id
stringlengths 27
33
| source
stringclasses 1
value | format
stringclasses 1
value | text
stringlengths 13
1.81M
|
|---|---|---|---|
warning/0004/hep-ph0004112.html
|
ar5iv
|
text
|
# The 𝜂_𝑐𝛾𝛾^∗ transition form factor in the decay 𝜂_𝑐→𝛾ℓ⁺ℓ⁻ and in the crossed channels 𝛾𝑒⁻→𝜂_𝑐𝑒⁻ and 𝑒⁺𝑒⁻→𝜂_𝑐𝛾.
## I Introduction
The radiative decays of mesons and baryons are an interesting source of important information for hadron spectroscopy and hadron electrodynamics. For example, it is worth recalling the role of such decays in determining the charmonium spectrum. Among radiative decays, especially intriguing are those where lepton pairs are produced. A classical example is the process $`\pi ^0\gamma e^+e^{}`$ , which is sensitive to the radius of the corresponding form factor. Such a decay is typical of any neutral pseudoscalar meson $`P`$ (= $`\eta ,\eta ^{},\eta _c`$, etc.):
$$P\gamma \mathrm{}^+\mathrm{}^{},\mathrm{}=e,\mu .$$
The corresponding transition electromagnetic form factors of $`P\gamma \gamma ^{}`$-decay are complex functions of $`t`$, the four-momentum transfer squared, in the time-like region $`4m_{\mathrm{}}^2<tm^2`$, where $`m`$ ($`m_{\mathrm{}}`$) is the mass of $`P`$ ($`\mathrm{}`$). In particular the decay $`\eta _c\gamma \mathrm{}^+\mathrm{}^{}`$ offers the possibility of studying the transition form factor in a relatively large region of time-like momentum transfer,
$$4m_{\mathrm{}}^2t9\text{GeV}^2.$$
A strong theoretical interest in the transition form factors of pseudoscalar mesons was stimulated by the experimental results of the CLEO collaboration , which measured these form factors at space-like momentum transfers, using the well known 2$`\gamma `$-exchange mechanism in the reaction $`e^+e^{}e^+e^{}P`$ (Fig. 1). If one lepton, say the positron, is not detected, we can study the process of photoproduction of pseudoscalar mesons on electrons, $`\gamma e^{}Pe^{}`$, in an unusual kinematics of colliding particles, $`\gamma `$ and $`e^{}`$. In the case of $`\pi ^0`$ production the corresponding form factor was measured up to $`t=8`$ GeV<sup>2</sup>, where, in principle, perturbative QCD applies. Recently also the $`\eta _c\gamma `$ transition form factor has been determined for two different values of $`t`$ in the spacelike region.
A special interest has the decay $`\eta _c\gamma \mathrm{}^+\mathrm{}^{}`$, as a generalisation of the decay $`\eta _c\gamma \gamma `$, whose branching ratio is known (see also predictions by dispersive methods). The availability of a form factor instead of a single constant, as in the case of $`\eta _c2\gamma `$ decay, allows a better insight of the structure of the $`\eta _c`$-meson and of perturbative QCD. It is also interesting to look at the predictions of the ”old” approach of Vector Dominance Model (VDM) in the unusual region of charmed particle decays.
This paper is devoted to a global approach to the study of all possible crossed channels, i. e.,
$$\eta _c\gamma \mathrm{}^+\mathrm{}^{},\gamma e^{}\eta _ce^{},e^+e^{}\eta _c\gamma ,$$
whose characteristics are driven by a single transition form factor. The numerical estimates are performed both within VDM and QCD, considering, in the latter case, the process $`c\overline{c}\gamma \mathrm{}^+\mathrm{}^{}`$ as the basic mechanism generating the $`\eta _c\gamma \gamma `$ transition form factor.
Section II is dedicated to the Dalitz plot distribution of the decay $`\eta _c\gamma \mathrm{}^+\mathrm{}^{}`$. In sect. III we propose two models for calculating the transition form factor $`\eta _c\gamma \gamma `$, one based on VDM and the other on QCD. In sect. IV we study the reaction $`e^+e^{}\eta _c\gamma `$. Sect. V is dedicated to the process $`\gamma e^{}\eta _ce^{}`$. Lastly in sect. VI we draw some conclusions.
## II The decay $`\eta _c\gamma \mathrm{}^+\mathrm{}^{}`$.
We start by considering the more general case of the decay of a meson $`M`$ into a real $`\gamma `$ and a lepton pair. The differential decay width reads
$$d\mathrm{\Gamma }=(2\pi )^4\frac{\overline{||^2}}{2m}\delta ^4(pqk_1k_2)\frac{d^3q}{(2\pi )^32\omega }\frac{d^3k_1}{(2\pi )^32E_1}\frac{d^3k_2}{(2\pi )^32E_2},$$
(1)
where $`\omega `$, $`E_1`$ and $`E_2`$ are the energies of the final $`\gamma `$, $`e^+`$ and $`e^{}`$ respectively.
The notation of the four-momenta is shown in Fig. 2; the line above the modulus squared of the matrix element $`||^2`$ denotes the sum over the polarisations of all the particles produced in the decay.
The standard form of the matrix element $``$ for the $`M\gamma \mathrm{}^+\mathrm{}^{}`$ decay in one-photon approximation is
$$=\frac{e^2}{t}\mathrm{}_\mu 𝒥^\mu ,$$
where $`t=(k_1+k_2)^2`$ is the effective mass squared of the leptonic pair, $`\mathrm{}_\mu =\overline{u}(k_2)\gamma _\mu u(k_1)`$ and $`𝒥_\mu `$ is the electromagnetic current in the decay $`M\gamma \gamma ^{}`$ In the decay considered here, $`𝒥_\mu `$ is proportional to the electric charge $`e`$..
According to the usual notation we obtain
$$\overline{||^2}=\frac{e^4}{t^2}L_{\mu \nu }W^{\mu \nu },$$
(2)
where
$$L_{\mu \nu }=\overline{\mathrm{}_\mu \mathrm{}_\nu ^{}},W^{\mu \nu }=\overline{𝒥^\mu 𝒥^\nu }.$$
(3)
Substituting (2) into (1), and integrating over all variables except $`E_{1,2}`$, we get the two-lepton Dalitz distribution:
$$d\mathrm{\Gamma }=\frac{\alpha ^2}{4\pi }\frac{L_{\mu \nu }W^{\mu \nu }}{t^2m}dE_1dE_2,$$
where $`\alpha `$ = $`e^2/4\pi `$. However for our aims it is more convenient to express $`E_1`$ and $`E_2`$ in terms of the dimensionless variables $`x=t/m^2`$ and $`y=2(E_1E_2)/m`$ , which are defined in the ranges $`\delta x1`$ and $`y_0yy_0`$, $`y_0=(1x)\sqrt{1\delta /x}`$ and $`\delta =4m_{\mathrm{}}^2/m^2`$. Furthermore we take into account, wherever possible, symmetry properties:
* Gauge invariance - $`k\mathrm{}`$=$`k𝒥`$=$`0`$ \- implies that the product $`L_{\mu \nu }W^{\mu \nu }`$ can be rewritten in terms of the space components of the two tensors above, i. e.,
$$L_{\mu \nu }W^{\mu \nu }=(L_{xx}+L_{yy})W_{xx}+\frac{t^2}{k_0^4}L_{zz}W_{zz},$$
(4)
where $`k_o`$ denotes the energy of the $`\gamma ^{}`$. Here we have adopted a reference frame in which the meson $`M`$ is at rest and the $`z`$axis is parallel to the momentum $`\stackrel{}{k}`$ of the virtual photon.
* Parity invariance allows to write
$$W_{ij}=(\delta _{ij}\widehat{k_i}\widehat{k_j})W_1(t)+\widehat{k_i}\widehat{k_j}W_2(t),\widehat{\stackrel{}{k}}=\stackrel{}{k}/|\stackrel{}{k}|,$$
where $`W_{1(2)}(t)`$ is the real structure function (SF) which describes the production of $`\gamma ^{}`$ with transverse (longitudinal) polarisation.
Taking into account all that, calculations yield
$$d\mathrm{\Gamma }=\frac{\alpha ^2}{16\pi }W_1(t)\frac{dxdy}{xm}\left[1+\frac{4m_{\mathrm{}}^2}{t}+R_L(t)\frac{t}{k_0^2}+y^2\frac{m^2}{4\stackrel{}{k}^2}\left(1R_L(t)\frac{t}{k_0^2}\right)\right],$$
$$R_L(t)=W_2(t)/W_1(t),k_0=\frac{m^2+t}{2m},\stackrel{}{k}^2=\frac{(m^2t)^2}{4m^2}.$$
(5)
One can see that the differential width, which characterises the Dalitz distribution for the decay $`M\gamma \mathrm{}^+\mathrm{}^{}`$, is symmetric with respect to the exchange $`E_1E_2`$, in agreement with $`C`$-invariance of the electromagnetic interaction.
The specific dependence of $`d^2\mathrm{\Gamma }/dxdy`$ on the variable $`y`$,
$$d^2\mathrm{\Gamma }/dxdy=a(x)+y^2b(x),$$
is a direct consequence of one-photon mechanism. It has the same origin as the $`\mathrm{cot}^2\frac{\theta _e}{2}`$ dependence of the differential cross section in electron-hadron scattering (where $`\theta _e`$ is the electron scattering angle in the laboratory system) and as the $`\mathrm{cos}^2\theta `$-dependence of the differential cross section for the inclusive process $`e^+e^{}hX`$, where $`h`$ denotes the final detected hadron and $`\theta `$ is the angle between the three-momenta of the hadron and of the electron in the overall CMS.
One can see that the value $`W_1(0)`$ determines the probability of the radiative decay width $`\mathrm{\Gamma }_0\mathrm{\Gamma }(M2\gamma )`$, i.e.,
$$\mathrm{\Gamma }_0=\frac{1}{2}(2\pi )^4e^2\frac{W_{xx}+W_{yy}}{2m}\delta ^4(pqk)\frac{d^3k}{(2\pi )^32\omega _1}\frac{d^3q}{(2\pi )^32\omega _2}=\frac{\alpha }{4m}W_1(0).$$
This allows to rewrite the Dalitz distribution in the following form:
$$\frac{1}{\mathrm{\Gamma }_0}\frac{d^2\mathrm{\Gamma }}{dxdy}(M\gamma \mathrm{}^+\mathrm{}^{})=\frac{\alpha }{4\pi }R_T(t)\frac{1}{x}\left[1+\frac{4m_{\mathrm{}}^2}{t}+R_L(t)\frac{t}{k_0^2}+\frac{y^2}{4\stackrel{}{k}^2}m^2\left(1R_L(t)\frac{t}{k_0^2}\right)\right],$$
(6)
where $`R_T(t)=W_1(t)/W_1(0)`$ and $`R_L(t)`$ is defined in Eq. (5).
Therefore the study of the energy distribution of leptons in $`M\gamma \mathrm{}^+\mathrm{}^{}`$ in the Dalitz plane allows to determine $`R_L(t)`$ and $`R_T(t)`$, i. e., the two fundamental quantities of the decay considered, as functions of the effective mass of the $`\mathrm{}^+\mathrm{}^{}`$-system.
The result of the integration in Eq. (6) over the variable $`y`$ determines the $`\mathrm{}^+\mathrm{}^{}`$ effective mass spectrum in the decay $`M\gamma \mathrm{}^+\mathrm{}^{}`$ :
$$\frac{1}{\mathrm{\Gamma }_0}\frac{d\mathrm{\Gamma }}{dx}(M\gamma \mathrm{}^+\mathrm{}^{})=\frac{2\alpha }{3\pi }R_T(t)\frac{(1x)}{x}\left(1+2\frac{m_{\mathrm{}}^2}{t}\right)\sqrt{1\frac{4m_{\mathrm{}}^2}{t}}\left[1+R_L(t)\frac{t}{2k_0^2}\right].$$
(7)
Turning to the specific decay $`\eta _c\gamma \gamma ^{}`$, in this case (and more generally for any neutral pseudoscalar meson) the hadron electromagnetic current reads
$$𝒥_\mu =\frac{F(t)}{m}ϵ_{\mu \alpha \beta \gamma }k^\alpha e^\beta q^\gamma ,$$
(8)
where $`e_\beta `$ is the polarisation four-vector of the real photon and $`F(t)`$ the electromagnetic form factor of the transition $`\eta _c\gamma \gamma ^{}`$, which has a nonzero imaginary part for $`t4m_\pi ^2`$. Note that we have defined the form factor differently from Lepage and Brodsky. Inserting Eq. (8) into the definition of the tensor $`W_{\mu \nu }`$, Eq. (3), we obtain
$$W_1(t)=|F(t)|^2(1x)^2\frac{m^2}{4},W_2(t)=0,$$
i.e. $`R_L(t)=0.`$
There are three different processes for studying the $`t`$dependence of $`F`$:
1. The photoproduction of $`\eta _c`$ in $`\gamma e^{}\eta _ce^{}`$ at space-like momentum transfers;
2. The decay $`\eta _c\gamma \mathrm{}^+\mathrm{}^{}`$, at time-like momentum transfers, $`4m_{\mathrm{}}^2tm^2`$;
3. The reaction $`e^+e^{}\eta _c\gamma `$, at time-like momentum transfers, $`tm^2`$.
Therefore the simultaneous study of all these different processes can allow to determine the $`t`$dependence of the electromagnetic form factor $`F(t)`$ for any value of $`t`$ (see Fig. 3). Moreover all the possible observables in the above mentioned processes depend only on $`|F(t)|^2`$. This means that the phase of $`F(t)`$ in the time-like region cannot be measured. But the determination of this phase as a function of $`t`$ is very important. For example, a dispersion relation for the form factor $`F(t)`$, considered as an analytical function of $`t`$, can be analysed only if both the modulus and the phase of $`F(t)`$ are known. The problem of measuring the phase of complex form factors of hadrons in the time-like region is not yet solved. This is a common problem of the electromagnetic form factor of the charged pion, of the neutral and charged kaons, etc..
The determination of the transition form factor $`F(t)`$ is necessary for calculating the following observables:
* The Dalitz-distribution in the variables $`x`$ and $`y`$;
* The spectrum of the effective masses in $`\eta _c\gamma \mathrm{}^+\mathrm{}^{}`$;
* The coefficient of internal conversion, i.e. the ratio of the decay width of $`\eta _c\gamma \mathrm{}^+\mathrm{}^{}`$ to the decay width of $`\eta _c\gamma \gamma `$:
$$I=\frac{\mathrm{\Gamma }(\eta _c\gamma \mathrm{}^+\mathrm{}^{})}{\mathrm{\Gamma }_0}=\frac{2\alpha }{3\pi }_\delta ^1𝑑x\left|\frac{F(t)}{F(0)}\right|^2\frac{(1x)^3}{x}\left(1+2\frac{m_{\mathrm{}}^2}{t}\right)\left(1\frac{4m_{\mathrm{}}^2}{t}\right)^{1/2}.$$
(9)
This coefficient derives its main contribution from small values of $`t`$, therefore it is sensitive to the mass of the lepton, in particular it is different in the decays $`\eta _c\gamma e^+e^{}`$ and $`\eta _c\gamma \mu ^+\mu ^{}`$.
## III Two models for $`F(t)`$
Now we consider the predictions of two models, the VDM and a pQCD-inspired non-relativistic model.
### A Vector Dominance Model
In the framework of VDM (Fig. 4), the $`t`$dependence of the electromagnetic form factor $`F(t)`$ can be written as
$$F(t)=\underset{V}{}\frac{g_V}{tm_V^2},g_V=m_V^2f_Vg(V\eta _c\gamma ),$$
(10)
having summed over the contributions of the vector mesons. $`m_V`$ is the vector meson mass, $`f_V`$ is the constant of the $`\gamma ^{}V`$-transition and $`g(V\eta _c\gamma )`$ is the constant in the vertex $`V\eta _c\gamma `$. If $`m_Vm`$, the decay $`V\eta _c\gamma `$ is possible and the corresponding branching ratio allows to determine the constant $`g(V\eta _c\gamma )`$, or, more exactly, as previously discussed, the modulus of this constant. Incidentally, notice that the constant $`f_V`$ determines the decay width of $`V\mathrm{}^+\mathrm{}^{}`$ (see Appendix).
The constants $`g_V`$ satisfy two important conditions:
$`{\displaystyle \underset{V}{}}{\displaystyle \frac{g_V}{m_V^2}}=F(0),`$ (11)
$`{\displaystyle \underset{V}{}}g_V=0.`$ (12)
The first one is a relation between different $`g_V`$, through an evident normalisation of $`F(t)`$. The constant $`F(0)`$ can be determined from the decay width of $`\eta _c2\gamma `$, i.e., from
$$\mathrm{\Gamma }_0=\frac{\alpha }{4}\left|F(0)\right|^2m,$$
(13)
which is experimentally known. Again, the value of the width $`\mathrm{\Gamma }_0`$ does not fix the sign of $`F(0)`$, which introduces a two-fold ambiguity in relation (11).
Relation (12) results from a specific asymptotic behaviour of $`F(t)`$: the structure of the current $`𝒥_\mu `$ shows that $`F(t)t^2`$ for large $`|t|`$. This differs from the standard $`|t|^1`$ behaviour of the elastic electromagnetic form factors of mesons, however it agrees with the quark counting rule, and it results from the presence in Eq. (8) of an additional four-momentum, owing to the electromagnetic current of the decay considered.
The simplest model that can be suggested here for $`F(t)`$ must contain the contribution of two vector mesons. The problem is how to select the most important contributions.
Taking into account the $`c\overline{c}`$-nature of $`\eta _c`$, the $`J/\psi `$ seems to be a good candidate, as confirmed by the value of the branching ratio $`Br(J/\psi \eta _c\gamma )=(1.3\pm 0.4)\%`$. Similarly the branching ratio $`Br(\eta _c\rho \rho )=(2.6\pm 0.9)\%`$ exhibits the importance of the $`\rho `$-contribution to the form factor $`F(t)`$. Therefore in such a two-pole model we find the following two parametrisations:
$$F(t)=\pm |F(0)|\frac{m_\rho ^2m_{J/\psi }^2}{m_{J/\psi }^2m_\rho ^2}\left(\frac{1}{tm_\rho ^2}\frac{1}{tm_{J/\psi }^2}\right),$$
(14)
which cannot be disentangled in the study of the above mentioned processes.
In order to avoid singularities at $`t`$=$`m_V^2`$, we have to substitute $`m_V^2m_V^2im_V\mathrm{\Gamma }_V`$, where $`\mathrm{\Gamma }_V`$ is the width of the $`V`$meson ($`V`$ = $`\rho `$, $`J/\psi `$).
These parametrisations offer a numerical estimation of all the observables of the $`\eta _c`$ photoproduction on electrons, $`\gamma e^{}\eta _ce^{}`$, and of the decay $`\eta _c\gamma \mathrm{}^+\mathrm{}^{}`$, where $`tm^2`$. In both processes the $`\rho `$-contribution is particularly important at relatively small momentum transfers, $`|t|1`$ GeV<sup>2</sup>. The $`J/\psi `$-contribution plays an important role in the decay $`\eta _c\gamma \mathrm{}^+\mathrm{}^{}`$, near the upper kinematical limit in the variable $`t`$, $`tm^2`$, where the $`J/\psi `$-pole is very close to the physical region.
On the other hand this simplified parametrisation is not convenient for the study of the annihilation process, $`e^+e^{}\eta _c\gamma `$, to which the entire set of $`\psi `$-resonances in principle could contribute. In this case formula (14) has to be generalised, taking into account, at least,
* the contribution of additional $`\psi `$ resonances;
* the widths of such resonances, as illustrated above.
In the framework of VDM it is in principle possible to write a five-pole representation for $`F(t)`$, with the contributions of $`\rho ,J/\psi ,\psi (3.77),\psi (4.04),\psi (4.41).`$ The constants $`g_V`$ of $`\rho ,J/\psi `$ and $`\psi (3.77)`$ can be found using the existing experimental data about the radiative decays: $`J/\psi \eta _c\gamma `$, $`\psi (3.77)\eta _c\gamma `$, $`\eta _c\rho \rho `$, $`\rho e^+e^{}`$, $`J/\psi e^+e^{}`$ and $`\psi (3.77)e^+e^{}`$. The details of the calculation are given in the Appendix. On the other hand, using relations (11) and (12), it is possible to find the constants $`g_V`$ of the $`\psi (4.04)`$ and the $`\psi (4.41)`$-resonances. As two choices are possible for the sign of $`F(0)`$ and for the constants $`g_V`$ for the first three resonances, $`\rho ,J/\psi `$ and $`\psi (3.77)`$, one finds a $`2^4=16`$-fold parametrisation for $`F(t)`$, which prevents any realistic estimation of the $`t`$behaviour of the cross section for the process $`e^+e^{}\eta _c\gamma `$. So our prediction of the various characteristics of the decay $`\eta _c\gamma e^+e^{}`$ will be based on the simplified two-pole model for $`F(t)`$.
### B QCD-Inspired Model
Now we consider the process of $`c\overline{c}`$-quark annihilation into $`\gamma \gamma ^{}`$, which receives the contribution of the two diagrams illustrated in Fig. 5.
In the limit of $`|\stackrel{}{p}|=0`$ (where $`\stackrel{}{p}`$ is the relative momentum of the $`c\overline{c}`$-system in the CMS), the $`c\overline{c}`$\- annihilation in a singlet $`S`$-state - which has the quantum numbers of the $`\eta _c`$-meson, $`J^P=0^{}`$ \- is described by a matrix element of the form
$$(c\overline{c})\stackrel{}{k}\stackrel{}{e}\times \stackrel{}{e^{}},$$
(15)
where $`\stackrel{}{e}`$ ($`\stackrel{}{e^{}}`$) is the polarisation vector of the real (virtual) photon. The following relations hold: $`\stackrel{}{e^{}}\stackrel{}{k}0`$ and $`\stackrel{}{e}\stackrel{}{k}=0`$. The result (15) can be found, without specific calculations, on the basis of symmetry properties relative to $`C`$ and $`P`$transformations for the process $`c\overline{c}\gamma \gamma ^{}`$. Due to the positive $`C`$parity of the two final $`\gamma `$’s, the sum of the total spin $`S`$ and the angular momentum $`\mathrm{}`$ of the ($`c\overline{c}`$)-system must be even. But in the limit of $`\stackrel{}{p}0`$ we have $`\mathrm{}`$ = 0, therefore $`S=0`$. In this limit the $`t`$dependence of the resulting matrix element for $`c\overline{c}\gamma \gamma ^{}`$ can be identified with the $`t`$dependence of the form factor $`F(t)`$ of the decay $`\eta _c\gamma \gamma ^{}`$. This dependence can be written in terms of the $`c`$quark propagators in the elementary process $`c\overline{c}\gamma \gamma ^{}`$, which are identical for both diagrams of Fig. 5. Therefore in the limit of $`\stackrel{}{p}=0`$ we have
$$F(t)\frac{1}{(pk)^2m_c^2}=\frac{1}{k^22pk}\stackrel{\stackrel{}{p}=0}{=}\frac{1}{k^22m_ck_0}=\frac{2}{k^24m_c^2},$$
which implies that $`F(t)`$ $``$ $`|t|^1`$ for large $`|t|`$, unlike the VDM.
The parameter $`4m_c^2`$ which appears in the denominator should be identified with a physical quantity. The simplest possibility is to set $`2m_c=m_{J/\psi }`$, which corresponds to the $`J/\psi `$-contribution to $`F(t)`$, appearing in the framework of VDM. A perturbative approach, on the basis of a factorisation of short- and long-distance physics , as considered by Feldmann and Kroll in the space-like region, results in
$$4m_c^2m^2+2<\stackrel{}{k}_{}^2>,$$
(16)
where $`\stackrel{}{k}_{}`$ denotes the transverse momentum of the $`c`$-quark in the $`\eta _c`$-meson . This substitution may be regarded as a possible procedure for taking into account the transverse momentum of the quark inside the meson. On the other hand, note that our QCD-inspired consideration above is valid for any $`t`$, both space-like and time-like, except for a small neighbourhood of $`t`$ = $`4m_c^2`$. Therefore, if we assume that the substitution (16) can be extended at least up to $`tm^2`$, $`F(t)`$ turns out to be very sensitive to $`<\stackrel{}{k}_{}^2>`$ near the upper kinematical limit of the Dalitz plot, as results from Fig. 6.
In this approach it is also possible to find the corresponding form factors of the radiative decays $`\chi _J\gamma \mathrm{}^+\mathrm{}^{}`$ of the P-wave charmonium states, $`J=0`$, $`1`$ and $`2`$. To this end it is necessary to find the matrix element of the process $`c\overline{c}\gamma \gamma ^{}`$ (see Fig. 5). The spin structure of the $`\chi `$-decay matrix elements is
$$\xi _1^{}\sigma _2\stackrel{}{\sigma }\stackrel{}{p}\xi _2\stackrel{}{e}\stackrel{}{e^{}}:\chi _0\gamma +\gamma ^{},$$
$$\xi _1^{}\sigma _2\stackrel{}{\sigma }\times \stackrel{}{p}\stackrel{}{e}\times \stackrel{}{k}\xi _2\stackrel{}{e^{}}\stackrel{}{k}:\chi _1\gamma +\gamma ^{},$$
$$\xi _1^{}\sigma _2(\sigma _ap_b+\sigma _bp_a\frac{2}{3}\delta _{ab}\stackrel{}{\sigma }\stackrel{}{p})\xi _2\left\{\begin{array}{cc}e_{1a}& e_{1b}\\ k_ak_b& \stackrel{}{e}\stackrel{}{e^{}}\\ e_{1a}k_b& \stackrel{}{e^{}}\stackrel{}{k}\end{array}\right\}:\chi _2\gamma +\gamma ^{},$$
where $`\xi _{1(2)}`$ is the two-component spinor of the $`\overline{c}`$ ($`c`$) quark.
In the general case the decays of $`\chi _0`$ and $`\chi _1`$ are characterised by one form factor, whereas in the case of $`\chi _2\gamma \gamma ^{}`$ three independent form factors are present. Notice also that the matrix element for $`\chi _1\gamma \gamma ^{}`$ is proportional to $`\stackrel{}{e^{}}\stackrel{}{k}`$, which vanishes for real photons, in agreement with the Landau-Yang theorem.
As one can see from Fig. 6, the VDM and QCD-inspired model predict essentially different $`t`$dependences for the transition form factor $`F(t)`$ in the whole region $`4m_{\mathrm{}}^2tm^2`$, where the decay $`\eta _c\gamma \mathrm{}^+\mathrm{}^{}`$ occurs. The coefficients of internal conversion of the decays $`\eta _c\gamma e^+e^{}`$ and $`\eta _c\gamma \mu ^+\mu ^{}`$ are also strongly model-dependent (see Table 1). These properties of $`F(t)`$ depend essentially on two factors: the role of the $`\rho `$meson in the VDM approach and the small mass difference between $`\eta _c`$ and $`J/\psi `$.
Also in the space-like region the two models describe the form factor quite differently; in particular the QCD-inspired model predicts for $`F(t)`$ a much slower decrease at increasing $`|t|`$, in agreement with present data.
## IV The reaction $`e^+e^{}\eta _c\gamma `$
The differential cross section in the overall CMS reads
$$\frac{d\sigma }{d\mathrm{\Omega }}=\frac{\overline{||^2}}{64\pi ^2t}\frac{|\stackrel{}{q}|}{|\stackrel{}{k}|}=\frac{\overline{||^2}}{64\pi ^2t}\left(1\frac{m^2}{t}\right),m_e=0,$$
(17)
where $`\stackrel{}{k}`$ and $`\stackrel{}{q}`$ are the three-momenta of the colliding leptons and of the produced $`\gamma `$ and $`t=(k_1+k_2)^2`$ is the square of the total energy of the colliding particles.
Taking into account Eq. (8), we get
$$\frac{d\sigma }{d\mathrm{\Omega }}=\frac{\alpha ^2}{32m^2}(1+\mathrm{cos}^2\theta )\left(1\frac{m^2}{t}\right)^3|F(t)|^2,tm^2.$$
(18)
The cross section for collinear kinematics (i.e. $`\theta =0`$ or $`\pi `$) is non-vanishing, in agreement with helicity conservation. Indeed, in the limit of $`m_e=0`$, owing to the properties of the electromagnetic current $`\overline{u}\gamma _\mu u`$, we have $`\lambda _i=\lambda _e^{}+\lambda _{e^+}=\pm 1`$; on the other hand for a real $`\gamma `$ we have $`\lambda _f=\lambda _\gamma =\pm 1`$.
In any process of the type $`e^+e^{}\gamma ^{}h_1\overline{h_2}`$, where $`h_1`$ and $`h_2`$ are hadrons or hadronic states, the one-photon mechanism is characterised by a definite angular dependence:
$$\frac{d\sigma }{d\mathrm{\Omega }}=A(t)+B(t)\mathrm{cos}^2\theta ,$$
where $`\theta `$ is the production angle of $`h_1`$ in the overall CMS and $`A(t)`$ and $`B(t)`$ are two structure functions which are quadratic combinations of the electromagnetic form factors of the transition $`\gamma ^{}h_1\overline{h_2}`$. In the case considered we have only one form factor for the vertex $`\gamma ^{}\eta _c\gamma `$, therefore $`B(t)/A(t)=1`$, as results from the angular dependence (18).
The threshold behaviour of the cross section is proportional to $`\left(1{\displaystyle \frac{m^2}{t}}\right)^3`$ and results from the magnetic dipole radiation in the transition $`\gamma ^{}\eta _c\gamma `$.
The factorisation of the $`t`$ and $`\mathrm{cos}\theta `$-dependences in the differential cross section of $`e^+e^{}\eta _c\gamma `$, Eq. (18), is due to the specific structure of the matrix element and to the choice of the CMS as a reference frame. Therefore the $`t`$dependence is the same for any production angle and coincides with the $`t`$dependence of the total cross section:
$$\sigma =\frac{\pi }{6m^2}\alpha ^2\left(1\frac{m^2}{t}\right)^3|F(t)|^2.$$
(19)
Formulae (18) and (19) can be re-written as
$$\frac{d\sigma }{d\mathrm{\Omega }}=\frac{\alpha }{2m^3}\mathrm{\Gamma }_0\left(1\frac{m^2}{t}\right)^3\left|\frac{F(t)}{F(0)}\right|^2(1+\mathrm{cos}^2\theta )=\frac{(1+\mathrm{cos}^2\theta )}{16\pi }3\sigma ,$$
$$\sigma =\frac{4}{3}\frac{\alpha \pi }{m^3}\mathrm{\Gamma }_0\left(1\frac{m^2}{t}\right)^3\left|\frac{F(t)}{F(0)}\right|^2.$$
In order to predict the $`t`$dependence of the total and differential cross sections, it is necessary to know the form factor $`F(t)`$ in the region $`tm^2`$, and especially where the contribution of all the $`\psi `$-resonances is important. From our previous discussion it turns out that the VDM can be applied here.
## V The process $`\gamma e^{}\eta _ce^{}`$.
The direct study of the reaction $`\gamma e^{}\eta _ce^{}`$ is experimentally difficult due to the large threshold, $`E_{\gamma ,thr}m^2/2m_e`$9 TeV. Therefore this process can be studied through the $`2\gamma `$ mechanism in the reaction $`e^\pm e^{}e^\pm e^{}\eta _c`$ (see Figs. 7 and 1) with a quasi-real photon produced in the $`e^\pm e^\pm \gamma `$-vertex. In this case, for the process $`\gamma e^{}\eta _ce^{}`$, it is necessary to consider a particular reference frame with colliding $`\gamma `$ and $`e^{}`$. Let us re-write the cross section of $`\gamma e^{}\eta _ce^{}`$ in terms of the Mandelstam variables $`s`$, $`t`$ and $`u`$, which can be defined in the standard form, according to the following notation of the particles four-momenta:
$$\gamma (q)+e^{}(k_1)e^{}(k_2)+\eta _c(p),$$
$$s=(k_1+q)^2,$$
$$u=(k_2q)^2,$$
$$t=(k_1k_2)^2,$$
where $`q`$, $`p`$ and $`k_1`$ ($`k_2`$) are the four-momenta of the photon, of $`\eta _c`$ and of the initial (final) electron. In the overall CMS the differential cross section can be written as
$$\frac{d\sigma }{d\mathrm{\Omega }_\eta }(\gamma e^{}\eta _ce^{})=\frac{\overline{||^2}}{64\pi ^2s}\left(1\frac{m^2}{s}\right).$$
In the one-photon exchange approximation the expression of $`|\overline{}|^2`$ can be derived from eq (5), by exploiting the crossing symmetry. In the direct channel we have
$$\overline{||^2}=\frac{e^4|F(t)|^2}{m^2t}(1+\mathrm{cos}^2\theta )(tm^2)^2,$$
(20)
where $`t`$ is the overall CMS energy squared.
Taking into account the relation
$$1\mathrm{cos}^2\theta =\frac{4su}{(tm^2)^2},$$
the invariant differential cross section for $`\gamma e^{}e^{}\eta _c`$ results to be
$$\frac{d\sigma }{dt}=\frac{\pi \alpha ^2}{4s^2m^2|t|}\left[(tm^2)^22us\right]|F(t)|^2=4\pi \frac{\alpha }{m^3}\mathrm{\Gamma }_0\left|\frac{F(t)}{F(0)}\right|^2\frac{(tm^2)^22us}{s^2|t|}.$$
(21)
Since the four-momentum transfer in this process is space-like, the reaction considered is especially sensitive to the $`\rho `$meson contribution, which is the lightest vector meson - in particular for $`|t|1`$ GeV<sup>2</sup>.
In the framework of one-photon mechanism a definite prediction for the polarisation observables can be derived. Independently of the kinematical conditions, the absorption asymmetry of a linearly polarised photon, i. e.,
$$\mathrm{\Sigma }=\frac{\sigma _{}\sigma _{}}{\sigma _{}+\sigma _{}},$$
where $`\sigma _{}`$ ($`\sigma _{}`$) is the differential cross section with linearly polarised photons, whose polarisations are orthogonal (parallel) to the reaction plane - is equal to 1. This result, which implies $`\sigma _{}`$ = 0, follows from parity conservation.
This prediction can be tested using a linearly polarised photon beam and detecting the final electron. The electron polarisation here does not carry any additional information, owing to the presence of a single electromagnetic form factor. However for two-spin polarisation observables non-trivial effects can be produced by the collisions of circularly polarised photons with longitudinally polarised electrons.
## VI Conclusions
We have considered some characteristics of the radiative decay $`\eta _c\gamma \mathrm{}^+\mathrm{}^{}`$, such as the Dalitz distribution with respect to the energies of the two final leptons, the effective mass spectrum of the lepton pair and the coefficient of internal conversion, i.e. the ratio $`\mathrm{\Gamma }(\eta _c\gamma \mathrm{}^+\mathrm{}^{})/\mathrm{\Gamma }(\eta _c2\gamma )`$. We have used two different models for the electromagnetic form factor $`F(t)`$. The VDM allows to find a relatively simple 2-pole representation for the form factor $`F(t)`$, with a definite normalisation at $`t=0`$ and asymptotic behaviour in $`t^2`$, in agreement with the quark counting rule. The VDM parametrisation contains the contribution of the $`\rho `$meson (which is important owing to the large width of the decay $`\eta _c\rho ^0\rho ^0`$) and the $`J/\psi `$-contribution. This two-pole representation can be used for analysing the decay $`\eta _c\gamma \mathrm{}^+\mathrm{}^{}`$, which corresponds to timelike $`t`$, and the photoproduction process $`\gamma e^{}\eta _ce^{}`$, where $`t`$ is space-like. However in the calculation of the energy dependence of the differential and total cross section of the process $`e^+e^{}\eta _c\gamma `$ the form factor $`F(t)`$ could be sensitive to the whole set of $`\psi `$-resonances. A suitable parametrisation of the form factor, in terms of the contribution of the vector mesons $`\rho `$, $`J/\psi `$, $`\psi (3.77)`$, $`\psi (4.04)`$ and $`\psi (4.41)`$, can be found in the framework of VDM, but with a $`2^4=16`$-fold ambiguities in the definition of the signs of the various constants. The QCD-inspired model for the form factor $`F(t)`$ on the basis of the process $`c\overline{c}\gamma \mathrm{}^+\mathrm{}^{}`$ (with two pole Feynman diagrams) results in a parametrisation which is very similar to the $`J/\psi `$-contribution alone. This model predicts an asymptotic behavour of the type $`|t|^1`$ for large $`|t|`$, in contrast with the VDM.
##
Here we consider the decays $`\eta _c\rho ^0\rho ^0`$, $`V\eta _c\gamma `$ and $`V\mathrm{}^+\mathrm{}^{}`$, whose decay constants are necessary for the calculation of the form factor $`F(t)`$ in VDM.
The matrix element for the process $`\eta _c\rho ^0\rho ^0`$ can be written as
$$=\frac{g(\eta _c\rho ^0\rho ^0)}{m}ϵ^{\mu \nu \alpha \beta }U_{1\mu }U_{2\nu }p_{1\alpha }p_{2\beta },$$
where $`U_1(U_2)`$ and $`p_1(p_2)`$ are the polarisation four-vectors and four-momenta of the two produced particles. Summing over the final polarisations, the partial decay width reads
$$\mathrm{\Gamma }(\eta _c\rho ^0\rho ^0)=\frac{m}{64\pi }|g(\eta _c\rho ^0\rho ^0)|^2\left(1\frac{4m_\rho ^2}{m^2}\right)^{3/2},$$
the constant $`g(\eta _c\rho ^0\rho ^0)`$ being dimensionless in the present normalisation.
The matrix element of the decay $`V\mathrm{}^+\mathrm{}^{}`$ (see Fig 8, which illustrates the notation of the particle four-momenta), can be written as
$$=e^2f_V\overline{u}(k_1)\gamma _\mu u(k_2)U^\mu ,$$
where $`U_\mu `$ is the polarisation four-vector of the $`V`$meson. Averaging over $`U_\mu `$ and summing over the final lepton polarisations, we find, in the limit of zero leptonic mass,
$$\mathrm{\Gamma }(V\mathrm{}^+\mathrm{}^{})=\alpha ^2f_V^2\frac{4\pi }{3}m_V.$$
The matrix element of the decay $`VP\gamma `$, where $`P`$ is a pseudoscalar meson (Fig 9), can be written as
$$=e\frac{g(VP\gamma )}{m_V}ϵ^{\mu \nu \alpha \beta }U_\mu e_\nu q_\alpha k_\beta .$$
Summing over the $`\gamma `$\- polarisations and averaging over the $`V`$-polarisations yields
$$\mathrm{\Gamma }(VP\gamma )=\frac{\alpha g^2(VP\gamma )}{24}m_V\left(1\frac{m^2}{m_V^2}\right)^3.$$
Finally $`g_V`$ is connected to $`g(VP\gamma `$) and $`f_V`$ through the relation
$$g_V=ef_Vg(VP\gamma )m_V^2.$$
We stress that this relation does not fix the sign of the constant $`g_V`$.
|
warning/0004/cond-mat0004081.html
|
ar5iv
|
text
|
# Applicability of Fermi golden rule and possibility of low-field runaway transport in nitrides
## I INTRODUCTION
One of the main factors determining electron transport characteristics in polar semiconductors is scattering of the electrons by polar optical phonons. For relatively weak electron-phonon interactions, when scattering events can be considered as independent, use of Fermi’s golden rule for the calculation of energy-dependent frequencies of electron transitions provides an adequate description of experimentally obtained velocity-field curves. Upon increasing of the interaction, however, the polaronic effects induced by autolocalization of an electron by the inertial part of the crystal polarization become more prominent and they determine the character of the scattering. Intensification of the electron-phonon interaction leads eventually to a situation where the average intercollision time becomes less than the duration of a collision. Such a strong coupling, therefore, requires proper account for the quantum interference effects and makes the problem of electron drift essentially nonlinear. This complicates dramatically the theoretical treatment of carrier scattering and field-dependent transport since, for the given case, the standard perturbation technique is not applicable.
To ensure energy conservation for the short-time perturbations, the inverse scattering rate, $`\tau `$, must be large enough to satisfy the inequality $`\tau \mathrm{}/\mathrm{\Delta }`$, where $`\mathrm{\Delta }`$ is the electron transition energy. This criterion, however, does not allow one to conclude whether or not a wide class of materials in which the transition energy may become of the order of the linewidth can be described successfully in the framework of standard perturbation approaches. To such a class of materials belong, in particular, nitrides of Al and Ga. These semiconductors have been investigated recently quite intensively due to a number of their unique properties that can be utilized in the current state-of-the-art semiconductor technology. However, most attempts to describe scattering processes in nitrides have been undertaken assuming the validity of Fermi golden rule. Moreover, despite the possibility of growing the nitrides of group III in zincblende-like structures, their crystal structure at ambient conditions is wurtzite-like. For wurtzites, it is generally necessary to account for optical anisotropy when considering the carrier-optical-phonon interactions. Since optical phonon spectra in wurtzites are far more complicated than those of cubic crystals, the majority of theoretical results have been obtained by ignoring the features of the phonons in optically anisotropic media. Recently, a formalism has been developed for evaluation of the scattering rates in bulk wurtzite-like semiconductors and heterostructures by taking into account peculiarities of the phonon spectra obtained in the framework of macroscopic dielectric continuum model. Due to the anisotropy-induced complexity of the problem, Ref. took advantage of the perturbation theory. However, as indicated by the previously-discussed considerations, the validity of such an approach requires independent confirmation.
The present paper demonstrates applicability of Fermi’s golden rule for describing adequately the electron-longitudinal-phonon interaction in polar materials by comparing the field-velocity dependences obtained in the frameworks of (a) the perturbation theory and (b) the non-perturbative path-integral approach of Thornber and Feynman (TF). A supplemental, but very important, result obtained from the present investigation is the discovery of the possibility for unique low-field long-distance runaway transport in materials characterized by strong electron-phonon interactions. Examples of such materials are the nitrides of Al and Ga.
## II MODEL
The most systematic and self-consistent approach for evaluating the long-range polaron ground-state energy $`G`$, effective mass $`m_0`$, and carrier energy dissipation for both strong- and weak-coupling limits of electron-phonon interaction has been developed by Feynman et al. In Ref. , the problem of electron drift in a parabolic band under steady state conditions is considered quantum mechanically assuming that all the energy losses are due to interaction of electrons with polar optical modes. Taking advantage of the Fröhlich’s polaron model, the authors used the path-integral method to eliminate the lattice coordinates from the momentum balance equation and obtained an expression for the magnitude of the electric field $`E`$ that is required to maintain a particular magnitude of electron velocity $`V`$ at arbitrary temperature and interaction strength characterized by the coupling constant
$$\alpha =\frac{e^2}{\mathrm{}}\left(\frac{1}{ϵ_{\mathrm{}}}\frac{1}{ϵ_0}\right)\left[\frac{m^{}}{2\mathrm{}\mathrm{\Omega }}\right]^{1/2}.$$
(1)
Herein, $`e`$ is the elementary charge. Evaluation of $`\alpha `$ requires four parameters: electron effective mass $`m^{}`$; frequency of the longitudinal phonon $`\mathrm{\Omega }`$; static dielectric constant $`ϵ_0`$; and high frequency dielectric constant $`ϵ_{\mathrm{}}`$. All these parameters can be measured experimentally and they are the only external parameters required for calculation of energy loss per unit distance $`eE`$ versus $`V`$.
Thornber and Feynman have calculated the dependences of $`eE(V)`$ for three coupling constants ($`\alpha =3,5,7`$) over a wide range of reciprocal temperatures $`\beta =\mathrm{}\mathrm{\Omega }/(k_BT)`$, where $`k_B`$ is the Boltzmann constant and $`T`$ is the temperature of the lattice. The general result of these calculations can be summarized briefly as follows. For each particular $`\alpha `$, $`eE(V)`$ has a maximum at some threshold value $`V_{th}`$. For $`\beta >1`$, location of this maximum becomes independent of temperature. For $`V<V_{th}`$, $`eE`$ is an increasing function of velocity. This interval of velocities corresponds to a stable situation when energy loss to the lattice due to the absorption and emission of optical phonons can be compensated by the energy gained by the electron from the applied field in such a way that at the given $`E`$, a small deviation $`\mathrm{\Delta }V`$ of the velocity from its steady state value $`V_s`$ creates a force, $`e[E(V_s)E(V_s\pm \mathrm{\Delta }V)]`$, which stabilizes the velocity at $`V_s`$. When the external field approaches the value $`E_{th}=E(V_{th})`$, the dependence tends to saturate since the magnitude of the energy loss due to interaction with optical phonons is finite. The case $`E>E_{th}`$ was excluded from consideration because no steady state conditions can be reached for such fields and electron would accelerate infinitely. The theory, however, predicts the existence of solutions for $`V>V_{th}`$. In this region, $`eE`$ is a decreasing function of the velocity which leads to an unstable steady state situation. For this case, any deviation of the velocity from $`V_s`$ would lead to either deceleration of the electron to velocity $`V<V_{th}`$ which is stable at the given field, or a gradually increasing acceleration if $`\mathrm{\Delta }V`$ leads to an increase in velocity. It is essential, that for $`V>V_{th}`$, the dependence $`eE(V)`$ can be interpreted as a time-dependent momentum loss in the absence of the external field. This loss would coincide with the rate of electron momentum loss if the criterion $`dV/dtV/`$(duration of the collision) is satisfied. For $`\beta >1`$, $`eE(V>V_{th})`$ also becomes independent on temperature.
In order to simplify the comparison, and taking into account that the strongest electron-polar-optical-phonon scattering is due to emission of the longitudinal optical (LO) phonons, we will consider the case when $`\beta 4`$. Due to high energy of LO phonons in the nitrides, such value of $`\beta `$ would correspond to room temperature in these materials. For GaAs, which we take as a reference point in our investigation, $`\beta =4`$ would correspond to lattice temperature of order of 104 K. Since $`\beta _{GaAs}`$ is slightly higher than 1 at the room temperature, the result obtained for maximum energy loss per unit distance can be compared to the experimental velocity-field dependences (see, for example, Ref. and citations therein). Indeed, at some threshold field $`E_{th}`$, the dependence $`V(E)`$ has a maximum caused by transitions of the carriers to an upper valley with a higher effective mass. In terms of the TF model, these transitions would start to occur when the energy supply from the external field would exceed the maximum loss to the lattice; i.e., at $`E>E_{th}`$.
Thus, if the average kinetic energy of electrons obtained in the framework of TF model at $`E_{th}`$ does not exceed the energy of bands separation and the effects of the increased effective mass due to non-parabolicity of $`\mathrm{\Gamma }`$ band can be neglected, the value of the argument at maximum of $`V(E)`$ dependence has to correlate with the extremum of the $`eE(V)`$.
Under the assumptions made above, we use the simplest model for estimation of energy loss to the lattice in the framework of the perturbation theory. In this model, we calculate the dependence of scattering rate $`1/\tau `$ due to the emission of LO phonons on a single electron kinetic energy $``$ using the Fermi’s golden rule. For an optically isotropic material, the scattering rate is
$`{\displaystyle \frac{1}{\tau }}=\left({\displaystyle \frac{2m^{}}{}}\right)^{1/2}{\displaystyle \frac{e^2\mathrm{\Omega }(N_q+1)}{\mathrm{}}}\left({\displaystyle \frac{1}{ϵ_{\mathrm{}}}}{\displaystyle \frac{1}{ϵ_0}}\right)`$ (2)
$`\mathrm{ln}\left[\sqrt{{\displaystyle \frac{}{\mathrm{}\mathrm{\Omega }}}1}+\sqrt{{\displaystyle \frac{}{\mathrm{}\mathrm{\Omega }}}}\right],`$ (3)
where $`N_q`$ is the phonon occupation number. Then, we assume that a carrier with velocity $`V=\sqrt{2/m^{}}`$ loses the energy $`\mathrm{}\mathrm{\Omega }`$ to the lattice in a distance $`\tau V`$.
## III RESULTS AND DISCUSSION
### A Limiting cases
The energy loss per unit distance vs. the electron velocity calculated for GaAs at room temperature in the framework of the model which uses Fermi’s golden rule is shown on Fig. 1 by the thin solid line. It is important, that as anticipated, the maximum of the dependence is in a good agreement with the experimental data which give the maximum of $`V(E)`$ at $`E3.43.9`$ kV/cm. One should note, that since calculations are made in a one-electron approximation and for single parabolic band, they overestimate the velocity obtained at the maximum. The experimentally measured values at maximum of $`V(E)`$ reflect the averaging of the velocities over bands with different effective masses as well as effects of non-parabolicity.
The curve calculated in the simplest model at 104 K ($`\beta =4`$) is given by the thick solid line. Surprisingly, the $`eE(V)`$ dependence computed for $`\beta =4`$ in the framework of the TF model (upper dashed line) exhibits a maximum located at somewhat higher fields. We assume that such a discrepancy occurs because in a weak-coupling limit the zero-order distribution for the electrons in this model reduces to a drifted quasi-Maxwellian. An essential requirement for such a distribution to be valid in the given case is the presence of high electron concentration and strong electron-electron interactions, which provide randomization of the direction of electron momentum between the scattering events. It is unlikely, however, that such a randomization can be achieved for scattering with emission of polar optical phonons. Indeed, in the weak-coupling limit the motion of the carrier in the near-threshold fields becomes essentially one-dimensional. Conservation of energy and momenta, valid in the weak-coupling limit, require the angle between the carrier and scattered phonon momenta to be no more than $`\mathrm{arccos}(\mathrm{}\mathrm{\Omega }/)`$. Due to this condition, emission of an LO phonon at threshold fields causes deviation of the direction of electron momentum from the direction of applied field by no more than $``$ 20 degrees. Additional focusing of the electron momentum in the direction of the electric field comes from the inverse dependence of the interaction matrix element on the phonon wavevector and leads to overestimation of energy losses when using the Maxwellian distribution.
In order to resolve this discrepancy, one can suggest - in analogy to the classical case - that at the same electron temperature, reduction of the dimension would correspond to a reduction in the average carrier energy. Since the deviation of the direction of the electron momentum relative to the direction of the electric field is small but finite, for the weak-coupling limit we have reduced the energy scale in TF model by a factor of two in order to match the maxima. This corresponds to decreasing the velocity and energy losses by factors $`2^{1/2}`$ and $`2^{3/2}`$, respectively. The corrected curve $`eE(V)`$ is shown by lower dashed line.
The differences in the shapes of the curves obtained in the simplest model and the TF model occur for the following reasons. The former model considers only the emission mechanism, whereas the latter model also takes into account absorption. As shown in the figure by dash-dotted line calculated for the uncorrected case of TF model for $`\beta 21`$, elimination of phonons absorption by reducing the lattice temperature yields the same slope of $`eE(V<V_{th})`$ as in the simple model. As expected, for this temperature interval, the temperature decrease does not affect the shape of the curve at and beyond the maximum. For $`V>V_{th}`$, the discrepancy in $`eE(V)`$ between the models appears to be due to the relatively high value of $`dV/dt`$ in the unstable region. In the framework of the TF model, one can estimate this value by $`eE/m_0`$, where $`m_0`$ can be obtained as $`m_0m^{}\nu ^2/\omega ^2`$; $`\nu `$ and $`\omega `$ are the parameters of TF model. We have computed these parameters from minimization of the free energy at zero temperature. Our estimations show that assuming the duration of the collision to be equal to $`\tau `$, the value of the derivative would be much less than $`V/\tau `$ only for $`V10^9`$ cm/sec. Thus, for $`V>V_{th}`$ the simple model cannot be used for GaAs under the conditions when an external field is applied.
Additional confirmation of the idea that the corrections required for the application of the TF model in the weak-coupling limit are induced by focusing of the carrier momentum comes from the fact, that in the case of strong coupling - when the directions of momenta are randomized due to strong electron-phonon interaction discussed previously - the model correctly explains the experimentally-obtained results. In Fig. 2 we depict the energy losses calculated for $`\mathrm{Al}_2\mathrm{O}_3`$. The maximum on the dependence obtained in the TF model (dashed curve) is in the excellent agreement with the experimentally obtained maximum losses in this material, 0.03 eV/$`\AA `$.
As expected, the maximum losses calculated in the perturbative model (solid line) are less by an order of magnitude. Of course, since for the given material $`\alpha >1`$, the simple model is not valid and we have presented here both dependences simply as a means of estimating the possible error which can be induced by a perturbative treatment and to demonstrate that no energy scale reduction can fit these dependences.
### B Intermediate case: nitrides
Due to the optical anisotropy inherent to wurtzites, the coupling parameter of polaron theory $`\alpha `$ becomes dependent on the angle $`\theta `$ between phonon wavevector and the optical axis. Assuming that $`ϵ_z^{\mathrm{}}=ϵ_t^{\mathrm{}}`$, we define this dependence as
$`\alpha (\theta )={\displaystyle \frac{e^2}{ϵ^{\mathrm{}}\mathrm{\Omega }}}\sqrt{{\displaystyle \frac{m^{}}{2(\mathrm{}\mathrm{\Omega })^3}}}[{\displaystyle \frac{\omega _{Lz}^2\omega _z^2}{(\mathrm{\Omega }^2\omega _z^2)^2}}\mathrm{cos}^2\theta `$ (4)
$`+{\displaystyle \frac{\omega _{Lt}^2\omega _t^2}{(\mathrm{\Omega }^2\omega _t^2)^2}}\mathrm{sin}^2\theta ]^1,`$ (5)
where $`\omega _{Lz}`$, $`\omega _z`$, $`\omega _{Lt}`$, and $`\omega _t`$ are the characteristic frequencies of the $`\mathrm{A1}(\mathrm{LO})`$, $`\mathrm{A1}(\mathrm{TO})`$, $`\mathrm{E1}(\mathrm{LO})`$, and $`\mathrm{E1}(\mathrm{TO})`$ modes, respectively. The phonon frequency as a function of $`\theta `$ can be obtained from the dispersion relation for the extraordinary bulk phonons. The dependence $`\alpha (\theta )`$ calculated for GaN is shown on Fig. 3.
To obtain the $`eE(V)`$ dependence in the TF model we have used $`\alpha =0.46`$. This value corresponds to the energy of LO phonon calculated for GaN in the cubic phase. The energy of the LO phonon in cubic AlN is taken to be 113 meV which is - as for the GaN case - between the energies of A1 and E1 LO modes in wurtzite phase. The scattering rates are calculated according to the formalism developed in Ref. . Comparison of the dependences obtained for the nitrides is given in Fig. 4. Again, the maxima obtained in the simple model correlate very well with the threshold electric fields of $`V(E)`$ dependences computed in the Monte Carlo technique for a three-valley model for the conduction band: 140 kV/cm for GaN and 450 kV/cm for AlN. Note that in order to match the maxima one needs to use the same reduction of the energy scale when calculating the energy losses in the TF model as for the case of GaAs. Additionally, one can see that the shapes of the dependences for $`VV_{th}`$ are almost the same. This agreement between the simple and the TF models is due to the extremely short duration of the collisions which can be estimated roughly as the inverted scattering rate, $`\tau 10^{14}`$ sec. The increase of the polaron effective mass cannot compensate the increase in the energy loss near the threshold value and, therefore, cannot reduce dV/dt. Nevertheless, due to frequent collisions, the criterion $`dV/dtV/\tau `$ is satisfied for the nitrides even at $`VV_{th}^+`$.
The comparison made here allowed us to conclude that in materials for which $`\tau \mathrm{\Omega }1`$, application of Fermi golden rule to explain transport phenomena is as good as in materials traditionally handled with the perturbation theory, i.e., in the materials for which much stronger criterion, $`\tau \mathrm{\Omega }1`$, is satisfied.
### C Low-field runaway transport
It is important to emphasize another result of our investigation. Figure 5 represents the dependencies of energy losses on electron kinetic energy obtained in the framework of the TF model at room temperature for the three materials considered: GaAs, GaN and AlN. On the figure, vertical arrows indicate the energy of the closest upper valley in the corresponding conduction band. The picture shows clearly the possibility to achieve unique pre - threshold - field runaway transport in the nitrides. Indeed, let us consider a GaN sample in an external electric field of 100 kV/cm. As shown in the figure, the two steady-state solutions for electron energy would correspond to such a situation. One of the solutions lies in the region $`V<V_{th}`$ and, consequently, it reflects a stable solution with respect to the electron energy fluctuations. Another one falls into the unstable area, $`V>V_{th}`$. Suppose an electron is injected into the sample with an energy somewhere in between the energy which corresponds to the second, unstable, solution and the threshold energy.
In this case, since the energy losses to the lattice would exceed the energy gain from the external field, the electron would decelerate until the stable solution at given field would be reached. If, however, the energy of the injected electron would just slightly exceed the value of high-energy steady-state solution, the electron would accelerate moving downwards on the unstable branch of the dependence until it gains enough energy, $`^{}`$, to appear in the nearest upper valley of the conduction band. The value of acceleration would gradually increase due to the increasing difference between the force caused by the external field, $`F=100`$ keV/cm, and the force caused by energy losses to the lattice represented by the $`eE()`$ dependence. In order to estimate the minimum runaway length $`L_r`$ , we have assumed that the energy of the injected electron is 0.7 eV and that the maximum force, $`FeE(^{})`$ , is constantly applied to the carrier. Under this assumption, we obtained $`L_r^{GaN}>220`$ nm.
It is easy to see that because the intervalley separation in AlN is smaller, the effect in this material is expected to be not as strong as in the previous case. Assuming injection energy 0.31 eV and applying field 300 kV/cm, we get $`L_r^{AlN}>39`$ nm. Our results also show that the previously-discussed runaway transport cannot be achieved in GaAs due to the small intervalley gap and broad peak on the $`eE()`$ dependence.
The results presented in this paper were obtained ignoring the non-parabolicity effects. These effects, however, would not change our main findings qualitatively. In order to improve accuracy of the expected quantities, further investigation is required. It is interesting to note, that the abstract possibility of the low-field runaway transport was mentioned initially by Thornber and Feynman. Nitrides of Ga and Al promise to be materials where such transport could be actually realized.
## IV SUMMARY
In present paper, we have compared the energy losses to the lattice calculated in different polar semiconductors within the frameworks of both non-perturbative and perturbative approaches. Our results reveal that application of Fermi golden rule for calculation of the scattering rates in nitrides, where $`\tau \mathrm{\Omega }1.5`$, is as appropriate as application of this standard perturbative treatment in the materials for which the well-known criterion $`\tau \mathrm{\Omega }1`$ is satisfied. This finding dramatically simplifies the analysis of transport phenomena in the wide-band polar semiconductors with intermediate magnitude of polaron coupling factor, $`\alpha <1`$.
Applying the non-perturbative path-integral approach of Thornber and Feynman to evaluation of field dependent electron energy dissipation in AlN, GaN and GaAs, we have found that pre-threshold low-field runaway electron transport can be realized in the nitrides. The conditions for such a transport can be formulated as follows: (a) the energy of the injected carrier should exceed the energy which corresponds to the solution of the momentum balance equation located on the unstable branch of $`eE(V)`$; and (b) the separation between the energy of injected carrier and the energy of the bottom of an upper valley must be high enough to provide a finite value of runaway length. It must be at least a few times higher than the energy of the polar optical phonon.
## V Acknowledgments
This study was supported, in part, by the Office of Naval Research and by the U.S. Army Research Office.
|
warning/0004/hep-ph0004163.html
|
ar5iv
|
text
|
# Second Cluster Integral and Excluded Volume Effects for the Pion Gas
## I Introduction
Thermal models (”fireball”) have been popular for decades (see, e.g. ) to fit the data on multiparticle production in high energy nucleus–nucleus collisions (see, e.g. and references therein). The ideal gas (IG) model of noninteracting hadrons and resonances which has mostly been empoyed to extract temperature $`T`$, baryonic chemical potential $`\mu _B`$, etc., from fits to the data is however not adequate for this purpose. This is among other things because of the following two reasons:
* The ideal gas model ignores the finite width of the resonances, while most of them have a width comparable to or even larger than the typical temperatures of the hadron gas $`120÷180`$ MeV. This leads to underestimation of the attraction between hadrons.
* The IG model does not take into account nonresonance interaction between hadrons, in particular the repulsion. As a result, in the description of the hadron yields data for the AGS and SPS energies the IG model leads to artificially large particle number densities, e.g. $`\rho 1.25`$ fm<sup>-3</sup> at $`T=185`$ MeV and $`\mu _B=270`$ MeV , which hardly can be consistent with a picture of a gas of point-like, noninteracting hadrons.
The procedure of Ref. introduces the Breit–Wigner mass spectrum of resonances. However, this widely used procedure works for narrow resonances only. It was found that it is insufficient in the realistic case as it does not take into account correlated particle pairs appearing along with resonances in the hadron gas. Therefore, the standard procedure underestimates tha attractive part of hadron interactions.
To solve the second problem the procedure, which allows to take into acount finite particcle volume, was proposed by Hagedorn and Rafelski . The excluded-volume Van der Waals equation of state was derived in Ref. and used by several authors (see, e.g. and references therein). Recently, this procedure was generalized to multicomponent and relativistic particle systems . Still, the proper particle volume was so far calculated by classical statistical mechanics formulae.
Our aim is to calculate Mayer’s cluster integrals (CIs) for the hadron gas from the available data on the hadron scatterings using correct quantum formulae and use them for fixing the parameters of the Van der Waals excluded volume model. In the present paper a first step in this direction is made, namely we calculate the second cluster integral (CI) in the case of a pure pion gas for a wide temperature range and consider the contribution of the repulsive part of the $`\pi \pi `$ interactions into the CI as an excluded volume of the Van der Waals model.
The article is organized as follows: in section II we derive the formula for the second cluster integral taking into account relativistic effects as well as the isospin of the pion. Section III is devoted to the hard-core repulsion at the quantum level. The domain of applicability of the classical formulae is found. The resonance attraction will be considered in section IV. The conditions which allow to use the narrow resonance approximation (NRA) and the Bright–Wigner formula of Ref. will be studied. In section V the CI for the pion gas is calculated from the experimental data on the $`\pi \pi `$-scattering. The results are compared with various approximations. In section VI the interacting pion gas is studied in the framework of the excluded-volume Van der Waals approach. The conclusions and a discussion of the results are given in section VII.
## II General formulae
Quantum mechanical formula for a calculation of Mayer’s second cluster integral $`b_2`$ in the case of nonrelativistic zero isospin $`I_0=0`$ particles was considered in Ref. (see also ). The pions, however, have nonzero isospin $`I_0=1`$ and the temperature of interest can be comparable to or larger than the pion mass. Therefore, for an adequate description of the pion gas a generalization of the formulae given in Refs. for relativistic particles carrying nonzero isospin is needed.
We start from the canonical partition function for N identical particles in the volume $`V`$ at the temperature $`T`$
$$Z(V,T,N)=d^{3N}r\underset{\alpha }{}\mathrm{\Psi }_\alpha ^{}(𝐫_1,𝐫_2,\mathrm{}𝐫_N)\mathrm{exp}\left(\frac{H}{T}\right)\mathrm{\Psi }_\alpha (𝐫_1,𝐫_2,\mathrm{}𝐫_N),$$
(1)
where $`H`$ is the Hamiltonian operator and $`\{\mathrm{\Psi }_\alpha \}`$ is a complete set of orthonormal wave functions in co-ordinate representation. With the notations
$$W_N(𝐫_1,𝐫_2,\mathrm{}𝐫_N)N!\underset{\alpha }{}\mathrm{\Psi }_\alpha ^{}(𝐫_1,𝐫_2,\mathrm{}𝐫_N)\mathrm{exp}\left(\frac{H}{T}\right)\mathrm{\Psi }_\alpha (𝐫_1,𝐫_2,\mathrm{}𝐫_N)$$
(2)
one gets
$$Z(V,T,N)=\frac{1}{N!}d^{3N}rW_N(𝐫_1,𝐫_2,\mathrm{}𝐫_N).$$
(3)
The function $`W_1(𝐫_1)`$ can be calculated in the thermodynamical limit $`V\mathrm{}`$
$`W_N(𝐫_1)`$ $`=`$ $`{\displaystyle \underset{𝐩,t_I}{}}{\displaystyle \frac{e^{i(𝐩,𝐫_1)}}{\sqrt{V}}}\mathrm{exp}\left({\displaystyle \frac{H}{T}}\right){\displaystyle \frac{e^{i(𝐩,𝐫_1)}}{\sqrt{V}}}`$ (4)
$`=`$ $`(2I_0+1){\displaystyle \frac{d^3p}{(2\pi )^3}\mathrm{exp}\left(\frac{\sqrt{𝐩^2+m^2}}{T}\right)}=g\varphi (T;m),`$ (5)
where $`I_0`$ and $`m`$ are, respectively, the particle isospin and mass ($`t_I=I_0,\mathrm{},+I_0`$ is the isospin projection), $`g(2I_0+1)`$ is the isospin degeneration factor<sup>*</sup><sup>*</sup>* If not only the isospin but also the spin has a nonzero value, $`J_0`$, the degeneration factor has the form $`g(2I_0+1)(2J_0+1)`$. and $`\varphi (T;m)`$ can be expressed via $`K_2`$ modified Bessel function
$$\varphi (T;m)=\frac{1}{2\pi ^2}_0^{\mathrm{}}p^2𝑑p\mathrm{exp}\left(\frac{\sqrt{p^2+m^2}}{T}\right)=\frac{m^2T}{2\pi ^2}K_2\left(\frac{m}{T}\right).$$
(6)
The asymptotics of $`\varphi (T;m)`$ in the nonrelativistic, $`m>>T`$, and ultra-relativistic, $`m<<T`$, limits are
$$\varphi (T;m)\{\begin{array}{cc}\left(\frac{mT}{2\pi }\right)^{3/2}\mathrm{exp}(m/T),\hfill & m>>T\hfill \\ & \\ \frac{T}{\pi ^2}\left(T^2\frac{m^2}{4}\right),\hfill & m<<T\hfill \end{array}$$
(7)
Following Refs. we introduce the functions $`U_l(𝐫_1,𝐫_2,\mathrm{}𝐫_l)`$:
$`W_1(𝐫_1)`$ $`=`$ $`U_1(𝐫_1)`$ (8)
$`W_2(𝐫_1,𝐫_2)`$ $`=`$ $`U_1(𝐫_1)U_1(𝐫_2)+U_2(𝐫_1,𝐫_2)`$ (9)
$`W_3(𝐫_1,𝐫_2,𝐫_3)`$ $`=`$ $`U_1(𝐫_1)U_1(𝐫_2)U_1(𝐫_3)+U_1(𝐫_1)U_2(𝐫_2,𝐫_3)`$ (11)
$`+U_1(𝐫_2)U_2(𝐫_3,𝐫_1)+U_1(𝐫_3)U_2(𝐫_1,𝐫_2)+U_3(𝐫_1,𝐫_2,𝐫_3)`$
etc. (12)
and define Mayer’s CIs The normalizations of the CIs in Eq. (13) are different from that of Ref. and correspond to the definition used in . The CIs (13) have dimensionality $`[\text{volume}]^{l1}`$, while in Ref. CIs are dimensionless.
$$b_l(V,T)=\frac{1}{l!V[g\varphi (T;m)]^l}d^{3l}rU_l(𝐫_1,𝐫_2,\mathrm{}𝐫_l).$$
(13)
It is easy to see that
$$b_11.$$
(14)
Substituting the expression (9) for the function $`W_2`$ into Eq. (3) one gets the two-particle partition function expressed via the CIs
$$Z(V,T,2)=g^2\left[\frac{1}{2}(b_1(V,T)\varphi (T;m)V)^2+b_2(V,T)\varphi ^2(T;m)V\right].$$
(15)
For arbitrary $`N`$ the expression of the partition function via $`b_l`$ reads
$$Z(V,T,N)=\underset{\{m_l\}}{}\underset{l=1}{\overset{N}{}}\frac{1}{m_l!}(b_l(V,T)\left[g\varphi (T;m)\right]^lV)^{m_l},$$
(16)
where the sum runs over all sets of nonnegative integer numbers $`\{m_l\}`$ satisfying the condition
$$\underset{l=1}{\overset{N}{}}lm_l=N.$$
(17)
Introducing the absolute activity , which in the relativistic case takes the form
$$zg\varphi (T;m)\mathrm{exp}\left(\frac{\mu }{T}\right),$$
(18)
where $`\mu `$ is the chemical potential, the grand canonical partition function has the form
$$𝒵(V,T,\mu )=\underset{N=1}{\overset{\mathrm{}}{}}\mathrm{exp}\left(\frac{\mu N}{T}\right)Z(V,T,N)=\mathrm{exp}\left(V\underset{l=1}{\overset{\mathrm{}}{}}b_l(V,T)z^l\right).$$
(19)
From Eq. (19) one can find the cluster expansion of the pressure and the particle density:
$`p(T,\mu )`$ $`=`$ $`T\underset{V\mathrm{}}{lim}{\displaystyle \frac{\mathrm{log}𝒵(V,T,\mu )}{V}}=T{\displaystyle \underset{l=1}{\overset{\mathrm{}}{}}}b_l(T)z^l,`$ (20)
$`n(T,\mu )`$ $`=`$ $`\underset{V\mathrm{}}{lim}{\displaystyle \frac{T}{V}}{\displaystyle \frac{\mathrm{log}𝒵(V,T,\mu )}{\mu }}={\displaystyle \underset{l=1}{\overset{\mathrm{}}{}}}lb_l(T)z^l,`$ (21)
where
$$b_l(T)\underset{V\mathrm{}}{lim}b_l(V,T).$$
(22)
Substituting the particle density (21) into the virial expansion Sometimes the term ’virial expansion’ is used in place of ’cluster expansion’ . We prefer to use the standard terminology : ’cluster expansion’ for the expansion in powers of the activity and ’virial expansion’ for that in powers of the particle density. for the pressure
$$p(T,n)=T\underset{i=1}{\overset{\mathrm{}}{}}a_in^i$$
(23)
and equating the coefficient of each power of $`z`$ with Eq.(20) one obtains the following expressions for the virial coefficients in terms of the CIs:
$`a_1`$ $`=`$ $`1`$ (24)
$`a_2`$ $`=`$ $`b_2`$ (25)
$`a_3`$ $`=`$ $`4b_2^22b_3`$ (26)
$`a_4`$ $`=`$ $`20b_2^3+18b_2b_33b_4`$ (27)
(28)
Let us represent the CIs as a sum of two terms:
$$b_l=b_l^{(0)}+b_l^{(i)},l>1,$$
(29)
where $`b_l^{(0)}`$ are the CIs for the IG and $`b_l^{(i)}`$ appear due to the particle interaction. In the classical (Boltzmann) gas one obtains $`b_l^{(0)}=0`$ for all $`l>1`$. In the quantum case, $`b_l^{(0)}`$ are nonzero due to Bose (Fermi) effects and can be easily found for arbitrary $`l`$. For noninteracting particles the logarithm of the expression (19) should coincide with the well-known expression for the logarithm of the ideal gas grand canonical partition function
$$\mathrm{log}𝒵^{(0)}(V,T,\mu )=\pm gV\frac{d^3p}{(2\pi )^3}\mathrm{log}\left[1\pm \mathrm{exp}\left(\frac{\mu \sqrt{𝐩^2+m^2}}{T}\right)\right]$$
(30)
(the upper (lower) sign corresponds to Fermi-Dirac (Bose-Einstein) statistics). One can expand the logarithm in the integrand and perform the integration
$`\mathrm{log}𝒵^{(0)}(V,T,\mu )`$ $`=`$ $`gV{\displaystyle \underset{l=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{(1)^{l+1}}{l}}{\displaystyle \frac{d^3p}{(2\pi )^3}\mathrm{exp}\left(\frac{l\left(\mu \sqrt{𝐩^2+m^2}\right)}{T}\right)}`$ (31)
$`=`$ $`gV{\displaystyle \underset{l=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{(1)^{l+1}}{l}}\mathrm{exp}\left({\displaystyle \frac{l\mu }{T}}\right)\varphi (T/l;m).`$ (32)
Comparing the last expression with Eq. (19) gives
$$b_l^{(0)}=\frac{(1)^{l+1}}{lg^{l1}}\frac{\varphi (T/l;m)}{[\varphi (T;m)]^l}.$$
(33)
In the nonrelativistic limit Eq.(33) is reduced to
$$b_l^{(0)}=(1)^{l+1}l^{5/2}\left[\frac{\lambda ^3}{g}\right]^{l1},$$
(34)
where $`\lambda `$ is the thermal wave length
$$\lambda =\sqrt{\frac{2\pi }{mT}}.$$
(35)
The expression (34) coincides for $`I_0=0`$ with the corresponding formulae of Ref. (up to the dimensional factor $`\lambda ^{3(l1)}`$, because of different normalization in Eq. (13)).
Using Eqs. (15) and (29) one can express $`b_2^{(i)}`$ via differences of the two-particle partition functions for real and ideal gases:
$$b_2^{(i)}=\frac{Z(V,T,2)Z^{(0)}(V,T,2)}{V[g\varphi (T;m)]^2}$$
(36)
Let us calculate $`Z(V,T,2)`$. A complete set of the orthonormal state vectors in the two particle system can be constructed from the following wave functions
$$|\alpha |𝐏,\stackrel{~}{\alpha }=\frac{e^{i(𝐏,𝐑)}}{\sqrt{V}}|\stackrel{~}{\alpha },$$
(37)
where $`𝐏`$ is the total momentum of the system, $`𝐑`$ is the radius-vector of its center of mass, and $`|\stackrel{~}{\alpha }`$ form a complete set of orthonormal state vectors of the system in the center of mass frame (c.m.f.) satisfying the Schrödinger equation
$$H|\stackrel{~}{\alpha }=\epsilon _{\stackrel{~}{\alpha }}|\stackrel{~}{\alpha }$$
(38)
with the normalization condition
$$\stackrel{~}{\alpha }^{}|\stackrel{~}{\alpha }=\delta _{\stackrel{~}{\alpha }^{}\stackrel{~}{\alpha }}.$$
(39)
The wave function (37) thus satisfies the following equations
$$H|𝐏,\stackrel{~}{\alpha }=\sqrt{𝐏^2+\epsilon _{\stackrel{~}{\alpha }}^2}|𝐏,\stackrel{~}{\alpha },$$
(40)
$$𝐏^{},\stackrel{~}{\alpha }^{}|\stackrel{~}{𝐏,\alpha }=\delta _{\stackrel{~}{\alpha }^{}\stackrel{~}{\alpha }}\delta _{𝐏^{}𝐏}.$$
(41)
The expression for $`Z(V,T,2)`$ in terms of the introduced wave functions has the form
$$Z(V,T,2)=\underset{𝐏,\stackrel{~}{\alpha }}{}𝐏,\stackrel{~}{\alpha }|\mathrm{exp}\left(\frac{H}{T}\right)|𝐏,\stackrel{~}{\alpha }=\underset{𝐏,\stackrel{~}{\alpha }}{}\mathrm{exp}\left(\frac{\sqrt{𝐏^2+\epsilon _{\stackrel{~}{\alpha }}^2}}{T}\right).$$
(42)
In the thermodynamical limit $`V\mathrm{}`$ the summation over $`𝐏`$ can be replaced by the integration and one finds
$$Z(V,T,2)=\underset{\stackrel{~}{\alpha }}{}V\frac{d^3P}{(2\pi )^3}\mathrm{exp}\left(\frac{\sqrt{𝐏^2+\epsilon _{\stackrel{~}{\alpha }}^2}}{T}\right)=V\underset{\stackrel{~}{\alpha }}{}\varphi (T;\epsilon _{\stackrel{~}{\alpha }}).$$
(43)
The states of two spinless particles in their c.m.f. can be enumerated by the following quantum numbers: the radial momentum $`q`$ (or, alternatively, the energy in c.m.f. $`\epsilon (q)=\sqrt{q^2+m^2})`$, the orbital angular momentum $`L`$, its projection $`m_L`$, the total isospin $`I`$ and its projection $`t_I`$, e.g. $`\stackrel{~}{\alpha }=(q,L,m_L,I,t_I)`$. Eq.(43) can be rewritten explicitly <sup>§</sup><sup>§</sup>§We assume that the particles do not form bound states with energy $`\epsilon <2m`$.
$$Z(V,T,2)=V\underset{I=0}{\overset{2I_0}{}}\underset{t_I=I}{\overset{I}{}}\underset{L}{}^{}\underset{m_L=L}{\overset{L}{}}_0^{\mathrm{}}𝑑qg_{Lm_LIt_I}(q)\varphi (T;\epsilon (q)),$$
(44)
where $`g_{Lm_LIt_I}(q)`$ is the density of states with the given set of quantum numbers. The sum $`^{}`$ extends over only those values of $`L`$ that satisfy the symmetry properties of the wave function. For spinless bosons it takes even value if the isospin part of the wave function is symmetric and odd values if it is antisymmetric. In the case of integer isospin particles like pions this means In the case of half-odd isospin scalar bosons we would have the opposite rule: even $`L`$ for odd $`I`$ and odd $`L`$ for even $`I`$.
$$L=\{\begin{array}{cc}0,2,4,6,\mathrm{}\hfill & \text{for even }I\hfill \\ 1,3,5,7,\mathrm{}\hfill & \text{for odd }I\hfill \end{array},$$
(45)
Substituting Eq.(44) into (36) one gets the following expression for the second CI
$$b_2^{(i)}=\frac{1}{[g\varphi (T;m)]^2}\underset{I=0}{\overset{2I_0}{}}\underset{t_I=I}{\overset{I}{}}\underset{L}{}^{}\underset{m_L=L}{\overset{L}{}}_0^{\mathrm{}}𝑑q(g_{Lm_LIt_I}(q)g_{Lm_LIt_I}^{(0)}(q))\varphi (T;\epsilon (q)),$$
(46)
where $`g_{Lm_LIt_I}^{(0)}`$ is the state density for the IG. The difference $`g_{Lm_LIt_I}(q)g_{Lm_LIt_I}^{(0)}(q)`$ in the thermodynamical limit can be expressed via phase shifts of two-particle scattering $`\delta _{Lm_LIt_I}(q)`$ :
$$g_{Lm_LIt_I}(q)g_{Lm_LIt_I}^{(0)}(q)=\frac{1}{\pi }\frac{d\delta _{Lm_LIt_I}(q)}{dq}.$$
(47)
Using the expression for $`\varphi (T;m)`$ one gets
$$b_2^{(i)}=\frac{2\pi }{m^4T[gK_2(m/T)]^2}\underset{I=0}{\overset{2I_0}{}}\underset{t_I=I}{\overset{I}{}}\underset{L}{}^{}\underset{m_L=L}{\overset{L}{}}_0^{\mathrm{}}𝑑q\epsilon ^2(q)\frac{d\delta _{Lm_LIt_I}(q)}{dq}K_2(\epsilon (q)/T).$$
(48)
In the case of hadron gas the phase shift does not depend on the angular momentum projection $`m_L`$ (no external fields) and the isospin projection $`t_I`$ (if only strong interactions are taken into account). This simplifies the last formula:
$$b_2^{(i)}=\frac{2\pi }{m^4T[gK_2(m/T)]^2}\underset{I=0}{\overset{2I_0}{}}\underset{L}{}^{}(2I+1)(2L+1)_0^{\mathrm{}}𝑑q\epsilon ^2(q)\frac{d\delta _{LI}(q)}{dq}K_2\left(\frac{\epsilon (q)}{T}\right).$$
(49)
Performing a partial integration and taking into account the properties of the Bessel functions, one gets
$$b_2^{(i)}=\frac{2\pi }{m^4T^2[gK_2(m/T)]^2}\underset{I=0}{\overset{2I_0}{}}\underset{L}{}^{}(2I+1)(2L+1)_{2m_\pi }^{\mathrm{}}𝑑\epsilon \epsilon ^2\delta _{L,I}(\epsilon )K_1\left(\frac{\epsilon }{T}\right).$$
(50)
In the nonrelativistic limit the formula (49) is reduced to
$$b_2^{(i)}=\frac{2\sqrt{2}}{\pi g^2}\lambda ^3\underset{I=0}{\overset{2I_0}{}}\underset{L}{}^{}(2I+1)(2L+1)_0^{\mathrm{}}𝑑q\frac{d\delta _{LI}(q)}{dq}\mathrm{exp}\left(\frac{q^2}{mT}\right),$$
(51)
which again at $`I_0=0`$ coincides with the corresponding formulae of Refs. , up to the factor $`\lambda ^3`$.
## III Hard core repulsion
The hard core repulsion plays an important role in the phenomenological description of the $`\pi \pi `$-scattering: the phase shift data for the isospin state $`I=2`$ can be successfully described assuming hard core repulsion between two particles . The best fit of the phase shift in $`S_0`$-state ($`I=L=0`$) can be obtained by assuming hard core repulsion in addition to resonance attraction . Hence we start our analysis from applying Eqs.(4951) to hard sphere model. Relativistic consideration of a hard sphere model by no means can be consistent. Still, as far as this model describes experimental data on low energy $`\pi \pi `$-scattering, we find it phenomenologically satisfactory.
The radial part of the wave function in the c.m.f. of two particles interacting by hard core potential in the state with orbital momentum $`L`$ and radial momentum $`q`$ can be represented in the following way
$$\varphi (r)=\{\begin{array}{cc}0\hfill & \text{for }rr_0\hfill \\ C(\mathrm{cos}\delta _Lj_L(qr)+\mathrm{sin}\delta _Ly_L(qr))\hfill & \text{for }r>r_0\hfill \end{array},$$
(52)
where $`r`$ is the distance between particle centers, $`r_0`$ is the minimal admitted value for $`r`$ (that is the doubled radius of the particle considered as a hard sphere), $`j_n(z)`$ and $`y_n(z)`$ are spherical Bessel functions, $`C`$ is the normalization constant and $`\delta _L`$ is fixed by the condition
$$\mathrm{cos}\delta _Lj_L(qr_0)+\mathrm{sin}\delta _Ly_L(qr_0)=0.$$
(53)
Using asymptotic properties of the spherical Bessel functions, it is easy to see that $`\delta _L`$ has a meaning of the phase shift describing scattering of two hard spheres:
$$\varphi (r)\frac{\mathrm{sin}\left(qr\frac{l\pi }{2}+\delta _L\right)}{qr},\text{ }r\mathrm{}.$$
(54)
The derivative of the phase shift is found to be
$$\frac{d\delta _L}{dq}=\frac{d}{dq}\mathrm{arctan}\left(\frac{j_L(qr_0)}{y_L(qr_0)}\right)=\frac{r_0}{(qr_0)^2[j_L^2(qr_0)+y_L^2(qr_0)]},$$
(55)
where the formula for the Wronskian
$$W(j_L(z),y_L(z))=z^2$$
(56)
has been used.
Let us introduce the functions
$$\kappa ^\pm (z)=\underset{L}{}^{}\frac{1}{z^2[j_L^2(z)+y_L^2(z)]},$$
(57)
where the sum $`_{L}^{}{}_{}{}^{}`$ runs over either even (superscript ‘$`+`$’) or odd (superscript ‘$``$’) nonnegative numbers.
Expanding $`\kappa ^\pm `$ around zero, one gets
$`\kappa ^+(z)`$ $``$ $`1+{\displaystyle \frac{5}{9}}z^4+O(z^6),`$ (58)
$`\kappa ^{}(z)`$ $``$ $`3z^23z^4+{\displaystyle \frac{682}{225}}z^6+O(z^8).`$ (59)
It has been checked numerically that the asymptotic behavior of $`\kappa ^\pm (z)`$ at large $`z`$ with a high accuracy can be presented by the formula
$$\kappa ^\pm (z)\frac{1}{3}z^2+\frac{\pi }{4}z+\frac{1}{3}+O(z^1).$$
(60)
The CI (49) for the case of hard sphere model can be represented in the form<sup>\**</sup><sup>\**</sup>\** In the simplest version of the hard sphere model when the core radius $`r_0`$ does not depend on $`I`$, the formula (61) could be reduced to $`b_2^{(i)}=b_2^\pm (r_0,T,m)`$. However, in the case of realistic $`\pi \pi `$ interaction every isospin state has its own $`r_0(I)`$ .
$$b_2^{(i)}=\frac{1}{g^2}\underset{I=0}{\overset{2I_0}{}}(2I+1)b_2^\pm (r_0,T,m),$$
(61)
where superscript ‘$`+`$’(‘$``$’) corresponds to even (odd) values<sup>††</sup><sup>††</sup>††Again, for the case of half-odd isospin particles the opposite rule would be valid. of $`I`$ and $`b_2^\pm (r_0)`$ is expressed via the function $`\kappa ^\pm (z)`$
$$b_2^\pm =\frac{2\pi r_0}{m^4T[K_2(m/T)]^2}_0^{\mathrm{}}𝑑q[\epsilon (q)]^2\kappa ^\pm (qr_0)K_2\left(\frac{\epsilon (q)}{T}\right).$$
(62)
In the nonrelativistic approximation the expression for $`b_2^\pm (r_0)`$ is reduced to
$$b_2^\pm (r_0,T,m)=\frac{2\sqrt{2}}{\pi }\lambda ^3r_0_0^{\mathrm{}}𝑑q\kappa ^\pm (qr_0)\mathrm{exp}\left(\frac{q^2}{mT}\right).$$
(63)
Substituting Eq. (58) into the last formula one gets the nonrelativistic expression for $`b_2^\pm (r_0,T,m)`$ at small $`r_0`$ ($`r_0<<\lambda `$)
$`b_2^+(r_0,T,m)`$ $``$ $`2\lambda ^2r_0\left(1+{\displaystyle \frac{5}{3}}\pi ^2(r_0/\lambda )^4+O\left[(r_0/\lambda )^6\right]\right),`$ (64)
$`b_2^{}(r_0,T,m)`$ $``$ $`6\pi r_0^3\left(13\pi (r_0/\lambda )^2+{\displaystyle \frac{682}{45}}\pi ^2(r_0/\lambda )^4+O\left[(r_0/\lambda )^6\right]\right),`$ (65)
in the opposite case $`r_0>>\lambda `$ the nonrelativistic expression for $`b_2^\pm (r_0,T,m)`$ can be obtained using Eq.(60):
$$b_2^\pm (r_0,T,m)=\frac{2}{3}\pi r_0^3\left(1+\frac{3\sqrt{2}}{4}\frac{\lambda }{r_0}+\frac{3}{2\pi }\frac{\lambda ^2}{r_0^2}+O\left[\left(\frac{\lambda }{r_0}\right)^3\right]\right)$$
(66)
The pion thermal wave length at the temperature $`T=50÷200`$ MeV ranges between $`3÷6`$ fm. From Eq. (66) one sees that the classical formula
$$b_2^\pm (r_0,T,m)\frac{2}{3}\pi r_0^3$$
(67)
(the particle volume multiplied by $`4`$) would give a reasonable approximation only at unrealistically large hard core radius $`r_050`$ fm. The hard core radii found from the $`\pi \pi `$-scattering are much smaller: $`r_0=0.60`$ fm in the $`S_0`$ state and $`r_0=0.17`$ fm at $`I=2`$ . (No evidence of hard core repulsion was found in $`P_1`$-state (I=L=1)). In this case the value of $`b_2^+(r_0,T,m)`$ can be estimated from formula (64). The results are presented in Fig.1. It is seen that in contrast to the classical case the quantum treatment leads to a rather strong dependence of the second CI on the temperature (approximately proportional to $`1/T`$) even in the nonrelativistic approximation. Numerical calculations show that relativistic effects make this dependence even stronger and essentially reduce at high temperature the CI with respect to its nonrelativistic value (see Fig. 1). Both relativistic and nonrelativistic quantum formulae give a much ($`1÷2`$ orders of magnitude) larger value than those given by the classical formula (67): $`0.45`$ fm<sup>3</sup> and $`0.01`$ fm<sup>3</sup> for $`r_0=0.60`$ fm $`r_0=0.17`$ fm, respectively.
As can be seen from Fig.1, the relativistic effects cannot be ignored even at relatively low temperatures. Therefore, only the relativistic formula (62) is used in the following calculations.
## IV Resonance attraction
The phase shifts of $`\pi \pi `$ elastic scattering can be approximately described by assuming that the attractive parts of the interaction appear due to the propagation of resonances in the $`s`$-channel of the reaction (see and references therein).
The distinctive feature of the resonance interaction is the rapid growth of the phase shift by $`\pi `$ radian in the vicinity of the pion momentum $`q_r`$, which is related to the resonance mass $`M_r`$:
$$M_r=2\sqrt{q_r^2+m^2}.$$
(68)
In the limit $`\mathrm{\Gamma }_r0`$ ($`\mathrm{\Gamma }_r`$ is the resonance width) the derivative of the phase shift can be approximated by the Dirac delta-function:
$$\frac{d\delta _{L,I}(q)}{dq}\pi \delta (qq_r).$$
(69)
In this approximation, which we will call the ’narrow resonance approximation’ (NRA), the expression (49) for $`b_2^{(i)}`$ yields:
$$b_2^{(i)}\frac{1}{g^2}\underset{r}{}(2I_r+1)(2L_r+1)\frac{\varphi (T;M_r)}{[\varphi (T;m)]^2},$$
(70)
where $`I_r`$, $`L_r`$ and $`M_r`$ are, respectively, the resonance’s isospin, spin and mass, with the index $`r`$ running over all resonances in the two-pion system.
Eq.(70) allows to rewrite the expression for the grand canonical partition function (19) in the following form
$$𝒵(V,T,\mu )=\mathrm{exp}\left[V\left(\underset{l=1}{\overset{\mathrm{}}{}}b_l^{(0)}z^l+\underset{r}{}z_r\right)\right],$$
(71)
where
$$z_r=\varphi (T;M_r)\mathrm{exp}\left(\frac{2\mu }{T}\right)$$
(72)
is the absolute activity of the resonance $`r`$ with degeneration factor $`g_r=(2I_r+1)(2L_r+1)`$. The expression (71) is nothing else than the partition function for a mixture of ideal gases of pions and two-pion resonances<sup>‡‡</sup><sup>‡‡</sup>‡‡ The fact that the resonance gases are classical is an artifact of the second cluster approximation. The quantum correction would appear from the 4-th and higher cluster integrals for the interacting pions.. This recovers the well known result of Ref. that narrow resonances contribute to the partition function as an ideal gas of stable particles. The quantitative criterion for an applicability of the NRA was found to be
$$\mathrm{\Gamma }_r<<T.$$
(73)
The resonances appearing in $`\pi \pi `$-scattering do not satisfy this criterion: most of them ($`\rho (770)`$, $`f_0(980)`$, $`f_2(1270)`$ , $`\rho _3(1690)`$) have widths comparable with a typical temperature of the hadron gas and the width of $`f_0(4001200)`$ (known also as the $`\sigma `$) is a few times larger then the temperature. Therefore, it is necessary to take into account the finite width of the resonances.
The scalar-isoscalar resonances contribution to the $`\pi \pi `$-phase shift in the $`S_0`$ state can be parametrized in the following way :
$$\mathrm{tan}\delta _r(q)=\frac{q}{q_r}\frac{M_r^2}{\epsilon (q)}\frac{\mathrm{\Gamma }_r}{M_r^2\epsilon ^2(q)},r=\sigma ,f_0(980).$$
(74)
For parametrization of nonzero (iso-)spin resonances we shall use the following formula
$$\mathrm{tan}\delta _r(q)=\left(\frac{q}{q_r}\right)^{2L_r+1}\frac{M_rx_r\mathrm{\Gamma }_r}{M_r^2\epsilon ^2(q)}\frac{D_{L_r}(q_rR_r)}{D_{L_r}(qR_r)},r=\rho (770),f_2(1270),\rho _3(1690),$$
(75)
where $`x_r`$ is the inelasticity, i.e. the decay fraction of the resonance into two pions, $`R_r`$ is the so-called interaction radius and the functions $`D_L(z)`$ are given by the formulae
$`D_1(z)`$ $`=`$ $`1+z^2`$ (76)
$`D_2(z)`$ $`=`$ $`9+3z^2+z^4`$ (77)
$`D_3(z)`$ $`=`$ $`225+45z^2+6z^4+z^6.`$ (78)
The resonance parameters are given in the Table I.
It is easy to see that if a resonance lies far from the threshold:
$$M_r2m>>\mathrm{\Gamma }_r$$
(79)
both formulae are reduced to
$$\mathrm{tan}\delta _r(q)\frac{\mathrm{\Gamma }_r/2}{M_r\epsilon }$$
(80)
In this case the activity of the resonance can be represented in the form
$$\underset{r}{}z_r=\underset{I=0}{\overset{2I_0}{}}\underset{L}{}^{}_{2m}^{\mathrm{}}𝑑\epsilon \zeta (\epsilon )(2I+1)(2L+1)\varphi (T;\epsilon )\mathrm{exp}\left(\frac{2\mu }{T}\right),$$
(81)
where the resonance profile function is given by the Breit-Wigner formula:
$$\zeta (\epsilon )=\frac{1}{2\pi }\frac{\mathrm{\Gamma }_r}{(\epsilon M_r)^2+(\mathrm{\Gamma }_r/2)^2}$$
(82)
From this we conclude that the procedure of Ref. where the profile function was postulated to be
$$\zeta (\epsilon )=\frac{\xi \mathrm{\Gamma }_r}{(\epsilon M_r)^2+(\mathrm{\Gamma }_r/2)^2}$$
(83)
with normalization constant $`\xi `$ fixed by the condition
$$_{2m}^{\mathrm{}}𝑑\epsilon \frac{\xi \mathrm{\Gamma }_r}{(\epsilon M_r)^2+(\mathrm{\Gamma }_r/2)^2}=1$$
(84)
becomes valid in the limit (79). The $`\sigma `$-resonance obviously does not satisfy this condition, even for the $`\rho (770)`$ the difference $`M_r2m`$ is only about $`3`$ times larger then the width. We have calculated the contributions of these two resonances into the second cluster integral $`b_2^{(i)}`$ using the parametrizations (74) and (75) and compare them with the corresponding approximate values found in the framework of the procedure of Ref. :
$$b_2^{(i)}=\frac{1}{m^4T[gK_2(m/T)]^2}\underset{r}{}_0^{\mathrm{}}𝑑q\epsilon ^2(q)\frac{\xi \mathrm{\Gamma }_r}{(\epsilon M_r)^2+(\mathrm{\Gamma }_r/2)^2}\frac{d\delta _{LI}(q)}{dq}K_2\left(\frac{\epsilon (q)}{T}\right).$$
(85)
and in the NRA (70). The results are shown in Figs. 2 and 3. As can be seen from Fig.2, for $`\sigma `$-resonance the both approximations essentially underestimate the CI. It is interesting to mention that at $`T>150`$ MeV the formula of Ref. gives slightly worse result than even that of the NRA. In the case of the $`\rho `$-resonance this formula systematically overestimates the CI in contrast to the $`\sigma `$ case. The role of the resonance width becomes small at high temperatures and all three formulae give comparable results in both ($`\sigma `$ and $`\rho `$) cases.
## V Interacting pion gas
Both type of interaction: hard core repulsion and resonance attraction are present in the pion gas. The phase shift for $`\pi \pi `$-scattering in the $`S_0`$ state at the center of mass energy below $`1`$ GeV can be represented as a sum of three terms :
$$\delta _{00}(q)=\delta _\sigma (q)+\delta _{f_0}(q)+\delta _{BG}(q).$$
(86)
The background term $`\delta _{BG}`$ is related to the hard core repulsion
$$\delta _{BG}(q)=r_0q,r_0=3.03\text{ GeV}^1$$
(87)
and two first terms describe attraction due to the resonances $`\sigma `$ and $`f_0`$ and are parametrized by the formula (74). The contributions of the $`S_0`$ state to the CI are shown in Fig.4. The attractive part is larger in absolute value than the repulsive part, so that total contribution of the $`S_0`$ state is positive.
The interaction in the $`S_2`$ state has a purely repulsive nature (no exotic resonances with isospin $`I=2`$ have been found). The phase shift can be successfully fitted by the hard core formula:
$$\delta _{02}(q)=r_0q,r_0=0.87\text{ GeV}^1.$$
(88)
As it is seen from Fig.4 the absolute value of the negative contribution of $`S_2`$ state into the CI is slightly larger than that of the positive contribution of the $`S_0`$ state, so that these two quantities almost cancel each other. The resulting contribution of the $`S`$-state into CI is negative and a few times smaller than those of the $`S_0`$\- and $`S_2`$-states separately. Due to the small value of the total $`S`$-state contribution into CI the $`P_1`$ state becomes important already at relatively low temperature.
The phase shifts of $`\pi \pi `$-scattering in $`P_1`$, $`D_2`$ and $`F_1`$ states can be parametrized by the formula (75) <sup>\**</sup><sup>\**</sup>\**The background and higher pole terms which are present in the formulae of Ref. were found to give a negligible contribution to the CI and are dropped in the present consideration.. The results are presented in Fig.5. At small temperatures $`T<80`$ MeV both the $`S`$\- and $`P`$-wave give comparable contributions to the CI. At higher temperatures, the $`P`$-wave dominates. The $`D`$\- and $`F`$-waves add small corrections to the CI at $`T>140`$ MeV. The contribution of higher waves is assumed to be negligible.
It should be mentioned that at very low temperatures, $`T<30`$ MeV, the total CI becomes negative, in agreement with the results of Ref. for the isotopically symmetric pion gas, while at high temperatures attraction dominates over repulsion.
The exact CI is also compared in Fig.6 to various approximations widely used for the hadron gas analysis. It is seen that ignoring repulsion between pions overestimates the CI by more than $`35`$%. On the other hand, the NRA underestimates the CI by at least $`20`$%. At low temperatures $`T<120`$ MeV both approximations become completely unreasonable. If one ignores both the finite resonance width and the repulsion (it corresponds to the ideal gas of pions and resonances) these two errors partially cancel each other. The simplest approximation, surprisingly enough, appears to give better results than the both more complicated ones. (There remains, however, discrepancy up to about $`15`$% at high temperatures). This means that both effects, the repulsion and the finite resonance width, should be taken into account simultaneously. Including either of these effects without another one increases rather than decreases the numerical errors with respect to the simplest IG model of pion and two-pion resonances.
Comparing the CI $`b_2^{(i)}`$ with the ideal gas CI $`b_2^{(0)}`$ (see Fig. 8) one observes that the interactions give essential contribution to the CI already at $`T=70`$ MeV. At $`T>150`$ MeV the interaction part $`b_2^{(i)}`$ clearly dominates over Bose effects related to $`b_2^{(0)}`$.
To estimate the influence of the second CI on the thermodynamical properties of the pion gas we have calculated the particle density in the second cluster approximation
$$n=n_0+2b_2^{(i)}z^2,$$
(89)
where
$$n_0=\underset{l=1}{\overset{\mathrm{}}{}}lb_l^{(0)}z^l=\frac{g}{2\pi ^2}_0^{\mathrm{}}𝑑pp^2\frac{1}{\mathrm{exp}\left(\frac{\sqrt{p^2+m^2}\mu }{T}\right)1}$$
(90)
is the density of the ideal pion gas (without resonances). The calculations were done assuming that the chemical equilibrium, $`\mu =0`$, is reached.
The temperature dependence of the ratio $`n/n_0`$ is shown in Fig.7. It can be seen that all approximations give consistent results up to $`5÷15\%`$. A rather small errors is explained by the fact that at low temperatures ($`T<120`$ MeV), where ignoring either the finite resonance width or the repulsion between pions leads to huge errors in the value of cluster integral, the activity of the equilibrium pion gas is small and the contribution of the second term of the cluster expansion into the value of particle density is not important. On the other hand, at large temperatures, where the second term becomes comparable with the first one, the both approximation provide more exact value of the cluster integral. Again, the ideal gas model provides the best approximation at all temperatures, except very large ones $`T>180`$ MeV.
This conclussion is close to that of Ref., where it was pointed out that the interacting pion gas in the second cluster approximation only slightly differs from the ideal gas of pions and $`\rho `$-mesons due to the nearly exact cancellation of the contributions from S-wave attractive and repulsive channels. That is the repulsive interactions and the contribution of the broad $`\sigma `$-resonans can be dropped simultaneously. The aim of our further consideration is to take properly into account the hard-core repulsive interactions. In this case the $`\sigma `$-meson contribution must be retained.
It is seen from Fig.7 that the contribution of the second term of cluster expansion to the particle density is comparable to that of the ideal gas. There is no reason to expect that the higher terms are negligible. The purpose of the next section is to go beyond the second cluster approximation.
## VI Van der Waals equation
Taking into account the attractive parts of higher cluster terms is straightforward: assuming that the attractive interaction of three and more pions is dominated by resonance interaction (as it was in the two pion case) we just add the activities of all the lightest pion resonances to the grand canonical partition function logarithm:
$$\mathrm{log}𝒵(V,T,\mu )=\mathrm{log}𝒵^{(0)}(V,T,\mu )+V\underset{r}{}z_r,$$
(91)
where the first term in the right hand side is the partition function of the ideal pion gas and the sum in the second term runs over not only the two pion resonances (Table I), but also includes the resonances decaying into three and more pions (Table II). For the activity of each resonance species we use the following expression
$$z_r=_{N_rm}^{\mathrm{}}𝑑\epsilon \zeta (\epsilon )g_r\varphi (T;\epsilon )\mathrm{exp}\left(\frac{\mu _r}{T}\right).$$
(92)
The chemical potential $`\mu _r`$ of a resonance decaying into $`N_r`$ pions is proportional to the pion chemical potential:
$$\mu _r=N_r\mu $$
(93)
In our calculations we assume chemical equilibrium: $`\mu _r=\mu =0`$. For the two-pion resonances from Table I we put
$$\zeta (\epsilon )=\frac{1}{\pi }\frac{d\delta }{d\epsilon }$$
(94)
and use the parametrization (74) and (75), so that the contribution of two pion resonances is reduced to the $`b_2^{(a)}z^2`$, where the $`b_2^{(a)}`$ is an attractive part of the CI $`b_2^{(i)}`$ shown in Fig. 8.
For the resonances from Table II we use the Breit-Wigner profile function (83) with the normalization (84) <sup>\*†</sup><sup>\*†</sup>\*†The formula of Ref. does not lead to large error because of the dominating contribution comes from the narrow resonance $`\omega (782)`$, for which both creteria (73) and (79) are fulfilled. On the other hand, lack of detailed experimental information on the phase shifts in the vicinity on the broad $`\pi \rho `$\- and $`\pi \sigma `$-resonances as well as large uncertainties in their masses, width and decay fractions make impossible and useless the application of the more exact formula.. In the NRA the expression (92) is reduced to Eq. (72).
It follows from Eq. (91) that the pressure and the pion density are calculated from the ideal gas model for pion and pion resonances with finite width (that is the repulsive interactions are ignored):
$`p(T,\mu )`$ $`=`$ $`p_0(T,\mu )+T{\displaystyle \underset{r}{}}z_r=p_0(T,\mu )+{\displaystyle \underset{r}{}}p_r(T,\mu _r)`$ (95)
$`n(T,\mu )`$ $`=`$ $`n_0(T,\mu )+{\displaystyle \underset{r}{}}N_rz_r=n_0(T,\mu )+{\displaystyle \underset{r}{}}N_rn_r(T,\mu _r),`$ (96)
where the ideal pion gas pressure $`p_0`$ is given by the formula
$$p_0(T,\mu )=gT\frac{d^3p}{(2\pi )^3}\mathrm{log}\left[1\mathrm{exp}\left(\frac{\mu \sqrt{𝐩^2+m^2}}{T}\right)\right],$$
(97)
with the particle density of the ideal pion gas given by Eq. (90).
To take into account the hard core repulsion between pions and resonances we use the excluded volume Van der Waals model . In the framework of this model the pressure $`p^{VdW}(T,\mu )`$ of one-component gas of particles with the excluded-volume parameter $`v_0`$ can be found from the transcendental equation
$$p^{VdW}(T,\mu )=p^{id}(T,\mu v_0p^{VdW}(T,\mu )),$$
(98)
where $`p^{id}(T,\mu )`$ is the pressure of corresponding ideal gas. The particle density $`n^{VdW}(T,\mu )`$ is related to that of ideal gas by the expression
$$n^{VdW}(T,\mu )=\frac{n^{id}(T,\mu v_0p^{VdW}(T,\mu ))}{1+v_0n^{id}(T,\mu v_0p^{VdW}(T,\mu ))},$$
(99)
The above model can be straightforwardly generalized to a multi-component gas, if one assumes that all particle species have the same excluded-volume parameter. In our calculations we put it to be the same for the pions and pion resonances. The standard procedure of derivation of the Van der Waals equation in the statistical physics shows that the excluded-volume parameter is equal to the absolute value of the repulsive part of the second virial coefficient . Therefore, the excluded-volume parameter for the pions can be identified with the repulsive part of the CI $`b_2^{(i)}`$ (See Fig. 8):
$$v_0=\left|b_2^{(r)}\right|.$$
(100)
Hence, to find the pressure of the interacting pion gas in the framework of the Van der Waals excluded volume model we solve the transcendental equation
$`p^{VdW}(T,\mu )`$ $`=`$ $`p_0(T,\stackrel{~}{\mu })+{\displaystyle \underset{r}{}}p_r(T,\stackrel{~}{\mu }_r)`$ (101)
$`\stackrel{~}{\mu }`$ $`=`$ $`\mu v_0p^{VdW}(T,\mu ),`$ (102)
$`\stackrel{~}{\mu }_r`$ $`=`$ $`\mu _rv_0p^{VdW}(T,\mu ).`$ (103)
The particle density of the pions is found from
$$n^{VdW}(T,\mu )=\frac{n_0(T,\stackrel{~}{\mu })+_rN_rn_r(T,\stackrel{~}{\mu }_r)}{1+v_0\left[n_0(T,\stackrel{~}{\mu })+_rn_r(T,\stackrel{~}{\mu }_r)\right]}$$
(104)
The result of the calculation is shown in Fig. 9. It is seen that essential deviation from the second order cluster expansion take place at the temperatures $`T140`$ MeV. A comparison with the ideal gas model of pions and pion resonances shows that the effects of the hard-core repulsion are not cancelled by the effects of the finite resonance width. This leads to an essential (up to $`30`$%) suppression of the pion density with respect to the ideal gas of pions and pion resonances.
## VII Conclusion
A quantum mechanical formula for the second cluster integral for the gas of relativistic particles with hard-core interaction was derived and analyzed. In the nonrelativistic classical limit, this formula is reduced to the expression used in Refs.. In the quantum case, however, the value of the cluster integral appears to be much larger in magnitude than the corresponding classical value and, in contrast to the classical case, depends on the temperature even in the nonrelativistic limit. It has been demonstrated that the second cluster integral for the pion gas all reasonable temperatures is far away from the classical limit. Its repulsive part, which can be interpreted as proper particle volume, is an order of magnitude larger than it could be expected from the classical evaluation. It should be mentioned that not only quantum effects but also relativistic ones are important in the case of pion gas. Surprisingly, they essentially modify the proper pion volume even at relatively low temperatures $`T30`$ MeV.
The role of finite resonance width in the second cluster integral was studied. It was established that the widely used add hoc formula with the normalized Breit-Wigner resonance profile is unsuitable for broad resonances lying close to the threshold, the parametrization of the experimental phase shifts should be used instead. The most striking example of this kind is the $`\sigma `$-resonance. Our analysis shows that in the case of $`\sigma `$-resonance the calculations with the normalized Breit-Wigner profile can give even worse result than simple zero-width approximation.
At the second order of cluster expansion, due to the presence of broad resonances in the $`\pi \pi `$-system, the negative contribution of the hard core repulsion into the cluster integral almost canceled by positive contributions of finite resonance widths in a rather broad temperature range. Because of this fact, the thermodynamical properties of the interacting pion gas in the second cluster approximation appear to be quite similar to those of the ideal gas of pions and two-pion resonances: the error in the value of the particle density does not exceed a few percents. Surprisingly, the account for finite resonance widths without account for the hard core repulsion as well as consideration of the hard core repulsion when the resonance widths are neglected worsen rather than improve a simple ideal gas model of pions and zero-width pion resonances. Both effects should be either neglected or taken into account simultaneously.
This does not mean, however, that we can restrict ourselves to the simple ideal gas picture of pions and zero-width pion resonances at all temperatures. As it has been demonstrated, when the temperature is sufficiently high ($`T140`$ MeV), the pion density becomes so large that the cluster expansion cannot be truncated at the second order. In contrast to the second cluster approximation, an appreciable deviation from the ideal gas model is observed, when the higher order terms are taken into account. In the framework of Van der Waals excluded-volume model, the pion density appears to be up to $`30`$% lower than that of the ideal pion-resonance gas. Hence, at high particle densities the correct model of the pion gas must include all pion resonances and the resonance width as well as the repulsive interactions between the particles must be taken into account.
It should be emphasized that, if the model takes properly into account the hard core repulsion, there is no reason to drop the $`\sigma `$-resonance. It must be included into the model along with other resonances.
The developed in the present paper approach will be used for calculation of excluded volumes of other hadrons. This will allow us to study the influence of hard core repulsion on the properties of realistic hadron gas including (anti-)nucleons and strange particles by means of multicomponent Van der Waals equation . We expect essentially larger excluded volume effects for nucleons: preliminary calculations show that the proper volume of the nucleon is essentially (by the factor $`2÷2.5`$) larger than that of the pion. Thus the hard core repulsion may essentially modify the particle number ratios in comparison to widely used ideal resonance gas model.
###### Acknowledgements.
We thank K.Bugaev for helpful discussions and comments. We acknowledge the financial support of DAAD and DFG, Germany. The research described in this publication was made possible in part by Award No. UP1-2119 of the U.S. Civilian Research & Development Foundation for the Independent States of the Former Soviet Union (CRDF).
|
warning/0004/astro-ph0004079.html
|
ar5iv
|
text
|
# The sources of extended continuum emission towards Q0151+048A : The host galaxy and the Damped Ly𝛼 Absorber Based on observations made with the Nordic Optical Telescope, operated on the island of La Palma jointly by Denmark, Finland, Iceland, Norway, and Sweden, in the Spanish Observatorio del Roque de los Muchachos of the Instituto de Astrofisica de Canarias.
## 1 Introduction
The study of the galaxy population at high redshifts has progressed rapidly during the last decade. Through the Lyman-break technique hundreds of normal (i.e. not dominated by active galactic nuclei), star forming galaxies at z=2–4 have been detected and studied with imaging as well as spectroscopy (Steidel et al. 1996). These so called Lyman-Break Galaxies (LBGs) have star formation rates (SFRs) in the range 4–55h<sup>-2</sup> M yr<sup>-1</sup> for $`\mathrm{\Omega }`$=1.0 or 20 – 270 M yr<sup>-1</sup> for $`\mathrm{\Omega }`$=0.2 (Pettini et al. 1998). Also, via the study of the class of high column density QSO absorption lines systems known as Damped Ly$`\alpha `$ Absorbers (DLAs) a wealth of information on the early chemical evolution of galaxies at z=2–4 has been obtained (e.g. Lu et al. 1996). The DLAs are in general forming stars at a significantly lower rate than the LBGs (Møller & Warren 1998, Fynbo et al. 1999).
Independent information about the formation of the brightest galaxies comes from detailed studies of the stellar populations of present day bright cluster ellipticals. These populations seems to have formed early (z$`>`$2) in strong burst of star formation (Bower et al. 1992). Studies of the fundamental plane for elliptical and lenticular galaxies in rich clusters at intermediate redshifts also indicate early formation times (z$`>`$5 for $`\mathrm{\Omega }`$=1, Jørgensen et al. 1999), and the fundamental plane for field ellipticals at similar redshifts is consistent with being the same as in clusters (Treu et al. 1999a). Studies of the globular cluster populations of faint elliptical galaxies also indicate rather early formation times (z$`>`$1), whereas for bright cluster ellipticals the globular cluster populations do not strongly constrain the possible formation scenarios (Kissler-Patig et al. 1998). Furthermore, the presence of seemingly old stellar populations in elliptical galaxies at z$`>`$1 proves that at least some elliptical galaxies formed very early in strong bursts of star formation (Spinrad et al. 1997, Treu et al. 1999b, see also Jimenez et al. 1999). For first-rank ellipticals star formation rates as high as SFR$``$10<sup>3</sup> M yr<sup>-1</sup> would then be possible. A reason why such high star formation rates have not been detected in galaxies at high redshift may be that these galaxies are the hosts of powerful QSOs and hence are hidden by the light from the QSOs (e.g. Terlevich & Boyle 1993). Support for a connection between QSOs and bright elliptical galaxies comes from the fact that radio quiet QSOs as well as radio loud QSOs and radio galaxies at z=0.1-0.3 are hosted by galaxies for which the light profiles are best fit by de Vaucouleurs profiles indicating that they are early stages of massive ellipticals (McLure et al. 1999). There is increasing evidence that QSOs at redshifts z$``$2 are embedded in extended emission that is consistent with the presence of a stellar population in the QSO host galaxies. In the case of radio loud QSOs host galaxies have been detected in the optical and infrared by Lehnert et al. (1992) and Carballo et al. (1998), and in the case of radio quiet QSOs host galaxies have been detected in the optical and near infrared by Aretxaga et al. (1998a,b). There does not seem to be any systematic differences between the host galaxies of radio loud and radio quiet QSOs. Both populations of host galaxies are extremely bright, R$``$21–22, and have optical-to-infrared colours in the range R-K$``$3–5. However, measured polarisation of the light from some radio galaxies show that scattered QSO light can also contribute significantly to the observed extended emission (e.g. Cimatti et al. 1998).
In 1996 we performed a narrow band study of the z$`{}_{abs}{}^{}`$z<sub>em</sub> Damped Ly$`\alpha `$ Absorber (DLA, Wolfe et al. 1986) towards Q0151+048A using the 2.56-m Nordic Optical Telescope (NOT) (Fynbo et al. 1999). The main result of this study was the detection of extended Ly$`\alpha `$ emission from the DLA. The Ly$`\alpha `$ emission line had prior to this been detected in a spectroscopic study of Q0151+048A (Møller et al. 1998), but the large extended nature of the DLA absorber was quite unexpected. U band data, also from the 1996 run, hinted at the existence of an extended broad band object, but the signal–to–noise ratio of the object was low. We have therefore obtained deeper imaging of Q0151+048 in broad band U, B and I filters in order to confirm or reject our tentative detection, and to measure the extend and luminosity of the broad band source if real.
In Sect. 2 below we describe our new observations. In Sect. 3 we describe in detail the two independent methods we have used to search for extended objects close to the quasar. First we describe the image-deconvolution, where we used the Magain et al. (1998, hereafter MCS) algorithm, secondly we describe the direct PSF subtraction, and Sect. 4 we discuss our results.
In this paper we adopt H=100 h km s<sup>-1</sup> Mpc<sup>-1</sup>, $`\mathrm{\Omega }_m`$=1.0 and $`\mathrm{\Omega }_\mathrm{\Lambda }`$=0 unless otherwise stated.
## 2 Observations and Data Reduction
The observations were performed during two observing runs in September and October 1998 with HiRAC (High Resolution Adaptive Camera) on the 2.56 m Nordic Optical Telescope. The CCD used was a 2048<sup>2</sup> back-side illuminated thinned Loral with a pixel size of 0.1082 arcsec.
All 6 nights during which the observations were performed were photometric and with good seeing (median 0.7 arcsec FWHM in the I–band). Integration times in U were 1000-1500 sec in order to avoid the total noise to be dominated by readout noise. In I and B the integration times were 250-300 sec and 300-500 sec respectively. Between exposures the telescope was moved 2-4 arcsec to minimize the effects of bad pixels and fringing. The total integration times in each filter are given in Table 1.
Also observed were several standard star sequences from Landolt (1992) and photometric solutions were obtained for each filter. For counts given as electrons per second we derive zero-points in the Landolt system of 22.77, 25.13 and 24.48 for u, B and I respectively. All magnitudes subsequently quoted in this paper are on the AB system. For the U–band we determined the colour equation $`u=U+0.17(UB)`$ relating the instrumental magnitude $`u`$ to the standard Johnson $`U`$. There was substantial scatter around the fit near $`UB=0`$ of $`0.05`$ mag., which is typical for this band (e.g. Bessell 1990). The instrumental magnitudes $`u`$ were converted to AB magnitudes using the equation $`u(AB)=u+0.58`$, determined by integrating the spectrum of the star GD71 over the passband. Here we have retained the lower case $`u`$ for the AB magnitude indicating that the effective wavelength of the filter lies significantly away from the standard value. The colour term for the I and B filters are consistent with zero i.e. $`i=I`$ and $`b=B`$, and we used the equations $`I(AB)=I+0.43`$ and $`B(AB)=B0.14`$ (Fukugita et al. 1995) to put the I and B magnitudes onto the AB system.
The data were bias-subtracted, and twilight sky frames were used to flatten the frames in the standard way.
For the deconvolution, the individual bias subtracted and flattened I–band images were divided into six groups (chronologically) and the frames of each group were then combined using the optimal combination code described by Møller & Warren (1993), which maximizes the signal–to–noise ratio for faint sources. These six combined images were used in the simultaneous deconvolution process (see Sect. 3.1 below). Furthermore, all the individual I–band images were combined into one combined image, which was used in the PSF subtraction described in Sect. 3.2. In the same way we divided the individual bias subtracted and flattened B–band frames in four groups and calculated combined images for each group and for all images. Finally we combined the ten individual bias subtracted and flattened U–band frames into one combined image. The details of the sky noise in the combined images are provided in Table 2.
## 3 Analysis
### 3.1 Deconvolution
#### 3.1.1 The deconvolution method
The images were deconvolved using the MCS algorithm. This method is based on the principle that the resolution of a deconvolved image must be compatible with its sampling, which is limited by the Nyquist frequency. The deconvolved image is decomposed into a sum of deconvolved point–sources plus a background smoothed on the length scale of the final resolution. The intensities and positions of the point–sources as well as an image of the more extended objects are given as output of the deconvolution procedure. Image decomposition allows objects blended with or even superposed on point–sources to be studied in some detail.
In order to check if the deconvolved model is compatible with the data, a residual map is computed. The residual map contains in each pixel the $`\chi ^2`$ of the fit of the model image (re–convolved with the PSF) to the original data of that pixel. The $`\chi ^2`$ image is used to determine the appropriate weight attributed to the smoothing of the image of extended sources in order to avoid under- or over-fitting of the data (see MCS for further details). A deconvolution compatible with the data should show a flat residual map with a mean value of 1 all over the image.
The MCS algorithm makes it possible to simultaneously deconvolve several frames. The advantage of this process is to derive the optimally constrained deconvolved frame which is simultaneously compatible with several different images of a given object. This results in a more accurate decomposition of the data than the deconvolution of one single combined frame. Moreover, applying the algorithm to many dithered frames leads to a deconvolved image with an improved sampling.
#### 3.1.2 Application to the data
Simultaneous deconvolution of the U–band data from 1996 had already strongly indicated extended broad band emission in the direction of Q0151+048A (see Fig. 1). However, although the shape of the extended emission was similar to the one found in narrow-band (Fynbo et al. 1999), it was unclear to which extent systematic errors in the determination of the PSF influenced the detection and hence the signal-to-noise ratio of the object was too uncertain to constrain its morphology and luminosity.
There are two bright stars in the field of Q0151+048, referred to as psfA and psfB (see Fynbo et al. 1999). However, our new deep I–band data revealed that psfB has a faint red companion star at a projected distance of 0$`\stackrel{}{.}`$7. In the I–band it is 4.3 magnitude fainter than psfB, in the B–band it is 6.3 magnitudes fainter than psfB, and in the U–band it remains undetected. In the following we only use psfA for the determination of the PSF.
We adopted for the deconvolved image a pixel size of 0$`\stackrel{}{.}`$0541 (half of the original one), and a final resolution of 3 pixels FWHM, or 0$`\stackrel{}{.}`$16 (the Nyquist limit is 2 pixels FWHM).
#### 3.1.3 Results
Fig. 2 shows the deconvolved images in all three bands I, B and U. The images show the five sources already known to be in the field, namely the three point–sources qA, qB and the star s, and the two faint galaxies gA and gB south west of qA (see Fynbo et al. 1999 for details). However, there is also significant extended emission under the point–source emission from qA in all three bands. This emission have nearly identical morphology in the I and B bands with contours centred on the position of qA and with a slight elongation with position angle $`20^o`$ east of north. The shape and intensity of the extended U–band emission under qA (Fig. 2 right panel) is consistent with the shape and intensity derived from the 1996 U–band data (Fig. 1). Since the two sets of U–band data have been obtained with two different instruments, the consistency between the two measurements makes strong systematic errors unlikely. The morphology of the U–band emission is significantly more extended than the B and I morphology, about 4.5$`\times `$2.2 arcsec<sup>2</sup>, and has a position angle of about 100 east of north. This is very similar to that of the Ly$`\alpha `$ source S4, which extends over 6$`\times `$3 arcsec<sup>2</sup> with position angle 98 east of north.
The extended emission towards qA is $``$4 magnitudes fainter than that of the point–source emission. This high contrast makes it difficult to determine the exact ratio between the luminosity of the extended source ($`L_{\mathrm{Ext}}`$) and that of qA ($`L_{\mathrm{QSO}}`$). Several deconvolution solutions with different luminosity $`L_{\mathrm{QSO}}/L_{\mathrm{Ext}}`$ ratios are compatible with the residual map constructed as described above. Thus, there is some degeneracy between the plausible solutions found by the algorithm.
In order to demarcate the range of plausible solutions, a grid of 15 deconvolved images in each band was calculated, representing 5 different luminosity ratios $`L_{\mathrm{QSO}}/L_{\mathrm{Ext}}`$ and with 3 different values of the Lagrangian smoothing parameter applied during the deconvolution (see MCS for a description of the Lagrangian smoothing parameter). The solutions with the highest $`L_{\mathrm{QSO}}/L_{\mathrm{Ext}}`$ ratios were unphysical since they have a ring-shaped morphology, i.e. a hole at the position of the QSO. The lowest values of $`L_{\mathrm{QSO}}/L_{\mathrm{Ext}}`$ were rejected by inspection of the residual map mentioned above. The resulting range of magnitudes for the extended emission is given in Table 3. The solutions shown in Fig.2 are those with the highest acceptable values of $`L_{\mathrm{QSO}}/L_{\mathrm{Ext}}`$. Our conclusions concerning the morphology of the extended emission in the three bands are, however, unchanged for all solutions within the acceptable range.
In the B-band there is also significant extended emission under the PSF of the fainter neighbour quasar qB.
### 3.2 Object based image decomposition
In conclusion of the previous section: i) There is clear evidence for extended broad band (U, B and I) emission in the vicinity of the quasars Q0151+048A,B; ii) the morphology of the extended object(s) is identical in B and I but significantly different in U; iii) the U–band morphology is more extended and similar to the morphology of the Ly$`\alpha `$ emission from the DLA absorbing galaxy (S4).
Those conclusions would suggest that the extended emission in this field is made up of three individual components: The DLA absorbing galaxy, the host galaxy of qA and the host galaxy of qB. The different morphology in the four different bands would then indicate that the objects have different spectral energy distributions (SEDs).
In order to investigate this further we decomposed the superimposed images into individual objects with different SEDs. For this image decomposition we applied the same procedure we used for the narrow band image analysis (Fynbo et al. 1999), but here we add more components. We consider point–sources, de Vaucouleurs profiles and exponential profiles. The best decomposition is determined as the minimum $`\chi ^2`$ fit following an iterative procedure as described below.
#### 3.2.1 B–band data
The B–band image is more than a magnitude deeper than the I and U–band images in terms of the background rms surface brightness. Our first step was therefore to produce optimized models of the galaxies from the B–band data. For the decomposition we considered the following 8 components: Three point–sources (qA, qB, s), four galaxies to be fitted (gA, gB and the host galaxies of qA and qB, in what follows named HGa and HGb), and one galaxy of “frozen” morphology (the DLA absorbing galaxy S4). For S4 we adopted the model determined from the narrow–band data (Fynbo et al. 1999). Note that most of the objects do not overlap significantly, thus allowing us to fit them independently.
The bright star psfA was used with DAOPHOT II (Stetson 1997) to define the PSF. We then employed the iterative $`\chi ^2`$ minimization procedure detailed in Fynbo et al. 1999, to decompose the image of qA into a point–source and a de Vaucouleurs galaxy model (convolved with the PSF). For the calculation of the $`\chi ^2`$ we excluded a circular region of radius 0.65 arcsec centred on qA due to the large PSF-subtraction residuals. After ten iterations a stable solution was found. The same procedure repeated with an exponential-disc profile instead of the de Vaucouleurs profile resulted in a much poorer fit. A significant positive residual, centred about 1 arcsec east of qA, was left after this procedure. The most plausible interpretation of the residual is that it originates from the DLA absorbing galaxy S4. As S4 is known to extend across the position of the bright quasar, measuring its flux from a direct aperture measurement is impossible because of the large PSF subtraction residuals. Instead we made a grid of models to determine the flux of S4 via minimum $`\chi ^2`$ fitting. For a given assumed B magnitude of S4 we first subtracted the correctly scaled exponential–disc model as determined from the original Ly$`\alpha `$ image (Fynbo et al. 1999). For that given B magnitude of S4 we then repeated the iterative fitting of a de Vaucouleurs profile to the remaining flux. The final model was chosen to be the model with the smallest $`\chi ^2`$ measured in an area excluding pixels less than 0.65 arcsec from qA. The improvement in the fit due to the inclusion of the S4 model was significant ($`\mathrm{\Delta }\chi ^2`$ = -21). We also fitted the profiles of the two galaxies gA and gB. For gA the best fit was obtained with an exponential-disc profile, whereas the best fit for gB was obtained with a de Vaucouleurs profile.
Fig. 3b shows a 14x14 arcsec<sup>2</sup> field of the area after subtraction of the qA and qB PSFs as determined from the minimum $`\chi ^2`$ fit. The residuals, after the additional subtraction of the final models for HGa, gA and gB can be seen directly below (Fig. 3e). The magnitude of HGb was measured on this final subtracted image.
#### 3.2.2 U and I band data
The well constrained galaxy models determined from the high signal–to–noise B image were subsequently used to decompose the U and I–band data. Since the combined seeing of the U–band data was poorer than that of the B image, we first smoothed the galaxy models to the seeing of the U–band data. For a large grid of U–band magnitudes of S4 and HGa we then subtracted scaled versions of the smoothed S4 and HGa galaxy models, fitted and subtracted the quasar point source component using DAOPHOT II, and finally calculated the $`\chi ^2`$ in an area excluding pixels less than 0.8 arcsec from qA. The final model was selected to be that which had the smallest $`\chi ^2`$. The U magnitudes of the galaxies gA and gB were determined in the same way. For the decomposition of the combined I–band data, which have a better seeing than the B–band data, we first smoothed the I–band image to the seeing of the B image and then proceeded as for the U–band data.
Results of this procedure can be seen in Fig. 3a,d and Fig. 3c,f for the I and U–bands respectively. As for the B–band data the upper frames show the fields after subtraction of final fits of qA and qB only, while in the lower frames the fitted models of galaxies HGa, gA and gB have also been subtracted. The magnitudes (and estimated associated errors) of objects resulting from the fitting procedure are listed in Table 3.
## 4 Summary and discussion
Our original interest in the field of Q0151+048A was to identify the DLA galaxy in front of it. This identification was accomplished via imaging in Ly$`\alpha `$ (Fynbo et al. 1999), but our broad band images left some questions open. The purpose of the deeper broad–band data presented in this paper was to clarify this situation. We shall here first summarise our findings, then briefly consider their implications.
### 4.1 Results summary
Our new data have unambiguously confirmed the presence of extended emission in the field in all three bands I, B and U. The different morphology seen in the three bands strongly suggest that we see three objects superimposed: The quasar, the DLA absorbing galaxy and the quasar host galaxy.
The superposition of three close objects of widely differing brightnesses causes considerable degeneracy for any attempt to determine the brightness of the faintest sources, and it is therefore impossible to find a unique solution for the flux of the faintest object (the DLA galaxy S4). Nevertheless, we find that S4 is clearly detected in the U image. The U–band magnitude of S4 determined via our minimum $`\chi ^2`$ procedure is fully consistent (to within 1 $`\sigma `$) with being caused by the known Ly$`\alpha `$ flux at 3565Å alone. The data are therefore consistent with a zero contribution from any continuum source in the U–band.
It is difficult to determine the exact errors on the I and B magnitudes of S4, but for both images we found a very significant improvement in the reduced $`\chi ^2`$ of the fit when we included S4. It is therefore likely that S4 is indeed a low surface brightness continuum source, but this question is going to be extremely hard to settle.
The existence of a separate extended continuum source centred on qA is, however, clearly demonstrated independently in all bands. This result was arrived at independently via image deconvolution, and via our iterative object fitting technique.
### 4.2 Discussion: Starburst Galaxy or Dust scattering
The distance modulus (for z=1.93) in the assumed cosmology with h=0.5 is 45.8. Assuming instead $`\mathrm{\Omega }`$=0.3 and $`\mathrm{\Omega }_\mathrm{\Lambda }=`$0.7 the corresponding distance modulus becomes 46.6. Hence, the absolute AB magnitudes of the host galaxy HGa is $`<`$-24.0(-24.8) in U (rest frame 1100–1300Å), $`<`$-24.5(-25.3) in B (rest frame 1300–1500Å) and $`<`$-24.0(-24.8) in I (rest frame 2300–3400Å). Such extremely bright magnitudes are in the local universe only connected with brightest cluster galaxies (for comparison M87 and Centaurus A both have absolute magnitudes of roughly -23 in the V–band). Brightest cluster members can be as bright as -26 (Oemler 1976). Interestingly we find that the absolute magnitude of HGa is similar to those of the extended ‘fuzz’ that have been detected around other high redshift QSOs by Lehnert et al. (1992), Carballo et al. (1998) and Aretxaga et al. (1998a,b).
The morphology of the host galaxy HGa is best fit by a de Vaucouleurs profile. The fit to an exponential-disc leads to a much poorer fit. A plausible interpretation of the data is therefore that we see the early stage of a massive elliptical galaxy in the process of forming the bulk of its stars. Assuming that all the light is coming from stars, and not e.g. scattered quasar light (see below), we can estimate the star formation rate (SFR) needed to explain the observed fluxes. In the case of continuous star formation we can adopt the relation between the SFR and the luminosity at 1500Å SFR = L$`{}_{1500}{}^{}/(1.3\times 10^{40}ergs^1`$Å<sup>-1</sup>) commonly used for LBGs (Pettini et al. 1998) and we hence infer a star formation rate of order 100(200) M yr<sup>-1</sup> for $`\mathrm{\Omega }`$=1(0.3) and $`\mathrm{\Omega }_\mathrm{\Lambda }`$=0(0.7). For instantaneous bursts we can use the Starburst99 package (Leitherer et al. 1999) to infer the colours of models calculated with solar metallicity and ages 1, 10 and 100 million years. The colours for these three models are given in Table 4.
The colours of the host, 0$`<`$u-B$`<`$1.1, -1.1$`<`$B-I$`<`$0.1 from Deconvolution and 0.9$`\pm `$0.3, -0.7$`\pm `$0.3 from PSF-subtraction, are roughly consistent with instantaneous bursts with ages in the range 10–100 Myr. The number of stars formed in the burst would be in the range from 10<sup>8</sup> to a few times 10<sup>9</sup> stars depending on the age of the burst and on the assumed cosmology.
Another interpretation of the extended fuzz frequently seen around quasars, is light from the quasar itself scattered by dust. This mechanism is well known from radio galaxies at high redshifts where scattering off dust grains has revealed the existence of “hidden” quasars in the galaxy cores. It is likely that radio quiet QSOs have similar non–isotropic radiation fields (see e.g. the discussion in Møller & Kjærgaard 1992), and in that case our line of sight is such that we look straight down the emission cone inside of which the scattering is taking place. In this case we therefore expect to see the quasar emission cone “end on” via forward scattered quasar light. The scattering process is expected to be essentially grey and recent calculations predict that as much as 10% of the quasar light could be scattered in this way (Witt & Gordon, 1999; Városi & Dwek, 1999; Vernet et al. in prep.). If considering a clumpy medium, we would expect dust scattered light to be emitted from inside a very large volume in front of the quasar. When taking the cone geometry into account one would expect its total flux to be roughly a few % of the quasar flux at any given wavelength (Fosbury, private communication). From Table 3 we find that the flux from HGa is 3, 6 and 2% of the flux from Q0151+048A in U, B and I respectively. Similar, but less significant, results are found for HGb. It is not yet known if the light profile of scattered light from a cone will reproduce a de Vaucouleurs profile, but since this seems to be a universally preferred profile it is not unlikely. One thing worth noting in Fig. 3e are the negative residuals surrounding the position of qA at a distance of 2-3 arcsec after subtraction of the fitted de Vaucouleurs profile. This indicates that the true profile of HGa in reality falls off steeper than a de Vaucouleurs profile. If model calculations were to show such a steep profile for forward scattered light in a radiation cone, that would be a strong hint towards the nature of the quasar fuzz.
However, the colours of the extended emission as seen in Table 3 and Fig. 4 are significantly different from those of the two QSOs, which argues against the scattering hypothesis. Hence, we conclude that at least a significant fraction of the observed extended emission must be caused by a star burst.
## Acknowledgments
We wish to thank Pierre Magain and Peter Stetson for making available the MCS-code and DAOHPOT-II respectively. We have benefitted from stimulating discussions with R. Fosbury and J. Vernet on the subject of dust scattering, and with C. Jean on the subject of reddening models of galaxy spectra. We thank the anonymous referee for valuable comments that clarified the paper on essential points. JUF thanks the European Southern Observatory for support from the ESO studentship programme. IB thanks the European Southern Observatory for support from the ESO visitor programme. IB was supported in part by contract ARC94/99-178 “Action de Recherche Concertée de la Communauté Française (Belgium)” and Pôle d’Attraction Interuniversitaire, P4/05 (SSTC, Belgium).
|
warning/0004/hep-ph0004161.html
|
ar5iv
|
text
|
# Search for new physics with ATLAS at the LHC
## 1 Introduction
ATLAS, a general purpose detector for proton-proton collisions, will be capable of exploring the new energy regime of 14 TeV which will become accessible at the Large Hadron Collider (LHC). The LHC will be installed in the existing LEP tunnel at CERN, and will run at a design luminosity of $`10^{34}`$ cm<sup>-2</sup>s<sup>-1</sup>. About 30 fb<sup>-1</sup> are expected to be collected during the first three years, when the machine will run at low luminosity ($`10^{33}`$ cm<sup>-2</sup>s<sup>-1</sup>), while about 100 fb<sup>-1</sup> per year will be collected when running at design luminosity. An ultimate integrated luminosity of at least 300 fb<sup>-1</sup> is achievable. The discovery potential of ATLAS for new phenomena is discussed in the following.
## 2 Higgs bosons
The experimental observation of one or several Higgs bosons will be fundamental for a better understanding of the mechanism of electroweak symmetry breaking. In the Standard Model (SM), one doublet of scalar fields is assumed, leading to the existence of one neutral scalar particle H. In the Minimal Supersymmetric Standard Model (MSSM), on the other hand, at least two Higgs doublets are required, corresponding to two charged (H<sup>±</sup>) and three neutral (h, H, A) physical states.
### 2.1 Standard Model Higgs boson
The dominant production mechanism of a SM Higgs boson at LHC energies is gluon-gluon fusion, which proceeds via a heavy quark triangle loop. For larger masses, also the WW fusion process contributes significantly.
