_2}(g_2)$$
(18)
This corresponds to *purely quantum* behavior of both sets of variables $`(x_1,y_1)`$ and $`(x_2,y_2)`$. The trivial character $`h_1=h_2=0`$ gives the family of one-dimensional (*purely classical*) representations parametrised by points of $`^{4n}`$:
$$\rho _{(q_1,p_1;q_2,p_2)}(s_1,x_1,y_1;s_2,x_2,y_2)=e^{2\pi \mathrm{i}(x_1p_1+y_1q_1+x_2p_2+y_2q_2)}$$
(19)
These cases for $`^n`$ were described above and studied in details in . A new situation appears when $`h_10`$ and $`h_2=0`$ corresponding to quantum behavior for $`(x_1,y_1)`$ and classical behavior for $`(x_2,y_2)`$. The choice $`h_1=0`$, $`h_20`$ swaps the quantum and classical parts. The quantum-classical representation is given by
$`\rho _{(h;q,p)}(g_1;g_2)`$ $`=`$ $`\rho _h(g_1)\rho _{(q,p)}(g_2)=\rho _h(s_1,x_1,y_1)e^{2\pi \mathrm{i}(qx_2+py_2)},`$ (20)
where $`q,p^n`$ and $`h\{0\}`$. In this representation a convolution (observable) on $`𝔻^n`$ generates a function on the classic phase space $`^{2n}`$ with values in space of quantum operators acting of $`L_2(^n)`$, cf. , or explicitly:
$`\rho _{(h;q,p)}k`$ $`=`$ $`{\displaystyle _{𝔻^n}}k(g_1;g_2)\rho _h(g_1)e^{2\pi \mathrm{i}(qx_2+py_2)}𝑑g_1𝑑g_2`$
$`=`$ $`{\displaystyle _^n}\widehat{k}_2(s_1,x_1,y_1;0,q,p)\rho _h(s_1,x_1,y_1)𝑑g_1`$
where $`\widehat{k}_2`$ is the partial Fourier transform of $`k`$ with respect to variables $`(s_2,x_2,y_2)(h,q,p)`$.
### 3.2 The Mixed Bracket
We define $`p`$-bracket in the case of $`𝔻^n`$ similarly to (13). Although this is not a unique option, some other similar definitions may be of interest as well.
###### Definition 1.
The *$`p`$-mechanical bracket* of two convolutions (observables) $`k_1(g_1;g_2)`$ and $`k_2(g_1;g_2)`$ on the group $`𝔻^n`$ is defined as follows:
$$\{[k_1,k_2]\}=(k_1k_2k_2k_1)(𝒜_1+𝒜_2),$$
(22)
where $``$ denotes the group convolution on $`𝔻^n`$. $`𝒜_1`$ and $`𝒜_2`$ are antiderivatives with respect to the variable $`s_1`$ and $`s_2`$ correspondingly, cf. (12).
Consistence of this definition, cf. , is given by:
###### Lemma 2.
The $`p`$-mechanical bracket (22) is linear, antisymmetric, satisfy Leibniz and Jacoby identities. Moreover $`p`$-mechanical bracket with a Hamiltonian has the dimensionality of time derivative.
We define *$`p`$-mechanisation* of a classical observable $`f(q,p)`$ is given by the Weyl (symmetrized) calculus defined on the generators as follows:
$$q_jQ_j=\delta _{x_j}^{}(g_1;g_2),p_1P_j=\chi _{s_k}^{}(s_1+s_2)\delta _{y_j}^{}(g_1;g_2),j=1,2\text{ and }k=3j,$$
(23)
where $`\delta _z^{}`$ is the derivative of the Dirac delta function with respect to the variable $`z`$ and $`\chi _{s_k}^{}`$ is the derivative of the Heaviside step function such that $`\chi _z^{}(z)=\delta (z)`$. Using the identity
$`{\displaystyle _{}}\chi _{s_k}^{}(s_1+s_2)e^{2\pi \mathrm{i}(h_1s_1+h_2s_2)}𝑑z={\displaystyle \frac{h_k}{h_1+h_2}}\text{ we get that }\{[Q_i,P_j]\}=\delta _{ij}I,`$ (24)
and all other brackets vanish. Representations of distributions (23) and the bracket (22) are:
$$\begin{array}{ccccc}& \rho _{(h_1;h_2)}& \rho _{(h;q,p)}& \hfill _{h_2}\rho _{(h;q,p)}|_{h_2=0}& \rho _{(q_1,p_1;q_2,p_2)}\\ & & & & \\ & & & & \\ Q_j& \frac{}{}_{x_j}\frac{\mathrm{i}h_j}{2}y_j& \begin{array}{c}_{x_1}\mathrm{i}hy_1/2\\ \mathrm{i}q\end{array}& \hfill \begin{array}{cc}0,\hfill & \text{if }j=1\hfill \\ _p/2,\hfill & \text{if }j=2\hfill \end{array}& q_j\\ & & & & \\ P_j& \frac{}{}\frac{h_k}{h_1+h_2}\left(_{y_j}+\frac{\mathrm{i}h_j}{2}x_j\right)& \begin{array}{c}0\\ \mathrm{i}p\end{array}& \hfill \begin{array}{cc}_{y_1}/h+\mathrm{i}x_1/2,\hfill & \text{if }j=1\hfill \\ \mathrm{i}p/h_q/2,\hfill & \text{if }j=2\hfill \end{array}& p_j\\ & & & & \\ \{[K_1,K_2]\}& \frac{}{}\left(\frac{1}{\mathrm{i}h_1}+\frac{1}{\mathrm{i}h_2}\right)[K_1,K_2]& \multicolumn{2}{c}{[K_1,K_2]_{qc}}& \{\widehat{k}_1,\widehat{k}_2\}\end{array}$$
(25)
where the bracket $`[,]_{qc}`$ in the quantum-classical case is defined by the expression:
$`[K_1,K_2]_{qc}`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{i}h}}[K_1,K_2]+{\displaystyle \frac{1}{2}}\left(\{K_1,K_2\}\{K_2,K_1\}\right)\mathrm{i}_{h_2}[K_1,K_2]|_{h_2=0}.`$ (26)
Calculations of the two first terms in (26) is similar to $`p`$-bracket , the third term is:
$`{\displaystyle \underset{𝔻^n}{}}{\displaystyle \underset{𝔻^n}{}}(k_1(g_1^{};g_2^{})k_2(g_1^1g_1;g_2^{\prime \prime })k_2(g_1^{};g_2^{})k_1(g_1^1g_1;g_2^{\prime \prime }))`$
$`\times (s_2^{\prime \prime }+s_2^{})e^{2\pi \mathrm{i}(qx_2^{}+py_2^{}+qx_2^{\prime \prime }+py_2^{\prime \prime })}dg_2^{}dg_2^{\prime \prime }dg_1^{}\rho _h(g_1)dg_1.`$
The complete derivation will be given elsewhere. The derivative $`_{h_2}`$ in (26) highlights the important difference between Aleksandrov’s and our approach: *quantum-classical observables are not operator valued functions on the classical phase space but rather first jets *\[*16\] of such functions*. This explain the appearance of the fourth column in (25).
By algebraic inheritance the quantum-classic bracket (26) enjoys all the properties from Lem. 2. Moreover quantum-classical bracket coincides with the Moyal bracket for purely quantum observables and the Poisson bracket for purely classical ones. Let a $`p`$-mechanical observable $`f(t;g_1;g_2)`$, which is a function on $`\times 𝔻^n`$, be a solution of the equation:
$$\frac{d}{dt}f(t;g_1;g_2)=\{[f,H]\}$$
(27)
with a Hamiltonian $`H(g_1,g_2)`$ on $`𝔻^n`$. Then $`f(t;g_1;g_2)`$ provide consistent dynamics (in the sense of ) under either representation (18)–(20).
###### Example 3 (Dynamics with two different Planck’s constants, cf. ).
Let $`p`$-mechanical Hamiltonian is defined by such a distribution on $`𝔻^n`$ (see definitions (23)):
$`H`$ $`=`$ $`Q_1P_2Q_2P_1=\chi _{s_1}^{}(s_1+s_2)\delta _{x_1,y_2}^{(2)}(g_1;g_2)\chi _{s_2}^{}(s_1+s_2)\delta _{x_2,y_1}^{(2)}(g_1;g_2).`$ (28)
In the classic-classical representation (19) it produces (see the last column of (25)) the quadratic Hamiltonian $`H_{cc}=q_1p_2q_2p_1`$, which generates a simple rotational dynamic:
$`q_1(t)=\mathrm{cos}tq_1(0)+\mathrm{sin}tq_2(0),`$ $`q_2(t)=\mathrm{sin}tq_1(0)+\mathrm{cos}tq_2(0),`$ (29)
$`p_1(t)=\mathrm{cos}tp_1(0)+\mathrm{sin}tp_2(0),`$ $`p_2(t)=\mathrm{sin}tp_1(0)+\mathrm{cos}tp_2(0).`$ (30)
In the quantum-quantum representation (18) defined by two Planck’s constants $`h_1`$ and $`h_2`$ ($`h_1h_20`$) the Hamiltonian becomes (see the first column of (25)):
$$H_{qq}=\frac{h_1}{h_1+h_2}\left(_{x_1}\frac{\mathrm{i}h_1}{2}y_1\right)\left(_{y_2}+\frac{\mathrm{i}h_2}{2}x_2\right)\frac{h_2}{h_1+h_2}\left(_{x_2}\frac{\mathrm{i}h_2}{2}y_2\right)\left(_{y_1}+\frac{\mathrm{i}h_1}{2}x_1\right).$$
The dynamic from the bracket $`\left(\frac{1}{\mathrm{i}h_1}+\frac{1}{\mathrm{i}h_2}\right)[H_{qq},f]`$ in (25) is the coordinate map on $`𝔻^n`$:
$`x_1(t)=\mathrm{cos}tx_1(0)+\mathrm{sin}tx_2(0),`$ $`x_2(t)=\mathrm{sin}tx_1(0)+\mathrm{cos}tx_2(0),`$ (31)
$`h_2y_1(t)=h_2\mathrm{cos}ty_1(0)+h_1\mathrm{sin}ty_2(0),`$ $`h_1y_2(t)=h_2\mathrm{sin}ty_1(0)+h_1\mathrm{cos}ty_2(0).`$ (32)
For $`h_1=h_2`$ this coincides with the standard quantisation of the classical dynamics (29)–(30).
The quantum-classic Hamiltonian is the $`1`$-jet (see the middle column of (25)):
$$H_{qc}=(\mathrm{i}(_{x_1}\frac{\mathrm{i}h}{2}y_1)p,\frac{\mathrm{i}}{h}\left(_{x_1}\frac{\mathrm{i}h}{2}y_1\right)\left(p\frac{\mathrm{i}h}{2}_q\right)\frac{\mathrm{i}}{h}q\left(_{y_1}+\frac{\mathrm{i}h}{2}x_1\right)).$$
Note, that Aleksandrov’s bracket (1) of $`H_{qc}`$ with $`\rho _{(h;q,p)}(Q_1)`$ vanish and thus do not generate any dynamics for this observable. However $`[H_{qq},\rho _{(h_1;h_2)}(Q_1)]=\frac{\mathrm{i}h_1h_2}{h_1+h_2}\rho _{(h_1;h_2)}(Q_2)`$ and thus the third term in the bracket (26) of $`H_{qc}`$ and $`\rho _{(h;q,p)}(Q_1)`$ is equal to $`\rho _{(h;q,p)}(Q_2)=\mathrm{i}q`$ (this is also the value of the entire bracket (26)). Together with the value of $`[H_{qc},\rho _{(h;q,p)}(Q_2)]_{qc}=\rho _{(h;q,p)}(Q_1)`$ this defines quantum-classic dynamics of coordinates as the partial Fourier transform $`x_2q`$ of the quantum-quantum coordinate map (31).
Similarly we calculate that $`[H_{qc},\rho _{(h;q,p)}(P_1)]_{qc}=\rho _{(h;q,p)}(P_2)=\mathrm{i}p`$ and $`[H_{qc},\rho _{(h;q,p)}(P_2)]_{qc}=\rho _{(h;q,p)}(P_1)`$. Note that $`\rho _{(h;q,p)}(P_1)`$ is the $`1`$-jet with the value $`(0,\frac{1}{h}_{y_1}+\frac{\mathrm{i}}{2}x_1)`$ according to the (25) and the quantum-classic bracket depends from its both components. The quantum-classic dynamic of momenta is obtained from (32) by prolongation into the $`1`$-jet space with respect to the variable $`h_2`$ at point $`h_2=0`$.
## 4 Conclusion
The sum (1) of first two terms in (26) was proposed as a version of quantum-classical bracket. It was also obtained by approximation arguments within $`p`$-mechanical approach in as a part of the true bracket unknown at that time. However the expression (1) violates the Jacobi identity and Leibniz rule (i.e. is not a derivative), as a consequence it could not be used for a consistent dynamic equation . Our new bracket (26) has one extra term which makes it satisfactory to this end. This term is of an analytical nature (i.e. involves a derivative in Planck’s constant) and is hard to guess from algebraic manipulations with the quantum commutator and Poisson’s bracket. For the same reasons our bracket (26) are immunised against the ‘‘no-go’’ theorem of the type proved in . We present an example of a dynamics (31)–(32), which mixes two quantum sectors with different Planck’s constants, and demonstrate the quantum-classic dynamics in the $`1`$-jet space.
I am grateful to Prof. G. Ingold and anonymous referees who helped to improve this letter.
|
warning/0506/gr-qc0506102.html
|
ar5iv
|
text
|
# Absolute Objects and Counterexamples: Jones-Geroch Dust, Torretti Constant Curvature, Tetrad-Spinor, and Scalar Density
## 1 Introduction
James L. Anderson analyzed the novelty of Einstein’s so-called General Theory of Relativity (GTR) as its lacking “absolute objects” \[Anderson, 1964, Anderson, 1967, Anderson, 1971\]. Metaphorically, absolute objects are often described as a fixed stage on which the dynamical actors play their parts. A review of Anderson’s definitions will be useful. Absolute objects are to be contrasted with dynamical objects. The values of the absolute objects do not depend on the values of the dynamical objects, but the values of the dynamical objects do depend on the values of the absolute objects \[Anderson, 1967, p. 83\]. Both absolute objects and dynamical objects are, mathematically speaking, geometrical objects or parts thereof; the importance of this requirement will appear later. It will be shown that Anderson’s definition is naturally amended to avoid the Jones-Geroch dust counterexample more or less as Friedman envisioned. The hitherto-unnoticed fact that Anderson’s analysis detects an absolute object in Torretti’s constant curvature spaces example blunts its force noticeably. A previously unnoticed counterexample involving spinor fields is proposed and tentatively resolved using an alternative spinor formalism. As a referee (who turns out to be Robert Geroch) suggested, however, GTR actually has an absolute object, using Friedman’s definition and an attractive choice of variables. Whether it is better to revise the notion of absolute object or revise that claim that GTR lacks them is presently unclear.
Before absolute objects can be defined, the notion of a covariance group must be outlined. Here it will prove helpful to draw upon the unjustly neglected work of Kip Thorne, Alan Lightman, and David Lee (TLL) \[Thorne et al., 1973\]; a useful companion paper (LLN) was written by Lee, Lightman and W.-T. Ni \[Lee et al., 1974\]. The TLL definition differs slightly from Anderson’s in its notion of faithfulness. According to TLL,
> A group $`𝒢`$ is a covariance group of a representation if (i) $`𝒢`$ maps \[kinematically possible trajectories\] of that representation into \[kinematically possible trajectories\]; (ii) the \[kinematically possible trajectories\] constitute “the basis of a faithful representation of $`𝒢`$” (i.e., no two elements of $`𝒢`$ produce identical mappings of the \[kinematically possible trajectories\]); (iii) $`𝒢`$ maps \[dynamically possible trajectories\] into \[dynamically possible trajectories\]. \[Thorne et al., 1973, p. 3567\]
One can now define absolute objects. They are, according to Anderson, objects with components $`\varphi _\alpha `$ such that
> (1) The $`\varphi _\alpha `$ constitute the basis of a faithful realization of the covariance group of the theory. (2) Any $`\varphi _\alpha `$ that satisfies the equations of motion of the theory appears, together with all its transforms under the covariance group, in every equivalence class of \[dynamically possible trajectories\]. \[Anderson, 1967, p. 83\]
Thus the components of the absolute objects are the same, up to equivalence under the covariance group,<sup>2</sup><sup>2</sup>2There seems to be no compelling reason to require a covariance group instead of a mere covariance groupoid, a structure that would be a group if it were meaningful to multiply every pair of elements. Einstein’s equations on a background space-time, once one imposes a consistent notion of causality, have a covariance groupoid that is not a group \[Pitts and Schieve, 2004\]. in every model of the theory. It is the dynamical objects that distinguish the different equivalence classes of the dynamically possible trajectories \[Anderson, 1967, p. 84\]. One notices that the components of the absolute object need be the same, up to equivalence under the covariance group, for all *dynamically* possible trajectories, not all *kinematically* possible trajectories. Might this matter have gone otherwise? For most purposes this choice makes no difference, because typically those objects whose components are the same for all dynamically possible trajectories share the same feature for all kinematically possible trajectories. This condition fails, however, in the context of Rosen’s and Sorkin’s deriving the flatness of a metric using a variational principle with a Lagrange multiplier, as will appear below.
It has been asserted that the novel and nontrivial sense in which GTR is generally covariant is its lack of absolute objects \[Anderson, 1967\] or “prior geometry” \[Misner et al., 1973, pp. 429-431\]. John Norton discusses this claim with some sympathy \[Norton, 1992, Norton, 1993, Norton, 1995\], though technical problems such as the Jones-Geroch dust and Torretti constant spatial curvature counterexamples are among his worries \[Norton, 1993, Norton, 1995\]. Anderson and Ronald Gautreau encapsulate the definition of an absolute object as an object that “affects the behavior of other objects but is not affected by these objects in turn.” \[Anderson and Gautreau, 1969, p. 1657\] Depending on how one construes “affects,” this summary might be serviceable, but only if used very cautiously. On other occasions absolute objects are said to “influence” dynamical objects but not *vice versa* \[Anderson, 1971, p. 169\]. Such terminology echoes Einstein and implies that absolute objects violate what Anderson calls a “generalized principle of action and reaction” \[Anderson, 1967, p. 339\] \[Anderson, 1971, p. 169\]. Norton has argued, rightly I think, that such a principle is hopelessly vague and arbitrary and that it should not be invoked to impart a spurious necessity to the contingent truth that our best current physical theory lacks them \[Norton, 1993, pp. 848, 849\]. One might also doubt whether terms such as “affects,” “influence” and “act” adequately capture what absolute objects typically do. These terms suggest that the dynamical objects in question would have well-defined behavior if the absolute objects could somehow be ‘turned off,’ so to speak (perhaps by replacing them with zero in the equations of motion), and that if the absolute objects were ‘turned on’ again, they would alter the well-defined behavior of the dynamical objects in much the way that an applied electric field alters the motion of a charged particle. But in important examples, such as Newtonian physics or special relativity, turning off many or all of the absolute objects destroys the theory: the equations of motion become degenerate or meaningless. The absolute objects do not so much alter an otherwise happy situation as provide conditions in which the dynamical objects can have well-defined behavior. Perhaps the stage metaphor for absolute objects is deeper than it seemed: presumably actors could put on a play on a stage consisting of a rubbery sheet or a giant pillow, or perhaps act in mid-air while falling freely, but it is easier to act on a firm wooden stage. Thus the claim that absolute objects have some defect knowable *a priori* easily may be taken too seriously. The fact that it is even possible to do without them, as supposedly holds in Einstein’s theory, should be something of a surprise (but instead turns out to be false, in light of the scalar density counterexample, on Friedman’s definition).
In Anderson’s framework, an important subgroup of a theory’s covariance group is its symmetry group \[Anderson, 1967, pp. 84-88\]. One first defines the symmetry group of a *geometrical object* as those transformations that leave the object unchanged. If the transformations are infinitesimal space-time mappings, then the Lie derivative of the geometrical object with respect to the relevant vector field vanishes for symmetries. The symmetry group of a physical system or theory—Anderson makes no distinction between them here—is
> the largest subgroup of the covariance group of this theory, which is simultaneously the symmetry group of its absolute objects. In particular, if the theory has no absolute objects, then the symmetry group of the physical system under consideration is just the covariance group of this theory. \[Anderson, 1967, p. 87\]
Thus, roughly speaking, the fewer absolute objects a theory has, the more of its covariance transformations are symmetry transformations. For the example of a massive real scalar field obeying the Klein-Gordon equation in flat space-time in arbitrary coordinates, the covariance group is the group of diffeomorphisms, while the symmetry group is the 10-parameter Poincaré group corresponding to the ten Killing vector fields of Minkowski space-time. For a massive real scalar field coupled to gravity in GTR, the covariance group is again the diffeomorphisms. The symmetry group is also the diffeomorphisms, because any diffeomorphism leaves the set of absolute objects invariant, trivially, because there are no absolute objects (or so one thought until Geroch’s scalar density counterexample appeared). The fact that the space-time metric in GTR + massive real scalar field has no symmetries in general, though quite true, plays no explicit role in determining the symmetry group of the theory insofar as the space-time metric is dynamical rather than absolute.
Finding Anderson’s definition obscure, Michael Friedman amended it in the interest of clarity \[Friedman, 1973, Friedman, 1983\]. Friedman takes his definition to express Anderson’s intuitions, so the target of analysis is shared between them. As it turns out, Friedman has made a number of changes to Anderson’s definitions, most of which seem to have received little comment by him or others, so some comparison will be worthwhile.
First, though Friedman’s and Anderson’s equivalence relations are laid out somewhat differently, a key difference between them is that Friedman’s equivalence relation, which he calls $`d`$-equivalence, comprises only diffeomorphism freedom \[Friedman, 1983, pp. 58-60\], not other kinds of gauge freedom such as local Lorentz freedom or electromagnetic or Yang-Mills gauge freedom, in defining the covariance group. But local Lorentz freedom is a feature of the standard version of Einstein’s GTR + spinors, for example. Anderson calls such groups besides diffeomorphisms “internal groups” \[Anderson, 1967, pp. 35, 36\], though the term does not always fit perfectly for the examples available today.<sup>3</sup><sup>3</sup>3In cases such as electromagnetic or Yang-Mills gauge freedom or local Lorentz invariance of an orthonormal tetrad, the name “internal” fits well, because the transformations happen independently at each space-time point. However, some symmetries that are not diffeomorphisms resist being called internal. One example is a theory with Einstein’s equations formulated with a background metric *tensor*. Then there are two symmetries: diffeomorphisms and gauge transformations, both of which involve derivatives of the fields to arbitrarily high order \[Grishchuk et al., 1984, Pitts and Schieve, 2004\] and so are nonlocal in their finite forms. More famously, supersymmetry (which appears in supergravity and superstring theory) nontrivially combines internal and external symmetries. Both examples became known after Anderson’s work. The taxonomy of TLL \[Thorne et al., 1973\] is more capacious, but still does not comfortably accommodate Einstein’s equations with a background metric. As with covariance transformations, symmetry transformations can form a groupoid that is not a group \[Pitts and Schieve, 2004\]. I find no argument for Friedman’s restricting the relevant equivalence relation to diffeomorphisms, so perhaps he was unaware of this departure from Anderson’s work. The goal is to distinguish physical sameness from conventional variation in descriptive fluff. Because these other symmetries involve descriptive fluff as much as diffeomorphisms do, it seems that Anderson was more successful than Friedman on this point. The role of internal groups in Anderson’s work seems to have escaped Norton’s notice \[Norton, 1993, pp. 847, 848\].
Second, Friedman’s mathematical language is less general than Anderson’s and fails to accommodate some useful mathematical entities that Anderson’s older component language permits. Anderson, a working physicist, knows what sorts of mathematical structures physicists actually use and need, while Friedman restricts his attention to that narrower collection of entities that all modern coordinate-free treatments of gravitation or (pseudo-)Riemannian geometry presently discuss, namely tensors and connections, but not, for example, tensor densities (especially of arbitrary real weight), which many such treatments neglect. Considering how frequently other authors agree with Friedman’s practice of neglecting tensor densities, it is worthwhile to recall how useful they are, if not essential in some applications. In the literature on modern nonperturbative canonical quantization of gravity with Ashtekar’s new variables and the like, tensor densities are used routinely. Some authors write densities in a way that makes their weight manifest: a weight $`2`$ density has two tildes over it, a weight $`1`$ density has a tilde below it, *etc.* Moreover, the use of a densitized lapse function has proven useful in 3+1-dimensional treatments of the initial value problem<sup>4</sup><sup>4</sup>4It is now customary in numerical general relativity to call the problem of inferring later or earlier states of a system from initial data the “Cauchy problem,” while the term “initial value problem” is reserved for the procedure of solving the constraint equations to get a set of initial data. This latter sort of problem exists only for constrained theories like GTR or Maxwell’s electromagnetism. in GTR and the dynamical preservation of the constraint equations \[Jantzen, 2004, Anderson and York, 1998\]. Perhaps these uses of densities are matters of convenience rather than necessary, because one can simulate tensor densities of integral weights using tensors. However, this procedure is not so obviously available for most densities of non-integral weight; it is generally unclear, for example, what a quantity with a third of an index or $`\pi `$ indices would mean. But tensor densities of fractional weight have been used in applications such as the conformal-traceless decomposition of André Lichnerowicz and James York in solving GTR’s initial value constraints in numerical general relativity \[York, 1972, Brown, 2005\], unimodular variants of GTR (discussed in \[Unruh, 1989, Earman, 2003\]), and quantum gravity \[Peres, 1963, DeWitt, 1967, Leonovich and Mladenov, 1993\]. Densities with *irrational* weights are, if not essential, at least very useful in work on massive variants of Einstein’s GTR \[Ogievetsky and Polubarinov, 1965, Pitts and Schieve, 2005\]. Thus Friedman’s mathematical language does not accommodate these quantities that physicists use and perhaps require. Hermann Weyl protested in 1920 against early clumsy efforts at component-free formalisms “which are threatening the peace of even the technical scientist” \[Weyl, 1952, p. 54\]. Fortunately some modern authors have accommodated densities of arbitrary weight in a modern fashion<sup>5</sup><sup>5</sup>5I thank referee Robert Geroch for emphasizing this point. It is worth noting that Spivak’s coordinate-free definition of arbitrary weight densities is spread out over pp. 314, 315, 391. One should also notice that tensor densities come in more than one kind; some can be of any real weight, while others are essentially of integral weight \[Gołab, 1974, Spivak, 1979\]. \[Spivak, 1979, Lang, 1995, Calderbank and Pedersen, 1999, Fatibene et al., 1997, Fatibene and Francaviglia, 2003\]. Both the Torretti counterexample and the scalar density counterexample (discussed below) that finds an absolute object in GTR are most readily discussed using tensor densities. Were tensor densities more widely discussed by philosophers of physics, likely Torretti’s counterexample would not have been overestimated for so long, while the scalar density counterexample would not have been overlooked for so long. Anderson did not neglect tensor densities, but simply erred in applying his definition of absolute objects to GTR by failing to consider the relevance of a simple change of variables to irreducible geometric objects. Thus we have examples of a problem noted by M. Ferraris, M. Francaviglia and C. Reina:
> In recent years, owing to their greater generality, geometric objects other than tensors began to enter physical applications, because in many cases using objects more general than tensors is essential \[list of references omitted\]. In fact, in spite of the widely known and systematic use of tensorial methods in mathematical physics, restricting ones \[*sic*\] attention to tensors may often turn out to be misleading. \[Ferraris et al., 1983, p. 120\]
Friedman’s mathematical language is also inadequate to express the techniques used by V. I. Ogievetskiĭ and I. V. Polubarinov in their atypical treatment of spinors coupled to gravity using a “square root of the metric” \[Ogievetskiĭ and Polubarinov, 1965\]. This spinor formalism should be useful in preventing the timelike leg of the orthonormal tetrad, which is typically used with spinors, from counting as an unwanted absolute object.
Third, while Friedman considers variously rich and spare versions of what is intuitively one theory (Newtonian gravity) and states a methodological preference for spare theories, his treatment lacks the firm resolve of Anderson’s demand that “irrelevant” variables be eliminated. This requirement is also imposed by TLL \[Thorne et al., 1973\] and discussed by John Norton \[Norton, 1993\]. One can readily adopt the Andersonian proscription of irrelevant variables to express Friedman’s intuitions about “natural” choices of variables \[Friedman, 1983, p. 59\] in relation to the Jones-Geroch dust counterexample.
A fourth difference pertains to the notion of standard formulations of a theory. Anderson argues (somewhat confusingly) that theories should be coordinate-covariant under arbitrary manifold mappings; this move seems to be offered as a substantive claim rather than a conventional choice. More understandably, TLL stipulate that the standard form of a theory be manifestly coordinate-covariant. Friedman, by contrast, takes as standard a form in which the absolute objects, if possible, have *constant components* \[Friedman, 1983, p. 60\] and so have limited coordinate freedom. Friedman implies that one can always choose coordinates such that the absolute objects (a) have constant components and (b) thus drop out of the theory’s differential equations, which then pertain to the dynamical objects alone. However, claim (a) is falsified by the counterexample of (anti-) de Sitter space-time as a background \[Rosen, 1978, Logunov et al., 1991\] for some *specific* curvature value. These space-times of constant curvature, at least for a fixed value of the curvature, satisfy Anderson’s and Friedman’s definitions of absolute objects for the space-time metric, but the components of the metric cannot be reduced to a set of constants. An analogous example with spatial curvature is also available. Anderson makes some effort to identify the ‘correct’ or best formulation of a theory, a task taken up in more detail by TLL \[Thorne et al., 1973\]. The latter authors’ “fully reduced generally covariant representation” of a theory, unlike Friedman’s “standard formulation” (p. 60), retains the full coordinate freedom by leaving the absolute objects as world tensors (or tensor densities, connections, or whatever they are). Friedman’s expectation that absolute objects be expressible using constant components is too strong to apply in every example. Claim (b) is falsified by the example of massive versions of Einstein’s theory \[Ogievetsky and Polubarinov, 1965, Freund et al., 1969, Babak and Grishchuk, 2003, Pitts and Schieve, 2005\]. After a lull from the mid-1970s to the mid-1990s, massive variants of gravity have received considerable attention from physicists lately, especially particle physicists. In those theories such that the background space-time metric is flat, its components can be reduced to a set of constants globally by a choice of coordinates, but the background metric still does not disappear from the field equations because it appears in them algebraically, not merely differentially as Friedman apparently assumed tacitly. Especially because (a) is false, the Thorne-Lee-Lightman fully reduced generally covariant formulation is therefore preferable to Friedman’s standard formulation, which fails to exist in some interesting examples. However, if one’s goal is more historical, so that Newtonian gravity and special relativity without gravity are the main theories of interest, then Friedman’s standard formulation suffices to illustrate the role of the Galilean and Poincaré groups, respectively.
Friedman’s expectation that the components of absolute objects could be reduced to constants in general, though incorrect, usefully calls attention to the role (or lack thereof) of Killing vector fields and the like in analyzing absolute objects. If the (anti-) de Sitter space-time examples show that constancy of components is too strict a criterion, the next best thing is to have a maximal set of 10 Killing vector fields in four space-time dimensions, whether commuting as in the flat space-time case or not as in the (anti-) de Sitter case. One could generalize requirements on Killing vector fields in various ways \[Kramer et al., 1980\]. Because absolute objects need not be metric tensors, the general notion is not Killing vector fields, but generalized Killing vector fields, that is, fields such that the Lie derivative of the absolute objects vanishes. Certainly some notion of constancy is one of the core intuitions that one has about absolute objects, though it plays no role in Anderson’s definition of absolute objects, as John Earman has noticed \[Earman, 1974\]. Newton’s claim that absolute space “remains similar and immovable” is suggestive of symmetry within a model \[Earman, 1989\], not merely similarity between models. Standard examples of absolute objects usually have a fair number of generalized Killing vector fields. In Anderson’s terminology, most typical theories will have fairly large symmetry groups. Usually at least a 7-parameter family of space- and time-translations and spatial rotations will be in the symmetry group, as in classical mechanics \[Goldstein, 1980\]. In GTR (including suitable matter fields), the lack (or scarcity, as the case may be) of absolute objects implies a vast symmetry group. This large group of all diffeomorphisms (or all volume-preserving ones) as symmetries of the absolute objects, in turn, leads to an embarrassment of riches concerning local conservation laws, albeit noncovariant and not unique \[Anderson, 1967, pp. 425, 426\]. From this fact follows the so-called nonlocalizability of gravitational energy. If time translation invariance were required for absolute objects, then that criterion could exclude Norton’s counterexample involving Robertson-Walker metrics \[Norton, 1993, p. 848\]. The most typical and plausible examples of absolute objects do not apply forces that violate conservation laws; those that do, might well be called miraculous.
## 2 Confined objects and global space-time topology
While absolute objects and dynamical objects are mutually exclusive, it is useful to have the third category of “confined” objects as well \[Thorne et al., 1973\]; these three categories are mutually exclusive and exhaustive, evidently. Some entities that seemed intuitively absolute but do not satisfy Anderson’s definition fit into the category of confined objects. “The confined variables are those which do *not* constitute the basis of a faithful representation of the \[manifold mapping group\]” \[Thorne et al., 1973, p. 3568\], which means (p. 3567) that there exist two distinct elements of the manifold mapping group that produce identical mappings of the confined ‘variables’. The requirement that absolute objects form a faithful realization of the theory’s covariance group is something that TLL carry over from Anderson \[Anderson, 1967, p. 83\], though they have different definitions of faithfulness \[Thorne et al., 1973, p. 3577\]. To avoid confusion with philosophical terminology (as a referee urged), let us call these new things “confined objects” rather than “confined variables.” TLL list universal constants as examples of confined objects. Indeed it is clear that structures that do not change at all under coordinate transformations are confined objects. Some other examples of things unaffected by coordinate transformations that come to mind include the identity matrix, the Lorentz matrix $`diag(1,1,1,1),`$ fixed Dirac $`\gamma ^\mu `$ matrices, Lie group structure constants, and Oswald Veblen’s “numerical tensors” (which, in Veblen’s usage, included tensor densities). The numerical tensors are the Kronecker $`\delta _\nu ^\mu `$ symbol, which is trivially a world tensor, and the Levi-Civita totally antisymmetric $`ϵ`$ symbol with values $`1,`$ $`1,`$ and $`0`$; these values are the components of both a contravariant tensor density of weight 1 and a covariant tensor density of weight -1 \[Veblen, 1933, Anderson, 1967, Spivak, 1979\]. It has been suggested by Harvey Brown that the signature of the metric is importantly like an absolute object \[Brown, 1997, Maidens, 1998\]. If the signature were an absolute object in the strict sense, then GTR would have an absolute object, contrary to Anderson’s diagnosis of the novelty of GTR (though that diagnosis will be imperiled below on other grounds). Anderson’s and Friedman’s works have no category for expressing this immutable, externally prescribed nature of the metric signature, because absolute objects are supposed to be geometric objects (tensor fields and the like). The fact that the spacetime metric signature is unaffected by diffeomorphisms suggests that it counts as a confined object in the richer TLL taxonomy. Restricting ourselves to space-time theories as usual, another issue worthy of consideration is the global topology of spacetime, which sometimes has been neglected (but see \[Hiskes, 1984, Friedman, 1983, Earman, 1974\] \[Stachel, 2002, pp. 298, 299\]). The global topology of spacetime is certainly untouched by diffeomorphisms, so it might be treated as a confined object.
## 3 Jones-Geroch counterexample and Friedman’s reply
With a clear grasp of absolute objects in hand, one can now consider the Jones-Geroch counterexample that claims that the 4-velocity of cosmic dust counts, absurdly, as an absolute object by Friedman’s or Anderson’s standards. Friedman concedes some force to this objection made by Robert Geroch and amplified by Roger Jones, here related by Friedman:
> …\[A\]s Robert Geroch has observed, since any two timelike, nowhere-vanishing vector fields defined on a relativistic space-time are $`d`$-equivalent, it follows that any such vector field counts as an absolute object according to \[Friedman’s criterion\]; and this is surely counter-intuitive. Fortunately, however, this problem does not arise in the context of any of the space-time theories I discuss. It could arise in the general relativistic theory of “dust” if we formulate the theory in terms of a quintuple $`M,D,g,\rho ,U`$, where $`\rho `$ is the density of the “dust” and $`U`$ is its velocity field. $`U`$ is nonvanishing and thus would count as an absolute object by my definition. But here it seems more natural to formulate the theory as a quadruple $`M,D,g,\rho U`$ where $`\rho U`$ is the momentum field of the “dust.” Since $`\rho U`$ does vanish in some models, it will not be absolute. (Geroch’s observation was conveyed to me by Roger Jones, who also suggested the example of the general relativistic theory of “dust.”…) \[Friedman, 1983, p. 59\]
Here $`D`$ is the torsion-free covariant derivative compatible with $`g`$. Other sources, including what Roger Jones reported hearing from Robert Geroch, indicate a qualification to *local* diffeomorphic equivalence of nonvanishing timelike vector fields \[Jones, 1981b, pp. 167, 168\] \[Jones, 1981a\] \[Trautman, 1965, p. 84\] \[Wald, 1984, p. 18\] \[Dodson and Poston, 1991, pp. 198-200\]. In any case nothing in my argument will depend on global *versus* merely local equivalence between arbitrary neighborhoods. Jones also distinguishes the local diffeomorphic equivalence of nonvanishing timelike vector fields, which holds in general, from the (local) diffeomorphic equivalence of their covariant derivatives of various orders, which typically does not hold.
Below I will argue that Friedman’s response is nearly satisfactory, though it has two weaknesses as he expressed it. First, the statement “$`\rho U`$ does vanish in some models” ought to have said “$`\rho U`$ does vanish in some neighborhoods in some models” to show that he is considering only genuine models of GTR + dust (in which dust vanishes in some neighborhoods in some models), rather than some models with (omnipresent?) dust and some degenerate models which nominally have dust but actually have no dust anywhere. The latter would seem to be a cheat. As it stands, the reader is left to wonder whether such a cheat is doing important work for Friedman (though John Norton correctly read Friedman’s proposal as “relying …on the possibility that $`\rho `$ vanishes somewhere” \[Norton, 1993, p. 848\]). Clearly some models with dust have neighborhoods lacking dust, and it is these models which will prevent the dust $`4`$-velocity from constituting an absolute object. Second, Friedman’s unfortunate notation $`\rho U`$ suggests that the mass current density (which I will call $`J^\mu `$) is logically posterior to $`\rho `$ and an everywhere nonvanishing timelike $`U^\mu .`$ If so, then one has not eliminated the absolute object after all. If a timelike nowhere vanishing $`U^\mu `$ exists in the theory, then it is absolute even if $`\rho U^\mu `$ vanishes somewhere and so is not absolute. Thus the significance of Friedman’s use of $`\rho U^\mu `$ is left obscure. Instead one can take $`J^\mu `$ to be the fundamental variable, while the timelike $`U^\mu `$ is a derived quantity defined wherever $`\rho 0.`$ Alternatively, one can take $`U^\mu `$ to be meaningful everywhere (and perhaps primitive), but vanishing where there is no dust. If Friedman had said that $`J^\mu `$ or $`U^\mu `$ “does vanish in some neighborhoods in some models,” then these two infelicities would have been avoided. Perhaps these expository imperfections led Roberto Torretti to judge Friedman’s reply *ad hoc* \[Torretti, 1984\] and John Norton to call it “a rather contrived escape” \[Norton, 1993, p. 848\]. Once these problems are removed, the merit of Friedman’s intuition shines brightly.
Below I shall review more discussion of this counterexample in the philosophical literature. Various neglected items from the physics literature will shed light on long-standing philosophical debates about absolute objects. Using the term “variational” for objects which are varied in an action principle \[Gotay et al., 2004\], one can safely follow Anderson in making “absolute” and “dynamical” mutually exclusive, while leaving open the connection between absoluteness and nonvariationality. It will be shown that there exist theories with variational absolute objects, at least if one does not exclude Rosen’s variational principle as somehow illegal. Such a theory can be obtained using Rosen’s trick to fulfill Maidens’s claim that the absolute special relativistic metric could be obtained variationally. However these theories arguably violate Anderson’s demand to eliminate irrelevant variables. A natural extension of the proscription of irrelevant variables serves to eliminate the Jones-Geroch counterexample: the dust $`4`$-velocity $`U^\mu `$ does not count as an absolute object for GTR + dust because $`U^\mu `$ does not exist where there is no dust.
## 4 Hiskes’s redefinition of absoluteness, Maidens’s worry, and Rosen’s answer in advance
Anne Hiskes proposed amending the definition of absolute objects so that no field varied in a theory’s action principle would be regarded as absolute \[Hiskes, 1984\]. Such a move makes use of what *prima facie* seems to be a true generalization about absolute and dynamical objects. This intuition was shared by the master. Anderson wrote:
> In addition to the differences between absolute and dynamical objects discussed in Section 4-3 there is another important difference that appears to be characteristic of these two types of objects. The equations of motion for the dynamical objects can often be derived from a variational principle, especially if these objects are fields. On the other hand, it appears to be the case, although we can give no proof of the assertion, that the equations of motion for the absolute objects do not have this property.…In the following discussion we will assume that the equations of motion for the dynamical objects of a theory follow from a variational principle and that those for the absolute elements do not. \[Anderson, 1967, pp. 88, 89\]
Thus Anderson suspected that most or all dynamical objects are variational, while no absolute object is variational. Similar intuitions are manifest in the TLL and LLN papers \[Thorne et al., 1973, Lee et al., 1974\]. Such a requirement also appears in their notion of being “Lagrangian-based” \[Thorne et al., 1973, p. 3573\]. Recently John Earman has found it convenient to use “absolute” to *mean* non-variational \[Earman, 2003\]. Anderson was quite sensitive to the possibility of reformulating what intuitively seems like the same theory using various different sets, and indeed increasingly large sets, of variables in an action principle \[Anderson, 1967, section 4.2\]. Unlike Hiskes, he strove to define a unique correct formulation that gave the expected answers.
More recently, Anna Maidens has entertained the idea that Hiskes’s redefinition could be deployed to remove the Jones-Geroch counterexample \[Maidens, 1998\]. If absolute objects must be nonvariational, while the dust $`4`$-velocity is variational, then the dust $`4`$-velocity is not absolute. Following Hawking and Ellis \[Hawking and Ellis, 1973\], Maidens indicates how the equations for the timelike vector field can be derived from a variational principle.<sup>6</sup><sup>6</sup>6One notices that Hawking and Ellis use a fluid variational principle with constrained variations, not the more familiar unconstrained variations. In some respects this is a disadvantage, though Schutz and Sorkin observe that it keeps one closer to the physical variables \[Schutz and Sorkin, 1977\]. They also observe that in many cases, including this one, one can eliminate the constraints on the variation (not to be confused with constraints in the sense of gauge theories \[Sundermeyer, 1982\]) using Lagrange multipliers. It seems to me that John Ray’s variational principle might be preferable in the present context, because it involves varying $`U^\mu `$ itself and uses unconstrained variations \[Ray, 1972\]. However Maidens is also sensitive to the large variety of choices of variables and even the number of field components in an action principle for what intuitively counts as a single theory. Thus she expected such a use of Hiskes’s redefinition to fail, because it eliminates the Jones-Geroch counterexample at the cost of introducing a new one. More specifically, Maidens has suggested that there might be some way to reformulate special relativistic theories such that the flat metric, which surely ought to count as absolute, is varied in the action principle. If that could be done, then Hiskes’s definition of absolute objects would prove to be too strict (the opposite problem from what the Jones-Geroch example suggests about Friedman’s), because it fails to count the metric tensor of special relativity as an absolute object. (Maidens presumably should envision a weakly generally covariant formulation of special relativity, though her notation is far from clear on that point.) “At this stage, however, we find a fly in the ointment, for its turns out that given suitable starting assumptions we can derive the Lorentz metric from an action principle.” \[Maidens, 1998, p. 262\] Supporting such claims would involve actually displaying a suitable Lagrangian density whose Euler-Lagrange equations give the desired results or else citing a source where such work had been done. Surprisingly, she fails to do either one. Success would involve finding an action principle for which the flatness of the metric holds for *all* models (her case (c)), not just some (her case (a), p. 265). A bit later she finds that “it is an open question as to whether the metric of special relativity is derivable from an action principle.” (p. 266) Two pages later she once again claims that “some of the physically necessary fixed background, e.g. the Lorentz metric, can also be derived from an action principle.” (p. 268) It is not easy to harmonize these fluctuating statements.
Fortunately Maidens’s expectation that the flatness of a metric (for all models) can be derived from a variational principle is in fact correct. The question was resolved by Nathan Rosen in the 1960s \[Rosen, 1966, Rosen, 1973\]. He used an action principle involving a Lagrange multiplier field with 20 components, a trick recently used also by Rafael Sorkin \[Sorkin, 2002\]. Thus requiring absolute objects to be nonvariational gives an excessively strict definition, so the Jones-Geroch counterexample is not adequately addressed thereby. Some objects that should count as absolute can be variational, as Maidens expected. Rosen includes the following term in an action principle (after a change in notation to $`\eta _{\mu \nu }`$ for the metric in question, which is *a priori* arbitrary apart from the signature) to force $`\eta _{\mu \nu }`$ to be flat:
$$S=d^4x\sqrt{\eta }R_{\rho \mu \nu \sigma }[\eta ]P^{\rho \mu \nu \sigma }.$$
(1)
This term is intended as a supplement to the action for a special relativistic theory, within which now $`\eta _{\mu \nu }`$ would be subject to variation as well. $`P^{\rho \mu \nu \sigma },`$ a tensor with the same symmetries as the Riemann tensor for $`\eta _{\mu \nu },`$ serves as a Lagrange multiplier. Varying $`P^{\rho \mu \nu \sigma }`$ immediately yields the flatness of $`\eta _{\mu \nu }.`$ Varying $`\eta _{\mu \nu }`$ takes more work and gives an equation of motion especially involving the second derivatives of $`P^{\rho \mu \nu \sigma }`$. That equation is not needed here. Rosen seems to make secret use of the Euler-Lagrange equations from $`P^{\rho \mu \nu \sigma }`$ to discard terms involving $`R_{\rho \mu \nu \sigma }[\eta ]`$ in his equations 10, 11, and 12; if so, then his equation 12 is not an “identity” as he claims. Alternately, he might be taking the metric to be flat before the variation but curved after it, as Sorkin proposes \[Sorkin, 2002\], if that is an intelligible alternative.<sup>7</sup><sup>7</sup>7 It is perhaps worth noting that varying $`P^{\rho \mu \nu \sigma }`$ gives an equation of motion for $`\eta _{\mu \nu }`$ and varying $`\eta _{\mu \nu }`$ gives an equation of motion primarily for $`P^{\rho \mu \nu \sigma }.`$ Thus one should avoid expressions like “the equations of motion for $`\eta _{\mu \nu }`$” or “the equations of motion for $`P^{\rho \mu \nu \sigma }`$” due to their ambiguity. As was noted above, Anderson’s requiring component equality (up to equivalence under the covariance group) only for *dynamically* possible trajectories is relevant here. Using Rosen’s trick, one has a geometric object such that its components agree for dynamically possible trajectories (“on-shell,” as physicists say) but not for kinematically possible trajectories (“off-shell”), because the metric is not flat for all kinematically possible trajectories.
Anderson briefly states that one must remove irrelevant variables from the theory under analysis. He writes:
> It is possible that a subset of the components of the \[geometrical object characterizing the kinematically possible trajectories of the theory\] do not appear in the equations of motion for the remaining components and furthermore can be eliminated from the theory without altering the structure of its equivalence classes. Such a subset is obviously irrelevant to the theory. We shall assume, therefore, that no subset of the components of \[that geometrical object\] is irrelevant in this sense.” \[Anderson, 1967, p. 83\]
Likewise TLL exclude the category of irrelevant variables \[Thorne et al., 1973, p. 3569\]. Anderson observes that
> one can always construct a hierarchy of theories all of which have the same equivalence-class structure in the sense that the equivalence classes of these theories can be put into one-to-one correspondence with each other. Two theories of such a hierarchy will differ both with regard to the mathematical quantities that describe their respective \[kinematically possible trajectories\] and their respective covariance groups. However, the set of mathematical quantities that describe the \[kinematically possible trajectories\] of a given theory in such a hierarchy will contain, as subsets, those of each theory that precedes it in the hierarchy. Likewise, its covariance group will contain, as a subgroup, the covariance group of each preceding theory.…
>
> The question then arises as to which theory of a hierarchy one should use to describe a given physical system. The answer rests, of course, in the final analysis, on the measurements that one can make on the system. It is necessary that each quantity used to describe the \[kinematically possible trajectories\] of a theory must, at least in principle, be measurable. \[Anderson, 1967, p. 81\]
Similar thoughts appear elsewhere in the text (pp. 306, 340). This requirement of observability, an unfortunate whiff of verificationism, presupposes that all the physics resides in the field equations.<sup>8</sup><sup>8</sup>8This last claim Anderson elsewhere implicitly appears to contradict when he considers boundary conditions (p. 75) and suggests (using “furthermore” on p. 83), surprisingly, that there could exist fields that do not appear in other fields’ equations of motion, but which help to determine the structure of the theory’s equivalence classes. As it happens, recent work on field formulations of Einstein’s equations provides an example: the flat metric does not appear essentially in the field equations, but it plays a role in the boundary conditions, topology, and the notion of gauge transformations \[Pitts and Schieve, 2004\]. Boundary conditions are important in string theory as well \[Braga et al., 2005\]. Thus Anderson is overly hasty in eliminating the background metric after deriving Einstein’s equations in flat space-time \[Anderson, 1967, pp. 303-306\] in the fashion of Kraichnan \[Kraichnan, 1955\]. While Kraichnan’s use of a background metric in no way requires that quantization occur by covariant perturbation theory \[Solov’ev, 1988\], historically the two projects have been linked in the minds of many. Anderson critiqued perturbative approaches to Einstein’s equations in response to a paper by Richard Arnowitt \[Arnowitt, 1963\]. But typically, fields that do useful work are observable, and Anderson’s requirement of observability, if not entirely on target, at least emphasizes the importance of excluding idle fields, such as $`P^{\rho \mu \nu \sigma }`$ appears to be.
While Rosen’s trick vindicates Maidens’s assertion that building nonvariationality into the notion of absolute objects is unsuccessful, Andersonian resources might be invoked to exclude Rosen’s trick as a form of cheating. Anderson’s prohibition of irrelevant variables appears to exclude theories making use of Rosen’s trick, because the dynamical evolution of the Lagrange multiplier $`P^{\rho \mu \nu \sigma }`$ has no effect on any other fields, whether gravitational or matter. $`P^{\rho \mu \nu \sigma }`$ appears to do nothing useful by Anderson’s standards. Making $`\eta _{\mu \nu }`$ variational and yet absolute could perhaps be useful in that it lets one treat the theory readily using the existing constrained dynamics formalism (*e.g.*, \[Sundermeyer, 1982\]), which has not made much room for nonvariational fields. Making $`\eta _{\mu \nu }`$ variational also allows one to define a conserved symmetric stress energy tensor without using the formal trick of the Rosenfeld approach, in which one replaces the flat metric by a curved one for taking a functional derivative and then restores flatness afterwards \[Deser, 1970\]. Whether Rosen’s trick or Rosenfeld’s is preferable is open to discussion, but an Andersonian elimination of the Lagrange multiplier field as irrelevant would be at least a defensible view.
Where does this dialectic leave us? Maidens proposed and rejected using Hiskes’s redefinition of absolute objects to exclude the Jones-Geroch counterexample to Friedman’s account of absolute objects. Maidens’s missing proof was supplied in advance by Rosen. But Rosen’s trick seems not to count against Anderson’s version of the intuition that absolute objects are nonvariational, because Anderson wisely has criteria for eliminating irrelevant variables. Does it follow that Anderson’s intuition, in the larger context of his project that excludes irrelevant variables, is vindicated? That is, if we accept Anderson’s definitions and proscriptions, should we also accept his intuition that fields are variational if and *only if* they are dynamical? As it turns out, Anderson’s generalization survives this alleged counterexample but might be threatened by another in which all fields are variational but there is still an absolute object. I have in mind parametrized theories \[Sundermeyer, 1982, Kuchař, 1973, Schmelzer, 2000, Arkani-Hamed et al., 2003, Norton, 2003, Earman, 2003\], in which preferred coordinates are rendered variational. One often calls the results “clock fields.” Perhaps some uses of clock fields could be excluded as irrelevant—not because the fields themselves are irrelevant, but because perhaps their variationality is. On the other hand, if clock fields are used to satisfy an appropriate notion of causality in bimetric theories like massive variants of Einstein’s equations \[Pitts and Schieve, 2005, Pitts and Schieve, 2004, Schmelzer, 2000\], then their variationality is relevant. Parametrized theories require more discussion than is appropriate here, however. The scalar density example below is, at present, another example of a variational yet absolute object.
## 5 Eliminating local irrelevance excludes the Geroch-Jones vector field
If Maidens’s proposed and rejected use of Hiskes’s redefinition is set aside for violation of Anderson’s prohibition of irrelevant variables, then the Jones-Geroch counterexample still remains to be addressed. Now it turns out that Anderson’s and TLL’s proscription of irrelevant variables, if it does not quite remove the Jones-Geroch counterexample, at least inspires a gentle amendment that does the job. This amendment seems especially appropriate after one notices that TLL replace \[Thorne et al., 1973, p. 3566\] Anderson’s notion of geometrical object \[Anderson, 1967, pp. 14-16\] with Andrzej Trautman’s notion of a geometric object \[Trautman, 1965\]. Presumably both notions aim to capture the same intuition.
Given the relative inaccessibility of Trautman’s lectures, it will be worthwhile to quote his definition of geometric objects in detail:
> Let $`X`$ be an $`n`$-dimensional differentiable manifold.…\[S\]ince tensors are not sufficient for all purposes in geometry and physics, \[*sic*\] for example scalar densities are not tensors, to avoid having to expand definitions and theorems whenever we need a new type of entity, it is convenient to define a more general entity, the geometric object, which includes nearly all the entities needed in geometry and physics, so that definitions and theorems can be given in terms of geometric objects so as to hold for all the more specialized cases that we may require.
>
> Let $`pX`$ be an arbitrary point of $`X`$ and let $`\{x^a\},\{x^a^{}\}`$ be two systems of local coordinates around $`p.`$ A geometric object field $`y`$ is a correspondence
>
> $$y:(p,\{x^a\})(y_1,y_2,\mathrm{}y_N)R^N$$
>
> which associates with every point $`pX`$ and every system of local coordinates $`\{x^a\}`$ around $`p`$, a set of $`N`$ real numbers, together with a rule which determines $`(y_1^{},\mathrm{}y_N^{})`$, given by
>
> $$y:(p,\{x^a^{}\})(y_1^{},\mathrm{}y_N^{})R^N$$
>
> in terms of the $`(y_1,y_2,\mathrm{}y_N)`$ and the values of \[*sic*\] $`p`$ of the functions and their partial derivatives which relate the coordinate systems $`\{x^a\}`$ and $`\{x^a^{}\}.`$…The $`N`$ numbers $`(y_1,\mathrm{}y_N)`$ are called the components of $`y`$ at $`p`$ with respect to the coordinates $`\{x^a\}`$. \[Trautman, 1965, pp. 84, 85\]
Trautman then notes that spinors are not geometric objects. He also notes that some objects that are not themselves geometric objects are nonetheless *parts* of geometric objects. *Pace* Friedman’s nonstandard usage \[Friedman, 1983, p. 359\], the class of geometric objects is not exhausted by tensors and connections. Trautman’s definition was fairly typical in its time, though a bit streamlined for physicists’ use. Geometric objects were considered with great thoroughness by Albert Nijenhuis \[Nijenhuis, 1952\]. A more recent treatment of them using modern differential geometry has been given by Ferraris, Francaviglia, and Reina \[Ferraris et al., 1983\].
The reader will notice that Trautman’s geometric objects are defined at every point in the space-time manifold. That fact is of special relevance for the dust example, because it implies that if a dust $`4`$-velocity timelike unit vector field $`U^\mu `$ is used as a variable in the theory, then a dust $`4`$-velocity timelike unit vector must be defined at every point in every model, *even if no dust exists in some neighborhoods in some models*. Here one recalls Anderson’s and TLL’s call for the elimination of irrelevant variables; Friedman also recognizes the value of eliminating surplus structure. It is not clear that existing notions of irrelevance apply strictly to the present case. The dust $`4`$-velocity is locally irrelevant, not globally irrelevant, one might say. Perhaps the authors had in mind fields that satisfy equations somewhat like the Klein-Gordon equation as their primary examples, as theoretical physicists often do; for such fields irrelevance is likely to be global. But now that the question is raised, it does seem clear that wherever there is no dust, there ought not to be a dust $`4`$-velocity timelike unit vector either—at least not if the task at hand is testing theories for absolute objects.
There seem to be three initially plausible alternatives concerning the dust $`4`$-velocity where the dust has holes in some model. First, one might retain a timelike $`4`$-velocity vector even in holes in the dust, while expecting the $`4`$-velocity values in the dust holes to be mere gauge fluff. It is noteworthy that at least some perfect fluid variational principles in the physics literature yield timelike unit vector $`4`$-velocities even where there is no fluid \[Ray, 1972\]. Perhaps mathematical convenience commends this option, though I find that Ray’s variational principle can be modified to lack a timelike $`4`$-velocity in holes in the fluid. Presumably one could show that the value of a timelike $`4`$-velocity vector is in fact gauge fluff in dust holes by using the Dirac-Bergmann constrained dynamics technology \[Sundermeyer, 1982\], though one might run into technical challenges with changes of rank or with the noncanonical Poisson brackets that can appear in fluid mechanics \[Morrison, 1998\]. In any case, the timelike dust $`4`$-velocity in dust holes has no physical meaning, yet leads one to conclude that the theory has an absolute object. Clearly any absolute object whose existence is inferred only by using physically meaningless quantities is spurious. If one allowed physically meaningless entities into a theory while testing for absolute objects, then one could take any theory and construct an empirically equivalent theory with as many absolute objects as one wants. One could concoct a version of GTR with Newton’s absolute space, for example. To permit such a procedure is just to give up Anderson’s program of analyzing the uniqueness of GTR, because analysis involves *trying* to get the intuitively known right answer as a consequence of some criteria. Anderson and TLL call for the elimination of irrelevant variables in order to address just this sort of problem. One might call the entities that they reject “globally irrelevant variables” because such entities play no role at any space-time point in any model. The Jones-Geroch example shows, I conclude, that one must also exclude “locally irrelevant variables,” entities that play no role in some neighborhoods in some models. One could consider whether mathematical entities that play no role at some space-time points or sets of measure zero should also be excluded as locally irrelevant, but there might be technical reasons for admitting them.
The two remaining options avoid this spurious absolute object in different ways. One option is to take the mass current density $`J^\mu `$ to be the primitive variable and regard $`U^\mu `$ and the dust density $`\rho `$ as derived. Then $`\rho `$ is defined by $`\rho =\sqrt{J^\mu g_{\mu \nu }J^\nu }.`$ The $`4`$-velocity $`U^\mu `$ is naturally defined by
$$U^\mu =\frac{J^\mu }{\sqrt{J^\nu g_{\nu \alpha }J^\alpha }},$$
so $`U^\mu `$ is only meaningful where the denominator $`\rho `$ is nonzero. That consequence is plausible on physical grounds and blocks the Jones-Geroch counterexample. The theory is thus formulated using a quadruple $`M,D,g,J,`$ not Friedman’s quadruple $`M,D,g,\rho U`$ or the quintuple $`M,D,g,\rho ,U.`$ In some models $`J^\mu `$ vanishes at some space-time points in some models of GTR + dust, so $`U^\mu `$ is undefined in such cases. Neither $`J^\mu `$ nor $`U^\mu `$ is a Gerochian nowhere vanishing timelike vector field for all models. By contrast, the mass current density $`J^\mu `$, which is equal to $`\rho U^\mu `$ where $`\rho 0,`$ automatically vanishes where there is no dust and is continuous at the transition from dust to vacuum. Thus Friedman’s suggestion that it is more “natural” to use the mass current density, once freed from the two infelicities noted at the beginning, is seen to be very reasonable.
The other option is to take $`U^\mu `$ to be meaningful but vanishing in those places in certain models where the dust has holes.<sup>9</sup><sup>9</sup>9 One need not commit oneself to $`J^\mu `$ as primitive and $`U^\mu `$ as derived. I am indebted to Don Howard for insightful probing about choices of primitive variables. If $`U^\mu `$ is allowed to vanish in some places, then it is not rightly everywhere called the dust $`4`$-velocity, as a referee notes. Although $`U^\mu `$ exists everywhere, it vanishes in some places in some models, so not every neighborhood in every model has $`U^\mu `$ that is gauge-equivalent to $`(1,0,0,0).`$ Anderson’s definition of absolute object requires that, for any component $`\varphi _\alpha `$ of an absolute object in a theory, “\[a\]ny $`\varphi _\alpha `$ that satisfies the equations of motion of the theory appears, together with all its transforms under the covariance group, in every equivalence class of \[dynamically possible trajectories\].” \[Anderson, 1967, p. 83\] Even if we drop Anderson’s requirement of global equivalence in favor of Hiskes’s (and Friedman’s \[Friedman, 1983, pp. 58-60\]) local equivalence, $`U^\mu `$ does not count as absolute. In dust-filled regions in a model, the dust $`4`$-velocity $`U^\mu `$ is diffeomorphic (at least in a neighborhood) to $`(1,0,0,0),`$ but in dust holes $`U^\mu `$ is diffeomorphic to $`(0,0,0,0)`$ instead. Thus $`U^\mu `$, like $`J^\mu `$, is not an absolute object. One might tolerate as harmless the surplus structure embodied in the vanishing $`U^\mu `$ vectors, though the mathematical discontinuity of the vector field makes it difficult to defend this option on grounds of mathematical convenience.
If one chooses to restrict one’s attention to models of GTR + dust that do have dust everywhere and always, such gerrymandering is simply changing the subject to consider a different theory. If one takes a semantic view of theories, then restricting attention to such a set of models is just to discuss some new theory besides GTR + dust, namely GTR + omnipresent dust. Manifestly GTR + omnipresent dust is a proper subset of GTR + dust. GTR + omnipresent dust has the peculiar feature of describing “dust” with such attributes as necessary existence, omnipresence and eternality, attributes more suited to a Deity than to dust. Moreover, GTR + omnipresent dust is not the set of cosmological models of GTR. For example, one can write down cosmological models in which dust is present but not omnipresent \[Feynman et al., 1995, p. 166\] \[Klein, 1971, Smoller and Temple, 2003\]. More realistic models include eras of radiation domination and perhaps dark energy, so dust is not even a good description of matter in every region of space-time in cosmological models in GTR. In short, GTR + omnipresent dust has no essential physical relevance to cosmology. Having suitably deflated expectations regarding the theory’s physical import, one can proceed to test it for absolute objects. The new theory GTR + omnipresent dust has an absolute object. But why shouldn’t it? Surely no one has well founded intuitions to the contrary. Any matter with the attributes of necessary existence, omnipresence and eternality just isn’t much like dust, but rather has the vaguely theological flavor that both friends and foes of absolute objects (such as Newton and Einstein in his Machian aspect, respectively–if the reader will pardon the anachronism) have sensed. Anderson anticipated the fact that one could consider a proper subset of models for which some field would count as absolute without counting as absolute for the full set of models. He wrote:
> We should perhaps emphasize that we are discussing here universal absolute objects, which must appear in the description of every \[dynamically possible trajectory\] of our space-time description. It is quite possible that, for a subclass of \[dynamically possible trajectories\], one or more dynamical objects satisfy the criteria of Section 4-3 and so play the role of absolute objects for those \[dynamically possible trajectories\].…The existence of such special subclasses of \[dynamically possible trajectories\] as those discussed above does not, of course, constitute a violation of the principle of general invariance as we have formulated it. Only the existence of universal absolute objects would do so. \[Anderson, 1967, pp. 339, 340\]
Thus Anderson reminds us that absolute objects are universal, not (so to speak) provincial like the dust $`4`$-velocity. While the dust $`4`$-velocity constitutes an absolute object for the theory GTR + omnipresent dust, it does not constitute an absolute object for GTR + dust due to the failure of universality. Thus Friedman’s intuition, as modified above, is vindicated. The alleged Jones-Geroch counterexample fails to count as an absolute object for GTR + dust and thus fails to undermine Friedman’s analysis after a slight amendment using Andersonian resources.
One might summarize Friedman’s reply, as amended above, as follows: Geroch’s merely mathematical vector field is irrelevant and eliminable because it does no physical work, while Jones’s dust application of the vector field does physical work but violates the condition of being meaningful and everywhere nonvanishing in all models and so violates the diffeomorphic equivalence needed for absoluteness. At this stage a summary might be useful. Physics literature previously unappreciated by philosophers of physics has been shown to shed light on the Jones-Geroch counterexample to Friedman’s (and likely Anderson’s or TLL’s) definition of absolute objects. An old result from Rosen vindicates Maidens’s claim that Hiskes’s redefinition of absolute objects could not be used to eliminate the Jones-Geroch counterexample without generating a new counterexample. The neglected but valuable paper by TLL and some infrequently attended parts of Anderson’s book proscribe irrelevant variables, a fact with important consequences. This proscription perhaps can be used to exclude Rosen’s trick for deriving flat space-time from a variational principle. Then Anderson’s generalization that absolute objects are variational and *vice versa* would seem to be rehabilitated, at least provisionally, though the clock fields of parametrized theories pose further questions(as does the scalar density example below). If variationality cannot be invoked to remove the Jones-Geroch counterexample, then some new move is required. Again the Anderson-TLL proscription of irrelevant variables is helpful, in spirit if not in letter. Excluding locally irrelevant values of the field $`U^\mu `$, which purports to be the $`4`$-velocity field of dust, would imply that $`U^\mu `$ is undefined wherever the dust vanishes, while the mass current $`J^\mu `$ vanishes there. Alternatively, $`U^\mu `$ and $`J^\mu `$ both vanish there. Either way, GTR + dust fails to have an everywhere nonvanishing timelike vector field that exists in all models. Thus a slight amendment of the Anderson-Friedman tradition using the Andersonian opposition to irrelevant variables eliminates the Jones-Geroch counterexample.
## 6 Torretti’s example of constant curvature spaces has Andersonian absolute object
A second long-standing worry concerning the Anderson-Friedman absolute objects project was suggested by Roberto Torretti \[Torretti, 1984\]. He considered a theory of modified Newtonian kinematics in which each model’s space has constant curvature, but different models have different values of that curvature. Because every model’s space has constant curvature, such a theory surely has something rather like an absolute object in it, Torretti’s intuition suggests. Though contrived, this example is relevantly like the cases of de Sitter or anti-de Sitter background metrics of constant curvature that are sometimes discussed in the physics literature (*e.g.*, \[Rosen, 1978, Logunov et al., 1991\]), where one often lumps together space-times with different values of constant curvature. The failure of the metrics to be locally diffeomorphically equivalent for distinct curvature values entails that the metric tensor does not satisfy Anderson’s or Friedman’s definition of an absolute object (or TLL’s, for that matter). Thus Torretti concludes that Anderson’s project is not adequate for achieving the goals that Friedman has or ought to have.
How seriously one takes Torretti’s objection will depend in part upon the degree that one shares Torretti’s expectations for absolute objects. Though Anderson evidently invented the term and defined it, Torretti expects a much broader array of applications that does Anderson. The justice of this expectation depends on what sorts of claims Friedman made on behalf of the Anderson-Friedman project, as well as how seriously one takes Anderson’s non-technical glosses about acting without being acted upon and the like. A homely example will help. A lawn mower is a modest but nontrivial tool for caring for the grass in one’s yard. One can imagine a more impressive machine that also trims around obstacles and pulls weeds, though no such machine exists presently. Anderson’s project, like a lawn mower, is a tool that largely succeeds in satisfying a modest but nontrivial goal. Torretti is more ambitious in his goal, but his tool, like a lawn mower that also trims around edges and pulls weeds, does not presently exist. In the absence of the more impressive tool, one might be content with the more modest tool that is presently available. It also seems peculiar that in Torretti’s example, the value of the curvature of space is contingent (varying across models), but necessarily (in every model) the value at one moment is the same as that at another moment. Perhaps the failure of an Anderson-Friedman definition of absolute objects to count the metric as absolute in Torretti’s example shows a quirk in the example rather than the definition.
Though neither Torretti nor later writers seem to have noticed, Anderson’s analysis, when applied to Torretti’s example, does yield a very specific and reasonable conclusion involving an absolute object. Though the spatial metric is not absolute, the conformal spatial metric density, a symmetric $`(0,2)`$ tensor density of weight $`\frac{2}{3}`$ (or its $`(2,0)`$ weight $`\frac{2}{3}`$ inverse) is an absolute object. This entity, when its components are expressed as a matrix, has unit determinant. It appears routinely in the conformal-traceless decomposition used in finding initial data in numerical studies of GTR. It defines angles and relative lengths of vectors at a point, but permits no comparison of lengths of vectors at different points. In three dimensions, conformal flatness of a metric is expressed by the vanishing of the Cotton tensor \[Aldersley, 1979, Garcia et al., 2004\], not the Weyl tensor, which vanishes identically. That the conformal metric density is an absolute object is shown in the following way. Every space with constant curvature is conformally flat \[Wolf, 1967, Robertson and Noonan, 1968, Misner et al., 1973\]. For conformally flat spatial metrics, manifestly the conformal parts are equal in a neighborhood up to diffeomorphisms. The conformal part just is the conformal metric density, so the conformal metric density is the same (within a diffeomorphism) locally for every model in Torretti’s theory. One could have the intuition that Anderson’s analysis captures as absolute everything that it ought to capture. I conclude that the force of this counterexample has been overestimated. Concerning Norton’s modification of Torretti’s example to Robertson-Walker metrics \[Norton, 1993, p. 848\], analogous comments could be made: these space-*times* are conformally flat \[Infeld and Schild, 1945, Tauber, 1967, Padmanabhan, 1993, Kuchowicz, 1973\] and so have as an absolute object the space-time conformal metric density.
## 7 Tetrad-spinor: Avoiding absolute object by eliminating irrelevant fields
One potential counterexample to the Anderson-Friedman example that seems not to have been noticed arises from the use of an orthonormal tetrad formalism, in which the metric tensor (or its inverse) is built out of four orthonormal vector fields $`e_A^\mu `$ by the formula $`g^{\mu \nu }=e_A^\mu \eta ^{AB}e_B^\nu `$ or the like. Four vector fields have among them 16 components, rather more than the 10 components of the metric, so there is some redundancy that leaves a new local Lorentz gauge freedom to make arbitrary position-dependent boosts and rotations of the tetrad. It is unnecessary to use a tetrad instead of a metric as the fundamental field when gravity (as described by GTR) is coupled to bosonic matter (represented by tensors, tensor densities or perhaps connections). However, it is widely believed to be necessary to use an orthonormal tetrad to couple gravity to the spinor fields that represent electrons, protons, and the like \[Weinberg, 1972, Deser and Isham, 1976, Fatibene and Francaviglia, 2003\]. The threat of a counterintuitive absolute object then arises. Given both local Lorentz and coordinate freedom, one can certainly bring the timelike leg into the component form $`(1,0,0,0)`$ at least in a neighborhood about any point. (Aligning the tetrad with the simultaneity hypersurfaces is known as imposing the time gauge on the tetrad \[Deser and Isham, 1976\].) Unlike the dust case, there cannot be any spacetime region in any model such that the timelike leg of the tetrad vanishes. Thus GTR coupled to a spinor field using an orthonormal tetrad gives an example of a Gerochian vector field: nowhere vanishing, everywhere timelike, gauge-equivalent to $`(1,0,0,0)`$, and (allegedly) required to couple the spinor and gravity and thus not irrelevant. Like clock fields, the timelike tetrad leg also appears to be both variational and absolute. If it is true that coupling spinors to gravity requires an orthonormal tetrad and that an orthonormal formalism for GTR yields an absolute object, then the intuitively absurd conclusion that GTR + spinors has an absolute object follows.
Before discussing the tetrad-spinor issue, it is worthwhile to consider Anderson’s treatment of spinors of the Dirac equation in a gravitational field (pp. 358-360). Anderson entertains the worry that $`\gamma ^\mu `$ might be an absolute object in flat spacetime, in fact one with a symmetry group smaller than the Poincaré group (though in this context $`\gamma ^\mu `$ is not a vector under *arbitrary* coordinate transformations, so it is not eligible to be an absolute object by Anderson’s standards, it would seem). Turning to curved spacetime, Anderson avoids using an orthonormal tetrad by using variable Dirac matrices $`\gamma ^\mu `$ satisfying $`\gamma ^\mu \gamma ^\nu +\gamma ^\nu \gamma ^\mu =2g^{\mu \nu }I.`$ What follows is a formalism with an internal symmetry group (apparently global) unrelated to the group of spacetime mappings. However, the implicit relationship between $`\gamma ^\mu `$ and $`g^{\mu \nu }`$ leaves obscure what a suitable action principle might be for deriving the Einstein-Dirac equations and what variables it would involve. Thus one can hardly even test Anderson’s formalism for absolute objects; his treatment of spinors is just incomplete. By contrast the tetrad-spinor formalism avoids such difficulties.
The tetrad-spinor example seems rather more serious a problem for definitions of absolute objects than the Jones-Geroch cosmological dust example was, because the spinor field is surely closer to being a fundamental field than is dust or any other perfect fluid. Spinors (actually vector-spinors for spin $`\frac{3}{2}`$) are also required in supergravity, where internal and external symmetries are combined, not to mention (super)string theory. On another occasion I expect to explain in more detail how to remove irrelevant variables here and thus avoid this unexpected absolute object. This removal is achieved using the alternative spinor formalism of V. I. Ogievetskiĭ and I. V. Polubarinov \[Ogievetskiĭ and Polubarinov, 1965\] to eliminate “enough” of the orthonormal tetrad as irrelevant that the timelike nowhere vanishing vector field disappears from the theory. A brief summary suffices here. Their formalism’s “square root of the metric” resembles an orthonormal tetrad gauge-fixed to form a symmetric matrix by sacrificing the local Lorentz freedom while preserving diffeomorphism freedom. The square root of the metric has only ten components rather than sixteen and can be computed using a binomial series expansion.
## 8 Scalar density example and unimodular GTR: Does GTR lack absolute objects?
Unimodular GTR was invented by Einstein, was discussed by Anderson along with David Finkelstein \[Anderson and Finkelstein, 1971\], and is rather well known today \[Earman, 2003\]. Still it turns out that consideration of unimodular GTR helps one to reach the startling conclusion that not only it, but GTR itself, has an absolute object on Friedman’s definition. (While serving as a referee, Robert Geroch proposed this counterexample, though using different mathematical variables.) Unimodular GTR comes in two flavors: the coordinate-restricted version in which only coordinates that fix the determinant of the metric components matrix to $`1,`$ and the weakly generally covariant version that admits any coordinates with the help of a nonvariational scalar density (usually of weight 1 or 2, but any nonzero weight suffices) and a dynamical conformal metric density, which is a $`(0,2)`$ tensor density of weight $`\frac{2}{n}`$ or a $`(2,0)`$ tensor density of weight $`\frac{2}{n}`$ in $`n`$ space-time dimensions. As Anderson and Finkelstein observe, a metric tensor as a geometric object is reducible into a conformal metric density and a scalar density. They have in mind an equation along these lines:
$$g_{\mu \nu }=\widehat{g}_{\mu \nu }\sqrt{g}^{\frac{2}{n}}$$
(2)
As usual, $`g`$ is the determinant of the matrix of components $`g_{\mu \nu }`$ of the metric tensor in a coordinate basis; $`g`$ is a scalar density of weight 2 and takes negative values because of the signature of the metric tensor. $`\widehat{g}_{\mu \nu }`$ is the conformal metric density. The new variables $`\widehat{g}_{\mu \nu }`$ (or its inverse) and $`\sqrt{g}`$ (or any nonzero power thereof) are those of Anderson and Finkelstein or are relevantly similar. They further observe that this scalar density is an absolute object in unimodular GTR. This observation seems unremarkable because that scalar density is not variational. For comparison, one recalls that Asher Peres rewrote the Lagrangian density for GTR in terms of the conformal metric density and a scalar density \[Peres, 1963\]; recently this idea was reinvented by M. O. Katanaev \[Katanaev, 2005\]. Surely the result is still GTR and not some other theory. To my knowledge, no one (prior to Geroch, in effect) has ever considered whether the scalar density, even if varied in an action principle for GTR, might still count as an absolute object. Once the question is raised about GTR with the Peres-type variables, a positive answer seems obvious: GTR has an absolute object! This absolute object is a scalar density of nonzero weight, because every neighborhood in every model space-time admits coordinates (at least locally) in which the component of the scalar density has a value of $`1.`$
Interesting conclusions follow. First, either Anderson’s claim that GTR’s novelty lay in its lack of absolute objects, or his analysis of absolute objects, is flawed. Second, the scalar density is absolute despite being variational, somewhat as clock fields might be. Perhaps some people assume that any field varied in an action principle is dynamical (that is, not absolute), even while officially employing Anderson’s definition of absoluteness. Third, it would be useful to combine hints from Anderson and Finkelstein about the (ir)reducibility of geometric objects with the notion of equivalent geometric objects \[Nijenhuis, 1952\] to accommodate changes of basic variables or, as a field theorist might say, field redefinitions. Finally, though some philosophers of physics profess to know absolute objects when they see them, even without an analysis, the case of GTR formulated using a conformal metric density and a scalar density suggests otherwise. Evidently no one has spotted the absolute scalar density in GTR simply by inspection. It follows that either one sometimes does not know an absolute object when one sees it, or that the Andersonian analysis of absolute objects gives the wrong answer for this example. If the latter horn is accepted, then Peres’s version of GTR in terms of a conformal metric density and a scalar density (both varied in the action principle) has no absolute object, whereas unimodular GTR in terms of a conformal metric density and a *nonvariational* scalar density *has* an absolute object, although the theories have the same geometric objects and nearly the same field equations (supplemented with Nöther identities). Such a claim requires justification. Perhaps those who claim to spot absolute objects by inspection merely detect nonvariational objects in this instance? Whether a theory has nonvariational objects is, at least in some important examples, merely a question of its formulation, because tricks such as Rosen’s Lagrange multiplier or the parametrization of preferred coordinates into clock fields can be employed to turn nonvariational fields into variational ones.<sup>10</sup><sup>10</sup>10One hesitates to generalize too broadly on this matter. In GTR in terms of the conformal metric density (or its inverse) and a scalar density, the latter counts as absolute, so one might be tempted not to vary it in the action principle. But then the field equations are changed: a cosmological constant enters as a constant of integration, as is well known. The reason pertains to the mathematical form of the Lie derivative of a scalar density \[Israel, 1979\]: for weight $`w,`$ $`\mathrm{\pounds }_\xi \varphi =\xi ^\mu \varphi ,_\mu +w\varphi \xi ^\mu ,_\mu `$ and the $`w`$ term opens the door to the constant of integration. For scalars, it makes no difference whether one varies them as clock fields or not, because the form of the generalized Bianchi identities \[Sundermeyer, 1982\] and the linear independence of the gradients of the clock fields ensures that the same equations hold either way. For Rosen’s flat metric tensor trick, a new Lagrange multiplier field is introduced. Thus the consequences of changing a nonvariational field into a variational one or *vice versa* depend on which sort of geometric object is involved. This matter could use further study, perhaps with an eye on work on first-order and second-order actions for theories in which all fields are variational \[Ray, 1975, Lindström, 1988\]. The absoluteness of the scalar density in GTR implies that it can be varied only at the cost of ceasing to call GTR “Lagrangian-based” (*c.f.* \[Thorne et al., 1973\]). If the having or lacking of absolute objects is merely a formal feature of a theory, then some new way of escaping the Kretschmann objection to the physical vacuity of general covariance \[Norton, 2003\] is needed. Absent much healthy competition, the Andersonian project is worthy of attention even if its widely advertised diagnosis of the novelty of GTR is incorrect.
If the novelty of GTR does not consist in its lacking absolute objects (given Anderson’s definition of them), still Anderson’s project of analyzing the novelty of GTR might be fixable. There are indeed interesting novel features of GTR that Anderson’s framework uncovers or suggests. For example, GTR apparently is novel in having an external symmetry group involving arbitrary functions of space and time and in having a group as large as the volume-preserving diffeomorphisms. While Hiskes’s proposal to invoke variational principles was too crude, some more sophisticated effort might succeed. It is not presently clear whether it is best to admit that GTR has an absolute object or to redefine absolute objects to keep GTR from having any, if possible, but it seems worthwhile to consider the question.
## 9 Conclusion
Reviewing the Anderson-Friedman absolute objects program and various possible counterexamples yields several conclusions. First, eliminating irrelevant fields or portions thereof vindicates Friedman’s resolution of the Jones-Geroch dust counterexample and apparently resolves the new tetrad-spinor counterexample. Second, limitation of the mathematics to tensor fields has been detrimental by obscuring from view the tetrad-spinor and scalar density cases, while leading to an overestimate of the force of Torretti’s constant curvature spaces example. The mathematical theory of geometric objects is important for consideration of absolute objects. In particular, the geometric objects used should be irreducible. Third, bringing into the philosophical discussion some neglected physics literature sheds light on various issues. Finally, the scalar density counterexample, which arguably is the only real problem for the Anderson-Friedman framework of the four considered here, shows that either GTR has an absolute object or the Andersonian definition of absolute objects is flawed.
## 10 Acknowledgments
The author thanks Don Howard, Jeremy Butterfield, Brandon Fogel, Roger Jones and David Malament for helpful discussion and comments on the manuscript; Harvey Brown and John Norton for useful discussions; A. Camp, Ray Jensen and P. Nelson for bibliographic assistance; and, as noted above, Robert Geroch for suggesting the scalar density counterexample.
|
warning/0506/cond-mat0506554.html
|
ar5iv
|
text
|
# Thermally Increasing Correlation/Modulation Lengths and Other Selection Rules in Systems with Long Range Interactions
## I Introduction
Systems with long range interactions abound in nature long . Several examples in the current condensed matter arena include Coulomb gases (plasmas) which encompass- amongst many others, the omnipresent free electron gas, quantum Hall systems lilly , QHE , Fogler , adatoms on metallic surfaces, amphiphilic systems amp , interacting elastic defects (dislocations and disclinations) in solids vortex3 , interactions amongst vortices in fluid mechanics vortex1 and superconductors vortex2 , crumpled membrane systems Seul , wave-particle interactions wavpart , interactions amongst holes in cuprate superconductors steve , us , zohar , Low , carlson , manganates and nickelates cheong , golosov , some theories of structural glasses dk , peter , Gilles , new , and colloidal systems der , reich . Needless to say, the list of systems (and works) goes far beyond the little outlined here. Much of the work to date focused on the character of the transitions in these systems and the subtle thermodynamics that is often observed (e.g., the equivalence between different ensembles in many such systems is no longer as obvious, nor always correct, as it is in the “canonical” short range case barre ). Other very interesting aspects of different systems have been addressed in azbel .
In the current article, we focus on translationally invariant systems harboring several interactions of different ranges. To avoid many of the subtleties related to pure long range interactions, we will (unless stated otherwise) examine situations wherein screening is present. With these ingredients in place, we will find that the correlation functions of systems with screened long range interactions exhibit, in the exactly solvable large n limit, a peculiar- rather universal- divergence of correlation lengths at high temperatures in many such systems. Notwithstanding this effect, none of the standard correlation functions exhibit any pathology- all correlators decay monotonically with increasing temperatures. In many such systems, there are new emergent modulation lengths governing the size of various domains. We find that these modulation lengths often also adhere to various scaling laws, sharp crossovers and divergences at various temperatures (with no associated thermodynamic transition). We also find that in such systems, correlation lengths generically evolve into modulation lengths (and vice versa) at various temperatures. The behavior of correlation and modulation lengths as a function of temperature will afford us with certain selection rules on the possible underlying microscopic interactions. In their simplest incarnation, for systems hosting two competing interactions, our central results are two fold: (i) In canonical systems harboring competing short and long range interactions, modulated patterns appear whose characteristic modulation length is minimal within the ground state and slowly slowly increases as as temperature is raised. The modulation length $`L_D`$ associated with these patterns diverges at a crossover temperature $`T^{}`$ above which a uniform phase with multiple correlation lengths appears. The largest correlation length monotonically increases even as $`T\mathrm{}`$ although the prefactor associated with these correlation rapidly diminishes. (ii) By sharp contrast, in systems with only finite range interactions, the system exhibits a constant number of modulation and correlation lengths at all temperatures. Furthermore, in these systems, canonically, modulations (if they transpire) span a maximal length scale within the ground state and in the case of two competing interactions, the modulation $`L_D`$ length decreases as the temperature is raised. Armed with these general characteristics, we may easily discern the viable microscopic interactions (exact or effective) which underlie temperature dependent patterns such as those displayed in Fig.(1). Taken at face value, our results on the modulation lengths would suggest that any two component interaction theories underlying panel A of Fig.(1) may involve a confluence of long and short range interactions whereas those underlying panel B might involve only effective short range interactions. This may be said without knowing, a priori, the detailed microscopic interactions driving these non-uniform patterns. The simple treatment presented below does not account for the curvature of bubbles and the like. These may be easily augmented by inspecting energy functionals (and their associated free energy extrema) of various continuum field morphologies under the the addition of detailed domain wall tension forms- e.g. explicit line integrals along the perimeter where surface tension exists- and the imposition of additional constraints via Lagrange multipliers.
## II Correlation Functions in the large N limit- general considerations
All of the results reported in this article were computed within the spherical or large $`n`$ limit kac . We focus on translationally invariant systems whose Hamiltonian is given by
$`H={\displaystyle \frac{1}{2}}{\displaystyle \underset{\stackrel{}{x},\stackrel{}{y}}{}}V(\stackrel{}{x},\stackrel{}{y})S(\stackrel{}{x})S(\stackrel{}{y}).`$ (1)
Here, the fields $`S(\stackrel{}{x})`$ may portray classical spins or bosonic fields. The sites $`\stackrel{}{x}`$ and $`\stackrel{}{y}`$ lie on a hypercubic lattice of size $`N`$ of unit lattice constant. In what follows, $`v(\stackrel{}{k})`$ and $`S(\stackrel{}{k})`$ will denote the Fourier transforms of $`V(\stackrel{}{x}\stackrel{}{y})`$ and $`S(\stackrel{}{x})`$. For analytic interactions, $`v(|\stackrel{}{k}|)`$ is a function of $`k^2`$ (to avoid branch cuts). The spins satisfy a single global spherical constraint,
$`{\displaystyle \underset{\stackrel{}{x}}{}}S^2(\stackrel{}{x})=N`$ (2)
enforced by a Lagrange multiplier $`\mu `$ which renders the model quadratic (as both Eqs.(1, 2) are) and thus solvable, see e.g. us . In the below we report the results for classical fields; the results for bosonic systems are qualitatively the same with additional Matsubara frequency summations in tow.
For our purposes, it suffices to note that the two spin correlator,
$`G(\stackrel{}{x})S(0)S(\stackrel{}{x})=k_BT{\displaystyle \frac{d^dk}{(2\pi )^d}\frac{e^{i\stackrel{}{k}\stackrel{}{x}}}{v(\stackrel{}{k})+\mu }},`$ (3)
with $`d`$ the spatial dimension. To complete the characterization of the correlation functions at different temperatures, we note that the Lagrange multiplier $`\mu (T)`$ is given by the implicit equation $`1=G(\stackrel{}{x}=0)`$ (Eq.(2) fused with translational invariance). We now investigate the general character of the correlation functions given by Eq.(3) for rotationally invariant systems. If the minimum (minima) of $`v(|\stackrel{}{k}|)`$ occur(s) at momenta $`q`$ far from the Brillouin zone boundaries of the cubic lattice then we may set the range of integration in Eq.(3) to be unrestricted. The correlation function is then dominated by the location of the poles (and/or branch cuts) of $`[v(k)+\mu ]`$. Specifically, if $`k^s[v(\stackrel{}{k})+\mu ]`$, with $`s`$ an integer, is a polynomial
$`P(z)={\displaystyle \underset{m=0}{\overset{M}{}}}a_mz^m`$ (4)
in $`z=k^2`$ then, upon insertion into Eq.(3) we will find that the correlators generally display a net of $`M`$ correlation and modulation lengths. At very special temperatures, the Lagrange multiplier $`\mu (T)`$ may be such that several poles degenerate into one- thus lowering the number of correlation/modulation lengths at those special temperatures. It is important to emphasize that this multiplicity of roots and thus of correlation/modulation lengths occurs generally for any $`v(k)`$ for which $`P(z)`$ is a polynomial of degree $`M2`$\- multiple length scales appear irrespective of any competing interactions (alternating or uniform signs in $`k^sv(k)`$). What underlies multiple length scales is the existence of terms of different ranges (different powers of $`z`$ in the illustration above)- not frustration.
## III General preliminaries: Short and long range interactions
A screened Coulomb interactions of screening length $`\lambda `$ (i.e. a two spin potential $`V=\frac{1}{8\pi }\frac{1}{|\stackrel{}{x}\stackrel{}{y}|}e^{\lambda |\stackrel{}{x}\stackrel{}{y}|}`$ in $`d=3`$, and $`V=\frac{1}{4\pi }e^{\lambda |\stackrel{}{x}\stackrel{}{y}|}\mathrm{ln}|\stackrel{}{x}\stackrel{}{y}|`$ in $`d=2`$) has the continuum Fourier transformed interaction kernel $`v(k)=[k^2+\lambda ^2]^1`$. On a hypercubic lattice, the nearest neighbor interactions in real space have the lattice lattice Laplacian
$`\mathrm{\Delta }(\stackrel{}{k})=2{\displaystyle \underset{l=1}{\overset{d}{}}}(1\mathrm{cos}k_l)`$ (5)
as their Fourier transform. The real lattice Laplacian
$$\stackrel{}{x}|\mathrm{\Delta }|\stackrel{}{y}=\{\begin{array}{cc}2d\hfill & \text{ for }\stackrel{}{x}=\stackrel{}{y}\hfill \\ 1\hfill & \text{ for }|\stackrel{}{x}\stackrel{}{y}|=1\hfill \end{array}$$
(6)
Notice that $`\stackrel{}{x}|\mathrm{\Delta }^R|\stackrel{}{y}=0\text{ for }|\stackrel{}{x}\stackrel{}{y}|>R`$, where $`R`$ is the spatial range over which the interaction kernel is non-vanishing. Eq.(6) corresponds to a system is of Range=2,
$`\stackrel{}{x}|\mathrm{\Delta }^2|\stackrel{}{y}\text{ }=`$ $`2d(2d+2)\text{ for }\stackrel{}{x}=\stackrel{}{y}`$
$`4d\text{ for }|\stackrel{}{x}\stackrel{}{y}|=1`$
$`2\text{ for }(\stackrel{}{x}\stackrel{}{y})=(\pm \widehat{e}_{\mathrm{}}\pm \widehat{e}_{\mathrm{}^{}})\text{ where }\mathrm{}\mathrm{}^{}`$
$`1\text{ for a }\pm 2\widehat{e}_{\mathrm{}}\text{ separation}.`$ (7)
In the continuum (small $`k`$) limit, $`\mathrm{\Delta }z=k^2`$.
For simplicity, many of the examples which we will employ to illustrate the general premise of the behavior of correlations in systems hosting long and short range interactions, the kernel $`v(\stackrel{}{k})`$ (the Fourier transform of $`V(\stackrel{}{x},\stackrel{}{y})`$ of Eq.(1)) will, in the continuum limit, be a simple function of $`k^2`$ or of $`\mathrm{\Delta }`$ whenever the finite lattice constant is kept.
Simple effects of tension may be emulated via a $`g(\varphi )^2`$ term in the Hamiltonian where $`\varphi `$ is a constant in a uniform domain. Upon Fourier transforming, such squared gradient terms lead to an effective $`k^2`$ in the $`\varphi `$ space kernel. Similarly, the effects of curvature notable in many mixtures and membrane systems are often captured by terms involving $`(^2h)`$ with $`h`$ a variable parameterizing the profile; at times the interplay of such curvature terms with others leads, in the aftermath, to a simple short range $`k^4`$ term in the interaction kernel. An excellent review of these issues is addressed in Seul . Screened dipolar interactions and others may be easily emulated by terms such as $`(k^2+\lambda ^2)^p`$ with $`p>0`$.
## IV multiple range interactions
We now summarize the situation wherein two dominant interactions compete whenever one of the interactions is of infinite range (albeit being screened) while the other is of finite range (i.e. $`V`$ strictly vanishes for separations $`|\stackrel{}{x}\stackrel{}{y}|>R`$). Such situations arise in many systems Seul .
1) The high temperature correlation functions are, in many instances, sum of several exponential pieces; e.g. the correlation function
$`S(\stackrel{}{x})S(\stackrel{}{y})={\displaystyle \frac{1}{|\stackrel{}{x}\stackrel{}{y}|^{d2}}}(A_1e^{|\stackrel{}{x}\stackrel{}{y}|/\xi _1}`$
$`+A_2e^{|\stackrel{}{x}\stackrel{}{y}|/\xi _2}+\mathrm{}).`$ (8)
\[At least one of the correlation lengths ($`\xi _i`$) diverges.\]
2) At low temperatures, translationally frustrated systems with competing interactions on different length scales display modulations (i.e. an oscillatory spatial dependence of the correlation functions), e.g.
$`S(\stackrel{}{x})S(\stackrel{}{y}){\displaystyle \frac{\mathrm{cos}(p|\stackrel{}{x}\stackrel{}{y}|)}{|\stackrel{}{x}\stackrel{}{y}|^{d2}}}e^{|\stackrel{}{x}\stackrel{}{y}|/\xi }+\mathrm{}`$ (9)
3) When present in systems with frustrating long-range interactions (e.g. the Coulomb Frustrated Ferromagnet originally introduced to portray frustrated charge separation in the cuprates, us , zohar , Low ) for which, in Eq.(1), $`v(k)=k^2+Qk^2`$, the modulation lengths monotonically increase with increasing temperatures. In the specific case of the Coulomb frustrated ferromagnet, the modulation length $`(2\pi /p)`$ monotonically increases with temperatures until it diverges at a disorder line temperature $`T^{}`$zohar . We find that the near this temperature (i.e. for $`T=T^{}`$), the modulation length scales as $`(T^{}T)^{1/2}`$. It should be emphasized that this divergence notwithstanding, the system does not exhibit a phase transition at $`T^{}`$. The free energy is analytic at this temperature. Nevertheless, the free energy can be made to have a singularity at $`T=T^{}`$ if the long range interaction is turned off ($`Q=0`$)- this allows us to view the divergence of the modulation length as sparked by a critical point which is “avoided”. us ,zohar
4) In most systems, the sum of the number of correlation and the number of modulation lengths is conserved as a function of temperature apart from special crossover temperatures. In the example of the Coulomb Frustrated Ferromagnet, at temperatures higher than this crossover temperatures $`T>T^{}`$ two correlation lengths appear as in (1). After diverging at $`T=T^{}`$ the modulation length turns into a correlation length at higher temperatures. All crossovers may be traced by examining the dynamics of the poles and branch cuts of $`1/[v(k)+\mu ]`$ as the temperature (implicit in $`\mu `$) is varied. The merger of two or more poles at special temperatures leads to a temporary annihilation of one (or more length) which is generically restored as the temperature is continuously varied.
5) In systems with only two finite range interactions (of different scales), the modulation length monotonically decreases with increasing temperatures. This, combined with (3), affords a stringent selection rule on the viable interactions underlying various experimentally observed modulation lengths. In canonical systems harboring only finite range interactions, the number of correlation lengths and the number of modulation lengths are both independently conserved at all temperatures (much unlike the case for long range interactions where only the sum of the two is conserved).
Although these characteristics are general, it is useful to provide expressions for specific cases. In what follows, we briefly sketch the behavior when two interactions of two different ranges, appear in unison. In section(V), we discuss the case of an infinite range interaction (wherein $`\stackrel{}{x}|V|\stackrel{}{y}0`$ for all $`\stackrel{}{x}\stackrel{}{y}`$) screened Coulomb interaction existing side by side with a short range nearest neighbor exchange interaction. We follow, in section(VI) by an investigation of a system hosting several finite range interactions. We the consider, in section(VII), with brief remarks concerning the behavior in the case of a system in which long range interactions compete. General remarks concerning the possibility of first order transitions (section(VIII)) in the modulation length and a rather universal domain length exponent (section(IX)) conclude the article.
## V Coexisting short and long range interactions
We begin with the “Screened Coulomb Ferromagnet”, capturing the competition between a repulsive screened Coulomb interactions and short range attractions. Here, the Fourier transform of the interaction kernel of Eq.(1) is
$`v(k)=k^2+{\displaystyle \frac{Q}{k^2+\lambda ^2}}`$ (10)
wherein a screened Coulomb interaction competes with a short range ferromagnetic interaction. Not too surprisingly, such an kernel naturally appears in many systems e.g. screened models of frustrated phase separation in the cuprates steve . Here, at high temperatures ($`T>T^{}`$ wherein a $`T^{}`$ is given by $`\mu (T^{})=\lambda ^2+2\sqrt{Q}`$), the pair correlator in dimension $`d=3`$
$`G(\stackrel{}{x})={\displaystyle \frac{k_BT}{4\pi |\stackrel{}{x}|}}{\displaystyle \frac{1}{\beta ^2\alpha ^2}}`$
$`\times [e^{\alpha |\stackrel{}{x}|}(\lambda ^2\alpha ^2)e^{\beta |\stackrel{}{x}|}(\lambda ^2\beta ^2)].`$ (11)
Here,
$`\alpha ^2,\beta ^2={\displaystyle \frac{\lambda ^2+\mu \sqrt{(\lambda ^2\mu )^24Q}}{2}}.`$ (12)
For $`T<T^{}`$,
$`G(\stackrel{}{x})={\displaystyle \frac{k_BT}{8\alpha _1\alpha _2\pi |\stackrel{}{x}|}}e^{\alpha _1|\stackrel{}{x}|}`$
$`\times [(\lambda ^2\alpha _1^2+\alpha _2^2)\mathrm{sin}\alpha _2|\stackrel{}{x}|+2\alpha _1\alpha _2\mathrm{cos}\alpha _2|\stackrel{}{x}|]`$ (13)
where $`\alpha =\alpha _1+i\alpha _2=\beta ^{}`$.
Similarly, in $`d=2`$, for $`T>T^{}`$,
$`G(\stackrel{}{x})={\displaystyle \frac{k_BT}{2\pi }}{\displaystyle \frac{1}{\beta ^2\alpha ^2}}[(\lambda ^2\alpha ^2)K_0(\alpha |\stackrel{}{x}|)`$
$`+(\beta ^2\lambda ^2)K_0(\beta |\stackrel{}{x}|)],`$ (14)
with the Bessel function $`K_0(x)=_0^{\mathrm{}}𝑑t\frac{\mathrm{cos}xt}{\sqrt{1+t^2}}`$. Much as in the three dimensional case, the high temperature correlator may be analytically continued to temperatures $`T<T^{}`$.
We alert the reader that two correlation lengths appear for $`\mu ^2>4Q`$ (including all unfrustrated screened attractive Coulomb ferromagnets (those with $`Q<0`$)). The evolution of the correlation functions may be traced by examining the dynamics of the poles in the complex $`k`$ plane as a function of temperature. At high temperatures, the correlation functions are borne by poles lying on the imaginary axis. In the high temperature limit ($`T\mathrm{}`$), one of the poles tends to $`k=0^+`$ leading to a divergent correlation length! No paradoxes are, however, encountered in this limit as the prefactor associated with this correlation length (as seen in Eqs.(12, 14) tends to zero as $`T\mathrm{}`$ and all correlations decay monotonically with increasing temperature. At $`T=T^{}`$ the poles merge in pairs at $`k=\pm i\sqrt{\lambda ^2+\sqrt{Q}}`$. Henceforth, at lower temperatures, the poles move off the imaginary axis (leading in turn to oscillations in the correlation functions). The norm of the poles, $`|\alpha |=(Q+\lambda ^2\mu (T))^{1/4}`$, becomes constant in the limit of vanishing screening ($`\lambda =0`$) wherein the after merging at $`T=T^{}`$, the poles slide along a circle. In the low temperature limit of the unscreened Coulomb ferromagnet, the poles hit the real axis at finite $`k`$, reflecting oscillatory modulations in the ground state. In the presence of screening, the pole trajectories are slightly skewed yet for $`Q>\lambda ^4`$, $`\alpha `$ tends to the ground state modulation wavenumber $`\sqrt{\sqrt{Q}\lambda ^2}`$. If the screening is sufficiently large, i.e. if the screening length is shorter than the natural period favored by a balance between the unscreened Coulomb interaction and the nearest neighbor attraction ($`\lambda >Q^{1/4}`$), then the correlation functions never exhibit oscillations. In such instances, the poles continuously stay on the imaginary axis and, at low temperatures, one pair of poles veers towards $`k=0`$ reflecting the uniform ground state of the heavily screened system. In Figs.(2,3), the evolution of the pole locations in traced at different temperatures $`(T>T^{},T<T^{})`$ in the limit of weak screening. To summarize the observed physics at hand, at high temperatures $`G(x)`$ is a sum of two decaying exponentials (one of which has a correlation length which diverges in the high temperature limit). For $`T<T^{}`$ in under-screened systems, one of the correlation lengths transforms into a modulation length characterizing low temperature oscillations. At the cross-over temperature $`T^{}`$, the modulation length is infinite. As the temperature is progressively lowered, the modulation length decreases in size- until it reaches its ground state value. Connoisseurs may recognize $`T^{}(Q,\lambda )`$ as a “disorder line” like temperature.
An analytical thermodynamic crossover does occur at $`T=T^{}`$. A large $`n`$ calculation of the free energy via equipartition reveals that the internal energy per partice
$`{\displaystyle \frac{U}{N}}={\displaystyle \frac{1}{2}}(k_BT\mu ),`$ (15)
To ascertain a crossover in $`U`$ and that in other thermodynamic functions, the forms of $`\mu `$ both above and below $`T^{}`$ may be easily gleaned from the spherical normalization condition to find that the real valued functional form of $`\mu (T)`$ changes explainlong .
We note, in passing, that the system orders at $`T=T_c`$ given by
$`{\displaystyle \frac{1}{k_BT_c}}={\displaystyle \frac{d^dk}{(2\pi )^d}\frac{1}{v(\stackrel{}{k})v(\stackrel{}{q})}}.`$ (16)
Here, for $`Q>\lambda ^4`$, the modulus of the minimizing (ground state) wavenumber ($`|\stackrel{}{q}|`$) is given by
$`q={\displaystyle \frac{2\pi }{L_D^g}}=\sqrt{\sqrt{Q}\lambda ^2},`$ (17)
with $`L_D^g`$ the ground state modulation length. Associated with this wavenumber is the kernel $`v(\stackrel{}{q})=2\sqrt{Q}\lambda ^2`$ to be inserted in Eq.(16) for an evaluation of the critical temperature $`T_c`$. Similarly, the ground state wavenumber $`\stackrel{}{q}=0`$ whenever $`Q<\lambda ^4`$. Needless to say, whenever $`Q>\lambda ^4`$ and modulations transpire for temperatures $`T<T^{}`$, the critical temperature at which the chemical potential of Eq.(3), $`\mu (T_c)=\lambda ^22\sqrt{Q}`$, is lower than the crossover temperature $`T^{}`$ (given by $`\mu (T^{})=\lambda ^2+2\sqrt{Q}`$) at which modulations first start to appear. The Yukawa Ferromagnet is found to have $`T_c(Q=\lambda ^4)>0\text{ in }d>4\text{ }`$ and in any dimension $`T_c(Q>\lambda ^4)=0`$. For small finite $`n`$ a first order Brazovskii transition may replace the continuous transition occurring at $`T_c`$ within the large $`n`$ limit Braz . Depending on parameter values such an equilibrium transition may or may not transpire before a possible glass transition may occur peter .
## VI Multiple short range interactions
Our central thesis is that in many models in which short range interactions compete with one another, the modulation length always varies with temperature. This is in contrast to in systems with infinite range interactions wherein the modulation length increases as temperature is raised.
We now illustrate this general premise by a few examples: (i) With the conventions of the Fourier transformed kernel of Eq.(1) and Eqs.( 5, 6), the Fourier transformed interaction kernel
$$v(\stackrel{}{k})=[\mathrm{\Delta }^3\mathrm{\Delta }_0^3]^2$$
(18)
corresponds to a finite range interaction linking sites which are, at most, six lattice units apart. In what follows we employ the shorthand $`c(\mu \mu _{min})`$ with $`\mu _{\mathrm{min}}=\mathrm{\Delta }_0^6`$. The poles, $`\{\mathrm{\Delta }_\alpha ^\pm \}_{\alpha =1}^3`$, of the correlator $`G(\stackrel{}{k})=k_BT[c+v(\stackrel{}{k})]^1`$ are given by
$`[\mathrm{\Delta }_0^3\pm ı\sqrt{c}]^{1/3}\mathrm{exp}(2\pi \alpha ı/3).`$ (19)
In the complex plane, the poles $`\{\mathrm{\Delta }_\alpha ^\pm \}`$ lie on the vertices of a hexagon. There exist, at least, two poles with different values of $`|Im\{k_i\}|`$ leading to, at least, two different correlation lengths, $`\xi _i=|Im\{k_i\}|^1`$ at all temperatures (and apart from special degenerate usually to three correlation lengths). The existence of multiple correlation lengths $`\{\xi _i\}`$ are the rule in systems of multiple range interactions- including unfrustrated systems with no competing interactions. Their presence (as well as the existence of multiple modulation lengths) enriches the scope of possible scaling functions, allowing us to construct a greater multitude of scaling functions $`F(\mathrm{\Gamma };\{\xi _i\},\{L_{D;i}\})`$, with $`\mathrm{\Gamma }`$ a set of external parameters, than those present in standard critical systems where only one length (a single correlation length) sets the scale. For a finite ranged interaction kernel $`V`$ which is a general polynomial of the lattice Laplacian $`\mathrm{\Delta }`$, the system possesses several modulation lengths and several correlation lengths at all temperatures; barring few exceptions - their net number is conserved as temperature is varied. Generically, for a finite ranged interaction which is a polynomial in the Lattice Laplacian $`\mathrm{\Delta }`$, the system possesses a fixed number of correlation lengths and a fixed number of modulation lengths; there is no sharp analogue of $`T_i`$ wherein modulation lengths turn into correlation lengths. Such multiple correlation lengths often present for such kernels $`v(\stackrel{}{k})`$ which are functions of the lattice Laplacian $`\mathrm{\Delta }`$ are accompanied by a $`T_c`$ discontinuity (an “avoided critical point” us , zohar ) in sufficiently high dimension when appropriate competition is present to allow real roots $`0<\{\mathrm{\Delta }_i\}<4d`$ when $`\mu =\mu _{min}`$. All cross-over temperatures \[$`T_{i=1,2,\mathrm{},p}`$\] (including low dimensional systems which possess no critical behavior for zero frustration), at which correlation lengths disappear and turn into modulation lengths, tend continuously to the avoided critical temperature (or its analytic continuation for low dimensions - in high dimensions such an “avoided critical temperature” us , zohar becomes critical for zero frustration). These crossover temperatures \[$`T_{i=1,2,\mathrm{},p}`$\] are more dramatic for non-analytic functions of $`\mathrm{\Delta }`$ such as the frustrated screened Coulomb interaction investigated in Section(V), wherein the domain length diverges.
(ii) We next consider a very prototypical interaction kernel appearing, amongst others, in amphiphile problems and crumpled elastic membrane systems, see e.g. Seul . For the short-range (Teubner-Strey) correlator
$`G^1(\stackrel{}{k})=a_2+c_1k^2+c_2k^4,`$ (20)
it is a simple matter to show that
$`G(\stackrel{}{x}){\displaystyle \frac{\mathrm{sin}\kappa x}{\kappa x}}\mathrm{exp}[x/\xi ],`$ (21)
where
$`\kappa =\sqrt{{\displaystyle \frac{1}{2}}\sqrt{{\displaystyle \frac{a_2}{c_2}}}{\displaystyle \frac{c_1}{4c_2}}}`$
$`\xi ^1=\sqrt{{\displaystyle \frac{1}{2}}\sqrt{{\displaystyle \frac{a_2}{c_2}}}+{\displaystyle \frac{c_1}{4c_2}}}.`$ (22)
In amphiphilic systems $`a_2`$ and $`c_{1,2}`$ are functions of amphiphile concentration (as well as temperature).
The above two examples reaffirm out claim that in many simple thermodynamical models, short range interactions, the modulation length (if it exists) typically increases as the temperature is lowered. The converse typically occurs in systems hosting competing infinite range and finite range interactions.
When, in the notation of the Fourier transformed kernel of Eq.(1), $`v(\stackrel{}{k})=k^4`$,
$`G(\stackrel{}{x})={\displaystyle \frac{1}{4\pi x\sqrt{\mu }}}\mathrm{exp}[x/\xi ]\mathrm{sin}\kappa x.`$ (23)
Thus within the spherical model of this range two system with a sole overall ferromagnetic interaction ($`v(\stackrel{}{k})=k^4`$) will display thermally induced oscillations. At $`T=0`$ the (ferromagnetic) ground state is unmodulated.
## VII Multiple Long Range Interactions
A different behavior is seen when two long range interactions compete (e.g. $`v(k)=Ak^2+Bk^4`$ with $`A<0`$ and $`B>0`$). In such instances, modulations are present at all temperatures. Moreover, by sharp contrast to the competing long-range and short range interactions investigated earlier, within the large $`n`$ limit, the modulations become more and more acute with a length which tend to zero in the high temperature limit.
## VIII First order transitions in the modulation length
In the examples furnished above and several of our general maxims in the introduction, we focused on systems in which only two interactions of different ranges exist in a single system. In these systems, we found within the large $`N`$ limit that the modulation lengths were always monotonic in temperature. Needless to say, this need not be the case yet within the large $`n`$ limit this generally requires the existence of interactions spanning more ranges. The ground state modulation lengths (the reciprocals of Fourier modes $`\{\stackrel{}{q}_i\}`$ minimizing the interaction kernel) need not be continuous as a function of the various parameters: a “first order transition” in the value of the ground state modulation lengths can occur. Such a possibility is quite obvious and need not be expanded upon in depth. Consider, for instance, the Range=3 interaction kernel
$`v(\stackrel{}{k})=a[\mathrm{\Delta }+ϵ]+{\displaystyle \frac{1}{2}}b[\mathrm{\Delta }+ϵ]^2+{\displaystyle \frac{1}{3}}c[\mathrm{\Delta }+ϵ]^3,`$ (24)
with \[$`0<ϵ1`$\] and $`c>0`$. If $`a>0`$ and $`b<0`$, then there are three minima, i.e. $`[\mathrm{\Delta }+ϵ]=0`$ and $`[\mathrm{\Delta }+ϵ]=\pm m_+^2`$ where $`m_+^2=\frac{1}{2c}[b+\sqrt{b^24ac}]`$. the locus of points in the $`ab`$ plane where the three minima are equal is determined by $`v(\stackrel{}{k})=0`$, which leads to $`m_+^2=\frac{4a}{b}`$. Thus, $`b=4\sqrt{ca/3}`$ is a line of “first order transitions”,in which the minimizing $`[\mathrm{\Delta }+ϵ]`$ (and thus the minimizing wavenumbers) changes discontinuously by an amount $`\mathrm{\Delta }m=(\frac{4a}{b})^{1/2}=(\frac{3a}{c})^{1/4}`$.
## IX A Universal Domain length exponent
Invoking the normalization condition $`G(\stackrel{}{x}=0)=1`$ in Eq.(3), we find that given the competing screened Coulomb and short range attraction of Eq.(10) and more general kernels $`v(\stackrel{}{k})`$ in which infinite-range interactions augment finite ones, in any problem in which the crossover temperature $`T^{}`$ is finite, the modulation length
$`L_D(T^{}T)^{\nu _L},`$ (25)
with the (large $`n`$) domain length exponent $`\nu _L=1/2`$ in any dimension $`d`$.
## X Conclusions
In conclusion, our major finding is a general evolution of modulation and correlation lengths as a function of temperature in different classes of systems, those harboring infinite range (including screened) interactions and those having only short range interactions. These “selection rules” impose constraints on candidate theories describing a system in which the empirical behavior of modulation lengths is known. We further elucidated on the peculiar thermal evolution of the multiple correlation lengths in many systems having long range interactions, the largest of which may increase monotonically in temperature (even as $`T\mathrm{}`$).
It is a pleasure to acknowledge many discussions with David Andelman, Lincoln Chayes, Daniel Kivelson, Steven Kivelson, Joseph Rudnick, Gilles Tarjus, and Peter Wolynes.
* Current address: Theoretical Division, Los Alamos National Laboratory, Los Alamos, NM 87545, USA
|
warning/0506/math0506508.html
|
ar5iv
|
text
|
# A Small-Gain Theorem for Monotone Systems with Multi-Valued Input-State Characteristics
## I Introduction
The recent extension of the theory of monotone dynamical systems to monotone input-output (i/o) systems has proven to be very useful in analyzing the global behavior of many important dynamics; see for example , and see Section II below for the relevant definitions. (See also for a detailed account of monotone dynamical systems.) Of particular interest in this literature are feedback interconnections of subsystems–or “modules”–that are monotone and that possess a unique globally asymptotically stable equilibrium, obviously depending on the particular (constant) input applied. This has lead to the introduction of the notion of input-state (i/s) characteristics, which are maps assigning to each constant input value the particular equilibrium point to which solutions converge. In many applications, this assignment is exactly the type of quantitative information that is available from experiments (such as gene expression levels, for instance). Monotonicity, on the other hand, may be considered as a qualitative or structural property of an i/o system; see the graphical tests for monotonicity in for example. These two ingredients, monotonicity of the subsystems and existence of characteristics, are key to proving the small-gain theorems in . (For small-gain theorems for nonlinear but not necessarily monotone systems, see .)
In practice however, many monotone i/o systems subject to constant inputs possess several equilibria and all solutions converge to one of them, although distinct solutions may converge to distinct equilibria. Such systems are sometimes called multi-stable. In fact, since monotone i/o systems subject to constant inputs are monotone dynamical systems, this type of global behavior is to be expected (see ). This suggests that the notion of an i/s characteristic ought to be generalized to a multi-valued map which assigns to each constant input value the set of all possible equilibria to which solutions converge.
This naturally leads to the question of whether the known small-gain theorem for monotone systems in remains valid if instead of the original notion of i/s characteristics, one assumes the existence of multi-valued characteristics for the subsystems. The purpose of our paper is to show that such an extension is indeed possible. In our main result, we prove that a negative feedback interconnection of monotone i/o subsystems with multi-valued characteristics is itself multi-stable, provided that all the solutions of a particular discrete-time inclusion (which is typically of much lower dimension than the subsystems) converge.
Our work provides a significant extension of the Angeli-Sontag monotone control systems theory because requires singleton-valued characteristics and therefore globally asymptotically stable equilibria. For other approaches to proving multi-stability, see (where positive feedback interconnections of monotone i/o subsystems are considered and the trajectories converge for almost all initial values) and (which is based on density functions and also concludes convergence for almost all initial values). This earlier work does not include ours because for example (a) our results provide global stabilization from all initial values, (b) we do not require any regularity such as singleton-valuedness, differentiability, or non-degeneracy for the i/s characteristics, and (c) our results are intrinsic in the sense that we make no use of Lyapunov or density functions.
This note is organized as follows. In Section II, we provide the necessary definitions and background for monotone control systems, multi-valued characteristics, weakly non-decreasing set-valued maps, and asymptotically autonomous systems. In Section III, we state our small-gain theorem and discuss its relationship to the small-gain theorems in . In Section IV, we prove our theorem and we illustrate our theorem in Section V. We close in Section VI with some suggestions for future research.
## II Background and Motivation
### II-A Monotonicity and Characteristics
We next provide the relevant definitions for monotone control systems and input-state characteristics. While our monotonicity definitions follow , our treatment of characteristics is novel because we allow discontinuous multi-valued characteristics and unstable equilibria. Our general setting is that of an input-output (i/o) system
$$\dot{x}=f(x,u),y=h(x),x𝒳,u𝒰,y𝒴$$
(1)
where $`𝒳^n`$ is the closure of its interior and partially ordered, $`𝒰`$ and $`𝒴`$ are subsets of partially ordered Euclidean spaces $`_𝒰`$ and $`_𝒴`$ respectively, and $`f`$ and $`h`$ are locally Lipschitz on some open set $`X`$ containing $`𝒳`$. We refer to $`𝒳`$ as the state space of (1), $`𝒰`$ as its input space, and $`𝒴`$ as its output space. In general, $`𝒳`$ will not be a linear space, since for example we often take $`𝒳=_0^n:=\{x^n:x_i0i\}`$. We use $``$ to denote the partial orders on all our spaces, bearing in mind that the partial orders on our various spaces could differ.
The set of control functions (also called inputs) for (1), which we denote by $`𝒰_{\mathrm{}}`$, consists of all locally essentially bounded Lebesgue measurable functions $`𝐮:𝒰`$, and we let $`t\varphi (t,x_o,𝐮)`$ denote the trajectory of (1) for any given initial value $`x_o𝒳`$ and $`𝐮𝒰_{\mathrm{}}`$. We always assume our dynamics $`f`$ are forward complete and $`𝒳`$-invariant, which means that $`\varphi (,x_o,𝐮)`$ is defined on $`[0,\mathrm{})`$ and valued in $`𝒳`$ for all $`x_o𝒳`$ and $`𝐮𝒰_{\mathrm{}}`$. Since we will be considering more than one dynamic, we often use sub- or superscripts to emphasize the state space variable or dynamic, so for example $`\varphi ^f`$ is the flow map for the dynamic $`f`$ and $`𝒴_z`$ is the output space for an i/o system with state variable $`z`$.
We always assume that our partial orders $``$ are induced by distinguished closed nonempty sets $`K`$ (called ordering cones) and we sometimes write $`K_𝒰`$ to indicate the cone inducing the partial order on the input space $`𝒰`$ and similarly for the other partial orders. We always assume $`K`$ is a pointed convex cone, meaning,
$$aKKa0,K+KK,K(K)=\{0\}.$$
When we say that a cone $`K`$ induces a partial order $``$, we mean the following: $`xy`$ if and only if $`yxK`$. This induces a partial order on the set of control functions $`𝒰_{\mathrm{}}`$ as follows: $`𝐮𝐯`$ if and only if $`𝐮(t)𝐯(t)`$ for Lebesgue almost all (a.a) $`t0`$. A function $`g`$ mapping a partially ordered space into another partially ordered space is called monotone provided: $`xy`$ implies $`g(x)g(y)`$. We say that (1) is single-input single-output (SISO) provided $`_𝒰=_𝒴=`$, taken with the usual order, i.e., the order induced by the cone $`K=[0,\mathrm{})`$.
###### Definition II.1
We say that (1) is monotone provided $`h`$ is monotone and
$$(pq\mathrm{and}𝐮𝐯)(\varphi (t,p,𝐮)\varphi (t,q,𝐯)t0)$$
holds for all $`p,q𝒳`$ and $`𝐮,𝐯𝒰_{\mathrm{}}`$.
We let $`\mathrm{Equil}(f)`$ denote the set of all equilibrium pairs for our dynamic $`f`$, namely, the set of all input-state pairs $`(\overline{u},\overline{x})`$ such that $`f(\overline{x},\overline{u})=0`$. For each $`(\overline{u},\overline{x})\mathrm{Equil}(f)`$, we let $`𝒟^f(\overline{u},\overline{x})`$ denote the domain of attraction of $`\dot{x}=f(x,\overline{u})`$ to $`\overline{x}`$, namely, the set of all $`p𝒳`$ for which $`\varphi (t,p,\overline{u})\overline{x}`$ as $`t+\mathrm{}`$, where $`\varphi `$ is the flow map for $`f`$. Since we are not assuming our equilibria are stable, the sets $`𝒟^f(\overline{u},\overline{x})`$ are not necessarily open and could even be singletons; see below for an example where $`𝒟^f(\overline{u},\overline{x})`$ is not open. Given $`(\overline{u},\overline{x})\mathrm{Equil}(f)`$, we say that $`f`$ is static Lyapunov stable at $`(\overline{u},\overline{x})`$ provided the following condition holds for all $`\epsilon >0`$: There exists $`\delta =\delta (\overline{u},\overline{x},\epsilon )>0`$ such that for all $`x_o𝒟^f(\overline{u},\overline{x})_\delta (\overline{x})(=`$ radius $`\delta `$ open ball centered at $`\overline{x}`$), we have $`|\varphi (t,x_o,\overline{u})\overline{x}|\epsilon `$ for all $`t0`$.
Recall the following notions from , in which we let $`f^{\overline{u}}`$ denote the constant input system $`f(,\overline{u})`$ for each $`\overline{u}𝒰`$. Given $`\overline{u}𝒰`$, we say that two nonempty (but not necessarily distinct) sets $`M_1,M_2𝒳`$ are $`f^{\overline{u}}`$-chained provided there exists a value $`y𝒳(M_1M_2)`$ and a trajectory $`x:𝒳`$ for $`f^{\overline{u}}`$ satisfying $`x(0)=y`$ whose $`\alpha `$-limit set $`\alpha (x):=\{\overline{x((\mathrm{},t])}:t0\}`$ lies in $`M_1`$ and whose $`\omega `$-limit set $`\omega (x):=\{\overline{x([t,+\mathrm{}))}:t0\}`$ lies in $`M_2`$. We say that a finite collection of nonempty sets $`M_1,M_2,\mathrm{},M_r𝒳`$ is $`f^{\overline{u}}`$-cyclically chained provided the following holds: If $`r=1`$, then $`M_1`$ is $`f^{\overline{u}}`$-chained to itself; and if $`r>1`$, then $`M_i`$ is $`f^{\overline{u}}`$-chained to $`M_{i+1}`$ for $`i=1,2,\mathrm{},r1`$ and $`M_r`$ is $`f^{\overline{u}}`$-chained to $`M_1`$. In this case, we call $`\{M_i\}`$ an $`f^{\overline{u}}`$-cycle. An $`f^{\overline{u}}`$-equilibrium is defined to be any point $`\overline{x}𝒳`$ such that $`f(\overline{x},\overline{u})=0`$. A set $`M𝒳`$ is called $`f^{\overline{u}}`$-invariant provided the flow map $`\varphi `$ for $`f`$ satisfies $`M=\{\varphi (t,x,\overline{u}):t0,xM\}`$. A compact $`f^{\overline{u}}`$-invariant set $`M𝒳`$ is called $`f^{\overline{u}}`$-isolated compact invariant provided there exists an open set $`𝒰𝒳`$ such that there is no compact $`f^{\overline{u}}`$-invariant subset $`\stackrel{~}{M}𝒳`$ satisfying $`M\stackrel{~}{M}𝒰`$ except $`M`$. We use the symbol $``$ to denote a set-valued map (also called a multifunction), e.g., $`F:𝒵_1𝒵_2`$ means that $`F`$ assigns each $`p𝒵_1`$ a nonempty set $`F(p)𝒵_2`$.
###### Definition II.2
We say that (1) is endowed with a static input-state (i/s) characteristic $`k_x:𝒰𝒳`$ provided:
1. $`\mathrm{Graph}(k_x)=\mathrm{Equil}(f)`$;
2. $`\{𝒟^f(\overline{u},\overline{x}):\overline{x}k_x(\overline{u})\}=𝒳`$ for all $`\overline{u}𝒰`$;
3. $`f`$ is static Lyapunov stable at each $`(\overline{u},\overline{x})\mathrm{Equil}(f)`$; and
4. For each $`\overline{u}𝒰`$, $`k_x(\overline{u})`$ consists of $`f^{\overline{u}}`$-isolated compact invariant $`f^{\overline{u}}`$-equilibria and contains no $`f^{\overline{u}}`$-cycles.
In this case, we also call $`k_y:=hk_x`$ an input-output (i/o) characteristic for (1).
This definition reduces to the usual singleton-valued i/s characteristic definition in when $`\mathrm{Card}\{k_x(\overline{u})\}=1`$ for all $`\overline{u}𝒰`$. We will not use the static Lyapunov stability property in the proof of our small-gain theorem per se, but we still include it to make our definition of i/s characteristics include the singleton-valued characteristic definition in . Condition 3 in our definition is not implied by the other conditions in the definition, even if $`f`$ has no controls, since it is well-known that $`f`$ could admit an unstable globally attractive equilibrium; see for example \[7, pp. 191-4\]. Condition 2 in the definition says for each $`\overline{u}𝒰`$ and each initial state, the corresponding $`f^{\overline{u}}`$-trajectory asymptotically approaches some state $`\overline{x}k_x(\overline{u})`$ (where $`\overline{x}`$ can in principle depend on the initial state of the trajectory). The stipulation in the static Lyapunov stability definition that $`x_o𝒟^f(\overline{u},\overline{x})_\delta (\overline{x})`$ is motivated by the fact that our domains of attraction $`𝒟^f(\overline{u},\overline{x})`$ may or may not be open, even if there are no controls. Condition $`4`$ is needed to apply the theory of asymptotically autonomous systems; see Section II-C for the relevant definitions and details.
Remark: Condition $`4`$, and in particular the “no cycles” part, may be hard to check in practice, at least if the system dimension is higher than $`2`$, but can often be checked using monotonicity arguments. Consider for instance a monotone system $`\dot{x}=f(x)`$ having two $`f`$-isolated compact invariant equilibria $`p`$ and $`q`$ and assume that $`pq`$ (where the latter means that $`qp`$ belongs to the interior of the order cone $`K`$, which is assumed to be nonempty). Then there exist neighborhoods $`N_p`$ and $`N_q`$ of $`p`$ and $`q`$ respectively such that $`n_pn_q`$ for all $`n_pN_p`$ and $`n_qN_q`$. We show that $`\{p,q\}`$ cannot be an $`f`$-cycle. Suppose it was a cycle. Then there exist points $`y`$ and $`z`$ such that $`\alpha (y)=\{p\}`$, $`\omega (y)=\{q\}`$ and $`\alpha (z)=\{q\}`$, $`\omega (z)=\{p\}`$. It follows in particular that there exists $`T>0`$ large enough such that $`n_p:=\varphi (T,y)N_p`$ and $`n_q:=\varphi (T,z)N_q`$. Consider the strictly ordered initial conditions $`n_pn_q`$ for the monotone system $`\dot{x}=f(x)`$. Since $`\omega (n_p)=\{q\}`$ and $`\omega (n_q)=\{p\}`$, there exists $`\stackrel{~}{T}>0`$ large enough so that $`\varphi (\stackrel{~}{T},n_p)N_q`$ and $`\varphi (\stackrel{~}{T},n_q)N_p`$ and thus $`\varphi (\stackrel{~}{T},n_q)\varphi (\stackrel{~}{T},n_p)`$, which contradicts monotonicity of the system. The same argument can be used to rule out cycles containing more than two equilibria, if we assume that the equilibria are totally ordered by $``$ (that is, either $`pq`$ or $`qp`$ whenever $`p`$ and $`q`$ are distinct equilibria).
### II-B Weakly Non-Decreasing Set-Valued Maps
A basic property of singleton-valued i/s characteristics $`k_x`$ is that they are non-decreasing in the relevant partial orders, in the sense that the following holds for all $`u,v𝒰_x`$: $`uv`$ implies $`k_x(u)k_x(v)`$; see for the elementary proof. It is therefore natural to inquire about whether set-valued i/s characteristics posses some analogous (but more general) order-preserving property. This motivates the following definition and lemma:
###### Definition II.3
Let $`𝒵_1`$ and $`𝒵_2`$ be partially ordered Euclidean spaces and $`F:𝒵_1𝒵_2`$ be any set-valued map. We say that $`F`$ is weakly non-decreasing provided the following holds for all $`p,q𝒵_1`$ such that $`pq`$: For each $`k_pF(p)`$ and $`k_qF(q)`$, there exist $`r_pF(p)`$ and $`r_qF(q)`$ such that $`r_pk_q`$ and $`k_pr_q`$.
###### Lemma II.4
If $`k_x`$ is an i/s characteristic for (1) and (1) is monotone, then $`k_x`$ is weakly non-decreasing.
###### Proof:
Let $`p,q𝒰_x`$ be such that $`pq`$, let $`k_pk_x(p)`$ and $`k_qk_x(q)`$, and let $`\varphi `$ denote the flow map of $`f`$. The corresponding trajectories for the constant inputs satisfy $`\varphi (t,k_q,p)\varphi (t,k_q,q)=k_q`$ for all $`t0`$, and $`\varphi (t,k_q,p)r_p`$ for some $`r_pk_x(p)`$ as $`t+\mathrm{}`$, so $`r_pk_q`$ follows because ordering cones are closed. The other order inequality is proved similarly. ∎
Definition II.3 reduces to non-decreasingness in the relevant orders when $`F`$ is singleton-valued. We are especially interested in solution sequences $`w_k`$ satisfying discrete set-valued inclusions $`w_{k+1}F(w_k)`$ for all $`k`$ where $`F`$ is weakly non-decreasing. To further motivate our study of weakly non-decreasing multifunctions, let us first assume that $`F:[0,1][0,1]`$ is a singleton-valued and non-decreasing map in the usual orders (that is, $`F(x)F(y)`$ when $`xy`$). Then it is obvious that every solution of $`x_{k+1}=F(x_k)`$ converges. Indeed, either $`x_0F(x_0)`$ and then $`x_0F(x_0)F^2(x_0)\mathrm{}F^k(x_0)`$ for all $`k`$, so the sequence $`\{F^k(x_0)\}`$ must converge since it is bounded above by $`1`$; or else $`F(x_0)x_0`$, which leads to a non-increasing sequence $`\{F^k(x_0)\}`$. That converges as well since it is bounded below by $`0`$. On the other hand, this simple dynamical behavior will not occur in general for multi-valued, weakly non-decreasing maps.
To see why, consider the following simple example. Assume that $`F:[0,1][0,1]`$ is a multi-valued map whose graph consists of the union of three straight line segments: one connecting $`A=(0,0)`$ with $`B=(1/2,1/4)`$, a second connecting $`B`$ to $`C=(1/4,1/2)`$ (of slope $`1`$), and a third connecting $`C`$ with $`D=(1,1)`$. This “inverted Zorro map” is illustrated in Figure 1 below and is weakly non-decreasing in the usual orders. Then the inclusion $`x_{k+1}F(x_k)`$ has periodic points of period $`2`$. For instance, the periodic sequence $`\{1/2,1/4,1/2,1/4,\mathrm{}\}`$ is a solution of the inclusion. In fact, to every initial condition $`x_0[1/4,1/2]`$ corresponds a periodic sequence of period $`2`$ satisfying the inclusion, namely $`\{x_0,3/4x_0,x_0,3/4x_0,\mathrm{}\}`$ (since $`3/4xF(x)`$ for all $`x[1/4,1/2]`$).
These periodic sequences are caused by the fact that the slope of the middle line segment of the graph of $`F`$ is $`1`$. Any slight decrease of this slope will destroy the periodic points and leads to solutions that converge to one of the fixed points. For example, for arbitrary $`ϵ>0`$ we can define $`F_ϵ`$ as the map whose graph consists of three straight line segments connecting $`A`$ to $`B`$, $`B`$ to $`E=((1+2ϵ)/(4+4ϵ),1/2)`$ (so the slope of this line segment is $`1ϵ`$), and $`E`$ to $`D`$. Then every solution of the inclusion $`x_{k+1}F_ϵ(x_k)`$ will converge to one of the three fixed points of $`F`$. In fact, each solution sequence of this inclusion converges to either $`0`$ or $`1`$, except for the constant sequence at the middle fixed point $`\stackrel{~}{x}=(3+2ϵ)/(4(2+ϵ))`$. To see why, notice that if $`x_o>1/2`$, then $`(x_k,F_ϵ(x_k))`$ remains on the segment $`\overline{ED}`$, so $`x_k1`$ by the argument for the singleton-valued case. Similarly, if $`x_o<(1+2ϵ)/(4+4ϵ)`$, then $`(x_k,F_ϵ(x_k))`$ remains on $`\overline{AB}`$ so $`x_k0`$ again by the singleton-valued case; while if $`x_k`$ stays in $`[(1+2ϵ)/(4+4ϵ),1/2]`$, then $`x_{k+1}=(1+ϵ)x_k+\frac{3}{4}+\frac{ϵ}{2}`$ for all $`k`$. Then either $`x_k\stackrel{~}{x}`$, or else $`|x_{k+1}x_k|=(1+ϵ)^k|x_1x_o|+\mathrm{}`$ as $`k+\mathrm{}`$ which is impossible. Therefore, either $`x_k`$ stays at $`\stackrel{~}{x}`$, or else $`x_k`$ exits $`[(1+2ϵ)/(4+4ϵ),1/2]`$ and then converges to either $`0`$ or $`1`$, as claimed.
### II-C Asymptotically Autonomous Systems
We will be especially interested in dynamics for which the asymptotic behavior under constant inputs is known. We will then obtain information about the trajectories for not-necessarily constant inputs using the theory of asymptotically autonomous systems. Before turning to this theory, first recall the following “Converging-Input Converging-State” (CICS) Property. This property was shown in and was used in to study the stability of interconnected monotone systems. We use the CICS property at the very end of the proof of our main result (on p.IV).
###### Lemma II.5
Let $`\overline{u}𝒰`$, and let $`\overline{x}`$ be an asymptotically stable equilibrium point for $`f^{\overline{u}}`$. Let $`𝒦`$ be a compact subset of $`𝒟^f(\overline{u},\overline{x})`$. If $`x:[0,\mathrm{})𝒳`$ is a $`𝒦`$-recurrent trajectory of $`f`$ for some continuous input $`u:[0,\mathrm{})𝒰`$, and if $`u(t)\overline{u}`$ as $`t+\mathrm{}`$, then $`x(t)\overline{x}`$ as $`t+\mathrm{}`$.
Here $`𝒦`$-recurrent means for each $`T>0`$, there exists $`t>T`$ such that $`x(t)𝒦`$. One of the requirements of asymptotic stability of $`\overline{x}`$ (in addition to the convergence condition) is the following stability property: For each $`\epsilon >0`$, there exists $`\delta >0`$ such that $`|\varphi (t,\xi ,\overline{u})\overline{x}|\epsilon `$ for all $`\xi _\delta (\overline{x})`$ and $`t0`$. The proof of the CICS property in uses the fact that $`𝒟^f(\overline{u},\overline{x})`$ is open, which follows from the assumption that $`\overline{x}`$ is a stable equilibrium.
However, in our more general setting where the i/s characteristics are multi-valued, the domains of attraction will not necessarily be open, so the CICS property does not apply. Instead, we prove our result using the theory of asymptotically autonomous systems developed by Thieme in . To this end, we first note that Condition 2 from our definition of i/s characteristics implies the following equilibrium condition (EC) from :
* + For each $`\overline{u}𝒰`$, the $`\omega `$-limit set of any pre-compact $`f^{\overline{u}}`$-trajectory on $`[0,\mathrm{})`$ consists of an $`f^{\overline{u}}`$-equilibrium.
By an asymptotically autonomous system, we mean a system $`\dot{x}=H(t,x)`$ that admits a second dynamic $`\dot{x}=\overline{H}(x)`$ (called a limiting dynamic) such that $`H(t,x)\overline{H}(x)`$ as $`t+\mathrm{}`$ locally uniformly in $`x`$. For example, if $`u𝒰_{\mathrm{}}`$ is continuous and $`\overline{u}𝒰`$ is such that $`u(t)\overline{u}`$ as $`t+\mathrm{}`$, then for our locally Lipschitz dynamic $`f`$, we know $`\dot{x}=H(t,x):=f(x,u(t))`$ is asymptotically autonomous with limiting dynamic $`\dot{x}=\overline{H}(x):=f(x,\overline{u})`$. Using this observation, the following is then immediate from \[12, Corollary 4.3\] and our i/s characteristic definition:
###### Lemma II.6
Assume (1) is endowed with an i/s characteristic. Let $`\overline{u}𝒰`$ and $`u:[0,\mathrm{})𝒰`$ be any locally Lipschitz function for which $`u(t)\overline{u}`$ as $`t+\mathrm{}`$. Let $`x:[0,\mathrm{})𝒳`$ be any bounded trajectory for (1) and this input $`u(t)`$. Then $`x(t)`$ converges towards an $`f^{\overline{u}}`$-equilibrium as $`t+\mathrm{}`$.
If one drops the “no cycle” part of condition $`4`$ in Definition II.2, then the conclusion of the above Lemma does not necessarily hold; see for an example.
## III Statement and Discussion of Small-Gain Theorem
We turn next to our small-gain theorem, which generalizes \[1, Theorem 3\]. The main novelty of our result lies in its applicability to cases where one of the interconnected systems has a multi-valued i/s characteristic, but see Remark 3 below for a further extension for cases where both subsystems have multi-valued i/s characteristics. In what follows, an equilibrium of a discrete inclusion $`w_{k+1}F(w_k)`$ is defined to be any value $`\overline{w}`$ such that $`\overline{w}F(\overline{w})`$; the set of all equilibria for this inclusion is denoted by $`(F)`$. A multi-function $`F`$ is called locally bounded provided it maps bounded sets into bounded sets. We say that a continuous time dynamics $`F`$ has a pointwise globally attractive set $`S`$ provided each maximal trajectory $`\zeta (t)`$ for $`F`$ asymptotically approaches some point in $`S`$ (which could in principle depend on the specific trajectory) as $`t+\mathrm{}`$.
###### Theorem 1
Consider the following interconnection of two SISO dynamic systems:
$$\begin{array}{cc}\dot{x}=f_x(x,w),\hfill & y=h_x(x)\hfill \\ \dot{z}=f_z(z,y),\hfill & w=h_z(z)\hfill \end{array}$$
(2)
with $`𝒰_x=𝒴_z`$ and $`𝒰_z=𝒴_x`$. Assume the following:
1. The first system is monotone when its input $`w`$ and output $`y`$ are ordered by the “standard order” induced by the positive real semi-axis.
2. The second system is monotone when its input $`y`$ is ordered by the standard order and its output $`w`$ is ordered by the opposite order (induced by the negative real semi-axis).
3. The respective static i/s characteristics $`k_x`$ and $`k_z`$ exist with $`k_x`$ singleton-valued and $`k_z`$ locally bounded.
4. Each trajectory of (2) is bounded; and each solution sequence $`\{v_k\}`$ of $`v_{k+1}(k_yk_w)(v_k)`$ converges.
Then (2) has the pointwise globally attractive set $`\{\{k_x(\overline{w})\}\times (k_zk_y)(\overline{w}):\overline{w}(k_wk_y)\}`$.
In this setting, $`k_y=h_xk_x`$ and $`k_w=h_zk_z`$.
Our theorem differs from the small-gain theorem \[1, Theorem 3\] mainly in that (a) we replaced the single valuedness of $`k_z`$ with local boundedness of $`k_z`$, (b) we replaced the discrete system $`w_{k+1}=(k_wk_y)(w_k)`$ from with a discrete inclusion, and (c) we conclude that (2) is attracted to a set of equilibrium points rather than a single point as in . Moreover, in contrast to , our theorem gives global convergence of the interconnection from all initial values.
###### Remark 2
Assumption 4 of our theorem is equivalent to the following: $`4^{}`$. Each trajectory of (2) is bounded; and $`\{k_y(w_k)\}`$ converges for each solution sequence $`\{w_k\}`$ of $`w_{k+1}(k_wk_y)(w_k)`$. In fact, if Assumption 4 holds and $`w_k`$ is any solution of $`w_{k+1}(k_wk_y)(w_k)`$, then $`k_y(w_k)`$ converges because $`v_k=k_y(w_k)`$ is a solution sequence for $`v_{k+1}(k_yk_w)(v_k)`$. Conversely, if Assumption $`4^{}`$ holds, and if $`v_k`$ is any solution sequence of $`v_{k+1}(k_yk_w)(v_k)`$, then we can inductively find a new sequence $`r_k`$ such that $`v_{k+1}k_y(r_k)`$ and $`r_{k+1}(k_wk_y)(r_k)`$ for all $`k`$, so $`v_k`$ converges. On the other hand, it could be that Assumption 4 holds but that there exists a divergent sequence $`w_k`$ for $`w_{k+1}(k_wk_y)(w_k)`$. See Remark 4 for an example where this occurs. However, if the trajectories of (2) are bounded, and if each solution of $`w_{k+1}(k_wk_y)(w_k)`$ converges, then Assumption $`4^{}`$ (or equivalently Assumption $`4`$) holds because $`k_y`$ is continuous (by the arguments from \[1, Proposition V.5\] and our assumption that $`k_x`$ is singleton valued).
## IV Proof of Small-Gain Theorem
The following key lemma generalizes \[1, Proposition V.8\] to systems with multi-valued characteristics. In it, we set $`u_{inf}:=lim\; inf_{t+\mathrm{}}u(t)`$ and $`u_{sup}:=lim\; sup_{t+\mathrm{}}u(t)`$ for any continuous scalar function $`u`$ on $`[0,\mathrm{})`$.
###### Lemma IV.1
Under the hypotheses of Theorem 1, if $`(x(t),z(t))`$ is any trajectory of (2) and $`\zeta \omega (z)`$, then there exist $`k_{}k_z(y_{inf})`$ and $`k_+k_z(y_{sup})`$ such that $`k_{}\zeta k_+`$.
###### Proof:
We only prove the existence of $`k_{}`$ since the proof of the existence of $`k_+`$ is similar. Set $`\mu =y_{inf}`$ and let $`\xi `$ be the initial value for $`z(t)`$. Let $`t_j+\mathrm{}`$ and $`\mu _j\mu `$ be sequences such that $`\mu _j𝒰_z`$ and $`y(t)\mu _j`$ for all $`tt_j`$ and all $`j`$. We have the following for all $`tt_j`$ and $`j`$:
$$z(t)=\varphi (t,\xi ,y)=\varphi (tt_j,\varphi (t_j,\xi ,y),y(+t_j))\varphi (tt_j,\varphi (t_j,\xi ,y),\mu _j),$$
(3)
where $`\varphi `$ is the flow map for $`f_z`$ and the last order inequality follows from the monotonicity of the $`z`$-subsystem. Therefore, if $`z(s_l)\zeta `$ for some sequence $`s_l+\mathrm{}`$, then we can set $`t=s_l`$ in (3) and use the closedness of order cones to find values $`v_jk_z(\mu _j)`$ such that
$$\zeta \underset{l\mathrm{}}{lim}\varphi (s_lt_j,\varphi (t_j,\xi ,y),\mu _j)=v_jj.$$
(4)
Since $`k_z`$ is assumed to be locally bounded and has a closed graph (by the continuity of the dynamic $`f_z`$ in all arguments), we can find $`k_{}k_z(\mu )`$ such that $`\zeta v_jk_{}`$, possibly by passing to a subsequence without relabelling. This proves the desired inequality. ∎
Returning to the proof of our small-gain theorem, notice that since the output $`w`$ is ordered by the negative real semi-axis, and since $`k_z`$ is weakly non-decreasing (by Lemma II.4), it follows that
$$\underset{k_pk_w(p)}{\mathrm{max}}\underset{k_qk_w(q)}{\mathrm{min}}(k_pk_q)(pq)\mathrm{\hspace{0.33em}0}p,q𝒰_z.$$
(5)
In other words, for each $`p,q𝒰_z`$ and $`k_pk_w(p)`$, we can find $`k_qk_w(q)`$, such that $`k_pk_q`$ and $`pq`$ have opposite signs. Also, $`k_y`$ is continuous and non-decreasing, as shown in \[1, Propositions V.5 and V.8\] and Lemma II.4. Choose any initial value $`\xi `$ for the interconnection (2), and let $`(x(t),z(t))`$ denote the corresponding trajectory for (2) starting at $`\xi `$. This trajectory is defined on $`[0,\mathrm{})`$ since we are assuming our trajectories are bounded. Set $`w_+=w_{sup}`$, $`w_{}=w_{inf}`$, and similarly define $`y_\pm `$. Let $`z_+`$ (resp., $`z_{}`$) $`\omega (z)`$ be such that $`w_{}=h_z(z_+)`$ (resp., $`w_+=h_z(z_{})`$). These limits exist because $`h_z`$ is continuous and $`z(t)`$ is bounded in the closed set $`𝒳_z`$. By Lemma IV.1, we can find $`k_+k_z(y_+)`$ and $`k_{}k_z(y_{})`$ such that $`k_{}z_{}`$ and $`z_+k_+`$. Setting $`r_+^{\left(0\right)}=h_z(k_+)`$ and $`r_{}^{\left(0\right)}=h_z(k_{})`$ and recalling that $`w`$ reverses order gives
$$k_w(y_+)r_+^{\left(o\right)}w_{}w_+r_{}^{\left(o\right)}k_w(y_{}).$$
(6)
Since we are assuming $`k_x`$ is singleton-valued, the proof of \[1, Theorem 3\] gives
$$k_y(w_{})y_{}y_+k_y(w_+).$$
(7)
Combining (6) and (7) and recalling that $`k_y`$ is non-decreasing gives
$`(k_yk_w)(y_+)k_y(r_+^o)=:s_+^{\left(1\right)}`$ $``$ $`k_y(w_{})y_{}`$
$``$ $`y_+k_y(w_+)s_{}^{\left(1\right)}:=k_y(r_{}^{\left(o\right)})(k_yk_w)(y_{}).`$
In summary,
$$(k_yk_w)(y_+)s_+^{\left(1\right)}y_{}y_+s_{}^{\left(1\right)}(k_yk_w)(y_{}).$$
(9)
Since $`y_+s_{}^{\left(1\right)}`$ and $`r_+^{\left(0\right)}k_w(y_+)`$, we can use (5) to find $`r_+^{\left(1\right)}k_w(s_{}^{\left(1\right)})k_w(k_yk_w)(y_{})`$ such that $`r_+^{\left(1\right)}r_+^{\left(0\right)}`$. Since $`k_y`$ is non-decreasing, (IV) therefore gives
$$y_{}k_y(r_+^{\left(0\right)})k_y(r_+^{\left(1\right)})=:s_{}^{\left(2\right)}(k_yk_w)^2(y_{}).$$
(10)
Similarly, since $`y_{}s_+^{\left(1\right)}`$ and $`r_{}^{\left(0\right)}k_w(y_{})`$, we can use (5) to find $`r_{}^{\left(1\right)}k_w(s_+^{\left(1\right)})k_w(k_yk_w)(y_+)`$ such that $`r_{}^{\left(o\right)}r_{}^{\left(1\right)}`$. Hence, (IV) also gives
$$y_+k_y(r_{}^{\left(0\right)})k_y(r_{}^{\left(1\right)})=:s_+^{\left(2\right)}(k_yk_w)^2(y_+).$$
(11)
Combining (10) and (11) gives
$$(k_yk_w)^2(y_{})s_{}^{\left(2\right)}y_{}y_+s_+^{\left(2\right)}(k_yk_w)^2(y_+).$$
Recalling (9) and proceeding inductively gives sequences $`\{s_\pm ^{\left(r\right)}\}`$ satisfying the following for all $`j`$:
$$(k_yk_w)^{2j}(y_{})s_{}^{\left(2j\right)}y_{}y_+s_+^{\left(2j\right)}(k_yk_w)^{2j}(y_+)$$
(12)
$$(k_yk_w)^{2j1}(y_+)s_+^{\left(2j1\right)}y_{}y_+s_{}^{\left(2j1\right)}(k_yk_w)^{2j1}(y_{}).$$
(13)
Notice that
$$s_\pm ^{\left(j\right)}(k_yk_w)^{j1}(s_\pm ^{\left(1\right)})j.$$
(14)
Therefore, Assumption 4 from our theorem provides $`\overline{r}_\pm `$ such that $`s_\pm ^{\left(j\right)}\overline{r}_\pm `$ as $`j+\mathrm{}`$. Letting $`j+\mathrm{}`$ in (12) shows that $`\overline{r}_{}\overline{r}_+`$. On the other hand, letting $`j+\mathrm{}`$ in (13) gives $`\overline{r}_+\overline{r}_{}`$. Thus,
$$\overline{r}_+=\overline{r}_{}=y_+=y_{}=:\overline{y}.$$
Applying Lemma II.6 to the $`z`$-subsystem $`f=f_z`$ and the input $`u(t)=y(t)\overline{y}`$ shows that $`z(t)\overline{z}`$ for some $`\overline{z}k_z(\overline{y})`$. Since $`h_z`$ is continuous, $`w(t)`$ converges as well; i.e., $`w_+=w_{}=:\overline{w}`$. Therefore, $`\overline{w}=h_z(\overline{z})k_w(\overline{y})`$ and (7) gives $`\overline{y}=k_y(\overline{w})`$. It follows that $`\overline{w}(k_wk_y)(\overline{w})`$, so $`\overline{w}(k_wk_y)`$. Therefore, our theorem will follow once we show that $`(x(t),z(t))`$ converges to some point in $`\{k_x(\overline{w})\}\times (k_zk_y)(\overline{w})`$ as $`t+\mathrm{}`$. To this end, first note that $`x(t)k_x(\overline{w})`$ as $`t+\mathrm{}`$ as a consequence of the CICS property (namely Lemma II.5 above) applied to the $`x`$-subsystem $`f=f_x`$ and the input $`u(t)=w(t)\overline{w}`$, because we are assuming that $`k_x`$ is singleton-valued. Since $`\overline{z}k_z(\overline{y})=k_z(k_y(\overline{w}))`$, this completes the proof of the theorem.
###### Remark 3
One can extend our theorem to cases where $`k_x`$ and $`k_z`$ are both multi-valued. For example, our theorem remains true if we replace its Assumption 3 by:
* The respective i/s characteristics $`k_x`$ and $`k_z`$ exist and are locally bounded.
In this case the conclusion of the theorem is that our interconnection (2) has the pointwise globally attractive set $`\{k_x(\overline{w})\times (k_zk_y)(\overline{w}):\overline{w}(k_wk_y)\}`$. The proof of this alternative formulation is similar to the proof we gave above and proceeds by a repeated application of
$$\underset{k_pk_y(p)}{\mathrm{min}}\underset{k_qk_y(q)}{\mathrm{max}}(k_pk_q)(pq)\mathrm{\hspace{0.33em}0}p,q𝒰_x.$$
(15)
Condition (15) follows because $`h_x`$ is monotone and $`k_x`$ is weakly non-decreasing. We leave the details of the proof of this more general version of our theorem to the reader.
## V Illustration
We next illustrate our theorem using the interconnection
$$\begin{array}{cc}\dot{x}=x+5+w,\hfill & y=x\hfill \\ \dot{z}=P(z)+y,\hfill & w=\frac{1}{1+z^2}\hfill \end{array}$$
(16)
evolving on $`[0,\mathrm{})\times [0,\mathrm{})`$, where $`P(z)=z(2z^29z+12)`$. We order $`x`$ and $`z`$ by the usual cone $`[0,\mathrm{})`$. This dynamic satisfies Conditions 1-2 from Theorem 1. Replacing $`w`$ with $`\frac{1}{1+w^2}`$ in (16) gives the planar positive feedback system
$$\begin{array}{cc}\dot{x}=x+5+\frac{1}{1+w^2},\hfill & y=x\hfill \\ \dot{z}=P(z)+y,\hfill & w=z.\hfill \end{array}$$
(17)
If we use superscripts o to label the characteristics of our original interconnection (16), and if we use $`k_x`$ and so on to denote the characteristics of (17), then $`k_x^o(\frac{1}{1+w^2})k_x(w)`$ and $`k_z^ok_z`$. Also, if $`u_{k+1}(k_w^ok_y^o)(u_k)`$ with $`u_k>0`$ for all $`k`$, then $`w_{k+1}(k_wk_y)(w_k)`$ for all $`k`$ when the $`w_k`$’s are chosen to satisfy
$$\frac{1}{1+w_k^2}=u_k$$
for all $`k`$. Moreover, since the output $`w`$ in (16) is always positive, $`(k_w^ok_y^o)(0)(0,\mathrm{})`$, so $`u_k>0`$ for all $`k1`$ along all solution sequences $`\{u_k\}`$ of $`u_{k+1}(k_w^ok_y^o)(u_k)`$. Therefore, if each solution sequence $`\{w_k\}`$ for $`w_{k+1}(k_wk_y)(w_k)`$ converges, then each solution sequence $`\{u_k\}`$ for $`u_{k+1}(k_w^ok_y^o)(u_k)`$ converges as well, which implies the required convergence of solutions of $`v_{k+1}(k_y^ok_w^o)(v_k)`$ by Remark 2. The fact that Condition 3 will also hold for the original interconnection (16) will then follow because (16) has the same trajectories as (17).
It therefore remains to show that (17) satisfies Condition 3 from our theorem, that all its trajectories are bounded, and that each solution of $`w_{k+1}(k_wk_y)(w_k)`$ converges. To this end, first note that since the outputs of both subsystems in (17) are also their states, i/s and i/o characteristics coincide for (17)-if they exist–so we can define
$$k_1=k_x=k_y,k_2=k_z=k_w$$
wherever the characteristics exist. The characteristic of the first subsystem in (17) is the singleton-valued function
$$k_1(w)=5+\frac{1}{1+w^2},w_+,$$
while the characteristic for the second subsystem is multi-valued and only determined implicitly as follows: $`k_2(y)=\{z:P(z)=y\}`$ for $`y_+`$. A bifurcation analysis of the scalar system $`\dot{z}=P(z)+y`$, treating $`y_+`$ as a bifurcation parameter, shows that $`k_2(y)`$ is a characteristic which is
1. single-valued if $`y[0,4)`$ or if $`y(5,\mathrm{})`$.
2. triple-valued if $`y(4,5)`$.
3. double-valued if $`y=4`$ or $`5`$: $`k_2(4)=\{1/2,2\}\text{ and }k_2(5)=\{1,5/2\}.`$
There are two saddle-node bifurcations, one at $`y=4`$ and the other at $`y=5`$. The four defining properties of a characteristic (see Definition II.2) can indeed be readily verified: For each $`y_+`$, the system $`\dot{z}=P(z)+y`$ has a finite number of isolated compact equilibria and no cycles (since the system is scalar), and every solution converges to one of the equilibria. It is also not hard to see that $`k_2`$ is locally bounded. In order to apply Theorem 1, we only need to verify that (17) satisfies Condition $`4`$ of our theorem.
To check that the trajectories of (17) (or equivalently of (16)) are bounded, it suffices to verify the following: Claim (G): If $`(x(t),z(t))`$ is any trajectory of (16) defined on some interval $`[0,T]`$, then there is a compact set $`D`$ depending only on $`(x(0),z(0))`$ (and not on $`T`$) such that $`(x(t),z(t))D`$ for all $`t[0,T]`$. Boundedness will follow from $`(G)`$ by standard results for extendability of solutions of ODE’s. To prove $`(G)`$, first note that the boundedness of $`w`$ on $`[0,T]`$ and the variations of parameters formula gives
$$|y(t)|=|x(t)||x(0)|+6$$
for all $`t[0,T]`$. Pick $`\stackrel{~}{z}>5/2`$ such that $`\stackrel{~}{z}=P^1(|x(0)|+6)`$ which exists because $`P`$ is one-to-one above $`5/2`$. It follows that if $`t[0,T)`$ is such that $`z(t)>\stackrel{~}{z}`$, then
$$(z(t))P(\stackrel{~}{z})=|x(0)|+6y(t),$$
so $`\dot{z}(t)0`$. Therefore, $`z(t)`$ stays below $`\stackrel{~}{z}`$ on $`[0,T]`$. Since $`\stackrel{~}{z}`$ depends only on $`x(0)`$, Claim (G) follows.
Next consider the discrete inclusion $`w_{k+1}\left(k_2k_1\right)(w_k)`$ and notice that it reduces to a discrete equation $`w_{k+1}=\left(k_2k_1\right)(w_k)`$ because $`k_1(w)>5`$ and $`k_2(y)`$ is single-valued when $`y>5`$. Notice also that for all $`w_0_+`$, the discrete equation gives $`w_k>5/2`$ for all $`k1`$. In particular, the interval $`(5/2,\mathrm{})`$ is forward invariant for the discrete equation. Finally, since $`|k_1^{}(w)|`$ is decreasing for $`w5/2`$, elementary calculus shows that
$$\left|k_2^{}(k_1(w))k_1^{}(w)\right|\frac{|k_1^{}(5/2)|}{P^{}(k_2k_1(w))}\frac{|k_1^{}(5/2)|}{P^{}(5/2)}=\frac{5}{(1+25/4)^2}\frac{2}{9}<\mathrm{\hspace{0.33em}1}w5/2,$$
so $`k_2k_1`$ is a contraction mapping on $`[5/2,\mathrm{})`$, hence the discrete equation has a unique globally attractive fixed point $`\overline{w}`$. Therefore, we know from Remark 2 that (17) satisfies Conditions 3-4 of our theorem, as claimed. Since
$$(k_w^ok_y^o)=\{\frac{1}{1+\overline{w}^2}:\overline{w}(k_2k_1)\},$$
we conclude that our original interconnection (16) has the unique globally attractive equilibrium
$$\left\{(5+\frac{1}{1+\overline{w}^2},k_2\left(5+\frac{1}{1+\overline{w}^2}\right))\right\}.$$
Figure 2 below illustrates this.
###### Remark 4
In the preceding example, the inclusion $`w_{k+1}(k_wk_y)(w_k)`$ had a unique equilibrium, but our theory applies to examples where $`(k_wk_y)`$ has more than one element as well. One such example is constructed by modifying the interconnection (17) in the following way: replace the $`x`$-subsystem with $`\dot{x}=x+R(w)`$ where $`R`$ consists of the line segments in the $`wy`$-plane joining $`(0,5)`$ to $`(.5,4.5)`$, $`(.5,4.5)`$ to $`(2.5,4.5)`$, and $`(2.5,4.5)`$ to $`(3.5,3)`$. With this change we get $`\mathrm{Card}\{(k_wk_y)\}=3`$, and the conclusion of our theorem remains true because $`\{k_y(w_k)\}`$ converges for each solution sequence $`\{w_k\}`$ of $`w_{k+1}(k_wk_y)(w_k)`$; see Remark 2. In fact, if $`w_o[.5,2.5]`$, then $`k_y(w_o)=4.5`$, so $`w_k(k_wk_y)`$ for all $`k`$, which gives $`k_y(w_k)=4.5`$ for all $`k`$. If $`w_o[0,.5]`$, then $`k_y(w_o)[4.5,5]`$, which gives $`w_1[.5,2.5]`$, so $`k_y(w_k)4.5`$ for all $`k2`$ as before. Finally, if $`w_o>5/2`$, then $`k_y(w_o)4.5`$, so $`w_1k_wk_y(w_o)[0,5/2]`$, so $`k_y(w_k)=4.5`$ for $`k3`$, by the previous two cases. On the other hand, one can find non-periodic divergent solution sequences of $`w_{k+1}(k_wk_y)(w_k)`$ when $`w_o[1/2,5/2]`$. The detailed analysis of this more complicated example is similar to the analysis of (17) and is left to the reader. Note that the convergence of the iterations in the preceding remark follows because $`R(w)`$ is a horizontal line, at least locally where it meets the other characteristic.
## VI Conclusion
We presented a new small-gain theorem for interconnections of monotone i/o systems with set-valued i/s characteristics. This corresponds to situations where the trajectory for a given constant input can converge to several possible equilibria, depending on the initial value for the trajectory. A key ingredient in the proof of our small-gain theorem is the theory of asymptotically autonomous systems, which requires in particular that the equilibria of the subsystems in the interconnection contain no chains. This suggests the question of how one might extend our theory to cases where the sets of equilibria of the subsystems are more general, e.g., where they contain chains or limit cycles. Research on this question is ongoing.
|
warning/0506/cond-mat0506365.html
|
ar5iv
|
text
|
# Scale invariant forces in 1𝑑 shuffled lattices
## I Introduction
The knowledge of the statistical properties of the force acting on a particle belonging to a gas and exerted by all the other particles provides important information in many physical contexts and applications. Typical examples are: (i) distribution of the gravitational force in a gas of masses in cosmological and stellar astrophysical applications chandra ; star ; fractal , (ii) distribution of molecular and dipolar interactions dipole in gas of particles, (iii) theory of defects in condensed matter physics dislo , and (iv) granular materials granular1 ; granular2 . The first seminal work in this field was due to Chandrasekhar chandra and deals, among many subjects, with the gravitational force probability distribution in a homogeneous Poisson spatial distribution of identical particles. By studying the characteristic function of the sum of the stochastic forces due to the single particles, the probability distribution of this total force is exactly found to be given by the so-called Holtzmark distribution, which is a three-dimensional analog of the one-dimensional fat tailed stable Lévy distributions. In star ; gauss-poisson ; dipole ; libro approximated extensions, to different branches of physics, of this approach can be found for more complex particle distribution (i.e., point processes) obtained by perturbing a homogeneous Poisson point process. In this paper we present a study of the total force probability distribution for a very different class of spatial particle distributions (i.e., point processes), the perturbed lattices of point particles, in the case in which the pair interaction decays spatially as a general power law. We think that this study can be very useful for application in both solid state physics (e.g., in the case of Coulomb or dipolar pair interaction) dipole , and cosmology where $`n`$-body gravitational simulations (introduced to study the problem of “structure formation” due to gravitational collapse from primordial cosmological mass density fluctuations) are performed usually starting from suitable perturbed lattice initial conditions n-body . We limit the study to the one-dimensional case in order to avoid difficulties related to the anisotropies of higher dimensional lattices. However the exact results we present in $`1d`$ are suggestive of the behavior of the same quantities in higher dimensions. In fact one can see 3d-ours that the change of spatial dimension only renders calculation not explicitly performable, keeping qualitatively the behavior we present below.
## II Definitions and formalism
In order to approach in the proper way the problem of the global force probability distribution in a perturbed lattice of particles interacting via a power law spatially decreasing pair interaction, let us consider firstly a gas of such identical particles with microscopic density
$$n(x)=\underset{i}{}\delta (xx_i),$$
where $`x_i`$ is the position of the $`i^{th}`$ particle. Let us assume that the average number density $`n_0=n(x)>0`$ (where the average $`\mathrm{}`$ is to be intended as an ensemble average) is well defined (i.e., the particle distribution is uniform on sufficiently large scales libro ). We then suppose that particles interact via a pair force $`f(x)`$ depending on the mutual pair distance $`x`$ as
$$f(x)=C\frac{x}{|x|^{\alpha +1}}.$$
This means that $`f(x)`$ gives the force exerted by a particle in the origin on another particle in $`x`$ (the force is attractive if $`C>0`$, and repulsive if $`C<0`$). Therefore the force-field in the point $`x`$ of the space will be
$$(x)=C\underset{i}{}\frac{x_ix}{|x_ix|^{\alpha +1}}C𝑑yn(y)\frac{yx}{|yx|^{\alpha +1}},$$
(1)
where the last integral is over all the space. However, from the last expression of Eq. (1), being $`n_0>0`$, we have that for a given realization of the stochastic density field $`n(x)`$, the infinite volume limit of $`(x)`$ is not univocally defined for $`\alpha 1`$ (i.e. the integral in absolutely diverging, and its value depend on how this limit is taken). The same feature is present in higher spatial dimensions. For example in $`d=3`$ the same problem is present for $`\alpha 3`$ and in general $`\alpha d`$ in $`d`$ dimensions. For instance this is the case of the gravitational force in a self-gravitating homogeneous gas of identical masses chandra , and of Coulomb interaction in the one component plasma (OCP) OCP of identical electrical charges both in the disordered and the ordered (i.e., the Coulomb lattice pines ) phases. This problem is well known in condensed matter physics about the OCP. However, in this case, the problem is automatically solved by the presence in the physical system of a uniform background charge density $`n_b(x)=n_0`$ with opposite sign with respect to the identical charged particles and such that to conserve global charge neutrality in the system. Once the attractive force of the background is considered on a charged particle together with the repulsive forces exerted by the other particles, the problem of the infinite volume limit of $``$ is solved and its value unique. For what concerns the self-gravitating systems, in Newtonian gravitation an analog of the uniform background of the OCP (i.e. a negative uniform mass density $`n_b(x)=n_0`$ such to generate a repulsive force on the particles) does not exist, and it has to be introduced in the system artificially to regularize the problem (an approach usually called Jeans’ swindle binney-tremaine ). However this negative background comes out naturally, as an effect of space expansion, when the gravitational motion of particles is described, starting from the equations of general relativity, in comoving coordinates in a quasi-uniform expanding Einstein - De Sitter universe peebles80 which is the main model of universe used in cosmology. In practice considering the presence of such balancing background will give for $``$ the following expression:
$$(x)=C𝑑y\delta n(y)\frac{yx}{|yx|^{\alpha +1}},$$
(2)
where $`\delta n(x)=n(x)n_0`$. This makes $``$ to be defined also for smaller $`\alpha `$ depending on the small $`k`$ behavior of the power spectrum $`S(k)|\stackrel{~}{\delta n}(k)|^2`$ of the density field where $`\stackrel{~}{\delta n}(k)`$ is the Fourier transform of $`\delta n(x)`$. By studying the large distance scaling behavior of the integrated fluctuations of $`n(x)`$ libro , it is simple to show that, assuming $`S(k)k^\beta `$ at small $`k`$, in $`d`$ dimension $``$ is a well defined statistical quantity (i.e., its value does not depend on the way in which the infinite volume limit is taken) for $`\alpha >(d\beta )/2`$ if $`\beta <1`$, and $`\alpha >(d1)/2`$ if $`\beta 1`$ (see also the discussion in Appendix II on the definiteness of the force $``$ respectively in the shuffled lattice and the homogeneous Poisson particle distributions). Note finally that taking the infinite volume limit symmetrically with respect to the point $`x`$ on which the force is calculated the background gives a zero net force on the point $`x`$. Therefore the value of $``$ obtained calculating Eq. (1) taking the infinite volume limit symmetrically with respect to the point $`x`$ gives automatically the well defined value obtained by Eq. (2) (i.e., subtracting the effect of the background). This preliminary discussion of the statistical definiteness of the force $``$ is useful to justify the symmetrical way in which we take the infinite volume limit in the one-dimensional shuffled lattice case we analyze in the rest of the paper. The statistical properties of $``$ we will find in this peculiar way (no background and symmetrical limit) coincide with those of the case in which the effect of a negative background is considered independently of the way in which the infinite volume limit is taken, which therefore can be considered the real physical case. Moreover, from the above considerations we deduce that, as in the shuffled lattice $`S(k)k^2`$ at small $`k`$ (see displa ), the results are valid for all values $`\alpha >0`$ in $`d=1`$.
Let us take, therefore, a $`1d`$ regular chain of $`2N+1`$ unitary mass particles with a lattice spacing $`a>0`$ (we will take eventually the limit $`N+\mathrm{}`$), i.e., the position of the $`n^{th}`$ particle is $`X_n=na`$. Therefore the microscopic density can be written as
$$n_{in}(x)=\underset{n=N}{\overset{N}{}}\delta (xna).$$
Clearly the average density of particles in the system is $`n_0=1/a`$. We now apply an uncorrelated displacement field (i.e., a random shuffling) to this system, i.e., a random displacement $`U_n`$ is applied to the generic $`n^{th}`$ particle independently of the other particles. This displacement field is completely characterized by the one-displacement probability density function (PDF) $`p(u)`$ (i.e., $`Prob(uU_n<u+du)=p(u)du`$). After the application of the displacements the new microscopic density will be:
$$n(x)=\underset{n=N}{\overset{N}{}}\delta (xnau_n),$$
(3)
the $`u_n`$’s being the realizations of the random variables $`U_n`$ all extracted from $`p(u)`$ independently one of each other. We consider the case $`p(u)=p(u)`$ for simplicity. Equation (3) says that the particle originally in $`X_n=na`$, after the displacement will be in $`X_n=na+U_n`$. For the analysis of spatial density correlations in such a system see displa . Let us call $`q_n(x)`$ the PDF of the position of the $`n^{th}`$ particle (i.e., $`Prob(xx_n<x+dx)=q_n(x)dx`$). Clearly it is given by
$$q_n(x)p(xna).$$
Let us now assume, as above, that the $`n^{th}`$ particle creates a force-field in the point $`x`$ of the type:
$$f_n(x)=C\frac{X_nx}{|X_nx|^{\alpha +1}}$$
with $`\alpha >0`$ and $`C`$ a constant. Therefore the total stochastic field $``$ generated at a generic point $`x`$ of the space by all the system particles is:
$$(x)=C\underset{n=N}{\overset{N}{}}\frac{X_nx}{|X_nx|^{\alpha +1}}.$$
(4)
Note that it is a sum of random variables. Let us call $`W_0(F)`$ its PDF; it will be given by
$`W_0(F)=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}\left[\underset{n=N}{\overset{N}{}}dx_np(x_nna)\right]}`$
$`\times \delta \left(FC{\displaystyle \underset{n0}{\overset{N,N}{}}}{\displaystyle \frac{x_nx}{|x_nx|^{\alpha +1}}}\right).`$
It is immediate to see that $`(x)`$ is the sum of independent random variables $`C(X_nx)/|X_nx|^{\alpha +1}`$. However, as the PDF’s $`q_n(x)`$ change with $`n`$, these variable are not identically distributed. As shown below, this, together with the fact that $``$ needs not a normalization in $`N`$ to be well defined in the large $`N`$ limit, are the reasons why we do not obtain in general an exact Gaussian or Lévy limit levy-stable ; feller for the $`W_0(F)`$. In order to study the asymptotic behavior in $`F`$ and $`N`$, it is usual to introduce the so-called characteristic function of $``$, i.e., the Fourier transform (FT) of $`W_0(F)`$:
$`\widehat{W}_0(k){\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑FW_0(F)e^{ikF}`$
$`={\displaystyle \underset{n=N}{\overset{N}{}}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑yp(yna)\mathrm{exp}\left(iCk{\displaystyle \frac{yx}{|yx|^{\alpha +1}}}\right).`$ (5)
By studying the small $`k`$ behavior of the single integrals in Eq. (5) and taking appropriately the limit $`N\mathrm{}`$ we can deduce the moments and the large $`F`$ behavior of $`W_0(F)`$.
However, we will study a slightly different and more difficult problem, which is of particular interest if we want to study the dynamics of the system particles under the effect of only this mutual force. We study directly the statistical properties of the stochastic force acting on one generic system particle. In particular we calculate the total force acting on the particle initially located at the origin of the space and displaced to $`X_0=U_0`$:
$$=\underset{n0}{\overset{N,N}{}}f_n(X_0)=C\underset{n0}{\overset{N,N}{}}\frac{X_nX_0}{|X_nX_0|^{\alpha +1}}.$$
(6)
In this way, taking the limit $`N\mathrm{}`$, we get the symmetric infinite volume limit with no negative background of Eq.(1) with $`x=X_0`$, which is, as explained in the first paragraph of this section, equal to Eq.(2) with arbitrary way of taking the infinite volume limit, and therefore giving the force on any particle belonging to the system with the uniform balancing background. Note that now, because of the presence of the variable $`X_0`$ in each term of the sum (6), $``$ is no more a sum of independent terms. However we show below how to reduce the problem to that of a sum of independent stochastic terms, by introducing the concept of conditional probability density function. The solution of this conditional problem will give also the way to face the study of the first unconditional case given by Eq. (4).
About $``$ in Eq. (6), we want again to find the PDF $`W(F)`$ of the value $`F`$ of this force. As before, since the displacements applied to the particles are independent one of each other, we have the exact relation:
$`W(F)={\displaystyle _{\mathrm{}}^+\mathrm{}\left[\underset{n=N}{\overset{N}{}}dx_np(x_nna)\right]}`$
$`\times \delta \left(FC{\displaystyle \underset{n0}{\overset{N,N}{}}}{\displaystyle \frac{x_nx_0}{|x_nx_0|^{\alpha +1}}}\right),`$
that, through a simple change of variables $`\mathrm{\Delta }_n=x_nx_0`$ or $`n0`$, can be rewritten as
$`W(F)={\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑x_0p(x_0)`$
$`{\displaystyle _{\mathrm{}}^+\mathrm{}\left[\underset{n0}{\overset{N,N}{}}d\mathrm{\Delta }_np(\mathrm{\Delta }_n+x_0na)\right]}`$
$`\times \delta \left(FC{\displaystyle \underset{n0}{\overset{N,N}{}}}{\displaystyle \frac{\mathrm{\Delta }_n}{|\mathrm{\Delta }_n|^{\alpha +1}}}\right).`$
Let us analyze the behavior of the conditional PDF $`P(F;x_0)`$, conditioned to the fact that the particle on which the force is evaluated is at $`X_0=x_0`$:
$`P(F;x_0)={\displaystyle _{\mathrm{}}^+\mathrm{}\left[\underset{n0}{\overset{N,N}{}}d\mathrm{\Delta }_np(\mathrm{\Delta }_n+x_0na)\right]}`$
$`\times \delta \left(FC{\displaystyle \underset{n0}{\overset{N,N}{}}}{\displaystyle \frac{\mathrm{\Delta }_n}{|\mathrm{\Delta }_n|^{\alpha +1}}}\right).`$ (7)
In this way, once $`x_0`$ is fixed, the total force $``$ is the sum of independent contributions
$$f_n=C\frac{\mathrm{\Delta }_n}{|\mathrm{\Delta }_n|^{\alpha +1}}.$$
(8)
It is interesting to see in which cases $``$ satisfies the central limit theorem. We will see that it never satisfies this theorem even when its PDF is rapidly decreasing at large values. More precisely, we will see that even in this last case its PDF is dependent on the details of $`p(u)`$.
For the sake of simplicity of notation let us assume in the rest of the paper that $`C=1`$ (the repulsive case $`C=1`$ will be trivially deduced from this). By performing the simple change of variables given by Eq. (8), it is possible to find the conditional (i.e., conditioned to $`X_0=x_0`$) PDF $`g_n(f;x_0)`$ of the single stochastic force generated by the particle in $`X_n`$ on the particle fixed in $`X_0=x_0`$:
$$g_n(f;x_0)=\frac{|f|^{11/\alpha }}{\alpha }p\left(\frac{f}{|f|^{1+1/\alpha }}+x_0na\right).$$
(9)
Clearly also the forces $`f_n`$ so distributed are independent of one each other. The support of $`g_n(f;x_0)`$ can be simply deduced from the one of $`p(u)`$.
## III Large $`F`$ analysis and limit theorems
Let us take the FT of Eq. (7) in order to evaluate the conditional characteristic function of the stochastic variable $``$:
$`\widehat{P}(k;x_0){\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑FP(F;x_0)e^{ikF}`$ (10)
$`={\displaystyle \underset{n0}{\overset{N,N}{}}}{\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑xp(x+x_0na)\mathrm{exp}\left(ik{\displaystyle \frac{x}{|x|^{\alpha +1}}}\right).`$
Note that the quantity
$`{\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑xp(x+x_0na)\mathrm{exp}\left(ik{\displaystyle \frac{x}{|x|^{\alpha +1}}}\right)`$ (11)
$`={\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑fg_n(f;x_0)e^{ikf}\{e^{ikf}\}_{n;x_0},`$
where $`g_n(f;x_0)`$, given by Eq. (9), is the conditional PDF of the field $`f`$ felt by the particle in $`x_0`$ due to the particle in $`x_n`$, is the conditional characteristic function of this field $`f`$. Moreover $`\{a(f)\}_{n;x_0}=_{\mathrm{}}^+\mathrm{}𝑑fg_n(f;x_0)a(f)`$. It is important to note that, if $`x_0=0`$ (i.e., the particle on which we calculate the force is stuck at the origin), the condition $`p(u)=p(u)`$ on the displacements of any particle would imply
$`{\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑xp(x+na)\mathrm{exp}\left(ik{\displaystyle \frac{x}{|x|^{\alpha +1}}}\right)`$
$`={\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑xp(xna)\mathrm{exp}\left(ik{\displaystyle \frac{x}{|x|^{\alpha +1}}}\right).`$
Consequently, if one fixes $`x_0=0`$ Eq. (10) becomes:
$$\widehat{W}(k)=\underset{n=1}{\overset{N}{}}\left|_{\mathrm{}}^+\mathrm{}𝑑xp(xna)\mathrm{exp}\left(ik\frac{x}{|x|^{\alpha +1}}\right)\right|^2.$$
However the shift $`x_00`$ of the particle initially at the origin, and on which we calculate the force, breaks this symmetry, which is anyway recovered when the average over $`p(x_0)`$ is performed to proceed from $`P(F;x_0)`$ to $`W(F)`$ (however we will see that this is a further source of noise when we calculate explicitly the variance of the force $``$).
In order to proceed into the analysis of the PDF of $``$, we have to distinguish two basically different cases:
1. Non-Overlapping condition (NOC): No particle can be found arbitrarily close to any other particle; i.e., the supports<sup>1</sup><sup>1</sup>1We call support of $`p(u)`$ simply the set of real values of $`u`$ such that $`p(u)>0`$. respectively of $`p(u)`$ and of $`p(una)`$, for all integer $`n0`$, have an empty overlap. The main case of physical interest in this class of displacement fields is when $`0<u_0<a/2`$ such that $`p(u)=0`$ for $`|u|>u_0`$;
2. Overlapping condition (OC): Particles can cross one each other and at least one pair of particles can be found arbitrarily close to one each other; i.e., the supports respectively of $`p(u)`$ and of $`p(una)`$, for at least an integer $`n0`$, have a non-zero overlap. The main case of physical interest in which this happens is when $`ϵ>0`$ such that $`p(u)>0`$ for all $`|u|<a/2+ϵ`$.
We will see that in the first case we obtain a rapidly decreasing $`W(F)`$ even though there is no constraint toward Gaussianity in the large $`N`$ limit, while in the second case we have a power law tailed $`W(F)`$ similarly to that of the three-dimensional Holtzmark distribution chandra .
## IV Detailed analysis of Eq. (10)
Let us analyze the single factor of Eq. (10) which, as aforementioned, is the FT of the conditional PDF $`g_n(f;x_0)`$ of the force felt by the particle in $`X_0=x_0`$ due to only the particle in $`X_n`$:
$`{\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑xp(x+x_0na)\mathrm{exp}\left(ik{\displaystyle \frac{x}{|x|^{\alpha +1}}}\right)`$
$`=\mathrm{exp}\left(ik{\displaystyle \frac{x}{|x|^{\alpha +1}}}\right)_{n,x_0},`$ (12)
where $`s(x)_{n,x_0}`$ denotes the average of the function $`s(x)`$ over the shifted PDF $`h_{n,x_0}(x)=p(x+x_0na)`$. In practice, if we indicate with simply $`s(u)=_{\mathrm{}}^+\mathrm{}𝑑up(u)s(u)`$, then we can say that
$$s(x)_{n,x_0}=s(u+nax_0).$$
We want to study Eq. (11) in the limit of small $`k`$. Similarly to what pointed out in the previous section, the small $`k`$ behavior of $`\mathrm{exp}\left(ik\frac{x}{|x|^{\alpha +1}}\right)_{n,x_0}`$ is different in the two cases in which, as a consequence of displacements, the pair of particles initially in $`x=0`$ and $`x=na`$ cannot or can be found arbitrarily close to one each other, i.e., respectively if the supports of $`p(u)`$ and $`p(una)`$ have an empty or a non-zero overlap.
Let us start with the case (i). If $`\mathrm{\hspace{0.17em}0}<u^{}<|n|a/2`$ such that $`p(u)=0`$ for $`|u|u^{}`$, the exponent of $`\mathrm{exp}\left(ik\frac{x}{|x|^{\alpha +1}}\right)`$ can take only limited values in the integral (12) (i.e., the support of $`g_n(f;x_0)`$ is restricted to only a finite interval of values of $`f`$). Note that in the given hypothesis (i), if $`n>0`$ the quantity $`x`$ can take only strictly positive values, while if $`n<0`$ it takes only strictly negative values. In this case, if $`n>0`$, we can write:
$`\mathrm{exp}\left(ik{\displaystyle \frac{x}{|x|^{\alpha +1}}}\right)_{n,x_0}`$
$`={\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑xp(x+x_0na){\displaystyle \underset{m=0}{\overset{+\mathrm{}}{}}}{\displaystyle \frac{(ikx^\alpha )^m}{m!}}`$
$`={\displaystyle \underset{m=0}{\overset{+\mathrm{}}{}}}{\displaystyle \frac{(ik)^m}{m!}}x^{\alpha m}_{n,x_0}`$
$`={\displaystyle \underset{m=0}{\overset{+\mathrm{}}{}}}{\displaystyle \frac{(ik)^m}{m!}}(u+nax_0)^{\alpha m},`$ (13)
where
$$(u+nax_0)^{\alpha m}=_u^{}^u^{}𝑑up(u)(ux_0+na)^{\alpha m}.$$
If instead $`n<0`$, Eq. (13) becomes
$`\mathrm{exp}\left(ik{\displaystyle \frac{x}{|x|^{\alpha +1}}}\right)_{n,x_0}={\displaystyle \underset{m=0}{\overset{+\mathrm{}}{}}}{\displaystyle \frac{(ik)^m}{m!}}(x)^{\alpha m}_{n,x_0}`$
$`={\displaystyle \underset{m=0}{\overset{+\mathrm{}}{}}}{\displaystyle \frac{(ik)^m}{m!}}(una+x_0)^{\alpha m},`$ (14)
Note that, as $`p(u)=p(u)`$, we have $`(una+x_0)^{\alpha m}=(una+x_0)^{\alpha m}`$. In both cases
$$\left|\left(\frac{x}{|x|^{\alpha +1}}\right)^m_{n,x_0}\right|<(|n|a+2u^{})^{\alpha m}$$
for any $`m0`$, and therefore the series in Eq. (13) absolutely converges. It is very important to note that, if $`u^{}<a/2`$ (i.e., no pair of particles can be found arbitrarily close to one each other) all the factors in Eq. (10) can be represented as a Taylor series (13) to all orders $`m`$. As shown below in more detail, this implies that when $`u^{}<a/2`$, $`W(F)`$ has all finite moments and therefore is rapidly decreasing at large $`F`$.
In the second case (ii) instead $`0<ϵ<a/2`$ such that, at least $`u`$ satisfying $`|n|a/2ϵ<|u|<|n|a/2+ϵ`$, we have $`p(u)>0`$. In this case the particle initially at $`\pm na`$ and the particle initially at the origin can be found arbitrarily close. This implies that the quantity $`x`$ (i.e., $`ux_0+na`$) in (12) is permitted to take arbitrary small values up to zero, and therefore in the last expression of Eq. (13) there would be an infinite number of diverging terms of the last series. In other words the Taylor series sum in the second expression of Eq. (13) cannot be exchanged with the average operation $`\mathrm{}_{n,x_0}`$, and we expect a singular part in the small $`k`$ expansion of the average $`\mathrm{exp}\left(ik\frac{x}{|x|^{\alpha +1}}\right)_{n,x_0}`$. In order to find it, we rewrite it as in Eq. (11):
$`\mathrm{exp}\left(ik{\displaystyle \frac{x}{|x|^{\alpha +1}}}\right)_{n;x_0}{\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑fg_n(f;x_0)e^{ikf}`$
$`={\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑xp(x+x_0na)\mathrm{exp}\left(ik{\displaystyle \frac{x}{|x|^{\alpha +1}}}\right)`$
Note that in this case, differently from the previous one, the support of $`g_n(f;x_0)`$ includes arbitrarily large values of $`|f|`$ for which, using Eq. (9), we have
$$g_n(f;x_0)=\frac{p(x_0na)}{\alpha }|f|^{(\alpha +1)/\alpha }+o(|f|^{(\alpha +1)/\alpha }).$$
Let us call $`M=[\alpha ^1]`$ the integer part of $`\alpha ^1`$. By using the results presented in Appendix I we can conclude that
$`\mathrm{exp}\left(ik{\displaystyle \frac{x}{|x|^{\alpha +1}}}\right)_{n;x_0}`$ (15)
$`={\displaystyle \underset{m=0}{\overset{M}{}}}{\displaystyle \frac{(ik)^m}{m!}}(u+nax_0)^{\alpha m}+S_n(k;x_0),`$
where $`S_n(k;x_0)`$ contains all the terms of order higher than $`M`$, including the singular part of the small $`k`$ expansion of $`\mathrm{exp}\left(ik\frac{x}{|x|^{\alpha +1}}\right)_{n;x_0}`$ which is of order $`1/\alpha `$ in $`k`$. By using Eq. (36), we can finally write:
$`S_n(k;x_0)=`$ (16)
$`\{\begin{array}{cc}\frac{(1)^{(M+1)/2}\pi p(x_0na)}{\alpha \mathrm{\Gamma }[(\alpha +1)/\alpha ]\mathrm{cos}\left(\frac{\alpha ^1M}{2}\pi \right)}k^{1/\alpha }+o(k^{1/\alpha })\hfill & \text{for odd }M\hfill \\ & \\ \frac{(1)^{M/2}\pi p(x_0na)}{\alpha \mathrm{\Gamma }[(\alpha +1)/\alpha ]\mathrm{sin}\left(\frac{\alpha ^1M}{2}\pi \right)}k^{1/\alpha }+o(k^{1/\alpha })\hfill & \text{for even }M\hfill \end{array}`$ (20)
Here, for simplicity, we have excluded the case in which exactly $`M=1/\alpha `$ for which we have logarithmic corrections in $`k`$ to the above equations.
## V Finding $`W(F)`$
At this point we can go further and classify the possible behaviors of $`P(F;x_0)`$ and $`W(F)`$.
Basically we again distinguish the following two cases:
1. $`ϵ>0`$ such that $`|u|>a/2ϵ`$ one has $`p(u)=0`$;
2. $`0<ϵ<a/2`$ such that, at least $`u`$ satisfying $`a/2ϵ<|u|a/2+ϵ`$, $`p(u)>0`$.
### V.1 Case 1: fast decreasing $`W(F)`$
In this case the system satisfies the NOC and all the factors in Eq. (10) can be expanded in the Taylor series (13) and (14) for all different $`n`$.
We can then write
$`\widehat{P}(k;x)`$ $``$ $`{\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑FP(F;x)e^{ikF}`$
$`=`$ $`{\displaystyle \underset{n=1}{\overset{N}{}}}e^{i\frac{k}{(ux+na)^\alpha }}e^{i\frac{k}{(u+x+na)^\alpha }}`$
$`=`$ $`{\displaystyle \underset{n=1}{\overset{N}{}}}[{\displaystyle \underset{m=0}{\overset{+\mathrm{}}{}}}{\displaystyle \frac{(ik)^m}{m!}}(ux+na)^{\alpha m}`$
$`{\displaystyle \underset{l=0}{\overset{+\mathrm{}}{}}}{\displaystyle \frac{(ik)^l}{l!}}(u+x+na)^{\alpha l}],`$
where in the average $`\mathrm{}`$ over the displacement $`u`$ we have used the symmetry property $`p(u)=p(u)`$. Note that from Eq. (V.1), we have $`\widehat{P}(k;x)=\widehat{P}^{}(k;x)`$, where $`A^{}`$ indicates the complex conjugate of $`A`$.
By calling again $`u^{}<a/2`$ the maximal permitted displacement for each particle (i.e., the support of $`p(u)`$ is included in $`[u^{},u^{}]`$), we can find $`\widehat{W}(k)=[W(F)]`$ by simply calculating the following average
$$\widehat{W}(k)=_u^{}^u^{}𝑑xp(x)\widehat{P}(k;x).$$
(22)
It is simple to verify that, as $`\widehat{P}(k;x)=\widehat{P}^{}(k;x)`$ and $`p(u)=p(u)`$, the function $`\widehat{W}(k)`$ is real and $`\widehat{W}(k)=\widehat{W}(k)`$. The Taylor expansion in $`k`$ of $`\widehat{W}(k)`$ is obtainable from Eqs.(V.1) and (22). Since it is a real function and is the characteristic function only even powers of $`k`$ are present. In particular the coefficient of the $`k^2`$ term is $`\overline{^2}/2`$ where
$$\overline{h()}=_{\mathrm{}}^+\mathrm{}𝑑FW(F)h(F).$$
Actually, rigorously speaking, we should show that all the coefficients of the Taylor expansion of $`\widehat{W}(k)`$ are convergent to finite values in the limit $`N\mathrm{}`$. It is simple to show it by expanding the terms $`(u\pm x_0+na)^{\alpha m}`$ of Eq. (V.1) in Taylor series of $`(u\pm x_0)/na`$ for $`n1`$ which is justified by the fact that in the given hypothesis $`|u|+|x_0|2u^{}<a`$, and considering that
$$\underset{m=0}{\overset{+\mathrm{}}{}}\frac{(ik)^m}{m!}(na)^{\alpha m}\times \underset{l=0}{\overset{+\mathrm{}}{}}\frac{(ik)^l}{l!}(na)^{\alpha l}=1,n1.$$
Therefore we conclude that we can write $`\widehat{W}(k)`$ in the following form:
$$\widehat{W}(k)=\underset{n=0}{\overset{+\mathrm{}}{}}(1)^n\frac{\overline{^{2n}}}{(2n)!}k^{2n}.$$
It is simple to see that
$`{\displaystyle \frac{\overline{^2}}{2}}={\displaystyle \underset{n=1}{\overset{N}{}}}\{(ux+na)^{2\alpha }_u_x`$ (23)
$`(ux+na)^\alpha _u(u+x+na)^\alpha _u_x\}`$
$`+{\displaystyle \underset{n<l}{\overset{1,N}{}}}[(ux+na)^\alpha _u(u+x+na)^\alpha _u]`$
$`\times [(ux+la)^\alpha _u(u+x+la)^\alpha _u]_x,`$
where for clarity we have redefined
$$\{\begin{array}{c}a(u)_u=_u^{}^u^{}𝑑ua(u)p(u)\hfill \\ a(x)_x=_u^{}^u^{}𝑑xa(x)p(x)\hfill \\ b(u,x)_u_x=_u^{}^u^{}_u^{}^u^{}𝑑u𝑑xb(u,x)p(u)p(x).\hfill \end{array}$$
It is matter of simple algebra to show that, for $`p(u)=p(u)`$, Eq. (23) can be rewritten as
$`{\displaystyle \frac{\overline{^2}}{2}}={\displaystyle \underset{n=1}{\overset{N}{}}}(ux+na)^{2\alpha }_u(ux+na)^\alpha _u^2_x`$
$`+{\displaystyle \underset{n,l}{\overset{1,N}{}}}(ux+na)^\alpha _u[(ux+la)^\alpha _u`$
$`(u+x+la)^\alpha _u]_x`$ (24)
We see that the force variance is composed of two different contributions: the former, given by the first sum in Eq. (24), is mainly due to the fluctuations in the displacements $`u`$ of all the sources of the force (in this term the average over $`x`$ is only a smoothing operation), while the latter, given by the second sum, is determined basically by the fluctuations created by the stochastic displacement $`x`$ of the particle initially in the origin on which we evaluate the force (in this term is the averages over $`u`$ to play a role of simple smoothing).
It is interesting and useful in applications to calculate all the above expressions by evaluating all the terms in the sums in Eq. (24) to the second order in $`(u\pm x)/na`$. In order to do this, we use the following second order Taylor expansion for $`BA`$:
$`(A+B)^\gamma =A^\gamma \left(1+{\displaystyle \frac{B}{A}}\right)^\gamma `$
$`=A^\gamma \left[1\gamma {\displaystyle \frac{B}{A}}+{\displaystyle \frac{\gamma (\gamma +1)}{2}}\left({\displaystyle \frac{B}{A}}\right)^2+o\left({\displaystyle \frac{B}{A}}\right)^2\right].`$
From this, substituting respectively $`A`$ with $`na`$ and $`B`$ with $`u\pm x`$, we have that
$`(u\pm x+na)^\gamma `$
$`(na)^\gamma \left[1\gamma {\displaystyle \frac{u\pm x}{na}}+{\displaystyle \frac{\gamma (\gamma +1)}{2}}\left({\displaystyle \frac{u\pm x}{na}}\right)^2\right].`$
Moreover we have that $`(u\pm x)^{2n+1}_u_x=0`$ for any integer $`n`$ due to the symmetry $`p(u)=p(u)`$, while we have that $`(u\pm x)^2_u_x=2\sigma ^2`$ where $`\sigma ^2=u^2_u=_u^{}^u^{}𝑑uu^2p(u)`$ is the variance of the single displacement. Therefore we can write
$$(u\pm x+na)^\gamma _u_x(na)^\gamma \left[1+\gamma (\gamma +1)\frac{\sigma ^2}{(na)^2}\right],$$
and
$$(u\pm x+na)^\alpha _u^2_x(na)^{2\alpha }\left[1+\alpha (3\alpha +2)\frac{\sigma ^2}{(na)^2}\right].$$
Henceforth
$`(u\pm x+na)^{2\alpha }_u(u\pm x+na)^\alpha _u^2_x`$
$`={\displaystyle \frac{\alpha ^2\sigma ^2}{(na)^{2(\alpha +1)}}}.`$
Moreover
$`(ux+na)^\alpha _u(u\pm x+la)^\alpha _u_x`$
$`(na)^\alpha (la)^\alpha `$
$`\times \left[1{\displaystyle \frac{\alpha ^2\sigma ^2}{(la)(na)}}+\alpha (\alpha +1)\sigma ^2\left({\displaystyle \frac{1}{(na)^2}}+{\displaystyle \frac{1}{(la)^2}}\right)\right];`$
from which
$`(ux+na)^\alpha _u[(ux+la)^\alpha _u`$ (25)
$`(u+x+la)^\alpha _u]_x{\displaystyle \frac{2\alpha ^2\sigma ^2}{(la)^{\alpha +1}(na)^{\alpha +1}}}.`$
Using all these results in all the terms of Eq. (24), we obtain
$$\frac{\overline{^2}}{2}\alpha ^2\sigma ^2\left\{\underset{n=1}{\overset{N}{}}\frac{1}{(na)^{2(\alpha +1)}}+2\left[\underset{n=1}{\overset{N}{}}\frac{1}{(na)^{\alpha +1}}\right]^2\right\}$$
(26)
It is simple to verify that both sums in Eq. (26) are converging for $`N+\mathrm{}`$ for all $`\alpha >0`$, for which we can then rewrite
$$\frac{\overline{F^2}}{2}\frac{\alpha ^2\sigma ^2}{a^{2(\alpha +1)}}\left[\zeta (2\alpha +2)+2\zeta ^2(\alpha +1)\right],$$
(27)
where $`\zeta (t)`$ is the Riemann zeta function (note that for $`t1^+`$ wee have $`\zeta (t)1/(t1)`$). Again the first term is due to the fluctuations in the position of the sources, while the second one is due to the fluctuations in the position of the particle on which we are calculating the force. In particular In Eq. (26) the generic term of the first sum give the relative weight of the $`n^{th}`$ nearest neighbor particles in determining the force on the particle in $`X_0`$. At last we can say that, in the case of displacements limited within a box well contained in a unitary cell around the initial lattice position, we can approximate $`W(F)`$ with a Gaussian PDF with zero mean and variance given by Eq. (27). However, as already pointed out, there is no constraint, in the limit $`N\mathrm{}`$, toward rigorous Gaussianity and non-Gaussian corrections are in general present.
### V.2 Case 2: power law tailed $`W(F)`$
As shown above, this is the case in which the OC is satisfied, i.e., particles are permitted to jump out of their initial lattice positions beyond the limit of the related unitary cell in such a way to be found arbitrarily close to some other particle. Note that this is always the case when the support of $`p(u)`$ is unlimited, i.e., if $`p(u)>0`$, $`uIR`$. However the same kind of $`W(F)`$ is also obtained if $`u^{}>a/2`$ such that $`p(u)>0`$, $`u[u^{},u^{}]`$ and zero outside. The difference between these two sub-cases is only in the amplitude of the power law tail of $`W(F)`$ but not in its exponent. In general, if the particle initially at the lattice site $`x=na`$ is permitted, through displacements, to be found arbitrarily close to the particle initially at $`x=0`$, it will contribute to the product (10) through a factor of the type (15). If instead this is not permitted, it will contribute to (10) through a factor of the form (13) or (14) depending respectively on whether $`n>0`$ or $`n<0`$. In any case if $`M=[1/\alpha ]`$, in order to find the main terms of the small $`k`$ expansion of $`\widehat{W}(k)`$ (so to determine the large $`F`$ tail of $`W(F)`$), it is sufficient to truncate all the small $`k`$ expansion of the different factors in Eq. (10) at most to the order $`M+1`$. For the sake of simplicity, let us limit the discussion to the case in which strictly $`\alpha M<1`$ in such a way to exclude logarithmic corrections in $`k`$. We can write
$`\widehat{P}(k;x_0)`$ (28)
$`{\displaystyle \underset{n}{\overset{\left(\text{OC}\right)}{}}}\left[{\displaystyle \underset{m=0}{\overset{M}{}}}{\displaystyle \frac{(ik)^m}{m!}}\left({\displaystyle \frac{x}{|x|^{\alpha +1}}}\right)^m_{n,x_0}+A(n,x_0,\alpha )k^{1/\alpha }\right]`$
$`\times {\displaystyle \underset{l}{\overset{(\text{NOC})}{}}}\left[{\displaystyle \underset{m=0}{\overset{M+1}{}}}{\displaystyle \frac{(ik)^m}{m!}}\left({\displaystyle \frac{x}{|x|^{\alpha +1}}}\right)^m_{l,x_0}\right],`$
where $`A(n,x_0,\alpha )`$ is the coefficient of the term $`k^{1/\alpha }`$ in Eq. (16), and the first product on $`n`$ is on the particles in $`X_n`$ with $`n0`$ which can be found arbitrarily near to the particle in $`X_0`$ (i.e., satisfying the OC with respect to the particle in $`X_0`$), while the product on $`l`$ is on the particles in $`X_l`$ with $`l0`$ which have a positive minimal distance to the same particle (i.e., satisfying the NOC with respect to the particle in $`X_0`$). If $`p(u)>0`$ $`uIR`$ all the system particles with $`n0`$ are included in the first product. If instead $`p(u)>0`$ for $`u[u^{},u^{}]`$ with $`u^{}>a/2`$ and zero outside, the first product include only contributions from the particles with $`2u^{}/a<n<2u^{}/a`$ and $`n0`$, while the others are included in the second product. The large $`F`$ behavior of $`P(F;x_0)`$ and consequently of $`W(F)`$ is completely determined by the singular term of order $`k^{1/\alpha }`$ in the small $`k`$ expansion of $`\widehat{P}(k;x_0)`$. It is simple to see that up to the order $`k^{1/\alpha }`$
$$\widehat{P}(k;x_0)=\underset{m=0}{\overset{M}{}}c_m(x_0)k^m+c_{1/\alpha }(x_0)k^{1/\alpha },$$
where the the coefficients $`c_m(x_0)`$ can be deduced by counting from Eq. (28) (in particular $`c_m(x_0)=i^m\stackrel{~}{^m}(x_0)/m!`$, where $`\stackrel{~}{^m}(x_0)=_{\mathrm{}}^+\mathrm{}𝑑FP(F;x_0)F^m`$ is the $`m^{th}`$ moment of $`P(F;x_0)`$ and $`mM`$) and
$$c_{1/\alpha }=\underset{2u^{}/a<n<2u^{}/a}{\overset{n0}{}}A(n;x_0),$$
where the formula includes also the case $`u^{}\mathrm{}`$. The small $`k`$ expansion up to the order $`k^{1/\alpha }`$ of $`W(F)`$ will be consequently
$$\widehat{W}(k)\underset{m=0}{\overset{M}{}}b_mk^m+b_{1/\alpha }k^{1/\alpha },$$
where
$`b_m`$ $`=`$ $`{\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑x_0p(x_0)c_m(x_0)={\displaystyle \frac{i^m}{m!}}\overline{^m}\text{ with }mM`$
$`b_{1/\alpha }`$ $`=`$ $`{\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑x_0p(x_0)c_{1/\alpha }(x_0),`$
where $`\overline{^m}=_{\mathrm{}}^+\mathrm{}𝑑FW(F)F^m=_{\mathrm{}}^+\mathrm{}𝑑x_0p(x_0)\stackrel{~}{^m}(x_0)\stackrel{~}{^m}(x_0)`$. It is possible to evaluate explicitly $`b_{1/\alpha }`$ by using Eq. (16).
We are now in the situation to connect the singular term $`b_{1/\alpha }k^{1/\alpha }`$ of the small $`k`$ expansion of $`\widehat{W}(k)`$ to the large $`F`$ tail of $`W(F)`$ by using directly the arguments in Appendix I. This gives simply
$$W(F)BF^{11/\alpha }$$
with
$$B=\frac{1}{\alpha }_{\mathrm{}}^+\mathrm{}𝑑x_0p(x_0)\underset{2u^{}/a<n<2u^{}/a}{\overset{n0}{}}p(x_0na).$$
(29)
Note that if the support of $`p(x_0na)`$ is much larger than $`a`$ and $`p(u)`$ is smooth (i.e., approximately constant) on the scale $`a`$, we can approximate Eq. (29) with
$$B=\frac{1p(0)a}{\alpha a}.$$
Note that this last approximated expression is not dependent on the details of $`p(u)`$ for $`u0`$. Finally, we can observe that we have obtained a power law tailed $`W(F)`$ characterized by the same exponent of the case of a homogeneous Poisson particle distribution presented in Appendix II. The only differences are the two following:
* The amplitude of this power law tail is reduced in the shuffled lattice with respect to that of the Poisson particle distribution, given by Eq. (45), of a factor
$$_{\mathrm{}}^+\mathrm{}𝑑x_0p(x_0)\underset{2u^{}/a<n<2u^{}/a}{\overset{n0}{}}ap(x_0na)1p(0)a.$$
* In the shuffled lattice we have this power law tail for each $`\alpha >0`$, while in the Poisson case the problem is not well defined for $`\alpha 1/2`$ (see Appendix II).
## VI Conclusions
We have presented a detailed study of the PDF $`W(F)`$ of the stochastic force $``$ generated by a randomly perturbed lattice of sources of a scale invariant attractive pair interaction field $`f(x)=Cx/|x|^{\alpha +1}`$ with $`\alpha >0`$ at distance $`x`$ from the source.
In general we distinguish two cases:
1. The NOC is satisfied and no pair of particles can be found at an arbitrarily small reciprocal distance;
2. The OC is satisfied and it exists at least one of such pairs of particles.
In the first case we have a fast decreasing $`W(F)`$ similar to a Gaussian PDF at large $`F`$, even though no constraint toward an exact Gaussian central limit theorem is found. In the second case a power law tailed $`W(F)`$ is found. The unique exponent of such power law is directly related to the pair interaction exponent $`\alpha `$, while its amplitude depends also on the lattice spacing $`a`$ (with respect to the unit distance through which we measure $`x`$ in $`f(x)`$) and in general on the shape of the perturbations PDF $`p(u)`$. In particular in this case $`W(F)`$ has a power law tail with the same exponent as the stable Lévy distribution found in the Poisson case (see Appendix II) but with a reduced amplitude, even though, analogously to the case (i), no constraint has been found toward the stable Lévy distribution.
Some further general considerations have can now be done:
* In the case in which the probability of finding arbitrarily close to one each other, the large $`F`$ behavior of $`W(F)`$ is basically determined by the small $`x`$ behavior of $`f(x)`$ and not at all by the the fact if it is long range or not. Therefore if we considered a fast decreasing $`f(x)`$ but with the same divergence in $`x=0`$ we would have deduced the same conclusions about the exponent of the large $`F`$ tail of $`W(F)`$ <sup>2</sup><sup>2</sup>2It is possible to show andrea-corr instead that the large $`x`$ behavior of the pair interaction $`f(x)`$ determines the large $`|xy|`$ behavior of the field-field correlation function $`(x)(y)`$..
* For this reason, even if we consider a lattice perturbed by correlated displacements, we expect to obtain the partition into the two cases (i) and (ii) above considered depending on the possibility or not to find pair of particles arbitrarily close.
* Actually different cases for the large $`F`$ tail of $`W(F)`$ between the Gaussian-like “fast decreasing” and Lévy-like “power law” tailed PDF with exponent $`\beta =(\alpha +1)/\alpha `$ are possible in very particular cases. These cases correspond to the choice of $`p(u)`$ such that $`p(u)>0`$ exactly for $`u[a/2,a/2]`$ and zero outside. By changing the limit behaviors of $`p(u)`$ when $`u\pm a/2`$ we can obtain different large $`F`$ behaviors of $`W(F)`$. In particular if $`p(a/2)>0`$ and finite (we consider $`p(u)=p(u)`$) we have the same case as (ii) described above with $`\beta =(\alpha +1)/\alpha `$. If instead $`p(a/2)=0`$, depending on the behavior of $`p(u)`$ for $`u(a/2)^{}`$ we will have different values of $`\beta `$ but in general larger than $`(\alpha +1)/\alpha `$. If, finally, $`p(a/2)=+\mathrm{}`$ (in such a way that $`p(u)`$ remains anyway integrable) in general we obtain $`\beta `$ smaller than $`(\alpha +1)/\alpha `$ (but always $`>1`$ so that $`W(F)`$ remains integrable).
## Appendix I: Fourier transform of power law tailed PDFs
We are interested in the small $`k`$ behavior of the characteristic function $`\widehat{f}(k)`$ of a given power law tailed PDF $`f(x)`$ which for large $`|x|`$ behaves as $`A|x|^\alpha `$ with $`\alpha >1`$. Let us call $`[\alpha ]=n1`$ the integer part of $`\alpha `$. In this hypothesis $`\widehat{f}(k)`$ has a regular Taylor expansion up to the order $`n1`$ followed by a singular term proportional to $`k^{\alpha 1}`$:
$$\widehat{f}(k)_{\mathrm{}}^+\mathrm{}𝑑xf(x)e^{ikx}=\underset{m=0}{\overset{n1}{}}\frac{(ik)^m}{m!}\overline{x^m}+\widehat{f}_s(k),$$
(30)
where $`\overline{x^m}=_{\mathrm{}}^+\mathrm{}𝑑xx^mp(x)`$ and $`\widehat{f}_s(k)`$ contains the singular part of $`\widehat{f}(k)`$ and at small $`k`$ is an infinitesimal of order $`\alpha 1`$ in $`k`$ (if $`\alpha `$ is an integer it contains also logarithmic corrections).
Now we show that effectively at sufficiently small $`k`$, $`\widehat{f}_s(k)Bk^{\alpha 1}`$ (where now $`a(k)b(k)`$ means that $`lim_{k0}[a(k)/b(k)]=1`$) giving an explicit expression for $`B`$ as a function of both the amplitude $`A`$ and the exponent $`\alpha `$.
First of all, let us study the case of a function $`h(x)`$ that can be written as
$$h(x)=B|x|^\beta +h_0(x),$$
(31)
where $`B>0`$, $`0<\beta <1`$ and $`h_0(x)`$ is a smooth function, integrable in $`x=0`$ and such that $`x^\beta h_0(x)0`$ for $`|x|\mathrm{}`$. This means that $`h(x)`$ presents an even power law tail. In this case the small $`k`$ behavior of $`\widehat{h}(k)=_{\mathrm{}}^+\mathrm{}𝑑xh(x)e^{ikx}`$ is completely determined by the Fourier transform of $`B|x|^\beta `$, i.e.:
$$\widehat{h}(k)B_{\mathrm{}}^+\mathrm{}𝑑x|x|^\beta e^{ikx}$$
In order to perform this Fourier transform we introduce the integral representation:
$$|x|^\beta =\frac{1}{\mathrm{\Gamma }(\beta )}_0^{\mathrm{}}𝑑zz^{\beta 1}e^{|x|z},$$
(32)
where $`\mathrm{\Gamma }(\beta )`$ is the Euler Gamma function. Using Eq. (32) we can write:
$`{\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑x|x|^\beta e^{ikx}={\displaystyle \frac{1}{\mathrm{\Gamma }(\beta )}}{\displaystyle _0^{\mathrm{}}}𝑑zz^{\beta 1}{\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑xe^{ikx|x|z}`$
$`={\displaystyle \frac{2k^{\beta 1}}{\mathrm{\Gamma }(\beta )}}{\displaystyle _0^{\mathrm{}}}𝑑q{\displaystyle \frac{q^\beta }{1+q^2}}.`$
Using the general relation valid for $`0<\beta <1`$
$$_0^{\mathrm{}}𝑑q\frac{q^\beta }{1+q^2}=\frac{\pi }{2\mathrm{cos}\left(\frac{\beta }{2}\pi \right)},$$
(33)
we can conclude
$$\widehat{h}(k)=\frac{B\pi }{\mathrm{\Gamma }(\beta )\mathrm{cos}\left(\frac{\beta }{2}\pi \right)}k^{\beta 1}+o(k^{\beta 1}).$$
(34)
If instead of using the integral representation (32) one used
$$|x|^\beta =\frac{1}{\mathrm{\Gamma }\left(\frac{\beta }{2}\right)}_0^{\mathrm{}}𝑑zz^{\beta /21}e^{x^2z},$$
one should obtain the alternative expression containing only Gamma functions:
$$\widehat{h}(k)=\frac{B\sqrt{\pi }\mathrm{\Gamma }\left(\frac{1\beta }{2}\right)}{2^\beta \mathrm{\Gamma }\left(\frac{\beta }{2}\right)}k^{\beta 1}+o(k^{\beta 1}).$$
Note that the coefficient of the term $`k^{\beta 1}`$ is real and positive. Another important case is when the function $`h(x)`$ has an odd non-integrable power law tail, i.e.:
$$h(x)=B|x|^\beta [2\theta (x)1]+h_0(x),$$
(35)
where $`\theta (x)`$ is the usual Heaviside step function, $`B>0`$, $`0<\beta <1`$, and $`h_0(x)`$ the same of Eq. (31). By using the same integral transformation leading to Eq. (34), we in this case we obtain:
$$\widehat{h}(k)=i\frac{B\pi }{\mathrm{\Gamma }(\beta )\mathrm{sin}\left(\frac{\beta }{2}\pi \right)}k^{\beta 1}+o(k^{\beta 1}).$$
At this point we can go back to the problem of finding the dominant small $`k`$ contribution of the term $`\widehat{f}_s(k)`$ in Eq. (30) for the PDF $`f(x)`$ decaying at large $`|x|`$ as $`A|x|^\alpha `$. Note that now we cannot apply directly the argument we have used for the above function $`h(x)`$. In fact if, from one side, also in this case we can write
$$f(x)=A|x|^\alpha +f_0(x)$$
with $`|x|^\alpha f_0(x)0`$ for $`|x|\mathrm{}`$, from the other side $`\alpha >1`$ (for definiteness of probability) and $`f_0(x)`$ contains a non integrable singularity at $`x=0`$ such that to cancel the non integrable contribution of the $`A|x|^\alpha `$ term at small $`x`$. In order to circumvent this difficulty we introduce the function
$$g(x)=x^nf(x),$$
where $`n`$ is the integer part of $`\alpha `$. In this way $`g(x)`$ is similar to the function $`h(x)`$ of Eq. (31) if $`n`$ is even and to the $`h(x)`$ of Eq. (35) if $`n`$ is odd. Therefore, by defining as usual $`\widehat{g}(k)=_{\mathrm{}}^+\mathrm{}𝑑xg(x)e^{ikx}`$, we can say that
$$\widehat{g}(k)=\frac{A\pi }{\mathrm{\Gamma }(\alpha n)\mathrm{cos}\left(\frac{\alpha n}{2}\pi \right)}k^{\alpha n1}+o(k^{\alpha n1})$$
if $`n`$ is even, and
$$\widehat{g}(k)=i\frac{A\pi }{\mathrm{\Gamma }(\alpha n)\mathrm{sin}\left(\frac{\alpha n}{2}\pi \right)}k^{\alpha n1}+o(k^{\alpha n1})$$
if $`n`$ is odd. Now in order to find the singular part $`\widehat{f}_s(k)`$ of $`\widehat{f}(k)`$ it is sufficient to integrate $`n`$ times $`\widehat{g}(k)`$ (the integration constants giving rise to the finite moments terms of $`\widehat{f}(k)`$ in Eq. (30)). In this way we obtain
$`\widehat{f}_s(k)`$ (36)
$`=\{\begin{array}{cc}\frac{(1)^{n/2}A\pi }{\mathrm{\Gamma }(\alpha )\mathrm{cos}\left(\frac{\alpha n}{2}\pi \right)}k^{\alpha 1}+o(k^{\alpha 1})\hfill & \text{for even }n\hfill \\ & \\ \frac{(1)^{(n+1)/2}A\pi }{\mathrm{\Gamma }(\alpha )\mathrm{sin}\left(\frac{\alpha n}{2}\pi \right)}k^{\alpha 1}+o(k^{\alpha 1})\hfill & \text{for odd }n,\hfill \end{array}`$ (40)
where we have used the following property of Gamma function: $`\mathrm{\Gamma }(x+1)=x\mathrm{\Gamma }(x)`$, hence $`[(\alpha 1)\mathrm{}(\alpha n)\mathrm{\Gamma }(\alpha n)]=\mathrm{\Gamma }(\alpha )`$. Note that in both case the coefficient of the term $`k^{\alpha 1}`$ is real.
## Appendix II: Force PDF in a Poisson particle distribution
Let us consider the case in which the particles are distributed on the line interval $`(L/2,L/2]`$ of length $`L`$ following a spatially stationary Poisson process with average density $`\rho _0>0`$. We want to know the PDF $`W_P(F)`$ of the field
$$F=\underset{i=1}{\overset{N}{}}\frac{x_i}{|x_i|^{\alpha +1}}$$
(41)
generated at the origin of the space by all the $`N`$ system particles (we can consider $`N=\rho _0L`$ as the fluctuations of order $`\sqrt{\rho _0L}`$, due to the Poisson statistics, are completely unimportant for this problem in the large $`L`$ limit). We will follow the procedure to find $`W_P(F)`$ in three dimensions used by Chandrasekhar in chandra for the gravitational force. Note that, as the positions of different particles are uncorrelated, the joint PDF $`p(x_1,\mathrm{},x_N)`$ of the positions of the $`N`$ system particles is simply:
$$p_N(x_1,\mathrm{},x_N)=\underset{i=1}{\overset{N}{}}p(x_i),$$
where $`p(x_i)=1/L`$. Therefore we can write
$$W_P(F)=_{L/2}^{L/2}\left[\underset{i=1}{\overset{N}{}}\frac{dx_i}{L}\right]\delta \left(F\underset{j=1}{\overset{N}{}}\frac{x_i}{|x_j|^{\alpha +1}}\right).$$
(42)
Let us now study the characteristic function
$$\widehat{W}_P(k)=[W_P(F)]=_{\mathrm{}}^+\mathrm{}𝑑FW_P(F)e^{ikF}.$$
By taking the FT of Eq. (42) we obtain
$$\widehat{W}_P(k)=\left[_{L/2}^{L/2}\frac{dx}{L}\mathrm{exp}\left(ik\frac{x}{|x|^{\alpha +1}}\right)\right]^N$$
By adding and subtracting 1 inside the square brackets, and taking the limit $`L\mathrm{}`$ with $`N=\rho _0L`$ we arrive at the final expression:
$$\widehat{W}_P(k)=\mathrm{exp}\left[\rho _0k^{1/\alpha }_{\mathrm{}}^+\mathrm{}𝑑t\left(1\mathrm{exp}\left(i\frac{t}{|t|^{\alpha +1}}\right)\right)\right].$$
Note that
$`{\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑t\left(1\mathrm{exp}\left(i{\displaystyle \frac{t}{|t|^{\alpha +1}}}\right)\right)=2{\displaystyle _0^{\mathrm{}}}𝑑t\left(1\mathrm{cos}(t^\alpha )\right)`$
$`={\displaystyle \frac{2}{\alpha }}{\displaystyle _0^{\mathrm{}}}𝑑uu^{11/\alpha }(1\mathrm{cos}u),`$
where the last passage is due to the change of variable $`t^\alpha =u`$. Let us now use, as in the previous appendix, the integral representation:
$$u^{11/\alpha }=\frac{1}{\mathrm{\Gamma }\left(\frac{\alpha +1}{\alpha }\right)}_0^{\mathrm{}}𝑑zz^{1/\alpha }e^{uz}.$$
Through this transformation we arrive finally to the relation
$`{\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑t\left(1\mathrm{exp}\left(i{\displaystyle \frac{t}{|t|^{\alpha +1}}}\right)\right)`$
$`={\displaystyle \frac{2}{\alpha \mathrm{\Gamma }\left(\frac{\alpha +1}{\alpha }\right)}}{\displaystyle _0^{\mathrm{}}}𝑑z{\displaystyle \frac{z^{1+1/\alpha }}{1+z^2}}.`$ (43)
Note that the last integral is diverging for $`\alpha 1/2`$, indicating that the problem is not well defined for these values of $`\alpha `$ as $`F`$ is not a well defined stochastic quantity. This means that the sum in Eq. (41) needs a $`L`$ dependent normalization to become a well defined stochastic variable. In fact, differently to the shuffled lattice case, where the typical mass fluctuation on regions of size $`R`$ is proportional to $`R^0`$, in the Poisson case this is due to the fact that such fluctuation is proportional to $`R^{1/2}`$. The field due to the mass fluctuation in a sphere of radius $`R`$ on the origin of the sphere is of order $`R^\alpha `$ for the shuffled lattice and $`R^{\alpha +1/2}`$ in the Poisson point process. This explains why for a shuffled lattice the problem is well defined for any $`\alpha >0`$ and not only for $`\alpha >1/2`$. This also says that for $`\alpha <1/2`$, in order to have a well defined stochastic field also in the Poisson case, we have to divide the field in Eq. (41) by $`L^{\alpha +1/2}`$ where $`L`$ is the system size. The same argument can be used in $`d`$ dimensions, for which the same mass fluctuations are respectively proportional to $`R^{(d1)/2}`$ in the shuffled lattice, and $`R^{d/2}`$ in the Poisson case. This says that the problem is well defined, without $`L`$ dependent normalization of the field, for $`\alpha >(d1)/2`$ for the shuffled lattice and $`\alpha >d/2`$ for the Poisson case (in the gravitational case in $`d=3`$ faced by Chandrasekhar chandra we have $`\alpha =2`$ and the field is a well defined stochastic quantity in both cases).
Let us now go back to Eq. (43) for $`\alpha >1/2`$. By using Eq. (33) we can rewrite it as
$$_{\mathrm{}}^+\mathrm{}𝑑t\left(1\mathrm{exp}\left(i\frac{t}{|t|^{\alpha +1}}\right)\right)=\frac{\pi }{\alpha \mathrm{\Gamma }\left(\frac{\alpha +1}{\alpha }\right)\mathrm{sin}\left(\frac{\pi }{2\alpha }\right)},$$
where we have also used $`\mathrm{cos}[\pi (1/\alpha 1)/2]=\mathrm{sin}[\pi /(2\alpha )]`$. Consequently
$$\widehat{W}_P(k)=\mathrm{exp}\left[\frac{\pi \rho _0}{\alpha \mathrm{\Gamma }\left(\frac{\alpha +1}{\alpha }\right)\mathrm{sin}\left(\frac{\pi }{2\alpha }\right)}k^{1/\alpha }\right],$$
(44)
which is exactly of the form of the characteristic function of a Lévy stable PDF levy-stable . By expanding Eq. (44) to the first non-vanishing order larger than zero, we have:
$$\widehat{W}_P(k)=1\frac{\pi \rho _0}{\alpha \mathrm{\Gamma }\left(\frac{\alpha +1}{\alpha }\right)\mathrm{sin}\left(\frac{\pi }{2\alpha }\right)}k^{1/\alpha }+o(k^{1/\alpha }).$$
If we now invert this Fourier transform as explained in the preceding appendix, we can conclude that
$$W_P(F)BF^{(\alpha +1)/\alpha }\text{ for large }F,$$
with
$$B=\frac{\rho _0}{\alpha }.$$
(45)
|
warning/0506/math0506025.html
|
ar5iv
|
text
|
# The 𝑁-eigenvalue Problem and Two Applications
## 1. Introduction
Consider the following two questions:
1. When does a compact Lie group $`GU(n)`$ have an element $`gG`$ possessing exactly two eigenvalues.
2. When does a compact Lie group $`GU(n)`$ have a cocharacter $`U(1)G`$ such that the composition $`U(1)U(n)`$ is a representation of $`U(1)`$ with exactly two weights.
A solution to the second problem gives a family of solutions to the first, by choosing $`g`$ to be almost any element of the image of $`U(1)`$. The converse is not true. For one thing, any non-central element of order $`2`$ in $`G`$ has exactly two eigenvalues. To eliminate these essentially trivial solutions, we can insist that the ratio between the two eigenvalues is not $`1`$. There remain interesting cases of finite groups $`G`$ satisfying the first (but obviously not the second) condition, especially when the ratio of eigenvalues is a third or fourth root of unity (see \[Bl\], \[Ko\], and \[W\] for classification results). On the other hand, when $`G`$ is infinite modulo center, the solutions of the two problems are essentially the same, though the historical reasons for considering them were quite different. The first problem was recently solved in the infinite-mod-center case by M. Freedman, M. Larsen, and Z. Wang \[FLW\] with an eye toward understanding representations of Hecke algebras. The second problem was solved by J-P. Serre \[Ser\] nearly thirty years ago in order to classify representations arising from Hodge-Tate modules of weight 1.
This paper is primarily devoted to an effort to understand the analogue of the first problem (the “$`N`$-eigenvalue problem” of the title) when the number $`N3`$ of eigenvalues is fixed and $`G`$ is infinite modulo its center. As a consequence, we also say something about the second problem. We are especially interested in the case $`N=3`$, both because the results can be made quite explicit and because it is especially relevant to the applications we have in mind. To specify our problem more precisely, we make the following definitions.
A *pair* $`(G,V)`$ consists of a compact Lie group $`G`$ and a faithful irreducible complex representation $`\rho :G\mathrm{GL}(V)`$. Let $`N`$ be a positive integer. We say a pair $`(G,V)`$ satisfies the *$`N`$-eigenvalue property* if there exists a *generating element*, i.e., an element $`gG`$ such that the conjugacy class of $`g`$ generates $`G`$ topologically and the spectrum $`X`$ of $`\rho (g)`$ has $`N`$ elements and satisfies the *no-cycle property*: for all roots of unity $`\zeta _n`$, $`n2`$, and all $`u^\times `$,
(1.1)
$$u\zeta _nX$$
Our goal is to classify pairs satisfying the $`N`$-eigenvalue property.
From the perspective of \[FLW\], the most obvious reason to consider the $`N`$-eigenvalue property is that certain naturally occurring representations of the Artin braid groups $`_n`$ satisfy this condition. The braid generators (half-twists) in the braid group form a generating conjugacy class, and given any braided tensor category $`𝒞`$ and any object $`x𝒞`$, we get a representation of $`\rho _{n,x}:_n\mathrm{GL}(V_{n,x})`$. When $`\rho _{n,x}`$ is unitary with respect to a hermitian form on $`V_{n,x}`$, the closure of $`\rho _{n,x}`$ is a compact Lie group endowed with a natural faithful representation and a generating conjugacy class. It is often possible to control the eigenvalues of half-twists, to guarantee the $`N`$-eigenvalue condition, and to guarantee irreducibility. In the case when the braided tensor category $`𝒞`$ is modular, we obtain in addition representations of the mapping class groups $`(\mathrm{\Sigma }_g)`$ of closed oriented surfaces $`\mathrm{\Sigma }_g`$ for each genus $`g`$. It is well-known that $`(\mathrm{\Sigma }_g)`$ is generated by the (mutually conjugate) Dehn twists $`D_c`$ on $`3g1`$ non-separating simple closed curves $`c`$ on $`\mathrm{\Sigma }_g`$ (see \[I\]). If $`𝒞`$ has $`m`$ simple object types, then each $`D_c`$ has at most $`m`$ distinct eigenvalues as the eigenvalues of $`D_c`$ consist of twists $`\theta _i`$ of the simple objects. When the values $`\theta _i`$ satisfy the no-cycle condition, it follows that each irreducible constituent of the representation of $`(\mathrm{\Sigma }_g)`$ arising from $`𝒞`$ defines a pair satisfying the $`N`$-eigenvalue property for some $`Nm`$.
The original motivation for the work of \[FLW\] was for applications to quantum computing. In \[FKLW\], topological models of quantum computing based on unitary topological quantum field theories (TQFTs) are proposed. Given a topological model of quantum computing, an important issue is whether or not this topological model is capable of simulating the universal circuit model of quantum computing \[NC\]. This question actually depends on the specific design of the topological quantum computer. But for the models based on braiding anyons in \[FKLW\], the universality question is translated into a question about the closures of the braid group representations. Quantum computing is the processing of information encoded in quantum state vectors in certain Hilbert spaces $`V_n`$ by unitary transformations. Universality is the ability to efficiently move any state vector $`vV_n`$ sufficiently close to any other state vector in $`V_n`$. A theorem of Kitaev-Solovay (see \[NC\]) guarantees efficiency if the available unitary transformations in $`U(V_n)`$ form a dense subset of $`SU(V_n)`$. Therefore, universality of topological models in \[FKLW\] is equivalent to the density of braid group representations.
The unitary Witten-Reshetikhin-Turaev Chern-Simons TQFTs based on the gauge groups $`\mathrm{SU}(N)`$ and $`\mathrm{SO}(N)`$ are of particular interests due to their relevance to braid statistics in condensed matter physics. It was discovered in the 1980s that in dimension 2, there are quasi-particles which are neither fermions nor bosons \[Wi\]. The most interesting of these *anyons* are non-abelian: when two such quasi-particles are exchanged, their wave function is changed by a unitary matrix, rather than a complex number, which depends on the exchanging paths (braiding). It is predicted by physicists that the braid statistics of quasi-particles in certain fractional quantum Hall liquids are described by Jones’ unitary braid groups representations or equivalently the braid representations coming from the $`\mathrm{SU}(2)`$ TQFTs. Physicists have also proposed models of braid statistics based on the $`\mathrm{SO}(3)`$ \[FF\] and $`\mathrm{SO}(5)`$ \[Wn\] TQFTs. Therefore, it may well be the case that both the Jones and the BMW braid group representations describe braid statistics of quasi-particles in nature. Experiments are proposed to confirm those predictions \[DFN\].
Problem (2) is significant partly because of its relation to problem (1), but in addition, there are number-theoretic applications, in the spirit of \[Ser\]. We mention a global one: assuming the Fontaine-Mazur conjecture, we can prove that if $`K`$ is a number field, $`\overline{K}`$ is an algebraic closure of $`K`$, $`G_K=\mathrm{Gal}(\overline{K}/K)`$, and $`X`$ is a non-singular projective variety over $`K`$, then $`E_8`$ does not occur as a factor of the identity component of the Zariski-closure of $`G_K`$ in the second étale cohomology group of $`\overline{X}`$.
The paper is organized as follows. The second and third sections treat the infinite-mod-center case of the $`N`$-eigenvalue problem. The second section gives the general shape of the solution for all $`N`$, and the third section gives an actual list for $`N=3`$. The fourth section shows that a fairly weak hypothesis on the actual eigenvalues is enough to guarantee that $`G`$ is infinite modulo its center. The fifth section gives applications to number theory, and the sixth section gives applications to braid group representations. We conclude with a discussion of future applications to topology and quantum computing.
### Acknowledgements
The second-named author would like to thank Hans Wenzl for many helpful correspondences.
## 2. Infinite groups
In this section, we consider the general $`N`$-eigenvalue problem for infinite compact groups. Our methods come directly from \[FLW\] and \[LW\].
###### Lemma 2.1.
Let $`V=V_1\mathrm{}V_k`$ be a complex vector space and $`T:VV`$ a linear transformation permuting the summands $`V_i`$ non-trivially. Then the spectrum of $`V`$ does not satisfy (1.1).
###### Proof.
Renumbering if necessary, we may assume $`V`$ permutes $`V_1,V_2,\mathrm{},V_r`$ cyclically, where $`r2`$. Let $`W=V_1\mathrm{}V_r`$, let $`\zeta _r=e^{\frac{2\pi i}{r}}`$, and let $`S:WW`$ act as the scalar $`\zeta _r^i`$ on $`V_i`$. Then
$$ST|_WS^1=\zeta _pT|_W,$$
so the spectrum of $`T|_W`$ is invariant under multiplication by $`\zeta _p`$. It is therefore a union of $`\zeta _p`$-cosets. ∎
###### Lemma 2.2.
Let $`(G_1,V_1)`$ and $`(G_2,V_2)`$ be pairs and let $`G`$ denote the image of $`G_1\times G_2`$ in $`\mathrm{GL}(V_1V_2)`$. If $`G`$ satisfies the $`N`$-eigenvalue property, then there exist integers $`N_1`$ and $`N_2`$ such that $`N_1+N_21N`$, and subgroups $`G_1^{}<G_1`$ and $`G_2^{}<G_2`$ such that $`(G_i^{},V_i)`$ satisfies the $`N_i`$-eigenvalue property and $`G_i^{}Z(G_i)=G_i`$ for $`i=1,2`$.
###### Proof.
Let $`gG`$ be a generating element, and let $`(g_1,g_2)G_1\times G_2`$ map to $`G`$. Let $`G_i^{}`$ denote the subgroup of $`G_i`$ generated by the conjugacy class of $`g_i`$. As the conjugacy class of $`g`$ generates $`G`$, $`(\mathrm{ker}G_iG)G_i^{}=G_i`$. By construction, $`\mathrm{ker}G_iGZ(G_i)`$.
The spectrum of $`\rho (g)`$ is the product of the spectra of $`\rho _i(g_i)`$. So the lemma reduces to the following claim: if $`X_1`$ and $`X_2`$ are finite subgroups of an abelian group $`A`$ such that $`X_1+X_2`$ does not contain a coset of a non-trivial subgroup of $`A`$, then $`|X_1+X_2||X_1|+|X_2|1`$. This is well-known (see, e.g., \[Ke\]). ∎
If $`(G,V)`$ arises in this way, we say it is *decomposable*; otherwise, it is *indecomposable*. Note that the tensor product of pairs which satisfy the $`N_1`$ and $`N_2`$-eigenvalue conditions need not satisfy the $`N_1+N_21`$-eigenvalue condition. For one thing, the product of sets of cardinality $`N_1`$ and $`N_2`$ could be as large as $`N_1N_2`$. For another, the product of sets satisfying the no-cycle property may itself fail to satisfy the no-cycle property.
###### Proposition 2.3.
Let $`(G,V)`$ be an indecomposable pair. If $`G`$ is infinite modulo its center and $`(G,V)`$ satisfies the $`N`$-eigenvalue property for some $`N`$, then $`G=G^{}Z(G)`$.
###### Proof.
Let $`gG`$ be a generating element. Then $`gG^{}`$ implies $`G=G^{}`$, in which case there is nothing to prove. If $`V|_G^{}`$ is not isotypic, then $`g`$ acts non-trivially on the isotypic factors, and by Lemma 2.1, the spectrum of $`g`$ fails to satisfy property (1.1). If $`V|_G^{}=W^n=WU`$, where $`G^{}`$ acts trivially on $`U`$ and irreducibly on $`W`$, then the span of $`\rho (G^{})`$ is $`\mathrm{End}(W)\mathrm{Id}_U\mathrm{End}(V)`$, so $`\rho (G)`$ lies in the normalizer of $`\mathrm{End}(W)\mathrm{Id}_U`$, which is $`\mathrm{End}(W)\mathrm{End}(U)`$. Thus $`\rho `$ maps $`G`$ to $`(\mathrm{GL}(W)\times \mathrm{GL}(U))/^\times `$. Let $`\stackrel{~}{G}`$ denote the cartesian square
$$\begin{array}{ccc}\stackrel{~}{G}& \stackrel{\stackrel{~}{\rho }}{}& \mathrm{GL}(W)\times \mathrm{GL}(U)\\ \pi & & & & \\ G& \stackrel{\rho }{}& (\mathrm{GL}(W)\times \mathrm{GL}(U))/^\times .\end{array}$$
If $`\stackrel{~}{g}\pi ^1(g)`$, then the projections of $`\stackrel{~}{\rho }(\stackrel{~}{g})`$ to $`\mathrm{GL}(W)`$ and $`\mathrm{GL}(U)`$ have spectra satisfying the no-cycle property, since the product of these spectra is the spectrum of $`\stackrel{~}{\rho }(\stackrel{~}{g})`$. If $`dimW`$ and $`dimU`$ are both $`2`$, then $`(G,V)`$ is decomposable, contrary to hypothesis. As $`G^{}`$ is not in the center of $`G`$, $`dimW2`$. It follows that $`dimU=1`$, i.e., the restriction of $`V`$ to $`G^{}`$ is irreducible. Thus every element of $`G`$ which commutes with $`G^{}`$ lies in $`Z(G)`$.
It follows that for every $`gG^{}Z(G)`$, conjugation by $`g`$ induces an automorphism of $`G^{}`$ which is not inner. By \[St, 7.5\], this implies that there exists a maximal torus $`T`$ of $`G^{}`$ such that $`gTg^1=T`$ but conjugation by $`g`$ induces a non-trivial automorphism of $`T`$. The characters of $`T`$ appearing in $`V|_T`$ span $`X^{}(T)`$ since $`V`$ is a faithful representation. Therefore, a non-trivial automorphism of $`T`$ must permute the weights of $`V`$ non-trivially. By Lemma 2.1, this implies that the spectrum of $`g`$ violates the no-cycle property, contrary to hypothesis. Thus $`gG^{}Z(G)`$, and since the conjugacy class of $`g`$ generates $`G`$, it follows that $`G=G^{}Z(G)`$. ∎
###### Proposition 2.4.
Let $`(G,V)`$ be as in Proposition 2.3. Then $`G`$ is the product of the derived group $`D`$ of $`G^{}`$ and a group of scalar matrices in $`V`$. The group $`D`$ is simple modulo its center, and the restriction of $`V`$ to $`D`$ is irreducible. If the highest weight $`\lambda `$ of $`V|_D`$ is written as a linear combination $`_ia_i\varpi _i`$, where $`\varpi _i`$ are the fundamental weights, then $`_ia_ib_iN1`$, where the $`b_i`$ are positive integers determined by the root system of $`D`$.
###### Proof.
As $`G^{}`$ is connected, $`G^{}=DZ(G^{})`$. As $`V|_G^{}`$ is irreducible, $`Z(G^{})`$ contains only scalars, as does $`Z(G)`$. Thus $`G=DZ(G^{})Z(G)`$, and the product $`Z(G^{})Z(G)`$ is scalar in $`\mathrm{GL}(V)`$. The centralizer of $`D`$ in $`\mathrm{GL}(V)`$ equals the centralizer of $`G^{}=DZ(G^{})`$ since $`Z(G^{})`$ is scalar. It follows that $`V|_D`$ is irreducible. Any tensor decomposition of $`V|_D`$ extends to $`G`$ since scalars respect any tensor decomposition; it follows that $`V|_D`$ is tensor indecomposable and therefore that $`D`$ is simple modulo its center. Let $`\lambda `$ denote its highest weight.
Let $`g`$ be a generating element, and let $`tD`$ be such that $`g^1t`$ is a scalar. $`T`$ be a maximal torus of $`D`$ containing $`t`$, $`R`$ the root system of $`D`$ with respect to $`T`$, and $`(,)`$ the Killing form on $`X^{}(T)`$. Let
$$\beta ,\alpha =\frac{2(\beta ,\alpha )}{(\alpha ,\alpha )},$$
and fixing a Weyl chamber, let $`\gamma `$ denote the root dual to the highest root in $`R`$. Thus $`\gamma `$ is the highest short root. By \[Bo, VIII, §7, Prop. 3(i)\], the maximal arithmetic progression of the form $`\lambda ,\lambda \gamma ,\lambda 2\gamma ,\mathrm{}`$ contained in the set of weights of $`V`$ has length
$$1+\lambda ,\gamma =1+\underset{i}{}a_ib_i,$$
where the positive integers $`b_i`$ are the coefficients in the representation of the highest root in $`R`$ in terms of the simple roots. If this sum exceeds $`N`$, then the geometric progression of values
$$\lambda (t),(\lambda \gamma )(t),(\lambda 2\gamma )(t),\mathrm{}$$
must either take $`N+1`$ distinct values, or fail (1.1), or be constant. The first two possibilities are ruled out by hypothesis, and it follows that $`\gamma (t)=1`$. If $`w`$ belongs to the Weyl group, the same considerations apply to the weight sequence $`w(\lambda ),w(\lambda )w(\gamma ),w(\lambda )2w(\gamma ),\mathrm{}`$, so $`w(\lambda )(t)=1`$. On the other hand, the short weights in a simple root system form a single Weyl orbit and generate the root lattice, so $`\alpha (t)=1`$ for all roots. This implies that $`t`$ lies in the center of $`G`$ and therefore that $`\rho (t)`$ is scalar, contrary to hypothesis. ∎
One can also formulate the $`N`$-eigenvalue property for complex Lie groups:
###### Definition 2.5.
Let $`G_{}`$ be a reductive complex Lie group and $`(\rho ,V)`$ a faithful irreducible complex representation of $`G_{}`$. Then $`(G_{},V)`$ satisfies the *$`N`$-eigenvalue property* if there exists a semisimple *generating element* $`g_{}G_{}`$ whose conjugacy class generates a Zariski-dense subgroup of $`G_{}`$, and such that the spectrum of $`\rho (g_{})`$ consists of $`N`$ eigenvalues satisfying the no-cycle condition.
###### Lemma 2.6.
Let $`G_{}`$ be a reductive complex Lie group and $`(\rho ,V)`$ a faithful irreducible complex representation of $`G_{}`$. Let $`G`$ be a maximal compact subgroup of $`G_{}`$. Then $`(G,V)`$ satisfies the $`N`$-eigenvalue property.
###### Proof.
Let $`T_{}`$ denote the Zariski-closure of the cyclic group $`g_{}`$ and $`TT_{}`$ the (unique) maximal compact subgroup. As $`T`$ can be regarded as the set of (real) points of a real algebraic group whose complex points give $`T_{}`$, $`T`$ is Zariski-dense in $`T_{}`$. We can decompose the restriction of $`V`$ to $`T_{}`$ as a direct sum of eigenspaces $`V_\chi `$ associated to characters $`\chi `$ of $`T_{}`$. There must be exactly $`N`$ such eigenspaces, since any coincidence among $`\chi _1(g_{}),\mathrm{},\chi _{N+1}(g_{})`$ gives the same coincidence for the characters on all of $`T_{}`$. The condition that $`\chi _i(t)\chi _j(t)`$ is open and non-empty in $`T_{}`$ as is the condition that $`\{\chi _1(t),\mathrm{},\chi _N(t)\}`$ satisfy the no-cycle condition. It follows that $`T`$ contains an element $`g`$ which satisfies both conditions.
As all maximal compact subgroups of $`G_{}`$ are conjugate, without loss of generality we may assume $`TG`$. We can regard $`G`$ as the group of real points of a real linear algebraic group whose complex points give $`G_{}`$ and $`TG`$ as a Zariski-closed subgroup. Let $`HG`$ denote the smallest normal Zariski-closed subgroup of $`G`$ containing $`g`$, or equivalently, $`T`$. Thus $`H`$ can be regarded as the group of real points of an algebraic group which is a normal subgroup of the algebraic group with real locus $`G`$. Let $`H_{}`$ denote the group of $``$-points of this subgroup. If $`HG`$, then $`H_{}G_{}`$, so $`g_{}T_{}H_{}`$ is contained in a proper normal subgroup of $`g_{}`$, contrary to hypothesis. It follows that $`g`$ is a generating element for $`(G,V)`$.
## 3. The $`3`$-eigenvalue problem
In this section, we give an explicit solution of the $`3`$-eigenvalue problem, assuming throughout that $`G`$ is a compact Lie group which is infinite modulo center.
###### Proposition 3.1.
If $`(G,V)`$ is a pair satisfying the 2-eigenvalue property, and $`\mathrm{\Phi }`$ denotes the root system of $`G`$ and $`\varpi `$ the highest weight of $`V`$ in the notation of \[Bo\], then $`(\mathrm{\Phi },\varpi )`$ is one of the following:
1. $`(A_r,\varpi _i)`$, $`1ir`$.