For Higgs masses below 150 GeV, the decay modes to $`\mathrm{b}\overline{\mathrm{b}}`$ ($`BR90\%`$) and to $`\tau \tau `$ ($`BR10\%`$) dominate. The decay to two photons is rather rare ($`BR10^3`$) and limited to the region $`90<m_\mathrm{H}<150`$ GeV. At larger masses ($`>180`$ GeV), the dominant decays are to WW ($`BR75\%`$) and to ZZ ($`BR20\%`$).
The overall sensitivity for the discovery of a SM Higgs boson is shown in Fig. 1 for various channels assuming an integrated luminosity of 100 fb<sup>-1</sup>. A SM Higgs boson can be discovered with the ATLAS experiment over the full mass range up to $``$1 TeV with a high significance. A $`5\sigma `$ discovery can be achieved —with two channels in most cases— over the full mass range even after a few years of running at low luminosity.
In the low mass region ($`m_\mathrm{H}<150`$ GeV), a SM Higgs boson can be discovered through the $`\mathrm{H}\gamma \gamma `$ channel with a signal significance of 5–7$`\sigma `$ over the continuous $`\gamma \gamma `$ background. The significance can be further enhanced by exploiting the associated production of the Higgs boson with a W or a $`\mathrm{t}\overline{\mathrm{t}}`$-pair. In the same mass range, a signal from the $`\mathrm{t}\overline{\mathrm{t}}\mathrm{H}`$, $`\mathrm{H}\mathrm{b}\overline{\mathrm{b}}`$ channel can also be observed with a significance $`>5\sigma `$ by exploiting the b-tagging capabilities of the detector.
The decay channel $`\mathrm{H}\mathrm{ZZ}^{}4\mathrm{}`$ provides a rather clean signature in the mass range between $``$120 GeV and $`2m_\mathrm{Z}`$, above which the gold-plated channel with two real Z bosons in the final states opens up. Both electrons and muons are considered in the final state, thus yielding eeee, ee$`\mu \mu `$ and $`\mu \mu \mu \mu `$ event topologies.
In the mass region $`150<m_\mathrm{H}<180`$ GeV, a pronounced dip occurs in the $`BR(\mathrm{H}\mathrm{ZZ}^{})`$, due to the opening of the WW decay mode. An excess of events from the $`\mathrm{H}\mathrm{WW}^{()}\mathrm{}\nu \mathrm{}\nu `$ channel can be used to enhance the significance in this region.
If a SM Higgs boson would be discovered at the LHC using the aforementioned analyses, its mass will be measured with a precision of 0.1% for $`m_\mathrm{H}<400`$ GeV and of 0.1–1% for $`400<m_\mathrm{H}<700`$ GeV. The Higgs boson width can be determined for masses above 200 GeV using the $`\mathrm{H}\mathrm{ZZ}4\mathrm{}`$ channel.
### 2.2 Minimal Supersymmetric Standard Model Higgs
The capability of the ATLAS experiment to detect MSSM Higgs bosons has been studied in depth over the last few years. Large sparticles masses have been assumed so that Higgs bosons are not allowed to decay to supersymmetric particles. In the MSSM, various decay modes accessible also in the case of the SM Higgs boson are predicted, such as $`\mathrm{h}\gamma \gamma `$, $`\mathrm{h}\mathrm{b}\overline{\mathrm{b}}`$, $`\mathrm{H}\mathrm{ZZ}^{()}4\mathrm{}`$. In addition, some channels are strongly enhanced at large $`\mathrm{tan}\beta `$, e.g. $`\mathrm{H}/\mathrm{A}\tau \tau `$ and $`\mathrm{H}/\mathrm{A}\mu \mu `$. Other potentially interesting channels, such as $`\mathrm{H}/\mathrm{A}\mathrm{t}\overline{\mathrm{t}}`$, $`\mathrm{A}\mathrm{Zh}`$, $`\mathrm{H}\mathrm{hh}`$ and $`\mathrm{H}^\pm \mathrm{tb}`$, were also studied. The $`5\sigma `$ discovery contours for individual channels in the $`(m_\mathrm{A},\mathrm{tan}\beta )`$ plane are shown in Fig. 2 for an integrated luminosity of 300 fb<sup>-1</sup>.
Complete coverage of the region shown will be possible at the LHC. Over a considerable fraction of the parameter space, at least two channels are accessible and/or more than one Higgs bosons can be observed. In most cases, the experiment will be capable of distinguishing between a SM and an MSSM Higgs boson.
## 3 Supersymmetry
If supersymmetry (SUSY) exists at the electroweak scale, then its discovery at the LHC should be straightforward. The SUSY cross section is dominated by gluinos and squarks production, which are strongly produced with large cross sections. Gluinos and squarks then decay via a series of steps into the LSP (which may itself decay if $`R`$-parity is violated). These decay chains lead to a variety of signatures involving multiple jets, leptons, photons, heavy flavors (e.g. see Fig. 4), W and Z bosons, and missing energy. The combination of a large production cross section and distinctive signatures makes it easy to separate SUSY from the Standard Model background. Therefore, the main challenge is not to discover SUSY, but to separate the many SUSY processes that occur and to measure the masses and other properties of the SUSY particles. In most cases, the backgrounds from other SUSY events dominate over the reducible SM backgrounds.
The approach followed has been the detailed investigation of signatures for particular points in the parameter spaces of the minimal supergravity (SUGRA), gauge mediated supersymmetry breaking (GMSB) and $`R`$-parity violating models. Methods such as looking for kinematic endpoints in mass distributions and using these to determine combination of masses have proven generally useful (e.g. see Fig. 4). By using these methods, the fundamental parameters of the underlying theory can be determined with precision of a few percent.
The starting point in the kind of analysis described earlier will be to look for characteristic deviations from the Standard Model. In SUGRA and some other models, there will be events with multiple jets and leptons plus large missing energy. In GMSB models, there would be events with prompt photons or quasi-stable sleptons. In $`R`$-parity violating models, there would be events with very high jet and/or leptons multiplicities.
## 4 Other physics scenarios
Various theoretical scenarios, in addition to supersymmetry, have been studied by ATLAS in order to establish the discovery potential. ATLAS will be sensitive to new resonances predicted in technicolor theories, up to the TeV range. Although the parameter space is very large, the number of potential channels allows for combinations of signatures to help in understanding the nature of the resonances, and determine the possible existence of techniparticles.
Other studies involve the discovery of excited quarks in the photon plus jet channel (masses up to 6 GeV), leptoquarks (masses up to 1.5 TeV) and compositeness probed by the high $`p_\mathrm{T}`$ jets (mass scale up to 40 TeV). New vector bosons can be discovered through their leptonic decays for masses up to 5–6 TeV. Monopoles can be probed via the $`\gamma \gamma \mathrm{X}`$ cross section for masses up to 20 TeV. It is also possible to investigate extra-dimensions scenarios by searching for missing energy plus jet or missing energy plus photon signatures.
## 5 Conclusions
ATLAS has a wide discovery potential for new physics and sensitivity to a large variety of signatures. A Standard Model Higgs boson can be discovered, if exists, over the whole mass range up to $``$1 TeV. MSSM Higgs bosons are accessible over a large part of the $`(m_\mathrm{A},\mathrm{tan}\beta )`$ parameter space.
Supersymmetric partners are expected to reveal themselves in signatures involving large missing energy, many leptons and/or many jets. These studies have been performed in the context of constrained supersymmetric scenarios, such as supergravity, gauge mediated supersymmetry breaking and $`R`$-parity violation.
Moreover, the possibility of discovering new physics in many other cases has been investigated by ATLAS, such as technicolor, compositeness and extra dimensions. Limits on the existence of excited quarks, leptoquarks, new gauge bosons, monopoles can be easily set up to the TeV scale.
## Acknowledgments
I would like to thank the organizers of the Winter Institute for their hospitality and support during my stay at Lake Louise. This work was supported in part by the Greek State Scholarships Foundation (I.K.Y.).
|
warning/0004/cond-mat0004405.html
|
ar5iv
|
text
|
# Coexisting Kondo singlet state with antiferromagnetic long-range order: A possible ground state for Kondo insulators
\[
## Abstract
The ground-state phase diagram of a half-filled anisotropic Kondo lattice model is calculated within a mean-field theory. For small transverse exchange coupling $`J_{}<J_{c1}`$, the ground state shows an antiferromagnetic long-range order with finite staggered magnetizations of both localized spins and conduction electrons. When $`J_{}>J_{c2}`$, the long-range order is destroyed and the system is in a disordered Kondo singlet state with a hybridization gap. Both ground states can describe the low-temperature phases of Kondo insulating compounds. Between these two distinct phases, there may be a coexistent regime as a result of the balance between local Kondo screening and magnetic interactions.
PACS numberes: 71.28.+d, 72.15.Qm, 75.20.Hr
\]
The Kondo lattice model is often considered as a theoretical model for heavy fermion materials. For this model, an important issue arises from the interplay between the Kondo screening and the magnetic interactions among localized spins mediated by the conduction electrons. The former effect favors a nonmagnetic Kondo singlet state in strong coupling limit, while the latter interactions tend to stabilize a magnetically long-range ordered state in weak coupling limit. The nature of such a transition between these two distinct phases has been a long standing issue since it was first suggested by Doniach . The one-dimensional model was intensively studied at half filling, showing that its ground state is always a disordered Kondo singlet state (or a spin liquid state). In higher dimensions, however, both antiferromagnetic long-range order (AFLRO) and disordered Kondo singlet states may occur . Recently, there have been more indications of such a transition from various approximate treatments for the Kondo or Anderson lattice models, including variational Monte Carlo calculation , higher-order series expansions , quantum Monte Carlo simulations , and infinite dimensional calculations .
In this paper, we would like to consider the issue whether the disordered Kondo singlet state can coexist with an AFLRO at half-filling. First of all, we introduce an anisotropic Kondo lattice model by distinguishing the longitudinal spin exchange interaction from the transverse one, because we notice that the longitudinal interaction describes a polarization of the conduction electrons by the localized spins, while the transverse one describes a spin-flip scattering of the conduction electrons off the localized spins. The former interaction is the origin of the magnetic interactions among localized spins, leading to an AFLRO; the latter interaction is responsible for the local Kondo screening effects, yielding a disordered Kondo singlet state. We also notice that the AFLRO and local Kondo singlet order operators form an irreducible representation of an SU(2) algebra, which can be regarded as the spectrum generating algebra of the half-filled Kondo lattice model. Within the framework of a mean field theory, the magnetic interactions and the Kondo screening are considered on an equal footing, and the ground-state phase diagram of the model Hamiltonian is calculated.
The symmetric Kondo lattice model with anisotropic exchange couplings is defined as:
$`H`$ $`=`$ $`{\displaystyle \underset{𝐤\sigma }{}}ϵ_𝐤c_{𝐤\sigma }^{}c_{𝐤\sigma }+H_{}+H_{}`$ (1)
$`H_{}`$ $`=`$ $`{\displaystyle \frac{J_{}}{4}}{\displaystyle \underset{i}{}}(d_i^{}d_id_i^{}d_i)(c_i^{}c_ic_i^{}c_i)`$ (2)
$`H_{}`$ $`=`$ $`{\displaystyle \frac{J_{}}{2}}{\displaystyle \underset{i}{}}(d_i^{}d_ic_i^{}c_i+d_i^{}d_ic_i^{}c_i),`$ (3)
where the pseudo-fermion representation for the localized spins $`S_i^z=(d_i^{}d_id_i^{}d_i)/2,`$ $`S_i^{}=d_i^{}d_i,`$ $`S_i^+=d_i^{}d_i,`$ has been used with a local constraint $`d_i^{}d_i+d_i^{}d_i=1`$. Note that $`H_{}`$ can also be rewritten in the form of
$`{\displaystyle \frac{J_{}}{4}}{\displaystyle \underset{i}{}}\left[(d_i^{}c_i+c_i^{}d_i)^2+(d_i^{}c_i+c_i^{}d_i)^2\right].`$$`H_{}`$ describes the polarization of conduction electrons by the local impurity spins, leading to a magnetic instability, while $`H_{}`$ corresponds to the spin-flip scatterings, giving rise to the local Kondo screening effect. The latter effect has been investigated by various approaches, in particular, those based on a 1/N expansion (N is the degeneracy of the localized spin). However, the former effect is not treated on an equal footing in these approaches because the magnetic interaction occurs there only at 1/N<sup>2</sup> order .
It has been known that the spectrum generating algebra plays an important role in analyzing complete spectra for a model. For instance, it has been used to study collective excitations and phase transitions in one-dimensional metals and two-dimensional Hubbard model . Here the spectrum generating algebra for the half-filled Kondo lattice model is given by generators of an SU(2) Lie group:
$`\tau ^+`$ $`=`$ $`{\displaystyle \underset{i}{}}e^{i𝐐𝐫_i}(c_i^{}d_i+d_i^{}c_i),`$ (4)
$`\tau ^{}`$ $`=`$ $`{\displaystyle \underset{i}{}}e^{i𝐐𝐫_i}(d_i^{}c_i+c_i^{}d_i),`$ (5)
$`\tau ^z`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{i}{}}\left[(c_i^{}c_ic_i^{}c_i)(d_i^{}d_id_i^{}d_i)\right],`$ (6)
which satisfy the commutation relations: $`[\tau ^z,\tau ^\pm ]=\pm \tau ^\pm ,`$ $`[\tau ^+,\tau ^{}]=2\tau ^z,`$ the SU(2) algebra. $`𝐐`$ is the AF reciprocal vector. We also find that an irreducible representation of this SU(2) algebra can serve as the order parameter operators for this model. They are given by:
$`K^+`$ $`=`$ $`{\displaystyle \underset{i}{}}(c_i^{}d_i+d_i^{}c_i),`$ (7)
$`K^{}`$ $`=`$ $`{\displaystyle \underset{i}{}}(d_i^{}c_i+c_i^{}d_i),`$ (8)
$`M^z`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{i}{}}e^{i𝐐𝐫_i}\left[(d_i^{}d_id_i^{}d_i)(c_i^{}c_ic_i^{}c_i)\right],`$ (9)
with the following commutation relations $`[\tau ^\pm ,K^{}]=2M^z,`$ $`[\tau ^\pm ,M^z]=K^\pm `$, $`[\tau ^z,K^\pm ]=\pm K^\pm ,`$ and $`[\tau ^z,M^z]=0,`$ where $`K^\pm `$ are the order parameter operators describing the local Kondo singlet state and $`M^z`$ is the staggered magnetization operator describing the AFLRO in terms of a commensurate spin-density wave. The expectation values of these order operators are the corresponding order parameters. Note that the model Hamiltonian does not satisfy this SU(2) symmetry defined by the spectrum generating algebra, which can however provide us with a guide to choose the order parameters in the model Hamiltonian.
Now we introduce a mean-field theory treating equally the magnetic interactions between the localized spins and the itinerant electron screening aspects. The longitudinal interaction term is approximated as:
$`{\displaystyle \frac{J_{}}{2}}[m_d{\displaystyle \underset{i}{}}e^{i𝐐𝐫_i}(c_i^{}c_ic_i^{}c_i)`$
$`\text{ }m_c{\displaystyle \underset{i}{}}e^{i𝐐𝐫_i}(d_i^{}d_id_i^{}d_i)]+J_{}m_cm_d𝒩,`$
where the staggered magnetizations have been introduced as the AF order parameters $`m_c=\frac{1}{2}<c_i^{}c_ic_i^{}c_i>e^{i𝐐𝐫_i}`$ and $`m_d=\frac{1}{2}<d_i^{}d_id_i^{}d_i>e^{i𝐐𝐫_i},`$ and a minus sign has been absorbed into the definition of $`m_c`$ for convenience. The transverse interaction term is approximated to be:
$`{\displaystyle \frac{J_{}V}{2}}{\displaystyle \underset{i\sigma }{}}(c_{i\sigma }^{}d_{i\sigma }+d_{i\sigma }^{}c_{i\sigma })+{\displaystyle \frac{J_{}}{2}}V^2𝒩,`$with the hybridization order parameter describing the local Kondo singlet state $`V=<c_i^{}d_i+d_i^{}c_i>`$ $`=<d_i^{}c_i+c_i^{}d_i>.`$ Moreover, we could also introduce a $`d`$-electron chemical potential term $`\mathrm{\Sigma }_{i,\sigma }`$ $`E_dd_{i\sigma }^{}d_{i\sigma }`$ to fix the $`d`$-electron density to be one on each site, but actually this is not necessary at half filling because the electron-hole symmetry automatically imposes $`E_d=0`$ . Therefore, the mean field Hamiltonian can be written in the following form
$$H=\underset{𝐤\sigma }{}{}_{}{}^{^{}}\mathrm{\Psi }_{𝐤\sigma }^{}\widehat{H}_{𝐤\sigma }\mathrm{\Psi }_{𝐤\sigma }+𝒩(J_{}m_cm_d+\frac{J_{}V^2}{2}),$$
(10)
where $`\mathrm{\Psi }_{𝐤\sigma }^{}=(c_{𝐤\sigma }^{},`$ $`c_{𝐤+𝐐\sigma }^{},`$ $`d_{𝐤\sigma }^{},`$ $`d_{𝐤+𝐐\sigma }^{})`$, its transposition $`\mathrm{\Psi }_{𝐤\sigma }`$, and the matrix
$$\widehat{H}_{𝐤\sigma }=\left[\begin{array}{cccc}ϵ_𝐤,\hfill & \frac{J_{}}{2}m_d\sigma ,\hfill & \frac{J_{}V}{2},\hfill & 0\hfill \\ \frac{J_{}}{2}m_d\sigma ,\hfill & ϵ_𝐤,\hfill & 0,\hfill & \frac{J_{}V}{2}\hfill \\ \frac{J_{}V}{2},\hfill & 0,\hfill & 0,\hfill & \frac{J_{}}{2}m_c\sigma \hfill \\ 0,\hfill & \frac{J_{}V}{2},\hfill & \frac{J_{}}{2}m_c\sigma ,\hfill & 0\hfill \end{array}\right].$$
(11)
The static staggered magnetizations partially break the translational symmetry and the Brillouin zone is folded in half, so that the summation over $`𝐤`$ is taken in the reduced Brillouin zone. The quasiparticle bands are determined by the equation: $`E\widehat{I}\widehat{H}=0,`$ giving rise to four bands with dispersions,
$`E_{\pm \pm }(𝐤)=\pm {\displaystyle \frac{1}{\sqrt{2}}}\sqrt{ϵ_𝐤^2+J_{}^2(m_c^2+m_d^2)/4+J_{}^2V^2/2\pm E^{}(𝐤)}`$with
$`E^{}(𝐤)`$ $`=`$ $`\{[ϵ_𝐤^2+J_{}^2(m_c^2+m_d^2)/4+J_{}^2V^2/2]^2`$
$`{\displaystyle \frac{1}{4}}(J_{}^2m_cm_d+J_{}^2V^2)^2J_{}^2m_c^2ϵ_𝐤^2\}^{1/2},`$
and the ground-state energy is given by
$`E_g=2{\displaystyle \underset{𝐤}{}}{}_{}{}^{^{}}[E_{}(𝐤)+E_+(𝐤)]+𝒩(J_{}m_cm_d+{\displaystyle \frac{J_{}V^2}{2}}),`$corresponding to completely filling the two negative energy bands with electrons. The self-consistent equations are obtained by minimizing the ground-state energy $`E_g`$ with respect to $`m_d`$, $`m_c`$, and $`V`$, respectively. After the summations over momenta are transformed into integrals over energies by assuming a constant density of states of the conduction electrons in $`[D,D]`$, the self-consistent equations are expressed in the form
$`{\displaystyle \frac{\lambda _{}}{8}}{\displaystyle \underset{1}{\overset{1}{}}}𝑑\epsilon {\displaystyle \frac{\text{ }2F_2(\epsilon )m_d+(\lambda _{}^2m_cm_d+\lambda _{}^2V^2)m_c}{2F_1(\epsilon )F_2(\epsilon )}}=m_c,`$ (12)
$`{\displaystyle \frac{\lambda _{}}{8}}{\displaystyle \underset{1}{\overset{1}{}}}𝑑\epsilon {\displaystyle \frac{\text{ }2(F_2(\epsilon )+2\epsilon ^2)m_c+(\lambda _{}^2m_cm_d+\lambda _{}^2V^2)m_d}{2F_1(\epsilon )F_2(\epsilon )}}=m_d,`$ (13)
$`{\displaystyle \frac{\lambda _{}}{4}}{\displaystyle \underset{1}{\overset{1}{}}}𝑑\epsilon {\displaystyle \frac{2F_2(\epsilon )+(\lambda _{}^2m_cm_d+\lambda _{}^2V^2)}{2F_1(\epsilon )F_2(\epsilon )}}=1,`$ (14)
where we have used the simplified expressions in terms of $`F_1`$ and $`F_2,`$
$`F_1(\epsilon )`$ $`=`$ $`\sqrt{\epsilon ^2+\lambda _{}^2(m_c^2+m_d^2)/4+\lambda _{}^2V^2/2+F_2(\epsilon ),}`$ (15)
$`F_2(\epsilon )`$ $`=`$ $`\sqrt{{\displaystyle \frac{1}{4}}\left(\lambda _{}^2m_cm_d+\lambda _{}^2V^2\right)^2+\lambda _{}^2m_c^2\epsilon ^2},`$ (16)
and we have also used the notations: $`\lambda _{}=J_{}/D,`$ $`\lambda _{}=J_{}/D`$, where $`2D`$ is the bandwidth of the conduction electrons.
When $`\lambda _{}\lambda _{}`$, the polarization effect described by $`H_{}`$ dominates, and the ground state has an AFLRO characterized by finite staggered magnetizations for both localized spins and the conduction electrons. Thus, the total staggered magnetization of the system should be given by $`M^z/𝒩=(m_d+m_c)`$, while the staggered magnetic moment is $`\mu =(m_dm_c)`$. Let us assume for the moment they are the only order parameters, i.e. $`V=0`$. All equations involved are greatly simplified. Due to unit cell doubling the conduction electron band is folded in half to form two new bands: $`ϵ_\pm (𝐤)=\pm \sqrt{ϵ_𝐤^2+(J_{}m_d)^2/4}.`$ There is a gap in the conduction electron excitation spectrum: $`\mathrm{\Delta }=J_{}m_d/2.`$ The charge gap is twice this value, $`\mathrm{\Delta }_{ch}=J_{}m_d`$. Moreover, the ground-state energy of the AFLRO state can also be found
$$\frac{E_g^{AF}}{𝒩D}=\frac{1}{2}\left[\sqrt{1+\left(\frac{\lambda _{}}{4}\right)^2}+\left(\frac{\lambda _{}}{4}\right)\frac{m_c}{m_d}\right],$$
(17)
where $`𝒩`$ is the total number of the lattice sites, and $`\frac{m_c}{m_d}=\frac{\lambda _{}}{4}\mathrm{ln}\left(\frac{4}{\lambda _{}}+\sqrt{1+(\frac{4}{\lambda _{}})^2}\right)`$with $`m_d=\frac{1}{2}`$.
In the opposite limit, when $`\lambda _{}\lambda _{}`$, the Kondo screening effect described by $`H_{}`$ dominates, and the ground state has a finite local Kondo order parameter $`V`$. If we assume again for the moment it is the only order parameter, the system is a nonmagnetic band insulator with an effective hybridization: $`\frac{1}{V}=\lambda _{}\mathrm{sinh}\left(\frac{1}{\lambda _{}}\right)`$. A similar result has already been given by the 1/N expansion approach , which is believed to describe the correct low-energy Kondo physics. The quasiparticle excitation spectrum is expressed in the form $`ϵ_\pm ^{}(𝐤)=\frac{1}{2}\left(ϵ_𝐤\pm \sqrt{ϵ_𝐤^2+\left(J_{}V\right)^2}\right),`$ and there is a small hybridization gap $`\mathrm{\Delta }_{hy}\left(J_{}V\right)^2/2D`$, which splits the Kondo resonance formed at the Fermi level. At half filling, each lattice site has one conduction electron and one localized spin, and they can form a local singlet state, thus the system becomes a disordered nonmagnetic insulator, a band-insulator with both charge and spin gap $`2\mathrm{\Delta }_{hy}`$. The ground-state energy can be also calculated
$$\frac{E_g^{KS}}{𝒩D}=\frac{1}{2}\mathrm{coth}\left(\frac{1}{\lambda _{}}\right).$$
(18)
Physically, this disordered Kondo singlet state is adiabatically connected to the usual Kondo spin liquid state .
Comparing the respective ground-state energies $`E_g^{KS}`$ and $`E_g^{AF}`$, we find the phase boundary between the two ground states:
$`\left({\displaystyle \frac{1}{\lambda _{}}}\right)_c=\mathrm{coth}^1[\sqrt{1+\left({\displaystyle \frac{\lambda _{}}{4}}\right)^2}`$ (19)
$`+\left({\displaystyle \frac{\lambda _{}}{4}}\right)^2\mathrm{ln}({\displaystyle \frac{4}{\lambda _{}}}+\sqrt{1+\left({\displaystyle \frac{4}{\lambda _{}}}\right)^2})],`$ (20)
which is displayed by the solid line in Fig.1. Below this line, the AFLRO state is more stable $`E_g^{AF}<E_g^{KS}`$; above this line the disordered Kondo singlet state is more stable. We note that this line intersects the diagonal $`\lambda _{}=\lambda _{}`$ at $`0.58`$. Taking into account that the cut-off parameter $`D=2t`$ with $`t`$ as the nearest neighbor hopping, in two dimensions, where our assumption of a constant density of states is better justified, we find $`J_c/t=1.16`$. This value is not far from the numerical results $`J_c/t=1.401.45`$, obtained for the isotropic model.
Now we turn to discuss the main new result of this paper, namely the possible coexistence of AFLRO with Kondo singlet state. In fact, our system of self-consistent equations allows solutions with all order parameters $`m_d,m_c`$ and $`V`$ being nonzero. Let’s start from the region where AF order dominates, with finite $`m_d`$ and $`m_c`$ to see where the instability towards the Kondo singlet state emerges. Assume $`V`$ being small, but non-zero, we obtain the following solution,
$`\left({\displaystyle \frac{1}{\lambda _{}}}\right)_{c2}={\displaystyle \frac{2}{\lambda _{}}}{\displaystyle \frac{m_c}{m_d}}`$ (21)
$`\text{ }+\sqrt{{\displaystyle \frac{1\frac{m_c}{m_d}}{1+\frac{m_c}{m_d}}}}\mathrm{tan}^1\left(\sqrt{{\displaystyle \frac{1\frac{m_c}{m_d}}{1+\frac{m_c}{m_d}}}}\mathrm{tanh}\left({\displaystyle \frac{m_c}{m_d}}{\displaystyle \frac{2}{\lambda _{}}}\right)\right),`$ (22)
where $`\frac{m_c}{m_d}=\frac{\lambda _{}}{4}\mathrm{ln}\left(\frac{4}{\lambda _{}}+\sqrt{1+(\frac{4}{\lambda _{}})^2}\right)`$. This instability line for the AF state is delineated by the thick dotted line in Fig.1. In a similar way, we can also determine the instability boundary in the Kondo singlet phase, which corresponds to the appearance of small AF order parameters $`m_c`$ and $`m_d`$. From the above self-consistent equations, we can derive the critical value of the coupling parameter
$$\left(\frac{1}{\lambda _{}}\right)_{c1}=\frac{\lambda _{}\mathrm{sinh}\left(\frac{2}{\lambda _{}}\right)2}{4\lambda _{}\left(\sqrt{\lambda _{}\mathrm{sinh}\left(\frac{2}{\lambda _{}}\right)1}1\right)},$$
(23)
and a relation between $`m_c`$ and $`m_d`$: $`\frac{m_c}{m_d}=\frac{\lambda _{}}{4\lambda _{}\lambda _{}}`$. This instability line for the disordered Kondo singlet state is displayed by a thin dotted line in Fig. 1.
Between these two instability lines $`(\lambda _{\text{ }c1}<\lambda _{}<\lambda _{\text{ }c2})`$, there is a narrow regime where the AF and the local Kondo singlet screening order parameters coexist to balance the energy gain from the transverse and longitudinal exchange couplings. The disordered Kondo singlet state and the AFLRO state can both be used to describe two distinct ground states of the conventional Kondo insulators. It seems to us that a new phase may be present when the exchange coupling parameters are tuned carefully. In such a new phase, the dynamical magnetic structure factor should have two contributions: a $`𝐪`$-independent single site slow component, which is typical of the localized Kondo-type excitations, and a strongly $`𝐪`$-dependent intersite fast component, reflecting the magnetic interactions. These features can be detected in inelastic neutron scattering experiments for some Kondo insulating materials. Moreover, since the localized spins are partially screened by the conduction electrons, and the residual localized spins still have weak AF long-range correlations mediated by the polarization of the conduction electrons. This novel feature may be related to the small magnitude of the induced staggered magnetic moments for URu<sub>2</sub>Si<sub>2</sub> and UPt<sub>3</sub>, the so-called heavy fermion micromagnetism.
We realize that our result is based on a mean field calculation, and its validity should be further checked by more rigorous analytic and numerical treatments. Eventually, it should be verified by experiments. We also note that the numerical results for the isotropic model indicate a second order phase transition between the Kondo singlet and AFLRO states. However, the order of phase transitions is a very sensitive issue, and the isotropic case may well be a degenerate point of the more general anisotropic model we consider here.
In conclusion, we have considered a half-filled anisotropic Kondo lattice model within a mean field theory. The ground-state phase diagram has been calculated. In addition to the AFLRO phase and the Kondo singlet phase, we have found that both of these two distinct phases can coexist as a result of the balance between the Kondo screening effects and the magnetic interactions, which provides a possible new ground state for the Kondo insulating compounds.
Acknowledgments
One of the authors (G. -M. Zhang) wishes to thank Zhi-Liang Cao, Qiang Gu, and Xiao-Bin Wang for their useful discussions and help, and would also like to express his gratitude to International Center for Theoretical Physics (Trieste, Italy) for the hospitality, where this work was initiated.
Figure Caption
Fig.1. The ground-state phase diagram of the half-filled anisotropic Kondo lattice model in the $`\lambda `$$`{}_{}{}^{}\lambda _{}`$ plane. The narrow area between the lines of $`(\lambda `$$`{}_{}{}^{})_{c1}`$ and $`(\lambda `$$`{}_{}{}^{})_{c2}`$ is the regime where the local Kondo singlet state and the AFLRO may coexist. The solid line corresponds to the boundary of $`E_g^{AF}=E_g^{KS}`$, and the thin dashed line denotes the isotropic limit of the model.
|
warning/0004/math0004173.html
|
ar5iv
|
text
|
# Eisenstein series twisted by modular symbols for the group SL_𝑛
## 1. Introduction
### 1.1.
Let $`\mathrm{\Gamma }`$ denote a finitely generated discrete subgroup of $`\mathrm{SL}_2()`$ that contains translations and acts on the upper halfplane $`𝔥`$. An automorphic form of real weight $`r`$ and multiplier $`\psi :\mathrm{\Gamma }𝐔`$ (here $`𝐔=\left\{w\right||w|=1\}`$ is the unit circle) is a meromorphic function $`F:𝔥`$ that satisfies
$$F(\gamma z)=\psi (\gamma )j(\gamma ,z)^rF(z)$$
for all $`\gamma =\left(\begin{array}{cc}a& b\\ c& d\end{array}\right)\mathrm{\Gamma }`$ with $`j(\gamma ,z)=cz+d`$. For $`r0`$, an integer, and $`F:𝔥`$ any function with sufficiently many derivatives, G. Bol proved the identity
$$\frac{d^r}{dz^{r+1}}\left((cz+d)^rF(\gamma z)\right)=(cz+d)^{r2}F^{(r+1)}(\gamma z),$$
which holds for all $`\gamma \mathrm{SL}_2()`$. It follows that if $`f(z)`$ is an automorphic form of weight $`r+2`$ and multiplier $`\psi `$, and if $`F`$ is any $`(r+1)`$-fold indefinite integral of $`f`$, then $`F`$ satisfies the functional equation
$$F(\gamma z)=\psi (\gamma )(cz+d)^r\left(F(z)+\varphi (\gamma ,z)\right),$$
where $`\varphi (\gamma ,z)`$ is a polynomial in $`z`$ of degree $`r`$ satisfying the cocycle condition
$$\varphi (\gamma _1\gamma _2,z)=\overline{\psi (\gamma _2)}j(\gamma _2,z)^r\varphi (\gamma _1,\gamma _2z)+\varphi (\gamma _1,z).$$
Such a function $`F`$ is called an *automorphic* (or *Eichler*) *integral*, and the corresponding polynomial $`\varphi (\gamma ,z)`$ is called a *period polynomial*.
### 1.2.
The Eisenstein series $`E^{}(z;r,\psi ,\varphi )`$ (twisted by a period polynomial $`\varphi `$) is defined by the infinite series
$$E^{}(z;r,\psi ,\varphi )=\underset{\gamma \mathrm{\Gamma }_{\mathrm{}}\backslash \mathrm{\Gamma }}{}\psi (\gamma )\varphi (\gamma ,z)j(\gamma ,z)^r.$$
The twisted Eisenstein series $`E^{}(z;r,\psi ,\varphi )`$ was first introduced by Eichler (1965). Automorphic integrals and Eisenstein series twisted by period polynomials were systematically studied by Knopp (1974). More recently (1995) nonholomorphic Eisenstein series twisted by modular symbols (period polynomials of degree 0) were introduced (cf. §5.3). O’Sullivan found (using Selberg’s method) the functional equation of these twisted Eisenstein series; very recently O’Sullivan and Chinta explicitly computed the scattering matrix occurring in the functional equation.
In this paper we show how to generalize the construction of Eisenstein series twisted by modular symbols to the group $`\mathrm{SL}_n`$. The basic properties and region of absolute convergence of such series are obtained in the case of the minimal parabolic subgroup. We conjecture that these series satisfy functional equations.
## 2. Eisenstein series
### 2.1.
In this section we recall the definition of cuspidal Eisenstein series following Langlands \[15, Ch. 4\]. We begin with some notation.
Let $`G=\mathrm{SL}_n()`$, let $`K=\mathrm{SO}_n()`$, and let $`\mathrm{\Gamma }G()`$ be an arithmetic group. Let $`P_0G`$ be the subgroup of upper-triangular matrices, and let $`A_0P_0`$ be the subgroup of diagonal matrices with each entry positive. For each decomposition $`n=n_1+\mathrm{}+n_k`$ with $`n_i>0`$, we have a standard parabolic subgroup
$$P=\left\{\left(\begin{array}{ccc}P_1& \mathrm{}& \\ & \mathrm{}& \mathrm{}\\ 0& & P_k\end{array}\right)\right|P_i\mathrm{GL}_{n_i}(),det(P_i)=1\}.$$
We fix a Langlands decomposition $`P=M_PA_PN_P`$ as follows: $`M_P`$ is the subgroup of block diagonal matrices, with each block an element of $`\mathrm{SL}_{n_i}^\pm ()`$; $`A_PP`$ is the subgroup with the $`i`$th block of the form $`a_iI_{n_i}`$, where $`a_i>0`$ and $`I_{n_i}`$ is the $`n_i\times n_i`$ identity; and $`N_PP`$ is the subgroup with the $`i`$th block equal to $`I_{n_i}`$. We transfer these decompositions to all rational parabolic subgroups by conjugation.
### 2.2.
Let $`𝔞_0`$, $`𝔞_P`$ be the Lie algebras of the groups $`A_0`$, $`A_P`$. Let $`\stackrel{ˇ}{𝔞}_0`$, $`\stackrel{ˇ}{𝔞}_P`$ be their $``$-duals, and denote the pairing by $`,`$. Let $`R=R^+R^{}\stackrel{ˇ}{𝔞}_0`$ be the roots of $`G`$, and let $`\mathrm{\Delta }R^+`$ be the standard set of simple roots. For any root $`\alpha `$, let $`\stackrel{ˇ}{\alpha }`$ be the corresponding coroot. For any parabolic subgroup $`P`$, let $`\rho _P=1/2_{\alpha R^+\stackrel{ˇ}{𝔞}_P}\alpha `$.
We recall the definition of the *height function* $`H_P:P𝔞_P`$. Given $`pP`$, write $`p=man`$, where $`mM_P`$, $`aA_P`$, and $`nN_P`$. Then $`H_P(p)`$ is defined via
$$e^{\chi ,H_P(p)}=a^\chi ,\text{for all }\chi \stackrel{ˇ}{𝔞}_P\text{.}$$
Using an Iwasawa decomposition $`G=PK`$, we extend the height function to a map $`H_P:G𝔞_P`$ by setting $`H_P(g)=H_P(p)`$, where $`g=pk`$, $`pP`$, $`kK`$.
### 2.3.
Fix a parabolic subgroup $`P`$, and let $`\mathrm{\Gamma }_P=\mathrm{\Gamma }P`$. Let $`fC^{\mathrm{}}(A_PN_P\backslash G)`$ be a $`\mathrm{\Gamma }_P`$-invariant, $`K`$-finite function such that for each $`gG`$, the function $`mf(mg)`$, $`mM_P`$, is a square-integrable automorphic form on $`M_P`$ with respect to $`\mathrm{\Gamma }_PM_P`$. Let $`\lambda (\stackrel{ˇ}{𝔞}_P)`$, and let $`gG`$.
###### Definition 2.4.
The *Eisenstein series* associated to the above data is
$$E_P(f,\lambda ,g)=\underset{\gamma \mathrm{\Gamma }_P\backslash \mathrm{\Gamma }}{}e^{\rho _P+\lambda ,H_P(\gamma g)}f(\gamma g).$$
It is known \[15, Lemma 4.1\] that this series converges absolutely and uniformly on compact subsets of $`G\times C`$, where
(1)
$$C=\left\{\lambda \right|\mathrm{}\lambda ,\stackrel{ˇ}{\alpha }>\rho _P,\stackrel{ˇ}{\alpha },\text{for all }\alpha \mathrm{\Delta }\};$$
here $`\mathrm{}`$ denotes real part.
## 3. Modular Symbols
### 3.1.
We recall the definition of modular symbols. Our definition is equivalent to that of Ash and Ash-Borel , but we need a slightly different formulation for our purposes.
Let $`V=^n`$ with the canonical $`G()`$-action. Let $`𝐰`$ be a tuple of subspaces $`(W_1,\mathrm{},W_k)`$, where $`W_iV`$. The *type* of $`𝐰`$ is the tuple $`(dimW_1,\mathrm{},dimW_k)`$. The tuple $`𝐰`$ is called *full* if $`dimW_i=n`$, and is called a *splitting* if $`V=_iW_i`$. Any splitting determines a rational flag
$$F_𝐰=\left\{\{0\}F_1\mathrm{}F_kV\right\}$$
by $`F_j=_{ij}W_i`$, and thus determines a rational parabolic subgroup $`P_𝐰`$, the stabilizer of $`F_𝐰`$. We abuse notation slightly and write $`P_𝐰=M_𝐰A_𝐰N_𝐰`$ for the associated Langlands decomposition. It is easy to check that, with our fixed decomposition, $`M_𝐰()`$ preserves each $`W_i`$.
### 3.2.
Let $`X`$ be the symmetric space $`G/K`$, and let $`\overline{X}`$ be the bordification of $`X`$ constructed by Borel-Serre . Then the cohomology $`H^i(\mathrm{\Gamma };)`$ may be identified with $`H^i(\mathrm{\Gamma }\backslash X;)`$ and $`H^i(\mathrm{\Gamma }\backslash \overline{X};)`$.
Let $`Y=\mathrm{\Gamma }\backslash X`$, $`\overline{Y}=\mathrm{\Gamma }\backslash \overline{X}`$, $`\overline{Y}=\overline{Y}Y`$, and let $`\pi :XY`$ be the canonical projection. Let $`d=(n^2+n)/21`$ be the dimension of $`Y`$. For all $`i`$, Lefschetz duality gives an isomorphism
$$H_{di}(\overline{Y},\overline{Y};)H^i(\mathrm{\Gamma };).$$
### 3.3.
Let $`𝐰`$ be a splitting, and let $`K_𝐰`$ be $`KM_𝐰A_𝐰`$. The inclusion $`M_𝐰A_𝐰G`$ induces a proper map
$$\iota :M_𝐰A_𝐰/K_𝐰X.$$
Let $`Y_𝐰`$ be the closure of $`(\pi \iota )(M_𝐰A_𝐰/K_𝐰)`$, and let $`d(𝐰)`$ be the dimension of $`Y_𝐰`$.
###### Definition 3.4.
Let $`𝐰=(W_1,\mathrm{},W_k)`$ be a full tuple of subspaces. Then the *modular symbol* $`\mathrm{\Xi }_𝐰`$ associated to $`𝐰`$ is defined as follows:
1. If $`𝐰`$ is a splitting, then $`\mathrm{\Xi }_𝐰H_{d(𝐰)}(\overline{Y},\overline{Y};)`$ is the fundamental class of $`Y_𝐰`$.
2. Otherwise, $`\mathrm{\Xi }_𝐰`$ is defined to be $`0H_{d(𝐰)}(\overline{Y},\overline{Y};)`$, where $`d(𝐰)`$ is the homological degree determined by any splitting with the same type as $`𝐰`$.
### 3.5.
We define a $`G()`$-action on tuples as follows. Given a full tuple $`𝐰=(W_1,\mathrm{},W_k)`$, let $`g𝐰`$ be the tuple $`(W_1,gW_2,\mathrm{},gW_k)`$. By abuse of notation we write $`g\mathrm{\Xi }_𝐰`$ for the modular symbol $`\mathrm{\Xi }_{g𝐰}`$.
Note that this is not a $`G()`$-action on modular symbols, since associativity does not hold. However, the definition $`g\mathrm{\Xi }_𝐰`$ will suffice for our construction.
Note also that $`g\mathrm{\Xi }_𝐰`$ is different from the modular symbol obtained via the natural $`G()`$-action defined by left translation of all subspaces in a tuple. In particular, let $`\gamma \mathrm{\Gamma }`$, and let $`𝐰^{}=(\gamma W_1,\mathrm{},\gamma W_k)`$. Then $`\mathrm{\Xi }_𝐰=\mathrm{\Xi }_𝐰^{}`$, but $`\mathrm{\Xi }_𝐰\gamma \mathrm{\Xi }_𝐰`$ in general.
## 4. Eisenstein series twisted by modular symbols
### 4.1.
Let $`𝐰=(W_1,\mathrm{},W_k)`$ be a full tuple of subspaces, and let $`P`$ be a rational parabolic subgroup. We say that $`P`$ and $`𝐰`$ are *compatible* if the following conditions hold: there is a splitting $`𝐰^{}=(W_1^{},\mathrm{},W_k^{})`$ such that $`P=P_𝐰^{}`$, the types of $`𝐰`$ and $`𝐰^{}`$ are equal, and $`W_1=W_1^{}`$.
Fix a rational parabolic subgroup $`P`$ and a compatible splitting $`𝐰`$. Let $`f`$, $`\lambda `$ be as in §2.3, and let $`\phi `$ be a $``$-valued linear form on $`H_{d(𝐰)}(\overline{Y},\overline{Y};)`$.
###### Definition 4.2.
The *twisted Eisenstein series* associated to the above data is
(2)
$$E_{P,\phi }^{}=E_{P,\phi }^{}(f,\lambda ,g,𝐰)=\underset{\gamma \mathrm{\Gamma }_P\backslash \mathrm{\Gamma }}{}\phi (\gamma \mathrm{\Xi }_𝐰)e^{\rho _P+\lambda ,H_P(\gamma g)}f(\gamma g).$$
We refer to §5 for examples of this series, and for a comparison with the construction in .
###### Proposition 4.3.
The series in (2) is well-defined.
###### Proof.
Let $`𝐰=(W_1,\mathrm{},W_k)`$ and let $`\gamma \mathrm{\Gamma }`$. We need to show that the modular symbol $`\gamma \mathrm{\Xi }_𝐰`$ depends only on the coset $`\mathrm{\Gamma }_P\gamma `$.
First we assume $`\gamma 𝐰`$ is a splitting. By the remarks at the end of §3.5, if $`\gamma \mathrm{\Gamma }`$ and $`𝐰^{}=(\gamma W_1,\mathrm{},\gamma W_k)`$ is the tuple obtained by left translation, then $`\mathrm{\Xi }_𝐰=\mathrm{\Xi }_𝐰^{}`$. From this it follows that if $`\gamma _P\mathrm{\Gamma }_P`$, then $`(\gamma _P\gamma )\mathrm{\Xi }_𝐰=\mathrm{\Xi }_𝐰`$. Indeed, $`\gamma _P\mathrm{\Xi }_𝐰=\mathrm{\Xi }_{𝐰^{\prime \prime }}`$, where $`𝐰^{\prime \prime }=(\gamma _P^1W_1,W_2,\mathrm{},W_k)`$, and any element of $`\mathrm{\Gamma }_P`$ preserves $`W_1`$.
Now assume that $`\gamma 𝐰`$ isn’t a splitting. There are two possibilities: (1) $`\gamma W_i\gamma W_j\{0\}`$ for some $`i,j>1`$; (2) $`\gamma W_i\gamma W_j=\{0\}`$ for all $`i,j>1`$ and $`W_1\gamma W_j\{0\}`$ for some $`j`$. In the first case, we have $`\gamma \mathrm{\Xi }_𝐰=0`$ for all $`\gamma `$, so the Eisenstein series is identically $`0`$. In the second case, we have $`(\gamma _P\gamma )\mathrm{\Xi }_w=0`$ for all $`\gamma _P\mathrm{\Gamma }_P`$, since left translation of the tuple $`\gamma _P\gamma 𝐰`$ by $`\gamma _P^1`$ preserves the incidence conditions satisfied by the $`W_i`$. This completes the proof.
### 4.4.
For the rest of this note, we will assume that $`P`$ is the minimal parabolic subgroup $`P_0`$, and will take $`f1`$. Although the functions $`E_{P,\phi }^{}`$ are not automorphic, a certain sum of them is.
###### Proposition 4.5.
Let $`W_i`$, $`i=0,\mathrm{},n`$ be $`1`$-dimensional subspaces of $`V`$, and let $`𝐰(i)`$ be the tuple $`(W_0,\mathrm{},\widehat{W}_i,\mathrm{},W_n)`$, where $`\widehat{W}_i`$ means delete $`W_i`$. Then
$$\phi (\mathrm{\Xi }_{𝐰(0)})E_P(f,\lambda ,g)=\underset{i=1}{\overset{n}{}}(1)^{i+1}E_{P,\phi }^{}(f,\lambda ,g,𝐰(i)).$$
###### Proof.
First, the twisted series on the right are well-defined, since if $`P`$ and $`𝐰`$ are compatible then so are $`P`$ and $`𝐰(i)`$ for each $`i1`$. We have the following basic relation among modular symbols for the minimal parabolic subgroup, from :
$$\mathrm{\Xi }_{𝐰(0)}=\underset{i=1}{\overset{n}{}}(1)^{i+1}\mathrm{\Xi }_{𝐰(i)}.$$
Note that the relations in imply that this equality holds true in $`H_{d(𝐰)}(Y_𝐰,Y_𝐰;)`$ for *any* collection of $`1`$-dimensional rational subspaces $`(W_0,\mathrm{},W_n)`$, even with the possibility that some $`𝐰(i)`$ aren’t splittings. The result follows immediately from Definition 4.2 and the fact that if $`𝐰^{}=(\gamma W_1,\mathrm{},\gamma W_n)`$ with $`\gamma \mathrm{\Gamma }`$, then $`\mathrm{\Xi }_𝐰^{}=\mathrm{\Xi }_{𝐰(0)}`$
###### Theorem 4.6.
Let $`P`$ be the minimal parabolic subgroup $`P_0`$, and let $`𝐰`$ be a compatible splitting. Let $`\phi `$ be a linear form on $`H_{n1}(\overline{Y},\overline{Y};)`$. Then the series (4.2) converges uniformly on compact subsets of $`G\times C`$, where $`C`$ is the cone (1).
###### Proof.
We begin by recalling some facts from the theory of modular symbols associated to the minimal parabolic subgroup. These facts are equivalent to results in , and are just reformulated in terms of tuples and splittings.
Let $`𝒲`$ be the set of all full tuples of $`1`$-dimensional subspaces. We define a function $`:𝒲`$ as follows. From each $`1`$-dimensional subspace $`W`$, we choose and fix a primitive vector $`v(W)^n`$. Then we set
$$𝐰=|det(v(W_1),\mathrm{},v(W_n))|.$$
Let $`𝒲_u𝒲`$ be the subset of tuples for which $`𝐰=1`$. The set $`\mathrm{\Gamma }\backslash 𝒲_u`$ is finite, where $`\mathrm{\Gamma }`$ acts by left translations. One can show that any modular symbol $`\mathrm{\Xi }_𝐰`$ can be written as a sum
(3)
$$\mathrm{\Xi }_𝐰=\underset{𝐰^{}S}{}\mathrm{\Xi }_𝐰^{},$$
where $`S`$ is a finite subset of $`𝒲_u`$ (depending on $`\mathrm{\Xi }_𝐰`$). Moreover, the cardinality of $`S`$ is bounded by $`p(\mathrm{log}𝐰)`$, where $`p`$ is a polynomial depending only on $`n`$ .
Let $`\gamma \mathrm{\Gamma }`$ and consider the modular symbol $`\gamma \mathrm{\Xi }_𝐰`$. Since $`𝐰`$ is compatible with $`P`$, the space $`W_1`$ is the span of the first basis element of $`V`$. Let us assume for the moment that for $`i>1`$, $`W_i`$ is the span of the $`i`$th standard basis element of $`V`$. This implies that $`\gamma 𝐰`$ is the absolute value of the determinant of a fixed $`(n1)\times (n1)`$ minor of $`\gamma `$. Hence
(4)
$$\gamma 𝐰<<\mathrm{max}\left\{|\gamma _{ij}|^{n1}\right|1i,jn\},$$
where the implied constant depends only on $`n`$.
Let $`M(\gamma )`$ be the right hand side of (4). It follows that there is a polynomial $`p_1`$, depending only on $`n`$, such that
$$p(\mathrm{log}\gamma 𝐰)<p_1(\mathrm{log}M(\gamma )).$$
Now consider the value $`\phi (\gamma \mathrm{\Xi }_𝐰)`$. Since $`\mathrm{\Gamma }\backslash 𝒲_u`$ is finite, there is a maximum value $`\phi _{\text{max}}`$ that $`|\phi |`$ attains on this set. Writing $`I(\gamma ,\lambda )=\mathrm{exp}(\rho _P+\lambda ,H_P(\gamma g))`$, we have
(5)
$$\underset{\gamma \mathrm{\Gamma }_P\backslash \mathrm{\Gamma }}{}|\phi (\gamma \mathrm{\Xi }_𝐰)I(\gamma ,\lambda )|<<\underset{\gamma \mathrm{\Gamma }_P\backslash \mathrm{\Gamma }}{}|p_1(\mathrm{log}M(\gamma ))I(\gamma ,\lambda )|,$$
where the implied constant depends on $`n`$ and $`\phi _{\text{max}}`$. The right of (5) has the same convergence properties as the usual Eisenstein series, and so the proof is complete under our assumption on $`𝐰`$.
Now assume $`W_i`$ is a general $`1`$-dimensional subspace of $`V`$ for $`i>1`$. Let $`v(W_i)_j`$ be the $`j`$th coordinate of $`v(W_i)`$, and let
$$M(𝐰)=\mathrm{max}\left\{|v(W_i)_j|\right|1i,jn\}.$$
Then
$$\gamma 𝐰<<M(\gamma ),$$
where the implied constant depends on $`n`$ and $`M(𝐰)`$. The rest of the proof proceeds as above.
## 5. Examples
### 5.1.
In this section we continue to assume that $`P`$ is the minimal parabolic subgroup $`P_0`$. We begin by discussing the connection between the construction in this note and that of .
Let $`\mathrm{}`$ be a positive integer, let $`G=\mathrm{SL}_2()`$, and let $`\mathrm{\Gamma }=\mathrm{\Gamma }_0(\mathrm{})`$. The space $`X=\mathrm{SL}_2()/\mathrm{SO}_2()`$ is the upper halfplane $`𝔥`$, and we let $`𝔥^{}=𝔥^1()`$ be the usual partial compactification obtained by adjoining cusps. Given a pair of cusps $`(q_1,q_2)`$, we can determine a full tuple $`(W_1,W_2)`$ by setting $`W_i`$ to be the subspace of $`^2`$ corresponding to the point $`q_i^1()`$. Slightly abusing notation, we denote the corresponding modular symbol by $`\mathrm{\Xi }(q_1,q_2)`$.
### 5.2.
We can construct an interesting linear form on the modular symbols as follows. Let $`f`$ be a fixed weight two holomorphic cuspform on $`\mathrm{\Gamma }`$. Then we set
$$\phi (\mathrm{\Xi }(q_1,q_2))=2\pi i_{q_1}^{q_2}f(z)𝑑z,$$
where the integration is taken along the ideal geodesic from $`q_1`$ to $`q_2`$. Note that if $`f`$ is a newform, then $`\phi (\mathrm{\Xi }(\mathrm{},0))`$ is the special value $`L(1,f)`$.
To compute the series (2), let $`\mathrm{\Gamma }_{\mathrm{}}=\mathrm{\Gamma }P`$, and let $`\mathrm{}:𝔥`$ be the imaginary part. Let $`\alpha \stackrel{ˇ}{𝔞}_0`$ be the standard positive root, so that $`\rho _P=\alpha /2`$. Write $`\lambda =t\alpha `$, where $`t`$. It is easy to check that $`e^{\lambda +\rho _P,H_P(g)}=\mathrm{}(z)^{t+1/2}`$, where $`z𝔥`$ is the point corresponding to $`g`$. Setting $`(q_1,q_2)=(\mathrm{},0)`$, we see that the corresponding tuple $`𝐰`$ is compatible with $`P`$. We obtain
(6)
$$E_{P,\phi }^{}(\lambda ,g,𝐰)=E_{P,\phi }^{}(t,z,𝐰)=\underset{\gamma \mathrm{\Gamma }_{\mathrm{}}\backslash \mathrm{\Gamma }}{}\phi (\gamma \mathrm{\Xi }_𝐰)\mathrm{}(\gamma z)^{t+1/2},t.$$
By Theorem 4.6, this converges for $`\mathrm{}t>1/2`$.
### 5.3.
To relate this to , we recall the pairing between classical modular symbols and cuspforms. One fixes a point $`z_0𝔥^{}`$, and defines a map
$`[]_f:`$ $`\mathrm{\Gamma }`$
$`\gamma 2\pi i{\displaystyle _{z_0}^{\gamma z_0}}f(z)𝑑z.`$
(In , this map is written as $`\gamma \gamma ,f`$.) One can show that this map is independent of $`z_0`$, vanishes on $`\mathrm{\Gamma }_{\mathrm{}}`$, and satisfies
$$[\gamma \gamma ^{}]_f=[\gamma ]_f+[\gamma ^{}]_f,\text{for }\gamma ,\gamma ^{}\mathrm{\Gamma }\text{.}$$
. Then the series in is defined by
(7)
$$E^{}(z,s)=\underset{\gamma \mathrm{\Gamma }_{\mathrm{}}\backslash \mathrm{\Gamma }}{}[\gamma ]_f\mathrm{}(\gamma z)^s,s,$$
which converges for $`\mathrm{}s>1`$.
To compare this with (6), let $`q_1=\mathrm{}`$ and $`q_2=z_0=0`$, and put $`s=t+1/2`$. Since
$$_{q_1}^{q_2}f+_{q_2}^{\gamma q_2}f=_{q_1}^{\gamma q_2}f,$$
we find
$$\phi (\mathrm{\Xi }(q_1,q_2))E(z,s)+E^{}(z,s)=E_{P,\phi }^{}(s1/2,z,𝐰),$$
where
$$E(z,s)=\underset{\gamma \mathrm{\Gamma }_{\mathrm{}}\backslash \mathrm{\Gamma }}{}\mathrm{}(\gamma z)^s$$
is the classical nonholomorphic Eisenstein series.
In it is shown that $`E^{}`$ is “automorphic up to a shift.” Precisely, if $`\gamma \mathrm{\Gamma }`$, then
$$E^{}(\gamma z,s)=E^{}(z,s)[\gamma ]_fE(z,s).$$
This is easily seen to be equivalent to Proposition 4.5 above.
### 5.4.
Now let $`G=\mathrm{SL}_3()`$, and let $`\mathrm{\Gamma }=\mathrm{\Gamma }_0(\mathrm{})`$. This is the arithmetic group defined to be the subgroup of $`G()`$ consisting of matrices with bottom row congruent to $`(0,0,)`$ mod $`\mathrm{}`$. The symmetric space $`X=\mathrm{SL}_3()/\mathrm{SO}_3()`$ is a $`5`$-dimensional smooth noncompact manifold, and our modular symbols live in $`H_2(\overline{Y},\overline{}Y;)`$.
To construct an interesting linear form on these modular symbols, we may use elements of the *cuspidal cohomology* $`H_{\mathrm{cusp}}^3(\mathrm{\Gamma };)`$. These are classes that, via the de Rham isomorphism, correspond to $`\mathrm{\Gamma }^{}`$-invariant differential forms $`\omega =_If_Id\omega _I`$, where the coefficients are cusp forms and $`\mathrm{\Gamma }^{}\mathrm{\Gamma }`$ is a torsionfree subgroup of finite index. In this context, $`H_{\mathrm{cusp}}^3(\mathrm{\Gamma };)`$ can alternatively be defined to be the kernel of the restriction map $`H^3(\overline{Y};)H^3(\overline{Y};)`$. We refer to for details.
### 5.5.
To explicitly construct classes in $`H_{\mathrm{cusp}}^3(\mathrm{\Gamma };)`$ that can be paired with modular symbols, we may use techniques of . There it is shown that $`H_{\mathrm{cusp}}^3(\mathrm{\Gamma };)`$ is isomorphic to a space $`W(\mathrm{\Gamma })`$ of functions $`f:^2(/\mathrm{})`$ satisfying certain relations \[4, Summary 3.23\]. A modular symbol $`\mathrm{\Xi }_𝐰`$ modulo $`\mathrm{\Gamma }`$ gives rise to a point $`p_𝐰^2(/\mathrm{})`$ by taking the bottom row of the matrix $`(v(W_1),v(W_2),v(W_3))`$ \[4, Prop. 3.12\]. Hence given an element $`\alpha H_{\mathrm{cusp}}^3(\mathrm{\Gamma };)`$ corresponding to a function $`f_\alpha W(\mathrm{\Gamma })`$, we obtain a linear form by setting
$$\phi (\mathrm{\Xi }_𝐰)=f_\alpha (p_𝐰).$$
This linear form is induced from the intersection pairing
$$H_3(\overline{Y})\times H_2(\overline{Y});$$
we refer to \[4, Prop. 3.24\] for details.
### 5.6.
For an explicit example, we may take $`\mathrm{}=53`$. This is the first level for which the cuspidal cohomology is nonzero; one finds that $`dimH_{\mathrm{cusp}}^3(\mathrm{\Gamma }_0(53);)=2`$. A sample element is given as a function in $`W(\mathrm{\Gamma })`$ in Table II of .
To compute $`E_{P,\phi }^{}`$, we may take $`\alpha H_{\mathrm{cusp}}^3(\mathrm{\Gamma }_0(53);)`$ to be a Hecke eigenclass. For a prime $`p`$ with $`(p,53)=1`$, the local $`L`$-factor of the representation corresponding to $`\alpha `$ has the form
$$(1a_pp^s+\overline{a}_pp^{12s}p^{33s})^1,$$
where $`s`$ and $`a_p`$ is the eigenvalue of a certain Hecke operator. If we fix an algebraic integer $`\rho `$ satisfying $`\rho ^2=11`$, we find that for our Hecke eigenclass
$$a_2=2\rho ,a_3=1+\rho ,a_5=1,a_7=3,\mathrm{}$$
If we represent $`\alpha `$ using a function $`fW(\mathrm{\Gamma })`$, and apply the formulæ in \[9, Ch. V and VII\], we can obtain a very explicit expression for $`E_{P,\phi }^{}`$.
In contrast to the $`\mathrm{SL}_2`$ case, the twisted Eisenstein series on $`\mathrm{SL}_3`$ isn’t simply automorphic up to a shift. If we consider the relation in Proposition 4.5, we see that a certain sum of *three* twisted Eisenstein series is equal to an automorphic function.
|
warning/0004/hep-ph0004020.html
|
ar5iv
|
text
|
# Neutrino Masses and Leptonic CP Violation
## I Introduction
The quark masses and the related flavor mixings are the most intriguing riddles to be understood in the Standard Model (SM). Within the SM, the quark masses and flavor mixing angles are not predictable. The flavor mixing angles arise since the quark states for which the weak interaction is diagonal are not mass eigenstates. Moreover, degenerate quarks of a given charge render the flavor mixing angles physically meaningless. Thanks to that fact, we get a hint that the flavor mixing angles can be related to the elements of quark mass matrix. As an attempt to provide any relationship between the flavor mixing angles and the elements of quark mass matrix, mass matrix ansatz has been suggested . With the help of a mass matrix ansatz, we may predict some free parameters in the SM.
On the other hand, recent neutrino experimental results and cosmological observations provide evidence for non-zero neutrino masses and the possible lepton flavor mixings. Then, the SM has to be enlarged and we have more free parameters to describe all fermion masses and their mixing angles. In this case, one may also reduce the number of free parameters by using separate lepton mass matrix ansatz. If it is possible to provide some quark-lepton symmetry in the quark and lepton mass matrices, one may reduce the number of free parameters much more. In recent work , we showed that the hierarchical quark mixing pattern as well as bimaximal lepton mixing pattern can arise from one single particular mass matrix based on the permutation symmetry with suitable breaking. Remarkably those different mixing patterns could be generated by using the same texture of the mass matrices for quarks and leptons but with different hierarchies. However, although non-zero neutrino masses and mixings can be interpreted as a solution to the solar and the atmospheric neutrino anomalies, the present neutrino experimental results do not pin down the values of neutrino masses and mixing angles in three-neutrino oscillation scheme. Moreover, the solution for the solar neutrino deficit may be either small or large mixing with different mass squared differences depending on whether we consider the matter effect (Mikheyev, Smirnov, and Wolfensten (MSW) effect) or not. Thus, one can at best estimate the hierarchy of neutrino mass patterns and their mixings case by case. In this work, we show in more detail how the lepton flavor mixing matrix can be obtained from a specific form of lepton mass matrix by assuming quark-lepton symmetry for the fermion
small mixing angle solutions for the solar neutrino deficit combined with the large mixing solution for the Super-Kamiokande atmospheric neutrino anomaly.
In addition, it will be very interesting to study the possible CP violation in the lepton sector, which arises due to the non-vanishing CP phase in the flavor mixing matrix . To do this, we will calculate the invariant leptonic CP violating quantity $`J_{\text{CP}}^l`$ from the phenomenological lepton flavor mixing matrix. As will be shown later, the invariant quantity, $`J_{\text{CP}}^l`$, depends on the neutrino masses as well as the CP phase. From the estimate of the neutrino mass bounds based on the neutrino experimental results, we will provide the possible range of $`J_{\text{CP}}^l`$.
## II Neutrino Mixing Matrix with a CP Violating Phase
Let us begin by assuming that the form of lepton mass matrix can be derived from the mass matrix ansatz based on the permutation symmetry with suitable breaking which is used in the quark sector . As shown in Ref. , the mass matrix has the following form:
$$M_H=\left(\begin{array}{ccc}0& A& 0\\ A& D& B\\ 0& B& C\end{array}\right).$$
(1)
The parameters $`A,B,C`$ and $`D`$ can be expressed in terms of the fermion mass eigenvalues and one free parameter $`ϵ^l`$. One can take the mass eigenvalues to be $`m_1,m_2`$ and $`m_3`$ with the following three conditions:
$`Tr(M_H)`$ $`=`$ $`m_1+m_2+m_3,`$ (2)
$`Det(M_H)`$ $`=`$ $`m_1m_2m_3,`$ (3)
$`\mathrm{and}Tr(M_H^2)`$ $`=`$ $`m_1^2+m_2^2+m_3^2.`$ (4)
The sign of the fermion mass is irrelevant since it can be changed by a chiral transformation. From those relations, we obtain the following form of fermion mass matrix:
$$M=\left(\begin{array}{ccc}0& \sqrt{\frac{m_1m_2m_3}{m_3ϵ^l}}& 0\\ \sqrt{\frac{m_1m_2m_3}{m_3ϵ^l}}& m_2m_1+ϵ^l& \zeta (m_2m_1+ϵ^l)\\ 0& \zeta (m_2m_1+ϵ^l)& m_3ϵ^l\end{array}\right),$$
(5)
in which the analytic relation between two parameters $`ϵ^l`$ and $`\zeta `$ is given by
$$\zeta ^2=\frac{ϵ^l(m_3m_2+m_1ϵ^l)(m_3ϵ^l)ϵ^lm_1m_2}{(m_3ϵ^l)(m_2m_1+ϵ^l)^2}.$$
(6)
With the help of the analytic form of the orthogonal matrix $`U`$ presented in Ref. , the real symmetric mass matrix $`M`$ can be diagonalized. The mass matrices for charged leptons and neutrinos have the same form of the mass matrix (3). In particular, notice that the parameter $`ϵ^l`$ will be taken to be identical in both the charged lepton mass matrix and the neutrino mass matrix, and will be determined from the neutrino experimental results. However, the parameters $`\zeta `$ are different in the two mass matrices because they depend on their fermion masses. Then, the neutrino mass matrix $`M_\nu `$ and charged lepton mass matrix $`M_l`$ can be brought to diagonal forms by the real unitary matrices $`U_\nu `$ and $`U_l`$,
$`U_\nu M_\nu U_\nu ^{}`$ $`=`$ $`\text{diag}(m_1,m_2,m_3),`$ (7)
$`U_lM_lU_l^{}`$ $`=`$ $`\text{diag}(m_e,m_\mu ,m_\tau ),`$ (8)
where $`m_1,m_2,`$ and $`m_3`$ are neutrino masses from now on. The lepton flavor mixing matrix $`V_{\text{CKM}}^l`$ is related to $`U_\nu `$ and $`U_l`$ as follows:
$$V_{\text{CKM}}^l=PU_lP^1U_\nu ^T,$$
(9)
where the phase matrix is $`P=\mathrm{diag}(e^{i\delta ^l},1,1)`$. More generally, we can also take the phase matrix, $`P`$, as $`\mathrm{diag}(e^{i\delta _1},e^{i\delta _2},e^{i\delta _3})`$. One may eliminate the phase $`\delta _3`$ by a phase transformation of fields. Because of the hierarchy of the charged lepton masses, Eq. (9) contains only the combination of phases in the form, $`\delta _1\delta _2`$, which will be identified as $`\delta ^l`$. To see easily how the lepton mixing pattern is related to the lepton mass hierarchy, first of all, we present the lepton flavor mixing matrix in the leading approximation. In the next section, the exact form derived from Eq. (5) will be used to determine the magnitudes of the elements of the mixing matrix.
Since the charged lepton family has pronounced mass hierarchy $`m_em_\mu m_\tau `$, the charged lepton mass matrix can be presented in the approximate form as
$$M_l\left(\begin{array}{ccc}0& \sqrt{m_em_\mu }& 0\\ \sqrt{m_em_\mu }& m_\mu & \sqrt{ϵ^lm_\tau }\\ 0& \sqrt{ϵ^lm_\tau }& m_\tau \end{array}\right).$$
(10)
From the unitary transformation, we obtain the approximate form of the matrix $`U_l`$ as follows
$$U_l\left(\begin{array}{ccc}1& \sqrt{\frac{m_e}{m_\mu }}& 0\\ \sqrt{\frac{m_e}{m_\mu }}& 1& \sqrt{\frac{ϵ^l}{m_\tau }}\\ 0& \sqrt{\frac{ϵ^l}{m_\tau }}& 1\end{array}\right),$$
(11)
where we assumed that the parameter $`ϵ^lm_\tau `$. On the other hand, the neutrino mass matrix can be obtained from the mixing pattern among three neutrinos and their mass hierarchy. As shown in Ref. , the large mixing between $`\nu _\mu `$ and $`\nu _\tau `$, which is a solution for the atmospheric neutrino anomaly, can be achieved by taking $`ϵ^lm_3/2`$ and $`m_1,m_2m_3`$ in Eq. (5). We also note that the large (small) mixing between $`\nu _e`$ and $`\nu _\mu `$, which is a solution for the solar neutrino deficit, can be obtained by taking $`m_1m_2(m_1<<m_2)`$. Keeping the next-to-leading order, the neutrino mass matrix, Eq. (5), becomes
$$M_\nu \left(\begin{array}{ccc}0& \sqrt{2m_1m_2}& 0\\ \sqrt{2m_1m_2}& \frac{m_3}{2}\left(1+2\frac{m_2m_1}{m_3}\right)& \frac{m_3}{2}\left(1\frac{m_2m_1}{m_3}\right)\\ 0& \frac{m_3}{2}\left(1\frac{m_2m_1}{m_3}\right)& \frac{m_3}{2}\end{array}\right),$$
(12)
and the form of $`U_\nu `$ is approximately given by
$$U_\nu \left(\begin{array}{ccc}w_2\left(1+\frac{m_1}{2m_3}\right)& \frac{w_1}{\sqrt{2}}\left(1+\frac{m_1}{2m_3}\right)& \frac{w_1}{\sqrt{2}}\left(1\frac{m_2}{m_3}\frac{m_1}{2m_3}\right)\\ w_1\left(1\frac{m_2}{2m_3}\right)& \frac{w_2}{\sqrt{2}}\left(1\frac{m_2}{2m_3}\right)& \frac{w_2}{\sqrt{2}}\left(1+\frac{m_2}{2m_3}+\frac{m_1}{m_3}\right)\\ \sqrt{\frac{m_1m_2}{m_3^2}}& \frac{1}{\sqrt{2}}\left(1+\frac{m_2m_1}{2m_3}\right)& \frac{1}{\sqrt{2}}\left(1\frac{m_2m_1}{2m_3}\right)\end{array}\right),$$
(13)
where
$$w_1\sqrt{\frac{m_1}{m_1+m_2}}\mathrm{and}w_2\sqrt{\frac{m_2}{m_1+m_2}}\mathrm{with}w_1^2+w_2^2=1.$$
From Eqs. (5), (7) and (9), the lepton flavor mixing matrix is expressed in the leading order in terms of the lepton masses, $`w_1`$, $`w_2`$, and the CP phase $`\delta ^l`$:
$$V_{\text{CKM}}^l\left(\begin{array}{ccc}w_2+w_1\sqrt{\frac{m_e}{2m_\mu }}e^{i\delta ^l}& w_1w_2\sqrt{\frac{m_e}{2m_\mu }}e^{i\delta ^l}& \sqrt{\frac{m_e}{2m_\mu }}e^{i\delta ^l}\\ \frac{w_1}{\sqrt{2}}+w_2\sqrt{\frac{m_e}{m_\mu }}e^{i\delta ^l}& \frac{w_2}{\sqrt{2}}+w_1\sqrt{\frac{m_e}{m_\mu }}e^{i\delta ^l}& \frac{1}{\sqrt{2}}\\ \frac{w_1}{\sqrt{2}}& \frac{w_2}{\sqrt{2}}& \frac{1}{\sqrt{2}}\end{array}\right).$$
(14)
The CP-violating rephasing-invariant quantity, $`J_{\text{CP}}^l`$, is presented by
$`J_{\text{CP}}^lIm[V_{11}^lV_{12}^lV_{21}^lV_{22}^l]=w_1w_2{\displaystyle \frac{1}{2}}\sqrt{{\displaystyle \frac{m_e}{2m_\mu }}}\mathrm{sin}\delta ^l.`$ (15)
Now let us express the lepton flavor mixing matrix in the standard parametrization . As is well known, in the quark sector the standard parametrization is given by
$`V_{\text{CKM}}^l=\left(\begin{array}{ccc}c_{12}c_{13}& s_{12}c_{13}& s_{13}e^{i\delta _{13}}\\ s_{12}c_{23}c_{12}s_{23}s_{13}e^{i\delta _{13}}& c_{12}c_{23}s_{12}s_{23}s_{13}e^{i\delta _{13}}& s_{23}c_{13}\\ s_{12}s_{23}c_{12}c_{23}s_{13}e^{i\delta _{13}}& c_{12}s_{23}s_{12}c_{23}s_{13}e^{i\delta _{13}}& c_{23}c_{13}\end{array}\right),`$ (19)
where $`s_{ij},c_{ij}`$ stand for $`\mathrm{sin}\theta _{ij}`$ and $`\mathrm{cos}\theta _{ij}`$, respectively. One can then relate the elements of the mixing matrix in the standard parametrization to the elements of the flavor mixing matrix (14) by using the fact that the magnitudes of the mixing matrix elements and the Jarlskog rephasing-invariant quantity, $`J_{\text{CP}}^l`$, are independent of the parametrization. Then, the mixing angles, $`\theta _{ij}`$, can be expressed by
$`\mathrm{tan}\theta _{12}`$ $`=`$ $`{\displaystyle \frac{|V_{12}^l|}{|V_{11}^l|}}\sqrt{{\displaystyle \frac{w_1^2+w_2^2(m_e/2m_\mu )2w_1w_2\sqrt{m_e/2m_\mu }\mathrm{cos}\delta ^l}{w_2^2+w_1^2(m_e/2m_\mu )+2w_1w_2\sqrt{m_e/2m_\mu }\mathrm{cos}\delta ^l}}},`$ (20)
$`\mathrm{sin}\theta _{13}`$ $`=`$ $`|V_{13}^l|\sqrt{{\displaystyle \frac{m_e}{2m_\mu }}},`$ (21)
$`\mathrm{tan}\theta _{23}`$ $`=`$ $`{\displaystyle \frac{|V_{23}^l|}{|V_{33}^l|}}1.`$ (22)
The magnitude of $`V_{13}^l`$ can be constrained by the CHOOZ experimental results and it turns out to be small, i.e., $`|V_{13}^l|0.22`$. Then, the lepton mixing matrix (19) is approximately written as
$$V_{\text{CKM}}^l\left(\begin{array}{ccc}c_{12}& s_{12}& s_{13}e^{i\delta _{13}}\\ s_{12}c_{23}& c_{12}c_{23}& s_{23}\\ s_{12}s_{23}& c_{12}s_{23}& c_{23}\end{array}\right),$$
(23)
where $`\theta _{12}`$ can be either large or small, depending on the solar neutrino oscillation solution.
Taking $`w_1w_2\sqrt{1/2}`$ (i.e., $`m_1m_2`$), one can obtain the nearly bimaximal mixing, which corresponds to
$$\theta _{12}\theta _{23}\pi /4(i.e.c_{12}=s_{12}=c_{23}=s_{23}=1/\sqrt{2}),$$
$$\mathrm{sin}\theta _{13}\sqrt{m_e/2m_\mu }\mathrm{and}\delta _{13}\delta ^l.$$
Then, the mixing matrix can be written as follows:
$`V_{\text{CKM}}^l\left(\begin{array}{ccc}\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}\sqrt{\frac{m_e}{m_\mu }}e^{i\delta ^l}\\ \frac{1}{2}& \frac{1}{2}& \frac{1}{\sqrt{2}}\\ \frac{1}{2}& \frac{1}{2}& \frac{1}{\sqrt{2}}\end{array}\right).`$ (27)
Note that our mixing matrix contains possible CP-violating phase with nonzero but small
$$|V_{13}^l|(=\mathrm{sin}\theta _{13})\frac{1}{\sqrt{2}}\sqrt{\frac{m_e}{m_\mu }}0.05,$$
which is still consistent with the bound obtained from present CHOOZ experiment . However, if it turns out that the neutrino mixing pattern is exact bimaximal mixing as suggested in Refs. , the element $`|V_{13}^l|`$ would become exactly zero and then we could not see any CP violation effects in the leptonic sector.
In the limit of small mass ratio $`m_1/m_2`$, which corresponds to the small mixing angle solution of the solar neutrino oscillation, the lepton mixing matrix (23) becomes
$$V_{\text{CKM}}^l\left(\begin{array}{ccc}c_{12}& s_{12}& \frac{1}{\sqrt{2}}\sqrt{\frac{m_e}{m_\mu }}e^{i\delta ^l}\\ \frac{s_{12}}{\sqrt{2}}& \frac{c_{12}}{\sqrt{2}}& \frac{1}{\sqrt{2}}\\ \frac{s_{12}}{\sqrt{2}}& \frac{c_{12}}{\sqrt{2}}& \frac{1}{\sqrt{2}}\end{array}\right),$$
(28)
where $`\mathrm{tan}\theta _{12}\sqrt{\frac{m_1}{m_2}+\frac{m_e}{2m_\mu }\sqrt{\frac{m_1m_e}{2m_2m_\mu }}\mathrm{cos}\delta ^l}`$. We note that the mixing angle $`\theta _{12}`$ is correlated with the phase $`\delta ^l`$ in this case.
## III Neutrino Masses and Leptonic CP Violation
In order to determine the magnitudes of the elements of the neutrino mass matrix and the possible range of CP violation, we have to obtain numerical values of the neutrino mass eigenvalues. Although we cannot obtain those values exactly, some possible ranges of the neutrino mass eigenvalues can be estimated from the recent experimental results by taking reasonable assumptions.
Most data on neutrino mixings are presented in the two-neutrino scheme. The results are expressed in $`(\mathrm{\Delta }m^2,\mathrm{sin}^22\theta )`$ plot. With a proper approximation we can use the data on solar and atmospheric neutrino oscillations to make analyses for the three-neutrino scheme . First we assume that the solar neutrino problems are solved by two-neutrino vacuum oscillations of $`\nu _e\nu _\mu `$. The survival probability for solar electron-neutrino in two-neutrino mixing scheme is given by
$$P(\nu _e\nu _e)=1\mathrm{sin}^22\theta _{sol}\mathrm{sin}^2\left(\frac{\mathrm{\Delta }m_{sol}^2}{4}\frac{L}{E}\right).$$
(29)
In the case of $`m_3^2L/E1`$ and $`|V_{13}^l|1`$, the survival probability for electron-neutrino in the three-neutrino mixing scheme may be written as
$$P(\nu _e\nu _e)14|V_{11}^lV_{12}^l|^2\mathrm{sin}^2\left(\frac{\mathrm{\Delta }m_{21}^2}{4}\frac{L}{E}\right).$$
(30)
Therefore, the mass squared difference and mixing angle for solar neutrino analysis in the two-neutrino scheme are related to the mass squared difference and the standard mixing angle, $`\theta _{12}`$, in the three-neutrino scheme:
$$\mathrm{\Delta }m_{sol}^2\mathrm{\Delta }m_{21}^2,\theta _{sol}\theta _{12}.$$
(31)
If we consider the matter effect in the Sun, the survival probability for electron neutrinos, Eqs. (29,30), is no longer valid. However, we can still make the connections between two- and three-neutrino oscillation parameters by Eq. (31) in this situation; the mixing angle $`\theta _{13}`$ is small, i.e. $`|V_{13}^l|1`$, and the third neutrino mass $`m_3`$ is so large that just one resonance conversion between $`m_1`$ and $`m_2`$ neutrino mass states can take place. In this case the three-neutrino mixing scheme may effectively be reduced to the two-neutrino mixing scheme and Eq. (31) remains valid. If three neutrino masses are degenerate such that the second resonance conversion could not be negligible, or the mixing element $`|V_{13}^l|`$ is large, then we have to analyze neutrino mixing data within full three-neutrino scheme.
Likewise, we can consider the atmospheric neutrino case. The atmospheric neutrino deficit seems to be explained by oscillation between $`\nu _\mu `$ and $`\nu _\tau `$ with large mixing. The survival probability for atmospheric muon-neutrino in the two-neutrino mixing scheme is given by
$$P(\nu _\mu \nu _\mu )=1\mathrm{sin}^22\theta _{atm}\mathrm{sin}^2\left(\frac{\mathrm{\Delta }m_{atm}^2}{4}\frac{L}{E}\right).$$
(32)
In the case of $`(m_2^2m_1^2)L/E1`$, we can write the survival probability for muon-neutrino in the three-neutrino mixing scheme as follows:
$$P(\nu _\mu \nu _\mu )14|V_{23}^l|^2(1|V_{23}^l|^2)\mathrm{sin}^2\left(\frac{\mathrm{\Delta }m_{31}^2}{4}\frac{L}{E}\right).$$
(33)
The mass squared difference and mixing angle for atmospheric neutrino analysis in the two-neutrino scheme are related to the mass squared difference and the standard mixing angle, $`\theta _{23}`$, in the three-neutrino scheme:
$$\mathrm{\Delta }m_{atm}^2\mathrm{\Delta }m_{31}^2,\theta _{atm}\theta _{23}.$$
(34)
Recent Super-Kamiokande experiments show evidence for oscillation of atmospheric neutrinos. The data exhibit a zenith angle dependent deficit of muon neutrinos, which is consistent with predictions based on the two-flavor $`\nu _\mu \nu _\tau `$ oscillations. At 90% confidence level the mass squared difference and mixing angle are
$`5\times 10^4`$ $`<\mathrm{\Delta }m_{atm}^2<`$ $`6\times 10^3\mathrm{eV}^2,`$ (35)
$`0.82`$ $`<\mathrm{sin}^22\theta _{atm}`$ $`1.`$ (36)
The best fit values are $`\mathrm{\Delta }m_{atm}^22.2\times 10^3`$ eV<sup>2</sup> and $`\mathrm{sin}^22\theta _{atm}=1`$. From the mixing matrix (9) the mixing angle $`\theta _{23}`$ is expressed in terms of the neutrino masses and the parameter $`ϵ^l`$ as
$$\mathrm{tan}\theta _{23}\sqrt{\frac{ϵ^l}{m_3ϵ^l}}.$$
(37)
We can constrain the ratio, $`ϵ^l/m_3`$, from Eqs. (34) and (36)
$$0.28\frac{ϵ^l}{m_3}0.5.$$
(38)
If we take the best fit value for the mass squared difference and assume the mass hierarchy of $`m_1,m_2m_3`$, we conclude that
$`m_3\sqrt{\mathrm{\Delta }m_{atm}^2}`$ $``$ $`4.7\times 10^2\mathrm{eV},`$ (39)
$`1.3\times 10^2`$ $`ϵ^l`$ $`2.4\times 10^2\mathrm{eV}.`$ (40)
As is well known, there are three oscillation solutions of the solar neutrino problems in the two-neutrino scheme :
* the Large Mixing Angle (LMA) MSW:
$`5\times 10^6`$ $`\mathrm{\Delta }m_{sol}^2`$ $`4\times 10^5\mathrm{eV}^2,`$ (41)
$`0.4`$ $`\mathrm{sin}^22\theta _{sol}`$ $`0.9.`$ (42)
* Vacuum Oscillations (VO):
$`5\times 10^{11}`$ $`\mathrm{\Delta }m_{sol}^2`$ $`10^{10}\mathrm{eV}^2,`$ (43)
$`0.67`$ $`\mathrm{sin}^22\theta _{sol}`$ $`1.`$ (44)
* the Small Mixing Angle (SMA) MSW:
$`3.8\times 10^6`$ $`\mathrm{\Delta }m_{sol}^2`$ $`10^5\mathrm{eV}^2,`$ (45)
$`3.5\times 10^3`$ $`\mathrm{sin}^22\theta _{sol}`$ $`1.4\times 10^2.`$ (46)
The intervals are 95% Confidence Level. We will not consider the MSW solution with low mass at $`\mathrm{\Delta }m_{sol}^2=7.9\times 10^8`$ eV<sup>2</sup>, $`\mathrm{sin}^22\theta _{sol}=0.96`$. All the solutions are consistent with the predictions of the standard solar model and with the observed average event rates in the Chlorine (Homestake) experiments , Kamiokande , Super-Kamiokande , Gallium (GALLEX and SAGE ) experiments. With these results from the solar neutrino oscillation solutions, we can investigate the bounds on the neutrino masses, $`m_1,m_2`$, and CP violation quantity, $`J_{\text{CP}}^l`$, based on the mass matrix ansatz (5). Although it is difficult to severely constrain the CP-violating phase from the results of solar and atmospheric experiments, we can get the possible ranges of magnitude of $`J_{\text{CP}}^l`$ for a given non-zero CP phase. To show this in detail, we will treat three cases of the solar neutrino oscillation solutions separately.
* LMA solution
Recent experimental results from Super-Kamiokande seem to provide some encouragement for considering the LMA solution of the MSW effect . The best fit values are at $`\mathrm{\Delta }m_{sol}^210^5\mathrm{eV}^2`$ and $`\mathrm{sin}^22\theta _{sol}0.6`$. The lepton flavor mixing matrix for this solution in the leading approximation is given by Eq. (27). From Eq. (20), the neutrino mixing angle $`\theta _{12}`$ is expressed in terms of the lepton masses, $`m_e,m_\mu ,m_1,m_2`$, and the CP phase $`\delta ^l`$. Since the LMA solution is the case of $`m_1m_2`$, one may ignore relatively small terms which contain the ratio $`m_e/m_\mu `$ in Eq. (20). Then, we approximately get
$$\theta _{sol}\theta _{12}\mathrm{arctan}\left(\frac{m_1}{m_2}\right),$$
(47)
which is in turn bounded by Eq. (42). Using the bounds of the mass squared difference (41) and mixing angle $`\theta _{12}`$, we obtain numerically allowed neutrino mass bounds:
$`3.0\times 10^4`$ $``$ $`m_12.0\times 10^3\mathrm{eV},`$ (48)
$`2.7\times 10^3`$ $``$ $`m_21.5\times 10^2\mathrm{eV}.`$ (49)
With the 90% confidence limit data on neutrino oscillation parameters we can estimate the possible ranges of the magnitude of the complex mixing matrix elements from the numerical analysis based on the exact form of the mixing matrix:
$$|V_{\text{LMA}}|=\left(\begin{array}{ccc}0.820.94& 0.350.55& 0.010.10\\ 0.250.50& 0.550.77& 0.560.70\\ 0.140.34& 0.500.65& 0.700.82\end{array}\right).$$
(50)
From these results, we can calculate the quantity $`J_{\text{CP}}^l`$ as a function of the CP phase $`\delta ^l`$. In Fig. 1.(A), we present our prediction for the allowed range of $`J_{\text{CP}}^l`$ as a function of $`\delta ^l`$, which is consistent with the solar and atmospheric neutrino experimental results.
* VO solution
This solution shows that there are well-separated two mass squared difference scales, $`\mathrm{\Delta }m_{atm}^22.2\times 10^3\mathrm{eV}^2`$ and $`\mathrm{\Delta }m_{sol}^28\times 10^{11}\mathrm{eV}^2`$. The mixing angle $`\theta _{12}`$ is also determined by the neutrino masses $`m_1`$ and $`m_2`$ as in the case of LMA solution (47). From the constraints Eqs. (43) and (44), we get, at best, the lower bounds on neutrino masses as follows:
$$m_10.24\times 10^5\mathrm{eV},m_20.93\times 10^5\mathrm{eV}.$$
(51)
The limits on magnitudes of the elements of the complex mixing matrix for this solution are
$$|V_{\text{VO}}|=\left(\begin{array}{ccc}0.700.90& 0.450.71& 0.020.06\\ 0.350.60& 0.470.75& 0.540.72\\ 0.200.50& 0.400.65& 0.700.84\end{array}\right).$$
(52)
Since the rephasing-invariant $`J_{\text{CP}}^l`$ given in Eqs. (15) actually depends on the ratio $`m_1/m_2`$ and the parameter $`ϵ^l`$, we can calculate the numerical values of $`J_{\text{CP}}^l`$ with the help of Eqs. (38), (44), and (47). The results are shown in Fig. 1.(B). As one can see from Fig. 1.(B), the allowed range of $`J_{\text{CP}}^l`$ for the VO solution is almost the same as that for the LMA solution. The reason is that both solutions have almost the same neutrino mixing angle $`\theta _{12}`$ and our ansatz leads to those solutions when we take the mass hierarchy $`m_1m_2`$. Therefore the mixing matrix may also be accommodated by Eq. (27) in the leading approximation as in the case of LMA solution.
* SMA solution
The small mixing angle $`\theta _{12}`$ implies small mass ratio, $`m_1/m_2`$, in Eq. (9). Different from the above two cases, the lepton mixing matrix for the SMA solution in the leading approximation is given by Eq. (28). In this case the angle depends sensitively on the phase $`\delta ^l`$ as well as on the ratio $`m_1/m_2`$. Note that the expression (47) does not hold in the extreme mass hierarchical case of $`m_2m_1`$, because we cannot neglect the $`m_e/m_\mu `$-terms in Eq. (20). Also, the dependence of CP phase should be taken into account when we calculate the mixing angle $`\theta _{12}\theta _{sol}`$. From the numerical analysis based on the exact form of the mixing matrix, it has been shown that the mass $`m_1`$ may have small value without a lower bound in the allowed parameter space given by Eqs. (45)-(46). The upper bound of mass $`m_1`$ depends on the value of CP-violating phase $`\delta ^l`$. When $`\delta ^l`$ is in the range of $`0<\delta ^l<\pi /2`$, the upper bound of $`m_1`$ may be up to $`4.2\times 10^5`$ eV. As $`\delta ^l`$ approaches $`\pi `$, the upper bound of $`m_1`$ becomes smaller, around $`3.0\times 10^7`$ eV. From Eq. (45), the lower mass bound of $`m_2`$ is to be around $`1.9\times 10^3`$ eV. The limits on magnitudes of the elements of the complex mixing matrix for this solution are
$$|V_{\text{SMA}}|=\left(\begin{array}{ccc}0.980.99& 0.030.06& 0.030.05\\ 0.030.08& 0.700.84& 0.540.71\\ 0.010.05& 0.540.71& 0.700.83\end{array}\right).$$
(53)
Fig. 2 shows the rephasing-invariant quantity $`J_{\text{CP}}^l`$ for the SMA solution as function of the CP phase $`\delta ^l`$. The allowed range of $`J_{\text{CP}}^l`$ is presented by the hatched region which comes from the constraints Eqs. (45) and (46). The shape of allowed region is different from that of large mixing of solar neutrinos. The maximum value of $`J_{\text{CP}}^l`$ for the SMA solution is $`1.3\times 10^3`$, while that for LMA and VO is of order $`0.01`$. In particular, as can be seen from Fig. 2, the magnitude of $`J_{\text{CP}}^l`$ is suppressed and severely constrained for the range of $`\pi /2<\delta ^l<\pi `$. Also, somewhat broad range of $`J_{\text{CP}}^l`$ for $`\delta ^l1`$ is obtained.
To summarize, we analyzed neutrino masses and mixings as the solutions of the solar neutrino problems and atmospheric neutrino deficits based on a mass matrix ansatz. Recent Super-Kamiokande results for atmospheric neutrino showed that the muon neutrino deficits may be explained by large mixing between $`\nu _\mu \nu _\tau `$. The solar neutrino problems have three possible solutions: small mixing MSW, large mixing MSW, and just-so vacuum oscillation solutions. Depending on the solutions to the solar neutrino problems, we have three possible mixing matrices in the three-neutrino scheme. For each case we investigated the neutrino mass bounds, the magnitudes of mixing matrix elements, and possible nonvanishing CP-violating rephasing-invariant quantity $`J_{\text{CP}}^l`$. We conclude that LMA-MSW and VO solutions may come from the mass matrix ansatz with the similar mass hierarchy: $`m_1m_2m_3`$. And $`J_{\text{CP}}^l`$ also has almost the same magnitude in the two cases, and may reach values up to values of 0.012. The origin of SMA-MSW solution may be attributed to the mass hierarchy: $`m_1m_2m_3`$ with our mass matrix ansatz. The magnitude of $`J_{\text{CP}}^l`$ depends on the CP phase $`\delta ^l`$. In the range of $`0\delta ^l\pi /2`$, the value of $`J_{\text{CP}}^l`$ may be up to $`1.3\times 10^3`$, which is small compared to LMA-MSW or VO solution. In $`\pi /2<\delta ^l<\pi `$, $`J_{\text{CP}}^l`$ is even more suppressed.
## Acknowledgments
We thank G. Cvetic and S. Pakvasa for careful reading of the manuscript and their valuable comments. The work of C.S.K. was supported in part by Grant No. 1999-2-111-002-5 from the Interdisciplinary research program of the KOSEF, in part by the BSRI Program of MOE, Project No. 99-015-DI0032, and in part by the KRF Sughak-research program, Project No. 1997-011-D00015. The work of J.D.K. was supported in part by a grant from the Natural Science Research Institute, Yonsei University in the year 1999.
|
warning/0004/math0004031.html
|
ar5iv
|
text
|
# Einstein–Weyl geometry, the dKP equation and twistor theory
## 1 Three-dimensional Einstein–Weyl spaces
The aim of this paper is to study the Einstein–Weyl (EW) equations in relation to integrable systems, and in particular the dispersionless Kadomtsev–Petviashvili equation.
We begin by collecting various definitions and formulae concerning three-dimensional Einstein–Weyl spaces (see for a fuller account). In section 2 we construct and characterise a class of new EW structures in 2+1 dimensions out of solutions to the dKP equation. We then show that the dKP solutions give rise to hyper-Kähler metrics in four dimensions. We abuse terminology and call hyper-Kähler (hyper-complex, hyper-Hermitian) metrics which in signature $`(++)`$ should be referred to as pseudo-hyper-Kähler (pseudo-hyper-complex, pseudo-hyper-Hermitian). A null vector field (with conformal weight) will play a central role in our discussion so most of our constructions only make sense for Einsetin-Weyl spaces with Lorentzian signature, or complex holomorphic EW spaces (i.e. the complexification of real analytic EW spaces) and for the most part we work with the latter and restrict to a real slice when reality conditions play a role.
In section 3 we construct some new examples of EW structures. We obtain all solutions of the dKP equation with the property that the associated EW space admits a family of divergence-free, shear-free geodesic congruences. These solutions give rise to new EW metrics depending on one arbitrary function of one variable.
In section 4 a Lax representation of the general EW equations is given, together with a reformulation of the EW equations in terms of a closed and simple two-form on the bundle of spinors. A full twistor characterisation of dKP Einstein–Weyl structures and the corresponding hyper-Kähler metrics will be given in section 5. In section 6 we summarise our present knowledge of conformal reductions of four-dimensional hyper-Kähler metrics in split signature. In the Appendix we show how to obtain the dKP equation as a reduction of Plebański’s second heavenly equation . Parts of this work appeared in the DPhil thesis of one of the authors (MD) .
Let $`𝒲`$ be a $`3`$-dimensional complex manifold (one can also define Weyl spaces in arbitrary dimension) with a torsion-free connection $`D`$ and a conformal metric $`[h]`$. We shall call $`𝒲`$ a Weyl space if the null geodesics of $`[h]`$ are also geodesics for $`D`$. This condition is equivalent to
$$D_ih_{jk}=\nu _ih_{jk}$$
(1.1)
for some one form $`\nu `$. Here $`h_{jk}`$ is a representative metric in the conformal class. The indices $`i,j,k,\mathrm{}`$ go from $`1`$ to $`3`$. If we change this representative by $`h\varphi ^2h`$, then $`\nu \nu +2\mathrm{d}\mathrm{ln}\varphi `$. The one-form $`\nu `$ ‘measures’ the difference between $`D`$ and the Levi-Civita connection $``$ of $`h`$:
$$D_iV^j=_iV^j\frac{1}{2}\left(\delta _i^j\nu _k+\delta _k^j\nu _ih_{ik}\nu ^j\right)V^k.$$
(1.2)
The Ricci tensor $`W_{ij}`$ and scalar $`W`$ of $`D`$ are related to the Ricci tensor $`R_{ij}`$ and scalar $`R`$ of $``$ by
$`W_{ij}`$ $`=`$ $`R_{ij}+_i\nu _j{\displaystyle \frac{1}{2}}_j\nu _i+{\displaystyle \frac{1}{4}}\nu _i\nu _j+h_{ij}\left({\displaystyle \frac{1}{4}}\nu _k\nu ^k+{\displaystyle \frac{1}{2}}_k\nu ^k\right),`$
$`W:`$ $`=`$ $`h^{ij}W_{ij}=R+2^k\nu _k{\displaystyle \frac{1}{2}}\nu ^k\nu _k.`$
A tensor object $`T`$ which transforms as $`T\varphi ^mT`$ when $`h_{ij}\varphi ^2h_{ij}`$ is said to be conformally invariant of weight $`m`$. The Ricci scalar $`W`$, and the Ricci tensor $`W_{ij}`$ have weights $`2`$ and $`0`$ respectively.
Let $`\beta `$ be a $`p`$-form of weight $`m`$. The covariant exterior derivative
$$\stackrel{~}{D}\beta :=\mathrm{d}\beta \frac{m}{2}\nu \beta $$
is a well-defined $`p+1`$-form of weight $`m`$. The formula for a covariant weighted derivative of a vector of weight $`m`$ is
$$\stackrel{~}{D}_iV^j=_iV^j\frac{1}{2}\delta _i^j\nu _kV^k\frac{m+1}{2}\nu _iV^j+\frac{1}{2}\nu ^jV_i.$$
(1.3)
We say that a vector $`K`$ is a symmetry of a Weyl structure if it preserves the conformal structure $`[h]`$, the Weyl connection, and the compatibility (1.1) between those two. These conditions imply
$$_Kh=\psi h,_K\nu =\mathrm{d}\psi ,$$
(1.4)
where $`(h,\nu )`$ is a Weyl structure, and $`_K`$ is the Lie derivative along $`K`$.
The conformally invariant Einstein–Weyl (EW) condition on $`(𝒲,h,\nu )`$ is
$$W_{(ij)}=\frac{1}{3}Wh_{ij}.$$
If the above equation is satisfied and $`\nu `$ is a gradient, then $`h`$ is conformal to a metric with constant curvature.
In terms of the Riemannian data the Einstein–Weyl equations are
$$\chi _{ij}:=R_{ij}+\frac{1}{2}_{(i}\nu _{j)}+\frac{1}{4}\nu _i\nu _j\frac{1}{3}\left(R+\frac{1}{2}^k\nu _k+\frac{1}{4}\nu ^k\nu _k\right)h_{ij}=0.$$
(1.5)
Here $`\chi _{ij}`$ is a conformally invariant tensor (the trace-free part of the Ricci tensor of the Weyl connection). Weyl spaces which satisfy (1.5) will be called Einstein–Weyl (or EW) spaces.
In three dimensions the general solution of (1.1)-(1.5) depends on four arbitrary functions of two variables . The equations of the Weyl geodesics are
$$\frac{\mathrm{d}}{\mathrm{d}s}\frac{}{\dot{x}^i}\frac{}{x^i}=F_i(x^j,\dot{x}^j)$$
where $`=(1/2)h_{ij}\dot{x}^i\dot{x}^j`$ and $`F_i=\dot{x}_i(\dot{x}^j\nu _j)(1/2)\nu _i(\dot{x}^j\dot{x}_j)`$. Here $`\dot{}=\mathrm{d}/\mathrm{d}s`$ stands for the derivative with respect to a parameter $`s`$. It is evident that for null $`\dot{x}^i`$ the geodesics coincide with the null geodesics for $`[h]`$.
## 2 Einstein–Weyl structures from the dKP equation
In this section we shall construct Einstein–Weyl structures out of solutions to the dKP equation. In subsection 2.1 we shall find a class of hyper-Kähler metrics in four dimensions which reduce to dKP EW metrics.
The full Kadomtsev–Petviashvili equation for $`U:=U(X^i),X^i=(X,Y,T)`$
$$(U_TUU_X(1/12)U_{XXX})_X=U_{YY}$$
(2.6)
arises as a compatibility condition for the linear system $`L_0\mathrm{\Psi }=L_1\mathrm{\Psi }=0`$, where $`\mathrm{\Psi }=\mathrm{\Psi }(X,Y,T)`$ and
$$L_0=_Y(1/2)_X^2U,L_1=_T(1/3)_X^3U_XW,$$
for some $`W=W(X,Y,T)`$. To take a dispersionless limit of (2.6) introduce the slow coordinates $`x^i:=ϵX^i`$ (note that our notation for ‘slow’ and ‘fast’ coordinates is different from the usual one), and define $`u(x^i):=U(X^i),w(x^i):=W(X^i)`$. The linear system is replaced by
$$S_y=(1/2)S_x^2+u,S_t=(1/3)S_x^3+uS_x+w.$$
(2.7)
Here $`S:=S(x^i)`$ is the action defined by $`\mathrm{\Psi }(X^i)=\mathrm{exp}[ϵ^1S(x^i)]`$, and higher order terms in $`ϵ`$ have been neglected. Formulae (2.7) can be treated as a pair of Hamilton–Jacobi equations $`S_{t_A}+H_A(S_x,x,t_A)=0`$, with $`t_A=(y,t)`$ and $`H_A=(H_2,H_3)`$ where
$$H_2:=\frac{\stackrel{~}{\lambda }^2}{2}+u,H_3:=\frac{\stackrel{~}{\lambda }^3}{3}+\stackrel{~}{\lambda }u+w$$
for $`u=u(x,y,t)`$ and $`w=w(x,y,t)`$.
Now $`x^i`$ and $`S/x^i=(\stackrel{~}{\lambda },H_2,H_3)`$ form a set of canonically conjugate variables on an ‘extended phase-space’, with the symplectic form
$$\mathrm{\Pi }=\mathrm{d}x^i\mathrm{d}\frac{S}{x^i}=\mathrm{d}x\mathrm{d}\stackrel{~}{\lambda }+\mathrm{d}y\mathrm{d}H_2+\mathrm{d}t\mathrm{d}H_3.$$
(2.8)
This two-form is closed by definition. It is also simple iff $`u`$ and $`w`$ satisfy
$$w_x=u_y,u_tuu_x=w_y.$$
Eliminating $`w`$ yields the dKP equation
$$(u_tuu_x)_x=u_{yy}.$$
(2.9)
The simplicity of $`\mathrm{\Pi }`$ implies $`[_y+X_{H_2},_t+X_{H_3}]=0`$ where $`X_H:=H_x_{\stackrel{~}{\lambda }}H_{\stackrel{~}{\lambda }}_x`$ denotes the Hamiltonian vector field with respect to $`\mathrm{d}\stackrel{~}{\lambda }\mathrm{d}x`$, holding $`t`$ and $`y`$ constant. This gives a Lax pair for the dKP equation in terms of Hamiltonian vector fields. To obtain a Lax pair which is linear in the spectral parameter put
$$L_0^{}:=_t+X_{H_3}\stackrel{~}{\lambda }(_y+X_{H_2})=_tu_x\stackrel{~}{\lambda }_y+u_y_{\stackrel{~}{\lambda }},L_1^{}:=_y+X_{H_2}=_y\stackrel{~}{\lambda }_x+u_x_{\stackrel{~}{\lambda }}.$$
(2.10)
The dKP equation is equivalent to
$$[L_0^{},L_1^{}]=u_xL_1^{}.$$
Define a triad of vectors
$$_{1^{}1^{}}:=_x,_{0^{}1^{}}:=_y,_{0^{}0^{}}:=_tu_x$$
so $`L_A^{}=\pi ^B^{}_{A^{}B^{}}+f_A^{}_{\stackrel{~}{\lambda }}`$, where $`\pi ^A^{}=(1,\stackrel{~}{\lambda })`$ and $`f_A^{}=(u_y,u_x)`$.
The next proposition shows that we can find a one form $`\nu `$ such that $`_{A^{}B^{}}`$ is a null triad for an EW metric:
###### Proposition 2.1
Let $`u:=u(x,y,t)`$ be a solution of the dKP equation (2.9). Then the metric and the one-form
$$h=\mathrm{d}y^24\mathrm{d}x\mathrm{d}t4u\mathrm{d}t^2,\nu =4u_x\mathrm{d}t$$
(2.11)
give an EW structure.
Proof. Let $`x^1:=t,x^2:=y,x^3:=x`$. Five (out of six) EW equations $`\chi _{ij}=0`$ are satisfied identically by ansatz (2.11). The equation $`\chi _{11}=0`$ is equivalent to (2.9). We also find $`W=3u_{xx}`$. $`\mathrm{}`$
Example: Solutions which yield EW structures conformal to Einstein metrics (i.e. those for which $`\nu `$ is exact) are of the form
$$u(x,y,t)=xf_1(t)+\frac{1}{2}\left(\frac{\mathrm{d}f_1(t)}{\mathrm{d}t}f_1(t)^2\right)y^2+f_2(t)y+f_3(t),$$
(2.12)
where $`f_1(t),f_2(t),f_3(t)`$ are arbitrary functions of one variable.
One can verify that the vector $`_x`$ in the EW space (2.11) is a covariantly constant null vector in the Weyl connection with weight $`1/2`$. Now we shall prove the converse, and show that solutions (2.11) are characterised by the existence of a constant weighted vector.
###### Proposition 2.2
If a three dimensional EW space has a constant weighted vector field $`l`$ then coordinates can be chosen to put the EW metric and 1-form in the form (2.11).
We shall need following lemma:
###### Lemma 2.3
Let $`l`$ be a constant weighted vector on a three-dimensional EW space. Then either the EW space is flat or $`l`$ is null (so on a real slice the signature is $`(+)`$) and has weight $`1/2`$.
Proof. Assume that $`(h,\nu )`$ is a complex EW structure (we shall specify the reality conditions later in the proof). Commuting the Weyl derivatives yields
$$[D_i,D_j]l^k=\frac{m}{2}(D_i\nu _jD_j\nu _i)l^k=W_{}^{k}{}_{mij}{}^{}l^m,$$
where $`W_{}^{k}{}_{mij}{}^{}`$ is the curvature of the Weyl connection, and $`m`$ is the weight of $`l^k`$. It can be decomposed as
$$W_{}^{k}{}_{mij}{}^{}=\epsilon _{ij}^{}{}_{}{}^{p}\epsilon _{m}^{}{}_{}{}^{kq}S_{pq}\delta _m^kF_{ij},$$
(2.13)
where $`F_{ij}=_{[i}\nu _{j]}`$, and $`S_{ij}`$ is a conformally invariant tensor of weight $`0`$. If the EW equations are satisfied $`S_{ij}`$ is given by
$$S_{ij}=\frac{1}{2}F_{ij}+\frac{W}{6}h_{ij}.$$
(2.14)
Equations (2.13) and (2.14) imply
$$(m+1)F_{ij}l^k=\frac{1}{2}\epsilon _{ij}^{}{}_{}{}^{p}l^m\epsilon _{m}^{}{}_{}{}^{kq}F_{pq}+\frac{W}{6}(\delta _i^kl_j\delta _j^kl_i).$$
(2.15)
In three dimensions any non-zero two-form $`F_{ij}`$ has a non-trivial kernel, i.e. there exists a non-zero vector $`L^j`$ with $`F_{ij}L^j=0`$, which implies
$$F_{ij}=F\epsilon _{ijk}L^k$$
(2.16)
for some non-zero $`F`$. We have to consider three cases:
* Suppose first that $`L^k`$ is a null vector and contract (2.15) with $`L^j`$ to find
$$0=\frac{1}{2}\epsilon _{ij}^{}{}_{}{}^{p}\epsilon _{m}^{}{}_{}{}^{kq}F\epsilon _{pqr}L^rl^mL^j+\frac{W}{6}(\delta _i^kl_jL^jL^kl_i).$$
(2.17)
Contracting this with $`L_k`$ yields $`Wl_jL^j=0`$. If $`W=0`$ then (2.17) implies that $`l^i`$ and $`L^i`$ are proportional, so $`l^i`$ is null. If $`W0`$, so that $`l_jL^j=0`$ then (2.17) reduces to
$$0=\frac{1}{2}FL^ql^mL_i\epsilon _{m}^{k}{}_{q}{}^{}\frac{W}{6}l_iL^k$$
from which again $`l^i`$ is null. Therefore $`l^i`$ and $`L^i`$ are both null and orthogonal and so (as we work in three dimensions) they have to be proportional. Now (2.17) forces $`W=0`$. Equation (2.15) is now satisfied only if $`m=1/2`$.
* If $`L^i`$ is not null, we can choose an orthogonal frame with $`F_{23}=F0`$ , and $`F_{12}=F_{13}=0`$, and use (2.15) to examine components of $`F_{ij}l^k`$ in this frame. This yields
$`Wl_1`$ $`=`$ $`0,Fl^1=0,{\displaystyle \frac{1}{2}}Fl^3+{\displaystyle \frac{1}{6}}Wl_2=0,{\displaystyle \frac{1}{2}}FV^2{\displaystyle \frac{1}{6}}Wl_3=0,`$ (2.18)
$`(m+1)Fl^1`$ $`=`$ $`0,(m+1)Fl^2={\displaystyle \frac{1}{6}}Wl_3={\displaystyle \frac{1}{2}}Fl^2,(m+1)Fl^3={\displaystyle \frac{1}{6}}Wl_2={\displaystyle \frac{1}{2}}Fl^3.`$
Therefore $`l^1=0`$, and (2.18) imply $`(m+1/2)Fl^2=0,(m+1/2)Fl^3=0.`$ But $`l^i0`$, so $`m=1/2`$. Equations ( 2.18) also imply that $`l^i`$ is null.
* If $`F=0=\mathrm{d}\nu =0`$ (Einstein case) choose a conformal gauge in which $`\nu =0`$. Now $`D_il^j=_il^j=0`$ implies $`R=0`$. Therefore the metric $`h`$ is flat and $`l^j`$ is a constant vector.
$`\mathrm{}`$
Proof of Proposition (2.2). Lemma 2.3 and the formula (1.3) with $`m=1/2`$ imply
$$\stackrel{~}{D}_il^j=D_il^j+\frac{1}{4}\nu _il^j=0.$$
(2.19)
Therefore $`D_il_j=(3/4)\nu _il_j`$, so $`\mathrm{d}𝐥=(3/4)\nu 𝐥`$ (here $`𝐥`$ is the one form dual to $`l`$).
This implies that we can rescale the metric and hence $`𝐥`$ so that $`𝐥=2\mathrm{d}t`$ for some function $`t`$. We must then have $`\nu =b\mathrm{d}t`$ for some function $`b`$. Choose coordinates $`x`$ and $`y`$ so that $`l(y)=0`$ and $`l(x)=1`$ and $`(x,y,t)`$ is a coordinate system. At this point we have
$$h=F\mathrm{d}y^2+G\mathrm{d}y\mathrm{d}t4\mathrm{d}x\mathrm{d}t4u\mathrm{d}t^2,\nu =b\mathrm{d}t,$$
where $`F,G,b`$ and $`u`$ are functions of $`x,y,t`$. The formulae (1.2) and (2.19) imply $`_il_j=(1/4)\nu _il_j(1/2)\nu _jl_i`$. Symmetrising this expression yields $`_{(i}l_{j)}=(1/4)\nu _il_j`$, which implies that $`F_x=G_x=0`$, and $`4u_x=b`$. We are still free to change $`xx+P(y,t)`$, which gives
$$h=F\mathrm{d}y^2+G\mathrm{d}y\mathrm{d}t4(\mathrm{d}x+P_y\mathrm{d}y+P_t\mathrm{d}t)\mathrm{d}t4u\mathrm{d}t^2\nu =4u_x\mathrm{d}t.$$
We can find $`K`$ such that $`\mathrm{d}\widehat{y}:=\sqrt{F}\mathrm{d}y+K\mathrm{d}t`$ is exact, and eliminate the $`\mathrm{d}\widehat{y}\mathrm{d}t`$ term in the metric by choosing $`4P_y=2K+G/\sqrt{F}`$. This (after redefining $`u`$ by adding to it a function of $`(\widehat{y},t)`$ so that $`\nu `$ remains unchanged) yields the EW structure (2.11). $`\mathrm{}`$
Remark: The above coordinate conditions fix the coordinates and $`u`$ only up to the freedom $`(x,y,t)(\stackrel{~}{x},\stackrel{~}{y},\stackrel{~}{t})`$, $`u(x,y,t)\stackrel{~}{u}(\stackrel{~}{x},\stackrel{~}{y},\stackrel{~}{t})`$ where
$`(x,y,t)`$ $`=`$ $`(\stackrel{~}{x}f^{}\stackrel{~}{y}g,\stackrel{~}{y}2f,\stackrel{~}{t}),`$
$`\stackrel{~}{u}(\stackrel{~}{x},\stackrel{~}{y},\stackrel{~}{t})`$ $`=`$ $`u(\stackrel{~}{x}f^{}\stackrel{~}{y}g,\stackrel{~}{y}2f,\stackrel{~}{t})\stackrel{~}{y}f^{\prime \prime }f^2g^{}.`$ (2.20)
where $`f`$ and $`g`$ are arbitrary functions of $`t`$ and $``$ denotes the derivative with respect to $`t`$.
Furthermore the conformal scale is only fixed up to arbitrary functions of $`t`$, $`h\stackrel{~}{h}=\mathrm{\Omega }^2h`$. Such a rescaling leads to a redefinition of $`t`$, $`t\stackrel{~}{t}`$ given by $`t=c(\stackrel{~}{t})`$ where $`\mathrm{\Omega }=c^{2/3}`$ where now and in the following $``$ denotes the derivative wrt $`\stackrel{~}{t}`$. This leads to the redefinitions $`(x,y,t)(\stackrel{~}{x},\stackrel{~}{y},\stackrel{~}{t})`$, $`u(x,y,t)\stackrel{~}{u}(\stackrel{~}{x},\stackrel{~}{y},\stackrel{~}{t})`$ given by
$`(x,y,t)`$ $`=`$ $`(c^{1/3}\stackrel{~}{x}+{\displaystyle \frac{c^{\prime \prime }}{6c^{2/3}}}\stackrel{~}{y}^2,c^{2/3}\stackrel{~}{y},c(\stackrel{~}{t})),`$
$`\stackrel{~}{u}(\stackrel{~}{x},\stackrel{~}{y},\stackrel{~}{t})`$ $`=`$ $`c^{2/3}u(c^{1/3}\stackrel{~}{x}+{\displaystyle \frac{c^{\prime \prime }}{6c^{2/3}}}\stackrel{~}{y}^2,c^{2/3}\stackrel{~}{y},c)+{\displaystyle \frac{c^{\prime \prime }\stackrel{~}{x}}{3c^{}}}+{\displaystyle \frac{\stackrel{~}{y}^2}{18}}\left({\displaystyle \frac{3c^{\prime \prime \prime }}{c^{}}}4\left({\displaystyle \frac{c^{\prime \prime }}{c^{}}}\right)^2\right).`$ (2.21)
From the point of view of the Einstein-Weyl spaces, the transformations above are equivalences, however from the point of view of the dKP equations, they map one solution of the dKP equations to another allowing one to deduce solutions depending on 3 functions of one variable from a given solution:
###### Corollary 2.4
Let $`u(x,y,t)`$ be a solution to the dKP equation, then $`\stackrel{~}{u}(\stackrel{~}{x},\stackrel{~}{y},\stackrel{~}{t})`$ is another solution where $`\stackrel{~}{u}`$ is given in terms of either of the formulae (2) or (2).
### 2.1 Hyper-Kähler structures from the dKP equation
In this subsection we shall show that EW structures given by (2.11) give rise to four-dimensional hyper-Kähler structures with symmetry. We shall start by summarising some results about anti-self-dual (ASD) four manifolds with Killing vectors, and the Lax representation of hyper-Hermitian four manifolds.
All three-dimensional EW spaces can be obtained as spaces of trajectories of conformal Killing vectors in four-dimensional manifolds with ASD conformal curvature:
###### Proposition 2.5 ()
Let $`(,\widehat{g})`$ be an ASD four-manifold with a conformal Killing vector $`K`$. The EW structure on the space $`𝒲`$ of trajectories of $`K`$ (which is assumed to be non-pathological) is defined by
$$h:=|K|^2\widehat{g}|K|^4𝐊𝐊,\nu :=s^{}\left(2|K|^2_{\widehat{g}}(𝐊\mathrm{d}𝐊)\right),$$
(2.22)
where $`|K|^2:=\widehat{g}_{ab}K^aK^b`$, $`𝐊`$ is the one form dual to $`K`$ and $`_{\widehat{g}}`$ is taken with respect to $`\widehat{g}`$ and $`s:𝒲`$ is an arbitrary section of the fibration $`𝒲`$. All EW structures arise in this way.
Conversely, let $`(h,\nu )`$ be a three–dimensional EW structure on $`𝒲`$, and let $`(V,\alpha )`$ be a pair consisting of a function of weight $`1`$ and a one-form on $`𝒲`$ which satisfy the generalised monopole equation
$$_h(\mathrm{d}V+(1/2)\nu V)=\mathrm{d}\alpha ,$$
(2.23)
where $`_h`$ is taken with respect to $`h`$. Then
$$g=Vh\pm V^1(\mathrm{d}z+\alpha )^2$$
(2.24)
is an ASD metric with an isometry $`K=_z`$. The minus sign in (2.24) is choosen if $`h`$ has signature $`(++)`$.
In what follows we shall consider ASD structures which are also (complexified) hyper-Hermitian.
A smooth manifold $``$ equipped with three almost complex structures $`(I,J,K)`$ satisfying the algebra of quaternions is called hyper-complex iff the almost complex structure $`𝒥_\lambda =aI+bJ+cK`$ is integrable for any $`(a,b,c)S^2`$. We use $`\lambda =(a+ib)/(c1)`$, a stereographic coordinate on $`S^2`$ which we view as a complex projective line $`^1`$. Let $`g`$ be a Riemannian metric on $``$. If $`(,𝒥_\lambda )`$ is hyper-complex and $`g(𝒥_\lambda X,𝒥_\lambda Y)=g(X,Y)`$ for all vectors $`X,Y`$ on $``$ then the triple $`(,𝒥_\lambda ,g)`$ is called a hyper-Hermitian structure.
We will in practice be interested in complexified or indefinite hyper-Hermitian metrics with signature $`(++)`$ for which the tensors $`(I,J,K)`$ must necessarily be complex. In signature $`(++)`$ we can arrange that one be real and the other two be pure imaginary, in the latter case they determine a pair of transverse null foliations.
We shall restrict ourselves to oriented four manifolds. In four dimensions a hyper-complex structure defines a conformal structure, which in explicit terms is represented by a conformal orthonormal frame of vector fields $`(X,IX,JX,KX)`$, for any $`XT`$. It is well known that this conformal structure is ASD with the orientation determined by the complex structures.
If there exists a choice of a conformal factor such that a two form $`\mathrm{\Sigma }_\lambda `$ defined by $`\mathrm{\Sigma }_\lambda (X,Y):=g(X,𝒥_\lambda Y)`$ is closed (with fixed $`\lambda `$) for all $`\lambda ^1`$ and all vectors $`(X,Y)`$ then $`(,𝒥_\lambda ,g)`$ is called hyper-Kähler.
We shall use the following characterisation of the hyper-Hermiticity condition:
###### Proposition 2.6 ()
Let $`_{AA^{}}`$ be four independent real vector fields on a four-dimensional real manifold $``$, and let
$$L_0=_{00^{}}\lambda _{01^{}},L_1=_{10^{}}\lambda _{11^{}},\text{where}\lambda ^1.$$
If
$$[L_0,L_1]=0$$
(2.25)
for every $`\lambda `$, then $`_{AA^{}}`$ is a null tetrad for a $`(++)`$ hyper-Hermitian metric on $``$. Every $`(++)`$ hyper-Hermitian metric arises in this way. Moreover, if the vectors $`_{AA^{}}`$ preserve a volume form $`\text{vol}_g`$ on $``$, then $`f^1_{AA^{}}`$ is a null tetrad for a $`(++)`$ hyper-Kähler metric on $``$. Here $`f^2=\text{vol}_g(_{00^{}},_{10^{}},_{01^{}},_{11^{}})`$.
Now we shall use (2.11) and Proposition 2.5 to construct ASD metrics out of solutions to the dKP equation, and Proposition 2.6 to show that they are hyper-Kähler.
Assume that $`h`$ and $`\nu `$ are as in (2.11). Taking the exterior derivative of the generalised monopole equation (2.23) yields
$`0`$ $`=`$ $`_i^iV+(1/2)(^i\nu _i)V+(1/2)\nu ^i_iV`$ (2.26)
$`=`$ $`V_{yy}V_{xt}+uV_{xx}+2u_xV_x+u_{xx}V`$
which is just a linearisation of the dKP equation (2.9) (note that for $`u=0`$ (2.26) is just the wave equation relative to the flat metric $`\mathrm{d}y^24\mathrm{d}x\mathrm{d}t`$). One solution is $`V=u_x/2`$. One could find a corresponding $`\alpha `$ and write down a metric using formula (2.24) (see the remarks after Proposition 2.7), but we shall present a different method based on the Lax operators.
Take the Lax operators (2.10) and introduce a new spectral parameter $`\lambda :=\stackrel{~}{\lambda }z`$ for some $`z`$. The function $`u(x,y,t)`$ does not depend on $`z`$ so we can replace $`_{\stackrel{~}{\lambda }}`$ by $`_z`$. This yields (with dropped primes and added tildes)
$`\stackrel{~}{L}_0`$ $`=`$ $`_tu_xz_y+u_y_z\lambda _y,`$
$`\stackrel{~}{L}_1`$ $`=`$ $`_yz_x+u_x_z\lambda _x.`$
To obtain a pair of exactly commuting operators take
$`L_1`$ $`:=`$ $`\stackrel{~}{L}_1=_yz_x+u_x_z\lambda _x,`$
$`L_0`$ $`:=`$ $`\stackrel{~}{L}_0+z\stackrel{~}{L}_1=_t(u+z^2)_x+(u_y+u_xz)_z\lambda (_y+z_x).`$
If $`u(x,y,t)`$ is a solution to (2.9) then these operators satisfy $`[L_0,L_1]=0`$ and so, by Proposition 2.6, the vectors
$$_{10^{}}=_yz_x+u_x_z,_{11^{}}=_x,_{00^{}}=_t(u+z^2)_x+(u_y+u_xz)_z,_{01^{}}=(_y+z_x),$$
form a hyper-Hermitian frame. The vectors $`_{AA^{}}`$ preserve the volume form $`\text{vol}_g=\mathrm{d}t\mathrm{d}y\mathrm{d}x\mathrm{d}z`$, and $`f^2=u_x/2`$. Therefore we have the following
###### Proposition 2.7
Let $`u=u(x,y,t)`$. The metric
$$g=\frac{u_x}{2}(\mathrm{d}y^24\mathrm{d}x\mathrm{d}t4u\mathrm{d}t^2)\frac{2}{u_x}(\mathrm{d}z\frac{u_x\mathrm{d}y}{2}u_y\mathrm{d}t)^2$$
(2.27)
is hyper-Kähler.
Remarks:
* The above metric has a Killing vector $`_z`$ with the dual
$$K=\frac{2}{u_x}(\mathrm{d}z\frac{u_x\mathrm{d}y}{2}u_y\mathrm{d}t),$$
and the formulae (2.22) gives rise to the Einstein–Weyl structure (2.11). The self-dual part of $`\mathrm{d}K`$ is a simple two-form. In section 5 we shall show that all hyper-Kähler metrics with such symmetries are locally given by (2.27).
* Note that $`u_x0`$ for (2.27) to be well defined. To obtain a flat metric take $`u=x/t`$ which is a special case of (2.12). The metric (2.27) becomes
$$g=2\mathrm{d}x\frac{\mathrm{d}t}{t}2x\frac{\mathrm{d}t^2}{t^2}+2t\mathrm{d}z^2+2\mathrm{d}z\mathrm{d}y.$$
Putting $`x=Xt+z^2t/2,y=Yzt`$ yields the flat metric
$$g=2\mathrm{d}X\mathrm{d}t+2\mathrm{d}z\mathrm{d}Y.$$
* The metric (2.27) could be found directly from the monopole equation (2.23) as follows: Rewrite the metric (2.11) in an orthonormal triad $`h=e_1^2+e_2^2e_3^2`$, where
$$e_1=\mathrm{d}y,e_2=\mathrm{d}x+(u1)\mathrm{d}t,e_3=\mathrm{d}x+(u+1)\mathrm{d}t.$$
The duality relations $`_he_1=e_3e_2,_he_2=e_1e_3,_he_3=e_1e_2`$ yield
$$_h\mathrm{d}t=\mathrm{d}t\mathrm{d}y,_h\mathrm{d}y=2\mathrm{d}t\mathrm{d}x,_h\mathrm{d}x=\mathrm{d}y\mathrm{d}x+2u\mathrm{d}y\mathrm{d}t.$$
(2.28)
Take $`V=u_x/2`$, and use the above relations to write the monopole equation (2.23) as
$$\frac{u_{xx}}{2}\mathrm{d}y\mathrm{d}x+u_{xy}\mathrm{d}t\mathrm{d}x+(u_x^2+uu_{xx}\frac{u_{xt}}{2})\mathrm{d}y\mathrm{d}t=\mathrm{d}\alpha .$$
Choosing the gauge in which $`\alpha =\alpha _1\mathrm{d}y+\alpha _2\mathrm{d}t`$ (this is always possible by redefining a coordinate $`z`$ along the orbits of a Killing vector) gives
$$(\alpha _1)_x=\frac{u_{xx}}{2},(\alpha _2)_x=u_{xy},(\alpha _2)_y(\alpha _1)_t=\frac{u_{xt}}{2}u_{yy}.$$
(2.29)
All solutions to this system of equations are gauge equivalent to
$$\alpha =\frac{u_x}{2}\mathrm{d}yu_y\mathrm{d}t.$$
Substituting $`V,\alpha `$ and $`h`$ to (2.24) yields (2.27).
* The Lax pair (2.10) can be obtained from the hyper-Kähler Lax pair by a symmetry reduction: The distribution $`(K,\stackrel{~}{L}_0,\stackrel{~}{L}_1)`$ is not integrable, as $`[K,\stackrel{~}{L}_0]=_y`$ and $`[K,\stackrel{~}{L}_1]=_x`$. To obtain an integrable distribution, one needs to lift $`K`$ to the correspondence space by $`\stackrel{~}{K}=K_\lambda `$. Then $`(\stackrel{~}{K},\stackrel{~}{L}_0,\stackrel{~}{L}_1)`$ is an integrable distribution, but $`\stackrel{~}{K}(\lambda )0`$, which forces us to introduce an invariant spectral parameter $`\stackrel{~}{\lambda }=\lambda +z`$. This implies that in the Lax pair we replace all $`_z`$ by $`\stackrel{~}{K}+_{\stackrel{~}{\lambda }}`$. Now we restrict ourselves to invariant solutions to $`\stackrel{~}{L}_0\mathrm{\Psi }=\stackrel{~}{L}_1\mathrm{\Psi }=0`$, and so we ignore $`\stackrel{~}{K}`$ in the Lax pair. The reduced Lax pair is given by (2.10).
In the covariantly constant primed spin frame the null tetrad is
$`e^{00^{}}`$ $`=`$ $`u_x\mathrm{d}t,e^{10^{}}={\displaystyle \frac{\mathrm{d}zu_y\mathrm{d}t}{u_x}},`$
$`e^{01^{}}`$ $`=`$ $`\mathrm{d}zu_x\mathrm{d}y(u_y+zu_x)\mathrm{d}t,e^{11^{}}=\mathrm{d}x+u\mathrm{d}t+z{\displaystyle \frac{\mathrm{d}zu_y\mathrm{d}t}{u_x}},`$
and the metric (2.27) is $`2(e^{00^{}}e^{11^{}}e^{01^{}}e^{10^{}})`$. The basis of SD two form is in this frame given by
$`\mathrm{\Sigma }^{0^{}0^{}}`$ $`=`$ $`\mathrm{d}z\mathrm{d}t,\mathrm{\Sigma }^{0^{}1^{}}=\mathrm{d}z\mathrm{d}y+\mathrm{d}(u+z^2)\mathrm{d}t,`$
$`\mathrm{\Sigma }^{1^{}1^{}}`$ $`=`$ $`u_x\mathrm{d}x\mathrm{d}yuu_x\mathrm{d}y\mathrm{d}t+u_y\mathrm{d}x\mathrm{d}t+\mathrm{d}(uz)\mathrm{d}t+\mathrm{d}z(\mathrm{d}x+z\mathrm{d}y+z^2\mathrm{d}t).`$
They satisfy
$$2\mathrm{\Sigma }^{0^{}0^{}}\mathrm{\Sigma }^{1^{}1^{}}=\mathrm{\Sigma }^{0^{}1^{}}\mathrm{\Sigma }^{0^{}1^{}},\mathrm{d}\mathrm{\Sigma }^{0^{}0^{}}=\mathrm{d}\mathrm{\Sigma }^{0^{}1^{}}=\mathrm{d}\mathrm{\Sigma }^{1^{}1^{}}=0,$$
which again implies that the metric (2.27) is hyper–Kähler. Note that the Killing vector $`K=_z`$ does not preserve the Kähler form $`\mathrm{\Sigma }^{0^{}1^{}}`$.
## 3 Examples
### 3.1 dKP EW spaces with $`S^1`$ symmetry
In this subsection we shall construct EW structures depending on one arbitrary function of one variable.
To find some explicit examples of (2.11) assume that $`u`$ is independent of $`y`$. Therefore it satisfies the simple equation $`uu_x=u_t`$, all solutions of which are given in an implicit form
$$u(x,t)=f(x+tu(x,t))$$
(more general hodograph transformations for dKP arising from its connection with equations of hydrodynamic type were studied in , and ).
Here $`f`$ is an arbitrary function of one variable $`s:=x+tu(x,t)`$. The idea is to write the Einstein–Weyl structure (2.11) making use of this ‘hodograph transformation’. We have
$$h=\mathrm{d}y^24\mathrm{d}t(\mathrm{d}x+u\mathrm{d}t)=\mathrm{d}y^24\mathrm{d}t(\mathrm{d}st\mathrm{d}u)=\mathrm{d}y^24\mathrm{d}t\mathrm{d}s+4t\mathrm{d}t\mathrm{d}f(s)$$
where we performed a coordinate transformation $`(x,y,t)(s,y,t)`$. Defining $`F(s):=\mathrm{d}f/\mathrm{d}s`$ and replacing $`u_x`$ by $`F/(1tF)`$ yields the EW structure
$$h=\mathrm{d}y^2+4(tF(s)1)\mathrm{d}t\mathrm{d}s,\nu =4\frac{F(s)}{tF(s)1}\mathrm{d}t,$$
(3.30)
which depends on one arbitrary function $`F(s)`$ (which we shall take to be strictly negative) of one variable. This structure has signature $`(++)`$. If $`t>0`$ then it is well-defined on $`S^1\times ^+\times `$.
We shall now show that formulae (3.30) give a class of $`EW`$ structures on principal $`S^1`$ bundles over Weyl manifolds.
###### Proposition 3.1
Let $`(𝒩,[H],\nu _H)`$ be a two-dimensional manifold with a Weyl structure of signature $`(+)`$ and let $`\pi :𝒲𝒩`$ be an $`S^1`$ bundle over $`N`$. If
$$h:=\mathrm{d}y^2+\pi ^{}H,\nu :=\pi ^{}\nu _H$$
(where $`y`$ is a coordinate on a fibre) is an EW structure on $`𝒲`$ then it can be put in the form (3.30).
Proof. We can use isothermal coordinates $`(\stackrel{~}{s},t)`$ on $`𝒩`$ and choose a representative of a conformal class $`[H]`$ such that $`h`$ and $`\nu `$ are
$$h=\mathrm{d}y^2+2G(\stackrel{~}{s},t)\mathrm{d}\stackrel{~}{s}\mathrm{d}t,\nu =K(\stackrel{~}{s},t)\mathrm{d}t.$$
(3.31)
Each EW structure of this form is equivalent to (3.30). This can be seen as follows: Equations $`\chi _{13}=0,\chi _{22}=0`$ imply that $`K=4G_t/G+f(t)`$. The function $`f(t)`$ can be absorbed in the definition of $`G`$. Then the vanishing of $`\chi _{33}`$ (all remaining EW equations are satisfied trivially) yields $`G(\stackrel{~}{s},t)=2F_1(\stackrel{~}{s})+2tF_2(\stackrel{~}{s})`$ for arbitrary $`F_1`$ and $`F_2`$. Now we define a new coordinate $`s`$ by $`\mathrm{d}s:=F_1(\stackrel{~}{s})\mathrm{d}\stackrel{~}{s}`$. Equivalence between (3.31) and (3.30) is finally obtained by putting $`F(s):=F_2(s)/F_1(s)`$. The metric (3.30) is not Einstein as $`G_{22}0,G_{13}0`$ and $`R=2F_s/(tF1)^3`$ is not constant (unless $`F`$ is constant). To visualise the two-dimensional surface $`𝒩`$ on which $`H`$ is defined one can restrict a flat $`(++)`$ metric on $`^4`$, $`g=\mathrm{d}f\mathrm{d}w\mathrm{d}s\mathrm{d}t`$ to the intersection of the paraboloid $`w=t^2/2`$ with the hyper-surface $`f=f(s)`$. $`\mathrm{}`$
The hyper-Kähler metric corresponding to (3.30) has an additional null Killing vector $`_y`$ and is (with definitions $`\mathrm{d}w:=F\mathrm{d}s,\widehat{F}(w):=F^1`$) given by
$$g=\mathrm{d}w\mathrm{d}t+\mathrm{d}z\mathrm{d}y+(t\widehat{F}(w))\mathrm{d}z^2$$
where $`\widehat{F}(w)`$ is arbitrary.
Other examples (without a Killing vector) can be obtained from
$$u=t\frac{\mathrm{d}A(t)}{\mathrm{d}t}\frac{x}{t}+\frac{y}{t}\sqrt{\frac{x}{t}+A(t)},$$
where $`A(t)`$ is arbitrary.
### 3.2 dKP metrics which are hyper-CR
Let us recall that that an EW metric is called hyper-CR (or special) if it admits a two-parameter family of shear-free, divergence-free geodesic congruences . All hyper-CR EW spaces arise as reductions of hyper-Kähler metrics by triholomorphic homotheties . In this section we shall find all EW metrics in 2+1 dimensions which are both dKP and hyper–CR. This will lead to a class of solutions to the dKP equation depending on one arbitrary function of one variable.
###### Proposition 3.2
All EW metrics which admit a constant weighted vector and a two parameter family of shear-free geodesic congruences with a vanishing divergence are either spaces of constant curvature or are locally of the form
$$h=\mathrm{d}y^24\mathrm{d}x\mathrm{d}t4\left(\frac{P(t)}{y}\frac{x^2}{y^2}\right)\mathrm{d}t^2,\nu =\frac{8x}{y^2}\mathrm{d}t,$$
(3.32)
where $`P`$ is an arbitrary function of $`t`$.
Proof. The hyper-CR condition for a metric is characterised by the existence of a scalar $`\rho `$ of weight $`1`$ which (together with the Einstein–Weyl one form $`\nu `$) satisfies the monopole equation
$$_h(\mathrm{d}\rho +\frac{1}{2}\nu \rho )=\mathrm{d}\nu ,$$
(3.33)
and the algebraic constraint
$$\rho ^2=\frac{8}{3}W.$$
(3.34)
We shall impose these conditions on the dKP metric (2.11). The monopole equation yields
$$(4u_{xx}2\rho _y)\mathrm{d}x\mathrm{d}t+\rho _x\mathrm{d}y\mathrm{d}x+(2\rho _xu\rho _t+2\rho u_x+4u_{xy})\mathrm{d}y\mathrm{d}t=0$$
which (together with (3.34)) gives four scalar equations:
$$\rho _y=2u_{xx},\rho _x=0,2\rho u_x\rho _t+4u_{xy}=0,\rho ^2=8u_{xx}.$$
(3.35)
If $`u_{xx}=0`$ then the last relation in (3.35) gives $`\rho =0`$. The monopole equation then implies that $`\nu `$ is closed, and the Einstein–Weyl metric is conformal to Einstein. Therefore we assume $`u_{xx}0`$. Differentiating the third equation in (3.35) with respect to $`x`$ (and using the first two equations) gives
$$\rho =2\frac{u_{xxy}}{u_{xx}}.$$
The integrability conditions to (the otherwise over-determined system) (3.35) are
$`u_{xxx}=0,u_{xxy}^2u_{xxyy}u_{xx}=u_{xx}^3,4u_{xxy}=\eta u_{xx}^3,`$ (3.36)
$`u_{xxy}u_{xxt}u_{xxyt}u_{xx}+2u_xu_{xx}u_{xxy}2u_{xy}u_{xx}^2=0.`$
The first condition implies $`u(x,y,t)=ax^2+bx+c`$. Here $`a,b,c`$ are functions of $`y`$ and $`t`$, which satisfy
$$a_{yy}+6a^2=0,$$
(3.37)
$$b_{yy}2a_t+6ab=0,$$
(3.38)
$$c_{yy}b_t+2ac+b^2=0,$$
(3.39)
$$a_y^2aa_{yy}2a^3=0,$$
(3.40)
$$a_y^2+4a^3=0,$$
(3.41)
$$aa_{yt}a_ya_t2aa_yb+2b_ya^2=0$$
(3.42)
Equations (3.37, 3.38, 3.39) follow from the dKP (2.9), and the other equations are the integrability conditions (3.36). Solve (3.41) to find $`a(y,t)=(yL(t))^2`$ (or $`a=0`$ which gives $`u_{xx}=0`$).
We can now perform the coordinate transformation (2) with $`f=L/2`$ and $`g=0`$ to set $`L(t)=0`$. One verifies that (3.37), and (3.41) are now also satisfied. Equation (3.38) gives $`b(y,t)=M(t)y^2+N(t)y^3,`$ but (3.42) implies $`N(t)=0`$. So far we have
$$h=\mathrm{d}y^24\mathrm{d}x\mathrm{d}t+4\left(c(y,t)\frac{xM(t)}{y^2}\frac{x^2}{y^2}\right)\mathrm{d}t^2,\nu =\frac{8x+4M(t)}{y^2}\mathrm{d}t.$$
The function $`M(t)`$ can be eliminated by the coordinate transformation (2) with $`g=M/2`$. Imposing (3.39) yields $`c(y,t)=P(t)/y+R(t)y^2`$ leaving
$$h=\mathrm{d}y^24\mathrm{d}x\mathrm{d}t+4\left(\frac{x^2}{y^2}+\frac{P(t)}{y}+R(t)y^2\right)\mathrm{d}t^2,\nu =\frac{8x}{y^2}\mathrm{d}t.$$
We eliminate $`R(t)`$ by performing the conformal rescaling and associated coordinate redefinitions of (2) with $`c(\stackrel{~}{t})`$ satisfying
$$R=\frac{c^{\prime \prime \prime }}{6c^3}+\frac{1}{4}\left(\frac{c^{\prime \prime }}{c^2}\right)^2.$$
This yields, dropping the tildes and with a redefinition of $`P`$,
$$u(x,y,t)=\frac{x^2}{y^2}+\frac{P(t)}{y}.$$
The Einstein–Weyl structure is therefore (3.32). The arbitrary function $`P(t)`$ can not be eliminated. This can be seen by finding the symmetries (1.4) of the EW structure (3.32). We summarise our findings in the table below:
| | Function $`P\left(t\right)`$ | Symmetries |
| --- | --- | --- |
| (i) | $`P\left(t\right)=0`$ | $`K_1,K_2,K_3,K_4`$ |
| (ii) | $`P\left(t\right)=const0`$ | $`K_1,K_2+3K_3,K_4`$ |
| (iii) | $`P\left(t\right)=\left(bt+c\right)^{\frac{3ab}{2b}}`$ | $`cK_1+aK_2+bK_3`$ |
| (iv) | general $`P\left(t\right)`$ | none |
where $`a,b,c`$ are constants, and
$$K_1=_t,K_2=(1/2)y_y+x_x,K_3=(1/2)y_y+t_t,K_4=ty_y+(y^2+2xt)_x+3t^2_t.$$
Note that in case $`(ii)`$ we can redefine coordinates to set $`P(t)=1`$. The vector fields $`K_1,K_2+3K_3,K_4`$ generate the Lie group of Bianchi type VIII, i.e. $`SU(1,1)`$, and the cases $`(i)`$ and $`(ii)`$ give homogeneous EW spaces. Case $`(iii)`$ can be reduced to $`P(t)=t^\alpha ,K=K_3+[(2\alpha +1)/3]K_2`$, where $`\alpha =const0`$. $`\mathrm{}`$
## 4 The twistor correspondences and Lax formulations
In this section we shall study the twistor theory of the EW spaces. We first discuss the twistor correspondence in the flat case. We then give a Lax formulation of the EW equations and derive from it the twistor correspondence. We study this correspondence in relation to reductions of the anti-self-duality equations on four-dimensional conformal structures. We then reformulate the Einstein–Weyl equations in terms of a certain two-form on the trivial $`^1`$ bundle over a Weyl space.
### 4.1 The flat correspondence
Let us begin by recalling Ward’s approach to twistors in (2+1)-dimensional flat space-times. Rearrange the space time coordinates $`(x,y,t)`$ as a symmetric two-spinor<sup>1</sup><sup>1</sup>1The use of primed (rather than unprimed) spinors in this section originates from the representation of Einstein–Weyl spaces as reductions of ASD (rather than SD) metrics in four dimensions. ASD structures (for which the covariantly constant self-dual spinors are conventionally denoted as having primed indices) are taken as basic because they arise from a natural choice of orientation and conformal structure on a Kähler manifold.
$$x^{A^{}B^{}}:=\left(\begin{array}{cc}t& y/2\\ y/2& x\end{array}\right),$$
such that the space-time metric and the volume form are :
$$h=2\mathrm{d}x_{A^{}B^{}}\mathrm{d}x^{A^{}B^{}},\text{vol}_h=\mathrm{d}x_{A^{}}^{}{}_{}{}^{B^{}}\mathrm{d}x_{C^{}}^{}{}_{}{}^{A^{}}\mathrm{d}x_{B^{}}^{}{}_{}{}^{C^{}}.$$
The two-dimensional spinor indices are raised and lowered with the symplectic form $`\epsilon _{A^{}B^{}}`$, such that $`\epsilon _{0^{}1^{}}=1`$ (see for a full account of the two-spinor formalism). We shall use the abstract index convention $`V^i=V^{(A^{}B^{})}=v^{(A^{}}\pi ^{B^{})}`$ based on an isomorphism $`T^i𝒲=S^{(A^{}}S^{B^{})}`$.
The projective mini-twistor space of $`^{2+1}`$ is the two-dimensional complex manifold $`𝒵=T^1`$ which is the total space of the line bundle $`𝒪(2)`$ of Chern class 2 over $`^1`$. Points of $`𝒵`$ correspond to null 2-planes in $`^{2+1}`$ via the incidence relation
$$x^{A^{}B^{}}\pi _A^{}\pi _B^{}=\omega .$$
(4.43)
Here $`(\omega ,\pi _0^{},\pi _1^{})`$ are homogeneous coordinates on $`𝒪(2)`$: $`(\omega ,\pi _A^{})(\rho ^2\omega ,\rho \pi _A^{})`$, where $`\rho ^{}`$. In the affine coordinates $`\stackrel{~}{\lambda }:=\pi _0^{}/\pi _1^{},\xi :=\omega /(\pi _1^{})^2`$ equation (4.43) is $`\xi =x+\stackrel{~}{\lambda }y+\stackrel{~}{\lambda }^2t.`$ First fix $`(\omega ,\pi _A^{})`$. If $`(\xi ,\stackrel{~}{\lambda })`$ are both real then (4.43) defines a null plane in $`^{2+1}`$. If both $`\xi `$ and $`\stackrel{~}{\lambda }`$ are complex then the solution to (4.43) is a time like curve in $`^{2+1}`$. We shall say that this curve is oriented to the future if $`\text{Im}\stackrel{~}{\lambda }>0`$ and to the past otherwise. If $`\stackrel{~}{\lambda }`$ is real and $`\xi `$ is complex then (4.43) has no solutions for finite $`x^{A^{}B^{}}`$.
An alternate interpretation of (4.43) is to fix $`x^{A^{}B^{}}`$. This determines $`\omega `$ as a function of $`\pi _A^{}`$ i.e. a section of $`𝒪(2)^1`$ when factored out by the relation $`(\omega ,\pi _A^{})(\rho ^2\omega ,\rho \pi _A^{})`$. These are embedded rational curves with normal bundle $`𝒪(2)`$. Two rational curves $`l_{p_1}`$ and $`l_{p_2}`$ (corresponding to $`(t_1,y_1,x_1)`$ and $`(t_2,y_2,x_2)`$ respectively) intersect at two points
$$\lambda _{1,2}=\frac{2R_2\sqrt{h(R,R)}}{2R_1},\text{where}R_i:=(t_1t_2,y_1y_2,x_1x_2).$$
Therefore the incidence of curves in $`𝒵`$ encodes the causal structure of $`^{2+1}`$ in the following sense: $`l_{p_1}`$ and $`l_{p_2}`$ intersect at (a) one point, (b) two real points, (c) two complex points conjugates of each other, iff $`p_1,p_2`$ are (a) null separated, (b) space-like separated, (c) time-like separated.
Examining the relevant cohomology groups shows that the moduli space of curves with normal bundle $`𝒪(2)`$ in $`𝒵`$ is $`^3`$. The real space-time $`^{2+1}`$ arises as the moduli space of curves that are invariant under the conjugation $`(\omega ,\pi _A^{})(\overline{\omega },\overline{\pi }_A^{})`$.
The correspondence space $`=^3\times ^1=\{(p,Z)^3\times 𝒵|Zl_p\}`$. By definition, it inherits fibrations over both $`^3`$ and $`𝒵`$ and the fibration of $`=^3\times ^1`$ over $`𝒵`$ has fibres spanned by the distribution $`L_A^{}=\pi ^B^{}_{A^{}B^{}}`$, where $`_{A^{}B^{}}x^{C^{}D^{}}=1/2(\epsilon _A^{}^C^{}\epsilon _B^{}^D^{}+\epsilon _B^{}^C^{}\epsilon _A^{}^D^{})`$. In the affine coordinates $`\pi ^A^{}=(1,\stackrel{~}{\lambda })`$ this distribution is
$$L_0^{}=_t\stackrel{~}{\lambda }_y,L_1^{}=_y\stackrel{~}{\lambda }_x$$
(we have ignored the constant factor $`\pi _1^{}`$). Note that this $`L_A^{}`$ is the special case $`u(x,y,t)=0`$ of the Lax pair (2.10) for the dKP equation.
We also define the correspondence space $`_W=^{2+1}\times ^1`$ for $`^{2+1}`$. Let $`𝒵_{}`$ be the sub-manifold of $`𝒵`$ preserved by the conjugation
$$(\omega ,\pi _0^{},\pi _1^{})(\overline{\omega },\overline{\pi _0^{}},\overline{\pi _1^{}}),$$
and let $`l_p`$ be the real line in $`𝒵_{}`$ that corresponds to $`p𝒲`$ and let $`Zl_p`$. The totally real correspondence space is a four-dimensional real manifold defined by $`_{}^4:=𝒵_{}\times ^{2+1}|_{Zl_p}`$ and can be represented as the set $`\stackrel{~}{\lambda }=\overline{\stackrel{~}{\lambda }}`$ or $`\pi _A^{}=\overline{\pi }_A^{}`$. The distribution $`L_A^{}\overline{L}_A^{}`$ is one dimensional, spanned by $`\overline{\pi }^A^{}\pi ^B^{}_{A^{}B^{}}`$, on the complement of $`_{}^4`$. On $`_{}^4`$ $`L_A^{}\overline{L}_A^{}`$ is two real dimensional as here $`L_A^{}=\overline{L}_A^{}`$. The real correspondence space $`_{}`$ divides $`_W=^{2+1}\times ^1`$ into two halves.
### 4.2 The Lax formulation and twistor correspondence
###### Proposition 4.1
Let $`V_1,V_2,V_3`$ be three independent holomorphic vector fields on a three dimensional complex manifold $`𝒲`$ such that
$$L_0^{}=V_1\stackrel{~}{\lambda }V_2+f_0^{}_{\stackrel{~}{\lambda }},L_1^{}=V_2\stackrel{~}{\lambda }V_3+f_1^{}_{\stackrel{~}{\lambda }}$$
(4.44)
is an integrable distribution for some functions $`f_0^{},f_1^{}`$, which are third-order polynomials in $`\stackrel{~}{\lambda }^1`$. Then there exists a one form $`\nu `$ such that the contravariant metric $`V_2V_21/2(V_1V_3+V_3V_1)`$ and $`\nu `$ give an EW structure on $`𝒲`$. Each EW structure arises in this way.
Remarks:
* The Lax pair (2.10) for the dKP equation is of course a special case of (4.44).
* The Lax formulations are widely applicable in the theory of integrable systems and so the above proposition can be applied outside twistor theory. It is however much easier to prove Proposition 4.1 using the twistor geometry, rather than an explicit calculation. This justifies adopting the spinor notation
$$_{A^{}B^{}}=\left(\begin{array}{cc}V_1& V_2\\ V_2& V_3\end{array}\right),f_A^{}=(f_0^{},f_1^{}),\pi ^A^{}=(1,\stackrel{~}{\lambda }),$$
in which the Lax pair has the compact form $`L_A^{}=\pi ^B^{}_{A^{}B^{}}+f_A^{}_{\stackrel{~}{\lambda }}`$. We shall use this notation in the proof of Proposition 4.1.
* The third order polynomials $`f_A^{}`$ contain eight functions not depending on $`\stackrel{~}{\lambda }`$. These can be reduced to four functions by choice of a suitable spin frame for which $`f_A^{}`$ become linear in $`\stackrel{~}{\lambda }`$. In this frame there exists a vector formula for $`\nu `$ in terms of $`\mathrm{\Gamma }_{ijk}`$, and $`f_A^{}`$.
* Proposition 4.1 holds for complex solutions and for any choice of signature for real space time.
Proof of Proposition 4.1. Assume that $`h=V_2V_21/2(V_1V_3+V_3V_1)`$ and $`\nu `$ gives an EW structure. Let $`V(\stackrel{~}{\lambda })=V_12\stackrel{~}{\lambda }V_2+\stackrel{~}{\lambda }^2V_3`$. Then $`g(V(\stackrel{~}{\lambda }),V(\stackrel{~}{\lambda }))=0`$ for all $`\stackrel{~}{\lambda }^1`$ so $`V(\stackrel{~}{\lambda })`$ determines a sphere of null vectors. Choose $`l_0^{}=V_1\stackrel{~}{\lambda }V_2,l_1^{}=V_2\stackrel{~}{\lambda }V_3`$ as a basis of the orthogonal complement of $`V(\stackrel{~}{\lambda })`$. For each $`\stackrel{~}{\lambda }^1`$ the vectors $`l_0^{},l_1^{}`$ give a null two-surface. It is well known that the EW equations on $`(h,\nu )`$ are equivalent to the integrability conditions of null, totally geodesic planes. Therefore the Frobenius theorem implies that the horizontal lifts
$$L_0^{}=V_1\stackrel{~}{\lambda }V_2+f_0^{}_{\stackrel{~}{\lambda }},L_1^{}=V_2\stackrel{~}{\lambda }V_3+f_1^{}_{\stackrel{~}{\lambda }}$$
of $`l_0^{},l_1^{}`$ to $`T(𝒲\times ^1)`$ span an integrable distribution. The functions $`f_0^{}`$ and $`f_1^{}`$ are third order in $`\stackrel{~}{\lambda }`$, because the Möbius transformations of $`^1`$ are generated by vector fields quadratic in $`\stackrel{~}{\lambda }`$, and $`l_0^{},l_1^{}`$ are linear $`\stackrel{~}{\lambda }`$.
The above argument can be made more explicit in spinor notation: let $`L_A^{}`$ be horizontal lift of $`l_A^{}=\pi ^B^{}_{A^{}B^{}}`$ to the weighted spin bundle (i.e. $`L_A^{}\pi _C^{}=0`$). This yields
$`L_A^{}`$ $`=`$ $`\pi ^B^{}_{A^{}B^{}}+\mathrm{\Gamma }_{A^{}B^{}C^{}D^{}}\pi ^B^{}\pi ^D^{}{\displaystyle \frac{}{\pi _C^{}}}`$ (4.45)
$`+{\displaystyle \frac{1}{2}}\nu _{B^{}D^{}}\pi ^B^{}\left(\pi ^D^{}{\displaystyle \frac{}{\pi ^A^{}}}{\displaystyle \frac{1}{2}}\pi _A^{}{\displaystyle \frac{}{\pi _D^{}}}\epsilon _{A^{}}^{}{}_{}{}^{D^{}}\pi {\displaystyle \frac{}{\pi }}\right),`$
where $`\mathrm{\Gamma }_{A^{}B^{}C^{}D^{}}`$ is spinor Levi–Civita connection defined by $`_{A^{}B^{}}\pi _C^{}=\mathrm{\Gamma }_{A^{}B^{}C^{}D^{}}\pi ^D^{}`$. The integrability conditions imply $`[L_A^{},L_B^{}]=0(\text{mod}L_A^{}).`$ The distribution $`L_A^{}`$, when projected to $`_𝒲`$ is given by (4.44), where
$$f_A^{}=\mathrm{\Gamma }_{A^{}B^{}C^{}D^{}}\pi ^B^{}\pi ^C^{}\pi ^D^{}+(1/4)\pi _A^{}\nu _{B^{}C^{}}\pi ^B^{}\pi ^C^{}.$$
$`\mathrm{}`$
The twistor space $`𝒵`$ for a solution to the EW equations on $`(𝒲,h,\nu )`$ associated to the Lax system on $`L_A^{}`$ as above is obtained by factoring the spin bundle $`𝒲\times ^1`$ by the twistor distribution (Lax pair) $`L_A^{}`$. This clearly has a projection $`q:𝒲\times ^1𝒵`$ and we have a double fibration
$$\begin{array}{ccccc}& & 𝒲\times ^1& & \\ & r& & q& \\ & 𝒲& & 𝒵& \end{array}$$
Each point $`p𝒲`$ determines a sphere $`l_p`$ made up of all the null totally geodesic two–surfaces through $`p`$. The normal bundle of $`l_p`$ in $`𝒵`$ is $`N=T𝒵|_{l_p}/Tl_p`$. This is a rank one vector bundle over $`^1`$, therefore it has to be one of the standard line bundles $`𝒪(n)`$.
###### Lemma 4.2
The holomorphic curves $`l_p:=q(_p^1)`$ where $`_p^1=r^1(p)`$, $`p𝒲`$, have normal bundle $`N=𝒪(2)`$.
Proof. To see this, note that $`N`$ can be identified with the quotient $`r^{}(T_p𝒲)/\{\mathrm{span}L_0^{},L_1^{}\}`$. In their homogeneous form the operators $`L_A^{}`$ have weight one, so the distribution spanned by them is isomorphic to the bundle $`^2𝒪(1)`$. The definition of the normal bundle as a quotient gives a sequence of sheaves over $`^1`$.
$$0^2𝒪(1)^3N0$$
and we see that $`N=𝒪(2)`$, because the last map, in the spinor notation, is given explicitly by $`V^{A^{}B^{}}V^{A^{}B^{}}\pi _A^{}\pi _B^{}`$ clearly projecting onto $`𝒪(2)`$. $`\mathrm{}`$
A generalisation of the flat mini-twistor correspondence to the 2+1 EW spaces is given by the following proposition
###### Proposition 4.3 ()
Any solution to the EW equations (1.5) is equivalent to a complex surface $`𝒵`$ with a family of rational curves with normal bundle $`𝒪(2)`$.
Points of $`𝒲`$ correspond to curves in $`𝒵`$ with self-intersection number 2. The Kodaira theorem applied to deformations preserving the real structure of $`𝒵`$ guarantees the existence of a three-dimensional complex family of such curves. Points of $`𝒵`$ correspond to totally geodesic hyper-surfaces in $`𝒲`$. Non-null geodesics in $`𝒲`$ consist of all the curves in $`𝒵`$ which intersect at two fixed points in $`𝒵`$. Null geodesics correspond to curves passing through one point with a given tangent direction. Thus the projective and conformal structures can be reconstructed. $`\mathrm{}`$
### 4.3 Mini-twistor spaces from twistor spaces
###### Proposition 4.4
All Einstein–Weyl spaces arise as symmetry reductions of hyper-Hermitian metrics (or indefinite hyper-Hermitian metrics) in four-dimensions.
Proof. Consider an EW structure with the corresponding Lax pair (4.44). Choose a spin frame in which $`f_A^{}`$ is linear in $`\stackrel{~}{\lambda }`$; $`f_A^{}=U_A^{}+\stackrel{~}{\lambda }W_A^{}`$ (this is always possible by making a suitable Möbius transformation of $`^1`$ and choosing an appropriate conformal scale) , and introduce a new spectral parameter $`\lambda :=\stackrel{~}{\lambda }z`$ for some $`z`$. Nothing in the $`L_A^{}`$ depends on $`z`$ so we can replace $`_{\stackrel{~}{\lambda }}`$ by $`_z`$. This yields (with a dropped prime)
$$L_A=_{A0^{}}\lambda _{A1^{}},$$
where
$`_{00^{}}`$ $`=`$ $`_{0^{}0^{}}+z_{0^{}1^{}}+(U_0^{}+zW_0^{})_z,`$
$`_{10^{}}`$ $`=`$ $`_{1^{}0^{}}+z_{1^{}1^{}}+(U_1^{}+zW_1^{})_z,`$
$`_{01^{}}`$ $`=`$ $`_{0^{}1^{}}+W_0^{}_z,`$
$`_{11^{}}`$ $`=`$ $`_{1^{}1^{}}+W_1^{}_z`$
where $`U_0^{},U_1^{},W_0^{},W_1^{}`$ are four functions not depending of $`\lambda `$. One is left with a Lax pair for a hyper-Hermitian four manifold because $`L_A`$ can be made to commute exactly (as in Proposition 2.6) by choosing two solution to the background coupled neutrino equation (see for details). This Lax pair has an obvious symmetry $`_z`$. $`\mathrm{}`$
Remark: All EW spaces arise as symmetry reductions of a pair of coupled PDEs , associated to hyper-Hermitian four manifolds. In Proposition 4.44 was proven using different methods for EW spaces of Riemannian signature.
The twistor construction of Hitchin can be viewed as a reduction of Penrose’s Nonlinear Graviton construction. It follows from (compare Proposition 2.5) that the mini-twistor space $`𝒵`$ corresponding to $`𝒲`$ is a factor space $`𝒫𝒯/𝒦`$ where $`𝒫𝒯`$ is the twistor space of $`(,g)`$ and $`𝒦`$ is a holomorphic vector field on $`𝒫𝒯`$ corresponding to a conformal Killing vector $`K`$.
Below we shall state the Penrose result extended to the Einstein and hyper-Hermitian cases:
###### Proposition 4.5
Let $`𝒫𝒯`$ be a three-dimensional complex manifold with a four-dimensional family of rational curves (invariant under a complex conjugation with fixed points) with normal bundle $`𝒪(1)𝒪(1)`$. Then the moduli space $``$ of these sections is equipped with an ASD conformal structure $`[g]`$ of signature $`(++)`$. Conversely given an ASD four-manifold there will always exists a corresponding twistor space. Moreover $``$ is:
* Hyper-Kähler iff there exists a projection $`\mu :𝒫𝒯^1`$, and each fibre of this projection is equipped with an $`\mu ^{}𝒪(2)`$ valued symplectic form (equivalently, we can require that the canonical bundle $`\kappa `$ of $`𝒫𝒯`$ is $`\kappa =\mu ^{}𝒪(4)`$).
* Hyper-Hermitian iff there is a projection $`\mu :𝒫𝒯^1`$ .
* Einstein ($`R_{ab}=\mathrm{\Lambda }g_{ab}`$) iff there exists a contact structure $`\tau \mathrm{\Lambda }^2(T^{}𝒫𝒯)𝒪(2)`$, where now $`𝒪(2)=\kappa ^{1/2}`$, and $`\kappa `$ is the canonical bundle $`\mathrm{\Omega }^3`$, such that $`\tau \mathrm{d}\tau =\mathrm{\Lambda }\xi `$ where $`\xi \mathrm{\Omega }^3\kappa ^1`$ .
#### 4.3.1 Construction of the two–form
Consider an ASD four–manifold $`(,[g])`$. Define the non-projective twistor space, $`𝒯`$, to be the total space of the line bundle $`\kappa ^{1/4}𝒫𝒯`$ where $`\kappa =\mathrm{\Omega }^3`$ is the canonical bundle. In the conformally-flat case $`𝒯`$ is the tautological line bundle $`𝒪(1)`$, i.e. $`^4^3`$, and we will also use this notation, $`𝒯=𝒪(1)`$ in the curved case. The nonprojective spin bundle $`S_A^{}`$ is defined to be the total space of the pullback of this line bundle to the correspondence space $`=\times ^1`$. Clearly $`S_A^{}=\times ^2`$. The fibration $`q:S^A^{}𝒯`$ is spanned by a lift of the twistor distribution or Lax pair. The non-projective spin bundle is the total space of a line bundle, which we will also denote by $`𝒪(1)`$, over $``$. (Note that in the hyper-Hermitian case the line bundles $`𝒪(n)`$ just defined will not be the same as $`\mu ^{}𝒪(n)`$ unless $`(,[g])`$ is in fact hyper-Kahler.)
The space $`𝒯`$ admits an Euler vector field $`\mathrm{{\rm Y}}`$ being the total space a of line bundle, and a tautological three-form, $`\xi `$ the pullback of the tautological three-form on $`\kappa `$. These satisfy $`_\mathrm{{\rm Y}}\xi =4\xi `$. Let $`\varphi =\mathrm{d}\xi `$, then $`\xi =4\varphi (\mathrm{{\rm Y}},\mathrm{},\mathrm{})`$. $`\xi `$ can be thought of as a form on $`𝒫𝒯`$ with values in the dual canonical bundle $`\kappa ^{}`$.
We now impose a symmetry: let $`K,\stackrel{~}{K}`$, and $`𝒦`$ be respectively: a conformal Killing vector on $``$, its lift to the correspondence space $`\times ^1`$, and the holomorphic vector field on $`𝒯`$ which is the push-forward of $`\stackrel{~}{K}`$.
###### Proposition 4.6
The two form $`\stackrel{~}{\mathrm{\Sigma }}:=q^{}\varphi (𝒦,\mathrm{{\rm Y}},\mathrm{},\mathrm{})\mathrm{\Lambda }^2(T^{}S^A^{})`$ satisfies
$$\stackrel{~}{\mathrm{\Sigma }}\stackrel{~}{\mathrm{\Sigma }}=0,\mathrm{d}\stackrel{~}{\mathrm{\Sigma }}=\beta \stackrel{~}{\mathrm{\Sigma }}_{\stackrel{~}{K}}\stackrel{~}{\mathrm{\Sigma }}=0$$
(4.46)
for some one-form $`\beta `$ homogeneous of degree $`0`$ in $`\pi ^A^{}`$.
Proof: It follows from the definition of $`\stackrel{~}{\mathrm{\Sigma }}`$ that the integrable twistor distribution belongs is the kernel of $`\stackrel{~}{\mathrm{\Sigma }}`$. Therefore equations (4.46) follow from Frobenius’ theorem. The one-form $`\beta `$ is defined up to the addition of $`\mathrm{d}(\mathrm{ln}\sigma )`$ where $`\sigma `$ is a twistor function homogeneous of degree 0. $`\mathrm{}`$
From $`_\mathrm{{\rm Y}}\stackrel{~}{\mathrm{\Sigma }}=4\stackrel{~}{\mathrm{\Sigma }}`$ and $`\mathrm{{\rm Y}}\text{}\stackrel{~}{\mathrm{\Sigma }}=0`$ it follows that $`\stackrel{~}{\mathrm{\Sigma }}`$ descends to $``$ where it takes values in $`𝒪(4)`$. Note however that $`\mathrm{d}\stackrel{~}{\mathrm{\Sigma }}`$ does not descend as $`\mathrm{{\rm Y}}\text{}\mathrm{d}\stackrel{~}{\mathrm{\Sigma }}=_\mathrm{{\rm Y}}\stackrel{~}{\mathrm{\Sigma }}0`$. To differentiate $`\stackrel{~}{\mathrm{\Sigma }}`$ on $``$ we need a nonzero section of $`𝒪(4)`$ in order to dehomogenise $`\stackrel{~}{\mathrm{\Sigma }}`$. When $`(,g)`$ is ASD Einstein or vacuum we can find a section of $`𝒪(4)`$ to dehomogenise $`\stackrel{~}{\mathrm{\Sigma }}`$. This section necessarily has zeroes, and so equivalently, this requires the existence of a divisor description of the dual canonical bundle. This can be seen from the twistor construction.
* Vacuum case: The twistor space fibres over $`^1`$ and so we can pull back $`\pi \mathrm{d}\pi `$ to $`𝒫𝒯`$.
Let $`𝒦`$ be a holomorphic vector field on $`𝒫𝒯`$ such that $`_𝒦\mathrm{\Sigma }_\lambda =\eta \mathrm{\Sigma }_\lambda `$ ($`𝒦`$ corresponds to a Homothetic Killing vector on $``$). The function $`D:=𝒦\text{}\pi \mathrm{d}\pi `$ is a section of $`𝒪(2)`$ and the two-form $`D^2𝒦\text{}\xi `$ descends to the mini-twistor space $`𝒵`$.
* Einstein case: Let $`𝒫𝒯_E`$ be the projective twistor space corresponding to a solution of the ASD Einstein equations. It is equipped with a contact structure $`\tau \mathrm{\Lambda }^2(T^{}𝒫𝒯_E)𝒪(2)`$ such that $`\tau \mathrm{d}\tau =\mathrm{\Lambda }\xi `$. $`\mathrm{d}\tau `$ defines a holomorphic symplectic structure on the non-projective twistor space $`𝒯_E`$. If $`K`$ is a Killing vector on an ASD Einstein manifold then the corresponding holomorphic vector field on the non-projective twistor space is Hamiltonian with respect to $`\mathrm{d}\tau `$. To see this, define a section of $`O(2)`$ by $`D:=𝒦\text{}\tau `$. We have $`\mathrm{d}D=_𝒦\tau 𝒦\text{}\mathrm{d}\tau =𝒦\text{}\mathrm{d}\tau `$ as $`𝒦`$ is a symmetry.
On the projective spin bundle $``$ define
$$\mathrm{\Pi }:=D^2\stackrel{~}{\mathrm{\Sigma }}.$$
We have the following result:
###### Proposition 4.7
The two-form $`\mathrm{\Pi }`$ is well defined on the Einstein–Weyl correspondence space $`_W`$. It satisfies
$$\mathrm{d}\mathrm{\Pi }=0,\mathrm{\Pi }\mathrm{\Pi }=0,$$
(4.47)
where $`\mathrm{d}=\mathrm{d}x^i_i+\mathrm{d}\stackrel{~}{\lambda }_{\stackrel{~}{\lambda }}`$ is the exterior derivative on $`_W`$. Any two linearly independent vectors $`L_A^{}`$ such that $`L_A^{}\text{}S=0`$ form a Lax pair for the EW equations.
Proof. The simplicity follows from $`\stackrel{~}{\mathrm{\Sigma }}\stackrel{~}{\mathrm{\Sigma }}=0`$. In the vacuum case the two form
$$\mathrm{\Pi }=q^{}\frac{𝒦\text{}\xi }{𝒦\text{}(\pi \mathrm{d}\pi )}$$
(4.48)
is a pull back of a closed and simple form on $`𝒫𝒯`$. In the Einstein case
$$\mathrm{\Pi }=D^2q^{}𝒦\text{}(\mathrm{\Lambda }\tau \mathrm{d}\tau )=\mathrm{d}(\mathrm{\Lambda }\tau /D).$$
Therefore Einstein–Weyl metrics which come from ASD Einstein and hyper-Kähler four manifolds give rise to the same structure on the reduced spin bundle. The form $`\mathrm{\Pi }`$ descends to $`_W`$ because $`\stackrel{~}{K}\text{}\mathrm{d}\mathrm{\Pi }=0`$ and $`\mathrm{d}(\stackrel{~}{K}\text{}\mathrm{\Pi })=0`$. $`\mathrm{}`$
Remark. In certain dispersionless integrable systems were expressed in terms of $`\mathrm{\Pi }`$ satisfying (4.47).
The two form $`\stackrel{~}{\mathrm{\Sigma }}`$ can be equivalently constructed from the data on $``$ as follows. Let $`K`$ be a Killing vector on a general ASD conformal manifold $`(,[g])`$, and let $`\mathrm{\Xi }`$ be a volume form on the non-projective primed spin bundle $`S^A^{}`$. Define the two form on $`S^A^{}`$
$$\stackrel{~}{\mathrm{\Sigma }}:=\mathrm{\Xi }(L_0,L_1,\stackrel{~}{K},\mathrm{{\rm Y}}_\mathrm{\Xi },\mathrm{},\mathrm{}).$$
(4.49)
Here $`\mathrm{{\rm Y}}_\mathrm{\Xi }=\pi ^A^{}/\pi ^A^{}`$ is the Euler vector field on $`S^A^{}`$, $`L_A`$ is the twistor distribution, and $`\stackrel{~}{K}`$ is a Lie lift of $`K`$ to $`S^A^{}`$. Now assume that $`(,g)`$ is also vacuum. Consequently $`_{AA^{}}K_{}^{A}{}_{B^{}}{}^{}=const`$ and the spin bundle is equipped with a canonical divisor<sup>2</sup><sup>2</sup>2We assume that $`_{AA^{}}K_{}^{A}{}_{B^{}}{}^{}0`$. If $`_{AA^{}}K_{}^{A}{}_{B^{}}{}^{}=0`$ then $`K`$ is triholomorphic and a section of $`𝒪(2)`$ which descends to the reduced spin bundle is $`(\iota \pi )^2`$ where $`\iota _A^{}`$ is any constant spinor. $`D:=\pi ^A^{}\pi ^B^{}_{AA^{}}K_{}^{A}{}_{B^{}}{}^{}𝒪(2)`$ which descends to the reduced spin bundle<sup>3</sup><sup>3</sup>3By the reduced spin bundle (correspondence space) we mean the space of orbits of $`\stackrel{~}{K}`$ in $`S^A^{}`$ (in $``$). (Figure 1). It is easy to prove that now
$`\stackrel{~}{\mathrm{\Sigma }}`$ $`=`$ $`\pi _A^{}\pi _B^{}\pi _C^{}\pi _D^{}\varphi ^{A^{}B^{}}\mathrm{\Sigma }^{C^{}D^{}}+\pi _A^{}\pi _B^{}\pi _C^{}\mathrm{d}\pi ^C^{}(K\text{}\mathrm{\Sigma }^{A^{}B^{}}),`$
$`\beta `$ $`=`$ $`{\displaystyle \frac{4\varphi _{A^{}B^{}}\pi ^A^{}\mathrm{d}\pi ^B^{}}{\pi _A^{}\pi _B^{}\varphi ^{A^{}B^{}}}}=\mathrm{d}\mathrm{ln}D^2`$
$`\mathrm{\Pi }`$ $`=`$ $`\mathrm{d}\lambda {\displaystyle \frac{K\text{}\mathrm{\Sigma }(\lambda )}{D^2}}{\displaystyle \frac{\mathrm{\Sigma }(\lambda )}{D}},\text{where}\mathrm{\Sigma }(\lambda )=\pi _A^{}\pi _B^{}\mathrm{\Sigma }^{A^{}B^{}}.`$ (4.50)
From the last formula it follows that to construct $`\mathrm{\Pi }`$ one should rewrite $`\mathrm{\Sigma }(\lambda )/D`$ in the coordinates in which $`K=_t`$, and then replace all $`\mathrm{d}t`$s by the differentials of a suitably defined invariant spectral parameter.
Example. We shall now illustrate the construction of $`\mathrm{\Pi }`$ with a simple example. Let $`2\mathrm{d}w\mathrm{d}\stackrel{~}{w}2\mathrm{d}z\mathrm{d}\stackrel{~}{z}`$ be a flat metric on $`^{2,2}`$ and let $`K=z_z\stackrel{~}{z}_{\stackrel{~}{z}}`$ be a Killing vector. The flat twistor distribution and the lifted symmetry are:
$$L_0=_{\stackrel{~}{w}}\lambda _z,L_1=_{\stackrel{~}{z}}\lambda _w,\stackrel{~}{K}=z_z\stackrel{~}{z}_{\stackrel{~}{z}}+\lambda _\lambda .$$
The volume form on $``$ and the two-form $`\mathrm{\Sigma }(\lambda )`$ are given by
$$\mathrm{\Xi }=\mathrm{d}\lambda \mathrm{d}z\mathrm{d}\stackrel{~}{z}\mathrm{d}w\mathrm{d}\stackrel{~}{w},\mathrm{\Sigma }(\lambda )=\lambda ^2\mathrm{d}\stackrel{~}{w}\mathrm{d}\stackrel{~}{z}+\lambda (\mathrm{d}w\mathrm{d}\stackrel{~}{w}\mathrm{d}z\mathrm{d}\stackrel{~}{z})+\mathrm{d}w\mathrm{d}z.$$
In the covariantly constant frame we introduce $`2r:=\text{ln}(z\stackrel{~}{z}),\mathrm{\hspace{0.33em}2}\varphi :=\text{ln}(z/\stackrel{~}{z})`$, so that $`\stackrel{~}{K}=_\varphi +\lambda _\lambda `$. In these coordinates
$$\mathrm{\Sigma }(\lambda )=\lambda ^2e^{r\varphi }\mathrm{d}\stackrel{~}{w}(\mathrm{d}r\mathrm{d}\varphi )+\lambda (\mathrm{d}w\mathrm{d}\stackrel{~}{w}+2e^{2r}\mathrm{d}r\mathrm{d}\varphi )+e^{r+\varphi }\mathrm{d}w(\mathrm{d}r+\mathrm{d}\varphi )$$
and (from (4.3.1))
$$\mathrm{\Pi }=e^r(\mathrm{d}\stackrel{~}{w}\mathrm{d}\stackrel{~}{\lambda }+\stackrel{~}{\lambda }^2\mathrm{d}w\mathrm{d}\stackrel{~}{\lambda }+\stackrel{~}{\lambda }\mathrm{d}\stackrel{~}{w}\mathrm{d}r\stackrel{~}{\lambda }^1\mathrm{d}w\mathrm{d}r)+2\stackrel{~}{\lambda }^1e^{2r}\mathrm{d}r\mathrm{d}\stackrel{~}{\lambda }\mathrm{d}w\mathrm{d}\stackrel{~}{w}$$
(4.51)
where $`\stackrel{~}{\lambda }=\lambda e^\varphi `$ is an invariant spectral parameter.
The two form $`\mathrm{\Pi }`$ can be also obtained as a pull-back from $`𝒫𝒯`$. Local inhomogeneous coordinates on $`𝒫𝒯`$ pulled back to $``$ are given by $`(\lambda ,\mu ^1=\lambda \stackrel{~}{w}+z,\mu ^0=\lambda \stackrel{~}{z}+w).`$ The holomorphic vector field on $`𝒫𝒯`$ is $`𝒦=\mu ^0_{\mu ^0}+\lambda _\lambda `$. From (4.48) we have
$$q^{}(𝒦\text{}(\mathrm{d}\lambda \mathrm{d}\mu ^0\mathrm{d}\mu ^1)=(\mu ^0\mathrm{d}\lambda \lambda \mathrm{d}\mu ^1)\mathrm{d}\mu ^1=\lambda ^2\mathrm{d}\mu ^1\mathrm{d}(\mu ^0/\lambda ).$$
Thus
$$\mathrm{\Pi }=\mathrm{d}\mu ^1\mathrm{d}(\mu ^0/\lambda )=\mathrm{d}P\mathrm{d}Q$$
which agrees with (4.51). Here $`P=\stackrel{~}{w}+\stackrel{~}{\lambda }^1e^r`$ and $`Q=\stackrel{~}{\lambda }e^r+w`$ are coordinates on mini-twistor space pulled back to the reduced spin bundle.
## 5 Twistor theory of the dKP Einstein-Weyl structures
Here we give an account of the twistor theory of the dKP EW metrics, and the dKP equation (some connections between a twistor theory and the dKP equations have been discussed in ). We shall also characterise all four dimensional hyper-Kähler and ASD Einstein metrics that give rise to the dKP EW structures.
Define the non-projective twistor space, $`𝒴`$ corresponding to a Weyl space $`𝒲`$, to be the total space of the line bundle $`\kappa ^{1/4}𝒵`$ where $`\kappa =\mathrm{\Omega }^2`$ is the canonical bundle of $`𝒵`$. The nonprojective spin bundle $`S_A^{}𝒲`$ is the rank two vector bundle defined to be the total space of the pullback of this line bundle to the correspondence space $`𝒲\times ^1`$. The fibration $`q:S^A^{}𝒴`$ is spanned by a lift of the mini-twistor distribution $`L_A^{}`$ (4.44).
Any shear-free null geodesic congruence of the Einstein-Weyl structure determines a one-dimensional sub-manifold in $`𝒵`$ (this is a reduction of the 4-dimensional Kerr theorem). A codimension–one submanifold determines a line bundle $`[D]`$ by the divisor construction; $`[D]`$ admits a section $`D`$ that vanishes precisely on the given submanifold.
When the Einstein–Weyl geometry arises from a solution of the dKP equation the dual canonical bundle $`\kappa ^1`$ of the minitwistor space admits a fourth root that is given by the divisor construction, that is it admits a section $`D`$ that vanishes on a codimension-one subset. In general, as seen above, if the Einstein-Weyl geometry is a reduction of an ASD Einstein, or hyper-Kähler four-manfiold, then $`\kappa ^{1/2}`$ admits a section whose zero set will generally have two components in the neighbourhood of a line. For an Einstein-Weyl dKP solution, the two ‘divisor curves’ in Fig (1) degenerate to one curve. This observation gives rise to a twistor characterisation of solutions to the dKP equation
###### Proposition 5.1
There is a one to one correspondence between Einstein-Weyl spaces obtained from solutions to the dKP equation and two-dimensional complex manifolds with
* A three parameter family of rational curves with normal bundle $`𝒪(2)`$.
* A global section $`l`$ of $`\kappa ^{1/4}`$, where $`\kappa `$ is the canonical bundle.
In order to obtain a real Einstein-Weyl structure, we require an antiholomorphic involution fixing a real slice, leaving a rational curve invariant and leaving the section of $`\kappa ^{1/4}`$ above invariant.
Proof. The global section $`l`$ of $`\kappa ^{1/4}`$, when pulled back to $`S_A^{}`$ determines a homogeneity degree one function on each fibre of $`S_A^{}`$ and so must, by globality, be given by $`l=\iota ^A^{}\pi _A^{}`$ and since $`l`$ is pulled back from twistor space, it must satisfy $`L_A^{}l=0`$. This implies $`\stackrel{~}{D}_{A^{}(B^{}}\iota _{C^{})}=0`$, and (after some algebraic manipulations)
$$\stackrel{~}{D}_{A^{}B^{}}\iota ^C^{}=0,$$
where $`\stackrel{~}{D}`$ is a covariant weighted derivative.
Therefore the null vector field $`l^a=\iota ^A^{}\iota ^B^{}`$ is covariantly constant. The Lemma 2.3 implies that the conformal weight of $`\iota ^A^{}`$ is $`1/4`$ and hence that of $`l^a`$ is $`1/2`$. This weight can be deduced from the correspondence as follows: the two form $`\stackrel{~}{\mathrm{\Sigma }}=\pi _A^{}\pi _B^{}e^{A^{}B^{}}\epsilon ^{C^{}D^{}}\pi _C^{}\mathrm{d}\pi _D^{}`$ has conformal weight $`0`$ on $`S^A^{}`$. $`e^{A^{}B^{}}`$ has weight 0, and $`\epsilon ^{A^{}B^{}}`$ weight $`1`$ so $`\pi _A^{}`$ has weight $`1/4`$. The global section $`\pi _A^{}\iota ^A^{}`$ is weightless so the weight of $`\iota ^A^{}`$ is $`1/4`$. Hence by Proposition 2.2 the corresponding Einstein-Weyl space arises from a solution to the dKP equation.
Conversely, given a solution to (2.9) one can obtain $`𝒵`$ as a factor space of $`𝒲\times ^1`$ by the distribution (2.10) and the covariant constant weighted null vector $`l^a=\iota ^A^{}\iota ^B^{}`$ gives rise to the section $`l=\iota ^A^{}\pi _A^{}`$ of $`\kappa ^{1/4}`$ $`\mathrm{}`$
Remark: Note that there is not a $`11`$ correspondence between such twistor spaces and solutions to the dKP equation on account of the coordinate freedom (2) and (2). The coordinate choices implicit in a solution to the dKP equation can be encoded on the twistor space in the choice of the coordinates near the divisor as follows.
Let $`\widehat{P},\widehat{Q}`$ be local coordinates on a neighbourhood of the divisor in $`𝒵`$ such that $`\widehat{Q}=0`$ on the divisor and, setting $`Q=\widehat{Q}^1,P=\widehat{P}/\widehat{Q}^2`$ on the complement of the divisor,we have
$$\mathrm{\Pi }=\mathrm{d}P\mathrm{d}Q=\widehat{Q}^4\mathrm{d}\widehat{P}\mathrm{d}\widehat{Q}.$$
Consider a graph of a rational curve $`\widehat{P}(\widehat{Q})`$. Parametrise the curve by $`(t,y,x)`$ as follows:
$$t:=\widehat{P}|_{\widehat{Q}=0},y:=\frac{\mathrm{d}\widehat{P}}{\mathrm{d}\widehat{Q}}|_{\widehat{Q}=0},x:=\frac{1}{2}\frac{\mathrm{d}^2\widehat{P}}{\mathrm{d}\widehat{Q}^2}|_{\widehat{Q}=0}.$$
Therefore the local coordinates $`P,Q`$ have the following expansion near $`\stackrel{~}{\lambda }=\mathrm{}`$
$$Q:=\stackrel{~}{\lambda }+\underset{i=1}{\overset{\mathrm{}}{}}u_i\stackrel{~}{\lambda }^i,P=\underset{i=1}{\overset{\mathrm{}}{}}w_iQ^i+x+Qy+Q^2t$$
(after performing an $`SL(2,)`$ transformation and choosing a spin frame such that the constant term in the Laurent expansion of $`Q`$ vanishes). When we pull the mini-twistor coordinates back to $``$, then $`u_i,w_i`$ become functions of $`(x,y,t)`$. The functions $`P`$ and $`Q`$ are solutions of Lax equations $`L_A^{}P=L_A^{}Q=0`$. They form a local Darboux atlas as $`\mathrm{\Pi }=\mathrm{d}P\mathrm{d}Q`$, where $`\mathrm{\Pi }`$ is given by (2.8).
$$\mathrm{\Pi }=\mathrm{d}x\mathrm{d}\stackrel{~}{\lambda }+\mathrm{d}y\mathrm{d}(\frac{\stackrel{~}{\lambda }^2}{2}+u_1)+\mathrm{d}t\mathrm{d}(\frac{\stackrel{~}{\lambda }^3}{3}+\stackrel{~}{\lambda }u_1+w_1).$$
The poles of $`\mathrm{\Pi }`$ occur on the divisor. Now $`\mathrm{\Pi }`$ is a pull back of a two-form from a two-dimensional manifold. Therefore is satisfies $`\mathrm{\Pi }\mathrm{\Pi }=0`$, which yields $`w_{1}^{}{}_{x}{}^{}=u_{1}^{}{}_{y}{}^{}`$ and the dKP equation (2.9) for $`u_1`$.
Thus, a solution to the dKP equation corresponds to a EW mini-twistor space as described in Proposition 5.1 together with a Darboux coordinate system as above on the third formal neighbourhood of the divisor. \[It seems likely that the Benney hierarchy will similarly correspond to the EW dKP minitwistor space as above together with the Darboux coordinate system on a neighbourhood of the divisor defined now to all orders.\]
Now we are in a position to give a characterisation of the hyper-Kähler metrics (2.27).
###### Proposition 5.2
Let $`g`$ be an indefinite hyper-Kähler metric with a symmetry $`K`$ satisfying $`\mathrm{d}K_+\mathrm{d}K_+=0`$. Then $`g`$ is locally of the form (2.27).
Proof. Let $`𝒦`$ be a vector field (corresponding to $`K`$) on a twistor space of $`(,g)`$. The divisor
$$𝒦\text{}\pi \mathrm{d}\pi =\pi _A^{}\pi _B^{}\varphi ^{A^{}B^{}}$$
descends to the minitwistor space. If $`\mathrm{d}K_+`$ is null then $`\varphi _{A^{}B^{}}=(1/2)_{AA^{}}K_B^{}^A=\iota _A^{}\iota _B^{}`$ for some constant spinor $`\iota ^A^{}`$. Therefore $`\pi \iota `$ on $`𝒫𝒯`$ defines a divisor in $`𝒵`$. It takes values in $`\kappa ^{1/4}`$ because the canonical bundle of $`𝒫𝒯`$ is the square of the pullback of the canonical bundle of $`^1`$. The assumptions of Proposition 5.1 are satisfied and so the EW structure corresponding to $`𝒵`$ is of the form (2.11). Therefore it follows from Proposition 2.5 that the metric $`g`$ is given by
$$g=\mathrm{\Omega }(\stackrel{~}{V}(\mathrm{d}\stackrel{~}{y}^24\mathrm{d}\stackrel{~}{x}\mathrm{d}\stackrel{~}{t}4\stackrel{~}{u}\mathrm{d}\stackrel{~}{t}^2)\stackrel{~}{V}^1(\mathrm{d}\stackrel{~}{z}+\stackrel{~}{\alpha })^2)=\mathrm{\Omega }\stackrel{~}{g},$$
where $`\stackrel{~}{u}(\stackrel{~}{x},\stackrel{~}{y},\stackrel{~}{t})`$ is a solution to dKP $`(\stackrel{~}{V},\stackrel{~}{\alpha })`$ is a solution to the monopole equation (2.23), and $`\mathrm{\Omega }`$ is a conformal factor. Calculating the scalar curvature of the metric $`\stackrel{~}{g}`$ yields
$$\stackrel{~}{R}=8(\stackrel{~}{V}_{\stackrel{~}{y}\stackrel{~}{y}}\stackrel{~}{V}_{\stackrel{~}{x}\stackrel{~}{t}}+(\stackrel{~}{u}\stackrel{~}{V})_{\stackrel{~}{x}\stackrel{~}{x}})\stackrel{~}{V},$$
and so $`\stackrel{~}{R}=0`$ because $`\stackrel{~}{V}`$ satisfies (2.26). However the metric $`g`$ is hyper-Kähler, therefore its scalar curvature also vanishes. As a consequence we deduce that $`\mathrm{\Omega }=\mathrm{\Omega }(\stackrel{~}{t})`$. Now we can use the coordinate freedom (2) to absorb $`\mathrm{\Omega }`$ in the solution to the dKP equation. This yields
$$g=(V(\mathrm{d}y^24\mathrm{d}x\mathrm{d}t4u\mathrm{d}t^2)V^1(\mathrm{d}z+\alpha )^2)=\mathrm{\Omega }\stackrel{~}{g},$$
(5.52)
where $`(V,\alpha )`$ is another solution to the monopole equation. In section 2.1 we showed that this metric is hyper-Kähler metric if $`V`$ is a multiple of $`u_x`$.
Consider the metric (5.52) with an arbitrary monopole $`V`$ (an arbitrary solution to the linearised dKP equation 2.26). The self-dual derivative of the isometry $`K=_z`$ is given by $`\varphi _{A^{}B^{}}=(u_x/V)\iota _A^{}\iota _B^{}`$, for some constant spinor $`\iota _A^{}`$. The well known identity $`_a_bK_c=R_{bcad}K^d`$ and the vacuum condition yield $`_a\varphi _{B^{}C^{}}=0`$. Therefore (5.52) is hyper-Kähler iff $`u_x/V=const`$. $`\mathrm{}`$
Remarks:
* This Proposition corrects an omission made in the classification of complexified hyper-Kähler spaces with symmetry. In the Appendix we shall demonstrate explicitly that the dKP equation is a reduction of the second heavenly equation considered in .
* Metrics (5.52) with $`Vconst\times u_x`$ are not vacuum, but they admit a covariantly constant real spinor. The full characterisation of these metrics will be given in our subsequent paper.
###### Proposition 5.3
All EW structures which arise from indefinite ASD Einstein metric with a symmetry $`K`$ satisfying $`\mathrm{d}K_+\mathrm{d}K_+=0`$ are locally of the form (2.11).
Proof. The canonical divisor $`D:=𝒦\text{}\tau `$ (where $`\tau `$ is the contact structure) descends to a mini-twistor space. Because $`\mathrm{d}K_+`$ is null the square root of $`D`$ exists and takes its values in $`\kappa ^{1/4}`$. $`\mathrm{}`$
## 6 Symmetry reductions of hyper-Kähler metrics in $`2+2`$ signature
Symmetry reductions of the hyper-Kähler condition on a real four-dimensional Riemannian metric have been completely classified:
* If the symmetry is tri-holomorphic, then the corresponding metric belongs to the Gibbons–Hawking class , and is given by a solution to the Laplace equation in three dimensions. The resulting Einstein–Weyl structures are trivial, and their mini-twistor space is $`T^1`$.
* Hyper-Kähler metrics with non-triholomorphic Killing vectors are given by solutions to the $`SU(\mathrm{})`$ Toda equation . The corresponding EW structures are characterised by the existence of a shear-free, twist-free geodesic congruence . Mini-twistor spaces are in this case equiped with a canonical divisor (two one-dimensional complex sub-manifolds) taking its values in $`𝒪(2)`$ . In EW Toda structures were characterised in terms of weighted vector fields.
* Hyper-Kähler metrics with tri-holomorphic conformal symmetries yield a class of EW structures (called hyper-CR EW structures) characterised by the existence of a sphere of shear-free, divergence-free geodesic congruences . The corresponding mini-twistor spaces are fibred over $`^1`$.
* Hyper-Kähler metrics with non-tri-holomorphic, conformal symmetry (and the resulting EW structures) are given by solutions to a certain second order integrable equation in three dimensions . This equation gives $`SU(\mathrm{})`$-Toda and hyper-CR Einstein-Weyl structures as limiting cases. The EW structures arising from conformal, non-tri-holomorphic reductions are characterised by the existence of a shear-free geodesic congruence for which the twist is a constant multiple of the divergence .
The above list is not complete if one considers Hyper-Kähler metrics in $`(++)`$ signature. The existence of null structures of various kinds allows two additional types of symmetries:
* Hyper-Kähler metrics for which the self-dual part of a derivative of a Killing vector is null correspond to solutions of the dispersionless Kadomtsev–Petviashvili equation (2.9). The corresponding EW structures are characterised by the existence of a constant weighted vector. The minitwistor spaces are such that the line bundle $`\kappa ^{1/4}`$ admits a section, where $`\kappa `$ is the canonical line bundle. The above statements have been proved in this paper.
* Hyper-Kähler metrics with conformal Killing vectors for which the self-dual part of a derivative of a conformal Killing vector is null.
The last possibility has not yet been investigated. The EW spaces will be given by a generalisation of the dKP equation. We intend to study this generalisation, and the corresponding EW geometries in a subsequent paper.
## 7 Outlook: a twistor theory for the full KP equation?
A combination of the dispersive limit of dKP with the twistor picture suggests a candidate for a twistor space for the full KP equation (2.6) (cf the similar proposal in ).
Let $`x`$ be a coordinate on a configuration space $`Q`$, and let $`\stackrel{~}{\lambda }`$ be the corresponding momentum. The extended six-dimensional phase-space $`T^{}(Q\times ^2)`$ is coordinatised by $`x^i=(x,y,t),p_i=(\stackrel{~}{\lambda },H_2,H_3)`$. Restrict the symplectic form $`\mathrm{\Pi }`$ on $`T^{}(Q\times ^2)`$ to the four-dimensional correspondence space $`^4`$ obtained by putting $`H_r:=H_r(x^i,\stackrel{~}{\lambda })`$, $`r=2,3`$. The (complexified) space $`^4`$ is foliated by sub-manifolds whose tangent vectors annihilate the symplectic form, which gives rise to a projection $`p:𝒵`$ such that $`\mathrm{\Pi }`$ descends to a symplectic form on $`𝒵`$. The two-dimensional complex manifold $`𝒵`$ is the mini-twistor space for the extended configuration space $`Q\times ^2`$ with its dKP Einstein–Weyl structure. It is believed that the Moyal quantisation of $`T^{}(Q\times ^2)`$ gives rise to the full KP equation. This suggests the conjecture that there exists a correspondence between solutions to the full KP equation and the Moyal deformations of $`𝒵`$.
It will be instructive to compare this approach to the twistor constructions for the full KP equations described in , and §12.6 of .
## 8 Acknowledgements
We thank David Calderbank for valuable comments which resulted in many improvements. Maciej Dunajski would like to thank Centro de Investigacion y de Estudios Avanzados in Mexico, where part of this work was done, for its financial support (3697 E, Proyecto de CONACYT). MD is also grateful to Bogdan Mielnik, Maciej Przanowski and Jerzy Plebański for their warm hospitality. LJM would like to acknowledge support from NATO collaborative Research Grant number CRG 950300.
Some results in sections 3 and 5 were obtained during the workshop Spaces of geodesics and complex methods in general relativity and geometry held in the summer of 1999 at the Erwin Schrödinger Institute in Vienna. We wish to thank ESI for the hospitality and for financial assistance.
## 9 Appendix
Here we shall demonstrate (by an explicit calculation) that the dKP equation (2.9) is a reduction of the second heavenly equation by a Killing vector with a null self-dual derivative.
Let $`\mathrm{\Theta }(z,t,q,y)`$ satisfy .
$$\mathrm{\Theta }_{zy}\mathrm{\Theta }_{tq}+\mathrm{\Theta }_{qq}\mathrm{\Theta }_{yy}\mathrm{\Theta }_{qy}^2=0.$$
(9.53)
Then
$$g=2(\mathrm{d}z\mathrm{d}y+\mathrm{d}q\mathrm{d}t\mathrm{\Theta }_{qq}\mathrm{d}z^2\mathrm{\Theta }_{yy}\mathrm{d}t^2+2\mathrm{\Theta }_{yq}\mathrm{d}z\mathrm{d}t)$$
(9.54)
is a hyper-Kähler metric. All hyper-Kähler metrics can locally be put in the form (9.54).
Let $`K`$ be a Killing vector such that $`\mathrm{d}K_+\mathrm{d}K_+=0`$. There is no loss of generality in choosing $`K=_z2z_q`$, in which case $`\mathrm{d}K_+=2\mathrm{d}t\mathrm{d}z`$.
The Killing equations yield $`(_K\mathrm{\Theta })_{yy}=(_K\mathrm{\Theta })_{qq}=0,(_K\mathrm{\Theta })_{yq}=1`$. They integrate to
$$\mathrm{\Theta }=zqy+yA(z,t)+qB(z,t)+C(z,t)+G(y,t,q+z^2).$$
(9.55)
The function $`C`$ is pure gauge and can be set to zero without loss of generality. Imposing (9.53) gives two equations: the first is $`A_z+B_t=2z^2`$, and we can deduce, without loss of generality, that $`A=z^3,B=z^2t`$, and the second is
$$uG_{tu}+G_{yy}G_{uu}G_{yu}^2=0,\text{where}u=(q+z^2).$$
(9.56)
The last equation is equivalent to the dKP equation. To see this write (9.56) as a closed system
$`\mathrm{d}G`$ $`=`$ $`G_u\mathrm{d}u+G_t\mathrm{d}t+G_y\mathrm{d}y,`$
$`0`$ $`=`$ $`u\mathrm{d}y\mathrm{d}t\mathrm{d}u+\mathrm{d}G_u\mathrm{d}y\mathrm{d}u\mathrm{d}G_y\mathrm{d}G_u\mathrm{d}t.`$ (9.57)
Now rewrite the first equation as $`\mathrm{d}(GuG_u)=G_t\mathrm{d}t+G_y\mathrm{d}yu\mathrm{d}G_u,`$ and perform a Legendre transform
$$x:=G_u,u=u(t,y,x),H(t,y,x):=G(t,y,u(t,y,x))+xu(t,y,x).$$
The relation $`\mathrm{d}H=H_t\mathrm{d}t+H_x\mathrm{d}x+H_y\mathrm{d}y`$ implies $`H_t=G_t,H_y=G_y,H_x=u`$. Equation (9.57) yields
$$H_x\mathrm{d}y\mathrm{d}t\mathrm{d}H_x+\mathrm{d}x\mathrm{d}y\mathrm{d}H_x+\mathrm{d}H_y\mathrm{d}x\mathrm{d}t=0,$$
which is equivalent to
$$H_xH_{xx}H_{xt}+H_{yy}=0.$$
(9.58)
Taking the $`x`$ derivative of the above equation and using $`H_x=u`$ yields
$$u_{xt}uu_{xx}u_x^2=u_{yy}$$
which is the dKP equation. To calculate the metric differentiate the relation $`x=G_u`$ with respect to $`x`$ and $`H_y=G_y`$ with respect to $`y`$,
$$1=G_{uu}u_x,0=G_{uy}+G_{uu}u_y,0=G_{ut}+G_{uu}u_t,G_{yy}=\frac{u_{y}^{}{}_{}{}^{2}}{u_x}+uu_xu_t$$
(we also used (9.58)). Therefore (from (9.55)) we have
$$\mathrm{\Theta }_{yy}=\frac{u_{y}^{}{}_{}{}^{2}}{u_x}+uu_xu_t,\mathrm{\Theta }_{yq}=\frac{u_y}{u_x}+z,\mathrm{\Theta }_{qq}=\frac{1}{u_x}.$$
The metric (9.54) in terms of $`u(x,y,t)`$ is
$`g`$ $`=`$ $`2(u_x\mathrm{d}x\mathrm{d}t+\mathrm{d}z\mathrm{d}y+2{\displaystyle \frac{u_y}{u_x}}\mathrm{d}z\mathrm{d}tu_y\mathrm{d}y\mathrm{d}t(uu_x+{\displaystyle \frac{u_y^2}{u_x}})\mathrm{d}t^2{\displaystyle \frac{1}{u_x}}\mathrm{d}z^2)`$
$`=`$ $`{\displaystyle \frac{u_x}{2}}(\mathrm{d}y^24\mathrm{d}x\mathrm{d}t4u\mathrm{d}t^2){\displaystyle \frac{2}{u_x}}(\mathrm{d}z{\displaystyle \frac{u_x\mathrm{d}y}{2}}u_y\mathrm{d}t)^2`$
which is (2.27).
|
warning/0004/math0004068.html
|
ar5iv
|
text
|
# Résultats sur la conjecture de dualité étrange sur le plan projectif
## 1 Introduction
La motivation principale de cet article est de fournir des exemples en faveur de la dualité étrange sur le plan projectif conjecturée par Le Potier (\[LeP7\]). On considère l’algèbre de Grothendieck $`\mathrm{K}(_2)`$ des classes de faisceaux algébriques cohérents sur $`_2`$. C’est un groupe abélien isomorphe à $`^3`$, un isomorphisme étant donné par le rang, la classe de Chern et la caractéristique d’Euler-Poincaré (ceci nous permet de désigner chaque classe $`c\mathrm{K}(_2)`$ par le triplet formé par son rang $`r`$, sa première classe de Chern $`c_1`$ et sa caractéristique d’Euler-Poincaré $`\chi `$, ou bien, lorsque c’est indiqué, sa deuxième classe de Chern $`c_2`$). Elle est munie d’une multiplication et d’une forme bilinéaire donnée par $`c,u=\chi (cu)`$. Pour deux classes $`c`$ et $`u`$ orthogonales, de rang $`r>0`$ et respectivement $`0`$, on note $`\mathrm{M}_c`$ et $`\mathrm{M}_u`$ les espaces de modules des faisceaux semi-stables sur $`_2`$ de classe $`c`$ et respectivement $`u`$. Sur chacun de ces espaces il existe un fibré inversible $`𝒟_u`$ et respectivement $`𝒟_c`$, appelé fibré déterminant. Alors le fibré produit tensoriel externe $`𝒟_u𝒟_c`$ sur $`\mathrm{M}_c\times \mathrm{M}_u`$ a une section canonique $`\sigma _{c,u}`$, qui fournit une application linéaire
$$\mathrm{D}_{c,u}:\mathrm{H}^0(\mathrm{M}_u,𝒟_c)^{}\mathrm{H}^0(\mathrm{M}_c,𝒟_u)$$
appelée morphisme de dualité étrange. Remarquons que le groupe $`\mathrm{SL}(3)`$ agit sur $`_2`$. Il agit ainsi sur les espaces de modules $`\mathrm{M}_c,\mathrm{M}_u`$, et sur les fibrés déterminants $`𝒟_c,𝒟_u`$. Le morphisme $`\mathrm{D}_{c,u}`$ est un morphisme de $`\mathrm{SL}(3)`$-représentations.
Conjecture (J. Le Potier) Si $`\mathrm{M}_c`$ est non-vide alors le morphisme de dualité étrange est un isomorphisme. On consacre la section 2 à la construction et l’interprétation géométrique du morphisme $`\mathrm{D}_{c,u}`$. On va se restreindre ici au cas $`c=(2,0,2n)`$ (qui recouvre le cas où la première classe de Chern est paire), et $`u=d(0,1,0)`$. Le cas $`d=1`$ et $`n19`$ a été analysé dans l’article \[D\]. Le résultat principal est:
###### Théorème 1.1
Soit $`𝔲=(1,0,0)\mathrm{K}(_2)`$. Si $`r=2,c_1=0`$ et $`n=c_25`$ (i.e. $`c=(2,0,2n)`$) et $`u=(0,d,0)=d𝔲`$, alors l’application linéaire
$$D_{c,d𝔲}:\mathrm{H}^0(\mathrm{M}_{d𝔲},𝒟_c)^{}\mathrm{H}^0(\mathrm{M}_c,𝒟_𝔲^d)$$
est un isomorphisme pour $`d=2,3`$, c’est-à-dire que dans ces conditions la conjecture de dualité étrange est vraie.
Au paragraphe 4 on utilise \[LeP3\] pour décrire les espaces de modules $`\mathrm{M}_{d𝔲}`$. Il existe un morphisme $`\pi :\mathrm{M}_{d𝔲}C_d`$ (espaces des courbes de degré $`d`$ dans $`_2`$) qui associe au faisceau $`G`$ l’équation de son support schématique. C’est un isomorphisme pour $`d=1,2`$ et un morphisme dont la fibre générique est de dimension $`1`$ pour $`d=3`$. Ceci nous permet de calculer $`\mathrm{H}^0(\mathrm{M}_{d𝔲},𝒟_c)`$.
Au paragraphe 5 on démontre l’injectivité du morphisme $`\mathrm{D}_{c,u}`$ en utilisant l’interprétation géométrique du théorème 2.1 (iv). On utilise les propriétés de $`\mathrm{M}_{d𝔲}`$ établies dans la section 4.
On calcule au paragraphe 6 les espaces $`\mathrm{H}^0(\mathrm{M}_c,𝒟_𝔲^2)`$ et $`\mathrm{H}^0(\mathrm{M}_c,𝒟_𝔲^3)`$ en tant que $`\mathrm{SL}(3)`$-représentations, selon la technique développée dans l’article \[D\].
###### Proposition 1.2
Avec les notations du théorème précédent, le $`\mathrm{SL}(3)`$-module $`\mathrm{H}^0(\mathrm{M}_c,𝒟_𝔲^2)`$ est isomorphe à $`\mathrm{S}^n(\mathrm{S}^2E)`$ et le $`\mathrm{SL}(3)`$-module $`\mathrm{H}^0(\mathrm{M}_c,𝒟_𝔲^3)`$ est isomorphe à $`\mathrm{S}^n(\mathrm{S}^3E)\mathrm{S}^{n2}(\mathrm{S}^3E)`$(où $`E=\mathrm{H}^0(_2,𝒪(1))`$ est la représentation standard de $`\mathrm{SL}(3)`$ ).
Ceci nous permet de conclure la preuve du théorème. De plus, nous calculons au paragraphe 7 la dimension des espaces de sections de $`𝒟_𝔲^k`$ pour $`n=3,4`$, qu’on écrit sous forme de série de Poincaré :
###### Théorème 1.3
i) Pour l’espace de modules $`\mathrm{M}_{(2,0,1)}`$ des faisceaux stables de rang $`2`$ et classes de Chern $`(0,3)`$, la série de Poincaré de $`𝒟_𝔲`$, $`P(t)=_{k0}t^kh^0(𝒟_𝔲^k)`$ est donnée par
$$P(t)=\frac{1+t^2+t^4}{(1t)^{10}}.$$
ii) Pour l’espace de modules $`\mathrm{M}_{(2,0,2)}`$ des faisceaux semi-stables de rang $`2`$ et classes de Chern $`(0,4)`$, la série de Poincaré de $`𝒟_𝔲`$ est donnée par
$$P(t)=\frac{1+t+7t^2+7t^3+22t^4+7t^5+7t^6+t^7+t^8}{(1t)^{14}}.$$
## 2 Morphisme de dualité étrange
L’objet de cette partie est de présenter la conjecture de Le Potier sur la dualité étrange.
### 2.1 L’algèbre de Grothendieck $`\mathrm{K}(_2)`$
Si $`S`$ est une variété algébrique, on désigne par $`\mathrm{K}(S)`$ le groupe de Grothendieck des classes de faisceaux algébriques cohérents sur $`S`$. Pour un faisceau $`F`$ on note $`[F]`$ sa classe dans le groupe $`\mathrm{K}(S)`$. Dans ce qui suit, on aura à considérer en particulier le groupe de Grothendieck $`K(_2)`$: c’est un groupe abélien libre de rang 3; l’application
$$\begin{array}{ccccc}\varphi & :& \mathrm{K}(_2)& & ^3\\ & & [F]& & (rg(F),c_1(F),\chi (F))\end{array}$$
qui à la classe d’un faisceau $`F`$ associe le rang $`r`$ de $`F`$, la classe de Chern $`c_1`$ de $`F`$, et la caractéristique d’Euler-Poincaré $`\chi `$ de $`F`$, est un isomorphisme de groupes abéliens. On notera un élément de $`\mathrm{K}(_2)`$ par son image $`(r,c_1,\chi )`$ dans $`^3`$. Si $`S`$ est lisse, une loi de multiplication sur $`\mathrm{K}(S)`$ est définie en prolongeant par linéarité la loi de multiplication définie pour $`F`$ et $`G`$ faisceaux algébriques cohérents sur $`S`$ par
$$FG=\underset{p}{}(1)^p\underset{¯}{Tor}_p(F,G)$$
Ce produit se réduit au produit tensoriel usuel si l’un des deux faisceaux $`F`$ ou $`G`$ est localement libre.
On parle alors d’algèbre de Grothendieck.
On note $`\eta =[𝒪_l]`$ la classe du faisceau structural d’une droite $`l`$, et $`\eta ^2=[𝒪_p]`$ celle du faisceau structural d’un point. En tant qu’algèbre, $`\mathrm{K}(_2)`$ est isomorphe à $`[\eta ]/(\eta ^3)`$. On munit $`\mathrm{K}(_2)`$ de la forme bilinéaire donnée par $`c,u=\chi (cu)`$. Dans la suite l’orthogonalité sera prise relativement à cette forme. On a aussi sur $`K(_2)`$ une involution $`uu^{}`$ qui associe à la classe d’un fibré vectoriel celle de son dual.
### 2.2 Espaces de modules et fibrés déterminants
Soit $`S`$ une variété. On note $`pr_1`$ la projection $`S\times _2S`$ et $`pr_2`$ la projection $`S\times _2_2`$. Pour un faisceau $``$ sur $`S\times _2`$, on note
$$pr_{1!}([]):=[R^0pr_1][R^1pr_1]+[R^2pr_1]$$
dans le groupe $`\mathrm{K}(S)`$. Cela définit, par linéarité, une application $`pr_{1!}:\mathrm{K}(S\times _2)\mathrm{K}(S)`$.
Si $`F`$ est un faisceau cohérent sur $`S`$ qui admet une résolution finie par des faisceaux localement libres $`A_i`$ :
$$0A_nA_{n1}\mathrm{}A_0F0$$
on introduit le faisceau inversible
$$detF=detA_0(detA_1)^1\mathrm{}(detA_n)^{(1)^n}.$$
L’application qui à $`F`$ associe son déterminant $`detF`$ est multiplicative sur les suites exactes.
Soient $`c\mathrm{K}(_2)`$ une classe de Grothendieck de rang $`r>0`$ et $`\mathrm{M}_c`$ l’espace de modules des faisceaux semi-stables de classe de Grothendieck $`c`$. C’est une variété algébrique projective irréductible normale, de dimension $`D=1c^{},c`$$`c^{}`$ est la classe duale de $`c`$. On note $`c^{}`$ le sous-espace de $`\mathrm{K}(_2)`$ des classes orthogonales à $`c`$. Dans \[Dréz2\], Drézet a construit un morphisme surjectif de groupes $`\lambda _c:c^{}\mathrm{Pic}(\mathrm{M}_c)`$ caractérisé par la propriété universelle suivante :
Pour toute famille plate $``$ de faisceaux semi-stables de classe de Grothendieck $`c`$, paramétrée par une variété algébrique $`S`$, la classe $`\lambda _{}(u)=detpr_{1!}(pr_2^{}(u))`$ définit un fibré inversible sur la variété $`S`$. Si $`f_{}:S\mathrm{M}_c`$ est le morphisme modulaire associé à la famille $``$, on a
$$f_{}^{}(\lambda _c(u))=\lambda _{}(u)$$
et le fibré $`\lambda _c(u)`$ est le seul à isomorphisme près qui satisfait à cette propriété pour toute famille plate $``$.
S’il existe un faisceau universel $``$ sur $`\mathrm{M}_c`$, d’après la propriété universelle de $`\lambda _c(u)`$, il résulte que $`\lambda _c(u)=\lambda _{}(u)`$. En général, on prouve l’existence de $`\lambda _c(u)`$ en écrivant $`\mathrm{M}_c=\mathrm{\Omega }^{ss}/\mathrm{SL}(H)`$ comme quotient d’un ouvert dans un schéma de Hilbert par l’action d’un groupe réductif, en considérant une famille universelle sur $`\mathrm{\Omega }^{ss}`$ et en utilisant un argument de descente.
On note $`\mathrm{K}_c`$ le sous-$``$-module libre de rang $`1`$ de $`c^{}`$ des classes de rang $`0`$. On appelle $`𝔲`$ le générateur positif de $`\mathrm{K}_c`$ (i.e. $`c_1(𝔲)`$ est un multiple positif de la classe hyperplane $`h`$ dans $`\mathrm{H}^2(_2,)`$). On a $`𝔲=(0,\frac{r^c}{\delta },\frac{c_1^c}{\delta })`$$`\delta =\mathrm{pgcd}(r^c,c_1^c)`$.
Le fibré $`𝒟=𝒟_𝔲=\lambda _c(𝔲)`$ s’appelle fibré déterminant de Donaldson sur $`\mathrm{M}_c`$. Pour toute classe $`u\mathrm{K}_c`$ on introduit plus généralement le fibré inversible $`𝒟_u=\lambda _c(u)`$; c’est donc un multiple du fibré déterminant de Donaldson.
Le problème du calcul de la dimension de l’espace de sections $`\mathrm{H}^0(\mathrm{M}_c,𝒟_u)`$ a conduit Le Potier (\[LeP3\]) à introduire $`\mathrm{M}_u`$, l’espace de modules des faisceaux semi-stables de dimension $`1`$ de classe de Grothendieck $`u`$. La notion de semi-stabilité (resp. stabilité) se généralise (cf.\[LeP3\]) pour les faisceaux algébriques cohérents $`F`$ de dimension $`1`$. On sait que $`\mathrm{M}_u`$ est encore une variété algébrique irréductible normale et que la classe $`c`$ (appartenant à $`u^{}`$) permet de construire un fibré inversible $`𝒟_c=\lambda _u(c)`$ sur $`\mathrm{M}_u`$, de la même manière que sur $`\mathrm{M}_c`$. Dans certains cas l’espace $`\mathrm{M}_u`$ et le fibré $`𝒟_c`$ sont plus faciles à décrire.
### 2.3 Construction du morphisme de dualité étrange
On considère deux classes $`c,u\mathrm{K}(_2)`$ dans l’algèbre de Grothendieck . On suppose qu’elles sont orthogonales, que $`r(c)>0`$, que $`r(u)=0`$ et $`c_1(u)>0`$. Le morphisme de dualité étrange est une conséquence de la construction simultanée des fibrés déterminants $`𝒟_u`$ sur $`\mathrm{M}_c`$ et $`𝒟_c`$ sur $`\mathrm{M}_u`$. Sa construction et son interprétation géométrique sont résumées dans le théorème suivant :
###### Théorème 2.1
i) Il existe une section canonique, définie à une constante près, $`\sigma _{c,u}\mathrm{H}^0(\mathrm{M}_c\times \mathrm{M}_u,𝒟_u𝒟_c)`$ qui s’annule exactement aux points $`([F],[G])`$ tels que $`h^0(_2,FG)=h^1(_2,FG)0`$.
ii) La section $`\sigma _{c,u}`$ définit une application linéaire
$$\mathrm{D}_{c,u}:\mathrm{H}^0(\mathrm{M}_u,𝒟_c)^{}\mathrm{H}^0(\mathrm{M}_c,𝒟_u).$$
iii) On note $`\sigma _F`$ la restriction de $`\sigma _{c,u}`$ à $`\{[F]\}\times \mathrm{M}_u`$. Si $`\sigma _{c,u}`$ n’est pas identiquement nulle, l’association $`F[\sigma _F]`$ définit une application rationnelle
$$\mathrm{\Phi }:\mathrm{M}_c\mathrm{H}^0(\mathrm{M}_u,𝒟_c).$$
Si en outre pour tout $`[F]\mathrm{M}_c`$, $`\sigma _F`$ n’est pas identiquement nulle, l’application $`\mathrm{\Phi }`$ est régulière.
iv) Si l’image du morphisme $`\mathrm{\Phi }`$ n’est pas contenue dans un hyperplan l’application $`\mathrm{D}_{c,u}`$ est injective.
La conjecture de Le Potier est alors :
###### Conjecture 2.2
Si $`\mathrm{M}_c`$ est non-vide alors le morphisme de dualité étrange $`\mathrm{D}_{c,u}`$ est un isomorphisme.
Preuve du théorème 2.1 :
On commence par rappeler les résultats suivants :
###### Lemme 2.3
Soit $`S`$ une variété algébrique. Soient $``$ et $`𝒢`$ des familles plates de faisceaux semi-stables sur $`_2`$ de classes de Grothendieck $`c`$ et respectivement $`u`$, paramétrées par $`S`$. Alors :
a) Le faisceau $`𝒢`$ a une résolution
$$0𝒬𝒢0$$
(1)
sur $`S\times _2`$ par des faisceaux localement libres $`𝒬`$ et $``$.
b) Le faisceau $``$ a une résolution
$$0𝒜0$$
(2)
sur $`S\times _2`$ par des faisceaux localement libres $`𝒜`$ et $``$. En plus, on peut choisir $``$ tel que
$$h^0(_s𝒢_s)=0\mathrm{pour}\mathrm{tout}sS.$$
(3)
c) $`\underset{¯}{\mathrm{Tor}}_i(,𝒢)=0`$ pour $`i>0`$.
Preuve :
La résolution pour $`𝒢`$ résulte du fait que $`𝒢`$ est pur de dimension $`1`$ sur chaque fibre. La résolution pour $``$ résulte du fait que la restriction de $``$ à chaque fibre est un faisceau sans torsion sur $`_2`$. On peut changer le faisceau $``$ dans la résolution de $``$ en $`[pr_2^{}𝒪(n)]^m`$, pour $`n0`$ et $`m`$ assez grand. Pour un choix de $`n`$ assez grand on obtient $`h^0(_s𝒢_s)=0`$ pour tout $`s`$. Il résulte d’après a), b), que $`\underset{¯}{\mathrm{Tor}}_i(,𝒢)=0`$ pour $`i2`$ et que $`\underset{¯}{\mathrm{Tor}}_1(,𝒢)`$ est inclus dans $`𝒢𝒜`$ et dans $`𝒬`$. Comme $`𝒢𝒜`$ est de torsion, et $`𝒬`$ sans torsion, on en déduit que $`\underset{¯}{\mathrm{Tor}}_1(,𝒢)=0`$. $`\mathrm{}`$
Soient $`S`$, $``$, $`𝒢`$ comme dans le lemme. D’après c) on a une suite exacte courte
$$0𝒜𝒢\stackrel{a}{}𝒢𝒢0.$$
(4)
On considère son image directe par $`pr_1`$. Le lemme 2.3 b) conduit à $`pr_1(𝒜𝒢)=pr_1(𝒢)=0`$. Le faisceau $`𝒢`$ est de dimension $`1`$ dans les fibres, donc $`R^2pr_1(𝒜𝒢)=R^2pr_1(𝒢)=0`$. Alors la suite
$$0pr_1(𝒢)R^1pr_1(𝒜𝒢)\stackrel{a}{}R^1pr_1(𝒢)R^1pr_1(𝒢)0$$
(5)
est exacte.
###### Lemme 2.4
Les faisceaux $`R^1pr_1(𝒜𝒢)`$ et $`R^1pr_1(𝒢)`$ sont localement libres de même rang sur $`S`$.
Preuve :
Puisque les familles $``$ et $`𝒢`$ sont $`S`$-plates, on obtient que $`𝒜𝒢`$, $`𝒢`$ et $`𝒢`$ sont des familles $`S`$-plates. Alors les suites (1), (4), (5) sont compatibles avec les changements de base $`S^{}S`$. En particulier, pour $`S^{}=\{s\}S`$, on obtient à partir de la suite (5) une suite exacte :
$$0\mathrm{H}^0(_s𝒢_s)\mathrm{H}^1(𝒜_s𝒢_s)\stackrel{a_s}{}\mathrm{H}^1(_s𝒢_s)\mathrm{H}^1(_s𝒢_s)0$$
(6)
et $`h^0(𝒜_s𝒢_s)=h^0(_s𝒢_s)=h^2(𝒜_s𝒢_s)=h^2(_s𝒢_s)=0.`$ Ces annulations et le fait que les familles $`𝒜𝒢`$ et $`𝒢`$ sont plates sur $`S`$ nous assurent que $`R^1pr_1(𝒜𝒢)`$ est localement libre de rang $`h^1(𝒜_s𝒢_s)=\chi (𝒜_s𝒢_s)`$. Le même argument montre que $`R^1pr_1(𝒢)`$ est localement libre de rang $`\chi (_s𝒢_s)`$. La suite (6) implique
$$\chi (𝒜_s𝒢_s)\chi (_s𝒢_s)=\chi (_s𝒢_s).$$
Le lemme 2.3 c) implique $`[_s𝒢_s]=[_s][𝒢_s]`$ dans $`\mathrm{K}(_2)`$. Comme $`c`$ et $`u`$ sont orthogonales, on a $`\chi (_s𝒢_s)=0`$. Alors les faisceaux $`R^1pr_1(𝒜𝒢)`$ et $`R^1pr_1(𝒢)`$ ont même rang. $`\mathrm{}`$
En utilisant la suite exacte (5) on définit un fibré $`𝒟_S`$ par
$$𝒟_S:=[detpr_{1!}(𝒢)]^{(1)}=detR^1pr_1(𝒢)[detR^1pr_1(𝒜𝒢)]^{(1)}.$$
L’application $`a`$ fournit une section $`\sigma _S`$ de ce fibré inversible sur $`S`$. Ni le fibré $`𝒟_S`$, ni la section $`\sigma _S`$, à une fonction inversible près, ne dépendent de la résolution choisie. La suite exacte (6) montre que la section $`\sigma _S`$ s’annule exactement aux points $`sS`$ où l’application linéaire $`a|_s`$ n’est pas inversible. Ces points sont ceux où $`h^0(_s𝒢_s)=h^1(_s𝒢_s)0`$.
On a vu que la suite (5) était compatible avec les changements de base $`S^{}S`$. Il résulte que le fibré inversible $`𝒟_S`$ et la section $`\sigma _S`$ le sont aussi. Soit $`\varphi :S\mathrm{M}_c\times \mathrm{M}_u`$ le morphisme modulaire associé aux familles $`,𝒢`$.
###### Proposition 2.5
Il existe un fibré inversible $`𝒟_{c,u}`$ sur $`\mathrm{M}_c\times \mathrm{M}_u`$ et une section $`\sigma _{c,u}`$ bien déterminée à une constante multiplicative près, qui vérifie: pour toute variété algébrique $`S`$ et pour toutes familles plates $``$, $`𝒢`$ de faisceaux semi-stables de classes $`c`$ respectivement $`u`$ dans $`\mathrm{K}(_2)`$, paramétrées par $`S`$, on a
$`𝒟_S`$ $`=`$ $`\varphi ^{}(𝒟_{c,u})\mathrm{et}`$
$`\sigma _S`$ $`=`$ $`\varphi ^{}(\sigma _{c,u})\text{ à une fonction inversible près.}`$
Preuve :
La preuve est classique. On commence par les deux lemmes suivants.
###### Lemme 2.6 (\[Simp\], \[LeP1\], \[LeP6\])
Il existe une variété lisse $`\mathrm{\Omega }_c`$, une famille plate $`_c`$ de faisceaux semi-stables de classe $`c`$ sur $`_2`$ paramétrée par $`\mathrm{\Omega }_c`$ et un groupe réductif $`G_c`$ qui agit sur $`\mathrm{\Omega }_c`$ tels que $`\mathrm{M}_c`$ soit un bon quotient de $`\mathrm{\Omega }_c`$ sous l’action de $`G_c`$.
La description de $`\mathrm{\Omega }_c`$ est la suivante. On introduit pour $`m`$ entier assez grand la caractéristique d’Euler-Poincaré $`P(m)`$ de $`c(m)`$, et la somme directe $`B`$ de $`N=P(m)`$ exemplaires du fibré inversible $`𝒪__2(m)`$. On considère le schéma de Hilbert $`\mathrm{Hilb}^c(B)`$ des faisceaux cohérents $`F`$ de classe de Grothendieck $`c`$ quotients de $`B`$. Le groupe $`G_c:=\mathrm{Aut}(B)`$ opère de manière naturelle sur $`\mathrm{Hilb}^c(B)`$. On note $`\mathrm{\Omega }_c:=\mathrm{\Omega }^{ss}`$ l’ouvert des points semi-stables pour l’action de $`G_c`$. Ces points représentent les faisceaux semi-stables $`F`$ pour lesquels le morphisme naturel $`\mathrm{H}^0(B(m))\mathrm{H}^0(F(m))`$ est un isomorphisme. L’ouvert $`\mathrm{\Omega }^{ss}`$ est invariant par l’action du groupe réductif $`\mathrm{Aut}(B)`$. C’est un ouvert lisse et sur $`\mathrm{\Omega }^{ss}\times _2`$ on dispose d’un faisceau quotient universel $``$, lequel est aussi muni d’une action de $`G_c`$. L’ouvert $`\mathrm{\Omega }^{ss}`$ admet pour bon quotient l’espace de modules grossier $`\mathrm{M}_c`$ des faisceaux semi-stables de classe $`c`$.
###### Lemme 2.7 (\[LeP3\])
Il existe une variété lisse $`\mathrm{\Omega }_u`$, une famille plate $`𝒢_u`$ de faisceaux semi-stables de classe $`u`$ sur $`_2`$ paramétrée par $`\mathrm{\Omega }_u`$ et un groupe réductif $`G_u`$ qui agit sur $`\mathrm{\Omega }_u`$, tels que $`\mathrm{M}_u`$ soit un bon quotient de $`\mathrm{\Omega }_u`$ sous l’action de $`G_u`$.
La description de la variété $`\mathrm{M}_u`$ est analogue à celle de la variété $`\mathrm{M}_c`$. Il existe un entier $`m`$ suffisamment grand pour que tout faisceau semi-stable $`G`$ sur $`_2`$ de classe $`u=(0,d,\chi )\mathrm{K}(_2)`$ vérifie: le faisceau $`G(m)`$ est engendré par ses sections globales et $`\mathrm{H}^1(G(m))=0`$. On considère $`H`$ un espace vectoriel de dimension $`n=dm+\chi `$ sur $``$. Soit $`\mathrm{Hilb}^u(B)`$ le schéma de Hilbert-Grothendieck des faisceaux quotients de $`B=H𝒪(m)`$ de classe $`u\mathrm{K}(_2)`$. Le groupe $`G_u:=\mathrm{GL}(H)`$ opère de manière naturelle sur $`\mathrm{Hilb}^u(B)`$. On considère l’ouvert $`\mathrm{\Omega }_u:=\mathrm{\Omega }^{ss}`$ des points semi-stables pour l’action de $`G_u`$: ces points correspondent aux faisceaux quotients de $`B`$ qui sont semi-stables et tels que le morphisme d’évaluation $`H\mathrm{H}^0(F(m))`$ soit un isomorphisme. Cet ouvert est lisse et $`\mathrm{M}_u`$ est le bon quotient de $`\mathrm{\Omega }^{ss}`$ pour l’action du groupe $`G_u`$.
On note $`\rho `$ la projection $`\mathrm{\Omega }_c\times \mathrm{\Omega }_u\mathrm{M}_c\times \mathrm{M}_u`$. On considère la construction précédente de $`𝒟_S`$ dans le cas où $`S:=\mathrm{\Omega }_c\times \mathrm{\Omega }_u,:=pr_{13}^{}(_c),𝒢:=pr_{23}^{}(𝒢_u)`$, où $`pr_{13}:\mathrm{\Omega }_c\times \mathrm{\Omega }_u\times _2\mathrm{\Omega }_c\times _2`$ et $`pr_{23}:\mathrm{\Omega }_c\times \mathrm{\Omega }_u\times _2\mathrm{\Omega }_u\times _2`$ sont les projections. On note $`𝒟_\mathrm{\Omega }`$ le fibré déterminant sur $`\mathrm{\Omega }_c\times \mathrm{\Omega }_u`$ ainsi obtenu et $`\sigma _\mathrm{\Omega }`$ sa section canonique, à une fonction inversible près.
De la compatibilité du fibré $`𝒟_S`$ aux changements de base $`S^{}S`$, il résulte une action du groupe $`G_c\times G_u`$ sur le fibré $`𝒟_\mathrm{\Omega }`$. La section $`\sigma _\mathrm{\Omega }`$ est équivariante. Dans le lemme qui suit on vérifie que la condition de descente est satisfaite pour $`𝒟_\mathrm{\Omega }`$.
###### Lemme 2.8
Soit $`(s_c,s_u)`$ un point de $`\mathrm{\Omega }_c\times \mathrm{\Omega }_u`$ tel que l’orbite $`G_cs_c\times G_us_u`$ soit fermée dans $`\mathrm{\Omega }_c\times \mathrm{\Omega }_u`$. Alors le stabilisateur $`G_{s_c}\times G_{s_u}`$ du point $`(s_c,s_u)`$ agit trivialement sur la fibre $`𝒟_{(s_c,s_u)}`$ du fibré $`𝒟_\mathrm{\Omega }`$ au point $`(s_c,s_u)`$.
Preuve :
Les points $`(s_c,s_u)`$ d’orbite fermée sont les points pour lesquels l’orbite $`G_cs_c`$ est fermée dans $`\mathrm{\Omega }_c`$ et l’orbite $`G_us_u`$ est fermée dans $`\mathrm{\Omega }_u`$. D’après le lemme 4.2 de \[D-N\], le faisceau $`_{s_c}`$ sur $`_2`$ est somme directe de faisceaux stables de même polynôme de Hilbert réduit, et de même pour $`𝒢_{s_u}`$. Écrivons :
$`_{s_c}`$ $`=`$ $`F_1^{m_1}\mathrm{}F_k^{m_k}`$
$`𝒢_{s_u}`$ $`=`$ $`G_1^{n_1}\mathrm{}G_l^{n_l}`$
pour des faisceaux stables $`F_i,G_j`$ différents deux par deux. Le stabilisateur du point $`(s_c,s_u)`$ est
$$G_{s_c}\times G_{s_u}=\mathrm{GL}(m_1)\times \mathrm{}\times \mathrm{GL}(m_k)\times \mathrm{GL}(n_1)\times \mathrm{}\times \mathrm{GL}(n_l).$$
D’après la définition de $`𝒟_S`$ on a
$$𝒟_{(s_c,s_u)}=[det\mathrm{H}^0(_2,_{s_c}𝒢_{s_u})]^1det\mathrm{H}^1(_2,_{s_c}𝒢_{s_u}).$$
On a également
$$\mathrm{H}^q(_2,_{s_c}𝒢_{s_u})=_{i=1}^k_{j=1}^l\mathrm{H}^q(_2,F_i^{m_i}G_j^{n_j})$$
pour $`q=0,1`$. Par conséquent, l’élément $`(g_1,\mathrm{},g_k,h_1,\mathrm{},h_l)`$ appartenant à $`G_{s_c}\times G_{s_u}`$, agit sur $`𝒟_{(s_c,s_u)}`$ par multiplication avec
$$\underset{i=1}{\overset{k}{}}\underset{j=1}{\overset{l}{}}(detg_ideth_j)^{h^0(_2,F_iG_j)+h^1(_2,F_iG_j)}.$$
Les faisceaux $`F_i,G_j`$ ont le même polynôme de Hilbert réduit que $`_{s_c}`$ respectivement $`𝒢_{s_u}`$. Puisque les classes $`c`$ et $`u`$ sont orthogonales, on obtient $`h^0(_2,F_iG_j)=h^1(_2,F_iG_j)`$. Donc le stabilisateur du point $`(s_c,s_u)`$ agit trivialement sur $`𝒟_{(s_c,s_u)}`$.$`\mathrm{}`$
Le lemme de Kempf (lemme de descente, th. 2.3, \[D-N\]) implique l’existence d’un unique fibré inversible $`𝒟_{c,u}`$ sur $`\mathrm{M}_c\times \mathrm{M}_u`$ qui satisfait $`\rho ^{}(𝒟_{c,u})=𝒟_\mathrm{\Omega }`$. Mais $`\mathrm{M}_c\times \mathrm{M}_u`$ est un bon quotient de $`\mathrm{\Omega }_c\times \mathrm{\Omega }_u`$ par l’action du groupe $`G_c\times G_u`$, donc
$$\mathrm{H}^0(\mathrm{M}_c\times \mathrm{M}_u,𝒟_{c,u})=\mathrm{H}^0(\mathrm{\Omega }_c\times \mathrm{\Omega }_u,𝒟_\mathrm{\Omega })^{G_c\times G_u}.$$
On peut choisir une résolution (2) $`G_c`$-équivariante. La suite exacte (5) sera alors $`G_c\times G_u`$-équivariante. Il résulte que la section $`\sigma _\mathrm{\Omega }`$, bien déterminée à fonction inversible près, est $`G_c\times G_u`$-équivariante, et donc bien déterminée à une constante multiplicative près. On déduit l’existence d’une section $`\sigma _{c,u}\mathrm{H}^0(\mathrm{M}_c\times \mathrm{M}_u,𝒟_{c,u})`$, bien déterminée à une constante multiplicative près.
Le fait que le fibré $`𝒟_{c,u}`$ et la section $`\sigma _{c,u}`$ satisfont la propriété d’universalité de l’énoncé résulte d’un argument classique (\[LeP2\], §2.13). Cela termine la preuve de la proposition 2.5.$`\mathrm{}`$
###### Lemme 2.9
Le fibré inversible $`𝒟_{c,u}`$ sur $`\mathrm{M}_c\times \mathrm{M}_u`$ est isomorphe au fibré inversible $`𝒟_u𝒟_c`$.
Preuve :
Prouvons que la restriction $`𝒟_{c,s_u}`$ du fibré $`𝒟_{c,u}`$ à $`\mathrm{M}_c\times \{s_u\}`$ est isomorphe au fibré $`𝒟_u`$ pour tout $`s\mathrm{M}_u`$. Soit $`G`$ un faisceau sur $`_2`$ dans la classe du point $`s_u`$. Soit $`S`$ une variété et $``$ une famille plate de faisceaux semi-stables de classe $`c\mathrm{K}(_2)`$, paramétrée par $`S`$. On note $`\phi :S\mathrm{M}_c`$ le morphisme modulaire associé à $``$ et $`\varphi =\phi \times s_u:S\mathrm{M}_c\times \mathrm{M}_u`$ le morphisme modulaire associé à $`,G`$. D’après la propriété d’universalité on a
$$\phi ^{}𝒟_{c,s_u}=\varphi ^{}𝒟_{c,u}=detpr_{1!}(pr_2^{}(G))^1.$$
Alors il y a un isomorphisme $`\phi ^{}𝒟_{c,s_u}\phi ^{}𝒟_{c,u}`$ pour tout couple $`(S,)`$. On déduit que $`𝒟_{c,s_u}`$ et $`𝒟_u`$ sont isomorphes. Nous prouvons de la même manière que la restriction de $`𝒟_{c,u}`$ à $`\{s_u\}\times \mathrm{M}_u`$ est isomorphe au fibré $`𝒟_c`$ sur $`\mathrm{M}_c`$. Les variétés $`\mathrm{M}_u`$ et $`\mathrm{M}_c`$ sont projectives et intègres. Le lemme suivant s’applique.
###### Lemme 2.10 (\[M-F\], p. 23, \[Milne\], §5, th. 5.1 et cor. 5.2)
Soit $`M`$ une variété algébrique intègre, $`N`$ une variété algébrique projective, et $``$ un faisceau inversible sur $`M\times N`$. On suppose que la classe d’isomorphisme de la restriction de $``$ à la fibre $`\{m\}\times N`$ est la même pour chaque $`mM`$. Alors $``$ s’écrit $`L_1L_2`$ pour deux faisceaux $`L_1\mathrm{Pic}(M)`$, $`L_2\mathrm{Pic}(N)`$.
Du lemme et du calcul des restrictions du fibré $`𝒟_{c,u}`$ aux fibres on trouve $`𝒟_{c,u}𝒟_u𝒟_c.`$ $`\mathrm{}`$
Cela prouve le point (i) du théorème 2.1. Les points (ii) et (iii) sont évidents. Le point (iv) résulte du lemme géométrique évident :
###### Lemme 2.11
Soient $`M`$ et $`N`$ deux variétés projectives, $`𝒟`$ et $``$ des fibrés inversibles sur $`M`$ respectivement $`N`$, et une section $`\sigma \mathrm{H}^0(M\times N,𝒟)`$. On note $`\sigma _m\mathrm{\Gamma }(N,)`$ la restriction de $`\sigma `$ à $`\{m\}\times N`$. On suppose que $`\sigma _m`$ n’est pas identiquement nulle pour tout $`mM`$. Alors :
i) La section $`\sigma `$ produit un morphisme $`D_{M,N}:\mathrm{H}^0(N,)^{}\mathrm{H}^0(M,𝒟)`$.
ii) La section $`\sigma `$ produit une application $`\mathrm{\Phi }:M\mathrm{H}^0(N,)`$ définie par $`m[\sigma _m]`$ qui vérifie $`\mathrm{\Phi }^{}𝒪(1)=𝒟`$.
iii) L’application $`\mathrm{\Phi }`$ induit sur les sections globales une application $`\mathrm{\Phi }^{}:\mathrm{H}^0(N,)^{}\mathrm{H}^0(M,𝒟)`$. Alors $`\mathrm{\Phi }^{}=D_{M,N}`$.
iv) Si l’image du morphisme $`\mathrm{\Phi }`$ n’est pas contenue dans un hyperplan (c’est-à-dire que les $`\sigma _m`$ engendrent $`\mathrm{H}^0(N,)`$), alors $`D_{M,N}`$ est injectif.
On applique le lemme pour $`M=\mathrm{M}_c,N=\mathrm{M}_u,𝒟=𝒟_u,=𝒟_c,\sigma =\sigma _{c,u}`$, lorsque $`\sigma _F`$ n’est pas identiquement nulle pour tout $`[F]\mathrm{M}_c`$. $`\mathrm{}`$
###### Remarques 2.12
1) D’après le point i) du théorème 2.1, le fait que $`\sigma _F`$ n’est pas identiquement nulle équivaut à: il existe $`G\mathrm{M}_u`$ tel que $`h^0(FG)=h^1(FG)=0`$. Dans les situations considérées cela sera vrai pour tout $`F\mathrm{M}_c`$. Le Potier a montré cette affirmation si $`2c_1^c=0\mathrm{mod}r^c`$, en utilisant l’existence d’une droite, ou bien d’une conique qui n’est pas de saut pour un faisceau stable générique $`F`$ de classe $`c`$.
2) Dans la proposition 3.3 de \[LeP5\], pour une classe $`c`$ donnée telle que l’espace $`M_c`$ est non-vide, on montre l’existence de $`uc^{}`$ de dimension $`1`$ telle que $`\sigma _{c,u}0`$. La démonstration repose sur un théorème de Flenner, qui donne le comportement de la semi-stabilité par restriction aux courbes de degré élevé, et la version effective d’un résultat de Faltings.
## 3 Préliminaires: le faisceau canonique sur $`\mathrm{M}_{d𝔲}`$ et sur $`\mathrm{M}_c`$
Le but de cette section est de donner des descriptions explicites pour les faisceaux canoniques sur $`\mathrm{M}_{d𝔲}`$ et sur $`\mathrm{M}_c`$. Ces résultats, et le théorème d’annulation de Kawamata-Viehweg, seront appliqués pour déduire des résultats d’annulation (thm. 3.7, prop. 4.9).
Nous allons étudier $`\mathrm{M}_u=\mathrm{M}_{d𝔲}`$, pour $`u=d𝔲=d(0,1,0)`$ et $`d1`$. On définit $`C_d=\mathrm{H}^0(_2,𝒪(d))`$, l’espace des courbes de degré $`d`$ dans $`_2`$. Pour un faisceau $`G\mathrm{M}_u`$, il existe une présentation
$$0Q\stackrel{a}{}RG0$$
(7)
et $`deta\mathrm{H}^0(_2,detRdetQ^1)=\mathrm{H}^0(_2,𝒪(d))`$ s’appelle l’équation du support schématique de $`G`$. Cela ne dépend pas, à une constante près, du choix de $`Q,R`$. On définit ainsi une application
$$\pi :\mathrm{M}_uC_d.$$
(8)
### 3.1 Le faisceau canonique sur $`\mathrm{M}_{d𝔲}`$
La description de la variété $`\mathrm{M}_{d𝔲}`$ comme quotient d’un ouvert lisse $`\mathrm{\Omega }^{ss}`$ d’un schéma de Hilbert-Grothendieck par l’action d’un groupe réductif $`\mathrm{SL}(H)`$ a été donnée au paragraphe 2.3, lemme 2.7.
###### Proposition 3.1
Le faisceau dualisant de la variété $`\mathrm{M}_u`$, $`\omega =\omega _{\mathrm{M}_u}`$, est inversible et isomorphe à $`\pi ^{}𝒪(3d)`$.
Preuve :
On peut supposer $`d3`$, puisque pour $`d=1,2`$, $`\pi `$ est un isomorphisme (voir prop. 4.4) et l’égalité est vérifiée. Il résulte du théorème de Boutot, \[Bout\], que $`\mathrm{M}_u`$ est une variété à singularités rationnelles. En particulier c’est une variété normale et de Cohen-Macaulay. Soient $`\mathrm{M}_u^s`$ l’ouvert de $`\mathrm{M}_u`$ des classes de faisceaux stables, $`j`$ l’inclusion canonique $`j:\mathrm{M}_u^s\mathrm{M}_u`$ et $`Y=C_{\mathrm{M}_u^s}`$ le complémentaire de $`\mathrm{M}_u^s`$ dans $`\mathrm{M}_u`$. Il est démontré dans \[LeP3\], prop 3.4, que $`\mathrm{codim}Y2`$. Soit $`\underset{¯}{p}Y`$ le point générique d’une sous-variété irréductible de $`Y`$. Puisque $`\mathrm{M}_u`$ est de Cohen-Macaulay on obtient $`\mathrm{prof}(\omega _{\underset{¯}{p}})2`$. Or on a l’énoncé suivant (\[Grot2\]) :
###### Théorème 3.2
Soit $`X`$ un schéma et $`YX`$ un fermé. Soit $`F`$ un faisceau algébrique cohérent sur $`X`$ dont le support est $`X`$. Soit $`n`$ un entier. Les conditions suivantes sont équivalentes :
i) Pour tout point $`xY`$, on a
$$\mathrm{prof}(F_x)n.$$
ii) Pour $`i<n`$
$$_Y^i(F)=0.$$
Cet énoncé entraîne que $`_Y^i(\mathrm{M}_u,\omega )=0`$ pour $`i=0,1`$. De la suite longue de cohomologie à support on déduit que $`\omega =j_{}(j^{}(\omega ))`$. Le même argument appliqué au faisceau inversible $`\pi ^{}𝒪(3d)`$ implique $`\pi ^{}𝒪(3d)=j_{}(j^{}(\pi ^{}𝒪(3d)))`$, donc il suffit de démontrer l’isomorphisme souhaité sur l’ouvert $`\mathrm{M}_u^s`$.
On note $`\mathrm{\Omega }^s`$ la préimage de $`\mathrm{M}_u^s`$ par le morphisme $`\rho :\mathrm{\Omega }^{ss}\mathrm{M}_u`$.
###### Lemme 3.3
L’application $`\rho ^{}:\mathrm{Pic}(\mathrm{M}_u^s)\mathrm{Pic}(\mathrm{\Omega }^s)`$ est injective.
Preuve :
L’action de $`\mathrm{SL}(H)`$ sur $`\mathrm{\Omega }^{ss}`$ se factorise à travers une action propre et libre du groupe $`G=\mathrm{PSL}(H)`$, et $`\mathrm{M}_u^s`$ est le quotient de cette action (cf. \[LeP3\], lemme 2.4). En appliquant le lemme de descente de Kempf (th. 2.3, p. 63, et remarque p. 66, \[D-N\]), on obtient un isomorphisme $`\mathrm{Pic}(\mathrm{M}_u^s)=\mathrm{Pic}^G(\mathrm{\Omega }^s)`$, où $`\mathrm{Pic}^G`$ désigne le groupe des fibrés inversibles munis d’une action de $`G`$. On a la suite exacte (\[LeP4\], §3.3) :
$$0\mathrm{H}^1(G,𝒪^{}(\mathrm{\Omega }^s))\mathrm{Pic}^G(\mathrm{\Omega }^s)\mathrm{Pic}(\mathrm{\Omega }^s)$$
$`\mathrm{H}^1(G,𝒪^{}(\mathrm{\Omega }^s))`$ est l’espace des morphismes croisés $`\varphi :G\times \mathrm{\Omega }^s^{}`$ , c’est-à-dire qui vérifient
$$\varphi (gg^{},x)=\varphi (g,g^{}x)\varphi (g^{},x).$$
Mais $`G=\mathrm{PGL}(H)`$ et les seules fonctions régulières inversibles sur $`\mathrm{GL}(H)`$ sont les caractères de $`\mathrm{GL}(H)`$, donc les seules fonctions régulières inversibles sur $`\mathrm{PGL}(H)`$ sont les constantes. Pour un morphisme croisé $`\varphi :G\times \mathrm{\Omega }^s^{}`$ on a $`\varphi (e,x)=1`$ et la fonction $`\varphi _x:G^{}`$ est régulière inversible. Cela implique que tout morphisme croisé est constant égal à $`1`$, et la conclusion. $`\mathrm{}`$
Par conséquent, il suffit de montrer
$$\rho ^{}\omega =\rho ^{}\pi ^{}𝒪(3d)$$
(9)
dans $`\mathrm{Pic}(\mathrm{\Omega }^s)`$. Il suffit encore de le prouver dans $`\mathrm{Pic}(\mathrm{\Omega }^s)`$, puisque $`\mathrm{Pic}(\mathrm{M}_u)=\mathrm{Pic}(\mathrm{M}_u^s)`$ est sans torsion (cf. th. 3.5, \[LeP3\]). On démontre cette affirmation en appliquant la formule de Riemann-Roch-Grothendieck. Si $`𝐓_{\mathrm{M}_u^s}`$ est le fibré tangent à $`\mathrm{M}_u^s`$ et $`𝒢`$ est la famille universelle de faisceaux stables de dimension $`1`$ paramétrée par $`\mathrm{\Omega }^s`$, on a $`\rho ^{}𝐓_{\mathrm{M}_u^s}=\underset{¯}{\mathrm{Ext}}_{pr_1}^1(𝒢,𝒢).`$ Puisque pour chaque $`s\mathrm{\Omega }^s`$, $`𝒢_s`$ est un faisceau stable, on a $`\mathrm{Hom}(𝒢_s,𝒢_s)=0`$ pour tout $`s`$, donc $`\underset{¯}{\mathrm{Hom}}(𝒢,𝒢)=𝒪`$. Alors $`\underset{¯}{\mathrm{Ext}}_{pr_1}^0(𝒢,𝒢)=pr_1\underset{¯}{\mathrm{Hom}}(𝒢,𝒢)=pr_1𝒪=𝒪_{\mathrm{\Omega }^s}`$. Chaque faisceau $`𝒢_s`$ est de dimension $`1`$, donc on a aussi $`\underset{¯}{\mathrm{Ext}}_{pr_1}^i(𝒢,𝒢)=0`$ pour $`i2`$. Alors
$$det\underset{¯}{\mathrm{Ext}}_{pr_1}^{}(𝒢,𝒢)=[det\underset{¯}{\mathrm{Ext}}_{pr_1}^1(𝒢,𝒢)]^1.$$
On peut calculer $`\rho ^{}\omega =\rho ^{}(det𝐓_{\mathrm{M}_u^s})^1=(det\rho ^{}𝐓_{\mathrm{M}_u^s})^1=det\underset{¯}{\mathrm{Ext}}_{pr_1}^{}(𝒢,𝒢)`$ dans $`\mathrm{Pic}(\mathrm{\Omega }^s)`$ avec la formule de Riemann-Roch-Grothendieck :
$$\rho ^{}\omega =pr_1([(ch𝒢)^{}(ch𝒢)Td(_2)]_3).$$
On a
$$[(ch𝒢)^{}(ch𝒢)Td(_2)]_3=(ch_1𝒢)^2Td_1(_2)=\frac{1}{2}c_1^2(𝒢)c_1(\omega __2).$$
(10)
Calculons $`c_1(𝒢)\mathrm{Pic}(\mathrm{\Omega }^s\times _2)`$. Par le corollaire 4.2 on a $`c_1(𝒢)=\rho ^{}\pi ^{}𝒪(1)𝒪(d)`$. On note $`𝔥=\rho ^{}\pi ^{}𝒪(1)`$ et $`h=pr_2^{}𝒪(1)`$ dans $`\mathrm{Pic}(\mathrm{\Omega }^s\times _2)`$. En notation additive on trouve $`c_1(𝒢)=𝔥+dh`$. On obtient dans (10) :
$$\rho ^{}\omega =pr_1([\frac{1}{2}(𝔥+dh)^2(3h)])=3d𝔥=\rho ^{}\pi ^{}𝒪(3d)$$
par la formule de projection. L’égalité (9) est prouvée, donc la proposition.$`\mathrm{}`$ $`\mathrm{}`$
### 3.2 Le faisceau canonique et le fibré déterminant de Donaldson sur $`\mathrm{M}_c`$
Le long de ce paragraphe $`c`$ désignera une classe générale $`c=(r,c_1,\chi )`$ dans $`\mathrm{K}(_2)`$ telle que $`r>0`$ et $`\mathrm{M}_c`$ soit non-vide. Le fibré $`𝒟=𝒟_𝔲`$ sera le fibré déterminant de Donaldson associé à la classe orthogonale à $`c`$, $`𝔲=(0,\frac{r}{\delta },\frac{c_1}{\delta })`$$`\delta =\mathrm{pgcd}(\mathrm{r},\mathrm{c}_1)`$.
###### Proposition 3.4
Le faisceau dualisant $`\omega _{\mathrm{M}_c}`$ est inversible et on a $`\omega _{\mathrm{M}_c}𝒟^{3\delta }`$ dans $`\mathrm{Pic}(\mathrm{M}_c)`$.
Preuve :
Soit $`\mathrm{M}_c^s`$ l’ouvert des classes représentant des faisceaux stables. Il y a deux cas à considérer: quand la codimension du complémentaire $`C_{\mathrm{M}_c^s}`$ de $`\mathrm{M}_c^s`$ dans $`\mathrm{M}_c`$ est $`2`$ et quand ce fermé est une hypersurface.
Dans le premier cas la démonstration est identique à celle de la proposition 3.1. On se ramène à prouver l’isomorphisme sur $`\mathrm{M}_c^s`$. Comme le groupe $`\mathrm{Pic}(\mathrm{M}_c)=\mathrm{Pic}(\mathrm{M}_c^s)`$ est sans torsion, il suffit de prouver l’isomorphisme dans $`\mathrm{Pic}(\mathrm{\Omega }^s)`$, où $`\mathrm{\Omega }^s`$ est l’image réciproque de $`\mathrm{M}_c^s`$ dans $`\mathrm{\Omega }^{ss}`$. Les calculs pour les premières classes de Chern des fibrés $`\rho ^{}(\omega _{\mathrm{M}_c})`$ et $`\rho ^{}(𝒟^{3\delta })`$ dans $`\mathrm{Pic}(\mathrm{\Omega }^s)`$ ont été faits par O’Grady (\[O’Gra\]) en utilisant la formule de Riemann-Roch-Grothendieck. Puisque ces classes coïncident, le résultat découle.
Une analyse du second cas a été faite par Drézet (\[Dréz1\]). Il est prouvé que : la classe $`c`$ est divisible par $`2`$, l’espace de modules $`\mathrm{M}_{\frac{c}{2}}`$ s’identifie à $`_2`$, le fibré déterminant $`𝒟_𝔲`$ sur $`\mathrm{M}_{\frac{c}{2}}`$ s’identifie à $`𝒪__2(1)`$ et l’espace de modules $`\mathrm{M}_c`$ s’identifie à $`_5`$. L’hypersurface des points strictement semi-stables est l’image du morphisme
$$\mathrm{Sym}^2(\mathrm{M}_{\frac{c}{2}})\mathrm{M}_c$$
qui associe aux classes $`[E],[F]`$ la classe $`[E][F]`$ dans $`\mathrm{M}_c`$. Lorsque la classe $`[F]`$ est fixée, le morphisme
$$\varphi _F:\mathrm{M}_{\frac{c}{2}}=_2\mathrm{M}_c=_5$$
qui associe à $`[E]`$ la classe $`[E][F]`$ est linéaire.
Cela suffit pour terminer la preuve de la proposition. En effet, le fibré canonique sur $`_5`$ est $`\omega __5=𝒪__5(6)`$ et il suffit de prouver que $`𝒟_𝔲`$ s’identifie à $`𝒪__5(1)`$. On prouve facilement que $`\varphi _F^{}(𝒟_𝔲)=𝒟_𝔲`$. Puisque $`\varphi _F`$ est linéaire on a aussi $`\varphi _F^{}(𝒪__5(1))=𝒪__2(1)`$. On avait vu que l’isomorphisme $`𝒟_𝔲𝒪__2(1)`$ était satisfait sur $`\mathrm{M}_{\frac{c}{2}}=_2`$. Puisque l’application $`\varphi _F^{}`$ est bijective on obtient que $`𝒟_𝔲=𝒪(1)`$ dans $`\mathrm{Pic}(\mathrm{M}_c)`$. $`\mathrm{}`$
###### Remarque 3.5
Le calcul du faisceau dualisant sur $`\mathrm{M}_c`$ a déjà été fait par Drézet à partir des monades, sans qu’il ait reconnu le rôle du fibré déterminant de Donaldson (\[Dréz2\], th. F), et par O’Grady (\[O’Gra\]), sur l’ouvert des points stables. Le calcul fait dans la proposition 3.1 pour le faisceau dualisant de $`\mathrm{M}_{d𝔲}`$ est calqué sur ce dernier.
###### Proposition 3.6
Le fibré $`𝒟`$ est nef et big.
Preuve:
D’après (\[LeP2\], \[Li\]) il existe un entier $`k`$ satisfaisant aux conditions suivantes:
-le fibré $`𝒟^k`$ est engendré par ses sections;
-considérons le morphisme associé
$$\varphi _k:\mathrm{M}_c_{}\mathrm{H}^0(\mathrm{M}_c,𝒟^k)$$
dans l’espace projectif des hyperplans de $`\mathrm{H}^0(\mathrm{M}_c,𝒟^k)`$. La restriction de $`\varphi _k`$ à l’ouvert des fibrés $`\mu `$-stables est à fibres finies.
De la première condition on déduit que $`𝒟`$ est nef, de la seconde que le nombre $`_{\mathrm{M}_c}c_1(𝒟)^{dim\mathrm{M}_c}`$ est strictement positif, c’est-à-dire que $`𝒟`$ est big.$`\mathrm{}`$
Par un théorème de Boutot (\[Bout\]), l’espace de modules $`\mathrm{M}_c`$ est à singularités rationnelles. Par ailleurs le théorème de Kawamata-Viehweg est valable sur les variétés à singularités rationnelles (\[E-V\]). Nous obtenons d’après les propositions 3.4 et 3.6 le théorème :
###### Théorème 3.7
Pour $`q>0`$ et $`k>3\delta `$, on a $`\mathrm{H}^q(\mathrm{M}_c,𝒟^k)=0`$.
## 4 Les espaces de modules $`\mathrm{M}_{d𝔲}`$
On considère la classe d’un point $`\eta ^2=(0,0,1)\mathrm{K}(_2)`$. C’est une classe orthogonale à $`u=d𝔲`$. Dans la suite, nous allons étudier les fibrés $`𝒟_c`$ sur $`\mathrm{M}_u`$, pour $`c,u=0`$. On peut écrire $`c`$ sous la forme $`r(c)[𝒪]n\eta ^2\mathrm{K}(_2)`$. Par additivité, il suffit d’étudier $`𝒟_{\eta ^2}`$ et $`𝒟_𝒪`$.
###### Proposition 4.1
On a $`\pi ^{}𝒪(1)=𝒟_{\eta ^2}^1`$.
Preuve :
On note $`\mathrm{\Xi }`$ l’hypersurface universelle dans $`C_d\times _2`$ paramétrée par $`C_d`$. La proposition 2.8 de \[LeP4\] affirme que $`𝒟_{\eta ^2}^1=\pi ^{}(\lambda _{𝒪_\mathrm{\Xi }}(\eta ^2))`$
$$\lambda _{𝒪_\mathrm{\Xi }}(\eta ^2)=detpr_{1!}(𝒪_\mathrm{\Xi }pr_2^{}(\eta ^2)).$$
En partant de la résolution de $`𝒪_\mathrm{\Xi }`$ sur $`C_d\times _2`$ :
$$0𝒪(1,d)𝒪𝒪_\mathrm{\Xi }0$$
on obtient $`\lambda _{𝒪_\mathrm{\Xi }}(\eta ^2)=𝒪(1)`$ sur $`C_d`$. $`\mathrm{}`$
Soit $`𝒢`$ une famille de faisceaux $`G`$ de dimension $`1`$ paramétrée par une variété algébrique intègre $`S`$, et
$$\varphi :S\mathrm{M}_{d𝔲}$$
le morphisme modulaire associé.
###### Corollaire 4.2
On a $`det𝒢=(\varphi \pi )^{}(𝒪(1))𝒪(d)`$ sur $`S\times _2`$.
Preuve du corollaire :
On note $`(det𝒢)_s`$ la restriction de $`det𝒢`$ à $`\{s\}\times _2`$ et $`(det𝒢)_x`$ la restriction de $`det𝒢`$ à $`S\times \{x\}`$. D’après le lemme 2.10 il suffit de prouver que $`(det𝒢)_s=𝒪(d)`$ sur $`_2`$ et que $`(det𝒢)_x=(\varphi \pi )^{}(𝒪(1))`$ sur $`S`$. On a évidemment $`(det𝒢)_s=𝒪(d)`$ car $`𝒢`$ est une famille de faisceaux de classe $`c_1=d`$ sur $`_2`$ paramétrée par $`S`$. Considérons la résolution (1) pour $`𝒢`$. On note $`𝒬_x`$, $`_x`$, $`𝒢_x`$ les restrictions de $`𝒬`$, $``$, et respectivement $`𝒢`$ à $`S\times \{x\}`$. On a une suite exacte :
$$0\underset{¯}{\mathrm{Tor}}_1^{S\times _2}(𝒢,pr_2^{}𝒪_x)𝒬_x_x𝒢_x0.$$
Chacun de ces faisceaux a un support inclus dans $`S\times \{x\}`$, donc leur image directe supérieure $`R^ipr_1`$ est nulle pour $`i>0`$. Il résulte que
$$pr_{1!}(𝒢pr_2^{}𝒪_x)=[pr_1(𝒢_x)][pr_1(\underset{¯}{\mathrm{Tor}}_1^{S\times _2}(𝒢,pr_2^{}𝒪_x))]=[_x][𝒬_x]$$
où on a identifié $`𝒬_x`$ avec $`pr_1(𝒬_x)`$ et $`_x`$ avec $`pr_1(_x)`$. Par la définition de $`𝒟_{\eta ^2}^1`$ on trouve :
$$\varphi ^{}𝒟_{\eta ^2}^1=\varphi ^{}𝒟_{[𝒪_x]}^1=(det𝒬_x)^1det_x.$$
(11)
D’après la proposition 4.1 il résulte que
$`(det𝒢)_s`$ $`=`$ $`(det𝒬_x)^1det_x`$
$`=`$ $`\varphi ^{}𝒟_{\eta ^2}^1`$
$`=`$ $`(\varphi \pi )^{}(𝒪(1)).\mathrm{}`$
On définit le fibré inversible $`\mathrm{\Theta }`$ sur $`\mathrm{M}_u`$ comme $`\mathrm{\Theta }=𝒟_𝒪=detpr_{1!}(u)^1`$. Tout ce qu’on utilisera dans la suite est contenu dans la proposition suivante, extraite de \[LeP3\], chap.2 :
###### Proposition 4.3 (\[LeP3\])
Le fibré $`\mathrm{\Theta }`$ a une section canonique $`\theta `$, unique à constante près, non identiquement nulle, qui s’annule aux points $`G`$ tels que $`h^0(_2,G)=h^1(_2,G)0`$.
### 4.1 Les cas $`d=1,2`$
###### Proposition 4.4 (\[LeP3\])
Pour $`d=1,2`$, le morphisme $`\pi `$ est un isomorphisme, et le fibré $`\mathrm{\Theta }`$ est trivial. En conclusion, les espaces $`\mathrm{H}^0(\mathrm{M}_{d𝔲},\mathrm{\Theta }^r(n))`$ et $`\mathrm{H}^0(C_d,𝒪(n))`$ s’identifient.
Preuve :
Le fait que $`\pi `$ est un isomorphisme dans ce cas est démontré dans \[LeP3\]. Plus précisément, l’inverse de $`\pi `$ est donné de la manière suivante: pour $`d=1`$, à une droite $`l`$ on associe le faisceau $`𝒪_l(1)`$, pour $`d=2`$ et pour une conique $`C`$ lisse, à $`C`$ on associe le faisceau $`𝒪_C(a)𝒪__1(1)`$ ($`a`$ point de $`C`$), pour $`d=2`$ et $`C`$ décomposable en deux droites $`l_1`$ et $`l_2`$, à $`C`$ on associe le faisceau $`𝒪_{l_1}(1)𝒪_{l_2}(1)`$. Pour chacun de ces faisceaux on a $`h^0=h^1=0`$. Il résulte que la section $`\theta `$ est partout non nulle sur $`\mathrm{M}_{d𝔲}`$. Donc le fibré $`\mathrm{\Theta }`$ est trivial. La conclusion résulte de la proposition 4.1, du fait que $`\pi `$ est un isomorphisme et de la trivialité du fibré $`\mathrm{\Theta }`$. $`\mathrm{}`$
On déduit en corollaire :
###### Corollaire 4.5
Pour $`c=2n\eta ^2`$ on a
$$\mathrm{H}^0(\mathrm{M}_𝔲,𝒟_c)=\mathrm{H}^0(_2^{},𝒪__2^{}(n))=\mathrm{S}^nE^{}$$
et
$$\mathrm{H}^0(\mathrm{M}_{2𝔲},𝒟_c)=\mathrm{H}^0(_5,𝒪__5(n))=\mathrm{S}^n(\mathrm{S}^2E^{}).$$
On rappelle que $`E=\mathrm{H}^0(_2,𝒪(1))`$.
### 4.2 Le cas $`d=3`$
Soit $`𝔲`$ la classe $`(0,1,0)\mathrm{K}(_2)`$ et $`c=(2,0,c_2=n)`$. L’objectif de ce paragraphe est de démontrer la :
###### Proposition 4.6
En tant que $`\mathrm{SL}(3)`$-représentation l’espace $`\mathrm{H}^0(\mathrm{M}_{3𝔲},𝒟_c)`$ s’identifie à $`\mathrm{S}^n(\mathrm{S}^3E^{})\mathrm{S}^{n2}(\mathrm{S}^3E^{})`$ ( où $`E=\mathrm{H}^0(_2,𝒪(1))`$).
Notre référence principale sera l’article \[LeP3\]. On considère la cubique universelle $`𝒞C_3\times _2`$. La projection $`𝒞C_3`$ induit par la propriété universelle de l’espace de modules $`\mathrm{M}_{3𝔲}`$ un morphisme $`s:C_3\mathrm{M}_{3𝔲}`$ qui associe à une cubique $`C`$ son faisceau structural $`𝒪_C`$. Le morphisme $`s`$ est une section de $`\pi `$ : la résolution
$$0𝒪(3)\stackrel{\underset{¯}{C}}{}𝒪𝒪_C0$$
démontre que $`(\pi s)(C)=C`$. Ici $`\underset{¯}{C}`$ désigne l’équation de la cubique $`C`$.
Prenons $`D_2\times _2^{}`$ la variété d’incidence et $`p`$ et $`q`$ les projections :
L’application $`Gq_{}(p^{}(G(2)))(1)`$ définit (cf. \[LeP3\]) un morphisme $`\varphi :\mathrm{M}_{3𝔲}\mathrm{M}_{(3,0,0)}^{}`$ dans l’espace de modules de faisceaux semi-stables de classe $`(r=3,c_1=0,\chi =0)`$ sur $`_2^{}`$. On note encore $`𝒟=𝒟_u`$ le fibré déterminant sur $`\mathrm{M}_{(3,0,0)}^{}`$, associé à la classe $`𝔲=(0,1,0)`$. La proposition suivante est prouvée dans \[LeP3\] :
###### Proposition 4.7
i) Le diviseur des zéros $`\mathrm{div}\theta `$ de la section $`\theta `$ coïncide avec l’image du morphisme $`s`$.
ii) Le morphisme $`\varphi `$ est l’éclatement d’un point lisse et la fibre exceptionnelle est $`\mathrm{div}\theta `$.
On obtient que les morphismes $`\pi ,s`$ établissent un isomorphisme entre $`\mathrm{div}\theta `$ et $`C_3`$. On identifiera $`\mathrm{div}\theta `$ et $`C_3`$ dans la suite. Puisque $`\mathrm{div}\theta `$ est le diviseur exceptionnel, son fibré conormal est $`𝒪_{C_3}(1)`$. De la suite exacte courte
$$0\mathrm{\Theta }^1\stackrel{\theta }{}𝒪𝒪_{\mathrm{div}\theta }0$$
(12)
il résulte que
$$\mathrm{\Theta }^1|_{\mathrm{div}\theta }𝒪_{C_3}(1).$$
(13)
###### Proposition 4.8
On a $`\mathrm{\Theta }(1)=\varphi ^{}𝒟`$ dans $`\mathrm{Pic}(\mathrm{M}_{3𝔲})`$.
Preuve :
Compte-tenu du fait que $`\mathrm{Pic}(\mathrm{M}_{3𝔲})`$ est sans torsion (cf. th. 3.5, \[LeP3\]) il suffit de démontrer que
$$\varphi ^{}𝒟^9=\mathrm{\Theta }^9(9).$$
Par la proposition 3.4, on a $`𝒟^9=\omega _{\mathrm{M}_{(3,0,0)}^{}}`$, et par la proposition 3.1, on a $`𝒪(9)=\omega _{\mathrm{M}_{3𝔲}}`$. L’égalité à démontrer devient
$$\varphi ^{}\omega _{\mathrm{M}_{(3,0,0)}^{}}=\omega _{\mathrm{M}_{3𝔲}}\mathrm{\Theta }^9.$$
Mais on a vu que $`\mathrm{\Theta }=𝒪(\mathrm{div}\theta )`$ et que $`\mathrm{div}\theta `$ est le diviseur exceptionnel. L’égalité à démontrer est
$$\omega _{\mathrm{M}_{3𝔲}}=\varphi ^{}\omega _{\mathrm{M}_{(3,0,0)}^{}}𝒪(9\mathrm{div}\theta )$$
qui est valable chaque fois qu’on éclate un point lisse dans une variété de dimension $`10`$ (\[Hart\], ex. II 8.5, p. 188).$`\mathrm{}`$
Puisque $`c=2n\eta ^2`$ dans $`\mathrm{K}(_2)`$, on a $`𝒟_c\mathrm{\Theta }^2(n)`$ sur $`\mathrm{M}_{3𝔲}`$. On passe au calcul de l’espace $`\mathrm{H}^0(\mathrm{M}_{3𝔲},\mathrm{\Theta }^2(n))`$.
###### Proposition 4.9
$`\mathrm{H}^q(\mathrm{M}_{3𝔲},\mathrm{\Theta }(n))=0`$ pour $`q1`$ et $`n8`$.
Preuve :
L’espace de modules $`\mathrm{M}_{3𝔲}`$ est le quotient d’une variété lisse par un groupe réductif, donc il est à singularités rationnelles, par un résultat de Boutot \[Bout\]. Le théorème de Kawamata-Viehweg s’applique (\[E-V\]). D’après la proposition 3.6 le fibré $`𝒟`$ est nef et big sur $`\mathrm{M}_{(3,0,0)}^{}`$ et d’après les propositions 4.7 et 4.8 le fibré $`\mathrm{\Theta }(1)`$ est l’image réciproque de $`𝒟`$ par un éclatement. On en déduit que le fibré $`\mathrm{\Theta }(1)`$ est big et nef sur $`\mathrm{M}_{3𝔲}`$. Le fibré $`𝒪(n1)=\pi ^{}(𝒪(n1))`$ est globalement engendré pour $`n1`$ donc $`\mathrm{\Theta }(n)`$ est big et nef pour $`n1`$. La proposition 3.1 fournit le faisceau dualisant sur $`\mathrm{M}_{d𝔲}`$ : $`\omega _{\mathrm{M}_{d𝔲}}=\pi ^{}(𝒪(3d))`$. Alors $`\mathrm{\Theta }(n)\omega _{\mathrm{M}_{d𝔲}}^1=\mathrm{\Theta }(n+9)`$ est big et nef pour $`n8`$. Le résultat en découle.$`\mathrm{}`$
On tensorise la suite (12) par $`\mathrm{\Theta }^2(n)`$. On obtient, après l’identification $`\mathrm{div}\theta =C_3`$, et en utilisant l’isomorphisme (13), la suite exacte courte sur $`\mathrm{M}_{3𝔲}`$ :
$$0\mathrm{\Theta }(n)\mathrm{\Theta }^2(n)𝒪_{C_3}(n2)0.$$
La proposition 4.9 conduit à une suite exacte courte sur les sections globales, pour $`n8`$ :
$$0\mathrm{H}^0(\mathrm{M}_{3𝔲},\mathrm{\Theta }(n))\mathrm{H}^0(\mathrm{M}_{3𝔲},\mathrm{\Theta }^2(n))\mathrm{H}^0(C_3,𝒪_{C_3}(n2))0.$$
(14)
###### Proposition 4.10
Soit $`u=3𝔲`$. Alors les morphismes :
$$\mathrm{H}^0(C_3,𝒪(n))\stackrel{\pi ^{}}{}\mathrm{H}^0(\mathrm{M}_u,𝒪(n))\stackrel{\theta }{}\mathrm{H}^0(\mathrm{M}_u,\mathrm{\Theta }(n))$$
sont des isomorphismes.
Preuve :
On considère l’ouvert $`UC_3`$ des courbes irréductibles. Son complémentaire $`C_U`$ est de codimension $`2`$. La proposition résulte des lemmes suivants :
###### Lemme 4.11
Le complémentaire de l’image réciproque $`\pi ^1(U)`$ de l’ouvert $`U`$ par le morphisme $`\pi `$ est de codimension au moins $`2`$ dans $`\mathrm{M}_u`$. En plus on a l’isomorphisme $`\pi _{}(𝒪_{\pi ^1(U)})=𝒪_U`$.
###### Lemme 4.12
Le morphisme $`\pi _{}𝒪_{\mathrm{M}_u}\stackrel{\theta }{}\pi _{}\mathrm{\Theta }`$ est un isomorphisme sur $`C_3`$.
Effectivement, le lemme 4.11 implique :
$`\mathrm{H}^0(\mathrm{M}_u,𝒪(n))`$ $`=\mathrm{H}^0(\pi ^1(U),𝒪(n))=\mathrm{H}^0(U,\pi _{}(𝒪(n)))`$
$`=\mathrm{H}^0(U,𝒪(n))=\mathrm{H}^0(C_3,𝒪(n)).`$
Il résulte du lemme 4.12 que $`\pi _{}𝒪(n)\stackrel{\theta }{}\pi _{}\mathrm{\Theta }(n)`$ est un isomorphisme. En prenant les sections globales sur $`C_3`$ on obtient que $`\mathrm{H}^0(\mathrm{M}_u,𝒪(n))\stackrel{\theta }{}\mathrm{H}^0(\mathrm{M}_u,\mathrm{\Theta }(n))`$ est un isomorphisme.
Preuve du lemme 4.11 :
Soit $`V\mathrm{M}_u`$ l’ouvert des faisceaux stables et localement libres sur leur support. Le lemme 3.2 et la proposition 3.4 de \[LeP3\] démontrent que $`\mathrm{codim}C_V2`$, où $`C_V`$ désigne le complémentaire de $`V`$. La proposition 2.8 de \[LeP3\] affirme que le morphisme $`V\stackrel{\pi }{}C_3`$ est lisse. Il résulte que ses fibres au-dessus de $`C_U`$ sont de dimension $`dim\mathrm{M}_udimC_3=1`$. L’inclusion $`\pi ^1(C_U)C_V(V\pi ^1(C_U))`$ entraîne $`\mathrm{codim}\pi ^1(C_U)2`$. Le théorème 9 de \[A-I-K\] nous assure que le morphisme projectif $`\pi ^1(U)U`$ est plat et à fibres intègres. D’où $`\pi _{}(𝒪_{\pi ^1(U)})=𝒪_U`$. $`\mathrm{}`$
Preuve du lemme 4.12 :
On déduit de la suite exacte (12) et de l’isomorphisme 13 la suite exacte courte :
$$0𝒪\stackrel{\theta }{}\mathrm{\Theta }𝒪_{C_3}(1)0$$
sur $`\mathrm{M}_{3𝔲}`$. On applique le foncteur $`\pi _{}`$. On obtient la suite exacte :
$$0\pi _{}(𝒪)\pi _{}(\mathrm{\Theta })\stackrel{\theta }{}𝒪_{C_3}(1)\stackrel{\delta }{}R^1\pi _{}(𝒪)$$
(15)
sur $`C_3`$.