2. $`(B_r,\varpi _r)`$.
3. $`(C_r,\varpi _1)`$.
4. $`(D_r,\varpi _i)`$, $`i=1,r1,r`$.
###### Proof.
This is the statement of \[FLW, 1.1\]. ∎
Before treating the general $`3`$-eigenvalue problem, we make a detailed study of the $`A_r`$ case.
###### Lemma 3.2.
Let $`(\rho ,V)`$ be an irreducible representation of $`\mathrm{SU}(n)`$ with highest weight $`\varpi `$, and $`t`$ a non-central element of $`\mathrm{SU}(n)`$. Suppose there are at most three eigenvalues of $`\rho (t)`$ and they satisfy the no-cycle property. Then one of the following is true:
1. For $`1in1`$, $`\varpi =\varpi _i`$, and $`t`$ has characteristic polynomial $`(x\lambda )^{n1}(x\lambda ^{1n})`$; the eigenvalues of $`\rho (t)`$ are $`\lambda ^i`$, $`\lambda ^{in}`$.
2. For $`1in1`$, $`\varpi =\varpi _i`$, and $`t`$ has characteristic polynomial $`(x\lambda _1)^{n2}(x\lambda _2)^2`$; the eigenvalues of $`\rho (t)`$ are $`\lambda _1^i`$, $`\lambda _1^{i1}\lambda _2`$, and $`\lambda _1^{i2}\lambda _2^2=\lambda _1^{in}`$.
3. For $`i\{1,2,n2,n1\}`$, $`\varpi =\varpi _i`$, and $`t`$ has eigenvalues $`\lambda _1`$ and $`\lambda _2`$; the spectrum of $`\rho (t)`$ is $`\{\lambda _1,\lambda _2\}`$, $`\{\lambda _1^2,\lambda _1\lambda _2,\lambda _2^2\}`$, $`\{\lambda _1^2,\lambda _1^1\lambda _2^1,\lambda _2^2\}`$, or $`\{\lambda _1^1,\lambda _2^1\}`$, if $`i`$ is $`1`$, $`2`$, $`n2`$, or $`n1`$ respectively.
4. For $`1in1`$, $`\varpi =\varpi _i`$, and $`t`$ has characteristic polynomial $`(x\lambda ^{n2})(x\lambda \mu )(x\lambda \mu ^1)`$; the eigenvalues of $`\rho (t)`$ are $`\lambda _1^i`$, $`\lambda _1^i\mu `$, and $`\lambda _1^i\mu ^1`$.
5. For $`i=1`$ or $`i=n1`$, $`\varpi =\varpi _i`$, and $`t`$ has eigenvalues $`\lambda _1,\lambda _2,\lambda _3`$; the eigenvalues of $`\rho (t)`$ are the $`\lambda _j`$ or the $`\lambda _j^1`$ if $`i=1`$ or $`i=n1`$ respectively.
6. For $`i=1`$ or $`i=n1`$, $`\varpi =2\varpi _i`$, and $`t`$ has eigenvalues $`\lambda _1`$ and $`\lambda _2`$, each of multiplicity at least $`2`$; the eigenvalues of $`\rho (t)`$ are $`\{\lambda _1^2,\lambda _1\lambda _2,\lambda _2^2\}`$ or $`\{\lambda _1^2,\lambda _1^1\lambda _2^1,\lambda _2^2\}`$ if $`i`$ is $`1`$ or $`n1`$ respectively.
7. The highest weight $`\varpi `$ is $`\varpi _1+\varpi _{n1}`$, and $`t`$ has eigenvalues $`\lambda _1`$ and $`\lambda _2`$, each of multiplicity at least $`2`$; the eigenvalues of $`\rho (t)`$ are $`\lambda _1/\lambda _2`$, $`1`$, and $`\lambda _2/\lambda _1`$.
8. For $`1ijn1`$, $`\varpi =\varpi _i+\varpi _j`$, and $`t`$ has characteristic polynomial $`(x\lambda )^{n1}(x\lambda ^{1n})`$; the eigenvalues of $`\rho (t)`$ are $`\lambda ^{i+j}`$, $`\lambda ^{i+jn}`$, and $`\lambda ^{i+j2n}`$.
In particular, only case (5) can give three eigenvalues not in geometric progression.
###### Proof.
By Proposition 2.4, if $`\rho (t)`$ has $`N3`$ eigenvalues, $`\varpi `$ is a sum of at most $`N1`$ fundamental weights. If $`\varpi =\varpi _i`$ and $`t`$ has eigenvalues $`\lambda _1,\mathrm{},\lambda _n`$, the eigenvalues of $`\rho (t)`$ are
$$\left\{\underset{sS}{}\lambda _s\right|S\{1,\mathrm{},n\},|S|=i\}.$$
Duality exchanges $`\varpi _i`$ and $`\varpi _{ni}`$ so without loss of generality we may assume $`in/2`$. If $`\lambda _1,\mathrm{},\lambda _4`$ are all distinct, and $`ni+3`$ (in particular, this holds if $`n5`$), then
$$\{\lambda _j\lambda _5\lambda _6\mathrm{}\lambda _{3+i}1j4\}$$
already contains four distinct elements. If $`n=4`$ and $`i=2`$, two products $`\lambda _i\lambda _j`$ and $`\lambda _k\lambda _l`$ are distinct unless $`\{i,j\}`$ and $`\{k,l\}`$ are complementary sets, in which case the equality implies $`\lambda _i\lambda _j=\pm 1`$. At least one of $`\lambda _1\lambda _j`$, $`2j4`$ is neither $`1`$ nor $`1`$, so there must be at least four elements in the set $`\{\lambda _1\lambda _2,\mathrm{},\lambda _3\lambda _4\}`$. If
$$\lambda _1=\lambda _2=\lambda _3\lambda _4=\lambda _5=\lambda _6,$$
and $`i3`$, then
$$\{\lambda _1^j\lambda _4^{3j}\lambda _7\mathrm{}\lambda _{3+i}0j3\}$$
contains a non-constant 4-term geometric progression in the spectrum of $`\rho (t)`$, contrary to hypothesis. If
$$\lambda _1=\lambda _2\lambda _3=\lambda _4\lambda _5\lambda _1,$$
then
$$\begin{array}{cc}\hfill \{\lambda _1^2\lambda _2\lambda _6\mathrm{}\lambda _{2+i},\lambda _1\lambda _2^2\lambda _6\mathrm{}\lambda _{2+i},& \lambda _1\lambda _2\lambda _3\lambda _6\mathrm{}\lambda _{2+i},\hfill \\ & \lambda _1^2\lambda _3\lambda _6\mathrm{}\lambda _{2+i},\lambda _2^2\lambda _3\lambda _6\mathrm{}\lambda _{2+i}\}\hfill \end{array}$$
contains at least four distinct elements unless $`\lambda _1\lambda _3=\lambda _2^2`$ and $`\lambda _2\lambda _3=\lambda _1^2`$, in which case it does not satisfy (1.1). The remaining possibilities are that $`t`$ has two distinct eigenvalues, one of multiplicity 1; two distinct eigenvalues, one of multiplicity 2; two distinct eigenvalues of arbitrary multiplicity, and $`i`$ (or $`ni`$) is $`2`$; three distinct eigenvalues, two of them of multiplicity 1; or three distinct eigenvalues of arbitrary multiplicity, and $`i`$ (or $`ni`$) is $`1`$.
These give rise to cases (1), (2), (3), (4), and (5) respectively. If $`\lambda =\varpi _i+\varpi _j`$, $`ij`$, is among the weights appearing in $`V_\varpi `$, then $`\varpi _{i1}+\varpi _{j+1}`$ also appears, where we define $`\varpi _0=\varpi _n=0`$. Thus if $`\varpi =\varpi _i+\varpi _j`$, $`ij`$, then either $`\varpi _{i+j}`$, $`\varpi _{2nij}`$ or $`\varpi _1+\varpi _{n1}`$ is among the weights of $`V_\varpi `$, as $`i+j`$ is less than, greater than, or equal to $`n`$.
First we consider the case $`i+j=n`$. If $`t`$ has three distinct eigenvalues $`\lambda _1,\lambda _2,\lambda _3`$, then
$$|\{\lambda _1/\lambda _2,\lambda _2/\lambda _1,\lambda _1/\lambda _3,\lambda _3/\lambda _1,\lambda _2/\lambda _3,\lambda _3/\lambda _2\}|3$$
implies that the set violates (1.1) with $`n=3`$. Thus, $`t`$ has exactly two eigenvalues $`\lambda _1`$ and $`\lambda _2`$. If $`i2`$ and $`\lambda _1`$ and $`\lambda _2`$ each occurs with multiplicity $`2`$, then
$$\{\lambda _1^2/\lambda _2^2,\lambda _1/\lambda _2,1,\lambda _2/\lambda _1,\lambda _2^2/\lambda _1^2\}$$
is contained in the spectrum of $`\rho (t)`$ since the Weyl orbits of $`\varpi _1+\varpi _{n1}`$ and $`\varpi _2+\varpi _{n2}`$ are subsets of the weights of $`V_\varpi `$. As $`\lambda _1\lambda _2`$, either this set contains $`5`$ distinct elements or it violates (1.1). The remaining cases are (7) and the $`i+j=n`$ case of (8).
If $`i+jn`$, replacing $`V_\varpi `$ by its dual if necessary, we can assume that $`i+j<n`$. If $`3i+jn3`$, then $`\varpi _{i+j}`$ is a weight of $`V_\varpi `$, so by the analysis above, $`t`$ has two eigenvalues, one with multiplicity one, and we are in case (8). If $`i+j=2`$, we see that $`2\varpi _1`$ and $`\varpi _2`$ are both weights of $`V_\varpi `$, so if $`\lambda _1,\lambda _2,\lambda _3`$ are eigenvalues of $`t`$,
$$\{\lambda _1^2,\lambda _2^2,\lambda _3^2,\lambda _2\lambda _3,\lambda _3\lambda _1,\lambda _1\lambda _2\}$$
is contained in the spectrum of $`\rho (t)`$, contrary to assumption. If there are exactly two eigenvalues, we get (6) and the $`i=j=1`$ case of (8).
If $`i+j=n2`$, then $`V_\varpi `$ contains all the weights of $`V_{\varpi _{n2}}`$, so $`t`$ may have only two eigenvalues, $`\lambda _1`$ and $`\lambda _2`$, by the analysis of the case that $`\varpi `$ is a fundamental weight, above. If each occurs with multiplicity $`2`$ and (without loss of generality) $`\lambda _1`$ occurs with multiplicity $`3`$, then
$$\{\lambda _2/\lambda _1^3,1/\lambda _1^2,1/\lambda _1\lambda _2,1/\lambda _2^2\}$$
is a $`4`$-term geometric progression contained in the spectrum of $`\rho (t)`$ contrary to hypothesis. If $`i+j=n1`$, then $`V_\varpi `$ contains all the weights of $`V_{\varpi _{n1}}`$. If $`\lambda _1`$ and $`\lambda _2`$ are eigenvalues of $`t`$ of multiplicity $`2`$, then the spectrum of $`\rho (t)`$ contains the 4-term geometric progression
$$\{\lambda _2/\lambda _1^2,1/\lambda _1,1/\lambda _2,\lambda _1/\lambda _2^2\}.$$
If $`t`$ has three distinct eigenvalues $`\lambda _1,\lambda _2,\lambda _3`$, then the spectrum of $`\rho (t)`$ contains
$$\{1/\lambda _1,1/\lambda _2,1/\lambda _3,\lambda _1/\lambda _2\lambda _3,\lambda _2/\lambda _1\lambda _3,\lambda _3/\lambda _1\lambda _2\}$$
which either violates the no-cycle condition or contains more than $`3`$ elements. It follows that $`t`$ has exactly two eigenvalues, one of multiplicity $`n1`$. So all of these possibilities are subsumed in case (8).
###### Theorem 3.3.
If $`(G,V)`$ is an indecomposable pair satisfying the $`3`$-eigenvalue property, $`\mathrm{\Phi }`$ denotes the root system of the derived group $`D`$ of $`G^{}`$, and $`\varpi `$ the highest weight of $`V`$, then $`(\mathrm{\Phi },\varpi )`$ is either one of the pairs enumerated in Proposition 3.1 or one of the following:
1. $`(A_r,\varpi _i+\varpi _j)`$, $`1ijr`$.
2. $`(B_r,\varpi _i)`$, $`1ir1`$.
3. $`(B_r,2\varpi _r)`$.
4. $`(C_r,\varpi _i)`$, $`2ir`$.
5. $`(C_r,2\varpi _1)`$.
6. $`(D_r,\varpi _i)`$, $`2ir2`$.
7. $`(D_r,\varpi )`$, $`\varpi \{2\varpi _{r1},\varpi _{r1}+\varpi _r,2\varpi _r\}`$.
8. $`(E_6,\varpi _i)`$, $`i=1,3,6`$.
9. $`(E_7,\varpi _i)`$, $`i=1,7`$.
10. $`(F_4,\varpi _4)`$.
11. $`(G_2,\varpi _2)`$.
If there exists a generating element with three eigenvalues which do not form a geometric progression, then $`(\mathrm{\Phi },\varpi )`$ is $`(A_r,\varpi _1)`$ or $`(A_r,\varpi _r)`$.
###### Proof.
By Proposition 2.4, the root system is simple and if $`\varpi =_ia_i\varpi _i`$ and the highest root is $`_ib_i\alpha _i`$, then $`a_ib_i2`$. By \[Bo, Planches\], this reduces the possibilities to those listed, together with:
1. $`(D_r,\varpi )`$, $`\varpi \{2\varpi _1,\varpi _1+\varpi _{r1},\varpi _1+\varpi _r\}`$.
2. $`(E_6,\varpi )`$, $`\varpi \{2\varpi _1,\varpi _2,\varpi _5,2\varpi _6,\varpi _1+\varpi _6\}`$.
3. $`(E_7,\varpi )`$, $`\varpi \{\varpi _2,\varpi _6,2\varpi _7\}`$.
4. $`(E_8,\varpi )`$, $`\varpi \{\varpi _1,\varpi _8\}`$
5. $`(F_4,\varpi _1)`$.
To see that the classical cases (1)–(7) above are achieved, we let $`G=D`$ and $`V`$ the indicated representation, and we choose the generating element as follows. For $`A_r`$, we let $`g`$ be the image of the diagonal element $`\mathrm{diag}(\lambda ^r,\lambda ,\mathrm{},\lambda )\mathrm{SU}(r+1)`$ in $`G`$. For $`B_r`$, we let $`g`$ denote the image of an element in $`\mathrm{Spin}(2r+1)`$ whose image in $`\mathrm{SO}(2r+1)`$ is $`\mathrm{diag}(\lambda ,\lambda ^1,1,\mathrm{},1)`$ For $`C_r`$, we let $`g`$ denote the image of the element $`(\lambda ,\lambda ^1,1\mathrm{},1)`$ in $`\mathrm{Sp}(2r)`$. For $`D_r`$, we let $`g`$ denote the image of an element in $`\mathrm{Spin}(2r)`$ whose image in $`\mathrm{SO}(2r)`$ is $`\mathrm{diag}(\lambda ,\lambda ^1,1,\mathrm{},1)`$.
Next we show that the excluded cases (12)–(16) above do not occur. For $`D_r`$, we consider an element $`g`$ whose image in $`\mathrm{SO}(2r)`$ has eigenvalues $`\lambda _1^{\pm 1}`$, $`\mathrm{}`$, $`\lambda _r^{\pm 1}`$. In $`V_{2\varpi }`$, the eigenvalues of $`g`$ are $`\lambda _i^{\pm 2}`$, $`\lambda _i^{\pm 1}\lambda _j^{\pm 1}`$, and 1. It is easy to see these represent at least 5 distinct values. A similar analysis rules out the remaining cases in (12).
For $`E_6`$ and $`F_4`$ we use the existence of equal rank semisimple subgroups of the form $`A_2^k`$. As these subgroups share a maximal torus with their ambient groups, every generating element $`g`$ can be conjugated into the subgroup. We use the branching rules tabulated in \[MP\] to compute the restrictions of $`G`$-representations via $`\mathrm{SU}(3)^kG`$; since the center of $`\mathrm{SU}(3)^k`$ has exponent $`3`$, and since we know that there are no $`2`$-eigenvalue solutions for $`F_4`$ and $`E_6`$, there can be no $`3`$-eigenvalue solutions coming from central elements of $`\mathrm{SU}(3)^k`$ and satisfying (1.1). If $`M(\lambda )\mathrm{SU}(3)`$ has eigenvalues $`\lambda ,\lambda ,\lambda ^2`$, then $`M(\lambda )\times M(\lambda ^1)`$ maps to an element of $`F_4`$ which has eigenvalues $`\lambda ^3,1,\lambda ^3`$ for $`V_{\varpi _4}`$. The restriction of $`F_4`$ to $`\mathrm{SU}(3)^2`$ is
$$V_{2\mu _2}V_{\mu _1}V_{2\mu _1}V_{\mu _2}V_{\mu _1+\mu _2}V_0V_0V_{\mu _1+\mu _2};$$
the image of any element non-central in both factors has at least four eigenvalues from the first summand; the image of any element central in the second factor but not the first has at least four eigenvalues from the first two summands; the image of any element central in the first factor but not in the second has at least four eigenvalues or the eigenvalues $`\{1,e^{\pm 2\pi i/3}\}`$ from the first two summands. For $`(E_6,\varpi _1)`$, the image of $`M(\lambda )\times M(\lambda )\times 1`$ has eigenvalues $`\{\lambda ^2,\lambda ,\lambda ^4\}`$, and it is not difficult to see that this is essentially the only way to get three eigenvalues. For $`(E_6,\varpi _2)`$, the image of $`M(\lambda )\times M(\lambda )\times 1`$ has eigenvalues $`\{\lambda ^3,1,\lambda ^3\}`$. To see that the excluded cases (13) do not give solutions to the $`3`$-eigenvalue problem, we note that
$$V_{\varpi _2}|_{A_2^3}=V_{\mu _1+\mu _2}V_{\mu _2}V_{\mu _2}V_{\mu _2}V_{\mu _1+\mu _2}V_{\mu _1}\mathrm{};$$
$$V_{2\varpi _1}|_{A_2^3}=V_{\mu _1+\mu _2}V_{\mu _2}V_{\mu _2}V_{\mu _2}V_{\mu _1+\mu _2}V_{\mu _1}\mathrm{};$$
$$V_{\varpi _1+\varpi _6}|_{A_2^3}=V_{2\mu _1}V_{\mu _1}V_{\mu _2}V_{\mu _1}V_{2\mu _1}V_{\mu _1}\mathrm{}.$$
These summands are already enough to guarantee that if $`(E_6,\varpi _2)`$, $`(E_6,2\varpi _1)`$, or $`(E_6,\varpi _1+\varpi _6)`$ satisfies the $`3`$-eigenvalue condition, any generating element in $`A_2^3`$ must be central in two of the three factors and have eigenvalues $`\lambda ,\lambda ,\lambda ^2`$, $`\lambda ^31`$, in the third. However, if $`\omega ^3=1`$, neither
$$\{\lambda ^3,1,\lambda ^3,\omega \lambda ,\omega \lambda ^2\}$$
nor
$$\{\lambda ^2,\lambda ^1,\lambda ^4,\omega \lambda ,\omega \lambda ^2\}$$
can have order $`3`$ and satisfy the no-cycle property.
For $`E_n`$, $`n7`$, we use the equal rank subgroups $`A_n`$. Again, \[MP\] gives the restriction of $`V_\varpi `$ to $`\mathrm{SU}(n+1)`$. The following table lists all irreducible components of these restrictions for all possible $`\varpi `$. It also specifies the eigenvalues in $`V_\varpi `$ for the image of the scalar matrix $`\zeta I`$ and the matrix $`M(\lambda )`$: $`\mathrm{\Phi }`$ $`\varpi `$ $`\{\mu _i\}`$ $`\zeta I`$ e-values $`M(\lambda )`$ e-values $`E_7`$ $`\varpi _1`$ $`\mu _1+\mu _7`$, $`\mu _4`$ $`\pm 1`$ $`\lambda ^8`$, $`\lambda ^4`$, $`1`$, $`\lambda ^4`$, $`\lambda ^8`$ $`E_7`$ $`\varpi _2`$ $`\mu _1+\mu _5`$, $`\mu _3+\mu _7`$, $`2\mu _1`$, $`2\mu _7`$ $`\pm i`$ $`\lambda ^{14}`$, $`\lambda ^{10}`$, $`\lambda ^6`$, $`\lambda ^2`$, $`\lambda ^2`$, $`\lambda ^6`$, $`\lambda ^{10}`$, $`\lambda ^{14}`$ $`E_7`$ $`\varpi _6`$ $`\mu _1+\mu _3`$, $`\mu _5+\mu _7`$, $`\mu _1+\mu _7`$, $`\mu _2+\mu _6`$ $`\pm 1`$ $`\lambda ^{12}`$, $`\lambda ^8`$, $`\lambda ^4`$, $`1`$, $`\lambda ^4`$, $`\lambda ^8`$, $`\lambda ^{12}`$ $`E_7`$ $`\varpi _7`$ $`\mu _2`$, $`\mu _6`$ $`\pm i`$ $`\lambda ^6`$, $`\lambda ^2`$, $`\lambda ^2`$, $`\lambda ^6`$ $`E_7`$ $`2\varpi _7`$ $`0`$, $`\mu _4`$, $`\mu _2+\mu _6`$, $`2\mu _2`$, $`2\mu _6`$ $`\pm 1`$ $`\lambda ^{12}`$, $`\lambda ^8`$, $`\lambda ^4`$, $`1`$, $`\lambda ^4`$, $`\lambda ^8`$, $`\lambda ^{12}`$ $`E_8`$ $`\varpi _1`$ $`\mu _3`$, $`\mu _6`$, $`\mu _1+\mu _8`$ $`1,e^{\pm 2\pi i/3}`$ $`\lambda ^9`$, $`\lambda ^6`$, $`\lambda ^3`$, $`1`$, $`\lambda ^3`$, $`\lambda ^6`$, $`\lambda ^9`$ $`E_8`$ $`\varpi _8`$ $`\mu _1+\mu _2`$, $`\mu _1+\mu _5`$, $`\mu _1+\mu _8`$, $`\mu _2+\mu _7`$, $`\mu _4+\mu _8`$, $`\mu _7+\mu _8`$ $`1,e^{\pm 2\pi i/3}`$ $`\lambda ^{15}`$, $`\lambda ^{12}`$, $`\lambda ^9`$, $`\lambda ^6`$, $`\lambda ^3`$, $`1`$, $`\lambda ^3`$, $`\lambda ^6`$, $`\lambda ^9`$, $`\lambda ^{12}`$, $`\lambda ^{15}`$
It follows that neither scalar matrices nor matrices of the form $`M(\lambda )`$ give rise to $`3`$-eigenvalue solutions. By Lemma 3.2, the only possible solutions to the $`3`$-eigenvalue problem for $`E_7`$ and $`E_8`$ are the pairs $`(E_7,\varpi _1)`$, $`(E_7,\varpi _7)`$, and $`(E_8,\varpi _1)`$. For the first, an element of $`\mathrm{SU}(8)`$ with eigenvalues $`\lambda `$, $`\lambda `$, $`\lambda `$, $`\lambda `$, $`\lambda `$, $`\lambda `$, $`\lambda ^3`$, $`\lambda ^3`$ maps to an element of $`E_7`$ with eigenvalues $`\lambda ^4`$, $`1`$, $`\lambda ^4`$. For the second, an element of $`\mathrm{SU}(8)`$ with eigenvalues $`\lambda `$, $`\lambda `$, $`\lambda `$, $`\lambda `$, $`\lambda ^1`$, $`\lambda ^1`$, $`\lambda ^1`$, $`\lambda ^1`$ maps to an element of $`E_7`$ with eigenvalues $`\lambda ^2`$, $`1`$, $`\lambda ^2`$. For $`(E_8,\varpi _1)`$, the only possibility is an element of $`\mathrm{SU}(9)`$ with $`\lambda _1`$ of multiplicity $`7`$ and $`\lambda _2`$ of multiplicity $`2`$. This maps to an element of $`E_8`$ with two three-term geometric progressions of eigenvalues: $`\lambda _1^3`$, $`\lambda _1^2\lambda _2`$, $`\lambda _1\lambda _2^2`$; and $`\lambda _1\lambda _2^1`$, $`1`$, $`\lambda _1^1\lambda _2`$. To have three eigenvalues in all, we must have $`\lambda _1^3=\lambda _1\lambda _2^1`$, which together with $`\lambda _1^7\lambda _2^2=1`$ implies that the eigenvalues are all equal, which we have already seen is not a possibility.
The case of $`G_2`$ is trivial.
When there are three eigenvalues not in geometric progression, the representations cannot be self-dual, and if $`\varphi =A_r`$, then $`\varpi \{\varpi _1,\varpi _r\}`$ by Lemma 3.2. The only remaining cases for which $`V_\varpi `$ is not self-dual are $`(D_r,V_{2\varpi _{r1}})`$ and its dual (when $`r`$ is odd) and $`(E_6,\varpi _1)`$ and its dual. In the first case, as $`r`$ is odd, the Weyl orbit of $`\varpi _1`$ lies in the set of weights of both $`V_{2\varpi _{r1}}`$ and $`V_{2\varpi _r}`$. The eigenvalues contributed by these weights come in mutually inverse pairs; if there are $`3`$ but not three in geometric progression, then there must be two: $`\lambda `$ and $`\lambda ^1`$, which are distinct from one another. Then the Weyl orbit of $`\varpi _3`$ also lies in the set of weights of $`V_\varpi `$, so $`\lambda ^3,\lambda ,\lambda ^1,\lambda ^3`$ are all eigenvalues of $`\rho (t)`$, which is absurd. In the second case, restricting from $`E_6`$ to $`\mathrm{SU}(6)\times \mathrm{SU}(2)`$, we get
$$V_{\varpi _1}V_{\varpi _1}V_{\varpi _4}V_0.$$
The second summand contributes $`2`$ eigenvalues or 3 eigenvalues not in geometric progression, so an inverse image $`(g_1,g_2)`$ of the generating element must be a scalar $`\zeta `$ in $`\mathrm{SU}(6)`$ (and therefore $`\zeta ^6=1`$). The eigenvalues of $`g`$ in the first summand are $`\{\zeta \lambda ,\zeta ^4,\zeta \lambda ^1\}`$ which are in geometric progression, contrary to assumption.
## 4. The asymptotic $`N`$-eigenvalue condition
In this section we consider what can be said when the eigenvalues of a generating element are sufficiently general. One hypothesis which is strong enough for our purposes is that the eigenvalues are distinct $`r`$th roots of unity where $`r`$ is a sufficiently large prime. We consider a somewhat more general condition.
###### Proposition 4.1.
Let $`TU(1)^d`$ be a torus and $`U`$ an open neighborhood of the identity in $`T`$. There exists a finite set $`S`$ of characters $`\chi :TU(1)`$ and an integer $`m`$ such that if $`n`$ is a positive integer and $`tT`$ an $`n`$-torsion point, at least one of the following must be true:
1. There exists $`\chi S`$ such that $`\chi (t)1`$ has order $`m`$.
2. There exists an integer $`k`$ relatively prime to $`n`$ such that $`t^kU`$.
###### Proof.
We use induction on dimension, the proposition being trivial in dimension 0.
By Urysohn’s lemma there exists a continuous function $`f:T[0,1]`$ such that $`f(x)=0`$ for $`xU`$ and $`f(x)=1`$ in some neighborhood of the identity. It is well-known (see, e.g. \[SW, VII Th. 1.7\]) that finite linear combinations of characters are dense in the $`L^{\mathrm{}}`$ norm on the set of continuous functions on $`T`$. It follows that there exists a real-valued finite sum $`f(x):=_{\chi S}a_\chi \chi (x)`$ such that $`f(x)<0`$ for all $`xTU`$ and $`a_0=f(x)𝑑x>0`$. Enlarging $`S`$ if necessary, we may assume without loss of generality that if $`n\chi S`$ for some positive integer $`n`$, then $`\chi S`$.
Suppose $`\chi (t)=1`$ for some non-trivial character $`\chi S`$. Let $`\lambda S`$ denote a primitive character in $`S`$ and $`k`$ a positive integer such that $`\chi =k\lambda `$. If $`m`$ is taken greater than the value of $`k`$ associated with any character in $`S`$, either (1) is satisfied or $`\lambda (t)=1`$. As $`\lambda `$ is primitive, $`\mathrm{ker}\lambda `$ is a subtorus of $`T`$. As there are only finitely many subtori arising in this way, the proposition follows by induction.
We may therefore assume that the order of $`\chi (t)`$ is greater than $`m`$ for each $`\chi S`$. We have
$$\underset{\{k[0,n](k,n)=1\}}{}f(t^k)=a_0\varphi (n)+\underset{\chi S\{0\}}{}a_\chi \underset{\{k[0,n](k,n)=1\}}{}\chi (t^k).$$
If $`n_\chi `$ is the order of $`\chi (t)`$, then
$$\underset{\{k[0,n](k,n)=1\}}{}\chi (t^k)=\frac{\varphi (n)}{\varphi (n_\chi )}\underset{\{k[0,n_\chi ](k,n_\chi )=1\}}{}\chi (t^k)=\frac{\mu (n)\varphi (n)}{\varphi (n_\chi )}.$$
Choosing $`m`$ large enough that for all $`n_\chi >m`$,
$$\underset{\chi S\{0\}}{}|a_\chi |\varphi (n_\chi )a_0,$$
we conclude that
$$\underset{\{k[0,n](k,n)=1\}}{}f(t^k)0$$
and therefore that $`t^kU`$ for some $`k`$ prime to $`n`$. ∎
###### Theorem 4.2.
For every integer $`N2`$ there exists an integer $`m`$ such that if $`(G,V)`$ satisfies the $`N`$-eigenvalue property with a generator $`g`$ with eigenvalues $`\lambda _1,\mathrm{},\lambda _N`$, and $`G`$ is finite modulo its center, then the group $`\lambda _i\lambda _j^1`$ generated by ratios of eigenvalues of $`\rho (g)`$ contains a non-trivial root of unity of order less than $`m`$.
###### Proof.
If $`G`$ is finite modulo its center and acts irreducibly on $`V`$, then either $`G^{}`$ is trivial or it consists of all scalars of absolute value 1. In the latter case, we can replace $`g`$ by $`det(g)^{1/dim(V)}g`$ for any choice of root, and the resulting conjugacy class still satisfies the $`N`$-eigenvalue property, generates a subgroup of $`G\mathrm{SL}(V)`$ (which is finite), and determines the same group of eigenvalue ratios $`\lambda _i\lambda _j^1`$. Without loss of generality, therefore, we may assume $`G`$ is finite.
Any automorphism of $``$ determines an automorphism of the abstract group $`\mathrm{GL}_n()`$ for each $`n`$. Consider the quotient $`T=U(1)^n/U(1)`$ of the diagonal unitary matrices by the unitary scalar matrices. Let $`UT`$ denote the image of $`A^n`$ in $`T`$, where $`A`$ is the arc from $`\pi /6`$ to $`\pi /6`$, and let $`n`$ be the order of the group generated by the eigenvalues of $`g`$. We apply Proposition 4.1 to obtain $`m`$ large enough that our hypotheses imply the existence of a field automorphism $`\sigma `$ of $``$ such that all the eigenvalues of $`\sigma (\rho (g))`$ lie in an arc of length $`\pi /3`$ on the unit circle. By \[Bl, Theorem 8\], this implies that the representation representation $`\sigma \rho `$ is imprimitive. As the conjugacy class of $`g`$ generates $`G`$, the element $`g`$ itself must satisfy the hypothesis of Lemma 2.1, and therefore the spectrum of $`\sigma (\rho (g))`$ does not satisfy (1.1). As this property is stable under Galois action, the spectrum of $`\rho (g)`$ fails to satisfy (1.1), contrary to hypothesis. ∎
###### Corollary 4.3.
For every integer $`N2`$ there exists an integer $`m`$ such that if $`(G,V)`$ satisfies the $`N`$-eigenvalue property with a generator $`g`$ of prime order $`r`$, then $`r<m`$ or $`G`$ is infinite modulo its center.
We remark that it is probably possible to prove a stronger version of this corollary, in which a good bound is given for $`m`$, using \[Z\] as a starting point.
## 5. Application to Hodge-Tate theory
Let $`\overline{}_p`$ be an algebraic closure of $`_p`$, and $`_p`$ denote the completion of $`\overline{}_p`$. Let $`K`$ and $`L`$ be subfields of $`\overline{}_p`$ finite over $`_p`$, and let $`\mathrm{\Gamma }_K:=\mathrm{Gal}(\overline{}_p/K)`$. Let $`V_LL^d`$ be a finite-dimensional $`L`$-vector space and $`\rho _L:\mathrm{\Gamma }_K\mathrm{GL}(V_L)`$ a continuous representation. Then $`\mathrm{\Gamma }_K`$ acts on both factors of $`V__p:=V_L_L_p`$. The representation is said to be *Hodge-Tate* if $`V__p`$ decomposes as a direct sum of factors $`V_{i_p}`$ such that $`\mathrm{\Gamma }_K`$ acts on $`V_i`$ through the $`i`$th tensor power of the cyclotomic character. If $`X`$ is a complete non-singular variety over $`K`$ and $`\overline{X}`$ is obtained from $`X`$ by extending scalars to $`\overline{}_p`$, then $`V_L:=H^k(\overline{X},L)`$ is Hodge-Tate for all non-negative integers $`k`$, and the factors $`V_{i_p}`$ are non-zero only if $`0ik`$ (\[Fa\]).
Let $`G_L`$ denote the Zariski-closure of the image of $`\rho _L(\mathrm{\Gamma }_K)`$ in $`\mathrm{GL}_d`$. By the axiom of choice, any two uncountable algebraically closed fields of characteristic zero whose cardinalities are the same are isomorphic. Therefore, $`_p`$, and extending scalars, we can view $`G_{}`$ as a complex algebraic group. Let $`G`$ denote a maximal compact subgroup of $`G_{}`$. The inclusion $`G_{}\mathrm{GL}(V_{})`$ gives $`G`$ a complex representation which we denote $`(\rho ,V)`$. If $`\rho _L`$ is absolutely irreducible, then $`V_{}`$ is an irreducible representation of $`G_{}`$ and therefore of $`G`$.
Although $`G_L`$ need not be connected, by passing to a finite extension $`K^{}`$ of $`K`$ (i.e., replacing $`\mathrm{\Gamma }_K`$ by a normal open subgroup) we can replace $`G_L`$ by its identity component. Therefore, in trying to understand what Lie algebras and Lie algebra representations can arise from Hodge-Tate structures with specified weights, without loss of generality we may assume that $`G_L`$ is connected.
###### Definition 5.1.
Let $`G_{}`$ be a connected reductive algebraic group over $``$, and $`V`$ a faithful complex representation of $`G_{}`$. We say that $`(G_{},V)`$ is of *$`N`$-eigenvalue type* if for every almost simple normal subgroup $`H_{}`$ of $`G_{}`$ and every irreducible factor $`W`$ of $`V|_H_{}`$, the image of $`H_{}`$ in $`\mathrm{GL}(W)`$ satisfies the $`N_W`$-eigenvalue property for some $`N_WN`$.
###### Lemma 5.2.
Let $`G_{}`$ be a connected reductive complex Lie group and $`(\rho ,V)`$ a faithful representation. Let $`g_iG_{}`$ be semisimple elements generating a Zariski-dense subgroup of $`G_{}`$, such that the spectrum of $`\rho (g_i)`$ has $`N`$ eigenvalues satisfying the no-cycle condition. Then $`(G_{},V)`$ is of $`N`$-eigenvalue type.
###### Proof.
Let $`D_{}`$ denote the derived group of $`G_{}`$. The universal cover $`\stackrel{~}{D}_{}`$ factors into simply connected, almost simple complex groups $`G_j`$. Every irreducible factor $`W`$ of $`V`$ restricts to an irreducible representation of $`\stackrel{~}{D}_{}`$ which decomposes as $`W_1\mathrm{}W_k`$, where $`W_j`$ is an irreducible representation of $`G_j`$.
Each $`g_i`$ in our generating set factors as $`d_iz_i`$, where $`z_i`$ lies in the center of $`G_{}`$. We choose $`\stackrel{~}{d}_i\stackrel{~}{D}_{}`$ lying over $`d_i`$, and let $`g_{ij}`$ denote the $`G_j`$ coordinate of $`\stackrel{~}{d}_i`$. For each $`j`$, there exists $`W`$ such that $`W_j`$ is non-trivial and $`i`$ such that $`g_{ij}`$ does not lie in the center of $`G_j`$. As $`g_i`$ is semisimple, the same is true of $`d_i`$ and therefore $`\stackrel{~}{d}_i`$ and therefore $`g_{ij}`$. Moreover, it has at most $`N`$ eigenvalues on $`W_j`$ and they satisfy (1.1), since if $`S`$ and $`T`$ are sets of complex numbers and the product set satisfies (1.1), then $`|S|,|T||ST|`$, and $`|S|`$ and $`|T|`$ satisfy (1.1). As $`g_{ij}`$ is not in the center of $`G_j`$, the conjugacy class of $`\rho _j(g_{ij})`$ generates a non-central normal subgroup of the almost simple group $`\rho _j(G_j)`$ and therefore generates the whole group.
###### Theorem 5.3.
If $`V_L`$ is an absolutely irreducible Hodge-Tate representation of $`G_K`$ with $`N`$ distinct weights, then $`(G_{}^{},V)`$ is of $`N`$-eigenvalue type.
###### Proof.
The grading of $`V_{}`$ which assigns $`V_i`$ degree $`i`$ uniquely determines a cocharacter $`h:𝔾_mG_{}`$ such that $`\rho h`$ acts isotypically on $`V_i`$ by the $`i`$th power character. By \[Sen\], $`G_L^{}`$ is the smallest $`L`$-algebraic subgroup of $`\mathrm{GL}_d`$ which contains $`h(𝔾_m)`$. Thus $`\{h^\sigma (𝔾_m)\sigma \mathrm{Aut}_L()\}`$ generates $`G_{}^{}`$. If $`u^\times `$ is of infinite order, then any element $`g_jh^{\sigma _j}(u)`$ (Zariski-topologically) generates $`h^{\sigma _j}(𝔾_m)`$. Together, the $`g_j`$ generate $`G_{}^{}`$. There are exactly $`N`$ distinct eigenvalues of $`\rho (g_j)`$ and they satisfy the no-cycle condition. The theorem now follows from Lemma 5.2. ∎
###### Theorem 5.4.
Assume that the Fontaine-Mazur conjecture \[FM, Conj. 5a\] holds. If $`X`$ is a complete non-singular variety over a number field $`K`$, $`k`$ is a non-negative integer, $`G_{}`$ is the complexification of the Zariski closure of $`\mathrm{Gal}(\overline{K}/K)`$ in $`\mathrm{Aut}(H^k(\overline{X},_p))`$, and $`V=\mathrm{Aut}(H^k(\overline{X},_p))__p`$, then $`(G_{}^{},V)`$ is of $`k`$-eigenvalue type.
###### Proof.
As $`X`$ has good reduction over $`K`$, there exists a rational integer $`M`$ such that $`X`$ is the generic fiber of a smooth proper scheme $`𝒳`$ over $`𝒪_K[1/M]`$, where $`𝒪_K`$ is the ring of integers of $`K`$. Thus, the homomorphism $`\mathrm{Gal}(\overline{K}/K)\mathrm{Aut}(H^k(\overline{X},_p))`$ factors through $`\rho :\mathrm{\Gamma }_{K,Mp}\mathrm{GL}_n(_p)`$, the Galois group over $`K`$ of the maximal subfield of $`\overline{K}`$ unramified over any prime of $`𝒪_K`$ not dividing $`Mp`$.
For each prime $`v`$ of $`𝒪_K`$ dividing $`Mp`$, we fix an embedding $`\overline{K}\overline{K}_v`$ and therefore an embedding $`\mathrm{\Gamma }_{G_v}\mathrm{\Gamma }_{K,Mp}`$. Let $`G`$, regarded as an algebraic group over $`_p`$, be the Zariski-closure of $`\rho (\mathrm{\Gamma }_{K,Mp})`$ in $`\mathrm{GL}_n`$, $`G_v`$ the Zariski-closure of $`\rho (\mathrm{\Gamma }_{G_v})`$, and $`G_p`$ the normal subgroup of $`G`$ generated by $`G_v^{}`$ for all $`v`$ lying over $`p`$. Replacing $`K`$ by a finite extension, we may assume that $`G_v`$ is connected for all such $`v`$, so $`G_p`$ is generated by conjugates of the $`G_v`$. By Theorem 5.3, the complexification $`G_p`$, together with its natural $`n`$ dimensional representation, is of $`k`$-eigenvalue type. If $`G_p`$ is of finite index in $`G`$, the theorem follows. Otherwise, there exists a homomorphism $`\mathrm{\Gamma }_{K,Mp}G(_p)/G_p(_p)`$ with Zariski-dense, and therefore infinite $`p`$-adic analytic image. By construction, this homomorphism is unramified at all primes over $`v`$. Such a homomorphism cannot exist according to the Fontaine-Mazur conjecture.
###### Corollary 5.5.
If the Fontaine-Mazur conjecture is true, then for every complex non-singular variety $`X`$ over a number field $`K`$, the Zariski closure of the image of $`\mathrm{Gal}(\overline{K}/K)`$ in $`\mathrm{Aut}(H^2(\overline{X},_p))`$ has no factor of type $`E_8`$.
## 6. Application to braid group representations
Artin’s braid group $`_m`$ is generated by $`\sigma _1,\mathrm{},\sigma _{m1}`$ subject to relations
$$\sigma _i\sigma _j=\sigma _j\sigma _i\text{if }|ij|2,\sigma _i\sigma _{i+1}\sigma _i=\sigma _{i+1}\sigma _i\sigma _{i+1}\text{ for }1im1.$$
In \[FLW\], the closed images of the unitary $`q=e^{2\pi i/\mathrm{}}`$ Hecke algebra representations of the braid groups are completely analyzed (completing a program initiated by Jones) for $`\mathrm{}5`$ and $`\mathrm{}6`$. In this section, we will carry out a similar analysis. We also discuss the situations in which the braid group representations arising from quantum groups at roots of unity satisfy the 3-eigenvalue condition.
### 6.1. Set-up
Given an irreducible unitary representation $`(\rho ,V)`$ of $`_m`$ there are three distinct possibilities for $`G=\overline{\rho (_m)}`$
1. $`G/Z(G)`$ is finite
2. $`\mathrm{SU}(V)G`$
3. $`G/Z(G)`$ is infinite, but $`\mathrm{SU}(V)G`$.
While the first (finite group) and third (non-dense) possibilities are interesting, we will focus on the second. There are a number of reasons for doing this. Firstly, we will see that $`\mathrm{SU}(V)G`$ is the generic situation, while the other (non-dense) cases require a case-by-case analysis that we will carry out in a separate work. Also, density is crucial for applications to quantum computing—our original motivation. Lastly, the application of Theorem 3.3 leads most directly to the conclusion $`\mathrm{SU}(V)G`$, *i.e.* by showing that $`(G,V)`$ is an indecomposable pair satisfying the 3-eigenvalue property for which the three eigenvalues do not form a geometric progression. Nearly all of the finite group/non-dense examples come from pairs having eigenvalues in geometric progression which will be considered in a forthcoming paper by the first two authors. We proceed with the following program:
1. Determine which representations have exactly three eigenvalues.
2. Determine conditions for the representations from (1) to be unitary.
3. Determine when the three eigenvalues from (1) and (2) satisfy the no-cycle condition. This will give us all pairs $`(G,V)`$.
4. Determine when the three eigenvalues from (1) and (2) are not in geometric progression. Although this does not ensure density, it does guarantee the pair $`(G,V)`$ is indecomposable, as three eigenvalues coming from a decomposable pair must be in geometric progression by Lemma 2.2.
5. Determine when $`G`$ is infinite modulo the center for the cases not excluded by (1)-(4).
### 6.2. $`BMW`$-algebra representations of the braid groups
We apply the strategy outlined above to $`BMW`$-algebras, first recalling what is well-known and then proceeding to the subsequent steps.
#### 6.2.1. Definitions and combinatorial results
Most of the material here can be found in \[Wz1\], and we summarize the details germane to the problem, carrying out steps (1) and (2) in the above program.
The Birman-Wenzl-Murakami (BMW) algebras are a sequence of finite dimensional algebras equipped with Markov traces. They can be described as quotients of the group algebra $`(r,q)_m`$ of Artin’s braid group where $`r`$ and $`q`$ are complex parameters. The precise definition of the BMW-algebra $`𝒞_m(r,q)`$ is:
###### Definition 6.1.
Let $`g_1,g_2,\mathrm{},g_{m1}`$ be invertible generators satisfying the braid relations $`(B1)`$ and $`(B2)`$ above and:
1. $`(g_ir^1)(g_iq)(g_i+q^1)=0`$
2. $`e_ig_{i1}^{\pm 1}e_i=r^{\pm 1}e_i`$, where
3. $`(qq^1)(1e_i)=g_ig_i^1`$ defines $`e_i`$.
The relations (R2) can be best understood by pictures where $`g_i`$ is the the braid generator $`\sigma _i`$ and $`e_i`$ is the $`i`$-th generator of the Temperley-Lieb algebra. Relation (R1) shows that the image of $`g_i`$ in any representation of $`𝒞_m(r,q)`$ has 3 eigenvalues: $`r^1,q`$ and $`q^1`$. When $`r\pm q^n`$ and $`q`$ is not a root of unity, each $`BMW`$-algebra $`𝒞_m(r,q)`$ is finite-dimensional and semisimple with simple components labeled by Young diagrams with $`m2j0`$ boxes for $`j`$. In other words, the $`BMW`$-algebra is a direct sum of full matrix algebras. For each simple component $`𝒞_{m,\lambda }`$ let $`V_{m,\lambda }`$ be the unique non-trivial simple $`𝒞_{m,\lambda }`$-module. Then the branching rule for restricting $`V_{m,\lambda }`$ to $`𝒞_{m1}(r,q)`$ is:
$$V_{m,\lambda }\underset{\mu \lambda }{}V_{m1,\mu }$$
where $`V_{m1,\mu }`$ is a simple $`𝒞_{m1}(r,q)`$-module and $`\mu `$ is a Young diagram with $`m12j0`$ boxes obtained from $`\lambda `$ by adding/removing a box to/from $`\lambda `$. This description of inclusions among to $`BMW`$-algebras can be neatly encoded in a graph called the *Bratteli diagram*. The graph consists of vertices labelled by $`(m,\lambda )`$ with $`|\lambda |=m2k`$ arranged in rows (labelled by integers $`m`$). Vertices in adjacent rows are connected if the their labels differ by 1 in the first entry and by one box in the second. The dimension of $`V_{m,\lambda }`$ can thus be computed by adding up the dimensions of the $`V_{m1,\mu }`$ whose labels are connected to $`(m,\lambda )`$ by an edge. We obtain representations of $`_m`$ on $`_\lambda V_{m,\lambda }`$ via the map $`\sigma _ig_i𝒞_m(r,q)`$.
We are interested in obtaining unitary representations of $`_m`$ from $`BMW`$-algebras, so we must consider semisimple quotients with $`r`$ and $`q`$ specialized at roots of unity. Specifically, we let $`r=q^n`$ for $`1n`$ and $`q=e^{\pi i/\mathrm{}}`$ ($`\mathrm{}1`$), *i.e.*, a primitive $`2\mathrm{}`$th root of unity. If a given irreducible representation is unitary for $`q=e^{\pi i/\mathrm{}}`$, it will remain so for $`q=e^{\pi i/\mathrm{}}`$. For other choices of primitive roots of unity we cannot expect to have unitarity. The quotient of each specialized $`BMW`$-algebra by the annihilator of the trace $`A_m:=\{a𝒞_m(r,q):tr(ab)=0\text{for all }b\}`$ is semisimple and we denote it $`𝒞_m(q^n,q)`$ (where $`q`$ is understood to be $`e^{\pi i/\mathrm{}}`$). The branching rules and simple decomposition described above for the generic case still essentially apply to $`𝒞_m(q^n,q)`$, except that some components no longer appear, and fewer Young diagrams are needed to describe the persisting components (for all $`m`$). Precisely which components survive depends on the values $`\mathrm{}`$ and $`n`$, and the derivation can be found in \[Wz1\], the results of which we will describe below. For now it is enough to note that each simple component (sector) that does survive the quotient gives us an irreducible representation of $`_m`$. Let $`\rho _{(m,\lambda )}^{(n,\mathrm{})}`$ acting on $`V_{(m,\lambda )}^{(n,\mathrm{})}`$ be the representation of $`_m`$ corresponding to the simple component of $`𝒞_m(q^n,q)`$ labeled by $`\lambda `$. Since the conjugacy class of $`\rho _{(m,\lambda )}^{(n,\mathrm{})}(\sigma _1)`$ generates the closed image of $`_m`$ topologically, there is a chance that the pair
$$(\overline{\rho _{(m,\lambda )}^{(n,\mathrm{})}(_m)},V_{(m,\lambda )}^{(n,\mathrm{})})$$
satisfies the 3-eigenvalue property.
As a first step we need to know the conditions under which the image of $`\sigma _1_m`$ under $`\rho _{(m,\lambda )}^{(n,\mathrm{})}`$ has 3 distinct eigenvalues. The answer is well-known to experts (see \[Wz1\]): *for $`m3`$, the image of $`\sigma _1`$ under the irreducible representation $`\rho _{(m,\lambda )}^{(n,\mathrm{})}`$ has 3 distinct eigenvalues precisely when $`|\lambda |<m`$ and $`𝒞_{3,\mathrm{}}`$ is three dimensional.* This is equivalent to the requirement that the corresponding simple component $`𝒞_{m,\lambda }`$ contains the simple component $`𝒞_{3,\mathrm{}}`$. This is most easily seen by considering the Bratteli diagram as described above. It is shown in \[Wz1\] that $`𝒞_m(q^n,q)/𝒜_m_m\overline{}_m(q^2)`$ where $`\overline{}_m(q^2)`$ is a quotient of the Iwahori-Hecke algebra of type $`A_{m1}`$, and $`_m`$ is the ideal generated by $`e_{m1}`$ (see \[Wz1\]). The Young diagrams labeling simple components of $`\overline{}_m(q^2)`$ have $`m`$ boxes, whereas those of $`_m`$ have $`m2j`$ boxes for some $`j1`$. The representations of $`_m`$ corresponding to the Iwahori-Hecke algebra part of $`𝒞_m(q^n,q)`$ have been studied in \[FLW\] where they are analyzed using the solution to the 2-eigenvalue problem. Thus the image of $`\sigma _1`$ on the irreducible representation $`V_{m,\lambda }`$ ($`m3`$) has (exactly) 3 distinct eigenvalues precisely when $`|\lambda |<m`$ and $`𝒞_{3,\mathrm{}}`$ is 3-dimensional in which case the eigenvalues are $`\{q^n,q,q^1\}`$. We can eliminate many redundant cases using isomorphisms (see \[TbW2\]):
(6.1)
$$𝒞_m(q^n,q)𝒞_m(q^n,q)𝒞_m(q^n,q)𝒞_m(q^n,q^1).$$
We describe the restrictions more precisely in the following, which is a reformulation of several results in \[Wz1\] and \[R1\]. Denote by $`\lambda _i`$ (resp. $`\lambda _i^{}`$) the number of boxes in the $`i`$th row (resp. column) of the Young diagram $`\lambda `$.
###### Proposition 6.2.
Let $`q=e^{\pi i/\mathrm{}}`$ and $`m3`$.
1. The matrix algebra $`𝒞_{3,\mathrm{}}`$ is a simple 3-dimensional subalgebra of $`𝒞_m(q^n,q)`$, provided one of the following conditions holds:
1. $`n=1`$ and $`\mathrm{}3`$
2. $`n=2`$ and $`\mathrm{}4`$
3. $`3n\mathrm{}3`$ (so $`\mathrm{}6`$)
4. $`4\mathrm{}n4`$, $`n`$ is even and $`\mathrm{}`$ is odd (so $`\mathrm{}9`$)
5. $`5\mathrm{}n5`$, $`n`$ is odd and $`\mathrm{}`$ is even (so $`\mathrm{}10`$)
Moreover, this list is exhaustive up to the isomorphisms 6.1.
2. The $`\lambda `$ for which $`𝒞_{m,\lambda }`$ may appear as a simple component in some $`𝒞_m(q^n,q)`$ are in the following sets of $`(n,\mathrm{})`$-admissible Young diagrams corresponding to each of the 5 cases above:
1. $`\{[1^2]\}\{[k]:k\}`$
2. $`\{[1^3]\}\{[k],[k,1]:1k\mathrm{}1\}`$
3. $`\{\lambda :\lambda _1+\lambda _2\mathrm{}n+1\text{and}\lambda _1^{}+\lambda _2^{}n+1\}\{[\mathrm{}n+1,1^{n1}]\}`$
4. $`\{\lambda :\lambda _1+\lambda _21n\text{and}\lambda _1^{}(\mathrm{}+n1)/2\}`$
5. $`\{\lambda :\lambda _1(1n)/2\text{and}\lambda _1^{}(\mathrm{}+n1)/2\}`$
3. Thus the image of $`\sigma _1`$ under the irreducible representation $`\rho _{(m,\lambda )}^{(n,\mathrm{})}`$ with $`|\lambda |<m`$ has 3 distinct eigenvalues provided $`n`$ and $`\mathrm{}`$ satisfy one of the conditions of (1) and $`\lambda `$ is in the corresponding set of admissible Young diagrams in (2). These representations are unitary except possibly in case (d).
###### Remark 6.3.
Observe that the set in 2(a) is infinite and independent of $`\mathrm{}`$. The other four labeling sets are finite, and it is easy to see that the corresponding Bratteli diagrams are periodic. In the case $`n=2`$ there is a slight exception to the rule for constructing the Bratteli diagram: the diagrams labeled by $`[\mathrm{}1,1]`$ and $`[\mathrm{}1]`$ are *not* connected by an edge (see \[Wz1\], Prop. 6.1). The fact that the representations in (a),(b),(c) and (e) are unitary was proved in \[Wz1\]. The full (reducible) representations of $`_m`$ factoring over $`𝒞_m(q^n,q)`$ corresponding to case (d) were shown in \[R1\] to be non-unitarizable *for any $`q`$* when $`\mathrm{}>2(n+1)`$. This leaves only finitely many possible $`\mathrm{}`$ for each fixed $`n`$, and even in these cases one can use the techniques of \[R1\] to show that for $`q=e^{\pi i/\mathrm{}}`$ one does not get unitarity except in degenerate cases. Restricting to the irreducible sectors one may get unitarizable representations, but not uniformly, so that for $`m0`$ no irreducible sector is unitary.
#### 6.2.2. Cycles and geometric progressions
The eigenvalues of any of the irreducible representations satisfying the conditions of Proposition 6.2 are $`\{q,q^1,q^n\}`$, with $`q=e^{\pi i/\mathrm{}}`$. Steps (3) and (4) of the program can be accomplished with simple computations. We have:
###### Lemma 6.4.
Let $`n`$, $`\mathrm{}`$ and $`\lambda `$ be as in Proposition 6.2. Then the eigenvalues of $`\rho _{(m,\lambda )}(\sigma _1)`$:
1. fail the no-cycle property if and only if $`n=1`$ or $`(n,\mathrm{})=(3,6)`$ and
2. are in geometric progression if and only if $`n\{3,\mathrm{}3,\pm \mathrm{}/2\}`$.
###### Proof.
The only way $`\{q,q^1,q^n\}`$ can fail the no-cycle condition is if it contains a coset of $`\{\pm 1\}`$ or $`\{1,e^{2\pi i/3},e^{4\pi i/3}\}`$. With the restrictions in Prop. 6.2 that $`\mathrm{}n+2`$ for $`n>0`$ and $`\mathrm{}4n`$ for $`n<0`$ as well as $`\mathrm{}3`$ one checks that only $`n=1`$ and $`(n,\mathrm{})=(3,6)`$ fail no-cycle. For the eigenvalues to be in geometric progression (still satisfying the conditions of Prop. 6.2) we check the solutions of $`\lambda _1\lambda _2(\lambda _3)^2=0`$ for the three possible assignments of $`\lambda _3`$. These yield the three solutions for $`n`$ above. ∎
###### Remark 6.5.
All of the exceptional cases $`n\{1,3,\mathrm{}3,\pm \mathrm{}/2\}`$ will be considered in a future work. As we remarked above the case $`n=1`$ is unique in that the labelling set of irreducible sectors is infinite. In fact, it is not hard to see, using the classification of $`m`$-dimensional irreducible representations of $`_m`$ found in \[FLSV\], that one obtains some finite group images for every $`m`$ when $`n=1`$. By the isomorphisms of $`BMW`$-algebras corresponding to $`rr^1`$ we see that the two cases $`n=3`$ and $`n=\mathrm{}3`$ are actually the same. Moreover, it can be shown that the (specialized quotient) $`BMW`$-algebras $`𝒞_m(q^3,q)`$ can be embedded (diagonally) in quotients of the tensor squares of Iwahori-Hecke algebras $`_m(q^2)`$. This indicates that the corresponding pairs may be tensor decomposable. In the subcase $`(n,\mathrm{})=(3,6)`$ work of Jones in \[J1\] shows that the images are all finite groups (essentially $`\mathrm{PSL}(2m,3)`$). The case $`n=\mathrm{}/2`$ sometimes also have finite group images *e.g.* when $`(n,\mathrm{})=(5,10)`$, see \[J2\].
#### 6.2.3. Infinite images and density
Finally, we need to determine, for representations not excluded by the steps (1)-(4) above, the values of $`m`$, $`\mathrm{}`$, $`n`$, and $`\lambda `$ for which the image of $`_m`$ under the unitary irreducible representation $`\rho _{(m,\lambda )}^{(n,\mathrm{})}`$ in $`𝒞_m(q^n,q)`$ with $`q=e^{\pi i/\mathrm{}}`$ is infinite modulo the center. Proposition 6.2 implies that a sufficient condition for $`\rho _{(m,\lambda )}^{(n,\mathrm{})}`$ to have infinite image is that the 3-dimensional representation $`\rho _{(3,\mathrm{})}^{(n,\mathrm{})}`$ have infinite image. So as a first step, we study this condition. For convenience of notation we denote this representation simply by $`\rho `$ despite its dependence on the parameters. A non-unitary realization of $`\rho `$ is given by:
$$\sigma _1A:=\left(\begin{array}{ccc}\frac{1}{q^n}& \frac{q^21}{q}& 0\\ 0& \frac{q^21}{q}& i\\ 0& i& 0\end{array}\right),\sigma _2B:=\left(\begin{array}{ccc}0& 0& i\\ 0& \frac{1}{q^n}& \frac{i(q^21)}{q^{n+1}}\\ i& 0& \frac{q^21}{q}\end{array}\right)$$
found in \[BW\].
Blichfeldt \[Bl\] has determined the irreducible finite subgroups of $`\mathrm{PSL}(3,)`$. Six are primitive groups of orders 36, 60, 72, 168, 216, and 360, and the imprimitive subgroups come in two infinite families isomorphic to extensions of $`S_3`$ and $`_3`$ by abelian groups.
###### Definition 6.6.
A group $`\mathrm{\Gamma }`$ is *primitive* if $`\mathrm{\Gamma }`$ has a faithful irreducible representation which cannot be expressed as a direct sum of subspaces which $`\mathrm{\Gamma }`$ permutes nontrivially.
By Lemma 2.1, a sufficient condition for $`G=\overline{\rho (_3)}`$ to be primitive is that the spectrum of $`\rho (\sigma _1)`$ satisfies the no-cycle property. So by Lemma 6.4 the image of $`\rho `$ is only imprimitive in the excluded cases $`n=1`$ and $`(n,\mathrm{})=(3,6)`$. So we may assume that the $`G`$ is primitive. We wish to determine when $`G`$ is infinite modulo the center. By rescaling the images of the generators $`\sigma _i`$ by the cube root of the determinant of $`\rho (\sigma _i)`$ we may assume that $`G\mathrm{SL}(3,)`$, and to determine the image modulo the center it suffices to consider the projective image. Thus $`G/Z(G)\mathrm{PSL}(3,)`$, and we may apply Blichfeldt’s classification. We state his result and include some useful information about orders of elements in:
###### Proposition 6.7.
The primitive subgroups of $`\mathrm{PSL}(3,)`$ are:
1. The *Hessian* group $`H`$ of order 216 or a normal subgroup of $`H`$ of order 36 or 72. The Hessian group is the subgroup of $`A_9`$ generated by $`(124)(568)(397)`$ and $`(456)(798)`$, and has elements of order $`\{1,2,3,4,6\}`$.
2. The simple group $`\mathrm{PSL}(2,7)A_7`$ of order 168. The orders of elements are $`\{1,2,3,4,7\}`$.
3. The simple group $`A_5`$ having elements of orders $`\{1,2,3,5\}`$.
4. The simple group $`A_6`$ having elements of orders $`\{1,2,3,4,5\}`$.
Using this result we have the following:
###### Theorem 6.8.
Let $`n`$ and $`\mathrm{}`$ be chosen so that $`\rho _{(3,\mathrm{})}^{(n,\mathrm{})}`$ is a 3-dimensional unitary irreducible representation of $`_3`$ with eigenvalues not in geometric progression and satisfying the no-cycle condition. That is, $`n`$ and $`\mathrm{}`$ satisfy the hypotheses of Proposition 6.2(1)(b),(c) or (e) in addition to $`n\{3,\mathrm{}3,\pm \mathrm{}/2\}`$. Let $`m3`$ and $`|\lambda |<m`$ with $`\lambda `$ $`(n,\mathrm{})`$-admissible. The closure of the group $`\rho _{(m,\lambda )}^{(n,\mathrm{})}(_m)`$ is infinite modulo the center with two exceptions: if $`(n,\mathrm{})\{(5,14),(9,14)\}`$ with $`(m,\lambda )\{(3,\mathrm{}),(4,[0])\}`$ then the projective images are isomorphic to $`\mathrm{PSL}(2,7)`$. Excluding these cases, if the dimension of the representation $`\rho _{(m,\lambda )}^{(n,\mathrm{})}`$ is $`k`$, then the closure of the image of $`_m`$ contains $`SU(k)`$.
###### Proof.
Knowing the specific eigenvalues of $`\rho (\sigma _1)`$ we compute its projective order $`t(n,\mathrm{})`$ as a function of $`\mathrm{}`$ and $`n`$ to be:
(6.2)
$$t(n,\mathrm{})=\{\begin{array}{cc}\mathrm{}/2\hfill & \text{if }\mathrm{}2(mod4)\text{ and }n3(mod4)\hfill \\ \mathrm{}\hfill & \text{if }\mathrm{}0(mod4)\text{ and }n\text{ even or}\hfill \\ & \text{ }\mathrm{}2(mod4)\text{ and }n1(mod4)\hfill \\ 2\mathrm{}\hfill & \text{otherwise}\hfill \end{array}$$
Under the stated hypotheses on $`n`$ and $`\mathrm{}`$ we consider cases, comparing with the list of possible orders of elements in Blichfeldt’s classification.
1. If $`\mathrm{}`$ is odd, then $`\mathrm{}5`$ in which case $`t(n,\mathrm{})10`$ which is too large.
2. If $`\mathrm{}0(mod4)`$ then $`\mathrm{}=8`$ is the smallest value not yet excluded which gives $`t(n,\mathrm{})8`$ which is again too large.
3. If $`\mathrm{}2(mod4)`$ then $`\mathrm{}6`$ and $`t(n,\mathrm{})12`$ unless $`n`$ is odd. If $`n1(mod4)`$ then $`\mathrm{}10`$ which gives us $`t(n,\mathrm{})=\mathrm{}10`$ which does not appear on the list. When $`\mathrm{}2(mod4)`$ and $`n3(mod4)`$ with $`n>0`$ we must have $`n7`$ which forces $`\mathrm{}18`$ since $`n\mathrm{}/2`$. For $`\mathrm{}2(mod4)`$ and $`n3(mod4)`$ with $`n<0`$ we must have $`\mathrm{}14`$ (since $`\mathrm{}=10`$ leads to $`n=5=\mathrm{}/2`$), which has the two possible values $`n=5`$ or $`n=9`$ which we claim gives rise to finite images. Observe that $`t(5,14)=t(9,14)=7`$.
To show that the projective images for $`(5,14)`$ and $`(9,14)`$ are both $`\mathrm{PSL}(2,7)`$ we first observe that it is enough by the isomorphism of 6.1 with $`rr^1`$ and $`q^5q^{14+5}=q^9`$ so these two cases give the same images. Then we use the explicit matrices $`A`$ and $`B`$ above to define $`S=B^1`$ and $`T=BAB`$ which then (projectively) satisfy the relations $`S^7=(S^4T)^4=(ST)^3=T^2=I_{3\times 3}`$ defining $`\mathrm{PSL}(2,7)`$. It is immediate from the Bratteli diagram that the representation of $`_4`$ corresponding to $`(4,[0])`$ is irreducible and isomorphic to the that of $`(3,\mathrm{})`$ when restricted to $`_3`$. Moreover, the representations of $`_4`$ corresponding to diagrams $`[1^2]`$ and $`[2]`$ each contain the representation of $`_3`$ corresponding to the Young diagram $`[1^2,1]`$ which was shown in \[FLW\] to have infinite image (modulo the center). For all of the infinite image cases the hypotheses of Theorem 3.3 are satisfied and the eigenvalues are not in geometric progression so density follows. ∎
### 6.3. Quantum Groups
In this subsection we consider braid group actions on centralizer algebras of representations of quantum groups at roots of unity. We find and analyze examples in which we may apply Theorem 3.3. We follow the general strategy in Subsection 6.1, but we note that as the representation spaces available to us are not necessarily simple subquotients of braid group algebras (unlike $`BMW`$-algebras) there is a subtlety regarding irreducibility.
#### 6.3.1. Braid group action on centralizer algebras
The Drinfeld-Jimbo quantum group $`U:=U_q𝔤`$ associated to a simple Lie algebra $`𝔤`$ is a ribbon Hopf-algebra. The so-called *universal $`R`$-matrix* that intertwines the coproduct with the opposite coproduct on $`U`$ can be used to construct representations of the braid group $`_n`$ on the morphism space $`\mathrm{End}_U(V^n)`$ for any finite dimensional highest weight $`U`$-module $`V`$ as follows. Fix such a $`U`$-module $`V`$ and define $`\stackrel{ˇ}{R}=P_VR_{VV}\mathrm{End}_U(V^2)`$ to be the $`U`$-isomorphism afforded us by composing the image of the universal $`R`$-matrix acting on $`VV`$ with the “flip” operator $`P_V:v_1v_2v_2v_1`$. Then define isomorphisms for each $`1in1`$ :
$$\stackrel{ˇ}{R}_i:=\mathrm{𝟏}^{(i1)}\stackrel{ˇ}{R}\mathrm{𝟏}^{(ni1)}\mathrm{End}_U(V^n)$$
so that the $`\stackrel{ˇ}{R}_i`$ satisfy the braid group relations. Then define a representation of $`_n`$ on $`\mathrm{End}_U(V^n)`$ by $`\sigma _i.f=\stackrel{ˇ}{R}_if`$.
Lusztig has defined a modified form of the quantum group $`U`$ so that one may specialize the quantum parameter $`q`$ to $`e^{\pm \pi i/\mathrm{}}`$. In fact, one may choose any $`q`$ so that $`q^2`$ is a primitive $`\mathrm{}`$th root of unity, but we will restrict our attention to $`q=e^{\pi i/\mathrm{}}`$ since these values (sometimes) yield unitary representations (see \[Wz2\]), which remain unitary for $`\overline{q}`$. The full representation category of $`U`$ at roots of unity is not semisimple, but has a semisimple subquotient category. This process is essentially due to Andersen and his coauthors (see \[A\] and references therein). This yields a semisimple ribbon category $``$ (see \[T\] for the definitions) with finitely many simple objects labeled by highest weights in a truncation of the dominant Weyl chamber, called the *Weyl alcove*. The braid group still acts on $`\mathrm{End}_U(V^n)`$ for any object $`V`$ as above, and for each quantum group we look for simple objects $`V_\lambda `$ so that the images of the braid generators on the irreducible subrepresentations of $`\mathrm{End}_U(V_\lambda ^n)`$ have 3 eigenvalues. Because the tensor product rules for objects labelled by weights near the upper wall of the Weyl alcove depends on $`\mathrm{}`$, we do not explicitly determine all $`V_\lambda `$ giving rise to pairs with the 3-eigenvalue property, and restrict our attention to weights near $`0`$. As in the $`BMW`$ algebra setting, we will always have an irreducible 3-dimensional representation of $`_3`$ to which we may reduce most questions. We sketch the idea (see *e.g.* \[TbW1\] Section 3): If $`V`$ is a simple object in (a finite semisimple ribbon category) $``$ such that $`VVV_0V_1V_2`$ is the decomposition into 3 inequivalent simple objects then $`\mathrm{End}_U(V^3)`$ has a 3-dimensional irreducible subrepresentation isomorphic to $`\mathrm{Hom}_U(V^3,W)`$ for a simple object $`W`$ appearing in $`V^3`$ with multiplicity three. Provided the (categorical) $`q`$-dimension of each of $`W`$, $`V`$ and $`V_i`$ are non-zero then this representation is irreducible and the image of $`\sigma _1`$ will have three distinct eigenvalues. As in the $`BMW`$-algebra situation we can construct a Bratteli diagram encoding the containments of the semisimple finite dimensional algebras:
$$\mathrm{End}_U(V)\mathrm{End}_U(VV)\mathrm{}\mathrm{End}_U(V^n)\mathrm{}$$
The simple components of $`\mathrm{End}_U(V^n)`$ will be isomorphic to $`\mathrm{Hom}_U(V^n,V_\mu )`$ where $`V_\mu `$ is a simple object appearing in the decomposition of $`V^n`$. The edges of the Bratteli diagram are determined by decomposing $`V_\gamma V`$ where $`V_\gamma `$ is a simple subobject of $`V^{(n1)}`$. There are techniques known for obtaining these decompositions, for example Littelman’s path basis technique \[L\], or crystal bases. However, when we consider the action of the braid group $`_n`$ on the spaces $`\mathrm{Hom}_U(V^n,V_\mu )`$ we have no guarantee that the action is irreducible. This is because $`\mathrm{End}_U(V^n)`$ might not be generated by the image of $`_n`$.
#### 6.3.2. Density results
We proceed to find pairs $`(X_r,\lambda )`$ so that the ribbon category corresponding to the quantum group $`U_q𝔤(X_r)`$ of Lie type $`X_r`$ has simple object $`V_\lambda `$ with $`V_\lambda ^3V_0V_1V_2`$ as above. We find that $`(A_r,\varpi _2)`$, $`(A_r,2\varpi _1)`$, $`(B_r,\varpi _1)`$, $`(C_r,\varpi _1)`$, $`(D_r,\varpi _1)`$ and $`(E_6,\varpi _1)`$ do satisfy these conditions (where the weights $`\varpi _i`$ are labeled as in \[Bo\]). With these in hand we compute the eigenvalues of the images of $`\sigma _i`$ in the corresponding representations. We use the following result found in \[LR\], Corollary 2.22(3) originally due to Reshetikhin. The form $`,`$ is the symmetric inner product on the root lattice normalized so that the square lengths of *short* roots is 2, and the weight $`\rho `$ is the half sum of the positive roots.
###### Proposition 6.9.
Suppose that $`V=V_\varpi `$ is an irreducible representation of the quantum group $`U_q𝔤`$ and that $`VV_\lambda `$ is multiplicity free for all $`V_\lambda `$ appearing in some $`V^n`$. Then for any $`V_\nu `$ appearing in $`V^2`$ we have:
$$\stackrel{ˇ}{R}_i_{V_\nu }=\pm q^{(1/2)\nu ,\nu +2\rho \varpi ,\varpi +2\rho }\mathrm{𝟏}_{V_\nu }$$
where the sign is $`+1`$ if $`V_\nu `$ appears in the symmetrization of $`V^2`$ and $`1`$ if $`V_\nu `$ appears in the anti-symmetrization of $`V^2`$.
We record the results in Table 1, where the notation follows \[Bo\]. The symbol $`11`$ denotes the unit object in the category. The necessary computations are standard and can be done by hand *e.g.* using the technique of \[L\].
The braid group representations corresponding to Lie types $`B,C`$ and $`D`$ are the same as those factoring over specializations of $`BMW`$-algebras, due to $`q`$-Schur-Weyl-Brauer duality, see \[Wz1\]. For this reason we ignore these cases in the following weaker version of Theorem 6.8.
###### Theorem 6.10.
Let $`(X_r,\lambda )`$ be as in Table 1 with $`X=A_r`$ or $`E_6`$. Then
1. For $`(A_r,\varpi _2)`$: provided $`r3`$ and $`\mathrm{}\mathrm{max}(\mathrm{r}+3,7)`$, $`\mathrm{Hom}_U((V_\lambda )^3,V_{\varpi _2+\varpi _4})`$ is unitary, irreducible and 3-dimensional and the image of $`\sigma _1`$ has 3 distinct eigenvalues. If $`V_\mu `$ appears in $`V_{\varpi _2+\varpi _4}V_\lambda ^{n3}`$ then $`\mathrm{Hom}_U(V^n,V_\mu )`$ *contains* an irreducible unitary representation of $`_n`$ with the 3-eigenvalue property. When $`\mathrm{}\{10,14\}`$, the eigenvalues of the image of $`\sigma _1`$ are not in geometric progression and the images of $`_n`$ are infinite modulo the center and so are dense in these cases.
2. For $`(A_r,2\varpi _1)`$: $`\mathrm{Hom}_U((V_\lambda )^3,V_{2\varpi _1+2\varpi _2})`$ is unitary, irreducible and 3-dimensional provided $`r1`$ and $`\mathrm{}r+5`$. If $`V_\mu `$ appears in $`V_{2\varpi _1+2\varpi _2}V_\lambda ^{n3}`$ then $`\mathrm{Hom}_U(V^n,V_\mu )`$ *contains* an irreducible unitary representation of $`_n`$ with the 3-eigenvalue property. When $`\mathrm{}\{6,10\}`$ the eigenvalues of the image of $`\sigma _1`$ are not in geometric progression and the images of $`_n`$ are infinite modulo the center and so are dense in these cases.
3. for $`(E_6,\varpi _1)`$: $`\mathrm{Hom}_U((V_\lambda )^3,V_{\varpi _1+\varpi _6})`$ is unitary, irreducible and 3-dimensional provided $`\mathrm{}14`$. If $`V_\mu `$ appears in $`V_{\varpi _1+\varpi _6}V_\lambda ^{n3}`$ then $`\mathrm{Hom}_U(V^n,V_\mu )`$ *contains* an irreducible unitary representation of $`_n`$ with the 3-eigenvalue property. Provided $`\mathrm{}18`$, the eigenvalues of the image of $`\sigma _1`$ are not in geometric progression and the images of $`_n`$ are infinite modulo the center and so are dense in these cases.
###### Proof.
For the object labelled by $`V_\nu `$ to be in the fundamental alcove, we must have $`\nu +\rho ,\theta <\mathrm{}`$ where $`\theta `$ is the highest root. This condition together with the requirement that the eigenvalues be distinct yield the first restrictions in each case. The unitarity of the representations is shown in \[Wz2\]. In each case the representation spaces $`\mathrm{Hom}_U(V^n,V_\mu )`$ described in the theorem contain the 3-dimensional representation spaces, so by restriction to $`_3`$ we see that the $`_n`$ representations must contain an irreducible unitary subrepresentation with the 3-eigenvalue property. Geometric progressions appear in each of the three cases if and only if $`\mathrm{}=10`$ in the first case, $`\mathrm{}=6`$ or $`10`$ in the second case and $`\mathrm{}=18`$ in the last case. Computing projective orders of the images of $`\sigma _1`$ and comparing as in the proof of Theorem 6.8 we find that the only finite group image that arises is in the first case with $`\mathrm{}=14`$. With these exceptions, the hypotheses of Theorem 3.3 are satisfied and we may conclude the images are dense. ∎
###### Remark 6.11.
To get sharper results we would need to describe the decompositions of the $`_n`$ representations $`\mathrm{Hom}_U(V^n,V_\mu )`$ that appear in the above theorem. This is in general quite complicated. In fact, the type $`E_6`$ case appears in an exceptional series discussed in \[Wz3\] (and extended slightly in \[R2\]). These give new semisimple finite dimensional quotients of the braid group algebras analogous to $`BMW`$-algebras about which little is known.
### 6.4. Concluding Remarks
In comparing this work to the 2-eigenvalue paper, it may be noted that we do not provide applications of our results to the distribution of values of the Kauffman polynomial in analogy with those given for the Jones polynomial in \[FLW, §5\]. That is, we do not consider the set of values $`F_L(a,z)`$ for fixed $`a`$ and $`z`$ and varying $`L`$. These values can be described as linear combinations of traces of any braid with closure $`L`$ in the different irreducible factors of a BMW algebra, just as in \[FLW\]. The difficulty is that our information on the closures of braid groups in BMW algebras is less detailed than the corresponding information for Hecke algebras. In particular, we have not completely determined these closures for the irreducible factors of the BMW-representations which are excluded in the statement of Theorem 6.8. Neither have we determined the equivalences and dualities existing between different irreducible factors in a fixed BMW-algebra. We certainly expect the limiting distributions to be Gaussian as for the Jones polynomial, but we do not yet have enough information to ensure that this is so.
In 1990s, Vertigan (see Theorems 6.3.5 and 6.3.6 of \[Wel\]) analyzed the classical computational complexity of exactly evaluating various knot polynomials at fixed complex values. With a few exceptions, all evaluations are $`\mathrm{\#}P`$-hard. At these few exceptional values, the link invariants have classical topological interpretations and can be computed in polynomial time. These results fit very well with the analysis of closed images of the braid group representations. In the case of unitary Jones representations of the braid groups at $`q`$, the closed image is dense in the corresponding special unitary groups exactly when computing the link invariants is $`\mathrm{\#}P`$-hard at $`q`$, while the finite image cases correspond to polynomial time computations. Part of the appeal of working out the exceptions to Theorem 6.8 is the hope of relating these cases to interesting special values of the Kauffman polynomial.
|
warning/0506/astro-ph0506041.html
|
ar5iv
|
text
|
# The DEEP2 Galaxy Redshift Survey: The Galaxy Luminosity Function to 𝑧∼ 1Based on observations taken at the W. M. Keck Observatory which is operated jointly by the University of California and the California Institute of Technology
## 1. Introduction
The luminosity function is an important tool to analyze redshift surveys since it provides a direct estimate of how much light is contained in galaxies. By characterizing the observed changes with redshift in the luminosity function of galaxies as a function of (rest-frame) wavelength, it is possible to measure how the star-formation rates (e.g., using ultra-violet data) and stellar masses (e.g., using $`K`$-band data) have changed as a function of time. These analyses quantify the observed changes undergone by the galaxies’ masses and mass-to-light ratios, thus providing valuable data for theories of galaxy formation.
The measurement of the galaxy luminosity function for samples of field galaxies (i.e., galaxies selected for redshift measurements independent of their local environment) has been made for almost every major redshift survey (see Binggeli et al. 1998, Tresse 1999 and de Lapparent et al. 2003 for reviews). Thanks to major surveys such as the Two-Degree Field Galaxy Redshift Survey (Colless et al. 2001) and the Sloan Digital Sky Survey (York et al. 2000), precise measurements of the luminosity function in the local ( $`z<0.3`$) universe are available (e.g., Norberg et al. 2002; Blanton et al. 2003; Bell et al. 2003), providing a benchmark to measure the luminosity function evolution.
The characterization of properties of galaxies at redshifts $`z`$ 1, a time when the universe was half its present age, is then an important step to fully understand how galaxies formed and evolve. The DEEP2 Redshift Survey (Davis et al. 2003) is a project that is measuring 50,000 galaxy redshifts in four widely separated regions of the sky, comprising a total area of 3.5 $`\mathrm{}\mathrm{°}`$, and is specifically designed to probe the properties of galaxies at redshifts beyond $`z`$ = 0.7.
In a series of two papers, the $`B`$-band galaxy luminosity function to $`z`$1 will be investigated using the DEEP2 Redshift Survey (this paper), followed by an analysis that combines DEEP2 with other major current surveys of distant galaxies (Wolf et al. 2003; Gabasch et al. 2004; Ilbert et al. 2005), discussing possible evolutionary scenarios for early and late type galaxies (Faber et al. 2005, hereafter Paper II). The choice of $`B`$-band rather than other rest-frame wavelengths is motivated by the large number of measurements in this spectral range both for local as well as for distant samples of galaxies. An additional advantage is that for most of the higher-redshift intervals considered in this work, observed $`R`$ and $`I`$ are sampling rest-frame $`B`$, thus minimizing the importance of K-corrections. The present paper uses data from $``$ 1/4 of the total DEEP2 survey to measure the galaxy luminosity function, and discusses the importance of several selection effects in its measurement. This analysis will also take advantage of the recently found $`bimodality`$ of galaxies in the color-magnitude diagram (Strateva et al. 2001; Hogg et al. 2003; Baldry et al. 2004 and references therein), where the predominantly red early-type galaxies occupy a distinct locus in color from the blue star-forming galaxies. This bimodality has been shown to extend to $`z`$1 (Im et al. 2002; Bell et al. 2004; Weiner et al. 2005) and beyond (e.g., Giallongo et al. 2005). A bimodal distribution is also seen for other parameters such as spectral class (Madgwick et al. 2002, 2003), morphologies, metallicities, and star formation rates (Kauffmann et al. 2003). The present paper will show that this bimodality persists to $`z`$ 1, and that it is related on how these two populations have evolved over the last 6 Gyr.
This paper is organized as follows: §2 presents the DEEP2 data used in this paper, describing the selection effects that are present in this sample; §3 describes the methods used to measure the luminosity function and its evolution, and the weighting scheme that was adopted to correct for the incomplete data sampling; §4 presents the analysis of DEEP2 data showing how the evolution of the galaxy luminosity function depends on the internal properties of galaxies, blue galaxies showing mainly luminosity evolution while the red galaxy luminosity function shows a decrease in number density toward higher redshifts. Two appendices follow, describing the method used to calculate the K-corrections and another estimating biases in the luminosity function calculation by making cuts at different limiting absolute magnitudes. Throughout this work, a ($`H_0,\mathrm{\Omega }_M,\mathrm{\Omega }_\mathrm{\Lambda })`$ = (70, 0.3, 0.7) cosmology is used. Unless indicated otherwise, magnitudes and colors are converted into the Vega system, following the relations shown in Table 1.
## 2. Data
This section gives a brief description of the DEEP2 data; for more details the reader is referred to Davis et al. (2003), who give an outline of the project, Faber et al. (in preparation), who describe the survey strategy and spectroscopic observations, Coil et al. (2004b), who describe the preparation of the source catalog, and Newman et al. (in preparation), where the spectroscopic reduction pipeline is described.
The photometric catalog for DEEP2 (Coil et al. 2004b) is derived from Canada-France-Hawaii Telescope (CFHT) images taken with the 12K $`\times `$ 8K mosaic camera (Cuillandre et al. 2001) in $`B`$, $`R`$ and $`I`$ in four different regions of the sky. The $`R`$-band images have the highest signal-to-noise and were used to define the galaxy sample, which has a limiting magnitude for image detection at $`R_{AB}25.5`$. Objects were identified using the $`imcat`$ software written by N. Kaiser and described by Kaiser, Squires & Broadhurst (1995). In addition to magnitudes, $`imcat`$ calculates other image parameters which are used in the object classification. The separation between stars and galaxies is based on magnitudes, sizes, and colors, which are used to assign each object a probability of being a galaxy ($`P_{gal}`$). For the DEEP2 fields, the cut is made at $`P_{gal}>`$ 0.2, $`i.e.`$, objects are considered as part of the sample whenever the probability of being a galaxy is greater than 20%. In Fields 2, 3, and 4, the spectroscopic sample is pre-selected using $`B`$, $`R`$, and $`I`$ to have estimated redshifts greater than 0.7, which approximately doubles the efficiency of the survey for galaxies near $`z1`$. The fourth field, the Extended Groth Strip (EGS), does not have this pre-selection but instead has roughly equal numbers of galaxies below and above $`z=0.7`$ selected using a well understood algorithm. In addition to the redshift pre-selection, a surface brightness cut defined as
$$SB=R_{AB}+2.5Log_{10}\{\pi (3r_g)^2\}26.5,$$
(1)
is applied when selecting spectroscopic candidates, where $`R_{AB}`$ is the $`R`$-band (AB) magnitude, and $`r_g`$ is the 1 $`\sigma `$ radius of the Gaussian fit to the image profile in the CFHT photometry; the minimum size for $`r_g`$ is fixed at 0<sup>′′</sup>.33, so that for compact objects with $`3r_g<1^{\prime \prime }`$, the surface brightness is measured within a circular aperture of 1<sup>′′</sup>. Finally, galaxies were selected to lie within bright and faint apparent-magnitude cuts of 18.5$`R_{AB}`$ 24.1.
The DEEP2 sample used here combines data from the first season of observations in Fields 2, 3, and 4 with about 1/4 of the total EGS data, which provides an initial sample at low redshifts. The total number of galaxies is 11284, with 4946 (45%) in EGS, 3948 (36%) in Field 4, 2299 (21%) in Field 3, and 91 (1%) in Field 2. Because of the $`BRI`$ redshift pre-selection, for $`z<0.8`$, only EGS is sampled well enough to be used, while data in all four fields are used for $`z0.8`$.
DEEP2 spectra were acquired with the DEIMOS spectrograph (Faber et al. 2003) on the Keck 2 telescope and processed by an automated pipeline that does the standard image reduction (division by flatfield, rectification of spectra, extraction of 2-D and 1-D spectra) and redshift determination (Newman et al., in preparation). The only human intervention occurs during redshift validation, where spectra and redshifts are visually examined, and redshifts are given a quality assessment that ranges from 1 (for completely indeterminate) to 4 (for ironclad). Only redshifts with quality 3 and 4 are used in this paper, which means that two or more features have been identified (the \[O II\] $`\lambda `$3727 doublet counts as two features). Duplicate observations and other tests indicate an rms accuracy of 30 km s<sup>-1</sup> and an unrecoverable failure rate of $``$1% for this sample.
The apparent color-magnitude (CM) diagram in $`R`$ versus $`RI`$ is shown in Figure 1$`a`$ for the DEEP2 parent catalog after applying the photometric-redshift cut in three of the fields and converting into Vega magnitudes (cf. Table 1). Figure 1$`b`$ shows the distribution of galaxies placed on masks, Figure 1$`c`$ shows galaxies with successful redshifts, and Figure 1$`d`$ shows galaxies with “failed” redshifts. Although failures are found in all parts of the diagram, the largest concentration is at faint and blue magnitudes. Independent data show that the great majority of these are beyond $`z1.4`$ (C. Steidel, private communication), corresponding to \[O II\] $`\lambda `$3727 passing beyond the DEEP2 wavelength window at that redshift. Redshift histograms corresponding to the rectangular regions outlined in Figure 1 are shown in Figure 2, where the vertical bars at the right of each diagram represent the number of failed redshifts in each bin. The increase in failures for faint and blue galaxies is apparent.
Figure 3$`a`$ plots $`UB`$ versus distance for the whole sample, where the rest-frame color is calculated using the K-correction procedure described in Appendix A. Throughout this paper, the rest-frame colors and magnitudes are corrected for Galactic extinction (Schlegel, Finkbeiner & Davis 1998) but not internal extinction. Color bimodality dividing red and blue galaxies is immediately apparent, extending to beyond $`z=1`$. Panel $`b`$ shows the EGS by itself, while panels $`c`$ and $`d`$ show the high effectiveness of the $`BRI`$ photometric selection in Fields 2, 3, and 4.
Figure 4 plots CM diagrams using $`UB`$ versus $`M_B`$ as a function of redshift. The solid line in each panel represents the limiting absolute magnitude at the high redshift end of each bin. The slope of this line changes with redshift because of the adoption of a fixed apparent magnitude limit ($`R`$) for the sample, with the color-redshift-dependence of the K-correction. At $`z`$ 0.4 the $`R`$-band filter used to select the sample coincides with rest $`B`$ but differs from it increasingly as the redshift is either greater or smaller than 0.4.
The bimodality in color-magnitude distribution is clearly seen; while red galaxies tend to be brighter on average than blue galaxies, it is clearly seen that blue galaxies dominate the sample when number of objects is considered. The upper dashed lines represent the cut used to separate red and blue galaxies, as explained in §4.1. Since the evidence of color evolution in DEEP2 data is slight, the zero-point and slope of this line with redshift is kept constant. The lower dashed lines have the same slope and are used to divide blue galaxies into two equal halves for further luminosity-function analysis; their zero-points are explained in §4.2.
An interesting feature of these diagrams is that, even though the detection of faint galaxies is favored at low redshifts, there are still very few red galaxies found with $`M_B>18`$, even at redshifts below $`z`$ 0.6, where they should be seen. The same absence was also seen by Weiner et al. (2005) in DEEP1 and by Kodama et al. (2004) in distant clusters. This point is discussed further in the context of COMBO-17 data in Paper II.
## 3. Methods
### 3.1. Luminosity Function Estimators
The luminosity function is defined as the number of galaxies per unit magnitude bin per unit co-moving volume, and is most frequently expressed using the Schechter (1976) parameterization, which in magnitudes is:
$`\varphi (M)dM`$ $`=`$ $`0.4ln10\varphi ^{}10^{0.4(M^{}M)(\alpha +1)}`$ (2)
$`\times exp\{10^{0.4(M^{}M)}\}dM,`$
where $`\varphi ^{}`$ represents the characteristic number density of galaxies per unit volume per unit magnitude, $`M^{}`$ the characteristic magnitude where the growth of the luminosity function changes from an exponential into a power law, and $`\alpha `$ the slope of this power law that describes the behavior of the faint end of this relation. Several estimators have been proposed to measure this statistic (e.g., Schmidt 1968; Lynden-Bell 1971; Turner 1979; Sandage, Tammann & Yahil 1979; Choloniewski 1986; Efstathiou, Ellis & Peterson 1988), and the relative merits of the different methods were explored by Willmer (1997) and Takeuchi, Yoshikawa & Ishii (2000) through the use of Monte-Carlo simulations.
In this work, the luminosity function calculation relies on two estimators. The first is the intuitive $`1/V_{max}`$ method where galaxies are counted within a volume. The calculation used here follows Eales (1993), Lilly et al. (1995), Ellis et al. (1996) and Takeuchi et al. (2000), which overcomes the bias identified by Felten (1976) and Willmer (1997). The integral luminosity function for an absolute magnitude bin between $`M_{bright}`$ and $`M_{faint}`$ is described as:
$$_{M_{bright}}^{M_{faint}}\varphi (M)𝑑M=\underset{i=1}{\overset{N_g}{}}\frac{\chi _i}{V_{max}(i)},$$
(3)
where $`\chi _i`$ is the galaxy weight that corrects for the sampling strategy used in the survey (discussed in detail in §3.3 below) and $`V_{max}(i)`$ is the maximum co-moving volume within which a galaxy $`i`$ with absolute magnitude $`M_i`$ may be detected in the survey:
$$V_{max}(i)=_\mathrm{\Omega }_{z_{min},i}^{z_{max},i}\frac{d^2V}{d\mathrm{\Omega }dz}𝑑z𝑑\mathrm{\Omega },$$
(4)
where $`z`$ is the redshift and $`\mathrm{\Omega }`$ the solid angle being probed. In a survey that is limited at bright ($`m_l`$) and faint ($`m_u`$) apparent magnitudes, the redshift limits $`z_{min},i`$ and $`z_{max},i`$ for galaxy $`i`$ are:
$$z_{max},i=min\{z_{max},z(M_i,m_u)\}$$
(5)
$$z_{min},i=max\{z_{min},z(M_i,m_l)\}$$
(6)
where the terms in braces are the redshift limits imposed either by the limits of the redshift bin being considered ($`z_{min}`$ and $`z_{max}`$) or by the apparent magnitude limits of the sample ($`m_l`$ and $`m_u`$). The Poisson error for the 1/$`V_{max}`$ method in a given redshift bin is given by:
$$\sigma _\varphi =\sqrt{\frac{\chi _i}{(V_{max}(i))^2}}.$$
(7)
In this paper, the $`1/V_{max}`$ method is calculated in absolute magnitude bins 0.5 mag wide, and redshift bins of width $`\mathrm{\Delta }z`$ = 0.2. The result is the average value of the luminosity function $`\varphi (M_k,z)`$ at redshift $`z`$ in magnitude bin $`k`$. The method makes no assumption about the shape of the luminosity function, therefore providing a non-parametric description of the data.
The second estimator is the most commonly used in luminosity function calculations – the parametric maximum-likelihood method of Sandage, Tammann and Yahil (1979, STY; Efstathiou, Ellis & Peterson 1988; Marzke, Huchra & Geller 1994). The STY method fits an analytic Schechter function (Equation 2), yielding values of the shape parameters $`M^{}`$ and $`\alpha `$ (but not the density normalization $`\varphi ^{}`$).
The probability density that a galaxy with absolute magnitude $`M_i`$ will be found in a redshift survey sample is proportional to the ratio between the differential luminosity function at $`M_i`$ and the luminosity function integrated over the absolute magnitude range that is detectable at redshift $`z_i`$. In the case of DEEP2 galaxies, the STY conditional probabilities were modified following Zucca, Pozzetti, & Zamorani (1994) to account for the galaxy weights, $`\chi _i1`$ (see §3.3 below), correcting for the sampling (e.g., Lin et al. 1999) and redshift success rates:
$$p(M_i,z_i)=[\frac{\varphi (M_i)dM}{_{M_{bright(z_i)}}^{M_{faint(z_i)}}\varphi (M)𝑑M}]^{\chi _i}.$$
(8)
Here $`M_{bright(z_i)}`$ and $`M_{faint(z_i)}`$ are the absolute magnitude limits at redshift $`z_i`$ accessible to a sample with apparent magnitude limits $`m_u`$ and $`m_l`$. $`M_{bright(z_i)}`$ and $`M_{faint(z_i)}`$ are implicitly a function of color (cf. Figure 4), which motivates the approach (used here) to divide galaxies at least broadly into two color bins. Implicitly, $`\varphi (M)`$ is assumed to vary with $`z`$; the analysis is carried out in fixed redshift bins in which $`\varphi (M,z)`$ is determined.
The likelihood function maximized by the STY method is defined by the joint probability of all galaxies in the sample belonging to the same parent distribution. The solution is obtained by assuming a parametric form for the luminosity function and maximizing the logarithm of the likelihood function relative to the product of the probability densities of the individual galaxies $`p(M_i,z_i)`$:
$$\mathrm{ln}=\mathrm{ln}\underset{i=1}{\overset{N_g}{}}p(M_i,z_i).$$
(9)
Because this method uses no type of binning, it preserves all information contained in the sample. Since the luminosity function normalization is canceled out (Equation 8), it is insensitive to density fluctuations in the galaxy sample. However, this also means that the normalization (defined by $`\varphi ^{}`$) must be estimated separately, using the procedure described in §3.2 below.
Another shortcoming of the STY method is that it does not produce a visual check of the fit. However, this can be done using the $`1/V_{max}`$ method, which shows the average number density of galaxies in bins of absolute magnitude and can be compared directly to the shape parameters of the STY results. The $`1/V_{max}`$ points also provide an independent check on the luminosity function normalization.
### 3.2. Luminosity Function Normalization
Since the STY probability estimator is defined from the ratio between the differential and integral luminosity functions, the density normalization is factored out and has to be estimated independently. The standard procedure for obtaining the luminosity function normalization ($`\varphi ^{}`$) measures the mean number density of galaxies in the sample, $`\overline{n}`$, which is then scaled by the integral of the luminosity function:
$$\varphi ^{}=\frac{\overline{n}}{_{M_{bright}}^{M_{faint}}\varphi (M)𝑑M}$$
(10)
where $`M_{bright}`$ and $`M_{faint}`$ are the brightest and faintest absolute magnitudes considered in the survey.
The method used to measure the mean density $`\overline{n}`$ is the unbiased minimum-variance estimator proposed by Davis & Huchra (1982):
$$\overline{n}=\frac{{\displaystyle \underset{i=1}{\overset{N_g}{}}}\chi _iN_i(z_i)w(z_i)}{_{z_{min}}^{z_{max}}s(z)w(z)\frac{dV}{dz}𝑑z},$$
(11)
which averages the redshift distribution of galaxies, $`N_i(z_i)`$, corrected by a weighting function, $`w(z_i)`$, that takes into account galaxy clustering; the selection function, $`s(z`$), that corrects for the unobserved portion of the luminosity function; and the sampling weight, $`\chi _i`$. The selection function is given by:
$$s(z)=\frac{_{max(M_{min(z_i)},M_{bright})}^{min(M_{max(z_i)},M_{faint})}\varphi (M)𝑑M}{_{M_{bright}}^{M_{faint}}\varphi (M)𝑑M}.$$
(12)
where $`M_{min(z_i)}`$ and $`M_{max(z_i)}`$ are the brightest and faintest absolute magnitudes at redshift $`z_i`$ contained within the apparent magnitude limits of the sample. The contribution due to galaxy clustering is accounted for by the second moment $`J_3`$ of the two-point correlation function $`\xi (r)`$ (e.g., Davis & Huchra 1982), which represents the mean number of galaxies in excess of random around each galaxy out to a distance $`r`$ (typically set at $``$ 30 Mpc):
$$w(z_i)=\frac{1}{1+\overline{n}J_3s(z)},J_3=_0^rr^2\xi (r)𝑑r.$$
(13)
Because of the small range of absolute magnitudes available at high redshift, the shape of the faint end slope, parameterized by $`\alpha `$ is not constrained by the fit, so we opted to keep the value of this parameter fixed, as discussed in §4.2. Thus, in the calculation of errors for the Schechter parameters only $`M^{}`$ and $`\varphi ^{}`$ are considered. Since the STY method factors out the density, it is also not suitable for calculating the correlated errors of $`\varphi ^{}`$ and $`M^{}`$, as, lacking $`\varphi ^{}`$, STY cannot take the high correlation between these two errors into account. These errors were therefore calculated from the 1-$`\sigma `$ error ellipsoid (Press et al. 1992) that resulted from fitting the Schechter function to the $`1/V_{max}`$ data points. Although the luminosity functions that result from the STY and $`1/V_{max}`$ methods are not quite identical (cf. Figure 7), the differences are small, and errors from $`1/V_{max}`$ should also be applicable to the STY method.
### 3.3. The Sampling Function and Galaxy Weights
An issue with every data set is the selection of weights to correct for missing galaxies. The adopted weights need to take into account the fact that (1) objects may be missing from the photometric catalog, (2) stars may be identified as galaxies and vice versa, (3) not all objects in the photometric catalog are targeted for redshifts (sampling rate) and (4) not all redshift targets yield successful redshifts (redshift success rate). In the case of DEEP2, since the limiting magnitude of the photometric catalog is 1.5 magnitudes fainter than the limit adopted for redshift selection, any effects due to incompleteness of the source catalog should be negligible. The loss of galaxies brighter than $`R_{AB}`$=24.1 but with surface brightness too low to admit them in the photometric catalog is ruled out from the inspection of HST images analyzed by Simard et al. (2002) for the EGS region in common with Groth Strip, which shows no large low-surface brightness galaxies. The loss of galaxies because of confusion with stars in well-defined regions of the color-magnitude diagram (item 2 above) is shown in §3.4 below to be negligible. Therefore, only factors (3) and (4) need to be taken into account in the weights. The basic assumption to deal with (3) is that all unobserved galaxies share the same average properties as the observed ones in a given color-magnitude bin. The last effect, factor (4), is dealt with by assigning a model redshift distribution to the failed galaxies.
A visual description of how the sampling rate and redshift success rates depend on the magnitude and color of galaxies is shown in Figure 5, which projects both rates averaged in a color-color-magnitude data cube onto the $`R`$ versus $`RI`$ plane. Both rates are shown separately for EGS and Fields 2-4 because of the different selection criteria. In the EGS, the average sampling rate of slits placed on galaxies is $``$60%, and the average redshift success is 73%. For Fields 2, 3, and 4, the average sampling rate is 59% (after foreground galaxies are eliminated via color pre-selection), and the redshift success is 73%.
To account for the unobserved galaxies and redshift failures we follow in this paper (and in Paper II for DEEP1 data) the method first applied by Lin et al. (1999) to the CNOC2 Redshift Survey. This defines around each galaxy $`i`$ a data cube in color-color-magnitude space and, from all attempted redshifts, counts the number of failed redshifts ($`N_f`$), the number ($`N_{z_h}`$) of galaxies with $`z>z_h`$, where $`z_h`$ is the high redshift limit of the sample; the number ($`N_{z_l}`$) of galaxies with $`z<z_l`$, where $`z_l`$ is the low redshift limit of the sample and the number ($`N_z`$) with good redshifts within the “legal” redshift range $`z_l`$ to $`z_h`$. For Fields 2-4, $`z_l=0.8`$ and $`z_h=1.4`$; and for EGS these are $`z_l=0.2`$ and $`z_h=1.4`$.
Next, for each galaxy in the photometric source catalog, the probability that it has a redshift in the legal redshift range is estimated. In the case of galaxies with good-quality redshifts, the probability that the redshift lies in the legal range is simply $`P(z_lzz_h)`$ = 1 when the galaxy has $`z_lzz_h`$ and $`P(z_lzz_h)`$ = 0 for $`z>z_h`$ or $`z<z_l`$. To get the probability for unobserved galaxies, however, some assumption must be made for the distribution of the failed redshifts. Two main models are used in the present work. The first assumes that all failed redshifts are beyond the high redshift cutoff of the sample, $`z_h`$ (the “minimal” model). In this case, the probability that an unobserved galaxy will be within the legal redshift range is the ratio of the number of good redshifts in the range divided by the sum of the number of successful redshifts plus failures:
$$P(z_lzz_h)=\frac{N_z}{N_z+N_{z_l}+N_{z_h}+N_f}.$$
(14)
The alternative model assumes that failures follow the same distribution as the observed sample (the “average” model). In this case, Equation 15 becomes:
$$P(z_lzz_h)=\frac{N_z}{N_z+N_{z_l}+N_{z_h}}.$$
(15)
Finally, the weight for each galaxy $`i`$ with an acceptable redshift is calculated by adding for all galaxies $`j`$ within the color-color-magnitude bin the probability that the redshift of galaxy $`j`$ is within the legal limits of the sample
$$\chi _i=\frac{\underset{j}{}P(z_lz_jz_h)}{N_z},$$
(16)
where $`j`$ includes both galaxies with and without attempted redshifts.
In the case of EGS, a final correction is applied to the weights to account for the different sampling strategy that was used, which includes low-redshift galaxies but de-weights them so that they do not dominate the sample. This (independently known) correction ($`f_m`$, Faber et al. in preparation; Newman et al. in preparation) depends on the location of the galaxy in $`BR`$ versus $`RI`$ and its apparent magnitude. From this correction, the probability that a galaxy will be placed on an EGS mask is given by:
$$P(mask)=0.33+0.43P_{gal}f_m,$$
(17)
where $`P_{gal}`$ is the probability that an object is a galaxy. For EGS galaxies the final probability weight is given by
$$\chi _i=\frac{\underset{j}{}P(z_{min}z_jz_{max})}{N_zP(mask)},$$
(18)
where $`j`$ includes both galaxies with and without redshifts.
The comparison of Equations 14 and 15 shows that weights in the average model are larger than in the minimal model. The weights and differences in weights between the minimal and average models are shown in Figure 6. These differences are typically of order 15-20% and most large differences occur for galaxies with extreme colors at faint magnitudes.
Based on the unpublished data of Steidel mentioned in §2, the minimal model more closely matches blue galaxies since most failed blue galaxies lie beyond the upper redshift limit of the survey $`z_h`$=1.4. In contrast, most failed red galaxies probably lie within the survey range and are better described by the average model. Because of this behavior, for the All galaxy sample we adopt a compromise “optimal” model, where blue galaxies have weights described by the minimal model, while red galaxies use the average model. However, since the All sample is dominated by blue galaxies, the differences between the optimal and minimal models are very small.
### 3.4. Other Sources of Incompleteness
Several tests were carried out to estimate the impact of what we believe are the principal sources of incompleteness, namely the surface brightness limit for slit assignment, the misclassification of objects, and the presence of dropouts in the $`B`$ band photometry.
To limit the rate of redshift failures, a surface brightness cut (Equation 1) was used to place galaxies on slits. This restriction eliminates both red ($`RI>`$ 1.25) and blue ($`RI`$ 1.25) galaxies, but the numbers are small. The overall fraction of red galaxies that lie below the surface brightness cut is $``$ 3%, increasing to 6% over the faintest 0.5 magnitude. For blue galaxies, the average number is $``$ 5%, increasing to 7% in the faintest 0.5 magnitude bin. In both cases, these numbers are accounted for by the weighting, since all galaxies that were not placed on slits are still counted when the weights are calculated, so no additional corrections are needed, as long as the characteristics of lower-surface brightness galaxies are similar to those of other galaxies situated in the same color-color-magnitude bin.
As mentioned in §2, the star-galaxy separation relies on the colors and sizes of detected objects to assign each one the probability of being a galaxy ($`P_{gal}`$). Since stars occupy a well defined locus in the $`RI`$ $`vs.`$ $`BR`$ diagram (Coil et al. 2004b), it is possible that DEEP2 galaxies with small apparent sizes and observed colors close to the stellar locus could be treated as stars ($`P_{gal}<`$0.2) in DEEP2 mask-making and thus be ignored in the analysis since the latter are not placed on masks. This loss is estimated using objects in common between DEEP2 and the structural catalog of Simard et al. (2002), which is derived from psf-corrected photometry using HST images in the original Groth Strip. When plotting the half-light radius $`versus`$ total-magnitude distribution (e.g., Figure 6 of Im et al. 2002), stars and galaxies are well separated down to the limiting magnitude $`R_{AB}`$= 24.1 adopted by DEEP2, which corresponds to approximately $`I814`$ 23.5. For red objects located close to the red stellar locus ($`RI1.25`$, 1.8 $`BR`$ 3.5), a total of 8 objects that are clearly galaxies in HST images are identified as stars ($`P_{gal}<`$0.2) in the DEEP2 source catalog, while 64 galaxies ($`P_{gal}`$0.2) are correctly identified within the same color boundaries. This corresponds to a loss of (8/64) or 13%. On the other hand, the number of spectroscopically observed stars misclassified as (red) galaxies, corresponds to $``$ 8% of the sample in the faintest magnitude bin ( 23.5 $`R_{AB}`$ 24.1). An examination of the distribution of surface brightnesses shows that all of these have $`SB`$ 25 $`R_{AB}`$ $`mag`$ $`arcsec^2`$, but that there are also galaxies in this range. Thus, the inspection of HST images and the distribution of sizes and surface brightnesses suggests that DEEP2 may be biased against high surface brightness red galaxies with small apparent sizes. However, no strong dependence with redshift was seen. Since the corrections for both effects are very uncertain, we opted not to apply them in the analysis.
A final systematic error is caused by the presence of $`B`$-band dropouts, which are objects that have good $`R`$ and $`I`$ magnitudes, but a low S/N or non-existent $`B`$ measurement. All three magnitudes ($`B,R,I`$) are needed to sort galaxies from stars; if $`B`$ is too dim and noisy, that object is never assigned to a slit. Moreover, as there are no $`BR`$ colors for the dropouts, such objects are also not accounted for in the weighting procedure described above which uses bins in color-color-magnitude space. Consequently, the weights were modified to account for the loss of these objects by counting the number of dropouts within each ($`R`$, $`RI`$) bin around a given galaxy and dividing this number by the total number of galaxies in the same bin. These corrections are typically less than 4%, though in some bins can reach $``$ 8%, and are applied to the final weights of each galaxy. The apparent $`RI`$ colors are consistent with most of these objects being part of the red sequence.
In summary, since most of these systematic effects due to incompleteness are small, they will not affect the final conclusions of this paper. Analyses carried out ignoring the last correction produce essentially identical results to those in the present paper.
## 4. Analysis
### 4.1. The Non-Parametric Luminosity Functions
The DEEP2 luminosity function is shown in Figure 7, the top row corresponding to the “All” galaxy function, while the second and third rows show the luminosity function determined for sub-samples of galaxies divided into “Blue” and “Red” by using the color bimodality. The weighting model (§3.3) adopted for each population is identified in the rightmost panel of each row.
For DEEP2 data, the color division between Red and Blue corresponds to the upper dotted line in Figure 4, which is given by:
$$UB=0.032(M_B+21.52)+0.4540.25.$$
(19)
This equation was derived from the van Dokkum et al. (2000) color-magnitude relation for red galaxies in distant clusters, converted to the cosmological model used in this paper and shifted downward by 0.25 mag in order to pass through the valley between red and blue galaxies. Although the colors of red galaxies may evolve with redshift, this effect is not strongly seen in DEEP2 colors, and a line with constant zero-point independent of redshift is adequate for all redshift bins. The constacy of $`UB`$ constrasts with the changes seen in the $`UV`$ $`vs.`$ $`M_V`$ of COMBO-17 (B04). However, when $`UB`$, $`UV`$ and $`BV`$ colors are plotted as a function of $`z`$ for the COMBO-17 sample, most of the color change can be traced to the $`BV`$ color (C. Wolf, private communication), implying that the stability of the DEEP2 color-magnitude relation over this redshift interval is not inconsistent with B04.
The separation between blue and red galaxies therefore is using a clear feature which is easily identified, even if its physical interpretation is not completely understood (e.g., Kauffmann et al. 2003).
Along the rows of Figure 7, each panel represents a different redshift bin, with $`z`$ increasing from left to right. The DEEP2 non-parametric luminosity function estimated using the 1/$`V_{max}`$ method is represented by the solid black squares. The sample used in the calculation of the luminosity function is shown in Figure 4. The absolute magnitude range is truncated at the faintest absolute magnitude which contains both red and blue galaxies, so that both populations are sampled in an unbiased way. A fully volume-limited sample for a given redshift bin would be obtained using the solid colored lines in Figure 4, which show limiting absolute magnitudes of the upper redshift of each bin, whereas the actually adopted limit (for the purpose of calculation of the luminosity functions), corresponds to the lower redshift limit of the bin. The slight loss of galaxies in the remainder of the bin does not affect the STY estimation since the range of absolute magnitudes accessible at any given $`z`$ is calculated on a galaxy-by-galaxy basis. In contrast, the 1/$`V_{max}`$ method will systematically underestimate the density of galaxies unless corrected, which was done by following Page & Carrera (2000), The error bars represent counting errors assuming Poisson statistics only. The uncertainty due to cosmic variance is shown as a separate error bar at the top left corner of each panel and was estimated following Newman & Davis (2002) who account for evolution of the correlation function using the mass power spectrum, and using the correct field geometry, that takes into account the elongated nature of DEEP2 fields which reduces the cosmic variance . The bias factors derived by Coil et al. (2004a) for red galaxies ($`b=1.32`$) and blue galaxies ($`b=0.93`$) relative to the mass are included in these cosmic variance estimates. To first order, cosmic variance should affect mainly the overall number density, $`\varphi ^{}`$, moving all points up and down together and leaving the shape of the function unchanged, whereas Poisson variance is random from point to point; therefore we show the Poisson and cosmic variance error bars separately. The dashed gray curves represent the DEEP2 luminosity function fits (§4.2) measured in the lowest redshift bin ($`0.2z<0.4`$), which are repeated in subsequent panels. The major conclusions are as follows:
All galaxies (top row): Relative to the low-$`z`$ Schechter function, the data in successive redshift bins march to brighter magnitudes ($`M_B^{}`$) but remain roughly constant in number density ($`\varphi ^{}`$). This visual assessment is confirmed by Schechter fits below. In short, for the whole population, galaxies are getting brighter with redshift, but their number density is remaining much the same, to $`z1`$.
Blue galaxies (middle row): The results found above for the All sample are repeated for the Blue sample, which is expected since blue galaxies dominate the total number of galaxies. This is shown in the middle row of Figure 7. The increasing separation between the points and black solid lines in each redshift bin relative to the DEEP2 fits at ($`0.2z<0.4`$) is easily seen, and the visual impression is that $`M_B^{}`$ brightens and $`\varphi ^{}`$ remains constant, again confirmed by Schechter fits below.
Red galaxies (bottom row): The bottom row of Figure 7 presents the data for red galaxies. As above, the dashed grey line represents the Schechter function fit to the lowest redshift bin of DEEP2 data. In contrast to blue galaxies, between $`z`$ 0.9 and $`z`$ 0.3, the luminosity function of red galaxies in DEEP2 shows no evidence for large changes, with most variations in the number density, particularly at low $`z`$, being within the margins of cosmic variance. The only bin that shows some hint of change is the highest-$`z`$ bin, centered at $`z`$= 1.1, but which is likely to be the most affected by incompleteness (see below). Therefore the results from the DEEP2 survey alone are consistent with rather little change in the raw counts of red galaxies at bright magnitudes. If $`M_B^{}`$ and $`\varphi ^{}`$ are changing, they must do so in coordinated fashion such that the counts at fixed magnitude remain roughly constant. This behavior differs markedly from that of blue galaxies, where counts increase at fixed $`M_B`$.
These results are fairly robust relative to the adopted weighting model. The black points in Figure 7 use the average model of §3.3, which assumes that red galaxies without redshifts follow the same distribution as the observed ones. For an extreme test, the weighting was changed to a model where failed red galaxies (comprising about 25% of the total red galaxy sample) are all placed in whatever redshift bin is being considered. Here, red galaxies are defined as all objects with apparent $`RI>1.33`$ (see line in Figure 1$`b`$). This extreme assumption clearly yields a strict upper limit to the red luminosity function in that bin. The test works well for red galaxies in the range $`z=0.71.1`$, which all cluster strongly near observed $`RI`$ = 1.5 (see Figure 1$`b`$).
This part of the apparent CM diagram thus contains all red galaxies that can possibly exist in this redshift range, unless large numbers are missing from the photometric catalog, which is unlikely, as discussed in §3.4. DEEP2 luminosity functions using this extreme incompleteness model are shown in Figure 8 as gray triangles. It is important to note that this model uses each failed red galaxy multiple times so the gray data points cannot be all valid simultaneously; they are strict upper limits. The new correction does not increase the number of galaxies very much in the All function, since the total counts are dominated by blue galaxies, and the Red function is significantly impacted in only the most distant bin. Quantitative conclusions are drawn below by fitting Schechter functions.
### 4.2. Schechter fits
The Schechter functions fits using the STY method are presented here. When splitting either galaxy sub-sample in narrow redshift bins, we see variations in the best-fitting faint-end slope that are not statistically significant, suggesting that we should average together slopes from several bins. In fact, the All galaxy function should show some trend because the ratio of red to blue galaxies changes with redshift and the shapes of the Red and Blue functions differ; however, the effect is small. As explained in more detail in Paper II, we decided to use the average faint-end slope values found within the range $`z=0.2`$ to 0.6 for the COMBO-17 sample, because of the much larger number of galaxies COMBO-17 contains in this redshift range in addition to there being no color pre-selection in that survey. The resulting values of the faint-end slope are $`\alpha =0.5`$ for the Red sample and $`\alpha =1.3`$ for the All and Blue samples; these were applied also to DEEP2 here. Even though several recent works have provided evidence of differential evolution between bright and faint red galaxies (e.g., McIntosh et al. 2005; Juneau et al. 2005; Treu et al. 2005), we adopt a fixed Schechter function in shape at all redshifts. The effect of varying the shape is small, as discussed in Paper II. The evolving Schechter parameters are presented in Table 3 for the All sample and in Tables 4 and 5 for the Blue and Red samples. Column (1) shows the central redshift of the bin; column (2) the number of galaxies used in the luminosity function calculation in each redshift bin; column (3) the value of the adopted faint-end slope, $`\alpha `$; column (4) the value of $`M_B^{}`$, followed by the upper and lower 68% Poisson errors in columns (5) and (6); the mean density $`\varphi ^{}`$ in column (7), followed by the 68% Poisson errors in columns (8) and (9); the square root of the cosmic variance error is shown in column (10); and (11) shows the luminosity density (in solar units) defined as
$$j_B(z)=L\varphi (L)𝑑L=L^{}\varphi ^{}\mathrm{\Gamma }(\alpha +2),$$
(20)
using $`M_B\mathrm{}`$=5.48 (Binney & Merrifield 1998), where $`\mathrm{\Gamma }`$ is the Gamma function, with the 68% Poisson error in column (12); column (13) indicates the weighting model (described in §3.3) used when calculating the fits. For the All and Red galaxy samples, the results using the the upper-limit method of §4.1 are also tabulated. The 68% Poisson errors for $`M_B^{}`$ and $`\varphi ^{}`$ were taken from the $`\mathrm{\Delta }\chi ^2`$ = 1 contour levels in the ($`M_B^{}`$, $`\varphi ^{}`$) plane, computed from the $`1/V_{max}`$ residuals and their errors relative to a given Schechter fit. Cosmic variance errors were computed as described above taking the volume and field geometry into account and using separate bias ($`b`$) values for Blue and Red relative to the All galaxy sample. Errors for $`j_B`$ were conservatively calculated by adding the fractional Poisson errors for $`M_B^{}`$ and $`\varphi ^{}`$ and cosmic variance in quadrature; these are an overestimate because this neglects the correlated errors in $`M_B^{}`$ and $`\varphi ^{}`$, which tend to conserve $`j_B`$. However, Poisson errors are generally smaller than cosmic variance, which is dominant, so this overestimate is small.
The changes of the Schechter parameters as a function of redshift are shown in Figure 11 for $`M_B^{}`$ (top row), $`\varphi ^{}`$ (middle row) and $`j_B`$ (bottom row) for the All, Blue and Red galaxy samples. The figure shows results separately for minimal and average models, and in the case of the All sample, using the optimal model. As expected from the raw counts in Figure 7, the Schechter parameters for blue and red galaxies evolve differently with redshift. The brightening of blue galaxies is clearly seen, while their number density ($`\varphi ^{}`$) holds fairly steady. In contrast, red galaxies evolve only modestly in either $`M_B^{}`$ or $`\varphi ^{}`$, and an increase in one quantity is balanced by the other keeping the total (red) luminosity density, $`j_B`$, roughly constant out to the very last bin, where it falls abruptly (see Table 5). The constancy of $`j_B`$ for red galaxies was noted by Bell et al. (2004), who drew the conclusion that the total stellar mass of the red sequence must be falling as a function of increasing redshift. Paper II provides further evidence for this. For now, we simply note that the DEEP2 red counts agree well with the raw COMBO-17 red counts (as shown in Paper II), and with the conclusion by Bell et al. (2004) that $`j_B`$ for red galaxies is constant.
The DEEP2 fitted values for $`\varphi ^{}`$ also show a formally significant drop back in time for red galaxies, a point which will be further discussed in Paper II.
The DEEP2 data were also used to explore if the different trends measured between red and blue galaxies can be detected when smaller subdivisions in the color-magnitude space are considered. For this, blue galaxies were subdivided using a line parallel to Equation (19) (which divides red from blue galaxies) but displaced downward in each redshift bin so it divides the blue galaxies into two equal halves. This line was calculated considering only galaxies brighter than $`M(z)=M_0Qz`$, where $`M_0`$ = -20, and $`Q`$ is the amount of luminosity evolution (measured in magnitudes) per unit redshift, so that only the statistically similar populations of galaxies would be used. This method was used in preference to a constant color cut, which would yield a spurious evolution in numbers simply because blue galaxies are reddening with time (cf. Figure 4). Although this division does not use a clear feature as that dividing blue and red galaxies, it is calculated at roughly the average color of blue galaxies at a given absolute magnitude, and it allows testing whether the degree of evolution is somehow correlated with the average color of galaxies. When calculating Schechter function fits for Moderately Blue and Very Blue galaxies, we find that the fixed faint-end slope $`\alpha `$ = -1.3 used for the Blue galaxy sample provides a good description of both sub-samples, neither population shows significant evidence that the faint-end slope is changing with redshift. The evolution of $`M_B^{}`$, $`\varphi ^{}`$ and $`j_B`$ for the subsamples of Moderately and Very Blue galaxies is shown in Figure 10. The top row shows how $`M_B^{}`$ changes with redshift, and it is readily apparent that the Moderately Blue galaxies are on average more luminous than the Very blue population. On the other hand, the number density of both populations (second row) does not show much evidence of significant changes; at all redshifts, the Very Blue galaxies present higher number densities than the Moderately Blue population. The luminosity density (bottom row) shows that, except for the highest redshift bin ($`z`$ 1.3), Moderately Blue galaxies output most of the optical light coming from the blue galaxy population. Overall both populations seem to evolve similarly, maintaining a constant offset in $`M_B^{}`$, while $`\varphi ^{}`$ holds constant for both halves separately. These results show that from $`z`$ 1 to the present, most of the light contributed by blue galaxies comes from galaxies with older stellar populations and/or greater dust reddening than the typical star-forming galaxy.
## 5. Summary
A sample of more than 11,000 DEEP2 galaxies from $`z=0.2`$-1.4 is used to study the evolution of galaxy luminosity functions. When DEEP2 galaxies are plotted on the color-magnitude diagram ($`M_B`$ vs. $`UB`$), blue and red galaxies occupy different loci, as seen in local samples, and this division is still clearly seen at $`z>`$ 1.0. The bimodality in the color-magnitude plane of galaxies is used to subdivide the DEEP2 sample to study how luminosity functions evolve as a function of galaxy color. In order to account for the partial sampling strategy and redshift success rate of DEEP2 as a function of color and magnitude, weights are calculated using different models describing how failed redshifts are distributed in $`z`$. The current data suggest that the vast majority of faint and blue galaxies in the DEEP2 sample for which no redshifts were successfully measured are at high redshift ($`z>1.4`$). In this work we make the assumption that red galaxies with failed $`z`$’s follow roughly the same redshift distribution as the good measurements. Given the nature of redshift failures, a compromise approach where blue failures are assumed to be at high redshift (minimal), while red failures are assumed to follow the average model is regarded as optimal. The conclusions of this work hold independently of the adopted model.
The results from this work show that populations of blue and red galaxies evolve differently. As an ensemble, blue galaxies show a larger amount of luminosity evolution, yet show little change in overall number density. Red galaxies show less change in luminosity, but a larger change in number density. When the luminosity density is considered, blue galaxies show a steady decrease toward lower redshifts, while the luminosity density of red galaxies is almost constant.
Finally, we divided the blue galaxies using the a sloping line that splits the population into two equal halves at each redshift. We find that both halves are still adequately described by a fixed faint end slope of $`\alpha `$= -1.3, and that both sub-populations evolve in a similar manner. Even in our highest redshift bins, the adopted shape of the faint end still provides a good description of the data, with no strong evidence of an increase in numbers of Very Blue galaxies at the lowest luminosity limit we probe.
A detailed comparison between the results obtained for the DEEP2 survey (this paper) with other works (Wolf et al. 2003; Bell et al. 2004; Gabasch et al. 2004; Ilbert et al. 2005) shows a good agreement. The combined results of these surveys are presented in Paper II (Faber et al. 2005), suggesting that the luminosity function of galaxies to $`z`$ 1, is currently well understood. The present paper presents the results using about a quarter of the planned DEEP2 data, and shows the potential that DEEP2 has in characterizing the properties of galaxy populations to $`z`$ 1.2. As the DEEP2 survey reaches completion, ancillary data coming from $`Z`$-band photometry by Lin and collaborators are also being obtained in DEEP2 Fields 2-4. These will allow measuring photometric redshifts for galaxies in these three fields, and will allow a far more precise characterization of the properties of galaxies with “failed” redshifts. This, combined with a 4 $`\times `$ larger sample with spectroscopic redshifts, will constitute for many years to come the main sample of galaxies at redshifts 0.7 $`z`$ 1.4, that can be used to study how galaxy populations change with time. The data for Fields 2-4 used in this paper can be retrieved from http//deep.berkeley.edu/DR1. The second data release, tentatively scheduled for late 2005, will included all the data which were used in the analysis of this paper.
The DEEP team thanks C. Wolf for several discussions regarding the color-separated luminosity function and C. Steidel for sharing unpublished redshift data. CNAW thanks G. Galaz, S. Rauzy, M. A. Hendry and K. D’Mellow for extensive discussions on the measurement of the luminosity function. Suggestions from the anonymous referee are gratefully acknowledged. The authors thank the Keck Observatory staff for their constant support during the several observing runs of DEEP2; the W. M. Keck Foundation and NASA for construction of the Keck telescopes. The DEIMOS spectrograph was funded by NSF grant ARI92-14621 and by generous grants from the California Association for Research in Astronomy, and from UCO/Lick Observatory. We also wish to recognize and acknowledge the highly significant cultural role and reverence that the summit of Mauna Kea has always had within the indigenous Hawaiian community. It is a privilege to be given the opportunity to conduct observations from this mountain. Support from National Science Foundation grants 00-71198 to UCSC and AST 00-71048 to UCB is gratefully acknowledged. SMF would like to thank the California Association for Research in Astronomy for a generous research grant and the Miller Institute at UC Berkeley for the support of a visiting Miller Professorship. JAN acknowledges support by NASA through Hubble Fellowship grant HST-HF-01132.01 awarded by the Space Telescope Science Institute, which is operated by AURA Inc. under NASA contract NAS 5-26555. Computer hardware gifts from Sun Microsystems and Quantum, Inc. are gratefully acknowledged. This research has made use of the NASA/IPAC Extragalactic Database (NED), which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. Finally, we acknowledge NASA’s (indispensable) Astrophysics Data System Bibliographic Services.
## Appendix A K-Corrections
The luminosity functions in this paper use Johnson rest-frame $`B`$ and $`UB`$ magnitudes and colors. Since $`B_{Johnson}`$ matches observed $`B`$, $`R`$ and $`I`$ only at certain redshifts, the transformation into rest-frame quantities requires the calculation of K-corrections (e.g., Oke & Sandage 1962; Hogg et al. 2002). Because of the rather limited number of bands (3 for DEEP2), the use of more robust techniques for the calculation of K-corrections as employed by COMBO-17 (Wolf et al. 2003) or SDSS (Blanton et al. 2003) is not possible. The procedure in this work is similar to that of Gebhardt et al. (2003), who used nearby galaxy SEDs from Kinney et al. (1996) to relate the observed color and magnitude at redshift $`z`$ to the rest-frame color and $`B`$-band magnitude.
We started with the 43 Kinney et al. SEDs whose spectra cover the range 1,100 Å $`\lambda `$ 10,000 Å without gaps, as listed in Table A8. Even though the Kinney et al. spectra are integrated only over a small aperture (10$`{}_{}{}^{\prime \prime }\times 20^{\prime \prime }`$) (in contrast to DEEP2 galaxy magnitudes and colors, which are close to total), this approach was chosen in preference to model spectra because the Kinney et al. data represent real spectra. The convolution between filter responses and galaxy SEDs followed Fukugita, Shimasaku & Ichikawa (1995) by resampling filters and spectra to the same dispersion (1 Å), using parabolic and linear interpolations respectively. Still following Fukugita et al. (1995), the curves for Johnson $`U`$ and $`B`$ filters come from Buser (1978) and Azusienis & Straizys (1969) respectively. The throughput curves for the CFHT $`12k\times 8k`$ DEEP2 $`BRI`$ imaging were calculated by Nick Kaiser, who provided filter transmission curves, CCD quantum efficiency curves, and the telescope response function. Normalized curves for the CFHT filters are shown in Figure A11. Calibration of these convolutions used the model atmosphere of Vega calculated by Kurucz that is distributed with the Bruzual & Charlot (2003) galaxy evolution synthesis package. The conversion between Vega and AB magnitudes (Table 1) simply compared the zero-points between the Vega calibration and that obtained using a flat spectrum in F($`\nu `$) converted into wavelength space (e.g., Fukugita et al. 1995).
Figure A12 compares synthesized $`UB`$ values for the Kinney et al. galaxies with $`UB`$ values for the same galaxies derived from the Third Revised Catalog of Galaxies (de Vaucouleurs et al. 1991, RC3). The latter were calculated using the RC3 raw total $`UB`$ colors, corrected only for Galactic absorption using the Schlegel et al. (1998) extinction values tabulated in the NASA Extragalactic Database<sup>1</sup><sup>1</sup>1http://nedwww.ipac.caltech.edu. Both sets of measurements are therefore consistent in being corrected for Galactic extinction though not for internal absorption or for a face-on geometry. The agreement is fairly good, even though the RC3 values refer to $`total`$ galaxy colors while the Kinney et al. spectra sample the center only. For the reddest templates, the synthetic spectra overerestimate $`UB`$ by $``$0.08 mag; this difference is in the expected direction of the natural internal color gradient. Overall, the good agreement in Figure A12 suggests that the zero-point of our synthetic $`UB`$ system is accurate to a few hundredths of a magnitude.
Figure A13 shows the calculated K-correction $`K_{RB}`$ (which converts $`R`$ into $`B_{Johnson}`$) as a function of synthetic observed $`RI`$ color for different redshift intervals, while Figure A14 shows calculated rest-frame $`UB`$ as a function of synthetic observed $`BR`$ in the same redshift intervals. Similar curves of $`UB`$ as a function of observed $`RI`$ for DEEP2 galaxies, are shown in Figure A15.
In general, relations are tight at redshifts where $`U`$ and $`B`$ are shifted close to the observed passbands but show more scatter as the match worsens. For redshifts beyond $``$0.7, where DEEP2 is focused, $`RI`$ color provides a much better estimate of rest-frame $`UB`$ and $`B`$ than $`BR`$.
Finally, Figure A16 compares synthetic DEEP2 $`BR`$ versus $`RI`$ colors from the Kinney et al. (1996) SEDs versus real data, binned by redshift. Observed galaxies are the red and green data points, while synthetic colors from the Kinney et al. templates are the black triangles; only 34 templates (identified in Table A1) are displayed here. A similar diagram using the whole set of 43 templates was used to select the final set. Whenever a template was an outlier compared to the observed galaxy distribution, it was flagged; templates flagged in more than two redshift bins were discarded. Galaxies that were discarded have a “no” in column (4) in Table A1 and are shown as asterisks in Figures A2 through A5.
The good agreement between observed and synthesized colors in Figure A16 suggests that, even though evolution of the template SEDs is being neglected in the present K-corrections, the errors introduced are probably small. A reason for this is that the observed color range of galaxies at all redshifts considered in this work is well covered by the spectral locus of the templates. A possible shortcoming of not using evolving SEDs for the K-correction, i.e., K+e corrections, is that at higher redshifts a portion of galaxies might shift into the wrong color class, as discussed by Wolf et al. (2003) and Bell et al. (2004). This problem is avoided in the present work by dividing galaxies into red and blue classes using the evolving “valley” of color bimodality. This does not prevent galaxies from changing color—indeed, the number of red galaxies may grow as blue galaxies migrate across the valley after star-formation quenching—but it does define classes of galaxies in a way that is independent of color zero-point errors.
Second-order polynomials were used to estimate $`UB`$ and the K-corrections from the observed colors. Custom fits were calculated (at the specific redshift of each observed galaxy) of $`UB`$ and the K-correction versus $`BR`$ and/or $`RI`$. Rest-frame parameters were obtained by entering the observed colors. The range of estimated colors (and K-corrections) was restricted to that covered by the template spectra, so that observed galaxies with extreme colors were forced to have reasonable rest-frame values. For DEEP2 galaxies, at redshifts where rest-frame $`UB`$ lies between $`BR`$ and $`RI`$, the K-corrections and rest-frame colors were obtained by interpolating between the $`BR`$ and $`RI`$ derived quantities. Otherwise, the rest-frame quantities were obtained using the closest pair of filters. The RMS error for estimated $`UB`$ ranges from 0.12 mag at $`z=1.2`$ (worst value) to 0.03 mag at redshifts where the observed filters best overlap $`UB`$. The RMS error in $`K_{RB}`$ ranges from $``$0.01 mag whenever one of the observed filters overlaps $`B_{Johnson}`$ to $``$0.15 mag at $`z1.5`$, where a large extrapolation is being used. The results obtained using the parabolic fits are comparable to the results using interpolations between SEDs (Lilly et al. 1995).
This procedure differs from that of Gebhardt et al. (2003) in two ways. First, Gebhardt et al. used nearly all the Kinney et al. (1996) templates after removing only two very deviant spectra. Second, the parabolic fit here between observed and rest-frame parameters is calculated at the exact redshift of each observed galaxy, whereas Gebhardt et al. attempted to calculate a more general polynomial that mapped the color transformation over the entire redshift range.
## Appendix B Stability of Schechter Parameters as a Function of Faint Limiting Magnitude
A limitation that is invariably present when calculating the luminosity function of galaxies is the smaller domain accessible in absolute magnitudes at higher redshifts. To examine this effect, Schechter fits were re-calculated for DEEP2 red galaxies considering three different lower-$`z`$ bins and raising the faint limit to brighter magnitudes to match the magnitude ranges accessible in the higher-$`z`$ bins. Any mismatch in the assumed shape of the luminosity function will result in a spurious drift of the fitted parameters as the magnitude limit is raised. The purpose of this test is to make sure that our measured evolutions in $`M_B^{}`$ and $`\varphi ^{}`$ for red galaxies are not contaminated by this kind of bias.
The results of this test are shown in Figure B17 for red galaxies, where $`\alpha `$ has been kept at the value $`0.5`$ used in the main text. Vertical gray lines show the limits of the data in bins $`z=`$0.8-1.0 and $`z=`$1.0-1.2. For all three lower bins, we see a drift of $`M_B^{}`$ of $``$0.1 mag toward fainter values as the samples are truncated, whereas the measured evolution is a brightening of $`M_B^{}`$ back in time. Thus, if anything, the true evolution in $`M_B^{}`$ is slightly more than claimed. The quantity $`\varphi ^{}`$ drifts upward by 0.1-0.15 dex with more truncation whereas the observed effect is a fall back in time, so again, the true evolution may be underestimated. Finally, the important quantity $`j_B`$ drifts upward by only 0.05-0.1 dex, confirming its essentially constant nature. A similar study was made for the blue galaxy sample, which shows the same behavior. This test also shows that 1/$`V_{max}`$ seems to provide a more robust estimate of the Schechter parameters than STY as the domain in absolute magnitudes decreases, though both agree within the estimated errors.
We conclude that errors caused by Schechter function mismatches are in all cases small compared to the measured evolutionary changes for red galaxies.
|
warning/0506/quant-ph0506118.html
|
ar5iv
|
text
|
# Microscopic models of quantum jump super-operators
## I Introduction
In the theory of continuous photodetection and continuous measurements the (one-count) quantum jump super-operator (QJS) is an essential part of the formalism carmichael ; plenio ; ueda1 ; ueda3 ; Gard92 ; WiseMilb ; Ban93 ; Garr94 ; agarwal ; ueda2 ; brun ; marsh , since it accounts for the loss of one photon from the electromagnetic field (EM) and corresponding photoelectron detection and counting within the measurement apparatus (MA). One of the main equations in this theory is the evolution equation of the field’s density operator $`\rho _t`$, or master equation, which reads in the simplest variant as
$$\frac{d\rho _t}{dt}=\frac{1}{i\mathrm{}}[H_0,\rho _t]\frac{\gamma }{2}\left(O^{}O\rho _t+\rho _tO^{}O2O\rho _tO^{}\right),$$
(1)
where $`H_0`$ is the EM field Hamiltonian, $`\gamma `$ is the field-MA coupling constant and $`O`$ is some lowering operator, representing the loss of a single photon from the field to the environment, that may be detected and counted by a duly constructed experimental setup. Defining the effective non-hermitian Hamiltonian as Mol75 ; Gisin92 ; Dum92 ; Molmer93
$$H_{eff}=H_0i\frac{\gamma }{2}O^{}O,$$
(2)
Eq. (1) can be written as (we set here $`\mathrm{}=1`$)
$$\frac{d\rho _t}{dt}=i\left(H_{eff}\rho _t\rho _tH_{eff}^{}\right)+\gamma O\rho _tO^{},$$
(3)
whose formal solution is (see, for example, carmichael ; Zol87 )
$`\rho _t`$ $`=`$ $`{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle _0^t}𝑑t_k{\displaystyle _0^{t_k}}𝑑t_{k1}\mathrm{}{\displaystyle _0^{t_2}}𝑑t_1e^{L\left(tt_k\right)}J`$ (4)
$`\times e^{L\left(t_kt_{k1}\right)}J\mathrm{}Je^{Lt_1}\rho _0,`$
where
$$L\rho _0=i\left[H_{eff}\rho _0\rho _0H_{eff}^{}\right],$$
$`\rho _0`$ being the density operator for the field state at $`t=0`$. The no-count super-operator $`\mathrm{exp}\left[L\left(t_kt_{k1}\right)\right]`$ evolves the initial state $`\rho _0`$ from time $`t_{k1}`$ to the latter time $`t_k`$ without taking out any photon from the field, it represents the field monitoring by a MA. The QJS $`J=\gamma OO^{}`$ is an operation which takes out instantaneously one photon from the field. Actually, Tr$`\left[J\rho _0\right]`$ is the rate of photodetection SD .
The explicit form of the QJS is not predetermined. In the phenomenological photon counting theory developed by Srinivas and Davies SD the QJS was introduced ad hoc as
$$J_{SD}=\gamma _{SD}aa^{}.$$
(5)
Later, Ben-Aryeh and Brif Aryeh and Oliveira et al. Oliveira considered QJS of the form
$$J_E=\gamma _EE_{}E_+,$$
(6)
where
$$E_{}=(a^{}a+1)^{1/2}a\mathrm{and}E_+=E_{}^{}$$
(7)
are the exponential phase operators of Susskind and Glogower susg ; CarNiet . These “non-linear” operators allow to remove some inconsistencies of the SD theory noticed by its authors.
However, the QJS (6) was introduced in Aryeh ; Oliveira also ad hoc. Therefore it is desirable to have not only a phenomenological theory, but also some microscopic models, which could justify the phenomenological schemes. The simplest example of such a model was considered for the first time in Imoto , where the QJS of Srinivas and Davies was derived under the assumption of highly efficient detection. The two fundamental assumptions of that model were: (a) infinitesimally small interaction time between the field and the MA, and (b) the presence of only few photons in the field mode. Only under these conditions one can use a simple perturbative approach and arrive at the mathematical expression for the QJS, which is independent of the details of interaction between the MA and the EM.
If the conditions (a) or (b) are not fulfilled, the QJS should depend on many factors, such as, for example, the kind of interaction between the field and MA, the interaction strength and the time $`T`$ of the interaction. Moreover, it should be emphasized, that the instant $`t_j`$ at which the quantum jump occurs cannot be determined exactly — it can happen randomly at any moment within $`T`$. Making different assumptions concerning the moment of ‘quantum jump’, one can obtain different formal expressions for the QJS. In AVS we have proposed a simple heuristic model for obtaining the ‘non-linear’ QJS of the form
$$J=\gamma F(a^{}a)aa^{}F(a^{}a).$$
(8)
In this connection, the aim of the present paper is to provide a more rigorous derivation of QJS´s, using a more sophisticated model that takes into account dissipation effects due to the ‘macroscopic part’ of the MA. Our approach is based on the hypothesis that the transition probability must be averaged over the interaction time $`T`$, during which a photon can be gobbled by the detector at any time in the interval $`(0,T)`$. Considering two different models of MA´s: a 2-level atom and a harmonic oscillator interacting with a single-mode EM field, we shall demonstrate that different kinds of interaction result in quite different QJS´s.
The plan of the paper is as follows. In Sec. II we derive the QJS using the modified Jaynes–Cummings model (with account of damping due to the spontaneous decay of the excited state) and calculating the time average of the transition operator. In Sec. III we apply the same scheme to the model of two coupled oscillators, showing explicitly how the variation of the relative strength of coupling constants results in the change of the function $`F(a^{}a)`$ in Eq. (8). Sec. IV contains a summary and conclusions.
## II Model of two-level atom detector
Let us consider first the model, which is a straightforward generalization of the one studied in Imoto . The role of the ‘system’ is played by a single mode of the electromagnetic field, while the ‘detector’ of the MA (sub-system constituting the MA that actually interacts with the EM field) consists of a single two-level atom. The Hamiltonian for the total system is chosen in the standard form of the Jaynes–Cummings model JCM
$$H_0=\frac{1}{2}\omega _0\sigma _0+\omega \widehat{n}+ga\sigma _++g^{}a^{}\sigma _{},$$
(9)
where the Pauli pseudo-spin operators $`\sigma _0`$ and $`\sigma _\pm `$ correspond to the atom ($`\sigma _+=|eg|`$, $`\sigma _{}=|ge|`$ and $`\sigma _0=|ee||gg|`$) and one considers that there were chosen two levels of the atom (the ground state $`|g`$ with frequency $`\omega _g`$ and the excited state $`|e`$ with frequency $`\omega _e=\omega _g+\omega _0`$); $`a`$, $`a^{}`$ and $`\widehat{n}=a^{}a`$ are the lowering, rising and number operators, respectively, of the EM field. Since the coupling between the field and the atom is weak, we assume that $`\omega |g|`$. Until now, the detector can absorb and emit photons back into the EM field, since the detector is not coupled to some macroscopic device that irreversibly absorbs the photons.
Therefore, we have to take into consideration that the detector is coupled to the ‘macroscopic part’ (MP) of the MA (e.g., phototube and associated electronics). Hence the detector suffers dissipative effects responsible for the spontaneous decay of the excited level of the detector (in this case of the atom). And it is precisely this physical process that represents a photodetection – the excited level of the detector decays, emitting a photoelectron into the MP of the MA, which is amplified by appropriate electronics and is seen as a macroscopic electrical current inside the MP of MA. We can take into account this dissipation effects by describing the whole photodetection process, including the spontaneous decay, by the master equation
$$\frac{d\rho _t}{dt}+i\left(H_{eff}\rho _t\rho _tH_{eff}^{}\right)=2\lambda \sigma _{}\rho _t\sigma _+,$$
(10)
which is the special case of Eq. (3), where $`O=\sigma _{}`$, $`O^{}=\sigma _+`$, $`H_{eff}=H_0i\lambda \sigma _+\sigma _{}`$, and $`2\lambda `$ is the coupling of the excited level of the atom (detector) to the MP of the MA (here we make a reasonable assumption that $`\lambda `$ has the same order of magnitude as $`|g|`$). The ‘sink’ term
$$R=2\lambda \sigma _{}\sigma _+$$
(11)
represents the $`|e|g`$ transition within the detector (the atomic decay process in this case). If $`\lambda =0`$, then the detector interacts with the EM field, but photoelectrons are not emitted (thus no counts happen), because the absorbed photons are emitted back to the field and then reabsorbed at a later time, periodically, analogously to the Rabi oscillations.
In the following, we shall use the quantum trajectories approach carmichael . The effective Hamiltonian (2) becomes
$`H_{eff}`$ $`=`$ $`Hi\lambda \sigma _+\sigma _{}={\displaystyle \frac{1}{2}}\left(\omega _0i\lambda \right)\sigma _0`$ (12)
$`+\omega \widehat{n}+ga\sigma _++g^{}a^{}\sigma _{}i\lambda /2`$
(where we have used $`\sigma _+\sigma _{}=(1+\sigma _0)/2`$) and the evolution of the system between two spontaneous decays is given by the no-count super-operator
$$𝒟_t\rho _0=U(t)\rho _0U^{}(t),U(t)=\mathrm{exp}\left(iH_{eff}t\right).$$
(13)
After a standard algebraic manipulation JCM ; Cress96 we obtain the following explicit form of the non-unitary evolution operator $`U(t)`$:
$`U(t)`$ $`=`$ $`e^{\lambda t/2}\mathrm{exp}\left[i\omega \left(\sigma _0/2+\widehat{n}\right)t\right]`$ (14)
$`\times \{{\displaystyle \frac{1}{2}}[C_{\widehat{n}+1}(t)i{\displaystyle \frac{\delta }{\left|g\right|}}S_{\widehat{n}+1}(t)](1+\sigma _0)`$
$`i{\displaystyle \frac{g}{\left|g\right|}}S_{\widehat{n}+1}(t)a\sigma _+i{\displaystyle \frac{g^{}}{\left|g\right|}}a^{}S_{\widehat{n}+1}(t)\sigma _{}`$
$`+{\displaystyle \frac{1}{2}}[C_{\widehat{n}}(t)+i{\displaystyle \frac{\delta }{\left|g\right|}}S_{\widehat{n}}(t)](1\sigma _0)\},`$
where
$$C_{\widehat{n}}(t)\mathrm{cos}\left(\left|g\right|B_{\widehat{n}}t\right),S_{\widehat{n}}(t)\mathrm{sin}\left(\left|g\right|B_{\widehat{n}}t\right)/B_{\widehat{n}},$$
(15)
$$B_{\widehat{n}}=\sqrt{\widehat{n}+\left(\delta /|g|\right)^2},\delta =\frac{1}{2}\left(\omega _0\omega i\lambda \right)$$
(16)
(note that parameter $`\delta `$ is complex and $`\widehat{n}`$ is an operator).
Assuming that the field state is $`\rho _0=\rho _F|gg|`$ at time $`t=0`$ or, analogously, the last photoemission occurred at $`t=0`$, the probability that the next photoelectron emission will occur within the time interval $`[t,t+\mathrm{\Delta }t)`$ is given by carmichael ; SD ; Cohen
$$P(t)=\mathrm{Tr}_{FD}\left[R𝒟_t\rho _0\right]\mathrm{\Delta }t,$$
(17)
(the subscripts $`F`$ and $`D`$ are a reminder that the trace operation is on field and detector spaces, respectively) where $`\mathrm{\Delta }t`$ is the time resolution of the MA. Tracing out first over the detector variables, the probability density for the next photoemission to occur at time $`t`$ will be Cohen
$$p(t)=\underset{\mathrm{\Delta }t0}{lim}\frac{P(t)}{\mathrm{\Delta }t}=\mathrm{Tr}_F\left[\mathrm{\Xi }(t)\rho _F\right],$$
(18)
where the time-dependent transition super-operator
$$\mathrm{\Xi }(t)=2\lambda \mathrm{\Gamma }(t)\mathrm{\Gamma }^{}(t),$$
(19)
acting on the EM field, stands for the photoelectron emission into the MP of the MA (i.e., the actual photodetection). Once again, the probability for detecting a photoelectron in $`[t,t+\mathrm{\Delta }t)`$ is $`P(t)=\mathrm{Tr}_F\left[\mathrm{\Xi }(t)\rho _F\right]\mathrm{\Delta }t`$ (now on we omit the subscript and write $`\rho _F\rho `$ for the field operator). In Eq. (19) $`\mathrm{\Gamma }(t)`$ is the time-dependent transition operator
$$\mathrm{\Gamma }(t)=e|U(t)|g,$$
(20)
that takes out a single photon from the field state. Substituting Eq. (14) into Eq. (20) we can write $`\mathrm{\Gamma }(t)`$ as
$$\mathrm{\Gamma }(t)=i\frac{g}{|g|}\mathrm{exp}\left(\lambda t/2i\omega \widehat{n}t\right)S_{\widehat{n}+1}(t)a,$$
(21)
so the time-dependent transition super-operator (19) becomes
$$\mathrm{\Xi }(t)\rho =2\lambda e^{\lambda t}e^{i\omega \widehat{n}t}S_{\widehat{n}+1}(t)a\rho a^{}S_{\widehat{n}+1}^{}(t)e^{i\omega \widehat{n}t}.$$
(22)
In the resonant case, $`\omega _0=\omega `$, we have
$$B_{\widehat{n}}=\sqrt{\widehat{n}\chi ^2},\chi \lambda /(2|g|).$$
(23)
If the interaction time $`\mathrm{\Delta }t`$ is small, and the number of photons in the field is not very high, in the sense that the condition
$$|g|\mathrm{\Delta }t\sqrt{n+1}1$$
(24)
is fulfilled for all eigenvalues of $`\widehat{n}`$, for which the probabilities $`p_n=n|\rho |n`$ are important, then one can replace the operator $`\mathrm{sin}\left(B_{\widehat{n}+1}|g|\mathrm{\Delta }t\right)`$ in Eq. (15) simply by $`B_{\widehat{n}+1}|g|\mathrm{\Delta }t`$ and arrive at the QJS
$$J\rho =e^{i\omega \widehat{n}\mathrm{\Delta }t}\left[2\lambda \left(|g|\mathrm{\Delta }t\right)^2a\rho a^{}\right]e^{i\omega \widehat{n}\mathrm{\Delta }t},$$
(25)
which has almost the Srinivas–Davies form (5), with the coupling constant
$$\gamma _{SD}=2\lambda \left(|g|\mathrm{\Delta }t\right)^2.$$
(26)
Taking $`2\lambda =(\mathrm{\Delta }t)^1`$ we obtain the same coupling constant $`\gamma _{SD}`$ as in Imoto , but this assumption is not the only possible. Note that super-operator (25) contains the factors $`\mathrm{exp}(\pm i\omega \widehat{n}\mathrm{\Delta }t)`$, which can be essentially different from the unit operator even under condition (24), for two reasons: (1) the condition $`|g|\mathrm{\Delta }t1`$ does not imply $`\omega \mathrm{\Delta }t1`$, because $`\omega |g|`$; (2) the condition (24) contains the square root of $`n`$, whereas the eigenvalues of $`\mathrm{exp}(\pm i\omega \widehat{n}\mathrm{\Delta }t)`$ depend on the number $`n`$ itself, which is much greater than $`\sqrt{n}`$ if $`n1`$. Consequently, even the simplest microscopic model gives rise to a QJS, which is, strictly speaking, different from the SD jump super-operator, coinciding with the former only for the diagonal elements $`|nn|`$ of the density matrix in the Fock basis.
If condition (24) is not satisfied, we propose that the QJS can be defined by averaging the transition super-operator (22) over the interaction time $`T`$, because the exact instant within $`(0,T)`$ at which the photodetection occurs in each run is unknown, so a reasonable hypothesis is that these events happen randomly with uniform probability distribution:
$$J_T\rho =\frac{1}{T}_0^T𝑑t\mathrm{\Xi }(t)\rho .$$
(27)
Writing the field density operator as
$$\rho =\underset{m,n=0}{\overset{\mathrm{}}{}}\rho _{mn}|mn|,$$
(28)
we have
$$J_T\rho =\underset{m,n=1}{\overset{\mathrm{}}{}}\rho _{mn}\sqrt{mn}f_{mn}|m1n1|,$$
(29)
where
$$f_{mn}=\frac{2\lambda }{T}_0^Te^{i\omega t(nm)\lambda t}S_m(t)S_n(t)𝑑t.$$
(30)
It is natural to suppose that the product $`\lambda T`$ is big enough, so that the photodetection can happen with high probability. Mathematically, it means that we assume that $`\mathrm{exp}(\lambda T)1`$. If $`\lambda \omega `$ (this is also a natural assumption), then the off-diagonal coefficients $`f_{mn}`$ with $`mn`$ are very small due to fast oscillations of the integrand in Eq. (27), so they can be neglected (a rough estimation gives for these terms the order of magnitude $`𝒪(\lambda /\omega )`$, compared with the diagonal coefficients $`f_{nn}`$). Consequently, the microscopic model leads to the nonlinear diagonal QJS of the form
$$J\rho =\gamma \text{diag}\left[F(\widehat{n})a\rho a^{}F(\widehat{n})\right],$$
(31)
where $`\text{diag}(\widehat{A})`$ means the diagonal part of the operator $`\widehat{A}`$ in the Fock basis. The function $`F(n)`$ can be restored from the coefficients $`f_{nn}`$ (apart the constant factor which can be included in the coefficient $`\gamma `$) as
$$F(n)=\sqrt{f_{n+1,n+1}}.$$
(32)
Under the condition $`\mathrm{exp}(\lambda T)1`$, the upper limit of integration in Eq. (27) can be extended formally to infinity, with exponentially small error. Then, taking into account the definition of the function $`S_n(t)`$ (15), we arrive at integrals of the form
$$_0^{\mathrm{}}𝑑te^{\lambda t}\times \{\begin{array}{cc}\mathrm{sin}^2(\mu t)/\mu ^2\hfill & \mathrm{for}\chi <1\hfill \\ t^2\hfill & \mathrm{for}\chi =1\hfill \\ \mathrm{sinh}^2(\mu t)/\mu ^2\hfill & \mathrm{for}\chi >1\hfill \end{array},$$
which can be calculated exactly (see, e.g., Eqs. 3.893.2 and 3.541.1 from Grad ). The final result does not depend on $`\lambda `$ or $`\chi `$ (and it is the same for either $`\chi <1`$ or $`\chi >1`$):
$$f_{nn}=(nT)^1.$$
(33)
Thus we obtain the QJS
$$J_T\rho =\gamma _T\underset{n=1}{\overset{\mathrm{}}{}}\rho _{nn}|n1n1|=\gamma _T\text{diag}\left(E_{}\rho E_+\right),$$
(34)
where $`\gamma _T=T^1`$, and the operators $`E_{}`$ and $`E_+`$ are defined by Eq. (7). Notice that, in principle, $`\gamma _T`$ is different from $`\gamma _{SD}`$. Moreover, the super-operator (34) derived from the microscopic model turns out to be different from the phenomenological QJS (6) studied in Oliveira ; AVS . The difference is that $`J_T`$ has no off-diagonal matrix elements, while $`J_E`$ has. We see that the microscopic model concerned (which can be justified in the case of big number of photons in the field mode) predicts that each photocount not only diminishes the number of photons in the mode exactly by one, but also destroys off-diagonal elements, which means the total decoherence of the field due to the interaction with MA.
Note, however, that the formula (33) holds under the assumption that the upper limit of integration in Eq. (30) can be extended to the infinity. But this cannot be done if parameter $`\chi `$ is very big. Indeed, for $`\chi >1`$ and $`\lambda T1`$, the integrand in (30) at $`t=T`$ is proportional to $`\mathrm{exp}\left[\lambda T\left(1\sqrt{1n/\chi ^2}\right)\right]`$, so it is not small when $`n/\chi ^21`$. Calculating the integral in the finite limits under the conditions $`n/\chi ^21`$ and $`\lambda T1`$, we obtain the approximate formula
$$f_{nn}=(Tn)^1\left\{1\mathrm{exp}\left[\lambda Tn/\left(2\chi ^2\right)\right]\right\},$$
(35)
which shows that $`f_{nn}`$ does not depend on $`n`$ if $`\lambda Tn/\left(2\chi ^2\right)1`$. Thus we see how the QJS (34) can be continuously transformed to the SD jump super-operator (5), when the number $`n`$ changes from big to relatively small values. It should be emphasized, nonetheless, that the off-diagonal coefficients $`f_{mn}`$ remain small even in this limit. Their magnitude approaches that of the diagonal coefficients only in the case of $`\lambda \omega `$, which does not seem to be very physical.
## III Model of harmonic oscillator detector
Now let us consider another model, where the role of the detector is played by a harmonic oscillator interacting with one EM field mode. This is a simplified version of the model proposed by Mollow mollow (for its applications in other areas see, e.g., DK96 and references therein). In the rotating wave approximation (whose validity was studied, e.g., in Ref. Estes68 ) the Hamiltonian is
$$H=\omega _aa^{}a+\omega _bb^{}b+gab^{}+g^{}a^{}b,$$
(36)
where the mode $`b`$ assumes the role of the detector and the mode $`a`$ corresponds to the EM field ($`\omega _b`$ and $`\omega _a`$ are the corresponding frequencies and $`g`$ is the detector-field coupling constant). In the following we shall repeat the same procedures we did in the section II. The dissipation effects due to the macroscopic part of the MA, associated to the mode $`b`$, can be taken into account by means of the master equation in the form
$$\frac{d\rho }{dt}+i\left[H_{eff}\rho \rho H_{eff}^{}\right]=2\lambda b\rho b^{}.$$
(37)
with the effective Hamiltonian
$`H_{eff}`$ $`=`$ $`Hi\lambda b^{}b=\left(\omega _bi\lambda \right)b^{}b`$ (38)
$`+\omega _aa^{}a+gba^{}+g^{}b^{}a.`$
The evolution operator $`U(t)=\mathrm{exp}(iH_{eff}t)`$ for the quadratic Hamiltonian (38) can be calculated by means of several different approaches Dbook . Here we use the algebraic approach Ban93 ; Wei ; alg ; inter , since Hamiltonian (38) is a linear combination of the generators of algebra $`su(1,1)`$
$$K_+b^{}a,K_{}ba^{},K_0(b^{}ba^{}a)/2,$$
$$[K_0,K_\pm ]=\pm K_\pm ,[K_{},K_+]=2K_0.$$
The evolution operator can be factorized as
$$U(t)=e^{i\mathrm{\Omega }tN}e^{A(t)K_+}e^{B(t)K_0}e^{C(t)K_{}},$$
(39)
where
$$N\left(b^{}b+a^{}a\right)/2,\mathrm{\Omega }\omega _b+\omega _ai\lambda .$$
The time-dependent coefficients are
$$A(t)=\frac{ig^{}\mathrm{sin}(\eta t)}{\eta \mathrm{{\rm Y}}\left(t\right)},C(t)=\frac{ig\mathrm{sin}(\eta t)}{\eta \mathrm{{\rm Y}}\left(t\right)},$$
(40)
$$B(t)=2\mathrm{ln}\mathrm{{\rm Y}}\left(t\right),$$
(41)
with
$$\mathrm{{\rm Y}}\left(t\right)=\mathrm{cos}(\eta t)+i\left[\omega _{ba}/(2\eta )\right]\mathrm{sin}(\eta t),$$
(42)
$$\omega _{ba}\omega _b\omega _ai\lambda ,\eta \left(|g|^2+\omega _{ba}^2/4\right)^{1/2}.$$
(43)
Assuming that the detector is in resonance with the EM field’s mode one gets $`\omega _{ba}=i\lambda `$ and
$$\mathrm{{\rm Y}}\left(t\right)=\mathrm{cos}(\eta _0t)+\left[\lambda /(2\eta _0)\right]\mathrm{sin}(\eta _0t),$$
(44)
$$\eta _0=\left(|g|^2\lambda ^2/4\right)^{1/2}.$$
(45)
If, initially, the detector oscillator is in the ground state $`|0_b`$, the time-dependent transition operator, corresponding to the absorption of one photon from the EM field, defined in (20), is
$`\mathrm{\Gamma }(t)`$ $`=`$ $`1_b|U(t)|0_b`$ (46)
$`=`$ $`A(t)\mathrm{exp}\left[{\displaystyle \frac{1}{2}}\left(i\mathrm{\Omega }t+B(t)\right)(a^{}a+1)\right]a`$
and the transition super-operator becomes
$`\mathrm{\Xi }(t)\rho `$ $`=`$ $`2\lambda |A(t)|^2\mathrm{exp}\left[{\displaystyle \frac{1}{2}}\left(i\mathrm{\Omega }t+B(t)\right)(a^{}a+1)\right]`$ (47)
$`\times a\rho a^{}\mathrm{exp}\left[{\displaystyle \frac{1}{2}}\left(i\mathrm{\Omega }^{}tB^{}(t)\right)(a^{}a+1)\right].`$
For “small” $`t=\mathrm{\Delta }t`$ and few photons in the cavity, the QJS (25) is recovered. Considering, instead, the time-averaged QJS, one has Eqs. (27)-(29). For $`\chi =\lambda /(2|g|)<1`$ (when the parameter $`\eta _0`$ is real) one can represent the coefficients $`f_{mn}`$ as (we consider the resonance case with $`\omega _a=\omega _b=\omega `$)
$`f_{mn}`$ $`=`$ $`{\displaystyle \frac{4\chi }{T(1\chi ^2)^{3/2}}}{\displaystyle _0^Z}𝑑z[\mathrm{cos}(z)+\xi \mathrm{sin}(z)]^{m+n2}`$ (48)
$`\times \mathrm{sin}^2(z)\mathrm{exp}\left[i\overline{\omega }z(nm)\xi z(m+n)\right],`$
where
$$\xi =\frac{\chi }{\sqrt{1\chi ^2}},\overline{\omega }=\frac{\omega }{|g|\sqrt{1\chi ^2}},Z=\frac{\lambda T}{2\xi }.$$
(49)
Since the parameter $`\overline{\omega }`$ is big, the off-diagonal coefficients $`f_{mn}`$ with $`nm`$ are very small due to the strongly oscillating factor $`\mathrm{exp}[i\overline{\omega }t(nm)]`$. Consequently, they can be neglected in the first approximation, and we arrive again at the diagonal QJS of the form (31).
We notice that the exact analytical expression for the integral in Eq. (48) is so complicated (even if $`m=n`$), that it is difficult to use it. For example, in the limit $`\chi 1`$ Eq. (48) can be reduced to the form
$$f_{nn}=\frac{4}{T}_0^{\lambda T/2}𝑑yy^2(1+y)^{2n2}\mathrm{exp}(2ny).$$
(50)
Replacing the upper limit by the infinity, we recognize the integral representation of the Tricomi confluent hypergeometric function $`\mathrm{\Psi }(a;c;z)`$ Bate . Thus we have (neglecting small corrections of the order of $`\mathrm{exp}(\lambda T)`$)
$$f_{nn}=\frac{8}{T}\mathrm{\Psi }(3;2n+2;2n).$$
(51)
Although the $`\mathrm{\Psi }`$-function in the right-hand side of Eq. (51) can be rewritten in terms of the associated Laguerre polynomials Bate as
$$\mathrm{\Psi }(3;2n+2;2n)=\frac{(2n)!}{2(2n)^{1+2n}}L_{2n2}^{(12n)}(2n),$$
(52)
neither Eq. (51) nor Eq. (52) help us to understand the behavior of the coefficient $`f_{nn}`$ as function of $`n`$. Therefore it is worth trying to find simple approximate formulas for the integral in (48).
If $`\chi 1`$, then also $`\xi 1`$, so we can neglect the term $`\xi \mathrm{sin}(z)`$ in the integrand of Eq. (48) and the function $`\mathrm{sin}^2(z)[\mathrm{cos}(z)]^{2n2}`$ can be replaced by its average value taken over the period $`2\pi `$ of fast (in the scale determined by the characteristic time $`\xi ^1`$) oscillations. After simple algebra we obtain (replacing the upper limit of integration $`Z`$ by infinity)
$$f_{nn}=\frac{4(2n2)!}{T(2^nn!)^2},\chi 1.$$
(53)
Using Stirling’s formula $`n!\sqrt{2\pi n}(n/e)^n`$, we can write for $`n1`$
$$f_{nn}\left(T\sqrt{\pi n^5}\right)^1.$$
(54)
This function corresponds to the QJS (31) with
$$F(\widehat{n})=F_5(\widehat{n})(\widehat{n}+1)^{5/4},\gamma =\gamma _5(T\sqrt{\pi })^1.$$
(55)
Thus, differently from the case of two-level detector, in the simplest version of the oscillator detector model the lowering operator contains the factor $`(\widehat{n}+1)^{5/4}`$, instead of $`(\widehat{n}+1)^{1/2}`$ as in the “E-model” (6) or simply $`\widehat{1}`$ as in the SD model (5).
The case $`\chi 1`$ is not very realistic from the practical point of view, since it corresponds to the detector with very low efficiency. However, we can calculate the integral (48) with arbitrary $`\xi `$ approximately, assuming that $`n1`$ and using the method of steepest descent. Rewriting the integrand as $`\mathrm{exp}[G(z)]`$, one can easily verify that the points of maxima of the function
$$G(z)=2\mathrm{ln}[\mathrm{sin}(z)]+2(n1)\mathrm{ln}[\mathrm{cos}(z)+\xi \mathrm{sin}(z)]2\xi nz$$
are given by the formula $`z_k=\pm z_0+k\pi `$, where
$$z_0=\mathrm{tan}^1(\mu ),\mu =\left(\xi ^2n+n1\right)^{1/2},$$
(56)
$`k=0,1,2,\mathrm{}`$ for the plus sign and $`k=1,2,\mathrm{}`$ for the minus sign. One can verify that
$$\mathrm{exp}\left[G(z_k)\right]=\frac{\mu ^2(1+\xi \mu )^{2n2}}{(1+\mu ^2)^n}\mathrm{exp}\left(2z_0\xi n2\xi \pi nk\right).$$
(57)
The second derivatives of the function $`G(z)`$ at the points of maxima do not depend on $`k`$:
$$G^{\prime \prime }(z_k)=\frac{4n(\xi ^2+1)}{1+\xi \mu }.$$
(58)
Using Eqs. (57) and (58) and performing summation over $`k`$ we find (taking $`Z=\mathrm{}`$)
$$f_{nn}=\frac{\chi \sqrt{8\pi }(1+\xi \mu )^{2n3/2}\mathrm{exp}\left(2z_0\xi n\right)}{T\sqrt{n}(n+\chi ^21)(1+\mu ^2)^n}\mathrm{coth}\left(\xi n\pi \right),$$
(59)
Although the application of the steepest descent method can be justified for $`n1`$, formula (59) seems to be a good approximation for $`n1`$, too. For example, for $`n=1`$ (when $`\mu =\xi ^1`$) it yields
$$Tf_{11}4\chi \sqrt{\pi }\mathrm{coth}(\pi \xi )\mathrm{exp}\left[2\xi \mathrm{tan}^1\left(\xi ^1\right)\right],$$
(60)
and the numerical values of (60) in the whole interval $`0<\chi <1`$ are not very far from the exact value $`Tf_{11}=1`$, which holds independently of $`\chi `$, as far as the upper limit of integration in (48) can be extended to the infinity.
For $`n1`$ (when $`\mu 1`$) Eq. (59) can be simplified as
$$f_{nn}(\chi )\frac{\chi \sqrt{8\pi }}{eT}n^{3/2}\mathrm{coth}\left(\frac{\chi n\pi }{\sqrt{1\chi ^2}}\right),\chi 1.$$
(61)
For $`\chi 1`$ the function (61) assumes the form (54), with slightly different coefficient $`\gamma ^{}=(eT)^1\sqrt{8/\pi }1.04\gamma _5`$.
For $`\chi >1`$ (when parameter $`\eta _0`$ is imaginary) we have, instead of (48), the integral (considering diagonal coefficients only)
$`f_{nn}`$ $`=`$ $`{\displaystyle \frac{4\chi }{T(\chi ^21)^{3/2}}}{\displaystyle _0^Y}𝑑z[\mathrm{cosh}(z)+\zeta \mathrm{sinh}(z)]^{2n2}`$ (62)
$`\times \mathrm{sinh}^2(z)\mathrm{exp}\left(2n\zeta z\right),`$
where
$$\zeta =\chi /\sqrt{\chi ^21},Y=\lambda T/(2\zeta ).$$
(63)
Applying again the steepest descent method, we have now the only point of maximum
$$z_{max}=\mathrm{tanh}^1(\nu ),\nu =\left[\left(\zeta ^21\right)n+1\right]^{1/2}.$$
(64)
Taking into account the value of the second derivative of the logarithm of integrand at this point,
$$G^{\prime \prime }(z_{max})=\frac{4n(\zeta ^21)}{1+\zeta \nu },\zeta \nu =\frac{\chi }{\sqrt{n+\chi ^21}},$$
(65)
we obtain
$$f_{nn}=\frac{\chi \sqrt{8\pi }(1+\zeta \nu )^{2n3/2}(1\nu )^{n(\zeta 1)}}{T\sqrt{n}(n+\chi ^21)(1+\nu )^{n(\zeta +1)}}.$$
(66)
One can check that the limit of formula (66) at $`\chi 1`$ coincides with the analogous limit of formula (59), so the transition through the point $`\chi =1`$ is continuous.
The asymptotical form of (66) for $`n\chi ^2`$ is the same as (61), except for the last factor:
$$f_{nn}(\chi )\frac{\chi \sqrt{8\pi }}{eT}n^{3/2},\chi 1,$$
(67)
Applying the steepest descent method to the integral (50) (for $`n1`$), we obtain the same result (67) with $`\chi =1`$. Thus for $`\chi 1`$ (not too small and not too big) we obtain the QJS in the form (31) with
$$F(\widehat{n})=F_3(\widehat{n})(\widehat{n}+1)^{3/4},\gamma =\gamma _3\frac{\chi \sqrt{8\pi }}{eT}.$$
(68)
For very big values of parameter $`\chi `$ (exceeding $`\sqrt{n}`$) the steepest descent method cannot be used, because the second derivative of the logarithm of integrand, given by Eq. (65), becomes small, and because the coordinate $`z_{max}`$, determined by Eq. (64), tends to infinity, while the upper limit $`Y`$ of integration in (62) tends to the fixed value $`\lambda T/2`$. For $`\chi 1`$, Eq. (68) holds for the values of $`n`$ satisfying approximately the inequality $`n>n_{}4\chi ^2\mathrm{exp}(\lambda T)`$. If $`n<n_{}`$, then it can be shown that Eq. (62) leads to the same approximate formula (35) as in the model of two-level detector, so the SD super-operator (however, without off-diagonal elements) is restored for relatively not very big values of $`n`$.
In Figures 1 and 2 we show the dependence of diagonal coefficients $`f_{nn}`$ on the number $`n`$ for different values of parameter $`\chi `$, obtained by numerical integration of (48) and (62) for the fixed value of the parameter $`\lambda T=10`$; in figure 3 we compare them with the approximate analytical formulas (59) and (66). We see that the coincidence is rather satisfactory for big values of $`n`$, although there are some differences for $`n1`$. We also see in Figure 2 that the increase of parameter $`\chi `$ results in the appearance of the SD plateau for small values of $`n`$, which goes into a slope corresponding to the power-law dependence for big values of $`n`$. The height of plateaus diminishes as $`\chi ^2`$ in accordance with Eq. (35), because big values of $`\chi `$ correspond (for fixed values of $`\lambda `$ and $`T`$) to small coupling coefficient $`|g|^2`$ between the field and MA and, consequently, low probability of photocount.
## IV Conclusions
Here we presented two microscopic models for deducting QJS’s. In the first one we supposed that the detector behaves like a 2-level atom, and in the second – as a harmonic oscillator. The main difference between our models and previous ones is that we take into account the dissipative effects that arise when one couples the actual detector to the phototube. This scheme includes the spontaneous decay of the detector with originated photoelectron emission inside the phototube, which is amplified and viewed as macroscopic electric current. Using quantum trajectories approach we deduced general time-dependent transition super-operator, responsible for taking out a single photon from the field. Since it depends explicitly on interaction time, we proposed two distinct schemes for obtaining time independent QJS´s from it. In the first case we assumed that the interaction time is small and that there are few photons in the cavity; in this situation we recovered the QJS proposed by Srinivas and Davies in both detector models. As a second scheme, we calculated time-averaged QJS on the time interval during which a photon is certainly absorbed; as the result, we obtained different non-linear QJS’s for the 2-level atom model and the model of harmonic oscillator. In particular, we have shown that for quantum states with the predominant contribution of Fock components with big values of $`n`$, the QJS has the nonlinear form (31) with the power-law asymptotic function $`F(\widehat{n})=(\widehat{n}+1)^\beta `$. However, the concrete value of the exponent $`\beta `$ is model-dependent. For the 2-level atom model we obtained $`\beta =1/2`$, whereas in the model of harmonic oscillator the values $`\beta =5/4`$ and $`\beta =3/4`$ were found, depending on the ratio between the spontaneous decay frequency of the excited state and the effective frequency of coupling between the detector and field mode. Also, we have demonstrated how the simple Srinivas–Davies QJS arises in the case of states with small number of photons. Another important result we obtained is that the QJS’s, when applied to density matrix’ non-diagonal elements, are null in average in both models due to the strong oscillations of the free field terms.
###### Acknowledgements.
Work supported by FAPESP (SP, Brazil) contracts # 00/15084-5, 04/13705-3. SSM and VVD acknowledge partial financial support from CNPq (DF, Brazil).
|
warning/0506/gr-qc0506137.html
|
ar5iv
|
text
|
# References
DAMTP-2005-58
June 2005
Nonlocal Effective Field Equations for Quantum Cosmology
H. W. Hamber <sup>1</sup><sup>1</sup>1e-mail address : HHamber@uci.edu
Department of Physics and Astronomy
University of California
Irvine, CA 92697-4575, USA
and
R. M. Williams <sup>2</sup><sup>2</sup>2e-mail address : R.M.Williams@damtp.cam.ac.uk
Girton College, Cambridge CB3 0JG, and
Department of Applied Mathematics and Theoretical Physics
Centre for Mathematical Sciences
Wilberforce Road, Cambridge CB3 0WA, United Kingdom.
ABSTRACT
The possibility that the strength of gravitational interactions might slowly increase with distance, is explored by formulating a set of effective field equations, which incorporate the gravitational, vacuum-polarization induced, running of Newton’s constant $`G`$. The resulting long distance (or large time) behaviour depends on only one adjustable parameter $`\xi `$, and the implications for the Robertson-Walker universe are calculated, predicting an accelerated power-law expansion at later times $`t\xi 1/H`$.
In the Standard Model of particle interactions, all gauge couplings are known to run with energy. Recent non-perturbative studies of quantum gravity have suggested that the gravitational coupling too may depend on a scale related to curvature, and therefore macroscopic in size. In this Letter, we investigate the effects of a running gravitational constant $`G`$ at large distances. This scale dependence is assumed to be driven by gravitational vacuum polarization effects, which produce an anti-screening effect some distance away from the primary source, and therefore tend to increase the strength of the gravitational coupling. A power law running of $`G`$ will be implemented via manifestly covariant nonlocal terms in the effective gravitational action and field equations.
We start by assuming that for Newton’s constant one has
$$G(r)=G(0)\left[\mathrm{\hspace{0.33em}1}+c_\xi (r/\xi )^{1/\nu }+O((r/\xi )^{2/\nu })\right],$$
(1)
where the exponent $`\nu `$ is generally related to the derivative of the beta function for pure gravity evaluated at the non-trivial ultraviolet fixed point. Recent studies have $`\nu ^1`$ varying between 3.0 and 1.7 . These estimates rely on three different, and unrelated, nonperturbative approaches to quantum gravity, based on the lattice path integral formulation, the two plus epsilon expansion of continuum gravity, and a momentum slicing scheme combined with renormalization group methods in the vicinity of flat space, respectively. In all three approaches a non-vanishing, positive bare cosmological constant is required for the consistency of the renormalization group procedure. The mass scale $`m=\xi ^1`$ in Eq. (1) is supposed to determine the magnitude of quantum deviations from the classical theory. It seems natural to identify $`1/\xi ^2`$ with either some very large average spatial curvature scale, or perhaps more appropriately with the Hubble constant (as measured today) determining the macroscopic expansion rate of the universe, via the correspondence
$$\xi =\mathrm{\hspace{0.33em}1}/H,$$
(2)
in a system of units for which the speed of light equals one. A possible concrete scenario is one in which $`\xi ^1=H_{\mathrm{}}=lim_t\mathrm{}H(t)=\sqrt{\mathrm{\Omega }_\mathrm{\Lambda }}H_0`$ with $`H_{\mathrm{}}^2=\frac{\mathrm{\Lambda }}{3}`$, where $`\mathrm{\Lambda }`$ is the observed cosmological constant, and for which the horizon radius is $`H_{\mathrm{}}^1`$.
As it stands, the formula for the running of $`G`$ is coordinate dependent, and we therefore replace it with a manifestly covariant expression involving the covariant d’Alembertian operator
$$\mathrm{}=g^{\mu \nu }_\mu _\nu ,$$
(3)
whose Green’s function in $`d`$ spatial dimensions is known to behave as
$$<x|\frac{1}{\mathrm{}}|y>\delta (rd(x,y|g))\frac{1}{r^{D2}},$$
(4)
where $`d`$ is the minimum distance between points $`x`$ and $`y`$ in a background with metric $`g_{\mu \nu }`$. We therefore write, in four dimensions,
$$GG(\mathrm{})=G(0)\left[\mathrm{\hspace{0.33em}1}+c_{\mathrm{}}\left(\frac{1}{\xi ^2\mathrm{}}\right)^{1/2\nu }+O((\xi ^2\mathrm{})^{1/\nu })\right].$$
(5)
One way of incorporating this is to replace the gravitational action
$$I=\frac{1}{16\pi G}𝑑x\sqrt{g}R$$
(6)
by
$$I=\frac{1}{16\pi G}𝑑x\sqrt{g}\left(1c_{\mathrm{}}\left(\frac{1}{\xi ^2\mathrm{}}\right)^{1/2\nu }+O((\xi ^2\mathrm{})^{1/\nu })\right)R.$$
(7)
The above prescription has in fact been used successfully to systematically incorporate the effects of radiative corrections in an effective action formalism . It should be noted that the coefficient $`c_\xi `$ in Eq. (1) is expected to be a calculable number of order one, not necessarily the same as the coefficient $`c_{\mathrm{}}`$, as $`r`$ and $`1/\sqrt{\mathrm{}}`$ are clearly rather different entities to begin with. One should recall here that in general the form of the covariant d’Alembertian operator $`\mathrm{}`$ depends on the specific tensor nature of the object it is acting on.
The details of the incorporation of this modified $`G`$ in the gravitational side of the Einstein equations are given elsewhere . Here we shall describe instead its incorporation on the matter side of Einstein’s equations, giving the effective field equations
$$R_{\mu \nu }\frac{1}{2}g_{\mu \nu }R+\mathrm{\Lambda }g_{\mu \nu }=\mathrm{\hspace{0.33em}8}\pi G\left(1+A(\mathrm{})\right)T_{\mu \nu },$$
(8)
where we have replaced $`G(r)`$ by $`G(0)(1+A(\mathrm{}))`$. These can be written in the form
$$R_{\mu \nu }\frac{1}{2}g_{\mu \nu }R+\mathrm{\Lambda }g_{\mu \nu }=\mathrm{\hspace{0.33em}8}\pi G\stackrel{~}{T}_{\mu \nu },$$
(9)
with $`\stackrel{~}{T}_{\mu \nu }=\left(1+A(\mathrm{})\right)T_{\mu \nu }`$ defined as an effective, or gravitationally dressed, energy-momentum tensor. Just like the ordinary Einstein gravity case, in general $`\stackrel{~}{T}_{\mu \nu }`$ might not be covariantly conserved a priori, $`^\mu \stackrel{~}{T}_{\mu \nu }\mathrm{\hspace{0.17em}0}`$, but ultimately the consistency of the effective field equations demands that it be exactly conserved in consideration of the Bianchi identity satisfied by the Riemann tensor. The ensuing new covariant conservation law
$$^\mu \stackrel{~}{T}_{\mu \nu }^\mu \left[\left(1+A(\mathrm{})\right)T_{\mu \nu }\right]=\mathrm{\hspace{0.33em}0}$$
(10)
can be then be viewed as a constraint on $`\stackrel{~}{T}_{\mu \nu }`$ (or $`T_{\mu \nu }`$) which, for example, in the specific case of a perfect fluid, will imply again a definite relationship between the density $`\rho (t)`$, the pressure $`p(t)`$ and the Robertson-Walker scale factor $`R(t)`$, just as it does in the standard case.
For simplicity we set the cosmological constant $`\mathrm{\Lambda }`$ to zero from now on and consider first the trace of the effective field equations
$$R=\mathrm{\hspace{0.33em}8}\pi G\left(1+A(\mathrm{})\right)T_\mu ^\mu .$$
(11)
The advantage of this is that, initially, we need to consider the action of $`\mathrm{}`$ only on a scalar function, $`S(x)`$ say, which is given by
$$\frac{1}{\sqrt{g}}_\mu g^{\mu \nu }\sqrt{g}_\nu S(x).$$
(12)
For the Robertson-Walker metric,
$$ds^2=dt^2+R^2(t)\left\{\frac{dr^2}{1kr^2}+r^2\left(d\theta ^2+\mathrm{sin}^2\theta d\phi ^2\right)\right\},$$
(13)
$`\mathrm{}`$ acting on a scalar function of $`t`$ only, is
$$\frac{1}{R^3(t)}\frac{}{t}\left[R^3(t)\frac{}{t}\right].$$
(14)
The energy-momentum tensor for a perfect fluid is given by
$$T_{\mu \nu }=\left(p(t)+\rho (t)\right)u_\mu u_\nu +g_{\mu \nu }p(t).$$
(15)
We consider a pressureless fluid $`p(t)=0`$ and assume the density and scale factor are given by powers of $`t`$, as in the classical solution for the RW metric: $`\rho (t)=\rho _0t^\beta ,R(t)=R_0t^\alpha `$. Then
$$\mathrm{}^n\left(T_\mu ^\mu \right)=\mathrm{}^n\left(\rho (t)\right)\mathrm{\hspace{0.33em}4}^n(1)^{n+1}\frac{\mathrm{\Gamma }(\frac{\beta }{2}+1)\mathrm{\Gamma }(\frac{\beta +3\alpha +1}{2})}{\mathrm{\Gamma }(\frac{\beta }{2}+1n)\mathrm{\Gamma }(\frac{\beta +3\alpha +1}{2}n)}\rho _0t^{\beta 2n}.$$
(16)
We may analytically continue the exponent to negative fractional $`n`$, and obtain with $`n=1/(2\nu )`$, an expression for $`(1+A(\mathrm{}))`$ acting on the trace of $`T_{\mu \nu }`$, given by
$$\left(1+c_\xi \left(\frac{t}{\xi }\right)^{1/\nu }\right)\rho _0t^\beta ,$$
(17)
with
$$c_\nu =\mathrm{\hspace{0.17em}4}^{1/2\nu }(1)^{11/2\nu }\frac{\mathrm{\Gamma }(\frac{\beta }{2}+1)\mathrm{\Gamma }(\frac{\beta +3\alpha +1}{2})}{\mathrm{\Gamma }(\frac{\beta }{2}+1+\frac{1}{2\nu })\mathrm{\Gamma }(\frac{\beta +3\alpha +1}{2}+\frac{1}{2\nu })}.$$
(18)
Using the value of the scalar curvature for the Robertson-Walker metric in the $`k=0`$ case,
$$R=\mathrm{\hspace{0.33em}6}\left(\dot{R}^2(t)+R(t)\ddot{R}(t)\right)/R^2(t),$$
(19)
gives
$$\frac{6\alpha \left(2\alpha 1\right)}{t^2}=\left(1+c_\xi \left(\frac{t}{\xi }\right)^{1/\nu }\right)\rho _0t^\beta .$$
(20)
For large $`t`$, when the correction term starts to take over, we see from the powers of $`t`$ that
$$\beta =21/\nu .$$
(21)
Next we will examine the full effective field equations (as opposed to just their trace part) of Eq. (8) with $`\mathrm{\Lambda }=0`$,
$$R_{\mu \nu }\frac{1}{2}g_{\mu \nu }R=\mathrm{\hspace{0.33em}8}\pi G\left(1+A(\mathrm{})\right)T_{\mu \nu }.$$
(22)
Here the d’Alembertian operator
$$\mathrm{}=g^{\mu \nu }_\mu _\nu $$
(23)
acts on a second rank tensor,
$`_\nu T_{\alpha \beta }=_\nu T_{\alpha \beta }\mathrm{\Gamma }_{\alpha \nu }^\lambda T_{\lambda \beta }\mathrm{\Gamma }_{\beta \nu }^\lambda T_{\alpha \lambda }I_{\nu \alpha \beta }`$
$$_\mu \left(_\nu T_{\alpha \beta }\right)=_\mu I_{\nu \alpha \beta }\mathrm{\Gamma }_{\nu \mu }^\lambda I_{\lambda \alpha \beta }\mathrm{\Gamma }_{\alpha \mu }^\lambda I_{\nu \lambda \beta }\mathrm{\Gamma }_{\beta \mu }^\lambda I_{\nu \alpha \lambda },$$
(24)
and would thus seem to require the calculation of 1920 terms, of which fortunately many vanish by symmetry. Assuming that $`T_{\mu \nu }`$ describes a pressureless perfect fluid, we obtain
$`\left(\mathrm{}T_{\mu \nu }\right)_{tt}`$ $`=`$ $`\mathrm{\hspace{0.33em}6}\rho (t)\left({\displaystyle \frac{\dot{R}(t)}{R(t)}}\right)^2\mathrm{\hspace{0.17em}3}\dot{\rho }(t){\displaystyle \frac{\dot{R}(t)}{R(t)}}\ddot{\rho }(t)`$
$`\left(\mathrm{}T_{\mu \nu }\right)_{rr}`$ $`=`$ $`{\displaystyle \frac{1}{1kr^2}}\left(2\rho (t)\dot{R}(t)^2\right)`$
$`\left(\mathrm{}T_{\mu \nu }\right)_{\theta \theta }`$ $`=`$ $`r^2(1kr^2)\left(\mathrm{}T_{\mu \nu }\right)_{rr}`$
$`\left(\mathrm{}T_{\mu \nu }\right)_{\phi \phi }`$ $`=`$ $`r^2(1kr^2)\mathrm{sin}^2\theta \left(\mathrm{}T_{\mu \nu }\right)_{rr},`$ (25)
with the remaining components equal to zero. Note that a non-vanishing pressure contribution is generated in the effective field equations, even if one assumes initially a pressureless fluid. As before, repeated applications of the d’Alembertian $`\mathrm{}`$ to the above expressions leads to rapidly escalating complexity (for example, eighteen distinct terms are generated by $`\mathrm{}^2`$ for each of the above contributions), which can only be tamed by introducing some further simplifying assumptions. In the following we will therefore assume as before that $`k=0`$, $`\rho (t)=\rho _0t^\beta `$, and $`R(t)=R_0t^\alpha `$. We obtain
$`\left(\mathrm{}T_{\mu \nu }\right)_{tt}`$ $`=`$ $`\left(6\alpha ^2\beta ^23\alpha \beta +\beta \right)\rho _0t^{\beta 2}`$
$`\left(\mathrm{}T_{\mu \nu }\right)_{rr}`$ $`=`$ $`\mathrm{\hspace{0.33em}2}R_0^2t^{2\alpha }\alpha ^2\rho _0t^{\beta 2},`$ (26)
which again shows that the $`tt`$ and $`rr`$ components get mixed by the action of the $`\mathrm{}`$ operator, and that a non-vanishing $`rr`$ component gets generated, even though it was not originally present.
The geometric side of the gravitational field equations, the Einstein tensor, has the following components for the RW metric:
$`G_{tt}`$ $`=`$ $`\mathrm{\hspace{0.33em}3}\dot{R}^2(t)/R^2(t)`$
$`G_{rr}`$ $`=`$ $`{\displaystyle \frac{1}{1kr^2}}\left(\dot{R}^2(t)+2R(t)\ddot{R}(t)\right)`$
$`G_{\theta \theta }`$ $`=`$ $`r^2\left(1kr^2\right)G_{rr}`$
$`G_{\phi \phi }`$ $`=`$ $`\mathrm{sin}^2\theta G_{\theta \theta }.`$ (27)
Then with $`k=0`$ and $`R(t)=R_0t^\alpha `$, these will all behave like $`t^2`$ so in fact a solution can only be achieved at order $`\mathrm{}^n`$ provided the exponent $`\beta `$ satisfies $`\beta =2+2n`$, or since $`n=1/(2\nu )`$,
$$\beta =2\mathrm{\hspace{0.17em}1}/\nu ,$$
(28)
as was found previously from the trace equation, Eqs. (11) and (21).
We must now determine $`\alpha `$. By repeated application of $`\mathrm{}`$, for general $`n`$ one can then write
$$\left(\mathrm{}^nT_{\mu \nu }\right)_{tt}c_{tt}(\alpha ,\nu )\rho _0t^2$$
(29)
and similarly for the $`rr`$ component
$$\left(\mathrm{}^nT_{\mu \nu }\right)_{rr}c_{rr}(\alpha ,\nu )R_0^2t^{2\alpha }\rho _0t^2.$$
(30)
But remarkably (see also Eq. (25) ) one finds for the two coefficients the simple identity
$$c_{rr}(\alpha ,\nu )=\frac{1}{3}c_{tt}(\alpha ,\nu )$$
(31)
as well as $`c_{\theta \theta }=r^2c_{rr}`$ and $`c_{\phi \phi }=r^2\mathrm{sin}^2\theta c_{rr}`$. Then for large times, when the quantum correction starts to become important, the $`tt`$ and $`rr`$ field equations reduce to
$$3\alpha ^2t^2=\mathrm{\hspace{0.33em}8}\pi Gc_{tt}(\alpha ,\nu )\rho _0t^2$$
(32)
and
$$\alpha (3\alpha \mathrm{\hspace{0.17em}2})R_0^2t^{2\alpha 2}=\mathrm{\hspace{0.33em}8}\pi Gc_{rr}(\alpha ,\nu )R_0^2t^{2\alpha }\rho _0t^2$$
(33)
respectively. But the identity $`c_{rr}=\frac{1}{3}c_{tt}`$ implies, simply from the consistency of the $`tt`$ and $`rr`$ effective field equations at large times,
$$\frac{c_{rr}(\alpha ,\nu )}{c_{tt}(\alpha ,\nu )}\frac{1}{3}=\frac{3\alpha 2}{3\alpha },$$
(34)
whose only possible solution finally gives the second sought-for result, namely
$$\alpha =\frac{1}{2}.$$
(35)
We still need to check whether the above solution is consistent with covariant energy conservation. With the assumed form for $`T_{\mu \nu }`$ it is easy to check that energy conservation yields for the $`t`$ component
$$\left(^\mu \left(\mathrm{}^nT_{\mu \nu }\right)\right)_t\left((3\alpha +\beta +\mathrm{\hspace{0.17em}1}/\nu )c_{tt}+\mathrm{\hspace{0.17em}3}\alpha c_{rr}\right)\rho _0t^{\beta +1/\nu 1}=\mathrm{\hspace{0.33em}0}$$
(36)
when evaluated for $`n=1/2\nu `$, and zero for the remaining three spatial components. But from the solution for the matter density $`\rho (t)`$ at large times one has $`\beta =21/\nu `$, so the above zero condition gives again $`c_{rr}/c_{tt}=(3\alpha 2)/3\alpha `$, exactly the same relationship previously implied by the consistency of the $`tt`$ and $`rr`$ field equations.
Let us emphasize that the values for $`\alpha =1/2`$ of Eq. (35) and $`\beta =21/\nu `$ of Eq. (28), determined from the effective field equations at large times, are found to be consistent with both the field equations and covariant energy conservation. More importantly, the above solution is also consistent with what was found previously by looking at the trace of the effective field equations, Eq. (11), which also implied the result $`\beta =21/\nu `$, Eq. (21).
The classical unmodified matter-dominated RW equations have solutions $`\alpha =2/3`$, $`\beta =2`$, which mean that the scale factor behaves as
$$R(t)t^\alpha t^{2/3}$$
(37)
and the density as
$$\rho (t)t^\beta t^2((R(t))^3.$$
(38)
This will also be the behaviour for our model at early times, but at later times, when the effect of the gravitational vacuum-polarization modification dominates, the behaviour is rather different: for the scale factor, we have
$$R(t)t^\alpha t^{1/2}$$
(39)
and for the density
$$\rho (t)t^\beta t^{21/\nu }\left(R(t)\right)^{2(2+1/\nu )}.$$
(40)
Thus the density decreases significantly faster in time than the classical value ($`\rho (t)t^2`$), a signature of an accelerating expansion at later times.
Within the Friedmann-Robertson-Walker (FRW) framework the gravitational vacuum polarization term behaves in many ways like a positive pressure term. The value $`\alpha =1/2`$ corresponds to $`\omega =1/3`$ in
$$\alpha =\frac{2}{3(1+\omega )},$$
(41)
(this follows from the consistency of the $`rr`$ and $`tt`$ equations in the general case) where we have taken the pressure and density to be related by $`p(t)=\omega \rho (t)`$, which is therefore characteristic of radiation. One can therefore visualize the gravitational vacuum polarization contribution as behaving like ordinary radiation, in the form of a dilute virtual graviton gas. It should be emphasized though that the relationship between density $`\rho (t)`$ and scale factor $`R(t)`$ is very different from the classical case.
Acknowledgements
The authors are grateful to Gabriele Veneziano for his close involvement in the early stages of this work, and for bringing to our attention the work of Vilkovisky and collaborators on nonlocal effective actions for gauge theories. Both authors also wish to thank the Theory Division at CERN for warm hospitality and generous financial support during the Summer of 2004. One of the authors (HWH) wishes to thank James Bjorken for useful discussions on issues related to the subject of this paper. The work of Ruth Williams was supported in part by the UK Particle Physics and Astronomy Research Council.
|
warning/0506/hep-th0506167.html
|
ar5iv
|
text
|
# Braneworld Flux Inflation
## I Introduction
Recent cosmological observations such as the WMAP results are consistent with the inflationary scenario. Hence, we are prompted to seek the inflaton in a unified theory of particle physics. Currently, it is widely believed that the most promising candidate for a unified theory is superstring theory. Interestingly, the superstring theory predicts the existence of the extra dimensions. In order to reconcile this prediction with our observed 4-dimensional universe, we need a mechanism to hide the extra dimensions. For a long time, the Kaluza-Klein compactification was considered to be the unique choice. However, recent developments in superstring theory suggest a braneworld picture where we are living on 4-dimensional hypersurface embedded in a higher dimensional spacetime Horava . This braneworld picture not only gives a way for the superstring theory to be phenomenologically viable but also suggests a new inflationary scenario, so-called brane inflation Dvali ; Kachru .
In the brane inflation scenario, the radion, the distance between branes, plays the role of the inflaton and inflation is terminated by the brane collision. This is nice because the inflation is realized purely geometrical manner without introducing an ad-hoc scalar field. In this scenario, however, branes are treated as test branes. On the other hand, relativistic cosmologists have studied the braneworld gravity intensively review . In these studies, the effect of the bulk geometry on the 4-dimensional braneworld cosmology have been a central concern, though inflation is usually assumed to be driven by the fundamental scalar field either on the brane or in the bulk. In this braneworld cosmology, the self-gravity of branes are properly treated and hence, it is clear how to calculate corrections due to the bulk effect MWBH ; LMWgw ; KKS ; HS . Taking a look at both approaches, we have come up with the idea of incorporating geometrical inflation in a simple Randall-Sundrum (RS) model RS1 .
In this paper, we would like to propose an inflationary scenario driven by the flux generated by a brane that is charged with respect to a five-form field strength. The idea is very similar to that of brane inflation but we take into account the self-gravity of branes. Our model is constructed in the RS framework RS1 . We suppose that initially two positive tension branes are inflating as de Sitter spacetimes embedded in an anti-de Sitter bulk. Eventually, they collide with each other and inflation will end. Subsequently, two branes are assumed to coallesce and evolve as a single $`Z_2`$-symmetric positive tension brane. The gravitational theory is non-singular and the model we construct is essentially 5-dimensional way. Except at the collision point, we can use the effective action obtained by the low energy approximation KS1 ; KS2 ; wiseman . But in the 4-dimensional Einstein frame, the evolution of the universe is discontinuous at the collision point, which means that the effective 4-dimensional theory breaks down. However at the collision point we can use 5-dimensional energy-momentum conservation to determine the dynamics LMW . We are thus able to analyze the spectrum of primordial scalar and tensor fluctuations produced after the collision. It turns out that a curvaton-type mechanism is required to generate the primordial density perturbations producing the present structure of the universe.
The organization of this paper is as follows. In sec.II, the basic setup is presented. In sec.III, our cosmological scenario is described. Inflation driven by the flux is analyzed both in the induced metric frame and the Einstein frame. The consistency analysis gives the expansion rate after the collision determined by the expansion rate of both branes before the collision. The cosmological history after the collision is briefly summarized. In sec.IV, the spectrum of fluctuations are calculated. The sec.V is devoted to conclusions. In the appendix, the detailed derivation of the effective action is provided.
## II Basic Setup
The point of brane inflation is that no fundamental scalar field is necessary and the exit from inflation is realized by the collision of branes. What we want to do is to incorporate this idea into the codimension-one RS braneworld model.
We consider a two-brane system where one has a $`Z_2`$ symmetry and the other does not. Hereafter, we call the former the boundary brane and the latter the bulk brane. As we show in an appendix such a set-up can be realised as the limiting case of a three-brane system, with two boundary branes, where one of the $`Z_2`$-symmetric branes is sent to infinity.
The model is described by the 5-dimensional action
$`S`$ $`=`$ $`{\displaystyle \frac{1}{2\kappa ^2}}{\displaystyle d^5x\sqrt{g}\left[\stackrel{(5)}{R}+2\mathrm{\Lambda }\right]}+{\displaystyle \underset{i=1}{\overset{2}{}}}{\displaystyle \frac{1}{\kappa ^2}}{\displaystyle d^4x\sqrt{h_i}K^i}`$ (1)
$`{\displaystyle \frac{1}{25!}}{\displaystyle d^5x\sqrt{g}F_5^2}+{\displaystyle \underset{i=1}{\overset{2}{}}}\mu _i{\displaystyle C_4}`$
$`+{\displaystyle \frac{1}{4!}}{\displaystyle d^5x_A\left(\sqrt{g}F^{ABCDE}C_{BCDE}\right)}`$
$`+{\displaystyle \underset{i=1}{\overset{2}{}}}{\displaystyle d^4x\sqrt{h_i}\left[\sigma _i+_{\mathrm{matter}}^i\right]}`$
where $`\kappa ^2`$ is the 5-dimensional gravitational coupling constant and $`\stackrel{(5)}{R}`$ is the 5-dimensional curvature. Both our branes have positive tension, $`\sigma _1,\sigma _2>0`$, but opposite charges, $`\mu _1>0,\mu _2<0`$, which couple to a 4-form potential field, $`C_4=(1/4!)C_{ABCD}dx^Adx^Bdx^Cdx^D`$. The bulk brane separates the bulk into regions $`I`$ and $`II`$ (see FIG.1).
We denote the induced metric on each brane by $`h_{i\mu \nu }`$ and $`K^i`$ denotes the trace-part of the extrinsic curvature of each brane. Here we have taken into account the Gibbons-Hawking boundary term instead of introducing delta-function singularities in the five-dimensional curvature. We incorporated the 5-form field $`F_5=dC_4`$ which can change the effective cosmological constant in the bulk $`\mathrm{\Lambda }`$ TS . The third line represents the surface term which is introduced to make the variation of the action with fixed $`F_5`$ consistent.
Let us take the coordinate system
$`ds^2=dy^2+g_{\mu \nu }(y,x)dx^\mu dx^\nu `$ (2)
The Latin indices $`\{A,B,\mathrm{}\}`$ and the Greek indices $`\{\mu ,\nu ,\mathrm{}\}`$ are used for tensors defined in the bulk and on the brane, respectively. The 5-form equation of motion becomes
$`_M\left(\sqrt{g}F^{0123y}\right)dx^M+{\displaystyle \underset{i=1}{\overset{2}{}}}\mu _i\delta (y\varphi _i(x))dy=0`$ (3)
where $`\varphi _i(x)`$ denotes the position of each brane. In each bulk ($`y\varphi _i`$), it is easy to solve Eq. (3) as
$`F_I^{0123y}={\displaystyle \frac{c_I}{\sqrt{g}}},F_{II}^{0123y}={\displaystyle \frac{c_{II}}{\sqrt{g}}}.`$ (4)
where $`c_I,c_{II}`$ are constants of integration. Because of the charge conservation, we have $`c_{II}=0`$ as is explained in the appendix. Thus we see $`F_5`$ has no local dynamics. As $`F_5`$ is not dynamical, we can eliminate it from the action by simply substituting the above solution into the original action. This can be done using the equations of motion to give
$`{\displaystyle \frac{1}{4!}}{\displaystyle d^5x_A\left(\sqrt{g}F^{ABCDE}C_{BCDE}\right)}`$
$`={\displaystyle \frac{1}{5!}}{\displaystyle d^5xF_5^2}+{\displaystyle d^5x_y\left(\sqrt{g}F^{0123y}\right)C_{0123}}`$
$`={\displaystyle _I}d^5xc_I^2{\displaystyle \underset{i=1}{\overset{2}{}}}\mu _i{\displaystyle d^4xC_{0123}}.`$ (5)
Substituting Eq. (5) into the original action (1), we see that 4-form potentials are cancelled and the effect of 5-form field strength is indistinguishable from a cosmological constant term in the bulk. Thus, the resultant action is
$`S`$ $`=`$ $`{\displaystyle \frac{1}{2\kappa ^2}}{\displaystyle _I}d^5x\sqrt{g}\left[\stackrel{(5)}{R}+2\mathrm{\Lambda }\kappa ^2c_I^2\right]`$ (6)
$`+{\displaystyle \frac{1}{2\kappa ^2}}{\displaystyle _{II}}d^5x\sqrt{g}\left[\stackrel{(5)}{R}+2\mathrm{\Lambda }\right]`$
$`+{\displaystyle \underset{i}{}}{\displaystyle \frac{1}{\kappa ^2}}{\displaystyle \sqrt{h_i}K^i}`$
$`+{\displaystyle \underset{i}{}}{\displaystyle d^4x\sqrt{h_i}\left[\sigma _i+_{\mathrm{matter}}^i\right]}.`$
The 5-form field strength $`F_5`$ in region $`I`$ acts like a change in the effective 5-dimensional cosmological constant in region $`I`$, $`\mathrm{\Lambda }_I\mathrm{\Lambda }\kappa ^2c_I^2/2`$. Consequently, two bulk regions have a different cosmological constant. The absolute value of the cosmological constant in region I is assumed to be small and therefore we shall see that the effective cosmological constants induced on both branes are positive.
If the expansion rate of the second brane is faster than that of first, both branes will eventually collide with each other and the opposing charges annihilate. The resulting brane tension is assumed to become (close to) the Randall-Sundrum tuning value so that the inflation ends and the universe becomes radiation dominated. We assume a completely inelastic collision so that the kinetic energy of the branes is transformed into radiation energy density on the brane. The subsequent evolution is same as that of the radiation dominated universe in the RSII brane model. Thus, the original flux generated by the charged branes has caused de Sitter inflation of branes and the exit from inflation is realized by the brane collision. In the following sections, we shall look at the details of this scenario.
## III Flux Inflation
The non-linear dynamics of de Sitter branes embedded in 5-dimensional anti-de Sitter space can be studied exactly without recourse to any approximation kaloper ; BCG ; radion . However the study of inhomogeneous perturbations about this background, when two de Sitter branes are in relative motion, is a much harder problem. Therefore we will use a low-energy approximation KS1 ; wiseman ; KS2 ; sugumi in this case, valid when the Hubble rate is smaller than the anti-de Sitter scale. In this case the only extra degree of freedom coming from the 5-dimensional gravitational field is the radion, a scalar in the 4-dimensional effective theory, describing the distance between the two branes.
Before the collision, the radion field is non-minimally coupled to the 4-dimensional gravitational field on either brane and the system is described by the scalar-tensor theory. Hence, it is useful to look at the cosmological evolution both from the induced metric frame on the bulk brane and the Einstein frame. After the collision the radion field vanishes and the low-energy system is described by 4-dimensional Einstein gravity. As the collision changes the theory discontinuously, the evolution of the universe in the Einstein frame looks strange. We find a contracting universe immediately before the collision which will start to expand abruptly after the collision. However, in the induced metric frame which is a natural frame for an observer, the universe is always expanding.
### III.1 Inflation in the induced metric frame
Except for the collision point, the low energy approximation can be applied KS1 ; wiseman ; KS2 ; sugumi . The detailed derivation of the effective action for our system can be found in the appendix. An alternative derivation is given in Ref. Ludo . The induced metric on the bulk brane is $`h_{\mu \nu }`$. The low-energy effective action on the bulk brane is
$`S={\displaystyle \frac{\mathrm{}_I}{2\kappa ^2}}{\displaystyle d^4x\sqrt{h}\left[\left(\mathrm{\Psi }^2+\alpha 1\right)R(h)+6(\mathrm{\Psi })^2V\right]},`$ (7)
where $`\mathrm{\Psi }`$ represents the radion field, with $`\mathrm{\Psi }1`$ when the branes are coincident. The AdS length scale $`\mathrm{}_I`$ is given in Eq. (47). For simplicity, we denote $`h^{\alpha \beta }_\beta \mathrm{\Psi }_\alpha \mathrm{\Psi }`$ as $`(\mathrm{\Psi })^2`$. The effective potential for the radion is given by
$`{\displaystyle \frac{\mathrm{}_I^2}{12}}V=\left(\beta _11\right)\mathrm{\Psi }^4\left({\displaystyle \frac{1}{\alpha }}1\beta _2\right).`$ (8)
Here we have defined the dimensionless parameters
$`\alpha ={\displaystyle \frac{\mathrm{}_{II}}{\mathrm{}_I}},\beta _1={\displaystyle \frac{\kappa ^2\sigma _1\mathrm{}_I}{6}},\beta _2={\displaystyle \frac{\kappa ^2\sigma _2\mathrm{}_I}{3}}.`$ (9)
We obtain the static case of single Minkowski brane at fixed distance in AdS when $`\beta _1=1`$, $`\beta _2=0`$ and $`\alpha =1`$. As we have $`c_{II}=0`$, the effective cosmological constant in region $`I`$, $`\mathrm{\Lambda }_I=\mathrm{\Lambda }\kappa ^2c_I^2/2`$, is always smaller than that in region $`II`$, $`\mathrm{\Lambda }`$, due to the flux in the region $`I`$. From (A2), we see $`\mathrm{}_I>\mathrm{}_{II}`$, i.e. $`\alpha <1`$.
The equation of motion for the radion is
$`\mathrm{}\mathrm{\Psi }{\displaystyle \frac{\mathrm{\Psi }^21+\alpha }{12(1\alpha )}}{\displaystyle \frac{V}{\mathrm{\Psi }}}+{\displaystyle \frac{\mathrm{\Psi }}{3(1\alpha )}}V(\mathrm{\Psi })=0.`$ (10)
The dynamics of the radion field thus appear non-trivial, and as the effective theory in the induced metric frame is a scalar-tensor theory, the cosmological dynamics will also depend on this non-trivial dynamics. However, the equations of motion for the induced metric can be written as
$`G_{\mu \nu }={\displaystyle \frac{6}{\mathrm{}_I^2(1\alpha )}}\left({\displaystyle \frac{1}{\alpha }}1\beta _2\right)h_{\mu \nu }+E_{\mu \nu }`$ (11)
where $`E_{\mu \nu }`$ represents the projected 5-dimensional Weyl tensor on the brane SMS and is determined by the radion field,
$`E_{\mu \nu }`$ $`=`$ $`{\displaystyle \frac{6}{\mathrm{}_I^2(1\alpha )}}\left({\displaystyle \frac{1}{\alpha }}1\beta _2\right)h_{\mu \nu }`$ (12)
$`+{\displaystyle \frac{2\mathrm{\Psi }}{\mathrm{\Psi }^21+\alpha }}\left(_\mu _\nu \mathrm{\Psi }h_{\mu \nu }\mathrm{}\mathrm{\Psi }\right)`$
$`{\displaystyle \frac{4}{\mathrm{\Psi }^21+\alpha }}\left(_\mu \mathrm{\Psi }_\nu \mathrm{\Psi }{\displaystyle \frac{1}{4}}h_{\mu \nu }(\mathrm{\Psi })^2\right)`$
$`{\displaystyle \frac{1}{2(\mathrm{\Psi }^21+\alpha )}}h_{\mu \nu }V.`$
This satisfies the traceless condition $`E^\mu {}_{\mu }{}^{}=0`$ and indicates the radion field behaves as the conformally invariant matter on the brane. Hence, for the isotropic and homogeneous Universe,
$`ds^2=dt^2+a^2(t)\left(dx^2+dy^2+dz^2\right),`$ (13)
the effect of the bulk gravity, $`E_{\mu \nu }`$, acts like a radiation fluid, and the Friedman equation is obtained as
$`H_2^2\mathrm{}_I^2={\displaystyle \frac{2}{1\alpha }}\left({\displaystyle \frac{1}{\alpha }}1\beta _2\right)+{\displaystyle \frac{C}{a^4}},`$ (14)
where $`H_2`$ denotes the Hubble parameter of the induced spacetime and the constant of integration $`C`$ is often called the dark radiation. As we know $`\alpha <1`$, in order to have the positive effective cosmological constant, we assume
$`{\displaystyle \frac{1}{\alpha }}1\beta _2>0.`$ (15)
In practice, once the universe starts to expand, the dark radiation term $`\frac{C}{a^4}`$ becomes soon negligible. Thus, we see that the induced spacetime on the brane rapidly approaches de Sitter at late times ($`a(t)\mathrm{}`$). We can obtain enough e-foldings on the brane, if we fine tune the initial brane positions (and hence the initial value of $`\mathrm{\Psi }`$, see Fig. 2). Thus, we have obtained an inflationary universe on the brane.
When $`\mathrm{\Psi }`$ reaches $`1`$, the inflation is suddenly terminated by the collision of branes in our scenario.
### III.2 View from the Einstein frame
We have shown that we can obtain a de Sitter inflationary universe in the induced metric frame due to the existence of the flux between the two branes in the bulk. It is interesting to see this in the conformally related Einstein frame nojiri . In the Einstein frame, the metric is $`\gamma _{\mu \nu }=[\mathrm{\Psi }^2+(\alpha 1)]h_{\mu \nu }`$, using the variable $`\mathrm{\Psi }^2=(1\alpha )\mathrm{coth}^2\psi `$. Then the action reduces to
$`S`$ $`=`$ $`{\displaystyle \frac{\mathrm{}_I}{2\kappa ^2}}{\displaystyle d^4x\sqrt{\gamma }\left[R(\gamma )6\left(\psi \right)^2U(\psi )\right]}`$ (16)
where the effective potential for the minimally-coupled radion $`\psi `$ is given by
$`{\displaystyle \frac{\mathrm{}_I^2}{12}}U=\left(\beta _11\right)\mathrm{cosh}^4\psi {\displaystyle \frac{\frac{1}{\alpha }1\beta _2}{(1\alpha )^2}}\mathrm{sinh}^4\psi .`$ (17)
The above potential is depicted in Figure 3. The unstable extremum corresponds to the static two de Sitter brane solution.
This extremum is located at $`\psi _0`$ determined by
$`(\beta _11)\mathrm{cosh}^2\psi _0={\displaystyle \frac{\frac{1}{\alpha }1\beta _2}{(1\alpha )^2}}\mathrm{sinh}^2\psi _0.`$ (18)
In that case, the potential energy becomes
$`{\displaystyle \frac{\mathrm{}_I^2}{12}}U|_{\psi =\psi _0}={\displaystyle \frac{\frac{1}{\alpha }1\beta _2}{(1\alpha )^2}}\mathrm{sinh}^2\psi _0.`$ (19)
The effective mass-squared is negative,
$`{\displaystyle \frac{\mathrm{}_I^2}{12}}{\displaystyle \frac{d^2U}{d\psi ^2}}|_{\psi =\psi _0}=8{\displaystyle \frac{\frac{1}{\alpha }1\beta _2}{(1\alpha )^2}}\mathrm{sinh}^2\psi _0.`$ (20)
From the above, one can read off the radion effective mass $`m_r^2=4H_0^2`$, where $`H_0`$ represents the Hubble parameter at $`\psi _0`$. This indicates the linearised instability of the static two de Sitter brane system, consistent with previous analyses GenSas ; CF ; Contaldi .
Let us assume two branes are separated enough initially, which means the radion is located at near the maximum. The radion then starts to roll down the hill and reaches the collision point $`\mathrm{coth}^2\psi =1/\sqrt{1\alpha }`$ which corresponds to $`\mathrm{\Psi }=1`$. In contrast to the other models discussing collisions Khoury ; Gen ; Bucher ; KSS ; GT ; Pillado , our model has no singularity at the collision because the bulk region never disappears in our model.
The Hubble parameter in the Einstein frame, $`H_E`$, is related to the Hubble parameter in the induced metric frame, $`H_2`$, as
$`H_E={\displaystyle \frac{1}{\sqrt{\mathrm{\Psi }^21+\alpha }}}H_2+{\displaystyle \frac{\mathrm{\Psi }\dot{\mathrm{\Psi }}}{\left(\mathrm{\Psi }^21+\alpha \right)^{3/2}}}.`$ (21)
The typical behavior of the Hubble parameter in the Einstein frame is depicted in Figure 4. While $`\mathrm{\Psi }`$ remains almost constant the Hubble rate is also nearly constant and we have almost exponential expansion in both the induced and Einstein frames. Due to the suppression factor $`1/\sqrt{\mathrm{\Psi }^21+\alpha }`$, the Hubble rate in the Einstein frame is much smaller than that in the induced frame. However, the e-folding number is almost the same as that in the Jordan frame because the time in the Einstein frame becomes longer by the conformal factor $`\sqrt{\mathrm{\Psi }^2+(\alpha 1)}`$. Shortly before the collision, we find that the effective potential in the Einstein frame becomes negative, the universe recollapses and immediately before the collision the universe is contracting in the Einstein frame, though the universe is always expanding in the induced metric frame.
### III.3 Graceful Exit Through Brane Collision
We need a fully 5-dimensional consideration to give a rule for evolution through the collision. We consider the simplest case of a completely inelastic collision where the bulk brane is absorbed by the boundary brane.
To describe the $`Z_2`$-symmetric collision of the branes it is useful to consider the complete system of four branes, consisting of the incoming $`Z_2`$ symmetric boundary brane, two copies of the bulk brane, and the outgoing $`Z_2`$ symmetric brane (see Figure 5).
A detailed analysis gives consistency conditions for the collision Neronov ; LMW . These are equivalent to relativistic energy-momentum conservation LMW
$`\rho _f=\sigma _1+2\gamma _{2|1}\sigma _2`$ (22)
where the pseudo-Lorentz factor between the colliding branes can be written as $`\gamma _{2|1}\mathrm{cosh}(\theta _2\theta _1)`$ where
$$\mathrm{sinh}\theta _1=H_1\mathrm{}_1,\mathrm{sinh}\theta _2=H_2\mathrm{}_1.$$
(23)
At low energy (small “angles”) this reduces to the “Newtonian” energy conservation law
$`\rho _f=\sigma _1+2\sigma _2,`$ (24)
plus the momentum conservation law
$`H_f={\displaystyle \frac{1}{\alpha }}\left[H_1(1\alpha )H_2\right].`$ (25)
Hence, in order to obtain an expanding universe after the collision we have to impose the constraint
$`H_1>(1\alpha )H_2.`$ (26)
In addition to this constraint, we require that the expansion rate of the bulk brane is bigger than the $`Z_2`$ symmetric brane, $`H_2>H_1`$, in order to cause a collision. It is easy to meet both these requirements, as we set $`\alpha <1`$.
### III.4 Cosmological Evolution After the Collision
After the collision, we assume the branes coalesce and behave as a single brane. We further assume the resulting brane tension is given by the RS value, $`\kappa ^2\sigma _f=6/\mathrm{}_{II}`$ so that the effective cosmological constant on the brane vanishes after the collision. This is the brane-world equivalent of the usual assumption that the inflaton potential is zero at its minimum.
At the collision, the additional energy density (above the RS brane tension) is assumed to be transferred to light degrees of freedom on the brane, i.e., radiation. Hence, the subsequent evolution will be governed by the standard Friedmann equation for the hot big bang with small Kaluza-Klein corrections.
As the radion disappears after the collision, the difference between the induced metric frame and the Einstein frame also disappears. In the Einstein frame, the abrupt change of the contracting phase to the expanding phase cannot be described within the 4-dimensional effective theory. This clearly shows that our model is different from a conventional 4-dimensional inflationary model.
## IV Perturbations
Having constructed a homogeneous cosmological model, we can now consider the spectrum of inhomogeneous perturbations that would be expected due to small-scale quantum fluctuations. As there is no singularity at the collision, we can unambiguously follow the evolution of fluctuations generated during inflation through the collision.
To study the behavior of fluctuations before the collision, it is convenient to work in the Einstein frame in which the radion is minimally coupled to the metric. We have the equations of motion in the Einstein frame
$`G_{\mu \nu }=6\left(_\mu \psi _\nu \psi {\displaystyle \frac{1}{2}}\gamma _{\mu \nu }(\psi )^2\right){\displaystyle \frac{1}{2}}\gamma _{\mu \nu }U(\psi )`$
$`\mathrm{}\psi {\displaystyle \frac{1}{12}}{\displaystyle \frac{dU}{d\psi }}=0.`$ (27)
The homogeneous and istropic background metric is
$`ds^2=b^2(\eta )\left[d\eta ^2+\delta _{ij}dx^idx^j\right].`$ (28)
and we have the Einstein equations become
$`3^2=3\psi ^2+{\displaystyle \frac{1}{2}}b^2U(\psi )`$ (29)
$`^2+2^{}=3\psi ^2+{\displaystyle \frac{1}{2}}b^2U(\psi )`$ (30)
$`\psi ^{\prime \prime }+2\psi ^{}+{\displaystyle \frac{b^2}{12}}{\displaystyle \frac{U}{\psi }}=0.`$ (31)
where a prime denotes a derivative with respect to the conformal time $`\eta `$ and $`=b^{}/b`$. During the de Sitter phase, the solution is
$`b={\displaystyle \frac{1}{H_0\eta }},\psi =\psi _0,`$ (32)
where $`bH_0=_0`$ and
$`\mathrm{}_I^2H_0^2=2{\displaystyle \frac{\frac{1}{\alpha }1\beta _2}{(1\alpha )^2}}\mathrm{sinh}^2\psi _0.`$ (33)
Now we can examine possible fluctuations separately.
### IV.1 Gravitational waves
Let us consider first the tensor perturbations
$`ds^2=b^2(\eta )\left[d\eta ^2+\left(\delta _{ij}+q_{ij}\right)dx^idx^j\right],`$ (34)
where the tensor perturbations satisfy $`q^i{}_{i}{}^{}=0,q_{ij,j}=0`$. We can reduce the Einstein equations to
$`q_{ij}^{\prime \prime }+2q_{ij}^{}+k^2q_{ij}=0.`$ (35)
Before the radion starts to roll down the hill, the background spacetime is the de Sitter spacetime (32). The positive frequency mode function is the standard one
$`q_{ij}(H_0\eta )^{3/2}H_{3/2}^{(1)}(k\eta ),`$ (36)
where $`H_{3/2}^{(1)}`$ is the Hankel’s function of the first kind. This gives the standard flat spectrum for the primordial gravitational waves. During the roll down phase, the universe will begin to contract rapidly from a numerical calculation, we see that the contracting phase is negligible in practice. During the collision process, as the gravitational waves are independent of gauge, we will have a flat spectrum on long wavelengths after the collision when the standard radiation dominated era begins.
### IV.2 Radion fluctuations
To study the behavior of the radion fluctuations, we express the metric perturbation in the Einstein frame as
$`ds^2=b^2[(1+2A)d\eta ^2+2_iBdx^id\eta `$
$`+((1+2)\delta _{ij}+2_i_jE)dx^idx^j],`$ (37)
where $`A,B,,E`$ represent the gauge-dependent scalar metric perturbations. A convenient gauge-invariant combination is the comoving curvature perturbation, which is the intrinsic curvature perturbation on uniform-radion hypersurfaces:
$`_c={\displaystyle \frac{\delta \psi }{\psi ^{}}}.`$ (38)
The second-order action for the curvature perturbation $`_c`$ is
$`S={\displaystyle \frac{1}{2}}{\displaystyle }d\eta d^3xz^2\left[_c^{}{}_{}{}^{2}_c^{|i}_{c|i}\right],`$ (39)
where
$`z=\sqrt{{\displaystyle \frac{3\mathrm{}_I}{2\kappa ^2}}}{\displaystyle \frac{b\psi ^{}}{}}.`$ (40)
The equation of motion for $`_c`$ is
$$_c^{\prime \prime }+2\frac{z^{}}{z}_c^{}+k^2_c=0.$$
(41)
Therefore, on large scales, $`_c`$ is constant.
Equivalently we can work in terms of the radion on uniform-curvature hypersurfaces
$`Q=\delta \psi {\displaystyle \frac{\psi ^{}}{}}={\displaystyle \frac{\psi ^{}}{}}_c.`$ (42)
During the de Sitter phase, the equation for $`Q`$ can be written as
$`Q^{\prime \prime }+2Q^{}+(k^24H_0^2b^2)Q=0,`$ (43)
where we have used Eq. (31). Hence, the positive frequency mode becomes
$`Q(H_0\eta )^{3/2}H_{5/2}^{(1)}(k\eta ),`$ (44)
where $`H_{5/2}^{(1)}`$ is the Hankel’s function of the first kind. Thus, we can read off the power spectrum of $`Q`$ as $`𝒫_Qk^2`$. Despite the exponential expansion the spectrum on large scales becomes red during the de Sitter inflation because the radion has negative effective mass-squared.
These field fluctuations can be translated to the curvature perturbations on comoving hypersurfaces via Eq. (42). The coefficient $`\psi ^{}/`$ is independent of scale and hence the comoving curvature perturbation shares the same red spectrum, $`𝒫_{}k^2`$. Near the collision, we expect the spectrum to be blue because of the rapid contraction, but this only affects small scales.
Finally, we need to calculate the curvature perturbation on uniform-density hypersurfaces, $`\zeta `$, on large scales after the collision. This should then remain constant on large scales for adiabatic perturbations after the collision, even in the brane-world Langlois , simply as a consequence of local energy conservation WMLL .
In the low-energy limit, energy conservation at the collision gives Eq. (24) which implies that the collision hypersurface will be a uniform-energy hypersurface. Thus $`\zeta `$ after the collision coincides with the curvature perturbation of this collision hypersurface.
We define the collision hypersurface in terms of the low-energy effective theory before the collision by the condition that $`\mathrm{\Psi }=1`$. Thus the collision hypersurface is a uniform-radion hypersurface and we can identify the curvature perturbation on this hypersurface as the comoving curvature perturbation, $`\zeta =_c`$ (where the negative sign comes from different historical conventions for the sign of the curvature perturbation).
One might worry that $`_c`$ was calculated in the Einstein frame and the collision hypersurface corresponds to a physical hypersurface on the two branes. In fact the comoving curvature perturbation is invariant under any conformal transformation that is function of the radion field, $`\mathrm{\Omega }^2(\mathrm{\Psi })`$, as the conformal transformation then corresponds to a uniform rescaling on uniform-$`\mathrm{\Psi }`$ hypersurfaces, leaving the comoving curvature perturbation conformally invariant. Hence $`_c`$ does describe the curvature perturbation on the physical collision hypersurface.
Alternatively we can calculate the curvature perturbation on the collision hypersurface using the $`\delta N`$ formalism Sasaki . This says that the curvature perturbation is given by the perturbed expansion with respect to an initial flat hypersurface, or $`\zeta =(dN/d\varphi _{})\delta \varphi _{}`$ where $`\delta \varphi _{}`$ is the field perturbation on the initial spatially flat hypersurface at horizon exit. This gives the usual expression $`_c=(H/\dot{\psi })Q`$ for the comoving curvature perturbation during inflation. Because the collision hypersurface is a uniform-$`\psi `$ hypersurface before the collision and a uniform-density hypersurface after the collision we have $`\zeta =_c=(H/\dot{\psi })Q`$.
However, we have shown that the comoving curvature perturbation has a steeply tilted red spectrum which means that these perturbations cannot be responsible for the formation of the large scale structure in our Universe.
### IV.3 Scale-invariant perturbations
The curvaton mechanism EnqSlo ; LW ; MT provides a simple example of how scale-invariant perturbations could be generated in our model. Both the bulk and boundary branes experience a de Sitter expansion before the collision. The radion field acquires a red spectrum due to its negative mass-squared, but any minimally-coupled, light scalar field living on either brane would acquire a scale-invariant spectrum of perturbations before the collision.
Thus we add to our basic model an additional degree of freedom which, to be specific, we assume lives on the bulk brane. The action for the curvaton $`\chi `$ in the induced metric frame takes the simple form
$`S_{\mathrm{curvaton}}={\displaystyle d^4x\sqrt{h}\left[\frac{1}{2}(\chi )^2\right]}`$ (45)
Vacuum fluctuations of a massless scalar field in de Sitter spacetime have the same time-dependence as the graviton (36), becoming frozen-in on large scales. Normalising to the Bunch-Davies vacuum state on small scales leads to the standard scale-invariant spectrum, $`(H_2/2\pi )^2`$, on super-Hubble scales. Hence, the fluctuations of the curvaton have a completely flat spectrum.
This is similar the way the curvaton mechanism was originally proposed in the pre-big bang scenario EnqSlo ; LW where axion fields have a non-trivial coupling to the dilaton field in the Einstein frame, but are minimally coupled in a conformally related frame, which can give rise to a scale-invariant spectrum if the expansion is de Sitter in that conformal frame CEW .
The curvaton mechanism requires that the curvaton has a non-zero mass after the collision, which seems entirely natural if the collision is associated with some symmetry-breaking in the degrees of freedom on the brane. The energy density of any massive scalar field will grow relative to the radiation density once the Hubble rate drops below its effective mass. Indeed the particles must decay before primordial nucleosynthesis to avoid an early matter domination in conflict with standard predictions of the abundances of the light elements. But if the curvaton comes to contribute a significant fraction of the total energy density before it decays then perturbations in the curvaton will result in primordial density perturbations on large scales capable of seeding the large scale structure in our Universe EnqSlo ; LW ; MT .
## V Conclusions
We have proposed a geometrical model of inflation where inflation on two branes is driven by the flux generated by the brane charge and terminated by the brane collision with charge annihilation. We assume the collision process is completely inelastic and the kinetic energy is transformed into the thermal energy. After the collision the two branes coalesce together and behave as a single brane universe with zero effective cosmological constant.
The four-dimensional, low-energy effective theory in the Einstein frame has to change abruptly at the collision point. Therefore, our model needs to be consistently described using 5-dimensional gravity. This can be done using consistency conditions at the collision that ensure energy-momentum conservation in the 5-dimensional theory. As the collision process has no singularity in the 5-dimensional gravity, we can unambiguously follow the evolution of inhomogeneous vacuum fluctuations about this homogenous background during the whole history of the universe. It turns out that the radion fluctuations produced during inflation have a steep red spectrum and cannot produce the present large-scale structure of our universe. Instead the curvature perturbations observed today must be generated by the curvaton or some similar mechanism from initially isocurvature excitations of massless degrees of freedom on one or other of the branes before the collision. The primordial gravitational waves produced are likely to be difficult to detect.
Our model does suffer from a significant fine tuning problem of the initial conditions. The radion field must start very close to an unstable extremum of its potential in order to obtain sufficient inflation. However, the radion field has a geometrical meaning so that the initial value is determined by the initial separation between two branes. Sufficient separation of the initial brane positions does not seem so unnatural, and the fine tuning valu of the radion may not be so serious. A quantum cosmological consideration GS ; Koyama would be needed to give a more insight on this initial condition problem. It might be that one can find an instanton solution to describe tunnelling to this extremum. Alternatively, in a semi-classical model such as stochastic inflation, it may be that quantum fluctuations drive the field to the top of the potential before classical evolution takes over and the field rolls down, leading to the branes to collide.
We stress that our model is fundamentally 5-dimensional in nature. It cannot be constructed starting from a 4-dimensional theory. It would of course be interesting to see if it is possible to embed our model into string theory model, but that goes beyond the scope of this paper.
Note added: While this work was being written up Koyama and Koyama KKKK submitted a related paper deriving an equivalent effective action for anti-D-branes in a type IIB string model.
###### Acknowledgements.
We wish to thank Roy Maartens for useful discussions. SK and JS would like to thank the Portsmouth ICG for its hospitality and financial support, under PPARC grant PPA/V/S/2001/00544. This work is supported by the Grant-in-Aid for the 21st Century COE ”Center for Diversity and Universality in Physics” from the Ministry of Education, Culture, Sports, Science and Technology (MEXT) of Japan. This work was also supported in part by Grant-in-Aid for Scientific Research Fund of the Ministry of Education, Science and Culture of Japan No. 155476 (SK), No.14540258 and No.17340075 (JS).
## Appendix A Derivation of effective action
### A.1 Static Solution
We shall start with the three-brane system and take the two-brane limit to derive the low energy effective action for the two-brane system. This allows us to use a simple moduli approximation method. Assuming there exists no matter on the third brane, this procedure can be justified.
In the low energy limit, the configuration should be almost static. As we have no matter in the bulk, the bulk metric should be anti-de Sitter spacetime. Hence, let us first consider the static solution of the form
$`ds_I^2=dy^2+b_I^2(y)\eta _{\mu \nu }dx^\mu dx^\nu `$
$`ds_{II}^2=dy^2+b_{II}^2(y)\eta _{\mu \nu }dx^\mu dx^\nu `$ (46)
where $`b_I=\mathrm{exp}(y/\mathrm{}_I)`$ and $`b_{II}=\mathrm{exp}(y/\mathrm{}_{II})`$ represents the warp factor in each region. The bulk equation of motion imply the relation between the curvature scale in the bulk and the flux as
$`{\displaystyle \frac{6}{\mathrm{}_I^2}}=\mathrm{\Lambda }{\displaystyle \frac{\kappa ^2}{2}}c_I^2,`$
$`{\displaystyle \frac{6}{\mathrm{}_{II}^2}}=\mathrm{\Lambda }{\displaystyle \frac{\kappa ^2}{2}}c_{II}^2.`$ (47)
We also have junction conditions for the metric which give the relation
$`\kappa ^2\sigma _1={\displaystyle \frac{6}{\mathrm{}_I}}`$ (48)
$`\kappa ^2\sigma _2={\displaystyle \frac{3}{\mathrm{}_{II}}}{\displaystyle \frac{3}{\mathrm{}_I}}`$ (49)
$`\kappa ^2\sigma _3={\displaystyle \frac{6}{\mathrm{}_{II}}}.`$ (50)
If we assume the inflating bulk brane $`\sigma _2>0`$, we have $`\mathrm{}_{II}<\mathrm{}_I`$. The above implies the relation
$`\sigma _1+2\sigma _2+\sigma _3=0.`$ (51)
The junction conditions for 5-form fields give
$`c_I={\displaystyle \frac{\mu _1}{2}},c_{II}=(\mu _2+{\displaystyle \frac{\mu _1}{2}})`$ (52)
$`\mu _1+2\mu _2+\mu _3=0.`$ (53)
The last relation is nothing but the charge conservation law. The static solution is realized under the relations Eqs. (51) and (53). In this paper, we set $`\mu _3=0`$. This implies $`c_{II}=0`$.
### A.2 Low energy effective action
Let us employ the moduli approximation method GPT which can be shown to be valid at low energy Soda ; Kanno . (An alternative derivation appears in Ref. Ludo .) In this method, we can use the factorized metric ansatz
$`ds^2`$ $`=`$ $`dy^2+b_I^2(y)g_{\mu \nu }^I(x)dx^\mu dx^\nu `$
$`ds^2`$ $`=`$ $`dy^2+b_{II}^2(y)g_{\mu \nu }^{II}(x)dx^\mu dx^\nu .`$ (54)
The brane positions are denoted by $`\varphi _1(x),\varphi _2(x)`$ and $`\varphi _3(x)`$. Now, we calculate the bulk action, the action for each brane tensions, and the Gibbons-Hawking terms, separately.
Let us start with the calculation of the bulk action. For the metric (54), using the Gauss equation, it is straightforward to relate the bulk Ricci scalar to the 4-dimentional Ricci scalar,
$`\stackrel{(5)}{R}={\displaystyle \frac{R(g^i)}{b_i^2}}{\displaystyle \frac{20}{\mathrm{}_i^2}},`$ (55)
we can write the contributions from each bulk as
$`S_{\mathrm{bulkI}}`$ $`=`$ $`{\displaystyle \frac{2}{2\kappa ^2}}{\displaystyle d^4x\sqrt{g^I}_{\varphi _1}^{\varphi _2}𝑑yb_I^4(y)\left[\frac{R(g^I)}{b_I^2}\frac{8}{\mathrm{}_I^2}\right]}`$ (56)
$`=`$ $`{\displaystyle \frac{1}{\kappa ^2}}{\displaystyle }d^4x\sqrt{g^I}[{\displaystyle \frac{\mathrm{}_I}{2}}\{b_I^2(\varphi _2)b_I^2(\varphi _1)\}R(g^I)`$
$`+{\displaystyle \frac{2}{\mathrm{}_I}}\{b_I^4(\varphi _2)b_I^4(\varphi _1)\}]`$
where we used the bulk equation of motion (47) in the first line. The factor 2 over $`\kappa ^2`$ comes from the $`Z_2`$ symmetry of this spacetime and we neglected the second order quantities. In the same way for region II, we have
$`S_{\mathrm{bulkI}\mathrm{I}}={\displaystyle \frac{2}{2\kappa ^2}}{\displaystyle d^4x\sqrt{g^{II}}_{\varphi _2}^{\varphi _3}𝑑yb_{II}^4(y)\left[\frac{R(g^{II})}{b_{II}^2}\frac{8}{\mathrm{}_{II}^2}\right]}`$
$`={\displaystyle \frac{1}{\kappa ^2}}{\displaystyle }d^4x\sqrt{g^{II}}[{\displaystyle \frac{\mathrm{}_{II}}{2}}\{b_{II}^2(\varphi _3)b_{II}^2(\varphi _2)\}R(g^{II})`$
$`+{\displaystyle \frac{2}{\mathrm{}_{II}}}\{b_{II}^4(\varphi _3)b_{II}^4(\varphi _2)\}].`$ (57)
In order to calculate the action for each brane tensions, we need to know the induced metric. The induced metric on the left boundary brane becomes
$`b^2(\varphi _1)g_{\mu \nu }^I+_\mu \varphi _1_\nu \varphi _1`$ (58)
and the similar formula holds for other branes. Then, the contributions from the brane tension terms become
$`S_1=\sigma _1{\displaystyle d^4x\sqrt{g^I}b_I^4(\varphi _1)\left[1+\frac{1}{2b_I^2(\varphi _1)}\left(\varphi _1\right)^2\right]},`$ (59)
$`S_2`$ $`=`$ $`2\sigma _2{\displaystyle d^4x\sqrt{g^I}b_I^4(\varphi _2)\left[1+\frac{1}{2b_I^2(\varphi _2)}\left(\varphi _2\right)^2\right]}`$ (60)
$`=`$ $`2\sigma _2{\displaystyle d^4x\sqrt{g^{II}}b_{II}^4(\varphi _2)\left[1+\frac{1}{2b_{II}^2(\varphi _2)}\left(\varphi _2\right)^2\right]}`$
and
$`S_3=\sigma _3{\displaystyle d^4x\sqrt{g^{II}}b_{II}^4(\varphi _3)\left[1+\frac{1}{2b_{II}^2(\varphi _3)}\left(\varphi _3\right)^2\right]},`$ (61)
where the factor 2 in front of $`\sigma _2`$ comes from the $`Z_2`$ symmetry.
Let us turn to the calculation of Gibbons-Hawking terms. The extrinsic curvature in leading order is given by
$`K_{\mu \nu }=n_y\left[{\displaystyle \frac{4}{\mathrm{}}}+{\displaystyle \frac{1}{b^2}}\mathrm{}\varphi +{\displaystyle \frac{1}{\mathrm{}b^2}}(\varphi )^2\right],`$ (62)
where $`n_y`$ is the normal vector to the brane defined by
$`n_y=\left(1+{\displaystyle \frac{1}{b^2}}(\varphi )^2\right)^{1/2}.`$ (63)
Hence, the Gibbons-Hawking terms are
$`S_{\mathrm{GH1}}={\displaystyle \frac{2}{\kappa ^2\mathrm{}_I}}{\displaystyle d^4x\sqrt{g^I}\left[4b_I^4(\varphi _1)+3b_I^2(\varphi _1)\left(\varphi _1\right)^2\right]},`$ (64)
$`S_{\mathrm{GH2}}={\displaystyle \frac{2}{\kappa ^2\mathrm{}_{II}}}{\displaystyle d^4x\sqrt{g^{II}}\left[4b_{II}^4(\varphi _2)+3b_{II}^2(\varphi _2)\left(\varphi _2\right)^2\right]}`$
$`{\displaystyle \frac{2}{\kappa ^2\mathrm{}_I}}{\displaystyle d^4x\sqrt{g^I}\left[4b_I^4(\varphi _2)+3b_I^2(\varphi _2)\left(\varphi _2\right)^2\right]},`$ (65)
and
$`S_{\mathrm{GH3}}={\displaystyle \frac{2}{\kappa ^2\mathrm{}_{II}}}{\displaystyle d^4x\sqrt{g^{II}}\left[4b_{II}^4(\varphi _3)+3b_{II}^2(\varphi _3)\left(\varphi _3\right)^2\right]},`$ (66)
where the factor 2 over $`\kappa ^2`$ in $`S_{\mathrm{GH2}}`$ again comes from the $`Z_2`$ symmetry
Substituting Eqs.(56), (57), (59), (60) (61), (64), (65) and (66) into the 5-dimensional action (1), the resultant 4-dimensional effective action can be summarized by the following. The curvature part becomes
$`S_\mathrm{R}={\displaystyle \frac{\mathrm{}_I}{2\kappa ^2}}{\displaystyle d^4x\sqrt{g^I}\left[b_I^2(\varphi _1)b_I^2(\varphi _2)\right]R(g^I)}`$
$`+{\displaystyle \frac{\mathrm{}_{II}}{2\kappa ^2}}{\displaystyle d^4x\sqrt{g^{II}}\left[b_{II}^2(\varphi _2)b_{II}^2(\varphi _3)\right]R(g^{II})}.`$ (67)
The kinetic part of radions are
$`S_\mathrm{K}={\displaystyle \frac{3}{\kappa ^2\mathrm{}_I}}{\displaystyle d^4x\sqrt{g^I}\left[b_I^2(\varphi _1)\left(\varphi _1\right)^2b_I^2(\varphi _2)\left(\varphi _2\right)^2\right]}`$
$`+{\displaystyle \frac{3}{\kappa ^2\mathrm{}_{II}}}{\displaystyle d^4x\sqrt{g^{II}}\left[b_{II}^2(\varphi _2)\left(\varphi _2\right)^2b_{II}^2(\varphi _3)\left(\varphi _3\right)^2\right]}.`$ (68)
The potential energy will be induced as
$`S_\mathrm{V}={\displaystyle }d^4x\sqrt{g^I}[({\displaystyle \frac{6}{\kappa ^2\mathrm{}_I}}\sigma _1)b_I^4(\varphi _1)`$
$`\{2\sigma _2{\displaystyle \frac{6}{\kappa ^2}}({\displaystyle \frac{1}{\mathrm{}_{II}}}{\displaystyle \frac{1}{\mathrm{}_I}})\}b_I^4(\varphi _2)]`$
$`{\displaystyle d^4x\sqrt{g^{II}}\left(\sigma _3+\frac{6}{\kappa ^2\mathrm{}_{II}}\right)b_{II}^4(\varphi _3)}.`$ (69)
Now we can write down the effective action for the bulk brane. The continuity condition of the metric is given by
$`h_{\mu \nu }=b_I^2(\varphi _2)g_{\mu \nu }^I=b_{II}^2(\varphi _2)g_{\mu \nu }^{II},`$ (70)
where $`h_{\mu \nu }`$ is the induced metric of the bulk brane. Defining the variables
$`\mathrm{\Psi }={\displaystyle \frac{b_I(\varphi _1)}{b_I(\varphi _2)}},\mathrm{\Phi }={\displaystyle \frac{b_{II}(\varphi _3)}{b_{II}(\varphi _2)}},`$ (71)
we obtain
$`S`$ $`=`$ $`{\displaystyle \frac{\mathrm{}_I}{2\kappa ^2}}{\displaystyle d^4x\sqrt{h}\left[\mathrm{\Psi }^21+\alpha \left(1\mathrm{\Phi }^2\right)\right]R(h)}`$ (72)
$`+{\displaystyle \frac{3\mathrm{}_I}{\kappa ^2}}{\displaystyle d^4x\sqrt{h}\left[\left(\mathrm{\Psi }\right)^2\alpha \left(\mathrm{\Phi }\right)^2\right]}`$
$`+{\displaystyle \frac{6}{\kappa ^2\mathrm{}_I}}{\displaystyle }d^4x\sqrt{h}[(1\beta _1)\mathrm{\Psi }^4`$
$`(1{\displaystyle \frac{1}{\alpha }}+\beta _2)({\displaystyle \frac{1}{\alpha }}+\beta _3)\mathrm{\Phi }^4].`$
Here we have defined the dimensionless parameters
$`\alpha ={\displaystyle \frac{\mathrm{}_{II}}{\mathrm{}_I}},`$ (73)
$`\beta _1={\displaystyle \frac{\kappa ^2\sigma _1\mathrm{}_I}{6}},\beta _2={\displaystyle \frac{\kappa ^2\sigma _2\mathrm{}_I}{3}},\beta _3={\displaystyle \frac{\kappa ^2\sigma _3\mathrm{}_I}{6}}.`$ (74)
As we have the relation $`\mathrm{}_{II}<\mathrm{}_I`$, the relation $`\alpha <1`$ also holds.