Soit $`WC_3`$ l’ouvert des cubiques lisses. La fibre $`P_C`$ du morphisme $`\pi `$ au-dessus de $`CW`$ s’identifie à la jacobienne de $`C`$, $`\mathrm{Jac}(C)`$. La restriction du fibré $`\mathrm{\Theta }`$ à $`P_C`$ s’identifie au fibré $`\mathrm{\Theta }`$ usuel sur $`\mathrm{Jac}(C)`$. Alors le morphisme $`\mathrm{H}^0(P_C,𝒪)\mathrm{H}^0(P_C,\mathrm{\Theta })`$ est un isomorphisme entre des espaces de dimension $`1`$. Puisque la fibration
$$\pi ^1(W)W$$
est plate, par le théorème de semi-continuité, le morphisme $`\pi _{}𝒪\pi _{}(\mathrm{\Theta })`$ est un isomorphisme de fibrés inversibles sur $`W`$.
Il résulte de la suite exacte (15) que le morphisme $`𝒪_{C_3}(1)\stackrel{\delta }{}R^1\pi _{}(𝒪)`$ est injectif sur $`W`$. Comme $`𝒪_{C_3}(1)`$ est un faisceau inversible, le morphisme $`\delta `$ est injectif partout sur $`C_3`$. Par conséquent le morphisme $`\pi _{}𝒪_{\mathrm{M}_u}\stackrel{\theta }{}\pi _{}\mathrm{\Theta }`$ est un isomorphisme partout sur $`C_3`$. $`\mathrm{}`$ $`\mathrm{}`$
D’après la proposition 4.10 et de la suite exacte (14) on obtient le
###### Corollaire 4.13
On a une suite exacte courte
$$0\mathrm{H}^0(C_3,𝒪(n))\stackrel{\theta ^2\pi ^{}}{}\mathrm{H}^0(\mathrm{M}_{3𝔲},\mathrm{\Theta }^2(n))\stackrel{s^{}}{}\mathrm{H}^0(C_3,𝒪(n2))0$$
pour $`n8`$.
D’où la proposition 4.6. $`\mathrm{}`$
## 5 Injectivité du morphisme $`\mathrm{D}_{c,u}`$
On commence par regarder le cas où la classe $`c`$ est de rang $`1`$. On utilise ensuite un argument de récurrence pour étendre le résultat au cas qui nous intéresse, où $`c`$ est de rang $`2`$. Le début de la récurrence utilise le cas $`c=(1,0,c_2=n)`$ étudié en préalable.
### 5.1 Le cas $`c=(1,0,c_2=n),u=d𝔲`$
###### Proposition 5.1
Pour $`d=1,2,3`$ et $`n0`$, l’image du morphisme $`\mathrm{\Phi }:\mathrm{M}_c\mathrm{H}^0(\mathrm{M}_{d𝔲},𝒟_c)`$ n’est pas contenue dans un hyperplan.
On a vu que le morphisme
$$\mathrm{H}^0(C_d,𝒪(n))\stackrel{\pi ^{}}{}\mathrm{H}^0(\mathrm{M}_u,𝒪(n))\stackrel{\theta }{}\mathrm{H}^0(\mathrm{M}_u,\mathrm{\Theta }(n))$$
était bijectif pour $`d=1,2,3`$. Le lemme suivant sera utile :
###### Lemme 5.2
Si $`F=I_Z`$ est l’idéal du sous-schéma $`Z`$ des $`n`$ points distincts $`a_1,\mathrm{},a_n`$ de $`_2`$, si $`[G]\mathrm{M}_{d𝔲}`$, et s’il existe un point $`a_k\mathrm{supp}G`$, alors
$$h^0(FG)=h^1(FG)0.$$
Preuve :
Sans restreindre la généralité, on peut supposer $`a_1,\mathrm{},a_i\mathrm{supp}G`$, $`a_{i+1},\mathrm{},a_n\mathrm{supp}G`$, pour un nombre $`i\{1,\mathrm{},n\}`$. On tensorise par $`G`$ la suite exacte :
$$0FI_{\widehat{Z}}_{j=1}^i𝒪_{a_j}0$$
$`I_{\widehat{Z}}`$ désigne l’idéal du sous-schéma des $`ni`$ points distincts $`a_{i+1},\mathrm{},a_n`$ et $`𝒪_{a_j}`$ le faisceau structural du point $`a_j`$. En utilisant l’annulation $`\underset{¯}{\mathrm{Tor}}_1(I_{\widehat{Z}},G)=0`$ (puisque $`I_{\widehat{Z}}`$ est trivial au voisinage du support de $`G`$), on obtient une inclusion $`0\underset{¯}{\mathrm{Tor}}_1(G,𝒪_{a_i})FG`$. Mais à partir de la résolution de longueur $`1`$ de $`G`$ par des faisceaux localement libres $`A`$ et $`B`$ :
$$0A\stackrel{\alpha }{}BG0$$
on obtient après tensorisation par $`𝒪_{a_i}`$ :
$$0\underset{¯}{\mathrm{Tor}}_1(G,𝒪_{a_i})A|_{a_i}\stackrel{\alpha |_{a_i}}{}B|_{a_i}G𝒪_{a_i}0$$
et $`det\alpha `$ est l’équation du support de $`G`$ donc $`det\alpha |_{a_i}=0`$ et $`\underset{¯}{\mathrm{Tor}}_1(G,𝒪_{a_i})0`$. Le faisceau $`\underset{¯}{\mathrm{Tor}}_1(G,𝒪_{a_i})`$ a pour support le point $`a_i`$, donc $`\mathrm{H}^0(\underset{¯}{\mathrm{Tor}}_1(G,𝒪_{a_i}))0`$, d’où $`\mathrm{H}^0(FG)0`$.$`\mathrm{}`$
On introduit quelques notations. Pour $`E=\mathrm{H}^0(_2,𝒪(1))`$, le point $`a_i`$ est un élément de $`(E^{})`$. Alors $`a_i^d`$ est un élément de $`(\mathrm{S}^dE^{})`$ et il représente, à une constante près, un élément dans $`\mathrm{H}^0(C_d,𝒪(1))`$. C’est l’équation $`H_{a_i}`$ de l’hyperplan des courbes de degré $`d`$ qui passent par le point $`a_i_2`$. Alors $`H_{a_1}\mathrm{}H_{a_n}\mathrm{S}^n\mathrm{H}^0(C_d,𝒪(1))=\mathrm{H}^0(C_d,𝒪(n))`$.
###### Lemme 5.3
Pour $`F=I_Z`$ comme dans le lemme 5.2 on a $`\sigma _F=\sigma _{c,u}(F)=cst\theta \pi ^{}(H_{a_1}\mathrm{}H_{a_n})`$, pour une constante $`cst^{}`$.
Ce lemme suffit pour démontrer la proposition 5.1, puisque $`\{H_{a_i}\}_i`$ engendrent $`\mathrm{H}^0(C_d,𝒪(1))`$ et les produits de $`\{H_{a_i}\}_i`$ engendrent $`\mathrm{S}^n\mathrm{H}^0(C_d,𝒪(1))`$.
Preuve du lemme 5.3 :
Le lemme 5.2 nous dit que $`\sigma _F`$ s’annule sur tous les faisceaux $`G`$ dont le support contient le point $`a_i`$. Ces faisceaux appartiennent à l’ensemble d’équation $`\pi ^{}(H_{a_i})=0`$. La section $`\alpha =\frac{\sigma _F}{_{i=1}^n\pi ^{}(H_{a_i})}`$ est une section rationnelle du fibré $`\mathrm{\Theta }`$ sur $`\mathrm{M}_{d𝔲}`$. Puisque $`\sigma _F`$ s’annule sur $`\pi ^1(\{H_{a_i}=0\})`$, $`\alpha `$ est une section régulière. La proposition 4.10 appliquée pour $`n=0`$ nous assure que $`\mathrm{H}^0(\mathrm{M}_{d𝔲},\mathrm{\Theta })=\mathrm{H}^0(\mathrm{M}_{d𝔲},𝒪)=`$. Donc $`\alpha =cst\theta .`$ $`\mathrm{}`$
On remarque, au passage, que la dualité étrange dans le cas $`r^c=1,d=1,2,3`$ a été prouvée.
###### Proposition 5.4
Le morphisme de dualité étrange est un isomorphisme dans le cas $`c=(1,0,1n)`$, $`u=(0,d,0)`$, pour $`d=1,2,3`$ et pour un entier positif $`n`$.
Preuve :
En effet, le premier membre de la dualité est $`\mathrm{H}^0(\mathrm{M}_{d𝔲},\mathrm{\Theta }(n))^{}`$, isomorphe par la proposition 4.10 à
$$\mathrm{H}^0(C_d,𝒪(n))^{}=\mathrm{S}^n(\mathrm{S}^dE).$$
Pour identifier le second membre on utilise le fait que $`\mathrm{M}_{(1,0,c_2=n)}`$ coïncide avec le schéma de Hilbert $`\mathrm{Hilb}^n(_2)`$ des sous-schémas finis de longueur $`n`$ de $`_2`$.
Soit $`\mathrm{S}^n(_2)`$ le quotient de la puissance $`n`$-ième $`_2^n`$ de $`_2`$ par le groupe symétrique $`𝔖_n`$. On dispose du morphisme de Hilbert-Chow $`HC:\mathrm{Hilb}^n(_2)\mathrm{S}^n(_2)`$ qui associe à un schéma fini $`Z`$ le cycle $`_{x_2}lgZ_xx`$. On note $`𝒪(1,1,\mathrm{},1)^{𝔖_n}`$ le quotient du fibré $`𝒪(1,1,\mathrm{},1)`$ par l’action de $`𝔖_n`$ .
###### Lemme 5.5
Les fibrés inversibles $`𝒟_𝔲`$ et $`HC^{}(𝒪(1,1,\mathrm{},1)^{𝔖_n})`$ sont isomorphes sur $`\mathrm{Hilb}^n(_2)`$.
Preuve :
On considère le diagramme :
Ici, $`C_1=_2^{}`$. Le morphisme $`\mathrm{\Psi }`$ est défini par $`\mathrm{\Psi }(a_1,\mathrm{},a_n)=[H_{a_1}\mathrm{}H_{a_n}]`$. Le lemme 5.3 prouve que
$$\mathrm{\Phi }=(\theta \pi ^{})\mathrm{\Psi }HC$$
sur l’ouvert des points distincts de $`\mathrm{Hilb}^n(_2)`$. Puisque cet ouvert est dense, le diagramme considéré est commutatif.
L’image réciproque du fibré $`𝒪(1)`$ de l’espace projectif $`\mathrm{H}^0(C_1,𝒪(n))`$, par $`\mathrm{\Psi }𝔖_n`$, est le fibré $`𝒪(1,1,\mathrm{},1)`$ sur $`_2^n`$ (on peut le vérifier sur chaque composante). On tient compte de l’isomorphisme (cf. \[LeP4\],§3.4) :
$$\mathrm{Pic}(\mathrm{S}^n(_2))\mathrm{Pic}(_2^n)^{𝔖_n}.$$
On obtient que l’image réciproque du fibré $`𝒪(1)`$ de $`\mathrm{H}^0(C_1,𝒪(n))`$ sur $`\mathrm{S}^n(_2)`$ est $`𝒪(1,1,\mathrm{},1)^{𝔖_n}`$. Par la commutativité du diagramme on obtient la conclusion.$`\mathrm{}`$
En passant aux puissances tensorielles supérieures, on trouve $`𝒟_{𝔲}^{}{}_{}{}^{d}=HC^{}(𝒪(d,d,\mathrm{},d)^{𝔖_n})`$. Cela entraîne que $`\mathrm{H}^0(\mathrm{Hilb}^m(_2)),𝒟_{𝔲}^{}{}_{}{}^{d})=\mathrm{S}^n(\mathrm{S}^dE)`$. L’injectivité de $`\mathrm{D}_{c,u}`$ a été prouvée dans la proposition 5.1.$`\mathrm{}`$
### 5.2 Le cas $`c=(2,0,c_2=n),u=d𝔲`$
###### Proposition 5.6
Pour $`1d3`$ et pour $`n2`$, l’image du morphisme $`\mathrm{\Phi }:\mathrm{M}_c\mathrm{H}^0(\mathrm{M}_{d𝔲},𝒟_c)`$ n’est pas contenue dans un hyperplan.
La proposition résulte des quatre lemmes suivants :
###### Lemme 5.7
La proposition est vraie pour $`n=2`$.
###### Lemme 5.8
L’application
$`\mathrm{H}^0(\mathrm{M}_{d𝔲},\mathrm{\Theta }^2(n))\mathrm{H}^0(C_d,𝒪(1))`$ $``$ $`\mathrm{H}^0(\mathrm{M}_{d𝔲},\mathrm{\Theta }^2(n+1))`$
$`st`$ $``$ $`s\pi ^{}t`$
est surjective.
###### Lemme 5.9
Soit $`n2`$ et $`\mathrm{M}_c^0\mathrm{M}_c`$ l’ouvert des points stables. Si l’image $`\mathrm{\Phi }(\mathrm{M}_c)`$ n’est pas contenue dans un hyperplan, alors $`\mathrm{\Phi }(\mathrm{M}_{c}^{}{}_{}{}^{0})`$ n’est pas contenue dans un hyperplan.
###### Lemme 5.10
Soit $`F\mathrm{M}_{(2,0,c_2=n)}^0`$, $`x_2`$ et $`a:F𝒪_x`$ un morphisme surjectif. Alors $`F^{}=\mathrm{Ker}a`$ est un faisceau semi-stable et on a $`\sigma _F^{}=\sigma _F\pi ^{}H_x`$ par l’application $`id\pi ^{}`$ du lemme 5.8. Ici
$$\sigma _F^{}\mathrm{H}^0(\mathrm{M}_{d𝔲},\mathrm{\Theta }^2(n+1)),\sigma _F\mathrm{H}^0(\mathrm{M}_{d𝔲},\mathrm{\Theta }^2(n))\text{ et }H_x\mathrm{H}^0(C_d,𝒪(1)).$$
Les lemmes 5.7, 5.8, 5.9 et 5.10 fournissent une démonstration par récurrence de la proposition 5.6. Le lemme 5.9 dit que les $`\sigma _F`$, pour $`F`$ stable, engendrent $`\mathrm{H}^0(\mathrm{M}_{d𝔲},\mathrm{\Theta }^2(n))`$. Mais les $`H_x`$ engendrent $`\mathrm{H}^0(C_d,𝒪(1))`$ lorsque $`x`$ varie. Par le lemme 5.8 on obtient que les $`\sigma _F\pi ^{}H_x`$ engendrent $`\mathrm{H}^0(\mathrm{M}_{d𝔲},\mathrm{\Theta }^2(n+1))`$. Le lemme 5.10 assure que de tels éléments sont de la forme $`\sigma _F^{}`$, donc des images par $`\mathrm{\Phi }`$ de $`\mathrm{M}_{(2,0,c_2=n+1)}`$. $`\mathrm{}`$
Preuve du lemme 5.8 :
Pour $`d=1,2`$, le diagramme suivant est commutatif :
Par la proposition 4.4, les morphismes horizontaux sont des isomorphismes. Le lemme résulte de la surjectivité du morphisme vertical gauche.
Pour $`d=3`$, on note $`H=\mathrm{H}^0(C_d,𝒪(1))`$. Le lemme est une conséquence du diagramme analogue, commutatif :
Les suites horizontales sont exactes d’après le corollaire 4.13. Les morphismes verticaux latéraux sont surjectifs, donc aussi le morphisme vertical central. $`\mathrm{}`$
Preuve du lemme 5.9 :
Ceci est évident, puisque $`\mathrm{M}_c^0`$ est un ouvert dense de $`\mathrm{M}_c`$. $`\mathrm{}`$
Preuve du lemme 5.10 :
Pour un sous-faisceau $`F^{\prime \prime }`$ de rang $`1`$ de $`F^{}`$ on a $`c_1(F^{\prime \prime })0`$ puisque $`F^{\prime \prime }`$ est aussi un sous-faisceau de $`F`$, et $`F`$ est stable. Si $`c_1(F^{\prime \prime })=0`$ alors $`\chi (F^{\prime \prime })<\frac{2n}{2}\frac{1}{2}=\frac{2(n+1)}{2}`$. Donc $`F^{}`$ est semi-stable.
On tensorise par $`G`$ la suite exacte:
$$0F^{}F𝒪_x0.$$
Soit $`[G]\mathrm{M}_{d𝔲}`$. En tenant compte du lemme 2.3c), on obtient une suite exacte
$$0\underset{¯}{\mathrm{Tor}}_1(G,𝒪_x)F^{}GFGG𝒪_x0.$$
Si $`x\mathrm{supp}G`$ alors $`\underset{¯}{\mathrm{Tor}}_1(G,𝒪_x)0`$ (voir la démonstration du lemme 5.2) et donc $`h^0(F^{}G)0`$. Sinon $`h^0(F^{}G)=h^0(FG)`$. On déduit que $`\sigma _F^{}=cst\sigma _FH_x`$. $`\mathrm{}`$
Preuve du lemme 5.7 :
Pour $`a,b_2`$, la classe du faisceau $`F=I_aI_b`$ appartient à $`\mathrm{M}_{(2,0,c_2=2)}`$. On commence par prouver que $`\sigma _F=cst\theta ^2\pi ^{}(H_aH_b)`$. La section $`\sigma _F`$ s’annule sur $`\pi ^1(\{H_a=0\})`$ et sur $`\pi ^1(\{H_b=0\})`$ d’après le lemme 5.2. Alors la section rationnelle $`\alpha =\frac{\sigma _F}{\pi ^{}(H_aH_b)}`$ de $`\mathrm{\Theta }^2`$ est régulière. Si $`d=1,2`$ on applique la proposition 4.4 pour avoir un isomorphisme $`\mathrm{H}^0(C_d,𝒪)=\mathrm{H}^O(\mathrm{M}_{d𝔲},\mathrm{\Theta }^2)`$. Pour $`d=3`$ cet isomorphisme découle du corollaire 4.13 appliqué pour $`n=0`$. Par conséquent $`\alpha =cst\theta ^2`$.
Puisque les produits $`H_aH_b`$ engendrent $`\mathrm{H}^0(C_d,𝒪(2))`$, le lemme 5.7 est prouvé pour $`d=1,2`$.
Pour $`d=3`$, en regardant la suite exacte du corollaire 4.13, on montre que les $`\sigma _F`$ engendrent le sous-espace $`\theta ^2\pi ^{}\mathrm{H}^0(C_3,𝒪(2))`$ de $`\mathrm{H}^0(\mathrm{M}_{d𝔲},\mathrm{\Theta }^2(2))`$. Pour montrer le lemme, en tenant compte du fait que $`h^0(C_3,𝒪)=1`$, il suffit de trouver un faisceau $`F`$ pour lequel $`s^{}\sigma _F0`$. Ceci équivaut à trouver une cubique $`C`$ pour laquelle $`h^0(F|_C)=h^1(F|_C)=0`$. Cette condition est satisfaite pour tout faisceau $`F`$ localement libre stable et toute cubique $`C`$. Effectivement, si on reconsidère la résolution de $`𝒪_C`$ :
$$0𝒪(3)\stackrel{\underset{¯}{C}}{}𝒪𝒪_C0,$$
comme $`\underset{¯}{\mathrm{Tor}}_1(F,𝒪_C)=0`$, on obtient
$$0F(3)\stackrel{\underset{¯}{C}}{}FF|_C0.$$
Alors la suite
$$\mathrm{H}^1(F)\mathrm{H}^1(F|_C)\mathrm{H}^2(F(3))$$
est exacte. Le nombre de Hodge $`h^1(F)=n2`$ est nul pour $`n=2`$ et par la dualité de Serre on a $`\mathrm{H}^2(F(3))=\mathrm{H}^0(F^{})=\mathrm{Hom}(F,𝒪)`$. Ce dernier groupe est nul, d’où $`h^1(F|_C)=0`$. L’annulation de $`\mathrm{Hom}(F,𝒪)`$ s’obtient ainsi: S’il existe un morphisme non nul $`m:F𝒪`$, alors $`\mathrm{im}m𝒪`$, donc $`\mathrm{im}m=I_Z`$, pour $`Z`$ sous-schéma dans $`_2`$. La stabilité de $`F`$ implique $`c_1(I_Z)=0`$ et $`\chi (I_Z)>0`$, soit que $`I_Z=𝒪`$. Alors $`\mathrm{Ker}m`$ est localement libre de $`c_1=0`$ et $`\chi =1`$, ce qui est contradictoire. Donc $`\mathrm{Hom}(F,𝒪)=0`$.$`\mathrm{}`$
## 6 Preuve de la proposition 1.2
On reprend la démarche et les notations de l’article \[D\]. Dans cet article, on se fixait un entier positif $`l`$, qui dans cette application sera toujours égal à $`1`$. On introduit la notion de système cohérent, qui consiste à considérer en même temps que le faisceau $`F`$, un sous-espace vectoriel $`\mathrm{\Gamma }`$ de son espace de sections $`\mathrm{H}^0(F)`$. La dimension de $`\mathrm{\Gamma }`$ donne l’ordre du système cohérent. À l’aide de résultats de Min He (\[He\]) sur les espaces de modules de systèmes cohérents $`(\mathrm{\Gamma },F(l))`$ d’ordre $`1`$, dont le faisceau sous-jacent est de rang $`2`$, et de classes de Chern $`c_1=2l,c_2=n+l^2`$, on se ramène au paragraphe 3 de \[D\], pour $`n`$ compris entre $`l(l1)`$ et $`(l+1)(l+2)`$, à l’étude de l’espace des sections d’un fibré vectoriel $`\mathrm{S}^{ld}𝔡^d`$ sur un ouvert $`U`$ du schéma de Hilbert $`\mathrm{Hilb}^m(_2)`$ des sous-schémas finis de longueur $`m=n+l^2`$. Si $`\mathrm{\Xi }\mathrm{Hilb}^m(_2)\times _2`$ est le sous-schéma universel, $`I_{\mathrm{\Xi }_m}`$ est le faisceaux d’idéaux associé, $`pr_1:\mathrm{Hilb}^m(_2)\times _2\mathrm{Hilb}^m(_2)`$, $`pr_2:\mathrm{Hilb}^m(_2)\times _2_2`$ sont les deux projections, le faisceau algébrique cohérent $``$ est défini par $`=R^1pr_1(I_{\mathrm{\Xi }_m}(2l3))`$. Ce faisceau est localement libre en dehors du fermé de Brill-Noether $`B`$ des schémas $`Z\mathrm{Hilb}^m(_2)`$ tels que $`h^0(I_Z(2l3))0`$. On note $`U`$ l’ouvert complémentaire de $`B`$. La codimension de $`B`$ est supérieure ou égale à $`2`$, donc les résultats de cohomologie locale nous permettent de passer de $`\mathrm{Hilb}^m(_2)`$ à $`U`$ pour le calcul d’un espace de sections. Le fibré $`𝔡=𝒟_𝔲`$ est le fibré déterminant sur le schéma de Hilbert $`\mathrm{Hilb}^m(_2)`$, identifié à l’espace de modules $`\mathrm{M}_{(1,0,c_2=n)}`$ comme dans le lemme 5.5. L’énoncé précis démontré dans \[D\] est:
###### Théorème 6.1
Soit $`n`$ un entier $`3`$. Soit $`l`$ un entier $`>0`$ tel que $`l(l1)n<(l+1)(l+2)`$. Alors on a un isomorphisme de $`\mathrm{SL}(3)`$–représentations
$$\mathrm{H}^0(\mathrm{M}_c,𝒟^d)=\mathrm{H}^0(U,\mathrm{S}^{ld}𝔡^d).$$
On désigne par $`E`$ l’espace de sections $`\mathrm{H}^0(_2,𝒪(1))`$. Au paragraphe 4 de \[D\] on montre que $`\mathrm{S}^{ld}𝔡^d`$ admet sur $`U`$ une résolution (\*) par un complexe $`K^i=\mathrm{\Lambda }^i\mathrm{S}^kE\mathrm{S}^{ld+i}(𝒪(k)^{^{[m]}})𝔡^d`$ pour $`i=0,\mathrm{},ld`$, où $`k=2l3`$, et $`𝒪(k)^{^{[m]}}`$ est défini par $`𝒪(k)^{^{[m]}}=pr_1(𝒪_\mathrm{\Xi }pr_2^{}(𝒪(k)))`$.
Il est prouvé dans \[D\] le théorème suivant:
###### Théorème 6.2
On a sur $`\mathrm{Hilb}^m(_2)`$ :
i) $`\mathrm{H}^0(𝔡^d)=\mathrm{S}^m(\mathrm{S}^dE)`$;
ii) $`\mathrm{H}^0(𝒪(k)^{^{[m]}}𝔡^d)=\mathrm{S}^{k+d}E\mathrm{S}^{m1}(\mathrm{S}^dE)`$;
iii) La $`\mathrm{SL}(3)`$-représentation $`\mathrm{H}^0(\mathrm{S}^2(𝒪(k)^{^{[m]}})𝔡^d)`$ est isomorphe à la représentation $`(\mathrm{S}^{2k+d}E\mathrm{S}^{m1}(\mathrm{S}^dE))(\mathrm{Ker}_k\mathrm{S}^{m2}(\mathrm{S}^dE))`$$`\mathrm{Ker}_k`$ est le noyau de la multiplication $`\mathrm{S}^2(\mathrm{S}^{k+d}E)\mathrm{S}^{2k+2d}E`$;
iv) La $`\mathrm{SL}(3)`$-représentation $`\mathrm{H}^0(\mathrm{S}^3(𝒪(k)^{^{[m]}})𝔡^d)`$ est le noyau du morphisme $`\alpha `$:
$$\alpha :\left[\mathrm{S}^{3k+d}E\mathrm{S}^{m1}(\mathrm{S}^dE)\right]\left[\mathrm{S}^{2k+d}E\mathrm{S}^{k+d}E\mathrm{S}^{m2}(\mathrm{S}^dE)\right]\left[\mathrm{S}^3(\mathrm{S}^{k+d}E)\mathrm{S}^{m3}(\mathrm{S}^dE)\right]$$
$$\left[\mathrm{S}^{3k+2d1}EE\mathrm{S}^{m2}(\mathrm{S}^dE)\right]\left[\mathrm{S}^{2k+2d}E\mathrm{S}^{k+d}E\mathrm{S}^{m3}(\mathrm{S}^dE)\right]$$
donné par la matrice:
$$\left(\begin{array}{ccc}\stackrel{~}{}& \stackrel{~}{D}& 0\\ 0& \rho & \stackrel{~}{\nu }\end{array}\right)$$
$`\stackrel{~}{}`$, $`\stackrel{~}{D}`$, $`\rho `$ et $`\stackrel{~}{\nu }`$ sont des opérateurs explicites.
Le point (i) est une conséquence du théorème de Kawamata-Viehweg (\[C-K-M\]), expliquée dans le lemme 5.5. Le point (ii) correspond au lemme 4.10 de \[D\]. Le point (iii) correspond au lemme 4.11 de \[D\]. Le point (iv) correspond à la proposition 5.13 de \[D\].
Le théorème 6.1 appliqué au cas $`l=1`$ nous donne un isomorphisme de $`\mathrm{SL}(3)`$-représentations
$$\mathrm{H}^0(\mathrm{M}_c,𝒟^d)=\mathrm{H}^0(\mathrm{Hilb}^{n+1}_2,\mathrm{S}^d𝔡^d)\text{ pour }3n5.$$
À partir de la présentation (\*) de $``$, appliquée pour $`k=2l3=1`$ et $`m=n+l^2=n+1`$, on obtient $`𝒪(1)^{^{[m]}}`$ et donc
$$\mathrm{H}^0(\mathrm{M}_c,𝒟^d)=\mathrm{H}^0(\mathrm{Hilb}^{n+1}_2,\mathrm{S}^d(𝒪(1)^{^{[m]}})𝔡^d)\text{ pour }3n5.$$
Pour $`d=2`$ on applique le théorème 6.2 iii) avec $`k=2l3=1`$, $`d=2`$ et $`m=n+l^2=n+1`$: $`\mathrm{H}^0(\mathrm{M}_c,𝒟^2)`$ est le noyau du morphisme surjectif
$$\mathrm{S}^n(\mathrm{S}^2E)\mathrm{S}^2E\mathrm{S}^{n1}(\mathrm{S}^2E)\stackrel{(0,id)}{}\mathrm{S}^2E\mathrm{S}^{n1}(\mathrm{S}^2E).$$
Par suite, la représentation $`\mathrm{H}^0(\mathrm{M}_c,𝒟^2)`$ est isomorphe à $`\mathrm{S}^n(\mathrm{S}^2E)`$ de dimension $`C_{n+5}^n`$. Pour $`d=3`$ on s’intéresse à l’espace $`\mathrm{H}^0(\mathrm{Hilb}^{n+1}_2,\mathrm{S}^3(𝒪(1)^{^{[m]}})𝔡^3)`$. Par le théorème 6.2 iv) on est amenés à étudier le noyau du morphisme $`\alpha `$ (on fait $`k=1`$, $`d=3`$ et $`m=n+1`$):
$`\alpha :\mathrm{S}^n(\mathrm{S}^3E)E\mathrm{S}^2E\mathrm{S}^{n1}(\mathrm{S}^3E)\mathrm{S}^3(\mathrm{S}^2E)\mathrm{S}^{n2}(\mathrm{S}^3E)`$
$`E\mathrm{S}^2E\mathrm{S}^{n1}(\mathrm{S}^3E)\mathrm{S}^4E\mathrm{S}^2E\mathrm{S}^{n2}(\mathrm{S}^3E).`$
On montre successivement selon une démarche analogue à celle des lemmes 5.16, 5.17, 5.19, 5.20 de l’article \[D\], que $`\stackrel{~}{D}`$ est un isomorphisme, que le noyau de $`\stackrel{~}{\nu }`$ est égal à $`\mathrm{S}^{2,2,2}E\mathrm{S}^{n2}(\mathrm{S}^3E)=\mathrm{S}^{n2}(\mathrm{S}^3E)`$ et son conoyau à $`\mathrm{S}^{5,1}E\mathrm{S}^{n2}(\mathrm{S}^3E)`$, et que le morphisme de liaison $`(0,\rho ):\mathrm{Ker}(\stackrel{~}{},\stackrel{~}{D})\mathrm{coker}\stackrel{~}{\nu }`$ est nul. Par conséquent, l’équivalent de la proposition 5.18 de \[D\] nous assure que le noyau de $`\alpha `$ est isomorphe à $`\mathrm{S}^n(\mathrm{S}^3E)\mathrm{S}^{n2}(\mathrm{S}^3E)`$ de dimension $`C_{n+9}^9+C_{n+7}^9`$. $`\mathrm{}`$
Ceci conclut la preuve du théorème 1.1.
## 7 Sections de $`𝒟^k`$ pour $`n=c_24`$
Le but de ce paragraphe est de calculer les dimensions des espaces de sections $`\mathrm{H}^0(\mathrm{M}_{(2,0,c_2=n)},𝒟_{𝔲}^{}{}_{}{}^{k})`$ pour $`n4`$.
Dans le cas $`n=2`$, le morphisme de Barth nous fournit un isomorphisme entre l’espace de modules $`\mathrm{M}_c`$ et $`_5`$, et le fibré déterminant s’identifie à $`𝒪(1)`$. Les cas intéressants sont donc $`n=3`$ et $`n=4`$.
On commence par un résultat général sur la fonction $`kh^0(\mathrm{M}_c,𝒟_u^k)`$ pour toute classe $`c=(r,c_1,c_2)`$ satisfaisant $`r>0`$ et $`\mathrm{M}_c`$ non-vide. On rappelle (§3.2) la notation $`𝔲=(0,\frac{r}{\delta },\frac{c_1}{\delta })`$$`\delta =\mathrm{pgcd}(\mathrm{r},\mathrm{c}_1)`$. D’après le théorème 3.7, la cohomologie supérieure $`\mathrm{H}^q(\mathrm{M}_c,𝒟_{𝔲}^{}{}_{}{}^{k})`$ s’annule pour $`k3\delta `$.
Ce résultat, l’isomorphisme $`\omega _{\mathrm{M}_c}𝒟_{𝔲}^{}{}_{}{}^{3\delta }`$ et la dualité de Serre nous assurent que
$`h^0(𝒟^j)`$ $`=`$ $`0\mathrm{si}j<0`$
$`h^q(𝒟^j)`$ $`=`$ $`0j\mathrm{si}0<q<D`$
$`h^D(𝒟^j)`$ $`=`$ $`0\mathrm{si}j>3\delta .`$
On note $`D=dim\mathrm{M}_c=1<c,c^{}>`$. On note $`𝒟=𝒟_𝔲`$.
###### Proposition 7.1
i) Supposons $`d2`$. Pour $`k>3\delta `$, la fonction
$$kh^0(\mathrm{M}_c,𝒟^k)$$
est un polynôme de degré $`D`$, de coefficient dominant
$$q_D=\frac{1}{D!}_{\mathrm{M}_c}c_1(𝒟)^D.$$
ii) La série de Poincaré $`P(t)`$ est de la forme
$$\frac{Q(t)}{(1t)^{D+1}}$$
$`Q`$ est un polynôme de degré $`D+13\delta `$, à coefficients entiers, tel que $`Q(1)=_{\mathrm{M}_c}c_1(𝒟)^D`$;
iii) Le polynôme $`Q`$ satisfait à la condition de symétrie
$$t^{D+13\delta }Q(\frac{1}{t})=Q(t).$$
Preuve :
La formule de Riemann-Roch pour des variétés éventuellement singulières (\[B-F-M\]) et le théorème 3.7 donnent
$$\begin{array}{ccccc}h^0(\mathrm{M}_c,𝒟^k)& =& \chi (\mathrm{M}_c,𝒟^k)& =& _{\mathrm{M}_c}ch(𝒟^k)Td(\mathrm{M}_c)\\ & =& _{\mathrm{M}_c}e^{kc_1(𝒟)}Td(\mathrm{M}_c)& =& _{0jD}\frac{k^j}{j!}_{\mathrm{M}_c}c_1(𝒟)^jTd(\mathrm{M}_c).\end{array}$$
Puisque $`𝒟`$ est big $`q_D>0`$, d’où i).
Posons $`a_j=_{\mathrm{M}_c}c_1(𝒟)^jTd(\mathrm{M}_c)`$. La série de Poincaré est
$$\underset{0jD}{}a_j(\underset{k0}{}\frac{k^j}{j!}t^k).$$
La somme de la série $`_{k0}\frac{k^j}{j!}t^k`$ (de rayon de convergence $`1`$) est une fonction rationnelle de la forme $`\frac{Q_j(t)}{(1t)^{j+1}}`$ où les polynômes $`Q_j`$ sont de degré inférieur où égal à $`j`$, et $`Q_j(1)=1`$. Ceci se voit par récurrence sur $`j`$. Il en résulte que la série de Poincaré est de la forme voulue. Puisque $`Q(t)=(1t)^{D+1}P(t)`$ et que $`P(t)`$ est une série formelle à coefficients entiers, le calcul de ce produit montre que $`Q(t)`$ est bien à coefficients entiers.
La relation classique (voir par exemple \[Fult\]) $`Td(V^{})ch(\lambda _1(V))=c_{top}(V^{})`$ appliquée au fibré $`V=W𝒟^1`$, où $`W`$ est un espace vectoriel de dimension $`m=D+1`$, prouve que
$$ch(\lambda _1(W𝒟^1))=c_{top}(W^{}𝒟)Td^1(W^{}𝒟).$$
Ici
$$\lambda _1(V)=\mathrm{\Lambda }^0V\mathrm{\Lambda }^1V+\mathrm{\Lambda }^2V\mathrm{}+(1)^{dimV}\mathrm{\Lambda }^{dimV}V$$
désigne la somme alternée des puissances extérieures du fibré $`V`$ et $`c_{top}`$ d’un fibré vectoriel de rang $`r`$ désigne la classe de Chern $`c_r`$ de ce fibré.
Mais $`V=W𝒟^1`$ est un fibré de rang $`D+1`$, donc sa classe de Chern maximale $`c_{D+1}(W𝒟^1)`$ appartient à l’espace de cohomologie $`\mathrm{H}^{2(D+1)}(\mathrm{M}_c)`$ qui est nul en raison de la dimension de $`\mathrm{M}_c`$ ($`dim\mathrm{M}_c=D`$). Donc $`ch(\lambda _1(W𝒟^1)𝒟^k)=0`$ et par la formule de Riemann-Roch on obtient
$$\chi (\lambda _1(W𝒟^1)𝒟^k)=_{\mathrm{M}_c}ch(\lambda _1(W𝒟^1)𝒟^k)Td(\mathrm{M}_c)=0$$
soit, en tenant compte de $`\mathrm{\Lambda }^i(W𝒟^1)=\mathrm{\Lambda }^iW𝒟^i`$,
$$S_k=\underset{i=0}{\overset{m}{}}(1)^iC_m^i\chi (𝒟^{ki})=0$$
quelque soit $`k`$.
Comme $`Q(t)=(1t)^{D+1}P(t)`$, le coefficient $`k`$-ième de $`Q`$ s’écrit
$`Q_k`$ $`=`$ $`\left(\begin{array}{c}m\\ 0\end{array}\right)h^0(𝒟^k)\left(\begin{array}{c}m\\ 1\end{array}\right)h^0(𝒟^{k1})+\mathrm{}+(1)^m\left(\begin{array}{c}m\\ m\end{array}\right)h^0(𝒟^{km})`$
$`=`$ $`{\displaystyle \underset{i=0}{\overset{m}{}}}(i)^iC_m^ih^0(𝒟^{ki})`$
(en tenant compte du fait que $`h^0(𝒟^j)=0`$ si $`j<0`$). Pour $`k<m3\delta `$ on obtient $`Q_k=S_k=0`$ en raison de l’annulation de la cohomologie supérieure $`\mathrm{H}^q(𝒟^j)`$ pour $`j<3\delta ,q>0`$ et de $`\mathrm{H}^0(𝒟^j)`$ pour $`j<0`$. D’où $`\mathrm{deg}Qm3\delta `$. La condition de symétrie s’exprime sur la symétrie des coefficients de $`Q`$ :
$$Q_k=Q_{m3\delta k}\mathrm{pour}0km3\delta .$$
Mais dans ce cas on a
$`Q_k`$ $`=`$ $`S_k{\displaystyle \underset{i=k+3\delta }{\overset{m}{}}}(i)^i\left(\begin{array}{c}m\\ i\end{array}\right)\chi (𝒟^{ki}).`$
La dualité de Serre s’écrit pour $`j3\delta `$ sous la forme $`\chi (𝒟^j)=(1)^Dh^0(𝒟^{j3\delta })`$. En l’appliquant dans la somme ci-dessus pour $`j=ki3\delta `$, et en faisant ensuite la transformation $`j=mi`$ on trouve
$`Q_k`$ $`=`$ $`S_k(1)^m{\displaystyle \underset{i=k+3\delta }{\overset{m}{}}}(i)^{mi}\left(\begin{array}{c}m\\ mi\end{array}\right)(1)^Dh^0(𝒟^{k+i3\delta })`$
$`=`$ $`S_k+{\displaystyle \underset{j=0}{\overset{mk3\delta }{}}}(i)^j\left(\begin{array}{c}m\\ j\end{array}\right)h^0(𝒟^{mk3\delta j})`$
$`=`$ $`S_k+Q_{m3\delta k}=Q_{m3\delta k}.`$
En particulier $`Q_0=Q_{m3\delta }=1`$ donc le degré de $`Q`$ est égal à $`m3\delta `$ exactement. $`\mathrm{}`$
Nous revenons aux cas particuliers qui nous intéressent. On prend $`c=(2,0,c_2=n)`$ pour $`n=3,4`$, $`\delta =2`$ et $`𝔲=(0,1,0).`$
Preuve du théorème 1.3 :
i) Ici, $`D=4c_23=9`$. La proposition précédente donne que $`P(t)`$ s’écrit sous la forme $`\frac{Q(t)}{(1t)^{10}}`$$`Q`$ est de degré $`4`$, vérifie la condition de symétrie et $`Q(1)`$ est égal à $`3=9!q_9`$$`q_9`$ est un nombre de Donaldson (\[Barth77\]). Le calcul de $`h^0(\mathrm{M}_c,𝒟^0)=1`$ et $`h^0(\mathrm{M}_c,𝒟)=10`$ (cf. chapitre 2) permet de conclure que $`Q(t)=1+t^2+t^4`$.
ii) Ce cas est analogue au précédent seulement il faut faire intervenir $`h^0(\mathrm{M}_c,𝒟)=15`$, et aussi $`h^0(\mathrm{M}_c,𝒟^2)=126`$ et $`h^0(\mathrm{M}_c,𝒟^3)=770`$ calculés dans 1.2. Ici $`Q(1)=54=13!q_{13}`$ (\[L-Q\]). $`\mathrm{}`$
###### Remarque 7.2
Dans le cas $`n=3`$, puisqu’on a obtenu $`h^0(\mathrm{M}_c,𝒟^d)`$ pour $`d=2,3`$ (cf. 1.2), on aurait pu déduire la valeur du nombre de Donaldson $`q_9`$.
Remerciements : Mes remerciements s’adressent à J. Le Potier, mon directeur de thèse, ainsi qu’à N. Dan. Je remercie D. Roessler pour la référence \[Fult\].
|
warning/0004/nlin0004013.html
|
ar5iv
|
text
|
# The critical 𝐴_{𝑛-1}⁽¹⁾ chain
## 1 Introduction
The one dimensional spin $`A_{n1}^{(1)}`$ chain is described by the Hamiltonian :
$`={\displaystyle \underset{k=\mathrm{}}{\overset{\mathrm{}}{}}}\left\{q{\displaystyle \underset{\genfrac{}{}{0pt}{}{a,b=0}{a>b}}{\overset{n1}{}}}e_{aa}^{(k+1)}e_{bb}^{(k)}+q^1{\displaystyle \underset{\genfrac{}{}{0pt}{}{a,b=0}{a<b}}{\overset{n1}{}}}e_{aa}^{(k+1)}e_{bb}^{(k)}{\displaystyle \underset{\genfrac{}{}{0pt}{}{a,b=0}{ab}}{\overset{n1}{}}}e_{ab}^{(k+1)}e_{ba}^{(k)}\right\},`$ (1.1)
where $`e_{ab}^{(k)}`$ is a matrix unit acting on $`k`$ th site. In this paper we consider the critical regime $`|q|=1`$. For the special case $`n=2`$ this model becomes the celebrated XXZ spin. M.Jimbo,H.Konno and T.Miwa established the trace construction of the correlation functions for the XXZ chain at critical regime $`|q|=1`$. In this paper we give the correlation functions for the higher rank generalization of the XXZ chain at the critical regime $`|q|=1`$. In order to write down the correlation functions, we need the ’dual’ vertex operators. This problem was absent from the XXZ chain, because the vertex operators are self-dual in this case. We give the pair of the vertex operators and their duals in this paper.
We give the $`N`$ point correlation functions which describe the ground-state average $`𝒪`$ of the local operator :
$`𝒪=e_{ϵ_1ϵ_1^{}}^{(1)}\mathrm{}e_{ϵ_Nϵ_N^{}}^{(N)}.`$ (1.2)
The correlation functions of the critical $`A_{n1}^{(1)}`$ chain are described by the following systems of difference equations :
$`G^{(N)}(\beta _1,\mathrm{},\beta _j\lambda i,\mathrm{},\beta _N|\beta _{N+1},\mathrm{},\beta _{2N})`$ (1.3)
$`=`$ $`R_{j,j1}^{V^{}V^{}}(\beta _j\beta _{j1}\lambda i)\mathrm{}R_{j,1}^{V^{}V^{}}(\beta _j\beta _1\lambda i)R_{j,2N}^{V^{}V}(\beta _j\beta _{2N})\mathrm{}R_{j,N+1}^{V^{}V}(\beta _j\beta _{N+1})`$
$`\times `$ $`R_{j,N}^{V^{}V^{}}(\beta _j\beta _N)\mathrm{}R_{j,j+1}^{V^{}V^{}}(\beta _j\beta _{j+1})G^{(N)}(\beta _1,\mathrm{},\beta _j,\mathrm{},\beta _N|\beta _{N+1},\mathrm{},\beta _{2N}),`$
and
$`G^{(N)}(\beta _1,\mathrm{},\beta _N|\beta _{N+1},\mathrm{},\beta _j+i\lambda ,\mathrm{}\beta _{2N})`$ (1.4)
$`=`$ $`R_{j+1,j}^{VV}(\beta _{j+1}\beta _j+\lambda i)\mathrm{}R_{2N.j}^{VV}(\beta _{2N}\beta _j+\lambda i)R_{1,j}^{V^{}V}(\beta _1\beta _j)\mathrm{}R_{N,j}^{V^{}V}(\beta _N\beta _j)`$
$`\times `$ $`R_{N+1,j}^{VV}(\beta _{N+1}\beta _j)\mathrm{}R_{j1,j}^{VV}(\beta _{j1}\beta _j)G^{(N)}(\beta _1,\mathrm{},\beta _N|\beta _{N+1},\mathrm{},\beta _j,\mathrm{}\beta _{2N}).`$
Here $`R_{ij}^{VV}(\beta )\mathrm{End}(V^{2N})`$ signifies the matrix acting as $`R^{VV}(\beta )`$ on th $`(i,j)`$th tensor components and as identity elsewhere. Here $`R^{VV}(\beta ),R^{V^{}V^{}}(\beta )`$ and $`R^{V^{}V}(\beta )`$ are given by (3.1), (3.2) and (3.4).
The correlation functions satisfy the restriction equation :
$`G^{(N)}(\beta +i(\pi \lambda ),\beta _2,\mathrm{},\beta _N|\beta _{N+1},\mathrm{},\beta _{2N})_{ϵ_1\mathrm{}ϵ_N,ϵ_N^{}\mathrm{}ϵ_1^{}}`$ (1.5)
$`=`$ $`\delta _{ϵ_1,ϵ_1^{}}G^{(N1)}(\beta _2,\mathrm{},\beta _N|\beta _{N+1},\mathrm{},\beta _{2N1})_{ϵ_2\mathrm{}ϵ_N,ϵ_N^{}\mathrm{}ϵ_2^{}}.`$
Here we have set
$`G^{(N)}(\beta _1\mathrm{}\beta _N|\beta _{N+1}\mathrm{}\beta _{2N})`$ (1.6)
$`=`$ $`{\displaystyle \underset{ϵ_1\mathrm{}ϵ_{2N}=0}{\overset{n1}{}}}v_{ϵ_1}\mathrm{}v_{ϵ_{2N}}G^{(N)}(\beta _1\mathrm{}\beta _N|\beta _{N+1}\mathrm{}\beta _{2N})_{ϵ_1\mathrm{}ϵ_{2N}}.`$
In this paper we set the deformation parameter as
$`q=\mathrm{exp}\left({\displaystyle \frac{\pi i}{\xi }}\right),`$ (1.7)
where $`\xi >1`$.
The ground state averages are obtaind from the components $`G^{(N)}`$, by taking $`\lambda =2\pi `$, and specializing the spectral parameters :
$`G^{(N)}(\beta +\pi i,\mathrm{},\beta +\pi i|\beta ,\mathrm{},\beta )_{ϵ_1\mathrm{}ϵ_N,ϵ_N\mathrm{}ϵ_1},`$ (1.8)
When we solve the diffence equations (1.3), (1.4), directly , we have a difficulty. The difficulty in this approach is that the solutions are determined only up to arbitrary periodic functions, so one has to single out in some way the correct solutions which correspond to the correlation functions. When we construct the solutions by the trace of the vertex operators, the ambiguity of solutions are resolved. In this paper we give the free boson realizations of the type-I vertex operators and their duals, and give the trace construction for the correlation functions of the critical $`A_{n1}^{(1)}`$ spin. In this connection, we should mention about the work , in which the authors give the free boson realizations of the dual type-II vertex operators of the $`A_{n1}^{(1)}`$ Toda field theory with imaginary coupling.
Now a few words about the organization of the paper. In section 2 we review the model briefly. In section 3 we give the defining relations of the vertex operators. In section 4 we give the free boson realizations of the vertex operators. In section 5 we give the proofs of the properties of the vertex operators. In section 6 we give the integral representations of the correlation functions. In Appendix A we summarized the multi Gamma functions. In Appendix B we summarized the normal ordering of the basic operators.
## 2 The critical $`A_{n1}^{(1)}`$ chain
Let us set the $`R`$-matrix by
$`R^{VV}(\beta )=r(\beta )\overline{R}(\beta ),r(\beta )={\displaystyle \frac{S_2(i\beta |\frac{2\pi }{n}\xi ,2\pi )S_2(i\beta +\frac{2\pi }{n}|\frac{2\pi }{n}\xi ,2\pi )}{S_2(i\beta |\frac{2\pi }{n}\xi ,2\pi )S_2(i\beta +\frac{2\pi }{n}|\frac{2\pi }{n}\xi ,2\pi )}},`$ (2.1)
where $`S_2(\beta |\omega _1,\omega _2)`$ is the Multi-Sine function given in Appendix A.
The matrix $`\overline{R}(\beta )`$ is given as follows :
$`\overline{R}(\beta )v_{k_1}v_{k_2}={\displaystyle \underset{j_1,j_2=0}{\overset{n1}{}}}v_{j_1}v_{j_2}\overline{R}(\beta )_{j_1j_2}^{k_1k_2},`$ (2.2)
where the nonzero entrise are
$`\overline{R}(\beta )_{jj}^{jj}`$ $`=`$ $`1,`$ (2.3)
$`\overline{R}(\beta )_{jk}^{jk}`$ $`=`$ $`{\displaystyle \frac{\mathrm{sh}\left({\displaystyle \frac{n}{2\xi }}\beta \right)}{\mathrm{sh}\left({\displaystyle \frac{n}{2\xi }}(\beta +{\displaystyle \frac{2\pi i}{n}})\right)}},(jk),`$ (2.4)
$`\overline{R}(\beta )_{jk}^{kj}`$ $`=`$ $`\{\begin{array}{cc}\frac{e^{{\scriptscriptstyle \frac{n}{2\xi }}\beta }\mathrm{sh}\left({\displaystyle \frac{\pi i}{\xi }}\right)}{\mathrm{sh}\left({\displaystyle \frac{n}{2\xi }}\left(\beta +{\displaystyle \frac{2\pi i}{n}}\right)\right)},& (j<k),\\ \frac{e^{{\scriptscriptstyle \frac{n}{2\xi }}\beta }\mathrm{sh}\left({\displaystyle \frac{\pi i}{\xi }}\right)}{\mathrm{sh}\left({\displaystyle \frac{n}{2\xi }}\left(\beta +{\displaystyle \frac{2\pi i}{n}}\right)\right)},& (j>k),\end{array}`$ (2.7)
The $`R`$-matrix satisfies the Yang-Baxter equations,
$`R_{12}^{VV}(\beta _1\beta _2)R_{13}^{VV}(\beta _1\beta _3)R_{23}^{VV}(\beta _2\beta _3)=R_{23}^{VV}(\beta _2\beta _3)R_{13}^{VV}(\beta _1\beta _3)R_{12}^{VV}(\beta _1\beta _2),`$ (2.8)
and the unitarity,
$`R_{12}^{VV}(\beta 1\beta _2)R_{21}^{VV}(\beta _2\beta _1)=id.`$ (2.9)
Let us set the monodromy matrix $`𝒯(\beta )`$ acting on the $`(N+1)`$ fold tensor product $`V_0V_1\mathrm{}V_N,(V_k=V=^n)`$,
$`𝒯(\beta )=R_{0,1}^{VV}(\beta )\mathrm{}R_{0,N}^{VV}(\beta )=\left(\begin{array}{ccccc}A_{1,1}& A_{1,2}& \mathrm{}& A_{1,n1}& A_{1,n}\\ A_{2,1}& A_{2,2}& \mathrm{}& A_{2,n1}& A_{2,n}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ A_{n1,1}& A_{n1,2}& \mathrm{}& A_{n1,n1}& A_{n1,n}\\ A_{n,1}& A_{n,2}& \mathrm{}& A_{n,n1}& A_{n,n}\end{array}\right),`$ (2.16)
where the partition into $`n\times n`$ blocks is according to the base of $`V_0`$.
Let us set the transfer matrix $`T(\beta )`$ acting on the $`N`$-th fold tensor product $`V_1\mathrm{}V_N`$ by
$`T(\beta )={\displaystyle \underset{j=1}{\overset{n}{}}}A_{j,j}=\mathrm{tr}_{V_0}(𝒯(\beta )).`$ (2.17)
From the Yang-Baxter equation, we know the transfer matrices commute each other.
$`[T(\beta _1),T(\beta _2)]=0.`$ (2.18)
In the thermodynamic limit $`N\mathrm{}`$, the logarithmic derivative of the transfer matrix and the Hamiltonian (1.1) have the following relation.
$`\left({\displaystyle \frac{d}{d\beta }}\mathrm{log}T\right)(0).`$ (2.19)
A.Doikou and R.I.Nepomechie computed by Bethe Ansatz the scattering matrix for the critical $`A_{n1}^{(1)}`$ spin chain. The scattering matrix $`S(\beta )`$ is given by
$`S(\beta )=s(\beta )\overline{S}(\beta ),s(\beta )={\displaystyle \frac{S_2(i\beta |\frac{2\pi }{n}(\xi 1),2\pi )S_2(i\beta +\frac{2(n1)\pi }{n}|\frac{2\pi }{n}(\xi 1),2\pi )}{S_2(i\beta |\frac{2\pi }{n}(\xi 1),2\pi )S_2(i\beta +\frac{2(n1)\pi }{n}|\frac{2\pi }{n}(\xi 1),2\pi )}}.`$ (2.20)
The matrix $`\overline{S}(\beta )`$ is given as follows :
$`\overline{S}(\beta )v_{k_1}v_{k_2}={\displaystyle \underset{j_1,j_2=0}{\overset{n1}{}}}v_{j_1}v_{j_2}\overline{S}(\beta )_{j_1j_2}^{k_1k_2},`$ (2.21)
where the nonzero entrise are
$`\overline{S}(\beta )_{jj}^{jj}`$ $`=`$ $`1,`$ (2.22)
$`\overline{S}(\beta )_{jk}^{jk}`$ $`=`$ $`{\displaystyle \frac{\mathrm{sh}\left({\displaystyle \frac{n}{2(\xi 1)}}\beta \right)}{\mathrm{sh}\left({\displaystyle \frac{n}{2(\xi 1)}}(\beta {\displaystyle \frac{2\pi i}{n}})\right)}},(jk),`$ (2.23)
$`\overline{S}(\beta )_{jk}^{kj}`$ $`=`$ $`\{\begin{array}{cc}\frac{e^{{\scriptscriptstyle \frac{n}{2(\xi 1)}}\beta }\mathrm{sh}\left({\displaystyle \frac{\pi i}{(\xi 1)}}\right)}{\mathrm{sh}\left({\displaystyle \frac{n}{2(\xi 1)}}\left(\beta {\displaystyle \frac{2\pi i}{n}}\right)\right)},& (j<k),\\ \frac{e^{{\scriptscriptstyle \frac{n}{2(\xi 1)}}\beta }\mathrm{sh}\left({\displaystyle \frac{\pi i}{(\xi 1)}}\right)}{\mathrm{sh}\left({\displaystyle \frac{n}{2(\xi 1)}}\left(\beta {\displaystyle \frac{2\pi i}{n}}\right)\right)},& (j>k),\end{array}`$ (2.26)
The scattering matrix of the present model agrees with the scattering matrix of the $`A_{n1}^{(1)}`$ Toda fields theory with imaginary coupling.
## 3 Vertex Operators
The type-I vertex operators satisfy
$`\mathrm{\Phi }_{j_2}(\beta _1)\mathrm{\Phi }_{j_1}(\beta _2)={\displaystyle \underset{k_1k_2=0}{\overset{n1}{}}}R^{VV}(\beta _1\beta _2)_{j_1j_2}^{k_1k_2}\mathrm{\Phi }_{k_1}(\beta _2)\mathrm{\Phi }_{k_2}(\beta _1).`$ (3.1)
The dual type-I vertex operators satisfy
$`\mathrm{\Phi }_{j_2}^{}(\beta _1)\mathrm{\Phi }_{j_1}^{}(\beta _2)={\displaystyle \underset{k_1k_2=0}{\overset{n1}{}}}R^{V^{}V^{}}(\beta _1\beta _2)_{j_1j_2}^{k_1k_2}\mathrm{\Phi }_{k_1}^{}(\beta _2)\mathrm{\Phi }_{k_2}^{}(\beta _1),`$ (3.2)
where we set
$`R^{V^{}V^{}}(\beta )_{j_1j_2}^{k_1k_2}=R^{VV}(\beta )_{j_2j_1}^{k_2k_1}.`$ (3.3)
The type-I and dual type-I vertex operators satisfy
$`\mathrm{\Phi }_{j_2}(\beta _1)\mathrm{\Phi }_{j_1}^{}\left(\beta _2\right)={\displaystyle \underset{k_1k_2=0}{\overset{n1}{}}}R^{V^{}V}(\beta _1\beta _2)_{j_1j_2}^{k_1k_2}\mathrm{\Phi }_{k_1}^{}(\beta _2)\mathrm{\Phi }_{k_2}(\beta _1),`$ (3.4)
where we set
$`R^{VV^{}}(\beta )=r^{}(\beta )\overline{R}^{}(\beta ),r^{}(\beta )={\displaystyle \frac{S_2(i\alpha +\pi |\frac{2\pi }{n}\xi ,2\pi )S_2(i\alpha +\pi +\frac{2\pi }{n}|\frac{2\pi }{n}\xi ,2\pi )}{S_2(i\alpha +\pi |\frac{2\pi }{n}\xi ,2\pi )S_2(i\alpha +\pi +\frac{2\pi }{n}|\frac{2\pi }{n}\xi ,2\pi )}}.`$ (3.5)
The matrix $`\overline{R}^{}(\beta )`$ is given as follows :
$`\overline{R}^{}(\beta )v_{k_1}v_{k_2}={\displaystyle \underset{j_1,j_2=0}{\overset{n1}{}}}v_{j_1}v_{j_2}\overline{R}^{}(\beta )_{j_1j_2}^{k_1k_2},`$ (3.6)
where the nonzero entrise are
$`\overline{R}^{}(\beta )_{jk}^{jk}`$ $`=`$ $`1,(jk),`$ (3.7)
$`\overline{R}^{}(\beta )_{jj}^{jj}`$ $`=`$ $`{\displaystyle \frac{\mathrm{sh}\left({\displaystyle \frac{n}{2\xi }}\beta \right)}{\mathrm{sh}\left({\displaystyle \frac{n}{2\xi }}(\beta +{\displaystyle \frac{2\pi i}{n}})\right)}},`$ (3.8)
$`\overline{R}^{}(\beta )_{jj}^{kk}`$ $`=`$ $`\{\begin{array}{cc}\frac{e^{{\scriptscriptstyle \frac{n}{2\xi }}\beta }\mathrm{sh}\left({\displaystyle \frac{\pi }{\xi }}i\right)}{\mathrm{sh}\left({\displaystyle \frac{n}{2\xi }}\left(\beta +{\displaystyle \frac{2\pi i}{n}}\right)\right)},& (j>k),\\ \frac{e^{{\scriptscriptstyle \frac{n}{2\xi }}\beta }\mathrm{sh}\left({\displaystyle \frac{\pi }{\xi }}i\right)}{\mathrm{sh}\left({\displaystyle \frac{n}{2\xi }}\left(\beta +{\displaystyle \frac{2\pi i}{n}}\right)\right)},& (j<k),\end{array}`$ (3.11)
The type-I and the dual type-I vertex operators satisfy
$`\mathrm{\Phi }_{j_2}^{}(\beta _1)\mathrm{\Phi }_{j_1}(\beta _2)={\displaystyle \underset{k_1k_2=0}{\overset{n1}{}}}R^{VV^{}}(\beta _1\beta _2)_{j_1j_2}^{k_1k_2}\mathrm{\Phi }_{k_1}(\beta _2)\mathrm{\Phi }_{k_2}^{}(\beta _1),`$ (3.12)
where we set
$`R^{VV^{}}(\beta )_{j_1j_2}^{k_1k_2}=R^{V^{}V}(\beta )_{k_1k_2}^{j_1j_2}.`$ (3.13)
The type-I vertex operator and it’s dual satisfy the inversion relation :
$`\mathrm{\Phi }_{j_1}(\beta )\mathrm{\Phi }_{j_2}^{}\left(\beta +\pi i\right)=g_n^1e^{\frac{2\pi i}{n}j_1}\delta _{j_1,j_2}id,(j_1j_2).`$ (3.14)
where we set
$`g_n^1=\left({\displaystyle \frac{2\xi }{n}}\right)^{n1}\mathrm{sh}\left({\displaystyle \frac{\pi i}{\xi }}\right)e^{\frac{\xi 1}{\xi }(\gamma +\mathrm{log}\frac{2\pi \xi }{n})n+\frac{\pi i}{n}(1n)}{\displaystyle \frac{\mathrm{\Gamma }(1/\xi )^{n1}}{\mathrm{\Gamma }(11/\xi )}}.`$ (3.15)
## 4 Free boson realizations
In this section we give the free boson realizations of the vertex operators.
Let us set bose-fields as
$`[b_j(t),b_k(t^{})]={\displaystyle \frac{1}{t}}{\displaystyle \frac{\mathrm{sh}\left(\frac{\pi }{n}(a_j|a_k)t\right)\mathrm{sh}\left(\frac{\pi }{n}(\xi 1)t\right)}{\mathrm{sh}\left(\frac{\pi }{n}t\right)\mathrm{sh}\left(\frac{\pi }{n}\xi t\right)}}\delta (t+t^{})`$ (4.1)
Let us set
$`b_1^{}(t)`$ $`=`$ $`{\displaystyle \underset{j=1}{\overset{n1}{}}}b_j(t){\displaystyle \frac{\mathrm{sh}\frac{(nj)\pi t}{n}}{\mathrm{sh}\pi t}},`$ (4.2)
$`b_{n1}^{}(t)`$ $`=`$ $`{\displaystyle \underset{j=1}{\overset{n1}{}}}b_j(t){\displaystyle \frac{\mathrm{sh}\frac{j\pi t}{n}}{\mathrm{sh}\pi t}}.`$ (4.3)
We have
$`[b_1^{}(t),b_j(t^{})]`$ $`=`$ $`\delta _{j1}{\displaystyle \frac{1}{t}}{\displaystyle \frac{\mathrm{sh}\left(\frac{\pi }{n}(\xi 1)t\right)}{\mathrm{sh}\left(\frac{\pi }{n}\xi t\right)}}\delta (t+t^{}),`$ (4.4)
$`[b_j(t),b_{n1}^{}(t^{})]`$ $`=`$ $`\delta _{j,n1}{\displaystyle \frac{1}{t}}{\displaystyle \frac{\mathrm{sh}\left(\frac{\pi }{n}(\xi 1)t\right)}{\mathrm{sh}\left(\frac{\pi }{n}\xi t\right)}}\delta (t+t^{}).`$ (4.5)
Let us set the basic operators as
$`U_j(\beta )`$ $`=`$ $`:\mathrm{exp}({\displaystyle _{\mathrm{}}^{\mathrm{}}}b_j(t)e^{i\beta t}dt):(1jn1),`$ (4.6)
$`U_0(\beta )`$ $`=`$ $`:\mathrm{exp}({\displaystyle _{\mathrm{}}^{\mathrm{}}}b_1^{}(t)e^{i\beta t}dt):,`$ (4.7)
$`U_n(\beta )`$ $`=`$ $`:\mathrm{exp}({\displaystyle _{\mathrm{}}^{\mathrm{}}}b_{n1}^{}(t)e^{i\beta t}dt):.`$ (4.8)
The bosonization of the type-I vertex operator is given by
$`\mathrm{\Phi }_j(\beta )`$ $`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\alpha _1}{2\pi i}}\mathrm{}{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\alpha _j}{2\pi i}}{\displaystyle \frac{e^{\frac{n}{2\xi }(\alpha _j\beta )}U_0(\beta )U_1(\alpha _1)\mathrm{}U_j(\alpha _j)}{_{k=1}^j\mathrm{sh}\left(\frac{n}{2\xi }(\alpha _{k1}\alpha _k\frac{\pi i}{n})\right)}}`$
$`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\alpha _1}{2\pi i}}\mathrm{}{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\alpha _j}{2\pi i}}{\displaystyle \frac{e^{\frac{n}{2\xi }(\alpha _j\beta )}U_j(\alpha _j)U_{j1}(\alpha _{j1})\mathrm{}U_0(\beta )}{_{k=1}^j\mathrm{sh}\left(\frac{n}{2\xi }(\alpha _k\alpha _{k1}\frac{\pi i}{n})\right)}},(0jn1)`$
where $`\alpha _0=\beta `$.
The bosonization of the dual type-I vertex operator is given by
$`\mathrm{\Phi }_j^{}(\beta )`$ $`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\alpha _{j+1}}{2\pi i}}\mathrm{}{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\alpha _{n1}}{2\pi i}}{\displaystyle \frac{e^{\frac{n}{2\xi }(\alpha _{j+1}\beta )}U_{j+1}(\alpha _{j+1})\mathrm{}U_{n1}(\alpha _{n1})U_n(\beta )}{_{k=j+1}^{n1}\mathrm{sh}\left(\frac{n}{2\xi }(\alpha _k\alpha _{k+1}\frac{\pi i}{n})\right)}},`$
$`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\alpha _{j+1}}{2\pi i}}\mathrm{}{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\alpha _{n1}}{2\pi i}}{\displaystyle \frac{e^{\frac{n}{2\xi }(\alpha _{j+1}\beta )}U_n(\beta )U_{n1}(\alpha _{n1})\mathrm{}U_{j+1}(\alpha _{j+1})}{_{k=j+1}^{n1}\mathrm{sh}\left(\frac{n}{2\xi }(\alpha _{k+1}\alpha _k\frac{\pi i}{n})\right)}},(0jn1),`$
where $`\alpha _n=\beta `$.
T.Miwa and Y.Takeyama gave the bosonization of some vertex operator, to construct the solutions of the critical quantum Knizhnik-Zamolodchikov equations at Level $`0`$. The calculation of $`S`$-matrix of the critical $`A_{n1}^{(1)}`$ chain teaches us that their vertex operators are the dual type-II vertex operators of the present model.
## 5 Proof
In this section we prove that the bosonizations of the vertex operators satisfy the commutation relations (3.1), (3.2), (3.4), (3.12) and the inversion relation (3.14). For convenience we list below formulae for the contractions of the basic operators.
$`U_j(\beta _1)U_k(\beta _2)=U_k(\beta _2)U_j(\beta _1),(|jk|2).`$ (5.1)
$`U_j(\beta _1)U_j(\beta _2)=U_j(\beta _2)U_j(\beta _1)r(\beta _1\beta _2),(j=0,n),`$ (5.2)
$`r(\beta )={\displaystyle \frac{S_2(i\alpha |\frac{2\pi }{n}\xi ,2\pi )S_2(i\alpha +\frac{2\pi }{n}|\frac{2\pi }{n}\xi ,2\pi )}{S_2(i\alpha |\frac{2\pi }{n}\xi ,2\pi )S_2(i\alpha +\frac{2\pi }{n}|\frac{2\pi }{n}\xi ,2\pi )}},`$ (5.3)
$`U_0(\beta _1)U_{n1}(\beta _2)=r^{}(\beta _1\beta _2)U_{n1}(\beta _2)U_0(\beta _1),`$ (5.4)
$`r^{}(\beta )={\displaystyle \frac{S_2(i\alpha +\pi |\frac{2\pi }{n}\xi ,2\pi )S_2(i\alpha +\pi +\frac{2\pi }{n}|\frac{2\pi }{n}\xi ,2\pi )}{S_2(i\alpha +\pi |\frac{2\pi }{n}\xi ,2\pi )S_2(i\alpha +\pi +\frac{2\pi }{n}|\frac{2\pi }{n}\xi ,2\pi )}}.`$ (5.5)
$`U_j(\beta _1)U_j(\beta _2)`$ $`=`$ $`H(\beta _1\beta _2)U_j(\beta _2)U_j(\beta _1),(1jn1),`$ (5.6)
$`H(\beta )`$ $`=`$ $`{\displaystyle \frac{\mathrm{sh}\left(\frac{n}{2\xi }(\beta +\frac{2\pi i}{n})\right)}{\mathrm{sh}\left(\frac{n}{2\xi }(\beta +\frac{2\pi i}{n})\right)}},`$ (5.7)
$`U_j(\beta _1)U_{j1}(\beta _2)`$ $`=`$ $`I(\beta _1\beta _2)U_{j1}(\beta _2)U_j(\beta _1),(1jn),`$ (5.8)
$`I(\beta )`$ $`=`$ $`{\displaystyle \frac{\mathrm{sh}\left(\frac{n}{2\xi }\left(\beta \frac{\pi i}{n}\right)\right)}{\mathrm{sh}\left(\frac{n}{2\xi }\left(\beta \frac{\pi i}{n}\right)\right)}}.`$ (5.9)
We are to prove the various properties of the vertex operators along the line of the papers . We are to prove the commutation relation of the type-I vertex operators (3.1). Consider the integral of the form :
$`{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑\alpha _1𝑑\alpha _2U_j(\alpha _1)U_j(\alpha _2)F(\alpha _1,\alpha _2).`$ (5.10)
Due to the commutation relation of $`U_j(\alpha )`$, the above integral equals to
$`{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑\alpha _1𝑑\alpha _2U_j(\alpha _1)U_j(\alpha _2)F(\alpha _2,\alpha _1)H(\alpha _2\alpha _1)`$ (5.11)
Observing this we define ’weak equality’ in the following sense. We say that the fuctions $`G_1(\alpha _1,\alpha _2)`$ and $`G_2(\alpha _1,\alpha _2)`$ are equal in weak sense if
$`G_1(\alpha _1,\alpha _2)+H(\alpha _2\alpha _1)G_1(\alpha _2,\alpha _1)=G_2(\alpha _1,\alpha _2)+H(\alpha _2\alpha _1)G_2(\alpha _2,\alpha _1).`$ (5.12)
We write
$`G_1(\alpha _1,\alpha _2)G_2(\alpha _1,\alpha _2).`$ (5.13)
To prove the commutation relations (3.1) and (3.2) it is enough to prove the equalities of the integrand parts in weakly sense.
Firsi we will show
$`\mathrm{\Phi }_\mu (\alpha _0)\mathrm{\Phi }_\mu (\alpha _0^{})=r(\alpha _0\alpha _0^{})\mathrm{\Phi }_\mu (\alpha _0^{})\mathrm{\Phi }_\mu (\alpha _0)(0\mu n1).`$ (5.14)
In what follows we use the notations :
$`b(\alpha )={\displaystyle \frac{\mathrm{sh}\rho \alpha }{\mathrm{sh}\rho (\alpha +\frac{2\pi i}{n})}},c(\alpha )={\displaystyle \frac{\mathrm{sh}\rho \frac{2\pi i}{n}}{\mathrm{sh}\rho (\alpha +\frac{2\pi i}{n})}},d(\alpha )={\displaystyle \frac{e^{\rho \alpha }}{\mathrm{sh}\rho (\alpha \frac{\pi i}{n})}},\rho ={\displaystyle \frac{n}{2\xi }}.`$ (5.15)
For $`\mu =0`$ case it is just the commutation relation of $`U_0(\alpha )`$. For $`\mu 1`$ case we will show that by induction. By using the commutation relations of the basic operators, we can rearrange the operator part as
$`U_0(\alpha _0)U_0(\alpha _0^{})U_1(\alpha _1)U_1(\alpha _1^{})\mathrm{}U_\mu (\alpha _\mu )U_\mu (\alpha _\mu ^{}).`$ (5.16)
The integrand function of (LHS) of (5.14) is given by
$`L_\mu (\alpha _0\alpha _0^{}\mathrm{}\alpha _\mu \alpha _\mu ^{})={\displaystyle \underset{k=1}{\overset{\mu }{}}}d(\alpha _k\alpha _{k1}){\displaystyle \underset{k=1}{\overset{\mu }{}}}d(\alpha _k^{}\alpha _{k1}^{}){\displaystyle \underset{k=1}{\overset{\mu }{}}}I(\alpha _k\alpha _{k1}^{})`$ (5.17)
The integrand of (RHS) is given by the exchange of variables $`\alpha _0\alpha _0^{}`$ of (LHS).
$`R_\mu (\alpha _0\alpha _0^{}\mathrm{}\alpha _\mu \alpha _\mu ^{})=L_\mu (\alpha _0^{}\alpha _0\mathrm{}\alpha _\mu \alpha _\mu ^{}).`$ (5.18)
For $`\mu =1`$ case the weakely identity for variables $`\alpha _1,\alpha _1^{}`$ can be shown by exact calculation.
$`L_1(\alpha _0\alpha _0^{}\alpha _1\alpha _1^{})R_1(\alpha _0\alpha _0^{}\alpha _1\alpha _1^{}).`$ (5.19)
Because the integrand function splits into two parts :
$`L_\mu (\alpha _0\alpha _0^{}\alpha _1\alpha _1^{}\mathrm{}\alpha _\mu \alpha _\mu ^{})=L_1(\alpha _0\alpha _0^{}\alpha _1\alpha _1^{})L_{\mu 1}(\alpha _1\alpha _1^{}\alpha _2\alpha _2^{}\mathrm{}\alpha _\mu \alpha _\mu ^{}),`$ (5.20)
the case $`\mu 2`$ follows from induction.
Next we will show
$`\mathrm{\Phi }_\mu (\alpha _0)\mathrm{\Phi }_\nu (\alpha _0^{})`$ (5.21)
$`=`$ $`r(\alpha _0\alpha _0^{})\left(b(\alpha _0\alpha _0^{})\mathrm{\Phi }_\nu (\alpha _0^{})\mathrm{\Phi }_\mu (\alpha _0)+e^{\mathrm{sgn}(\nu \mu )\rho (\alpha _0\alpha _0^{})}c(\alpha _0\alpha _0^{})\mathrm{\Phi }_\mu (\alpha _0^{})\mathrm{\Phi }_\nu (\alpha _0)\right).`$
We prove the case $`\nu >\mu `$. The case $`\mu >\nu `$ is similar. For the case $`\nu >\mu =0`$, the equality (5.21) follows from the following integrand equality, which can be derived by direct calculations.
$`d(\alpha _1^{}\alpha _0^{})=b(\alpha _0\alpha _0^{})I(\alpha _1^{}\alpha _0)d(\alpha _1^{}\alpha _0^{})+c(\alpha _0\alpha _0^{})e^{\rho (\alpha _0\alpha _0^{})}d(\alpha _1^{}\alpha _0).`$ (5.22)
For the case $`\nu >\mu 1`$, the equality (5.21) follows from the weak equality with respect to the variables $`(\alpha _1\alpha _1^{}),(\alpha _2\alpha _2^{})\mathrm{}(\alpha _\mu \alpha _\mu ^{})`$ :
$`A_\mu (\alpha _0\alpha _0^{}\mathrm{}\alpha _{\mu +1}\alpha _{\mu +1}^{})+B_\mu (\alpha _0\alpha _0^{}\mathrm{}\alpha _{\mu +1}\alpha _{\mu +1}^{})+C_\mu (\alpha _0\alpha _0^{}\mathrm{}\alpha _{\mu +1}\alpha _{\mu +1}^{})0,`$ (5.23)
where we set
$`A_\mu (\alpha _0\alpha _0^{}\mathrm{}\alpha _{\mu +1}\alpha _{\mu +1}^{})={\displaystyle \underset{k=0}{\overset{\mu 1}{}}}I(\alpha _{k+1}\alpha _k^{}){\displaystyle \underset{k=0}{\overset{\mu 1}{}}}d(\alpha _{k+1}\alpha _k){\displaystyle \underset{k=0}{\overset{\mu }{}}}d(\alpha _{k+1}^{}\alpha _k^{}),`$ (5.24)
$`B_\mu (\alpha _0\alpha _0^{}\mathrm{}\alpha _{\mu +1}\alpha _{\mu +1}^{})`$ $`=`$ $`b(\alpha _0\alpha _0^{})\times I(\alpha _{\mu +1}^{}\alpha _\mu ^{})\left\{{\displaystyle \underset{k=1}{\overset{\mu 1}{}}}d(\alpha _{k+1}\alpha _k^{})\right\}I(\alpha _1\alpha _0)`$
$`\times `$ $`d(\alpha _{\mu +1}^{}\alpha _\mu )\left\{{\displaystyle \underset{k=1}{\overset{\mu 1}{}}}d(\alpha _{k+1}\alpha _k)\right\}d(\alpha _1\alpha _0^{})\left\{{\displaystyle \underset{k=1}{\overset{\mu 1}{}}}d(\alpha _{k+1}^{}\alpha _k^{})\right\}d(\alpha _1^{}\alpha _0),`$
and
$`C_\mu (\alpha _0\alpha _0^{}\mathrm{}\alpha _{\mu +1}\alpha _{\mu +1}^{})`$ $`=`$ $`c(\alpha _0\alpha _0^{})e^{\rho (\alpha _0\alpha _0^{})}\times \left\{{\displaystyle \underset{k=1}{\overset{\mu 1}{}}}I(\alpha _{k+1}\alpha _k^{})\right\}I(\alpha _1\alpha _0)`$
$`\times `$ $`\left\{{\displaystyle \underset{k=1}{\overset{\mu }{}}}d(\alpha _{k+1}^{}\alpha _k^{})\right\}d(\alpha _1^{}\alpha _0)\left\{{\displaystyle \underset{k=1}{\overset{\mu }{}}}d(\alpha _{k+1}\alpha _k)\right\}d(\alpha _1\alpha _0^{}).`$
We are to prove the equation (5.23) by induction of $`\mu `$. Inserting the equation (5.23) for $`\mu 1`$ to $`B_\mu `$, we have we have the equation in weakly sense with respect to the variables $`(\alpha _1\alpha _1^{})\mathrm{}(\alpha _\mu \alpha _\mu ^{})`$ :
$`B_\mu (\alpha _0\alpha _0^{}\mathrm{}\alpha _{\mu +1}\alpha _{\mu +1}^{})A_\mu ^{}(\alpha _0\alpha _0^{}\mathrm{}\alpha _{\mu +1}\alpha _{\mu +1}^{})+C_\mu ^{}(\alpha _0\alpha _0^{}\mathrm{}\alpha _{\mu +1}\alpha _{\mu +1}^{}),`$ (5.27)
where we set
$`A_\mu ^{}(\alpha _0\alpha _0^{}\mathrm{}\alpha _\mu \alpha _{\mu +1}^{})={\displaystyle \frac{b(\alpha _0\alpha _0^{})}{b(\alpha _1^{}\alpha _1)}}\left\{{\displaystyle \underset{k=2}{\overset{\mu }{}}}d(\alpha _{k+1}^{}\alpha _k^{})\right\}d(\alpha _2^{}\alpha _1)d(\alpha _1\alpha _0^{})`$
$`\times \left\{{\displaystyle \underset{k=2}{\overset{\mu 1}{}}}d(\alpha _{k+1}\alpha _k)\right\}d(\alpha _2\alpha _1^{})d(\alpha _1^{}\alpha _0)\left\{{\displaystyle \underset{k=2}{\overset{\mu 1}{}}}I(\alpha _{k+1}\alpha _k^{})\right\}I(\alpha _2\alpha _1)I(\alpha _1\alpha _0),`$ (5.28)
and
$`C_\mu ^{}(\alpha _0\alpha _0^{}\mathrm{}\alpha _{\mu +1}\alpha _{\mu +1}^{})={\displaystyle \frac{b(\alpha _0\alpha _0^{})c(\alpha _1^{}\alpha _1)e^{\rho (\alpha _1^{}\alpha _1)}}{b(\alpha _1^{}\alpha _1)}}`$
$`\times `$ $`\left\{{\displaystyle \underset{k=1}{\overset{\mu }{}}}d(\alpha _{k+1}^{}\alpha _k^{})\right\}d(\alpha _1^{}\alpha _0)\left\{{\displaystyle \underset{k=1}{\overset{\mu 1}{}}}d(\alpha _{k+1}\alpha _k)\right\}d(\alpha _1\alpha _0^{})\left\{{\displaystyle \underset{k=1}{\overset{\mu 1}{}}}I(\alpha _{k+1}\alpha _k^{})\right\}I(\alpha _1\alpha _0).`$
Using the following relation : $`b(\alpha )H(\alpha )=b(\alpha ),`$ we have
$`A_\mu ^{}(\alpha _0\alpha _0^{}\mathrm{}\alpha _{\mu +1}\alpha _{\mu +1}^{})A_\mu ^{\prime \prime }(\alpha _0\alpha _0^{}\mathrm{}\alpha _{\mu +1}\alpha _{\mu +1}^{}),`$ (5.30)
where
$`A_\mu ^{\prime \prime }(\alpha _0\alpha _0^{}\mathrm{}\alpha _{\mu +1}\alpha _{\mu +1}^{})`$
$`=`$ $`{\displaystyle \frac{b(\alpha _0\alpha _0^{})}{b(\alpha _1^{}\alpha _1)}}\left\{{\displaystyle \underset{k=0}{\overset{\mu }{}}}d(\alpha _{k+1}^{}\alpha _k^{})\right\}\left\{{\displaystyle \underset{k=0}{\overset{\mu }{}}}d(\alpha _{k+1}\alpha _k)\right\}\left\{{\displaystyle \underset{k=0}{\overset{\mu 1}{}}}I(\alpha _{k+1}\alpha _k^{})\right\}I(\alpha _1^{}\alpha _0).`$
We have
$`A_\mu +A_\mu ^{\prime \prime }`$ $``$ $`{\displaystyle \frac{\mathrm{sh}\left(\rho (\alpha _0^{}\alpha _1^{}\frac{\pi i}{n})\right)\mathrm{sh}\left(\rho (\alpha _0\alpha _1+\frac{\pi i}{n})\right)\mathrm{sh}\left(\rho (\alpha _0+\alpha _1\alpha _0^{}\alpha _1^{})\right)\mathrm{sh}\left(\rho (\frac{2\pi i}{n})\right)}{\mathrm{sh}\left(\rho (\alpha _0^{}\alpha _1\frac{\pi i}{n})\right)\mathrm{sh}\left(\rho (\alpha _1^{}\alpha _0+\frac{2\pi i}{n})\right)\mathrm{sh}\left(\rho (\alpha _0\alpha _1^{}\frac{\pi i}{n})\right)\mathrm{sh}\left(\rho (\alpha _0\alpha _0^{}+\frac{2\pi i}{n})\right)}}`$ (5.32)
$`\times `$ $`{\displaystyle \frac{1}{b(\alpha _1^{}\alpha _1)}}\left\{{\displaystyle \underset{k=0}{\overset{\mu }{}}}d(\alpha _{k+1}^{}\alpha _k^{})\right\}\left\{{\displaystyle \underset{k=0}{\overset{\mu 1}{}}}d(\alpha _{k+1}\alpha _k)\right\}\left\{{\displaystyle \underset{k=1}{\overset{\mu 1}{}}}I(\alpha _{k+1}\alpha _k^{})\right\},`$
where we have used the relation :
$`I(\alpha _1\alpha _0^{})b(\alpha _1^{}\alpha _1)I(\alpha _1^{}\alpha _0)b(\alpha _0\alpha _0^{})`$
$`=`$ $`{\displaystyle \frac{\mathrm{sh}\left(\rho (\alpha _0^{}\alpha _1^{}\frac{\pi i}{n})\right)\mathrm{sh}\left(\rho (\alpha _0\alpha _1+\frac{\pi i}{n})\right)\mathrm{sh}\left(\rho (\alpha _0+\alpha _1\alpha _0^{}\alpha _1^{})\right)\mathrm{sh}\left(\rho (\frac{2\pi i}{n})\right)}{\mathrm{sh}\left(\rho (\alpha _0^{}\alpha _1\frac{\pi i}{n})\right)\mathrm{sh}\left(\rho (\alpha _1^{}\alpha _0+\frac{2\pi i}{n})\right)\mathrm{sh}\left(\rho (\alpha _0\alpha _1^{}\frac{\pi i}{n})\right)\mathrm{sh}\left(\rho (\alpha _0\alpha _0^{}+\frac{2\pi i}{n})\right)}}.`$
We have
$`C_\mu +C_\mu ^{}=b(\alpha _0\alpha _0^{}){\displaystyle \frac{\mathrm{sh}\left(\rho (\frac{2\pi i}{n})\right)\mathrm{sh}(\rho (\alpha _0+\alpha _1\alpha _0^{}\alpha _1^{})))}{\mathrm{sh}\left(\rho (\alpha _1^{}\alpha _1)\right)\mathrm{sh}(\rho (\alpha _0\alpha _0^{})))}}`$ (5.34)
$`\left\{{\displaystyle \underset{k=1}{\overset{\mu }{}}}d(\alpha _{k+1}^{}\alpha _k^{})\right\}d(\alpha _1^{}\alpha _0)\left\{{\displaystyle \underset{k=1}{\overset{\mu 1}{}}}d(\alpha _{k+1}\alpha _k)\right\}d(\alpha _1\alpha _0^{})\left\{{\displaystyle \underset{k=1}{\overset{\mu 1}{}}}I(\alpha _{k+1}\alpha _k^{})\right\}I(\alpha _1\alpha _0),`$
where we have used the relation :
$`{\displaystyle \frac{c(\alpha _1^{}\alpha _1)}{b(\alpha _1^{}\alpha _1)}}e^{\rho (\alpha _1^{}\alpha _1)}{\displaystyle \frac{c(\alpha _0\alpha _0^{})}{b(\alpha _0\alpha _0^{})}}e^{\rho (\alpha _0\alpha _0^{})}={\displaystyle \frac{\mathrm{sh}\left(\rho (\frac{2\pi i}{n})\right)\mathrm{sh}(\rho (\alpha _0+\alpha _1\alpha _0^{}\alpha _1^{})))}{\mathrm{sh}\left(\rho (\alpha _1^{}\alpha _1)\right)\mathrm{sh}(\rho (\alpha _0\alpha _0^{})))}}.`$ (5.35)
We arrive at
$`A_\mu +B_\mu +C_\mu A_\mu +A_\mu ^{\prime \prime }+C_\mu +C_\mu ^{}0.`$ (5.36)
Now we have proved the commutation relation (3.1). As the same arguments as the above we can show the commutation relations (3.2), (3.4) and (3.12).
Next we prove the inversion relations.
The bosonization of the type I vertex operator is deformed as
$`\mathrm{\Phi }_j(\beta )`$ $`=`$ $`\left({\displaystyle \frac{i}{\pi }}\right)^je^{j\frac{\xi 1}{\xi }(\gamma +\mathrm{log}\frac{2\pi \xi }{n})}{\displaystyle _{C_j}}{\displaystyle \frac{d\alpha _j}{2\pi i}}\mathrm{}{\displaystyle _{C_1}}{\displaystyle \frac{d\alpha _1}{2\pi i}}e^{\frac{n}{2\xi }(\alpha _j\beta )}`$
$`\times `$ $`:U_0(\beta )\mathrm{}U_j(\alpha _j):{\displaystyle \underset{k=1}{\overset{j}{}}}\mathrm{\Gamma }({\displaystyle \frac{n}{2\pi \xi }}i(\alpha _k\alpha _{k1})+{\displaystyle \frac{1}{2\xi }})\mathrm{\Gamma }({\displaystyle \frac{n}{2\pi \xi }}i(\alpha _k\alpha _{k1})+{\displaystyle \frac{1}{2\xi }}),`$
where $`\alpha _0=\beta `$. Here the contour $`C_k,(k=1,\mathrm{},j)`$ is taken $`(\mathrm{},\mathrm{})`$ except that the poles
$`\alpha _k\alpha _{k1}={\displaystyle \frac{\pi i}{n}}+{\displaystyle \frac{2\pi \xi i}{n}}l,(l0),`$ (5.38)
of $`\mathrm{\Gamma }\left(\frac{n}{2\pi \xi }i(\alpha _k\alpha _{k1})+\frac{1}{2\xi }\right)`$ are above $`C_k`$ and the poles
$`\alpha _k\alpha _{k1}={\displaystyle \frac{\pi i}{n}}{\displaystyle \frac{2\pi \xi i}{n}}l,(l0),`$ (5.39)
of $`\mathrm{\Gamma }\left(\frac{n}{2\pi \xi }i(\alpha _k\alpha _{k1})+\frac{1}{2\xi }\right)`$ are below $`C_k`$.
The bosonization of the dual type-I vertex operator is deformed as
$`\mathrm{\Phi }_j^{}(\beta )`$ $`=`$ $`\left({\displaystyle \frac{i}{\pi }}\right)^{nj1}e^{(nj1)\frac{\xi 1}{\xi }(\gamma +\mathrm{log}\frac{2\pi \xi }{n})}{\displaystyle _{C_{j+1}^{}}}{\displaystyle \frac{d\alpha _{j+1}}{2\pi i}}\mathrm{}{\displaystyle _{C_{n1}^{}}}{\displaystyle \frac{d\alpha _{n1}}{2\pi i}}e^{\frac{n}{2\xi }(\alpha _{j+1}\beta )}`$
$`\times `$ $`:U_{j+1}(\alpha _{j+1})\mathrm{}U_n(\beta ):{\displaystyle \underset{k=j+1}{\overset{n1}{}}}\mathrm{\Gamma }({\displaystyle \frac{n}{2\pi \xi }}i(\alpha _{k+1}\alpha _k)+{\displaystyle \frac{1}{2\xi }})\mathrm{\Gamma }({\displaystyle \frac{n}{2\pi \xi }}i(\alpha _{k+1}\alpha _k)+{\displaystyle \frac{1}{2\xi }}),`$
where $`\alpha _n=\beta `$. Here the contour $`C_k^{},(k=j+1,\mathrm{},n1)`$ is taken $`(\mathrm{},\mathrm{})`$ except that the poles
$`\alpha _k\alpha _{k+1}={\displaystyle \frac{\pi i}{n}}+{\displaystyle \frac{2\pi \xi i}{n}}l,(l0),`$ (5.41)
of $`\mathrm{\Gamma }\left(\frac{n}{2\pi \xi }i(\alpha _{k+1}\alpha _k)+\frac{1}{2\xi }\right)`$ are above $`C_k^{}`$ and the poles
$`\alpha _k\alpha _{k+1}={\displaystyle \frac{\pi i}{n}}{\displaystyle \frac{2\pi \xi i}{n}}l,(l0),`$ (5.42)
of $`\mathrm{\Gamma }\left(\frac{n}{2\pi \xi }i(\alpha _{k+1}\alpha _k)+\frac{1}{2\xi }\right)`$ are below $`C_k^{}`$.
Let us prove the relation (3.14). For $`j_1<j_2`$ we have
$`\mathrm{\Phi }_{j_1}(\beta _1)\mathrm{\Phi }_{j_2}^{}(\beta _2)=r^{}(\beta _1\beta _2)\mathrm{\Phi }_{j_2}^{}(\beta _2)\mathrm{\Phi }_{j_1}(\beta _1).`$ (5.43)
As $`\beta _2\beta _1+\pi i`$, $`r^{}(\beta _1\beta _2)`$ becomes zero.
Next we prove the case $`j_1=j_2`$. We have
$`\mathrm{\Phi }_j(\beta _1)\mathrm{\Phi }_j^{}(\beta _2)`$ $`=`$ $`r^{}(\beta _1\beta _2){\displaystyle _{C_1}}{\displaystyle \frac{d\alpha _1}{2\pi i}}\mathrm{}{\displaystyle _{C_j}}{\displaystyle \frac{d\alpha _j}{2\pi i}}{\displaystyle _{C_{j+1}^{}}}{\displaystyle \frac{d\alpha _{j+1}}{2\pi i}}\mathrm{}{\displaystyle _{C_{n1}^{}}}{\displaystyle \frac{d\alpha _{n1}}{2\pi i}}`$
$`\times `$ $`U_n(\beta _2)U_{n1}(\alpha _{n1})\mathrm{}U_1(\alpha _1)U_0(\beta _1)`$
$`\times `$ $`e^{\frac{n}{2\xi }(\alpha _j+\alpha _{j+1}\beta _1\beta _2)}\mathrm{sh}\left({\displaystyle \frac{n}{2\xi }}(\alpha _j\alpha _{j+1}{\displaystyle \frac{\pi i}{n}})\right){\displaystyle \underset{k=1}{\overset{n}{}}}{\displaystyle \frac{1}{\mathrm{sh}\left(\frac{n}{2\xi }(\alpha _k\alpha _{k1}\frac{\pi i}{n})\right)}}.`$
When we take the limit $`\beta _2\beta _1+\pi i`$, the contour is pinched but the function $`r^{}(\beta _1\beta _2)`$ has a zero. Therefore the limit is evaluated by successively taking the residues at $`\alpha _k=\alpha _{k1}+\frac{\pi i}{n}`$ for $`k=1,\mathrm{},j`$, and $`\alpha _k=\alpha _{k+1}\frac{\pi i}{n}`$ for $`k=j+1,\mathrm{},n1`$, successively. The following relation is usefull.
$`U_n(\beta +\pi i)U_{n1}\left(\beta +{\displaystyle \frac{n1}{n}}\pi i\right)\mathrm{}U_1\left(\beta +{\displaystyle \frac{\pi i}{n}}\right)U_0(\beta )=e^{\frac{\xi 1}{\xi }(\gamma +\mathrm{log}\frac{2\pi \xi }{n})n}\mathrm{\Gamma }(1/\xi )^n.`$ (5.45)
Now we have proved the inversion relation (3.14).
## 6 Correlation functions
In this section we derive a solution of the system of difference equations (1.3) and (1.4), algebraically, and obtain an integral representation of it.
Let us introduce the degree operator $`D`$ on the Fock space :
$`Db_j(t)|0=tb_j(t)|0,(t>0).`$ (6.1)
We have
$`e^{\lambda D}b(t)e^{\lambda D}=e^{\lambda t}b(t).`$ (6.2)
Therefore the degree operator $`D`$ has the homogeneity condition.
$`e^{\lambda D}U_j(\beta )e^{\lambda D}=U_j(\beta +i\lambda ).`$ (6.3)
The vertex operator and the degree operator enjoy the homogeneity property.
$`e^{\lambda D}\mathrm{\Phi }_j(\beta )e^{\lambda D}=\mathrm{\Phi }_j(\beta +i\lambda ),e^{\lambda D}\mathrm{\Phi }_j^{}(\beta )e^{\lambda D}=\mathrm{\Phi }_j^{}(\beta +i\lambda ).`$ (6.4)
Now let us consider the trace functions for $`\lambda >0`$ defined by
$`G(\beta _1\mathrm{}\beta _N|\beta _{N+1}\mathrm{}\beta _{2N})_{ϵ_1\mathrm{}ϵ_{2N}}`$ (6.5)
$`=`$ $`{\displaystyle \frac{\mathrm{tr}_{}\left(e^{\lambda D}\mathrm{\Phi }_{ϵ_1}^{}(\beta _1)\mathrm{}\mathrm{\Phi }_{ϵ_N}^{}(\beta _N)\mathrm{\Phi }_{ϵ_{N+1}}(\beta _{N+1})\mathrm{}\mathrm{\Phi }_{ϵ_{2N}}(\beta _{2N})\right)}{\mathrm{tr}_{}\left(e^{\lambda D}\right)}},`$
where the space $``$ is the Fock space of the free bosons.
By using the homogeneity condition (6.4) and the commutation relations (3.1), (3.2), (3.12), it is shown that the above trace function satisfies the desired difference equations (1.3), (1.4). The inversion relation (3.14) means the restriction equation (1.5).
Using the free boson realizations, we shall treat the trace of the form : $`\mathrm{tr}_{}(e^{\lambda D}𝒪)`$. The calculation is simplified by the technique of Clavelli and Shapiro . Their prescription is as follows. Introduce a copy of the bosons $`a(t)(t)`$ satisfying the relation $`[a(t),b(t^{})]=0`$ and the same commutation relation as the $`b(t)`$. Let
$`\stackrel{~}{b}(t)={\displaystyle \frac{b(t)}{1e^{\lambda t}}}+a(t),\stackrel{~}{b}(t)={\displaystyle \frac{a(t)}{e^{\lambda t}1}}+b(t),(t>0).`$ (6.6)
For a linear operator $`𝒪`$ on the Fock space $`_b=[b(t)]`$, let $`\stackrel{~}{𝒪}`$ be the operator on $`_b_a,(_a=[a(t)])`$ obtaind by substituting $`\stackrel{~}{b}(t)`$ for $`b(t)`$. We have then
$`{\displaystyle \frac{\mathrm{tr}__b\left(e^{\lambda D}𝒪\right)}{\mathrm{tr}__b(e^{\lambda D})}}=\stackrel{~}{0}|\stackrel{~}{𝒪}|\stackrel{~}{0},`$ (6.7)
where the left hand side denotes the usual expectation value with respect to the Fock vacuum $`|\stackrel{~}{0}=|0|0`$, $`\stackrel{~}{0}|\stackrel{~}{0}`$. In what follows we use the following abbreviation :
$`𝒪_\lambda ={\displaystyle \frac{\mathrm{tr}_{}\left(e^{\lambda D}𝒪\right)}{\mathrm{tr}_{}(e^{\lambda D})}}.`$ (6.8)
We have the following.
$`b_j(t)b_k(t^{})_\lambda ={\displaystyle \frac{e^{\lambda t}}{e^{\lambda t}1}}[b_j(t),b_k(t^{})].`$ (6.9)
We have the following formula usefull to evaluate the function (6.5).
$`{\displaystyle \frac{\mathrm{tr}_H(e^{\lambda D}:e^{_{\mathrm{}}^{\mathrm{}}c(t)e^{i\beta _1t}𝑑t}::e^{_{\mathrm{}}^{\mathrm{}}d(t)e^{i\beta _1t}𝑑t}:)}{\mathrm{tr}_H(e^{\lambda D})}}=\mathrm{exp}\left({\displaystyle _0^{\mathrm{}}}A(t){\displaystyle \frac{\mathrm{ch}(i(\beta _1\beta _2)+\frac{\lambda }{2})t}{\mathrm{sh}\frac{\lambda t}{2}}}𝑑t\right),`$ (6.10)
where $`c(t)`$ and $`d(t)`$ are bosons satisfying $`[c(t),d(t^{})]=A(t)\delta (t+t^{})`$ and $`A(t)=A(t)`$. In order to understand an integral of the right hand side, see the Appendix B.
The basic trace functions are evaluated as following.
$`U_0(\alpha _1)U_0(\alpha _2)_\lambda =U_n(\alpha _1)U_n(\alpha _2)_\lambda =\mathrm{Const}.E_\lambda (\alpha _1\alpha _2),`$ (6.11)
$`U_0(\alpha _1)U_n(\alpha _2)_\lambda =U_n(\alpha _1)U_0(\alpha _2)_\lambda =\mathrm{Const}.E_\lambda ^{}(\alpha _1\alpha _2).`$ (6.12)
Here we set
$`E_\lambda (\alpha )={\displaystyle \frac{S_3(i\alpha +\frac{2\pi }{n})S_3(i\alpha +\frac{2\pi }{n}+\lambda )}{S_3(i\alpha +\frac{2\pi }{n}\xi )S_3(i\alpha +\frac{2\pi }{n}\xi +\lambda )}}`$ (6.13)
$`E_\lambda ^{}(\alpha )={\displaystyle \frac{S_3(i\alpha +\pi )S_3(i\alpha +\lambda +\pi )}{S_3(i\alpha +\frac{2\pi }{n}+\pi )S_3(i\alpha +\lambda +\pi +\frac{2\pi }{n})}},`$ (6.14)
where
$`S_3(\beta )=S_3\left(\beta |2\pi ,{\displaystyle \frac{2\pi }{n}}\xi ,\lambda \right).`$ (6.15)
$`U_j(\alpha _1)U_{j1}(\alpha _2)_\lambda =U_{j1}(\alpha _1)U_j(\alpha _2)_\lambda `$ (6.16)
$`=`$ $`\mathrm{Const}.\phi (\alpha _1\alpha _2)\times {\displaystyle \frac{1}{\mathrm{sh}\left({\displaystyle \frac{n}{2\xi }}(\alpha _1\alpha _2{\displaystyle \frac{\pi i}{n}}+\lambda i)\right)}},`$
$`U_j(\alpha _1)U_j(\alpha _2)_\lambda =\mathrm{Const}.\psi (\alpha _1\alpha _2)\times `$
$`\times `$ $`\mathrm{sh}\left({\displaystyle \frac{n}{2\xi }}\alpha \right)\mathrm{sh}\left({\displaystyle \frac{n}{2\xi }}(\alpha +{\displaystyle \frac{2\pi i}{n}})\right)\mathrm{sh}\left({\displaystyle \frac{\pi }{\lambda }}(\alpha +{\displaystyle \frac{2\pi i}{n}})\right)\mathrm{sh}\left({\displaystyle \frac{\pi }{\lambda }}(\alpha +{\displaystyle \frac{2\pi i}{n}})\right).`$
Here we set
$`\phi (\alpha )`$ $`=`$ $`{\displaystyle \frac{1}{S_2\left(i\alpha +{\displaystyle \frac{\pi }{n}}|\lambda ,{\displaystyle \frac{2\pi }{n}}\xi \right)S_2\left(i\alpha +{\displaystyle \frac{\pi }{n}}|\lambda ,{\displaystyle \frac{2\pi }{n}}\xi \right)}},`$ (6.18)
$`\psi (\alpha )`$ $`=`$ $`{\displaystyle \frac{1}{S_2\left(i\alpha {\displaystyle \frac{2\pi }{n}}|\lambda ,{\displaystyle \frac{2\pi }{n}}\xi \right)S_2\left(i\alpha {\displaystyle \frac{2\pi }{n}}|\lambda ,{\displaystyle \frac{2\pi }{n}}\xi \right)}}.`$ (6.19)
The functions $`\phi (\alpha )`$ and $`\psi (\alpha )`$ become the integral kernel of the correlation functions.
The function $`\phi (\alpha )`$ has poles at
$`\alpha =\pm i\left(n_1\lambda +n_2{\displaystyle \frac{2\pi }{n}}\xi +{\displaystyle \frac{\pi }{n}}\right),(n_1,n_20).`$ (6.20)
The function $`\psi (\alpha )`$ has poles at
$`\alpha =\pm i\left(n_1\lambda +n_2{\displaystyle \frac{2\pi }{n}}\xi {\displaystyle \frac{2\pi }{n}}\right),(n_1,n_20).`$ (6.21)
The trace of the vertex operators (6.5) is evaluated by applying the Wick’s theorem.
The one-point correlation functions are evaluated as follows. In what follows we set $`\rho =\frac{n}{2\xi }`$.
$`G(\beta _1|\beta _1^{})_{ϵ,ϵ}`$ $`=`$ $`E_\lambda ^{}(\beta _1\beta _1^{}){\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\alpha _1}{2\pi i}}\mathrm{}{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\alpha _{n1}}{2\pi i}}{\displaystyle \underset{k=1}{\overset{n}{}}}\phi (\alpha _k\alpha _{k1})e^{\rho (\alpha _ϵ+\alpha _{ϵ+1}\beta _1\beta _1^{})}`$
$`\times `$ $`\mathrm{sh}\left(\rho (\alpha _{ϵ+1}\alpha _ϵ{\displaystyle \frac{\pi i}{n}})\right){\displaystyle \underset{k=1}{\overset{n}{}}}{\displaystyle \frac{1}{\mathrm{sh}\left(\rho (\alpha _k\alpha _{k1}{\displaystyle \frac{\pi i}{n}})\right)\mathrm{sh}\left(\rho (\alpha _k\alpha _{k1}{\displaystyle \frac{\pi i}{n}}+\lambda i)\right)}},`$
where $`\alpha _n=\beta _1,\alpha _0=\beta _1^{}`$. Here we omit an irrelevant constant factor.
The $`N`$ point correlation functions are evaluated as follows.
$`G(\beta _1\mathrm{}\beta _N|\beta _N^{}\mathrm{}\beta _1^{})_{ϵ_1\mathrm{}ϵ_N,ϵ_N,\mathrm{},ϵ_1}=E(\beta _1\mathrm{}\beta _N|\beta _N^{}\mathrm{}\beta _1^{}){\displaystyle \underset{j,r}{}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\alpha _{j,r}}{2\pi i}}I_{ϵ_1\mathrm{}ϵ_N}(\left\{\alpha _{j,r}\right\}).`$ (6.23)
We associate the variables $`\alpha _{j,r},(1rN,0jϵ_r)`$ to the basic operator $`U_j(\alpha _{j,r})`$ contained in the vertex operators $`\mathrm{\Phi }_{ϵ_r}(\beta _r^{})`$ and the variables $`\alpha _{j,r},(1rN,ϵ_r+1jn)`$ to the basic operator $`U_j(\alpha _{j,r})`$ contained in the vertex operators $`\mathrm{\Phi }_{ϵ_r}^{}(\beta _r)`$.
We set
$`\alpha _{n,r}=\beta _r,`$ $`\alpha _{0,r}=\beta _r^{},`$ (6.24)
$`𝒩_j^{}=\{k|ϵ_kj1\},`$ $`𝒩_j=\{k|ϵ_kj\}.`$ (6.25)
The function $`E(\beta _1\mathrm{}\beta _N|\beta _N^{}\mathrm{}\beta _1^{})`$ is given by
$`E(\beta _1\mathrm{}\beta _N|\beta _N^{}\mathrm{}\beta _1^{})`$ (6.26)
$`=`$ $`e^{\rho (\beta _1+\mathrm{}+\beta _N+\beta _1^{}+\mathrm{}+\beta _N^{})}{\displaystyle \underset{1j<kN}{}}E_\lambda (\beta _j\beta _k){\displaystyle \underset{1j<kN}{}}E_\lambda (\beta _k^{}\beta _j^{}){\displaystyle \underset{j,k=1}{\overset{N}{}}}E_\lambda ^{}(\beta _j\beta _k^{}).`$
The integrand is given by
$`I_{ϵ_1\mathrm{}ϵ_N}(\{\alpha _{j,r}\})=K(\{\alpha _{j,r}\})g(\{\alpha _{j,r}\})h(\{\alpha _{j,r}\}).`$ (6.27)
Here the integral kernel is given by
$`K(\{\alpha _{j,r}\})={\displaystyle \underset{j=1}{\overset{n1}{}}}{\displaystyle \underset{\genfrac{}{}{0pt}{}{r,s=1}{r>s}}{\overset{N}{}}}\psi (\alpha _{j,r}\alpha _{j,s})\times {\displaystyle \underset{j=1}{\overset{n1}{}}}{\displaystyle \underset{k,l=1}{\overset{N}{}}}\left\{{\displaystyle \underset{\genfrac{}{}{0pt}{}{kN_j^{}}{lN_{j1}}}{}}\phi (\alpha _{j,k}\alpha _{j,l}){\displaystyle \underset{\genfrac{}{}{0pt}{}{kN_{j1}^{}}{lN_j}}{}}\phi (\alpha _{j,k}\alpha _{j,l})\right\}`$
$`\times {\displaystyle \underset{j=1}{\overset{n1}{}}}{\displaystyle \underset{k,l=1}{\overset{N}{}}}\left\{{\displaystyle \underset{\genfrac{}{}{0pt}{}{kN_j}{lN_{j1}}}{}}\phi (\alpha _{j,k}\alpha _{j,l}){\displaystyle \underset{\genfrac{}{}{0pt}{}{kN_j^{}}{lN_{j1}^{}}}{}}\phi (\alpha _{j,k}\alpha _{j,l})\right\}^2.`$ (6.28)
Here we set
$`h(\{\alpha _{j,r}\})={\displaystyle \underset{j=1}{\overset{n1}{}}}{\displaystyle \underset{\genfrac{}{}{0pt}{}{r,s=1}{r>s}}{\overset{N}{}}}\left\{\mathrm{sh}\left({\displaystyle \frac{\pi }{\lambda }}(\alpha _{j,r}\alpha _{j,s}+{\displaystyle \frac{2\pi i}{n}})\right)\mathrm{sh}\left({\displaystyle \frac{\pi }{\lambda }}(\alpha _{j,s}\alpha _{j,r}+{\displaystyle \frac{2\pi i}{n}})\right)\right\},`$ (6.29)
and
$`g(\{\alpha _{j,r}\})`$
$`=`$ $`{\displaystyle \underset{j=1}{\overset{n1}{}}}{\displaystyle \underset{\genfrac{}{}{0pt}{}{r>s}{sN_j}}{}}\left\{\mathrm{sh}\left(\rho (\alpha _{j,r}\alpha _{j,s})\right)\mathrm{sh}\left(\rho (\alpha _{j,r}\alpha _{j,s}+{\displaystyle \frac{2\pi i}{n}})\right)\right\}`$
$`\times `$ $`{\displaystyle \underset{j=1}{\overset{n1}{}}}{\displaystyle \underset{\genfrac{}{}{0pt}{}{r>s}{sN_j^{}}}{}}\left\{\mathrm{sh}\left(\rho (\alpha _{j,s}\alpha _{j,r})\right)\mathrm{sh}\left(\rho (\alpha _{j,s}\alpha _{j,r}+{\displaystyle \frac{2\pi i}{n}})\right)\right\}`$
$`\times `$ $`{\displaystyle \underset{j=1}{\overset{n}{}}}{\displaystyle \underset{\genfrac{}{}{0pt}{}{rN_j}{sN_{j1}}}{}}\left\{\mathrm{sh}\left(\rho (\alpha _{j,r}\alpha _{j1,s}{\displaystyle \frac{\pi i}{n}}+\lambda i)\right)\mathrm{sh}\left(\rho (\alpha _{j1,r}\alpha _{j,s}{\displaystyle \frac{\pi i}{n}}+\lambda i)\right)\right\}^1`$
$`\times `$ $`{\displaystyle \underset{j=1}{\overset{n}{}}}{\displaystyle \underset{\genfrac{}{}{0pt}{}{rN_j^{}}{sN_{j1}^{}}}{}}\left\{\mathrm{sh}\left(\rho (\alpha _{j,r}\alpha _{j1,s}{\displaystyle \frac{\pi i}{n}}+\lambda i)\right)\mathrm{sh}\left(\rho (\alpha _{j1,r}\alpha _{j,s}{\displaystyle \frac{\pi i}{n}}+\lambda i)\right)\right\}^1`$
$`\times `$ $`{\displaystyle \underset{j=1}{\overset{n}{}}}\left\{{\displaystyle \underset{\genfrac{}{}{0pt}{}{rN_j^{}}{sN_{j1}}}{}}\mathrm{sh}\left(\rho (\alpha _{j,r}\alpha _{j1,s}{\displaystyle \frac{\pi i}{n}}+\lambda i)\right){\displaystyle \underset{\genfrac{}{}{0pt}{}{rN_{j1}^{}}{sN_j}}{}}\mathrm{sh}\left(\rho (\alpha _{j,r}\alpha _{j1,s}{\displaystyle \frac{\pi i}{n}}+\lambda i)\right)\right\}^1`$
$`\times `$ $`{\displaystyle \underset{r=1}{\overset{N}{}}}e^{\rho (\alpha _{ϵ_r}+\alpha _{ϵ_r+1})}{\displaystyle \underset{r=1}{\overset{N}{}}}\mathrm{sh}\left(\rho (\alpha _{ϵ_r+1,r}\alpha _{ϵ_r,r}{\displaystyle \frac{\pi i}{n}})\right){\displaystyle \underset{j=1}{\overset{n}{}}}{\displaystyle \underset{r=1}{\overset{N}{}}}\left\{\mathrm{sh}\left(\rho (\alpha _{j,r}\alpha _{j1,r}{\displaystyle \frac{\pi i}{n}})\right)\right\}^1.`$
Here we omit an irrelevant constant factor.
Acknowledgements. This work was partly supported by Grant-in-Aid for Encouragements for Young Scientists (A) from Japan Society for the Promotion of Science. (11740099)
## Appendix A Multi-Gamma functions
Here we summarize the multiple gamma and the multiple sine functions, following Kurokawa .
Let us set the functions $`\mathrm{\Gamma }_1(x|\omega ),\mathrm{\Gamma }_2(x|\omega _1,\omega _2)`$ and $`\mathrm{\Gamma }_3(x|\omega _1,\omega _2,\omega _3)`$ by
$`\mathrm{log}\mathrm{\Gamma }_1(x|\omega )+\gamma B_{11}(x|\omega )`$ $`=`$ $`{\displaystyle _C}{\displaystyle \frac{dt}{2\pi it}}e^{xt}{\displaystyle \frac{\mathrm{log}(t)}{1e^{\omega t}}},`$ (A.1)
$`\mathrm{log}\mathrm{\Gamma }_2(x|\omega _1,\omega _2){\displaystyle \frac{\gamma }{2}}B_{22}(x|\omega _1,\omega _2)`$ $`=`$ $`{\displaystyle _C}{\displaystyle \frac{dt}{2\pi it}}e^{xt}{\displaystyle \frac{\mathrm{log}(t)}{(1e^{\omega _1t})(1e^{\omega _2t})}},`$ (A.2)
$`\mathrm{log}\mathrm{\Gamma }_3(x|\omega _1,\omega _2,\omega _3)+{\displaystyle \frac{\gamma }{3!}}B_{33}(x|\omega _1,\omega _2,\omega _3)`$ $`=`$ $`{\displaystyle _C}{\displaystyle \frac{dt}{2\pi it}}e^{xt}{\displaystyle \frac{\mathrm{log}(t)}{(1e^{\omega _1t})(1e^{\omega _2t})(1e^{\omega _3t})}},`$ (A.3)
where the functions $`B_{jj}(x)`$ are the multiple Bernoulli polynomials defined by
$`{\displaystyle \frac{t^re^{xt}}{_{j=1}^r(e^{\omega _jt}1)}}={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{t^n}{n!}}B_{r,n}(x|\omega _1\mathrm{}\omega _r),`$ (A.4)
more explicitly
$`B_{11}(x|\omega )`$ $`=`$ $`{\displaystyle \frac{x}{\omega }}{\displaystyle \frac{1}{2}},`$ (A.5)
$`B_{22}(x|\omega )`$ $`=`$ $`{\displaystyle \frac{x^2}{\omega _1\omega _2}}\left({\displaystyle \frac{1}{\omega _1}}+{\displaystyle \frac{1}{\omega _2}}\right)x+{\displaystyle \frac{1}{2}}+{\displaystyle \frac{1}{6}}\left({\displaystyle \frac{\omega _1}{\omega _2}}+{\displaystyle \frac{\omega _2}{\omega _1}}\right).`$ (A.6)
Here $`\gamma `$ is Euler’s constant, $`\gamma =lim_n\mathrm{}(1+\frac{1}{2}+\frac{1}{3}+\mathrm{}+\frac{1}{n}\mathrm{log}n)`$.
Here the contor of integral is given by
Let us set
$`S_1(x|\omega )`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{\Gamma }_1(\omega x|\omega )\mathrm{\Gamma }_1(x|\omega )}},`$ (A.7)
$`S_2(x|\omega _1,\omega _2)`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Gamma }_2(\omega _1+\omega _2x|\omega _1,\omega _2)}{\mathrm{\Gamma }_2(x|\omega _1,\omega _2)}},`$ (A.8)
$`S_3(x|\omega _1,\omega _2,\omega _3)`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{\Gamma }_3(\omega _1+\omega _2+\omega _3x|\omega _1,\omega _2,\omega _3)\mathrm{\Gamma }_3(x|\omega _1,\omega _2,\omega _3)}}`$ (A.9)
We have
$`\mathrm{\Gamma }_1(x|\omega )=e^{(\frac{x}{\omega }\frac{1}{2})\mathrm{log}\omega }{\displaystyle \frac{\mathrm{\Gamma }(x/\omega )}{\sqrt{2\pi }}},S_1(x|\omega )=2\mathrm{s}\mathrm{i}\mathrm{n}(\pi x/\omega ),`$ (A.10)
$`{\displaystyle \frac{\mathrm{\Gamma }_2(x+\omega _1|\omega _1,\omega _2)}{\mathrm{\Gamma }_2(x|\omega _1,\omega _2)}}={\displaystyle \frac{1}{\mathrm{\Gamma }_1(x|\omega _2)}},{\displaystyle \frac{S_2(x+\omega _1|\omega _1,\omega _2)}{S_2(x|\omega _1,\omega _2)}}={\displaystyle \frac{1}{S_1(x|\omega _2)}},{\displaystyle \frac{\mathrm{\Gamma }_1(x+\omega |\omega )}{\mathrm{\Gamma }_1(x|\omega )}}=x.`$ (A.11)
$`{\displaystyle \frac{\mathrm{\Gamma }_3(x+\omega _1|\omega _1,\omega _2,\omega _3}{\mathrm{\Gamma }_3(x|\omega _1,\omega _2,\omega _3)}}={\displaystyle \frac{1}{\mathrm{\Gamma }_2(x|\omega _2,\omega _3)}},{\displaystyle \frac{S_3(x+\omega _1|\omega _1,\omega _2,\omega _3)}{S_3(x|\omega _1,\omega _2,\omega _3)}}={\displaystyle \frac{1}{S_2(x|\omega _2,\omega _3)}}.`$ (A.12)
$`\mathrm{log}S_2(x|\omega _1\omega _2)={\displaystyle _C}{\displaystyle \frac{\mathrm{sh}(x\frac{\omega _1+\omega _2}{2})t}{2\mathrm{s}\mathrm{h}\frac{\omega _1t}{2}\mathrm{sh}\frac{\omega _2t}{2}}}\mathrm{log}(t){\displaystyle \frac{dt}{2\pi it}},(0<\mathrm{Re}x<\omega _1+\omega _2).`$ (A.13)
$`S_2(x|\omega _1\omega _2)={\displaystyle \frac{2\pi }{\sqrt{\omega _1\omega _2}}}x+O(x^2),(x0).`$ (A.14)
$`S_2(x|\omega _1\omega _2)S_2(x|\omega _1\omega _2)=4\mathrm{s}\mathrm{i}\mathrm{n}{\displaystyle \frac{\pi x}{\omega _1}}\mathrm{sin}{\displaystyle \frac{\pi x}{\omega _2}}.`$ (A.15)
## Appendix B Normal Ordering
Here we list the formulas of the form
$`X(\beta _1)Y(\beta _2)=C_{XY}(\beta _1\beta _2):X(\beta _1)X(\beta _2):,`$ (B.1)
where $`X,Y=U_j`$, and $`C_{XY}(\beta )`$ is a meromorphic function on $``$. These formulae follow from the commutation relation of the free bosons. When we compute the contraction of the basic operators, we often encounter an integral
$`{\displaystyle _0^{\mathrm{}}}F(t)𝑑t,`$ (B.2)
which is divergent at $`t=0`$. Here we adopt the following prescription for regularization : it should be understood as the countour integral,
$`{\displaystyle _C}F(t){\displaystyle \frac{\mathrm{log}(t)}{2\pi i}}𝑑t,`$ (B.3)
where the countour $`C`$ is given by
The contractions of the basic operators have the following forms.
$`U_j(\alpha _1)U_j(\alpha _2)=h_{jj}(\alpha _1\alpha _2):U_j(\alpha _1)U_j(\alpha _2):\left(\mathrm{Im}(\alpha _2\alpha _1)<{\displaystyle \frac{2\pi }{n}},j=0,n\right),`$ (B.4)
$`h_{00}(\alpha )=h_{nn}(\alpha )=e^{\gamma \frac{\xi 1}{\xi }\frac{n1}{n}}{\displaystyle \frac{\mathrm{\Gamma }_2(i\alpha +\frac{2\pi }{n}\xi |\frac{2\pi }{n}\xi ,2\pi )\mathrm{\Gamma }_2(i\alpha +2\pi |\frac{2\pi }{n}\xi ,2\pi )}{\mathrm{\Gamma }_2(i\alpha +\frac{2\pi }{n}|\frac{2\pi }{n}\xi ,2\pi )\mathrm{\Gamma }_2(i\alpha +2\pi +\frac{2\pi }{n}\xi \frac{2\pi }{n}|\frac{2\pi }{n}\xi ,2\pi )}},`$ (B.5)
$`U_j(\alpha _1)U_j(\alpha _2)`$ $`=`$ $`h_{jj}(\alpha _1\alpha _2):U_j(\alpha _1)U_j(\alpha _2):\left(\mathrm{Im}(\alpha _2\alpha _1)<{\displaystyle \frac{2\pi }{n}}(\xi 1),1jn1\right),`$ (B.6)
$`h_{jj}(\alpha )`$ $`=`$ $`e^{2\gamma \frac{\xi 1}{\xi }}{\displaystyle \frac{\alpha }{i}}{\displaystyle \frac{\mathrm{\Gamma }_1(i\alpha +\frac{2\pi }{n}\xi \frac{2\pi }{n}|\frac{2\pi }{n}\xi )}{\mathrm{\Gamma }_1(i\alpha +\frac{2\pi }{n}|\frac{2\pi }{n}\xi )}},`$ (B.7)
$`U_j(\alpha _1)U_{j1}(\alpha _2)`$ $`=`$ $`h_{jj1}(\alpha _1\alpha _2):U_{j1}(\alpha _1)U_j(\alpha _2):\left(\mathrm{Im}(\alpha _2\alpha _1)<{\displaystyle \frac{\pi }{n}},1jn\right),`$ (B.8)
$`h_{jj1}(\alpha )`$ $`=`$ $`e^{\gamma \frac{\xi 1}{\xi }}{\displaystyle \frac{\mathrm{\Gamma }_1(i\alpha +\frac{\pi }{n}|\frac{2\pi }{n}\xi )}{\mathrm{\Gamma }_1(i\alpha +\frac{\pi }{n}(2\xi 1)|\frac{2\pi }{n}\xi )}},`$ (B.9)
and
$`U_0(\alpha _1)U_n(\alpha _2)=h_{0n}(\alpha _1\alpha _2):U_0(\alpha _1)U_n(\alpha _2):,(\mathrm{Im}(\alpha _2\alpha _1)<{\displaystyle \frac{2\pi }{n}})`$ (B.10)
$`h_{0n}(\alpha )=e^{\frac{\gamma }{n}\frac{\xi 1}{\xi }}{\displaystyle \frac{\mathrm{\Gamma }_2(i\alpha +\pi +\frac{2\pi }{n}\xi \frac{2\pi }{n}|\frac{2\pi }{n}\xi ,2\pi )\mathrm{\Gamma }_2(i\alpha +\pi +\frac{2\pi }{n}|\frac{2\pi }{n}\xi ,2\pi )}{\mathrm{\Gamma }_2(i\alpha +\pi |\frac{2\pi }{n}\xi ,2\pi )\mathrm{\Gamma }_2(i\alpha +\pi +\frac{2\pi }{n}\xi |\frac{2\pi }{n}\xi ,2\pi )}},`$ (B.11)
|
warning/0004/nucl-th0004062.html
|
ar5iv
|
text
|
# Application of realistic effective interactions to the structure of the Zr isotopes
## I INTRODUCTION
The Zr isotopes undergo a clear and smooth shape transition with increasing neutron number. The isotopes which are displayed in Figure 1(a) span from pure spherical nuclei that can be described in terms of simple shell-model configurations to the strongly deformed nucleus <sup>102</sup>Zr. Evidence for coexisting shapes has been reported around <sup>100</sup>Zr . In the intermediate region, both <sup>96</sup>Zr and <sup>98</sup>Zr present evidence for sub-shell closure. These isotopes have a remarkably large gap between the ground state and the $`2_1^+`$ level, $`1.751`$ MeV in <sup>96</sup>Zr and $`1.223`$ MeV in <sup>98</sup>Zr. They also have a relatively low level density below 3 MeV. Empirically, the Zr isotopes are fairly well established. Lhersonneau and collaborators have recently performed careful experimental studies of the Zr isotopes and neighbouring nuclei, and they have made major contributions to the identification of levels in <sup>97</sup>Zr and <sup>99</sup>Zr .
For comparison the empirical Sr spectra are sketched in Figure 1(b). The Sr isotopes differ from the Zr isotopes by only two protons but are qualitatively quite different. Compared to Zr the Sr isotopes have a much smoother behaviour with a stable $`0^+2^+`$ spacing similar to that observed in tin. Large gaps due to sub-shell closure as observed in <sup>96</sup>Zr and <sup>98</sup>Zr do not occur in Sr. However, as in Zr there is a clear transition from spherical to deformed shape around $`N=60`$. The low-lying spectrum of <sup>100</sup>Sr is a nearly perfect rotational band. It is a theoretical challenge to describe a sequence of isotopes with such big changes in the structure from one nucleus to another as in Zr. For a proper description of the Zr isotopes one has to allow for proton excitations. In particular, the protons seem to play a dominant role in the $`0_2^+`$ state. Thus, the common choice of inert core has been <sup>88</sup>Sr, though with large variations in the size of model space, truncation scheme and effective interactions, Refs. . The early calculation by Auerbach and Talmi was carried out with valence protons filling the ($`1p_{1/2},0g_{9/2}`$) oribitals and valence neutrons filling the ($`1d_{5/2}`$) orbital. In particular if one wants to describe more neutron rich Zr isotopes a larger model space is required as the interaction between ($`0g_{9/2}`$) protons and ($`2d_{3/2}`$) and ($`0g_{7/2}`$) neutrons becomes increasingly important . To our knowledge the present work is the first shell-model calculation of the Zr-isotopes which includes nuclei up to <sup>98</sup>Zr within a non-truncated $`p:(1p_{1/2},0g_{9/2})`$ $`n:(1d_{5/2},2s_{1/2},1d_{3/2},0g_{7/2},0h_{11/2})`$ model space and with a fully realistic effective interaction.
We will in the present work perform a systematic shell-model study of the Zr isotopes from N=50 to N=58, and will in particular pay attention to the nuclei around the closure of the neutron $`(1d_{5/2})`$ and $`(2s_{1/2})`$ sub-shells, <sup>96-98</sup>Zr. In several works we have performed thorough analyses of the effective two-body interaction. We have derived shell-model effective interactions based on meson exchange models for the free nucleon-nucleon interaction, using many-body perturbation theory as described below. The systems that have been studied are reaching from the oxygen region to the tin isotopes and the N=82 isotones, Refs. . For the lighter systems, such as the $`sd`$\- and $`pf`$-shell nuclei, we obtained markedly better results for nuclei with one kind of valence nucleons than for nuclei with both kinds. For the heavier systems we have so far restricted ourselves to nuclei with like valence nucleons, such as the Sn isotopes and the $`N=82`$ isotones. This way we have managed to keep the dimensionality of the eigenvalue problem within tractable limits. Further, we have seen the need for establishing confidence in the $`T=1`$ interaction before considering systems with both valence protons and neutrons, where the proton-neutron interaction may play a crucial role. In fact, the Zr isotopes represent a challenge on both these accounts.
Let us in terms of a simplified model like the weak coupling scheme, in which the proton-neutron interaction is assumed to be weak, see if one can gain insight into the qualitative properties of this interaction. In Figure 2 we demonstrate the validity of this scheme by seeing how well it describes properties of <sup>92,94,96</sup>Zr. In column one the empirical spectrum of <sup>90</sup>Zr (dashed lines), which represents the proton degrees of freedom, is put on top of the <sup>90</sup>Sr spectrum (solid lines), which represents the neutron degrees of freedom. This would represent the <sup>92</sup>Zr spectrum in the weak coupling limit and should be compared with the empirical <sup>92</sup>Zr spectrum in column two. Similarly the weak coupling spectra $`{}_{}{}^{92}\mathrm{Sr}+^{90}`$Zr and $`{}_{}{}^{94}\mathrm{Sr}+^{90}`$Zr are compared with the empirical <sup>94</sup>Zr and <sup>96</sup>Zr spectra, respectively. Most states in <sup>92</sup>Zr and <sup>94</sup>Zr are well reproduced by the weak coupling scheme. The model does however collapse in <sup>96</sup>Zr, due to the presumed closure of the $`1d_{5/2}`$ sub-shell which does not have a counterpart in <sup>94</sup>Sr. The fact that the weak coupling scheme is fairly successful in describing <sup>92</sup>Zr and <sup>94</sup>Zr leads us to believe that the proton-neutron part of the effective interaction is either rather weak or state-independent.
The paper is organized as follows. In Sect. 2 we give a summary of the calculation of the effective interaction and the shell model. Then the results are presented and discussed in Sect. 3 and conclusions are drawn in Sect. 4.
## II THEORETICAL FRAMEWORK
The aim of microscopic nuclear structure calculations is to derive various properties of finite nuclei from the underlying hamiltonian describing the interaction between nucleons. When dealing with nuclei, such as the Zr isotopes with $`A=90100`$, the full dimensionality of the many-body Schrödinger equation
$$H\mathrm{\Psi }_i(1,\mathrm{},A)=E_i\mathrm{\Psi }_1(1,\mathrm{},A),$$
(1)
becomes intractable and one has to seek viable approximations to Eq. (1). In Eq. (1), $`E_i`$ and $`\mathrm{\Psi }_i`$ are the eigenvalues and eigenfunctions for a state $`i`$ in the Hilbert space.
One is normally only interested in solving Eq. (1) for certain low-lying states. It is then customary to divide the Hilbert space into a model space defined by the operator $`P`$ and an excluded space defined by a projection operator $`Q=1P`$
$$P=\underset{i=1}{\overset{d}{}}|\psi _i\psi _i|Q=\underset{i=d+1}{\overset{\mathrm{}}{}}|\psi _i\psi _i|,$$
(2)
with $`d`$ being the size of the model space and such that $`PQ=0`$. The assumption is that the low-lying states can be fairly well reproduced by configurations consisting of a few particles occupying physically selected orbitals, defining the model space. In the present work, the model space to be used both in the shell-model calculation and in the derivation of the effective interaction is given by the proton orbitals $`1p_{1/2}`$ and $`0g_{9/2}`$ and the neutron orbitals $`2s_{1/2}`$, $`1d_{5/2}`$, $`1d_{3/2}`$, $`0g_{7/2}`$ and $`0h_{11/2}`$.
Eq. (1) can then be rewritten as a secular equation
$$PH_{\mathrm{eff}}P\mathrm{\Psi }_i=P(H_0+V_{\mathrm{eff}})P\mathrm{\Psi }_i=E_iP\mathrm{\Psi }_i,$$
(3)
where $`H_{\mathrm{eff}}`$ now is an effective hamiltonian acting solely within the chosen model space. The term $`H_0`$ is the unperturbed hamiltonian while the effective interaction is given by
$$V_{\mathrm{eff}}=\underset{i=1}{\overset{\mathrm{}}{}}V_{\mathrm{eff}}^{(i)},$$
(4)
with $`V_{\mathrm{eff}}^{(1)}`$, $`V_{\mathrm{eff}}^{(2)}`$, $`V_{\mathrm{eff}}^{(3)}`$,… being effective one-body, two-body, three-body interactions etc. It is also customary in nuclear shell-model calculations to add the one-body effective interaction $`V_{\mathrm{eff}}^{(1)}`$ to the unperturbed part of the hamiltonian so that
$$H_{\mathrm{eff}}=\stackrel{~}{H}_0+V_{\mathrm{eff}}^{(2)}+V_{\mathrm{eff}}^{(3)}+\mathrm{},$$
(5)
where $`\stackrel{~}{H}_0=H_0+V_{\mathrm{eff}}^{(1)}`$. This allows us to replace the eigenvalues of $`\stackrel{~}{H}_0`$ by the empirical single-particle energies for the nucleon orbitals of our model space, or valence space. Thus, the remaining quantity to calculate is the two- or more-body effective interaction $`_{i=2}^{\mathrm{}}V_{\mathrm{eff}}^{(i)}`$. In this work we will restrict our attention to the derivation of an effective two-body interaction
$$V_{\mathrm{eff}}=V_{\mathrm{eff}}^{(2)},$$
(6)
using the many-body methods discussed in Ref. and briefly reviewed below.
### A Effective interaction
Our procedure for obtaining an effective interaction for the Zr isotopes starts with a free nucleon-nucleon interaction $`V^{(2)}`$ which is appropriate for nuclear physics at low and intermediate energies. At present, there are several potentials available. The most recent versions of Machleidt and co-workers , the Nimjegen group and the Argonne group have all a $`\chi ^2`$ per datum close to $`1`$. In this work we will thus choose to work with the charge-dependent version of the Bonn potential models, see Ref. . The potential model of Ref. is an extension of the one-boson-exchange models of the Bonn group , where mesons like $`\pi `$, $`\rho `$, $`\eta `$, $`\delta `$, $`\omega `$ and the fictitious $`\sigma `$ meson are included. In the charge-dependent version of Ref. , the first five mesons have the same set of parameters for all partial waves, whereas the parameters of the $`\sigma `$ meson are allowed to vary.
The next step in our perturbative many-body scheme is to handle the fact that the strong repulsive core of the nucleon-nucleon potential $`V`$ is unsuitable for perturbative approaches. This problem is overcome by introducing the reaction matrix $`G`$ given by the solution of the Bethe-Goldstone equation
$$G=V+V\frac{Q}{\omega QTQ}G,$$
(7)
where $`\omega `$ is the unperturbed energy of the interacting nucleons, and $`H_0`$ is the unperturbed hamiltonian. The operator $`Q`$, commonly referred to as the Pauli operator, is a projection operator which prevents the interacting nucleons from scattering into states occupied by other nucleons. In this work we solve the Bethe-Goldstone equation for five starting energies $`\omega `$, by way of the so-called double-partitioning scheme discussed in e.g., Ref. . The $`G`$-matrix is the sum over all ladder type of diagrams. This sum is meant to renormalize the repulsive short-range part of the interaction. The physical interpretation is that the particles must interact with each other an infinite number of times in order to produce a finite interaction.
Finally, we briefly sketch how to calculate an effective two-body interaction for the chosen model space in terms of the $`G`$-matrix. Since the $`G`$-matrix represents just the summmation to all orders of ladder diagrams with particle-particle intermediate states, there are obviously other terms which need to be included in an effective interaction. Long-range effects represented by core-polarization terms are also needed. The first step then is to define the so-called $`\widehat{Q}`$-box given by
$$P\widehat{Q}P=PGP+P\left(G\frac{Q}{\omega H_0}G+G\frac{Q}{\omega H_0}G\frac{Q}{\omega H_0}G+\mathrm{}\right)P.$$
(8)
The $`\widehat{Q}`$-box is made up of non-folded diagrams which are irreducible and valence linked. A diagram is said to be irreducible if between each pair of vertices there is at least one hole state or a particle state outside the model space. In a valence-linked diagram the interactions are linked (via fermion lines) to at least one valence line. Note that a valence-linked diagram can be either connected (consisting of a single piece) or disconnected. In the final expansion including folded diagrams as well, the disconnected diagrams are found to cancel out . This corresponds to the cancellation of unlinked diagrams of the Goldstone expansion . These definitions are discussed in Refs. . We can then obtain an effective interaction $`H_{\mathrm{eff}}=\stackrel{~}{H}_0+V_{\mathrm{eff}}^{(2)}`$ in terms of the $`\widehat{Q}`$-box , with
$$V_{\mathrm{eff}}^{(2)}(n)=\widehat{Q}+\underset{m=1}{\overset{\mathrm{}}{}}\frac{1}{m!}\frac{d^m\widehat{Q}}{d\omega ^m}\left\{V_{\mathrm{eff}}^{(2)}(n1)\right\}^m,$$
(9)
where $`(n)`$ and $`(n1)`$ refer to the effective interaction after $`n`$ and $`n1`$ iterations. The zeroth iteration is represented by just the $`\widehat{Q}`$-box. Observe also that the effective interaction $`V_{\mathrm{eff}}^{(2)}(n)`$ is evaluated at a given model space energy $`\omega `$, as is the case for the $`G`$-matrix as well. Here we choose $`\omega =20`$ MeV. Moreover, although $`\widehat{Q}`$ and its derivatives contain disconnected diagrams, such diagrams cancel exactly in each order , thus yielding a fully connected expansion in Eq. (9). Less than $`10`$ iterations were needed in order to obtain a numerically stable result. All non-folded diagrams through third order in the interaction $`G`$ are included. For further details, see Ref. .
### B Shell model
The effective two-particle interaction can in turn be used in shell-model calculations. Our approach in solving the eigenvalue problem is the Lanczos algorithm, which is an iterative method that gives the solutions of the lowest eigenstates. The technique is described in detail in Refs. . The shell-model code developed by us is designed for an $`m`$-scheme Slater determinant (SD) representation. Even with a rather restricted single-particle basis the size of the shell-model problem grows rapidly with increasing number of active valence particles. Table I shows how the number of configurations in an SD-basis grows with the number of valence particles acting within the ($`1p_{1/2}0g_{9/2}`$) proton shell and the ($`2s1d0g_{7/2}0h_{11/2}`$) neutron shell relative to the <sup>88</sup>Sr-core. Note that the <sup>98</sup>Zr system consists of more than 26.000.000 basis states. Only a few years ago shell-model calculations on such systems were not tractable. Even with today’s fast computers and effective algorithms this kind of calculation is still rather time consuming.
The single-neutron energies are taken to be those deduced from <sup>89</sup>Sr in Refs. . In the literature the single-proton energy splitting $`\epsilon (1p_{1/2})\epsilon (0g_{9/2})`$ varies from 0.839 MeV to 1.0 MeV . As the final results show little sensitivity to variations within this energy interval we let the $`0g_{9/2}`$ single-proton energy relative to $`1p_{1/2}`$ be 0.9 MeV. The single-particle energies used in this work are listed in Table II.
## III RESULTS AND DISCUSSIONS
In this section we present the calculations of the Zr isotopes. Firstly, we analyze some systematics of the even Zr isotopes. The results are displayed in Figures 3 and 4 (for some selected states) and in more detail in Tables IX, X and XI. Then, we proceed with the discussion of the odd Zr isotopes, in Figure 7 and Tables VII and VIII. A major aim is to investigate the effective interaction that has been derived for this mass region. Since the odd nuclei are generally more sensitive to the underlying assumptions made, this may give an even more severe test of the interaction and the foundations on which our model is based. At the end of this section we discuss problems concerning the binding energies. In order to study the effect of the proton degrees of freedom we have also performed a more restricted calculation, considering only valence neutrons with respect to a <sup>90</sup>Zr-core.
### A Even isotopes
The experimental spectra of the two nuclei, <sup>92</sup>Zr and <sup>94</sup>Zr, show very similar features. The shell-model calculation does also provide $`0_1^+`$, $`2_1^+`$ and $`4_1^+`$ levels in <sup>92</sup>Zr and <sup>94</sup>Zr with quite similar features, although the three levels are too compressed compared to their experimental counterparts. Comparison of the results obtained with <sup>90</sup>Zr- and <sup>88</sup>Sr-cores indicates that these levels are little affected by proton excitations. In <sup>96</sup>Zr the calculated $`0_1^+`$, $`2_1^+`$ and $`4_1^+`$ levels are still too compressed, though not as pronounced as in <sup>92</sup>Zr and <sup>94</sup>Zr, while in <sup>98</sup>Zr the calculated spectrum is more open than the experimental one.
The energy of the empirical $`3_1^{}`$ level, Figure 1(a), is monotonously reduced with increasing neutron number. In <sup>92</sup>Zr and <sup>94</sup>Zr the calculated $`3_1^{}`$ level is obtained at about 2 MeV, while in <sup>96</sup>Zr and <sup>98</sup>Zr the $`3_1^{}`$ level is obtained much too high at about 3.5 MeV. Due to the extension of the model space from a <sup>90</sup>Zr-core to a <sup>88</sup>Sr-core the $`3_1^{}`$ level in <sup>92</sup>Zr and <sup>94</sup>Zr undergoes a considerable lowering, yielding results close to the experimental values. From the occupation numbers in Table III it is clear that the $`3_1^{}`$ state undergoes a structural change from <sup>94</sup>Zr to <sup>96</sup>Zr. The $`0g_{7/2}`$ and $`2s_{1/2}`$ orbitals start to play a more important role. For comparison, the observed $`3_1^{}`$ levels in Sr, Figure 1(b), are all situated around 2 MeV. Their calculated counterparts are located too high, at about 3 MeV excitation energy. The structure of the $`3_1^{}`$ level in Zr is totally different from the structure in Sr, Figure 5. Let us look in detail into these states by comparing the $`3_1^{}`$ levels in <sup>92</sup>Sr and <sup>94</sup>Zr. Both nuclei have four valence neutrons. In <sup>92</sup>Sr the dominant configuration is $`[(1d_{5/2})^30h_{11/2}]_{J=3^{}}`$ whereas in <sup>94</sup>Zr the dominant configuration is $`[(1p_{1/2}0g_{9/2})_{J_\pi =4^{},5^{}}(1d_{5/2})_{J_\nu =2^+}^4]_{J=3^{}}`$. The difference can be ascribed to the single-particle energies. In Zr the $`3_1^{}`$ state is created by exciting a proton into the $`0g_{9/2}`$ orbital instead of exciting a neutron into the $`0h_{11/2}`$ orbital. Because the $`0h_{11/2}`$ orbital is located very high in the single-particle spectrum it is more favourable to excite a proton rather than a neutron.
The situation for the $`5_1^{}`$ level is more stable throughout the whole sequence of Zr isotopes, with empirical values between 2.5 and 3.0 MeV. The <sup>90</sup>Zr-core calculations provide level energies that are 1.0 - 1.5 MeV too high, while the <sup>88</sup>Sr-core calculations give energies too low by about 1 MeV. This state consists predominantly of configurations with a proton excited into the $`0g_{9/2}`$ orbital and the neutrons remaining in the lowest possible single-particle orbitals.
As pointed out in the introduction, there are strong variations in the structure from one nucleus to another, reflected in for instance the $`0^+2^+`$ spacing. Qualitatively we reproduce the variation in the $`0^+2^+`$ spacing quite well, as shown in Figure 6, although in <sup>90-96</sup>Zr the gap is $`200400`$ keV less than the experimental spacing.
In spite of clear differences in the experimental Zr and Sr spectra, the shell-model calculation provides rather similar results. The calculated Zr spectra are in far better agreement with experimental data than the calculated Sr spectra, which may indicate that the core is in a different condition in the two systems. In Sr the core seems to be relatively soft, whereas in Zr the two additional protons tend to stabilize the core.
#### 1 Proton configurations
The proton degrees of freedom are crucial in order to describe certain energy levels. For example, the first excited $`0_2^+`$ state in <sup>92</sup>Zr has strong components of proton excitations. From the occupation numbers in Table IV we see that the character of $`0_1^+`$ state is totally different from the $`0_2^+`$ state. The proton parts of their wave functions are almost orthogonal to each others. In the ground state the protons are most likely to be found in the $`1p_{1/2}`$-orbital, while for the excited $`0^+`$-state it is more probable to find the protons in the $`0g_{9/2}`$-orbital. However, in <sup>96</sup>Zr the two shell-model $`0^+`$ states have nearly the same proton structure, almost pure $`(1p_{1/2}^2)_\pi `$ configuration. The $`0^+`$ levels in <sup>98</sup>Zr show similar structure as in <sup>96</sup>Zr, but the $`0_2^+`$ state has slightly stronger $`0g_{9/2}`$ mixing than in <sup>96</sup>Zr. The third $`0^+`$ state in both <sup>96</sup>Zr and <sup>98</sup>Zr have $`(0g_{9/2})`$ as the predominant proton configuration. The change in the proton configuration with increasing neutron number was observed in pick-up experiments by Saha et al. .
#### 2 Sub-shell closure
Clear signs of sub-shell closure are seen in <sup>96</sup>Zr and also in <sup>98</sup>Zr, due to filling of the $`1d_{5/2}`$ and the $`2s_{1/2}`$ orbitals, respectively. From <sup>94</sup>Zr to <sup>96</sup>Zr the gap between the ground state and the $`2_1^+`$ state is doubled. The experimental $`0_1^+2_1^+`$ gap increases, from 0.919 MeV in <sup>94</sup>Zr to 1.751 MeV in <sup>96</sup>Zr, and the calculated gap is also more than doubled, from 0.520 MeV to 1.426 MeV. In <sup>98</sup>Zr the calculated spacing is larger than the experimental one. The experimental $`0^+2^+`$ spacing is 1.223 MeV, and the corresponding calculated spacing 1.463 MeV. The <sup>96</sup>Zr ground state $`1d_{5/2}`$ occupation number is 5.66, and in <sup>98</sup>Zr the $`1d_{5/2}`$ and $`2s_{1/2}`$ occupation numbers are 5.76 and 1.87, respectively.
The total impression of the <sup>98</sup>Zr shell-model results displayed in Figure 4 is disappointing. Only the $`2_1^+`$ level is reasonably reproduced. As an alternative to the extremely time and space consuming calculation presented in Figure 4, we may close the $`1d_{5/2}`$ orbital and perform a <sup>94</sup>Sr-core shell-model calculation of the system. The results, shown in Table XI, are much improved.
#### 3 E2 transition rates
Experimental and calculated E2 transition rates are tabulated in Table V. In order to bring the theoretical results into agreement with the measured $`2_1^+0_1^+`$ transition rates we have employed effective charges of 1.8$`e`$ and 1.5$`e`$ for the protons and neutrons, respectively. These values are consistent with the effective charges obtained in Refs. , however a bit overestimated compared to the fitting to data on <sup>92</sup>Mo and <sup>91</sup>Zr done by Halse in Ref. . To be mentioned, the calculation of effective charges based on perturbative many-body methods gives much smaller values, $`1.1e`$ and in the range $`0.5e0.7e`$ for the proton and neutron effective charge, respectively.
We adopt Halse’s effective value of the oscillator parameter $`b=2.25`$ fm. The value was choosen by reference to measurements for the radii of the single-particle orbitals in <sup>89</sup>Sr and of the charge distributions in <sup>92-96</sup>Mo, Refs. and .
With effective charges and the oscillator parameter as described above, the transition rates between yrast states are fairly well reproduced. In the former discussion we have focused on the $`0_2^+`$ state, in particlular its proton structure. In <sup>92</sup>Zr and <sup>94</sup>Zr we totally fail in reproducing the transition rates involving the $`0_2^+`$ state. The experimental transition rates between $`0_2^+`$ and $`2_1^+`$ in <sup>92</sup>Zr and <sup>94</sup>Zr are relatively strong, 14.3(5) and 9.3(4) W.u., respectively, whereas the calculated transition rates are two to three orders of magnitude smaller. Similarly, in <sup>98</sup>Zr there is an experimental transition rate between $`0_3^+`$ and $`2_1^+`$ with strength 51(5) W.u. The corresponding calculated transition rate is negligible. In fact, the occupation numbers in Table IV show that the $`0_3^+`$ state in <sup>98</sup>Zr has similar proton structure to the $`0_2^+`$ state in <sup>92</sup>Zr and <sup>94</sup>Zr. Leaving out the neutron contributions, as done in column 6 of Table V, by setting the effective neutron charge equal to zero, we observe that the proton part of the wave function contributes more to the transitions involving the excited $`0^+`$ states than what it does to the other transition rates. The contribution is however far from sufficient and there is a cancellation effect between the proton and neutron contributions. Contributions to the transition rates between yrast states do on the other hand mainly stem from the neutron degrees of freedom. In conclusion, it seems that the $`0_2^+`$ states in <sup>92</sup>Zr and <sup>94</sup>Zr and the $`0_3^+`$ state in <sup>98</sup>Zr contain strong collective components not reproduced by the present shell model calculation.
### B Odd isotopes
It is somehow surprising to notice that the shell model gives a much better description of the odd than of the even Zr isotopes. The reproduction of the low-lying positive parity states are overall satisfactory. On the other hand the shell model has some problems in describing the negative parity states. Several of the negative parity states in <sup>91,93</sup>Zr are calculated up to 1 MeV too low. Only a few negative parity states are known in <sup>95,97</sup>Zr. Thus a detailed comparison is difficult, but the $`11/2_1^{}`$ state is reproduced in nice agreement with experiment.
We will make a detailed study of <sup>93</sup>Zr. This nucleus is not too simple and not too complex (two protons and three neutrons outside the closed core), and useful information can therefore be extracted from a few central and relatively simple configurations.
The shell-model calculation provides three states below 600 keV, $`5/2_1^+`$, $`3/2_1^+`$ and $`9/2_1^+`$. From experiments, only two states are known, $`5/2_1^+`$ and $`3/2_1^+`$. There are however theoretical arguments supporting a low-lying $`9/2_1^+`$-state. The configuration that requires the least energy has all particles in the lowest possible single-particle orbital. In the case of <sup>93</sup>Zr the two protons occupy the $`1p_{1/2}`$ single-particle orbital, and couple to $`J=0`$. The three neutrons occupy the $`1d_{5/2}`$-orbital $`(1d_{5/2}^3)_\nu `$. The three neutrons can then couple to $`J^\pi =3/2^+`$, $`5/2^+`$ or $`9/2^+`$. Such states will be located well below 1 MeV. The lowest experimental $`9/2^+`$ candidate observed up to now is seen at 1.46 MeV.
As many as four $`1/2^+`$ states are observed within a small energy interval of 250 keV in the region from 0.95 MeV to 1.22 MeV. This observation has no shell-model counterpart. Our calculation provides only one $`1/2^+`$ level at 1.40 MeV.
We already pointed out that our model has difficulty in describing the negative parity states. Consider for example the structure of the $`11/2_1^{}`$ state in <sup>93</sup>Zr, which comes 500 keV lower than the experimental position. Odd parity states are constructed by configurations with an odd number of particles in negative parity states. Within our model, protons can occupy the $`1p_{1/2}`$-orbital and neutrons can occupy the $`0h_{11/2}`$-orbital to produce negative parity states. Both the first and second excited $`11/2^{}`$ states in <sup>93</sup>Zr are predominantly based on the proton configuration $`(1p_{1/2}0g_{9/2})_\pi `$. The same is true for the $`11/2_1^{}`$ state in the other Zr isotopes.
Finally, we examine <sup>97</sup>Zr. As we would expect for a system caught in between two “magic” nuclei we recognize pronounced single-particle structure. From the occupation numbers, listed in Table VI, we see that both the ground state, $`1/2_1^+`$, and the next level, $`3/2_1^+`$, are pure one-quasiparticle states built on a full $`1d_{5/2}`$-orbital, i.e. <sup>96</sup>Zr-core. Also the $`7/2_1^+`$ state is a one-quasiparticle state with the $`1d_{5/2}`$-orbital nearly closed, though its calculated position is about 0.7 MeV too high. This means that a somewhat lower single-particle energy $`\epsilon _{0g_{9/2}}`$might be more appropriate for our effective interaction. The $`5/2_1^+`$ state can be regarded as a $`1d_{5/2}`$-hole state relative to the <sup>98</sup>Zr-core. All in all the yrast states apart from $`7/2^+`$ are very well reproduced. In the other, non-yrast states, the <sup>96</sup>Zr-core breaks up and two neutrons are distributed equally among the $`2s_{1/2}`$ and $`1d_{3/2}`$ orbitals.
### C Binding energies
The binding energies calculated by the formula
$`\mathrm{BE}(^{90+n}\mathrm{Zr})`$ $`=`$ $`\mathrm{BE}(^{90+n}Zr)\mathrm{BE}(^{88}\mathrm{Sr})`$ (10)
$``$ $`n(\mathrm{BE}(^{89}\mathrm{Sr})\mathrm{BE}(^{88}\mathrm{Sr}))`$ (11)
$``$ $`2(\mathrm{BE}(^{89}\mathrm{Y})\mathrm{BE}(^{88}\mathrm{Sr}))`$ (12)
are plotted in Figure 8. In Eq. (12) $`n`$ is the number of valence neutrons. The experimental binding energies show a parabola structure with a minimum at <sup>96</sup>Zr, whereas the calculated binding energies increase linearly down to <sup>98</sup>Zr. With increasing neutron number the systems become far too strongly bound. This phenomenon of overbinding of nuclear systems when effective interactions from meson theory are used has been much discussed in the literature, for example in Ref. . The solutions to the problem has been that such matrix elements must be modified in order to reproduce the binding energies correctly. The so-called centroid matrix elements should be corrected in order to reproduce experiment. However, there is no well-defined recipe for doing this.
The curve labeled “no pn-int” in Figure 8 shows the binding energies with the proton-neutron interaction switched off. Now, the systems are too weakly bound, which tells us that the proton-neutron part of the interaction contributes strongly to the overbinding. We have made the pn-interaction less attractive by adding an overall constant to the diagonal matrix elements. This will not affect the excitation energies relative to the ground state. The constant is chosen such as to fit the experimental binding energy of <sup>90</sup>Y. Thus, a constant 0.3 MeV added to the original diagonal proton-neutron matrix elements, $`V_{\mathrm{abab}}^{\mathrm{mod}}(\mathrm{pn})=V_{\mathrm{abab}}^{\mathrm{eff}}(\mathrm{pn})+0.3\mathrm{MeV}`$, gives binding energies as shown in Figure 8, labeled “modified pn-int.”. The fit to the experimental values are much improved, although there is a linear rather than parabolic dependence on the particle number.
## IV CONCLUSIONS
In this work we have performed a full $`(1p_{1/2}0g_{9/2})`$ proton and $`(2s1d0g_{7/2}0h_{11/2})`$ neutron shell-model calculation of the zirconium isotopes ranging from N=52 to N=60. For the first time we present results from calculations with a proton-neutron effective interaction in such heavy nuclei.
We have succeeded in obtaining a qualitative reproduction of important properties, although there are also shortcomings. The odd isotopes are very well described by our shell model, in fact better than the calculated even isotopes. Both for the odd and the even nuclei we have difficulties in reproducing the negative parity states well.
For comparison we have presented shell-model results for the neighbouring strontium isotopes. The qualitative features are much better reproduced in Zr than in Sr. For example, the shell model fails in reproducing the very stable $`0^+2^+`$ spacing in Sr. Differently from Zr, there is no sign of N=56 and N=58 sub-shell closures in Sr. The quality of the closed-shell core (<sup>88</sup>Sr) may in fact be different for the two sets of isotopes. It is likely that the additional protons in Zr give a more balanced system and serve to stabilize the core. In Sr the core seems to be softer and more unstable.
The empirical Zr spectra can to a certain extent be interpreted in a weak coupling scheme. The isotopes <sup>92</sup>Zr and <sup>94</sup>Zr are well described in terms of this model, which in turn indicates that the proton-neutron interaction should not be too strong. The simple weak coupling picture collapses however in <sup>96</sup>Zr.
In order to obtain results for <sup>98</sup>Zr, we performed calculations that are extremely heavy and time consuming. All the efforts gave final results that were far off, and in fact a much simpler calculation based on a <sup>94</sup>Sr-core provided results in better agreement with the experimental data.
The occupation numbers give a hint that the $`0g_{7/2}`$ and the $`0h_{11/2}`$ neutron orbitals are of minor importance. The major properties are in fact fairly well described within a reduced basis $`(1p_{1/2},0g_{9/2})_\pi `$ $`(2s_{1/2},1d_{5/2},1d_{3/2})_\nu `$.
In order to further test the wave functions we calculated E2 transition rates in the even Zr isotopes. Transitions between yrast states are fairly well reproduced, whereas transitions involving certain excited $`0^+`$ states are calculated far too small, indicating that these states contain strong components not accounted for by the present shell model.
As in other mass regions we fail in reproducing bulk properties such as the binding energies. With increasing number of valence nucleons the systems become far too strongly bound. We have demonstrated that this problem can be cured by simple adjustments of some selected matrix elements.
The calculations have been carried out at the IBM cluster at the University of Oslo. Support for this from the Research Council of Norway (NFR) is acknowledged.
## A Tables
The low-lying experimental and calculated energy-levels in <sup>90-98</sup>Zr are listed in Tables IX, X and XI.
|
warning/0004/math0004045.html
|
ar5iv
|
text
|
# Ray Singer Analytic Torsion of Calabi Yau manifolds I.
## 1 Introduction.
One of the most remarkable formula that I encounter is the Kronecker limit formula. It states that if
$`E(s)=\underset{n,m}{^{}}\frac{1}{|n+m\tau |^{2s}}`$
where $`\tau `$, $`\mathrm{Im}\tau >0`$ and ’ means that the sum is taken over all pair of integers $`(m,n)(0,0),`$ then $`E(s)`$ has a meromorphic continuation in $``$ with only one pole at $`s=1`$ and $`\mathrm{exp}(\frac{d}{ds}E(s)|_{s=0})=\left(\mathrm{Im}\tau \right)^2\left|\eta \right|^4`$ where $`\eta `$ is the Dedekind eta function.
It is a well know fact that in the case of elliptic curves $`\{E_\tau =/(n+m\tau ),`$ $`\mathrm{Im}\tau >0\},`$ $`E(s)`$ is the zeta function of the Laplacian of the flat metric on the elliptic curves $`E_\tau ,`$ the regularized determinant of the Laplacian is $`\mathrm{exp}(\frac{d}{ds}E(s)|_{s=0})`$ and $`\eta ^{24}`$ is equal to the discriminant of the elliptic curve $`E_\tau .`$ $`\eta ^{24}`$ vanishes at $`\mathrm{},`$ which corresponds to an elliptic curve with the node. Thus the Kronecker limit formula is an explicit formula for the determinant of the Laplacian of an elliptic curve and gives a relation between the spectrum of the Laplacian and the discriminant of elliptic curves. The Kronecker limit formula has a modern interpretation as the Quillen norm of a section of the determinant line bundle.
There is a simple non formal explanation of the above mentioned fact. It is a well known fact that the spectrum of the Laplacian of a Riemannian metric on a compact manifold is discrete. When the manifold acquires singularities than the specter becomes continuous. This phenomenon suggests that when the metric ”degenerates” together with the manifold, then the regularized determinant vanishes on the points that parametrize the singular varieties. The problem is how to relate the specter of the Laplacian with the discriminant locus. The relation is suggested by the theory of determinant line bundles on the moduli space, their Quillen metrics and the Ray-Singer torsion as developed recently by Quillen, Donaldson, Bismut, Gillet and Soulé and others.
The problem that we are going to study in a series of two papers is to find the generalization of the analogue of the Dedekind eta function for odd dimensional CY manifolds.
The idea on which these two papers are based is very simple. The Quillen metric is related to the spectral properties of the Laplacian acting on (0,q) forms in case of Kähler manifolds. The main question is when the Ray Singer analytic torsion is the Quillen metric of some holomorphic section of the determinant line bundle. It is easy to prove that if the index of the $`\overline{}`$ operator is zero, then one can construct a non vanishing $`C^{\mathrm{}}`$ section $`det(\overline{})`$ of the determinant line bundle $``$ up to a constant whose Quillen norm is exactly the analytic Ray Singer torsion. We will show that knowing the existence of the non vanishing section $`det(\overline{})`$ implies that there exists a holomorphic section $`\eta ^N`$ of some power of the determinant line bundle which vanishes on $`𝒟_{\mathrm{}}=\overline{(M)}`$ $`\backslash `$$`(M),`$ where $`\overline{(M)}`$ is some projective compactification of $`(M)`$ such that $`𝒟_{\mathrm{}}=\overline{(M)}`$ $`\backslash `$$`(M)`$ is a divisor with normal crossings. According to Viewheg $`(M)`$ is a quasi projective variety. See .
For this program we need the analogue of the variational formulas for the determinant of the Laplacian of a CY metric acting on (0,q) forms. The variational formulas are very important in the construction of the holomorphic section $`\eta ^N`$ mentioned before.
In this paper we will generalize our variational formulas that were proved for the Laplacian of a CY metric acting on functions to (0,q) forms. See and .
We discussed the problem of finding the relations between the spectral properties of the Laplacian of CY metric on K3 surfaces in a series of joint papers with J. Jorgenson. (See , , and .)<sup>1</sup><sup>1</sup>1The mistakes that appeared in these papers are corrected in . The results of these papers showed that the problem of relating the spectral properties of the Laplacian of CY metric on even dimensional CY manifold is very delicate one. For example, in the case of algebraic polarized K3 surfaces we showed that the determinant of the Laplacian defines the discriminant locus of polarized K3 surfaces for polarization classes $`e`$ such that the Baily Borel compactification of the moduli space of algebraic pseudo-polarized K3 surfaces $`\mathrm{\Gamma }_n\backslash SO_0(2,19)/SO(2)\times SO(19)`$ contains only one zero dimensional cusp. In the other cases the discriminant locus can not be recovered from the spectral properties of the Laplacians of CY metrics. The difficulties in the even dimensional case are based on the fact that Ray Singer Analytic torsion is zero and the index of the $`\overline{}`$ is equal to 2. So we can not find a non vanishing canonical section of the determinant line bundle.
The analytic torsion for Enriques surfaces is discussed in from point of view of string theory and in from mathematical point of view. Based on these two papers one should consider the Enriques surfaces from the point of view of the spectral properties of the Laplacian of CY metric as an odd dimensional CY manifolds.
There are relations between the results our results in the series of the two papers and the results of in . Some of these relations are discussed in The results and the conjectures stated in thes two papers are related to the results in , and .
This article is organized as follows. In Section II we introduce the basic definition and some notations. In Section III we review Kodaira-Spencer-Kuranishi deformation theory of Calabi Yau manifolds following . In Section IV we introduce some canonical identifications of different Hilbert spaces on a CY manifold. We prove that some operators are of trace class and compute their traces. In Section V we proved that $`Tr(\mathrm{exp}(t\mathrm{\Delta }_q))=\left(\genfrac{}{}{0pt}{}{n}{q}\right)Tr(\mathrm{exp}(t\mathrm{\Delta }_0)),`$ where $`\mathrm{\Delta }_q`$ is the Laplacian of the Calabi Yau metric acting on (0,q) forms. Our proof is based on Bochner type technique. In Section VI we formulate and prove the main Theorem of this article, namely that $`\mathrm{log}(det(\mathrm{\Delta }_q))`$ is a potential for the Weil -Petersson metric. In Section VII we gave some applications of the technique and results that we used in the previous section. We prove that the coefficients $`a_k`$ in the short term asymptotic expansion of the trace of the heat kernel of CY metric for $`0kdim_{}`$M are constants. In Section VIII show that Ray Singer analytic torsion of CY metric I(M) is bounded.
###### Acknowledgement 1
The author wants to thank G. Moore for many stimulating conversations about the topic in this paper. These conversations inspired many of the ideas and the results in this paper. I want to thank my friend J. Jorgenson for introducing me to the exciting world of determinants of Laplacians. I am grateful to S. Donaldson, S.-T. Yau, G. Zuckerman, D. Kazhdan, S. Lang, B. Lian, J. Li, K. Liu, Y. Eliashberg and R. Donagi for their encouragements, useful comments and support. I want to thank Sinan Unver for his help. Special thanks to the National Center for Theoretical Sciences (Taiwan) for their hospitality during the preparation of this article. I want to express my special thanks to Prof. Chang and Prof. Wang. I want to thank Yale University for their hospitality and the opportunity to lecture on part of the material included in this paper. I want to thank G. Moore, D. Mostow and G. Zuckreman for making a number of useful comments. I want to thank P. Deligne for his useful and critical remarks.
## 2 Some Remarks, Notations and Preliminary Results.
### 2.1 Definition of the Regularized Determinant
Let (M,g) be an n dimensional Riemannian manifold. Let $`\mathrm{\Delta }_q=dd^{}+d^{}d`$ be the Laplacian acting on the space of q forms on M. It is a well known fact that the spectrum of the Laplacian $`\mathrm{\Delta }_q`$ is positive and discrete. This means that the non zero eigen values of $`\mathrm{\Delta }_q`$ are $`0<\lambda _1\lambda _2\mathrm{}\lambda _n\mathrm{}`$ We will define the zeta function of $`\mathrm{\Delta }_q`$ as follows: $`\zeta _q(s)=_{i=1}^{\mathrm{}}\lambda _i^s.`$ It is a well known fact that $`\zeta _q(s)`$ is a well defined analytic function for $`\mathrm{Re}(s)C,`$ it has a meromorphic continuation in the complex plane and $`0`$ is not a pole of $`\zeta _q(s).`$ Then we define $`det(\mathrm{\Delta }_q)=\mathrm{exp}\left(\frac{d}{ds}\left(\zeta _q(s)\right)|_{s=0}\right).`$
### 2.2 Definitions and Notations
Let M be a n-dimensional Kähler manifold with a zero canonical class. Suppose that $`H^k(`$M,$`𝒪_\text{M})=0`$ for 1$`k<n.`$ Such manifolds are called Calabi-Yau manifolds. A pair (M,$`L`$) will be called a polarized CY manifold if M is a CY manifold and $`LH^2(`$M,$``$)<sup>2</sup><sup>2</sup>2Notice that $`H^{1,1}(`$M,$`)=H^2(`$M,$`)`$ since $`H^2`$(M,$`𝒪_\text{M})=0`$ for CY manifolds. is a fixed class such that it represents the imaginary part of a Kähler metric on M.
Yau’s celebrated theorem asserts the existence of a unique Ricci flat Kähler metric g on M such that the cohomology class \[Im(g)\]=$`L`$. From now on we will consider polarized CY manifolds of odd dimension. The polarization class $`L`$ determines the CY metric g uniquely. We will denote by $`_q=\overline{}^{}\overline{}+\overline{}\overline{}^{}`$ the associated Laplacians that act on smooth $`(0,q)`$ forms on M for $`0qn`$. $`\overline{}^{}`$ is the adjoint operator of $`\overline{}`$ with respect to the CY metric g. The determinant of these operators $`_q,`$ defined through zeta function regularization, will be denoted by det$`\left(_q\right).`$
The Hodge decomposition theorem asserts that $`\mathrm{\Gamma }(`$M,$`\mathrm{\Omega }^{0,q})=\mathrm{Im}(\overline{})\mathrm{Im}(\overline{}^{})`$ for $`1qdim_{}M1.`$ The restriction of $`_q`$ on $`\mathrm{Im}(\overline{})`$ will be denoted by $`_q^{^{}}=\overline{}\overline{}^{}`$, and the restriction of $`\mathrm{\Delta }_q`$ on $`\mathrm{Im}(\overline{}^{})`$ will be denoted by $`_q^\mathrm{"}=\overline{}^{}\overline{}.`$ Hence we have $`Tr(\mathrm{exp}(t_q)=Tr(\mathrm{exp}(t_q^{^{}})+Tr(\mathrm{exp}(t_q^\mathrm{"}).`$ This implies that $`\zeta _q(s)=_{k=1}^{\mathrm{}}\lambda _k^s=\zeta _q^{^{}}(s)+\zeta _q^\mathrm{"}(s),`$ where $`\lambda _k>0`$ are the positive eigen values of $`_q`$ and $`\zeta _q^{^{}}(s)`$ & $`\zeta _q^\mathrm{"}(s)`$ are the zeta functions of $`_q^{^{}}`$ and $`_q^\mathrm{"}.`$ From here and the definition of the regularized determinant we obtain that $`\mathrm{log}det(_q)=\mathrm{log}det(_q^{^{}})+\mathrm{log}det(_q^\mathrm{"}).`$ It is a well known fact that the action of $`_q^^\mathrm{"}`$ on $`\mathrm{Im}\overline{}^{}`$ is isospectral to the action of $`_{q+1}^{^{}}`$ on $`\mathrm{Im}\overline{},`$ which means that the spectrum of $`_q^^\mathrm{"}`$ is equal to the spectrum of $`_{q+1}^{^{}}.`$ So we have the equality $`det(_q^\mathrm{"})=det(_{q+1}^{^{}}).`$
###### Notation 2
Let f be a map from a set A to a set B and let g be a map from the set B to the set C, then the compositions of those two maps we will denote by f$``$g.
## 3 Kodaira-Spencer-Kuranishi Theory for CY
### 3.1 Basic Definitions
In and was developed the local deformation theory of CY manifolds. We will review the results in and in this section.
Let M be an even dimensional C manifold. We will say that M has an almost complex structure if there exists a section $`IC^{\mathrm{}}(M,Hom(T^{},T^{})`$ such that $`I^2=id.`$ $`T`$ is the tangent bundle and $`T^{}`$ is the cotangent bundle on M. This definition is equivalent to the following one: Let M be an even dimensional C manifold. Suppose that there exists a global splitting of the complexified cotangent bundle $`T^{}𝐂=\mathrm{\Omega }^{1,0}\mathrm{\Omega }^{0,1}`$, where $`\mathrm{\Omega }^{0,1}=\overline{\mathrm{\Omega }^{1,0}}.`$ Then we will say that M has an almost complex structure. We will say that an almost complex structure is an integrable one, if for each point $`x`$M there exists an open set $`U`$M such that we can find local coordinates $`z^1,..,z^n,`$ such that $`dz^1,..,dz^n`$ are linearly independent in each point $`mU`$ and they generate $`\mathrm{\Omega }^{1,0}|_U.`$
###### Definition 3
Let M be a complex manifold. Let $`\varphi \mathrm{\Gamma }(`$M,$`Hom(\mathrm{\Omega }^{1,0},\mathrm{\Omega }^{0,1}))`$, then we will call $`\varphi `$ a Beltrami differential.
Since $`\mathrm{\Gamma }(`$M,$`Hom(\mathrm{\Omega }^{1,0},\mathrm{\Omega }^{0,1}))\mathrm{\Gamma }(`$M,$`\mathrm{\Omega }^{0,1}T^{1,0})`$, we deduce that locally $`\varphi `$ can be written as follows: $`\varphi |_U=\varphi _{\overline{\alpha }}^\beta \overline{dz}^\alpha \frac{}{z^\beta }`$. From now on we will denote by
$`A_\varphi =\left(\begin{array}{cc}id& \varphi (\tau )\\ \overline{\varphi (\tau )}& id\end{array}\right).`$
We will consider only those Beltrami differentials $`\varphi `$ such that det($`A_\varphi )0.`$ The Beltrami differential $`\varphi `$ defines an integrable complex structure on M if and only if the following equation holds: $`\overline{}\varphi +\frac{1}{2}[\varphi ,\varphi ]=0,`$ where
$`[\varphi ,\varphi ]|_U:=_{\nu =1}^n_{1\alpha <\beta n}\left(_{\mu =1}^n\left(\varphi _{\overline{\alpha }}^\mu \left(_\mu \varphi _{\overline{\beta }}^\nu \right)\varphi _{\overline{\beta }}^\mu \left(_\nu \varphi _{\overline{\alpha }}^\nu \right)\right)\right)\overline{dz}^\alpha \overline{dz}^\beta \frac{}{dz^\nu }.`$ (See .)
### 3.2 Kuranishi Space and Flat Local Coordinates
Kuranishi proved the following Theorem:
###### Theorem 4
Let $`\left\{\varphi _i\right\}`$ be a basis of harmonic $`(0,1)`$ forms of $`^1(`$M$`,T^{1,0})`$ on a Hermitian manifold M. Let $`G`$ be the Green operator and let $`\varphi (\tau ^1,..,\tau ^N)`$ be defined as follows:
$`\varphi (\tau ^1,..,\tau ^N)=_{i=1}^N\varphi _i\tau ^i+\frac{1}{2}\overline{}^{}G[\varphi (\tau ^1,..,\tau ^N),\varphi (\tau ^1,..,\tau ^N)]`$,
then there exists $`\epsilon >0`$ such that if $`\tau =(\tau ^1,..,\tau ^N)`$ satisfies $`|\tau _i|<\epsilon `$, then $`\varphi (\tau ^1,..,\tau ^N)`$ is a global $`C^{\mathrm{}}`$ section of the bundle $`\mathrm{\Omega }^{(0,1)}T^{1,0}`$.(See .)
Based on the Theorem 4 we proved in the following Theorem:
###### Theorem 5
Let M be a CY manifold and let $`\left\{\varphi _i\right\}`$ be a basis of harmonic $`(0,1)`$ forms with coefficients in $`T^{1,0}`$ of $`^1(`$M$`,T^{1,0}),`$ then the equation: $`\overline{}\varphi +\frac{1}{2}[\varphi ,\varphi ]=0`$ has a solution in the form:
$`\varphi (\tau _1,..,\tau _N)=_{i=1}^N\varphi _i\tau ^i+_{|I_N|2}\varphi _{I_N}\tau ^{I_N}=_{i=1}^N\varphi _i\tau ^i+\frac{1}{2}\overline{}^{}G[\varphi (\tau ^1,..,\tau ^N),\varphi (\tau ^1,..,\tau ^N)],`$
$`\overline{}^{}\varphi (\tau _1,..,\tau _N)=0`$, where $`I_N=(i_1,..,i_N)`$ is a multi-index, $`\varphi _{I_N}C^{\mathrm{}}(`$M$`,\mathrm{\Omega }^{0,1}T^{1,0})`$, $`\tau ^{I_N}=(\tau ^i)^{i_1}..(\tau ^N)^{i_N}`$ and for some $`\epsilon >0`$ $`\varphi (\tau )C^{\mathrm{}}(`$M,$`\mathrm{\Omega }^{0,1}T^{1,0})`$ if $`|\tau ^i|<\epsilon `$ and $`i=1,..,N.`$ See and $`\text{[21]}.`$
It is a standard fact from Kodaira-Spencer-Kuranishi deformation theory that for each $`\tau =(\tau ^1,..,\tau ^N)`$ as in Theorem 5 the Beltrami differential $`\varphi (\tau ^1,..,\tau ^N)`$ defines a new integrable complex structure on M, i.e. the points of $`𝒦,`$ where $`𝒦:\{\tau =(\tau ^1,..,\tau ^N)|`$ $`|\tau ^i|<\epsilon \}`$ defines a family of operators $`\overline{}_\tau `$ on the $`C^{\mathrm{}}`$ family $`𝒦\times MM,`$ parametrized by $`𝒦`$ and $`\overline{}_\tau `$ are integrable in the sense of Newlander-Nirenberg. Moreover it was proved by Kodaira, Spencer and Kuranishi that we get a complex analytic family of CY manifolds $`\pi :𝒳𝒦,`$ where as $`C^{\mathrm{}}`$ manifold $`𝒳𝒦\times M.`$ The family $`\pi :𝒳𝒦`$ is called the Kuranishi family. The operators $`\overline{}_\tau `$ are defined as follows:
###### Definition 6
Let $`\{𝒰_i\}`$ be an open covering of M, with local coordinate system in $`𝒰_i`$ given by $`\{z_i^k\}`$ with $`k=1,\mathrm{},n=`$dim<sub>C</sub>M. Assume that: $`\varphi (\tau ^1,..,\tau ^N)|_{𝒰_i}`$ is given by:
$`\varphi (\tau ^1,..,\tau ^N)=_{j,k=1}^n(\varphi (\tau ^1,..,\tau ^N))_{\overline{j}}^kd\overline{z}^j\frac{}{z^k}.`$
Then we define $`(\overline{})_{\tau ,\overline{j}}=\frac{\overline{}}{\overline{z^j}}_{k=1}^n(\varphi (\tau ^1,..,\tau ^N))_{\overline{j}}^k\frac{}{z^k}.`$
###### Definition 7
The coordinates $`\tau =(\tau ^1,..,\tau ^N)`$ defined in Theorem 5 will be fixed from now on and will be called the flat coordinate system in $`𝒦`$.
### 3.3 Weil-Petersson Metric
It is a well known fact from Kodaira-Spencer-Kuranishi theory that the tangent space $`T_{\tau ,𝒦\text{ }}`$at a point $`\tau 𝒦`$ can be identified with the space of harmonic (0,1) forms with values in the holomorphic vector fields $`^1(`$M$`{}_{\tau }{}^{},T`$). We will view each element $`\varphi ^1(`$M$`{}_{\tau }{}^{},T`$) as a pointwise linear map from $`\mathrm{\Omega }_{\text{M}_\tau }^{(1,0)}`$ to $`\mathrm{\Omega }_{\text{M}_\tau }^{(0,1)}.`$ Given $`\varphi _1`$ and $`\varphi _2^1(`$M$`{}_{\tau }{}^{},T`$), the trace of the map: $`\varphi _1\overline{\varphi _2}:`$ $`\mathrm{\Omega }_{\text{M}_\tau }^{(0,1)}\mathrm{\Omega }_{\text{M}_\tau }^{(0,1)}`$ at the point $`m`$M<sub>τ</sub> with respect to the metric g is simply:
$`Tr(\varphi _1\overline{\varphi _2})(m)=_{k,l,m,p=1}^n(\varphi _1)_{\overline{l}}^k(\overline{\varphi )_{\overline{p}}^m}g^{\overline{l},p}g_{k,\overline{m}}.`$
###### Definition 8
We will define the Weil-Petersson metric on $`𝒦`$ via the scalar product:
$`<\varphi _1,\varphi _2>=_\text{M}Tr(\varphi _1\overline{\varphi _2})vol(g).`$
We proved in that the coordinates $`\tau =(\tau ^1,..,\tau ^N)`$ as defined in Definition 7 are flat in the sense that the Weil-Petersson metric is Kähler and in these coordinates we have that the components $`g_{i,\overline{j}}`$ of the Weil Petersson metric are given by the following formulas in these coordinates:
$`g_{i,\overline{j}}=\delta _{i,\overline{j}}+R_{i,\overline{j},l,\overline{k}}\tau ^l\overline{\tau ^k}+O(\tau ^3).`$
On page 332 of the following results is proved:
###### Lemma 9
Let $`\varphi ^1(`$M$`,T`$) be a harmonic form with respect to the CY metric g. Let
$`\varphi |_U=_{k,l=1}^n\varphi _{\overline{k}}^l\overline{dz}^k\frac{}{z^l},`$
then $`\varphi _{\overline{k},\overline{l}}=_{j=1}^ng_{j,\overline{k}}\varphi _{\overline{l}}^j=_{j=1}^ng_{j,\overline{l}}\varphi _{\overline{k}}^j=\varphi _{\overline{l},\overline{k}}.`$
We will use Lemma 9 to prove the following theorem:
### 3.4 Infinitesimal Deformation of the Imaginary Part of the Weil-Petersson Metric
###### Theorem 10
Near the point $`\tau =0`$ of the Kuranishi space $`𝒦`$ the imaginary part $`\mathrm{Im}(g)`$ of the CY metric $`g`$ has the following expansion in the coordinates $`\tau :=(\tau ^1,..,\tau ^N)`$: $`\mathrm{Im}(g)(\tau ,\overline{\tau })=\mathrm{Im}(g)(0)+O(\tau ^2).`$
PROOF: In we proved that the forms $`\theta _\tau ^k=dz^k+_{l=1}\varphi (\tau ^1,..,\tau ^N)_{\overline{l}}^kd\overline{z^l}`$ ($`k=1,.,n)`$ form a basis of $`(1,0)`$ forms relative to the complex structure defined by $`\tau 𝒦`$ in $`𝒰`$M. Let
$`\mathrm{Im}(g_\tau )=\sqrt{1}_{1kln}g_{k,\overline{l}}(\tau ,\overline{\tau })`$ $`\theta _\tau ^k\overline{\theta _\tau ^l}.`$
and
$`g_{k,\overline{l}}(\tau ,\overline{\tau })=g_{k,\overline{l}}(0)+_{i=1}^N\left(\left(g_{k,\overline{l}}(1)\right)_i\tau ^i+\left(g_{k,\overline{l}}^{^{}}(1)\right)_i\overline{\tau ^i}\right)+O(2)`$.
Substituting in the expression for $`\mathrm{Im}(g_\tau )`$ the expressions for $`\theta _\tau ^k`$ we get the following formula:
$`\mathrm{Im}(g_\tau )=\sqrt{1}_{1kln}g_{k,\overline{l}}(\tau ,\overline{\tau })\theta _\tau ^k\overline{\theta _\tau ^l}=\sqrt{1}_{1kln}g_{k,\overline{l}}(0)dz^k\overline{dz^l}+`$
$`+_{i=1}^N\tau ^i\sqrt{1}\left(_{1kln}\left(\left(g_{k,\overline{l}}(1)\right)_idz^k\overline{dz^l}+_{m=1}^n(g_{k,\overline{m}}\overline{\varphi _{i,\overline{l}}^m}g_{l,\overline{m}}\overline{\varphi _{i,\overline{k}}^m})dz^kdz^l\right)\right)`$
$`+_{i=1}^N\overline{\tau ^i}\overline{\sqrt{1}(_{1kln}(\left(g_{k,\overline{l}}(1)\right)_idz^k\overline{dz^l}+(_{m=1}^n(g_{k,\overline{m}}\overline{\varphi _{i,\overline{l}}^m}g_{l,\overline{m}}\overline{\varphi _{i,\overline{k}}^m})dz^kdz^l))}.`$
From Lemma 9 we conclude that $`_{m=1}^n(g_{k,\overline{m}}\overline{\varphi _{i,\overline{l}}^m}g_{l,\overline{m}}\overline{\varphi _{i,\overline{k}}^m})=0`$ and so:
$`\mathrm{Im}(g_\tau )=\sqrt{1}_{1kln}g_{k,\overline{l}}(0)dz^k\overline{dz^l}+`$
$`_{i=1}^N\tau ^i\sqrt{1}\left(_{1kln}\left(g_{k,\overline{l}}(1)\right)_idz^k\overline{dz^l}\right)+_{i=1}^N\overline{\tau ^i}\sqrt{1}\overline{_{1kln}\left(g_{k,\overline{l}}(1)\right)_idz^k\overline{dz^l}}+O(2).`$
Let us define (1,1) forms $`\psi _i:`$
$`\psi _i=\sqrt{1}\left(_{1kln}\left(g_{k,\overline{l}}(1)\right)_idz^k\overline{dz^l}\right).`$
Since
\[$`\mathrm{Im}(g_\tau )`$\]=\[$`\mathrm{Im}(g_0)+_{i=1}^N\tau ^i\psi _i+_{i=1}^N\overline{\tau ^i\psi _i}+O(\tau ^2)]=[\mathrm{Im}(g_0)]`$
we deduce that each $`\psi _i`$ is an exact form, i.e.: $`\psi _i=\sqrt{1}\overline{}f_i,`$ where $`f_i`$ are globally defined functions on M. If we prove that $`\psi _i=0`$ our theorem will follow. In we proved that: det$`(g_\tau )=`$det$`(g_0)+O(2).`$ From this result we deduce by direct computations that:
det$`(g_\tau )=`$det$`(g_0)+_{i=1}^N\tau ^i\left(\sqrt{1}_{k,l}g^{\overline{l},k}_k\overline{_l}(f_i)\right)+_{i=1}^N\overline{\tau ^i}(`$complex conjugate)+$`O(2).`$
Hence we obtain that for each i we have: $`_{k,l}g^{\overline{l},k}_k\overline{_l}(f_i)=\mathrm{}(f_i)=0,`$ where $`\mathrm{}`$ is the Laplacian of the metric g. From the maximum principle, we deduce that all $`f_i`$ are constants. Theorem 10 is proved. $`\mathrm{}.`$
## 4 Hilbert Spaces and Operators of Trace Class.
### 4.1 Spectral Canonical Identifications of Some Hilbert Spaces
###### Definition 11
We will denote by $`L_{0,q1}^2(\mathrm{Im}(\overline{}^{}))`$ the Hilbert subspace in $`L^2(`$M,$`\mathrm{\Omega }^{(0,q1)})`$ which is the $`L^2`$ completion of $`\overline{^{}}`$ exact forms in $`C^{\mathrm{}}(`$M,$`\mathrm{\Omega }^{(0,q1)})`$ for $`q1.`$ In the same manner we will denote by $`L_{0,q}^2(\mathrm{Im}(\overline{}))`$ the Hilbert subspace in $`L^2(\mathrm{\Omega }^{(0,q)})`$ which is the $`L^2`$ completion of $`\overline{}`$ exact $`(0,q)`$ forms in $`C^{\mathrm{}}(`$M,$`\mathrm{\Omega }^{(0,q)})`$ for $`q0`$ and by $`L_{1,q1}^2(\mathrm{Im}())`$ we denote the Hilbert subspace in $`L^2(\mathrm{\Omega }^{(1,q1)})`$ which is the $`L^2`$ competition of the $``$ exact $`(1,q1)`$ forms in $`C^{\mathrm{}}(`$M,$`\mathrm{\Omega }^{(1,q1)})`$ . All the completions are with respect to the scalar product on the bundles $`\mathrm{\Omega }^{p,q}`$ defined by the CY metric g.
Let $`\varphi (\tau ^1,..,\tau ^N`$ ) be the solution of the equation $`\overline{}\varphi (\tau ^1,..,\tau ^N`$ )$`=\frac{1}{2}[\varphi (\tau ^1,..,\tau ^N`$ ),$`\varphi (\tau ^1,..,\tau ^N`$ $`)]`$ established in Theorem 5. From the Definition 3 of the Beltrami differential we know that $`\varphi (\tau ^1,..,\tau ^N`$ ) defines a linear fibrewise map $`\varphi (\tau ^1,..,\tau ^N):\mathrm{\Omega }^{(1,0)}\mathrm{\Omega }^{(0,1)}`$. So
$`\varphi (\tau ^1,..,\tau ^N`$ )$`C^{\mathrm{}}(M,Hom(\mathrm{\Omega }^{(1,0)},\mathrm{\Omega }^{(0,1)}.`$
We define the following linear map between the vector bundles $`\varphi id:`$ $`\mathrm{\Omega }^{(1,q1)}\mathrm{\Omega }^{(0,q)}`$ as $`\varphi (dz^i\alpha )=\varphi (dz^i)\alpha .`$
###### Definition 12
For each $`1qn,`$ $`\varphi id`$ defines a natural operator F(q,$`\varphi )`$ between the Hilbert spaces $`L^2(`$M, $`\mathrm{\Omega }^{(1,q1)})`$ and $`L^2`$(M,$`\mathrm{\Omega }^{(0,q)}`$).
###### Definition 13
The restriction of the map F(q,$`\varphi `$) on the subspace $`\mathrm{Im}()L^2(`$M, $`\mathrm{\Omega }^{(1,q1)}))`$ to $`\mathrm{Im}(\overline{)}L^2`$(M,$`\mathrm{\Omega }^{(0,q)}`$) will be denoted by $`F^{}(q,\varphi )`$.
###### Lemma 14
The Hilbert subspaces $`L_{0,q1}^2(\mathrm{Im}(\overline{}^{})),`$ $`L_{0,q}^2(\mathrm{Im}(\overline{}))`$ and $`L_{1,q1}^2(\mathrm{Im}())`$ are invariant with respect to the Laplacians
$`\mathrm{}_{q1}^\mathrm{"}=\overline{}_{q1}^{}\overline{}_q,`$ $`\mathrm{}_q^{^{}}=\overline{}_q\overline{}_{q+1}^{},`$
i.e.
$`\mathrm{}_{q1}^\mathrm{"}(L_{0,q1}^2(\mathrm{Im}(\overline{}^{}))=L_{0,q1}^2(\mathrm{Im}(\overline{}^{})),`$ $`\mathrm{}_q^{^{}}(L_{0,q}^2(\mathrm{Im}(\overline{}))=L_{0,q}^2(\mathrm{Im}(\overline{}))`$ and $`\mathrm{}_q^\mathrm{"}(L_{1,q1}^2(\mathrm{Im}())=L_{0,q1}^2(\mathrm{Im}())`$
PROOF: The proof of this lemma is standard fact from Kähler geometry. The first two identities followed from directly from the definition of the Laplacian. The last equality follows from the fact that in Kähler geometry the Laplacians $`\mathrm{}_{q1}=\overline{}_{q1}\overline{}_q^{}+\overline{}_{q1}^{}\overline{}_q`$ and $`\mathrm{}_{q1}^{^{}}=_{q1}_q^{}+_{q1}^{}_q`$ coincide, $`\overline{}_q^{}=[\mathrm{\Lambda },_q]`$ and $`_q^{}=[\mathrm{\Lambda },\overline{}_q]`$. See and . Our lemma is proved. $`\mathrm{}.`$
Let us denote by $`\{\omega _i(0,q1)|`$ $`i=1,\mathrm{},\mathrm{}\}`$ all the eigen forms of the Laplacian $`\mathrm{}_{q1}`$in the Hilbert space $`L_{0,q1}^2(\mathrm{Im}(\overline{}^{}))`$ with norm equal to one. In the same way we will denote by $`\{\omega _i(0,q)|`$ $`i=1,\mathrm{},\mathrm{}\}`$ all the eigen forms of the Laplacian $`\mathrm{}_q`$ with norm one in the Hilbert space $`L_{0,q}^2(\mathrm{Im}(\overline{}))`$ and by $`\{\omega _i(1,q1)|`$ $`i=1,\mathrm{},\mathrm{}\}`$ all the eigen forms of norm one of the Laplacian $`\mathrm{}_q`$in the Hilbert space $`L_{1,q1}^2(\mathrm{Im}()).`$
###### Lemma 15
The forms $`\{\omega _i(0,q1)|`$ $`i=1,\mathrm{},\mathrm{}\},`$ $`\{\omega _i(0,q)|`$ $`i=1,\mathrm{},\mathrm{}\}`$ and $`\{\omega _i(1,q1)|`$ $`i=1,\mathrm{},\mathrm{}\}`$ form orthonormal bases in the Hilbert spaces $`L_{0,q1}^2(\mathrm{Im}(\overline{}^{})),`$ $`L_{0,q}^2(\mathrm{Im}(\overline{}))`$ and $`L_{0,q1}^2(\mathrm{Im}()).`$
PROOF: The proof of this lemma is standard fact from the theory of self-adjoint compact operators in Hilbert spaces. See . $`\mathrm{}.`$
###### Lemma 16
Let (M,g) be a Kähler manifold with a Kähler metric g. Let
($`L_{0,q1}^2(\mathrm{Im}(\overline{^{}})),\{\omega _i(0,q1)\})`$, $`(L_{0,q}^2(\mathrm{Im}(\overline{})),\{\omega _i(0,q)\}`$
and $`(L_{1,q1}^2(\mathrm{Im}()),\{\omega _i(1,q1)\})`$ be the Hilbert spaces with orthonormal bases defined in Definition 11 for q$`1`$. Then
$`\overline{}\left(\frac{\omega _i(0,q1)}{\overline{}\omega _i(0,q1)^2}\right)=\lambda _i\omega _i(0,q)`$ and $`\left(\frac{\omega _i(0,q1)}{\omega _i(0,q1)^2}\right)=\lambda _i\omega _i(1,q1).`$
PROOF: This a standard fact which can be found in . $`\mathrm{}.`$
###### Remark 17
Lemma 16 gives a natural identification of the Hilbert spaces $`L_{0,q1}^2(\mathrm{Im}(\overline{^{}}))`$, $`L_{0,q}^2(\mathrm{Im}(\overline{}))`$ and $`L_{1,q1}^2(\mathrm{Im}())`$ because we can choose natural bases of all these Hilbert spaces by choosing an orthonormal basis consisting of eigen forms of the Laplacians. We are using the following orthonormal bases to get the above identifications:
$`\{\omega _i(0,q\},`$ $`\{\frac{(\omega _i(0,q)}{(\omega _i(0,q)^2}:=e_i\}`$ and $`\{\frac{\overline{}(\omega _i(0,q)}{\overline{}(\omega _i(0,q)^2}:=f_i\}.`$
### 4.2 Trace Class Operators in Hilbert Spaces
We will define the trace of the operator F$`^{^{}}`$(q,$`\varphi `$):$`L_{1,q1}^2(\mathrm{Im}())L_{0,q}^2(\mathrm{Im}(\overline{}))`$ (if this trace exists) acting on the identified Hilbert spaces as the usual trace of an operator acting on a Hilbert space. For example we define the trace of the operator F$`^{^{}}`$(q,$`\varphi `$) with respect to the orthonormal bases
$`\{\frac{(\omega _i(0,q)}{(\omega _i(0,q)^2}=\omega _i(1,q):=e_i\}`$ and $`\{\frac{\overline{}(\omega _i(0,q)}{\overline{}(\omega _i(0,q)^2}=\omega _i(0,q+1):=f_i\}.`$
###### Theorem 18
Let F’(q,$`\varphi )`$ be defined as in Definition 13, then F $`^{^{}}`$(q,$`\varphi `$) are operators of trace class.
PROOF: From the Definition 13 of the operators F$`{}_{}{}^{^{}}(q,\varphi )`$ we know that they are induced by the fibrewise linear maps $`\varphi id`$ :$`\mathrm{\Omega }^{1,q1}\mathrm{\Omega }^{0,q}.`$
Since M is a compact manifold we can choose $`N_{1,q1}`$ global $`C^{\mathrm{}}`$ forms $`\psi _i`$ of type (1,q-1) such that they span at each point $`yM,`$ the space $`\mathrm{\Omega }_y^{1,q1}.`$ In the same way we can find $`N_{0,q}`$ forms $`\sigma _j`$ of type (0,q) such that they span at each point $`yM,`$ the space $`\mathrm{\Omega }_y^{1,q1}.`$ Without lost of generality we may assume that both $`\psi _i`$ and $`\sigma _j`$ are linearly independent vectors in the identified Hilbert spaces $`L_{1,q1}^2(\mathrm{Im}())`$ $`\&`$ $`L_{0,q}^2(\mathrm{Im}(\overline{})).`$
Then the maps F(q,$`\varphi )`$:$`L_{1,q1}^2(\mathrm{Im}())L_{0,q}^2(\mathrm{Im}(\overline{}))`$ are given by $`N_{1,q1}\times N_{0,q}`$ matrix. So the maps F(q,$`\varphi )`$ are linear operators between finite dimensional spaces therefore they are of trace class. Since F$`{}_{}{}^{^{}}(q,\varphi )`$ are the restriction of the trace class operators F(q-1,$`\varphi )`$, we deduce that F$`{}_{}{}^{^{}}(`$q,$`\varphi )`$ are of trace class too. Theorem 18 is proved. $`\mathrm{}.`$
###### Corollary 19
The operator $`\overline{}^1F^{^{}}(q,\varphi )`$ is of trace class.
PROOF: We have identified the Hilbert spaces
$`L_{0,q1}^2(\mathrm{Im}(\overline{}^{}))`$, $`L_{0,q}^2(\mathrm{Im}(\overline{}))`$ and $`L_{1,q1}^2(\mathrm{Im}())`$
in Remark 17.
The operators $`\overline{}^1F^{^{}}(q,\varphi )`$ which act on $`L_{0,q1}^2(\mathrm{Im}(\overline{^{}}))`$ can be considered as a composition of a differential operator, operators with a smooth kernel and integral operator by using the above identification. From Proposition 2.45 page 96 in the book it follows directly that the operator $`\overline{}^1F^{^{}}(q,\varphi )`$ is of trace class. Cor. 19 is proved. $`\mathrm{}.`$
###### Theorem 20
For $`t>0`$ and q$`1`$ the following equality holds
$`Tr(\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"}))\overline{}^1F^{^{}}(q,\varphi ))=Tr(\mathrm{exp}(t(\mathrm{}_q^{^{^{}}})F^{^{}}(q,\varphi ))=_{i=1}^{\mathrm{}}\mathrm{exp}(t\lambda _i)a_{ii}.`$
where $`\lambda _i`$ are eigen values of $`\mathrm{}_{q1}^^\mathrm{"}`$ and we have the following expression for the trace:
$`Tr\left(F^{^{}}(q,\varphi )\right)=_{i=1}^{\mathrm{}}a_{ii}`$
in the orthonormal bases consisting of eigen vectors of the corresponding Laplacians as defined in Lemma 15.
PROOF: Theorem 18 and Corollary 19 imply that the operators $`\overline{}^1F^{^{}}(q,\varphi )`$ and $`F^{^{}}(q,\varphi )`$ are of trace class. The proof of this theorem is based on the direct computation of the traces of the operators $`\overline{}^1F^{^{}}(q,\varphi )`$ and $`F^{^{}}(q,\varphi )`$ with respect to the standard bases of orthonormal vectors
$`\left\{\frac{\omega _i(0,q1)}{\omega _i(0,q1)^2}=\omega _i(1,q1)\right\}`$ and $`\left\{\frac{\overline{}\omega _i(0,q1)}{\overline{}\omega _i(0,q1)^2}=\omega _i(0,q)\right\},`$
where $`\mathrm{\Delta }_{q1}\left(\omega _i(0,q1)\right)=\lambda _i\omega _i(0,q1),`$ $`\mathrm{\Delta }_q\omega _i(1,q1)=\lambda _i\omega _i(1,q1)`$ and $`\mathrm{\Delta }_q\omega _i(0,q)=\lambda _i\omega _i(0,q).`$(See Lemma 16). Let
$`F^{^{}}(q,\varphi )(\omega _i(1,q1))=_{j=1}^{\mathrm{}}a_{ij}(\omega _j(0,q)).`$
###### Lemma 21
We have the following formula:
Tr ($`\overline{}^1F^{^{}}(q,\varphi )`$)$`=`$Tr($`F^{^{}}(q,\varphi ))=_{i=1}^{\mathrm{}}a_{ii}`$ and q$`1.`$
PROOF: The operator $`\overline{}^1F^{^{}}(q,\varphi )`$ act on the Hilbert space $`L_{0,q1}^2(\mathrm{Im}(\overline{^{}}))`$ with an orthonormal basis of non zero eigen vectors of the Laplacian $`\mathrm{\Delta }_{q1}`$ $`\{\omega _i(0,q1)\}.`$ Recall that $`\omega _i(0,q1)=\overline{}\omega _i(0,q1)=\sqrt{\lambda _i}.`$ So we have
$`\omega _i(0,q1)=\sqrt{\lambda _i}\omega _i(1,q1)`$ and $`\overline{}\omega _i(0,q1)=\sqrt{\lambda _i}\omega _i(0,q).`$
From the expression $`F^{^{}}(q,\varphi )(\omega _i(1,q1))=_{j=1}^{\mathrm{}}a_{ij}(\omega _j(0,q))`$ and above equalities we obtain the following formula for the matrix of the operator $`\overline{}^1F^{^{}}(q,\varphi )`$ in the basis $`\{\omega _i(0,q1)\}`$
$`\overline{}^1F^{^{}}(q,\varphi )((\omega _i(0,q1))=\overline{}^1F^{^{}}(q,\varphi )\left(\sqrt{\lambda _i}\omega _i(1,q1)\right)=`$
$`\sqrt{\lambda _i}\left(\overline{}^1_{j=1}^{\mathrm{}}a_{ij}(\omega _j(0,q))\right)=\left(\sqrt{\lambda _i}\right)_{j=1}^{\mathrm{}}a_{ij}(\overline{}^1\omega _j(0,q))=`$
Substituting in the last formula the expression $`\frac{\overline{}\omega _j(0,q1)}{\sqrt{\lambda _j}}=\omega _j(0,q)`$ we obtain that
$`\overline{}^1F^{^{}}(q,\varphi )((\omega _i(0,q1))=\sqrt{\lambda _i}_{j=1}^{\mathrm{}}a_{ij}(\overline{}^1\frac{\overline{}\omega _j(0,q1)}{\sqrt{\lambda _j}})=.`$
$`=_{j=1}^{\mathrm{}}\frac{\sqrt{\lambda _j}}{\sqrt{\lambda _i}}a_{ij}(\omega _j(0,q1)).`$
So
$`Tr\left(\overline{}^1F^{^{}}(q,\varphi )\right)=_{i=1}^{\mathrm{}}_{j=1}^{\mathrm{}}\frac{\sqrt{\lambda _j}}{\sqrt{\lambda _i}}a_{ij}(\omega _j(0,q1)),\omega _i(0,q1)=`$
$`_{i=1}^{\mathrm{}}a_{ii}_{i=1}^{\mathrm{}}F^{}(q,\varphi )(\omega _i(0,q1)),\omega _i(0,q1).`$
On the other hand we know from Theorem 18 that the operator $`F^{}(q,\varphi )`$ is of trace class. From the canonical identifications of the Hilbert spaces $`L_{1,q1}^2(\mathrm{Im})`$ and $`L_{0,q}^2(\mathrm{Im}\overline{})`$ by the orthonormal eigen forms with a non zero eigen forms we deduce that
$`Tr(F^{}(q,\varphi ))=_{i=1}^{\mathrm{}}F^{^{}}(q,\varphi )(\omega _i(1,q1)),\omega _i(0,q)=_{i=1}^{\mathrm{}}_{j=1}^{\mathrm{}}a_{ij}(\omega _j(0,q)),\omega _i(0,q)`$.
Lemma 21 is proved. $`\mathrm{}.`$
The End of the Proof of Theorem 20: The formulas $`\mathrm{}_{q1}^^^\mathrm{"}\left(\omega _j(0,q1)\right)=\lambda _i\omega _j(0,q1)`$ imply that
$`\mathrm{exp}(t(\mathrm{}_{q1}^^^\mathrm{"})\left(\omega _j(0,q1)\right)=\mathrm{exp}(t\lambda _j)\omega _j(0,q1).`$
From the expression $`\overline{}^1F^{^{}}(q,\varphi )(\omega _i(0,q1))=a_{ij}\frac{\sqrt{\lambda _j}}{\sqrt{\lambda _i}}\omega _j(0,q1)`$ proved in Lemma 21 we deduce:
$`Tr(\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})\overline{}^1F^{^{}}(q,\varphi ))=`$
$`_{i=1}^{\mathrm{}}(exp(t(\mathrm{}_{q1}^^\mathrm{"})\overline{}^1(F^{^{}}(q,\varphi ))(\omega _i(0,q1)),\omega _i(0,q1)=`$
$`_{i=1}^{\mathrm{}}_{j=1}^{\mathrm{}}exp(t(\lambda _j)a_{ij}\frac{\sqrt{\lambda _j}}{\sqrt{\lambda _i}}\omega _j(0,q1)),\omega _i(0,q1)=`$
$`_{i=1}^{\mathrm{}}a_{ii}\mathrm{exp}(t\lambda _i).`$
So we obtain $`Tr`$($`\mathrm{exp}(t(\mathrm{}_q^^\mathrm{"})\frac{}{\tau _i}\left(\overline{}_\tau \right)|{}_{\tau =0}{}^{})=_{i=1}^{\mathrm{}}a_{ii}\mathrm{exp}(t\lambda _i).`$ From the expression $`\mathrm{}_q^{^{^{}}}\left(\overline{}\omega _i(0,q1)\right)=\lambda _i\left(\overline{}\omega _i(0,q1)\right)`$ we obtain that $`\mathrm{exp}(t\mathrm{}_q^{^{}})\omega _i(1,q1)=\mathrm{exp}(\lambda _it)\omega _i(1,q1).`$ From the formula $`F^{^{}}(q,\varphi )(\omega _i(1,q1)))=_{j=1}^{\mathrm{}}a_{ij}(\omega _j(0,q))`$ we conclude:
$`\left(\mathrm{exp}(t\mathrm{}_q^{^{}})F^{^{}}(q,\varphi )\right)(\omega _i(1,q1))=_{j=1}^{\mathrm{}}a_{ij}\mathrm{exp}(\lambda _jt)\omega _i(0,q)`$
and so
$`Tr(\mathrm{exp}(t\mathrm{}_q^{^{}})F^{^{}}(q,\varphi ))=_{j=1}^{\mathrm{}}a_{ii}\lambda _i\mathrm{exp}(\lambda _it)=Tr(\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})\overline{}^1F^{^{}}(q,\varphi )).`$
Theorem 20 is proved. $`\mathrm{}.`$
###### Corollary 22
We have the following formula:
$`Tr(\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})\overline{}^1F^{^{}}(q,\varphi ))Tr(\mathrm{exp}(t\mathrm{}_q^{^{}})F^{^{}}(q,\varphi ))=`$
$`Tr\left(\mathrm{exp}(t\mathrm{}_q^{^{}})F^{^{}}(q,\varphi )\right)=_{j=1}^{\mathrm{}}a_{ii}\mathrm{exp}(\lambda _it).`$
###### Corollary 23
a<sub>ii</sub> tends to zero with $`i\mathrm{}`$ exponentially fast.
Repeating the arguments that we used to prove Theorem 20 we get the following Theorem:
###### Theorem 24
For $`t>0`$ and q$`1`$ the following equality holds
$`Tr\left(\mathrm{exp}\left(t(\mathrm{}_{q1}^^\mathrm{"})\right)^1\overline{F^{^{}}(q,\varphi _j)}F^{^{}}(q_i,\varphi )\right)=`$
$`Tr(\mathrm{exp}(t(\mathrm{}_q^{^{^{}}})\overline{F^{^{}}(q,\varphi _j)}F^{^{}}(q_i,\varphi ))=_{j=1}^{\mathrm{}}c_{ii}\mathrm{exp}(\lambda _it)`$
where $`Tr\left(\overline{F^{^{}}(q,\varphi _j)}F^{^{}}(q_i,\varphi )\right)=_{j=1}^{\mathrm{}}c_{ii}.`$
## 5 Bochner’s Formulas for CY manifolds.
We will now give explicit expression for Ray-Singer torsion using 26. In order to use 26 we need to have information about the relations between $`Tr(\mathrm{exp}(\mathrm{}_q))`$ and $`Tr(\mathrm{exp}(\mathrm{}_0)).`$ We will use Bochner technique to find these relations.
###### Theorem 25
$`Tr(\mathrm{exp}(t\mathrm{}_q))=\left(\genfrac{}{}{0pt}{}{n}{q}\right)Tr(\mathrm{exp}(t\mathrm{}_0)).`$
PROOF: In order to prove Theorem 25we will use the following formulas proved in on page 119: Let M be a Kähler manifold and let $`\mathrm{}`$ be the Laplacian of a Kähler metric defined on (p,q) form
$`\varphi =\frac{1}{p!q!}\varphi _{i_1,.,i_p;\overline{j}_1,.,\overline{j}_q}dz^{i_1}.dz^{i_p}\overline{dz}^{j_1}.\overline{dz}^{j_q},`$
then
$`(\mathrm{}\varphi )_{i_1,.,i_p;\overline{j}_1,.,\overline{j}_q}=_{i,j}g^{\overline{j},i}_i\overline{}_j\varphi _{i_1,.,i_p;\overline{j}_1,.,\overline{j}_q}+`$
$`+_k_l_{m,n}R^m`$ $`_{i_k,\overline{j}_l}`$ $`{}_{}{}^{\overline{n}}\varphi _{i_1,.,i_{k1},m,i_k,.i_p\overline{j}_1,.,\overline{j}_{l1},\overline{n},\overline{j}_{l+1},..,,\overline{j}_q}^{}`$
$`_{k=1}^n_mR_{\overline{j}_k}`$ $`{}_{}{}^{\overline{m}}\varphi _{i_1,.,i_p\overline{j}_1,.,\overline{j}_{k1},\overline{m},\overline{j}_{k+1},..,,\overline{j}_q}^{},`$
where $`R_{i,\overline{j},k,\overline{l}}`$ is the curvature of the Kähler metric g, $`\overline{}_j=\overline{}_j`$ and $`_i`$ is the covariant derivative in the direction $`\frac{}{z^i}`$ and
$`R_{\overline{n}}`$ $`{}_{}{}^{\overline{m}}=_{k=1}^ng^{\overline{m},k}R_{k,\overline{m}},`$
where $`R_{k,\overline{m}}`$ is the Ricci curvature. If M is a CY manifold and g is a CY metric, then $`R_{k,\overline{m}}=0.`$ When $`\varphi `$ is a form of type $`(0,q),`$ then from the above mentioned formulas we obtain that:
$`(\mathrm{}\varphi )_{\overline{j}_1,.,\overline{j}_q}=_{n,m}g^{\overline{m},n}_n\overline{}_m\varphi _{\overline{j}_1,.,\overline{j}_q}.`$
On page 110 in the following formula is proved:
$`(\overline{}^{}\varphi )_{\overline{j}_1,.,\overline{j}_q}=(1)^p_{m,n}g^{\overline{m},n}_n\varphi _{\overline{j}_1,.,\overline{j}_q}.`$
Using all these formulas we get
$`\left(\varphi \right)_{\overline{j}_1,.,\overline{j}_q}=\overline{}^{}\overline{}(\varphi _{\overline{j}_1,.,\overline{j}_q})=_0\left(\varphi _{\overline{j}_1,.,\overline{j}_q}\right).`$
From here Theorem 25 follows directly, i.e. $`Tr(\mathrm{exp}(\mathrm{}_q))=\left(\genfrac{}{}{0pt}{}{n}{q}\right)Tr(\mathrm{exp}(\mathrm{}_0)).`$ Our Theorem is proved. $`\mathrm{}.`$
## 6 Variational Formulas.
Let $`<\varphi _i,\varphi _j>`$ be defined as in Definition 8, then
###### Theorem 26
The following variational formulas hold for CY manifolds of complex dimension $`n2:`$
i. $`(\frac{d^2}{\tau _i\overline{\tau _i}}\mathrm{log}(det(\mathrm{}_q^\mathrm{"}))(0)=\left(\genfrac{}{}{0pt}{}{n1}{q1}\right)<\varphi _i,\varphi _j>`$ for 1$`qn1.`$
ii. $`(\frac{d^2}{\tau _i\overline{\tau _i}}(\mathrm{log}(det(\mathrm{}_q^\mathrm{"}))(0)=<\varphi _i,\varphi _j>`$ for $`q=1`$ or $`n.`$
### 6.1 Ideas of the Proof
Let $`q1.`$ The proof of Theorem 26 is based on the fact that $`\zeta _{q1}(s)`$ is the Mellin transform of $`Tr\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"}),`$ i.e. we have
$`\zeta _{q1}(s)=\frac{1}{\mathrm{\Gamma }(s)}_0^{\mathrm{}}(Tr\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"}))t^{s1}dt.`$
The definition of det($`\mathrm{}_{q1}^^\mathrm{"})=\frac{d}{ds}(\zeta _{q1}(s))|_{s=0}`$, and the power series expansion of zeta function $`\zeta _{q1}(s)=\zeta _{q1}(0)+\frac{d}{ds}\left(\zeta _{q1}(0)\right)s+\mathrm{}`$ suggest that in order to compute $`\frac{^2}{\overline{\tau _i}\text{ }\tau _i}(det\left(\mathrm{log}(\mathrm{}_{q1}^^\mathrm{"})\right))|_{\tau =0}`$ we need to compute
$`\frac{d}{ds}\left(\frac{^2}{\overline{\tau _i}\text{ }\tau _i}\frac{1}{\mathrm{\Gamma }(s)}_0^{\mathrm{}}Tr\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})t^{s1}dt)\right|{}_{s=0,\tau =0}{}^{}.`$
First we will compute $`\frac{d}{d\tau _i}Tr\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})`$ and will prove that
$`\frac{d}{d\tau _i}Tr\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})=t\frac{d}{dt}Tr\mathrm{exp}((t(\mathrm{}_{q1}^^\mathrm{"})\overline{}^1F^{}(q,\varphi _i))=t_{i=1}^{\mathrm{}}\mathrm{exp}(\lambda _it)\lambda _ia_{ii},`$
where $`Tr(F^{}(q,\varphi _i))=_{i=1}^{\mathrm{}}a_{ii}.`$ By integrating by parts and following closely the arguments from the book on page 257-260 we will obtain
$`\frac{}{\tau _i}(\frac{1}{\mathrm{\Gamma }(s)}_0^{\mathrm{}}Tr\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})t^{s1}dt)|_{\tau =0}=\frac{s}{\mathrm{\Gamma }(s)}_0^{\mathrm{}}Tr(\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})\overline{}^1F^{}(q,\varphi _i))t^{s1}dt.`$
From the last formula we will obtain that:
$`\frac{d}{ds}\left(\frac{^2}{\overline{\tau _i}\text{ }\tau _i}\frac{1}{\mathrm{\Gamma }(s)}_0^{\mathrm{}}Tr\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})t^{s1}dt)\right|{}_{s=0,\tau =0}{}^{}=`$
$`\underset{t0}{lim}\frac{\overline{}}{\overline{\tau _i}}Tr(\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})\overline{}^1F^{}(q,\varphi _i)).`$
Direct computation will show that:
$`\underset{t0}{lim}\frac{\overline{}}{\overline{\tau _i}}Tr(\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})\overline{}^1F^{}(q,\varphi _i))=`$
$`\underset{t0}{lim}Tr(\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"}))^1\overline{F^{}(q,\varphi _j)}F^{}(q,\varphi _i)))=`$
$`Tr\left(\overline{F^{}(q,\varphi _j)}F^{}(q,\varphi _i)\right)=\varphi _i,\varphi _j.`$
### 6.2 Preliminary Results
###### Lemma 27
The following formulas are true:
$`\frac{}{\tau _i}\left(\overline{_\tau }\right)|_{\tau =0}=`$F$`{}_{}{}^{^{}}(q,\varphi _i).`$
PROOF: From the expression in Definition 6 we conclude that
$`\delta _i^{^{}}(`$ $`\overline{_\tau })=\frac{}{\tau _i}\left(\overline{}_\tau \right)=\frac{}{\tau ^i}(\frac{\overline{}}{\overline{z^j}}_{m=1}^N(\tau ^m_{k=1}^N(\varphi _m)_{\overline{j}}^k\frac{}{z^k}))+O(\tau ^2)).`$
So
$`\frac{}{\tau _i}\left(\overline{_\tau }\right)|_{\tau =0}=_{k=1}^N(\varphi _i)_{\overline{j}}^k\frac{}{z^k}.`$
Lemma 27 follows directly from this expression and the Definition 13 of F$`{}_{}{}^{^{}}(q,\varphi _i)`$. $`\mathrm{}.`$
###### Lemma 28
$`\frac{}{\tau _i}\left(\overline{}_\tau ^{}\right)|_{\tau =0}=0.`$
PROOF: We know from Kähler geometry that $`(\overline{_\tau })^{}=[\mathrm{\Lambda }_\tau ,_\tau ],`$ where $`\mathrm{\Lambda }_\tau `$ is the contraction with (1,1) vector filed$`:`$
$`\frac{\sqrt{1}}{2}_{k,l=1}^n`$g$`{}_{\tau }{}^{\overline{k},l}(\theta _\tau ^l)_{}^{}(\overline{\theta _\tau ^k})^{}.`$
on M<sub>τ</sub> and $`(\theta _\tau ^l)^{}`$ is (1,0) vector field on M<sub>τ</sub> dual to the (1,0) form $`\theta _\tau ^i=dz^i+_{j=1}^N\tau ^j(_{k=1}^n(\varphi _j)_{\overline{k}}^i\overline{dz}^k)).`$ Theorem 10 implies $`\frac{}{\tau _i}(\mathrm{\Lambda }_\tau )|_{\tau =0}=0.`$ On the other hand $`_\tau `$ depends antiholomorphically on $`\tau `$, i.e. it depends on $`\overline{\tau }=(\overline{\tau _1},.,\overline{\tau _N}).`$ So we deduce that:
$`\frac{}{\tau _i}((\overline{_\tau })^{})|_{\tau =0}=\left([\frac{}{\tau _i}(\mathrm{\Lambda }_\tau ),_\tau ]+[\mathrm{\Lambda }_\tau ,\frac{}{\tau _i}(_\tau )]\right)|_{\tau =0}=0.`$
Lemma 28 is proved. $`\mathrm{}.`$
### 6.3 Computation of holomorphic derivative
###### Theorem 29
The following formula is true
$`\frac{}{\tau _i}(Tr(\mathrm{exp}(t\mathrm{}_{\tau ,q1}^\mathrm{"}))|_{\tau =0}=tTr(\frac{d}{dt}\left(\mathrm{exp}(t\mathrm{}_{q1}^\mathrm{"})\right)\overline{}^1F^{}(q,\varphi _i)_\tau )|_{\tau =0}.`$
PROOF: Direct computations show that:
$`\frac{}{\tau _i}(Tr(\mathrm{exp}(t\mathrm{}_{\tau ,q1}^\mathrm{"}))=t\mathrm{exp}((t\mathrm{}_{\tau ,q1}^\mathrm{"})\frac{d}{d\tau _i}\left(\mathrm{}_{\tau ,q1}^\mathrm{"}\right))|_{\tau =0}.`$
Lemma 27 and 28 imply that
$`\frac{d}{d\tau _i}\left(\mathrm{}_{\tau ,q1}^\mathrm{"}\right)|_{\tau =0}=\left(\overline{}_\tau ^{}\frac{d}{d\tau _i}\overline{}_\tau \right)|_{\tau =0}=\left(\overline{}^{}F^{}(q,\varphi _i)\right)|_{\tau =0}.`$
Since for CY manifolds of complex dimension $`3`$ the operators $`\overline{}_\tau `$ give isomorphisms between the spaces of non constant functions on $`M_\tau `$ $`C^{\mathrm{}}(M_\tau )/`$ and the space of $`C^{\mathrm{}}`$ $`\overline{}_\tau `$ closed (0,1) forms on $`M_\tau .`$ So we have that for $`\overline{}_\tau ^{}`$ is well defined on the space of $`C^{\mathrm{}}`$ $`\overline{}_\tau `$ closed (0,1) forms on $`M_\tau `$ and we have the following formula on $`\mathrm{Im}\overline{}`$:
$`\overline{}_\tau ^{}=(\mathrm{}_{\tau ,q1}^\mathrm{"})\overline{}_\tau ^1.`$
Using all this information we get by direct substitutions that:
$`\frac{}{\tau _i}(Tr(\mathrm{exp}(t\mathrm{}_{q,\tau }))|_{\tau =0}=tTr\left(\mathrm{exp}((t\mathrm{}_{\tau ,q1}^\mathrm{"})\frac{d}{d\tau _i}(\overline{}_\tau ^{}\overline{}_\tau ))\right)|_{\tau =0}=`$
$`tTr(\mathrm{exp}(t\mathrm{}_{q1}^\mathrm{"})(\overline{}^{}F^{}(q,\varphi _i)))=`$
$`tTr\left(\mathrm{exp}\left(t\mathrm{}_{q1}^\mathrm{"}\right)\mathrm{}_{q1}^\mathrm{"}\overline{}^1F(q,\varphi _i)\right)=tTr\left(\frac{d}{dt}\mathrm{exp}\left(t\mathrm{}_{q1}^\mathrm{"}\right)\overline{}^1F(q,\varphi _i)\right).`$
Theorem 29 is proved. $`\mathrm{}.`$
###### Lemma 30
The following formula is true:
$`\frac{}{\tau _i}(\frac{1}{\mathrm{\Gamma }(s)}_0^{\mathrm{}}Tr\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})t^{s1}dt)|_{\tau =0}=\frac{s}{\mathrm{\Gamma }(s)}_0^{\mathrm{}}Tr(\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})\overline{}^1F^{}(q,\varphi _i))t^{s1}dt.`$
PROOF: Theorem 29 imply that we have:
$`\frac{}{\tau _i}(\frac{1}{\mathrm{\Gamma }(s)}_0^{\mathrm{}}Tr\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})t^{s1}dt)|_{\tau =0}=\frac{1}{\mathrm{\Gamma }(s)}_0^{\mathrm{}}Tr(\frac{d}{dt}(\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"}))\overline{}^1F^{}(q,\varphi _i))t^sdt.`$
Taking into account that Theorem 20 implies that
$`\underset{t0}{lim}(Tr\left(\frac{d}{dt}(\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"}))\overline{}^1F^{}(q,\varphi _i))\right)=\underset{t0}{lim}(t_{i=1}^{\mathrm{}}\mathrm{exp}(\lambda _it)\lambda _ia_{ii})=0`$
and by integrating by parts as in we derived the formula stated in Lemma 30. Lemma 30 is proved. $`\mathrm{}.`$
###### Corollary 31
The following formula is true:
$`\frac{d}{ds}\left(\frac{}{\tau _i}(\zeta _{q1}(s))\right)|_{s=0}=\underset{t0}{lim}Tr(\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})\overline{}^1F^{}(q,\varphi _i)).`$
PROOF: Lemma 30 implies:
$`\frac{d}{d\tau _i}\left(\zeta _{q1}(s)\right)=\frac{s}{\mathrm{\Gamma }(s)}_0^{\mathrm{}}Tr(\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})\overline{}^1F^{}(q,\varphi _i))t^{s1}dt.`$
We already computeted the trace of the operator $`\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})\overline{}^1F^{}(q,\varphi _i)`$ so we obtain:
$`\underset{t0}{lim}Tr(\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})\overline{}^1F^{}(q,\varphi _i))=\underset{t0}{lim}\mathrm{exp}(t\lambda _i)\lambda _ia_{ii}=`$
$`\lambda _ia_{ii}<\mathrm{}.`$
From the fact that $`\frac{s}{\mathrm{\Gamma }(s)}=s^2+O(s^3)`$ and direct easy computations we conclude that
$`\frac{d}{ds}\left(\frac{}{\tau _i}(\zeta _{q1}(s))\right)|_{s=0}=\underset{t0}{lim}Tr(\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})\overline{}^1F^{}(q,\varphi _i)).`$
Corollary 31 is proved. $`\mathrm{}.`$
### 6.4 Computation of the Antiholomorphic Derivative
Corollary 31 implies that we need to compute the antiholomorphic derivative $`\frac{\overline{}}{\overline{\tau }_j}Tr(\mathrm{exp}(t(\mathrm{}_{q1}^^\mathrm{"})\overline{}^1F^{}(q,\varphi _i))`$ in order to finish the proof of Theorem 26. The computations of the antiholomorphic derivative are based on the arguments of Quillen as modified in .
###### Definition 32
We define the function k$`{}_{\tau }{}^{\mathrm{\#}}(w,z,t)`$ in a neighborhood of the diagonal M in MxM as follows: Let $`\rho _\tau `$ be the injectivity radius on M$`_\tau .`$ Let $`d_\tau (w,z)`$ be the distance between the points $`w`$ and $`z`$ on M<sub>τ</sub> with respect to CY metric g<sub>τ</sub> and let $`𝒫_\tau (w,z)(q)`$ be the parallel transport of the bundle $`\mathrm{\Omega }_\tau ^{0,q+1}`$along the minimal geodesic joining the point $`w`$ and $`z`$ with respect to the Levi Cevita connection of the CY metric. We suppose that $`|\tau |<\epsilon .`$ Let $`\delta `$ be such that $`\delta >\rho _\tau .`$ Then we define the function k$`{}_{\tau }{}^{\mathrm{\#}}(w,z,t)`$ as a C function using partition of unity as follows:
k$`{}_{\tau }{}^{\mathrm{\#}}(w,z,t)=\{\begin{array}{cc}(4\pi t)^{\frac{n}{2}}\mathrm{exp}\left(\frac{d_\tau ^2(w,z)}{4t}\right)𝒫_\tau (w,z)(q)\hfill & if\text{ }d_\tau (w,z)<\rho _\tau \hfill \\ 0\hfill & if\text{ }d_\tau (w,z)>\delta .\hfill \end{array}`$
It was proved in on page 87 that we can represent the operator $`\mathrm{exp}(t\mathrm{}_{\tau ,q})`$ by an integral kernel k$`{}_{t}{}^{}(w,z,\tau )`$ where
k$`{}_{t}{}^{}(w,z,\tau )=(4\pi t)^{\frac{n}{2}}\mathrm{exp}(\frac{d_\tau ^2(w,z)}{4t})(𝒫_\tau (w,z)(q)+O(t)).`$
We will denote by M$`{}_{\mathrm{\Delta }}{}^{}`$M$`\times `$M the diagonal in M$`\times `$M. Following the arguments from page 258 of , we will prove the following theorem:
###### Theorem 33
The following formula holds:
$`\frac{}{\tau _i}(\mathrm{log}(det\mathrm{}_{\tau ,q}^\mathrm{"})|_{\tau =0}=\underset{t0}{lim}_\text{M}\left(Tr(\left(k_\tau ^\mathrm{\#}(w,z,t)\right|{}_{\mathrm{Im}(\overline{})}{}^{})F^{^{}}(q,\varphi _i))|_{\tau =0}\right)vol(g(0)).`$
PROOF: An easy calculation, using the fact that $`\frac{\overline{}}{\overline{\tau _i}}(F^{^{}}(\varphi ,q))|_{\tau =0}=0,`$ Theorem 20, and the definition of $`\epsilon (w,z,t)=\mathrm{exp}(\mathrm{}_q)`$k$`{}_{0}{}^{\mathrm{\#}}(w,z,t),`$ show that on the diagonal of MxM for $`t>0`$ we have
$`\frac{}{\tau _i}\mathrm{log}\left(det\mathrm{}_{\tau ,q}^\mathrm{"}\right)|_{\tau =0}=`$
$`\underset{t0}{lim}(_{\text{M}_\mathrm{\Delta }}\left(Tr\left(\mathrm{exp}(t\left(\mathrm{}_{\tau ,q}\right)|{}_{\mathrm{Im}(\overline{})}{}^{}F_{}^{^{}}(q,\varphi _i))|_{\tau =0}\right)vol\right)=`$
$`\underset{t0}{lim}\left(_{\text{M}_\mathrm{\Delta }}(Tr\left(k_\tau ^\mathrm{\#}(w,z,t)\right|{}_{\mathrm{Im}(\overline{})}{}^{})F^{^{}}(q,\varphi _i))vol\right)+`$
$`\underset{t0}{lim}_\text{M}\epsilon _0(w,z,t)|{}_{\mathrm{Im}(\overline{})}{}^{}F_{}^{^{}}(q,\varphi _i)vol=`$
$`\underset{t0}{lim}\left((4\pi t)^{\frac{n}{2}}_{\text{M}_\mathrm{\Delta }}Tr\left(\left(\mathrm{exp}(\frac{d_\tau ^2(w,z)}{4t})𝒫_\tau (w,z)(q)|_{\tau =0}\right)\right|{}_{\mathrm{Im}(\overline{})}{}^{}F^{^{}}(q,\varphi _i))vol\right)`$
$`\underset{t0}{lim}_{\text{M}_\mathrm{\Delta }}Tr(\left(\epsilon _0(w,z,t)\right|{}_{\mathrm{Im}(\overline{})}{}^{})F^{^{}}(q,\varphi _i))vol.`$
On the other hand, the definition of k$`{}_{\tau }{}^{\mathrm{\#}}(w,z,t)`$ implies that $`\mathrm{exp}(\mathrm{}_q)`$k$`{}_{0}{}^{\mathrm{\#}}(w,z,t)=\epsilon _0(w,z,t)`$ is bounded and tends to zero away from the diagonal, as $`t`$ tends to zero. From here we deduce that
$`\underset{t0}{lim}_{\text{M}_\mathrm{\Delta }}Tr(\left(\epsilon _0(w,z,t)\right|{}_{\mathrm{Im}(\overline{})}{}^{})F^{^{}}(q,\varphi _i))vol=0.`$
uniformly in z. Thus, to calculate the limit
$`\underset{t0}{lim}_{\text{M}_\mathrm{\Delta }}Tr\left(\mathrm{exp}(t\mathrm{}_{\tau ,q})\right|{}_{\mathrm{Im}(\overline{})}{}^{}F^{^{}}(q,\varphi _i))|_{\tau =0}vol`$
we may replace the Heat kernel $`\mathrm{exp}(t\mathrm{}\mathrm{"}_{\tau ,q})`$ by its explicit approximation k$`{}_{\tau }{}^{\mathrm{\#}}(w,z,t).`$ So we deduce that
$`\frac{}{\tau _i}(\mathrm{log}(det\mathrm{}_{\tau ,q}^\mathrm{"})|_{\tau =0}=\underset{t0}{lim}_{\text{M}_\mathrm{\Delta }}Tr\left(k_\tau ^\mathrm{\#}(w,z,t)\right|{}_{\mathrm{Im}(\overline{})}{}^{}F^{^{}}(q,\varphi _i))|_{\tau =0}vol.`$
This proves Theorem 33. $`\mathrm{}.`$
###### Corollary 34
Let $`\mathrm{Pr}_q`$ be the projection operator from $`L^2(`$M,$`\mathrm{\Omega }^{0,q})`$ to $`L_{(0,q)}^2(\mathrm{Im}(\overline{}))`$, then we have the following formula:
$`\frac{d^2}{\tau _i\overline{\tau _i}}(\mathrm{log}(det_q^\mathrm{"}(0))=`$
$`\underset{t0}{lim}\frac{1}{\left(4\pi t\right)^n}_\text{M}Tr((\mathrm{Pr}_q(\mathrm{exp}\frac{d_\tau ^2(w,z)}{4t})\delta _j^\mathrm{"}\left(𝒫_\tau (w,z)(q)\right|{}_{\tau =0}{}^{}))F^{^{}}(q,\varphi _i))vol.`$
PROOF: The proof of the corollary follows directly from Theorem 33 and the fact that computation of the trace of a kernel means to restrict the kernel to the diagonal. From here we deduce that: $`\frac{\overline{}}{\overline{\tau _j}}\left(d_\tau ^2(w,z)\right)|_{w=z}=0.`$ Now Corollary34 follows directly. $`\mathrm{}.`$
###### Theorem 35
$`\frac{\overline{}}{\overline{\tau _j}}𝒫_\tau (0,z)(q)=\overline{\text{F}^{\text{}}\text{(q},\varphi _j)}+O(\overline{\tau }),`$where F$`^{^{}}`$(q,$`\varphi _j`$) is defined in Definition 13.
PROOF: We will prove the theorem first for q=1. In this case the operator $`𝒫_\tau `$($`w,z)`$ is the parallel transportation for the bundle $`\mathrm{\Omega }^{0,1}`$ and it defines a linear map: $`𝒫_\tau (w,a):\mathrm{\Omega }_{w,\tau }^{0,1}\mathrm{\Omega }_{z,\tau }^{0,1}.`$ Once we prove Theorem 35 for q=1, the general case will follow directly from standard facts from linear algebra.
Since g<sub>τ</sub> is a Kähler metric, the parallel transport operator $`𝒫_\tau (w,z)`$ preserves the splitting of the complexified cotangent bundle of M into (1,0) and (0,1) forms. So the operator $`𝒫_\tau (w,z)`$ maps linearly $`\mathrm{\Omega }_{w,\tau }^{0,1}`$ to $`\mathrm{\Omega }_{z,\tau }^{0,1}.`$ The parallel transportation operators $`𝒫_\tau (w,z)`$ are defined by the Levi-Chevita connection $`_\tau `$ of the metrics g<sub>τ</sub>. We are going to study the local expansion of $`𝒫_\tau (w,z)`$ in terms of $`\tau .`$ It is a standard fact that
$`_\tau =_\tau ^{1,0}+_\tau ^{0,1}=(_\tau \left(g_\tau ^1_\tau g_\tau \right))+\overline{}_\tau .`$
In order to define the parallel transportation between $`\mathrm{\Omega }_{w,\tau }^{0,1}`$ to $`\mathrm{\Omega }_{z,\tau }^{0,1}`$ we need to join the points $`w`$ and $`z`$ by geodesics. We will suppose that $`w`$ and $`z`$ are ”close”. This assumption can be made since we need to compute a trace of an operator given by some kernel. So our computations will be done on the diagonal M$``$M$`\times `$M. So from here it follows that we can joint $`w`$ and $`z`$ with a unique geodesic. The parallel transportation of the (0,1) form $`\eta \mathrm{\Omega }_{\tau =0}^{0,1}|_{\tau \text{N}},`$ from a point $`w`$ to a point $`z`$ is given by solving the equations for fix $`\tau :`$
$`_\tau ^{0,1}\left(\pi _\tau ^{(0,1)}\eta \right):=\overline{_\tau }(\pi _\tau ^{(0,1)}\eta )=0`$
$`_\tau ^{1,0}\left(\pi _\tau ^{(0,1)}\eta \right)=(_\tau \left(g_\tau ^1_\tau g_\tau \right)\left(\pi _\tau ^{(0,1)}\eta (t)\right)=0`$ $`\&`$ $`\eta (0)=\eta .`$
where $`\pi _\tau ^{(1,0)}`$ and $`\pi _\tau ^{(0,1)}`$ are the projection operators on (1,0) and (0,1) forms on the complex manifold M<sub>τ</sub>. Without loss of generality we can assume that $`w=0.`$ So we can write the following expression for the parallel transportation operator:
$`𝒫_\tau (0,z)(\eta )=\pi _\tau ^{(1,0)}(\eta )+\pi _\tau ^{(0,1)}(\eta )+z(_\tau (\eta ))+O(z^2)`$
for point $`z`$M<sub>τ</sub> near the fix point $`0`$M<sub>τ</sub> and $`_\tau `$ is a linear operator depending on $`\tau `$, i.e.: $`_\tau `$:$`\mathrm{\Omega }_{0,\tau }^{(0,1)}\mathrm{\Omega }_{z,\tau }^{(0,1)}.`$ Kodaira-Spencer deformation theory implies that $`\mathrm{\Omega }_\tau ^{1,0}`$ depends holomorphically on $`\tau .`$ This fact implies that: $`\delta _j^\mathrm{"}\left(\pi _\tau ^{(1,0)}\right)=0.`$ So we obtain the following formula:
$`\frac{\overline{}}{\overline{\tau _j}}(𝒫_\tau (0,z)(\eta ))=\delta _j^\mathrm{"}(\pi _\tau ^{(0,1)}(\eta ))+O(z)`$
for $`z`$ near 0. It is easy to see from the definition of the tangent space to a point of the Grassmanian of $`\mathrm{\Omega }_{0,\tau }^{(0,1)}T_0^{}`$ that $`\delta _j^\mathrm{"}(\pi _\tau ^{(0,1)})|{}_{\tau =0}{}^{}=\overline{\varphi _j}.`$ This implies that for $`z=0`$ we get
$`\frac{\overline{}}{\overline{\tau _i}}(𝒫_\tau (0,z)(\eta )|{}_{\tau =0}{}^{}=\frac{\overline{}}{\overline{\tau _i}}(\pi _\tau ^{(0,1)})(\eta )|{}_{\tau =0}{}^{}=\overline{\varphi _j}(\eta ),`$
where $`\varphi _j:C^{\mathrm{}}(M,\mathrm{\Omega }^{1,0})C^{\mathrm{}}(M,\mathrm{\Omega }^{0,1})`$ is the Beltrami operator when $`z=0.`$ So our Theorem 35 is proved for q=1.
In order to prove Theorem 35 for any $`q>1,`$ we to notice that for $`n_1<..<n_q:`$
$`(\overline{dz}^{n_1}.\overline{dz}^{n_q})=_{j=1}^n(1)^{j1}(\overline{dz}^{n_1}.((\overline{dz}^{n_j})).\overline{dz}^{n_q}).`$
From here the last formula in Theorem 35 follows directly once it is established for the case q=1. Our theorem is proved.$`\mathrm{}.`$
We are now ready to end the proof of Theorem 26.
###### Theorem 36
The following formula is true for q$`>`$0:
$`\frac{d^2}{\tau _i\overline{\tau _i}}\left(\mathrm{log}det\mathrm{\Delta }_{\tau ,q1}^\mathrm{"}\right)|_{\tau =0}=Tr\left(F^{^{}}(q,\varphi _i)\overline{F^{^{}}(q,\varphi _j)}\right).`$
PROOF: Corollary 34 and Theorem 35 implies that we have the following formula:
$`\frac{d^2}{\tau _i\overline{\tau _i}}\left(\mathrm{log}det\mathrm{\Delta }_{\tau ,q1}^\mathrm{"}\right)|_{\tau =0}=`$
$`\underset{t0}{lim}\frac{1}{\left(4\pi t\right)^n}_{\text{M}_\mathrm{\Delta }}(Tr\left(\mathrm{Pr}_q((\mathrm{exp}\frac{d_\tau ^2(w,z)}{4t})\delta _j^\mathrm{"}\left(𝒫_\tau (w,z)(q)\right))\right|{}_{\tau =0}{}^{})F^{^{}}(q,\varphi _i))vol=`$
$`=\underset{t0}{lim}\left(\frac{1}{\left(4\pi t\right)^n}_{\text{M}_\mathrm{\Delta }}Tr\left(\mathrm{Pr}_q\left(\mathrm{exp}\frac{d_\tau ^2(w,z)}{4t}\right)\overline{F^{^{}}(q,\varphi _j)}F^{^{}}(q,\varphi _i)\right)vol\right).`$
Since
$`\underset{t0}{lim}\left(\frac{1}{\left(4\pi t\right)^n}Tr\left(\mathrm{Pr}_q(\mathrm{exp}\frac{d_\tau ^2(w,z)}{4t})\right)\right)|{}_{w=z}{}^{}=Dirac(\delta )(z),`$
where $`Dirac(\delta )`$ is the Dirac delta function. So we obtain that
$`\frac{d^2}{\tau _i\overline{\tau _i}}\left(\mathrm{log}\left(det_{q1}^\mathrm{"}(0)\right)\right)|_{\tau =0}=Tr\left(F^{^{}}(q,\varphi _i)\overline{F^{^{}}(q,\varphi _j)}\right)`$
This proves Theorem 36. $`\mathrm{}.`$
In order to end the proof of the Theorem 26 we will need the following lemma from linear algebra:
###### Lemma 37
Let F be a linear map of a vector space V of dimension n. Then the linear operator F$`id`$ on $`^q`$V for 1$`<qn`$ has a trace given by the formula: Tr(F$`id`$)$`=\left(\genfrac{}{}{0pt}{}{n1}{q1}\right)Tr(`$F).
PROOF: The proof is obvious. $`\mathrm{}.`$
###### Theorem 38
$`Tr\left(\text{F}^{^{}}(q,\varphi _i)\overline{\text{F}^{^{}}(q,\varphi _j)}\right)=\left(\genfrac{}{}{0pt}{}{n1}{q1}\right)<\varphi _i,\varphi _j>.`$
PROOF: Applying the variational formula from Theorem 36 for q=n-1 we get that
$`\frac{d^2}{\tau _i\overline{\tau _i}}\left(\mathrm{log}\left(det_{n1}^\mathrm{"}(0)\right)\right)|_{\tau =0}=Tr\left(\text{F}^{^{}}(n,\varphi _i)\overline{F^{^{}}(n,\varphi _j)}\right)=Tr\left(\text{F}^{^{}}(n,\varphi _i)\overline{\text{F}^{^{}}(n,\varphi _j)}\right).`$
It is easy to see that the composition of the maps F$`{}_{}{}^{^{}}(n,\varphi _i)`$ and $`\overline{F^{^{}}(n,\varphi _j)}`$ is defined on the Hilbert space L$`{}_{}{}^{2}(M,\mathrm{\Omega }^{0,n}),`$ i.e.
F$`{}_{}{}^{^{}}(n,\varphi _i)\overline{F^{^{}}(n,\varphi _j)}:`$ L$`{}_{}{}^{2}(M,\mathrm{\Omega }^{0,n})`$L$`{}_{}{}^{2}(M,\mathrm{\Omega }^{0,n}).`$
From Hodge theorem and the definition of CY manifold we deduce that: $`L^2(M,\mathrm{\Omega }^{0,n})=\mathrm{Im}(\overline{})\overline{\omega _\tau }.`$ From here and from Lemma 37 we obtain that
$`\frac{d^2}{\tau _i\overline{\tau _i}}\left(\mathrm{log}\left(det_{n1}^\mathrm{"}(0)\right)\right)|_{\tau =0}=Tr\left(F(1,\varphi _i)\overline{\text{F(1,}\varphi _j)}\right).`$
On the other hand, the Hodge star operator $``$ for CY metric on CY manifold M, gives us a spectral isomorphism between the Hilbert spaces
$`:L^2(`$M,$`\mathrm{\Omega }^{0,0})L^2(`$M,$`\mathrm{\Omega }^{0,n}).`$
Since the antiholomorphic form $`\overline{\omega _\tau }`$ is a parallel form with respect to the Levi-Chevita connection of the CY metric we can deduce that the Hodge $``$ operator gives a spectral isometry between those two spaces. From this fact and the fact that
$`Tr(F(1,\varphi _i)\overline{\text{F(1,}\varphi _j)}).=<\varphi _i,\varphi _j>`$
we conclude that:
$`\frac{d^2}{\tau _i\overline{\tau _i}}\left(\mathrm{log}\left(det_0^\mathrm{"}(0)\right)\right)|_{\tau =0}=\frac{d^2}{\tau _i\overline{\tau _i}}\left(\mathrm{log}\left(det_{n1}^\mathrm{"}(0)\right)\right)|_{\tau =0}=Tr(F(1,\varphi _i)\overline{\text{F(1,}\varphi _j)}).=<\varphi _i,\varphi _j>.`$
This proves Theorem 26 when q=0 and q=n-1. The formula we just proved, i.e.
$`Tr(F(1,\varphi _i)\overline{\text{F(1,}\varphi _j)}).=<\varphi _i,\varphi _j>`$
Theorem 36 and Lemma 37 directly imply Theorem 26 for any q. $`\mathrm{}.`$
## 7 Some Applications.
### 7.1 Computation of the Analytic Torsion
###### Theorem 39
Let M be an odd dimensional CY manifold, then
$`\mathrm{log}(I(M))=2\mathrm{log}(det(\mathrm{}_0)).`$
PROOF: It is a standard fact that
$`\mathrm{log}(I(`$M$`))=_{q=0}^{n=2m+1}(1)^qq\mathrm{log}(det(\mathrm{}_q))=_{q=1}^{n=2m+1}(1)^q\mathrm{log}(det(\mathrm{}_q^{^{}}))`$
From the formulas:
$`\mathrm{log}(det(\mathrm{}_q))=\zeta _\mathrm{}_q^{}(0),`$
$`\zeta _\mathrm{}_q(s)=\frac{1}{\mathrm{\Gamma }(s)}_\text{0}^{\mathrm{}}Tr(\mathrm{exp}(t\mathrm{}_q))t^{s1}𝑑t`$
$`\zeta _\mathrm{}_q^{^{}}(s)=\frac{1}{\mathrm{\Gamma }(s)}_\text{0}^{\mathrm{}}Tr(\mathrm{exp}(t\mathrm{}_q^{^{}}))t^{s1}𝑑t`$
and Theorems 25 & 26 we deduce that $`\mathrm{log}(det(\mathrm{}_q^{^{}}))=\left(\genfrac{}{}{0pt}{}{n1}{q}\right)\mathrm{log}(det(\mathrm{}_0)).`$ From here we obtain that
$`\mathrm{log}(I(`$M$`))=_{q=1}^{n1}(1)^q\left(\genfrac{}{}{0pt}{}{n1}{q}\right)\mathrm{log}(det(\mathrm{}_0))+(1)^{n+1}\mathrm{log}(det(\mathrm{}_0)).`$
From the equality:
$`(11)^{n1}=\left(_{q=0}^{n1}(1)^q\left(\genfrac{}{}{0pt}{}{n1}{q}\right)\right)=0`$
we conclude that $`\mathrm{log}(I(M))=2\mathrm{log}(det(\mathrm{}_0)).`$ Theorem 39 is proved. $`\mathrm{}.`$
### 7.2 Some Invariants of the Short Term Asymptotic Expansion of the Heat Kernel
From the well know fact that for small $`t`$ we have $`Tr(\mathrm{exp}(t\mathrm{}_0))=\frac{vol(N)}{t^n}+\frac{k(g)}{t^{n1}}+..+a_0+h(t,\tau ,\overline{\tau }),`$ we will deduce that:
###### Theorem 40
Suppose that M is a CY manifold and g is a CY metric with a fixed class of cohomology, then the coefficients a<sub>k</sub> for k=0,.,n in the expression defined above are constant which depends only on the CY manifolds and the fixed class of cohomology of the CY metric
#### Idea of the Proof
We know that the moduli space of CY metrics g with fix class of cohomology is the same as the moduli space of complex structures. This follows directly from the uniqueness and existence of the solution of the Calabi problem. See . From here and results of Kodaira it follows that $`Tr(\mathrm{exp}(t\mathrm{}_0))`$ is a smooth function with respect the coordinates $`\tau =(\tau _1,.,\tau _N)`$ of the Kuranishi space $`𝒦`$(M). If we prove that
$`\underset{t0}{lim}\frac{}{\tau }\left(Tr(\mathrm{exp}(t\mathrm{}_0))\right)=\underset{t0}{lim}\frac{}{\tau }(\frac{vol(N)}{t^n}+\frac{k(g)}{t^{n1}}+..+a_0+h(t,\tau ,\overline{\tau }))=c<\mathrm{},`$
then this implies that $`a_k`$ for $`k=0,..,n=dim_{}M`$ are constants on the moduli space.
PROOF: Let $`F^{}(1,\varphi )`$ be the operator defined in Definition 13 and let $`a_{ii}`$ be its trace, then we have:
###### Lemma 41
The following formula is true:
$`\frac{}{\tau }\left(Tr(\mathrm{exp}\left(t\mathrm{}_{\tau ,0}\right))\right)|_{\tau =0}=t_{i=1}^{\mathrm{}}\mathrm{exp}(t\lambda _i)\lambda _ia_{ii}<\mathrm{},`$
for all $`t0.`$ $`\lambda _i`$ are eigen values of the Laplacian $`\mathrm{}_0`$ on M<sub>0</sub> and .
PROOF: According to Theorem 29 the following formula is true:
$`\frac{}{\tau }(Tr(\mathrm{exp}(t\mathrm{}_{\tau ,0}^\mathrm{"}))|_{\tau =0}=tTr(\frac{d}{dt}\left(\mathrm{exp}(t\mathrm{}_0^\mathrm{"})\right)\overline{}^1F^{}(1,\varphi )).`$
According to Theorem 20 we have
$`Tr\left(\frac{d}{dt}\left(\mathrm{exp}\left(t\mathrm{}_0^\mathrm{"}\right)\right)\overline{}^1F^{}(1,\varphi )\right)=t_{i=1}^{\mathrm{}}\lambda _i\mathrm{exp}(t\lambda _i)a_{ii}<\mathrm{}.`$
Lemma 41 is proved. $`\mathrm{}.`$
Since $`\left\{\lambda _n\right\}`$ for $`n1`$ are the eigen values of the Laplacian $`\mathrm{}_0`$ then
$`\underset{n\mathrm{}}{lim}\lambda _n\left((n^{\frac{2}{dimM}})^1\right)=C>0,`$
where dim<sub>C</sub>M is the complex dimension of M. From here and the fact that the kernel of the operator $`\mathrm{\Phi }`$ is a matrix with C coefficients we derive that $`_{i=1}^{\mathrm{}}\lambda _ia_{ii}<\mathrm{}.`$ So Lemma 41 implies that
$`\underset{t0}{lim}\frac{}{\tau }(Tr(\mathrm{exp}(t\mathrm{}_0))=\underset{t0}{lim}t_{i=1}^{\mathrm{}}\lambda _i\mathrm{exp}(t\lambda _i)a_{ii}=0.`$
This will imply that $`\frac{}{\tau }\left(a_k\right)=0,`$ for $`k=0,.,n.`$ Since $`Tr(\mathrm{exp}(t\mathrm{}_0))`$ is a real function when $`t`$ and $`t>0.`$ So $`\frac{}{\tau }\left(a_k\right)=0`$ implies Theorem 40 directly. Theorem 40 is proved. $`\mathrm{}.`$
## 8 The Analytic Torsion on CY Manifolds is Bounded.
###### Theorem 42
Let M be any CY manifold, then 0$`det(\mathrm{}_q)C_q`$ for $`0qn=dimM.`$
### 8.1 Outline of the Proof that of the Ray Singer Torsion on CY Manifolds is bounded
Theorem 25 implies that in order to prove Theorem 42 it is enouph to bound $`det(\mathrm{}_0).`$ The bound of $`det(\mathrm{}_0)`$ is based on the following expression for the zeta function of the Laplacian acting on functions:
$`\zeta _0(s)=\frac{1}{\mathrm{\Gamma }(s)}_0^{\mathrm{}}(Tr(\mathrm{exp}(t\mathrm{}_0))t^{s1}dt=b_0+b_1s+O(s^2).`$
From the definition of $`det(\mathrm{}_0)`$ it follows that $`det(\mathrm{}_0)=\mathrm{exp}(b_1).`$ So if we bound $`b_1`$ Theorem 42 will be proved. The bound $`b_1`$ is based on two facts. The first one is the following asymptotic expansion of the Tr(exp($`t\mathrm{}_0):`$
$`Tr(\mathrm{exp}(t\mathrm{}_0))=\left(_{k=0}^n\frac{a_k}{t^k}\right)+O(t)=\frac{vol(N)}{t^n}+\frac{k(g)}{t^{n1}}+..+a_0+O(t),`$
where n=dim$`{}_{}{}^{}M`$ and k(g) is the scalar curvature of g. See on page 79. The second one is the explicit formula for $`b_1`$ in :
$`b_1=\gamma a_0+_{k=1}^n\frac{a_k}{k}+_0^1\left(Tr(\mathrm{exp}(t\mathrm{}_0))_{k=0}^n\frac{a_k}{t^k}\right)\frac{dt}{t}+_1^{\mathrm{}}Tr(\mathrm{exp}(t\mathrm{}_0))\frac{dt}{t},`$
where $`\gamma `$ is the Euler constant. We will show that $`b_1=C+\psi (t,\tau ,\overline{\tau }),`$ where $`\psi (t,\tau ,\overline{\tau })0.`$ From here we will obtain that $`det(\mathrm{\Delta }_0)\mathrm{exp}(C).`$
### 8.2 Proof of that the Analytic Torsion is Bounded
###### Remark 43
From now on we will consider the following situation: We will restrict our function $`h(t,\tau ,\overline{\tau })`$ on an one dimensional disk in the Teichmüller space $`𝒯`$(M) of M and this disk is defined as follows: $`\varphi ^1(M,T^{1,0})`$ and let
$`\varphi (\tau ):=\varphi \tau +\frac{1}{2}\overline{}^{}G[\varphi (\tau ),\varphi (\tau )].`$
then we know that the Beltrami differential $`\varphi \left(\tau \right)`$ is well defined $`C^{\mathrm{}}`$ section of $`C^{\mathrm{}}(M_0,Hom(\mathrm{\Omega }^{1,0},\mathrm{\Omega }^{0,1}))`$ in a small disk for $`|\tau |<\epsilon `$ in the Teichmüller space $`𝒯`$(M).
###### Theorem 44
$`\mathrm{log}(det(\mathrm{\Delta }_0))=b_1(\tau ,\overline{\tau })=C+\psi (\tau ,\overline{\tau }),`$ where $`C`$ is a constant and $`\psi (\tau ,\overline{\tau })0.`$
PROOF: Let us define $`\psi _1(\tau ,\overline{\tau }):=\frac{\overline{}^2}{\tau \text{ }\overline{\tau }}b_1(\tau ,\overline{\tau }).`$ Theorem 26 implies that $`\psi _1(\tau ,\overline{\tau })0.`$ Let us define
$`\psi (\tau ,\overline{\tau }):=\frac{1}{2\pi \sqrt{1}}_{|w\tau |1}\psi _1(\tau ,\overline{\tau })G(\tau ,w)d(w)\overline{d(w)},`$
where $`G(\tau ,w)=\mathrm{log}|w\tau |`$ is the Green kernel of the Laplacian $`\frac{\overline{}^2}{\tau \text{ }\overline{\tau }}.`$ Clearly since $`\psi _1(\tau ,\overline{\tau })0`$ and $`G(\tau ,w)0`$ for $`|\tau w|1`$ we can conclude that $`\psi (\tau ,\overline{\tau })0.`$ From the definition of the Green kernel we obtain that $`\frac{\overline{}^2}{\tau \text{ }\overline{\tau }}\psi (\tau ,\overline{\tau })=\frac{\overline{}^2}{\tau \text{ }\overline{\tau }}b_1(\tau ,\overline{\tau })=\psi _1(\tau ,\overline{\tau }).`$ This fact implies that b$`{}_{1}{}^{}(\tau ,\overline{\tau })=`$ $`\psi (\tau ,\overline{\tau })+g(\tau )+\overline{g(\tau )},`$ where $`g(\tau )`$ is a complex analytic function in the disk $`D`$ defined in Remark 43.
###### Lemma 45
$`g(\tau )=const.`$
PROOF: According to we have the following expression for $`Tr(\mathrm{exp}(t\mathrm{}_0)):`$
$`Tr(\mathrm{exp}(t\mathrm{}_0))=\left(_{k=0}^n\frac{a_k}{t^k}\right)+O(t)=\frac{vol(N)}{t^n}+\frac{k(g)}{t^{n1}}+..+a_0+h(t,\tau ,\overline{\tau }).`$
According to we have the following formula for $`b_1(\tau ,\overline{\tau }):`$
$`\left(\frac{d}{ds}\zeta _0(s)\right)|_{s=0}=b_1(\tau ,\overline{\tau })=\gamma a_0+_{k=1}^n\frac{a_k}{k}+_0^1\left(Tr(\mathrm{exp}(t\mathrm{}_0))_{k=0}^n\frac{a_k}{t^k}\right)\frac{dt}{t}+_1^{\mathrm{}}Tr(\mathrm{exp}(t\mathrm{}_0))\frac{dt}{t}.`$
Next we will compute $`\frac{d}{d\tau }\left(\frac{d}{ds}\zeta _0(s)\right).`$ We have for large $`s`$ the following formula for
$`\zeta _0(s):=\frac{1}{\mathrm{\Gamma }(s)}_0^1\left(a_0+_{k=1}^n\frac{a_k}{t^k}\right)t^{s1}𝑑t+`$
$`\frac{1}{\mathrm{\Gamma }(s)}\left(_0^1\left(Tr(\mathrm{exp}(t\mathrm{}_0))_{k=0}^n\frac{a_k}{t^k}\right)t^s\frac{dt}{t}+_1^{\mathrm{}}Tr(\mathrm{exp}(t\mathrm{}_0))t^s\frac{dt}{t}\right).`$
So direct computations and Theorem 40 show that we have
$`\frac{d}{d\tau }\zeta _0(s)=\frac{d}{d\tau }\frac{1}{\mathrm{\Gamma }(s)}\left(_0^1Tr(\mathrm{exp}(t\mathrm{}_0))t^s\frac{dt}{t}+_1^{\mathrm{}}Tr(\mathrm{exp}(t\mathrm{}_0))t^s\frac{dt}{t}\right)=`$
$`\frac{d}{d\tau }\frac{1}{\mathrm{\Gamma }(s)}_0^{\mathrm{}}Tr(\mathrm{exp}(t\mathrm{}_0))t^s\frac{dt}{t}=\frac{1}{\mathrm{\Gamma }(s)}_0^{\mathrm{}}Tr(\frac{d}{d\tau }\mathrm{exp}(t\mathrm{}_0))t^s\frac{dt}{t}`$
According to Lemma 41:
$`\frac{d}{d\tau }\left(Tr(\mathrm{exp}\left(t\mathrm{}_\tau \right))\right)=t_{i=1}^{\mathrm{}}\lambda _i\mathrm{exp}(t\lambda _i)a_{ii}<\mathrm{}.`$
where $`Tr(F(1,\varphi )=_{i=1}^{\mathrm{}}a_{ii}.`$ Combining these facts we conclude that
$`\frac{d}{d\tau }b_1(\tau ,\overline{\tau }))=\frac{d}{ds}(\zeta _{\mathrm{\Delta }_0}(s))|_{s=0}=\frac{d}{ds}\left(\frac{1}{\mathrm{\Gamma }(s)}_0^{\mathrm{}}\left(Tr(\frac{d}{d\tau }\mathrm{exp}(t\mathrm{}_0))\right)t^s\frac{dt}{t}\right)|_{s=0}=`$
$`\frac{d}{ds}\left(\frac{1}{\mathrm{\Gamma }(s)}_0^{\mathrm{}}\left(t_{i=1}^{\mathrm{}}\lambda _i\mathrm{exp}(t\lambda _i)a_{ii}\right)t^s\frac{dt}{t}\right)|_{s=0}=\frac{d}{ds}\left(_{i=1}^{\mathrm{}}\lambda _i^sa_{ii}\right)|_{s=0}=_{i=1}^{\mathrm{}}\mathrm{log}\left(\lambda _i\right)a_{ii}.`$
Kodaira proved that the positive eigen values of the Laplacians $`\overline{}_\tau `$ $`\overline{}_\tau `$ depend on a $`C^{\mathrm{}}`$ manner in a small neighborhood of $`\tau _0D.`$ See . From here and the formula:
$`\frac{d}{d\tau }b_1(\tau ,\overline{\tau }))=\frac{d}{ds}(\zeta _{\mathrm{\Delta }_0}(s))|_{s=0}=\frac{d}{d\tau }\left(g(\tau )\right)+\frac{d}{d\tau }\left(\psi (t,\tau ,\overline{\tau })\right)=_{i=1}^{\mathrm{}}\mathrm{log}(\lambda _i)a_{ii}`$
we can conclude that $`\frac{d}{d\tau }\left(g(\tau )\right)=0.`$ Lemma 45 is proved. $`\mathrm{}.`$ Lemma 45 implies Theorem 42 directly. $`\mathrm{}.`$
|
warning/0004/hep-th0004037.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The charged black hole solutions of string theory in four dimensions have been investigated for different purposes. From the point of view of General Relativity (GR) they generalize the Reissner-Nordstrom (RN) solutions of the Einstein-Maxwell theory. From the point of view of string theory they represent low-energy geometrical structures that should give some information about the fundamental string dynamics .
More recently, these solutions have become interesting also from the point view of the Anti-de Sitter (AdS)/Conformal Field Theory (CFT) duality . In fact, it is well known that in the near-horizon, near-extremal regime the RN-like charged black hole solutions of string theory behave like $`AdS_2\times S^2`$. They can be, therefore, very useful for trying to understand better the puzzles of the AdS<sub>2</sub>/CFT<sub>1</sub> duality . In Ref. a detailed study was performed about the arising of $`AdS_2`$ as near-horizon geometry of the RN solution of the Einstein-Maxwell theory. It was found that the finite-energy excitations of $`AdS_2`$ are suppressed. Only zero-energy configuration survive, whose degeneration should, in principle, be able to explain the entropy of the near-extremal RN black hole.
In this paper we extend the discussion of Ref. to the black hole solutions in the context of heterotic string theory. Owing to the string/string /string triality of heterotic, type IIA and type IIB strings in four dimensions , our discussion also holds for the black hole solutions of type IIA and type IIB string theory. It is well known that the most general solution of this kind represents a generalization of the RN black hole. The new feature with respect to the RN case is represented by the presence of scalar fields (the dilaton and the moduli).
Using duality symmetry arguments we argue that the relevant information about the solutions is encoded in a single-scalar single-U(1)-field solution. We analyze in detail both from the geometrical and thermodynamical point of view this black hole solution. We find that, whereas the geometrical structure of the the near-horizon solutions is the same as in the pure RN case, the presence of the dilaton allows both for solutions with constant dilaton and solutions with non-constant dilaton. Whereas in the former case finite-energy excitations of $`AdS_2`$ are still suppressed in the latter they are allowed.
The structure of the paper is the following. In Sect. 2 we show as the general black hole solution of heterotic string theory can be written as a single-scalar single-U(1) field solutions with RN causal structure. In Sect. 3 we discuss the geometrical and the thermodynamical behaviour of the solutions in the near-extremal, near-horizon limit.
## 2 Black hole solutions
The truncated version of the bosonic action for the heterotic string compactified on a six-torus is the following
$`S`$ $`=`$ $`{\displaystyle \frac{1}{16\pi G}}{\displaystyle }d^4x\sqrt{g}\{R{\displaystyle \frac{1}{2}}[(\eta )^2+(\tau )^2+(\rho )^2]`$ (1)
$`{\displaystyle \frac{e^\eta }{4}}[e^{\tau \rho }F_1^2+e^{\tau +\rho }F_2^2+e^{\tau +\rho }F_3^2+e^{\tau \rho }F_4^2]\}.`$
In this action, we have set to zero the axion fields and all the $`U(1)`$ fields but four, two Kaluza-Klein fields $`F_1`$, $`F_2`$ and two winding modes $`F_3`$, $`F_4`$. The scalar fields are related to the standard definitions of the string coupling, Kähler form and complex structure of the torus,
$$e^\eta =ImSe^\tau =ImTe^\rho =ImU.$$
(2)
The most general, non-extremal (dyonic) solution in the Einstein-Hilbert frame is given by
$`ds^2`$ $`=`$ $`\left(H_1H_2H_3H_4\right)^{1/2}fdt^2+\left(H_1H_2H_3H_4\right)^{1/2}\left(f^1dr^2+r^2d\mathrm{\Omega }_2^2\right),`$
$`e^{2\eta }`$ $`=`$ $`{\displaystyle \frac{H_1H_3}{H_2H_4}},e^{2\tau }={\displaystyle \frac{H_1H_4}{H_2H_3}},e^{2\rho }={\displaystyle \frac{H_1H_2}{H_3H_4}}`$
$`F_1`$ $`=`$ $`dH_1dt,\stackrel{~}{F}_2=dH_2dt,F_3=dH_3dt,\stackrel{~}{F}_4=dH_4dt`$ (3)
where $`\stackrel{~}{F}_2=e^{\eta \tau +\rho }{}_{}{}^{}F_{2}^{},\stackrel{~}{F}_4=e^{\eta +\tau \rho }{}_{}{}^{}F_{4}^{}`$ ( denotes the Hodge dual) and $`H_i,f`$, $`i=1\mathrm{}4`$, are given in terms of harmonic functions,
$$H_i=\left(g_i+\frac{\mu \mathrm{sinh}^2\alpha _i}{r}\right)^1,f=1\frac{\mu }{r}.$$
(4)
The extremal limit is obtained by
$$\mu 0\mathrm{sinh}^2\alpha _i\mathrm{}\mu \mathrm{sinh}^2\alpha _iq_i.$$
(5)
For particular values of the parameters the solutions (3) can be put in correspondence with the solutions of the effective dilaton gravity action
$$S_{eff}=\frac{1}{16\pi G}d^4x\sqrt{g}\left[R2\left(\mathrm{\Phi }\right)^2+e^{2a\mathrm{\Phi }}F^2\right]$$
(6)
with $`a`$ given by one of the following four values
$$a=0a=\frac{1}{\sqrt{3}}a=1a=\sqrt{3}$$
(7)
The case $`a=0`$ describes the Reissner-Nordström black hole of GR and corresponds to $`H_1=H_2=H_3=H_4`$ in Eq. (3), whereas $`a=\sqrt{3},a=1a=1/\sqrt{3}`$ correspond, respectively, to $`H_2=H_3=H_4=1`$, $`H_1=H_2,H_3=H_4=1`$, $`H_1=H_2=H_3,H_4=1`$.
A very interesting proposal is the so called compositeness idea, according to which the $`a=\sqrt{3}`$ solution can be seen as a fundamental state of which the other solutions are bound states with zero binding energy . This idea stems basically from the higher dimensional interpretation of black holes as intersections of D-branes. The number of individual components is denoted by $`n`$. Elementary $`n=1`$ solutions correspond to dilaton gravity theories with $`a=\sqrt{3}`$; $`n=2`$ bound states correspond to $`a=1`$ solutions, $`n=3`$ to $`a=1/\sqrt{3}`$ and finally $`n=4`$ correspond to $`a=0`$. Moreover, the compositeness idea has been used together with the duality symmetries (in particular the $`O(3,Z)`$ duality group) of the model to generate the whole spectrum of BPS states .
The purpose of this paper is to study in detail the thermodynamical and geometrical behaviour of the stringy composite black holes (3) and to establish which relation they have towards Reissner-Nordström black holes.
The general solution (3) is very complicated to study. It contains nine arbitrary integration constants (four moduli $`g_i`$, four $`U(1)`$-charges and the mass); thanks to the $`O(3,Z)`$ duality symmetry of the model, all the relevant information about the nature of these black holes can be obtained by studying some simplified models which we get as we move in the moduli and in the charge space. Following the spirit of the compositeness idea, we can study those solutions we get if we equate some of the charges and of the moduli. In this way we will construct solutions with $`\alpha _i0`$ that describe single-scalar single-U(1)-field black holes and whose strong (or weak) coupling regime is exactly given by the dilaton gravity solutions of the model (6).
This can be done in a systematic way by exploiting the $`O(3,Z)`$ duality symmetry of the model in a way similar to that followed in Ref. in dealing with the case of some null charges. The solutions can be characterized by giving the number $`m`$ of equal moduli and charges (we consider only solutions with the same number of equal charges and moduli). It is evident that $`m`$ is invariant under the action of the $`O(3,Z)`$ duality group described in Ref. . It can be therefore used to label different representations of the duality group. Because $`m=1`$ is equivalent to $`m=3`$ we will have three multiplets on which the duality group $`O(3,Z)`$ will act by changing the scalar and the $`U(1)`$-field but leaving the geometry of the solution unchanged.
Hence, it will be enough to consider just one representative solution for each multiplet, the whole multiplet can be obtained acting on this solution with the $`O(3,Z)`$ group. Because $`m=4`$ is nothing but the well-known Reissner-Nordstrom solution, in the following we will consider only $`m=N=1,2,3`$. We will set $`\alpha _1=\alpha _2=\alpha _4`$, $`g_1=g_2=g_4`$ for $`N=1,3`$ and $`\alpha _1=\alpha _3`$, $`\alpha _2=\alpha _4`$, $`g_1=g_3`$, $`g_2=g_4`$ for $`N=2`$. Requiring the solution to be asymptotically Minkowskian will impose an additional constraint on the moduli: $`_{i=1}^4g_i=1`$.
The solution can be written in a simple form by introducing the parameters $`\lambda _i`$
$$\lambda _i=\frac{\mu \mathrm{sinh}^2\alpha _i}{g_i},$$
the scalar charge $`\sigma `$ and the parameters $`r_\pm `$, defined as follows:
$`r_{}`$ $`=`$ $`\lambda _1,r_+=\mu +\lambda _1,2\sigma L_P=\lambda _3\lambda _1,\mathrm{for}N=1,`$
$`r_{}`$ $`=`$ $`\lambda _1,r_+=\mu +\lambda _1,2\sigma L_P=\lambda _2\lambda _1,\mathrm{for}N=2,`$
$`r_{}`$ $`=`$ $`\lambda _3,r_+=\mu +\lambda _3,2\sigma L_P=\lambda _1\lambda _3,\mathrm{for}N=3,`$ (8)
where $`L_P`$ is the Planck length. The $`U(1)`$-charges $`q_i`$ can be written in terms of the other parameters as follows (no summation on i)
$$q_i^2=g_i^2\lambda _i\left(\lambda _i+\mu \right)$$
(9)
With the previous positions and performing the coordinate change $`rrr_{}`$ the solution (3) becomes:
$`ds^2`$ $`=`$ $`{\displaystyle \frac{(rr_+)(rr_{})}{r^{(4N)/2}(r+2\sigma L_P)^{N/2}}}dt^2+{\displaystyle \frac{r^{(4N)/2}(r+2\sigma L_P)^{N/2}}{(rr_+)(rr_{})}}dr^2`$
$`+`$ $`r^{(4N)/2}(r+2\sigma L_P)^{N/2}d\mathrm{\Omega }_2^2`$
$`e^\eta `$ $`=`$ $`e^{\eta _0}\left(1+{\displaystyle \frac{2\sigma L_P}{r}}\right)^\gamma ,`$ (10)
where $`\gamma =1`$ for $`N=2`$ and $`\gamma =\pm 1/2`$ for $`N=1,3`$ respectively. For $`N=1,3`$ we also have $`\tau =\rho =\eta `$ while for $`N=2`$ we have $`\tau =\rho =0`$ as can be easily seen from eq. (3). The duality group $`O(3,Z)`$ acting on the solutions (2) with a given $`N`$ generates the corresponding multiplet. The duality acts on the scalars whereas the metric part of the solution remains unchanged. For instance, the $`\tau _S`$ duality ($`\eta \eta ,F_1\stackrel{~}{F}_3,F_3\stackrel{~}{F}_1,F_2\stackrel{~}{F}_4,F_4\stackrel{~}{F}_2`$), acting on the solution $`N=2`$, exchanges the electric solutions with the magnetic ones.
The parameters $`r_+,r_{},\sigma `$ are related to the mass $`M`$ and U(1)-charges $`q_i`$ by the following equations
$$L_P^2Q^2=r_+r_{},2ML_P^2=r_++r_{}+N\sigma L_P,$$
(11)
where we have introduced the adimensional charge $`Q`$ defined as
$$Q\frac{q_i}{L_Pg_i}$$
(12)
where $`i=1`$ for $`N=1,2`$ and $`i=3`$ for $`N=3`$. From eq. (11) follows the relation
$$r_\pm =L_P\left(L_PM\frac{N}{2}\sigma \pm \sqrt{(L_PM\frac{N}{2}\sigma )^2Q^2}\right)$$
(13)
as well as the extremality condition
$$L_PMQ+\frac{N}{2}\sigma .$$
(14)
The solutions (2) represent a four-parameters generalization of the RN solution of GR. In the heterotic string context, the RN solution is described by the $`m=4`$ multiplet, and it is obtained by putting $`N=0`$ in Eq. (2).
One can easily show that the causal structure of the solutions (2) is the same as that of the RN solutions, the effect of the scalar charge $`\sigma `$ being a shift of the $`r=0`$ singularity. Solutions of this kind have been already discussed in the literature .
It is interesting to notice that the bound-state solutions with $`n=1,2,3`$ elementary constituents can be obtained as strong (or weak for the dual solutions) coupling regime of the corresponding $`m=1,2,3`$ solution. In fact, the former solutions are characterized by some vanishing $`U(1)`$-charges, whereas from Eq. (2) and (5) it follows that $`q=0`$ can be achieved by $`g\mathrm{}`$. This fact has a natural explanation if one uses the bound state interpretation of the black holes. Since the bound state has zero binding energy, the mass of the composite state is the sum of the masses $`M_i`$ of the elementary constituents. But $`M_i`$ behaves as $`q_i/g_i`$ so that in the strong coupling regime some of the elementary constituents of the $`n=4`$ solution do not contribute, leaving an effective $`n=1,2,3`$ bound state.
## 3 Thermodynamical behaviour
An important property of the RN black holes is that the semiclassical analysis of their thermodynamical behaviour breaks down very near extremality : the formulae for the entropy and temperature are given by
$`S_{BH}`$ $`=`$ $`{\displaystyle \frac{\pi r_+^2}{L_P^2}}`$
$`T_H`$ $`=`$ $`{\displaystyle \frac{1}{4\pi }}\left({\displaystyle \frac{g_{00}}{r}}\right)_{r=r_+}={\displaystyle \frac{r_+r_{}}{4\pi r_+^2}}.`$ (15)
Near extremality, the excitation energy above extremality $`E`$ and temperature $`T_H`$ are related in the following way
$$E2\pi ^2Q^3T_H^2L_P.$$
(16)
At an excitation energy of
$$E_{gap}=\frac{1}{Q^3L_P}$$
(17)
the semiclassical analysis ceases to be valid. The nature of this breakdown is well understood in string theory: the black hole develops a mass gap and (17) is the energy of its lowest-lying excitation. Maldacena, Michelson and Strominger studied both the geometry and the thermodynamical behaviour of the near-extremal RN black holes in the near-horizon limit. They showed that the spacetime always factorizes as $`AdS_2\times S^2`$, with $`AdS_2`$ endowed with the Robinson-Bertotti metric. Moreover, they found that in this limit all the finite-energy excitations of $`AdS_2`$ are suppressed.