After taking the two-brane limit $`a(\varphi _3)0`$, namely, $`\mathrm{\Phi }0`$, we obtain the low energy effective action for the two-brane system.
$`S={\displaystyle \frac{\mathrm{}_I}{2\kappa ^2}}{\displaystyle d^4x\sqrt{h}\left(\mathrm{\Psi }^2+\alpha 1\right)R(h)}`$
$`+{\displaystyle \frac{3\mathrm{}_I}{\kappa ^2}}{\displaystyle d^4x\sqrt{h}\left(\mathrm{\Psi }\right)^2}`$ (75)
$`+{\displaystyle \frac{6}{\kappa ^2\mathrm{}_I}}{\displaystyle d^4x\sqrt{h}\left[\left(1\beta _1\right)\mathrm{\Psi }^4\left(1\frac{1}{\alpha }+\beta _2\right)\right]}.`$
The scalar variable $`\mathrm{\Psi }`$ which describes the distance between two positive tension branes is called as the radion.
|
warning/0506/quant-ph0506098.html
|
ar5iv
|
text
|
# Measure of phonon-number moments and motional quadratures through infinitesimal-time probing of trapped ions
## 1 Introduction
A single ion in a radio-frequency trap is an ideal system for the investigation of the basic and intriguing features of quantum mechanics . Recent advances in the manipulation of the internal and external degrees of freedom of the trapped particle allowed, for example, the realization of laser cooling to the ground state for extended periods of time , the creation of nonclassical motional states , entanglement between external and internal degrees of freedom of one ion or several ions , quantum gates for quantum computation , and single photon emission on demand .
An important aspect at this advanced level of quantum state engineering is the ability to obtain as much information as possible, and with high efficiency, about the motional state of the trapped atom. Moreover, since the Lamb-Dicke (LD) regime, where the motion of the particle is restricted to a region smaller than a wavelength, is sometimes difficult to achieve in a trap, it is desirable to have measurement schemes working also outside this regime .
So far, several techniques have been proposed to gain information about the motional quantum state of a trapped particle. The central idea in all these methods is the mapping of the ionic external dynamics to an internal degree of freedom, where the latter can be read out, for example, with electron shelving techniques . Some of these procedures are applicable only inside the LD regime , whereas others work also beyond this limit , allowing to derive in some cases the complete motional density operator. However, a typical requirement of conventional schemes is that the measurements have to be performed during relatively long interaction times, through several Rabi oscillations in the resonant case or slow phase shifts in the dispersive case, where decoherence mechanisms cannot be neglected.
In this paper, we present a method that allows to determine the expectation value of the phonon-number moments $`\widehat{n}^p`$ of a trapped ion for any positive integer $`p`$, as well as any generalized motional nonlinear quadrature $`\frac{1}{2}\widehat{f}(\widehat{n})\widehat{a}e^{i\varphi }+\widehat{a}^{}\widehat{f}(\widehat{n})e^{i\varphi }`$ <sup>1</sup><sup>1</sup>1Generalized nonlinear quadratures can be related to nonlinear coherent states, theoretically studied in . A method for generating nonlinear coherent states in trapped ions, but not for measuring them, can be found in .. Here, $`\widehat{a}`$ and $`\widehat{a}^{}`$ are the phonon annihilation and creation operators, respectively, $`\widehat{n}=\widehat{a}^{}\widehat{a}`$, $`\widehat{f}(\widehat{n})`$ is a function of $`\widehat{n}`$, and $`\varphi `$ is an arbitrary phase factor. In particular, we are able to measure the expectation value of the linear quadratures, i.e., when $`\widehat{f}(\widehat{n})=1`$, including the ionic position and momentum. Our technique is not restricted to the LD regime and, moreover, can be extended easily to all spatial dimensions, or to systems containing more than one trapped ion. Furthermore, in contrast to the methods presented so far, our proposal allows to obtain the required information in an extremely short motion-probe interaction time, which is useful in particular in presence of strong decoherence processes.
In Sec. 2, we show for a single trapped ion how to obtain information about the expectation values of the phonon-number moments (Sec. 2.1) and the generalized motional nonlinear and linear quadratures (Sec. 2.2) through infinitesimal-time motion-probe interactions. In Sec. 3, we extend the method to the $`N`$-ion case. In Sec. 4, we summarize our central results.
## 2 Single ion case
### 2.1 Phonon-number moments
We consider a single two-level ion trapped in a harmonic potential. For the sake of simplicity, we confine our treatment to one motional degree of freedom, though all results are easily generalizable to three dimensions. The system is described by the Hamiltonian
$$\widehat{H}_0=\mathrm{}\nu (\widehat{n}+1/2)+\mathrm{}\omega _0|ee|,$$
(1)
where $`\nu `$ is the harmonic oscillator frequency, $`|e`$ and $`|g`$ are the electronic upper and lower states of the two-level ion, respectively, and $`\omega _0`$ is the associated transition frequency.
The ion is excited by a travelling laser beam resonant with the carrier electronic transition, leaving unchanged the motional populations. In the usual rotating-wave approximation, and in the interaction picture, the Hamiltonian reads
$$\widehat{H}_{\mathrm{int}}=\frac{1}{2}\mathrm{}\mathrm{\Omega }_L(\widehat{\sigma }^++\widehat{\sigma }^{})\widehat{f}_0(\widehat{n};\eta ),$$
(2)
where $`\mathrm{\Omega }_L`$ is the Rabi frequency, $`\widehat{\sigma }^\pm `$ are the electronic two-level flip operators, $`\eta `$ is the LD parameter ($`\eta =k_xx_0`$, $`x_0`$ being the extension of the ground state of the motional mode and $`k_x`$ the projection of the laser wave vector on the trap axis), and $`\widehat{f}_0(\widehat{n};\eta )`$ is given by
$$\widehat{f}_0(\widehat{n};\eta )=e^{\eta ^2/2}\underset{l=0}{\overset{\mathrm{}}{}}\frac{(i\eta )^{2l}}{l!^2}\frac{\widehat{n}!}{(\widehat{n}l)!}.$$
(3)
At any time $`t`$, the probability of finding the ion in the excited state $`|e`$ is given by
$$P_e(t)=\mathrm{Tr}\left[\widehat{\rho }(t)|ee|\right]=|ee|,$$
(4)
where $`\widehat{\rho }(t)`$ is the system density operator, describing the internal and external degrees of freedom of the trapped particle.
Considering that, for any operator $`\widehat{A}`$,
$$\frac{d}{dt}\widehat{A}=\frac{1}{i\mathrm{}}[\widehat{A},\widehat{H}]+\frac{\widehat{A}}{t},$$
(5)
we get, for $`\widehat{A}=|ee|`$,
$$\frac{dP_e}{dt}=\frac{1}{i\mathrm{}}[|ee|,\widehat{H}].$$
(6)
In our case, we have $`[|ee|,\widehat{H}_0]=0`$, so that
$$\frac{dP_e}{dt}=\frac{1}{i\mathrm{}}[|ee|,\widehat{H}_{\mathrm{int}}].$$
(7)
Since, in the interaction picture,
$$[|ee|,\widehat{H}_{\mathrm{int}}]=\frac{\mathrm{}\mathrm{\Omega }_L}{2}(\widehat{\sigma }^+\widehat{\sigma }^{})\widehat{f}_0(\widehat{n},\eta ),$$
(8)
we obtain, from (7) and (8),
$$\frac{dP_e}{dt}=\frac{\mathrm{\Omega }_L}{2i}\mathrm{Tr}\left[\widehat{\rho }(t)(\widehat{\sigma }^+\widehat{\sigma }^{})\widehat{f}_0(\widehat{n},\eta )\right].$$
(9)
Next, we consider that the ion is prepared initially in the state
$$\widehat{\rho }(0)=|\pm _\varphi \pm _\varphi |\widehat{\rho }_f,$$
(10)
where
$$|\pm _\varphi =\frac{1}{\sqrt{2}}(|g\pm e^{i\varphi }|e),$$
(11)
and $`\widehat{\rho }_f`$ is the phonon state we aim to characterize. Then, a straightforward calculation yields
$$\mathrm{Tr}\left[\widehat{\rho }(0)(\widehat{\sigma }^+\widehat{\sigma }^{})\widehat{f}_0(\widehat{n},\eta )\right]=i\mathrm{sin}(\varphi )\mathrm{Tr}\left[\widehat{\rho }_f\widehat{f}_0(\widehat{n},\eta )\right],$$
(12)
and thus, using Eq. (9),
$$\frac{dP_e^{\pm _\varphi }}{d\tau }|_{\tau =0}=\mathrm{sin}(\varphi )\widehat{f}_0(\widehat{n},\eta ),$$
(13)
where $`\tau `$ is the dimensionless time $`\mathrm{\Omega }_Lt/2`$ and the token $`\pm _\varphi `$ stands for the two parameters (sign $`\pm `$ and phase $`\varphi `$) defining state $`|\pm _\varphi `$ of Eq. (11).
Eq. (13) shows that the mean value of the nonlinear operator $`\widehat{f}_0(\widehat{n};\eta )`$ is determined by the time derivative of the population of the excited state at the initial interaction time $`\tau =0`$. This allows to gain information about the ionic motional state, in particular, as will be shown below, about the phonon-number moments. Moreover, since the time derivative of the excitation probability $`P_e^{\pm _\varphi }`$ is evaluated at $`\tau =0`$, this information can be obtained in a very short motion-probe interaction time, even before decoherence mechanisms can affect the initial coherent evolution. Remark further that the “contrast” in the measurement of $`\widehat{f}_0(\widehat{n},\eta )`$, as seen in Eq. (13), can be tuned with the proper choice of the phase $`\varphi `$.
For small LD parameters, a series expansion of the nonlinear operator $`\widehat{f}_0(\widehat{n};\eta )`$ leads to the expression
$`\widehat{f}_0(\widehat{n};\eta )1\eta ^2\widehat{n}+{\displaystyle \frac{\eta ^4}{4}}\widehat{n}^2.`$ (14)
Thus, if we repeat the measurements with two known LD parameters, $`\eta _1`$ and $`\eta _2`$ (varying, e.g., the angle between the laser beam and the trap axis), we can derive $`\widehat{n}`$ and $`\widehat{n}^2`$ by solving the linear system
$$\{\begin{array}{cc}\hfill \widehat{f}_0(\widehat{n},\eta _1)& =1\eta _1^2\widehat{n}+\frac{\eta _1^4}{4}\widehat{n}^2,\hfill \\ \hfill \widehat{f}_0(\widehat{n},\eta _2)& =1\eta _2^2\widehat{n}+\frac{\eta _2^4}{4}\widehat{n}^2,\hfill \end{array}$$
(15)
from which, for example, the Fano-Mandel parameter
$$Q=\frac{\widehat{n}^2\widehat{n}^2}{\widehat{n}}$$
(16)
can be extracted. It is known that the Fano-Mandel parameter allows to determine the “classicality” of a given motional state, since quantum states with $`Q>1`$ are considered as classical states, and those with $`Q<1`$ as nonclassical states, due to their sub-Poissonian phonon-number distribution.
This low order moment determination scheme can be generalized to higher order vibronic moments $`\widehat{n}^p`$ ($`p>0`$). Indeed, for higher LD parameters, the series expansion of $`\widehat{f}_0(\widehat{n},\eta )`$ in Eq. (3) will extend beyond the $`\widehat{n}^2`$ term, up to a certain moment $`\widehat{n}^N`$ <sup>2</sup><sup>2</sup>2The series of Eq. (3) is convergent for all $`\eta `$. When the sum of this series is truncated at $`l=N`$, the coefficient of the $`\widehat{n}^N`$ term is given by $`(1)^Ne^{\eta ^2/2}\eta ^{2N}/N!^2`$. For any $`\eta `$ value, this coefficient can be made arbitrarily small by a proper choice of $`N`$ and the series may be truncated within good approximation up to this term.. We could repeat then the measurement technique, as in Eq. (15), for $`N`$ known LD parameters, yielding a linear system from which the $`N`$ first vibronic moments become accessible. Each $`\widehat{n}^p`$, $`0<pN`$, could also be determined by measuring only once the initial time derivative of $`P_e^{\pm _\varphi }`$. To show this, we require that the ion is submitted to $`N`$ simultaneous laser interactions, each of them resonant with the electronic transition and leaving the motional state unchanged. In the interaction picture, they give rise to the simultaneous action of Hamiltonians
$$\widehat{H}_{\mathrm{int}}^j=\frac{1}{2}\mathrm{}\mathrm{\Omega }_j\widehat{\sigma }^+\widehat{f}_0(\widehat{n};\eta _j)+\mathrm{H}.\mathrm{c}.,j=1,\mathrm{},N,$$
(17)
where $`\mathrm{\Omega }_j`$ and $`\eta _j`$ are, respectively, the electronic Rabi frequency and the LD parameter of laser $`j`$. The total Hamiltonian will be given in this case by
$$\widehat{H}_{\mathrm{int}}=\underset{j=1}{\overset{N}{}}\widehat{H}_{\mathrm{int}}^j=\frac{\mathrm{}\mathrm{\Omega }_L}{2}(\widehat{\sigma }^++\widehat{\sigma }^{})\widehat{F}_0(\widehat{n}),$$
(18)
where $`\mathrm{\Omega }_L=\mathrm{max}(\mathrm{\Omega }_j)`$ and
$$\widehat{F}_0(\widehat{n})=\underset{j=1}{\overset{N}{}}\frac{\mathrm{\Omega }_j}{\mathrm{\Omega }_L}\widehat{f}_0(\widehat{n};\eta _j).$$
(19)
If the system is again initially prepared in the state $`|\pm _\varphi \pm _\varphi |\widehat{\rho }_f`$, we get, similarly to Eq. (13),
$$\frac{dP_e^{\pm _\varphi }}{d\tau }|_{\tau =0}=\mathrm{sin}(\varphi )\widehat{F}_0(\widehat{n}).$$
(20)
It has been shown by de Matos Filho and Vogel that $`\widehat{F}_0(\widehat{n})`$ may be rewritten in the form of a Taylor series
$$\widehat{F}_0(\widehat{n})=\underset{p=0}{\overset{\mathrm{}}{}}c_p\widehat{n}^p,$$
(21)
where the Taylor coefficients $`c_p`$ are given by
$$c_p=\{\begin{array}{cc}\underset{j=1}{\overset{N}{}}e^{\eta _j^2/2}\frac{\mathrm{\Omega }_j}{\mathrm{\Omega }_L},& \mathrm{if}p=0,\hfill \\ (1)^p\underset{j=1}{\overset{N}{}}e^{\eta _j^2/2}\frac{\mathrm{\Omega }_j}{\mathrm{\Omega }_L}\left(\underset{m=p}{\overset{\mathrm{}}{}}a_p^m\frac{\eta _j^{2m}}{m!^2}\right),& \mathrm{if}p0,\hfill \end{array}$$
(22)
with
$$a_p^m=\{\begin{array}{cc}1,& \mathrm{if}p=m,\hfill \\ \underset{j_{i_1}<j_{i_2}<\mathrm{}<j_{i_{mp}}=1}{\overset{m1}{}}j_{i_1}j_{i_2}\mathrm{}j_{i_{mp}}& \mathrm{if}p<m.\hfill \end{array}$$
(23)
This means that for given values of the $`N`$ LD parameters $`\eta _j`$ (fixed by the geometry of the laser beams regarding to the trap axis up to the maximum value $`\eta _{\mathrm{max}}=kx_0`$, $`k`$ being the wave vector of the lasers), the $`c_p`$ coefficients are linear combination of all Rabi frequencies $`\mathrm{\Omega }_j`$. In this way, the use of $`N`$ lasers allows to fix at will $`N`$ coefficients of the Taylor series. In particular, if the $`N`$ first coefficients are set to 0, we obtain
$$\widehat{F}_0(\widehat{n})=𝒪(\frac{\eta _{\mathrm{max}}^{2N}}{N!^2}\widehat{n}^N).$$
(24)
Note that it is always possible to choose the $`N`$ value such that $`𝒪(\frac{\eta _{\mathrm{max}}^{2N}}{N!^2}\widehat{n}^N)`$ is negligible, even outside the LD regime where $`\eta _{\mathrm{max}}`$ can be greater than 1. The mean value of $`𝒪(\frac{\eta _{\mathrm{max}}^{2N}}{N!^2}\widehat{n}^N)`$ can be verified experimentally by measuring the time derivative of the population of the excited state at interaction time $`\tau =0`$. According to Eq. (20), the outcome of this measurement yields $`𝒪(\frac{\eta _{\mathrm{max}}^{2N}}{N!^2}\widehat{n}^N)`$ so that we can check if it is indeed negligible for the phonon distribution we aim to characterize (in this case it is recommendable to set $`\varphi =\pm \pi /2`$).
The Rabi frequencies of the $`N`$ lasers could also be chosen in such a manner that only a single coefficient $`c_p`$ ($`p<N`$) is equal to 1, the others remaining equal to 0. In this case, according to Eq. (21) and considering that all terms $`\widehat{n}^r`$, $`rN`$, are negligible,
$$\widehat{F}_0(\widehat{n})=\widehat{n}^p,$$
(25)
and the measurement of $`\frac{dP_e^{\pm _\varphi }}{d\tau }|_{\tau =0}`$ yields directly the vibronic moment $`\widehat{n}^p`$. This step can be reproduced for any $`p<N`$.
It is noteworthy to mention that the knowledge of the phonon-number moments $`\widehat{n}^p`$ for all positive integers $`p`$ allows to derive the complete phonon distribution $`p(n)n|\widehat{\rho }_f|n`$, as known from classical statistics . Also, the generalization to the $`3D`$ case provides measurement schemes for $`\widehat{n}_x`$, $`\widehat{n}_y`$, $`\widehat{n}_z`$, $`\widehat{n}_x^2`$, $`\widehat{n}_y^2`$, $`\widehat{n}_z^2`$, $`\widehat{n}_x\widehat{n}_y`$, $`\widehat{n}_x\widehat{n}_z`$, $`\widehat{n}_y\widehat{n}_z`$, $`\widehat{n}_x\widehat{n}_y\widehat{n}_z`$, and so on. Recently, the relevance of measuring different photon-number moments, through quantum field homodyning, for determining the “classicality” of arbitrary quantum states has been considered .
### 2.2 Generalized nonlinear quadratures
Next, we show that by using a red or blue sideband excitation, expectation values of generalized nonlinear quadrature moments, $`\frac{1}{2}\widehat{f}(\widehat{n})\widehat{a}e^{i\varphi }+\widehat{a}^{}\widehat{f}(\widehat{n})e^{i\varphi }`$, can be determined using similar techniques. In particular, we have access to the expectation values of the generalized linear quadrature moments, among them position $`\widehat{x}`$ and momentum $`\widehat{p}`$ <sup>3</sup><sup>3</sup>3In this limit our results for trapped ions are analogous to the ones derived for measuring field quadratures in cavity QED setups: see .. As will be shown below, these expectation values can be obtained inside and outside of the LD regime.
Let us consider that the ion is excited with a red-sideband detuned laser (the blue-sideband case can be treated equally, leading to similar results). In the usual rotating-wave approximation, and in the interaction picture, the Hamiltonian reads
$$\widehat{H}_{\mathrm{int}}=\frac{i}{2}\mathrm{}\mathrm{\Omega }_L\eta \widehat{\sigma }^+\widehat{f}_1(\widehat{n};\eta )\widehat{a}+\mathrm{H}.\mathrm{c}.,$$
(26)
where
$$\widehat{f}_1(\widehat{n};\eta )=e^{\eta ^2/2}\underset{l=0}{\overset{\mathrm{}}{}}\frac{(i\eta )^{2l}}{l!(l+1)!}\frac{\widehat{n}!}{(\widehat{n}l)!}.$$
(27)
If the ion system is initially prepared in the state $`|\pm _\varphi \pm _\varphi |\widehat{\rho }_f`$, we obtain straightforwardly from Eq. (6)
$$\frac{dP_e^{\pm _\varphi }}{d\tau }|_{\tau =0}=\pm \frac{1}{2}\widehat{f}_1(\widehat{n};\eta )\widehat{a}e^{i\varphi }+\widehat{a}^{}\widehat{f}_1(\widehat{n};\eta )e^{i\varphi },$$
(28)
where $`\tau `$ is the dimensionless time $`\eta \mathrm{\Omega }_Lt/2`$. In the LD regime, $`\widehat{f}_1(\widehat{n};\eta )=1`$ and Eq. (28) reduces to the generalized quadrature $`\widehat{X}_\varphi (\widehat{a}e^{i\varphi }+\widehat{a}^{}e^{i\varphi })/2`$,
$$\frac{dP_e^{\pm _\varphi }}{d\tau }|_{\tau =0}=\pm \widehat{X}_\varphi =\pm \frac{1}{2}\left(\mathrm{cos}(\varphi )\frac{\widehat{x}}{x_0}+\mathrm{sin}(\varphi )\frac{\widehat{p}}{p_0}\right),$$
(29)
where $`\widehat{x}`$ and $`\widehat{p}`$ are the trapped ion position and momentum, respectively, and $`x_0`$ and $`p_0`$ are the spreads of these quantities in the ground state of the trap potential. Eq. (29) shows that any quadrature moment $`\widehat{X}_\varphi `$ can be experimentally determined from the measurement of the initial time derivative of the population of the excited state, whereas, in particular, $`\widehat{x}`$ and $`\widehat{p}`$ can be obtained by choosing $`\varphi =0,\pi `$ and $`\varphi =\pm \pi /2`$, respectively.
Outside the LD regime, a similar information is gained using $`N`$ simultaneous red-sideband detuned lasers. In the interaction picture, they give rise to the Hamiltonians
$$\widehat{H}_{\mathrm{int}}^j=\frac{i}{2}\mathrm{}\mathrm{\Omega }_j\eta _j\widehat{\sigma }^+\widehat{f}_1(\widehat{n};\eta _j)\widehat{a}+\mathrm{H}.\mathrm{c}.,j=1,\mathrm{},N,$$
(30)
where $`\mathrm{\Omega }_j`$ and $`\eta _j`$ are, respectively, the electronic Rabi frequency and the LD parameter of laser $`j`$. The total Hamiltonian is given in this case by
$$\widehat{H}_{\mathrm{int}}=\underset{j=1}{\overset{N}{}}\widehat{H}_{\mathrm{int}}^j=\frac{i}{2}\mathrm{}\mathrm{\Omega }_L\widehat{\sigma }^+\widehat{F}_1(\widehat{n})\widehat{a}+\mathrm{H}.\mathrm{c}.,$$
(31)
where $`\mathrm{\Omega }_L=\mathrm{max}(\mathrm{\Omega }_j)`$ and
$$\widehat{F}_1(\widehat{n})=\underset{j=1}{\overset{N}{}}\frac{\mathrm{\Omega }_j\eta _j}{\mathrm{\Omega }_L}\widehat{f}_1(\widehat{n};\eta _j).$$
(32)
If the ion is initially in the state $`|\pm _\varphi \pm _\varphi |\widehat{\rho }_f`$, we obtain, similarly to Eq. (28), with $`\tau =\mathrm{\Omega }_Lt/2`$,
$$\frac{dP_e^{\pm _\varphi }}{d\tau }|_{\tau =0}=\pm \frac{1}{2}\widehat{F}_1(\widehat{n})\widehat{a}e^{i\varphi }+\widehat{a}^{}\widehat{F}_1(\widehat{n})e^{i\varphi }.$$
(33)
According to de Matos Filho and Vogel , $`\widehat{F}_1(\widehat{n})`$ may be again rewritten in the form of a Taylor series
$$\widehat{F}_1(\widehat{n})=\underset{p=0}{\overset{\mathrm{}}{}}c_p\widehat{n}^p,$$
(34)
where the $`c_p`$ coefficients are linear combination of the $`N`$ laser Rabi frequencies $`\mathrm{\Omega }_j`$. Similarly to the case of a resonant laser described in the preceding section, we can choose the $`N`$ value in such a way that $`𝒪(c_N\widehat{n}^N)`$ is negligible and fix next the $`N`$ Rabi frequencies to set all coefficients $`c_p`$ ($`p<N`$) to 0, except $`c_0`$ to $`1`$. In this case, $`\widehat{F}_1(\widehat{n})1`$, Eq. (33) reduces to Eq. (29) and the quadrature moments (and more specifically $`\widehat{x}`$ and $`\widehat{p}`$) are determined as done in the case of the LD regime. Clearly, by following a similar procedure, we could also engineer $`\widehat{F}_1(\widehat{n})`$ to describe an arbitrary polynomial.
## 3 $`N`$-ion case
We now briefly discuss how our proposal can be generalized to a chain of $`N`$ identical two-level ions in a linear Paul trap. This system is described by the Hamiltonian
$$\widehat{H}_0=\underset{j=1}{\overset{N}{}}\mathrm{}\nu _j(\widehat{n}_j+1/2)+\mathrm{}\omega _0\underset{k=1}{\overset{N}{}}|e_ke|,$$
(35)
where $`\nu _j`$ are the frequencies associated with the collective motional modes, $`\widehat{n}_j=\widehat{a}_j^{}\widehat{a}_j`$ are the respective phonon-number operators, $`|e_k`$ and $`|g_k`$ are the electronic states of the two-level ion $`k`$, and $`\omega _0`$ is the electronic transition frequency of each ion.
The ions are illuminated by a laser beam resonant with their electronic transition while leaving unchanged the phonon population, realizing a nonlinear carrier excitation. In the usual rotating-wave approximation and in the interaction picture, the Hamiltonian reads
$$\widehat{H}_{\mathrm{int}}=\frac{1}{2}\mathrm{}\mathrm{\Omega }_L\underset{k=1}{\overset{N}{}}(\widehat{\sigma }_k^++\widehat{\sigma }_k^{})\underset{j=1}{\overset{N}{}}\widehat{f}_0(\widehat{n}_j;\eta _j),$$
(36)
where $`\mathrm{\Omega }_L`$ is the electronic Rabi frequency, $`\widehat{\sigma }_k^\pm `$ are the electronic two-level flip operators of atom $`k`$, $`\eta _j`$ is the LD parameter related to the collective mode $`j`$ ($`\eta _j=k_xx_0_j`$, $`x_0_j`$ being the extension of the ground state of mode $`j`$ and $`k_x`$ the projection of the laser wave vector on the trap axis), and
$$\widehat{f}_0(\widehat{n}_j;\eta _j)=e^{\eta _j^2/2}\underset{l=0}{\overset{\mathrm{}}{}}\frac{(i\eta _j)^{2l}}{l!^2}\frac{\widehat{n}_j!}{(\widehat{n}_jl)!}.$$
(37)
In the following, we denote the product $`_{j=1}^N\widehat{f}_0(\widehat{n}_j;\eta _j)`$ by $`\widehat{}_0`$. The generalized relation
$$\frac{d(P_e)_k}{dt}=\frac{1}{i\mathrm{}}[|e_ke|,\widehat{H}],$$
(38)
can be easily obtained, as in Eq. (6), where $`(P_e)_k`$ is the probability of finding ion $`k`$ in its excited state. As $`|e_ke|`$ commutes with $`\widehat{H}_0`$ and any operator associated with other ions $`k^{}`$, we get immediately
$$\frac{d(P_e)_k}{dt}=\frac{1}{i\mathrm{}}[|e_ke|,\frac{1}{2}\mathrm{}\mathrm{\Omega }_L(\widehat{\sigma }_k^++\widehat{\sigma }_k^{})\widehat{}_0],$$
(39)
and thus
$$\frac{d(P_e)_k}{dt}=\frac{\mathrm{\Omega }_L}{2i}\mathrm{Tr}\left[\widehat{\rho }(t)(\widehat{\sigma }_k^+\widehat{\sigma }_k^{})\widehat{}_0\right],$$
(40)
where $`\widehat{\rho }(t)`$ is the $`N`$-ion system density operator.
Let us consider the following initial state
$$\widehat{\rho }(0)=\widehat{\rho }_k\widehat{\rho }_A\widehat{\rho }_f,$$
(41)
where
$$\widehat{\rho }_k=|\pm _\varphi _k\pm _\varphi |$$
(42)
is the electronic density operator of the $`k`$-th ion, with
$$|\pm _\varphi _k=\frac{1}{\sqrt{2}}(|g_k\pm e^{i\varphi }|e_k),$$
(43)
$`\widehat{\rho }_A`$ is an arbitrary electronic density operator of the remaining ions, and $`\widehat{\rho }_f`$ is the collective motional density operator associated with the $`N`$ eigenmodes. Following similar steps as in the previous sections, we obtain
$`\mathrm{Tr}\left[\widehat{\rho }(0)(\widehat{\sigma }_k^+\widehat{\sigma }_k^{})\widehat{}_0\right]=i\mathrm{sin}(\varphi )\mathrm{Tr}\left[\widehat{\rho }_f\widehat{}_0\right],`$ (44)
similar to Eq. (12), and thus, using Eq. (40),
$$\frac{d(P_e)_k}{d\tau }|_{\tau =0}=\mathrm{sin}(\varphi )\widehat{}_0,$$
(45)
where $`\tau `$ is the dimensionless time $`\tau =\mathrm{\Omega }_Lt/2`$. Eq. (45) tells us that by measuring the time derivative of the population of the excited state of one ion at $`\tau =0`$, we are able to gain information about the collective nonlinear operator $`\widehat{}_0`$.
## 4 Summary
In conclusion, we have proposed a method that allows to determine the phonon-number moments and the motional nonlinear and linear quadratures of a trapped ion, in particular the ionic position and momentum. Our method makes use of the nonlinear behavior of the ion-laser interaction in harmonic traps. The measurement of the phonon-number moments $`\widehat{n}^p`$ requires resonant carrier interaction, with no phonon gain or loss, while the measurement of the nonlinear quadrature moments $`\frac{1}{2}\widehat{f}(\widehat{n})\widehat{a}e^{i\varphi }+\widehat{a}^{}\widehat{f}(\widehat{n})e^{i\varphi }`$ demands the use of red or blue sideband excitations. In contrast to methods presented so far, our proposal is designed for measurements realized in very short probe laser interaction times, thus preventing the noisy action of decoherence processes. In addition, we have shown that our scheme works inside and outside the LD regime, and that it can be generalized to the case of $`N`$ ions. In the latter case, information about a collective property is gained by the measurement of the excited state population of a single ion.
TB acknowledges support from the Belgian Institut Interuniversitaire des Sciences Nucléaires (IISN), and thanks JVZ and ES for the hospitality at the University of Erlangen, Erlangen, and Max-Planck-Institut für Quantenoptik, Garching, Germany. ES acknowledges support from EU through RESQ project.
## References
|
warning/0506/astro-ph0506095.html
|
ar5iv
|
text
|
# Discovery and Monitoring of the likely IR Counterpart of SGR 1806–20 during the 2004 𝛾–ray burst-active stateThe results reported in this Letter are based on observations carried out at ESO, VLT, Chile (programs 072.D–0297, 073.D–0381 and 274.D–5018)
## 1 Introduction
Soft Gamma–ray Repeaters (SGRs) were discovered in the seventies through the detection of short ($`<`$1s), recurrent, and intense bursts of high energy emission peaked in the soft $`\gamma `$ rays. Only four confirmed SGRs are known, three in the Galaxy and one in the Large Magellanic Cloud (for a review see e.g. Woods & Thompson 2004). The detection of a $`8`$ s periodicity in the decaying tail of a very intense ($``$10<sup>44</sup> ergs) and long (several minutes) event, known as giant flare, from SGR 0526–66 on 1979 March 5th (Mazets et al. 1979) suggested the association of SGRs with neutron stars. A small sample of peculiar X–ray pulsars, namely the Anomalous X–ray Pulsars (AXPs) has been proposed to be closely related to SGRs based on similar properties, namely their period P (in the 5–12 s range), their period derivative Ṗ (10<sup>-10</sup>–10<sup>-13</sup> s s<sup>-1</sup> range), and X–ray bursts (Kouveliotou et al. 1998; Kaspi et al. 2003; Gavriil et al. 2002).
Both SGRs and AXPs have been proposed to be powered by the decay of strong magnetic fields that characterise these neutron stars (B $``$ 10<sup>14</sup>–10<sup>15</sup> G; Duncan & Thompson 1992; Thompson & Duncan 1995). The “magnetar” model is founded on two observational facts: firstly, the rotational energy loss inferred from the SGR and AXP spin–down is insufficient to power their persistent X–ray luminosity of $``$10<sup>34</sup>–10$`{}_{}{}^{36}\text{ergs s}\text{-1}`$; secondly, there is no evidence for a companion stars which could provide the mass to power the X–ray emission through accretion.
Bursting activity from SGR 1806–20 resumed at the end of 2003 displaying an increase in both the $`\gamma `$–ray burst rate and the hard X–ray persistent emission (Mereghetti et al. 2005a) throughout 2004, and culminating with the giant flare of 27th December 2004 (Borkowski et al. 2004), during which $``$10<sup>47</sup> ergs were released (for a distance of about 10kpc; Cameron et al. 2005; McClure-Griffiths & Gaensler 2005). Few days after this event, SGR 1806–20 was observed and detected in the radio band for the first time, providing very accurate positions (VLA; Cameron et al. 2005; Gaensler et al. 2005).
In this work we report on the results of an extended Target of Opportunity (ToO) observational campaign on SGR 1806–20 carried out during 2004 with the ESO VLT. In particular, we report on the likely discovery of the IR counterpart to SGR 1806–20 based on positional coincidence with the radio and Chandra positions and flux variability. (Preliminary results were reported in Israel et al. 2004, 2005a, 2005b, before and independently from Kosugi et al. 2005.) We briefly compare the IR emission properties of SGR 1806–20 with those of related objects.
## 2 NAOS-CONICA observations at VLT
The observations presented here were performed as part of an ESO Target of Opportunity program extending over October 2003 – September 2004, and a Director Discretionary Time observation on October 2004. Observations were triggered following the detection of intense $`\gamma `$–ray bursts or during epochs of increased burst rate (Götz et al. 2004; Golenetskii et al. 2004a,2004b). The data were acquired at Yepun VLT with the Nasmyth Adaptive Optics System and the High Resolution Near IR Camera providing a pixel size of 0$`\stackrel{}{.}`$027 (NAOS-CONICA; see Table 1 for details). For all observations we used an exposure time of 40 s and a number of frames per image of 3 with a random offset of 7″ among images in order to perform background subtraction of the variable IR sky. VLT NACO science images were reduced based on the standard tools provided by the ESO - Eclipse package (Devillard 1997). As a result of the presence of the Ks=8.9 magnitude object LBV 1806–20 (see Eikenberry et al. 2004) close to the edge of the NACO field of view (FOV), artificial ring–like ghost structures were clearly detected in the image at coordinates R.A.=18<sup>h</sup> 08<sup>m</sup> 40$`\stackrel{s}{.}`$31; Dec.=$``$20 24′ 41$`\stackrel{}{.}`$21 (equinox 2000), 13″ away from the LBV 1806–20 position. In order to reduce as much as possible the effects of contamination due to nearby objects, relative aperture and Point Spread Function (PSF) photometry was obtained within narrow annuli (about 1–1.5 FWHM depending on the seeing conditions), while the background was evaluated close to the object under analysis. Absolute photometry was derived by analysis of the best seeing frames. Finally, we cross–checked our absolute magnitudes by means of archival ISAAC data of the same region and about $`100`$ isolated stars taken from the 2MASS catalog and within the instrument FOV: the results were in agreement to within 0.05 Ks magnitudes.
In order to register the Chandra and VLA coordinates of SGR 1806–20 on our IR images, we obtained the image astrometry by using the positions of about 10 stars selected from the 2MASS catalogs and within the $``$30$`\mathrm{}\times `$ 30$`\mathrm{}`$ NACO FOV of final images. The residual in the fit was of 0$`\stackrel{}{.}`$06 in each coordinate, converting to $`0\stackrel{}{.}1`$ once the 2MASS absolute accuracy was included<sup>1</sup><sup>1</sup>1http://www.ipac.caltech.edu/2mass/releases/allsky/doc. Fig. 1 shows the $``$ 1$`\stackrel{}{.}`$5$`\times `$1$`\stackrel{}{.}`$5 Ks band region around the Chandra and VLA positions (1$`\sigma `$ confidence level radius of 0$`\stackrel{}{.}`$3, 0$`\stackrel{}{.}`$04, respectively; Kaplan et al. 2002; Gaensler et al. 2005). However, given that the Gaensler et al. (2005) radio position refers to about 20 days after the giant flare of SGR 1806–20, and that the source from which is originating the radio emission is moving at about 4 mas/day (Taylor et al. 2005), we also plot the VLA position obtained after 7 days by Cameron et al. (2005; 1$`\sigma `$ radius if 0$`\stackrel{}{.}`$1), corrected for about 30 mas in right ascension (following Taylor et al. 2005); this corresponds to a final 1$`\sigma `$ confidence level radius of 0$`\stackrel{}{.}`$14. Source $`A`$, a relatively faint (Ks$``$20) object, at the sky position R.A.= 18<sup>h</sup> 08<sup>m</sup> 39$`\stackrel{s}{.}`$337, Dec.= $``$20 24′ 39$`\stackrel{}{.}`$85 (equinox 2000, 90% uncertainty of 0$`\stackrel{}{.}`$06 ), is found to be consistent with the Chandra and VLA positional uncertainty circles superimposed on our IR astrometry–corrected frame. Objects $`B`$ and $`C`$ ($``$0$`\stackrel{}{.}`$23 and 0$`\stackrel{}{.}`$27 away from $`A`$, respectively; by looking at the contour lines, we note that object $`B`$ might be the blend of two unresolved objects) are only marginally consistent with the X–ray and radio positions, even though statistically plausible. Object $`A`$ was not detected in the J and H images; 3$`\sigma `$ upper limits of magnitude 21.2 and 19.5 were derived, respectively. We note that the SGR 1806–20 IR counterpart $`A`$ (Ks magnitudes are listed in Table 1) plus objects $`B`$ and $`C`$ (Ks=19.07$`\pm `$0.04 and 18.77$`\pm `$0.04, respectively) are all within a radius of $``$0$`\stackrel{}{.}`$25 from the VLA positions, and consistent with being unresolved components of candidate $`B`$ in the IR images of Eikenberry et al. (2001; K=18.6$`\pm `$1.0).
Light curves of the $`A`$, $`B`$ and $`C`$ objects marked in Fig. 1 are shown in Fig. 1 (right plot). Candidate $`A`$ is the only one showing a clear brightening (a factor of $``$2) in the IR flux between June and October 2004. Objects $`B`$ and $`C`$ show a fairly constant flux<sup>2</sup><sup>2</sup>2For the faintest component of blended object B we can reasonably exclude any IR variability, similar to that shown by $`A`$, for any Ks magnitude brighter than about 22.5 .. The upper panel of Fig. 1 (right plot) shows the closest reference star (1$`\stackrel{}{.}`$6 away form the target) used for relative photometry across Ks images: the object is constant to within the photometric uncertainties. We thus conclude that object $`A`$ is variable. We checked for a similar variability also in the X–ray flux of SGR 1806–20. Both the XMM–Newton (Mereghetti et al. 2005b) and INTEGRAL (Mereghetti et al. 2005a) persistent fluxes of SGR 1806–20 showed an increase across the two semesters of 2004 by a factor of 1.94$`{}_{}{}^{+0.01}{}_{0.02}{}^{}`$ and 1.7$`{}_{}{}^{+0.4}{}_{0.3}{}^{}`$ in the 2–10 keV and 20–100 keV bands, respectively<sup>3</sup><sup>3</sup>3for the 2–10 keV 2004 first semester flux we assumed that of October 2003, based on the unvaried INTEGRAL flux between October 2003 and February–April 2004.. During the same time interval the NACO Ks flux increased by a factor of 2.4$`{}_{}{}^{+0.9}{}_{0.5}{}^{}`$, consistent with high energy flux variations. This further supports the identification of object A as the correct IR counterpart of SGR 1806–20.
Recently, independent from our work, the object $`A`$ has been proposed as the IR counterpart to SGR 1806–20 (Kosugi et al. 2005; their object $`B3`$). A comparison of their photometry with our we shows that nearly all the Ks magnitudes have an offset of about 0.2, with the important exception of objects $`A`$ and $`C`$ which are 1.6 and 0.6 Ks magnitudes brighter than the corresponding objects $`B3`$ and $`B1`$ in Kosugi et al. (2005), respectively. Even though we do not have a clear explanation for the observed differences, we note that in our images we did not see any evidence for (i) a brightening of objects $`B`$ and $`C`$, and (ii) an increase of the local background around object $`A`$ (based on our best datasets with FWHM $``$0$`\stackrel{}{.}`$1), in contrast to Kosugi et al. (2005). An unusually high background level (regardless of its origin) may of course result in a flux underestimation of a source that lies in the same area.
## 3 Discussion
The deep and high spatial resolution NACO images allowed us to identify the likely IR counterpart of SGR 1806–20, and monitor its IR flux for seven months in 2004, during which an increase by a factor of $``$2 was detected, correlated with the flux in the high energy bands. In fact, the IR flux of SGR 1806–20 was fairly constant until mid-June 2004, while it grew rapidly between June (Ks=20.01$`\pm `$0.14) and October 2004 (Ks=19.32$`\pm `$0.16; 1$`\sigma `$ uncertainties; these values override the preliminary ones reported in Israel et al. 2005b).
IR variability has been detected in nearly all AXPs with known IR counterpart. In particular, for 1E 1048.1$``$5937, XTE J1810$``$197 and 1E 2259$`+`$586, IR variability has been found, or suspected, to be correlated with the persistent X–ray emission (Israel et al. 2002; Rea et al. 2004; Tam et al. 2004). Based on the NACO results we can conclude that the IR/X–ray correlation observed in AXPs also holds for SGR 1806–20. The total fluence of the IR enhancement between June and October 2004 is about $`10^{41}`$ ergs (we assumed $`A_V`$=29$`\pm `$2; see Eikenberry et al. 2004), a factor of about 100 smaller than that in the 2–10 keV band.
Based on the above reported findings we note that the SGR 1806–20 emission varies in a similar fashion (in terms of timescale and amplitude of variation) over more than five orders of magnitude in photon energy. The similar flux variation in the IR and X–ray bands suggests that the emission in the two bands has a similar, if not the same, origin. Moreover, it has become evident that X–ray flux enhancement of the persistent emission of SGRs is correlated with their burst rate, making it difficult to compare the fluxes among different SGRs without knowing their burst history (see Woods & Thompson 2004). Tam et al. (2004) argued that IR thermal surface emission (within the magnetar model) is ruled out during the correlated X–ray/IR flux decay phases of 1E 2259$`+`$586 (implausibly high implied brightness temperature), suggesting the magnetospheric origin for the IR enhancement. Alternatively, the IR flux can be due to reradiation by material in the vicinity of the the pulsar. This model naturally predicts a correlation between the the IR and the X–ray flux (Perna, Hernquist & Narayan 2000; Rea et al. 2004).
This is the first time that the broad band energy properties of an SGR can be compared, over a similar energy band, with those of other classes of isolated neutron stars, such as AXPs and radio pulsars. In Fig. 2 we show the “nearly simultaneous” broad band energy spectrum of SGR 1806–20 from the IR to $`\gamma `$ rays (high energy data are taken form Mereghetti et al. 2005a; see caption for details). The high energy part of the spectrum is clearly consistent with being non–thermal emission (a power–law model is generally used) from the source. We also plot the spectrum from the AXP 1E 1841$``$045, for which 20–200 keV band data are available (Kuiper et al. 2004); a similar non–thermal component is displayed by the source. Non–thermal components are also seen in radio pulsars and modelled with power–law components (see Kaspi, Roberts & Harding 2004 for a recent review). In some cases there is a smooth connection between optical, X–rays and $`\gamma `$–ray emission (Crab), while in other cases the extrapolation is plausible (Vela; see Fig. 2). It is worth noting the similar flux ratios in the IR and hard X–ray bands for the three objects, and the significant difference of the characteristic temperature of thermal soft X–ray components between radio pulsars ($``$0.1 keV) and SGRs/AXPs (0.4–0.8 keV for a BB fit and 0.2–0.5 keV for a magnetic atmosphere fit, Perna et al. 2001), suggesting a significantly larger energy injection on the neutron star surface in “magnetar” candidates than in radio pulsars.
Future detailed multi–wavelength observations campaigns of AXPs and SGRs will likely help clarifying the link between IR and high energy bands. Furthermore, the detection of the quiescent IR flux level of SGR 1806–20 will allow to compare the net energy released by the source in the IR and X–ray/$`\gamma `$–ray bands during its bursting active phase.
###### Acknowledgements.
We thank the ESO Director’s Discretionary Time Committee for accepting the observation of SGR 1806–20 few hours after the 5th October 2004 intense X–ray burst. We are also indebted with VLT personnel for their continuous help in optimising and performing the NACO observations. We thanks D. Dobrzycka and W. Hummel for their help in clarifying the artificial nature of the rings in the NACO images. This work was partially supported through Agenzia Spaziale Italiana (ASI), Ministero dell’Istruzione, Università e Ricerca Scientifica e Tecnologica (MIUR – COFIN), and Istituto Nazionale di Astrofisica (INAF) grants. N.R. is supported by a Marie Curie Trainig Grant (HPMT-CT-2001-00245).
|
warning/0506/astro-ph0506260.html
|
ar5iv
|
text
|
# Chandra Observation of the Cluster of Galaxies MS 0839.9+2938 at 𝑧=0.194: the Central Excess Iron and SN Ia Enrichment
## 1 Introduction
The metallicity of the intra-cluster medium (ICM) in clusters of galaxies is a sensitive tracer of both the history of the hierarchical clustering and the chemical evolutions of member galaxies and clusters themselves. Since the ICM has been enriched by the supernova explosions, the study of the metal abundance gradient in clusters, especially in those distant ones, may provide us with crucial constraints on the SN rate, stellar mass loss rate, and thus star formation rate. Along with the metal enrichment, a tremendous amount of the supernova energy has been inputted into the ICM, primarily in the form of gas dynamic energy that may be used to heat the ICM. Thus detailed measurements of the metal distribution in clusters as a function of redshift may also provide us with valuable hints to investigate how the ICM has been heated, a puzzling question that has plagued astronomers for a long time since the absence of massive cooling gas has been confirmed in many clusters (see, e.g., Makishima et al. 2001 for a review).
In the past five years, the inward iron abundance increase, first discovered with Ginga, ROSAT and ASCA in some nearby clusters that usually host a giant cD galaxy (e.g., Koyama et al. 1991; Fukazawa et al. 1994; Ezawa et al. 1997; Xu et al. 1997; Ikebe et al. 1999; Finoguenov et al. 2000), has been revealed in more sources (e.g., a sample of 12 nearby rich clusters at $`z_{}^<0.1`$ that have cool cores; De Grandi et al. 2004) and in much more details by the high spatial resolution observations of Chandra and XMM, which allows us to measure the abundances of iron and other metals with sufficient accuracy to deduce the SN Ia and SN II enrichments and their spatial variations. In clusters that exhibit a central iron excess, it has been found that the average iron abundance is typically $`0.40.6`$ solar within a radius of 30–80 kpc, which descends to about $`0.20.3`$ solar in the outer regions. In several most neighboring sources, such as M87/the Virgo cluster (Matsushita et al. 2003), NGC 1399/the Fornax cluster (Buote 2002), NGC 5044/the NGC 5044 group (Buote et al. 2003), NGC 1129/AWM 7 (Furusho et al. 2003), and NGC 4696/the Centaurus cluster (Sanders & Fabian 2002), the average iron abundance is found to increase further to 0.7–1.5 solar within the central 15–35 kpc, although in the innermost 2 kpc of M87 and NGC 4696 the iron abundance decreases down to $`{}_{}{}^{<}0.5`$ solar to form a central dip. In M87 (Matsushita et al. 2003), NGC 1399 (Buote 2002), NGC 5044 (Buote et al. 2003) and a sample of 19 clusters (Tamura et al. 2004), the spatial distributions of the silicon and sulphur abundances are likely to follow that of iron. The distribution of the oxygen abundance, on the other hand, is roughly uniform within the errors. These results strongly support the idea that there has been an excess SN Ia enrichment in the cluster’s inner region.
For the time being, the spatial distributions of metal abundances in clusters at $`z>0.1`$ are less understood due to the instrumental limitations, although there is evidence that the same central iron abundance excess may also exist (e.g., Allen et al. 2001, 2002; Iwasawa et al. 2001). As a step to approach more distant clusters, in this paper we present the Chandra study of the iron abundance distribution in MS 0839.9+2938, an intermediately distant cluster located at $`z=0.194`$ (Stocke et al. 1991). The cluster was listed as an extended X-ray source and identified as a cluster in the Einstein Extended Medium Sensitivity Survey (Gioia et al. 1990). It hosts a giant cD galaxy at its dynamic center, which is a weak, compact radio source with an intense, extended $`H_\alpha `$ emission (Nesci et al. 1995). The advent of studying this source is that since its average gas temperature is low ($`3.6`$ keV; see §4.2), it is easier to measure the abundance gradient accurately due to the large equivalent widths of Fe-$`L`$ and -$`K`$ lines.
We organize the paper as follows. In §2, we describe the observation and data reduction. In §3 and §4, we present the imaging and spectral analyses, respectively. In §5, we investigate the spatial distributions of the gas and gravitating mass as well as gas cooling. Finally, we discuss and summarize our results in §6 and §7, respectively. Throughout the paper, we adopt the cosmological parameters $`H_0=70`$ km s<sup>-1</sup> Mpc<sup>-1</sup>, $`\mathrm{\Omega }_m=0.3`$ and $`\mathrm{\Omega }_\mathrm{\Lambda }=0.7`$, so that $`1^{\prime \prime }`$ corresponds to $`3.2h_{70}^1`$ kpc at the redshift of the cluster. In order to compare directly with previous results, we use the older solar abundances from Anders & Grevesse (1989), where the iron abundance relative to hydrogen is $`4.68\times 10^5`$ in number, which is about 46% higher than the currently preferred meteoritic value (McWilliam 1997; Grevesse & Sauval 1999). Unless stated otherwise, the quoted errors are the 90% confidence limits.
## 2 Observation and Data Reduction
The Chandra observation of MS 0839.9+2938 was carried out on January 29, 2001 for a total exposure of 29.7 ks with the chips 2, 3, 6, 7, 8, and 9 of the Advanced CCD Imaging Spectrometer (ACIS). The center of the cluster was positioned close to the nominal aim point on the S3 chip (CCD 7), so that most of the X-ray emission of the galaxy was covered by the S3 chip. The events were collected with frame times of 3.2 s and telemetered in the Faint mode. The focal plane temperature was set to be -120 C. In this work, we used the Chandra data analysis package CIAO software (version 3.0) to process the data extracted from the S3 chip only. We kept events with ASCA grades 0, 2, 3, 4, and 6, and removed all the bad pixels, bad columns, and columns adjacent to bad columns and node boundaries. In order to detect occasional background flares, whose effects are particularly significant on the backside-illuminated S1 and S3 chips, we examined the lightcurves of the background regions on the S3 chip both in 2.5–7 keV, where the background flares are expected most significant, and in 4–7 keV, where the source contamination is least. We found that there was almost no strong background flares that increased the count rate to over 20% more than its mean value. By excluding the high background intervals we obtained a net exposure of 27.1 ks for the analyses. We restrict our spectral study in 0.7–8.0 keV to avoid the calibration uncertainties at the low energy end and the instrumental background at the high energy end. In the spectral fittings, we have taken into account corrections for the ACIS quantum efficiency degradation and used the XSPEC v11.2.0 software.
## 3 Imaging Analysis
### 3.1 X-ray Morphology
In Figure 1, we present both the smoothed 0.7–8.0 keV image of the central $`4.9^{}`$ ($`944h_{70}^1`$ kpc) of MS 0839.9+2938 in the logarithmic scale, and the corresponding DSS optical image on which the X-ray contours are overlaid. The smoothed X-ray image was created by using the CIAO tool csmooth with a minimum significance of 3 and a maximum significance of 5, and has been background-subtracted and exposure-corrected. As can be seen clearly, the diffuse X-ray emission is strongly peaked at the cluster’s center. Within the innermost $`8^{\prime \prime }`$ ($`26h_{70}^1`$ kpc), it appears to be elongated approximately in the northwest-southeast direction, where it follows the distribution of the optical lights. Outside this region the diffuse emission is elongated in the north-south direction out to $`1.1^{}`$ ($`210h_{70}^1`$ kpc). The X-ray emission of the cluster extends to at least $`4.2^{}`$ ($`800h_{70}^1`$ kpc), where the background begins to dominate the flux. There is no apparent substructures in X-rays, which indicates that the cluster has not experienced any major merger events in the recent past, and is likely to be in a relaxed state. Since the astrometric errors of the Chandra observation are about $`1^{\prime \prime }`$, the position of the X-ray peak (RA=08h42m56.0s, Dec=+29d27m27.2s, J2000) is in excellent agreement with the optical center of the cD galaxy 2MASX J08425596+2927272 (Becker et al. 1995) to within $`<0.5^{\prime \prime }`$.
Based on the studies of Brandt et al. (2000) and Mushotzky et al. (2000), we estimate that there may be 5–10 unrelated X-ray sources in the $`r<4.2^{}`$ region of the cluster. By utilizing a wavedetect detection algorithm and a source detection threshold of 10<sup>-6</sup>, and then crosschecking the results both by using the celldetect detection algorithm and by eye, we detected 10 X-ray point sources in $`r<4.2^{}`$, among which less than 1 is expected to be a false detection. We find that two of them have optical and radio counterparts within $`1^{\prime \prime }`$. One is CGCG 150-019, an Sb galaxy located at $`z=0.0276`$ (Wegner et al. 2001), and the other is FIRST J084251.9+292825, a Seyfert galaxy in MS 0839.9+2938 (Hutchings & Edwards 2000). In the analysis that follows, all the detected X-ray point sources were excluded.
### 3.2 Central Excess Emission
In Figure 2, we show the azimuthally-averaged surface brightness profile of the cluster in 0.7–8 keV that has been corrected with a weighted exposure map. All the detected point sources have been excluded, while the background has not been subtracted. We find that in $`15.8800h_{70}^1`$ kpc, the observed X-ray surface brightness profile can be well described ($`\chi _\nu ^2=1.05`$) by a model consisting of a standard $`\beta `$ component (Jones & Forman 1984) and a spatially uniform background as
$$S(r)=S_0[1+(r/r_\mathrm{c})^2]^{0.53\beta }+S_{\mathrm{bkg}},$$
(1)
where $`r_\mathrm{c}`$ is the core radius, $`\beta `$ is the slope, $`S_{\mathrm{bkg}}`$ is the background, and $`S_0`$ is the normalization. The best-fitting parameters are $`r_\mathrm{c}=51.7\pm 0.5`$ $`h_{70}^1`$ kpc and $`\beta =0.62\pm 0.01`$. However, when we extrapolate the model inwards, the fit becomes worse ($`r_\mathrm{c}=41.7\pm 0.4`$ $`h_{70}^1`$ kpc, $`\beta =0.59\pm 0.01`$, and $`\chi _\nu ^2=2.84`$), which significantly underestimates the data in $`r<4.9^{\prime \prime }`$ ($`15.8h_{70}^1`$ kpc), leaving an obvious central excess emission above the model. Such central excess emission has been seen in many clusters, e.g., the Centaurus cluster (Ikebe et al. 1999), Abell 1795 (Xu et al. 1998), Abell 1983 (Pratt & Arnaud 2003), and Cl0024+17 (Ota et al. 2004). It may be attributed to either a temperature drop, or a metal abundance increase, or an excess of the gas density, all within the central region. In §5.1, we will present detailed model fittings to describe the observed X-ray surface brightness profile by taking into account both the temperature and abundance gradients obtained from the spectral analyses (§4).
## 4 Spectral Analysis
### 4.1 Two-Dimensional Hardness Ratio Distribution
Before the direct spectral fittings, we first study the two-dimensional hardness ratio distribution of the hot diffuse gas in the cluster. The hardness ratio HR is defined as the ratio of the background-subtracted and exposure-corrected counts in 1.4–8.0 keV to those in 0.7–1.4 keV. As can be seen in Figure 3, in $`100300h_{70}^1`$ kpc the hardness ratio is approximately constant at a value of $`1.001.30`$, with typical errors of $`\pm 0.06`$. Assuming an average metallicity of 0.4 solar and the Galactic absorption $`N_\mathrm{H}=4.04\times 10^{20}`$ cm<sup>-2</sup> (Dickey & Lockman 1990), this is consistent with the hardness ratio of an isothermal plasma at $`kT4.06.5`$ keV. In $`r70140h_{70}^1`$ kpc, there are two relatively colder (90% confidence level) regions located at the southeast and northwest of the center, respectively. In the southeast colder region the gas temperature is estimated to be $`2.83.2`$ keV (HR$`=0.80\pm 0.04`$), which is close to that of the gas in the northwest colder region ($`3.03.2`$ keV, HR$`=0.83\pm 0.04`$). In $`r_{}^<40h_{70}^1`$ kpc, where the hardness ratio is $`0.600.83`$ with typical errors of $`\pm 0.04`$, the gas appears to be significantly colder than in the outer regions, indicating the existence of a plasma at $`kT2.23.1`$ keV. These results agree very well with those obtained in the detailed spectral fittings (§4.3).
### 4.2 Total Spectrum
We extracted the total spectrum of the cluster using a circular region filter with a radius of $`2.3^{}`$ ($`440h_{70}^1`$ kpc), which is centered at the X-ray peak, excluding all the detected point sources. The background spectrum was extracted from a separate region on the S3 chip that is far away enough from the cluster. We also have used the $`\mathrm{𝐶ℎ𝑎𝑛𝑑𝑟𝑎}`$ blank fields for background, and obtained essentially the same results as those shown below. At the boundary of the so defined cluster region, the count flux is about 3 times more than that of the background. The resulting background-subtracted spectrum contains a total of about 14,100 counts in 0.7–8.0 keV, showing a strong line feature at $`5.6`$ keV that corresponds to a rest-frame energy of about 6.7 keV where the K<sub>α</sub> lines of He-like iron should reside. The appearance of this Fe K<sub>α</sub> feature indicates that the average gas temperature of the cluster should be within 2.7–5.2 keV.
We have fit the spectrum with a model consisting of a single absorbed isothermal component (Table 1). First we choose the APEC model (Smith et al. 2001) as the isothermal component, and fix the redshift and absorption to $`z=0.194`$ and the Galactic value, respectively. The fit is marginally acceptable ($`\chi _\nu ^2=1.18`$), which gives an average gas temperature of $`kT=3.65_{0.20}^{+0.19}`$ keV, and an average metal abundance of $`Z=0.44_{0.08}^{+0.09}`$ solar. In order to improve the fit we have attempted to apply the VAPEC model instead of the APEC model. For MS 0839.9+2938, the redshift is large enough to move the Ly<sub>α</sub> lines of silicon to lower energies to avoid the calibration uncertainties caused by the mirror Ir edge. Thus in the fitting, we leave both the Fe and Si abundances free, and tie the abundance of Ni to that of Fe. Although we have applied the contamination file from the latest CALDB to correct for the ACIS deficiency in low energies, the degradation of energy resolution still does not recover. For an approximate estimation, we tie the abundances of other $`\alpha `$-elements (O, Ne, Mg, S, Al, Ar and Ca ) together. The model yields a gas temperature of $`kT=3.56_{0.20}^{+0.21}`$ keV, and abundances of $`Z_{\mathrm{Fe}}=0.46_{0.09}^{+0.10}`$ solar, $`Z_{\mathrm{Si}}=0.42_{0.29}^{+0.29}`$ solar, and $`Z_\alpha =1.20_{0.42}^{+0.48}`$ solar that is mostly dominated by the S abundance. Although the VAPEC fit ($`\chi _\nu ^2=1.08`$) is better than the APEC fit, there are still residuals remaining at about 0.9 keV, where the model underestimates the data. These residuals may mostly caused by the excess Fe-$`L`$ emissions arising from the colder gas in the central region, whose existence has been inferred in the study of the hardness ratio distribution (§4.1). In either the APEC or VAPEC fitting, even if the absorption is left free, the obtained column density is still consistent with the Galactic value, and the fit is not improved.
We also have examined if the fit to the total spectrum can be further improved by the use of a model consisting of two isothermal spectral components, both subjected to a common absorption (Table 1). For this purpose we employ two VAPEC components, whose absorption is again fixed to the Galactic value because allowing it to vary does not improve the fit. The metal abundances of each VAPEC component are grouped in the same way as in the single VAPEC fitting, and the abundances of the same element of the two VAPEC components are tied together. We perform the F-test and find that at the 95% confidence level the two-temperature fit ($`\chi _\nu ^2=1.01`$) is better than the one-temperature APEC or VAPEC fit. The obtained gas temperatures for the two components are $`kT_{\mathrm{low}}=2.10_{1.34}^{+0.75}`$ keV and $`kT_{\mathrm{high}}=6.04_{2.24}^{+4.81}`$ keV, and the metal abundances are $`Z_{\mathrm{Fe}}=0.38_{0.10}^{+0.12}`$ solar, $`Z_{\mathrm{Si}}=0.37_{0.24}^{+0.27}`$ solar, and $`Z_\alpha =0.96_{0.50}^{+0.47}`$ solar. With these best-fit parameters we calculate that the unabsorbed 0.5–10 keV flux of the cluster is $`3.9_{0.3}^{+0.2}\times 10^{12}`$ erg cm<sup>-2</sup> s<sup>-1</sup>, corresponding to a luminosity of $`4.1_{0.2}^{+0.4}\times 10^{44}h_{70}^2`$ erg s<sup>-1</sup>.
### 4.3 Projected and Deprojected Analyses
In order to investigate the temperature and abundance gradients, we divided the cluster into five annular regions, all centered at the X-ray peak. As the first step, we study the projected spectra extracted in these annuli. The background spectrum was again extracted in a region on the S3 chip far away enough from the cluster; the use of the Chandra blank fields yields essentially the same results. Assuming that the gas in each annulus is in the thermal equilibrium, we fit the spectra using a single absorbed APEC model with the absorption fixed to the Galactic value; allowing the absorption to vary does not improve the fits. As is shown in Figure 4 and Table 2, the fit is acceptable for each region. The gas temperature is found to be approximately constant at $`4.3`$ keV outside about $`130h_{70}^1`$ kpc, while it descends to $`3.2`$ keV in the central $`37h_{70}^1`$ kpc, as has been expected in the study of the hardness ratio distribution (§4.1). On the other hand, the average metal abundance, which is mainly determined by iron, increases towards the center from a roughly constant value of 0.4 solar in the outer regions to about 0.8 solar within the central $`37h_{70}^1`$ kpc. Instead of the APEC model, we have also attempted to apply the VAPEC model (Fig. 4 and Table 2). For better statistics, all the abundances of the $`\alpha `$-elements are tied together, and the absorption is again fixed to the Galactic value. We find that the fits are acceptable too, and the obtained gas temperatures agree nicely with those obtained with the APEC model. In the regions outside $`r=74h_{70}^1`$ kpc, the iron abundance varies slowly in a range of 0.25–0.42 solar, while in the central region, it increases drastically to about 0.8 solar. The spatial distribution of the average abundance of the $`\alpha `$-elements, however, is not well determined due to large errors.
Next, we perform the deprojected analysis of the spectra extracted in the inner four regions by using the standard ”onion peeling” method (Fig. 4 and Table 2). Assuming that the gas temperature and abundance are uniform in each spherical shell, and that the gas density follows the best-fit profile derived in §5.1, we subtract emissions from the outer shells projected on the inner shells. We fit the deprojected spectra with either the APEC or VAPEC model whose absorption is fixed to the Galactic value. We find that the deprojected spectra can be well fitted with either of the models. The gas temperatures obtained with the APEC model agree very well with those obtained with the VAPEC model, and both the average metal abundance of the APEC model and the iron abundance of the VAPEC model show significant central increase in the innermost region as has been found in the projected analysis. For each model the best-fit gas temperatures and metal abundances obtained in the projected and deprojected analyses are consistent with each other. In the innermost region, the deprojected temperatures are only slightly lower than, but still consistent with the projected values.
### 4.4 Temperature and Iron Abundance Gradients in the ICM
As we have shown above, both the projected and deprojected spectral analyses indicate that in MS 0839.9+2938 there is a central temperature drop within a radius of about $`74h_{70}^1`$ kpc. According to the deprojected APEC fit, the gas temperature descends from $`kT4.2`$ keV in $`>130h_{70}^1`$ kpc to $`2.7`$ keV in $`<37h_{70}^1`$ kpc, which can be approximated by a power-law profile $`T(R)=T_0(R/R_0)^\alpha `$, where $`T_0=3.0`$ keV, $`\alpha =0.17`$ and $`R_0=35.4h_{70}^1`$ kpc (Fig. 4a). In the same region where the cold gas is present, the iron abundance shows a clear tendency to increase. In the projected analysis, both the iron-dominated metal abundance of the APEC model and the iron abundance of the VAPEC model in the central $`37h_{70}^1`$ kpc region are significantly higher than their counterparts in $`74130h_{70}^1`$ kpc at the 90% confidence level. They are also apparently higher than the abundances in $`3774h_{70}^1`$ kpc and $`130230h_{70}^1`$ kpc, respectively, with the 90% confidence ranges overlap with each other only slightly. In the deprojected analysis, although the obtained errors are much larger, the average metal abundance and iron abundance still show a tendency to increase inward. At the 90% confidence level, their values inferred for the $`<37h_{70}^1`$ kpc region are significantly higher than those for the $`74130h_{70}^1`$ kpc region.
In order to crosscheck these results with better statistics, we divide the cluster into the inner ($`<74h_{70}^1`$ kpc) and outer ($`74444h_{70}^1`$ kpc) regions, and fit their deprojected spectra with the an absorbed VAPEC model (Table 3). In the fitting the absorption is fixed to the Galactic value, and the metal abundances are grouped in the same way as in the fittings of the total spectrum. We find that the fits are acceptable for both regions. At the 90% confidence level, the central temperature decrease and abundance increase can be confirmed. The silicon abundance and the average abundance of other $`\alpha `$-elements, however, show no tendency to vary between the two regions.
Since the iron abundance is determined as much by the Fe-L emission lines keV as by the Fe-K lines, the obtained inward iron abundance increase might be an artifact, which is actually caused by the enhancement of the Fe-L blend due to an inward temperature drop that is even steeper than has been modeled. We have examined this possibility. First we exclude the Fe-$`L`$ lines completely from the spectra and use the Fe-$`K`$ lines only to measure the iron abundance. The obtained iron abundance does show a tendency to increase inwards, although it is less significant due to poorer statistics. Second, we fit the deprojected spectrum of the innermost region by using a single absorbed VAPEC model with the absorption fixed to the Galactic value and the iron abundance fixed to 0.47 solar (the upper limit of the iron abundance in the third annular region; Table 2). We do obtain a lower temperature of $`2.38\pm 0.30`$ keV as is expected. However, the fit appears to be worse and should be ruled out at the 90% confidence level in terms of F-test, showing obvious spectral residuals in $`4.57.0`$ keV as well as at $`0.9`$ keV, where the fits to the Fe-$`K`$ and Fe-$`L`$ lines are worse. This indicates that a central abundance increase is much more preferred than an even steeper temperature gradient. Third, we study the two-dimensional fit-statistic contours of temperature and iron abundance at the 68%, 90% and 99% confidence levels for the innermost region and the third region, respectively, all obtained in the deprojected VAPEC fitting (Fig. 5). As can be clearly seen the central iron abundance is higher than that of the third region at a significance of 99%. Consequently, we conclude that the possibility of the detected central iron abundance increase being an artifact should be excluded. Note that the coexistence of the temperature drop and abundance increase may indicate that they are physically correlated, since the higher the metallicity is, the faster the gas cools down.
## 5 Mass Distributions and Cooling of Gas
### 5.1 Gas Density Profile
Assuming that 1) the gas is ideal and in the thermal equilibrium state, and 2) the spatial distribution of the gas is spherically asymmetric, we deduce the three-dimensional gas density distribution $`n_g(R)`$ by fitting the observed radial X-ray surface brightness profile $`S(r)`$, which was extracted in 0.7–8.0 keV, azimuthally-averaged, and exposure-corrected. The calculations was performed in a projected way by using
$$S(r)=S_0_r^{\mathrm{}}\mathrm{\Lambda }(T,Z)n_g^2(R)\frac{RdR}{\sqrt{R^2r^2}}+S_{\mathrm{bkg}},$$
(2)
where $`\mathrm{\Lambda }(T,Z)`$ is the cooling function calculated by taking into account both the temperature and abundance gradients as is modeled by the best-fit deprojected APEC parameters, and $`S_{\mathrm{bkg}}`$ is the background.
First we assume that the gas density follows the $`\beta `$ distribution
$$n_g(R)=n_{g,0}\left[1+(R/R_c)^2\right]^{3\beta /2},$$
(3)
where $`R_c`$ is the core radius and $`\beta `$ is the slope. We find $`R_c=47.3\pm 0.3h_{70}^1`$ kpc and $`\beta =0.61\pm 0.01`$. The fit, however, is unacceptable ($`\chi _\nu ^2=1.63`$), showing strong residuals in $`<15.8h_{70}^1`$ kpc where the model significantly underestimates the data (Fig. 2). In fact, the gas density derived in such a way is lower than that obtained in the deprojected spectral analysis (§4.3) by about 45%. This confirms the result in §3.2, although the central brightness excess above the model is less prominent since the effects of the central temperature drop and abundance increase are both counted in, which enhance the model emission in the central region.
Next we examine if $`S(r)`$ can be reproduced by adopting the universal gravitating mass distribution as has been found in the cosmological N-body simulations (Navarro et al. (1997)
$$\rho (R)=\frac{\delta _c\rho _{crit}}{(R/R_s)(1+R/R_s)^2},$$
(4)
where $`\rho (R)`$ is the gravitating mass density, $`R_s`$ is the scale radius, $`\rho _{crit}`$ is the critical density of the universe at the observed redshift, and $`\delta _c`$ is the characteristic density contract that can be expressed in term of the equivalent concentration parameter c as
$$\delta _c=\frac{200}{3}\frac{c^3}{[\mathrm{ln}(1+c)c/(1+c)]}.$$
(5)
As is shown in Makino et al. (1998) and Wu et al. (2000), under the assumptions of 1) spherical symmetry, 2) hydrostatic equilibrium, and 3) ideal and isothermal gas, the gas distribution can be given in an analytic form
$$\stackrel{~}{n}_g(x)=\stackrel{~}{n}_g(0)\frac{(1+x)^{\alpha /x}1}{e^\alpha 1},$$
(6)
where $`x=R/R_s`$ is the dimensionless radius, $`\alpha =4\pi G\mu m_p\delta _c\rho _{crit}R_s^2/kT`$ is the index, and $`\mu =0.609`$ is the average molecular weight for a fully ionized gas. Note that in Eq.(6) a background density at infinity has been subtracted to avoid the divergence in integrating the X-ray emission along the line of sight. As are shown in Figure 2 and Table 4. We find that, comparing with the $`\beta `$ model, the fit is improved in the central region, giving $`R_s=195.1\pm 0.5h_{70}^1`$ kpc, $`c=6.5\pm 0.1`$, and $`\chi _\nu ^2=1.12`$. However, it still systematically underestimates the data in $`<15.8`$ kpc.
In order to improve the fit in the central region, we have attempted to add an additional $`\beta `$ component to the $`\beta `$ model to account for the central excess emission, for which the gas density distribution is
$$n_g(R)=\left\{n_{g1,0}^2\left[1+(R/R_{c1})^2\right]^{3\beta _1}+n_{g2,0}^2\left[1+(R/R_{c2})^2\right]^{3\beta _2}\right\}^{1/2}.$$
(7)
This empirical two-component model has been successfully applied to many clusters of galaxies as well as some groups of galaxies (see, e.g., Makishima et al. 2001 and references therein). We find that the model can best fit the observed surface brightness profile throughout the spatial range spanned by the cluster and yield an acceptable fit. The best-fit parameters are $`R_{c1}=65.9_{0.5}^{+0.6}h_{70}^1`$ kpc and $`\beta _1=0.66_{0.01}^{+0.01}`$ for one $`\beta `$ component, which mostly describes the surface brightness profile in the outer region, and $`R_{c2}=15.5_{0.4}^{+0.3}`$$`h_{70}^1`$ kpc and $`\beta _2=0.59_{0.01}^{+0.01}`$ for another, which describes the central excess brightness.
### 5.2 Gas and Gravitating Mass Distributions
The distribution of the diffuse X-ray emission in different energy bands is roughly symmetric outside to about $`4.2^{}`$ ($`800h_{70}^1`$ kpc) in MS 0839.9+2938, indicating that the cluster is nearly relaxed. In this case the X-ray imaging spectroscopy can be considered as a reliable tool to deduce the mass distributions in the cluster. Assuming that the gas is ideal and the cluster is in the hydrostatic equilibrium state, and utilizing the best-fit temperature gradient obtained in the deprojected analyses and the gas density profile deduced with the two-$`\beta `$ model, we calculate the three-dimensional gravitating mass distribution by using
$$M(R)=\frac{kTR}{G\mu m_p}\left(\frac{d\mathrm{ln}n_g}{d\mathrm{ln}R}+\frac{d\mathrm{ln}T}{d\mathrm{ln}R}\right),$$
(8)
where $`M(R)`$ is the total mass within the radius $`R`$, G is the gravitational constant, and $`m_p`$ is the proton mass. Here the effects of non-thermal turbulent, magnetic pressure, and cosmic ray pressure are not taken into account. The resulting distribution of the total gravitating mass is shown in Figure 6, along with the 90% errors determined by performing the Monte-Carlo simulations that account for the range of temperature and gas profiles allowed by the data. The total mass distribution shows an excess beyond that predicted by the $`\beta `$ model in $`R_{}^<30h_{70}^1`$ kpc, which is roughly the optical boundary of the cD galaxy (Nesci et al. 1989) and the region where the excess X-ray emission, temperature drop, and iron abundance increase are observed. The excess mass ($`6\times 10^{11}M_{}`$) is comparable to the mass of the cD galaxy as is inferred from its optical luminosity ($`1\times 10^{12}M_{}`$ within $`30h_{70}^1`$ kpc; Nesci et al. 1989). At the virial radius $`R_{200}`$ ($`1640h_{70}^1`$ kpc), within which the average mass density is 200 times the current critical density of the universe, the extrapolated gravitating mass is $`M_{200}=6.1\times 10^{14}M_{}`$. In Figure 6, we also show the calculated gas mass fraction along with the 90% errors. It increases from about 0.05 at the center to about 0.1 at $`100h_{70}^1`$ kpc, and then keeps as a constant in the outer regions. For $`\mathrm{\Omega }_bh^2=0.024\pm 0.001`$ (Spergel et al. 2003), we obtain $`\mathrm{\Omega }_m=0.42_{0.14}^{+0.13}`$, which is self-consistent with the value adopted in this work.
### 5.3 Cooling of Gas and Mass Deposit Rate
With the gas density $`n_g(R)`$ deduced with the best-fit two-$`\beta `$ model (§5.1), we estimate the gas cooling time as
$$t_{\mathrm{cool}}=2.4\times 10^{10}\mathrm{yr}\left(\frac{\mathrm{𝑘𝑇}}{\mathrm{keV}}\right)^{1/2}\left(\frac{n_g}{10^3\mathrm{cm}^3}\right)^1$$
(9)
(cf, Sarazin 1986), where both continuum and line emissions are included (Fig. 6). We find that $`t_{\mathrm{cool}}10^8`$ yr at the center, which is much shorter than the age of the universe at the cluster’s redshift. The cooling radius $`R_{\mathrm{cool}}`$, defined as the radius where the cooling time is equal to the age of the universe, is $`150h_{70}^1`$ kpc. This is roughly the radius within which the gas temperature begins to descend inwards.
In order to determine the mass deposit rate of the cooling processes, we fit the spectrum extracted in the two innermost annuli ($`<74.1h_{70}^1`$ kpc) where the gas temperature is significantly lower than that of the outer regions. We use the cooling flow model MKCFLOW, which is absorbed by a column density fixed to the Galactic value (Fig. 7). We find that when the low temperature $`kT_{\mathrm{low}}`$ of MKCFLOW is fixed to 0.08 keV, the model yields a rather poor fit to the data ($`\chi _\nu ^2=1.69`$), giving rise to too much cool gas and thus overestimating the data in $`{}_{}{}^{<}0.8`$ keV. When $`kT_{\mathrm{low}}`$ is left free, the fit becomes acceptable ($`\chi _\nu ^2=1.00`$), but it requires $`kT_{\mathrm{low}}=1.14_{0.42}^{+0.31}`$ keV, which may suggest that the spectrum is dominated by a hot component. We then add an additional ambient isothermal component (MEKAL) into the model, tying its temperature to $`kT_{\mathrm{high}}`$ of MKCFLOW. When $`kT_{\mathrm{low}}`$ is fixed to 0.08 keV, we obtain $`kT_{\mathrm{high}}=3.40_{0.25}^{+0.27}`$ keV and $`\dot{M}=53_{44}^{+43}M_{}`$ yr<sup>-1</sup>; when $`kT_{\mathrm{low}}`$ is free, we obtain $`kT_{\mathrm{low}}=0.56_{0.48}^{+1.53}`$ keV, $`kT_{\mathrm{high}}=3.42_{0.19}^{+0.25}`$ keV, and $`\dot{M}=67_{33}^{+59}M_{}`$ yr<sup>-1</sup>. Whether or not $`kT_{\mathrm{low}}`$ is fixed, the MKCFLOW+MEKAL fit is acceptable ($`\chi _\nu ^2=0.99`$ and $`1.00`$, respectively), but $`kT_{\mathrm{low}}`$ and $`\dot{M}`$ are far from being well constrained. On the other hand, the two-temperature APEC model also gives a good fit ($`\chi _\nu ^2=0.99`$) to the data. The derived best-fit low and high temperatures are $`kT_{\mathrm{low}}=2.17_{0.66}^{+1.08}`$ keV and $`kT_{\mathrm{high}}=4.85_{1.38}^{+0.65}`$ keV. Thus, given the current data quality and instrumental capabilities, the two-phase isothermal model cannot be distinguished from the single MKCFLOW model with a free $`kT_{\mathrm{low}}`$, or the MEKAL+MKCFLOW model. Although not well constrained, the cooling flow model cannot be ruled out for this cluster.
## 6 Discussions
### 6.1 Metal Enrichment in ICM and Role of SNe Ia in Inner Region
The inward iron abundance increase found in many nearby clusters and groups around/in the giant, dominating elliptical galaxy at the center seems to follow a somewhat universal profile. Actually, when we compare the iron abundances in the inner ($`{}_{}{}^{<}70h_{70}^1`$) and outer regions ($`Z_{\mathrm{Fe},\mathrm{in}}`$ and $`Z_{\mathrm{Fe},\mathrm{out}}`$, respectively) of the clusters in which an inner abundance enhancement is observed (Buote et al. 2003; Nevalainen et al. 2001; Pratt $`\&`$ Arnaud, 2003; Takahashi $`\&`$ Yamashita 2003; Finoguenov et al. 2004; Tamura et al. 2004), we find that within the errors the ratio $`Z_{\mathrm{Fe},\mathrm{in}}/Z_{\mathrm{Fe},\mathrm{out}}`$ is roughly a constant as a function of either the ICM temperature or the cluster’s B-band luminosity (Fig. 8). All these results strongly imply that the observed central iron excess has probably been caused by the same enrichment process associated with the central dominating galaxy, which is a cD galaxy in most cases. Meanwhile, the high spatial resolution observations of the nearby giant ellipticals M87 (Matsushita et al. 2003) and NGC 5044 (Buote et al. 2003) revealed that the spatial distribution of oxygen abundance is approximately uniform out to about 40–70 kpc, inferring that the metals synthesized in SNe II are evenly distributed in the cluster/group. Thus, the central iron abundance increase should be ascribed mainly to SNe Ia.
By performing both the projected and deprojected spectral analyses, we find that in MS 0839.9+2938 the measured average iron abundance is about 0.4 solar in the outer regions, while it increases to $`1`$ solar within the central $`37h_{70}^1`$ kpc at the 90% confidence level. Reports about detections of such a significant central iron excess in clusters at $`z_{}^>0.2`$ are rare in literatures. In 4C+55.16 ($`z=0.240`$; Iwasawa et al. 2001) the metal abundance increases from $`{}_{}{}^{<}0.5`$ solar in outer regions to $`>1`$ solar within the central $`34h_{70}^1`$ kpc. In Abell 2390 ($`z=0.23`$; Allen et al. 2001), it increases from $`0.23_{0.13}^{+0.12}`$ solar in $`143714h_{70}^1`$ kpc to $`0.48_{0.10}^{+0.11}`$ solar within $`<71h_{70}^1`$ kpc. However, note that the results of the latter two cases were acquired by performing projected analysis at the 68% confidence level, and the abundance is an average value for both iron and $`\alpha `$-elements, although it is iron-dominated.
In MS 0839.9+2938, the excess iron mass within $`R<74h_{70}^1`$ kpc is estimated to be $`M_{\mathrm{Fe},\mathrm{excess}}3.0\times 10^8M_{}`$, which is similar to that of Abell 2390 ($`M_{\mathrm{Fe},\mathrm{excess}}2.5\times 10^8M_{}`$ within $`R<71h_{70}^1`$ kpc, based on the data of Allen et al. 2001). The derived $`M_{\mathrm{Fe},\mathrm{excess}}`$ obviously exceeds the iron enrichment caused only by the stellar winds and ram pressure stripping of metal-enriched gas (e.g., Trager et al. 2000; Ciotti et al. 1991). Therefore, we speculate that a significant part of the excess iron may have been blown into the ICM directly by the SN Ia explosions.
By reproducing the observed iron-to-silicon abundance ratio in the central $`74h_{70}^1`$ kpc of MS 0839.9+2938 (Fe/Si=$`1.29_{0.52}^{+1.40}`$ solar at the 68% confidence level), where a clear iron abundance increase is seen, we have attempted to estimate the SN Ia contribution to the iron enrichment. The current supernova models give very different yields to the metals (Gibson et al. 1997). When we adopt the theoretical SN Ia yields obtained with the W7 model and the weighted SN II yields in Nomoto et al. (1997), where an IMF slope of $`x=1.35`$ (Salpeter 1955) and a stellar mass range of $`1050M_{}`$ were assumed for the SN II progenitors, we find that the SN Ia contribution to the total iron mass is $`90_{18}^{+10}\%`$, corresponding to an iron abundance of $`Z_{\mathrm{Fe},\mathrm{SNIa}}=0.56_{0.11}^{+0.06}`$ solar. This fraction is consistent with those found in NGC 5044 ($`6779`$%; Buote et al. 2003), M87 ($`5597`$%; Matsushita et al. 2003), NGC 1399 ($`80\%`$; Buote 2002), and the RGH 80 group ($`80\%`$; Xue et al. 2004). We are not able to derive reasonable results by using the SN Ia yields obtained with either the WDD1 or WDD2 model in Nomoto et al. (1997), because both models give rise to too much silicon.
Using the same method as above we estimate that for the $`74<R<444h_{70}^1`$ kpc region, where the observed Fe/Si ratio is $`0.84_{0.47}^{+2.71}`$ solar (68% confidence), the SN Ia contributions is $`75_{55}^{+25}\%`$, or $`Z_{\mathrm{Fe},\mathrm{SNIa}}=0.23_{0.17}^{+0.08}`$ (W7 model). By comparing these results with those for the inner region, we find that the spatial distribution of the SN II contribution to iron is roughly constant at about $`0.08`$ solar, which is close to that in Finoguenov et al. (2002; $`Z_{\mathrm{Fe},\mathrm{SNII}}=0.10.15`$ solar) and Matsushita et al. (2003; $`Z_{\mathrm{Fe},\mathrm{SNII}}=0.20.4`$ solar).
It is crucial to examine if there has been enough time for SNe Ia to enrich the cluster’s central region with the observed amount of excess iron. Since the SN Ia enrichment of iron in the ICM is determined both by iron blown out directly into the ICM during the SN Ia explosions and by iron lost in the stellar winds, we calculate the enrichment time $`t_{\mathrm{enrich}}`$ (B$`\ddot{\mathrm{o}}`$hringer et al. 2004) using
$$t_{\mathrm{enrich}}=M_{\mathrm{Fe},\mathrm{SNIa}}/[L_\mathrm{B}\times (R_{\mathrm{SNIa}}+R_{\mathrm{wind}})],$$
(10)
where $`M_{\mathrm{Fe},\mathrm{SNIa}}`$ is the iron mass contributed by SNe Ia, $`L_\mathrm{B}`$ ($`2.2\times 10^{11}h_{70}^2\mathrm{L}_{}`$; Nesci et al. 1989) is cluster’s B-band luminosity, $`R_{\mathrm{SNIa}}`$ is the direct SN Ia iron-enriching rate in units of $`M_{}`$ yr<sup>-1</sup> $`L_{}^1`$, and $`R_{\mathrm{wind}}`$ is the iron-enriching rate due to the stellar mass loss in units of $`M_{}`$ yr<sup>-1</sup> $`L_{}^1`$. As is shown above, we assume that SNe II evenly enriched the gas throughout cluster, yielding an iron abundance of 0.08 solar, so by subtracting the SN II contribution, we obtain $`M_{\mathrm{Fe},\mathrm{SNIa}}=5.2\pm 0.8\times 10^8M_{}`$. In SN Ia explosions the rate of iron mass that has been directly transferred into the ICM is
$$R_{\mathrm{SNIa}}=SR10^{12}\mathrm{L}_{\mathrm{B},}^1\eta _{\mathrm{Fe}},$$
(11)
where $`\eta _{\mathrm{Fe}}`$ is the iron yield per SN Ia event, and $`SR`$ is the SN Ia rate in units of SNu (1 SNu = 1 supernova $`(10^{10}L_{B,})^1`$ century<sup>-1</sup>). In calculations, we choose $`\eta _{\mathrm{Fe}}=0.5`$ and 0.7 $`M_{}`$, respectively, and adopt either a temporally constant $`SR`$ of $`0.18h_{70}^2`$ SNu (Cappellaro et al. 1999), or an observation-constrained time-dependent $`SR`$ that increases with the redshift until $`z1`$ and then drops (Dahlen et al. 2004). In the stellar winds, the iron loss rate is
$$R_{\mathrm{wind}}=1.5\times 10^{11}\mathrm{L}_{\mathrm{B},}^1t_{15}^{1.3}\gamma _{\mathrm{Fe}},$$
(12)
where $`\gamma _{\mathrm{Fe}}`$ is the iron mass fraction in stellar winds, and $`t_{15}`$ is the age in units of 15 Gyr (Ciotti et al. 1991). We list the calculated enrichment times and redshifts at which the SN Ia enrichment is expected to start in Table 5. We find that when the supernova rate was a constant in the past, the derived enrichment time is too close to or even longer than the Hubble time at the cluster’s redshift ($`11.1\times 10^9`$ yr). When the observed time-dependent supernova rate provided in Dahlen et al. (2004) is adopted, the derived enrichment times are 7.9 Gyr and 6.4 Gyr for $`\eta =0.5`$ and $`0.7`$ $`M_{}`$, respectively. For comparison, we also perform similar calculations for the inner $`71h_{70}^1`$ kpc region of Abell 2390 and find that the resulting enrichment time is 6.6–10.7 Gyr. These results agree with those deduced for the Virgo cluster, the Centaurus cluster, the Perseus cluster and Abell 1795 ($`t_{\mathrm{enrich}}=511`$ Gyr; B$`\ddot{\mathrm{o}}`$hringer et al. 2004), indicating that the excess iron can be entirely produced by the central brightest galaxy (De Grandi & Molendi 2001; De Grandi et al. 2004). Since only few SNe Ia are observed at $`z_{}^>1.8`$, we predict that the most distant clusters showing a significant central iron excess should be detected at no farther than $`z0.5`$.
### 6.2 Scaling Relations
The derived X-ray temperature, X-ray luminosity and virial mass at $`R_{200}`$ of MS 0839.9+2938 are $`kT=3.65_{0.20}^{+0.19}`$ keV, $`L_\mathrm{X}=4.1_{0.2}^{+0.4}\times 10^{44}`$ $`h_{70}^2`$ erg s<sup>-1</sup>, and $`M_{200}=6.1\times 10^{14}h_{70}^1M_{}`$, respectively. Within the errors, these values are consistent with the observed luminosity-mass relation (e.g., Markevitch et al. 1998) and mass-temperature relation (e.g., Sanderson et al. 2003) after the effect of gas cooling is compensated. Using the gas density obtained with the best-fit two-$`\beta `$ model (§5.1) and the analytical power-law temperature profile (§4.4), we calculate the azimuthally averaged gas entropy $`S=kT/n^{2/3}`$ in the cluster and plot its spatial distribution in Figure 9. In $`R>0.03R_{200}`$ ($`49h_{70}^1`$ kpc), the profile can be approximated by a power-law form with an index of 1.1, as has been expected from shock heating that occurred during the spherical collapse (Tozzi & Norman 2001). At $`0.1R_{200}`$, the derived gas entropy is $`167_{32}^{+22}h_{70}^{1/3}`$ keV cm<sup>2</sup>, which is consistent with the entropy-temperature relation for a sample of 66 relaxed clusters presented in Ponman et al. (2003).
## 7 Summaries
We present the Chandra ACIS study of the intermediately distant ($`z=0.194`$) cluster of galaxies MS 0839.9+2938. By performing both the projected and deprojected spectral analyses, we find that the emission-measure weighted gas temperature is approximately constant at about 4 keV in $`>130h_{70}^1`$ kpc. In the inner regions, the gas temperature descends towards the center, reaching $`{}_{}{}^{<}3`$ keV within the innermost $`37h_{70}^1`$ kpc. Along with the temperature drop, we detect a significant inward iron abundance increase from about 0.4 solar in the outer regions to $`1`$ solar within the central $`37h_{70}^1`$ kpc. Besides MS 0839.9+2938, another cluster at $`z0.2`$ that has been found to show a similar, strong central iron excess is Abell 2390 ($`z=0.23`$; Allen et al. 2001). We argue that most of the excess iron is contributed by SNe Ia. By using the observed SN Ia rate and stellar mass loss rate, we estimate that in MS 0839.9+2938 the time needed to enrich the central region with excess iron is $`6.47.9`$ Gyr, which is similar to those found for the nearby clusters. In almost the same region where the temperature drop and abundance increase are seen, the observed X-ray surface brightness profile shows an excess beyond the distribution expected by either the $`\beta `$ model or the NFW model, and can be well fitted with an empirical two-$`\beta `$ model. The origins of all these phenomena can be correlated with each other (cf, Makishima et al. 2001). The excess amount of gas, a part of which comes from the supernova explosions with heavy metals, is trapped around the central dominating galaxy to form the observed surface brightness excess. The relatively high gas density and high metallicity both speed up the gas cooling process, causing the observed inward gas temperature drop.
This work was supported by the National Science Foundation of China (Grant No. 10273009 & 10233040), the Ministry of Science and Technology of China (Grant No. NKBRSF G19990754), and Shanghai Key Projects in Basic Research (No. 04JC14079).
Figure Captions
Fig.1 – (a): Central $`4.9^{}`$ ($`944h_{70}^1`$ kpc) of MS 0839.9+2938 in 0.7–8 keV, which is plotted in the logarithmic scale. The image has been background-subtracted, exposure-corrected, and smoothed with a minimum significance of 3 and a maximum significance of 5. X-ray intensity contours are spaced from $`1.4\times 10^{10}`$ to $`4.7\times 10^7`$ photons cm<sup>-2</sup> s<sup>-1</sup> pixel<sup>-2</sup>. The inner $`1^{}`$ region, which is marked by a box, is magnified in (b), where the contour levels are from $`5.7\times 10^9`$ to $`2.5\times 10^7`$ photons cm<sup>-2</sup> s<sup>-1</sup> pixel<sup>-2</sup>. (c): Corresponding DSS optical image for the same sky field as in (a) on which the X-ray contours are overlaid for comparison.
Fig.2 – Azimuthally-averaged radial surface brightness profile of MS 0839.9+2938 in 0.7–8 keV. The background is assumed to be spatially uniform and is represented by the dotted line. In $`15.8800h_{70}^1`$ kpc, the surface brightness profile can be well described by the $`\beta `$ model (dashed line), showing an obvious central emission excess. The fits of the two-$`\beta `$, $`\beta `$ and NFW models to the whole cluster region, and the fit of the $`\beta `$ model to the outer region only are shown with the solid, dashed, dot-dashed and dot-dot-dot-dashed lines, respectively.
Fig.3 – Two-dimensional distribution of the hardness-ratio of the diffuse emission in MS 0839.9+2938, which is defined as the ratio of the background-subtracted and exposure-corrected counts in 1.4–8.0 keV to those in 0.7–1.4 keV. The cross represents the peak of X-ray emission, and the circle ($`r=74h_{70}^1`$ kpc) marks the region in which significant temperature drop and abundance increase are detected. The image is plotted in linear scale, and has been smoothed in the same way as in Figure 1.
Fig.4 – (a): Projected (solid) and deprojected (dashed) radial distributions of the gas temperature, which are obtained with the absorbed APEC model. The best-fit temperature profile for the deprojected distribution is shown as a dotted line. (b): The same as (a) but for the absorbed VAPEC model. (c): Projected (solid) and deprojected (dashed) radial distributions of the average metal abundance, which are obtained with the absorbed APEC model. (d): The same as (c) but for the iron abundance obtained with the absorbed VAPEC model.
Fig.5 – Confidence contours at the 68.3%, 90% and 99% confidence levels for the gas temperature and iron abundance derived with the deprojected VAPEC model for the innermost ($`R<37.3h_{70}^1`$ kpc) and the third ($`74.1<R<130.1h_{70}^1`$ kpc) annular regions.
Fig.6 – Upper and middle panels: calculated radial distributions of the total gravitating mass, gas mass and gas mass fraction, along with the 90% error limits (shown as dotted lines). The total mass was calculated by using both $`\beta `$ and the best-fit two-$`\beta `$ model. The gas mass and gas fraction were calculated by using the best-fit two-$`\beta `$ model. Lower panel: the calculated cooling time as a function of radius. The horizontal, dashed line represents the age of the universe at the redshift of cluster. Radii within which the gas temperature drops and iron abundance increases significantly ($`R_\mathrm{a}`$), and at which the gas temperature begins to decrease inwards ($`R_\mathrm{b}`$), are indicated by vertical lines.
Fig.7 – Upper and middle panels: Spectrum extracted in $`<74.1h_{70}^1`$ kpc, where the gas temperature is significantly lower than that of the outer regions, along with the best-fit two-temperature APEC model and residuals. Lower panel: model ratios of the single MKCFLOW model and MEKAL+MKCFLOW model to the two-temperature APEC model. $`kT_{\mathrm{low}}`$ of the MKCFLOW component is fixed to 0.08 keV.
Fig.8 – Ratios of the iron abundance in cluster’s inner region to that in the outer region Z<sub>in</sub>/Z<sub>out</sub> as a function of the cluster’s (a) average gas temperature and (b) optical luminosity for nearby clusters (solid; Tamura et al 2004; Girardi et al. 2000; David et al. 1995; Pratt & Arnaud 2003) and two intermediately distant clusters, i.e., MS 0839.9+2938 (dashed) and Abell 2390 (dot-dashed. This is the average abundance ratio rather than iron abundance ratio; Allen et al. 2001). The ratio for the nearby clusters is shown as a horizontal dotted line.
Fig.9 – Radial distribution of the gas entropy (crosses and the dashed line) along with the theoretical expectation quoted from Tozzi et al. 2001 (solid line).
|
warning/0506/math0506572.html
|
ar5iv
|
text
|
# The isomorphism problem for Coxeter groups
## 1. Introduction
Coxeter groups are important in several mathematical areas. It is therefore a bit surprising that the isomorphism problem for those groups does not seem to have been considered before the late 1990’s. They only earlier reference known to the author where this problem has been asked is . The first major contributions to it are and . In a rigidity result is proved for a certain class of Coxeter groups. Rigidity means that the Coxeter generating sets are all conjugate. In diagram twists have been introduced. Those provide non-trivial examples of non-rigid Coxeter groups. The question about which Coxeter systems are rigid arises naturally as well as the more general question about the isomorphism problem for Coxeter groups.
The purpose of the present paper is to give a survey about what is known at present about the isomorphism problem. The main motivation for writing this survey is provided by a recent result obtained by the author in collaboration with Bob Howlett. This result reduces the isomorphism problem to its ‘reflection-preserving version’. For the solution of the latter there is a conjecture stated in . Considerable progress towards a proof of this conjecture was made in and by recent work of Pierre-Emmanuel Caprace in there is reasonable hope that this conjecture will be proved in the near future. Due to these facts there is now a clear picture of what the solution of the isomorphism problem should look like. In fact, at present there is a solution if one assumes that there are no irreducible spherical residues of rank 3. We state two conjectures in Section 5. The first one is known to be true for all Coxeter systems having no $`H_3`$-subsystems; the second is a refinement of Conjecture 8.1 in already mentioned. Under the assumption that both conjectures are true, we give an algorithm for the solution of the isomorphism problem.
### Two versions of the isomorphism problem
Let $`W`$ be a group and let $`SW`$ be a set of involutions. Then $`M(S)`$ denotes the square matrix $`(o(ss^{}))_{s,s^{}S}`$ where $`o(w)`$ denotes the order of an element $`wW`$. The matrix $`M(S)`$ is called the type of $`S`$. As the elements of $`S`$ are involutions we have the following.
* For all $`s,s^{}S`$ we have $`o(ss^{})𝐍\{\mathrm{}\}`$;
* for all $`ss^{}S`$ we have $`o(ss^{})=o(s^{}s)2`$;
* for all $`sS`$ we have $`o(ss)=1`$.
Hence, the matrix $`M(S)`$ is a symmetric square matrix with entries in the set $`𝐍\{\mathrm{}\}`$ where all entries on the main diagonal are equal to one and all remaining entries are strictly greater than one. Such a matrix is called a Coxeter matrix over $`S`$.
Let $`(W,S)`$ be as above. We call $`(W,S)`$ a Coxeter system (of type $`M(S)`$) if $`S=W`$ and if the relations $`((ss^{})^{o(ss^{})}=1_W)_{s,s^{}S}`$ provide a presentation of $`W`$. For a given Coxeter matrix $`M=(M_{ij})_{i,jI}`$ over a set $`I`$, we define the Coxeter group of type $`M`$ by setting $`W(M):=I((ij)^{m_{ij}}=1)_{i,jI}`$. It is a basic fact that the pair $`(W(M),I)`$ is a Coxeter system of type $`M`$ (i.e. that $`o(ij)=m_{ij}`$ in $`W(M)`$ for all $`i,jI`$).
In this paper we will consider the isomorphism problem for finitely generated Coxeter groups. Thus, if we talk about a Coxeter system $`(W,S)`$ or a Coxeter matrix $`M`$ over $`I`$ it is always understood that the sets $`S`$ and $`I`$ are finite.
Here are two versions of the isomorphism problem for Coxeter groups.
Problem 1 Given two Coxeter matrices $`M`$ and $`M^{}`$, decide whether the groups $`W(M)`$ and $`W(M^{})`$ are isomorphic.
Problem 2 Given two Coxeter matrices $`M`$ and $`M^{}`$, find all isomorphisms from $`W(M)`$ onto $`W(M^{})`$.
At first sight, Problem 1 seems to be a more natural question than Problem 2. The latter is just a more general version of the first. Roughly speaking, the solution of Problem 2 is equivalent to the solution of Problem 1 and a description of the automorphism group of $`W(M)`$ for any Coxeter matrix $`M`$. This is in fact the main motivation to consider Problem 2. It turns out that for certain Coxeter matrices $`M`$ a good understanding of the automorphism group of the group $`W(M)`$ is only possible if a solution of Problem 2 is available for all Coxeter matrices $`M^{}`$.
### Content
In Section 2 we recall some definitions, fix notation and mention some basic facts concerning Coxeter groups. In Section 3 we will consider the rigidity problem for Coxeter groups. This is an interesting special case of the isomorphism problem. In this section we will provide examples of non-rigid Coxeter systems which will play an important role later. Section 4 is devoted to explaining the results obtained in and and how these results reduce the isomorphism problem to its ‘reflection-preserving version’ which will then be treated in Section 5. In Section 6 we explain an algorithm to solve the isomorphism problem under the assumption that Conjectures 1 and 2 of Section 5 hold. Finally, in Section 7 we will make some remarks on the automorphism groups of Coxeter groups.
Remark: It was mentioned above that there is no contribution to the isomorphism problem for Coxeter groups before the late 1990’s. Since then, however, there are several publications concerning this subject. For instance, Problem 1 has been solved completely in the case where $`M`$ is assumed to be even (i.e. no odd entries) by P. Bahls and M. Mihalik (see and the references given there).
In this survey paper we do not attempt to give a systematic description of all contributions to the isomorphism problem for Coxeter groups. We mention results (or consequences of them) whenever it will be convenient. However, we try to include all references on the subject in the bibliography. Thus, quite a few references will be mentioned only there.
### Acknowledgement
The content of this paper is based on my talk at the Coxeter Legacy Conference at Toronto in May 2004. I thank the organizers for the invitation to present this survey at this conference.
## 2. Preliminaries
### Coxeter diagrams
With a Coxeter matrix $`M=(m_{ij})_{i,jI}`$ we associate its diagram $`\mathrm{\Gamma }(M)`$. It is the edge-labelled graph $`(I,E(M))`$, where the edge-set is $`E(M):=\{\{i,j\}m_{ij}3\}`$ and where an edge $`\{i,j\}E(M)`$ has the label $`m_{ij}`$. We do not distinguish between a Coxeter matrix and its diagram since they carry the same information. We call a Coxeter matrix irreducible if its associated Coxeter diagram is connected. An irreducible component of $`M`$ is a subset $`J`$ of $`I`$, which is a connected component of the diagram. A Coxeter matrix $`M`$ is called spherical if $`W(M)`$ is finite. The irreducible spherical Coxeter diagrams have been classified by H.S.M. Coxeter in ; we will use the Bourbaki notation for denoting them with the exception that we denote rank 2 diagrams for the dihedral groups of order $`2n`$ by $`I_2(n)`$. Thus we have the four series $`A_n,C_n=B_n,D_n`$ and $`I_2(n)`$ and the 6 exceptional diagrams $`E_6,E_7,E_8,F_4,H_3`$ and $`H_4`$.
An isomorphism from a Coxeter diagram $`M=(m_{ij})_{i,jI}`$ onto a Coxeter diagram $`M^{}=(m_{ij}^{})_{i,jI^{}}`$ is a graph isomorphism which preserves the edge-labels.
Let $`M=(m_{ij})_{i,jI}`$ be a Coxeter matrix over $`I`$ and let $`J`$ be a subset of $`I`$. Then we put $`M_J:=(m_{jk})_{j,kJ}`$ and $`J^{}:=\{kIm_{kj}=2\text{ for all }jJ\}`$.
A Coxeter matrix $`M`$ is called right-angled if all edge-labels of $`\mathrm{\Gamma }(M)`$ are infinite; it is called 2-spherical if there are no infinities; it is called even if there are no odd labels and it is called of large type if the diagram is a complete graph (hence if there are no 2’s in $`M`$).
### Coxeter systems
Let $`(W,S)`$ be a Coxeter system. The set of its reflections is defined to be the set $`S^W:=\{wsw^1sS\text{ and }wW\}`$. The length of $`wW`$ is the length of a shortest product of elements in $`S`$ representing $`w`$; it is denoted by $`l(w)`$. We call $`(W,S)`$ right-angled, 2-spherical, even or of large type if this is the case for $`M(S)`$.
We list some facts about Coxeter systems which are important in the sequel. Facts 1 and 2 are basic and can be found in any standard reference on Coxeter groups (see or ); Fact 3 is a non-trivial exercise in but it follows also from the fact that the Davis-complex of a Coxeter system is CAT(0); Fact 4 is contained in ; Fact 5 can be shown by considering the geometric representation and Fact 6 is just an easy consequence of the definition of a Coxeter system.
* If $`JS`$, then $`(J,J)`$ is a Coxeter system.
* Let $`JS`$ and $`l:W𝐍`$ be the length function of $`(W,S)`$. Then the following are equivalent:
+ $`(J,J)`$ is finite;
+ there is an element $`\rho _J`$ such that $`l(\rho _J)>l(x)`$ for all $`\rho _JxJ`$.
Moreover, if these two conditions are satisfied, then $`\rho _J^2=1_W`$.
* If $`XW`$ is a finite subgroup, then there exist $`wW`$ and $`JS`$ such that $`X^wJ`$ and such that $`J`$ is a spherical subset of $`S`$ (i.e. $`J`$ finite).
* Let $`rW`$ be an involution. Then there exist $`wW`$ and $`JS`$ such that $`J`$ is spherical, $`w\rho _Jw^1=r`$ and such that $`\rho _J`$ is central in $`J`$.
* Suppose that $`J`$ is a spherical subset of $`S`$ such that $`\rho _J`$ is central in $`J`$. Then the normalizer of $`J`$ in $`W`$ and the centralizer of $`\rho _J`$ in $`W`$ coincide.
* Let $`(W,S)`$ be a Coxeter system. Then each permutation $`\pi `$ of $`S`$ which is an automorphism of $`M(S)`$ extends uniquely to an automorphism $`\gamma _\pi `$ of $`W`$.
Let $`(W,S)`$ be a Coxeter system. By Fact 6 we can identify the stabilizer of $`S`$ in $`\mathrm{Aut}(W)`$ with the group of automorphisms of $`M(S)`$; this subgroup will be denoted by $`\mathrm{\Gamma }_S(W)`$ and its elements are called the graph-automorphisms of $`(W,S)`$. The group $`\mathrm{\Gamma }_S`$ has trivial intersection with the group $`\mathrm{Inn}(W)`$ of inner automorphisms. An automorphism of $`W`$ will be called inner-by-graph if it can be written as a product of an inner automorphism and a graph-automorphism.
## 3. Rigidity
Let $`G`$ be a group and $`RG`$ a set of involutions. Recall that the Coxeter matrix $`M(R)`$ is called the type of $`R`$; the set $`R`$ is called universal if $`(R,R)`$ is a Coxeter system; it is called a Coxeter generating set of $`G`$ if it is universal and $`G=R`$.
A Coxeter matrix $`M`$ is called rigid if for each Coxeter generating set $`R`$ of $`W(M)`$ the Coxeter diagrams $`M(R)`$ and $`M`$ are isomorphic. It is called strongly rigid if any two Coxeter generating sets of $`W(M)`$ are conjugate in $`W(M)`$.
Clearly, strong rigidity implies rigidity. If a Coxeter diagram is (strongly) rigid, then we call the corresponding Coxeter group and Coxeter system (strongly) rigid as well.
If one can show that the Coxeter diagram $`M`$ of Problem 1 is rigid, then this problem is trivially solved. The answer is just that the Coxeter diagram $`M^{}`$ has to be isomorphic to $`M`$.
Similarly, if one can show that the Coxeter diagram $`M`$ is strongly rigid, then Problem 2 is solved. An isomorphism onto $`W(M^{})`$ exists if and only if $`M^{}`$ and $`M`$ are isomorphic. Moreover, the automorphism group of $`W(M)`$ is just the semi-direct product of the group of inner automorphisms with the group of graph-automorphisms of $`W(M)`$; in other words: all automorphisms of $`W`$ are inner-by-graph.
There are several interesting classes of Coxeter systems which are not rigid. Before describing them we present some positive results. The first is due to D. Radcliffe .
###### Theorem 3.1
Right-angled Coxeter systems are rigid.
Although we fixed the convention that all Coxeter systems in this paper are by definition of finite rank it is appropriate to mention that the theorem above has been generalized to right-angled Coxeter systems of arbitrary rank by A. Castella (see ). The next result about strong rigidity is the result of R. Charney and M. Davis already mentioned in the introduction (see ).
###### Theorem 3.2
Let $`(W,S)`$ be a Coxeter system. If $`W`$ is capable of acting effectively, properly and cocompactly on some contractible manifold, then $`(W,S)`$ is strongly rigid. In particular, Coxeter groups of affine and compact hyperbolic type are strongly rigid.
The next result is very recent. An important step towards a proof of it was already made in ; in the version presented here it is a consequence of the main results in and .
###### Theorem 3.3
Suppose that $`(W,S)`$ is irreducible, non-spherical and 2-spherical, then $`(W,S)`$ is strongly rigid.
In the following we describe two ways to manipulate the generating set of a given Coxeter system in order to produce a new one whose type is possibly non-isomorphic to the type of the original one. It is conjectured (and known to be true in a lot of special cases) that Coxeter systems are rigid up to these manipulations.
### Pseudo-Transpositions
Let $`k1`$ be a natural number and put $`n:=2(2k+1)`$. We consider the dihedral group $`W`$ of order $`2n`$ as the group of isometries preserving a regular $`n`$-gon in the euclidian plane. Let $`s,tW`$ be two reflections whose axes intersect in an angle $`\frac{\pi }{n}`$ and let $`\rho `$ be the central symmetry. Then it is easily seen that $`\{s,t\}`$ and $`\{s,tst,\rho \}`$ are both Coxeter generating sets for $`W`$ of type $`I_2(n)`$ and $`I_2(2k+1)\times A_1`$ respectively. Thus, the dihedral group of order $`2n`$ is a non-rigid Coxeter group because it has two Coxeter generating sets of different types. This example is of course trivial and a bit cheating because one of the two Coxeter matrices is not irreducible. However, it can be used to produce more general examples by taking direct products or free products. In pseudo-transpositions have been introduced in order to describe the general feature.
Let $`(W,S)`$ be a Coxeter system and let $`\tau S`$. We call $`\tau `$ a pseudo-transposition if the following holds.
* There is a unique $`tS`$ such that $`o(\tau t)=2(2k+1)`$ for some natural number $`k1`$.
* For all $`sS\{\tau ,t\}`$ one has $`o(\tau s)\{2,\mathrm{}\}`$ and if $`o(s\tau )=2`$, then $`o(st)=2`$ as well.
The following is an easy observation about pseudo-transpositions.
###### Lemma 3.4
Let $`(W,S)`$ be a Coxeter system, let $`\tau S`$ be a pseudo-transposition of $`(W,S)`$ and let $`tS`$ be as in the definition above. Then $`S\{\tau \}\{\tau t\tau ,\rho _{\{\tau ,t\}}\}`$ is a Coxeter generating set of $`W`$.
There is also another kind of pseudo-transpositions for Coxeter systems based on the fact that the Coxeter groups $`W(C_n)`$ and $`W(D_n\times A_1)`$ are isomorphic for odd $`n`$. They yield also non-isomorphic Coxeter generating sets in a similar way. We refer to for the details.
Let $`(W,S)`$ be a Coxeter system, let $`\tau S`$ be a pseudo-transposition and let $`R`$ be the ‘new’ Coxeter generating set as described in the lemma above. Then we call the Coxeter system $`(W,R)`$ an elementary reduction of $`(W,S)`$. A Coxeter system $`(W,S^{})`$ will be called a reduction of $`(W,S)`$ if it can be obtained from $`(W,S)`$ by a sequence of elementary reductions. Finally, we call $`(W,S)`$ reduced, if there are no pseudo-transpositions. It is easy to see that each Coxeter system has a reduced reduction.
Given a Coxeter diagram $`M`$ over a set $`I`$, then a Coxeter diagram $`M^{}`$ over $`I^{}`$ is called an elementary reduction of $`M`$ if there is an elementary reduction of the Coxeter system $`(W(M),I)`$ whose type is isomorphic to $`M^{}`$; we call $`M^{}`$ a reduction of $`M`$ if $`M^{}`$ can be obtained from $`M`$ by a sequence of elementary reductions and we call $`M`$ reduced if the system $`(W(M),I)`$ has no pseudo-transpositions.
Clearly, any rigid Coxeter system has to be reduced in view of Lemma 3.4 above. The following result is due to M. Mihalik and is based on earlier work of P. Bahls ; it states that the converse is true for even Coxeter systems.
###### Theorem 3.5
An even Coxeter system is rigid if and only if there is no pseudo-transposition.
Note that this result generalizes Theorem 3.1.
### Twistings
In this subsection we describe twistings as they were introduced in and we give some further definitions concerning them.
Let $`(W,S)`$ be a Coxeter system and let $`J,KS`$. We call the pair $`(J,K)`$ an $`S`$-admissible pair if the following holds.
* $`J`$ is a spherical subset of $`S`$ and $`K(JJ^{})=\mathrm{}`$.
* For all $`kK`$ and $`lL:=S(JJ^{}K)`$ the order of $`kl`$ is infinite.
An $`S`$-admissible pair $`(J,K)`$ is called trivial if $`K`$ or $`L`$ are empty. For a $`S`$-admissible pair $`(J,K)`$ we put $`T_{(J,K)}(S):=JJ^{}K\{\rho _Jl\rho _JlL\}`$.
The following lemma is not too difficult to prove (see ).
###### Lemma 3.6
Let $`(W,S)`$ be a Coxeter system and let $`(J,K)`$ be a $`S`$-admissible pair. Then $`T_{(J,K)}(S)`$ is a Coxeter generating set of $`W`$ which is contained in $`S^W`$.
Let $`(W,S),(J,K)`$ and $`S^{}:=T_{(J,K)}(S)`$ be as in the previous lemma. If $`\rho _J`$ is central in $`J`$, then it is easily verified that $`M(S)`$ is isomorphic to $`M(S^{})`$. If $`\rho _J`$ is not central in $`J`$, then $`M(S)`$ is not isomorphic to $`M(S^{})`$ in the generic case. The following example of such a situation was given in .
Example: Let $`(W,S)`$ be a Coxeter system such that $`S=\{s_1,s_2,s_3,s_4\}`$ and such that $`o(s_1s_2)=o(s_2s_3)=o(s_3s_4)=3`$ and $`o(s_1s_3)=o(s_1s_4)=o(s_2s_4)=\mathrm{}`$. We put $`J:=\{s_2,s_3\}`$ and $`K:=\{s_1\}`$. It follows that
$$S^{}:=T_{(J,K)}(S):=\{s_1^{}:=s_1,s_2^{}:=s_2,s_3^{}:=s_3,s_4^{}:=s_2s_3s_2s_4s_2s_3s_2\}$$
and that $`o(s_1^{}s_2^{})=o(s_2^{}s_3^{})=o(s_2^{}s_4^{})=3`$ and $`o(s_1^{}s_3^{})=o(s_1^{}s_4^{})=o(s_3^{}s_4^{})=\mathrm{}`$. Thus $`M(S)`$ and $`M(S^{})`$ are not isomorphic.
Let $`S,R`$ be Coxeter generating sets of a group $`W`$; we call $`R`$ a twist of $`S`$ if there is a $`S`$-admissible pair $`(J,K)`$ such that $`R=T_{(J,K)}(S)`$. It is readily verified that $`R`$ is a twist of $`S`$ if and only if $`S`$ is a twist of $`R`$ and that $`S^W=R^W`$ in this case. A Coxeter generating set $`S`$ is called twist-rigid if there are no non-trivial $`S`$-admissible pairs; i.e. if there are no twists of $`S`$ which are not conjugate to $`S`$ in $`W`$.
Let $`M`$ be a Coxeter matrix over $`I`$. A Coxeter matrix $`M^{}`$ is called a twist of $`M`$ if there is a twist $`I^{}`$ of $`I`$ in the Coxeter system $`(W(M),I)`$ such that $`M(I^{})`$ is isomorphic with $`M^{}`$. As before one verifies that $`M^{}`$ is a twist of $`M`$ if and only if $`M`$ is a twist of $`M^{}`$.
We close this section with a result about strong rigidity for Coxeter groups. Obviously, if $`(W,S)`$ is a strongly rigid Coxeter system, then $`S`$ has to be twist-rigid. The following theorem provides the converse under the additional assumption that all Coxeter generating sets $`R`$ of $`W`$ are contained in $`S^W`$. In view of Corollary 4.2 below there are ‘a lot of examples’ where this assumption holds.
###### Theorem 3.7
Let $`M`$ be a non-spherical, irreducible Coxeter diagram over $`I`$ such that there is no subdiagram of type $`H_3`$. Suppose that $`I`$ is a twist-rigid subset of $`W(M)`$ and that all Coxeter generating sets of $`W(M)`$ are contained in $`I^{W(M)}`$. Then $`M`$ is strongly rigid.
This theorem was first proved in the large-type case ($`m_{ij}>2`$ for all $`i,j`$) in ; the result as it is stated above has been obtained recently by P.-E. Caprace .
## 4. The reduction to the restricted isomorphism problem
The restricted isomorphism problems for Coxeter groups are the following:
Problem 3: Given a Coxeter system $`(W,S)`$ and a Coxeter matrix $`M`$, decide whether there is a Coxeter generating set $`RS^W`$ of $`W`$ such that $`M(R)=M`$.
Problem 4: Given a Coxeter system $`(W,S)`$ and a Coxeter matrix $`M`$, find all Coxeter generating sets $`RS^W`$ of $`W`$ with $`M(R)=M`$.
In Problems 1 and 2 of the introduction have been reduced to Problems 3 and 4 respectively. This reduction is based on the results on the finite continuation of a reflection in a Coxeter group, which have been obtained in . The purpose of this section is to describe the results obtained in both references. The original motivation for the investigations in was to find a tool to characterize reflections in abstract Coxeter groups. We first provide some examples, where an abstract Coxeter group does not determine ‘its set of reflections’.
We have already seen examples, where an abstract Coxeter group has different Coxeter generating sets yielding different sets of reflections. If $`(W,S)`$ is not reduced and if $`R`$ is an elementary reduction of $`S`$, then $`S^WR^W`$ and $`R^WS^W`$. We will now obtain further examples by producing automorphisms of Coxeter groups which do not preserve reflections. There are two kinds of such automorphisms, namely $`s`$-transvections and $`J`$-local automorphisms.
### $`s`$-Transvections
Let $`(W,S)`$ be a Coxeter system and let $`sS`$. We define the odd connected component of $`s`$ in the diagram $`\mathrm{\Gamma }(S)`$ to be the set of all elements $`tS`$ for which there is a path from $`s`$ to $`t`$ such that all its edge-labels are odd. We denote the odd component of $`s`$ by $`\mathrm{odd}(s)`$ and we put
$$\mathrm{eodd}(s):=\mathrm{odd}(s)\{tSo(tt^{})\mathrm{}\text{ for some }t^{}\mathrm{odd}(s)\}.$$
Let $`J_s`$ denote the irreducible component of $`\mathrm{eodd}(s)`$ which contains $`s`$ and let $`K_s`$ denote the union of all spherical irreducible components of $`\mathrm{eodd}(s)`$ which do not contain $`s`$.
Let $`z`$ be an element in the center of $`K_s`$. We define the mapping $`\theta _{s,z}:SW`$ by setting $`\theta _{s,z}(t)=tz`$ if $`t\mathrm{odd}(s)`$ and by setting $`\theta _{s,z}(t)=t`$ for the remaining $`tS`$. One readily verifies that this mapping extends to an involutory automorphism of $`W`$ and that $`sz`$ is not contained in $`S^W`$. Hence $`\theta _{s,z}(S)`$ is a Coxeter generating set of $`W`$ providing a different set of reflections.
The involutory automorphism described above is called an $`s`$-transvection of the Coxeter system $`(W,S)`$. In fact, the definition of an $`s`$-transvection given in is slightly more general. This is due to particular instances which might arise when there are subsystems of type $`C_3`$. Due to these instances the formal definition of an $`s`$-transvection is somewhat involved and will be omitted here. Nevertheless, we give an example of such a $`C_3`$-transvection because - unlike for the other kinds of automorphisms - it is not an ‘obvious automorphism easily seen from the diagram’.
Example Let $`(W,S)`$ be a Coxeter system where $`S=\{s,t,t^{},c\}`$ such that $`o(st)=o(st^{})=3`$, $`o(ct)=o(ct^{})=4`$, $`o(sc)=2`$ and $`o(tt^{})=\mathrm{}`$. Define $`\theta :SW`$ by setting $`\theta (c):=c`$, $`\theta (s):=sc`$, $`\theta (t):=stcsts`$ and $`\theta (t^{}):=st^{}cst^{}s`$. One verifies that $`\theta `$ extends uniquely to an involutory automorphism of $`W`$.
### $`J`$-local automorphisms
Let $`(W,S)`$ be a Coxeter system. A subset $`J`$ of $`S`$ is called a graph factor of $`(W,S)`$ if $`J`$ is spherical and if for all $`tSJ`$ either $`tj=jt`$ for all $`jJ`$ or $`o(tj)=\mathrm{}`$ for all $`jJ`$.
Let $`J`$ be a graph factor of $`(W,S)`$ and let $`\alpha `$ be an automorphism of $`J`$. Then it is readily verified that there is a unique automorphism of $`W`$ stabilizing the subgroup $`J`$, inducing $`\alpha `$ on it and inducing the identity on $`SJ`$. We call such an automorphism a $`J`$-local automorphism.
This observation can be used to produce non-reflection preserving automorphisms. There are lots of examples of finite Coxeter groups, having automorphisms which are not reflection preserving. Obvious examples are the elementary abelian 2-groups. A particularly interesting example is of course the exceptional automorphism of $`\mathrm{Sym}(6)`$ which is the Coxeter group of type $`A_5`$.
### The finite continuation of a reflection
Let $`(W,S)`$ be a Coxeter system. As $`S`$ is supposed to be finite and as each finite subgroup of $`W`$ is conjugate to a subgroup of some spherical standard parabolic subgroup it follows that there is an upper bound for the order of any finite subgroup of $`W`$. This implies that there is for any subgroup $`X`$ of $`W`$ a unique maximal normal finite subgroup of $`X`$ which we denote by $`O_{\text{fin}}(X)`$.
Let $`rW`$ be an involution of $`W`$; by the result of Richardson mentioned in Section 2 (Fact 4) we know that $`r`$ is conjugate to some $`\rho _J`$ for some spherical subset $`J`$ of $`S`$ and such that $`\rho _J`$ is central in $`J`$. Now one knows that $`N_W(J)=C_W(\rho _J)`$ (Fact 5) and hence $`J`$ is contained in $`O_{\text{fin}}(C_W(\rho _J))`$. These considerations show that $`r`$ must be a reflection if $`O_{\text{fin}}(C_W(r))=r`$. Hence we have found a handy criterion which ensures that a given involution of an abstract Coxeter group is a reflection for any Coxeter generating set of that group.
This idea was the starting point for the results obtained in . It soon turned out that it is more convenient to work with the finite continuation $`\mathrm{FC}(r)`$ rather than with the group $`O_{\text{fin}}(C_W(r))`$. This is defined to be the intersection of all maximal finite subgroups of $`W`$ containing $`r`$. The main result of is the following theorem. Its proof is based on a careful analysis of the centralizer of a reflection which had been desrcibed in detail in .
###### Theorem 4.1
Let $`(W,S)`$ be a Coxeter system and let $`sS`$. Then $`\mathrm{FC}(s)`$ is known. Moreover, if $`\mathrm{FC}(s)=s`$ , then $`s`$ is a reflection for each Coxeter generating set of $`W`$.
The description of $`\mathrm{FC}(s)`$ may become complicated if there are subsystems of type $`C_3`$ or $`D_4`$. If this is not the case, one can describe $`\mathrm{FC}(s)`$ by means of the subsets $`J_s`$ and $`K_s`$ defined in the paragraph on $`s`$-transvections as follows.
###### Corollary 4.2
Let $`(W,S)`$ be a Coxeter system and suppose that there is no subsystem of type $`C_3`$ or $`D_4`$. Let $`sS`$. If $`J_s`$ is spherical, then $`\mathrm{FC}(s)=J_sK_s`$; in the remaining cases one has $`\mathrm{FC}(s)=\{s\}K_s`$. In particular, if $`K_s=\mathrm{}`$ and $`J_s`$ is non-spherical, then $`s`$ is a reflection for each Coxeter generating set of $`W`$.
### The reduction theorem
Let $`(W,S)`$ be a Coxeter system. We call $`sS`$ $`\mathrm{FC}`$-centered if $`\mathrm{FC}(s)=J`$ for some $`JS`$. A fundamental reflection might not be $`\mathrm{FC}`$-centered if there are subsystems of type $`C_3`$ or $`D_4`$. Moreover, the group of automorphisms of $`W`$ which stabilize the subset $`S^W`$ is denoted by $`\mathrm{Ref}_S(W)`$. We are now able to state the main result of .
###### Theorem 4.3
Let $`(W,S)`$ be a reduced Coxeter system. For each $`\mathrm{FC}`$-centered $`sS`$, let $`T_s`$ denote the group of all $`s`$-transvections of $`(W,S)`$. For each graph factor $`JS`$ let $`L_J`$ denote the group of all $`J`$-local automorphisms of $`(W,S)`$. Let $`\mathrm{\Sigma }`$ be the subgroup of $`\mathrm{Aut}(W)`$ which is generated by all $`T_s`$ and all $`L_J`$, where $`s`$ runs through the $`\mathrm{FC}`$-centered elements of $`S`$ and $`J`$ runs through the set of graph factors of $`(W,S)`$. Let $`\stackrel{~}{\mathrm{\Sigma }}`$ be the subgroup of $`\mathrm{Aut}(W)`$ which stabilizes $`\mathrm{FC}(s)`$ for all $`sS`$. Then we have the following:
* The group $`\stackrel{~}{\mathrm{\Sigma }}`$ is finite and $`\mathrm{\Sigma }\stackrel{~}{\mathrm{\Sigma }}`$. In particular, $`\mathrm{\Sigma }`$ is a finite subgroup of $`\mathrm{Aut}(W)`$.
* Given a reduced Coxeter system $`(W^{},S^{})`$ and an isomorphism $`\alpha :WW^{}`$, then there exists $`\sigma \mathrm{\Sigma }`$ such that $`\alpha (\sigma (S))S^W^{}`$.
* The group $`\mathrm{\Sigma }`$ (and hence also the group $`\stackrel{~}{\mathrm{\Sigma }}`$) is a finite supplement of $`\mathrm{Ref}_S(W)`$ in $`\mathrm{Aut}(W)`$.
Part b) of the theorem above says in particular, that if $`(W,S)`$ and $`(W^{},S^{})`$ are Coxeter systems which are both reduced and if there is an isomorphism from $`W`$ onto $`W^{}`$, then there is also an isomorphism between them which maps $`S^W`$ onto $`S^W^{}`$. This yields the reduction of Problem 1 to Problem 3 for reduced Coxeter systems. Moreover, given any reduced Coxeter system $`(W,S)`$, then its group of automorphism can be written as $`\mathrm{\Sigma }\mathrm{Ref}_S(W)`$, hence Problem 2 is reduced to Problem 4 for reduced Coxeter systems.
## 5. The restricted isomorphism problem
In view of the reduction result described in the previous section it suffices to solve Problems 3 and 4 in order to solve Problems 1 and 2 respectively. Thus we are led to the following question.
Question: Let $`(W,S)`$ be a Coxeter system and let $`RS^W`$ be a Coxeter generating set of $`W`$. What can be said about $`R`$?
We have to consider Coxeter generating sets whose elements are reflections in a given Coxeter system. The following is a first observation which can be shown by using the geometric representation of a Coxeter group.
###### Lemma 5.1
Let $`(W,S)`$ be a Coxeter system, let $`RS^W`$ be a Coxeter generating set of $`W`$ and let $`XR`$ be such that $`X`$ is finite. Then there exists a subset $`J`$ of $`S`$ and an element $`wW`$ such that $`X^w=J`$. In particular, if $`r,r^{}R`$ are such that $`o(rr^{})=n\mathrm{}`$, then there exist $`s,s^{}S`$ such that $`o(ss^{})=n`$ and such that the subgroups $`r,r^{}`$ and $`s,s^{}`$ are conjugate.
Let $`(W,S)`$ be a Coxeter system and let $`RS^W`$ be a Coxeter generating set. We call $`R`$ sharp-angled with respect to $`S`$ if for any two reflections $`r,r^{}R`$ there exists $`wW`$ such that $`\{r,r^{}\}^wS`$.
Let $`W`$ be the dihedral group of order $`2n`$ for some natural number $`n2`$. We consider $`W`$ as the group of automorphisms of the regular $`n`$-gon in the euclidean plane. Let $`S=\{s,t\}`$, where $`s`$ and $`t`$ are reflections whose axes intersect in an angle $`\frac{\pi }{n}`$. Given $`rr^{}S^W`$, then $`\{r,r^{}\}`$ is sharp-angled with respect to $`S`$ if the reflection axes of $`r`$ and $`r^{}`$ intersect in an angle $`\frac{\pi }{n}`$.
### Angle-deformations
Let $`(W,S)`$ be a Coxeter system, let $`stS`$ be such that $`st`$ has finite order, let $`xs,t`$ be such that $`s,xtx^1=s,t`$ and put $`Y:=S(\{s,t\}\{s,t\}^{})`$. Let $`Y_s`$ be the set of all $`yY`$ for which there exists a sequence $`y_1,\mathrm{},y_k=y`$ in $`Y`$ such that $`o(sy_1),o(y_1y_2),\mathrm{},o(y_{k1}y_k)`$ are finite and define $`Y_t`$ analogously. We define the mapping $`\delta _x:SW`$ by setting $`\delta _x(r):=r`$ if $`rS(Y_t\{t\})`$ and $`\delta _x(r)=xrx^1`$ in the remaining cases. The following is easy to verify.
###### Lemma 5.2
If $`Y_sY_t=\mathrm{}`$ then $`\delta _x`$ extends uniquely to an automorphism of $`W`$ which stabilizes the set $`S^W`$.
If $`\{s,xtx^1\}`$ is not sharp-angled with respect to $`\{s,t\}`$ and if $`\delta _x`$ is as above, then $`\delta _x(S)`$ is not sharp-angled with respect to $`S`$. We therefore call the automorphisms of the lemma above angle-deformations.
The following result can be obtained by using rigidity of Fuchsian Coxeter groups in a similar way as it was done in .
###### Proposition 5.3
Let $`(W,S)`$ be a Coxeter system and suppose that there is no 3-subset $`J`$ of $`S`$ such that $`M(J)=H_3`$. Let $`\mathrm{\Delta }`$ be the group generated by all angle deformations of $`(W,S)`$. Given a Coxeter generating set $`RS^W`$, then there exists $`\delta \mathrm{\Delta }`$ such that $`\delta (R)`$ is sharp-angled with respect to $`S`$.
In view of the previous proposition the following conjecture is known to be true for Coxeter systems having no subsystem of type $`H_3`$.
Conjecture 1: Let $`(W,S)`$ be a Coxeter system and $`RS^W`$ be a Coxeter generating set. Then there exists an automorphism $`\alpha `$ of $`W`$ such that $`\alpha (S^W)=S^W`$ and such that $`\alpha (R)`$ is sharp-angled with respect to $`S`$.
### Twist-equivalence
Let $`(W,S)`$ be a Coxeter system and let $`RS^W`$ be a Coxeter generating set of $`W`$. Recall that $`R^{}S^W`$ is called a twist of $`R`$ if there is an $`R`$-admissible pair $`(J,K)`$ such that $`R^{}=T_{(J,K)}(R)`$. Moreover, $`R^{}`$ is a twist of $`R`$ if and only if $`R`$ is a twist of $`R^{}`$. By taking the transitive closure we obtain an equivalence relation on the set of the Coxeter generating sets contained in $`S^W`$ which is called twist-equivalence.
If $`R^{}`$ is a twist of $`RS^W`$, then $`R^{}R^W`$ and $`R^{}`$ is sharp-angled with respect to $`R`$. Hence, if $`R^{}`$ is twist-equivalent with $`RS^W`$, then $`R^{}R^W`$ and $`R^{}`$ is sharp-angled with respect to $`R`$. There is some evidence that the converse is also true. This is the content of the conjecture below. This conjecture is a refinement of Conjecture 8.1 in .
Conjecture 2: Let $`(W,S)`$ be a Coxeter system and $`RS^W`$ a Coxeter generating set of $`W`$ which is sharp-angled with respect to $`S`$. Then $`R`$ is twist-equivalent to $`S`$.
At present, the following two theorems are known by recent work of P.-E. Caprace. The first improves earlier results obtained in , and .
###### Theorem 5.4
Conjecture 2 holds for all Coxeter systems which do not contain an irreducible spherical subsystem of rank 3.
###### Theorem 5.5
If $`(W,S)`$ is a Coxeter system such that $`M(J^{})`$ is 2-spherical for each spherical subset $`J`$ of $`S`$, Conjecture 2 holds for $`(W,S)`$.
The main tool to prove Conjecture 2 in the references above is known to the experts as ‘Kac Conjugacy Theorem for root bases’. This theorem is proved in for affine and compact hyperbolic groups. A proof for all Coxeter groups is given in .
## 6. The solution of Problem 1
Let $`M`$ be a Coxeter diagram over a set $`I`$. Recall that $`M^{}`$ is called a twist of $`M`$ if there is a twist $`I^{}`$ of $`IW(M)`$ such that $`M(I^{})`$ is isomorphic to $`M^{}`$. Again, $`M^{}`$ is a twist of $`M`$ if and only if $`M`$ is a twist of $`M^{}`$ and by taking the transitive closure we obtain an equivalence relation on the set of Coxeter matrices which is called twist-equivalence as well.
The following lemma is easy to prove.
###### Lemma 6.1
Let $`(W,S)`$ be a Coxeter system and let $`M`$ be a Coxeter matrix. Then the following are equivalent.
* There exists a Coxeter generating set $`RS^W`$ such that $`M(R)`$ is isomorphic to $`M`$ and such that $`R`$ is twist-equivalent to $`S`$.
* The matrices $`M(S)`$ and $`M`$ are twist-equivalent.
Using the previous lemma one obtains the following theorem, which yields the solution of Problem 3.
###### Theorem 6.2
Let $`(W,S)`$ and $`(W^{},S^{})`$ be Coxeter systems and suppose that Conjectures 1 and 2 hold for $`(W,S)`$. Then the following are equivalent.
* $`M(S)`$ and $`M(S^{})`$ are twist-equivalent.
* There exists an isomorphism $`\alpha :W^{}W`$ such that $`\alpha (S^{})S^W`$
We recall that a Coxeter system $`(W,S)`$ is reduced if the set $`S`$ contains no pseudo-transposition, that there is a natural notion of a Coxeter system or a Coxeter matrix to be a reduction of another and that it is always possible to produce a reduced reduction of a Coxeter system or Coxeter matrix by an easy algorithm. Now the previous theorem and Theorem 4.3 yield the following.
###### Theorem 6.3
Let $`M`$ and $`M^{}`$ be irreducible Coxeter matrices of rank at least 3 and let $`(W,S)`$ be a Coxeter system of type $`M`$. If Conjectures 1 and 2 hold for $`(W,S)`$, then the following are equivalent.
* The groups $`W(M)`$ and $`W(M^{})`$ are isomorphic.
* If $`M_1`$ is a reduced reduction of $`M`$ and if $`M_1^{}`$ is a reduced reduction of $`M^{}`$, then $`M_1`$ and $`M_1^{}`$ are twist equivalent.
In view of Theorem 5.4 and Proposition 5.3 we have the following corollary.
###### Corollary 6.4
Let $`M`$ and $`M^{}`$ be Coxeter matrices and suppose that $`M`$ has no subdiagram of type $`A_3,C_3`$ or $`H_3`$, then the following are equivalent:
* The groups $`W(M)`$ and $`W(M^{})`$ are isomorphic.
* If $`M_1`$ is a reduced reduction of $`M`$ and if $`M_1^{}`$ is a reduced reduction of $`M^{}`$, then $`M_1`$ and $`M_1^{}`$ are twist equivalent.
## 7. On automorphisms of Coxeter groups
The previous section shows that there is—under the hypothesis that Conjectures 1 and 2 are true—a satisfactory solution of Problem 1. Unfortunately, we cannot offer a satisfactory description of the automorphism groups of Coxeter groups under the same assumptions which would yield a solution of Problem 2 as well. In fact, the author has serious doubts whether such a handy description exists in the general case. Nevertheless there are several natural subgroups of the automorphism group of a Coxeter group which are quite well understood. In most of the ‘interesting’ cases, the understanding of these subgroups suffices to understand the group of automorphisms as a whole. Our discussion will be restricted to those subgroups. Before going more into the details we would like to mention that the automorphism groups of Coxeter groups had been determined in various special cases.
* A presentation of the automorphism groups of right-angled Coxeter groups was given in . This work was based on the results obtained in and the latter is a far reaching generalization of the result in .
* The automorphism groups of 2-spherical Coxeter groups are ‘trivial’ (i.e. all automorphisms are inner-by-graph) if there is no direct factor which is spherical. This result was accomplished in and . A ‘virtual’ result in this direction has been obtained already in and the main tool developed there was used again in .
* The automorphism groups of several classes of Coxeter groups which are ‘almost spherical’ have been described in , , and . In a complete description of the automorphism groups of the irreducible spherical Coxeter groups is given.
Given an abstract Coxeter group $`W`$, then there is always a Coxeter generating set $`SW`$ such that $`(W,S)`$ is reduced. Thus, there is no loss of generality if we consider only reduced Coxeter systems in this section. Let $`(W,S)`$ be a reduced Coxeter system. We define the following subgroups:
* $`\mathrm{Ref}_S(W):=\{\alpha \mathrm{Aut}(W)\alpha (S^W)=S^W\}`$,
* $`\mathrm{Ang}_S(W):=\{\alpha \mathrm{Ref}_S(W)\alpha (S)\text{ sharp-angled with respect to }S\}`$,
* $`\stackrel{~}{\mathrm{\Sigma }}_S(W):=\{\alpha \mathrm{Aut}(W)\alpha (\mathrm{FC}(s))=\mathrm{FC}(s)\text{ for all }sS\}`$
* $`\mathrm{\Gamma }_S(W):=\{\alpha \mathrm{Aut}(W)\alpha (S)=S\}`$
In view of Theorem 4.3 we have $`\mathrm{Aut}(W)=\stackrel{~}{\mathrm{\Sigma }}_S(W)\mathrm{Ref}_S(W)`$ and the group $`\stackrel{~}{\mathrm{\Sigma }}_S(W)`$ is a finite group. Thus, there is a finite supplement of $`\mathrm{Ref}_S(W)`$ in $`\mathrm{Aut}(W)`$. There is the natural question about minimal supplements (or even complements) of $`\mathrm{Ref}_S(W)`$ in $`\mathrm{Aut}(W)`$. The example of the Coxeter group of type $`A_1^k`$ shows that there are not always complements. However, a careful analysis of several special cases provides some evidence for the following conjecture.
Conjecture 3: Let $`(W,S)`$ be a reduced Coxeter system. Then there exists a subgroup $`\mathrm{\Omega }\stackrel{~}{\mathrm{\Sigma }}_S(W)`$ such that $`\mathrm{\Pi }:=\mathrm{\Omega }\mathrm{Ref}_S(W)\mathrm{\Gamma }_S(W)`$ and such that $`\mathrm{\Omega }`$ is a supplement of $`\mathrm{Ref}_S(W)`$ in $`\mathrm{Aut}(W)`$. Moreover, there is a normal 2-subgroup $`U`$ of $`\mathrm{\Omega }`$ and a complement of $`L`$ of $`U`$ in $`\mathrm{\Omega }`$ such that $`L=L_1\times L_2\times \mathrm{}L_k`$ where $`L_i`$ is isomorphic to $`\mathrm{GL}(n_i,2)`$ for some natural number $`n_i`$ for $`1ik`$ and $`\mathrm{\Pi }L_i`$ is just the set of permutation matrices.
There is a canonical candidate for the choice of the group $`\mathrm{\Omega }`$ and based on this choice the validity of the conjecture is not difficult to see in several special cases. However, the arguments become somewhat involved in the general case.
### Reflection-preserving automorphisms
As $`\mathrm{Ref}_S(W)`$ has a finite supplement, a big part of $`\mathrm{Aut}(W)`$ is understood if $`\mathrm{Ref}_S(W)`$ is understood. A first observation is that $`\mathrm{Ang}_S(W)`$ is a normal subgroup of finite index in $`\mathrm{Ref}_S(W)`$ and therefore a similar remark holds for $`\mathrm{Ang}_S(W)`$. We do not know whether $`\mathrm{Ang}_S(W)`$ always has a finite supplement in $`\mathrm{Ref}_S(W)`$ but we believe that there are examples where this is not the case. If there is no $`H_3`$-subdiagram, then the group $`\mathrm{Ref}_S(W)`$ is generated by the angle-deformations of $`(W,S)`$ and $`\mathrm{Ang}_S(W)`$. We expect this to be true in general with a suitable definition of angle-deformations in the case where there are $`H_3`$-subdiagrams.
In the following we will consider the group $`\mathrm{Ang}_S(W)`$. Let
$$𝐑:=\{RS^WR\text{ sharp-angled Coxeter generating set of }W\text{ with respect to }S\}$$
and call two elements $`RR^{}`$ in $`𝐑`$ adjacent if one is a twist of the other. This yields a graph which we call $`𝐂`$. Conjecture 2 is equivalent to the statement that the graph $`𝐂`$ is connected.
We consider first the special case where $`M(S)`$ is even in which case Conjectures 1 and 2 are known to be true. If $`M(S)`$ is even, there is for each neighbor $`R`$ of $`S`$ in the graph $`𝐂`$ a canonical involution $`\theta _R`$ in $`\mathrm{Ang}_S(W)`$ which switches $`S`$ and $`R`$. Setting $`X:=\theta _RR\text{ neighbor of }S`$, one verifies that $`𝐂`$ is the Cayley graph of $`X`$ with respect to this generator set and that $`\mathrm{\Gamma }_S`$ is a complement of $`X`$ in $`\mathrm{Ang}_S(W)`$. It is probably possible to generalize the arguments given in in order to give a presentation of the group $`\mathrm{Ang}_S(W)`$. The key ingredient of such a generalization would be the observation that the group $`\mathrm{Ang}_S(W)`$ is something like a ‘generalized Coxeter group’ as it is in the right-angled case.
Let’s consider the general case under the assumption that Conjecture 2 holds. The situation becomes more complicated. The graph $`𝐂`$ is no longer the Cayley graph of a group but of a groupoid. We do not go into the details here. But it is worth mentioning that a similar situation occurs if one is interested in the normalizer of a parabolic subgroup in a Coxeter group. These normalizers had been described in and in a satisfactory way. The key observation in is that they are finite index subgroups of a groupoid which one might call a Coxeter groupoid in view of its properties which are quite similar to those of Coxeter groups. We believe, that a presentation of $`\mathrm{Ang}_S(W)`$ can be given by using analogous ideas. It would be based on the observation that the graph $`𝐂`$ is the Cayley-graph of a generalized Coxeter groupoid of which $`\mathrm{Ang}_S(W)`$ is a subgroup of finite index. However, a concrete description of such a presentation might become rather involved.
|
warning/0506/cond-mat0506602.html
|
ar5iv
|
text
|
# Potential of mean force and the charge reversal of rodlike polyions
## I Introduction
It is our pleasure to contribute this paper to the special issue of Molecular Physics dedicated to celebrate Ben Widom’s outstanding contributions to Physical Chemistry and Statistical Mechanics. Ben’s work is characterized by a profound physical insight, combined with an ability to abstract the most complex physical phenomena into a simple model. From scaling and criticality Widom (1965) to microemulsions Widom (1984) and the hydrophobic effect Kolomeisky and Widom (1999), Ben’s sagacity has opened new frontiers of Physical Chemistry. While it is impossible to compete with Ben’s intuition, one can at least try to follow his example. In this paper we will, therefore, study a simple model of interaction between a polyion and multivalent counterions inside a polyelectrolyte solution.
Thermodynamic systems in which long range Coulomb interactions play the dominant role pose an outstanding challenge to Physical Chemistry Levin (2002). Even such basic question as the possible existence of a liquid-gas phase separation in a restricted primitive model has been positively settled only quite recently Levin (2002). Even so, the order of this transition still remains a source of an outstanding debate Luijten et al. (2002). For strongly asymmetric electrolytes such as aqueous colloidal suspensions, even the existence of a liquid-liquid phase separation continues to be controversial van Roij and Hansen (1997); Levin et al. (1998); Diehl et al. (2001); Deserno and Grünberg (2002); Tamashiro and Schiessel (2003); Trizac and Levin (2004).
When aqueous colloidal suspensions or polyelectrolyte solutions contain multivalent counterions other curious phenomena appear. For example, it is found that for sufficiently small separations two like-charged polyions can attract one another Rouzina and Bloomfield (1996); Grønbech-Jensen et al. (1997); Ha and Liu (1997); Arenzon et al. (1999); Hansen and Löwen (2000); Gelbart et al. (2000); Solis and de la Cruz (2001); Angelini et al. (2003). If an external electric field is applied to such a suspension the electrophoretic mobility of colloidal particles is often found to be reversed, so that the particles move in the direction opposite to the one expected based purely on their chemical charge Lozada-Cassou et al. (1982); Messina et al. (2002); Grosberg et al. (2002); Levin (2002); Martin-Molina et al. (2003). Both of these phenomena are a consequence of strong electrostatic coupling between the polyions and the counterions.
The counterions inside the suspension can be divided into two categories: those which are associated (condensed) to the colloidal particle and those which are free. The condensed counterions contribute to the effective, renormalized, charge of the polyion-counterion complex, while the free counterions and coions result in screening of the electrostatic interactions inside the suspension Levin (2002). In this paper we will explore the potential of mean force between a rodlike polyion with $`n`$ associated counterions and a counterion located at a transverse distance $`d`$ from the polyion center, Fig. 1.
## II The model
Consider a rodlike polyion of $`Z`$ (even) monomers, each carrying a charge $`q`$, inside an aqueous suspension containing multivalent counterions and salt. The monomers are located uniformly with separation $`b`$ along the rod. Strong electrostatic coupling between the polyion and the counterions results in a condensation of $`n`$ $`\alpha `$-valent counterions onto the polyion. The condensed counterions are free to hop between the monomers of the polyion Arenzon et al. (1999). If a monomer has an associated counterion, its charge is renormalized to $`(\alpha 1)q`$. The free, uncondensed, counterions and coions screen the electrostatic interactions, changing the potential between the two charges $`q_1`$ and $`q_2`$ from the Coulomb to the Debye-H ckel Debye and Hückel (1923) form
$$V(r)=\frac{1}{ϵ}\frac{q_1q_2\mathrm{exp}(\kappa r)}{r},$$
(1)
where $`ϵ`$ is the dielectric constant of the solvent and $`\kappa `$ is the inverse Debye length. The question that we would like to address in this paper is what is the potential of mean force between the polyion-counterion complex containing $`n`$ condensed $`\alpha `$-ions and an additional $`\alpha `$-valent counterion located transversely at distance $`d`$ from the polyion center, see Fig. 1.
To proceed, we assign to each monomer $`i`$ a lattice-gas variable $`\sigma _i`$, such that $`\sigma _i`$ is equal to $`1`$ if a counterion is condensed onto cite $`i`$ and $`0`$ otherwise. For a given configuration $`\{\sigma \}`$, the interaction Hamiltonian between the complex and a counterion located at a transverse distance $`d`$ from its center is
$``$ $`=`$ $`{\displaystyle \frac{1}{D}}{\displaystyle \underset{i=1}{\overset{Z}{}}}{\displaystyle \frac{\alpha q^2(\sigma _i\alpha 1)}{\sqrt{r_i^2+d^2}}}\mathrm{exp}(\kappa \sqrt{r_i^2+d^2})`$ (2)
$`+{\displaystyle \frac{1}{2D}}{\displaystyle \underset{i,i^{}=1,ii^{}}{\overset{Z}{}}}{\displaystyle \frac{q^2(\sigma _i\alpha 1)(\sigma _i^{}\alpha 1)}{b|ii^{}|}}\mathrm{exp}(\kappa b|ii^{}|),`$
where
$$r_i=\frac{2i1Z}{2}b.$$
It is convenient to define the reduced distance between the polyion and the counterion $`x=d/b`$, the reduced inverse Debye length $`k=\kappa b`$, and the Manning parameter Manning (1969, 1978) as $`\xi =q^2/ϵk_BTb`$. In terms of these adimensional variables the reduced Hamiltonian, $`H\beta /\xi `$, becomes
$`H`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{Z}{}}}(\sigma _i\alpha 1)[{\displaystyle \frac{2\alpha }{\sqrt{(2i1Z)^2+4x^2}}}\mathrm{exp}(k\sqrt{(2i1Z)^2+4x^2})`$ (3)
$`+{\displaystyle \frac{1}{2}}{\displaystyle \underset{i^{}=1,ii^{}}{\overset{Z}{}}}{\displaystyle \frac{\sigma _i^{}\alpha 1}{|ii^{}|}}\mathrm{exp}(k|ii^{}|)].`$
The partition function is a trace over all possible distributions of $`n`$ condensed counterions among the $`Z`$ polyion sites. There is a total of
$$N_c=\frac{Z!}{(Zn)!n!}$$
such configurations. The partition function is then
$$Q=\underset{\{\sigma \}}{}^{}\mathrm{exp}[\xi H],$$
where the sum is over the $`N_c`$ configurations $`\{\sigma \}`$ which obey the constraint $`_{i=1}^Z\sigma _i=n`$, denoted by the prime. It is convenient to order the terms in the Hamiltonian by the distances between the pair of interacting charges. This results in
$$H=\underset{i=1}{\overset{Z/2}{}}2\alpha [(\sigma _i+\sigma _{Zi+1})\alpha 2]\left[\frac{\mathrm{exp}(k\sqrt{(2i1Z)^2+4x^2)}}{\sqrt{(2i1Z)^2+4x^2}}\right]$$
$$+\underset{j=1}{\overset{Z1}{}}\underset{i=1}{\overset{Zj}{}}(\sigma _i\alpha 1)(\sigma _{i+j}\alpha 1)\frac{\mathrm{exp}(kj)}{j}.$$
If we now define the Boltzmann factors
$$x_j=\mathrm{exp}\left[\frac{\xi \mathrm{exp}(k\sqrt{(2j1Z)^2+4x^2})}{\sqrt{(2j1Z)^2+4x^2}}\right],$$
and
$$y_j=\mathrm{exp}\left[\frac{\xi \mathrm{exp}(kj)}{j}\right],$$
the contribution of each configuration to the partition function will be a product of these factors raised to exponents which are polynomials in $`\alpha `$, that is
$$Q=\underset{i=1}{\overset{N_c}{}}\underset{j=1}{\overset{Z/2}{}}x_j^{v_{i,j}}\underset{j=1}{\overset{Z1}{}}y_j^{u_{i,j}}.$$
(4)
The polynomials $`v_{i,j}=a_{i,j}\alpha +b_{i,j}\alpha ^2`$ and $`u_{i,j}=c_{i,j}d_{i,j}\alpha +e_{i,j}\alpha ^2`$, have integer non-negative coefficients. The advantage of the simple model constructed above is that for not too large values of $`Z`$ and $`n`$ the partition function can be evaluated exactly with a help of a computer.
The potential of mean force (measured in units of $`q^2/ϵb`$) between a polyion-counterion complex and an $`\alpha `$-ion located at $`x`$ is,
$$\varphi (\xi ,k,\alpha ,x)=\frac{1}{\xi }\mathrm{ln}\frac{Q(x)}{Q(\mathrm{})}.$$
(5)
The potential is normalized so that $`\varphi (\mathrm{})=0`$.
The computer code which generates the partition function for given values of $`Z`$ and $`n`$ determines the set of integer coefficients of the polynomials defined following the Eq.(4). Each set of polynomial coefficients may correspond to more than one internal configuration of the polyion, so that the degeneracy must also be taken into account. All the data is stored on the computer and used to perform a floating point calculation of the free energy.
## III Results and Discussion
In Fig. 2 the potential of mean force between various complexes and an $`\alpha `$-ion is plotted. The complexes are composed of a polyion of charge $`10q`$ and $`n`$ associated divalent counterions. Notice that for $`n=5`$ (neutral complex) the potential is a monotonically increasing function of $`x`$, so that the sixth counterion is always attracted to the complex.
For an overcharged complex with $`n=6`$ condensed counterions, the potential of mean force develops a barrier. At large distances the seventh counterion is repelled from the complex, while at short distances it is attracted to it. The minimum of the free energy, however, is reached when the seventh counterion is located at $`x=0`$. The potential of mean force, therefore, favors counterion condensation. The size of the barrier increases with $`n`$ and the minimum at $`x=0`$ becomes metastable for $`n=8`$. For $`n=9`$ the potential is a monotonically decreasing function of $`x`$, and the tenth counterion is always repelled from the complex. We next study the dependence of the depth of the potential well and the height of the barrier on the parameters of the model.
In Fig. 2, we saw that when the complex is overcharged $`n>Z/\alpha `$, the potential can have two minima, one located at $`x=\mathrm{}`$ and another at $`x=0`$. Which one of the two minima is the global one is determined by the sign of $`\varphi (0)`$. Figs. 3 and 4 show the behavior of $`\varphi (0)`$ as a function of $`\xi `$ and $`k`$. When $`\varphi (0)<0`$ the position at $`x=0`$ is the absolute minimum, while when $`\varphi (0)>0`$, $`x=0`$ is at most metastable. We should note, however, that the present discussion is not sufficient to define the absolute number of condensed counterions. For a counterion to be condensed the depth of the potential well must be sufficiently large, compared to the thermal energy $`k_BT`$, to prevent its rapid escape from the polyion surface. At the level of the present discussion this criterion is arbitrary. Thus, in this paper we will not consider the absolute number of condensed counterions but only the conditions which favor or disfavor the counterions condensation. From Figs. 3 and 4, we see that for a polyion of $`Z=10`$ and $`n=4`$ condensed trivalent counterions the minimum at $`x=0`$ is the global one for the parameters plotted. Approach of an additional fifth counterion to this already overcharged complex is, therefore, energetically favorable.
The depth of the global minimum $`|\varphi (0)|`$ is a monotonically increasing function of the Manning parameter, see Fig. 3. The dependence on the salt concentration, however, is not monotonic. From Fig. 4 we see that small concentrations of salt favor counterion condensation, i.e. $`\varphi (0)`$ becomes more negative for small $`k`$. Larger concentrations of salt, however, have a destabilizing effect on the counterion condensation.
This is even clearer for complexes composed of a polyion with $`Z=10`$ and $`n=5`$ condensed trivalent counterions. Fig. 5 shows that the position of the free energy minimum is a non-trivial function of salt concentration. Depending on the Manning parameter $`\xi `$ and the concentration of salt $`k`$, association of an additional, sixth, counterion can be either favored or disfavored. On the other hand, for $`Z=10`$, $`n=6`$, and $`\alpha =3`$, $`\varphi (0)`$ is always positive so that a complex with $`n=7`$ condensed counterions can be at most metastable.
We next explore the dependence of the barrier height $`\varphi (x_m)`$, where $`x_m`$ is the position of the maximum of the potential of mean force see Fig. 2, on the parameters of the model. In Fig. 6, $`\varphi (x_m)`$ is depicted as a function of the Manning parameter $`\xi `$ for polyion of size $`Z=10`$ with $`n=4`$ associated counterions. We see that the barrier height diminishes with the increase of $`\xi `$ and the amount of salt inside the suspension.
To explore the dependence of the barrier height on the size of the polyion $`Z`$, in Fig. 7 we plot $`\varphi (x_m)`$ as a function of $`Z`$ for complexes composed of a polyion and $`n^{}`$ condensed trivalent counterions, such that $`\varphi _n^{}(0)=0`$. While in the absence of salt the barrier height shows a significant dependence on the polyion size, at finite salt concentration this dependence weakens and $`\varphi (x_m)`$ seems to saturate when the polyion size is significantly larger than the Debye length. For large $`Z`$ and small concentration of electrolyte, however, the kinetic barrier can be many $`k_BT`$, providing a significant limitation to overcharging Levin and Arenzon (2003).
Charge reversal is a consequence of strong positional correlations between the counterions. These correlations are induced by the electrostatic repulsion between the particles. Thus, we expect that both the barrier height and the relative depth of the absolute minimum will be strongly dependent on the counterion valence. In Figs. 8 and 9 we show the dependence of the barrier height and the depth of the potential well on the valence of the counterions.
Although all the overcharged complexes depicted in Figs. 8 and 9 have the same net charge $`4q`$, the depth of the potential well and the height of the kinetic barrier depend on $`\alpha `$. As expected, larger counterion charge leads to stronger positional correlations and favors the counterion condensation and the charge reversal ($`\varphi (0)`$ becomes more negative with increasing $`\alpha `$). The barrier height, however, once again shows a nontrivial dependence on the salt concentration. For small amounts of salt and large $`Z`$, increased counterion valence leads to larger kinetic barriers.
## IV Conclusions
We have studied the potential of mean force between a polyion and an $`\alpha `$-valent counterion inside a polyelectrolyte solution containing multivalent counterions and monovalent salt. The model is sufficiently simple that the partition function can be calculated exactly. It is found that for an overcharged polyion the potential of mean force can have two minima, one located at $`x=0`$ and another $`x=\mathrm{}`$. Which one of the minima is the global one depends on the charge density of the polyion and the amount of salt inside the suspension. When the global minimum is at $`x=0`$, a counterion from the bulk finds it energetically favorable to approach the polyion surface. To reach $`x=0`$, however, the counterion must overcome a free energy barrier. For small salt concentrations, this barrier can be sufficiently large to provide a kinetic limitation to the extent of charge reversal. Furthermore, even if the counterion reaches $`x=0`$, whether or not it will become condensed will depend on the depth of the potential well. Counterion condensation will occur only if $`\varphi (x_m)\varphi (0)1/\xi `$. Otherwise, the thermal fluctuations will lead to a fast escape of the counterion from the $`x=0`$ minimum.
For suspensions containing rodlike polyelectrolytes and the multivalent counterions micro phase separation is observed under certain conditions Tang and Janmey (1996); Bruinsma (2001). The polyions aggregate forming bundles with a well defined crossectional area. It has been argued that bundle formation is an activated process and the size of the bundles is kinetically controlled Shklovskii (1999); Ha and Liu (1999); Stilck et al. (2002). It should then be quite interesting to explore the dependence of the barrier height on the concentration of monovalent electrolyte using a theory similar to the one presented above.
To conclude, the extent of the charge reversal is strongly dependent on the amount of monovalent salt present in the suspension. Small concentrations of salt will enhance the overcharging while an excessive amount of salt will hinder the charge reversal. Furthermore, even if the minimum of the free energy corresponds to an overcharged state, we find that depending on the polyion charge density and the amount of salt in the suspension, there can be significant kinetic limitations to the overcharging.
This work was supported in part by the Brazilian agencies CNPq and FAPERJ. JFS acknowledges funding by project Pronex-CNPq-FAPERJ/171.168-2003.
|
warning/0506/physics0506104.html
|
ar5iv
|
text
|
# Long range scattering resonances in strong-field seeking states of polar molecules
## I Introduction
Conventional spectroscopy of atoms and molecules begins with the premise that the energy levels of the species being probed are fixed although they may shift slightly in electromagnetic fields. These levels are then interrogated by energy-dependent probes such as photons or charged or neutral particles. Information on the energetics and structure of the molecules is extracted from absorption energies, oscillator strengths, selection rules, etc. In these investigations, the study of resonances has played a central role.
With the advent of ultracold environments for atoms and molecules, this general view of spectroscopy can be inverted. Cold collisions provide a nearly monochromatic probe of near-threshold intermolecular interactions, with resolution set by the milliKelvin or microKelvin temperature of the gas. In this case the energy levels of nearby resonant states can be tuned into resonance with the zero-energy collisions. For cold gases of alkali atoms, this strategy is already in widespread use. It exploits the fact that the Zeeman effect can shift the internal energies of the atoms over ranges orders of magnitude larger than the collision energy itself. In this way, atoms can be made to resonate when they would not naturally do so (i.e., in zero field). Measurement of these “Feshbach” resonances (more properly, Fano-Feshbach resonances) has yielded the most accurate determination of alkali-alkali potential energy surfaces for near-threshold processes potentials .
The current experimental and theoretical push to create and study ultracold molecules cold-rev will lead to many more opportunities for this novel kind of spectroscopy, since molecules possess many more internal degrees of freedom than do atoms. There will be, for example, numerous Fano-Feshbach resonances in which one or both of the collision partners becomes vibrationally or rotationally excited Forrey ; Deskevich ; Bala ; statistical arguments suggest that these resonances will be quite narrow in energy, a fact related to their abundance Deskevich . A second class of resonant states will occur when the constituents are excited into higher-lying fine structure or hyperfine structure states, more reminiscent of the resonances observed in the alkali atoms. In molecules, these resonances are naturally also tunable in energy using magnetic fields krems .
In this paper, we are primarily interested in a third class of resonance: potential resonances that are engendered by altering the intermolecular potential energy surface itself. This capability becomes especially prominent in cold collisions of heteronuclear polar molecules, whose dipolar interactions are quite strong on the scale of the low translational temperatures of the gas. At the same time, the dipole moments of the individual dipoles can be strengthened or weakened as well as aligned by an applied electric field.
This kind of resonance includes the shape resonances, discussed in the context of cold atoms polarized by strong electric fields shape-you ; Deb . A second set, dubbed “field-linked” resonances, has been studied in some detail in Refs AA\_PRL ; AA\_PRA ; AA\_FL . These states appear in potential energy surfaces (PES’s) that correlate to weak electric field-seeking states of the free molecules. They are weakly bound, long-ranged in character, and indeed do not appear to exist without an electric field present. Similar resonances are predicted to occur in metastable states of the alkaline earth elements at low temperatures derev ; santra .
A rich set of potential resonances emerges among strong-field seeking states, and this is the subject of the present paper. Strong-field seekers are of increasing importance experimentally, since they would enable molecules to be trapped in their absolute ground states where no 2-body inelastic collision processes are available to harm the gas. In this case, the colliding molecules are free to approach to within a small internuclear distance of one another; the resulting potential resonances therefore can probe detailed intermolecular dynamics near threshold. The resulting data, consisting of scattering peaks as a function of electric field, can be thought of as a kind of “Stark spectroscopy”. Just such a tool has been applied previously in precision measurements of alkali Rydberg spectra Ryd1 ; Ryd2 .
In this paper we explore such spectra in ultracold molecules, finding that the spectrum is dominated by a quasi-regular series in the electric field values. Such a series is the fundamental building block of molecular Stark spectroscopy and plays a role analogous to the Rydberg series in atomic spectroscopy. In both cases, the series lays out the fundamental structure of the unperturbed, long-range physics between interacting entities. In the case of atoms, the deviation from an unperturbed Rydberg series, encapsulated in the quantum defects, yields information on the electron-core interaction Seaton ; qdt . Similarly, it is expected that differences in observed Stark spectra from those presented here will probe the short-range intermolecular interactions.
To emphasize the generality of these resonances, we first briefly develop in Sec. II a general formalism for arbitrary polar collision partners, covering explicitly Hund’s cases (a) and (b) as well as asymmetric rotor molecules. Using this formalism, we introduce in Sec. III the potential resonances using a simplified version of the molecular gas in which all dipoles are assumed to be perfectly aligned and where molecular fine structure plays no role. In Sec. IV, we consider the case of more realistic molecules, where fine structure does intervene. We show that the structure of the potential resonances is unchanged, but that additional narrow Fano-Feshbach resonances do appear. Throughout the article we emphasize how the various types of resonance can be classified and organized by simple considerations involving the WKB approximation.
## II General Form of the Stark Effect and dipolar interaction
In this section, we briefly recapitulate our scattering model which is presented in Refs.AA\_PRA ; CT1 ; CT\_thesis . The molecular scattering Hamiltonian can be written as
$$H=T_0+V_{SR}+V_{LR}+H_S+H_{int}$$
(1)
Where $`T_0`$ is the kinetic energy operator, $`V_{SR}`$ is the short range potential energy surface (PES), $`V_{LR}`$ is the long range interaction of the polar molecules, $`H_S`$ is the Stark Hamiltonian, and $`H_{int}`$ is the Hamiltonian describing the internal degrees of freedom of the molecules. For the purpose of this paper we disregard $`V_{SR}`$. The long range interaction is dominated by the dipole-dipole interaction AA\_PRA . Furthermore an electric field can significantly change the structure of the interaction and be used to control the molecular collisions AA\_PRA ; CT1 .
The first step in performing a scattering calculation is to construct a basis of molecular energy eigenstates in an external field. Here we generalize the approach slightly to make it applicable to various types of polar molecules. Once the Stark Hamiltonian is obtained, the result can be used in the Hamiltonian of the dipolar interaction, showing their common physical origin in terms of electric fields.
To simplify the analysis and calculations, we assume that the vibrational degrees of freedom are frozen out at low temperatures; hence can treat the molecules as rigid rotors. The molecular state is described in terms of the state $`|JM_J\mathrm{\Omega }`$, where $`J`$ is the molecule’s rotational plus electronic angular momenta, $`M_J`$ is the projection of $`J`$ onto the lab axis, and $`\mathrm{\Omega }`$ is $`J`$’s projection onto the molecular axis. To describe the rigid rotor molecular wave function, we use $`\alpha ,\beta ,\gamma |JM_J\mathrm{\Omega }=\sqrt{\frac{2J+1}{8\pi ^2}}D_{M_J\mathrm{\Omega }}^J(\alpha ,\beta ,\gamma )`$, where $`\alpha ,\beta ,`$ and $`\gamma `$ are the Euler angles defining the molecular axis and $`D_{M_J\mathrm{\Omega }}^J`$ is a Wigner D-function Brown .
### II.1 The Stark effect
Since the electric field is a true vector (as opposed to a pseudovector), it only couples states of opposite parity. This implies that the Stark energies vary quadratically with low electric field; they vary linearly only at higher fields once the Stark energy is greater than the energy splitting of the two states. The Stark Hamiltonian has the form
$$H_S=\stackrel{}{\mu }\stackrel{}{}.$$
(2)
For the current discussion, we pursue a general approach and allow the field to point in any direction in the laboratory reference frame. This general approach allows the matrix elements to be used in the dipolar interaction. However when explicitly considering the molecular states in an electric field, we that the field lies in the $`\widehat{z}`$ direction.
The Stark interaction can be evaluated by decomposing the electric field in spherical coordinates and then rotating the dipole into the lab frame. In this representation, the Stark Hamiltonian has the form $`H_S=_q\mu D_{q0}^1`$, where $`\mu `$ is the electric dipole moment of the molecule and $`D^1`$ is a Wigner D-function. $`D_{q0}^1`$ is equal to $`(1)^{2q}C_q^1`$, a reduced spherical harmonic Brink , and $`q`$ is the projection quantum number of $``$ onto the lab axis. At the heart of evaluating the Stark effect, we find the operator $`D^1`$ coupling two molecular states, which themselves are described by D functions. The Stark matrix element is an integral of three D-functions over the molecular coordinates, averaging over molecular orientation. The evaluation of the integral results in selection rules whose details depend on the molecular specifics. For a valuable qualitative discussion on the Stark effect see Schreel .
A given molecule may also have a nuclear spin that generates a hyperfine structure. In this case, it is more appropriate to present the matrix elements in the hyperfine basis, where $`F`$ and $`M_F`$ define the state. Here $`F`$ is the sum of $`J`$ and the nuclear spin $`I`$ and $`M_F`$ is the projection of $`F`$ in the lab frame. We use the Wigner-Eckart theorem to compute the Stark matrix elements in a compact form. The matrix elements of the Stark effect are written as
$`\alpha FM_F|H_S|\alpha ^{}F^{}M_F^{}=\mu \alpha FM_F|D_{q0}^1|F^{}M_F^{}\alpha ^{},`$ (3)
which contains a purely geometrical matrix element
$`\alpha FM_F|D_{q0}^1|F^{}M_F^{}\alpha ^{}=[F](1)^{1+M_F+F^{}}`$
$`\times \left(\begin{array}{ccc}F^{}& 1& F\\ M_F^{}& M_FM_F^{}& M_F\end{array}\right)\alpha FD_0^1\alpha ^{}F^{}.`$ (6)
Here $`\alpha FD_0^1\alpha ^{}F`$ is the reduced matrix element and $`\alpha `$ represents all remaining quantum numbers needed to uniquely determine the quantum state. $`[j]`$ is a shorthand notation for $`\sqrt{2j+1}`$. We have left $`q`$ in the matrix element simply as a place holder. Its value is assumed to be $`q=M_FM_F^{}`$ in accordance with conservation of angular momentum. If we consider the case where the electric field points in the $`\widehat{z}`$ direction, then $`q=0`$, and we find that $`M_F=M_F^{}`$. This means that $`M_F`$ is a conserved quantum number. We have used the convention of Brink and Satchler to define the Wigner-Eckart theorem and reduced matrix elements Brink .
### II.2 Molecular examples
We present a few examples of reduced matrix elements for specific molecular symmetries. First, consider a Hund’s case (a) molecule with $`\mathrm{\Omega }0`$. The OH radical, with ground state $`{}_{}{}^{2}\mathrm{\Pi }_{3/2}^{}`$, is a good example of this. The energy eigenstates of such a molecule in zero electric field are eigenstates of parity, i.e., $`|JM_J\overline{\mathrm{\Omega }}ϵ`$ where $`ϵ=+()`$ represents the $`e`$ ($`f`$) parity state and $`\overline{\mathrm{\Omega }}=|\mathrm{\Omega }|`$. The parity of this molecule is $`ϵ(1)^{J1/2}`$ if $`J`$ is a half integer or $`ϵ(1)^J`$ if $`J`$ is an integer. (For details see Refs. Schreel ; CT1 ; CT\_thesis .) Taking this molecular structure into account, we find that the reduced matrix element is
$`\alpha FD_0^1F^{}\alpha ^{}=()^{1+I+F+J+J^{}\overline{\mathrm{\Omega }}}[F^{},J,J^{}]`$
$`\times \left\{\begin{array}{ccc}F& F^{}& 1\\ J^{}& J& I\end{array}\right\}\left(\begin{array}{ccc}J^{}& 1& J\\ \overline{\mathrm{\Omega }}& 0& \overline{\mathrm{\Omega }}\end{array}\right)`$ (11)
$`\times \left({\displaystyle \frac{1+ϵϵ^{}(1)^{J+J^{}+2\overline{\mathrm{\Omega }}+1}}{2}}\right).`$ (12)
Here the index $`\alpha `$ represents the set of quantum numbers $`ϵ,\overline{\mathrm{\Omega }}`$ and $`J`$. We have introduced another notation: $`[j_1,j_2,\mathrm{},j_N]=`$ $`\sqrt{(2j_1+1)(2j_2+1)\mathrm{}(2j_N+1)}`$. This reduced matrix element is the same for any case (a) molecule.
On the other hand consider a Hund’s case (b) molecule with $`L=0`$. Many molecules fit this mold such as heteronuclear alkali dimers and SrO with $`{}_{}{}^{2S+1}\mathrm{\Sigma }`$ ground states. Here the parity of a state is directly identified by the value of $`J`$, where $`parity=(1)^J`$. The Stark effect therefore mixes the ground state with the next rotational state. We find the reduced matrix element to be
$`\alpha FD_0^1F^{}\alpha ^{}=(1)^{F+I+J+J^{}+S+N+N^{}}`$
$`\times [N,N^{},J,J^{},F^{}]\left\{\begin{array}{ccc}F& F^{}& 1\\ J^{}& J& I\end{array}\right\}`$ (15)
$`\times \left\{\begin{array}{ccc}J& J^{}& 1\\ N^{}& N& S\end{array}\right\}\left(\begin{array}{ccc}N^{}& 1& N\\ 0& 0& 0\end{array}\right)`$ (20)
where the index $`\alpha `$ represents the set of quantum numbers $`N`$ and $`S`$.
This formalism is easily extended to include asymmetric rotors by including the rotational Hamiltonian to construct molecular eigenstates. For an asymmetric rotor, there are three distinct moments of inertia and therefore three distinct rotational constants. The rotational Hamiltonian is $`H_{rot}=A𝐉_a^2+B𝐉_b^2+C𝐉_c^2`$, where $`A>B>C`$ and $`a,b`$ and $`c`$ are the axis labels in the molecular frame. This additional structure mixes $`\mathrm{\Omega }`$ such that it is not a good quantum number, implying we that need to diagonalize the rotational Hamiltonian along with the Stark Hamiltonian to obtain the molecular eigenstates. We consider the primary effect of this additional structure to change the progression of rotationally excited states. For a complete discussion on asymmetric rotors see Refs. Hain ; tinkham .
### II.3 Dipole-dipole interaction
The intrigue of polar molecules is their long range anisotropic scattering properties whose origin is the dipole-dipole interaction. The interaction in vector form is
$$H_{\mu \mu }=\frac{3(\widehat{𝐑}\widehat{\mu }_1)(\widehat{𝐑}\widehat{\mu }_2)\widehat{\mu }_1\widehat{\mu }_2}{R^3},$$
(21)
where $`\widehat{\mu }_i`$ is the electric dipole of molecule $`i`$, $`R`$ is the intermolecular separation, and $`\widehat{𝐑}`$ is the unit vector defining the intermolecular axis. This interaction is conveniently rewritten in terms of tensorial operators in the laboratory frame as:
$$H_{\mu \mu }=\frac{\sqrt{6}}{R^3}\underset{q}{}(1)^qC_q^2(\mu _1\mu _2)_q^2.$$
(22)
Here $`C_q^2(\theta ,\varphi )`$ is a reduced spherical harmonic that acts on the relative angular coordinate of the molecules, while $`(\mu _1\mu _2)_q^2`$ is the second rank tensor formed from the two rank one operators determining the individual dipoles. For this reason, matrix elements of the interaction are also given conveniently in terms of the matrix elements in Eq. (II.1). The matrix elements of the dipolar interaction are:
$`12lm_l|H_{\mu \mu }|1^{}2^{}l^{}m_l^{}=(1)^{m_l+m_l^{}+1}\left({\displaystyle \frac{\mu ^2\sqrt{6}}{R^3}}\right)`$
$`\times lm_l|C_{(m_lm_l^{})}^2|l^{}m_l^{}\alpha _1F_1M_{F_1}|D_{q0}^1|\alpha _1^{}F_1^{}M_{F_1}^{}`$
$`\times \alpha _2F_2M_{F_2}|D_{q^{}0}^1|\alpha _2^{}F_2^{}M_{F_2}^{}`$
$`\times \left(\begin{array}{ccc}1& 1& 2\\ M_{F_1}M_{F_1}^{}& M_{F_2}M_{F_2}^{}& m_lm_l^{}\end{array}\right)`$ (25)
where
$`lm_l|C_{(m_lm_l^{})}^2|l^{}m_l^{}=(1)^{m_l}[l,l^{}]\left(\begin{array}{ccc}l& 2& l^{}\\ 0& 0& 0\end{array}\right)`$ (28)
$`\times \left(\begin{array}{ccc}l& 2& l^{}\\ m_l& m_lm_l^{}& m_l^{}\end{array}\right).`$ (31)
Equation (25) shows that once the Stark Hamiltonian has been constructed for a particular molecule, then the Hamiltonian describing the dipolar interaction can be achieved with little extra effort. This result reflects the fact that in Eq. (25) each dipole is acted on by the electric field of the other. We have also used a shorthand to represent the channel, $`|12lm_l`$, where $`|1`$ denotes the quantum state of the first molecule and as $`|2`$ for second molecule. At this point, we disregard all interactions between molecules except the dipolar interaction.
To exploit an analogy with Rydberg atoms, the long range Coulombic interaction is well understood. With this understanding of the long range physics, a solution to the complete problem is achieved by matching it to the short range solution or a parameterization of the short range interaction qdt . This idea was, in fact, implemented as a numerical tool in Ref.Deb , which dealt with collisions of atoms in strong electric fields. To this end, we pursue an understanding of the long range characteristics of the dipolar scattering. We then envision merging the long range physics with the short range physics, or a parametrization thereof, to offer insight into the short range interaction and to explore the dynamic interaction of polar molecules. With this idea in mind we first explore “pure” dipolar scattering.
## III Dipolar Scattering
Our primary interest in polar molecule scattering is how the strong anisotropic interaction affects the system. As a first step illustrating the influence of dipolar interactions, we present a simple model composed of polarized dipoles with no internal structure. Strictly speaking, this system is created by an infinitely strong electric field that completely polarizes the molecules and raises all other internal states to experimentally unattainable energies. Thus the only label required for a channel is its partial wave, $`l=\{0,2,4,\mathrm{}\}`$ in the numerical example given here. The matrix elements of the dipole-dipole interaction are taken to be $`12l0|H_{\mu \mu }|12l^{}0=0.32\mu ^2l0|C_0^2|l^{}0`$, which is typical for molecules like RbCs or SrO. We then artificially vary the dipole moment $`\mu `$ to see the effect of an increasingly strong dipolar interaction. Pragmatically speaking, varying $`\mu `$ parallels changing the electric field. The intention of this model is to focus on the effect of direct anisotropic couplings between the degenerate channels, as measured by their effect on the partial wave channels.
For this model, we use a reduced mass of $`m_r=10^4`$ a.u., typical of very light molecules. We moreover assume that the molecules approach one another along the laboratory z-axis, so that only the projection $`m_l=0`$ of orbital angular momentum is relevant. To set a concrete boundary condition at small $`R`$, we apply “hard wall” boundary conditions, $`\psi (R_{in})=0`$ at a characteristic radius $`R_{in}=20`$ a.u. (In Sec. V we will relax this restriction.) We pick the collision energy to be nearly zero, namely $`10^{12}`$ K. To converge the calculation for this model requires inclusion of partial waves up to $`l=14`$, and numerical integration of the Schrödinger equation out to $`R=R_{\mathrm{}}=1\times 10^5`$ a.u., using the log-derivative propagator method of Johnson Johnson .
To get a sense of the influence of increasing the dipole moment, we first look at adiabatic curves of the system. Figure 1 (a) shows two different sets of adiabatic curves: a gray set with $`\mu =0.1`$ (a.u.) and a black set $`\mu =1.0`$ (a.u.). In each set, the four lowest adiabatic curves are shown. Looking at these curves, we can see two characteristic effects of increasing $`\mu `$. First, the lowest curve becomes much deeper. Second, the higher adiabatic curves, originating form non-zero partial waves, may support bound states at short distance states, i.e. within the centrifugal barrier. Both these effect may generate bound states, leading to distinct classes of scattering resonances as $`\mu `$ is varied. The deepening of the lowest adiabatic curves induces potential resonances, whereas the higher-lying curves lead to narrow shape resonances, wherein the molecules must tunnel through the centrifugal barrier.
The different classes of resonances can clearly be seen in cross section, as shown in Fig. 1 (b). The broad quasi-regular set of resonances seen in the cross section are the potential resonances originating from the lowest adiabatic curve. The narrow shape resonances appearing sporadically in the spectrum originate from the higher-lying curves. For the purpose of this paper, we focus on the wide potential resonances and simply acknowledge the existence of the narrow shape resonances.
To show that the broad resonances primarily belong to the lowest potential and the narrow shape resonances belong to the higher-lying potentials, we use an eigenphase analysis. The eigenphase can be thought of as the sum of the phase shifts for all of the channels; thus it tracks the behavior of all the channels simultaneously. The eigenphase is defined as
$$\varphi _{eigen}=\underset{i}{}\mathrm{tan}^1(\lambda _i^K).$$
(32)
Here $`\lambda _i^K`$ are the eigenvalues of the $`K`$ matrix from the full-scattering calculation Taylor . (The $`K`$ matrix is related to the more familiar scattering matrix by $`S=(1+iK)/(1iK)`$.) When the system gains a bound state, it appears as a $`\pi `$ jump in eigenphase. The eigenphase of the system is shown in Fig. 2 (a), as the solid line with many abrupt steps.
To analyze this situation further, we construct an approximate eigenphase as follows. First, we asses the total phase accumulated in each adiabatic channel, using a WKB prescription:
$$\varphi _{WKB}^{(i)}(\mu )=_{R_{in}}^{R_{out}}\sqrt{2m_rV_{AD}^{(i)}(\mu ,R)/\mathrm{}^2}.$$
Here $`(i)`$ stands for the $`i^{th}`$ adiabatic curve. For the lowest adiabatic curve, which is always attractive, the range of integration is from $`R_{in}`$ to $`R_{out}=\mathrm{}`$. For high-lying channels that possess a barrier to scattering at zero collision energy, the limits of integration are from $`R_{in}`$ to the inner classical turning point of the barrier. This will yield some information on shape resonances trapped behind the barrier, but we will not make much of this in the analysis to follow. Finally, we add together the individual WKB phases to produce an approximate eigenphase shift, which we dub the “adiabatic WKB phase” (AWP):
$$\varphi _{WKB}(\mu )=\underset{i}{}\varphi _{WKB}^{(i)}(\mu ),$$
(33)
Since we are not concerned with properties of the phase associated with the higher-lying adiabatic curves, we do not consider the connection formula now.
The total AWP for this system is shown as a dashed line in Fig. 2 (a). It tracks the eigenphase well but offers more information if we decompose the AWP into its contributions. Figure 2 (b) shows the individual contributions of the sum. The largest contribution is the phase from the lowest adiabatic curve, which can be associated with potential resonances. A black bracket appears above this phase contribution with vertical marks indicating when it passes through an integer multiple of $`\pi `$, i.e., when we expect to see a potential resonance in the cross section. This same bracket is plotted in Fig. 1 and shows good agreement between the locations of the potential resonances and the AWP predictions. We conclude from this that the main resonance features in the Stark spectrum arise primarily from this single potential curve.
In Fig. 2 (b) the two remaining phase contributions originate from non-zero partial wave channels that possess centrifugal barriers; see Fig. 1(b). The gray bracket indicates where the sum of these two contributions pass through an integer multiple of $`\pi `$, and thus represent a guess for where the shape resonances lie. This gray bracket is also shown in Fig. 1. The agreement with the position of the narrow resonance features in the cross section is not nearly as good as it is with the broad potential resonances. This indicates a more involved criterion for shape resonances. Nevertheless, the AWP predicts 15 shape resonances, and there are 13 in the range of $`\mu `$ shown. The AWP appears to offer a means to roughly predict the number of shape resonances in this system, even though it does not predict the locations exactly.
A main point of this analysis is that the AWP in the lowest adiabatic channel alone is sufficient to locate the potential (as opposed to shape) resonances, without further modification. For the rest of this paper, we focus on the potential resonances in more realistic molecules with internal molecular structure. The general analysis in terms of a single-channel AWP will still hold, but an additional phase shift will be required to describe the spectrum.
## IV Strong Field Seekers
Strong-field-seeking molecules are approximately described as polarized in the sense of the last section, because their dipole moments are aligned with the field. They will, however, contain a richer resonance structure owing to the presence of low-lying excited states that can alter the dipolar potential energy surface at small-$`R`$.
For concreteness, we focus here on molecules with a $`{}_{}{}^{1}\mathrm{\Sigma }`$ ground state. Heteronulcear alkalis fit into this category and are rapidly approaching ground state production with various species alkali . As examples, we pick RbCs and SrO in their ground states. Ground state RbCs has been produced experimentally Sage . As for SrO, promising new techniques should lead to experimental results soon Demille . For simplicity we include only the $`J=0`$, $`J=1`$ rotational states, and freeze the projection of molecular angular momentum to $`M_J=0`$. This restricts the number of scattering thresholds to three, identified by the parity quantum number of the molecules in zero field. The parity quantum numbers for the three thresholds are ($`,+,++`$). This model is similar to the one presented in Ref. AA\_PRA which can be easily constructed for any rigid rotor when only including two molecular states. One immediate consequence of multiple thresholds is the presence of rotational Fano-Feshbach resonances in the collisional spectrum.
The first example is RbCs, whose physical parameters are $`\mu =1.3`$ D, $`m_r=110`$ a.m.u., and $`B_e=0.0245(K)`$ Sage . As before, we apply vanishing boundary condition at $`R_{in}=20`$ a.u. To converge this calculation over the field range considered, we require partial waves up to $`l=30`$. We first look at the adiabatic curves of the system to get an understanding of how the real system deviates from the simple model presented above. In Fig. 3 (a) we plot the 6 lowest adiabatic curves for the RbCs system with only four partial waves, so the figure is more easily interpreted. The sets of adiabatic curves shown are for two different field values: the gray set has $``$=0 and the black set has $``$= 5000 (V/cm). There are two important features that differ from the dipole example. First, there are two higher thresholds, and the electric field shifts these apart in energy as the field is increased. Second,the electric field dramatically changes the radial dependence of the Hamiltonian.
The difference in thresholds can be seen clearly in Fig. 3 (a) where the lowest excited threshold moves from $`0.05(K)`$ at $``$=0 to $`0.18(K)`$ at $``$=5000 (V/cm). The difference in radial dependences for the two cases is seen more clearly in a log-log plot of the two lowest adiabatic curves for both fields, as shown in Fig. 3 (b). The gray set corresponding to zero field has two distinct asymptotic radial powers. At large $`R`$ the lowest adiabatic curve has a $`1/R^6`$ behavior asymptotically because of couplings with channels far away in energy \[$`0.05(K)`$\]. However as $`R`$ approaches zero the dipolar interaction has overwhelmed the rotational energy separation and the radial dependence becomes $`1/R^3`$ in character at about R=100 (a.u.). For reference, the dashed line is proportional to $`1/R^3`$. With $``$= 5000 (V/cm), the two black curves show the radial dependence of the adiabatic curves. The lowest curve now has nearly $`1/R^3`$ over the whole range shown. Asymptotically when the centrifugal barrier is larger, the radial dependence will change to $`1/R^4`$ AA\_PRA . The second lowest adiabatic curve is also significantly altered by the strong dipolar interaction as can be seen in Fig. 3 (b).
In Fig. 4 (a), we plot the cross section for the model RbCs system. This spectrum is riddled with narrow Fano-Feshbach resonances, but is still dominated by a series of potential resonances similar to the one in Fig. 1 (b). There are two sets of AWP predictions shown as over-brackets. To understand their difference, we look to Fig. 4 (b). The AWP for the lowest adiabatic curve is shown in Fig. 4 (b) for two different zero-field phase values. The gray curve is the AWP that is directly computed from the method described above, Eq. (33). The locations where it passes through an integer multiple of $`\pi `$ are indicated by the gray triangles. Referring back to Fig. 4 (a), where the same gray triangles appear we see that this simple estimate does not reproduce the resonance position.
We can, however, introduce an additional overall phase shift to account for the difference in short-range interactions from the pure polarized case. The shifted AWP reads
$$\stackrel{~}{\varphi }_{WKB}^{(1)}()=\varphi _{WKB}^{(1)}()+\pi \delta _{defect}.$$
(34)
By treating $`\delta _{defect}`$ as a fitting parameter, we can obtain the resonance positions indicated by the black bracket in Fig. 4, which agree quite well with the resonance positions in the close-coupled calculation. To do so requires, in this case, a phase shift $`\delta _{defect}=0.14`$. In analogy with Rydberg spectroscopy we consider the shift we have added to be a “quantum defect” that accounts for the effect of the short-range interaction. The additional phase shift reflects the influence of short range physics on the scattering, such as curve crossings with curves from higher thresholds.
The AWP also saturates with field, as can be seen in Fig. 4 (b). This occurs because the electric field eventually fully polarizes the molecules, so the dipole moment cannot increase further. The effect can also be seen in the spacing of the potential resonances. At low fields, the potential resonances occur frequently in field. Then, as the field is further increased, the resonances occur less often in field, which is a signature of dipole moment saturating and therefore an increasing field having less effect on the molecular interaction.
As a second example, we consider SrO, which has the physical parameters $`m_r`$=52 a.m.u, $`\mu `$=8.9 D, and $`B_e=`$ 0.5 K Demille . We choose this molecule for its comparatively large mass and dipole moment, which guarantee a large number of resonances. Fig. 5 we have plotted the cross section for SrO, which is dominated by the quasi-regular potential resonance series. As before, the black bracket indicates where the phase shifted AWP predicts the potential resonances, and we see the agreement is good. Furthermore, the series has not terminated since we have not completely polarized the dipole. The series of potential resonances saturates at 17.5 (kV/cm) after a total of 43 potential resonances have been induced. To line up the AWP’s predictions and the actual cross section requires a defect of $`\delta _{defect}=0.215`$.
We have picked two examples to illustrate how the potential resonances will appear in the context of collisional spectroscopy. These resonance will occur to varying degrees in the strong-field seeking collisions of all polar molecules. For example, we can also make similar predictions for an asymmetric rotor molecule such as formaldehyde $`(H_2CO)`$. We find that this molecule should possess six potential resonances in the field interval from 0 to 50 $`(kV/cm)`$.
It is worth noting that portions of similar resonance series was anticipated in cold atomic gases subjected to electric fields Deb . However very few such resonances are likely to be observed, owning to the enormous fields ($`MV/cm`$) required to generate them. In polar molecules, by contrast, the entire series should be readily observable.
## V Collisional Spectroscopy
Through the course of this work we have shown that the zero-energy cross section of weak-field-seeking molecules is dominated by a set of broad potential resonances. Even though these resonances are themselves intriguing, their properties can be exploited to learn much more about the system. The general structure of the potential resonances is governed by the long range dipolar interaction, which has a predictable and common form. With a clear understanding of this interaction and how it induces resonances, it could be exploited to learn about the short range interaction of the molecules. This is because details of where the lines appear must also depend on the boundary condition experienced by the wave function at small values of $`R`$. Therefore the spectrum contains information on the small $`R`$ intermolecular dynamics. Thus by studying the potential resonance series we can extract information about the short range dynamics.
This idea is similar to quantum defect theory, which, for example, has been very successful in Rydberg spectroscopy. The short range physics of the electron interacting with the nucleus is complicated and not easily solved. However once the electron is out of the small $`R`$ region, it enters into a pure coulomb potential where the motion of the electron is well understood. The effect of the short range must be merged with the long range physics to form a complete solution. To account for the short range interaction, the energy can be parameterized by replacing the principal quantum number by an effective quantum number $`n^{}=n\mu `$. This procedure is tantamount to identifying an additional phase shift due to the interaction of the electron with the atomic core. The idea of merging standard long range physics with complicated short range behavior has been applied successfully not only in Rydberg states of atoms qdt ; Fano and molecules Jungen , but also in atomic collisions Mies , cold collisions Greene2 ; Mies2 ; Gao , and dipole-dipole interactions of the type we envision here Deb .
As a simple expression of this idea, we can alter the boundary condition applied at $`R_{in}`$ when performing the full scattering calculation and note its influence on the field dependent spectrum. For the above calculations we have imposed the standard vanishing boundary condition, $`\psi (R_{in})=0`$ for all channels. We now replace this condition with a uniform logarithmic derivative, $`b=(\frac{d}{dR}\psi )\psi ^1`$, at $`R_{in}`$. Thus previously we set $`b=\mathrm{}`$, but now we allow $`b`$ to vary. The log-derivative is conveniently represented as a phase:
$$b=\mathrm{cotan}(\pi \beta ),$$
(35)
where $`\beta `$ can lie between zero and one, covering all values of $`b`$ from $`\mathrm{}`$ to $`+\mathrm{}`$. For $`\beta =0`$, the boundary condition is the one employed above, $`\psi (R_{in})=0`$, whereas for $`\beta =0.5`$, the boundary condition is $`\frac{d}{dR}\psi (R_{in})=0`$.
We have re-computed the collisional spectrum of RbCs for several different initial conditions, and plotted the field values of the potential resonances, $`^{(b)}`$, in Fig. 6 as sets of points. The values of $`\beta `$ for the different calculations are $`0`$ (filled circle), $`0.11`$ (filled square), $`0.22`$ (filled triangle), $`0.56`$ (hollow circle), $`0.78`$ (hollow square), and $`0.89`$ (hollow triangle). The filled circles are resonant locations for cross section in Fig. 4 (a).
We next wish to demonstrate that each such spectrum can be identified by a single quantum defect parameter, as was done in the previous section. This entails picking a value of $`\delta _{defect}`$ and then setting the phase $`\stackrel{~}{\varphi }_{WKB}^{(1)}()`$ equal to an integer multiple of $`\pi `$. This yields a set of resonant field values, $`_{WKB}(\delta _{defect})`$. Each $`\delta _{defect}`$ corresponds to a particular approximate spectrum. The set of curves, $`_{WKB}(\delta _{defect})`$, generated by continuously varying $`\delta _{defect}`$ are shown are shown in Fig. 6 as solid lines. The bracket in Fig. 4 (a) corresponds to the set of points where a vertical line intersects $`_{WKB}(\delta _{defect})`$ with $`\delta _{defect}=0.14`$.
We can compare the the resonant field values predicted by the AWP, $`_{WKB}(\delta _{defect})`$, and resonant field values given by the full calculation with different initial conditions, $`^{(b)}`$. To plot $`^{(b)}`$ we have varied the height at which the set of points $`^{(b)}`$ is plotted until it aligns with $`_{WKB}(\delta _{defect})`$. Doing this we are able to to see how $`b`$ and $`\delta _{defect}`$ are related. Thus in fig. 6 we can see that even with different boundary conditions, the single AWP curve in fig 4 (b) can be used to predict the spacing between the potential resonances by varying a single parameter, $`\delta _{defect}`$. This shows the AWP represents the long range scattering physics well, and that empirically extracted parameters like $`\delta _{defect}`$ will carry information about the short-range physics such as that embodied in $`b`$.
## VI Conclusion
A number of resonant processes may occur when two polar molecules meet in an ultracold gas. We have focused here on the dominant, quasi-regular, series of potential resonances between weak-field seeking states. These potential resonances originate in the direct deformation of the potential energy surface upon which the molecules scatter. Observation of these resonances may offer a direct means for probing the short range interaction between molecules. We have provided a means of analyzing this system with an adiabatic WKB phase integral. This method shows how the system evolves with the application of an electric field.
###### Acknowledgements.
This work was supported by the NSF and by a grant from the W. M. Keck Foundation. The authors thank D. Blume for a critical reading of this manuscript.
|
warning/0506/quant-ph0506011.html
|
ar5iv
|
text
|
# Non-classical light emission by a superconducting artificial atom with broken symmetry
## I Introduction
The symmetry of a quantum system determines the selection rules of its transitions. For instance, all states of a generic atom must have a well-defined parity, and one-photon absorption (emission) due to the electric-dipole interaction can only happen for non-degenerate states with opposite parities. For second-order processes, a two-photon transition requires that these states have the same parities. Thus single- and two-photon transitions between two given energy levels cannot coexist.
Most investigations so far have focused on either $`\mathrm{\Lambda }`$, or $`\mathrm{\Xi }`$ (ladder), or $`V`$-type transitions scullybook ; jpm when studying three-level atomic systems. These notations, defined according to the transition configuration, are well known to physicists studying atoms and optics. For example, a $`\mathrm{\Lambda }`$-type transition atom means that there are optical transitions from the top energy level to the two lower energy levels, respectively; however, the optical transition between the two lower energy levels is forbidden.
The $`\mathrm{\Delta }`$-type three-level systems with cyclic transitions (CT) cyclic , in which one-photon and two photon processes coexist, are less common. It is of interest to explore the possibility of the coexistence of single photon and two photon processes. For chiral and other broken-symmetry systems, the lack of inversion center allows the CT to occur in realistic physical processes cyclic . It has been shown that such quantum systems can be experimentally implemented by left- and right-handed chiral molecules cyclic . With CT, the populations of the different energy levels can be selectively transferred by controlling classical fields.
In an atomic system, $`\mathrm{\Delta }`$-type transitions can also be formed fleischhauer by applying three classical pulses: a pair of Raman pulses and an additional detuning pulse. It was shown fleischhauer that the physical mechanism of the cyclic stimulated Raman adiabatic passage is not an adiabatic rotation of the dark state, but the rotation of a higher-order trapping state in a generalized adiabatic basis.
Most recently, the microwave control of the quantum states has been investigated for “artificial atoms” made of superconducting three-junction flux qubit circuits liu , which possess discrete energy levels. The optical selection rule of microwave-assisted transitions was analyzed liu for this artificial atom. It was shown liu that the microwave assisted transitions can appear for any two different states when the bias magnetic flux is near the optimal point but not equal to $`1/2`$ (the value of the optimal point is $`1/2`$). This is because the center of inversion symmetry of the potential energy of the artificial atom is broken when the bias is not equal to $`1/2`$. Then, so-called $`\mathrm{\Delta }`$-type or cyclic transition can be formed for the lowest three energy levels.
The $`\mathrm{\Delta }`$-type transitions can also be obtained from the model of the single-junction flux qubit zhou ; zafiris ; migliore . In any $`\mathrm{\Delta }`$type artificial atom, the population can be cyclically transferred by adiabatically controlling both the amplitudes and phases of the applied microwave pulses. However, the population transfer in the $`\mathrm{\Lambda }`$type artificial atom murali requires that two classical fields induce the transitions from the top energy level to other two lower energy levels, and transitions between two lower energy levels should be forbidden. This condition can be easily satisfied in the usual atoms due to the electric-dipole transition rule and its well defined party. However, in artificial atoms, these two fields can also induce a transition between two lower energy levels when we study a $`\mathrm{\Lambda }`$-type artificial atom liu . If some phase conditions are satisfied, $`\mathrm{\Delta }`$-type transitions can be formed even with only two classical fields. This is a basic difference between the usual atom fleischhauer and the artificial atom liu .
Here, we investigate new phenomena of a cyclic artificial atom, coupled to a quantized microwave field and controlled by two classical fields. We will explore the CT mechanism to create a single-mode photon state, or a macroscopic Schröedinger cat state which is the entangled state between a macroscopic quantum two level system (macroscopic qubit) and non-classical photon states. Our approach is robust because the working space is spanned by the ground state, or the two lowest energy levels, of the artificial atom. Because the ground state is not easy to be excited by the environment in low temperature limit. Also our scheme is more controllable than either $`\mathrm{\Lambda }`$, or $`\mathrm{\Xi }`$, or $`V`$-type atoms, since the extra coupling between the external field and the two lowest energy levels offers a new controllable parameter.
Our paper is organized as follows. In Sec. II, we describe how to model the superconducting flux qubit circuit as a three-level artificial system with $`\mathrm{\Delta }`$type (or cyclic) transitions, which are induced by the microwave electromagnetic fields. In Sec. III, we consider the case with large detuning. In this case, the top energy level can be adiabatically removed and an effectively driving field can be applied to the single-mode quantized field, then nonclassical states can be generated by the driving quantized field. In Sec. IV, it is demonstrated that the standard Schrödinger cat state, which is an entangled state between the inner states of the artificial atom and the quasi-classical photon state, can be generated. Finally, in Sec. V, we give conclusions and discuss possible applications.
## II Model and Hamiltonian
The artificial atom you considered here, described in Fig. 1(a), is a superconducting loop with three Josephson junctions orlando ; yu ; saito . Two junctions have the same Josephson energies and capacitances, which are $`\alpha `$ times larger than that of the third one. Then, the Hamiltonian can be written as liu ; orlando
$$H^{}=\frac{P_p^2}{2M_p}+\frac{P_m^2}{2M_m}+U(\phi _p,\phi _m,f),$$
(1)
with the effective masses $`M_p=2C_\mathrm{J}(\mathrm{\Phi }_0/2\pi )^2`$ and $`M_m=M_p(1+2\alpha )`$. The effective potential $`U(\phi _p,\phi _m,f)`$ is
$`U(\phi _p,\phi _m,f)`$ $`=`$ $`2E_\mathrm{J}(1\mathrm{cos}\phi _p\mathrm{cos}\phi _m)`$ (2)
$`+`$ $`\alpha E_\mathrm{J}[1\mathrm{cos}(2\pi f+2\phi _m)]`$
where $`\phi _p=(\phi _1+\phi _2)/2`$ and $`\phi _m=(\phi _1\phi _2)/2`$ are defined by the phase drops $`\phi _1`$ and $`\phi _2`$ across the two larger junctions; $`f=\mathrm{\Phi }_\mathrm{e}/\mathrm{\Phi }_0`$ is the reduced bias magnetic flux through the qubit loop, and $`\mathrm{\Phi }_0`$ is the magnetic flux quantum.
The potential energy $`U(\varphi _p,\varphi _m,f)`$ is an even function of the canonical variable $`\varphi _p`$, and naturally has the mirror symmetry for $`\varphi _p=\varphi _p`$. For other variable $`\varphi _m`$, the symmetry is completely determined by the reduced bias magnetic flux $`f`$. This is shown in Fig. 1(b), comparing $`f=0.5`$ and $`f=0.45`$, for a given $`\varphi _p=0.9`$. When $`2f=n`$ with an integer $`n`$, the potential energy $`U`$ has an inversion symmetry with respect to both phase variables $`\varphi _m`$ and $`\varphi _p`$; that is,
$$U(\varphi _m,\varphi _p,\mathrm{\hspace{0.17em}2}f=n)=U(\varphi _m,\varphi _p,\mathrm{\hspace{0.17em}2}f=n),$$
(3)
and thus the parities of the eigenstates are well-defined. However, the inversion symmetry with $`\phi _p`$ and $`\phi _m`$ is broken when $`2fn`$, that is,
$$U(\varphi _m,\varphi _p,\mathrm{\hspace{0.17em}2}fn)U(\varphi _m,\varphi _p,\mathrm{\hspace{0.17em}2}f=n).$$
(4)
Ref. liu computed the $`f`$-dependent energy spectrum, with the lowest three energy levels, denoted by $`|b`$, $`|c`$, and $`|e`$, well separated from the other upper-energy levels. Since microwave-assisted transitions can occur among the lowest three energy levels liu , this artificial atom allows, cyclic or $`\mathrm{\Delta }`$-shaped, transitions when $`f0.5`$.
Besides the bias magnetic flux $`\mathrm{\Phi }_e`$, we also apply another magnetic flux $`\mathrm{\Phi }_a`$, consisting of a quantized field and two classical fields. To realize the strong coupling of the flux qubit to a quantized field, now the flux qubit is coupled to a one-dimensional transmission line resonator. This can be realized by replacing the charge-qubit in the circuit QED architecture liu-epl ; wallraff ; liu-pra ; yliu by a flux qubit. Then a single-mode quantized magnetic field can be provided by the transmission line resonator. All three fields are assumed to induce transitions among the lowest three energy levels of the artificial atom to form the $`\mathrm{\Delta }`$-shaped configuration mentioned above. The frequencies of the quantized and two classical fields are assumed to be $`\omega `$, $`\mathrm{\Omega }_e`$, and $`\mathrm{\Omega }_c`$, respectively.
The Hamiltonian of the three-level artificial atom interacting with the three fields can be written as
$`H`$ $`=`$ $`\omega _e|ee|+\omega _c|cc|+\omega a^{}a`$
$`+`$ $`(g|ec|a+Ge^{i\mathrm{\Omega }_et}|be|+\lambda e^{i\mathrm{\Omega }_ct}|bc|+\mathrm{H}.\mathrm{c}.).`$
Here, we take $`\mathrm{}=1`$. The quantized field is assumed to couple the transition between $`|e`$ and $`|c`$, while the two classical fields are applied between $`|e`$ and $`|b`$, as well as between $`|c`$ and $`|b`$, respectively. $`\omega _e`$ ($`\omega _c`$) are transition frequencies between $`|e`$ ($`|c`$) and $`|b`$ (see the Fig. 2). The detuning between the transition frequency $`\omega _e`$ (or $`\omega _c`$) and the frequency of the classical field $`\mathrm{\Omega }_e`$ (or $`\mathrm{\Omega }_c`$) is denoted by
$$\mathrm{\Delta }_e=\omega _e\mathrm{\Omega }_e\mathrm{or}\mathrm{\Delta }_c=\omega _c\mathrm{\Omega }_c.$$
(6)
$`a`$ and $`a^{}`$ are the annihilation and creation operators of the quantized mode, $`G`$ and $`\lambda `$ are the Rabi-frequencies of the classical fields, $`g`$ denotes the vacuum Rabi-frequency of the quantized mode. Without loss of generality, we assume that all Rabi frequencies are real numbers. Here, we assume that the frequencies of the three fields satisfy the condition
$$\mathrm{\Omega }_e\mathrm{\Omega }_c=\omega .$$
(7)
This condition is required such that the equivalent Hamiltonian in a “rotating” reference frame (defined below) will be time-independent. In this case, the evolution of the quantum system will remain in the adiabatic subspace when the Rabi frequencies are adiabatically changed to transfer the quantum information, carried by photons, to the artificial atoms.
Figure 2 illustrates the transitions induced by the interactions of the artificial atom with the three fields. This cyclic or $`\mathrm{\Delta }`$-shaped transitions define a new type of atom, different from the $`\mathrm{\Lambda }`$ (or $`\mathrm{\Xi }`$, or $`V`$)-type atoms scullybook ; jpm . In a “rotating” reference frame of a time-dependent unitary transformation
$$W(t)=\mathrm{exp}[it(\mathrm{\Omega }_e|ee|+\mathrm{\Omega }_c|cc|+\omega a^{}a)],$$
(8)
the Hamiltonian in Eq. (II) can be rewritten as
$`H`$ $`=`$ $`\mathrm{\Delta }_c|cc|+\mathrm{\Delta }_e|ee|`$ (9)
$`+`$ $`(g|ec|a+G|eb|+\lambda |bc|+\mathrm{H}.\mathrm{c}.),`$
where the the frequencies-matching condition $`\mathrm{\Omega }_e\mathrm{\Omega }_c=\omega `$ has been used.
The population of the three-level artificial atom can be cyclically transferred by adiabatically applying three classical fields liu . However, in the presence of a quantized field, the transitions
$`|e,n`$ $``$ $`|c,n+1|b,n+1`$
$``$ $`|e,n+1|c,n+2|e,n+2\mathrm{}`$
cannot form a closed cycle because each cycle produces a one photon excitation. The triangular or $`\mathrm{\Delta }`$-shaped geometry of the transitions is shown in Fig. 3, where the classical fields can only induce transitions in the plane of each triangle of atom-photon joint states, while the quantized field drives the transitions from one plane to another, by increasing or decreasing one photon.
## III Mechanism to generate nonclassical photon states
In this section, we will consider the possibility to utilize the above $`\mathrm{\Delta }`$-shaped three level artificial atom as a basic single photon device. It is well known that there has been considerable interest in the generation of non-classical light using solid-state devices for highly sensitive metrology and quantum information. Some solid-state lasers have been proposed to emit non-classical light with photon number squeezing, but the present proposal, based on $`\mathrm{\Delta }`$-shaped artificial atoms, is essentially a macroscopic quantum device, which, in principle, could be easily controlled by only using classical parameters (e.g., the magnetic flux).
To intuitively describe the main mechanism of how to create the quasi-classical and non-classical photon states by using the transition configuration shown in Fig. 3, we first rewrite the sub-Hamiltonian in Eq. (9)
$$H_s=\mathrm{\Delta }_c|cc|+[\lambda |cb|+\mathrm{H}.\mathrm{c}.],$$
(10)
into
$$H_s=ϵ_+|++|+ϵ_{}||$$
(11)
with two dressed states
$`|+`$ $`=`$ $`\mathrm{cos}\left({\displaystyle \frac{\theta }{2}}\right)|c+\mathrm{sin}\left({\displaystyle \frac{\theta }{2}}\right)|b,`$
$`|`$ $`=`$ $`\mathrm{sin}\left({\displaystyle \frac{\theta }{2}}\right)|c+\mathrm{cos}\left({\displaystyle \frac{\theta }{2}}\right)|b,`$
where we have defined the mixing angle
$$\theta =\mathrm{arctan}\left(\frac{2\lambda }{\mathrm{\Delta }_c}\right).$$
(12)
It is obvious that $`\theta `$ can be controlled through the detuning $`\mathrm{\Delta }_c`$ by changing the frequency of the classical field. The states $`|\pm `$ are the eigenstates of $`H_s`$ corresponding to the eigenvalues
$$ϵ_\pm =\frac{\mathrm{\Delta }_c}{2}\pm \omega ^{},$$
(13)
with the dressed frequency
$$\omega ^{}=\sqrt{\lambda ^2+\frac{\mathrm{\Delta }_c^2}{4}}.$$
(14)
Then, in this dressed basis, the total Hamiltonian in Eq. (9)
$$H=H_0+H_1$$
(15a)
can be rewritten as
$$H_0=\mathrm{\Delta }_e|ee|+ϵ_+|++|+ϵ_{}||$$
(15b)
and
$$H_1=g(\theta )A|e+|G(\theta )B|e|+\mathrm{H}.\mathrm{c}.$$
(15c)
with the displaced boson operators $`A=a+\xi `$ and $`B=A\zeta `$, and the controllable parameters
$`g(\theta )=g\mathrm{cos}\left({\displaystyle \frac{\theta }{2}}\right),`$ $`G(\theta )=g\mathrm{sin}\left({\displaystyle \frac{\theta }{2}}\right),`$
$`\xi (\theta )={\displaystyle \frac{G}{g}}\mathrm{tan}\left({\displaystyle \frac{\theta }{2}}\right),`$ $`\zeta (\theta )={\displaystyle \frac{G}{g}}\mathrm{tan}^1\left({\displaystyle \frac{\theta }{2}}\right).`$
The Hamiltonian (15a) describes the $`\mathrm{\Lambda }`$-like transition atom shown in Fig. 4(a). Instead of the usual $`\mathrm{\Lambda }`$-type atom, the transitions between states $`|e`$ and $`|`$ are induced by two fields, one is a quantized light field with coupling strength $`g\mathrm{sin}(\theta /2)`$, described by a displaced annihilation operator $`a`$, another is a classical field with the Rabi frequency $`G\mathrm{cos}(\theta /2)`$.
Figure 4(a) schematically describes the creation of quasi-classical and non-classical photon states based on the CT process. Due to the coherent $`|c`$-$`|b`$ interaction with the coupling strength $`\lambda `$, as in Eq. (10), the system can be described by the driven JC model shown in Eq. (15a). However, for large detunings, in Eqs. (15a-15c), i.e., $`\mathrm{\Delta }_eϵ_\pm g(\theta ),G(\theta )`$, we can adiabatically separate the excited state $`|e`$ and then a coherent transition between states $`|c`$ and $`|b`$ is induced by the quantized field originally applied between $`|e`$ and $`|c`$. This is very similar to the usual Jaynes-Cummings (JC) model, obtained by adiabatically eliminating the highest third energy level in the stimulated Raman scattering of intense laser light. In the dressed states $`|\pm `$ basis, and for the above large detuning condition, there exist three types of subspaces, related to states $`|+`$, $`|`$, and $`|e`$, respectively. These subspaces are depicted in Fig. 4(b) by vertical lines linking the vertices of the triangles. Corresponding to each state, e.g., $`|+`$, the photon mode is driven by an effective external force depending on the coherent $`|c`$-$`|b`$ interaction, and thus the single mode photon states can be produced from the vacuum state.
## IV Adiabatic generation of Schrödinger cat states
In order to better understand the above-mentioned mechanism to generate non-classical photon states from these controllable artificial atoms, we demonstrate the adiabatic generation of Schrödinger cat states. In the large detuning limit, we can adiabatically eliminate the terms causing transitions from $`|e`$ to $`|+`$ and $`|`$.
The adiabatic elimination can be done by using the Fröhlich-Nakajima transformation (FNT) fro ; nakajima , which is applied to achieve the effective electron-electron interaction Hamiltonian in the BCS theory. To consider the validity of this method, we will show that it is equivalent to a result of the second order perturbation in the Appendix A. In the FNT method, we define a transformation by the operator $`V=`$ $`\mathrm{exp}(S)`$, with an anti-Hermitian operator $`S`$ to be determined. Then we apply this transformation $`V`$ to the original Hamiltonian (15a) to given an equivalent Hamiltonian $`H_V=V^{}HV.`$
We assume that the operator $`S`$ to be the perturbation term with the same order as $`H_I`$, and then we can expand $`H_V`$ in the series of $`S`$. In general, we can consider the Hamiltonian of an interacting system, described by a sum of free Hamiltonian $`H_0`$ and the interaction Hamiltonian $`H_1`$ as $`H=H_0+H_1`$, shown in Eq. (15a). By comparing with the free part $`H_0`$, the interaction part $`H_1`$ can be regard as a perturbation term. Let us perform the transformation $`V=`$ $`\mathrm{exp}(S)`$ on the Hamiltonian $`H=H_0+H_1`$. Then, we can derive an approximately equivalent Hamiltonian $`H_V`$ as
$$H_VH_0+\frac{1}{2}[H_1,S],$$
(16)
where the operator $`S`$ can be determined by
$$H_1+[H_0,S]=0.$$
(17)
The transformation, by which one can obtain the effective Hamiltonian in Eq. (16) from the Hamiltonian in Eq. (15a), is the so-called generalized Fröhlich transformation (for details, see Appendix A).
If we replace $`H_0`$ and $`H_1`$ in Eq. (17) by the explicit expressions in Eqs. (15b) and (15c), and assume
$`S`$ $`=`$ $`\mathrm{\Gamma }_1A|e+|+\mathrm{\Gamma }_2B|e|`$ (18)
$`+`$ $`\mathrm{\Gamma }_3A^{}|+e|+\mathrm{\Gamma }_4B^{}|e|,`$
for parameters $`\mathrm{\Gamma }_i(i=1,\mathrm{\hspace{0.17em}2},\mathrm{\hspace{0.17em}3},\mathrm{\hspace{0.17em}4})`$ to be determined, then the parameters $`\mathrm{\Gamma }_i(i=1,\mathrm{\hspace{0.17em}2},\mathrm{\hspace{0.17em}3},\mathrm{\hspace{0.17em}4})`$ can be obtained as
$`\mathrm{\Gamma }_1`$ $`=`$ $`\mathrm{\Gamma }_3={\displaystyle \frac{g(\theta )}{ϵ+\mathrm{\Delta }}},`$ (19a)
$`\mathrm{\Gamma }_2`$ $`=`$ $`\mathrm{\Gamma }_4={\displaystyle \frac{G(\theta )}{ϵ\mathrm{\Delta }}},`$ (19b)
with
$$\mathrm{\Delta }=\frac{1}{2}\sqrt{(\mathrm{\Delta }_c\mathrm{\Delta }_e)^2+4\lambda ^2},ϵ=\frac{1}{2}(\mathrm{\Delta }_c+\mathrm{\Delta }_e).$$
(20)
Then, using the expressions of $`S`$, $`H_0`$, and $`H_1`$ in Eqs. (18), (15b), and (15c), we can obtain an effective Hamiltonian from Eq. (16) as
$$H_VH_e|ee|+H_{bc}.$$
(21a)
Here, the Hamiltonians $`H_e`$ and $`H_{bc}`$ can be expressed as
$$H_e=\mathrm{\Delta }_e+\mathrm{\Omega }_AAA^++\mathrm{\Omega }_BBB^+$$
(21b)
and
$`H_{bc}`$ $`=`$ $`(ϵ_+\mathrm{\Omega }_AA^+A)|++|`$ (21c)
$`+(ϵ_{}\mathrm{\Omega }_BB^+B)||`$
$`+\mathrm{\Gamma }[AB^+|+|+A^+B|+|],`$
with
$$\mathrm{\Gamma }=\frac{G(\theta )g(\theta )}{2\mathrm{\Delta }_{}\mathrm{\Delta }_+}(2\mathrm{\Delta }_e\mathrm{\Delta }_c).$$
(22)
The effective frequencies
$$\mathrm{\Omega }_A=\frac{g^2(\theta )}{\mathrm{\Delta }_+},\mathrm{\Omega }_B=\frac{G^2(\theta )}{\mathrm{\Delta }_{}}$$
(23)
represent the Stark shifts with $`\mathrm{\Delta }_\pm =\mathrm{\Delta }_eϵ_\pm `$.
According to former definitions of the operators $`A`$ and $`B`$ in Eq. (15c), the Hamiltonian $`H_e`$ can be rewritten as
$`H_e`$ $`=`$ $`(\mathrm{\Omega }_A+\mathrm{\Omega }_B)aa^{}`$
$`+[(\xi \mathrm{\Omega }_A\eta \mathrm{\Omega }_B)a^{}+\mathrm{H}.\mathrm{c}.]`$
after neglecting the constant terms $`\mathrm{\Delta }_e+|\xi |^2+|\eta |^2`$. It is clear that the Hamiltonian $`H_e`$ describes a driven harmonic oscillator. Then, when the total system can be adiabatically kept in the excited state $`|e`$, $`H_e`$ describes the creation of a coherent photon state from the vacuum scullybook . However, due to the spontaneous emission of excited states, it is difficult to keep the artificial atom in its excited state $`|e`$. Thus, let us now consider how to generate non-classical photon states by only using the more robust lower states $`|\pm `$.
The last term in $`H_{bc}`$ oscillates in a larger frequency range: $`|ϵ_+ϵ_{}|2\omega ^{}`$. Thus, in the rotating wave approximation, we have
$`H_{bc}`$ $`=`$ $`(ϵ_+\mathrm{\Omega }_AA^+A)|++|`$ (25)
$`+`$ $`(ϵ_{}\mathrm{\Omega }_BB^+B)||.`$
This is the standard Hamiltonian to describe the dynamical generation of Schrödinger cat states (e.g., Ref. cqed ). Since the bare ground state $`|b`$ is easy to be initialized, we can assume that the artificial atom is initially in the bare ground state $`|b=\mathrm{sin}(\theta /2)|++\mathrm{cos}(\theta /2)|`$, while the cavity field is initially in the vacuum state $`|0`$. Then at time $`\tau `$, the whole system can evolve into
$`|\psi (\tau )`$ $`=`$ $`\mathrm{exp}(iH_{bc}t)[\mathrm{sin}(\theta /2)|++\mathrm{cos}(\theta /2)|]|0`$ (26)
$`=`$ $`\mathrm{sin}{\displaystyle \frac{\theta }{2}}\mathrm{exp}[i\xi ^2\mathrm{exp}(i\mathrm{\Omega }_At)]|\alpha (\xi ,t)|+`$
$`+`$ $`\mathrm{cos}{\displaystyle \frac{\theta }{2}}\mathrm{exp}[i\zeta ^2\mathrm{exp}(i\mathrm{\Omega }_Bt)]|\alpha (\zeta ,t)|.`$
where $`|\alpha (x,t)=`$ $`|\alpha =\alpha (x,t)`$ (and $`x=\xi ,\zeta `$) denotes coherent states with
$$\alpha (x,t)=x[1\mathrm{exp}(i\mathrm{\Omega }_xt)]$$
(27)
and $`\mathrm{\Omega }_\xi =\mathrm{\Omega }_{A,}`$ $`\mathrm{\Omega }_\zeta =\mathrm{\Omega }_B`$. By adjusting the coupling constant $`\lambda `$ between $`|c`$ and $`|b`$, in this “cyclic atom”, one can control dynamical processes to obtain the cat states of the qubit subsystem consisting of the two dressed states $`|\pm `$ entangled with the quantized field.
To show the existence of the “cat”, we need to calculate the overlap
$$F(\lambda ,t)=|\alpha (\zeta ,t)|\alpha (\xi ,t)|=\mathrm{exp}[y(\lambda ,t)]$$
(28)
for two coherent states $`|\alpha (\xi ,t)`$ and $`|\alpha (\zeta ,t)`$, where
$`y(\lambda ,t)`$ $`=`$ $`2(\zeta +\xi )^24\zeta \xi \mathrm{sin}^2\left[{\displaystyle \frac{t}{2}}(\mathrm{\Omega }_B\mathrm{\Omega }_A)\right]`$
$`2\zeta (\zeta +\xi )\mathrm{cos}(\mathrm{\Omega }_Bt)2\xi (\zeta +\xi )\mathrm{cos}(\mathrm{\Omega }_At).`$
In Fig. 5, the time evolution of $`y(t)`$ is plotted for given parameters, e.g., $`\mathrm{\Delta }_e=3\lambda `$, $`G=0.9\lambda `$, $`g=0.8\lambda `$ for different values of $`\theta =\mathrm{arctan}(2\lambda /\mathrm{\Delta }_c)=\pi /2,\pi /4,\pi /6`$. It shows that $`y(t)`$ can periodically reach its maximum value, which means that $`|\alpha (\zeta ,t)|\alpha (\xi ,t)|`$ becomes minimum at these times, with period $`2\pi `$. The period of the function $`y(t)`$ is determined by three frequencies $`\mathrm{\Omega }_A`$, $`\mathrm{\Omega }_B`$, and $`\mathrm{\Omega }_B\mathrm{\Omega }_A`$, so Fig. 5 shows the small modulation overimposed on the larger modulation. We find that a larger detuning $`\mathrm{\Delta }_c`$ corresponds to a larger maximum value when other parameters are fixed. However, $`y(t)`$ needs a longer period to reach these maximum points. The above result demonstrates that macroscopic Schrödinger cat states, an entanglement between a macroscopic quantum two-level system (macroscopic qubit) and the non-classical photon states, can be generated by superconducting quantum devices. These cat states are different from the usual Schrödinger cat states, an entanglement between a microscopic two level atom and the quasi-classical photon states, which are created by using the atomic cavity QED cqed .
The above setup can also be used to create a coherent state if the cyclic artificial atom is initially prepared in the state $`|`$. In this case, the total system should be adiabatically kept in the ground state $`|`$. The effective Hamiltonian becomes $`H_{}\mathrm{\Omega }_BB^+B`$, by dropping the constant term. More explicitly,
$$H_{}\mathrm{\Omega }_Ba^+a+f(\theta )(a+a^+)$$
(30)
realizes a driven harmonic oscillator. The driving force $`f(\theta )`$ can be expressed as $`f(\theta )=G(\theta )g(\theta )/\mathrm{\Delta }_{}`$, and it depends on the coupling constant $`\lambda `$. Starting from the vacuum $`|0`$, with a duration $`t`$, the single-mode quantized field will evolve into a coherent state $`|\phi (t)=|\alpha `$ with $`\alpha =\zeta [1\mathrm{exp}(i\mathrm{\Omega }_Bt]`$, where a time-dependent global phase $`\mathrm{exp}[i\zeta ^2(\mathrm{sin}(\mathrm{\Omega }_Bt)\mathrm{\Omega }_Bt)]`$ has been neglected. From the expression of the photon number
$$N(t)=\alpha |a^+a|\alpha =\zeta ^2|1\mathrm{exp}(i\mathrm{\Omega }_Bt))|^2,$$
(31)
we can calculate the generation rate of the photons in the quantized mode:
$$r(t)=\left|\frac{\mathrm{d}N(t)}{\mathrm{d}t}\right|=\frac{2g^2(\theta )}{\mathrm{\Delta }_{}}\mathrm{sin}(\mathrm{\Omega }_bt).$$
(32)
This result shows that, the coupling strength $`\lambda `$ of the interaction between $`|c`$ and $`|b`$, caused by the symmetry-breaking, can be used to enhance the probability of creating single-mode photons. If there is no interaction between $`|c`$ and $`|b`$, the external force $`f(\theta )`$ would vanish accordingly and then the dynamic evolution cannot automatically produce coherent photon states.
## V Conclusions
In an artificial atom represented by a superconducting quantum circuit, we briefly review optical transitions and their selection rules. It is shown that all transitions are possible in such artificial atom when the applied bias magnetic flux is not at the optimal point liu . Then cyclic or $`\mathrm{\Delta }`$-shaped transitions can be realized for the lowest three energy levels in this artificial atom. Using this cyclic population transfer mechanisms, we have studied how to create nonclassical single-mode photon states and a macroscopic Schrödinger cat states. We show that this approach is controllable, because either the ground state or the two lowest-energy levels are utilized through their coherent coupling to external fields, which can be used to control the parameters of the system. For example, if the Rabi frequency $`\lambda =0`$, then the classical field, which induces transitions between states $`|c`$ and $`|b`$, is set to zero. Thus, our model can be referred to $`\mathrm{\Lambda }`$type atom driven by a quantized and a classical fields parkins . In this case parkins , neither the cat state nor the coherent state can be generated from the initial state of the whole system with the bare ground state of atom and the vacuum of the quantized field. Starting from either the vacuum or a coherent state, it is a deterministic scheme to generate nonclassical photon states via CT manipulations.
The large detuning limit implies a relatively weak coupling constants $`g(\theta )`$ and $`G(\theta )`$, as well as relatively large detuning $`\mathrm{\Delta }_eϵ_\pm `$, shown in Eqs. (15b-15c). Thus it limits the efficient adjustment of the dynamic processes. Therefore, a generic version of our proposal might not be very efficient. Fortunately, in our proposal, the strength $`\lambda `$ of the controlling-field coupling between $`|c`$ and $`|b`$ is adjustable. Thus, one can feasibly manipulate this parameter for our goal without violating the requirement of large detuning.
Another question is the problem of decoherence. An efficient scheme requires the decoherence time comparable with the characteristic time of the effective frequencies $`\mathrm{\Omega }_B`$ and $`\mathrm{\Omega }_A`$. It is known that the interaction strength $`g`$ between the qubit and quantized field can reach about $`100`$ MHz if we use the transmission line resonator in the circuit QED gir ; wallraff . According to the definitions of $`\mathrm{\Omega }_B`$ and $`\mathrm{\Omega }_A`$ in Eq. (23), they can be of the order of $`10100`$ MHz, if we choose appropriate detunings $`\mathrm{\Delta }_e`$ and $`\mathrm{\Delta }_c`$. Then, it is possible to realize our proposal within the experimental values gir1 for $`T_17\mu `$s and $`T_2800`$ ns.
Finally, we should point out the relation between our present work and the quantum Carnot engine (QCE) in Ref. scully ; quan . In the QCE proposal, the $`\mathrm{\Lambda }`$-type atoms are prepared as a superposition of two lower states. In the $`\mathrm{\Delta }`$-type transition configuration, the superposition of the two lowest states can be naturally produced by the interaction between the field and the artificial atom, and hence the cyclic three-level atom is a good candidate to demonstrate the QCE.
## VI Acknowledgments
We acknowledge the partial support of the US NSA and ARDA under AFOSR contract No. F49620-02-1-0334, and the NSF grant No. EIA-0130383. The work of CPS is also partially supported by the NSFC and Fundamental Research Program of China with No. 2001CB309310.
## Appendix A Generalized Fröhlich-Nakajima transformation and its equivalence to perturbation theory
Let us consider a Hamiltonian $`H`$ of a given system with its free part $`H_0`$ and a perturbation term $`H_I`$
$$H=H_0+\lambda H_I.$$
(33)
Here, $`\lambda `$ is the so-called perturbation parameter introduced to characterize the order of the perturbation. At the end of calculation, $`\lambda `$ is taken as unity.
The crucial point of the generalized Fröhlich-Nakajima transformation is to choose a proper unitary transformation $`V(\lambda )=\mathrm{exp}(\lambda S)`$, where $`S`$ is an anti-Hermitian operator, to be determined. The inverse transformation of $`V(\lambda )`$ makes the states $`|\mathrm{\Psi }`$, governed by the Hamiltonian in Eq. (33), change to a new state
$$|\mathrm{\Phi }=V(\lambda )|\mathrm{\Psi }=\mathrm{exp}(\lambda S)|\mathrm{\Psi }.$$
(34)
And the evolution of the state $`|\mathrm{\Phi }`$ is governed by the transferred Hamiltonian
$$H_\lambda =e^{\lambda S}He^{\lambda S}.$$
(35)
It is well known that the unitary transformation does not change the dynamics of the system, and then the Hamiltonians $`H_\lambda `$ and $`H`$ describe the same physical process. Here the operator $`S`$ should be appropriately chosen such that it has the same order as the perturbation term $`H_I`$. Physically, the effect of the Hamiltonian $`H_I`$ on the final result is so small that it can be neglected.
Using the Baker-Campbell-Hausdorff formula, the Hamiltonian $`H_\lambda `$ can be expressed in a series of the parameter $`\lambda `$ as
$$H_\lambda =H_0+\underset{n=1}{}\frac{\lambda ^n(1)^{n1}}{(n1)!}\underset{n1}{\underset{}{[S,[S,\mathrm{}[S}},H_I]]].$$
(36)
Second order perturbation theory can be realized by imposing the condition
$$H_I+[H_0,S]=0$$
(37)
on Eq. (36). Eq. (37) can be used to determine the operator $`S`$. For the sake of simplicity, the eigenstates of $`H_0`$ are assumed to be non-degenerate. Let $`|n`$ be the eigenstate of the Hamiltonian $`H_0`$ with the eigenvalue $`E_n`$. Taking the matrix elements of Eq. (37 ) with respect to the basis $`\{|n\}`$ as
$$m|H_I|n+\left(E_mE_n\right)m|S|n=0,$$
(38)
we can find the explicit expression of matrix elements for the operator $`S`$
$$S_{mn}=m|S|n=\frac{m|H_I|n}{E_nE_m}.$$
(39)
Thus, the representation of the operator $`S`$ in the $`\{|n\}`$ basis can be
$$S=\underset{mn}{}\frac{m|H_I|n}{E_nE_m}|mn|.$$
(40)
From Eqs. (36) and (37), we obtain the effective Hamiltonian
$$H_\lambda H_0+\frac{1}{2}[H_I,S]$$
(41)
up to second order in $`H_I`$. Using a matrix representation, $`H_\lambda `$ can be expressed as
$`H_\lambda `$ $`=`$ $`{\displaystyle \underset{n}{}}E_n|nn|`$ (42)
$`+`$ $`{\displaystyle \underset{l(ln),m}{}}{\displaystyle \frac{m|H_I|ll|H_I|n}{2(E_nE_l)}}|mn|`$
in the $`\{|n\}`$ basis. We can see that the Fröhlich-Nakajima transformation is only applicable to a systems with $`m|H_I|m=0`$. Actually we can decompose the total Hamiltonian $`H`$ such that $`H_0`$ only includes all diagonal elements in the $`\{|n\}`$ basis of eigenstates for the Hamiltonian $`H_0`$ while the off-diagonal ones are included in $`H_I`$.
It is easy to obtain the eigenvalues of the transferred Hamiltonian in Eq. (41) or (42), up to second order in $`H_I`$, as
$`E_n^{(0)}`$ $`=`$ $`n|H_0|n+{\displaystyle \frac{1}{2}}n|[H_I,S]|n`$ (43)
$`=`$ $`E_n+{\displaystyle \underset{ln}{}}{\displaystyle \frac{|l|H_I|n|^2}{E_nE_l}},`$
which correspond to the zero-order eigenstates of the Hamiltonian $`H_\lambda `$. The second term in the right side of Eq. (43) is the so-called self-energy term.
In fact, from Eq. (42), it can be found that zero-order eigenstates $`|\mathrm{\Psi }_n^{(0)}`$ of the Hamiltonian $`H_\lambda `$ are just the eigenstates $`|n`$ of the Hamiltonian $`H_0`$, i.e., $`|\mathrm{\Psi }_n^{(0)}=|n`$. The eigenvalues in Eq. (43) provide energy corrections using the time-independent perturbation theory.
To consider the relation between the Fröhlich-Nakajima transformation and the time-independent perturbation theory, we can transfer eigenstates $`|\mathrm{\Psi }_n^{(0)}`$ back to the original picture. In this case, the first order eigenstates $`|\mathrm{\Psi }_n^{(1)}`$ of the Hamiltonian $`H`$ can be obtained by
$`|\mathrm{\Psi }_n^{(1)}`$ $`=`$ $`V\left(\lambda \right)|\mathrm{\Psi }_n^{(0)}=\left(1+S\right)|n`$
$`=`$ $`|n+{\displaystyle \underset{mn}{}}{\displaystyle \frac{m|H_I|n}{E_nE_m}}|m,`$
where the expansion $`V\left(\lambda \right)`$ is kept up to first order in $`\lambda `$.
It is easy to prove that $`|\mathrm{\Psi }_n^{(1)}`$ are just the first-order eigenstates of the original Hamiltonian $`H`$, with respect to the perturbation decomposition of $`H_0`$ and $`H_I`$. Since we have chosen that $`H_I`$ does not have diagonal terms, the first correction to the energy is zero, and then $`E_n`$ is also the result of the first correction of the energy for the Hamiltonian $`H`$.
The eigenvalues in Eq. (43) are up to the second order corrections. Correspondingly, the eigenstates $`|\mathrm{\Psi }_n^{(2)}`$ of $`H`$ corresponding to the second-order energy corrections can be given by acting $`V(\lambda )`$ on the first order eigenstates $`|\mathrm{\Psi }_n^{(1)}`$ of the Hamiltonian $`H_\lambda `$. That is
$`|\mathrm{\Psi }_n^{(2)}`$ $`=`$ $`V(\lambda )|\mathrm{\Phi }_n^{(1)}=V^1(\lambda )|\mathrm{\Psi }_n^{(1)}`$
$`=`$ $`\left(1+S+{\displaystyle \frac{S^2}{2}}\right)|\mathrm{\Psi }_n^{(1)}`$
$`=`$ $`|n+{\displaystyle \underset{mn}{}}{\displaystyle \frac{m|H_I|n}{E_nE_m}}|m`$
$`+`$ $`{\displaystyle \underset{l,m,ln}{}}{\displaystyle \frac{m|H_I|ll|H_I|n}{2\left(E_lE_m\right)\left(E_nE_l\right)}}|m.`$
|
warning/0506/gr-qc0506073.html
|
ar5iv
|
text
|
# Charged black holes in compactified spacetimes
## I Introduction
String theory predicts that our world has more dimensions than the 4 we see around us every day. The usual way out of this apparent discrepancy is to say that the additional dimensions are compactified and small. In the simplest case the extra dimensions are just circles. String theory is also a theory of gravity. It is therefore a natural and interesting question to ask what is the behavior of the classical solutions of General Relativity in spacetimes where one or more dimensions are compactified. This question is however surprisingly hard to answer. The periodic analogue of a Schwarzschild black hole in a space of topology $`R3\times S1`$ has been st udied more or less independently by Majumdar Majumdar , Papapetrou Papapetrou , Myers Myers , Korotkin and Nicolai Korotkin1 ; Korotkin2 as well as by Frolov and Frolov Frolov . In higher dimensions the full solution is known only implicitly HO1 ; HO3 ; HO4 ; Kol5 ; Kol6 .
The problem becomes even more interesting when one realizes, following Gregory and Laflamme GL1 ; GL2 , that there is a competition between different gravitational configurations with the same mass and charges but with different symmetries and even of different horizon topology. Which configuration is stable depends on the particular value of the mass and the charges. For instance, a long and thin black string becomes unstable and will ”decay” into a new configuration with higher entropy. This new configuration could for instance be a nontranslationally invariant black string or an infinite array of black holes. The phase diagram of black objects in higher dimensional circle space times has a very rich structure Kol1 ; HO2 ; Kol3 ; HO5 . Only parts of it is accessible to analytical methods and the most interesting pices has been calculated only numerically Wiseman1 ; Wiseman2 ; Kol2 ; Kol4 ; Wiseman3 ; Wiseman4 or using perturbation theory Gubser . It would therefore clearly be of great interest to have a better analytical understanding of the various solutions that could appear.
In this paper we perform a more modest task. We generalize the solution given in Myers ; Korotkin1 ; Frolov to arbitrary electric and NUT charge. In doing so we find a new parameter of the solution which has not been recognized before. We analyze the physical meaning of this parameter. We also notice that adding NUT charge to the solution makes it more well behaved. Maybe this is a feature which would persist in higher dimensions? Our method heavily relies on Weyl coordinates. However, as shown in Emparan , because of symmetry reasons, Weyl coordinates are expected to be useful at most in 4 and 5 dimensions. This is something which would have to be overcome if one would like to generalize our results to higher dimensions.
This paper is organized as follows. Section 2 is a general discussion of properties of the four dimensional Einstein-Maxwells equations important for our problem. In Section 3 we discuss the (nonperiodic) Reissner-Nordström-NUT solution. In Section 4 we find its periodic generalization and discuss the metric in various limits. In Section 5 we give our conclusions.
## II Einstein-Maxwells equations for stationary spacetimes
We write a general stationary metric with timelike Killing vector $`\xi ^a`$ as
$$g_{ab}=f^2\mu _a\mu _b+f^2h_{ab},$$
(1)
where
$`f=\sqrt{\xi ^a\xi _a}`$ (2)
$`\mu _a=f^2\xi _a,`$ (3)
which implies that $`h_{ab}\xi ^b=0`$ and that $`f`$, $`\mu _a`$ and $`h_{ab}`$ are stationary fields, i.e. Lie dragged by $`\xi ^a`$. We now decompose the electromagnetic field strength $`F_{ab}`$ by defining the electric and magnetic fields $`E_a`$ and $`B_a`$ with respect to $`\xi ^a`$:
$`E_a=F_{ab}\xi ^b`$ (4)
$`B_a=F_{ab}\xi ^b,`$ (5)
where $`F_{ab}`$ is the dual field strength
$$F_{ab}=\frac{1}{2}ϵ_{ab}{}_{}{}^{cd}F_{cd}^{}.$$
(6)
It then follows that
$`F_{ab}`$ $`=2E_{[a}\mu _{b]}ϵ_{ab}{}_{}{}^{cd}B_{c}^{}\mu _d`$ (7)
$`F_{ab}`$ $`=2B_{[a}\mu _{b]}+ϵ_{ab}{}_{}{}^{cd}E_{c}^{}\mu _d.`$ (8)
Assuming that $`F_{ab}`$ is stationary as well as source-free so that $`F_{ab}`$ is closed, then both $`E_a`$ and $`B_a`$ are gradients of stationary scalars. Hence we introduce electric and magnetic potentials $`v`$ and $`u`$ according to
$$E_a=_av,B_a=_au.$$
(9)
Moreover, with the electromagnetic field being the only matter source, a twist scalar $`\chi `$, likewise stationary, can be introduced by<sup>1</sup><sup>1</sup>1We use units such that the four-dimensional Einstein-Maxwell equations read $`G_{ab}=2T_{ab}`$ where $`T_{ab}=F_a{}_{}{}^{c}F_{bc}^{}\frac{1}{4}F^{cd}F_{cd}g_{ab}`$.
$$_a\chi =ϵ_{abcd}\xi ^b^d\xi ^c+2(u_avv_au).$$
(10)
The tensor $`h_{ab}`$ can be viewed as a metric on the three-dimensional manifold $`\mathrm{\Sigma }`$ of Killing orbits and we shall use that the full 4D Einstein-Maxwell equations are now equivalent to the 3D equations obtained from the action
$$S=d^3x\sqrt{h}\left[{}_{}{}^{(3)}R2h^{ab}\gamma _{AB}D_aX^AD_bX^B\right],$$
(11)
where $`{}_{}{}^{(3)}R`$, $`h^{ab}`$ and $`D_a`$ are the Ricci scalar, inverse and Levi-Civita connection of $`h_{ab}`$. This action describes three-dimensional Euclidean gravity coupled to a certain sigma model. More precisely, the effective matter part of the action is that of a wave map from $`\mathrm{\Sigma }`$ to a four-dimensional target space with metric
$$d\sigma ^2=\gamma _{AB}dX^AdX^B=\frac{\frac{1}{2}dd\overline{}+\psi d\overline{\psi }d+\overline{\psi }d\psi d\overline{}(+\overline{})d\psi d\overline{\psi }}{2\left[\frac{1}{2}(+\overline{})+\psi \overline{\psi }\right]^2},$$
(12)
where $``$ and $`\psi `$ are the complex Ernst potentials
$$=f^2v^2u^2+i\chi ,\psi =v+iu.$$
(13)
This is an Einstein metric, whose Ricci tensor obeys
$$R_{AB}=6\gamma _{AB}$$
(14)
If we use as target space coordinates $`X^A`$ the real variables $`f,A,S,\chi `$, where $`A`$ and $`S`$ are the amplitude and phase of $`\psi `$, i.e.
$$\psi =Ae^{iS},$$
(15)
the target metric takes the neat form
$$d\sigma ^2=\frac{1}{f^2}(df^2dA^2A^2dS^2)+\frac{(d\chi +2A^2dS)^2}{4f^4},$$
(16)
which makes it evident that it has signature $`(,,+,+)`$. Now, varying the action (11) with respect to $`h^{ab}`$ gives
$${}_{}{}^{(3)}R_{ab}^{}=2\gamma _{AB}D_aX^AD_bX^B,$$
(17)
while variation with respect to the target space coordinates $`X^A`$ leads to the wave map equation
$$h^{ab}(D_aD_bX^C+\mathrm{\Gamma }^C{}_{AB}{}^{}D_{a}^{}X^AD_bX^B)=0,$$
(18)
with $`\mathrm{\Gamma }^C_{AB}`$ being the Christoffel symbols of $`\gamma _{AB}`$. For our purposes it is important to note that if the $`X^A`$ all depend solely on one free function $`\omega `$, then eqs. (17) and (18) become
$`{}_{}{}^{(3)}R_{ab}^{}=2\gamma _{AB}{\displaystyle \frac{dX^A}{d\omega }}{\displaystyle \frac{dX^B}{d\omega }}D_a\omega D_b\omega `$ (19)
$`{\displaystyle \frac{dX^C}{d\omega }}h^{ab}D_aD_b\omega +\left({\displaystyle \frac{d^2X^C}{d\omega ^2}}+\mathrm{\Gamma }^C{}_{AB}{}^{}{\displaystyle \frac{dX^A}{d\omega }}{\displaystyle \frac{dX^B}{d\omega }}\right)h^{ab}D_a\omega D_b\omega =0.`$ (20)
Now, eq. (20) tells us that the curve $`X^A(\omega )`$ is a geodesic, since, as $`h_{ab}`$ is supposed to be a positive-definite metric, it cannot happen that $`h^{ab}D_a\omega D_b\omega `$ vanishes identically unless $`\omega `$ is constant. Using the freedom to reparametrize the geodesic, we can assume that it is affinely parametrized and arrive at the equations
$`ϵ:=\gamma _{AB}{\displaystyle \frac{dX^A}{d\omega }}{\displaystyle \frac{dX^B}{d\omega }}=\mathrm{constant}`$ (21)
$`{\displaystyle \frac{d^2X^C}{d\omega ^2}}+\mathrm{\Gamma }^C{}_{AB}{}^{}{\displaystyle \frac{dX^A}{d\omega }}{\displaystyle \frac{dX^B}{d\omega }}=0`$ (22)
$`{}_{}{}^{(3)}R_{ab}^{}=2ϵD_a\omega D_b\omega `$ (23)
$`h^{ab}D_aD_b\omega =0.`$ (24)
In this paper we focus exclusively on this class of solutions, which we shall refer to as geodesic solutions.
### II.1 Axisymmetry and Weyl coordinates
Let us now assume that the stationary spacetime is also axisymmetric with an axisymmetry generator $`\eta ^a`$ that commutes with the timelike Killing vector $`\xi ^a`$. We can then introduce coordinates $`(t,\rho ,z,\varphi )`$ such that $`\xi ^a=(/t)^a`$, $`\eta ^a=(/\varphi )^a`$ while the three-metric $`h_{ab}`$ and one-form $`\mu _a`$ take the forms
$`dl^2=e^{2k}(d\rho ^2+dz^2)+W^2d\varphi ^2`$ (25)
$`𝝁=dt+\mathrm{\Omega }d\varphi .`$ (26)
The metric functions $`f`$, $`k`$, $`W`$ and $`\mathrm{\Omega }`$ obviously depend on $`\rho `$ and $`z`$ only. If we assume that $`\omega `$ also depends solely on $`\rho `$ and $`z`$ (although it could have a linear dependence on $`\varphi `$ with constant coefficient, a possibility which we do not consider here), the $`\varphi \varphi `$-component of eq. (23) becomes
$$W_{,\rho \rho }+W_{,zz}=0,$$
(27)
making it possible to choose $`\rho `$ and $`z`$ such that $`W=\rho `$, a choice well-known as Weyl (canonical) coordinates. The great advantage of being able to use these coordinates is that the unknown metric function $`k`$ does not enter the Laplace equation (24) (nor the more general wave map equation (18)), which will thus be identical to the Laplace equation in flat space, expressed in cylindrical coordinates, for an unknown function that is independent of $`\varphi `$;
$$\omega _{,\rho \rho }+\rho ^1\omega _{,\rho }+\omega _{,zz}=0.$$
(28)
With the use of this equation, the remaining components of eq. (23) become
$$\begin{array}{cc}& k_{,\rho }=ϵ\rho \left[(\omega _{,\rho })^2(\omega _{,z})^2\right]\hfill \\ & k_{,z}=2ϵ\rho \omega _{,\rho }\omega _{,z},\hfill \end{array}$$
(29)
while eq. (10) gives
$$\begin{array}{cc}& \mathrm{\Omega }_{,\rho }=4p_\chi \rho \omega _{,z}\hfill \\ & \mathrm{\Omega }_{,z}=4p_\chi \rho \omega _{,\rho },\hfill \end{array}$$
(30)
with $`p_\chi `$ being a constant equal to the conserved geodesic momentum associated with the cyclic coordinate $`\chi `$ of the target space metric. Explicitly,
$$p_\chi =\frac{{\displaystyle \frac{d\chi }{d\omega }}+2A^2{\displaystyle \frac{dS}{d\omega }}}{4f^4}.$$
(31)
A recipe for finding a solution of the type considered here is thus to choose an appropriate geodesic of the target space and an appropriate solution to the flat space Laplace equation (28). The geodesic then gives the four target space functions as functions of $`\omega `$, including the metric component $`f`$ and the potentials $`v`$ and $`u`$ which completely determine the electromagnetic field via eqs. (7) and (9). What then remains is to integrate the two pairs of equations (29) and (30) to determine the remaining spacetime metric components $`k`$ and $`\mathrm{\Omega }`$. Note that eq. (28) is the integrability condition for both pairs of equations (when $`ϵ0p_\chi `$).
## III The Reissner-Nordström-NUT solution
Using Schwarzschild type coordinates, the Reissner-Nordström-NUT (RNN) solution with mass $`M`$, electric charge $`Q`$ and NUT (gravitational monopole) charge $`l`$ is
$`ds^2=f^2(dt+\mathrm{\Omega }d\varphi )^2+f^2dr^2+(r^2+l^2)(d\theta ^2+\mathrm{sin}^2\theta d\varphi ^2),`$ (32)
$`𝑭=Q{\displaystyle \frac{r^2l^2}{(r^2+l^2)^2}}(dt+\mathrm{\Omega }d\varphi )dr+{\displaystyle \frac{2Qlr}{r^2+l^2}}\mathrm{sin}\theta d\theta d\varphi .`$ (33)
where
$`f^2={\displaystyle \frac{r^22Mr+Q^2l^2}{r^2+l^2}}={\displaystyle \frac{(rr_+)(rr_{})}{r^2+l^2}}`$ (34)
$`\mathrm{\Omega }=2l\mathrm{cos}\theta +\mathrm{\Omega }^{}.`$ (35)
Here we have defined
$$r_\pm =M\pm \mathrm{\Delta },\mathrm{\Delta }=\sqrt{M^2+l^2Q^2},$$
(36)
while $`\mathrm{\Omega }^{}`$ is a constant which should be set to $`2l`$ ($`2l`$) to make the half-axis $`\theta =0`$ ($`\theta =\pi `$) explicitly regular, leaving the other half-axis – the Misner string – singular, as $`d\varphi `$ is not a well-behaved one-form at $`\theta =0,\pi `$. However, it is well-known that both half-axes can be made regular, since changing $`\mathrm{\Omega }^{}`$ from $`2l`$ to $`2l`$ can be mimicked by changing the time coordinate from $`t`$ to $`t^{}=t4l\varphi `$. The price to pay is closed timelike curves since it requires that $`t`$ and $`t^{}`$ should both be periodic with period $`8\pi l`$.
The three-metric $`h_{ab}`$ for this solution is seen to be
$$dl^2=dr^2+(r^22Mr+Q^2l^2)d\mathrm{\Omega }^2=dr^2+(rr_+)(rr_{})d\mathrm{\Omega }^2$$
(37)
and the three remaining target space scalars are given by
$`\chi ={\displaystyle \frac{2l(rM)}{r^2+l^2}}`$ (38)
$`A={\displaystyle \frac{Q}{\sqrt{r^2+l^2}}}`$ (39)
$`S=\mathrm{arctan}\left({\displaystyle \frac{l}{r}}\right),`$ (40)
Since all the $`X^A`$ depend on $`r`$ only, this is a geodesic solution. In the non-extremal case $`\mathrm{\Delta }>0`$, the geodesic is spacelike ($`ϵ>0`$) and explicitly given by
$`f={\displaystyle \frac{\mathrm{\Delta }}{\sqrt{(M\mathrm{sinh}\overline{\omega }\mathrm{\Delta }\mathrm{cosh}\overline{\omega })^2+l^2\mathrm{sinh}^2\overline{\omega }}}}={\displaystyle \frac{r_+r_{}}{\sqrt{(r_{}e^{\overline{\omega }}r_+e^{\overline{\omega }})^2+l^2(e^{\overline{\omega }}e^{\overline{\omega }})^2}}}`$ (41)
$`\chi ={\displaystyle \frac{l\mathrm{\Delta }\mathrm{sinh}2\overline{\omega }}{(M\mathrm{sinh}\overline{\omega }\mathrm{\Delta }\mathrm{cosh}\overline{\omega })^2+l^2\mathrm{sinh}^2\overline{\omega }}}={\displaystyle \frac{2l\mathrm{\Delta }(e^{2\overline{\omega }}e^{2\overline{\omega }})}{(r_{}e^{\overline{\omega }}r_+e^{\overline{\omega }})^2+l^2(e^{\overline{\omega }}e^{\overline{\omega }})^2}}`$ (42)
$`A={\displaystyle \frac{Q\mathrm{sinh}\overline{\omega }}{\sqrt{(M\mathrm{sinh}\overline{\omega }\mathrm{\Delta }\mathrm{cosh}\overline{\omega })^2+l^2\mathrm{sinh}^2\overline{\omega }}}}={\displaystyle \frac{Q(e^{\overline{\omega }}e^{\overline{\omega }})}{\sqrt{(r_{}e^{\overline{\omega }}r_+e^{\overline{\omega }})^2+l^2(e^{\overline{\omega }}e^{\overline{\omega }})^2}}}`$ (43)
$`S=\mathrm{arctan}\left({\displaystyle \frac{l\mathrm{sinh}\overline{\omega }}{M\mathrm{sinh}\overline{\omega }\mathrm{\Delta }\mathrm{cosh}\overline{\omega }}}\right)=\mathrm{arctan}\left[{\displaystyle \frac{l(e^{\overline{\omega }}e^{\overline{\omega }})}{r_{}e^{\overline{\omega }}r_+e^{\overline{\omega }}}}\right],`$ (44)
where $`\overline{\omega }`$ is the arclength parameter for the geodesic, and here the function that satisfies the Laplace equation (24), namely
$$\overline{\omega }=arccoth\left(\frac{rM}{\mathrm{\Delta }}\right)=\frac{1}{2}\mathrm{ln}\left(\frac{rr_+}{rr_{}}\right).$$
(45)
However, we shall instead think of the geodesic as parametrized by the rescaled affine parameter
$$\omega =\frac{M}{\mathrm{\Delta }}\overline{\omega },$$
(46)
which implies that the norm of the geodesic tangent vector has norm $`ϵ=\mathrm{\Delta }/M`$. The reason for this is that we will then straightforwardly be able to treat the extremal case – which corresponds to a null geodesic ($`ϵ=0`$) – as the $`\mathrm{\Delta }0`$ limit of the non-extremal case. Indeed, taking this limit for the above formulae, we find that the spacelike geodesic goes over into a lightlike one;
$`f={\displaystyle \frac{M}{\sqrt{M^2(1\omega )^2+l^2\omega ^2}}}`$ (47)
$`\chi ={\displaystyle \frac{2lM\omega }{M^2(1\omega )^2+l^2\omega ^2}}`$ (48)
$`A={\displaystyle \frac{Q\omega }{\sqrt{M^2(1\omega )^2+l^2\omega ^2}}}`$ (49)
$`S=\mathrm{arctan}\left[{\displaystyle \frac{l\omega }{M(1\omega )}}\right],`$ (50)
while $`\omega `$ goes over into
$$\omega =\frac{M}{rM}.$$
(51)
This limiting procedure would not have worked, had we used $`\overline{\omega }`$ as the affine parameter. Moreover, $`\omega `$ has the large $`r`$ asymptotic behaviour
$$\omega =\frac{M}{r}+O(r^2),$$
(52)
which means that it is $`\omega `$, rather than $`\overline{\omega }`$, that corresponds to a Newtonian gravitational potential. However, although we will think of the geodesic as parametrized by $`\omega `$, we will often work with $`\overline{\omega }`$ in subsequent calculations, since the factor $`M/\mathrm{\Delta }`$ would otherwise often appear merely as an annoying appendage.
For future reference, we here finally calculate the conserved geodesic momentum $`p_\chi `$ according to the formula (31), to find the simple relation
$$p_\chi =\frac{l}{2M}.$$
(53)
### III.1 Charges
A special and somewhat surprising feature of the RNN solution is that although no sources are included for the electromagnetic field, the result of calculating the electric charge $`Q_{\mathrm{el}}`$ by integrating $`F_{ab}`$ over a two-sphere of constant $`t`$ and $`r`$ and dividing by $`4\pi `$ is not what one would naively expect from Stoke’s theorem, i.e. the result is not independent of the choice of two-sphere but depends on the radius;
$$Q_{\mathrm{el}}=Q\frac{r^2l^2}{r^2+l^2}.$$
(54)
We also calculate the magnetic charge $`Q_{\mathrm{mag}}`$ by integrating $`F_{ab}`$ over the same sphere to find
$$Q_{\mathrm{mag}}=\frac{2Qlr}{r^2+l^2}.$$
(55)
Obviously we have $`Q_{\mathrm{el}}=Q`$, $`Q_{\mathrm{mag}}=0`$ asymptotically as $`r\mathrm{}`$, so at infinity the solution is purely electric with charge $`Q`$. It is worth noting that the relation
$$Q_{\mathrm{el}}^{\mathrm{\hspace{0.17em}2}}+Q_{\mathrm{mag}}^{\mathrm{\hspace{0.17em}2}}=Q^2$$
(56)
holds as an identity, for all $`r`$. The reason why it is possible that electric and magnetic charges depend on $`r`$ is that the gravitomagnetic vector potential $`\mathrm{\Omega }d\varphi `$ enters not only the metric but also the electromagnetic field strength (33). Thus the latter is singular on the Misner string, i.e. on $`\theta =0`$ or $`\theta =\pi `$ (or both) depending on how $`\mathrm{\Omega }^{}`$ is chosen. As the Misner string goes through every two-sphere that we integrate over, we cannot expect Stoke’s theorem to imply that the charges should be independent of the choice of sphere. Interestingly, if we refrain from removing the singularity by introducing periodic time, the Misner string must be thought of as a concrete physical object as it carries electric and magnetic charge density.
### III.2 Transformation to Weyl coordinates
For the RNN solution, the transformation from Schwarzschild type coordinates to Weyl coordinates reads
$`\rho =\sqrt{(rM)^2\mathrm{\Delta }^2}\mathrm{sin}\theta `$ (57)
$`z=(rM)\mathrm{cos}\theta ,`$ (58)
with inverse transformation
$`r=\lambda +M`$ (59)
$`\mathrm{cos}\theta ={\displaystyle \frac{z}{\lambda }},`$ (60)
where
$$\lambda =\frac{1}{2}(\lambda _++\lambda _{}),\lambda _\pm =\sqrt{\rho ^2+(z\pm \mathrm{\Delta })^2}.$$
(61)
The functions $`\omega `$, $`k`$ and $`\mathrm{\Omega }`$ are now given by
$`\omega ={\displaystyle \frac{M}{2\mathrm{\Delta }}}\mathrm{ln}\left({\displaystyle \frac{\lambda \mathrm{\Delta }}{\lambda +\mathrm{\Delta }}}\right)`$ (62)
$`k={\displaystyle \frac{1}{2}}\mathrm{ln}\left({\displaystyle \frac{\lambda ^2\mathrm{\Delta }^2}{\lambda ^2\eta ^2}}\right),\eta ={\displaystyle \frac{1}{2}}(\lambda _+\lambda _{})`$ (63)
$`\mathrm{\Omega }={\displaystyle \frac{2lz}{\lambda }}+\mathrm{\Omega }^{}.`$ (64)
In the non-extremal case $`\mathrm{\Delta }>0`$, the function $`\omega `$ is the potential of an infinitely thin rod located at $`\rho =0`$, $`|z|\mathrm{\Delta }`$ and having a line density $`M/(2\mathrm{\Delta })`$ per unit length, which means that the line density can be interpreted as the mass line density. Note that in the extremal case $`\mathrm{\Delta }=0`$, the function $`k`$ vanishes identically leaving the three-metric $`h_{ab}`$ flat. Moreover $`\lambda `$ reduces to $`\sqrt{\rho ^2+z^2}`$ and $`\omega `$ to the standard monopole solution $`M/\lambda `$.
## IV Periodic analogue of the Reissner-Nordström-NUT solution
We will now construct compactified versions of the RNN solution, which generalize the compactified Schwarzschild solution discussed in Myers ; Korotkin1 ; Korotkin2 ; Frolov . The approach we take can be summerized as follows:
* For every member of the RNN family of solutions, compactify the function $`\omega `$ by using canonical Weyl coordinates and taking the $`z`$-periodic analogue of the original solution to the Laplace equation (28). This is straightforward due to the linearity of the latter.
* Insert $`\omega `$ into the *same* geodesic that defines the original RNN solution.
* Integrate eqs. (29) and (30) to find the remaining metric functions $`k`$ and $`\mathrm{\Omega }`$.
Before we start, some general remarks about the first of these steps are in order. As proved by Korotkin and Nicolai Korotkin1 ; Korotkin2 , if $`{}_{}{}^{0}\omega (\rho ,z)`$ is a solution to the Laplace equation with the asymptotic behaviour
$${}_{}{}^{0}\omega (\rho ,z)=\frac{M}{\stackrel{~}{r}}+O(\stackrel{~}{r}^2)\text{as }\stackrel{~}{r}\mathrm{},$$
(65)
where $`\stackrel{~}{r}=\sqrt{\rho ^2+z^2}`$ and $`M`$ is some constant (in our case the mass), then the series
$$\omega (\rho ,z)=\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\left[{}_{}{}^{0}\omega (\rho ,z+nL)+a_n\right],a_n=\{\begin{array}{cc}\frac{M}{L|n|}\hfill & \text{if }n0\hfill \\ 0\hfill & \text{if }n=0\hfill \end{array},$$
(66)
is convergent for any $`(\rho ,z)`$ such that $`(\rho ,z+nL)`$ does not coincide with a singular point of $`{}_{}{}^{0}\omega (\rho ,z)`$ for any integer $`n`$. The resulting function $`\omega `$ then obviously defines a $`z`$-periodic solution to the Laplace equation (28) with period $`L`$. The constants $`a_n`$ are essential as they make the series converge, but one cannot say that they are uniquely determined as one could add to $`a_n`$ any other $`n`$-dependent constant $`b_n`$ that falls off faster than $`|n|^1`$ as $`|n|\mathrm{}`$ so that
$$B:=\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}b_n$$
(67)
is a finite constant. In other words, the $`z`$-periodic analogue of the function $`{}_{}{}^{0}\omega (\rho ,z)`$ is a priori only defined up to an additive constant $`B`$. Korotkin and Nicolai fix this constant by taking eq. (66) as it stands ($`B=0`$) to define $`\omega (\rho ,z)`$, but this choice has not been made by the other workers that have studied the compactified Schwarzschild black hole; while MyersMyers makes an explicitly different choice of $`a_n`$ leading to a certain non-zero $`B`$, Frolov and FrolovFrolov instead use a Green’s function method to compactify which in effect corresponds to a third choice of $`B`$. Now, since the function $`\omega (\rho ,z)`$ is periodic in $`z`$, it does not make sense to study its behaviour for large $`\stackrel{~}{r}`$, but for large $`\rho `$ it has the asymptotic behaviour
$$\omega =\frac{2M}{L}\mathrm{ln}\rho +O(1).$$
(68)
If $`\omega `$ would tend to a constant in this limit, it would be natural to make that constant vanish by a suitable choice of $`B`$. However, since $`\omega `$ instead diverges logarithmically (unless $`M=0`$), we see no physical motivation for fixing $`B`$ at any particular value and hence we will keep it as a free parameter.
We will now implement the compactification scheme as outlined above. We thus take $`{}_{}{}^{0}\omega `$ to be the function given by eq. (62). According to the above, its periodic analogue is
$$\omega =\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\left[\frac{M}{2\mathrm{\Delta }}\mathrm{ln}\left(\frac{\lambda _n\mathrm{\Delta }}{\lambda _n+\mathrm{\Delta }}\right)+a_n\right]+B,$$
(69)
where
$`\lambda _n={\displaystyle \frac{1}{2}}\left[\sqrt{\rho ^2+(z_n+\mathrm{\Delta })^2}+\sqrt{\rho ^2+(z_n\mathrm{\Delta })^2}\right],z_n=znL`$ (70)
$`a_n=\{\begin{array}{cc}{\displaystyle \frac{M}{L|n|}},\hfill & \text{if }n0\hfill \\ 0\hfill & \text{if }n=0.\hfill \end{array}`$ (73)
We take $`z`$ to have the range $`z[L/2,L/2]`$ with the end points of the interval identified. The horizon, where $`\omega `$ diverges to minus infinity, will thus be located at $`\rho =0`$, $`|z|\mathrm{\Delta }`$, just as in the non-compactified case. Since we shall not consider the case when the horizon overlaps itself, we require that $`\mathrm{\Delta }L/2`$. It will prove useful to introduce the dimensionless variable
$$\beta =\frac{2\mathrm{\Delta }}{L},$$
(74)
having the range $`\beta [0,1]`$. For convenience, we will, for the time being, work with $`\overline{\omega }`$ and use the dimensionless coordinates
$$x=\frac{\rho }{\mathrm{\Delta }},y=\frac{z}{\mathrm{\Delta }}.$$
(75)
Clearly $`\overline{\omega }`$ is then given by
$$\overline{\omega }=\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\left[\frac{1}{2}\mathrm{ln}\left(\frac{\lambda _n^{}1}{\lambda _n^{}+1}\right)+\overline{a}_n\right]+\overline{B},$$
(76)
where
$`\lambda _n^{}={\displaystyle \frac{1}{2}}\left[\sqrt{x^2+(y_n+1)^2}+\sqrt{x^2+(y_n1)^2}\right],y_n=y2n\beta ^1`$ (77)
$`\overline{a}_n={\displaystyle \frac{\mathrm{\Delta }}{M}}a_n=\{\begin{array}{cc}{\displaystyle \frac{\beta }{2|n|}},\hfill & \text{if }n0\hfill \\ 0\hfill & \text{if }n=0,\hfill \end{array}`$ (80)
$`\overline{B}={\displaystyle \frac{\mathrm{\Delta }}{M}}B.`$ (81)
We shall now explicitly evaluate $`\overline{\omega }`$ close to the symmetry axis $`x=0`$. As the behaviour on the horizon section $`|y|1`$ of the axis is different from the off-horizon section $`1<|y|\beta ^1`$, these two sections will have to be treated differently, but in both cases we shall express the result in terms of the function
$$\phi _\beta (\xi )=\frac{\beta }{2\pi }\mathrm{sin}\left(\frac{\pi \beta \xi }{2}\right)\mathrm{\Gamma }\left(\frac{\beta \xi }{2}\right)^2,$$
(82)
having the properties
$`\underset{\beta 0}{lim}\phi _\beta (\xi )={\displaystyle \frac{1}{\xi }}`$ (83)
$`\phi _\beta (\xi )\phi _\beta (\xi )={\displaystyle \frac{1}{\xi ^2}}.`$ (84)
Moreover, for reasons that will become clear, we shall replace the free constant $`\overline{B}`$ by a free constant $`\overline{u}`$ according to
$$\overline{B}=\overline{u}\beta \gamma \frac{1}{2}\mathrm{ln}[2\phi _\beta (2)],$$
(85)
where $`\gamma `$ is the Euler constant. Now, for small $`x`$ and $`|y|1`$, we find that $`\overline{\omega }`$ is given by
$$e^{2\overline{\omega }}=\frac{e^{2\overline{u}}}{4H(y)}x^2+O(x^4),$$
(86)
where $`H(y)`$ is the function
$$H(y)=\frac{2\phi _\beta (2)}{\phi _\beta (1+y)\phi _\beta (1y)}.$$
(87)
For $`1<|y|\beta ^1`$, on the other hand, one finds that
$$e^{2\overline{\omega }}=\frac{e^{2\overline{u}}}{2\phi _\beta (2)}\frac{\phi _\beta (|y|+1)}{\phi _\beta (|y|1)}+O(x^2).$$
(88)
This is all that is needed to generate the whole solution. Indeed, for both axis sectors, $`\overline{\omega }`$ is a solution to the Laplace equation which can be expanded in terms of a constant $`C`$ and a function $`A(y)`$ as
$$\overline{\omega }=C\mathrm{ln}x+\underset{k=0}{\overset{\mathrm{}}{}}\frac{(1)^k}{(2^kk!)^2}\frac{d^{2k}A(y)}{dy^{2k}}x^{2k}.$$
(89)
Inserting this expansion into the right hand sides of eqs. (29) and (30) (using eq. (53) for the latter), we find that $`k`$ and $`\mathrm{\Omega }`$ are determined up to additive constants $`k^{}`$ and $`\mathrm{\Omega }^{}`$;
$`k=k^{}+C^2\mathrm{ln}x+2CA(y){\displaystyle \frac{1}{2}}\left[CA^{\prime \prime }(y)+A^{}(y)^2\right]x^2+O(x^4)`$ (90)
$`\mathrm{\Omega }=\mathrm{\Omega }^{}+2l\left[Cy+{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{2k(1)^k}{(2^kk!)^2}}{\displaystyle \frac{d^{2k1}A(y)}{dy^{2k1}}}x^{2k}\right],`$ (91)
where we refrain from giving the whole series for $`k`$ since it depends nonlinearly on $`\overline{\omega }`$, although it would not be difficult to write down at least a few more terms. The value of $`k^{}`$ has to be determined from the requirement that the axis be regular. Starting with the off-horizon section, regularity means absence of a conical singularity which is in turn means that $`k`$ should vanish as $`x0`$, at least if we assume that $`\mathrm{\Omega }`$ vanishes in that limit, which can always be achieved as we shall see below. Thus, since the constant $`C`$ is zero in this case, we directly obtain $`k^{}=0`$. For the horizon section we instead have $`C=1`$. In this case we may determine $`k^{}`$ by inspecting the horizon metric, which reads
$$ds_\mathrm{H}^2=R^2\left[\frac{e^{2(k^{}+\overline{u})}}{16}\frac{dy^2}{H(y)}+H(y)d\varphi ^2\right],$$
(92)
while $`R`$ is the constant
$$R=\sqrt{r_+^2+l^2}e^{\overline{u}}$$
(93)
Now, since
$$H^{}(\pm 1)=2,$$
(94)
it follows that for the horizon’s polar points $`y=\pm 1`$ to be regular, we must set
$$k^{}=\overline{u}+\mathrm{ln}4.$$
(95)
When we are on the horizon $`|y|1`$, the constant $`\mathrm{\Omega }^{}`$ in eq. (91) precisely corresponds to the constant $`\mathrm{\Omega }^{}`$ in eqs. (35) and (64) and should be set to $`2l`$ ($`2l`$) to make $`\mathrm{\Omega }`$ vanish at the horizon pole $`y=1`$ ($`y=1`$). Continuity at $`y=\pm 1`$ then requires us to set $`\mathrm{\Omega }^{}=0`$ for $`y>1`$ ($`y<1`$) and $`\mathrm{\Omega }^{}=4l`$ ($`\mathrm{\Omega }^{}=4l`$) for $`y<1`$ ($`y>1`$). Again, these two choices do not correspond to different physics, but are related by a change of time $`tt^{}=t4l\varphi `$ which results in time periodicity with the period $`8\pi l`$. However, note that unlike the non-periodic case, both choices imply that $`\mathrm{\Omega }`$ has a jump discontinuity where $`y=\beta ^1`$ is periodically identified with $`y=\beta ^1`$. This discontinuity is of course not of a physical nature either, since we can shift it to any other position on the $`y`$-axis. For instance, we may put the discontinuity somewhere on the horizon and obtain a completely regular off-horizon section on which $`\mathrm{\Omega }`$ is everywhere zero.
### IV.1 Fourier expansions
In principle, the full RNN solution is given as the $`x=0`$ expansions (89) - (91), but since these series converge very slowly for large $`x`$ and are quite useless when it comes to determining the asymptotic behaviour as $`x\mathrm{}`$, we shall now follow Frolov and Frolov and represent the solution in terms of fourier series, thus making explicit use of the periodicity of the coordiante $`y`$. Still using the rescaled coordinates $`(x,y)`$, we assume fourier expansions for the functions $`\overline{\omega }`$ and $`k`$ of the forms
$`\overline{\omega }={\displaystyle \underset{q=0}{\overset{\mathrm{}}{}}}\overline{\omega }_q(x)\mathrm{cos}(\pi q\beta y)`$ (96)
$`k={\displaystyle \underset{q=0}{\overset{\mathrm{}}{}}}k_q(x)\mathrm{cos}(\pi q\beta y)`$ (97)
Since $`\overline{\omega }`$ is a solution to the Laplace equation (28), it follows that
$$\overline{\omega }_q^{\prime \prime }(x)+x^1\overline{\omega }_q^{}(x)(\pi q\beta )^2\overline{\omega }_q(x)=0.$$
(98)
Disqualifying solutions that diverges exponentially as $`x\mathrm{}`$, we obtain that $`\overline{\omega }_0(x)`$ is a linear function of $`\mathrm{ln}x`$ while, for $`q>0`$,
$$\overline{\omega }_q(x)K_0(\pi q\beta x),$$
(99)
where $`K_0`$ is the modified Bessel function of the second kind. For small $`x`$, we have
$$K_0(\pi q\beta x)=\gamma +\mathrm{ln}\left(\frac{\pi q\beta x}{2}\right)+O(x^2).$$
(100)
We can now use that we already know $`\overline{\omega }`$ for small $`x`$ to find that
$`\overline{\omega }_0(x)=\overline{u}{\displaystyle \frac{1}{2}}\mathrm{ln}[2\phi _\beta (2)]+\beta \mathrm{ln}\left({\displaystyle \frac{\beta x}{4}}\right)`$ (101)
$`\overline{\omega }_q(x)={\displaystyle \frac{2}{\pi q}}\mathrm{sin}(\pi q\beta )K_0(\pi q\beta x)\text{for }q>0.`$ (102)
We use this fourier expansion for $`\overline{\omega }`$ to plot the metric function $`f`$ for two choices of the parameters $`M`$, $`Q`$ and $`l`$ (figure 1).
Due to the nonlinearity of eqs. (29) which governs $`k`$, we shall not attempt to give its full fourier series. However, since all $`k_q(x)`$ with $`q>0`$ decay exponentially as $`x\mathrm{}`$, it will suffice to determine $`k_0(x)`$ to obtain the asymptotic behaviour in that limit. Now, the zeroth order fourier term of the first of eqs. (29) reads
$$k_0^{}(x)=x\left[(\overline{\omega }_{,x})^2(\overline{\omega }_{,z})^2\right]_0=x\left\{\overline{\omega }_0^{}(x)^2+\frac{1}{2}\underset{q=1}{\overset{\mathrm{}}{}}\left[\overline{\omega }_q^{}(x)^2(\pi q\beta )^2\overline{\omega }_q(x)^2\right]\right\}$$
(103)
Inserting the determined expressions for the $`\overline{\omega }_q(x)`$ gives
$$k_0^{}(x)=\beta ^2x\left\{\frac{1}{x^2}+2\underset{q=1}{\overset{\mathrm{}}{}}\mathrm{sin}^2(\pi q\beta )\left[K_1(\pi q\beta x)^2K_0(\pi q\beta x)^2\right]\right\},$$
(104)
which can be integrated to yield
$$k_0(x)=k_{}+\beta ^2\left\{\mathrm{ln}x+x^2\underset{q=1}{\overset{\mathrm{}}{}}\mathrm{sin}^2(\pi q\beta )\left[2K_1(\pi q\beta x)^2K_0(\pi q\beta x)K_2(\pi q\beta x)K_0(\pi q\beta x)^2\right]\right\},$$
(105)
where $`k_{}`$ is some constant yet to be determined. Using now that
$$x^2\left[2K_1(\pi k\beta x)^2K_0(\pi k\beta x)K_2(\pi k\beta x)K_0(\pi k\beta x)^2\right]=\frac{2}{(\pi k\beta )^2}\left[1+\gamma +\mathrm{ln}\left(\frac{\pi k\beta x}{2}\right)\right]+O(x^2),$$
(106)
we can compare $`k_0(x)`$ as given by eq. (105) to the zeroth order fourier term of the exact expression for $`k`$ we have near $`x=0`$ to find, after some manipulations, that $`k_{}`$ is the $`\beta `$-dependent constant
$$k_{}=\beta ^2\left[1+\mathrm{ln}\left(\frac{\beta }{4}\right)\right]\beta \mathrm{ln}[\phi _\beta (2)]+_0^\beta \mathrm{ln}[\phi _\beta ^{}(2)]𝑑\beta ^{}.$$
(107)
Finally, from eqs. (30) and (53) we obtain that $`\mathrm{\Omega }`$ is given by the following “quasi-fourier” series:
$`\mathrm{\Omega }=\mathrm{\Omega }^{}+2l\beta y+4l{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\pi k}}\mathrm{sin}\left(\pi k\beta \right)xK_1\left(\pi k\beta x\right)\mathrm{sin}\left(\pi k\beta y\right),`$ (108)
where $`\mathrm{\Omega }^{}`$ is the same constant that appears in eq. (91) for the case $`|y|1`$.
### IV.2 Properties of the event horizon
As noted above, the (outer) event horizon has the geometry
$$ds_\mathrm{H}^2=^2\left[\frac{dy^2}{H(y)}+H(y)d\varphi ^2\right],=e^{\overline{u}}\sqrt{r_+^2+l^2}.$$
(109)
It directly follows that the horizon area and Gaussian curvature (half the Ricci scalar) are given by the simple formulae
$`𝒜=4\pi ^2`$ (110)
$`K={\displaystyle \frac{1}{2}}H^{\prime \prime }(y)^2.`$ (111)
It is worth noting that $`H(y)`$ behaves for small $`\beta `$ as
$$H(y)=1y^2\frac{1}{2}\zeta (3)(1y^2)^2\beta ^3+O(\beta ^5),$$
(112)
while for $`\beta `$ close to unity it behaves as
$$H(y)=4(1\beta )+4\left[\mathrm{\Psi }\left(\frac{1+y}{2}\right)+\mathrm{\Psi }\left(\frac{1y}{2}\right)+2\gamma +1\right](1\beta )^2+O((1\beta )^3),$$
(113)
where $`\mathrm{\Psi }`$ is the digamma function (the logarithmic derivative of the gamma function). Clearly, for $`\beta =0`$, the metric is an exact two-sphere of radius $``$. Note that $`\beta 0`$ corresponds to two independent and physically different limits, namely the extremal limit $`\mathrm{\Delta }0`$ as well as the limit of infinite coordinate diameter $`L\mathrm{}`$. This is natural since both limits, makes the *proper* distance between the poles of the black hole infinite, keeping the black hole from being distorted “by itself”, i.e. by the gravitational field from its periodic copies. As $`\beta `$ is increased, the sphere becomes deformed in a prolate manner. For small $`\beta `$ the deformation is very small, however, as the first correction term in eq. (112) enters at order $`\beta ^3`$. As $`\beta `$ is further increased the deformed sphere gets stretched out to a long cigar, and for $`\beta `$ close to unity the geometry is almost everywhere a flat cylinder of length $`/\sqrt{1\beta }`$ and circumference $`4\pi \sqrt{1\beta }`$, except at the poles $`y=\pm 1`$ where the curvature is very high (note that the mean Gaussian curvature is always $`^2`$). Up to a constant conformal factor the geometry only depends on the parameter $`\beta `$ and hence we can refer the reader to Frolov and FrolovFrolov for a further discussion of its properties, as well as embedding diagrams. Now, the horizon area $`𝒜`$ is clearly modified by a factor $`e^{2\overline{u}}`$ compared to the non-compactified case. This is in agreement with the general treatment of distorted black holes with (electric) charge as presented by Fairhurst and KrishnanFairhurst:2000xh , based on Geroch and Hartle’s treatment of the case without chargeGeroch . Although it may not have been obvious from the way we introduced it, the parameter $`\overline{u}`$ (denoted the same way in Fairhurst:2000xh but with an unbarred $`u`$ in Geroch and Frolov ) can be defined as
$$\overline{u}=\delta \overline{\omega }|_{\rho =0,z=\pm \mathrm{\Delta }},\delta \overline{\omega }=\overline{\omega }{}_{}{}^{0}\overline{\omega },$$
(114)
i.e. $`\overline{u}`$ is the value of $`\delta \overline{\omega }`$ evaluated at either of the poles of the event horizon, with $`\delta \overline{\omega }`$ being the difference between the function $`\overline{\omega }`$ for the deformed black hole (whether the deformation is due to compactification as in our case, or to external matter) and the function $`{}_{}{}^{0}\overline{\omega }`$ for the undeformed RNN solution with the same $`M`$, $`Q`$ and $`l`$. Since as $`\overline{\omega }\mathrm{}`$, we have
$$f=\frac{2\mathrm{\Delta }}{\sqrt{r_+^2+l^2}}e^{\overline{\omega }}+O(e^{3\overline{\omega }}),$$
(115)
it follows that $`\overline{u}`$ is closely related to the change in horizon pole redshift factor that the deformation produces. In the Schwarzschild case $`Q=l=0`$, the relation is more direct since $`f=e^{\overline{\omega }}`$ in that case. The term “horizon redshift factor” must not be taken too literally, however, since it normally refers to an asymptotically flat situation when the value of $`f^1`$ gives the redshift factor between the point of evaluation and infinity where $`f=1`$. Anyhow, we shall refer to $`\overline{u}`$ as the *redshift parameter*. In table 1 we display how $`\overline{u}`$ has in effect been fixed by previous workers that have studied the compactified Schwarzschild solution. As we see no reason why any one of these choices should be better than the other two, we prefer to keep $`\overline{u}`$ as a free parameter.
Other quantities that are interesting to calculate on the horizon are the surface gravity $`\kappa `$ and Komar mass $`M_{\mathrm{Komar}}`$ with respect to the Killing vector $`\xi ^a`$, as well as the electric and magnetic charges and potentials. We find that these quantities are given by
$`\kappa ={\displaystyle \frac{\mathrm{\Delta }}{^2}}`$ (116)
$`M_{\mathrm{Komar}}=\mathrm{\Delta }`$ (117)
$`Q_{\mathrm{el}}=Q{\displaystyle \frac{r_+^2l^2}{r_+^2+l^2}}`$ (118)
$`Q_{\mathrm{mag}}={\displaystyle \frac{2Qlr_+}{r_+^2+l^2}}`$ (119)
$`v={\displaystyle \frac{Qr_+}{r_+^2+l^2}}`$ (120)
$`u={\displaystyle \frac{Ql}{r_+^2+l^2}}`$ (121)
We note that the surface gravity is constant over the event horizon and thus the zeroth law of thermodynamics holds. Moreover the electric and magnetic potentials are constant as well. In fact the electromagnetic charges and potentials take the exact same values as in the noncompactified case. For a discussion of further thermodynamic properties, we refer the reader to the general framwork for distorted charged black holesFairhurst:2000xh ; Yazadjiev:2000by . Unfortunately, to our knowledge no such framework exists in the case of nonzero NUT charge, but providing it here would be beyond the scope of this paper.
### IV.3 Proper distance between black hole poles
A quantity that characterizes the off-horizon section of the symmetry axis is the proper spatial separation between the poles of the black hole. It can be calculated as
$$L_{\mathrm{sep}}=2\mathrm{\Delta }_1^{\beta ^1}f^1𝑑y,$$
(122)
with $`f`$ evaluated at $`x=0`$. To get a general idea of how this quantity depends on the choice of charges $`M`$, $`Q`$ and $`l`$, we have held the quotients $`Q/M`$ and $`l/M`$ fixed at four different values and, with the redshift parameter $`\overline{u}`$ set to zero, plotted $`L_{\mathrm{sep}}/L`$ as a function of $`\beta `$ (figure 2). Since $`L_{\mathrm{sep}}`$ depends nontrivially on how one chooses $`\overline{u}`$ to depend on $`\beta `$, our plot for the case $`Q=l=0`$ looks different than the one given by Frolov and Frolov Frolov . In particular, in Frolov $`L_{\mathrm{sep}}`$ tends to a finite value rather than zero as $`\beta 1`$, which is counter-intuitive since in that limit the polar points reach each other as the event horizon fills the whole axis. Again, this is directly related to the behaviour of $`\overline{u}`$ in the same limit. However, any choice of $`\overline{u}`$ which stays *finite* results in a vanishing $`L_{\mathrm{sep}}`$ as $`\beta 1`$, just like for $`\overline{u}0`$. To show the effect of different choices of $`\overline{u}`$, we have also plotted $`L_{\mathrm{sep}}`$ as a function of $`\beta `$ for the different choices of $`\overline{u}`$ collected in table 1, with the choice $`\overline{u}0`$ again included for comparison (figure 3).
### IV.4 Large distance asymptotics
In the limit $`\rho \mathrm{}`$, our compactified black hole solution approaches an exact solution corresponding to the field of a homogeneous line mass with electric as well as NUT line charge. This solution is obtained simply by truncating all terms that fall off exponentially with $`\rho `$, which is easy to do given the fourier expansions derived in subsection IV.1. Explicitly,
$`\overline{\omega }=\overline{u}{\displaystyle \frac{1}{2}}\mathrm{ln}[2\phi _\beta (2)]+\beta \mathrm{ln}\left({\displaystyle \frac{\rho }{2L}}\right)`$ (123)
$`k=k_{}+\beta ^2\mathrm{ln}\left({\displaystyle \frac{\rho }{\mathrm{\Delta }}}\right)`$ (124)
$`\mathrm{\Omega }={\displaystyle \frac{4lz}{L}}+\mathrm{\Omega }^{},`$ (125)
so the spacetime metric for our compactified Reisser-Nordström-NUT solution, in the nonextremal case $`\mathrm{\Delta }>0`$, rapidly approaches
$$ds^2=f^2(dt+\mathrm{\Omega }d\varphi )^2+f^2\left[\stackrel{~}{C}^2(\rho /\rho _0)^{2\beta ^2}(d\rho ^2+dz^2)+\rho ^2d\varphi ^2\right],$$
(126)
where
$$f^2=\frac{(r_+r_{})^2}{\left[r_{}(\rho /\rho _0)^\beta r_+(\rho /\rho _0)^\beta \right]^2+l^2\left[(\rho /\rho _0)^\beta (\rho /\rho _0)^\beta \right]^2}$$
(127)
As before, $`r_\pm =M\pm \mathrm{\Delta }`$ with $`\mathrm{\Delta }=\sqrt{M^2+l^2Q^2}`$, but we have here also introduced the constants
$`\rho _0=\left[2\phi _\beta (2)e^{2\overline{u}}\right]^{\frac{1}{2\beta }}2L`$ (128)
$`\stackrel{~}{C}=\left[{\displaystyle \frac{\phi _\beta (2)e^{2\overline{u}}}{2}}\right]^{\frac{\beta }{2}}e^{\beta ^2+_0^\beta \mathrm{ln}[\phi _\beta ^{}(2)]𝑑\beta ^{}}.`$ (129)
If we set both the electric charge $`Q`$ and NUT charge $`l`$ to zero, the function $`f^2`$ simplifies to
$$f^2=\left(\frac{\rho }{\rho _0}\right)^{2\beta },$$
(130)
which shows that this is a special case of Levi-Civita’s static cylindrically symmetric vacuum solution. As discussed for instance by Bic̆ák et al. Bicak:2004fw , the Levi-Civita metric contains two essential parameters $`m`$ and $`𝒞`$ and can be written in the standard form
$$ds^2=\stackrel{ˇ}{\rho }^{2m}d\stackrel{ˇ}{t}^2+\stackrel{ˇ}{\rho }^{2m(m1)}(d\stackrel{ˇ}{\rho }^2+d\stackrel{ˇ}{z}^2)+\frac{1}{𝒞^2}\stackrel{ˇ}{\rho }^{2(1m)}d\varphi ^2,$$
(131)
where the coordinates $`\stackrel{ˇ}{t}`$, $`\stackrel{ˇ}{\rho }`$ and $`\stackrel{ˇ}{z}`$ as well as the *conicity parameter* $`𝒞`$ in general carry dimension in a nonstandard way. The conicity parameter can obviously be set to unity by rescaling the coordinate $`\varphi `$ as $`\varphi 𝒞\varphi `$, but since we insist that $`\varphi `$ be a periodic coordinate with the standard range $`\varphi [0,2\varphi )`$, the value of $`𝒞`$ cannot be viewed as a coordinate choice. In fact, since we are dealing with a compactified spacetime for which $`z`$ is also a periodic coordinate with a fixed range, the general static cylindrically symmetric solution – what one might call the *compactified Levi-Civita solution* – has instead three essential parameters, which for instance can be taken to be our $`\beta `$, $`\rho _0`$ and $`\stackrel{~}{C}`$, with $`\beta `$ directly corresponding to $`m`$.
Setting the electric charge $`Q`$ to zero while having a non-vanishing NUT charge $`l`$ makes it possible to reexpress $`f^2`$ as
$$f^2=\frac{2\mathrm{\Delta }}{r_+(\rho /\rho _0)^{2\beta }r_{}(\rho /\rho _0)^{2\beta }}=\frac{\mathrm{\Delta }}{l}\frac{1}{\mathrm{cosh}[2\beta \mathrm{ln}(\rho /\rho _1)]},$$
(132)
where (note that $`r_{}<0`$ in this case)
$$\rho _1=\left(\frac{r_{}}{r_+}\right)^{\frac{1}{4\beta }}\rho _0.$$
(133)
As should be expected, this vacuum solution is the “cylindrical analogue of NUT space” that was constructed and studied by Nouri-ZonozNouri-Zonoz:1997ms . According to Nouri-Zonoz, the “gravitomagnetic charge per unit length” is (up to sign) equal to the coefficient of $`z`$ in the expression for $`\mathrm{\Omega }`$, i.e. in our case $`4l/L`$. However, what we have constructed here is the asymptotic field of a black hole with NUT charge $`l`$ in a space with a compactified dimension of ($`z`$-coordinate) length $`L`$, so we expect the NUT charge per unit $`z`$-length to be $`l/L`$. We believe that the discrepancy of a factor of $`4`$ is only a matter of definitions (i.e. Nouri-Zonoz’ “gravitomagnetic charge” is four times NUT charge, when the latter is definied so that its value for the RNN metric (32) is $`l`$.
Setting $`l=0`$ but keeping a non-zero electric charge $`Q`$, the general asymptotic solution (126) instead becomes the cylindrically symmetric Einstein-Maxwell solution with a purely electric Maxwell field. This is one of the solutions constructed and studied by Richterek et al. Richterek:2000dh as well as Miguelote et al. Miguelote:2000qi . The function $`f^2`$ can in this case be written as
$$f^2=\frac{\mathrm{\Delta }^2}{Q^2}\frac{1}{\mathrm{sinh}^2[\beta \mathrm{ln}(\rho /\rho _2)]},\rho _2=\left(\frac{r_+}{r_{}}\right)^{\frac{1}{2\beta }}\rho _0.$$
(134)
Note that $`f^2`$ diverges at a finite cylinder radius $`\rho =\rho _2`$, indicating that there should be a curvature singularity there. This is indeed the case, as noticed in Richterek:2000dh ; Miguelote:2000qi , but let us see this explicitly by calculating the energy density of the electromagnetic field with respect to the unit timelike vector field $`u^a=f^1\xi ^a`$ where $`\xi ^a=(/t)^a`$, as before. The result, in the general case with all three charges $`M`$, $`Q`$ and $`l`$ arbitrary, is
$$\mu _{\mathrm{EM}}=u^au^bT_{ab}=\frac{1}{2}u^au^bG_{ab}=2\left(\frac{Q}{\stackrel{~}{C}\rho _0L}\right)^2(\rho /\rho _0)^{2(1+\beta ^2)}f^4.$$
(135)
Clearly $`\mu _{\mathrm{EM}}`$ diverges at finite $`\rho `$ exactly when $`f^2`$ does. However, the curvature singularity at $`\rho =\rho _2`$ disappears as soon as a NUT charge $`l`$ is turned on, since the denominator of the right hand side of eq. (127) is then the sum of two squares which cannot vanish simultaneously. Of course, one can make the curvature grow arbitrarily large close to $`\rho =\rho _2`$ by choosing the quotient $`l/Q`$ sufficiently small. This feature is not only present in the asymptotic solution discussed here, but also in the full compactified RNN solution with the $`z`$-dependent fourier terms turned back on. The reason for this is that the range of the function $`\omega `$ changes from $`\omega (\mathrm{},0)`$ to $`\omega (\mathrm{},\mathrm{})`$ when we compactify the spacetime by making the coordinate $`z`$ periodic; In both cases $`\omega \mathrm{}`$ occurs at the event horizon, i.e. at $`\rho =0`$, $`|z|\mathrm{\Delta }`$, but $`\rho \mathrm{}`$ sends $`\omega `$ to $`+\mathrm{}`$ rather than zero in the compactified case. In particular, for $`Q0=l`$, the compactified $`\overline{\omega }`$ will take on the positive value
$$\overline{\omega }_{\mathrm{crit}}=\frac{1}{2}\mathrm{ln}\left(\frac{r_+}{r_{}}\right),$$
(136)
for which $`f^2`$ diverges, as can be seen from eq. (41). Again, this gives a curvature singularity which however is regularized by turning on a NUT charge $`l`$, however small. The difference, compared to the asymptotic solution, is that the curvature singularity now occurs on some torus $`\rho =f(z)`$ which is not a simple flat torus $`\rho =\mathrm{constant}`$. The asymptotic solution also has a curvature singularity on the axis $`\rho =0`$ for $`\beta (0,1)`$, but this is in contrast to the full compactified RNN solution which by construction is regular on that axis. Another interesting feature related to the NUT parameter $`l`$ is that although the spacetime metric (126) is cylindrically symmetric in the sense that no physically measurable quantities depend on the $`z`$, this coordiate still appears in the expression (125) for $`\mathrm{\Omega }`$. What is interesting is that if we demand that the metric be explicitly periodic in $`z`$ with period $`L`$, then the metric must be unchanged under $`\mathrm{\Omega }^{}\mathrm{\Omega }^{}+4nl`$ for any integer $`n`$. This is true, granted that $`t`$ is periodic with period $`8\pi l`$ (just like for the non-compactified RNN solution we started out with), since we can mimick the shift in $`\mathrm{\Omega }^{}`$ by letting $`tt+4nl\varphi `$.
Turning finally to the extremal case $`\mathrm{\Delta }=0`$, we must use $`\omega =M\overline{\omega }/\mathrm{\Delta }`$ instead of $`\overline{\omega }`$ before taking the limit $`\mathrm{\Delta }0`$. The limit then gives
$`\omega ={\displaystyle \frac{2M}{L}}\left[u^{}+\gamma +\mathrm{ln}\left({\displaystyle \frac{\rho }{2L}}\right)\right],u^{}=\underset{\beta 0}{lim}\beta ^1\overline{u}`$ (137)
$`k=0`$ (138)
$`\mathrm{\Omega }={\displaystyle \frac{4lz}{L}}+\mathrm{\Omega }^{},`$ (139)
so the asymptotic spacetime metric is in this case
$$ds^2=f^2(dt+\mathrm{\Omega }d\varphi )^2+f^2\left(d\rho ^2+dz^2+\rho ^2d\varphi ^2\right),$$
(140)
with, using eq. (47),
$$f^2=\frac{1}{\left[1{\displaystyle \frac{2M}{L}}\mathrm{ln}(\rho /\rho _3)\right]^2+\left[{\displaystyle \frac{2l}{L}}\mathrm{ln}(\rho /\rho _3)\right]^2},\rho _3=2Le^{u^{}\gamma }.$$
(141)
Just like in the nonextremal case, this asymptotic solution fails to be asymptotically flat, albeit in a weaker manner. Also, the discussion concerning the curvature singularity at finite $`\rho `$ applies here as well. In particular, from eq. (47) it follows that if $`l=0`$, the asymptotic solution has a curvature singularity at $`\rho =\rho _3e^{L/(2M)}`$, but the singularity is removed when the NUT charge $`l`$ is turned back on.
## V Conclusions
Having constructed and analyzed what we consider to be the natural periodic analogue of the Reissner-Nordström-NUT solution, a question that naturally arises is whether or not any given asymptotically flat black hole has a periodic analogue and, if so, whether or not the periodic analogue is unique, given the period $`L`$ of the compactified coordinate. By identifying the target space geodesic that the non-compactified solution defines and assuming that the very same geodesic should be used for the compactified version, we have found that the problem essentially reduces to finding the appropriate axisymmetric solution to the Laplace equation in flat space, just as in the Schwarzschild case. There is thus a simple superposition principle at play, which makes compactification tractable when Weyl coordinates are used. However, the low dimension requires a regularization procedure which introduces a one-parameter ambiguity as there is no obvious way to fix the freedom to add an arbitrary constant to the solution to the Laplace equation. In all previous works on the periodic Schwarzschild solution, this constant – the redshift parameter – was always fixed at an early stage and without any physical motivation for the specific choice made. As we have seen there are three different choices in the existing papers on the subject, which we take as support for instead keeping the parameter free, or at least fixing it at a later stage from physical considerations. One way of fixing it would be to require that a set of relations that hold for horizon quantities (for instance the relation between the horizon area and the black hole charges) should remain the same after compactification. If one could motivate why this should be the case, our method would give a neat one-to-one correspondance between standard and periodic Reissner-Nordström-NUT black holes. Of course, we have not given any proof that our method is correct, in some suitably defined sense. One could perhaps argue that it is not, based on the fact that our solution family has a curvature singularity surrounding the horizon in the case of electric but no NUT charge. However, we see no way of avoiding such a singularity. Indeed, the singularity in question is actually a general feature not just for our full solution with a nontrivial dependence on the periodic coordinate, but also for the cylindrically symmetric solution that is approached for large cylinder radii. Since there simply is no other cylindrically symmetric solution available that could work as an asymptotic solution for a compactified Reissner-Nordström black hole, it seems at least likely that the exponentially decaying corrections to the asymptotic solution cannot smooth out the singularity.
One can think of several ways to extend our work while staying in four dimensions. One way would be to include a dilaton field with arbitrary coupling parameter. This would be straightforward to do, as our method would still be applicable with the dilaton simply entering as an extra target space coordinate. It would be interesting to see whether the inclusion of a dilaton could prevent the above-mentioned singularity from occuring. A general framework for distorted charged dilaton black holes has been provided by YazadjievYazadjiev:2000by .
It would be even more interesting if the periodic analogue of the Kerr solution could be constructed. Some indications of how this could be done were given in Korotkin2 , but no explicit formulae were given there, except for a few first steps. Since the noncompactifed Kerr solution traces out a target space two-surface rather than a geodesic, it would no longer be possible to linearize the problem and thus more sophisticated methods, such as Bäcklund transformations, are needed.
It would of course be even more interesting if our work could give some insights into how things work in higher dimensions. Consider the Einstein-Maxwell equations for a static $`(d+1)`$-dimensional spacetime. Dimensionally reducing with respect to the static Killing vector, one arrives at Einstein gravity in $`d`$ euclidean dimensions, coupled to a two-dimensional (three-dimensional if a dilaton is included) sigma model. One question one could ask is whether or not the target space geodesic that the $`(d+1)`$-dimensional Reissner-Nordström solution corresponds to would also be traced out when the solution is compactified on a circle. The work of Harmark and Obers actually suggests that this could be the case, as their ansatz is such that the target space coordinates are functionally dependent, thus corresponding to geodesic solutions. This does not mean that it would be easy to make analytical progress for $`d>3`$, but nevertheless it could give a coordinate invariant way of thinking about the ansatz of these authors which in turn could lead to a deeper understanding of the general problem. It would be very interesting to investigate this matter further.
## Acknowledgements
The work of RvU was supported by the Ministry of Education of the Czech Republic under the project MSM 0021622409. The work of MK was supported by a post doctoral fellowship of the Faculty of Science of Masaryk University.
|
warning/0506/cond-mat0506294.html
|
ar5iv
|
text
|
# Universal Scaling Relations in Strongly Anisotropic Materials
## Abstract
We consider the critical temperature in strongly anisotropic antiferromagnetic materials, with weak coupling between stacked planes, in order to determine the interplane coupling constant from experimentally measured susceptibilities. We present theoretical arguments for a universal relation between interplane coupling and susceptibility shown numerically by Yasuda et. al., Phys. Rev. Lett. 94, 217201 (2005). We predict a more general scaling function if the system is close to a quantum critical point, a similar relation for other susceptibilities than considered in Yasuda et. al., and the validity of these relations for more general phase transitions.
Many materials display at low temperatures strongly spatially anisotropic responses to magnetic or electronic probes. This fact has motivated the theoretical study of low dimensional quantum systems on their own right. Solving one- or two-dimensional quantum systems can be useful to understand intermediary regimes of temperature in which fluctuations are dominated by the subsystem of lower dimensionality. Three-dimensionality is effectively restored once the temperature is lowered below the lowest energy scale characterizing the anisotropy.
For magnetic systems this scale can be the temperature $`1/\beta _{AF}^{}`$ ($`k_B^{}=\mathrm{}=1`$) below which antiferromagnetic (AF) long-range order manifests itself. In this context, one of the most studied model is perhaps a stacking in three dimensions of chains or square lattices on each of which a nearest-neighbor quantum spin-$`S`$ Heisenberg model $`H_J^{}`$ with AF exchange coupling $`J>0`$ is defined. To model a strong spatial anisotropy, one assumes that there exists a nearest-neighbor AF exchange coupling $`J^{}`$ in the directions transverse to the chains or planes that is much weaker than $`J`$, $`JJ^{}>0`$. The three-dimensional quantum Hamiltonian is $`H_{3d}^{}`$. Many efforts have been invested for the last 30 years in calculating the $`J^{}`$–dependence of $`1/\beta _{AF}^{}`$Scalapino75 ; Schulz96 ; Irkhin97 ; Bocquet02 ; Yasuda05 . In this letter we shall address the question: are there some universal relations that relate $`J^{}`$ and some observables of $`H_J^{}`$ or $`H_{3d}^{}`$?
The motivation for this question comes from the work by Yasuda et al. in Ref. Yasuda05, in which the Néel temperature $`1/\beta _{AF}^{}`$ of $`H_{3d}^{}`$, the $`n`$-dimensional static staggered susceptibility $`\chi _s^{(n)}=\chi ^{zz}(𝑸,\omega =0;\beta _{AF}^{})`$ of $`H_J^{}`$ where $`𝑸=\pi `$ if $`n=1`$ or $`𝑸=(\pi ,\pi )`$ if $`n=2`$, and $`J^{}\chi _s^{(n)}=1/\zeta _n^{}(J^{})`$ were computed numerically as a function of $`0<J^{}/J1`$. In quasi-two-dimension, it was observed that
$`J^{}\chi _s^{(n=2)}=1/\zeta _{n=2}^{}`$ (1)
becomes independent of $`J^{}/J`$ when $`J^{}/J<0.1`$ and that the constant value that it takes, although not 1/2 as predicted by mean-field theory,Scalapino75 ; Schulz96 is independent of the magnitude of the spin and even takes the same value for a classical ($`S=\mathrm{}`$) model. Although the evidence is less pronounced the same conclusion was reached in quasi-one-dimension.
We want to construct a tractable model that reproduces qualitatively these findings and we want to understand how these results can be useful to establish experimentally the implied universality. We present a theoretical argument that as $`J^{}/J0`$, the function $`\zeta _{n=2}`$ converges to a constant. We make the following additional predictions.
First, we consider more general AF models in the plane, and we consider the case in which, by tuning parameters, it is possible to tune the planar model close to a quantum phase transition, so that the zero-temperature AF order of the two-dimensional model (without interplane couplings) becomes small. Then, we predict the scaling function
$`J^{}\chi _s^{(2)}=F_1^{}(c\beta _{AF}^{}/\xi ^{(2)}),`$ (2)
for some scaling function $`F_1^{}`$, in the limit $`J^{}/J0`$, where $`\xi ^{(2)}`$ is the correlation length of the two-dimensional model at temperature $`1/\beta _{AF}^{}`$ and $`c`$ is a spin-wave velocity defined below. Note that in the system considered by Yasuda et. al. the planar model is in the renormalized classical regime so that $`c\beta _{AF}^{}/\xi ^{(2)}`$ is exponentially small in $`c\beta _{AF}^{}`$ and converges to zero as $`J^{}/J0`$. Therefore, $`F_1^{}(c\beta _{AF}^{}/\xi ^{(2)}=0)=1/\zeta _2^{}`$.
Second, we predict a similar scaling relation that will be valid for quantities which are easier to access experimentally. The susceptibility $`\chi _s^{(2)}`$ defined above is that of the two-dimensional model without the interlayer couplings, and cannot be measured in most real materials. We define $`\chi _{\pi ,\pi ,0}^{(3)}=\chi ^{zz}(\pi ,\pi ,0,\omega =0;\beta _{AF}^{})`$ to be the static susceptibility in the layered system at wave vector $`(\pi ,\pi )`$ in the plane and wave vector $`0`$ perpendicular to the plane at temperature $`1/\beta _{AF}^{}`$. Then, we predict that
$`J^{}\chi _{\pi ,\pi ,0}^{(3)}=F_2^{}(c^{(3)}\beta _{AF}^{}/\xi _{\pi ,\pi ,0}^{(3)}),`$ (3)
for some scaling function $`F_2^{}`$, in the limit $`J^{}/J0`$, where $`c^{(3)}`$ and $`\xi _{\pi ,\pi ,0}^{(3)}`$ are the in-plane spin-wave velocity and correlation length of $`H_{3d}^{}`$ at temperature $`1/\beta _{AF}^{}`$ near wave vector $`(\pi ,\pi ,0)`$, respectively. As it is the instantaneous structure factor $`S_{\pi ,\pi ,0}^{}`$ that is most readily measured Ronnow99 , we note that the product $`J^{}S_{\pi ,\pi ,0}^{}\beta _{AF}`$ also should obey a scaling law of the form (3) for some scaling function $`F_3^{}`$.
Third, we predict that similar scaling results hold for other layered models.
In quasi one-dimension, we expect that similar scaling results will also hold. This does not, however, help us understand the results of Yasuda et. al. in quasi one-dimension. The scaling functions $`F_1^{},F_2^{}`$ imply that the classical and quantum models will show the same $`\zeta ^{(1)}`$ only if $`c\beta _{AF}^{}\xi ^{(1)}`$. However, as the one-dimensional Heisenberg model on a chain is not in the renormalized classical regime but rather quantum critical, it should have some non-zero $`c\beta _{AF}^{}/\xi ^{(1)}`$ and should show a different $`\zeta ^{(1)}`$ than the classical model. Thus, the one-dimensional results remain a puzzle.
Physical Motivation— Here, we present a physical motivation for the results above and a brief microscopic derivation of the relevant non-linear sigma model. In the next section, we show these scaling results using a renormalization group (RG) for this non-linear sigma model. The reason for which we use this model is that we want to illustrate the effects of field renormalization and the non-linear sigma model RG already has a nontrivial field renormalization at leading order in the coupling constant, while such a renormalization is not seen until order $`ϵ^2`$ ($`1/N`$) in a $`4ϵ`$ (large $`N`$) expansion.
Since the interplane interaction is weak, we can treat it perturbatively at the microscopic level. Following standard steps, in the absence of the interplane interaction, we can first derive the partition function for the two-dimensional $`O(N)`$ quantum non-linear sigma model ($`2d`$QNLSM) with field $`𝒏_k^{}(𝒓,\tau )`$, where $`k`$ is a discrete index labelling individual planes, $`𝒓`$ is a two-dimensional vector describing coordinates in the plane, and $`\tau `$ is imaginary time. The relevant action for plane $`k`$ is $`S_k^{}=S_k^{(1)}+S_k^{(2)}`$ where
$`S_k^{(1)}:={\displaystyle _k^{(1)}}{\displaystyle \underset{0}{\overset{\beta }{}}}𝑑\tau {\displaystyle \underset{a}{\overset{L}{}}}d^2𝒓{\displaystyle \frac{c}{2ag}}\left(_\mu ^{}𝒏_k^{}\right)^2`$ (4a)
and
$`S_k^{(2)}:={\displaystyle _k^{(2)}}{\displaystyle \underset{0}{\overset{\beta }{}}}𝑑\tau {\displaystyle \underset{a}{\overset{L}{}}}d^2𝒓{\displaystyle \frac{c}{a^3}}Z_h^{}𝒉𝒏_k^{}.`$ (4b)
Here, the lattice spacing $`a`$ plays the role of the microscopic ultra-violet (UV) cutoff, i.e., $`\mathrm{\Lambda }1/a`$ that of an upper cutoff on momenta. The linear size $`L`$ of the plane is the largest length scale of the problem. The derivative $`_\mu =(_{c\tau },\mathbf{})`$ depends on the spin-wave velocity $`c`$ in the plane and is of order $`Ja`$. The dimensionless coupling constant $`g`$ depends on the microscopic details of the intraplane interactions. The dimensionless background field $`𝒉`$, where $`h=|𝒉|`$, is the external source for a static staggered magnetic field conjugate to the planar AF order parameter of the underlying lattice model. It breaks the $`O(N)`$ symmetry of Lagrangian (4a) down to $`O(N1)`$ and as such acts as an infra-red (IR) regulator. The dimensionless coupling $`Z_h^{}`$ is the field renormalization constant associated to $`𝒏_k^{}`$. The use of the continuum limit within each of the planes labelled by $`k`$ is justified if we are after the physics on length scales much longer than $`a`$.
The interplane nearest-neighbor AF coupling $`J^{}`$ gives the characteristic interplane spin-wave velocity $`c^{}J^{}a`$ and length scale $`a^{}(J/J^{})^{1/2}a`$. The couplings $`J^{},g`$ get renormalized as discussed below, so the velocity $`c^{}`$ changes at longer length scales. For a very weak nearest-neighbor interplain AF coupling, $`J^{}J`$, the physics on length scales much larger than $`a`$ but yet not much larger than $`a^{}`$ is captured by the partition function
$`Z={\displaystyle \underset{^N}{}}\left[{\displaystyle \underset{k}{}}𝒟[𝒏_k^{}]\delta (𝒏_k^21)\right]\mathrm{exp}({\displaystyle \underset{k}{}}{\displaystyle _k^{}}),`$ (5a)
$`_k^{}=_k^{(1)}+_k^{(2)}+_k^{(3)}.`$ (5b)
The Lagrangian $`_k^{(3)}`$ encodes the effect of the microscopic nearest-neighbor interplane AF interaction $`J^{}`$. To compute this, we use a Hubbard-Stratonovich transformation to replace the microscopic interaction between any two spins in the cubic lattice with coordinates $`(i,j,k)`$ and $`(i,j,k+1)`$ by an interaction of each spin with a fluctuating magnetic field $`𝑯_{i,j,k}^{}`$. When $`J^{}J`$ and on intermediary length scales $`a\lambda a^{}`$, the dominant mode for the Hubbard-Stratonovich field is near momentum $`(\pi ,\pi )`$ in the plane. The action for the spins in the presence of this field is the same as Eq. (4b) with field $`𝒉`$ replaced by $`𝑯_k^{}`$. In this approximation, integrating the Hubbard-Stratonovich field gives the short-range interplane interaction term
$`_k^{(3)}={\displaystyle \frac{J^{}Z^{}}{2a^2}}\left(𝒏_k^{}𝒏_{k+1}^{}\right)^2,`$ (5c)
where $`Z^{}`$ renormalizes as $`Z_h^2`$ to lowest order in $`J^{}/J`$,
$`Z^{}=Z_h^2\left[1+𝒪(J^{}/J)\right].`$ (6)
This defines the so-called three-dimensional strongly anisotropic $`O(N)`$ QNLSM ($`3d`$SAQNLSM).
This derivation of the $`3d`$SAQNLSM considered interactions between spins positioned directly above and below each other in neighboring planes. It is possible to treat more complicated interplane interactions. For example, going back to the cubic lattice, let there be an AF interaction $`J_1^{}`$ between the spin at site $`(i,j,k)`$ with that at $`(i,j,k\pm 1)`$ and another AF interactions $`J_2^{}`$ to the spins at sites $`(i\pm 1,j,k\pm 1)`$ and $`(i,j\pm 1,k\pm 1)`$. Then, if $`J_1^{}J`$ and $`J_2^{}J`$, we can derive the $`3d`$SAQNLSM with the interplane coupling $`J^{}=J_1^{}4J_2^{}`$.
Renormalization Group— Here, we present an RG analysis of the non-linear sigma model (5). The most important result in this section is that the identity (6) is preserved under the RG flow up to the length scale at which the scale dependent effective anisotropy (13) is of order 1.
We perform a RG analysis following Polyakov for convenience Polyakov75 . In each plane labelled by $`k`$, we write
$`𝒏_k^{}`$ $`=`$ $`𝒎_k^{}\left(1\mathit{\varphi }_k^2\right)^{1/2}+{\displaystyle \underset{a=1}{\overset{N1}{}}}𝒆_k^a\varphi _k^a.`$ (7)
The field of unit length $`𝒎_k^{}`$ encodes the planar AF order expected in the limit $`g/c0`$, while the $`N1`$ fields $`𝒆_k^a`$ capture the deviations away from the direction $`𝒎_k^{}`$ of symmetry breaking, i.e., the $`N1`$ $`𝒆_k^a`$ form an orthonormal basis of vectors orthogonal to $`𝒎_k^{}`$. The $`N1`$ coefficients $`\varphi _k^a`$ make up the vector $`\mathit{\varphi }_k^{}`$. To leading order in an expansion in powers of $`g/c`$ of the parametrization (7), the field $`𝒎_k^{}`$ is the slow mode while the $`N1`$ fields $`\varphi _k^a`$ represent fast modes with characteristic 2-momenta $`\stackrel{~}{\mathrm{\Lambda }}<|𝒑|\mathrm{\Lambda }`$. Substituting Eq. (7) into Eq. (5) gives the Lagrangian
$`_k^{(1)}+_k^{(2)}={\displaystyle \frac{c}{2ag}}[(_\mu ^{}\varphi _k^aA_{k\mu }^{ab}\varphi _k^b)^2+\left(B_{k\mu }^a\right)^2`$ (8)
$`+B_{k\mu }^aB_{k\mu }^b(\varphi _k^a\varphi _k^b\mathit{\varphi }_k^2\delta ^{ab})]{\displaystyle \frac{c}{a^3}}Z^{}_h𝒉𝒎^{}_k(1\mathit{\varphi }_k^2)^{1/2}`$
to leading order in an expansion in powers of $`g/c`$. The $`N1`$ coefficients $`B_{k\mu }^a`$ are defined by $`_\mu ^{}𝒎_k^{}=_{a=1}^{N1}B_\mu ^a𝒆_k^a`$. The $`(N1)(N2)/2`$ independent coefficients $`A_{k\mu }^{ab}=(_\mu ^{}𝒆_k^b)𝒆_k^a`$. The RG flows of the dimensionless couplings $`g`$, $`Z_h^{}h`$, and $`t1/(J\beta )`$ that follow after integration over the fast modes $`\mathit{\varphi }_k^{}`$ in the limit of no interplane interactions were computed by Chakravarty, Halperin, and Nelson to leading order in $`g/c`$ (see Fig. 1) CHN89 . To this order, $`c`$ is unchanged.
To quantify the very weak microscopic interplanar coupling, we define the anisotropy $`\stackrel{~}{\alpha }`$ as the ratio of the importance of $`_k^{(1)}`$ to $`_k^{(3)}`$ when the upper cutoff on the momenta is $`\stackrel{~}{\mathrm{\Lambda }}`$. By assumption, this anisotropy is strong at the microscopic level (upper cutoff $`\mathrm{\Lambda }`$),
$`\alpha =gJ^{}Z^{}/J1.`$ (9)
Next, we consider the renormalization of $`_k^{(1)}`$ in Eq. (Universal Scaling Relations in Strongly Anisotropic Materials) when $`𝒉=0`$ and of the interplane interaction
$`_k^{(3)}`$ $`=`$ $`{\displaystyle \frac{J^{}Z^{}}{2a^2}}[(1\mathit{\varphi }_k^2)^{1/2}𝒎_k^{}(1\mathit{\varphi }_{k+1}^2)^{1/2}𝒎_{k+1}^{}`$ (10)
$`+{\displaystyle \underset{a=1}{\overset{N1}{}}}(\varphi _k^a𝒆_k^a\varphi _{k+1}^a𝒆_{k+1}^a)]^2`$
after averaging over the fast modes $`\mathit{\varphi }_k^{}`$. To this end, we shall introduce the renormalized values
$`{\displaystyle \frac{1}{\stackrel{~}{g}}}`$ $`=`$ $`{\displaystyle \frac{1}{g}}\left({\displaystyle \frac{\mathrm{\Lambda }}{\stackrel{~}{\mathrm{\Lambda }}}}\right)\left(1+\varphi _k^a\varphi _k^b\mathit{\varphi }_k^2\delta ^{ab}\right),`$ (11a)
$`{\displaystyle \frac{1}{\stackrel{~}{t}}}`$ $`=`$ $`{\displaystyle \frac{1}{t}}\left(1+\varphi _k^a\varphi _k^b\mathit{\varphi }_k^2\delta ^{ab}\right),`$ (11b)
$`\stackrel{~}{Z}_h^{}`$ $`=`$ $`Z_h^{}\left(1{\displaystyle \frac{1}{2}}\mathit{\varphi }_k^2+\mathrm{}\right),`$ (11c)
at the scale $`\stackrel{~}{\mathrm{\Lambda }}`$ as a result of averaging over the fast modes $`\mathit{\varphi }_k^{}`$. For $`\alpha 1`$, this average over fast modes is $`\varphi _k^a\varphi _l^b=\delta _{kl}^{}\delta ^{ab}\mathrm{ln}(\mathrm{\Lambda }/\stackrel{~}{\mathrm{\Lambda }})\frac{g}{4\pi }\mathrm{coth}(g/2t)`$. Furthermore,
$`_k^{(3)}{\displaystyle \frac{J^{}Z^{}}{2a^2}}{\displaystyle \frac{\stackrel{~}{Z}_h^2}{Z_h^2}}\left(𝒎_k^{}𝒎_{k+1}^{}\right)^2`$ (12)
from which follows the renormalizations
$`\stackrel{~}{\alpha }`$ $`=`$ $`\left({\displaystyle \frac{\mathrm{\Lambda }}{\stackrel{~}{\mathrm{\Lambda }}}}\right)^3{\displaystyle \frac{\stackrel{~}{g}\stackrel{~}{Z^{}}}{gZ^{}}}\alpha ,{\displaystyle \frac{\stackrel{~}{Z^{}}}{Z^{}}}={\displaystyle \frac{\stackrel{~}{Z}_h^2}{Z_h^2}},`$ (13)
so long as $`\stackrel{~}{\alpha }1`$.
Let us follow the RG flows encoded by Eqs. (11) and Eqs. (13) starting from the initial values $`g>0`$, $`t1/(J\beta _{AF}^{})<\mathrm{}`$, and $`1\alpha >0`$, see Eq. (9), corresponding to a point on the phase boundary between the Néel and paramagnetic phase (see Fig. 1). Aside from the thermal de Broglie wavelength of the spin waves $`c\beta _{AF}^{}`$, the initial values $`g`$ and $`t`$ define a second characteristic length scale, the correlation length $`\xi ^{(2)}`$ in the $`2d`$QNLSM, in view of $`1\alpha >0`$. We shall distinguish two cases. In the renormalized classical regime $`c\beta _{AF}^{}/\xi ^{(2)}1`$. In the quantum critical regime $`c\beta _{AF}^{}/\xi ^{(2)}1`$. Finally, we denote by $`\xi _{\mathrm{cross}}^{}`$ the RG length scale at which $`\stackrel{~}{\alpha }1`$ and beyond which the RG flows Eqs. (11) and Eqs. (13) should be replaced by the flows of the isotropic $`3d`$QNLSM; naive scaling gives $`\xi _{\mathrm{cross}}a^{}`$, but the RG flows above will change this scaling. Any two of these characteristic length scales, $`c\beta _{AF},\xi ^{(2)},`$ and $`\xi _{\mathrm{cross}}`$, fix the third one since the RG flows are constrained to the boundary between the Néel and paramagnetic phases by assumption. Without loss of generality, we shall consider the case $`\mathrm{\Lambda }>1/(c\beta _{AF}^{})`$. As we lower the upper momentum cutoff, the RG scale $`\stackrel{~}{\mathrm{\Lambda }}^1`$ will eventually become larger than $`c\beta _{AF}^{}`$. We shall consider RG scales $`\stackrel{~}{\mathrm{\Lambda }}^1c\beta _{AF}^{}`$, for which the quantum fluctuations are important.
We begin with the renormalized classical regime of the $`3d`$SAQNLSM. As is illustrated in Fig. 1, the running coupling constants $`\stackrel{~}{g}`$ in Eq. (11a) and $`\stackrel{~}{t}`$ in Eq. (11b) flow towards zero and $`\mathrm{}`$, respectively, as $`\stackrel{~}{\mathrm{\Lambda }}`$ decreases but so long $`\stackrel{~}{\mathrm{\Lambda }}^1c\beta _{AF}^{}`$. By Eq. (13), the effective anisotropy decreases; using naive scaling which is valid for $`\stackrel{~}{g}1`$, we have $`\stackrel{~}{\alpha }(\mathrm{\Lambda }/\stackrel{~}{\mathrm{\Lambda }})^2\alpha `$. Beyond the RG length scale $`\stackrel{~}{\mathrm{\Lambda }}^1c\beta _{AF}^{}`$ we can replace the $`2d`$QNLSM in each plane by a classical $`2d`$NLSM with the effective coupling $`\stackrel{~}{g}_{cl}^{}`$, where $`\stackrel{~}{g}_{cl}^{}=\stackrel{~}{g}`$ at the scale $`\stackrel{~}{\mathrm{\Lambda }}^1c\beta _{AF}^{}`$. The effective anisotropy of the classical $`3d`$SANLSM continues to decrease as $`\stackrel{~}{\alpha }=(\mathrm{\Lambda }/\stackrel{~}{\mathrm{\Lambda }})^2(\mathrm{\Lambda }c\beta _{AF}^{})[\stackrel{~}{g}_{cl}^{}\stackrel{~}{Z^{}}/(gZ^{})]\alpha =(\mathrm{\Lambda }/\stackrel{~}{\mathrm{\Lambda }})^2\stackrel{~}{g}_{cl}^{}\stackrel{~}{Z^{}}J^{}\beta _{AF}^{}`$ continues to grow until it reaches the isotropic RG scale $`\stackrel{~}{\alpha }1`$. Equation (6) can then no longer hold. However, since there is only a finite range of scales over which $`\stackrel{~}{\alpha }`$ is non-negligible but still less than unity, we deduce that, at the scale $`\stackrel{~}{\alpha }1`$, $`\stackrel{~}{Z^{}}\stackrel{~}{Z}_h^2`$, up to some constant of order unity. Furthermore, $`\stackrel{~}{g}_{cl}^{}1`$ at this scale also since it lies at some point on the phase boundary between the Néel and paramagnetic phases. But $`\stackrel{~}{g}_{cl}^{}1`$ tells us that the corresponding RG scale $`\stackrel{~}{\mathrm{\Lambda }}`$ is of the order of the correlation length of the $`2d`$QNLSM. In turn, this allows us to infer that the static staggered susceptibility of the $`2d`$QNLSM is given by $`\chi _s^{(2)}(\mathrm{\Lambda }/\stackrel{~}{\mathrm{\Lambda }})^2\stackrel{~}{Z}_h^2\beta _{AF}^{}`$, up to universal corrections of order unity. We now multiply $`\chi _s^{(2)}`$ by $`J^{}`$ estimated from the anisotropy $`\alpha `$ of the classical $`3d`$SANLSM, $`J^{}\chi _s^{(2)}\stackrel{~}{\alpha }(\stackrel{~}{Z}_h^2/\stackrel{~}{Z^{}})/\stackrel{~}{g}_{cl}^{}`$. Using Eq. (6), and the fact that $`\stackrel{~}{g}_{cl}^{}1`$ and $`\stackrel{~}{\alpha }1`$, we arrive at Eq. (1) in the renormalized classical regime.
Note that each of these relations, such as $`\stackrel{~}{\alpha }1`$ and $`\stackrel{~}{g}_{cl}^{}1`$, is defined up to a multiplicative constant that depends on the details of how we define the RG. However, the dimensionless combination in Eq. (1) is universal. The reason for the universality is that all the microscopic details of the Heisenberg model are encoded into the three independent quantities $`\stackrel{~}{g}`$, $`\stackrel{~}{Z^{}}`$, and $`\stackrel{~}{Z}_h^{}`$ on any length scale much larger than $`a`$. Let us perform the RG flow to some scale such that $`\stackrel{~}{\alpha }`$ is much less than unity. Then, the identity (6) relates $`\stackrel{~}{Z^{}}`$ to $`\stackrel{~}{Z}_h^{}`$, leaving only two quantities independent in the classical regime, say $`\stackrel{~}{g}_{cl}^{}`$ and $`\stackrel{~}{\alpha }`$. The requirement of criticality relates $`\stackrel{~}{g}_{cl}^{}`$ to $`\stackrel{~}{\alpha }`$, leaving only one independent quantity, say $`\stackrel{~}{\alpha }`$. Choosing the renormalization scale to be some given fraction of the correlation length in the two-dimensional model fixes the last quantity, and thus there are no independent parameters left.
Near a quantum critical point and as is the case for the renormalized classical regime, the length scale at which $`\stackrel{~}{\alpha }1`$ is of the order $`\xi ^{(2)}`$. Now, however, there is no significant separation of scales between $`c\beta _{AF}^{}`$ and $`\xi ^{(2)}`$ anymore, i.e., $`\stackrel{~}{\alpha }1`$ already at $`c\beta _{AF}^{}`$. Correspondingly, there will be universal corrections to (6) in the form $`\stackrel{~}{Z^{}}/Z^{}=\kappa (1,g(1))\stackrel{~}{Z}_h^2/Z_h^2`$ where the function $`\kappa `$ of $`\alpha `$ and $`g`$ is universal with $`\kappa (0,g)=1`$. The deviations in the quantum critical regime from the limiting value of $`J^{}\chi _s^{(2)}`$ in the classical renormalized regime define the universal scaling function $`F_1^{}`$ of $`c\beta _{AF}^{}/\xi ^{(2)}`$.
Similarly, the correlation at the $`(\pi ,\pi ,0)`$ point is of order $`\stackrel{~}{\mathrm{\Lambda }}^1`$ in the plane while it is of order a single interlayer spacing between the planes. Thus, $`\chi _{\pi ,\pi ,0}^{(3)}\chi _s^{(2)}`$ and $`\xi _{\pi ,\pi ,0}^{(3)}\xi ^{(2)}`$, and so Eq. (3) follows.
We close by noting that all arguments presented here for a non-linear sigma model with $`O(N)`$ symmetry extend to non-linear sigma models defined on Riemannian manifolds with a positive curvature tensor. For example, we expect similar universal scaling relations for a stacking of AF Heisenberg models on a triangular lattice.
Discussion— We have provided a field-theoretic basis for understanding the result of Yasuda et. al in the quasi-two-dimensional case, generalized it to deal with situations near quantum critical points, and given a version thereof expressed in terms of experimentally accessible quantities. The calculation of the scaling functions $`F_{1,2,3}^{}`$ within a field theoretic approach will only be approximate, and the best estimate for $`F_1^{}(0)`$ is given by numerical calculations. For example, it can be shown that the mean-field result $`F_2^{}(0)=1/2`$ follows from the large $`N`$ limit of the $`O(N)`$ $`3d`$SAQNLSM. It would be very valuable to perform a numerical study of $`F_{2,3}^{}`$, in both the renormalized classical and quantum critical regimes.
Acknowledgements— We thank M. Troyer for explaining Yasuda05 and H. M. Ronnow for useful discussions. This work was supported by DOE grant W-7405-ENG-36.
|
warning/0506/math0506122.html
|
ar5iv
|
text
|
# Nonlinear problems with boundary blow-up: a Karamata regular variation theory approach
## 1. Introduction and main results
Let $`\mathrm{\Omega }𝐑^N`$ $`(N3)`$ be a smooth bounded domain.
Consider the semilinear elliptic equation
(1.1)
$$\mathrm{\Delta }u+au=b(x)f(u)\text{in}\mathrm{\Omega },$$
where $`fC^1[0,\mathrm{})`$, $`a𝐑`$ is a parameter and $`bC^{0,\mu }(\overline{\mathrm{\Omega }})`$ satisfies $`b0`$, $`b0`$ in $`\mathrm{\Omega }`$. Such equations are also known as the stationary version of the Fisher equation and the Kolmogoroff–Petrovsky–Piscounoff equation and they have been studied by Kazdan–Warner , Ouyang , del Pino and Du–Huang .
Note that if $`f(u)=u^{(N+2)/(N2)}`$, then (1.1) originates from the Yamabe problem, which is a basic problem in Riemannian geometry (see, e.g., ).
The existence of positive solutions of (1.1) subject to the Dirichlet boundary condition, $`u=0`$ on $`\mathrm{\Omega }`$, has been intensively studied in the case $`f(u)=u^p`$, $`p>1`$ (see , , , , and ); this problem is a basic population model (see ) and it is also related to some prescribed curvature problems in Riemannian geometry (see and ). Moreover, if $`b>0`$ in $`\overline{\mathrm{\Omega }}`$, then it is referred to as the logistic equation and it has a unique positive solution if and only if $`a>\lambda _1(\mathrm{\Omega })`$, where $`\lambda _1(\mathrm{\Omega })`$ denotes the first eigenvalue of $`(\mathrm{\Delta })`$ in $`H_0^1(\mathrm{\Omega })`$.
In the understanding of (1.1) an important role is played by the interior of the zero set of $`b`$:
$$\mathrm{\Omega }_0:=\mathrm{int}\{x\mathrm{\Omega }:b(x)=0\}.$$
We assume, throughout this paper, that $`\mathrm{\Omega }_0`$ is connected (possibly empty), $`\overline{\mathrm{\Omega }}_0\mathrm{\Omega }`$ and $`b>0`$ in $`\mathrm{\Omega }\overline{\mathrm{\Omega }}_0`$. Note that we allow $`b0`$ on $`\mathrm{\Omega }`$. Let $`\mathrm{\Omega }_0`$ satisfy an exterior cone condition and $`\lambda _{\mathrm{},1}`$ be the first Dirichlet eigenvalue of $`(\mathrm{\Delta })`$ in $`H_0^1(\mathrm{\Omega }_0)`$ (with $`\lambda _{\mathrm{},1}=\mathrm{}`$ if $`\mathrm{\Omega }_0=\mathrm{}`$).
By a *large* (or *blow-up*) solution of (1.1), we mean any non-negative $`C^2(\mathrm{\Omega })`$-solution of (1.1) such that $`u(x)\mathrm{}`$ as $`d(x):=\mathrm{dist}(x,\mathrm{\Omega })0`$.
Assuming that $`f`$ satisfies
($`A_1`$)
$$fC^1[0,\mathrm{})\text{is non-negative and }f(u)/u\text{is increasing on }(0,\mathrm{}),$$
then, necessarily $`f(0)=0`$, and by the strong maximum principle, any non-negative classical solution of (1.1) is positive in $`\mathrm{\Omega }`$ unless it is identically zero. Consequently, any large solution of (1.1) is positive. Moreover, it is well known (see, e.g., Remark 1.1 in ) that in this situation, the Keller–Osserman condition
($`A_2`$)
$$_1^{\mathrm{}}\frac{dt}{\sqrt{F(t)}}<\mathrm{},\text{where}F(t)=_0^tf(s)𝑑s$$
is necessary for the existence of large solutions of (1.1).
When $`(A_1)`$ and $`(A_2)`$ hold, Theorem 1.1 in shows that (1.1) possesses large solutions if and only if $`a<\lambda _{\mathrm{},1}`$. The hypothesis $`(A_1)`$ is inspired by , where it is developed an exhaustive study of positive solutions of (1.1), subject to $`u=0`$ on $`\mathrm{\Omega }`$.
Our major goal is to advance innovative methods to study the uniqueness and asymptotic behavior of large solutions of (1.1). We develop the research line opened up in to gain insight into the two-term asymptotic expansion of the large solution near $`\mathrm{\Omega }`$. Our approach relies essentially on the *regular variation theory* (see and section 2) not only in the statement but in the proof as well. This enables us to obtain significant information about the qualitative behavior of the large solution to (1.1) in a general framework that removes previous restrictions in the literature.
We point out that, despite a long history and intense research on the large solutions, the regular variation theory arising in probability theory has not been exploited before in this context.
Singular value problems having large solutions have been initially studied for the special case $`f(u)=e^u`$ by Bieberbach (if $`N=2`$). Problems of this type arise in Riemannian geometry. More precisely, if a Riemannian metric of the form $`|ds|^2=e^{2u(x)}|dx|^2`$ has constant Gaussian curvature $`g^2`$ then $`\mathrm{\Delta }u=g^2e^{2u}`$. This study was continued by Rademacher (if $`N=3`$) in connection with some concrete questions arising in the theory of Riemann surfaces, automorphic functions and in the theory of the electric potential in a glowing hollow metal body.
The question of large solutions was later considered in $`N`$-dimensional domains and for other classes of nonlinearities (see , , , , , , , , , , , , ).
In higher dimensions the notion of Gaussian curvature has to be replaced by the scalar curvature. It turns out that if a metric of the form $`|ds|^2=u(x)^{4/(N2)}|dx|^2`$ has constant scalar curvature $`g^2`$, then $`u`$ satisfies (1.1) for $`f(u)=u^{(N+2)/(N2)}`$, $`a=0`$ and $`b(x)=[(N2)g^2]/[4(N1)]`$. In a celebrated paper, Loewner and Nirenberg described the precise asymptotic behavior at the boundary of large solutions to this equation and used this result in order to establish the uniqueness of the solution. Their main result is derived under the assumption that $`\mathrm{\Omega }`$ consists of the disjoint union of finitely compact $`C^{\mathrm{}}`$ manifolds, each having codimension less than $`N/2+1`$. More precisely, the uniqueness of a large solution is a consequence of the fact that every large solution $`u`$ satisfies
(1.2)
$$u(x)=(d(x))+o((d(x)))\text{as}d(x)0,$$
where $``$ is defined by
(1.3)
$$_{(t)}^{\mathrm{}}\frac{ds}{\sqrt{2F(s)}}=\left(\frac{(N2)g^2}{4(N1)}\right)^{1/2}t,\text{for all}t>0.$$
Kondrat’ev and Nikishkin established the uniqueness of a large solution for the case $`a=0`$, $`b=1`$ and $`f(u)=u^p`$ ($`p3`$), when $`\mathrm{\Omega }`$ is a $`C^2`$-manifold and $`\mathrm{\Delta }`$ is replaced by a more general second order elliptic operator.
Dynkin showed that there exist certain relations between hitting probabilities for some Markov processes called superdiffusions and maximal solutions of (1.1) with $`a=0`$, $`b=1`$ and $`f(u)=u^p`$ ($`1<p2`$). By means of a probabilistic representation, a uniqueness result in domains with non-smooth boundary was established by le Gall when $`p=2`$. We point out that the case $`p=2`$ arises in the study of the subsonic motion of a gas. In this connection the question of uniqueness is of special interest.
Recently, gives the uniqueness and exact two-term asymptotic expansion of the large solution of (1.1) in the special case $`f(u)=u^p`$ ($`p>1`$), $`b>0`$ in $`\mathrm{\Omega }`$ and $`b0`$ on $`\mathrm{\Omega }`$ such that
(1.4)
$$b(x)=C_0[d(x)]^\gamma +o([d(x)]^\gamma )\text{as}d(x)0,\text{for some constants}C_0,\gamma >0.$$
It was shown there that the degenerate case $`b0`$ on $`\mathrm{\Omega }`$ is a *natural* restriction for $`b`$ inherited from the logistic equation.
To present our main results, we briefly recall some notions from Karamata’s theory (see or ); more details are provided in section 2.
A positive measurable function $`R`$ defined on $`[A,\mathrm{})`$, for some $`A>0`$, is called regularly varying with index $`q𝐑`$, written $`RRV_q`$, provided that
$$\underset{u\mathrm{}}{lim}\frac{R(\lambda u)}{R(u)}=\lambda ^q,\text{for all}\lambda >0.$$
When the index $`q`$ is zero, we say that the function is slowly varying.
Clearly, if $`RRV_q`$, then $`L(u):=R(u)/u^q`$ is a slowly varying function.
Let $`𝒦`$ denote the set of all positive, non-decreasing $`kC^1(0,\nu )`$ that satisfy
$$\underset{t0}{lim}\left(\frac{_0^tk(s)𝑑s}{k(t)}\right):=\mathrm{}_0\text{and}\underset{t0}{lim}\left(\frac{_0^tk(s)𝑑s}{k(t)}\right)^{}:=\mathrm{}_1.$$
Notice that $`\mathrm{}_0=0`$ and $`\mathrm{}_1[0,1]`$, for every $`k𝒦`$. Thus, $`𝒦=𝒦_{(01]}𝒦_0`$, where
$$𝒦_{(01]}=\{k𝒦:0<\mathrm{}_11\}\text{and}𝒦_0=\{k𝒦:\mathrm{}_1=0\}.$$
The exact characterization of $`𝒦_{(01]}`$ and $`𝒦_0`$ will be provided in section 3.
If $`H`$ is a non-decreasing function on $`𝐑`$, then we define the (left continuous) inverse of $`H`$ by
$$H^{}(y)=inf\{s:H(s)y\}.$$
Our first result establishes the uniqueness of the large solution of (1.1).
###### Theorem 1.1.
Let $`(A_1)`$ hold and $`fRV_{\rho +1}`$ with $`\rho >0`$. Suppose there exists $`k𝒦`$ such that
(1.5)
$$b(x)=k^2(d)+o(k^2(d))\text{as }d(x)0.$$
Then, for any $`a(\mathrm{},\lambda _{\mathrm{},1})`$, $`(\text{1.1})`$ admits a unique large solution $`u_a`$. Moreover, the asymptotic behavior is given by
(1.6)
$$u_a(x)=[2(2+\mathrm{}_1\rho )/\rho ^2]^{1/\rho }\phi (d)+o(\phi (d))\text{as }d(x)0,$$
where $`\phi `$ is defined by
(1.7)
$$\frac{f(\phi (t))}{\phi (t)}=\frac{1}{\left(_0^tk(s)𝑑s\right)^2},\text{for }t>0\text{small}.$$
Under the assumptions of Theorem 1.1, let $`r(t)`$ satisfy $`lim_{t0}\left(_0^tk(s)𝑑s\right)^2r(t)=1`$ and $`\widehat{f}(u)`$ be chosen such that $`lim_u\mathrm{}\widehat{f}(u)/f(u)=1`$ and $`j(u)=\widehat{f}(u)/u`$ is non-decreasing for $`u>0`$ large. Then, $`lim_{t0}\phi (t)/\widehat{\phi }(t)=1`$, where $`\phi `$ is defined by (1.7) and $`\widehat{\phi }(t)=j^{}(r(t))`$ for $`t>0`$ small.
The behavior of $`\phi (t)`$ for small $`t>0`$ will be described in section 3. In particular, if $`k𝒦`$ with $`\mathrm{}_10`$, then $`\phi (1/u)RV_{2/(\rho \mathrm{}_1)}`$. In contrast, if $`k𝒦`$ with $`\mathrm{}_1=0`$, then $`\phi (1/u)RV_q`$, for all $`q𝐑`$ (see Remark 3.3).
###### Remark 1.1.
Theorem 1.1 improves the main result in , where assuming that $`f^{}RV_\rho `$ (which yields $`fRV_{\rho +1}`$), we prove
(1.8)
$$u_a(x)=\xi _0h(d)+o(h(d))\text{as }d(x)0,$$
where $`\xi _0=\left(\frac{2+\mathrm{}_1\rho }{2+\rho }\right)^{1/\rho }`$ and $`h`$ is given by
(1.9)
$$_{h(t)}^{\mathrm{}}\frac{ds}{\sqrt{2F(s)}}=_0^tk(s)𝑑s,\text{for }t>0\text{ small}.$$
###### Remark 1.2.
Theorem 1.1 recovers the uniqueness results of and . Note that for $`k(t)=[(N2)g^2/4(N1)]^{1/2}`$ in (1.5) and $`f(u)=u^{(N+2)/(N2)}`$, (1.6) reduces to relation (1.2), prescribed by Loewner and Nirenberg for their problem. Moreover, if $`f(u)=u^p`$ (with $`p=\rho +1>1`$) and $`k(t)=\sqrt{C_0}t^{\gamma /2}`$ ($`C_0,\gamma >0`$), then we regain the uniqueness result of .
The next objective is to find the two-term blow-up rate of $`u_a`$ when (1.5) is replaced by
(1.10)
$$b(x)=k^2(d)(1+\stackrel{~}{c}d^\theta +o(d^\theta ))\text{as }d(x)0,$$
where $`\theta >0`$, $`\stackrel{~}{c}𝐑`$ are constants. To simplify the exposition, we assume that $`f^{}RV_\rho `$ $`(\rho >0)`$, which is equivalent to $`f(u)`$ being of the form
(1.11)
$$f(u)=Cu^{\rho +1}\mathrm{exp}\left\{_B^u\frac{\varphi (t)}{t}𝑑t\right\},uB,$$
for some constants $`B,C>0`$, where $`\varphi C[B,\mathrm{})`$ satisfies $`lim_u\mathrm{}\varphi (u)=0`$. In this case, $`f(u)/u`$ is increasing on $`[B,\mathrm{})`$ provided that $`B`$ is large enough.
We prove that the two-term asymptotic expansion of $`u_a`$ near $`\mathrm{\Omega }`$ depends on the chosen subclass for $`k𝒦`$ and the additional hypotheses on $`f`$ (by means of $`\varphi `$ in (1.11)).
Let $`\rho 2<\eta 0`$ and $`\tau ,\zeta >0`$. We define
$`_{\rho \eta }`$ $`=\{f^{}RV_\rho (\rho >0):\text{either}\varphi RV_\eta \text{or}\varphi RV_\eta \},`$
$`_{\rho 0,\tau }`$ $`=\{f_{\rho 0}:\underset{u\mathrm{}}{lim}(\mathrm{ln}u)^\tau \varphi (u)=\mathrm{}^{}𝐑\},`$
$`𝒦_{(01],\tau }`$ $`=\{k𝒦_{(01]}:\underset{t0}{lim}(\mathrm{ln}t)^\tau \left[\left({\displaystyle \frac{_0^tk(s)𝑑s}{k(t)}}\right)^{}\mathrm{}_1\right]:=L_{\mathrm{}}𝐑\},`$
$`𝒦_{0,\zeta }`$ $`=\{k𝒦_0:\underset{t0}{lim}{\displaystyle \frac{1}{t^\zeta }}\left({\displaystyle \frac{_0^tk(s)𝑑s}{k(t)}}\right)^{}:=L_{}𝐑\}.`$
Further in the paper, $`\eta `$, $`\tau `$ and $`\zeta `$ are understood in the above range.
For the sake of comparison, we state here the following result.
###### Theorem 1.2.
Suppose $`(A_1)`$, (1.10) with $`k𝒦_{0,\zeta }`$, and one of the following growth conditions at infinity:
1. $`f(u)=Cu^{\rho +1}`$ in a neighborhood of infinity (i.e., $`\varphi 0`$ in $`(\text{1.11})`$);
2. $`f_{\rho \eta }`$ with $`\eta 0`$;
3. $`f_{\rho 0,\tau _1}`$ with $`\tau _1=\varpi /\zeta `$, where $`\varpi =\mathrm{min}\{\theta ,\zeta \}`$.
Then, for any $`a(\mathrm{},\lambda _{\mathrm{},1})`$, the two-term blow-up rate of $`u_a`$ is
(1.12)
$$u_a(x)=\xi _0h(d)(1+\chi d^\varpi +o(d^\varpi ))\text{as }d(x)0$$
where $`h`$ is given by (1.9), $`\xi _0=[2/(2+\rho )]^{1/\rho }`$ and
$$\chi =\{\begin{array}{cc}& \frac{L_{}}{2}\mathrm{Heaviside}(\theta \zeta )\frac{\stackrel{~}{c}}{\rho }\mathrm{Heaviside}(\zeta \theta ):=\chi _1\text{if}(\mathrm{i})\text{or}(\mathrm{ii})\text{holds},\hfill \\ & \chi _1\frac{\mathrm{}^{}}{\rho }\left[\frac{\rho \zeta L_{}}{2(1+\zeta )}\right]^{\tau _1}\left(\frac{1}{\rho +2}+\mathrm{ln}\xi _0\right)\text{if }f\text{ obeys }(\mathrm{iii}).\hfill \end{array}$$
Theorem 1.2 is a consequence of \[14, Theorem 1\] and Proposition 3.4.
###### Theorem 1.3.
Suppose $`(A_1)`$, (1.10) with $`k𝒦_{(01],\tau }`$, and one of the following conditions:
1. $`f_{\rho \eta }`$ with $`\eta L_{\mathrm{}}0`$;
2. $`f_{\rho 0,\tau }`$ with $`[\mathrm{}^{}(\mathrm{}_11)]^2+L_{\mathrm{}}^20`$.
Then, for any $`a(\mathrm{},\lambda _{\mathrm{},1})`$, the two-term blow-up rate of $`u_a`$ is
(1.13)
$$u_a(x)=\xi _0h(d)[1+\stackrel{~}{\chi }(\mathrm{ln}d)^\tau +o((\mathrm{ln}d)^\tau )]\text{as }d(x)0,$$
where $`h`$ is given by (1.9), $`\xi _0=[(2+\mathrm{}_1\rho )/(2+\rho )]^{1/\rho }`$ and
(1.14)
$$\stackrel{~}{\chi }=\{\begin{array}{cc}& \frac{L_{\mathrm{}}}{2+\rho \mathrm{}_1}:=\chi _2\text{if}(\mathrm{i})\text{holds},\hfill \\ & \chi _2\frac{\mathrm{}^{}}{\rho }\left(\frac{\rho \mathrm{}_1}{2}\right)^\tau \left[\frac{2(1\mathrm{}_1)}{(\rho +2)(\rho \mathrm{}_1+2)}+\mathrm{ln}\xi _0\right]\text{if }f\text{ obeys}(\mathrm{ii}).\hfill \end{array}$$
###### Remark 1.3.
Note that Theorems 1.2 and 1.3 distinguish from Theorem 1 in , which treats the particular case $`f(u)=u^p`$ $`(p>1)`$, $`\mathrm{\Omega }_0=\mathrm{}`$, $`k(t)=\sqrt{C_0t^\gamma }`$ ($`C_0,\gamma >0`$) and $`\theta =1`$ in (1.10). The second term in the asymptotic expansion of $`u_a`$ near $`\mathrm{\Omega }`$ involves in both the distance function $`d(x)`$ and the mean curvature of $`\mathrm{\Omega }`$.
Theorem 1.2 admits the case $`f(u)=u^p`$ assuming that $`k𝒦_{0,\zeta }`$, while the alternative (ii) of Theorem 1.3 includes the case $`k(t)=\sqrt{C_0t^\gamma }`$ (when $`L_{\mathrm{}}=0`$) provided that $`f_{\rho 0,\tau }`$ with $`\mathrm{}^{}0`$. Relations (1.12) and (1.13) show how dramatically changes the two-term asymptotic expansion of $`u_a`$ from the result in . Our approach is completely different from that in , as we use essentially Karamata’s theory.
We point out that the asymptotic general results stated in the above theorems do not concern the difference or the quotient of $`u(x)`$ and $`\psi (d(x))`$, as established in , , , for $`a=0`$ and $`b=1`$, where $`\psi `$ is a large solution of
$$\psi ^{\prime \prime }(r)=f(\psi (r))\text{on}(0,\mathrm{}).$$
For instance, Bieberbach and Rademacher proved that $`|u(x)\psi (d(x))|`$ is bounded in a neighborhood of the boundary. Their result was improved by Bandle and Essén who showed that $`lim_{d(x)0}\left(u(x)\psi (d(x))\right)=0`$.
The rest of the paper is organized as follows. In section 2.1 we collect the notions and properties of regularly varying functions that are invoked in our proofs. In section 2.2 we prove some auxiliary results including Lemmas 1 and 2 in , which have only been stated there. In Section 3 we characterize the class $`𝒦`$ as well as its subclasses $`𝒦_{0,\zeta }`$ and $`𝒦_{(01],\tau }`$ that appear in Theorems 1.2 and 1.3. Sections 4 and 5 are dedicated to the proof of Theorems 1.1 and 1.3.
## 2. Preliminaries
### 2.1. Properties of regularly varying function
The theory of regular variation was instituted in 1930 by Karamata and subsequently developed by himself and many others. Although Karamata originally introduced his theory in order to use it in Tauberian theorems, regularly varying functions have been later applied in several branches of Analysis: Abelian theorems (asymptotic of series and integrals—Fourier ones in particular), analytic (entire) functions, analytic number theory, etc. The great potential of regular variation for probability theory and its applications was realised by Feller and also stimulated by de Haan . The first monograph on regularly varying functions was written by Seneta , while the theory and various applications of the subject are presented in the comprehensive treatise of Bingham, Goldie and Teugels .
We give here a brief account of the definitions and properties of regularly varying functions involved in our paper (see or for details).
###### Definition 2.1.
A positive measurable function $`Z`$ defined on $`[A,\mathrm{})`$, for some $`A>0`$, is called *regularly varying (at infinity) with index* $`q𝐑`$, written $`ZRV_q`$, provided that
$$\underset{u\mathrm{}}{lim}\frac{Z(\xi u)}{Z(u)}=\xi ^q,\text{for all}\xi >0.$$
When the index $`q`$ is zero, we say that the function is *slowly varying*.
###### Remark 2.1.
Let $`Z:[A,\mathrm{})(0,\mathrm{})`$ be a measurable function. Then
1. $`Z`$ is regularly varying if and only if $`lim_u\mathrm{}Z(\xi u)/Z(u)`$ is finite and positive for each $`\xi `$ in a set $`S(0,\mathrm{})`$ of positive measure (see \[48, Lemma 1.6 and Theorem 1.3\]).
2. The transformation $`Z(u)=u^qL(u)`$ reduces regular variation to slow variation. Indeed, $`lim_u\mathrm{}Z(\xi u)/Z(u)=u^q`$ if and only if $`lim_u\mathrm{}L(\xi u)/L(u)=1`$, for every $`\xi >0`$.
###### Example 2.1.
Any measurable function on $`[A,\mathrm{})`$ which has a positive limit at infinity is slowly varying. The logarithm $`\mathrm{log}u`$, its iterates $`\mathrm{log}\mathrm{log}u`$ ($`=\mathrm{log}_2u`$), $`\mathrm{log}_mu`$ ($`=\mathrm{log}\mathrm{log}_{m1}u`$) and powers of $`\mathrm{log}_mu`$ are non-trivial examples of slowly varying functions. Non-logarithmic examples are given by $`\mathrm{exp}\left\{(\mathrm{log}u)^{\alpha _1}\right\}`$, where $`\alpha _1(0,1)`$ and $`\mathrm{exp}\left\{(\mathrm{log}u)/\mathrm{log}\mathrm{log}u\right\}`$.
In what follows $`L`$ denotes a slowly varying function defined on $`[A,\mathrm{})`$. For details on Propositions 2.12.5, we refer to .
###### Proposition 2.1 (Uniform Convergence Theorem).
The convergence $`\frac{L(\xi u)}{L(u)}1`$ as $`u\mathrm{}`$ holds uniformly on each compact $`\xi `$-set in $`(0,\mathrm{})`$.
###### Proposition 2.2 (Representation Theorem).
The function $`L(u)`$ is slowly varying if and only if it can be written in the form
(2.1)
$$L(u)=M(u)\mathrm{exp}\left\{_B^u\frac{y(t)}{t}𝑑t\right\}(uB)$$
for some $`B>A`$, where $`yC[B,\mathrm{})`$ satisfies $`lim_u\mathrm{}y(u)=0`$ and $`M(u)`$ is measurable on $`[B,\mathrm{})`$ such that $`lim_u\mathrm{}M(u):=\overline{M}(0,\mathrm{})`$.
The Karamata representation (2.1) is non-unique because we can adjust one of $`M(u)`$, $`y(u)`$ and modify properly the other one. Thus, the function $`y`$ may be assumed arbitrarily smooth, but the smothness properties of $`M(u)`$ can ultimately reach those of $`L(u)`$. If $`M(u)`$ is replaced by its limit at infinity $`\overline{M}>0`$, we obtain a slowly varying function $`L_0C^1[B,\mathrm{})`$ of the form
$$L_0(u)=\overline{M}\mathrm{exp}\left\{_B^u\frac{y(t)}{t}𝑑t\right\}(uB),$$
where $`yC[B,\mathrm{})`$ vanishes at infinity. Such a function $`L_0(u)`$ is called a *normalised* slowly varying function.
As an important subclass of $`RV_q`$, we distinguish $`NRV_q`$ defined as
(2.2)
$$NRV_q=\{ZRV_q:\frac{Z(u)}{u^q}\text{is a }\text{normalised}\text{ slowly varying function}\}.$$
Notice that $`L(u)`$ given by (2.1) is asymptotic equivalent to $`L_0(u)`$, which has much enhanced properties. For instance, we see that $`y(u)=\frac{uL_0^{}(u)}{L_0(u)}`$, for all $`uB`$. Conversely, any function $`L_0C^1[B,\mathrm{})`$ which is positive and satisfies
(2.3)
$$\underset{u\mathrm{}}{lim}\frac{uL_0^{}(u)}{L_0(u)}=0$$
is a normalised slowly varying. More generally, if the right hand side of (2.3) is $`q𝐑`$, then $`L_0NRV_q`$.
###### Proposition 2.3 (Elementary properties of slowly varying functions).
If $`L`$ is slowly varying, then
1. For any $`\alpha >0`$, $`u^\alpha L(u)\mathrm{}`$, $`u^\alpha L(u)0`$ as $`u\mathrm{}`$;
2. $`(L(u))^\alpha `$ varies slowly for every $`\alpha 𝐑`$;
3. If $`L_1`$ varies slowly, so do $`L(u)L_1(u)`$ and $`L(u)+L_1(u)`$.
From Proposition 2.3 (i) and Remark 2.1 (ii), $`lim_u\mathrm{}Z(u)=\mathrm{}`$ (resp., $`0`$) for any function $`ZRV_q`$ with $`q>0`$ (resp., $`q<0`$).
###### Remark 2.2.
Note that the behavior at infinity for a slowly varying function cannot be predicted. For instance,
$$L(u)=\mathrm{exp}\left\{(\mathrm{log}u)^{1/3}\mathrm{cos}((\mathrm{log}u)^{1/3})\right\}$$
exhibits infinite oscillation in the sense that
$$\underset{u\mathrm{}}{lim\; inf}L(u)=0\text{and}\underset{u\mathrm{}}{lim\; sup}L(u)=\mathrm{}.$$
###### Proposition 2.4 (Karamata’s Theorem; direct half).
Let $`ZRV_q`$ be locally bounded in $`[A,\mathrm{})`$. Then
1. for any $`j(q+1)`$,
(2.4)
$$\underset{u\mathrm{}}{lim}\frac{u^{j+1}Z(u)}{_A^ux^jZ(x)𝑑x}=j+q+1.$$
2. for any $`j<(q+1)`$ (and for $`j=(q+1)`$ if $`^{\mathrm{}}x^{(q+1)}Z(x)𝑑x<\mathrm{}`$)
(2.5)
$$\underset{u\mathrm{}}{lim}\frac{u^{j+1}Z(u)}{_u^{\mathrm{}}x^jZ(x)𝑑x}=(j+q+1).$$
###### Proposition 2.5 (Karamata’s Theorem; converse half).
Let $`Z`$ be positive and locally integrable in $`[A,\mathrm{})`$.
1. If $`(\text{2.4})`$ holds for some $`j>(q+1)`$, then $`ZRV_q`$.
2. If $`(\text{2.5})`$ is satisfied for some $`j<(q+1)`$, then $`ZRV_q`$.
For a non-decreasing function $`H`$ on $`𝐑`$, we define the (left continuous) inverse of $`H`$ by
$$H^{}(y)=inf\{s:H(s)y\}.$$
###### Proposition 2.6 (see Proposition 0.8 in ).
We have
1. If $`ZRV_q`$, then $`lim_u\mathrm{}\mathrm{log}Z(u)/\mathrm{log}u=q`$.
2. If $`Z_1RV_{q_1}`$ and $`Z_2RV_{q_2}`$ with $`lim_u\mathrm{}Z_2(u)=\mathrm{}`$, then
$$Z_1Z_2RV_{q_1q_2}.$$
3. Suppose $`Z`$ is non-decreasing, $`Z(\mathrm{})=\mathrm{}`$, and $`ZRV_q`$, $`0<q<\mathrm{}`$. Then
$$Z^{}RV_{1/q}.$$
4. Suppose $`Z_1`$, $`Z_2`$ are non-decreasing and $`q`$-varying, $`0<q<\mathrm{}`$. Then for $`c(0,\mathrm{})`$
$$\underset{u\mathrm{}}{lim}\frac{Z_1(u)}{Z_2(u)}=c\text{if and only if }\underset{u\mathrm{}}{lim}\frac{Z_1^{}(u)}{Z_2^{}(u)}=c^{1/q}.$$
### 2.2. Auxiliary results
Based on regular variation theory, we prove here two results that have only been stated in .
###### Remark 2.3.
If $`fRV_{\rho +1}`$ ($`\rho >0`$) is continuous, then
(2.6)
$$\mathrm{\Xi }(u):=\frac{\sqrt{F(u)}}{f(u)_u^{\mathrm{}}[F(s)]^{1/2}𝑑s}\frac{\rho }{2(\rho +2)}\text{as }u\mathrm{},$$
where $`F`$ stands for an antiderivative of $`f`$. Indeed, by Proposition 2.4, we have
(2.7)
$$\underset{u\mathrm{}}{lim}\frac{F(u)}{uf(u)}=\frac{1}{\rho +2}\text{and}\underset{u\mathrm{}}{lim}\frac{u[F(u)]^{1/2}}{_u^{\mathrm{}}[F(s)]^{1/2}𝑑s}=\frac{\rho }{2}.$$
###### Lemma 2.1 (Properties of $`h`$).
If $`fRV_{\rho +1}`$ $`(\rho >0)`$ is continuous and $`k𝒦`$, then $`h`$ defined by (1.9) is a $`C^2`$-function satisfying the following:
1. $`\underset{t0}{lim}{\displaystyle \frac{h^{\prime \prime }(t)}{k^2(t)f(h(t)\xi )}}={\displaystyle \frac{2+\rho \mathrm{}_1}{\xi ^{\rho +1}(2+\rho )}}`$, for each $`\xi >0`$;
2. $`\underset{t0}{lim}{\displaystyle \frac{h(t)h^{\prime \prime }(t)}{[h^{}(t)]^2}}={\displaystyle \frac{2+\rho \mathrm{}_1}{2}}`$ and $`\underset{t0}{lim}{\displaystyle \frac{\mathrm{ln}k(t)}{\mathrm{ln}h(t)}}={\displaystyle \frac{\rho (\mathrm{}_11)}{2}}`$;
3. $`\underset{t0}{lim}{\displaystyle \frac{h^{}(t)}{th^{\prime \prime }(t)}}={\displaystyle \frac{\rho \mathrm{}_1}{2+\rho \mathrm{}_1}}`$ and $`\underset{t0}{lim}{\displaystyle \frac{h(t)}{t^2h^{\prime \prime }(t)}}={\displaystyle \frac{\rho ^2\mathrm{}_1^2}{2(2+\rho \mathrm{}_1)}}`$;
4. $`\underset{t0}{lim}{\displaystyle \frac{h(t)}{th^{}(t)}}=\underset{t0}{lim}{\displaystyle \frac{\mathrm{ln}t}{\mathrm{ln}h(t)}}={\displaystyle \frac{\rho \mathrm{}_1}{2}}`$;
5. $`\underset{t0}{lim}t^jh(t)=\mathrm{},`$ for all $`j>0`$, provided that $`k𝒦_0`$. If, in addition, $`k𝒦_{0,\zeta }`$ then $`\underset{t0}{lim}{\displaystyle \frac{1}{\zeta t^\zeta \mathrm{ln}h(t)}}=\underset{t0}{lim}{\displaystyle \frac{h^{}(t)}{t^{\zeta +1}h^{\prime \prime }(t)}}={\displaystyle \frac{\rho L_{}}{2(\zeta +1)}}.`$
###### Proof.
By (1.9), the function $`hC^2(0,\nu )`$, for some $`\nu >0`$, and $`lim_{t0}h(t)=\mathrm{}`$.
For any $`t(0,\nu )`$, we have $`h^{}(t)=k(t)\sqrt{2F(h(t))}`$ and
(2.8)
$$h^{\prime \prime }(t)=k^2(t)f(h(t))\left\{1+2\mathrm{\Xi }(h(t))\left[\left(\frac{_0^tk(s)𝑑s}{k(t)}\right)^{}1\right]\right\}.$$
Using Remark 2.3 and $`fRV_{\rho +1}`$, we reach (i).
(ii). By (i) and (2.7), we get
(2.9)
$$\underset{t0}{lim}\frac{h(t)h^{\prime \prime }(t)}{[h^{}(t)]^2}=\underset{t0}{lim}\frac{h^{\prime \prime }(t)}{k^2(t)f(h(t))}\frac{h(t)f(h(t))}{2F(h(t))}=\frac{2+\rho \mathrm{}_1}{2},$$
respectively
(2.10)
$$\underset{t0}{lim}\frac{k^{}(t)}{k(t)}\frac{h(t)}{h^{}(t)}=\underset{t0}{lim}\frac{h(t)f(h(t))}{F(h(t))}\frac{k^{}(t)\left(_0^tk(s)𝑑s\right)}{k^2(t)}\mathrm{\Xi }(h(t))=\frac{\rho (\mathrm{}_11)}{2}.$$
(iii). Using (i) and Remark 2.3, we find
$$\underset{t0}{lim}\frac{h^{}(t)}{th^{\prime \prime }(t)}=\frac{2(2+\rho )}{2+\rho \mathrm{}_1}\underset{t0}{lim}\frac{_0^tk(s)𝑑s}{tk(t)}\mathrm{\Xi }(h(t))=\frac{\rho \mathrm{}_1}{2+\rho \mathrm{}_1},$$
which, together with (2.9), implies that
$$\underset{t0}{lim}\frac{h(t)}{t^2h^{\prime \prime }(t)}=\underset{t0}{lim}\frac{h(t)h^{\prime \prime }(t)}{[h^{}(t)]^2}\left[\frac{h^{}(t)}{th^{\prime \prime }(t)}\right]^2=\frac{\rho ^2\mathrm{}_1^2}{2(2+\rho \mathrm{}_1)}.$$
(iv). If $`\mathrm{}_10`$, then by (iii), we have
$$\underset{t0}{lim}\frac{h(t)}{th^{}(t)}=\underset{t0}{lim}\frac{h(t)}{t^2h^{\prime \prime }(t)}\frac{th^{\prime \prime }(t)}{h^{}(t)}=\frac{\rho \mathrm{}_1}{2}.$$
If $`\mathrm{}_1=0`$, then we derive
(2.11)
$$\underset{t0}{lim}\frac{k(t)}{tk^{}(t)}=\underset{t0}{lim}\frac{k^2(t)}{k^{}(t)\left(_0^tk(s)𝑑s\right)}\frac{_0^tk(s)𝑑s}{tk(t)}=0.$$
This and (2.10) yield $`lim_{t0}\frac{h(t)}{th^{}(t)}=0`$, which concludes (iv).
(v). If $`k𝒦_0`$, then using (iv), we obtain $`lim_{t0}\mathrm{ln}[t^jh(t)]=\mathrm{}`$, for all $`j>0`$.
Suppose $`k𝒦_{0,\zeta }`$, for some $`\zeta >0`$. Then, $`lim_{t0}\frac{_0^tk(s)𝑑s}{t^{\zeta +1}k(t)}=\frac{L_{}}{\zeta +1}`$ and
(2.12)
$$\frac{L_{}}{\zeta +1}=\underset{t0}{lim}\frac{_0^tk(s)𝑑s}{t^{\zeta +1}k(t)}\frac{k^2(t)}{k^{}(t)\left(_0^tk(s)𝑑s\right)}=\underset{t0}{lim}\frac{k(t)}{t^{\zeta +1}k^{}(t)}=\frac{1}{\zeta }\underset{t0}{lim}\frac{1}{t^\zeta \mathrm{ln}k(t)}.$$
By (2.9), (2.10) and (2.12), we deduce
$$\underset{t0}{lim}\frac{h^{}(t)}{t^{\zeta +1}h^{\prime \prime }(t)}=\underset{t0}{lim}\frac{h(t)}{h^{}(t)t^{\zeta +1}}=\underset{t0}{lim}\frac{k^{}(t)h(t)}{k(t)h^{}(t)}\frac{k(t)}{t^{\zeta +1}k^{}(t)}=\frac{\rho L_{}}{2(\zeta +1)}.$$
This completes the proof of the lemma. ∎
Let $`\tau >0`$ be arbitrary and $`f`$ be as in Remark 2.3. For $`u>0`$ sufficiently large, we define
(2.13)
$$T_{1,\tau }(u)=\left[\frac{\rho }{2(\rho +2)}\mathrm{\Xi }(u)\right](\mathrm{ln}u)^\tau \text{and}T_{2,\tau }(u)=\left[\frac{f(\xi _0u)}{\xi _0f(u)}\xi _0^\rho \right](\mathrm{ln}u)^\tau .$$
###### Remark 2.4.
When $`f(u)=Cu^{\rho +1}`$, we have $`T_{1,\tau }(u)=T_{2,\tau }(u)=0`$.
###### Lemma 2.2.
Assume that $`f_{\rho \eta }`$ (where $`\rho 2<\eta 0`$). The following hold:
1. If $`f_{\rho 0,\tau }`$, then
$$\underset{u\mathrm{}}{lim}T_{1,\tau }(u)=\frac{\mathrm{}^{}}{(\rho +2)^2}\text{and}\underset{u\mathrm{}}{lim}T_{2,\tau }(u)=\xi _0^\rho \mathrm{}^{}\mathrm{ln}\xi _0.$$
2. If $`f_{\rho \eta }`$ with $`\eta 0`$, then
$$\underset{u\mathrm{}}{lim}T_{1,\tau }(u)=\underset{u\mathrm{}}{lim}T_{2,\tau }(u)=0.$$
###### Proof.
Using the second limit in (2.7), we obtain
$$\underset{u\mathrm{}}{lim}T_{1,\tau }(u)=\frac{\rho }{2}\underset{u\mathrm{}}{lim}\frac{\frac{\rho }{2(\rho +2)}_u^{\mathrm{}}[F(s)]^{1/2}𝑑s\frac{\sqrt{F(u)}}{f(u)}}{u[F(u)]^{1/2}(\mathrm{ln}u)^\tau }.$$
By L’Hospital’s rule, we arrive at
$$\underset{u\mathrm{}}{lim}T_{1,\tau }(u)=\underset{u\mathrm{}}{lim}\left[\frac{\rho +1}{\rho +2}\frac{F(u)f^{}(u)}{f^2(u)}\right](\mathrm{ln}u)^\tau :=\underset{u\mathrm{}}{lim}Q_{1,\tau }(u).$$
A simple calculation shows that, for $`u>0`$ large,
$`Q_{1,\tau }(u)`$ $`={\displaystyle \frac{(\mathrm{ln}u)^\tau }{\rho +2}}\left[\rho +1{\displaystyle \frac{uf^{}(u)}{f(u)}}\right]+{\displaystyle \frac{uf^{}(u)}{f(u)}}\left[{\displaystyle \frac{1}{\rho +2}}{\displaystyle \frac{F(u)}{uf(u)}}\right](\mathrm{ln}u)^\tau `$
$`=:{\displaystyle \frac{1}{\rho +2}}Q_{2,\tau }(u)+{\displaystyle \frac{uf^{}(u)}{f(u)}}Q_{3,\tau }(u).`$
Since (1.11) holds with $`\varphi RV_\eta `$ or $`\varphi RV_\eta `$, we can assume $`B>0`$ such that $`\varphi 0`$ on $`[B,\mathrm{})`$. For any $`u>B`$, we have $`Q_{2,\tau }(u)=\varphi (u)(\mathrm{ln}u)^\tau `$ and
$$Q_{3,\tau }(u)=\stackrel{~}{C}\frac{(\mathrm{ln}u)^\tau }{uf(u)}+\frac{_B^uf(s)\varphi (s)𝑑s}{(\rho +2)uf(u)\varphi (u)}\varphi (u)(\mathrm{ln}u)^\tau ,$$
where $`\stackrel{~}{C}𝐑`$ is a constant. Since either $`f\varphi RV_{\rho +\eta +1}`$ or $`f\varphi RV_{\rho +\eta +1}`$, by Proposition 2.4,
$$\underset{u\mathrm{}}{lim}\frac{uf(u)\varphi (u)}{_B^uf(x)\varphi (x)𝑑x}=\rho +\eta +2.$$
If (i) holds, then $`lim_u\mathrm{}Q_{2,\tau }(u)=\mathrm{}^{}`$ and $`lim_u\mathrm{}Q_{3,\tau }(u)=\mathrm{}^{}(\rho +2)^2`$. Thus,
$$\underset{u\mathrm{}}{lim}T_{1,\tau }(u)=\underset{u\mathrm{}}{lim}Q_{1,\tau }(u)=\mathrm{}^{}/(\rho +2)^2.$$
If (ii) holds, then by Proposition 2.3, we have $`lim_u\mathrm{}(\mathrm{ln}u)^\tau \varphi (u)=0`$. It follows that
$$\underset{u\mathrm{}}{lim}Q_{2,\tau }(u)=\underset{u\mathrm{}}{lim}Q_{3,\tau }(u)=0$$
which yields $`lim_u\mathrm{}T_{1,\tau }(u)=0`$. Note that the proof is finished if $`\xi _0=1`$, since $`T_{2,\tau }(u)=0`$ for each $`u>0`$.
Arguing by contradiction, let us suppose that $`\xi _01`$. Then, by (1.11),
$$T_{2,\tau }(u)=\xi _0^\rho \left[\mathrm{exp}\left\{_u^{\xi _0u}\frac{\varphi (t)}{t}𝑑t\right\}1\right](\mathrm{ln}u)^\tau ,u>B/\xi _0.$$
But, $`lim_u\mathrm{}\varphi (us)/s=0`$, uniformly with respect to $`s[\xi _0,1]`$. So
$$\underset{u\mathrm{}}{lim}_u^{\xi _0u}\frac{\varphi (t)}{t}𝑑t=\underset{u\mathrm{}}{lim}_1^{\xi _0}\frac{\varphi (su)}{s}𝑑s=0$$
which leads to
$$\underset{u\mathrm{}}{lim}T_{2,\tau }(u)=\xi _0^\rho \underset{u\mathrm{}}{lim}\left(_u^{\xi _0u}\frac{\varphi (t)}{t}𝑑t\right)(\mathrm{ln}u)^\tau .$$
If (i) occurs, then by Proposition 2.1, we have
$$\underset{u\mathrm{}}{lim}T_{2,\tau }(u)=\xi _0^\rho \underset{u\mathrm{}}{lim}(\mathrm{ln}u)^\tau \varphi (u)_1^{\xi _0}\frac{\varphi (tu)}{\varphi (u)}\frac{dt}{t}=\xi _0^\rho \mathrm{}^{}\mathrm{ln}\xi _0.$$
If (ii) occurs, then by Proposition 2.3, we infer that
$$\underset{u\mathrm{}}{lim}T_{2,\tau }(u)=\frac{\xi _0^\rho }{\tau }\underset{u\mathrm{}}{lim}\left[\varphi (\xi _0u)\varphi (u)\right](\mathrm{ln}u)^{\tau +1}=0.$$
The proof of Lemma 2.2 is now complete. ∎
###### Lemma 2.3.
If $`k𝒦_{(01],\tau }`$ and $`f`$ satisfies either (i) or (ii) of Theorem 1.3, then
(2.14)
$$(t):=(\mathrm{ln}t)^\tau \left(1\frac{k^2(t)f(\xi _0h(t))}{\xi _0h^{\prime \prime }(t)}\right)\rho \stackrel{~}{\chi }\text{as }t0,$$
where $`\stackrel{~}{\chi }`$ is defined by (1.14).
###### Proof.
Using (2.8), we write $`(t)=\frac{k^2(t)f(h(t))}{h^{\prime \prime }(t)}_{i=1}^3_i(t)`$, for $`t>0`$ small, where
$$\{\begin{array}{cc}\hfill _1(t)& :=2\mathrm{\Xi }(h(t))(\mathrm{ln}t)^\tau \left[\left(\frac{_0^tk(s)𝑑s}{k(t)}\right)^{}\mathrm{}_1\right],\hfill \\ \hfill _2(t)& :=2(1\mathrm{}_1)\left(\frac{\mathrm{ln}t}{\mathrm{ln}h(t)}\right)^\tau T_{1,\tau }(h(t))\text{and}_3(t):=\left(\frac{\mathrm{ln}t}{\mathrm{ln}h(t)}\right)^\tau T_{2,\tau }(h(t)).\hfill \end{array}$$
By Remark 2.3, we find $`lim_{t0}_1(t)=\rho L_{\mathrm{}}/(\rho +2)`$.
*Case* (i) (that is, $`f_{\rho \eta }`$ with $`\eta L_{\mathrm{}}0`$). By Lemmas 2.1 and 2.2, it turns out that
$$\underset{t0}{lim}_2(t)=\underset{t0}{lim}_3(t)=0\text{and}\underset{t0}{lim}(t)=\frac{\rho L_{\mathrm{}}}{2+\rho \mathrm{}_1}=:\rho \stackrel{~}{\chi }.$$
*Case* (ii) (that is, $`f_{\rho 0,\tau }`$ with $`[\mathrm{}^{}(\mathrm{}_11)]^2+L_{\mathrm{}}^20`$). By Lemmas 2.1 and 2.2, we get
$$\underset{t0}{lim}_2(t)=\frac{2(1\mathrm{}_1)\mathrm{}^{}}{(\rho +2)^2}\left(\frac{\rho \mathrm{}_1}{2}\right)^\tau \text{and}\underset{t0}{lim}_3(t)=\frac{\mathrm{}^{}(2+\rho \mathrm{}_1)}{(2+\rho )}\left(\frac{\rho \mathrm{}_1}{2}\right)^\tau \mathrm{ln}\xi _0.$$
Thus, we arrive at
$$\underset{t0}{lim}(t)=\frac{\rho L_{\mathrm{}}}{2+\rho \mathrm{}_1}\mathrm{}^{}\left(\frac{\rho \mathrm{}_1}{2}\right)^\tau [\frac{2(1\mathrm{}_1)}{(\rho +2)(2+\rho \mathrm{}_1)}+\mathrm{ln}\xi _0]=:\rho \stackrel{~}{\chi }.$$
This finishes the proof. ∎
## 3. Characterization of $`𝒦`$ and its subclasses
Definition 2.1 extends to *regular variation at the origin*. We say that $`Z`$ is regularly varying (on the right) at the origin with index $`q`$ (and write, $`ZRV_q(0+))`$ if $`Z(1/u)RV_q`$. Moreover, by $`ZNRV_q(0+)`$ we mean that $`Z(1/u)NRV_q`$. The meaning of $`NRV_q`$ is given by (2.2).
###### Proposition 3.1.
We have $`k𝒦_{(01]}`$ if and only if $`k`$ is non-decreasing near the origin and $`k`$ belongs to $`NRV_\alpha (0+)`$ for some $`\alpha 0`$ (where $`\alpha =1/\mathrm{}_11`$).
###### Proof.
If $`k𝒦_{(01]}`$, then from the definition
$$\underset{t0^+}{lim}\frac{_0^tk(s)𝑑s}{k(t)}/t=\underset{t0^+}{lim}\left(\frac{_0^tk(s)𝑑s}{k(t)}\right)^{}=\mathrm{}_1,$$
which implies that
$$\underset{u\mathrm{}}{lim}\frac{u\frac{d}{du}k(1/u)}{k(1/u)}=\underset{t0^+}{lim}\frac{tk^{}(t)}{k(t)}=\frac{\mathrm{}_11}{\mathrm{}_1}.$$
Thus $`k(1/u)`$ belongs to $`NRV_{11/\mathrm{}_1}`$. Conversely, if $`k`$ belongs to $`NRV_\alpha (0+)`$ with $`\alpha 0`$, then $`k`$ is a positive $`C^1`$-function on some interval $`(0,\nu )`$ and
(3.1)
$$\underset{t0^+}{lim}\frac{tk^{}(t)}{k(t)}=\alpha .$$
By Proposition 2.4, we deduce
(3.2)
$$\underset{t0^+}{lim}\frac{_0^tk(s)𝑑s}{tk(t)}=\underset{u\mathrm{}}{lim}\frac{_u^{\mathrm{}}x^2k(1/x)𝑑x}{u^1k(1/u)}=\frac{1}{1+\alpha }.$$
Combining (3.1) and (3.2), we get $`lim_{t0^+}\left(_0^tk(s)𝑑s/k(t)\right)^{}=1/(1+\alpha )`$. If, in addition, $`k`$ is non-decreasing near $`0`$, then $`k𝒦`$ with $`\mathrm{}_1=1/(1+\alpha )`$. Note that by (3.1), $`k`$ is increasing near the origin if $`\alpha >0`$; however, when $`k`$ is slowly varying at $`0`$, then we cannot draw any conclusion about the monotonicity of $`k`$ near the origin (see Remark 2.2). ∎
###### Remark 3.1.
By Propositions 3.1 and 2.1, we deduce $`k𝒦_{(01]}`$ if and only $`k`$ is of the form
(3.3)
$$k(t)=c_0t^\alpha \mathrm{exp}\left\{_t^{c_1}\frac{E(y)}{y}𝑑y\right\}(0<t<c_1),\text{for some }0\alpha (=1/\mathrm{}_11)$$
where $`c_0,c_1>0`$ are constants, $`EC[0,c_1)`$ with $`E(0)=0`$ and (only for $`\mathrm{}_1=1`$) $`E(t)\alpha `$.
###### Proposition 3.2.
We have $`k𝒦_{(01],\tau }`$ if and only if $`k`$ is of the form (3.3) where, in addition,
(3.4)
$$\underset{t0}{lim}(\mathrm{ln}t)^\tau E(t)=\mathrm{}_{\mathrm{}}𝐑\text{with }\mathrm{}_{\mathrm{}}=(1+\alpha )^2L_{\mathrm{}}.$$
###### Proof.
Suppose $`k`$ satisfies (3.3) and (3.4). A simple calculation leads to
(3.5)
$$\underset{t0}{lim}(\mathrm{ln}t)^\tau \left[\frac{1\mathrm{}_1}{\mathrm{}_1}\frac{tk^{}(t)}{k(t)}\right]=\underset{t0}{lim}(\mathrm{ln}t)^\tau E(t)=\mathrm{}_{\mathrm{}}.$$
By L’Hospital’s rule, we find
(3.6) $`\underset{t0}{lim}(\mathrm{ln}t)^\tau \left[\mathrm{}_1{\displaystyle \frac{_0^tk(s)𝑑s}{tk(t)}}\right]`$ $`=\underset{t0}{lim}{\displaystyle \frac{(\mathrm{}_11)+\mathrm{}_1tk^{}(t)/k(t)}{(\mathrm{ln}t)^\tau \left[1+\frac{tk^{}(t)}{k(t)}\frac{\tau }{\mathrm{ln}t}\right]}}`$
$`=\mathrm{}_1^2\underset{t0}{lim}(\mathrm{ln}t)^\tau \left[{\displaystyle \frac{1\mathrm{}_1}{\mathrm{}_1}}{\displaystyle \frac{tk^{}(t)}{k(t)}}\right]={\displaystyle \frac{\mathrm{}_{\mathrm{}}}{(\alpha +1)^2}}.`$
We see that, for each $`t>0`$ small,
(3.7)
$$\left(\frac{_0^tk(s)𝑑s}{k(t)}\right)^{}\mathrm{}_1=\frac{tk^{}(t)}{k(t)}\left[\mathrm{}_1\frac{_0^tk(s)𝑑s}{tk(t)}\right]+\mathrm{}_1\left[\frac{1\mathrm{}_1}{\mathrm{}_1}\frac{tk^{}(t)}{k(t)}\right].$$
By (3.5)–(3.7), we infer that $`k𝒦_{(01],\tau }`$ with $`L_{\mathrm{}}=\mathrm{}_{\mathrm{}}/(1+\alpha )^2`$.
Conversely, if $`k𝒦_{(01],\tau }`$, then $`k`$ is of the form (3.3). Moreover, we have
(3.8)
$$\underset{t0}{lim}(\mathrm{ln}t)^\tau \left(\frac{_0^tk(s)𝑑s}{tk(t)}\mathrm{}_1\right)=\underset{t0}{lim}\frac{\left(_0^tk(s)𝑑s/k(t)\right)^{}\mathrm{}_1}{(\mathrm{ln}t)^\tau \left(1\frac{\tau }{\mathrm{ln}t}\right)}=L_{\mathrm{}}.$$
By (3.7) and (3.8), we deduce
$$L_{\mathrm{}}=\alpha L_{\mathrm{}}+\frac{1}{\alpha +1}\underset{t0}{lim}(\mathrm{ln}t)^\tau E(t).$$
Consequently, $`lim_{t0}(\mathrm{ln}t)^\tau E(t)=(1+\alpha )^2L_{\mathrm{}}`$. Hence, (3.4) holds. ∎
###### Proposition 3.3.
We have $`k𝒦_0`$ if and only if $`k`$ is of the form
(3.9)
$$k(t)=d_0\left(\mathrm{exp}\left\{_t^{d_1}\frac{dx}{x𝒲(x)}\right\}\right)^{}(0<t<d_1),$$
where $`d_0,d_1>0`$ are constants and $`0<𝒲C^1(0,d_1)`$ satisfies $`lim_{t0}𝒲(t)=lim_{t0}t𝒲^{}(t)=0`$.
###### Proof.
If $`k𝒦_0`$, then we set
(3.10)
$$𝒲(t)=\frac{_0^tk(s)𝑑s}{tk(t)},\text{for }t(0,d_1).$$
Hence, $`lim_{t0}𝒲(t)=0`$ and, for $`t>0`$ small,
$$t𝒲^{}(t)=\left(\frac{_0^tk(s)𝑑s}{k(t)}\right)^{}\frac{_0^tk(s)𝑑s}{tk(t)}.$$
It follows that $`lim_{t0}t𝒲^{}(t)=0`$. By (3.10), we find
$$_t^{d_1}\frac{dx}{x𝒲(x)}=\mathrm{ln}\left(_0^{d_1}k(s)𝑑s\right)\mathrm{ln}\left(_0^tk(s)𝑑s\right),t(0,d_1)$$
so that (3.9) is fulfilled. Conversely, if (3.9) holds, then $`lim_{t0}_t^{d_1}\frac{dx}{x𝒲(x)}=\mathrm{}`$ and
(3.11)
$$_0^tk(s)𝑑s=d_0\mathrm{exp}\left\{_t^{d_1}\frac{dx}{x𝒲(x)}\right\}=tk(t)𝒲(t),t(0,d_1).$$
This, together with the properties of $`𝒲`$, shows that $`k𝒦_0`$. ∎
###### Proposition 3.4.
We have $`k𝒦_{0,\zeta }`$ if and only if $`k`$ is of the form (3.9) where, in addition,
(3.12)
$$\underset{t0}{lim}t^{1\zeta }𝒲^{}(t)=\mathrm{}_{}\text{with }\mathrm{}_{}=\zeta L_{}/(1+\zeta ).$$
###### Proof.
If $`k𝒦_{0,\zeta }`$, then (3.9) and (3.11) are fulfilled. Therefore,
$$L_{}=\underset{t0}{lim}\frac{(t𝒲(t))^{}}{t^\zeta }=\underset{t0}{lim}\frac{𝒲(t)+t𝒲^{}(t)}{t^\zeta }\text{and }\frac{L_{}}{\zeta +1}=\underset{t0}{lim}\frac{_0^tk(s)𝑑s}{k(t)t^{\zeta +1}}=\underset{t0}{lim}\frac{𝒲(t)}{t^\zeta },$$
from which (3.12) follows. Conversely, if (3.9) and (3.12) hold, then $`lim_{t0}𝒲(t)/t^\zeta =\mathrm{}_{}/\zeta `$. By (3.11), we infer that
$$\frac{1}{t^\zeta }\left(\frac{_0^tk(s)𝑑s}{k(t)}\right)^{}=\frac{1}{t^\zeta }(𝒲(t)+t𝒲^{}(t))\frac{\mathrm{}_{}(\zeta +1)}{\zeta }\text{as }t0.$$
Thus, $`k𝒦_{0,\zeta }`$ with $`L_{}=\mathrm{}_{}(\zeta +1)/\zeta `$. ∎
###### Remark 3.2.
If $`k𝒦_0`$ or $`k𝒦_{(01],\tau }`$ with $`(1\mathrm{}_1)^2+L_{\mathrm{}}^20`$, then
(3.13)
$$\underset{t0}{lim}\frac{k^{}(t)}{k(t)t^{\theta 1}}=\mathrm{},\text{for every}\theta >0.$$
Indeed, if $`k𝒦_0`$, then $`lim_{t0}\frac{tk^{}(t)}{k(t)}=\mathrm{}`$. Assuming that $`k𝒦_{(01],\tau }`$, we deduce (3.13) from (3.1) when $`\mathrm{}_11`$, otherwise from (3.4) when $`L_{\mathrm{}}0`$ since
$$\underset{t0}{lim}\frac{k^{}(t)}{k(t)t^{\theta 1}}=\underset{t0}{lim}E(t)t^\theta =L_{\mathrm{}}\underset{t0}{lim}\frac{t^\theta }{(\mathrm{ln}t)^\tau }=\mathrm{}.$$
###### Definition 3.1 (see ).
A non-decreasing function $`U`$ is $`\mathrm{\Gamma }`$-varying at $`\mathrm{}`$ if $`U`$ is defined on an interval $`(A,\mathrm{})`$, $`lim_x\mathrm{}U(x)=\mathrm{}`$ and there is $`g:(A,\mathrm{})(0,\mathrm{})`$ such that
$$\underset{y\mathrm{}}{lim}\frac{U(y+\lambda g(y))}{U(y)}=e^\lambda ,\lambda 𝐑.$$
The function $`g`$ is called an auxiliary function and is unique up to asymptotic equivalence.
###### Remark 3.3.
Under the assumptions of Theorem 1.1, we have
1. Suppose $`lim_{t0}\left(_0^tk(s)𝑑s\right)^2r(t)=1`$ and let $`\widehat{f}(u)`$ be such that $`lim_u\mathrm{}\widehat{f}(u)/f(u)=1`$ and $`j(u):=\widehat{f}(u)/u`$ is non-decreasing for $`u>0`$ large. Then $`lim_{t0}\widehat{\phi }(t)/\phi (t)=1`$, where $`\phi (t)`$ is given by (1.7) and $`\widehat{\phi }(t)=j^{}(r(t))`$ for $`t>0`$ small.
2. If $`k𝒦`$ with $`\mathrm{}_10`$, then $`\phi (1/u)RV_{2/(\rho \mathrm{}_1)}`$.
3. If $`k𝒦_0`$, then $`\phi (1/u)`$ is $`\mathrm{\Gamma }`$-varying at $`u=\mathrm{}`$ with auxiliary function $`{\displaystyle \frac{\rho u^2_0^{1/u}k(s)𝑑s}{2k(1/u)}}`$.
4. $`lim_{t0}\phi (t)/h(t)=[2(\rho +2)/\rho ^2]^{1/\rho }`$, where $`h(t)`$ is given by (1.9).
Indeed, by Proposition 2.6 we find $`(f(u)/u)^{}RV_{1/\rho }`$ and $`lim_u\mathrm{}(f(u)/u)^{}/j^{}(u)=1`$. Then, by Proposition 2.1 we deduce (a). We see that (b) follows by Proposition 2.6 since $`\left(_0^{1/u}k(s)𝑑s\right)^2RV_{2/\mathrm{}_1}`$ (cf. Proposition 3.1) and $`f(u)/uRV_\rho `$. If $`k𝒦_0`$, then by Proposition 3.3 and \[47, p. 106\], we get $`\left(_0^{1/u}k(s)𝑑s\right)^2`$ is $`\mathrm{\Gamma }`$-varying at $`u=\mathrm{}`$ with auxiliary function $`u𝒲(1/u)/2`$. By \[47, p. 36\], we conclude (c). Notice that $`Y(u):=\left(1/_u^{\mathrm{}}[2F(s)]^{1/2}𝑑s\right)^2RV_\rho `$ and $`Y(h(t))=\left(_0^tk(s)𝑑s\right)^2`$ for $`t>0`$ small. We have $`lim_u\mathrm{}f(u)/[uY(u)]=2(\rho +2)/\rho ^2`$ (cf. Remark 2.3). By Proposition 2.6, we achieve (d).
## 4. Proof of Theorem 1.1
Fix $`a(\mathrm{},\lambda _{\mathrm{},1})`$. By \[12, Theorem 1.1\], equation (1.1) has at least a large solution.
In what follows, we will prove that (1.6) holds for any large solution. Hence, a standard argument leads to the uniqueness (see, for instance, or ).
By virtue of Remark 3.3 (d), it is enough to demonstrate (1.8). Let $`u_a`$ denote an arbitrary large solution of (1.1). Fix $`\epsilon (0,1/2)`$ and choose $`\delta >0`$ such that
1. $`d(x)`$ is a $`C^2`$ function on the set $`\{x\mathrm{\Omega }:d(x)<\delta \}`$;
2. $`k`$ is non-decreasing on $`(0,\delta )`$;
3. $`1\epsilon <b(x)/k^2(d(x))<1+\epsilon `$, $`x\mathrm{\Omega }`$ with $`0<d(x)<\delta `$ (since (1.5) holds);
4. $`h^{}(t)<0`$ and $`h^{\prime \prime }(t)>0`$ for each $`t(0,\delta )`$ (cf. Lemma 2.1).
Define $`\xi ^\pm =\left[\frac{2+\mathrm{}_1\rho }{(12\epsilon )(2+\rho )}\right]^{1/\rho }`$ and $`u^\pm (x)=\xi ^\pm h(d(x))`$, for any $`x`$ with $`d(x)(0,\delta )`$.
The proof of (1.8) will be divided into three steps:
*Step* 1. There exists $`\delta _1(0,\delta )`$ small such that
(4.1)
$$\{\begin{array}{cc}& \mathrm{\Delta }u^++au^+(1\epsilon )k^2(d)f(u^+)0,x\text{ with }d(x)(0,\delta _1)\hfill \\ & \mathrm{\Delta }u^{}+au^{}(1+\epsilon )k^2(d)f(u^{})0,x\text{ with }d(x)(0,\delta _1).\hfill \end{array}$$
Indeed, for every $`x\mathrm{\Omega }`$ with $`0<d(x)<\delta `$, we have
(4.2) $`\mathrm{\Delta }u^\pm +au^\pm (1\epsilon )k^2(d)f(u^\pm )`$
$`=\xi ^\pm h^{\prime \prime }(d)(1+a{\displaystyle \frac{h(d)}{h^{\prime \prime }(d)}}+\mathrm{\Delta }d{\displaystyle \frac{h^{}(d)}{h^{\prime \prime }(d)}}(1\epsilon ){\displaystyle \frac{k^2(d)f(u^\pm )}{\xi ^\pm h^{\prime \prime }(d)}})=:\xi ^\pm h^{\prime \prime }(d)B^\pm (d).`$
By Lemma 2.1, we deduce $`lim_{d0}B^\pm (d)=\epsilon /(12\epsilon )`$, which proves (4.1).
*Step* 2. There exists $`M^+`$, $`\delta ^+>0`$ such that
$$u_a(x)u^+(x)+M^+,x\mathrm{\Omega }\text{with}0<d<\delta ^+.$$
For $`x\mathrm{\Omega }`$ with $`d(x)(0,\delta _1)`$, we define $`\mathrm{\Psi }_x(u)=aub(x)f(u)`$ for each $`u>0`$. By Lemma 2.1,
(4.3)
$$\underset{d(x)0}{lim}\frac{b(x)f(u^+(x))}{u^+(x)}=\underset{d0}{lim}\frac{k^2(d)f(u^+)}{\xi ^+h^{\prime \prime }(d)}\frac{h^{\prime \prime }(d)}{h(d)}=\mathrm{}.$$
From this and $`(A_1)`$, we infer that there exists $`\delta _2(0,\delta _1)`$ such that, for any $`x`$ with $`0<d(x)<\delta _2`$,
$$u\mathrm{\Psi }_x(u)\text{is decreasing on some interval }(u_x,\mathrm{})\text{with }0<u_x<u^+(x).$$
Hence, for each $`M>0`$, we have
(4.4)
$$\mathrm{\Psi }_x(u^+(x)+M)\mathrm{\Psi }_x(u^+(x)),x\mathrm{\Omega }\text{ with }0<d(x)<\delta _2.$$
Fix $`\sigma (0,\delta _2/4)`$ and set $`𝒩_\sigma :=\{x\mathrm{\Omega }:\sigma <d(x)<\delta _2/2\}`$.
We define $`u_\sigma ^{}(x)=u^+(d\sigma ,s)+M^+`$, where $`(d,s)`$ are the local coordinates of $`x𝒩_\sigma `$. We choose $`M^+>0`$ large enough such that
$$u_\sigma ^{}(\delta _2/2,s)=u^+(\delta _2/2\sigma ,s)+M^+u_a(\delta _2/2,s),\sigma (0,\delta _2/4)\text{ and }s\mathrm{\Omega }.$$
By (ii), (iii), (4.1) and (4.4), we obtain
$`\mathrm{\Delta }u_\sigma ^{}(x)`$ $`au^+(d\sigma ,s)(1\epsilon )k^2(d\sigma )f(u^+(d\sigma ,s))`$
$`au^+(d\sigma ,s)b(x)f(u^+(d\sigma ,s))`$
$`a(u^+(d\sigma ,s)+M^+)b(x)f(u^+(d\sigma ,s)+M^+)`$
$`=au_\sigma ^{}(x)b(x)f(u_\sigma ^{}(x))\text{in }𝒩_\sigma .`$
So, uniformly with respect to $`\sigma `$, we have
(4.5)
$$\mathrm{\Delta }u_\sigma ^{}(x)+au_\sigma ^{}(x)b(x)f(u_\sigma ^{}(x))\text{in }𝒩_\sigma .$$
Since $`u_\sigma ^{}(x)\mathrm{}`$ as $`d\sigma `$, from \[12, Lemma 2.1\], we get $`u_au_\sigma ^{}`$ in $`𝒩_\sigma `$, for every $`\sigma (0,\delta _2/4)`$. Letting $`\sigma 0`$, we achieve the assertion of Step 2 (with $`\delta ^+(0,\delta _2/2)`$ arbitrarily chosen).
*Step* 3. There exists $`M^{}`$, $`\delta ^{}>0`$ such that
(4.6)
$$u_a(x)u^{}(x)M^{},x=(d,s)\mathrm{\Omega }\text{with}0<d<\delta ^{}.$$
For every $`r(0,\delta )`$, define $`\mathrm{\Omega }_r=\{x\mathrm{\Omega }:0<d(x)<r\}`$.
Fix $`\sigma (0,\delta _2/4)`$. We define $`v_\sigma ^{}(x)=\lambda u^{}(d+\sigma ,s)`$ for $`x=(d,s)\mathrm{\Omega }_{\delta _2/2}`$, where $`\lambda (0,1)`$ is chosen small enough such that
(4.7)
$$v_\sigma ^{}(\delta _2/4,s)=\lambda u^{}(\delta _2/4+\sigma ,s)u_a(\delta _2/4,s),\sigma (0,\delta _2/4),s\mathrm{\Omega }.$$
Notice that $`lim\; sup_{d0}(v_\sigma ^{}u_a)(x)=\mathrm{}`$. By (ii), (iii), (4.1) and $`(A_1)`$, we have
$`\mathrm{\Delta }v_\sigma ^{}(x)+av_\sigma ^{}(x)`$ $`=\lambda (\mathrm{\Delta }u^{}(d+\sigma ,s)+au^{}(d+\sigma ,s))`$
$`\lambda (1+\epsilon )k^2(d+\sigma )f(u^{}(d+\sigma ,s))(1+\epsilon )k^2(d)f(\lambda u^{}(d+\sigma ,s))`$
$`b(x)f(v_\sigma ^{}(x)),x=(d,s)\mathrm{\Omega }_{\delta _2/4}.`$
Using \[12, Lemma 2.1\], we derive $`v_\sigma ^{}u_a`$ in $`\mathrm{\Omega }_{\delta _2/4}`$. Letting $`\sigma 0`$, we get
(4.8)
$$\lambda u^{}(x)u_a(x),x\mathrm{\Omega }_{\delta _2/4}.$$
By Lemma 2.1, $`lim_{d0}k^2(d)f(\lambda ^2u^{})/u^{}=\mathrm{}`$. Thus, there exists $`\stackrel{~}{\delta }(0,\delta _2/4)`$ such that
(4.9)
$$k^2(d)f(\lambda ^2u^{})/u^{}\lambda ^2|a|,x\mathrm{\Omega }\text{with}0<d\stackrel{~}{\delta }.$$
Choose $`\delta _{}(0,\stackrel{~}{\delta })`$, sufficiently close to $`\stackrel{~}{\delta }`$, such that
(4.10)
$$h(\delta _{})/h(\stackrel{~}{\delta })<1+\lambda .$$
For each $`\sigma (0,\stackrel{~}{\delta }\delta _{})`$, we define $`z_\sigma (x)=u^{}(d+\sigma ,s)(1\lambda )u^{}(\delta _{},s)`$, where $`x=(d,s)\mathrm{\Omega }_\delta _{}`$. We prove that $`z_\sigma `$ is positive in $`\mathrm{\Omega }_\delta _{}`$ and
(4.11)
$$\mathrm{\Delta }z_\sigma +az_\sigma b(x)f(z_\sigma )\text{in }\mathrm{\Omega }_\delta _{}.$$
By (iv), $`u^{}(x)`$ decreases with $`d`$ when $`d<\stackrel{~}{\delta }`$. This and (4.10) imply that
(4.12)
$$1+\lambda >\frac{u^{}(\delta _{},s)}{u^{}(\stackrel{~}{\delta },s)}\frac{u^{}(\delta _{},s)}{u^{}(d+\sigma ,s)},x=(d,s)\mathrm{\Omega }_\delta _{}.$$
Hence,
(4.13)
$$z_\sigma (x)=u^{}(d+\sigma ,s)\left(1\frac{(1\lambda )u^{}(\delta _{},s)}{u^{}(d+\sigma ,s)}\right)\lambda ^2u^{}(d+\sigma ,s)>0,x\mathrm{\Omega }_\delta _{}.$$
By (4.1), (ii) and (iii), we see that (4.11) follows if
(4.14)
$$(1+\epsilon )k^2(d+\sigma )\left[f(u^{}(d+\sigma ,s))f(z_\sigma (d,s))\right]a(1\lambda )u^{}(\delta _{},s),(d,s)\mathrm{\Omega }_\delta _{}.$$
The Lagrange mean value theorem and $`(A_1)`$ show that
(4.15)
$$f(u^{}(d+\sigma ,s))f(z_\sigma (d,s))(1\lambda )u^{}(\delta _{},s)f(z_\sigma (x))/z_\sigma (x)$$
which, combined with (4.9) and (4.13), proves (4.14).
Notice that $`lim\; sup_{d0}(z_\sigma u_a)(x)=\mathrm{}`$. By (4.8), we have
$$z_\sigma (x)=u^{}(\delta _{}+\sigma ,s)(1\lambda )u^{}(\delta _{},s)\lambda u^{}(\delta _{},s)u_a(x),x=(\delta _{},s)\mathrm{\Omega }.$$
By \[12, Lemma 2.1\], $`z_\sigma u_a`$ in $`\mathrm{\Omega }_\delta _{}`$, for every $`\sigma (0,\stackrel{~}{\delta }\delta _{})`$. Letting $`\sigma 0`$, we conclude Step 3.
Thus, by Steps 2 and 3, we have
$$\xi ^{}\underset{d(x)0}{lim\; inf}\frac{u_a(x)}{h(d(x))}\underset{d(x)0}{lim\; sup}\frac{u_a(x)}{h(d(x))}\xi ^+.$$
Taking $`\epsilon 0`$, we obtain (1.8). This finishes the proof of Theorem 1.1. ∎
## 5. Proof of Theorem 1.3
Fix $`a<\lambda _{\mathrm{},1}`$ and denote by $`u_a`$ the unique large solution of (1.1).
Let $`\epsilon (0,1/2)`$ be arbitrary and $`\delta >0`$ be such that (i), (ii), (iv) from §4 are satisfied.
By (1.10) and Remark 3.2, we can diminish $`\delta >0`$ such that
(5.1)
$$\{\begin{array}{cc}& 1+(\stackrel{~}{c}\epsilon )d^\theta <b(x)/k^2(d)<1+(\stackrel{~}{c}+\epsilon )d^\theta ,x\mathrm{\Omega }\text{with}d(0,\delta ),\hfill \\ & k^2(t)\left[1+(\stackrel{~}{c}\epsilon )t^\theta \right]\text{is increasing on}(0,\delta ).\hfill \end{array}$$
Define $`u^\pm (x)=\xi _0h(d)\left[1+\chi _\epsilon ^\pm (\mathrm{ln}d)^\tau \right]`$ for $`x\mathrm{\Omega }`$ with $`d(0,\delta )`$, where $`\chi _\epsilon ^\pm =\stackrel{~}{\chi }\pm \epsilon `$.
We can assume $`u^\pm (x)>0`$ for every $`x\mathrm{\Omega }`$ with $`d(x)(0,\delta )`$.
By the Lagrange mean value theorem, we obtain
$$f(u^\pm (x))=f(\xi _0h(d))+\xi _0\chi _\epsilon ^\pm \frac{h(d)}{(\mathrm{ln}d)^\tau }f^{}(\mathrm{\Psi }^\pm (d)),$$
where $`\mathrm{\Psi }^\pm (d)=\xi _0h(d)\left[1+\chi _\epsilon ^\pm \lambda ^\pm (d)(\mathrm{ln}d)^\tau \right],`$ for some $`\lambda ^\pm (d)[0,1]`$.
Since $`f(u)/u^{\rho +1}`$ is slowly varying, by Proposition 2.1 we find
(5.2)
$$\underset{d0}{lim}\frac{f(\mathrm{\Psi }^\pm (d))}{f(\xi _0h(d))}=\underset{d0}{lim}\frac{f(u^\pm (d))}{f(\xi _0h(d))}=1.$$
*Step* 1. There exists $`\delta _1(0,\delta )`$ so that
(5.3)
$$\{\begin{array}{cc}& \mathrm{\Delta }u^++au^+k^2(d)[1+(\stackrel{~}{c}\epsilon )d^\theta ]f(u^+)0,x\mathrm{\Omega }\text{ with }d<\delta _1,\hfill \\ & \mathrm{\Delta }u^{}+au^{}k^2(d)[1+(\stackrel{~}{c}+\epsilon )d^\theta ]f(u^{})0,x\mathrm{\Omega }\text{ with }d<\delta _1.\hfill \end{array}$$
For every $`x\mathrm{\Omega }`$ with $`d(0,\delta )`$, we have
(5.4)
$$\mathrm{\Delta }u^\pm +au^\pm k^2(d)\left[1+(\stackrel{~}{c}\epsilon )d^\theta \right]f(u^\pm )=\xi _0\frac{h^{\prime \prime }(d)}{(\mathrm{ln}d)^\tau }𝒥^\pm (d)$$
where
$`𝒥^\pm (d):=`$ $`[\chi _\epsilon ^\pm \mathrm{\Delta }d{\displaystyle \frac{h^{}(d)}{h^{\prime \prime }(d)}}+{\displaystyle \frac{h^{}(d)}{dh^{\prime \prime }(d)}}(d(\mathrm{ln}d)^\tau \mathrm{\Delta }d{\displaystyle \frac{2\tau \chi _\epsilon ^\pm }{\mathrm{ln}d}})+a{\displaystyle \frac{h(d)}{h^{\prime \prime }(d)}}(\chi _\epsilon ^\pm +(\mathrm{ln}d)^\tau )`$
$`+{\displaystyle \frac{\tau \chi _\epsilon ^\pm h(d)}{d^2h^{\prime \prime }(d)\mathrm{ln}d}}\left(1+{\displaystyle \frac{\tau +1}{\mathrm{ln}d}}d\mathrm{\Delta }d\right)+(\stackrel{~}{c}\pm \epsilon )d^\theta (\mathrm{ln}d)^\tau {\displaystyle \frac{k^2(d)f(\xi _0h(d))}{\xi _0h^{\prime \prime }(d)}}`$
$`+(\stackrel{~}{c}\pm \epsilon )\chi _\epsilon ^\pm d^\theta {\displaystyle \frac{k^2(d)h(d)f^{}(\mathrm{\Psi }^\pm (d))}{h^{\prime \prime }(d)}}+(d)+𝒥_1^\pm (d)].`$
Here $``$ is defined by (2.14), while
$$𝒥_1^\pm (d):=\chi _\epsilon ^\pm \left(1\frac{k^2(d)h(d)f^{}(\mathrm{\Psi }^\pm (d))}{h^{\prime \prime }(d)}\right).$$
By Lemma 2.1 and (5.2), we infer that
$$\underset{d0}{lim}\frac{k^2(d)h(d)f^{}(\mathrm{\Psi }^\pm (d))}{h^{\prime \prime }(d)}=\underset{d0}{lim}\frac{\mathrm{\Psi }^\pm (d)f^{}(\mathrm{\Psi }^\pm (d))}{f(\mathrm{\Psi }^\pm (d))}\frac{k^2(d)f(\xi _0h(d))}{\xi _0h^{\prime \prime }(d)}=\rho +1.$$
Hence, $`lim_{d0}𝒥_1^\pm (d)=\rho \chi _\epsilon ^\pm :=\rho (\stackrel{~}{\chi }\pm \epsilon )`$. Using Lemmas 2.1 and 2.3, we find
$$\underset{d0}{lim}𝒥^+(d)=\rho \epsilon <0\text{and}\underset{d0}{lim}𝒥^{}(d)=\rho \epsilon >0.$$
Therefore, by (5.4) we conclude (5.3).
*Step* 2. There exists $`M^+`$, $`\delta ^+>0`$ such that
$$u_a(x)u^+(x)+M^+,x\mathrm{\Omega }\text{with}0<d<\delta ^+.$$
We only recover (4.5), the rest being similar to the proof of Step 2 in Theorem 1.1. Indeed, by (5.3), (5.1) and (4.4), we obtain
$`\mathrm{\Delta }u_\sigma ^{}(x)`$ $`au^+(d\sigma ,s)[1+(\stackrel{~}{c}\epsilon )(d\sigma )^\theta ]k^2(d\sigma )f(u^+(d\sigma ,s))`$
$`au^+(d\sigma ,s)[1+(\stackrel{~}{c}\epsilon )d^\theta ]k^2(d)f(u^+(d\sigma ,s))`$
$`au^+(d\sigma ,s)b(x)f(u^+(d\sigma ,s))`$
$`a(u^+(d\sigma ,s)+M^+)b(x)f(u^+(d\sigma ,s)+M^+)`$
$`=au_\sigma ^{}(x)b(x)f(u_\sigma ^{}(x))\text{in }𝒩_\sigma .`$
*Step* 3. There exists $`M^{}`$, $`\delta ^{}>0`$ such that
$$u_a(x)u^{}(x)M^{},x\mathrm{\Omega }\text{with}0<d<\delta ^{}.$$
We proceed in the same way as for proving (4.6). To recover (4.8) (with $`\lambda `$ given by (4.7)), we show that $`\mathrm{\Delta }v_\sigma ^{}+av_\sigma ^{}b(x)f(v_\sigma ^{})`$ in $`\mathrm{\Omega }_{\delta _2/4}`$. Indeed, using (5.1), (5.3) and $`(A_1)`$, we find
$`\mathrm{\Delta }v_\sigma ^{}(x)+av_\sigma ^{}(x)`$ $`=\lambda (\mathrm{\Delta }u^{}(d+\sigma ,s)+au^{}(d+\sigma ,s))`$
$`\lambda k^2(d+\sigma )[1+(\stackrel{~}{c}+\epsilon )(d+\sigma )^\theta ]f(u^{}(d+\sigma ,s))`$
$`k^2(d)[1+(\stackrel{~}{c}+\epsilon )d^\theta ]f(\lambda u^{}(d+\sigma ,s))`$
$`b(x)f(v_\sigma ^{}(x)),x=(d,s)\mathrm{\Omega }_{\delta _2/4}.`$
Since $`lim_{d0}k^2(d)f(\lambda ^2u^{}(x))/u^{}(x)=\mathrm{}`$, there exists $`\stackrel{~}{\delta }(0,\delta _2/4)`$ such that
(5.5)
$$k^2(d)[1+(\stackrel{~}{c}+\epsilon )d^\theta ]f(\lambda ^2u^{})/u^{}\lambda ^2|a|,x\mathrm{\Omega }\text{with}0<d\stackrel{~}{\delta }.$$
By Lemma 2.1, we infer that $`u^{}(x)`$ decreases with $`d`$ when $`d(0,\stackrel{~}{\delta })`$ (if necessary, $`\stackrel{~}{\delta }>0`$ is diminished). Choose $`\delta _{}(0,\stackrel{~}{\delta })`$ close enough to $`\stackrel{~}{\delta }`$ such that
(5.6)
$$\frac{h(\delta _{})(1+\chi _\epsilon ^{}(\mathrm{ln}\delta _{})^\tau )}{h(\stackrel{~}{\delta })(1+\chi _\epsilon ^{}(\mathrm{ln}\stackrel{~}{\delta })^\tau )}<1+\lambda .$$
Hence, we regain (4.12), (4.13) and (4.15).
By (5.1) and (5.3), we see that (4.11) follows if
(5.7)
$$k^2(d+\sigma )[1+(\stackrel{~}{c}+\epsilon )(d+\sigma )^\theta ]\left[f(u^{}(d+\sigma ,s))f(z_\sigma (d,s))\right]a(1\lambda )u^{}(\delta _{},s)$$
for each $`(d,s)\mathrm{\Omega }_\delta _{}`$. Using (4.15), together with (5.5) and (4.13), we arrive at (5.7). From now on, the argument is the same as before. This proves the claim of Step 3.
By Steps 2 and 3, it follows that
(5.8)
$$\{\begin{array}{cc}& \chi _\epsilon ^+\left[1+\frac{u_a(x)}{\xi _0h(d)}\right](\mathrm{ln}d)^\tau \frac{M^+(\mathrm{ln}d)^\tau }{\xi _0h(d)},x\mathrm{\Omega }\text{with}d<\delta ^+\hfill \\ & \chi _\epsilon ^{}\left[1+\frac{u_a(x)}{\xi _0h(d)}\right](\mathrm{ln}d)^\tau +\frac{M^{}(\mathrm{ln}d)^\tau }{\xi _0h(d)},x\mathrm{\Omega }\text{with}d<\delta ^{}.\hfill \end{array}$$
Using Lemma 2.1, we have
$$\underset{t0}{lim}\frac{(\mathrm{ln}t)^\tau }{h(t)}=\underset{t0}{lim}\left(\frac{\mathrm{ln}t}{\mathrm{ln}h(t)}\right)^\tau \frac{(\mathrm{ln}h(t))^\tau }{h(t)}=\left(\frac{\rho \mathrm{}_1}{2}\right)^\tau \underset{u\mathrm{}}{lim}\frac{(\mathrm{ln}u)^\tau }{u}=0.$$
Passing to the limit $`d0`$ in (5.8), we obtain
$$\chi _\epsilon ^{}\underset{d0}{lim\; inf}\left[1+\frac{u_a(x)}{\xi _0h(d)}\right](\mathrm{ln}d)^\tau \underset{d0}{lim\; sup}\left[1+\frac{u_a(x)}{\xi _0h(d)}\right](\mathrm{ln}d)^\tau \chi _\epsilon ^+.$$
By sending $`\epsilon `$ to 0, the proof of Theorem 1.3 is finished. ∎
|
warning/0506/astro-ph0506163.html
|
ar5iv
|
text
|
# Summary
## Summary
Various techniques are being used to search for extra-solar planetary signatures, including accurate measurement of radial velocity and positional (astrometric) displacements, gravitational microlensing, and photometric transits. Planned space experiments promise a considerable increase in the detections and statistical knowledge arising especially from transit and astrometric measurements over the years 2005–15, with some hundreds of terrestrial-type planets expected from transit measurements, and many thousands of Jupiter-mass planets expected from astrometric measurements.
Beyond 2015, very ambitious space (Darwin/TPF) and ground (OWL) experiments are targeting direct detection of nearby Earth-mass planets in the habitable zone and the measurement of their spectral characteristics. Beyond these, ‘Life Finder’ (aiming to produce confirmatory evidence of the presence of life) and ‘Earth Imager’ (some massive interferometric array providing resolved images of a distant Earth) appear as distant visions.
This report, to ESA and ESO, summarises the direction of exo-planet research that can be expected over the next 10 years or so, identifies the roles of the major facilities of the two organisations in the field, and concludes with some recommendations which may assist development of the field.
The report has been compiled by the Working Group members and experts (page iii) over the period June–December 2004.
Introduction & Background
Following an agreement to cooperate on science planning issues, the executives of the European Southern Observatory (ESO) and the European Space Agency (ESA) Science Programme and representatives of their science advisory structures have met to share information and to identify potential synergies within their future projects. The agreement arose from their joint founding membership of EIROforum (http://www.eiroforum.org) and a recognition that, as pan-European organisations, they served essentially the same scientific community.
At a meeting at ESO in Garching during September 2003, it was agreed to establish a number of working groups that would be tasked to explore these synergies in important areas of mutual interest and to make recommendations to both organisations. The chair and co-chair of each group were to be chosen by the executives but thereafter, the groups would be free to select their membership and to act independently of the sponsoring organisations. The first working group to be established was on the topic of Extra-Solar Planet research, both detection and physical study, over a period extending from now until around 2015. The group worked on its report from June until December 2004 and reported its conclusions and recommendations to a second ESA-ESO meeting, held at ESA HQ in Paris in February 2005.
Terms of Reference and Composition
The goals set for the working group were to provide:
* A survey of the field: this will comprise: (a) a review of the methods used or envisaged for extra-solar planet detection and study; (b) a survey of the associated instrumentation world-wide (operational, planned, or proposed, on-ground and in space); (c) for each, a summary of the potential targets, accuracy and sensitivity limits, and scientific capabilities and limitations.
* An examination of the role of ESO and ESA facilities: this will: (a) identify areas in which current and planned ESA and ESO facilities will contribute; (b) analyse the expected scientific returns and risks of each; (c) identify areas of potential scientific overlap, and thus assess the extent to which the facilities complement or compete; (d) identify open areas which merit attention by one or both organisations (for example, follow-up observations by ESO to maximise the return from other major facilities); (e) conclude on the scientific case for the very large facilities planned or proposed.
The working group membership was established by the chair and co-chair: the report is not a result of consultation with the community as a whole. The experts contributed considerable information for the report, but the conclusions and recommendations are the responsibility of the members.
Catherine Cesarsky (ESO) Álvaro Giménez Cañete (ESA)
March 2005
## 1 Survey of the Field
### 1.1 Introduction
The field of exo-planet research has exploded dramatically since the discovery of the first such systems in 1995. Underlying this huge interest three main themes of exo-planet research can be identified: (a) characterising and understanding the planetary populations in our Galaxy; (b) understanding the formation and evolution of planetary systems (e.g., accretion, migration, interaction, mass-radius relation, albedo, distribution, host star properties, etc.); (c) the search for and study of biological markers in exo-planets, with resolved imaging and the search for intelligent life as ‘ultimate’ and much more distant goals.
Detection methods for extra-solar planets can be broadly classified into those based on: (i) dynamical effects (radial velocity, astrometry, or timing in the case of the pulsar planets); (ii) microlensing (astrometric or photometric); (iii) photometric signals (transits and reflected light); (iv) direct imaging from ground or space in the optical or infrared; and (v) miscellaneous effects (such as magnetic superflares, or radio emission). Each have their strengths, and advances in each field will bring specific and often complementary discovery and diagnostic capabilities. Detections are a pre-requisite for the subsequent steps of detailed physical-chemical characterisation demanded by the emerging discipline of exo-planetology.
As of December 2004, 135 extra-solar planets have been discovered from their radial velocity signature, comprising 119 systems of which 12 are double and 2 are triple. One of these planets has also been observed to transit the parent star. Four additional confirmed planets have been discovered through transit detections using data from OGLE (and confirmed through radial velocity measurements), and one, TrES-1, using a small 10-cm ground-based telescope. One further, seemingly reliable, planet candidate has been detected through its microlensing signature. The planets detected to date (apart from those surrounding radio pulsars, which are not considered further in this report) are primarily ‘massive’ planets, of order 1 $`M_\mathrm{J}`$, but extending down to perhaps 0.05 $`M_\mathrm{J}`$ (around 15 $`M_{}`$) for three short-period systems, although the inclination (and hence true mass) of two of these is unknown<sup>1</sup><sup>1</sup>1 the following notation is used: $`M_\mathrm{J}`$ = Jupiter mass; $`M_{}`$ = Earth mass $`0.003M_\mathrm{J}`$..
Detection methods considered to date are summarised in Figure 1, which also gives an indication of the lower mass limits which are likely to be reached in the foreseeable future for each method. More information and ongoing projects are given in Jean Schneider’s www page: http://www.obspm.fr/encycl/searches.html.
An earlier ESO Working Group on the ‘Detection of Extra-Solar Planets’ submitted a report with detailed recommendations in 1997 (Paresce et al., 1997). A summary and status of these recommendations is attached as Appendix C.
### 1.2 The Search for Earth-Mass Planets and Habitability
The search for planets around stars in general, and Earth-mass planets in particular, is motivated by efforts to understand their frequency of occurrence (as a function of mass, semi-major axis, eccentricity, etc.) and their formation mechanism and, by analogy, to gain an improved understanding of the formation of our own Solar System. Search accuracies will progressively improve to the point that the detection of telluric planets in the ‘habitable zone’ will become feasible, and there is presently no reason to assume that such planets will not exist in large numbers. Improvements in spectroscopic abundance determinations, whether from Earth or space, and developments of atmospheric modelling, will lead to searches for planets which are progressively habitable, inhabited by micro-organisms, and ultimately by intelligent life (these may or may not prove fruitful). Search strategies will be assisted by improved understanding of the conditions required for development of life on Earth.
Very broadly, the search for potentially habitable planets is being concentrated around Sun-like stars (spectral type and age), focussing on Earth-mass planets, in low-eccentricity orbits at about 1 AU representing the ‘continuously habitable zone’ (the habitable zone is the distance range from the parent star over which liquid water is likely to be present; the continuously habitable zone is the region throughout which liquid water should have been present over a significant fraction of the star’s main-sequence lifetime). Further details and potential spectral diagnostics of life are given in this section. Such considerations may imply that the fraction of habitable planets is small, but they should add to the knowledge of where to look.
Assessment of the suitability of a planet for supporting life, or habitability, is based on our knowledge of life on Earth. With the general consensus among biologists that carbon-based life requires water for its self-sustaining chemical reactions, the search for habitable planets has therefore focused on identifying environments in which liquid water is stable over billions of years. Earth’s habitability over early geological time scales is complex, but its atmosphere is thought to have experienced an evolution in the greenhouse blanket of CO<sub>2</sub> and H<sub>2</sub>O to accommodate the 30% increase in the Sun’s luminosity over the last 4.6 billion years in order to sustain the presence of liquid water evident from geological records. In the future, the Sun will increase to roughly three times its present luminosity by the time it leaves the main sequence, in about 5 Gyr.
The habitable zone is consequently presently defined by the range of distances from a star where liquid water can exist on the planet’s surface. This is primarily controlled by the star-planet separation, but is affected by factors such as planet rotation combined with atmospheric convection. For Earth-like planets orbiting main-sequence stars, the inner edge is bounded by water loss and the runaway greenhouse effect, as exemplified by the CO<sub>2</sub>-rich atmosphere and resulting temperature of Venus. The outer boundary is determined by CO<sub>2</sub> condensation and runaway glaciation, but it may be extended outwards by factors such as internal heat sources including long-lived radionuclides (U<sup>235</sup>, U<sup>238</sup>, K<sup>40</sup> etc., as on Earth), tidal heating due to gravitational interactions (as in the case of Jupiter’s moon Io), and pressure-induced far-infrared opacity of H<sub>2</sub>, since even for effective temperatures as low as 30 K, atmospheric basal temperatures can exceed the melting point of water. These considerations result, for a $`1M_{}`$ star, in an inner habitability boundary at about 0.7 AU and an outer boundary at around 1.5 AU or beyond. The habitable zone evolves outwards with time because of the increasing luminosity of the Sun with age, resulting in a narrower width of the continuously habitable zone over $`4`$ Gyr of around 0.95–1.15 AU. Positive feedback due to the greenhouse effect and planetary albedo variations, and negative feedback due to the link between atmospheric CO<sub>2</sub> level and surface temperature may limit these boundaries further. Migration of the habitable zone to much larger distances, 5–50 AU, during the short period of post-main-sequence evolution corresponding to the sub-giant and red giant phases, has been considered.
Within the $`1`$ AU habitability zone, Earth ‘class’ planets can be considered as those with masses between about 0.5–10 $`M_{}`$ or, equivalently assuming Earth density, radii between 0.8–2.2 $`R_{}`$. Planets below this mass in the habitable zone are likely to lose their life-supporting atmospheres because of their low gravity and lack of plate tectonics, while more massive systems are unlikely to be habitable because they can attract a H-He atmosphere and become gas giants. Habitability is also likely to be governed by the range of stellar types for which life has enough time to evolve, i.e. stars not more massive than spectral type A. However, even F stars have narrower continuously habitable zones because they evolve more strongly (and rapidly), while planets orbiting in the habitable zones of late K and M stars become trapped in synchronous rotation due to tidal damping, which may preclude life apart from close to the light-shadow line. Mid- to early-K and G stars may therefore be optimal for the development of life.
Owen (1980) argued that large-scale biological activity on a telluric planet necessarily produces a large quantity of O<sub>2</sub>. Photosynthesis builds organic molecules from CO<sub>2</sub> and H<sub>2</sub>O, with the help of H<sup>+</sup> ions which can be provided from different sources. In the case of oxygenic bacteria on Earth, H<sup>+</sup> ions are provided by the photodissociation of H<sub>2</sub>O, in which case oxygen is produced as a by-product. However, this is not the case for anoxygenic bacteria, and thus O<sub>2</sub> is to be considered as a possible but not a necessary by-product of life (for this signature of biological activity, as well as for any other, a key issue is that of false positives, i.e. cases where the signature is detected but there is no actual life on the planet, while the case of false negatives, when there is some life on the planet but the signature is absent, is significantly less ‘serious’). Indeed, Earth’s atmosphere was O<sub>2</sub>-free until about 2 billion years ago, suppressed for more than 1.5 billion years after life originated. Owen (1980) noted the possibility, quantified by Schneider (1994) based on transit measurements, of using the 760-nm band of oxygen as a spectroscopic tracer of life on another planet since, being highly reactive with reducing rocks and volcanic gases, it would disappear in a short time in the absence of a continuous production mechanism. Plate tectonics and volcanic activity provide a sink for free O<sub>2</sub>, and are the result of internal planet heating by radioactive uranium and of silicate fluidity, both of which are expected to be generic whenever the mass of the planet is sufficient and when liquid water is present. For small enough planet masses, volcanic activity disappears some time after planet formation, as do the associated oxygen sinks.
O<sub>3</sub> is itself a tracer of O<sub>2</sub> and, with a prominent spectral signature at 9.6 $`\mu `$m in the infrared where the planet/star contrast is significantly stronger than in the optical ($`1.4\times 10^7`$ rather than $`2\times 10^{10}`$ for the Earth/Sun case), should be easier to detect than the visible wavelength lines. These considerations are motivating the development of infrared space interferometers for the study of bands such as H<sub>2</sub>O at 6–8 $`\mu `$m, CH<sub>4</sub> at 7.7 $`\mu `$m, O<sub>3</sub> at 9.6 $`\mu `$m, CO<sub>2</sub> at 15 $`\mu `$m and H<sub>2</sub>O at 18 $`\mu `$m. Higher resolution studies might reveal the presence of CH<sub>4</sub>, its presence on Earth resulting from a balance between anaerobic decomposition of organic matter and its interaction with atmospheric oxygen; its highly disequilibrium co-existence with O<sub>2</sub> could be strong evidence for the existence of life.
The possibility that O<sub>2</sub> and O<sub>3</sub> are not unambiguous identifications of Earth-like biology, but rather a result of abiotic processes, has been considered in detail by Léger et al. (1999) and Selsis et al. (2002). They considered various production processes such as abiotic photodissociation of CO<sub>2</sub> and H<sub>2</sub>O followed by the preferential escape of hydrogen from the atmosphere. In addition, cometary bombardment could bring O<sub>2</sub> and O<sub>3</sub> sputtered from H<sub>2</sub>O by energetic particles, depending on the temperature, greenhouse blanketing, and presence of volcanic activity. They concluded that a simultaneous detection of significant amounts of H<sub>2</sub>O and O<sub>3</sub> in the atmosphere of a planet in the habitable zone presently stands as a criterion for large-scale photosynthetic activity on the planet. Such an activity on a planet illuminated by a star similar to the Sun, or cooler, is likely to be a significant indication that there is local biological activity, because this synthesis requires the storage of the energy of at least 2 photons (8 in the case on Earth) prior to the synthesis of organic molecules from H<sub>2</sub>O and CO<sub>2</sub>. This is likely to require delicate systems that have developed during a biological evolutionary process. The biosignature based on O<sub>3</sub> seems to be robust because no counter example has been demonstrated. It is not the case for the biosignature based on O<sub>2</sub> (Selsis et al., 2002), where false positives can be encountered. This puts a hierarchy between observations that can detect O<sub>2</sub> and those that can detect O<sub>3</sub>.
Habitability may be further confined within a narrow range of \[Fe/H\] of the parent star (Gonzalez, 1999b). If the occurrence of gas giants decreases at lower metallicities, their shielding of inner planets in the habitable zone from frequent cometary impacts, as occurs in our Solar System, would also be diminished. At higher metallicity, asteroid and cometary debris left over from planetary formation may be more plentiful, enhancing impact probabilities. Gonzalez (1999a) has also investigated whether the anomalously small motion of the Sun with respect to the local standard of rest, both in terms of its pseudo-elliptical component within the Galactic plane, and its vertical excursion with respect to the mid-plane, may be explicable in anthropic terms. Such an orbit could provide effective shielding from high-energy ionising photons and cosmic rays from nearby supernovae, from the X-ray background by neutral hydrogen in the Galactic plane, and from temporary increases in the perturbed Oort comet impact rate.
### 1.3 Present Limits: Ground and Space
Figure 2 illustrates the detection domains for the radial velocity, astrometry, and transit methods as a function of achievable accuracy. It also shows the location of the exo-planets known to date, in a mass-orbital radius (period) diagram.
The fundamental accuracy limits of each method are not yet firmly established, although such knowledge is necessary to predict the real performances of dedicated surveys on ground and in space. Granular flows and star spots on the surface of late-type stars place specific limits on the photometric stability, the stability of the photocentric position, and the stability of spectroscopically-derived radial velocities, whether these observations are made from ground or space. A series of hydrodynamical convection models covering stellar objects from white dwarfs to red giants has been used to give estimates of the photometric and photocentric stellar variability in wavelength-integrated light across the HR diagram (Svensson & Ludwig, 2005).
(a) Radial velocity experiment accuracies are close to the values of around 1–3 m s<sup>-1</sup> at which atmospheric circulation and oscillations limit measurement precision, implying mass detection limits only down to 0.01–0.1 $`M_\mathrm{J}`$ (depending on orbital period); detection of an Earth in the habitable zone would require accuracies of $`0.030.1`$ m s<sup>-1</sup>. Observations from space will not improve these limits, and no high-precision radial velocity measurements from space have been proposed.
The idea of ‘stacking up’ many radial velocity observations to average the effects of stellar oscillations is appealing, but faces several complications: (i) even if p-mode oscillation effects can be minimised, beating amongst these modes may induce large radial-velocity variations (up to 10 m s<sup>-1</sup> peak-to-peak) over timescales of a few hours, specifically some 5–6 hours for $`\mu `$ Arae (Bouchy, private communication). The star will therefore need to be observed over several hours for each epoch (radial velocity point); (ii) simulations by Bouchy (private communication) show that a precision of $`1`$ m s<sup>-1</sup> is reached in about 15–20 min, while the gain is much less rapid with increasing observation time. A precision of 0.1 m s<sup>-1</sup> (still insufficient for the detection of the Earth around the Sun) will therefore be very expensive in terms of telescope time; (iii) a wavelength calibration precision from night-to-night is then needed at the level of the long-term precision targeted. Reference calibration to 0.1 m s<sup>-1</sup> will require further improvements in calibration techniques. With HARPS, a precision of about 0.5 m s<sup>-1</sup> is reached, as illustrated by asteroseismology results on $`\mu `$ Arae with 250 observations each. Investigations are ongoing (Udry, private communication) into the possibility of having a reference at the 0.02 m s<sup>-1</sup> level for an instrument on OWL, while the HARPS GTO programme anyway pushing in this direction will soon help to better characterise the question. In conclusion, a very high radial velocity precision seems possible, but at a very high cost.
There is a significant difference in the case of transiting candidates: now the period and phase are known, and with $`e0`$ for short period planets, a series of accumulated measurements can be used to constrain the radial velocity semi-amplitude. With HARPS at a precision of 1 m s<sup>-1</sup>, for short-period planets, it is expected that limits of a few Earth-masses, for $`P<10`$ days, can be reached. If the transiting object is larger, then the radial velocity effect will be larger and easier to detect. False positive detections will be the main problem.
(b) Photometric (transit) limits below the Earth’s atmosphere are typically a little below the 1% photometric precision, limited by variations in extinction, scintillation and background noise (depending on telescope aperture size), corresponding to masses of about 1 $`M_\mathrm{J}`$ for solar-type stars. One main challenge is to reach differential photometric accuracies of around 1 mmag over a wide field of view, in which airmass, transparency, differential refraction and seeing all vary significantly. The situation improves above the atmosphere, and a number of space experiments are planned to reach the 0.01% limits required for the detection of Earth-mass planets. HST can place much better limits on transit photometry than is possible from the ground, as exemplified by HD 209458 (see Sections 2.1.2 and 2.2.1). Simulations have been made by the COROT teams in order to estimate the transit detection threshold due to stellar activity. In the case of a very active star, the detection of an Earth (80 ppm) is not possible. In the case of a quiet star (like the Sun), it is possible if several transits are summed. In the case of COROT, 1.6 $`M_{}`$ is detected after 10–30 transits. Another complication is again false-positives, where statistical effects, stellar activity, and background binaries can all mimic transit events, and which call for independent confirmation of detections in general.
(c) Astrometric measurements do not yet extend below the 1 milli-arcsec of Hipparcos, implying current detectability limits typically above 1–10 $`M_\mathrm{J}`$. Even with the expected advent of narrow-field ground-based astrometry at 10 micro-arcsec (e.g. PRIMA), detections would be well short of Earth-mass planets, even within 10 pc. Above the atmosphere, astrometric accuracy limits improve significantly. The studies of Svensson & Ludwig (2005) indicate that for $`\mathrm{log}g4.4`$, resulting displacements are around $`10^710^8`$ AU suggesting, for example, that this effect will not degrade the Gaia measurements, with the exception of nearby ($`<100`$ pc) red giants. Nevertheless, that work treats only the variability caused by the evolution of stellar surface inhomogeneities driven by thermal convection (stellar granulation). At lower temporal frequencies, the variability is much higher (but not yet treatable by hydrodynamic models), caused by magnetic stellar activity, spottiness, and rotation, all of which may make substantial additional contributions to the astrometric (and photometric) variability.
(d) Microlensing searches are not limited by current measurement accuracies for Earth-mass planets, which can produce relatively large amplitude photometric signals (a few tenths of a magnitude or larger), though small amplitude signals are more frequent. The limitations of this method are rather of statistical nature: even if all stars acting as microlenses have planets, only a small subset of them would show up in the microlensed lightcurve, depending on the projected separation and the exact geometry between relative path and planetary caustic. Space measurements help significantly by reducing the photometric confusion effects resulting from observations in very crowded regions (such as the Galactic bulge) which are favoured fields to improve the statistics of detectable events.
## 2 The Period 2005–2015
### 2.1 Ground Observations: 2005–2015
There are many ongoing ground-based surveys. At the time of writing, the Planets Encyclopaedia www page (http://www.obspm.fr/encycl/searches.html) lists as either ongoing or planned: 18 radial velocity searches, 15 transit searches, 5 microlensing programmes, 10 imaging/direct detection programmes, 2 radio surveys, and 3 astrometric efforts. An overview of efforts and expected results is given in this section, with a particular focus on the ESA and ESO contributions.
#### 2.1.1 Radial Velocity Searches
A summary of ongoing or planned radial velocity experiments is given in Table 1. It is certainly incomplete, and is intended only to give a flavour for the activity in the field. The vast majority of extra-solar planets discovered so far have been found by radial velocity searches, which have a natural bias for the discovery of massive planets orbiting close to their central star (hot Jupiters). As the surveys continue for longer periods of time they become more and more sensitive to planets having longer periods, and to additional planets in systems in which one hot Jupiter is already known. Dedicated designs have brought spectrographs very close to the practical accuracy limit for ground-based radial velocity searches of $`1`$ m s<sup>-1</sup>, allowing detection of lower mass planets. HARPS has recently detected a second planet around $`\mu `$ Arae, with a period of 9.5 days, a velocity semi-amplitude of less than 5 m s<sup>-1</sup> (Santos et al., 2004), and a derived $`M\mathrm{sin}i`$ of only 14 $`M_{}`$ (about one Uranus mass), making it the lowest mass planet found so far (as of December 2004). A planet of $`17M_{}`$ has been discovered orbiting the M dwarf GJ 436 based on Keck data (Butler et al., 2004), and of $`18M_{}`$ reported for 55 Cnc using HET observations (McArthur et al., 2004).
Another trend is that larger telescopes are being used for the radial velocity searches. As a result the limiting magnitude of such searches has increased from typically $`V=7.5`$ a few years ago to $`V12`$, with the number of stars thus available for radial velocity study increasing by almost two orders of magnitude. For example, the N2K Consortium is using the Keck, Magellan and Subaru telescopes to track the next 2000 (N2K) closest ($`<`$110 pc), brightest, and most metal-rich FGK stars not on current Doppler surveys for new hot Jupiters. Started in early 2004, and with a precision of 4–7 m s<sup>-1</sup>, the first Saturn-mass planet from this survey has recently been reported (Fischer et al., 2005).
Table 5 summarises predictions of the numbers of planets which might be detected by each of the methods discussed in this section, including radial velocity measurements, out to 2008–10. Although such a table is open to misinterpretation and debate (being sensitive to planetary and instrumental hypotheses) it provides some indication of the development of exo-planet statistics over the next few years.
To derive these estimates for radial velocity observations, the following arguments have been used: (a) during the past year (2003) the number of target stars monitored has at least doubled (HARPS, new Elodie programme, HET, others). New programmes will also probably start in the coming years (e.g. Sophie at OHP). The number of presently known planets with masses in the range $`0.510M_\mathrm{J}`$ (‘easy planets’) can then probably be multiplied by $`34`$; (b) the typical precision is improving, and the number of planets with lower masses will thus further increase. Uncertainty remains on the existence or frequency of low-mass planets in short-period orbits (presently unknown), and in the detailed effects of stellar jitter; (c) the number of long-period planets in the present surveys already ongoing for several years will increase, as the distribution of planet numbers increases with period. However, the maximum mass of detected planets also increases with period, such that many high-mass planets will be found, probably with masses $`>10M_\mathrm{J}`$. Earth-masses will be out of reach due to stellar jitter, except maybe for short-$`P`$, low-mass transiting planets with $`P`$ and $`e`$ fixed, for which a large number of measurements can be stacked at the appropriate phase (see Section 1.3).
More details are included of the ESO contribution, HARPS, since it typifies the state-of-the-art technical and scientific objectives:
HARPS: The Observatoire de Genève together with the Physikalisches Institut der Universität Bern, the Observatoire de Haute-Provence, and the Service d’Aéronomie du CNRS and in collaboration with ESO, developed the HARPS spectrograph installed on ESO’s 3.6-m Telescope at La Silla. The instrument is a high-resolution, high-efficiency fibre-fed echelle spectrograph designed to efficiently search for extra-solar planets reaching a precision of 1 m s<sup>-1</sup> on radial-velocity measurements. Typically, this precision is reached in 1 minute for a $`V=7.5`$ G-dwarf star. The long-term precision is ensured by the spectrograph’s stability: the spectrograph resides in a pressure- and temperature-controlled vacuum tank, with a drift usually well below 1 m s<sup>-1</sup> during one night, which can be further corrected using the simultaneous thorium technique. HARPS has been available to the community since October 2003.
For the development of this instrument, the HARPS consortium has been granted 500 guaranteed nights over 5 years (100 nights per year). The HARPS survey is designed to address several specific questions:
(a) only a few of the hundred detected planets have masses less than the mass of Saturn, and due to the present precision of radial velocity surveys the distribution of planetary masses is heavily biased (or completely unknown) for masses less than half the mass of Jupiter. The high precision of HARPS will allow searches for low-mass planets: for a sample of preselected non-active solar-type stars (from the Coralie planet-search sample), the aim is to explore the domain of the mass-function for short-period planets below the mass of Saturn down to a few Earth masses;
(b) in a continuation of the planet-search programmes conducted over $``$10 years, a quick screening of a large volume-limited sample of $``$1000 still unobserved stars will be performed in order to identify new ‘hot Jupiters’ and other Jovian-type planets. Increasing the list of ‘hot Jupiters’ will improve the prospects of finding further stars with a planetary transit among relatively bright stars. Better statistics are needed to identify new properties of the distribution of exo-planet parameters. This part of the programme has already revealed two new short-period planets (Pepe et al., 2004);
(c) a systematic search for planets will be made for a volume-limited sample of slowly-rotating non-binary M-dwarfs closer than 11 pc. Such a survey of very low mass stars should constrain the frequency of planets as a function of stellar mass. Up to now only one planetary system orbiting an M-dwarf is known. For the less massive stars short-period planets of only a few times the mass of the Earth could be detected. Since most of these objects are faint, high efficiency is required. These objects are of prime importance for future astrometric studies to be carried out with the VLTI or SIM;
(d) stars with detected giant planets exhibit an impressive excess of metallicity in contrast to stellar samples without giant planets. The excess of metallicity does not seem related to the mass of the convective zone and probably originates in the chemical composition of the primordial molecular cloud. To add new constraints to the link between star chemical composition and frequency (or properties) of exo-planets, two programmes are being carried out. The first is a search for exo-planets orbiting solar-type stars with notable metal deficiency (for most of them \[Fe/H\] between $`0.5`$ and $`1.0`$). Among the existing detections of exo-planets only two or three have been found with metallicity in that range. The aim is to estimate the frequency of exo-planets in that domain of metallicity and, if possible, to compare their characteristics (masses, orbits) to planets orbiting metal-rich stars;
(e) the second ‘abundance-related’ programme aims at exploring the link between stellar metallicity and properties of exo-planets. Visual binaries with solar-type stars of almost identical magnitudes have been selected. For those including giant planets a detailed chemical analysis will be done for both stellar components to search for possible differences in their chemical compositions;
(f) follow-up radial velocity measurements for stars with planetary transits detected by COROT will be made with HARPS (where the photometric transit provides an estimate of the radius of the transiting planet as well as the orbital period and phase). Complementary ground-based spectroscopic measurements with HARPS will constrain the planetary mass and thus the planet mean density. The main scientific return for the planetary programme of the COROT mission will come from this combination of photometric and radial velocity data.
#### 2.1.2 Transit Searches
The transit method aims at detecting the dimming of the stellar light by occultation due to an orbiting planet. Transit experiments offer a number of very important contributions: (i) searches can be conducted over wide fields over long periods (5 years or more), and are therefore potentially efficient at detecting previously unknown systems; (ii) from the ground they are able to detect massive transiting planets, especially the ‘hot Jupiters’, while from space planets down to Earth-mass or below can be detected; (iii) spectroscopy during the planet transit can yield physical diagnostics of the transiting planets. A summary of ongoing or planned transit experiments is given in Table 2; see also the recent review by Horne (2003). Again, Table 5 summarises predictions of the numbers of planets that might be detected by this method out to 2008–12.
Transit measurements can only detect planets with a favourable orientation of their orbital plane, implying that only a small fraction of planets can ever be detected or monitored using this technique. In particular a nearby census can only reveal a small fraction of existing systems. The probability of viewing a planetary system edge-on depends on the distance of the planet to the central star. For close-in orbits it is about 10%, decreasing for more distant planets. Transit searches therefore try to maximise the number of stars they can observe simultaneously. This can be reached by either small telescopes with wide field of view observing relatively bright stars, or large telescopes providing deep exposures to increase the number of targets monitored. The current transit search surveys therefore group into two classes, using small or larger ($`>70`$ cm) telescopes. Support for the implicit hypothesis that the orbits of planetary systems are randomly distributed in the Galaxy comes from the fact that stellar rotation axes are themselves randomly distributed (obtained by $`v\mathrm{sin}i`$ distributions, and independently confirmed by magnetic-field orientation studies).
The wide-angle survey teams follow STARE and Vulcan in using small (10 cm) wide-angle (10) CCD cameras with a pixel size of order 1 arcsec or larger, sacrificing angular resolution to expand the field of view. The faint limit, at $`V1213`$ reaches to $`d300500`$ pc, comparable to the disk scale height, so that target fields cover the entire sky, which may contain some 1000 hot transiting Jupiters to this limit (Horne, 2003). The deep surveys use (mosaic) cameras on 1–4 m telescopes, reaching $`V1921`$ and $`d45`$ kpc, so that Galactic plane and open cluster fields are primary targets. Horne (2003) predicts up to 200 hot Jupiters per month being discovered by ongoing ground transit surveys in the future (perhaps by the year 2010).
The size of planets detectable from transits with ground-based searches is limited by the Earth’s atmosphere (Section 1.3). The photometric precision of typical lightcurves is a little under 1%, which corresponds to about Jupiter-sized planets for solar-type stars. Ground-based surveys are further limited in their time coverage of potential transiting planets by daytime and bad weather periods. The small number of transit surveys operating over more than a few months so far, and the lack of continuous observations on the target fields, are the most likely reason why only few planets have been discovered through transit detections so far, somewhat in contrast with the large number of surveys listed in Table 2. The situation could be significantly improved by ground-based networks of telescopes spanning a range of longitudes to ensure continuous observational coverage of target fields.
The discovery of a temporary dimming of the stellar lightcurve alone is not sufficient to secure the detection of a transiting planet. Grazing eclipsing binary stars, background binaries, brown dwarfs and stellar spots can cause lightcurves similar to transiting planets. Follow-up measurements, in particular radial velocity measurements, and determination of stellar parameters, therefore play an important role in the detection of planetary transits to exclude other causes of light dimming.
Up to now, five confirmed planets have been discovered by ground-based transit searches: four using the 1.3 m OGLE telescope, and one with a 10 cm ground-based system (TrES-1). The OGLE experiment (Optical Gravitational Lensing Experiment) uses the 1.3 m Warsaw Telescope at Las Campanas Observatory, Chile. It is equipped with a mosaic of 8 CCDs of 2k$`\times `$4k each, giving a field of view of 35 arcmin square with 0.26 arcsec/pixel. The telescope is primarily used to search for microlensing events by viewing near the Galactic centre, but significant time (more than three months) was made available for transit searches. It has monitored some 52 000 disk stars for 32 nights, reporting some 100 transit candidates with periods ranging from 1–9 days (e.g. Udalski et al. 2002a; 2002b) based solely on the dimming of stellar lightcurves. Most of them were quickly identified as stellar binary systems. Radial velocity follow-up for many of the candidate stars was difficult because of the faintness of the stars (down to $`I16`$ mag). Nevertheless, four events were confirmed as transiting planets by radial velocity follow-up measurements (Konacki et al., 2003; Bouchy et al., 2004; Pont et al., 2004).
The TrES telescopes belong to the class of small telescopes dedicated to transit searches. All three telescopes of this transit search programme are small aperture (10 cm) wide-field ($`6^{}`$) systems. They are located at Tenerife, Lowell Observatory, and Palomar Mountain, and thus span a range of longitudes (Alonso et al., 2004). Recently, the first planet found by this system has been announced. The discovery is again based on the lightcurves and radial velocity confirmation, showing that small-scale systems indeed have the potential to find transiting planets, providing their observational coverage is sufficiently high.
These examples show the potential of the transit method to find a large number of planets, including small planets, in an unbiased sample of stars. The full potential will be exploited in future space missions, from which Earth-mass planets can be detected (Section 2.2).
In addition to the geometric information derived rather directly from accurate photometric measurements of planetary transits, high-cadence, high S/N spectroscopy of transit events can reveal properties of the planetary atmosphere and exosphere. Extensive work on the first transiting planet, HD 209458b, has shown the level of current photometric and spectroscopic capabilities, principally using HST before the failure of STIS. Ground-based photometry was able to reach a precision of 0.2% (Henry et al., 2000; Charbonneau et al., 2002; Jha et al., 2000; Deeg et al., 2001) while HST/STIS has achieved $`0.01`$% (Brown et al., 2001). The high-cadence capability of the HST Fine Guidance Sensor (FGS) is also being exploited for transit timing (Schultz et al., 2004).
The use of time-resolved spectroscopy by Charbonneau et al. (2002) showed that the HD 209458b transit was $`2.3\times 10^4`$ deeper when observed at the sodium D lines. Again using STIS, Vidal-Madjar et al. (2003) detect a very large (15%) transit depth at Ly-$`\alpha `$, showing that the planet is losing mass. Subsequently the same group (Vidal-Madjar et al., 2004) reported a detection of C and O in the exosphere. In addition to these lines, the possibility exists, in the optical band, of looking for the effects of water (longward of 500 nm) and of Rayleigh scattering in the blue. Moutou et al. (2003) searched unsuccessfully for He I 1083 nm absorption using the VLT with ISAAC. Their upper limit of 0.5% at 3$`\sigma `$ for a 0.3 nm bandwidth was limited by the detector fringing properties at this wavelength. An alternative approach is to search for an infrared signature during secondary eclipse. This method, applied by Richardson et al. (2003) which they call ‘occultation spectroscopy’, searches for the disappearance and reappearance of weak spectral features due to the exo-planet as it passes behind the star. They argue that at the longest infrared wavelengths, this technique becomes preferable to conventional transit spectroscopy. They observed the system in the wing of the strong $`\nu _3`$ band of methane near 3.6 $`\mu `$m during two secondary eclipses, using the VLT/ISAAC spectrometer at a spectral resolution of 3300 but were unable to detect a signal.
A recent study by Holman & Murray (2005) has shown that for many planets discovered by transit surveys, accurate timing measurements between successive transits (of accuracies between 0.1–100 minutes) will allow for the detection of additional planets in the system (not necessarily transiting) via their gravitational interaction with the transiting planet. The transit time variations depend on the mass of the additional planet, and in some cases Earth-mass planets will produce a measurable effect. This effect is particularly prominent for long-period transiting systems, where the ‘perturber’ (e.g. an Earth-mass planet) is close to orbital resonance (Agol et al., 2005).
The possibilities for future follow-up studies of this nature at optical and UV wavelengths are seriously compromised by the failure of STIS on HST. Longer wavelength absorption spectroscopy (1–28 $`\mu `$m) should be possible with the NIRSpec and MIRI instruments on JWST provided that these are configured to allow efficient high-cadence and high S/N observations. If HST is followed by another, similar aperture optical UV telescope before 2015, it is likely that strong arguments will be made to equip it to allow STIS-type transit spectroscopy.
CRIRES: CRIRES is a cryogenic high-resolution infrared echelle spectrometer to be installed at the UT1 of the VLT in 2005. It covers the wavelength region 950–5200 nm at a maximum resolution of 100 000 (0.2 arcsec slit). The instrument has been designed for stability, and will be suited for radial velocity studies. Furthermore it may be the most powerful ground-based instrument for transit spectrosocopy in the infrared. Interference from the atmosphere is a severe complication at infrared wavelengths and the gain in resolution of factor 30 with respect to ISAAC will help alleviate this problem. Käufl (2002) discusses the use of OH lines in the K band for the detection of extra-solar planet atmospheres. Hydrocarbons like C<sub>2</sub>H<sub>2</sub> or CH<sub>4</sub> are prominent constituents in the atmospheres of Jupiter-like planets in the solar system, and also provide lines in the operating range of CRIRES. The isotope shift of <sup>12</sup>CO and <sup>13</sup>CO will be well resolved (Boogert et al., 2002, 2004). CRIRES will also allow analysis of the atmospheres of planets, moons and comets in our solar system, some of which have a rich organic chemistry. Performed in close collaboration with the solar system community, a survey at high-spectral resolution will result in a reference library for study of extra-solar planets. Comparsion with measurements from space will result in a much better understanding of the relation between integrated spectrum and local physical conditions in the atmospheres.
#### 2.1.3 Reflected Light
Additional phase-dependent effects, such as the modulation of light reflected from a planet, should also be detectable by accurate photometric satellites with a precision of order $`10^4`$. The method can be used, in principle, both for independent detection of planets, and for studying known hot Jupiters. It might be more effective than transit searches in some cases, because the effect may be observable for a wide range of inclinations, and is not confined to a narrow angle around $`90^{}`$. The technique is applicable in two cases:
(a) the stellar light reflected by the planet. The ratio of reflected light to the stellar light is of the order of $`A(R/2d)^2F(\varphi )`$, where $`A`$ is the albedo of the planet, $`d`$ is the planetary orbital distance, and $`F(\varphi )`$ is a function of the orbital phase, of the order of unity. For a 2.5-day period, Jupiter-radius planet at an orbital separation of $`7R_{}`$ around a solar-type star, this ratio is $`5\times 10^5A`$, which is almost detectable with COROT or Kepler. For $`R=1.4R_\mathrm{J}`$, as in HD 209458, the ratio is $`10^4A`$. Arnold & Schneider (2004) point out that the shape of the modulation is sensitive to whether the planet has rings.
(b) a stronger effect might be the black-body emission of a close planet. If the planet rotation is synchronised with the orbital motion, one side of the planet faces the parent star all the time, resulting in different temperatures of the two sides of the planet. Estimates suggest differences of up to 1500 K or more, depending on the distance from the star and on the circulation streams in the planetary atmosphere. The hot side of the planet can contribute some fraction of the light of the system, again with a sinusoidal modulation. In the infrared tail of the black-body radiation, the ratio of the planet to star emission could be of the order of 1/400 (Mazeh, private communication), assuming that the far side of the planet is cold. The effect can be seen only in the infrared, and only if there is a large temperature difference between the two sides of the planet.
#### 2.1.4 Microlensing Searches
Microlensing searches for extra-solar planets take advantage of the very characteristic temporal magnification of a (bright) background star, due to a (faint) star-plus-planet system passing in front of it, as seen from Earth. This method is very different from all the other techniques used for planet searching. There are a number of apparent disadvantages: (i) very small probability for a planetary microlensing event, even if all stars have planets (of order $`10^8`$ for background stars in the Galactic bulge); (ii) potential planets are much more distant than those found with other techniques (of order a few kpc), which means subsequent more detailed investigations of the planet are close to impossible; (iii) the duration of the planet-induced deviation in the microlensing lightcurve can be very short (typically hours to days), and the measurement is not repeatable: it is a once-and-only event; (iv) lightcurve shapes caused by extra-solar planets can be very diverse and do not always yield a unique planet mass/separation fit; (v) the derived property is not the planet mass, but the mass ratio between host star and planet. However these apparent disadvantages of the microlensing method can largely be ‘overcome’, and are more than balanced by its many advantages:
(i) no bias for pre-selected nearby host stars: microlensing will provide a fair ‘mass-selected’ sample of the planet population in the Milky Way;
(ii) no strong bias for planets with large masses: the duration of the planetary signal in the lightcurve is roughly proportional to the square root of the planet mass (with a wide spread); its amplitude, however, is independent of the planet mass to first order (though affected by the finite size of the source star);
(iii) Earth-bound method sensitive down to (almost) Earth-masses: Microlensing is sensitive to lower-mass planets than most other methods (except pulsar searches). In principle, it is possible to even detect Earth-mass planets with ground-based monitoring via microlensing. In practice, however, this would mean extremely high monitoring frequency and photometric accuracy;
(iv) most sensitive for planets within ‘lensing zone’, which overlaps with habitable zone: microlensing is not sensitive to very close-in planets: the signal would be undistinguishable from a star with the combined mass. In the current mode of operation (planet searching with high monitoring frequency of ‘alerted’ events), far out planets are not detected either. The most likely detection range is the so-called lensing zone (between 0.6–1.6 Einstein radii), roughly corresponding to projected separations of a couple of AU;
(v) multiple planet systems detectable: The detection of more than one planet per system is certainly possible with microlensing, though its probability is probably another order of magnitude smaller than for a single planet;
(vi) ‘instantanous’ detection of large semi-major axes possible: The measured (projected) distance between planet and host-star of typically a few AU is, though, only a lower limit to the real semi-major axis;
(vii) detection of free-floating planets (i.e. isolated bodies of planetary mass) possible: space-based searches will have high enough photometric accuracy and monitoring frequency to detect and characterise any existing free-floating planets;
(viii) ultimately best statistics of galactic population of planets: Gravitational microlensing will ultimately provide the best statistics for planets in the Milky Way; it is not bias-free, but the biases in the search technique are of very different character from those of all other methods, can easily be quantified and are more favorable for global (i.e. Galactic) statistics.
These characteristics of the microlensing technique mean that it is complementary to the other search methods.
The first planet detection with the microlensing technique was published in May 2004 (Bond et al., 2004): The OGLE- and MOA-teams detected a clear caustic-crossing microlensing signal in event OGLE-2003-BLG-235 or MOA-2003-BLG-53 which could only be reproduced with a binary-lens model involving a mass ratio of $`q=0.0039_{07}^{+11}`$. The planetary deviation lasted about one week, with a measured maximum magnification of more than a factor of 12. Assuming a low-mass main sequence primary, this would correspond to a planet of about $`m_\mathrm{P}1.5M_\mathrm{J}`$ at a projected separation of about $`d3`$ AU.
A summary of ongoing or planned microlensing experiments is given in Table 3. (Stellar) microlensing events continue to be observed at large rates, particularly towards the Galactic bulge (which is the prime search direction due to the high density of background stars). Both MOA and OGLE regularly post their microlensing alerts on their web sites. In the context of exo-planets, specialized networks have been established (PLANET, MicroFUN) which perform follow-up observations of ‘alerted’ events at high time resolution and look for possible planetary perturbations in the stellar microlensing light curves. In the 2004 observing season, OGLE alone had detected and alerted on more than 600 stellar microlensing events. With the MOA inauguration of a new 1.8-m dedicated telescope in December 2004, more than 1000 stellar microlensing events will be found per season from 2005 onwards.
A number of studies looked into the statistics of planets from microlensing searches. They come in two kinds, either providing detection/exclusion probabilities for planets in individual lightcurves or for ensembles of events:
In the case of MACHO 98–BLG–3, the estimated probability for explaining the data without a planet is $`<1`$%. The best planetary model has a planet of $`0.41.5M_{}`$ at a projected radius of either 1.5 or 2.3 AU (Bond et al., 2002).
Very high magnification events are well-suited for showing signatures of planets, because the relative track is very close to the central caustic which should be slightly perturbed by the existence of any planet (at the same time, such planets are difficult to characterise uniquely): In the case of MOA 2003–BLG–32 = OGLE 2003–BLG–219 (Abe et al., 2004), with a peak magnification of more than 500, continuous observations around the maximum did not show any planetary signature. This enabled the authors to put very stringent limits on the probability of a companion: planets of $`m_\mathrm{P}=1.3M_{}`$ are excluded from more than 50% of the projected annular region from $`2.33.6`$ AU surrounding the lens star, Uranus-mass planets from 0.9–8.7 AU, and planets 1.3 more massive than Saturn are excluded from 0.2–60 AU.
The best published statistical limits on the frequency of Jupiter-mass planets from (lack of) microlensing signatures can be found in Gaudi et al. (2002), based on the first five years (1995–99) of PLANET team data. They concluded that less than 33% of the M-dwarfs in the Galactic bulge have Jupiter-mass companions with a projected separation between 1.5–4 AU. Many more stellar events have been monitored subsequently, and improved limits should become available soon.
The availability of the VST may soon provide a means for ESO to support a massive microlensing search for planets (cf. Sackett 1997, Appendix C of the Final report of the ESO Working Group on the Detection of Planets).
A space-based microlensing mission (MPF/GEST) is discussed in Section 2.2.3.
#### 2.1.5 Astrometry
The principle of planet detection with astrometry is similar to that underlying the Doppler technique: the presence of a planet is inferred from the motion of its parent star around the common centre of gravity. In the case of astrometry the two components of this motion are observed in the plane of the sky; this gives sufficient information to solve for the orbital elements without $`\mathrm{sin}i`$ ambiguity. Astrometry also has advantages for a number of specific questions, because this method is applicable to all types of stars, and more sensitive to planets with larger orbital semi-major axes. Astrometric surveys of young and old planetary systems will therefore give unparalleled insight into the mechanisms of planet formation, orbital migration and evolution, orbital resonances, and interaction between planets. Interferometric techniques should improve astrometric precision well beyond current capabilities.
Specific applications are:
(a) mass determination for planets detected in radial velocity surveys (without the $`\mathrm{sin}i`$ factor). The radial velocity method gives only a lower limit to the mass, because the inclination of the orbit with respect to the line-of-sight remains unknown. Astrometry can resolve this ambiguity, because it measures two components of the orbital motion, from which the inclination can be derived;
(b) confirmation of hints for long-period planets in radial velocity surveys. Many of the stars with detected short-period planets also show long-term trends in the velocity residuals (Fischer et al., 2001). These are indicative of additional long-period planets, whose presence can be confirmed astrometrically;
(c) inventory of planets around stars of all masses. The radial velocity technique works well only for stars with a sufficient number of narrow spectral lines, i.e., fairly old stars with $`M<1.2M_{}`$. Astrometry can detect planets around more massive stars and complete a census of gas and ice giants around stars of all masses;
(d) detection of gas giants around pre-main-sequence stars, signatures of planet formation. Astrometry can detect giant planets around young stars, and thus probe the time of planet formation and migration. Observations of pre-main-sequence stars of different ages can provide a test of the formation mechanism of gas giants. Whereas gas accretion on $`10M_{}`$ cores requires $`10`$ Myr, formation by disk instabilities would proceed rapidly and thus produce an astrometric signature even at very young stellar ages;
(e) detection of multiple systems with masses decreasing from the inside out. Whereas the astrometric signal increases linearly with the semi-major axis $`a`$ of the planetary orbit, the radial velocity signal scales with $`1/\sqrt{a}`$. This leads to opposite detection biases for the two methods. Systems in which the masses increase with $`a`$ (e.g., $`\upsilon `$ And) are easily detected by the radial velocity technique because the planets’ signatures are of similar amplitudes. Conversely, systems with masses decreasing with $`a`$ are more easily detected astrometrically;
(f) determine whether multiple systems are coplanar or not. Many of the known extra-solar planets have highly eccentric orbits. A plausible origin of these eccentricities is strong gravitational interaction between two or several massive planets. This could also lead to orbits that are not aligned with the equatorial plane of the star, and to non-coplanar orbits in multiple systems.
Astrometric observations by interferometry are based on measurements of the delay $`D=D_{\mathrm{int}}+(\lambda /2\pi )\varphi `$, where $`D_{\mathrm{int}}=D_2D_1`$ is the internal delay measured by a metrology system, and $`\varphi `$ the observed fringe phase. Here $`\varphi `$ has to be unwrapped, i.e., not restricted to the interval $`[0,2\pi )`$. In other words, one has to determine which of the sinusoidal fringes was observed. This can, for example, be done with dispersed-fringe techniques (Quirrenbach, 2001). $`D`$ is related to the baseline $`\stackrel{}{B}`$ by $`D=\stackrel{}{B}\widehat{s}=B\mathrm{cos}\theta `$, where $`\widehat{s}`$ is a unit vector in the direction towards the star, and $`\theta `$ the angle between $`\stackrel{}{B}`$ and $`\widehat{s}`$. Each data point is thus a one-dimensional measurement of the position of the star $`\theta `$, provided that the length and direction of the baseline are accurately known. The second coordinate can be measured with a separate baseline at a roughly orthogonal orientation. The photon noise limit for the precision $`\sigma `$ of an astrometric measurement is given by $`\sigma =(1/\mathrm{SNR})(\lambda /(2\pi B))`$. Since high signal-to-noise ratios can be obtained for bright stars, $`\sigma `$ can be orders of magnitude smaller than the resolution $`\lambda /B`$ of the interferometer. For example, the resolution of SIM ($`B=10`$ m) is about 10 milli-arcsec, but the astrometric precision should approach 1 micro-arcsec; for PRIMA, $`B=200`$ m, the resolution is 2 milli-arcsec, and the astrometric precision should approach 10 micro-arcsec.
Because of the short coherence time of the atmosphere, precise astrometry from the ground requires simultaneous observations of the target and an astrometric reference. In a dual-star interferometer, each telescope accepts two small fields and sends two separate beams through the delay lines. The delay difference between the two fields is taken out with an additional short-stroke differential delay line; an internal laser metrology system is used to monitor the delay difference (which is equal to the phase difference multiplied with $`\lambda /2\pi `$). For astrometric observations, this delay difference $`\mathrm{\Delta }D`$ is the observable of interest, because it is directly related to the coordinate difference between the target and reference stars; it follows that $`\mathrm{\Delta }DD_tD_r=\stackrel{}{B}\left(\widehat{s}_t\widehat{s}_r\right)=B(\mathrm{cos}\theta _t\mathrm{cos}\theta _r)`$, where the subscript $`t`$ is used for the target, and $`r`$ for the reference. To get robust two-dimensional position measurements, observations of the target with respect to several references and with a number of baseline orientations are required.
Measurements of the delay difference between two stars give relative astrometric information; this means that the position information is not obtained in a global reference frame, but only with respect to nearby comparison stars, which define a local reference frame on a small patch of sky. This approach greatly reduces the atmospheric errors, and some instrumental requirements are also relaxed. The downside is that the information that can be obtained in this way is more restricted, because the local frame may have a motion and rotation of its own. This makes it impossible to measure proper motions. Moreover, all parallax ellipses have the same orientation and axial ratio, which allows only relative parallaxes to be measured.
Specific instrument approaches are discussed in Section 2.1.6 (NAOS-CONICA, Planet Finder, PRIMA) and Section 2.2.2 (Gaia, SIM, etc.).
No planets have been discovered using this technique to date.
#### 2.1.6 Direct Detection
The light coming from an extra-solar planet is much fainter (of order $`10^9`$ in the optical, and a factor 10–100 less in the infrared) than the signal from the star. Therefore the challenge is to build instruments that are able to provide extremely high contrast and spatial resolution. The different approaches are summarised below and in Table 4. The first direct detection of a young planet may already have been achieved by a team using NACO on the VLT. An object detected close to 2MASS WJ1207334–393254 is either a planet or possibly a brown dwarf (Chauvin et al., 2004). Regardless of the exact nature of this particular object, it is likely that imaging of more massive, young extra-solar planets will become more feasible in the near future.
A number of programmes are using, or planning to use, interferometry to achieve high spatial resolution. Destructive interference can be used to remove most of the light from the central star (nulling). ESA and ESO are collaborating on a nulling demonstrator for Darwin called GENIE which will be used on the VLT. Other searches use a coronographic approach to block out the star’s light. The Lyot project’s coronograph has now been declared fully operational and will conduct a survey of 300 nearby stars.
NAOS-CONICA: NAOS-CONICA (NACO) is installed on UT3 at the VLT. It is an adaptive optics system working in the 1–5 $`\mu `$m range, with a Shack-Hartmann wavefront analyser operating in the visible or near-infrared. The instrument is equipped with a large collection of broad- and narrow-band filters for imaging, and a set of grisms for low-dispersion spectroscopy. It also features a polarimetric system. In good conditions, with a bright reference star, a Strehl ratio of $`0.5`$ is achievable.
Two modes are of special interest for planetary observations: (1) a coronographic mode: a classical Lyot-type instrument with a circular focal spot (of 0.7 or 1.4 arcsec) is used together with an undersized pupil mask. A four quadrant phase mask (which introduces a shift of $`\pi `$ to the wavefront) is being commissioned. It reduces the light of the central star by a factor of $`70`$, and permits observations within 0.35 arcsec of the centre, i.e. much closer than with a classic mask; (2) simultaneous differential imaging over $`5\times 5`$ arcsec<sup>2</sup>: four images are obtained simultaneously through 3 narrow-band filters. Two are taken in the 1.625 $`\mu `$m methane feature, and the two others at 1.575 and 1.600 $`\mu `$m, outside the spectral line. The data are registered simultaneously in the four channels, and the point-spread function (including all its residual aberrations, and the speckles, including super-speckles) are identical in all four images. This mode was designed to search for methane-rich objects near very bright stars, with a contrast of 50 000 accessible. With these characteristics, a Jupiter-like planet in a Jupiter-like orbit around a very nearby star (within 5 pc) should be just detectable at its largest elongation. Detection performances are limited by uncorrected phase residuals (mainly low-order aberrations). NACO should be considered as a prototype permitting the study of novel techniques that will be used in dedicated instruments such as ESO’s Planet Finder.
Planet Finder: The next step beyond the VLT AO facility NAOS-CONICA (NACO) would be a dedicated VLT instrument optimised for the detection of extra-solar planets. Two independent design studies are currently underway for such a Planet Finder instrument at the VLT. The prime goal is to gain at least an order of magnitude with respect to NACO in the detection of faint objects very close to a bright star, ideally reaching giant planets. Much higher Strehl ratios than NAOS, around 0.9 in the K-band, are targetted. Planet Finder will combine high-order adaptive optics with differential detection techniques; multi-waveband imaging, integral-field spectroscopy, and imaging polarimetry are foreseen for the focal plane instruments. Planet Finder could become operational around 2009. A review board for the assessment of the Phase A studies met on 16–17 December 2004.
Planet Finder may discover giant planets in different phases of their evolution. During the ongoing contraction and accretion phases, the internal luminosity of these planets exceeds the reflected light contribution by several orders of magnitude, e.g. a Jupiter mass planet will be 10<sup>3</sup> times brighter at 1 Myr than at 1 Gyr. This raises the possibility of detecting young planets around the closest young stars, in spite of their relatively large distances. Planet Finder will also search for old planets in the Solar neighbourhood. The S/N for the detection of exo-planets drops rapidly with distance, due to the combined effects of inverse-square brightness losses and the reduction in stellar-planet angular separation. The most promising targets for old systems are therefore within 5–10 pc, exploring the range in separation down to $`35`$ AU. For this reason, possible targets for Planet Finder may be found among stars known to have planetary systems from high-precision radial velocity surveys. Even more promising is the synergy with planet searches using astrometric perturbations: giant planets detected by Planet Finder should give a signal in the tens of milli-arcsecond range, clearly measurable with PRIMA and/or future space missions (SIM, Gaia). This will provide an independent estimate of planetary masses.
The detailed science cases will probably differ between both groups because of different AO system performances and focal instruments. In the French-led proposal, typical targets are: (a) stars in young associations (more than 100 candidates), which will be looked at much closer than with NACO (with an interference coronagraph, at 1–2 $`\lambda /D`$, achievable contrast $`>10^5`$); (b) close-by solar-type stars at moderate ages (1–2 Gyr): contrast: a few $`10^5`$; about 200 objects; for some nearby objects performances should be much better; (c) late-type stars of all ages: with better performances on the youngest and closest ones; (d) other science: disks and stellar environments. In the German-led proposal (CHEOPS) the primary goals are to find mature (old, Jupiter-like) planets in nearby systems (within 15 pc), with polarimetry possible; and young, still-warm planets in the nearest star-forming regions (within 100 pc) with an integral-field spectrograph operating in the J and H bands. Secondary goals are to observe brown dwarfs, young stellar object disks, debris disks, etc.
PRIMA: The ESO VLT Interferometer consists of four 8 m Unit Telescopes and four moveable 1.8 m Auxiliary Telescopes, which can form baselines up to 200 m in length. The PRIMA (Phase-Reference Imaging and Micro-Arcsecond Astrometry, Quirrenbach et al. (1998)) facility will implement dual-star astrometry at the VLTI; it is expected to become operational in 2007. Its goal is to measure the masses and orbital inclinations of planets already known from radial velocity surveys. In addition, a survey with PRIMA will be conducted to establish the frequency of planets along the main sequence and through time.
The principles of interferometric astrrometry are summarised in Section 2.1.5. Differential phase observations with a near-IR interferometer offer a way to obtain spectra of extra-solar planets. The method makes use of the wavelength dependence of the interferometer phase of the planet/star system, which depends both on the interferometer geometry and on the brightness ratio between the planet and the star. The differential phase is strongly affected by instrumental and atmospheric dispersion effects. Difficulties in calibrating these effects might prevent the application of the differential phase method to systems with a very high contrast, such as extra-solar planets. A promising alternative is the use of spectrally resolved closure phases, which are immune to many of the systematic and random errors affecting the single-baseline phases.
Figure 3 shows the predicted response of the AMBER instrument at the VLTI to a realistic model of the 51 Peg system, taking into account a theoretical spectrum of the planet as well as the geometry of the VLTI. Joergens & Quirrenbach (2004) have presented a strategy to determine the geometry of the planetary system and the spectrum of the extra-solar planet from such closure phase observations in two steps. First, there is a close relation between the nulls in the closure phase and the nulls in the corresponding single-baseline phases: every second null of a single-baseline phase is also a null in the closure phase. This means that the nulls in the closure phase do not depend on the spectrum but only on the geometry, so that the geometry of the system can be determined by measuring the nulls in the closure phase at three or more different hour angles. In the second step, the known geometry can then be used to extract the planet spectrum directly from the closure phases.
ALMA: ALMA is an interferometer in the mm-wavelength range, which will consist of sixty-four 12 m diameter antennae located in northern Chile, at 5050 m altitude. The antennae can be spaced from a compact configuration with a maximum separation of 150 m to a very extended configuration where the maximum spacing is 16 km, providing a resolution of 10 milli-arcsec at shortest wavelengths. The receivers will cover the atmospheric windows in the 35–1000 GHz range (350 $`\mu `$m – 7 mm) with a bandwidth of 8 GHz in two polarizations, and a resolution of 32000 channels. ALMA will be powerful for studying the disks around young stars, able to image such disks out to several hundred parsecs, providing density and temperature profiles (through measurements of thermal dust emission), and providing constraints on disk dynamics and chemistry (through measurements of spectral lines). In the case of protoplanetary disks, ALMA will be able to image gaps and holes caused by protoplanets.
In terms of direct detection of the planet themselves, however, ALMA is rather limited. At its best resolution (widest configuration), the system will be able to resolve a Jupiter-like planet from its star out to 100–150 pc. The main limitation comes from the flux of the planet. The best frequency for planet observation, which optimises the combination of expected detector noise characteristics, the spectrum of the objects, and the site characteristics, is at 350 GHz. The flux density of the planet is directly related to their temperature, size and distance as $`F_{350}=6.10^8TR_\mathrm{J}^2/D^2`$, where the distance $`D`$ is in pc, $`R_\mathrm{J}`$ the radius expressed in Jupiter units, and $`T`$ the temperature in K. Together with the expected sensitivity performances of ALMA, this indicates that a Jupiter would be detectable only out to about 1 pc. In the case of a ‘hot Jupiter’ ($`R=1.5R_\mathrm{J}`$, $`T=1000`$ K), this limit is pushed to a few parsecs, but still not far enough to actually encompass any useful star. On the other hand, a proto-Jupiter (with $`R=30R_\mathrm{J}`$ and $`T=2500`$ K) would be detectable in a matter of minutes to hours out to several tens of parsecs.
The contrast between the star and the surrounding bodies, a critical factor at shorter wavelengths (visible–infrared), becomes an advantage in the case of ALMA. The contrast factor is of the order of 1000, which is well within the dynamic range of the detectors, and the bright central source helps maintaining the optimal coherence of the interferometer. So, while ALMA is not expected to contribute significantly to the study of mature planets, its contribution will be very significant for studying early stages of planet formation, from nebula to protoplanet.
#### 2.1.7 Other Searches
The upper right-hand section of Figure 1 indicates some miscellaneous search methods. Jupiter’s magnetosphere is known to produce strong emission of radio waves. These decametric bursts are targeted by a number of collaborative efforts in the radio community, as summarised in Table 4. As noted in Section 2.2.2, Gaia might also detect a few protoplanetary collisions photometrically (Zhang & Sigurdsson, 2003), although whether they could ever be recognised as such, buried within such a vast volume of other variables, has not been assessed.
### 2.2 Space Observations: 2005–2015
In the near-term future of space missions, there are two principal detection approaches: transits (exemplified by the COROT, Kepler and Eddington missions), and astrometry (exemplified by Gaia and SIM). A non-approved concept, MPF (originally GEST), uses microlensing to expand on the parameter space for which statistical information on planet frequency would be provided.
#### 2.2.1 Space Transit Measurements: COROT, Kepler and Eddington
COROT: COROT is a French-European-ESA collaboration, led by CNES, comprising a 27 cm telescope with a CCD camera, with a launch planned for June 2006. After MOST, it will be the second satellite dedicated to long-term high-accuracy photometric monitoring from space. COROT will combine the study of asteroseismology with the search for exo-planetary transits. The observation of 60 000 stars (12 000 stars simultaneously for 150 days each) is expected to result in the detection of a few hot telluric planets.
As a general remark, which applies to the other predictions in this section, as well as to the radial velocity and ground-based transit searches discussed previously, it should be noted that the expected rate of exo-planet detection is difficult to quantify, given both their unknown frequency of occurrence, and detection uncertainties due to stellar activity.
The numbers in Table 5 are based on the number of dwarf stars in the COROT fields, the mass and period distribution of known exo-planets, the probability of transits, the fact that no more than 1 short-period exo-planet per star is expected, and the expected accuracy of COROT as a function of magnitude. Table 3 in Bordé et al. (2003) assumes one planet per dwarf star, and a uniform orbital distribution law. This leads to a significantly larger number of predicted detections, and is presumably an overestimate of the actual number of planets expected (as they also note). In practice, the critical factor in all these predictions is the unknown number of low-mass planets per star.
Kepler: improved prospects for photometric transit detections will come with NASA’s Kepler mission, due for launch around 2007. Kepler is a 0.95 m aperture, differential photometer with a 105 square degree field of view. It focuses on the detection of Earth-size planets or larger in or near the habitable zone of a wide variety of stellar spectral types, monitoring some $`10^5`$ main-sequence stars brighter than 14 mag. Detection of some 50–640 terrestrial inner-orbit planet transits are predicted, depending on whether their typical radii lie in the range $`R1.02.2R_\mathrm{E}`$, determining the distribution of sizes and orbital characteristics. Kepler will assist the preparation of future programmes like SIM and Darwin/TPF by identifying the common stellar characteristics of host stars, and defining the volume of space needed to search.
The numbers in Table 5 are taken from the Kepler www site, and are based on the following assumptions: 100 000 main-sequence stars observed with a precision of better than $`5\times 10^5`$; typical variability of 75% of the stars is similar to that of our Sun; most main-sequence stars, including binaries, have terrestrial planets in or near the habitable zone; on average, two Earth-size or larger planets exist between 0.5–1.5 AU; transit probability for planets in the habitable zone is 0.5% per planet; the transit is near grazing in a 1-year orbit; each star has one giant planet in an outer orbit; on average 1% of the main-sequence stars have giant planets in orbits shorter than 1 week and comparable numbers in periods of 1–4 weeks and 1–12 months; mission life time of 4 years. Results for giant planets are expected around 2007, with those on terrestrial planets (which will require more careful verification) around 2010.
Eddington: ESA’s Eddington mission was originally proposed for launch around 2008. It entered ESA’s science programme as a ‘reserve’ mission, was approved in 2002, but cancelled in November 2003 due to overall financial constraints. The Eddington payload was composed of three identical, co-aligned telescopes with a $`0.7`$ m aperture and identical $`3\times 2`$ mosaic CCD cameras, a total collecting area of $`0.75`$ m<sup>2</sup> and a field of view of $`20`$ deg<sup>2</sup>. Each telescope had a slightly different bandpass, allowing colour information to be derived for high S/N transits.
The baseline lifetime was 5 years (extended operations possible), of which three would be dedicated to a single long observation (currently baselined in Lacerta), and two would be used for short (one to a few months) observations of other fields. Planet searches were to be conducted during the three years observation, in $`10^5`$ stars, of which $`10^4`$ would be observed with sufficient accuracy to detect Earth-like planets. During the shorter observations, a further $`4\times 10^5`$ stars would be searched for planets, allowing many shorter-period planets to be discovered.
The numbers in Table 5 are the results of detailed Monte Carlo simulations, which assume a ‘standard’ planet function of the form $`f(a,m|M)a^\alpha m^\beta M^\gamma `$ where $`a`$ is the orbital radius of the planet, $`m`$ its mass, and $`M`$ the mass of the parent star, and using the mass-radius relationship obtained from solar system objects. The simulations used $`\gamma =0`$, i.e. the planet function is independent of the stellar mass. As both the Doppler planet population and the solar system objects are consistent with $`\alpha =1`$, $`\beta =1`$, the simulations have also used $`f(a,m)a^1m^1`$ normalised to 0.01 hot Jupiters per star (as derived from the Doppler surveys). In this assumption, each star has 0.25 normal Jupiters ($`r>R_\mathrm{J}`$), 0.01 hot Jupiters ($`r>R_\mathrm{J}`$, $`P=35`$ day), 0.85 Earths ($`R>R_{}`$) at any orbital distance, and 0.07 ‘habitable planets’ (with liquid water temperatures). This is considered as a conservative mass function, as the typical star probably has fewer rocky planets than the solar system. Also, the simulations are stopped at $`R=1R_\mathrm{J}`$ (because of the emphasis on the smaller planets), so that the number of gas giants is under-predicted. The result of the Monte Carlo simulation predicts, for the single 3-year observation: 14 000 planets in total (in 12 500 planetary systems), of which 8000 hot planets, of which 5600 with $`R>5R_{}`$, 660 Earths ($`0.5<R<2.0R_{}`$); 160 habitable zone planets, of which 20 ‘Earths’. A larger number of hot Jupiters would also be found in the short (asteroseismology) observations. Predictions for multiple systems depend sensitively on the assumed distribution of relative orbital inclinations.
The following missions, not dedicated to transits, can also be noted:
HST: although suitably placed above the atmosphere such that low-mass planets could be detected by HST using this technique in principle, HST is not a dedicated transit discovery instrument, and its discovery efficiency is constrained by its limited field of view, and available observing time. Any searches using HST will therefore likely be restricted to observations of especially high-surface density regions. Observations of 47 Tuc (34 000 main-sequence stars monitored for 8.3 days) failed to detect planets (Gilliland et al., 2000), whereas 17 would have been expected based on radial velocity surveys. This non-result is currently attributed to effects of metallicity (Gonzalez, 1998), ultraviolet evaporation (Armitage, 2000), or collisional disruption (Bonnell et al., 2001) of the protoplanetary disks in this crowded stellar environment. An advance in transit statistics should come from observations of the Galaxy bulge with HST by Sahu et al., monitoring 100 000 stars to $`V=23`$ over 7 days in February 2004, with results expected in spring 2005.
Nevertheless HST, and its successor JWST, have considerable potential for follow-up observations of transiting systems discovered by other methods, for example by Kepler. This issue is developed further in Section 4.
MOST: the Canadian satellite MOST is a 15 cm telescope launched on 30 June 2003. Dedicated to the long-term photometric monitoring of a small number of stars primarily for asteroseismology studies, it has a photometric performance just a factor of 2 better than from ground (Matthews et al., 2004). Although with limited transit discovery potential, it will nevertheless aim to detect reflected light of a few known hot Jupiters.
#### 2.2.2 Space Astrometry Missions: Gaia and SIM
As noted previously, astrometric planet detection involves detecting the system’s photocentric motion on the plane of the sky, in the same way that radial velocity detection involves detecting the system’s photocentric motion along the line of sight. The amplitude of the displacement, and therefore the system’s detectability, can be characterised by the system’s ‘astrometric signature’, $`\alpha =(M_{\mathrm{Planet}}/M_{\mathrm{star}})(a/d)`$. This signature is measured in arcsec when the orbital radius $`a`$ is measured in AU and the distance $`d`$ is measured in pc. Astrometric measurements can provide the planetary mass directly rather than $`M\mathrm{sin}i`$ (as provided by radial velocity techniques) if $`d`$ is determined and if $`M_{\mathrm{star}}`$ is estimated from stellar evolutionary theory.
Figure 4 shows the astrometric signature versus orbital period for the known exo-planets, where the size of the circles indicates the planetary mass. The horizontal line at the top of the figure indicates the Hipparcos astrometric accuracy, and shows immediately why Hipparcos was unable to detect new planetary systems. Neverthess, Hipparcos data was useful for placing some constraints on the masses of planet candidates (Zucker & Mazeh, 2001).
Gaia (ESA) and SIM (NASA) are two very different approaches to space astrometry, both approved and under development:
Gaia: Gaia is a scanning, survey-type instrument, with a launch around 2011 (Perryman et al., 2001). Its detectability domains are shown in Figure 4: periods below about 0.2 yr will not be detectable because of the relatively long times between successive observations dictated by the scanning law, while periods longer than about 12 yr will result in photocentric motions indistinguishable from rectilinear motion over the mission’s measurement duration (about 5 years). As seen in the figure, Gaia will therefore contribute substantially to the large-scale systematic detection of Jupiter-mass planets (or above) in Jupiter-period orbits (or smaller); some 10–20 000 detections out to 150–200 pc are expected (Lattanzi et al., 2000; Sozzetti et al., 2001), including confirmation of most of the (longer-period) radial velocity detections known to date. Planetary masses, $`M`$, rather than $`M\mathrm{sin}i`$, will be obtained. Full orbital parameters will be obtained for some 5000 (higher S/N) systems. Relative inclinations can in principle be obtained for multiple systems with favourable orbits (Sozzetti et al., 2001), important for studies of formation scenarios and orbital stability of multiple systems. Some 4–5000 transit systems, of the hot-Jupiter type, might also be detected (Robichon, 2002). Gaia might also detect a handful of protoplanetary collisions photometrically (Zhang & Sigurdsson, 2003). Gaia cannot observe systems at epochs other than those determined by its fully deterministic scanning law, and will not detect planets with masses much below 10–20 $`M_{}`$ unless such systems exist within 10–20 pc.