The previous features hold for the RN solution. One would like to know if they represent a general feature of the solutions with $`AdS_2\times S^2`$ near-horizon geometry. Let us therefore consider the near-extremal, near-horizon thermodynamical behaviour of the black hole solutions (2).
The Hawking temperature of these solutions is given by
$$T_H=\frac{r_+r_{}}{4\pi r_+^2}\left(\frac{r_++2\sigma L_P}{r_+}\right)^{N/2},$$
(18)
whereas for the entropy we have,
$$S=\frac{\pi r_+^2}{L_P^2}\left(\frac{r_++2\sigma L_P}{r_+}\right)^{N/2}.$$
(19)
As we are going to see see in detail soon, in the near-extremal, near-horizon regime the solutions (2) always behave as $`AdS_2\times S^2`$. In this situation the energy-temperature relation is
$$E2\pi ^2T_H^2L_PQ^3\left(\frac{Q+2\sigma }{Q}\right)^N$$
(20)
and the energy (17) at which the semiclassical analysis breaks down is given by
$$E_{gap}=\frac{1}{Q^3L_P}\left(\frac{Q}{Q+2\sigma }\right)^N.$$
(21)
Eqs. (20) and (21) represent a generalization to the heterotic string of the formulae (16) and (17) of the pure RN case. The only difference between the two cases is the presence of the scalar charge $`\sigma `$, which is a consequence of the presence of a non-trivial dilaton. The near-horizon limit can be achieved as in Ref. by letting $`L_P0`$ but holding fixed some of the remaining parameters $`E,T_H,Q,\sigma `$. In principle one could also consider more general limits for which $`T_H`$ is not fixed but, as shown in Ref. , these cases present intricate features, in particular the geometry becomes singular. We performed a detailed analysis of the various limits for the three cases ($`N=1,2,3`$) and found out that for finite $`\sigma `$ the results obtained by Maldacena and others for the RN black hole hold true also for these other solutions: the near-horizon geometry is always $`AdS_2\times S^2`$ and there are no finite-energy excitations, despite the presence of the scalar charge. This result was somehow expected but never proved explicitly in literature.
The new feature appears when we take the limit $`L_P0`$ together with $`\sigma \mathrm{}`$, holding $`E,T_H`$ and $`Q`$ fixed. In this case the near-horizon geometry is again $`AdS_2\times S^2`$ and finite energy excitation of $`AdS_2`$ are allowed holding $`L_P\sigma ^N`$ fixed. However in this case $`AdS_2`$ is endowed with a non-constant, linearly varying, dilaton. The resulting two-dimensional model is similar to that corresponding to the $`a=1/\sqrt{3}`$ case in Eq. (6), which has been shown to be also relevant for the $`AdS_2/CFT_1`$ correspondence . We do not discuss here the limit $`L_p0`$, $`Q\mathrm{}`$, $`(T_H,\sigma )`$ fixed because there is nothing new with respect to the pure RN case discussed in Ref. . The relevant limits to be discussed are: (a) $`L_p0`$, $`(T_H,Q,\sigma )`$ fixed; (b) $`L_p0`$, $`\sigma \mathrm{}`$, $`(T_H,Q)`$ fixed. In both cases the near-horizon geometry is $`AdS_2\times S^2`$ but whereas in case (a) the dilaton is constant near the horizon in case (b) we have a non-constant, linearly varying dilaton.
(a) $`L_p0`$, $`(T_H,Q,\sigma )`$ fixed
Defining
$$U=\frac{rr_+}{L_P^2},$$
(22)
and performing the limit in Eq. (2) keeping $`U`$ fixed, we get
$`{\displaystyle \frac{ds^2}{Lp^2}}`$ $`=`$ $`{\displaystyle \frac{U^2+4\pi R^2T_HU}{R^2}}dt^2+{\displaystyle \frac{R^2}{U^2+4\pi R^2T_HU}}dU^2+R^2d\mathrm{\Omega }_2^2,`$
$`e^\eta `$ $`=`$ $`e^{\eta _0}\left({\displaystyle \frac{R}{Q}}\right)^{4\gamma /N},`$ (23)
where $`R`$ is given in terms of the two charges $`\sigma ,Q`$:
$$R=Q\left(1+\frac{2\sigma }{Q}\right)^{N/4},$$
(24)
and $`\gamma `$ is given as in Eq. (2). Hence, in this case the near-horizon geometry is the same as in the pure RN case, the dilaton being a constant near the horizon. What changes is just the radius $`R`$ of the transverse two-sphere. In our case it is a function of both the $`U(1)`$\- and scalar charges. From Eqs. (20) and (21) we obtain that in this limit the excitation energy goes to zero, whereas $`E_{gap}`$ diverges. Analogously to the pure RN case we cannot have finite energy excitations of $`AdS_2`$.
(b) $`L_p0`$, $`\sigma \mathrm{}`$, $`(T_H,Q,E)`$ fixed
From Eqs. (20) and (21) it is evident that we can hold $`(T_H,Q,E)`$ fixed while $`L_p0`$ if we allow $`\sigma \mathrm{}`$, with $`L_p\sigma ^Nconst.`$ Let us define
$$V=\left(\frac{E_{gap}}{L_P^3Q}\right)^{1/2}(rr_+).$$
(25)
Performing the limit while keeping $`V`$ fixed, the solution (2) becomes,
$$\frac{ds^2E_{gap}}{Lp}=(V^2+4\pi T_HV)dt^2+\frac{1}{V^2+4\pi T_HV}dV^2+d\mathrm{\Omega }_2^2,\eta \mathrm{ln}V.$$
(26)
As in the previous case the near-horizon geometry is $`AdS_2\times S^2`$, but now the dilaton is not constant. Thus, heterotic string black holes allow for finite-energy excitations of $`AdS_2`$, but they require a non constant dilaton. This feature has been also noticed in Ref. in the analysis of the pure RN case.
It is interesting to notice that $`AdS_2`$ with a nonconstant dilaton has already emerged as the near-horizon geometry of the $`n=3`$ ($`a=1/\sqrt{3}`$ in Eq. (6)) heterotic string black hole . $`AdS_2`$ endowed with a nonconstant dilaton has been also used to give a realization of the $`AdS_2/CFT_1`$ correspondence. Here we have shown that this kind of models can emerge also as near-horizon limit of RN-like four dimensional geometries. Moreover, the fact that in this context finite energy excitations of $`AdS_2`$ are allowed could be useful to circumvent some of the problems of the pure RN case.
Until now we have considered only solutions with all the four $`U(1)`$-charges different from zero. Let us now consider the solutions we obtain when some of the charges go to zero, i.e the composite black hole solutions with $`n=1,2,3`$. Although the thermodynamics of these black holes has been already discussed in the literature , it is interesting to compare it with the thermodynamical behaviour of the $`N=1,2,3`$ solutions.
The solutions describing bound states of $`n=1,2,3`$ elementary black holes can be obtained as the $`g\mathrm{}`$, strong coupling regime, of the solution with $`N=1,2,3`$. Because of Eq. (2) the corresponding solutions can be obtained putting $`r_{}=0`$ in Eq.(2). The inner horizon disappears and the causal structure of the solutions becomes radically different from that of the RN-like black holes. In the extremal limit the horizon matches the singularity which is timelike for $`n=1`$ and null for $`n=2,3`$.
The Hawking temperature associated with the black hole is
$$T=\frac{1}{4\pi }\left(\frac{g_{00}}{r}\right)_{r=r_+}=\frac{1}{4\pi }\left(r_+\right)^{\frac{n}{2}1}\left(r_++2\sigma L_P\right)^{n/2}.$$
(27)
In the near-extremal limit $`r_+=2EL_P^2`$ and thus we have the following energy-temperature relations
$`E`$ $``$ $`\left(64\pi ^2L_P^3\sigma T^2\right)^1,\mathrm{for}n=1,`$
$`T`$ $``$ $`{\displaystyle \frac{1}{8\pi \sigma L_P}},\mathrm{for}n=2,`$
$`E`$ $``$ $`64\pi ^2T^2L_P\sigma ^3,\mathrm{for}n=3.`$ (28)
For $`n=1`$ the specific heat is negative and this is probably related with the nature of the singularity, which in the extremal limit is timelike. For $`n=2`$ there is no dependence of the excitation energy on the temperature. This behaviour can be explained, at least in principle, in terms of the underlying two-dimensional model . Finally, for $`n=3`$, the energy-temperature relation is similar to that of Eq. (16). This relation indicates that model has a sensible description in terms of an effective two-dimensional model that admits $`AdS_2`$ as solution . The main difference with the RN-like case is that here the dilaton is not constant near the horizon.
|
warning/0004/astro-ph0004051.html
|
ar5iv
|
text
|
# Quasars: a supermassive rotating toroidal black hole interpretation
## 1 INTRODUCTION
The commonly cited method of producing ultra-relativistic bi-directional jets as observed in quasars is the mechanism described by Blandford & Znajek (1978), whereby magnetic field lines thread the poles of a rotating BH as they descend towards the event horizon. Rotational energy may be extracted from the BH by this technique which is ejected in the form of radiation and matter travelling at high velocity along the BH’s spin axis. Critical assessments by Ghosh & Abramowicz (1997) and Livio, Ogilvie & Pringle (1999) suggest that the role of the Blandford-Znajek mechanism has been generally overestimated and inadequately accounts for the larger double radio lobe structures. Numerical simulations indicate that the observed gamma ray energy release along quasar jets is four orders of magnitude more energetic than the Blandford-Znajek mechanism predicts. Issues that are difficult to reconcile with this model are the variability of jet dispersion angles, the finite quasar lifetime and the multiplicity of red-shifts in the very metallic absorption spectra. Any viable alternative model must simultaneously cater for all features.
Speculation concerning the fundamental processes governing quasars invariably involves discussion of compact massive central bodies, the consensus being that these are rotating spheroidal BHs of mass $`10^6`$$`10^9\text{M}_{}`$. Profiles of stellar orbital velocities within AGN haloes lend weight to the premise that massive objects reside at the galactic nuclei. Inactive galactic nuclei (IGN) have yielded comparable velocity profiles, suggesting that the masses of AGN and IGN are similar, if not identical. The vastly differing activity levels could signify an inherent defect with current AGN models. Little theoretical progress seems to have been made over past decades towards a full explanation for these exceedingly energetic phenomena. Closer inspection and revision of existing theories may be necessary to attain a consistent understanding of quasars and AGN. The crucial test of any theory is the correspondence between predictions and observations, it is argued that existing models are struggling in this respect. The purpose of this discussion is to advocate a new model and describe how, with relatively minor theoretical embellishments, compelling explanations for AGN and anisotropic gamma-ray bursts accompanying core-collapse supernovae can be developed. Possible mechanisms responsible for originating, accelerating and collimating jets are then discussed.
The well known black hole uniqueness theorems rest on the classical theory of general relativity. It is thought that any non-distorted and asymptotically flat black hole spacetime can be represented by the Kerr-Newman set of solutions. But, while an ultimate theory remains elusive, one cannot be entirely sure of the validity of the uniqueness theorems. Hence, it is important to explore possibilities beyond those anticipated by purely classical calculations. Small departures from classical physics, e.g. of quantum mechanical origin, might lead to profound macroscopic changes, even to the extent that the topology of black hole horizons can be altered. Similarities between a rotating toroidal black hole (TBH) at the galactic centre accreting matter from its surroundings will be compared with observational evidence from quasars, Seyferts, BL Lacertae and blazars, which have long been suspected to be manifestations of the same underlying astrophysical phenomena. Attention will be paid to the formation of such a TBH, its long-term stability in our universe, jet production and its evolution with time. The possibility is examined that quasars may have been present at some stage of almost every galaxy’s development in the early universe. A diagram encapsulating the life-cycles of toroidal BHs, in qualitative agreement with those of quasars is finally presented.
## 2 ROTATING TOROIDAL BLACK HOLES AND THEIR FORMATION
It is proposed that the central component of the quasar mechanism is a rapidly rotating black hole with a toroidal event horizon. First, the possible embryonics of formation are addressed. The constituent stars of most observable galaxies are concentrated in the plane of galactic rotation. Direct observations of the cosmic microwave background by the COBE satellite indicates that matter was very evenly distributed throughout the cosmos in earlier times. Thereafter, on the scale of inter-galactic distances, matter must have collapsed under the action of gravity, triggering the emergence of protogalaxies composed of low-density hydrogen and helium gas. Because most galaxies are observed to rotate, these protogalaxies would generally have possessed angular momentum.
Protogalactic gas clouds then draw towards the plane of rotation. Molecules have random velocities but collisions are relatively infrequent owing to low particle densities. Those with small velocities tend to accumulate at the galactic centre due to net gravitational attraction. However, these particles gain kinetic energy as they proceed towards the centre, preventing the majority from occupying orbits confined to the centremost regions of the galaxy. Instead they tend to cluster in elliptical orbits of larger radii resulting in a relative underabundance of particles at the core, see Fig. 1, curve labelled $`t=1`$. For a given instant in time, the density within the galactic plane is at a local minimum at the centre, increases with radius to a maximum and thereafter tapers off. As time progresses, the gas distribution becomes more pronounced and collisions between molecules more frequent. The evolution of the distribution in this scenario is qualitatively depicted by the series of gas density curves in Fig. 1. The curve labelled $`t=4`$ represents what may be identified as a toroidal gas cloud. This ultra-low metallicity toroidal proto-star continues to condense as more gas molecules amass until the density and pressure are sufficient for nuclear fusion. The toroidal stars of this model, Fig. 2(a), will exhaust their nuclear fuel exceedingly rapidly as regions suitable for fusion occupy a greater volumetric fraction of toroidal stars than of spherodial stars.
By comparison with the upper size and mass limits for spheroidal neutron stars, an upper limit can be estimated for the minor radius $`R_2`$ of a neutron torus above which collapse to a TBH will result<sup>1</sup><sup>1</sup>1At the time of publication, this TBH formation route seemed to me the most likely, avoiding the need for an SBH$``$TBH transition. Since then, I am of the opinion that the transition can occur in either direction and that the toroidal progenitor is merely an interesting possibility, not an essential ingredient of the model.. Simplifying assumptions are employed. First, the density of neutron degenerate material is assumed to remain constant and independent of pressure as for an incompressible fluid. Hydrostatic equilibrium is reached whereby the pressure of the fluid counteracts gravitational compression at all locations. Newtonian approximations will be used to derive the surface gravity. The torus is assumed to have a major radius much larger than the minor radius, $`R_1R_2`$, so that an infinitely long cylinder approximation is valid. Rotation is neglected. The gravitational field within a sphere of constant density tails off linearly from the surface to the centre, even according to general relativity. It is useful to confirm that gravity within an infinitely long solid cylinder of constant mass density is linearly related to the radial distance from the axis according to a Newtonian analysis.
For points external to spherically symmetric objects, the gravity is known to be equivalent to that of a point particle of equal total mass located at the centre of symmetry, regardless of any radial density variations. Similarly, the external gravity of an infinitely long cylindrically symmetric mass is equivalent to that of a line mass of infinite length located on the axis. To prove this, assign an outer radius to the cylinder of $`R_\mathrm{T}`$, a longitudinal coordinate $`x`$ along the cylinder’s length and angular coordinate $`\vartheta `$. The gravitational field strength at some radius $`a<R_\mathrm{T}`$ inside the cylinder with longitudinal coordinate $`x=0`$ and angular coordinate $`\vartheta =0`$ is sought. This location is external to a cylinder of radius $`a`$ and internal to a cylindrical shell of radial thickness $`ba`$. Because Newtonian gravity obeys the principle of superposition it is first demonstrated that the gravitational field vanishes at all points located within an infinitesimally thin and infinitely long cylindrical shell with constant mass per unit area $`\sigma `$. Then, by integrating the gravitational contribution of internal cylindrical shells, gravity inside an infinite homogeneous cylinder is observed to vary linearly with radius, as is the familiar variation within homogeneous spheres.
Let the thin cylindrical shell have radius $`b>a`$. Integrating the radially directed gravitational field contributions of elemental masses of constant radius $`b`$ over the integration variables $`x`$ and $`\vartheta `$, an expression is obtained of the form:
$`g(a)=2G\sigma {\displaystyle _0^\pi }{\displaystyle _{\mathrm{}}^+\mathrm{}}{\displaystyle \frac{b(ab\mathrm{cos}\vartheta )}{(a^2+b^22ab\mathrm{cos}\vartheta +x^2)^{3/2}}}𝑑x𝑑\vartheta `$ (1)
Integrating with respect to $`x`$ gives:
$$\begin{array}{c}g(a)=2G\sigma \left[\frac{x}{\sqrt{x^2+a^22ab\mathrm{cos}\vartheta +b^2}}\right]_{\mathrm{}}^+\mathrm{}\hfill \\ \hfill \times _0^\pi \left(\frac{abb^2\mathrm{cos}\vartheta }{a^22ab\mathrm{cos}\vartheta +b^2}\right)d\vartheta \\ \hfill \text{thus, }g(a)=4G\sigma _0^\pi \frac{abb^2\mathrm{cos}\vartheta }{a^22ab\mathrm{cos}\vartheta +b^2}𝑑\vartheta \end{array}$$
(2)
Splitting the integral in two, integrating with respect to $`\vartheta `$ and recalling that $`b>a`$ yields:
$`g(a)=4\pi G\sigma ab\left({\displaystyle \frac{1}{|b^2a^2|}}{\displaystyle \frac{1}{b^2a^2}}\right)=0`$ (3)
It is therefore possible to ignore the gravitational contribution of cylindrical shells with radii larger than $`a`$ and consider only the internal cylinder of radius $`a`$. When the previous integral is recalculated for the case where $`a>b`$ it is found that:
$`g(a)={\displaystyle \frac{4\pi G\sigma b}{a}}`$ (4)
To transform to a volumetric calculation, $`\sigma `$ is replaced by a three-dimensional mass density $`\rho `$ and $`g`$ is summed over cylindrical shells from $`b=0`$ to $`b=R_\mathrm{T}`$ to find the surface gravity of a torus in an infinite cylinder approximation:
$`g_\mathrm{T}={\displaystyle _0^{R_\mathrm{T}}}{\displaystyle \frac{4\pi G\rho b}{R_\mathrm{T}}}𝑑b=2\pi G\rho R_\mathrm{T}`$ (5)
which also shows that the gravity within an infinite cylinder varies linearly with radius. Mass density $`\rho `$ is assigned to both the toroidal and spherical neutron stars. The radius of the neutron sphere is $`R_\mathrm{S}`$ and the minor radius of the neutron torus $`R_\mathrm{T}`$. For a sphere, surface gravity $`g_\mathrm{S}`$ can be immediately calculated:
$`g_\mathrm{S}={\displaystyle \frac{GM_{\mathrm{sphere}}}{R_\mathrm{S}^2}}={\displaystyle \frac{4\pi G\rho R_\mathrm{S}}{3}}`$ (6)
By calculating the pressure at the centre of the sphere ($`P_{\mathrm{SC}}`$) and torus ($`P_{\mathrm{TC}}`$), then equating the two values, it will be possible to compare the limiting radii at which further gravitational collapse takes place. The surface pressures are assumed to be zero and integration is performed over infinitesimally thin (spherical or cylindrical) shells of matter. The pressure difference between the inner and outer surface of a shell is given by the weight of the shell divided by the area of the inner shell surface. Noting that the weight of the shell depends on the local value of gravity, which is constant throughout the shell and a linear function of radius from zero at the centre to $`g_\mathrm{S}`$ or $`g_\mathrm{T}`$ at the surface, one can write:
$`P_{\mathrm{SC}}`$ $`={\displaystyle \frac{\rho g_\mathrm{S}}{R_\mathrm{S}}}{\displaystyle _0^{R_\mathrm{S}}}r𝑑r={\displaystyle \frac{\rho g_\mathrm{S}R_\mathrm{S}}{2}}={\displaystyle \frac{2\pi G\rho ^2R_\mathrm{S}^2}{3}}`$ (7)
$`P_{\mathrm{TC}}`$ $`={\displaystyle \frac{\rho g_\mathrm{T}}{R_\mathrm{T}}}{\displaystyle _0^{R_\mathrm{T}}}r𝑑r={\displaystyle \frac{\rho g_\mathrm{T}R_\mathrm{T}}{2}}=\pi G\rho ^2R_\mathrm{T}^2`$ (8)
Equating $`P_{\mathrm{SC}}`$ and $`P_{\mathrm{TC}}`$ allows the determination of an upper limit for the minor radius of a neutron torus in terms of the maximum neutron sphere radius. Note that this result is independent of the density of neutron star matter and that the reliability of the result is improved by the balancing of the Newtonian approximations:
$`R_{\mathrm{T}_{\mathrm{max}}}=\sqrt{{\displaystyle \frac{2}{3}}}\times R_{\mathrm{S}_{\mathrm{max}}}8.5\mathrm{km}`$ (9)
As might be expected, the minor radius of an infinitely long neutron cylinder must be smaller than the maximum spherical radius. In circumstances where the infinite cylinder approximation is invalid, the minor radius will be further constrained. If general relativity were to be used then the pressure gradient for a spherical star would be given by the standard equation describing hydrostatic equilibrium:
$`{\displaystyle \frac{dP}{dr}}=G{\displaystyle \frac{(\rho +P/c^2)[m(r)+4\pi r^3P/c^2]}{r[r2Gm(r)/c^2]}}`$ (10)
Here $`P`$ is the pressure at some radius $`r`$ and $`m(r)`$ is the mass enclosed by the 2-sphere defined by $`r`$, whose internal density may vary with radius according to a chosen equation of state. General relativity requires larger neutron degeneracy pressures if gravity is to be resisted, but the estimate of (9) is adequate for the present discussion.
Confining the discussion to those toroidal stars whose gravitational implosion directly results in a TBH rather than intermediate white dwarf or neutron density phases, the pre-collapse seed star is assumed to be incompressible and of solar density $`1400`$kg m<sup>-3</sup>. For the purposes of approximation, the surface areas of extremal Kerr BH event horizons are equated with the surface areas of TBHs with equal mass, alternatively this may be viewed as equating the entropy of the BHs. The TBH is assumed to have an angular momentum equal to the extremal Kerr BH of equal mass. In addition, the TBH geometry will be taken to be that of an Euclidean torus parameterised by the major and minor radii $`R_1`$ and $`R_2`$ respectively. This crude model permits the preparation of order of magnitude estimates.
An extremally rotating Kerr BH has $`r_+=m`$ so its area is $`A=4\pi r_+^2=4\pi m^2`$. The surface area of an Euclidean torus is $`A=4\pi ^2R_1R_2`$ so, to a good approximation, the rotating TBH mass is related to the TBH area by equating these two expressions for area and, after restoring the natural constants ($`r_+=Gm/c^2`$), it is found:
$`M_{\mathrm{TBH}}{\displaystyle \frac{c^2}{G}}\sqrt{\pi R_1R_2}`$ (11)
Now consider a (low density) toroidal star (TS) of Euclidean geometry whose major radius is $`R_1`$ as before, but with a minor radius $`R_3`$. Evidently $`R_3>R_2`$ otherwise the TS is a TBH and $`R_3<R_1`$ ensures the star is toroidal. The TS undergoes gravitational collapse once its nuclear fuel is exhausted and the resulting TBH is assumed to have the same major radius $`R_1`$ as the TS. Since the volume of the TS is $`V_{\mathrm{TS}}=2\pi R_1R_3^2`$, and the TS is composed of constant density material $`\rho 1400`$kg m<sup>-3</sup> then the mass of the toroidal star will be $`M_{\mathrm{TS}}=2\pi \rho R_1R_3^2`$. Following a supernova (SN) implosion of the star, typically most of the mass will have been ejected. A parameter $`\eta `$ represents the fraction of the original TS mass remaining in the TBH after the SN. The remaining mass is identified with the mass of the resultant TBH so that:
$`{\displaystyle \frac{c^2}{G}}\sqrt{{\displaystyle \frac{R_2}{4\pi R_1}}}\rho \eta R_3^2`$ (12)
Having already determined the maximum minor radius of a neutron torus in (9) this implies:
$`R_1>R_28.5\text{km}`$ (13)
Taking the limit as $`R_3R_1`$ with $`R_3<R_1`$ in (11) and using the relation $`M_{\mathrm{TBH}}=\eta M_{\mathrm{TS}}`$ with $`\eta =0.1`$ (90% mass ejection) gives a limit for TBH formation:
$`{\displaystyle \frac{R_1^5}{R_2}}>{\displaystyle \frac{c^4}{4\pi \rho ^2\eta ^2G^2}}7.4\times 10^{48}\text{m}^4`$ (14)
Allowing $`R_28.5`$km with this condition gives a lower bound for $`R_1`$:
$`R_136\times 10^9\text{m}`$ (15)
If the TS grows too large and too massive, then it will become a TBH without an implosion or electron/neutron degeneracy supported phases. Since the area of the TS is larger than the area of a TBH of the same mass, then $`R_3>R_2`$ because $`R_1`$ is common to both. Consideration of (12) in the case where $`R_3R_2`$ then leads to:
$`R_1R_2^3<{\displaystyle \frac{c^4}{4\pi \rho ^2\eta ^2G^2}}7.4\times 10^{48}\text{m}^4`$ (16)
The SN is assumed to shed 90% of the original star’s mass during the implosion (this assumption is the least reliable and easily dominates the combined errors of the remaining assumptions). In special cases where $`R_2`$ approaches $`R_1`$ and the inequalities hold then almost no mass is lost because the star does not collapse much before the event horizon engulfs it. Since it has been assumed that the major radius is unchanged during collapse, conservation of angular momentum dictates that the angular velocity of the resulting BH will match that of the seed star. Hence, less mass ejection is anticipated than in more familiar SN events wherein a star collapses to form a spheroidal BH with very high angular velocity. For a given $`R_2/R_1`$ ratio, the permissible range of toroidal star masses which can gravitationally collapse to form a TBH range is typically quite broad (Fig. 9). This issue is returned to later.
These massive toroidal stars would have rapidly exhausted their nuclear fuel. Regions suitable for fusion reactions occupy a much larger proportion of the total stellar volume in toroidal stars than in spherical stars. The end result would be a supernova-like implosion, most likely the first SN event of its host galaxy, presumably localised to one portion of the torus initially, Fig. 2(b). Because the implosion is limited by the speed of light, it could take several hours for the implosion to propagate around the torus in both directions, Fig. 2(c), until the implosion fronts meet at the opposite end of the torus. During the implosion a thin tubular event horizon expands along the torus, eventually encountering itself and sealing to provide a stable TBH, Fig. 2(d). The illustrations of Fig. 2 are not based upon precise physical calculations, they are merely intended show the progression of the gravitational implosion around the torus.
The mass of the toroidal star is such that if as much as 90% of its mass is outwardly expelled during the SN implosion, there will still remain enough mass to construct what must inevitably become a BH rather than a neutron star remnant. Suppose that much more of the mass is ejected during the SN, perhaps 99%, then what may remain could conceivably be a toroidal white dwarf or toroidal neutron star. In either case, turbulence and dissipative processes are unlikely to leave these delicate structures unchanged. If macroscopic axisymmetry is retained, e.g. through electromagnetic confinement of the torus, then after a brief period the torus will evolve to a smaller major radius. As this occurs, an increase in either its minor radius, its density or, more likely both ensues. Therefore, toroidal white dwarves could become toroidal neutron stars and toroidal neutron stars could become toroidal black holes. Toroidal neutron stars of masses $`10^6\text{M}_{}`$ are precluded as serious AGN candidates by their limited lifespan, slender geometry and inability to endure sustained accretion. Smith and Mann (1997) have recently investigated gravitational collapse as a TBH formation mechanism starting with collisionless particles of random velocities but zero net angular momentum.
Quasar observations yield spectra with very strong metallic absorption lines. The population II stars of the galactic centre would mainly consist of Hydrogen and Helium, which has previously troubled spheroidal BH quasar models. Because SNe are efficient at generating heavy elements, the TBH creation SN would have scattered a substantial amount of metallic elements into the ambient galactic environment, imparting its signature on the radiation spectrum of the central engine.
## 3 STABILITY OF ROTATING TOROIDAL BLACK HOLES
For some time, following the work of Hawking (1972) and Hawking & Ellis (1973), it was thought that TBHs were unstable, albeit marginally. This somewhat contra-intuitive result assumed that Einstein’s cosmological constant ($`\mathrm{\Lambda }`$) was zero. Numerical computations of collisionless particles resulting in a transient toroidal event horizon (terminating in a sub-extremal Kerr BH) and assuming $`\mathrm{\Lambda }=0`$ were performed by Abrahams et al (1994), Hughes et al (1994) and Shapiro, Teukolsky & Winicour (1995). These results were consistent with the topological censorship theorem of Friedman, Schleich & Witt (1993) which implies that a light ray cannot pass through the central toroidal aperture before the topology becomes spherical. More recently, papers by Huang & Liang (1995), Aminneborg et al (1996), Mann (1997), Vanzo (1997) and Brill (1997) have provided mathematical descriptions of TBHs within the framework of general relativity. These equations assume that the cosmological constant is negatively valued to admit stability for the TBH and is literally constant throughout the spacetime described, which has an anti-de Sitter (AdS) background. The Vanzo paper claims that a TBH can exist in a virtually flat spacetime because the TBH size is determined by the mass and conformal class of the torus, not by the cosmological constant. Rotating charged black (cosmic) strings have been described by Lemos & Zanchin (1996). A spacetime metric for a rotating, uncharged TBH presented by Klemm, Moretti & Vanzo (1998), is hereafter referred to as the KMV metric. This metric is not unique, but it is the first generalisation to admit rotation of TBHs. Holst & Peldan (1997) showed that rotating Banados-Teitelboim-Zanelli (BTZ) BHs cannot be described in terms of a 3+1 split of spacetime, instead spacetimes of non-constant curvature are required.
Physical measurements to date have been unable to establish conclusively whether $`\mathrm{\Lambda }`$ is positive or negative. The accelerating cosmological recession of distant SNe favours a positive value, though whether this recession is attributable to a cosmological constant is the subject of continuing debate. Arguments against TBH stability have assumed that in our universe, the constant is precisely zero everywhere. The weak energy condition is assumed to be satisfied, although it is known to be violated in certain situations e.g. Casimir effect and Hawking radiation. Topological BHs in anti de-Sitter spacetimes are now known not to conflict with the Principle of Topological censorship, for a recent discussion, see Galloway et al (1999). Intuitively, rotating TBHs are not dissimilar to Kerr BHs in that both contain ring singularities whose radii are determined by the angular momentum assuming constant BH mass. One extra parameter is necessary to characterise a stationary TBH in addition to the mass, angular momentum and charge of the Kerr-Newman metric. This parameter determines the exact geometry of the torus and can be expressed as the ratio of the minor and major radii $`R_2/R_1`$ (as used here) or the ratio akin to a Teichmüller parameter presented by the KMV paper. Stationarity is preserved only when this parameter achieves a balance with the TBH mass and angular momentum, and to a lesser degree the charge.
It was demonstrated by Gannon (1976) that for non-stationary BHs in asymptotically flat spacetimes, the topology of the event horizon must be either spherical or toroidal. A rotating TBH located at the centre of a galaxy surrounded by accreting matter is manifestly non-stationary. The stationary BH metrics containing physical singularities are acutely idealised, the Kerr metric contains a ring singularity surrounded by vacuum i.e. a universe devoid of other matter. This is a gross simplification of what would be found in nature. Inside the inner event horizon $`r_{}`$, particles are not compelled to collapse towards the singularity, but are free to explore all radii $`0rr_{}`$. Suppose a Kerr BH forms by the collapse of a non-rotating neutron star. The matter at the surface of the neutron star can reach the singularity in a finite proper time. On the other hand, viewed from infinity, this matter never crosses the event horizon, less still reaches the singularity. According to distant observers, the matter is frozen fractionally above the event horizon. Just as infalling matter experiences the crossing of outer then inner event horizons in finite time, it also witnesses the end of the external universe before nearing the singularity. The only possible answer to the question: “when does the BH become stationary to distant observers?” is never. Indeed, one might venture that truly stationary spacetimes are forbidden. It is dangerous to be guided by predictions about BH stability which rely on stationarity as one of the underlying assumptions.
Perhaps there is some deeper significance underlying the unobtainability of stationarity. Consider a closed universe approaching a big crunch and contracting rapidly in all directions. The surface defining the outer reaches of this universe could be considered as the event horizon of a BH beyond which spacetime does not exist in the usual sense. This is a BH that could conceivably approach stationarity in a short and finite time as measured by the clocks of all internal observers, there being no external observers. A singularity develops which is accessible to all the infalling matter. The outermost layers of the imploding universe catch up with the innermost layers at the Cauchy horizon, the surface of infinite blue shift. A vacuum develops in the region surrounding the singularity as it swiftly becomes devoid of matter and stationarity is achieved. The singularity now contains the entire mass of the pre-collapse universe and the Pauli exclusion principle does not participate in the physics of the singularity. What grounds are there for discarding the Pauli exclusion principle? This principle has successfully predicted the existence of white dwarfs and neutron stars. Could quark degeneracy arise? What might string theory predict? There are obvious similarities between the Pauli exclusion principle and the premise that stationary BHs are forbidden. If it is true that BHs truly abhor stationarity then presumably re-expansion would be the only option.
Suppose that a TBH with a substantial central aperture is rotating in asymptotically flat space with a near maximal angular momentum (event horizon velocity approaching the speed of light). In principle, there is no reason why the rotational energy of this TBH cannot be arbitrarily larger than the TBH’s rest mass, whereas a Kerr BH can only hold at most 29% of its total energy in rotational form, the remainder being the irreducible mass. According to topological censorship, the TBH must become spheroidal before a light ray can traverse the aperture. The fate of the excess rotational energy is something of a conundrum. Is the excess energy hastily expelled by some undiscovered mechanism? Is topological censorship flawed? Would the TBH break up into multiple co-rotating spheroidal BHs? Does the Kerr BH rotate above the extremal limit, and if so is the singularity revealed? These problems can be circumvented for now by assuming a negative $`\mathrm{\Lambda }`$.
It seems somewhat coincidental that the cosmological constant is so nearly zero and not very much larger in value, on purely theoretical grounds a value 120 orders of magnitude greater than observational limits might have been expected. One plausible suggestion was proposed by Coleman (1988). According to the author, macroscopic cancellation mechanisms operate on the zero point energies under normal circumstances and these result in a zero expectation for $`\mathrm{\Lambda }`$. The situation is, however, complicated in the presence of intense gravitational fields generated by BHs, particularly in the immediate vicinity of the singularities residing within the event horizons. Under such conditions, the cancellation of zero point energies operates imperfectly and gives rise to what may be considered a localised but substantial cosmological ‘constant’. By this means, TBH stability could be ensured within a universe where elsewhere $`\mathrm{\Lambda }`$ is small. It was also suspected that zero point energy might play a part in the physics of curved spacetimes because of imperfect cancellations complicating the assumptions underlying the quantum mechanical technique of renormalization (Misner, Thorne & Wheeler 1973). It comes as little surprise that quantum effects may play a prominent role when the spacetime of classical general relativity becomes singular, the stability of the TBH structure could prove to be the only direct evidence of this.
The KMV metric has axial symmetry and the horizons are Riemannian surfaces of constant gravity that obey the familiar BH entropy-area laws. Utilising the membrane paradigm approach (Thorne, Price & MacDonald 1986) simplifies the consideration of the physics of these BHs outside the event horizon. The fact that these objects are thermodynamically well behaved, whilst interesting, is of little relevance to the present discussion. Parallels between rotating TBH solutions and the Kerr solutions for spinning spheroidal BHs may be drawn, for instance both have ergoregions external to their event horizons and the maximum angular momentum of each is bounded for a given mass. Conservation of mass and angular momentum is known to be satisfied. A maximally rotating Kerr BH has a static limit extending to $`2m`$, double the radius of the outer event horizon. The equator of the static limit surface circles with the speed of light at extremality. The ergoregion occupies the region between the static limit and the event horizon within which everything is compelled to co-rotate with the BH due to the spacelike character of the time coordinate. Similarly, the maximal KMV metric determines the ratio $`r_\mathrm{s}/r_+`$ to be 1.59. Orbits within 300% of the extremal Kerr event horizon radius are unstable and matter (the accretion disk) tends to be drawn towards the BH. For a maximally rotating BH, entering the ergoregion becomes impossible because incoming particles would have to travel ‘faster’ than light and possess an infinite amount of energy. Similarly, if the BH is rotating slightly below this rate then only a tiny fraction of the external particles will penetrate the ergoregion, those with very high kinetic energies.
An ergoregion enshrouds the toroidal event horizon of the KMV metric. Fig. 3 depicts a cross-sectional view of a rotating TBH. The event horizon will be enshrouded by an ergoregion which, depending upon the precise geometry of the TBH, might entirely seal the central aperture. Beyond the ergoregion lies what is sometimes referred to as a zone of unstable orbits within which particles are unable to establish repeating orbital patterns by following geodesic pathways. The ergoregion does not intersect the event horizon at any point, as it does at the poles of a Kerr BH. Particles cannot penetrate the ergoregion of a maximally rotating TBH, whichever trajectory is attempted. The maximal rotation rate will not be achieved in practice because the BH is able to reduce its rotation rate by several methods which are relevant to jet formation and several theoretical reasons such as the fact that the internal singularity would become naked, even as viewed from infinity.
## 4 METRIC OF ROTATING TOROIDAL BLACK HOLE
The KMV metric of an uncharged rotating TBH in asymptotically anti de-Sitter (AdS) spacetime is tentatively forwarded as a model for the naturally occurring TBH. The primary reservations concerning the physical applicability of the topological BH metrics in AdS gravity are that $`\mathrm{\Lambda }`$ is assumed to be independent of location and, contrary to the most reliable observations, negative in value. Given the uncertainty regarding the role of quantum mechanics in BH physics, these assumptions may be invalid. The metric describes a vacuum solution of Einstein’s equation which has reached equilibrium after an infinite coordinate time has elapsed. It possesses a ring singularity, but no provision for accreting matter has been made. Indeed, a massive accretion disk may act as a stabilising influence on a TBH within an asymptotically flat spacetime (further discussed in appendix B). A negative cosmological constant may be thought of as contributing a cosmological attraction. In its absence, the combination of a host galaxy’s matter and the nearby massive accretion disk surrounding the outer periphery of a TBH located within a galactic nucleus may provide a natural substitute for the stabilising negative $`\mathrm{\Lambda }`$ used in the AdS metrics.
With these considerations in mind, attention is focused on the KMV metric which, for convenience, is now recalled:
$`ds^2=N^2dt^2+{\displaystyle \frac{\rho ^2}{\mathrm{\Delta }_r}}dr^2+{\displaystyle \frac{\rho ^2}{\mathrm{\Delta }_P}}dP^2+{\displaystyle \frac{\mathrm{\Sigma }^2}{\rho ^2}}(d\varphi \omega dt)^2`$ (17)
where the following relations apply:
$`\rho ^2`$ $`=r^2+a^2P^2`$ (18)
$`\mathrm{\Sigma }^2`$ $`=r^4\mathrm{\Delta }_Pa^2P^4\mathrm{\Delta }_r`$ (19)
$`\mathrm{\Delta }_r`$ $`=a^22mr+r^4/l^2`$ (20)
$`\mathrm{\Delta }_P`$ $`=1+{\displaystyle \frac{a^2P^4}{l^2}}`$ (21)
$`N^2`$ $`={\displaystyle \frac{\rho ^2\mathrm{\Delta }_P\mathrm{\Delta }_r}{\mathrm{\Sigma }^2}}`$ (22)
$`\omega `$ $`={\displaystyle \frac{\mathrm{\Delta }_rP^2+r^2\mathrm{\Delta }_P}{\mathrm{\Sigma }^2}}a`$ (23)
Here, the angular velocity is $`\omega `$, the equatorial angle is $`\varphi `$, $`P`$ is another angular variable with some period $`T`$, $`r`$ is a pseudo-radial coordinate, $`a`$ is the angular momentum per unit mass and $`l`$ is defined as $`\sqrt{3/\mathrm{\Lambda }}`$. The ratio of $`T`$ to $`2\pi `$ is analogous to the Teichmüller parameter describing a flat torus in Riemannian geometry. TBH mass by the ADM definition is $`M=mT/2\pi `$ and angular momentum $`J=Ma`$. The coordinate $`r`$, as in the Boyer-Lindquist form of the Kerr metric, is only a true radial coordinate as $`r\mathrm{}`$ with $`r=0`$ the location of a ring singularity not corresponding to zero radius. Unlike the Kerr BH, as $`r\mathrm{}`$ within the equatorial plane, this point is outside the event horizon. It would be preferable to introduce a coordinate transformation whereby $`r^{}=f(r)`$ such that $`f(\mathrm{})=\mathrm{}`$, $`f(0)=a`$ (say) and $`f(\mathrm{})=0`$ and select $`f(r)`$ such that $`r^{}`$ is an affine parameter, but this is beyond the scope of the present discussion.
In order for the metric to describe a torus, $`P`$ is a periodic variable with period $`T`$ and is covered by four patches $`P=\lambda \mathrm{sin}\vartheta `$ at $`\vartheta =0`$ and $`\vartheta =\pi `$, and $`P=\lambda \mathrm{cos}\vartheta `$ at $`\vartheta =\pi /2`$ and $`\vartheta =3\pi /2`$ where $`\lambda `$ is a constant such that $`T=2\pi \lambda `$. Between these points the behaviour is defined by $`\mathrm{cos}\vartheta `$ being some $`C^{\mathrm{}}`$ function (infinitely differentiable) of $`\mathrm{sin}\vartheta `$ and vice versa.
Upon inspection of the metric, (17), it can be seen that $`\mathrm{\Delta }_r`$ in (20) becomes zero at the event horizon. Inner and outer event horizons exist as real and positive roots of the quartic equation with real coefficients: $`r^42ml^2r+a^2l^2=0`$ along with two other physically less meaningful complex conjugate roots. For the extremally rotating case, $`a=a_c=\sqrt{3}\times \sqrt[3]{lm^2/4}`$ and these two roots coincide. It is straightforward to verify that the real roots are $`r_\pm =\sqrt[3]{ml^2/2}`$.
The ergoregion is defined as the region between the outer event horizon and the static limit hypersurface at which the metric coefficient of $`dt^2`$ vanishes altogether, i.e. $`g^{tt}=\omega ^2\mathrm{\Sigma }^2/\rho ^2N^2`$ which is solved for $`r`$ by another quartic $`r^42ml^2ra^4P^4=0`$. This polynomial has one real and positive root, one real and negative root and two complex conjugate roots. The real positive root has a minimum value of $`\sqrt[3]{ml^2}`$ for $`P=0`$ which is larger than $`r_+`$ by a factor of 1.59. The static limit hypersurface is well separated from the event horizon and, unlike the poles of the Kerr situation, these surfaces are nowhere contiguous. Therefore, a substantial ergoregion is observed.
Examining the equatorial plane by setting $`dP=P=0`$ the metric reduces to:
$`ds^2=\left({\displaystyle \frac{r^2}{l^2}}{\displaystyle \frac{2m}{r}}\right)dt^2+\left({\displaystyle \frac{r^2}{l^2}}{\displaystyle \frac{2m}{r}}+{\displaystyle \frac{a^2}{r^2}}\right)^1dr^2`$ (24)
In order to determine the trajectories of null geodesics within this hypersurface use can be made of the Euler-Lagrange equations with $`K=ds^2/2`$:
$`{\displaystyle \frac{K}{x^a}}{\displaystyle \frac{d}{du}}\left({\displaystyle \frac{K}{\dot{x}^a}}\right)=0`$ (25)
the overdot denoting differentiation with respect to some affine parameter $`u`$. The equatorial metric (24) may be partial differentiated with respect to $`t`$ and $`\varphi `$ respectively then integrated with respect to $`u`$ to give:
$$\left(\frac{2m}{r}\frac{r^2}{l^2}\right)\dot{t}a\dot{\varphi }=𝜶$$
(26)
$$r^2\dot{\varphi }a\dot{t}=𝜷$$
(27)
Where $`𝜶`$ and $`𝜷`$ are constants of integration. A third constant $`𝜸=𝜶/𝜷`$ can be defined and used to relate both equations:
$`\left({\displaystyle \frac{2m}{r}}{\displaystyle \frac{r^2}{l^2}}\right)\dot{t}a\dot{\varphi }=𝜸(r^2\dot{\varphi }a\dot{t})`$ (28)
which upon rearrangement reads:
$`{\displaystyle \frac{d\varphi }{dt}}={\displaystyle \frac{\dot{\varphi }}{\dot{t}}}={\displaystyle \frac{2ml^2r^3+𝜸arl^2}{arl^2+𝜸l^2r^3}}`$ (29)
Next, boundary conditions are imposed by considering the extremal case $`a=a_c`$ for which the angular velocity at the event horizon $`r_+`$ is, using the expressions for $`a_c`$ and $`r_+`$ and noting that $`\omega =a/r^2`$ in the equatorial plane.
$`{\displaystyle \frac{d\varphi }{dt}}|_{r=r_+}=\mathrm{\Omega }_H={\displaystyle \frac{a_c}{r_+^2}}={\displaystyle \frac{\sqrt{3}}{l}}={\displaystyle \frac{2ml^2r_+^3+𝜸ar_+l^2}{ar_+l^2+𝜸l^2r_+^3}}`$ (30)
Some algebra reveals that the constant of proportionality $`𝜸`$ obeys the relations:
$`𝜸={\displaystyle \frac{3\sqrt[3]{2}ml2\sqrt{3}a\sqrt[3]{ml^2}}{\sqrt{3}\sqrt[3]{2}ml^2+2al\sqrt[3]{ml^2}}}={\displaystyle \frac{3r_+^2\sqrt{3}al}{\sqrt{3}lr_+^2+al^2}}`$ (31)
So now the rate of change of $`\varphi `$ with respect to coordinate time $`t`$ is fully determined. Following straight from the metric and the condition that $`ds=0`$ for null geodesics:
$`({\displaystyle \frac{2m}{r}}{\displaystyle \frac{r^2}{l^2}})\dot{t}^2+\left({\displaystyle \frac{r^2}{l^2}}{\displaystyle \frac{2m}{r}}+{\displaystyle \frac{a^2}{r^2}}\right)^1\dot{r}^2+r^2\dot{\varphi }^22a\dot{\varphi }\dot{t}=0`$ (32)
Dividing throughout by $`(dt/du)^2`$ eliminates the affine variable allowing $`dr/dt`$ to be found using the previously derived expression for $`d\varphi /dt`$:
$`{\displaystyle \frac{dr}{dt}}=\sqrt{[2a{\displaystyle \frac{d\varphi }{dt}}r^2{\displaystyle \frac{d\varphi }{dt}}^2+{\displaystyle \frac{r^2}{l^2}}{\displaystyle \frac{2m}{r}}]\left[{\displaystyle \frac{r^2}{l^2}}{\displaystyle \frac{2m}{r}}{\displaystyle \frac{a^2}{r^2}}\right]}`$ (33)
It would be possible to continue this analysis by integrating with respect to $`t`$ for each variable $`r`$ and $`\varphi `$, resulting in cumbersome mathematical terms. It is sufficient for now to say that these equations allow the null congruences of the equatorial plane to be readily determined by numerical methods.
## 5 ROTATING TOROIDAL BLACK HOLE IN ASYMPTOTICALLY FLAT SPACE
In order to visualise a TBH in asymptotically flat space and its effect on local spacetime, a method which approximates the time dilation at locations in space surrounding arbitrarily complex mass configurations is now introduced. First, the time dilation is derived for the Schwarzschild spacetime with metric:
$`ds^2=(1{\displaystyle \frac{2m}{r}})dt^2(1{\displaystyle \frac{2m}{r}})^1dr^2r^2(d\vartheta ^2+\mathrm{sin}^2\vartheta d\vartheta ^2)`$ (34)
The event horizon of this static spacetime occurs when $`g^{rr}`$ becomes infinite, or $`r_+=2m`$ in geometrical units. Consider the time dilation of a stationary particle located at some constant $`r`$, $`\vartheta `$ and $`\varphi `$. The metric interval $`ds`$ can be interpreted as the proper time of particles travelling on timelike paths so that $`d\tau =ds`$. The time dilation may be read from the metric at once as:
$$\frac{d\tau }{dt}=\sqrt{\left(1\frac{2m}{r}\right)}=\sqrt{\left(1\frac{r_+}{r}\right)}=\sqrt{\psi }$$
(35)
$$\text{where }\psi =\left(1\frac{2m}{r}\right)$$
(36)
As $`r\mathrm{}`$ notice that $`d\tau /dt=1`$ whilst $`d\tau /dt`$ decreases towards zero as the event horizon is approached, as expected. Now, the particle is allowed to undergo radial motion $`dr0,d\vartheta =d\varphi =0`$. The metric is divided throughout by $`dt^2`$ and the particle’s radial velocity in local coordinates is given by $`v_p=dr/d\tau `$. This yields a similar equation to the last but with the introduction of a $`v_p`$ dependent term:
$`{\displaystyle \frac{d\tau }{dt}}={\displaystyle \frac{\psi }{\sqrt{\psi +v_p^2}}}`$ (37)
For the Schwarzschild BH, the event horizon and stationary limit coincide at $`r=r_+`$ and $`d\tau /dt`$ becomes zero there. A radial velocity can affect the time dilation but cannot alter the location of the hypersurface at which the time dilation approaches zero. Conversely, in the Kerr case which is now briefly addressed, $`d\tau /dt`$ becomes zero for stationary particles outside the event horizon on the stationary limit, the outermost boundary of the ergosphere. Particles motionless with respect to distant observers will appear to freeze at the static limit but particles in prograde orbits can both penetrate and escape the ergosphere in a finite coordinate time. For retrograde orbits, the time dilation approaches zero at radii beyond the static limit so the location of the stationary limit is meaningful only for particles with zero coordinate velocity. The ergosphere is a zone where some particles are able to travel on spacelike trajectories — these trajectories becoming increasingly probable close to the outer event horizon. Whereas negative energy states are only available within the ergosphere of a Kerr BH, a charged Kerr-Newman BH offers negative energy states beyond the static limit.
Returning to the Schwarzschild metric, situations where the particle undergoes transverse (azimuthal) motion are examined by setting $`dr=0`$ and $`d\vartheta =0`$. Noting that $`v_p=r^2d\varphi /d\tau `$ this leads to:
$`{\displaystyle \frac{d\tau }{dt}}=\sqrt{{\displaystyle \frac{\psi }{1+v_p^2}}}`$ (38)
The next task in this analysis is to derive approximations for the time dilation experienced by observers nearby a moving point mass where the clock at infinity is motionless relative to the nearby observers. This cannot be read directly from the Schwarzschild coordinates since the mass of the singularity is stationary with respect to observers at infinity. By taking the limit as $`m0`$ one obtains $`\psi 1`$ and both the previous equations reduce to the time dilation of special relativity when two objects are in relative motion. These limits are used to introduce a contribution to the time dilation equivalent to inducing a motion of the clocks at infinity. The situation then describes clocks at infinity moving with velocity $`v_p`$, clocks of local observers moving with velocity $`v_p`$ and a motionless point mass. Since all inertial frames are equivalent, one can think of this as stationary clocks and a moving mass with velocity $`v_p`$ in the opposite direction. The various possibilities are depicted in Fig. 4. Condition 6 has been determined by taking the ratio of the expression in condition 4 with the expression in condition 2. Likewise, condition 7 has been determined by taking the ratio of the expression in condition 5 with the expression in condition 3. By taking these ratios, Lorentzian boosts are applied which remove the time dilation contributions of expressions 4 and 5 which were purely due to the relative motions of the clocks. What remains are motionless clocks in the presence of a moving point mass.
The expressions in conditions 2 and 3 are identical implying that time dilation between observers in the absence of gravity is independent of the direction of motion. Conditions 2 and 3 are limiting cases of conditions 4 and 5 respectively in the absence of matter. The parity between the expressions of conditions 1 and 7 suggests that only the component of the mass’s velocity towards the local clock (not the clock at $`\mathrm{}`$ since this is always unaffected by the mass) contributes to the time dilation of the local clock relative to the clock at infinity. Condition 6 can then be used to calculate the time dilation precisely in more general circumstances providing that $`v_p`$ is the velocity component towards the local clock. Note also that there is no requirement for the clocks and the mass to be aligned as they are in Fig. 4, the expressions presented are valid for all configurations owing to the perfect spherical symmetry of the Schwarzschild geometry.
Suppose the Schwarzschild point singularity is subdivided into $`N`$ smaller but not necessarily equal masses, each point mass being located at $`r=0`$, the same spatial position as the parent singularity. In order to accurately recover the time dilation of (35), one is obliged to perform $`N`$ summations of the ratios $`r_+/r`$ where $`r_+`$ relates to the Schwarzschild radius of the mass of each child singularity in turn according to the equation $`r_+=2m_{\mathrm{child}}`$. This will be generalised for the purposes of approximation such that the point masses are not coincident but are located separately in space. Thus the distance $`r`$ will in general be different for each point mass. Restoring natural constants, the following equation is obtained:
$`{\displaystyle \frac{d\tau }{dt}}=\sqrt{1{\displaystyle \frac{2G}{c^2}}\times {\displaystyle \underset{n}{}}{\displaystyle \frac{M_n}{R_n}}}`$ (39)
This may be thought of as a pseudo-principle of superposition and these results may be used to approximate an asymptotically flat spacetime containing a ring singularity. Firstly, the discussion is confined to a momentarily stationary ring singularity i.e. one with zero angular velocity and a radius $`R_1`$ whose derivative with respect to time is momentarily zero. An expression for the time dilation relative to observers at infinity experienced by a spatially fixed observer due to the momentarily motionless ring singularity is derived. The singularity is assigned a constant mass per unit length $`b`$ and radius $`a`$ such that the total mass is $`2\pi ab`$. The time dilation within the plane of the ring is first considered. By symmetry, the only independent coordinate is the radius $`r`$ and the time dilation $`d\tau /dt`$ at that point is approximated by:
$`{\displaystyle \frac{d\tau }{dt}}=\sqrt{1{\displaystyle \frac{2abG}{c^2}}{\displaystyle _0^{2\pi }}{\displaystyle \frac{d\varphi }{\sqrt{r^2+a^2+2ar\mathrm{cos}\varphi }}}}`$ (40)
Setting $`\phi =\varphi /2`$ the time dilation can be expressed in terms of a complete elliptic integral of the first kind, K(k):
$$\frac{d\tau }{dt}=\sqrt{1\frac{8abG}{(a+r)c^2}_0^{\pi /2}\frac{d\phi }{\sqrt{1k^2\mathrm{sin}^2\phi }}}$$
(41)
$$\text{o}r\frac{d\tau }{dt}=\sqrt{1\frac{8abG}{(a+r)c^2}K(k)}\text{ where }k=\frac{2\sqrt{ar}}{a+r}$$
(42)
The point at the centre of the ring ($`r=0`$) is a special case which is readily integrated to give:
$`{\displaystyle \frac{d\tau }{dt}}=\sqrt{1{\displaystyle \frac{4\pi bG}{c^2}}}`$ (43)
In order to describe a TBH, the ring density $`b`$ must be smaller than $`c^2/4\pi G`$. By writing $`(\sqrt{a}\sqrt{r})^20`$ and expanding it is found that $`k1`$ in all cases of (42) satisfying the requirements of the elliptic integral, $`k`$ being the ratio of the geometric and arithmetic means of the parameters $`r`$ and $`a`$. For the static case, the term within the square root becomes zero at the event horizon. If rotation is allowed, the time dilation will become infinite not at the event horizon but at the static limit, the external boundary of the ergoregion where the invariant interval of motionless particles is lightlike.
When the ring singularity rotates with constant angular velocity $`\omega `$, the velocity of a point on the ring is taken to be $`v_r=a\omega `$. Consider the component of this velocity directed towards the observer $`P`$ situated at some radius $`r`$ from the centre of the ring and within the equatorial plane. This component contributes to the time dilation experienced by the observer according to the expression presented in condition 6 of Fig. 4. The terms within the square root causing a deviation from parity of proper and coordinate time are once more summed. The ring’s total mass $`2\pi ab`$ as before. Assuming the centre of mass to be located at the centre of symmetry, the following estimate of the ring’s angular velocity shall be used: $`a^2\omega ^22\pi bG`$ so that the angular velocity is $`\omega \sqrt{2\pi bG}/a`$. Recalling the expression for time dilation of condition 6 and substituting $`\psi =1(2m/r)=1r_+/r`$ then rewriting in such a way as to give a separate and integrable deviation from unity within the square root gives:
$`{\displaystyle \frac{d\tau }{dt}}=\psi \sqrt{{\displaystyle \frac{1+v_p^2}{\psi +v_p^2}}}=\sqrt{1{\displaystyle \frac{r_+}{r}}\left[{\displaystyle \frac{(1+2v_p^2)\frac{r_+}{r}(1+v_p^2)}{1+v_p^2(\frac{r_+}{r})}}\right]}`$ (44)
As $`rr_+`$, $`d\tau /dt0`$ which means only particles travelling at the speed of light can remain on the horizon, as expected. As it stands, this formula allows the deviation from unity within the square root to be summed for an arbitrarily large number of point masses, regardless of the mass contained by each. Simplification is possible if it is assumed that all these point masses are infinitesimally small so that the Schwarzschild radius of each is negligible compared to the distance between each mass and the local clock where $`d\tau /dt`$ is to be determined, $`r_+r`$. Implementing this simplification and including the integration symbol to emphasise the fact that the point masses should be vanishingly small yields:
$`{\displaystyle \frac{d\tau }{dt}}=\sqrt{1{\displaystyle \frac{r_+}{r}\left(\frac{1+2v_p^2}{1+v_p^2}\right)}}`$ (45)
The time dilation relative to observers at spatial infinity is now derived for points surrounding the rotating ring singularity. These test points are assigned cylindrical coordinates $`(r,\varphi ,z)`$, and are not confined to the equatorial plane. By symmetry the $`\varphi `$ coordinate is redundant. The resulting time dilation resembles the previously derived expression containing an elliptic integral but with additional complexity:
$`{\displaystyle \frac{d\tau }{dt}}=\sqrt{12ab{\displaystyle _0^{2\pi }}I(\varphi )𝑑\varphi }`$ (46)
substituting $`u=a^2+r^2+z^2`$ the integrand reads:
$`I(\varphi )={\displaystyle \frac{u+2ar\mathrm{cos}\varphi +2a^2r^2\omega ^2\mathrm{sin}^2\varphi }{\sqrt{u+2ar\mathrm{cos}\varphi }\left(u+2ar\mathrm{cos}\varphi +a^2r^2\omega ^2\mathrm{sin}^2\varphi \right)}}`$ (47)
Note that the time dilation at the centre of the ring ($`r=0,z=0`$) is still given by (43) because the velocity of each point mass is perpendicular to the line connecting the point mass to the local observer and that this holds along the entire axis of rotation. It would be possible, but more complicated, to determine the approximate location of event horizons using a similar method. One would need to transform the local observers to those of a locally non rotating frame (LNRF). This would be achieved in the equatorial plane by relating the angular velocity of the ring to that of the local observers. Starting with condition 6 of Fig. 4, one would generalise to the case where the local clock and point mass have separate (non-zero) velocities with respect to the clock at spatial infinity by applying a Lorentzian boost to the coordinate clock. Then, equivalents to the expression in (43) and (45) would need to be found.
The time dilation can be computed numerically but care is needed when selecting the ring’s angular velocity $`\omega `$ otherwise the situation becomes unphysical with frame-dragging velocities in excess of $`c`$. This formula was used to determine the shape of the ergoregion in Fig. 3 when viewed in cross section. Fig. 5 presents a 3-dimensional projection of the time dilation as viewed by observers located at spatial infinity for the equatorial plane intersecting a rotating TBH. This embedding diagram portrays local time dilation (as viewed by distant observers) due to the presence of mass as the deviation from an otherwise flat plane according to (47). The TBH drags local spacetime with it in synchrony with the event horizon. Accordingly, inertial test particles travelling within the equatorial plane along initially radial geodesics from spatial infinity are compelled to orbit the TBH until their angular velocity reaches that of the event horizon. This occurs at the moment the horizon is crossed. Colour is used to denote the angular velocity of locally non-rotating observers as measured by distant observers, colours of longer wavelengths representing angular velocities approaching that of the TBH. A section of the outer funnel has not been plotted to provide visibility of the TBH aperture region. The ergoregions have not been identified here.
## 6 JET FORMATION FROM TOROIDAL BLACK HOLES
The near-maximally rotating TBH undergoing accretion provides an excellent mechanism for the formation of ultra-relativistic (Lorentz factor $`10`$) bi-directional jets as have been observed in quasars. The purpose of this section is not to explore the behaviour of the jets as they travel towards the distant radio lobes, the magnetohydrodynamics of which has been studied in great detail elsewhere, nor to analyse the myriad of particle interactions capable of extracting rotational energy from the TBH ergoregion. Rather, the essential differences between existing models and the accretion of matter onto a rotating TBH shall be outlined. Supermassive BHs have long been thought to reside at the heart of quasars and active galactic nuclei. Though masses as large as $`10^9\text{M}_{}`$ have been associated with these, a typical rotating galaxy possesses ample angular momentum to spin up BHs of this size to maximal rotation. An upper limit on rotational velocity exists because the peak velocity of the event horizon cannot exceed that of light. In practice, the maximum rotation rate will not be reached, not least because the internal singularity would be revealed. Transportation of angular momentum away from the TBH by jets imposes more practical limitations. Hence the formation of jets, an intriguing feature of many AGN, is of paramount importance. Potential mechanisms responsible for their origination are now presented within the framework of the TBH model.
A nascent galaxy may harbour a TBH whose spin rate is increasing. Once the spin reaches a plateau after a short delay (in cosmological terms), equilibrium is achieved and the accretion process is balanced by the angular momentum released by the TBH due to gravitational radiation, the production of jets and growth through capture of mass and angular momentum. Of these, the outflow of angular momentum is typically dominated by jet generation processes. This maintains a rapidly rotating TBH, but implies that accreting matter rotates with greater angular velocity than the spacetime near the TBH. Essentially, the interaction between this accreting mass and the enormous flywheel of the rotating TBH constitutes the basis for jet energy release. Fig. 6 illustrates a rotating TBH surrounded by an accretion disk. Since the TBH is able to shed any excess angular momentum by several mechanisms, its angular velocity is suppressed relative to the accretion disk $`\omega _{tbh}<\omega _{disk}`$. Apart from gradual mass accumulation, one can picture the TBH as being largely unaffected during periods of sustained activity, acting somewhat like a catalyser for the expulsion of angular momentum along the jet axis. Jet formation can therefore progress for substantial periods of time: $`10^610^9`$yrs.
Within the central aperture, spacetime is dragged in concordance with TBH. The central aperture is a negative gravitational potential well, a spacetime vortex containing deeply negative energy states. Matter negotiating this central aperture will be obliged to travel along geodesics which appear to the external universe to be rotating. Particles capable of escaping to infinity require relativistic velocities closely aligned to the axis of rotation. Matter is able to travel in either direction along the rotational axis in order to achieve this, and angular momentum is transported away from the TBH equally by each jet. Of primary interest is a TBH whose central aperture is sufficiently small to provide powerful, collimated jets.
Matter travelling through the aperture will undergo gravitational slingshot and could be propelled outwards along the jets, however the importance of this will be to impart kinetic energy leading to frictional heating. Temperatures of at least several million degrees (and perhaps as much as $`10^{11}\mathrm{°}`$K) will be realised in the aperture, transforming the contained matter into a plasma emitting X-rays and some gamma rays.
Ejection mechanisms such as the Penrose process (Penrose 1969), superradiant scattering (photonic counterpart of the Penrose process) and their analogues (e.g. due to particle-particle collisions) could dominate in the formation of jets. For convenience, the term Penrose process is used loosely to refer to all variations. The Penrose process exploits the existence of negative energy orbits inside the ergoregion of a rotating BH, permitting the extraction of energy to infinity at the expense of the rotational/kinetic energy of the BH. A particle travelling through the ergosphere might disintegrate into two particles, one of which plunges headlong towards the event horizon whilst the other emerges from the ergosphere and escapes to ‘infinity’, e.g. as part of a jet. Energy is extracted if the emergent particle fragment has more energy than the originally intact particle, with the captured fragment carrying negative energy into the BH. The Penrose process efficiency improves if the particles have relativistic incident velocities, particularly those opposing the BH’s rotation. A physical example of particle disintegration occurs within a high energy plasma when neutral hydrogen atoms are stripped of their electrons. More generally, ergoregional particle-particle collisions in which angular momentum and total energy are conserved may cause one of the resultant particles to be ejected to infinity (Piran, Shaham & Katz 1975; Piran & Shaham 1977). Particles are always ejected in a way that reduces the BH’s angular momentum and rotational energy which increases the surface area of the event horizon, in keeping with the entropy law. As has been described by Wagh, Dhurandhar & Dadhich (1985) and Bhat, Dhurandar & Dadhich (1985) the presence of an electromagnetic field and/or charged particles can dramatically increase the efficiency of the Penrose process, easily to a level where net rotational energy may be extracted from the BH.
The astrophysical significance of the Penrose process has been traditionally questioned, partly because BHs of spherical topology are not expected to retain significant electrical charge. It is argued that tori exhibit a vital difference. When a rotating torus accumulates charge, the circulating current establishes a poloidal magnetic field. Lines of magnetic flux encircle the torus but nowhere intersect its surface. Nearby the surface, flux lines are orthogonal to the current flow and parallel to the surface itself. Sufficiently intense magnetic fields constrain the motion of accreting plasma, obliging its constituent particles to follow helical trajectories which wind about lines of flux. Whereas charged spheroids are rapidly neutralised by plasma guided directly towards the surface by flux lines, neutralisation of charged tori is strongly inhibited due to the absence of flux lines intersecting the surface.
According to the membrane paradigm, one can imagine the TBH’s (infinitesimally stretched) event horizon to be an electrically conducting surface where electric fields incident to this membrane are terminated by an appropriate surface electric charge density. Also, the surface current will be such that the magnetic field parallel to the surface is terminated, in this way there will be no parallel magnetic field inside the event horizon. The effective surface resistivity will be of the order of several hundred ohms. The accreting matter contains neither magnetism nor net charge initially. The TBH is, however, spinning and dragging local spacetime around with it. The charged particles entering the ergoregions are mainly electrons and protons. Low efficiency Penrose processes will preferentially eject particles of larger charge/mass ratios (the electrons) and a net positive circulating charge will emerge hovering above the horizon. The toroidal membrane rotates and drags these positive charges around with it thus forming a circular electrical circuit. Current flowing in the circuit gives rise to an axial magnetic field through the central aperture, modulating the efficiency of particle emissions via the Penrose process thereby reinforcing the circulating charge and magnetic field. This magnetic field also plays a role in collimating the jets as they are launched, with charged particles spiralling along the magnetic field lines generating synchrotron radiation. Similarly, the dipolar magnetic field of the TBH channels free charged particles from the outer accretion disk into the central aperture, spiralling along the lines of magnetic flux.
The structure of the magnetic field encompassing a conducting toroidal shell is shown in Fig. 7. Lines of flux are illustrated which arise when a current circulates around the toroidal shell. Because the aperture can become arbitrarily small, if the total charge of the torus remains constant, the flux density and hence magnetic field can become arbitrarily large within this region. Computed plots of magnetic field strength along the equatorial plane of the torus are given for four separate toroidal geometries in Fig. 8(a). These have been calculated using the usual Biot-Savart relations and assume a constant and uniform surface current density $`J`$. Analytically, the magnetic field strength perpendicular to and within the equatorial plane at some displacement from the axis $`a`$, making use of symmetry, is given by the double integral:
$$B(a)=_0^{2\pi }_0^{2\pi }\frac{\mu _0Jt(a\mathrm{cos}\varphi t\mathrm{cos}2\varphi )}{4\pi (a^2+t^2+R_2^2\mathrm{sin}^2\vartheta 2at\mathrm{cos}\varphi )^{3/2}}𝑑\varphi 𝑑\vartheta $$
(48)
$$\text{where }t=R_1+R_2\mathrm{cos}\vartheta $$
(49)
The Biot-Savart law simplifies at the centre of a circular current loop carrying a current $`I`$ and it is straightforward to verify that the field strength at that point is:
$`B(0)={\displaystyle \frac{\mu _0I}{2R_1}}`$ (50)
Nearby the toroidal surface, Ampere’s circuital law (51) states that the current enclosed by a closed path determines the sum of the magnetic field along the same closed path so the field strength is always finite at the shell’s surface.
$`{\displaystyle B𝑑s}=\mu _0\times I`$ (51)
The same law demonstrates that the integral (48) is independent of $`R_2`$ providing $`a<R_1R_2`$ or $`a>R_1+R_2`$, therefore some simplification is available by setting $`R_20`$ whilst a constant current circulates. The integral of (47) can be expressed in terms of multiple elliptic integrals. Numerical computations have been used to derive plots in Fig. 8(a) which show that the field within the central aperture is generally stronger than in the outer periphery of the torus as measured by local inertial observers. This is particularly true for the geometries where $`R_2R_1`$ that would produce tightly collimated jets. The charged particles spiral around the strong field lines of the aperture achieving high velocities and alternate between contra-rotation and co-rotation during each cycle of their spiral. During the contra-rotation phase, they are especially likely to participate in ergoregional particle collisions in which energy and momentum is transferred to the jets at the expense of the angular momentum of the TBH.
To better approximate the magnetic field of a charged TBH, the gravitational time dilation must be taken into account. Restricting analysis to the time dilation generated by a momentarily static ring singularity whose event horizon is transiently toroidal, the horizon must coincide in the equatorial plane with the surfaces of the electrically conducting toroidal shell. Essentially, this is achieved by precise adjustment of the radius of the ring singularity ($`R_{ring}>R_1`$) and the ring singularity’s mass per unit length. The magnetic field strength plots of Fig. 8(a) are then recalculated, this time taking account of the local time dilation (lapse function) relative to a distant observer. The situation is analogous to earth based quasar observations because this magnetic field directly modulates the non-thermal radiation emanating mainly from within the TBH ergoregion. Note also, the negative energy states within the ergoregion of the central aperture will be more negative than elsewhere within the ergoregion but that the increased time dilation counteracts this. Results are presented in Fig. 8(b). Again, the electrical surface current density $`J`$ has been held constant but the TBH mass is different in each case.
Several factors need to be considered which potentially impact the results. Time dilation diminishes more rapidly than flux density with axial displacement. Improved models would involve non-uniform surface current densities for the TBH membranes, a means of estimating the TBH charge and a more accurate determination of the TBH angular velocity, which may require a theory of quantum gravity.
These facts imply that the Penrose process will occur predominantly within the central aperture of the TBH, and less so in the outer regions. Ergoregional particle emissions in the outer regions are largely reabsorbed by interactions with the accretion disk, whereas the rarefied central aperture allows relatively unimpeded passage to scattered particles. Thus, the most visibly energetic TBHs will be those with tightly focused jets.
As discussed previously, because charge neutralisation is inhibited for tori, intrinsically stronger magnetic fields are to be anticipated in the vicinity of a TBH as compared to the rotating spheroidal BH situation. The presence of the magnetic field and the plasma gives rise to a force-free magnetosphere within the TBH’s central aperture providing the plasma is sufficiently rarefied. The accumulation of circulating positively charged particles moderated by time dilation nearby the TBH event horizon is, for present purposes, identical to the situation where the TBH itself is charged. Supposing that the equilibrium magnetic field stabilises at large values $`10^{12}`$ G or larger, then vacuum breakdown may play a part in the formation of jets. Energy stored in the TBH magnetosphere would then be tunnelled quantum mechanically, creating pairs of charged particles and anti-particles. These virtual particles would then be separated by the intense electromagnetic forces before they could silently annihilate one another. The detection or otherwise of a significant positron population in the jets is a useful tool for resolving the issue of whether vacuum breakdown has a role to play.
Energetic photons (X-rays and gamma rays) generated by the plasma of the central aperture would be emitted in all directions. Ergoregional processes would be capable of promoting them to higher energies because they are travelling at the speed of light, often in retrograde trajectories about the TBH. Resulting photons could inhabit the gamma ray spectrum at energies as high as the TeV range. The Penrose process should also act on the high velocity electrons and atomic nuclei of the high-energy plasma occupying the ergoregion of the aperture. At the expense of the rotational energy of the TBH, bi-directional jets with relativistic velocities are therefore likely to result. Matter and radiation must necessarily emerge from either side of the central aperture, where it begins its journey along one of the two jets. Depending on the geometry of the TBH, the jets could be tightly focused and penetrating or conversely, spluttering weakly over a broad solid angle. Particles are preferentially ejected in close alignment with the modulating magnetic field, in this case along the spin axis of the TBH. The toroidally originated magnetic field provides for deeper negative energy states within the ergoregion whilst extending the region of occurrence well beyond the static limit surface. Bhat, Dhurandar and Dadhich demonstrated that when charged particles are involved in Penrose process interactions there exists virtually no upper bound on the efficiency of energy extraction.
The jets are to some extent able to collimate themselves if they are sufficiently focused at the source by trapping the magnetic field internally. Magnetohydrodynamic studies have had much success in explaining the characteristics of these outflows which emerge supersonically and travel for several millions of light years before ploughing into radio lobes formed at the ICM/jet interface. The knots frequently visible along jets are readily interpreted as the result of substantial short-term matter ingestion from stellar collisions with the TBH or instabilities in the accretion flow, one further possibility being that these knots may also be related to rapid fluctuations in TBH charge and instabilities in the TBH mechanism itself. The rotating jets will cause a net outflow of angular momentum from the TBH, which is counterbalanced by the net inflow of angular momentum due to accretion around the TBH’s periphery. Jets transport angular momentum from the TBH because particles ejected from the ergoregions are rotating with the TBH having been launched by the Penrose process within the ergoregion, thereby generating a decelerational torque (recoil) on the TBH. By the mechanisms described, a significant portion of the mass and kinetic energy of accreting matter and radiation is available for jet production. For detailed analysis, numerical simulations will be required.
The Penrose process reaches maximum efficiency when one of the particles heads directly towards the event horizon along the shortest path (i.e. it has the most negative energy state possible). Similarly, when the negative energy state arises due to the presence of charge on a particle, one of the particles emerging from the collision ideally scatters towards the event horizon along the shortest path. When the trajectory of the other scattered particle is considered for the purely gravitational Penrose process, the particle will head directly away from the event horizon, which for the TBH central aperture is typically a poor escape route. The situation is altered for electromagnetically dominated Penrose process interactions as the potential of the charged particle within an electric field should be considered. In order to recoil with maximal energy extraction, the charged particle will follow a path that leads towards greatest electrical potential which, for the aperture of a charged and rotating TBH, is aligned axially with the magnetic field. These ejected particles will frequently collide with the accretion flow streaming from the outermost periphery of the TBH. The jets will be sufficiently strong to overcome this inward accretion flow in the regions nearest the spin axis. An almost identical scenario was analysed (Blandford & Rees, 1974) wherein hot, relativistic plasma escapes anisotropically through orifices punctured in a cool surrounding gas resulting in beams of collimated plasma.
Data gleaned from quasar observations is consistent with the present TBH model. High energy gamma rays at energies up to $`20`$GeV $`1`$Tev have been detected within the jets. These may correspond to photons ejected by the Penrose process, rather than by some secondary acceleration mechanism within the jets. The variation in jet dispersion angles is related to the $`R_2/R_1`$ ratio of the TBH and is naturally accounted for by the opening angle of the TBH. Quasar spectra can contain three separate red-shifted portions; the TBH model displays a similar complexity:
* The plasma of the central aperture is buried in a deep gravitational potential well.
* Jets travel relativistically in opposite directions, one of which is usually not directly detectable.
* Radiation passes through the metal enriched clouds generated by the SN of the TBH creation event.
* The remoteness of the QSO galaxy correlates to a cosmological recession and red-shift.
A carefully considered numerical study of gravitating fluids (Marcus, Press & Teukolsky 1977) reveals a bifurcation from the Maclaurin ellipsoids to lower energy state ’Maclaurin toroids’ at high angular momenta which the authors suggest may be stable against all small perturbations. Alternatively, toroidal density distributions (TBHs or transient neutron tori) may develop in the dynamically collapsing cores of moderately rotating progenitor stars (rotary core collapse). Butterworth & Ipser (1976) demonstrated that ergoregions can form when relativistic stars spin rapidly although absolute event horizons are absent. With the notable exception of long-term stability, charged neutron tori could share many similarities with the charged TBH central engine of quasars. Although the neutron torus could exhibit various non-axisymmetric instabilities, it appears that the electromagnetic structure could significantly counteract these effects for core-collapse timescales if electrical charge gathers on the torus. Moreover, stability will clearly be reinforced if the composition of the torus becomes superfluidic and superconducting, as is likely for the least soft equations of state.
Rotating, electrically charged neutron tori are able to generate immensely strong axial magnetic fields. When ergoregions arise and accreting matter is available, relativistic bi-directional jets should emerge. Given a steady supply of mass and angular momentum, as from an accretion disk fuelled by a binary companion star, microquasar behaviour could result — though long-term stability arguments favour a TBH engine. Certain explosive core-collapse SNe (hypernovae) accompanied by anisotropic gamma-ray bursts (GRBs) could be interpreted as the outcome of a neutron/BH torus forming during rotary core collapse. Metastable configurations can be envisaged in which a white dwarf accretes matter from a companion star, the angular momentum accumulates until the core eventually collapses to a neutron torus causing a brief but intense outburst of mass and angular momentum sufficient for the star to resume its quiescent spheroidal white dwarf state. Recovery from neutron to white dwarf densities is permitted, but event horizons grow irreversibly. Rotary core collapse may curtail the birth of TBHs at masses substantially $`below`$ the lower limits commonly adopted for BHs.
## 7 EVOLUTION BEYOND OPTICALLY BRIGHT QSO PHASE
It is well documented that quasar populations decline rapidly at co-moving red-shifts below $`z2.5`$. What is not easily explained by leading models reliant on a central supermassive spheroidal BH is why quasar activity terminates so abruptly in recent times. This is especially puzzling given that the BHs themselves cannot have altered other than gain yet more mass from their surroundings. Ideas have been proposed such as advection dominated accretion flow (ADAF), the ‘spin’ paradigm and various ‘state transitions’ related to accretion efficiency. A significant diminution of BH angular momentum is unlikely given the tendency of accretion disks to transfer rotational energy to a BH. If fundamentally different modes of accretion do operate, lower efficiency modes will be obliterated whenever a massive body approaches the disk/engine environment. Mass injection of this kind would sporadically re-establish brief quasar behaviour in previously dormant galaxies, including neighbouring galaxies which are observed to be emphatically inactive. Evidence for such transient activity is absent both in local and distant galaxies. This is exceedingly troublesome for standard AGN models when accounting for the quiescent cores of nearby giant ellipticals.
According to the present model, the TBH will transition to a spheroidal BH once accretion has inflated the event horizon or decreased the angular momentum sufficiently. This provides a natural mechanism for the termination of quasar-like TBH activity within the universe and is amply supported by observations. The swelling of the toroidal event horizon due to mass capture generally overcomes the increasing angular momentum of the TBH by the same process. Although the major radius of the torus may be increasing, the minor radius will eventually catch up leading to a topological transition. Immediately prior to the extinction of the TBH, the most energetic and tightly collimated jets are anticipated to form, albeit with enhanced gravitational red-shifts as seen from infinity.
Fig. 9 illustrates a number of features of the TBH quasar model and has been constructed using the inequality relations (12) to (15) from section 2. The shaded wedge represents the area within which TBHs can come into being directly from the implosion of a toroidal star. Here, it has been assumed that the seed star has constant density of $`1400`$kg m<sup>-3</sup> and that 90% of the star’s mass is ejected during the implosion ($`\eta =0.1`$). TBH creation at radii $`R_1`$ below about $`36\times 10^9`$m is prohibited because electron or neutron degeneracy would halt the collapse, as it would slightly beyond the left hand edge of the diagram at about $`R_2/R_110^6`$ and beyond, so that the wedge shape does not continue indefinitely. Above the shaded wedge, it is impossible for the toroidal star to be sufficiently massive if, as it must be, $`R_3<R_1`$ for the seed star and densities above $`1400`$kg m<sup>-3</sup> are disallowed. Lightly shaded regions above and below the main wedge show how the diagram would be altered if different values were taken for $`\eta `$.
This diagram identifies the region at which quasar-like behaviour is to be expected from TBHs (progressively shaded vertical section) where $`R_2`$ is almost as large as $`R_1`$ and narrow jets are formed. The line defined by $`R_2=R_1`$ is the quasar extinction boundary where the TBH becomes a spheroidal BH. Lines have been plotted to indicate contours of equal TBH mass and similarly for constant minor radius $`R_2`$. Because the mass of the TBH will not diminish with time, constant mass boundaries can only be traversed in one direction. There is a sufficiently broad birth zone spanning several orders of magnitude on each axis which enhances the probability that TBH creation is widespread at the centre of typical protogalaxies.
TBH birth is anticipated to occur at lower masses and lower $`R_2/R_1`$ ratios because the seed stars required are very large even for these. One typical case has been presented on the diagram. For this example a toroidal star of mass $`6\times 10^6\text{M}_{}`$ and radii $`R_1=1.8\times 10^{11}`$m and $`R_3=8.7\times 10^{10}`$m implodes after exhausting its fuel on a very short timescale to leave a TBH of mass $`6\times 10^5\text{M}_{}`$ and radii $`R_1=1.8\times 10^{11}`$m and $`R_2=2.8\times 10^6`$m. The accretion rate within this young galaxy is increasing so the TBH mass swiftly increases as does its angular momentum and angular velocity. The arrows of the evolutionary trace depict the evolutionary rate, fastest at the start then slowing down at higher masses such that the QSO phase can exist for a timescale several orders of magnitude greater than the formation time of the TBH. Somewhat inevitably, when $`R_1`$ and $`R_2`$ equalize, the quasar phase is discontinued, in this example when the mass reaches about $`10^9\text{M}_{}`$. For a given mass, the relationship between $`R_1`$ and $`R_2`$ will depend upon the angular velocity of the TBH which in turn is related to the angular momentum inflow due to accretion. In order to sustain a toroidal event horizon indefinitely, an ever increasing supply of angular momentum may be required. The accretion rate might be relatively low at the time of TBH birth, rising swiftly before peaking and slowly decreasing thereafter. Accordingly, the jet formation phase is predicted to terminate due to the topological transition at the boundary where $`R_2R_1`$.
## 8 DISCUSSION
It has been qualitatively described how rotating TBHs might evolve from protogalactic gas clouds and accrete matter from the galactic centre until their inner apertures contract and highly focused relativistic jets form. The viability of the model can be tested by observing the evolution of jet collimation with red-shift as this model predicts the degree of collimation continues to increase (though jet energetics may decline at later times) until the topological transition. Such an approach could further address the issue of whether the entire AGN population or some subset thereof is accounted for by a TBH model. The direct observations of AGN and quasars in our universe suggests that TBHs are more than abstract mathematical constructs. Classical general relativity still remains to be unified satisfactorily with quantum mechanics. Evidently, TBH stability is intertwined with this issue. As direct experiments cannot be performed in intensely curved spacetimes, astrophysical observations must be our guide. The proposed stability of TBHs allows the event horizon’s interior to be metaphorically unveiled, providing clues to the nature of quantum gravity and grand unification theories.
General relativity demands that the cosmological constant $`\mathrm{\Lambda }`$ be sufficiently negative in order to provide long-term TBH stability. Observational estimates of $`\mathrm{\Lambda }`$ based upon universal expansion reliant on the behaviour of general relativity in weak field environments suggest that its value is very small but probably positive. One possibility is that $`\mathrm{\Lambda }`$ is primarily a function of local spacetime curvature. Alternatively, the presence of external matter (accretion disk and galaxy) may provide the necessary curvature and non-stationarity permitting TBH stability over indefinite periods.
The TBH model, despite its controversial nature, presents a promising means of understanding the following characteristics of quasars: extreme jet energies, varying jet emergence angles, abrupt extinction, high gamma-ray radiation, the presence of heavy elements and the multiplicity of red-shifts in absorption spectra. None of these features are readily explained by spheroidal BH models. It is encouraging that the TBH model also appears to lead to plausible models for macroscopic processes within supernovae, microquasars and gamma-ray bursts. Fortunately, neutron tori – unlike TBHs – are exempt from topological censorship. Hence, these will certainly exist in astrophyscial circumstances, if only very briefly. Betraying their existence in distinctive and overtly energetic ways, these curiosities should be amenable to observational identification and study.
Numerical simulations are crucial if accurate comparisons with further detailed observations are to be made. Gravitational wave detectors and planned optical/X-ray interferometer technology will be sufficiently advanced in forthcoming decades to conclusively resolve the question of whether toroidal black holes truly exist.
## ACKNOWLEDGEMENTS
I wish to thank my father, Andrew Spivey for numerous enlightening discussions and continued assistance. Thanks also to Antony Hewish of Cambridge University for initial comments.
APPENDIX A
General relativity does not restrict the distribution of mass-energy external to event horizons. The question is not whether, but to what extent can toroidal arrangements of neutron degenerate matter be temporarily stabilised by rotation. The purpose of this appendix is to outline and prepare quantitative estimates of the circumstances leading to the creation of neutron tori using extensive simplifying assumptions due to the complexity of the situation. Although it seems that rather extreme conditions are necessary for tori to form, and dynamic evolution could be both violent and rapid, it will be seen that rotary core collapse supernovae are a natural setting for the birth of dense tori. Although the analysis presented is relevant to a broad range of stellar densities, including the formation of toroidal black hole cores, it must be stressed that tori of neutron density and below cannot be instantly dismissed by topological censorship - the primary objection to the formation of TBHs. It is known that ergoregions can form when relativistic stars form with high angular momentum despite the absence of either apparent or absolute event horizons. Potentially, this permits the proposed quasar mechanism to operate in a number of seemingly unrelated astrophysical phenomena.
Many stars, particularly the brighter Type Oe and Be, rotate considerably faster than the Sun. Equatorial velocities in the range $`300700`$km s<sup>-1</sup> are not uncommon as compared to $`2`$km s<sup>-1</sup> for the Sun. It is thought that the majority of stars have high initial angular velocities but that coupling between solar winds, magnetic fields and the inter-stellar medium cause a gradual decline in angular momentum. The brighter, more massive stars can be very short-lived and will retain a large angular momentum once their nuclear fuel is spent. It is therefore worthwhile studying the internal structure of rapidly rotating stars undergoing gravitational collapse to determine the conditions best suited for producing toroidal core configurations.
In order to preserve analyticity, a simplified model is presented. Later it will be apparent that removal of any simplifications necessitates a numerical treatment such as the one presented by Marcus, Press & Teukolsky. A uniformly rotating (constant angular velocity) ellipsoid would consist of ellipsoidal shells for which the assumption of uniform shell density is invalidated in situations of interest. Instead, uniformly rotating infinite cylinders are first considered. A cross section is illustrated in Fig. A.1(a) of the proposed structure of a rotating star composed of two immiscible, incompressible fluids with densities $`\rho _1`$ and $`\rho _2`$ where $`\rho _2>\rho _1`$. The contours denote lines of equal hydrostatic pressure increasing from zero at the surface of the spheroidal envelope to a peak within the higher density toroidal core.
If the pressure and density within a cylinder composed of two fluids can be shown to reach a maximum at a finite distance from the axis of rotation, Fig. A.1(b), and that such a distribution is in equilibrium, then the some degree of stability can be ascribed to the arrangement in Fig. A.1(a). A Newtonian analysis permits exact solutions by virtue of the superposition of cylindrical shells of differing densities, the linearity of (constant density) cylindrical gravity with radius, the null gravity within infinite cylindrical shells and the gravitational equivalence of cylindrical shells to axial line masses in the exterior regions. Cylindrical coordinates $`(r,\varphi ,z)`$ are employed to consider two immiscible and incompressible fluids of densities $`\rho _1`$ and $`\rho _2`$ with $`\rho _2>\rho _1`$ rotating smoothly and uniformly. Temperature is neglected because degenerate materials are of particular concern. The gravitational profile is not linear with radius because it is caused by three zones of different density: regions a), b) and c) with radii $`R_1`$, $`R_2`$ and $`R_3`$ respectively indicated in Fig. A.1(b). When equilibrium is achieved, the resultant force on each fluid element is zero. In this analysis, the individual forces acting on the elements are due to the pressure gradient, the centripetal acceleration and the gravitational attraction which, by virtue of the cylindrical symmetry, need only be considered in the radial direction. It is elementary to derive expressions for the derivatives of pressure, $`P_a,P_b`$ and $`P_c`$, with respect to radius for each of the three regions a), b) and c) respectively:
$`{\displaystyle \frac{dP_a}{dr}}`$ $`=r\rho _1(\omega ^22\pi G\rho _1)`$ (A.52)
$`{\displaystyle \frac{dP_b}{dr}}`$ $`=r\rho _2\left\{\omega ^22\pi G\left[\rho _2+\left({\displaystyle \frac{R_1^2}{r^2}}\right)(\rho _1\rho _2)\right]\right\}`$ (A.53)
$`{\displaystyle \frac{dP_c}{dr}}`$ $`=r\rho _1\left\{\omega ^22\pi G\left[\rho _2+\left({\displaystyle \frac{R_1^2R_2^2}{r^2}}\right)(\rho _1\rho _2)\right]\right\}`$ (A.54)
These expressions are readily integrated using the following boundary conditions: $`P_c(R_3)=0,P_b(R_2)=P_c(R_2)`$ and $`P_a(R_1)=P_b(R_1)`$. It is immediately apparent from (A.52) that the pressure increases with radius from $`r=0`$ providing a certain minimum angular velocity is exceeded: $`\omega >\omega _{\mathrm{min}}=\sqrt{2\pi G\rho _1}`$. For physically meaningful results, the pressure must not become negative at radii occupied by matter. There are two circumstances where this might first arise: at the centre (when the cylinder becomes hollow) and immediately beneath the surface (centripetal forces overcome gravitational forces leading to mass shedding). The latter condition is simply expressed as $`dP_c/dr>0`$ at $`rR_3`$ permitting the definition of a maximum angular velocity $`\omega _{\mathrm{max}}>\omega _{\mathrm{min}}`$ which is conveniently expressed as:
$`\omega _{\mathrm{max}}=\omega _{\mathrm{min}}\times \sqrt{1+\left({\displaystyle \frac{\rho _2\rho _1}{\rho _1}}\right)\left({\displaystyle \frac{R_2^2R_1^2}{R_3^2}}\right)}`$ (A.55)
In general, a wide range of angular velocities are available if the densities are very dissimilar whereas, as the densities of the two fluids approach one another, there is a much narrower range of values that $`\omega `$ can occupy above $`\omega _{\mathrm{min}}`$. A specific example is given in which the radii are in the ratio 1:2:3 for $`R_1:R_2:R_3`$ and the densities 1:2 for $`\rho _1`$ and $`\rho _2`$. Results are plotted in Fig. A.2.
The diagram presents the pressure variation along the radius of a rotating infinite cylinder. Several curves have been plotted which correspond to different rates of rotation. For the example given, internal pressure remains positive up to $`\omega 1.155\omega _{\mathrm{min}}`$, the mass shedding limit. The stability of these results is trivial because the assumption of equilibrium was inherent in the model, all solutions are in neutral equilibrium including those at low angular velocity and the non-rotating case. The significance of $`\omega _{\mathrm{min}}`$ is that stability cannot be achieved below this if the fluids become infinitesimally compressible because the density and pressure distributions would be qualitatively different.
When a homogeneous rotating cylinder of compressible fluid is considered, it transpires that stability of off-axis peak density arrangements is unattainable if the assumption of uniform rotation is retained. This is evident upon inspection of (A.52) for cases where the pressure and density are positively correlated i.e. $`dP/d\rho >0`$. This does not mean that an axial pressure-density peak will always result, once the central angular velocity exceeds a certain value $`\omega _c\sqrt{2\pi G\rho _c}`$ then axial density peaks are also unstable and differential rotation will occur (non-constant angular velocity). Furthermore, the angular velocity will generally decrease with radius for these systems, coinciding with the most physically realistic situations exemplified by the frame-dragging of the Kerr spacetime. Hence a non-axial density peak akin to toroidal solutions of rotating gravitating spheroids results. A numerical treatment must be employed for systems of even this complexity, before realistic interactions such as viscosity, radiation pressure, temperature variations and magnetic braking are incorporated in the model. Analytical limits of relevance to infinite cylinders exist in (i) the differentially rotating incompressible fluid approximation and (ii) the uniformly rotating compressible fluid regime whereby iterative solution of Volterra integral equations is in principle achievable. Numerical techniques become mandatory for differentially rotating compressible fluids in the infinite cylinder approximation.
It should be stressed that long term stability is not the issue here – all that is required is for a differentially rotating toroidal structure to transiently exist during the inherently dynamic and unstable collapse phase of stellar evolution - the necessary timescale for ‘stability’ is briefer than that required for dissipative processes to restore uniform rotation, e.g. magnetic braking and viscosity.
When a rapidly rotating star collapses, and the collapse originates at the core where gravity first defeats pressure, strong differential rotation accompanies a radial inrush of material - this environment facilitates the formation of a toroidal core. Because angular momentum is conserved for all collapsing shells, the angular velocity in the core increases appreciably to a level in excess of $`\omega _{\mathrm{min}}`$ and the angular velocity beyond the core declines radially. If the core is sufficiently dense, relativistic frame dragging contributes to the differential rotation. Radially decreasing rotation coupled to the fact that the pressure (and therefore the density) must increase with radius at the centre means that toroidal cores are practically inevitable during the collapse of rapidly rotating stars. Axisymmetric tori often exhibit non-axisymmetric instabilities in numerical simulations, the resultant gravitational radiation being of fascination to gravitational wave astronomy. The perturbational influence of orbiting companions could also disturb the symmetry. A simple equation of state is insufficient to model superfluidity and superconductivity, properties that neutron stars are widely expected to possess, rendering inapplicable many models constructed to investigate gravitational wave driven instabilities. The self-gravity of a torus can sustain a state of pseudo-equilibrium. If the torus is electrically charged, the exterior magnetic field acts as a barrier both to surface winds consisting of charged particles and to neutralising inflows – reinforcing stability. The outcome for situations where both $`\omega _{\mathrm{min}}`$ and $`\omega _{\mathrm{max}}`$ is exceeded is qualitatively unchanged. Mass expelled from the outermost periphery of the torus is not ejected to infinity but forms an equatorial disk orbiting the torus. If sufficient mass is shed, this disk can become geometrically thick, i.e. the torus expands.
For slowly rotating ellipsoids, the value of $`\omega _{\mathrm{min}}`$ increases to $`\sqrt{4\pi G\rho _c}`$. If the neutron core following gravitational collapse of the Sun has a density in excess of $`3\times 10^{18}`$kg m<sup>-3</sup> then it is conceivable that the core could become toroidal. If a more attainable collapse density of $`10^{14}`$kg m<sup>-3</sup> were specified, then the Sun would only need to rotate at a moderately higher angular velocity of $`5.5\omega _{\mathrm{sun}}`$ for $`\omega _{\mathrm{min}}`$ to be attained during core collapse. The Sun’s internal pressure would first vanish within the surface (mass shed into keplerian orbit) were it to rotate at a rate $`212\omega _{\mathrm{sun}}`$, so the available rotation range for toroidal core collapse is generally broad and attainable for typical massive stars.
A TBH embedded in a collapsing stellar envelope resembles a scaled-down version of the described quasar environment. In contrast with the sustained activity of active galaxies, the tremendously accelerated accretion onto the core causes a violent and intense anisotropic explosion. Such a TBH could be of relatively low mass, perhaps below $`0.1\text{M}_{}`$ as efficient jet generation processes begin to operate and stall its growth. There is therefore a definite possibility that a large fraction of the remaining star is propelled along the jets. Together with the first active galaxies, there are obvious implications for cosmological reionization which appears to have occurred at red-shifts of $`z6`$. A neutron torus formed during rotary core collapse can ‘recover’ following the ejection of substantial mass from the stellar envelope to a lower density object e.g. a white dwarf. On the other hand, charged neutron tori that do not recover will subsequently collapse to charged spheroidal neutron stars forming highly magnetised pulsars or Kerr-Newman BHs. Objects classified as ‘magnetars’ or anomalous X-ray pulsars (AXPs) have been observed. The inferred magnetic field strengths due to the spin-down rate of these objects is $`10^{14}`$G, providing a useful clue as to the degree of electrical charge of the neutron tori during SNe and thus, by inference, the net TBH charge in AGN circumstances. This level of charge can have a significant bearing on the spacetime geometry – this is discussed in appendix B.
White dwarves containing toroidal neutron cores and isolated rotating neutron tori will often form during the core collapse of moderately rotating progenitors. The rotation rate should be sufficient for these dense tori to generate ergoregions thence accumulate charge sustaining magnetospheres which bolster the negative energy states of the ergoregion. These neutron tori are unlikely to be long-lived as they are susceptible to various instabilities and their differential rotation will eventually be erased by dissipative processes, but they are certainly of interest in more dynamic environments. The significance of these short-lived neutron tori or equivalently stellar mass toroidal black holes (SMTBH) are now addressed in three separate astrophysical settings.
Firstly, a binary system consisting of a SMTBH and a stellar companion could operate as follows: an accretion disk forms around the SMTBH composed of material transported from the nearby star by gravitational and electromagnetic interactions. The central aperture of the SMTBH contains an ergoregion and an intense dipolar magnetic field due to a net electrical charge. This then gives rise to anti-parallel jets aligned with the axis of rotation in a very similar manner to the quasar albeit on a smaller scale. Microquasars have been observed within the confines of the Milky Way and are so called because they seem to obey simple scaling laws applied to quasars. Accretion of material from the companion star and jet formation will combine to decrease the overall angular momentum of the SMTBH on a shorter timescale than that of the quasars, the unusual behaviour terminating when the topology transitions to spheroidal after the angular momentum has been partially jettisoned. Unlike pulsars, the magnetic fields generated by charged tori will be robustly aligned with the rotation axis and, with the possible exception of radiation emanating from an accretion disk, periodic bursts of radiation will not be observed.
Relativistic galactic jet sources and their similarities with quasar outflows have been reviewed by Mirabel & Rodriguez (1999). From the limited microquasar observations available, it appears that the jet velocities have a bimodal distribution classified by $`\nu _{jet}0.3c`$ and $`\nu _{jet}0.9c`$. Whether a corresponding distribution exists for the jets of active galaxies is currently unknown. An interesting feature of some microquasars which is absent in quasars is their behaviour as the accretion disk is exhausted resulting in a sudden ejection of condensations (Mirabel et al 1998). Existing steady state MHD models with continuous jets have difficulty accounting for this, relying on a disk-supported magnetic field. This problem is resolved in the present model because the magnetic field of the torus remains when the disk disintegrates and confines the remaining plasma to circulate above the event horizon of the SMTBH, weaving repeatedly through the central aperture until it emerges in the form of jets aligned with the spin axis.
Secondly, the mechanism could participate in the most energetic SNe - those which have been dubbed ‘hypernovae’ with energies two orders of magnitude above ‘ordinary’ SNe and thought to coincide with longer duration gamma-ray bursts. GRB980425 has been associated with SN1998bw providing evidence for a common mechanism (Cen, 1999). For example, consider a massive, rotating and fuel starved star undergoing rotary core collapse. A neutron torus (or SMTBH) develops in the core embedded within a lower density envelope. As before, charge accumulates on the torus which cannot be quickly neutralised on a timescale comparable to that of the implosion. A strong magnetic field threads the central aperture of the torus and strongly negative energy states are available in the ergoregion. The mechanism results in a ferocious outward explosion of matter from the centre of the SN in which a significant proportion of the star’s mass is expelled anisotropically. Jets from SNe have been inferred from nearby hot-spots detected by optical speckle interferometry (Cen 1999, and references therein). Evidence of highly anisotropic ejecta is provided by polarimetric SN observations (Wang et al, 1999). Collapsar models attempting to account for jet formation in core collapse SNe are hampered by the spherical core topology as jets, if they form at all, are unlikely to penetrate through imploding shells for the following reasons:
* The inefficiency of the Blandford-Znajek mechanism is now recognised.
* The topology dictates that plasma flowing along magnetic flux lines efficiently neutralises the core — but in the case of toroidal cores, neutralisation paths are orthogonal to flux lines, and neutralisation is thereby inhibited.
* The equatorial plane, which features additional centrifugal forces, is no more unlikely to feature outflows than polar regions if the star is assumed to remain electrically neutral.
Thirdly, pseudo-periodic gamma ray bursts (GRBs) could be generated by a mechanism similar to the microquasar. Consider a rapidly rotating white dwarf with a companion star providing a steady supply of material to an orbiting accretion disk. Metastable oscillations could be established whereby the core of the white dwarf collapses to a neutron torus once sufficient matter and angular momentum has accumulated. This results in a brief period of jet activity in which enough mass and angular momentum is expelled to restore the star to a pure spheroidal white dwarf. Accretion of matter from the binary companion then repopulates the accretion disk, with the mass and angular momentum of the white dwarf slowly increasing until the cycle repeats. It may be that some of the shorter-duration gamma ray bursts can be attributed to situations like this.
APPENDIX B
A year has elapsed since this paper was deposited on the archive. During that time, explanations for the stability of toroidal black holes have not been ventured. An open discussion along these lines could have been included in the original submission but was not, partly to avoid polarising the views of a sympathetic audience and partly because of the bewildering array of possibilities. Though no claim is made of a satisfactory resolution, this final version devotes an appendix offering a ‘snapshot’ of my thoughts on this issue which have benefited from twelve months of distilled cogitation. Parallels between the Kerr BH and Newtonian analogues comprising a self-gravitating annulus are explored. A simple quantized Newtonian model is used to investigate possible consequences in relativistic settings. The intention is not to denigrate the existing research on black hole spacetimes which has been the focus of much solemn effort by dedicated practitioners of general relativity, rather to communicate some pertinent concerns in a straightforward and hopefully thought-provoking manner.
First it is shown that the ring singularity of the Kerr geometry travels with the velocity of light, regardless of the degree of rotation. Then it is demonstrated that a Newtonian equivalent of the Kerr geometry is a homogeneous self-gravitating ring of infinite density rotating at infinite velocity, implying an unphysically large angular momentum and kinetic energy. This is interpreted as an inevitable consequence of the simplistic model which entirely disregards the microscopic quantum nature of the ring. When a more realistic analysis is pursued in the Newtonian setting, it is observed that the velocity required for equilibrium is logarithmically relaxed - a macroscopically observable consequence of the quantum world. By implication it is then argued that a natural relativistic counterpart deviates from the Kerr solution, and that this deviation would be particularly evident at high angular momenta. Astrophysical environments of interest are then addressed to illustrate the possibility that event horizon topologies may not be restricted to 2-spheres.
In pseudo-Cartesian coordinates $`(\overline{t},x,y,z)`$, the Kerr metric reads:
$$\begin{array}{c}ds^2=d\overline{t}^2+dx^2+dy^2+dz^2+\frac{2mr^3}{r^4+a^2z^2}\hfill \\ \hfill \times \left[d\overline{t}+\frac{(rx+ay)dx+(ryax)dy}{a^2+r^2}+\frac{z}{r}dz\right]^2\end{array}$$
(B.56)
and in Boyer-Lindquist coordinates $`(t,r,\theta ,\varphi )`$:
$$\begin{array}{c}ds^2=\frac{\mathrm{\Delta }}{\rho ^2}(a\mathrm{sin}^2\theta d\varphi dt)^2\hfill \\ \hfill +\frac{\mathrm{sin}^2\theta }{\rho ^2}[(r^2+a^2)d\varphi adt]^2+\frac{\rho ^2}{\mathrm{\Delta }}dr^2+\rho ^2d\theta ^2\end{array}$$
(B.57)
$`\mathrm{\Delta }`$ $`=r^22mr+a^2`$ (B.58)
$`\rho ^2`$ $`=r^2+a^2\mathrm{cos}^2\theta `$ (B.59)
The spatial coordinates of the two metrics obey the transformations:
$`x`$ $`=r\mathrm{sin}\theta \mathrm{cos}\varphi +a\mathrm{sin}\theta \mathrm{sin}\varphi `$ (B.60)
$`y`$ $`=r\mathrm{sin}\theta \mathrm{sin}\varphi a\mathrm{sin}\theta \mathrm{cos}\varphi `$ (B.61)
$`z`$ $`=r\mathrm{cos}\theta `$ (B.62)
A true spherical polar coordinate $`R`$, coinciding asymptotically with $`r`$ at large radii, can be defined as $`R^2=x^2+y^2+z^2`$ which transforms to $`R^2=r^2+a^2\mathrm{sin}^2\theta `$. The ring singularity of the Kerr BH resides at $`(r=0,\theta =\pi /2)`$ or $`(R=a,\theta =\pi /2)`$. Because the Kerr solution is stationary, the circle on which the singularity lies is a geodesic. To investigate its velocity, it is assumed that $`r`$ and $`\mathrm{cos}\theta `$ are small, $`r^2`$ is negligible and $`dr=d\theta =0`$. The Boyer-Lindquist metric then becomes:
$`ds^2=2mr{\displaystyle \frac{(a\mathrm{sin}^2\theta d\varphi dt)^2}{a^2\mathrm{cos}^2\theta }}`$ (B.63)
As $`\mathrm{cos}\theta `$ becomes infinitesimally small, to preserve $`ds^21`$, the numerator must converge as rapidly as $`\mathrm{cos}^2\theta `$ to zero, so the term within parentheses vanishes yielding
$`v_{\mathrm{ring}}=R{\displaystyle \frac{d\varphi }{dt}}={\displaystyle \frac{\sqrt{r^2+a^2\mathrm{sin}^2\theta }}{a\mathrm{sin}^2\theta }}=1`$ (B.64)
showing that the singularity travels with coordinate velocity $`c`$ when $`a^2>0`$. This is to be compared with the coordinate velocity of particles and photons remaining on the equator of the Kerr BH event horizon but being dragged around with the horizon. To investigate this, after setting $`dr=d\theta =0`$ and $`\theta =\pi /2`$ the metric reduces to:
$`rds^2=(r2m)dt^2+4amdtd\varphi (r^3+ra^2+2ma^2)d\varphi ^2`$ (B.65)
To remain on the horizon, $`ds^2`$ is necessarily zero and the radius at the event horizon is given by $`r_+=m+\sqrt{m^2a^2}`$ so that $`R_+^{\mathrm{\hspace{0.17em}2}}=2m(m+\sqrt{m^2a^2})=2mr_+`$. Solving for $`d\varphi /dt`$ allows the coordinate velocity of the equatorial event horizon to be determined:
$`v_{\mathrm{eeh}}=R_+{\displaystyle \frac{d\varphi }{dt}}|_{r_+}=\left[{\displaystyle \frac{a^2}{2m(m+\sqrt{m^2a^2})}}\right]^{1/2}`$ (B.66)
which varies from $`v_{\mathrm{eeh}}=0.5a/m`$ as $`a0`$ to $`v_{\mathrm{eeh}}0.7a/m`$ as $`am`$. The Kerr-Newman metric endowed with charge $`Q`$ features an outer event horizon at $`r_+=M+\sqrt{m^2Q^2a^2}`$ and, after restoring constants, remains sub-extremal if
$`G^2m^2GQ^2+c^2a^2`$ (B.67)
so charge and rotation act repulsively and in unison to drive the BH away from spherical symmetry. This conclusion seems inevitable without abandoning the equivalence principle (Petkov, 2001), which in any case invalidates the metric. Since extremality cannot be surpassed, general relativity restricts static particles to having $`Q/M\zeta _{\mathrm{max}}=\sqrt{G}`$. Gross violations are witnessed in particle physics, for protons $`Q/M>10^{13}\zeta _{\mathrm{max}}`$ and for electrons $`Q/M>10^{16}\zeta _{\mathrm{max}}`$. Besides quantum scales, this means that a tiny electrical charge can profoundly affect large-scale spacetime geometries. For instance, if an imbalance of one electron/proton exists for every $`10^{13}`$ neutrons, the time dilation at the surface of a neutron star is almost entirely eliminated. Electromagnetism can thus play an important role in the stability of charged tori.
Consider a planar constellation of $`N`$ identical satellites of total mass $`M`$ evenly and symmetrically distributed on a circle (Fig. B.1). According to Newtonian mechanics, equilibrium is given by the balancing of gravitational and centripetal forces acting on each satellite such that the radius, $`r`$, of the circular orbits remains constant. All satellites rotate at constant angular velocity $`\omega `$ so that their individual velocities are $`v=\omega r`$. Equilibrium is attained when:
$`{\displaystyle \frac{GM}{4Nr^2}}{\displaystyle \underset{n=1}{\overset{N1}{}}}{\displaystyle \frac{1}{\mathrm{sin}(\frac{n\pi }{N})}}={\displaystyle \frac{v^2}{r}}`$ (B.68)
In the limit $`N\mathrm{}`$, the series can be expressed in the form of a definite integral:
$`\underset{N\mathrm{}}{lim}{\displaystyle \frac{1}{N}}{\displaystyle \underset{n=1}{\overset{N1}{}}}{\displaystyle \frac{1}{\mathrm{sin}(\frac{n\pi }{N})}}={\displaystyle _0^\pi }{\displaystyle \frac{dx}{\mathrm{sin}x}}={\displaystyle \frac{4rv^2}{GM}}`$ (B.69)
Although the series converges for $`N<\mathrm{}`$, the integral is divergent since
$`{\displaystyle _0^\pi }{\displaystyle \frac{dx}{\mathrm{sin}x}}=\left[\mathrm{log}\left|\mathrm{tan}{\displaystyle \frac{x}{2}}\right|\right]_0^\pi =2\times \underset{x+0}{lim}\mathrm{log}\left({\displaystyle \frac{2}{x}}\right)`$ (B.70)
Hence, a ring of finite total mass $`M`$ composed of an infinitude of individual satellites is obliged to rotate at infinite velocity in order to maintain radial equilibrium. This in turn implies an infinite angular momentum and kinetic energy for homogeneous rings. Planetary ring systems, such as Saturn’s, are nevertheless observed. Despite their self-gravity, the rotational velocity remains modest – evidence that their microscopic structure is particulate, not homogeneous. Where self-gravitating rings of matter are concerned, homogeneity is an unreasonable assumption, but one which is often adopted in general relativity as exemplified by the Kerr metric.
Some of the following statements are necessarily speculative, but it should be kept in mind that the overly complacent alternative is to trust in mathematical models when conditions are far more extreme than those accessible to experimental tests. Acceleration of the Kerr singularity to the speed of light formally requires an infinite amount of energy. This energy must have been supplied by the gravitational potential energy of the matter which the black hole consumes. It is implausible that this energy source is truly unbounded, allowing the spacetime to attain infinite curvature. Rather, the collapse must halt at some limiting density, at which time the kinetic energy remains finite. Therefore, though the Newtonian analysis presented is very simplistic, aspects of this simplicity are also shared by its relativistic counterpart, the Kerr black hole.
In an attempt to circumvent these limitations and prepare rough estimates, suppose that the ring structure is subdivided into a non-infinite number of satellites. For instance, the maximum number of neutrons contained in a mass of $`10^9\text{M}_{}`$ is $`ϵ^1=1.2\times 10^{66}`$. In this case, the equilibrium satellite velocity remains finite with magnitude
$`v\sqrt{{\displaystyle \frac{GM}{4r}}{\displaystyle _ϵ^{\pi ϵ}}{\displaystyle \frac{dx}{\mathrm{sin}x}}}8.7\sqrt{{\displaystyle \frac{GM}{r}}}`$ (B.71)
But in these situations, neutrons are plausibly replaced by Planck-scale particles. The characteristic Planck length is $`l_\mathrm{p}=\sqrt{\mathrm{}G/c^3}1.6\times 10^{35}`$m and the Planck mass is $`m_\mathrm{p}=\sqrt{\mathrm{}c/G}2.2\times 10^8`$kg. A Kerr BH of mass $`M`$ contains a singularity of maximal radius when $`a=M`$ or $`R=MG/c^2`$ in natural units. Suppose that the singularity is a crystalline structure of Planck ‘particles’ in a circular arrangement whose further collapse is resisted by quantum mechanical repulsion. One might object that this arrangement is too idealised or that it requires an infinite universal time to elapse - perhaps so, but it is here argued that some form of uncertainty principle or holographic correspondence prevents external observers from distinguishing between the present assumption and any other model. The mean separation between adjacent Planck particles is determined to be independent of the total mass and given by $`d_{\mathrm{sep}}=2\pi Rm_\mathrm{p}/M=2\pi \sqrt{\mathrm{}G/c^3}=2\pi l_\mathrm{p}`$, a coincidence substantiating the original premise of a Planckian singularity. For a $`10^9\text{M}_{}`$ BH, the total number of particles is $`ϵ^1=9\times 10^{46}`$ which yields $`v7.4\sqrt{GM/r}=7.4`$c. The equivalent relativistic velocity is $`\mathrm{\Gamma }=8.4`$ or $`v_{\mathrm{rel}}0.993`$c (for a stellar mass BH, $`v_{\mathrm{rel}}0.992`$c). It is known that the ratio $`a/m`$ for a Kerr BH can realistically approach $`0.998`$ in astrophysical circumstances (Thorne, 1974), allowing scope for deviation from the Kerr geometry due to Planck scale phenomena.
Though the margin for this to occur seems slender for isolated horizons, indicative of almost negligible deviation, the assumption that the singularity is quantum-mechanically sustained at the Planck scale introduces additional considerations which, in typical settings, are favourable for significant deformation of the Kerr geometry. The spacetime curvature in the vicinity of the singularity is tamed by the repulsion between Planck particles (e.g. Louko & Matschull, (2001) where some success has been claimed regarding the quantization of geometry). This repulsion presumably grows as the ring’s radius and $`a/m`$ decreases. Whereas the Kerr singularity is infinitely distant from the black hole’s exterior due to the immensely strong curvature (Thorne, Price & MacDonald, 1986), the Planckian singularity is susceptible to the influence of other matter through tidal gravity. Indeed, the surface gravity of a Kerr-Newman BH diminishes with increasing charge and rotation, disappearing altogether at extremality. So without infinite curvature, the singularity would otherwise be deformable.
Most modern attempts at unification invoke extra spatial dimensions. At high energies, gravity is thought to ‘leak’ away from our 3 dimensions. This has yet to be confirmed experimentally, due to practical difficulties. Measurements have verified the inverse square law down to submillimetre scales, equating to an energy scale $`10^2`$eV : well short of particle accelerator energies $`100`$GeV, the sypersymmetry scale $`10^{16}`$GeV and the Planck scale $`m_\mathrm{p}c^210^{19}`$GeV. Diminution of gravitational interactions at small scales would inevitably cause the orbital velocities of BH singularities to decrease further, quite feasibly by a substantial amount. General relativity will certainly break down at the highest energies, otherwise collapse cannot even be halted at the Planck scale. Since energy conditions are known to be violated by the Casimir effect and Hawking radiation, it is unlikely that they will be satisfied everywhere within a black hole. Conversely, the extreme pathology of closed timelike curves exhibited by the stationary BH metrics is tacitly embraced by investigators.
Consider an accretion disk orbiting a rapidly rotating BH. The disk gravity induces tidal stresses on the Planckian singularity which tends to stretch it radially. In addition to the accretion disk, AGN reside within a molecular torus and host galaxy, each equatorially oriented with respect to the BH. In the context of rotary core collapse SNe, the outer shells collapse towards the plane of rotation where a substantial proportion of the debris forms a thick disk. Its mass and proximity to the core impose tidal forces whose influence on the central BH will be more pronounced than in AGN situations. The gravitational potential at radius $`r`$ in the plane of a circular hoop of radius $`Rr0`$ whose linear mass density is $`\lambda `$ is given by the following function containing a complete elliptic integral of the first kind K(k):
$`\mathrm{\Phi }_{\mathrm{hoop}}(r)={\displaystyle \frac{R}{r}}\mathrm{\Phi }_{\mathrm{hoop}}\left({\displaystyle \frac{R^2}{r}}\right)={\displaystyle \frac{4G\lambda R}{R+r}}K\left({\displaystyle \frac{2\sqrt{Rr}}{R+r}}\right)`$ (B.72)
$`\mathrm{\Phi }_{\mathrm{hoop}}`$ decreases monotonically from the centre ($`r=0`$) to negative infinity as $`rR`$. In elementary functions, it can be proven (by using series expansions for the elliptic integral and differentiating) that the internal ($`g_{\mathrm{int}}`$) and external ($`g_{\mathrm{ext}}`$) gravitational accelerations are bounded as follows:
$`\pi {\displaystyle \frac{(R^{\mathrm{\hspace{0.17em}2}}r^2)g_{\mathrm{int}}}{rG\lambda }}4{\displaystyle \frac{(r^2R^{\mathrm{\hspace{0.17em}2}})g_{\mathrm{ext}}}{RG\lambda }}2\pi `$ (B.73)
The lower bound corresponds to $`rR`$, the central limit to $`rR`$ and the upper bound is the asymptotic behaviour as $`r\mathrm{}`$. These functions are readily integrated in many situations and permit the simple construction of models in which one may be interested in calculating conservative estimates. For increased accuracy, interpolations are available whose maximum errors are $`0.2\%`$ at $`r/R0.57`$ for the internal gravity and $`0.9\%`$ at $`r/R1.29`$ for the external gravity, which is oppositely directed:
$`g_{\mathrm{int}}(r)`$ $`G\lambda r\left[{\displaystyle \frac{4}{R^{\mathrm{\hspace{0.17em}2}}r^2}}{\displaystyle \frac{4\pi }{R(R^{\mathrm{\hspace{0.17em}2}}r^2)^{1/2}}}\right]`$ (B.74)
$`g_{\mathrm{ext}}(r)`$ $`G\lambda R\left[{\displaystyle \frac{4}{R^{\mathrm{\hspace{0.17em}2}}r^2}}{\displaystyle \frac{2\pi 4}{r^{3/2}(r^2R^{\mathrm{\hspace{0.17em}2}})^{1/4}}}\right]`$ (B.75)
Of immediate interest are truncated disks of constant areal mass densities $`\sigma `$. Integration leads to the following approximation for the gravity within a thin truncated disk ($`0rR_\alpha <R_\beta `$) where $`R_\alpha `$ and $`R_\beta `$ are the disk’s radii at the inner and outer rims respectively:
$$\begin{array}{c}g2G\sigma \mathrm{log}\left[\frac{(R_\alpha +r)(R_\beta r)}{(R_\alpha r)(R_\beta +r)}\right]\hfill \\ \hfill (4\pi )G\sigma \left[\mathrm{cos}^1\left(\frac{r}{R_\beta }\right)\mathrm{cos}^1\left(\frac{r}{R_\alpha }\right)\right]\end{array}$$
(B.76)
It is often assumed that BH accretion disks are truncated some distance from the event horizon because stable circular orbits are forbidden within $`r_{\mathrm{ms}}`$, the innermost radius of marginal stability. For Schwarzschild BHs, $`r_{\mathrm{ms}}=6M`$, for retrograde extremal Kerr orbits, $`r_{\mathrm{ms}}=9M`$, and for prograde extremal Kerr orbits, $`r_{\mathrm{ms}}M`$. Instability implies that material penetrating within these boundaries is accelerated towards the event horizon, so that these regions can scarcely be totally vacated. The acceleration experienced by infalling material, according to Newton’s third law, is counterbalanced by a smaller but oppositely directed acceleration of the Planckian singularity towards the surrounding disk. When the singularity is teased towards larger radii, the BH is likely to accumulate further angular momentum. Hence, continued accretion and tidal distortion can reinforce the distortion, making it conceivable that nonstationary accretion into the event horizon, coupled with continuous disk replenishment, could maintain a TBH against collapse for prolonged periods. According to this view, a TBH might evolve from a spheroidal BH and vice versa — topological transitions of both kinds being possible. This enhances the probability of TBH formation, circumventing the need for collapse of a toroidal dust cloud or star as outlined in section 2. It also raises the question of whether the nuclei of some galaxies are revived into activity following a dormant stage e.g. by tidal interaction with nearby galaxies or more directly, following mergers. Elliptical galaxies, believed to be the outcomes of mergers, host an overabundance of AGN.
In conclusion, though Kerr’s metric for rotating black holes is rightly celebrated as one of the landmark discoveries of 20th century science, the subsequent failure to satisfactorily unite the classical theory of general relativity with quantum mechanics remains a severe impediment to an understanding of nature. Until this fundamental obstacle can be overcome, one must be fully aware of the inherent drawbacks of myopic alternatives. Black hole topology is a subject unsuitable for purely classical computations, however sophisticated the mathematical veneer. Extreme caution should be employed if results are to be usefully construed. The rule of thumb calculations presented here are suggestive that an ultimate theory might accommodate the possibility of toroidal black holes, temporarily stabilised on astrophysically relevant timescales by the action of accretion, tidal deformation and Planck scale repulsion. The basis of the scenario advocated for supporting TBH stability is summarised below:
* Black holes will often be formed with nearly maximal rotation e.g. from collapse of moderately rotating stellar progenitors. Accretion via a thin disk can also lead to a rapid build-up angular momentum. Astrophysically realistic ratios of $`a/m`$ can approach 0.998. Near-extremal black holes are therefore predicted to exist in rotary core-collapse supernovae, galactic nuclei and accreting black hole binary systems. In all situations, tidal stresses are exerted by matter exterior to the event horizon which orbits in the equatorial plane. Disk truncation decreases as extremality is approached, the radius of the innermost stable orbit and the singularity converge towards that of the event horizon. Mass steadily migrates into the event horizon by accretion, bridging the zone of disk truncation and bolstering the tidal forces acting on the singularity.
* The Kerr geometry is in some respects unphysical — it contains a homogeneous ring singularity rotating at the speed of light, infinite spacetime curvature at the location of the singularity, closed timelike paths within its Cauchy horizon and makes no provision whatsoever for quantum mechanics. Furthermore, all attempts to match the exterior metric with realistic non-vacuum internal distributions of matter have thus far failed.
* If collapse halts at the Planck-scale, the velocity of rotation could decrease to $`0.993`$c, suppressing the spacetime curvature in the vicinity of the singularity making it susceptible to tidal deformation by surrounding structures e.g. thin disks, thick disks and tori. High energy effects such as gravity ‘leaking’ into higher dimensions may additionally modify the rotational velocity of the singularity. Radial elongation of the Planck singularity is conducive to the accumulation of angular momentum beyond that of undistorted configurations, acting as a barrier against collapse should accretion be interrupted. Sustained accretion and tidal distortion might ultimately result in toroidal event horizon topology.
|
warning/0004/cond-mat0004458.html
|
ar5iv
|
text
|
# Generalized Numerical Renormalization Group for Dynamical Quantities
## Abstract
In this paper we introduce a new approach for calculating dynamical properties within the numerical renormalization group. It is demonstrated that the method previously used fails for the Anderson impurity in a magnetic field due to the absence of energy scale separation. The problem is solved by evaluating the Green function with respect to the reduced density matrix of the *full* system, leading to accurate spectra in agreement with the static magnetization. The new procedure (denoted as DM-NRG) provides a unifying framework for calculating dynamics at any temperature and represents the correct extension of Wilson’s original thermodynamic calculation.
Quantum impurity models and their low-temperature properties are of central importance in condensed matter physics. They show characteristic many-body effects like the screening of a local moment by conduction electrons (the Kondo effect) which was first observed in measurements on dilute magnetic impurities in metals (see ). More recently, artificial nanostructures (quantum dots or surface atoms probed by STM ) with tunable parameters provided new representations of the Anderson or Kondo model . In theory, a very fruitful line of research was opened by the development of dynamical mean-field theory (DMFT) where correlated lattice systems are mapped onto effective impurity models which are then accessible in a controlled way .
In all the above areas, progress depends sensitively on the existence of a reliable calculational method that can provide static and dynamic (spectral) properties in the full energy range. Wilson’s numerical renormalization group gave the first quantitative description of the Kondo effect. In systems with very different energy scales (small Kondo temperature $`T_K`$, large bandwidth) it is the only technique that can do so. In the original calculation Wilson focused on obtaining thermodynamic expectation values like the impurity susceptibility by iterative diagonalization. Each iteration step was shown to correspond to a certain temperature where expectation values could be obtained with great precision. Later, the method was extended to zero temperature dynamical properties by several groups and applied to a variety of problems , including recent DMFT calculations (e.g. ). In these calculations the additional assumption had to be made that transitions from the *ground state* to higher excitations are already correctly described in the first few iterations. Accurate results in agreement with sum rules were obtained for the single particle spectrum in the absence of external fields. In the following, however, we demonstrate that this procedure is not rigorous and fails for the Anderson impurity model in a magnetic field. To remedy the defect, we introduce a new approach based on the concept of the *reduced density matrix*. This procedure (which in the following will be denoted as DM-NRG) makes use of the full information contained in iterative diagonalization and can therefore be considered as the true extension of Wilson’s original work to dynamical quantities.
To be specific, we consider the spin $`1/2`$ Anderson model $`H=H_0+H_{\mathrm{imp}}`$ where the impurity part is given by
$`H_{\mathrm{imp}}=V(f_\sigma ^{}c_{0\sigma }^{}+h.c.)+Un_fn_fϵ_fn_f`$ $`hS_f^z.`$ (1)
Here we have introduced a local magnetic field $`h`$ coupled to the impurity spin $`S_f^z`$, an on-site Coulomb repulsion $`U`$, and a hybridization $`\mathrm{\Delta }=\pi V^2/2`$ to the conduction band orbital $`c_{0\sigma }`$. Units are chosen as $`\mathrm{}=k_B=g=\mu _B=1`$. Depending on the energy of the impurity level, $`ϵ_f`$, different physical behaviour is realized. In the following, we focus on the symmetric ($`ϵ_f=U/2`$) and mixed valence ($`|ϵ_f|\mathrm{\Delta }`$) regimes. The conduction band (extending in the range $`[1,1]`$) is already written in the *linear chain* representation characteristic for NRG
$$H_0=\underset{n=0}{\overset{\mathrm{}}{}}\xi _n(c_{n\sigma }^{}c_{n+1\sigma }^{}+h.c.)$$
(2)
where the hopping matrix elements decay exponentially $`\xi _n\mathrm{\Lambda }^{n/2}`$ due to a logarithmic discretization of the conduction band. This model – while still a nontrivial many-body problem – can now be solved by iterative diagonalization, keeping in each step only the lowest, most relevant levels. The number of iterations then corresponds to the temperature one is interested in according to $`T_N=c\mathrm{\Lambda }^{(N1)/2}`$, where $`c`$ is a constant of order one. For calculating static quantities, all necessary information is thus obtained because only excitations on the scale $`T_N`$ are relevant. For dynamical properties, however, an additional energy scale is provided by the frequency $`\omega `$ which may be much larger than the temperature. Let us focus on the spin resolved single particle spectral density
$$A_\sigma (\omega )=\underset{nm}{}|m|f_\sigma ^{}|n|^2\delta \left(\omega E_m+E_n\right)\frac{e^{\beta E_m}+e^{\beta E_n}}{Z}$$
(3)
in the Lehmann representation where the $`|n`$ are the many-particle eigenstates of $`H`$ and $`Z`$ is the partition function. Obviously, spectral information at frequencies $`\omega T_N`$ requires matrix elements between low-lying states and excitations which in iteration $`N`$ are not available anymore (they have already been lost by truncation). To circumvent this difficulty, the following “ad hoc” procedure has been used so far: In calculating $`A(\omega )`$, expression (3) was simply evaluated in iteration step $`N^{}N`$ where $`T_N^{}\omega `$, assuming that for an analysis of this spectral regime the low energy levels were described well enough. There is no rigorous argument to justify this assumption, as, for example, the crossover to the strong coupling fixed point and the corresponding change in the ground state may occur at a much lower temperature scale $`T_KT_N^{}`$.
In fig. 1 we present results for the symmetric model (1) at $`T=0`$ which have been calculated in this way. Without external field, one obtains the well-known three-peak structure characteristic for a small Kondo temperature $`T_K`$. Switching on a small magnetic field $`h=𝒪(T_K)`$ only affects the quasiparticle peak, while the high energy spectrum is almost unchanged. This result is easily understood: In the iterations where the atomic levels are calculated, the NRG procedure does not yet “know” about the tiny perturbation that eventually breaks the spin symmetry of the ground state. One can, however, easily verify that this result is incorrect: Calculating the ground-state magnetization $`m`$ (a static quantity) directly as a thermodynamic expectation value $`(n_fn_f)`$ and comparing with the value derived from the spectrum
$$m=_{\mathrm{}}^0𝑑\omega A_{}(\omega )_{\mathrm{}}^0𝑑\omega A_{}(\omega )$$
(4)
the results do not agree (see table in fig. 3). Physically, the strong polarization of the impurity due to a small magnetic perturbation should suppress the upper atomic level because no particle excitations are possible anymore. This suppression is drastically underestimated by the method used so far (indeed, in the limit of vanishing Kondo temperature $`T_K`$ it will not be seen at all).
In order to capture this effect it is clearly necessary to obtain the correct ground state *before* calculating the spectra. This is achieved by the following two-stage procedure:
1) NRG iterations are performed down to the temperature $`T_N`$ of interest, in particular we choose $`T_NT_K`$ to calculate ground-state properties. In each iteration step, we keep the information on the transformation between one set of eigenstates and the next, i.e. we save the corresponding unitary matrix. After obtaining the relevant excitations at temperature $`T_N`$ one can define the density matrix
$$\widehat{\rho }=\underset{m}{}e^{E_m^N/T_N}|m_Nm|$$
(5)
which completely describes the physical state of the system. In particular, the equilibrium Green’s function can be written as
$$G_{}(t)=i\theta (t)\mathrm{Tr}\left(\widehat{\rho }\{f_{}^{}(t),f_{}^{}(0)\}\right)$$
(6)
2) Now we repeat the iterative diagonalization for the same parameters. Each iteration step $`N^{}`$ yields the single-particle excitations (and matrix elements of $`f^{}`$) relevant at a frequency $`\omega T_N^{}`$. But instead of using (3), we now employ (6) and evaluate the spectral function with respect to the correct *reduced density matrix* : As depicted in fig. 2, the complete chain is split into a smaller cluster of length $`N^{}`$ and an *environment* containing the remaining degrees of freedom. In the product basis of these two subsystems, the full density matrix has the form
$$\widehat{\rho }=\underset{\genfrac{}{}{0pt}{}{m_1m_2}{n_1n_2}}{}\rho _{m_1n_1,m_2n_2}|m_1_{\mathrm{env}}|n_1_{\mathrm{sys}}n_2|m_2|$$
(7)
which is in general not diagonal anymore. Performing a partial trace on the environment then yields the density submatrix
$$\widehat{\rho }^{\mathrm{red}}=\underset{n_1n_2}{}\rho _{n_1n_2}^{\mathrm{red}}|n_1_{\mathrm{sys}}n_2|$$
(8)
with
$$\rho _{n_1n_2}^{\mathrm{red}}=\underset{m}{}\rho _{mn_1,mn_2}$$
(9)
This projection is easily done using the previously stored unitary transformation matrices. Note that $`\rho ^{\mathrm{red}}`$ – defined only on the shorter chain – contains all the relevant information about the quantum mechanical state of the *full* system. This concept has been applied very successfully in the density matrix renormalization group (DMRG) , where the projection on a smaller subsystem is essential for obtaining eigenstates of the model. In NRG, on the other hand, diagonalization can be performed directly due to the logarithmic discretization, but to describe the effect of the chain degrees of freedom on the impurity (or a small cluster) one has to determine $`\rho ^{\mathrm{red}}`$. In the following, we therefore refer to the calculational scheme presented here as DM-NRG.
In fig. 1 we compare the spectrum calculated by the DM-NRG to the one obtained with the NRG version used so far in the literature (the same number of levels has been used in both calculations). The strong shift of spectral weight due to the polarized impurity is now clearly seen, as well as a slight change in the height and shape of the quasiparticle peak. The magnetization has been calculated from (4) for different values of $`h`$ and is in good agreement with the static calculation (see fig. 3). The remaining deviation of about three percent is due to an error in the total spectral weight.
The resulting field dependence of the spectrum is displayed in fig. 4. With increasing $`h`$, the Kondo resonance is suppressed and eventually merges with the lower atomic level. Regarding the total density of states (DOS) $`A(\omega )=_\sigma A_\sigma (\omega )`$, the Kondo peak is split by the field and the DOS at the Fermi level strongly reduced. This effect has been observed directly in measurements of the differential conductance through a quantum dot .
So far calculations have been at $`T=0`$. Upon increase of the temperature at a finite magnetic field, we expect a reduction of the average impurity magnetization due to thermal fluctuations. As a consequence, particle excitations with polarization in the field direction should gain spectral weight. In fig. 5, this effect is obvious: At temperatures $`Th`$, the asymmetry in $`A_{}(\omega )`$ is strongly reduced. Note that in finite temperature NRG calculations, no spectral information can be obtained at frequencies $`\omega T`$. In this region data have to be fitted. This important fact will be discussed in detail in a subsequent publication.
Results for an asymmetric impurity close to the mixed valence regime are shown in fig. 6. The almost complete shift of spectral weight to the particle (hole) sector is again observed for the two spin polarizations, which in this case are not symmetric anymore. In the total DOS (fig. 7), changes are less prominent. We merely observe a suppression of the quasiparticle peak and a redistribution of the corresponding weight to higher frequencies.
Comparing our findings to previous calculations, it should be pointed out that up to now only the modified perturbation theory and the Quantum Monte Carlo method (QMC) have been applied to calculate the impurity spectrum in a magnetic field. The former is limited to small $`U`$, while QMC calculations have so far only been done in the mixed valence regime (and at temperatures $`TT_K`$) due to the increase in computing cost for the symmetric case. In a recent NRG calculation on the Kondo model , the problems discussed here did not occur due to the absence of atomic levels. Apart from these restrictions, we find qualitative agreement with our results, which do not suffer from similar limitations.
In conclusion, we have presented a new method for calculating dynamical properties at arbitrary temperatures within the numerical renormalization group. It has been demonstrated that – despite logarithmic discretization – energy scale separation is in general not valid in the case of spectral quantities. This effect is neglected in the NRG scheme used so far in the literature. Within our generalized procedure (DM-NRG), based on the reduced density matrix, we can now account for changes in the ground state occuring at energies far below the external frequency scale.
The method introduced here has been applied to the Anderson impurity in an external magnetic field, which is of great interest in view of recent transport measurements of quantum dots. Nonperturbative $`T=0`$ studies have not been performed so far, mainly because of technical difficulties in extending NRG to systems with broken spin symmetry. Our spectral results are in excellent agreement with the sum rule provided by the (static) magnetization. In the total density of states we find the splitting and suppression of the quasiparticle peak which is also observed experimentally.
Future applications of the DM-NRG include DMFT calculations for phases with long range order, where symmetry-breaking perturbations and their effect on the spectrum have to be treated reliably. In addition, more complex impurity systems including orbital degeneracy may be studied, which (due to the rapid advances in nanoscale preparation techniques ) are of growing experimental interest.