SIM: SIM is a pointed interferometer with a launch around 2010 (Danner & Unwin, 1999): accuracies of a few micro-arcsec down to 20 mag are projected. Such faint observations will be expensive in terms of observing time, and brighter target stars are likely to be the rule. Of 15 key projects and mission scientist programmes currently studied, three focus on planetary systems: (1) A Search for Young Planetary Systems (Beichman): this will survey 200 stars with ages from 1–100 Myr (mostly in star-forming regions at 125–140 pc, but including TW Hya at 50 pc) which expects to find anywhere between 10–200 planets, depending on whether the occurrence rate is the canonical 5–7 % from current radial velocity surveys, or 100 % of all young stars. The survey will be sensitive to $`M_\mathrm{J}`$ planets at orbital distances of 1–5 AU. (2) Discovery of Planetary Systems (Marcy): this will focus on searches for 1–3 $`M_{}`$ planets within 8 pc, and for 3–20 $`M_{}`$ planets within 8–30 pc. The survey is also considered as a reconnaissance for TPF. Target stars will be selected from ongoing surveys of the nearest 900 GKM main-sequence stars in the northern hemisphere with the Lick 3-m and the Keck 10-m telescopes, the nearest 200 GK stars in the south with the AAO 3.9-m, and the planned 6.5-m Magellan survey extending to a further 600 GKM stars in the south. (3) Extra-Solar Planet Interferometric Surveys (Shao): this major SIM survey programme comprises a deep survey of about 75 nearby main-sequence stars within about 10 pc of the Sun, of which one third are G dwarfs and the remainder are inactive main-sequence stars of other spectral types (mostly K and M but including a few A and F). Over the mission lifetime, each target will be observed some 70 times, each of twenty 1-minute observations resulting in a final accuracy of about 1 micro-arcsec. Each pointing will be accompanied by the observation of typically 28 additional bright nearby stars (within 25 pc), with single 1-minute observations leading to some 2000 stars observed with accuracies of about 4 micro-arcsec. These will be from diverse types: all main-sequence spectral types, binaries, a broad range of age and metallicity, dust disks, white dwarfs, planets from radial velocity surveys, etc. Preparatory programmes for this survey include radial velocity monitoring and adaptive optic imaging. The total expected number of new detections from SIM, for any given planetary mass and orbital radius is again not straightforward to predict, and depends on the (unknown) mass distribution of exo-planets versus orbital radius at $`a1`$ AU. Estimates are given in Table 5.
The NASA SMEX proposal AMEX (which followed on from the FAME study) aimed at 150 micro-arcsec accuracy at 9 mag and 3 milli-arcsec at 15 mag and, with a proposed launch in 2007–08, would have provided limited prospects for planet detection through the comparison of proper motions with Hipparcos, including some 600 detections to 30 pc down to K5V stars, and transits to V =11 mag. AMEX was not selected by NASA in 2003. In mid-2004 NASA announced the selection of nine studies for future mission concepts within its ‘Astronomical Search for Origins Program’, including the ‘Origins Billion Star Survey’ (OBSS) focussing on a census of giant extra-solar planets using the principles of Gaia. If OBSS is selected, its contribution to astrometric exo-planet research would not surpass those of Gaia.
The Japanese mission JASMINE, and a potential prototype nano-JASMINE, have been under discussion at a low level in Japan for several years (Gouda et al., 2002). Originally conceived as a mini-Gaia but able to concentrate on the Galactic centre by operating in the infrared, the mission’s technical feasibility has been improved in the past few months (although its scientific niche has been weakened) with the move to CCD detector technology.
#### 2.2.3 Space-Based Microlensing: MPF
Proposals for exo-planet detection through their microlensing signatures have been made. GEST (Galactic Exo-Planet Survey Telescope (Bennett & Rhie, 2002)) was proposed for a NASA mission in 2001–02 (a Survey for Terrestrial Exo-Planets (STEP) was also submitted to NASA’s Extra-Solar Planets Advanced Concepts Program at the same time). It was not selected in 2002, but was re-submitted during 2004 under the name of Microlensing Planet Finder (MPF), using HgCdTe and Si-PIN detectors in place of the earlier CCDs (Bennett et al., 2004).
A 1.2-m aperture telescope with a 2 deg<sup>2</sup> field of view continuously monitors $`10^8`$ Galactic bulge main-sequence stars. In about one case out of a million, sources in the bulge are lensed by foreground (bulge or disk) stars which are accompanied by the planets being sought. Observing high surface-density sky regions improves lensing probabilities to sufficient levels that successful detections can be expected over reasonable observing times. Space observations are considered mandatory to permit the high photometric accuracy required for detection even in very crowded regions where seeing limits the achievable photometric accuracy and hence detectability achievable from the ground.
Microlensing probes particular exo-planet domains: for example, low-mass planets can be detected, albeit usually at very large distances of typically 5–8 kpc. The sensitivity of such measurements is highest at (projected) orbital separations of 0.7–10 AU, but it will also detect systems with larger separations, masses as low as that of Mars, large moons of terrestrial planets, and some 50 000 giant planets via transits with orbital separations of up to 20 AU (the prime sensitivity of a transit survey extends inward from 1 AU, while the sensitivity of microlensing extends outwards). There are theoretical reasons to believe that free-floating planets may be abundant as a by-product of planetary formation, and MPF/GEST will also detect these.
The planetary lensing events have a typical duration of 2–20 hr (compared to the typical 2–20 weeks duration for lensing events due to stars), and must be sampled by photometry of $`1`$% accuracy several times per hour over a period of several days, and with high angular resolution because of the high density of bright main-sequence stars in the central bulge. The proposed polar orbit is oriented to keep the Galactic bulge in the continuous viewing zone. Most of the multiple-planet detections in the simulations of Bennett & Rhie (2002) are systems in which both ‘Jupiter’ and ‘Saturn’ planets are detected. Since multiple orbits are generally stable only if they are close to circular, a microlensing survey will be able to provide information on the abundance of giant planets with nearly circular orbits by measuring the frequency of double-planet detections and the ratios of their separations.
Just over 100 Earths would be detected if each lens star has one in a 1 AU orbit. The peak sensitivity is at an orbital distance of 2.5 AU, with 230 expected detections if each lens star had a planet in such an orbit. Although the prime quantity obtained from a microlensing detection is the mass ratio between planet and star, additional information or hypotheses can be combined to estimate the mass of the host star, the planetary mass, the distance to the host star, and the planet-star separation in the plane of the sky.
One of the disadvantages of lensing experiments is that a planet event, once observed, can never (in practice) be seen again – follow-up observations for further characterisations are not feasible (unlike the case for any of the other principal detection methods). Nevertheless, a mission like MPF/GEST will provide important observational and statistical data on the occurrence of low-mass planets (Earth to Jupiter masses), low-mass planets at larger orbital radii, multiple systems and, significantly, free-floating planets formed as a by-product of the system formation.
Detection by microlensing could in principle also be included in the Eddington mission, but the Eddington team has made no detailed evaluation of feasibility, and its inclusion would be likely to drive instrumental requirements in a non-trivial manner.
#### 2.2.4 Other Space Missions: JWST, Spitzer, SOFIA
JWST: JWST (http://www.jwst.nasa.gov/) is a collaboration between NASA, ESA and the CSA. It will be a passively-cooled (40–50 K) observatory spacecraft with an 18-segment primary mirror having an effective aperture of about 6.5 m and diffraction-limited performance at 2 $`\mu `$m, equipped with four principal science instruments and a fine guidance sensor. It is scheduled for launch in 2011 into an L2 Lissajous orbit. The observatory is optimised for the 1–5 $`\mu `$m band, but will be equipped to cover 0.6–28 $`\mu `$m with a combination of imaging (through fixed and tunable filters) and low- to moderate-resolution ($`100<R<3000`$) spectroscopy. The ‘Origin and Evolution of Planetary Systems’ is one of JWST’s five science themes, and the science requirements are drafted accordingly.
The Design Reference Mission (http://www.jwst.nasa.gov/ScienceGoals.htm) describes the exo-planet programmes currently foreseen for JWST. The survey programmes are aimed at finding giant planets and isolated objects using direct imaging, and bound planets using coronography. Follow-up studies are planned using tunable filter imaging ($`R100`$) and slit spectroscopy. For isolated sources, objects at AB = 30 mag can be reached using the near-infrared camera (NIRCam). The tunable filter can reach AB = 27 mag, while mid-infrared spectroscopy (with MIRI) can reach AB = 23 mag at $`R3000`$. For widely-separated giant planets, $`R100`$ coronography will provide preliminary temperature estimates and for these and for isolated systems, $`R1000`$ near-IR spectroscopy will access metallicity indicators. Synoptic observations of bodies in our own Solar System, such as Titan, over the 10-year lifetime of JWST will begin the study of secular surface and atmospheric changes.
A report on ‘Astrobiology and JWST’ (Seager & Lunine, 2004) listed three areas where the technical capabilities of JWST should be optimised for the follow-up of transit events: (1) in principle, JWST can measure the transmission spectra of giant planet atmospheres during planet transits of bright stars (7–14 mag) but this requires capabilities of rapid detector readout and high instrument duty-cycle in order to achieve very high S/N over a typical transit time (12 hr). If Earth-sized planets are common and detected in transit around stars brighter than 6 mag, the JWST near-IR spectrograph (NIRSpec) could detect atmospheric biomarker signatures; (2) the collection of $`10^8`$ photons per image for NIRSpec, by spreading photons over $`10^5`$ spatial+spectral pixels, would enable JWST to characterise atmospheres during the transit of a terrestrial planet in the habitable-zone of a solar-type star; (3) NIRSpec is important for characterising transiting extra-solar planets. Many transiting extra-solar planets are expected to be found in the next several years with both ground-based and space-based telescopes (including Kepler). The short wavelength end is especially important for detecting scattered light and characterising planetary albedos. In particular, NIRSpec’s long-slit configuration is essential for these observations. Thus JWST, even more so than HST, has considerable potential for follow-up observations of transiting systems discovered by other methods, for example by Kepler. This issue is developed further in Section 4.
Spitzer: Spitzer (ex-SIRTF, http://www.spitzer.caltech.edu/) is an 85 cm aperture, liquid helium cooled telescope in an Earth-trailing heliocentric orbit. Launched in August 2003, it has a projected lifetime (minimum) of 2.5 years with a goal of 5 years or more. The instrument complement provides the capabilities for imaging/photometry from 3–180 $`\mu `$m, spectroscopy from 5–40 $`\mu `$m and spectrophotometry from 50–100 $`\mu `$m. Spitzer’s expected contribution to the field of exo-planet research lies in its ability to measure excess radiation from dust disks over the critical mid-infrared wavelength range. The imaging capability is determined by the diffraction limit of the relatively small telescope (1.5 arcsec at 6.5 $`\mu `$m). The sensitivity, however, allows the detection of dust masses to below the mass in small grains inferred in our Kuiper Belt ($`6\times 10^{22}`$ gm) surrounding a Solar-type star at 30 pc.
One of the six Legacy Programs is concerned explicitly with the formation and evolution of planetary systems (see: ‘The Formation and Evolution of Planetary Systems: Placing Our Solar System in Context’ http://feps.as.arizona.edu/). This uses 350 hr of photometric and spectroscopic Spitzer time to detect and characterise the dust disk emission from two samples of solar-like stars, the first consisting of objects within 50 pc spanning an age range from 100–3000 Myr and the second containing objects between 15–180 pc spanning ages from 3–100 Myr.
The first call for Guest Observer programmes resulted in 8 accepted exo-planet proposals out of a total of 202 programmes in all subject areas. This includes one to characterise the atmosphere and evolution of the transiting extra-solar planet HD 209458b. The Cycle 2 call for proposals had a deadline of 12 Feb 2005. The accepted Guest Observer proposals related to exo-planets were:
* Ultracool Brown Dwarfs and Massive Planets Around Nearby White Dwarfs
* The SIM/TPF Sample: Comparative Planetology of Neighbouring Solar Systems
* Evolution of Gaseous Disks and Formation of Giant and Terrestrial Planets
* Searching the Stellar Graveyard for Planets and Brown Dwarfs with SST
* A Search for Planetary Systems around White Dwarf Merger Remnants
* Characterising the Atmosphere and Evolution of HD 209458b
* Survey for Planets and Exozodiacal Dust Around White Dwarfs
* Mineralogy, Grain Growth and Dust Settling in Brown Dwarf Disks
SOFIA: SOFIA (Stratospheric Observatory for Infrared Astronomy) is a joint endeavour of NASA and the German DLR (http://www.sofia.arc.nasa.gov/). A modified Boeing 747SP carrying a 2.7-m telescope will operate at an altitude of about 12 km. The first call for proposals will be issued in August 2005 for the first observing cycle starting in January 2006. A total of nine instrument have been selected, providing imaging and spectroscopy in the range 1–600 $`\mu `$m. Operating out of NASA Ames Research Center, the facility is to observe three or four nights a week for at least twenty years. Its location above the bulk of Earth’s atmosphere will provide access to the mid-infrared region without the limitations of observatories on the ground. Its long projected lifetime will make it possible to conduct muli-epoch observations for variable or evolving objects and the ability to easily exchange instruments will ensure that the latest technology can be incorporated as it becomes available.
The science with SOFIA will revolve around cold matter in our solar system, the interstellar medium, stars and galaxies. Similar to Spitzer, SOFIA is not expected to contribute directly to the discovery of extra-solar planets, but it will enhance understanding of planet formation by studying circumstellar disks. Furthermore it will provide interesting information on the infrared spectra of solar system bodies including the chemistry of atmospheres.
### 2.3 Summary of Prospects 2005–2015
The prospects for the main search experiments described in this section are summarised in Table 5. This table presents only a simplified picture of planet detection capabilities, ignoring the comparative importance of finding large numbers of exo-planets with only an estimation of $`M\mathrm{sin}i`$ or $`r/R`$, or more comprehensively characterising a smaller number of planets (with mass, radius, albedo, and age). It also ignores the fact that different objects will be detected by different methods, and that different methods supply complementary astrophysical information.
While confidence is developing in statistical distributions of planets above about 0.05 $`M_\mathrm{J}`$, the occurrence of lower-mass, and specifically Earth-mass planets remains a matter for speculation. In this spirit, the lower range of detection limits for low-mass planets for Kepler, Eddington and SIM is indicated in the table as 0.
While these transit and astrometric discovery predictions must therefore be taken with certain caveats, they promise a major advance in the detection and knowledge of the statistical properties of a wide range of exo-planets: ranging from massive (Jupiter-mass) planets in long-period (Jupiter-type) orbits, and the occurrence and properties of multiple systems (via astrometry), through to Earth-mass planets in the habitable zone (via transits or microlensing).
## 3 The Period 2015–2025
### 3.1 Ground Observations: 2015–2025
#### 3.1.1 OWL/ELT
The 100-m diameter Overwhelmingly Large Telescope (OWL) is being studied by ESO. S/N = 10 will be reached at 35 mag ($`t=1`$ hr) for imaging, and at 30 mag for $`R=1000`$ spectroscopy ($`t=3`$ hr). The resolution, $`\lambda /D1`$ milli-arcsec in the V-band, is a factor 40 improvement over HST. Detection of Earth-mass exo-planets is part of the scientific case for the 100 m OWL/ELT project (see the on-line case for the European Large Telescope, ELT, at www-astro.physics.ox.ac.uk/$``$imh/ELT, from which some of the following has been taken).
This section focuses on the science case and technical issues for Earth-mass planet detection. For this, OWL must be equipped with advanced adaptive optics systems to compensate for atmospheric seeing and the production of a near-diffraction limited image or image core, expected to deliver Strehl ratios $`S`$ (the peak image brightness relative to that in a near-perfect image) from several to many tens of percent. In the near-infrared (1–5 $`\mu `$m), values near 90% may be attainable, and are needed for planet detection. Ultimate detection capabilities depend sensitively on whether such high Strehl ratios can be delivered in practice (such advanced AO systems do not yet exist). In addition, the telescope design (e.g. number and shape of elements) has an impact on the final image quality, even with co-phasing techniques, so that the exo-planet objectives must be considered early in the project.
Applied to a nearby solar-type star, adaptive optics produces an image with several components. A central ‘spike’ resembling a diffraction-limited image contains some $`S`$ per cent of the total light, less a modest fraction diverted into the other, wider-spread components of the telescope point-spread function. The spike is surrounded by a residual halo containing the remaining light distributed like an unmodified seeing disc, i.e. with a generally Lorentzian distribution. This halo overlies, and within the seeing disk generally dominates, the fainter wings of the telescope point-spread function. These wings combine the light diffused by the small-scale imperfections of the optics and the accumulated dust, as well as the light diffracted by the central obstruction and geometric edges (mirrors, and supporting structures in the beam). The diffraction pattern is strongly dominated by the secondary mirror supporting structure, and by the edges of the mirror segments, which also produce strong secondary peaks.
The diffracted light can locally strongly dominate the uncorrected seeing light. The detailed structure of the halos in adaptive optics-corrected images is still being explored. At large radii they are composed of the rapidly varying diffraction-spot-sized ‘classical’ speckles. Less well-understood ‘super-speckles’ occur closer to the first 2–3 diffraction rings; larger, brighter, less rapidly variable and therefore less likely to average out on a long-exposure image. At the wavelengths of interest they should not, however, occur more than 10 milli-arcsec from the image centre, corresponding to 0.1 AU at a distance of 10 pc. Additionally, techniques such as simultaneous differential imaging may completely cancel these super-speckles. Nevertheless, their noise contribution remains even after subtraction, and is typically the strongest noise source in the 5–15 $`\lambda /D`$ region. This can only be reduced by improving the Strehl ratio of the adaptive optics system.
Detection of a terrestrial-like planet in the presence of the stellar glare is made possible in principle by the relatively large angular separation between the image of a habitable planet and the central diffraction peak of its parent star. Thus at separations of 100 milli-arcsec (1 AU at 10 pc) only the sky background, the wider scattering components of the intrinsic point-spread function, and the adaptive optics halo contribute to the background. The orbital radius at which an exo-planet is in practice detectable is then bounded by two effects. At the inner extreme, the bright inner structures of the stellar image will drastically reduce sensitivity at angular separations below 10–20 milli-arcsec, corresponding to an orbital radius of 1 AU at 50 pc or 5 AU at 250 pc, so that beyond these distances only self-luminous exo-planets (‘young Jupiters’) could be detected. At the outer extremes, the brightness of the starlight reflected by the planet falls off with increasing orbital distance from the parent star, even though it is well separated from it. Some relevant scales are shown in Table 6.
Various techniques to increase the contrast of the images are at different stages of development, from theoretical concepts to working prototypes. The most promising are: (i) classic coronography, which involves masking the star in the focal and pupil planes; advanced Lyot stop studies taking into account the segmented mirror suggest that contrasts of $`10^9`$ are achievable (neglecting diffusion by dust, mirror micro-roughness and the atmosphere); (ii) nulling interferometry (using the coherence of the star light to eliminate it interferometrically), (iii) extreme adaptive optics and multi-conjugated adaptive optics (which result in a higher Strehl ratio and cleaner point-spread function), (iv) simultaneous differential imaging (using the contrast of the target between nearby spectral bands and/or the polarisation of the target). Some of these methods can be combined to reach yet higher contrast (Codona & Angel, 2004).
The wavelength range where a broad range of planets may best be detected and studied with OWL is probably the near infrared J-band at 1.25 $`\mu `$m (where achievable Strehl ratios should approach 90% and where strong spectral features of water are available as diagnostics), and the far red Z and I bands extending down to 700 nm (less favourable for adaptive optics, but containing the critical O<sub>2</sub> B-band absorption complex at 760 nm). In the centres of the stronger absorption bands saturated lines will obscure the signal from an exo-Earth. However, in the wings numerous narrow unsaturated but detectable lines will move in and out of coincidence as the two planets move around their respective suns with a modulated Doppler shift of up to 50 km s<sup>-1</sup>. As discussed in Section 1.2, free oxygen in the exo-planet atmosphere would be a strong indicator of life (i.e. of photosynthetic biochemistry), while its absence would not be conclusive evidence for an abiotic world.
Table 8 indicates the S/N that can be expected from a 24 hr OWL observation of an Earth at 10 pc, in the optical and near-infrared. Clearly, even if Earths exist at distances as close as only 10 pc, detection and spectroscopy will be challenging even with OWL. Concerning the number of target stars accessible for investigation by OWL, the Darwin study identifies some 500 F5–K9 stars out to 25 pc, of which some 285 are single. There are only some 2–5 G0V–G2V stars within 10 pc, and some 21 single, non-variable G0V–G4V within 15 pc.
The distance at which an Earth will be observable will depend on various parameters (Strehl ratio achievable, the performance of nulling/subtraction techniques, e.g. phase-induced apodization coronography could yield attenuation of $`10^9`$, etc.) but in any case is a strong function of mirror diameter. Assuming, as above, that a telescope can see a planet beyond an angular distance of $`5\lambda /D`$ from the parent star, the volume of space explored (i.e. the number of stars) is proportional to $`D^3`$, and the numbers for G stars go from about 25 for a 30 m telescope to about 900 for a 100 m. The time to achieve the same S/N for different size telescopes scales as (S/N)<sup>-2</sup>; the object lies within the uncorrected light from the star, so measurements are in the background-limited regime, i.e. $`SD^2`$, background $`D^2`$, pixel size $`D^2`$ (being diffraction limited) so S/N $`D^2/\sqrt{D^2\times D^2}=D^2`$ and $`tD^4`$. Thus a 30 m aperture would take 123 times longer than a 100 m to observe an object that both would separate from the parent star. This is also the reason why a lower limit of about 70 m diameter is considered useful for spectroscopy of the very nearest exo-solar systems, and why 30 m telescopes do not include Earth-like exo-planets as part of their scientific goals.
The limiting distances at which imaging and spectroscopic observations can be performed must take into account the photon noise from the star, sky and planet, the speckle noise for a realistically high Strehl ratio, and slowly varying aberrations that contaminate the image subtraction. From these distances, the number of host candidates are obtained from star catalogues (e.g., keeping only single G and early K stars). Results are shown in Table 7. The actual number of planets which will be discovered is a function of the (unknown) fraction of planets per star. The conclusions of the OWL studies are that the number of stars accessible to a 100 m telescope is large enough to secure spectroscopic measurements of a large number of planets. In the case of spectroscopy of Earths at 1 AU, however, the number of accessible stars is just large enough to guarantee that a few planets should be observable. These observations will be difficult (but hopefully feasible) for a 100 m telescope, but are out of reach of smaller telescopes. These preliminary conclusions clearly all require further evaluation.
After the mechanical assembly of the OWL telescope is completed, the mirror cell will start to be populated with segments. It is expected that that phase will take a few years ($`3`$), during which the telescope will already be available for scientific observations, although with a reduced collecting area. The configuration of the segments in the cell during this filling phase is still under discussion, but an attractive option is to place them in such way that OWL could be used as a ‘hyper-telescope’, i.e. an interferometer with densified pupil, assuming that proper phasing can be achieved in such a scheme. In such configuration, the resolution and imaging characteristics are very similar to that of a filled aperture telescope with the full diameter (cf. Riaud et al. (2002) and Appendix B), but with a very small accessible field (of the order of $`\lambda /d`$ of a single segment), i.e. $`0.1`$ arcsec, which is very suitable for a planet detection. Science time during the mirror assembly phase could then be used to perform a broad survey for planets around nearby stars. Assuming 100 nights of observations per year for 2–3 years, 1 hr per single observation, and 8–10 epochs per star in order to sample the (potential) planet orbit, of the order of 300 stars could be searched for planets. In that way, a catalogue of targets would be made available for further studies. While the detection limit would not be as deep as that of the full, completed OWL, objects discovered with the early, hypertelescope version would have characteristics (separation and intrinsic magnitude) that would permit further physical studies with the completed telescope.
In conclusion, it is stressed that there remain technical uncertainties in inferring that OWL can indeed observe Earth-type planets: the major assumptions which must be clarified over the coming years are that adaptive optics can work as needed, producing the required Strehl ratios, and providing adequate S/N ratios in both continuum and spectral diagnostic lines. Tracking is not seen as a comparable issue: if the telescope can track to sub-arcsec accuracy (it does according to analysis, at some 0.3 arcsec rms with 10 m s<sup>-1</sup> wind and taking into account friction in drives, and without field stabilisation with M6), then the problem of pointing stability is a control loop issue comparable to adaptive optics but with a much lower frequency.
Even though smaller ELTs, such as the 30-m TMT (Thirty-Metre Telescope), will not reach Earth-like planets, there is still a scientific niche for them between the objectives of VLT-Planet Finder and OWL: notably, detection and characterisation of giant planets, of all ages, and close to their parents stars. Evidently, their technical challenges will be easier to reach than those for a 100-m aperture.
The astrometric capabilities of OWL remain to be assessed, but narrow-angle astrometric prospects should be very good, and provision of an astrometric facility in the instrument plan may be appropriate.
#### 3.1.2 Observations at an Antarctic Site
Excellent ground-based astronomical sites in terms of telescope sensitivity at infrared and submillimeter wavelengths are located on the Antarctic Plateau, where high atmospheric transparency and low sky emission result from the extremely cold and dry air; these benefits are well characterised at the South Pole station. The relative advantages offered by three potentially superior sites, Dome C, Dome F, and Dome A, located higher on the Antarctic Plateau, are quantified by Lawrence (2004). In the near- to mid-infrared, sensitivity gains relative to the South Pole of up to a factor of 10 are predicted at Dome A, and a factor of 2 for Dome C. In the mid- to far-infrared, sensitivity gains relative to the South Pole up to a factor of 100 are predicted for Dome A and 10 for Dome C. These values correspond to even larger gains (up to 3 orders of magnitude) compared to the best mid-latitude sites, such as Mauna Kea and the Chajnantor Plateau.
Thus these Antarctic sites appear to offer extremely good conditions for future interferometric projects and for (robotic) photometric surveys, yielding one long observing ‘night’ per year (completely dark in June–July, decreasing to 7 hr darkness during March and October, although with an extended ‘twilight’ period due to the fact that the Sun does not set very far below the horizon). Table 8 shows that an Antarctic OWL might achieve comparable performances to Darwin/TPF in the near-infrared; currently this is viewed as somewhat hypothetical, given the complex logistics and formidable meteorological conditions for construction and operation.
Dome C (elevation 3250 m, latitude $`75^{}`$) is the site of the French-Italian Concordia station, whose main characteristics have been studied over the last few years. It provides: (i) an ambient temperature ranging from 195 K (winter) to 235 K (summer), resulting in low, stable thermal emission; (ii) extremely dry air (250 $`\mu `$m precipitable water vapour typical), resulting in enlarged and improved infrared transmission windows; (iii) very low surface winds (median wind speed of 2.7 m s<sup>-1</sup>, and below 5 m s<sup>-1</sup> for more than 90% of the time), resulting in very low free-air turbulence, with a quasi-absence of jet streams since Dome C is located inside the polar vortex; (iv) some 80% of clear skies. The combination of coldness and dryness for the atmosphere results in infrared photometric gains that peak at about a factor of 25 in the K and L bands, i.e. an Antarctic 1.8-m telescope is more efficient than an 8-m telescope at a temperate site. In the H and N bands the gain (of order 3) is also notable although less dramatic.
Exceptional winter conditions were confirmed by teams from the University of Nice (Aristidi et al., 2003) and Australia UNSW (Lawrence et al., 2004): over a 3-month period, the median seeing was 0.27 arcsec, with 0.15 arcsec achieved for more than 25% of the time. As most of the turbulence is located near the ground, the isoplanatic angle is enlarged (5.9 arcsec under normalised conditions, compared with 2.9 arcsec for Paranal) which improves the field of view (and the likelihood of finding a reference star) for adaptive optics. This has consequences in the feasibility of adaptive optics for ELTs, where multiconjugate systems and laser guide stars may no longer be needed. Because the turbulence is generated by low-velocity winds, it is slow, meaning better sensitivity and improved phase correction for adaptive optics systems. There is still some debate about the normalised coherence time (whose median value is 2.6 ms on Paranal): indirect measurements with the MASS scintillometer indicate a median value of 5.9 ms, while direct DIMM measurements show a correlation of the image motions beyond 250 ms (as expected from $`r_0=50`$ cm and a typical 2.5 m s<sup>-1</sup> wind speed).
Even taking the most pessimistic value (5.9 ms), the coherence volume that drives the sensitivity of adaptive optics and interferometric systems is improved by a typical factor of 20, if the AO wavefront sensor operates in the K-band. When combined with the factor of 25 photometric gain, this results in an expected sensitivity for interferometers and AO systems 500 times better than a similarly equipped temperate site. For bright sources, higher dynamic range observations (either by coronography or nulling) can be achieved due to the reduced phase and background noise.
Current adaptive optics mainly concerns the correction of the phase of the wavefront to achieve diffraction-limited imaging. Even after a perfectly plane wavefront has been produced, diffraction effects caused by scintillation may dominate in the far wings and halos of the image. Scintillation, originating in inhomogeneities of the higher atmosphere, causes patterns of ‘flying shadows’ on the ground (with intensity contrasts of perhaps only a few percent), carried across the telescope pupil at the windspeed of the originating atmospheric layer. In principle, this could be corrected by second-order adaptive optics (modulating both the phase and amplitude), but the task is not trivial. An additional advantage for Dome C (assuming that ‘ordinary’ adaptive optics will be fully operational) is that such sites also appear to have very much less scintillation: not only do the low windspeeds (no jet streams such as prevail over Chile and Hawaii) imply much less energy deposited in atmospheric turbulence, but the winter-time atmospheric structure is such that the tropopause effectively reaches the ground, with no significant temperature discontinuities in the higher atmosphere.
In the most favourable cases (K band and an L band extended to 2.8–4.2 $`\mu `$m) Dome C thus provides a near space-quality environment, presumably at a small fraction of the cost, and with almost no limitations in weight or volume. The logistics for the station were developed by the French (IPEV) and Italian (PNRA) polar institutes: humans reach the site by plane with a total travel time of about 48 hr from Europe, while the heavy equipment is shipped in standard containers by boat to the coast and then by pack trains of Caterpillar trucks onto the glacier slopes up to Concordia.
It is still the accepted wisdom that a census of exo-Earths and the characterisation of their atmosphere (in search for biomarkers) will require a space mission such as Darwin. The survey part of the Darwin programme could perhaps be undertaken from Dome C, given the extreme phase and background stability of the site. A preliminary study should be undertaken to determine whether this could be achieved, either with a coronographic ELT in the visible or near-IR, or a large interferometer in the thermal infrared. This would enable Darwin to concentrate on the spectroscopy of identified targets, i.e. at wavelengths 6–18 $`\mu `$m where clear biomarkers exist and which for the most part are not accessible from Dome C. A simplified interferometer (two 1 m telescopes) equipped with a GENIE-type instrument could provide good-quality spectra of hot (Jupiter- or Neptune-class) bodies and the characterisation of exo-zodiacal light around main-sequence nearby stars – precursor science for Darwin that otherwise necessitates extensive use of VLT UT time. Other applications of Dome C for exo-planets include astrometry and transit photometry. The ultimate astrometric accuracy of an interferometer was estimated by Swain et al. (2003) to be a factor of some 30 better at Dome C than at Paranal. However, before this potential gain is realised, the technique must have matured to the point where astrometry on temperate sites is limited by the atmosphere only, and not by systematic effects. Transit photometry could benefit from lower scintillation, and longer continuous time coverage (polar night), than on temperate sites.
The nature of the site, halfway between ground and space, makes it an especially interesting area for collaboration between ESO and ESA. Although Europe is naturally in a leadership position due to the presence of the Concordia station, Dome C operation is made in a fully international (Antarctica Treaty), politically interesting (‘continent of science’) context, which may help to gather resources and foster international cooperation (Australia, USA, China) for a large project. A major facility could be proposed at the EU level in 2007, as part of the FP7 programme, and in coincidence with the International Polar Year.
While an interferometer along the lines of GENIE seems very appealing, building OWL at Dome C is a different matter. Exo-planet studies are going to be one of several science drivers for any ELT. Limited sky coverage, and logistics, may be the limiting factors.
Meanwhile, specific concepts being studied include:
(a) a 0.8-m Italian project, featuring a mid-infrared imager (5–25 $`\mu `$m) and possibly a 1–2.5 $`\mu `$m spectrograph. Construction at Dome C has started, and first light is scheduled for December 2005;
(b) a large Australian Dome C development proposal including various groups involved in a 2-m optical/IR telescope (PILOT). Associated structures at Dome C are already advancing, with the first winter-over starting in 2005;
(c) a robotic reflective Schmidt telescope for Dome C by Strassmeier et al. (2004);
(d) a successful bid for NSF study money for Phase 1 of an Antarctic Planetary Interferometer (API), to be placed at Dome C (PI: Mark Swain). Phase 1 is intended to be capable of atmospheric spectroscopy of gas giants, whereas it will take a full Phase 2 system to target Earth-like planets;
(e) a French concept for a Dome C interferometer called KEOPS (Kiloparsec Explorer for Optical Planet Search, PI: Farrokh Vakili). Proposed timescales cover the current study phase (telescopes, beam stabilisation, co-phasing, delay lines, recombiner, coronograph); a single 1.5-m telescope operation in 2007; a simple interferometer with 2 telescopes in 2008–09; first 6-telescope ring system, possibly in association with API, around 2015; and a full 36-telescope system in the more distant future.
### 3.2 Space Observations: 2015–2025
The literature makes no reference to transit missions beyond Kepler and Eddington, nor astrometric missions beyond SIM and Gaia, all due for completion around 2015. Rather, space missions projected for 2015–20 and beyond fall into the category of ‘imaging’ or ‘direct detection’ concepts, notably Darwin and/or TPF. ‘Imaging’ here generally refers to imaging of the exo-planetary system, i.e. direct detection of the exo-planet as a point source of light distinct from that of the parent star, and not to resolved imaging of the exo-planet surface.
The next major break-through in exo-planetary science will be the detection and detailed characterisation of Earth-like planets in habitable zones. The prime goals would be to detect light from Earth-like planets and to perform low-resolution spectroscopy of their atmospheres in order to characterise their physical and chemical properties. The target samples would include about 200 stars in the Solar neighbourhood. Follow-up spectroscopy covering the molecular bands of CO<sub>2</sub>, H<sub>2</sub>O, O<sub>3</sub>, and CH<sub>4</sub> will deepen understanding of Earth-like planets in general, and may lead to the identification of unique biomarkers. The search for life on other planets will enable us to place life as it exists today on Earth in the context of planetary and biological evolution and survival.
In the more distant future, perhaps well beyond 2025, a successful Darwin/TPF would logically be followed by ‘life finders’ and true ‘planet imagers’. At present they appear only as more distant goals, and a brief discussion of them is given in Appendix B. They are unlikely to affect ESA/NASA policy over the next decade or more, at least until the prospects for the success of Darwin/TPF can be quantified, except in the areas of advanced technology studies. Similarly, ideas beyond Darwin/TPF are unlikely to influence the choices or prospects for the very large (50–100 m) telescopes on ground.
#### 3.2.1 Darwin
Darwin is the ESA mission concept aiming at the direct detection of exo-planets, and is focused on an interferometer configuration. It was originally conceived as a set of eight spacecraft at L2 (6 telescopes, one beam combination unit, and one communication unit) that would survey 100 of the closest stars in the infrared, searching for Earth-like planets and analysing their atmospheres for the chemical signature of life (Fridlund, 2000), scientific objectives in common with those of TPF. More recent studies have identified a simplified option, employing 4 telescopes separated by up to 50–100 m operated in a ‘dual-Bracewell’ configuration (or possibly $`3\times 3.5`$ m telescopes), requiring a dual Soyuz-Fregat launch, and a target launch date of 2015. The mission foresees a detection phase of 2 years (allowing the follow-up of 150–200 stars), and a spectroscopy phase of 3 years. Specific precursor efforts include a possible space mission (Smart-3) to demonstrate the concept of formation flying. The emphasis on the infrared rather than the optical is guided by: (i) the less-stringent requirements on technology (contrast and resolution); (ii) infrared spectroscopy allows characterisation in a direct and unambiguous manner; (iii) infrared interferometry allows a larger sample of objects to be surveyed for Earth-like planets (150–200 versus 32 for visual coronography); (iv) it provides the heritage for the next generation of missions which will demand higher resolution spatial imaging. Figure 6 is a simulated image of our Solar System viewed from 10 pc.
Table 8, from Angel (2003), is a summary of detection capabilities for an Earth-Sun system at 10 pc for various experiments being studied, including TPF and Darwin. It shows that even for the most ambitious projects being planned at present, direct detection of even a nearby Earth represents a huge challenge.
Table 9 summarises the estimated integration times for a variety of stellar types (target stellar types are F–G with some K–M), and a range of distances. Broadly, these are consistent with the simplified summary given in Table 8. Typical distance limits are out to 25 pc. A terrestrial planet is considered as $`R_{\mathrm{max}}2R_{}`$. The spectral range is 6–18 $`\mu `$m, covering the key absorption lines, with a spectral resolution of 20 (50 is required for CH<sub>4</sub> in low abundance). Integration time estimates are given for a S/N = 5 in imaging (i.e. detected planetary signal/total noise) and S/N = 7 in the faintest part of spectrum for spectroscopy, for the following assumptions: total collecting area for all telescopes of 40 m<sup>2</sup>; telescopes at 40 K; effective planetary temperature 265 K (signal scales as $`T_{\mathrm{eff}}^4`$); $`R=R_{}`$; integrated over 8–16 $`\mu `$m; 10 times the level of zodiacal dust than in the inner solar system; all of the habitable zone is searched; Spitzer Si-As detectors. Times given are based on a 90% confidence level for a non-detection based on 3 observations (a positive detection at S/N = 5 then takes one third of the times given). To detect the Earth at 265 K (instead of 290 K) around the Sun at 20 pc would take 9 hr at S/N = 5, and 36 hr at S/N = 10. For a K5V star at 10 pc these times are about 1–4 hr respectively. For spectroscopy, the S/N varies between 7 and significantly higher, depending on the atmosphere. For an Earth atmosphere (but at 265 K) at 20 pc, an integration time of 54 hr provides a S/N$`715`$.
One of the strongest noise sources is the leakage of photons from the resolved stellar surface. This is a function of both stellar temperature and diameter (distance, spectral type). It is most clearly seen for nearby stars where the detection time drops strongly. For 20 pc detection times do not change much since the star does not leak very much for any diameter. Detection times rise rapidly for K5V stars at 30 pc, because of the 90% confidence requirement, which itself demands more changes in array size because the habitable zone is very close to the star.
The number of candidates accessible and visible to Darwin (two $`\pm 45^{}`$ caps near the ecliptic poles are inaccessible) is estimated as follows. Considering only single stars, there are 211 K and 82 G candidates out to 25 pc, with another 30 F-type stars plus many M-dwarfs. Combined with the integration time estimates, Darwin should survey more than 150 stars in 2 years. Within 5 years, all 293 single and accessible solar-type stars stars out to 25 pc can be surveyed, with a spectrum obtained for each system in the case of a planet prevalence of $``$10%. Up to 1000 systems in the solar neighbourhood could be surveyed if the planetary fraction is even smaller.
#### 3.2.2 The Darwin Ground-Based Precursor: GENIE
GENIE is a nulling interferometer under development by ESA and ESO for the VLTI. GENIE will allow the development and demonstration of the technology required for nulling interferometry, allowing testing of the Darwin technology in an integrated and operational system (amplitude control loops, high-accuracy optical path difference control loop, dispersion control, polarisation compensation, background subtraction, and internal modulation). GENIE is considered by the Darwin project as necessary for demonstrating these technical concepts in advance of launch.
Two competitive instrument definition studies were due for completion by the end of 2004, with the scientific case being prepared in collaboration between ESO and ESA. GENIE will be considered by the ESO Council in April 2005, and could be operational by mid-2008. If not accepted by ESO, a separate laboratory technology demonstrator would be needed to validate the Darwin concepts.
GENIE is also considered mandatory as a specific instrumental configuration to survey southern-hemisphere target candidates, in order to decide which targets are most suitable for observation by Darwin, and specifically to characterise the level of exo-zodiacal light, which must be below certain limits necessary for exo-planetary detections. A corresponding programme is planned to be undertaken at the Keck telescope in the northern hemisphere for preparations for TPF. Studies by den Hartog et al. (2004) show that the capabilities of GENIE are sufficient to detect the planet around $`\tau `$ Boo within 1 hour, and by inference some 5 other candidate ‘hot Jupiter’ planets. GENIE could use either the VLT UTs or ATs, at a wavelength of 3.6 $`\mu `$m, and will require a significant number of observing nights (of order 50).
The main limiting factors for GENIE are the phase and thermal background stability on Paranal, as well as system complexity issues involved with its integration into the existing VLTI environment. If GENIE were located on the high Antarctic plateau (Dome C) where the thermal background is lower, and the seeing both better and slower (see Section 3.1.2), the performance required to perform its science programme could perhaps be achieved using smaller (of order 1 m) telescopes and a simpler, dedicated system architecture. This has to be balanced against the reduced sky coverage at $`75^{}`$ latitude, and probably more complex logistics.
#### 3.2.3 Terrestrial Planet Finder (TPF)
NASA’s TPF roughly parallels the ESA Darwin study, with close discussions taking place between the two teams. TPF was conceived to take the form of either a coronograph operating at visible wavelengths or a large-baseline interferometer operating in the infrared (Beichman, 2003). There are two aspects of this choice which should be distinguished: (a) the scientific aspect: whether reflected (visible and near IR) light or thermal emission (mid-IR) is the best regime to characterise planets (albedo, temperature, colour, key species that can be identified e.g. CO<sub>2</sub> etc.; see Schneider (2003) for a recent discussion); (b) the instrumental aspects: whether an interferometer or a coronograph is the best. Here, the NASA Technology Plan for TPF stated that ‘Technology readiness, rather than a scientific preference for any wavelength region, will probably be the determining factor in the selection of a final architecture’. In May 2002, two architectural concepts were selected for further evaluation: an infrared interferometer (multiple small telescopes on a fixed structure or on separated spacecraft flying in precision formation and utilising nulling), and a visible light coronograph (utilising a large optical telescope, with a mirror three to four times bigger and at least 10 times more precise in wave-front error than the Hubble Space Telescope).
In April 2004, NASA announced that it would embark on a $`6\times 3.5`$ m<sup>2</sup> (more recently changed to $`8\times 3.5`$ m<sup>2</sup>) visual coronograph in 2014 (TPF-C), with a wavelength range 0.6–1.06 $`\mu `$m, and targeting a full search of 32 nearby stars and an incomplete search for 130 stars (more recently quoted as 80). A free-flying interferometer, in collaboration with ESA, would be considered before 2020 (TPF-I).
A visible light system can be smaller (some 10 m aperture) than a comparable infrared interferometer, however advances in mirror technology are required: mirrors must be ultra-smooth ($`\lambda /\mathrm{15\hspace{0.17em}000}`$; a number stated in the TPF documentation, although values an order of magnitude inferior appear more plausible) to minimise scattered light, and in addition active optics would be needed to maintain low and mid-spatial frequency mirror structure at acceptable levels. Infrared interferometry would require either large boom technology or formation flying, typically with separation accuracies at the cm-level with short internal delay lines. For the detection of ozone at distances of 15 pc and S/N$``$25, apertures of about 40 m<sup>2</sup>, and observing times of 2–8 weeks per object, are indicated.
Many ideas for scientific and technological precursors for TPF have been examined. The many possible solutions involve combinations of adaptive wavefront correction, coronographs, apodization, interferometers, and large free-flying occulters. The main contenders are summarised in Appendix A for completeness, although with the recent (April 2004) NASA announcement on TPF strategy, it is not clear whether any of these concepts will be developed further.
### 3.3 ESA Themes: 2015–2025
In mid-2004, ESA solicited a call for ideas for its scientific programme beyond 2015, resulting in several outline proposals in the area of exo-planets:
A Large UV Telescope (Lecavelier): the proposal considers a large UV telescope as a promising way to conduct a deep search for bio-markers in Earth-mass exo-planets, through the detection of relevant atmospheric signatures (such as ozone, water, CO, and CO<sub>2</sub>) in a very large number of exo-planets, and planetary satellites, up to several hundred parsec distant. It notes that the detection of atmospheric signatures in HD 209458b has been made through space UV observations, and stresses that large numbers of prime targets for these observations will be detected by Gaia. It underlines the possibility of time-resolved spectroscopy (over the 10-min typical transit time of an Earth-mass planet) providing spatial information of atmospheric constituents along the planet surface (poles versus equator, presence of continents). A JWST-size telescope is proposed, and some quantitative expectations are given.
Search for Planets and Life in the Universe (Léger): the proposal makes the general case for the continued development of an ESA strategy for the improved detection and characterisation of exo-planets, primarily through the techniques of transits and direct detection. This proposal was submitted on behalf of more than 200 individuals and institutions.
Astrometric detection of Earth-Mass Planets (Perryman): the proposal points out that, beyond the micro-arcsec astrometric accuracies of ESA’s Gaia mission, 10 nano-arcsec accuracies would permit the detection of astrometric perturbations due to Earth-mass planets around Sun-like stars at 100 pc. If a survey-type mission were feasible, the concept could lead to the systematic survey of many hundreds of thousands of stars for Earth-mass planets – important for the generation of target objects if the fraction of Earth-mass planets turns out to be very small. Earth-mass perturbations around a solar-mass star are 300 nano-arcsec at 10 pc, or 30 nano-arcsec at 100 pc, the latter requiring an instantaneous measurement accuracy a factor 3 better, i.e. 10 nano-arcsec at, say, 12 mag. This is a factor of some 1000 improvement with respect to Gaia. Keeping all other mission parameters (efficiency, transverse field of view, mission duration, total observing time per star, and image pixel sampling, etc.) unchanged, we can consider reaching this accuracy simply through a scaling up of the primary mirror size. The Gaia primary mirror has an along-scan dimension $`D=1.4`$ m and a transverse dimension $`H=0.5`$ m; the final accuracy scales as $`\sigma D^{3/2}H^{1/2}`$. These desired accuracies would therefore require a primary mirror size of order $`50\times 12`$ m<sup>2</sup>, and a focal length (scaling with $`D`$) of about 1600 m, similar to the scale of the optics derived for the mini-version of Life Finder. Accuracy levels of $`10`$ nano-arcsec are still above the noise floors due to interplanetary and interstellar scintillation in the optical, or stochastic gravitational wave noise. Astrometric accuracy limits due to surface granular structure and star spots were discussed in Section 1.3. To reach the astrometric precision of 300 nano-arcsec at 10 pc means that the photocentre of a star must be determined to $`5\times 10^5`$ m, or $`1.5\times 10^4R_{}`$. This is still significantly above the photometric stability limits of $`10^710^8`$ AU, or $`10^310^4`$ m, derived by Svensson & Ludwig (2005) for stars with $`\mathrm{log}g=4.4`$. Note, however, that this only concerns thermally-driven granulation, thus determining only the lowest possible level of stellar astrometric stability.
Understanding the Planetary Population of our Galaxy (Piotto): the proposal underlines the importance of the field to ESA and to Europe. It argues that most disciplines start with the discovery and study of individual objects, before moving on to the more mature stage in which large, unbiased samples of objects are studied. In the decade 2015–25, an open issue will be the discovery and characterisation of large unbiased samples of planets down to habitable systems. This should provide information necessary to understand how the environment and nature of the parent star affects the resulting planetary population; how stellar metallicity, multiplicity, crowding, and population influence the nature, habitability and survival of planetary systems. The proposal suggests a transit-type follow-up to Kepler/Eddington, using a significantly larger aperture and improved detector technologies to obtain much improved samples/statistics; combined with the availability of large ($`>25`$ m) ground-based telescopes for follow-up spectroscopic observations.
Exo-planet Detection and Characterisation (Surdej): the proposal underlines the general importance of the field to ESA and to Europe, and proposes to intensify efforts towards a Darwin-like mission, including further emphasis on coronography, perhaps as an ESA instrument on TPF-C. The proposal also underlines the importance to astrobiology of sample-return missions in the Solar System, for example to Mars, Europa, and Titan. Otherwise, no specific instrument approach or design is considered.
Evolution of Atmospheres and Ionospheres of Planets and Exo-Planets (André): the proposal considers a combination of in situ observations of Solar System planets and remote sensing of exo-planets in order to advance understanding of the long-term evolution of the atmospheres and ionospheres of planets (and large moons), and to identify the important sources and sinks of atmospheres and ionospheres of planets during different parts of their evolution. No specific instrument approach or design is considered.
Planetary Habitability in the Solar System and Beyond (Bertaux): the proposal underlines the general importance of the field to ESA and to Europe. It includes suggestions for intensifying the search for (past) life on Mars through robotic exploration in order to refine probability estimates of life beyond the Solar System. It also notes the potential problem that ESA’s Darwin mission is presently designed as its own candidate surveyor, and argues that a preliminary programme, possibly ground-based, should undertake the advanced search so that Darwin can focus its observing time on spectral acquisition of Earth-mass planets. The suggestion is to conduct this survey, under ESA responsibility, using Doppler measurements, reaching the required accuracies of 0.1–0.8 m s<sup>-1</sup> through the accumulation of large number of individual 1–2 m s<sup>-1</sup> measurements. The feasibility of this proposal is addressed in Section 1.3.
The Hypertelescope Path (Labeyrie): Appendix B provides an introduction to the concepts and goals of the ‘densified pupil multi-aperture imaging interferometer’ or ‘hypertelescope’. The proposal refers to the fact that space testing of versions as small as 20 cm mirror diameter, with mass below 0.5 kg, could be considered. Progressive versions could be used to study stellar surface resolution (20 cm apertures spanning 100 m in geostationary orbit); detection of exo-Earths at visible and infrared wavelengths (spanning a few hundred metres at L2); much larger versions will be needed to resolve surface features of exo-Earths.
Lamarck – an International Space Interferometer for Exo-Life Studies (Schneider): the proposal stresses the importance of the optical (rather than infrared) domain for space interferometry, and outlines a concept for a temporaily ‘evolving’ interferometric station, with the participation of other countries like China, Japan, India and Russia, employing an initial collecting area of around 40 m<sup>2</sup>, baselines above 3 km, a spectral range of 0.3–3 $`\mu `$m and $`R=100`$. The mission strategy would be to detect Earth-like planets with baselines up to 1 km, imaging of the most promising candidates with very long baselines, then interferometer upgrades with subsequently-launched free flyers.
The resulting recommendations of the ESA Astronomy Working Group are contained in ASTRO(2004)18 of 19 Oct 2004. The 47 responses were assigned to three themes: (1) Other worlds and life in the Universe; (2) the early Universe; (3) the evolving violent Universe. The relevant part of the document is reproduced here verbatim:
1.1 From exo-planets to biomarkers
After the first discovery of an extra-solar planet in 1995, there has been steady progress towards detecting planets with ever smaller masses, and towards the development of a broader suite of techniques to characterise their properties. There is no doubt that this trend will continue into the next two decades, as substantial technological challenges are progressively overcome. After Corot will have opened the way to telluric planet finding, the Eddington mission would get a first census of the frequency of Earth-like planets. Gaia will deliver important insights into the frequency of giant planets; the existence and location of such planets is crucial for the possible existence of Earth-like planets in the habitable zone. Gaia will also further improve our understanding of the stellar and Galactic constraints on planet formation and existence.
The next major break-through in exo-planetary science will be the detection and detailed characterisation of Earth-like planets in habitable zones. The prime goals would be to detect light from Earth-like planets and to perform low-resolution spectroscopy of their atmospheres in order to characterise their physical and chemical properties. The target sample would include about 200 stars in the Solar neighbourhood. Follow-up spectroscopy covering the molecular bands of CO<sub>2</sub>, H<sub>2</sub>O, O<sub>3</sub>, and CH<sub>4</sub> will deepen our understanding of Earth-like planets in general, and may lead to the identification of unique biomarkers. The search for life on other planets will enable us to place life as it exists today on Earth in the context of planetary and biological evolution and survival. Being aware of the technical challenge to overcome the high brightness ratio between star and planet and the advancements in technology reached during the past few years by ESA studies, the AWG strongly and unanimously recommends the implementation of a mission addressing these objectives through a nulling interferometer for the wavelength range between 6–20 $`\mu `$m. Such a mission should be implemented around 2015, making Europe highly competitive in the field.
The next step in this well-defined roadmap would be a complete census of all terrestrial extra-solar planets within 100 pc, for example through the use of high-precision astrometry. Longer-term goals will include the direct detection and high-resolution spectroscopy of these planets with large telescopes in IR, optical and UV wavelengths and finally the spatially resolved imaging of exo-Earths, leading to the new field of comparative exo-planetology.
The conclusions of the document state: For the very first part of the 2015-2025 decade, two observatory-class missions are recommended: (1) a mid-infrared nulling interferometer for the detection and characterisation of Earth-like planets in the Solar neighbourhood. In order to ensure a timely implementation of such a mission, the AWG gives high priority to the continuation of the present fruitful technology program for this project, and strongly and unanimously recommends that all steps be taken to ensure that such a mission can be flown as early as possible in the 2015 time frame; (2) a far-infrared observatory, etc…
The conclusions also comment: The AWG reiterates its unanimous support to a rapid implementation of the Eddington mission, which would constitute an important intermediate step towards the scientific goals formulated in theme 1.
### 3.4 Other Concepts and Future Plans
Although it is of course impossible to look forward a number of years and to predict the status then, more or less plausible novel (i.e., as yet unproven) methods might conceivably be applied for exo-planet studies.
The use of ‘new’ physical principles for the imaging of exo-planets may develop. Not long ago, it was realised that light (photons) may carry orbital angular momentum (in addition to the ordinary photon spin associated with circular polarization). Interference of such light may produce a dark spot on the optical axis, but high intensities outside (but still inside the ‘classical’ diffraction spot). By suitable manipulation of starlight, different amounts of angular momentum might be induced to light from the central star, relative to its nearby planet, thus enhancing the observable contrast by several orders of magnitude. For an introduction to light’s orbital angular momentum, see Padgett et al. (2004); for a discussion of high-contrast imaging using such methods, see Swartzlander (2001); for reference in the context of exo-planet studies, see Harwit (2003).
### 3.5 Summary of Prospects: 2015–2025
The major space and ground-based experiments so far considered for the 2015–25 time-frame do not include survey missions beyond Kepler/Eddington and Gaia, but focus on the detection and characterisation of a few low-mass planets in the Solar neighbourhood.
The detection prospects are summarised in Table 8, compiled by Angel (2003), and include Darwin/TPF and OWL. An assessment of the S/N of planet detections with OWL has been made independently by Hainaut & Gilmozzi (ESO), and is in broad agreement with those in the table. The Darwin S/N assessments for both imaging and spectroscopy, detailed in Table 9, are also broadly in agreement with this summary table.
Accepting the (unproven) hypothesis that there are Earth-mass planets around solar-type stars within 10–20 pc, their detection and characterisation even with Darwin/TPF and/or OWL will be challenging. At this time it would seem appropriate to continue the search with both ground and space techniques, at least until their respective technical limitations and costs are better understood.
## 4 The Role of ESO and ESA Facilities
### 4.1 The Expected Direction of Research
As detailed in the previous sections, plans for the forthcoming decade are focused on surveys that attempt to detect large numbers of objects by various methods (transits, astrometry, microlensing). The aim is to obtain information on thousands of systems, and thus to characterise the population frequency of extra-solar planets according to mass, orbital radius, etc., essential for refining theories of planet formation and evolution.
Beyond 2015, programmes targeting further characterisation of the large-scale statistical distribution of planets do not (yet) exist, probably since the requirements for large-scale surveys of other aspects of planetary parameter space have not yet been carefully considered. Instead, post-2015 projects currently address the existence and characterisation of Earth-mass planets around a number of carefully selected candidate stars, both from ground and space, and questions related to habitability and the existence of life. Both ground and space approaches will be highly challenging and expensive, and their technical feasibility is still being evaluated.
A synthesis of prospects is presented in Figure 7, which tabulates the space (second column) and ground (fourth column) techniques, the resulting candidate detections (third column) and statistical knowledge (first column), and possibilities for follow-up observations.
In this section the following questions are addressed:
(a) what follow-up observations and facilities are required to characterise these systems more completely (Section 4.2);
(b) what does the resulting (statistical) knowledge of exo-planet distributions imply for the targeted observations of Darwin and OWL (Section 4.3);
(c) what information will be available, or should be anticipated, for a deeper astrophysical characterisation of the host stars of planetary systems (Section 4.4);
(d) what is the potential overlap amongst the major facilities currently planned or studied by ESO and ESA (Section 4.5);
(e) are there specific long-lead time space or ground facilities which should be considered to fill observational gaps anticipated over the next 10–20 years (Section 4.6);
(f) are there other considerations that ESO/ESA should investigate for proper interpretation of the data which will be generated by these two European organisations, or others, and which might limit the development of the field unless suitably coordinated (Section 4.7).
### 4.2 Follow-Up Observations
With reference to Figure 7, over the next 5–10 years ongoing or approved survey experiments are expected to generate:
(a) high-mass ($`M_\mathrm{J}`$) candidates: some hundreds from COROT, Kepler and Eddington; thousands with Gaia astrometry and thousands of transiting systems with Gaia photometry; some hundreds from ongoing and future ground-based radial velocity surveys; and hundreds (possibly thousands) of hot Jupiters from ground-based transit surveys;
(b) low-mass ($`13M_{}`$) candidates: a small number of hot terrestrial planets from COROT around 2008; with tens to hundreds from Kepler expected to be available to the community after about 2010.
In principle, the information required for further characterisation of a detected planetary system is independent of the star or planet mass: as discussed elsewhere in this report: (a) radial velocity measurements of transit detections to eliminate false alarms due to grazing eclipsing binaries, triple stars, star spots, and false positives; (b) transit spectroscopy from ground or space for the determination of atmospheric properties; (c) the combination of transit or astrometry data with radial velocity information allowing the determination of the true mass of the planet; (d) photometric or spectroscopic information needed to characterise the parent star; (e) follow-up imaging of transiting candidates with high spatial resolution adaptive optics to minimise the possibility that the object is actually multiple, or that a foreground or background binary system is causing the dimming of the light; (f) all the above will provide candidates for ground-based imaging by VLT, VLTI, OWL, etc.
For microlensing candidates, no follow-up observations are generally possible. Nevertheless, due to the relative proper motion, the lens star will increase its angular separation to the source star and become visible after some time. One case is known: MACHO LMC–5, in which the lens star was imaged after about 8 years or so (Alcock et al., 2001). So in principle, there exists a possibility to study the host star of a microlensing planet, depending on the mass/apparent brightness of the lensing (i.e., host) star, and on the relative transverse velocity.
In practice, the problem is distinct for high-mass or low-mass planets.
#### 4.2.1 High-Mass Planets
For high-mass objects (of order $`1M_\mathrm{J}`$), ground-based follow-up transit measurements will generally be technically feasible with the current generation of instruments. COROT, Kepler, Eddington, Gaia, OGLE, possibly HST, and the ground-based transit networks will supply thousands of targets which can be followed from the ground. Small telescopes, preferably robotic, and substantial amounts of observing time will be required. Existing transit measurement facilities and amateur networks could contribute, although some new dedicated provision for follow-up is probably needed, in particular enlarging networks in longitude to significantly improve the efficiency of ground-based transit observations.
Radial velocity follow-up will be needed for all high-mass candidates, of which Gaia will generate a very large number. Thousands of the more massive planets could be observed by ground-based radial velocity instruments, assisting the determination of multiple planets, relative orbital inclinations, etc. ESO could consider a coordinated large-scale follow-up of such radial velocity observations, and whether additional facilities will be needed. An assessment of telescope time/aperture needed for these projects has not been made — probably a combination of existing facilities and larger dedicated instruments would be needed.
#### 4.2.2 Low-Mass Planets
Low-mass (of order $`1M_{}`$) planets will hopefully be detected by Kepler and Eddington, perhaps in moderately large numbers (some hundreds).
They will typically be too low-amplitude for ground-based photometric follow-up, which in any case are unlikely to improve on the S/N of weak transit events detected by Kepler or Eddington – these missions will have years of lightcurves on a star, and if the signal is still weak after phasing and adding all data, it will be difficult to improve from the ground in a reasonable time. This is a potential problem, since such follow-up observations will be needed: (i) to confirm the reality of the lower S/N detections; (ii) to search for planetary periods for candidates for which only one transit is detected; (iii) to confirm candidate detections for which two or possibly more transits were detected (since periods and transit times are known, ground-based follow-up may be more feasible); (iv) to search for period changes due to planetary moons etc.; (v) to characterise the transit systems in terms of chemical abundances. Probably the only prospect is follow up with HST and/or JWST (see Section 2.2.1) for transits, and possibly SIM for astrometry, although many of the transit candidates will be too distant even for these instruments.
Radial velocity measurements are again needed in principle to supplement the orbital information. Improvements in radial velocity precision for transits may be achieved by the stacking of repeated observations at the known planet period, as described in Section 1.3. If Eddington is approved (or for the study of Kepler candidates), the development of new telescope facilities should be considered, such as a high-precision spectrograph (like HARPS) based on an 8-m (or larger) telescope. Given the low expected surface density of accessible candidates, it is unlikely that a multi-object instrument would be effective and so a highly-optimised single object instrument offering a precision of $`1`$ m s<sup>-1</sup> would be preferred. The HARPS detection of a 14 $`M_{}`$ planet demonstrates that it may be possible to characterise all the exo-planets detected by COROT (massive Earths with short periods) in this way.
#### 4.2.3 Summary of Follow-Up Facilities Required
To summarise the requirements for these follow-up programmes, the following facilities will be needed:
(a) high-precision radial velocity instrumentation for the follow-up of astrometric and transit detections, to ensure the detection of a planet by a second independent method, and to determine its true mass. For Jupiter-mass planets, existing instrumentation may be technically adequate but observing time may be inadequate; for Earth-mass candidates, special-purpose instrumentation (like HARPS) on a large telescope, would be required. Requirements for automated telescopes for precise radial-velocity measurements have been discussed by Pepe & Mayor (2004);
(b) photometric monitoring of large numbers of high-mass planet candidates coming from space experiments: a combination of existing and new facilities will be needed;
(c) very high-resolution (adaptive optics) imaging of transit candidates to reject the ‘false-positives’ generated by confusing sources;
(d) photometric and spectroscopic characterisation of all candidate planetary systems, providing fundamental stellar parameters such as temperature, gravity and metallicity. A combination of planned (Gaia) and new (spectroscopic survey) instrumentation will probably be required;
(e) follow-up of candidates from microlensing studies will be necessary to identify host-star properties that maximise the probability of finding a planetary system.
A first (and incomplete) assessment of necessary telescope time can be made as follows:
(a) for the photometric monitoring: assuming a telescope optimized for fast photometric measurements (i.e. with low overhead), up to 6 measurements per hour can be expected, i.e. 15 000 measurements per year. A dedicated instrument, following the concept of the ‘Gamma-Ray burst Optical/Near infrared Detector’ (GROND, http://www.mpe.mpg.de/$``$jcg/GROND/), permitting simultaneous measurements in up to 7 photometric bands, could be duplicated for a 4-m and for a 1–2-m telescope in order to optimize the aperture of the telescope with the brightness of the target. The photometric monitoring would then be performed in about 1 year (full time) assuming two telescopes.
(b) for the radial velocity measurements: assuming an instrument like HARPS after optimization for fast observations, 4–5 stars could be measured each hour (i.e. about 5 minutes preset and overhead, and 5–10 minutes exposure), thus 40–50 measurements per night, or up to 15 000 measurements per year, assuming the instrument is operated full time. This follow up will then require this telescope for a few years in order to measure each star at least 3–4 times at different orbital phases.
### 4.3 Statistics of Exo-Planets: Implications for Darwin/OWL
As noted under Appendix A, the McKee-Taylor Decadal Survey Committee (McKee & Taylor, 2000) qualified its endorsement of the TPF mission with the condition that the abundance of Earth-size planets be determined prior to the start of the TPF mission. A similar strategy will be desirable for Darwin.
The various missions and experiments discussed throughout this report will contribute to detailed statistical information of planetary distributions over different mass ranges, but cannot provide complete information on the occurrence of Earth-mass planets in the Solar neighbourhood (say, to 15–20 pc) in advance of the planned Darwin launch: transit measurements are highly incomplete, the Gaia lower mass limit is above 10 $`M_{}`$, and the SIM survey will presumably also be incomplete down to the levels of 1–2 $`M_{}`$. Gaia and SIM may assist in identifying nearby stars accompanied by Jupiter-type planets (i.e., Jupiter mass in a Jupiter orbit), which may be a representative pre-requisite for the development of life. Nano-arcsec astrometry could provide the necessary information in principle, but practical implementation lies some years in the future.
Nevertheless the SIM and COROT/Kepler/Eddington results should clarify whether Earth-mass planets are common or not. If they are not detected, and therefore not common, Darwin’s task in identifying Earth-mass planets within 20–30 pc will be even more challenging. Then, either additional identification strategies are required (for example, the development of an Antarctic facility dedicated to the purpose, such as a coronographic ELT in the visible/near infrared, or a large interferometer in the thermal infrared), or it is accepted that a significant fraction of Darwin observations is devoted to a search for its own spectroscopic candidates. Thus Darwin could be launched relying on a statistical estimate of the number, and size, of the terrestrial planets that are expected around the stellar target list, combined with the GENIE results on levels of exo-zodiacal emission. If statistics indicate that the number of Earth-mass planets accessible to Darwin is of the order of several tens, this strategy would be acceptable. Conversely, it could also be argued that a null result from a 200-star survey would be a valuable scientific result in its own right.
Current understanding suggests that metallicity is a crucial factor influencing the formation of planets. Higher metallicity therefore seems to be the most promising stellar characteristic that would indicate whether a star is likely to harbour planets. The recently published Geneva-Copenhagen survey (Nordström et al., 2004), mostly aimed at stellar kinematics, has also provided valuable information on metallicities for over 14 000 F and G type dwarf stars using Stromgren photometry. A more detailed measurement of stellar parameters, including metallicities, requires medium- to high-resolution spectra over a large wavelength range. For extra-solar planet research an accurate knowledge of the properties of the host star is essential, even more so when looking for the weak signatures of the planetary atmosphere. A full spectral characterisation of stars in the solar neighbourhood would be a valuable step towards a target list for major projects such as TPF, Darwin or OWL.
### 4.4 Astrophysical Characterisation of Host Stars
Several thousand planetary systems should be discovered by a combination of COROT, Kepler, Eddington, SIM and Gaia over the years 2006–2015. There will be a need for detailed astrophysical characterisation of large numbers of stars (some tens of thousands), in both the northern and southern hemispheres. Specifically:
(a) physical characteristics of the host star via photometry: it will be important to characterise, homogeneously, luminosity ($`T_{\mathrm{eff}}`$, $`\mathrm{log}g`$), metallicity, and micro-variability. Large-scale multi-colour, multi-epoch photometry for this purpose may be available from Pan-STARRS (Panoramic Survey Telescope and Rapid Response System) or LSST (Large Synoptic Survey Telescope) and from Gaia’s 11-band multi-colour photometry, although not before 2012;
(b) physical characteristics of the host star via spectroscopy: the goal is again to determine the spectroscopic parameters of the central star: $`T_{\mathrm{eff}}`$, $`\mathrm{log}g`$, \[Fe/H\]. Medium-high spectral resolving power is needed (20 000–60 000) in the visible range;
(c) spatial distribution and kinematics: Gaia will provide highly-accurate distances essential for luminosity calibration (e.g., 1% at 1 kpc at 15 mag), characterised as a function of the local environment (e.g., for stars within the core or halo of open clusters), and kinematic motions (e.g., with respect to the Local Standard of Rest, velocity with respect to the Galactic mid-plane, etc.). This characterisation is important because of possible links with habitability (as discussed in Section 1.2).
(d) kinematic radial velocities for fainter stars. Hipparcos would have greatly benefited from a coordinated acquisition of radial velocities from the ground. Key programmes went part way towards this, but only for half the sky, for a subset of spectral types, and on a schedule out of phase with the Hipparcos Catalogue publication in 1997—the Geneva-Copenhagen survey of 14 000 F and G dwarfs has just been published (Nordström et al., 2004). Gaia partly addresses the problem by acquiring radial velocities on-board, but higher precision and a fainter limiting magnitude would demand a concerted and coordinated campaign on the ground.
With the caveats on the time-scales of data availability, planned surveys (Gaia, if not others) should generally provide detailed astrophysical characterisation of the host stars of all detected systems, with the possible exception of kinematic radial velocities for faint stars ($`>20`$ mag), and a dedicated spectral survey.
#### 4.4.1 A Dedicated Spectral Survey
ESO has capabilities to perform a full spectral characterisation of all F–K type stars in the southern hemisphere within a radius of say 50 pc. Facilities at ESO could be used to obtain a complete spectrum of the stars from the atmospheric cut-off to 5 $`\mu `$m. By combining the Echelle spectrographs FEROS (2.2-m), UVES, and CRIRES the complete wavelength range can be observed at $`R\mathrm{50\hspace{0.17em}000}`$. A cursory study shows that such a project could be undertaken at the level of a (very) large programme, of the order of 100 nights in total. Selecting stars within 50 pc (Hipparcos distances) and at declination south of 30 degrees North, a total of about 2500 stars would have to be observed. 75% of these have V magnitudes between 5 and 9 making them ideal targets for FEROS, while the remaining fainter ones would be observed with UVES. Using the ESO exposure time estimators for the respective instruments we found that a S/N of about 100 will be achieved within minutes. Using about 110 hr of time on FEROS and about 40 hr on UVES per period to obtain the spectra from 350–980 nm the optical survey could be completed within two years with existing facilities, assuming an ESO ‘Large Programme’ using only late twilight and periods of bright moon and/or poor seeing. CRIRES is being assembled at this point, therefore no estimate of the time required for the infrared observation is given here. Because the majority of stars are rather bright a very substantial fraction (30–50%) of the stars can be observed during twilight while the remainder can be observed during bright time. It is evident that a high degree of homogeneity of spectral data needs to be achieved across the full set of stars in order to achieve the scientific goals of the project. In particular the combination of spectra calls for excellent calibration across instruments. To this end the ECF’s Instrument Physical Modelling Group which is involved in the calibration of instruments on HST and at ESO is prepared to make major contributions thereby ensuring maximum quality of the data and the resulting high level products.
A comparable ‘Nearby Stars’ project (PI: R.E. Luck) is being conducted in the northern hemisphere (http://astrwww.cwru.edu/adam/) using the McDonald 2.1-m Struve reflector and the Sandiford Echelle Spectrograph.
A dedicated survey would provide data useful for other applications in astrophysics. Results would include a very fine spectral library which, combined with e.g. Gaia information on the luminosity function and kinematics, would be very powerful.
### 4.5 Potential Overlap and Competition
Eddington and Kepler
The arguments for undertaking the Eddington mission are unchanged from the time that the ESA Astronomy Working Group placed it at the top of the list of astronomy projects to be undertaken if financing can be made available. Essentially, Kepler is designed to study a single low-latitude target field and thus a single population of stars, in a limited mass range. In the same way that the star formation history of the Galaxy cannot be ascertained by observing a single field, the acceptance of Kepler results as being of general applicability to the complete Galaxy is questionable. This is exemplified by the absence of planets in 47 Tuc, and is a question that can only be resolved by further observations.
Without any further mission dedicated to terrestrial planet searches, there will be no real understanding of the distribution of such planets in general (e.g., in other parts of the Galaxy or in clusters), and the number of targets discovered by Kepler may in any case be very small. Eddington would observe stars in a variety of environments, i.e., field stars and stars in open clusters, and will thus consolidate the search for planets across different stellar parameters (including metallicity, age, mass, binary status, density of surrounding stellar population, etc.).
More generally, experiments need repeatability: single experiment results always provoke more questions, and if the results are controversial it is only reasonable that they are checked independently. Having just one mission aimed at the detection of terrestrial planets appears risky and inadequate, in particular considering the effort being made for future missions that aim at studying these planets (Darwin/TPF). The present working group emphasises the value of aiming to launch such a mission at the earliest opportunity, preferably before 2010. Should it turn out that it is only possible to launch Eddington significantly later, ESA should consider a larger-scale post-Kepler mission on a longer time schedule.
OWL and Kepler/Eddington
The OWL science case is currently not so much based on finding Earth-like planets, but rather on imaging or spectroscopy of nearby candidates detected by other means. To this extent OWL’s primary science case is as a follow-up instrument. On the other hand there are no prospects other than Darwin/TPF for finding all nearby Earth-mass planets (transits can only discover a small fraction), so its use for Earth-mass planet searches in a survey-type mode would be valuable, and should add to its contribution to the field. Such a survey might be undertaken during the mirror-filling phase, although detailed feasibility studies of this approach have not yet been made.
OWL/Dome C versus Darwin/TPF
As summarised in Section 3.1.2 and Table 8, appropriate Antarctic observations could at least partially compete with some aspects of the Darwin/TPF missions. However, some key spectral regions are accessible only from space, e.g. O<sub>3</sub> at 9.6 $`\mu `$m and CO<sub>2</sub> at 15 $`\mu `$m. Trade-offs would depend on the number of objects visible, the S/N ultimately attainable from Darwin, and the accessible spectral range and the resulting scientific information that can be inferred for exo-planetology and exobiology. Further investigations will be needed in order to understand whether such Antarctic-based observations could be considered as realistic in the medium term.
### 4.6 Open Areas: Survey Mission Beyond Kepler/Eddington
The absence of specific survey missions beyond 2015 noted in Section 4.1 can only be justified once all interesting targets in the sky to be studied in the future have been discovered. For small terrestrial planets, the Kepler mission is the only approved discovery mission. Even with Eddington, statistics of the occurrence of Earth-mass planets will still be poorly known, and the need for a larger, more performant, space transit survey seems already rather compelling. A case could also be made for an all-sky transit survey from space, in order to find the nearest transiting terrestrial planets, facilitating follow-up with astrometry and spectroscopy. The arguments presented in Section 3.3, ‘Understanding the Planetary Population of our Galaxy’, also draw attention to a lack of further search programmes for terrestrial planets beyond Kepler.
### 4.7 Other Considerations
#### 4.7.1 Fundamental Physical Data
It is not evident that present knowledge of basic atomic data is adequate to understand the high-resolution spectra of normal host stars over the broad wavelength region discussed in Section 4.4.1, and the spectral signatures of planets and their atmospheres. Preliminary investigations suggest that there is a need for improved atomic and molecular lines particularly in the near infrared. To illustrate the point and the potential risk of inadequate atomic data we give two examples: (i) Li in low mass post-AGB stars; expanded wavelength tables for Ce have shown that the line at 670.8 nm previously identified with (slightly redshifted) Li is coincident with a Ce II line at 670.8099 nm. Abundance analysis subsequently showed that for all practical purposes no Li is present in these stellar photospheres (Reyniers et al., 2002). This is an important finding for stellar evolution and nucleosynthesis since an elaborate mechanism had to be evoked in order to explain the presence of Li in these evolved stars; (ii) variability of the fine structure constant: evidence from the absorption features in distant quasars suggested a variability of the fine structure constant over time. Originally, these studies had been hampered by limited accuracy of laboratory wavelengths for the relevant species (Pickering et al., 2000). Most recently a further complication introduced by the sensitivity to the abundance of heavy isotopes has been described for Mg (Ashenfelter et al., 2004), where uncertainties in the isotope ratios might imitate a variation of the fine structure constant at the observed level. Nearer to the issues of exo-planets, the debate about the nature of Jupiter’s interior centres on the (unknown) equation of state of H/He mixtures at high pressures.
Regardless of the exact details these examples illustrate that the lack of accurate and reliable atomic data can lead to serious misinterpretation of astrophysical data. When dealing with fundamental issues such as the properties of extra-solar planets and their suitability for supporting life such pitfalls can be rather embarrassing and damaging to the credibility of the field. In order to assess the current situation properly, further consultation will be needed between the astronomical community and the relevant atomic and molecular physics community involved in the laboratory work. Links between these communities are also being investigated in the context of the Virtual Observatory.
#### 4.7.2 Fundamental Planetary Data
The solar system research community has the best knowledge of the only planetary system studied in detail, with extensive data archives like the Planetary Data System (PDS). For extra-solar planet research full use of that information must be made without being limited by it. Most of the planetary systems found so far are quite different from ours. Items of particular relevance might be: (i) distribution of minor bodies and dust, relevant for zodiacal light; (ii) mass radius relation for various kinds of planets; (iii) reflective properties of different planetary surfaces (gaseous/solid, rock/ice) in particular integrated over the sphere (and thus the influence of weather); (iv) spectral signatures of planetary atmospheres; (v) mass-loss from atmospheres; documented by planetary probes for Mars, indication for mass loss has also been found in HST observations of the transit of HD 209458; (vi) signatures of volcanism, like SO<sub>2</sub> (cf. Earth and Io).
Scientific results on solar system bodies and extra-solar planets are presented at various conferences of both communities, which in part have already established ‘cross-discipline’ sessions to support the exchange of knowledge between the two fields (e.g., the EGU and DPS). ESO and ESA might consider fostering the exchange of information and collaboration by organising joint cross-disciplinary workshops, including ESA’s Earth observation community. Such dedicated workshops could aim at the specific needs of extra-solar planet research and serve to define the needs for further observations of solar system bodies (e.g., atmospheres of planets and moons). In addition, building a data base collecting, for example, the spectra on planets, moons and minor bodies in our solar system as a reference for extra-solar planets could be discussed within both communities at such dedicated meetings.
#### 4.7.3 Amateur Networks
As described above, thousands of candidate planets will be discovered by ongoing surveys both from the ground and in space. Follow-up of all these candidates will be essential in order to verify their existence and to derive additional information about their properties. Observations of transits in particular will be valuable for candidates found by radial velocity or astrometric measurements. Dedicated (groups of) amateurs with state of the art equipment at reasonably large ($`>`$60 cm) telescopes could contribute in this area. The AAVSO is already calling on its observers to engage in such activity. Very recently, a dedicated campaign to observe possible transits of GJ 876b in the system IL Aqr was announced spanning a ten-hour period starting 21 Oct 2004 (http://www.aavso.org/news/ilaqr.shtml). ESO and ESA could take a leading role here by establishing and coordinating a dedicated Amateur Transit Network, an interesting opportunity for scientific reasons as well as for its potential for public outreach.
Scientifically, the most attractive aspect is that a global network (10–20 individual observatories) could provide very significant amounts of observing time and excellent time coverage. The latter is also essential for finding planetary signatures in microlensing events. Such a network probably will be the most cost-effective and efficient way to achieve a large increase in follow-up observations of candidates. The major challenge is to ensure that observations from different sources can successfully be combined for analysis. To achieve the required homogeneity, ESO/ESA could consider the use of common hardware (CCD and filters) by the members of the network and possibly software support. Relatively modest expenditures would be required to equip each participating observatory with a high-quality photometric system, very beneficial in terms of data quality. Although the effort required to coordinate such a network should not be underestimated, if properly coordinated and supported, an ESO/ESA amateur transit network could provide a cost-effective way to substantially increase the capablility for follow-up observations of candidates. Such an increase seems necessary in order to keep up with the demand projected to materialise over the next decade.
Sky & Telescope (December 2004, p18) reported that only eight days after the transiting system TrES–1 had been announced, a transit was observed by a Belgian amateur as well as by two observers from Slovenia. The first transit system HD 209458 is fairly bright at $`V=7.6`$ mag, but TrES–1 is only 11.8 mag. The detector hardware used for the discovery of TrES–1 is quite comparable to what advanced amateurs have. One main difference is the software which enables the monitoring of thousands of stars, and therefore acceptable chances for discovery. In this context, one aspect of Section 2.1.2 is recalled: Holman & Murray (2005) and Agol et al. (2005) describe how repeated observations of transits may lead to the discovery of Earth-mass planets: the gravitational influence of the small planet will induce pertubations in the orbit of the much larger planet causing the transit; over a period of time this will lead to a shift of the observed transit times on the order of minutes. The effect is more pronounced for long-period massive planets, with low-mass companions in orbital resonance.
Involvement of amateurs in extra-solar planet research also opens a very attractive field in terms of public outreach and eduction. The general public clearly is very interested in the topic of extra-solar planets. The subject therefore is well suited to provide information and eductational material explaining the prospects and limitations of searches for extra-solar planets to the public and decision makers. Given the very substantial investments required to realise the next generation of extremely large telescopes, astronomy should make a dedicated effort to take advantage of this overlap of public and research interest. While an outreach effort on extra-solar planets will be worthwhile in its own right, we expect that an outreach effort centered around an amateur (public) involvement in science will be much more effective thanks to its novelty and emotional appeal. In order to get some feedback from the amateur community directly one of us (FK) gave a presentation at the 8 November 2004 meeting of the astronomy club ‘Max Valier’ in Bolzano, Italy. The club has more than 100 members and is well-equipped with an 80 cm telescope and modern CCDs. Their new observatory at Gummer was built with funds of the local government and is now run as a public observatory (http://www.sternwarte.it) with an extensive program of tours and observing sessions for schools and the general public. Further details of this meeting, and a more detailed proposal for follow-up, are available.
## 5 Recommendations
Europe has already established leadership in major areas of the exo-planet field, including radial velocity (HARPS), transit searches (COROT), and astrometry (Gaia).
The first goal of future actions should be to take full advantage of this situation, with an offensive policy to optimize the scientific return of instruments already built or foreseen in the near future \[the ongoing or planned ESO instrument programmes (HARPS, UVES, CRIRES, OmegaTranS, PRIMA, GENIE, Planet Finder, etc.) are not considered further here\].
The second goal is to prepare new initiatives. Suggested directions are:
1. ESA
1.1. Eddington: provide a clearer message to the community about the plans and schedule for an Eddington-type exo-planetary mission \[Section 2.2.1\].
1.2. Gaia: recognise that the highest accuracy will supply the most comprehensive information on low-mass planets in the Solar neighbourhood (down to about 10 $`M_{}`$), thus assisting the identification of targets for Darwin/OWL/PRIMA, and the largest number of detections, scaling as the third power of the accuracy \[Section 2.2.2\].
1.3. Darwin: maintain the research and development plan for this key project in the domain of exo-planet science at high priority, consistent with the current 2015 target launch date, cf. TPF-C launch planned for 2014 \[Section 3.2\].
1.4. JWST: support the ‘Astrobiology and JWST’ panel recommendations, to ensure that JWST can follow up low-mass transits discoveries \[Section 2.2.4\].
1.5. Encourage the community to submit mission proposals covering the important themes in Section 3.3: e.g. a future transit survey mission significantly more performant than currently planned; an all-sky transit mission; a UV spectroscopic mission for transit spectroscopy; astrometric detection of Earth-mass planets out to 20–30 pc, etc. \[Section 3.3\].
2. ESO
2.1. Support experiments to improve radial velocity mass detection limits, e.g. based on experience from HARPS, down to those imposed by stellar surface phenomena \[Section 1.3\].
2.2. Characterise nearby potential planet host stars, e.g. $`T_{\mathrm{eff}}`$, log $`g`$, \[Fe/H\], $`M`$, radius, etc. \[Section 4.4\].
2.3. Improve the capabilities of main-stream VLT instruments for high cadence, high S/N transit spectroscopy in the visible and infrared \[Section 2.1.2\].
2.4. Evaluate observation time for follow-up observations over the next 10 years of transit candidates obtained by space missions (COROT, Kepler, Eddington-like) and ground-based observations, including high-resolution imaging and photometry, on small to very large telescopes \[Section 4.2.3\].
2.5. In addition to the use of OWL as a follow-up facility, consider prospects for OWL searches for Earth-mass planets in the solar neighbourhood, including during the filling phase \[Section 3.1.1\].
2.6. Investigate the astrometric capabilities of OWL, possibly leading to the inclusion of an astrometric facility in the instrumentation plan \[Section 3.1.1\].
3. ESO–ESA – Joint Initiatives
3.1. Consider radial velocity follow-up of COROT (and possibly Kepler or Eddington-type missions – but note caveats of Kepler visibility) transit candidates . This would involve hundreds of targets and requires: (1) a major time allocation of La Silla 3.6-m telescope to HARPS; (2) probably an instrument with a precision similar to HARPS but on a larger telescope \[Section 4.2.3\].
3.2. Consider facilities for radial velocity follow-up of the large number of candidates (20–30 000) from Gaia, requiring observing time on relatively small, possibly robotic, telescopes for a few years of full-time operations \[Section 4.2.3\].
3.3. Consider the photometric (transit) monitoring of large number of high-mass planet candidates which will come from space experiments \[Section 4.2.3\].
3.4. Support valid observing time requests for preparatory work to space exo-planet missions, e.g., field characterisation for an Eddington-like mission \[Section 4.2.3\].
3.5. Evaluate the prospects of implementing a small interferometric array, along the lines of GENIE, at Dome C/A, given that it could radically change the capabilities of ELTs if (a second) OWL were located there \[Section 3.1.2\].
3.6. Consider coordination of amateur networks, along the lines of AAVSO, for the follow-up of hot Jupiters detected from ground transit surveys, from Gaia astrometry and photometry, and for surveys for longer-period objects \[Section 4.7.3\].
3.7. Foster cooperation between the solar system research community and the extra-solar planets community, e.g. by supporting or establishing joint meetings of both communities addressing common topics, such as formation, comparative planetology, emergence and evolution of life on Earth and elsewhere, etc. \[Section 4.7.2\].
3.8. Coordinate exo-planet public communication, with some common ‘code of conduct’, e.g.: (i) no press release without a supporting scientic paper; (ii) claims should not be overstated, with support sought from external organisations or universities; (iii) retractions should be posted in the same form to retain credibility.
## Index to ESA and ESO facilities:
Descriptions of each experiment or facility are given as follows:
#### ESA:
COROT: see Section 2.2.1
Eddington: see Section 2.2.1
Gaia: see Section 2.2.2
HST (ESA/NASA): see Section 2.2.1
JWST (ESA/NASA): see Section 2.2.4
#### ESO:
ALMA: see Section 3
CRIRES: see Section 2.1.2
GENIE: see Section 3.2.2
HARPS: see Section 2.1.1
NAOS-CONICA: see Section 2.1.6
OWL: see Section 3.1.1
Planet Finder: see Section 2.1.6
PRIMA: see Section 2.1.6
VLTI: see Section 2.1.6
#### NASA/Other:
FAME/AMEX/OBSS: see Section 2.2.2
JASMINE (Japan): see Section 2.2.2
Kepler: see Section 2.2.1
MPF/GEST: see Section 2.2.3
SIM: see Section 2.2.2
SOFIA: see Section 2.2.4
Spitzer (ex-SIRTF): see Section 2.2.4
## Appendix A Space Precursors: Interferometers, Coronographs and Apodizers
The McKee-Taylor Decadal Survey Committee (McKee & Taylor, 2000) qualified its endorsement of the TPF mission with the condition that the abundance of Earth-size planets be determined prior to the start of the TPF mission. Many ideas for scientific and technological precursors for TPF have been examined. The main contenders are summarised here for completeness, although with the recent (April 2004) NASA announcement on TPF strategy, it seems unlikely that any of these concepts will be developed further:
Eclipse (coronography) is a proposed NASA Discovery-class mission to perform a direct imaging survey of nearby planetary systems, including a complete survey for Jovian-sized planets orbiting 5 AU from all stars of spectral types A–K within 15 pc of the Sun (Trauger et al., 2003). Its optical design incorporates a telescope with an unobscured aperture of 1.8 m, a coronographic camera for suppression of diffracted light, and precision active optical correction for suppression of scattered light, and imaging/spectroscopy. A three-year science mission would survey the nearby stars accessible to TPF. Eclipse may be resubmitted for NASA’s Discovery round in 2004.
Jovian Planet Finder (JPF) was a MIDEX proposal to directly image Jupiter-like planets around some 40 nearby stars using a 1.5-m optical imaging telescope and coronographic system, originally on the International Space Station (ISS) (Clampin et al., 2002). Its sensitivity results from super-smooth optical polishing, and should be sensitive to Jovian planets at typical distances of 2–20 AU from the parent star, and imaging of their dusty disks – potentially solar system analogues. A 3-yr mission lifetime was proposed.
Extra-Solar Planet Imager (ESPI) is another proposed precursor to TPF (Lyon et al., 2003). Originally proposed as a NASA Midex mission as a $`1.5\times 1.5`$ m<sup>2</sup> apodized square aperture telescope, reducing the diffracted light from a bright central source, and making possible observations down to 0.3 arcsec from the central star. Jupiter-like planets could be detected around 160–175 stars out to 16 pc, with S/N $`>5`$ in observations lasting up to 100 hours. Spectroscopic follow-up of the brightest discoveries would be made. The Extra-Solar Planet Observatory (ExPO) is a similar concept proposed as a Discovery-class mission (Gezari et al., 2003).
Self-luminous Planet Finder (SPF) is a further TPF precursor under study by N. Woolf and colleagues, aiming at the search for younger or more massive giant planets in Jupiter/Saturn like orbits, where they will be highly self-luminous and bright at wavelengths of 5–10 $`\mu `$m, where neither local nor solar system zodiacal glow will limit observations. SPF will demonstrate the key technologies of passive cooling associated with interferometric nulling and truss operation that are required for a TPF mission. SPF targets young Jupiter-like planets both around nearby stars such as $`ϵ`$ Eri, and around A and early F stars.
Fourier-Kelvin Stellar Interferometer (FKSI) is a concept under study at NASA GSFC (Danchi et al., 2003). It is a space-based mid-infrared imaging interferometer mission concept being developed as a precursor for TPF. It aims to provide 3 times the angular resolution of JWST and to demonstrate the principles of interferometry in space. In its minimum configuration, it uses two 0.5-m apertures on a 12.5-m baseline, and predicts that some 7 known exo-planets will be directly detectable, with low-resolution spectroscopy ($`R20`$) possible in favourable cases.
Optical Planet Discoverer (OPD) is a concept midway between coronography and Bracewell nulling (Mennesson et al., 2003).
Phase-Induced Amplitude Apodization (PIAA, Guyon (2003)) is an alternative to classical pupil apodization techniques (using an amplitude pupil mask). An achromatic apodized pupil is obtained by reflection of an unapodized flat wavefront on two mirrors. By carefully choosing the shape of these two mirrors, it is possible to obtain a contrast better than $`10^9`$ at a distance smaller than $`2\lambda /d`$ from the optical axis. The technique preserves both the angular resolution and light-gathering capabilities of the unapodized pupil, and claims to allow efficient detection of terrestrial planets with a 1.5-m telescope in the visible.
Occulting masks are another approach to tackle in a conceptually simple manner the basic problem of how to separate dim sources from bright ones, and have been considered as precursor missions to Darwin/TPF. Interest in this approach at NASA level currently appears limited.
UMBRAS (Umbral Missions Blocking Radiating Astronomical Sources) refers to a class of missions (Schultz et al., 2003), currently designed around a 4-m telescope and a 10-m occulter, with earlier concepts including a 5–8 m screen (CORVET), or as NOME (Nexus Occulting Mission Extension) a modification to Nexus, itself foreseen as a test of technologies for JWST.
BOSS (Big Occulting Steerable Satellite (Copi & Starkman, 2000)) consists of a large occulting mask, typically a $`70\times 70`$ m<sup>2</sup> transparent square with a 35 m radius, and a radially-dependent, circular transmission function inscribed, supported by a framework of inflatable or deployable struts. The mask is used by appropriately aligning it with a ground- or space-based observing telescope. In combination with JWST, for example, both would be in a Lissajous-type orbit around the Sun-Earth Lagrange point L2, with the mask steered to observe a selected object using a combination of solar sailing and ion or chemical propulsion. All but about $`4\times 10^5`$ of the light at 1 $`\mu `$m would be blocked in the region of interest around a star selected for exo-planet observations. Their predictions suggest that planets separated by as little as 0.1–0.2 arcsec from their parent star could be seen down to a relative intensity of $`1\times 10^9`$ for a magnitude 8 star. Their simulations indicate that for systems mimicking our solar system, Earth and Venus would be visible for stars out to 5 pc, with Jupiter and Saturn remaining visible out to about 20 pc.
## Appendix B Beyond 2025: Life Finder and Planet Imager
Within NASA’s Origins Program HST, Spitzer and others are referred to as ‘precursor missions’, with SIM and JWST as ‘First Generation Missions’ leading to the ‘Second Generation Mission’ TPF which will begin to examine the existence of life beyond our Solar System. Once habitable planets are identified, a ‘Life Finder’ type of mission would expand on the TPF principles to detect the chemicals that reveal biological activities. And once a planet with life is found, ‘Planet Imager’ would be needed to observe it. These ‘Third Generation Missions’, Life Finder and Planet Imager, are currently just visions because the required technology is not on the immediate horizon. Short descriptions are included here for completeness.
Life Finder: Taking pictures of the nearest planetary system (Darwin/TPF) is considered to be a reasonable goal on a 10-year timescale, with low-quality spectra a realistic by-product. Life Finder, which would only be considered after Darwin/TPF results are available, and once oxygen or ozone has been discovered in the atmosphere, would aim to produce confirmatory evidence of the presence of life, searching for an atmosphere significantly out of chemical equilibrium, for example through its oxygen (20% abundance on Earth) and methane ($`10^6`$ abundance on Earth). Some pointers to the technology requirements and complexity of Life Finder have been described in the ‘Path to Life Finder’ (Woolf et al., 2001). Given that Darwin/TPF will take low-resolution low-S/N spectra, a large area high angular resolution telescope will be needed for detailed spectral study in order to confirm the presence of life. Recalling that the target objects will be as faint as the Hubble Deep Field galaxies, buried in the glare of their parent star some 0.05–0.1 arcsec away, the light collecting area of Life Finder will have to be substantially larger than TPF’s 50 m<sup>2</sup>: a useful target is 500–5000 m<sup>2</sup>. One of the primary technical challenges will be to produce such a collecting area at affordable cost and mass.
The required development of new low mass and better wavefront optics, coronography versus nulling, pointing control by solar radiation pressure, sunshield, vibration damping, and space assembly, were addressed by Woolf et al. (2001). According to their study, a ‘mini-Life Finder’ might be a $`50\times 10`$ m<sup>2</sup> telescope, made with 12 segments of $`8.3\times 5`$ m<sup>2</sup>, made of 5 kg m<sup>-2</sup> glass, piezo-electric controlled adaptive optics, and a total mass (optics and structure) of about 10 tons. Cooling would be by an attached sunshade also used for solar pressure pointing, in a ‘sun orbiting fall away’ orbit to avoid the generation of thruster heat needed to maintain the L2-type orbit. There are still unsolved complexities underlying the actual science case for Life Finder: if the goal is to detect the 7.6 $`\mu `$m methane feature — which is not definitively the relevant goal; see, for example discussions of the use of the ‘vegetation signature’ in Arnold et al. (2002) — the required collecting area accelerates from a plausible 220 m<sup>2</sup> (four or five 8-m telescopes) for a planet at 3.5 pc, to a mighty 4000 m<sup>2</sup> (eighty 8-m telescopes) even at only 15 pc. A new proposal to study Life Finder has recently been submitted to NASA by Shao, Traub, Danchi & Woolf (N. Woolf, private communication). Various reports on related studies can be found under NASA’s Institute for Advanced Concepts (NIAC) www pages (http://www.niac.usra.edu/) including ‘Very large optics for the study of Extra-Solar Terrestrial Planets’ (N. Woolf); and ‘A structureless extremely large yet very lightweight swarm array space telescope’ (I. Bekey). The former includes an outline technology development plan for Life Finder, with costs simply stated as $``$$2 billion.
Planet Imager: TPF aims to obtain images of planetary systems in which the planets appear as point sources. Resolving the surface of a planet is, at best, a far future goal requiring huge technology development that is not yet even in planning. Much longer baselines will be required, from tens to hundreds of km in extent. Formation flying of these systems will require technology development well beyond even the daunting technologies of Darwin/TPF – complex control systems, ranging and metrology, wavefront sensing, optical control and on-board computing. Having accepted that we are now peering into a much more distant and uncertain future, we can examine some of the ideas which are being discussed.
Life Finder studies (Woolf et al., 2001) have been used to evaluate the requirements for Planet Imager which, they consider, would require some 50–100 Life Finder telescopes used together in an interferometric array. Their conclusions were that ‘the scientific benefit from this monstrously difficult task does not seem commensurate with the difficulty’. This echoes the conclusions of Bender & Stebbins (1996) who undertook a partial design of a separated spacecraft interferometer which could achieve visible light images with $`10\times 10`$ resolution elements across an Earth-like planet at 10 pc. This called for 15–25 telescopes of 10-m aperture, spread over 200 km baselines. Reaching $`100\times 100`$ resolution elements would require 150–200 spacecraft distributed over 2000 km baselines, and an observation time of 10 years per planet. These authors noted that the resources they identified would dwarf those of the Apollo Program or the Space Station, concluding that it was ‘difficult to see how such a program could be justified’. The effects of planetary rotation on the time variability of the spectral features observed by an imager, complicates the imaging task although may be tractable, while more erratic time variability (climatic, cloud coverage, etc.) will greatly exacerbate any imaging attempts.
Hypertelescopes/OVLA:
Parallel to the Planet Imager studies, in Europe the LISE group (Laboratoire d’Interférométrie Stellaire et Exo-planétaire) carries out research in the area of high-resolution astronomical imaging, including imaging extra-solar planets. The group is studying several complementary projects for ‘hypertelescopes’ on Earth and in space (Riaud et al., 2002; Gillet et al., 2003). The steps needed to reach this goal are set out as requiring: (1) a hypertelescope on Earth – the OVLA (Optical Very Large Array); (2) a 100-m precursor geostationary version in space; (3) a km-scale version in a higher orbital location; (4) a 100 km version, including dozens of mirrors of typically 3 m aperture. Labeyrie et al. proposed the mission ‘Epicurus’, an extra-solar Earth imager, to ESA in 1999 in response to the F2/F3 call for mission proposals.
Their basic ‘hypertelescope’ design involves a dilute array of smaller apertures (an imaging interferometer) having a ‘densified’ exit pupil, meaning that the exit pupil has sub-pupils having a larger relative size than the corresponding sub-apertures in the entrance pupil (see Fig. 1 of Pedretti et al. (2000)). Their applicability extends to observing methods highly sensitive to the exit pupil shape, such as phase-mask coronography.
In the most recent published studies (Riaud et al., 2002) the hypertelescope is combined with such a coronograph to yield attenuations at levels of $`10^8`$. Simulations of 37 telescopes of 60 cm aperture distributed over a baseline of 80 m in the infrared, observing the 389 Hipparcos M5–F0 stars out to 25 pc (with simulated contributions from zodiacal and exo-zodiacal background) yields 10-hour snapshot images in which an Earth-like planet is potentially detectable around 73% of the stars. Gains of a factor 20–30 with respect to a simple Bracewell nulling interferometer are reported.
In space, the plans call for a flotilla of dozens or hundreds of small elements, deployed in the form of a large dilute mosaic mirror. Pointing is achievable by globally rotating the array, which is slowly steerable with small solar sails attached to each element. A ‘moth-eye’ version allows full sky coverage with fixed elements, using several moving focal stations (Labeyrie, 1999b, a). The geostationary precursor hypertelescope could be a version of TPF. An exo-Earth discoverer would require a 100–1000 m hypertelescope with coronograph, while an exo-Earth imager would require a 150 km hypertelescope with coronograph. In the approach of Labeyrie (1999b) a 30-min exposure using a hypertelescope comprising 150 3-m diameter mirrors in space with separations up to 150 km, would be sufficient to detect ‘green’ spots similar to the Earth’s Amazon basin on a planet at 10 light-years, although these vegetation features are more prominent in the infrared (Arnold et al., 2002).
## Appendix C ESO 1997 Working Group on Extra-Solar Planets
The ESO 1997 Working Group on the Detection of Extra-Solar Planets consisted of 16 members, and met on four occasions during 1996 and 1997. The assignment was to ‘advise ESO \[…\] on how to help designing a competitive strategy in this field that is predicted to expand dramatically in the next years, and to become one of the leading fields of astronomy in the next century.’ The 1997 Working Group and its task were somewhat comparable to the current effort, with the notable difference that this time the Working Group was asked to provide recommendations to both ESA and ESO, thereby encompassing both space- and ground-based astronomy.
This appendix looks briefly at the findings and recommendations of their final report, published on 10 September 1997. This may provide some understanding of why some recommendations were successfully implemented while others were not, and what lessons can be learned from the past.
In the introduction of their report the 1997 Working Group states: ‘Only by allocating a major fraction of time on some of its telescopes and developing new technology — and doing it now – will ESO fully exploit its potential in this field and be truly competitive.’ They then laid out their plan to achieve this goal. They identified six areas in which ESO could play a critical role, namely: radial velocity searches, narrow-angle astrometry, microlensing, direct detection, transits and timing of eclipses. The recommendations, planned timeframe, and status and achievements of each of these is summarised in Table 10.
Radial Velocity Searches: The main recommendation was to devote a major fraction of the observing time at the CAT and the 3.6 m to monitor the radial velocities of about 1000 stars over the next 5–10 years. They called for the development of a dedicated spectrograph providing an accuray of about 1 m/s. Such a survey was considered indispensable to provide targets for VLTI which was assumed to become operational by 2002. They also advocated high-fidelity calibration of the iodine cell for UVES and the timely development of CRIRES to provide a survey using twilight observations in order to obtain complimentary information in the IR.
Narrow-Angle Astrometry: The Working Group considered astrometry a very attractive method for the study of planets because it allows for determination of orbits and planet masses directly. It was considered complimentary to other search methods exploring different regions of parameter space. Achieving an accuracy of about 10 micro-arcsec was considered feasible. Realisation of the technique was regarded as technically very challenging, requiring knowledge of the length of the baseline to within 50 $`\mu `$m and the knowledge of the delay between the two stars with 5 nm precision over a 100 m baseline.
Microlensing: Microlensing was identified as a method which is in principle able to detect Earth-mass planets. Since searches for Jovian planets were already in the development phase in 1997 the Working Group suggested that ESO should focus on detecting Earth-mass planets. They suggested that a dedicated 2.5 m telescope should monitor the bulge with a large (1) field detector (16k by 16k CCD) during a 120 night season. It would observe a few million uncrowded stars achieving 1 % photometric accuracy reaching $`V=20`$ in a 4-minute exposure. A high sampling rate is crucial to obtain the characteristic of the short (5 hr) planet event on the microlensing light curve. No specific time frame was given but the Working Group called for an aggressive ESO-based campaign.
When VST/OmegaCam becomes operational in 2006, the technical capabilities for the above programme will be available. Microlensing searches for Earth-mass planets, though, are not part of the major science goals for VST/OmegaCam.
Direct Detection: Direct detection was considered essential for deriving many physical properties of extra-solar plants such as size, temperature, chemical composition etc. The Working Group stressed that it would be extremely challening to achieve the required contrast levels in particular from the ground. They identified a very powerful adaptive optics system as a key ingredient in combination with coronographic instrumentation and sophisticated data processing. They pointed out that spectroscopic signatures of planetary atmospheres should be detectable with high-resolution spectroscopy in the NIR (CRIRES) via their time-dependent Doppler shift. Other spectral features will be unique to the planet and therefore appear as ‘alien’ features in the stellar spectrum. They called for further studies of instrumental requirements, and an early realization of an instrument like CRIRES.
Transits and Timing Eclipses: The Working Group noted the potential of this method for detecting planets down to Uranus size, along with planets with rings and moons of giant planets. They mention the possibility of obtaining spectra of planets’ atmospheres during transits. They also noted that timing of eclipses in binary stars was a simple method for detecting giant planets in these systems.
|
warning/0506/cond-mat0506078.html
|
ar5iv
|
text
|
# Adiabatic pumping in a Superconductor-Normal-Superconductor weak link
## Abstract
We present a formalism to study adiabatic pumping through a superconductor - normal - superconductor weak link. At zero temperature, the pumped charge is related to the Berry phase accumulated, in a pumping cycle, by the Andreev bound states. We analyze in detail the case when the normal region is short compared to the superconducting coherence length. The pumped charge turns out to be an even function of the superconducting phase difference. Hence, it can be distinguished from the charge transferred due to the standard Josephson effect.
In a mesoscopic conductor a dc charge current can be obtained, in the absence of applied voltages, by cycling in time two parameters which characterize the system thouless83 ; switkes99 . This transport mechanism is called pumping. If the time scale over which the time-dependent parameters vary is large compared to the typical electron dwell time of the system, then pumping is adiabatic, and the pumped charge does not depend on the detailed timing of the cycle, but only on its geometrical properties. Different formulations have been developed to describe adiabatic pumping, for example, based on scattering theory in Refs. brouwer98 ; makhlin01 ; buttiker02 or Green’s function methods in Refs. zhou99 ; entin02 . In the scattering approach the pumped charge per cycle can be expressed in terms of derivatives of the scattering amplitudes with respect to the pumping parameters (Brouwer’s formula brouwer98 ). This result is based on the so-called emissivities of the system buttiker94 , which express the charge that flows from a lead in response to the variation of one parameter in the scattering region. This formulation requires the presence of terminals which provide propagating channels. The scattering approach has been later extended to hybrid systems containing superconducting (S) terminals. In Refs. wang01 ; blaauboer02 a two terminal structure comprising one superconducting lead was considered. Subsequently, this approach was generalized to multiple-superconducting-lead systems, where at least one normal lead is present taddei04 . The presence of a normal lead is essential for generalizing Brouwer’s formula to these hybrid structures.
If only superconducting leads are present, at low enough temperature, pumping is due to the adiabatic transport of Cooper pairs. Besides the dependence of the pumped charge on the cycle, in the superconducting pumps there is a dependence on the superconducting phase difference(s) (the overall process is coherent). Moreover, in addition to Cooper-pair pumping, in equilibrium, there is a contribution due to the Josephson effect if the phase difference between the two superconducting leads is different from zero. Up to now adiabatic Cooper pair pumping was studied only in the Coulomb Blockade regime for all-superconducting systems (superconducting islands weakly connected to superconducting leads) geerligs91 ; pekola99 ; fazio03 ; niskanen03 . In this Letter we would like to (partially) fill this gap, and consider adiabatic pumping between two superconducting terminals connected through a normal (N) region. We focus on the regime of an open structure (SNS weak link), where charging effects are negligible. This is relevant when the normal region is, for example, a chaotic cavity as those used for normal, electronic pumps switkes99 .
The derivation of a formula for the pumped charge makes use of the connection between Berry’s phase berry84 and the pumped charge avron00 ; aunola03 ; bender05 , which we prove to be valid also for the SNS weak link. The resulting expression for the charge pumped in a period can be written in terms of derivatives of the Andreev-bound-state wavefunctions with respect to the pumping parameters. We point out that there is a close analogy of the problem studied here with that of pumping in a Aharonov-Bohm ring moskaletsring .
The system under investigation (depicted in Fig. 1) consists of a SNS junction, with the weak link occupying the region $`W/2<x<W/2`$. The superconducting order parameter is given by $`\mathrm{\Delta }_0e^{i\phi /2}`$ and $`\mathrm{\Delta }_0e^{i\phi /2}`$ for the superconductor on the left-hand-side and right-hand-side, respectively. The properties of the weak link can be varied, for example, by realizing it with a semiconductor and operating on two independent external gates, indicated by $`X_1(t)`$ and $`X_2(t)`$ in the figure.
The state of such a hybrid structure $`|\psi (t)`$ is the solution of the time-dependent Bogoliubov-de Gennes equation:
$$i\mathrm{}_t|\psi (t)=H(t)|\psi (t),$$
(1)
where the Hamiltonian
$$H(t)=\left(\begin{array}{cc}\frac{\mathrm{}^2}{2m}\stackrel{}{}^2+U(\stackrel{}{r},t)\mu & \mathrm{\Delta }(\stackrel{}{r})e^{i\varphi (\stackrel{}{r})}\\ \mathrm{\Delta }(\stackrel{}{r})e^{i\varphi (\stackrel{}{r})}& \frac{\mathrm{}^2}{2m}\stackrel{}{}^2U(\stackrel{}{r},t)+\mu \end{array}\right)$$
(2)
depends on time through the two parameters: $`H(t)=H[X_1(t),X_2(t)]`$. In Eq. (2) $`U(\stackrel{}{r},t)`$ is the potential that takes into account the effect of the time-varying external gate voltages, $`\varphi (\stackrel{}{r})=\phi /2[\theta (xW/2)\theta (xW/2)]`$, $`\mathrm{\Delta }(\stackrel{}{r})=\mathrm{\Delta }_0/2[\theta (xW/2)+\theta (xW/2)]`$ and $`\mu `$ is the superconductor chemical potential (equal for the two S leads). We now assume that the state $`|\psi (t)`$ evolves adiabatically and that at any time $`t`$ it is in an instantaneous eigenstate of the Hamiltonian. The instantaneous solutions are defined by the equation:
$$H(t)|\stackrel{~}{\psi }(t)=ϵ(t)|\stackrel{~}{\psi }(t),$$
(3)
whereby $`t`$ plays the role of a parameter. After a cycle of period $`\tau `$, the states returns to the initial one, but with an added phase factor $`\mathrm{\Phi }`$:
$$|\psi (\tau )=e^{i\mathrm{\Phi }}|\psi (0),$$
(4)
The phase $`\mathrm{\Phi }`$ contains both a geometrical (Berry’s) and a dynamical contribution $`\mathrm{\Phi }=\gamma _\text{B}\gamma _\text{D}`$. The dynamical phase is simply given by $`\gamma _\text{D}=\frac{1}{\mathrm{}}_0^\tau 𝑑t\stackrel{~}{\psi }|H|\stackrel{~}{\psi }`$
For the SNS system, where Andreev bound states are formed, the condition of validity of the adiabatic approximation is that the frequency $`\mathrm{}\omega `$ of the time-dependent parameters be much smaller than the energy difference between any pair of Andreev bound states, or between any Andreev bound states and the continuum of states above the gap. This implies that the pumping frequency needs at least to be smaller than the superconducting gap $`\mathrm{\Delta }_0`$.
It is possible to show explicitly that the charge current $`J_{\text{ch}}`$ carried by a Bogoliubov-de Gennes eigenstate $`|\stackrel{~}{\psi }`$ is given by the expectation value of the derivative of the Hamiltonian $`H`$ with respect to the superconducting phase difference ludin99 :
$$J_{\text{ch}}=\frac{2e}{\mathrm{}}\stackrel{~}{\psi }|\frac{H}{\phi }|\stackrel{~}{\psi }.$$
(5)
The charge transferred per cycle is then defined as $`Q=_0^\tau J_{\text{ch}}(t)𝑑t.`$ By assuming adiabatic evolution of the state, and making use of Eq. (5), the following relation between the accumulated phase and the charge transferred in a cycle can be written:
$$Q=2e\frac{}{\phi }(\gamma _\text{D}\gamma _\text{B}).$$
(6)
The first term corresponds to the instantaneous Josephson current integrated over one period, while the second represents the pumped charge. Using Green’s theorem, $`\gamma _\text{B}`$ can be written in terms of derivatives with respect to the pumping parameter of the instantaneous eigenfunctions:
$$\gamma _\text{B}=2_\text{S}𝑑X_1𝑑X_2\text{Im}\left[\frac{\stackrel{~}{\psi }}{X_1}|\frac{\stackrel{~}{\psi }}{X_2}\right],$$
(7)
$`S`$ being the area in the parameter space spanned by the parameters over one cycle. In Eq. (7) the notation $`|`$ stands for a space integration defined by $`A|B=𝑑\stackrel{}{r}A^{}(\stackrel{}{r})B(\stackrel{}{r})`$, $`A`$ and $`B`$ being vectors in the Nambu space.
In the short junction limit (i.e. when the distance $`W`$ between the two superconducting interfaces is much smaller than the superconducting coherence length) only the superconducting regions contribute to the space integration in Eq. (7). The instantaneous eigenfunction, corresponding to the Andreev-bound-state energy $`ϵ_j`$, in the superconducting regions can be written as:
$$\stackrel{~}{\psi _j}(\stackrel{}{r})=\{\begin{array}{cc}\stackrel{~}{\psi }_{\text{S,L},j}(\stackrel{}{r})\hfill & \hfill xW/2\\ \stackrel{~}{\psi }_{\text{S,R},j}(\stackrel{}{r})\hfill & \hfill xW/2\end{array},$$
(8)
with
$$\stackrel{~}{\psi }_{\text{S,L},j}(\stackrel{}{r})=\underset{n}{}\left(𝐛_{\text{L}n,j}^+e^{ik_{n,j}^+x}+𝐛_{\text{L}n,j}^{}e^{ik_{n,j}^{}x}\right)\chi _n(y,z)$$
(9)
and
$$\stackrel{~}{\psi }_{\text{S,R},j}(\stackrel{}{r})=\underset{n}{}\left(𝐛_{\text{R}n,j}^+e^{ik_{n,j}^+x}+𝐛_{\text{R}n,j}^{}e^{ik^{}{}_{n}{}^{},jx}\right)\chi _n(y,z),$$
(10)
$`n`$ being the transverse channel index relative to the transverse wavefunction $`\chi _n(y,z)`$ with transverse subband energies $`E_n`$ note0 . The index $`j`$ labels the different Andreev bound states. In Eqs. (9) and (10) $`k_{n,j}^\pm `$ are particle(hole)-like quasiparticle wavevectors given, in the Andreev approximation ($`\mathrm{\Delta }_0\mu E_n`$ for any $`E_n`$), by $`k_{n,j}^\pm =k_{\text{F}n}\left(1\pm \frac{i}{2}\sqrt{\frac{\mathrm{\Delta }_0^2ϵ_j^2}{(\mu E_n)^2}}\right)`$, with $`k_{\text{F}n}=\sqrt{\frac{2m}{\mathrm{}^2}(\mu E_n)}`$. The Nambu-space vectors $`𝐛_{\nu n,j}^\beta `$ can be calculated with the following procedure: i) the eigenfunction in the fictitious leads in the normal regions adjacent to the superconducting interface (see Fig. 1) are calculated along the lines of Ref. beenakker92 ; ii) the wavefunction in the superconductor is obtained by imposing the continuity equations at the interfaces within the Andreev approximation; iii) the normalization condition $`_{\nu =\text{L,R}}\stackrel{~}{\psi }_{\text{S},\nu ,j}|\stackrel{~}{\psi }_{\text{S},\nu ,j}=1`$ is imposed.
Making use of the Andreev approximation, the pumped charge reduces to
$$Q_{\text{}}=4e\frac{}{\varphi }_\text{S}𝑑X_1𝑑X_2\underset{\stackrel{\stackrel{\beta =+,}{\nu =\text{L,R}}}{n,j}}{}\left\{\frac{1}{2\text{Im}k_{n,j}^+}\text{Im}\left[\frac{𝐛_{\nu n,j}^\beta }{X_1}\frac{𝐛_{\nu n,j}^\beta }{X_2}\right]+\frac{1}{(2\text{Im}k_{n,j}^+)^2}\text{Re}\left[𝐛_{\nu n,j}^\beta \frac{𝐛_{\nu n,j}^\beta }{X_2}\frac{k_{n,j}^+}{X_1}+\frac{𝐛_{\nu n,j}^\beta }{X_1}𝐛_{\nu n,j}^\beta \frac{k_{n,j}^+}{X_2}\right]\right\},$$
(11)
where the sum over $`j`$ runs over the Andreev bound states. This is the central result of this Letter, and a few comments are in order. We have succeeded in expressing the pumped charge as a function of the instantaneous Andreev-bound-state eigenfunction. The vectors $`𝐛_{\text{L(R)}n,j}^\pm `$ depend only on the parameters of the system, such as the normal region scattering matrix $`S_0`$, the superconducting gap, and the superconducting phase difference. It is clear that the pumped current can be written in terms of the elements of the normal-region scattering matrix $`S_0`$. Equation (11) is a zero temperature result, and contains only the contribution to the pumped charge due to the Andreev bound states. At finite temperatures, but still smaller than the gap, the contributions of the different Andreev bound states $`ϵ_j`$ are weighted by the thermal occupation $`12f(ϵ_j)`$, being $`f`$ the Fermi function. At temperatures of the order of the gap, there is an additional contribution, not contained in Eq. (11), due to the propagating quasi particles with energies above the gap. When superconductivity is suppressed only the latter contribution, which is described by Brouwer’s formula, remains, leading to the usual result for the pumped charge through a normal region connected to normal leads.
Now let us consider the following single-channel parametrization for the normal-conductor scattering matrix
$$S_0=\left(\begin{array}{cc}e^{i\alpha }\sqrt{1g}& i\sqrt{g}\\ i\sqrt{g}& e^{i\alpha }\sqrt{1g}\end{array}\right),$$
(12)
choosing $`g`$ and $`\alpha `$ as pumping fields ($`X_1=g`$, and $`X_2=\alpha `$). The instantaneous Andreev-bound-state energy $`ϵ_0`$ is simply related to the transmission probability $`g`$ by beenakker92
$$ϵ_0(t)=\mathrm{\Delta }_0\sqrt{1g(t)\mathrm{sin}^2(\frac{\phi }{2})},$$
(13)
so that $`\gamma _\text{D}=1/\mathrm{}_0^\tau ϵ_0(t)`$.
The charge transfered due to the Josephson current (in the following named also Josephson charge) reads
$$Q_{\text{Jos}}=2e\frac{\mathrm{\Delta }_0}{\mathrm{}}_0^{\frac{2\pi }{\omega }}𝑑t\frac{g(t)}{2}\frac{\mathrm{sin}\left(\frac{\phi }{2}\right)\mathrm{cos}\left(\frac{\phi }{2}\right)}{\sqrt{1g(t)\mathrm{sin}^2\left(\frac{\phi }{2}\right)}}.$$
(14)
Notice that it depends on the pumping frequency $`\omega `$. On the other hand, the pumped charge does not depend on the pumping frequency, but only on the geometrical properties of the pumping cycle, and it reads
$$Q_\text{p}=4e_S𝑑g𝑑\alpha \frac{}{\phi }\text{Im}\left[\frac{\stackrel{~}{\psi }}{g}|\frac{\stackrel{~}{\psi }}{\alpha }\right].$$
(15)
Interestingly, the integrand of Eq. (15) turns out to be independent of $`\alpha `$.
We now consider the following sinusoidal pumping cycle: $`g(t)=\overline{g}+\mathrm{\Delta }g\mathrm{sin}(\omega t)`$ and $`\alpha (t)=\overline{\alpha }+\mathrm{\Delta }\alpha \mathrm{sin}(\omega t+\varphi _0)`$. In the weak pumping limit we assume that $`\mathrm{\Delta }g/\overline{g}1`$ so that the integrand of Eqs. (14) and (15) vary negligibly during the cycle. As far as the frequency is concerned, the maximum value of $`\mathrm{}\omega `$ in order for the adiabatic hypothesis to hold is $`\mathrm{}\omega <\mathrm{\Delta }_0\overline{ϵ}_0`$, with $`\overline{ϵ}_0=\mathrm{\Delta }_0\sqrt{1\overline{g}\mathrm{sin}^2(\phi /2)}`$. In order to compare $`Q_\text{p}`$ with $`Q_{\text{Jos}}`$, we choose $`\mathrm{}\omega =0.1(\mathrm{\Delta }_0\stackrel{~}{ϵ}_0)`$, $`\stackrel{~}{ϵ}_0`$ being equal to the value of $`\overline{ϵ}_0`$ at $`\phi =\pi /2`$. Note that the adiabatic condition breaks down for $`\phi =0`$ or $`\overline{g}=0`$, when the Andreev bound state is at the gap boundary. Thus in our analysis we shall avoid small values of those variables.
Figure 2 shows the pumped charge as a function of the superconducting phase difference $`\phi `$ for different values of $`\overline{g}`$. $`Q_\text{p}`$ is a non-monotonous function of $`\phi `$ exhibiting a maximum at $`\phi =\pi `$. For comparison, in the inset of Fig. 2, we plot the transferred charge due to the Josephson current, whose absolute value is larger, with respect to the pumped current, by a factor of order $`\mathrm{}\omega /\mathrm{\Delta }_0`$. However, while $`Q_\text{p}(\phi )`$ is an even function of $`\phi `$, $`Q_{\text{Jos}}(\phi )`$ is odd, so that a measure of $`[Q(\phi )+Q(\phi )]/2`$ will single out only the pumped contribution. The particular parity of $`Q_\text{p}`$ is due to the fact that a time-reversal operation implies not only the reversal of phase but also of the pumping trajectory in parameter space. It has also to be noticed (see Fig. 2) that the pumped charge is not quantized. In absence of Coulomb blockade, charge quantization occurs only for very specific pumping cycles (see, for example, Ref. makhlin01 ). In addition the global superconducting phase-coherence of the system, which leads to wave functions extending in the two superconducting electrodes, also hinders charge quantization.
To study the effect of an external magnetic field in the normal region, we change slightly the parametrization of $`S_0`$ introducing a phase factor $`\mathrm{exp}\pm i\beta `$ in the transmission amplitudes. As a result both the pumped and the Josephson charge are shifted along the $`\phi `$ axis by $`2\beta `$, i.e. $`Q_{\text{p/Jos}}(\phi )J_{\text{p/Jos}}(\phi 2\beta )`$. For example, the maximum pumped charge is now reached for $`\phi =2\beta \pi `$.
As far as detection is concerned, we wish to stress that for realistic parameters, using Al as superconductor, one can attain sizable pumped currents of the order of 5 nA. A sensitive setup to currents of even smaller size can be realized by inserting the SNS pump in a arm of a SQUID.
In conclusion, we have presented a formalism to study adiabatic charge pumping in a SNS weak link. The pumped charge is related to the Berry’s phase accumulated in a pumping cycle by the Andreev bound-state wavefunctions, which can be written as a function of the scattering matrix of the normal region. In the short junction limit, the pumped charge is even with respect to the superconducting phase difference. Hence, it can be easily distinguished from the charge transferred by the Josephson current.
###### Acknowledgements.
We acknowledge support from Institut Universitaire de France (F.W.J.H.) and from EC through grants EC-RTN Nano, EC-RTN Spintronics and EC-IST-SQUIBIT2 (M.G., F.T. and R.F.).
|
warning/0506/gr-qc0506088.html
|
ar5iv
|
text
|
# Glafka 2004: Spacetime topology from the tomographic histories approach I: Non-relativistic Case
## 1 Introduction with Motivational Remarks
In the standard formulation of relativity theory, the spacetime topology is a priori fixed by the theorist to that of a continuous manifold; hence, it is not an observable entity. Only the metric structure is traditionally supposed to be dynamically variable. With the exception of Wheeler’s celebrated, but largely heuristic, spacetime foam scenario , there is no well developed theory in which the spacetime topology can be regarded as a dynamical variable proper, with quantum traits built into the theory from the very start. However, one may try to consider idealized situations where certain topological features are represented as quantum variables that can in principle be observed and measured . Even in General Relativity (GR), where no variable quantity is supposed to be quantum—*i.e.*, subject to coherent quantum superpositions and associated uncertainty in its determinations, we need histories (*e.g.*, material particles’ causal geodesic trajectories) to actually define the topology of spacetime. This is because the concept of neighborhood turns out to be something which someone (:an observer), located at some point in spacetime, deduces for regions that belong to her causal past. Similarly, the concept of distance can be established only if information (:causal signals, or actual travelling material particles) is (causally) transmitted from one point to another. All in all, the causal nexus of the world determines both its topological and metric structures.
On the other hand, an interesting feature of quantum mechanics is that we may be able to make and verify statements about topology from a single-time case, as long as we are allowed to repeat experiments (and in principle we are allowed to do that indefinitely, if only in a theoretical, idealized, ‘gedanken’/theoretical fashion) in order to get the relative frequencies. In the classical (*i.e.*, non-quantum mechanical) case, one-time measurements do not give any information about global properties, such as the background topology.
Having said that, a remarkable consequence of quantum mechanics is that the wavefunction is a non-local entity, so that we may be in a position to deduce topological properties of the background, provided that we have enough repetitions of the experiment to reconstruct the relative frequencies. Thus, instead of saying that the wavefunction is a square integrable function on a topological space and use this to deduce probabilities about experimental outcomes (:events), we hereby propose to do the converse. *We start from probabilities and the continuity assumption for events, and from this information we derive the structure of the topological space in which these events are supposed to happen*. A word of caution is due here: the continuity assumption is normally taken to presuppose a topology—for how else can one talk about a *continuous* wavefunction? Well, and here is the crux of the inverse scenario: our assumption is that the wavefunction must be continuous with respect to the topology *to be deduced* from the relative frequencies of events. In other words, *the (continuity of the) wavefuction is born with the topology being deduced*.
In what follows, we do the same for the 4-dimensional case and recover ‘spacetime’. We should point out that no matter that we talk about space-time, we are still in the non-relativistic regime. We just speak of space points labelled by their ‘absolute’, Galilean time of occurrence. The relativistic case will be considered in a forthcoming paper .
One could say that histories are still needed to define global properties, such as the topology; however, here we maintain instead that they are needed in order to extract the form of the wavefunction. Of course, prima facie one can counter our arguments by holding that the assumption about (complete) knowledge of the wavefunction immediately leads to EPR-type of paradoxes. Our retort is that EPR-phenomena do not arise in our setting, while causality is rescued by the fact that we need classical communication to recover the full (:complete) state, as for example in various (quantum) teleportation scenarios—see, *e.g.*, Aharonov and Vaidman .
Let us outline the contents of the present paper. In the first part we introduce the consistent histories approach whereby we are given a configuration space for the system, its full Hamiltonian (including interactions), as well as the initial conditions (generally speaking, ‘exosystemic’ parameters of the problem traditionally supposed to be determined by an experimenter external to the experimentee—the physical system under experimental focus), and from these we calculate the probabilities for histories to occur. In our *inverse*alias, ‘*tomographic*’—approach, we are given the sets of observed histories together with their relative frequencies, and from these we reconstruct (some of) the parameters of the problem, with no allusion to external/internal systemic distinctions, as befits the histories approach. Then, certain issues about topology and the character of various possible indeterminacies of the derived topology that are involved in our approach are highlighted.
The main part of the paper follows, where we present *what* we are able to recover and *how* we do that. In this paper we specifically develop the *non-relativistic* case and focus on what can be said about topology using the set of histories alone, and also what needs some further measurements in order to be ‘sharply’ determined. Finally, we illustrate all this by virtue of two toy-models. The first is our variant of the usual double-slit experiment, both when the particle is detected at the slit, and when it is not. The second is an example of an environment involving a ‘bath of sensors’.
But before we delve into the paper, we feel that the new term ‘*tomography*’ ought to be further explained; otherwise, there is no reason to have it only for the sake of fancy neologisms and lexiplacy. We believe that its use can be justified on the following semantic grounds: experiments and their records may be thought of as ‘cuts’ (:‘$`\tau o\mu \stackrel{´}{ϵ}\varsigma `$ in Greek) incurred on the quantum system.<sup>4</sup><sup>4</sup>4Recall the Heisenberg ‘schnitts’ (German for ‘cuts’) in the standard Copenhagean quantum theory. From (the results of) these ‘observational measurement-slices’ and their relative frequencies of occurrence, we ‘retro-write’ (:‘redraw’, or ‘reconstruct retrodictorily’ so to speak)—as it were, ‘after the fact’—the (spacetime) topology. Moreover, in Greek, the verb ‘to write’ (or ‘to draw’, generically speaking) is $`\gamma \rho \stackrel{´}{\alpha }\varphi \omega `$. Hence ‘*tomo-graphy*’:<sup>5</sup><sup>5</sup>5In Greek, ‘$`\tau o\mu o`$-$`\gamma \rho \alpha \varphi \stackrel{´}{\iota }\alpha `$:=‘slice-wise writing/skethching/drawing’). we are re(tro)sketching spacetime topology from ‘experimental cuts’ exercised on the quantum system(!) All in all, this etymological dissection of the word ‘tomography’ accords with the title of the paper: “*spacetime topology (derived, or effectively re-sketched) from ‘tomographic’, inverse histories*”.
## 2 Histories and inverse histories approach
The decoherent histories approach to quantum mechanics deals with the kind of questions that may be asked about a closed system, without the assumption of wavefunction collapse (upon measurement). It tells us, in a non-instrumentalist way, under what conditions we may meaningfully talk about statements concerning histories of our system, by using ordinary logic. This approach was mainly developed by Gell-Mann and Hartle , and it was largely inspired by the original work of Griffiths and Omnès .
In this section, after we briefly recall useful rudiments of the standard histories scheme, we introduce its ‘converse’ theoretical scenario that interests us presently: *the inverse histories approach*. Pictorially, the two schemes are related as follows:
### 2.1 The HPO version of the standard histories approach
The formulation of the standard histories scenario that we follow presently is due to Isham *et al.* (*e.g.*, see ), and it is called HPO (History Projection Operator) approach. It consists of a space of histories $`𝒰𝒫`$, which is the space of all possible histories of the closed system in question, and a space of decoherence functionals $`𝒟`$. Parenthetically, the space of histories is usually assumed to be a tensor product of copies of the standard QM Hilbert space. Two histories are called disjoint, write $`\alpha \beta `$, if the realization of the one excludes the other. Two disjoint histories can be combined to form a third one $`\gamma =\alpha \beta `$ (for $`\alpha \beta `$). A complete set of histories is a set $`\{\alpha _i\}`$ such that $`\alpha _i\alpha _j(\alpha _i,\alpha _j,ij)`$, and $`\alpha _1\alpha _2\mathrm{}\alpha _i\mathrm{}=\mathrm{𝟏}`$
A decoherence functional is a complex valued function $`d:𝒰𝒫\times 𝒰𝒫`$ with the following properties:
* Hermiticity: $`d(\alpha ,\beta )=d^{}(\beta ,\alpha )`$
* Normalization: $`d(1,1)=1`$
* Positivity: $`d(\alpha ,\alpha )0`$
* Additivity: $`d(\alpha ,\beta \gamma )=d(\alpha ,\beta )+d(\alpha ,\gamma )`$ for any $`\beta \gamma `$
A complete set of histories $`\{\alpha _i\}`$ is said to obey the decoherence condition, *i.e.*, $`d(\alpha _i,\alpha _j)=\delta _{ij}p\left(\alpha _i\right)`$ while $`p\left(\alpha _i\right)`$ is interpreted as the probability for that history to occur *within the context of this complete set*.
The decoherence functional encodes the initial condition as well as the evolution of the system. Here we should also note that the topology of the space-time is presupposed when we group histories into complete sets, *i.e.*, in collections of partitions of unity.
In standard QM, histories correspond to time ordered strings of projections and to combination of these when they are disjoint. An important issue here is the relation between decoherence and records. Namely, it can be shown that if a set of histories decoheres, there exists a set of projection operators on the final time that are perfectly correlated with these histories and vice versa.<sup>6</sup><sup>6</sup>6This is the case for a *pure* initial state, and we restrict ourselves to it. These projections are called records. It is this concept that figures mainly in our approach (*e.g.*, see Halliwell ).
To sum things up, in the standard histories approach
* The system is given, as well as its environment. The latter is represented by prescribing initial conditions and in some cases final conditions.
* The space, its topological structure in particular, is presupposed.
* The interactions are given in terms of the decoherence functional, which encodes the dynamical information. For the complete dynamics, the full Hamiltonian must be known.
### 2.2 Tomographic histories approach
In our approach things are different, as we solve the *inverse* problem. While in standard histories one is given the Hamiltonian, initial conditions, as well as the space on which they are defined, and the aim is to predict probabilities for histories, we do the opposite thing. We make repetitions to get the frequencies for different records. Then, by making certain assumptions about these records, namely, that they are nothing but records of events, we recover the topological structure of the underlying configuration space. This means that from a set of events, with no other structure presupposed (:a priori imposed from outside the system), we end up with a causal set representing the discretized version of the *extended configuration space* of the system in question.
The extended configuration space that we get will be an ‘*effective*’ one, and in a sense it accounts for certain properties of the Hamiltonian, such as interactions with other objects not controlled by the experimenter. For instance, the latter could be some kind of ‘repulsive’ field that prohibits the system to go somewhere (:in a region of its configuration space), which can then be recovered as a hole (:a dynamically inaccessible region) in that space.
To compare the two approaches, let us review for a moment the standard histories approach where the decoherence functional, as well as the space of histories, are given. For these, one is assumed to be given the initial conditions, the configuration space, and the Hamiltonian of the system in focus—*i.e.*, generally speaking, the parameters of the system. When we are able to perform multiple runs of the experiment and we choose a decohering set of histories, the decoherence functional yields the probability for each history to occur, which, in turn, corresponds to the history’s relative frequency with respect to the set chosen.
Having the same Hamiltonian and the same initial conditions, we may consider another decohering set of histories, not necessarily compatible with the previous one, for which again the probabilities can be calculated. This is in broad terms what the usual histories approach accomplishes.
We on the other hand will be tackling the inverse problem. The essence of our approach is the following. Since we can carry out our experiment sufficiently many times, we have access to the following two things—the set of possible histories and the relative frequencies for each history to occur for every initial state. From this we recover the parameters of the experiment, namely, the effective topology of the extended configuration space.
One thing to highlight here is what corresponds to a decoherent set of histories in our inverse scenario. It is one particular partition of unity of the record space. Our freedom of choosing a particular basis in which to measure things will in general give different decoherent sets than had we chosen a different one (:different basis, different decoherent sets). Note also that since we consider histories *operationalistically*, we always deal with histories that are contained in a decoherent set, namely, the set that corresponds to the set of records that we choose to analyze.
In our setup we shall assume that *the records capture the spatio-temporal properties* (of the system in focus). This means that the histories are coarse-grained trajectories of the system, belonging to a space whose topological properties we ultimately wish to deduce. We shall then claim that the whole concept of spacetime, as a background structure, does not make sense in finer-grained situations. In this way, all the histories are single-valued on our discretized version of ‘effective spacetime’. One should note here that we may still have histories that have the particle in a superposition of different position eigenstates, but only if the latter are ‘finer’ than the degree of our coarse-graining. With the coarse-graining we effectively identify (*i.e.*, we group into an ‘equivalence class’ of some sort) the points that we cannot distinguish operationally, with the resulting equivalence class of ‘*operationally indistinguishable points*’ corresponding to a ‘blown up’, ‘fat point’ in our discretized version of ‘effective spacetime’.
### 2.3 Classical versus quantum indeterminacy of topology
In this subsection we would like to emphasize that there are two essentially different kinds of indeterminacy involved in derivations of the effective topology.
The first one is of a ‘classical’ character, that is, it comes from the lack of *our* knowledge about the systems’ configuration space, as for instance when we do not have sufficiently many repetitions of the experiment. For example, the configuration space might appear to be a segment of a straight line, when in fact it is a circle. This could be due to incomplete information that we gather from an insufficiently repeated experiment, which could result to some points at the end of the segment, that would ultimately make the configuration space a circle, not to be detected. Another way that classical indeterminacy could arise would be when some records are simply not accessible<sup>7</sup><sup>7</sup>7This is not the case in this paper. In our setup we assume that we have access to all records that are related to detectable events. when, as a matter of fact, the interaction of our system with the ‘record space’ is supposed to capture all the spatiotemporal features or properties of the system. In toto, as befits the epithet ‘classical’, this type of indeterminacy in effective topology determinations is an ‘*epistemic*’ one: it reflects *our* ignorance, our partial experimental knowledge about and control over the quantum system.
The second type of indeterminacy, like the one arising in Quantum Theory, is due to a fundamental ‘quantum dichotomy’ of our experimental settings and determinations. For instance, the topology of coordinate and momentum space of a quantum particle may be different from each other, so that what we recover in the end depends on what we initially choose to measure: coordinates or momenta. Plainly, this reflects the fundamental quantum duality between the position and momentum observables in standard QM, which in turn is a reflection of the *ontological* (as opposed to epistemic) nature of quantum indeterminacy and uncertainty. Our setup simply limits our freedom to measure anything we want to what is produced by a decoherent set of histories, and therefore it is associated with a projection operator on our record space. We must emphasize however that we still have some freedom, since incompatible consistent sets have incompatible records in the record space, so that our choice of what basis to measure in the record space is still in force. This issue is addressed in more detail in section 3.3.
### 2.4 The operationalistic underpinnings of our scenario
Our approach is essentially *operationalistic*. The notion of *record space* is regarded as the only source of information we possess about the system we wish to explore. The effective topology then refers to the configuration space of the system in question. In our tomographic approach, we are given the sets of observed histories together with their relative frequencies, from which then we reconstruct the parameters of the problem.
We assume that some of the records may be identified with particular events, *i.e.*, spacetime ‘points’. Furthermore, we claim that this is the only case we may speak of a configuration space proper. That is, if we do *not* have access to events even in principle, we *cannot* speak about their support or their topological and causal nexus, as, say, in the causal set scenario (causet). Then, relative frequencies are recovered by repetition of the whole histories involved: by restarting the system in an identical environment and letting it evolve for the same amount of time.<sup>8</sup><sup>8</sup>8From our vantage, ‘*history could in principle repeat itself*’ (pun intended). In our operationalistic (ultimately, relational-algebraic) view, the only way one can talk about some background structure such as ‘spacetime’, is relative to something else. More precisely, we use our data (records) to (re)construct an ‘arena’ for a particular subsystem of the universe that we are interested in, and it is *only* in this sense that we may speak of ‘spacetime’. Retrodictorily, *‘spacetime’ is where and when ‘it’ must have happened, if we judge by our records, and the latter are the only data we have got*. Thus, philologically speaking, ‘*quantum tomography is spacetime archaeology*’.
More precisely, we have a system (call it ‘particle’), which is placed into an appropriate experimental environment, and we are able to
* Repeat the experiment with *the same* initial conditions. In this way we get the relative frequencies of the records.
* Vary the initial conditions of the system in question, leaving all the environment (and records) the same. For each initial condition of the system, we rerun the experiment. These first two steps give us the set of all possible histories (coarse-grained trajectories) of the particle, as well as their relative frequencies.
Another basic ingredient is the space of records. It is a space of data resulting from controlled environment tampering with the system, and it is supposed to capture its spatiotemporal properties. Records are interpreted as *distinguishable events*. That is to say,
* We can distinguish them spatiotemporally. Although we do not know the structure of the set of records that corresponds to events, we can identify each record corresponding to a spacetime point as being different from the others. Thus, while we know nothing a priori about their causal or spatial (topological) ordering, events can be labelled so that we do not have identification problems. For instance, we may consider photons of different frequencies, each frequency mode coming from one point. In the examples to follow this will become more transparent.
* We can vary each record corresponding to a particular event independently. The variation is in some sense small—this may be effectuated by a ‘small energy’ variation of the record. The latter is assumed to be small enough not to affect the ‘topology’ of the records (*i.e.*, neighborhoods in the set of records remain the same). By ‘topology’ we mean a reticular structure associated with appropriate coarse-graining of a region of the extended configuration space we explore. The said variations give us the proximity relations between events.
Experiments are carried out repeatedly and multiply. We label the runs by initial conditions of the system, number of run and ‘positions’ of events.<sup>9</sup><sup>9</sup>9By this we mean whether or not we varied one record corresponding to an event. Each run gives us a history, *i.e.*, a sequence of causally related events. Note here that for the same initial conditions of the system, the different histories group together to form decoherent sets.
To conclude, from our experiments we get the following information:
1. The set of histories of the system associated with a fixed set of initial conditions. We call this set of histories fiducial set. Here we emphasize that these correspond to coarse-grained ‘trajectories’.<sup>10</sup><sup>10</sup>10The inverted commas are added to the word ‘trajectories’, since the space on which they are defined is not presupposed. We define the set of all histories to be $`𝒞`$, while each history that is contained in it is denoted by $`C_i`$. We therefore obtain the set $`𝒞`$ as well as the set $`𝒫`$ which is the set of all possible events, or else the set of ‘spacetime’ points<sup>11</sup><sup>11</sup>11We remind here that we just speak of space points labelled by their Galilean time of occurrence.
2. The relative frequencies of outcome of these histories depending on the initial conditions. This is a function $`f_j:𝒞[0,1]`$ which gives the (normalized) relative frequency of histories for each particular initial condition (corresponding to $`j^{th}`$ initial state of the system).
3. The change in the relative frequencies when one event is varied. This is a function $`f_j^p:𝒞[0,1]`$ which is the *new* relative frequencies when the event $`p`$ has been varied. This will lead us to the proximity relation between the points produced by the fiducial set of histories.
It is important to note that *before* we vary the records, we already have the fiducial set of histories. It provides us the set on which the topology is imposed.
## 3 Non-relativistic case
We reconstruct the *effective* topology of the extended configuration space. But let us explain what we do in a bit more detail.
#### Effective versus ‘real’ topology.
In our approach, we consider the effective topology which we derive from our observations. That means the following. Believing in Einstein’s theory, we posit that the physical processes take place in spacetime, which is a topological space with certain *‘real’* topology. However, there is no way for us to measure this ‘real’ topology exactly. That is why we are speaking of *effective topology*—the topology of a model of configuration space which accords with our experiments and fits their outcomes.
An important issue should be emphasized at this point. Suppose we have derived a non-trivial topology for the configuration space—say for instance that it has a defect, such as a hole. This indicates to us merely that we have non-contractible loops, nothing more. Why these loops fail to be contractible—due to the existence of a ‘real hole’, or because of, say, the presence of a potential barrier—such a question is, as a matter of principle, not verifiable within our approach.
As a consequence, we may admit transitions between 3-dimensional surfaces of different number of components (with respect to the effective topology), without regarding this as being unphysical.
#### The record space.
As noted before, we rely solely on operationalistic means to recover the effective topology. In turn, this means that we are able to control the preparation of the initial state (see section 2.4) and then read out the observation which is carried out by a specified device. The state space of this device we shall call record space. Here it should be pointed out that we assume certain things about this record space. In the case of the examples in section 4, we specify the main features of the interaction Hamiltonian of the system with the record. More generally, we need only to assume that it captures the spatiotemporal properties of the system and therefore that it leaves records of events. Other records, outside our record space, may exist and they specify other features of the particle, such as its spin or electrical charge. If some records of the spatiotemporal properties are elsewhere then we may end up with an incomplete topology reflecting the classical indeterminacy mentioned earlier.
In closing this subsection we should also stress that since the device is anyway a quantum system, reading out the records causes some loss of information about the system in focus. Moreover, our choice of what to read out may also affect the resulting topology, which is related to the aforementioned ontological quantum indeterminacy.
### 3.1 Extended configuration space and algebraic considerations
We have a classical or quantum physical system, and we observe it for a period of time $`(t_0,t_1)`$. If $`𝐌`$ is the configuration space of the system, then the Cartesian product
$$=𝐌\times (t_0,t_1)$$
(1)
is the *extended configuration space*. Moreover, we also take into account a more general situation in which the topology of the configuration space $`𝐌`$ may change in time and the extended configuration space $``$ is no longer decomposable into a product like (1).
Assume for a moment that the configuration space is at all times connected. This is not a trivial statement, as we are talking of ‘effective extended configuration space’ which in principle allows for transitions from connected to non-connected subsets in different moments of time. By considering $`𝒞`$, the set of all histories, we may deduce the spatial slices as the subsets $`S_i`$ of points no pair of which is contained in the same history (trajectory).
$$p,qS_i\mathrm{}C_j𝒞p,qC_j$$
(2)
Moreover, we regard ‘maximal’ slices as being the ‘time-slices’, *i.e.*, any extension of the spatial surface will move the subset outside the class of spatial surfaces.
$$\mathrm{}r𝒫rS_i=S_j$$
(3)
Note here that the relation indicating that two points do not belong to the same history is *transitive* in the case we have only one component. It should also be noted that we cannot determine the order of the slices merely from the set of histories (*i.e.*, without varying the records), neither can we deduce any other topological feature within each of these slices. Thankfully, the latter is not the case in the relativistic situation since the upper bound in the speed of transmission of information leads to a notion of proximity in each spatial surface. This will be explored in a later publication .
Returning to the general case, in which transitions from connected to non-connected spaces are allowed, the above procedure will produce ambiguities. Two ‘events’ could never be contained in the same history due to the fact that they are in separate connected components and *not* because they ‘occur’ at the same time. Trying then to form maximal subsets of $`𝒫`$ that are not contained pairwise to any history, will not lead to a unique partitioning of the set of ‘spacetime’ points . This is due to the fact that the property of two points not belonging to the same history is not transitive anymore.
#### An example of ambiguity in partitioning:
We have a non-connected space. Say we have two boxes separated by a rigid partition (*e.g.*, an infinite potential barrier). The thick line in the graph represents the partition. Apart from the obstructing partition, all histories are allowed which do not contain points of the same ‘horizontal’ line corresponding to ‘same time’. If the particle is in one time in point $`a_1`$ at the left side of the partition, then it can never be in any of the points on the right hand side of the partition, as *e.g.*, point $`b_5`$. This, according to the previous definition of ‘time-slice’, means that $`a_1`$ is in the same slice with all the points on the right hand side of the partition no matter which instant they are measured at.
The latter would lead to contradiction, since clearly $`a_5`$ and $`b_5`$ are not in the same time-slice as there is a history joining them. If we stick to the proper definition of ‘time-slice’, *i.e.*, a maximal set of points pairwise not belonging to the same history, we will end up having point $`a_1`$ in one of the following ‘slices’: ($`a_1,a_2,a_3,a_4,a_5,a_6`$), or ($`a_1,a_2,a_3,b_4,b_5,b_6`$), or, finally ($`a_1,a_2,a_3,c_4,c_5,c_6`$). Any of these obey the definition of ‘spatial-slice’; therefore, just from the set of histories we will end up with some ambiguity about what a spatial-slice or a ‘moment of time’ is.
In general, this would not be a problem since we could consider each component separately. But, in our effective set up we may have the two disconnected components becoming connected in the future. For example, the separation was made from ice and it melted (or from an unstable radioactive substance which quickly decomposed!). An example of this situation will be examined later.
We could therefore already make one non-trivial statement about the topology, just by considering the set of histories. Namely, that if there is a unique way of ‘foliating’ the points of the ‘spacetime’ into slices, the space is connected. Furthermore, we will be able to determine the number of different components of the ‘4-dimensional’ configuration space, and on top of this, the number of components of one particular ‘spatial’ surface.
### 3.2 Extracting connected components
Having the set of decoherent histories, we can already extract some information about the effective topology. Let us first show how connected components are detected. In order to do this, recall that, given a connected component $`K`$ of a topological space $`X`$, the relation $`a\sigma b:=\{a,bK\}`$ is an equivalence relation on $`X`$.
#### 4-dimensional connectedness.
In our setup, we are given the relation $`aHb:=\{C𝒞a,bC\}`$, which means that there exists a history containing both $`a`$ and $`b`$. The relation $``$ is an equivalence, *i.e.*, a symmetric, reflexive and transitive relation on $`X`$. However, the relation $`H`$ is symmetric and reflexive, but not transitive. Thus, the relation $`\sigma `$ can be obtained as the transitive closure of the relation $`H`$. In general, finding the transitive closure is an infinite operation; however, here we deal with histories containing a finite number of events, hence the transitive closure can be delimited in a finite number of steps.
A possible algorithm to find the transitive closure can be devised using Boolean matrix machinery . Namely, we can define the relation $`H`$ by its Boolean matrix (denote it by the same symbol $`H`$), then $`\sigma `$—the transitive closure of $`H`$—is obtained as a Boolean matrix power $`H^{|A|}`$ of $`H`$, where $`|A|`$ is the number of antichains. So, effectively the procedure of extracting connected components goes as follows:
* Form the Boolean matrix of the relation $`H`$ ‘to belong to the same history’
$$aHb:=\{C𝒞a,bC\}$$
* Calculate its $`|E|`$’s power using Boolean arithmetics rather than $``$ and $``$:
$$\sigma =H^{|E|}$$
Recall that the Boolean operations has the following rules: $`1+0=0+1=1+1=1`$, $`0+0=0`$, $`11=1`$, $`10=01=00=0`$.
* The resulting matrix $`\sigma `$ is always block-diagonal and the blocks of entries are in 1–1 correspondence with the connected components of the space $`E`$ of events.
#### Components of a spatial surface.
The procedure just described would account for the number of components our ‘4-dimensional’ configuration space has. Note that, since we speak of ‘effective configuration space’, we may as well have transitions, in some particular time, from a number of components to another. It would then be of interest to consider the number of components a spatial surface has.
To this end we should point out that there is some ambiguity about what a spatial surface is, thus this ambiguity will also be present in the considerations to follow.
* We let $`S_i`$ be a spatial surface. $`p𝒫\backslash S_i`$, we consider the following:
$`\{S_i^pS_iqS_i^pC_j𝒞p,qC_j\}`$
* We will then end up with a family of subsets of $`S_i`$, call it $`𝒮_{s_i}`$. Note that some of these will be identical, while others may contain others. We declare them ‘open’.
* From this family we generate a topology by taking arbitrary unions and finite intersections of the subsets. The resulting topology is denoted by $`𝒯_{s_i}`$.
* We then consider a sub-selection of the open subsets of $`𝒯_{s_i}`$ such that:
1. It covers all $`S_i`$, *i.e.*, their union is $`S_i`$.
2. They are disjoint.
3. They are ‘minimal’: that is, they contain the smallest of the subsets in the family $`𝒯_{s_i}`$.
This is a disjoint open covering of $`S_i`$ that is also a basis for the topology $`𝒯_{s_i}`$.
* Finally, each of these subsets corresponds to one component of the spatial surface $`S_i`$.
To clarify things, and without wishing to repeat ourselves, we describe the above in words. We chose the surface in question. Then, for each point in space we see which part of the surface is causally connected to it. Then we pick the smallest family of subsets of the surface that covers the surface. Since the separate components do not overlap, we need to secure that this family is also disjoint. That is why we need to generate a family bigger than $`𝒮_{s_i}`$, namely, $`𝒯_{s_i}`$, while from this we are guaranteed to have a basis that consists of the relevant components, which basis would a fortiori be a disjoint covering.
#### Illustrative example of recovering the components of a spatial surface:
Here the space is connected when seen 4-dimensionally. The partitions that exist forbid for example a history containing $`a_5`$ and $`b_3`$ (the lower one), or $`a_5`$ and $`c_1`$ (the higher partition). Note that even without the ‘d-column’, the space is connected when viewed ‘4-dimensionally’, as the ‘transitive closure’ of any point is the set itself.
Now we follow the steps described above. We pick the spatial slice that corresponds to the b-horizontal line ($`b_1,b_2\mathrm{},b_6`$), and we are looking for its components.
First we consider the set $`𝒮_{s_i}`$, which in this case is the set containing the following subsets: $`\{(b_1,b_2),(b_1,b_2,b_3,b_4),(b_3,b_4,b_5,b_6),(b_5,b_6),(b_1,b_2,b_3,b_4,b_5,b_6)\}`$. Note that the subset $`(b_3,b_4)`$ does not belong to $`𝒮_{s_i}`$. The result we would like to have is that there are three components, namely, $`\{(b_1,b_2),(b_3,b_4),(b_5,b_6)\}`$. To obtain this, we have to follow section 3.2 and extend $`𝒮_{s_i}`$ to $`𝒯_{s_i}`$, which is the topology induced by $`𝒮_{s_i}`$ if we consider unions and intersections. In the latter, the subset $`(b_3,b_4)`$ is also included as it is the intersection of $`(b_1,b_2,b_3,b_4)`$ and $`(b_3,b_4,b_5,b_6)`$.
We then need to pick a sub-selection of the elements of $`𝒯_{s_i}`$ that is disjoint and covers the surface (*i.e.*, the horizontal $`b`$). There are two possible choices: either $`\{(b_1,b_2),(b_3,b_4),(b_5,b_6)\}`$, or $`\{(b_1,b_2,b_3,b_4,b_5,b_6)\}`$. The second is not ‘minimal’, *i.e.*, it does not contain the smallest sets and therefore it is not a basis for the topology $`𝒯_{s_i}`$. Finally, we are left with $`\{(b_1,b_2),(b_3,b_4),(b_5,b_6)\}`$, which is the desired result.
A final note just to mention that the above discussion is liable to ambiguities that come from the fact that there is not a unique definition of spatial surface. Instead of the $`b`$-horizontal as a surface, we could have taken as spatial surface for example the subset $`\{(c_1,c_2,b_3,b_4,b_5,b_6)\}`$, and we would end up with similar results.
### 3.3 Reconstruction of topology—statistical approach
As mentioned earlier, different decohering sets of histories may lead us to different effective topologies. It follows that the effective topology is a result of our measurements. We could claim that our system is in a superposition of different effective ‘spacetimes’<sup>12</sup><sup>12</sup>12Here we still assume that we are in the non-relativistic regime. and our choice of measurement causes a ‘reduction’ to one particular (or to a particular subspace of all the possible) ‘spacetime’. In the description above we carry out a measurement in record space on the ‘basis’ that is related with spacetime points, *i.e.*, events. If this assumption is not satisfied, the actual choice of our measurements would affect the resulting topology. It should be pointed out here that this is the generic case, since we cannot have full knowledge about whether or not our records capture only configuration space properties and not, possibly incompatible, momentum space as well. On the other hand, if our measurements are sufficiently coarse, we could have compatible ‘position’ and ‘momentum’ measurements.
Now we are in a position to address how to recover topology assuming that we can vary slightly one event independently from the others, and repeat the runs of the experiment. The result of such variations will be certain changes of the relative frequencies, that is why we call this process statistical reconstruction of topology. This procedure fixes the ambiguities about the ‘time-slices’ that existed due to the non-connected spatial surfaces, as well as the order of these slices.
* We have the relative frequencies, $`f_j(C_i)`$ of each history $`C_i`$ with initial condition‘$`j`$’.
* We vary slightly one event say event $`p𝒫`$ and repeat the procedure to get the new relative frequencies of histories $`f_j^p(C_i)`$. It is important to note that, provided the variation is small, the set of histories is the same and only their relative frequencies change. Therefore, all the considerations that were already made from the mere set of histories still apply galore.
By observing which histories have changed their frequencies compared to the undisturbed event case, we can deduce a few things—for starters, some notion of closeness (or proximity). The histories whose frequencies are significantly affected by the perturbation are in some sense ‘close’.
* We consider *each initial condition separately*<sup>13</sup><sup>13</sup>13This to avoid problems related with the following. Assume that we vary a point $`a`$ in a way that it has the same distance with one neighboring point $`b`$. Then the overall probability of the $`b`$ due to symmetry will be invariant, but depending on which is the initial condition of the system the probabilities of some histories will increase while other will decrease with a net probability unchanged. In this way we would fail to recognize $`b`$ as a neighbor of $`a`$..For each initial condition we see the probabilities of which histories alter significantly.
* After we vary the ‘event’, we repeat the experiment exactly with same initial condition as before and then, by considering the change in relative frequency of events (not of histories), we can deduce which events are neighbors (call them $`j`$-neighbors, whereby the label ‘$`j`$’ stands for the initial condition we consider). We repeat this for all possible initial conditions. We then consider the union of all these $`j`$-neighborhoods to get the total neighborhood of the point we varied.
In other words we take a small positive number $`ϵ1`$. We define another function, the difference function, as follows:
$$\delta _j^p:𝒞[0,1]:f_j(C_i)f_j^p(C_i)$$
(4)
We then consider all the points belonging to the histories $`C_i𝒞`$ that
$`\delta _j^p(C_i)>ϵ`$. We name them j-neighbors of $`p`$. So we have:
$$qN_j^pqC_i,C_i𝒞\delta _j^p(C_i)>ϵ$$
We then consider different initial conditions ‘$`j`$’ and we group all the neighbors together to form the neighbors of ‘$`p`$’ , $`N^p`$.
$$qN^pjqN_j^p$$
* We already know which of these neighbors are (definitely) not in the same time-slice (:those that both belong to at least one history), and we can coin them ‘temporal neighbors’. Events being in different path-connected components will never affect each other. Note here that the neighbors that will be affected, and are definitely not in the same time, are only to the future of the event in question. Thus, properly speaking, we should talk about ‘future temporal neighbors’. With these in hand, we may get the order of the histories.<sup>14</sup><sup>14</sup>14Note that since we get a direction from the fact that only the ‘future’ neighbors are affected, helps us recover the order with no doubt about the overall direction.
* Then we mark the events that are neighbors, but not *temporal* neighbors, as ‘spatial-neighbors’, and use them to define proximity in the ‘time-slice’ in focus.
So we define *spatial neighborhood* of ‘$`p`$’ to be:
$$SN^pq[N^p_iC_i],pC_ii$$
* We repeat this procedure varying slightly one by one all the ‘events’.
* From the proximity we deduce the topology of each time slice in the usual way—*e.g.*, as it is done in metric spaces.
* We will have obtained the topology of each spatial components. We can then choose an arbitrary partitioning of these slices to get the total 4-dimensional case. We then check that we do not have contradiction.This contradiction could be due to, for example, some event being affected by a change in an event to its future rather than to its past (:‘advanced’ and ‘retarded’ contradiction, respectively).If a contradiction arises, we pick another ‘partitioning’, so on and so forth, until the correct one is obtained.
* By patching all the slices together, we recover the topology of our ‘spacetime’, or more precisely, of its reticular substitute .
Alternatively, we may consider the closest neighbors to define a cover of each time-slice, and then find the finitary substitute of the underlying continuous topology. To find only the closest neighbors, we need to ‘tune’ the parameter ‘$`ϵ`$’ to be sufficiently big so that it gives only the number of closest neighbors we want (4 to get a 3-dimensional space in the triangulations scheme). Along these lines, we first get a *prebase* from which the topology is unambiguously reconstructed. We should note here that the above construction makes ‘heavy’ use of the relative frequencies, not only of the set of histories. It effectively uses the former to define neighborhoods.
## 4 Toy models
### 4.1 Double-slit experiment
We consider a discrete version of the registration screen. This means that our data will be a discrete distribution of registered events, each ‘column’ being discretely labelled. We consider the case that we do not detect which slit the particle passes through, as well as the case that we do. In both cases we have always the same initial conditions—a particle is emitted far away from a barrier bearing two holes. Note also that the particle in question is assumed not to be a photon, so that it can be detected on the slit without being absorbed.
#### Case I: Not detected on the slit.
The particle passes through the slit. Then it is absorbed by a film so that we can identify different events by measuring the position of the excited grain on the film. Note that we need only distinguish the different events and not their actual position. So, somebody could have cut the film and glued it back with different order (but the same for all the repetitions). The correspondence between this gedanken-experimental scenario and our theoretical scenario above is the following:
* single experiment—emitting one particle and registering it
* the record space—the real line (position-loci of registered events)
* a particular history—an event
To recover the configuration space, we (a) assume continuity of the distribution, and (b) move slightly one point of registration on the film. By observing the probabilities of the events that are significantly altered, we define the neighborhood of this ‘point’ (:proximity neighborhood).
We recover several segments of a line representing the configuration space. The fact that it is not the whole real line that is being recovered, is due to the fact that there exist dark fringes, *i.e.*, regions where the particle is never detected.
#### Case II: Detected on the slit.
The particle passes through the slit, and a photon, whose frequency depends on which slit the particle passes through, is emitted. This happens because we have an oscillator of different frequency on each slit, and when a particle passes, the oscillator increases its energy level. Then, upon relaxing back to its ground state, it emits a photon. Then the particle is absorbed by the discrete screen. The correspondence between this gedanken-experimental scenario and our theoretical scenario above is the following:
* single experiment—emitting one particle, and subsequently registering it as well as the photon carrying information about which slit the particle passed.
* the record space—the real line (position-loci of registered events) and the detector of the photon (or the oscillators). Note here that we can distinguish all events from each other, but not know anything else about their topological structure.
* a particular history—a photon with frequency depending on which slit the particle passes through, followed by a position on the discrete screen line.
To recover the configuration space, we (a) assume continuity of the distribution, and (b) we move slightly one point of registration on the film, or one of the oscillators. As usual, we recover neighborhoods by small variations of the established ‘records’.
#### What is eventually recovered.
Two points separated from one another, and at a later time<sup>15</sup><sup>15</sup>15The order of the events is recovered, due to the fact that varying events to the past and only affects the relative frequencies of the future and NOT visa-versa. a segment of straight line (actually a syncopated version of it). Note that the line does not decay into disjoint segments, as we have no interference and therefore there are no ‘dark fringes’.
### 4.2 Bath of sensors
Here we consider a thought experiment that better illustrates the foregoing ideas. We have a closed box, and in it there are many (say, $`n`$) different oscillators, all of different frequency. We require this to be able to distinguish our ‘points’, but note that we do not know anything about their structure. We inject a particle into the box that has the following property: when it is sufficiently close to one oscillator, the oscillator increases its energy level. At the final time when we measure things, the oscillators will relax to their ground state, emitting one photon of the same frequency as the oscillator that had been excited.
The only other thing we need to assume is that somehow the signals (photons records) emitted from each oscillator can be distinguished from those emitted by the same oscillator at a different time (we need that to have spatio-temporal labels). This may be done by having, for example, a moving film around the box, and earlier or later signals from the same oscillator would be identified by different positions on the film. The fact that each oscillator may have significantly different ‘half-life’ before it relaxes, means that signals from different times may be confused. The important point here is that we only care about the order of photons coming from the same spatial event, since others can be distinguished by their different frequency. Finally, we will end up with a set of different records corresponding to different events, and all of them will be spatio-temporally distinguishable.
We then infuse the particle into the box, with different momentum and from different points. Each repetition (with the same initial condition) gives one possible history. We do it many times for each initial setup, and we record the results in our data-sheets (experimental protocols). After this is accomplished, we obtain the set of possible histories and their relative frequencies. This is sufficient, in our non-relativistic case, for deriving the number of different components each spatial slice has.
We may then vary slightly each oscillator separately, and repeat the experiment. By this, we will obtain information needed to recover the topology . In this setting we have the following correspondences with our theoretical scheme:
* single experiment—emitting the particle in the box with some given initial condition, and with a specific setup of the oscillators, and then record at the final time the photons having been emitted (maybe read them out directly from the moving film).
* the record space—photons of different frequency on different positions on the registering film.
* a particular history—a collection of photons of different frequencies (possibly also of different positions on the film, if a moving film is required for distinguishing events in time).
To recover the configuration space, we do the following:
We assume continuity of the distribution and move slightly one oscillator at a particular ‘time’. By noting the probabilities of which events are significantly altered, we define the neighborhood of this ‘point’. Here we need to take into account the union of all the neighborhoods related to all the possible initial conditions. We then specify the ‘temporal’ and ‘spatial’ neighborhoods. We repeat this procedure varying slightly all the ‘events’ one by one. Using the proximity relation, we deduce the topology of this slice, as was described in section 3.3.
By patching all the slices together, we recover the topology of our ‘spacetime’.
#### What is eventually recovered.
We get the effective topology of the interior of the box. This includes other objects that were not known to be there, as well as their time evolution. So if for example there was a cube of ice in the box that melted, this will be represented by a cubic hole in the configuration space that gradually changed shape to become flat.
## 5 Conclusions
Let us summarize what we have done. We have a laboratory in which we explore a physical system whose configuration space is unknown. We are able to run the experiments sufficiently many times, either by leaving the initial conditions unchanged, or by varying them. We also have another physical system, whose configuration space is coined record space. As a result of each run of the experiment, the record space acquires a state (a *quantum* state, in general). In each run, we perform a measurement over the record space. Which measurement in particular, this is a matter of our choice.
After multiple runs, we have a set of protocols (data-sheets). Each protocol tells us which events occurred within a particular experiment. This set of events is referred to as a history. When the initial conditions remain unchanged, the arising set of histories is treated as a decohering set.
Initially, as a result of our observations, we have histories and, in addition, their relative frequencies. This primary set of histories we call fiducial set.
From the fiducial set, we deduce the number of components of our ‘spacetime’ (extended configuration space) as well as the number of components in each ‘spatial’ surface (i.e. moment of time).
We then allow for variations of the records. This yields new histories which make it possible to deduce proximity on the fiducial set and hence the topological properties of the ‘spacetime’.
As a result, we reconstruct the *effective* topology of the ‘spacetime’ region involved in our observations. ‘Effective’ means that we can say nothing about the ‘true’ topology, and that all our statements are consequences of our observations. The working definition of configuration space that we employ is the following. Configuration space is the space of all possible configurations of our system.
The topology we recover—the ‘effective’ one—may include holes and other topological features that result from existing ‘potentials’ that we do not vary. What could be referred to as the ‘true’ topology would be something that takes into account only the background manifold. In this sense it would be like saying that we may vary not only the initial state of the system in question, but the initial state of any potential except the gravitational (which is supposed to account for the ‘geometry’ of the background manifold).
We emphasize once again that we recover histories *operationalistically*. The record space is *the only* source of information we possess about the system we explore. The effective topology is then regarded as the ‘best possible’ (:as realistic, or as pragmatic a) picture of the actual configuration space of the system in focus as one can acquire from her ‘experimental intercourse’ with it.
Last but not least, some loose, anticipatory connections with the forthcoming paper are due here. In the latter, we develop the relativistic version of our ‘topology-from-inverse histories’ theoretical scheme. This essentially means that, due to a physical upper bound in the transmission/propagation of (material) signals, one is forced to focus more on recovering the causal topology of spacetime from ‘inverse causal histories’, rather than on just recovering the topology of ‘frozen, absolute, fat spatial slices’ (*i.e.*, merely of ‘space’) as we did presently.
#### Acknowledgments.
We are grateful to Charis Anastopoulos, Jonathan Halliwell and Serguei Krasnikov, for their interest in our work and for related feedback, as well as to Chris Isham, for reading an early draft and making many critical comments and useful suggestions. The basic ideas underlying this work were originally conceived shortly prior to and during the ‘Glafka-2004: Iconoclastic Approaches to Quantum Gravity’ meeting in Athens (Greece), generously sponsored by Qualco Technologies (c/o Dr Orestis Tsakalotos, Qualco’s CEO) and by a European Commission Reintegration Grant (ERG-505432) awarded to the first author. An outline of those preliminary ideas was presented briefly at the said meeting by the third author (RRZ). Subsequently, a major part of the paper was written under the aegis of the program *Tête-à-tête in St.Petersburg*, supported by the Euler Mathematical Institute of the Russian Academy of Sciences. RRZ acknowledges the support from the research grant No. 04-06-80215a from RFFI (Russian Basic Research Foundation). PW acknowledges partial funding from the Leventis Foundation. Finally, IR acknowledges financial support from the EC via the aforementioned grant.
|
warning/0506/quant-ph0506049.html
|
ar5iv
|
text
|
# Entanglement, Purity, and Information Entropies in Continuous Variable Systems22footnote 2 Presented at the International Conference “Entanglement, Information & Noise”, Krzyżowa, Poland, June 14–20, 2004.
## 1. Introduction
According to Erwin Schrödinger, quantum entanglement is not *“one but rather the characteristic trait of quantum mechanics, the one that enforces its entire departure from classical lines of thought”* . Entanglement has been widely recognized as a fundamental aspect of quantum theory, stemming directly from the superposition principle and quantum non factorizability. Remarkably, it is now also acknowledged as a fundamental physical resource, much on the same status as energy and entropy, and as a key factor in the realization of information processes otherwise impossible to implement on classical systems. Thus the degree of entanglement and information are the crucial features of a quantum state from the point of view of Quantum Information Theory . Indeed, the search for proper mathematical frameworks to quantify such features in generic (mixed) quantum states cannot be yet considered accomplished. In view of such considerations, it is clear that the full understanding of the relationships between the quantum correlations contained in a multipartite state and the global and local (i.e. referring to the reduced states of the subsystems) degrees of information of the state, is of critical importance. In particular, it would represent a relevant step towards the clarification of the nature of quantum correlations and, possibly, of the distinction between quantum and classical correlations of mixed states . The main question we want to address in this work is:
> What can we say about the quantum correlations existing between the subsystems of a quantum multipartite system in a mixed state, if we know the degrees of information carried by the global and the reduced states?
We can anticipate that the answer will be *“almost everything”* in the context of Gaussian states of continuous variable systems. To this aim, we will start by briefly reviewing in Sec. 2. the properties of Gaussian states in infinite–dimensional Hilbert spaces, and the concepts of information and entanglement. In Sec. 3. we will show, step by step, how the entanglement of two–mode Gaussian states can be accurately characterized by the knowledge of global and marginal degrees of information, quantified by the purities, or by the generalized entropies of the global state and of the reduced states of the subsystems. In Sec. 4. we will give a brief sketch of the generalization of our methods to the quantification of multipartite entanglement in multimode Gaussian states under symmetry. Finally, in Sec. 5. we will summarize our results and discuss future perspectives.
## 2. Gaussian states: general properties
We consider a continuous variable (CV) system consisting of $`N`$ canonical bosonic modes, associated to an infinite-dimensional Hilbert space and described by the vector $`\widehat{X}=\{\widehat{x}_1,\widehat{p}_1,\mathrm{},\widehat{x}_N,\widehat{p}_N\}`$ of the field quadrature (“position” and “momentum”) operators. The quadrature phase operators are connected to the annihilation $`\widehat{a}_i`$ and creation $`\widehat{a}_i^{}`$ operators of each mode, by the relations $`\widehat{x}_i=(\widehat{a}_i+\widehat{a}_i^{})`$ and $`\widehat{p}_i=(\widehat{a}_i\widehat{a}_i^{})/i`$. The canonical commutation relations for the $`\widehat{X}_i`$’s can be expressed in matrix form: $`[\widehat{X}_i,\widehat{X}_j]=2i\mathrm{\Omega }_{ij}`$, with the symplectic form $`\mathrm{\Omega }=_{i=1}^n\omega `$ and $`\omega =\delta _{ij1}\delta _{ij+1},i,j=1,2`$.
Quantum states of paramount importance in CV systems are the so-called Gaussian states, i.e. states with Gaussian characteristic functions and quasi–probability distributions . The interest in this special class of states (important examples are vacua, coherent, squeezed and thermal states of the electromagnetic field) stems from the feasibility to produce and control them with linear optics, and from the increasing number of efficient proposals and successful experimental implementations of CV quantum information and communication processes involving multimode Gaussian states (see for a recent review). By definition, a Gaussian state is completely characterized by first and second moments of the canonical operators. When addressing physical properties invariant under local unitary transformations, such as the mixedness and the entanglement, one can neglect first moments and completely characterize Gaussian states by the $`2N\times 2N`$ real covariance matrix (CM) $`𝝈`$, whose entries are $`\sigma _{ij}=1/2\{\widehat{X}_i,\widehat{X}_j\}\widehat{X}_i\widehat{X}_j`$. Throughout the paper, $`𝝈`$ will be used indifferently to indicate the CM of a Gaussian state or the state itself. A real, symmetric matrix $`𝝈`$ must fulfill the Robertson-Schrödinger uncertainty relation
$$𝝈+i\mathrm{\Omega }0,$$
(1)
to be a bona fide CM of a physical state. Symplectic operations (i.e. belonging to the group $`Sp_{(2N,)}=\{SSL(2N,):S^T\mathrm{\Omega }S=\mathrm{\Omega }\}`$) acting by congruence on CMs in phase space, amount to unitary operations on density matrices in Hilbert space. In phase space, any $`N`$-mode Gaussian state can be transformed by symplectic operations in its Williamson diagonal form $`𝝂`$ , such that $`𝝈=S^T𝝂S`$, with $`𝝂=\mathrm{diag}\{\nu _1,\nu _1,\mathrm{}\nu _N,\nu _N\}`$. The set $`\mathrm{\Sigma }=\{\nu _i\}`$ constitutes the symplectic spectrum of $`𝝈`$ and its elements must fulfill the conditions $`\nu _i1`$, following from Eq. (1) and ensuring positivity of the density matrix associated to $`𝝈`$. We remark that the full saturation of the uncertainty principle can only be achieved by pure $`N`$-mode Gaussian states, for which $`\nu _i=1i=1,\mathrm{},N`$. Instead, mixed states such that $`\nu _{ik}=1`$ and $`\nu _{i>k}>1`$, with $`1kN`$, only partially saturate the uncertainty principle, with partial saturation becoming weaker with decreasing $`k`$. The symplectic eigenvalues $`\nu _i`$ can be computed as the orthogonal eigenvalues of the matrix $`|i\mathrm{\Omega }𝝈|`$, so they are determined by $`N`$ symplectic invariants associated to the characteristic polynomial of such a matrix. Two global invariants which will be useful are the determinant $`\mathrm{Det}𝝈=_i\nu _i^2`$ and the *seralian* $`\mathrm{\Delta }=_i\nu _i^2`$, which is the sum of the determinants of all the $`2\times 2`$ submatrices of $`𝝈`$ related to each mode.
The degree of information about the preparation of a quantum state $`\varrho `$ can be characterized by its *purity* $`\mu \mathrm{Tr}\varrho ^2`$. For a Gaussian state with CM $`𝝈`$ one has simply $`\mu =1/\sqrt{\mathrm{Det}𝝈}`$ . In general, the mixedness, or lack of information about the preparation of the state, can be quantified by generalized entropic measures, such as the Bastiaans–Tsallis entropies $`S_p(1\mathrm{Tr}\varrho ^p)/(p1)`$, which reduce to the linear entropy $`S_L=1\mu `$ for $`p=2`$, and the Rényi entropies $`S_p^R(\mathrm{log}\mathrm{Tr}\varrho ^p)/(1p)`$. Both entropic families are parametrized by $`p>1`$ and it can be easily shown that $`lim_{p1+}S_p=lim_{p1+}S_p^R=\mathrm{Tr}(\varrho \mathrm{log}\varrho )S_V`$, so that also the Shannon-von Neumann entropy $`S_V`$ can be defined in terms of generalized entropies. The quantity $`S_V`$ is additive on tensor product states and provides a further convenient measure of mixedness of the quantum state $`\varrho `$.
As for the entanglement, we recall that positivity of the CM’s partial transpose (PPT) is a necessary and sufficient condition of separability for $`(N+1)`$-mode Gaussian states with respect to $`1\times N`$-mode partitions . In phase space, partial transposition amounts to a mirror reflection of one quadrature associated to the single-mode partition. If $`\{\stackrel{~}{\nu }_i\}`$ is the symplectic spectrum of the partially transposed CM $`\stackrel{~}{𝝈}`$, then a $`(N+1)`$-mode Gaussian state with CM $`𝝈`$ is separable if and only if $`\stackrel{~}{\nu }_i1`$ $`i`$. A convenient measure of CV entanglement is the *logarithmic negativity* $`E_𝒩\mathrm{log}\stackrel{~}{\varrho }_1`$, $`_1`$ denoting the trace norm, which constitutes an upper bound to the distillable entanglement of the quantum state $`\varrho `$. It can be readily computed in terms of the symplectic spectrum $`\stackrel{~}{\nu }_i`$ of $`\stackrel{~}{𝝈}`$, yielding
$$E_𝒩=\{\begin{array}{cc}0,& \stackrel{~}{\nu }_i1i;\\ _{i:\stackrel{~}{\nu }_i<1}\mathrm{log}\stackrel{~}{\nu }_i,& \text{else .}\end{array}$$
(2)
The logarithmic negativity quantifies the extent to which the PPT condition $`\stackrel{~}{\nu }_i1`$ is violated.
## 3. Characterizing two–mode entanglement by information measures
### 3.1. Parametrization of Gaussian states with symplectic invariants
Two–mode Gaussian states represent the prototypical quantum states of CV systems. Their CM can be written is the following block form
$$𝝈\left(\begin{array}{cc}𝜶& 𝜸\\ 𝜸^T& 𝜷\end{array}\right),$$
(3)
where the three $`2\times 2`$ matrices $`𝜶`$, $`𝜷`$, $`𝜸`$ are, respectively, the CMs of the two reduced modes and the correlation matrix between them. It is well known that for any two–mode CM $`𝝈`$ there exists a local symplectic operation $`S_l=S_1S_2`$ which takes $`𝝈`$ to the so called standard form $`𝝈_{sf}`$
$$S_l^T𝝈S_l=𝝈_{sf}\left(\begin{array}{cccc}a& 0& c_+& 0\\ 0& a& 0& c_{}\\ c_+& 0& b& 0\\ 0& c_{}& 0& b\end{array}\right).$$
(4)
States whose standard form fulfills $`a=b`$ are said to be symmetric. Let us recall that any pure state is symmetric and fulfills $`c_+=c_{}=\sqrt{a^21}`$. The uncertainty principle Ineq. (1) can be recast as a constraint on the $`Sp_{(4,)}`$ invariants $`\mathrm{Det}𝝈`$ and $`\mathrm{\Delta }(𝝈)=\mathrm{Det}𝜶+\mathrm{Det}𝜷+2\mathrm{Det}𝜸`$, yielding $`\mathrm{\Delta }(𝝈)1+\mathrm{Det}𝝈`$. The symplectic eigenvalues of a two–mode Gaussian state will be named $`\nu _{}`$ and $`\nu _+`$, with $`\nu _{}\nu _+`$ in general. A simple expression for the $`\nu _{}`$ can be found in terms of the two $`Sp_{(4,)}`$ invariants
$$2\nu _{}^2=\mathrm{\Delta }(𝝈)\sqrt{\mathrm{\Delta }(𝝈)^24\mathrm{Det}𝝈}.$$
(5)
The standard form covariances $`a`$, $`b`$, $`c_+`$, and $`c_{}`$ can be determined in terms of the two local symplectic invariants
$$\mu _1=(\mathrm{Det}𝜶)^{1/2}=1/a,\mu _2=(\mathrm{Det}𝜷)^{1/2}=1/b,$$
(6)
which are the marginal purities of the reduced single–mode states, and of the two global symplectic invariants
$$\mu =(\mathrm{Det}𝝈)^{1/2}=[(abc_+^2)(abc_{}^2)]^{1/2},\mathrm{\Delta }=a^2+b^2+2c_+c_{},$$
(7)
which are the global purity and the seralian, respectively. Eqs. (6-7) can be inverted to provide the following physical parametrization of two–mode states in terms of the four independent parameters $`\mu _1,\mu _2,\mu `$, and $`\mathrm{\Delta }`$ :
$`a={\displaystyle \frac{1}{\mu _1}},b`$ $`=`$ $`{\displaystyle \frac{1}{\mu _2}},c_\pm ={\displaystyle \frac{\sqrt{\mu _1\mu _2}}{4}}\left(ϵ_{}\pm ϵ_+\right),`$ (8)
$`\mathrm{with}ϵ_{}`$ $``$ $`\sqrt{\left[\mathrm{\Delta }{\displaystyle \frac{(\mu _1\mu _2)^2}{\mu _1^2\mu _2^2}}\right]^2{\displaystyle \frac{4}{\mu ^2}}}.`$
The uncertainty principle and the existence of the radicals appearing in Eq. (8) impose the following constraints on the four invariants in order to describe a physical state
$`0`$ $``$ $`\mu _{1,2}\mathrm{\hspace{0.17em}\hspace{0.17em}1},`$ (9)
$`\mu _1\mu _2`$ $``$ $`\mu {\displaystyle \frac{\mu _1\mu _2}{\mu _1\mu _2+\left|\mu _1\mu _2\right|}},`$ (10)
$`{\displaystyle \frac{2}{\mu }}+{\displaystyle \frac{(\mu _1\mu _2)^2}{\mu _1^2\mu _2^2}}`$ $``$ $`\mathrm{\Delta }\mathrm{\hspace{0.17em}\hspace{0.17em}1}+{\displaystyle \frac{1}{\mu ^2}}.`$ (11)
The physical meaning of these constraints, and the role of the extremal states (i.e. states whose invariants saturate the upper or lower bounds of Eqs. (10-11)) in relation to the entanglement, will be carefully investigated in the next subsections.
In terms of symplectic invariants, partial transposition corresponds to flipping the sign of $`\mathrm{Det}𝜸`$, so that $`\mathrm{\Delta }`$ turns into $`\stackrel{~}{\mathrm{\Delta }}=\mathrm{\Delta }4\mathrm{Det}𝜸=\mathrm{\Delta }+2/\mu _1^2+2/\mu _2^2`$. The symplectic eigenvalues of the CM $`𝝈`$ and of its partial transpose $`\stackrel{~}{𝝈}`$ are promptly determined in terms of symplectic invariants
$$2\nu _{}^2=\mathrm{\Delta }\sqrt{\mathrm{\Delta }^2\frac{4}{\mu ^2}},2\stackrel{~}{\nu }_{}^2=\stackrel{~}{\mathrm{\Delta }}\sqrt{\stackrel{~}{\mathrm{\Delta }}^2\frac{4}{\mu ^2}}.$$
(12)
The PPT criterion yields a state $`𝝈`$ separable if and only if $`\stackrel{~}{\nu }_{}1`$. A bona fide measure of entanglement for two–mode Gaussian states should thus be a monotonically decreasing function of $`\stackrel{~}{\nu }_{}`$ , quantifying the violation of the previous inequality. A computable entanglement monotone for generic two-mode Gaussian states is provided by the logarithmic negativity Eq. (2)
$$E_𝒩=\mathrm{max}\{0,\mathrm{log}\stackrel{~}{\nu }_{}\}.$$
(13)
In the special instance of symmetric Gaussian states, the *entanglement of formation* is also computable but, being again a decreasing function of $`\stackrel{~}{\nu }_{}`$, it provides the same characterization of entanglement and is thus fully equivalent to the logarithmic negativity in this subcase.
### 3.2. Entanglement vs Information (I) – Maximal entanglement at fixed global purity
The first step towards giving an answer to our original question is to investigate the properties of extremally entangled states at a given degree of global information. Let us mention that, for two–qubit systems, the existence of maximally entangled states at fixed mixedness (MEMS) was first found numerically by Ishizaka and Hiroshima . The discovery of such states spurred several theoretical works , aimed at exploring the relations between different measures of entanglement and mixedness (strictly related to the questions of the ordering of these different measures , and to the volume of the set of mixed entangled states ).
Unfortunately, it is easy to show that a similar analysis in the CV scenario is meaningless. Indeed, for any fixed, finite global purity $`\mu `$ there exist infinitely many Gaussian states which are infinitely entangled. As an example, we can consider the class of (nonsymmetric) two–mode squeezed thermal states. Let $`S_r=\mathrm{exp}(\frac{1}{2}ra_1a_2\frac{1}{2}ra_1^{}a_2^{})`$ be the two mode squeezing operator with real squeezing parameter $`r0`$, and let $`\varrho _{\nu _i}^{_{}}`$ be a tensor product of thermal states with CM $`𝝂_\nu _{}=\mathrm{𝟙}_2\nu _{}\mathrm{𝟙}_2\nu _+`$, where $`\nu _{}`$ is, as usual, the symplectic spectrum of the state. Then, a nonsymmetric two-mode squeezed thermal state $`\xi _{\nu _i,r}`$ is defined as $`\xi _{\nu _i,r}=S_r\varrho _{\nu _i}^{_{}}S_r^{}`$, corresponding to a standard form with
$`a`$ $`=`$ $`\nu _{}\mathrm{cosh}^2r+\nu _+\mathrm{sinh}^2r,b=\nu _{}\mathrm{sinh}^2r+\nu _+\mathrm{cosh}^2r,`$ (14)
$`c_\pm `$ $`=`$ $`\pm {\displaystyle \frac{\nu _{}+\nu _+}{2}}\mathrm{sinh}2r.`$
For simplicity we can consider the symmetric instance ($`\nu _{}=\nu _+=1/\sqrt{\mu }`$) and compute the logarithmic negativity Eq. (13), which takes the expression $`E_𝒩(r,\mu )=(1/2)\mathrm{log}[\mathrm{e}^{4r}/\mu ]`$. Notice how the completely mixed state ($`\mu 0`$) is always separable while, for any $`\mu >0`$, we can freely increase the squeezing $`r`$ to obtain Gaussian states with arbitrarily large entanglement. For fixed squeezing, as naturally expected, the entanglement decreases with decreasing degree of purity of the state, analogously to what happens in discrete–variable MEMS .
### 3.3. Entanglement vs Information (II) – Maximal entanglement at fixed local purities
The next step in the analysis is the unveiling of the relation between the entanglement of a Gaussian state of CV systems and the degrees of information related to the subsystems. Maximally entangled states for given marginal mixednesses (MEMMS) have been recently introduced and analyzed in detail in the context of qubit systems by Adesso et al. . The MEMMS provide a suitable generalization of pure states, in which the entanglement is completely quantified by the marginal degrees of mixedness.
For two–mode Gaussian states, it follows from the expression Eq. (12) of $`\stackrel{~}{\nu }_{}`$ that, for fixed marginal purities $`\mu _{1,2}`$ and seralian $`\mathrm{\Delta }`$, the logarithmic negativity is strictly increasing with increasing $`\mu `$. By imposing the saturation of the upper bound of Eq. (10), $`\mu =\mu ^{\mathrm{max}}(\mu _{1,2})(\mu _1\mu _2)/(\mu _1\mu _2+\left|\mu _1\mu _2\right|)`$, we determine the most pure states for fixed marginals; moreover, choosing $`\mu =\mu ^{\mathrm{max}}(\mu _{1,2})`$ immediately implies that the upper and the lower bounds on $`\mathrm{\Delta }`$ of Eq. (11) coincide and $`\mathrm{\Delta }`$ is uniquely determined in terms of $`\mu _{1,2}`$. This means that the two–mode states with maximal purity for fixed marginals are indeed the Gaussian maximally entangled states for fixed marginal mixednesses (GMEMMS). They can be seen as the CV analogues of the MEMMS.
In Fig. 1 the logarithmic negativity of GMEMMS is plotted (a) as a function of the marginal linear entropies $`S_{L1,2}1\mu _{1,2}`$, in comparison (b) with the behaviour of the tangle (an entanglement monotone equivalent to the entanglement of formation for two qubits ) as a function of $`S_{L1,2}`$ for discrete variable MEMMS. Notice, as a common feature, how the maximal entanglement achievable by quantum mixed states rapidly increases with increasing marginal mixednesses (like in the pure–state instance) and decreases with increasing difference of the marginals. This is natural, because the presence of quantum correlations between the subsystems implies that they should possess similar amounts of quantum information. Let us finally mention that the “minimally” entangled states for fixed marginals, which saturate the lower bound of Eq. (10) ($`\mu =\mu _1\mu _2`$), are just the tensor product states, i.e. states without any (quantum or classical) correlations between the subsystems.
### 3.4. Entanglement vs Information (III) – Extremal entanglement at fixed global and local purities
What we have shown so far, by simple analytical bounds, is a general trend of increasing entanglement with increasing global purity, and with decreasing marginal purities and difference between them. We now wish to exploit the joint information about global and marginal degrees of purity to achieve a significative characterization of entanglement, both qualitatively and quantitatively. Let us first investigate the role played by the seralian $`\mathrm{\Delta }`$ in the characterization of the properties of Gaussian states. To this aim, we analyse the dependence of the eigenvalue $`\stackrel{~}{\nu }_{}`$ on $`\mathrm{\Delta }`$, for fixed $`\mu _{1,2}`$ and $`\mu `$:
$$\frac{\stackrel{~}{\nu }_{}^2}{\mathrm{\Delta }}|_{\mu _1,\mu _2,\mu }=\frac{1}{2}\left(\frac{\stackrel{~}{\mathrm{\Delta }}}{\sqrt{\stackrel{~}{\mathrm{\Delta }}^2\frac{1}{4\mu ^2}}}1\right)>0.$$
(15)
The smallest symplectic eigenvalue of the partially transposed state is strictly monotone in $`\mathrm{\Delta }`$. Therefore the entanglement of a generic Gaussian state $`𝝈`$ with given global purity $`\mu `$ and marginal purities $`\mu _{1,2}`$, strictly increases with decreasing $`\mathrm{\Delta }`$. The seralian $`\mathrm{\Delta }`$ is thus endowed with a direct physical interpretation: at given global and marginal purities, it determines the amount of entanglement of the state. Moreover, due to inequality (11), $`\mathrm{\Delta }`$ is constrained both by lower and upper bounds; therefore, not only maximally but also *minimally* entangled Gaussian states exist. This fact admirably elucidates the relation between quantum correlations and information in two–mode Gaussian states: the entanglement of such states is tightly bound by the amount of global and marginal purities, with only one remaining degree of freedom related to the invariant $`\mathrm{\Delta }`$ .
We now aim to characterize *extremally* (maximally and minimally) entangled Gaussian states for fixed global and marginal purities. Let us first consider the states saturating the lower bound in Eq. (11), which entails *maximal* entanglement and defines the class of Gaussian most entangled states for fixed global and local purities (GMEMS). It is easily seen that such states belong to the class of asymmetric two–mode squeezed thermal states Eq. (14), with squeezing parameter and symplectic spectrum
$`\mathrm{tanh}2r`$ $`=`$ $`2(\mu _1\mu _2\mu _1^2\mu _2^2/\mu )^{1/2}/(\mu _1+\mu _2),`$ (16)
$`\nu _{}^2`$ $`=`$ $`{\displaystyle \frac{1}{\mu }}+{\displaystyle \frac{(\mu _1\mu _2)^2}{2\mu _1^2\mu _2^2}}{\displaystyle \frac{|\mu _1\mu _2|}{2\mu _1\mu _2}}\sqrt{{\displaystyle \frac{(\mu _1\mu _2)^2}{\mu _1^2\mu _2^2}}+{\displaystyle \frac{4}{\mu }}}.`$ (17)
Nonsymmetric two–mode thermal squeezed states turn out to be *separable* in the range
$$\mu \frac{\mu _1\mu _2}{\mu _1+\mu _2\mu _1\mu _2}.$$
(18)
As a consequence, all Gaussian states whose purities fall in the *separable region* defined by inequality (18) are not entangled.
We next consider the states that saturate the upper bound in Eq. (11). They determine the class of Gaussian least entangled states for given global and local purities (GLEMS) and, outside the separable region (where every Gaussian state can be considered a GLEMS having zero entanglement), they fulfill $`\mathrm{\Delta }=1+1/\mu ^2`$. This relation implies that the symplectic spectrum of these states takes the form $`\nu _{}=1`$, $`\nu _+=1/\mu `$. We thus find that GLEMS are mixed Gaussian states of partial minimum uncertainty, so in some sense they are the most classical ones and this is consistent with their property of having minimal entanglement. According to the PPT criterion, GLEMS are separable only if $`\mu \mu _1\mu _2/\sqrt{\mu _1^2+\mu _2^2\mu _1^2\mu _2^2}`$. Therefore, in the range
$$\frac{\mu _1\mu _2}{\mu _1+\mu _2\mu _1\mu _2}<\mu \frac{\mu _1\mu _2}{\sqrt{\mu _1^2+\mu _2^2\mu _1^2\mu _2^2}}$$
(19)
both separable and entangled states can be found. Instead, the region
$$\mu >\frac{\mu _1\mu _2}{\sqrt{\mu _1^2+\mu _2^2\mu _1^2\mu _2^2}}$$
(20)
can only accomodate *entangled* states. The very narrow region defined by inequality (19) is thus the only *region of coexistence* of both entangled and separable Gaussian mixed states. The discrimination of the different zones provides strong necessary or sufficient conditions for the entanglement in terms of the degrees of information, and allows to classify the separability of all two-mode Gaussian states according to their global and marginal purities, as shown in Fig. 2.
Knowledge of $`\mu _{1,2}`$ and $`\mu `$ thus accurately *qualifies* the entanglement of Gaussian states: as we will now show, quantitative knowledge of the local and global purities provides a reliable *quantification* of entanglement as well. Outside the separable region (see Table I in Fig. 2), GMEMS attain maximum logarithmic negativity $`E_{𝒩max}(\mu _{1,2},\mu )`$, while, in the entangled region, GLEMS acquire minimum logarithmic negativity $`E_{𝒩min}(\mu _{1,2},\mu )`$, where
$`E_{𝒩max}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\mathrm{log}[{\displaystyle \frac{1}{\mu }}+\left({\displaystyle \frac{\mu _1+\mu _2}{2\mu _1^2\mu _2^2}}\right)\left(\mu _1+\mu _2\sqrt{(\mu _1+\mu _2)^2{\displaystyle \frac{4\mu _1^2\mu _2^2}{\mu }}}\right)],`$ (21)
$`E_{𝒩min}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\mathrm{log}[{\displaystyle \frac{1}{\mu _1^2}}+{\displaystyle \frac{1}{\mu _2^2}}{\displaystyle \frac{1}{2\mu ^2}}{\displaystyle \frac{1}{2}}\sqrt{\left({\displaystyle \frac{1}{\mu _1^2}}+{\displaystyle \frac{1}{\mu _2^2}}{\displaystyle \frac{1}{2\mu ^2}}{\displaystyle \frac{1}{2}}\right)^2{\displaystyle \frac{1}{\mu ^2}}}].`$ (22)
Knowledge of the full CM, i.e. including the symplectic invariant $`\mathrm{\Delta }`$ or all the cross-correlations, would allow for an exact quantification of the entanglement. However, we will now show that an estimate based only on the knowledge of the experimentally measurable global and marginal purities turns out to be quite accurate. We will quantify the entanglement of Gaussian states with given global and marginal purities by the *“average logarithmic negativity”* $`\overline{E}_𝒩`$
$$\overline{E}_𝒩(\mu _{1,2},\mu )\frac{E_{𝒩max}(\mu _{1,2},\mu )+E_{𝒩min}(\mu _{1,2},\mu )}{2}.$$
(23)
We can then also define the relative error $`\delta \overline{E}_𝒩`$ on $`\overline{E}_𝒩`$ as
$$\delta \overline{E}_𝒩(\mu _{1,2},\mu )\frac{E_{𝒩max}(\mu _{1,2},\mu )E_{𝒩min}(\mu _{1,2},\mu )}{E_{𝒩max}(\mu _{1,2},\mu )+E_{𝒩min}(\mu _{1,2},\mu )}.$$
(24)
It is easily seen that this error decreases exponentially both with increasing global purity and decreasing marginal purities, i.e. with increasing entanglement. For ease of graphical display, let us consider the important case of symmetric Gaussian states, for which the reduction $`\mu _1=\mu _2\mu _i`$ occurs.
In Fig. 4, $`E_{𝒩max}(\mu _i,\mu )`$ of Eq. (21) and $`E_{𝒩min}(\mu _i,\mu )`$ of Eq. (22) are plotted versus $`\mu _i`$ and $`\mu `$. In the case $`\mu =1`$ the upper and lower bounds correctly coincide, since for pure states the entanglement is completely quantified by the marginal purity. For mixed states this is not the case, but, as the plot shows, knowledge of the global and marginal purities strictly bounds the entanglement both from above and from below. The relative error $`\delta \overline{E}_𝒩(\mu _i,\mu )`$ given by Eq. (24) is plotted in Fig. 4 as a function of the ratio $`\mu /\mu _i`$. It decays exponentially, dropping below $`5\%`$ for $`\mu >\mu _i`$. Thus the reliable quantification of quantum correlations in genuinely entangled states is always assured by this method, except at most for a small set of states with very weak entanglement (states with $`E_𝒩1`$). Moreover, the accuracy is even greater in the general non-symmetric case $`\mu _1\mu _2`$, because the maximal entanglement decreases in such an instance (see Fig. 1). This analysis shows that the average logarithmic negativity $`\overline{E}_𝒩`$ is a reliable estimate of the logarithmic negativity $`E_𝒩`$, improving as the entanglement increases. This allows for an accurate quantification of CV entanglement by knowledge of the global and marginal purities. The purities may be in turn directly measured experimentally, without the full tomographic reconstruction of the whole CM, by exploiting quantum networks techniques or single–photon detections without homodyning .
### 3.5. Entanglement vs Information (IV) – Extremal entanglement at fixed global and local generalized entropies
In this section we introduce a more general characterization of the entanglement of two–mode Gaussian states in terms of the degrees of information, by exploiting the generalized Tsallis $`p`$entropies
$$S_p\frac{1\mathrm{Tr}\varrho ^p}{p1},p>1,$$
(25)
as measures of global and marginal mixedness. Such an analysis can be carried out along the same lines of the previous section, by studying the explicit behavior of the global invariant $`\mathrm{\Delta }`$ at fixed global and marginal entropies, and its relation with the logarithmic negativity $`E_𝒩`$. Let us remark that the $`S_p`$’s can be computed for a generic Gaussian state in terms of the symplectic eigenvalues , namely
$$\mathrm{Tr}\varrho ^p=\underset{i=1}{\overset{n}{}}g_p(\nu _i),g_p(x)=\frac{2^p}{(x+1)^p(x1)^p}.$$
(26)
We begin by observing that the standard form CM $`𝝈`$ of a generic two–mode Gaussian state Eq. (4) can be parametrized by the following quantities: the two marginals $`\mu _{1,2}`$ (or any other marginal $`S_{p_{1,2}}`$ because all the local, single-mode entropies are equivalent for any value of the integer $`p`$), the global $`p`$entropy $`S_p`$ (for some chosen value of the integer $`p`$), and the global symplectic invariant $`\mathrm{\Delta }`$.
After somewhat lengthy but straightforward calculations (the details can be found in Ref. ), one finds that the entanglement is still bounded from above and from below by functionals of the global and marginal $`p`$entropies, and the two extremal classes of states are again the nonsymmetric squeezed thermal states (GMEMS) and the mixed states of partial minimum uncertainty (GLEMS). Nevertheless, the seralian $`\mathrm{\Delta }`$ is no longer monotonically related to the entanglement of the state, at fixed generalized entropies. In particular, for any $`p>2`$ (i.e. with the exception of the linear and Von Neumann entropies), there exists a unique *nodal surface* $`𝒮_p^n(S_{p_{1,2}})`$ such that
$$\frac{\stackrel{~}{\nu }_{}}{\mathrm{\Delta }}|_{S_{p_1},S_{p_2},S_p}\mathrm{is}\{\begin{array}{cc}>0,\hfill & \text{ when}S_p>𝒮_p^n(S_{p_{1,2}});\hfill \\ =0,\hfill & \text{ when}S_p=𝒮_p^n(S_{p_{1,2}});\hfill \\ <0,\hfill & \text{ when}S_p<𝒮_p^n(S_{p_{1,2}}).\hfill \end{array}$$
(27)
While in the first case of Eq. (27) GMEMS and GLEMS retain their property of being, respectively, maximally and minimally entangled Gaussian states for fixed degrees of information, in the last case they exchange their role: two–mode squeezed states become minimally entangled states for fixed $`p`$entropies, and the states of partial minimum uncertainty are those with maximal entanglement. Even more remarkably, the entanglement of all Gaussian states (including GMEMS, GLEMS and infinitely many other different states) whose $`p`$entropies lay on the nodal surface of inversion $`S_p=𝒮_p^n(S_{p_{1,2}})`$, does not depend on $`\mathrm{\Delta }`$ and is therefore completely quantified in terms of the global and marginal generalized degrees of information. Thus the state parametrization by the Tsallis $`p`$entropies (and, presumably, by the companion family of the Rényi entropies as well), provides a remarkable inversion between GMEMS and GLEMS, and allows to identify a class of Gaussian states whose entanglement is quantified exactly by the knowledge of the global and marginal degrees of information. Moreover, even outside the nodal surface of inversion, the gap between maximal and minimal entanglement for fixed $`p`$entropies decreases with increasing $`p`$, so that the accuracy of the quantitative estimate of the logarithmic negativity as a function of the $`p`$entropies increases accordingly, as shown in Fig. 5. Despite this interesting features, measurements of purity, or equivalently of linear entropy ($`p=2`$), remain the best candidates for a direct estimation of CV entanglement in realistic experiments. This is due to the fact that measuring the global $`p`$entropy ($`p2`$) of a state requires the full tomographic reconstruction of the whole CM, thus nullifying the advantages of the analysis presented in Sec. 3.4..
## 4. Quantification of multimode entanglement under symmetry
For two–mode Gaussian states of CV systems we have shown that the measurement of the three purities (or generalized entropies), out of the four independent standard form covariances, suffices in providing a reliable quantitative characterization of the entanglement. It is intuitively evident that the efficiency of such a quantitative estimation in terms of information entropies should improve significantly with increasing number $`N`$ of modes, because the ratio between the total number of covariances and the total number of global and marginal degrees of information quickly scales with $`N`$. Moreover, the structure of multipartite entanglement that can arise in multimode settings is much richer than the basic bipartite CV entanglement. Here we briefly discuss the simplest multipartite setting in which the purities again successfully bound the entanglement with great accuracy.
We consider highly symmetric $`(N+1)`$-mode Gaussian states of generic single-mode systems $`𝜶`$ coupled to fully symmetric $`N`$-mode systems $`𝝈_{\beta ^N}`$, resulting in a global CM $`𝝈`$ of the form
$$𝝈=(\begin{array}{cc}𝜶& 𝚪\\ 𝚪^𝖳& 𝝈_{\beta ^N}\end{array}),𝝈_{\beta ^N}=(\begin{array}{cccc}𝜷& 𝜺& \mathrm{}& 𝜺\\ 𝜺& 𝜷& 𝜺& \mathrm{}\\ \mathrm{}& 𝜺& \mathrm{}& 𝜺\\ 𝜺& \mathrm{}& 𝜺& 𝜷\end{array}),𝚪=(\underset{N}{\underset{}{𝜸\mathrm{}𝜸}}).$$
(28)
The state $`𝝈`$ is determined by six independent parameters, three of which are related to the fully symmetric $`N`$-mode block $`𝝈_{\beta ^N}`$ (or, equivalently, to any of its two–mode subblocks $`𝝈_{\beta ^2}`$). The remaining parameters are determined by the single-mode purity $`\mu _\alpha (\mathrm{Det}𝜶)^{1/2}`$ and by the two global $`Sp_{(2N+2,)}`$ invariants $`\mathrm{Det}𝝈1/(\mu _\sigma ^2)=_{i=1}^{N+1}\nu _i^2`$ and $`\mathrm{\Delta }_\sigma \mathrm{\Delta }_\alpha +\mathrm{\Delta }_{\beta ^N}=_{i=1}^{N+1}\nu _i^2`$. Here $`\mu _\sigma `$ is the global purity of the state $`𝝈`$, the $`\nu _i`$’s constitute its symplectic spectrum, $`\mathrm{\Delta }_\alpha \mathrm{Det}𝜶+2N\mathrm{Det}𝜸`$, and $`\mathrm{\Delta }_{\beta ^N}N(\mathrm{Det}𝜷+(N1)\mathrm{Det}𝜺)`$.
To compute the multimode $`1\times N`$ logarithmic negativity of state $`𝝈`$ between the mode $`𝜶`$ and the $`N`$-mode block $`𝝈_{\beta ^N}`$, it is convenient to perform the local symplectic operation that brings $`𝝈_{\beta ^N}`$ in its Williamson normal form, which is characterized by a $`(N1)`$-times degenerate eigenvalue $`\nu _{}=\sqrt{(be_1)(be_2)}`$ (the same smallest symplectic eigenvalue of the two–mode subblock $`𝝈_{\beta ^2}`$) and a nondegenerate eigenvalue $`\nu _{+^{(N)}}=\sqrt{(b+(N1)e_1)(b+(N1)e_2)}=1/(\nu _{}^{N1}\mu _{\beta ^N})`$, where $`\mu _{\beta ^N}`$ is the purity of the $`N`$-mode fully symmetric block . The crucial point here is that this local operation actually decouples the mode $`𝜶`$ from all the $`(N1)`$ modes corresponding to the eigenvalue $`\nu _{}`$ and concentrates the whole $`1\times N`$ entanglement between only two modes. The state Eq. (28) is thus brought in the form $`𝝈=\left(\begin{array}{cc}𝝈^{eq}& \mathrm{𝟎}\\ \mathrm{𝟎}& 𝝂_{}\end{array}\right),`$ where $`𝝂_{}`$ is a diagonal $`(2N2)\times (2N2)`$ matrix with all entries equal to $`\nu _{}`$, and the *equivalent* two–mode state $`𝝈^{eq}`$ is characterized by its invariants
$$\mu _1^{eq}=\mu _\alpha ,\mu _2^{eq}=\nu _{}^{N1}\mu _{\beta ^N},\mu ^{eq}=\nu _{}^{N1}\mu _\sigma ,\mathrm{\Delta }^{eq}=\mathrm{\Delta }_\alpha +(\nu _{}^{N1}\mu _{\beta ^N})^2.$$
(29)
These invariants are, in turn, determined in terms of the six invariants of the original $`N`$-mode state $`𝝈`$, but we can exploit the previous two–mode analysis (see Sec. 3.4.) to conclude that, even without the explicit knowledge of $`\mathrm{\Delta }^{eq}`$, and so of the coupling $`\mathrm{Det}𝜸`$, the multimode entanglement under symmetry can be quantified through the average logarithmic negativity Eq. (23) of the equivalent state $`𝝈^{eq}`$, with the *same* accuracy demonstrated for generic two–mode states (see Fig. 4).
Thus the degrees of information of a multipartite Gaussian state again provide a strong characterization and a reliable quantitative estimate of the entanglement. Moreover, the method of the two–mode reduction can be used to compute the $`1\times K`$ entanglement between the mode $`𝜶`$ and any $`K`$mode subblock of $`𝝈_{\beta ^N}`$, with $`K=1,\mathrm{},N`$, in order to establish a multipartite entanglement hierarchy in the $`(N+1)`$-mode state of the form Eq. (28) .
## 5. Summary and Outlook
In this work we aimed at unveiling the close relation between the entanglement encoded in a quantum state and its degrees of information. We have shown in detail how the knowledge of the global degree of information alone, or of the marginal informations related to the subsystems of a multipartite system, results in a qualitative characterization of the entanglement: the latter increases with decreasing global mixedness, with increasing marginal mixednesses, and with marginal mixednesses as close as possible. We then proved how the simultaneous knowledge of all the global and local degrees of information of a Gaussian state leads to the identification of extremally (maximally and minimally) entangled states at fixed mixednesses (purities or generalized information entropies), providing an accurate quantitative characterization of CV entanglement. It is worth remarking that, out of the subset of Gaussian states, very little is known about entanglement and information in generic states of CV systems. Nevertheless, most of the results presented here (including the sufficient conditions for entanglement based on information measures), derived for CM’s using the symplectic formalism in phase space, retain their validity for generic states of CV systems. For instance, any two-mode state with a CM corresponding to an entangled Gaussian state is itself entangled too . So our methods may serve to detect entanglement in a broader class of states of infinite-dimensional Hilbert spaces.
The generalization of this analysis, connecting entanglement and information, to highly symmetric multimode Gaussian states of $`1\times N`$-mode partitions has been briefly sketched, presenting a simple method to estimate CV multimode entanglement by measurements of purity in an equivalent two–mode state. The extension of this method to the quantification of multipartite entanglement in Gaussian states with respect to generic $`M\times N`$ bipartitions of the modes , as well as a deeper understanding of the structure of genuine multipartite CV entanglement and its relation with multiple degrees of information beyond the symmetry constraints, are being currently investigated.
## Acknowledgements
G. A. acknowledges stimulating discussions with Frank Verstraete, Reinhard F. Werner and Karol Życzkowski at EIN04 in Krzyżowa.
|
warning/0506/math0506044.html
|
ar5iv
|
text
|
# Free-differentiability conditions on the free-energy function implying large deviations
## 1 Introduction
Let $`(\mu _\alpha )`$ be a net of Radon sub-probability measures on a Hausdorff topological space $`X`$, and $`(t_\alpha )`$ be a net in $`]0,+\mathrm{}[`$ converging to $`0`$. Let $`(X)`$ (resp. $`𝒞(X)`$) denote the set of $`[\mathrm{},+\mathrm{}[`$-valued Borel measurable (resp. continuous) functions on $`X`$. For each $`h(X)`$, we define
$$\underset{¯}{\mathrm{\Lambda }}(h)=\mathrm{log}lim\; inf\mu _\alpha ^{t_\alpha }(e^{h/t_\alpha })$$
and
$$\overline{\mathrm{\Lambda }}(h)=\mathrm{log}lim\; sup\mu _\alpha ^{t_\alpha }(e^{h/t_\alpha })$$
where $`\mu _\alpha ^{t_\alpha }(e^{h/t_\alpha })`$ stands for $`(_Xe^{h(x)/t_\alpha }\mu _\alpha (dx))^{t_\alpha }`$, and write $`\mathrm{\Lambda }(h)`$ when both expressions are equal. When $`X=`$, for each pair of reals $`(\lambda ,\nu )`$, let $`h_{\lambda ,\nu }`$ be the function defined on $`X`$ by $`h_{\lambda ,\nu }(x)=\lambda x`$ if $`x0`$ and $`h_{\lambda ,\nu }(x)=\nu x`$ if $`x0`$ (we write simply $`h_\lambda `$ in place of $`h_{\lambda ,\lambda }`$). For each real $`\lambda `$, we put $`L(\lambda )=\mathrm{\Lambda }(h_\lambda )`$ when $`\mathrm{\Lambda }(h_\lambda )`$ exists.
A well-known problem of large deviations in $``$ (usually stated for sequences of probability measures) is the following: assuming that $`L(\lambda )`$ exists and is finite for all $`\lambda `$ in an open interval $`G`$ containing $`0`$, and that the map $`L_G`$ is not differentiable on $`G`$, what conditions on $`L_G`$ do imply large deviations, and with which rate function?
In relation with this problem, R. S. Ellis posed the following question (): assuming that $`\mathrm{\Lambda }(h_{\lambda ,\nu })`$ exists and is finite for all $`(\lambda ,\nu )^2`$, what conditions on the functional $`\mathrm{\Lambda }_{\{h_{\lambda ,\nu }:(\lambda ,\nu )^2\}}`$ do imply large deviations with rate function $`J(x)=sup_{(\lambda ,\nu )^2}\{h_{\lambda ,\nu }(x)\mathrm{\Lambda }(h_{\lambda ,\nu })\}`$ for all $`xX`$ ?
In this paper, we solve the above problem by giving conditions on $`L_G`$ involving only its left and right derivatives; the rate function is obtained as an abstract Legendre-Fenchel transform $`\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}`$, where $`𝒮`$ can be any set in $`𝒞(X)`$ containing $`\{h_\lambda :\lambda G\}`$ (Theorem 3). When $`𝒮=\{h_\lambda :\lambda G\}`$, we get a strengthening of the Gärtner-Ellis theorem by removing the usual differentiability assumption (Corollary 1). The answer to the Ellis question is obtained with $`𝒮=\{h_{\lambda ,\nu }:(\lambda ,\nu )^2\}`$ (Corollary 2).
The techniques used are refinements of those developed in previous author’s works (, ), where variational forms for $`\underset{¯}{\mathrm{\Lambda }}(h)`$ and $`\overline{\mathrm{\Lambda }}(h)`$ are obtained with $`h(X)`$ satisfying the usual Varadhan’s tail condition ($`X`$ a general space). We consider here the set $`𝒞_𝒦(X)`$ of elements $`h`$ in $`𝒞(X)`$ for which $`\{yX:e^{h(x)}\epsilon e^{h(y)}e^{h(x)}+\epsilon \}`$ is compact for all $`xX`$ and $`\epsilon >0`$ with $`e^{h(x)}>\epsilon `$. The first step is Theorem 2, which establishes that for any $`𝒯𝒞_𝒦(X)`$, and under suitable conditions (weaker than vague large deviations), there exist some reals $`m,M`$ such that
$$\mathrm{\Lambda }(h)=\underset{x\{mhM\}}{sup}\{h(x)l_1(x)\}\text{for all}h𝒯,$$
where $`l_1(x)=\mathrm{log}inf\{lim\; inf\mu _\alpha ^{t_\alpha }(G):xGX,G\text{ open}\}`$ for all $`xX`$; in particular, $`\mathrm{\Lambda }(h)`$ exists and has the same form as when large deviations hold. Note that when $`X=`$ and $`𝒯=\{h_\lambda :\lambda G\}`$ with $`0G`$, then the $`sup`$ in the above expression can be taken on a compact set (if $`0G`$, this follows from the exponential tightness). It turns out that any subnet of $`(\mu _\alpha ^{t_\alpha })`$ has a subnet $`(\mu _\gamma ^{t_\gamma })`$ satisfying the above conditions. The second step consists then in applying Theorem 2 with $`X=`$, $`𝒯=\{h_\lambda :\lambda G\}`$ and all these subnets. More precisely, we show that if $`x`$ is the left or right derivative of $`L`$ at some point $`\lambda _xG`$, then $`l_1^{(\mu _\gamma ^{t_\gamma })}(x)\lambda _xxL(\lambda _x)`$, whence
$$l_1^{(\mu _\gamma ^{t_\gamma })}(x)L_{G}^{}{}_{}{}^{}(x)$$
(1)
(Proposition 1). Let $`𝒮`$ be any set in $`𝒞(X)`$ containing $`\{h_\lambda :\lambda G\}`$, and assume that $`\mathrm{\Lambda }(h)`$ exists for all $`h𝒮`$. It is easy to see that
$$L_{G}^{}{}_{}{}^{}\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}l_0^{(\mu _\gamma ^{t_\gamma })}l_1^{(\mu _\gamma ^{t_\gamma })},$$
(2)
where $`l_0^{(\mu _\gamma ^{t_\gamma })}(x)=\mathrm{log}inf\{lim\; sup\mu _\gamma ^{t_\gamma }(G):xGX,G\text{open}\}`$ for all $`xX`$. Putting together (1) and (2) give
$$L_{G}^{}{}_{}{}^{}(x)=\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}(x)=l_0^{(\mu _\gamma ^{t_\gamma })}(x)=l_1^{(\mu _\gamma ^{t_\gamma })}(x)$$
(3)
for all $`x`$ in the image of the left (resp. right) derivative of $`L_G`$; consequently, if the set of these images contains $`\{\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}<+\mathrm{}\}`$, then $`(\mu _\gamma ^{t_\gamma })`$ satisfies a vague (narrow if $`0G`$) large deviation principle with powers $`(t_\gamma )`$ and rate function $`\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}`$, which moreover coincides with $`L_{G}^{}{}_{}{}^{}`$ on its effective domain. By compactness and Hausdorffness arguments, we conclude that the same result holds for the net $`(\mu _\alpha ^{t_\alpha })`$. Furthermore, $`\{\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}<+\mathrm{}\}`$ can be replaced by its interior, when $`\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}`$ is proper convex and lower semi-continuous, which is the case when $`𝒮=\{h_\lambda :\lambda G\}`$; this allows us to improve a strong version of Gärtner-Ellis theorem given by O’ Brien.
Various generalizations are given in order to get large deviations with a rate function coinciding with $`\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}`$ and $`L_{G}^{}{}_{}{}^{}`$ only on its effective domain. Note that all our results hold for general nets of sub-probability measures and powers.
The paper is organized as follows. Section 2 fixes the notations and recall some results on large deviations and convexity; Section 3 deals with the variational forms of the functionals $`\mathrm{\Lambda }`$; Section 4 treats the case $`X=`$.
## 2 Preliminaries
Throughout the paper, the notations $`\underset{¯}{\mathrm{\Lambda }}`$, $`\overline{\mathrm{\Lambda }}`$, $`\mathrm{\Lambda }`$, $`l_0`$, $`l_1`$ refer to the net $`(\mu _\alpha ^{t_\alpha })`$. We shall write $`l_1^{(\mu _\beta ^{t_\beta })}`$ when in the definition of $`l_1`$, $`(\mu _\alpha ^{t_\alpha })`$ is replaced by the subnet $`(\mu _\beta ^{t_\beta })`$. We do not make such distinction for the map $`\mathrm{\Lambda }`$, since it does not depend on the subnet along which the limit is taken. We recall that $`l_0`$ and $`l_1`$ are lower semi-continuous functions.
###### Definition 1
* $`(\mu _\alpha )`$ satisfies a (narrow) large deviation principle with powers $`(t_\alpha )`$ if there exists a $`[0,+\mathrm{}]`$-valued lower semi-continuous function $`J`$ on $`X`$ such that
$$lim\; sup\mu _\alpha ^{t_\alpha }(F)\underset{xF}{sup}e^{J(x)}\text{for all closed }FX$$
(4)
and
$$\underset{xG}{sup}e^{J(x)}lim\; inf\mu _\alpha ^{t_\alpha }(G)\text{for all open }GX;$$
$`J`$ is a rate function for $`(\mu _\alpha ^{t_\alpha })`$, which is said to be tight when it has compact level sets. When ”closed” is replaced by ”compact” in (4), we say that a vague large deviation principle holds.
* $`(\mu _\alpha )`$ is exponentially tight with respect to $`(t_\alpha )`$ if for each $`\epsilon >0`$ there exists a compact set $`K_\epsilon X`$ such that $`lim\; sup\mu _\alpha ^{t_\alpha }(X\mathtt{\backslash }K_\epsilon )<\epsilon `$.
The following results are well-known for a net $`(\mu _\epsilon ^\epsilon )_{\epsilon >0}`$, with $`\mu _\epsilon `$ a Radon probability measure (); it is easy to see that the proofs work also for general nets of sub-probability measures and powers.
###### Lemma 1
* Let $`X`$ be locally compact Hausdorff. Then, $`(\mu _\alpha )`$ satisfies a vague large deviation principle with powers $`(t_\alpha )`$ if and only if $`l_0=l_1`$. In this case, $`l_0`$ is the rate function.
* If $`(\mu _\alpha )`$ satisfies a vague large deviation principle with powers $`(t_\alpha )`$, and $`(\mu _\alpha )`$ is exponentially tight with respect to $`(t_\alpha )`$, then $`(\mu _\alpha )`$ satisfies a large deviation principle with same powers and same rate function.
A capacity on $`X`$ is a map $`c`$ from the powerset of $`X`$ to $`[0,+\mathrm{}]`$ such that:
* $`c(\mathrm{})=0`$.
* $`c(Y)=sup\{c(K):KY,K\text{ compact}\}`$ for all $`YX`$.
* $`c(K)=inf\{c(G):KGX,G\text{open}\}`$ for all compact $`KX`$.
The vague topology on the set of capacities is the coarsest topology for which the maps $`cc(Y)`$ are upper (resp. lower) semi-continuous for all compact (resp. open) $`YX`$. Let $`\mathrm{\Gamma }(X,[0,1])`$ denote the set of $`[0,1]`$-valued capacities on $`X`$ provided with the vague topology, and note that $`(\mu _\alpha ^{t_\alpha })`$ is a net in $`\mathrm{\Gamma }(X,[0,1])`$. For each $`[0,+\mathrm{}]`$-valued lower semi-continuous function $`l`$ on $`X`$, we associate the element $`c_l`$ in $`\mathrm{\Gamma }(X,[0,1])`$ defined by $`c_l(Y)=sup_{xY}e^{l(x)}`$ for all $`YX`$. We refer to for the first assertion in the following lemma; the second one is the mere transcription of the definition of a vague large deviation principle in terms of capacities.
###### Lemma 2
* If $`X`$ is locally compact Hausdorff, then $`\mathrm{\Gamma }(X,[0,1])`$ is a compact Hausdorff space.
* $`(\mu _\alpha )`$ satisfies a vague large deviation principle with powers $`(t_\alpha )`$ and rate function $`J`$ if and only if $`(\mu _\alpha ^{t_\alpha })`$ converges to $`c_J`$ in $`\mathrm{\Gamma }(X,[0,1])`$.
For any $`[\mathrm{},+\mathrm{}]`$-valued (not necessary convex) function $`f`$ defined on some topological space, we put $`𝒟\text{om}(f)=\{f<+\mathrm{}\}`$ (the so-called effective domain), and denote by $`\text{int}𝒟\text{om}(f)`$ (resp. $`\text{bd}𝒟\text{om}(f)`$) the interior (resp. boundary) of $`𝒟\text{om}(f)`$. The range of $`f`$ is denoted by $`\text{ran}f`$.
A $`[\mathrm{},+\mathrm{}]`$-valued convex function $`f`$ on $``$ is said to be proper if $`f`$ is $`]\mathrm{},+\mathrm{}]`$-valued and takes a finite value on at least one point. The Legendre-Fenchel transform $`f^{}`$ of $`f`$ is defined by $`f^{}(x)=sup_\lambda \{\lambda xf(\lambda )\}`$ for all $`x`$; note that $`f^{}`$ is convex lower semi-continuous, and proper when $`f`$ is proper. Let $`I`$ be a nonempty interval, and $`f_I`$ be a $`]\mathrm{},+\mathrm{}]`$-valued convex function on $`I`$. We denote by $`\widehat{f_I}`$ the convex function on $``$ which coincides with $`f_I`$ on $`I`$, and takes the value $`+\mathrm{}`$ out $`I`$; in this case we write simply $`f_{I}^{}{}_{}{}^{}`$ in place of $`\widehat{f_I}^{}`$. The left and right derivatives of $`f_I`$ at some point $`x𝒟\text{om}(f_I)`$ are denoted by $`f_{I}^{}{}_{}{}^{}(x)`$ and $`f_{I}^{}{}_{+}{}^{}(x)`$ respectively. A proper convex function $`f`$ on $``$ is said to be essentially smooth if $`\text{int}𝒟\text{om}(f)\mathrm{}`$, $`f`$ is differentiable on $`\text{int}𝒟\text{om}(f)`$, and $`lim|f^{}(x_n)|=+\mathrm{}`$ for all sequences $`(x_n)`$ in $`\text{int}𝒟\text{om}(f)`$ converging to some $`x\text{bd}𝒟\text{om}(f)`$ ().
If $`L(\lambda )`$ exists and is finite for all $`\lambda `$ in a nonempty open interval $`G`$, then $`L_G`$ is convex; if moreover $`0G`$, then $`(\mu _\alpha )`$ is exponentially tight with respect to $`(t_\alpha )`$. If $`L(\lambda )`$ exists for all reals $`\lambda `$, then $`L`$ is a $`[\mathrm{},+\mathrm{}]`$-valued convex function on $``$; if moreover $`0\text{int}𝒟\text{om}(L)`$, then $`L`$ is proper (the proof of these facts is obtained by modifying suitably the one of Lemma 2.3.9 in ).
###### Lemma 3
Let $`f`$ be a proper convex lower semi-continuous function on $``$. Then,
$$\underset{yG}{inf}f(y)=\underset{yG\text{int}𝒟\text{om}(f)}{inf}f(y)$$
for all open sets $`G`$.
###### Proof.
Let $`G`$ be an open subset of $``$. If $`G𝒟\text{om}(f)=\mathrm{}`$, then the conclusion holds trivially ($`inf\mathrm{}=+\mathrm{}`$ by convention). Assume that $`G𝒟\text{om}(f)\mathrm{}`$. By Corollary 6.3.2 of , $`G\text{int}𝒟\text{om}(f)\mathrm{}`$. By Theorem VI.3.2 of , for each $`x𝒟\text{om}(f)`$ we can find a sequence $`(x_n)`$ in $`\text{int}𝒟\text{om}(f)`$ converging to $`x`$ and such that $`limf(x_n)=f(x)`$, which implies $`inf_{G𝒟\text{om}(f)}f=inf_{G\text{int}𝒟\text{om}(f)}f`$, and the lemma is proved since $`inf_{G𝒟\text{om}(f)}f=inf_Gf`$. ∎
## 3 Variational forms for $`\mathrm{\Lambda }`$ on $`𝒞_𝒦(X)`$
We begin by defining a notion, which will appear as a key condition in the sequel; it is nothing else but a uniform version of the tail condition in Varadhan’s theorem.
###### Definition 2
We say that a set $`𝒯(X)`$ satisfies the tail condition for $`(\mu _\alpha ^{t_\alpha })`$ if for each $`\epsilon >0`$, there exists a real $`M`$ such that
$$lim\; sup\mu _\alpha ^{t_\alpha }(e^{h/t_\alpha }1_{\{h>M\}})<\epsilon \text{for all }h𝒯.$$
For each $`h(X)`$, each $`xX`$ and each $`\epsilon >0`$, we put $`F_{e^{h(x)},\epsilon }=\{yX:e^{h(x)}\epsilon e^{h(y)}e^{h(x)}+\epsilon \}`$ and $`G_{e^{h(x)},\epsilon }=\{yX:e^{h(x)}\epsilon <e^{h(y)}<e^{h(x)}+\epsilon \}`$. The following expressions are known when $`(\mu _\alpha )`$ is a net of probability measures, and when $`𝒯`$ has only one element, say $`h`$ (see and for the first and the second assertion, respectively). The proofs reveal that the constant $`M`$ comes from the above tail condition (assumed to be satisfied by $`h`$), so that the uniform versions for a general $`𝒯`$ follow immediately; they moreover work as well for the sub-probability case.
###### Theorem 1
Let $`𝒯(X)`$ satisfying the tail condition for $`(\mu _\alpha ^{t_\alpha })`$. There is a real $`M`$ such that for each $`h𝒯`$,
$$e^{\underset{¯}{\mathrm{\Lambda }}(h)}=lim\; inf\underset{xX,\epsilon >0}{sup}\{(e^{h(x)}\epsilon )\mu _\alpha ^{t_\alpha }(G_{e^{h(x)},\epsilon })\}=\underset{\epsilon 0}{lim}lim\; inf\underset{x\{hM\}}{sup}\{e^{h(x)}\mu _\alpha ^{t_\alpha }(G_{e^{h(x)},\epsilon })\}$$
and
$$e^{\overline{\mathrm{\Lambda }}(h)}=\underset{xX,\epsilon >0}{sup}\{(e^{h(x)}\epsilon )lim\; sup\mu _\alpha ^{t_\alpha }(G_{e^{h(x)},\epsilon })\}=\underset{x\{hM\},\epsilon >0}{sup}\{(e^{h(x)}\epsilon )lim\; sup\mu _\alpha ^{t_\alpha }(G_{e^{h(x)},\epsilon })\}.$$
In the above expressions, $`G_{e^{h(x)},\epsilon }`$ can be replaced by $`F_{e^{h(x)},\epsilon }`$.
Part (a) of the following theorem shows that under conditions strictly weaker than large deviations, $`\mathrm{\Lambda }(h)`$ exists and has the same form as when large deviations hold, since in this case the rate function coincides with $`l_1`$ (Lemma 1); it can be seen as a vague version of Varadhan’s theorem. Note that the hypothesis $`h𝒞_𝒦(X)`$ cannot be dropped: consider a vague large deviation principle for a net of probability measures with rate function $`J+\mathrm{}`$, take $`h0`$ and get $`\mathrm{\Lambda }(h)=0`$ and $`sup_X\{h(x)J(x)\}=\mathrm{}`$. Note also that the condition $`(ii)`$ holds in particular when $`(\mu _\alpha ^{t_\alpha })`$ converges in $`\mathrm{\Gamma }(X,[0,1])`$.
###### Theorem 2
Let $`𝒯𝒞(X)`$ with $`X`$ locally compact Hausdorff, and assume that the following hold:
* $`𝒯`$ satisfies the tail condition for $`(\mu _\alpha ^{t_\alpha })`$.
* $`lim\; sup\mu _\alpha ^{t_\alpha }(K)lim\; inf\mu _\alpha ^{t_\alpha }(G)`$ for each compact $`KX`$ and each open $`GX`$ with $`KG`$.
* $`inf_{h𝒯}\overline{\mathrm{\Lambda }}(h)>m`$ for some real $`m`$.
The following conclusions hold.
* If $`𝒯𝒞_𝒦(X)`$, then $`\mathrm{\Lambda }(h)`$ exists for all $`h𝒯`$, and there is a real $`M`$ such that
$$\mathrm{\Lambda }(h)=\underset{x\{mhM\}}{sup}\{h(x)l_1(x)\}=\underset{xX}{sup}\{h(x)l_1(x)\}\text{for all }h𝒯.$$
(5)
* If $`(\mu _\alpha )`$ is exponentially tight with respect to $`(t_\alpha )`$, then $`\mathrm{\Lambda }(h)`$ exists for all $`h𝒯`$, and there is a real $`M`$ and a compact $`KX`$ such that
$$\mathrm{\Lambda }(h)=\underset{xK\{mhM\}}{sup}\{h(x)l_1(x)\}=\underset{xX}{sup}\{h(x)l_1(x)\}\text{for all }h𝒯.$$
(6)
###### Proof.
Assume $`𝒯𝒞_𝒦(X)`$. By $`(i)`$ and Theorem 1, there is a real $`M^{}`$ such that for each $`h𝒯`$,
$$\underset{x\{hM^{}+\mathrm{log}2\}}{sup}e^{h(x)}e^{l_1(x)}\underset{xX}{sup}e^{h(x)}e^{l_1(x)}e^{\underset{¯}{\mathrm{\Lambda }}(h)}$$
(7)
$$e^{\overline{\mathrm{\Lambda }}(h)}=\underset{x\{hM^{}\},\epsilon >0}{sup}\{(e^{h(x)}\epsilon )lim\; sup\mu _\alpha ^{t_\alpha }(F_{e^{h(x)},\epsilon })\}.$$
Put $`M=\mathrm{log}2+M^{}`$, and suppose that
$$\underset{x\{hM\}}{sup}e^{h(x)}e^{l_1(x)}+\nu <\underset{x\{hM^{}\},\epsilon >0}{sup}\{(e^{h(x)}\epsilon )lim\; sup\mu _\alpha ^{t_\alpha }(F_{e^{h(x)},\epsilon })\}$$
for some $`h𝒯`$ and some $`\nu >0`$. Then there exists $`x_0\{hM^{}\}`$ and $`\epsilon _0>0`$ with $`e^{h(x_0)}>\epsilon _0`$ such that
$$\underset{x\{hM\}}{sup}e^{h(x)}e^{l_1(x)}<(e^{h(x_0)}\epsilon _0\nu )lim\; sup\mu _\alpha ^{t_\alpha }(F_{e^{h(x_0)},\epsilon _0}).$$
(8)
By continuity and local compactness, for each $`xF_{e^{h(x_0)},\epsilon _0}`$, there exist some open sets $`V_x`$ and $`V_x^{}`$ satisfying $`xV_x\overline{V_x}V_x^{}`$ with $`\overline{V_x}`$ compact, and such that $`e^{h(y)}>e^{h(x_0)}\epsilon _0\nu `$ for all $`yV_x^{}`$. Note that $`h(x)M`$ for each $`xF_{e^{h(x_0)},\epsilon _0}`$, since $`e^{h(x_0)}+\epsilon _0<2e^M^{}`$. By (8), for each $`xF_{e^{h(x_0)},\epsilon _0}`$, there exist some open sets $`W_x`$ and $`W_x^{}`$ satisfying $`xW_x\overline{W_x}W_x^{}`$ with $`\overline{W_x}`$ compact, and such that
$$e^{h(x)}lim\; inf\mu _\alpha ^{t_\alpha }(W_x^{})<(e^{h(x_0)}\epsilon _0\nu )lim\; sup\mu _\alpha ^{t_\alpha }(F_{e^{h(x_0)},\epsilon _0}).$$
(9)
Put $`G_x=W_xV_x`$ for all $`xF_{e^{h(x_0)},\epsilon _0}`$. Since $`F_{e^{h(x_0)},\epsilon _0}`$ is compact, there is a finite set $`AF_{e^{h(x_0)},\epsilon _0}`$ such that $`F_{e^{h(x_0)},\epsilon _0}_{xA}G_x`$; thus, for some $`xA`$ we have
$$(e^{h(x_0)}\epsilon _0\nu )lim\; sup\mu _\alpha ^{t_\alpha }(F_{e^{h(x_0)},\epsilon _0})e^{h(x)}lim\; sup\mu _\alpha ^{t_\alpha }(G_x)$$
$$e^{h(x)}lim\; sup\mu _\alpha ^{t_\alpha }(\overline{W_x})e^{h(x)}lim\; inf\mu _\alpha ^{t_\alpha }(W_x^{})$$
(where the third inequality follows from $`(ii)`$), which contradicts (9). Therefore, all inequalities in (7) are equalities, that is for each $`h𝒯`$, $`\mathrm{\Lambda }(h)`$ exists and
$$\mathrm{\Lambda }(h)=\underset{x\{hM\}}{sup}\{h(x)l_1(x)\}=\underset{xX}{sup}\{h(x)l_1(x)\}=\underset{x\{mhM\}}{sup}\{h(x)l_1(x)\},$$
(where the third equality follows from $`(iii)`$), which proves $`(a)`$. For $`(b)`$, the above proof works verbatim replacing $`\{hM\}`$ and $`F_{e^{h(x_0)},\epsilon _0}`$ by $`\{hM\}K`$ and $`F_{e^{h(x_0)},\epsilon _0}K`$ respectively, where $`K`$ is some compact set given by the exponential tightness. ∎
The following definition extends the usual notion of Legendre-Fenchel transform (when $`X`$ is a real topological vector space and $`𝒮`$ its topological dual) and its generalization proposed in (with $`X=`$ and $`𝒮=\{h_{\lambda ,\nu }:(\lambda ,\nu )^2\}`$); it coincides with our preceding notations since for $`𝒮=\{h_\lambda :\lambda G\}`$ with $`G`$ a nonempty open interval, we have
$$L_{G}^{}{}_{}{}^{}(x)=\underset{\lambda }{sup}\{\lambda x\widehat{L_G}(\lambda )\}=\underset{\lambda G}{sup}\{\lambda xL(\lambda )\}=\underset{\{h_\lambda :\lambda G\}}{sup}\{h_\lambda (x)\mathrm{\Lambda }(h_\lambda )\}=\mathrm{\Lambda }_{S}^{}{}_{}{}^{}(x).$$
In (Corollary 2), we proved that for $`X`$ completely regular (not necessary Hausdorff), a rate function has always the form $`\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}`$, where $`𝒮`$ is any set in $`𝒞(X)`$ stable by translation, separating suitably points and closed sets, and such that each $`h𝒮`$ satisfies the tail condition for $`(\mu _\alpha ^{t_\alpha })`$; this is proved in for $`X`$ normal Hausdorff and $`𝒮`$ the set of all bounded continuous functions on $`X`$ (this case was known under exponential tightness hypothesis as a part of the conclusion of Bryc’s theorem). We will identify in the next section others sets $`𝒮`$ for which the rate function is given by $`\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}`$.
###### Definition 3
Let $`𝒮(X)`$ such that $`\mathrm{\Lambda }(h)`$ exists for all $`h𝒮`$. The map $`\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}`$ defined by
$$\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}(x)=\underset{h𝒮}{sup}\{h(x)\mathrm{\Lambda }(h)\}\text{for all }xX,$$
is the abstract Legendre-Fenchel transform of $`\mathrm{\Lambda }_𝒮`$.
## 4 The case $`X=`$
In this section, we take $`X=`$ and apply Theorem 2 with $`𝒯=\{h_\lambda :\lambda G\}`$ where $`G`$ is a nonempty open interval. This allows us to compare the values of $`l_1^{(\mu _\gamma ^{t_\gamma })}`$ and those of $`L_{G}^{}{}_{}{}^{}`$ on $`\text{ran}L_{G}^{}{}_{}{}^{}\text{ran}L_{G}^{}{}_{+}{}^{}`$, where $`(\mu _\gamma ^{t_\gamma })`$ is a suitable subnet of $`(\mu _\alpha ^{t_\alpha })`$ (Proposition 1). By means of a compactness argument, we then derive sufficient conditions for large deviations, involving only the left and right derivatives of $`L_G`$; the rate function is given by an abstract Legendre-Fenchel transform $`\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}`$ (Theorem 3). The strengthening of Gärtner-Ellis theorem (Corollary 1) and the solution to the Ellis question (Corollary 2) are obtained by taking suitable $`𝒮`$.
###### Proposition 1
Let $`\lambda _0`$, and assume that $`L(\lambda )`$ exists and is finite for all $`\lambda `$ in an open interval $`G`$ containing $`\lambda _0`$. Then, $`(\mu _\alpha ^{t_\alpha })`$ has a subnet $`(\mu _\gamma ^{t_\gamma })`$ such that
$$l_1^{(\mu _\gamma ^{t_\gamma })}(L_{G}^{}{}_{}{}^{}(\lambda _0))\lambda _0L_{G}^{}{}_{}{}^{}(\lambda _0)L(\lambda _0)$$
and
$$l_1^{(\mu _\gamma ^{t_\gamma })}(L_{G}^{}{}_{+}{}^{}(\lambda _0))\lambda _0(L_{G}^{}{}_{+}{}^{}(\lambda _0))L(\lambda _0).$$
Whence,
$$l_1^{(\mu _\gamma ^{t_\gamma })}(x)L_{G}^{}{}_{}{}^{}(x)\text{for all }x\text{ran}L_{G}^{}{}_{}{}^{}\text{ran}L_{G}^{}{}_{+}{}^{}.$$
###### Proof.
Let $`G_0`$ be an open interval such that $`\lambda _0G_0\overline{G_0}G`$. Let $`\lambda _1`$ and $`\lambda _2`$ in $`G\mathtt{\backslash }\{0\}`$ such that $`\lambda _1<\lambda <\lambda _2`$ for all $`\lambda G_0`$. There exists $`\gamma >1`$ such that $`\{\gamma \lambda _1,\gamma \lambda _2\}𝒟\text{om}(L)`$ so that $`h_{\lambda _1}`$ and $`h_{\lambda _2}`$ satisfy (individually) the tail condition by Lemma 4.3.8 of (the proof given there for probability measures works as well for the sub-probability case). Therefore, for each $`\epsilon >0`$ and for each $`i\{1,2\}`$ there exists $`M_{i,\epsilon }`$ such that
$$lim\; sup\mu _\alpha ^{t_\alpha }(e^{h_{\lambda _i}/t_\alpha }1_{\{h_{\lambda _i}>M_{i,\epsilon }\}})<\epsilon .$$
Put $`M_\epsilon =M_{1,\epsilon }M_{2,\epsilon }`$, and get for each $`\lambda G_0`$,
$$_{\{x:\lambda x>M_\epsilon \}}e^{\lambda x/t_\alpha }\mu _\alpha (dx)=_{\{x:\lambda x>M_\epsilon \}_{}}e^{\lambda x/t_\alpha }\mu _\alpha (dx)+_{\{x:\lambda x>M_\epsilon \}_+}e^{\lambda x/t_\alpha }\mu _\alpha (dx)$$
$$_{\{x:\lambda _1x>M_{1,\epsilon }\}_{}}e^{\lambda _1x/t_\alpha }\mu _\alpha (dx)+_{\{x:\lambda _2x>M_{2,\epsilon }\}_+}e^{\lambda _2x/t_\alpha }\mu _\alpha (dx),$$
whence
$$\lambda G_0,lim\; sup\mu _\alpha ^{t_\alpha }(e^{h_\lambda /t_\alpha }1_{\{h_\lambda >M_\epsilon \}})$$
$$lim\; sup\mu _\alpha ^{t_\alpha }(e^{h_{\lambda _1}/t_\alpha }1_{\{h_{\lambda _1}>M_{1,\epsilon }\}})lim\; sup\mu _\alpha ^{t_\alpha }(e^{h_{\lambda _2}/t_\alpha }1_{\{h_{\lambda _2}>M_{2,\epsilon }\}})<\epsilon .$$
It follows that $`\{h_\lambda :\lambda G_0\}`$ satisfies the tail condition for $`(\mu _\alpha ^{t_\alpha })`$. Since $`L_G`$ is continuous and $`\overline{G_0}`$ compact, $`L_{G_0}`$ is bounded and in particular $`inf_{\lambda G_0}L(\lambda )>m`$ for some real $`m`$. Let $`(\mu _\gamma ^{t_\gamma })`$ be a subnet of $`(\mu _\alpha ^{t_\alpha })`$ converging in $`\mathrm{\Gamma }(X,[0,1])`$ (given by Lemma 2), put $`𝒯=\{h_\lambda :\lambda G_0\}`$, and note that all the hypotheses of Theorem 2 hold for $`𝒯`$ and $`(\mu _\gamma ^{t_\gamma })`$, with moreover $`𝒯𝒞_𝒦(X)`$. If $`\lambda _00`$ (say $`\lambda _0>0`$), then $`\lambda _1`$ and $`\lambda _2`$ can be chosen such that $`0<\lambda _1<\lambda <\lambda _2`$ for all $`\lambda G_0`$. Since for each real $`Mm`$, there is a compact $`K_M`$ such that $`_{\lambda G_0}\{mh_\lambda M\}K_M`$, by Theorem 2 (a) we get a compact $`K`$ such that
$$L(\lambda )=\underset{xK}{sup}\{\lambda xl_1^{(\mu _\gamma ^{t_\gamma })}(x)\}\text{for all }\lambda G_0.$$
(10)
If $`\lambda _0=0`$, then $`(\mu _\alpha )`$ (resp. $`(\mu _\gamma )`$) is exponentially tight with respect to $`(t_\alpha )`$ (resp. $`(t_\gamma )`$), and we apply Theorem 2 (b) to get (10). Therefore, for each $`\lambda G_0`$ there exists $`x_\lambda K`$ such that $`L(\lambda )=\lambda x_\lambda l_1^{(\mu _\gamma ^{t_\gamma })}(x_\lambda )`$. Put $`x=L_{G}^{}{}_{+}{}^{}(\lambda _0)`$, and let $`(x_{\lambda ^{}+\lambda _0})`$ be a subnet of $`(x_{\lambda +\lambda _0})_{\lambda +\lambda _0G_0,\lambda >0}`$. Since $`x_{\lambda +\lambda _0}K`$ for all $`\lambda +\lambda _0G_0`$, $`(x_{\lambda ^{}+\lambda _0})`$ has a subnet $`(x_{\lambda ^{\prime \prime }+\lambda _0})`$ converging to some point $`x^{\prime \prime }K`$ when $`\lambda ^{\prime \prime }0^+`$, so that
$$x=\underset{\lambda ^{\prime \prime }0^+}{lim}\frac{L(\lambda ^{\prime \prime }+\lambda _0)L(\lambda _0)}{\lambda ^{\prime \prime }}=\underset{\lambda ^{\prime \prime }0^+}{lim}\frac{(\lambda ^{\prime \prime }+\lambda _0)x_{\lambda ^{\prime \prime }+\lambda _0}l_1^{(\mu _\gamma ^{t_\gamma })}(x_{\lambda ^{\prime \prime }+\lambda _0})L(\lambda _0)}{\lambda ^{\prime \prime }}$$
$$=x^{\prime \prime }+\underset{\lambda ^{\prime \prime }0^+}{lim}\frac{\lambda _0x_{\lambda ^{\prime \prime }+\lambda _0}l_1^{(\mu _\gamma ^{t_\gamma })}(x_{\lambda ^{\prime \prime }+\lambda _0})L(\lambda _0)}{\lambda ^{\prime \prime }},$$
which implies $`x^{\prime \prime }=x`$ and
$$0=\underset{\lambda ^{\prime \prime }0^+}{lim}\lambda _0x_{\lambda ^{\prime \prime }+\lambda _0}l_1^{(\mu _\gamma ^{t_\gamma })}(x_{\lambda ^{\prime \prime }+\lambda _0})L(\lambda _0)\lambda _0xl_1^{(\mu _\gamma ^{t_\gamma })}(x)L(\lambda _0),$$
which proves the assertion concerning $`L_{G}^{}{}_{+}{}^{}(\lambda _0)`$. A similar proof works for $`L_{G}^{}{}_{}{}^{}(\lambda _0)`$. ∎
###### Theorem 3
Let $`𝒮𝒞(X)`$ and $`GX`$ be a nonempty open interval such that $`𝒮\{h_\lambda :\lambda G\}`$, and assume that $`\mathrm{\Lambda }(h)`$ exists for all $`h𝒮`$ with $`L(\lambda )`$ finite for all $`\lambda G`$.