The author would like to thank D. Vollhardt, R. Bulla and H. Kontani for valuable discussions. This work was supported in part by the DFG through SFB 484.
|
warning/0004/astro-ph0004192.html
|
ar5iv
|
text
|
# DIRBE Minus 2MASS: Confirming the Cosmic Infrared Background at 2.2 Microns
## 1 Introduction
The Diffuse InfraRed Background Experiment (DIRBE) on the COsmic Background Explorer (COBE, see Boggess et al. (1992)) observed the entire sky in 10 infrared wavelengths from 1.25 to 240 $`\mu `$m. Hauser et al. (1998) discuss the determination of the Cosmic InfraRed Background (CIRB) by removing foreground emission from the DIRBE data. This paper detected the CIRB at 140 and 240 $`\mu `$m, but only gives upper limits at shorter wavelengths. From 5 to 100 $`\mu `$m, the zodiacal light foreground due to thermal emission from interplanetary dust grains is so large that no reliable estimates of the CIRB can be made from a position 1 AU from the Sun (Kelsall et al., 1998). In the shorter wavelengths from 1.25 to 3.5 $`\mu `$m, the zodiacal light is fainter, but uncertainties in modeling the foreground due to Galactic stars are too large to allow a determination of the CIRB (Arendt et al., 1998). Recently, Gorjian et al. (2000) removed the Galactic star foreground by directly measuring the stars in a $`2{}_{}{}^{}\times 2^{}`$ box using ground-based telescopes and then subtracting the stellar contribution from the DIRBE intensity on a pixel-by-pixel basis. This field, a DIRBE dark spot, was selected using DIRBE data to have a minimal number of bright Galactic stars. In addition, Wright & Reese (2000) used a histogram fitting method to remove the stellar foreground from the DIRBE data in a less model-dependent way than that used by Arendt et al. (1998). Gorjian et al. (2000) and Wright & Reese (2000) obtained consistent estimates of the CIRB at 2.2 and 3.5 $`\mu `$m. With the recent 2<sup>nd</sup> incremental release of 2MASS data, it is now possible to apply the direct subtraction method of Gorjian et al. (2000) to four additional DIRBE dark spots scattered around the North and South Galactic polar caps.
Kashlinsky & Odenwald (2000) have claimed a detection of the fluctuations of the CIRB. Kashlinsky & Odenwald (2000) also give the range 0.05 to 0.1 for the ratio of the fluctuations in the DIRBE beam to the mean intensity for the CIRB. But this ratio and fluctuation combine to give a range of CIRB values that is incompatible with the Hauser et al. (1998) upper limits on the CIRB, especially at 1.25 $`\mu `$m. Furthermore, the claimed cosmic fluctuations are larger than the residuals in the DIRBE$``$2MASS fits presented in §3. In this paper, Kashlinsky & Odenwald (2000) is treated as an upper limit on the CIRB which is compatible with previous limits and the results found here. Wright (2001) will discuss the possible cosmic fluctuation signal in the DIRBE$``$2MASS residuals in more detail.
## 2 Data Sets
The two main datasets used in this paper are the DIRBE maps and the 2MASS point source catalog (PSC).
The DIRBE weekly maps were used: DIRBE\_WKnn\_P3B.FITS for $`04\text{nn}44`$. These data and the very strong no-zodi principle described by Wright (1997) were used to derive a model for the interplanetary dust foreground that is described in Wright (1998) and Gorjian et al. (2000). The zodiacal light model was then subtracted from each weekly map, and the remainders were averaged into mission averaged, zodiacal subtracted maps. At 1.25 and 2.2 $`\mu `$m, no correction for interstellar dust emission is needed (Arendt et al., 1998). The pixels in these mission averaged, zodiacal subtracted maps provide the DIRBE data, $`D_i`$.
DIRBE has a $`0.7^{}\times 0.7^{}`$ square beam with a diagonal of $`1^{}`$. Thus a thick buffer ring is needed around any studied field to keep bright stars outside the field from influencing the measured DIRBE intensity. One can minimize the resulting inefficiency by studying fields with a large area:perimeter ratio. Large circular regions have the largest area:perimeter ratio, and for circles as large as $`3^{}`$ diameter it is still possible to find fields that have no stars brighter than the 2MASS saturation limit. To find such fields, a list of DIRBE dark spots was generated by smoothing the zodi-subtracted 3.5 $`\mu `$m map with a kernel given by
$$\mathrm{log}_2W=\frac{|\widehat{n}\widehat{n}^{}|^2}{3(0.03023^2|\widehat{n}\widehat{n}^{}|^2)}$$
(1)
where $`\widehat{n}`$ and $`\widehat{n}^{}`$ are unit vectors. This kernel and all of its derivatives are continuous, and it vanishes identically for radii greater than $`2\mathrm{sin}^1(0.03023/2)=1.732^{}`$. The FWHM is 3. The 20 faintest spots of the smoothed map in the Northern Galactic Hemisphere and the 20 faintest spots in the Southern Galactic Hemisphere were then located. The darkest spot is in the Northern Hemisphere and was studied by Gorjian et al. (2000) using ground-based data.
The 2MASS data were obtained over the WWW between 16 Mar 2000 and 23 Mar 2000. The IRSA interface was used to search the catalog. This allows a maximum search area of a 1 radius circle, which is too small for a comparison to the DIRBE data taken with a 0.7 beam. Therefore, a total of seven cone searches in a pattern consisting of a hexagon plus the central position were made around each DIRBE dark spot. These searches were restricted to stars brighter than $`K_s=14`$. The seven resulting files were combined by stripping the table headers, concatenating, sorting, and then using the UNIX uniq filter. If there are no gaps in the 2MASS coverage near a dark spot, this yields a catalog that covers a 2 radius circle around the dark spot plus six small “ears.” Only stars within the circle were used in this analysis. However, there are usually gaps in the 2MASS coverage. Searching the 20 darkest spots in each of the Galactic polar caps produced only four usable fields out of 40 candidates. These are listed in Table 1. $`\beta `$ is the ecliptic latitude. The combined catalogs for each field were converted into star charts which were checked for missing 2MASS data. Two of these four fields have coverage gaps near the edge of the 2 radius circle, and thus have $`r<2^{}`$ and a smaller but still useful number of pixels. Figure 1 shows the 2MASS catalog stars superimposed on the DIRBE map for the dark spot at $`(l,b)=(107.7{}_{}{}^{},57.7{}_{}{}^{})`$.
## 3 Analysis
The DIRBE data for the $`i^{th}`$ pixel is $`D_i`$, and should be the sum of the zodiacal light, $`Z_i`$; the cataloged stars, $`B_i`$; the faint stars, $`F_i`$; and the CIRB, $`C`$. The cataloged star contribution is found using the method of Gorjian et al. (2000) on all stars brighter than $`K=14`$. In this method, the DIRBE beam center is assumed to be uniformly distributed within the area of the $`i^{th}`$ pixel, and the DIRBE beam orientation is assumed to be uniformly distributed in position angle. Under these assumptions, the probability that the $`j^{th}`$ star contributes to the signal in the $`i^{th}`$ pixel is $`p_{ij}`$, and the cataloged star contribution is
$$B_i=\mathrm{\Omega }_b^1\underset{j}{}p_{ij}S_j$$
(2)
where $`S_j`$ is the flux of the $`j^{th}`$ star and $`\mathrm{\Omega }_b`$ is the solid angle of the DIRBE beam.<sup>1</sup><sup>1</sup>1Gorjian et al. (2000) used $`\mathrm{\Omega }_p`$ instead of $`\mathrm{\Omega }_b`$ in Equation 2: this is appropriate for $`p`$’s normalized to the total flux as in Table 5 of Wright & Reese (2000), but for $`p`$’s normalized to a peak of unity as in Figure 2 the beam solid angle must be used for the normalization of Equation 2. Figure 2 shows the probability $`p_{ij}`$ for stars located in the center, near an edge, or near a corner of a pixel located near the center, an edge, or a corner of a cube face in the quadrilateralized spherical cube pixel scheme. The standard deviation of the bright star contribution is given by
$$\sigma ^2(B_i)=\mathrm{\Omega }_b^2\underset{j}{}[p_{ij}(1p_{ij})+p_{ij}^2(0.1+(0.4\mathrm{ln}10)^2\sigma ^2(m_j))]S_j^2.$$
(3)
The first term in the $`[]`$’s in Equation 3 is the “flicker noise” caused by a star that is on the edge of the DIRBE beam. The second term represents the flux error for the $`j^{th}`$ star, and it includes a generous allowance for variability between the time of the DIRBE observations and the the time of the 2MASS observations: the “0.1” corresponds to $`\sigma =0.34`$ magnitudes. This standard deviation is shown by the error bars in Figures 3 and 4. The actual noise on the DIRBE data is negligible: $`1\text{kJy sr}\text{-1}`$ at $`1.25\mu \text{m}`$ and $`1.2\text{kJy sr}\text{-1}`$ at $`2.2\mu \text{m}`$ (Hauser et al., 1998). The uncertainty in the DIRBE zero level, or offset, is also negligible (Hauser et al., 1998). The CIRB should be isotropic, and $`F`$ should vary only slightly within a 2 radius of a dark spot. But both the zodi-subtracted data, $`\text{DZ}_i`$, and the bright star contribution fluctuate greatly from DIRBE beam to DIRBE beam due to the confusion noise caused by overlapping stars. Figures 3 and 4 show plots of $`\text{DZ}_i`$ vs. $`B_i`$ for each of the four fields. The gray points with large error bars are pixels with a bright star at the edge of the beam. These bright stars usually saturate the 2MASS survey, and are assigned a nominal magnitude of 4 and a nominal flux error of a factor of ten. This large flux error guarantees that the pixels affected by saturated stars have no effect on the subsequent analysis. The lines have unit slope with intercepts determined using a weighted median procedure. The average values of the data D<sub>i</sub>, the zodi-subtracted data DZ<sub>i</sub>, and the derived intercepts DZ(0) for each of the four fields are given in Table 1.
The 2MASS $`K_s`$ magnitudes were converted to DIRBE fluxes using the $`F_{}(K)=614\text{Jy}`$ derived by Gorjian et al. (2000) from the median of determinations using seven bright red stars. This assumes that there is no significant $`K_sK`$ color difference between the red calibration stars used by Gorjian et al. (2000) and the stars in the dark spots. The 2MASS observations saturate \[even in the first read\] on stars brighter than $`5^{th}`$ magnitude. At this level, the DIRBE data are still substantially affected by confusion noise, so a direct comparison of DIRBE and 2MASS on the same stars will not give a high precision result, but the overall agreement between the unity slope lines and the data in Figures 3 and 4 shows that a direct DIRBE to 2MASS comparison is consistent with the Gorjian et al. (2000) calibration. Figure 5 shows a histogram of the values $`\text{DZ}_iB_i\text{DZ(0)}`$ for the four dark spots combined. Note that the standard deviations derived from the three highest bins of these histograms of the individual histograms are 1.27, 1.93, 2.82, and 2.93 kJy sr<sup>-1</sup>, while the standard deviation of the Gaussian fit in Figure 5 is 1.81 kJy sr<sup>-1</sup>. Since this histogram includes the DIRBE detector noise, any small scale errors in the zodiacal light model, any small scale structure in the faint star contribution, a contribution from stellar variability, and a calibration error contribution in addition to any real extragalactic fluctuation, one can take $`1.81\text{kJy sr}\text{-1}=2.47\text{nW m}\text{-2}\text{ sr}\text{-1}`$ as an upper limit on the extragalactic fluctuation $`\delta C_{rms}`$.
Stars fainter than $`K=14`$ contribute a small amount which must be subtracted from the intercepts. This contribution was evaluated using the Wainscoat et al. (1992) star count model. But Wright & Reese (2000) and Gorjian et al. (2000) find that this model overpredicts high latitude star counts by 10% in the $`6<K<12`$ range. After applying this 10% correction, which assumes that the same ratio applies to $`K>14`$, the faint star corrections are $`F=1.58`$, 1.87, 1.50, and 2.03 kJy sr<sup>-1</sup> in the four fields. An uncertainty of 10% of the total model prediction is assigned to this correction, and listed in Table 2 under “Faint Source”. Galaxies brighter than $`K=14`$ may be subtracted incorrectly, and their contribution should be added back into the CIRB. Galaxies with $`K<14`$ add up to 0.35 kJy sr<sup>-1</sup>, according to the empirical fits of Gardner et al. (1993). A fraction of these galaxies will be in the 2MASS PSC and, since galaxies should not be subtracted, these incorrectly subtracted galaxies should be added back to the CIRB. The fluxes in the 2MASS Extended Source Catalog objects with $`K_s<14`$ add up to 0.25 kJy sr<sup>-1</sup>, indicating that the correction for galaxies in the PSC should be on the order of $`0.1`$ kJy sr<sup>-1</sup>. In addition, 30% of the Extended Source Catalog objects are coincident with PSC objects suggesting that the compact galactic nuclei – which are in the PSC but should not be subtracted from the CIRB – account for 0.1 kJy sr<sup>-1</sup>. Thus the CIRB estimates shown in Table 1 are given by $`\text{DZ(0)}\text{F}+0.1\text{kJy sr}\text{-1}`$. An uncertainty of 100% of this correction is included in Table 2 under “Galaxies”.
Note that the statistical uncertainties in the intercepts are all $`0.4\text{kJy sr}\text{-1}`$ and thus negligible compared to systematic errors in the interplanetary dust model. Gorjian et al. (2000) adopt an uncertainty of 5% of the zodiacal intensity at the ecliptic poles, and this gives an uncertainty of $`\pm 3.79\text{kJy sr}\text{-1}`$. The effect of a $`\pm 10\%`$ calibration error between the DIRBE flux scale and the 2MASS magnitude scale would be a systematic $`\pm 2.58\text{kJy sr}\text{-1}`$ change in the CIRB. The precision of the Gorjian et al. (2000) calibration of the DIRBE $`K`$-band fluxes to the standard infrared magnitude is $`\pm 2.1\%`$. This DIRBE flux calibration agreed with Arendt et al. (1998) calibration to better than $`1\%`$. Thus a $`10\%`$ uncertainty in the DIRBE vs. 2MASS calibration appears to be conservative. This uncertainty is listed in Table 2 under “Calibration”. The mean of the CIRB estimates in Table 1 is $`14.79\pm 0.51\text{kJy sr}\text{-1}`$. This standard deviation of the mean of the 4 fields is listed in Table 2 under “Scatter”. Finally, the largest uncertainty is the zodiacal light modeling uncertainty. Adding the errors in Table 2 in quadrature gives a result of $`14.8\pm 4.6`$ kJy sr<sup>-1</sup> or $`20.2\pm 6.3`$ nW m<sup>-2</sup> sr<sup>-1</sup>.
## 4 J Band
For the $`J`$-band the contribution from stars with $`J<14`$ and $`K<14`$ was calculated on a pixel by pixel basis. This dual wavelength magnitude selection is essentially equivalent to a simple $`J<14`$ selection. There are very few stars in high Galactic latitude fields with color $`JK<0`$. Tests in three fields using deeper samples from the 2MASS catalog show that imposing the $`K<14`$ cut on a $`J<14`$ sample reduces the intensity by $`<0.3`$%. Table 3 gives the photometric quantities for the four dark fields at 1.25 $`\mu `$m. The $`(10\pm 10)\%`$ correction of the faint star contribution derived from $`K`$-band star counts has been applied to $`F`$. Assuming a color of $`JK=1`$ for galaxies gives an estimate of $`0.05\text{kJy sr}\text{-1}`$ for the improperly subtracted faint galaxy contribution. This has been added to the CIRB estimates in the table. The mean of the CIRB estimates is $`12.04\pm 1.49\text{kJy sr}\text{-1}`$, and this would change by $`3.10\text{kJy sr}\text{-1}`$ for $`\pm 10\%`$ changes in the flux of 1512 Jy for $`0^{th}`$ magnitude at $`1.25\mu \text{m}`$ used in this paper. This value is the median of calibrations based on $`\beta `$ And, $`\alpha `$ Tau, $`\alpha `$ Aur, $`\alpha `$ Boo, $`\alpha `$ Her, and $`\beta `$ Peg. This calibration has an uncertainty of $`\pm 2\%`$ and differs from the Arendt et al. (1998) calibration by -2.4%. A $`\pm 10\%`$ uncertainty in the calibration of DIRBE vs. 2MASS has been adopted giving the error $`\pm 3.10\text{kJy sr}\text{-1}`$ listed in Table 2 . The systematic error due to interplanetary dust modeling is 5% of the ecliptic pole zodiacal intensity or $`\pm 5.87\text{kJy sr}\text{-1}`$. Adding the errors in quadrature gives a result of $`12.0\pm 6.8`$ kJy sr<sup>-1</sup> or $`28.9\pm 16.3`$ nW m<sup>-2</sup> sr<sup>-1</sup>. This is obviously not a significant detection due to the large uncertainty in the zodiacal foreground. However, a $`2\sigma `$ upper limit is $`\nu I_\nu <62\text{nW m}\text{-2}\text{ sr}\text{-1}`$ which is a slight improvement on Hauser et al. (1998).
The width of the combined histogram of the residuals for the $`J`$-band data is $`2.32\text{kJy sr}\text{-1}`$ or $`5.6\text{nW m}\text{-2}\text{ sr}\text{-1}`$ which gives an upper limit on extragalactic fluctuations since it also includes detector noise and star subtraction errors.
A CIRB of $`12.0\pm 6.8`$ kJy sr<sup>-1</sup> at 1.25 $`\mu `$m is fainter than the prediction of Dwek & Arendt (1998), whose correlation gives $`23.5\pm 8.6\text{kJy sr}\text{-1}`$ for this paper’s $`K`$-band CIRB. Given the combined uncertainties in the difference, this is $`<1.2\sigma `$ higher than the result in this paper.
## 5 Discussion
Subtracting the 2MASS catalog from the zodi-subtracted DIRBE data yields a statistically significant, isotropic background at 2.2 $`\mu `$m of $`14.8\pm 4.6`$ kJy sr<sup>-1</sup> which is consistent with the earlier results from Gorjian et al. (2000) $`(16.4\pm 4.4\text{kJy sr}\text{-1})`$ and Wright & Reese (2000) $`(16.9\pm 4.4\text{kJy sr}\text{-1})`$ within the systematic error associated with the modeling the zodiacal dust cloud. Averaging the results of Gorjian et al. (2000), Wright & Reese (2000) and this paper gives a CIRB at 2.2 $`\mu `$m of $`16\pm 4\text{kJy sr}\text{-1}`$. This averaging has not reduced the estimated error which is dominated by systematic effects that affect all three results equally. The foreground due to interplanetary dust at 1.25 $`\mu `$m is too large to allow a CIRB detection, but an improved upper limit is found. Note that the Zodi-Subtracted Mission Average maps which used the Kelsall et al. (1998) zodiacal light model give a CIRB that is 13.75 kJy sr<sup>-1</sup> larger at 1.25 $`\mu `$m and 6.08 kJy sr<sup>-1</sup> larger at 2.2 $`\mu `$m than results obtained here using the zodiacal light model described in Wright (1998) and Gorjian et al. (2000) based on the very strong no-zodi principle of Wright (1997). Figure 6 shows a plot of the $`J`$-band intensity vs. $`K`$-band intensity averaged over the four dark spots analyzed in this paper. Three values are plotted: the average total intensity $`\text{D}`$, the average zodi-subtracted intensity $`\text{DZ}`$, and the CIRB estimates. The Hauser et al. (1998) upper limits on the CIRB, the Dwek & Arendt (1998) correlation and the $`1\sigma `$ error bars from this paper are shown as well. This figure emphasizes the large subtractions that are involved in determining the CIRB from data taken 1 AU from the Sun: the zodiacal light is about 16 times larger than the CIRB at 1.25 $`\mu `$m and 8 times larger than the CIRB at 2.2 $`\mu `$m. Galactic stars are a problem in the large DIRBE beam, but in the selected dark spots the effect of stars is 4 times less than that of the zodiacal light.
Bernstein (1999) has measured the optical extragalactic background light and obtained results at $`\lambda =0.8,\mathrm{\hspace{0.33em}0.55},\text{\&}\mathrm{\hspace{0.33em}0.3}\mu \text{m}`$ which are consistent with a reasonable extrapolation through the uncertain $`J`$-band result found here, as shown in Figure 7. Both Bernstein (1999) and this work face challenging and uncertain corrections for the zodiacal light, but the two papers use very different techniques and should not have systematic errors in common. Thus the lack of a discontinuity in the spectrum between 0.8 and 1.25 $`\mu `$m is an indication in favor of the background level reported here. The model shown in Figure 7 is the $`\mathrm{\Lambda }`$CDM-Salpeter model from Primack et al. (1999) which appears to fit the observed far IR to near IR to optical ratios. But the model was multiplied by 1.84 to match the level of the observed background.
The COBE datasets were developed by the NASA Goddard Space Flight Center under the guidance of the COBE Science Working Group and were provided by the NSSDC. 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, funded by the National Aeronautics and Space Administration and the National Science Foundation.
|
warning/0004/astro-ph0004171.html
|
ar5iv
|
text
|
# Resolving SNR 0540-6944 from LMC X-1 with Chandra
## 1 Introduction
Recent observations of the Large Magellanic Cloud (LMC) have opened up a new era in near-extragalactic studies of supernova remnants (SNRs). Thanks to advances in instrumentation over the last decade, we have been able to observe this large and varied sample of remnants without the difficulties in distance determination and obscuration for Galactic remnants, or those in resolution and sensitivity for ones in more distant galaxies. This has allowed us both to examine individual remnants in detail (e.g., Williams et al. 1997, 1999a; Chu et al. 1997, 2000) and as a group (e.g., Williams et al. 1999b; Williams 1999) to explore their interactions with the surrounding interstellar medium (ISM).
The completeness of the LMC sample, however, is a vexing question. Many SNRs lack one or more of the traditional SNR signatures (high \[S II\]/H$`\alpha `$ ratios, nonthermal radio emission, and X-ray emission) and must be identified through other means (e.g., Chu 1997). New LMC SNRs continue to be found (e.g., Smith et al. 1994; Chu et al. 1993, 1995, 1999). Each discovery not only adds to the list of known SNRs, but provides additional insight into how to detect other such SNRs. This has profound implications for both the completeness of the LMC sample and the assemblage of samples from other galaxies, and thus for our understanding of the contribution from SNRs to the energetics and dynamics of the ISM.
One of these discoveries, SNR 0540$``$6944, may prove particularly informative in this regard. The SNR, discovered by Chu et al. (1997), is difficult to observe in optical and radio due to emission from the surrounding H II region N159; the X-ray emission was similarly obscured by that from the nearby X-ray binary LMC X-1. The SNR’s expansion was serendipitously uncovered in optical echelle observations, and its SNR nature confirmed by using soft-band (0.1-1.0 keV) ROSAT PSPC images to separate the SNR’s thermal emission from the much harder emission of LMC X-1. However, the SNR remained spatially and spectrally confused with LMC X-1 at the resolution of the ROSAT observations.
The unprecedented combination of spatial and spectral resolution and sensitivity from the Chandra (formerly AXAF) X-ray Observatory and its instruments allows us, for the first time, to get a picture of SNR 0540$``$6944 distinct from that of LMC X-1. Using these data, we are able to present images and spectra from the object itself. The results confirm the SNR identification of Chu et al. 1997; provide spatial and spectral data on the SNR itself; and illustrate the information that may be gleaned by separating objects whose close proximity confuses their emission. It provides a striking demonstration of the many new areas of investigation made possible by Chandra ’s power, even in a brief observation.
## 2 Observations
LMC X-1 was observed as a calibration source by Chandra in late 1999 and early 2000. For our spectral and soft-band imaging studies, we used the Advanced CCD Imaging Spectrometer (ACIS) on-axis observation (sequence number 490002, 6.5 ksec) to avoid additional problems with off-axis distortions. Additionally, the target was focused on chip S3, a back-illuminated chip, which allowed for slightly greater spectral sensitivity, while avoiding the problems with radiation damage to which the front-illuminated chips were subject (Orbital Calibration reports, ASC, 1999). Only about 3 ksec of data are available for analysis; other datasets available for this object were off-axis, contained processing errors, or both. The ACIS (using a back-illuminated chip) has an angular resolution of 1″, an energy resolution of 100 eV at 1.0 keV (E/$`\mathrm{\Delta }`$E =9), an energy range of 0.2-10 keV, and an effective area of 600 cm<sup>-2</sup> at 1.0 keV (Chandra Observatory Guide, ASC, 1997). Data reduction and analysis were performed using the CIAO (Chandra X-ray Center software) and FTOOLS, XSPEC, and XIMAGE data-processing routines (Arnaud 1996).
## 3 Analysis
While SNR 0540$``$6944 can be spatially separated from LMC X-1 using the ACIS-S, the flux from the SNR is very low, and close to the background level. As a result, both spatial and spectral analysis are difficult. In addition, due to frame transfer effects, a “spike” of emission from LMC X-1 intersects the circle of emission from the SNR. These factors complicate attempts to isolate the the SNR itself from its much brighter neighbor.
### 3.1 Morphological Analysis
A first look at the image of this region on the Chandra ACIS-S is disappointing. Even using the S3 back-illuminated chip, more sensitive to low-energy emission, the SNR’s presence is barely detectable next to that of LMC X-1. However, the emission from SNR 0540$``$6944 is likely to fall largely in the energy band between 0.1 and 3 keV, as is expected for a thermal plasma. We therefore make exposure-corrected images in the soft (0.3-3.0 keV) and hard (3.0-9.0 keV) bands and compare them. (Fig. 1a-b; note that we have used a low-energy cutoff of 0.3 keV to reduce contributions from the soft X-ray background. No counts are expected from the LMC below this cutoff due to the intervening column density.) We do indeed see emission at the position of SNR 0540$``$6944 in the soft image; but it is still overwhelmed by that from LMC X-1.
In order to eliminate some of the contamination by LMC X-1, we subtract the hard map from the soft map, thus removing emission from areas where hard X-rays dominate. In order to bring up the contrast in the resulting map, we divide by a total map. What remains is a “softness ratio” map in the form (soft - hard) / (soft + hard). This can be used to discern the structure of the soft emission, presumably that from the SNR (Fig. 1c-d). What is thus revealed is a roughly circular structure centered at J2000.0 coordinates 05<sup>h</sup>40<sup>m</sup>05<sup>s</sup>, -69°44′07″. The circle has a rough diameter of $``$1$`{}_{}{}^{}.25`$, or 19 pc at the distance to the LMC (50 kpc). The emission is distributed over the remnant, suggesting a thick-shelled structure. The remnant is considerably brighter in a small ($``$9″ radius) region to the northeast.
For the purpose of optical comparison we obtained archival Hubble Space Telescope images in the H$`\alpha `$ and \[O III\] emission lines (Fig. 1e-f; PEP ID 6535; H$`\alpha `$: four 300 sec exposures; \[O III\]: four 230 sec exposures). These images show a roughly circular, highly filamentary structure amidst the emission from the rest of the N159 region. This structure corresponds very well to the position and extent of X-ray emission revealed by the ACIS ratio image, suggesting that the optical emission comes from the cooling shell of the SNR.
### 3.2 Spectral Analysis
In order to separate the SNR emission from that of LMC X-1, it becomes useful to consider the emission of the X-ray binary itself. In this we are aided by the availability of data from the ASCA SIS (ad43004000, 12.5 ksec). ASCA is insensitive at energies below 0.7 keV, so the contribution from the SNR to the data is expected to be minimal. We can, therefore, take the ASCA data as representative of LMC X-1 alone.
We therefore approached the problem using four separate spectra. Two were from ASCA observations using the SIS0 and SIS1 instruments; the region covered includes both LMC X-1 and SNR 0540$``$6944. A third spectrum was extracted from Chandra ACIS-S data, similarly covering a region including both LMC X-1 and SNR 0540$``$6944. A fourth spectrum, also extracted from Chandra ACIS-S data, covered SNR 0540$``$6944 only.
From previous studies of LMC X-1 (e.g., Schlegel et al. 1994) we know that the spectrum of this X-ray binary is well represented by the combination of disk-blackbody and power-law models. Previous studies (e.g., Schmidtke, Ponder, & Cowley 1999) suggest that there are no significant long-term variations in the spectrum of LMC X-1, allowing us to meaningfully compare datasets taken at different times. We expect the SNR contribution to be reasonably well modeled by a thermal plasma model (Raymond & Smith 1977). Our combined model for the region, then, has four components: one for photoelectric absorption (based on Morrision & McCammon 1983), applied to a combined Raymond-Smith, disk-blackbody and power-law model. Abundances for the Raymond-Smith model were set to 0.3 solar, as appropriate to the ISM of the LMC (Russel & Dopita 1992).
This model was simultaneously fit to our four spectra. The model parameters were linked, with the exceptions of the normalizations, which were allowed to fit independently to the four spectra. For the ASCA spectra, the Raymond-Smith normalizations were set to zero, as little thermal contribution was expected. Likewise, the normalizations for the disk-blackbody and power-law components were set to zero for the Chandra spectrum of 0540$``$6944 alone, as the contributions from LMC X-1 were expected to be minimal.
The best-fit parameters are given below, with the 90% confidence ranges given in parentheses. The best fit for the X-ray absorption of the region, $`N_H=7.2(6.97.5)\times 10^{21}`$ cm<sup>-2</sup>, is similar to that found by Schlegel et al. (1994) and Schmidtke et al. (1999) for LMC X-1; it is also somewhat atypically high for the LMC. The best-fit parameters for LMC X-1 are a blackbody temperature of kT$`{}_{bb}{}^{}=`$0.82 (0.81-0.83) and a power-law index of $`\mathrm{\Gamma }`$=2.3 (2.2-2.4), again consistent with the results from Schlegel et al. (1994) and Schmidtke et al. (1999). The thermal plasma component yields a best-fit temperature of kT<sub>rs</sub>=0.31 (0.18-0.43), reasonable for an older SNR (Fig. 2). Based on these spectral results, we computed a flux from the SNR in the 0.3-3 keV range of 2.8 $`\times 10^{14}`$ erg cm<sup>-2</sup> s<sup>-1</sup>, and a luminosity in the same range of 8.4 $`\times 10^{33}`$ erg s<sup>-1</sup>. When we look at the data for SNR 0540$``$6944 and LMC X-1 combined, we find a flux in the 0.3-3 keV range of 1.9 $`\times 10^{12}`$ erg cm<sup>-2</sup> s<sup>-1</sup> and a luminosity of 5.7 $`\times 10^{35}`$ erg s<sup>-1</sup>. Thus, within the energy range specified, the SNR contributes about 1.5% of the X-ray flux from this region.
## 4 Discussion
Morphologically, the SNR is without a sharply defined shell in X-rays. The large and distributed X-ray structure, interior to much of the optical emission, implies an older, Sedov-stage SNR. This picture is strengthened by the relatively low temperature and low luminosity of the X-ray emission. The velocity of $``$180 km s<sup>-1</sup> found by Chu et al. (1997) indicates a slow expansion consistent with this picture. If this velocity is representative of the actual expansion velocity of the remnant, and the SNR is undergoing Sedov-like expansion, we would expect the shock velocity to be $`v_{shock}=4/3v_{exp}=240`$ km s<sup>-1</sup>.
This shock velocity, in turn, would imply a temperature of $`kT=3/16\mu v_{shock}^2`$, where $`\mu `$ is the mean molecular weight (assumed 1.1$`m_H`$). For the shock velocity given above, this gives approximately 0.12 keV. The temperature derived from X-ray spectral fits is somewhat higher; the X-ray temperature would imply $`v_{exp}=280`$ (220-340) km s<sup>-1</sup>. Similar discrepancies have been noted for other LMC remnants (Williams 1999). One possible explanation for this discrepancy is that the highest-velocity material may be too faint to show up clearly in the echelle spectrum. Another possibility is that the expansion velocity is indeed accurately reflected by the echelle spectroscopy and that this slow shock speed is insufficient to produce X-ray emission at the shock front. In this latter case, the remnant may have entered the shell-forming phase, as evidenced by the pronounced H$`\alpha `$ shell. The observed X-rays in such a case are likely to be “fossil” radiation produced by the cooling of gas shocked to high temperatures earlier in the remnant’s expansion.
Using the Sedov equation and the velocity from Chu et al. (1997), and assuming constant external density, we find an age of $``$20,000 yr for this SNR. The velocity derived from the X-ray temperature would give an age of $``$12,000 yr. These must be regarded as only approximate figures. For instance, if the SNR is expanding in the wind-blown bubble formed by its progenitor - a quite plausible scenario, as the remnant is within an H II region - it may very well be considerably younger than these estimates. Given this estimate of age, we do not expect a causal connection between SNR 0540$``$6944 and LMC X-1. The projected distance between the center of the SNR and LMC X-1 is about 1$`\stackrel{}{\mathrm{.}}`$75, or a minimum of 26 pc separation. To reach this distance within the estimated age of the SNR, the compact object would have had to travel at a constant speed of over 1200 km s<sup>-1</sup>.
Our fits to the X-ray spectrum use a normalization constant directly related to the emissivity, $`K=10^{14}n_en_H𝑑V/(4\pi D^2)`$. Here D is the distance to the remnant, $`n_e`$ and $`n_H`$ the electron and particle densities (we assume $`n_e=1.1n_H`$), and $`V`$ the remnant volume. This allows us to make estimates of the density, energy and pressure within the X-ray emitting gas of this remnant. Using the fitted value of $`K=5.66\times 10^4`$ and assuming a volume filling factor for the gas of 10% (corresponding to a shell thickness of about 0.33 pc, 3% of the radius, a reasonable value for a middle-aged remnant), we calculated the gas density at about $`n`$=1.2 cm<sup>-3</sup>. Using the formula $`E_{th}=(3/2)NkT`$, where N is the total particle number in the hot cavity, we find a thermal energy of about $`10^{49}`$ erg, suitable for a remnant in which much of the remaining energy is tied up in the kinetic energy of expansion. The thermal pressure of the hot gas, according to $`P=nkT`$, is about 6 $`\times 10^{10}`$ dyne cm<sup>-2</sup>. These figures should, of course, only be regarded as rough estimates, as there is perhaps an order of magnitude uncertainty in the actual volume occupied by the hot gas, as well as additional uncertainties in the fitted temperature and emissivity.
The reason for the brightening of X-rays in the northeast section of the remnant is unclear. A search for timed emission revealed no significant peaks in the power spectrum; however, given the short exposure and high background for this observation, further investigation is indicated. The X-ray brightening occurs near an optical “knot” of bright emission, and may indicate a region where the SNR shock is encountering denser material.
In summary, we find that Chandra data are sufficient to distinguish SNR 0540$``$6944 from the emission of the nearby X-ray binary LMC X-1. While these findings are preliminary, as there remain uncertainties in the spatial and spectral responses, they are still informative. Given that this SNR has a lower luminosity than most LMC SNRs (eg Williams 1999), the results point out the capacity of Chandra to distinguish faint objects in confused regions. As observations continue, we may uncover an entire population of low-luminosity, older SNRs, in the LMC and even in more distant galaxies. This has the potential to substantially increase our estimates of SNR rates, which to date have been largely based on the fraction of SNRs that are more readily detected. SNR 0540$``$6944 also suggests the possibility of finding previously undetected SNRs within the crowded environs of H II regions, where we would indeed expect a high population of Type II SNRs.
The authors thank Eric Schlegel for his helpful comments as referee, as well as the CXC team for their work in making the data and analysis tools available. We acknowledge the support of the National Resarch Council (RMW) and NASA ADP Grant NAG 5-7003 (RMW & YHC).
|
warning/0004/hep-th0004092.html
|
ar5iv
|
text
|
# Mirror symmetry by O3-planes
## 1 Introduction
In , Intriligator and Seiberg found a new duality, the so-called “mirror symmetry”, between two different $`N=4`$ gauge theories in three dimensions. There exists such a mirror duality in three dimensions due to several special properties. First, the $`N=4`$ theory has a global R-symmetry $`SO(4)`$ which can be rewritten as $`SU(2)_L\times SU(2)_R`$, i.e., as the direct product of two independent $`SU(2)`$ factors. This is one crucial property for mirror symmetry because one action of the mirror duality is to simply interchange these two $`SU(2)`$ factors<sup>1</sup><sup>1</sup>1 When we discuss the mirror duality of $`N=2`$ theory in three dimensions, we must enhance the explicit $`U(1)`$ global R-symmetry to two $`U(1)`$’s, i.e., $`U(1)\times U(1)`$. Otherwise there is no good way to define the mirror theory. For details see .. Under the global R-symmetry, the vector multiplet is in the adjoint of $`SU(2)_L`$ and is invariant under $`SU(2)_R`$ while the hypermultiplet is in the adjoint of $`SU(2)_R`$ and is invariant under $`SU(2)_L`$ (notice that both multiplets have four scalars if we dualize the gauge field $`A_\mu `$ in three dimensions to a scalar). Furthermore, the mass parameter transforms as $`(3,1)`$ of $`SU(2)_L\times SU(2)_R`$ and the FI-parameter as $`(1,3)`$. So after mirror duality, the Coulomb branch and mass parameter of one theory change to the Higgs branch and FI-parameter of the other and vice versa. Such mapping has an immediate application: because the Higgs branch is not renormalized by quantum effects , we can get the exact result about the Coulomb branch of one theory which is corrected by quantum effects by studying the Higgs branch of the mirror theory which can be studied at the classical level. Because of this and other good applications of mirror duality (for details, see ), a lot of work has been done in this topic to try and find new mirror pairs.
There are several ways to construct the mirror pairs. The first way is to use the arguments coming from field theory . This method gives a lot of details how fields and parameters map to each other under the mirror duality. However, this method requires a lot of results which are not easy to get in field theory, so it is hard to use it to construct general mirror pairs. The second way is to use M-theory to construct the mirror pairs as done by Porrati and Zaffaroni in . The third way is to use the geometric realization in . The fourth way, which is also the most popular way in the construction of mirror pairs, is given in by using brane setups. The brane setup has the good property of making many quantities in field theory more visible. For example, the R-symmetry $`SU(2)_L\times SU(2)_R`$ corresponds to rotations in planes $`X^{345}`$ and $`X^{789}`$. The Coulomb branch and Higgs branch become the positions of D3-branes in NS-branes and D5-branes. The mass parameter and FI-parameter also have similar geometric correspondences. These geometric pictures give us some intuition to understand the problem better (for more applications of brane setups, see review ). The key observation in is that the mirror duality is just the S-duality in string theory. Using the known property of S-dual transformation of various kinds of branes we can easily find the mirror pairs. In this paper we will follow the last method.
Because we will use the brane setup to find the mirror theory, let us talk more about the general idea of the brane construction. Given a gauge theory with gauge group and some matter contents, first we try to find a proper brane setup which represents the gauge theory (usually it is the Coulomb branch given explicitly in the brane setup). After that, we move to the Higgs branch<sup>2</sup><sup>2</sup>2Usually, we can break all gauge symmetries by Higgs mechanism. However, in some cases after Higgsing there are still some massless gauge fields. We call the latter case “incomplete Higgsing of the theory by splitting the D3-branes between NS-branes and D5-branes. Then we make the S-duality transformation (mirror transformation) which changes the NS-brane to D5-brane, D5-brane to NS-brane and D3-brane to itself, while perform the electric-magnetic duality in the world volume theory of D3-branes. When the brane setup involves an orientifold or $`ON`$ plane, we need to know the S-duality rule for them too. Finally, we read out the corresponding gauge theory given by the S-dual brane setup—it is the mirror theory which we want to find.
In applications, it is straight forward to use the above procedure to give the mirror theory of $`U(n)`$ gauge theory with some flavors or the product of $`U(n)`$’s with some bifundamentals because the brane setup of those theories involve only NS5-branes, D5-branes and D3-branes and we know how to deal with them. However, when we try to find the mirror for a gauge group $`Sp(k)`$ or $`SO(n)`$, we must use an orientifold plane in the brane setup. Now a problem arises because sometimes we do not know how to read out the gauge theory of the S-dual brane setup of these orientifolds. The orientifolds which are involved in the construction can be divided into two types: the orientifold three plane (O3-plane) and the orientifold five plane (O5-plane). Sen has given an answer about the gauge theory under the $`ON`$-projection, which is the S-dual of the $`O5^{}`$ plane plus a physical D5-brane, in . Using this result, we can get the mirror theory for $`Sp(k)`$ by using the orientifold five plane in the initial brane setup. For $`SO(k)`$, if we insist on using the orientifold five plane again in the brane setup, we must know what is the gauge theory under the $`ON^+`$ projection which is the S-dual of $`O5^+`$ plane. It is still an open problem to read it out.
In the above paragraph, we mention that there is a difficulty to use orientifold five-plane to construct the mirror theory of $`SO(n)`$ gauge group. However, for constructing the $`Sp(k)`$ or $`SO(n)`$ gauge theory we can use an O3-plane instead of the O5-plane. Because under S-duality the O3-plane changes into another O3-plane, we know how to read out the gauge theory (unlike the O5-plane which becomes $`ON`$ plane under S-duality). Motivated by this observation, in this paper we use O3-planes to investigate the mirror theory of $`SO(n)`$ and $`Sp(k)`$ gauge groups. In particular, we get the mirror theory for $`SO(n)`$ gauge group which is a completely new result. Furthermore, our proposal for the construction of the mirror theory predicts a nontrivial strong coupling limit of field theories with eight supercharges.
The contents of the paper are as follows. In section 2, we discuss some basic facts on Op-planes which will set the stage for calculating the mirrors. These include the four kinds of O3-planes and the $`s`$-configuration involving 1/2NS-brane and 1/2D5-brane. In section 3 we discuss the splitting of physical D5-branes on O3-planes. It is a crucial ingredient in our construction of mirror theory. By S-duality, we get the rules for how a physical NS-brane can split into two 1/2NS-branes or conversely how two 1/2NS-branes can combine into a physical NS-brane. The latter predicts a nontrivial transition of strongly coupled field theories. After these preparations, we give the mirror theory of a single gauge group with some flavors: $`Sp(k)`$ in section 4, $`Sp^{}(k)`$ in section<sup>3</sup><sup>3</sup>3There are two ways to get $`Sp(k)`$ gauge group: by $`O3^+`$-plane or $`\stackrel{~}{O3^+}`$-plane. We denote the theory given by $`O3^+`$-plane as $`Sp(k)`$ and the theory given by $`\stackrel{~}{O3^+}`$-plane as $`Sp^{}(k)`$. 5, $`SO(2k)`$ in section 6 and $`SO(2k+1)`$ in section 7. In sections eight and nine we generalize the mirror construction to products of two gauge groups: $`Sp(k)\times SO(2m)`$ in section 8 and $`Sp^{}(k)\times SO(2m+1)`$ in section 9. Finally, we give conclusions in section 10.
## 2 Some facts concerning O3-planes
In this section, we summarize some facts about the O3-plane which will be useful for the mirror construction later.
### 2.1 The four kinds of O3-planes
There are four kinds of O3-planes which we will meet in this paper (for a more detailed discussion, see ): $`O3^+,O3^{},\stackrel{~}{O3^+},\stackrel{~}{O3^{}}`$. However, before entering the specific discussion of $`O3`$-planes let us start from general $`Op`$-planes. When $`p5`$, there exist four kinds of orientifolds $`Op^+,Op^{},\stackrel{~}{Op^+},\stackrel{~}{Op^{}}`$. Among these four we are very familiar with $`Op^+,Op^{},\stackrel{~}{Op^{}}`$. They can be described perturbatively as the fixed planes of the orientifold projection $`\mathrm{\Omega }`$ which acts on the world sheet as well as the Chan-Paton factors. By different choices of the action $`\mathrm{\Omega }`$ on the Chan-Paton factors we get two kinds of projections which we denote as $`\pm `$ projection. In the $`+`$ case, we can put only an even number of 1/2Dp-branes and the corresponding plane is the $`Op^+`$ plane. In the $``$ case, we can put an even or odd number of 1/2Dp-branes and the corresponding plane is $`Op^{}`$ for even number of 1/2Dp-branes and $`\stackrel{~}{Op^{}}`$ for odd number of 1/2Dp-branes. For $`\stackrel{~}{Op^{}}`$, because there is an odd number of 1/2Dp-branes, one 1/2Dp-brane must be stuck on the orientifold plane so that sometimes we consider the $`\stackrel{~}{Op^{}}`$ as the bound state of the $`Op^{}`$ and the 1/2Dp-brane (for more detailed discussion, the reader is referred to ). The $`\stackrel{~}{Op^+}`$ is more complicated and is discussed in detail by Witten in . In that paper, Witten observes $`O3`$-planes from a more unified point of view, namely discrete torsion (he deals with $`O3`$-planes. However the discussion can be easily generalized to other $`Op`$-planes). We can distinguish $`Op`$-planes by two $`Z_2`$ charges $`(b,c)`$ with the definition $`b=_{RP^2}B_{NS}`$ and $`c=_{RP^{5p}}C^{5p}`$ (the $`(b,c)`$ is defined under modular two and the discussion presented here comes from lecture already given by one of the authors at ITP, Santa Barbara; see also ). The second charge $`c`$ exists only for $`p5`$. For $`p>5`$, it can not be defined and we are left only with two types of $`Op`$-planes (it is a little mysterious that $`\stackrel{~}{Op^{}}`$ does not exist for $`p>5`$, some arguments can be found in ). We summarize the properties of these four $`Op`$-planes according the discrete torsions $`(b,c)`$ in Table 1 (where S-duality is applied only to $`p=3`$).
These four kinds of $`O`$-planes are not unrelated to each other and in fact change to each other when they pass through the 1/2NS-brane or 1/2D-brane . The change is shown in Figure 1: when $`Op^{}`$($`\stackrel{~}{Op^{}}`$) passes through the 1/2NS-brane, it changes to $`Op^+`$($`\stackrel{~}{Op^+}`$) and vice versa; when $`Op^{}`$($`Op^+`$) passes through the 1/2D(p+2) -brane, it changes to $`\stackrel{~}{Op^{}}`$($`\stackrel{~}{Op^+}`$) and vice versa.
After the discussion of general $`Op`$-planes, we focus on $`O3`$-planes which will be used throughout this paper. For $`O3`$-planes, the charge of $`O3^{}`$ is $`1/4`$ while the charges of $`O3^+,\stackrel{~}{O3^{}},\stackrel{~}{O3^+}`$ are $`1/4`$. The fact that the charges for the latter three $`O3`$-planes are identical is not a coincidence and they are related to each other by the $`SL(2,Z)`$ duality symmetry in Type IIB. In particular, under S-duality $`O3^+`$ and $`\stackrel{~}{O3^{}}`$ transform to each other while $`\stackrel{~}{O3^+}`$ transforms to itself. $`O3^{}`$ transforms to itself also under S-duality because it is the only O3-plane with $`1/4`$ charge. One immediate application of the above S-duality property is that the change of O3-planes crossing the 1/2NS-brane is exactly S-dual to the change of O3-planes crossing the 1/2D5-brane. So our rule is consistent. The above discussions will be useful later in the study of mirror symmetry.
### 2.2 The supersymmetric configuration
In the procedures involved in mirror transformations, we need to break the D3-branes between the NS-brane and D5-brane to avoid the so called $`s`$-rule . Furthermore, to read out the mirror theory from the brane setup it is convenient to move a 1/2NS-brane along the $`X^6`$ direction (our notations and conventions for the brane setups for all kinds of branes is given in the caption of Figure 1.) to pass through the 1/2D5-brane such that the D3-branes ending on the 1/2NS-branes are annihilated in order to keep the linking number between 1/2NS-brane and 1/2D5-brane invariant. All these actions require the understanding of supersymmetric configurations in the presence of O3-planes. We summarize these results in this subsection. The tool in our discussion of $`s`$-configuration is still the conservation of linking number between 1/2NS-brane and 1/2D5-brane. The formula of linking number for 1/2NS-brane and 1/2D5-brane is
$$\begin{array}{ccc}L_{NS}\hfill & =\hfill & \frac{1}{2}(R_{D5}L_{D5})+(L_{D3}R_{D3})\hfill \\ L_{D5}\hfill & =\hfill & \frac{1}{2}(R_{NS}L_{NS})+(L_{D3}R_{D3})\hfill \end{array}$$
(1)
where $`R_{D5}(L_{D5})`$ is the D5-charge to the right (left) of NS-brane (1/2D5-brane has 1/2 charge) and similar definition to others. Because we have four kinds of O3-planes we will have four kinds of supersymmetric configurations including one 1/2-NS brane and one 1/2-D5 brane. These four different cases are:
$$\begin{array}{cccc}(1)\hfill & O3^{}\hfill & (1/2D51/2NS)or(1/2NS1/2D5)\hfill & \stackrel{~}{O3^+},\hfill \\ (2)\hfill & O3^+\hfill & (1/2D51/2NS)or(1/2NS1/2D5)\hfill & \stackrel{~}{O3^{}},\hfill \\ (3)\hfill & \stackrel{~}{O3^{}}\hfill & (1/2D51/2NS)or(1/2NS1/2D5)\hfill & O3^+,\hfill \\ (4)\hfill & \stackrel{~}{O3^+}\hfill & (1/2D51/2NS)or(1/2NS1/2D5)\hfill & O3^{},\hfill \end{array}$$
(2)
where the configuration $`O3^{}(1/2D51/2NS)or(1/2NS1/2D5)\stackrel{~}{O3^+}`$ means that the $`O3^{}`$ plane is at the left, $`\stackrel{~}{O3^+}`$ at the right. In the middle we put 1/2NS-brane and 1/2D5-brane according to the order 1/2D5-1/2NS from left to right (see part (a) of Figure 2) or 1/2NS-1/2D5 (see part (b) of Figure 2).
The general pattern for the above four supersymmetric configurations is shown in Figure 2, where we assume the number of connected D3-branes (in physical units) from the left 1/2D5-brane to the right 1/2NS-brane is $`N`$ and from the left 1/2NS-brane to the right 1/2 D5-brane is $`\stackrel{~}{N}`$. So a configuration to be supersymmetric is equivalent to the solution of $`N,\stackrel{~}{N}0`$ such that they conserve the linking number after crossing.
#### 2.2.1 The first case: $`O3^{}\stackrel{~}{O3^+}`$
In this case we start from the brane setup (a) of Figure 2 with $`O3^{}`$ plane at the left, $`\stackrel{~}{O3^{}}`$ plane in the middle and $`\stackrel{~}{O3^+}`$ plane at the right. The linking numbers are $`L_{1/2D5}=\frac{1}{2}(\frac{1}{2}0)+[(\frac{1}{4})(\frac{1}{4}+N)]=N\frac{1}{4}`$ and $`L_{1/2NS}=\frac{1}{2}(0\frac{1}{2})+[(\frac{1}{4}+N)(\frac{1}{4})]=N\frac{1}{4}`$. Now we move the 1/2D5 along $`X^6`$ direction to pass through 1/2NS and get the (b) of Figure 2 with $`O3^{}`$ plane at the left, $`O3^+`$ plane in the middle and $`\stackrel{~}{O3^+}`$ plane at the right. For the latter we have linking numbers as $`L_{1/2D5}=\frac{1}{2}(0\frac{1}{2})+[(\stackrel{~}{N}+\frac{1}{4})(\frac{1}{4})]=\stackrel{~}{N}\frac{1}{4}`$ and $`L_{1/2NS}=\frac{1}{2}(\frac{1}{2}0)+[(\frac{1}{4})(\stackrel{~}{N}+\frac{1}{4})]=\stackrel{~}{N}\frac{1}{4}.`$ Comparing these two linking numbers we get
$$N=\stackrel{~}{N}$$
(3)
It is a highly constraining equation. For the supersymmetric configuration, the only solution is $`N=\stackrel{~}{N}=0`$. This means that when we break the D3-brane to go to the Higgs branch, we can not put D3-brane between 1/2NS-brane and 1/2D5-brane in this orientifold configuration.
#### 2.2.2 The second case: $`O3^+\stackrel{~}{O3^{}}`$
Starting from brane setup (a) of Figure 2 with $`O3^+`$ at the left, $`\stackrel{~}{O3^+}`$ in the middle and $`\stackrel{~}{O3^{}}`$ at the right, we find the linking numbers as $`L_{1/2D5}=\frac{1}{2}(\frac{1}{2}0)+[(\frac{1}{4})(\frac{1}{4}+N)]=N+\frac{1}{4}`$ and $`L_{1/2NS}=\frac{1}{2}(0\frac{1}{2})+[(\frac{1}{4}+N)(\frac{1}{4})]=N\frac{1}{4}`$. Again by moving the 1/2D5-brane to pass through 1/2NS-brane we get the brane setup as (b) with the middle O3-plane changed from $`\stackrel{~}{O3^+}`$ in (a) to $`O3^{}`$ in (b) (the left and right O3-plane are invariant under the motion). The linking numbers for the latter are $`L_{1/2D5}=\frac{1}{2}(0\frac{1}{2})+[(\stackrel{~}{N}\frac{1}{4})(\frac{1}{4})]=\stackrel{~}{N}\frac{3}{4}`$ and $`L_{1/2NS}=\frac{1}{2}(\frac{1}{2}0)+[(\frac{1}{4})(\stackrel{~}{N}\frac{1}{4})]=\stackrel{~}{N}+\frac{3}{4}`$. From these relations we find the equation
$$N+1=\stackrel{~}{N}.$$
(4)
So for a consistent supersymmetric configuration there are three solutions: $`(N,\stackrel{~}{N})=(0,1);(\frac{1}{2},\frac{1}{2});(1,0)`$.
#### 2.2.3 The third case: $`\stackrel{~}{O3^{}}O3^+`$
For the third case, we start from the brane setup (a) with $`\stackrel{~}{O3^{}}`$ at the left, $`O3^{}`$ in the middle and $`O3^+`$ at the right. The linking numbers are $`L_{1/2D5}=\frac{1}{2}(\frac{1}{2}0)+[(\frac{1}{4})(\frac{1}{4}+N)]=N+\frac{3}{4}`$ and $`L_{1/2NS}=\frac{1}{2}(0\frac{1}{2})+[(\frac{1}{4}+N)(\frac{1}{4})]=N\frac{3}{4}`$. Now we move the 1/2D5-brane to pass through 1/2NS-brane and get the brane setup (b) with $`\stackrel{~}{O3^+}`$ in the middle. The linking numbers become $`L_{1/2D5}=\frac{1}{2}(0\frac{1}{2})+[(\stackrel{~}{N}+\frac{1}{4})(\frac{1}{4})]=\stackrel{~}{N}\frac{1}{4}`$ and $`L_{1/2NS}=\frac{1}{2}(\frac{1}{2}0)+[(\frac{1}{4})(\stackrel{~}{N}+\frac{1}{4})]=\stackrel{~}{N}+\frac{1}{4}`$. By comparing these relations we have
$$N+1=\stackrel{~}{N}.$$
(5)
So again there are three solutions: $`(N,\stackrel{~}{N})=(0,1);(\frac{1}{2},\frac{1}{2});(1,0)`$.
#### 2.2.4 The fourth case: $`\stackrel{~}{O3^+}O3^{}`$
For the last case we start from the brane setup (a) with $`\stackrel{~}{O3^+}`$ at the left, $`O3^+`$ in the middle and $`O3^{}`$ at the right. The linking numbers are $`L_{1/2D5}=\frac{1}{2}(\frac{1}{2}0)+[(\frac{1}{4})(\frac{1}{4}+N)]=N+\frac{1}{4}`$ and $`L_{1/2NS}=\frac{1}{2}(0\frac{1}{2})+[(\frac{1}{4}+N)(\frac{1}{4})]=N+\frac{1}{4}`$. Now we move the 1/2D5-brane to pass through 1/2NS-brane and get the brane setup (b) with $`\stackrel{~}{O3^{}}`$ in the middle. The linking numbers change to $`L_{1/2D5}=\frac{1}{2}(0\frac{1}{2})+[(\stackrel{~}{N}+\frac{1}{4})(\frac{1}{4})]=\stackrel{~}{N}+\frac{1}{4}`$ and $`L_{1/2NS}=\frac{1}{2}(\frac{1}{2}0)+[(\frac{1}{4})(\stackrel{~}{N}+\frac{1}{4})]=\stackrel{~}{N}+\frac{1}{4}`$. From these relations we have
$$N=\stackrel{~}{N}.$$
(6)
The only solution is $`(N,\stackrel{~}{N})=(0,0)`$ as in the first case.
Let us summarize the results in the last four subsections. When the charge of $`O3`$-planes at the two sides are the same (case two and case three), the condition is $`N+\stackrel{~}{N}=1`$, so there is annihilation or creation of D3-branes in crossing. When the charge of $`O3`$-planes at the two sides are different (case one and case four), the condition is $`N=\stackrel{~}{N}=0`$, so there can not be any D3-branes between the 1/2NS-brane and 1/2D5-brane.
## 3 The splitting of the physical brane
To construct the mirror theory by brane setups, we can follow the procedure given in the introduction . However, in the presence of the O3-plane, we need one new input: how to split the physical D5-brane into two 1/2D5-branes on the O3-plane. Initially, the physical D5-brane can be placed off the O3-plane in pairs of 1/2D5-branes (see Figure 3). We can move the pair of 1/2D5-branes to touch the O3-plane. After touching the O3-plane, in principle every 1/2D5-brane can move freely on the O3-plane. We call such an independent motion of the 1/2D5-brane as “splitting” of the physical D5-brane. We want to emphasize that the splitting of a physical D5-brane into two 1/2D5-branes is a nontrivial dynamical process in string theory and can be applied to many situations. Here we need the splitting because in the mirror theory, the gauge theory is given by D3-branes ending on 1/2NS-branes which are the S-dual of 1/2D5-branes in the original theory. In this paper, we give only a preliminary discussion. We found some novel results: sometimes there is a creation of one physical D3-brane between these two 1/2D5-branes; sometimes there is an annihilation and sometimes, no creation and no annihilation. We found these results by matching the Higgs branch moduli of the $`Sp`$ or $`SO`$ theory with the correct dimension of moduli space.
### 3.1 The splitting of D5-branes without ending D3-branes
Before going to the general situation let us discuss the splitting of D5-branes which do not have any D3-branes ending on them. First we discuss the case where there is only one physical D5-brane and O3-plane (see Figure 3). Before splitting, every 1/2D5-brane has linking number zero. After splitting, there can be $`N`$ physical D3-branes between these two 1/2D5-branes (to keep supersymmetry, there can not be anti-D3-branes between them; furthermore, because here we do not have any D3-branes initially, there can not be annihilation either). Let us calculate the linking number after splitting:
$$\begin{array}{ccc}O3beforesplitting& \mathrm{\Delta }L_{left}& \mathrm{\Delta }L_{right}\\ O3^+& N& N\\ \stackrel{~}{O3^+}& N& N\\ O3^{}& \frac{1}{2}N& \frac{1}{2}+N\\ \stackrel{~}{O3^{}}& \frac{1}{2}N& \frac{1}{2}+N\end{array}$$
(7)
In BPS states, we have the tension of D5-branes proportional to their charge (linking number). To have the minimum tension configuration, it is natural to have $`N=0`$ for the first three cases. However, for the last case, $`N=0`$ and $`N=1`$ are equally favorable just from the point view of tension. We will fix the ambiguity in the next paragraph. However, before we end this paragraph, we want to emphasize that no matter what case it is, the total change in linking number is always
$$\mathrm{\Delta }L=0or\mathrm{\Delta }L=\pm \frac{1}{2}.$$
We can fix the ambiguity for the last case by considering Higgsing. Starting from the $`SO(3)`$ gauge theory with one flavor, we can Higgs it to $`SO(2)`$ with one singlet (there are $`31=2`$ gauge fields which get mass, so we leave only $`3\times 12=1`$ singlet). In part (a) of Figure 4 we assume $`N=0`$ in the splitting process and go to the Higgs branch. By moving 1/2D5-branes outside we find the final theory is $`SO(2)`$ without singlets in part (b). This means that our assumption is wrong. Choosing the other assumption $`N=1`$ in part (c), by moving 1/2D5-branes outside we get the final theory is $`SO(2)`$ with a singlet in part (d) which is exactly what we expect from the field theory. This shows that, for matching the correct moduli dimensions of Higgs branch, in last case of (7) there should be a D3-brane created in the splitting.
The discussion of the splitting of physical D5-branes becomes more complex if there are more than one D5-brane to be split. The complexity manifests in the last two cases in (7) because in these cases there is a change of linking number ($`\mathrm{\Delta }L=\pm \frac{1}{2}`$) for every 1/2D5-brane. Before splitting, we have, for example, $`2n`$ 1/2D5-branes with linking number zero. After splitting, we have $`n`$ 1/2D5-branes with linking number $`\frac{1}{2}`$ and $`n`$, with linking number $`\frac{1}{2}`$. The different order of linking number gives different physical content, i.e., the order determines when there should be D3-branes created and when there are no D3-branes created.
To illustrate our idea, let us see Figure 5. After splitting one D5-brane according to the analysis in the last paragraph, we continue to split the second D5-brane. However, in this case, we have two choices. In the first choice, the second D5-brane is far away from the first D5-brane in $`X^6`$ direction like part (a). So locally the splitting should be the same as the first D5-brane and we get part (b). Notice that the order of linking number of 1/2D5-branes is $`\frac{1}{2},+\frac{1}{2},\frac{1}{2},+\frac{1}{2}`$. In the second choice, the second D5-brane is in the middle of the pair of first 1/2D5-branes as part (c). Naively, the second D5-brane will see the $`O3^{}`$-plane (in fact, D5-brane will see more) so the splitting looks like to go as part (d) with the order of linking number $`\frac{1}{2},\frac{1}{2},+\frac{1}{2},+\frac{1}{2}`$. However, part (d) is not consistent with the Higgs branch of $`SO(3)`$ with two flavors. Furthermore, because there are eight supercharges, the different positions of D5-branes should not effect the physics. So we argue that from part (c) we should get part (b) too. In part (c), the second D5-brane sees not only the $`O3^{}`$-plane, but also the one created D3-branes, 1/2D5-branes at left with $`\mathrm{\Delta }=\frac{1}{2}`$ and 1/2D5-branes at right with $`\mathrm{\Delta }=+\frac{1}{2}`$. This more complete information determines that the second D5-brane will split to part (b).
From the above observation, we propose that the correct order of linking number should be $`\frac{1}{2},+\frac{1}{2},\frac{1}{2},+\frac{1}{2},\mathrm{}.,\frac{1}{2},+\frac{1}{2}`$ (notice the alternating fashion of $`\frac{1}{2}`$ and $`+\frac{1}{2}`$). We make such a suggestion because it is the only correct order which can produce the consistent Higgs pattern for $`SO(K)`$ gauge group with $`N`$ flavors. It will be very interesting if we can derive such a rule from string theory. Furthermore, this proposal will give very interesting predictions which we will discuss later.
Let us pause a moment to summarize the results we have obtained above. Without the D3-brane ending on D5-branes, (1) the change of linking number of 1/2D5-branes is $`\mathrm{\Delta }L=0`$ for $`O3^+,\stackrel{~}{O3^+}`$ and $`\mathrm{\Delta }L=\pm \frac{1}{2}`$ for $`O3^{},\stackrel{~}{O3^{}}`$; (2) for the splitting of a bunch of D5-branes, the order of linking number is $`\frac{1}{2},+\frac{1}{2},\frac{1}{2},+\frac{1}{2},\mathrm{}.,\frac{1}{2},+\frac{1}{2}`$.
### 3.2 The splitting of D5-branes with ending D3-branes
After the discussion of the splitting of D5-branes without D3-branes ending on them, we consider the case that there are $`N`$ D3-branes ending on them. The results for this latter case can be derived from the results in the last subsection. For example, let us discuss the case of one D5-brane with one ending D3-brane in Figure 6. We can add one 1/2NS-brane such that the D3-brane ending on it as part (a). Then we can move D5-brane to the right of 1/2NS-brane and annihilate the D3-brane as part (b). Now the part (b) is the case we discussed in the last subsection. We can split the physical D5-brane and move two 1/2D5-branes to left of 1/2NS-brane by using the result in section 2. By this loop, we finally get the splitting of D5-brane with one ending D3-brane. For more D3-branes ending on D5-branes we can add more 1/2NS-brane and repeat the above procedure.
Although the above trick solves our problem completely, it is too tedious and we need a more direct way to see it. Notice that the change of the linking number of 1/2D5-branes happens only at splitting. So we can use the changing of linking number as the rule to determine the splitting of D5-brane. In general there will be $`N_L`$ D3-branes ending on D5-brane from the left and $`N_R`$ D3-branes, from right. The rule depends only on the absolute difference between $`N_L,N_R`$, i.e., $`N=|N_LN_R|`$. We summarize the rule in Table 2.
### 3.3 The splitting of NS-branes and novel predictions of field theory in the strong coupling limit
Making S-duality, we can get the rules of splitting physical NS-branes into 1/2NS-branes on O3-plane as Table 3.
From Table 3, we get two predictions of $`N=4`$ three dimensional field theory in the strong coupling limit (see Figure 7). In the first case (part (a) of Figure 7), the field theory is $`SO(2k)\times Sp(k)\times SO(2k)`$ with two bifundamentals. From the brane setup in part (a), we see that, by reversing the process of the splitting of the NS-brane, we can move two middle 1/2NS-branes to meet together and leave $`O3^{}`$-plane. In field theory, moving two middle 1/2NS-branes together corresponds to the strong coupling limit of $`Sp(2k)`$ gauge theory, and moving NS-brane off the $`O3^{}`$-plane corresponds to turning on “FI-parameters” <sup>4</sup><sup>4</sup>4In fact, it is a hidden “FI-parameters”. We will discuss it more in section 4.3. So our brane configuration predicts that, at the strong coupling limit of $`Sp(2k)`$ and the turning of FI-parameter, the original theory $`SO(2k)\times Sp(k)\times SO(2k)`$ with two bifundamentals will flow to $`SO(2k)`$ without any flavor. The second case is given in part (b) of Figure 7. By the similar arguments, we predict that at the strong coupling limit of $`SO(2k+2)`$ and the turning of FI-parameter, the field theory $`Sp(k)\times SO(2k+2)\times Sp(k)`$ with two bifundamentals will flow to $`Sp(k)`$ without any flavor. It will be interesting to check these two predictions from the field theory point of view.
## 4 The mirror of $`Sp(k)`$ gauge theory
Now we start to construct mirror pairs using the above knowledge. First let us discuss $`Sp(k)`$ gauge theory with $`N`$ fundamental flavors. In this case, the brane setup is as follows: we put $`2k`$ 1/2D3-branes, i.e., branes and their images under the $`O`$-plane (extended in $`X^{0126}`$) ending on two 1/2NS branes (extended in $`X^{012345}`$) along $`X^6`$ while $`2N`$ 1/2D5-branes (extended in $`X^{012789}`$) are put in the middle (see (a) of figure 8). Then $`O3`$-planes from the left to right read as $`O3^{},O3^+,O3^{}`$. This $`O3`$-plane configuration reminds us of the special $`s`$-configuration discussed in the last section. Because in the presence of $`O3`$-planes the $`s`$-configuration is a little different from the known $`s`$-rule in , we will demonstrate the detailed steps for the mirror construction for $`Sp(1)`$ with three flavors. Thereafter we quickly go to the general $`Sp(k)`$ case.
### 4.1 $`Sp(1)`$ with the $`3`$ flavors
For the $`Sp(1)`$ gauge theory with $`3`$ flavors we have the following information about the moduli space of the Higgs branch and the Coulomb branch as well as the FI-parameters and mass parameters:
$$\begin{array}{ccc}d_v\hfill & =\hfill & 1,\hfill \\ d_H\hfill & =\hfill & 3\times 23=3,\hfill \\ \mathrm{\#}m\hfill & =\hfill & 3,\hfill \\ \mathrm{\#}\zeta \hfill & =\hfill & 0\hfill \end{array}$$
(8)
After the mirror map, we should have a mirror theory which has $`d_v=3,d_H=1,\mathrm{\#}\zeta =3,\mathrm{\#}m=0`$, i.e. the Coulomb branch and the Higgs branch are interchanged while the mass parameters and FI-parameters are exchanged . However, when we count these parameters, sometimes we meet nontrivial situations, such as the “hidden FI-term” explained in . We will see later that these “hidden parameters” arise in our construction and will discuss them in more detail later.
The details of the mirror construction are given in Figure 8. Let us go step by step. Part (a) is just the brane setup for $`Sp(1)`$ with three fundamental flavors. By moving the physical D5-brane to touch the orientifold $`O3^+`$ plane, i.e., setting the masses to zero, we can split them into 1/2D5-branes as in part (b). Now we go to the Higgs branch by splitting the D3-branes between those 1/2NS-branes and 1/2D5-branes. However, from (3) and (6), we must split these D3-branes as given by part (c). The crucial point is that there is no D3-brane connected between the 1/2NS-brane and its nearest 1/2D5-brane because it is prohibited by the supersymmetric configuration discussed in section 2. Now we can use the rules (3) and (6) to move the left 1/2NS-brane crossing the neighboring right 1/2D5-brane and the right 1/2NS-brane crossing the neighboring left 1/2D5-brane. The result is given by part (d). Notice that in such a process, no D3-brane is created or annihilated. Applying (4) and (5) to move the 1/2NS-brane across 1/2D5-brane, we reach part (e). In this process, the physical D3-brane which connects the 1/2NS-brane and 1/2D5-brane is annihilated. Now we can apply the mirror transformation to give the result shown in part (f). However, it is a little hard to read out the final gauge theory because of the $`O3^{}`$ and $`\stackrel{~}{O3^{}}`$ projections in the same interval. We can get rid of this ambiguity by applying (3) and (6) again to reach the result in part (g).
Now we have the brane setup for the mirror theory in part (g) of figure 8. We can read out the theory directly from the brane setup according the standard rule: For $`2k`$ 1/2D3-brane stretching between two 1/2NS-branes with $`O3^{},\stackrel{~}{O3^{}},O3^+,\stackrel{~}{O3^+}`$ planes we get $`SO(2k),SO(2k+1),Sp(k),Sp^{}(k)`$ gauge groups respectively. For one 1/2D5-brane between two 1/2NS-branes it contributes one fundamental half-hypermultiplet for that gauge group. For one physical D5-brane between two 1/2NS-branes it contributes one fundamental hypermultiplet for the gauge group. For two gauge groups which have a common 1/2NS-brane there is a bifundamental (in the presence of $`O3`$-plane, such bifundamental is, more exactly, half-hypermultiplet). Applying the above rules we immediately get the mirror theory as $`SO(2)\times Sp(1)\times SO(2)`$ with two bi-fundamentals and one fundamental for $`Sp(1)`$. Here we want to emphasize that in general we get only half fundamental hypermultiplets coming from the 1/2D5-brane. The unusual point for this explicit example is that the two 1/2D5-branes are in the same interval such that they can combine together and leave the orientifold (see section 3). Now let us calculate the moduli spaces and parameters to see if they are really mirror to each other. For the mirror theory in the part(h) of Figure 8, it is easy to get the dimensions of moduli spaces as $`d_v=1+1+1=3`$ and $`d_H=(2\times 2+2\times 2)/2+1\times 2(1+1+3)=65=1`$, so we see the results match when comparing to 8. However, when we turn to calculate the mass parameters and FI-parameters, a mismatch occurs. In the mirror theory, we have two bifundamentals and one fundamental. For the two bifundamentals we do not know how to turn the mass parameters so we get the $`\mathrm{\#}m=1`$. Because there are no $`U(1)`$ factors in the mirror theory, it seems that we should get $`\mathrm{\#}\zeta =0`$. Now comparing with the original theory, we find a mismatch in the mass parameters and FI-parameters. The solution of the above mismatch is given by the concept of “hidden FI-term” which we will discuss later .
### 4.2 Another method to go to the Higgs branch
In the above procedure, we split D5-branes first, then went to the Higgs branch by splitting the D3-branes. However, we can go to the Higgs branch in another way by splitting the D3-brane first on the physical D5-brane and then splitting the D5-brane on the O3-plane. The procedure of this second method is drawn in Figure 9. In part (a) , we keep the D5-branes off the O3-plane and split the D3-branes to go the Higgs branch (such splitting is very familiar to us already, see ). By moving the physical D5-brane to cross the 1/2NS-brane, we can get rid of the D3-brane ending one D5-brane and 1/2NS-brane. The result is shown in part (b). Now we move the D5-brane to the O3-plane and split them. For consistency with the first method in the last subsection we must require the splitting of D5-brane with one D3-brane ending on it as the rule given in section 3.2. In fact, as we discussed above, we find all rules in section 3.2 in this way. It is easy to check that in this example we should get the same result as part (e) of Figure 8.
### 4.3 The “hidden FI-term”
We have met the mismatch of mass and FI parameters in the above mirror pair. It is time for us to talk more about it in this subsection. In fact, such a mismatch of mass and FI parameters in mirror pair is not new to us. Kapustin found this problem in . In that paper, he considers the mirror of $`Sp(k)`$ with an antisymmetric tensor and $`n`$ fundamental flavors. He found that when $`n=2,3`$ the quivers of the mirror theory are in fact affine $`A_1`$ for $`n=2`$ and affine $`A_3`$ for $`n=3`$. However, it is a well-known fact that a gauge theory given by an affine $`A_n`$ quiver has one mass parameter. On the other hand, classically the original $`Sp(k)`$ theory does not have any FI parameters. Kapustin suggests the concept of “hidden FI term” to resolve the conflict. Such a term arises as the deformation in the infrared limit and has the same quantum number as a FI-term. Because it is a quantum effect, these deformations need not have a Lagrangian description in the ultraviolet. To count the number of hidden FI deformation we simply count the mass parameters in the mirror theory. Now applying Kapustin’s explanation to our example, we find there is one “hidden FI-term” for the original theory and three “hidden FI-terms” for the mirror theory. This result is consistent with Kapustin’s result. Notice that for $`k=1`$ the antisymmetric tensor of $`Sp(k)`$ does not exist, so his theory is in exact agreement with our original theory and we both find one “hidden FI-term”. Hereafter we do not discuss the matching of the mass and FI parameters anymore, but we will mention the case when there exists a “hidden FI-term” for the original theory.
The appearance of the “hidden FI term” indicates another important aspect of the possible enhanced hidden global symmetry. In , the authors observed that the fixed point can have global symmetries which are manifest in one description but hidden in another (i.e., can be seen only quantum mechanically). For example, the $`U(1)`$ with two flavors is a self-mirror theory. On a classical level we have $`SU(2)\times U(1)`$ global symmetry, where $`SU(2)`$ is the flavor symmetry and $`U(1)`$ is the global symmetry connecting one FI-parameter (FI-parameter can be considered as a component in the background vector supermultiplet of $`U(1)`$). However, at the fixed point, the $`U(1)`$ global symmetry is enhanced to $`SU(2)`$. This enhanced symmetry can be easily seen in the brane setup of the mirror theory because in this special case ($`U(1)`$ with two flavors), the two D5-branes (the S-dual of two NS-branes in original symmetry) meet in same interval. This is another advantage of brane setup because we can see a lot of nontrivial phenomena pictorially. In later sections, when we find the case where there is a “hidden FI term”, we will also discuss the enhanced global symmetry.
There is another interesting aspect which is worth mentioning. If our construction is right, it seems that we have two different theories which are mirror to the same one because in we can construct the mirror of $`Sp(k)`$ gauge theory by using the $`O5^{}`$ plane. This is also met by Kapustin in . He noticed that two theories, (1) the $`Sp(k)`$ gauge theory with an antisymmetric tensor plus two or three fundamental and (2) the $`U(k)`$ gauge theory with an adjoint plus two or four fundamental flavors, are mirror to the same affine $`A_1`$ or $`A_3`$ quiver theory. Because mirror symmetry is a property in the infrared limit of gauge theory, such a non-uniqueness is allowed. Actually the brane picture provides a definition of the theory beyond the infrared limit and the non-uniqueness can be seen in nature by having two different brane representations of the same field theory.
### 4.4 $`Sp(2)`$ with $`6`$ flavors
With the experience of $`Sp(1)`$ gauge theory, we can deal with the $`Sp(2)`$ with $`6`$ flavors very quickly. The moduli spaces for the original theory have $`d_v=2`$ and $`d_H=6\times 410=14`$ (as mentioned above, in the following discussion we do not discuss the issue of mass parameters and FI-parameters). The steps for getting the mirror theory is in Figure 10. From the brane setup (d) in Figure 10 we read out that the mirror theory as $`SO(2)\times Sp(1)\times SO(4)\times Sp(2)\times SO(5)\times Sp(2)\times SO(4)\times Sp(1)\times SO(2)`$ with $`8`$ bifundamentals and two fundamental half-hypermultiplets one for each $`Sp(2)`$ gauge theory. By an easy calculation, we can check the moduli spaces as: $`d_v=4\times 1+5\times 2=14`$ and $`d_H=(2\times 4+2\times 8+2\times 16+2\times 20+2\times 4)/2(2+2\times 3+2\times 6+3\times 10)=5250=2`$.
### 4.5 The general case
Now we discuss the general case, i.e., $`Sp(k)`$ with $`N`$ fundamental flavors (to get the complete Higgsing, we have to assume that $`N2k`$). The moduli space has $`d_v=k`$ and $`d_H=2kNk(2k+1)`$. The steps for getting the mirror theory are shown in Figure 11.
From it we can read out that the mirror theory are $`SO(2)\times Sp(1)\times SO(4)\times Sp(2)\mathrm{}\times Sp(k1)\times SO(2k)\times (Sp(k)\times SO(2k+1))^{n2k1}\times Sp(k)\times SO(2k)\times Sp(k1)\mathrm{}\times Sp(1)\times SO(2)`$ with bifundamentals and one fundamental half-hypermultiplet for each the first and the last $`Sp(k)`$ gauge groups. For clarity, the corresponding quiver diagram of the above mirror theory is also drawn in part (c) of this figure. Now we can calculate the moduli spaces of the mirror as
$$\begin{array}{ccc}d_v\hfill & =\hfill & 4_{n=1}^{k1}n+(2N4k+1)k=2Nkk(2k+1),\hfill \\ d_H\hfill & =\hfill & [\frac{1}{2}\times 2_{n=1}^{k1}((2n)^2+2n(2n+2))+\frac{1}{2}(2(2k)^2+(2N4k2)2k(2k+1))+2k]\hfill \\ & \hfill & [2_{n=1}^{k1}(n(2n1)+n(2n+1))+(2N4k1)k(2k+1)+2k(2k1)]\hfill \\ & =\hfill & k.\hfill \end{array}$$
(9)
As mentioned above, for general $`N,k`$ we get only half-hypermultiplets coming from the 1/2D5-branes in the mirror theory. However, there are two degenerate cases where one fundamental hypermultiplet does exist instead of two half-hypermultiplets. The first case is when $`N=2k`$. In this case, we do not need to move the 1/2NS brane further from part (a) to part (b) in Figure 11. Instead, we can make the mirror transformation directly from part(a). In the mirror theory, we get only one $`SO(2k)`$ gauge group but with one flavor for this $`SO(2k)`$. As explained above, such a flavor hints a “hidden FI-term” in the original theory. The second case is when $`N=2k+1`$, where we get only one $`Sp(k)`$ gauge group in mirror theory, but also with one flavor of the $`Sp(k)`$ which also suggests a “hidden FI-term” in the original theory. For $`k=1`$, the two cases where a “hidden FI-term” shows is given in . For $`k2`$ it is a new result.
As we mentioned in section 4.3, in the case where a “hidden FI-term” shows we should consider the possible enhancement of global symmetry. In general the theory has global $`SO(2N)`$ flavor symmetry. When $`N=2k`$, the global symmetry will be enhanced to $`SO(2N)\times Sp(1)`$. The factor $`Sp(1)`$ can be seen from the mirror theory, where two 1/2D5-branes meet and give one flavor to the $`SO(2k)`$ gauge group (notice the flavor symmetry for $`Sp(k)`$ gauge groups is $`SO(2N)`$, for $`Sp^{}(k)`$ gauge groups, $`SO(2N+1)`$, for $`SO(2k)`$ gauge groups, $`Sp(N)`$ and for $`SO(2k+1)`$ gauge groups, $`Sp^{}(N)`$). When $`N=2k+1`$, the global symmetry will be enhanced to $`SO(2N)\times SO(2)`$ because in this case, the one extra flavor in mirror theory belongs to the $`Sp(k)`$ gauge group.
## 5 The mirror of $`Sp^{}(k)`$ gauge theory
We know that the $`O3^+`$ and $`\stackrel{~}{O3^+}`$ projections both give $`Sp(k)`$ gauge theory. To distinguish them, we denote the gauge theory given by $`O3^+`$ projection as $`Sp(k)`$ and that by $`\stackrel{~}{O3^+}`$ as $`Sp^{}(k)`$. After the discussion of the mirror of $`Sp(k)`$ gauge group in the last section, we now address the $`Sp^{}(k)`$ case in this section. The brane setup of $`Sp^{}`$ is just to replace the $`O3^\pm `$ in $`Sp(k)`$ by $`\stackrel{~}{O3^\pm }`$ (for example, see figure 12). However, by such a replacement, the theory becomes $`Sp(k)`$ gauge theory with $`n`$ flavors plus two half-hypermultiplets contributed from the $`\stackrel{~}{O3^{}}`$ at the two sides (notice that $`\stackrel{~}{O3^{}}`$ can be considered as $`O3^{}`$ plus a 1/2D3-brane). We will start the discussion also from a simple example, then go to the general case. Furthermore, we will compare the mirror of $`Sp(k)`$ and $`Sp^{}(k)`$ and show that in fact they give the same mirror theory.
### 5.1 $`Sp^{}(1)`$ with 3 fundamental flavors
In this example, the theory is $`Sp(1)`$ gauge group with three hypermultiplets and two half-hypermultiplets. The moduli space has $`d_v=1`$ and $`d_H=3\times 2+2\times 2/23=5`$. The steps for finding the mirror theory are drawn in Figure 12. First we go to the Higgs branch. Now equations (4) and (5) allow us the break the D3-branes between the 1/2NS-branes and the neighboring 1/2D5-brane as part (a). Form part (a) we move 1/2NS-brane inside to pass one 1/2D5-brane and get part (b). In this step, the 1/2NS-branes get rid of the D3-brane ending on them already. However, this brane setup does not readily give the correct mirror theory and we need go to the next step, i.e., moving 1/2NS-brane one step further inside as in part (c). Finally, we make the S-duality transformation and get the mirror theory in part (d). The mirror theory is $`SO(2)\times Sp(1)\times SO(3)\times Sp(1)\times SO(2)`$ with four bifundamentals and two half-hypermultiplets one for each $`Sp(1)`$ gauge group. We can check the moduli spaces of the mirror theory as having $`d_v=5`$ and $`d_H=(2\times 4+2\times 6+2\times 2)/2(2+3\times 3)=1211=1`$.
### 5.2 The general case
Now we discuss the general case, i.e., $`Sp^{}(k)`$ with $`n`$ hypermultiplets and two half-hypermultiplets. The moduli spaces have $`d_v=k`$ and $`d_H=2nk+2kk(2k+1)=2nkk(2k1)`$. The main steps to get the mirror theory are in Figure 13.
We can read out the mirror theory from the quiver diagram in part(b) and check the moduli space as having
$$\begin{array}{ccc}d_v\hfill & =\hfill & 4_{t=1}^kt+k(2n4k1)=2nkk(2k1),\hfill \\ d_H\hfill & =\hfill & [2_{t=1}^{k1}((2t)^2+2t(2t+2))+2(2k)^2+(2n4k)2k(2k+1)+2(2k)]/2\hfill \\ & \hfill & [2_{t=1}^{k1}(t(2t1)+t(2t+1))+2k(2k1)+(2n4k+1)k(2k+1)]\hfill \\ & =\hfill & k\hfill \end{array}$$
(10)
As the case of $`Sp`$ gauge group when $`n=2k`$, the mirror theory has only one $`Sp(k)`$ gauge group and the two half-hypermultiplets combine together to give one flavor for $`Sp(k)`$. It means that we have a “hidden FI-term” in the original theory. However, it is not the end of the story. By careful observation, we find that when $`n=2k1`$, the mirror theory has only one $`SO(2k)`$ gauge group and two half-hypermultiplets also combine together to give one flavor for $`SO(2k)`$ (this happens because in this case, we do not need move 1/2NS-brane one further step as we did from part (b) to part (c) in Figure 12). So we get a “hidden FI-term” in this case also. This is not expected initially because it seems that for $`n=2k1`$ we can not get the complete Higgs branch, but this is not true. By studying the part (a) of Figure 12, we find that for $`n=1`$ in $`Sp^{}(1)`$ we indeed get complete Higgsing. Furthermore by the discussion in the next subsection we will see more clearly the reason why $`n=2k1`$ gives a “hidden FI-term”.
Now let us discuss the global symmetry. The results are very similar to those at the end of section 4. In the general case we have global $`SO(2N+1)`$ flavor symmetry<sup>5</sup><sup>5</sup>5From the discussion in the next subsection, the mirrors of single $`Sp^{}(k)`$ with $`N`$ flavors and single $`Sp(k)`$ with $`N+1`$ flavors are identical. In the latter case, the flavor symmetry is $`SO(2N+2)`$, but in the former case, we see only an obvious $`SO(2N+1)`$ flavor symmetry. However, in current situation of product gauge theories the argument of section 5.3 can not be applied directly. There is true distinguishing between $`Sp(k)`$ and $`Sp^{}(k)`$ gauge theories. When $`N=2k1`$, the global symmetry goes to $`SO(2N+1)\times Sp(1)`$. When $`N=2k`$, the global symmetry goes to $`SO(2N+1)\times SO(2)`$.
### 5.3 Comparing the mirror of $`Sp(k)`$ and $`Sp^{}(k)`$
In the above, we have discussed the mirror of two kinds of $`Sp`$ gauge groups, i.e., $`Sp(k)`$ and $`Sp^{}(k)`$. We want to ask ourselves whether there is any relation between the mirrors of these two $`Sp`$ gauge groups? By checking the two quivers in Figure 11 and Figure 13, we find that these two quivers are exactly the same, except that $`N`$ flavors in $`Sp^{}(k)`$ should correspond to $`N+1`$ flavors in $`Sp(k)`$. This is reasonable because for $`Sp^{}(k)`$ with $`N`$ flavors there are two half-hypermultiplets which give the same degrees of freedom as one flavor. However, in principle there is a difference between one flavor and two half-hypermultiplets: for the former we can involve one mass parameter, but for the latter there is no such mass parameter. We will show, in the case of $`Sp^{}(k)`$, that the two half-hypermultiplets do combine to give one flavor with the mass parameter. To see this, we move one 1/2D5-brane from infinity at each side to pass the 1/2NS-brane. By using the $`s`$-configuration in section 2, we get the brane setup of $`Sp(k)`$ with an additional flavor. The whole discussion is shown in figure 14. Furthermore, it is easily to show that the two cases where a “hidden FI-parameter” shows in $`Sp(k)`$ and $`Sp^{}(k)`$ exactly match each other.
## 6 The mirror of $`SO(2k)`$ gauge theory
After the discussion of the mirror theories for $`Sp(k)`$ gauge groups, we now discuss $`SO(2k)`$. There are no known results for the mirror of $`SO(2k)`$ gauge groups and it is the main motivation of this paper to calculate it using the $`O3`$ plane. As in the last two sections, we first present the simple case of $`SO(2)`$ with three flavors, then give the general results for $`SO(2k)`$ with $`N`$ flavors.
### 6.1 $`SO(2)`$ with 3 flavors
For $`SO(2)`$ gauge theory with three flavors, the moduli spaces have $`d_v=1`$ and $`d_H=3\times 21=5`$. The steps for the mirror transformation are given in Figure 15. In part (a) , we break the D3-branes by preserving the supersymmetric configurations, then use (4) and (5) to move the 1/2NS-brane passing the 1/2D5-brane to get part (b). Unlike the $`Sp`$ case, part (b) is already convenient for the mirror transformation, so we can make S-duality directly and get part (c) . From the brane setup in part (c) we read out the mirror theory to be $`Sp(1)\times SO(2)\times Sp(1)\times SO(2)\times Sp(1)`$ with four bifundamentals and two half-hypermultiplets for the leftmost $`Sp(1)`$ and two half-hypermultiplets for the rightmost $`Sp(1)`$. Here we want to emphasize that in the two half-hypermultiplets for the leftmost $`Sp(1)`$, one comes from the 1/2D5-brane and the other from the $`\stackrel{~}{O3^{}}`$ projection (same for the rightmost $`Sp(1)`$). That the half-hypermultiplets come from different sources is a general phenomenon in $`SO(2k)`$. However, for our simple example, we can combine these two half-hypermultiplets together by moving the 1/2D5-brane in part (c) to go part (d). Now we have one flavor of $`Sp(1)`$ given by one physical D3-brane stuck between the 1/2NS-brane and the 1/2D5-brane. We need to emphasize that because the physical D3-brane is stuck between the 1/2NS-brane and the 1/2D5-brane, it does not contribute to the mass parameter. It will be interesting to compare it with the discussion in section 4.3, where we find that the two half-hypermultiplets of $`Sp^{}(k)`$ can combine to give a flavor with free mass parameter. Finally, we calculate the dimension of the moduli spaces of mirror theory as $`d_v=5`$ and $`d_H=(4\times 4/2+2\times 2)(2+3\times 3)=1211=1`$.
### 6.2 An exotic example: $`SO(2)`$ with 2 flavors
In this subsection, we discuss the mirror of $`SO(2)`$ with two flavors. This theory will show one nontrivial phenomenon. The moduli are $`d_v=1`$ and $`d_H=2\times 21=3`$. According to the standard procedure introduced in the last subsection we get the Higgs branch as part (a) in Figure 16 and the mirror theory in part (b). The dimensions of moduli spaces of the mirror theory in part (b) are $`d_v=1+1+1=3`$ and $`d_H=2\times 4/2+4\times 2/2(3+3+1)=98=1`$.
However it seems we can get another possible Higgs branch in part (c) by moving the 1/2NS-brane one further step inside from part (a). If these two 1/2NS-branes do not meet together, the brane setup is not convenient to perform S-duality to get the mirror theory and we must go back to part (a). But in this special example, these two 1/2NS-branes do meet together. Now if these two 1/2NS-brane can combine to leave the $`O3^+`$ plane, we do get another mirror theory like part (d). Let us assume it is correct first and calculate the moduli spaces. In the part (d), the mirror theory is $`Sp(1)\times SO(3)\times Sp(1)`$ with two bifundamentals, two half-hypermultiplets for the two $`Sp(1)`$ and one fundamental for $`SO(3)`$, so the moduli are $`d_v=3`$ and $`d_H=2\times 6/2+2\times 2/2+3(3+3+3)=119=2`$. Therefore the results do not match. There is another inconsistent result because in the mirror theory of part(d) we get one “hidden FI-term” which does not exist in the mirror theory of part (b).
What is the resolution for the above inconsistency? Notice the combination of two 1/2NS-branes on the $`O3^+`$ is S-dual to the combination of two 1/2D5-branes on the $`\stackrel{~}{O3^{}}`$. We have discussed this configuration in section 3.1, where we showed, only when there is an extra physical D3-brane between these two 1/2D5-branes (1/2NS-branes) can they combine and leave the O3-plane. So the conclusion is that the two 1/2NS-branes in part (c) can not combine and leave the O3-plane. We are left with only one correct mirror theory in part (b).
### 6.3 The general $`SO(2k)`$ with $`N`$ flavors
With the experience of the $`SO(2)`$ case, we can now work on the general $`SO(2k)`$ with $`N`$ flavors. The moduli for this theory are $`d_v=k`$ and $`d_H=2kNk(2k1)`$. The steps for the mirror theory are given in Figure 17.
Again, we first break the D3-branes according to the supersymmetric configuration, then move the 1/2NS-branes inside to go to part(a). The brane setup in part(a) can be considered as the brane setup of S-duality just by exchanging the roles of the 1/2NS-brane and the 1/2D5-brane and putting in a proper O3-plane. For clarity, we draw the quiver diagram of the mirror theory in part(b). Let us check the result again by calculating the moduli of the mirror theory as
$$\begin{array}{ccc}d_v\hfill & =\hfill & 4_{n=1}^{k1}n+k(2N4k+3)=2kNk(2k1),\hfill \\ d_H\hfill & =\hfill & [\frac{1}{2}\times 2_{n=1}^{2k2}(n+1)(n+2)+\frac{1}{2}4k^2(2N4k+2)+\frac{1}{2}(2k+2k+2+2)]\hfill \\ & \hfill & [4_{n=1}^{k1}n(2n+1)+k(2k+1)(N2k+2)+k(2k1)(N2k+1)]\hfill \\ & =\hfill & k.\hfill \end{array}$$
(11)
By checking part (a) in Figure 17, we find that there is a “hidden FI-parameter” in the original theory when $`N=2k1`$ because two 1/2NS-branes will meet in same interval of $`\stackrel{~}{O3^+}`$ plane. For general $`N,k`$, the global symmetry is an $`Sp(N)`$ flavor symmetry, but in the case $`N=2k1`$ it is enhanced to $`Sp(N)\times SO(3)`$. We want to point out that there is only one case where “hidden FI-parameters” show in $`SO(2k)`$ while for $`Sp(k)`$ and $`Sp^{}(k)`$ there are two cases. This difference can be seen very clearly in part (c) of Figure 16. In that case two 1/2NS-branes do meet in same interval, but they can not combine and leave $`O3^+`$-plane. So there is no “hidden FI-parameters”.
## 7 The mirror of $`SO(2k+1)`$ gauge theory
In this section, we discuss the mirror theories of $`SO(2k+1)`$ to complete our study of single gauge groups. We first present the simple example of $`SO(3)`$ with two flavors, then give the general results for $`SO(2k+1)`$ with $`N`$ flavors.
### 7.1 $`SO(3)`$ with 2 flavors
For $`SO(3)`$ with two flavors, the dimensions of moduli space are $`d_v=1`$ and $`d_H=2\times 33=3`$. The steps to get the mirror theory are shown in Figure 18. In part (a) we split the physical D5-branes into the 1/2D5-branes according the rules given in section two. In such a process we see the generation of two physical D3-branes which is necessary to account for the correct Higgs branch. In part (b) we split the D3-brane between the 1/2D5-branes and 1/2NS-branes to go to the Higgs branch. Notice that there is no D3-branes connecting 1/2NS-brane and the nearest 1/2D5-brane which is required by $`s`$-rule. In part (c) we move the 1/2NS-branes inside to get rid of the D3-branes ending on them. Now we can make S-duality to give the mirror theory in part (d). However, in our example, there is a special property: two 1/2D5-branes can combine together and leave the $`\stackrel{~}{O3^+}`$-plane to give one flavor.
Now we can read out the mirror theory as $`SO(3)\times Sp(1)\times SO(3)`$ with two bifundamentals and one flavor for $`Sp(1)`$. Let us calculate the dimension of moduli space. For the Coulomb branch, we have $`d_v=1+1+1=3`$ which matches the Higgs branch of the original theory. For the Higgs branch, naively we should have $`d_H=[\frac{1}{2}(6+6)+2][3+3+3]=1`$. However, the dimension can never be negative. The negative result means that our naive calculation is wrong. The reason is that in our naive calculation we assumed that there is complete Higgsing. However, in our example, there is no complete Higgsing in the mirror theory. After Higgsing, we still keep two $`SO(2)`$ gauge groups which give the correct $`d_H=[8][92]=1`$ and match the Coulomb branch in the original theory. Furthermore, in our example, we have one flavor in the mirror theory which means that there is a “hidden FI-term” in the original theory.
### 7.2 The general case: $`SO(2k+1)`$ with $`N`$ flavors
Now let us discuss the mirror of $`SO(2k+1)`$ with $`N`$ flavors. The dimensions of moduli spaces are $`d_v=k`$ and $`d_H=(2k+1)Nk(2k+1)`$. The steps to get the mirror theory are given in Figure 19. In part (a), we give the brane setup of the Higgs branch. In fact, we can consider it as well as the brane setup of the mirror theory by just changing the role of the vertical line and cross line (in Higgs branch, vertical lines denote 1/2D5-branes and cross lines, 1/2NS-branes; in the mirror theory, vertical lines denote 1/2NS-branes and cross lines, 1/2D5-branes). For convenience, we give the quiver diagram of the mirror theory in part (b).
Let us calculate the dimensions of the moduli spaces of the mirror theory to see if they match the dimensions of the moduli spaces of the original theory. The calculations are given as
$$\begin{array}{ccc}d_v\hfill & =\hfill & [_{i=1}^{k1}4i]+k(N2k+3)+(k+1)(N2k)\hfill \\ & =\hfill & (2k+1)Nk(2k+1)\hfill \\ d_H\hfill & =\hfill & 2_{i=1}^{k1}[\frac{(2i+1)2i}{2}+\frac{2i(2i+3)}{2}2i(2i+1)]\hfill \\ & +\hfill & (2N4k)\frac{2k(2k+2)}{2}(N2k+1)k(2k+1)(N2k)(k+1)(2(k+1)1)\hfill \\ & +\hfill & 2\frac{2k}{2}+2\frac{2k(2k+1)}{2}2k(2k+1)\hfill \\ & +\hfill & N\hfill \\ & =\hfill & [2k(k1)]+[(N2k)k(2k+1)]+[2k]+[N]\hfill \\ & =\hfill & k\hfill \end{array}$$
(12)
Notice that we add $`N`$ when we calculate $`d_H`$ because after the Higgsing, the mirror theory still keep $`N`$ $`SO(2)`$ gauge group. Furthermore, from the part (a) in Figure 19 we see when $`N=2k`$, two 1/2-branes can combine together and leave the orientifold plane. This means that when $`N=2k`$ there is a “hidden FI-term” in the original theory. This also means that in the special case, the original theory has an enhanced global $`Sp^{}(N)\times SO(3)`$ symmetry instead of $`Sp^{}(N)`$ flavor symmetry in general.
### 7.3 Comparing the mirrors of $`SO(2k)`$ and $`SO(2k+1)`$
At the end of this section, let us compare the mirror theories of $`SO(2k)`$ and $`SO(2k+1)`$. First we can start from the $`SO(2k+1)`$ with $`N+1`$ flavors to go to $`SO(2k)`$ with $`N`$ flavors by Higgsing one flavor. At the other side, by comparing the quivers in Figure 17 and Figure 19, it is obvious that if we change the $`SO(d)`$ gauge group in Figure 19 to $`SO(d2)`$ while keeping the $`Sp(d/2)`$ gauge group we get exactly the quiver in Figure 17. In particular, the two $`SO(3)`$ gauge group in Figure 19 go to $`SO(1)`$ and disappear as a gauge group but add two half-hypermultiplets to two $`Sp(1)`$ at the two ends of quiver in Figure 17. This pattern can also be found if we higgs $`SO(2k)`$ with $`N`$ flavors to $`SO(2k1)`$ with $`N1`$ flavors. In the latter case, we change the $`Sp(d/2)`$ gauge group in Figure 17 to $`Sp(d/21)`$ gauge group while keeping $`SO(d)`$ gauge group. After such a change, the quiver in Figure 17 becomes exactly the quiver in Figure 19 (the two nodes at the ends in Figure 17 disappear). Notice that the Higgsing in the original theory should correspond to the reduction of the Coulomb branch in the mirror theory. The change of gauge group is exactly the required reduction of the Coulomb branch in the mirror theory.
The above pattern passes another consistency check. Notice that for $`SO(2k+1)`$ gauge theory with $`N+1`$ flavors, it has an enhanced $`SO(3)`$ global symmetry when $`N+1=2k`$. After Higgsing, we get $`SO(2k)`$ with $`N`$ flavors. For the latter, it has an enhanced $`SO(3)`$ global symmetry exactly when $`N=2k1`$. We see such hidden global symmetry is not broken by the Higgs mechanism as it should be.
## 8 The mirror of $`Sp(k)\times SO(2m)`$
We have discussed the mirror for a single $`Sp`$ or $`SO`$ group above. In this section, we generalize the above construction to the case of the product of $`Sp`$ and $`SO`$ gauge groups. Because after crossing the 1/2NS-brane $`O3^\pm `$($`\stackrel{~}{O3^\pm }`$) change to $`O3^{}`$($`\stackrel{~}{O3^{}}`$) and vise versa, we get two series of products $`SO(2n_1)\times Sp(k_1)\times SO(2n_2)\times Sp(k_2)..`$ and $`SO(2n_1+1)\times Sp^{}(k_1)\times SO(2n_2+1)\times Sp^{}(k_2)..`$. In this section, we discuss the first series and leave the second series to next section. For simplicity, we will discuss only the product of two gauge groups, i.e., $`Sp(k)\times SO(2m)`$ (the case of more product groups can be directly generalized). For this simple case, we still have two choices, the so called “elliptic model” ($`X^6`$ direction is compactified) , or the “non-elliptic model” ($`X^6`$ direction is not compactified). We discuss these two models one by one.
### 8.1 The non-elliptic model
For the non-elliptic model, there are $`N`$ fundamentals for $`SO(2m)`$, $`H`$ fundamentals for $`Sp(k)`$ and one bifundamental (for simplicity we assume that $`N,H`$ are sufficiently large. For $`N,H`$ too small, there are a lot of special cases which need to be discussed individually and are tedious without providing too much new insight). The moduli are $`d_v=m+k`$ and $`d_H=2mN+2kH+2mkm(2m1)k(2k+1)`$. In constructing the mirror theory, we need to study three cases: $`m>k`$, $`m=k`$ and $`m<k`$. Let us start with the case of $`m>k`$. The mirror theory is given in Figure 20.
When we go to the Higgs branch, we can connect the D3-branes at the two sides of middle 1/2NS-brane. Because $`m>k`$, we can connect only $`k`$ D3-branes such that they end on the left and the right 1/2NS-branes. There are still $`mk`$ D3-branes ending on the middle 1/2NS-brane from the right. To get rid of those D3-branes, we must move the middle 1/2NS-brane to the right. The final Higgs branch after such a motion is given in part (a) of Figure 20 and the quiver diagram of the mirror, in part (b). The moduli of the mirror can be calculated as
$$\begin{array}{ccc}d_v\hfill & =\hfill & [_{p=1}^{p=k1}2p]+(2H2k+1)k+[2_{p=k+1}^{m1}p]\hfill \\ & +\hfill & (2N4m+2k+3)m+[2_{p=1}^{p=m1}p]\hfill \\ & =\hfill & 2mN+2kH+2mkm(2m1)k(2k+1),\hfill \\ d_H\hfill & =\hfill & [_{i=1}^{i=k1}(2i\times 2i+2i\times (2i+2))/2i(2i1)i(2i+1)]\hfill \\ & +\hfill & [4k^2/2+(2H2k1)2k(2k+1)/2(Hk)2k(2k+1)k(2k1)]\hfill \\ & +\hfill & [_{i=1}^{mk1}(2k+2i1)(2k+2i)/2+(2k+2i)(2k+2i+1)/22(k+i)(2k+2i+1)]\hfill \\ & +\hfill & [2m(2m1)/2+4m^2(2N4m+2k+2)/2\hfill \\ & & (N2m+k+1)(m(2m1)+m(2m+1))m(2m+1)]\hfill \\ & +\hfill & [_{i=1}^{m1}2i(2i+1)/2+(2i+1)(2i+2)/22i(2i+1)]\hfill \\ & +\hfill & [2k/2+2m/2+2/2+2m/2]\hfill \\ & =\hfill & [k(k1)]+[2k^2]+[(km)(mk1)]+[2m]+[m^21]+[k+2m+1]\hfill \\ & =\hfill & k+m\hfill \end{array}$$
(13)
From part (a) of Figure 20, we see that when $`2N4m+2k+2=0`$, the two 1/2NS-branes meet together which indicates a “hidden FI-parameter” in the original theory.
After the discussion of the $`m>k`$ case, we go to the $`m=k`$ case. Here, by connecting the D3-branes between the two sides of the middle 1/2NS-brane, we get the Higgs branch looking like part (a) in Figure 21.
From the quiver diagram part (b) we recalculate the moduli space as:
$$\begin{array}{ccc}d_v\hfill & =\hfill & [_{p=1}^{p=k1}2p]+(2H2k1+2N2m+2+1)k+[2_{p=k+1}^{m1}p]\hfill \\ & =\hfill & 2kN+2kH2k^2\hfill \\ & =\hfill & 2mN+2kH+2mkm(2m1)k(2k+1)whenm=k,\hfill \\ d_H\hfill & =\hfill & [_{i=1}^{i=k1}(2i\times 2i+2i\times (2i+2))/2i(2i1)i(2i+1)]\hfill \\ & +\hfill & [4k^2/2+(2H2k2)2k(2k+1)/2(2H2k1)k(2k+1)k(2k1)]\hfill \\ & +\hfill & [(2N2m+2)4m^2/2(Nm+1)(m(2m+1)+m(2m1))]\hfill \\ & +\hfill & [_{i=1}^{m1}2i(2i+1)/2+(2i+1)(2i+2)/22i(2i+1)]\hfill \\ & +\hfill & [2k/2+2m/2+2/2+2m/2]\hfill \\ & =\hfill & [k(k1)]+[2k^2]+[0]+[m^21]+[k+2m+1]\hfill \\ & =\hfill & k+m.\hfill \end{array}$$
(14)
From the figure again, when $`2H2k=0`$ or $`2N2m+2=0`$, the two 1/2NS-branes meet together to give a “hidden FI-term” in the original theory.
Now we are left with only one case, i.e., $`m<k`$. In this last case, to get rid of the D3-branes, the middle 1/2NS brane should move to the left direction. The result is shown in Figure 22.
The moduli of the mirror theory are
$$\begin{array}{ccc}d_v\hfill & =\hfill & [_{p=1}^{p=k1}2p]+(2H4k+2m+1)k+[2_{p=m+1}^{k1}p]\hfill \\ & +\hfill & (2N2m+3)m+[2_{p=1}^{p=m1}p]\hfill \\ & =\hfill & 2mN+2kH+2mkm(2m1)k(2k+1),\hfill \\ d_H\hfill & =\hfill & [_{i=1}^{i=k1}(2i\times 2i+2i\times (2i+2))/2i(2i1)i(2i+1)]\hfill \\ & +\hfill & [4k^2/2+(2H4k+2m2)2k(2k+1)/2+4k^2/2\hfill \\ & & (2H4k+2m1)k(2k+1)2k(2k1)]\hfill \\ & +\hfill & [_{i=1}^{km1}2(m+i)2(m+i)/2+2(m+i)2(m+i+1)/2\hfill \\ & & (m+i)(2(m+i)1)(m+i)(2(m+i)+1)]\hfill \\ & +\hfill & [2m(2m+2)/2+4m^2(2N2m+2)/2\hfill \\ & & (Nm+1)(m(2m1)+m(2m+1))m(2m+1)]\hfill \\ & +\hfill & [_{i=1}^{m1}2i(2i+1)/2+(2i+1)(2i+2)/22i(2i+1)]\hfill \\ & +\hfill & [2k/2+2k/2+2/2+2m/2]\hfill \\ & =\hfill & [k(k1)]+[2k^2+k]+[(k+m)(km1)]+[m]+[m^21]+[2k+m+1]\hfill \\ & =\hfill & k+m\hfill \end{array}$$
(15)
There is also a possible “hidden FI-term” in original theory when $`2H4k+2m2=0`$ which can be explicitly seen in part(a) in Figure 22.
### 8.2 The elliptic model
In the elliptic model, the $`X^6`$ direction is compactified such that for consistency, we must have an even number of 1/2NS-branes and an even number of gauge groups where half of them are $`Sp`$ gauge groups and the other half, $`SO`$ gauge groups. We discuss the case of only two gauge groups, i.e., $`Sp(k)\times SO(2m)`$ with $`H`$ fundamentals for $`Sp(k)`$, $`N`$ fundamentals for $`SO(2m)`$ and two bifundamentals. The moduli for this theory are $`d_v=k+m`$ and $`d_H=2mN+2kH+4mkm(2m1)k(2k+1)`$. The mirror theory for the elliptic model is similar to the non-elliptic model. The only difference is that in the non-elliptic model we can connect the D3-branes only at the middle 1/2NS-brane, but here in the elliptic model we can connect the D3-branes to all 1/2NS-branes (here two 1/2NS-branes). Again we divide into three cases to discuss. The simple one is the case $`m=k`$. In this case, we can connect all D3-branes such that no D3-brane is left to end on the 1/2NS-branes. The Higgs branch and the quiver of the mirror are given in parts (a) (b) of Figure 23 and the moduli are:
$$\begin{array}{ccc}d_v\hfill & =\hfill & (2H+2N)k\hfill \\ & =\hfill & 2mN+2kH+4mkm(2m1)k(2k+1)whenm=k,\hfill \\ d_H\hfill & =\hfill & [(2H2)2k(2k+1)/2(H1)2k(2k+1)k(2k+1)]\hfill \\ & +\hfill & [(2N+2)4k^2/2N(k(2k+1)+k(2k1))k(2k1)]\hfill \\ & +\hfill & [2k/2+2k/2]\hfill \\ & =\hfill & [k(2k+1)]+[2k^2+k]+[2k]=k+m\hfill \end{array}$$
(16)
There is still one case where the “hidden FI-term” appears, namely when $`H=1`$. In this case, two 1/2NS-branes in part(a) of Figure 23 meet together.
Now we move to the case of $`k>m`$. The Higgs branch looks like the superposition of the Higgs branch of the $`m=k`$ case together with that of a single $`Sp(km)`$ gauge theory. We give the result in Figure 24. To check the result, we calculate the moduli as
$$\begin{array}{ccc}d_v\hfill & =\hfill & [4_{i=1}^{i=km1}(m+i)]+k(2H4(km)+1)+m(2N+3)\hfill \\ & =\hfill & 2mN+2kH+4mkm(2m1)k(2k+1),\hfill \\ d_H\hfill & =\hfill & [2_{i=1}^{i=km1}2(m+i)2(m+i)/2+2(m+i)2(m+i+1)/2\hfill \\ & & (m+i)(2(m+i)1)(m+i)(2(m+i)+1)]\hfill \\ & +\hfill & [2\times 4k^2/2+(2H4(km)2)2k(2k+1)/2\hfill \\ & & (2H4(km)1)k(2k+1)2k(2k1)]\hfill \\ & +\hfill & [(2N+2)4m^2/2+2\times 2m(2m+2)/2(N+1)(m(2m+1)+m(2m1))m(2m+1)]\hfill \\ & +\hfill & [2\times 2k/2]\hfill \\ & =\hfill & [2(k^2m^2km)]+[2k^2+k]+[2m^2+3m]+[2k]\hfill \\ & =\hfill & k+m.\hfill \end{array}$$
(17)
When $`2H4(km)2=0`$, two 1/2NS-branes will meet together in part(a) of Figure 24. This is the condition that a “hidden FI-term” exists.
Now the remainder case is $`m>k`$. In this case, the Higgs branch looks like the superposition of two Higgs branches: that of the $`m=k`$ case and that of a single $`SO(2(mk))`$ theory. The result can be found in Figure 25. The moduli of the mirror theory are
$$\begin{array}{ccc}d_v\hfill & =\hfill & [4_{i=1}^{i=mk1}(k+i)]+k(2H+1)+m(2N4(mk)+3)\hfill \\ & =\hfill & 2mN+2kH+4mkm(2m1)k(2k+1),\hfill \\ d_H\hfill & =\hfill & [2_{i=1}^{i=mk1}2(k+i)(2k+2i+1)/2+(2k+2i+1)(2k+2i+2)/2\hfill \\ & & 2(k+i)(2k+2i+1)]\hfill \\ & +\hfill & [(2N4(mk)+2)4m^2/2(N2(mk)+1)(m(2m+1)+m(2m1))\hfill \\ & & m(2m+1)]\hfill \\ & +\hfill & [2h2k(2k+1)/2+2(2k+1)(2k+2)/2(2H+1)k(2k+1)]\hfill \\ & +\hfill & [2\times 2m/2]\hfill \\ & =\hfill & [2m^22(k+1)^2]+[2m^2m]+[2k^2+5k+2]+[2k]\hfill \\ & =\hfill & k+m\hfill \end{array}$$
(18)
When $`2N4(mk)+2=0`$, there is a “hidden FI-term” in the original theory.
## 9 The mirror of $`Sp^{}(k)\times SO(2m+1)`$
For completion, we give one more example: the mirror theory of $`Sp^{}(k)\times SO(2m+1)`$. We assume that there are $`H`$ flavors for $`Sp^{}(k)`$ gauge theory and $`N`$ flavors for $`SO(2m+1)`$ gauge theory. Besides, there are one or two bifundamentals and half-hypermultiplet for $`Sp^{}(k)`$ depend on different situations. Again we divide our discussion into two parts: non-elliptic model and elliptic model.
### 9.1 The non-elliptic model
Let us start from the non-elliptic model. In this case, the dimensions of moduli spaces are $`d_v=k+m`$ and $`d_H=2kH+(2m+1)N+k(2m+1)+kk(2k+1)m(2m+1)=2kH+(2m+1)N2k^2+k2m^2m+2km`$ (here again, for simplicity we assume $`N,H`$ are sufficiently large to avoid special cases). The mirror theory depends on whether $`m>k`$ , $`m=k`$ or $`m<k`$. We first give the mirror of the case $`m=k`$ because in this particular case, we can combine the D3-branes at the two sides of middle 1/2NS-brane such that there is no D3-branes ending on the middle 1/2NS-brane anymore. The mirror theory is given in Figure 26.
Let us check it by calculating the dimensions of moduli spaces of the mirror theory:
$$\begin{array}{ccc}d_v\hfill & =\hfill & 2_{i=1}^{k1}i+(2H2k+2)k+(Nk)(k+k+1)+k+2_{i=1}^{k1}i\hfill \\ & =\hfill & 2kH+(2k+1)N2k^2\hfill \\ d_H\hfill & =\hfill & _{i=1}^{k1}[\frac{2i2i}{2}+\frac{2i(2i+2)}{2}i(2i1)i(2i+1)]\hfill \\ & +\hfill & _{i=1}^{k1}[\frac{(2i+1)2i}{2}+\frac{2i(2i+3)}{2}2i(2i+1)]\hfill \\ & +\hfill & \frac{2k(2k+1)}{2}(2H2k)(Hk)2k(2k+1)\hfill \\ & +\hfill & \frac{2k(2k+2)}{2}(2N_2k)(Nk)(k(2k+1)+(k+1)(2k+1))\hfill \\ & +\hfill & \frac{2k2k}{2}k(2k+1)k(2k1)+\frac{2k(2k+1)}{2}k(2k+1)\hfill \\ & +\hfill & 3\frac{2k}{2}+N\hfill \\ & =\hfill & [k^2k]+[k^2k]+[0]+[(Nk)]+[2k^2]+[3k+N]\hfill \\ & =\hfill & 2k,\hfill \end{array}$$
(19)
where when we calculate the $`d_H`$ we add the term $`N`$ to account for the remaining $`H`$ $`SO(2)`$ gauge groups after Higgsing (this happens for latter examples so we will not mention it every time). When $`2H2m=0`$ there is a “hidden FI-term” in the original theory.
Now we go to the case that $`k>m`$. In this case, after connecting the D3-branes at the two sides of the middle 1/2NS-brane, we still have $`km`$ D3-brane ending on the middle 1/2NS-brane from the left. The mirror theory is given in Figure 27.
The dimensions of the moduli space of the mirror theory are
$$\begin{array}{ccc}d_v\hfill & =\hfill & [2_{i=1}^{k1}i]+[2_{i=1}^{m1}i]+(2H4k+2m+3)k\hfill \\ & +\hfill & [2_{i=1}^{km1}(m+i)]+[(Nm+2)m+(Nm)(m+1)]\hfill \\ & =\hfill & [k^2k]+[m^2m]+[2kH4k^2+2km+3k]\hfill \\ & +\hfill & [k^2m^2km]+[(2m+1)N2m^2+m]\hfill \\ & =\hfill & 2kH+(2m+1)N2k^22m^2+km+2km\hfill \\ d_H\hfill & =\hfill & _{i=1}^{k1}[\frac{2i2i}{2}+\frac{2i(2i+2)}{2}i(2i1)i(2i+1)]\hfill \\ & +\hfill & 2\frac{2k2k}{2}+\frac{2k(2k+1)}{2}(2H4k+2m)(2H4k+2m+1)k(2k+1)2k(2k1)\hfill \\ & +\hfill & _{i=1}^{km1}[\frac{(2(m+i))^2}{2}+\frac{2(m+i)2(m+i+1)}{2}(m+i)(2(m+i)1)(m+i)(2(m+i)+1)]\hfill \\ & +\hfill & \frac{2m(2m+2)}{2}(2N2m+1)(Nm+1)m(2m+1)(Nm)m(2(m+1)1)\hfill \\ & +\hfill & _{i=1}^{m1}[\frac{(2i+1)2i}{2}+\frac{2i(2i+3)}{2}2i(2i+1)]\hfill \\ & +\hfill & \frac{2m(2m+1)}{2}m(2m+1)+2\frac{2k}{2}+\frac{2m}{2}+N\hfill \\ & =\hfill & [k^2k]+[2k^2+k]+[k^2m^2km]+[N+2m]+[m^2m]+[N+2k+m]\hfill \\ & =\hfill & k+m.\hfill \end{array}$$
(20)
After the discussion of above two cases, we go to the last case: $`k<m`$. In this case, because $`k<m`$, after the combination of D3-branes at the two sides of middle 1/2NS-brane, we still leave $`mk`$ D3-brane ending on it from the right. The mirror theory is given in Figure 28.
Let us calculate the dimensions of moduli spaces:
$$\begin{array}{ccc}d_v\hfill & =\hfill & [2_{i=1}^{k1}i]+[2_{i=1}^{m1}i]+(2H2k+2)k\hfill \\ & +\hfill & 2_{i=1}^{mk1}(k+i)+m(N2m+k+3)+(m+1)(N2m+k)\hfill \\ & =\hfill & [k^2k]+[m^2m]+[2kH2k^2+2k]+[m^2k^2km]\hfill \\ & +\hfill & [(2m+1)N4m^2+m+k+2km]\hfill \\ & =\hfill & 2kH+(2m+1)N+2km2k^22m^2m+k\hfill \\ d_H\hfill & =\hfill & _{i=1}^{k1}[\frac{2i2i}{2}+\frac{2i(2i+2)}{2}i(2i1)i(2i+1)]\hfill \\ & +\hfill & _{i=1}^{m1}[\frac{(2i+1)2i}{2}+\frac{2i(2i+3)}{2}2i(2i+1)]\hfill \\ & +\hfill & \frac{(2k)^2}{2}+\frac{2k(2k+1)}{2}(2H2k)k(2k1)(2H2k+1)k(2k+1)\hfill \\ & +\hfill & _{i=1}^{mk1}[\frac{(2k+2i+1)2(k+i)}{2}+\frac{2(k+i)(2k+2i+3)}{2}2(k+i)(2(k+i)+1)]\hfill \\ & +\hfill & [2\frac{2m(2m+1)}{2}+\frac{2m(2m+2)}{2}(2N4m+2k)\hfill \\ & & (N2m+k+3)m(2m+1)(N2m+k)(m+1)(2m+1)]\hfill \\ & +\hfill & \frac{2k(2k+3)}{2}+k+2m+N\hfill \\ & =\hfill & [k^2k]+[m^2m]+[2k^2]+[m^2k^2km]\hfill \\ & +\hfill & [N2m^2+mk]+[2k^2+4k+2m+N]\hfill \\ & =\hfill & m+k.\hfill \end{array}$$
(21)
### 9.2 The elliptic model
In this section, we discuss the mirror theory of $`Sp^{}(k)\times SO(2m+1)`$ in the elliptic model. Now because $`X^6`$ is compact, the matter contents are $`H`$ flavors for $`Sp^{}(k)`$, $`N`$ flavors for $`SO(2m+1)`$ and two bifundamentals. The dimensions of the moduli spaces are $`d_v=k+m`$ and $`d_H=2kH+(2m+1)N+2k(2m+1)k(2k+1)m(2m+1)=2kH+(2m+1)N+4km2k^22m^2m+k`$. Again, our investigation will be divided into three cases $`k=m`$, $`k>m`$ and $`k<m`$.
Let us start from the case $`k=m`$. In this case, because we can combine all D3-branes at the two sides of 1/2NS-branes, it makes the mirror theory very simple as shown in Figure 29.
Let us check the dimensions of moduli spaces:
$$\begin{array}{ccc}d_v\hfill & =\hfill & 2kH+Nk+N(k+1)=2kH+(2k+1)N\hfill \\ d_H\hfill & =\hfill & [\frac{2k(2k+1)}{2}2H2Hk(2k+1)]+[N]+[2\frac{2k}{2}]\hfill \\ & +\hfill & [\frac{2k(2k+2)}{2}2NNk(2k+1)N(k+1)(2k+1)]\hfill \\ & =\hfill & 2k.\hfill \end{array}$$
(22)
Now we go to the case of $`k>m`$. In this case, After combining the D3-branes, we still leave $`km`$ D3-branes in the interval of $`\stackrel{~}{O3^+}`$-plane. The mirror theory is given in Figure 30.
The dimensions of moduli spaces are
$$\begin{array}{ccc}d_v\hfill & =\hfill & 2_{i=1}^{km1}2(m+i)+k(2H4k+4m+3)+m(N+1)+(m+1)N\hfill \\ & =\hfill & [2k^22m^22k2m]+[2kH+(2m+1)N+4km4k^2+3k+m]\hfill \\ & =\hfill & 2kH+(2m+1)N+4km2k^22m^2+km\hfill \\ d_H\hfill & =\hfill & 2_{i=1}^{km1}[\frac{(2(m+i))^2}{2}+\frac{2(m+i)2(m+i+1)}{2}(m+i)(2m+2i1)(m+i)(2m+2i+1)]\hfill \\ & +\hfill & 2\frac{(2k)^2}{2}+\frac{2k(2k+1)}{2}(2H4k+4m)(2H4k+4m+1)k(2k+1)2k(2k1)\hfill \\ & +\hfill & 2k+N\hfill \\ & =\hfill & [2k^22m^22k2m]+[2k^2+k]+[N+2m^2+3m]+[N+2k]\hfill \\ & =\hfill & k+m\hfill \end{array}$$
(23)
We are left only one more example, i.e., the case of $`k<m`$. For this case, after the combination, we still have $`mk`$ D3-branes in the interval of $`\stackrel{~}{O3^{}}`$-plane. The mirror theory is given in Figure 31.
The dimensions of moduli spaces are
$$\begin{array}{ccc}d_v\hfill & =\hfill & 2_{i=1}^{mk1}2(k+i)+k(2H+1)++m(N2m+2k+3)+(m+1)(N2m+2k)\hfill \\ & =\hfill & [2m^22k^22k2m]+[2kH+(2m+1)N+4km+m+3k]\hfill \\ & =\hfill & 2kH+(2m+1)N2m^22k^2m+k\hfill \\ d_h\hfill & =\hfill & 2_{i=1}^{mk1}[\frac{(2k+2i)(2k+2i+1)}{2}+\frac{(2k+2i)(2k+2i+3)}{2}2(k+i)(2k+2i+1)]\hfill \\ & +\hfill & [2\frac{2m(2m+1)}{2}+\frac{2m(2m+2)}{2}(2n4m+4k)\hfill \\ & & (N2m+2k+3)m(2m+1)(N2m+2k)(m+1)(2m+1)]\hfill \\ & +\hfill & \frac{2k(2k+1)}{2}(2H)(2H+1)k(2k+1)+2\frac{2k(2k+3)}{2}\hfill \\ & +\hfill & 2m+N\hfill \\ & =\hfill & [2m^22k^22k2m]+[N2m^22k+m]+[2k^2+5k]+[2m+N]\hfill \\ & =\hfill & k+m.\hfill \end{array}$$
(24)
## 10 Conclusion
In this paper, we give the mirror theories of $`Sp(k)`$ and $`SO(n)`$ gauge theories. In particular, for the first time the mirror of $`SO(n)`$ gauge theory is given. In the construction of the mirror, we have made an assumption about the splitting of D5-branes on O3-planes in the brane-plane system<sup>6</sup><sup>6</sup>6For more details about the brane-plane system see . We want to emphasize that because the splitting of D5-brane on O3-plane is a nontrivial dynamical process and we do not fully understand it at this moment, we can not really prove our assumption by calculation. However, although our discussions in this paper indicate that our assumption is consistent, the other independent checks are favorable. This gives one direction of further work as to prove our observation.
Furthermore, as we discussed in section three, our rules observed in this paper about the splitting of physical brane predict some nontrivial strong coupling limit of a particular field theory. It will be very interesting to use the Seiberg-Witten curve to show whether it is true.
There is another direction to pursue our investigation. By rotating one of the 1/2NS-branes we break the $`N=4`$ theory in three dimensions to an $`N=2`$ theory. Then we can discuss the mirror of $`N=2`$ in three dimension as we have done in this paper. However, because there is less supersymmetry in the $`N=2`$ case, things become more complex (for a detailed explanation of new features in $`N=2`$, see ). Indeed, we can even break the supersymmetry further to discuss the mirror symmetry in the $`N=1`$ case .
## Acknowledgements
We would like to extend our sincere gratitude to A. Kapustin, K. Intriligator and A. Uranga for fruitful discussions. Furthermore A. Hanany would like to thank A. Zaffaroni for discussion and B.Feng would like to thank Yang-Hui He, A. Naqvi, and J. S. Song for their suggestions and help.
This project is supported by the DOE under grant no. DE-FC02-94ER40818. A.H is also partially supported by the National Science Foundation under grant no. PHY94-07194, by an A.P.Sloan Foundation Fellowship and by a DOE OJI award.
|
warning/0004/astro-ph0004384.html
|
ar5iv
|
text
|
# Scalar fields in an anisotropic closed universe.
## I Introduction
Current observations of luminosity-redshift relations of type Ia supernovas and measurements of the anisotropy cosmic background radiation and mass power spectrum provide evidence that the total matter density of the universe coincides with its critical value. This, agrees with the theoretical arguments derived from inflation, where it is suggested that our universe should become soon flat after a short period of inflation.
Since, astronomical observations give rise to the bound $`\mathrm{\Omega }_M0.3`$, in which baryons and Cold Dark Matter are included, we are in front of a problematic situation. There exist a sort of ”missing energy” that should represent something like the $`70\%`$ of the critical value.
It has been argued that the simplest explication, a cosmological constant (vacuum energy density) is consistent with these results. Other alternatives have been considered. For instance, bulk pressure that is significantly negative, i.e., $`\alpha 1/3`$, where $`p=\alpha \rho `$ is the effective equation of state, in which $`p`$ is the pressure and $`\rho `$ is the energy density. Here, this sort of matter could correspond to a network of topological defects (such that strings or walls) or an evolving scalar field (referred as quintessence), $`Q(t)`$, in which case the pressure and the energy density become defined by $`p_Q={\displaystyle \frac{1}{2}}\dot{Q}V(Q)`$ and $`\rho _Q={\displaystyle \frac{1}{2}}\dot{Q}^2+V(Q)`$, respectively. Here, $`V(Q)`$ represents the scalar potential associated to the scalar field $`Q`$ and the overdots specify derivatives with respect to time.
The main difference between these two sort of models i.e., the cosmological constant and the scalar field with a negative pressure, is that the latter is spatially inhomogeneous and thus can cluster gravitationally, where the former is totally spatial uniform. In this respect, the fluctuation of the scalar field could have an important effect on large scale structure of the universe.
Since the total energy density equals the critical density, then the spatial part of the metric is supposed to correspond to a flat Friedmann-Robertson-Walker (FRW) metric. However, It has been mentioned that the observations referred above, i.e. those related to type Ia supernovas, do not rule out a different type of geometry. There, it was advanced that these measurements allow an open universe in which the cosmological constant is vanished.
From the theoretical point of view, it seems that quantum field theory is more consistent on compact spatial surfaces that in hyperbolic spaces. On the other hand, in quantum cosmology the ”birth” of universes have been described under the assumption that the three-geometry is characterized by a close spatial surface. In this way, motivated by quantum cosmology and by the short period of inflation that the universe underwent at early time in its evolution, we describe in this paper the conditions under which a closed universe model may look flat al low redshift. This kind of situation has been considered in the literature. There, a closed universe with $`\mathrm{\Omega }_0<1`$ was studied. Here, $`\mathrm{\Omega }_0`$ represents the density parameter associated to the total mass of the universe. Openness is obtained by adding to the matter density texture or tangled strings with equation of state $`p=\rho /3`$. Here, the additional energy density is redshifted as $`a^2`$, similar to the curvature term in a closed universe, where $`a`$ is the scale factor. Kolb studied this sort of matter, arising to the important conclusion that a closed universe may expand forever at constant speed.
It is natural to assume the geometry at very early epoch more general than just the isotropic and homogeneous FRW. Although the universe, on large scale, seems homogeneous and isotropic at present, there is no observational data that guarantees the isotropy in an era prior to the recombination. In fact, it is possible to begin with an anisotropic universe which isotropizes during its evolution.
In relation with the matter that we could take into account in an anisotropic background, may have many possible sources. For instant, populations of collisionless particles, gravitons, electric, or magnetic fields, or by topological defects.
The anisotropic dynamics can in general encode either relative velocity effects or dissipative effects or both. In this respect, it is possible to start with an anisotropic universe that eventually isotropizes at later time in the evolution of the universe due to dissipative processes involving the matter that it contains. Also, this kind of model seems to be more appropriated when adiabatic theory of galaxy formation is considered . Thus, it seems quite natural to include in this study a matter component with this kind of property, in a background which in essence is anisotropic.
The aim of the present paper is to study a closed anisotropic cosmological model, with metric corresponding to Kanstowski-Sachs, where the matter content is composed by an imperfect fluid together with a scalar field whose equation of state parameter $`\alpha `$ remains negative during the evolution of the universe.
## II The field equations
We start by considering the effective Einstein Lagrangian given by
$$=\frac{1}{\kappa }R+\frac{1}{2}(_\mu Q)^2V(Q)+_M,$$
(1)
where, $`\kappa =16\pi G`$, with $`G`$ the Newton’s gravitational constant, $`R`$ the scalar curvature, $`Q`$ the quintessence scalar field with associated potential $`V(Q)`$, and $`_M`$ represents the matter Lagrangian density. We assume that the matter lagrangian density $`_{}`$ is associated to a fluid (characterized by the pressure and energy density, $`p_M`$ and $`\rho _M`$, respectively) which presents a shear viscosity. By taking a preferred timelike vector field (four velocity) $`u^\alpha `$, which satisfies $`u^\alpha u_\alpha =1`$ and it is a Ricci eigenvector, we can write the following matter energy-momentum tensor
$`T_{\alpha \beta }=(\rho _M+p_M)u_\alpha u_\beta p_Mg_{\alpha \beta }+2\eta _M\sigma _{\alpha \beta },`$ (2)
where $`\eta _M`$ and $`\sigma _{\alpha \beta }`$ are the shear viscosity (or coefficient of dynamic viscosity, $`\eta _M0`$) and the traceless shear tensor, respectively. The shear tensor has the form
$`\sigma _{\alpha \beta }=h_\alpha ^\gamma u_{(\gamma ;\delta )}h_\beta ^\delta {\displaystyle \frac{1}{3}}\theta h_{\alpha \beta },`$ (3)
where $`\theta =u_{;\alpha }^\alpha `$ is the scalar expansion and $`h_{\alpha \gamma }`$ is the projection tensor defined from the expression $`h_{\alpha \beta }=g_{\alpha \beta }u_\alpha u_\beta `$, with signature for the metric $`(+,,,)`$.
In this paper we consider a spatially homogeneous background spacetime of Kantowski-Sachs type, which, as far as it is known, it is the only spatially homogeneous models that it is not included in the Bianchi classification, thus we have
$$ds^2=dt^2a^2(t)\left(d\theta ^2+sin^2(\theta )d\varphi ^2\right)b^2(t)dr^2,$$
(4)
where $`a`$ and $`b`$ are the scale factors which describe the anisotropy of the model. This sort of metric combines spherical symmetry with a translational symmetry in the ”radial” direction. The metric (4) has been studied by many authors that have considered different sort of matter components. As examples, it has been considered a homogeneous shear-free cosmological models with an imperfect fluid matter content . On the other hand, a energy-effective-action related to string theory has been studied . Here, when the pseudoscalar axion field is time depending only, it reduces to that of a stiff perfect-fluid cosmology. Also, a scalar field for a convex positive scalar potential was taken into account , among others.
Since the metric (4) is spatially homogeneous the scalar field $`Q`$ can only depend on time, and thus the time-time component of Einstein’s field equations is
$$\left(\frac{\dot{a}}{a}\right)^2+\left(\frac{\dot{a}}{a}\right)\left(\frac{\dot{b}}{b}\right)+\frac{1}{a^2}=\frac{2\kappa }{3}\left(\rho _M+\rho _Q\right),$$
(5)
where, as was mentioned above, the dots stand for derivatives with respect to the cosmological time $`t`$. From the metric (4), and considering the comoving frame, i.e., $`u^\alpha =\delta _0^\alpha `$, we find that the components of the shear tensor are given by
$`\sigma _{11}={\displaystyle \frac{2}{3}}b^2\left({\displaystyle \frac{\dot{a}}{a}}{\displaystyle \frac{\dot{b}}{b}}\right),`$ (6)
$`\sigma _{22}={\displaystyle \frac{1}{3}}a^2\left({\displaystyle \frac{\dot{b}}{b}}{\displaystyle \frac{\dot{a}}{a}}\right),`$ (7)
$`\sigma _{33}={\displaystyle \frac{1}{3}}sin^2(\theta )a^2\left({\displaystyle \frac{\dot{b}}{b}}{\displaystyle \frac{\dot{a}}{a}}\right).`$ (8)
Here, $`\sigma _{00}=0`$ and $`\sigma _\alpha ^\alpha =0`$. The other components of Einstein’s field equations are
$`2{\displaystyle \frac{\ddot{a}}{a}}+\left({\displaystyle \frac{\dot{a}}{a}}\right)^2+\left({\displaystyle \frac{1}{a}}\right)^2`$ (9)
$`=\kappa (p_M+p_Q){\displaystyle \frac{4}{3}}\kappa \eta _M\left({\displaystyle \frac{\dot{a}}{a}}{\displaystyle \frac{\dot{b}}{b}}\right),`$ (10)
and
$`{\displaystyle \frac{\ddot{b}}{b}}+{\displaystyle \frac{\ddot{a}}{a}}+{\displaystyle \frac{\dot{a}\dot{b}}{ab}}=\kappa (p_M+p_Q)+{\displaystyle \frac{2}{3}}\kappa \eta _M\left({\displaystyle \frac{\dot{a}}{a}}{\displaystyle \frac{\dot{b}}{b}}\right).`$ (11)
In order to solve this set of equations, we need to supply this set with equation of state for the matter content and the scalar field. We assume that the matter content satisfies the relation $`p_M=\gamma \rho _M`$, where $`\gamma `$ may bee a time depending quantity and its (present) value depends on the characteristics of matter content. In the following we assume that this constant lies in the range $`0\gamma 1`$, where the extremes correspond to dust and stiff fluid, respectively. In the same way, we shall assume that the scalar field $`Q`$ satisfies a similar effective equation of state, i.e., $`p_Q=\alpha \rho _Q`$, where now the parameter $`\alpha `$ is assumed to be negative.