* If
$$\text{ran}L_{G}^{}{}_{}{}^{}\text{ran}L_{G}^{}{}_{+}{}^{}𝒟\text{om}(l_0)\{l_1>\overline{\mathrm{\Lambda }}(0)\},$$
(11)
then $`(\mu _\alpha )`$ satisfies a vague large deviation principle with powers $`(t_\alpha )`$ and rate function $`J`$ satisfying
$$J(x)=L_{G}^{}{}_{}{}^{}(x)=\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}(x)\text{for all }x𝒟\text{om}(J)\{J>\overline{\mathrm{\Lambda }}(0)\}.$$
(12)
If moreover $`0G`$, then the principle is narrow and
$$J(x)=L_{G}^{}{}_{}{}^{}(x)=\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}(x)\text{for all }x𝒟\text{om}(J).$$
(13)
* If
$$\text{ran}L_{G}^{}{}_{}{}^{}\text{ran}L_{G}^{}{}_{+}{}^{}𝒟\text{om}(l_0),$$
(14)
then $`(\mu _\alpha )`$ satisfies a vague large deviation principle with powers $`(t_\alpha )`$ and rate function $`J`$ satisfying
$$J(x)=L_{G}^{}{}_{}{}^{}(x)=\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}(x)\text{for all }x𝒟\text{om}(J).$$
(15)
If moreover $`0G`$, then the principle is narrow.
* If
$$\text{ran}L_{G}^{}{}_{}{}^{}\text{ran}L_{G}^{}{}_{+}{}^{}𝒟\text{om}(\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{})\{l_1>\overline{\mathrm{\Lambda }}(0)\},$$
(16)
then $`(\mu _\alpha )`$ satisfies a vague large deviation principle with powers $`(t_\alpha )`$ and rate function $`J`$ satisfying
$$J(x)=\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}(x)\text{for all }x\{J>\overline{\mathrm{\Lambda }}(0)\},$$
(17)
and
$$J(x)=L_{G}^{}{}_{}{}^{}(x)\text{for all }x𝒟\text{om}(\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{})\{J>\overline{\mathrm{\Lambda }}(0)\}.$$
(18)
If moreover $`0G`$, then the principle is narrow with $`J=\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}`$ satisfying
$$J(x)=L_{G}^{}{}_{}{}^{}(x)\text{for all }x𝒟\text{om}(J).$$
(19)
* If
$$\text{ran}L_{G}^{}{}_{}{}^{}\text{ran}L_{G}^{}{}_{+}{}^{}𝒟\text{om}(\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}),$$
(20)
then $`(\mu _\alpha )`$ satisfies a vague large deviation principle with powers $`(t_\alpha )`$ and rate function $`J=\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}`$ satisfying
$$J(x)=L_{G}^{}{}_{}{}^{}(x)\text{for all }x𝒟\text{om}(J).$$
(21)
If moreover $`0G`$, then the principle is narrow.
* If $`l_0`$ is proper convex, then $`(a)`$ (resp. $`(b)`$) holds verbatim replacing the symbol $`𝒟\text{om}`$ by $`\text{int}𝒟\text{om}`$ in (11), (12), (13) (resp. (14), (15)).
* If $`\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}`$ is proper convex and lower semi-continuous, then $`(c)`$ (resp. $`(d)`$) holds verbatim replacing the symbol $`𝒟\text{om}`$ by $`\text{int}𝒟\text{om}`$ in (16), (18), (19)) (resp. (20), (21)).
###### Proof.
For all $`h𝒮`$ and all $`xX`$ we have by Theorem 1 (since $`\mathrm{\Lambda }(h)\overline{\mathrm{\Lambda }}(h1_{\{hM\}}+(\mathrm{})1_{\{h>M\}})`$ for all reals $`M`$),
$$\mathrm{\Lambda }(h)h(x)\underset{M}{sup}\underset{\{hM\}}{sup}\{h(y)l_0(y)\}h(x)\underset{yX}{sup}\{h(y)l_0(y)\}h(x)l_0(x),$$
so that
$$L_{G}^{}{}_{}{}^{}(x)\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}(x)l_0(x)\text{for all }xX.$$
(22)
Assume that (11) holds, and let $`(\mu _\beta ^{t_\beta })`$ be a subnet of $`(\mu _\alpha ^{t_\alpha })`$. By Proposition 1 applied to $`(\mu _\beta ^{t_\beta })`$ in place of $`(\mu _\alpha ^{t_\alpha })`$, $`(\mu _\beta ^{t_\beta })`$ has a subnet $`(\mu _\gamma ^{t_\gamma })`$ such that
$$l_1^{(\mu _\gamma ^{t_\gamma })}(x)L_{G}^{}{}_{}{}^{}(x)\text{for all }x𝒟\text{om}(l_0)\{l_1>\overline{\mathrm{\Lambda }}(0)\}.$$
(23)
Since
$$l_0l_0^{(\mu _\gamma ^{t_\gamma })}l_1^{(\mu _\gamma ^{t_\gamma })},$$
(24)
(22) and (23) imply
$$l_0^{(\mu _\gamma ^{t_\gamma })}(x)=l_1^{(\mu _\gamma ^{t_\gamma })}(x)=L_{G}^{}{}_{}{}^{}(x)=\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}(x)=l_0(x)\text{for all }x𝒟\text{om}(l_0)\{l_1>\overline{\mathrm{\Lambda }}(0)\}.$$
(25)
If $`x𝒟\text{om}(l_0)`$, then $`l_0^{(\mu _\gamma ^{t_\gamma })}(x)=l_1^{(\mu _\gamma ^{t_\gamma })}(x)=+\mathrm{}`$ by (24). If $`l_1(x)\overline{\mathrm{\Lambda }}(0)`$, then
$$l_0^{(\mu _\gamma ^{t_\gamma })}(x)=l_1^{(\mu _\gamma ^{t_\gamma })}(x)=l_0(x)=l_1(x)=\overline{\mathrm{\Lambda }}(0).$$
Therefore, $`l_0^{(\mu _\gamma ^{t_\gamma })}(x)=l_1^{(\mu _\gamma ^{t_\gamma })}(x)`$ for all $`xX`$. By Lemma 1 applied to $`(\mu _\gamma ^{t_\gamma })`$, $`(\mu _\gamma )`$ satisfies a vague large deviation principle with powers $`(t_\gamma )`$ and rate function
$$J(x)=\{\begin{array}{cc}\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}\hfill & \text{if}x𝒟\text{om}(l_0)\{l_1>\overline{\mathrm{\Lambda }}(0)\}\hfill \\ & \\ \overline{\mathrm{\Lambda }}(0)\hfill & \text{if}l_1(x)\overline{\mathrm{\Lambda }}(0)\hfill \\ & \\ +\mathrm{}\hfill & \text{if}x𝒟\text{om}(l_0).\hfill \end{array}$$
(26)
By Lemma 2 (b), $`(\mu _\gamma ^{t_\gamma })`$ converges to $`c_J`$ in $`\mathrm{\Gamma }(X,[0,1])`$. Since $`(\mu _\beta ^{t_\beta })`$ is arbitrary, we have proved that any subnet of $`(\mu _\alpha ^{t_\alpha })`$ has a subnet converging vaguely to $`c_J`$. By Lemma 2 (a), it follows that $`(\mu _\alpha ^{t_\alpha })`$ converges vaguely to $`c_J`$, which proves the first assertion of (a) ((12) follows from (25) and (26), since $`J=l_0=l_1`$). If $`0G`$, then (13) follows from (22) and (26) since $`L(0)L_{G}^{}{}_{}{}^{}`$, and the principle is narrow by exponential tightness. The proofs of (b),(c),(d) are similar. Assume that $`l_0`$ is proper convex, and
$$\text{ran}L_{G}^{}{}_{}{}^{}\text{ran}L_{G}^{}{}_{+}{}^{}\text{int}𝒟\text{om}(l_0)\{l_1>\overline{\mathrm{\Lambda }}(0)\}.$$
In the same way as above we get
$$l_0^{(\mu _\gamma ^{t_\gamma })}(x)=l_1^{(\mu _\gamma ^{t_\gamma })}(x)=L_{G}^{}{}_{}{}^{}(x)=\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}(x)=l_0(x)\text{for all }x\text{int}𝒟\text{om}(l_0)\{l_1>\overline{\mathrm{\Lambda }}(0)\}.$$
(27)
Suppose that $`l_1^{(\mu _\gamma ^{t_\gamma })}(x)>l_0(x)`$ for some $`x\{l_1>\overline{\mathrm{\Lambda }}(0)\}`$. Since $`l_1`$ and $`l_1^{(\mu _\gamma ^{t_\gamma })}`$ are lower semi-continuous, there is an open set $`G_0`$ containing $`x`$ such that
$$\underset{G_0\{l_1>\overline{\mathrm{\Lambda }}(0)\}}{inf}l_1^{(\mu _\gamma ^{t_\gamma })}>\underset{G_0\{l_1>\overline{\mathrm{\Lambda }}(0)\}}{inf}l_0=\underset{G_0\{l_1>\overline{\mathrm{\Lambda }}(0)\}\text{int}𝒟\text{om}(l_0)}{inf}l_0,$$
where the equality follows from Lemma 3 applied to $`l_0`$ and $`G_0\{l_1>\overline{\mathrm{\Lambda }}(0)\}`$. Then, there exists $`yG_0\{l_1>\overline{\mathrm{\Lambda }}(0)\}\text{int}𝒟\text{om}(l_0)`$ such that $`l_1^{(\mu _\gamma ^{t_\gamma })}(y)>l_0(y)`$, which contradicts (27). We then have $`l_1^{(\mu _\gamma ^{t_\gamma })}(x)l_0(x)`$ for all $`x\{l_1>\overline{\mathrm{\Lambda }}(0)\}`$, and by (24),
$$l_0^{(\mu _\gamma ^{t_\gamma })}(x)=l_1^{(\mu _\gamma ^{t_\gamma })}(x)=l_0(x)\text{for all }x\{l_1>\overline{\mathrm{\Lambda }}(0)\}.$$
Since
$$l_0^{(\mu _\gamma ^{t_\gamma })}(x)=l_1^{(\mu _\gamma ^{t_\gamma })}(x)=l_0(x)=l_1(x)=\overline{\mathrm{\Lambda }}(0)\text{for all }x\{l_1\overline{\mathrm{\Lambda }}(0)\},$$
it follows as above that $`(\mu _\alpha ^{t_\alpha })`$ converges vaguely to $`c_J`$, with $`J=l_0=l_1`$ satisfying by (27),
$$J(x)=L_{G}^{}{}_{}{}^{}(x)=\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}(x)\text{for all }x\text{int}𝒟\text{om}(J)\{J>\overline{\mathrm{\Lambda }}(0)\}.$$
(28)
If $`0G`$, then $`L(0)L_{G}^{}{}_{}{}^{}`$, and by (22) and (28) we get
$$J(x)=L_{G}^{}{}_{}{}^{}(x)=\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}(x)\text{for all }x\text{int}𝒟\text{om}(J).$$
This proves the assertion of (e) concerning (a); the one concerning (b) is proved similarly. Assume that $`\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}`$ is proper convex lower semi-continuous, and
$$\text{ran}L_{G}^{}{}_{}{}^{}\text{ran}L_{G}^{}{}_{+}{}^{}\text{int}𝒟\text{om}(\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{})\{l_1>\overline{\mathrm{\Lambda }}(0)\}.$$
As above we get
$$l_0^{(\mu _\gamma ^{t_\gamma })}(x)=l_1^{(\mu _\gamma ^{t_\gamma })}(x)=L_{G}^{}{}_{}{}^{}(x)=\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}(x)\text{for all }x\text{int}𝒟\text{om}(\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{})\{l_1>\overline{\mathrm{\Lambda }}(0)\}.$$
(29)
The same reasoning as in the proof of (e) (with $`\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}`$ in place of $`l_0`$) gives $`l_1^{(\mu _\gamma ^{t_\gamma })}(x)\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}(x)`$ for all $`x\{l_1>\overline{\mathrm{\Lambda }}(0)\}`$, and by (22),
$$l_0^{(\mu _\gamma ^{t_\gamma })}(x)=l_1^{(\mu _\gamma ^{t_\gamma })}(x)=\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}(x)=l_0(x)\text{for all }x\{l_1>\overline{\mathrm{\Lambda }}(0)\}.$$
(30)
Since
$$l_0^{(\mu _\gamma ^{t_\gamma })}(x)=l_1^{(\mu _\gamma ^{t_\gamma })}(x)=\overline{\mathrm{\Lambda }}(0)\text{for all }x\{l_1\overline{\mathrm{\Lambda }}(0)\},$$
it follows as above that $`(\mu _\alpha ^{t_\alpha })`$ converges vaguely to $`c_J`$, with $`J`$ satisfying (17). Since $`J=l_1`$, (29) gives
$$J(x)=L_{G}^{}{}_{}{}^{}(x)\text{for all }x\text{int}𝒟\text{om}(\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{})\{J>\overline{\mathrm{\Lambda }}(0)\}.$$
(31)
Since $`0G`$ implies $`L(0)L_{G}^{}{}_{}{}^{}`$, by (22), (30), (31), we obtain $`J=\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}`$ and
$$J(x)=L_{G}^{}{}_{}{}^{}(x)\text{for all }x\text{int}𝒟\text{om}(\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}).$$
This proves the assertion of (f) concerning (c); the one concerning (d) is proved similarly. ∎
The standard Gärtner-Ellis theorem deals with the case where $`(\mu _\alpha )`$ is a sequence of Borel probability measures; it states that if $`L(\lambda )`$ exists for all reals $`\lambda `$, $`L`$ is lower semi-continuous essentially smooth and $`0\text{int}𝒟\text{om}(L)`$, then $`(\mu _\alpha )`$ satisfies a large deviation principle with powers $`(t_\alpha )`$ and rate function $`L^{}`$ (, Theorem 2.3.6, , ). A stronger version has been given by O’ Brien (, Theorem 5.1): if $`L(\lambda )`$ exists and is finite for all $`\lambda `$ in a nonempty open interval $`G`$ and if $`\widehat{L_G}`$ is essentially smooth, then $`(\mu _\alpha )`$ satisfies a vague large deviation principle with powers $`(t_\alpha )`$ and rate function $`L_{G}^{}{}_{}{}^{}`$; if moreover $`0G`$, then the principle is narrow. The former version is recovered by taking $`G=\text{int}𝒟\text{om}(L)`$ (the hypotheses implying $`L^{}=L_{G}^{}{}_{}{}^{}`$ with $`\widehat{L_G}`$ essentially smooth). The improvements consists in the obtention of the vague large deviations, and in the fact that $`L`$ in not assumed to exist out $`G`$ (even when $`L`$ exists on $`X`$, it is not assumed to be lower semi-continuous).
The following corollary summarizes the case where $`𝒮=\{h_\lambda :\lambda G\}`$ in Theorem 3, and where large deviations hold with rate function $`L_{G}^{}{}_{}{}^{}`$ ($`=\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}`$). It strengthens the O’ Brien’s version of Gärtner-Ellis theorem by obtaining the same conclusions, with the essential smoothness hypothesis replaced by the weaker condition (32) (or (33) when $`0G`$); in particular, there is no differentiability assumption. Furthermore, it works for general nets of Radon sub-probability measures.
###### Corollary 1
We assume that $`L(\lambda )`$ exists and is finite for all $`\lambda `$ in a nonempty open interval $`GX`$.
* If
$$\text{ran}L_{G}^{}{}_{}{}^{}\text{ran}L_{G}^{}{}_{+}{}^{}\text{int}𝒟\text{om}(L_{G}^{}{}_{}{}^{}),$$
(32)
then $`(\mu _\alpha )`$ satisfies a vague large deviation principle with powers $`(t_\alpha )`$ and rate function $`L_{G}^{}{}_{}{}^{}`$. The condition (32) is satisfied in particular when $`\widehat{L_G}`$ is essentially smooth.
* If $`0G`$ and
$$\text{ran}L_{G}^{}{}_{}{}^{}\text{ran}L_{G}^{}{}_{+}{}^{}\text{int}𝒟\text{om}(L_{G}^{}{}_{}{}^{})\{l_1>L(0)\},$$
(33)
then $`(\mu _\alpha )`$ satisfies a large deviation principle with powers $`(t_\alpha )`$ and rate function $`L_{G}^{}{}_{}{}^{}`$.
###### Proof.
(b) and the first assertion of (a) follow from Theorem 3 (f) with $`𝒮=\{h_\lambda :\lambda G\}`$. Assume that $`\widehat{L_G}`$ is essentially smooth. Extend $`L_G`$ by continuity to a convex function $`L_{\overline{G}}`$ on $`\overline{G}`$, so that $`\widehat{L_{\overline{G}}}`$ is a proper convex lower semi-continuous function on $`X`$ with $`G=\text{int}𝒟\text{om}(\widehat{L_{\overline{G}}})`$; moreover, $`\widehat{L_{\overline{G}}}`$ is essentially smooth. By Theorem 26.1 and Corollary 26.4.1 of ,
$$\text{ran}L_{\overline{G}}^{}{}_{}{}^{}\text{int}𝒟\text{om}(L_{\overline{G}}^{}{}_{}{}^{}),$$
(34)
which gives (32) since $`\text{ran}L_{\overline{G}}^{}{}_{}{}^{}=\text{ran}L_{G}^{}{}_{}{}^{}`$ and $`L_{\overline{G}}^{}{}_{}{}^{}=L_G^{}`$. ∎
The solution to the Ellis question (with in fact weaker hypotheses) is a direct consequence of Theorem 3 $`(c)`$, by taking $`𝒮=\{h_{\lambda ,\nu }:(\lambda ,\nu )^2\}`$.
###### Corollary 2
Put $`𝒮=\{h_{\lambda ,\nu }:(\lambda ,\nu )^2\}`$, and assume that $`\mathrm{\Lambda }(h_{\lambda ,\nu })`$ exists for all $`(\lambda ,\nu )^2`$ and is finite for all pairs $`(\lambda ,\lambda )`$ with $`\lambda `$ in some open interval $`G`$ containing $`0`$. If $`\text{ran}L_{G}^{}{}_{}{}^{}\text{ran}L_{G}^{}{}_{+}{}^{}𝒟\text{om}(\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{})\{l_1>L(0)\}`$, then $`(\mu _\alpha )`$ satisfies a large deviation principle with powers $`(t_\alpha )`$ and rate function $`J=\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}`$. Moreover,
$$J(x)=L_{G}^{}{}_{}{}^{}(x)\text{for all }x𝒟\text{om}(J).$$
The following example is often cited as a typical case not covered by the Gärtner-Ellis theorem (, ).
###### Example 1
Consider the sequence $`(\mu _n^{1/n})`$ where $`\mu _n\{1\}=\mu _n\{1\}=\frac{1}{2}`$ for all $`n`$. Then $`L(\lambda )=|\lambda |`$ for all reals $`\lambda `$. Take $`𝒮=\{h_{\lambda ,\nu }:(\lambda ,\nu )^2\}`$ and compute
$$\mathrm{\Lambda }(h_{\lambda ,\nu })=\lambda \nu \text{for all }(\lambda ,\nu )^2,$$
whence
$$\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}(x)=\{\begin{array}{cc}0\hfill & \text{if}|x|=1\hfill \\ +\mathrm{}\hfill & \text{if}|x|1.\hfill \end{array}$$
Then, $`\text{ran}L_{}^{}\text{ran}L_+^{}=\{1,1\}𝒟\text{om}(\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{})`$, and by Corollary 2, $`(\mu _n)`$ satisfies a large deviation principle with powers $`(1/n)`$ and rate function $`J=\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}`$. Since
$$L^{}(x)=\{\begin{array}{cc}0\hfill & \text{if}|x|1\hfill \\ +\mathrm{}\hfill & \text{if}|x|>1,\hfill \end{array}$$
we have $`J(x)=L^{}(x)`$ for all $`x\{1,1\}=𝒟\text{om}(J)`$. Note that for any nonempty open set $`G]1,1[`$,
$$\text{ran}L_{G}^{}{}_{}{}^{}\text{ran}L_{G}^{}{}_{+}{}^{}\text{int}𝒟\text{om}(L_G^{})\{J>0\}]1,1[,$$
and the condition (33) of Corollary 1 does not hold.
The following example exhibits a situation with convex rate function, where both above corollaries do not work; we then apply theorem 3 with another set $`𝒮`$.
###### Example 2
Consider the net $`(\mu _\epsilon ^\epsilon )_{\epsilon >0}`$, where $`\mu _\epsilon `$ is the probability measure on $`X`$ defined by $`\mu _\epsilon (0)=12p_\epsilon `$, $`\mu _\epsilon (\epsilon \mathrm{log}p_\epsilon )=\mu _\epsilon (\epsilon \mathrm{log}p_\epsilon )=p_\epsilon `$, and assume that $`lim\epsilon \mathrm{log}p_\epsilon =\mathrm{}`$. Put $`Q_n(x)=n|x|e^{|x|}x`$ for all $`n`$ and all $`xX`$, and take $`𝒮=\{Q_n:n\}\{h_\lambda :\lambda ]1,1[\}`$. Easy calculations give $`\mathrm{\Lambda }(Q_n)=0`$ for all $`n`$, and
$$L(\lambda )=\{\begin{array}{cc}0\hfill & \text{if }|\lambda |1\hfill \\ +\mathrm{}\hfill & \text{if }|\lambda |>1,\hfill \end{array}$$
so that
$$L_{]1,1[}^{}{}_{}{}^{}(x)=L^{}(x)=|x|\text{for all }xX,$$
and
$$\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}(x)=\underset{n}{sup}\{Q_n(x)\mathrm{\Lambda }(Q_n)\}L_{]1,1[}^{}{}_{}{}^{}(x)=\{\begin{array}{cc}0\hfill & \text{if }x=0\hfill \\ +\mathrm{}\hfill & \text{otherwise}.\hfill \end{array}$$
Then, $`\text{ran}L_{]1,1[}^{}{}_{}{}^{}=\{0\}𝒟\text{om}(\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{})`$, and by Theorem 3 $`(d)`$, $`(\mu _\epsilon )`$ satisfies a large deviation principle with powers $`(\epsilon )_{\epsilon >0}`$ and rate function $`J=\mathrm{\Lambda }_{𝒮}^{}{}_{}{}^{}`$. Note that $`J`$ is convex but $`JL^{}`$ (however, $`J`$ coincides with $`L^{}`$ on $`𝒟\text{om}(J)`$); in particular, $`L`$ is not essentially smooth and the Gärtner-Ellis theorem does not work. Furthermore, for any nonempty open set $`G]1,1[`$,
$$\{0\}=\text{ran}L_{G}^{}{}_{}{}^{}\text{int}𝒟\text{om}(L_{G}^{}{}_{}{}^{})\{J>0\}X\mathtt{\backslash }\{0\}$$
and the condition (33) of Corollary 1 does not hold either. We observe also that Corollary 2 does not apply; indeed, the set $`\{h_{\lambda ,\nu }:(\lambda ,\nu )^2\}`$ is not suitable since
$$\mathrm{\Lambda }(h_{\lambda ,\nu })=\{\begin{array}{cc}0\hfill & \text{if }\lambda 1\text{ and }\nu 1\hfill \\ +\mathrm{}\hfill & \text{otherwise}\hfill \end{array}$$
gives $`\mathrm{\Lambda }_{\{h_{\lambda ,\nu }:(\lambda ,\nu )^2\}}^{}{}_{}{}^{}(x)=L^{}(x)`$ for all $`xX`$.
|
warning/0506/astro-ph0506297.html
|
ar5iv
|
text
|
# 1 Fourier frequency spectra showing the consecutive steps of prewhitening for the four known 𝛽 Cephei stars in the ASAS-3 catalogue: IL Vel, NSV 24078, V 1449 Aql, and SY Equ.
Pulsating Stars in the ASAS-3 Database
I. $`\beta `$ Cephei Stars
by
A. P i g u l s k i
Instytut Astronomiczny Uniwersytetu Wrocławskiego,
Kopernika 11, 51-622 Wrocław, Poland
E-mail: pigulski@astro.uni.wroc.pl
Received 2005 May 30
ABSTRACT
We present results of an analysis of the ASAS-3 data for short-period variables from the recently published catalogue of over 38 000 stars. Using the data available in the literature we verify the results of the automatic classification related to $`\beta `$ Cephei pulsators. In particular, we find that 14 stars in the catalogue can be classified unambiguously as new $`\beta `$ Cephei stars. By means of periodogram analysis we derive the frequencies and amplitudes of the excited modes. The main modes in the new $`\beta `$ Cephei stars have large semi-amplitudes, between 35 and 80 mmag. Up to four modes were found in some stars. Two (maybe three) new $`\beta `$ Cephei stars are members of southern young open clusters: ASAS 164409$``$4719.1 belongs to NGC 6200, ASAS 164630$``$4701.2 is a member of Hogg 22, and ASAS 164939$``$4431.7 could be a member of NGC 6216.
We also analyze the photometry of four known $`\beta `$ Cephei stars in the ASAS-3 catalogue, namely IL Vel, NSV 24078, V1449 Aql and SY Equ. Finally, we discuss the distribution of $`\beta `$ Cephei stars in the Galaxy.
Keywords: stars: $`\beta `$ Cephei – stars: pulsations – stars: classification
1. Introduction
The idea of monitoring the whole sky for variability with small wide-field cameras, initiated and popularized by Paczyński (1997, 2000, 2001), resulted in several successful projects of which the All Sky Automated Survey (ASAS) conducted by Dr. Grzegorz Pojmański (Pojmański 1997) is probably the most fruitful. At the third stage of the project, ASAS-3, its ultimate goal was achieved: the whole southern sky and partly the northern sky were monitored for variability. The preliminary lists of variable stars resulting from the analysis of the photometric data obtained within the ASAS-3 project have been recently published in a series of four papers (Pojmański 2002, 2003; Pojmański and Maciejewski 2004, 2005) and are available in the Internet. Hereafter, we will refer to the tables and data presented in these papers as the ASAS-3 catalogue. The catalogue includes over 38 000 variable stars brighter than $``$14 mag in $`V`$.
The $`V`$-band photometry of the ASAS-3 catalogue forms an excellent homogeneous source of data that can be used to study properties of different types of variable stars in the Galaxy. In this series of papers, we will present the analysis of the ASAS-3 photometry starting with the short-period early B-type pulsators, the $`\beta `$ Cephei stars.
According to the newest review paper on $`\beta `$ Cephei stars (Stankov and Handler 2005), 93 certain objects of this type are presently known in the Galaxy. About half of them were found in young open clusters and OB associations; many, including the prototype, are naked-eye objects. Typical periods range from 3 to 7 hours, the semi-amplitudes are usually smaller than 0.03 mag in $`V`$. Multiperiodic behavior is rather a rule than exception. The pulsations are identified with radial and nonradial low-order $`p`$ and/or $`g`$ modes.
From the analysis of the ASAS-3 $`V`$-band photometry and the information available in the literature, we will show that 18 objects in the ASAS-3 database are bona fide $`\beta `$ Cephei stars, of which only four were previously known. The remaining 14 variable stars are new large-amplitude members of this interesting group of pulsators.
2. Selection of Objects
A preliminary selection and classification of variable stars were already done by the authors of the ASAS-3 catalogue. Since several million stars were monitored during the ASAS-3 survey, fast algorithm for extracting variable objects had to be implemented. For this reason, the first selection of variables was made by the authors of the catalogue using the magnitude–dispersion diagram. Only stars showing an excess of dispersion were selected as subjects of the subsequent periodogram analysis. The analysis allowed to find dominant periodicity reported in the catalogue. Initially, the classification was made using Fourier coefficients, periods, amplitudes and visual inspection of the light curves (Pojmański 2002). Starting with the second paper of the series (Pojmański 2003), the authors of the catalogue used additional information, i.e., IRAS and 2MASS infrared photometry and galactic coordinates, to assign automatically the periodic variable stars to about a dozen predefined classes. This classification was only provisional, but could be very helpful in extracting objects of a given type. The classification types were not exclusive, that is, many stars were assigned to more than one class.
The class BCEP, denoting $`\beta `$ Cephei stars, was included in the classification scheme in the third part of the ASAS-3 catalogue (Pojmański and Maciejewski 2004). Eight BCEP stars were reported in that paper, another 14 were added in the fourth paper (Pojmański and Maciejewski 2005). Adding four stars with positive declinations that appear only in the Internet version of the ASAS-3 catalogue, ten stars with secondary or tertiary BCEP classification, and the known $`\beta `$ Cephei star IL Vel classified as DSCT, denoting $`\delta `$ Scuti star, we get 37 stars that appear in the Internet list of $`\beta `$ Cephei stars in the ASAS-3 catalogue. As we will show below, only 12 of them can be actually classified as bona fide $`\beta `$ Cephei stars. On the other hand, some $`\beta `$ Cephei stars could be included in the other classes, DSCT being the most probable one. The shortest periods observed in $`\beta `$ Cephei stars are of the order of 0.06 d, the median value for all known Galactic objects is about 0.17 d (Stankov and Handler 2005). Periods of $`\delta `$ Scuti stars are, on the average, shorter. Their shortest periods are equal to about 0.02 d, but there are $`\delta `$ Scuti stars with periods longer than 0.2 d (Rodríguez and Breger 2002). Both classes show multiperiodic behavior and small amplitudes. Because of the wide range of overlapping periods, $`\beta `$ Cephei and $`\delta `$ Scuti stars cannot be distinguished solely from the observed periods. The only subgroup of $`\delta `$ Scuti stars that is relatively easy to indicate using periods and light curves are the high-amplitude $`\delta `$ Scuti (HADS) stars. They show large-amplitude variations, characteristic shape of the light curve and, typically, only one or two radial modes.
The easiest method to distinguish between $`\beta `$ Cephei and $`\delta `$ Scuti stars is using the information on their MK spectral types. This is because $`\beta `$ Cephei stars cover a very narrow range of spectral types, between B0 and B2.5, while $`\delta `$ Scuti pulsators are A or early F-type stars. Some photometric methods can also be used. For example, the two classes separate well in the two-color ($`UB`$) vs. ($`BV`$) diagram. Unfortunately, the $`UBV`$ photometry is available only for some ASAS stars. Pojmański and Maciejewski (2004, 2005) used a homogeneous set of infrared measurements from the 2MASS survey as an additional filter in the classification. In particular, they used ($`JH`$) vs. ($`HK`$) and $`\mathrm{log}`$(period) vs. ($`JH`$) diagrams. While some $`\beta `$ Cephei stars with small reddenings can be separated from $`\delta `$ Scuti objects in this way, this method fails for more reddened stars.
In this situation, we decided to proceed in the following way. First, we analyzed the ASAS-3 photometry of all 37 stars that were classified as BCEP in the ASAS-3 catalogue. We also checked their $`UBV`$ photometry (if available), H$`\beta `$ photometry and spectral types. In this way, we found that 12 stars from this sample can be reliably classified as $`\beta `$ Cephei pulsators. Four of them were already known as stars of this type. Furthermore, we checked the available MK spectral types and $`UBV\beta `$ photometry for all stars in the ASAS-3 catalogue that had: (i) full amplitude, $`\mathrm{\Delta }V`$, smaller than 0.3 mag (note that BW Vul, $`\beta `$ Cephei star with the exceptionally large amplitude has $`\mathrm{\Delta }V`$ 0.2 mag), and (ii) dominating period shorter than 0.5 d (for all classes of pulsating stars as well as the variables classified as ACV and MISC) or 1 d (for three classes of eclipsing binaries: EC, ESD, ED). There were about 1700 stars in the ASAS-3 catalogue satisfying both criteria. As a result, we found six additional $`\beta `$ Cephei stars. All eighteen stars of this type found in the ASAS-3 database are listed in Table 1.
T a b l e 1
$`\beta `$ Cephei stars in the ASAS-3 database
| HD | CPD/BD | ASAS name | ASAS classification | $`V`$ | MK sp. type | Notes |
| --- | --- | --- | --- | --- | --- | --- |
| 80383 | $``$52<sup>o</sup>2185 | 091731–5250.3 | DSCT | 9.14 | B2 III (1) | IL Vel |
| 164340 | $``$40<sup>o</sup>8357 | 180233–4005.2 | BCEP/EC/ESD | 9.28 | B0 III (2) | NSV 24078 |
| 180642 | $`+`$00<sup>o</sup>4159 | 191715+0103.6 | BCEP | 8.27 | B1.5 II-III (3) | V1449 Aql |
| 203664 | $`+`$09<sup>o</sup>4973 | 212329+0955.9 | BCEP/EC/ESD | 8.57 | B0.5 IIIn (3) | SY Equ |
| | $``$62<sup>o</sup>2707 | 122213–6320.8 | BCEP | 10.06 | B2 III (4) | |
| 133823 | $``$65<sup>o</sup>2993 | 150955–6530.4 | BCEP | 9.62 | B2 IV (4) | |
| | $``$50<sup>o</sup>9210 | 161858–5103.5 | BCEP/EC | 10.33 | B2 II (5) | |
| 328862 | $``$47<sup>o</sup>7861 | 164409–4719.1 | BCEP/DSCT | 10.13 | B0.5 III (6) | in NGC 6200 |
| | $``$46<sup>o</sup>8213 | 164630–4701.2 | RRC/DSCT/EC/ESD | 10.86 | | in Hogg 22 |
| 328906 | | 164939–4431.7 | DSCT/EC/ESD | 11.22 | \[B2\] (7) | in NGC 6216? |
| 152077 | $``$43<sup>o</sup>7731 | 165314–4345.0 | BCEP/DSCT | 9.08 | B1 II (4) | |
| 152477 | $``$47<sup>o</sup>7958 | 165554–4808.8 | ESD/RRC/EC | 9.03 | B1 II (4) | |
| 155336 | $``$32<sup>o</sup>4389 | 171218–3306.1 | BCEP/DSCT | 9.46 | B1/2 Ib (8) | |
| 165582 | $``$34<sup>o</sup>7600 | 180808–3434.5 | BCEP | 9.39 | B1 II (4) | |
| 167743 | $``$15<sup>o</sup>4909 | 181716–1527.1 | BCEP=DSCT | 9.59 | B2 Ib (9) | |
| | | 182610–1704.3 | DSCT | 10.21 | | ALS 5036 |
| | | 182617–1515.7 | DSCT | 10.73 | | ALS 5040 |
| | $``$14<sup>o</sup>5057 | 182726–1442.1 | EC/ESD | 9.99 | | |
References to MK spectral types in Table 1: (1) Houk (1978), (2) Hill et al. (1974), (3) Walborn (1971), (4) Garrison et al. (1977), (5) FitzGerald (1987), (6) Whiteoak (1963), (7) Spectral type on the Harvard system, Nesterov et al. (1995), (8) Houk (1982), (9) Houk and Smith-Moore (1988).
3. The Analysis
The ASAS-3 data for a given star consist of the aperture photometry made through five different apertures. Since both magnitudes and their errors are reported in each aperture, we chose for analysis the data made in the aperture which had the smallest mean error. The photometry for a star in the catalogue comprises typically several independent subsets of measurements, as the star was usually observed in several different fields. These subsets may differ in the mean magnitude, as pointed out by the authors of the catalogue. In addition, a quality grade (from A to D) was assigned to each measurement. The procedure of analyzing the data was the following. First, the individual subsets were extracted. We used only the measurements that were flagged A or B. Because of the magnitude offsets between subsets, the mean magnitude was subtracted from each subset and then the subsets were combined. These combined data were subject of preliminary periodogram analysis.
After subtracting all significant signals from the combined data, the residuals were checked and the following operations were performed: (i) The outliers were rejected from the original data using 3-$`\sigma `$ clipping in residuals. (ii) The mean offset for each subset was calculated from the residuals and, along with the mean magnitude derived earlier, subtracted from the original data. (iii) The long-term trend was removed from the data. This was done by calculating average residuals in 200-day intervals that were used to derive a smoothed curve with a cubic spline fit. This smoothed curve was removed from the original data.
The combined, cleaned and detrended data were used in the final analysis. The analysis included: (i) calculating Fourier periodogram in the range from 0 to 30 d<sup>-1</sup>, (ii) fitting a sinuoid with the frequency of the highest peak, as well as all frequencies found earlier, (iii) improving frequencies, amplitudes and phases by means of the non-linear least-squares method. The residuals from the fit were used in the next iteration as the input file. The extraction of frequencies was performed until the signal-to-noise ratio (S/N) in the periodogram became smaller than 4. However, some frequencies with S/N barely exceeding 4 were not included in the final solution. By the final solution we mean the results of fitting the data with a series of sinusoidal terms in the form of
$$\underset{i=1}{\overset{n}{}}A_i\mathrm{sin}[2\pi F_i(tT_0)+\varphi _i],$$
(1)
where
$$F_i=\underset{j=1}{\overset{k}{}}m_jf_j,$$
(2)
is a linear combination of independent frequencies, $`f_j`$, extracted from the periodograms. This form was used to allow the combination frequencies and harmonics found in the periodograms to be fitted. In the above equations, $`A_i`$ denote semi-amplitude, $`t`$ is the time reckoned from the initial epoch $`T_0`$ = HJD 2450000.0, $`\varphi _i`$ is the phase, and $`m_j`$ is an integer. The parameters of the fits are given in Table 2 for all 18 $`\beta `$ Cephei stars listed in Table 1. However, instead of listing phases $`\varphi _i`$, we provide the times of maximum light, $`T_{\mathrm{max}}^iT_0`$, closest to the mean epoch of all observations. It Table 2, $`\sigma _{\mathrm{res}}`$ denotes standard deviation of the residuals, DT is the detection threshold corresponding to S/N = 4. The other columns are self-explanatory. The r.m.s. errors of the last digits are given in parentheses.
The Fourier periodograms showing the consecutive steps of prewhitening are shown in Figs. 1–4 and commented in the next section. As can be judged from these figures, the daily aliases in the periodograms of the ASAS-3 data are quite strong. Consequently, there is some ambiguity in the derived frequencies, especially for the low-amplitude terms.
4. Notes on Individual Stars
4.1. Known $`\beta `$ Cephei-Type Stars
IL Vel = HD 80383 = CPD $``$52<sup>o</sup>2185, B2 III. The pulsations of IL Vel were discovered by Haug (1977, 1979). The most extensive photometry was obtained by Heynderickx and Haug (1994) and Handler et al. (2003). The latter authors showed convincingly that the pulsational spectrum of IL Vel is dominated by two large-amplitude modes closely spaced in frequency. Handler et al. (2003) found a third mode, with a much lower amplitude. We detected the two large-amplitude modes in the ASAS-3 data (Fig. 1). Our frequencies (see Table 2) agree with those of Handler et al. (2003), but the amplitudes are $``$10% smaller. The $`V`$ amplitude of the third mode found by Handler et al. (2003) amounted to 6.5 $`\pm `$ 0.7 mmag. We do not detect this mode despite the fact that the detection threshold for this star in the ASAS-3 data is equal to 4.8 mmag.
NSV 24078 = HD 164340 = CPD $``$40<sup>o</sup>8357 = HIP 88352, B0 III. The star was suspected to be variable from the Hipparcos data. The data from this satellite were analyzed by Molenda-Żakowicz (private communication), who found two periodicities. Consequently, the star was included into the catalogue of Stankov and Handler (2005). We detected the same two frequencies in the ASAS-3 data (Fig. 1). There is an indication of a third frequency ($`f_3`$ 3.9278 d<sup>-1</sup> or one of the daily aliases), but the S/N for this frequency equals to about 4.5, that is, it only slightly exceeds the adopted detection level (S/N = 4). We therefore did not include this frequency in the final solution.
V1449 Aql = HD 180642 = ALS 10235 = HIP 94793, B1.5 II-III. This star was found to be variable in the Hipparcos data (Waelkens et al. 1998, Aerts 2000). A single frequency of 5.4871 d<sup>-1</sup> and $`V`$ amplitude of 39 mmag was found by Aerts (2000). This frequency with almost the same amplitude is confirmed in the ASAS-3 data (Table 2, Fig. 1). No other periodicities with an amplitude exceeding 6 mmag were found.
SY Equ = HD 203664 = BD +9<sup>o</sup>4973, B0.5 IIIn. This is another star found by Hipparcos and reobserved by Aerts (2000). She found a single mode with frequency of 6.0289 d<sup>-1</sup> in the Hipparcos data. We confirm this frequency from the ASAS-3 observations (Fig. 1, Table 2).
4.2. New $`\beta `$ Cephei-Type Stars
CPD $``$62<sup>o</sup>2707 = ALS 2653. The MK spectral type of this star (B2 III) was given by Garrison et al. (1977). The $`UBV`$ photometry of Schild et al. (1983) and the $`UBV\beta `$ photometry of Klare and Neckel (1977, hereafter KN77) indicate that it is indeed an early B-type star. The pulsation spectrum consists of a large-amplitude single mode with frequency $`f_1`$ = 7.0589 d<sup>-1</sup> whose harmonic, 2$`f_1`$, is also detected in the ASAS-3 data (Fig. 2).
HD 133823 = CPD $``$65<sup>o</sup>2993. Two MK spectral types are reported for this star in the literature. Houk and Cowley (1975) classify HD 133823 as a B3 II star, while Garrison et al. (1977) give B2 IV. We find a single frequency of $`f_1`$ = 5.6804 d<sup>-1</sup> in the ASAS-3 data. A significant signal at 0.4989 d<sup>-1</sup> appears in the periodogram after removing $`f_1`$ (Fig. 2), but it is likely to be spurious due to its proximity to 0.5 d<sup>-1</sup>.
CPD $``$50<sup>o</sup>9210 = ALS 3547. The MK spectral type given by FitzGerald (1987) is B2 II. The available $`UBV`$ photometry (FitzGerald 1987, Reed and Vance 1996) are also typical for an early B-type star. We detect two close frequencies in the ASAS-3 data (Fig. 2, Table 2).
T a b l e 2
Parameters of the sine-curve fits to the $`V`$ magnitudes of the $`\beta `$ Cephei stars in the ASAS-3 database
| Star | $`f`$ | $`N_{\mathrm{obs}}`$ | $`f_i`$ | $`A_i`$ | $`T_{\mathrm{max}}^iT_0`$ | $`\sigma _{\mathrm{res}}`$ / DT |
| --- | --- | --- | --- | --- | --- | --- |
| | | | \[d<sup>-1</sup>\] | \[mmag\] | \[d\] | \[mmag\] |
| IL Vel | $`f_1`$ | 367 | 5.459781(08) | 37.9(09) | 2741.3991(07) | 12.8 / 4.8 |
| | $`f_2`$ | | 5.363255(09) | 34.7(10) | 2741.4018(08) | |
| NSV 24078 | $`f_1`$ | 270 | 6.377727(14) | 27.5(10) | 2673.2062(09) | 11.7 / 5.0 |
| | $`f_2`$ | | 6.538740(18) | 20.9(10) | 2673.2280(12) | |
| V 1449 Aql | $`f_1`$ | 182 | 5.486928(14) | 36.8(12) | 2854.3232(09) | 11.3 / 5.9 |
| SY Equ | $`f_1`$ | 106 | 6.028753(24) | 29.7(12) | 2893.1065(10) | 8.5 / 5.8 |
| CPD $``$62<sup>o</sup>2707 | $`f_1`$ | 380 | 7.058920(05) | 55.1(10) | 2695.1646(04) | 13.9 / 5.1 |
| | 2$`f_1`$ | | 14.117840 | 9.0(10) | 2695.1675(13) | |
| HD 133823 | $`f_1`$ | 287 | 5.680437(10) | 52.9(15) | 2692.0050(08) | 18.2 / 7.5 |
| CPD $``$50<sup>o</sup>9210 | $`f_1`$ | 280 | 4.866859(13) | 39.5(15) | 2687.1942(13) | 17.3 / 7.4 |
| | $`f_2`$ | | 4.879365(27) | 19.4(14) | 2687.1390(25) | |
| HD 328862 | $`f_1`$ | 262 | 4.948815(07) | 82.7(17) | 2701.9660(07) | 19.2 / 8.4 |
| | $`f_2`$ | | 4.924633(34) | 15.1(16) | 2701.9864(38) | |
| | $`f_3`$ | | 5.398964(49) | 11.6(18) | 2701.9490(42) | |
| | $`f_1+f_2`$ | | 9.873448 | 10.5(18) | 2701.9762(26) | |
| CPD $``$46<sup>o</sup>8213 | $`f_1`$ | 232 | 4.460172(13) | 64.4(23) | 2723.1942(13) | 25.3 / 11.8 |
| HD 328906 | $`f_1`$ | 256 | 5.630744(23) | 41.8(28) | 2679.1349(19) | 31.8 / 14.1 |
| HD 152077 | $`f_1`$ | 405 | 4.911496(10) | 51.0(13) | 2584.6736(08) | 17.7 / 6.4 |
| | $`f_2`$ | | 4.851653(17) | 25.0(12) | 2584.5983(17) | |
| | $`f_3`$ | | 4.886405(24) | 18.5(13) | 2584.6192(22) | |
| | $`f_1+f_2`$ | | 9.763149 | 9.1(13) | 2584.7367(22) | |
| | $`f_4`$ | | 3.985732(66) | 7.7(13) | 2584.6873(65) | |
| HD 152477 | $`f_1`$ | 313 | 3.773723(10) | 35.4(10) | 2702.3268(12) | 12.5 / 5.0 |
| HD 155336 | $`f_1`$ | 620 | 5.531762(08) | 46.5(08) | 3007.3924(05) | 13.7 / 5.6 |
| | $`f_2`$ | | 5.400986(32) | 10.6(08) | 3007.4848(23) | |
| | $`f_3`$ | | 3.980814(38) | 8.5(08) | 3007.3636(37) | |
| HD 165582 | $`f_1`$ | 222 | 4.747411(23) | 38.9(18) | 2874.1057(15) | 18.0 / 9.0 |
| | $`f_2`$ | | 7.390277(40) | 22.5(18) | 2874.0508(17) | |
| | $`f_3`$ | | 4.725015(50) | 16.3(18) | 2874.1073(37) | |
| | $`f_4`$ | | 3.381100(67) | 15.0(17) | 2874.2119(59) | |
| | $`f_1+f_3`$ | | 9.472426 | 10.1(17) | 2874.1081(31) | |
| HD 167743 | $`f_1`$ | 300 | 4.823737(10) | 41.6(12) | 2662.1794(09) | 14.2 / 5.8 |
| | $`f_2`$ | | 5.096927(17) | 26.4(12) | 2662.1918(14) | |
| | $`f_3`$ | | 4.975822(30) | 13.8(13) | 2662.1817(26) | |
| ALS 5036 | $`f_1`$ | 520 | 4.917343(08) | 56.5(10) | 2722.9005(06) | 15.2 / 4.9 |
| | $`f_2`$ | | 4.919087(33) | 13.3(10) | 2722.8908(24) | |
| ALS 5040 | $`f_1`$ | 365 | 4.973861(14) | 50.0(17) | 2695.9661(11) | 22.2 / 8.5 |
| | $`f_2`$ | | 5.071912(35) | 20.2(17) | 2695.9670(26) | |
| | $`f_3`$ | | 4.523109(69) | 10.6(17) | 2696.0228(56) | |
| BD $``$14<sup>o</sup>5057 | $`f_1`$ | 323 | 4.163681(09) | 43.9(10) | 2674.0466(09) | 12.7 / 5.0 |
HD 328862 = CPD $``$47<sup>o</sup>7861 = ALS 3721 = NGC 6200 #4. The MK spectral type reported by Whiteoak (1963) is B0.5 III, while FitzGerald et al. (1977) classified the star as B1 III:. We detect three frequencies in the ASAS-3 data (Table 2, Fig. 2) and the combination term, $`f_1+f_2`$. (Table 2). The amplitude of the main mode is the largest among all stars reported in this paper and is comparable to that of BW Vul. The star is a member of the loose open cluster NGC 6200 in Ara (FitzGerald et al. 1977).
CPD $``$46<sup>o</sup>8213 = Hogg 22 #67. No MK spectral type is available for this star. However, from the $`UBV`$ photometry of Forbes and Short (1996) we may conclude that it is a reddened early B-type star and a likely member of a very young open cluster Hogg 22 (Hogg 1965). In the sky, the cluster is located 6 off NGC 6204, another open cluster, but is twice as distant, much younger and more reddened than NGC 6204 (Whiteoak 1963, Moffat and Vogt 1973, Forbes and Short 1996). The age of Hogg 22 was estimated by Forbes and Short (1996) to be 5 $`\pm `$ 2 Myr. In the ASAS-3 data of CPD $``$46<sup>o</sup>8213 we detect a single mode with a frequency of about 4.4602 d<sup>-1</sup> (Fig. 2).
HD 328906 = CD $``$44<sup>o</sup>11167. Neither MK spectral type not the $`UBV`$ photometry is available for HD 328906. However, it has the Harvard spectral type of B2 (Nesterov et al. 1995). This allows us to classify the star as a new $`\beta `$ Cephei variable. We detect a mode with frequency $`f_1`$ 5.6307 d<sup>-1</sup> in the ASAS-3 data. There is an indication of the presence of a second mode, at frequency 5.2587 d<sup>-1</sup> (Fig. 3), as it stands slightly above the adopted detection threshold. The star is located about 12 north of the young open cluster NGC 6216. The age of the cluster is 35 $`\pm `$ 15 Myr (Piatti et al. 2000), still young enough to contain a $`\beta `$ Cephei star. However, the membership of HD 328906 has to be verified.
HD 152077 = CPD $``$43<sup>o</sup>7731 = ALS 3793. According to Houk (1978), the MK spectral type is B2 Iab/Ib, but Garrison et al. (1977) give B1 II. However, the Strömgren $`\beta `$ index measured by KN77 and Knude (1992), amounting to 2.593 and 2.610, respectively, is rather too large for a supergiant. The $`UBV`$ photometry is available from KN77, Dachs et al. (1982) and Schild et al. (1983). The pulsational spectrum of HD 152077 is quite rich; we detect four independent modes including a triplet near 4.9 d<sup>-1</sup> (Fig. 3). The triplet is, however, not equidistant in frequency as one would expect for a rotationally split mode. The difference $`f_1f_3`$ = 0.02509 $`\pm `$ 0.00003 d<sup>-1</sup>, while $`f_3f_2`$ = 0.03475 $`\pm `$ 0.00003 d<sup>-1</sup>. The fourth detected frequency, $`f_4`$ = 3.9857 d<sup>-1</sup>, is far from the triplet, but it is quite likely that it is the 1 d<sup>-1</sup> alias at 4.9884 d<sup>-1</sup> that is the correct frequency. We also detect a combination frequency, $`f_1+f_2`$.
HD 152477 = CPD $``$47<sup>o</sup>7958. The star was classified as B1 Ib by Houk (1978) and B1 II by Garrison et al. (1977). The $`UBV`$ photometry was obtained by KN77 and Schild et al. (1983). KN77 measured also its $`\beta `$ index to be 2.608. We find a single mode with frequency 3.7737 d<sup>-1</sup> (Fig. 3).
HD 155336 = CPD $``$32<sup>o</sup>4389 = ALS 3961. Roslund (1966) and Houk (1982) give B3 III and B1/B2 Ib, respectively. The $`UBV`$ photometry was first reported by Roslund (1964) and $`UBV\beta `$ photometry, by KN77. We find three modes in the ASAS-3 data of HD 155336. As for HD 152077, we suspect that the 1 d<sup>-1</sup> alias of $`f_3`$, at 4.9808 d<sup>-1</sup>, may be the correct value. The ASAS-3 data for this star are distributed less evenly than for the other ones, so that the alias structure is also different. This is because the field with this star was observed more frequently during six nights around JD 2453200.
HD 165582 = CPD $``$34<sup>o</sup>7600 = ALS 4668. It is certainly an early B-type star as confirmed by three MK spectral classifications available in the literature: B1 II (Garrison et al. 1977), B1 Ib (Houk 1982), and B0.5 III (Clayton et al. 2000). The star’s $`UBV\beta `$ photometry (KN77, Dachs et al. 1982, Schild et al. 1983) is consistent with the MK classification. The frequency spectrum, shown in Fig. 3, is quite complicated. We detect four modes spread over a large range in frequency. Owing to the severe aliasing problem, it is possible that for some modes we did not extract the correct frequency. This applies especially to $`f_2`$. We tried different alias frequencies for $`f_2`$, but the solutions other than that shown in Table 2 always comprised more than five modes. The solution we provide seems to be the best in the sense that it includes the frequencies of the highest peaks in the periodograms and consists of the smallest number of frequencies. However, the ambiguity remains, especially because $`f_2f_4`$ 4 d<sup>-1</sup>. The $`f_3`$ term suffers from the same ambiguity despite the fact that we detect its combination with $`f_1`$, namely $`f_1+f_3`$ (Fig. 3).
HD 167743 = BD $``$15<sup>o</sup>4909 = ALS 9453. Houk and Smith-Moore (1988) classified this star as B2 Ib. No $`UBV`$ photometry is available, however. The pulsational spectrum (Fig. 4) consists of a triplet. The components, like those for HD 152077, are not equidistant in frequency: $`f_3f_1`$ 0.152 d<sup>-1</sup>, $`f_2f_3`$ 0.121 d<sup>-1</sup>.
ALS 5036. No MK spectral type is available, but the $`UBV`$ photometry reported by Reed (1993) clearly indicates that we are dealing with an early B-type star. The frequency spectrum consists of two very close frequencies (Fig. 4), the beat period is of the order of 570 days. This is very long, but a similar case is already known among $`\beta `$ Cephei stars: the $`f_3`$ mode in 16 Lac (Jerzykiewicz and Pigulski 1996).
ALS 5040. No MK spectral type is available. However, Reed and Vance (1996) provide the $`UBV`$ photometry, allowing us to conclude that it is a $`\beta `$ Cephei star. The power spectrum (Fig. 4) reveals three independent modes.
BD $``$14<sup>o</sup>5057 = ALS 9636. No MK spectral type is available, but the $`UBV`$ photometry of Lahulla and Hilton (1992) indicates an early B-type. We detect a single frequency (Fig. 4).
As a general comment to the $`\beta `$ Cephei stars that have no MK spectral types, we present Fig. 5, showing the location of the $`\beta `$ Cephei stars described in this paper in the two-color diagram. Only two stars, HD 328906 and HD 167743, out of 18 listed in Table 1, are not shown in the figure, because they lack $`UBV`$ photometry. However, both have spectral types that allow us to classify them unambiguously as $`\beta `$ Cephei variables. We see from Fig. 5 that all 16 stars, despite large range in reddening, fit well between the reddening lines for a B0 and B2.5 V star. This confirms that they are early B-type stars and supports their classification as $`\beta `$ Cephei variables.
4.3. Reclassified Stars
As we indicated in the Introduction, out of 37 stars classified as BCEP in the ASAS-3 catalogue, only 12 were verified as $`\beta `$ Cephei stars. The remaining 25 stars were excluded because: (i) they have A or F MK spectral type indicating either $`\delta `$ Scuti or W UMa type (10 stars), (ii) the range of variability is too large ($`\mathrm{\Delta }V`$ = 0.20–0.65 mag) and/or the shape of the light curve indicated other type of variable (13 stars). Of the large-amplitude variables classified as BCEP, nine are HADS, two are RR Lyrae stars of RRc type and two are W UMa eclipsing binaries. It is interesting to note that these stars are typically much fainter than $`\beta `$ Cephei stars from Table 1, their $`V`$ magnitudes range between 11.4 and 13.3.
There remain only two stars without MK types or $`UBV`$ photometry, ASAS 202543+0948.0 and ASAS 213518+1047.6, that have relatively small amplitudes and the $`V`$ magnitudes of about 10.8. For these two stars a chance that they are $`\beta `$ Cephei stars remains. We detect a single sinusoidal variation in ASAS 202543+0948.0 and two frequencies in ASAS 213518+1047.6. However, their dominant periods are equal to about 0.1 d, so that the $`\delta `$ Scuti classification seems to be more likely for them. Their Galactic latitudes are equal to $``$15.8<sup>o</sup> and $``$29.2<sup>o</sup>, respectively, also indicating that at least the second one is not a $`\beta `$ Cephei star. There are many stars similar to these two in the ASAS-3 catalogue, but they are usually classified as $`\delta `$ Scuti stars. Since neither the MK spectral type nor the $`UBV`$ photometry is available for them, we do not include the two, and the other stars with similar properties, in our list of $`\beta `$ Cephei stars.
5. Discussion
5.1. Location in the Galaxy
As we pointed out in the Introduction, the homogeinity of the ASAS-3 catalogue allows some general considerations. First, we would like to comment on the distribution of the $`\beta `$ Cephei stars in the sky. Their location in the Galactic coordinates is shown in Fig. 6. As expected, they concentrate around the Galactic plane, with only four stars located at relatively high Galactic latitudes. Three are bright, and therefore nearby, while the fourth, HN Aqr = PHL 346 (Waelkens and Rufener 1988, Hambly et al. 1996, Dufton et al. 1998, Lynn et al. 2002), is the only known $`\beta `$ Cephei star located far from the Galactic plane.
Two other features in the distribution of the Galactic $`\beta `$ Cephei stars can be noted. The first is the large asymmetry in the distribution: about 80% of them are located in the southern sky. This is partly due to the observational selection effects, but the dependence of the driving mechanism on metallicity, combined with decreasing metallicity at larger galactocentric distances, may also play an important role (see Pigulski et al. 2002, Pigulski 2004).
The second feature seen in Fig. 6 is an apparent strip of bright $`\beta `$ Cephei stars between Galactic longitudes $`l`$ = 200<sup>o</sup> and 360<sup>o</sup>, inclined with respect to the Galactic plane. It was already indicated by Lesh and Aizenman (1973) and explained as a part of the Gould Belt, a nearby (less than 1 kpc) apparent disk of OB stars surrounding the Sun and inclined by $``$20<sup>o</sup> with respect to the Galactic plane (see, e.g., Stothers and Frogel 1974). A part of the Gould Belt, at 300$`{}_{}{}^{\mathrm{o}}<l<`$ 360<sup>o</sup>, is known as Sco-Cen OB association and forms an extended complex of young stars at a distance of 0.1–0.2 kpc (e.g. Sartori et al. 2003).
As the new $`\beta `$ Cephei stars are concerned, they are all located close to the Galactic plane. Taking into account the absolute magnitudes of $`\beta `$ Cephei stars, the range of their apparent magnitudes (9–11 mag, see Table 1) translates into the range of distances i.e. of 1–3 kpc. Because the Galactic longitudes of the new $`\beta `$ Cephei stars cover $`l`$ between 300<sup>o</sup> and 20<sup>o</sup>, we may conclude that they lie in the Sagittarius-Carina arm of the Galaxy. The fact that no new $`\beta `$ Cephei star was discovered in the interval 200$`{}_{}{}^{\mathrm{o}}<l<`$ 300<sup>o</sup> may also be explained. At these Galactic longitudes and the distances of 1–3 kpc we probe the areas where there is no pronounced spiral arm and therefore no young population is present.
This is even better illustrated in Fig. 7, where $`\beta `$ Cephei stars (except for HN Aqr) are shown in the projection onto the Galactic plane. The distances used in this figure were estimated using a simplified method of estimating absolute magnitude, $`M_\mathrm{V}`$. First, the ($`UB`$) and ($`BV`$) colors were dereddened. Then, we assumed that $`M_\mathrm{V}`$ depends linearly on ($`BV`$)<sub>0</sub>. For open clusters we used the distances from the literature. The clusters containing at least several $`\beta `$ Cephei stars are shown with large filled circles and labeled.
We see from Fig. 7 that, as pointed out above, the new $`\beta `$ Cephei stars populate mainly the Sagittarius-Carina arm of the Galaxy, where the three southern open clusters rich in $`\beta `$ Cephei stars, i.e., NGC 3293, NGC 4755, and NGC 6231, are located too.
5.2. The amplitudes
As can be judged from Figs. 1–4, all $`\beta `$ Cephei stars found in the ASAS-3 catalogue have large amplitudes. This can be seen from Fig. 8 which shows the $`V`$-filter semi-amplitudes of the known $`\beta `$ Cephei stars (Stankov and Handler 2005) and the 14 new members described in this paper. Among the previously known $`\beta `$ Cephei stars, BW Vul has indeed the large amplitude.Note that for BW Vul, Stankov and Handler (2005) report semi-amplitude, while for the other stars, the full amplitude. This explains why BW Vul does not stand out in their Figs. 4 and 8. The remaining stars have semi-amplitudes in $`V`$ below 40 mmag. The new stars fill the gap, because their main modes have semi-amplitudes larger than 35 mmag. This is certailny due to the selection effect. The detection threshold for all but two stars analyzed by us range between 5–10 mmag, the average value being 7 mmag. Owing to the richness of the fields, the two stars in clusters are the only in our sample that have detection thresholds exceeding 10 mmag (see Table 2). We conclude therefore with the statement which was also clear for the authors of the ASAS-3 catalogue: many $`\beta `$ Cephei stars (and other low-amplitude variables) can be found in the ASAS-3 data once the periodograms will be used as the search method, instead of the dispersion-magnitude diagram.
6. Conclusions
The ASAS-3 photometry covering the whole southern sky turned out to be very efficient in finding new bright variable stars. Analysis of the ASAS-3 catalogue and the available literature led us to the conclusion that 14 stars from this catalogue can be now safely classified as new $`\beta `$ Cephei stars. All these stars have large amplitudes, many are multiperiodic. They are therefore excellent targets for follow-up spectroscopy and photometry. As they are also bright ($`V`$ 9–11 mag), this tasks can be performed by means of relatively small telescopes.
Acknowledgements. This work was supported by the KBN grant 1 P03D 016 27. This research has made use of the SIMBAD database, operated at CDS, Strasbourg, France. We wish to thank Prof. M. Jerzykiewicz for helpful comments.
REFERENCES
Aerts, C. 2000, Astron. Astrophys., 361, 245.
Clayton, G.C., Gordon, K.D., and Wolff, M.J. 2000, Astrophys. J. Suppl., 129, 147.
Dachs, J., Kaiser, D., Nikolov, A., and Sherwood, W.A. 1982, Astron. Astrophys. Suppl., 50, 261.
Dufton, P.L., Keenan, F.P., Kilkenny, D., et al. 1998, M.N.R.A.S., 297, 565.
FitzGerald, M.P. 1987, M.N.R.A.S., 229, 227.
FitzGerald, M.P., Jackson, P.D., and Moffat, A.F.J. 1977, The Observatory, 97, 129.
Forbes, D., and Short, S. 1996, Astron. J., 111, 1609.
Garrison, R.F., Hiltner, W.A., and Schild, R.E. 1977, Astrophys. J. Suppl., 35, 111.
Hambly, N.C., Wood, K.D., Keenan, F.P., et al. 1996, Astron. Astrophys., 306, 119.
Handler, G., Shobbrook, R.R., Vuthela, F.F., Balona, L.A., Rodler, F., and Tshenye, T. 2003, M.N.R.A.S., 341, 1005.
Haug, U. 1977, The ESO Messenger, 9, 14.
Haug, U. 1979, Astron. Astrophys., 80, 119.
Heynderickx, D., and Haug, U. 1994, Astron. Astrophys. Suppl., 106, 79.
Hill, P.W., Kilkenny, D., and van Breda, I.G. 1974, M.N.R.A.S., 168, 451.
Hogg, A.R. 1965, Publ. Astron. Soc. Pacific, 77, 440.
Houk, N. 1978, Michigan Spectral Survey, Ann Arbor, Dep. Astron., Univ. Michigan, 2.
Houk, N. 1982, Michigan Spectral Survey, Ann Arbor, Dep. Astron., Univ. Michigan, 3.
Houk, N., and Cowley, A.P. 1975, Michigan Spectral Survey, Ann Arbor, Dep. Astron., Univ. Michigan, 1.
Houk, N., and Smith-Moore, M. 1988, Michigan Spectral Survey, Ann Arbor, Dep. Astron., Univ. Michigan, 4.
Jerzykiewicz, M., and Pigulski, A. 1996, M.N.R.A.S., 282, 853.
Klare, G., and Neckel, T. 1977, Astron. Astrophys. Suppl., 27, 215 (KN77).
Knude, J. 1992, Astron. Astrophys. Suppl., 92, 841.
Lahulla, J.F., and Hilton, J. 1992, Astron. Astrophys. Suppl., 94, 265.
Lesh, J.R., and Aizenman, M.L. 1973, Astron. Astrophys., 26, 1.
Lynn, B.B., Dufton, P.L., Keenan, F.P., et al. 2002, M.N.R.A.S., 336, 1287.
Moffat, A.F.J., and Vogt, N. 1973, Astron. Astrophys. Suppl., 10, 135.
Nesterov, V.V., Kuzmin, A.V., Ashimbaeva, N.T., Volchkov, A.A., Röser, S., and Bastian, U. 1995, Astron. Astrophys. Suppl., 110, 367.
Paczyński, B. 1997, Proc. of the 12th IAP Colloquium: ”Variable stars and the astrophysical returns of microlensing surveys”, eds. R. Ferlet, J.-P. Maillard, and B. Raban, p. 357.
Paczyński, B. 2000, P.A.S.P., 112, 1281.
Paczyński, B. 2001, A.S.P. Conf. Ser., 246, 45.
Piatti, A.E., Clariá, J.J., and Bica, E. 2000, Astron. Astrophys., 360, 529.
Pigulski, A. 2004, Comm. in Asteroseismology, 145, 72.
Pigulski, A., Kopacki, G., Kołaczkowski, Z., and Jerzykiewicz, M. 2002, A.S.P. Conf. Ser., 259, 146.
Pojmański, G. 1997, Acta Astron., 47, 467.
Pojmański, G. 2002, Acta Astron., 52, 397.
Pojmański, G. 2003, Acta Astron., 53, 341.
Pojmański, G., and Maciejewski, G. 2004, Acta Astron., 54, 153.
Pojmański, G., and Maciejewski, G. 2005, Acta Astron., 55, 97.
Reed, B.C. 1993, Astron. J., 106, 2291.
Reed, B.C., and Vance, S.J. 1996, Astron. J., 112, 2855.
Roslund, C. 1964, Arkiv für Astron., 3, 357.
Roslund, C. 1966, Arkiv für Astron., 4, 73.
Rodríguez, E., and Breger, M. 2002, A.S.P. Conf. Ser., 259, 328.
Sartori, M.J., Lépine, J.R.D., and Dias, W.S. 2003, Astron. Astrophys., 404, 913.
Schild, R.E., Garrison, R.F., and Hiltner, W.A. 1983, Astrophys. J. Suppl., 51, 321.
Stankov, A., and Handler, G. 2005, Astrophys. J. Suppl., 158, 193.
Stothers, R., and Frogel, J.A. 1974, Astron. J., 79, 456.
Waelkens, C., and Rufener, F. 1988, Astron. Astrophys., 201, L5.
Waelkens, C., Aerts, C., Kestens, E., Grenon, M., and Eyer, L. 1998, Astron. Astrophys., 330, 215.
Walborn, N.R. 1971, Astrophys. J. Suppl., 23, 257.
Whiteoak, J.B. 1963, M.N.R.A.S., 125, 105.
|
warning/0506/cond-mat0506755.html
|
ar5iv
|
text
|
# Spin-polarized electron transport in ferromagnet/semiconductor heterostructures: Unification of ballistic and diffusive transport
## I Introduction
Considerable attention has been devoted in the past few years to the study of spin-polarized electron transport in hybrid nanostructures composed of different types of material, such as nonmagnetic or magnetic semiconductors, normal metals, ferromagnets, and superconductors. The motivation behind these efforts derives from the desire to understand the principles of operation, to assess the performance, and to explore the field of possible applications, of solid-state devices relying on the control and manipulation of electron spin (“spintronic devices”).wol01 ; gru02 ; joh02 ; zut04 Particular emphasis in spintronics research is currently placed on the study of spin-polarized transport in heterostructures formed of a nonmagnetic semiconductor and two (metallic or semiconducting) ferromagnetic contacts.zut04 ; aws02 ; sch02 ; sch05 Structures of this kind are considered promising candidates for future technological applications. For the actual design of spintronic devices, a detailed theoretical understanding of spin-polarized transport in ferromagnet/semiconductor heterostructures is indispensable. Up to now, a number of pertinent studies have been undertaken, which mostly rely on the drift-diffusion model.
Schmidt et al.sch00 describe the spin polarization by the same diffusion equation for the chemical potential as used to treat spin-polarized transport in ferromagnet/normal-metal heterojunctions.joh87 ; son87 ; joh88 ; val93 For a metallic-ferromagnet/semiconductor heterojunction with a perfect interface (no interface resistance or spin scattering), they find that, as a consequence of the large conductance mismatch, the injected current spin polarization is very low. Filip et al.fil02 and Rashbaras00 suggest that efficient spin injection can be obtained by introducing spin-selective interface resistances, for example, in the form of tunnel barriers. This idea is pursued in a number of detailed theoretical investigations in which the interface resistances are taken into account either phenomenologically by introducing discontinuities in the chemical potentials at the interface,smi01 ; fer01 ; ras02a ; yuf02 ; yuf02a , or explicitly by treating the Schottky barrier arising from band bending in the interface depletion region.alb02 ; alb03 Yu and Flattéyuf02 ; yuf02a have derived a drift-diffusion equation for the density spin polarization, which allows, in particular, the effect of applied electric fields to be studied. A formalism taking into account the effect of the electron-electron interaction on spin-polarized transport in metals and doped (degenerate) semiconductors in the diffusive regime has been introduced by D’Amico and Vignale,ami02 and has subsequently ami04 been generalized to include the effect of applied electric fields. Spin injection under conditions where, in the semiconductor, ballistic transport prevails over drift-diffusion has been studied by Kravchenko and Rashbakra03 within a Boltzmann equation approach; they find that in the absence of spin-selective interface resistances, the Sharvin interface conductancesha65 controls the injection efficiency. Spin injection across a Schottky barrier, arising from thermionic emission as well as tunneling injection, has been treated by Shen at al.she04 within a Monte Carlo model. Phase-coherent transport in the ballistic limit has been studied, e.g., in Refs. tan00, ; kir01, ; hum01, ; hun01, ; mir01, ; mat02, ; zwi03, .
It emerges that the theory of spin-polarized electron transport in ferromagnet/semiconductor heterostructures has reached a level of considerable sophistication. Nevertheless, we believe that certain aspects of the semiconductor part of the transport problem require a more detailed, unified treatment, such as the interplay of spin relaxation and transport mechanism all the way from the diffusive to the ballistic regime, and the effects of the spatial variation of the electrostatic potential profile. In the present work, we provide a comprehensive framework for systematically dealing with these aspects.
The starting point is our unified semiclassical description of (spinless) ballistic and diffusive electron transport in parallel-plane semiconductor structures,lip03 in which the idea of a thermoballistic transport mechanism was introduced. The latter relies on the concept of a “thermoballistic current” inside the semiconducting sample. This current consists of electrons which move ballistically in the electric field arising from internal (built-in) and external electrostatic potentials, and which are thermalized at randomly distributed equilibration points (with mean distance equal to the mean free path, or momentum relaxation length) due to coupling to the background of impurity atoms and phonons. The current-voltage characteristic for nondegenerate systems as well as the zero-bias conductance for degenerate systems are expressed in terms of a reduced resistance; for arbitrary momentum relaxation length and arbitrary potential profile, the latter quantity is determined from a resistance function which is obtained as the solution of an integral equation. The thermoballistic chemical potential and current are derived from this solution as well. The chemical potential exhibits discontinuities at the boundaries of the semiconductor, which are related to the Sharvin interface conductance.
In order to develop, within the unified description, a theory of spin-polarized electron transport in (nondegenerate) semiconductors, we introduce the “thermoballistic spin-polarized current” which generalizes the thermoballistic current of Ref. lip03, so as to allow spin relaxation to take place during the ballistic motion between the equilibration points. The thermoballistic spin-polarized current is constructed in terms of a “spin transport function” that determines the spin polarization inside the semiconductor for arbitrary potential profile and arbitrary values of the momentum and spin relaxation lengths. This function satisfies an integral equation which follows from the balance equation connecting the thermoballistic spin-polarized current and density. The appearance of an integral equation in the unified description of electron transport (with or without account of the spin degree of freedom) reflects the nonlocal character of the transport across the ballistic intervals between the equilibration points. For electron transport in a homogeneous semiconductor without space charge, driven by an external electric field, the integral equation for the spin transport function can be converted, in an approximation that is adequate for demonstrating the principal effects of the transport mechanism, into a second-order differential equation that generalizes the standard spin drift-diffusion equation. The spin polarization along a ferromagnet/semiconductor heterostructure is obtained by invoking continuity of the current spin polarization at the interfaces and matching the spin-resolved chemical potentials on the ferromagnet sides of the latter, with allowance for spin-selective interface resistances.
As a prerequisite to developing a theory of spin-polarized electron transport in semiconductors within the unified transport model of Ref. lip03, , we have to modify and complete the formulation given in that reference. This will be done in the next section. In Sec. III, the spin degree of freedom is introduced into the unified description. The integral equation for the spin transport function inside a semiconducting sample is derived, and the generalized spin drift-diffusion equation is obtained. Spin-polarized transport in heterostructures formed of a nonmagnetic, homogeneous semiconductor and two ferromagnetic contacts is treated in Sec. IV. We demonstrate the procedure for the calculation of the current and density spin polarizations across a heterostructure in the zero-bias limit and of the injected spin polarizations for field-driven transport. Various examples are considered which illustrate the effects of transport mechanism and electric field and exhibit the relation of the unified description to previous descriptions by other authors. In Sec. V, the contents of the paper are summarized and our conclusions as well as an outlook towards applications and extensions of the present work are presented. In the Appendix, details of the extended formulation of the unified transport model outlined in Sec. II are worked out.
## II Unification of ballistic and diffusive transport in semiconductors
The unified description of (spinless) ballistic and diffusive electron transport developed in Ref. lip03, has yielded, for the nondegenerate case, the current-voltage characteristic for a semiconducting sample enclosed between two plane-parallel contacts. There, the discontinuity of the chemical potential has been placed at the interface at one or the other end of the sample; this gave rise to an ambiguity in the behavior of the chemical potential inside the sample. In order, nevertheless, to obtain a unique current-voltage characteristic, the reduced resistance determining the latter was subjected to a heuristic symmetrization procedure (see Sec. IV.C of Ref. lip03, ).
In the following, we extend in a systematic way the unified description by treating simultaneously the effects of the two interfaces on an equal footing. In this way, we arrive at unique chemical potentials, currents, and densities inside as well as at the ends of the semiconducting sample. This is prerequisite to the study of spin-polarized electron transport in ferromagnet/semicinductor heterostructures, which is the principal aim of the present work. As in Ref. lip03, , we work within the semiclassical approximation, ignoring all coherence effects.
### II.1 Thermoballistic transport
We consider a semiconducting sample bordering on a left contact at $`x=x_1`$ and on a right contact at $`x=x_2`$ (see Fig. 1), so that $`S=x_2x_1`$ is the sample length. The geometry of the set-up is one-dimensional, whereas the electron motion is treated in three-dimensional space. Reformulating the unified description of electron transport in a nondegenerate semiconductor,lip03 we introduce the (net) electron current density (electron current, for short) $`J(x^{},x^{\prime \prime })`$ across the ballistic interval $`[x^{},x^{\prime \prime }]`$ between two equilibration points $`x^{}`$ and $`x^{\prime \prime }`$,
$`J(x^{},x^{\prime \prime })`$ $`=`$ $`v_eN_ce^{\beta E_c^m(x^{},x^{\prime \prime })}\left[e^{\beta \mu (x^{})}e^{\beta \mu (x^{\prime \prime })}\right]`$
($`x_1x^{}<x^{\prime \prime }x_2`$), which is the difference of the ballistic current $`J^l(x^{},x^{\prime \prime })`$ injected at the left end at $`x^{}`$,
$$J^l(x^{},x^{\prime \prime })=v_eN_ce^{\beta [E_c^m(x^{},x^{\prime \prime })\mu (x^{})]},$$
(2)
and the analogous ballistic current $`J^r(x^{},x^{\prime \prime })`$ injected at the right end at $`x^{\prime \prime }`$,
$$J^r(x^{},x^{\prime \prime })=v_eN_ce^{\beta [E_c^m(x^{},x^{\prime \prime })\mu (x^{\prime \prime })]}.$$
(3)
Here, the function $`\mu (x)`$ is the chemical potential at the equilibration point $`x`$. \[In Ref. lip03, , the term quasi-Fermi energy (symbol $`E_F`$) was used for the chemical potential as defined, e.g., in Ref. ash76, .\] Furthermore, $`v_e=(2\pi m^{}\beta )^{1/2}`$ is the emission velocity, $`N_c=2(2\pi m^{}/\beta h^2)^{3/2}`$ is the effective density of states at the conduction band edge, $`m^{}`$ is the effective mass of the electrons, and $`\beta =(k_BT)^1`$. The currents (2) and (3) contain only the electrons transmitted across the sample, i.e., the electrons with sufficient energy to surmount the potential barrier determined by
$$\widehat{E}_c^m(x^{},x^{\prime \prime })=E_c^m(x^{},x^{\prime \prime })E_c^0;$$
(4)
here, $`E_c^m(x^{},x^{\prime \prime })`$ is the maximum value of the potential profile $`E_c(x)`$ in the interval $`[x^{},x^{\prime \prime }]`$, and $`E_c^0`$ is its overall minimum across the sample. The profile $`E_c(x)`$ is, in general, a self-consistent solution of a nonlinear Poisson equation involving the conduction band edge potential and the external electrostatic potential.
Expressions (2) and (3) for the currents injected at the left and right ends of a ballistic interval follow from Eqs. (20) and (21) of Ref. lip03, if classical transmission probabilities are used. Tunneling effects can be included by replacing the classical probabilities with their quantal (WKB) analogues, as done, e.g., in Ref. lip01, . In the present paper, we do not consider this possibility.
It is convenient to rewrite Eq. (LABEL:eq:10xx1) in the form
$$J(x^{},x^{\prime \prime })=e^{\beta \widehat{E}_c^m(x^{},x^{\prime \prime })}[𝒥(x^{})𝒥(x^{\prime \prime })];$$
(5)
the quantity
$$𝒥(x^{})=v_eN_ce^{\beta [E_c^0\mu (x^{})]}$$
(6)
is the current injected at the left end $`x^{}`$ of the ballistic interval into the right direction in a flat profile $`E_c(x)=E_c^0`$, and similarly for the current $`𝒥(x^{\prime \prime })`$ injected at the right end $`x^{\prime \prime }`$ into the left direction.
The ballistic density, at position $`x`$ in the interval $`[x^{},x^{\prime \prime }]`$, of transmitted electrons injected at the left end at $`x^{}`$ is given by
$`n^l(x^{},x^{\prime \prime };x)`$ $`=`$ $`{\displaystyle \frac{N_c}{2}}(2\beta /\pi m^{})^{1/2}`$
$`\times {\displaystyle _0^{\mathrm{}}}dpe^{\beta [p^2/2m^{}+E_c(x)\mu (x^{})]}`$
$`\times \mathrm{\Theta }\mathbf{(}p^2/2m^{}+E_c(x)E_c^m(x^{},x^{\prime \prime })\mathbf{)}`$
$`=`$ $`{\displaystyle \frac{N_c}{2}}C(x^{},x^{\prime \prime };x)e^{\beta [E_c^m(x^{},x^{\prime \prime })\mu (x^{})]},`$
where
$`C(x^{},x^{\prime \prime };x)`$ $`=`$ $`e^{\beta [E_c^m(x^{},x^{\prime \prime })E_c(x)]}`$
$`\times \mathrm{erfc}\mathbf{(}(\beta [E_c^m(x^{},x^{\prime \prime })E_c(x)])^{1/2}\mathbf{)};`$
the function $`\mathrm{erfc}(x)`$ is the complementary error function,abr65 and $`0<C(x^{},x^{\prime \prime };x)1`$ \[the conserved currents (2) and (3) are, of course, independent of $`x`$\]. Analogously, the ballistic electron density in the transmitted current injected at the right end at $`x^{\prime \prime }`$ is
$`n^r(x^{},x^{\prime \prime };x)`$ $`=`$ $`{\displaystyle \frac{N_c}{2}}C(x^{},x^{\prime \prime };x)e^{\beta [E_c^m(x^{},x^{\prime \prime })\mu (x^{\prime \prime })]}.`$
Dividing the current (2) by the density (LABEL:eq:16yzz1) \[or (3) by (LABEL:eq:16nico)\], we obtain for the average velocity $`v(x^{},x^{\prime \prime };x)`$ at position $`x`$ of the electrons injected at either end of, and transmitted across, the interval $`[x^{},x^{\prime \prime }]`$
$$v(x^{},x^{\prime \prime };x)=\frac{J^{l,r}(x^{},x^{\prime \prime })}{n^{l,r}(x^{},x^{\prime \prime };x)}=\frac{2v_e}{C(x^{},x^{\prime \prime };x)},$$
(10)
which depends only on the potential profile. For constant profile, one has $`C(x^{},x^{\prime \prime };x)=1`$, and the electrons move with the average velocity $`2v_e`$ from one or the other end to its opposite. This is the average velocity of the injected electrons also in the case of a position-dependent profile. However, some of these electrons are reflected back, so that the average velocity at position $`x`$ of those electrons which have sufficient energy to pass over the top of the potential profile must be higher than $`2v_e`$, namely, equal to the velocity given by Eq. (10).
In analogy to Eq. (LABEL:eq:10xx1), the sum of the densities $`n^l(x^{},x^{\prime \prime };x)`$ and $`n^r(x^{},x^{\prime \prime };x)`$ is the ballistic density $`n(x^{},x^{\prime \prime };x)`$ of transmitted electrons in the interval $`[x^{},x^{\prime \prime }]`$,
$`n(x^{},x^{\prime \prime };x)`$ $`=`$ $`{\displaystyle \frac{N_c}{2}}C(x^{},x^{\prime \prime };x)e^{\beta E_c^m(x^{},x^{\prime \prime })}`$ (11)
$`\times \left[e^{\beta \mu (x^{})}+e^{\beta \mu (x^{\prime \prime })}\right].`$
The density $`n(x^{},x^{\prime \prime };x)`$, like the current $`J(x^{},x^{\prime \prime })`$, comprises only electrons that participate in the transport.
From the ballistic current (5), the thermoballistic current $`J(x)`$ at position $`x`$ inside the semiconductor is constructed by summing up the weighted contributions of the ballistic intervals $`[x^{},x^{\prime \prime }]`$ for which $`x_1x^{}<x<x^{\prime \prime }x_2`$ \[see Eq. (23) of Ref. lip03, \],
$`J(x)`$ $`=`$ $`w_1(x_1,x_2;l)[𝒥_1𝒥_2]`$
$`+`$ $`{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}w_2(x^{},x_2;l)[𝒥(x^{})𝒥_2]`$
$`+`$ $`{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{}}{l}}w_2(x_1,x^{};l)[𝒥_1𝒥(x^{})]`$
$`+`$ $`{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{\prime \prime }}{l}}w_3(x^{},x^{\prime \prime };l)[𝒥(x^{})𝒥(x^{\prime \prime })],`$
with
$$w_n(x^{},x^{\prime \prime };l)=p_n(|x^{\prime \prime }x^{}|/l)e^{\beta \widehat{E}_c^m(x^{},x^{\prime \prime })}$$
(13)
$`(n=0,1,2,3)`$, where $`l`$ is the momentum relaxation length of the electrons, which comprises the effect of relaxation due to electron scattering by impurity atoms and phonons. The probabilities $`p_n(x/l)`$ of occurrence of the ballistic intervals depend on the dimensionality of the transport \[note that $`w_n(x^{},x^{\prime \prime };l)`$ is symmetric with respect to an interchange of $`x^{}`$ and $`x^{\prime \prime }`$\]. In Eq. (LABEL:eq:11a), the quantities $`𝒥_{1,2}=𝒥(x_{1,2})`$ are fixed by the chemical potentials $`\mu _{1,2}=\mu (x_{1,2})`$ on the contact sides of the interfaces at $`x_{1,2}`$, i.e., immediately outside of the sample,
$$𝒥_{1,2}=v_eN_ce^{\beta (E_c^0\mu _{1,2})}.$$
(14)
For later convenience, we introduce a symbolic operator $`𝕎(x^{},x^{\prime \prime };l)`$ to write expression (LABEL:eq:11a) in the condensed form
$`J(x)`$ $`=`$ $`{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{\prime \prime }}{l}}𝕎(x^{},x^{\prime \prime };l)[𝒥(x^{})𝒥(x^{\prime \prime })],`$
which, in view of Eqs. (6) and (14), may also be written as
$`J(x)`$ $`=`$ $`v_eN_ce^{\beta E_c^0}{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{\prime \prime }}{l}}𝕎(x^{},x^{\prime \prime };l)`$ (16)
$`\times \left[e^{\beta \mu (x^{})}e^{\beta \mu (x^{\prime \prime })}\right].`$
Analogously, we introduce the thermoballistic density inside the semiconductor, $`n(x)`$, as
$`n(x)`$ $`=`$ $`{\displaystyle \frac{N_c}{2}}e^{\beta E_c^0}{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{\prime \prime }}{l}}𝕎_C(x^{},x^{\prime \prime };l;x)`$ (17)
$`\times \left[e^{\beta \mu (x^{})}+e^{\beta \mu (x^{\prime \prime })}\right],`$
where
$$𝕎_C(x^{},x^{\prime \prime };l;x)=𝕎(x^{},x^{\prime \prime };l)C(x^{},x^{\prime \prime };x).$$
(18)
Again, the current $`J(x)`$ and the density $`n(x)`$ comprise only electrons that participate in the transport.
The thermoballistic current (LABEL:eq:11a) by itself is not, in general, conserved, but together with the background currentlip03 $`J^{back}(x)`$ it adds up to the conserved physical current $`J`$:
$$J(x)+J^{back}(x)=J=\mathrm{const}.$$
(19)
The background current is confined within the sample and, therefore, must vanish when integrated over the latter, which implies that the thermoballistic current $`J(x)`$ averaged over the sample yields the physical current $`J`$,
$$\frac{1}{x_2x_1}_{x_1}^{x_2}𝑑xJ(x)=J.$$
(20)
The non-conservation of the thermoballistic current can be viewed as arising from a source term $`Q(x)`$ associated with the gain of thermoballistic electron density due to the coupling to the background, as expressed by the equation
$$\frac{dJ(x)}{dx}=Q(x).$$
(21)
In the background, the source term appears as a sink term describing the loss of electron density,
$$Q^{back}(x)=Q(x).$$
(22)
Again, since the background electrons are confined to the sample, the integral of $`Q^{back}(x)`$ over the sample must vanish and, therefore, also that of the thermoballistic source term $`Q(x)`$,
$$_{x_1}^{x_2}𝑑xQ(x)=0.$$
(23)
Owing to Eq. (21), this implies
$$J(x_1^+)=J(x_2^{})\kappa J,$$
(24)
that is, the thermoballistic current entering at one end of the sample, $`x=x_1^+`$, must be the same as the one leaving at the other end, $`x=x_2^{}`$. The quantity $`\kappa `$ has been introduced for later convenience; it normalizes the thermoballistic current on the sample sides of the ferromagnet/semiconductor interfaces to the physical current $`J`$. We remark that, in analogy to the thermoballistic current, the thermoballistic density as well as other thermoballistic quantities introduced later in the development all have their background complement.
Condition (20) has been used in Ref. lip03, to obtain the fundamental integral equation for the determination of the thermoballistic current. Condition (24) is new, and provides us with an extension of the formalism of Ref. lip03, which allows us to establish a unique thermoballistic chemical potential inside the sample, as will be shown in the following.
### II.2 Thermoballistic chemical potential, current, and density
Substituting expression (LABEL:eq:11a) in condition (20), we obtain
$`{\displaystyle \frac{x_2x_1}{l}}J`$ $`=`$ $`\left[u_1(x_1,x_2;l)+{\displaystyle _{x_1}^{x_2}}{\displaystyle \frac{dx^{}}{l}}u_2(x_1,x^{};l)\right]𝒥_1`$ (25)
$``$ $`\left[u_1(x_1,x_2;l)+{\displaystyle _{x_1}^{x_2}}{\displaystyle \frac{dx^{}}{l}}u_2(x^{},x_2;l)\right]𝒥_2`$
$`+`$ $`{\displaystyle _{x_1}^{x_2}}{\displaystyle \frac{dx^{}}{l}}[\text{}u_2(x^{},x_2;l)u_2(x_1,x^{};l)`$
$`+{\displaystyle _{x_1}^{x_2}}{\displaystyle \frac{dx^{\prime \prime }}{l}}u_3(x^{},x^{\prime \prime };l)]𝒥(x^{}),`$
where
$$u_n(x^{},x^{\prime \prime };l)=\frac{x^{\prime \prime }x^{}}{l}w_n(x^{},x^{\prime \prime };l)$$
(26)
\[note that $`u_n(x^{},x^{\prime \prime };l)`$ is antisymmetric with respect to an interchange of $`x^{}`$ and $`x^{\prime \prime }`$\]. Equation (25) is a basic condition on the function $`𝒥(x)`$, and hence, via Eq. (6), on the chemical potential $`\mu (x)`$, whose determination allows all relevant transport quantities to be obtained.
For given current $`J`$, only the value of the chemical potential at one of the interfaces with the contacts can be prescribed. If, at the interface at $`x_1`$, we prescribe the value of $`\mu _1`$, i.e., that of $`𝒥_1`$ \[case (i)\], we can find the value of $`𝒥(x)`$ at the other interface at $`x_2`$ by re-expressing in Eq. (25) $`𝒥_2`$ as $`𝒥(x_2)`$ and replacing $`x_2`$ with the variable $`x`$, thereby obtaining an integral equation for the function $`𝒥(x)`$ in the range $`x_1<xx_2`$. If, now, $`𝒥(x_2)`$ is required to assume a preassigned value for which we re-introduce the symbol $`𝒥_2`$, then the current $`J`$ on the left-hand side of Eq. (25) is fixed at some value $`J_1`$. We denote the associated solution of the integral equation by $`𝒥_1(x)`$.
On the other hand, prescribing the value $`\mu _2`$ for the chemical potential at the interface at $`x_2`$ \[case (ii)\], we re-express in Eq. (25) $`𝒥_1`$ as $`𝒥(x_1)`$. Then, replacing $`x_1`$ with the variable $`x`$, we obtain an integral equation for the function $`𝒥(x)`$ in the range $`x_1x<x_2`$. With $`𝒥(x_1)`$ required to assume a preassigned value $`𝒥_1`$, the current $`J`$ is now fixed at the value $`J_2`$. The associated solution of the integral equation is denoted by $`𝒥_2(x)`$.
To determine the functions $`𝒥_{1,2}(x)`$ explicitly, we proceed in the following way. In case (i), we define the “resistance function”lip03
$$\chi _1(x)=\frac{𝒥_1𝒥_1(x)}{J_1};\chi _1(x_1)=0,$$
(27)
and obtain from Eq. (25), following the procedure outlined above,
$`{\displaystyle \frac{xx_1}{l}}`$ $``$ $`\left[u_1(x_1,x;l)+{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}u_2(x^{},x;l)\right]\chi _1(x)`$
$`+`$ $`{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}[\text{}u_2(x^{},x;l)u_2(x_1,x^{};l)`$
$`+{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{\prime \prime }}{l}}u_3(x^{},x^{\prime \prime };l)]\chi _1(x^{})=0,`$
which is a linear, Volterra-type integral equation for $`\chi _1(x)`$. Letting $`xx_1^+`$ in Eq. (LABEL:eq:11bb), we find, using the properties of $`u_n(x^{},x^{\prime \prime };l)`$,
$`\chi _1(x_1^+)`$ $`=`$ $`{\displaystyle \frac{𝒥_1𝒥_1(x_1^+)}{J_1}}=e^{\beta [E_c(x_1)E_c^0]}\chi _1(x_1).`$
With this discontinuity incorporated in it, the solution $`\chi _1(x)`$ is unique and continuous for $`x_1<xx_2`$. The solution of Eq. (LABEL:eq:11bb) can be obtained in closed form under special conditions; in the general case, one solves this equation efficiently by discretization and numerical propagation, using the initial value $`\chi _1(x_1^+)`$ given by Eq. (LABEL:eq:8).
In case (ii), we define the resistance function
$$\chi _2(x)=\frac{𝒥_2(x)𝒥_2}{J_2};\chi _2(x_2)=0,$$
(30)
which satisfies the integral equation
$`{\displaystyle \frac{x_2x}{l}}`$ $``$ $`\left[u_1(x,x_2;l)+{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{}}{l}}u_2(x,x^{};l)\right]\chi _2(x)`$
$``$ $`{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{}}{l}}[\text{}u_2(x^{},x_2;l)u_2(x,x^{};l)`$
$`+{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{\prime \prime }}{l}}u_3(x^{},x^{\prime \prime };l)]\chi _2(x^{})=0.`$
The solution $`\chi _2(x)`$ is discontinuous at $`x=x_2`$:
$`\chi _2(x_2^{})`$ $`=`$ $`{\displaystyle \frac{𝒥_2(x_2^{})𝒥_2)}{J_2}}=e^{\beta [E_c(x_2)E_c^0]}\chi _2(x_2).`$
It follows from Eqs. (LABEL:eq:11bb) and (LABEL:eq:11bbb), using the properties of $`u_n(x^{},x^{\prime \prime };l)`$, that the functions $`\chi _1(x)`$ and $`\chi _2(x)`$ are related by
$$\chi _2(x)=\chi _1^{}(x_0x),$$
(33)
where $`x_0=x_1+x_2`$; the asterisk attached to $`\chi _1`$ indicates that this function is to be calculated using the reverse of the profile $`E_c(x)`$, given by $`E_c^{}(x)=E_c(x_0x)`$. If the profile is symmetric, $`E_c^{}(x)=E_c(x)`$, the functions $`\chi _1(x)`$ and $`\chi _2(x)`$ are the reverse of one another, $`\chi _2(x)=\chi _1(x_0x)`$.
The two functions $`𝒥_{1,2}(x)`$ are not, in general, equal and yield different chemical potentials $`\mu _{1,2}(x)`$ via Eq. (6). Then, in view of Eq. (14), Eq. (LABEL:eq:8) implies $`\mu _1(x_1^+)\mu _1`$, and the chemical potential $`\mu _1(x)`$ is discontinuous at the interface at $`x=x_1`$, i.e., its value on the semiconductor side of the interface is not equal to its value at the interface itself. Analogously, $`\mu _2(x_2^{})\mu _2`$. The ambiguity thus found is a generalization of the ambiguity of the chemical potential in the ballistic limit $`l/S\mathrm{}`$,dat95 where it may either be associated with the current injected at $`x_2`$, in which case it is discontinuous at $`x=x_1`$ (Sharvin resistance at the interface at $`x=x_1`$), or with the current injected at $`x=x_1`$, in which case it is discontinuous at $`x=x_2`$ (Sharvin resistance at the interface at $`x=x_2`$). In the Appendix, details of the construction of a unique thermoballistic chemical potential $`\mu (x)`$, current $`J(x)`$, and density $`n(x)`$ in terms of the two solutions $`\chi _{1,2}(x)`$ are presented. Here, we only summarize the results.
A quantity $`\chi `$ is introduced as
$$\chi =\widehat{a}_1\chi _1(x_2)+\widehat{a}_2\chi _2(x_1),$$
(34)
where $`\widehat{a}_1+\widehat{a}_2=1`$. The coefficients $`\widehat{a}_1`$ and $`\widehat{a}_2`$ are given by
$$\widehat{a}_{1,2}=\frac{a_{1,2}}{a},$$
(35)
where
$`a_1`$ $`=`$ $`{\displaystyle _{x_1}^{x_2}}{\displaystyle \frac{dx^{}}{l}}\{w_2(x_1,x^{};l)[\chi _2(x^{})\chi _2(x_1)]`$ (36)
$`+w_2(x^{},x_2;l)\chi _2(x^{})\},`$
$`a_2`$ $`=`$ $`{\displaystyle _{x_1}^{x_2}}{\displaystyle \frac{dx^{}}{l}}\{w_2(x_1,x^{};l)\chi _1(x^{})`$ (37)
$`+w_2(x^{},x_2;l)[\chi _1(x^{})\chi _1(x_2)]\},`$
and
$$a=a_1+a_2;$$
(38)
for a symmetric potential profile $`E_c(x)`$, we have $`\widehat{a}_1=\widehat{a}_2=1/2`$.
The current $`𝒥(x)`$ is expressed as
$$𝒥(x)=\frac{1}{2}(𝒥_1+𝒥_2)J\chi _{}(x),$$
(39)
where
$`\chi _{}(x)`$ $`=`$ $`\widehat{a}_1\left[\chi _1(x){\displaystyle \frac{1}{2}}\chi _1(x_2)\right]\widehat{a}_2\left[\chi _2(x){\displaystyle \frac{1}{2}}\chi _2(x_1)\right]`$
$`(x_1xx_2)`$. The currents $`𝒥_1`$ and $`𝒥_2`$ satisfy the relation
$$𝒥_1𝒥_2=J\chi .$$
(41)
With the use of Eq. (14), the current-voltage characteristic is then obtained in the form
$$J=v_eN_ce^{\beta E_p}\frac{1}{\stackrel{~}{\chi }}\left(1e^{\beta eV}\right),$$
(42)
where
$$V=\frac{\mu _1\mu _2}{e}$$
(43)
is the voltage bias between the contacts, and $`E_p=E_c^m(x_1,x_2)\mu _1`$; the quantity $`\stackrel{~}{\chi }`$, given by
$$\stackrel{~}{\chi }=e^{\beta \widehat{E}_c^m(x_1,x_2)}\chi ,$$
(44)
is the “reduced resistance”.lip03 It replaces, in the present extended unified description, expression (58) of Ref. lip03, , which was obtained, in a heuristic way, by taking the mean value of the reduced resistances corresponding to case (i) and (ii), respectively. To determine $`\stackrel{~}{\chi }`$ in the diffusive and ballistic regimes, we evaluate the functions $`\chi _{1,2}(x)`$ by following the development given in the Appendix of Ref. lip03, . In the diffusive regime $`l/S1`$, we find $`\chi _1(x_2)=\chi _2(x_1)=\chi `$, which leads to
$$\stackrel{~}{\chi }=\frac{1}{2p_0(0)}_{x_1}^{x_2}\frac{dx}{l}e^{\beta [E_c^m(x_1,x_2)E_c(x)]},$$
(45)
where the values of $`p_0(0)`$ for one-, two-, and three-dimensional transport are given in Sec. II of Ref. lip03, . In the ballistic limit $`l/S\mathrm{}`$, we have $`\stackrel{~}{\chi }=1`$.
According to Eqs. (6), (14), (39), and (41), the thermoballistic chemical potential $`\mu (x)`$ is given by
$`e^{\beta \mu (x)}`$ $`=`$ $`\left[{\displaystyle \frac{1}{2}}{\displaystyle \frac{\chi _{}(x)}{\chi }}\right]e^{\beta \mu _1}+\left[{\displaystyle \frac{1}{2}}+{\displaystyle \frac{\chi _{}(x)}{\chi }}\right]e^{\beta \mu _2}`$ (46)
$`=`$ $`\eta _+2{\displaystyle \frac{\chi _{}(x)}{\chi }}\eta _{}`$
$`(x_1xx_2)`$, where
$$\eta _\pm =\frac{1}{2}\left(e^{\beta \mu _1}\pm e^{\beta \mu _2}\right),$$
(47)
and the thermoballistic equilibrium electron density $`\overline{n}(x)`$ by
$$\overline{n}(x)=N_ce^{\beta [E_c(x)\mu (x)]}.$$
(48)
The thermoballistic chemical potential $`\mu (x)`$ \[and hence the thermoballistic equilibrium density $`\overline{n}(x)`$\] are discontinuous at the interfaces at $`x_{1,2}`$,
$$e^{\beta [\mu (x_1^+)\mu _1]}1=\widehat{a}_1\frac{\beta e^2J}{𝒢_1},$$
(49)
$$e^{\beta [\mu (x_2^{})\mu _2]}1=\widehat{a}_2\frac{\beta e^2J}{𝒢_2},$$
(50)
as can be shown with the help of Eqs. (LABEL:eq:8) and (LABEL:eq:8a), respectively, and Eqs. (LABEL:eq:12s) and (42). Here,
$$𝒢_{1,2}=\beta e^2v_e\overline{n}_{1,2}$$
(51)
are the respective Sharvin interface conductances,sha65 ; lip03 with $`\overline{n}_{1,2}=\overline{n}(x_{1,2})`$. The discontinuities of the functions $`\mathrm{exp}[\beta \mu (x)]`$ and $`\overline{n}(x)`$ are proportional to $`J`$. In the diffusive regime $`l/S1`$, where, according to Eqs. (42) and (45), $`Jl/S`$, the discontinuities approach zero together with $`l/S`$.
The thermoballistic current $`J(x)`$ is obtained in terms of $`\chi _{}(x)`$ and $`\chi `$ by substituting expression (39) in combination with Eq. (41) in Eq. (LABEL:eq:11aba) \[or, more explicitly, in Eq. (LABEL:eq:11a)\],
$`J(x)=J\{\text{}w_1(x_1,x_2;l)\chi `$
$`+`$ $`{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}w_2(x^{},x_2;l)\left[{\displaystyle \frac{\chi }{2}}\chi _{}(x^{})\right]`$
$`+`$ $`{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{}}{l}}w_2(x_1,x^{};l)\left[{\displaystyle \frac{\chi }{2}}+\chi _{}(x^{})\right]`$
$`+`$ $`{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{\prime \prime }}{l}}w_3(x^{},x^{\prime \prime };l)[\chi _{}(x^{\prime \prime })\chi _{}(x^{})]\}.`$
The thermoballistic density $`n(x)`$ is found in a similar fashion from Eq. (17),
$`n(x)={\displaystyle \frac{J}{2v_e}}\{\text{}\chi \mathrm{coth}(\beta eV/2)𝔚(x_1,x_2;l;x)`$
$``$ $`{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}𝔴_2(x^{},x_2;l;x)\left[{\displaystyle \frac{\chi }{2}}+\chi _{}(x^{})\right]`$
$`+`$ $`{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{}}{l}}𝔴_2(x_1,x^{};l;x)\left[{\displaystyle \frac{\chi }{2}}\chi _{}(x^{})\right]`$
$``$ $`{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{\prime \prime }}{l}}𝔴_3(x^{},x^{\prime \prime };l;x)[\chi _{}(x^{\prime \prime })+\chi _{}(x^{})]\},`$
where
$`𝔚(x_1,x_2;l;x)=𝔴_1(x_1,x_2;l;x)`$ (54)
$`+`$ $`{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}𝔴_2(x^{},x_2;l;x)`$
$`+`$ $`{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{}}{l}}𝔴_2(x_1,x^{};l;x)`$
$`+`$ $`{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{\prime \prime }}{l}}𝔴_3(x^{},x^{\prime \prime };l;x)`$
and
$`𝔴_n(x^{},x^{\prime \prime };l;x)=w_n(x^{},x^{\prime \prime };l)C(x^{},x^{\prime \prime };x).`$ (55)
In the zero-bias limit $`V0`$, expression (LABEL:eq:11ahab) reduces to the form $`n(x)=\overline{n}_1𝔚(x_1,x_2;l;x)`$, from which the physical meaning of the function $`𝔚(x_1,x_2;l;x)`$ becomes apparent.
In the diffusive regime $`l/S1`$, we have
$$e^{\beta \mu (x)}=\eta _+\frac{I_c(x_1,x)I_c(x,x_2)}{I_c(x_1,x_2)}\eta _{},$$
(56)
where
$$I_c(x^{},x^{\prime \prime })=_x^{}^{x^{\prime \prime }}𝑑ze^{\beta E_c(z)},$$
(57)
and
$`J(x)=J`$ $`=`$ $`{\displaystyle \frac{\nu }{e}}\overline{n}(x){\displaystyle \frac{d\mu (x)}{dx}}`$
$`=`$ $`{\displaystyle \frac{\nu }{e}}\left[\overline{n}(x){\displaystyle \frac{dE_c(x)}{dx}}+{\displaystyle \frac{1}{\beta }}{\displaystyle \frac{d\overline{n}(x)}{dx}}\right],`$
where $`\nu =2ev_e\beta l`$ is the electron mobility, and $`l=p_0(0)l`$ is the effective momentum relaxation length.lip03 Integrating Eq. (LABEL:eq:4wasg), we retrieve the current-voltage characteristic (42) with $`\stackrel{~}{\chi }`$ given by Eq. (45). The thermoballistic density $`n(x)`$ becomes equal to the equilibrium density $`\overline{n}(x)`$ given by Eq. (48).
In the ballistic limit $`l/S\mathrm{}`$, we have
$`e^{\beta \mu (x)}=\eta _+\{2e^{\beta E_c^m(x_1,x_2)}`$
$`\times [\widehat{a}_1e^{\beta E_c^m(x_1,x)}\widehat{a}_2e^{\beta E_c^m(x,x_2)}](\widehat{a}_1\widehat{a}_2)\}\eta _{},`$
where $`\widehat{a}_1`$ and $`\widehat{a}_2`$ are to be calculated from Eqs. (35)–(38) with
$$\chi _1(x)=e^{\beta \widehat{E}_c^m(x_1,x)}$$
(60)
and
$$\chi _2(x)=e^{\beta \widehat{E}_c^m(x,x_2)}.$$
(61)
Further,
$$J(x)=J=2v_eN_ce^{\beta E_c^m(x_1,x_2)}\eta _{}$$
(62)
and
$$n(x)=N_cC(x_1,x_2;x)e^{\beta E_c^m(x_1,x_2)}\eta _+.$$
(63)
As our model demands, expressions (62) and (63) are identical to the original ballistic expressions (5) and (11) with $`x^{}=x_1`$ and $`x^{\prime \prime }=x_2`$.
### II.3 Field-driven transport in a homogeneous semiconductor
To illustrate the formalism developed so far, we now consider electron transport in a homogeneous semiconductor without space charge. The electrons are assumed to be driven by a (constant) electric field $``$ directed antiparallel to the $`x`$-axis, i.e., they are moving in a linearly falling potential of the form
$$E_c(x)=E_c(x_1)e||(xx_1),$$
(64)
in which case
$`C(x^{},x^{\prime \prime };x)`$ $``$ $`C\mathbf{(}ϵ(xx^{})\mathbf{)}`$
$`=`$ $`e^{ϵ(xx^{})}\mathrm{erfc}\mathbf{(}[ϵ(xx^{})]^{1/2}\mathbf{)},`$
where $`ϵ=\beta e||`$. The latter quantity is related to the voltage bias $`V`$ via $`ϵS=\beta eV`$, so that
$$\overline{n}_1=\overline{n}_2=\overline{n},$$
(66)
where use has been made of Eqs. (43) and (48).
In order to calculate the thermoballistic chemical potential $`\mu (x)`$ from Eq. (46) \[or, equivalently, the thermoballistic equilibrium electron density $`\overline{n}(x)`$ from Eq. (48)\], the thermoballistic current $`J(x)`$ from Eq. (LABEL:eq:11aha), and the thermoballistic density $`n(x)`$ from Eq. (LABEL:eq:11ahab), we first have to solve the integral equations (LABEL:eq:11bb) and (LABEL:eq:11bbb) numerically for the functions $`\chi _1(x)`$ and $`\chi _2(x)`$, respectively. For convenience, we use in the integral equations the probabilities $`p_n(x/l)`$ in their one-dimensional form, $`p_n(x/l)=e^{x/l}`$ \[see Eq. (10) of Ref. lip03, \], so that from Eq. (13) and (26)
$`u_n(x^{},x^{\prime \prime };l)`$ $`=`$ $`{\displaystyle \frac{x^{\prime \prime }x^{}}{l}}e^{|x^{\prime \prime }x^{}|/l}e^{ϵ[x_2\mathrm{min}(x^{},x^{\prime \prime })]}`$
(this simplification has only minor effect, see Ref. lip03, ). Owing to the scaling properties of the function $`C(x^{},x^{\prime \prime };x)`$ given by Eq. (LABEL:eq:29csu) and of the function $`u_n(x^{},x^{\prime \prime };l)`$ given by Eq. (LABEL:eq:14adac), the results of the calculations can be expressed essentially in terms of three dimensionless quantities, for example, $`x/S`$, $`ϵS`$, and $`l/S`$. In Figs. 2, 3, and 4, we show the dependence of $`\mu (x)`$, $`\overline{n}(x)`$, $`J(x)`$, and $`n(x)`$ on $`x/S`$ (assuming $`x_1=0`$) for $`ϵS=1`$ and various values of $`l/S`$ .
From Fig. 2 (lower panel), it is seen that in the diffusive limit $`l/S0`$ the chemical potential $`\mu (x)`$ decreases linearly with $`x/S`$ and is continuous at the interfaces; this behavior persists in the case of arbitrary $`ϵS`$, where $`\beta [\mu (x)\mu _1]=ϵx`$ for $`0x/S1`$. In combination with the identical decrease of the potential profile $`E_c(x)`$, this implies a position-independent equilibrium density $`\overline{n}(x)`$ (see upper panel of Fig. 2). When $`l/S`$ rises towards the ballistic limit $`l/S\mathrm{}`$, discontinuities of $`\mu (x)`$ develop at the interfaces, which increase in magnitude, and the slope of $`\mu (x)`$ becomes smaller. This results in a rise of the equilibrium density $`\overline{n}(x)`$ across the sample. As a function of $`ϵS`$, the discontinuities of $`\mu (x)`$ become smaller in magnitude if $`ϵS\mathrm{}`$, and larger if $`ϵS0`$, such that in the latter case $`\mu (x)`$ becomes independent of $`x`$ for $`0<x/S<1`$.
The ratio $`J(x)/J`$ shown in Fig. 3 is close to unity across the whole sample. As the ballistic limit $`l/S\mathrm{}`$ is approached, $`J(x)/J`$ becomes more and more symmetric about $`x/S=1/2`$, and $`J(x)/J1`$ in the full range $`0x/S1`$. A somewhat peculiar behavior of $`J(x)`$ is observed in the diffusive regime $`l/S1`$. Here, $`J(x)/J`$ is very close to unity inside the sample, except for the immediate vicinity of the ends at $`x/S=0`$ and $`x/S=1`$, where some structure develops, and $`J(x)/J`$ converges towards a value smaller than unity when $`x/S0`$ or $`x/S1`$. It is important to keep in mind the latter feature when, within the unified treatment of spin-polarized transport, the injected spin polarization at ferromagnet/semiconductor interfaces is defined (see Sec. IV.C). When $`ϵS`$ is varied, the qualitative behavior of $`J(x)/J`$ persists for all values of $`l/S`$ considered.
In Fig. 4, we show the ratio $`n(x)/n_0`$, where $`n_0`$ is the constant value that $`n(x)`$ \[as well as $`\overline{n}(x)`$\] assume in the diffusive limit $`l/S0`$. For increasing $`l/S`$, i.e., as the ballistic contribution to the transport mechanism increases, the ratio $`n(x)/n_0`$ decreases as a whole. This effect can be interpreted as reflecting the fact that ballistically, the density decreases rapidly as the velocity rises along the sample \[see Eq. (10)\]; this is impeded at the equilibration points, which lie very dense when $`l/S1`$ \[slow decrease of $`n(x)`$\] and are widely spread when $`l/S1`$ \[rapid decrease of $`n(x)`$\]. For $`l/S>1`$, the behavior of $`n(x)`$ is largely determined by the function $`C(ϵx)`$ \[in the ballistic limit $`l/S\mathrm{}`$, we have $`n(x)=(\overline{n}/2)(1+e^{ϵS})C(ϵx)`$\]. Again, the qualitative behavior of $`n(x)`$ does not change when $`ϵS`$ is varied.
The zero-bias limit $`ϵ0`$ can be treated analytically. The solution of Eq. (LABEL:eq:11bb) is found to be
$$\chi _1(x)=1+\frac{xx_1}{2l}.$$
(68)
Since $`\chi _2(x)=\chi _1(x_0x)`$ and $`\widehat{a}_1=\widehat{a}_2=1/2`$ to zeroth order in $`ϵS=\beta eV`$, we obtain
$$\chi =1+\frac{S}{2l},\chi _{}(x)=\frac{xx_0/2}{2l}.$$
(69)
From Eq. (LABEL:eq:11aha), the thermoballistic current $`J(x)`$ is found to be constant, $`J(x)=J`$. For the current $`J`$, we have from Eqs. (42) and (48), to first order in $`\beta eV`$,
$$J=\frac{2l}{2l+S}v_e\overline{n}\beta eV,$$
(70)
where $`\overline{n}`$ is the common value of the thermoballistic equilibrium density at either end of the sample \[see Eq. (66)\]. Then, the conductance $`G=eJ/V`$ becomes
$$G=\frac{2l}{2l+S}𝒢,$$
(71)
where $`𝒢`$ is the Sharvin interface conductance \[see Eq. (51)\]. Relation (71) generalizes the Ohm conductance $`G=(2l/S)𝒢=\sigma /S`$ (valid in the diffusive regime, $`l/S1`$), where $`\sigma =2\beta e^2v_e\overline{n}l`$ is the conductivity.lip03 The thermoballistic equilibrium density $`\overline{n}(x)`$ and the thermoballistic density $`n(x)`$ are obtained from expressions (48) and (LABEL:eq:11ahab), respectively, as
$$\overline{n}(x)=n(x)=\overline{n},$$
(72)
i.e., they are, in the present case of zero bias, both independent of position and equal to the equilibrium density at the ends of the sample.
## III Spin-polarized transport within the unified description
Having established, in the preceding section, a unified description of spinless electron transport in semiconductors in terms of a unique thermoballistic chemical potential, we will now extend this scheme by including the spin degree of freedom. We allow spin relaxation to take place during the motion of the electrons across the ballistic intervals. Spin relaxation is generally governed by the equation of balance connecting the spin-polarized current with the spin-polarized density. In the unified description, it is the thermoballistic current and density which enter into this equation. The solution of the balance equation is found in terms of a spin transport function that is related to the spin-resolved thermoballistic chemical potentials.
### III.1 Balance equation and transport mechanism
In a stationary situation, the total electron current $`J=J_{}(x)+J_{}(x)`$ composed of its spin-resolved parts $`J_{}(x)`$ and $`J_{}(x)`$ is conserved, whereas the spin-polarized current $`J_{}(x)=J_{}(x)J_{}(x)`$, or rather its off-equilibrium part $`\widehat{J}_{}(x)=J_{}(x)\stackrel{~}{J}_{}(x)`$, where $`\stackrel{~}{J}_{}(x)`$ is the relaxed part of the spin-polarized current, is connected with the off-equilibrium spin-polarized density $`\widehat{n}_{}(x)`$ through the balance equation
$$\frac{d\widehat{J}_{}(x)}{dx}+\frac{\widehat{n}_{}(x)}{\tau _s}=0.$$
(73)
Here, $`\widehat{n}_{}(x)`$ is defined in analogy to $`\widehat{J}_{}(x)`$, and $`\tau _s`$ is the spin relaxation time. For a complete description of spin-polarized transport, the balance equation (73) is to be supplemented with a relation between the current and the density, which reflects the specific transport mechanism.
In the ballistic limit, the electron currents $`J_{}(x)`$ are proportional to the densities $`n_{}(x)`$ of the electrons participating in the transport,
$$J_{}(x)=v(x)n_{}(x),$$
(74)
where $`v(x)`$ is the average velocity of the electrons at position $`x`$ \[we disregard spin splitting of the conduction band edge potential $`E_c(x)`$, so that $`v(x)`$ is independent of spin\]. This relation holds also for the off-equilibrium spin-polarized current and density,
$$\widehat{J}_{}(x)=v(x)\widehat{n}_{}(x).$$
(75)
Use of this equation in Eq. (73) yields
$$\frac{d\widehat{J}_{}(x)}{dx}+C(x)\frac{\widehat{J}_{}(x)}{l_s}=0,$$
(76)
where $`l_s=2v_e\tau _s`$ is the (ballistic) spin relaxation length, which comprises the overall effect of the various underlying microscopic spin relaxation mechanisms,ell54 ; yaf63 ; dya71 ; yuk05 and where Eq. (10) has been used (omitting the positions $`x^{},x^{\prime \prime }`$ of the end points) to express $`v(x)`$ in terms of $`C(x)`$, i.e., of the potential $`E_c(x)`$ in which the electrons move.
In the diffusive regime, the off-equilibrium spin-polarized current and density are connected by the relation
$$\widehat{J}_{}(x)=\frac{\nu }{e}\left[\widehat{n}_{}(x)\frac{dE_c(x)}{dx}+\frac{1}{\beta }\frac{d\widehat{n}_{}(x)}{dx}\right]$$
(77)
\[see Eq. (LABEL:eq:4wasg)\]. We then find from Eq. (73)
$`{\displaystyle \frac{d^2\widehat{n}_{}(x)}{dx^2}}`$ $`+`$ $`\beta {\displaystyle \frac{dE_c(x)}{dx}}{\displaystyle \frac{d\widehat{n}_{}(x)}{dx}}`$
$`+`$ $`\beta {\displaystyle \frac{d^2E_c(x)}{dx^2}}\widehat{n}_{}(x){\displaystyle \frac{1}{L_s^2}}\widehat{n}_{}(x)=0,`$
where
$$L_s=\sqrt{ll_s}$$
(79)
is the spin diffusion length.
In the unified description, the total current and density inside the semiconducting sample are taken to be the thermoballistic current $`J(x)`$ and density $`n(x)`$, between which no direct relation generally exists. Instead, Eqs. (16) and (17) express these two quantities separately in terms of the chemical potential $`\mu (x)`$. The connection between the off-equilibrium thermoballistic spin-polarized current $`\widehat{J}_{}(x)`$ and density $`\widehat{n}_{}(x)`$ can be established along similar lines, as will be described in the following.
### III.2 Thermoballistic spin-polarized current and density
In order to include spin relaxation in the unified description, we begin by introducing the thermoballistic equilibrium densities $`\overline{n}_{}(x^{})`$ for spin-up and spin-down electrons at an equilibration point $`x^{}`$. It is convenient to express $`\overline{n}_{}(x^{})`$ in terms of the spin-independent thermoballistic equilibrium density $`\overline{n}(x^{})`$ and a “spin fraction” $`\alpha _{}(x^{})`$ via
$$\overline{n}_{}(x^{})=\overline{n}(x^{})\alpha _{}(x^{}),$$
(80)
with $`\alpha _{}(x^{})+\alpha _{}(x^{})=1`$. In analogy to Eq. (48), we define spin-resolved thermoballistic chemical potentials $`\mu _{}(x^{})`$ via
$$\overline{n}_{}(x^{})=N_ce^{\beta [E_c(x^{})\mu _{}(x^{})]},$$
(81)
which implies
$$e^{\beta \mu _{}(x^{})}=e^{\beta \mu (x^{})}\alpha _{}(x^{}).$$
(82)
The spin fraction $`\alpha _{}(x^{})`$ also enters into the definition of the spin-resolved ballistic current $`J_{}^l(x^{},x^{\prime \prime })`$ injected at the left end at $`x^{}`$ of the interval $`[x^{},x^{\prime \prime }]`$,
$`J_{}^l(x^{},x^{\prime \prime })`$ $`=`$ $`v_eN_ce^{\beta [E_c^m(x^{},x^{\prime \prime })\mu _{}(x^{})]}`$ (83)
$`=`$ $`J^l(x^{},x^{\prime \prime })\alpha _{}(x^{}).`$
We emphasize that expression (83) for $`J_{}^l(x^{},x^{\prime \prime })`$ holds only at the left end, since this current is not conserved owing to spin relaxation, and becomes position-dependent inside the interval. There, we write it in the form
$$J_{}^l(x^{},x^{\prime \prime };x)=J^l(x^{},x^{\prime \prime })\alpha _{}^l(x^{},x^{\prime \prime };x).$$
(84)
The function $`\alpha _{}^l(x^{},x^{\prime \prime };x)`$ is the spin fraction at position $`x`$ of spin-up (spin-down) electrons injected into the ballistic interval $`[x^{},x^{\prime \prime }]`$ at its left end at $`x^{}`$, with $`\alpha _{}^l(x^{},x^{\prime \prime };x)+\alpha _{}^l(x^{},x^{\prime \prime };x)=1`$. Here, the dependence on the position $`x^{\prime \prime }`$ (at the end of the ballistic interval $`[x^{},x^{\prime \prime }]`$ opposite to that at position $`x^{}`$ where the electrons are injected) is due to the effect of the potential barrier embodied in the function $`E_c^m(x^{},x^{\prime \prime })`$ in expression (2).
When $`x`$ coincides with the injection point $`x^{}`$, the current (84) becomes identical to the current (83), so that
$$\alpha _{}^l(x^{},x^{\prime \prime };x^{})=\alpha _{}(x^{}).$$
(85)
We now introduce the “spin fraction excess” $`\alpha _{}(x^{})=\alpha _{}(x^{})\alpha _{}(x^{})`$ and the off-equilibrium spin fraction excess $`\widehat{\alpha }_{}(x^{})=\alpha _{}(x^{})\stackrel{~}{\alpha }_{}`$, where $`\stackrel{~}{\alpha }_{}=\stackrel{~}{\alpha }_{}\stackrel{~}{\alpha }_{}`$, and $`\stackrel{~}{\alpha }_{}`$ are the relaxed parts of the spin fractions ($`\stackrel{~}{\alpha }_{}=0`$ for nonmagnetic semiconductors). With the off-equilibrium spin fraction excess $`\widehat{\alpha }_{}^l(x^{},x^{\prime \prime };x)`$ defined in an analogous way, we write the off-equilibrium ballistic spin-polarized current $`\widehat{J}_{}^l(x^{},x^{\prime \prime };x)`$ as
$$\widehat{J}_{}^l(x^{},x^{\prime \prime };x)=J^l(x^{},x^{\prime \prime })\widehat{\alpha }_{}^l(x^{},x^{\prime \prime };x).$$
(86)
The spin relaxation of the electrons injected at the left equilibration point $`x^{}`$ into the ballistic interval $`[x^{},x^{\prime \prime }]`$ is governed by Eq. (76), which, owing to Eq. (84), becomes a differential equation for $`\widehat{\alpha }_{}^l(x^{},x^{\prime \prime };x)`$,
$$\frac{d\widehat{\alpha }_{}^l(x^{},x^{\prime \prime };x)}{dx}+C(x^{},x^{\prime \prime };x)\frac{\widehat{\alpha }_{}^l(x^{},x^{\prime \prime };x)}{l_s}=0.$$
(87)
The solution of Eq. (87) is
$$\widehat{\alpha }_{}^l(x^{},x^{\prime \prime };x)=\widehat{\alpha }_{}(x^{})e^{(x^{},x^{\prime \prime };x^{},x)/l_s},$$
(88)
where
$$(x^{},x^{\prime \prime };z_1,z_2)=_{z_<}^{z_>}𝑑zC(x^{},x^{\prime \prime };z),$$
(89)
with $`z_<=\mathrm{min}(z_1,z_2)`$ and $`z_>=\mathrm{max}(z_1,z_2)`$. We then have for the off-equilibrium ballistic spin-polarized current at position $`x`$ of electrons injected at $`x^{}`$
$`\widehat{J}_{}^l(x^{},x^{\prime \prime };x)`$ $`=`$ $`J^l(x^{},x^{\prime \prime })\widehat{\alpha }_{}(x^{})e^{(x^{},x^{\prime \prime };x^{},x)/l_s};`$
analogously, we find
$`\widehat{J}_{}^r(x^{},x^{\prime \prime };x)`$ $`=`$ $`J^r(x^{},x^{\prime \prime })\widehat{\alpha }_{}(x^{\prime \prime })e^{(x^{},x^{\prime \prime };x,x^{\prime \prime })/l_s}`$
for the off-equilibrium ballistic spin-polarized current injected at $`x^{\prime \prime }`$.
Separating out the relaxed part, we now write the (net) ballistic spin-polarized current, in analogy to Eq. (LABEL:eq:10xx1), in the form
$$J_{}(x^{},x^{\prime \prime };x)=\widehat{J}_{}(x^{},x^{\prime \prime };x)+J(x^{},x^{\prime \prime })\stackrel{~}{\alpha }_{},$$
(92)
with the off-equilibrium ballistic spin-polarized current
$`\widehat{J}_{}(x^{},x^{\prime \prime };x)=v_eN_ce^{\beta \widehat{E}_c^m(x^{},x^{\prime \prime })}`$
$`\times `$ $`\left[A(x^{})e^{(x^{},x^{\prime \prime };x^{},x)/l_s}A(x^{\prime \prime })e^{(x^{},x^{\prime \prime };x,x^{\prime \prime })/l_s}\right];`$
here, we have introduced the “spin transport function”
$$A(x^{})=e^{\beta [E_c^0\mu (x^{})]}\widehat{\alpha }_{}(x^{})$$
(94)
at the equilibration point $`x^{}`$ ($`x_1x^{}x_2`$). Using the relation
$$\mu _{}(x^{})=\frac{1}{\beta }\mathrm{ln}\left(\frac{1+\alpha _{}(x^{})}{1\alpha _{}(x^{})}\right),$$
(95)
between the splitting $`\mu _{}(x^{})=\mu _{}(x^{})\mu _{}(x^{})`$ of the spin-up and spin-down chemical potentials and the spin fraction excess $`\alpha _{}(x^{})`$, which follows from Eq. (82), we have
$`A(x)`$ $`=`$ $`e^{\beta [E_c^0\mu (x)]}`$
$`\times `$ $`\left[\mathrm{tanh}\left({\displaystyle \frac{\beta \mu _{}(x)}{2}}\right)\mathrm{tanh}\left({\displaystyle \frac{\beta \overline{\mu }_{}(x)}{2}}\right)\right].`$
For the ballistic spin-polarized density, we have, in analogy to Eq. (92),
$$n_{}(x^{},x^{\prime \prime };x)=\widehat{n}_{}(x^{},x^{\prime \prime };x)+n(x^{},x^{\prime \prime };x)\stackrel{~}{\alpha }_{},$$
(97)
where
$`\widehat{n}_{}(x^{},x^{\prime \prime };x)={\displaystyle \frac{N_c}{2}}C(x^{},x^{\prime \prime };x)e^{\beta \widehat{E}_c^m(x^{},x^{\prime \prime })}`$
$`\times `$ $`\left[A(x^{})e^{(x^{},x^{\prime \prime };x^{},x)/l_s}+A(x^{\prime \prime })e^{(x^{},x^{\prime \prime };x,x^{\prime \prime })/l_s}\right]`$
is the off-equilibrium ballistic spin-polarized density \[see Eqs. (11) and (LABEL:eq:22a)\].
For the thermoballistic spin-polarized current $`J_{}(x)`$ passing through the point $`x`$, we find
$$J_{}(x)=\widehat{J}_{}(x)+J(x)\stackrel{~}{\alpha }_{},$$
(99)
where the off-equilibrium thermoballistic spin-polarized current $`\widehat{J}_{}(x)`$ is obtained from the off-equilibrium ballistic spin-polarized current (LABEL:eq:22a) by summing up the weighted contributions of the ballistic intervals,
$`\widehat{J}_{}(x)=v_eN_c{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{\prime \prime }}{l}}𝕎(x^{},x^{\prime \prime };l)`$
$`\times `$ $`\left[A(x^{})e^{(x^{},x^{\prime \prime };x^{},x)/l_s}A(x^{\prime \prime })e^{(x^{},x^{\prime \prime };x,x^{\prime \prime })/l_s}\right]`$
$`(x_1<x<x_2)`$. Similarly, the thermoballistic spin-polarized density $`n_{}(x)`$ at the point $`x`$ is
$$n_{}(x)=\widehat{n}_{}(x)+n(x)\stackrel{~}{\alpha }_{},$$
(101)
where the off-equilibrium thermoballistic spin-polarized density $`\widehat{n}_{}(x)`$ is obtained from (LABEL:eq:20a) as
$`\widehat{n}_{}(x)={\displaystyle \frac{N_c}{2}}{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{\prime \prime }}{l}}𝕎_C(x^{},x^{\prime \prime };l;x)`$
$`\times `$ $`\left[A(x^{})e^{(x^{},x^{\prime \prime };x^{},x)/l_s}+A(x^{\prime \prime })e^{(x^{},x^{\prime \prime };x,x^{\prime \prime })/l_s}\right]`$
$`(x_1<x<x_2)`$.
In the diffusive regime, $`l/l_s1`$, $`l/S1`$, the integrals over $`x^{}`$ and $`x^{\prime \prime }`$ in Eqs. (LABEL:eq:23) and (LABEL:eq:24) can be evaluated explicitly, yielding
$$\widehat{J}_{}(x)=2v_eN_cle^{\beta [E_c(x)E_c^0]}\frac{dA(x)}{dx}$$
(103)
and
$$\widehat{n}_{}(x)=N_ce^{\beta [E_c(x)E_c^0]}A(x).$$
(104)
Eliminating the function $`A(x)`$ from these two equations, we obtain the standard drift-diffusion relation (77).
In the current $`\widehat{J}_{}(x)`$ and density $`\widehat{n}_{}(x)`$, the spin relaxation in each ballistic interval is described in terms of the values of the spin transport function $`A(x)`$ at the end points $`x=x^{}`$ and $`x=x^{\prime \prime }`$. Since $`\widehat{J}_{}(x)`$ and $`\widehat{n}_{}(x)`$ are linearly connected with $`A(x)`$, they are linearly connected with each other. Thus, it appears that the function $`A(x)`$ \[and not the chemical-potential splitting $`\mu _{}(x)`$\] is the key quantity for treating spin transport within the unified description. It remains to find an equation for the determination of this function.
### III.3 Integral equation for the spin transport function
The required equation is provided by the basic balance equation, Eq. (73), which we now read in terms of the off-equilibrium thermoballistic spin-polarized current (LABEL:eq:23) and density (LABEL:eq:24) of the unified transport description. Since the derivative with respect to $`x`$ of the terms in the brackets of expression (LABEL:eq:23) for $`\widehat{J}_{}(x)`$ is compensated by the term $`\widehat{n}_{}(x)/\tau _s`$ (this reflects the fact that spin relaxation in the ballistic intervals has already been taken into account), only the derivative on the limits of integration in expression (LABEL:eq:23) remains, and we have
$`{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{}}{l}}𝕎(x,x^{};l)\left[A(x)A(x^{})e^{(x,x^{})/l_s}\right]`$
$``$ $`{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}𝕎(x^{},x;l)\left[A(x^{})e^{(x^{},x)/l_s}A(x)\right]=0,`$
where
$$(x^{},x^{\prime \prime })=(x^{},x^{\prime \prime };x^{},x^{\prime \prime }).$$
(106)
With the action of the symbolic operator $`𝕎(x^{},x^{\prime \prime };l)`$ explained by comparison of Eqs. (LABEL:eq:11a) and (LABEL:eq:11aba), Eq. (LABEL:eq:26) reads explicitly
$`𝒲_2(x_1,x;l,l_s)A_1`$ $`+`$ $`𝒲_2(x,x_2;l,l_s)A_2`$
$``$ $`W(x_1,x_2;x;l)A(x)`$
$`+`$ $`{\displaystyle _{x_1}^{x_2}}{\displaystyle \frac{dx^{}}{l}}𝒲_3(x^{},x;l,l_s)A(x^{})=0,`$
where
$$𝒲_n(x^{},x^{\prime \prime };l,l_s)=w_n(x^{},x^{\prime \prime };l)e^{(x^{},x^{\prime \prime })/l_s},$$
(108)
$`W(x_1,x_2;x;l)`$ $`=`$ $`w_2(x_1,x;l)+w_2(x,x_2;l)`$ (109)
$`+`$ $`{\displaystyle _{x_1}^{x_2}}{\displaystyle \frac{dx^{}}{l}}w_3(x^{},x;l),`$
and $`A_{1,2}=A(x_{1,2})`$. Equation (LABEL:eq:26a) is a linear, Fredholm-type integral equation for the spin transport function $`A(x)`$. Its solution for $`x_1<x<x_2`$ determines the spin-polarized electron transport inside the semiconducting sample, and is obtained in terms of the values $`A_1`$ and $`A_2`$ at the interfaces at the ends of the sample. The latter are determined by the off-equilibrium spin fraction excesses $`\widehat{\alpha }_{}(x_{1,2})`$ and the chemical potentials $`\mu _{1,2}=\mu (x_{1,2})`$ in the contacts \[see Eq. (94)\]. The function $`A(x)`$ is not, in general, continuous at the interfaces, $`A(x_1^+)A_1`$, $`A(x_2^{})A_2`$ (“Sharvin effect”), as will be demonstrated in Sec. III.D by way of a particular example. The discontinuities of $`A(x)`$ arise from the joint effect of the discontinuities of the spin-independent thermoballistic chemical potential $`\mu (x)`$ and those of the spin fraction excess $`\alpha _{}(x)`$ \[or, equivalently, of the spin-resolved thermoballistic chemical potentials $`\mu _{}(x)`$\].
Substituting $`A(x)`$ in Eqs. (LABEL:eq:23) and (LABEL:eq:24), we obtain the off-equilibrium thermoballistic spin-polarized current $`\widehat{J}_{}(x)`$ and density $`\widehat{n}_{}(x)`$, respectively; the thermoballistic spin-polarized current $`J_{}(x)`$ and density $`n_{}(x)`$ then follow from Eqs. (99) and (101), respectively. Dividing by the corresponding total thermoballistic current and density, Eqs. (16) and (17), respectively, we get the current spin polarization
$$P_J(x)=\frac{J_{}(x)}{J(x)}=\frac{\widehat{J}_{}(x)}{J(x)}+\stackrel{~}{\alpha }_{}$$
(110)
and the density spin polarization
$$P_n(x)=\frac{n_{}(x)}{n(x)}=\frac{\widehat{n}_{}(x)}{n(x)}+\stackrel{~}{\alpha }_{}$$
(111)
inside the sample. These polarizations are written in terms of the thermoballistic current and density; however, we take their magnitudes to be also those of the physical polarizations, for the following reason. The underlying assumption of our approach is that the equilibration process, i.e., the coupling between the thermoballistic and background currents, is independent of spin (this is clearly true for the D’yakonov-Perel’ spin relaxation mechanism,dya71 but remains to be examined for the other mechanisms). Therefore, the relative spin content is the same in these two currents, and thus equal to that of their sum, viz., the physical current. Hence, we may take the polarizations $`P_J(x)`$ and $`P_n(x)`$ of Eqs. (110) and (111), respectively, for the physical polarizations.
The integral equation (LABEL:eq:26a) constitutes the central result of the present work. It allows the calculation of the spin polarization in semiconductors for any value of the momentum and spin relaxation lengths as well as for arbitrary band edge potential profile. The fact that we are led, in the unified description of spin-polarized transport, to an integral equation is connected with the introduction of the momentum relaxation length $`l`$ as an independent parameter of arbitrary magnitude, which gives rise to nonlocal ballistic effects. The basic parameters controlling the transport in the unified description are the equilibrium densities $`\overline{n}_{1,2}`$, the momentum relaxation length $`l`$, and the spin relaxation length $`l_s`$, whereas in the standard drift-diffusion model one uses the conductivity $`\sigma `$ and the spin diffusion length $`L_s`$.
### III.4 Differential equation for the spin transport function
In order to interpret our unified description of spin-polarized transport and relate it to previous, less general descriptions, we consider in the following the case of field-driven transport in a homogeneous semiconductor without space charge. As in Sec. II.C, we take the probabilities $`p_n(x/l)`$ in their one-dimensional form. Then, Eq. (LABEL:eq:26a) can be converted into an integrodifferential equation for the spin transport function $`A(x)`$. In an approximation which is adequate for the present purposes, the latter equation reduces to a second-order differential equation.
#### III.4.1 General form and diffusive regime
For a potential of the form (64), the integral equation (LABEL:eq:26a) reduces, with the help of Eqs. (LABEL:eq:29csu), (LABEL:eq:14adac), (89), and (106), to
$`f_1(xx_1)A_1+f_2(x_2x)A_2`$ (112)
$``$ $`f(xx_1)A(x)+{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}f_1(xx^{})A(x^{})`$
$`+`$ $`{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{}}{l}}f_2(x^{}x)A(x^{})=0,`$
where
$$f_1(x)=e^{[ϵ+1/l+𝔠(ϵx)/l_s]x},$$
(113)
$$f_2(x)=e^{[1/l+𝔠(ϵx)/l_s]x},$$
(114)
$$f(x)=\frac{1}{1+ϵl}\left\{2+ϵl\left[1+e^{(ϵ+1/l)x}\right]\right\},$$
(115)
and
$$𝔠(\zeta )=\frac{1}{\zeta }_0^\zeta 𝑑\zeta ^{}C(\zeta ^{})$$
(116)
with $`0<𝔠(\zeta )1`$, $`𝔠(\zeta )1`$ for $`\zeta 0`$, and $`𝔠(\zeta )2(\pi \zeta )^{1/2}`$ for $`\zeta \mathrm{}`$.
By supplementing the inhomogeneous integral equation (112) with the equations obtained by forming its first and second derivative with respect to $`x`$, and eliminating from this set of equations the quantities $`A_1`$ and $`A_2`$, we can convert Eq. (112) into a homogeneous integrodifferential equation for the spin transport function $`A(x)`$. \[This procedure could also be applied to Eq. (LABEL:eq:26a), but does not seem to be helpful in the general case\]. Now, the latter equation can be simplified by replacing the function $`𝔠(\zeta )`$ with a position-independent average value $`\overline{𝔠}`$, so that the coefficient functions $`f_1(x)`$ and $`f_2(x)`$ in Eq. (112) reduce to pure exponentials. With this approximation, the integrodifferential equation for $`A(x)`$ becomes a second-order differential equation of the form
$$b_0(x)\frac{d^2A(x)}{dx^2}+b_1(x)\frac{dA(x)}{dx}+b_2(x)A(x)=0,$$
(117)
where
$$b_0(x)=2+ϵl[1+b(x)],$$
(118)
$$b_1(x)=ϵ(2+ϵl)[1b(x)],$$
(119)
$`b_2(x)`$ $`=`$ $`{\displaystyle \frac{1}{l\overline{l}\stackrel{~}{l}^2}}\{2[\stackrel{~}{l}^2l\overline{l}+ϵl\stackrel{~}{l}(\stackrel{~}{l}\overline{l})]`$ (120)
$`+`$ $`\text{}ϵ\overline{l}(\stackrel{~}{l}l)(l+\stackrel{~}{l}+ϵl\stackrel{~}{l})[1+b(x)]\},`$
with
$$b(x)=e^{(ϵ+1/l)(xx_1)}$$
(121)
and
$$\frac{1}{\overline{l}}=\frac{1}{l}+\frac{1}{l_s},\frac{1}{\stackrel{~}{l}}=\frac{1}{l}+\frac{1}{\stackrel{~}{l}_s},\stackrel{~}{l}_s=\frac{l_s}{\overline{𝔠}}.$$
(122)
Since, owing to the presence of the factor $`e^{x/l}`$ in the functions $`f_1(x)`$ and $`f_2(x)`$, only the values of $`𝔠(ϵx)`$ within the range $`0xl`$ contribute appreciably, we choose $`\overline{𝔠}`$ as the average of $`𝔠(ϵx)`$ over an $`x`$-interval of length equal to the momentum relaxation length $`l`$,
$$\overline{𝔠}=\frac{1}{l}_0^l𝑑x𝔠(ϵx)=\frac{1}{ϵl}_0^{ϵl}𝑑\zeta \mathrm{ln}(ϵl/\zeta )C(\zeta ).$$
(123)
In the right-hand integral of this equation, the range of small $`ϵx`$, where $`C(ϵx)1`$, is emphasized because of the weight factor $`\mathrm{ln}(ϵl/\zeta )`$. For large $`ϵl`$ (in the ballistic regime and/or for strong fields), the variation of $`𝔠(ϵx)`$ with $`x`$ becomes essential, and a more detailed study of the validity of the approximation leading to Eq. (117) will be necessary. For the present purpose of solely demonstrating the principal effects of the transport mechanism, we consider this approximation, in conjunction with the choice (123) for $`\overline{𝔠}`$, to be sufficiently accurate.
In the diffusive regime characterized by the conditions $`l/l_s1`$, $`l/S1`$, and $`ϵl1`$, we have $`\overline{𝔠}=1`$ and $`\stackrel{~}{l}=\overline{l}`$, and Eq. (117) reduces to
$$\frac{d^2A(x)}{dx^2}+ϵ\frac{dA(x)}{dx}\frac{1}{L_s^2}A(x)=0.$$
(124)
In view of Eq. (104), Eq. (124) can be rewritten in terms of $`\widehat{n}_{}(x)`$ and then agrees with Eq. (LABEL:eq:5aa), and with Eq. (2.8) of Yu and Flattéyuf02a if the intrinsic spin diffusion length $`L`$ of that reference is identified with the spin diffusion length $`L_s=\sqrt{ll_s}`$. It thus turns out that Eq. (117) generalizes the usual spin drift-diffusion equation to the case of arbitrary values of the ratio $`l/l_s`$.
#### III.4.2 Zero-bias limit
In the zero-bias limit $`ϵ0`$, the integral equation (112) reduces to
$`e^{(xx_1)/\overline{l}}A_1+e^{(x_2x)/\overline{l}}A_2`$ (125)
$``$ $`2A(x)+{\displaystyle _{x_1}^{x_2}}{\displaystyle \frac{dx^{}}{l}}e^{|xx^{}|/\overline{l}}A(x^{})=0,`$
from which one derives the differential equation
$$\frac{d^2A(x)}{dx^2}\frac{1}{L^2}A(x)=0;$$
(126)
here,
$$L=\sqrt{\overline{l}l_s}=\frac{L_s}{\sqrt{1+l/l_s}}$$
(127)
is the generalization of the spin diffusion length (79), which includes ballistic effects via the renormalization factor $`1/\sqrt{1+l/l_s}`$. The length $`L`$ becomes equal to the spin diffusion length proper, $`L=L_s`$, in the diffusive regime where $`\overline{l}=l`$, and to the spin relaxation length, $`L=l_s`$, in the ballistic limit $`l/l_s\mathrm{}`$ where $`\overline{l}=l_s`$.
Equation (126) has the general solution, for $`x_1<x<x_2`$,
$$A(x)=C_1e^{(xx_1)/L}+C_2e^{(x_2x)/L}.$$
(128)
With this expression substituted for $`A(x)`$ in Eq. (125), the set of two equations resulting from writing down this equation for $`x=x_1`$ and $`x=x_2`$, respectively, can be solved for the coefficients $`C_{1,2}`$,
$$C_1=\frac{1}{D}\left[(1+\gamma )e^{S/L}A_1(1\gamma )A_2\right],$$
(129)
$$C_2=\frac{1}{D}\left[(1\gamma )A_1(1+\gamma )e^{S/L}A_2\right],$$
(130)
where
$$D=(1+\gamma )^2e^{S/L}(1\gamma )^2e^{S/L},$$
(131)
with
$$\gamma =\frac{L}{l_s}=\frac{\overline{l}}{L}=\sqrt{\frac{l}{l+l_s}}1.$$
(132)
It follows from Eqs. (128)–(132) that the function $`A(x)`$ is discontinuous at $`x=x_{1,2}`$,
$$\mathrm{\Delta }A_1A(x_1^+)A_1=\frac{1}{2}(gA_1hA_2),$$
(133)
$$\mathrm{\Delta }A_2A_2A(x_2^{})=\frac{1}{2}(hA_1gA_2),$$
(134)
where
$$g=\frac{2\gamma }{D}\left[(1+\gamma )e^{S/L}+(1\gamma )e^{S/L}\right]1$$
(135)
and
$$h=\frac{4\gamma }{D}\frac{2}{1+\gamma }e^{S/L}.$$
(136)
In the diffusive regime, one has $`L=L_s=l_s\sqrt{l/l_s}`$ and $`\gamma =0`$, and therefore
$$A(x)=A_1e^{(xx_1)/L_s}+A_2e^{(x_2x)/L_s}.$$
(137)
In the ballistic limit, one has $`L=l_s`$ and $`\gamma =1`$, so that
$$A(x)=\frac{1}{2}\left[A_1e^{(xx_1)/l_s}+A_2e^{(x_2x)/l_s}\right].$$
(138)
The discontinuity of $`A(x)`$, e.g., at $`x=x_1`$, is $`\mathrm{\Delta }A_1=A_2\mathrm{exp}(S/L_s)`$ in the diffusive regime, and $`\mathrm{\Delta }A_1=\frac{1}{2}[A_1+A_2\mathrm{exp}(S/l_s)]`$ in the ballistic limit.
## IV Spin-polarized transport in ferromagnet/semiconductor heterostructures
We now turn to the unified description of spin-polarized electron transport in heterostructures formed of a semiconductor and two ferromagnetic contacts (cf. Fig. 1). We treat the ferromagnets as fully degenerate Fermi systems. The semiconductor is taken to be nonmagnetic \[i.e., $`\stackrel{~}{\alpha }_{}=0`$, and hence $`\widehat{\alpha }_{}(x^{})=\alpha _{}(x^{})`$\] and homogeneous without space charge. We disregard spin-flip scattering at the interfaces, but spin-selective interface resistances are included in our description by introducing discontinuities into the spin-resolved chemical potentials, in the same way as in previous descriptionssmi01 ; fer01 ; ras02a ; yuf02a within the drift-diffusion model. Of course, for realistic applications, it is necessary to treat the effect of interface barriers explicitly,han03 using potential profiles $`E_c(x)`$ which, generally, must be calculated self-consistently from a nonlinear Poisson equation. In some cases, however, it may be sufficient to perform non-self-consistent calculations using appropriately modeled band edge profiles.alb02 ; alb03 ; she04 In any case, this would require the spin transport function $`A(x)`$ to be determined by numerically solving the integral equation (LABEL:eq:26a). This task will be deferred to future work.
In order to obtain the position dependence of the spin polarization across the heterostructure, the current spin polarization and the chemical potential in the semiconductor are to be connected with the corresponding quantities in the left and right ferromagnets. The current spin polarization $`P_J(x)`$ \[see Eq. (110)\], as expressed by the ratio of the thermoballistic currents $`\widehat{J}_{}(x)`$ and $`J(x)`$, is equal to the physical current spin polarization \[see the discussion following Eq. (111)\]. It is, therefore, continuous across the whole heterostructure, in particular, at the interfaces, so that the current spin polarizations in the semiconductor and the ferromagnets can be equated directly there. On the other hand, in the presence of ballistic contributions, the thermoballistic chemical potential $`\mu (x)`$ and the spin transport function $`A(x)`$ are not continuous at the interfaces. The discontinuities at the interfaces are taken into account when the functions $`\mu (x)`$ and $`A(x)`$ inside the semiconductor are calculated in terms of their values $`\mu _{1,2}`$ and $`A_{1,2}`$, respectively. The latter values are to be equated with the values of the corresponding quantities in the ferromagnet.
We begin with a brief summary of the standard description (see, e.g., Ref. yuf02a, ) of the spin polarization in the ferromagnets.
### IV.1 Current spin polarization in the ferromagnets
In the (semi-infinite) left ferromagnet located in the range $`x<x_1`$, the spin-up and spin-down chemical potentials $`\mu _{}(x)`$ are given by
$$\mu _{}(x)=\frac{e^2J}{\sigma _1}(x_1x)\pm \frac{C_1}{\sigma _{}^{(1)}}e^{(x_1x)/L_s^{(1)}},$$
(139)
where $`L_s^{(1)}`$ is the spin diffusion length. The quantities $`\sigma _{}^{(1)}`$ are the conductivities for spin up and spin down, which are independent of position, and $`\sigma _1=\sigma _{}^{(1)}+\sigma _{}^{(1)}`$. We then have $`C_1=\sigma _{}^{(1)}\mu _{}(x_1^{})=\sigma _{}^{(1)}\mu _{}(x_1^{})`$, and therefore $`C_1=(\sigma _{}^{(1)}\sigma _{}^{(1)}/\sigma _1)\mu _{}(x_1^{})`$, where $`\mu _{}(x)=\mu _{}(x)\mu _{}(x)`$. With
$$J_{}(x)=\frac{\sigma _{}^{(1)}}{e^2}\frac{d\mu _{}(x)}{dx},$$
(140)
we now find for the current spin polarization
$$P_J(x)=P_1\frac{G_1}{2e^2J}\mu _{}(x_1^{})e^{(x_1x)/L_s^{(1)}},$$
(141)
where $`P_1=(\sigma _{}^{(1)}\sigma _{}^{(1)})/\sigma _1`$ is the relaxed (current or density) spin polarization in the left ferromagnet, and
$$G_1=\frac{4\sigma _{}^{(1)}\sigma _{}^{(1)}}{\sigma _1L_s^{(1)}}=\frac{\sigma _1}{L_s^{(1)}}\left(1P_1^2\right)$$
(142)
is a transport parameter of the ferromagnet, which has the dimension of interface conductance. Analogously, we obtain
$$P_J(x)=P_2+\frac{G_2}{2e^2J}\mu _{}(x_2^+)e^{(xx_2)/L_s^{(2)}}$$
(143)
for the current spin polarization in the right ferromagnet located in the range $`x>x_2`$.
In the absence of spin-selective interface resistances, the chemical-potential splitting $`\mu _{}(x)`$ is continuous at the interface, $`\mu _{}(x_1^{})=\mu _{}(x_1)`$ and $`\mu _{}(x_2)=\mu _{}(x_2^+)`$, where $`\mu _{}(x_{1,2})`$ are its values at the interface itself. The latter are to be set equal to the corresponding values in the semiconductor, which yields
$`[\mu _{}(x_{1,2})]_{\mathrm{ferromagnet}}`$ $`=`$ $`[\mu _{}(x_{1,2})]_{\mathrm{semiconductor}}`$ (144)
$`=`$ $`{\displaystyle \frac{1}{\beta }}\mathrm{ln}\left({\displaystyle \frac{1+\alpha _{1,2}}{1\alpha _{1,2}}}\right),`$
where the right-hand part of this equation follows from Eq. (95) for $`x^{}=x_{1,2}`$, and $`\alpha _{1,2}=\alpha _{}(x_{1,2})`$. For the current spin polarizations at the interfaces, $`P_J(x_{1,2})`$, we have from Eqs. (141) and (143)
$$P_J(x_1)=P_1\frac{G_1}{2\beta e^2J}\mathrm{ln}\left(\frac{1+\alpha _1}{1\alpha _1}\right),$$
(145)
$$P_J(x_2)=P_2+\frac{G_2}{2\beta e^2J}\mathrm{ln}\left(\frac{1+\alpha _2}{1\alpha _2}\right),$$
(146)
which are to be set equal to the corresponding polarizations of the semiconductor.
Spin-selective interface resistances $`\rho _{}^{(1,2)}`$ are introduced via discontinuities of the spin-resolved chemical potentials on the contact sides of the interfaces. At $`x=x_1`$, for example, the discontinuity has the form
$$\mu _{}(x_1^{})\mu _{}(x_1)=e^2J_{}(x_1)\rho _{}^{(1)}.$$
(147)
The corresponding interface resistance is located between $`x=x_1^{}`$ and $`x_1`$ (in the ferromagnetic contact), and thus is adjacent to the Sharvin interface resistance between $`x=x_1`$ and $`x_1^+`$ (in the semiconductor). The quantity $`\mu _{}(x_1^{})`$ to be substituted in Eq. (141) is obtained, using Eqs. (139), (140), and (142), as
$`\mu _{}(x_1^{})`$ $`=`$ $`{\displaystyle \frac{1}{1+G_1\rho _+^{(1)}/4}}`$
$`\times \left\{\mu _{}(x_1)+{\displaystyle \frac{e^2J}{2}}\left[P_1\rho _+^{(1)}+\rho _{}^{(1)}\right]\right\},`$
where $`\rho _\pm ^{(1)}=\rho _{}^{(1)}\pm \rho _{}^{(1)}`$. The connection of $`\mu _{}(x_1)`$ with the spin fraction excess $`\alpha _1`$ is, as before, given by Eq. (144). The same procedure applies mutatis mutandis to the interface at $`x=x_2`$.
### IV.2 Spin polarization across a heterostructure in the zero-bias limit
In the zero-bias limit $`J0`$, we now demonstrate the procedure for calculating the current and density spin polarizations across a ferromagnet/semiconductor heterostructure.
Evaluating expressions (LABEL:eq:23) and (LABEL:eq:24), respectively, with $`A(x)`$ given by Eq. (128), we find for the thermoballistic spin-polarized current in the semiconductor
$$J_{}(x)=2v_eN_c\overline{l}\frac{dA(x)}{dx},$$
(149)
and for the thermoballistic spin-polarized density
$$n_{}(x)=N_cA(x)$$
(150)
$`(x_1<x<x_2)`$. For zero bias, one has $`J(x)=J=\mathrm{const}.`$ and $`n(x)=\overline{n}=\mathrm{const}.`$ \[see Eqs. (70) and (72)\], so that, by combining expressions (149) and (150), we obtain the relation
$$P_J(x)=\frac{2v_e\overline{n}\overline{l}}{J}\frac{dP_n(x)}{dx}$$
(151)
between the current and density spin polarizations. Furthermore, $`J_{}(x)`$ and $`n_{}(x)`$ both satisfy Eq. (126), and so do the polarizations $`P_J(x)`$ and $`P_n(x)`$ given by Eqs. (110) and (111), respectively. Differentiation of Eq. (151) then yields, together with Eq. (126) for $`P_n(x)`$,
$$P_n(x)=\frac{l_sJ}{2v_e\overline{n}}\frac{dP_J(x)}{dx}.$$
(152)
From Eq. (110) with $`J(x)=J`$, we find, using Eqs. (128) and (149), the explicit form of the current spin polarization as
$`P_J(x)`$ $`=`$ $`{\displaystyle \frac{2v_eN_c\overline{l}}{LJ}}\left[C_1e^{(xx_1)/L}C_2e^{(x_2x)/L}\right].`$
The density spin polarization is obtained from Eqs. (128) and (150) \[or, equivalently, from Eqs. (152) and (LABEL:eq:40c)\] as
$$P_n(x)=\frac{N_c}{\overline{n}}\left[C_1e^{(xx_1)/L}+C_2e^{(x_2x)/L}\right].$$
(154)
The coefficients $`C_{1,2}`$ in Eqs. (LABEL:eq:40c) and (154) can be expressed via Eqs. (129)–(132), using Eq. (94), in terms of the spin fraction excesses $`\alpha _{1,2}`$ on the contact sides of the interfaces.
In order to determine the quantities $`\alpha _{1,2}`$, we consider the current spin polarization (LABEL:eq:40c) on the semiconductor sides of the interfaces,
$$P_J(x_1^+)=\frac{𝒢}{\beta e^2J}(g\alpha _1h\alpha _2),$$
(155)
$$P_J(x_2^{})=\frac{𝒢}{\beta e^2J}(h\alpha _1g\alpha _2);$$
(156)
here, $`𝒢`$ is the Sharvin interface conductance given by Eq. (51), and the coefficients $`g`$ and $`h`$ are given by Eqs. (135) and (136), respectively. As mentioned before, $`P_J(x_1)=P_J(x_1^+)`$ and $`P_J(x_2^{})=P_J(x_2)`$, and the connection with the polarization in the contacts is made by equating expressions (145) and (155), and expressions (146) and (156) \[note that, if spin-selective interface resistances are included, expression (145) for $`P_J(x_1)`$ is to be replaced with the general expression obtained by using expression (LABEL:eq:karol) for $`\mu _{}(x_1^{})`$ in Eq. (141), and analogously for $`P_J(x_2)`$\]. In the zero-bias limit, when $`J(x_{1,2})=J`$ \[or $`\kappa =1`$ in Eq. (24)\], we have $`|\alpha _{1,2}|1`$, and this procedure then results in the system of coupled linear equations
$$\left(g+\stackrel{~}{G}_1\right)\alpha _1h\alpha _2=\frac{\beta e^2J}{𝒢}P_1,$$
(157)
$$h\alpha _1\left(g+\stackrel{~}{G}_2\right)\alpha _2=\frac{\beta e^2J}{𝒢}P_2,$$
(158)
where
$$\stackrel{~}{G}_{1,2}=\frac{G_{1,2}}{𝒢}.$$
(159)
The solutions of Eqs. (157) and (158) are found to be
$$\alpha _1=\frac{\beta e^2J}{𝒢\mathrm{\Delta }}\left[\left(g+\stackrel{~}{G}_2\right)P_1hP_2\right],$$
(160)
$$\alpha _2=\frac{\beta e^2J}{𝒢\mathrm{\Delta }}\left[hP_1\left(g+\stackrel{~}{G}_1\right)P_2\right],$$
(161)
where
$$\mathrm{\Delta }=\left(g+\stackrel{~}{G}_1\right)\left(g+\stackrel{~}{G}_2\right)h^2.$$
(162)
Expressions (160) and (161) determine the spin fraction excesses $`\alpha _1`$ and $`\alpha _2`$ in terms of the current $`J`$, of the polarizations $`P_1`$ and $`P_2`$ in the left and right ferromagnet, respectively, and of material parameters, such as the conductivities $`\sigma _{1,2}`$ and the spin diffusion lengths $`L_s^{(1,2)}`$ of the ferromagnets (via $`\stackrel{~}{G}_{1,2}`$), and the momentum relaxation length $`l`$ and the spin relaxation length $`l_s`$ of the semiconductor as well as its length $`S`$ (via $`g`$ and $`h`$) and the equilibrium density $`\overline{n}`$ (via $`𝒢`$). Since the quantities $`\alpha _{1,2}`$ are proportional to the current $`J`$, the current spin polarization $`P_J(x)`$ is independent of $`J`$, while the density spin polarization $`P_n(x)`$ is proportional to $`J`$.
The current spin polarization along the entire heterostructure, $`P_J(x)`$, is now obtained as follows. In the semiconductor, it is given by expression (LABEL:eq:40c), with $`C_{1,2}`$ calculated from $`\alpha _{1,2}`$ as explained there. In the ferromagnets, the expressions for the current spin polarization are provided by Eqs. (141) and (143), respectively, where the quantities $`\mu _{}(x_1^{})`$ and $`\mu _{}(x_2^+)`$ are calculated from Eq. (LABEL:eq:karol) and from its analogue for $`\mu _{}(x_2^+)`$, respectively. Analogously, the density spin polarization $`P_n(x)`$ in the semiconductor is given by expression (152). We do not write down the density spin polarizations in the ferromagnets, but only mention that they do not, in general, match the polarizations $`P_n(x_1^+)`$ and $`P_n(x_2^{})`$ on the semiconductor sides of the interfaces.
In order to demonstrate the effect of the transport mechanism (characterized by the magnitude of the ratios $`l/l_s`$ and $`l/S`$), we show in Fig. 5 the zero-bias current spin polarization $`P_J(x)`$ for a symmetric ferromagnet/semiconductor/ferromagnet heterostructure with sample length $`S=1`$ $`\mu `$m at $`T=300`$ K as a function of $`x`$ for various values of the momentum relaxation length $`l`$. For the parameters of the ferromagnets, we adopt from Ref. yuf02a, the values $`\sigma _1=\sigma _2=10^3`$ $`\mathrm{\Omega }^1`$ cm<sup>-1</sup> for the bulk conductivities and $`L_s^{(1)}=L_s^{(2)}=60`$ nm for the spin diffusion lengths; the bulk polarizations are chosen as $`P_1=P_2=0.8`$. For the material parameters of the semiconductor, we take the values $`m^{}=0.067m_e`$ for the effective electron mass, $`l_s=2.5`$ $`\mu `$m for the ballistic spin relaxation length (corresponding to n-doped GaAs; see Refs. kim01, and bec05, ), and $`\overline{n}=5.0\times 10^{17}`$ cm<sup>-3</sup> for the equilibrium electron density. Clearly, in a specific semiconducting system, the value of the momentum relaxation length $`l`$ is fixed. Therefore, when varying $`l`$, we are considering the above parameter values to be representative for a whole class of semiconductors (regarded as nondegenerate; at room temperature, this should be an acceptable working hypothesisami04 ) that differ in the strength of impurity and phonon scattering and hence in the magnitude of $`l`$.
The momentum relaxation length $`l`$ affects the results shown in Fig. 5 in a twofold way. (i) It determines the conduction in the semiconductor. For small values of $`l`$, the conductance of the latter is small, and thus the conductance mismatch with the ferromagnets is large, leading to a small injected current spin polarization $`P_J(0)`$. (ii) It determines the generalized spin diffusion length $`L=[ll_s/(1+l/l_s)]^{1/2}`$, which acts as the polarization decay length, so that for small $`l`$ the polarization dies out rapidly inside the semiconductor. The degree of polarization may be raised considerably all along the semiconductor when the value of $`l`$ is increased up to a length of the order of the sample length, in which case the ballistic component becomes prevalent. Figure 5 also shows that, by introducing appropriately chosen spin-selective interface resistances, one may offset the suppression of the injected polarization due to the conductance mismatch for small $`l`$; however, the rapid decay of the polarization inside the semiconductor cannot be prevented in this way.
For the case of Fig. 5, we show in Figs. 6 and 7, respectively, the zero-bias current spin polarization $`P_J(x)`$ for various values of the equilibrium density $`\overline{n}`$ and the spin relaxation length $`l_s`$. It is seen that varying $`\overline{n}`$ has about the same overall effect on $`P_J(x)`$ as varying the momentum relaxation length $`l`$, whereas varying $`l_s`$ affects mainly the rate of decay of $`P_J(x)`$.
### IV.3 Injected spin polarization for field-driven transport
We introduce the “injected spin polarization” as the spin polarization at one of the interfaces, e.g., at $`x=x_1`$, generated by the bulk polarization $`P_1`$ of the left ferromagnet regardless of the influence of the right ferromagnet. More precisely, we define the injected current spin polarization as the current spin polarization $`P_J(x_1^+)`$ given by Eq. (110) in the limit $`S/L\mathrm{}`$. Similarly, the injected density spin polarization is defined as the polarization $`P_n(x_1^+)`$ of Eq. (111) in the same limit. The injected spin polarization at $`x=x_1^+`$ provides the initial value of the left-generated polarization in the semiconductor, which propagates into the region $`x>x_1`$ while being degraded by the effect of spin relaxation.
We now consider the injected spin polarization for electron transport driven by an external electric field, i.e., a potential profile of the form (64).
#### IV.3.1 General case
In order to obtain the spin transport function $`A(x)`$, we have to solve Eq. (117) numerically under the condition $`A(x)\mathrm{exp}(x/\lambda )`$ for $`x\mathrm{}`$. The decay length $`\lambda `$ is determined by solving Eq. (117) in the range $`xx_1(ϵ+1/l)^1`$ where the function $`b(x)`$ in the coefficient functions $`b_0(x)`$, $`b_1(x)`$, and $`b_2(x)`$ can be disregarded,
$`\lambda `$ $`=`$ $`\left\{{\displaystyle \frac{ϵ}{2}}+\left[{\displaystyle \frac{ϵ^2}{4}}+{\displaystyle \frac{1+ϵ\stackrel{~}{l}}{\stackrel{~}{l}^2}}{\displaystyle \frac{1+ϵl}{l\overline{l}}}{\displaystyle \frac{2+ϵ\overline{l}}{2+ϵl}}\right]^{1/2}\right\}^1.`$
It can be shown that for any combination of parameter values, $`\lambda `$ is a real number.
For calculating the injected current spin polarization from Eq. (110), we determine the thermoballistic spin-polarized current at the interface, $`J_{}(x_1^+)`$, from Eq. (LABEL:eq:23). Using Eq. (94) and fixing the normalization of the function $`A(x)`$ in terms of $`A_1`$ with the help of Eq. (112), we find
$$J_{}(x_1^+)=v_e\overline{n}\mathrm{\Gamma }_J\alpha _1,$$
(164)
where
$$\mathrm{\Gamma }_J=\frac{A(x_1^+)\overline{A}}{A(x_1^+)\overline{A}/2}$$
(165)
and
$$\overline{A}=_{x_1}^{\mathrm{}}\frac{dx}{l}e^{(xx_1)/\stackrel{~}{l}}A(x).$$
(166)
To find the total thermoballistic current at the interface, $`J(x_1^+)`$, we go back to Eq. (24). Expressing the current $`J`$ in the form
$$J=\frac{1}{\stackrel{~}{\chi }}v_e\overline{n},$$
(167)
which follows, for $`ϵ>0`$ and $`S/L\mathrm{}`$, from the current-voltage characteristic (42) with $`\beta eV=ϵS`$ and $`N_c\mathrm{exp}(\beta E_p)=\overline{n}`$, we obtain
$$J(x_1^+)=\frac{\kappa }{\stackrel{~}{\chi }}v_e\overline{n}.$$
(168)
This expression is conveniently evaluated by using for $`\kappa `$ and $`\stackrel{~}{\chi }`$ the closed-form representations
$$\kappa =\frac{1+ϵl}{2+ϵl},\stackrel{~}{\chi }=\frac{(1+ϵl)^2}{ϵl(2+ϵl)},$$
(169)
which have been inferred from the results of systematic numerical calculations for fixed $`ϵl>0`$ and very large values of $`S/L`$. For the injected current spin polarization, we now find
$$P_J(x_1^+)=\frac{J_{}(x_1^+)}{J(x_1^+)}=\frac{\stackrel{~}{\chi }}{\kappa }\mathrm{\Gamma }_J\alpha _1,$$
(170)
which, by continuity, is equal to $`P_J(x_1)`$. Setting the right-hand side of Eq. (170) equal to expression (145) \[or to the more general expression including spin-selective interface resistances; see the remark following Eqs. (155) and (156)\] for the injected spin polarization in terms of the contact parameters, we arrive at
$$P_1\frac{\stackrel{~}{G}_1}{2}\stackrel{~}{\chi }\mathrm{ln}\left(\frac{1+\alpha _1}{1\alpha _1}\right)=\frac{\stackrel{~}{\chi }}{\kappa }\mathrm{\Gamma }_J\alpha _1.$$
(171)
This is a nonlinear equation for $`\alpha _1`$ which is to be solved for given values of the parameters $`ϵ`$, $`P_1`$, $`G_1`$, $`\overline{n}`$, $`l`$, and $`l_s`$.
Turning to the calculation of the injected density spin polarization, we determine the thermoballistic spin-polarized density at the interface, $`n_{}(x_1^+)`$, from Eq. (LABEL:eq:24), using again Eqs. (94) and (112), and obtain
$$n_{}(x_1^+)=\frac{\overline{n}}{2}\mathrm{\Gamma }_n\alpha _1,$$
(172)
where
$$\mathrm{\Gamma }_n=\frac{A(x_1^+)}{A(x_1^+)\overline{A}/2}.$$
(173)
For the total thermoballistic density at the interface, $`n(x_1^+)`$, we find from Eq. (LABEL:eq:11ahab), using Eqs. (48), (LABEL:eq:11aha), and (167),
$$n(x_1^+)=\overline{n}\left(1\frac{\kappa J}{2v_e\overline{n}}\right)=\overline{n}\left(1\frac{\kappa }{2\stackrel{~}{\chi }}\right).$$
(174)
The injected density spin polarization now follows as
$$P_n(x_1^+)=\frac{n_{}(x_1^+)}{n(x_1^+)}=\frac{\stackrel{~}{\chi }}{2\stackrel{~}{\chi }\kappa }\mathrm{\Gamma }_n\alpha _1,$$
(175)
where the spin fraction excess $`\alpha _1`$ is again to be determined by solving Eq. (171).
Figure 8 shows the injected current spin polarization $`P_J(x_1)`$ for $`S/L\mathrm{}`$ as a function of the electric-field parameter $`ϵ`$ for various values of the momentum relaxation length $`l`$; the remaining parameter values are the same as in Fig. 5. In calculating $`P_J(x_1)`$ from Eq. (170), we have used $`\kappa =1`$. This choice has been made because expression (170) with $`\kappa `$ given by Eq. (169) does not represent a meaningful injected polarization in the diffusive limit (see below); instead, one must set $`\kappa =1`$ in this limit. For simplicity, we have used this value throughout. In conformity with the drift-diffusion results of Ref. yuf02a, , the injected polarization generally rises with increasing $`ϵ`$; however, as in Fig. 5, the main effect is due to the variation of $`l`$.
#### IV.3.2 Diffusive regime
In order to relate our treatment of the injected spin polarization at ferromagnet/semiconductor interfaces to previous treatments within the drift-diffusion model, in particular to that of Yu and Flatté,yuf02a we consider the diffusive regime, $`l/l_s1`$ and $`ϵl1`$, in some detail. In that regime, the spin transport function $`A(x)`$ is determined by Eq. (124), whose solution is $`A(x)\mathrm{exp}(x/L_s^ϵ)`$, with the field-dependent spin diffusion length $`L_s^ϵ`$ given by
$$\frac{1}{L_s^ϵ}=\frac{ϵ}{2}+\left(\frac{ϵ^2}{4}+\frac{1}{L_s^2}\right)^{1/2}.$$
(176)
We then obtain $`\mathrm{\Gamma }_J=2l/L_s^ϵ`$ and $`\mathrm{\Gamma }_n=2`$. Furthermore, from Eqs. (169) for $`ϵl1`$, we find $`\kappa =1/2`$ and $`\stackrel{~}{\chi }=1/2ϵl`$.
At this point, some analysis is required regarding the definition of the injected current spin polarization in the diffusive regime. In the definition introduced above, first the functions $`J_{}(x)`$ and $`J(x)`$ are evaluated for $`xx_1^+`$, and subsequently the diffusive limit is approached. This procedure results, in particular, in the value $`\kappa =J(x_1^+)/J=1/2`$. A closer look at the function $`J(x)/J`$ (see Fig. 3), however, shows that in the diffusive regime this function is virtually equal to unity inside the semiconducting sample and tends to smaller values only within a (very short) distance of order $`l`$ from the interfaces. Therefore, it is indicated here to define the injected current spin polarization in terms of a position $`x>l`$ inside the sample, where $`J(x)/J=1`$ is the relevant value for the propagation of the spin polarization into the semiconductor. Thus, in the diffusive regime, we adopt the effective value $`\kappa =1`$ in the calculation of the injected spin polarization. While in the ballistic limit (and now also in the diffusive regime) the choice $`\kappa =1`$ is unique, in the range of intermediate $`l`$-values a meaningful definition of the injected spin polarization requires an appropriate choice of the position inside the sample at which the thermoballistic current and spin-polarized current are to be evaluated.
With the choice $`\kappa =1`$, the injected current spin polarization in the diffusive regime is obtained from Eq. (170) as
$$P_J(x_1)=\frac{1}{ϵL_s^ϵ}\alpha _1,$$
(177)
where the spin fraction excess $`\alpha _1`$ is now to be calculated from Eq. (171) with $`\kappa =1`$. Since $`\kappa \stackrel{~}{\chi }`$ for $`ϵl1`$, the injected density spin polarization in the diffusive regime follows from Eq. (175) as
$$P_n(x_1^+)=\alpha _1,$$
(178)
where $`\alpha _1`$ again must be calculated from Eq. (171) with $`\kappa =1`$.
Comparing our results for the injected spin polarization in the diffusive regime to the results of Yu and Flattéyuf02a based on standard drift-diffusion theory, we find that the field-dependent spin diffusion length $`L_s^ϵ`$ given by Eq. (176) agrees with the “up-stream” spin diffusion length $`L_u`$ given by Eq. (2.23b) of Ref. yuf02a, , provided the intrinsic spin diffusion length $`L`$ of that reference is identified with the spin diffusion length $`L_s=\sqrt{ll_s}`$ of the present work. Then, by expressing the conductivity of the semiconductor in Eq. (3.5) of Ref. yuf02a, (with the interface resistances set equal to zero) as $`\sigma _s=2\beta e^2v_e\overline{n}_1l`$, we recognize the equivalence of that equation with our Eq. (171) in the diffusive regime. This, in turn, implies that the injected current and density spin polarizations of either work are formally identical. Numerical calculations have confirmed this result.
#### IV.3.3 Zero-bias limit
We now consider the injected current spin polarization in the zero-bias limit, in which $`|\alpha _1|1`$ and $`J(x)=J`$, i.e., $`\kappa =1`$. Here, the spin transport function $`A(x)`$ is determined by Eq. (126), i.e., $`A(x)\mathrm{exp}(x/L)`$, so that $`\mathrm{\Gamma }_J=2\gamma /(1+\gamma )\gamma _J`$ and $`\mathrm{\Gamma }_n=2/(1+\gamma )\gamma _n`$. From Eq. (171), we then have
$$\alpha _1=\frac{P_1}{\stackrel{~}{\chi }\gamma _J(1+G_1/\gamma _J𝒢)}.$$
(179)
Combining this with Eq. (170), we find for the injected current spin polarization
$$P_J(x_1)=\frac{1}{1+G_1/\gamma _J𝒢}P_1.$$
(180)
It is instructive to consider expression (180) in the diffusive and ballistic regimes.
In the diffusive regime $`l/l_s1`$, we have $`\gamma _J=2\sqrt{l/l_s}`$ and, therefore,
$$P_J(x_1)=\frac{1}{1+G_1/𝒢_0}P_1.$$
(181)
Here, $`𝒢_0=2𝒢\sqrt{l/l_s}=\sigma _0/L_s`$, where Eq. (79) has been used, and $`\sigma _0=2𝒢l`$ is the conductivity of the semiconductor \[see the remarks following Eq. (71)\]. The quantity $`𝒢_0`$ is seen to be the semiconductor analogue of the ferromagnet parameter $`G_1`$ defined by Eq. (142). Choosing $`P_1=0.8`$ and adopting the valuesyuf02a $`\sigma _1=10^3`$ $`\mathrm{\Omega }^1\mathrm{cm}^1`$, $`L_s^{(1)}=60`$ nm, $`\sigma _0=10`$ $`\mathrm{\Omega }^1\mathrm{cm}^1`$, and $`L_s=2`$ $`\mu `$m, we have $`G_1/𝒢_0=1.2\times 10^3`$ and hence $`P_J(x_1)0.6\times 10^3`$. The large value of the ratio $`G_1/𝒢_0`$ reflects the “conductance mismatch” which appears to be the determining parameter of the injected spin polarization in the diffusive regime.sch02 ; sch05 ; sch00
On the other hand, in the ballistic limit $`l/l_s\mathrm{}`$, we have $`\gamma _J=\gamma =1`$, so that
$$P_J(x_1)=\frac{1}{1+G_1/𝒢}P_1.$$
(182)
Here, the Sharvin interface conductance $`𝒢`$ takes the place of the quantity $`𝒢_0`$ in Eq. (181). Assuming $`m^{}=0.067m_e`$ and $`T=300`$ K, we have $`𝒢=3.2\times 10^{11}`$ $`\mathrm{\Omega }^1`$m<sup>-2</sup> for $`\overline{n}=5\times 10^{17}`$ cm<sup>-3</sup> and $`𝒢=0.64\times 10^{10}`$ $`\mathrm{\Omega }^1`$m<sup>-2</sup> for $`\overline{n}=10^{16}`$ cm<sup>-3</sup>. This results in $`P_J(x_1)0.3`$ and $`0.8\times 10^2`$, respectively. The first example, where the large Sharvin interface conductance entails a large injected spin polarization, is fictitious since the high doping concentration needed to obtain an electron density of $`5\times 10^{17}`$ cm<sup>-3</sup> (for example, in GaAs) would imply such small values of the momentum relaxation length $`l`$ that ballistic transport is all but ruled out. Only semiconducting materials with unusually large mobilities would make this a realistic case. The second example with the lower electron density of $`10^{16}`$ cm<sup>-3</sup> would be more favorable to a ballistic transport mechanism, but leads to a very small injected spin polarization; this confirms the conclusion of Kravchenko and Rashbakra03 stating that spin injection is suppressed even in the ballistic regime unless spin-selective interface resistances are introduced.
## V Concluding remarks
We have developed a unified semiclassical theory of spin-polarized electron transport in heterostructures formed of a nondegenerate semiconductor and two ferromagnetic contacts. In this theory, the spin polarization inside the semiconductor is obtained for a general transport mechanism that covers the whole range between the purely diffusive and purely ballistic mechanisms and is controlled by the momentum relaxation length of the electrons.
The basis of the present work is provided by our previously developed unified model of (spinless) electron transport in semiconductors, in which diffusive and ballistic transport are combined in the concept of the thermoballistic electron current. As a prerequisite to the extension of the spinless unified model to spin-polarized transport, we have modified and completed its formulation in such a way that an unambiguous description of electron transport in terms of a uniquely defined thermoballistic chemical potential is achieved. From the chemical potential, the unique thermoballistic current and density are obtained; numerical calculations show that, for typical parameter values, the thermoballistic current is close to the physical current.
In order to treat spin-polarized transport in semiconductors within the unified description, we have introduced a thermoballistic spin-polarized current and a thermoballistic spin-polarized density by allowing spin relaxation to take place during the ballistic electron motion. These are expressed in terms of a spin transport function which comprises in a compact form the information contained in the spin-resolved thermoballistic chemical potentials. Using the balance equation that connects the thermoballistic spin-polarized current and density, we have derived an integral equation for the spin transport function, from which the latter can be calculated in terms of its values at the interfaces of the semiconductor with the contacts. The spin transport function determines, in conjunction with the spin-independent thermoballistic chemical potential, all spin-dependent quantities in the semiconductor, in particular, the position dependence of the current and density spin polarization. The spin polarization all across a ferromagnet/semiconductor heterostructure is determined by making use of the continuity of the current spin polarization at the contact-semiconductor interfaces and connecting the spin-resolved chemical potentials there. Thereby, a unified description of spin-polarized transport emerges that provides a basis for the systematic study of the interplay of spin relaxation and transport mechanism in heterostructures relevant to spintronic applications.
To interpret the formalism developed here and to relate it to previous, less general formulations, we have considered spin-polarized electron transport in a homogeneous semiconductor without space charge, driven by an external electric field. Within a judicious approximation, the integral equation for the spin transport function can then be reduced to a second-order differential equation which generalizes the standard spin drift-diffusion equation to the case of arbitrary values of the ratio of momentum to spin relaxation length. In the zero-bias limit, the position dependence of the spin polarizations across a heterostructure is obtained in closed form.
The generalized spin drift-diffusion equation has been used in calculations of the current spin polarization across a symmetric ferromagnet/semiconductor/ferromagnet heterostructure with material parameters in the range of interest for spintronic devices. The dependence on the transport mechanism in the semiconductor has been exhibited by varying the momentum relaxation length over several orders of magnitude. It was found that the ballistic regime favors sizeable (large) spin polarizations. The same picture emerges from calculations of the injected current spin polarization as a function of an applied electric field. While the field also serves to raise the polarization in the semiconductor, the main effect still is due to the variation of the momentum relaxation length, i.e., to the influence of the ballistic component of the transport mechanism. In order to exploit the potentiality of varying the transport mechanism with the aim to improve the efficiency of spintronic devices, the identification and design of novel semiconducting materials is called for.
In the present work, emphasis has been placed on a careful elaboration of the formalism underlying the unified description of spin-polarized electron transport in ferromagnet/semiconductor heterostructures. In the illustrative calculations, we have restricted ourselves to the simplest cases. In future work, applications of the present formalism will have to be based on the solution of the fundamental integral equation for the spin transport function in its general form. These should include the treatment of magnetic semiconducting samples (characterized, in the unified description, by nonzero values of the relaxed spin fraction excess) and of interface barriers, like Schottky or tunnel barriers (represented by appropriately chosen potential profiles). As to possible extensions of the theory, setting up a formalism for the treatment of degenerate semiconductors appears to have first priority.
*
## Appendix A Unique thermoballistic functions
In this Appendix, we present details of the construction of a unique thermoballistic chemical potential $`\mu (x)`$, current $`J(x)`$, and density $`n(x)`$ in terms of the solutions $`\chi _1(x)`$ and $`\chi _2(x)`$ of Eqs. (LABEL:eq:11bb) and (LABEL:eq:11bbb), respectively.
Evaluating expression (LABEL:eq:11a) with the function $`𝒥_1(x)`$ following from the solution of Eq. (LABEL:eq:11bb) \[case (i)\], with $`𝒥_2`$ set equal to $`𝒥_1(x_2)`$, we obtain the thermoballistic current
$`J_1(x)=w_1(x_1,x_2;l)[𝒥_1𝒥_1(x_2)]`$
$`+`$ $`{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}w_2(x^{},x_2;l)[𝒥_1(x^{})𝒥_1(x_2)]`$
$`+`$ $`{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{}}{l}}w_2(x_1,x^{};l)[𝒥_1𝒥_1(x^{})]`$
$`+`$ $`{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{\prime \prime }}{l}}w_3(x^{},x^{\prime \prime };l)[𝒥_1(x^{})𝒥_1(x^{\prime \prime })]`$
and, similarly, for case (ii), the current
$`J_2(x)=w_1(x_1,x_2;l)[𝒥_2(x_1)𝒥_2]`$
$`+`$ $`{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}w_2(x^{},x_2;l)[𝒥_2(x^{})𝒥_2]`$
$`+`$ $`{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{}}{l}}w_2(x_1,x^{};l)[𝒥_2(x_1)𝒥_2(x^{})]`$
$`+`$ $`{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{\prime \prime }}{l}}w_3(x^{},x^{\prime \prime };l)[𝒥_2(x^{})𝒥_2(x^{\prime \prime })],`$
which are not, in general, equal. This ambiguity is removed by introducing a unique thermoballistic current $`J^{(u)}(x)`$ as a superposition of the currents $`J_1(x)`$ and $`J_2(x)`$,
$$J^{(u)}(x)=\widehat{a}_1\frac{J}{J_1}J_1(x)+\widehat{a}_2\frac{J}{J_2}J_2(x),$$
(185)
where
$$J_{1,2}=\frac{1}{x_2x_1}_{x_1}^{x_2}𝑑xJ_{1,2}(x).$$
(186)
The current $`J^{(u)}(x)`$ has to satisfy Eq. (20),
$$\frac{1}{x_2x_1}_{x_1}^{x_2}𝑑xJ^{(u)}(x)=J.$$
(187)
Similarly, Eq. (24) is to be replaced with
$$J^{(u)}(x_1^+)=J^{(u)}(x_2^{})\kappa J.$$
(188)
The conditions (187) and (188) determine the coefficients $`\widehat{a}_1`$ and $`\widehat{a}_2`$. We find from Eq. (187), using Eqs. (185) and (186),
$$\widehat{a}_1+\widehat{a}_2=1.$$
(189)
In order to apply condition (188), we evaluate the currents (LABEL:eq:11ax) and (LABEL:eq:11ay) at the ends of the sample, using Eqs. (27) and (30),
$`{\displaystyle \frac{J_1(x_1^+)}{J_1}}`$ $`=`$ $`w_1(x_1,x_2;l)\chi _1(x_2)`$ (190)
$`+`$ $`{\displaystyle _{x_1}^{x_2}}{\displaystyle \frac{dx^{}}{l}}w_2(x_1,x^{};l)\chi _1(x^{}),`$
$`{\displaystyle \frac{J_2(x_1^+)}{J_2}}`$ $`=`$ $`w_1(x_1,x_2;l)\chi _2(x_1)`$
$`+`$ $`{\displaystyle _{x_1}^{x_2}}{\displaystyle \frac{dx^{}}{l}}w_2(x_1,x^{};l)[\chi _2(x_1)\chi _2(x^{})],`$
$`{\displaystyle \frac{J_1(x_2^{})}{J_1}}`$ $`=`$ $`w_1(x_1,x_2;l)\chi _1(x_2)`$
$`+`$ $`{\displaystyle _{x_1}^{x_2}}{\displaystyle \frac{dx^{}}{l}}w_2(x^{},x_2;l)[\chi _1(x_2)\chi _1(x^{})],`$
$`{\displaystyle \frac{J_2(x_2^{})}{J_2}}`$ $`=`$ $`w_1(x_1,x_2;l)\chi _2(x_1)`$ (193)
$`+`$ $`{\displaystyle _{x_1}^{x_2}}{\displaystyle \frac{dx^{}}{l}}w_2(x^{},x_2;l)\chi _2(x^{}).`$
Employing Eqs. (185), (188), and (190)–(193), we obtain the coefficients $`\widehat{a}_{1,2}`$ as given by Eqs. (35)–(38).
Now, introducing, in analogy to Eq. (185),
$$𝒥^{(u)}(x)=\widehat{a}_1\frac{J}{J_1}𝒥_1(x)+\widehat{a}_2\frac{J}{J_2}𝒥_2(x)$$
(194)
for $`x_1<x<x_2`$, and, in addition,
$$𝒥^{(u)}(x_1)=\widehat{a}_1\frac{J}{J_1}𝒥_1+\widehat{a}_2\frac{J}{J_2}𝒥_2(x_1),$$
(195)
$$𝒥^{(u)}(x_2)=\widehat{a}_1\frac{J}{J_1}𝒥_1(x_2)+\widehat{a}_2\frac{J}{J_2}𝒥_2,$$
(196)
we may write the unique thermoballistic current (185), using Eqs. (LABEL:eq:11ax), (LABEL:eq:11ay) and (194)–(196), in a symbolic form analogous to expression (LABEL:eq:11aba),
$`J^{(u)}(x)`$ $`=`$ $`{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{\prime \prime }}{l}}𝕎(x^{},x^{\prime \prime };l)`$ (197)
$`\times \left[𝒥^{(u)}(x^{})𝒥^{(u)}(x^{\prime \prime })\right];`$
here, the values $`𝒥^{(u)}(x_{1,2})`$ are to be identified with their physical values $`𝒥_{1,2}`$ in the contacts,
$$𝒥^{(u)}(x_{1,2})=𝒥_{1,2}.$$
(198)
In line with the definition (6) of the current $`𝒥(x)`$ in terms of the chemical potential $`\mu (x)`$, we now define a unique chemical potential $`\mu ^{(u)}(x)`$ via relation (194) by
$$e^{\beta \mu ^{(u)}(x)}=\frac{1}{v_eN_c}e^{\beta E_c^0}𝒥^{(u)}(x)$$
(199)
for $`x_1xx_2`$, where now
$$\mu ^{(u)}(x_{1,2})=\mu _{1,2}.$$
(200)
The chemical potential $`\mu ^{(u)}(x)`$ is the key quantity in the extended unified description of electron transport inside the sample. In terms of $`\mu ^{(u)}(x)`$, the unique ballistic current across the interval $`[x^{},x^{\prime \prime }]`$ appearing in the expression for the thermoballistic current (197) is given by Eq. (LABEL:eq:10xx1).
For the explicit calculation of the chemical potential $`\mu ^{(u)}(x)`$, we use Eq. (198) in Eqs. (195) and (196). We then find, with the help of Eqs. (27) and (30),
$$\left(1\widehat{a}_1\frac{J}{J_1}\right)𝒥_1\widehat{a}_2\frac{J}{J_2}𝒥_2=\widehat{a}_2J\chi _2(x_1),$$
(201)
$`\widehat{a}_1{\displaystyle \frac{J}{J_1}}𝒥_1+\left(1\widehat{a}_2{\displaystyle \frac{J}{J_2}}\right)𝒥_2`$ $`=`$ $`\widehat{a}_1J\chi _1(x_2).`$
(202)
By subtracting Eq. (202) from Eq. (201), we obtain Eq. (41). On the other hand, adding Eqs. (201) and (202) results in
$`\widehat{a}_1{\displaystyle \frac{J}{J_1}}𝒥_1+\widehat{a}_2{\displaystyle \frac{J}{J_2}}𝒥_2`$ $`=`$ $`{\displaystyle \frac{1}{2}}(𝒥_1+𝒥_2)`$
$`+`$ $`{\displaystyle \frac{J}{2}}[\widehat{a}_1\chi _1(x_2)\widehat{a}_2\chi _2(x_1)].`$
Then, expressing the current $`𝒥^{(u)}(x)`$ in terms of the quantities $`𝒥_{1,2}`$ and the functions $`\chi _{1,2}(x)`$ by combining Eq. (194) with Eqs. (27) and (30),
$`𝒥^{(u)}(x)`$ $`=`$ $`\widehat{a}_1{\displaystyle \frac{J}{J_1}}[𝒥_1J_1\chi _1(x)]`$
$`+\widehat{a}_2{\displaystyle \frac{J}{J_2}}[𝒥_2+J_2\chi _2(x)]`$
$`=`$ $`\widehat{a}_1{\displaystyle \frac{J}{J_1}}𝒥_1+\widehat{a}_2{\displaystyle \frac{J}{J_2}}𝒥_2`$
$`J[\widehat{a}_1\chi _1(x)\widehat{a}_2\chi _2(x)],`$
we obtain Eq. (39) with $`\chi _{}(x)`$ given by Eq. (LABEL:eq:12s). Using Eqs. (39) and (41) to eliminate the total current $`J`$ as well as Eqs. (199) to go over to the unique chemical potential $`\mu ^{(u)}(x)`$, we find
$`e^{\beta \mu ^{(u)}(x)}`$ $`=`$ $`\left[{\displaystyle \frac{1}{2}}{\displaystyle \frac{\chi _{}(x)}{\chi }}\right]e^{\beta \mu _1}+\left[{\displaystyle \frac{1}{2}}+{\displaystyle \frac{\chi _{}(x)}{\chi }}\right]e^{\beta \mu _2},`$
where $`\chi `$ is defined by Eq. (34). The corresponding thermoballistic current $`J^{(u)}(x)`$ and density $`n^{(u)}(x)`$ are obtained by substituting expression (LABEL:eq:12ijx) in Eq. (16) and (17), respectively.
In the main body of the paper, we always deal with the unique chemical potential, current, and density, and omit the superscript $`u`$; we have already adhered to this convention when referring from the Appendix to the equations of Sec. II.B.
|
warning/0506/math0506454.html
|
ar5iv
|
text
|
# A note on infinitely distributive inverse semigroupsResearch supported in part by FEDER and FCT through CAMGSD.
( )
By an *infinitely distributive* inverse semigroup will be meant an inverse semigroup $`S`$ such that for every subset $`XS`$ and every $`sS`$, if $`X`$ exists then so does $`(sX)`$, and furthermore $`(sX)=sX`$.
One important aspect is that the infinite distributivity of $`E(S)`$ implies that of $`S`$; that is, if the multiplication of $`E(S)`$ distributes over all the joins that exist in $`E(S)`$ then $`S`$ is infinitely distributive. This can be seen in Proposition 20, page 28, of Lawson’s book . Although the statement of the proposition mentions only joins of nonempty sets, the proof applies equally to any subset.
The aim of this note is to present a proof of an analogous property for binary meets instead of multiplication; that is, we show that for any infinitely distributive inverse semigroup the existing binary meets distribute over all the joins that exist.
A useful consequence of this lies in the possibility of constructing, from infinitely distributive inverse semigroups, certain quantales that are also locales (due to the stability of the existing joins both with respect to the multiplication and the binary meets), yielding a direct connection to étale groupoids via the results of . The consequences of this include an algebraic construction of “groupoids of germs” from certain inverse semigroups, such as pseudogroups, and will be developed elsewhere.
###### Lemma.
Let $`S`$ be an inverse semigroup, and let $`x,yS`$ be such that the meet $`xy`$ exists. Then the join
$$f=\{g(xx^1yy^1)gx=gy\}$$
exists, and we have $`xy=fx=fy`$.
###### Proof.
Consider the set $`Z`$ of lower bounds of $`x`$ and $`y`$,
$$Z=\{zSzx,zy\},$$
whose join is $`xy`$. By \[1, Prop. 17, p. 27\], the (nonempty) set
$$F=\{zz^1zZ\}$$
has a join $`f=F`$ that coincides with $`(Z)(Z)^1=(xy)(xy)^1`$. Hence, $`xy=fx=fy`$. The lemma now follows from the fact that the elements $`zz^1`$ with $`zZ`$ are precisely the idempotents $`g(xx^1yy^1)`$ such that $`gx=gy`$.
Under the assumption of infinite distributivity we have a converse:
###### Lemma.
Let $`S`$ be an infinitely distributive inverse semigroup, and let $`x,yS`$ be such that the join
$$f=\{g(xx^1yy^1)gx=gy\}$$
exists. Then the meet $`xy`$ exists, and we have $`xy=fx=fy`$.
###### Proof.
By \[1, Prop. 20, p. 28\], the join
$$\{gxgxx^1yy^1,gx=gy\}$$
exists and it equals $`fx`$. Similarly, the join
$$\{gygxx^1yy^1,gx=gy\}$$
exists and it equals $`fy`$. But the two sets of which we are taking joins are the same due to the condition $`gx=gy`$, and thus $`fx=fy`$. The element $`fx`$ is therefore a lower bound of both $`x`$ and $`y`$. Let $`z`$ be another such lower bound. Then $`z=zz^1x=zz^1y`$, and thus $`zz^1f`$, which implies $`zfx`$. Hence, $`fx`$ is the greatest lower bound of $`x`$ and $`y`$.
###### Theorem.
Let $`S`$ be an infinitely distributive inverse semigroup, let $`xS`$, and let $`(y_i)`$ be a family of elements of $`S`$. Assume that the join $`_iy_i`$ exists, and that the meet $`x_iy_i`$ exists. Then, for all $`i`$ the meet $`xy_i`$ exists, the join $`_i(xy_i)`$ exists, and we have
$$x\underset{i}{}y_i=\underset{i}{}(xy_i).$$
###### Proof.
Let us write $`y`$ for $`_iy_i`$, $`e_i`$ for $`y_iy_i^1`$, and let $`f`$ be the idempotent
$$f=\{g(xx^1yy^1)gx=gy\},$$
which exists, by the first lemma. Furthermore, also by the first lemma, we have $`xy=fx=fy`$. We shall prove that $`xy_i`$ exists for each $`i`$, and that it equals $`e_i(xy)=e_ifx=e_ify`$. By the second lemma, it suffices to show that for each $`i`$ the join
$$f_i=\{g(xx^1y_iy_i^1)gx=gy_i\}$$
exists and equals $`e_if`$. Consider $`g(xx^1y_iy_i^1)=(xx^1e_i)`$ such that $`gx=gy_i`$. The condition $`g(xx^1e_i)`$ implies that $`ge_i`$, and thus $`g=ge_i`$. Hence, since $`y_i=e_iy`$, the condition $`gx=gy_i`$ implies $`gx=ge_iy=gy`$, and thus $`gf`$ because furthermore $`g(xx^1yy^1)`$. Hence, we have both $`ge_i`$ and $`gf`$, i.e., $`ge_if`$, meaning that $`e_if`$ is an upper bound of the set
$$X=\{g(xx^1e_i)gx=gy_i\}.$$
In order to see that it is the least upper bound it suffices to check that $`e_if`$ belongs to $`X`$, which is immediate: first, $`e_ifxx^1`$ because $`fxx^1`$, and thus $`e_if(xx^1e_i)`$; secondly,
$$(e_if)x=(e_ie_if)y=(e_ife_i)y=(e_if)(e_iy)=(e_if)y_i.$$
Hence, $`e_ifX`$, and thus $`e_ifx=xy_i`$. In addition, the join $`_ie_i`$ exists and it equals $`yy^1`$, by \[1, Prop. 17, p. 27\], and thus, using infinite distributivity and the fact that $`fyy^1`$, we obtain
$$xy=fx=yy^1fx=(\underset{i}{}e_i)fx=\underset{i}{}(e_ifx)=\underset{i}{}(xy_i).\text{ }$$
|
warning/0506/gr-qc0506010.html
|
ar5iv
|
text
|
# Kink Stability of Self-Similar Solutions of Scalar Field in 2+1 Gravity
## I Introduction
The studies of non-linearity of the Einstein field equations near the threshold of black hole formation reveal very rich phenomena , which are quite similar to critical phenomena in statistical mechanics and quantum field theory . In particular, by numerically studying the gravitational collapse of a massless scalar field in $`3+1`$-dimensional spherically symmetric spacetimes, Choptuik found that the mass of such formed black holes takes the form ,
$$M_{BH}=C(p)\left(pp^{}\right)^\gamma ,$$
(1.1)
where $`C(p)`$ is a finite constant with $`C(p^{})0`$, and $`p`$ parameterizes a family of initial data in such a way that when $`p>p^{}`$ black holes are formed, and when $`p<p^{}`$ no black holes are formed. It was shown that, in contrast to $`C(p)`$, the exponent $`\gamma `$ is universal to all the families of initial data studied, and was numerically determined as $`\gamma 0.37`$. The solution with $`p=p^{}`$, usually called the critical solution, is found also universal. Choptuik’s studies were soon generalized to other matter fields . From all the work done so far, the collapse in general falls into two different types, depending on whether the black hole mass takes the scaling form (1.1) or not. When it takes the form, the corresponding collapse is called Type $`II`$ collapse, and when it does not it is called Type $`I`$ collapse. In the type $`II`$ collapse, all the critical solutions found so far have either discrete self-similarity (DSS) or homothetic self-similarity (HSS), depending on the matter fields. In the type $`I`$ collapse, the critical solutions have neither DSS nor HSS. For certain matter fields, these two types of collapse can co-exist. A critical solution in both two types has one and only one unstable mode. This now is considered as one of the main criteria for a solution to be critical.
The studies of critical collapse have been mainly numerical so far, and analytical ones are still highly hindered by the complexity of the problem, even after imposing some symmetries. Lately, some progress has been achieved in the studies of critical collapse of a scalar field in an anti-de Sitter background in $`2+1`$-dimensional spacetimes both numerically and analytically . This serves as the first analytical model in critical collapse. In particular, Garfinkle first found a class of exact solutions to Einstein-scalar field equations, denoted by $`S[n]`$, and later Garfinkle and Gundlach (GG) studied their linear perturbations and found that the solution with $`n=2`$ has only one unstable mode . By definition this is a critical solution, and the corresponding exponent $`\gamma `$ in Eq.(1.1) can be read off from the expression
$$\gamma =\frac{1}{|k_1|},$$
(1.2)
from which it was found $`\gamma =4/3`$, where $`k_1`$ denotes the unstable mode. Although the exponent $`\gamma `$ is close to that found numerically by Pretorius and Choptuik , $`\gamma 1.2\pm 0.05`$ (but not to the one of Husain and Olivier, $`\gamma 0.81`$), this solution is different from the numerical critical solution . Using different boundary conditions, Hirschmann, Wang and Wu (HWW) found that the solution with $`n=4`$ has only one unstable mode . As first noted by Garfinkle , this $`n=4`$ solution matches extremely well with the numerical critical solution found by Pretorius and Choptuik . However, the corresponding exponent $`\gamma `$ now is given by $`\gamma =|k_1|^1=4`$, which is significantly different from the numerical ones. The boundary conditions used by HWW are : (a) The perturbations must be free of spacetime singularity on the symmetry axis; (b) They are analytical across the self-similarity horizon, as the background solutions do; (c) No matter field come out of the already formed trapped region . GG considered only Conditions (a) and (b) .
In this paper we shall study another important issue of critical collapse for a scalar field, the kink stability. The kink modes result from the existence of critical characteristic lines (they are also referred to as self-similarity horizons, and sonic lines), along which discontinuities of (higher order) derivatives of some physical quantities can be developed and propagate. The instability is characterized by the divergence of the discontinuity, and the blow-up may imply the formation of shock waves . An example that discontinuities of derivatives can propagate along a sonic line is given by the linear perturbation, $`\delta \phi (\tau ,z)=\phi _1(z)e^{k\tau }`$, of the massless scalar field in $`2+1`$ gravity, which satisfies the following equation ,
$$z\left(1z\right)\phi _{1}^{}{}_{}{}^{\prime \prime }+\frac{1}{2}\left[(1+2k)z(3+2k)\right]\phi _{1}^{}{}_{}{}^{}\frac{1}{2}k\phi _1=f(z),$$
(1.3)
where a prime denotes the ordinary differentiation with respect to the indicated argument, $`f(z)`$ is a smooth function of $`z`$, and $`z=1`$ is the location of the sonic line (cf. Eq.(112) in ). From the above equation we can see that it is possible for $`\phi _1`$ to has discontinuous derivatives only across the line $`z=1`$. In fact, assume that $`\phi _1`$ is continuous across $`z=1`$ (but not its first-order derivative), we can write it in the form
$$\phi _1(z)=\phi _1^+(z)H(z1)+\phi _1^{}(z)\left[1H(z1)\right],$$
(1.4)
where $`H(x)`$ denotes the Heavside (step) function, defined as
$$H(x)=\{\begin{array}{cc}1,\hfill & x>0\text{,}\hfill \\ 0,\hfill & x<0\text{.}\hfill \end{array}$$
(1.5)
Then, we find that
$`\phi _{1}^{}{}_{}{}^{}`$ $`=`$ $`\phi _{1}^{+}{}_{}{}^{}H(z1)+\phi _{1}^{}{}_{}{}^{}\left[1H(z1)\right],`$ (1.6)
$`\phi _{1}^{}{}_{}{}^{\prime \prime }`$ $`=`$ $`\phi _{1}^{+}{}_{}{}^{\prime \prime }H(z1)+\phi _{1}^{}{}_{}{}^{\prime \prime }\left[1H(z1)\right]+\left[\phi _{1}^{}{}_{}{}^{}\right]^{}\delta (z1),`$ (1.7)
where $`\delta (x)`$ denotes the Dirac delta function, and
$$\left[\phi _{1}^{}{}_{}{}^{}\right]^{}\underset{z1^{+0}}{lim}\left(\frac{d\phi _1^+(z)}{dz}\right)\underset{z1^0}{lim}\left(\frac{d\phi _1^{}(z)}{dz}\right).$$
(1.8)
Substituting Eq.(1.6) into Eq.(1.3) and considering the facts
$`H^m(x)=H(x),\left[1H(x)\right]^m=\left[1H(x)\right],`$ (1.9)
$`\left[1H(x)\right]H(x)=0,x\delta (x)=0,`$ (1.10)
where $`m`$ is an integer, one can see that Eq.(1.3) holds also on the horizon $`z=1`$ even when $`\left[\phi _{1}^{}{}_{}{}^{}\right]^{}0`$. This is because $`(1z)\left[\phi _{1}^{}{}_{}{}^{}\right]^{}\delta (z1)=0`$, as long as $`\left[\phi _{1}^{}{}_{}{}^{}\right]^{}`$ is finite. Thus, we have
$$\left[\delta \phi _{,z}\right]^{}=\left[\phi _{1}^{}{}_{}{}^{}\right]^{}e^{k\tau },$$
(1.11)
where $`()_{,z}()/z`$. The above expression shows clearly how the discontinuity of the first derivative of the perturbation $`\delta \phi (\tau ,z)`$ propagate along the sonic line $`z=1`$. When $`Re(k)>0`$ the perturbation grows exponentially as $`\tau \mathrm{}`$, and is said unstable with respect to the kink perturbation. When $`Re(k)<0`$ the perturbation decays exponentially and is said stable.
Note that if the discontinuity happened on other places, say, $`z=z_01`$, clearly Eq.(1.3) would not hold on $`z=z_0`$, because now $`(1z)\left[\phi _{1}^{}{}_{}{}^{}\right]^{}\delta (zz_0)0`$. This explains why the discontinuities are allowed only along the sonic lines. The above analysis also shows that the kink perturbations are different from the ones considered in and , because there it was required that $`\phi _1(z)`$ is analytical across $`z=1`$, that is,
$$\left[\phi _{1}^{}{}_{}{}^{(m)}\right]^{}=0,(m=1,2,\mathrm{})$$
(1.12)
where $`\phi _{1}^{}{}_{}{}^{(m)}`$ denotes the $`m`$-th order derivatives of $`\phi _1`$. Therefore, kink perturbations were excluded in the studies of linear perturbations of and .
Ori and Piran first studied kink stability of self-similar solutions in newtonian gravity , and lately Harada generalized such a study to the relativistic case and found that the critical self-similar solutions of a perfect fluid with the equation of state $`P=k\rho `$ are not stable against kink perturbations for $`k0.89`$, where $`P`$ and $`\rho `$ denote, respectively, the pressure and energy density of the fluid . More recently, Harada and Maeda showed that in four-dimensional spherically symmetric case the self-similar massless scalar solution found lately by Brady et al is also not stable against kink perturbations .
In this paper, we study the kink stability of the scalar field in $`2+1`$ gravity. Instead of assuming that $`\delta \phi (\tau ,z)`$ is $`c^0`$ across the sonic line, as we did in the above example, following Harada , and Harada and Maeda , we shall assume that $`\delta \phi (\tau ,z)`$ is $`c^1`$, that is, $`\delta \phi (\tau ,z)`$ and its first-order derivative with respect to $`z`$ are continuous across the sonic line, but not its second-order derivative. We shall first show that perturbations obtained along the sonic line allow the existence of unstable modes. However, when we consider perturbations outside the sonic line and take the ones obtained along the sonic line as their boundary conditions, we find that these conditions together with the ones on the symmetry axis do not allow any non-trivial perturbations in the regions outside the sonic line. Therefore, the consideration of perturbations in the whole spacetime limits the unstable mode found along the sonic line. Thus, all the self-similar solutions of the massless scalar field are stable against kink perturbations in $`2+1`$ gravity. As a result, the critical solution for the scalar collapse remains critical, even after the kink perturbations are taken into account.
Specifically, the paper is organized as follows: In Sec. II we give a brief review of the self-similar solution, which is needed in the studies of linear perturbations in Sec. III, in which we first consider the linear perturbations of the self-similar solutions along the sonic line, and then the perturbations outside the sonic line. In Sec. IV, we summarize the main results obtained in this paper and then present our concluding remarks.
## II The Einstein-Scalar Field Equations
The general form of metric for a ($`2+1`$)-dimensional spacetime with circular symmetry can be cast in the form,
$$ds^2=2e^{2\sigma (u,v)}dudv+r^2(u,v)d\theta ^2,$$
(2.1)
where $`(u,v)`$ is a pair of null coordinates varying in the range $`(\mathrm{},\mathrm{})`$, and $`\theta `$ is the usual angular coordinate with the hypersurfaces $`\theta =0,\mathrm{\hspace{0.33em}2}\pi `$ being identified. $`\xi _{(\theta )}=_\theta `$ is a Killing vector. It should be noted that the form of the metric is unchanged under the coordinate transformations,
$$u=u(\overline{u}),v=v(\overline{v}).$$
(2.2)
To have circular symmetry, some conditions on the symmetry axis needed to be imposed. In general this is not trivial. As a matter of fact, only when the axis is free of spacetime singularity, do we know how to impose these conditions. Since in this paper we are mainly interested in gravitational collapse, we shall assume that the axis is regular at the beginning of the collapse. In particular, we impose the following conditions:
(i) There must exist a symmetry axis, which can be expressed as
$$X\left|\xi _{(\theta )}^\mu \xi _{(\theta )}^\nu g_{\mu \nu }\right|0,$$
(2.3)
as $`vf(u)`$, where we assumed that the axis is located at $`r(v=f(u),u)=0`$.
(ii) The spacetime near the symmetry axis is locally flat, which can be written as
$$\frac{X_{,\alpha }X_{,\beta }g^{\alpha \beta }}{4X}1,$$
(2.4)
as $`vf(u)`$, where $`()_{,\alpha }()/x^\alpha `$.
The corresponding Einstein-scalar field equations for the metric (2.1) take the form,
$`r_{,uu}2\sigma _{,u}r_{,u}`$ $`=`$ $`8\pi Gr\varphi _{,u}^2,`$ (2.5)
$`r_{,vv}2\sigma _{,v}r_{,v}`$ $`=`$ $`8\pi Gr\varphi _{,v}^2,`$ (2.6)
$`r_{,uv}+2r\sigma _{,uv}`$ $`=`$ $`8\pi Gr\varphi _{,u}\varphi _{,v},`$ (2.7)
$`r_{,uv}`$ $`=`$ $`0,`$ (2.8)
while the equation of motion for the scalar field is given by
$$2\varphi _{,uv}+\frac{1}{r}\left(r_{,u}\varphi _{,v}+r_{,v}\varphi _{,u}\right)=0.$$
(2.9)
To study self-similar solutions, we first introduce the dimensionless variables, $`z`$ and $`\tau `$, via the relations
$$z=\frac{v}{(u)},\tau =\mathrm{ln}\left(\frac{(u)}{u_0}\right),$$
(2.10)
where $`u_0`$ is a dimensional constant with the dimension of length, and the above relations are assumed to be valid only in the region $`v0,u0`$. We will refer to this region as Region $`I`$ \[cf. Fig. 1\].
Self-similar solutions are given by
$$F(\tau ,z)=F_{ss}(z),$$
(2.11)
where $`F\{\sigma ,s,\phi \}`$, and
$`r(u,v)`$ $``$ $`(u)s(\tau ,z),`$ (2.12)
$`\varphi (u,v)`$ $``$ $`c\mathrm{ln}\left|u\right|+\phi (\tau ,z),`$ (2.13)
with $`c`$ being an arbitrary constant. A class of such solutions was first found by Garfinkle , which can be written as
$`\sigma _{ss}(u,v)`$ $`=`$ $`{\displaystyle \frac{1}{2}}\mathrm{ln}\left\{{\displaystyle \frac{\left[v^{1/2}+ϵ(u)^{1/2}\right]^{4\chi }}{(uv)^\chi }}\right\}+\sigma _0^1,`$ (2.14)
$`r_{ss}(u,v)`$ $`=`$ $`(u)v,`$ (2.15)
$`\varphi _{ss}(u,v)`$ $`=`$ $`2c\mathrm{ln}\left|v^{1/2}+ϵ(u)^{1/2}\right|+\varphi _0^1,`$ (2.16)
where $`ϵ=\pm 1`$, $`\sigma _0^1`$ and $`\phi _0^1`$ are integration constants, and $`\chi 8\pi Gc^2`$. As shown in , the hypersurface $`v=0`$ for the solutions with $`1>\chi 1/2`$ represents a sonic line, and the solutions can be extended across the hypersurface, whereby they can be interpreted as representing the gravitational collapse of a scalar field, in which a black hole is finally formed. The extension can be realized by introducing two new coordinates $`\overline{u}`$ and $`\overline{v}`$ via the relations
$$\overline{u}=(u)^{1/2n},\overline{v}=v^{1/2n},$$
(2.17)
where $`n1/[2(1\chi )]1`$. In order to have the extension unique, we require that it be analytical across the hypersurface $`v=0`$, which, in turn, requires $`n`$ to be an integer and satisfy the condition,
$$n=\frac{1}{2(1\chi )}=\{\begin{array}{cc}2l,\hfill & ϵ=1\text{,}\hfill \\ 2l+1,\hfill & ϵ=1\text{,}\hfill \end{array}$$
(2.18)
where $`l`$ is another integer. For the detail, we refer readers to . In these new coordinates, the metric and the massless scalar field are given by
$`ds^2`$ $`=`$ $`2e^{2\overline{\sigma }_{ss}(\overline{u},\overline{v})}d\overline{u}d\overline{v}+r_{ss}^{}{}_{}{}^{2}(\overline{u},\overline{v})d\theta ^2,`$ (2.19)
$`\overline{\sigma }_{ss}(\overline{u},\overline{v})`$ $`=`$ $`\sigma _{ss}(u,v)+{\displaystyle \frac{1}{2}}\mathrm{ln}\left\{4n^2\left(\overline{u}\overline{v}\right)^{2n1}\right\}`$ (2.20)
$`=`$ $`{\displaystyle \frac{1}{2}}\mathrm{ln}\left\{4n^2\left|f(\overline{u},\overline{v})\right|^{4\chi }\right\}+\sigma _0^1,`$ (2.21)
$`r_{ss}(\overline{u},\overline{v})`$ $`=`$ $`(\overline{u})^{2n}\overline{v}^{2n}`$ (2.22)
$`\varphi _{ss}(\overline{u},\overline{v})`$ $`=`$ $`2c\mathrm{ln}\left|f(\overline{u},\overline{v})\right|+\varphi _0^1,`$ (2.23)
where $`f(\overline{u},\overline{v})\overline{v}^n+ϵ(\overline{u})^n`$. Note that the symmetry axis (the vertical line $`r=0`$ in Fig. 1) is located at $`\overline{v}=\overline{u}`$, for which conditions (2.3) requires $`\sigma _0^1=\frac{1}{2}(14\chi )\mathrm{ln}(2)`$. The corresponding Penrose diagram is given by Fig. 1.
## III Linear Perturbations of Self-Similar Solutions: Kink Stability
In this section, we consider the linear perturbations of the self-similar solutions given by Eq.(2.19). For the sake of simplicity, we shall drop all the bars from $`\overline{\sigma },\overline{u}`$ and $`\overline{v}`$, so that the background solutions can be written as,
$`ds^2`$ $`=`$ $`2e^{2\sigma _{ss}(u,v)}dudv+r_{ss}^2(u,v)d\theta ^2,`$ (3.1)
$`\sigma _{ss}(u,v)`$ $`=`$ $`{\displaystyle \frac{1}{2}}\mathrm{ln}\left(4n^2\left|f(u,v)\right|^{4\chi }\right)+\sigma _0^1,`$ (3.2)
$`r_{ss}(u,v)`$ $`=`$ $`(u)^{2n}v^{2n},`$ (3.3)
$`\varphi _{ss}(u,v)`$ $`=`$ $`2c\mathrm{ln}\left|f(u,v)\right|+\varphi _0^1,`$ (3.4)
$`f(u,v)`$ $``$ $`v^n+ϵ(u)^n.`$ (3.5)
Let us first divide the spacetime in Fig. 1 into three different regions, $`\mathrm{\Omega }^\pm `$ and $`\mathrm{\Sigma }`$, defined, respectively, by $`\mathrm{\Omega }^+=\{x^\alpha :u,v0,vu\}`$, $`\mathrm{\Omega }^{}=\{x^\alpha :u0,v0,v|u|\}`$, and $`\mathrm{\Sigma }=\{x^\alpha :v=0\}`$. Then, for any given $`C^1`$ function $`f(u,v)`$, we can write it as
$$f(u,v)=f^+(u,v)\left[1H(v)\right]+f^{}(u,v)H(v),$$
(3.6)
where $`f^\pm `$ denote the functions, defined, respectively, in the regions $`\mathrm{\Omega }^\pm `$. In the present case, we have
$$f^\pm (u,v)=f_{ss}^\pm (u,v)+\delta f^\pm (u,v),$$
(3.7)
where $`f_{ss}\{\sigma _{ss},r_{ss},\varphi _{ss}\}`$ denotes the background solutions given by Eq.(3.1), which are analytical across $`v=0`$,
$$\underset{v0^{}}{lim}\frac{^mf_{ss}^+(u,v)}{v^m}=\underset{v0^+}{lim}\frac{^mf_{ss}^{}(u,v)}{v^m},(m=0,1,2,\mathrm{}).$$
(3.8)
Since $`f(u,v)`$ is $`C^1`$, we must have
$`\underset{v0^{}}{lim}\delta f^+(u,v)=\underset{v0^+}{lim}\delta f^{}(u,v)\delta f_c(u),`$ (3.9)
$`\underset{v0^{}}{lim}\delta f_{,v}^+(u,v)=\underset{v0^+}{lim}\delta f_{,v}^{}(u,v)\delta f_c^{(1)}(u).`$ (3.10)
Then, we find
$`f_{,v}(u,v)`$ $`=`$ $`f_{,v}^+(u,v)\left[1H(v)\right]+f_{,v}^{}(u,v)H(v),`$ (3.11)
$`f_{,vv}(u,v)`$ $`=`$ $`f_{,vv}^+(u,v)\left[1H(v)\right]+f_{,vv}^{}(u,v)H(v).`$ (3.12)
Inserting Eqs.(3.7)-(3.11) into Eqs.(2.5)-(2.9) and considering Eq.(1.9), to the first order of $`\delta f`$, we obtain
$`\delta r_{,uu}2\left(\sigma _{ss,u}\delta r_{,u}+r_{ss,u}\delta \sigma _{,u}\right)=8\pi G\left(2r_{ss}\varphi _{ss,u}\delta \varphi _{,u}+\varphi _{ss,u}^2\delta r\right),`$ (3.13)
$`\delta r_{,vv}2\left(\sigma _{ss,v}\delta r_{,v}+r_{ss,v}\delta \sigma _{,v}\right)=8\pi G\left(2r_{ss}\varphi _{ss,v}\delta \varphi _{,v}+\varphi _{ss,v}^2\delta r\right),`$ (3.14)
$`2\left(r_{ss}\delta \sigma _{,uv}+\sigma _{ss,uv}\delta r\right)=8\pi G\left[r_{ss}\left(\varphi _{ss,u}\delta \varphi _{,v}+\varphi _{ss,v}\delta \varphi _{,u}\right)+\varphi _{ss,u}\varphi _{ss,v}\delta r\right],`$ (3.15)
$`\delta r_{,uv}=0,`$ (3.16)
$`2r_{ss}\delta \varphi _{,uv}+2\varphi _{ss,uv}\delta r+r_{ss,u}\delta \varphi _{,v}+r_{ss,v}\delta \varphi _{,u}+\varphi _{ss,u}\delta r_{,v}+\varphi _{ss,v}\delta r_{,u}=0,`$ (3.17)
where the quantities $`f_{ss}`$ and $`\delta f`$ should be understood as $`f_{ss}^+`$ and $`\delta f^+`$ in $`\mathrm{\Omega }^+`$, and $`f_{ss}^{}`$ and $`\delta f^{}`$ in $`\mathrm{\Omega }^{}`$.
### A Kink Stability
Kink stability is the study of the linear perturbations of Eqs.(3.13) - (3.17) along the sonic line $`v=0`$. To solve these equations for $`\delta f_c(u)`$, following Ori and Piran (See also ), we impose the following conditions: Assume that the perturbations turn on at the moment, say, $`u=u_0`$, then we require
(A) the perturbations initially vanish in the interior,
$$\delta f^{}(u_0,v)=0,v\mathrm{\Omega }^{},$$
(3.18)
(B) the perturbations and their first-order derivatives be continuous everywhere, and in particular across the sonic line,
$$\left[\delta f\right]^{}=0,\left[\delta f_{,v}\right]^{}=0,\left(v=0\right),$$
(3.19)
(C) $`\delta \varphi _{}^{\pm }{}_{,vv}{}^{}`$ and $`\delta \sigma _{}^{\pm }{}_{,vv}{}^{}`$ be discontinuous across the sonic line,
$$\delta \varphi _c^{\prime \prime }(u)\left[\delta \varphi _{,vv}\right]^{}0,\delta \sigma _c^{\prime \prime }(u)\left[\delta \sigma _{,vv}\right]^{}0,\left(v=0\right).$$
(3.20)
From the above we first note that Eq.(3.18) remains true for all $`u>u_0`$. In fact, $`\delta f^{}(u,v)=0`$ are indeed solutions of Eqs.(3.13) - (3.17) in $`\mathrm{\Omega }^{}`$. Then, from Eqs.(3.18) and (3.19) we find
$`\delta f^+(u,0)=0,\delta f_{,v}^+(u,0)=0,`$ (3.21)
$`\delta \varphi _c^{\prime \prime }(u)=\delta \varphi _{,vv}^+(u,0),\delta \sigma _c^{\prime \prime }(u)=\delta \sigma _{,vv}^+(u,0).`$ (3.22)
Taking the limit $`v0^{}`$ in Eqs.(3.13)-(3.17) and considering the above equation we find that
$$\left[\delta r_{,vv}\right]^{}=\delta r_{,vv}^+(u,0)=0.$$
(3.23)
On the other hand, taking derivatives of Eqs.(3.17) and (3.15) with respect to $`v`$, and then taking the limit $`v0^{}`$, we obtain
$`2r_{ss}\left(\delta \varphi _c^{\prime \prime }\right)_{,u}+r_{ss,u}\delta \varphi _c^{\prime \prime }=0,`$ (3.24)
$`\left(\delta \sigma _c^{\prime \prime }\right)_{,u}=4\pi G\varphi _{ss,u}\delta \varphi _c^{\prime \prime },`$ (3.25)
along the sonic line $`v=0`$. Substituting Eq.(3.1) into the above equations and then integrating them, we obtain
$`\delta \varphi _c^{\prime \prime }(u)`$ $`=`$ $`{\displaystyle \frac{A}{(u)^n}}={\displaystyle \frac{A}{u_0^{1/2}}}e^{\tau /2},`$ (3.26)
$`\delta \sigma _c^{\prime \prime }(u)`$ $`=`$ $`{\displaystyle \frac{8\pi GcA}{(u)^n}}={\displaystyle \frac{8\pi GcA}{u_0^{1/2}}}e^{\tau /2},`$ (3.27)
where $`A`$ is an integration constant. Since $`n1`$, from the above expressions we can see that both $`\delta \varphi _c^{\prime \prime }(u)`$ and $`\delta \sigma _c^{\prime \prime }(u)`$ diverge as $`u0^{}(\text{or}\tau \mathrm{})`$, or in other words, the self-similar solutions are not stable against the kink perturbations.
It should be noted that $`\delta f^+(u,v)`$ cannot be zero identically in $`\mathrm{\Omega }^+`$, because we already have $`\delta f^{}(u,v)=0`$ in $`\mathrm{\Omega }^{}`$ and
$$\delta f_c^{\prime \prime }(u)=\delta f_{,vv}^+(u,0^{})0.$$
(3.28)
Then, a natural question rises: Do the perturbations given by Eq.(3.26) match to the ones in region $`\mathrm{\Omega }^+`$? To answer this question, in the next subsection we shall consider the linear perturbations of Eqs.(3.13)-(3.17) in region $`\mathrm{\Omega }^+`$, by considering Eqs.(3.21), (3.23) and (3.26) as their boundary conditions at $`v=0`$.
### B Linear Perturbations in $`\mathrm{\Omega }^+`$
To study the linear perturbations in $`\mathrm{\Omega }^+`$, it is found convenient to use the dimensionless variables $`\tau `$ and $`z`$, defined by Eq.(2.10). However, they are valid only in region $`\mathrm{\Omega }^{}`$. In region $`\mathrm{\Omega }^+`$ we define them as
$$\stackrel{~}{\tau }=\mathrm{ln}\left(\frac{\stackrel{~}{u}}{u_0}\right),\stackrel{~}{z}=\frac{\stackrel{~}{v}}{\stackrel{~}{u}},$$
(3.29)
where $`\stackrel{~}{u},\stackrel{~}{v}0`$ in $`\mathrm{\Omega }^+`$, and
$$\stackrel{~}{u}(u)^{2n},\stackrel{~}{v}(v)^{2n}.$$
(3.30)
The null coordinates $`u`$ and $`v`$ in Eq.(3.30) should be understood as the ones, $`\overline{u}`$ and $`\overline{v}`$, defined by Eq.(2.17). In terms of $`\stackrel{~}{\tau }`$ and $`\stackrel{~}{z}`$, the background solutions (3.1) in $`\mathrm{\Omega }^+`$ can be written in the form,
$`s_0(\stackrel{~}{z})`$ $`=`$ $`1\stackrel{~}{z},`$ (3.31)
$`\sigma _0(\stackrel{~}{z})`$ $`=`$ $`2\chi \mathrm{ln}\left(\stackrel{~}{z}^{1/4}+\stackrel{~}{z}^{1/4}\right)+\sigma _0^1,`$ (3.32)
$`\phi _0(\stackrel{~}{z})`$ $`=`$ $`2c\mathrm{ln}\left(1+\stackrel{~}{z}^{1/2}\right)+\phi _0^1,`$ (3.33)
with
$`\sigma _{ss}(u,v)`$ $`=`$ $`\sigma _0(\stackrel{~}{z})+{\displaystyle \frac{1}{2}}\mathrm{ln}\left[4n^2(uv)^{2n1}\right],`$ (3.34)
$`r_{ss}(u,v)`$ $`=`$ $`(u)^{2n}s_0(\stackrel{~}{z}),`$ (3.35)
$`\varphi _{ss}(u,v)`$ $`=`$ $`\phi _0(\stackrel{~}{z})+c\mathrm{ln}\left[(u)^{2n}\right].`$ (3.36)
Again, $`u`$ and $`v`$ in Eq.(3.34) should be understood as $`\overline{u}`$ and $`\overline{v}`$ defined by Eq.(2.17). For detail, we refer readers to Eqs.(57)-(59) in . Without causing any confusions, in the following we shall drop the tildes from $`\stackrel{~}{\tau }`$ and $`\stackrel{~}{z}`$. Then, writing the perturbations as
$`\delta r`$ $`=`$ $`(\stackrel{~}{u})s_1(z)e^{k\tau },`$ (3.37)
$`\delta \sigma `$ $`=`$ $`\sigma _1(z)e^{k\tau },`$ (3.38)
$`\delta \varphi `$ $`=`$ $`\phi _1(z)e^{k\tau },`$ (3.39)
it can be shown that the linearized perturbations given by Eqs.(3.13)-(3.17) reduce exactly to the ones of (67)-(71) of , the general solutions of which are Eqs.(110)-(118) for $`k=1`$, and Eqs.(120)-(125) for $`k1`$, given in . In particular, $`s_1(z)`$ is given by
$$s_1(z)=\{\begin{array}{cc}\beta \mathrm{ln}(z)+s_1^0,\hfill & k=1\text{,}\hfill \\ \beta z^{1k}+s_1^0,\hfill & k1\text{,}\hfill \end{array}$$
(3.40)
where $`\beta `$ and $`s_1^0`$ are the integration constants.
However, since here we consider the kink stability, the boundary conditions are different from the ones used in . In particular, in it was required that the perturbations be analytical across the surface $`v=0`$, while in the present case these conditions should be replaced by Eqs.(3.21), (3.23) and (3.26), which can be written as
$`k`$ $`=`$ $`{\displaystyle \frac{1}{2}},`$ (3.41)
$`s_1(z)`$ $``$ $`O\left(z^3\right),`$ (3.42)
$`\sigma _1(z)`$ $``$ $`{\displaystyle \frac{4\pi GcA}{2u_0^{1/2}}}z^2+O\left(z^3\right),`$ (3.43)
$`\phi _1(z)`$ $``$ $`{\displaystyle \frac{A}{2u_0^{1/2}}}z^2+O\left(z^3\right),`$ (3.44)
as $`z0`$. Thus, the instability of the perturbations along the hypersurface $`v=0`$ found in the last subsection is due to a single mode, $`k=1/2`$. In addition to the above conditions, we also need to impose some conditions on the symmetry axis $`r=0`$, so that the local-flatness conditions (2.3) and (2.4) are satisfied. In terms of $`f_1(z)`$, these conditions are exactly the ones given by Eq.(105) in ,
$`s_1(z)|_{z=1}`$ $`=`$ $`0,`$ (3.45)
$`\sigma _1(z)|_{z=1}`$ $``$ $`\mathrm{finite}`$ (3.46)
$`\left\{(1z){\displaystyle \frac{d\phi _1(z)}{dz}}2k\phi _1(z)\right\}|_{z=1}`$ $``$ $`\mathrm{finite},(r=0).`$ (3.47)
From Eqs.(3.40) and (3.45) we find that $`s_1^0=\beta `$, while Eq.(3.41) requires $`\beta =0`$, for which the solutions of $`\sigma _1(z)`$ and $`\phi _1(z)`$ with $`k=1/2`$ are given by
$`\sigma _1(z)`$ $`=`$ $`{\displaystyle \frac{2\chi }{c}}(1z^{1/2})\left[z^{1/2}(1+z^{1/2}){\displaystyle \frac{d\phi _1(z)}{dz}}+{\displaystyle \frac{1}{2}}\phi _1\right],`$ (3.48)
$`\phi _1(z)`$ $`=`$ $`c_1F({\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}};1;z)+c_2F({\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}};1;1z),`$ (3.49)
where $`c_1`$ and $`c_2`$ are two arbitrary constants, and $`F(a,b;c;z)`$ denotes the ordinary hypergeometric function with $`F(a,b;c;0)=1`$. From the expression ,
$$F(\frac{1}{2},\frac{1}{2};1;z)=\frac{1}{\pi ^2}\underset{n=0}{\overset{\mathrm{}}{}}\frac{2\mathrm{\Gamma }^2\left(\frac{1}{2}+n\right)}{\left(n!\right)^2}\left\{\psi (n+1)\psi \left(n+\frac{1}{2}\right)\frac{1}{2}\mathrm{ln}(1z)\right\}(1z)^n,$$
(3.50)
we find
$$F(\frac{1}{2},\frac{1}{2};1;z)\frac{1}{\pi }\mathrm{ln}(1z),$$
(3.51)
as $`z1`$. Then, Eq.(3.45) requires $`c_1=0`$. On the other hand, from Eq.(3.50) we also find
$$F(\frac{1}{2},\frac{1}{2};1;1z)\frac{1}{\pi }\mathrm{ln}(z),$$
(3.52)
as $`z0`$. Thus, the conditions of Eq.(3.41) yield $`c_2=0`$. In review of all the above, we find that the boundary conditions (3.41) and (3.45) require
$$s_1(z)=\sigma _1(z)=\phi _1(z)=0.$$
(3.53)
That is, non-trivial perturbations in $`\mathrm{\Omega }^+`$ are not allowed by the boundary conditions (3.41) and (3.45). Then, we must have $`\delta f_c^{\prime \prime }(u)=0`$. In other words, the consideration of the perturbations in $`\mathrm{\Omega }^+`$ limits the unstable mode of the perturbations along the sonic line $`v=0`$.
## IV Conclusions
In this paper, we have studied the kink stability of the self-similar solutions of a massless scalar field in $`2+1`$ gravity, and found that perturbations along the sonic line (self-similar horizon) indeed allow the existence of an unstable mode.
In the study of kink stability, it is assumed that the spacetime inside the sonic line is not perturbed, that is, $`\delta f^{}(u,v)=0`$ identically . Then, $`\delta f^+(u,v)`$ must be non-zero outside the sonic line, in order to have non-vanishing perturbations along the sonic line. However, the perturbations outside the sonic line cannot be arbitrary. In particular, they have to match to the ones along the sonic line. In addition, they need also satisfy some physical/geometrical conditions, such as, the local-flatness conditions on the symmetry axis. A natural question now is: After considering all these, does the spectrum of the perturbations obtained along the sonic line still remain the same?
To answer this question, in Sec. III we have studied the perturbations outside the sonic line, by taking the ones obtained along the sonic line as their boundary conditions. We have shown explicitly that these conditions, together with the ones on the symmetry axis, indeed alter the spectrum of the perturbations along the sonic line, and in particular, they limit all the unstable modes. Thus, all the self-similar solutions of the massless scalar field in $`2+1`$ gravity is stable against kink perturbations. As a result, the critical solution for the scalar collapse remains critical even after the kink perturbations are taken into account.
Finally, we note that in the newtonian gravity the spectrum of the perturbations along the sonic line remains the same, even after the perturbations outside the sonic line are taken into account . It would be very interesting to see if this is still the case in four-dimensional spacetimes in the framework of Einstein’s theory of gravity.
## Acknowledgments
One of the authors (AW) would like to express his gratitude to M.W. Choptuik, D. Garfinkle, and T. Harada for their valuable discussions and suggestions. Part of the work was done when one of the authors (YW) was visiting the Department of Mathematics, Baylor University. She would like to express her gratitude to the Department for hospitality. The authors also thank the Astrophysics Center, Zhejiang University of Technology for hospitality.
|
warning/0506/astro-ph0506383.html
|
ar5iv
|
text
|
# The Excess Far-Infrared Emission of AGN in the Local Universe
## 1 Introduction
In the mid 1980’s, spectroscopic follow up of galaxies observed by the IRAS satellite revealed that infrared-luminous galaxies consist of a mixture of star-forming galaxies and galaxies with an active nucleus (AGN). De Grijp et al. (1985) found that fraction of AGN could be maximized by selecting IRAS sources with relatively warm 25 to 60 micron colours and they then used this criterion to construct a catalog of 80 Seyfert 1 and 141 Seyfert 2 galaxies (De Grijp et al. 1992).
A considerable amount of effort has been devoted to understanding the physical mechanisms responsible for the infrared emission in AGN. The current paradigm asserts that type 1 and type 2 Seyfert galaxies are drawn from the same parent population. In type 1 Seyfert galaxies the subparsec-scale continuum source is viewed directly, but in type 2 Seyferts this source is blocked from view by a structure (commonly referred to as the “torus”) with a size of between 1 and 10 parsec, which is optically thick to radiation from X-ray to near-IR wavelengths. The size of the obscuring region, together with energy conservation arguments, then imply that most of the radiation absorbed by this structure will be re-radiated in the mid-IR as thermal emission from dust (Pier & Krolik 1992; Granato & Danese 1994).
In recent years it has become clear that star formation and AGN activity frequently occur together in galaxies (e.g. Cid Fernandes et al. 2001; Kauffmann et al. 2003c; Heckman et al. 2004). As a result, it is natural to suppose that a substantial fraction of IR emission from AGN may be produced by star formation. It is commonly believed that star formation is likely to dominate the emission at long wavelengths ($`>60\mu `$m) and that emission from the torus prevails in the mid-IR. Rowan-Robinson & Crawford (1989) modelled the far-IR spectra of IRAS galaxies using three components: a “disc component” due to the interstellar dust illuminated by the galaxy’s starlight; a “starburst component” arising from hot stars in optically thick dust clouds, and a “Seyfert component” due to a power-law continuum source within the torus. In most cases ($``$ 61$`\%`$), the IRAS spectra are well fitted by a combination of disc and starburst components. Only $``$ 24$`\%`$ of the galaxies in the samples required the Seyfert component. Nevertheless, as discussed by Silva, Granato & Maiolino (2004), the intrinsic far-IR properties of AGN remain subject to strong uncertainties (only large-aperture data from ISO or IRAS are available at $`\lambda >20\mu `$m) and there is also substantial freedom in the dusty torus models at these wavelengths.
The relative importance of dust heating by star formation and by AGN has also been a topic of considerable controversy in the study of ultraluminous IRAS galaxies (ULIRGs), which are the most powerful galaxies detected by IRAS with IR luminosities in excess of $`10^{12}L_{}`$. A significant fraction of ULIRGs exhibit nuclear optical emission line spectra characteristic of Seyfert galaxies (Sanders et al. 1988; Armus et al. 1990), but the infrared, millimeter and radio characteristics of these systems are very similar to those of ordinary starbursts (Rieke et al. 1985; Rowan-Robinson & Crawford 1989; Condon et al. 1991). Genzel et al. (1998) used mid-infrared spectra from the ISO satellite to demonstrate that 70-80% of the ULIRGS in their sample were predominantly powered by star formation and 20-30% by a central AGN. These conclusions were based on an analysis of the ratio of high- to low-excitation mid-IR emission lines as well as the strength of the 7.7 $`\mu `$m PAH feature in these systems.
The same question is relevant when attempting to convert from the luminosity function of sub-millimeter sources observed at high redshifts to estimates of the total star formation rate density (e.g. Blain et al. 1999; Rowan-Robinson 2001). The contribution from AGN has been investigated by looking for overlap between sub-mm galaxies detected by SCUBA and hard X-ray sources found by Chandra. Only around 10% of the sub-mm sources are found to have an X-ray counterpart (Severgnini et al. 2000; Barger et al. 2001). More recently Silva et al. (2004) have used a combination of X-ray data on the evolution of the AGN luminosity function and spectral energy distributions drawn from their models to show that around 95% of the total IR background is likely produced by star formation.
In this paper, we study the far-infrared properties of AGN in the local Universe. As a result of recent large redshift surveys such as the Sloan Digital Sky Survey (SDSS), it is possible to compile unprecedentedly large samples of galaxies with both high quality optical spectra and IRAS fluxes. It then becomes possible to adopt a purely statistical approach to understanding the origin of the IR emission in AGN. In this paper, we analyze a sample of over a thousand galaxies drawn from the SDSS Data Release 2 (DR2) with IRAS detections. Optical emission line ratios are used to classify the galaxies into AGN and “normal” galaxies. We then create subsamples of normal galaxies and AGN that have been carefully matched in terms of key physical properties such as stellar mass, redshift, galaxy structural parameters and mean stellar age, and we quantify whether there are systematic differences between the mean IR luminosities of the galaxies in the matched subsamples.
## 2 The galaxy samples
### 2.1 The SDSS Spectroscopic Sample
The Sloan Digital Sky Survey (York et al. 2000; Stoughton et al. 2002, and references therein) is an optical imaging (u,g,r,i,z bands) and spectroscopic survey of about a quarter of the extragalactic sky, being carried out at the Apache Point Observatory. The spectroscopic sample considered in this paper is a sample of about 212,000 objects with magnitudes $`14.5<r<17.77`$, spectroscopically confirmed to be galaxies. This sample of galaxies is described by Brinchmann et al. (2004) . The galaxies have a median redshift of $`z0.1`$.
The SDSS spectra cover an observed wavelength range of 3800 to 9200Å, at an instrumental velocity resolution of about 65km s<sup>-1</sup>. The spectra are obtained through fibres of about 3-arcsecond diameter, which corresponds to 5.7 kpc at a redshift of 0.1; at this redshift the spectra therefore represent a large proportion (up to 50%) of the total galaxy light, whilst for the very lowest redshift objects they are more dominated by the nuclear emission.
As described by Brinchmann et al. (2004), many properties of these galaxies have been parameterised, with the derived catalogues of parameters being publically available on the web. Derived parameters include: fundamental galaxy parameters such as total stellar masses, sizes, surface mass densities, concentration indices, mass-to-light ratios, 4000Å break strengths, dust attenuation measurements, and H$`\delta `$ absorption measurements (Kauffmann et al. 2003a) ; accurate emission line fluxes, after subtraction of the modelled stellar continuum to account for underlying stellar absorption features (Kauffmann et al. 2003a; Tremonti et al. 2004); galaxy metallicities (Tremonti et al. 2004); star formation rates (Brinchmann et al. 2004); parameters measuring optical AGN activity, such as emission line ratios, and galaxy velocity dispersions (hence black hole mass estimates; Kauffmann et al. 2003c, Heckman et al. 2004). The sample analysed in this paper includes star-forming galaxies and those AGN in which non-stellar continuum light from the nucleus has a negligible effect on the physical parameters derived for the host galaxy (see Kauffmann et al. 2003c for more detailed discussion). In the rest of the paper we use the term “AGN” to refer to these objects. <sup>1</sup><sup>1</sup>1Note that this sample explicitly excludes those objects classified by the SDSS spectroscopic pipeline as QSOs. In such cases the SDSS spectrum is dominated by light from the AGN.
### 2.2 Cross-identification with the IRAS catalogues
We cross-identified the SDSS galaxies with the IRAS Faint-Sources Catalogue (FSC) using the web search engine GATOR (available at http://irsa.ipac.caltech.edu/applications/Gator). We initially assumed a search radius of 1 arcmin in order to account for the IRAS positional uncertainties. Initially, no constraint on the IRAS flux density or the quality of the IRAS flux detections was applied and the query returned a total of 5765 matches. Because of the large search radius, a substantial number of IRAS sources (1205 or 21%) were matched to more than one SDSS galaxy. These sources were eliminated from our catalogue in order to maximize the reliability of our sample. Our catalogue is thus incomplete and somewhat biased against galaxies in high density regions, but this is not particularly important for the applications discussed in this paper.
We restricted our analysis to IRAS sources with reliable flux densities \[with flux quality from moderate (2) to good (3), which means that they have been detected more than once in the repeat scans\]. As discussed in detail in Kauffmann et al. (2003c), we use the \[NII\]/H$`\alpha `$ versus \[OIII\]/H$`\beta `$ emission line ratio diagnostic diagram (Baldwin, Phillips & Terlevich 1981) to classify our galaxies into AGN and normal galaxies. In order for a galaxy to be placed on the BPT diagram, the four emission lines \[NII\],\[OIII\],H$`\alpha `$ and H$`\beta `$ must be detected with S/N$`>3`$. Galaxies in which these four lines are not detected with sufficient signal-to-noise are classified as “normal”. The normal galaxy sample may thus contain some AGN with emission lines that are too weak to classify. We are not concerned with this here, because the aim of this paper is to characterize the infrared emission from more powerful Seyfert galaxies. Figure 1 shows the distribution in the BPT diagram of the emission-line galaxies in our sample with 60 and 100 $`\mu `$m IRAS detections. Red crosses indicate galaxies that are classified as AGN. As can be seen, most of the AGN in our sample lie in the region of the diagram occupied by Seyfert galaxies or by “composite” systems in which there is both an active nucleus and ongoing star formation (see Kauffmann et al. 2003c for a more detailed discussion of how the general population of AGN in the SDSS populate the BPT diagram). There are almost no IRAS-selected AGN in the region of the diagram occupied by LINERs.
The infrared luminosities of the galaxies in our sample were computed using the formulae given in Helou et al. (1988) and Sanders & Mirabel (1996):
FIR = 1.26 $`\times `$ \[F(60$`\mu `$m) $`+`$ F(100$`\mu `$m)\] W m<sup>-2</sup>
F(60$`\mu `$m) = 2.58 $`\times `$ 10$`{}_{}{}^{14}f_{\nu }^{}`$(60$`\mu `$m) W m<sup>-2</sup>
F(100$`\mu `$m) = 1.00 $`\times `$ 10$`{}_{}{}^{14}f_{\nu }^{}`$(100$`\mu `$m) W m<sup>-2</sup>
L<sub>FIR</sub> = 4$`\pi `$D$`{}_{L}{}^{}{}_{}{}^{2}`$FIR
L<sub>100</sub> = 4$`\pi `$D$`{}_{L}{}^{}{}_{}{}^{2}`$F(100$`\mu `$m)
L<sub>60</sub> = 4$`\pi `$D$`{}_{L}{}^{}{}_{}{}^{2}`$F(60$`\mu `$m)
where $`f_\nu `$(60$`\mu `$m) and $`f_\nu `$(100$`\mu `$m) are the IRAS flux densities and we have adopted a cosmolgy with H<sub>o</sub> = 70 km s<sup>-1</sup> Mpc<sup>-1</sup>, $`\mathrm{\Omega }=0.3`$ and $`\mathrm{\Lambda }=0.7`$
The left panel of Figure 2 shows the FIR luminosities (in solar units) plotted as a function of redshift for all the galaxies with reliable flux measurements at both 60 and 100 $`\mu `$m. The IRAS flux limit means that the IR luminosities of the objects in our sample increase from L<sub>FIR</sub> $`10^{10}L_{}`$ at $`z`$ = 0.05 to $`10^{11}L_{}`$ at $`z=0.10.2`$. This is true for both normal galaxies and for AGN. In the right panel of Fig. 2, we plot the derived FIR luminosity as a function of the distance in arcseconds between the position of the IRAS source and the position of the matching SDSS galaxy.
#### 2.2.1 Further checks on the reliability of the sample
In our sample, the signal-to-noise of the detected sources ranges from 3.5 up to 50. Follow-up of IRAS FSC sources have shown that at $`S/N<8`$, a significant fraction (more than 20%) of the sources turn out to be false detections. Because of the large matching radius that is employed, one might worry that some of these false sources may contaminate our sample. Another source of potential contamination is cirrus emission from interstellar dust in our own Galaxy. In the IRAS catalogue, this is parametrized by the “cirrus flag”, which indicates the number of 100 $`\mu `$m sources detected within a 30 arcmin radius of the FSC source.
Figure 3 shows the distribution of positional offsets (in arcseconds) between the IRAS FSC source and the matched SDSS galaxy for different cuts in S/N and in cirrus flag. As the S/N of the sources increases there is a slight shift towards smaller offsets, indicating that some false identifications have been eliminated. The shifts are quite small, however, indicating that the majority of low S/N objects are in fact real. Different cuts in the cirrus flag have no effect on the distribution of offsets.
We have generated catalogues of randomly distributed sources in the area of sky covered by DR2 and we use these to evaluate the likelihood that a false detection will be matched with an SDSS galaxy from the main spectroscopic sample. Results are shown in Figure 4 as a function of the matching radius. Since, within the footprint covered by DR2, the FSC contains 4560 sources associated with galaxies, Figure 4 indicates that about 288 false matches occur for a matching radius of 60 arcseconds. At a 30 arcsecond matching radius, the number of randomly matched galaxies drops to about 46, clearly a more acceptable level of contamination. From Figure 3 we see that a cut at at a positional offset of 30 arcseconds eliminates only around a quarter of the galaxies in our sample. We conclude that a cut in positional offset is a more efficient way to eliminate contaminating sources than a cut in signal-to-noise.
Another way to evaluate the reliability of the sample is to study the scatter in the relations between properties of galaxies derived from the SDSS spectra and those derived from the IRAS fluxes. In Figure 5 we study the effect of different cuts in positional offset and in redshift on the relation between the 4000 Å break strength D<sub>n</sub>(4000) and the “normalized” FIR luminosity $`L_{\mathrm{FIR}}/M_{}`$. Both quantities are a measure of the age of the stellar population in the galaxy and will be used extensively in the analysis that follows.
In the top panels of Fig. 5, we hold the positional offset cut constant at R $`<`$ 30 arcseconds and we show what happens if different cuts in redshift are imposed on the sample. At high redshifts , the sample contains only extremely IR-luminous objects. Fig. 1 shows there are a handful of galaxies with IR luminosities in excess of $`10^{12}L_{}`$ at $`z>0.15`$. These are the ULIRGs, which have been traditionally considered as a separate class of galaxy in their own right. Since there are very few of these objects in our sample, we exclude them by imposing a cut at $`z=0.15`$.
At low redshifts the SDSS spectra, which are obtained using 3 arcsecond diameter fibres, sample only the inner regions of the galaxies and aperture effects may influence our analysis. Kewley, Jansen & Geller (2005) have recently carried out a detailed study of aperture effects on the spectra of galaxies using a sample of 101 nearby galaxies with both global and nuclear spectra. They conclude that so long as the fibre captures more than $``$ 20% of the total light, the differences between physical parameters derived using nuclear spectra and those derived using global spectra are modest. For SDSS, Kewley et al. recommend that redshift cut $`z>0.04`$ be imposed. We choose a somewhat more conservative cut ($`z>0.06`$). The top panels in Fig. 5 show that this cut eliminates a population of “normal” galaxies with high values of $`L_{\mathrm{FIR}}/M_{}`$. These are primarily low mass galaxies ($`<10^{10}M_{}`$) that are currently experiencing a strong burst of star formation. There are very few AGN that have stellar masses lower than $`10^{10}M_{}`$ (Kauffmann et al. 2003c), so eliminating these objects from our sample will not affect the analysis presented in this paper.
In the bottom panels, we fix the redshift interval at $`0.06<z<0.15`$ and we investigate what happens if the cut in positional offset is allowed to vary. As can be seen, if matches with positional offsets as large as 60 arcseconds are accepted, there is a “tail” of outlying galaxies with large 4000 Å break strengths (indicative of an old stellar population), but with high normalized FIR luminosity. These outliers largely disappear when the offset is restricted to be less than 30 arcseconds.
To summarize, the final cuts we impose on our sample are the following:
(i) Only IRAS sources with flux quality flag $``$ 2 are retained.
(ii) The positional offset between the IRAS source and the SDSS galaxy must be less than 30”.
(iii) All sources that have more than one SDSS match within a matching radius of 30” are eliminated.
(iv) All galaxies with $`z<0.06`$ and $`z>0.15`$ are also excluded.
We created two catalogues: one with flux measurements at 60 $`\mu `$m only (1090 galaxies of which 553 are AGN) and a second with flux measurements at both 60 and 100 $`\mu `$m (526 objects of which 284 are AGN). All sources in the 60 $`\mu `$m catalogue have an IRAS flux quality of 3, while all galaxies in the 100 $`\mu `$m are characterised by a quality of 2.
## 3 Properties of the IRAS-selected AGN
In Figure 6 we show the distributions of some of the basic properties of the AGN and the normal star-forming galaxies in our sample, such as their redshifts, IR luminosities, stellar masses, concentrations, stellar surface densities and 4000 Å break strengths. Unless specified otherwise, we show results for the catalogue with reliable flux measurements at both 60 and 100 $`\mu `$m. The histograms representing the normal galaxies and the AGN have been shaded in black and in red, respectively.
The main result seen in Figure 6 is that the IRAS-selected AGN have larger stellar masses than the “‘normal” galaxies in the sample. As discussed by Kauffmann et al. (2003c), the AGN fraction among normal galaxies falls off very steeply at stellar masses less than a few $`\times 10^{10}M_{}`$ and this is apparent in the third panel of Figure 6. As discussed in Kauffmann et al. (2003b), more massive galaxies have higher concentrations and surface densities and their stellar populations are also older. It is therefore not surprising that Figure 6 shows that the AGN are biased to higher values of $`C`$, $`\mu _{}`$ and D<sub>n</sub>(4000) when compared to the star-forming galaxies.
In Figure 7, we plot the “normalized” IR-luminosity $`L_{FIR}/M_{}`$ as a function of stellar mass, concentration, stellar surface mass density and 4000 Å break strength. Once again solid black points indicate normal galaxies and red crosses AGN. It should be noted that the strong correlation between $`L_{FIR}/M_{}`$ and stellar mass and surface density is a selection effect caused by our redshift cut and by the IRAS flux detection limit. However, it possible to conclude from this plot that the largest normalized IR-luminosities are obtained for galaxies with the lowest masses, concentrations and surface densities. There are few AGN in this region of parameter space. At a fixed value of $`M_{}`$, $`C`$ or $`\mu _{}`$, however, the differences between AGN and non-AGN are much more subtle. There does not appear to be a significant difference between the normalized IR-luminosities of AGN and non-AGN at fixed stellar mass or concentration, but AGN do appear to be offset to slightly higher values of $`L_{\mathrm{FIR}}/M_{}`$ at a fixed value of D<sub>n</sub>(4000).
We now divide our samples of AGN and normal star-forming galaxies (non-AGN) into bins of stellar mass with a width of $`\mathrm{\Delta }`$Log<sub>10</sub>(M) = 0.2. We only consider bins that contain at least 10 AGN and 10 normal star-forming galaxies. We compute the mean Log<sub>10</sub>(L<sub>FIR</sub>/M) of the AGN and the normal star-forming galaxies in each bin and we define the “AGN excess” $`\mathrm{\Delta }`$Log<sub>10</sub>(L<sub>FIR</sub>/M) as
$`<`$Log<sub>10</sub>(L<sub>FIR</sub>/M)$`>_{AGN}`$ \- $`<`$Log<sub>10</sub>(L<sub>FIR</sub>/M)$`>_{nonAGN}`$
To analyze whether our results depend on wavelength, we have also calculated $`\mathrm{\Delta }`$Log<sub>10</sub>(L<sub>100</sub>/M) and $`\mathrm{\Delta }`$Log<sub>10</sub>(L<sub>60</sub>/M), which use only the 100 or 60 $`\mu `$m fluxes rather than the combination of the two quantities. Our results are shown in Figure 8. The error bars have been computed using standard bootstrap resampling techniques. As can be seen, at a fixed stellar mass there is very little difference in normalized IR-luminosity between AGN and normal star-forming galaxies at any wavelength.
In the next step of the analysis, we divide our sample into bi-dimensional bins in stellar mass and D<sub>n</sub>(4000). Once again, we only include bins with more than 10 AGN and normal star-forming galaxies in our analysis. The adopted bin sizes are 0.5 dex in $`\mathrm{log}M_{}`$ and 0.1 in D<sub>n</sub>(4000). The left hand panels of Figure 9 show histograms of the distribution of $`\mathrm{\Delta }`$Log<sub>10</sub>(L<sub>FIR</sub>/M), $`\mathrm{\Delta }`$Log<sub>10</sub>(L<sub>60</sub>/M), and $`\mathrm{\Delta }`$Log<sub>10</sub>(L<sub>100</sub>/M) for the bins with sufficient galaxies and AGN to perform the comparison. In the right hand panels, the same quantities are plotted as a function of the value of D<sub>n</sub>(4000) at the center of each bin. We conclude that if galaxies and AGN are matched in both stellar mass and 4000 Å break strength, AGN tend to be brighter than star-forming galaxies by an amount which is larger at 60 $`\mu `$m than at 100 $`\mu `$m. However, this infrared excess is significant at a 2$`\sigma `$ level at 60 $`\mu `$m and only at 1$`\sigma `$ level at 100 $`\mu `$m and in L<sub>FIR</sub>. The reason the excess is not seen in Figure 8 is because AGN are found in galaxies with larger 4000 Å break strengths than star-forming galaxies of the same mass.
## 4 Matched Pair Analysis
The differences between Figures 8 and 9 teach us that in order to compare the far-IR properties of AGN and normal star-forming galaxies in an unbiased way, it is important to match the properties of the host galaxies of these two kinds of systems as closely as possible. The relatively small number of galaxies in our sample means that it is not feasible to bin in more than two dimensions. Instead, we choose to create a sample of galaxy-AGN pairs that are closely matched in stellar mass $`M_{}`$, D<sub>n</sub>(4000), concentration index $`C`$, stellar surface mass density $`\mu _{}`$ (hence physical size) and redshift $`z`$. In order to maximize the number of pairs, we first analyze the catalogue with reliable flux measurements at 60 $`\mu `$m. We accept pairs if $`\mathrm{\Delta }\mathrm{log}M_{}<0.25`$, $`\mathrm{\Delta }`$D<sub>n</sub>(4000)$`<0.04`$, $`\mathrm{\Delta }C<0.1`$, $`\mathrm{\Delta }\mathrm{log}\mu _{}<0.25`$ and $`\mathrm{\Delta }z<`$ 0.03. This leaves us with a sample of 254 unique galaxy-AGN pairs.
In Figure 10, we plot $`\mathrm{\Delta }\mathrm{log}L_{60}/M_{}`$ (the difference between the normalized 60 $`\mu `$m luminosity of the AGN and the matched galaxy ) for each pair as a function of a number of parameters describing the AGN. The red points on the plot show the average value of $`\mathrm{\Delta }\mathrm{log}L_{60}/M_{}`$ calculated in bins of each parameter (the red point is positioned at the center of the bin). The close pair analysis indicates that there is a 0.2 dex excess in $`\mathrm{\Delta }\mathrm{log}L_{60}/M_{}`$ for the AGN relative to the normal star-forming galaxies. The excess does not depend on redshift (showing that is not caused by aperture effects) or on structural properties such as galaxy concentration or stellar surface mass density. It does appear to be larger for lower mass AGN with smaller 4000 Å breaks and for more powerful AGN with larger extinction-corrected \[OIII\] luminosities.
In order to assess the error in the measured far-IR excess of AGN, we generated 5 different galaxy/AGN pair samples by starting the search for pairs from different points in the catalogue. In addition, we created 300 bootstrap resamplings of each of the 5 pair samples in order to assess whether our results were sensitive to the presence of a few outliers in the distribution. The derived 60 $`\mu `$m excess for each of the 5 samples and the associated error estimated from the bootstrap resamplings are listed in Table 1. As can be seen, the variance among the 5 samples is consistent with the errors calculated using the bootstrap technique and is around 0.03 dex. We thus conclude that our measured 60 $`\mu `$m excess of 0.2 dex is statistically significant. In the same way, we derived an error of 0.035 dex on the excess measured at 100 $`\mu `$m and in L<sub>FIR</sub>. This implies that the excesses found at 60 and 100 $`\mu `$m differ at about the 2$`\sigma `$ level.
Figure 11 shows histograms of the distribution of $`\mathrm{\Delta }\mathrm{log}L_{60}/M_{}`$ , $`\mathrm{\Delta }\mathrm{log}L_{100}/M_{}`$ and $`\mathrm{\Delta }\mathrm{log}L_{\mathrm{FIR}}/M_{}`$. Note that the results for $`L_{100}`$ and $`L_{\mathrm{FIR}}`$ are based on a sample of 111 pairs from the catalogue with reliable flux measurements in both the 60 and 100 $`\mu `$m bands. Fig. 11 demonstrates the excess IR emission in AGN is larger at 60 $`\mu `$m than at 100$`\mu `$m.
## 5 Interpretation and Discussion
We have demonstrated that when IRAS-selected AGN and normal star-forming galaxies are carefully matched in terms of parameters such as stellar mass, size, concentration, redshift and 4000 Å break, the AGN exhibit “excess” far-IR emission af about 0.18 dex in $`\mathrm{log}L_{\mathrm{FIR}}/M_{}`$.
What causes this excess emission in AGN? One possibility is that the excess originates from the active nucleus itself. The question one might then ask is whether this is energetically feasible. In Type 2 AGN, the central engine is obscured and the AGN luminosity can only be estimated indirectly. Heckman et al. (2004) have discussed how the \[OIII\] luminosity can be used as a tracer of AGN activity. They used the bolometric correction to the \[OIII\] luminosity derived for Type 1 AGN to estimate the average mass accretion rate onto black holes in the local Universe. The implied bolometric correction was $`L_{Bol}/L_{O3}3500`$, where $`L_{O3}`$ was the \[OIII\] luminosity in solar units, uncorrected for extinction due to dust. The quoted uncertainty on this conversion was $`0.4`$ dex.
In the left panel of Figure 12, we plot $`L_{\mathrm{FIR}}`$ as a function of the raw, uncorrected \[OIII\] luminosity for the AGN in our sample with reliable flux densities at 60 and 100 $`\mu `$m. We have found that these AGN have on average a $`0.18`$ dex excess in $`\mathrm{log}L_{\mathrm{FIR}}/M_{}`$. This would imply that one third of the total FIR luminosity is from the AGN. In the left panel of Figure 12, we see that the median ratio of FIR to \[OIII\] luminosity is 5 dex. In the assumption that the quoted $`0.18`$ dex excess in $`\mathrm{log}L_{\mathrm{FIR}}/M_{}`$ is entirely due to the AGN, a FIR-\[OIII\] luminosity ratio of 5 dex would imply a typical ratio of FIR (AGN) to \[OIII\] of several 10<sup>4</sup>. This is about an order-of-magnitude greater than the entire bolometric luminosity of the AGN if we assume the bolometric correction of Heckman et al. At first glance, this would appear to rule out the hypothesis that the observed excess could be produced by the active nucleus. We caution, however, that our IRAS-selected AGN are not typical of the general population of AGN in the SDSS spectroscopic sample. In particular, they have considerably higher H$`\alpha `$/H$`\beta `$ ratios, which implies that the amount of extinction for the \[OIII\] line is larger. In the right hand panel of Figure 12 we plot $`L_{\mathrm{FIR}}`$ as a function of the extinction-corrected \[OIII\] luminosity. The average extinction correction to \[OIII\] for the AGN in our sample is a factor of $`100`$. This is five times larger than the correction derived for the general population of SDSS AGN with raw \[OIII\] luminosities in the same range as those in our SDSS/IRAS sample. It is not unreasonable to suppose that strong systematic effects may arise when applying a calibration derived for “typical” Type 1 AGN to a sample of very dusty Type 2 AGN. Figure 12 shows that if the extinction correction is applied, the median ratio of FIR to \[OIII\] luminosity is $``$3 dex. The FIR excess attributed to the AGN would now be only several hundred times the extinction-corrected \[OIII\] luminosity. This is still rather large to account for the measured infrared excess of AGN.
An alternative explanation is that the excess infrared emission is caused by an extra component of star formation, which is not reflected in the measured 4000 Å break strengths. Kauffmann et al. (2003) showed that the distribution of AGN with strong \[OIII\] emission in the plane of H$`\delta `$ absorption line strength versus 4000 Å break strength was systematically different than that of ordinary star-forming galaxies of the same stellar mass. The AGN were offset to higher values of H$`\delta _A`$ at a given value of D<sub>n</sub>(4000), indicating that they were more likely to have experienced a recent burst of star formation. In the left panel of Figure 13, we compare the H$`\delta `$ equivalent widths of IRAS-selected AGN and normal star-forming galaxies and find that there is no significant offset in mean equivalent width between the two populations. Both populations are characterized by strong H$`\delta `$ absorption lines and irregular morphologies (Pasquali et al., in preparation) and we conclude that IRAS selection favours galaxies with bursty star formation histories irrespective of whether or not they are classified as an AGN. If there is an excess component of star formation in the AGN in our sample, it clearly cannot be detected using standard indicators in the optical part of the spectrum. In the left panel of Fig. 13, we compare the Balmer decrements of the AGN and the normal star-forming galaxies. The AGN exhibit a small offset towards larger values of $`H\alpha /H\beta `$. This may indicate that AGN contain slightly more dust, but an offset in this direction is also expected because of the different ionization conditions in these systems. The offset in Balmer decrement between AGN and normal star-forming galaxies is not sufficient to explain the offset in FIR luminosity; we have verified this by creating AGN/galaxy samples that are closely matched in redshift ($`<0.03`$), stellar mass ($`<0.25`$ dex), D4000 ($`<0.07`$) and H$`\alpha `$/H$`\beta `$ ($`<0.005`$ dex) and we find that the mean offset in $`L_{\mathrm{FIR}}/M_{}`$ changes very little.
At infrared wavelengths, the only diagnostics available to us are the IRAS 60 and 100 $`\mu `$m fluxes. It is possible to study whether AGN exhibit a greater degree of scatter in their far-IR properties relative to normal star-forming galaxies. We have created samples of AGN/AGN and galaxy/galaxy pairs using the exact same matching criterion used for the galaxy/AGN pairs. We then compare the scatter in 60 $`\mu `$m luminosity differences for the two pair samples. Figure 14 presents the results of this analysis and shows clearly that the AGN exhibit greater spread in their IR luminosities than the star-forming galaxies in our sample. This is consistent with the idea that the star formation in AGN is more episodic but it does not prove the hypothesis that an excess population of young stars is present in the AGN. One could imagine that dust heating by the central source might also be subject to temporal fluctuations. What is required in order to understand our results in more detail is high resolution imaging of these galaxies at infrared or radio wavelengths. This would enable us to compare the distributions of the dust emission and star-forming sites in the two classes of galaxies. With the combination of spatial information and good statistics, it should be possible to disentangle the physical mechanisms responsible for the excess far-infrared emission observed in nearby Seyfert 2 galaxies.
## Acknowledgments
We would like to thank S. Charlot and C. Tremonti for helpful discussions, and an anonymous referee for useful comments that 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/astro-ph0506378.html
|
ar5iv
|
text
|
# Radiative transfer in moving media
## 1 Introduction
Radiation is the major source of information about stars. A number of stellar properties can be obtained by comparing observed spectra with the synthetic ones calculated from model stellar atmospheres. A key problem in computing synthetic spectra is solving the radiative transfer equation in stellar atmospheres, since the emergent radiation is formed in these regions. There exists a large number of methods for solving the transfer equation in a static one-dimensional case, which are, unfortunately, inappropriate for stars with stellar winds, rapidly rotating stars, accretion discs, and also for nebulae. Our main interest is to study various aspects of the stellar wind in hot stars, therefore we want to develop a method for solving the radiative transfer equation that is well suited for this case, where the symmetries enabling a one-dimensional approximation are broken.
There are not many methods available to solve static multidimensional radiative transfer problems. For optically thin regions, the Monte Carlo method (Boissé boisse (1990)) may be used. On the other hand, in the optically thick regions it is possible to solve the transfer equation using a method that employs the diffusion approximation (Kneer & Heasley difuzemulti (1979)). Neither of these methods are suitable for stellar atmospheres, where the transfer problem must be solved from optically thick regions to regions where the optical thickness is very small. The classical way of solving the radiative transfer equation in more dimensions is the long characteristics method (Cannon dlouhy (1970)). This method fully describes the radiation field, but the computer time necessary to obtain a solution may become long. For this reason, Kunasz & Auer (kratky (1988)) developed the short characteristics method, which is the best currently available multidimensional method. There exist several applications of short characteristics methods in a Cartesian grid (Fabiani Bendicho (zaklad, 2003, and references therein)) and 2D axially symmetric geometry (Georgiev et. al bulhar (2003)). Dullemond & Turolla (dullemond (2000)) developed a “rotating plane” method for axially symmetric problems based on the short characteristics method, too. An efficient approach is to use adaptive grids following Folini et al. (doris (2003)) and Steinacker et al. (steinray (2003)). A review by Auer (2003b ) provides an excellent insight into the grounds for astrophysical multidimensional radiative transfer.
Another possibility is to apply the finite element method, which has been recently used for multidimensional radiative transfer by, e.g., Richling et al. (femulti (2001)). However, the finite element method is not used very often for radiative transfer in stellar atmosphere studies due to its convergence problems caused by the about ten orders of magnitude changes of physical quantities (e.g. opacity, density) in stellar atmospheres. Dykema et al. (DFEIII (1996)) tried to overcome the instability and convergence problems using a modification of the finite element method, namely, the discontinuous finite element method. The discontinuous finite element method was also used in our previous paper (Korčáková & Kubát kk (2003), hereafter Paper I) for one-dimensional radiative transfer.
If the velocity field is present, the situation is more complicated. The changes in opacity and emissivity along a ray in spectral lines can be large due to the Doppler shift, and we have to take this into consideration. In the continuum, we can simply use the static equation, since the Doppler shift has only a small effect on the opacity and emissivity coefficients. For simplicity the methods assuming monotonic velocity field have been used very often. The multidimensional problem with velocity fields can be solved in the observer frame (the frame connected with the center of a star) or in the comoving frame (the frame connected with the outflowing material). Radiative transfer in more dimensions has usually been solved in the observer frame (e.g. Carlsson & Stein multitubingen (2003)). One of the firsts works to do this is Mihalas, Auer & Mihalas (mihalas78 (1978)), who were able to solve a periodic velocity field in two-dimensional planar geometry. Recently, van Noort et al. (vnapj (2002, 2003)) solved this problem in two dimensions. Their code is able to cope with both spherical and cylindrical geometry for Cartesian coordinate systems. The observer frame is appropriate only for small velocity gradients. The velocity difference between neighbouring depth points has to be smaller than several times the thermal velocity. If the velocity gradient is large, the Sobolev approximation (Sobolev Sobolev (1947)) is often used. This approximation was used for a 3D case by Folini et al. (doris (2003)). Both cases of small and large velocity gradients can be solved in the comoving frame. For a 3D accretion disc, the radiative transfer in the comoving frame was solved by Papkalla (papkalla (1995)).
However, there exists another effect that is usually neglected or taken into account in an approximate manner, namely the rotation of stars. Stellar rotation is usually taken into account as a convolution of the line profile, obtained from the plane-parallel static solution, and of the rotation profile (cf. Gray konvoluce (1976)). In this technique one must make use of analytical expressions for limb darkening, which do not give a correct description of the angular dependence of the specific intensity. It also does not take into account the fact that limb darkening is strongly frequency dependent across the line profile (cf. Hadrava & Kubát hadku (2003)). In order to describe the radiation field and its influence on the stellar atmosphere correctly, it is also necessary to take into account the Doppler shift of lines in a rotating atmosphere for the transfer of radiation along nonradial rays.
In this paper we present a new method to formally solve the radiative transfer equation in axial symmetry, which does not employ the Sobolev approximation. Our method is based on the Local Lorentz Transformation method (LLT) described in Paper I for the one-dimensional case. For studies of stellar wind, accretion discs or stellar rotation we need a method that uses a more general geometry than plane-parallel or spherical. However, it is not necessary to treat the whole three dimensional space in detail. We employ axial symmetry. The results will be more accurate than in a plane-parallel or spherical geometry and the calculation will be faster than for the full 3D problem.
In the first part of this paper we present our axially symmetrical code. The geometry of the model is described in detail, followed by some tests of our code. We compare these results with a spherically symmetric model atmosphere from Kubát (ATAmod (2003)). We also present some test calculations with a velocity field. Our line profiles are compared with convolved profiles for the case of stellar rotation. Some results for the stellar wind are also shown. Then we test the dependence of the computing time on the number of grid points. In the last part, we comment on the possibilities of the application for our method.
## 2 Method
We solve the radiative transfer equation for the specific intensity of radiation $`I`$
$$nI(r,n,\nu )=\chi (r,\nu )\left[I(r,n,\nu )S(r,\nu )\right],$$
(1)
where $`\chi `$ is the opacity, $`S`$ is the source function, $`n`$ is the direction of radiation, $`r`$ is the radius vector, and $`\nu `$ denotes the frequency. We assume axial symmetry and we allow for a nonzero velocity field.
The basic idea of this method is to solve the radiative transfer problem not in the whole star, but in separated planes intersecting the star.
### 2.1 The spatial grid
Let us consider the spherical coordinate system ($`r`$, $`\theta `$, $`\varphi `$). We choose as the axis of symmetry $`\theta =0`$ (see Fig. 1). First, we introduce the discretization of the radial distance $`r`$ and angle $`\theta `$. Due to the axial symmetry, physical quantities do not depend on the angle $`\varphi `$. The grid is chosen to give the best description of the system, depending on the studied object. As an example, for the study of stellar winds, the grid of angles $`\theta `$ can be equidistant. On the other hand, for accretion discs it must be finer near the equatorial plane, where the disc resides. In radial distance $`r`$ we choose the grid (very similarly to a 1D problem) to be equidistant in the logarithm of the radial optical depth. A suitable choice is about 5 points per decade. We assume the opacity and emissivity of the stellar material to be known at the grid points.
To reduce the 3D problem to a 2D one, we do not solve the radiative transfer equation directly in the 3D grid, but in a set of “longitudinal planes” intersecting the star parallel to the plane $`\varphi =0`$ (we “slice the star” – see Fig. 1). In thin stellar atmospheres it is favourable to choose a finer grid of longitudinal planes closer to stellar limb, since the limb darkening is better described by such a choice.
The primary grid described above helps us only to interpolate the quantities to the selected planes and finally to calculate the emergent flux. To solve the transfer equation we need a slightly modified grid. For each longitudinal plane we choose the polar coordinate system and define a grid of concentric *grid circles* and *radial grid lines* (see Fig. 1). The grid of concentric circles corresponds to the original 3D grid in the planes, which intersect the opaque stellar core where we do not solve the transfer equation and where we apply the lower boundary condition following from the diffusion approximation. However, choosing an appropriate grid for planes which do not intersect the opaque stellar core may become difficult, since the grid chosen must correspond to the geometry of the problem as well as resolving the velocity field. This is shown in Fig. 2. We must provide an appropriate grid
for both geometrically thin and extended atmospheres.
For geometrically thin atmospheres the grid of radial distances in the longitudinal plane is chosen using intersection points with grid circles in the primary 3D grid (left panel in Fig. 2). For an extended atmosphere it is better to define grid points where the plane intersects the radial grid lines (right panel in Fig. 2). The reason why we need an appropriate grid in the planes that do not intersect the inner boundary is the necessity to ensure sufficient spatial resolution for the radiative transfer equation solution. This may be difficult in the central region of the plane. Moreover, similar problem involves by the velocity field. The velocity gradient in the planes far from the center can be large, so it is necessary to have a sufficient number of grid points.
The opacity, emissivity and the source function are interpolated to the new coordinate system. Linear interpolation is used here, since the grid is not orthogonal. The radiative transfer equation is solved for each longitudinal plane independently (Section 2.3). The whole radiative field is obtained by rotating the separate planes around the axis of symmetry (Section 2.4).
### 2.2 The frequency grid
The frequency grid is to be chosen to enable the most efficient description of the radiation field. We set the frequency interval using both the largest line width and the total Doppler shift caused by the global motion. For example, for Doppler profile this interval is
$`\nu `$ $`\nu _0\left(1(\text{v}_{\mathrm{}}+\text{v}_{rot_0})/c\right)2\mathrm{\Delta }\nu _D,`$
$`\nu _0(1+(\text{v}_{\mathrm{}}+\text{v}_{rot_0})/c)+2\mathrm{\Delta }\nu _D,`$ (2)
where $`\text{v}_{\mathrm{}}`$ is the terminal velocity, $`\text{v}_{rot_0}`$ the rotation velocity in photosphere. $`\mathrm{\Delta }\nu _D`$ denotes the broadest Doppler halfwidth, which corresponds to the highest temperature. For very extended atmospheres or planetary nebulae (optically thin medium) we must take into account that the regions with $`\text{v}_{\mathrm{}}`$ and $`\text{v}_{\mathrm{}}`$ “see” each other. In that case, we must multiply $`\text{v}_{\mathrm{}}`$ in Eq. (2) by two. We take the frequency step to be equidistant for the case of the Doppler line profile. This step must be small enough to describe the line profile variations in the region with low temperature. The step is determined in the following way. We first typically set 5 points per line equidistantly at the depth point with the narrowest line profile (which is usually the depth point with the lowest temperature). Then we cover the whole frequency interval (2) with frequency points using this frequency step. This frequency grid is then used in other depth points where the line profile is broader. For a Voigt profile we need to extend the frequency interval (2), but the frequency step in the extended part of the interval may be larger.
### 2.3 Solution in longitudinal planes
In the given longitudinal plane we introduce the polar coordinate system using grid circles and radial grid lines defined in the preceding subsection (2.1). The values of temperature, density and velocity are known at the grid points. First, we solve the transfer equation in the plane starting at the outer boundary. Once the radiation field in the direction towards the center is known, we can continue to solve the transfer equation in the opposite direction, i.e. from the inner boundary towards the outer one. At the inner boundary the diffusion approximation may be taken for planes intersecting the opaque stellar center as a boundary condition. In other planes (that do not intersect the stellar center), we adopt as the lower boundary condition the intensity taken from the previously calculated solution from the outer boundary inwards.
#### 2.3.1 The solution from the outer boundary to the central regions
We begin the calculations at the outer boundary (stellar surface), where the boundary condition (i.e. the incoming intensity $`I`$) is known. At each inner grid point we choose several rays per quadrant (see Fig. 3). The rays start at the outer grid circle and end at a given grid point.
Along these rays we solve the transfer equation. The angle distribution of these rays may be the same as often used in the case of the plane-parallel geometry, where the angle cosines $`\mu =\mathrm{cos}\alpha `$ ($`\alpha `$ is the angle between the ray and the radial direction, see Fig. 3) are chosen to be the roots of Legendre polynomials in the interval $`(0,1)`$ ($`\mu =0.8872983346`$, $`0.5`$, $`0.1127016654`$) to ensure better numerical accuracy of the angle integration (Press et. al (recipes, 1986, section 4.5)). Three rays at a given point is usually sufficient, because the whole radiation field is obtained by summing the information from all longitudinal planes intersecting the given point (see Fig. 7). However, in some situations it is better to use more (up to 9) rays per quadrant to overcome a numerical error introduced by the necessary interpolation of intensity described below. As one can see in Fig. 3, the rays start at the preceding grid circle. This means that the rays do not usually start at a grid point and that they may also intersect some radial grid lines.
The solution diagram is illustrated in Fig. 4. We perform a linear interpolation of the source function and opacity to obtain their values at points $`A`$, $`B`$, and $`C`$ and of the incoming intensity for the value at point $`A`$.
The optical depth difference $`\mathrm{\Delta }\tau `$ is calculated along the ray between the individual intersection points (abscissas $`AB`$, $`BC`$, and $`CD`$) using the linear approximation,
$$\mathrm{\Delta }\tau _{(AB)}=\frac{\chi _A+\chi _B}{2}\mathrm{\Delta }s_{(AB)},$$
(3)
and similarly for $`\mathrm{\Delta }\tau _{(BC)}`$ and $`\mathrm{\Delta }\tau _{(CD)}`$. Here $`\chi _A`$ and $`\chi _B`$ are opacities at respective points and $`\mathrm{\Delta }s_{(AB)}`$ is the geometrical distance between points $`A`$ and $`B`$.
We solve the equation of radiative transfer by parts between all intersection points along the ray. We assume a linear dependence of the source function on the optical depth between the intersection points. For the interval $`AB`$ the solution is
$$I_{(B)}=I_{(A)}e^{\mathrm{\Delta }\tau _{(AB)}}+_0^{\mathrm{\Delta }\tau _{(AB)}}S(t)e^{[(\mathrm{\Delta }\tau _{(AB)}t)]}𝑑t,$$
(4)
and similarly for intervals $`BC`$ and $`CD`$. The final intensity $`I_{(D)}`$ at the grid point $`D`$ is thus determined by three successive applications of the equation (4). In this manner we obtain the specific intensity in the downward solution at every grid point. It may, of course, happen that the geometric distance along the ray in the given cell is very small (the ray near the point $`B`$ in the Fig. 4). In this case we do not solve the transfer problem in this cell and we simply set $`I_{(B)}=I_{(A)}`$. Doing this, we eliminate a numerical instability. Usually, it is enough if the condition $`|\mathrm{cos}(\pi /2\theta \alpha )|>10^7`$ ($`\alpha `$ has the same meaning as in Fig. 3) is fulfilled, even if it is much more accurate to estimate it using the optical depth difference.
There are several reasons why we prefer only a linear dependence of the source function on other higher order approximations. First, it is numerically stable and second, it is much faster. The higher order optical depth interpolation may sometimes lead to numerical errors by adding new extrema (cf. Auer 2003a ). They may become significant especially in moving medium. There is only one reason why linear interpolation should not be used: because the diffusion approximation in deep optically thick layers would be inaccurate. Since we do not solve the transfer problem using the short characteristics method, this inaccuracy is not large in our case. Our characteristics are longer and the transfer equation is solved in several steps, so the information from the farther regions is naturally included. Therefore we chose the safer linear interpolation.
#### 2.3.2 The solution from the central regions to the outer boundary
The upward solution is very similar to the previous step. The procedure is depicted in Fig. 5. We lead the rays to the grid points
using the same angles as in the previous case. The rays start either at the preceding grid circle (closer to the center) or at the same circle as the grid point. At the intersection points we interpolate the opacity and the source function. Between these points we calculate the optical depth using (3) and solve the transfer equation using (4) as in the previous case of downward integration.
From Fig. 1, one can see that there exist planes that do not intersect the inner boundary region. For these planes we adopt the intensity calculated from the previous step (solution from the upper boundary to the stellar center) as the lower boundary condition. However, the solution of the transfer equation in the central grid circle must be performed with care. The situation is shown in the Fig. 6. First, we solve the radiative transfer equation from the intersection point $`A`$ to the center of the abscissa $`AC`$ (point $`B`$) and then to point $`C`$. The physical quantities at point $`B`$ we obtain by linear interpolation from points $`A`$ and $`C`$. The optical depth difference is calculated using (3) and the specific intensity is determined using (4). The value of the intensity at point $`C`$ determines the lower boundary condition for further solutions outwards.
For planes, which intersect the region of validity of the diffusion approximation (the stellar core), the situation is easier. We simply use the appropriate lower boundary condition.
### 2.4 The full radiation field
To obtain the whole radiation field we take the advantage of the symmetry of the problem. We know the radiation field at grid points in all directions lying inside the longitudinal planes. The radiation field in the whole star is then obtained by rotating the longitudinal planes around the axis of symmetry $`\theta =0`$ (see Fig. 7). This gives a sufficient description of the specific intensity.
The emergent radiation flux towards the observer is calculated by integrating specific emergent intensity over the stellar disc.
### 2.5 Velocity field
The radiative transfer equation in moving media can be solved either in the observer frame or in the comoving frame. Since the coefficients of emissivity and opacity are angle dependent in the observer frame, we solve this problem rather in the comoving frame. This frame, which is coupled with the moving medium, is generally a non-inertial frame. The form of the radiative transfer equation in the comoving frame we can obtain either from the general relativity or from the special relativity by assuming a set of Local Lorentz Transformations.
In the Paper I we introduced an LLT (Local Lorentz Transformation) method for a solution of the radiative transfer equation in one-dimensional moving media based on an application of Lorentz transformations on cell boundaries and a static radiative transfer equation solution between them. The method was tested by a comparison with an exact solution of the radiative transfer equation using the discontinuous finite element method (see Fig. 8 in Paper I), which solves the comoving transfer equation including the $`/\nu `$ (Doppler) and $`/\mu `$ (aberration) terms. Here we apply the same idea for the case of a two-dimensional arbitrarily moving medium.
In our method, we replace the solution of the radiative transfer equation in the comoving frame by a set of Local Lorentz Transformations and a solution of the transfer equation in corresponding local inertial frames. In these inertial frames we solve a static radiative transfer equation, since it is Lorentz invariant. To do this, we must consider the constant velocity of every cell and the change of the velocity we allow only at the cell boundary (see Fig. 8).
Two other conditions must be fulfilled there. First, velocity with respect to the observer frame must be low to neglect aberration. This permits us to solve the radiative transfer along straight characteristic lines. This assumption is valid, if $`\text{v}/c0.01`$ (Mihalas et al. CMF3 (1976), Hauschildt et al. H95 (1995)). Second, let us assume a steady-state fluid. Due to this assumption we can neglect the time delay between different parts of the medium. The correctness of this approach has been discussed in more detail in Paper I, where this method was tested for a plane-parallel geometry.
The basic solution scheme within cells, which was described in Section 2.3, is not affected by the presence of the velocity field. We project the velocity field to the longitudinal planes. At cell boundaries we perform the Lorentz transformation of frequency $`\nu `$ (see also Eq. (24) in paper I)
$$\nu ^{}=\nu \sqrt{1\frac{\left(\mathrm{\Delta }\text{v}\right)^2}{c^2}}\left(1\frac{n\mathrm{\Delta }\text{v}}{c}\right)$$
(5)
($`\mathrm{\Delta }\text{v}`$ is the velocity difference between cells). Since the velocity field is not relativistic, we can simply use the classical Doppler law, which speeds up the solution. The Lorentz transformation of intensity $`I^{}\left(\nu ^{}\right)=I\left(\nu \right)\left(\nu ^3/\nu ^3\right)`$ (see Eq, (23) in Paper I) at the cell boundary has a negligible effect, and we do not take it into account, since the intensity is proportional to the third power of the frequency ratio. This is close to one in most stellar applications. Since the equation of the radiative transfer is Lorentz invariant, we solve the static equation of the radiative transfer (4) within each cell.
Since we solve the transfer equation in the comoving frame, we have to do one additional step. We know the source function and opacity in the frequency grid points in the rest frame coupled with a given cell. The incoming intensity from a neighbouring cell is known at different frequencies, which are Doppler shifted according to the velocity difference between the cells. This means that we must interpolate the intensity to frequency points coupled to the rest frame of a cell.
To ensure a correct treatment of line transfer we have to guarantee that the frequency shift at the cell boundaries due to Doppler effect is smaller than a quarter of a Doppler halfwidth (see Paper I). If it is not, we must make the grid finer. This condition must be fulfilled even if we calculate with the Voigt profile since the center of the line has a Doppler core. This is the only limitation of this method. If the velocity gradient is too high, we must refine the grid, which slows down the calculation.
## 3 Test calculations
Tests of the method are based on a model atmosphere of a main sequence B star with effective temperature $`T_{\mathrm{eff}}=17\times 10^3\mathrm{K}`$, gravitation acceleration $`\mathrm{log}g=4.12`$, and radius $`3.26R_{}`$. We obtain the state parameters, electron density and temperature, using the hydrostatic spherically symmetric model atmosphere code ATA (Kubát ATAmod (2003)). We consider no incoming radiation as the outer boundary condition and the diffusion approximation as a lower boundary condition.
### 3.1 Static case
For a basic comparison we took a model atmosphere calculated using the static computer code ATA. The radiative transfer in ATA is solved using the long characteristics method and Feautrier variables (see Kubát ATAdis (1993, 1994, 2003)). We compare the result from the ATA code with the flux obtained from our new code (Fig. 9). The line profile was chosen to be Doppler here and the input parameters in the axially symmetric code are independent of $`\theta `$. This comparison for this simplest case proves the basic correctness of the new code.
#### 3.1.1 Limb darkening
An important result arising from the solution of the transfer equation is the limb darkening law. In Fig. 10 we plotted the limb darkening law across the $`\mathrm{H}\alpha `$ line profile calculated from our code. We choose as an x-axis the distance from the center of star in star radius units ($`x=1`$ for $`r=R_{}`$, where $`R_{}`$ is the stellar radius). At this scale the effect of limb darkening is more clearly visible.
To show this effect in more detail, we extracted the dependence of the specific intensity on the distance from the center of the stellar disc for a continuum frequency (Fig. 11) and for the central frequency of the $`\mathrm{H}\alpha `$ line (Fig. 12).
The obtained data for the continuum are compared with functions usually used to express the limb darkening law. The first function ($`f(x)`$) is adopted from Allen’s Astrophysical Quantities (allen (1963)),
$$I(x)=1ab+a(1x^2)^{1/2}+b(1x^2)f(x).$$
(6)
Using the least squares method we obtain the parameters $`a=0.55\pm 0.02`$ and $`b=0.20\pm 0.01`$ to fit our results to Eq. (6). Limb darkening is also often described by the law (Gray konvoluce (1976)),
$$I(x)=(1ϵ)+ϵ(1x^2)^{1/2}g(x).$$
(7)
Fitting this case we obtain the value of the parameter $`ϵ=0.277\pm 0.008`$.
Figure 12 shows the limb darkening in the center of the $`\mathrm{H}\alpha `$ line.
As one can see, brightening instead of darkening is observed near the stellar limb, which corresponds to the effect of flash spectra in solar chromosphere. The same result was obtained also by Hadrava and Kubát (hadku (2003)).
### 3.2 Stellar wind
The case of the moving atmosphere is first tested for a spherically symmetric stellar wind, for which we adopt the beta law for the dependence of the wind velocity v on radius $`r`$ (see, e.g., Lamers & Cassinelli CL99 (1999))
$`\text{v}(r)=\text{v}_{\mathrm{}}\left\{1\left[1\left({\displaystyle \frac{\text{v}_R}{\text{v}_{\mathrm{}}}}\right)^{\frac{1}{\beta }}\right]{\displaystyle \frac{R}{r}}\right\}^\beta .`$ (8)
We choose a velocity in the photosphere of $`\text{v}_R=200\mathrm{km}\mathrm{s}^1`$ and a terminal velocity $`\text{v}_{\mathrm{}}=2000\mathrm{km}\mathrm{s}^1`$.
To check the consistency of our new method we compare it with the plane-parallel method described in Paper I. The plane-parallel radiative transfer equation is solved using the discontinuous finite element method. The velocity field is included using the Local Lorentz Transformation similarly to this paper. In Paper I the Local Lorentz Transformation method was compared with the solution of the radiative transfer equation using the discontinuous finite element method, where the frequency term was consistently included into the solution as an independent variable. Since the velocity field is non-relativistic, calculations with aberration give the same results as without it (in agreement with Mihalas et al. CMF3 (1976), Hauschildt et al. H95 (1995)).
The result from plane-parallel geometry is plotted using the dashed line in Fig. 13. As we can see, the agreement with the new (axially symmetric – solid line) results is not good. The reason is in the geometry used. Rays with large angle (low $`\mu `$) in a curved atmosphere (spherically symmetric or axially symmetric) do not even touch regions that are optically thick in continuum, whereas for a plane parallel geometry *all* rays with $`\mu >0`$ finally reach the continuum optically thick regions (see Fig. 14). As a consequence, there is more continuum radiation in a plane parallel geometry, which results in shallower line profiles. This has been well known for a long time (see, e.g., Kunasz et al. spher2 (1975)).
This is why we plot another line profile, which also is obtained using the plane-parallel geometry, but in which the intensity incoming from directions with $`\mu <0.26`$ is set to zero. We can see a very good agreement in the blue wing of the line profile, where we see central parts of the star and where the assumption of plane-parallel atmosphere is acceptable. However, the sphericity effects, limb darkening and limb brightening influence the red wing of the line. A different profile from a axially symmetric code is in agreement with the fact the line profile obtained in spherical geometry is different from that obtained in a plane parallel geometry. The difference depends on the sphericity effects on the temperature structure (see, e.g., Kunasz et al. 1975, Gruschinske & Kudritzki 1979, Kubat 1995, 1999).
In Fig. 15 we compare the line profiles for three different values of the parameter $`\beta `$ ($`0.5`$, $`1`$ and $`2`$). To show the possibility of our code to solve a more general velocity field we plot in Fig. 15 the line profile obtained using a decelerating velocity field. We choose a linear dependence of the velocity on the radial distance in a logarithmic scale. The photospheric and terminal velocities are 2000 $`\mathrm{km}\mathrm{s}^1`$ and 200 $`\mathrm{km}\mathrm{s}^1`$, respectively.
The most important point is the velocity gradient in the region of line formation. For this reason, lines corresponding to different values of the $`\beta `$ (in Fig. 15) parameter have different profiles, even if the values of the photospheric and infinity velocity are the same. We do not see the P Cygni profile, since the input parameters are based on the hydrostatic model, which produces a geometrically thin atmosphere. To obtain a P Cygni profile, not only enough high velocity gradient, but a sufficiently extended line formation region are necessary. Consequently, only a blue shifted absorption profile appears in our case. Note that similar line profile shapes were obtained by Noerdlinger & Rybicki (ppvel (1974)) for a plane-parallel expanding atmosphere.
### 3.3 Stellar rotation
To test the stellar rotation we assume a spherical star and we consider the power-law dependence of the rotation velocity on the radius,
$`\text{v}(r)=\text{v}(R_{})\left({\displaystyle \frac{r}{R_{}}}\right)^j,`$ (9)
where $`R_{}`$ is the stellar radius and $`\text{v}(R_{})`$ is the rotation velocity in the photosphere. We choose the parameter $`j`$ to be equal to $`1`$, which corresponds to equatorial discs formed by a stellar wind (Kroll & Hanuschik kroll (1997)) using conservation of the angular momentum. Stars which rotate near their critical velocity are far from being spherically symmetric. For example, the ratio of the polar and equatorial radii of $`\alpha `$ Eridani was determined to be about one half (Domiciano de Souza achernar (2003)) or even $`1/5`$ (Jackson et al. ach (2004)). In these stars the approximation of spherical symmetry fails and we must solve the transfer problem in a more general geometry. However, although gravity darkening is neglected in our test case, our code is able to easily handle this effect.
We compare line profiles calculated using our model with a rotational velocity field and detailed radiative transfer with other possibilities of calculation of the rotationally broadened profile.
The standard and most commonly used way is to solve a detailed static plane-parallel radiative transfer equation for a static atmosphere and to convolve the resulting profile with a rotation profile using the relation (see Gray konvoluce (1976)),
$`{\displaystyle \frac{F_\nu }{F_c}}=H(\nu )G(\nu )={\displaystyle _{\mathrm{}}^{\mathrm{}}}H(\nu \mathrm{\Delta }\nu )G(\mathrm{\Delta }\nu )𝑑\mathrm{\Delta }\nu .`$ (10)
Here, $`F_\nu `$ is the flux for a given frequency, $`F_c`$ is the continuum flux, $`H(\nu )`$ is the normalized flux from the nonrotating star. The rotation profile $`G(\nu )`$ is equal to
$$\begin{array}{c}G(\nu )=\frac{2(1ϵ)\sqrt{1\left(\frac{\mathrm{\Delta }\nu }{\mathrm{\Delta }\nu _{\mathrm{max}}}\right)^2}+\frac{1}{2}\pi ϵ\sqrt{1\left(\frac{\mathrm{\Delta }\nu }{\mathrm{\Delta }\nu _{\mathrm{max}}}\right)^2}}{\mathrm{\Delta }\nu _{\mathrm{max}}\pi \left(1\frac{ϵ}{3}\right)}\hfill \end{array}$$
(11)
in the interval $`|\mathrm{\Delta }\nu |<\mathrm{\Delta }\nu _{\mathrm{max}}`$, and $`G(\nu )=0`$ elsewhere. This expression assumes the limb darkening law in the form (7) and the parameter $`ϵ`$ is the same as in the Eq. (7).
A more exact and also computationally more expensive approach also uses the emergent radiation from the static atmosphere, but takes into account the angle dependence of the specific intensity. The resulting profile is then calculated by integrating the specific intensity across the stellar disc. This approach will be called “integrated static profile” hereafter. In these two latter approaches the radius dependence of the rotational velocity (9) cannot be taken into account and the photosphere is tacitly assumed to rotate as a rigid body.
The most exact solution is to calculate the emergent radiation using a full solution of the radiative transfer equation in a moving atmosphere, as has been done using our method. Results for the rotation velocity described by Eq. (9) with $`\text{v}(R_{})=0.2\text{v}_c`$ are shown in Fig. 16. Here, $`\text{v}_c=\sqrt{GM_{}/R_{}}`$ is the critical rotational velocity, which is equal to $`540\mathrm{km}\mathrm{s}^1`$ for our case.
In this figure we plotted the profile of the $`\mathrm{H}\alpha `$ line in the static case, the profile obtained from convolving the static profile with the rotation profile (11) using values $`ϵ=0`$ and $`1`$, the one from the integration of the static profile over the stellar disc, and the “true” profile calculated with the full influence of the rotation velocity field. The ratio of the equivalent width in the static case and the flux obtained from integrating of the line profile over the disc is of the order of $`10^5`$, which represents good numerical accuracy.
Both line profiles obtained using convolution (10) are deeper than those calculated from our code. The difference between line depth obtained by convolution and line depth obtained from our code $`(I_{\mathrm{our}}I)/(1I_{\mathrm{our}})`$ is about $`5`$% for parameter $`ϵ=0`$ and almost $`20`$% for $`ϵ=1`$. The error of five percent is not too large, since sometimes the observed spectra have a lower accuracy. However, for high S/N spectra one may have an accuracy better than $`1\%`$, which makes the error of $`5\%`$ significant. There is no doubt about the detectability of a $`20\%`$ difference. The commonly used value of the parameter, $`ϵ0.6`$, yields remarkable differences as well. This difference is a consequence of the dependence of limb darkening on frequency across the line profile. As we can see from Fig. 10, the continuum intensity decreases with increasing distance from the center of the stellar disk. On the other hand, the intensity in the center of the line is significantly changed very close to the limb (see Fig. 12). These effects cannot be described by the simple formulae (6) and (7). The difference between the line profile calculated by integrating the specific intensity over the disc and the profile which includes the velocity field and detailed radiative transfer is very small (see the magnified part of Fig. 16), as expected, since the radial velocity gradient is relatively small and the line forming region is very thin. For more extended and rapidly rotating sources these effects will be amplified.
### 3.4 Accretion discs
Our method is also appropriate for solving the radiative transfer equation in accretion discs. It can take advantage of the accretion disc geometry. Since the disc is densest and, consequently, optically thickest close to the equatorial plane, we can employ the possibility of unevenly distributed angles $`\theta `$ in the definition of the primary grid.
In addition to having a better resolution of the disc, the chosen grid is also able to describe possible jets. Using this method we can also include in the calculations the boundary region, winds from this region and from the inner disc, as well as the hot corona beyond the disc. Another advantage of this method is the ability to handle both optically thin and thick discs. Emergent spectra from the accretion disk of a cataclysmic variable calculated using our method were presented in Korčáková et al. (2004b ).
### 3.5 The tests of the grid
The grid tests show a linear dependence of the computing time on the number of geometrical depth points $`D`$ (see Fig. 17, left panel). The right panel of the same figure shows the dependence of the computing time on the number of frequency points $`N`$. The fitting function is a polynomial of the third order.
Although one may expect only linear dependence of the computing time on a number of frequency points, higher order dependence is due to the necessity of interpolating the intensity for each frequency point at each boundary of the cells due to the Doppler shift. In Fig. 18 we show the dependence of computing time on the number of angular grid points $`I`$. The fitting function is also a straight line.
## 4 Conclusions
We presented a new method for solving the radiative transfer equation in axial symmetry with the possibility of including arbitrary velocity fields. The basic idea is to solve the transfer equation in planes, that intersect the object. In a given plane a combination of long and short characteristics methods is used. This method allows us to better describe the global character of the radiation field and the resulting computing time is not too long. The velocity field is taken into account using the Lorentz transformation of frequency, which allows us to solve the transfer problem in the region with a small velocity gradient as well as for high velocity gradients. This technique is very useful for studying stellar wind, where it is applicable to the stellar wind region together with the stellar photosphere.
Tests of this method were performed for a model atmosphere of a B type star with $`T_{\mathrm{eff}}=17\times 10^3\mathrm{K}`$, gravitational acceleration $`\mathrm{log}g=4.12`$, and radius $`3.26R_{}`$. For a static spherically symmetric atmosphere our code gives correct results, as can be seen from Figure 9. We also present the limb darkening law for our model (Fig. 10). Note that the line profile shows limb brightening at the central line frequency (Fig. 12). This result is in agreement with the results of Hadrava and Kubát (hadku (2003)).
Further tests were performed in the presence of a velocity field. For an expanding atmosphere (stellar wind) we adopt the classical $`\beta `$-velocity law (8). The resulting line profile is shown in Fig. 15. Since we take the input parameters (temperature, density) from the hydrostatic code, we do not obtain a P Cygni profile. However, the line profile is shifted to the blue part and deformed. Note that from these blue parts of ultraviolet lines one can determine the wind terminal velocity.
In section 3.3 we studied the application of our method to the problem of stellar rotation. Fig. 16 shows the necessity of including the frequency dependence of limb darkening in calculating rotationally broadened line profiles. On the other hand, the difference between the flux calculated by integrating the Doppler shifted static profile across the disc and the “true” flux that included the solution of the transfer problem with a velocity field is very small in a thin atmosphere.
The dependence of computational time on the number of angular points (Fig. 18) as well as on the number of geometrical points is linear (Fig. 17, left panel). The dependence on the number of frequency points is more complicated, which stems from the interpolation of intensity due to the Doppler shift at cell boundaries.
The limitation of this method is only in an axially symmetrical approach and not very steep velocity gradient. In the latter case it is necessary to refine the grid and the computing time becomes longer. It is not possible to use it for the relativistic velocities too, since the aberration effect must be included in this case.
Our method for solving the radiative transfer equation is especially suitable for including stellar winds, since it is able to handle the outer region of a stellar wind together with the stellar photosphere. This is necessary, above all, for line-driven winds of hot stars. The process of initiating this type of wind near the photosphere is not fully understood yet. Calculations that involve the wind together with the quiet and possibly almost static stellar atmosphere may be able to resolve this problem in the future. Our method can also be used to solve the transfer problem in very rapidly rotating stars. Gravitational darkening is too large in these objects, so it is not possible to neglect the dependence of physical properties on the angle and to calculate using the assumption of spherical symmetry. The results of the application of our method can become useful for interpreting interferometric observations. The geometrical flexibility of our method is also very useful for studying extremely nonspherical objects such a accretion discs. It is possible to simultaneously include the disc, central object, jets and hot corona. Another advantage of this method is the possibility of solving both optically thin and thick discs.
###### Acknowledgements.
The authors would like to thank the referee for valuable comments on the manuscript and Dr. Adéla Kawka for her comments. This research has made use of NASA’s Astrophysics Data System. This work was supported by grants GA ČR 205/01/1267 and 205/04/P224. The Astronomical Institute Ondřejov is supported by projects K2043105 and Z1003909.
|
warning/0506/astro-ph0506085.html
|
ar5iv
|
text
|
# Optical and Infrared Signatures of Ultra-luminous X-ray Sources
## 1 INTRODUCTION
The presence of point-like, non-nuclear and extremely luminous X-ray sources in local galaxies has been recognised for some time (Fabbiano & White,, 2003; Swartz et al.,, 2004). These ultra-luminous X-ray sources (ULX) can have luminosities $`10^{40}`$ergs s<sup>-1</sup>, and their true nature is yet to be understood. This luminosity greatly exceeds the Eddington luminosity $`L_{Edd}`$ of a 10 M black hole (BH) if the emission is isotropic. This has led to the suggestion that the accreting object could be an intermediate mass black hole (IMBH) with mass $`501000`$ M (Colbert & Mushotzky,, 1999; Makishima et al.,, 2000). If so, these objects are a link between the established population of stellar mass black holes, and the supermassive black holes in active galactic nuclei. The presence of cool X-ray spectral components, and the timescales of rapid variability in some systems (e.g. NGC 4559 X-7, Cropper et al., 2004) is consistent with the IMBH scenario. However, the existence of IMBH is still under debate, as some observations appear contradictory. For example, some systems require high accretion disk temperatures in model fits to the X-ray data (Ebisawa et al.,, 2003). It has also been noted that the emission is not likely to be isotropic, but collimated after to a greater or lesser extent (King et al.,, 2001; Körding et al.,, 2002; Fabrika,, 2004). Alternatively, it has also been argued by Begelman, (2002) that an accretion disk dominated by radiation pressure would exhibit strong density inhomogeneities on scales much smaller than the disk scale height, so that an inhomogeneous accretion disk could permit escaping flux to exceed $`L_{Edd}`$ by a factor of up to $`10100`$.
It seems increasingly likely that the ULX population is heterogeneous, with evidence to support beaming with stellar mass BHs in some cases and IMBH in others (Fabbiano,, 2004). The QPO reported by Strohmayer & Mushotzky, (2003) is incompatible with the beaming hypothesis, for example, as is the association of some ULXs with diffuse H$`\alpha `$ nebulae, suggesting isotropic illumination of the interstellar medium by the ULX (Pakull & Mirioni,, 2002; Miller et al.,, 2003). In this paper we assume an IMBH interpretation for ULX. Furthermore, we seek to describe the very brightest objects in the ULX population – specifically, those that have luminosities of $`10^{40}`$ergs s<sup>-1</sup> or greater (e.g. in M82, Matsumoto et al., 2001; Kaaret et al., 2001). We find that wind accretion from a companion star is insufficient to supply the accretion rate required for this luminosity. We therefore assume the BH accretes via Roche lobe overflow from a necessarily massive or giant companion.
The IMBH interpretion depends on the X-ray emission from the disk in the immediate environment of the ULX, which is model dependent. Moreover, the physics at these high accretion rates near the event horizon is far from understood. We therefore seek an alternative channel to X-rays by which the nature of the ULX can be explored, using optical/infrared (optical/IR) properties. We argue that these properties will be strongly influenced by the proximity of such an intense X-ray radiation field, and this can be used as a diagnostic. If the ULX is in a binary system, the X-ray emission will modify the optical/IR characteristics of the companion star and accretion disk. In particular, this will induce intensity and colour shifts compared to normal stars, and these will vary at orbital periods. These both identify the true optical counterpart and also provide relatively direct indications of the masses of the components.
X-ray irradiation can drive evolution in XRB (Podsiadlowski,, 1991; Ruderman et al.,, 1989), and cause significant colour and magnitude changes of the optical counterpart. This has been observed in the sub-Eddington regime. One example is the Her X-1 system, an X-ray binary consisting of a neutron star accreting matter from a non-degenerate stellar companion. The X-ray luminosity is a third of the Eddington luminosity (Howarth & Wilson,, 1983), and the binary period is $`1.7`$ days. The neutron star accretes matter via Roche lobe overflow through an accretion disk (Vrtilek et al.,, 2001), and so is a good analogy to the ULX systems described in this paper. The star has been observed to change spectral type from A to B over the binary period. Bahcall & Bahcall, (1972) observed a B magnitude amplitude of $`1.5`$ mag and interpreted this variation as a result of X-ray heating of a late A-type star. Other authors interpret the variation in terms of heating of the star and a tilted, precessing accretion disk (Gerend & Boynton,, 1976; Howarth & Wilson,, 1983).
We consider a binary model for ULX to investigate the radiation effects. We calculate the optical/IR observables from the companion star and accretion disk and discuss the implications of the results. We apply the results to the NGC 4559 candidates to either eliminate them as possibilities or constrain the parameters of any binary system of which they may be part. We also predict the IR properties for future observations.
We describe our model in Section 2, and examine the results of this model in Section 3. In Section 4 we compare the model to observations of ULX X-7 in NGC 4559.
## 2 MODEL
The accretion rate required by the brightest ULXs exceeds that which could be supplied by a stellar wind. We therefore assume the matter is transferred onto the compact object through Roche lobe overflow. We will use this assumption to constrain the geometry of the system. We assume that the system is in a quasi-steady state, and the irradiated surfaces are in thermal, radiative, and hydrostatic equilibrium. This requires that the irradiated layers necessarily re-emit all of the radiation falling on them.
### 2.1 The radiative transfer formulation
We consider the effects of radiative transport and radiative equilibrium in the irradiated surface under X-ray illumination. We consider a plane-parallel model and adopt the radiative transport formulation of Milne, (1926) and Wu et al., (2001) to describe the heated stellar surface and accretion disk. We detail this formulation in Appendix A.
#### 2.1.1 The heated stellar surface
The heating of any point on the stellar surface is described by the equations in Appendix A. We calculated the total heating by dividing the surface of the star into discrete cells and calculating the magnitude of the effect for each cell. The Roche surface for a given mass ratio was calculated over a grid of cells. Each cell has a flat surface, the size of which is dependent on the distance to neighbouring points. The angles $`\alpha `$ and $`\theta `$ (required components of the heating equations) can be calculated through appeal to the angle between the normal vector to the surface at the point in question, and the vector incident on the point originating at the BH ($`\alpha `$) or the observer ($`\theta `$). The irradiation temperature is calculated at this point, and taken to be the temperature over the entire surface of the cell.
The shape of the Roche Lobe is determined solely by the mass ratio $`q`$, so the angles and relative positions of the points are independent of the scale of the system. The system can therefore be easily scaled appropriately to calculate the heating, according to the binary separation $`a`$. This in turn is directly inferred from the binary period when the mass ratio is known. If the system is semi-detached then the size of the star itself is also determined by the scale of the system.
We calculate the luminosity of the star by summing the emergent radiation in the direction of the observer for each cell. This includes a component as a result of irradiation, as well as the original stellar luminosity, which includes both limb darkening and gravity darkening (Von Zeipel,, 1924) effects.
#### 2.1.2 The heated accretion disk - Dubus et al. prescription
We also include the emission from an irradiated accretion disk as an additional optical source. We have considered two models. The first follows Dubus et al., (1999) to describe this disk. There, the irradiation temperature $`T_{irr}`$ varies as
$$T_{irr}^4=C\frac{\dot{M}c^2}{4\pi \sigma R^2}$$
(1)
where $`\dot{M}`$ is the accretion rate and $`R`$ is the distance from the accreting source.
For accreting black holes,
$$L_x=\eta \dot{M}c^2,$$
(2)
where the efficiency parameter $`\eta 0.1`$. If we take an X-ray albedo of $`0.9`$ and assume that $`\eta =0.1`$, then the value of $`C`$ in Equation 1 would be $`2.57\times 10^3`$, with the geometry of a thin disk (Dubus et al.,, 1999; de Jong et al.,, 1996). Here and hereafter we use Equation 2 with $`\eta =0.1`$ to determine the value of $`\dot{M}`$ from $`L_x`$.
Using the fact that the radiation-transfer equations are linear, we can use the principle of superposition to calculate the disk temperature from the combination of the irradiative heating and the viscous heating in the disk in the absence of X-ray irradiation. We calculate the temperature as a function of disk radius, and then sum the flux from a series of blackbody annuli to describe the overall disk flux. We take the inner disk radius to be the last stable circular orbit around the BH we are describing. We take the outer disk radius to be the ‘tidal truncation radius’, beyond which Keplerian orbits intersect. This is weakly dependent on the mass ratio (Paczyński,, 1977) but is generally taken to be between $`0.6`$ and $`0.7`$ of the Roche lobe radius. We consider it sufficient in our model to fix the outer disk radius to be $`0.6`$ of the Roche lobe radius.
#### 2.1.3 The heated accretion disk - radiative transfer formulation
The alternative description of the accretion disk directly applies the radiative transfer formulation of Wu et al., (2001) that we have modified for for the star (see Appendix A). We consider a thin disk and determine a radial temperature profile in absence of irradiation using the Shakura & Sunyaev, (1973) prescription. We assume the local flare angle is given by $`h(r)r^{9/7}`$ (Dubus et al.,, 1999), where $`h`$ is the disk scale height.
We calculate the luminosity of the disk in a manner identical to that of the star. We divide the disk surface into cells and calculate the heating effect on each as determined by the incident flux and the angle of incidence $`\alpha `$. We sum the emergent radiation from each cell in the direction of the observer in order to calculate the disk luminosity.
We will refer to both disk models throughout this paper. The prescription of Dubus et al., (1999) will be referred to as the first model, and the radiative transfer formulation will be referred to as the second.
## 3 MODEL RESULTS
In this section we examine the dependence of our model on various parameters. Table 1 contains masses, radii and luminosities for some early-type main sequence (MS) stars and some supergiants. We use these stellar parameters in our model. We keep the primary (BH) mass as an input parameter, and we constrain the scale of the Roche lobes by setting the volume radius of the secondary lobe equal to the radius of the star in Table 1, hence fulfilling the condition for Roche lobe overflow.
We take the X-ray luminosity of the BH to be a constant $`10^{40}`$ergs s<sup>-1</sup> emitted isotropically. The soft X-rays are easily absorbed at the disk surface, whereas the hard X-rays are less easily absorbed but are scattered. They penetrate the star to greater optical depths until the photons are down scattered to lower energies. In our model, we parametrise the absorption coefficients for the hard and soft X-rays by means of two parameters $`k_s`$ and $`k_h`$ and denote the hard and soft components of the X-ray flux as $`S_h`$ and $`S_s`$ respectively. We also define a hardness-ratio parameter $`\xi =S_h/S_s`$. We choose values of $`2.5`$ and $`0.01`$ for the two parameters $`k_s`$ and $`k_h`$ respectively throughout and allow the band boundary of the hard and soft X-ray bands as parameters to be determined. For an input spectrum consisting of a blackbody and a power law component, we find the boundary of the soft and hard band to be $`1.5`$keV. We set the gravity darkening parameter $`\beta `$ to be $`0.25`$, representing a star with a purely radiative outer envelope.
In Figure 1 we show the intensity variation over the surface of an O5V star and disk when we take the BH mass to be $`10`$$`M_{}`$. In Figure 2 we use the same BH mass with a G0I star. We use the quantity $`B(\tau )`$ as a measure of intensity (equation 11), setting $`\tau `$ to $`2/3`$, and we show projections of the star and disk in the orbital plane. The stellar maps show both the irradiative and darkening effects. We use the Dubus prescription of Section 2.1.2 to describe the disk.
We see that the combined surface intensity is significantly higher than would be expected for an unirradiated star. There is however a noticeable difference between the two figures. The stellar intensity of the G0I star increases in the direction of the L1 point, reaching a peak there. On the other hand, in the O5V figure the darkening effects dominate at the L1 point, so that the intensity at that point is less than the surrounding surface. Here we are using a low hardness ratio of $`\xi =0.01`$, and so little flux penetrates to an optical depth of $`\tau =2/3`$. We find that if we increase the hardness ratio the intensity distribution becomes similar to that of the G0I star. As the BH mass increases, the separation increases, the irradiating flux decreases and the intensity distribution over the surface of both stars tends towards that shown in Figure 1(b). Note that we have not included here any shadowing of the accretion disk on the stellar surface, which should magnify any darkening at the L1 point.
In Figure 3(a) we illustrate the change in effective luminosity of an O5V star. We show the $`V`$ band absolute magnitude against the BH mass for an unirradiated star and three different sets of irradiated star calculations. As well as our model described above, we shows the results from an extension to our model where we have included the effect of a disk which is completely opaque to the radiation incident on it and is hence shadowed on the stellar surface. We also show the effects of in addition taking into account the irradiation pressure on the star – we will discuss this in section 3.1. The disk height at the outer disk radius $`R_{out}`$ is taken to be $`0.2R_{out}`$ (de Jong et al.,, 1996). This shielding effect results in a reduced stellar magnitude. In this figure we have set the phase angle to be zero (so the star is in superior conjunction) and set the inclination of the system such that $`\mathrm{cos}i=0.5`$.
Figure 3(a) shows that the heating effect on the star decreases with increasing BH mass, which may be counter-intuitive. This relationship is a consequence of constraining the volume radius of the secondary Roche lobe to the radius of the undistorted star. As the mass ratio decreases, the Roche lobe geometry requires the binary separation $`a`$ to increase. The result is a decrease in the amount of flux incident on the stellar surface.
Figure 3(b) is a sample lightcurve for the O5V star. Here we use a BH mass of $`100`$$`M_{}`$ and an inclination such that $`\mathrm{cos}i=0.5`$. This figure shows both the ellipsoidal variation of an unirradiated star, as well as the combination of both ellipsoidal and irradiative effects. We include a third line showing the magnitude when the irradiated accretion disk is included. We will discuss the effect of the disk on the lightcurve in Section 3.2.
### 3.1 Irradiation pressure effects
We now consider the effect of the X-ray irradiation on the geometrical shape of the secondary. We apply the prescription of Phillips & Podsiadlowski, (2002), which involves a modification of the Roche potential in which the radiation pressure force is parameterised using the ratio of the radiation to the gravitational force. Equation 6 of Phillips & Podsiadlowski, (2002) shows how we can combine the gravitational and radiation pressure forces from the BH as a ’reduced’ gravitational force equal to $`(1\delta )F_{grav}`$, where $`F_{grav}`$ is the gravitational force. $`\delta `$ is the product of a parameter dependent on the X-ray luminosity and binary mass ratio (calculated using equation 25 of Phillips & Podsiadlowski, 2002) and the cosine of the angle between the surface normal and the direction of the flux vector.
After calculating the position of each cell on the Roche surface in our usual way, we calculate a $`\delta `$ value for each cell, replace the gravitational force in the Roche potential with the reduced gravitational force and recalculate the position of each cell. This needs to be repeated for a number of iterations in order to find the solution since the surface normal for each cell changes for each calculation. When the surface has been determined we calculate the heating at each point in our normal way. We take the accretion disk to be opaque to the radiation, so the inner Lagrangian point is shadowed and therefore the radiation pressure will not cause the star to become detached from this point.
The greater the flux incident on the stellar surface, the greater the pressure effect. We therefore see the most significant distortion when the binary separation is at its lowest, which occurs when we use a MS star and a low BH mass. We illustrate this effect in Figure 4, in which we use a $`150`$$`M_{}`$ BH and an O5V star. We show the $`V`$ magnitude dependence on increasing BH mass in Figure 3(a).
Note that we plot values only for BH masses of $`100`$$`1000`$$`M_{}`$. This is because we observe that for a BH mass of less than $`100`$$`M_{}`$ the flux incident on the surface is extremely high and the Phillips & Podsiadlowski, (2002) formulation becomes inappropriate to describe the stellar shape, since the formulation does not allow for any surface motion. In reality the external irradiation will drive circulatory currents in the stellar surface. A full treatment will require hydrodynamical motions to be considered.
### 3.2 Inclusion of the accretion disk
We now investigate further the additional flux from the accretion disk. An increased BH mass leads to a larger binary separation and thus to a corresponding increase in the size of the accretion disk, since the outer disk radius is constrained by the Roche lobe size through tidal effects. The net result is that the disk total luminosity increases with BH mass, and hence compensates for the decreasing stellar total luminosity.
We find that the luminosity class of the irradiated star is the most important factor in determining which component dominates. To illustrate this we show in Figure 5 the absolute magnitude dependence on BH mass for a O5V and a G0I star, along with the corresponding disk magnitudes. We use the Dubus et al., (1999) disk prescription for Figure 5(a,b), and the Wu et al., (2001) disk model for Figure 5(c,d). We find the magnitude of this second disk model is strongly dependent on the hardness parameter $`\xi `$. A low value of $`\xi =0.01`$ produces a disk almost identical to the Dubus disk in terms of magnitude and colour. We use a value of $`\xi =0.1`$ in (c) and (d) to illustrate the effect of a harder X-ray spectrum. In Figure 6 we further illustrate the effect of varying disk hardness on the disk magnitude for different combinations of star and BH.
If we examine the stellar luminosity change as a function of BH mass in Figure 5 first, we note that while the BH + MS star changes by a few tenths of a magnitude over the BH mass range, the BH + supergiant decreases by two magnitudes over that same range. The supergiant has a much larger radius, and so for a low binary separation the flux incident on the stellar surface will be high. However, when the mass ratio is decreased, this larger radius leads to a correspondingly larger binary separation than we see in the MS systems.
If we now examine the disk intensity dependence on BH mass, we find the reverse is true. If the donor is a MS star, the Dubus disk (a) increases in $`V`$ magnitude by more than 2.5 magnitudes over the mass range. When we use harder X-ray radiation (c) the result is a more luminous disk, with approximately the same increase in magnitude over the mass range. In contrast, the Dubus disk accompanying the supergiant (b) increases in $`V`$ magnitude by less than a magnitude over the mass range (the exact increase depends on the method we use to determine $`R_{out}`$). The disk irradiated by the harder X-rays (d) increases by about 1.5 magnitudes. These can be explained by the fact that the large supergiant leads to a large Roche lobe for all BH masses. Hence even a low BH mass results in a very bright disk, and since the temperature of the disk decreases with increasing disk radius, the effect of making a large disk larger still has a smaller effect in terms of total disk luminosity. In constrast, when the companion star is on the main sequence, the smaller size of the system at low BH masses results in a small and faint disk. When the BH mass is increased and the disk grows, the effect on its magnitude is much more significant.
We also show on Figure 5 the $`V`$ magnitude dependence on BH mass of the disk and star combined. It is interesting to note that were we actually observing an O5V system, it would be much easier to constrain the BH mass with the disk component included. The same cannot be said for the system with the G0 supergiant. The gradient of the luminosity change with increasing BH mass is still dictated by the decreasing stellar luminosity, but the curve is rendered shallower by the disk component.
We now investigate the effect of the accretion disk on the amplitude of the lightcurve. We include a line in Figure 3(b) showing the magnitude when the irradiated accretion disk is included. In our thin disk approximation the contribution of the disk will be constant for any phase, so the shape of the lightcurve will not be affected. The exception to this will be when the inclination is such that the disk is partially or fully eclipsed by the star. The relative amplitude of the lightcurve will be affected, depending on the luminosity of the disk. In Figure 3(b) for example, the amplitude decreases from $`0.1`$ V magnitudes to $`0.07`$. For any given star the magnitude of the disk will increase with BH mass, and so the lightcurve relative amplitude will decrease. The lightcurve of the star alone will decrease in amplitude with increasing BH mass due to the decrease in X-ray flux incident on the star – the addition of the disk will reinforce this. We found that the amplitude of the lightcurve for a G0I star with $`L_x=10^{40}`$ergs s<sup>-1</sup>, $`\mathrm{cos}(i)=0.5`$ and $`\xi =0.01`$ drops from $`1.5`$ Mag to $`0.2`$ Mag as we increase the BH mass from $`10`$ to $`1000`$$`M_{}`$. An O5V star under the same set of conditions produces a lightcurve with an amplitude of $`0.17`$ Mag for a BH mass of $`10`$$`M_{}`$. This drops below $`0.1`$ Mag as the mass is increased to $`100`$$`M_{}`$ as shown in Figure 3(b), and at a mass of $`1000`$$`M_{}`$ the lightcurve is dominated by the ellipsoidal variation and has an amplitude of $`0.05`$ Mag.
Figure 6 shows the disk magnitude for a hardness ratio over the range of $`\xi =10^4`$$`10^4`$ using the Wu et al., (2001) formulation. We show the magnitude for a combination of a $`10`$, $`100`$ and $`1000`$$`M_{}`$ BH with a O5V (a) and a G0I (b) star. The change in disk luminosity over this hardness range is large, demonstrating the importance of this factor.
To summarise this section, we find that the stellar luminosity component is at its greatest for low BH masses and the disk component is at its greatest for high BH masses. If we consider separately a MS star, a supergiant star, a disk in a BH/MS system and a disk in a BH/supergiant system, we find the biggest changes in magnitude over the BH mass range occur for a supergiant star or a BH/MS disk. In general then, while the emission will always consist of a disk and a star component, the stellar component will dominate for a MS star / low mass BH combination, and the disk component will dominate in the case of a supergiant / high mass BH. This assumes the X-ray radiation is soft – when the hardness of the X-rays is increased, the contribution of the disk component will increase for all BH masses, and in the supergiant systems in particular we begin to see domination by the disk component over the entire mass range.
### 3.3 Irradiation effects at infrared wavelengths
We now use our model to examine the change in the parameters of a ULX system by making predictions at other wavelengths, extending into the infrared.
We have examined the magnitude change for a star and disk for a wavelength of $`0.54.0\mu `$m, encompassing the $`V`$, $`R`$, $`I`$, $`J`$, $`H`$, $`K`$ and $`L`$ wavebands. We use the Johnson filter convention, with the Kron/Cousins convention for the $`R`$ and $`I`$ bands. We show plots in Figure 7 for the stars in Table 1, using three different BH masses and the Wu et al., (2001) disk model with $`\xi =0.01`$. As noted before, this is essentially interchangeable with the Dubus disk. We use a $`10`$$`M_{}`$ BH in Figure 7(a), a $`100`$$`M_{}`$ BH in Figure 7(b) and a $`1000`$$`M_{}`$ BH in Figure 7(c). We incorporate shadowing of the star by the disk into our stellar irradiation model.
Firstly, we see that there is a very large range in magnitude between these different systems. Secondly, we notice that as the mass of the BH increases, it becomes progressively harder to distinguish between different star/disk combinations with a $`V`$ band observation alone. Thirdly, we see that there is a much more clear distinction when we extend observations to longer wavelengths. Note that there is a clear separation between the MS stars and the supergiants which becomes more apparent as BH mass is increased. This suggests that infrared observations will have more diagnostic power in determining the characteristics of the ULX than observations at optical wavelengths.
## 4 APPLICATION TO ULX X-7 IN NGC 4559
Soria et al., (2005) used HST data to study the optical environment of ULX X-7 in NGC 4559. They found eight possible candidates for the ULX optical counterpart, listing the $`B`$, $`V`$ and $`I_C`$ standard magnitudes for each in table 2 of that paper. In this section we apply our model to this system with the aim of further constraining the candidate population.
We use colour-magnitude diagrams to compare our model predictions with the observations of Soria et al., (2005). We again use the parameters of Table 1. This set is sufficient for us to make generalisations about the spectral type and luminosity class of the donor star. We also add to this set the parameters inferred in Soria et al., (2005) for ‘Star No.1’; the most likely candidate for the optical counterpart. These are determined from the evolutionary tracks of Lejeune & Schaerer, (2001) for a non-irradiated, isolated star with the observed $`B`$, $`V`$ and $`I`$ colours. The mass is found to be $`1530`$$`M_{}`$(we take the mass to be in the middle of this range), the bolometric luminosity is $`1.4\pm 0.2\times 10^5`$$`L_{}`$ and the effective temperature is $`T_{eff}=16000\pm 5000K`$. By additionally using these parameters we can make predictions for a slightly evolved star, based on a current evolutionary model.
In each colour-magnitude diagram we plot the colours and magnitudes of seven of the eight candidate stars. No data are available for star no.7, which fell on a hot HST/WFPC2 pixel. The error on the colour measurement is taken from the errors on the individual magnitude measurements, with the assumption that these errors are not correlated between bands. We then plot a line for each set of stellar parameters which we used in our model. The line shows the effect of varying the BH mass. The mass is varied from $`101000`$$`M_{}`$, and we indicate on each line where the mass is equal to $`10`$, $`100`$ and $`1000`$$`M_{}`$.
Figure 8 compares the stars of Table 1 with the observations for systems for different inclinations, since we have no knowledge of the orientation of the system. We take the inclination to be $`\mathrm{cos}(i)=0.0`$ in Figure 8(a) and $`\mathrm{cos}(i)=0.5`$ in Figure 8(b,c). As before we use two disk models – the first of which describes both the prescription of Section 2.1.2 and the model of Section 2.1.3 when $`\xi =0.01`$. The second disk model uses the formulation of Section 2.1.3 but with $`\xi =0.1`$, giving a brighter disk. We have again taken the phase angle of the binary to be such that the star is at superior conjunction. We have examined other phase angles but we find that this does not affect the general results described in this section. We also find that the effect of including radiation pressure for BH masses of $`100`$$`1000`$$`M_{}`$ is small. We incorporate shadowing of the star by the disk into our stellar irradiation model.
We first examine Figure 8(a). This describes the case where $`\mathrm{cos}(i)=0.0`$, so we see no accretion disk component. Examining the lines for the Table 1 stars, we see a clear distinction between the supergiants and the MS stars. The MS stars occupy the left hand side of the plot only, and show much less change in colour and magnitude over the BH mass range. If we now look at the observed stars, we can see most of the candidate stars seem to fit into one of the two regimes. It is possible that either star 5 or 8 could be an irradiated MS star. Stars 2, 3 and 4 could all be irradiated supergiants (albeit for a very large BH mass). Star 1 could be an irradiated supergiant, or an irradiate, evolved MS star (track 1).
The picture changes significantly when we consider the case where $`\mathrm{cos}(i)=0.5`$. As well as the irradiated star, there is a disk component present. We look first at the fainter disk as described in Figure 8(b). As in Figure 5, we see that the inclusion of the disk results in an increase in $`V`$ magnitude with increasing BH mass. Stars 1, 5 and 8 can still be described by our models.. However, the addition of the disk flux has a large effect on the magnitude of the supergiant and high-mass BH systems. The large luminosity decrease with increasing BH mass is curtailed, and the observations of stars 2, 3 and 4 are too faint for supergiants to be candidates. We additionally examined the case where $`\mathrm{cos}(i)=1.0`$, and found the difference between this and the $`\mathrm{cos}(i)=0.5`$ case to be minor. Again, stars 1, 5 and 8 fit with predictions of a star and disk. Stars 2, 3 and 4 are too faint to fit with the predictions of the model.
We now consider the brighter disk caused by increasing $`\xi `$ by an order of magnitude, as described in Figure 8(c). This disk completely dominates the optical emission – the only role the star plays is to constrain the disk brightness by determining the Roche lobe geometry. We draw the same conclusions from these plots as we did from Figure 8(b) – Stars 1, 5 and 8 fit within the predictions of these star and disk combinations. Stars 2, 3 and 4 do not.
We find therefore that the observed colours of stars 2, 3 and 4 are marginally consistent with irradiated red supergiants when we assume a large BH mass and set $`\mathrm{cos}(i)=0.0`$. In this case there will be temporal variations which will act as the candidate signature. When we move to lower inclinations the increased flux from the accretion disk in our models indicates that the optical counterpart cannot be a red star (redder than $`BV0.6`$), and so we suggest that star 1, 5 or 8 is more likely to be the counterpart. In this case, the optical observations are unable to discriminate between different BH masses. We note that Soria et al., (2005) suggested without taking the effects of a disk or irradiated companion into account that star 1 is the most likely candidate for the optical counterpart of X-7 in NGC 4559. In addition, taking the results of Section 3.3 into account, we note that the irradiated system could be very bright in the IR. Observations at these wavelengths could be useful in determining the counterpart.
## 5 CONCLUSIONS
We have constructed a model describing the heating effect of a ULX on a Roche lobe filling companion star. We plan to apply this model to the problem of positively identifying ULX optical counterparts, and then use observations of that counterpart to constrain the BH mass and the nature of the companion.
Our model uses a radiative transfer formulation to account for the X-ray nature of the incident radiation and the distribution of the re-radiated emission. We incorporate the distorted Roche lobe filling geometry of the star and account for the limb and gravity darkening effects. We also include the additional luminosity from an irradiated accretion disk, for two different disk models.
We have illustrated how our model can be applied in a number of different ways. First, we assume that the donor star is filling its Roche lobe, as suggested by the high X-ray luminosity. We have then shown that for a given spectral type, the effects of irradiation decrease as the mass of the black hole in the system (and hence binary separation) is increased. Measurement of the amplitude and period of the lightcurve of a counterpart will allow us to determine many of the free parameters of the system – we have shown how our model incorporates both the ellipsoidal and irradiative effects and that we can separate out these components. We have discussed how the accretion disk can be the dominant component in some systems and have shown that measurement of the periodicity of the binary can distinguish between these regimes. We have shown how the importance of the disk is linked with the luminosity class of the companion star, and predict that observations at infrared wavelengths makes it easier to distinguish between different systems without the need for temporal observations.
Finally, we have applied our model to a set of observations of potential candidates for the counterpart of ULX X-7 in NGC 4559. While the available data is not sufficient to draw any firm conclusions, we see that if the observations contain any appreciable accretion disk component then only three of the possible candidates fit our model. These candidates imply the donor star is an early-type MS star. Supergiant companions are only possible for a high inclination and a BH mass of $`1000`$$`M_{}`$.
## ACKNOWLEDGEMENTS
We would like to thank the referee (G. Dubus) for his comments, which led to a number of important improvements to this paper. We would also like to thank Richard Mushotzky for interesting discussions.
## Appendix A The radiative transfer formulation
We consider the effects of radiative transport in the irradiated surface under X-ray illumination. We consider a plane-parallel model and adopt the radiative transport formulation of Milne, (1926) and Wu et al., (2001). Milne’s original formulation was for incident radiation at optical wavelengths and cooling via emission of optical radiation. This was modified by Wu et al., (2001) to account for incident radiation at X-ray wavelengths. The modification is important because soft X-rays will be absorbed close to the surface of the star by neutral and weakly ionized matter via bound-free transitions. Hard X-rays will only be attenuated at great depths when the matter density is significantly higher. The soft and hard X-ray components will subsequently have higher or larger opacities than for the optical radiation. The formulation is linear and therefore the principle of superposition is applicable. This allows us to derive the total emission using the irradiated and non-irradiated components.
We take the incident radiation to be parallel beams of soft and hard X-rays, with effective fluxes $`\pi S_s`$ and $`\pi S_h`$ per unit area normal to the beams, and making an angle $`\alpha `$ to the normal to the stellar surface. The absorption coefficients of the soft and hard X-rays are $`k_s\kappa `$ and $`k_h\kappa `$ respectively, where $`\kappa `$ is the absorption coefficient of the optical radiation ($`k_s>1`$ and $`k_h<1`$ defines our soft/hard X-ray convention in this study, following Wu et al., 2001).
The total blackbody radiation flux is a combination of a component $`B_x(\tau )`$ as a result of irradiative heating by the incident X-rays and the component of the radiation from the star in the absence of irradiative heating $`B_s(\tau )`$, where $`\tau `$ is the optical depth. The irradiative heating component $`B_x(\tau )`$ was solved in the limit of a semi infinite plane by the method of successive approximations and was found to be
$$B_x(\tau )=ab_s\mathrm{exp}(k_s\tau \mathrm{sec}\alpha )b_h\mathrm{exp}(k_h\tau \mathrm{sec}\alpha )$$
(3)
in the second approximation (Wu et al.,, 2001), where $`a`$, $`b_s`$ and $`b_h`$ are constants to be determined by the boundary conditions. For a semi-infinite slab opaque at optical wavelengths, the emergent optical/IR radiation in the direction $`\theta `$ is the Laplace transform of $`B_x(\tau )`$
$`I(0,\mu )`$ $`=`$ $`\underset{\tau _{tot}\mathrm{}}{lim}\left[{\displaystyle _0^{\tau tot}}𝑑\tau B_x(\tau )\mathrm{exp}(\tau /\mathrm{cos}\theta )\right]`$ (4)
$`=`$ $`ab_s𝒜_s\left[𝒜_s+\mu \right]^1b_h𝒜_h\left[𝒜_h+\mu \right]^1`$
where $`𝒜_s`$ and $`𝒜_h`$ are $`\mathrm{cos}\alpha /k_s`$ and $`\mathrm{cos}\alpha /k_h`$ respectively, and $`\mu =\mathrm{cos}\theta `$.
Here $`a`$,$`b_s`$ and $`b_h`$ are obtained by solving the radiative-equilibrium and radiative transfer equations for the conditions $`b_s0`$ when $`S_s0`$ and $`b_h0`$ when $`S_h0`$:
$$a=\frac{1}{2}\left[k_sS_s𝒜_sf_s(\alpha )+k_hS_h𝒜_hf_h(\alpha )\right]$$
(5)
$$b_s=\frac{1}{2}k_sS_s\left[𝒜_s\frac{1}{2}\right]f_s(\alpha )$$
(6)
$$b_h=\frac{1}{2}k_hS_h\left[𝒜_h\frac{1}{2}\right]f_h(\alpha )$$
(7)
where the functions $`f_s(\alpha )`$ and $`f_h(\alpha )`$ are given by
$`f_s(\alpha )=\left[1𝒜_s+𝒜_s\left(𝒜_s{\displaystyle \frac{1}{2}}\right)\mathrm{ln}(1+k_s\mathrm{sec}\alpha )\right]^1`$ (8)
$`f_h(\alpha )=\left[1𝒜_h+𝒜_h\left(𝒜_h{\displaystyle \frac{1}{2}}\right)\mathrm{ln}(1+k_h\mathrm{sec}\alpha )\right]^1`$ (9)
The hardness of the X-ray source is defined in terms of a hardness parameter $`\xi =S_h/S_s`$, with the total X-ray flux $`S_x=S_s+S_h`$. By expressing $`B_x(\tau )`$ in terms of this parameter we obtain
$`B_x(\tau )={\displaystyle \frac{1}{2}}S_x\{k_sf_s(\alpha )\left({\displaystyle \frac{\xi }{1+\xi }}\right)[𝒜_s(𝒜_s{\displaystyle \frac{1}{2}})e^{\tau /𝒜_s}]`$ (10)
$`+k_hf_h(\alpha )\left({\displaystyle \frac{1}{1+\xi }}\right)[𝒜_h(𝒜_h{\displaystyle \frac{1}{2}})e^{\tau /𝒜_h}]\}.`$
As the radiative-transfer equations are linear, the local temperature stratification is given by
$$T(\tau )=\left\{\frac{\pi }{\sigma }[B_x(\tau )+B_s(\tau )]\right\}^{1/4}\left(\frac{\pi }{\sigma }B(\tau )\right)^{1/4}.$$
(11)
The surface temperature of a star is effectively the temperature at an optical depth of $`\tau =2/3`$. Hence, we find when it is viewed at a given inclination angle $`\alpha `$, the effective temperature of the surface under irradiation is
$$T_{eff}=\left\{\frac{\pi }{\sigma }B_x(2/3)+T_{unirr}^4\right\}^{1/4}$$
(12)
where $`T_{unirr}`$ is the effective temperature in the absence of any irradiation.
|
warning/0506/cond-mat0506429.html
|
ar5iv
|
text
|
# A critical reexamination of the non-reciprocal x-ray gyrotropy in V2O3.
## I Introduction
The crystal and electronic structure of V<sub>2</sub>O<sub>3</sub> has been the subject of an intensive experimental and theoretical study throughout the 70’s that led to the identification of V<sub>2</sub>O<sub>3</sub> as the prototype of Mott-Hubbard systems. This compound shows a metal-insulator transitions, due to the interplay between band formation and electron Coulomb correlation, from a paramagnetic metallic (PM) phase at room temperature to a paramagnetic insulating (PI) phase at higher temperature ($`500K`$) and to an antiferromagnetic insulating (AFI) phase at lower temperature ($`T_c150K`$ for the stoichiometric compound). Associated with this latter there is also a structural transition from a corundum cell (rhombohedral system) to a monoclinic one (see Fig. 1).
In this investigation a correct description of the properties of the AFI ground state has been of paramount importance for both the physical understanding of the insulating phase and for throwing light on the mechanism of the metal insulator transition. Indeed Castellani, Natoli and Ranninger (CNR), were the first to realize the importance of the orbital degrees of freedom for the explanation of the peculiar antiferromagnetic spin structure that breaks the trigonal symmetry of the corundum lattice in the high temperature phase. Evidence for their role was subsequently found in a series of inelastic neutron scattering experiments by Wei Bao et al. and NMR nuclear spin relaxation measurements in the high temperature paramagnetic phases. The ensuing physical picture was one in which the orbital degrees of freedom are frozen in the AFI phase, giving rise to the peculiar spin structure, and are responsible for the short range magnetic fluctuations in the high temperature phases, providing much of the necessary entropy content for the transition to occur. Even though the Hubbard model used by CNR was based on a spin S = 1/2 for the $`V^{3+}`$ ion in a doubly degenarate orbital $`e_g`$ state, the subsequent picture based on spin S=1, implied by some experiments and worked out by Mila et al. and Di Matteo et al., did not change the above physical description, but only redefined the basic building blocks of the problem. Indeed in the S=1 picture the elementary units are the vertical biatomic molecules in a doubly degenerate ground state, strongly coupled to spin S=2, and weakly interacting via electronic kinetic hopping in the basal planes.
The introduction of the extra degrees of freedom in the problem might be expected to change the magnetic space group of the ordered AFI phase as inferred from the lattice geometry and spin moments only. As explained in the following section, the magnetic space group of the monoclinic cell with the atomic sites occupied by the magnetic moments in the way found by Moon is the magnetic space group $`P2/a+\widehat{T}\{\widehat{E}|t_0\}P2/a`$, containing explicitly the time reversal operator $`\widehat{T}`$ and the inversion $`I`$ ($`\widehat{E}`$ is the identity operator, $`t_0`$ is the body-centered translation).
In this context the observation of a nonreciprocal x-ray linear dichroism measured at the K-edge of the Vanadium atom in the AFI phase of V<sub>2</sub>O<sub>3</sub> seemed to shed new light on the properties of the insulating ground state, since the effect can only be observed when neither the inversion nor the time reversal are separately symmetry operations of the system, thus implying that the system should be magnetoelectric. Even though not explicitly stated in Ref. \[\] this finding was pointing to a role of the orbital degrees of freedom in lowering the magnetic symmetry of V<sub>2</sub>O<sub>3</sub>, all the more that an independent observation of the forbidden (111)<sub>m</sub> reflection in the $`V`$ K-edge anomalous diffraction had been interpreted as implying some kind of orbital ordering.
However a closer consideration of all the pieces of evidence of the problem revealed some inconsistencies in the overall interpretative frame, as discussed at length in Ref. \[\]. First of all the theoretical model used in Ref. \[\] to interpret the orbital ordering was based on the old CNR S=1/2 model, which was later shown to be inadequate to describe the physics of V<sub>2</sub>O<sub>3</sub>. In the new correlated model based on spin S=1 the orbital ordering in the AFI ground state was again found compatible with the $`P2/a+\widehat{T}\{E|t_0\}P2/a`$ group. In the meantime growing evidence was accumulating that the forbidden (111)<sub>m</sub> reflection is of pure magnetic origin and has nothing to do with orbital ordering. Last but not least, a new direct measurement of the magnetoelectric effect in V<sub>2</sub>O<sub>3</sub> has provided a negative result, confirming the previous result by Astrov.
Concomitant to all this, our efforts to understand at a deeper level the implications of the nonreciprocal x-ray gyrotropy experiment led us to note some paradoxical consequences of this result and to look for alternative interpretations. It is the purpose of this paper to present the conclusions of this investigation and to point up at least the need to repeat the experiment under better controlled conditions.
This paper is organized as follows. In section II, we discuss in some detail, for the convenience of the reader, the structural and magnetic properties of the system in its monoclinic phase and derive the corresponding magnetic space group, listing all the magnetoelectric subgroups. In section III we derive explicitly the response functions corresponding to a linear dichroism experiment both in the dipole-dipole (E1-E1) and the dipole-quadrupole (E1-E2) channel. We show that the corresponding signal is proportional to the average of certain operators in the final scattering state of the excited photoelectron, having definite transformation properties under inversion and time reversal, and discuss in the light of the possible magnetic space groups the conditions under which such a signal is different from zero. We find that contrary to what implied in Ref. \[\] one expects a significant signal from the E1-E1 channel, which is inversion and time reversal even. We further argue that its existence is a necessary consequence of the Templeton scattering at the (10$`\overline{1}`$)<sub>m</sub> forbidden reflection in the $`\sigma \sigma `$ channel observed by Paolasini et al. in the monoclinic phase of the same compound. We estimate the dichroic signal in the frame of multiple scattering (MS) theory, which should be adequate to describe transitions to the itinerant conduction $`p`$ states in V<sub>2</sub>O<sub>3</sub>. Surprisingly enough its shape is quite similar to the “nonreciprocal” gyrotropy signal reported in Ref. \[\]. We speculate on such similarity and give arguments whereby the reversal of the signal under magnetic field could in fact be due to an uncontrolled geometrical effect. In section IV we summarize our conclusions.
## II The monoclinic structure of V<sub>2</sub>O<sub>3</sub>.
Pure V<sub>2</sub>O<sub>3</sub>, as well as the 2.8% chromium doped sample used in the x-ray experiments mentioned in the introduction, is a paramagnet at room temperature that crystallize in the trigonal system. At T$`{}_{c}{}^{}`$ 150 K (180 K for the doped compound), it undergoes a structural transition to the monoclinic system and an antiferromagnetic order sets in. This doubles the unit cell to four formula units, instead of the two formula units of the trigonal cell. Moreover the transition breaks the trigonal symmetry in such a way that one of the three originally equivalent vanadium-vanadium bonds in the hexagonal plane of the corundum cell becomes ferromagnetic and longer of about $`4\%`$ than the other two, which are antiferromagnetic. Because of this, three possible monoclinic twin domains exist, depending on which of the three equivalent bonds stretches and becomes ferromagnetic. They are related by a rotation of 120<sup>o</sup> around the corundum c<sub>H</sub>-axis. Experimentally one finds that V<sub>2</sub>O<sub>3</sub> has the tendency to crystallize prevalently in one of the three monodomains, more than in a statistical average of the three and that the choice among them is more or less random, depending on the size, shape and boundary conditions of the crystal.
We give here a detailed description of the monoclinic cells and of its lattice and magnetic symmetries, as we shall use them in the subsequent analysis of the structure factor.
The crystal positions are taken from Dernier and Marezio, and the magnetic structure of the monoclinic cell from Moon later confirmed by Wei Bao.
The monoclinic cell, shown in Fig. 1, belongs to the body-centered crystallographic space group I2/a. Using the reference frame and the numbering of vanadium atoms of Fig. 1, we can divide the eight atoms of the monoclinic cell into two groups of four, with opposite orientation of the magnetic moment, $`V_1=(1/2u,v,w),V_2=(1/2+u,v,w),V_3=(u,v,w),V_4=(u,v,w)`$, and $`V_1^{}`$, $`V_2^{}`$, $`V_3^{}`$, $`V_4^{}`$, with coordinates obtained by adding the vector (1/2, 1/2, 1/2) to the first group. Here $`u=0.3438,v=0.0008,w=0.2991`$ are the fractional coordinates of the atoms in unit of the monoclinic axis. Note that, neglecting the magnetic moments, the two groups of atoms with their oxygen environments are translationally equivalent.
From these data we can infer that the magnetic space group of the monoclinic cell is the magnetic space group $`P2/a+\widehat{T}\{E|t_0\}P2/a`$, where $`\widehat{T}`$ is the time-reversal operator and the monoclinic group $`P2/a`$ contains the identity $`\widehat{E}`$, the inversion $`\widehat{I}`$, the two-fold rotation about the monoclinic b<sub>m</sub>-axis $`\widehat{C}_{2b}`$ and the reflection $`\widehat{m}_b`$ with respect to the plane perpendicular to this axis. Of course, the appropriate translation is associated to each of these operators. Choosing the origin of the system as in Fig. 1, we have the following symmetry operations:
$`\begin{array}{ccc}1)\widehat{E},\widehat{I}\hfill & \hfill & \mathrm{No}\mathrm{translation}\hfill \\ 2)\widehat{C}_{2b},\widehat{m}_b\hfill & \hfill & \frac{1}{2}(\stackrel{}{b}_m+\stackrel{}{c}_m)\hfill \\ 3)\widehat{T},\widehat{T}\widehat{I}\hfill & \hfill & \frac{1}{2}(\stackrel{}{a}_m+\stackrel{}{b}_m+\stackrel{}{c}_m)\hfill \\ 4)\widehat{T}\widehat{C}_{2b},\widehat{T}\widehat{m}_b\hfill & \hfill & \frac{1}{2}\stackrel{}{a}_m.\hfill \end{array}`$ (5)
Notice that, if one considers only the lattice and magnetic moments, the presence of the time-reversal symmetry is a necessary consequence of the translational equivalence of the two groups of atoms. Now, as anticipated in the introduction, one of the consequences of the observation of non-reciprocal effects by linear dichroism is the indication that other degrees of freedom may play a role in determining the symmetry of the insulating ground state. In fact a necessary condition to detect the non-reciprocal gyrotropy tensor is that the system be magnetoelectric, this fact implying in turn the breakdown of time-reversal and inversion symmetry. Orbital degrees of freedom are the natural candidates in the physics of V<sub>2</sub>O<sub>3</sub> to obtain such a symmetry reduction, although new, still unsuspected, ingredients might come into play. One other possibility, already debated in the literature, yet very controversial, concerns the presence of an out-of-plane (a<sub>m</sub>-c<sub>m</sub> plane) component of the magnetic moment directed along the $`y`$ axis.. This fact would at least break the glide-plane symmetry. Three independent neutron scattering experiments were not able either to discard or accept this hypotesis. However this out of plane component would contradict the argument given in Refs. \[\] to explain the absence of any dipole signal at the (111)<sub>m</sub> reflection in the energy region of the conduction band of $`p`$ symmetry, that can only be explained by the presence of the glide-plane symmetry together with the fact that $`<L_y>=0`$. For this reason, we can conclude that x-ray measurements point toward the absence of an out-of-plane component of the magnetic moment.
Here we list for completeness all the possible space magnetic subgroups of $`P2/a+\widehat{T}\{E|t_0\}P2/a`$ that do admit magnetoelectricity:
$`\begin{array}{ccc}1)(\widehat{E},\widehat{T}\widehat{I},\widehat{T}\widehat{C}_{2b},\widehat{m}_b)\hfill & P2^{}/a\hfill & \\ 2)(\widehat{E},\widehat{T}\widehat{I},\widehat{C}_{2b},\widehat{T}\widehat{m}_b)\hfill & P2/a^{}\hfill & \\ 3)(\widehat{E},\widehat{T}\widehat{I})\hfill & & \\ 4)(\widehat{E},\widehat{T}\widehat{C}_{2b})\hfill & & \\ 5)(\widehat{E},\widehat{C}_{2b})\hfill & & \\ 6)(\widehat{E},\widehat{T}\widehat{m}_b)\hfill & & \\ 7)(\widehat{E},\widehat{m}_b)\hfill & & \\ 8)(\widehat{E})\hfill & & \end{array}`$ (14)
We have indicated in parenthesis the symmetry operations of the associated point groups and for simplicity we have not written the corresponding translations which are the same as above. Note that all the two elements subgroups actually belong to the triclinic system. On the basis of these groups we shall discuss in the next section the conditions for the observation of linear dichroism in absorption.
## III Normal and nonreciprocal linear dichroism
Linear dichroism in the x-ray range is the differential spectrum obtained by subtracting two core absorption spectra taken with orthogonal linear polarizations. According to whether it is of dipole-dipole (E1-E1) or dipole-quadrupole (E1-E2) origin we shall distinguish, for convenience and reasons to become apparent later, between respectively: Normal (NXLD) and Nonreciprocal (NRXLD) X-ray Linear Dichroism. The two contributions can obviously be both present in the same spectrum. The experiment we are interested in was performed with the x-ray beam directed along the hexagonal c<sub>h</sub>-axis, or (20$`\overline{2}`$)<sub>m</sub> in monoclinic Miller indices, with the two orthogonal directions of the linear polarizations lying in the hexagonal plane without specification of their orientation with respect to the in plane crystallographc axes.
### A Dichroic operators and related sum rules
As is well known, at the Vanadium K-edge the x-ray absorption cross section is given by:
$$\sigma _i=4\pi ^2\alpha \mathrm{}\omega \underset{n}{}|\mathrm{\Psi }_n^{(i)}|\widehat{O}|\mathrm{\Psi }_0^{(i)}|^2\delta (\mathrm{}\omega (E_nE_0))$$
(15)
The operator $`\widehat{O}\widehat{ϵ}\stackrel{}{r}(1+\frac{i}{2}\stackrel{}{k}\stackrel{}{r})`$ is the usual matter-radiation interaction operator expanded up to the quadrupolar term, with the usual notation for the photon polarisation $`\widehat{ϵ}`$ and the wave vector $`\stackrel{}{k}`$. $`\mathrm{\Psi }_0^{(i)}`$ ($`\mathrm{\Psi }_n^{(i)}`$) is the ground (excited) state of the crystal, $`E_0`$ ($`E_n`$) its energy and the index $`i`$ indicates the lattice site of the Vanadium photoabsorbing atom. The sum is extended over all the excited states of the system and $`\mathrm{}\omega `$ is the energy of the incoming photon, while $`\alpha `$ is the fine-structure constant.
Before analyzing Eq. (15) on the basis of the crystal space magnetic symmetry, we derive explicitly the cross section and the consequent sum rule for non-reciprocal and normal linear dichroism, on the basis of the formalism developed in Ref. \[\]. In both cases linear dichroism is defined as:
$$\sigma _i^{LD}\sigma _i(ϵ_a)\sigma _i(ϵ_b)$$
(16)
where $`ϵ_a`$ and $`ϵ_b`$ are ususally chosen orthogonal. With reference to Sec. 7 of Ref. \[\], we can write the signal up to the E1-E2 channel as the scalar product of two tensors, representing respectively the properties of the light (T) and of the matter (M):
$`\sigma _i^{LD}{\displaystyle \underset{p,q}{}}()^{1+p+q}[T_q^{(p)}(ϵ_a,ϵ_a,k)M_q^{(p)}`$ (17)
$`T_q^{(p)}(ϵ_b,ϵ_b,k)M_q^{(p)}]`$ (18)
where p=1,2,3. The explicit expression of the tensors $`T_q^{(p)}`$ is given in Sec. 7 of Ref. \[\]. Moreover, in terms of the tensors expressing the properties of the matter (M, for E1-E1 and $`\widehat{M}`$ for E1-E2) we obtain:
$$\sigma _i^{NXLD}\frac{1}{\sqrt{5}}(M_2^{(2)}+M_2^{(2)})$$
(19)
$`\sigma _i^{NRXLD}((\widehat{M}_2^{(3)}+\widehat{M}_2^{(3)}c.c.)+`$ (20)
$`{\displaystyle \frac{1}{\sqrt{2}}}(\widehat{M}_2^{(2)}\widehat{M}_2^{(2)}c.c.))`$ (21)
In both sets of equations we have omitted for simplicity the index $`i`$ in the right hand side. To evaluate the matter tensor we shall adopt the one particle approach in the framework of multiple scattering theory. However, as shown in Ref. \[\], all the expressions remain valid in a second quantization scheme, if one takes the multiple scattering functions as the one particle basis for the matter operators. Around each site the photoelectron wave function can be expanded in spherical harmonics and spin states as:
$$|\mathrm{\Psi }_i(\stackrel{}{k}_e)>=\underset{l,m,\sigma }{}B_{lm\sigma }^i(\stackrel{}{k}_e)R_{lm\sigma }^i(r;E)Y_{lm}(\widehat{r})\chi _\sigma $$
(22)
where $`\stackrel{}{k}_e`$ is the photoelectron wave vector associated to the final scattering state with kinetic energy $`Ek_e^2`$ and $`\chi _\sigma `$ is the spin state. Using this expression it is possible to write the tensors $`M_q^{(p)}`$ in terms of the scattering amplitudes $`B_{lm\sigma }^i(\stackrel{}{k})`$. In the case of interest (K-edge) $`l`$ is either 1 or 2 so that we find in the geometrical setting of the experiment of Ref. \[\]:
$`\sigma _i^{NXLD}`$ $`=`$ $`4\pi ^2\alpha \mathrm{}\omega M_l^2(E){\displaystyle 𝑑\widehat{k}_e\underset{m,\sigma }{}|m|(B_{1m\sigma }^i)^{}B_{1,m\sigma }^i}`$ (24)
$`{\displaystyle 𝑑\widehat{k}_e<\mathrm{\Psi }_i(\stackrel{}{k}_e)|(L_x^{}^2L_y^{}^2)|\mathrm{\Psi }_i(\stackrel{}{k}_e)>}`$
$`\sigma _i^{NRXLD}=4\pi ^2\alpha \mathrm{}\omega M_l^2(E){\displaystyle 𝑑\widehat{k}_e}`$ (25)
$`{\displaystyle \underset{m,\sigma }{}}|m|(B_{2m\sigma }^i)^{}B_{1,m\sigma }^ic.c.`$ (26)
$`{\displaystyle 𝑑\widehat{k}_e<\mathrm{\Psi }_i(\stackrel{}{k}_e)|(L_x^{}^2L_y^{}^2)\mathrm{\Omega }_z|\mathrm{\Psi }_i(\stackrel{}{k}_e)>}c.c.`$ (27)
where, indicating by $`R_{mt}`$ the muffin-tin radius, the proportionality factor is
$$\left[\frac{M_l(E)}{_0^{R_{mt}}r^2𝑑rR_l^2(r;E)}\right]^2.$$
In the previous equations we have ignored the weak $`m\sigma `$ dependence of the radial wave functions, as justified for deep core (1s) electron transitions. We have introduced the dipole radial transition matrix element, $`M_l(E)`$ and integrated over all the direction of the escaping photoelectron ($`\widehat{k}_e`$) at a fixed photon energy, as appropriate for absorption. Moreover we have used the equality $`(L_x^2L_y^2)=(L_+^2+L_{}^2)`$ and the fact that the magnetic quantum number $`m`$ runs from –1 to 1. The labels $`x^{}`$ and $`y^{}`$ indicate two arbitrary orthogonal directions of the two polarizations $`ϵ_a`$ and $`ϵ_b`$ in the hexagonal plane. As discussed in Ref. \[\] the energy integrated spectrum provides the expectation value of the same operators over the ground state, a situation that is in common with all kinds of dichroisms.
Equations (24) and (27) show that linear dichroism at the energy $`\mathrm{}\omega `$ of the incoming photon measures the expectation value of a physical operator over the final scattering state corresponding to that energy. In particular, this operator is proportional to $`L_x^{}^2L_y^{}^2`$ in the dipole-dipole channel and to $`(L_x^{}^2L_y^{}^2)\mathrm{\Omega }_zc.c.`$ in the dipole-quadrupole channel. In the last expression $`\mathrm{\Omega }_z(\stackrel{}{L}\times \widehat{n}\widehat{n}\times \stackrel{}{L})_z`$, where $`\stackrel{}{L}`$ is the usual angular momentum operator and $`\widehat{n}`$ the radial unit vector operator. It is the raising/lowering operator for the orbital quantum number $`l`$ when applied to a spherical harmonic with the same $`m`$ projection, and is known as the anapole or toroidal moment operator.
Note that the linear dichroism signal is time-reversal even and inversion even in the dipole-dipole channel, while it is time-reversal odd and inversion odd in the dipole-quadrupole channel, as expected on the basis of general considerations.
### B Extinction rules for linear dichroism.
In the monoclinic crystal, the total absorption signal is obtained by summing over all the eight inequivalent vanadium sites $`I_C=_{i=1}^8\sigma _i`$ in the unit cell. Therefore, indicating with the label $`LD`$ both types of dichroism we obtain
$`I_{LD}={\displaystyle \underset{i=1}{\overset{8}{}}}\sigma _i^{LD}`$ (28)
By exploiting the symmetry operations of the system, we can now relate one another the various $`\sigma _i^{LD}`$ in the unit cell, exactly as done for the anomalous scattering amplitude in Ref. \[\]. Absorption is in fact the imaginary part of the forward scattering amplitude. A symmetry operation then acts both on the site index and on the operator appearing in the averages (24) and (27), assuming that the final scattering state $`|\mathrm{\Psi }_i(\stackrel{}{k}_e)>`$ does not break the symmetry of the ground state. This latter assumption is physically plausible since in the one electron approximation the rest of the system is seen by the excited photoelectron as a static scattering potential. Note that the symmetry considerations in this subsection do not depend on the one particle approximation introduced in the previous subsection, but hold a more general validity, relying only on the symmetry group of V<sub>2</sub>O<sub>3</sub>..
The site transformation rules for the largest magnetic group $`P2/a+\widehat{T}\{E|t_0\}P2/a`$ are given in the following table:
$`\begin{array}{ccccccccc}& & & & & & & & \\ \widehat{E}:\hfill & \sigma _1& \sigma _2& \sigma _3& \sigma _4& \sigma _1^{}& \sigma _2^{}& \sigma _3^{}& \sigma _4^{}\\ & & & & & & & & \\ \widehat{I}:\hfill & \sigma _2& \sigma _1& \sigma _4& \sigma _3& \sigma _2^{}& \sigma _1^{}& \sigma _4^{}& \sigma _3^{}\\ & & & & & & & & \\ \widehat{C}_{2b}:\hfill & \sigma _4^{}& \sigma _3^{}& \sigma _2^{}& \sigma _1^{}& \sigma _4& \sigma _3& \sigma _2& \sigma _1\\ & & & & & & & & \\ \widehat{m}_b:\hfill & \sigma _3^{}& \sigma _4^{}& \sigma _1^{}& \sigma _2^{}& \sigma _3& \sigma _4& \sigma _1& \sigma _2\\ & & & & & & & & \\ \widehat{T}:\hfill & \sigma _1^{}& \sigma _2^{}& \sigma _3^{}& \sigma _4^{}& \sigma _1& \sigma _2& \sigma _3& \sigma _4\\ & & & & & & & & \\ \widehat{T}\widehat{I}:\hfill & \sigma _2^{}& \sigma _1^{}& \sigma _4^{}& \sigma _3^{}& \sigma _2& \sigma _1& \sigma _4& \sigma _3\\ & & & & & & & & \\ \widehat{T}\widehat{C}_{2b}:\hfill & \sigma _4& \sigma _3& \sigma _2& \sigma _1& \sigma _4^{}& \sigma _3^{}& \sigma _2^{}& \sigma _1^{}\\ & & & & & & & & \\ \widehat{T}\widehat{m}_b:\hfill & \sigma _3& \sigma _4& \sigma _1& \sigma _2& \sigma _3^{}& \sigma _4^{}& \sigma _1^{}& \sigma _2^{}\end{array}`$
Therefore, eg, $`\widehat{T}\sigma _1^{NRXLD}=\sigma _1^{}^{NRXLD}`$ and similarly for all the other cases.
Using these relations we can then express $`I_{LD}`$ in term of one or two absorption site according to the symmetry group chosen. We consider the three cases of the groups: $`P2/a+\widehat{T}\{E|t_0\}P2/a`$, $`P2^{}/a`$ and $`P2/a^{}`$ which are pertinent to the monoclinic phase.
In the case of the original group $`P2/a+\widehat{T}\{E|t_0\}P2/a`$ (non ME), the global signal can be expressed in terms of only one independent center of absorption. Thus Eq. (28) can be written as:
$$I_{LD}=(1+\widehat{T}\widehat{I})(1+\widehat{T}\widehat{m}_b)(1+\widehat{T})\sigma _4^{LD}$$
(29)
In the case of a ME subgroups we use only the part of the table referring to the operators of each subgroup so that
$$I_{LD}(P2^{}/a)=(1+\widehat{T}\widehat{I})(1+\widehat{T}\widehat{m}_b)(\sigma _4^{LD}+\sigma _4^{}^{LD})$$
(30)
and
$$I_{LD}(P2/a^{})=(1+\widehat{T}\widehat{I})(1+\widehat{m}_b)(\sigma _4^{LD}+\sigma _4^{}^{LD})$$
(31)
in terms of two independent centers of absorption. In these formulas it is now intended that the group operators have already acted on the site but not on the physical observables.
In all cases it is straightforward to derive that no circular dichroism can be detected for the system, as experimentally found in Ref. \[\], since all the expressions are proportional to $`(1+\widehat{T}\widehat{I})`$ and the product of time-reversal and inversion symmetries is always -1 for circular dichroism ($`\widehat{I}=1`$ and $`\widehat{T}=1`$ in the E1-E1 channel and $`\widehat{I}=1`$ and $`\widehat{T}=1`$ in the E1-E2 channel). As expected, the three magnetic groups present different extinction rules for NRXLD. Since in this case the relevant operator is time-reversal and inversion odd, the signal is zero for the non ME group Eq. (29), due to the term $`(1+\widehat{T})`$, and different from zero for the other two, since $`\widehat{T}\widehat{I}=+1`$.
On the contrary there is no case in which the linear dichroism in the E1-E1 channel (NXLD) is forbidden by the crystal structure, because in this channel both the action of the time-reversal and the action of the inversion operators give +1. Thus, since the magnetic symmetry around each vanadium ion is $`\widehat{C}_1`$ (ie, no symmetry at all) a non zero atomic NXLD is in principle expected. Such a dichroic signal is zero only when each pair of vanadium ions contribute to the total intensity with an opposite sign. This latter occurrence is possible only for the particular direction at an angle $`\pi /4`$ with respect to the glide plane, in which case for each pair of vanadium ions related by this symmetry operation (eg 1 and 3’, 2 and 4’, 3 and 1’, 4 and 2’, as inferred by the transformation table given above) $`L_x^2L_y^2`$ and viceversa, leading to the extinction of the total crystal signal as seen from Eqs. (29), (30) and (31). Note that any reduction of the symmetry to a group that does not contain the glide plane (either multiplied by the time-reversal or not) implies that the NXLD signal is never zero. Note also that the non-reciprocal signal, proportional to $`(L_x^2L_y^2)\mathrm{\Omega }_zh.c.`$, at the particular angle $`\pi `$/4 vanishes if the magnetic group contains the glide-plane alone as a symmetry operation (like $`P2/a^{}`$) and may be different from zero with groups non containing it (like $`P2^{}/a`$).
We dwelt a bit on this analysis, since it highlights an unexpected feature in the nonreciprocal XLD spectra as presented in Ref. \[\]. In fact, since one cannot disentangle the E1-E1 from the E1-E2 channel one would expect the superposition of two linear dichroic signals in any experimental spectrum, one of which is time-reversal and inversion even (coming from the E1-E1 transition) and a second, which is time-reversal and inversion odd (coming from the E1-E2 transition). As a consequence the residual signal obtained by taking the half sum of the two nonreciprocal XLD spectra, respectively with the external magnetic field oriented parallel/antiparallel to the hexagonal $`c_H`$ axis, (ie \[XLD(H<sup>+</sup>) + XLD(H<sup>-</sup>)\]/2), should give the normal (time-reversal and inversion even) linear dichroism signal, as correctly recognized by the authors in Ref. \[\]. From Fig. 2 of the same reference one can infer that this residual signal is about ten times smaller than the nonreciprocal signal (which is itself between 0.0-1.0 % of the main absorption), in practice nearly coinciding with the background noise, ’due perhaps to the slightly distorted monoclinic structure’, in their own words.
We show in the following section that a multiple scattering realistic simulation of the NXLD in the monoclinic lattice provides a signal which can be up to 4% of the main absorption. The presence of such a signal is also corroborated by the sizable Templeton scattering at the (10$`\overline{1}`$)<sub>m</sub> forbidden reflection in the $`\sigma \sigma `$ channel observed by Paolasini et al. in the monoclinic phase. We performed a numerical simulation of the (10$`\overline{1}`$)<sub>m</sub> reflection (see next subsection) and found that the signal is almost entirely due to the E1-E1 channel. Its existence implies a different value of the the $`x`$ and $`y`$ dipole matrix elements, which is at the basis of the NXLD. Notice that the Templeton scattering is zero in the corundum (paramagnetic) phase, as experimentally verified in the same work, due to the existence of the trigonal symmetry axis $`C_3`$, since then the $`x`$ and $`y`$ components of the electron scattering wavefunctions are degenerate in the real basis representation of the spherical harmonics. As a consequence mixed matrix elements of the kind $`<\mathrm{\Psi }_0|x|\mathrm{\Psi }_n><\mathrm{\Psi }_n|y|\mathrm{\Psi }_0>`$, which are at the origin of the Templeton effect in the $`\sigma \sigma `$ channel at the (10$`\overline{1}`$)<sub>m</sub> reflection (see next subsection), vanish.
The only possibility to have a situation like the one depicted in Ref. \[\] is in the case that by mere accident the two orthogonal linear polarizations of the incoming photon beam in the geometrical setting of the experiment lay at an angle of $`\pi /4`$ relative to the monoclinic $`b`$ axis normal to the glide plane (whether or not multiplied by the time-reversal operator). At the same time to avoid the extinction of the NRXLD signal, the magnetoelectric group should be $`P2^{}/a`$, since in this case the operator $`\widehat{T}\widehat{m}_b`$ does not change the sign of $`(L_x^2L_y^2)\mathrm{\Omega }_zh.c.`$.
Under these assumptions the search for this particular geometrical setting would have directly determined the ground state symmetry of the AFI phase. However there is no trace in Ref. \[\] of such investigation, so that their result cannot be unambiguously interpreted.
### C Multiple Scattering simulations of NXLD
The MS simulations of the normal linear dichroism in V<sub>2</sub>O<sub>3</sub> were carried out in the framework of the multiple scattering theory within the muffin-tin approximation. We chose a cluster containing 135 atoms, ie 54 Vanadium and 81 oxygen atoms in the correct ratio 2 to 3, having a radius of 6.9 Å, enough to get convergence with the cluster size. This latter is assured by convoluting the spectrum calculated with the real part of the Hedin-Lundqvist (HL) potential with a lorentian function having an energy dependent damping $`\mathrm{\Gamma }(E)`$ derived from the universal mean free path curve by the relation $`\lambda (E)=1/k_eE/\mathrm{\Gamma }(E)`$, where $`E`$ is the photoelectron kinetic energy (in Rydbergs) and $`k_e=\sqrt{E}`$ its wave vector. This procedure seems to give better results than calculating the spectrum directly with the complex HL potential, as recently found in other cases. The actual calculations were performed independently with two different codes, the FDMNES program which is non muffin-tin but incorporates a muffin-tin version and the CONTINUUM program, with similar results. As anticipated in the introduction, the independent particle approximation should be adequate to describe transitions to delocalized conduction states with $`p`$ symmetry around the photoaborber. Only strong quadrupole transitions to empty correlated $`d`$ states might not be well described, together with $`p`$ transitions to hybridized $`pd`$ states in the same energy range. However we are only interested in the higher energy region where the nonreciprocal gyrotropic effect seems to be more pronounced. Moreover in the same energy range the muffin-tin approximation is reasonably accurate, due to the close-packed structure of V<sub>2</sub>O<sub>3</sub> (despite the Vanadium voids) and the moderately high kinetic energy of the excited photoelectron. In order to eliminate any doubt on the confidence of the muffin-tin calculations, we also performed non-muffin-tin calculations on a substantially smaller cluster (eleven atoms) with the FDMNES program and compared with the muffin-tin counterpart. General shapes and, more important, amplitudes of the various signals discussed below did indeed compare very favorably. (Larger cluster calculations with no local symmetry, like in the present case, are probitively long in cpu time and require a lot of computer memory).
Even though the formalism of spherical tensors is very convenient to write down the response functions of x-ray spectroscopies, as illustrated in section III B, working in cartesian coordinates leads to immediately readable formulas. In the geometry of Ref. \[\] the direction of the incoming x-ray beam coincides with the trigonal axis of the high-temperature phase of V<sub>2</sub>O<sub>3</sub>. Thus, in the reference frame of Fig. 1, the two polarizations have components: $`ϵ_a(\mathrm{cos}\alpha ,\mathrm{sin}\alpha ,0)`$ and $`ϵ_b(\mathrm{sin}\alpha ,\mathrm{cos}\alpha ,0)`$.
The atomic linear dichroism in the E1-E1 channel is then given by:
$`\sigma _i^{NXLD}`$ $``$ $`{\displaystyle \underset{n}{}}\delta (\omega _n)(\mathrm{\Psi }_0^i|\stackrel{}{ϵ}_a\stackrel{}{r}|\mathrm{\Psi }_n\mathrm{\Psi }_n|\stackrel{}{ϵ}_a\stackrel{}{r}|\mathrm{\Psi }_0^i`$ (32)
$``$ $`\mathrm{\Psi }_0^i|\stackrel{}{ϵ}_b\stackrel{}{r}|\mathrm{\Psi }_n\mathrm{\Psi }_n|\stackrel{}{ϵ}_b\stackrel{}{r}|\mathrm{\Psi }_0^i)`$ (33)
$`=`$ $`{\displaystyle \underset{n}{}}\delta (\omega _n)\{\mathrm{cos}2\alpha (x|x_iy|y_i)`$ (34)
$`+`$ $`\mathrm{sin}2\alpha Re[x|y_i]\}`$ (35)
writing for brevity $`\omega _n=\mathrm{}\omega (E_nE_0)`$, $`\mathrm{\Psi }_0^i|x|\mathrm{\Psi }_n\mathrm{\Psi }_n|x|\mathrm{\Psi }_0^ix|x_i`$ and similarly for the other components.
When summing over all sites of the unit cell, only the part proportional to $`\mathrm{cos}2\alpha `$ survives, since the mixed matrix elements $`Re[x|y_i]`$ give opposite contributions for pairs related by the glide plane. Therefore the crystal absorption is given by:
$$\sigma _C^{NXLD}\mathrm{cos}2\alpha \underset{n}{}\delta (\omega _n)(x|xy|y)$$
(36)
This is in contrast to what happens for Templeton scattering in the $`\sigma \sigma `$ channel at the (10$`\overline{1}`$)<sub>m</sub> reflection in the same E1-E1 channel. In this case the incoming and outgoing polarization have components $`ϵ_\sigma =(\mathrm{cos}\beta ,\mathrm{sin}\beta ,0)`$, so that the scattering amplitude for the atom at site $`i`$ reads:
$`f_i(\omega )`$ $`=`$ $`(\mathrm{}\omega )^2{\displaystyle \underset{n}{}}{\displaystyle \frac{\omega _n+i\mathrm{\Gamma }}{\omega _n^2+\mathrm{\Gamma }^2}}(\mathrm{cos}^2\beta x|x_i+\mathrm{sin}^2\beta y|y_i`$ (37)
$`+`$ $`\mathrm{sin}2\beta Re[x|y_i])`$ (38)
Since the crystal structure factor at the (10$`\overline{1}`$)<sub>m</sub> reflection substracts the amplitudes coming from pairs of atoms related by the glide plane symmetry, the net crystal amplitude $`F_C(\omega )`$ turns out to be:
$`F_C(\omega )(\mathrm{}\omega )^2{\displaystyle \underset{n}{}}{\displaystyle \frac{\omega _n+i\mathrm{\Gamma }}{\omega _n^2+\mathrm{\Gamma }^2}}\mathrm{sin}2\beta Re[x|y]`$ (39)
as anticipated above. Therefore in $`C_3`$ symmetry both linear dichroism and Templeton scattering at the (10$`\overline{1}`$)<sub>m</sub> reflection are zero, whereas if this latter does not vanish a normal linear dichroism is also expected.
However it is a bit hard to derive the size of the linear dichroism from the size of Templeton scattering, due to the extinction corrections in this latter spectroscopy. Our aim here is to show that the energy shape of the Templeton intensity is mainly of dipolar origin. To this purpose we present in Fig. 2 the E1-E1 contribution of our simulation for the $`\sigma \sigma `$ channel at the (10$`\overline{1}`$)<sub>m</sub> reflection. The agreement with the experimental spectrum of Ref. \[\] is rather good, in the sense that we obtain peaks and valleys at the right energy position. Only the intensities disagree with the experiment. However one should observe that fig.(2b) of Ref. \[\] represents the raw data not corrected for absorption. The amplitude of the E1-E2 contribution has been calculated to be ten times smaller than that in the E1-E1 channel, due to radial matrix elements effects and the very small weight of the $`nd`$ ($`n>3`$) scattering amplitudes in the conduction band of $`p`$ symmetry. Since in charge scattering the two contributions do not interfere, if one neglects magnetic order, the ratio of intensities in the two channels is 1 to one hundred.
The main results of the numerical calculation for absorption and linear dichroism are shown in Figs. 3 and 4. Figure 3 shows the absorption signals in Mbarn with the incoming photon polarizations in the hexagonal plane and directed along b<sub>m</sub> and perpendicular to it. The general shape follows quite well the experimental counterpart. In Fig. 4 instead we show on an expanded scale the variation of the dichroic signal (in Mbarn) as a function of the angle $`\alpha `$. The full line curve with negative maximum corresponds to $`\alpha =0^o`$ ($`<x|x><y|y>`$), while the long-dashed line with positive maximum is for $`\alpha =90^o`$ ($`<y|y><x|x>`$), in keeping with the $`\mathrm{cos}2\alpha `$ variation. The other two curves are for $`\alpha =15^o`$ and $`\alpha =75^o`$ for later use. It is immedately evident that the dichroic signal vanish and then reverts its sign when the polarization angle crosses 45<sup>o</sup>. Its shape is fairly similar to the non reciprocal experimental signal reported in Ref. \[\] while the maximum changes from 0 to 4$`\%`$ of the main absorption and is about 3$`\%`$ for $`\alpha =75^o`$, to be compared with the experimental 1$`\%`$. In both figures we have not corrected for the unknown experimental resolution.
This finding is to be contrasted with an intensity more than 10 times smaller for the E1-E2 signal. This latter is in fact zero in the paramagnetic phase and picks up intensity only because of the magnetic ordering. Therefore on top of the factor one tenth in the matrix element, mentioned above when discussing Templeton scattering, one should add a further reduction factor originating from magnetic effects. We did not attempt to calculate the atomic signal as given by Eq. (27) for comparison, since the global signal depends on the crystal magnetic space group, as discussed above. Not knowing the mechanism that determines the lowering of the magnetic symmetry makes it impossible a meaningful estimate of the crystal magnetic effect. However this discussion should indicate that the non reciprocal gyrotropic effect in V<sub>2</sub>O<sub>3</sub>, if it exists, is more than one order of magnitude smaller than normal linear dichroism.
The relevance of these calculations to the argument we are discussing stems from the fact that there is a remote chance that the reversal of the dichroic signal with the magnetic field might be a geometrical artifact due to a lack of full control of the experimental conditions. In fact, since the procedure of magnetoelectric annealing is rather ineffective due to the non magnetoelectricity of V<sub>2</sub>O<sub>3</sub> and the sizable conductivity of the sample in the paramagnateic insulating phase as discussed in Ref. , it is not hard to conceive that the procedure performed in Ref. could in reality involve two twins rotated by 120<sup>o</sup> degrees, so that the two curves refer to linear polarizations at 15<sup>o</sup> and 75<sup>o</sup> degrees with respect to the monoclinic b<sub>m</sub>-axis. Therefore the control of the orientation of the crystal domain after annealing is essential in order to eliminate any ambiguity in the interpretation of the results.
Admittedly this is a quite remote possibility, however not more remote than the one depicted at the end of the previous subsection under the assumption that the observed effect was a truly nonreciprocal effect. Clearly the whole argument hinges on the existence of a sizable linear dichroic effect of E1-E1 character, which is anyway a prediction of the present work subject to experimental test. In fact a measurement of the angular variation of the linear dichroism in the AFI phase without magnetoelectric annealing might settle the question.
## IV Conclusions
In the preceding sections we have argued about the existence of a measurable normal (ie time-reversal and inversion even) linear dichroism in the AFI monoclinic phase of V<sub>2</sub>O<sub>3</sub>, both on the basis of realistic simulations in the framework of MS theory and experimental evidence coming from the observation of Templeton scattering at the (10$`\overline{1}`$)<sub>m</sub> forbidden reflection in the $`\sigma \sigma `$ channel. Therefore it comes to a surprise that no such a signal is present in the nonreciprocal linear dicroism experimental spectrum measured by Goulon et al. in the same sample. From the analysis of the crystal absorption cross section according to various ground state magnetic symmetry groups we have concluded that there exists only one group ($`P2^{}/a`$) and a particular geometrical setting capable of eliminating the NXLD signal and leaving in evidence the nonreciprocal dichroic effect.
Even accepting the fact that by a quite remote chance the authors in Ref. had fallen on the correct setting, there still remain some paradoxical aspects with their result. The group $`P2^{}/a`$ is ME; however a recent measurement of the ME effect has given a negative result. Moreover this group does not seem to be among the possible candidates of the magnetic symmetry of the ground state in the AFI phase, as obtained by the minimization of an effective Hubbard Hamiltonian with degenerate bands. Finally, the reduction to the group $`P2^{}/a`$ seems not compatible with the features of the Bragg-forbidden reflection (111)<sub>m</sub>: the signal is present only at the quadrupole energies (5465 eV) and it is absent at the dipole energies (5470–5490 eV), while no extinction rules are present at the dipolar energies for the $`P2^{}/a`$ group.
Faced to these paradoxes, we have tried to provide an alternative interpretation of the “nonreciprocal” results, invoking a geometrical origin for the reversal of the linear dichroism with the external magnetic field. However the question can only be settled by experimental tests in which one has the possibility to check the twin after each transition to the AFI phase, by means of an x-ray scattering equipment: this would give a full geometrical control of the system to be analyzed. Given the implications of the nonreciprocal effect, if confirmed, for the physics of V<sub>2</sub>O<sub>3</sub>, its unambiguous experimental determination is of the outmost importance.
We would like to acknowledge interesting discussions with L. Paolasini on his measurements of the Templeton scattering in V<sub>2</sub>O<sub>3</sub>.
|
warning/0506/astro-ph0506394.html
|
ar5iv
|
text
|
# X-ray Cluster Associated with the z=1.063 CSS Quasar 3C 186: The Jet is NOT Frustrated.
## 1 Introduction
Powerful radio sources, that are compact on galaxy scales, the Giga-Hertz Peaked Spectrum (GPS) and Compact Steep Spectrum (CSS) sources, comprise a significant fraction of the bright radio source population (10-20$`\%`$, O’Dea 1998). They are strong candidates for being the progenitors of large-scale radio sources (e.g. Fanti et al. 1995, O’Dea & Baum 1997, O’Dea 1998), but this connection has not been firmly established. Their radio morphologies show compact emission on arcsec (VLA resolution) scales while on milliarcsec scales (VLBI) the sources look remarkably like scaled down large radio galaxies, where the entire radio structure (1-10 kpc) is enclosed within the host galaxy. Since the first GPS samples have been constructed there has been a clear controversy regarding their nature (see O’Dea 1998 and references therein). In the evolution model the source size and the characteristic spectral break at GHz radio frequencies could be an indication of young age, while in the other model the radio jet could be frustrated by a dense confining medium. Recent observations (e.g. measured expansion timescales of $`<1000`$ years, Owsianik et al. 1998, Polatidis & Conway 2003) give more weight to the evolution model (Readhead et al. 1996, Snellen et al. 2000, Alexander 2000), although there has been no definite observational evidence to rule out either of the models and both interpretations are still viable.
Here, we report the Chandra discovery of extended X-ray emission associated with the compact steep spectrum (CSS) quasar, 3C 186 (Q0740+380, z=1.063). The Chandra spectrum of the diffuse emission contains 741$`\pm 40`$ counts and a strong (EW$``$412eV) iron emission line at the quasar redshift characteristic of thermal emission. This is the first observation of thermal emission associated with a CSS quasar at high redshift and gives us a rare opportunity to study interactions between an expanding CSS radio source and the cluster medium. As we show in Section 6.2 that the pressure of the hot cluster gas is too low, by 2-3 orders of magnitudes, to confine the radio source. This is direct observational evidence that the radio source is not thermally confined, and so instead is presumably young so that we are observing it at an early stage of its evolution.
Over the last decade attempts have been made to find X-ray clusters associated with radio-loud sources at high redshift (e.g. O’Dea 2000, Siemiginowska et al 2003 for studies related to GPS and CSS). If the clusters are found around large number of radio-loud sources then this could be used to place constraints on structure-formations models at large redshift. Bremer, Fabian & Crawford (1997) describe the model for an onset of a powerful radio source in the center of a cooling flow cluster. Such a cluster could also confine a compact radio source. The limited capabilities of the available X-ray telescopes allowed only for a few detections of extended X-ray emission around radio sources at redshifts $`z>0.3`$ (Hardcastle & Worrall 1999, Crawford & Fabian 2003, Worrall et al. 2001). High dynamic range observations are required to detect faint diffuse emission in the vicinity of a bright powerful source. The Chandra X-ray Observatory can resolve spatially distinct X-ray emission components in the vicinity of a strong X-ray source with $``$1 arcsec resolution and a high dynamic range, as evidenced, for example, by the discovery of many resolved quasar X-ray jets (e.g. Schwartz et al. 2000, Siemiginowska et al. 2002, Sambruna et al. 2004, Marshall et al. 2005).
The large-scale X-ray emission observed in radio sources can result from several processes, confusing detections of X-ray clusters at $`z>1`$ (e.g. Celotti & Fabian 2004). Radio synchrotron emission implies the presence of a population of relativistic particles that produce high-energy emission via inverse Compton scattering of the cosmic microwave background (CMB) photons. The energy density of the CMB increases with redshift as $`(1+z)^4`$, so the surface brightness of the inverse Compton emission is approximately constant with redshift. This is in contrast with thermal cluster emission where the surface brightness drops with redshift. X-ray spectral and spatial information are key to identifying different emission components and measuring diffuse thermal X-ray emission.
At the highest redshifts Carilli et al. (2002) describe extended emission possibly associated with thermal emission from shock-heated gas within $`150`$ kpc of the $`z=2.156`$ radio galaxy PKS 1138-262. Fabian et al. (2003) report $``$100 kpc-scale emission around the nucleus of the redshift $`z=`$1.786 radio galaxy 3C 294 and give several possible explanations for the origin of this emission. In both cases the poor quality of the X-ray spectrum does not allow confirmation of the thermal nature of the diffuse X-ray emission. At redshifts $`0.5<z<1.0`$ thermal confirmation has been possible in some sources, e.g. 3C 220.1 (Worrall et al. 2001)
Several deep images of nearby X-ray clusters (Fabian et al. 2003, Forman et al. 2003, Nulsen et al. 2004) obtained recently with Chandra provide evidence that intermittent AGN outbursts with an average power of $`10^{45}`$ erg sec<sup>-1</sup> supply energy into the cluster preventing its cooling (McNamara et al. 2005). These deeply imaged clusters contain only relatively low power AGN ($`10^{40}10^{43}`$ erg sec<sup>-1</sup>) that are interpreted as having been active in the past in order to drive the observed cluster morphology. In contrast to these low redshift clusters, the 3C 186 cluster is observed during the quasar’s active phase while heating the cluster medium.
3C 186 is a very luminous quasar (L$`{}_{bol}{}^{}10^{47}`$ erg sec<sup>-1</sup>). It has a strong big blue bump in the optical-UV band and broad optical emission lines (Netzer et al. 1997, Simpson & Rawlings 2000, Kuraszkiewicz et al. 2002, Evans & Koratkar 2004). It is therefore a typical quasar except for its radio properties. The radio morphology shows two components separated by 2$`\mathrm{}`$ and a jet connecting the core and NW component (Cawthorne et al., 1986). Murgia et al. (1999) estimated the age of the CSS source to be of the order of $`10^5`$ years based on the spectral age of the radio source. Our observation provides X-ray morphology and spectral information for the quasar and an associated X-ray cluster. The quasar core is so bright in X-rays that any X-ray emission associated with the radio components is not spatially resolved. The diffuse X-ray cluster emission is detected beyond the quasar core and the CSS source.
We describe the Chandra observation, our analysis techniques and the results for the extended component in Sec.2. Section 3 and 4 present the archival radio and optical data. The quasar X-ray spectrum is presented in Sec.5. Section 6 contains the discussion.
Throughout this paper we use the cosmological parameters based on the WMAP measurements (Spergel et al. 2003): H$`{}_{0}{}^{}=`$71 km sec<sup>-1</sup> Mpc<sup>-1</sup>, $`\mathrm{\Omega }_M=0.27`$, and $`\mathrm{\Omega }_{\mathrm{vac}}=0.73`$. At $`z=1.063`$, 1″ corresponds to $``$8.2 kpc.
## 2 Chandra Observations
3C 186 was observed for $`38`$ ksec with the Chandra Advanced CCD Imaging Spectrometer (ACIS-S, Weisskopf et al. 2002) on 2002 May 16 (ObsID 3098). The source was located $`35\mathrm{}`$ from the default aim-point position (to avoid node boundaries) on the ACIS-S backside illuminated chip S3 (Proposer’s Observatory Guide (POG)<sup>1</sup><sup>1</sup>1http://asc.harvard.edu/proposer/POG/index.html). The 1/8 subarray CCD readout mode of one CCD only was used resulting in 0.441 sec frame readout time. The observation was made in VFAINT mode with the standard 5x5 pixel island used to assign the event grades by the pipeline. This mode allows for a more efficient way of determining the background events and cleaning the background, especially at the higher energies. After standard filtering the effective exposure time for this observation was 34,398 sec. Given the ACIS-S count rate of 0.025 counts s<sup>-1</sup> frame<sup>-1</sup> the pileup fraction was low $`<2\%`$ (see PIMMS<sup>2</sup><sup>2</sup>2http://asc.harvard.edu/toolkit/pimms.jsp).
Figure 1 shows the Chandra ACIS-S image overlayed with the quasar core and background regions. Figure 2 shows the ACIS-S image of 3C 186 adaptively smoothed with the CIAO tool CSMOOTH. The X-ray emission is more extended than the X-ray emission of a typical radio-quiet quasar observed with Chandra. The diffuse emission extends up to $`120`$ kpc in radius. Figure LABEL:fig:comparisons show the smoothed images in soft (0.5-2 keV) and hard (2-7 keV) energy bands. Only the NE quadrant is visible in the hard band, while all of the extended emission is visible in the soft. Below we describe in detail the imaging and spectral analysis of this emission.
### 2.1 Imaging Analysis
The X-ray data analysis was performed in CIAO 3.2<sup>3</sup><sup>3</sup>3http://cxc.harvard.edu/ciao/ with the calibration files from the CALDB 3.0 data base. Note that the ACIS-S contamination file acisD1999-08-13contamN0003.fits was included in our analysis; this accounts for the temporal, but not spatial, variation of the contamination layer on the optical-blocking filter of ACIS<sup>4</sup><sup>4</sup>4http://cxc.harvard.edu/ciao/why/acisqedeg.html. Since the quasar is observed close to the aim-point, which is where the spatially-invariant contamination model was calibrated, the results will not change if the data were re-analyzed using the spatially-dependent contamination model. We used Sherpa (Freeman et al. 2001) for all spectral and image modeling and fitting.
We ran acis\_process\_events to remove pixel randomization and to obtain the highest resolution image data. The X-ray position of the quasar (J2000: 07 44 17.47 +37 53 17.11) agrees with the radio position (Li & Jin 1996) to better than 0.1$`\mathrm{}`$, (which is smaller than Chandra’s 90$`\%`$ pointing accuracy of 0.6 arcsec, Weisskopf et al. 2003), so we have high confidence in the source identification.
To determine the size of the extended emission we ran a ray trace using CHaRT<sup>5</sup><sup>5</sup>5http://cxc.harvard.edu/chart/ and then MARX<sup>6</sup><sup>6</sup>6http://space.mit.edu/CXC/MARX/ to create a high S/N simulation of a point source. We modeled the quasar core as a point source with the energy spectrum given by the fitting described in Section 5. We then extracted a radial profile from both the Chandra data and the simulated point source image assuming annuli separated by 1 arcsec and centered on the quasar. For the simulation we added 7$`\%`$ errors to account for uncertainty in the raytrace model (Schwartz et al 2000a, Jerius et al 2004). The PSF was normalized to match the peak surface brightness of the core. The resultant profiles are shown in Fig 4 and clearly illustrate that the observed emission (empty squares) is highly inconsistent with a point source (solid line and empty triangles).
We apply a $`\beta `$-model to the one-dimensional surface brightness profile beyond $`3\mathrm{}`$ radius to estimate a core radius and $`\beta `$ parameter for the diffuse emission. The best fit model is represented in Figure 5 with $`\beta `$=0.64$`{}_{0.07}{}^{}{}_{}{}^{+0.11}`$ and a core radius $`r_c=5.8`$$`{}_{1.7}{}^{}{}_{}{}^{+2.1}`$ which corresponds to 47$`{}_{}{}^{+13}{}_{14}{}^{}`$ kpc.
Since the core radius found in the one-dimensional analysis is small, a two-dimensional fit was made to see what influence the quasar emission and any non-sphericity of the cluster emission has on the $`\beta `$-model parameters. A model consisting of the ChaRT-generated PSF and a two-dimensional $`\beta `$ model was fitted to the data, using the default pixel size of 0.492″. The Cash statistic (Cash 1979) was used, since the number of counts per pixel was low outside the core, and the cluster center, ellipticity, and position angle were allowed to vary as well as the normalization, core radius, and $`\beta `$ parameter. The best-fit cluster location differs from that of the quasar by 0.2″, which is within the one-sigma error circle of the position (0.3″). The best-fit model is elliptical, with an ellipticity of $`0.24_{0.07}^{+0.06}`$ and position angle of $`47\pm 10`$ degrees, but the core radius and $`\beta `$ values are similar to the one-dimensional results, with $`r_c=5.5_{1.2}^{+1.5}`$″ (45$`{}_{}{}^{+12}{}_{10}{}^{}`$ kpc) and $`\beta =0.58_{0.05}^{+0.06}`$.
### 2.2 Spectral Analysis of the Extended emission
We extracted the energy spectrum of the extended emission from an annulus of radii 2.7″ and 15″ centered on the quasar. There are 1189$`\pm 34`$ total counts and 741.4$`\pm 40.4`$ net source counts in this region in the full Chandra energy band. The background spectrum was taken from the annulus of radii 20″ and 30″ . Because the background increases at low and high energies we modeled only the spectrum within a 0.3-7 keV energy range. The total number of counts in this energy range was 876$`\pm 31`$ with 691.0$`\pm 32.5`$ net counts. In all spectral modeling we simultaneously fit background and source data applying $`\chi ^2`$ statistics with the weighting described by Primini et al. (1994) as implemented in Sherpa. All errors quoted below are $`90\%`$ errors for a single parameter calculated with the projection routine in Sherpa. Table 1 lists the applied models and the best fit values.
We first fitted the spectrum of the diffuse emission with an absorbed power-law model assuming the equivalent column of hydrogen in the Galaxy of 5.68$`\times 10^{20}`$ atoms cm<sup>-2</sup> (COLDEN<sup>7</sup><sup>7</sup>7Stark et al 1992). We then tested for excess absorption. The fitted absorbing column is in agreement with the Galactic value (see Table 1). Modeling the data with a power law indicated an excess at the Fe-line energy characteristic of thermal emission at the quasar redshift. We added a gaussian line to the model and obtained the line equivalent width of EW=412 eV at the observed energy E<sub>obs</sub>=3.18$`\pm 0.07`$ keV (90$`\%`$ errors) corresponding to the rest-frame energy of 6.56$`\pm 0.14`$ keV. Note that this line energy indicated that the line is emitted by the hot ionized plasma.
We next applied the RAYMOND and MEKAL XSPEC plasma models in Sherpa and obtained gas temperatures 4.4-5.2 keV for two choices of abundance, solar and 0.3 solar, using the abundances of Anders & Grevesse (1989). Figure 6 shows the best-fit thermal model together with the residuals. The observed unabsorbed flux assuming these models is of order 6.2$`\pm 0.3\times 10^{14}`$erg sec<sup>-1</sup> cm<sup>-2</sup> for the 0.5-2 keV energy range and 5$`\pm 0.7\times 10^{14}`$erg sec<sup>-1</sup> cm<sup>-2</sup> for the 2-10 keV energy range (the flux errors are based on 5$`\%`$ and 14$`\%`$ counts uncertainties in each band respectively).
The 0.5-2 keV rest frame X-ray luminosity of the cluster emission is equal to $`3\times 10^{44}`$ erg sec<sup>-1</sup> (K-corrected). This is only the luminosity calculated based on the emission in the adopted annulus extending to about 122 kpc. This is about 52$`\%`$ of the total cluster luminosity as estimated from the radial profile fitting described in the previous section. Thus the total cluster luminosity is approximately $`6\times 10^{44}`$ erg sec<sup>-1</sup>.
To investigate a possible non-thermal contribution to the spectrum due to inverse Compton scattering of CMB photons we fixed the MEKAL model at the best value and added a power law component with $`\mathrm{\Gamma }=1.7`$ to the model. The 3$`\sigma `$ upper limit for the contribution of this component at 1 keV is then 3.4$`\times 10^6`$ photons cm<sup>-2</sup> sec<sup>-1</sup>, i.e. less than $`12\%`$ of the extended emission can be non-thermal.
To investigate possible temperature variations in the radial direction we extracted the spectra from two annuli: the inner one spans 2.7″-7.8″ and the outer one spans 7.8″-15″. Table 2 summarizes the results. The hardness radio indicates that there are more soft counts in the inner region than in the outer one. The RAYMOND model fit to these spectra indicate a slight temperature decrease, by $`\mathrm{\Delta }\mathrm{kT}0.3`$ keV, towards the inner radii. However the errorbars are larger than any apparent deviations. The available data do not allow for fitting of cooling flow models (see Sec.6.1 for cooling-flow discussion).
To further investigate possible temperature variations of the extended emission in the azimuthal direction we divided the 2.7″ to 15″ annulus into four sectors and extracted counts and spectra from these sectors. The sectors, illustrated in Fig.7a, were chosen to follow the non-symmetrical shape of the extended emission as apparent in the smoothed image (Fig.2 and Fig.3). We calculated the surface brightness profile in each sector which confirmed that the emission in the NE-SW direction (sectors 1 and 3) is stronger (by a factor of 1.25-1.5) than in NW-SE direction (sectors 4 and 2). The surface brightness values are indicated in Figure 7b.
The X-ray emission properties of the sectors are presented in Table 2. The hardness ratios indicate possible spectral differences between sectors. We fit the Chandra spectrum of each sector with a thermal emission model. There is a slight temperature variation (with a minimum of 3.7 keV and a maximum of 4.9 keV) between sectors in the best-fit values, however the error bars are approximately 1.5-2 keV, so that the fitted variations are all within 90$`\%`$ errors.
In summary, the diffuse emission is non-symmetric and elongated in the NE-SW direction with the harder emission towards NE (see Fig.3). Note that this structure is orthogonal to the radio emission described in the next section.
## 3 Radio Observations
We searched for radio emission corresponding to the diffuse X-ray emission revealed in the Chandra image by reanalyzing VLA 1.5 GHz data published by van Breugel et al.. (1992). The multi-configuration (A and B) data were obtained in 1987 and amounted to a total integration time of about 80 minutes split almost equally between the two configurations. At the resolution of our image (Figure 8a), the radio source is dominated by a 1.8” double source aligned at a position angle of about –37 deg. Our self-calibrated dataset does not show extended radio emission on the scale of the X-ray emission detected above a 3$`\sigma `$ rms noise of 0.75 mJy/beam in the naturally weighted image; the off-source ($`>`$15”) rms achieved in the image is about a factor of two smaller.
We also analyzed a 410 second VLA 15 GHz A configuration dataset of 3C 186 obtained on 13 Dec 1992 (program AL280) in order to examine the radio jet more clearly. The image shows similar features to previous high resolution maps (Cawthorne et al.. 1986; Spencer et al. 1991), i.e., a one-sided jet to the northwest connecting a flat-spectrum core to a diffuse radio lobe, and a bright radio lobe in the southeast direction (see Fig. 8b).
## 4 Hubble Space Telescope Data and Optical Emission
A cosmic-ray rejected WFPC2 Associations<sup>8</sup><sup>8</sup>8http://archive.stsci.edu/hst/manual image with a total integration time of 8,000 seconds was downloaded from the HST archive. The F675W image, obtained in Program 6491, was centered on one of the lower resolution wide field (WF) chips. We modeled the central source with elliptical isophotes utilizing the ELLIPSE task in the STSDAS<sup>9</sup><sup>9</sup>9http://www.stsci.edu/resources/software\_hardware/stsdas package and subtracted it from the image in order to show the nearest lying objects more clearly. The resultant image (Fig.9) shows several sources within the X-ray cluster emission indicated by a 15$`\mathrm{}`$ circle. Unfortunately only a single band image of 3C 186 was taken with the WFPC2 camera and we cannot identify the colors of these sources.
## 5 Quasar X-ray emission
The quasar core emission dominates the overall X-ray emission in the vicinity of 3C 186. We define the quasar emission region as a circle with 1.75″ radius and assume background emission from an annulus with radii 20″and 30″ (see Fig.1). Based on the PSF modeling we estimate that $``$98% of the point-source counts are included in this source region. The extracted quasar spectrum contains 1968.7$`\pm 44.3`$ net counts (1905.4 net counts in energy range between 0.3-7 keV). We model the spectrum with an absorbed power law. The best-fit power-law model has a photon index $`\mathrm{\Gamma }=2.01\pm 0.07`$ and a 2-10 keV flux of 1.7$`\times 10^{13}`$ ergs cm<sup>-2</sup> sec<sup>-1</sup> which corresponds to a quasar X-ray luminosity L<sub>X</sub>(2-10 keV) = 1.2$`\times 10^{45}`$erg sec<sup>-1</sup> and L<sub>X</sub>(0.5-2 keV)=1.1$`\times 10^{45}`$erg sec<sup>-1</sup> (unabsorbed and K-corrected luminosity). We do not detect any significant neutral absorbing column intrinsic to the quasar with the 3$`\sigma `$ upper limit to the equivalent column of Hydrogen of $`<9.0\times 10^{20}`$atoms cm<sup>-2</sup>.
The power-law fit to the data leaves some residuals at $`3`$ keV indicating a possible emission line at this energy. We added the Gaussian line component to the model and obtained the best fit location for the narrow emission line (FWHM$`<0.23`$ keV) at E<sub>obs</sub>=3.07$`{}_{0.11}{}^{}{}_{}{}^{+0.06}`$ keV corresponding to E<sub>rest</sub>=6.33$`\pm 0.06`$ keV. The line equivalent width is equal to EW=162 eV and it can be identified with Fe-K$`\alpha `$ emission. The best-fit model and residuals are shown in Fig.10. (Note that the residuals between 1 and 2 keV are due to calibration uncertainties.) We can estimate the flux contribution to this line from the extended thermal emission by extrapolating the radial profile of the extended emission into the central circular region assumed for the quasar emission. The contribution from the thermal cluster emission to the quasar spectrum is of order 10$`\%`$. Given the line flux in both components we estimate that only about $``$5$`\%`$ of the line emission can come from the thermal cluster emission, thus the line is dominated by the nuclear emission. Note that the energy, E<sub>rest</sub>=6.33$`\pm 0.06`$ keV, of the detected Fe-line in the quasar spectrum is in agreement with being emitted by the neutral medium, while the energy of the Fe-line emitted by the cluster gas, E<sub>rest</sub>=6.56$`\pm 0.14`$ keV, indicates an emission from ionized gas.
## 6 Discussion
### 6.1 X-ray Cluster
Our Chandra observation reveals X-ray cluster emission at the redshift of the quasar 3C 186. The X-ray properties of the cluster are summarized in Table 3. We compare the cluster temperature and its luminosity with results for the other clusters at high redshift using the MEKAL model with the abundance set to 0.3 as in Vikhlinin et al.(2002). The cluster temperature of $`5.2_{0.9}^{+1.2}`$ keV and the total X-ray luminosity of L$`{}_{X}{}^{}(0.52\mathrm{keV})`$ $`6\times 10^{44}`$ erg sec<sup>-1</sup> agree with the temperature-luminosity relation typically observed in high redshift ($`z>0.7`$) clusters (e.g. Vikhlinin et al. 2002, Lumb et al. 2004).
We can estimate physical properties of the cluster using standard formulae (Donahue et al. 2003; Worrall & Birkinshaw, 2004). Approximating the X-ray diffuse emission as spherical, we calculate the cluster central electron density ($`n_0`$) to be approximately 0.044$`\pm 0.006`$ cm<sup>-3</sup> (errors are only due to the uncertainty in normalization) for the best-fit 1D beta model parameters: $`\beta `$=0.64, a core radius of $`\theta _c=5.8\mathrm{}`$ (Sec.2.1), a gas temperature of 5 keV and the spectral normalization of 3.5$`(\pm 0.4)\times `$10<sup>-4</sup> based on the MEKAL model fit to the spectrum of the diffuse emission (assuming the annulus between 2.7″and 15″radii). These parameters imply that the mass of the gas enclosed within 1 Mpc radius of the isothermal sphere is $``$2.2$`\times 10^{13}`$M. The total gravitational mass enclosed within 1 Mpc is $``$2.6$`\times 10^{14}`$M assuming an isothermal-sphere model. The cluster gravitational mass is comparable to the mass of the other high redshift clusters recently measured with Chandra and XMM-Newton (Donahue et al. 2003, Worrall & Birkinshaw 2003, Vikhlinin et al 2002). The gas fraction, i.e. $`10\%`$, broadly agrees with the gas fraction usually found in high redshift (z$`>`$0.7) clusters.
We found a relatively small core radius, $`50`$ kpc, for the 3C 186 cluster while most redshift $`z>0.7`$ clusters have typical core radii larger than $`100`$ kpc. If gas of the same X-ray luminosity as found within a radius of 123 kpc of 3C 186 were distributed with a beta model of more typical core radius (150 kpc), the gas mass contained within a radius of 1 Mpc would increase by a factor of about 2.
Is there a cooling flow in this cluster? We estimate the gas mass enclosed within the core radius of $`50`$ kpc to be of the order of 2.2$`\times 10^{11}`$M. Given the gas central density and the temperature of 5 keV the cooling time for this core is $`1.6\times 10^9`$ years and without a heat source there would be a cooling flow with the cooling-flow rate $`50`$ M yr<sup>-1</sup> (Fabian & Nulsen 1977, Fabian 1994).
Our image analysis indicates that the cluster X-ray emission follows an elliptical distribution (see Section 2.1). The angle between the major axes of the two-dimensional ellipse and the radio jet axes is $`84_6^{+7}`$ degrees. This type of X-ray vs. radio morphology is often seen in lower redshift clusters and it is interpreted as due to interactions between the radio plasma and the cluster medium (as for example in Hydra cluster in Nulsen et al 2002). The current radio data do not show any radio emission on scales similar to the observed X-ray cluster emission. However, the X-ray morphology suggests that there could be old plasma there possibly associated with previous activity of the quasar. The detection of a relic at low radio frequency observation would allow studies of the evolution timescales and the feedback between the quasar activity and the cluster medium.
### 6.2 Optical Environment of 3C 186
Optical observations of the quasar 3C 186 indicated that the source is located in a rich galaxy environment. Sánchez and González-Serrano (2002) studied an over-density and clustering of galaxies in the optical field of several high redshift radio sources. They include a K-band image of the 3C 186 field and indicate that the surface density of the galaxies in a possible cluster peaks to the NE about 50$`\mathrm{}`$ from the quasar. While the reported location of the peak density is located just outside the Chandra FOV we do not detect any significant X-ray emission towards the optical peak away from the quasar. In fact the quasar seems to be centrally located with respect to the X-ray diffuse emission.
Recent optical study of the cluster environments of radio-loud quasars at $`0.6<`$z$`<1.1`$ by Barr et al (2003) shows that these quasars do not reside in the center of the galaxy distributions. Because the X-ray emitting gas traces the cluster’s gravitational potential, the X-ray observations can confirm whether the quasar is located in the center of the potential well of the cluster. The Chandra observations 3C 186 indicate that the quasar is located at the center of the X-ray emission and offset from the center of the galaxy distribution.
### 6.3 Lack of Confinement of the CSS source
The diffuse emission on a $``$120 kpc scale discovered in this Chandra observation indicates the presence of a hot Intercluster medium (ICM) surrounding this powerful CSS quasar. The X-ray spectrum of the cluster is dominated by a thermal component with a strong Fe-line at the quasar redshift. In one scenario the small size of CSS radio sources is associated with a dense environment that prevents the expansion of the source and confines the radio lobes to the size of a host galaxy (Wilkinson et al. 1981, van Breugel et al. 1984, O’Dea et al. 1991). Is this cluster then responsible for confining the CSS source?
Based on the cluster central density and temperature, we estimate a central thermal pressure of $`5\times 10^{11}`$ dyn cm<sup>-2</sup>. If this pressure is higher than the pressure within the expanding radio components of the CSS source then the cluster gas may be responsible for confining the radio source and its small size.
The radio source is dominated by a double at 1.5 GHz, which straddles a central flat-spectrum radio core as seen in the higher resolution 15 GHz map (Figure 8). Using DIFMAP’s MODELFIT (Shephard, Pearson & Taylor 1994) routine, the 1.5 GHz data are best-fit with two 0.59 Jy elliptical Gaussians with dimensions of 0.7” x 0.3”, and a separation of 1.8”. Taking the spectral index to be 1, and approximating each component as a homogeneous spheroid, we estimate the minimum pressure in each radio component to be $``$10<sup>-8</sup> dyn cm<sup>-2</sup> (see also Murgia et al 1999). Thus the radio source is highly overpressured by about 2-3 orders of magnitude with respect to the thermal cluster medium.
The new X-ray detection gives direct observational evidence that the radio source is not thermally confined as posited in the ”frustrated” scenario for CSS sources. Instead, at least in this CSS quasar, it appears that the radio source may indeed be young (Readhead & Hewish 1976; Phillips & Mutel 1982, Carvalho 1985) and that we are observing an early stage of radio source evolution.
The jet can also interact and be stopped by a clumpy cold medium of the host galaxy, as described by Carvalho et al 1998, De Young 1991 or Jeyakumar et al 2005. The absorption due to this cold medium should be detectable in the quasar X-ray spectrum. We can use the 3$`\sigma `$ upper limit of 9$`\times 10^{20}`$ cm<sup>-2</sup> on the total absorbing column density intrinsic to the quasar (see Sec.5 and Table 4) to estimate the total size of the clumpy medium. The frustrated jet models require clouds with densities of 1-30 cm<sup>-3</sup>. The absorption limit gives the size of the clumpy medium of order 10-100 pc compared with the 16 kpc diameter of the radio source. Any such region cannot significantly limit the expansion of the radio source and frustrate the jet (see also discussion in Guainazzi et al 2004).
Our X-ray observation of 3C 186 indicates that the CSS radio source is not confined, but it is at its early stage of the evolution into a large scale radio source.
### 6.4 The CSS and the Cluster Heating
The CSS radio components are overpressured with respect to the thermal cluster gas. Thus the expansion of these components into the cluster medium could potentially heat the center of this cluster. The energy dissipated into the cluster by the expanding radio components has been widely discussed in the context of the low redshift clusters, where there is evidence for the repetitive outbursts of an AGN. However, the details of the dissipation process are undecided with ion viscosity and “sound” waves being possible candidates (Fabian et al. 2005). On the other hand the mechanical energy released during the shock wave propagation throughout the cluster could be transfered into the cluster thermal energy at the location of the shock (P.Nulsen private communication).
We can estimate the energy content of the hot cluster gas assuming a total emitting volume of 2.3$`\times 10^{71}`$cm<sup>3</sup> (contained by an annulus with 3 and 15$`\mathrm{}`$ radii, assuming spherical geometry) and $`kT5`$ keV, to be of the order of $`\frac{3}{2}kTnV`$ 2$`\times 10^{61}`$ ergs (where $`n`$ is the average gas particle density in the cluster). We can estimate the jet power from the relation between the radio luminosity and the jet power given by Willott et al (1999, Eq.(12)). Using the 151 MHz flux density of 5.9$`\times 10^{24}`$ erg sec<sup>-1</sup> cm<sup>-2</sup> Hz<sup>-1</sup> (Hales et al. 1993) which accounts for the total radio emission from the jet and hot spots (the core is already almost completely self-absorbed at 1.7 GHz, Spencer et al. 1991) the jet kinetic power is of order $`L_{jet}10^{46}`$erg sec<sup>-1</sup>. If the expanding radio source dissipated the jet’s energy, $`L_{jet}10^{46}`$erg sec<sup>-1</sup>, into the cluster’s central 120 kpc region then the heating time would be $``$10<sup>8</sup> years. We can also estimate the amount of mechanical work done by the jet and radio components during the expansion to the current radio size ($`2\mathrm{}\times 0.3\mathrm{}2.3\times 10^{66}`$cm<sup>3</sup>) as $`pdV10^{56}`$ ergs. If the expansion velocity is of the order of 0.1$`c`$ then the radio source has been expanding for about $`5\times 10^5`$ years with an average power of 6$`\times 10^{42}`$ erg sec<sup>-1</sup>. The estimated jet power is $`3`$ orders of magnitude higher.
### 6.5 X-ray emission of the core
The evolution of radio-source expansion within host galaxies and clusters has been considered by Heinz, Reynolds and Begelman (1998). They simulated interactions between a growing radio source and the interstellar and intergalactic medium. For the highly supersonic expansion of the young source a shock forms around the expanding source and it heats up the medium to X-ray temperatures. As a result a “cocoon” of hot medium surrounds the radio source. Depending on the density of the medium and the strength of the shock a source of the size of 16 kpc can emit $`10^{45}`$ erg sec<sup>-1</sup> in the Chandra band.
The double CSS radio source and the radio jet remain unresolved in the Chandra observation of 3C 186. Thus the “cocoon” region is located within the unresolved X-ray core and thus could be giving a significant contribution to the total observed X-ray luminosity of 10<sup>45</sup> erg s<sup>-1</sup>. The amount of this contribution depends on the physical parameters of the expanding radio source and the ISM. A weak Fe-line is the only emission feature present in the otherwise featureless X-ray spectrum, however the observed spectral photon index is quite steep for a typical radio loud quasar (Elvis et al 1985, Bechtold et al 1994). We therefore estimated a possible thermal contribution to the quasar luminosity using the RAYMOND plasma model fits to the quasar spectrum. We concluded that about $`15\%`$ of the 0.5-2 keV luminosity, e.g. 1.5$`\times 10^{44}`$ erg sec<sup>-1</sup> could be due to the thermal emission. Of course there are other possible contribution to the observed X-ray spectrum from the radio jets knots and hot spots. Consistent theoretical modeling of the expansion and the emission of the jet components in the future may help in understanding their relative contributions to the X-ray spectrum. This is necessary in order to disentangle “true” quasar emission related directly to the accretion flow.
The quasar optical-UV (big blue bump) luminosity of 5.7$`\times 10^{46}`$erg sec<sup>-1</sup> (based on measurements in Simpson & Rawlings, 2000) is dominated by the typical quasar emission related to the accretion onto a supermassive black hole. We can therefore estimate the central black hole mass and required accretion rate based on that luminosity. Assuming that the quasar is emitting at the Eddington luminosity the black hole mass should be of the order $`4.5\times 10^8`$M. Based on CIV FWHM measurements of Kuraszkiewicz et al (2002) and the Vestergaard (2002) scaling relationship for the black hole mass, the estimated mass of the black hole is approximately a factor of 10 higher, $`3.2\times 10^9`$M. In any case the accretion rate required by the observed UV luminosity, and assuming 10$`\%`$ efficiency of converting the gravitational energy into radiation, is equal to $`10`$ Myear<sup>-1</sup>. Given the age of the radio source of $`5\times 10^5`$ years, a total of $`5\times 10^6`$M should have been accreted onto the black hole to support the current “outburst”.
Recent studies of formation of galaxies and growth of a supermassive black hole suggest that the quasar activity is a result of a merger event (e.g. DiMatteo, Springel & Hernquist 2005). In this model a short phase of quasar activity is a direct consequence of the increased fuel supply onto a central black hole. Based on the radio aging and the current size of 3C 186 its activity started $`10^6`$ years ago. The X-ray cluster emission of 3C 186, although elliptical in shape does not show any signatures of a merger event responsible for that activity. The cooling time for this cluster is $`10^9`$ years, so if this cluster is forming then the gas flowing into the center could onset the radio activity (Bremer et al. 1997). Such a powerful outburst can easily prevent the cooling of the cluster core. Although, the exact process of transferring the outburst energy into the cluster gas is unknown, the available outburst energy exceeds by $`>2`$ orders of magnitude the luminosity of the cluster core. The feedback between the jet heating and the cooling of the cluster can regulate the growth of the central black hole. On the other hand the short timescale of the current outburst may also suggest that the intermittent AGN behaviour could be related to the physics of the accretion flow (Siemiginowska, Czerny & Kostyunin 1996, Janiuk et al 2004) instead of the properties of a large scale environment triggering the quasar outburst. This is the first cluster observed at the early phase of the quasar radio activity giving us a potential to study the early stages of interactions between the cluster and AGN.
### 6.6 Old halo: relic
Is the current 3C 186 outburst the first phase of the quasar activity in this cluster? The large mass of the central black hole suggests that the supermassive black hole must have been formed much earlier than 10<sup>5</sup> years ago. As we show in the previous section only a small fraction of the mass has been accreted during the current active phase.
One way to accommodate the observed statistics of source sizes in the radio source evolution is to invoke intermittent activity (Reynolds & Begelman 1997) with average timescales of $``$10<sup>4</sup> – 10<sup>5</sup> yrs. In this scenario, a diffuse radio halo filled with old electrons from previous periods of activity should be apparent around CSS sources like 3C 186 in deep, low-frequency radio maps. Most CSS and GPS sources do not show very large-scale (100’s kpc) extended radio emission in present data (O’Dea 1998 and references therein). X-ray observations of nearby clusters suggest a longer period of the intermittency, e.g. 10$`{}_{}{}^{6}10^7`$ years (e.g. Fabian et al 2003, Forman et al 2004). For these longer timescales 3C 186 would still be at an early phase of the cycle.
If there was a previous outburst we might be able to detect “bubbles” filled with the old radio plasma (seen as depressions in the X-ray surface brightness) in deeper X-ray observations in the future. The “bubbles” observed in nearby clusters indicate locations of radio plasma injected into the cluster in the past (Churazov et al 2001). They are buoyant, moving within the cluster gas and last for at least one cycle, as indicated by more than one pair of “bubbles” observed in some X-ray clusters (Fabian et al. 2002, Belsole et al 2001, Owen et al. 2000). Even if the radio source outburst lasted only $``$10<sup>5</sup> years the radio plasma would still be present within buoyant “bubbles” loosing its energy via synchrotron radiation. Thus if there was a previous outburst we might be able to detect the radio relic in low-frequency observations.
At large redshifts, large-scale radio halos may produce a non-thermal X-ray component via inverse Compton scattering off the CMB. In the case of 3C 186, at z=1.063, the X-ray spectral fits do not require such a power-law component in addition to the thermal emission. From our data (Sec.2.2), we estimate that a non-thermal component contributes less than 12$`\%`$ of the total emission, F(1 keV)$``$few nJy.
No large-scale extended radio emission is visible in our 1.5 GHz VLA image – we estimate that there is less than 0.3 Jy coming from a possible halo by integrating the 3$`\sigma `$ rms limit over the extent of the detected X-ray emission ($``$15”). The electrons scattering CMB photons to 1 keV ($`\gamma `$10<sup>3</sup>) radiate synchrotron radio emission at much lower frequencies ($`\nu `$ (B / 1 $`\mu `$G ) MHz). A deep low-frequency radio observation aimed at detecting extended radio emission on the scale of the X-ray cluster would put a useful lower limit on the cluster magnetic field assuming equipartition (see Carilli & Taylor 2002).
## 7 Summary
Our main results can be summarize as follows:
* We have observed the CSS quasar 3C 186 for 38 ksec and detected X-ray cluster emission extending out to $`120`$ kpc from the quasar.
* The cluster temperature and luminosity follows the relationship observed in the other high redshift clusters, its gas mass is relatively small. The low mass could be the result of a small core radius measured in this cluster in comparison to the other clusters at $`z>0.7`$.
* The estimated pressure of the cluster gas is 2-3 order of magnitude lower than the pressure of the radio components, thus the cluster gas cannot confine the expanding radio source.
* Non-thermal and thermal emission associated with the radio components cannot be separated spatialy from the unresolved X-ray emission that is measured. We have placed an upper limit of 1.5$`\times 10^{44}`$ erg sec<sup>-1</sup> on the contribution from the gas that is shock heated as a result of the radio-source expansion. Detailed modeling is required to estimate the relative contribution of several possible components identified in radio observations.
* Future low frequency radio observation may provide information about the large scale distribution of an old electron population from a previous outburst of the quasar activity in this source.
We thank Paul Nulsen and Brian McNamara for useful discussions. We thank the anonymous referee for a careful reading of the manuscript and comments. This research is funded in part by NASA contract NAS8-39073. Partial support for this work was provided by the National Aeronautics and Space Administration through Chandra Award Number GO-01164X and GO2-3148A issued by the Chandra X-Ray Observatory Center, which is operated by the Smithsonian Astrophysical Observatory for and on behalf of NASA under contract NAS8-39073. The VLA is a facility of the National Radio Astronomy Observatory is operated by Associated Universities, Inc. under a cooperative agreement with the National Science Foundation. This work was supported in part by NASA grants GO2-3148A, GO-09820.01-A and NAS8-39073.
|
warning/0506/cond-mat0506705.html
|
ar5iv
|
text
|
# H2 dissociation over Au-nanowires and the fractional conductance quantum
## Abstract
The dissociation of H<sub>2</sub> molecules over stretched Au nanowires and its effect on the conductance are analyzed using a combination of Density Functional (DFT) total energy calculations and non-equilibrium Keldysh-Green function methods. Our DFT simulations reproduce the characteristic formation of Au monoatomic chains with a conductance close to $`G_0=2e^2/h`$. These stretched Au nanowires are shown to be better catalysts for H<sub>2</sub> dissociation than Au surfaces. This is confirmed by the nanowire conductance evidence: while insensitive to molecular hydrogen, atomic hydrogen induces the appearance of fractional conductances ($`G0.5G_0`$) as observed experimentally.
Gold surfaces are chemically inert and regarded as poor catalysts at variance with other metal surfaces. The low reactivity of molecular hydrogen on noble metal surfaces, like Au and Cu, seems to be well understood Norskov ; Stromquist98 ; Brivio99 ; Lemaire02 . Density Functional (DFT) calculations Norskov have shown that the dissociation of $`H_2`$ on Au or Cu is an activated process: for large molecule-surface distance, $`d`$, the interaction energy is repulsive with a high barrier of 1.1 eV (around $`d`$ = 1.5 Å), which the molecule has to overcome to move along the reaction path $`\mathrm{H}_2\mathrm{H}+\mathrm{H}`$ and reach the final atomic chemisorption state, with a total adsorption energy (2.07 eV per atom Lemaire02 ) that is less than the $`H_2`$ binding energy ( 4.75 eV). Compared to surfaces, small particles are known to be better catalysts. Au is, in particular, considered as an exceptional catalyst when prepared as nanoparticles on a variety of support materialsHaruta02 . Understanding this strong catalytic activity is still the subject of an extensive research effort with different possible explanations, including the particle shape or perimeter, support effects and the metal oxidation state Haruta02 ; Molina03 ; Goodman04 ; Yoon05 .
Nanowires are a good example of systems whose size is so small that one can expect their reactivity with molecules to be increased considerably. The formation of metallic nanocontacts has been analyzed in detail thanks to the gentle control of the distance at the atomic scale provided by both the Scanning Tunneling Microscope and the Mecanically Controllable Break Junction Rubio01 ; Scheer97 . While, in many metallic contacts, the formation of nanowires during the last stages of the stretching, just before the breaking point, is characterized by an atomic dimer geometry (Al is a paradigmatic case Jelinek03 ; Jelinek05 ), in Au the final geometry seems to be a chain of several atoms between the two electrodes with a conductance close to the conductance quantum unit $`G_o=\frac{2e^2}{h}`$ Yanson98 . Although the formation and stability of monoatomic Au chains has been addressed by several authors Torres ; Hakkinen ; daSilva01 , certain relevant aspects –in particular, the changes in the structural and transport properties of the nanocontacts induced by the presence of impurities Bahn ; Novaes ; Legoas ; Barnett04 – are not yet fully understood.
Recently , Csonka et al Csonka have analyzed the interaction of $`H_2`$ with a breaking gold nanowire and have found new fractional peaks (in units of $`G_o`$) in the conductance histogram. Moreover, conductance traces in a stretched nanowire demonstrate a reversible transition between fractional and integer conductances, in a time scale of milliseconds or seconds, suggesting successive adsorption and desorption of hydrogen on the chain. These experiments do not show these effects for Cu and Ag, where stable single-atom chains are not formed. This suggests the great importance of the Au chains in the variation of gold conductance in the presence of $`H_2`$. Csonka et al Csonka have proposed a possible explanation for the observed behaviour in terms of a dimerization effect which has been theoretically predicted for idealized clean Au nanowires Hakkinen ; Okamoto . However this dimerization has neither been observed in conductance histogram measurements nor in more complex theoretical simulations Sanchez-Portal ; daSilva01 ; Rubio01 .
In this Letter, we show that there is, in fact, a strong link between the enhanced reactivity of the stretched monoatomic gold chains and the appearance of the fractional conductance peaks. First, we have simulated the whole deformation process for an Au nanocontact upon stretching, finding the formation and the final breaking of a 4-atom Au chain. The realistic nanocontact configurations, calculated in this way, are then used to investigate whether the new fractional peaks in the conductance are associated with adsorbed molecular or atomic hydrogen. As our analysis suggests that only atomic hydrogen can be responsible for these changes in the nanowire conductance, we have also investigated how the chemical reaction $`\mathrm{H}_2\mathrm{H}+\mathrm{H}`$ is affected by the presence of a freely suspended Au-wire. This simplified model captures the key ingredients in the real nanocontact structure and provides a natural playground to explore the influence of the low dimensionality and the strain in the different steps of the dissociation process. These calculations show that a stretched Au-nanowire is much more reactive than the Au surfaces, with a small activation barrier, around 0.1 eV, for the H<sub>2</sub> dissociation and larger chemisorption energies. Our results for the nanocontact conductance, combined with the low value we have calculated for the $`H_2`$-reaction activation barrier, strongly suggest that the molecule dissociates on a Au-nanowire and that the observed fractional conductance upon adsorption of molecular hydrogen is basically due to the atomic hydrogen produced in the reaction.
Our calculations for stretched Au-nanowires have been performed using a fast local-orbital DFT-LDA code (Fireball2004 Fir04 ). This code offers a very favorable accuracy/efficiency balance if the basis of excited pseudoatomic orbitals basis is chosen carefully. The electrical conductance of the Au nanocontacts has been calculated using a Keldysh-Green function approach based on the first-principles tight-binding Hamiltonian obtained from the Fireball code, at each point of the deformation path (see references Jelinek03 ; Jelinek05 for details).
First, we have analyzed the formation of a Au-nanowire obtained by stretching a thick Au wire having four layers, with three atoms in each layer, sandwiched between two (111)-oriented metal electrodes as shown in figure 1 (configuration A represents the initial relaxed configuration). We use a supercell approach, where periodic boundary conditions along different directions are introduced: parallel to the surface we have considered a 3$`\times `$3-periodicity, while in the perpendicular direction we join artificially the last layers of both electrodes(see refs. Jelinek03 ; Jelinek05 ). Figure 1 shows the total energy of the system as a function of the stretching displacement relax\_details : notice the energy jumps associated with the irreversible deformations and structural rearrangements the wire has during this process. This figure (see also fig. 2) shows several snapshots for different geometries corresponding to the labels in the energy curve. Our DFT simulations reproduce the characteristic formation of Au monoatomic chains (with up to four atoms) found in the experiments. Similar results have been recently obtained using parametrized tight-binding molecular dynamics daSilva04 .
The inset of figure 1 shows the conductance of the system along the stretching process, as well as the different channels contributing to it. Notice that these results compare well with the experimental evidence Rubio01 . In particular, we reproduce (a) the long conductance plateau associated with the formation of the monoatomic chain (configurations C $``$ I), where the conductance is basically controlled by a single channel associated mostly with the Au s-electrons; and (b) the conductance oscillations during the elongation process. The very good agreement between both our structural and conductance results for the evolution of the Au nanocontact and the experimental evidence provides strong support to the remaining simulations presented in this paper.
In a second step, we have analyzed the nanowire conductance upon the adsorption of molecular and atomic hydrogen. Starting with the different geometries corresponding to the points E, F and G in figure 1, we have analyzed, via our local-orbital DFT code, how molecular and atomic hydrogen are adsorbed on those geometries and, then, how the nanowire conductance is modified according to the new optimized structures (see Figure 2). Molecular hydrogen is weakly adsorbed on all the different nanowires, with energies of around 0.3 eV. Notice that, in all these cases, molecular hydrogen does not penetrate the nanowire too much (see figure 2) and the conductance properties of the nanowire are not affected practically by the adsorption of molecular hydrogen. In particular, for the three cases shown in figure 2, the nanowire conductance takes the values 1.05 $`G_o`$, 1.05 $`G_o`$ and 0.95 $`G_o`$, respectively. Atomic hydrogen introduces more dramatic changes: in particular, for the cases shown in figure 2, the adsorption energies are 3.3 eV ,3.5 eV and 3.9 eV for the chains with two, three or four atoms in the nanowire, respectively. These values are much larger than the ones found for H adsorbed on a surface (around 2.1 eV) Norskov ; Lemaire02 . Moreover, we find significant modifications in the conductance of these three cases, with total values of 0.68 $`G_o`$, 0.65 $`G_o`$ and 0.55 $`G_o`$, respectively. The eigenchannel analysis for all these cases shows that the transport is dominated, as in the clean nanowire, by a channel mostly associated with the Au s-electrons but with a reduced transmitivity. This reduction is related to the significant displacement of the density of states (DOS) to lower energies (particularly evident for the d bands) for the Au atoms bonded to hydrogen that results in a reduction of the total DOS at E<sub>F</sub>. The nanowires with three or four atoms have been reanalyzed assuming that two hydrogen atoms are simultaneously adsorbed on the chain: the case of three atoms (case F) presents a conductance of 0.45 $`G_o`$, while for a four-atom nanowire (case G) the conductance is 0.2 $`G_o`$.
The enhanced reactivity of Au-chains, with respect to Au-surfaces, is due to the change in the Au DOS. Fig. 3b shows this DOS for the atoms of the Au(111)-surface, and for the Au-atoms of the four atom chain in the nanowire: Au-atoms with lower coordination form directional bonds and present a narrower DOS shifted towards the Fermi level. Then, for H chemisorbed on the chain, its DOS (see fig. 3a) presents a tightly bound state at -7.3 eV below the Fermi level, whereas the DOS for H on the Au(111)-surface presents a broadened resonance with some contribution from antibonding states just below the Fermi level Norskov . Notice that the reactivity of Au chains is further increased by the wire stretching due to a further shift of the Au-bands towards the Fermi level.
It is remarkable that, in the geometries shown in figure 2, neither the molecule nor the atom penetrates the nanowire, breaking the bond between two Au atoms. Recent work by Barnett et al Barnett04 showed that, for an essentially broken Au wire (with $`G0.02G_0`$), a barrierless insertion of the H<sub>2</sub> molecule into the contact is possible. This would correspond, in our case, to a wire state beyond the configuration I in figure 1, where the conductance is still very close to $`G_o`$. They then used this configuration as a starting point for a detailed study of the structure, orientation and stability of the molecule upon compression of the wire. Our approach is different, since we are interested in the wire-H<sub>2</sub> interaction during the chain formation process. We have calculated the energy barriers the molecule experiences when moving from the geometries shown in figure 2 to inserted sites. Our calculations yield values larger than 0.5 eV.
Both the conductance and the total energy results discussed so far support the dissociation of molecular hydrogen in the Au monoatomic chains. As the energetics were discussed in terms of the local orbital code, using the LDA approximation for the exchange-correlation, we have re-analyzed the dissociation mechanism on freely suspended Au-wires, with H (or $`H_2`$) adsorbed on them, using CASTEP castep ; technical , with a gradient corrected approximation Perdew91 (GGA) for the exchange-correlation functional. This simplified model for the nanowire is dictated by the computational resources needed for this full calculation, but it offers the possibility to discuss the relative contribution of the low dimensionality and the strain in the enhanced reactivity of the nanowire.
We have considered a freely suspended Au wire with six independent atoms per chain and periodic boundary conditions chosen to produce a stretch deformation in order to simulate the breaking process. After relaxing the free wire for each strain condition, we have studied how molecular hydrogen interacts with it: in particular, we have explored the possibility of having the following reaction, $`\mathrm{H}_2\mathrm{H}+\mathrm{H}`$ on the suspended wire. Figure 4 shows our main results for a wire with a 11% strain in an arbitrary configurational path: the case A in the figure represents the geometry of $`\mathrm{H}_2`$ interacting with the freely suspended wire; this is a physisorbed state with an adsorption energy of around 0.02 eV. Case B corresponds to an intermediate state in which a H-atom is chemisorbed between two Au-atoms , and the other H is still bonded to the Au-atom on which the molecule was initially physisorbed: this case has an adsorption energy of 0.3 eV; we have found, however, a barrier of 0.1 eV between states A and B . Case C corresponds to the final reaction state, in which the two H are adsorbed between two Au-atoms; the chemisorption energy of this final state is 1.4 eV, but again we find an energy barrier of 0.4 eV between states B and C. The reaction path drawn in figure 4 shows that molecular hydrogen sees a total barrier of 0.1 eV for the reaction $`\mathrm{H}_2\mathrm{H}+\mathrm{H}`$ on a freely suspended Au-wire. In a Au-surface, the energy barrier for that reaction is around 1.0 eV, and the adsorption energy, with respect to $`H_2`$ , is negative, around 0.5 eV. These numbers show the extreme importance that a stretched Au-nanowire has in the reactivity of molecular hydrogen: its reaction energy barrier and its chemisortion energy are lowered at least by 1 eV by the nanowire. Notice that the reduced dimensionality of the nanowire (compared to the surface) is not enough to induced this high reactivity, as shown by similar calculations for a non-strained chain, where we find an initial bound state for the molecule with energy -0.10 eV and quite large barriers ($`0.5`$ eV) for the dissociation.
In conclusion, stretched Au nanowires are much more reactive with molecular hydrogen than Au surfaces. Our DFT-GGA calculations show that, in a stretched Au nanowire, the activation barrier for H<sub>2</sub> dissociation is very small, around 0.10 eV. Notice that this is a reliable upper bound, as the use of the the GGA approximation (that only partially corrects the gross overestimation of barriers in LDA), the freely suspended wire geometry considered, and the neglect of possible quantum tunneling effects in H tend to overestimate the calculated barrier. This enhanced reactivity can also be expected in the case of Pt, where the formation of chains of several atoms has been also observed Smit01 . Complementary evidence is given by Csonka et al Csonka data for the conductance of a stretched Au nanowire with adsorbed molecular hydrogen. Our calculations indicate that only atomic hydrogen can be responsible of the changes observed by those researchers in the nanowire conductance. It is the combination of these two results, our calculated activation barrier and the changes in the nanowire conductance, that strongly suggest that molecular hydrogen dissociates when adsorbed on stretched Au nanowires.
###### Acknowledgements.
P.J. gratefully acknowledges financial support by the Ministerio de Educacion y Ciencia of Spain and the Ministry of Education, Youth and Sports of Czech Republic. This work has been supported by the DGI-MCyT (Spain) under contracts MAT2002-01534 and MAT2004-01271. Part of these calculations have been performed in the Centro de Computación Científica de la UAM. We thank Dr. J.C. Conesa and Prof. Sidney Davidson for helpful comments.
|
warning/0506/math0506419.html
|
ar5iv
|
text
|
# Adaptation and Parameter Estimation in Systems with Unstable Target Dynamics and Nonlinear Parametrization
## 1 Introduction
Results in adaptive control theory and systems identification are most frequently used in control engineering, but have potentially a much wider significance. In particular these theories are of great potential relevance for sciences such as physics and biology . On the other hand, it is in these areas that the current limitations of control theory are most strongly felt. Whereas effective procedures are available in case the system is static ,,,, adequate solutions for dynamical systems have been proposed under conditions that may not be adequate for most scientific applications. These conditions require that systems are linear in their parameters, the target dynamics is stable in the Lyapunov sense, and a Lyapunov function of the target dynamics can be given , , , ,, ,. Each of these restrictions alone is limiting the role of control theory in the scientific arena; together they constitute the ”standard” approach that confines control theory to a limited role, even within the realm of engineering.
Whereas in artificial system design, nonlinear parametrizations could often be avoided, physical and biological models often require the inclusion of nonlinearly parametrized uncertainties ,,,, . Proposed solutions to the nonlinear parametrization problem have cemented the standard approach, in that they eliminate any hopes of escaping from the stable target dynamics requirement. Nonlinearity is traditionally solved by invoking dominance of the nonlinear terms , . Dominance inevitably overcompensates the nonlinearity inherent to the system. This is undesirable if the system’s target motions require such nonlinearities. It is in particularly unhelpful, in case the system equations embody certain physical laws or other regularities that necessitate nonlinear parametrization. Terms overcompensated by dominance are likely to be exactly the ones postulated by these laws and regularities. In order to enable nonstable and in a sense more delicate target dynamics, more gentle control is needed: one that enables a system to reach the desired dynamical state by modification of, rather than destroying, its intrinsic motions.
Alternatives to dominance are available, but they face a variety of restrictions that make them appear less satisfactory. Often they apply to a narrowly defined class of parameterizations, e.g. convex functionals as in . When a broader class of nonlinearities is considered, for instance Hammerstain (Wiener) models, ,,,, the functions are restricted to static input (output) nonlinearities. Perhaps the most promising approach so far involves local linear (nonlinear) modelling techniques , , . The resulting models, on the other hand, are not always physically plausible. In case fairly general nonlinear state dependent functions are allowed , the class of dynamical systems is limited to those modeled by the first-order ordinary differential equations with nonlinear parameterized terms that are Lipschitz in time. The last restriction but not least is that the majority of these methods rely on the assumption of stable target dynamics.
There are physical and biological systems, however, which do not meet the requirement of stable target dynamics , . Multistability and coexistence of multiple attractors ,,, are well-known examples where a system could, at best, be only locally stable. In biophysics amplification by oscillatory instability is believed to be a general mechanism of signal detection in sensory systems . Furthermore, as demonstrated in , instability (intermittency) offers a solution to the longstanding binding problem in the biology of vision . In fact, also in artificial systems unstable target dynamics are sometimes required , . For instance, in , the effectiveness of using chaotic dynamics in solving the path finding problems in robotics is shown. In computer science unstable (intermittent) synchrony was shown to be an effective paradigm for solving the image segmentation problems .
Control-theoretic motivation and successful solutions to the problem of adaptive regulation to unstable dynamical states are provided in , for linear systems with linear parameterization. For nonlinear systems with nonlinear parameterization, and, possibly, unstable target dynamics the problems of adaptive regulation and parameter estimation need further development.
In our present paper we aim to provide a unified tool capable of solving the problems of adaptation and parameter estimation
$``$ in the presence of nonlinear state-dependent parameterization;
$``$ with non-trivial target sets (namely, surfaces in the systems’s state space)
$``$ with potentially unstable target dynamics, and therefore
$``$ without requiring for knowledge (or existence) of Lyapunov functions of the desired motions<sup>1</sup><sup>1</sup>1This problem, as mentioned in , was long remained an open theoretical challenge.
Previous efforts to deal with nonlinearity by adopting domination functions ,, or low-order mathematical models have tried to address the most general case. In contrast to this, we restrict ourselves in advance to a certain class of nonlinearities. This class, however, is wide enough to cover a variety of relevant models in physics, mechanics, physiology and neural computation. In particular, it includes models of stiction, slip and surface dependent friction, nonlinearities in dampers, smooth saturation, dead-zones in mechanical systems, and nonlinearities in models of bio-reactors , , , , .
In order to deal with unstable and non-equilibrium target dynamics, without invoking the knowledge (or even existence) of the corresponding Lyapunov funcion, we employ operator formalism in the functional spaces rather than conventional tools<sup>2</sup><sup>2</sup>2We refer here to common practice to fit the derivative of the goal functionals (Lyapunov candidates) to specific algebraic inequalities leading to the property of Lyapunov stability.. In particular, we consider desired dynamics in terms of input-output mappings in the specific functional spaces. The only requirement we impose on these mappings is existence of nonlinear operator gains that bound functional norms of the outputs, given norm-bounded inputs. The inputs for these mappings are the mismatches between the modeled uncertainty and a compensator. The outputs are the state $`𝐱`$ and a function $`\psi (𝐱,t)`$, not necessarily definite in state, which is considered a measure of deviation from the target set. This system-theoretic point of view allows us to formulate the problem of adaptation as a problem of regulation of the mismatches to specific functional spaces followed, if possible, by minimization of their functional norm.
We show that the solution to this problem for the given class of parameterizations does not require continuity of the corresponding operator gains. This, in turn, suggests that stability of the desired dynamics, which in many cases is synonymous to continuity of the input-output operators , is not a necessary requirement for our approach. Furthermore, given that $`\psi (𝐱,t)`$ may not be definite, this new point of view on the adaptation allows us to lift conventional state-space metric restrictions on the goal functionals<sup>3</sup><sup>3</sup>3Which are usually defined as positive-definite and radially unbounded functions of state ,,.
Under standard and intuitively clear additional hypotheses (i.e. persistent excitation of a certain functional of state), we show that the proposed adaptation procedures solve the problem of parameter estimation for nonlinearly parameterized uncertainties. In this case convergence is exponential. The estimates of the convergence rates are based on the results of and provided here for consistency. In case the conditions specifying the class of nonlinear in parameter uncertainties hold only locally, we show that sufficiently high frequency of excitation still ensures convergence. For cases where the standard persistent excitation property does not hold, we formulate a new version of nonlinear persistent excitation condition . With this new property it is still possible to show asymptotic convergence of the estimates to the actual values of unknown parameters. Whether the convergence is exponential is not answered in this paper.
The paper is organized as follows. Section 2 describes notations and conventions we are using in the paper; in Section 3 we formulate the problem. For the sake of compact exposition of our results we restrict ourselves to systems that are affine in control, although some non-affine cases are discussed towards the end of the paper. Section 4 contains the main results of the paper. We discuss several immediate extensions of the present results in Section 5. In Section 6 we provide a practically relevant application of our method, and Section 7 concludes the paper.
## 2 Notation
According to the standard convention, symbol $``$ defines the field of real numbers and $`_c=\{x|xc\}`$, $`_+=_0`$; symbol $``$ defines the set of natural numbers; symbol $`^n`$ stands for a linear space $`()`$ over the field of reals with $`\mathrm{dim}\{()\}=n`$; $`𝐱`$ denotes the Euclidian norm of $`𝐱^n`$; $`𝒞^k`$ denotes the space of functions that are at least $`k`$ times differentiable. Symbol $`𝒦`$ denotes the space of all functions $`\kappa :_+_+`$ such that $`\kappa (0)=0`$, and that $`x^{}>x^{\prime \prime }`$, $`x^{},x^{\prime \prime }_+`$ implies that $`\kappa (x^{})\kappa (x^{\prime \prime })>0`$. By symbol $`L_p^n[t_0,T]`$, where $`T>0`$, $`p1`$ we denote the space of all functions $`𝐟:_+^n`$ such that
$$𝐟_{p,[t_0,T]}=\left(_0^T𝐟(\tau )^p𝑑\tau \right)^{1/p}<\mathrm{}$$
Symbol $`𝐟_{p,[t_0,T]}`$ denotes the $`L_p^n[t_0,T]`$-norm of vector-function $`𝐟(t)`$. By $`L_{\mathrm{}}^n[t_0,T]`$ we denote the space of all functions $`𝐟:_+^n`$ such that
$$𝐟_{\mathrm{},[t_0,T]}=\mathrm{ess}sup\{𝐟(t),t[t_0,T]\}<\mathrm{},$$
and $`𝐟_{\mathrm{},[t_0,T]}`$ stands for the $`L_{\mathrm{}}^n[t_0,T]`$ norm of $`𝐟(t)`$.
Let $`𝐟:^n^m`$ be given. Function $`𝐟(𝐱):^n^m`$ is said to be locally bounded if for any $`𝐱<\delta `$ there exists constant $`D(\delta )>0`$ such that the following holds: $`𝐟(𝐱)D(\delta )`$.
Let $`\mathrm{\Gamma }`$ be an $`n\times n`$ square matrix, then $`\mathrm{\Gamma }>0`$ denotes a positive definite (symmetric) matrix, and $`\mathrm{\Gamma }^1`$ is the inverse of $`\mathrm{\Gamma }`$. By $`\mathrm{\Gamma }0`$ we denote a positive semi-definite matrix. Symbols $`\lambda _{\mathrm{min}}(\mathrm{\Gamma })`$, $`\lambda _{\mathrm{max}}(\mathrm{\Gamma })`$ stand for the minimal and maximal eigenvalues of $`\mathrm{\Gamma }`$ respectively. By symbol $`I`$ we denote the identity matrix. We reserve symbol $`𝐱_\mathrm{\Gamma }^2`$ to denote the following quadratic form: $`𝐱^T\mathrm{\Gamma }𝐱`$, $`𝐱^n`$. Notation $`||`$ stands for the module of a scalar. The solution of a system of differential equations $`\dot{𝐱}=𝐟(𝐱,t,𝜽,𝐮),𝐱(t_0)=𝐱_0`$, $`𝐮:_+^m`$, $`𝜽^d`$ for $`tt_0`$ will be denoted as $`𝐱(t,𝐱_0,t_0,𝜽,𝐮)`$, or simply as $`𝐱(t)`$ if it is clear from the context what the values of $`𝐱_0,𝜽`$ are and how the function $`𝐮(t)`$ is defined.
Let $`𝐮:^n\times ^d\times _+^m`$ be a function of state $`𝐱`$, parameters $`\widehat{𝜽}`$, and time $`t`$. Let in addition both $`𝐱`$ and $`\widehat{𝜽}`$ be functions of $`t`$. Then in case the arguments of $`𝐮`$ are clearly defined by the context, we will simply write $`𝐮(t)`$ instead of $`𝐮(𝐱(t),\widehat{𝜽}(t),t)`$.
The (forward complete) system $`\dot{𝐱}=𝐟(𝐱,t,𝜽,𝐮(t))`$, is said to have an $`L_p^m[t_0,T]L_q^n[t_0,T]`$, gain ($`Tt_0`$, $`p,q_1\mathrm{}`$) with respect to its input $`𝐮(t)`$ if and only if $`𝐱(t,𝐱_0,t_0,𝜽,𝐮(t))L_q^n[t_0,T]`$ for any $`𝐮(t)L_p^m[t_0,T]`$ and there exists a function $`\gamma _{q,p}:^n\times ^d\times _+_+`$ such that the following inequality holds:
$$𝐱(t)_{q,[t_0,T]}\gamma _{q,p}(𝐱_0,𝜽,𝐮(t)_{p,[t_0,T]})$$
Function $`\gamma _{q,p}(𝐱_0,𝜽,𝐮(t)_{p,[t_0,T]})`$ is assumed to be non-decreasing in $`𝐮(t)_{p,[t_0,T]}`$, and locally bounded in its arguments.
For notational convenience when dealing with vector fields and partial derivatives we will use the following extended notion of Lie derivative of a function. Let it be the case that $`𝐱^n`$ and $`𝐱`$ can be partitioned as follows $`𝐱=𝐱_1𝐱_2`$, where $`𝐱_1^q`$, $`𝐱_1=(x_{11},\mathrm{},x_{1q})^T`$, $`𝐱_2^p`$, $`𝐱_2=(x_{21},\mathrm{},x_{2p})^T`$, $`q+p=n`$, and $``$ denotes concatenation of two vectors. Define $`𝐟:^n^n`$ such that $`𝐟(𝐱)=𝐟_1(𝐱)𝐟_2(𝐱)`$, where $`𝐟_1:^n^q`$, $`𝐟_1()=(f_{11}(),\mathrm{},f_{1q}())^T`$, $`𝐟_2:^n^p`$, $`𝐟_2()=(f_{21}(),\mathrm{},f_{2p}())^T`$. Then symbol $`L_{𝐟_i(𝐱)}\psi (𝐱,t)`$, $`i\{1,2\}`$ denotes the Lie derivative of function $`\psi (𝐱,t)`$ with respect to vector field $`𝐟_i(𝐱,𝜽)`$:
$$L_{𝐟_i(𝐱)}\psi (𝐱,t)=\underset{j}{\overset{dim𝐱_i}{}}\frac{\psi (𝐱,t)}{x_{ij}}f_{ij}(𝐱,𝜽)$$
Symbol $`\mathrm{sign}()`$ denotes the signum-function:
$$\mathrm{sign}(s)=\{\begin{array}{cc}1,\hfill & s>0\hfill \\ 0,\hfill & s=0\hfill \\ 1,\hfill & s<0\hfill \end{array}$$
## 3 Problem Formulation
Let the following system be given:
$`\dot{𝐱}_1`$ $`=`$ $`𝐟_1(𝐱)+𝐠_1(𝐱)u,`$
$`\dot{𝐱}_2`$ $`=`$ $`𝐟_2(𝐱,𝜽)+𝐠_2(𝐱)u,`$ (1)
where
$$𝐱_1=(x_{11},\mathrm{},x_{1q})^T^q$$
$$𝐱_2=(x_{21},\mathrm{},x_{2p})^T^p$$
$$𝐱=(x_{11},\mathrm{},x_{1q},x_{21},\mathrm{},x_{2p})^T^n$$
$`𝜽\mathrm{\Omega }_\theta ^d`$ is a vector of unknown parameters, and $`\mathrm{\Omega }_\theta `$ is a closed bounded subset of $`^d`$; $`u`$ is the control input, and functions $`𝐟_1:^n^q`$, $`𝐟_2:^n\times ^d^p`$, $`𝐠_1:^n^q`$, $`𝐠_2:^n^p`$ are locally bounded. Vector $`𝐱^n`$ is a state vector, and vectors $`𝐱_1`$, $`𝐱_2`$ are referred to as uncertainty-independent and uncertainty-dependent partitions of $`𝐱`$, respectively.
For the sake of compactness we introduce the following alternative description for (3):
$`\dot{𝐱}=𝐟(𝐱,𝜽)+𝐠(𝐱)u,`$ (2)
where
$$𝐠(𝐱)=(g_{11}(𝐱),\mathrm{},g_{1q}(𝐱),g_{21}(𝐱),\mathrm{},g_{2p}(𝐱))^T$$
$$𝐟(𝐱)=(f_{11}(𝐱),\mathrm{},f_{1q}(𝐱),f_{21}(𝐱,𝜽),\mathrm{},f_{2p}(𝐱,𝜽))^T$$
Our goal is to derive both the control function $`u(𝐱,t)`$ and estimator $`\widehat{𝜽}(t)`$, such that all trajectories of the system are bounded and state $`𝐱(t)`$ converges to the desired domain in $`^n`$. In addition, we would like to find conditions ensuring that the estimate $`\widehat{𝜽}(t)`$ converges to unknown $`𝜽\mathrm{\Omega }_\theta `$ asymptotically. In order to ensure boundedness of the trajectories, we should design an input $`u(𝐱,t)`$ that restricts all possible motions of system (2) to an admissible bounded domain $`\mathrm{\Omega }^n`$ in the system state space, and if possible steers trajectories $`𝐱(t)`$ to the specific set $`\mathrm{\Omega }_0\mathrm{\Omega }`$.
As a measure of closeness of trajectories $`𝐱(t)`$ to the desired state we introduce the error function $`\psi :^n\times _+,\psi 𝒞^1`$ such that
$$\mathrm{\Omega }_0=\{𝐱(t)^n|\psi (𝐱(t),t)=0\}$$
(3)
In conventional theories it is usually required that function $`\psi (𝐱,t)`$ satisfies (algebraic) metric restrictions:
$$\nu _1(𝐱𝝃(t))\psi (𝐱,t)\nu _2(𝐱𝝃(t)),\nu _1,\nu _2𝒦_{\mathrm{}},$$
(4)
where function $`𝝃:_+^n`$, $`𝝃𝒞^0`$ is, for instance, the reference trajectory. Function $`\psi (𝐱,t)`$ in this case serves as the Lyapunov candidate for the controlled system under the assumption that $`𝜽`$ is known. The problem, however, is that finding such a Lyapunov candidate is not a trivial task. Furthermore, the desired trajectories $`𝝃(t)`$ as functions of time may only be partially specified. In case no reference function is available (e.g $`𝝃(t)=0`$) and the task is to steer the state $`𝐱`$ of system (3) to a non-trivial set $`\mathrm{\Omega }_0^n`$, it is often difficult to find a goal functional $`\psi (𝐱,t)`$ such that both (3) and (4) are satisfied. In addition, in physical and nonlinear systems the desired dynamics (e.g. dynamics of convergence of trajectories $`𝐱(t)`$ to the reference $`𝝃(t)`$) could be unstable in Lyapunov sense , although it may posess certain degree of attraction , and bounded deviation from the reference.
In order to tackle these complex, but still possible, phenomena we propose to replace the conventional goal functionals (4) with new and less restrictive ones. In particular, we propose to replace the standard norms $``$ in $`^n`$ in (4) with functional norms $`𝐱(t)_{p,[t_0,T]}`$, $`Tt_0`$ in the functional spaces $`L_p[t_0,T]`$, $`Tt_0`$, $`p_1\mathrm{}`$. In the other words, we replace algebraic inequality (4) with operator relations. This will allow us to keep function $`\psi (𝐱,t)`$ as a measure of closeness of trajectories $`𝐱(t)`$ to the desired set $`\mathrm{\Omega }_0`$ without imposing state-metric restrictions (4) on the function $`\psi (𝐱,t)`$. On the other hand we will be able to derive bounds for $`𝐱(t)`$ from the values of functional $`L_p^1[t_0,T]`$-norms of the function $`\psi (𝐱(t),t)`$. Let us formally introduce this requirement as follows:
###### Assumption 1 (Target operator)
For the given function $`\psi (𝐱,t)𝒞^1`$ the following property holds:
$$𝐱(t)_{\mathrm{},[t_0,T]}\stackrel{~}{\gamma }(𝐱_0,𝜽,\psi (𝐱(t),t)_{\mathrm{},[t_0,T]})$$
(5)
where $`\stackrel{~}{\gamma }(𝐱_0,𝛉,\psi (𝐱(t),t)_{\mathrm{},[t_0,T]})`$ is a locally bounded and non-negative function of its arguments.
Assumption 1 can be interpreted as a sort of unboundedness observability property \[Jiang\_1994\] of system (3) with respect to the “output” function $`\psi (𝐱,t)`$. It can also be viewed as a bounded input - bounded state assumption for system (3) along the constraint $`\psi (𝐱(t,𝐱_0,t_0,𝜽,u(𝐱(t),t)),t)=\upsilon (t)`$, where signal $`\upsilon (t)`$ serves as the new input<sup>4</sup><sup>4</sup>4If, however, boundedness of the state is not required explicitly (i.e. it is guaranteed by additional control or follows from the physical properties of the system itself), Assumption 1 can be removed from the statements of our results.. In order to illustrate this consider the equations of a spring-mass system with nonlinear damping:
$$\begin{array}{cc}\hfill \dot{x}_1& =x_2\hfill \\ \hfill \dot{x}_2& =k_0x_1+f(x_2,t)+u(t),k_0<0\hfill \end{array}$$
(6)
where $`f:\times _+`$, $`f(,)𝒞^1`$ is the nonlinear time-varying damping term. Equations of this type arise in broad areas of engineering ranging from active suspension control to haptic interfaces and identification of the muscle dynamics . Let the desired dynamics of (6) be an exponentially fast convergence of $`x_1(t)`$, $`x_2(t)`$ to the origin. This requirement is satisfied for the following target set:
$$\mathrm{\Omega }_0=\{𝐱^2|𝐱:x_1+\lambda x_2=0\},\lambda ,\lambda >0$$
Therefore, function $`\psi (𝐱,t)`$ could be chosen as $`\psi (𝐱,t)=x_1+\lambda x_2`$. Let $`\psi (𝐱(t),t)L_{\mathrm{}}^1[t_0,T]`$, i. e. $`x_1(t)+\lambda x_2(t)=\upsilon (t)`$, $`\upsilon (t)L_{\mathrm{}}^1[t_0,T]`$. An equivalent description of system (6) in accordance with this constraint is given by
$$\dot{x}_1=\lambda ^1x_1+\lambda ^1\upsilon (t),\lambda x_2(t)+x_1(t)=\upsilon (t)$$
(7)
It is clear that system (7) has the bounded input - bounded state property with respect to input $`\upsilon (t)`$ as $`x_1(t)_{\mathrm{},[t_0,T]}|x_1(t_0)|+\upsilon (t)_{\mathrm{},[t_0,T]}`$ and $`x_2(t)_{\mathrm{},[t_0,T]}\lambda ^1(x_1(t)_{\mathrm{},[t_0,T]}+\upsilon (t)_{\mathrm{},[t_0,T]})`$. This automatically implies that Assumption 1 holds for system (6) with $`\psi (𝐱,t)=x_1+\lambda x_2`$, $`\lambda >0`$. In particular, the following estimate holds
$$𝐱(t)_{\mathrm{},[t_0,T]}(1+\lambda ^1)|x_1(t_0)|+(1+2\lambda ^1)\psi (𝐱(t),t)_{\mathrm{},[t_0,T]}$$
Let us specify a class of control inputs $`u`$ which, in principle, can ensure boundedness of solutions $`𝐱(t,𝐱_0,t_0,𝜽,u)`$ for every $`𝜽\mathrm{\Omega }_\theta `$ and $`𝐱_0^n`$. According to (5), boundedness of $`𝐱(t,𝐱_0,t_0,𝜽,u)`$ is ensured if we find a control input $`u`$ such that $`\psi (𝐱(t),t)L_{\mathrm{}}^1[t_0,\mathrm{}]`$. To this objective consider the dynamics of system (2) with respect to $`\psi (𝐱,t)`$:
$`\dot{\psi }=L_{𝐟(𝐱,𝜽)}\psi (𝐱,t)+L_{𝐠(𝐱)}\psi (𝐱,t)u+{\displaystyle \frac{\psi (𝐱,t)}{t}},`$ (8)
Assuming that the inverse $`\left(L_{𝐠(𝐱)}\psi (𝐱,t)\right)^1`$ exists everywhere, we may choose the control input $`u`$ in the following class of functions:
$$\begin{array}{cc}\hfill u(𝐱,\widehat{𝜽},𝝎,t)& =(L_{𝐠(𝐱)}\psi (𝐱,t))^1\left(L_{𝐟(𝐱,\widehat{𝜽})}\psi (𝐱,t)\phi (\psi ,𝝎,t)\frac{\psi (𝐱,t)}{t}\right)\hfill \\ \hfill \phi :& \times ^w\times _+\hfill \end{array}$$
(9)
where $`𝝎\mathrm{\Omega }_\omega ^w`$ is a vector of known parameters of function $`\phi (\psi ,𝝎,t)`$. Denoting $`L_{𝐟(𝐱,𝜽)}\psi (𝐱,t)=f(𝐱,𝜽,t)`$ and taking into account (9) we may rewrite equation (8) in the following manner:
$`\dot{\psi }=f(𝐱,𝜽,t)f(𝐱,\widehat{𝜽},t)\phi (\psi ,𝝎,t)`$ (10)
Hence, feedback (9) renders the original system (3) into the well-known nonlinear error model form <sup>5</sup><sup>5</sup>5The error models (10) have proven to be convenient representations of systems with nonlinear parametrization in the problems of adaptation and parameter estimation .
In practical applications, state $`𝐱`$ of original model (3) is hardly ever available. Furthermore, imprecise physical models of the processes and measurement noise often lead to the presence of unmodeled dynamics in (10). Although we do not address these issues in detail in the present article, we do allow additive perturbations that are functions of time from $`L_2^1[t_0,\mathrm{}]`$ in the right-hand side of (10). In particular, instead of (10) we consider the following equation:
$`\dot{\psi }=f(𝐱,𝜽,t)f(𝐱,\widehat{𝜽},t)\phi (\psi ,𝝎,t)+\epsilon (t),`$ (11)
where, if not stated overwise, the function $`\epsilon :_+`$, $`\epsilon L_2^1[t_0,\mathrm{}]C^0`$. One of the immediate advantages of (11) in comparison with (10) is that it allows us to take the presence of state observers in the system into consideration. This clearly widens the range of possible applications of our results.
Let us now specify the desired properties of function $`\phi (\psi ,𝝎,t)`$ in (9), (11). The majority of known algorithms for parameter estimation and adaptive control assume global (Lyapunov) stability of system (11) for $`𝜽\widehat{𝜽}`$. In our study, however, we would like to refrain from this standard and at the same time restrictive requirement. Instead we propose that finite energy of the signal $`f(𝐱(t),𝜽,t)f(𝐱(t),\widehat{𝜽}(t),t)`$, defined for example by its $`L_2^1[t_0,\mathrm{}]`$ norm with respect to the variable $`t`$, results in finite deviation from the target set given by equality $`\psi (𝐱,t)=0`$. Formally this requirement is introduced in Assumption 2:
###### Assumption 2 (Target dynamics operator)
Consider the following system:
$$\dot{\psi }=\phi (\psi ,𝝎,t)+\zeta (t),$$
(12)
where $`\zeta :_+`$ and $`\phi (\psi ,𝛚,t)`$ is from (11). Then for every $`𝛚\mathrm{\Omega }_\omega `$ system (12) has $`L_2^1[t_0,\mathrm{}]L_{\mathrm{}}^1[t_0,\mathrm{}]`$ gain with respect to input $`\zeta (t)`$. In the other words,
$$\zeta (t)L_2^1[t_0,\mathrm{}]\psi (t,\psi _0,t_0,𝝎)L_{\mathrm{}}^1[t_0,\mathrm{}],\psi _0$$
and there exists a function $`\gamma _{\mathrm{},2}`$ such that
$$\psi (t)_{\mathrm{},[t_0,T]}\gamma _{\mathrm{},2}(\psi _0,𝝎,\zeta (t)_{2,[t_0,T]}),\zeta (t)L_2^1[t_0,T]$$
(13)
In contrast to conventional approaches, Assumption 2 does not require global asymptotic stability of the origin of (unperturbed, i.e for $`\zeta (t)=0`$) system (12). In fact, system (12) is allowed to have Lyapunov-unstable equilibria. Moreover, there may be no equilibria at all in (12), or it can even exhibit chaotic dynamics. Examples of such systems, which potentially inherit chaotic behavior but still satisfy Assumption 2, are the well-known Lorenz and Hindmarsh-Rose oscillators. The last system models ion current through a membrane in the living cell, and is widely used in artificial neural networks, for instance, for processing of the visual information .
When the stability of the target dynamics $`\dot{\psi }=\phi (\psi ,𝝎,t)`$ is known a-priori, one of the benefits of Assumption 2 is that there is no need to know the particular Lyapunov function of the unperturbed system. Apart from being, in some sense, a more friendly and less invasive concept, this enables us to design adaptive/parameter estimation procedures for systems with externally-driven uncontrolled multistability <sup>6</sup><sup>6</sup>6In systems with externally driven multistability, i.e. when there are multiple coexistent attractors and trajectories switch from one attractor to another depending on the external perturbation, parameter estimation/control algorithms based on the knowledge of a specific Lyapunov function require additional information about the instant dynamical state (attractors and their allocation) of the system itself. This leads to a necessity to identify current dynamical state of the system prior to control/identification of its parameters..
The differences between conventional restrictions on the goal functionals and alternative requirements formulated in Assumptions 1, 2 are further illustrated with Figure 1.
For simplicity it is assumed that function $`\psi (𝐱,t)`$ does not depend on $`t`$ explicitly and therefore its zeroes form a (set) surface in $`^n`$. For the conventional approaches this set should additionally satisfy metric conditions (4) in $`^n`$, Fig. 1. a. These conditions often restrict class of the possible target sets to the points in $`^n`$. In case Assumptions 1, 2 are satisfied this restriction does not apply any more. Indeed, given that $`\zeta (t)L_2^1[t_0,\mathrm{}]`$ we can bound $`\psi (𝐱(t),t)_{\mathrm{},[t_0,\mathrm{}]}\gamma _{\mathrm{},2}(\psi _0,𝝎,\zeta (t)_{2,[t_0,\mathrm{}]})=M`$. Therefore, according to Assumption 2, the state $`𝐱(t)`$ is bounded and belongs to the sphere $`\mathrm{\Omega }_𝐱=\{𝐱^n|𝐱:𝐱(t)\stackrel{~}{\gamma }(𝐱_0,𝜽,M)=\mathrm{\Delta }\}`$. On the other hand, the state $`𝐱(t)`$ belongs to the domain $`\mathrm{\Omega }_\psi =\{𝐱^n|𝐱:|\psi (𝐱,t)|M\}`$. This implies that the segments of trajectory $`𝐱(t,𝐱_0,t_0,𝜽,𝐮(t))`$, for $`tt_0`$ will remain in the bounded domain $`\mathrm{\Omega }_𝐱\mathrm{\Omega }_\psi `$ (shadowy volume in Fig. 1. b.) for all $`t>t_0`$.
The Figure 1 also emphasizes the difference between the proposed operator framework and known approaches in adaptive control based on geometrical representations . Indeed, the results based on coordinate transformations around the target manifold (3) are applicable only in a subset of $`^n`$ where $`\psi (𝐱,t)`$ does not depend explicitly on $`t`$, and rank of $`\psi (𝐱,t)`$ is constant. In this respect these results are local. On the other hand, Assumptions 1, 2 do not require constant rank conditions and allow both time-varying $`\psi (𝐱,t)`$ and $`\phi (\psi ,𝝎,t)`$. This makes Assumptions 1, 2 a suitable replacement to conventional approaches for systems with non-stationary dynamics, or ones which are far away from equilibrium or invariant target manifolds.
So far we have introduced basic assumptions on system (3) dynamics and the class of feedback considered in this article. Let us now specify the class of functions $`f(𝐱,𝜽,t)`$ in (11). Since general parametrization of function $`f(𝐱,𝜽,t)`$ is methodologically difficult to deal with but solutions provided for a restricted class of nonlinearities (for instance to those which allow linear re-parametrization) often yield physically implausible models, we have opted for a new class of parameterizations. This class shall include a sufficiently broad range of physical models, in particular those with nonlinear parametrization; the proposed parameterizations will also, in principle, be able to handle arbitrary state nonlinearity in the class of functions from $`𝒞^1`$. As a candidate for such a parametrization we suggest nonlinear functions that satisfy the following assumption:
###### Assumption 3 (Monotonicity and Growth Rate in Parameters)
For the given function $`f(𝐱,𝛉,t)`$ in (11) there exists function $`𝛂(𝐱,t):^n\times _+^d,𝛂(𝐱,t)𝒞^1`$ and positive constant $`D>0`$ such that
$$(f(𝐱,\widehat{𝜽},t)f(𝐱,𝜽,t))(𝜶(𝐱,t)^T(\widehat{𝜽}𝜽))0$$
(14)
$$|f(𝐱,\widehat{𝜽},t)f(𝐱,𝜽,t)|D|𝜶(𝐱,t)^T(\widehat{𝜽}𝜽)|$$
(15)
The first inequality (14) in Assumption 3 holds, for example, for every smooth nonlinear function which is monotonic with respect to a linear functional $`\mathit{\varphi }(𝐱)^T𝜽`$ over a vector of parameters:
$$f(𝐱,𝜽,t)=f_m(𝐱,\mathit{\varphi }(𝐱)^T𝜽,t)$$
$$\mathrm{sign}\left(\frac{f_m(𝐱,\lambda ,t)}{\lambda }\right)=\mathrm{const}$$
Hence function $`𝜶(𝐱,t)`$ satisfying (14) could be chosen in the following form: $`𝜶(𝐱,t)=M\mathit{\varphi }(𝐱)\kappa (𝐱,t)`$, where $`\kappa :^n\times _+_+`$, $`\kappa (𝐱,t)𝒞^1`$.
The second inequality, (15), is satisfied if the function $`f(𝐱,\mathit{\varphi }(𝐱)^T𝜽,t)`$ does not grow faster than a linear function in variable $`\mathit{\varphi }(𝐱)^T𝜽`$ for every $`𝐱^n`$. This requirement holds, for example, for those functions $`f(𝐱,\mathit{\varphi }(𝐱)^T𝜽,t)`$ which are globally Lipschitz in $`\mathit{\varphi }(𝐱)^T𝜽`$:
$$|f_m(𝐱,\mathit{\varphi }(𝐱)^T𝜽,t)f_m(𝐱,\mathit{\varphi }(𝐱)^T𝜽^{},t)|D_\theta (𝐱,t)|\mathit{\varphi }(𝐱)^T(𝜽𝜽^{})|$$
In particular, inequalities (14), (15) hold for the function $`f(𝐱,\mathit{\varphi }(𝐱)^T𝜽,t)`$ with $`𝜶(𝐱,t)=MD_\theta (𝐱,t)\mathit{\varphi }(𝐱)`$. A graphical illustration of the choice of function $`𝜶(𝐱,t)`$ is given in Figure 2.
This set of conditions naturally extends from systems that are linear in parameters to those with nonlinear parametrization. Assumption 3 covers (at least for bounded $`𝜽,\widehat{𝜽}\mathrm{\Omega }_\theta `$) a considerable variety of practically relevant models with nonlinear parametrization. These include effects of stiction forces , slip and surface dependent friction given by the “magic formula” or physics-inspired model of the tyre , nonlinear processes in dampers for automotive suspension , smooth saturation, and dead-zones in mechanical systems. It further includes nonlinearities in models of bio-reactors . The class of functions $`f(𝐱,𝜽,t)`$ specified in Assumption 3 can also serve as nonlinear replacement of functions that are linear in their parameters in a variety of piecewise approximation models. Last but not least, this set of functions includes sigmoid and Gaussian nonlinearities, which are favored in neuro and fuzzy control and mathematical models of neural processes . Table 1 provides some of the parametric nonlinearities that occur in these processes and their corresponding functions $`𝜶(𝐱,t)`$.
Assumption 3 bounds the growth rate of the difference $`|f(𝐱,𝜽,t)f(𝐱,\widehat{𝜽},t)|`$ by the functional $`D|𝜶(𝐱,t)^T(\widehat{𝜽}𝜽)|`$. This will help us to find a parameter estimation algorithm such that the estimates converge to $`𝜽`$ sufficiently fast for the solutions of (3), (11) to remain bounded with non-dominating feedback (9). On the other hand, parametric error $`\widehat{𝜽}𝜽`$ can be inferred from the changes in the variable $`\psi (𝐱,t)`$, according to (11), only by means of the difference $`f(𝐱,𝜽,t)f(𝐱,\widehat{𝜽},t)`$. Therefore, as long as convergence of the estimates $`\widehat{𝜽}`$ to $`𝜽`$ is expected, it is also useful to have the estimate of $`|f(𝐱,𝜽,t)f(𝐱,\widehat{𝜽},t)|`$ from below, as specified in Assumption 4<sup>7</sup><sup>7</sup>7Despite Assumption 4 requires that (16) holds for every $`𝐱^n`$, $`𝜽^d`$, and $`t_+`$, we will see later that for a variety of problems it is sufficient that it is satisfied only locally.:
###### Assumption 4
For the given function $`f(𝐱,𝛉,t)`$ in (11) and function $`𝛂(𝐱,t)`$, satisfying Assumption 3, there exists a positive constant $`D_1>0`$ such that
$$|f(𝐱,\widehat{𝜽},t)f(𝐱,𝜽,t)|D_1|𝜶(𝐱,t)^T(\widehat{𝜽}𝜽)|$$
(16)
In problems of parameter estimation, effectiveness of the algorithms often depends on how ”good” the nonlinearity $`f(𝐱,𝜽,t)`$ is, and how predictable locally is the system’s behavior. As a measure of goodness and predictability usually the substitutes as smoothness, boundedness, and Lipschitz conditions are considered. In our study, we distinguish several such specific properties of functions $`f(𝐱,𝜽,t)`$ and $`\phi (\psi ,𝝎,t)`$. These properties are provided below
###### H 1
Function $`f(𝐱,𝛉,t)`$ is locally bounded with respect to $`𝐱`$, $`𝛉`$ uniformly in $`t`$.
###### H 2
Function $`f(𝐱,𝛉,t)𝒞^1`$, and $`f(𝐱,𝛉,t)/t`$ is locally bounded with respect to $`𝐱`$, $`𝛉`$ uniformly in $`t`$.
###### H 3
Function $`f(𝐱,𝛉,t)`$ is globally Lipschitz with respect to $`𝛉`$ uniformly in $`𝐱`$, $`t`$:
$$D_\theta >0:|f(𝐱,𝜽,t)f(𝐱,\widehat{𝜽},t)|D_\theta 𝜽\widehat{𝜽}$$
###### H 4
Let $`U_x^n`$, $`U_\theta ^d`$ be given and $`U_x`$, $`U_\theta `$ are bounded. Then there exists constant $`D_{U_x,U_\theta }>0`$ such that for every $`𝐱U_x`$ and $`𝛉,\widehat{𝛉}U_\theta `$ Assumption 4 is satisfied with $`D_1=D_{U_x,U_\theta }`$.
###### H 5
Function $`\phi (\psi ,𝛚,t)`$ is locally bounded in $`\psi `$, $`𝛚`$ uniformly in $`t`$.
In the next section we present novel algorithms for adaptive control and parameter estimation in nonlinear dynamical systems (2) which satisfy Assumptions 1, 2, 3, and 4. We show that under an additional structural requirement, which relates properties of function $`𝜶(𝐱,t)`$ and vector-filed $`𝐟(𝐱,𝜽)=𝐟_1(𝐱,𝜽)𝐟_2(𝐱,𝜽)`$ in (3), (2), the following desired property holds:
$$𝐱(t)L_{\mathrm{}}^n[t_0,\mathrm{}];f(𝐱(t),𝜽,t)f(𝐱,\widehat{𝜽}(t),t)L_2^1[t_0,\mathrm{}]$$
(17)
After boundedness of the solutions is guaranteed, we prove that
$$\underset{t\mathrm{}}{lim}\psi (𝐱(t),t)=0$$
In addition, we show that
$$\underset{t\mathrm{}}{lim}\widehat{𝜽}(t)=𝜽$$
(18)
In particular we demonstrate that the standard persistent excitation condition is sufficient to guarantee the convergence. Furthermore, in the case that Assumptions 3,4 hold only locally (for $`𝐱`$ from a domain of $`^n`$) we demonstrate that sufficiently high excitation in the system still leads to the desired estimates.
## 4 Main Results
Standard approaches in parameter estimation and adaptation problems usually assume feedback and a parameter adjustment algorithm in the following form
$$\begin{array}{cc}& u=u(𝐱,\widehat{𝜽},t)\hfill \\ & \dot{\widehat{𝜽}}=𝒜_{\mathrm{lg}}(\psi ,𝐱,t)\hfill \end{array}$$
(19)
The most favorite strategy of finding these, known also as certainty-equivalence principle, is a two-stage design prescription. First, construct uncertainty-dependent feedback $`u(𝐱,𝜽,t)`$, $`𝜽\mathrm{\Omega }_\theta `$ which ensures boundedness of the trajectories $`𝐱(t)`$. Second, replace $`𝜽`$ with $`\widehat{𝜽}`$ in $`u(𝐱,𝜽,t)`$ and, given the constraints (e.g., $`\dot{𝐱}`$, $`𝜽`$ cannot be measured explicitly, while state $`𝐱`$ is available), design function $`𝒜_{\mathrm{lg}}:\times ^n\times _+^d`$ which guarantees (17), (18), or/and $`\psi (𝐱(t),t)0`$.
With this strategy, the design of the feedback $`u(𝐱,𝜽,t)`$ is generally independent<sup>8</sup><sup>8</sup>8In particular, it is the standard requirement that function $`u(𝐱,𝜽,t)`$ should guarantee Lyaponov stability of the system for $`\widehat{𝜽}=𝜽`$, while parameter adjustment algorithms use this property in order to ensure stability of the whole system. No other properties are required from the function $`u(𝐱,𝜽,t)`$. of the specific design of the parameter estimation algorithm $`𝒜_{\mathrm{lg}}(\psi ,𝐱,t)`$. This allows the full benefit of contemporary nonlinear control theory in designing feedbacks $`𝐮(𝐱,𝜽,t)`$. On the other hand, this strategy equally benefits from conventional parameter estimation and adaptation theories which provide a list of the ready-to-be-implemented algorithms under the assumption that feedback $`𝐮(𝐱,𝜽,t)`$ ensures stability of the system.
Ironically, the power of the certainty-equivalence principle – simplicity and relative independence of the stages of design – is also its Achilles’ heel. This principle does not take into account the possible interactions between stabilizing control and parameter estimation procedures. It has been reported in that an additional “interaction” term $`\widehat{𝜽}_P(𝐱,t):^n\times _+^d`$ added to the parameters $`𝜽`$ in function $`u(𝐱,𝜽,t)`$: $`u(𝐱,𝜽+\widehat{𝜽}_P(𝐱,t),t)`$ introduces new properties to the system. Unfortunately, straightforward introduction of this ”interaction” term as a new variable of the design affects its simplicity, internal order, and so much favored independence of the design stages (control and estimation).
An alternative strategy which introduces a new design paradigm is proposed in . Its main idea is that adaptation algorithms in (19) are initially allowed to depend on unmeasurable variables $`\dot{\psi },\dot{𝐱},𝜽`$
$$\dot{\widehat{𝜽}}=𝒜_{\mathrm{lg}}^{}(\psi ,\dot{\psi },𝐱,\dot{𝐱},𝜽,t)$$
(20)
For this reason we refer to such algorithms as virtual algorithms. If the desired properties (17), (18) are ensured with (20) then, taking into account properties of the vector-fields $`𝐟(𝐱,𝜽)`$, $`𝐠(𝐱)`$ in (3), we convert the unrealizable algorithm (20) into an equivalent representation in integro-differential, or finite, form :
$$\begin{array}{cc}& \widehat{𝜽}=\mathrm{\Gamma }(\widehat{𝜽}_P(𝐱,t)+\widehat{𝜽}_I(t)),\mathrm{\Gamma }^{d\times d},\mathrm{\Gamma }>0\hfill \\ & \dot{\widehat{𝜽}}_I=𝒜_{\mathrm{lg}}(\psi ,𝐱,t)\hfill \end{array}$$
(21)
This approach preserves the convenience of the certainty-equivalence principle, as the feedback $`u(𝐱,𝜽,t)`$ could, in principle, be built independently of the subsequent parameter adjustment procedure. At the same time, it provides the necessary interaction term $`\widehat{𝜽}_P(𝐱,t)`$ ensuring the required properties (17), (18) of the closed-loop system even if function $`f(𝐱,𝜽,t)`$ in (11) is nonlinear in $`𝜽`$.
In this paper we propose the following class of virtual adaptation algorithms<sup>9</sup><sup>9</sup>9Choice of the virtual algorithm in the form of equation (22) is motivated by our previous study of derivative-dependent algorithms for systems with uncertainties that are nonlinear in their parameters :
$$\dot{\widehat{𝜽}}=\mathrm{\Gamma }(\dot{\psi }+\phi (\psi ,𝝎,t))𝜶(𝐱,t)+𝒬(𝐱,\widehat{𝜽},t)(𝜽\widehat{𝜽}),\mathrm{\Gamma }^{d\times d},\mathrm{\Gamma }>0$$
(22)
where $`𝒬(𝐱,\widehat{𝜽},t):^n\times ^d\times _+^{d\times d}`$, $`𝒬()𝒞^0`$. As a candidate for finite form realization (21) of algorithms (22) we select the following set of equations:
$$\begin{array}{cc}\hfill \widehat{𝜽}(𝐱,t)& =\mathrm{\Gamma }(\widehat{𝜽}_P(𝐱,t)+\widehat{𝜽}_I(t));\mathrm{\Gamma }^{d\times d},\mathrm{\Gamma }>0\hfill \\ \hfill \widehat{𝜽}_P(𝐱,t)& =\psi (𝐱,t)𝜶(𝐱,t)\mathrm{\Psi }(𝐱,t)\hfill \\ \hfill \dot{\widehat{𝜽}}_I& =\phi (\psi (𝐱,t),𝝎,t)𝜶(𝐱,t)+(𝐱,\widehat{𝜽},u(𝐱,\widehat{𝜽},t),t),\hfill \end{array}$$
(23)
where function $`\mathrm{\Psi }(𝐱,t):^n\times _+_d`$, $`\mathrm{\Psi }(𝐱,t)𝒞^1`$ satisfies Assumption 5
###### Assumption 5
There exists function $`\mathrm{\Psi }(𝐱,t)`$ such that
$$\frac{\mathrm{\Psi }(𝐱,t)}{𝐱_2}\psi (𝐱,t)\frac{𝜶(𝐱,t)}{𝐱_2}=(𝐱,t),$$
(24)
where $`(𝐱,t):^n\times _+^{d\times p}`$ is either zero or, if $`𝐟_2(𝐱,𝛉)`$ is differentiable in $`𝛉`$, satisfies the following:
$$(𝐱,t)(𝐱,𝜽,𝜽^{})0𝜽,𝜽^{}\mathrm{\Omega }_\theta ,𝐱^n$$
$$(𝐱,𝜽,𝜽^{})=_0^1\frac{𝐟_2(𝐱,𝐬(\lambda ))}{𝐬}𝑑\lambda ,𝐬(\lambda )=𝜽^{}\lambda +𝜽(1\lambda )$$
Function $`(𝐱,\widehat{𝜽},u(𝐱,\widehat{𝜽},t),t):^n\times ^d\times \times _+^d`$ in (23) is given as follows:
$`(𝐱,u(𝐱,\widehat{𝜽},t),t)`$ $`=`$ $`\mathrm{\Psi }(𝐱,t)/t\psi (𝐱,t)(𝜶(𝐱,t)/t)`$ (25)
$`(\psi (𝐱,t)L_{𝐟_1}𝜶(𝐱,t)L_{𝐟_1}\mathrm{\Psi }(𝐱,t))(\psi (𝐱,t)L_{𝐠_1}𝜶(𝐱,t)L_{𝐠_1}\mathrm{\Psi }(𝐱,t))u(𝐱,\widehat{𝜽},t)`$
$`+(𝐱,t)(𝐟_2(𝐱,\widehat{𝜽})+𝐠_2(𝐱)u(𝐱,\widehat{𝜽},t)).`$
Functions $`\mathrm{\Psi }(𝐱,t)`$ and $`(𝐱,\widehat{𝜽},u(𝐱,\widehat{𝜽},t),t)`$ are introduced into (23) in order to shape the derivative $`\dot{\widehat{𝜽}}(𝐱,t)`$ to fit equation (22). The role of function $`\mathrm{\Psi }(𝐱,t)`$ in (23) is to compensate for the uncertainty-dependent term $`\psi (𝐱,t)L_{𝐟_2(𝐱,𝜽)}𝜶(𝐱,t)`$, and equation (24) is the condition when such compensation is possible<sup>10</sup><sup>10</sup>10We show below, in the proof of Theorem 1 (see Appendix), that Assumption 5 is indeed sufficient for the function $`\widehat{𝜽}(𝐱,t)`$ to be a realization of (22).. With the function $`(𝐱,\widehat{𝜽},u(𝐱,\widehat{𝜽},t),t)`$ we eliminate the influence of the uncertainty-independent vector fields $`𝐟_1(𝐱)`$, $`𝐠_1(𝐱)`$, and $`𝐠_2(𝐱)`$ on the desired form of the time-derivative $`\dot{\widehat{𝜽}}(𝐱,t)`$. The properties of system (3), together with control (9) and this new adaptation algorithm (23), (25), are summarized in Theorem 1 and Theorem 2.
###### Theorem 1 (Boundedness)
Let system (3), (11), (23), (25) be given and Assumptions 3,4,5 be satisfied. Then the following properties hold
P1) Let for the given initial conditions $`𝐱(t_0)`$, $`\widehat{𝛉}_I(t_0)`$ and parameters vector $`𝛉`$, interval $`[t_0,T^{}]`$ be the (maximal) time-interval of existence of solutions in the closed loop system (3), (11), (23), (25). Then
$$f(𝐱(t),𝜽,t)f(𝐱(t),\widehat{𝜽}(t),t))L_2^1[t_0,T^{}]$$
and
$$\begin{array}{cc}& f(𝐱(t),𝜽,t)f(𝐱(t),\widehat{𝜽}(t),t))_{2,[t_0,T^{}]}D_f(𝜽,t_0,\mathrm{\Gamma },\epsilon (t)_{2,[t_0,T^{}]});\hfill \\ & D_f(𝜽,t_0,\mathrm{\Gamma },\epsilon (t)_{2,[t_0,T^{}]})=\left(\frac{D}{2}𝜽\widehat{𝜽}(t_0)_{\mathrm{\Gamma }^1}^2\right)^{0.5}+\frac{D}{D_1}\epsilon (t)_{2,[t_0,T^{}]}\hfill \\ & 𝜽\widehat{𝜽}(t)_{\mathrm{\Gamma }^1}^2\widehat{𝜽}(t_0)𝜽_{\mathrm{\Gamma }^1}^2+\frac{D}{2D_1^2}\epsilon (t)_{2,[t_0,T^{}]}^2\hfill \end{array}$$
(26)
In addition, if Assumptions 1 and 2 are satisfied then
P2) $`\psi (𝐱(t),t)L_{\mathrm{}}^1[t_0,\mathrm{}]`$, $`𝐱(t)L_{\mathrm{}}^n[t_0,\mathrm{}]`$ and
$$\psi (𝐱(t),t)_{\mathrm{},[t_0,\mathrm{}]}\gamma _{\mathrm{},2}(\psi (𝐱_0,t_0),𝝎,D_f(𝜽,t_0,\mathrm{\Gamma },\epsilon (t)_{2,[t_0,\mathrm{}]})+\epsilon (t)_{2,[t_0,\mathrm{}]})$$
(27)
P3) if properties H1, H5 hold, and system (12) has $`L_2^1[t_0,\mathrm{}]L_p^1[t_0,\mathrm{}]`$, $`p>1`$ gain with respect to input $`\zeta (t)`$ and output $`\psi `$ then
$$\epsilon (t)L_2^1[t_0,\mathrm{}]L_{\mathrm{}}^1[t_0,\mathrm{}]\underset{t\mathrm{}}{lim}\psi (𝐱(t),t)=0$$
(28)
If, in addition, property H2 holds, and functions $`𝛂(𝐱,t)`$, $`\psi (𝐱,t)/t`$ are locally bounded with respect to $`𝐱`$ uniformly in $`t`$, then
P4) the following limiting relation holds
$$\underset{t\mathrm{}}{lim}f(𝐱(t),𝜽,t)f(𝐱(t),\widehat{𝜽}(t),t)=0$$
(29)
Proofs of Theorem 1 and subsequent results are given in the Appendix.
Before we proceed with discussion of the results of Theorem 1, we wish to comment on Assumption 5. Assumption 5 links the possibility to design the parameter adjustment algorithm in the form of equation (22), with the properties of functions $`𝜶(𝐱,t)`$ and $`\psi (𝐱,t)`$. These functions depend on the properties of nonlinearity $`f(𝐱,𝜽,t)`$ itself (function $`𝜶(𝐱,t)`$) and, importantly, on the chosen specification of the desired target set:
$$\{𝐱^n|\psi (𝐱,t)=0\}\mathrm{\Omega }_0$$
given by function $`\psi (𝐱,t)`$. Specific properties of the functions $`f(𝐱,𝜽,t)`$ and $`\psi (𝐱,t)`$ are interrelated through the possibility to solve partial differential equation (24) for the function $`\mathrm{\Psi }(𝐱,t)`$. Let $`(𝐱,t)=\mathrm{col}(_1(𝐱,t)`$,$`\mathrm{}`$, $`_d(𝐱,t))`$, and $`𝜶(𝐱,t)𝒞^2`$, $`𝜶(𝐱,t)=\mathrm{col}(\alpha _1(𝐱,t),\mathrm{},\alpha _d(𝐱,t))`$, then necessary and sufficient conditions for existence of the function $`\mathrm{\Psi }(𝐱,t)`$ follow from the Poincar$`\stackrel{´}{\mathrm{e}}`$ lemma:
$$\frac{}{𝐱_2}\left(\psi (𝐱,t)\frac{\alpha _i(𝐱,t)}{𝐱_2}+_i(𝐱,t)\right)=\left(\frac{}{𝐱_2}\left(\psi (𝐱,t)\frac{\alpha _i(𝐱,t)}{𝐱_2}+_i(𝐱,t)\right)\right)^T$$
(30)
This relation, in the form of conditions of existence of the solutions for function $`\mathrm{\Psi }(𝐱,t)`$ in (24), takes into account structural properties of system (3), (11). Indeed, let $`(𝐱,t)=0`$ and consider partial derivatives $`\alpha _i(𝐱,t)/𝐱_2`$, $`\psi (𝐱,t)/𝐱_2`$ with respect to vector $`𝐱_2=(x_{21},\mathrm{},x_{2p})^T`$. Let
$$\begin{array}{cc}\hfill \frac{\psi (𝐱,t)}{𝐱_2}& =\left(\begin{array}{cccccccc}0& 0& \mathrm{}& 0& & 0& \mathrm{}& 0\end{array}\right)\hfill \\ \hfill \frac{\alpha _i(𝐱,t)}{𝐱_2}& =\left(\begin{array}{cccccccc}0& 0& \mathrm{}& 0& & 0& \mathrm{}& 0\end{array}\right)\hfill \end{array}$$
(31)
where symbol $``$ denotes a function of $`𝐱`$ and $`t`$. Then condition (31) guarantees that equality (30) (and, subsequently, Assumption 5) holds. Whether or not Assumption 5 holds, depends, roughly speaking, on how large is the part of partition $`𝐱_2`$ that enters the arguments of functions $`\psi (𝐱,t)`$, $`𝜶(𝐱,t)`$. In the case of $`\alpha (𝐱_1𝐱_2,t)/𝐱_2=0`$, Assumption 5 holds for arbitrary $`\psi (𝐱,t)𝒞^1`$. If $`\psi (𝐱,t)`$, $`𝜶(𝐱,t)`$ depend on just a single component of $`𝐱_2`$, for instance $`x_{2k},k\{0,\mathrm{},p\}`$, then conditions (31) hold and function $`\mathrm{\Psi }(𝐱,t)`$ can be derived explicitly by integration
$$\mathrm{\Psi }(𝐱,t)=\psi (𝐱,t)\frac{𝜶(𝐱,t)}{x_{2k}}𝑑x_{2k}$$
(32)
In all other cases, the existence of the required function $`\mathrm{\Psi }(𝐱,t)`$ follows from (30).
The necessity to satisfy Assumption 5 may seem to be a critical restriction, which limits applicability of our approach. However, we notice that it holds in the relevant problem settings<sup>11</sup><sup>11</sup>11See, for example, the problem setting in for parameter estimation in the presence of nonlinear state-dependent parametrization. This problem setting, according to our knowledge, is by far one of the most general available in the literature. for arbitrary $`𝜶(𝐱,t),\psi (𝐱,t)𝒞^1`$. Consider, for instance , where the class of systems is restricted to (33):
$$\dot{x}=\varrho (x,u)x+f(𝜽,u,x),\varrho (x,t)>\varrho _{\mathrm{min}}>0,x$$
(33)
The dimension of the state in system (33) coincides with that of the uncertainty-dependent partition and equals to unit ($`dim\{𝐱\}=dim\{𝐱_2\}=1`$). Hence, according to (32), and in case functions $`\psi (x,t),𝜶(x,t)𝒞^1`$, there will always exist a function $`\mathrm{\Psi }(x,t)`$ satisfying equality (24) with $`(x,t)=0`$.
In the general case, when $`dim\{𝐱_2\}>1`$, the problems of finding function $`\mathrm{\Psi }(𝐱,t)`$ satisfying condition (24) can be avoided (or converted into one with an already known solutions such as (30), (32)) by the embedding technique proposed in . The main idea of the method is to introduce an auxiliary system
$$\begin{array}{cc}\hfill \dot{𝝃}& =𝐟_𝝃(𝐱,𝝃,t),𝝃^z\hfill \\ \hfill 𝐡_\xi & =𝐡_\xi (𝝃,t),^z\times _+^h\hfill \end{array}$$
(34)
such that
$$f(𝐱(t),𝜽,t)f(𝐱_1(t)𝐡_\xi (t)𝐱_2^{}(t),𝜽,t)L_2^1[t_0,\mathrm{}]$$
(35)
and $`dim\{𝐡_\xi \}+dim\{𝐱_2^{}\}=p`$. Then (11) can be rewritten as follows:
$`\dot{\psi }=f(𝐱_1𝐡_\xi 𝐱_2^{},𝜽,t)f(𝐱_1𝐡_\xi 𝐱_2^{},\widehat{𝜽},t)\phi (\psi ,𝝎,t)+\epsilon _\xi (t),`$ (36)
where $`\epsilon _\xi (t)L_2^1[t_0,\mathrm{}]`$, and $`dim\{𝐱_2^{}\}=ph<p`$. In principle, the dimension of $`𝐱_2^{}`$ could be reduced to $`1`$ or $`0`$. As soon as this is ensured, Assumption 5 will be satisfied and the results of Theorem 1 follow. Sufficient conditions ensuring the existence of such an embedding in general case are provided in . For systems in which the parametric uncertainty can be reduced to vector fields with low-triangular structure the embedding is given in .
An alternative way to construct system (34) with the desired properties is to use (possible, high-gain, discontinuous) robust observers. In order to illustrate this approach consider the rather general case when function $`𝐟_2(𝐱,𝜽)`$ in (3) is given as $`𝐟_2(𝐱,𝜽)=\overline{𝐟}_2(𝐱)+\mathit{\varphi }(𝐱,𝜽)`$, and function $`\mathit{\varphi }(𝐱,𝜽)`$ is bounded.
Let in addition there exist continuous functions $`𝐡_ϵ:^p^p`$, $`𝐡_\xi :^p^p`$ such that the following inequality is satisfied
$$𝐡_ϵ(𝝃𝐱_2)|f(𝐱_1𝐡_\xi (𝝃),𝜽,t)f(𝐱_1𝐱_2,𝜽,t)|$$
(37)
As a candidate for yet unknown tracking system (34) we select the following
$$\dot{𝝃}=\overline{𝐟}(𝐱)+𝐟_ϵ(𝝃𝐱_2)+𝐠_2(𝐱)u+𝝊$$
(38)
where function $`𝐟_ϵ:^p^p`$ and auxiliary input $`𝝊^p`$ are the design parameters. Subtracting equations for $`\dot{𝐱}_2`$ in (3) from (38) yields:
$$\begin{array}{cc}\hfill \dot{\mathit{ϵ}}& =𝐟_ϵ(\mathit{ϵ})\mathit{\varphi }(𝐱(t),𝜽)+𝝊\hfill \\ \hfill 𝐲_ϵ& =𝐡_ϵ(\mathit{ϵ})\hfill \end{array}$$
(39)
where $`\mathit{ϵ}=𝝃𝐱_2`$. Let us finally choose the function $`𝐟_ϵ`$ in (39) such that the system $`\dot{\mathit{ϵ}}=𝐟_ϵ(\mathit{ϵ})+𝝊`$ is strictly passive with a positive definite storage function $`V(\mathit{ϵ},t):`$
$$\dot{V}(\mathit{ϵ},t)𝐲_ϵ^T𝝊\beta 𝐲_ϵ^2,\beta >0$$
(40)
According to <sup>12</sup><sup>12</sup>12In one extra assumption on the function $`𝐟_ϵ(\mathit{ϵ})`$ in (39) is imposed. In particular it is required that the system $`\dot{\mathit{ϵ}}=𝐟_ϵ(\mathit{ϵ})+\upsilon `$ is strongly zero-detectable with respect to inputs $`\upsilon `$ and output $`𝐲_ϵ`$. In our case, however, the limiting relations $`lim_t\mathrm{}𝐲_ϵ(t)=0`$, $`lim_t\mathrm{}\mathit{ϵ}(t)=0`$ are not necessary. Therefore, as follows from the proof of Theorem 2 in , in order to show just $`𝐲_\epsilon (t)L_2^p[t_0,\mathrm{}]`$ the assumption of strong zero-delectability can be omitted.(page 1484, Theorem 2) inequality (40) guarantees that there always exists input $`𝝊(t)`$ in (38) such that $`𝐡_ϵ(\mathit{ϵ}(t))L_2^p[t_0,\mathrm{}]`$. Then taking into account (37) we can conclude that condition (35) holds (with $`𝐱_2^{}:dim\{𝐱_2^{}\}=0`$). This implies that the original error model (11) can be converted into (36), which in our case satisfies Assumption 5 ($`\alpha _i(𝐱,t)𝐱_2=0`$ for the corresponding $`\alpha _i(𝐱,t)`$ in (36)).
Let us now briefly comment on the results of Theorem 1. The theorem ensures a set of relevant properties for both control (P2, P3) and parameter estimation problems (P1, P4). These properties, as illustrated with (26)–(29), provide conditions for boundedness of the solutions $`𝐱(t,𝐱_0,t_0,𝜽,u(t))`$, reaching the target set $`\mathrm{\Omega }_0`$, and exact compensation of the uncertainty term $`f(𝐱,𝜽,t)`$ even in the presence of unknown disturbances $`\epsilon (t)L_2^1[t_0,\mathrm{}]L_{\mathrm{}}^1[t_0,\mathrm{}]`$. These characterizations are the consequence of the fact that $`(f(𝐱(t),𝜽,t)f(𝐱(t),\widehat{𝜽}(t),t)))L_2^1[t_0,\mathrm{}]`$, which in turn is guaranteed by properties (14), (15), (16) of the function $`f(𝐱,𝜽,t)`$ in Assumptions 3, 4. Among these properties, estimate (16) in Assumption 4 is particulary important for allowing disturbances (potentially unbounded) from $`L_2^1[t_0,\mathrm{}]`$. When no disturbances are present it is possible to show that P1–P4 hold without involving Assumption 4.
###### Corollary 1
Let system (3), (11),(23), (25) be given, $`\epsilon (t)=0`$, and Assumptions 3,5 hold. Then
P5) norm $`𝛉\widehat{𝛉}(t)_{\mathrm{\Gamma }^1}^2`$ is non-increasing and properties P1–P4<sup>13</sup><sup>13</sup>13In this case, however, the bound for $`\psi (𝐱(t),t)_{\mathrm{},[t_0,\mathrm{}]}`$ will be different from the one given by equation (27) in Theorem 1. Its new estimate is given by formula (77) in Appendix 1 of Theorem 1 hold with $`\epsilon (t)=0`$ respectively.
In addition to the fact that $`|f(𝐱,𝜽,t)f(𝐱,\widehat{𝜽},t)|`$ is not required to be bounded from below as in (16), Corollary 1 ensures that $`𝜽\widehat{𝜽}(t)_{\mathrm{\Gamma }^1}^2`$ is not growing with time when $`\epsilon (t)=0`$. The practical relevance of the corollary is that it will allow us to guarantee desired convergence (18) with a much weaker, local version of Assumption 4. It will also help us to establish conditions for (semi-global) exponential stability in the unperturbed system, which in turn will enable (small) disturbances from $`L_{\mathrm{}}^1[t_0,\mathrm{}]`$ in the right-hand side of (11).
Another consequence of Theorem 1 concerns the specific case when $`\epsilon (t)L_{\mathrm{}}^1[t_0,\mathrm{}]L_2^1[t_0,\mathrm{}]`$.
###### Corollary 2
Let system (3), (11),(23), (25) be given, Assumptions 1, 35 hold, $`\epsilon (t)L_{\mathrm{}}^1[t_0,\mathrm{}]L_2^1[t_0,\mathrm{}]`$, and property H3 holds. Let in addition, system (12) has $`L_p^1[t_0,\mathrm{}]L_{\mathrm{}}^1[t_0,\mathrm{}]`$, $`p2`$ gain. Then
P6) $`\psi (𝐱(t),t)L_{\mathrm{}}^1[t_0,\mathrm{}]`$, $`𝐱(t)L_{\mathrm{}}^n[t_0,\mathrm{}]`$;
P7) if properties H1, H5 hold, and system (12) has $`L_p^1[t_0,\mathrm{}]L_q^1[t_0,\mathrm{}]`$, $`q>1`$ gain with respect to input $`\zeta (t)`$ and state $`\psi `$, then $`lim_t\mathrm{}\psi (𝐱(t),t)=0`$;
If in addition property H2 is satisfied, functions $`𝛂(𝐱,t)`$, $`\psi (𝐱,t)/t`$ are locally bounded with respect to $`𝐱`$ uniformly in $`t`$, then limiting relation (29) holds as well.
Corollary 2 extends applicability of algorithms (23), (25) to systems (11) with defined $`L_p^1[t_0,\mathrm{}]L_{\mathrm{}}^1[t_0,\mathrm{}]`$ gains for arbitrary $`p2`$.
Let us formulate conditions ensuring convergence of the estimates $`\widehat{𝜽}(t)`$ to $`𝜽`$ in the closed loop system (3), (11), (23), (25). When the mathematical model of the uncertainties is linear in its parameters, i.e. $`f(𝐱,𝜽,t)=𝜻(𝐱,t)^T𝜽`$, the usual requirement for convergence is that signal $`𝜻(𝐱(t),t)`$ is persistently exciting :
###### Definition 1 (Persistent Excitation)
Let function $`𝛇:_+^k`$ be given. Function $`𝛇(t)`$ is said to be persistently exciting iff there exist constants $`\delta >0`$ and $`L>0`$ such that for all $`t_+`$ the following holds
$$_t^{t+L}𝜻(\tau )𝜻(\tau )^T𝑑\tau \delta I$$
(41)
The conventional notion of persistent excitation requires specific properties (i.g. the integral inequality (41)) from signal $`𝜻(t)`$ as a function of time. In the closed loop system, however, relevant signals in the model of uncertainty $`f(𝐱,𝜽,t)`$ can depend on state $`𝐱`$ and parameters. In particular, they depend on initial conditions, parameters of the feedback, and initial time $`t_0`$. In order to address this issue it is suggested in to use the notion of uniform persistent excitation:
###### Definition 2 (Uniform Persistent Excitation)
Let function $`𝛇:^n\times _+^k`$ be given, and $`𝐱(t,𝐱_0,t_0,𝛉_0)`$ be a solution of (3), where the vector $`𝛉_0^s`$ stands for all possible parameters of (3) and feedback (9), (23), (25). Function $`𝛇(𝐱(t,𝐱_0,t_0,𝛉_0),t)`$ is said to be uniformly persistently exciting iff there exist constants $`\delta >0`$ and $`L>0`$ such that for all $`t,t_0_+`$, $`𝐱_0^n`$, $`𝛉_0^s`$ the following holds
$$_t^{t+L}𝜻(𝐱(\tau ,𝐱_0,t_0,𝜽_0),\tau )𝜻(𝐱(\tau ,𝐱_0,t_0,𝜽_0),\tau )^T𝑑\tau \delta I$$
(42)
When dealing with nonlinear parameterization, it is also useful to have a characterization which takes into account nonlinearity in the model. In the linear case, persistent excitation of signal $`𝜻(𝐱(t),t)`$ (inequality (41)) implies that the following property holds
$$t^{}[t,t+L]:|𝜻(𝐱(t^{}),t^{})^T(𝜽_1𝜽_2)|\delta 𝜽_1𝜽_2$$
(43)
In the other words, the difference $`|𝜻(𝐱(t),t)^T(𝜽_1𝜽_2)|`$ is proportional to the distance $`𝜽_1𝜽_2`$ in parameter space for some $`t^{}[t,t+L]`$. In the nonlinear case it is natural to replace the linear term $`𝜻(𝐱(t^{}),t^{})^T(𝜽_1𝜽_2)`$ in (43) with its nonlinear substitute $`f(𝐱(t^{}),𝜽_1,t^{})f(𝐱(t^{}),𝜽_2,t^{})`$ as has been done, for example, in for systems with convex/concave parametrization. It is also natural to replace the proportion $`\delta 𝜽_1𝜽_2`$ in the right-hand side of (43) with a nonlinear function. Therefore, as a candidate for the nonlinear persistent excitation condition we propose the following notion:
###### Definition 3 (Nonlinear Persistent Excitation)
The function $`f(𝐱(t),𝛉,t):^n\times ^d\times _+`$ is said to be persistently excited with respect to parameters $`𝛉\mathrm{\Omega }_\theta ^d`$ iff there exist constant $`L>0`$ and function $`\varrho :_+_+,\rho 𝒦C^0`$ such that for all $`t_+`$, $`𝛉_1,𝛉_2\mathrm{\Omega }_\theta `$ the following holds:
$$t^{}[t,t+L]:|f(𝐱(t^{}),𝜽_1,t^{})f(𝐱(t^{}),𝜽_2,t^{})|\varrho (𝜽_1𝜽_2)$$
(44)
Properties (41) and (44) in Definitions 1 and 3 can be considered as alternative characterizations of excitation in dynamical systems. While inequality (41) accounts for specific properties of the signals in the uncertainty, inequality (44) accounts for possibility to detect parametrical difference from the difference $`f(𝐱(t),𝜽_1,t)f(𝐱(t),𝜽_2,t)`$. Taking into account these two equally possible but still rather distinct characterizations of excitation in nonlinear systems, in Theorem 2 below we present a set of alternatives for parameter convergence in system (3), (10), (23), (25).
###### Theorem 2 (Convergence)
Let system (3), (10), (23), (25) satisfy Assumptions 13. Let, in addition, Assumption 5 hold with $`(𝐱,t)=0`$. Then $`𝐱(t)L_{\mathrm{}}^n[t_0,\mathrm{}]`$, $`\widehat{𝛉}(t)L_{\mathrm{}}^d[t_0,\mathrm{}]`$. Moreover the limiting relation:
$$\underset{t\mathrm{}}{lim}\widehat{𝜽}(𝐱(t),t)=𝜽$$
is ensured if $`𝛂(𝐱,t)`$ is locally bounded in $`𝐱`$ uniformly in $`t`$, and one of the following alternatives hold:
1) function $`𝛂(𝐱(t),t)`$ is persistently exciting, and hypothesis H4 holds;
2) function $`f(𝐱(t),𝛉,t)`$ is nonlinearly persistently exciting, i. e. it satisfies condition (44); it satisfies hypotheses H1, H2; function $`\phi (\psi ,𝛚,t)`$ satisfies H5; function $`\psi (𝐱,t)/t`$ be locally bounded in $`𝐱`$ uniformly in $`t`$;
In case alternative 1) is satisfied, the estimates $`\widehat{𝛉}(𝐱(t),t)`$ converge to $`𝛉`$ exponentially fast. If, in addition, $`𝛂(𝐱(t),t)`$ is uniformly persistently exciting and Assumption 4 holds, then convergence is uniform. The rate of convergence can be estimated as follows:
$$\widehat{𝜽}(t)𝜽e^{\rho t}\widehat{𝜽}(t_0)𝜽D_\mathrm{\Gamma }$$
(45)
$$\rho =\frac{\delta D_1\lambda _{\mathrm{min}}(\mathrm{\Gamma })}{2L(1+\lambda _{\mathrm{max}}^2(\mathrm{\Gamma })L^2D^2\alpha _{\mathrm{}}^4)},D_\mathrm{\Gamma }=\left(\frac{\lambda _{\mathrm{max}}(\mathrm{\Gamma })}{\lambda _{\mathrm{min}}(\mathrm{\Gamma })}\right)^{\frac{1}{2}},\alpha _{\mathrm{}}=\underset{𝐱𝐱(t)_{\mathrm{},[t_0,\mathrm{}]},tt_0}{sup}𝜶(𝐱,t)$$
Notice that Theorem 2 considers error models (10) where no disturbance term $`\epsilon (t)`$ is present. Despite this Theorem 2 can be straightforwardly extended to error models with disturbance (11). Indeed, as follows from alternative 1), the parameter estimation subsystem becomes exponentially stable in case function $`\alpha (𝐱(t),t)`$ is (uniformly) persistently exciting. This in turn allows (sufficiently small) additive disturbances in the right-hand side of (10). In case the excitation is uniform, convergence of the estimates $`\widehat{𝜽}(t)`$ to a neighborhood of $`𝜽`$ is guaranteed for every $`\epsilon (t)L_{\mathrm{}}^1[t_0,\mathrm{}]`$ by inverse Lyapunov stability theorems .
In case of alternative 2), nonlinear persistent excitation condition (44) guarantees convergence (18) without invoking Assumption 4 or H4. In this case, however, the convergence may not be robust, which seems to be a natural tradeoff between generality of nonlinear parameterizations $`f(𝐱,𝜽,t)`$ and robustness with respect to unknown disturbances $`\epsilon (t)`$.
## 5 Discussion
So far we have shown that, for the class of nonlinearly parameterized systems, there exist a control function and parameter adjustment algorithms, such that solutions of the whole system are bounded and parametric uncertainty is decreasing in time. We have shown also that in case of persistently excited functions $`𝜶(𝐱,t)`$ the estimates $`\widehat{𝜽}(t)`$ in (23) converge exponentially fast to vector $`𝜽`$. These results, however, are not necessarily limited to functions satisfying Assumptions 3 or 4. Due to space limitations, however, we provide just the main ideas of possible extensions, leaving out the technical details. Let us first examine the case where these assumptions hold only in some domains of the system state space.
Nonlinear functions satisfying Assumptions 3, 4 in a domain of $`^n`$. Let, in particular, for the given nonlinear function $`f(𝐱,𝜽,t)`$ there exits the following partition of the state space:
$$\mathrm{\Omega }_𝐱=\mathrm{\Omega }_M\mathrm{\Omega }_A,\mathrm{\Omega }_M=\underset{j}{}\mathrm{\Omega }_{M,j},\mathrm{\Omega }_A=\mathrm{\Omega }_𝐱/\mathrm{\Omega }_M$$
where $`\mathrm{\Omega }_{M,j}`$ are the subsets of $`^n`$ where Assumptions 3, 4 are satisfied for every $`𝜽\mathrm{\Omega }_\theta `$ with the corresponding functions $`𝜶_j(𝐱,t)`$ and constants $`D_j`$, $`D_{1,j}`$. Let us also assume that $`\mathrm{\Omega }_M`$ contains an open set.
A typical example of a nonlinear function which satisfies this assumption is $`\mathrm{sin}(\theta x)`$, where the unknown parameter $`\theta `$ belongs to a bounded interval. Let, for instance, the system dynamics is given by
$$\begin{array}{cc}\hfill \dot{x}_1=& x_2\hfill \\ \hfill \dot{x}_2=& \mathrm{sin}(\theta x_1)+u,\hfill \end{array}$$
(46)
where parameter $`\theta \mathrm{\Omega }_\theta =[0.6,1.4]`$ is unknown a-priori. For the given bounds of $`\mathrm{\Omega }_\theta `$ the domain $`\mathrm{\Omega }_M`$ can be derived as follows:
$$\begin{array}{cc}\hfill \mathrm{\Omega }_M& =\{𝐱|x_1[3.38,2.59]\}\{𝐱|x_1[1.14,1.14]\}\{𝐱|x_1[2.59,3.38]\}\hfill \\ & =\mathrm{\Omega }_{M,1}\mathrm{\Omega }_{M,2}\mathrm{\Omega }_{M,3}\hfill \end{array}$$
(47)
and the function $`𝜶(𝐱,t)`$, satisfying Assumptions 3, 4 in $`\mathrm{\Omega }_M`$ is defined as
$$𝜶(𝐱,t)=\{\begin{array}{cc}x_1,\hfill & 𝐱\mathrm{\Omega }_{M,2}\hfill \\ x_1\hfill & 𝐱\mathrm{\Omega }_{M,1}\mathrm{\Omega }_{M,3}\hfill \end{array}$$
Another example is $`x^\theta `$, $`\theta [t_0,\mathrm{})`$. The last parametrization is widely used in modelling physical “power law” phenomena in nature (see, for example , where this function models effects of nonlinear damping in muscles).
The fact that Assumptions 3, 4 hold in the domain $`\mathrm{\Omega }_M^n`$, allows us to guarantee decrease of the norm $`𝜽\widehat{𝜽}(t)_{\mathrm{\Gamma }^1}^2`$ only if the state belongs to $`\mathrm{\Omega }_M`$. Therefore, extra control is needed in order to steer state $`𝐱`$ back into the domain $`\mathrm{\Omega }_M`$. Let us, for example, pick point $`𝐱^{}\mathrm{\Omega }_M`$, such that $`\mathrm{dist}\{𝐱^{},\mathrm{\Omega }_A\}>r,r_+`$. Let, in addition, there exists control function $`u_0(𝐱,t)`$ such that it steers state $`𝐱`$ of system (3) from the initial point $`𝐱_0`$ into the $`\delta _0`$-neighborhood $`U(\delta _0,𝐱^{})`$ of $`𝐱^{}`$ in finite time $`T_0(𝐱_0)`$. Suppose also that $`\delta _0<r`$. As follows from Theorems 1, 2, $`\widehat{𝜽(t)}`$ is bounded for every segment of the solution which starts from $`U(\delta _0,𝐱^{})`$ at $`t=t_i`$ and leaves the domain $`U(r,𝐱^{})`$ at $`t=t_{i+1}`$. Furthermore the bound for $`\widehat{𝜽}_{\mathrm{},[t_i,t_{i+1}]}`$ can be estimated a-priory from the bounds on $`𝜽`$ and $`\epsilon (t)_{2,[t_0,\mathrm{}}`$ (see also (26)). Given that the right-hand side of system (3), (9), (23), (25) is locally bounded we can conclude that the time interval $`t_{i+1}t_i`$ will always be separated from zero. Taking into account the results of Theorem 2, equation (45), we may conclude that sufficiently high excitation, defined by the ratio $`\delta /L`$, will guarantee that $`\widehat{𝜽}(t_{i+1})𝜽<\kappa \widehat{𝜽}(t_i)𝜽`$, $`\kappa ,0<\kappa <1`$. If the time sequence $`\{t_i\}`$ is infinite (i.e. the system always escapes the ball $`U(r,𝐱^{})`$) then convergence is asymptotic. In case the sequence $`\{t_i\}`$ is finite (i.e. $`t^{}>0:𝐱(t)\mathrm{\Omega }_Mt>t^{}`$) convergence is exponential, this follows from Theorem 2.
In principle, the size of $`\mathrm{\Omega }_M`$ and its location in $`^n`$ depend on the bounds of $`\mathrm{\Omega }_\theta `$. In fact, the larger the bounds, the smaller the volume of $`\mathrm{\Omega }_M`$. Moreover, the size of $`\mathrm{\Omega }_M`$ as a function of the bounds of $`\mathrm{\Omega }_\theta `$ depends on specific properties of nonlinearity $`f(𝐱,𝜽,t)`$. These observations suggest that in order to handle a broader class of nonlinearities (or functions with higher degree of uncertainty in $`𝜽`$) within the strategy proposed above, one needs to increase the excitation in functions $`𝜶_j(𝐱,t)`$. This is consistent with previously reported results on parameter convergence in nonlinearly parameterized systems. Whether the extension of the class of nonlinearities to more general functions renders it necessary to increase excitation, however, is still an open issue<sup>14</sup><sup>14</sup>14An example is given in , where nonlinear persistent excitation condition holds for the given parametrization, while linear persistent excitation condition for linear parametrization with respect to the same parameter-independent function is not satisfied..
Functions $`f(𝐱,𝛉,t)`$ with nonlinear incremental growth rates in $`𝛉`$. Another direction to extend the class nonlinear functions suitable for our method is to allow nonlinear bounds for the growth rates in (14), (16) in Assumptions 3, 4. The most straightforward generalizations, which do not change dramatically the machinery of technical proofs of Theorems 1, 2, are provided in Assumptions 6, 7 below:
###### Assumption 6
For the given function $`f(𝐱,𝛉,t)`$ in (11) there exist function $`𝛂(𝐱,t):^n\times _+^d,𝛂(𝐱,t)𝒞^1`$, function $`𝛔:^d^d`$, $`𝛔(𝐳)=(\sigma _1(z_1),\sigma _2(z_2),\mathrm{},\sigma _d(z_d))^T`$
$$\sigma _i(\xi )\xi 0,\underset{S\mathrm{}}{lim}_0^S\sigma _i(\xi )𝑑\xi =\mathrm{},$$
and function $`\overline{\gamma }𝒦`$ such that
$$(f(𝐱,\widehat{𝜽},t)f(𝐱,𝜽,t))𝜶(𝐱,t)^T𝝈(\widehat{𝜽}𝜽)0$$
(48)
$$|f(𝐱,\widehat{𝜽},t)f(𝐱,𝜽,t)|\overline{\gamma }(|𝜶(𝐱,t)^T𝝈(𝜽\widehat{𝜽})|)$$
(49)
###### Assumption 7
For the given function $`f(𝐱,𝛉,t)`$ in (11) and function $`𝛂(𝐱,t)`$ satisfying Assumption 6 there exists function $`\underset{¯}{\gamma }𝒦`$ such that
$$\underset{¯}{\gamma }(|𝜶(𝐱,t)^T𝝈(𝜽\widehat{𝜽})|)|f(𝐱,𝜽,t)f(𝐱,\widehat{𝜽},t)|$$
Choosing for simplicity $`\mathrm{\Gamma }=I`$, denoting $`\mathrm{\Delta }f(\widehat{𝜽},𝜽,𝐱,t)=|f(𝐱,\widehat{𝜽},t)f(𝐱,𝜽,t)|`$, letting $`|\epsilon (t)|`$ in (11) be such that $`_0^{\mathrm{}}\gamma _ϵ(|\epsilon (\tau )|)𝑑\tau <\mathrm{}`$, $`\gamma _ϵ𝒦`$ and replacing $`V_{\widehat{𝜽}}(\widehat{𝜽},𝜽,t)`$ in (67) (proof, of Theorem 1, Appendix 1) with
$$V_{\widehat{𝜽}}(\widehat{𝜽},𝜽,t)=\underset{i=1}{\overset{d}{}}_0^{\sigma (\widehat{\theta }_i\theta _i)}\sigma (\xi )𝑑\xi +_t^{\mathrm{}}\gamma _ϵ(|\epsilon (\tau )|)𝑑\tau $$
we can obtain the following estimate:
$`\dot{V}`$ $`=`$ $`𝝈(\widehat{𝜽}𝜽)^T𝜶(𝐱,t)(f(𝐱,\widehat{𝜽},t)f(𝐱,𝜽,t))+\epsilon (t)𝝈(\widehat{𝜽}𝜽)^T𝜶(𝐱,t)\gamma _ϵ(|\epsilon (t)|)`$ (50)
$``$ $`\overline{\gamma }^1(\mathrm{\Delta }f(\widehat{𝜽},𝜽,𝐱,t))\mathrm{\Delta }f(\widehat{𝜽},𝜽,𝐱,t)+|\epsilon (t)|\underset{¯}{\gamma }^1(\mathrm{\Delta }f(\widehat{𝜽},𝜽,𝐱,t))\gamma _ϵ(|\epsilon (t)|)`$
Boundedness of $`\widehat{𝜽}`$ follows from (50) if we can resolve the following inequality for unknown $`\gamma _ϵ(|\epsilon (t)|)`$:
$$\overline{\gamma }^1(\mathrm{\Delta }f(\widehat{𝜽},𝜽,𝐱,t))\mathrm{\Delta }f(\widehat{𝜽},𝜽,𝐱,t)+|\epsilon (t)|\underset{¯}{\gamma }^1(\mathrm{\Delta }f(\widehat{𝜽},𝜽,𝐱,t))\gamma _ϵ(|\epsilon (t)|)0,$$
If in addition there exists $`\gamma _f𝒦`$ such that
$$\overline{\gamma }^1(\mathrm{\Delta }f(\widehat{𝜽},𝜽,𝐱,t))\mathrm{\Delta }f(\widehat{𝜽},𝜽,𝐱,t)+|\epsilon (t)|\underset{¯}{\gamma }^1(\mathrm{\Delta }f(\widehat{𝜽},𝜽,𝐱,t))\gamma _ϵ(|\epsilon (t)|)\gamma _f(|\mathrm{\Delta }f(\widehat{𝜽},𝜽,𝐱,t)|)$$
(51)
then we can guarantee that $`\mathrm{\Delta }f(\widehat{𝜽}(t),𝜽,𝐱(t),t)L_{\gamma _f}^1[t_0,\mathrm{}]`$, where $`L_{\gamma _f}^n[t_0,\mathrm{}]`$ is the space of all functions $`𝐟_0(t):_+^n`$ with finite integral $`_0^{\mathrm{}}\gamma _f(𝐟_0(t))𝑑t\mathrm{}`$. Therefore, if system (12) has $`L_{\gamma _f}^1[t_0,\mathrm{}]L_{\gamma _ϵ}^1L_{\mathrm{}}^1`$ gain, we can conclude, invoking this new modified Assumption 2, that $`\psi (𝐱(t),t)L_{\mathrm{}}^1`$, $`𝐱(t)L_{\mathrm{}}^n`$.
Notice that letting functions $`\underset{¯}{\gamma }`$, $`\overline{\gamma }`$ linear ($`\underset{¯}{\gamma }(|s|)=D|s|`$, $`\overline{\gamma }(|s|)=D_1|s|`$), allows us straightforwardly obtain, as in (8), that choice $`\gamma _ϵ=\frac{D}{4D_1^2}\epsilon ^2`$ ensures the following inequality:
$$\dot{V}\frac{1}{D}\left(|f(𝐱,\widehat{𝜽},t)f(𝐱,𝜽,t)|\frac{D}{2D_1}\epsilon (t)\right)^2$$
This implies that properties similar to P1)–P7) can be derived for the case where Assumptions 3, 4 are replaced with Assumptions 6, 7 and functions $`\overline{\gamma }()`$, $`\underset{¯}{\gamma }()`$ are linear (for the proofs of Theorem 1, and Corollaries 1, 2 see the Appendix). Parameter convergence in this case can also be deduced from Theorem 2, alternative 2). Notice that, due to the nonlinearities in $`𝝈(𝜽\widehat{𝜽})`$, convergence in general may not be exponentially fast.
For the nonlinear functions $`\underset{¯}{\gamma }`$, $`\overline{\gamma }`$ in Assumptions 6, 7 the formulation of the results will require the notions of $`L_{\gamma _f}`$ spaces introduced above. The machinery behind the statements, however, will remain the same. The practical relevance of these results with nonlinear functions $`\underset{¯}{\gamma }`$, $`\overline{\gamma }`$ is that they enable us to take into account the specific properties of the signal $`\epsilon (t)`$ when designing control, adaptation and parameter estimation procedures. If, for example, disturbance $`\epsilon (t)`$ is due to observers, we might derive requirements on convergence rates for the observer-induced errors $`\epsilon (t)`$ (i. e. $`\epsilon (t)L_{\gamma _ϵ}^1[t_0,\mathrm{}]`$, and $`\gamma _ϵ()`$ satisfies inequality (51)). Given these rates and the fact that $`\mathrm{\Delta }f(𝐱(t),\widehat{𝜽}(t),t)L_{\gamma _f}^1[t_0,\mathrm{}]`$, the target dynamics
$$\dot{\psi }=\phi (\psi ,𝝎,t)+\zeta (t),\zeta (t)L_{\gamma _f}^1[t_0,\mathrm{}]L_{\gamma _ϵ}^1[t_0,\mathrm{}]$$
should be chosen in order to guarantee boundedness of $`\psi (t)`$ for all $`\zeta (t)L_{\gamma _f}^1[t_0,\mathrm{}]L_{\gamma _ϵ}^1[t_0,\mathrm{}]`$. This will allow synergy at all stages of the design and analysis of adapting systems.
Singularities in control (9), and non-affine models. In the problem statement we restricted the class of nonlinear systems of interest models (3) that are affine in control and furthermore, we assumed that inverse $`\left(L_{𝐠(𝐱)}\psi (𝐱,t)\right)^1`$ exists everywhere. Even though this restriction holds in wide variety of practically relevant situations, the question is whether the proposed approach could be extended to more general classes of systems. Let us, for instance, assume that either $`L_{𝐠(𝐱)}\psi (𝐱,t)=0`$ for some $`𝐱^n`$, or the right-hand side of (3) is not affine in control, e.g.
$$\dot{𝐱}=𝐟(𝐱,𝜽,u)$$
(52)
Obviously, control function (9), which transforms (3) into (10), is not relevant any more. Despite that, it is still possible to transform system (3) into an error model, similar to (10). In order realize this transformation without invoking the use of linearity in the control or taking inverse $`\left(L_{𝐠(𝐱)}\psi (𝐱,t)\right)^1`$, it should be possible to find a function $`u(𝐱,𝜽,𝝎,t)`$ such that the following invariance condition is satisfied:
$$\frac{\psi (𝐱,t)}{𝐱}𝐟(𝐱,𝜽,u(𝐱,𝜽,𝝎,t))=\phi (\psi (𝐱,t),𝝎,t)\frac{\psi (𝐱,t)}{t}$$
(53)
Denoting
$$\frac{\psi (𝐱,t)}{𝐱}𝐟(𝐱,𝜽,u(𝐱,\widehat{𝜽},𝝎,t))=f^{}(𝐱,𝜽,\widehat{𝜽},𝝎,t)$$
(54)
and taking into account (52), (53), and (54) we can calculate derivative $`\dot{\psi }`$ in the following form
$`\dot{\psi }`$ $`=`$ $`f^{}(𝐱,𝜽,\widehat{𝜽},𝝎,t)+f^{}(𝐱,𝜽,𝜽,𝝎,t)f^{}(𝐱,𝜽,𝜽,𝝎,t)+{\displaystyle \frac{\psi (𝐱,t)}{t}}`$ (55)
$`=`$ $`\phi (\psi (𝐱,t),𝝎,t)+f^{}(𝐱,𝜽,𝜽,𝝎,t)f^{}(𝐱,𝜽,\widehat{𝜽},𝝎,t)`$
The main difference between error models (55) and (10) is that function $`f^{}(𝐱,𝜽,\widehat{𝜽},𝝎,t)`$ in (55) depends on additional parameters $`𝜽`$, $`𝝎`$. Despite this difference our approach can still be applied to models (55) if inequalities (14), (16) in Assumption 3 (or/and 4) hold for function $`f^{}(𝐱,𝜽,\widehat{𝜽},𝝎,t)`$ for any $`𝝎\mathrm{\Omega }_\omega `$. Adaptation algorithms in this case can straightforwardly be derived from (8), (8) (in Appendix 1) and will have the form similar to (23), (25).
In the next section we illustrate the application to and main steps in the design of our algorithms for the optimal slip identification problem in brake control systems.
## 6 Example
Consider the problem of minimizing the braking distance for a single wheel rolling along a surface. The surface properties can vary depending on the current position of the wheel. The wheel dynamics can be given by the following system of differential equations :
$`\dot{x}_1`$ $`=`$ $`{\displaystyle \frac{1}{m}}F_s(F_n,𝐱,\theta ),`$
$`\dot{x}_2`$ $`=`$ $`{\displaystyle \frac{1}{J}}(F_s(F_n,𝐱,\theta )ru)`$
$`\dot{x}_3`$ $`=`$ $`{\displaystyle \frac{1}{x_1}}(({\displaystyle \frac{1}{m}}(1x_3)+{\displaystyle \frac{r^2}{J}})F_s(F_n,𝐱,\theta ){\displaystyle \frac{r}{J}}u),`$ (56)
$`x_1`$ is longitudinal velocity, $`x_2`$ is angular velocity, $`x_3=(x_1rx_2)/x_1`$<sup>15</sup><sup>15</sup>15Given this functional relation, variable $`x_3`$ in (6) can be viewed as an output of the reduced system with state $`(x_1,x_2)`$. is wheel slip, $`𝐱=(x_1,x_2,x_3)^T`$, $`m`$ is mass of the wheel, $`J`$ is moment of inertia, $`r`$ is radius of the wheel, $`u`$ is control input (brake torque), $`F_s(F_n,𝐱,\theta )`$ is a function specifying the tyre-road friction force depending on the surface-dependent parameter $`\theta `$ and the load force $`F_n`$. This function, for example, can be derived from steady-state behavior of the LuGre tyre-road friction model :
$$F_s(F_n,𝐱,\theta )=F_n\mathrm{sign}(x_2)\frac{\frac{\sigma _0}{L}g(x_2,x_3,\theta )\frac{x_3}{1x_3}}{\frac{\sigma _0}{L}\frac{x_3}{1x_3}+g(x_2,x_3,\theta )},$$
(57)
$$g(x_2,x_3,\theta )=\theta (\mu _C+(\mu _S\mu _C)e^{\frac{|rx_2x_3|}{|1x_3|v_s}}),$$
(58)
where $`\mu _C`$, $`\mu _S`$ are Coulomb and static friction coefficients, $`v_s`$ is the Stribeck velocity, $`\sigma _0`$ is the normalized rubber longitudinal stiffness, $`L`$ is the length of the road contact patch. In order to avoid singularities we assume, as suggested in , that the system is turned off when velocity $`x_1`$ reaches a small neighborhood of zero (in our example we stopped simulations as soon as $`x_1`$ becomes less than $`5`$ m/sec). Moreover, given that functions (57), (58) are bounded for the relevant set of the system parameters, it is always possible to design control function $`u(𝐱,t)`$ in (6) such that
$$x_3(t)[\delta ,1\delta ],\delta ,\delta >0$$
(59)
for all $`t:x_1(t)\delta _1`$, $`\delta _1=5`$ m/sec.
While the majority of the model parameters can be estimated a-priori, the tyre-road parameter $`\theta `$ is dependent on the properties of the road surface. Therefore, on-line identification of the parameter $`\theta `$ is desirable in order to compute the optimal slip value
$`x_3^{}=\mathrm{arg}\underset{x_3}{\mathrm{max}}F_s(F_n,𝐱,\theta )`$ (60)
which ensures the maximum deceleration force and therefore results in the shortest braking distance.
The main loop controller is derived in accordance with the standard certainty-equivalence principle and can be written as follows:
$$u(𝐱,\widehat{\theta },x_3^{})=\frac{J}{r}((\frac{1}{m}(1x_3)+\frac{r^2}{J})F_s(F_n,𝐱,\widehat{\theta })K_sx_1(x_3x_3^{})),K_s>0$$
In order to estimate parameter $`\theta `$ by measuring the values of variables $`x_1,x_2`$ and $`x_3`$, we construct the following subsystem:
$$\dot{\widehat{x}}_3=\frac{1}{x_1}((\frac{1}{m}(1x_3)+\frac{r^2}{J})F_s(F_n,𝐱,\widehat{\theta })\frac{r}{J}u)+(x_3\widehat{x}_3)$$
and consider the dynamics of error function $`\psi (𝐱,t)=\psi (x_3,\widehat{x}_3)=x_3\widehat{x}_3`$:
$`\dot{\psi }=\psi +{\displaystyle \frac{1}{x_1}}({\displaystyle \frac{1}{m}}(1x_3)+{\displaystyle \frac{r^2}{J}})(F_s(F_n,𝐱,\theta )F_s(F_n,𝐱,\widehat{\theta }))`$ (61)
The desired dynamics of system (61) is
$$\dot{\psi }=\psi +\xi (t)$$
(62)
where $`\xi (t)`$ is to be from $`L_2^1[t_0,\mathrm{}]`$. Let us check Assumptions 1, 2 for the function $`\psi (𝐱,t)=x_3\widehat{x}_3(t)`$ and system (62). Notice first that state $`𝐱`$ of system (6) is bounded according to the physical laws governing the dynamics of (6). In addition, boundedness of $`\psi (𝐱,t)`$ implies that $`\widehat{x}_3(t)`$ is bounded. Hence Assumption 1 holds. System (62), obviously, has $`L_2^1[t_0,\mathrm{}]L_{\mathrm{}}[t_0,\mathrm{}]`$ gain. We can conclude that Assumption 2 also holds. Let us check Assumptions 3, 4.
Taking into account (59) we can conclude that function $`\frac{1}{x_1}(\frac{1}{m}(1x_3)+\frac{r^2}{J})`$ in (61) is positive and, furthermore, is separated from zero for all $`x_1>\delta _1`$. Therefore, taking this into account equations (57), (58) we can conclude that function $`\frac{1}{x_1}(\frac{1}{m}(1x_3)+\frac{r^2}{J})F_s(F_n,𝐱,\theta )`$ in (61) satisfies Assumptions 3, 4 with
$$\alpha (𝐱,t)=\mathrm{const}=1,x_1:\delta _1<x_1<x_1(t_0)$$
Therefore, in order to design an estimation scheme satisfying assumptions of Theorem 2 we shall find functions $`\mathrm{\Psi }(𝐱,t)`$, $`(𝐱,t)`$ such that Assumption 5 holds. It is easy to see that this assumption is satisfied with $`\mathrm{\Psi }(𝐱,t)=\mathrm{const}`$, and $`(𝐱,t)`$=0. Let us choose, therefore, $`\mathrm{\Psi }(𝐱,t)=0`$. Then according to (23) and (61), a parameter adjustment algorithm will be given by the following system:
$`\widehat{\theta }=\gamma ((x_3\widehat{x}_3)+\widehat{\theta }_I),\dot{\widehat{\theta }}_I=(x_3\widehat{x}_3),\gamma =100`$ (63)
An important fact about algorithm (63) is that it is a parametric linear proportional-integral scheme. According to Theorem 2 the estimates (63) converge to $`\theta `$ exponentially fast in the domain specified by equation (59), and inequality $`x_1(t_0)x_1(t)>\delta _1`$. The last inequality is satisfied as, according to (6), time-derivative of the variable $`x_1(t)`$ is non-positive and the system is turned “off“ when $`x_1(t)\delta _1`$.
We simulated system (6) – (63) with the following setup of parameters and initial conditions: $`\sigma _0=200`$, $`L=0.25`$, $`\mu _C=0.5`$, $`\mu _S=0.9`$, $`v_s=12.5`$, $`r=0.3`$, $`m=200`$, $`J=0.23`$, $`F_n=3000`$, $`K_s=30`$. The effectiveness of estimation algorithm (63) could be illustrated with Figure 4. Estimates $`\widehat{\theta }`$ approach the actual values of parameter $`\theta `$ sufficiently fast for the controller to calculate the optimal slip value $`x_3^{}`$ and steer the system toward this point in real braking time. Effectiveness of the proposed identification-based control can be confirmed by comparing the braking distance in the system with on-line estimation of $`x_3^{}`$ according to formula (60) with the one, in which the values of $`x_3^{}`$ were kept constant (in the interval $`[0.1,0.2]`$). For model parameters as presently given and road condition given by the piece-wise constant function
$$\theta (s)=\{\begin{array}{cc}0.3,\hfill & s[0,8]\hfill \\ 1.3,\hfill & s(8,16]\hfill \\ 0.7,\hfill & s(16,24]\hfill \\ 0.4,\hfill & s(24,32]\hfill \\ 1.5,\hfill & s(32,40]\hfill \\ 0.6,\hfill & s(40,\mathrm{}]\hfill \end{array},s=_0^tx_1(\tau )𝑑\tau $$
the simulated braking distance obtained with our on-line estimation procedure of $`x_3^{}`$ is $`54.95`$ meters. This result compares favorably with the values obtained for preset values of $`x_3^{}`$, which range between $`57.52`$ and $`55.32`$ (for $`x_3^{}=0.1`$ and $`x_3^{}=0.2`$ respectively).
## 7 Conclusion
In the present article we provided new tool for the design and analysis of adaptive/parameter estimation schemes for dynamic systems with possible Lyapunov-unstable desired dynamics and nonlinear parameterization. In our method we consider adaptation as a process of asymptotic compensation of the uncertainty, or as control in functional spaces, rather than as simply reaching of a control goal. In particular, we wished to achieve that mismatches between the modeled uncertainty and compensator vanish asymptotically with time or belong to specific functional spaces. This understanding of adaptation naturally leads to the possibility to describe the desired dynamics of adapting systems in terms of an operator, which maps these mismatches into error functions, as functions of time from a functional space. Continuity of this target operator is not required. Hence stability of the desired dynamics, as a substitute of continuity, is not necessary for our approach. The adaptation mechanism itself could be viewed as control in functional spaces. In the other words, the aim of adaptation consists in ensuring that the uncertainty-induced errors of the compensator belong to a specific functional space. This idea leads to classes of adaptive systems, where applications require gentle, non-dominating control and where the desired dynamical state can be unstable.
When the desired motions in the system are known to be Lyapunov stable, our approach allows to design adaptation procedures without knowledge of the particular Lyapunov function. As mentioned in , this was one of the open theoretical challenges in the theory of adaptive cotnrol.
Another contribution of our present study is that we proposed a new class of parameterizations for nonlinearly parameterized models. Instead of aiming at a general solution for the problem of nonlinearity in the parameters, parametrization was restricted to a set of smooth functions, which are monotonic with respect to a linear functional in the parameters. For this new class, adaptation/estimation algorithms were introduced and analyzed. It was shown that standard linear persistent excitation conditions suffice to ensure exponentially fast convergence of the estimates to the actual values of unknown parameters. If, however, the monotonicity assumption holds only locally in the system state space, excitation with sufficiently high-frequency of oscillations still is able to ensure cpnvergence. In addition to the analysis of the effects of conventional persistent excitation on convergence, we also formulated a much weaker property - nonlinear persistent excitation condition. With this new property we established conditions for asymptotic convergence of the estimates. It is also desirable to notice that in case of linear parametrization the proposed parameter estimation schemes allow to estimate the unknowns in a dynamical system without asking for the usual filtered transformations, thus reducing the number of integrators in the estimator.
An application of our results, which is relevant to the parameter estimation problems for systems with nonlinear parameterization, was provided as an example. In this example we did not cover all solutions to every theoretical problem we were targeting in this article. In particular, it covers only the problem of nonlinear parameterization. The main rationale, however, was to illustrate all steps of our method. Last but not the least, the application presents a practically relevant solution to an important engineering problem. The effectiveness of the solution to this problem leads us to expect that our newly proposed method can successfully be implemented in a variety of other applications.
## 8 Appendix 1. Proofs of the theorems and auxiliary results
Proof of Theorem 1. Let us first show that property P1) holds. Consider solutions of system (3), (11), (23), (25) passing through the point $`𝐱(t_0)`$, $`\widehat{𝜽}_I(t_0)`$ for $`t[t_0,T^{}]`$<sup>16</sup><sup>16</sup>16According to the theorem formulation, interval $`[t_0,T^{}]`$ is the interval of existence of the solutions. Let us calculate formally the time-derivative of function $`\widehat{𝜽}(𝐱,t)`$: $`\dot{\widehat{𝜽}}(𝐱,t)=\mathrm{\Gamma }(\dot{\widehat{𝜽}}_P+\dot{\widehat{𝜽}}_I)=\mathrm{\Gamma }(\dot{\psi }𝜶(𝐱,t)+\psi \dot{𝜶}(𝐱,t)\dot{\mathrm{\Psi }}(𝐱,t)+\dot{\widehat{𝜽}}_I)`$. Notice that
$`\psi \dot{𝜶}(𝐱,t)\dot{\mathrm{\Psi }}(𝐱,t)+\dot{\widehat{𝜽}}_I=\psi (𝐱,t){\displaystyle \frac{𝜶(𝐱,t)}{𝐱_1}}\dot{𝐱}_1+\psi (𝐱,t){\displaystyle \frac{𝜶(𝐱)}{𝐱_2}}\dot{𝐱}_2+\psi (𝐱,t){\displaystyle \frac{𝜶(𝐱,t)}{t}}`$
$`{\displaystyle \frac{\mathrm{\Psi }(𝐱,t)}{𝐱_1}}\dot{𝐱}_1{\displaystyle \frac{\mathrm{\Psi }(𝐱,t)}{𝐱_2}}\dot{𝐱}_2{\displaystyle \frac{\mathrm{\Psi }(𝐱,t)}{t}}+\dot{\widehat{𝜽}}_I`$ (64)
According to Assumption 5, $`\frac{\mathrm{\Psi }(𝐱,t)}{𝐱_2}=\psi (𝐱,t)\frac{𝜶(𝐱,t)}{𝐱_2}+(𝐱,t)`$. Then taking into account (8), we can obtain
$`\psi \dot{𝜶}(𝐱,t)\dot{\mathrm{\Psi }}(𝐱,t)+\dot{\widehat{𝜽}}_I=\left(\psi (𝐱,t){\displaystyle \frac{𝜶(𝐱,t)}{𝐱_1}}{\displaystyle \frac{\mathrm{\Psi }}{𝐱_1}}\right)\dot{𝐱}_1+\psi (𝐱,t){\displaystyle \frac{𝜶(𝐱,t)}{t}}{\displaystyle \frac{\mathrm{\Psi }(𝐱,t)}{t}}`$
$`(𝐱,t)(𝐟_2(𝐱,𝜽)+𝐠_2(𝐱)u)+\dot{\widehat{𝜽}}_I`$ (65)
Notice that according to the proposed notation we can rewrite the term $`\left(\psi (𝐱,t)\frac{𝜶(𝐱,t)}{𝐱_1}\frac{\mathrm{\Psi }}{𝐱_1}\right)\dot{𝐱}_1`$ in the following form: $`\left(\psi (𝐱,t)L_{𝐟_1}𝜶(𝐱,t)L_{𝐟_1}\mathrm{\Psi }(𝐱,t)\right)+\left(\psi (𝐱,t)L_{𝐠_1}𝜶(𝐱,t)L_{𝐠_1}\mathrm{\Psi }(𝐱,t)\right)u(𝐱,\widehat{𝜽},t)`$. Hence it follows from (23) and (8) that $`\psi \dot{𝜶}(𝐱,t)\dot{\mathrm{\Psi }}(𝐱,t)+\dot{\widehat{𝜽}}_I=\phi (\psi )𝜶(𝐱,t)(𝐱,t)(𝐟_2(𝐱,𝜽)𝐟_2(𝐱,\widehat{𝜽}))`$. Therefore derivative $`\dot{\widehat{𝜽}}(𝐱,t)`$ can be written in the following way:
$`\dot{\widehat{𝜽}}=\mathrm{\Gamma }((\dot{\psi }+\phi (\psi ))𝜶(𝐱,t)(𝐱,t)(𝐟_2(𝐱,𝜽)𝐟_2(𝐱,\widehat{𝜽})))`$ (66)
Consider the following positive-definite function:
$$V_{\widehat{𝜽}}(\widehat{𝜽},𝜽,t)=\frac{1}{2}\widehat{𝜽}𝜽_{\mathrm{\Gamma }^1}^2+\frac{D}{4D_1^2}_t^{\mathrm{}}\epsilon ^2(\tau )𝑑\tau $$
(67)
Its time-derivative according to equations (66) can be obtained as follows:
$`\dot{V}_{\widehat{𝜽}}(\widehat{𝜽},𝜽,t)=(\phi (\psi )+\dot{\psi })(\widehat{𝜽}𝜽)^T𝜶(𝐱,t)(\widehat{𝜽}𝜽)^T(𝐱,t)(𝐟_2(𝐱,𝜽)𝐟_2(𝐱,\widehat{𝜽})){\displaystyle \frac{D}{4D_1^2}}\epsilon ^2(t)`$ (68)
Let $`(𝐱,t)0`$, then consider the following difference $`𝐟_2(𝐱,𝜽)𝐟_2(𝐱,\widehat{𝜽})`$. Applying Hadamard’s lemma we represent this difference in the following way:
$$𝐟_2(𝐱,𝜽)𝐟_2(𝐱,\widehat{𝜽})=_0^1\frac{𝐟_2(𝐱,𝐬(\lambda ))}{𝐬}𝑑\lambda (𝜽\widehat{𝜽}),𝐬(\lambda )=𝜽\lambda +\widehat{𝜽}(1\lambda )$$
Therefore, according to Assumption 5 function $`(\widehat{𝜽}𝜽)^T(𝐱,t)(𝐟_2(𝐱,𝜽)𝐟_2(𝐱,\widehat{𝜽}))`$ is positive semi-definite, hence using Assumptions 3, 4 and equality (11) we can estimate derivative $`\dot{V}_{\widehat{𝜽}}`$ as follows:
$`\dot{V}_{\widehat{𝜽}}(\widehat{𝜽},𝜽,t)(f(𝐱,\widehat{𝜽},t)f(𝐱,𝜽,t)+\epsilon (t))(\widehat{𝜽}𝜽)^T𝜶(𝐱,t){\displaystyle \frac{D}{4D_1^2}}\epsilon ^2(t)`$
$`{\displaystyle \frac{1}{D}}(f(𝐱,\widehat{𝜽},t)f(𝐱,𝜽,t))^2+{\displaystyle \frac{1}{D_1}}|\epsilon (t)||f(𝐱,\widehat{𝜽},t)f(𝐱,𝜽,t)|{\displaystyle \frac{D}{4D_1^2}}\epsilon ^2(t)`$ (69)
$`{\displaystyle \frac{1}{D}}\left(|f(𝐱,\widehat{𝜽},t)f(𝐱,𝜽,t)|{\displaystyle \frac{D}{2D_1}}\epsilon (t)\right)^20`$
It follows immediately from (8), (67) that
$$\widehat{𝜽}(t)𝜽_{\mathrm{\Gamma }^1}^2\widehat{𝜽}(t_0)𝜽_{\mathrm{\Gamma }^1}^2+\frac{D}{2D_1^2}\epsilon (t)_{2,[t_0,\mathrm{}]}^2$$
(70)
In particular, for $`t[t_0,T^{}]`$ we can derive from (67) that
$$\widehat{𝜽}(t)𝜽_{\mathrm{\Gamma }^1}^2\widehat{𝜽}(t_0)𝜽_{\mathrm{\Gamma }^1}^2+\frac{D}{2D_1^2}\epsilon (t)_{2,[t_0,T^{}]}^2$$
Therefore $`\widehat{𝜽}(t)L_{\mathrm{}}^2[t_0,T^{}]`$. Furthermore $`|f(𝐱(t),\widehat{𝜽}(t),t)f(𝐱(t),𝜽,t)|\frac{D}{2D_1}\epsilon (t)L_2^1[t_0,T^{}]`$. In particular
$$\begin{array}{c}\hfill |f(𝐱(t),\widehat{𝜽}(t),t)f(𝐱(t),𝜽,t)|\frac{D}{2D_1}\epsilon (t)_{2,[t_0,T^{}]}^2\frac{D}{2}𝜽\widehat{𝜽}(t_0)_{\mathrm{\Gamma }^1}^2+\frac{D^2}{4D_1^2}\epsilon (t)_{2,[t_0,T^{}]}^2\end{array}$$
(71)
Hence $`f(𝐱(t),\widehat{𝜽}(t),t)f(𝐱(t),𝜽,t)L_2^1[t_0,T^{}]`$ as a sum of two functions from $`L_2^1[t_0,T^{}]`$. In order to estimate the upper bound of norm $`f(𝐱(t),\widehat{𝜽}(t),t)f(𝐱(t),𝜽,t)_{2,[t_0,T^{}]}`$ from (71) we use the Minkowski inequality:
$$f(𝐱(t),\widehat{𝜽}(t),t)f(𝐱(t),𝜽,t)|\frac{D}{2D_1}\epsilon (t)_{2,[t_0,T^{}]}\left(\frac{D}{2}𝜽\widehat{𝜽}(t_0)_{\mathrm{\Gamma }^1}^2\right)^{0.5}+\frac{D}{2D_1}\epsilon (t)_{2,[t_0,T^{}]}$$
and then apply the triangle inequality to the functions from $`L_2^1[t_0,T^{}]`$:
$$\begin{array}{cc}& f(𝐱(t),\widehat{𝜽}(t),t)f(𝐱(t),𝜽,t)_{2,[t_0,T^{}]}f(𝐱(t),\widehat{𝜽}(t),t)f(𝐱(t),𝜽,t)\frac{D}{2D_1}\epsilon (t)_{2,[t_0,T^{}]}+\hfill \\ & \frac{D}{2D_1}\epsilon (t)_{2,[t_0,T^{}]}\left(\frac{D}{2}𝜽\widehat{𝜽}(t_0)_{\mathrm{\Gamma }^1}^2\right)^{0.5}+\frac{D}{D_1}\epsilon (t)_{2,[t_0,T^{}]}\hfill \end{array}$$
(72)
Therefore, property P1) is proven.
Let us prove property P2). In order to do this we have to check first if the solutions of the closed loop system are defined for all $`t_+`$, i.e. they do not reach infinity in finite time. We prove this by a contradiction argument. Indeed, let there exists time instant $`t_s`$ such that $`𝐱(t_s)=\mathrm{}`$. It follows from P1), however, that $`f(𝐱(t),\widehat{𝜽}(t),t)f(𝐱(t),𝜽,t)L_2^1[t_0,t_s]`$. Furthermore, according to (72) the norm $`f(𝐱(t),\widehat{𝜽}(t),t)f(𝐱(t),𝜽,t)_{2,[t_0,t_s]}`$ can be bounded from above by a continuous function of $`𝜽,\widehat{𝜽}(t_0)`$, $`\mathrm{\Gamma }`$, and $`\epsilon (t)_{2,[t_0,\mathrm{}]}`$. Let us denote this bound by symbol $`D_f`$. Notice that $`D_f`$ does not depend on $`t_s`$. Consider system (11) for $`t[t_0,t_s]`$:
$$\dot{\psi }=f(𝐱,𝜽,t)f(𝐱,\widehat{𝜽},t)\phi (\psi ,𝝎,t)+\epsilon (t)$$
Given that both $`f(𝐱(t),𝜽,t)f(𝐱(t),\widehat{𝜽}(t),t),\epsilon (t)L_2^1[t_0,t_s]`$ and taking into account Assumption 2, we automatically obtain that $`\psi (𝐱(t),t)L_{\mathrm{}}^1[t_0,t_s]`$. In particular, using the triangle inequality and the fact that function $`\gamma _{\mathrm{},2}(\psi (𝐱_0,t_0),𝝎,M)`$ in Assumption 2 is non-decreasing in $`M`$, we can estimate the norm $`\psi (𝐱(t),t)_{\mathrm{},[t_0,t_s]}`$ as follows:
$$\psi (𝐱(t),t)_{\mathrm{},[t_0,t_s]}\gamma _{\mathrm{},2}(\psi (𝐱_0,t_0),𝝎,D_f+\epsilon (t)_{2,[t_0,\mathrm{}]}^2)$$
(73)
According to Assumption 1 the following inequality holds:
$$𝐱(t)_{\mathrm{},[t_0,t_s]}\stackrel{~}{\gamma }(𝐱_0,𝜽,\gamma _{\mathrm{},2}(\psi (𝐱_0,t_0),𝝎,D_f+\epsilon (t)_{2,[t_0,\mathrm{}]}^2))$$
(74)
Given that a superposition of locally bounded functions is locally bounded, we can conclude that $`𝐱(t)_{\mathrm{}[t_0,t_s]}`$ is bounded. This, however, contradicts to the previous claim that $`𝐱(t_s)=\mathrm{}`$. Taking into account inequality (70) we can derive that both $`\widehat{𝜽}(𝐱(t),t)`$ and $`\widehat{𝜽}_I(t)`$ are bounded for every $`t_+`$. Moreover, according to (73), (74), (70) these bounds are (locally bounded) functions of initial conditions and parameters. Therefore, $`𝐱(t)L_{\mathrm{}}^n[t_0,\mathrm{}]`$, $`\widehat{𝜽}(𝐱(t),t)L_{\mathrm{}}^d[t_0,\mathrm{}]`$. Inequality (27) follows immediately from (72), (13), and the triangle inequality. Property P2) is proven.
Let us show that P3) holds. It is assumed that system (12) has $`L_2^1[t_0,\mathrm{}]L_p^1[t_0,\mathrm{}]`$, $`p>1`$ gain. In addition, we have just shown that $`f(𝐱(t),𝜽,t)f(𝐱(t),\widehat{𝜽}(t),t),\epsilon (t)L_2[t_0,\mathrm{}]`$. Hence, taking into account equation (11) we conclude that $`\psi (𝐱(t),t)L_p^1[t_0,\mathrm{}]`$, $`p>1`$. On the other hand, given that $`f(𝐱,\widehat{𝜽},t)`$, $`\phi (\psi ,𝝎,t)`$ are locally bounded with respect to their first two arguments uniformly in $`t`$, and that $`𝐱(t)L_{\mathrm{}}^n[t_0,\mathrm{}]`$,$`\psi (𝐱(t),t)L_{\mathrm{}}^1[t_0,\mathrm{}]`$, $`\widehat{𝜽}(t)L_{\mathrm{}}^d[t_0,\mathrm{}]`$, $`𝜽\mathrm{\Omega }_\theta `$, signal $`\phi (\psi (𝐱(t),t),𝝎,t)+f(𝐱(t),𝜽,t)f(𝐱(t),\widehat{𝜽}(t),t)`$ is bounded. Then $`\epsilon (t)L_{\mathrm{}}^1[t_0,\mathrm{}]`$ implies that $`\dot{\psi }`$ is bounded, and P3) is guaranteed by Barbalat’s lemma.
To complete the proof of the theorem (property P4) consider the time-derivative of function $`f(𝐱,\widehat{𝜽},t)`$:
$$\frac{d}{dt}f(𝐱,\widehat{𝜽},t)=L_{𝐟(𝐱,𝜽)+𝐠(𝐱)u(𝐱,\widehat{𝜽},t)}f(𝐱,\widehat{𝜽},t)+\frac{f(𝐱,\widehat{𝜽},t)}{\widehat{𝜽}}\mathrm{\Gamma }(\phi (\psi ,𝝎,t)+\dot{\psi })𝜶(𝐱,t)+\frac{f(𝐱,\widehat{𝜽},t)}{t}$$
Taking into account that $`𝐟(𝐱,𝜽)`$, $`𝐠(𝐱)`$, function $`f(𝐱,𝜽,t)`$ is continuously differentiable in $`𝐱`$, $`𝜽`$; derivative $`f(𝐱,𝜽,t)/t`$ is locally bounded with respect to $`𝐱`$, $`𝜽`$ uniformly in $`t`$; functions $`𝜶(𝐱,t)`$, $`\psi (𝐱,t)/t`$ are locally bounded with respect to $`𝐱`$ uniformly in $`t`$, then $`d/dt(f(𝐱,𝜽,t)f(𝐱,\widehat{𝜽},t))`$ is bounded. Then given that $`f(𝐱(t),𝜽,t)f(𝐱(t),\widehat{𝜽}(t),t)L_2^1[t_0,\mathrm{}]`$ by applying Barbalat’s lemma we conclude that $`f(𝐱,𝜽,\tau )f(𝐱,\widehat{𝜽},\tau )0`$ as $`t\mathrm{}`$. The theorem is proven.
Proof of Corollary 1. Let $`\epsilon (t)=0`$. Then choosing the function $`V_{\widehat{𝜽}(\widehat{𝜽,𝜽,t})}`$ as in (67), using (68), and invoking Assumption 3, we obtain that
$$\dot{V}_{\widehat{𝜽}(\widehat{𝜽},𝜽,t)}(f(𝐱,𝜽,t)f(𝐱,\widehat{𝜽},t))𝜶(𝐱,t)^T(𝜽\widehat{𝜽})\frac{1}{D}(f(𝐱,𝜽,t)f(𝐱,\widehat{𝜽},t))^2$$
(75)
Equality (75) and the fact that $`\epsilon (t)=0`$ in (67) imply that the norm $`\widehat{𝜽}𝜽_{\mathrm{\Gamma }^1}^2`$ is non-increasing. Furthermore, (75) implies that
$$f(𝐱(t),𝜽,t)f(𝐱(t),\widehat{𝜽}(t),t)_{2,[t_0,T^{}]}\left(\frac{D}{2}\widehat{𝜽}(t_0)𝜽_{\mathrm{\Gamma }^1}^2\right)^{0.5}$$
(76)
This proves property P1). Taking into account (76) and given that Assumptions 1, 2 are satisfied we can conclude that $`𝐱(t)L_{\mathrm{}}^n[t_0,\mathrm{}]`$, $`\psi (𝐱(t),t)L_{\mathrm{}}^1[t_0,\mathrm{}]`$, and the following estimate holds:
$$\psi (𝐱(t),t)_{\mathrm{},[t_0,\mathrm{}]}\gamma _{\mathrm{},2}(\psi (𝐱_0,t_0),𝝎,\left(\frac{D}{2}\widehat{𝜽}(t_0)𝜽_{\mathrm{\Gamma }^1}^2\right)^{0.5})$$
(77)
Hence P2) is also proven. Properties P3),P4) follow by the same arguments as in the proof of Theorem 1. Therefore, P5) is proven. The corollary is proven.
Proof of Corollary 2. Let us show that P6) holds. Without the loss of generality assume that solutions of the system exist over the following time interval $`[t_0,T^{}]`$. According to Theorem 1, property P1), the norm $`𝜽\widehat{𝜽}(t)`$ is bounded from above by a function of initial conditions $`\widehat{𝜽}(t_0)`$, parameters $`\mathrm{\Gamma }`$, $`D`$, $`D_1`$, and $`\epsilon (t)_{2,[t_0,\mathrm{}]}`$. Let us denote this bound by symbol $`B_\theta `$. Notice that $`B_\theta `$ does not depend on $`T^{}`$. On the other hand, according to Hypothesis H3 the following estimate holds:
$$D_\theta >0:|f(𝐱,𝜽,t)f(𝐱,\widehat{𝜽},t)|D_\theta 𝜽\widehat{𝜽}$$
Hence $`|f(𝐱(t),𝜽,t)f(𝐱(t),\widehat{𝜽}(t),t)|L_{\mathrm{}}^1[t_0,T^{}]`$ and moreover
$$f(𝐱(t),𝜽,t)f(𝐱,\widehat{𝜽}(t),t)_{\mathrm{},[t_0,T^{}]}D_\theta B_\theta $$
(78)
Consider the following signal $`\mu (t)=f(𝐱(t),𝜽,t)f(𝐱(t),\widehat{𝜽}(t),t)+\epsilon (t)`$. Signal $`\mu (t)L_2^1[t_0,T^{}]L_{\mathrm{}}^1[t_0,T^{}]`$, and let $`M_{\mathrm{}}`$, $`M_2_+`$ be the bounds for its $`L_{\mathrm{}}^1[t_0,T^{}]`$, $`L_2^1[t_0,T^{}]`$ norms respectively. According to (72), (78), these bounds can be estimated as follows: $`M_{\mathrm{}}=D_\theta B_\theta +\epsilon (t)_{\mathrm{},[t_0,\mathrm{}]}`$, $`M_2=\left(\frac{D}{2}𝜽\widehat{𝜽}(t_0)_{\mathrm{\Gamma }^1}^2\right)^{0.5}+\left(\frac{D}{D_1}+1\right)\epsilon (t)_{2,[t_0,\mathrm{}]}`$. Therefore, given $`p2`$ one can derive that
$$_{t_0}^{\mathrm{}}\mu ^p(\tau )𝑑\tau =_{t_0}^{\mathrm{}}\mu ^{p2}(\tau )\mu ^2(\tau )𝑑\tau M_{\mathrm{}}^{p2}M_2^2$$
Hence, $`\mu (t)L_p^1[t_0,T^{}]`$ and its $`L_p[t_0,T^{}]`$-norm is bounded from above by $`M_{\mathrm{}}^{p2}M_2^2`$, where the bounds $`M_{\mathrm{}}`$, $`M_2`$ both do not depend on $`T^{}`$. According to the corollary formulation, system (12) has $`L_p^1[t_0,\mathrm{}]L_{\mathrm{}}^1[t_0,\mathrm{}]`$ gain and therefore $`\psi (𝐱(t),t)L_{\mathrm{}}[t_0,T^{}]`$. Then applying the same argument as in the proof of property P2) of Theorem 1 and using Assumption 1 we can immediately obtain that $`𝐱(t)L_{\mathrm{}}^n[t_0,\mathrm{}]`$, $`\psi (𝐱(t),t)L_{\mathrm{}}^1[t_0,\mathrm{}]`$, and $`\widehat{𝜽}(t)L_{\mathrm{}}^d[t_0,\mathrm{}]`$. Thus property P6) is proven. Property P7) can now be proven in the same way as property P3) in Theorem 1. The corollary is proven.
Proof of Theorem 2. According to the theorem formulation, Assumptions 1,2, 3, 5 hold. Hence, applying Corollary 1 we can conclude that $`\widehat{𝜽}(t)L_{\mathrm{}}^d[t_0,\mathrm{}]`$ and $`𝐱(t)L_{\mathrm{}}^n[t_0,\mathrm{}]`$.
Let us show that limiting relation (18) holds in case alternative 1) is satisfied. To this purpose consider derivative $`\dot{\widehat{𝜽}}`$:
$$\dot{\widehat{𝜽}}=\mathrm{\Gamma }(\dot{\psi }+\phi (\psi ))𝜶(𝐱,t)=\mathrm{\Gamma }(f(𝐱,𝜽,t)f(𝐱,\widehat{𝜽},t))𝜶(𝐱,t).$$
(79)
Given that $`\widehat{𝜽}(t)L_{\mathrm{}}^d[t_0,\mathrm{}]`$ and $`𝐱(t)L_{\mathrm{}}^n[t_0,\mathrm{}]`$, and that Hypothesis H4 holds, the function $`f(𝐱,𝜽,t)`$ satisfies the following inequality for some $`D`$, $`D_1,_+`$:
$`D_1|𝜶(𝐱,t)^T(\widehat{𝜽}𝜽)||f(𝐱,𝜽,t)f(𝐱,\widehat{𝜽},t))|D|𝜶(𝐱,t)^T(\widehat{𝜽}𝜽)|;`$
$`𝜶(𝐱,t)^T(\widehat{𝜽}𝜽)(f(𝐱,\widehat{𝜽},t)f(𝐱,𝜽,t))0`$
Therefore, there exists function $`\kappa :_+_+`$, $`D_1\kappa ^2(t)D`$ such that
$$\dot{\widehat{𝜽}}=\kappa ^2(t)\mathrm{\Gamma }𝜶(𝐱,t)^T(\widehat{𝜽}𝜽)𝜶(𝐱,t)=\kappa ^2(t)\mathrm{\Gamma }𝜶(𝐱,t)𝜶(𝐱,t)^T(\widehat{𝜽}𝜽)$$
(80)
Notice that matrix $`\mathrm{\Gamma }`$ is a positive definite and symmetric matrix. It, therefore, can be factorized as follows $`\mathrm{\Gamma }=\mathrm{\Gamma }_0\mathrm{\Gamma }_0^T`$: where $`\mathrm{\Gamma }_0`$ is nonsingular $`n\times n`$ real matrix. Let us define $`\stackrel{~}{𝜽}=\mathrm{\Gamma }_0^1(\widehat{𝜽}𝜽)`$. In these new coordinates, equation (80) will have the following form:
$$\dot{\stackrel{~}{𝜽}}=\kappa (t)^2\mathrm{\Gamma }_0^1\mathrm{\Gamma }_0\mathrm{\Gamma }_0^T𝜶(𝐱,t)𝜶(𝐱,t)^T(\widehat{𝜽}𝜽)=\kappa ^2(t)\mathrm{\Gamma }_0^T𝜶(𝐱,t)𝜶(𝐱,t)^T\mathrm{\Gamma }_0\stackrel{~}{𝜽}$$
(81)
Denoting $`\kappa (t)\mathrm{\Gamma }_0^T𝜶(𝐱,t)=\mathit{\varphi }(𝐱,t)`$ we can rewrite equation (81) as follows:
$$\dot{\stackrel{~}{𝜽}}=\mathit{\varphi }(𝐱,t)\mathit{\varphi }(𝐱,t)^T\stackrel{~}{𝜽}$$
(82)
where function $`\mathit{\varphi }(𝐱,t):^n\times _+^d`$ satisfies equality:
$`𝜼^T{\displaystyle _t^{t+L}}\mathit{\varphi }(𝐱(\tau ),\tau )\mathit{\varphi }(𝐱(\tau ),\tau )𝑑\tau 𝜼=𝜼^T\mathrm{\Gamma }_0^T\left({\displaystyle _t^{t+L}}\kappa ^2(\tau )𝜶(𝐱(\tau ),\tau )𝜶(𝐱(\tau ),\tau )^T𝑑\tau \right)\mathrm{\Gamma }^0𝜼`$ (83)
for all $`𝜼^d`$. Taking into account that function $`𝜶(𝐱(t),t)`$ is persistently exciting, $`\mathrm{\Gamma }=\mathrm{\Gamma }_0^T\mathrm{\Gamma }_0`$, and that $`\kappa ^2(t)D_1`$ we can obtain the following bound for quadratic form (83):
$`𝜼^T{\displaystyle _t^{t+L}}\mathit{\varphi }(𝐱(\tau ),\tau )\mathit{\varphi }(𝐱(\tau ),\tau )𝑑\tau 𝜼\delta D_1\mathrm{\Gamma }_0𝜼^2=\delta D_1𝜼^T\mathrm{\Gamma }𝜼\delta D_1\lambda _{\mathrm{min}}(\mathrm{\Gamma })𝜼^2=\delta _\varphi 𝜼^2`$ (84)
Hence, function $`\mathit{\varphi }(𝐱(t),t)`$ is also persistently exciting. Notice also that $`\varphi (𝐱,t)`$ is bounded from above:
$$\begin{array}{cc}\hfill \mathit{\varphi }(𝐱,t)& =\kappa (t)𝜶(𝐱,t)\mathrm{\Gamma }_0\lambda _{\mathrm{max}}(\mathrm{\Gamma }_0)\kappa (t)𝜶(𝐱,t)\lambda _{\mathrm{max}}(\mathrm{\Gamma }_0)D\alpha _{\mathrm{}}\hfill \\ \hfill \alpha _{\mathrm{}}& =\underset{𝐱𝐱(t)_{\mathrm{},[t_0,\mathrm{}]},tt_0}{sup}𝜶(𝐱,t)\hfill \end{array}$$
In order to show that $`lim_t\mathrm{}\stackrel{~}{𝜽}(t)=0`$ exponentially fast we invoke the following useful lemma from (Lemma 5, page 18):
###### Lemma 1
Let system (82) be given, condition (84) holds (uniformly), and $`\mathbf{\varphi }(𝐱,t)`$ in (82) be bounded $`(\mathbf{\varphi }(𝐱,t))\varphi _M`$. Then system (82) is (uniformly) exponentially stable and, furthermore:
$$\stackrel{~}{𝜽}(t)e^{Kt}\stackrel{~}{𝜽}(t_0),K=\frac{\delta _\varphi }{L(1+L\varphi _M^2)^2}$$
(85)
According to Lemma 1 solutions of system (82) converge to the origin exponentially fast with a rate of convergence defined by (85), where
$$\delta _\varphi =\lambda _{\mathrm{min}}(\mathrm{\Gamma })D_1\delta ,\varphi _M^2=\lambda _{\mathrm{max}}(\mathrm{\Gamma })D^2\alpha _{\mathrm{}}^2$$
(86)
Taking into account equation (85), (86) and observing that $`(1+L\varphi _M^2)^22(1+\varphi _M^4L^2)`$, $`\lambda _{\mathrm{max}}(\mathrm{\Gamma }_0)^2=\lambda _{\mathrm{max}}(\mathrm{\Gamma })`$ we can estimate $`\stackrel{~}{𝜽}(t)`$ as follows:
$$\stackrel{~}{𝜽}(t)e^{K_1t}\stackrel{~}{𝜽}(t_0),K_1=\frac{\delta D_1\lambda _{\mathrm{min}}(\mathrm{\Gamma })}{2L(1+\lambda _{\mathrm{max}}^2(\mathrm{\Gamma })L^2D^2\alpha _{\mathrm{}}^4)}$$
(87)
Given that $`\mathrm{\Gamma }_0\stackrel{~}{𝜽}(t)=(\widehat{𝜽}(t)𝜽)`$, and using (87) we derive the following bounds for $`(\widehat{𝜽}(t)𝜽)`$:
$$(\widehat{𝜽}(t)𝜽)\mathrm{\Gamma }_0\stackrel{~}{𝜽}(t)\lambda _{\mathrm{max}}(\mathrm{\Gamma }_0)e^{K_1t}\mathrm{\Gamma }_0^1(\widehat{𝜽}(t_0)𝜽)\left(\frac{\lambda _{\mathrm{max}}(\mathrm{\Gamma }_0)}{\lambda _{\mathrm{min}}(\mathrm{\Gamma }_0)}\right)e^{K_1t}\widehat{𝜽}(t_0)𝜽$$
This proves alternative 1) of the theorem.
Let us prove alternative 2). It follows immediately from Corollary 1 of Theorem 1 that
$$\underset{t\mathrm{}}{lim}f(𝐱(t),𝜽,t)f(𝐱(t),\widehat{𝜽}(t),t)=0$$
(88)
Furthermore, given that $`\dot{\widehat{𝜽}}=\mathrm{\Gamma }(f(𝐱(t),𝜽,t)f(𝐱(t),\widehat{𝜽}(t),t))𝜶(𝐱,t)`$, $`𝐱(t)L_{\mathrm{}}^n[t_0,\mathrm{}]`$ and $`𝜶(𝐱,t)`$ is locally bounded in $`𝐱`$ uniformly in $`t`$, we can conclude that $`\dot{\widehat{𝜽}}0`$ as $`t\mathrm{}`$. Let us divide the $`_+`$ into the following union of subintervals:
$$_+=\underset{i=1}{\overset{\mathrm{}}{}}\mathrm{\Delta }_i,\mathrm{\Delta }_i=[t_i,t_i+T],t_0=0,t_{i+1}=t_i+T,i$$
The fact that $`\dot{\widehat{𝜽}}0`$ as $`t\mathrm{}`$ ensures that
$$\underset{i\mathrm{}}{lim}\widehat{𝜽}(s_i)\widehat{𝜽}(\tau _i)=0,s_i,\tau _i\mathrm{\Delta }_i$$
(89)
In order to show this let us integrate equation (79)
$$\widehat{𝜽}(s_i)\widehat{𝜽}(\tau _i)=\mathrm{\Gamma }_{s_i}^{\tau _i}(f(𝐱(\tau ),𝜽,\tau )f(𝐱(\tau ),\widehat{𝜽}(\tau ),\tau ))\alpha (𝐱(\tau ),\tau )𝑑\tau $$
(90)
Applying the Cauchy-Schwartz inequality to (90) and subsequently using the mean value theorem we can obtain the following estimate:
$`\widehat{𝜽}(s_i)\widehat{𝜽}(\tau _i)`$ $``$ $`{\displaystyle _{t_i}^{t_i+T}}\mathrm{\Gamma }|f(𝐱(\tau ),𝜽,\tau )f(𝐱(\tau ),\widehat{𝜽}(\tau ),\tau )|\alpha (𝐱(\tau ),\tau )𝑑\tau `$ (91)
$`=`$ $`\mathrm{\Gamma }T|f(𝐱(\tau _i^{}),𝜽,\tau _i^{})f(𝐱(\tau _i^{}),\widehat{𝜽}(\tau _i^{}),\tau _i^{})|𝜶(𝐱(\tau _i^{}),\tau _i^{}),\tau _i^{}\mathrm{\Delta }_i`$
Given that limiting relation (88) holds, $`𝐱(t)L_{\mathrm{}}^n[t_0,\mathrm{}]`$, and $`𝜶(𝐱,t)`$ is locally bounded uniformly in $`t`$ we can conclude from (91) that limiting relation (89) holds.
Let us choose a sequence of points from $`_+`$: $`\{\tau _i\}_{i=1}^{\mathrm{}}`$ such that $`\tau _i\mathrm{\Delta }_i`$, $`i`$. As follows from the nonlinear persistent excitation condition (inequality (44)), for every $`\widehat{𝜽}(\tau _i)`$, $`\tau _i\mathrm{\Delta }_i`$ there exists a point $`t_i^{}\mathrm{\Delta }_i`$ such that the following inequality holds
$$f(𝐱(t_i^{}),𝜽,t_i^{})f(𝐱(t_i^{}),\widehat{𝜽}(\tau _i),t_i^{})\varrho (𝜽\widehat{𝜽}(\tau _i))0$$
(92)
Let us consider the following differences:
$$f(𝐱(t_i^{}),\widehat{𝜽}(\tau _i),t_i^{})f(𝐱(t_i^{}),\widehat{𝜽}(t_i^{}),t_i^{}),\tau _i,t_i^{}\mathrm{\Delta }_i$$
It follows immediately from H1, H2, and (89) that
$$\underset{i\mathrm{}}{lim}f(𝐱(t_i^{}),\widehat{𝜽}(\tau _i),t_i^{})f(𝐱(t_i^{}),\widehat{𝜽}(t_i^{}),t_i^{})=0,\tau _i,t_i^{}\mathrm{\Delta }_i$$
(93)
Taking into account (93) and (88) we can derive that
$`\underset{i\mathrm{}}{lim}f(𝐱(t_i^{}),𝜽,t_i^{})f(𝐱(t_i^{}),\widehat{𝜽}(\tau _i),t_i^{})=\underset{i\mathrm{}}{lim}(f(𝐱(t_i^{}),𝜽,t_i^{})f(𝐱(t_i^{}),\widehat{𝜽}(t_i^{}),t_i^{}))+`$
$`\underset{i\mathrm{}}{lim}f(𝐱(t_i^{}),\widehat{𝜽}(t_i^{}),t_i^{})f(𝐱(t_i^{}),\widehat{𝜽}(\tau _i),t_i^{})=0`$ (94)
According to (8) and (92), sequence $`\{\varrho (𝜽\widehat{𝜽}(\tau _i))\}_{i=1}^{\mathrm{}}`$ is bounded from above and below by two sequences converging to zero. Hence, $`lim_i\mathrm{}\varrho (𝜽\widehat{𝜽}(\tau _i))=0`$. Notice that $`\varrho ()𝒦C^0`$ which implies that
$$\underset{i\mathrm{}}{lim}𝜽\widehat{𝜽}(\tau _i)=0$$
(95)
In order to show that $`lim_t\mathrm{}(𝜽\widehat{𝜽}(t))=0`$ notice that
$$𝜽\widehat{𝜽}(t)𝜽\widehat{𝜽}(s_i),s_i=\mathrm{arg}\underset{s\mathrm{\Delta }_i}{\mathrm{max}}𝜽\widehat{𝜽}(s)t\mathrm{\Delta }_i$$
Hence, applying the triangle inequality $`𝜽\widehat{𝜽}(s_i)𝜽\widehat{𝜽}(\tau _i)+\widehat{𝜽}(\tau _i)\widehat{𝜽}(s_i)`$ and using equations (89), (95) we can conclude that $`𝜽\widehat{𝜽}(t)`$ is bounded from above and below by two functions converging to zero. Hence, $`𝜽\widehat{𝜽}(t)0`$ as $`t\mathrm{}`$ and limiting relation (18) holds. The theorem is proven.
|
warning/0506/astro-ph0506611.html
|
ar5iv
|
text
|
# Outflows in Infrared-Luminous Starbursts at 𝑧<0.5. II. Analysis and Discussion1footnote 11footnote 1Some of the observations reported here were obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among the California Institute of Technology, the University of California, and the National Aeronautics and Space Administration. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. 2footnote 22footnote 2Some of the observations reported here were obtained at the MMT Observatory, which is a joint facility of the Smithsonian Institution and the University of Arizona. 3footnote 33footnote 3Some of the observations reported here were obtained at the Kitt Peak National Observatory, National Optical Astronomy Observatory, which is operated by the Association of Universities for Research in Astronomy, Inc. (AURA) under cooperative agreement with the National Science Foundation.
## 1 INTRODUCTION
The number of theoretical papers discussing or incorporating galactic winds has grown sharply in recent years. Superwinds have been invoked to explain a myriad of issues relating to contemporary cosmology (Veilleux & Rupke, 2004; Veilleux et al., 2005). Numerical simulations and semi-analytic models typically incorporate simplified prescriptions for superwinds, rather than detailed microphysics. This is appropriate not only due to the current level of computing power, but also because hydrodynamic simulations of superwinds cannot yet accurately predict many large-scale properties of the outflows. Thus, theorists depend on observers to demonstrate the macroscopic behavior of real outflows over a range of galaxy types and environments.
Previous large surveys of local starburst galaxies have shown that superwinds are common. Lehnert & Heckman (1995, 1996) undertook an imaging and spectroscopic survey of $``$50 edge-on galaxies with infrared luminosities $`L_{\mathrm{IR}}<10^{12}`$ $`L_{\mathrm{}}`$ and redshifts $`cz<15000`$ km s<sup>-1</sup>. Evidence for superwinds in these galaxies includes: extended line emission, shock-like line ratios, broad emission lines, velocity shear, and blue emission line asymmetries. Several of these properties also show a positive correlation with infrared activity ($`=L_{\mathrm{IR}}/\mathrm{L}_{\mathrm{opt}}`$). Lehnert & Heckman (1996) find that the average (deprojected) outflow speed of the ionized gas in these galaxies is $`170_{80}^{+150}`$ km s<sup>-1</sup>.
Using the complementary technique of optical absorption-line spectroscopy, Heckman et al. (2000) studied a sample of 32 infrared-luminous galaxies with $`z0.1`$. Their survey includes galaxies whose luminosities are dominated by a starburst as well as some powered by an active galactic nucleus (AGN). They include galaxies with a range of luminosities, including five ultraluminous infrared galaxies (ULIRGs; $`L_{\mathrm{IR}}`$ $`10^{12}`$ $`L_{\mathrm{}}`$). Heckman et al. (2000) also find that the detection rate of winds is high. They show that the outflows are dusty, and measure outflow velocities up to $`400600`$ km s<sup>-1</sup>; however, they do not measure an individual mass outflow rate for each galaxy and fit only a single velocity component to each absorption feature.
Most recently, detailed absorption-line studies of $`1020`$ ULIRGs (Rupke et al., 2002; Martin, 2005) have found that the detection rate of winds in ULIRGs is higher than that in luminous infrared galaxies (LIRGs; $`L_{\mathrm{IR}}`$ $`10^{11}`$ $`L_{\mathrm{}}`$). These studies show that ULIRGs have high mass outflow rates, of tens to hundreds of M yr<sup>-1</sup>. Martin (2005) also argues that wind velocity scales with star formation rate and galactic mass.
At higher redshifts, directly studying winds is possible only using deep observations of bright or gravitationally-lensed galaxies. One of the best-studied high-redshift galaxy populations is the Lyman-break galaxies (LBGs). Extensive spectroscopic observations of LBGs have shown that most of them host high-velocity outflows (Pettini et al., 2001; Shapley et al., 2003; Adelberger et al., 2003). Lyman-$`\alpha `$ emission is redshifted (perhaps due to resonant scattering on the receding half of the outflow), while UV absorption lines are blueshifted due to interstellar absorption in the outflow. The average offset of the two is 614 km s<sup>-1</sup> (Adelberger et al., 2003), implying that an average (projected) outflow velocity is approximately 300 km s<sup>-1</sup>. Early evidence suggested that these outflows are quite large ($`r500`$ kpc) and metal-enriched (Adelberger et al., 2003), but more recent work has revised the probable sizes of these outflows to more modest radii ($`r100`$ kpc; Adelberger et al. 2005).
Though the observational data set on outflows is steadily increasing, both in quality and quantity, it is in many ways still quite limited. Over most of cosmic history, the frequency of occurrence and impact of superwinds remains unquantified. We also know very little about how the properties of superwinds depend on the properties of their host galaxies. Observations have typically focused on dwarf galaxies or edge-on disk galaxies with moderate star formation rates (as a well as a set of high-redshift galaxies with as yet uncertain properties). ULIRGs are of special interest, as they may host over 50% of global star formation at $`z2`$ (Pérez-González et al., 2005). By studying ULIRGs in the local universe, we can learn how winds are affecting the cosmos at high redshift, when many of the present-day stars were formed and much of the metal enrichment of the IGM may have occurred.
These observations are needed in part to inform theoretical prescriptions of winds in numerical simulations. Many theorists assume that the mass outflow rate in the wind and star formation rate in the corresponding host galaxy are comparable and linearly proportional to each other (e.g., Kauffmann & Charlot, 1998; Aguirre et al., 2001a, b; Silk, 2003). In other words, the ratio of the mass outflow rate to the star formation rate in a galaxy is generically unity. This is presumably based on the claim of some observers that this is so, for luminous infrared galaxies (Heckman et al., 2000) and dwarfs or spirals (Martin, 1999). However, preliminary data from our pilot study suggests that this ratio is an order of magnitude lower in ULIRGs (Rupke et al., 2002), at least for the neutral gas phase.
The role of this work is to fill in the gaps mentioned above and produce a systematic study of large-scale outflows. We have made the largest survey to date of massive outflows in infrared-luminous galaxies. Our survey includes over 100 galaxies and extends over a broad redshift range ($`z=0.00.5`$). We have included 78 starburst galaxies with a wide range of properties and, for the first time, searched for superwinds in a large number (43) of starburst-dominated ULIRGs. (In a forthcoming paper, we discuss observations of 26 ULIRGs which have Seyfert nuclei and compare them to the starburst ULIRGs; Rupke et al. 2005b.)
The existence of outflows can be inferred from the presence of absorption lines that are blueshifted with respect to the systemic velocity of the host galaxy. We apply this technique using moderately high resolution spectroscopy (FWHM $`=6585`$ km s<sup>-1</sup>) of the Na I D $`\lambda \lambda 5890,5896`$ doublet. Our observations were obtained with echelle and long-slit spectrographs on the Keck II, MMT, and KPNO 4m telescopes. We perform detailed profile fitting using multiple velocity components and assuming a Gaussian in optical depth; from these fits we measure velocity, velocity width, optical depth, covering fraction, and column density for each component. The details of the sample selection, observations, and data reduction are discussed in a companion paper (Rupke et al., 2005a, hereafter Paper I). In Paper I, we also present the absorption- and emission-line spectra and a discussion of the absorption-line fitting.
The size and makeup of our sample allows us to study the properties of these winds as a function of host galaxy properties over a large dynamic range. Wind properties that we can measure include: velocity; mass, momentum, energy, and their outflow rates; and mass entrainment efficiency, which is a measure of the relative efficiency with which these winds entrain interstellar gas clouds. The relevant galaxy properties are discussed in Paper I, and include star formation rate, optical and near-infrared luminosity, circular velocity, spectral type, and redshift.
In §§2.2$``$2.4, we describe in detail the distributions of velocity, optical depth, column density, and covering fraction in our sample. §§2.1, 2.6, and 2.7 cover our measurements of outflow detection rate; mass, momentum, energy, and their outflow rates; and mass entrainment efficiency. §2.5 discusses the spatial distribution of absorbing gas. In §3, we look at the dependence of outflow detection rate and properties on host galaxy properties. We also compare absorption and emission lines in §3.4. §4 covers, in this order: our discussion of alternatives to the superwind scenario, the global covering factor and escape fraction of outflows, comparison to theory, and discussion of redshift evolution. We summarize in §5.
For all calculations, we assume present-day values for the cosmological parameters of $`H_0=75`$ km s<sup>-1</sup> Mpc<sup>-1</sup> and the standard $`\mathrm{\Omega }_m=0.3`$, $`\mathrm{\Omega }_\mathrm{\Lambda }=0.7`$ cosmology. All wavelengths quoted are vacuum wavelengths (except those used as labels for spectral lines) and are generally taken from the NIST Atomic Spectra Database<sup>1</sup><sup>1</sup>1http://physics.nist.gov/cgi-bin/AtData/main\_asd. (The vacuum wavelengths of Na I D are 5891.58 and 5897.55 Å.)
## 2 OUTFLOW PROPERTIES
In this section, we discuss our measurements of outflow properties. Table 1 lists the average outflow properties for our three subsamples. Table 2 lists the outflow properties of each galaxy. As discussed in Paper I, the full sample of 78 galaxies is divided into three subsamples based on infrared luminosity and redshift: (a) the IRGs (InfraRed Galaxies), with $`L_{\mathrm{IR}}`$ $`<10^{12}`$ $`L_{\mathrm{}}`$; (b) the low-$`z`$ ULIRGs, with $`L_{\mathrm{IR}}`$ $`10^{12}`$ $`L_{\mathrm{}}`$ and $`z<0.25`$; and (c) the high-$`z`$ ULIRGs, with $`L_{\mathrm{IR}}`$ $`10^{12}`$ $`L_{\mathrm{}}`$ and $`0.25<z<0.50`$.
These outflow quantities are calculated from the fitted parameters of the Na I D feature in each galaxy, as listed in the tables of Paper I. Paper I describes the details of the Na I D fitting procedure. In short, we fit multiple velocity components to the Na I D feature in each object. The large velocity dispersions in the galaxies we observe cause the components of the Na I D doublet, as well as velocity components if there are more than one, to blend with one another. Our fitting procedure addresses this problem. We assume Gaussians in optical depth, which translates into observable non-Gaussian intensity profiles for optical depths greater than unity. We also fit a constant covering fraction for each velocity component. The measured parameters resulting from our fits are: outflow velocity, $`\mathrm{\Delta }v`$; Doppler width, $`b`$; central optical depth of the Na I D<sub>1</sub> line, $`\tau _{1,c}`$; and the covering fraction, $`C_f`$. The errors in these parameters are determined using Monte Carlo simulations.
In Paper I, we also dicuss the method of computation of the Na I and H column densities from these fitted parameters. The values of $`N`$(Na I) and $`N`$(H) for each galaxy are listed in Table 4 of Paper I.
### 2.1 Rate of Detection
The current survey contains over $`50\%`$ more starburst galaxies than previous superwind surveys of infrared sources (Lehnert & Heckman, 1995, 1996; Heckman et al., 2000; Rupke et al., 2002; Martin, 2005). It also has a much higher number of ULIRGs (43 vs. $`<`$20). We are thus in a better position to quantify the frequency of occurrence of winds in ULIRGs, especially as a function of galaxy properties.
We use a velocity cutoff of $`50`$ km s<sup>-1</sup> to delineate galaxies with superwinds, a slightly more liberal cutoff than in our pilot study (in which we used $`70`$ km s<sup>-1</sup>). Velocity components blueshifted by more than 50 km s<sup>-1</sup> are assumed to be outflowing. This cutoff is chosen to avoid contamination due to systematic errors and measurement errors in wavelength calibration ($``$10 km s<sup>-1</sup>), line fitting (10 km s<sup>-1</sup> on average), and redshift determination (15 km s<sup>-1</sup>). The vast majority of redshifted components have $`|\mathrm{\Delta }v|<50`$ km s<sup>-1</sup>, suggesting that this cutoff is a good one.
Table 1 lists the detection rate of massive outflows in each subsample. We observe that outflows are common in each subsample, are detected in a strong majority of ULIRGs, and are detected more frequently in ULIRGs than in less-luminous galaxies (see §§3.1 and 4.2). The high-$`z`$ ULIRGs also show a lower detection rate than the low-$`z`$ ULIRGs; we discuss this further in §4.6.
Applying this same cutoff to the galaxies in Heckman et al. (2000) that are optically classified as H II galaxies or LINERs and have $`L_{\mathrm{IR}}`$ $`<10^{12}`$ $`L_{\mathrm{}}`$, we compute a wind detection rate of ($`32\pm 12`$)% ($`7/22`$ galaxies, with a median log\[$`L_{\mathrm{IR}}`$/$`L_{\mathrm{}}`$\] $`=11.0`$). This is consistent with the measurement from our IRG subsample. Our measurement of ($`42\pm 8`$)% is slightly higher than that of Heckman et al. (2000), but so is the median infrared luminosity of our subsample. Our detection rate in ULIRGs, $`(70\pm 7)`$% overall, is consistent with those from our preliminary report, $`(75\pm 15)`$% (Rupke et al., 2002), and another recent survey, $`(85\pm 8)`$% (Martin, 2005). High-$`z`$ ($`z3`$) galaxies of comparable SFR to the IRGs, the LBGs (with SFR $``$ 10$``$100 M yr<sup>-1</sup>; Pettini et al. 1998), have a much higher detection rate than in our IRG subsample; almost 100% of LBGs have detected outflows (Adelberger et al., 2003).
In §4.2, we consider the detection rates in light of the global covering factor of the gas. In short, these detection rates are lower limits to the actual fraction of these galaxies that host outflows, which may be close to unity.
### 2.2 Velocity Distribution
Previous surveys of winds in starbursting galaxies that measured the distribution of outflow velocities in galaxies include Heckman et al. (2000) and Adelberger et al. (2003). The latter show that the distribution of the velocity offset between the Ly$`\alpha `$ line and UV absorption lines is approximately a Gaussian of mean 614 km s<sup>-1</sup> and width $`\sigma =316`$ km s<sup>-1</sup>, including a slight upturn at $`\mathrm{\Delta }v=0`$. This implies that every $`z3`$ Lyman-break galaxy has outflowing gas, with an average projected outflow velocity of $``$300 km s<sup>-1</sup>.
The distribution of gas velocities in local galaxies of comparable SFR is different from that of LBGs, however. The strong-stellar-contamination subsample of Heckman et al. (2000) shows a roughly symmetric Na I D velocity distribution around 0 km s<sup>-1</sup>. The distribution of the interstellar-dominated subsample is asymmetric; it lies completely in the blue for $`|\mathrm{\Delta }v|>100`$ km s<sup>-1</sup>.
In Figure 1$`a`$, we show the distributions of the central velocity of each component in our sample using 100 km s<sup>-1</sup> bins. The distribution of velocities for IRGs is fairly symmetric around $`\mathrm{\Delta }v=0`$ but has a clear tail toward blue velocities. The blue asymmetry in the distribution of velocities for ULIRGs above $`|\mathrm{\Delta }v|=100`$ km s<sup>-1</sup> is even more pronounced than in IRGs. Note also that there are several components with redshifted velocities greater than 100 km s<sup>-1</sup> in the ULIRG distributions (though not in the IRG distribution).
The distributions of the velocity of the highest column density outflowing gas in each galaxy, $`\mathrm{\Delta }v_{maxN}`$, are shown in Figure 1$`c`$. (Essentially, these distributions show the $`\mathrm{\Delta }v`$’s which are less than -50 km s<sup>-1</sup>, but we pick only one component from each galaxy.) The median value for each subsample is listed in Table 1. The median value for our full sample is -140 km s<sup>-1</sup>, with a dispersion of 120 km s<sup>-1</sup> and a median error of 7 km s<sup>-1</sup>. The range of $`|\mathrm{\Delta }v_{maxN}|`$ is $`50520`$ km s<sup>-1</sup>. Kolmogorov-Smirnov (K-S) and Kuiper tests show that the distributions of $`\mathrm{\Delta }v_{maxN}`$ for the IRGs and low-$`z`$ ULIRGs are different at $`90\%`$ confidence; ULIRGs have higher $`\mathrm{\Delta }v_{maxN}`$ on average.
The median $`\mathrm{\Delta }v_{maxN}`$ of the IRGs, 100 km s<sup>-1</sup>, is smaller than the deprojected velocities of ionized gas in the halos of edge-on galaxies of comparable SFR ($`170_{80}^{+150}`$ km s<sup>-1</sup>; Lehnert & Heckman 1996) and the mean projected velocity of ionized gas in high-$`z`$ LBGs (300 km s<sup>-1</sup>; Adelberger et al. 2003). We address the relative velocities of ionized and neutral gas in §3.4.
The distributions of ‘maximum’ velocity in each galaxy ($`\mathrm{\Delta }v_{max}\mathrm{\Delta }v\mathrm{FWHM}/2`$, computed for the most blueshifted component) are shown in Figure 1$`d`$. ($``$90% of the detectable gas in the wind has a velocity less than or equal to $`\mathrm{\Delta }v_{max}`$.) The median value for each subsample is listed in Table 1. The median value for our full sample is $`350`$ km s<sup>-1</sup>, with a dispersion of 170 km s<sup>-1</sup> and median error of 20 km s<sup>-1</sup>. The range for $`|\mathrm{\Delta }v_{max}|`$ is $`130600`$ km s<sup>-1</sup>, though there is a single galaxy (F10378$`+`$1108) which has $`\mathrm{\Delta }v_{max}=1100`$ km s<sup>-1</sup>. K-S and Kuiper tests show that the distributions of each subsample are consistent with being drawn from the same parent distribution.
The Doppler widths ($`b=\mathrm{FWHM}/[2\sqrt{\mathrm{ln}2}]`$) of each velocity component are shown as a histogram in Figure 1$`b`$. These values have been corrected downward (in quadrature) for the instrumental resolution. The range of values is 50 km s<sup>-1</sup> (the resolution limit) up to almost 700 km s<sup>-1</sup>, with a median error of 25 km s<sup>-1</sup>. The median value for the IRGs is 150 km s<sup>-1</sup>, while the average value in the low-$`z`$ ULIRGs is higher, at 200 km s<sup>-1</sup>. K-S and Kuiper tests show that the overall distributions of the different subsamples are not significantly different, however.
The large values that we measure for $`b`$ imply that the broadening is due to the sum of large-scale motions in the wind. These values are far higher than those expected from thermal broadening of warm Na gas ($`b_{th}=2.7`$ km s<sup>-1</sup> at $`T=10^4`$ K). However, our spectral resolution is not high enough to resolve clouds with widths $`b50`$ km s<sup>-1</sup>; the wind may be a superposition of such clouds, each at a different velocity. Schwartz & Martin (2004) find that the narrowest Na I D component in a sample of dwarf starbursts is 15 km s<sup>-1</sup>. In a future paper (Rupke & Veilleux 2005, in prep.), we will present high-resolution observations designed to search for such components in ULIRGs.
What about projection effects? If the wind is not spherically symmetric and the wind velocities are not purely radial, the measured velocities will be smaller than the actual wind velocities due to the inclination of the wind with respect to the line of sight. Consider a simple model of a constant-velocity wind emerging perpendicular to the disk of the galaxy, where the highest velocity we observe ($``$600 km s<sup>-1</sup>) is the wind velocity. Lower velocities could then be due to the inclination of the wind. Assuming a uniform distribution of inclinations, we can compute the expected distribution of velocities for such a model. In Figure 2, we compare the observed velocities with those predicted from this simple model. The observed and predicted distributions differ at $`>`$99% confidence in each case. Obviously projection effects are more complex than this simple model. If there really is a uniform distribution of inclinations, the real maximum velocities must be typically smaller than 600 km s<sup>-1</sup>.
### 2.3 Equivalent Width, Optical Depth, and Column Density
Figure 3$`a`$ plots the distribution of the total rest-frame equivalent width of the Na I D feature for every galaxy nucleus in our sample. These values are based on our fits to the data. Table 2 of Paper I lists these values explicitly. We compute a median of 3.3 Å, a dispersion of 2.1 Å, and a maximum of 9.1 Å.
Figure 3$`b`$ plots the distribution of the central optical depth of the Na I D<sub>1</sub> line (the D<sub>2</sub> line has twice the optical depth of the D<sub>1</sub> line). The distributions for IRGs and ULIRGs are quite similar, and K-S and Kuiper tests show that they are consistent with being drawn from the same parent distribution. We see that the peaks of the distributions occur at $`\tau 1`$; in other words, most of the gas is moderately optically thick (if our assumptions about a constant covering fraction and a Maxwellian velocity distribution are correct; see Paper I). The median error in an individual measurement of $`\tau `$ is $``$20%.
The distributions of Na I and H column densities are shown in Figure 4$`a`$ and $`b`$. The distributions peak at log\[$`N`$(Na I)\] $`13.514`$ and log\[$`N`$(H)\] $`21`$. As a check on our results, we have one galaxy (NGC 1614) in common with Schwartz & Martin (2004). Our Na I column density measurement matches theirs within 0.25 dex (despite their much higher spectral resolution). Our measurement of the difference in velocity between the red and blue components in this galaxy ($`\mathrm{\Delta }v=200`$ km s<sup>-1</sup>) also matches theirs.
### 2.4 Covering Fraction
In Figure 3$`c`$, we show the distributions of covering fraction in each subsample. Recall that $`C_f`$ is assumed to be constant within each velocity component. The mean value of $`C_f`$ is $`0.370.45`$ for all three subsamples, and the error in an individual measurement is $`0.050.1`$. For the IRGs, the $`C_f`$ distribution peaks in the $`0.250.5`$ bin. The distribution of $`C_f`$ in ULIRGs has a broader peak in the range $`0.00.5`$, and a possible second peak in the $`0.751.00`$ bin. K-S tests show that the differences between these distributions are not significant, though these tests are not as sensitive to data far from the mean (such as the hypothesized second peak at $`C_f=0.751.00`$ for ULIRGs). The Kuiper test, which is more evenly weighted, shows a significant difference between the IRGs and low-$`z`$ ULIRGs, at 95% confidence.
The physical model behind $`C_f`$ is described in Paper I. In the superwind context, $`C_f`$ may reflect (a) the clumpiness of the wind and/or (b) the global solid angle subtended by the wind with respect to the galaxy center. A covering fraction less than unity may also be caused by light scattered into the line-of-sight, or by the wind having a small size as projected against the background stellar continuum. If the latter is the case, we would expect $`C_f`$to vary significantly with redshift. However, we observe only a negligible dependence on $`z`$ (Table 1). See §4.2 for further discussion of $`C_f`$ in the context of the global covering factor of the outflow.
### 2.5 Spatial Distribution of Absorbing Gas
We know from observations of local galaxies that superwind gas extends over a range of scales from sub-kiloparsec to 10$`+`$ kpc (e.g., Veilleux et al., 2003). The hot wind outer radius in starburst galaxies apparently increases as galaxy size and mass increase (Strickland et al., 2004). In the large, massive galaxies we are studying, we expect winds to extend to radii of several kpc or greater. For instance, recent integral field spectroscopy and deep Chandra imaging and spectroscopy of the nearest ULIRG, Arp 220, show that it contains a superwind with a velocity of 200 km s<sup>-1</sup> at radii of a few kpc (Arribas et al., 2001; McDowell et al., 2003). NGC 6240, a nearby high-luminosity LIRG, contains a superbubble with radius 5 kpc and a bow shock at 20 kpc (Veilleux et al., 2003).
As we discuss below (§2.6), our mass, momentum, energy, and mass entrainment efficiency measurements are sensitive to the radial distribution of the gas. The inner radius of the wind is constrained by the size of the starburst region, and this radius cannot be smaller than a few hundred pc (Downes & Solomon, 1998). Are there any constraints on either the inner or outer radius of the wind from our spectra, obtained by measuring the size of the absorbing region across the spatially extended background continuum light?
The continua in the spectra of most ULIRGs in our sample have angular sizes comparable to the seeing limit. There are a few nearby exceptions. F17207$``$0014 ($`z=0.043`$) possesses blueshifted absorption across the entirety of the spatial profile, implying that this absorption occurs at projected radii of $``$5 kpc. IRAS 20046$``$0623 ($`z=0.084`$) shows a velocity gradient in the Na I D profile across $`6\mathrm{}`$, or $``$9 kpc. The gradient in Na I D matches the emission-line profile, which has a blueshifted tail westward of the western nucleus (see also Murphy et al., 2001). The velocity of this tail matches that of the blueshifted absorption in the western nucleus, which extends over $``$2 kpc. As we discuss below, F10190$`+`$1322:E ($`z=0.076`$) has redshifted absorption that is extended on scales of $`56`$ kpc (§4.1.3) and arises in the disk of the western nucleus. Finally, we observe blueshifted absorption in F10565$`+`$2448 ($`z=0.043`$) that indicates projected radii of $``$2 kpc.
Many more of the IRGs than the ULIRGs have spatially extended continua due to their lower redshifts. Several show clear evidence for stellar rotation in their Na I D profiles; typically these rotation components are narrow. A few IRGs also show more complex structures. The blueshifted Na I D in F16504$`+`$0228 (NGC 6240; $`z=0.024`$) is constant in velocity across most of the continuum, but it becomes more blueshifted in concert with the line emission to the east of the galaxy. This suggests absorbing radii of up to 8 kpc in projection. In F08354$`+`$2555 (NGC 2623; $`z=0.018`$), the blueshifted absorption extends across 22$`\mathrm{}`$, which corresponds to 4 kpc in projected radius (though as we argue below, this is probably not outflowing gas; §4.1.3).
These values are consistent with our assumption of a ‘thin-shell’ outflow located at radii of a few kpc (§2.6). However, they do not rule out a thicker shell extended over several kpc in radius. Furthermore, the absorbing shell cannot be too thin, due to the broad profiles we observe (§2.2). The large global covering factors we measure (§4.2) also argue against narrow radial filaments, which would tend to have small global covering factors.
### 2.6 Mass, Momentum, and Energy
#### 2.6.1 Formulas
We would like to measure the mass, momentum, and energy (and their outflow rates) in these winds for the phase of the ISM that we probe with our observations (roughly, the neutral gas phase). We choose a simple model for the wind that depends on the physical parameters output from our fitting. An alternative is to produce a database of synthetic profiles based on a more complex model; comparing these synthetic profiles directly to our spectra would then yield properties of the wind such as mass, density, etc. Unfortunately, a complex model of this sort is under-constrained by our data.
As in our preliminary report (Rupke et al., 2002), we assume a spherically symmetric mass-conserving free wind, with an instantaneous mass outflow rate and velocity that are independent of radius within the wind and zero outside. The mass outflow rate at a radius $`r`$ within the wind is given by
$$dM/dt(r)=\mathrm{\Omega }\mu m_pn(r)vr^2,$$
(1)
where $`\mathrm{\Omega }`$ is the solid angle subtended by the wind as seen from its origin (i.e., the wind’s global covering factor), $`\mu m_p`$ is the mass per particle ($`m_p`$ being the proton mass and $`\mu `$ a correction for relative He abundance), $`n(r)`$ is the wind number density, and $`v`$ is the wind velocity. We wish to re-write this equation in terms of the (measurable) column density $`N`$ rather than the space density $`n`$. Rearranging to solve for $`n(r)`$, we require a density profile $`n(r)r^2`$ (e.g., an isothermal sphere).
For a wind of finite thickness (inner and outer radii $`r_1`$ and $`r_2`$), the total observed column density in the outflow is the integral of $`n(r)`$ along our line-of-sight:
$$N=_{r_1}^{r_2}n(r)𝑑r.$$
(2)
Solving eq. (1) for $`n(r)`$, substituting this into eq. (2), integrating, and rearranging leaves us with the instantaneous mass outflow rate across any radius $`r`$ within the wind,
$$dM/dt_{thick}^{inst}=\underset{i}{}\mathrm{\Omega }_i\mu m_pN_iv_i\frac{r_1r_2}{r_2r_1}.$$
(3)
The sum is taken over the number of outflowing components in the galaxy: $`\mathrm{\Omega }_i`$ is the solid angle subtended by the component as seen from the wind’s origin, $`N_i`$ is the column density of each component, and $`v_i`$ is the central wind velocity in each component. The corresponding mass of the wind, obtained by simple volume integration of the density profile, is
$$M_{thick}=\underset{i}{}\mathrm{\Omega }_i\mu m_pN_ir_1r_2.$$
(4)
To average the mass outflow rate across a radius $`r<r_1`$ over the wind lifetime, we divide $`M_{thick}`$ by $`t_{wind}=r_2/v_i`$ to give
$$dM/dt_{thick}^{avg}=\underset{i}{}\mathrm{\Omega }_i\mu m_pN_iv_ir_1.$$
(5)
Suppose we assume instead a thin shell, with $`r_1r_2=r`$ and $`\mathrm{\Delta }r=r_2r_1`$. Equations (3)$``$(5) then become
$`dM/dt_{thin}^{inst}`$ $`={\displaystyle \underset{i}{}}\mathrm{\Omega }_i\mu m_pN_iv_i{\displaystyle \frac{r^2}{\mathrm{\Delta }r}},`$ (6)
$`dM/dt_{thin}^{avg}`$ $`={\displaystyle \underset{i}{}}\mathrm{\Omega }_i\mu m_pN_iv_ir,`$ (7)
$`\mathrm{and}`$ $`M_{thin}`$ $`={\displaystyle \underset{i}{}}\mathrm{\Omega }_i\mu m_pN_ir^2.`$ (8)
The energies and momenta and their outflow rates are computed using similar sums:
$`dp/dt`$ $`={\displaystyle \underset{i}{}}(dM/dt)_iv_i,`$ (9)
$`p`$ $`={\displaystyle \underset{i}{}}M_iv_i,`$ (10)
$`dE/dt`$ $`={\displaystyle \underset{i}{}}(dM/dt)_i\times (v_i^2/2+3\sigma _i^2/2),`$ (11)
$`\mathrm{and}`$ $`E`$ $`={\displaystyle \underset{i}{}}M_i(v_i^2/2+3\sigma _i^2/2),`$ (12)
where the energy includes both the ‘bulk’ kinetic energy due to the outflowing gas and ’turbulent’ kinetic energy (where we assume the same $`\sigma `$ in each dimension).
#### 2.6.2 Wind Geometry, Cloud Lifetimes, and Time-Averaging
The masses, momenta, and energies we compute are sensitive to the assumed geometry. In Rupke et al. (2002), we assumed a thick wind that extends from an inner radius $`r_{}`$ to infinity. Here, we will revert to a thin shell of uniform radius 5 kpc. This radius is motivated both by observations of local starbursts (e.g., Strickland et al., 2004; Veilleux et al., 2005) and by our own data (§2.5).
The ‘thin shell’ geometry is, however, uncertain. Theory and simulations predict a thin shell of warm gas behind the forward shock of the wind (e.g., Castor et al., 1975; Suchkov et al., 1994; Strickland & Stevens, 2000). This gas may or may not be accompanied by neutral material (see the discussion in §7 of Paper I), but the shell is a natural interpretation of our simple model. The shell may be partly or completely broken up due to Rayleigh-Taylor instabilities (Suchkov et al., 1994). Observations of local starbursts also find dusty filaments (which presumably contain neutral gas) entrained on the edges of superwind bicones (see Veilleux et al., 2005, and references therein). Our model does not account for these filaments, which would have a broader radial range and subtend a much smaller angle.
Clouds may also exist in the wind interior. These clouds can be ablated and destroyed by thermal evaporation (conduction) or hydrodynamic ablation. Simple formulas for thermal evaporation in the absence of magnetic fields (Cowie et al., 1981) and cloud destruction by hydrodynamic ablation (Klein et al., 1994; Poludnenko et al., 2002) show that small clouds of size 1 pc and density 100 cm<sup>-2</sup> can survive for a time of order $`10^6`$ yr in a wind fluid of density 10<sup>-3</sup> cm<sup>-2</sup> and temperature $`3\times 10^7`$ K. This is not much smaller than the wind lifetime ($`5\times 10^65\times 10^7`$ yr) of a 5 kpc shell with a constant velocity of $`1001000`$ km s<sup>-1</sup>. Various effects can increase cloud lifetimes, including: (1) the presence of azimuthal magnetic fields, which make cloud evaporation less likely, as they decrease conduction relative to the classical Spitzer value (as observed in 30 Dor; Smith & Wang 2004); (2) conversely, increased heat conduction, which may inhibit hydrodynamic instabilities (Marcolini et al., 2005); (3) clouds stop evaporating when they are crushed to high optical depth (Ferrara & Shchekinov, 1993); and (4) Kelvin-Helmholtz instabilities are suppressed by ablation of small bumps in the cloud (Schiano et al., 1995). Cloud lifetimes can be decreased, however, by interactions among nearby clouds (Poludnenko et al., 2004).
In short, the survival times of clouds in the wind are not yet certain, but it is entirely possible that they may survive for a substantial fraction of the wind lifetime and be observable. If they do exist throughout the wind’s interior, then a thick wind model is more applicable. Assuming a thick wind instead of a thin wind will decrease our measured masses, momenta, energies, and outflow rates, since the radial factor in the above equations is the inner radius in the thick wind case, rather than the outer radius. For instance, a thick wind extending from $`15`$ kpc has a mass 5 times lower than the $`r=5`$ kpc thin shell that we assume. However, there are limits to the inner radius of the wind based on the size of the starburst (e.g., Downes & Solomon, 1998), such that our masses, etc., will not decrease by more than a factor of $``$10. More complicated geometries (e.g., irregular, filamentary structures) will alter our results in ways that we cannot easily predict, though we do not expect dramatic departures from our predictions.
A second geometric consideration is the solid angle subtended by the wind, as seen from the wind’s origin (i.e., its global covering factor). Local winds emerging from disk galaxies are typically biconical along the galaxy’s minor axis. We account for this by letting the solid angle subtended by the wind $`\mathrm{\Omega }`$ be less than $`4\pi `$. Furthermore, the wind may be clumpy, rather than a smooth shell. In our model, we divide $`\mathrm{\Omega }`$ into two parts, the large-scale covering factor (related to the wind’s opening angle), given by $`C_\mathrm{\Omega }`$, and the local covering factor (related to the wind’s clumpiness), given by $`C_f`$. Thus, $`\mathrm{\Omega }/4\pi =C_\mathrm{\Omega }C_f`$. For the IRGs, we assume $`C_\mathrm{\Omega }=0.4`$, where 0.4 is based on the average opening angle of winds in local starbursts ($``$65$`\mathrm{°}`$; e.g., Veilleux et al. 2005). For ULIRGs, we use a larger value of $`C_\mathrm{\Omega }=0.8`$, which is motivated by our detection rate (§4.2). We assume that the measured covering fraction $`C_f`$ describes the wind’s local clumpiness.
Furthermore, for the outflow rates, we must specify whether we wish to compute the instantaneous values in the wind or the values averaged over the lifetime of the wind. In Rupke et al. (2002), we quoted instantaneous mass outflow rates. In this paper, we will quote time-averaged outflow rates, which are a more useful quantity when comparing to the star formation rate. We address the relative wind and starburst lifetimes in §2.7.
#### 2.6.3 Results
A final numerical assumption is an average particle mass of 1.4$`m_p`$ to account for the contribution of He. The resulting numerical formulas are
$`M`$ $`=5.6\times 10^8{\displaystyle \left(\frac{C_\mathrm{\Omega }}{0.4}C_f\right)\left(\frac{r^2}{100\mathrm{kpc}^2}\right)\left(\frac{N(\mathrm{H})}{10^{21}\mathrm{cm}^2}\right)M_{\mathrm{}}},`$ (13)
$`dM/dt`$ $`=11.5{\displaystyle \left(\frac{C_\mathrm{\Omega }}{0.4}C_f\right)\left(\frac{r}{10\mathrm{kpc}}\right)\left(\frac{N(\mathrm{H})}{10^{21}\mathrm{cm}^2}\right)\left(\frac{|\mathrm{\Delta }v|}{200\mathrm{km}\mathrm{s}^1}\right)M_{\mathrm{}}\mathrm{yr}^1},`$ (14)
$`p`$ $`=2.2\times 10^{49}{\displaystyle \left(\frac{C_\mathrm{\Omega }}{0.4}C_f\right)\left(\frac{r^2}{100\mathrm{kpc}^2}\right)\left(\frac{N(\mathrm{H})}{10^{21}\mathrm{cm}^2}\right)\left(\frac{|\mathrm{\Delta }v|}{200\mathrm{km}\mathrm{s}^1}\right)\mathrm{dyne}\mathrm{s}},`$ (15)
$`dp/dt`$ $`=1.4\times 10^{34}{\displaystyle \left(\frac{C_\mathrm{\Omega }}{0.4}C_f\right)\left(\frac{r}{10\mathrm{kpc}}\right)\left(\frac{N(\mathrm{H})}{10^{21}\mathrm{cm}^2}\right)\left(\frac{|\mathrm{\Delta }v|}{200\mathrm{km}\mathrm{s}^1}\right)^2\mathrm{dyne}},`$ (16)
$`E`$ $`=2.2\times 10^{56}{\displaystyle \left(\frac{C_\mathrm{\Omega }}{0.4}C_f\right)\left(\frac{r^2}{100\mathrm{kpc}^2}\right)\left(\frac{N(\mathrm{H})}{10^{21}\mathrm{cm}^2}\right)}`$
$`\times \left[\left({\displaystyle \frac{|\mathrm{\Delta }v|}{200\mathrm{km}\mathrm{s}^1}}\right)^2+1.5\left({\displaystyle \frac{b}{200\mathrm{km}\mathrm{s}^1}}\right)^2\right]\mathrm{erg},`$
$`\mathrm{and}`$ $`dE/dt`$ $`=1.4\times 10^{41}{\displaystyle \left(\frac{C_\mathrm{\Omega }}{0.4}C_f\right)\left(\frac{r}{10\mathrm{kpc}}\right)\left(\frac{N(\mathrm{H})}{10^{21}\mathrm{cm}^2}\right)\left(\frac{|\mathrm{\Delta }v|}{200\mathrm{km}\mathrm{s}^1}\right)}`$
$`\times \left[\left({\displaystyle \frac{|\mathrm{\Delta }v|}{200\mathrm{km}\mathrm{s}^1}}\right)^2+1.5\left({\displaystyle \frac{b}{200\mathrm{km}\mathrm{s}^1}}\right)^2\right]\mathrm{erg}\mathrm{s}^1.`$
The sums here are performed over the outflowing velocity components in each galaxy. We remind the reader that we assume $`r=5`$ kpc universally, but different values of $`C_\mathrm{\Omega }`$ for the LIRGs (0.4) and the ULIRGs (0.8). We record these quantities for each galaxy in Table 2, their median values and dispersions in Table 1, and their distributions in Figures 4$`ch`$. K-S and Kuiper tests show that the distributions for each subsample are consistent with being drawn from the same parent distribution.
### 2.7 Mass Entrainment Efficiency
The mass entrainment efficiency compares the amount of gas entrained by the hot wind fluid to the amount of gas being turned into stars. Star formation in these galaxies occurs at the galactic nucleus in dense concentrations of molecular gas. However, most of the gas in a superwind is entrained from the disk and halo of the host galaxy as the wind propagates outward, either by acceleration of cold clouds or mass-loading of these clouds into the wind fluid (Suchkov et al., 1994; Strickland & Stevens, 2000). The mass entrainment efficiency is thus a useful quantitative description of how the wind’s evolution is connected to its power source (the starburst).
The star formation rates for our sample are computed from the usual Kennicutt formula relating SFR to the total infrared luminosity (Kennicutt, 1998),
$$\mathrm{SFR}=\alpha \frac{L_{\mathrm{IR}}}{5.8\times 10^9L_{\mathrm{IR}}}.$$
(19)
The model for this scaling assumes continuous star formation, with a starburst age of $`10100`$ Myr and a Salpeter initial mass function (IMF) with stellar mass limits 1 and 100 $`M_{\mathrm{}}`$. We correct for AGN contribution to $`L_{\mathrm{IR}}`$; $`\alpha `$ equals the fraction of the infrared luminosity powered by star formation. Infrared Space Observatory (ISO) observations show that $`70\%95\%`$ of the infrared luminosity of a typical ULIRG is dust-reprocessed light from young stars (Genzel et al., 1998). These values apply to our sample, which we have selected to include objects whose spectra indicate vigorous starbursts. We therefore assume $`\alpha =0.8`$ for our ULIRGs. For the IRGs, we use $`\alpha =1.0`$.
The mass entrainment efficiency is the mass outflow rate normalized to the corresponding star formation rate of a galaxy:
$$\eta \frac{dM/dt}{\mathrm{SFR}}.$$
(20)
Figure 4$`i`$ shows the distributions of $`\eta `$ for each subsample. We measure median values of 0.33, 0.19, and 0.09 in our IRG, low-$`z`$ ULIRG, and high-$`z`$ ULIRG subsamples, respectively. The full range of $`\eta `$ is three orders of magnitude, $`0.0110`$. The distribution of $`\eta `$ for the IRGs is different from that of the ULIRGs at $`>`$99% confidence. As with $`dM/dt`$, $`\eta `$ is sensitive to the assumed wind geometry and to our assumptions about the local physical state of the gas.
Note that $`dM/dt`$ and SFR are time-averages over the wind and starburst lifetimes, respectively. The superwinds in these galaxies are powered initially by stellar winds, and by supernovae at times $``$10<sup>7</sup> yr (Leitherer et al., 1992). The wind lifetimes we calculate ($`t_{wind}=r_2/v`$) are of order $`550`$ Myr; thus, the winds are generally supernovae-powered. The starburst lifetimes ($``$100 Myr) are comparable to the wind lifetimes, as the gas consumption timescales for the molecular gas in the nuclei of ULIRGs indicate. (ULIRGs have molecular gas masses $``$10<sup>10</sup> $`M_{\mathrm{}}`$ \[Sanders et al. 1991; Solomon et al. 1997\] and form stars at rates $``$150 M yr<sup>-1</sup>.) Thus, it is appropriate to directly compare our computed $`dM/dt`$ and SFR. In this model, $`\eta `$ is also a time-average over the lifetime of the wind.
## 3 OUTFLOW PROPERTIES AND HOST GALAXY PROPERTIES
One of the primary purposes of this survey is to look for dependence of outflow properties on the properties of the host galaxies. This allows us to (1) better understand the physics of outflows and (2) describe their properties using approximate analytic functions. The latter is especially useful as input to theoretical analysis and simulations. The galaxy properties that we use in this analysis are discussed in Paper I. They include star formation, optical and near-infrared luminosity, circular velocity (mass), and spectral type. Ideally, it would be useful to also look for a dependence on star formation rate surface density. However, since most of our objects are not well-resolved, we cannot do this reliably.
In determining the presence of a correlation between two quantities, we use three tests: Pearson’s correlation coefficient, Spearman’s correlation coefficient, and a weighted least-squares fit. We accept as significant those correlations for which the slope in the fit ($`a\pm \delta a`$, where log $`X`$ $`=a`$ log $`Y+Y_0`$) satisfies $`a>(3\times \delta a)`$ and for which the probability of no correlation using Pearson’s $`r`$ is less than $`5\%`$ (in all cases for which it is $`<`$0.05, it is also $`<`$0.01). The probability for a null result using Spearman’s $`r`$ is also typically (though not always) low in these cases. In Table 3, we list the power-law slopes of the computed fits and correlation coefficients for each significant correlation.
In this analysis, we include a few LIRGs from Heckman et al. (2000), ULIRGs from Martin (2005), and dwarf starbursts from Schwartz & Martin (2004). We only use velocities from Heckman et al. (2000), since the fits to individual galaxies do not account for covering fraction, and we only include galaxies with LINER or H II spectral types (4 galaxies total). From the Martin (2005) data, we only use galaxies with measured spectral types of LINER or H II and Na I D doublet ratio ratios $`R>1.1`$ which are not already in our sample (6 galaxies total; see next paragraph for justification of the doublet ratio criterion). We use our equations to compute the properties of these galaxies starting with $`\mathrm{\Delta }v`$, $`\tau `$, $`b`$, and $`C_f`$ for each galaxy. We do the same for the four dwarf starbursts with detectable winds in Schwartz & Martin (2004). For the Martin (2005) ULIRGs, we assume $`C_\mathrm{\Omega }=0.8`$ and $`r=5`$ kpc as we do for our ULIRGs. For the dwarf starbursts, we assume $`C_\mathrm{\Omega }=0.4`$ (as for the LIRGs) and use the measured shell radii listed in Schwartz & Martin (2004), with the radius changing from object to object and ranging from 0.1$``$1 kpc. To estimate errors where none are given in these references, we use median values from our own observations at similar resolution.
We note as a caveat that two of the dwarf galaxies have Na I D doublet ratios $`R1.1`$, and thus the lines are optically thick (Schwartz & Martin, 2004). The technique used by Schwartz & Martin (2004) to recover the column densities assumes Gaussians in intensity and thus does not self-consistently treat cases of high $`\tau `$. The resulting column densities are thus overestimates to the actual values under the assumption of a Maxwellian velocity distribution (see Paper I for extensive discussion of this issue).
### 3.1 Detection Rate
Figure 5 shows the dependence of detection rate on star formation rate, $`K`$\- or $`K^{}`$-band magnitude, $`R`$-band magnitude, and circular velocity. We arrive at the bins in this figure by dividing the full range of values into three bins of equal size. The results show that detection rate increases as SFR increases. (This conclusion is bolstered by dividing the IRGs by IR luminosity; see below.) There is only weak or no dependence of detection rate on $`M_{K^{()}}`$, $`M_R`$, or $`v_c`$. Note that, although Figure 5 does not show it, there is a decrease in the detection rate at the highest SFR, as seen by comparing the low-$`z`$ and high-$`z`$ ULIRGs (Table 1). We discuss this further in §4.6.
In §4.2, we argue that the detection rate likely reflects the wind opening angle rather than the frequency of occurrence of winds (which we suggest is close to unity in infrared-luminous galaxies). The increase of detection rate with SFR is probably a result of the different wind geometry in ULIRGs, which have high SFR. The observed difference in wind geometry could in turn reflect the nature of ULIRGs as merging galaxies. The fact that we observe no dependence of detection rate on galactic mass implies that other physics besides the gravitational potential determine whether or not an outflow is formed.
The detection rate does not depend on spectral type. In each subsample, the detection rate of winds in LINERs is statistically indistinguishable from that in H II nuclei. Overall, the detection rate is identical in LINERs and H II nuclei: $`50\%`$. (We find winds in 20 of 40 LINER nuclei and 23 of 43 H II nuclei.)
For the 1 Jy galaxies, interaction classes are available (Veilleux et al., 2002). Most or all of these ULIRGs are involved in a major merger, and these classes describe what stage of the interaction that each galaxy is in. We find that galaxies which are in the latest stages of interaction (classes IV = ‘merger’ and V = ‘old merger’) have the same detection rate: $`(82\pm 9)\%`$ for class IV galaxies, and $`(86\pm 13)\%`$ for class V galaxies. Class III (= ‘pre-merger’) galaxies, which are at an earlier interaction stage and have at least two identifiable nuclei, show the same detection rate within the errors: $`(67\pm 19)\%`$.
The IRGs do not have systematic classifications attached to them. However, from examination of DSS2 and 2MASS images, there is evidence of a major interaction or minor merger in many of them, as seen in the presence of tidal tails, very irregular stellar morphology, and/or large companion galaxies that appear to be interacting with the host. (Many of those galaxies for which we do not observe evidence of an interaction are too distant for the images to show interesting features.) Of the IRGs which appear to be undergoing or to have recently undergone a major interaction or minor merger, we detect a wind in 5 of 11, or $`45\%`$, of them. This percentage is identical to the overall detection rate of $`43\%`$ in the IRGs.
Alternatively, Ishida (2004) finds that all galaxies with $`L_{\mathrm{IR}}>10^{11.5}L_{\mathrm{}}`$ are undergoing a major interaction. If we divide our IRG subsample by luminosity, then 8/13 galaxies (62$`\pm `$13%) with $`L_{\mathrm{IR}}>10^{11.5}L_{\mathrm{}}`$ host winds, vs. 7/22 galaxies (32$`\pm `$10%) with $`L_{\mathrm{IR}}<10^{11.5}L_{\mathrm{}}`$. This could indicate some dependence of detection rate on the presence of a major interaction; however, without detailed morphological studies of the galaxies in our sample, this result is inconclusive.
In short, we conclude that there is no strong dependence of detection rate on merger stage. The change in detection rate with SFR, which reflects ouflow geometry, could ultimately be due to the presence/absence of an interaction or strength of interaction, although we do not have firm evidence to demonstrate this.
### 3.2 Outflow Velocity
In Figure 6, we plot for each galaxy the logarithm of the maximum outflow velocity $`\mathrm{\Delta }v_{max}`$ as a function of SFR, $`M_{K^{()}}`$, $`M_R`$, and $`v_c`$. These figures show that $`\mathrm{\Delta }v_{max}`$ is not significantly correlated with any of these quantities in our data. However, if we add four dwarf galaxies (Schwartz & Martin, 2004), increasing the range of galaxy properties probed (e.g., the range in SFR increases from 2 to 4 orders of magnitude), we measure significant correlations of maximum velocity with each galaxy property. The strongest correlation we find is with $`v_c`$: $`|\mathrm{\Delta }v_{max}|v_c^{0.8\pm 0.2}`$. The other dependences we compute are weak ($`|a|0.10.2`$).
In Figure 7, we do the same for the velocity of the highest column density outflowing gas in each galaxy. Again, no correlations emerge in our data. If we add the dwarf galaxies, interesting behavior emerges. Allowed velocities increase slowly but smoothly with SFR, luminosity, and mass, but at some characterisitic value of these galaxy properties there is a sharp increase in the allowed velocities of the optically thickest gas. The lower right-hand corners of panels $`eh`$ in this figure are not populated, but this is a selection effect; we do not include points with $`\mathrm{\Delta }v_{maxN}>50`$ km s<sup>-1</sup> in our analysis because of measurement uncertainties.
In the Discussion (§4.4), we discuss the interpretation of these correlations of velocity with galaxy properties.
Some authors have suggested that the shock-like line ratios found in the nuclei of galaxies classified as LINERs are due to the presence of shocks in outflowing gas (Veilleux et al., 1995; Lutz et al., 1999; Taniguchi et al., 1999). We can test this hypothesis by comparing the velocities in LINERs and H II galaxies.
We find that the median maximum velocity $`|\mathrm{\Delta }v_{max}|`$ in LINER nuclei is higher than in H II nuclei by 125 km s<sup>-1</sup> (393 vs. 267 km s<sup>-1</sup>, respectively). The median velocity of the highest column density gas, $`|\mathrm{\Delta }v_{maxN}|`$, is higher in LINERs by 100 km s<sup>-1</sup> (229 vs. 119 km s<sup>-1</sup>, respectively). Finally, the median velocity width $`b`$ is higher in LINERs by 70 km s<sup>-1</sup> (224 vs. 152 km s<sup>-1</sup>). The overall distributions of maximum velocity and velocity width in all LINERs in our sample are different from those of H II nuclei at $`>`$95% confidence (see Figure 8). However, the confidence level is too low for $`\mathrm{\Delta }v_{maxN}`$ to demonstrate a convincing difference.
This difference in outflow velocity and velocity dispersion could explain some or all of the physical differences between the LINER and H II optical spectral classes. As Dopita & Sutherland (1995) demonstrate, a modest increase in shock velocity can increase the \[N II\] $`\lambda 6583`$/H$`\alpha `$ line ratio and push a galaxy from having an H II galaxy classification to a LINER classification. However, other effects may also contribute to the line ratios in some LINERs, including a weak AGN.
We also observe an increase in the median maximum velocity in the latest merger stages in the 1 Jy ULIRGs. In interaction classes III, IV, and V, we observe median values of $`|\mathrm{\Delta }v_{max}|`$ of 360, 350, and 450 km s<sup>-1</sup>, respectively. This conclusion is tentative, since we have $``$10 galaxies with outflowing gas in each class. Furthermore, the median value of $`\mathrm{\Delta }v_{maxN}`$ peaks in interaction class IV ($`\mathrm{\Delta }v_{maxN}`$$`=`$ 130, 210, and 160 km s<sup>-1</sup> for interaction classes III$``$V, respectively). If there are really larger velocities later in the merger, then the acceleration of cold gas clouds is somehow more efficient in these stages. This could be due simply to more time for the acceleration to occur, more available energy, or less pressure from surrounding gas (which has been cleared out by tidal forces or previous action of winds, or has decreased because of a decrease in density with radius).
### 3.3 Mass, Momentum, Energy, and $`\eta `$
In Figure 9 we plot outflowing column density of Na I as a function of SFR, $`M_{K^{()}}`$, $`M_R`$, and $`v_c`$. This figure shows that $`N`$(Na I) is weakly correlated with SFR, $`M_R`$, and galactic mass ($`v_c^2`$): $`|a|0.30.6`$.
In Figures $`\text{10}\text{15}`$, we plot mass, momentum, energy, and their outflow rates as a function of galaxy properties. These figures show that for our sample considered alone, there are no significant correlations, except possibly between $`E`$ and SFR. However, if we also include the four dwarf galaxies from Schwartz & Martin (2004), we find significant correlations in almost every case. The measured correlations with $`R`$\- and $`K`$-band luminosity are weak ($`|a|1`$). The overall slopes of mass and momentum with respect to SFR are approximately linear ($`a=1.11.4`$, with $`\delta a0.1`$). However, the slope of energy with respect to SFR is steeper ($`a=1.61.8`$). The dependence of mass, momentum, and energy on galactic mass is the strongest ($`a=1.52.5`$).
Above a characteristic star formation rate (SFR $`10100`$ M yr<sup>-1</sup>), luminosity, and mass, these correlations disappear. In other words, the masses, momenta, and energies become approximately constant as a function of galaxy properties.
In Figure 16, we plot mass entrainment efficiency as a function of galaxy properties. The mass entrainment efficiency is roughly constant as a function of SFR, luminosity, and mass. However, the scatter is large (over two orders of magnitude). Furthermore, at the highest values of SFR (i.e., in our sample alone), $`\eta `$ decreases as SFR and $`M_{K^{()}}`$ increase.
How are these correlations affected by our use of the luminosity-metallicity relationship (Paper I)? If instead we assume solar metallicity, we find that the resulting correlations are unaffected. The changes in slope are less than 1$`\sigma `$ in each case. This convincingly confirms that non-solar metallicities do not drive the trends of outflow properties vs. galaxy properties that we observe. However, assuming solar metallicity does move the normalization of the observed correlations upward by a factor of $``$2, since the galaxies in our sample have twice solar metallicity on average. Thus, the hydrogen column density, mass, momentum, energy, and mass entrainment efficiency are higher by a factor of $``$2 on average under the assumption of solar metallicity.
In the Discussion (§4.4), we interpret these correlations of mass, momentum, energy, and mass entrainment efficiency with galaxy properties.
### 3.4 Correlations with Emission Lines
Resolved outflows in the local universe present distinctive properties when observed in optical line emission, including limb-brightened bipolar structures and line-splitting in velocity space. The surface brightnesses of these winds are typically much smaller than those of their background galaxies, making it difficult to see emission-line evidence of small superbubbles in many distant galaxies. However, very luminous, extended emission-line nebulae exist in nearby LIRGs and ULIRGs such as NGC 6240 and Arp 220 (Heckman et al., 1987; Armus et al., 1990; Veilleux et al., 2003; Gerssen et al., 2004) and may be powered by starburst-driven outflows.
In Figure 17, we plot inverted profiles of the \[N II\] $`\lambda \lambda 6548,6583`$ and \[O III\] $`\lambda 5007`$ lines below the Na I D profile for 17 galaxies where the blue and red emission line wings have asymmetric profiles (i.e., one has a higher maximum velocity and/or more flux than the other) . For each emission line, we fitted and subtracted the continuum using a low-order polynomial. The extraction apertures for these data are the same as those for the Na I D spectra, which cover most of the visible continuum. The \[N II\] line profile is a combination of the red half of the $`\lambda 6583`$ line and the blue half of the $`\lambda 6548`$ line. The latter is not seriously contaminated by H$`\alpha `$, since they are 700 km s<sup>-1</sup> apart.
In this subset of 17 galaxies, only 1 has a red emission-line asymmetry in the profile wings (F10190$`+`$1322:E, discussed in §2.5 and §4.1.3). Thirteen of the other 16 galaxies, or 81%, have Na I D components that are blueshifted by more than 50 km s<sup>-1</sup>. Two of these are not interpreted as winds on the basis of other considerations (§4.1), leaving 11 of 16, or 69%.
Thus, in galaxies with blue-asymmetric emission-line wings in \[N II\] and/or \[O III\], 69% possess a superwind on the basis of Na I D. The corollary: of the 45 galaxies in our sample that possess superwinds on the basis of Na I D, at least 11, or $``$25%, possess blue-asymmetric emission-line wings. More galaxies may possess these wings, but they are not detected at our S/N.
These blue-asymmetric emission-line wings may very well represent the ionized phase of the outflow, which is seen prominently in the halos of nearby edge-on starbursts. Interestingly, in half (6 out of 11) of these cases, the ionized gas has a higher maximum velocity by a few hundred km s<sup>-1</sup> (in the other half the ionized and neutral velocities are roughly the same). The emission line velocities reach up to $``$1000 km s<sup>-1</sup> or higher in several cases. Furthermore, in most cases the \[O III\] and \[N II\] profiles match well, suggesting that the low- and high-ionization states generally have the same velocities, at least in starburst outflows.
Not only are the maximum ionized gas velocities larger than the maximum neutral gas velocities, but the velocities of the bulk of the outflowing gas are, as well. Deprojected velocities of ionized gas in edge-on starbursting galaxies are $``$170 km s<sup>-1</sup> on average (Lehnert & Heckman, 1996); for galaxies of comparable SFR (the IRGs), the velocities of the bulk of the neutral gas are 100 km s<sup>-1</sup> on average (§2.2 and Table 1). The latter do not need to be deprojected significantly, since they are typically detected while close to face-on (§4.2). Thus the ionized gas velocities are $``$70% greater than the neutral-gas velocities in these galaxies.
### 3.5 Comparison to H I 21 cm Spectra
Given the high column densities of neutral H I that we infer with Na I D ($`10^{21}`$ cm<sup>-2</sup>), we might expect to also observe outflowing gas directly in emission or absorption using the H I 21 cm line. Since H I data exists for many of the objects in our IRG subsample, we can search for direct evidence of neutral, outflowing H I in the form of blue wings in the absorption profiles.
The kinematics of the H I gas in F10565$`+`$2448 are the most interesting. The systemic velocity that we measure from nebular emission lines agrees with the narrow, deep absorption trough in the H I 21 cm line at $``$12900 km s<sup>-1</sup> (Mirabel & Sanders, 1988). There appears to be a broad, blueshifted component in H I that extends to roughly the same outflow velocities as seen in Na I D. There is also redshifted H I in emission at $``$13100 km s<sup>-1</sup>. If the Na I D and broad H I absorption represent a bubble expanding in our direction along the line of sight, the redshifted H I emission line could indicate a counter-bubble that is expanding away from us.
Three other objects in our sample (F01417$`+`$1651, F02512$`+`$1446:S, and F03359$`+`$1523) also show H I absorption in Mirabel & Sanders (1988). The S/N of the H I spectra for these objects are comparable to that of F10565$`+`$2448, but there are no obvious broad, blueshifted components.
## 4 DISCUSSION
### 4.1 Alternative Explanations
The zeroth order interpretation of the blueshifted absorption lines in these galaxies is that they are produced by starburst-driven outflows. Alternative explanations should produce velocity distributions that are symmetric about $`\mathrm{\Delta }v=0`$ km s<sup>-1</sup>, for reasons we discuss below. Thus, the maximum number of blueshifted components attributable to other phenomena is a mirror reflection of the distribution of red components seen in Figure 1$`a`$, and is therefore not significant.
#### 4.1.1 Gas in Rotation
We observe rotation in the emission lines of most IRGs and in a sizable fraction of ULIRGs, suggesting that there are ordered gas disks in most of these systems. In only a few cases is rotation also observed in Na I D. In a large number of these galaxies ($``$17 total), we see blueshifted Na I D that is at or near the blue rotation arm of the galaxy in velocity space. Rarely (in only $``$3 cases) do we observe the Na I D near the red rotation arm. Since there is no reason for this asymmetry in a rotating disk scenario, we conclude that this gas is in general not in simple rotation.
#### 4.1.2 Tidal Debris
The red (and some of the blue) components that we observe in ULIRGs may be tidal debris, gas stirred up by vigorous interactions. The simulations of Barnes & Hernquist (1991) show that much of the gas in an equal-mass prograde merger flows suddenly to a compact region at the merger center, and large tidal arms are spun out. However, much of the extended tidal debris falls back to the merger center over Gyr time scales. In NGC 7252, radial velocities of up to $`\pm 200`$ km s<sup>-1</sup> are observed due to velocity gradients along tidal tails, the redshifted velocities representing gas falling back to the disk (Hibbard & Mihos, 1995).
From examination of the velocity distribution of tidal material in the simulations of Hibbard & Mihos (1995), we expect a narrow velocity width from a tidal tail in projection, while we observe mostly broad profiles. However, there are two illustrative cases of possible tidal debris in ULIRGs. F09039$`+`$0503 possesses a narrow (FWHM $``$ 30 km s<sup>-1</sup>; Rupke & Veilleux 2005, in prep.) component that is redshifted by $`180190`$ km s<sup>-1</sup> with respect to systemic. The velocity is comparable to those measured for tidal features in NGC 7252 (Hibbard & Mihos, 1995), and its small width is suggestive of a compact feature such as a tidal tail. In NGC 6240, there is an emission- and absorption-line feature to the east of the disk; this feature lies atop a stellar structure that appears tidal in origin (e.g., Veilleux et al., 2003). However, the evidence is only circumstantial; this galaxy also has an emission-line superbubble to the west of the galaxy (e.g., Veilleux et al., 2003), and it is conceivable that the Na I D absorption and line emission to the west of the galaxy trace a counterbubble.
#### 4.1.3 Multiple Nuclei and Overlapping Disks
We observe evidence of overlapping disks in several double-nucleus galaxies. In five galaxies, there is a redshifted absorbing component in the spectrum of one nucleus that is within 40 km s<sup>-1</sup> of the redshift of the other nucleus; the nuclei with these components are F01417$`+`$1651:S, F02411$``$0353:NE, F16333$`+`$4630:W, F16474$`+`$3430:N, and F23234$`+`$0946:E. A sixth galaxy, F08354$`+`$2555, contains Na I D that is blueshifted from the H I velocity but is coincident with the emission-line peak, which also corresponds to the velocity of a compact stellar object $`510`$″ S of the nucleus; the Na I D may be associated with either the galaxy nucleus or the southern object. Finally, F10190$`+`$1322:E has redshifted absorption at $`+320`$ km s<sup>-1</sup> with respect to systemic, as well as extended, redshifted emission. The velocity of the part of the western disk that overlaps the eastern nucleus matches the redshifted absorption in velocity space (Murphy et al., 2001).
This suggests that some of the components in our velocity distribution are due to the projection of gas disks along the line of sight. Notably, these seven galaxies still contain three cases of high-velocity blueshifted absorption that are consistent with the outflow hypothesis. In our analysis, we have removed the components that appear to be due to these overlapping disks, as we find this hypothesis more compelling than the outflow one for this small subset of components.
#### 4.1.4 Merger-Induced Winds
Recently, Cox et al. (2004) have hypothesized that the shocks produced as gas funnels to the center of the merger of two equal-mass galaxies can produce outflows of hot gas. These shocks heat a large amount of gas to $`10^{67}`$ K, which, in analogy to a starburst-driven superwind, expands radially outward at speeds $``$200 km s<sup>-1</sup>. This gas could potentially entrain cold gas clouds and evolve much as a starburst-driven superwind. However, the energy injection region may be on larger scales (several kpc vs. $``$1 kpc for a superwind). Further exploration of this idea is warranted, but it is currently indistinguishable in our data from the superwind hypothesis.
### 4.2 Frequency of Occurrence and Global Covering Factor
The detection rate $`D`$ is a function of both the actual frequency of occurrence of winds $`F`$ and the global angular covering factor of the wind, $`\mathrm{\Omega }`$. To better understand this, we make the assumption that a wind is detected if our line-of-sight lies within the opening angle of the wind. (For an outflow emerging perpendicular to a galactic disk, the opening angle is delineated by the biconical structure of the wind.) Within that opening angle, there may be local clumping of the wind material. The global angular covering factor, or solid angle subtended by the wind, is the product of the these two factors (the opening angle and local clumping). We assume that the opening angle factor $`C_\mathrm{\Omega }`$ (a fraction of 4$`\pi `$) is given by the detection rate (which is the case if the frequency of occurrence of winds is 100%), and that the local clumping is given by the measured covering fraction along the line-of-sight, $`C_f`$. The global covering factor is then given by $`\mathrm{\Omega }/4\pi =C_\mathrm{\Omega }C_f=DC_f`$ (see §2.6.2 and Crenshaw et al. 2003). In the more general case, where $`F`$ is not necessarily unity, we can set lower limits to $`F`$ and $`C_\mathrm{\Omega }`$ using our detection rate: $`D<F<1`$ and $`D<C_\mathrm{\Omega }<1`$.
We already know that $`C_\mathrm{\Omega }`$ is less than unity in local galaxies ($``$0.4 on average; e.g., Veilleux et al. 2005) and is similar to our detection rate (§2.1). We also observe marginally significant correlations between the outflow detection rate and the apparent ellipticity of the galaxy in the IRG subsample. We find that we are very likely to observe winds in galaxies that are almost face-on (8 outflows in 15 galaxies) but not at all likely in edge-on galaxies (2 out of 12 galaxies). This is consistent with values of $`C_\mathrm{\Omega }`$ close to $`1/2`$, under the assumptions that winds occur in all starbursts and emerge perpendicular to the galaxy disk.
In ULIRGs, however, both $`C_\mathrm{\Omega }`$ and $`F`$ must be at least 0.7, since $`D0.7`$. Thus, the geometry of outflows in ULIRGs is not the same as in local disk galaxies.
In either case, we conclude that winds occur in almost all starbursting, infrared-luminous galaxies (i.e., $`F1`$). To get the global covering factor, $`\mathrm{\Omega }`$, we then fold together our detection rates and measured average covering fractions along the line of sight (see Table 1). We thus compute $`\mathrm{\Omega }/4\pi 0.15`$ for the IRGs and $`\mathrm{\Omega }/4\pi 0.30`$ for ULIRGs.
The high-redshift Lyman-break galaxies have star formation rates similar to the IRGs (Pettini et al., 1998). However, the detection rate of winds in LBGs is almost 100% (Adelberger et al., 2003). The method of detection involves both absorption and emission lines, but it seems to imply that the opening angle and global covering factor of these winds are much higher than that in local galaxies of similar SFR. On large scales, the opening angles of LBG winds are smaller, however, perhaps closer to 4$`\pi `$/3 (Adelberger et al., 2005); this value is similar to that of IRGs.
### 4.3 Gas Escape Fraction
In our preliminary report (Rupke et al., 2002), we tentatively claimed that the fraction of gas in ULIRG superwinds that escapes the host galaxy ($`f_{esc}`$) is high, perhaps up to $`4050\%`$. We computed this value by estimating the circular velocity, $`v_c`$, in each galaxy and comparing it to the escape velocity, $`v_{esc}`$ (computed using a singular isothermal sphere). However, we previously included two Seyfert 2s with high outflow velocities, whose winds are not necessarily starburst-driven (Rupke et al., 2005b).
In this work, we continue to use a singular isothermal sphere with $`r_{max}/r=10100`$. If the gas absorbs at radius $`r`$, then the escape velocity is parameterized uniquely by $`v_c`$ and $`r_{max}/r`$. Our procedure to calculate $`f_{esc}`$ is as follows: (a) to get $`v_c`$, use measured values where possible; otherwise, for ULIRGs, use an average value (K. Dasyra, private communication; see Paper I for more details); (b) use $`v_c`$ to compute $`v_{esc}`$; (c) compute the mass or mass outflow rate of gas that has a velocity above $`v_{esc}`$; (d) sum $`dM/dt`$ and $`dM/dt_{esc}`$ over all galaxies with measured $`v_c`$; and (e) divide $`dM/dt_{esc}^{total}`$ by $`dM/dt^{total}`$. (Note that $`M`$ and $`dM/dt`$ are interchangeable in this algorithm; the values of $`f_{esc}`$ computed using $`M`$ are comparable to those computed using $`dM/dt`$, but smaller by a factor of 2. The discrepancy is due to the extra factor of $`\mathrm{\Delta }v`$ in the definition of $`dM/dt`$; see §2.6.3.)
By examining Figure 18$`b`$, we see that a significant number of galaxies have maximum velocities close to or above the predicted $`v_{esc}`$ for a singular isothermal sphere. Ignoring halo drag (Silich & Tenorio-Tagle, 2001) or acceleration of the wind, we compute that between 5% and 20% of the neutral material in these winds will escape the galaxy and enter the IGM, for $`r_{max}/r=10100`$. These numbers are smaller than our previous estimates (Rupke et al., 2002), but are still significant. The hot, freely-expanding wind that drives the entrained, neutral material (and carries the majority of the metals; Martin et al. 2002) is even more likely to escape. If we correct for possible projection effects in the velocities, this escape fraction is likely to increase.
Some of the maximum velocities that we measure are close to plausible values for $`v_{esc}`$. Is this coincidental or suggestive of further physics? We suggest that there is a good possibility that there is (or was) material of higher velocities, as is seen in local starburst-driven superwinds like that in NGC 3079 at velocities of up to 1500 km s<sup>-1</sup> (Veilleux et al., 1994; Cecil et al., 2001). This high-velocity gas may be better probed by emission lines, which are more sensitive to higher-temperature, lower-density gas. We do observe higher emission-line than absorption-line velocities in several galaxies (§3.4). Assuming that this gas moves radially (though it may form vortices; see Cecil et al. 2001), it will expand more quickly than the low-velocity gas and dissipate as it reaches large radius, giving it a low cross-section in the line-of-sight and making it difficult to detect using absorption-line probes. It may be that the single galaxy in our sample with $`\mathrm{\Delta }v_{max}>1000`$ km s<sup>-1</sup> is an example of this, where the gas has not completely escaped to large radii. However, the Na I D feature in this galaxy is quite broad and deep, and for this reason not obviously consistent with this interpretation.
### 4.4 Comparison with Theory
A substantial number of numerical simulations of superwinds have been performed over the years (e.g., Suchkov et al. 1994; Strickland & Stevens 2000). These models do not typically make predictions about values of mass, momentum, and energy in the cold gas. They do show that there are large quantities of warm gas in the superwind, distributed both in filaments and clouds entrained from the disk and as a swept-up shell surrounding the wind (which is disrupted by Rayleigh-Taylor instabilities at ‘blowout’). Uncertainties in these models still exist because they treat the ISM as a continuous medium and do not include the microphysics of wind/cloud interactions (such as conduction and ablation leading to mass-loading of the wind fluid and cloud destruction, small-scale hydrodynamic instabilities, and radiative cooling). New simulations are underway (Cooper, Sutherland, & Bicknell 2005, in prep.) incorporating a fractal gas distribution, which will better treat the physics of entrainment and thus the gas mass in the wind. This should in turn lead to accurate predictions of mass outflow rates.
In lieu of comparison to detailed wind simulations, there is a limited amount of analytic work to which we can compare our results (e.g., Silk, 2003; Murray et al., 2005). Furthermore, we can also compare the properties of the entrained gas to the properties of the hot gas we expect to be produced by the starburst (Leitherer et al., 1999). We will also compare to predictions based on radiation pressure.
#### 4.4.1 Velocities
The gas in these winds may be driven by ram pressure from the hot free wind, by radiation pressure from the starburst on the gas and dust in the wind, or by some combination of the two. In the case that ram pressure drives the wind, the properties of the hot gas help determine the final wind velocities. Murray et al. (2005, eq. A3) predict the ‘characteristic’ velocities of clouds accelerated by the hot gas, as well as the cloud velocity as a function of radius. The maximum velocity that these clouds can attain is close to the characteristic velocity and is largely constrained by the velocity of the hot wind that drives them.
In Figure 18$`a`$, $`c`$, we plot the characteristic velocity $`v_{cl}`$ for clouds of column density $`N`$(H) $`=10^{20}`$ and 10<sup>21</sup> cm<sup>-2</sup>. We assume a constant hot wind velocity $`v_{hot}=600`$ km s<sup>-1</sup> (or $`2\times 10^7`$ K) to be consistent with the highest velocities we measure, a cloud initial radius of 1 kpc, $`C_\mathrm{\Omega }=0.4`$, $`(dM/dt_{hot})/SFR=0.33`$ (Leitherer et al., 1999), and a mean particle mass $`\mu =1.4`$. Note that $`v_{cl}`$ is set to $`v_{hot}`$ if it rises above it. We see that $`\mathrm{\Delta }v_{max}`$ is less than or equal to $`v_{cl}`$ for $`N`$(H) $`=10^{20}`$ cm<sup>-2</sup>, consistent with the idea that the clouds are accelerated by the hot wind until they reach a velocity near $`v_{cl}`$. The range of measured $`N`$(H) roughly matches that predicted by the $`v_{cl}`$ lines.
To better incorporate column density and circular velocity information, we compute the expected characteristic and maximum velocities in this model for each galaxy. We plot the histogram of the differences between expected and observed velocities in Figure 19. We find that the agreement between the model and observations is in general poor. This is best indicated by the large dispersion ($`200300`$ km s<sup>-1</sup>) in the distributions. Furthermore, in one-third to one-half of the cases the observed velocity is larger than the predicted velocity. We thus conclude that we are missing important galaxy or wind properties that determine the cloud velocities, such as knowledge of the hot wind temperature in individual galaxies or the cloud initial radius.
In Figure 18$`b`$, $`d`$, we plot lines for the case that an optically thick shell of gas is driven by radiation pressure. In this case, the outflow velocity is linearly proportional to the galaxy’s circular velocity, for a constant ratio of the galaxy’s luminosity to the critical luminosity for driving of an outflow (eq. of Murray et al. 2005). In most cases, the measured values of $`\mathrm{\Delta }v_{max}`$ fall between the lines for $`L/L_{crit}=1.05`$ and 2.0, assuming that the observed clouds are at 10 times the initial radius.
The consequences of these results for distinguishing between ram-pressure-driven and radiation-pressure-driven winds are ambiguous. The ram pressure model clearly needs modification before it can be matched to the data. The radiation pressure model may fit the data, but it is underconstrained.
#### 4.4.2 Mass, Momenta, and Energy
In §3.3, we discuss the positive correlations we observe between outflow properties and galaxy properties. Here, we address the interpretation of these correlations.
In Figure 20, we plot the mass, momentum, and energy injection rates ($`dM_{hot}/dt`$, $`dp_{hot}/dt`$, and $`dE_{hot}/dt`$) from stellar winds and supernovae as dashed lines. We assume that these scale linearly with SFR, and we take the normalizations from the stellar synthesis models of Leitherer et al. (1999). These normalizations assume a continuous starburst of age $`40`$ Myr or more, a Salpeter IMF with mass limits 1 and 100 $`M_{\mathrm{}}`$, and twice solar metallicity (to match the average metallicity of our sample as computed in Paper I). The dotted line in the plot of $`dp/dt`$ vs. SFR assumes optically thick radiation pressure driving over the same timescale.
The approximately linear proportionality of $`M`$, $`dM/dt`$, $`p`$, and $`dp/dt`$ on SFR (Table 3) is consistent with the neutral gas being driven by the hot wind fluid ejected by the starburst if the supernova rate scales with SFR. The steeper dependence of $`E`$ and $`dE/dt`$ on SFR implies that galaxies with SFR $``$ 10$``$100 M yr<sup>-1</sup> thermalize the kinetic energy of their supernovae more efficiently than dwarf starbursts and/or accelerate the cold clouds in the wind more efficiently. The latter requires that there be fractionally more energy in other wind phases in the dwarf galaxies, since a smaller fraction of the available energy is transferred to the neutral gas clouds. If our assumptions about the wind geometry are correct, then the increase with increasing SFR of thermalization efficiency and/or fraction of energy in the cold wind is $``$20 on average over two orders of magnitude in SFR.
The flattening of these relationships at the highest star formation rates (SFR $`10100`$ M yr<sup>-1</sup>) implies some ‘saturation’ effect. The simplest explanation is that there is not enough interstellar mass in the galaxies with the highest SFR to keep up with the trend (see the plot of $`M`$ vs. SFR in Figure 10). The outflowing gas masses in ULIRGs are $``$10<sup>9</sup> $`M_{\mathrm{}}`$, which is already $``$1% of the dynamical mass of a ULIRG (Tacconi et al., 2002). Note that the maximum wind velocity also saturates at $``$600 km s<sup>-1</sup> (Figure 6). There is more energy and momentum available in the hot wind at high SFR. However, it doesn’t move into the colder gas, either because there isn’t enough gas to entrain or because the clouds can’t be accelerated above a certain velocity (Murray et al., 2005; Martin, 2005). Finally, the thermalization efficiency of the supernova kinetic energy may be smaller in the most luminous galaxies than in galaxies with moderate luminosity (by a factor of 10 on average). This could result from the massive amounts of molecular gas in ULIRGs (Sanders et al., 1991; Solomon et al., 1997), which may absorb energy and momentum and inhibit the outflow. The kinematics of this molecular phase may not be probed by our observations, and some of its energy will be radiated away due to its high density.
The strong dependence of these quantities on galactic mass is striking. The increase of mass, momentum, and energy with $`v_c`$ implies that, contrary to expectations, the large potentials of massive galaxies do not inhibit the formation of strong, energetic winds (see also §3.1). The fact that higher-mass galaxies have much larger and more energetic winds is related to the greater amount of star-forming and radiative energy available in massive galaxies, as well as the greater amount of ambient gas available for entrainment. The flattening at the highest masses, as we discuss above, may imply a removal of most of the ambient gas in the way of the wind, and is also consistent with a velocity ceiling or a decrease in thermalization efficiency at the highest masses.
How do the magnitude of the predictions of the hot wind model compare with the data? Comparing the mass in the hot and cold gas (using $`dM/dt`$), we see that the relative amount of entrained gas ($`dM/dt_{cold}/dM/dt_{hot}`$) is less than unity ($``$0.5) on average. However, the range is quite large, from 0.001 to 10. The ratio of gas mass in cold and hot gas is not typically well-known for even nearby galaxies; we show here that the range of values is probably large. The momentum in the cold gas is always smaller or equal to the available momentum in the hot wind fluid, consistent with conservation of momentum in the ram pressure driven model. Some galaxies appear to have most of their momentum in the cold, entrained gas. However, on average the amount of momentum in the entrained gas is a factor of 10 less than the momentum injected into the hot wind fluid. The range of $`p_{cold}/p_{hot}`$ is $``$0.001$``$1. The theoretical plots of energy and $`dE/dt`$ vs. SFR show that the observed values are 1% or less on average than the injected values (though again the range is large, from 0.01% to almost 100%). This implies that the hot gas must carry a substantial fraction of the wind’s energy and/or that the thermalization efficiency is low ($``$10% or less).
Silk (2003) argues that the mass outflow rate of entrained gas may be proportional to the porosity of the ISM. The porosity $`Q`$ is related to the filling factor of hot gas in the ISM by $`f_{hot}1e^Q`$. In this prescription, $`dM/dt_{cold}=\beta f_{hot}\times \mathrm{SFR}`$, where $`\beta `$ represents the amount of entrainment. We find that $`dM/dt_{cold}=0.2\mathrm{SFR}^{1.1\pm 0.1}`$ (see Figure 11 and Table 3), consistent with this interpretation. Since we find that $`dM/dt_{cold}0.5dM/dt_{hot}`$ on average, $`\beta 0.5`$. In this model, $`f_{hot}0.4`$, which means that the volume filled with hot gas is around 40% of the total. The porosity is thus $`Q0.5`$. Note, however, that $`\beta `$ is subject to considerable uncertainty, and this result should be considered tentative.
How does our data distinguish between driving of the wind by a hot wind versus radiation pressure? Figure 20 shows the expected momentum injection rates from radiation pressure as a function of SFR. These have the same linear dependence on SFR as the injection rates from stellar winds and supernovae, but the amount of momentum in radiation pressure is lower by a factor $``$10. Thus, the winds with the highest momenta per unit SFR need some other source of momentum besides radiation pressure. Winds with lower momenta may have some contribution to $`p`$ and $`dp/dt`$ from radiation pressure, however.
#### 4.4.3 Mass Entrainment Efficiency
The ‘mass entrainment efficiency’ is an important parameter as input to simple prescriptions in cosmological simulations, and is typically assumed to be of order unity (e.g., Kauffmann & Charlot, 1998; Aguirre et al., 2001a, b). Our results show that, on average, $`\eta `$ is up to an order of magnitude smaller than unity in IRGs and ULIRGs ($`0.10.3`$). However, it is comparable to the ‘reheating efficiency’ of the galaxy (equal to $`dM/dt_{hot}`$ / SFR), which is $``$0.33 for a starburst of age $``$40 Myr and twice solar metallicity with a Salpeter IMF (mass limits 1 and 100 M yr<sup>-1</sup>; Leitherer et al. 1999). The dispersion in $`\eta `$ is large, ranging from 0.001 to 10.
This quantity is independent of galactic mass, as shown by our data. However, there is some evidence that it is not constant with star formation rate or $`K`$-band magnitude (Figure 16$`a`$, $`b`$) for SFR $``$10 M yr<sup>-1</sup> and $`M_K<24`$, but decreases as $`\eta `$ SFR<sup>-0.5±0.2</sup> and $`\eta M_K^{0.4\pm 0.1}`$. (Note that the dependence on SFR is not a 3$`\sigma `$ result and thus does not appear in Table 3.) This decrease in $`\eta `$ may be due to the exhaustion of the gas reservoirs available for entrainment, the saturation of velocity at high SFR/luminosity, or a decrease in thermalization efficiency at high SFR/luminosity (§4.4.2).
The normalization of $`\eta `$ is uncertain, due to possible errors in wind geometry and local physical conditions (§2.6). Changes in our assumptions act in different directions, however. Assuming a thick instead of thin wind decreases $`\eta `$, while increasing the ionization fraction increase it. If our values are correct, theorists typically overestimate the mass entrainment efficiency(e.g., Kauffmann & Charlot, 1998; Aguirre et al., 2001a, b; Silk, 2003). Furthermore, cosmological simulations need to account for the large dispersion in $`\eta `$ (four orders of magnitude!) and the variation of $`\eta `$ with SFR that we observe, rather than assuming a single value.
### 4.5 Superwinds in Mergers
It has been postulated that superwinds play a role in the evolution of gas-rich mergers. ULIRGs may evolve into quasars when a buried AGN turns on and breaks the obscuring screen of dust (e.g., Sanders et al., 1988; Veilleux et al., 2002). Given their high frequency of occurrence in ULIRGs, outflows could easily play a role in redistributing dust and gas, thereby increasing the escape fraction of nuclear continuum light. If most ULIRGs do evolve into ellipticals, this gas redistribution may also destroy the central gas density spikes predicted in numerical simulations of mergers (Mihos & Hernquist, 1994). If left in place, these spikes would evolve into sharp upturns in the stellar surface brightness distribution in evolved ellipticals at small radii; breaks like this are not typical of elliptical surface brightness profiles (Hibbard & Yun, 1999). The magnitudes of $`dM/dt`$ that we measure ($``$100 M yr<sup>-1</sup> on average) would be able to evacuate $`10^{10}`$ $`M_{\mathrm{}}`$ of gas if they operate over $``$100 Myr (comparable to or greater than the gas consumption timescale in ULIRGs). Roughly $`10^{10}`$ $`M_{\mathrm{}}`$ of molecular gas are observed in the centers of ULIRGs (e.g., Sanders et al., 1991; Solomon et al., 1997; Downes & Solomon, 1998; Hibbard & Yun, 1999).
Many of the best-studied and clearest examples of superwinds in the local universe are found in starbursting disk galaxies with $`L_{\mathrm{IR}}`$ $`<10^{11}`$ $`L_{\mathrm{}}`$. Given that many LIRGs, and most ULIRGs, are mergers in which the morphology and kinematics of the galaxy are highly disturbed (e.g., Kim et al., 2002; Arribas et al., 2004; Ishida, 2004), one might suppose that the physical picture of a symmetric, bipolar superwind along the galaxy’s minor axis would not apply. However, there is evidence from observations of resolved, infrared-luminous merging galaxies that these galaxies produce ordered superwind structures analogous to those in quiescent disk galaxies. Examples include (in order of increasing $`L_{\mathrm{IR}}`$): NGC 520 (Hibbard & van Gorkom 1996; Hibbard et al. 2000), Arp 299 (Heckman et al., 1999; Hibbard et al., 2000), NGC 6240 (Heckman et al., 1987; Veilleux et al., 2003; Gerssen et al., 2004), and Arp 220 (Heckman et al., 1987, 1996; Hibbard et al., 2000; Arribas et al., 2001; McDowell et al., 2003). These galaxies all show spectacular signs of merging, including tidal tails and bridges and multiple nuclei. NGC 520, Arp 299, and Arp 220 possess large-scale rotating H I disks, and the H I data for these galaxies show gaps along the minor axis and H I-poor regions in the stellar tidal tails; this gas may have been evacuated by a superwind (Hibbard et al., 2000). Other evidence for superwinds comes from emission-line and X-ray data, which show minor-axis extensions, bubble-like and bow shock structures, kinematic broadening and/or line-splitting, shock-like emission line ratios, and/or tight correlations between optical and X-ray emitting filaments. Most recently, diffuse and extended X-ray-emitting gas has been discovered in perhaps the best-studied merger, the Antennae; this gas may have a superwind origin (Fabbiano et al., 2004).
However, the large opening angles we infer for ULIRG winds ($`C_\mathrm{\Omega }0.7`$) are not obviously consistent with the picture of a biconical superwind. Alternatively, there may be multiple superwind episodes of this kind over a short period of time, which will fill out the area surrounding the galaxy.
### 4.6 Redshift Evolution
One of the original purposes of this study was to explore the properties of superwinds deeper into redshift space. Given that the number density of ULIRGs evolves strongly upward with increasing redshift (e.g., Kim & Sanders, 1998; Chapman et al., 2003; Cowie et al., 2004; Pérez-González et al., 2005), we expect that winds from ULIRGs had a strong impact on the intergalactic medium and galaxy evolution at $`z>1`$. However, this assumes that the properties of winds in ULIRGs do not change with $`z`$. Apart from our study, there are currently no observations of winds in ULIRGs outside of the local universe, except in a single hyperluminous infrared galaxy at high redshift (Smail et al., 2003).
We have partially accomplished this goal by observing a substantial number of galaxies at redshifts up to $`z=0.5`$. We do observe differences between our high-$`z`$ and low-$`z`$ subsamples. Notably, winds are less frequently observed in the high-$`z`$ ULIRGs (with a detection rate of $`46\pm 14\%`$ vs. $`80\pm 7\%`$ for the low-$`z`$ ULIRGs), and less efficiently entrain gas (with a lower median $`\eta `$ than the low-$`z`$ ULIRGs by a factor of 2). However, the high-$`z`$ galaxies have higher star formation rates (or equivalently, luminosities) than the low-$`z`$ ULIRGs, on average (389 M yr<sup>-1</sup> and 225 M yr<sup>-1</sup>, respectively). This is a selection effect, since the most distant ULIRGs observed by the Infrared Astronomical Satellite (IRAS) are necessarily those with the highest intrinsic luminosities.
The differences in mass entrainment efficiency between the low- and high-$`z`$ subsamples are primarily a result of the variation of $`\eta `$ with SFR at high star formation rates (§3.3), rather than redshift evolution. In other words, the average mass entrainment efficiency in our high-$`z`$ ULIRGs is lower simply because they follow a trend of $`\eta `$ vs. SFR that is independent of redshift (at least for $`z1`$).
The change in the rate of wind detection is less certainly attributable to SFR variations. It is quite possible that the wind frequency of occurrence drops above a certain star formation rate. Other factors may also play a role; for instance, more extended continuum light in a distant galaxy may leak into the slit and wash out the Na I D absorption line. The high-$`z`$ spectra generally have lower S/N than those of nearer galaxies, as well, diminishing our ability to detect Na I D. Finally, the global wind covering factor in the highest-luminosity starbursts may simply be lower (§4.2).
Assuming that the relationships between wind properties and galaxy properties, as well as the properties of ULIRG winds, remain constant with redshift, the cosmological impact of winds in ULIRGs depends largely on their density evolution. Studies to date (e.g., Kim & Sanders, 1998; Chapman et al., 2003; Cowie et al., 2004; Pérez-González et al., 2005) show that the number density of ULIRGs rises strongly with increasing redshift and that they host a large fraction of star formation at $`z>1`$ . Thus, we expect that winds in ULIRGs, since they are common, massive, and energetic, have the potential to strongly impact the intergalactic and intracluster medium at redshifts $`>`$1. Properly assessing their actual impact, however, will require a better understanding of how much of their gas escapes into the IGM/ICM.
The highest redshift winds (at $`z3`$) have higher velocities than those in galaxies of comparable SFR in the local universe (§2.2). However, their global covering factors may be similar (§4.2). The higher velocities may reflect a difference in the structure of these early star forming galaxies or their surroundings from those of today.
## 5 SUMMARY
We have surveyed 78 starburst-dominated infrared-luminous galaxies, the largest systematic study to date of superwinds at $`z3`$. Our primary goal has been to study the detection rate and properties of winds as a function of the properties of the winds’ host galaxies. This provides insights into the physics and properties of superwinds that studies of individual galaxies do not allow. Furthermore, by studying a large sample of ultra-luminous infrared galaxies (ULIRGs), we can quantify how superwinds affect galaxy evolution and the intergalactic medium at high redshift, where ULIRGs may host most of the star formation in the universe (Pérez-González et al., 2005).
(1) Detection Rate. We find that superwinds are present in almost all infrared-luminous starburst galaxies. Our detection rates are 43% and 70% for our IRG ($`L_{\mathrm{IR}}=10^{11.36}L_{\mathrm{}}`$) subsample and ULIRGs, respectively. However, these detection rates are lower limits to the actual frequency of occurrence of outflows in these galaxies and also to the opening angle of the outflows. Because wind opening angles in local disk galaxies (as a fraction of 4$`\pi `$) equal the detection rate in our IRG subsample, we assume that these winds are found in almost all of our sample and that the detection rate exclusively reflects the wind geometry. We thus compute that the global covering factor of neutral gas outflows is higher in ULIRGs ($`\mathrm{\Omega }/4\pi 0.3`$) than in LIRGs ($`\mathrm{\Omega }/4\pi 0.15`$). The opening angle of high-$`z`$ starbursts is uncertain, but may be close to several tenths of 4$`\pi `$, similar to the IRGs Adelberger et al. 2003, 2005.
At the highest star formation rates, the detection rate in ULIRGs decreases (from 80% to 46%). However, the overall trend is for detection rate to increase with SFR. Detection rate does not depend on optical or near-infrared luminosity, galactic mass, spectral type, or on the merger stage in ULIRGs.
Some of the absorption we detect is related to overlapping gas disks in a multiple system or to tidal debris; these components are removed from the formal analysis. In a handful of cases we can demonstrate that these explanations are more likely than the outflow hypothesis. However, most alternative explanations should produce a roughly symmetric velocity distribution or one skewed to the red, and the one that we observe in our data has a strong blueward asymmetry. The number of outflowing components that we can attribute to these alternative explanations is at most a mirror image of the distribution of red components, and thus much less than the total number of blueshifted components.
(2) Velocity. The maximum velocities for these winds range from 150 to 600 km s<sup>-1</sup> (and up to 1100 km s<sup>-1</sup> in a single case), with a median value of 350 km s<sup>-1</sup>. The distribution of velocities is not consistent with a simple constant-velocity model of projection effects. The velocity of the gas with the highest column density in each galaxy has a median value of 140 km s<sup>-1</sup>, less than half that of $`\mathrm{\Delta }v_{max}`$. The average projected velocities of ionized gas in local edge-on starbursts of SFR comparable to the IRGs ($``$170 km s<sup>-1</sup>; Lehnert & Heckman 1996) are higher than the average value of $`\mathrm{\Delta }v_{maxN}`$ in IRGs (100 km s<sup>-1</sup>). High-redshift starbursts of star formation rate similar to the IRGs have higher mean velocities than both the IRGs and the ULIRGs: $``$300 km s<sup>-1</sup>, vs. $``$200 km s<sup>-1</sup> (Adelberger et al., 2003; Shapley et al., 2003).
The maximum velocities we measure are often close to or slightly larger than the escape velocity of the galaxy. The escape fraction of neutral gas is thus non-zero when averaged over the entire sample: $`f_{esc}520\%`$. Furthermore, we often observe higher velocities in the ionized gas (as probed by emission lines), which has velocities exceeding 1000 km s<sup>-1</sup> in several galaxies.
We show that LINERs have higher median values of $`\mathrm{\Delta }v_{max}`$ and Doppler width $`b`$ (by $``$100 km s<sup>-1</sup>) than H II galaxies. Thus, the spectral classification for these systems, based on line ratios, may simply reflect high-velocity vs. low-velocity shocks. We also show that for ULIRGs, wind velocity may increase as the merger progresses (though the statistics are admittedly low).
The outflow velocities in starbursts rise slowly with star formation rate, luminosity, and galactic mass when we consider both dwarf starbursts (Schwartz & Martin, 2004) and luminous galaxies. However, in our sample alone, velocity is independent of galaxy properties. This is consistent with a limit to the terminal velocity of neutral gas clouds above an SFR of $`10100`$ M yr<sup>-1</sup> (Murray et al., 2005; Martin, 2005). However, detailed models of ram pressure driven clouds which incorporate this limit are not consistent with our data (Murray et al., 2005). Models of radiation pressure driven gas are presently underconstrained (Murray et al., 2005).
(3) Mass, Momentum, Energy. We find that mass, momentum, energy, and their outflow rates depend on star formation rate roughly linearly, though these relationships flatten at high SFR ($``$10$``$100 M yr<sup>-1</sup>). This dependence is consistent with ram pressure driving of a hot gas or radiation pressure driving, since both the supernova rate and the galaxy luminosity scale with SFR. Modulo the flattening at high SFR, the dependence of energy on star formation rate is SFR<sup>1.6</sup>, suggesting that galaxies with SFR $``$ 10$``$100 M yr<sup>-1</sup> have higher thermalization efficiency than lower-SFR galaxies, or accelerate the cold gas more effectively and thus have fractionally less energy in other wind phases.
These wind properties correlate also with on galaxy luminosity and mass. The relationships to galactic mass are quite strong, with power law slopes of $`1.52.5`$. Thus, the wind properties are sensitive to the galaxy mass. However, the winds get larger and more powerful with increasing mass, rather than smaller.
At high SFR, luminosity, and galactic mass, we observe a flattening in the increase of wind properties (velocity, mass, momentum, and energy) with galaxy properties. This flattening may be due to entrainment of all the available gas clouds, the existence of a terminal velocity in ULIRGs above which the wind cannot be accelerated, and/or a reduction in thermalization efficiency at high SFR (perhaps due to the interception of the wind by dense ambient material, like the massive quantities of molecular gas feeding the starburst).
The magnitudes of mass, momentum, energy, and their outflow rates in neutral gas are compared to those in the hot gas, as predicted by starburst models (Leitherer et al., 1999). We find that the range of entrainment factors in the wind, $`M_{cold}/M_{hot}`$, is $``$0.01$``$10, with an average of a few. The measured momenta are, on average, a factor of $``$10 less than, and rarely greater than, the momentum in the hot wind. They are sometimes greater than the momentum available from radiation pressure, however, suggesting that, in many galaxies, the hot wind is likely to dominate the driving of the wind rather than radiation pressure. The measured energies in the neutral gas are significantly less than the energy in the hot wind on average, by factors of $``$1$``$1000; thus, the hot wind must carry most of the wind’s energy and/or the thermalization efficiency must on average be small ($``$10%), especially in dwarf galaxies.
The mass entrainment efficiency, $`\eta (dM/dt)/`$SFR, is found to range from $`0.00110`$ but is roughly $`0.1`$ on average. These numbers, however, are sensitive to the wind’s geometry and the local physics. This quantity is independent of galaxy properties on average but declines with star formation rate at SFR $``$ 10$``$100 M yr<sup>-1</sup> and with $`M_K`$ at $`M_K<24`$. The large observed range of $`\eta `$ and the dependence on SFR/luminosity show that the prescriptions and assumptions employed in numerical simulations of galaxy formation are not completely correct. These prescriptions typically assume a constant $`\eta `$ of approximately unity.
(4) Outlook. There is evidence for starburst-driven superwinds in local infrared-luminous mergers such as NGC 6240. However, the complex kinematics of the ionized gas in these galaxies make it sometimes difficult to distinguish starburst-driven winds from tidal motions directly produced by the interaction. In this sense, our data provides the most unambiguous evidence to date for frequent, massive, and energetic outflows in these systems.
To interpret our data more completely, however, we require better knowledge of local merging systems. Hydrodynamic modeling of winds in evolving environments such as these would allow comparison of observational data to numerical models, as is possible in more quiescent systems. More sensitive observations to probe the morphology and kinematics of faint structures in these galaxies, and three-dimensional kinematic observations of multiple gas phases will also help to disentangle the motions of different components (e.g., tidal tails vs. expanding superbubbles) as has been done for Arp 220 (Arribas et al., 2001; McDowell et al., 2003; Colina et al., 2004).
A better understanding of the neutral phase of these outflows is also needed, especially its geometry and small-scale structure. We need to know how the clouds in these winds are distributed, and how they stand up under radiation, evaporation, and shock ablation. Understanding these better will enable us to more precisely measure the properties and fate of these clouds from global measurements.
Many ULIRGs have infrared luminosities which are dominated by dust-reprocessed radiation from an AGN, rather than a starburst. AGN also power outflows, and it would be instructive to compare the properties of outflows in starburst-dominated ULIRGs and AGN-dominated ULIRGs. In a forthcoming paper, we present Na I D observations of 26 ULIRGs with Seyfert nuclei and perform this comparison (Rupke et al., 2005b).
We thank Kalliopi Dasyra and Dong Chan Kim for supplying useful data prior to publication. DSR is supported by NSF/CAREER grant AST-9874973. SV is grateful for partial support of this research by a Cottrell Scholarship awarded by the Research Corporation, NASA/LTSA grant NAG 56547, and NSF/CAREER grant AST-9874973. This research has made use of the NASA/IPAC Extragalactic Database (NED), which is operated by the JPL, Caltech, under contract with the NASA. It also makes use of data products from the Two Micron All Sky Survey, which is a joint project of the University of Massachusetts and IPAC/Caltech, funded by the NASA and the NSF. The authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of Mauna Kea has always had within the indigenous Hawaiian community. We are most fortunate to have the opportunity to conduct observations from this mountain.
|
warning/0506/cond-mat0506325.html
|
ar5iv
|
text
|
# Critical dynamics of diluted relaxational models coupled to a conserved density (diluted model C)
## Abstract
We consider the influence of quenched disorder on the relaxational critical dynamics of a system characterized by a non-conserved order parameter coupled to the diffusive dynamics of a conserved scalar density (model C). Disorder leads to model A critical dynamics in the asymptotics, however it is the effective critical behavior which is often observed in experiments and in computer simulations and this is described by the full set of dynamical equations of diluted model C. Indeed different scenarios of effective critical behavior are predicted.
Pacs numbers: 05.70.Jk; 64.60.Ht; 64.60.Ak
The critical behavior of pure systems might be changed by introducing imperfections like dilution, defects, etc. into a critical system. If such a change can be expected is answered by the Harris criterion Harris74 stating that a new diluted critical behavior appears if the specific heat of the pure system is diverging. The diluted critical behavior then has a nondiverging specific heat. Since the borderline value $`n_c`$ between a diverging and nondiverging specific heat at space dimensions $`d=3`$ lies between order parameter (OP) dimensions $`n=1`$ (Ising model) and $`n=2`$ (XY model) only the Ising case belongs to a new universality class. In consequence this result led to the conclusion that for the critical dynamics the coupling of conserved quantities to the OP is of no relevance Krey77 ; Lawrie84 . The argument was the following: For the critical dynamics of a relaxational model it was shown hahoma74 ; FoMo03 ; Folk04 that the coupling to a conserved density (e.g. the energy density) is relevant if the specific heat diverges. Due to dilution this is never the case and therefore the coupling is of no relevance. Therefore most of the papers considered only the relaxational dynamics of Ising systems Grinstein77 ; Prudnikov92 ; Oerding95 ; Janssen95
However this argumentation is based on the asymptotic properties of the diluted model. Experimental data and computer simulations made clear that in most cases one observes non-asymptotic critical behavior, described often by dilution dependent effective exponents (see e.g. review ; perumal03 ). In such a case the Harris criterion does not hold and therefore one has to consider in the dynamics the coupling to the conserved density and its effects on the effective critical behavior. In addition one is not restricted to the Ising case since already in statics the effective critical behavior for $`n>1`$ is different from the pure case perumal03 .
There are two relevant parameters of model C: (i) the static coupling $`\gamma `$ of the OP to the conserved density and (ii) a dynamic parameter, the time scale ratio $`w=\mathrm{\Gamma }/\lambda `$ where $`\mathrm{\Gamma }`$ is the relaxation rate of the OP and $`\lambda `$ is the diffusion rate of the conserved density. From the renormalization group (RG) treatment of model C one knows that the one loop order does not give reliable results due to the stability of a fixed point with the time scale ratio $`w=\mathrm{}`$. In two loop (and higher) order it turns out that this fixed point is unstable and model C is characterized by strong and weak scaling regions for the dynamics at $`d=3`$ FoMo03 ; Folk04 . Moreover it was shown that non-asymptotic effects are already present in model C Folk04 . In the following we will consider how these aspects are influenced by disorder.
Model C hahoma74 ; FoMo03 describes the relaxational dynamics of a system characterized by an $`n`$-component nonconserved OP $`\stackrel{}{\phi }_0(x,t)`$ coupled to the diffusive dynamics of a conserved scalar density $`m_0(x,t)`$. The structure of the equations of motions is not changed by the presence of disorder. They read:
$$\frac{\stackrel{}{\phi }_0}{t}=\stackrel{̊}{\mathrm{\Gamma }}\frac{}{\stackrel{}{\phi }_0}+\stackrel{}{\theta }_\phi ,\frac{m_0}{t}=\stackrel{̊}{\lambda }^2\frac{}{m_0}+\theta _m$$
(1)
where <sub>0</sub> or $`\stackrel{̊}{}`$ denote unrenormalized quantities. The stochastic forces in (1) satisfy the Einstein relations:
$`<\theta _{\phi _i}(x,t)\theta _{\phi _j}(x^{},t^{})>`$ $`=`$ $`2\stackrel{̊}{\mathrm{\Gamma }}\delta (xx^{})\delta (tt^{})\delta _{ij},`$ (2)
$`<\theta _m(x,t)\theta _m(x^{},t^{})>`$ $`=`$ $`2\stackrel{̊}{\lambda }^2\delta (xx^{})\delta (tt^{}).`$ (3)
Equilibrium is described by the static functional $``$ of the disordered magnetic system
$``$ $`=`$ $`{\displaystyle }d^dx\{{\displaystyle \frac{1}{2}}\stackrel{̊}{\stackrel{~}{r}}|\stackrel{}{\phi }_0|^2+V(x)|\stackrel{}{\phi }_0|^2+{\displaystyle \frac{1}{2}}{\displaystyle \underset{i=1}{\overset{n}{}}}(\phi _{i,0})^2`$
$`+`$ $`{\displaystyle \frac{\stackrel{̊}{\stackrel{~}{u}}}{4!}}|\stackrel{}{\phi }_0|^4+{\displaystyle \frac{1}{2}}a_mm_{0}^{}{}_{}{}^{2}+{\displaystyle \frac{1}{2}}\stackrel{̊}{\gamma }m_0|\stackrel{}{\phi }_0|^2\stackrel{̊}{h}_mm_0\},`$
where $`V(x)`$ is an impurity potential which introduces disorder to the system, $`d`$ is the spatial dimension. It contains a coupling $`\stackrel{̊}{\gamma }`$ to the secondary density which can be integrated out. Thus static critical properties described by the functional (Critical dynamics of diluted relaxational models coupled to a conserved density (diluted model C)) are equivalent to those of the functional $`=d^dx\left\{\frac{1}{2}\stackrel{̊}{r}|\stackrel{}{\phi }_0|^2+V(x)|\stackrel{}{\phi }_0|^2+\frac{1}{2}_{i=1}^n(\phi _{i,0})^2+\frac{\stackrel{̊}{u}}{4!}|\stackrel{}{\phi }_0|^4\right\}`$. The parameters $`\stackrel{̊}{r}`$ and $`\stackrel{̊}{u}`$ are related to $`\stackrel{̊}{\stackrel{~}{r}}`$, $`\stackrel{̊}{\stackrel{~}{u}}`$, $`a_m`$, $`\stackrel{̊}{\gamma }`$ and $`\stackrel{̊}{h}_m`$ by $`\stackrel{̊}{r}=\stackrel{̊}{\stackrel{~}{r}}+\stackrel{̊}{\gamma }\stackrel{̊}{h}_m/a_m`$ and $`\stackrel{̊}{u}=\stackrel{̊}{\stackrel{~}{u}}3\stackrel{̊}{\gamma }^2/a_m`$.
$`\stackrel{̊}{r}`$ is proportional to the temperature distance from the mean field critical temperature, $`\stackrel{̊}{u}`$ is positive. The properties of the random potential $`V(x)`$ are governed by a Gaussian distribution with width $`\stackrel{̊}{\mathrm{\Delta }}`$ ($`<<V(x)V(x^{})>>=4\stackrel{̊}{\mathrm{\Delta }}\delta (xx^{})`$, the double angular brackets means averaging over disorder). If $`\stackrel{̊}{\gamma }0`$ equations (1) describes dynamical properties of a purely relaxational model (model A) in the presence of disorder Lawrie84 .
We treat the critical dynamics of the disordered models within the field theoretical RG method Bausch76 , where the appropriate Lagrangians of the models are studied. The average over the random potential generates new terms in the Lagrangians with coupling $`\stackrel{̊}{\mathrm{\Delta }}`$, which correspond to the static coupling terms in $``$ generated by the disorder. The renormalization of the Lagrangian leads to the RG functions, describing the critical dynamics of our models. We use the minimal subtraction scheme with dimensional regularization to calculate these functions.
For renormalization of the OP $`\stackrel{}{\phi }_0`$, fourth-order couplings $`\stackrel{̊}{u},\stackrel{̊}{\mathrm{\Delta }}`$ and correlation functions with $`\stackrel{}{\phi }_0^2`$ insertion we introduce renormalization $`Z`$-factors as $`\stackrel{}{\phi }_0=Z_\phi ^{1/2}\stackrel{}{\phi }`$, $`\stackrel{̊}{u}=\mu ^ϵZ_\phi ^2Z_uuA_d^1`$, $`\stackrel{̊}{\mathrm{\Delta }}=\mu ^ϵZ_\phi ^2Z_\mathrm{\Delta }\mathrm{\Delta }A_d^1`$, and $`|\stackrel{}{\phi }_0|^2=Z_{\phi ^2}|\stackrel{}{\phi }|^2`$ ($`\mu `$ is the scale, $`ϵ=4d`$ and $`A_d`$ is a geometric factor). Within dynamics renormalization factor for the OP kinetic coefficient $`\stackrel{̊}{\mathrm{\Gamma }}=Z_\mathrm{\Gamma }\mathrm{\Gamma }`$ is introduced. The $`Z`$-factors introduced so far are enough to renormalize the diluted model A. For model C one needs to introduce additional
renormalization factors. The secondary density $`m_0`$ and coupling parameter $`\stackrel{̊}{\gamma }`$ are renormalized by $`a_m^{1/2}m_0=Z_mm`$ and $`a_m^{1/2}\stackrel{̊}{\gamma }=\mu ^{ϵ/2}Z_{\phi ^2}Z_m\gamma A_d^{1/2}`$ with $`Z_m^2(u,\mathrm{\Delta },\gamma )=1+\gamma ^2A_{\phi ^2}(u,\mathrm{\Delta })`$. The kinetic coefficient $`\lambda `$ renormalizes as $`a_m\stackrel{̊}{\lambda }=Z_m^2\lambda `$.
Defining the $`\zeta `$-functions as $`d\mathrm{ln}Z^1/d\mathrm{ln}\mu `$, where $`Z`$ represents any renormalization factor, one obtains the flow equations for the renormalized static and dynamic parameters. The flow equations for $`u`$ and $`\mathrm{\Delta }`$ decouple from the remaining parameters and are equal to expressions obtained for any $`n`$ in the diluted Ginsburg-Landau-Wilson (GLW) model statics . For the additional static parameter $`\gamma `$ appearing in diluted model C we have
$$l\frac{d\gamma }{dl}=\gamma \left[\frac{ϵ}{2}+\zeta _{\phi ^2}(u,\mathrm{\Delta })+\frac{1}{2}\gamma ^2B_{\phi ^2}(u,\mathrm{\Delta })\right].$$
(5)
The flow parameter $`l`$ is related to the reduced temperature and is consequently a measure for the distance to the critical temperature. The function $`\zeta _{\phi ^2}(u,\mathrm{\Delta })`$ is known from statics in the diluted model statics . The function $`B_{\phi ^2}(u,\mathrm{\Delta })`$, which is defined by the additive renormalization of the specific heat within the GLW-model, in the diluted case reads $`B_{\phi ^2}(u,\mathrm{\Delta })=n/2+𝒪(u^2,\mathrm{\Delta }^2,u\mathrm{\Delta })`$.
The flow equation for the time scale ratio (we introduce $`\rho =w/(1+w)`$ instead) is
$$l\frac{d\rho }{dl}=\rho (1\rho )\left[\zeta _\mathrm{\Gamma }(u,\mathrm{\Delta },\gamma ,\rho )\gamma ^2B_{\phi ^2}(u,\mathrm{\Delta })\right],$$
(6)
where the dynamic $`\zeta `$-function $`\zeta _\mathrm{\Gamma }`$ in two loop order reads
$`\zeta _\mathrm{\Gamma }(u,\mathrm{\Delta },\gamma ,\rho )=\zeta _\mathrm{\Gamma }^{(C)}(u,\gamma ,\rho )+4\mathrm{\Delta }{\displaystyle \frac{n+2}{3}}u\mathrm{\Delta }+20\mathrm{\Delta }^2`$
$`+2\mathrm{\Delta }\rho \gamma ^2\left[3\left(1\mathrm{ln}(1\rho )\right)+\rho \mathrm{ln}{\displaystyle \frac{\rho }{1\rho }}{\displaystyle \frac{\rho }{1\rho }}\mathrm{ln}\rho \right].`$
The corresponding explicit two loop expression for the $`\zeta `$-function of model C, $`\zeta _\mathrm{\Gamma }^{(C)}(u,\gamma ,\rho )=\zeta _w^{(C)}(u,\gamma ,\rho )+n\gamma ^2/2`$, is given in Folk04 (see $`\zeta _w`$ in Eq. (50) there). In these terms also the pure model A terms are included, which taken together with the last three terms in the first line of Eq. (Critical dynamics of diluted relaxational models coupled to a conserved density (diluted model C)) recover the $`\zeta `$-function for the diluted model A Lawrie84 ; Prudnikov92 . The zeros of the right hand sides of Eqs.(5), (6), and the corresponding equations for $`u`$ and $`\mathrm{\Delta }`$ give the possible fixed points.
It turns out that for all values of $`n`$ the stable fixed point value $`\gamma ^{}=0`$, and the well-known asymptotic static results are reproduced. Thus the flow in the space of the static couplings $`u`$, $`\mathrm{\Delta }`$ and $`\gamma `$ for $`n<n_c<2`$ looks like the flow for $`n=1`$ (see Fig. 1) whereas for $`n>n_c`$ it looks like the flow for $`n=3`$ (see Fig. 2). The fixed points for $`n=1`$ are indicated in Tab. 1. Only the mixed fixed point (M in Fig.1) is stable. However depending on the initial conditions a rich crossover behavior is observed. The same is true for $`n=3`$ although now the pure fixed point (P in Fig. 2) is stable.
The stable fixed points are found by calculating the stability matrix and its eigenvalues $`\omega _i`$ ($`i`$ represents $`u`$, $`\mathrm{\Delta }`$, $`\gamma `$ or $`\rho `$) . They govern the ”velocity” of the flow near the fixed points. A small stability exponent at the stable fixed point indicates a slow approach of the asymptotic behavior. This is the case for all $`n`$ and can be seen from Tab.2 for $`n=1`$. It is the slow approach in the $`\gamma `$ direction which characterizes the static flow within model C (see the extremely small values $`\omega _\gamma =0.0018`$ for $`n=1`$ \[at the fixed point M in Tab.2\] and for $`n=3`$ still $`\omega _\gamma =0.1109`$). In addition to the small values of the static stability exponents near the fixed points one has also small values $`\omega _\rho `$ coming from the dynamic parameter $`\rho `$ (however only near unstable fixed points). Thus one expects depending on the initial values of static and dynamic parameters a complex behavior in the non-asymptotic region.
The effective dynamical exponent is of special interest here, it is found by inserting the solutions of the flow equations into the expression
$$z_{eff}(l)=2+\zeta _\mathrm{\Gamma }(u(l),\mathrm{\Delta }(l),\gamma (l),\rho (l)).$$
(8)
One reaches the universal asymptotic value $`z`$ when the flow comes very near the stable fixed point. One observes that the stable fixed point value of the time ratio is always zero, independently of the specific heat exponent value of the pure model. Consequently, the results for model A are recovered in the asymptotics (either in the diluted universality class for $`n<n_c`$ or in the pure model A universality class for $`n>n_c`$). This combines with the fact that even in the region where dilution changes the static critical behavior the stable fixed point value of $`\gamma ^{}`$ is always zero. In consequence the secondary density is for all $`n`$ asymptotically decoupled and has an asymptotic dynamical exponent $`z_m=2`$.
The non-asymptotic behavior is however quite different as can be seen from Figs. 3 and 4. Due to the static non-asymptotic behavior also the dynamics is dominated by different non-asymptotic effects. This can be seen by comparing different $`z_{eff}(l)`$ for different initial conditions. The curves a, b, and c in Fig. 1 and all curves in Fig. 2 reach large values of the coupling $`\gamma `$ and/or $`\mathrm{\Delta }`$ and this leads to the typical maximum in the effective exponents independent of the initial value of $`\rho `$ (for the statics see e.g. perumal03 ). However an additional fixed point P is present at $`n=1`$. This leads for curve d in Fig. 1 almost to a plateau of $`z_{eff}`$ at its value for the unstable fixed point P. This plateau is more pronounced when the flow comes nearer to P where it stays longer because of the small transient exponent $`\omega _\rho `$. For curve c both effects (the maximum and the effect of fixed point P) are combined leading to the minimum in $`z_{eff}`$.
Consider now the contributions to the effective dynamical critical exponent $`z_{eff}`$ of different origin: (i) from the terms already present in model A (dashed curve in Fig. 5), (ii) from terms present in pure model C only (short dashed curve) and (iii) and from terms present in the diluted model C only (dashed-short dashed curve). The above contributions may add up to almost the asymptotic value of the exponent although the parameters are far away form their asymptotic values. This is an important point since the appearance of an asymptotic value in one physical quantity does not mean that other quantities have also reached the asymptotics. This is due to the different dependence of physical quantities on the model parameters. Another special feature of the diluted model C is that already in one loop order one observes qualitatively the same behavior as in two loop order of course with changed values for the exponents and the borderline value $`n_c`$, which in one loop is at $`n_c=4`$.
In concluding we remark that contrary to the general belief the coupling of a conserved density to the order parameter is relevant for the calculation of the dynamical critical behavior of diluted systems since this coupling is important to describe non-asymptotic effects. These effects have been seen in the experiments on physical systems perumal03 ; EX as well as in Monte Carlo simulations MC .
We acknowledge support from the Fonds zur Förderung der wissenschaftlichen Forschung (project P16574)
|
warning/0506/hep-th0506081.html
|
ar5iv
|
text
|
# On Graviphoton F-terms of 𝒩=1 𝑆𝑈(𝑁) SYM with Fundamental Matter
## 1 Introduction
F-terms of four-dimensional supersymmetric gauge theories in supergravity and graviphoton backgrounds have attracted much attention in recent years. On the one hand they are related to certain exactly computable amplitudes of two gravitons and graviphotons. On the other hand they are computed by second quantized partition functions of topological strings , and have an interesting mathematical structure . Gravitational F-terms are directly related to the partition function of two-dimensional non-critical strings . Recently, gravitational F-terms have been related to the computation of certain $`𝒩=2`$ black hole partition functions .
In this paper we will consider the gravitational and graviphoton correlation functions in the context of four-dimensional $`𝒩=1`$ supersymmetric gauge theories. Dijkgraaf and Vafa suggested a matrix model description, where the gravitational F-terms can be computed by summing up the non-planar matrix diagrams . The assumption made is that the relevant fields are the glueball superfields $`S_i`$ and the F-terms are holomorphic couplings of the glueball superfields to gravity and the graviphoton. The DV matrix proposal has been proven diagrammatically in .
The gravitational and graviphoton F-terms of interest are of the form
$`\mathrm{\Gamma }_1`$ $`=`$ $`{\displaystyle \underset{g=0}{\overset{\mathrm{}}{}}}{\displaystyle d^4xd^2\theta (F_{\alpha \beta }F^{\alpha \beta })^gN_i\frac{F_g(S)}{S_i}},`$ (1)
$`\mathrm{\Gamma }_2`$ $`=`$ $`{\displaystyle \underset{g=1}{\overset{\mathrm{}}{}}}g{\displaystyle d^4xd^2\theta G^2(F_{\alpha \beta }F^{\alpha \beta })^{g1}F_g(S)},`$ (2)
where $`G_{\alpha \beta \gamma }`$ is the $`𝒩=1`$ Weyl superfield and $`F_{\alpha \beta }`$ is the graviphoton. According to the DV proposal, $`F_g(S_i)`$ is the partition function of the corresponding matrix model evaluated by summing the genus $`g`$ diagrams with $`S_i`$ being the ’t Hooft parameters.
The approach we will take is to use the generalized Konishi anomaly equations and some knowledge on the correlation functions to obtain information about the perturbative contribution to the correlation functions involving gravity and the graviphoton. In general, it is not clear in which cases the generalized Konishi anomaly equations are sufficient in order to determine the gravitational and graviphoton F-terms. We will study $`𝒩=1`$ $`SU(N_c)`$ supersymmetric gauge theory with fundamental matter and show that the anomaly equations are insufficient to obtain the full perturbative correlation functions (and the F-terms). We will also discuss the field theoretic perturbative diagrammatic computation.
Other recent works on the computation of gravitational and graviphoton correlation functions and F-terms are .
The paper is organized as follows. In section 2 we review the computational scheme. Then in section 3 we apply the scheme to the computation of correlation functions in $`SU(N_c)`$ SYM. Later we demonstrate the problems with the diagrammatic computation in section 4.
## 2 The computational scheme
Here we review the scheme used for computing the correlation functions in the presence of either graviphoton or gravity backgrounds.
### 2.1 The chiral ring
We consider here an $`𝒩=1`$ supersymmetric gauge theory with chiral matter multiplets coupled to it. Chiral operators are operators annihilated by the covariant derivative $`\overline{D}_{\dot{\alpha }}`$. All such operators modulo terms which are $`\overline{D}_{\dot{\alpha }}`$ exact form the ring structure of the chiral ring.
Denoting by $`W_\alpha =\frac{1}{4}\overline{D}^2e^VD_\alpha e^V`$ the spinor field-strength superfield of the vector superfield $`V`$ one has in flat space that in the chiral ring (i.e., up to $`\overline{D}_{\dot{\alpha }}`$ exact terms)
$$\{W_\alpha ,W_\beta \}=0.$$
(3)
This relation is modified in a background of gravity and the graviphoton field. Let $`G_{\alpha \beta \gamma }`$ be the $`𝒩=1`$ Weyl superfield and $`F_{\alpha \beta }`$ be the graviphoton field which together form the $`𝒩=2`$ Weyl superfield $`H_{\alpha \beta }=F_{\alpha \beta }+\widehat{\theta }^\gamma G_{\alpha \beta \gamma }`$, where $`\widehat{\theta }`$ is the additional supercoordinate of $`𝒩=2`$ superspace. In the presence of these either the supercoordinates become non-anti-commutative or (3) is modified to
$$\{W_\alpha ,W_\beta \}=F_{\alpha \beta }+2G_{\alpha \beta \gamma }W^\gamma .$$
(4)
This deformation of the chiral ring leads to the chiral ring relations
$`[W_\alpha ,W^2]`$ $`=`$ $`2F_{\alpha \beta }W^\beta ,`$ (5)
$`\{W_\alpha ,W^2\}`$ $`=`$ $`{\displaystyle \frac{2}{3}}(G^2W_\alpha +G_{\alpha \beta \gamma }F^{\beta \gamma }),`$ (6)
$`W_\alpha W^2`$ $`=`$ $`F_{\alpha \beta }W^\beta {\displaystyle \frac{1}{3}}G^2W_\alpha {\displaystyle \frac{1}{3}}G_{\alpha \beta \gamma }F^{\beta \gamma },`$ (7)
$`(G^2)^2`$ $`=`$ $`0.`$ (8)
In addition, for a chiral superfield in the fundamental or anti-fundamental representation
$$W_{\alpha a}{}_{}{}^{b}Q_{b}^{i}=\stackrel{~}{Q}_i^aW_{\alpha a}{}_{}{}^{b}=0$$
(9)
in the chiral ring.
### 2.2 The Konishi anomaly equations
The classical Konishi equations in the chiral ring for a field transformation $`\delta Q_a^i`$ are
$$\frac{W_{\mathrm{tree}}}{Q_a^i}\delta Q_a^i=0.$$
(10)
These are modified quantum mechanically to
$$\frac{W_{\mathrm{tree}}}{Q_a^i}\delta Q_a^i+\left(\frac{1}{32\pi ^2}W_{\alpha a}{}_{}{}^{b}W_{}^{\alpha }{}_{b}{}^{}{}_{}{}^{c}+\frac{1}{32\pi ^2}\frac{G^2}{3}\delta _a^c\right)\frac{\delta Q_c^i}{Q_a^i}=0.$$
(11)
As argued in , the Konishi anomaly equations are not modified in the presence of the graviphoton field. The argument is based on the graviphoton being of dimension three so any Lorentz scalar with smooth limits of the dimensional parameters of the theory constructed from it would have a dimension greater then three. All the terms in the Konishi anomaly equation are of dimension three, therefore the anomaly equation for a field transformation $`\delta Q_a^i`$ remains of the form (11) and is not modified.
In the next section we will use the Konishi anomaly equations in order to obtain perturbative information on the correlation functions. As will be seen, the Konishi equations are not sufficient to completely determine the correlations functions, but they do provide some constraints on their general form.
## 3 $`SU(N_c)`$ SYM with fundamental matter
In this section we consider the $`𝒩=1`$ $`SU(N_c)`$ SYM theory with chiral matter multiplets $`Q_a^i`$ and $`\stackrel{~}{Q}_i^a`$ ($`a,b,\mathrm{}=1,\mathrm{},N_c`$ are color indices and $`i,j,\mathrm{}=1,\mathrm{},N_f`$ are flavor indices) in the fundamental and anti-fundamental representation, respectively, considered in . The theory has the tree-level superpotential
$$W_{\mathrm{tree}}=m\mathrm{tr}M+\lambda \mathrm{tr}M^2,$$
(12)
where $`M_i{}_{}{}^{j}=\stackrel{~}{Q}_i^aQ_a^j`$ are the gauge-invariant meson operators. For simplicity we will take the case of a single flavor ($`N_f=1`$), but the results should be readily extendable to $`N_f>1`$.
### 3.1 The anomaly equations
We first look at the field transformation $`\delta Q_a=Q_aM^k`$. It yields the equations obtained in
$$mM^{k+1}+2\lambda M^{k+2}+\frac{N_c+k}{3}G^2M^k=SM^k.$$
(13)
(Here and henceforth we redefine the Weyl superfield $`G^2G^2/32\pi ^2`$.)
Since the graviphoton does not appear explicitly in the anomaly equation, the only way to obtain graviphoton dependence is via the chiral ring relations (5)–(7) or by including it in the transformation $`\delta Q_a`$. However, using Lorentz invariant graviphoton terms such as $`F_{\alpha \beta }F^{\alpha \beta }`$ in the transformation will only result in multiplying the entire anomaly equation by these expressions and not yield any new independent equations.
$`\delta Q`$ should be a scalar in the fundamental representation of the gauge group, so $`W_\alpha `$ can be incorporated into it only as $`W_{\alpha a}{}_{}{}^{b}W_{}^{\alpha }{}_{b}{}^{}{}_{}{}^{c}Q_{c}^{}`$, which vanishes in the chiral ring, or as $`S=\frac{1}{32\pi ^2}\mathrm{Tr}(W_\alpha W^\alpha )`$ which will only multiply the equation by $`S`$. Another possible scalar can be constructed using the graviphoton: $`F_{\alpha \beta }W^\alpha W^\beta `$. By using the symmetry of $`F_{\alpha \beta }`$ and (4) one has
$$F_{\alpha \beta }W^\alpha W^\beta =\frac{1}{2}F_{\alpha \beta }F^{\alpha \beta }+F_{\alpha \beta }G^{\alpha \beta }{}_{\gamma }{}^{}W_{}^{\gamma }.$$
(14)
This is in the adjoint representation so an appropriate term can be obtained by acting with it on $`Q`$ which vanishes in the chiral ring due to (9) or by tracing over the gauge indices, making the second term vanish as $`W_\alpha `$ is traceless.
In general, transformations yielding the graviphoton will either vanish because of (9) or will lead to dependent anomaly equations. This is unlike the case of the theory with matter in the adjoint representation of the gauge group since in that case transformations such as $`W_\alpha \mathrm{\Phi }`$ do not vanish and can combine with the $`W^2`$ term in the anomaly equation to generate coupling to the graviphoton by the chiral ring relations.
### 3.2 Correlation functions without gravity
#### 3.2.1 Solution of the anomaly equations
The generalized Konishi anomaly equations for the theory considered were found in the presence of gravity but no graviphoton backgrounds in and with gravity turned off are of the form
$$SM^k=mM^{k+1}+2\lambda M^{k+2}.$$
(15)
Since the graviphoton background does not modify the anomaly equations , these remain valid even with the graviphoton turned on.
By performing a $`z`$-transform of (15) one obtains the single equation
$$S\underset{k=0}{\overset{\mathrm{}}{}}\frac{1}{k!}M^kz^k=m\underset{k=0}{\overset{\mathrm{}}{}}\frac{1}{k!}M^{k+1}z^k+2\lambda \underset{k=0}{\overset{\mathrm{}}{}}\frac{1}{k!}M^{k+2}z^k.$$
(16)
We now define the generating function for the meson operator correlation functions
$$f(z)=\underset{k=0}{\overset{\mathrm{}}{}}\frac{1}{k!}M^kz^k.$$
(17)
It follows immediately that
$`{\displaystyle \frac{df(z)}{dz}}`$ $`=`$ $`{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{k!}}M^{k+1}z^k,`$ (18)
$`{\displaystyle \frac{d^2f(z)}{dz^2}}`$ $`=`$ $`{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{k!}}M^{k+2}z^k.`$ (19)
Hence, the infinite set of equations (15) can be written as the ordinary differential equation
$$Sf(z)=m\frac{df(z)}{dz}+2\lambda \frac{d^2f(z)}{dz^2},$$
(20)
whose general solution is
$$f(z)=A_+\mathrm{exp}\left(\frac{m+\sqrt{m^2+8\lambda S}}{4\lambda }z\right)+A_{}\mathrm{exp}\left(\frac{m\sqrt{m^2+8\lambda S}}{4\lambda }z\right),$$
(21)
where $`A_+`$ and $`A_{}`$ are coefficients which may depend on the couplings as well as the glueball superfield $`S`$ and the graviphoton. Particularly, note that any possible graviphoton dependence may enter through these coefficients alone.
Thus, we conclude that the correlation functions are of the form
$$M^k=A_+\left(\frac{m+\sqrt{m^2+8\lambda S}}{4\lambda }\right)^k+A_{}\left(\frac{m\sqrt{m^2+8\lambda S}}{4\lambda }\right)^k.$$
(22)
Because correlation functions of chiral operators factorize in the absence of the graviphoton and gravity, in the limit $`F_{\alpha \beta }0`$ either $`A_+1`$ and $`A_{}0`$ or the other way around depending on whether the Higgsed vacuum is considered or not. Hence, in the un-Higgsed vacuum
$$A_+=1+O(F_{\alpha \beta }F^{\alpha \beta }),A_{}=O(F_{\alpha \beta }F^{\alpha \beta })$$
(23)
and
$$A_+=O(F_{\alpha \beta }F^{\alpha \beta }),A_{}=1+O(F_{\alpha \beta }F^{\alpha \beta })$$
(24)
in the Higgsed vacuum.
#### 3.2.2 The form of $`A_\pm `$
As noted before, the graviphoton dependence can only enter via the coefficients $`A_\pm `$. In general, these should depend on the couplings $`m`$ and $`\lambda `$ and on the background fields $`S`$ and $`F_{\alpha \beta }`$. Hence, the $`A_\pm `$ should be sums of terms of the form
$$\lambda ^nm^pS^q(F_{\alpha \beta }F^{\alpha \beta })^r.$$
From holomorphicity we expect $`n`$, $`p`$, $`q`$ and $`r`$ to be integers. Also since the coefficients are dimensionless, the powers must satisfy
$$n+p+3q+6r=0.$$
(25)
The limit $`F_{\alpha \beta }0`$ must be regular so that the ordinary correlation functions without graviphoton background obtained in are recovered. Thus $`r0`$. The classical limit $`S0`$, in which the Konishi anomaly vanishes, must be smooth so $`q0`$. In the limit of $`\lambda 0`$ the correlation function $`M`$ is either smooth for the case of the un-Higgsed vacuum (the one corresponding to the plus sign solution) or diverges as $`1/\lambda `$ in the Higgsed vacuum (the minus sign vacuum). Thus, we do not expect additional, higher order divergence as $`\lambda 0`$ so $`n0`$.
Finally, the flavor symmetries $`U(1)_Q`$ and $`U(1)_{\stackrel{~}{Q}}`$ are broken at tree-level by the superpotential. These can be restored by assigning charges to the couplings as given in the table
| | $`U(1)_Q`$ | $`U(1)_{\stackrel{~}{Q}}`$ |
| --- | --- | --- |
| $`Q`$ | $`1`$ | $`0`$ |
| $`\stackrel{~}{Q}`$ | $`0`$ | $`1`$ |
| $`S`$ | $`0`$ | $`0`$ |
| $`m`$ | $`1`$ | $`1`$ |
| $`\lambda `$ | $`2`$ | $`2`$ |
| $`F_{\alpha \beta }`$ | $`0`$ | $`0`$ |
Requiring the correlation functions to be invariant under these restored symmetries yields
$$2n+p=0.$$
(26)
Putting all of this together we have
$$n=q+2r$$
(27)
and the terms in the power series expansion of $`A_\pm `$ are of the form
$$\left(\frac{\lambda }{m^2}\right)^{q+2r}(F_{\alpha \beta }F^{\alpha \beta })^rS^q.$$
(28)
It should be noted that only terms which vanish in the limit $`m\mathrm{}`$ are allowed — in accordance with one’s expectation that the matter completely decouples in this limit, leaving only the pure gauge theory coupled to gravity and to the graviphoton.
### 3.3 Correlation functions with gravity and graviphoton backgrounds
#### 3.3.1 Solution of the anomaly equations
By performing a $`z`$-transform on (13) the following equation is obtained
$`S{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{z^k}{k!}}M^k`$ $`=`$ $`m{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{z^k}{k!}}M^{k+1}+2\lambda {\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{z^k}{k!}}M^{k+2}+{\displaystyle \frac{N_c}{3}}G^2{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{z^k}{k!}}M^k+`$ (29)
$`+{\displaystyle \frac{1}{3}}G^2{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{kz^k}{k!}}M^k.`$
Using the generating function (17) one finds that
$$z\frac{df(z)}{dz}=\underset{k=0}{\overset{\mathrm{}}{}}\frac{kz^k}{k!}M^k$$
(30)
and together with the relations (18) and (19) (13) can be cast into the ordinary differential equation
$$Sf(z)=m\frac{df(z)}{dz}+2\lambda \frac{d^2f(z)}{dz^2}+\frac{N_c}{3}G^2f(z)+\frac{1}{3}G^2z\frac{df(z)}{dz}.$$
(31)
The function $`f(z)`$ can be expanded in powers of $`G^2`$ using the chiral ring relation $`(G^2)^2=0`$
$$f(z)=f_0(z)+G^2f_1(z)$$
(32)
and upon substitution in (31) we get two differential equations for $`f_0(z)`$ and $`f_1(z)`$:
$`Sf_0(z)`$ $`=`$ $`m{\displaystyle \frac{df_0(z)}{dz}}+2\lambda {\displaystyle \frac{d^2f_0(z)}{dz^2}},`$ (33)
$`Sf_1(z)`$ $`=`$ $`m{\displaystyle \frac{df_1(z)}{dz}}+2\lambda {\displaystyle \frac{d^2f_1(z)}{dz^2}}+{\displaystyle \frac{N_c}{3}}f_0(z)+{\displaystyle \frac{1}{3}}z{\displaystyle \frac{df_0(z)}{dz}}.`$ (34)
Defining
$$\alpha _\pm =\frac{m\pm \sqrt{m^2+8\lambda S}}{4\lambda },$$
(35)
the solution of the equation for $`f_0(z)`$ is (21)
$$f_0(z)=A_+e^{\alpha _+z}+A_{}e^{\alpha _{}z}.$$
(36)
Plugging this in (34) the following equation is obtained
$`Sf_1`$ $`=`$ $`m{\displaystyle \frac{df_1}{dz}}+2\lambda {\displaystyle \frac{d^2f_1}{dz^2}}+{\displaystyle \frac{N_c}{3}}\left(A_+e^{\alpha _+z}+A_{}e^{\alpha _{}z}\right)+`$ (37)
$`+{\displaystyle \frac{1}{3}}z\left(\alpha _+A_+e^{\alpha _+z}+\alpha _{}A_{}e^{\alpha _{}z}\right).`$
By considering a solution of the form
$$f_1(z)=c_+(z)e^{\alpha _+z}+c_{}(z)e^{\alpha _{}z}$$
(38)
and further assuming that the equation thus obtained can be divided into separate equations for the unknown functions $`c_\pm (z)`$ we have
$$2\lambda \frac{d^2c_\pm }{dz^2}+(m+4\lambda \alpha _\pm )\frac{dc_\pm }{dz}+(m\alpha _\pm +2\lambda \alpha _\pm ^2S)c_\pm +\frac{N_c}{3}A_\pm +\frac{1}{3}z\alpha _\pm A_\pm =0,$$
(39)
whose solution is given by
$`c_\pm (z)`$ $`=`$ $`\pm {\displaystyle \frac{C_{1\pm }}{\sqrt{m^2+8\lambda S}}}{\displaystyle \frac{N_c}{3}}A_\pm \left(\pm {\displaystyle \frac{z}{\sqrt{m^2+8\lambda S}}}{\displaystyle \frac{2\lambda }{m^2+8\lambda S}}\right)`$ (40)
$`{\displaystyle \frac{1}{6}}\alpha _\pm A_\pm \left[\pm {\displaystyle \frac{8\lambda ^2}{(m^2+8\lambda S)^{3/2}}}{\displaystyle \frac{4\lambda z}{m^2+8\lambda S}}\pm {\displaystyle \frac{z^2}{\sqrt{m^2+8\lambda S}}}\right]+`$
$`+C_{2\pm }\mathrm{exp}\left({\displaystyle \frac{\sqrt{m^2+8\lambda S}}{2\lambda }}z\right),`$
where $`C_{1\pm }`$ and $`C_{2\pm }`$ are integration constants which may depend on the couplings, the glueball superfield and the graviphoton.
#### 3.3.2 Constraints on the coefficients
As shown in , there are two types of related effective F-terms coupling the glueball to gravity and the graviphoton
$`\mathrm{\Gamma }_1`$ $`=`$ $`{\displaystyle d^4xd^2\theta W_0}={\displaystyle \underset{g=0}{\overset{\mathrm{}}{}}}{\displaystyle d^4xd^2\theta (F_{\alpha \beta }F^{\alpha \beta })^gN_i\frac{F_g(S)}{S_i}},`$ (41)
$`\mathrm{\Gamma }_2`$ $`=`$ $`{\displaystyle d^4xd^2\theta G^2W_1}={\displaystyle \underset{g=1}{\overset{\mathrm{}}{}}}g{\displaystyle d^4xd^2\theta G^2(F_{\alpha \beta }F^{\alpha \beta })^{g1}F_g(S)}.`$ (42)
Since the function $`F_g(S)`$ is found in both, in the case of unbroken gauge symmetry the two are related as
$$\frac{W_0}{u}=N_c\frac{W_1}{S},$$
(43)
where we have set $`u=F_{\alpha \beta }F^{\alpha \beta }`$.
The correlation functions $`M`$ and $`M^2`$ can be obtained from the effective superpotential $`W_{\mathrm{eff}}=W_0+W_1G^2`$ by differentiating it with respect to the couplings,
$$M=\frac{W_{\mathrm{eff}}}{m},M^2=\frac{W_{\mathrm{eff}}}{\lambda }.$$
(44)
Hence, $`f_0(z)`$ and $`f_1(z)`$ must satisfy the relations
$`{\displaystyle \frac{^2f_0}{uz}}|_{z=0}`$ $`=`$ $`N_c{\displaystyle \frac{^2f_1}{Sz}}|_{z=0},`$ (45)
$`{\displaystyle \frac{^3f_0}{uz^2}}|_{z=0}`$ $`=`$ $`N_c{\displaystyle \frac{^3f_1}{Sz^2}}|_{z=0}.`$ (46)
The constraint on the integration constants from the first of these is
$`0`$ $`=`$ $`{\displaystyle \frac{2N_c\lambda \left[(1+N_c)m^2+4(2N_c1)\lambda S2N_cm\sqrt{m^2+8\lambda S}\right]}{3(m^2+8\lambda S)^{5/2}}}A_{}`$ (47)
$`{\displaystyle \frac{2N_c\lambda \left[(1+N_c)m^2+4(2N_c1)\lambda S+2N_cm\sqrt{m^2+8\lambda S}\right]}{3(m^2+8\lambda S)^{5/2}}}A_++`$
$`+{\displaystyle \frac{N_cm(C_1C_{1+})}{(m^2+8\lambda S)^{3/2}}}+{\displaystyle \frac{N_c(C_{2+}C_2)}{\sqrt{m^2+8\lambda S}}}+`$
$`+{\displaystyle \frac{m+\sqrt{m^2+8\lambda S}}{4\lambda }}{\displaystyle \frac{A_+}{u}}+{\displaystyle \frac{m\sqrt{m^2+8\lambda S}}{4\lambda }}{\displaystyle \frac{A_{}}{u}}+`$
$`+{\displaystyle \frac{N_c\left[4\lambda S+N_c\left(m^2+8\lambda S+m\sqrt{m^2+8\lambda S}\right)\right]}{6(m^2+8\lambda S)^{3/2}}}{\displaystyle \frac{A_+}{S}}+`$
$`+{\displaystyle \frac{N_c\left[4\lambda SN_c\left(m^2+8\lambda Sm\sqrt{m^2+8\lambda S}\right)\right]}{6(m^2+8\lambda S)^{3/2}}}{\displaystyle \frac{A_{}}{S}}+`$
$`+{\displaystyle \frac{N_c(m\sqrt{m^2+8\lambda S})}{4\lambda \sqrt{m^2+8\lambda S}}}{\displaystyle \frac{C_{1+}}{S}}{\displaystyle \frac{N_c(m+\sqrt{m^2+8\lambda S})}{4\lambda \sqrt{m^2+8\lambda S}}}{\displaystyle \frac{C_1}{S}}+`$
$`+{\displaystyle \frac{N_c(m+\sqrt{m^2+8\lambda S})}{4\lambda }}{\displaystyle \frac{C_{2+}}{S}}+{\displaystyle \frac{N_c(m\sqrt{m^2+8\lambda S})}{4\lambda }}{\displaystyle \frac{C_2}{S}},`$
while the second one yields
$`0={\displaystyle \frac{N_cm\left[(1+2N_c)m(m+\sqrt{m^2+8\lambda S})+4(14N_c)\lambda S\right]A_{}}{6(m^2+8\lambda S)^{5/2}}}+`$ (48)
$`+{\displaystyle \frac{N_cm\left[(1+2N_c)m(m+\sqrt{m^2+8\lambda S})4(14N_c)\lambda S\right]A_+}{6(m^2+8\lambda S)^{5/2}}}+`$
$`+{\displaystyle \frac{2N_cS(C_1C_{1+})}{(m^2+8\lambda S)^{3/2}}}+{\displaystyle \frac{N_c(m\sqrt{m^2+8\lambda S})C_2}{2\lambda \sqrt{m^2+8\lambda S}}}{\displaystyle \frac{N_c(m+\sqrt{m^2+8\lambda S})C_{2+}}{2\lambda \sqrt{m^2+8\lambda S}}}+`$
$`+{\displaystyle \frac{(m+\sqrt{m^2+8\lambda S})^2}{16\lambda ^2}}{\displaystyle \frac{A_{}}{u}}+{\displaystyle \frac{(m+\sqrt{m^2+8\lambda S})^2}{16\lambda ^2}}{\displaystyle \frac{A_+}{u}}+`$
$`+{\displaystyle \frac{N_c\left[N_cm^2(m+\sqrt{m^2+8\lambda S})+2(4N_c1)m\lambda S+2(6N_c+1)\lambda S\sqrt{m^2+8\lambda S}\right]}{12\lambda (m^2+8\lambda S)^{3/2}}}{\displaystyle \frac{A_{}}{S}}`$
$`{\displaystyle \frac{N_c\left[N_cm^2(m\sqrt{m^2+8\lambda S})+2(4N_c1)m\lambda S2(6N_c+1)\lambda S\sqrt{m^2+8\lambda S}\right]}{12\lambda (m^2+8\lambda S)^{3/2}}}{\displaystyle \frac{A_+}{S}}+`$
$`+{\displaystyle \frac{N_c(m+\sqrt{m^2+8\lambda S})^2}{16\lambda ^2\sqrt{m^2+8\lambda S}}}{\displaystyle \frac{C_1}{S}}{\displaystyle \frac{N_c(m+\sqrt{m^2+8\lambda S})^2}{16\lambda ^2\sqrt{m^2+8\lambda S}}}{\displaystyle \frac{C_{1+}}{S}}`$
$`{\displaystyle \frac{N_c(m+\sqrt{m^2+8\lambda S})^2}{16\lambda ^2}}{\displaystyle \frac{C_2}{S}}{\displaystyle \frac{N_c(m+\sqrt{m^2+8\lambda S})^2}{16\lambda ^2}}{\displaystyle \frac{C_{2+}}{S}}.`$
## 4 The field theory graphs
The entire Lagrangian of the $`C`$-deformed field theory is not known. Hence it is not clear how to compute the correlation functions in presence of graviphoton background. In this section we make some assumptions about the Lagrangian and demonstrate the difficulties arising from these assumptions.
According to the Dijkgraaf–Vafa conjecture the perturbative expansion of the graviphoton correction terms should be obtainable by computing the non-planar graphs of the theory. In a scheme was given for computing such graphs for matter in the adjoint representation of the gauge group. This scheme does not appear to be applicable in the case of matter in the fundamental representation since the scheme includes the selection rule that a non-vanishing graph with $`h`$ holes should have gaugino insertions in $`h1`$ of its holes. This selection rule originated from the requirement that the graphs be path-ordering independent. Basically, for a graph to be path-ordering independent one must have
$$_{\gamma _i}p_\alpha =0$$
(49)
in that graph, where $`p_\alpha `$ is the world-sheet current of space-time supersymmetry and $`\gamma _i`$ is the contour of the $`i`$-th hole. This was enforced by inserting $`h1`$ gaugino insertions
$$\underset{i=1}{\overset{h1}{}}\left(_{\gamma _i}W^\alpha p_\alpha \right)^2$$
and utilizing the fact that
$$\underset{i=1}{\overset{h}{}}_{\gamma _i}p_\alpha =0.$$
(50)
However, (50) does not hold in the case of matter in the fundamental representation. This is most easily seen in the field theory limit, in which (50) takes the form
$$\underset{i=1}{\overset{h}{}}\underset{I}{}s_IL_{Ii}\pi _\alpha ^I=0,$$
(51)
where the index $`I`$ denotes the propagator and $`L_{Ii}`$ is a matrix relating index-loop momenta to the propagator momenta. Unlike the double-line propagators of matter in the adjoint representation, propagators of matter in the fundamental representation have only a single color line. Therefore, whereas in the adjoint case each propagator is traversed twice in opposite directions and thus contributes two identical terms with opposite signs to the sum, so (51) is satisfied, in the fundamental case it is traversed once and obviously the sum can no longer vanish. Note that in the string theory description, similar subtleties may be encountered due to the way fundamental matter is engineered geometrically via singular Calabi-Yau compactification, or D-brane wrappings.
Thus we are forced to take a different approach. If one assumes that the Lagrangian of the field theory is such that the graviphoton does not couple to the fields in a way that modifies the propagator of the chiral superfields, the techniques employed in can still be used in the absence of gravity. Under such assumptions the graviphoton dependence will show up as a result of applying the chiral ring relations. The chiral superfield propagator is given by
$$\frac{1}{p^2+W^\alpha \pi _\alpha +m}=_0^{\mathrm{}}𝑑se^{s(p^2+W^\alpha \pi _\alpha +m)},$$
(52)
where $`p`$ and $`\pi _\alpha `$ are the bosonic and fermionic momenta, respectively, and using holomorphicity $`\overline{m}`$ has been taken to be $`1`$. The vertices of the theory are read directly from the tree-level superpotential. The computation then proceeds by integrating over both the bosonic and fermionic loop-momenta. The bosonic integral is a simple Gaussian integral yielding a determinant depending on the Schwinger parameters $`s_I`$ while the fermionic one brings down insertions of $`W_\alpha `$ which combine using the chiral ring relations to form the graviphoton and glueball dependent terms multiplying a polynomial of the Schwinger parameters. According to the conjecture the fermionic and bosonic $`s`$-dependence should cancel leaving only a vector model computation. The $`W_\alpha `$ insertions must be path-ordered as these are no longer anti-commutative.
However, as we will soon demonstrate this approach fails. An example of this is one of the graphs for the correlation function $`M`$
The bosonic integral is found to be
$$Z_B=\frac{1}{(4\pi )^4(s_1+s_3)^2s_2^2},$$
(53)
while the fermionic integration with the origin of path-ordering taken to be the $`M`$ insertion yields by utilizing the chiral ring relations
$`Z_F`$ $`=`$ $`{\displaystyle \frac{1}{16}}s_2^2\left[(s_3^2+s_1^2)\mathrm{Tr}(W^2W^2)+2s_1s_3\mathrm{Tr}(W^\alpha W^2W_\alpha )\right]=`$ (54)
$`=`$ $`{\displaystyle \frac{N_c}{32}}s_2^2(s_1s_3)^2F_{\alpha \beta }F^{\alpha \beta },`$
whose $`s`$-dependence does not cancel with that of $`Z_B`$. But taking the path-ordering origin at the vertex one has
$$Z_F=\frac{N_c}{32}s_2^2(s_1+s_3)^2F_{\alpha \beta }F^{\alpha \beta }$$
(55)
leading to the exact cancellation of the $`s`$-dependence of this graph. One is drawn to conclude that this diagram should be taken to be zero by some selection rule in order to remove this ambiguity.
Other graphs also feature such problems. For example, the correlation function $`M^2`$ includes contribution from the graph
The bosonic integral in this case is
$$Z_B=\frac{1}{(4\pi )^6s_1^2(s_2+s_4)^2s_3^2},$$
(56)
while the fermionic integration with the origin of the path-ordering taken to be at the $`M^2`$ insertion is
$$Z_F=\frac{\pi ^2}{4}s_1^2s_3^2(s_2s_4)^2F_{\alpha \beta }F^{\alpha \beta }S.$$
(57)
Taking the origin of the path-ordering at the other vertex yields the same result. It can be seen that again in this case the $`s`$-dependence does not cancel.
We conclude that either the field theory Lagrangian must also be deformed in some way in addition to the $`C`$-deformation, or an appropriate scheme has to be developed within the framework of the above assumptions.
###### Acknowledgments.
We would like to thank P. A. Grassi, V. Kaplunovsky, K. S. Narain and H. Ooguri for valuable discussions. This work is supported by the ISF and the GIF.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.