In order to have a universe which is closed, but still have a matter density content corresponding to a flat universe, we impose the following relations:
$$\kappa \rho __Q=a^2,$$
(12)
and
$$\eta ^{}=\eta __M+\frac{1}{2}\frac{\rho __Q}{\overline{\sigma }},$$
(13)
where $`\overline{\sigma }=\left({\displaystyle \frac{\dot{a}}{a}}{\displaystyle \frac{\dot{b}}{b}}\right)`$ and $`\alpha `$ has been chosen to be equal to $`1/3`$. Under these conditions the Einstein’s field equations become
$$\left(\frac{\dot{a}}{a}\right)^2+\left(\frac{\dot{a}}{a}\right)\left(\frac{\dot{b}}{b}\right)=\frac{2\kappa }{3}\rho _M,$$
(14)
$`2{\displaystyle \frac{\ddot{a}}{a}}+\left({\displaystyle \frac{\dot{a}}{a}}\right)^2=\kappa \gamma \rho _M{\displaystyle \frac{4}{3}}\kappa \eta ^{}\overline{\sigma },`$ (15)
and
$`{\displaystyle \frac{\ddot{b}}{b}}+{\displaystyle \frac{\ddot{a}}{a}}+{\displaystyle \frac{\dot{a}}{a}}{\displaystyle \frac{\dot{b}}{b}}=\kappa \gamma \rho _M+{\displaystyle \frac{2}{3}}\kappa \eta ^{}\overline{\sigma }.`$ (16)
This set of equations is similar to that of a matter fluid with shear viscosity immersed in a background corresponding to a flat axisymmetric cosmological model.
## III Solution to the field equations and some consequences
In the following we will describe solutions to the set of equations (14)–(16) in the cases in which $`\eta ^{}=\mathrm{\hspace{0.17em}0}`$, i.e., where there is not generation of entropy and $`\eta ^{}\mathrm{\hspace{0.17em}0}`$, where exist generation of entropy. For the former solutions we calculate the angular distance-redshifts relations (specifically, for the stiff model with $`\eta ^{}=0`$) which are compared with its analogous results corresponding to the flat spacetime, and for the latter, we calculate the generation of entropy (for the cases in which $`\eta ^{}0`$).
### A Cases $`\eta ^{}=\mathrm{\hspace{0.17em}0}`$
In this case we describe two possible solutions. One of these is the vacuum Kasner solution in which $`p_M=\rho _M=0`$, and a stiff fluid with equation of state $`p_M=\rho _M0`$, i.e., $`\gamma =1`$. In the former case it is found that $`a(t)=a_0`$ and $`b(t)=t`$ is possible solution to the field equations. Here, the scalar field $`Q`$ remains constant and thus $`\rho _Q=const.=V_0`$ and $`p_Q={\displaystyle \frac{1}{3}}V_0`$. On the other hand, $`\eta _M`$ becomes $`\eta _M={\displaystyle \frac{1}{2}}{\displaystyle \frac{V_0}{t}}`$. It seems that we could have others Kasner solution, such that $`a(t)=t^{2/3}`$ and $`b(t)=t^{1/3}`$. However, this sort of solution gives rise to a shear viscosity which is essentially negative, since $`\eta ={\displaystyle \frac{1}{2\kappa }}{\displaystyle \frac{1}{t^{1/3}}}`$. Therefore, we disregard this type of solution.
In the latter case, in which $`p_M=\rho _M`$, i.e., $`\gamma =1`$, we find a possible solution in which
$$a(t)=t^n\text{}b(t)=t^{12n},$$
(17)
where $`n`$ is a positive number. Here, we get
$$p_M=\rho _M=\frac{n(23n)}{\kappa t^2}$$
(18)
and
$$\eta _M=\frac{1}{2\kappa }\frac{1}{(13n)}\frac{1}{t^{2n1}}.$$
(19)
In order to obtain $`\eta >0`$ we demand $`n<1/3`$. Here, it is found that the scalar field growth as
$$Q(t)=\sqrt{(}\frac{2}{3\kappa })\frac{1}{1n}t^{1n},$$
(20)
and its corresponding potential is given by
$$V(Q)=\left(\frac{2^{2n1}}{3\kappa }\right)^{\frac{1}{1n}}Q^{\frac{2n}{1n}}.$$
(21)
Notice that this potential decreases when $`Q`$ increases, since $`n<1/3`$.
At this moment, we would like to calculate the luminosity distance , $`d_L(z)`$, as a function of the redshift $`z`$. This concept plays a crucial role in describing the geometry and matter content of the universe. From the metric (4) we observe that, light emitted by the an object of luminosity $``$ and located at the coordinate distance $`\theta `$, at a time $`t`$ is received by an observer (assumed located at $`\theta =0`$) at the time $`t=t_0`$. The time coordinates are related by the cosmological redshift $`z`$ in the $`\theta `$-direction by the expression, $`1+z=a(t_0)/a(t)a_0/a(t)`$. The luminosity flux reaching the observer is $`={\displaystyle \frac{}{4\pi d_L^2}}`$, where $`d_L`$ is the luminosity distance to the object, given by $`d_L(z)=a_0sin(\theta (z))(1+z)`$.
In order to obtain an explicit expression for the angular size, let us now consider an object aligned to the $`\varphi `$-direction and proper length $`l`$, so that its ”up” and ”down” coordinates are $`(\theta ,\varphi +\delta \varphi ,0)`$ and $`(\theta ,\varphi ,0)`$. The proper length of the object is obtained by setting $`t=const.`$ in the line-element metric (4), $`ds^2=l^2=a^2(t)sin^2(\theta )\delta \varphi ^2`$. Thus, the angular size becomes
$`\delta \varphi ={\displaystyle \frac{l}{d_L(z)}}(1+z)^2,`$ (22)
with $`d_L`$ defined above.
From the solutions represented by equation (17) we obtain for the angular size,
$$\delta \varphi _n=\frac{l}{a_0}\frac{\left(\mathrm{\hspace{0.17em}1}+z\right)}{sin\left[\frac{n}{1n}\frac{1}{a_0H_0}\left[\mathrm{\hspace{0.17em}1}\left(\mathrm{\hspace{0.17em}1}+z\right)^{\frac{1n}{n}}\right]\right]},$$
(23)
where $`H_0`$ is a parameter defined by $`H_0={\displaystyle \frac{n}{a_0^{1/n}}}`$.
Fig. 1 shows the angular size as a function of the redshift in the range $`0.05z2.80`$ for three different values of the parameters $`n`$. Here, we have used de value $`a_0H_0=\sqrt{2/5}`$. In this plot we have added the graph of the angular size corresponding to the isotropic closed FRW model, with a matter dominated by a quintessence component defined by $`\alpha =1/3`$. Notice that, for different values of the parameter $`n`$, those curves near to the value $`n=1/3`$ become closer to that corresponding of the isotropic FRW model.
In Fig. 2 we show the angular sizes as a function of the redshift $`z`$ for a flat and close anisotropic universe models. Notice that at low redshift both curves become similar. We could distinguish them at $`z0.5`$.
### B Cases $`\eta ^{}\mathrm{\hspace{0.17em}0}`$
As before, in this case we consider two different solutions. We start by describing a quasi-anisotropic and an exponential growing solutions.
In the former case, we found as a possible solution $`a(t)=t^{2/3}`$ and $`b(t)=t^{2/3}\left[1+(t/t_0)^n\right]`$, where $`n`$ and $`t_0`$ are two arbitrary constants. Notice that at large time $`b(t)`$ approach to $`a(t)`$ and thus the universe isotropizes. Thus, asymptotically the universe approaches to an homogeneous isotropic flat universe which is filled by dust, i.e., $`b(t)a(t)=t^{2/3}`$.
For this solution it is found that,
$$\rho _M=\frac{4}{3\kappa t^2}\left[1\frac{n}{1+\left(\frac{t}{t_0}\right)^n}\right],$$
(24)
$$p_M=\frac{2n(1n)}{3\kappa t^2\left[1+\frac{n}{1+\left(\frac{t}{t_0}\right)^n}\right]},$$
(25)
and
$$\eta __M=\frac{1}{2n\kappa t}\left[n\kappa (n1)+t^{2/3}\left[1+\frac{n}{1+\left(\frac{t}{t_0}\right)^n}\right]\right].$$
(26)
Notice that $`\gamma `$ becomes a time depending quantity in this case,
$$\gamma (t)=\frac{n}{2(1n)}\left[1n+\left(\frac{t}{t_0}\right)^n\right]^1.$$
(27)
Notice also that $`\gamma (t)0`$ for $`t\mathrm{}`$, in agreement with the remark described above.
The effective shear viscosity becomes $`\eta ^{}={\displaystyle \frac{n1}{2\kappa t}}`$, and in order to be positive the parameter $`n`$ should be bounded from below, i.e., $`n1`$.
The corresponding solution for the scalar field is
$$Q(t)=\sqrt{\frac{6}{\kappa }}\left(\frac{t}{t_0}\right)^{\frac{1}{3}}Q_0\left(\frac{t}{t_0}\right)^{\frac{1}{3}},$$
(28)
and the potential becomes
$$V(Q)=V_0\left(\frac{Q_0}{Q}\right)^4,$$
(29)
where the constant $`V_0`$ is given by $`V_0=\frac{1}{9}Q_0^3`$. This sort of solution was described in ref where scalar fields in FRW metric were studied.
A second possible solution is that in which the scale factor $`a`$ grows exponentially, i.e.,
$`a(t)=e^{Ht},`$ (30)
and
$`b(t)=e^{Ht/2}sin\left({\displaystyle \frac{3Ht}{2}}\right),`$ (31)
where $`H`$ is a constant to be determined later on.
This solution corresponds to a universe filled with a viscous dust, since $`p_M=0`$ and $`\eta ^{}0`$ at any time. The energy density and the effective shear viscosity become
$$\rho __M=3H^2cot\left(\frac{3Ht}{2}\right).$$
(32)
and
$$\eta ^{}=\frac{3H}{2\left[cot\left(\frac{3Ht}{2}\right)1\right]},$$
(33)
respectively.
In order to have $`\eta ^{}0`$ we must impose that $`0{\displaystyle \frac{3Ht}{2}}{\displaystyle \frac{\pi }{4}}`$. This result in an age for the universe given by $`t_0={\displaystyle \frac{\pi }{6}}H^1`$, which could be used for fixing the value of $`H`$.
Notice that the solutions (30) and (31) give rise to the following Hubble expansion rates,
$`H_1={\displaystyle \frac{\dot{a}}{a}}=H`$ (34)
and
$$H_2=\frac{\dot{b}}{b}=\frac{H}{2}\left[3cot\left(\frac{3Ht}{2}\right)1\right],$$
(35)
and thus the Hubble horizon related to the $`\theta `$-$`\varphi `$ plane remains constant.
The corresponding scalar field is found to evolve as
$`Q(t)={\displaystyle \frac{1}{H}}\sqrt{{\displaystyle \frac{2}{3\kappa }}}\left[e^{Ht_0}e^{Ht}\right]+Q_0,`$ (36)
where $`Q_0`$ is the value of $`Q(t)`$ at $`t=t_0`$. The corresponding scalar potential $`V(Q)`$ becomes
$`V(Q)=V_0\left[1\sqrt{{\displaystyle \frac{3\kappa }{2}}}(QQ_0)\right],`$ (37)
where $`V_0`$ is a constant defined by $`V_0={\displaystyle \frac{2}{3\kappa }}e^{2Ht_0}`$. From this expression we see that this potential decreases when $`Q`$ increases, similar to the other case.
It is well known that the production of entropy could be related to the anisotropy of the universe. In the following we proceed to calculate this production in the cases described above. In order to do this, we introduce the entropy current four-vector $`S^\mu `$ as
$$S^\mu =n__bk__B\lambda u^\mu ,$$
(38)
where as before $`u^\mu `$ represents the four-velocity, $`n__b`$ the baryon number density, $`k__B`$ is the Boltzmann’s constant and $`\lambda `$ the nondimensional entropy per baryon. It could be shown that
$$S_{;\mu }^\mu =\frac{2\eta }{T}\sigma _{\mu \nu }\sigma ^{\mu \nu },$$
(39)
where in the first case we get
$$S_{;\mu }^\mu =\frac{2n^2(n1)}{3\kappa Tt^3\left[1+(t/t_0)^n\right]^2}.$$
(40)
The left-hand side of this expression gives, in the comoving frame of reference
$$S_{;\mu }^\mu =k__Bn__b\dot{\lambda },$$
(41)
where we have used the conservation equation for baryon number, $`(n__bu^\mu )_{;\mu }=0`$. Thus, we get
$$\dot{\lambda }=\frac{2n^2(n1)}{3n_bk__BTt^3\left[1+(t/t_0)^n\right]^2}.$$
(42)
Note that this expression decrease when $`t`$ increase and becomes zero for $`t\mathrm{}`$, similar to $`\eta _M`$, $`p_M`$ and $`\rho _M`$.
From expression (42) evaluated at $`t=t_{_{1000}}`$ (equivalent to 1000 s) and $`t=t_{rec}`$ (time at recombination) we get
$`{\displaystyle \frac{\dot{\lambda }_{_{1000}}}{\dot{\lambda }_{rec}}}={\displaystyle \frac{n_b^{rec}}{n_b^{1000}}}{\displaystyle \frac{T_{rec}}{T_{_{1000}}}}\left({\displaystyle \frac{t_{rec}}{t_{_{1000}}}}\right)^3[{\displaystyle \frac{1+t_{rec}^n/t_0^n}{1+t_{_{1000}}^n/t_0^n}}.]^2`$ (43)
With the data given in ref. and taking $`n=2`$, we obtain for $`t_0`$ the value $`t_04\times 10^{10}`$ s. This value, together with the age of the universe, $`t_c5\times 10^{17}`$ s, allows us to obtain the ratio between the shear, $`\sigma `$, and the scalar expansion, $`\theta `$, given by
$`\left({\displaystyle \frac{\sigma }{\theta }}\right)_{t_c}={\displaystyle \frac{n}{\sqrt{3}(2(t_c/t_0)^n+2n)}}5\times 10^{15},`$ (44)
which is inside of the bound expressed by COBE measurements, that gives $`\left({\displaystyle \frac{\sigma }{\theta }}\right)_{t_c}6.9\times 10^{10}`$ .
In the second case, and following a similar process we find that
$$\dot{\lambda }=\frac{9H^3}{4n_bk__B\kappa T}\left[cot\left(\frac{3Ht}{2}\right)1\right].$$
(45)
By using the observational data specified above we get that
$$\frac{\dot{\lambda }_{_{1000}}}{\dot{\lambda }_{rec}}\mathrm{\hspace{0.17em}6}\times \mathrm{\hspace{0.17em}10}^{25}.$$
(46)
Thus, the generation of entropy has decreased more than $`10^{25}`$ times the value at recombination during the period from $`t_{rec}`$ to $`t1000s`$.
## IV Conclusions
We have studied an anisotropic universe cosmological model described by the metric (4). We included in our model negative anisotropic pressures motivated by quintessence cosmological scenarios. This component was represented by a scalar field $`Q`$, whose equation of state was considered to be given by $`p__Q=\alpha \rho __Q`$, where the parameter $`\alpha `$ was considered to be equal to $`1/3`$.
In order that our closed universe scenario could resemble a flat model, we imposed the conditions specified by equations (12) and (13). Under these conditions, we have determined, in different cases, explicit expressions for the scalar potential $`V(Q)`$. In all these cases we have found that these potential decrease as a function of the scalar field $`Q`$. In this respect, would be interesting to study the cosmological consequences that this sort of potential may have during the evolution of the universe. Specially, the influence that it carried during the rapid expansion (inflation) that the universe is believed to present at early time of its evolution.
In the cases in which the shear viscosity was vanished we have determined the angular sizes for different values of the parameters. Here, we found, similar to the isotropic case, that our closed model looks similar to a flat model at low redshifts.
On the other hand, solutions in which the shear viscosity was not vanished, we have determined the generation of entropy. Here, we have found that our results agree with the bound imposed by the observational data.
## Acknowledgments
MC was supported by COMICION NACIONAL DE CIENCIAS Y TECNOLOGIA through Grant FONDECYT N<sup>0</sup> 1990601, also by Dirección de Promoción y Desarrollo de la Universidad del Bío-Bío and in part by Dicyt (Universidad de Santiago de Chile). SdC was supported from the COMICION NACIONAL DE CIENCIAS Y TECNOLOGIA through Grant FONDECYT N<sup>0</sup> 1971157 and also from UCV-DGIP 123.744/99.
|
warning/0004/astro-ph0004005.html
|
ar5iv
|
text
|
# 3C radio sources as they’ve never been seen before
## 1. Introduction
Low-frequency radio surveys play a key role in selecting samples of radio sources, dominated by optically thin synchrotron emission, which are free of orientation biases. Examples of such samples, which have been pivotal in advancing our understanding of the nature of radio sources, are the celebrated 3C sample (revised by Laing, Riley and Longair 1983) and very recently the much fainter 7C sample (Rawlings et al. 1998). Following recent technical innovation at the VLA, it is now possible to make spatially well-resolved images of radio sources at low frequencies: at 74 MHz, images can now routinely be made with an angular resolution of 25<sup>′′</sup>. It is in this low-frequency regime that models of the energy supply to the lobes from the hotspots in classical doubles may be tested, and where their energy budgets may be investigated, for example by the detection or otherwise of steep-spectrum halos surrounding these objects.
## 2. 3C 84 and its halo
The 74 MHz image (Fig. 1, left) shows a low surface brightness halo surrounding 3C 84. The signal to noise of the image is insufficient to reveal whether the halo has a definite boundary, though this does appear to be hinted at in the 330 MHz image (see Fig. 1 right, and Burns et al 1992) and at 1.4 GHz \[Ger de Bruyn, priv. comm.\]. The spectrum of the halo is not particularly steep, averaging $`\alpha _{330}^{74}1.1`$ (defined such that the flux density $`S_\nu `$ at frequency $`\nu `$ is given by $`S_\nu \nu ^\alpha `$), with (e.g. the region 200<sup>′′</sup> west and 100<sup>′′</sup> south of the core) in places a spectral index as flat as 0.7. However, curious new features are seen (only) in the 74 MHz image: protrusions apparently emanating from the core region (at 2 o’clock and 6 o’clock) have very steep spectra ($`\alpha \stackrel{>}{}2)`$. It is possible that these could be outflows from the core, or merely static structures with very strange spectra. A full analysis of these images and those of the other objects observed will be presented in Kassim, Perley & Blundell (in prep).
## 3. 3C 129 and its twin tails
While in the inner few arcmin this radio source is a wide-angle tailed source with oppositely directed jets close to the core (see Rudnick & Burns 1981) on larger scales the jets follow each other quite closely and have the appearance of a narrow-angle tailed source (see Fig. 2 and Kassim et al. 1993). The integrated spectral index of both the entire source, and of just the extended tails, between 74 MHz and 327 MHz is $`\alpha 1.1`$.
## 4. 3C 219 and other classical doubles at low-frequency
Fig. 3 shows that the images of the classical double 3C 219 at 74 MHz and at 1.5 GHz are more remarkable for their similarities than for their differences. Just as in the cases of 3C 98 and 3C 390.3 we presented in Blundell et al. (1999b) there is no evidence of any extended emission at low-frequency which is not already seen at GHz frequencies; this appears to be the case for all the classical doubles imaged to date. Jenkins & Scheuer (1976) pointed out that if synchrotron cooling played a part in determining the spectral shapes of lobes, then lobes should be observed to extend further at low frequency than at high frequency. Our images thus suggest that the Lorentz factor particles responsible for the 74 MHz emission are entirely co-spatial with those responsible for the 1.4 GHz emission.
In a recent paper, Blundell & Rawlings (2000) discussed a contribution to spectral steepening along the lobe from the hotspot to the core which is separate from the traditionally assumed simple synchrotron cooling picture. This contribution is particularly important in explaining spectral gradients measured a decade below fitted break frequencies, for example that in Cygnus A by Kassim et al. (1996). This comes from a gradient in magnetic field causing different parts of the underlying curved energy distribution to be ‘illuminated’ (see figure 4 in Blundell & Rawlings 2000) in the different regions along the lobe.
We conclude by noting that low-frequency observations of 3C radio sources have yet to reveal extended emission associated with a classical double source beyond that seen at GHz frequencies. Halos around the rather more amorphous types of object are not uncommon and in the case of 3C 84 close to the core discrete regions with very steep spectra ($`\alpha \stackrel{>}{}2`$) have been discovered at 74 MHz. A full analysis of these data will appear in a forthcoming paper.
It is a pleasure to thank the conference organisers for the splendid way in which they organised the Pune meeting. K.M.B. thanks the Royal Commission for the Exhibition of 1851 for a Research Fellowship. Basic research in radio astronomy at the US Naval Research Laboratory is supported by the US Office of Naval Research. The VLA is a facility of the NRAO operated by Associated Universities, Inc., under co-operative agreement with the NSF.
## References
Blundell, K.M. & Rawlings, S., 2000, AJ, 119, 1111
Blundell, K.M., Rawlings, S. & Willott, C.J., 1999a, AJ, 117, 677
Blundell, K.M., Rawlings, S., Willott, C.J., Kassim, N.E. & Perley, R.A., 1999b, in ‘Lifecycles of Radio Galaxies’, New Astronomy Reviews eds J. Biretta et al. \[astro-ph/9910158\]
Burns, J.O., Sulkanen, M.E., Gisler, G.R. & Perley, R.A., 1992, ApJ, 388, L49
Jenkins, C.J. & Scheuer, P.A.G., 1976, MNRAS, 174, 327
Laing, R.A., Riley, J.M., & Longair, M.S., 1983, MNRAS, 204, 151
Kassim, N, Perley, R.A., Eriksen, W. & Dwarakanath, K.S., 1993, AJ, 106, 2218
Kassim, N.E., Perley, R.A., Carilli, C.L., Harris, D.E. & Erickson, W.C., 1996, in ‘Cygnus A – Study of a radio galaxy’, p182, eds Carilli & Harris
Rawlings, S., Blundell, K.M., Lacy, M., Willott, C.J. & Eales, S.A., in ‘Observational Cosmology with the new radio surveys’, eds M.N. Bremer, N. Jackson and I. Pérez-Fournon, Kluwer Academic Publishers, p171, (1998)
Rudnick, L. & Burns, J.O., 1981, ApJ, 246, L69
|
warning/0004/hep-ph0004039.html
|
ar5iv
|
text
|
# 1 𝛾(𝑇)/𝛾_{𝑝𝑒𝑟𝑡}(𝑇) as function of log{𝑔_{𝑇=0}}.
TUM-HEP-372/00
A consistent nonperturbative approach to thermal damping-rates<sup>1</sup><sup>1</sup>1Supported by the SFB 375 für Astroteilchenphysik der Deutschen Forschungsgemeinschaft
Bastian Bergerhoff<sup>2</sup><sup>2</sup>2bberger@physik.tu-muenchen.de and Jürgen Reingruber<sup>3</sup><sup>3</sup>3reingrub@physik.tu-muenchen.de
Institut für Theoretische Physik
Technische Universität München
James-Franck-Strasse, D-85748 Garching, Germany
## Abstract
We propose a nonperturbative scheme for the calculation of thermal damping-rates using exact renormalization group (RG)-equations. Special emphasis is put on the thermal RG where first results for the rate were given in . We point out that in order to obtain a complete result that also reproduces the known perturbative behaviour one has to take into account effects that were neglected in . We propose a well-defined way of doing the calculations that reproduces perturbation theory in lowest order but goes considerably beyond perturbative results and should be applicable also at second order phase-transitions.
Pacs-No.s: 11.10.Wx,11.15.Tk,05.10.Cc,05.70.Ln
Perturbation theory at finite temperatures is often invalidated by bad infrared behaviour. These problems can be solved by different resummation schemes. One of the most powerful approaches to resummation is the renormalization group of Wilson and others . This involves introducing an external scale and deriving functional differential equations for the dependence of generating functionals on this scale. The right hand side of such an RG-equation can formally be interpreted as a one-loop expression in the sense that it is of order $`\mathrm{}`$ compared to the left hand side. Nevertheless the Wilsonian RG constitutes a nonperturbative method and the RG-equations are exact functional relations. The approach is well known to correctly reproduce the infrared-behaviour of theories even at second-order phase-transitions.
Even though the RG-equations define a nonperturbative approach, it is possible to reproduce perturbation theory. This is straightforward for one-loop results, but becomes rather tedious if one goes to higher orders. Quantities which are well described perturbatively for a range of temperatures but for which perturbation theory fails e.g. at a phase-transition constitute a major challenge for RG-equations. This holds in particular for quantities where the one-loop contribution vanishes. The thermal damping-rate in $`\phi ^4`$-theory is an example for a quantity where the lowest order perturbative contribution is two-loop. If one aims at a reliable calculation of such quantities at all temperatures, one has to reproduce two-loop perturbation theory as the leading term and apply sensible nonperturbative resummations close to a possible phase-transition.
A recent formulation of Wilsonian RG-equations in the Schwinger-Keldysh (CTP-) formalism of real-time thermal field theory is particularily suited for calculations of nonstatic thermal Green-functions . It makes use of the fact that the propagators in the CTP-formalism separate into the usual zero-temperature and a finite-T part. The finite-T parts only contribute on-shell and depend only on the three-momenta. The cutoff modifies the thermal part and one introduces cut-off propagators of the form
$`D_\mathrm{\Lambda }(k)`$ $`=`$ $`\left(\begin{array}{cc}\mathrm{\Delta }& 0\\ 0& \mathrm{\Delta }^{}\end{array}\right)+(\mathrm{\Delta }\mathrm{\Delta }^{})\left(\begin{array}{cc}0& \mathrm{\Theta }(k_0)\\ \mathrm{\Theta }(k_0)& 0\end{array}\right)+`$ (8)
$`+(\mathrm{\Delta }\mathrm{\Delta }^{})\mathrm{\Theta }(|\stackrel{}{k}|,\mathrm{\Lambda })N(|k_0|)\left(\begin{array}{cc}1& 1\\ 1& 1\end{array}\right)`$
where $`\mathrm{\Theta }(|\stackrel{}{k}|,\mathrm{\Lambda })`$ is a possibly smeared out step function. We will in the following use a sharp cutoff, i.e. set $`\mathrm{\Theta }(|\stackrel{}{k}|,\mathrm{\Lambda })=\mathrm{\Theta }(|\stackrel{}{k}|\mathrm{\Lambda })`$. Thus for finite $`\mathrm{\Lambda }`$, the propagation of thermal modes with three-momemtum small compared to $`\mathrm{\Lambda }`$ (soft modes) is supressed, while the hard thermal modes are unmodified. Inserting this propagator into the usual expression for the generating functional $`Z[J]`$ one obtains a $`\mathrm{\Lambda }`$-dependent functional $`Z_\mathrm{\Lambda }`$. Introducing the modified Legendre-transform
$`\mathrm{\Gamma }_\mathrm{\Lambda }[\mathrm{\Phi }]=i\mathrm{ln}Z_\mathrm{\Lambda }[J]J\mathrm{\Phi }{\displaystyle \frac{1}{2}}\mathrm{\Phi }\left(D_\mathrm{\Lambda }\right)^1\mathrm{\Phi }`$ (9)
one readily obtains for the scale-dependence of $`\mathrm{\Gamma }_\mathrm{\Lambda }`$ the exact functional differential equation
$`\mathrm{\Lambda }{\displaystyle \frac{\mathrm{\Gamma }_\mathrm{\Lambda }[\mathrm{\Phi }]}{\mathrm{\Lambda }}}={\displaystyle \frac{i}{2}}\text{Tr}\mathrm{\Lambda }{\displaystyle \frac{D_\mathrm{\Lambda }^1}{\mathrm{\Lambda }}}\left(D_\mathrm{\Lambda }^1+{\displaystyle \frac{\delta ^2\mathrm{\Gamma }_\mathrm{\Lambda }[\mathrm{\Phi }]}{\delta \mathrm{\Phi }\delta \mathrm{\Phi }}}\right)^1`$ (10)
This is the thermal renormalization group equation (TRG) and will be the starting point of our discussion.
Being a differential equation, (10) of course has to be supplemented by boundary conditions. As discussed in , in the limit $`\mathrm{\Lambda }\mathrm{}`$, the effective action $`\mathrm{\Gamma }_\mathrm{\Lambda }`$ is trivially obtained from the $`(T=0)`$-effective action of the theory. Also, in the limit $`\mathrm{\Lambda }0`$ the full finite temperature CTP-effective action $`\mathrm{\Gamma }[\mathrm{\Phi }]`$ is obtained.
RG-equations for Green-functions are simply obtained from (10) by taking functional derivatives with respect to $`\mathrm{\Phi }`$. For the damping-rate, we are interested in the imaginary part of the two-point function. The flow-equation for the two-point function reads in terms of the thermal fields $`\phi _i`$ with $`i=1,2`$ ($`\phi _0`$ is some background configuration)
$`\mathrm{\Lambda }{\displaystyle \frac{}{\mathrm{\Lambda }}}{\displaystyle \frac{\delta ^2\mathrm{\Gamma }_\mathrm{\Lambda }}{\delta \phi _i\delta \phi _j}}|_{\phi _0}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\text{Tr}K_\mathrm{\Lambda }{\displaystyle \frac{\delta ^4\mathrm{\Gamma }_\mathrm{\Lambda }}{\delta \phi _i\delta \phi _j\delta \mathrm{\Phi }\delta \mathrm{\Phi }}}|_{\phi _0}`$ (11)
$`({\displaystyle \frac{1}{2}}\text{Tr}K_\mathrm{\Lambda }{\displaystyle \frac{\delta ^3\mathrm{\Gamma }_\mathrm{\Lambda }}{\delta \phi _i\delta \mathrm{\Phi }\delta \mathrm{\Phi }}}𝒟{\displaystyle \frac{\delta ^3\mathrm{\Gamma }_\mathrm{\Lambda }}{\delta \phi _j\delta \mathrm{\Phi }\delta \mathrm{\Phi }}}|_{\phi _0}+(ij))`$
where $`K_\mathrm{\Lambda }=i𝒟\mathrm{\Lambda }\frac{D_\mathrm{\Lambda }^1}{\mathrm{\Lambda }}𝒟`$ is the kernel of the TRG and $`𝒟=\left(D_\mathrm{\Lambda }^1+\mathrm{\Gamma }_\mathrm{\Lambda }^{(2)}\right)^1`$ is the full two-point function . Even though in this representation the flow-equation appears like a plain one-loop equation, due to the fact that all $`n`$-point functions involved are full $`n`$-point functions it is actually an exact, nonperturbative expression.
Let us then assume that we want to calculate the imaginary part of the thermal two-point function in a scalar theory with unbroken $`Z_2`$-symmetry. Perturbatively in this case the lowest contribution to the imaginary part of the self-energy occurs on two-loop level. As out in , in such a situation the contribution to the imaginary part of (11) has to come from an imaginary part of the full four-point function (by the $`Z_2`$-symmetry, the second contribution in (11) vanishes identically for all $`\mathrm{\Lambda }`$). We thus have to solve a coupled system of at least two flow-equations, (11) and a corresponding equation for the imaginary part of the four-point function.
First however let us make some remarks concerning the thermal indice which appear in real-time formulations of thermal field-theory . The physical fields always carry a thermal index 1 and one is in general interested in the calculation of Green-functions for those fields. Since the propagators are nondiagonal in the thermal indice, in the calculation of higher loop contributions one has to include vertice of the 2-fields nevertheless – neglecting those contributions yields singular expressions. If we use the TRG to compute Green-functions, we have to use full vertice according to (10). In this case, one has to allow for vertice with mixed thermal structure.
Furthermore, what enters the physical self-energy (and thus the damping-rate) is not the Green-function $`\frac{\delta ^2\mathrm{\Gamma }}{\delta \phi _1\delta \phi _1}`$, but rather a specific combination of the $`\frac{\delta ^2\mathrm{\Gamma }}{\delta \phi _i\delta \phi _j}`$ , known as $`\overline{\mathrm{\Pi }}(p)`$ in the literature. Assuming a Schwinger-Dyson equation for the full propagator it is easy to show that for the real- and imaginary parts of $`\overline{\mathrm{\Pi }}`$ the following relations hold:
$`\overline{\mathrm{\Pi }}(p)`$ $`=`$ $`{\displaystyle \frac{\delta ^2\mathrm{\Gamma }}{\delta \phi _1(p)\delta \phi _1(p)}}`$
$`\overline{\mathrm{\Pi }}(p)`$ $`=`$ $`ϵ(p_0)\left({\displaystyle \frac{\delta ^2\mathrm{\Gamma }}{\delta \phi _1(p)\delta \phi _1(p)}}+{\displaystyle \frac{\delta ^2\mathrm{\Gamma }}{\delta \phi _1(p)\delta \phi _2(p)}}\right)`$ (12)
($`ϵ(p_0)`$ is the sign function). $`\overline{\mathrm{\Pi }}`$ can also be obtained as
$`\overline{\mathrm{\Pi }}(p)`$ $`=`$ $`{\displaystyle \frac{1}{1+2N(|p_0|)}}{\displaystyle \frac{\delta ^2\mathrm{\Gamma }}{\delta \phi _1(p)\delta \phi _1(p)}}`$ (13)
Note that this expression explicitely involves a distribution function, whereas the second equation of (12) does not. We will below use (12) for a calculation of the physical self-energy.<sup>1</sup><sup>1</sup>1In , (13) was used for the calculation of the rate. We believe for a number of reasons that will become clear below that it is preferrable to use (12). Nevertheless, in principle for an exact solution of the TRG, the outcome should be identical. Another important feature of real-time thermal field-theories is that the effective action has the following symmetry
$`\mathrm{\Gamma }[\phi _1,\phi _2]=\mathrm{\Gamma }^{}[\phi _2^{},\phi _1^{}]`$ (14)
For real, momentum independent vertice (in the present theory with a real scalar field in configuration space), this gives the following relations:
$`\mathrm{\Gamma }_{i_1i_2\mathrm{}i_n}^{(n)}=\mathrm{\Gamma }_{\overline{i}_1\overline{i}_2...\overline{i}_n}^{(n)}`$ (15)
where $`\overline{i}=2`$ for $`i=1`$ and vice versa. Furthermore one may introduce a functional $`\overline{\mathrm{\Gamma }}[\varphi ]`$ by
$`{\displaystyle \frac{\delta \overline{\mathrm{\Gamma }}[\varphi ]}{\delta \varphi }}={\displaystyle \frac{\delta \mathrm{\Gamma }}{\delta \phi _1}}|_{\phi _1=\varphi ,\phi _2=\phi _2[\varphi ]}`$ (16)
where $`\phi _2[\varphi ]`$ is the solution of the field-equation for $`\phi _2`$. For momentum-independent vertice, it is easy to see that one has
$`\overline{\mathrm{\Gamma }}^{(3)}={\displaystyle \underset{i,j=1,2}{}}\mathrm{\Gamma }_{1ij}^{(3)}|_{\phi _a=\varphi };\overline{\mathrm{\Gamma }}^{(4)}={\displaystyle \underset{i,j,k=1,2}{}}\mathrm{\Gamma }_{1ijk}^{(4)}|_{\phi _a=\varphi }`$ (17)
and so on. These relations trivially generalize to $`\mathrm{\Gamma }_\mathrm{\Lambda }`$ and we will make use of them below.
Let us now turn to the calculation of the imaginary part of the self-energy of (at $`T=0`$) massless $`\phi ^4`$-theory. In order to obtain an equation for the scale-dependence of $`\overline{\mathrm{\Pi }}`$, we plug the flow-equations for the two-point functions with 1-1 and 1-2 external legs into the second relation in (12). We now make a crucial approximation: We neglect the imaginary part of the self-energy on the right hand side of all flow-equations (for the present work we also assume $`\overline{\mathrm{\Pi }}`$ to be momentum-independent, which is however not crucial for the results discussed and could be straightforwardly improved). This may be viewed as a ”quasiparticle-approximation”. As we will see below, it does not influence the leading behaviour for small couplings. Making this approximation however means that we will not be resumming imaginary parts in the calculations – a fact important to keep in mind.
In this case, the kernel takes the form
$`K_\mathrm{\Lambda }(k)`$ $`=`$ $`2\pi \delta (k^2m_\mathrm{\Lambda }^2)\mathrm{\Lambda }\delta (|\stackrel{}{k}|\mathrm{\Lambda })N(|k_0|)\left(\begin{array}{cc}1& 1\\ 1& 1\end{array}\right)`$ (20)
($`m_\mathrm{\Lambda }`$ is the real part of the self-energy) and is purely real. It is thus immediatly clear that all contributions to the flow of $`\overline{\mathrm{\Pi }}`$ are from the imaginary part of the full four-point function. Also, due to the specific combination of $`\mathrm{\Gamma }_{\mathrm{\Lambda },11}^{(2)}`$ and $`\mathrm{\Gamma }_{\mathrm{\Lambda },12}^{(2)}`$ appearing in (12), taking the trace over thermal indice in the flow-equation for $`\overline{\mathrm{\Pi }}`$ leaves us with the following result:
$`\mathrm{\Lambda }{\displaystyle \frac{\overline{\mathrm{\Pi }}_\mathrm{\Lambda }(p)}{\mathrm{\Lambda }}}={\displaystyle \frac{\mathrm{\Lambda }^3}{4\pi ^2}}{\displaystyle \frac{N(\omega _\mathrm{\Lambda })}{\omega _\mathrm{\Lambda }}}ϵ(p_0){\displaystyle \underset{i,j,k=1,2}{}}\mathrm{\Gamma }_{\mathrm{\Lambda },1ijk}^{(4)}(p,p,Q_\mathrm{\Lambda },Q_\mathrm{\Lambda })`$ (21)
In (21) we use the notation
$`\omega _\mathrm{\Lambda }=\sqrt{\mathrm{\Lambda }^2+m_\mathrm{\Lambda }^2},Q_\mathrm{\Lambda }=(\omega _\mathrm{\Lambda },|\stackrel{}{Q_\mathrm{\Lambda }}|=\mathrm{\Lambda })`$ (22)
For the quantities appearing on the right hand side of (21), we again need the RG-equations governing their scale-dependence. Let us first discuss the flow-equation for the coupling appearing in (21). We use the notation
$`{\displaystyle \underset{i,j,k=1,2}{}}\mathrm{\Gamma }_{\mathrm{\Lambda },1ijk}^{(4)}(p,p,q,q)=\overline{\mathrm{\Gamma }}_\mathrm{\Lambda }^{(4)}(p,p,q,q)`$ (23)
which is suggested by (17). We now make a further approximation, which is on equal footing with the quasiparticle-approximation done above: We neglect the imaginary parts of all couplings on the right hand side of the flow-equation for $`\overline{\mathrm{\Gamma }}_\mathrm{\Lambda }^{(4)}`$. This should of course not be done in (21), since there one would disregard the leading (and in fact only) contribution. Within this approximation we find:
$`\mathrm{\Lambda }{\displaystyle \frac{}{\mathrm{\Lambda }}}\overline{\mathrm{\Gamma }}_\mathrm{\Lambda }^{(4)}(p,p,q,q)`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Lambda }}{2}}\left(\overline{\mathrm{\Gamma }}_\mathrm{\Lambda }^{(4)}\right)^2{\displaystyle }{\displaystyle \frac{d^4l}{(2\pi )^2}}\delta (l^2m_\mathrm{\Lambda }^2)\theta (|\stackrel{}{l}|^2+m_\mathrm{\Lambda }^2)\delta (|\stackrel{}{l}|\mathrm{\Lambda })N(|l_0|)\times `$ (24)
$`\times {\displaystyle \underset{Q=\pm q}{}}\delta ((l+p+Q)^2m_\mathrm{\Lambda }^2)ϵ(l_0+p_0+Q_0)`$
Note that (24) is a completely well-defined expression. Furthermore the vertex appearing here is just the combination defined in (17), and we have taken into account vertice with both 1- and 2-external legs. This is only possible for the specific sum in (23). Neglecting vertice with 2-legs within the present approach would yield ill-defined products of distributions. We have checked that the inclusion of trilinear couplings in the present approximations (i.e. considering a theory with broken symmetry at low temperatures) is possible along the same line of arguments. Thus the results depicted in (21) and (24), together with a flow-equation for the real part of the effective action at constant fields, constitute a well-defined nonperturbative method for the calculation of thermal rates. This is the first important point in the present work.
Even though the approach using Wilsonian RG-equations is nonperturbative one wants to recover the leading perturbative results in situations where perturbation theory is valid. This is a nontrivial problem, since $`\overline{\mathrm{\Pi }}`$ in the theory under study vanishes to one-loop order.
The loop expansion can be reconstructed from Wilsonian RG-equations iteratively, making use of the fact that the right hand side of (10) is down by a factor of $`\mathrm{}`$ compared to the left hand side. In order to obtain a result for $`\overline{\mathrm{\Pi }}`$ to order $`\mathrm{}^2`$, we need the right hand side of (21) to order $`\mathrm{}`$. The quantities appearing on the right hand side of (21) are the thermal mass $`m_\mathrm{\Lambda }`$ and the imaginary part of the four-point function at finite $`\mathrm{\Lambda }`$. This imaginary part is a pure quantum effect and thus is itself of order $`\mathrm{}`$, so we need not consider corrections to the mass.
In order to find the imaginary part of the four-point function (which has to be calculated for the specific choice of momenta entering in (21) and at nonvanishing external scale $`\mathrm{\Lambda }`$), we turn to the corresponding flow-equation (24). To order $`\mathrm{}`$ we may neglect all loop corrections to the quantities appearing on the right hand side of (24). Thus to this order the couplings and masses are $`\mathrm{\Lambda }`$-independent and the integration of (24) may be performed noting that $`\delta (|\stackrel{}{l}|\mathrm{\Lambda })=_\mathrm{\Lambda }\theta (|\stackrel{}{l}|\mathrm{\Lambda })`$. One finds
$`\overline{\mathrm{\Gamma }}_\mathrm{\Lambda }^{(4)}(p,p,q,q)=\overline{\mathrm{\Gamma }}_{\mathrm{\Lambda }=\mathrm{}}^{(4)}(p,p,q,q)+`$
$`+{\displaystyle \frac{1}{2}}\left(\overline{\mathrm{\Gamma }}^{(4)}\right)^2{\displaystyle }{\displaystyle \frac{d^4l}{(2\pi )^2}}\delta (l^2m^2)\theta (|\stackrel{}{l}|^2+m^2)\theta (|\stackrel{}{l}|\mathrm{\Lambda })N(|l_0|)\times `$
$`\times {\displaystyle \underset{Q=\pm q}{}}\delta ((l+p+Q)^2m^2)ϵ(l_0+p_0+Q_0)`$ (25)
Plugging this result into the right hand side of the flow-equation for $`\overline{\mathrm{\Pi }}`$ and performing the remaining $`\mathrm{\Lambda }`$-integration should then give the desired imaginary part of the self energy to order $`\mathrm{}^2`$.
The result of a perturbative calculation may be taken e.g. from where one finds
$`\overline{\mathrm{\Pi }}(p)=\pi ϵ(p_0){\displaystyle \frac{\left(\overline{\mathrm{\Gamma }}^{(4)}\right)^2}{6}}{\displaystyle }{\displaystyle \frac{d^3\stackrel{}{k}}{(2\pi )^3}}{\displaystyle \frac{d^3\stackrel{}{q}}{(2\pi )^3}}{\displaystyle \frac{1}{8\omega _k\omega _q\omega _r}}\times `$
$`\times \{[\delta (p_0+\omega _k+\omega _q+\omega _r)\delta (p_0\omega _k\omega _q\omega _r)]\times `$
$`\times \left[1+N_k+N_q+N_r+N_kN_q+N_kN_r+N_qN_r\right]+`$
$`+\left[\delta (p_0+\omega _k+\omega _q\omega _r)\delta (p_0\omega _k\omega _q+\omega _r)\right]\left[N_r+N_kN_r+N_qN_rN_kN_q\right]+`$
$`+\left[\delta (p_0+\omega _k\omega _q+\omega _r)\delta (p_0\omega _k+\omega _q\omega _r)\right]\left[N_q+N_kN_q+N_rN_qN_kN_r\right]+`$
$`+[\delta (p_0\omega _k+\omega _q+\omega _r)\delta (p_0+\omega _k\omega _q\omega _r)][N_k+N_kN_q+N_kN_rN_qN_r]\}`$ (26)
where again $`\omega _p=\sqrt{\stackrel{}{p}^2+m^2}`$ and $`\stackrel{}{r}=\stackrel{}{k}+\stackrel{}{q}+\stackrel{}{p}`$. $`N_p`$ is the Bose-distribution with energy $`\omega _p`$. The part $`1`$ is of course the $`(T=0)`$-imaginary part. In the present method this part enters via the boundary conditions and is not calculable within the TRG. The other parts however have to come out of our result from (21) and (24).
An important observation at this point is that for the one-loop contribution in (25) one has two parts, one being the boundary part at $`\mathrm{\Lambda }=\mathrm{}`$ which is given by the one-loop $`(T=0)`$-result and the second one being proportional to $`N`$ and representing the contribution from thermal fluctuations. If we simply start with the tree-level effective action at $`T=0`$ – that is set the imaginary part of the four point function to 0 for $`\mathrm{\Lambda }\mathrm{}`$ – and plug the remaining contribution from (25) into (21), we only reproduce the contributions to $`\overline{\mathrm{\Pi }}`$ which are bilinear in the distribution functions. Dropping all $`(T=0)`$-quantum contributions for the boundary value of the flow of $`n`$-point functions is however routinely done in applications of the TRG (). Since the thermal damping-rate vanishes to one-loop, one does not reproduce the leading order perturbative result (26) and will not be able to give quantitatively reliable results for all temperatures.
To end the discussion of the perturbative result, we thus need the $`\mathrm{\Lambda }\mathrm{}`$ or $`(T=0)`$-value of the imaginary part of the four-point function to one loop. This is easily found to be
$`\overline{\mathrm{\Gamma }}_{\mathrm{\Lambda }=\mathrm{}}^{(4)}(p,p,q,q)`$ $`=`$ $`{\displaystyle \underset{Q=\pm q}{}}{\displaystyle \frac{\pi }{8}}\left(\overline{\mathrm{\Gamma }}_{T=0}^{(4)}\right)^2{\displaystyle }{\displaystyle \frac{d^3\stackrel{}{k}}{(2\pi )^3}}{\displaystyle \frac{1}{\omega _k\omega _{k+p+Q}}}\times `$ (27)
$`\times \left[\delta (p_0+Q_0\omega _k\omega _{k+p+Q})\delta (p_0+Q_0+\omega _k+\omega _{k+p+Q})\right]`$
Using this result in (25) and plugging the resulting expression for $`\overline{\mathrm{\Gamma }}_\mathrm{\Lambda }^{(4)}(p,p,Q_\mathrm{\Lambda },Q_\mathrm{\Lambda })`$ into the flow-equation for $`\overline{\mathrm{\Pi }}(p)`$ given in (21) one may indeed do the integration with respect to $`\mathrm{\Lambda }`$ to find for $`\mathrm{\Lambda }=0`$ the perturbative result (26) up to its $`(T=0)`$-part.<sup>2</sup><sup>2</sup>2In order to obtain the explicit form given in (26), symmetrization in the three-momenta is necessary. The calculation is however straightforward.
In order to go beyond the leading order two-loop result, we will take into account the thermal renormalization of the real parts of both the self-energy as well as the momentum-independent parts of the vertice. We perform a derivative expansion of the full effective action $`\overline{\mathrm{\Gamma }}_\mathrm{\Lambda }`$ and approximate it by the effective potential and a standard kinetic term. This approximation is discussed in detail in where also the resulting flow-equations may be found. We solve the flow-equation for the effective potential as discussed in and thus resum the momentum-independent (thermal) parts of $`n`$-point functions with an arbitrary number of external legs – this resummation goes considerably beyond the usual daisy- or superdaisy-resummations. It can be systematically improved by relaxing the derivative-approximation.
As the boundary condition for the flow of the effective potential we will use the tree-level effective potential of a massless $`\phi ^4`$-theory, $`U_{\mathrm{\Lambda }=\mathrm{}}=\frac{g_{T=0}}{24}\varphi ^2`$. This amounts to neglecting all $`(T=0)`$-loop corrections to the effective potential. Together with the boundary conditions and flow-equations for the imaginary parts as discussed above we thus have a closed system of differential equations<sup>3</sup><sup>3</sup>3Note that for the solution of the flow-equation for $`\overline{\mathrm{\Gamma }}_\mathrm{\Lambda }^{(4)}(p,p,q,q)`$, the external momentum $`q`$ is not given by (22) with the running values of $`\mathrm{\Lambda }`$ and $`\omega _\mathrm{\Lambda }`$. Instead it has to be considered fixed for some independent combination $`(\mathrm{\Lambda }^{},\omega _\mathrm{\Lambda }^{})`$. which may be solved numerically.
The resulting imaginary part of the self-energy is connected to the thermal damping-rate $`\gamma (T)`$ through
$`\gamma (T)={\displaystyle \frac{\overline{\mathrm{\Pi }}_{\mathrm{\Lambda }=0}(p_0=m_T,\stackrel{}{p}=0)}{2m_T}}`$ (28)
Perturbatively, using daisy-resummed perturbation theory one finds for a massless theory
$`\gamma _{pert}(T)={\displaystyle \frac{g_{T=0}^{3/2}}{64\sqrt{24}\pi }}\sqrt{1{\displaystyle \frac{3g_{T=0}^{1/2}}{\sqrt{24}\pi }}}T`$ (29)
In figure 1 we display the ratio of the result obtained using the flow-equations and (29) as a function of $`g_{T=0}`$. Indeed the TRG reproduces daisy-resummed perturbation theory for $`g_{T=0}0`$, even though the leading perturbative result is two-loop. For finite values of the coupling, the different resummations yield different results. Let us again point out that the result of the renormalization group-calculations performed here can be understood as a fully resummed result in leading order in an expansion in the anomalous dimension and the quasiparticle approximation. It goes beyond the resummations used in the literature (it in particular trivially includes the daisy- and superdaisy-schemes ).
Figure 2 shows the dependence of the real- and imaginary parts of the self-energy on the external scale $`\mathrm{\Lambda }`$ at fixed $`g_{T=0}`$. We note the typical behaviour: Small corrections to the $`(T=0)`$-values for large $`\mathrm{\Lambda }/T`$ due to Boltzmann-suppression of the thermal fluctuations and saturation of the values for $`\mathrm{\Lambda }m_\mathrm{\Lambda }`$. The limit $`\mathrm{\Lambda }0`$ is completely safe.
We close this letter by commenting on one of the most important qualitative points made in , namely that a calculation of the damping-rate from the TRG reproduces critical slowing down, i.e. the fact that the rate vanishes as the critical temperature of a theory with (at $`T=0`$) spontaneously broken $`Z_2`$-symmetry is approached. We believe that even though this conclusion was reached on the basis of an incomplete calculation it still holds if one uses the consistent approach laid out in the present paper. The reason why this should be the case is easy to see: The rate is an on-shell quantity with vanishing external momentum and as the critical temperature of the (second-order) phase-transition is approached, the external energy thus vanishes. On the other hand, in the framework of Wilsonian RG-equations internal propagators are never massless for any $`\mathrm{\Lambda }>0`$. Close to the phase-transition, the masses and coupling exhibit three-dimensional scaling behaviour, i.e. the renormalized mass and coupling vanish as
$`m_T(TT_c)^\nu ;g_T(TT_c)^\nu `$ (30)
In such a case, the limit $`m_T0`$ should be completely regular and the scaling arguments also given in should still hold: The perturbative result for the imaginary part of the self-energy behaves for small $`m_T`$ as $`\overline{\mathrm{\Pi }}(m_T,0)g_{T=0}^2\mathrm{ln}m_T`$ and thus one has for the rate
$`\gamma (T){\displaystyle \frac{g_{T=0}^2}{m_T}}\mathrm{ln}m_T`$ (31)
Consistent resummation using the Wilsonian RG replaces the coupling $`g_{T=0}`$ with the renormalized coupling $`g_T`$ and one obtains with $`t=TT_c`$
$`\gamma (TT_c)t^\nu \mathrm{ln}t0`$ (32)
for positive $`\nu `$ (in the present model, $`\nu `$ is found to be $`0.63`$ ).
Within the scheme presented here we now have for the first time a complete approach that allows us to study linear-response and static quantities in a nonperturbative manner also in the critical regime, while reproducing the perturbative results where they are valid. As shown in , the TRG can also be formulated for theories involving fermionic degrees of freedom. It should be very interesting to apply this nonperturbative method e.g. to questions related to the physics of nuclear matter as the chiral phase-transition is approached.
Acknowledgements: We thank M. Pietroni for helpful discussions on his results.
|
warning/0004/astro-ph0004121.html
|
ar5iv
|
text
|
# Centrifugal buoyancy as a mechanism for neutron star glitches
## 1 Introduction.
The ultimate motivation for this article is the problem of explaining one of the salient observational features of isolated (non-binary) pulsars, which is that comparatively long periods of continuous “spin down” of the observed frequency $`\mathrm{\Omega }`$ are occasionally interrupted by small “glitches”. Such a glitch consists of a sudden small increase, $`\delta \mathrm{\Omega }`$ say, that partially cancels the continuous negative variation $`\mathrm{\Delta }\mathrm{\Omega }`$ that has been accumulated since the preceding glitch.
Since very soon after its discovery in 1968, it has been generally agreed that the pulsar phenomenon is attributable to a strong magnetic field anchored in the outer crust layers of a central neutron star. The observed frequency $`\mathrm{\Omega }`$ is to be interpreted as the rotation frequency of the outer crust layer, whose continuous spin down is evidently due to the continuous decrease of the angular momentum $`J`$ due to radiation from the external magnetosphere. After thirty years of work, two basic problems remain.
The first is to account for the spectrum (from radio to X-ray and beyond) and the detailed pulse structure of the radiation, which are presumed to depend on the still very poorly understood workings of the magnetosphere.
The second problem – the one with which the present article is concerned – is to account for the frequency “glitches”. It is generally recognised that the glitches must be explained in terms of what goes on in the interior of the neutron star, and it is also generally believed that the glitch phenomenon is essentially related to the property of solidity that is predicted (on the basis of simple, generally accepted theoretical considerations) to characterise the crust of the neutron star after it has fallen below the relevant extremely high melting temperature, which occurs very soon after its formation.
The purpose of this article is to draw attention to the potential importance, as a mechanism for glitches, of the stresses induced in the crust just by the effective force arising from the deficit of centrifugal buoyancy that will be present whenever there is differential rotation.
It is to be noticed that centrifugal buoyancy is a phenomenom that has been previously considered in the context of neutron stars, at least with reference to one of its possible consequences, namely Ekman pumping. This is a mechanism that can considerably shorten the timescale needed for the redistribution of angular momentum (in comparison with viscous diffusion characterized by the timescale given by $`\tau _{\mathrm{visc}}R_{}^{\mathrm{\hspace{0.17em}2}}/\nu _{}`$ where $`\nu _{}`$ is the typical kinetic viscosity coefficient and $`R_{}`$ is the relevant stellar radial length scale) and thus the damping of differential rotation in cases for which (as will be the case in a typical pulsar) the star is rotating fast enough for the corresponding rotation timescale $`\tau _{\mathrm{rot}}=2\pi /\mathrm{\Omega }`$ to be short compared with $`\tau _{\mathrm{visc}}`$. In such circumstances, “Ekman pumping” will supplement the very slow diffusive transport by more rapid convective transport propelled by centrifugal buoyancy forces. The ensuing “Ekman timescale” $`\tau __\mathrm{E}`$ for the effective damping of differential rotation in such cases will be given roughly by the geometric mean of the pure diffusion and rotation timescales, i.e. $`\tau __\mathrm{E}\sqrt{\tau _{\mathrm{rot}}\tau _{\mathrm{visc}}}`$.
While it has been recognized that either Ekman pumping or magnetic coupling is in general efficient to bring into corotation the core plasma with the crust , it is expected that Ekman pumping is quite inefficient (see e.g. ) for the uncharged crust neutron superfluid that is believed (see e.g. ) to permeate the lower layers of the crust in the density range from $`10^{11}`$ to about $`10^{14}`$ gm/cm<sup>3</sup>. This means that the convectively accelerated Ekman timescale, $`\tau __\mathrm{E}R_{}\sqrt{2\pi /\nu _{}\mathrm{\Omega }}`$, is too long to prevent the development of significant differential rotation. The negligibility, in such cases, of Ekman pumping is attributable to the effective negligibility of viscosity, but should not be construed as implying the negligibility of centrifugal buoyancy forces. In previous discussions of such scenarios – and in particular of the simplified strictly stationary limit in which the effective viscosity is neglected, so that no possibility of Ekman pumping can arise at all – the role of centrifugal buoyancy forces has been rather generally overlooked. The upshot of the present investigation of stationary differentially rotating configurations is to show that in such cases the general neglect of the centrifugal buoyancy effect is quite unjustified, and that on the contrary this effect is potentially capable by itself of providing the dominant contribution to the crust stresses that are ultimately released in “glitches”.
## 2 Glitches driven by the spheroidality mechanism.
In the first years after the problem of accounting for neutron star glitches was posed, attention was concentrated on what is describable as the “spheroidality mechanism” . This mechanism depends on the supposition that the solidity forces will not be strong enough to allow the stellar equilibrium configuration to differ very much from a perfectly fluid equilibrium state, which would be spherical in the absence of rotation, but which will actually have the form of an oblate spheroid with ellipticity proportional to $`\mathrm{\Omega }^2`$. The moment of inertia, defined as the ratio $`I=J/\mathrm{\Omega }`$, will be given for a slowly rotating fluid by an expression of the form
$$I=I_0\left(1+\mathrm{\Omega }^2/\mathrm{\Omega }_{}^{\mathrm{\hspace{0.17em}2}}\right),$$
(1)
where $`I_0`$ is its spherical limit value and $`\mathrm{\Omega }_{}`$ is a constant characterising the rather high angular frequency needed for relative deviations from spherical symmetry to be of order unity. A more accurate formula involving higher order corrections would be needed for a star with angular velocity near the critical value $`\mathrm{\Omega }^2\mathrm{\Omega }_{}^{\mathrm{\hspace{0.17em}2}}`$, but the cases in which glitches have been observed so far are all characterised by
$$\mathrm{\Omega }^2\mathrm{\Omega }_{}^{\mathrm{\hspace{0.17em}2}}.$$
(2)
For a perfectly fluid star model, a continuous angular momentum variation $`\mathrm{\Delta }J<0`$ would bring about a corresponding momentum of inertia variation $`\mathrm{\Delta }I`$ that would be given by
$$\frac{\mathrm{\Delta }I}{I}2\frac{\mathrm{\Omega }^2}{\mathrm{\Omega }_{}^{\mathrm{\hspace{0.17em}2}}}\frac{\mathrm{\Delta }\mathrm{\Omega }}{\mathrm{\Omega }}<0.$$
(3)
Due to the solidity of the crust, which tends to preserve the more highly elliptic initial configuration, the actual change in the moment of inertia will fall short of what is predicted by this formula, but at some stage the strain will build up to the point at which the solid structure will break down (see Fig. 1) . It is predicted that there will then be a “crustquake” in which the solid structure suddenly changes towards what the perfect fluid structure would have been, thereby changing the moment of inertia by an amount
$$\delta I=\epsilon \mathrm{\Delta }I,$$
(4)
where $`\epsilon `$ is an efficiency factor that should presumably lie somewhere in the range
$$0<\epsilon \mathrm{}<1.$$
(5)
Since the amount of angular momentum loss during the very short duration of the glitch will be negligible, the corresponding discontinuous angular velocity change will be given by
$$\frac{\delta \mathrm{\Omega }}{\mathrm{\Omega }}=\frac{\delta I}{I}.$$
(6)
Its value will therefore be expressible in terms of the order of unity efficiency factor $`\epsilon `$ by
$$\delta \mathrm{\Omega }=2\epsilon \frac{\mathrm{\Omega }^2}{\mathrm{\Omega }_{}^{\mathrm{\hspace{0.17em}2}}}\mathrm{\Delta }\mathrm{\Omega },$$
(7)
in which it is to be recalled that $`\mathrm{\Delta }\mathrm{\Omega }`$ denotes the continuous (negative) change in angular velocity since the preceding glitch. This mechanism must presumably operate, and may account for some observed glitches, but it soon became clear that even if this mechanism is maximally efficient, with
$$\epsilon 1,$$
(8)
the magnitude predicted by (7) is much too low for such a mechanism to be able to account for the comparatively large glitches that are frequently observed in cases such as that of the Vela pulsar.
## 3 Glitches driven by differential rotation.
Soon after the empirical discovery of glitches too large to be accounted for by the “spheroidality” mechanism, it came to be recognised by theorists that a plausible explanation involved the superfluid property of the deeper layers of sufficiently cool neutron stars. This property makes it possible to conceive that an interior neutron superfluid layer with moment of inertia, $`I_\mathrm{n}`$ say, can rotate with an angular velocity, $`\mathrm{\Omega }_\mathrm{n}`$ say, that may differ from the externally observable angular velocity $`\mathrm{\Omega }`$ that characterises the part of the star that corotates with the crust, with its own moment of inertia
$$I_\mathrm{c}=II_\mathrm{n}.$$
(9)
In such a case it can be supposed that when an external braking mechanism causes the corotating crust component to undergo an angular velocity change $`\mathrm{\Delta }\mathrm{\Omega }`$, the angular velocity $`\mathrm{\Omega }_\mathrm{n}`$ of the independently rotating neutron superfluid layer may in the short run be unaffected, with negligible variation expressible by
$$\mathrm{\Delta }\mathrm{\Omega }_\mathrm{n}=0,$$
(10)
but that, when the ensuing angular velocity difference between the corotating crust component and the neutron superfluid layer exceeds some critical value there will be a discontinuous adjusment whereby this angular velocity difference is reduced by some process involving a transfer of angular momentum between the two components. Such a process will evidently entail a negative adjustment $`\delta \mathrm{\Omega }_\mathrm{n}`$ of the angular velocity of the neutron superfluid layer and an accompanying positive adjustment $`\delta \mathrm{\Omega }`$ of the (observable) angular velocity of the corotating crust component, whereby the latter increases its angular momentum by an amount $`I_\mathrm{c}\delta \mathrm{\Omega }`$ that is equal to the amount $`I_\mathrm{n}\delta \mathrm{\Omega }_\mathrm{n}`$ that is lost by the neutron superfluid component, so that the total angular momentum change during the discontinuous ‘glitch’ process is zero, i.e.
$$I_\mathrm{c}\delta \mathrm{\Omega }+I_\mathrm{n}\delta \mathrm{\Omega }_\mathrm{n}=0.$$
(11)
If this adjustment process were a hundred per cent efficient, the net variation $`\mathrm{\Delta }\mathrm{\Omega }+\delta \mathrm{\Omega }`$ of the corotating crust angular velocity would be exactly matched by the net neutron superfluid angular velocity variation, which by (10) will be simply given by $`\delta \mathrm{\Omega }_\mathrm{n}`$, so that one would have
$$\delta \mathrm{\Omega }_\mathrm{n}=\epsilon (\mathrm{\Delta }\mathrm{\Omega }+\delta \mathrm{\Omega }),$$
(12)
with $`\epsilon 1`$. In practice one would expect that there would typically be an incomplete adjustment, still expressible by a relation of the form (12), but with an efficiency factor $`\epsilon `$ having some lower value in the range (5). By substituting (11) in (12) it can be seen that the observable glitch magnitude will be given by
$$\delta \mathrm{\Omega }=\frac{\epsilon I_\mathrm{n}\mathrm{\Delta }\mathrm{\Omega }}{I_\mathrm{c}+\epsilon I_\mathrm{n}},$$
(13)
and hence, by (9), that for an efficiency factor $`\epsilon `$ with any value in the range (5) the glitch magnitude will satisfy the inequality
$$\delta \mathrm{\Omega }\mathrm{}>\epsilon \frac{I_\mathrm{n}}{I}\mathrm{\Delta }\mathrm{\Omega }.$$
(14)
By comparing (14) with (7), it can be seen that, for a given assumed value of the efficiency factor $`\epsilon `$, the differential rotation adjustment mechanism characterised by (14) can give rise to a much larger glitch magnitude $`\delta \mathrm{\Omega }`$ than is possible by the spheroidality adjusment mechanism characterised by (7), because the factor $`I_\mathrm{n}/I`$ in (14) can be of order unity, whereas the corresponding factor in (7), namely $`2\mathrm{\Omega }^2/\mathrm{\Omega }_{}^{\mathrm{\hspace{0.17em}2}}`$ is very small compared to unity in even the most rapidly rotating pulsars. Thus, unlike the spheroidality mechanism, mechanisms involving angular momentum transfer between differentially rotating components can plausibly be considered as candidates for explaining the frequent large glitches observed in the Vela pulsar.
## 4 Glitch mechanisms due to the vortices.
In the context of a glitch due to differential rotation, the question that arises is what physical mechanism can increase the effective coupling between the superfluid component and the crust, in order to generate a transfer of angular momentum.
The explanations that exist in the literature are based on an important property of a superfluid neutron star, which we have not yet mentioned in this article: the existence of an array of vortex lines in the rotating neutron superfluid component, each vortex carrying a quantum of vorticity $`\kappa =h/(2m_\mathrm{n})`$ (where $`m_\mathrm{n}`$ is the neutron mass). The vortex number density (per unit area) $`n__\mathrm{V}`$ is directly related to the superfluid angular velocity $`\mathrm{\Omega }_\mathrm{n}`$ by the expression
$$n__\mathrm{V}=\frac{2\mathrm{\Omega }_\mathrm{n}}{\kappa }$$
(15)
(for uniform rotation).
The kind of angular momentum transfer mechanism that has for many years been generally considered to offer the most likely explanation for large glitches is based on the supposition that these vortices will be “pinned” in the sense of being effectively anchored in the lower crust, either by pinning in the strict sense or by a sufficiently strong friction force . The braking of the crust will thus have the effect of slowing down the vortices relatively to the underlying superfluid, thereby giving rise to a Magnus force tending to move them out through the superfluid layer and thus slow it down as well. However this tendency to move out will be thwarted by the same anchoring effect that gave rise to it in the first place. This conflict will cause the pinning forces to build up to a critical point at which there will be a breakdown bringing about a discontinuous readjustment of the kind described by the analysis of the preceding section, and in particular by the formula (14).
The breakdown can occur in two different manners:
(a) There can be a sudden unpinning of many vortices, due to the breaking of the pinning bonds , .
(b) Another possibility is that the crust lattice breaks before vortex lines can unpin from it, as suggested in and studied in detail by Ruderman .
Finally, we would like to mention another interesting glitch mechanism due to Link and Epstein , which may be relevant for the present work:
(c) their thermally driven glitch mechanism is based on the so-called vortex creep model , in which the coupling between the vortices and the crust is strongly temperature dependent. A sudden local increase of the inner crust temperature, such as may be due to a crustquake, can then be shown to induce a glitch.
It must be emphasized that all these three mechanisms, even if corresponding to some breaking of the crust as in the scenarios (b) and (c), are very different from the mechanism of section 2, in the sense that they all are in the context of a two-component star, with the neutron superfluid rotating faster than the crust and thus acting as a reservoir of angular momentum. In the following sections, we will consider a mechanism which is not based on the presence of vortices, but still in the context of differential rotation.
Finally, let us mention the question of how big is $`I_\mathrm{n}`$ compared with $`I`$, in other words how much of the neutron fluid is effectively free to rotate independently of the rest? In the unpinned part the vortices can move out freely so as to establish corotation, so $`I_\mathrm{n}`$ may be relatively small , representing the moment of inertia just of the small fraction of the neutron fluid that interpenetrates the deeper layers of the solid crust where pinning is expected to be most effective. However effective pinning may not be confined to the solid crust: it may also be achieved by forces exerted by quantised magnetic field lines (resulting from superfluidity of the protons) in the layers below the crust, in which case the relevant value of $`I_\mathrm{n}`$ might be much larger . Another question (which applies also to the less important spheroidality mechanism discussed above) is that of the absolute values of the discontinuous changes. The foregoing reasonning is concerned just with the ratio of $`\delta \mathrm{\Omega }`$ to $`\mathrm{\Delta }\mathrm{\Omega }`$ but does not tackle the harder problem of their absolute values.
## 5 Potential importance of the centrifugal buoyancy mechanism.
So far we have only been summarising what has long well known to workers in this field. We now come to what seems to us to be an important point that has been overlooked, which is that independently of vortex pinning there is another, comparably powerful mechanism, that can also cause discontinuous angular momentum transfer to a solid crust from an independently rotating superfluid layer. This mechanism does not depend on superfluidity in the strict sense but merely requires perfect fluidity in the sense that the effective viscosity should be low enough for the slowdown of the neutron fluid to lag behind the slowdown (due to its coupling with the radiating magnetosphere) of the solid outer layers. The point is that if the outer layers were also effectively fluid, there would be a convective readjustment, in which annular rings of fluid would change their relative positions, each retaining its separate angular momentum, in such a way that those with less angular momentum per unit mass, and thus with less “centrifugal buoyancy” would move towards the axis while those with more would move out so as to establish a state of equilibrium in which, provided the pressure depends only on the density, the angular velocity would decrease outwards as a function just of cylindrical radius, in accordance with the well known Taylor-Proudman theorem (see, e.g., ). The effect of crust solidity will be to temporarily postpone such readjustments, by the development of the anisotropic stresses needed to balance the centrifugal buoyancy forces. However when such stresses have built up to the critical point at which the solid structure breaks down, the pent up centrifugal buoyancy forces will produce a “starquake” in which the convective readjustment that would have ocurred continuously in the fluid case, is finally achieved in a discontinuous transition.
It is to be noticed that in contrast with the vortex pinning effect (in the following, for easier comparison, we will have in mind the scenario (b) of Section 4), which tends to pull the more slowly rotating crust material outwards from the axis towards the equator (see Fig. 2), the effect of the centrifugal buoyancy deficit in the crust is to pull the crust material inwards towards the axis of the star, where it will finally be subducted into the fluid interior (see Fig. 3). Although the centrifugal buoyancy effect produces convective circulation in just the opposite direction to that produced by vortex pinning (which if it were strong enough would lead to subduction at the equator rather than the axis ) its effect on the angular momentum distribution would be similar, i.e. the net effect of a centrifugal buoyancy crustquake will be a discontinuous transfer angular momentum to the crust from the more rapidly rotating fluid layer. This means that the crude quantitative estimate given by equation (14) is applicable just as well to the effect of a centrifugal buoyancy crustquake as to a vortex pinning crustquake.
The main point we want to emphasise is that whereas vortex pinning may indeed be the main driving force for the build up of the stress that is relaxed in crustquakes, the extent to which it really is depends on detailed considerations about the strength of vortex pinning. On the other hand the opposing centrifugal buoyancy mechanism will always function whenever there is differential rotation. It will be seen in the next section that when it is fully effective the oppositely directed pinning mechanism will be strong enough to overwhelm (i.e. to more than cancel) the buoyancy mechanism, but the latter mechanism is more robust in the sense that it will always make a significant contribution.
Our tentative conclusion – which we are proposing as a subject for debate and further investigation – is that the hitherto neglected centrifugal buoyancy effect may be the dominant cause of the crustquakes that are observed as pulsar glitches, while vortex pinning crustquakes, if they occur at all, are relatively rare. This does not mean that vortex pinning is unimportant for the phenomenon, because it is likely to be what determines the magnitude of the relevant moment of inertia contribution $`I_\mathrm{n}`$ in the estimate (14) for the ratio of $`\delta \mathrm{\Omega }`$ to $`\mathrm{\Delta }\mathrm{\Omega }`$. However what it means is that the vortex pinning stresses are not what is immediately responsible for the discontinuous breakdown, and hence not what is of dominant relevance for estimating the absolute values of $`\mathrm{\Delta }\mathrm{\Omega }`$ at which it is likely to occur.
## 6 The working of the centrifugal buoyancy deficit mechanism.
An accurate treatment of neutron star would of course require a general relativistic analysis , but as a first step towards the estimation of the stress forces needed to maintain equilibrium where the crust constituent is interpenetrated by an independently rotating fluid constituent, it will suffice for our present purpose to work in a Newtonian framework, using a highly idealised two-constituent model in which the corotating crust component (including the protons and electrons, as well as a fraction of the neutrons that is bound into atomic type nuclei) and the neutron superfluid are considered as independent material media having respective mass densities
$$\rho _\mathrm{c}=mn_\mathrm{c},\rho _\mathrm{n}=mn_\mathrm{n}$$
(16)
and spatial velocity components $`v_\mathrm{c}^i`$ and $`v_\mathrm{n}^i`$ ($`i=1,2,3`$) where $`m`$ is the proton mass and $`n_\mathrm{c}`$ and $`n_\mathrm{n}`$ are the corresponding baryon number densities. For an approximate description of the kind of scenario envisaged by Alpar et al in which the rigidly corotating constituent consists not just of the crust lattice but also of the proton superfluid in the core which will be locked to the crust by electromagnetic interactions we adopt a simplified treatment in which it is postulated that the dynamics is governed by Euler type equations of motion of the familiar form
$$\rho _\mathrm{c}\left(__0v_\mathrm{c}^i+v_\mathrm{c}^j_jv_\mathrm{c}^i\right)=^iP_\mathrm{c}\rho _\mathrm{c}^i\varphi +f_\mathrm{c}^i,$$
(17)
$$\rho _\mathrm{n}\left(__0v_\mathrm{n}^i+v_\mathrm{n}^j_jv_\mathrm{n}^i\right)=^iP_\mathrm{n}\rho _\mathrm{n}^i\varphi +f_\mathrm{n}^i,$$
(18)
using $`__0`$ to denote partial differentiation with respect to Newtonian time, where $`\varphi `$ is the Newtonian gravitational potential, and where $`P_\mathrm{c}`$, $`P_\mathrm{n}`$ and $`f_\mathrm{c}^i`$, $`f_\mathrm{n}^i`$ respectively denote the relevant pressure scalars and force density vectors. In a lowest order approximation in which both components can be considered to obey barotropic equations of state giving their energy densities $`\epsilon _\mathrm{c}`$ and $`\epsilon _\mathrm{n}`$ as functions respectively of $`n_\mathrm{c}`$ and of $`n_\mathrm{n}`$, they will be characterised by corresponding chemical potentials
$$\mu _\mathrm{c}=\frac{d\epsilon _\mathrm{c}}{dn_\mathrm{c}},\mu _\mathrm{n}=\frac{d\epsilon _\mathrm{n}}{dn_\mathrm{n}},$$
(19)
from which the associated pressure contributions can be evaluated as
$$P_\mathrm{c}=\mu _\mathrm{c}n_\mathrm{c}\epsilon _\mathrm{c},P_\mathrm{n}=\mu _\mathrm{n}n_\mathrm{n}\epsilon _\mathrm{n}.$$
(20)
This implies that the required gradient terms will be given by
$$^iP_\mathrm{c}=n_\mathrm{c}^i\mu _\mathrm{c},^iP_\mathrm{n}=n_\mathrm{n}^i\mu _\mathrm{n}.$$
(21)
(It is to be remarked that in a more detailed analysis the baryon chemical potential $`\mu _\mathrm{c}`$ in the component corotating with the crust would be interpretable as the sum of proton and electron contributions, $`\mu _\mathrm{c}=\mu _\mathrm{p}+\mu _\mathrm{e}`$.)
Although adequate for the fluid constituent, a purely barotropic description will not be sufficiently accurate for the crust constituent in which we want to allow for the effects of solidity. The usual way to do this is to replace the isotropic pressure gradient term $`_iP_\mathrm{c}`$ by a stress gradient term of the form $`_jT_{\mathrm{c}i}^j`$ where $`T_{\mathrm{c}j}^j`$ is the total stress tensor. It will be convenient for our purpose to decompose the latter in the form
$$T_{\mathrm{c}i}^j=P_\mathrm{c}\delta _i^js_i^j,$$
(22)
where the extra anisotropic stress contribution $`s_i^j`$ is a correction term that will be small compared with the dominant isotropic contribution $`P_\mathrm{c}\delta _j^i`$. This means that while $`f_\mathrm{n}^i`$ is to be interpreted as the interaction force density, if any, exerted on the neutron superfluid component by effects such as vortex pinning, on the other hand the term $`f_\mathrm{c}^i`$ in (17) will consist, not just of the equal and opposite interaction term $`f_\mathrm{n}^i`$ but also of an extra correction term $`f_\mathrm{s}^i`$ due to the anisotropic stress correction representing the effect of the solidity property, i.e, we shall have
$$f_\mathrm{c}^i=f_\mathrm{s}^if_\mathrm{n}^i,f_\mathrm{s}^i=_js_i^j.$$
(23)
The anisotropic stress contribution $`s_i^j`$ and the associated force density $`f_\mathrm{s}^i`$ might also include an allowance for magnetic effects, such as are ultimately responsible for the external braking mechanism and for locking the proton superfluid in the core to the outer crust lattice. However for the equilibrium of the strictly stationary states with which we shall be concerned here such magnetic effects are not important, so it may be considered that the stress force density $`f_\mathrm{s}^i`$ arises just from the Coulomb lattice rigidity in the crust, and that it vanishes in the high density core.
Let us now restrict our attention to configurations that are stationary, so that the terms acted on by $`__0`$ will vanish, and let us suppose the motion consists just of a circular motion about the $`x^_3`$ axis, so that each comoving particle moves with a fixed value of the cylindrical radius $`\varpi =\left(x^{{}_{1}{}^{}\mathrm{\hspace{0.17em}2}}+x^{{}_{2}{}^{}\mathrm{\hspace{0.17em}2}}\right)^{1/2}`$. This means that the velocity gradient terms in the equations of motion will be given by
$$v_\mathrm{c}^j_jv_\mathrm{c}^i=\frac{_1}{^2}\mathrm{\Omega }_\mathrm{c}^{\mathrm{\hspace{0.17em}2}}^i\varpi ^2,v_\mathrm{n}^j_jv_\mathrm{n}^i=\frac{_1}{^2}\mathrm{\Omega }_\mathrm{n}^{\mathrm{\hspace{0.17em}2}}^i\varpi ^2,$$
(24)
where $`\mathrm{\Omega }_\mathrm{c}`$ is the local angular velocity of the crust constituent and $`\mathrm{\Omega }_\mathrm{n}`$ is the local angular velocity of the superfluid constituent. Under these conditions the Euler equations (17) and (18) can be rewritten in the form
$$\frac{_1}{^2}\mathrm{\Omega }_\mathrm{c}^{\mathrm{\hspace{0.17em}2}}^i\varpi ^2^i\left(\varphi +m^1\mu _\mathrm{c}\right)=\rho _\mathrm{c}^1\left(f_\mathrm{n}^if_\mathrm{s}^i\right),$$
(25)
and
$$\frac{_1}{^2}\mathrm{\Omega }_\mathrm{n}^{\mathrm{\hspace{0.17em}2}}^i\varpi ^2^i\left(\varphi +m^1\mu _\mathrm{n}\right)=\rho _\mathrm{n}^1f_\mathrm{n}^i.$$
(26)
If vortex pinning were effective, it would contribute to $`f_\mathrm{n}^i`$ the force density needed to counteract Joukowsky-Magnus type lift force density $`f_\mathrm{J}^i`$ that would be exerted on the vortices by the Magnus effect, which would be given by
$$f_\mathrm{J}^i=\rho _\mathrm{n}\left(\mathrm{\Omega }_\mathrm{n}\mathrm{\Omega }_\mathrm{c}\right)\mathrm{\Omega }_\mathrm{n}^i\varpi ^2,$$
(27)
but in the absence of vortex pinning or other coupling forces, the right hand side of (26) will simply vanish, in which case it can be seen that the fluid will satisfy the Taylor-Proudman condition, meaning that its angular velocity $`\mathrm{\Omega }_\mathrm{n}`$ and also the combination $`m\varphi +\mu _\mathrm{n}`$ must vary as a function only of the cylindrical radius $`\varpi `$.
Since the interaction force density $`f_\mathrm{n}^i`$ will cancel out of the linear combination of (25) and (26) obtained from the direct sum of (17) and (18), it follows that this combination will take a simple form that is conveniently expressible – independently of whether vortex pinning is actually effective or not – in terms of the “would-be” Joukowsky force density (27) as
$$^iP+\rho \left(^i\varphi \frac{_1}{^2}\mathrm{\Omega }_\mathrm{c}^{\mathrm{\hspace{0.17em}2}}^i\varpi ^2\right)=f_\mathrm{J}^i+f_\mathrm{s}^i\frac{_1}{^2}\rho _\mathrm{n}\left(\mathrm{\Omega }_\mathrm{n}\mathrm{\Omega }_\mathrm{c}\right)^2^i\varpi ^2,$$
(28)
in which the total pressure $`P`$ and mass density pressure $`\rho `$ are defined in the obvious way as
$$P=P_\mathrm{c}+P_\mathrm{n},\rho =\rho _\mathrm{c}+\rho _\mathrm{n}.$$
(29)
In a systematic calculation by successive approximations, the first stage would be to obtain a zeroth order solution of the stellar equilibrium problem in which the (first order) crust rigidity and differential rotation contributions on the right hand side of (28) would simply be neglected. What we are interested in here is the next stage, which involves the first order equation (from which the zeroth order part has cancelled out) that is obtainable by taking the difference of (25) and (26).
Before going ahead it is necessary to stress that, since only weak interactions are involved, it cannot be taken for granted that the relevant nuclear transitions involved in the “neutron drip” process whereby matter is transferred between the ionic crust material and the interpenetrating neutron superfluid will be very rapid compared with the “secular evolution” timescales on which the state under consideration is significantly modified. If the “neutron drip” process were sufficiently rapid one would obtain not just mechanical equilibrium, such as expressed by equations (25) and (26), but also thermodynamical equilibrium in the rest frame of the crust, in the sense that the energy per baryon of the “normal” matter corotating with the crust, which is just $`\mu _\mathrm{c}`$, would be the same as the energy per baryon of the neutron fluid with respect to the crust corotating frame, which has the value $`\mu _\mathrm{n}+\frac{_1}{^2}m\left(\mathrm{\Omega }_\mathrm{n}\mathrm{\Omega }_\mathrm{c}\right)^2\varpi ^2`$. In practice however, due to the slowness of the relevant nuclear transitions , it is necessary to allow for the possibility of a finite deviation,
$$\mathrm{\Delta }\mu =\mu _\mathrm{c}\mu _\mathrm{n}\frac{_1}{^2}m\left(\mathrm{\Omega }_\mathrm{n}\mathrm{\Omega }_\mathrm{c}\right)^2\varpi ^2,$$
(30)
from exact thermodynamic equilibrium. Estimates of the likely values for such a chemical potential excess due to the simple spheroidality adjustment mechanism, discussed above in Section 2, have been provided by the recent work of Reisenegger . Significantly larger values are likely to arise from the differential rotation mechanisms considered here due to the resulting tendency for the crust constituent to be convected relative to the neutron fluid constituent.
Including allowance for the possibility of a neutron drip delay contribution
$$f_\mathrm{x}^i=n_\mathrm{c}^i(\mathrm{\Delta }\mu ),$$
(31)
representing the force density due to the chemical potential excess (30) if any, the solid stress force density $`f_\mathrm{s}^i`$ ultimately responsible for the glitches in which we are interested can be seen to be given by the first order equation obtained by subtracting (26) from (25), which will be expressible in the form
$$f_\mathrm{s}^i=f_\mathrm{x}^i+\frac{\rho }{\rho _\mathrm{n}}f_\mathrm{n}^i+f_\mathrm{b}^i.$$
(32)
The final term in the above equation is what can be interpreted as the extra force needed to compensate for the buoyancy deficit of the crust due to its lack of rotation velocity relative to the neutron superfluid, and is given by
$$f_\mathrm{b}^i=\rho _\mathrm{c}\left(\mathrm{\Omega }_\mathrm{n}\mathrm{\Omega }_\mathrm{c}\right)\left(^i(\varpi ^2\mathrm{\Omega }_\mathrm{n})\varpi ^2^i\mathrm{\Omega }_\mathrm{c}\right).$$
(33)
## 7 Estimation of the centrifugal buoyancy deficit force density.
The solidity property of the crust implies that, in a stationary state, its rotation must be rigid, i.e.
$$\mathrm{\Omega }_\mathrm{c}=\mathrm{\Omega },^i\mathrm{\Omega }=0,$$
(34)
where $`\mathrm{\Omega }`$ is a uniform angular velocity value (the one that is actually observable from outside), so the formula (33) for the buoyancy deficit force density can be immediately simplified to the form
$$f_\mathrm{b}^i=\rho _\mathrm{c}\left(\mathrm{\Omega }_\mathrm{n}\mathrm{\Omega }\right)^i(\varpi ^2\mathrm{\Omega }_\mathrm{n}).$$
(35)
If the superfluid were macroscopically irrotational, i.e. if there were no vortices present, then $`\varpi ^2\mathrm{\Omega }_\mathrm{n}`$ would have a uniform value so the right hand side of (35) would also vanish, i.e. the effective buoyancy deficit force density $`f_\mathrm{b}^i`$ would be zero.
What we actually anticipate in the context of the pulsar slowdown problem is that $`\mathrm{\Omega }_\mathrm{n}`$ will be approximately uniform (representing rigid rather than irrotational motion) with a value equal to that of the crust component at a rather earlier stage, perhaps just after the previous glitch, and that the velocity difference will therefore be small compared with the total angular velocity
$$|\mathrm{\Omega }_\mathrm{n}\mathrm{\Omega }||\mathrm{\Omega }|.$$
(36)
Thus, by neglecting corrections of quadratic order in this velocity difference, we see that (35) can be conveniently approximated by the simpler formula
$$f_\mathrm{b}^i\rho _\mathrm{c}\left(\mathrm{\Omega }_\mathrm{n}\mathrm{\Omega }\right)\mathrm{\Omega }_\mathrm{n}^i(\varpi ^2),$$
(37)
which will be accurate to linear order in the difference $`\mathrm{\Omega }_\mathrm{n}\mathrm{\Omega }`$. It is to be remarked that, to the same order of accuracy, the neutron drip delay force contribution (31) will be given by the approximation
$$f_\mathrm{x}^in_\mathrm{c}^i(\mu _\mathrm{c}\mu _\mathrm{n}).$$
(38)
It is to be observed that the formula (37) for the buoyancy deficit force density closely ressembles the Joukowsky formula (27) for the lift force density $`f_\mathrm{J}^i`$ that would be exerted on the vortices by the Magnus effect if they are pinned to the crust: this Joukowsky force density is evidently related to the buoyancy deficit force density by the simple proportionality relation
$$f_\mathrm{J}^i\frac{\rho _\mathrm{n}}{\rho _\mathrm{c}}f_\mathrm{b}^i.$$
(39)
It follows that, in terms of the effective (centrifugally adjusted) gravitational potential
$$\psi _\mathrm{c}=\varphi \frac{_1}{^2}\mathrm{\Omega }_\mathrm{c}^{\mathrm{\hspace{0.17em}2}}\varpi ^2,$$
(40)
the basic stellar equilibrium equation (28) will reduce to the form
$$^iP+\rho ^i\psi _\mathrm{c}f_\mathrm{J}^i+f_\mathrm{s}^i,$$
(41)
in which the zeroth order terms are grouped on the left and the first order terms are on the right (while the final second order term on the right of (28) has been neglected). Since the left hand side consists just of the small difference left over after the approximate cancellation of the dominant zeroth order terms, this equation does not provide any utilisable information about the solid force density $`f_\mathrm{s}^i`$ in which we are interested: on the contrary, after $`f_\mathrm{s}^i`$ has been evaluated by other means, (41) can be used to calculate the corresponding first order adjustments to the zeroth order pressure and density distributions.
The equation that does supply the relevant information about the solid stress force density $`f_\mathrm{s}^i`$ in which we are interested is the first order equilibrium condition (32), whose terms can be instructively regrouped in the form
$$f_\mathrm{s}^if_\mathrm{x}^i=f_\mathrm{b}^i+\frac{\rho }{\rho _\mathrm{n}}f_\mathrm{n}^i,$$
(42)
in which it can be seen from (37) that the right hand side will always be approximately proportional to the first order difference $`\mathrm{\Omega }_\mathrm{n}\mathrm{\Omega }`$ whether or not pinning is effective. (This shows incidentally that differential rotation would be impossible if both the rigidity force $`f_\mathrm{s}^i`$ and the chemical delay contribution $`f_\mathrm{x}^i`$ were negligible.)
In particular, the relation (42) shows, by (39) that in the pinned case, i.e. when the superfluid is submitted to a force density
$$f_\mathrm{n}^if_\mathrm{J}^i.$$
(43)
on the crust, the stress force density $`f_\mathrm{s}^i`$ necessary for equilibrium will be given by the simple formula
$$f_\mathrm{s}^i=f_\mathrm{J}^i+f_\mathrm{x}^i,$$
(44)
in which the first term on the right is just the Joukowsky-Magnus contribution, as is assumed in the conventional presentation of the vortex pinning theory of pulsar glitches.
The formula (44) is potentially misleading in that it gives the false impression that if the pinning were ineffective, so that instead of being given by (43) the force exerted on the superfluid by the crust were simply zero,
$$f_\mathrm{n}^i=0,$$
(45)
then the first term on the right of the stress force density formula would similarly disappear, whereas in fact substitution of (45) in (32) leads to the replacement of (44) by the formula
$$f_\mathrm{s}^i=f_\mathrm{b}^i+f_\mathrm{x}^i,$$
(46)
in which, instead of the Joukowsky-Magnus contribution $`f_\mathrm{J}^i`$, the right hand side is now given by the oppositely directed buoyancy deficit force contribution $`f_\mathrm{b}^i`$.
Our reasonning so far does not make it obvious whether or not the crust will develop a sufficiently non-uniform chemical potential excess $`\mathrm{\Delta }\mu `$ to provide a significant chemical excess force $`f_\mathrm{x}^i`$. If it is a good approximation to suppose that chemical excess force in the crust vanishes,
$$f_\mathrm{x}^i=0,$$
(47)
(as seems to have been implicitly asumed in most previous works but which needs to be confirmed or infirmed quantatively) then, in the case where pinning would not be effective (as has been advocated by Jones contrarily to earlier works), it follows from (46) that there will still be a solid stress force density given by
$$f_\mathrm{s}^if_\mathrm{b}^i,$$
(48)
in which the centrifugal buoyancy deficit force density on the right is given by equation (37). This formula can be seen to differ from the (alternative) well known formula – for the stress due to pinning –, deduced from (44) by the same assumption (47),
$$f_\mathrm{s}^if_\mathrm{J}^i,$$
(49)
with the Joukowsky-Magnus term on the right hand side given by (27), only by having the opposite sign and by having a proportionality factor given by the density $`\rho _\mathrm{c}`$ of the corotating crust component instead of the density $`\rho _\mathrm{n}`$ of the differentially rotating neutron superfluid component.
## 8 Discussion and conclusions.
In the lower crust region that seems most likely to be relevant for the explanation of the large glitches observed in the Vela pulsar one would expect the corotating constituent to be characterised by a density $`\rho _\mathrm{c}`$ (attributable mainly to protons and bound neutrons in the atomic type ions forming a solid lattice) having a range of values that is roughly comparable with that of the corresponding neutron superfluid density $`\rho _\mathrm{n}`$ (quantitatively round about $`10^{13}`$ g/cm<sup>3</sup>). Thus although they are of opposite sign (tending to push the crust material outward in the case (43) of vortex pinning, but to push it inwards in the case (45) for which pinning is absent) the alternative formulae (48) and (49) both predict the same rough order of magnitude for the stress induced on the crust by the existence of a difference between the angular velocity $`\mathrm{\Omega }_\mathrm{n}`$ of the neutron superfluid constituent and the (externally observable) angular velocity $`\mathrm{\Omega }`$ characterising the crust.
The implication is that, as a candidate for explaining the large magnitude of the discontinuous changes $`\delta \mathrm{\Omega }`$ that are commonly observed in a pulsar such as Vela, the previously overlooked buoyancy deficit mechanism characterised by the formula (48), i.e.
$$f_\mathrm{s}^i\rho _\mathrm{c}\left(\mathrm{\Omega }_\mathrm{n}\mathrm{\Omega }\right)\mathrm{\Omega }_\mathrm{n}^i\varpi ^2,$$
(50)
(pushing outward along the cylindrical radial direction) seems at first sight to be just as promising as the more thoroughly investigated vortex pinning mechanism, which, if the chemical contribution $`f_\mathrm{x}^i`$ were unimportant, would be given according to (49) by
$$f_\mathrm{s}^i\rho _\mathrm{n}\left(\mathrm{\Omega }_\mathrm{n}\mathrm{\Omega }\right)\mathrm{\Omega }_\mathrm{n}^i\varpi ^2,$$
(51)
(pushing inward along the cylindrical radial direction). In order to obtain definitive conclusions it is clear however that much more work on both kinds of mechanism will be needed. In particular it will be necessary to pay more attention than hitherto to the role of the chemical excess force (31).
The present situation can be summarised by the statement that the large magnitude of the observed glitches in Vela provides strong evidence for the existence of angular velocity differences – and hence for the existence of superfluidity – in the pulsar interior, but that it is premature to claim it also provides strong evidence for vortex pinning because stresses of comparable magnitude could be produced in the absence of pinning by the centrifugal buoyancy deficit mechanism.
Acknowledgments
We wish to thank R. Prix and P. Hansel for very valuable discussions. One of us (D.S.) would like to acknowledge “Jumelage France-Arménie” exchange programme for financial support.
|
warning/0004/nucl-th0004066.html
|
ar5iv
|
text
|
# Jets and produced particles in 𝑝𝑝 collisions from SPS to RHIC energies for nuclear applications
## ACKNOWLEDGMENTS
We thank M. Werlen, P. Aurenche, and D. Soper for stimulating discussions. This work was supported in part by U.S. DOE grant DE-FG02-86ER-40251, Hungarian OTKA Grant No. T032796, FKFP grant 0220/2000, and by the US-Hungarian Joint Fund No. 652. Partial support by the Domus Hungarica program of the Hungarian Academy of Sciences and by the Research Council of Kent State University is gratefully acknowledged.
Fig. 1. Energy and transverse-momentum dependent $`K`$ factor, $`K_{jet}(s,p_T)`$ with $`Q=p_T/2`$, $`R=0.7`$, $`R_{sep}=2R`$. Dashed lines represent Eq. (2).
Fig. 2. $`K`$ factor, $`K_{jet}(s,p_T)`$ with $`R=1.0`$ ($`Q=p_T/2`$ and $`R_{sep}=2R`$). Dashed lines correspond to Eq. (3).
Fig. 3. $`K`$ factor for pions, $`K_\pi (s,p_T)`$ (solid line) and for kaons, $`K_K(s,p_T)`$ (dashed), after hadronization of jets calculated with $`R=0.7`$ ($`Q=p_T/2`$, $`R_{sep}=2R`$).
Fig. 4. Comparison of $`K`$ factors, $`K_{jet}(s,p_T)`$ (dotted lines) and $`K_\pi (s,p_T)`$ (full lines) at $`\sqrt{s}=24`$ GeV, with $`R=0.7`$, $`R=0.85`$ and $`R=1.0`$ ($`Q=p_T/2`$, $`R_{sep}=2R`$).
Fig. 5. $`K`$ factor for photons, $`K_\gamma (s,p_T)`$ at energies $`\sqrt{s}=20200`$ GeV.
|
warning/0004/hep-ph0004191.html
|
ar5iv
|
text
|
# Neutrino conversions in cosmological gamma-ray burst fireballs
## I Introduction
Recently, cosmological fireballs are suggested as a possible production site for gamma-ray bursts as well as the high energy ($`E\stackrel{>}{_{}}\mathrm{\hspace{0.17em}10}^6`$ GeV) neutrinos . Although, the origin of these cosmological Gamma-Ray Burst fireballs (GRB) is not yet understood, the observations suggest that generically a very compact source of linear scale $`\mathrm{\hspace{0.17em}10}^7`$ cm through internal or/and external shock propagation produces these gamma-ray bursts (as well as the burst of high energy neutrinos) . Typically, this compact source is hypothesized to be formed possibly due to the merging of binary neutron stars or due to collapse of a supermassive star.
The main source of high energy tau neutrinos in GRBs is the production and decay of $`D_S^\pm `$. The production of $`D_S^\pm `$ may be through $`p\gamma `$ and/or through $`pp`$ collisions. In , the $`\nu _e`$ and $`\nu _\mu `$ flux is estimated in $`pp`$ collisions, whereas in , the $`\nu _e`$ and $`\nu _\mu `$ flux is estimated in $`p\gamma `$ collisions for GRBs. In $`pp`$ collisions, the flux of $`\nu _\tau `$ may be obtained through the main process of $`p+pD_S^++X`$. The $`D_S^+`$ decays into $`\tau ^+`$ lepton and $`\nu _\tau `$ with a branching ratio of $`\mathrm{\hspace{0.17em}3}\%`$. This $`\tau ^+`$ lepton further decays into $`\nu _\tau `$. The cross-section for $`D_S^+`$ production, which is main source of $`\nu _\tau `$’s, is $``$ 4 orders of magnitude lower than that of $`\pi ^+`$ and/or $`\pi ^{}`$ which subsequently produces $`\nu _e`$ and $`\nu _\mu `$. The branching ratio for $`\nu _e`$ and/or $`\nu _\mu `$ production is higher up to an order of magnitude than that for $`\nu _\tau `$ production (through $`D_S^\pm `$). These imply that the $`\nu _\tau `$ flux in $`pp`$ collisions is suppressed up to 5 orders of magnitude relative to corresponding $`\nu _e`$ and/or $`\nu _\mu `$ fluxes. In $`p\gamma `$ collisions, the main process for the production of $`\nu _\tau `$ may be $`p+\gamma D_S^++\mathrm{\Lambda }^0+\overline{D}^0`$ with similar relevant branching ratios and corresponding suppression for cross-section values. Here the corresponding main source for $`\nu _e`$ and $`\nu _\mu `$ production is $`p+\gamma \mathrm{\Delta }^+\pi ^++n`$. Therefore, in $`p\gamma `$ collisions, the $`\nu _\tau `$ flux is also suppressed up to 5 orders of magnitude relative to $`\nu _e`$ and/or $`\nu _\mu `$ flux. Thus, in both type of collisions, including the relevant kinematics, the intrinsic $`\nu _\tau (\overline{\nu }_\tau )`$ flux is estimated to be rather small relative to $`\nu _e(\overline{\nu }_e)`$ and/or $`\nu _\mu (\overline{\nu }_\mu )`$ fluxes from GRBs, typically being, $`F_\tau ^0/F_{e,\mu }^0<10^5`$ .
In this paper, we consider the possibility of obtaining higher $`\nu _\tau `$ flux, that is, $`F_\tau ^0/F_{e,\mu }^0>\mathrm{\hspace{0.17em}10}^5`$, from GRBs through neutrino conversions as compared to no conversion situation. The present study is particularly useful as the new under ice/water Čerenkov light neutrino telescopes namely AMANDA and Baikal (also the NESTOR and ANTARES) will have energy, angle and flavor resolution for high energy neutrinos originating at cosmological distances . Recently, there are several discussions concerning the signatures of a possible neutrino burst from GRBs correlated in time and angle . In particular, there is a suggestion of measuring $`\nu _\tau `$ flux from cosmologically distant sources through a double shower (bang) event or through a small pile up of up ward going $`\mu `$-like events near (10$`{}_{}{}^{4}10^5`$) GeV .
The plan of this paper is as follows. In Sect. II, we briefly describe the matter density and magnetic field in the vicinity of GRBs. In Sect. III, we discuss in some detail, the range of neutrino mixing parameters that may give rise to relatively large precession/conversion probabilities resulting from neutrino flavor/spin-flavor conversions. In Sect. IV, we give estimates for separable but contained double shower event rates induced by high energy $`\nu _\tau `$’s originating from GRBs without/with conversions for km<sup>2</sup> surface area under water/ice neutrino telescopes for illustrative purposes and finally in Sect. V, we summarize our results.
## II matter density and magnetic field in the vicinity of GRB
According to , the isotropic high energy neutrino production may take place in the vicinity of $`r_p\mathrm{\Gamma }^2c\mathrm{\Delta }t`$. Here $`\mathrm{\Gamma }`$ is the Lorentz factor (typically $`\mathrm{\Gamma }\mathrm{\hspace{0.17em}300}`$) and $`\mathrm{\Delta }t`$ is the observed GRB variability time scale (typically $`\mathrm{\Delta }t\mathrm{\hspace{0.17em}1}`$ ms). In the vicinity of $`r_p`$, the fireball matter density is $`\rho \mathrm{\hspace{0.17em}10}^{13}`$ g cm<sup>-3</sup> . In these models, the typical distance traversed by neutrinos may be taken as, $`\mathrm{\Delta }r\stackrel{<}{_{}}(10^41)`$ pc, in the vicinity of GRB, where 1 pc $`\mathrm{\hspace{0.17em}3}\times 10^{18}`$ cm. Matter effects on neutrino oscillations are relevant if $`G_F\rho /m_N\delta m^2/2E`$. Using $`\rho `$ from Ref. , it follows that matter effects are absent for $`\delta m^2\stackrel{>}{_{}}𝒪(10^{10})`$ eV<sup>2</sup>. Matter effects due to coherent forward scattering of neutrinos off the background for high energy neutrinos originating from GRBs are not expected to be important in the neutrino production regions around GRBs and will not be further discuss here.
Taking the observed duration of the typical gamma-ray burst as, $`\mathrm{\Delta }t\stackrel{<}{_{}}\mathrm{\hspace{0.17em}1}`$ ms, we obtain the mass of the source as, $`M_{GRB}\stackrel{<}{_{}}\mathrm{\Delta }t/G_N`$, where $`G_N`$ is the gravitational constant. For the relatively shorter observed duration of gamma-ray burst from a typical GRB, $`\mathrm{\Delta }t\mathrm{\hspace{0.17em}0.2}`$ ms, implying $`M_{GRB}\mathrm{\hspace{0.17em}40}M_{}`$ (where $`M_{}\mathrm{\hspace{0.17em}2}\times 10^{33}`$ g is solar mass). We use $`M_{GRB}\mathrm{\hspace{0.17em}2}\times 10^2M_{}`$ in our estimates.
The magnetic field in the vicinity of a GRB is obtained by considering the equipartition arguments . We use the following profile of magnetic field, $`B_{GRB}`$, as an example, for $`r>r_p`$
$$B_{GRB}(r)B_0\left(\frac{r_p}{r}\right)^2,$$
(1)
where $`B_0L^{1/2}c^{1/2}(r_p\mathrm{\Gamma })^1`$ with $`L`$ being the total wind luminosity (typically $`L\mathrm{\hspace{0.17em}10}^{51}`$ erg s<sup>-1</sup>).
## III Neutrino conversions in GRB
### A Flavor oscillations
In the framework of three flavor analysis, the flavor precession probability from $`\alpha `$ to $`\beta `$ neutrino flavor is
$$P(\nu _\alpha \nu _\beta )P_{\alpha \beta }=\underset{i=1}{\overset{3}{}}|U_{\alpha i}|^2|U_{\beta i}|^2+\underset{ij}{}U_{\alpha i}U_{\beta i}^{}U_{\alpha j}^{}U_{\beta j}\mathrm{cos}\left(\frac{2\pi L}{l_{ij}}\right),$$
(2)
where $`\alpha ,\beta =e,\mu ,`$ or $`\tau `$. $`U`$ is the 3$`\times `$3 MNS mixing matrix and can be obtained in usual notation through
$$UR_{23}(\theta _1)\text{diag}(e^{i\delta /2},1,e^{i\delta /2})R_{31}(\theta _2)\text{diag}(e^{i\delta /2},1,e^{i\delta /2})R_{12}(\theta _3),$$
(3)
thus coinciding with the standard form given by the Particle Data Group . In Eq. (3), $`l_{ij}4\pi E/\delta m_{ij}^2`$ with $`\delta m_{ij}^2|m_i^2m_j^2|`$ and $`L`$ is the distance between the source and the detector. For simplicity, we will assume here a vanishing value for $`\theta _{31}`$ and CP violating phase $`\delta `$ in $`U`$.
At present, the atmospheric muon and solar electron neutrino deficits can be explained with oscillations among three active neutrinos . For this, typically, $`\delta m^2𝒪(10^3)`$ eV<sup>2</sup> and $`\mathrm{sin}^22\theta 𝒪(1)`$ for the explanation of atmospheric muon neutrino deficit, whereas for the explanation of solar electron neutrino deficit, we may have $`\delta m^2𝒪(10^{10})`$ eV<sup>2</sup> and $`\mathrm{sin}^22\theta 𝒪(1)`$ \[just so\] or $`\delta m^2𝒪(10^5)`$ eV<sup>2</sup> and $`\mathrm{sin}^22\theta 𝒪(10^2)`$ \[SMA (MSW)\] or $`\delta m^2𝒪(10^5)`$ eV<sup>2</sup> and $`\mathrm{sin}^22\theta 𝒪(1)`$ \[LMA (MSW)\]. The present status of data thus permits multiple oscillation solutions to solar neutrino deficit. We intend to discuss here implications of these mixings for high energy cosmic neutrino propagation.
In the above explanations, the total range of $`\delta m^2`$ is $`10^{10}\delta m^2/`$ eV$`{}_{}{}^{2}10^3`$ irrespective of neutrino flavor. The typical energy span relevant for possible flavor identification for high energy cosmic neutrinos is $`2\times 10^6E/`$GeV$`2\times 10^7`$ (see Sect. IV). Taking a typical distance between the GRB and our galaxy as $`L1000`$ Mpc, we note that $`\mathrm{cos}`$ term in Eq. (2) vanishes and so Eq. (2) reduces to
$$P_{\alpha \beta }\underset{i=1}{\overset{3}{}}|U_{\alpha i}|^2|U_{\beta i}|^2.$$
(4)
It is assumed here that no relatively dense objects exist between the GRB and the earth so as to effect significantly this oscillations pattern. Since $`P_{\alpha \beta }`$ in above Eq. is symmetric under the exchange of $`\alpha `$ and $`\beta `$ indices implying that essentially no $`T`$ (or $`CP`$) violation effects arise in neutrino vacuum flavor oscillations for high energy cosmic neutrinos .
Let us denote by $`F_\alpha ^0`$, the intrinsic neutrino fluxes. From the discussion in the previous Sect., it follows that $`F_e^0:F_\mu ^0:F_\tau ^0=1:2:<10^5`$. For simplicity, we take these ratios as 1 : 2 : 0. In order to estimate the final flux ratios on earth for high energy cosmic neutrinos originating from cosmologically distant GRBs, let us introduce a 3$`\times `$3 matrix of vacuum flavor precession probabilities such that
$$F_\alpha =\underset{\beta }{}P_{\alpha \beta }F_\beta ^0,$$
(5)
where the unitarity conditions for $`P_{\alpha \beta }`$ read as
$`P_{ee}+P_{e\mu }+P_{e\tau }`$ $`=`$ $`1,`$ (6)
$`P_{e\mu }+P_{\mu \mu }+P_{\mu \tau }`$ $`=`$ $`1,`$ (7)
$`P_{e\tau }+P_{\mu \tau }+P_{\tau \tau }`$ $`=`$ $`1.`$ (8)
The explicit form for the matrix $`P`$ in case of just so flavor oscillations as solution to solar neutrino problem along with the solution to atmospheric neutrino deficit in terms of $`\nu _\mu `$ to $`\nu _\tau `$ oscillations with maximal depth is
$$P=\left(\begin{array}{ccc}1/2& 1/4& 1/4\\ 1/4& 3/8& 3/8\\ 1/4& 3/8& 3/8\end{array}\right).$$
(9)
Using Eq. (9) and Eq. (5), we note that $`F_e:F_\mu :F_\tau =1:1:1`$ at the level of $`F_e^0`$. Also, Eq. (8) is satisfied. The same final flux ratio is obtained in remaining two cases for which the corresponding $`P`$ matrices are \[for SMA (MSW)\]
$$P=\left(\begin{array}{ccc}1& 0& 0\\ 0& 1/2& 1/2\\ 0& 1/2& 1/2\end{array}\right),$$
(10)
whereas in case of LMA (MSW),
$$P=\left(\begin{array}{ccc}5/8& 3/16& 3/16\\ 3/16& 13/32& 13/32\\ 3/16& 13/32& 13/32\end{array}\right).$$
(11)
Thus, independent of the oscillation solution for solar neutrino problem, we have $`F_e:F_\mu :F_\tau =1:1:1`$. Importantly, these ratios do not depend on neutrino energy or $`\delta m^2`$ at least in the relevant energy interval.
Summarizing, although intrinsically the high energy cosmic tau neutrino flux is negligibally small however because of vacuum flavor oscillations it becomes comparable to $`\nu _e`$ flux thus providing some prospects for its possible detection.
### B Spin-flavor oscillations
We consider here an example of a possibility that may lead to an energy dependence and/or change in the above mentioned final flux ratios. We consider spin-flavor oscillations resulting from an interplay of possible Violation of Equivalence Principle (VEP) parameterized by $`\mathrm{\Delta }f`$ and the magnetic field in the vicinity of GRBs. We obtain the range of neutrino mixing parameters giving $`F_\tau /F_{e,\mu }>\mathrm{\hspace{0.17em}10}^5`$. The possibility of VEP arises from the realization that different flavors of neutrinos may couple differently to gravity .
Consider a system of two mixed neutrinos ($`\overline{\nu }_e`$ and $`\nu _\tau `$) for simplicity. The difference of diagonal elements of the $`2\times 2`$ effective Hamiltonian describing the dynamics of the mixed system of these oscillating neutrinos in the basis $`\psi ^T=(\overline{\nu }_e,\nu _\tau )`$ for vanishing vacuum and gravity mixing angles is
$$\mathrm{\Delta }H=\delta V_G,$$
(12)
whereas each of the off diagonal elements is $`\mu B`$ ($`\mu `$ is magnitude of neutrino magnetic moment). In Eq. (12), $`\delta =\delta m^2/2E`$, where $`\delta m^2=m^2(\nu _\tau )m^2(\overline{\nu }_e)>\mathrm{\hspace{0.17em}0}`$. Here $`V_G`$ is the effective potential felt by the neutrinos at a distance $`r`$ from a gravitational source of mass $`M`$ due to VEP and is given by
$$V_G\mathrm{\Delta }f\varphi (r)E,$$
(13)
where $`\mathrm{\Delta }f=f_3f_1`$ is a measure of the degree of VEP and $`\varphi (r)=G_NMr^1`$ is the gravitational potential in the Keplerian approximation. In Eq. (13), $`f_3G_N`$ and $`f_1G_N`$ are respectively the gravitational couplings of $`\nu _\tau `$ and $`\overline{\nu }_e`$, such that $`f_1f_3`$. We assume here same gravitational couplings for $`\nu `$ and $`\overline{\nu }`$ for simplicity. This implies that the sign of $`V_G`$ is same for $`\overline{\nu }_e\nu _\tau `$ and $`\nu _e\overline{\nu }_\tau `$ conversions . If this is not the case then the sign of $`V_G`$ will be different for the two conversions and only one of the two conversions can occur. This may lead to a change in expected $`\nu _e/\overline{\nu }_e`$ ratio. We briefly comment on the possibility of empirical realization of this situation in next Sect..
There are at least three relevant $`\varphi (r)`$’s that need to be considered . The effect of $`\varphi (r)`$ due to supercluster named Great Attractor with $`\varphi _{SC}(r)`$ in the vicinity of GRB; $`\varphi (r)`$ due to GRB itself, which is, $`\varphi _{GRB}(r)`$, in the vicinity of GRB and the galactic gravitational potential, which is, $`\varphi _G(r)`$. Therefore, we use, $`\varphi (r)\varphi _{SC}(r)+\varphi _{GRB}(r)+\varphi _G(r)`$. However, $`\varphi _G(r)\varphi _{SC}(r),\varphi _{GRB}(r)`$ in the vicinity of GRB. Thus, $`\varphi (r)\varphi _{SC}(r)+\varphi _{GRB}(r)`$. If the neutrino production region $`r_p`$ is $`\stackrel{<}{_{}}10^{13}`$ cm then at $`rr_p`$, we have $`\varphi _{GRB}(r)>\varphi _{SC}(r)`$. At $`r\mathrm{\hspace{0.17em}6}\times 10^{13}`$ cm, $`\varphi _{GRB}(r)\varphi _{SC}(r)`$ and for $`r\stackrel{>}{_{}}\mathrm{\hspace{0.17em}10}^{14}`$ cm, $`\varphi _{GRB}(r)<\varphi _{SC}(r)`$. If the supercluster is a fake object then $`\varphi (r)\varphi _{GRB}(r)`$. Here we assume the smallness of the effect of $`\varphi (r)`$ due to an Active Galactic Nucleus (AGN), if any, nearby to GRB.
The possibility of vanishing gravity and vacuum mixing angle in Eq. (12) allows us to identify the range of $`\mathrm{\Delta }f`$ relevant for the neutrino magnetic moment effects only. Latter in this Sect., we briefly comment on the ranges of relevant neutrino mixing parameters for non vanishing gravity mixing angle with vanishing neutrino magnetic moment.
The case of $`\overline{\nu }_\mu \nu _\tau `$ can be studied by replacing $`\overline{\nu }_e`$ with $`\overline{\nu }_\mu `$ along with corresponding changes in $`V_G`$ and $`\delta m^2`$. The intrinsic flux of $`\overline{\nu }_\mu `$ may be greater than that of $`\overline{\nu }_e`$ by a factor of $``$ 2 , thus also possibly enhancing the expected $`\nu _\tau `$ flux from GRBs through $`\overline{\nu }_\mu \nu _\tau `$. However, we have checked that observationally this possibility leads to quite similar results in terms of event rates and are therefore not discussed here further. We now study in some detail, the various possibilities arising from relative comparison between $`\delta `$ and $`V_G`$ in Eq. (12).
Let us first ignore the effects of VEP ($`\mathrm{\Delta }f=\mathrm{\hspace{0.17em}0}`$). For constant $`B`$, the spin-flavor precession probability $`P(\overline{\nu }_e\nu _\tau )`$ is obtained using Eq. (12) as
$$P(\overline{\nu }_e\nu _\tau )=\left[\frac{(2\mu B)^2}{(2\mu B)^2+\delta ^2}\right]\mathrm{sin}^2\left(\sqrt{(2\mu B)^2+\delta ^2}\frac{\mathrm{\Delta }r}{2}\right).$$
(14)
We take $`\mu \mathrm{\hspace{0.17em}10}^{12}\mu _B`$ or less, where $`\mu _B`$ is Bohr magneton, which is less than the stringent astrophysical upper bound on $`\mu `$ based on cooling of red giants . We here consider the transition magnetic moment, thus allowing the possibility of simultaneously changing the relevant neutrino flavor as well as the helicity. Therefore, the precessed $`\nu _\tau `$ is an active neutrino and interacts weakly. In Eq. (14), $`\mathrm{\Delta }r`$ is the width of the region with $`B`$. If $`\delta <\mathrm{\hspace{0.17em}2}\mu B`$, then, for $`E\mathrm{\hspace{0.17em}2}\times 10^6`$ GeV and using Eq. (1), we obtain $`\delta m^2<\mathrm{\hspace{0.17em}5}\times 10^8`$ eV<sup>2</sup>. We take, $`\delta m^2\mathrm{\hspace{0.17em}10}^9`$ eV<sup>2</sup>, as an example and consequently we obtain from Eq. (14) an energy independent large ($`P>\mathrm{\hspace{0.17em}1}/2`$) spin-flavor precession probability for $`\mu \mathrm{\hspace{0.17em}10}^{12}\mu _B`$ with $`10^4\stackrel{<}{_{}}\mathrm{\Delta }r/\text{pc}\stackrel{<}{_{}}\mathrm{\hspace{0.17em}1}`$. This relatively small value of $`\delta m^2`$ is also interesting in the context of sun and supernovae . Thus, for $`\mu `$ of the order of $`10^{12}\mu _B`$, the $`\nu _\tau `$ flux may be higher than the expected one from GRBs, that is, $`F_\tau /F_e>\mathrm{\hspace{0.17em}10}^5`$ due to neutrino spin-flavor precession effects. The neutrino spin-flavor precession effects are essentially determined by the product $`\mu B`$ so one may rescale $`\mu `$ and $`B`$ to obtain the same results. For $`\delta 2\mu B`$ and $`\delta >\mathrm{\hspace{0.17em}2}\mu B`$, we obtain from Eq. (14), an energy dependent $`P`$ such that $`P<\mathrm{\hspace{0.17em}1}/2`$.
With non vanishing $`\mathrm{\Delta }f`$ ($`\mathrm{\Delta }f\mathrm{\hspace{0.17em}0}`$), a resonant character in neutrino spin-flavor precession can be obtained for a range of values of relevant neutrino mixing parameters<sup>*</sup><sup>*</sup>*From above discussion, it follows that $`E`$ dependent/independent spin-flavor precession may also be obtained for non zero $`\mathrm{\Delta }f`$, however, given the current status of the high energy neutrino detection, for simplicity, we ignore these possibilities which tend to overlap with this case for a certain range of relevant neutrino mixing parameters; for details of these possibilities in the context of AGN, see .. Two conditions are essential to obtain a resonant character in neutrino spin-flavor precession: the level crossing and the adiabaticity at the level crossing (resonance). The level crossing condition is obtained by taking $`\mathrm{\Delta }H=\mathrm{\hspace{0.17em}0}`$ and is given by:
$$\delta m^2\mathrm{\hspace{0.17em}10}^3\text{eV}^2\left(\frac{|\mathrm{\Delta }f|}{10^{28}}\right).$$
(15)
These $`\mathrm{\Delta }f`$ values are well below the relevant upper limits on $`\mathrm{\Delta }f`$ which are typically in the $`10^{20}`$ range . Conversely speaking, the prospective detection of high energy neutrinos from cosmologically distant GRBs may be sensitive to $`\mathrm{\Delta }f`$ values as low as $`10^{28}`$. The other essential condition, namely, the adiabaticity in the resonance reads
$$\kappa \frac{2(2\mu B)^2}{|\text{d}V_G/\text{d}r|}\stackrel{>}{_{}}\mathrm{\hspace{0.17em}1}.$$
(16)
Note that here $`\kappa `$ depends explicitly on $`E`$ through $`V_G`$ unlike the case of ordinary neutrino spin-flip induced by the matter effects. A resonant character in neutrino spin-flavor precession is obtained if $`\kappa \stackrel{>}{_{}}\mathrm{\hspace{0.17em}1}`$ such that Eq. (15) is satisfied. We notice that $`B_{ad}/B_{GRB}\stackrel{<}{_{}}\mathrm{\hspace{0.17em}1}`$ for $`\mu 10^{12}\mu _B`$. Here $`B_{ad}`$ is obtained by setting $`\kappa \mathrm{\hspace{0.17em}1}`$ in Eq. (16). The general expression for relevant neutrino spin-flavor conversion probability is given by
$$P(\overline{\nu }_e\nu _\tau )=\frac{1}{2}\left(\frac{1}{2}P_{LZ}\right)\mathrm{cos}2\theta _f\mathrm{cos}2\theta _i,$$
(17)
where $`P_{LZ}=\mathrm{exp}(\frac{\pi }{4}\kappa )`$ and $`\mathrm{tan}2\theta _i=(2\mu B)/\mathrm{\Delta }H`$ is being evaluated at the high energy neutrino production site in the vicinity of GRB, whereas $`\mathrm{tan}2\theta _f=(2\mu B)/\delta `$ is evaluated at the exit. In Fig. 1, we plot $`P(\overline{\nu }_e\nu _\tau )`$ given by Eq. (17) as a function of $`\mathrm{\Delta }f`$ as well as $`\delta m^2`$ with $`E5\times 10^6`$ GeV. Four equi $`P`$ contours are also shown in Fig. 1. Note that the resonant spin-flavor precession probability is relatively small ($`P<1/2`$) for $`\mathrm{\Delta }f\stackrel{>}{_{}}10^{26}`$ essentially irrespective of $`\delta m^2`$ values. The expected spectrum $`F_\tau `$ of the high energy tau neutrinos originating from GRBs due to spin-flavor conversions is calculated as
$$F_\tau P(\overline{\nu }_e\nu _\tau )F_e^0.$$
(18)
The energy dependence in $`F_\tau `$ is now evident \[as compared to $`F_\tau `$ given by Eq. (5)\] when we convolve $`P(\overline{\nu }_e\nu _\tau )`$ given by Eq. (17) with $`F_e^0`$ taken from Ref. . The degree of energy dependence clearly depends on the extent of spin-flavor conversions. With the improved information on either $`\mathrm{\Delta }f`$ and/or $`\mu `$, one may be able to distinguish between the situations of resonant and non resonant spin-flavor precession induced by an interplay of VEP and $`\mu `$ in $`B_{GRB}`$.
Let us now consider briefly the effects of non vanishing gravity mixing angle $`\theta _G`$ for vanishing neutrino magnetic moment. In the case of massless or degenerate neutrinos, the corresponding vacuum flavor oscillation analog for $`\nu _e\nu _\tau `$ is obtained through $`\theta \theta _G`$ and $`\frac{\delta m^2}{4E}V_G`$ in the standard flavor precession probability formula in 2 flavor approxiamtion. For maximal $`\theta _G`$, the sensitivity of $`\mathrm{\Delta }f`$ may be estimated by equating the argument of second $`\mathrm{sin}`$ factor equal to $`\pi /2`$ in the corresponding expression for $`P`$ . This implies $`\mathrm{\Delta }f\mathrm{\hspace{0.17em}10}^{41}`$ with $`\varphi (r)\varphi _{SC}(r)`$. This value of $`\mathrm{\Delta }f`$ is of the same order of magnitude as that expected for neutrinos originating from AGNs. In case of non zero $`\delta m^2`$, a resonant or/and non resonant flavor conversion between $`\nu _e`$ and $`\nu _\tau `$ in the vicinity of a GRB is also possible due to an interplay of vanishing/non vanishing vacuum and gravity mixing angles. For instance, a resonant flavor conversion between $`\nu _e`$ and $`\nu _\tau `$ may be obtained if $`\mathrm{sin}^22\theta _G\mathrm{\hspace{0.17em}0.25}`$ with $`\mathrm{\Delta }f10^{31}`$ ($`\theta \mathrm{\hspace{0.17em}0}`$). Here the relevant level crossing may occur at $`r\mathrm{\hspace{0.17em}0.1}`$ pc with corresponding $`\delta m^2\mathrm{\hspace{0.17em}10}^6`$ eV<sup>2</sup>.
## IV Signatures of high energy $`\nu _\tau `$ in neutrino telescopes
The km<sup>2</sup> surface area under water/ice high energy neutrino telescopes may be able to obtain first examples of high energy $`\nu _\tau `$, through double showers, originating from GRBs correlated in time and direction with corresponding gamma-ray burst or may at least provide relevant useful upper limits . The first shower occurs because of deep inelastic charged current interaction of high energy tau neutrinos near/inside the neutrino telescope producing the tau lepton (along with the first shower) and the second shower occurs due to (hadronic) decay of this tau lepton.
The calculation of down ward going contained but separable double shower event rate for a km<sup>2</sup> surface area under ice/water neutrino telescope can be carried out by replacing the muon range expression with the tau one ($`E(1y)\tau c/m_\tau c^2`$) and then subtracting it from the linear size of a typical high energy neutrino telescope in the event rate formula while using the expected $`\nu _\tau `$ flux spectrum given by Eq. (5) and/or by Eq. (18). Here, $`y`$ is the fraction of the neutrino energy carried by the hadrons in lab frame. Thus, $`(1y)`$ is the fraction of energy transferred to the associated tau lepton having life time $`\tau c`$ and mass $`m_\tau c^2`$. We take here $`y0.25`$ . The condition of containdness of the two showers is obtained by requiring that the separation between the two showers is less than the typical $``$ km size of the neutrino telescope. It is obtained by equating the range of tau neutrino induced tau leptons with the linear size of detector implying $`E\stackrel{<}{_{}}\mathrm{\hspace{0.17em}2}\times 10^7`$ GeV. The condition of separableness of the two showers is obtained by demanding that the separation between the two showers is larger than the typical spread of the showers such that the amplitude of the second shower is essentially 2 times the first shower. This leads to $`E\stackrel{>}{_{}}\mathrm{\hspace{0.17em}2}\times 10^6`$ GeV . Thus, the two showers may be separated by a $`\mu `$-like track within these energy limits. To calculate the event rates, we use Martin Roberts Stirling (MRS 96 R<sub>1</sub>) parton distributions and present event rates in units of yr<sup>-1</sup> sr<sup>-1</sup>. We have checked that other recent parton distributions give quite similar event rates and are therefore not depicted here. Following , we present in Table I, the expected contained but separable double shower event rates for down word going $`\nu _\tau `$ in km<sup>2</sup> size under water/ice Čerenkov high energy neutrino telescopes for illustrative purposes. In Table I, the vacuum oscillation situation is essentially independent of the choice of the oscillation solution to solar neutrino problem. From Table I, we notice that the event rates for neutrino flavor/spin-flavor precession are up to $``$ 5 orders of magnitude higher than that for typical intrinsic (no oscillations) tau neutrino flux.
The possibility of measuring the contained but separable double shower events may enable one to distinguish between the high energy tau neutrinos and electron and/or muon neutrinos originating from cosmologically distant GRBs while providing useful information about the relevant energy interval at the same time. The chance of having double shower events induced by electron and/or muon neutrinos is negligibly small for the relevant energies . Collective information about directionality of the source, rate and energy dependence of neutrino fluxes will be needed to possibly isolate the mechanism of neutrino oscillation. The up ward going tau neutrinos at these energies may lead to a small pile up of up ward going $`\mu `$-like events near (10$`{}_{}{}^{4}10^5`$) GeV with fairly flat zenith angle dependence .
We now briefly discuss the potential of the under water/ice high energy neutrino telescopes to possibly determine an observational consequence of neutrino spin-flip in GRB induced by VEP. In the electron neutrino channel, the $`\overline{\nu }_e`$ interaction rate (integrated over all angles) is estimated to be an order of magnitude higher than that of $`(\nu _e+\overline{\nu }_e)`$ per Megaton year . This an order of magnitude difference in interaction rate of down ward going $`\overline{\nu }_e`$ is due to Glashow resonance encountered by $`\overline{\nu }_e`$ with $`E\stackrel{>}{_{}}\mathrm{\hspace{0.17em}10}^6`$ GeV when $`\overline{\nu }_e`$ interact with electrons inside the detector as compared to corresponding deep inelastic scattering. The up ward going $`\overline{\nu }_e`$, on the other hand, while passing through the earth, at these energies, are almost completely absorbed by the earth mainly due to same resonant effect. Thus, for instance, if $`E\mathrm{\hspace{0.17em}6.4}\times 10^6`$ GeV, an energy resolution $`\mathrm{\Delta }E/E\mathrm{\hspace{0.17em}2}\mathrm{\Gamma }_W/M_W\mathrm{\hspace{0.17em}1}/20`$, where $`\mathrm{\Gamma }_W`$ 2 GeV is the width of Glashow resonance and $`M_W`$80 GeV, may be needed to empirically differentiate between $`\overline{\nu }_e`$ and $`(\nu _e+\overline{\nu }_e)`$. The existing/planned high energy neutrino telescopes may thus in principle attempt to measure the $`\nu _e/\overline{\nu }_e`$ ratio in addition to identifying ($`\nu _\tau +\overline{\nu }_\tau `$) and ($`\nu _\mu +\overline{\nu }_\mu `$) events separately.
This feature may be utilized, for instance, to explain a situation in which a change in $`\nu _e/\overline{\nu }_e`$ ratio is observed as compared to GRB neutrino flux predictions in . This situation, if realized obsevationally may be an evidence for the neutrino spin-flip in GRB due to VEP, provided if neutrinos and antineutrinos couple differently to gravity. This follows from the possibility discussed in previous Sect. that an interplay between VEP and neutrino magnetic moment in $`B_{GRB}`$ may leads to conversions in either $`\nu _e`$ or $`\overline{\nu }_e`$ channel but not in both channels simultaneously.
## V Results and discussion
1. Intrinsically, the flux of high energy cosmic tau neutrinos is quite small, relative to non tau neutrino flavor, typically being $`F_\tau ^0/F_{e,\mu }^0<\mathrm{\hspace{0.17em}10}^5`$ (whereas $`F_e^0/F_\mu ^01/2`$) from cosmologically distant GRBs.
2. Because of neutrino oscillations, this ratio can be greatly enhanced. In the context of three flavor neutrino mixing scheme which can accommodate the oscillation solutions to solar and atmospheric neutrino deficits in terms of oscillations between three active neutrinos, the final down ward going ratio of fluxes of high energy cosmic neutrinos on earth is $`F_eF_\mu F_\tau F_e^0`$, essentially irrespective of the oscillation solution to solar neutrino problem.
The (vacuum) flavor oscillations leads to an essentially energy independent flux of high energy neutrinos of all flavors originating from cosmologically distant GRBs at the level of electron neutrino flux, whereas spin-flavor precessions/conversions may lead to an energy dependence or/and change in this situation.
The spin-flavor conversions may occur possibly through several mechanisms. We have discussed in some detail mainly the spin-flavor precession/conversion situation induced by a non zero neutrino magnetic moment and by a relatively small VEP as an example to point out the possibility of obtaining some what higher tau neutrino fluxes as compared to no oscillations/conversions scenarios from GRBs.
The matter density in the vicinity of GRB is quite small (up to 4$``$5 orders of magnitude) to induce any resonant flavor/spin-flavor neutrino conversion due to normal matter effects. We have pointed out that a resonant character in the neutrino spin-flavor conversions may nevertheless be obtained due to possible VEP. The corresponding degree of VEP may be $`(10^{35}10^{25}`$) depending on $`\delta m^2`$ value for vanishing gravity mixing angle.
3. This enhancement in high energy cosmic tau neutrino flux may lead to the possibility of its detection in km<sup>2</sup> surface area high energy neutrino telescopes. For $`2\times 10^6E`$/GeV $`2\times 10^7`$, the down ward going high energy cosmic tau neutrinos may produce a double shower signature because of charged current deep inelastic scattering followed by a subsequent hadronic decay of associated tau lepton
The double shower event rate for intrinsic (no oscillations/conversions) high energy tau neutrinos originating from GRBs turns out to be small as compared to that due to precession/conversion effects up to a factor of $`\mathrm{\hspace{0.17em}10}^5`$. Thus, the high energy neutrino telescopes may possibly provide useful upper bounds on intrinsic properties of neutrinos such as mass, mixing and magnetic moment, etc.. The relevant tau neutrino energy range for detection in km<sup>2</sup> surface area under water/ice neutrino telescopes may be $`2\times 10^6\stackrel{<}{_{}}E/\text{GeV}\stackrel{<}{_{}}\mathrm{\hspace{0.17em}2}\times 10^7`$ through characteristic contained but separable double shower events.
Observationally, the high energy $`\nu _\tau `$ burst from a GRB may possibly be correlated to the corresponding gamma-ray burst/highest energy cosmic rays (if both have common origin) in time and in direction thus raising the possibility of its detection. If the range of neutrino mixing parameters pointed out in this study is realized terrestially/extraterrestially then a relatively large (energy dependent) $`\nu _\tau `$ flux from GRBs is expected as compared to no oscillation/conversion scenario.
#### Acknowledgments.
The author thanks Japan Society for the Promotion of Science for financial support.
|
warning/0004/cond-mat0004469.html
|
ar5iv
|
text
|
# Exact versus mean-field description of the Bose-Einstein condensate: a model study
## I Introduction
Recent advances in the trapping techniques have renewed interest in various aspects of many body theory. In fact a cloud of weakly interacting trapped atoms is an ideal system for which various aspects of many body theory can be tested and verified. The ideal bosonic gas undergoes the Bose-Einstein condensation if the phase-space density exceeds one. This phenomenon manifests itself by the macroscopic occupation of the single particle ground state. In the case of an interacting system the condensate wave function can be defined by the spectral decomposition of the one-body density matrix. This decomposition is closely related to the off-diagonal long range order or the existence of the order parameter, i.e., the ‘classical’ field with given amplitude and phase commonly used in the theory of superfluidity . Realization of decomposition procedure is practically impossible because it requires a full solution of the many-body problem. Mean-field approaches are commonly used instead. The basic idea for a mean-field description of the dilute, weakly interacting Bose gas below transition temperature was introduced by Bogoliubov . Most of the results for the interacting Bose-Einstein condensate are obtained within the Bogoliubov theory which in many cases provides a reliable quantitative description of the quantum Bose gas. Indeed, the low energy excitation spectrum of the trapped condensate well below transition temperature agrees remarkably well with predictions based on the Bogoliubov theory . On the other hand the mean-field Bogoliubov approach fails to reproduce excitation spectrum at higher temperatures – close to the transition point . Therefore, the question about limits of validity of the Bogoliubov method is of great importance. One possible way to test the quality of this approximation is to go beyond the mean field theory. Another possibility to assess the usefulness of Bogoliubov’s theory is to study the exactly soluble models and to compare their predictions with those based on the approximate method. The models provide not only a unique soluble many body problem but also allow to verify various approximations. This will shed some light on the validity and exactness of the Bogoliubov method, widely used in many body physics .
There are only few exactly soluble models of quantum systems where the interactions between atoms is chosen in the form allowing for the exact analytic solution. These are: (i) the one-dimensional model of impenetrable bosons introduced by Girardeau , (ii) its contact potential version formulated by Lieb ; (iii) the model of particles interacting by harmonic forces . Although in the first two cases the formal solution is given but in practice the problem is still quite complicated and quantitative calculations can be done for a very small number of particles only . The latter case seems to be much simpler because, as it has been shown in , it can be reduced to the problem of noninteracting particles in a harmonic trap. Therefore in the following we are going to examine, within this exactly soluble model, various concepts and methods related to the interacting Bose-Einstein condensate.
This paper is organized as follows. In Sec. II we present exact results regarding properties of harmonically interacting bosons trapped within harmonic potential. The results are obtained within a soluble model that was developed in . In the second part of Sec. II we find the analytic expression for the order parameter and study the effect of quantum depletion of the condensate as well as quantum fluctuations at zero temperature. In the third part we analyze the thermal properties of the system. In Sec. III we apply the Bogoliubov method to our model. Within this approximation we first determine a condensate wave function and an excitation spectrum. Using the Bogoliubov spectrum we calculate occupation of the condensate and its fluctuation at finite temperatures. We compare these mean-field results to the results obtained within the exact model. We finish in Sec. IV with some concluding remarks.
## II Exact results
### A Ground state and excitation spectrum
In our previous paper we have shown the algebraic method of diagonalization of the Hamiltonian describing a system of many particles interacting via harmonic forces. The system under consideration consists of many particles confined by an external harmonic potential interacting by harmonic forces, i.e., two body interaction potential has the form:
$$V(𝐱_i𝐱_j)=\frac{\sigma }{2}\mathrm{\Omega }^2(𝐱_i𝐱_j)^2,$$
(1)
where $`\mathrm{\Omega }`$ defines the interaction strength and $`\sigma =+1`$ signifies the attractive interaction of particles placed at positions $`𝐱_i`$ and $`𝐱_j`$ whereas $`\sigma =1`$ – corresponds to repulsive interactions. The total Hamiltonian of the $`N`$-particle system has therefore the following form:
$$H=\underset{i=1}{\overset{N}{}}\frac{1}{2}(𝐩_i^2+𝐱_i^2)+\underset{i<j}{}V(𝐱_i𝐱_j).$$
(2)
Let us first recall the exact results of . For the sake of simplicity we denote the set of all particle positions vectors by $`𝐗_N=(𝐱_1,\mathrm{},𝐱_N)`$. The Hamiltonian can be easily diagonalized if one introduces collective variables:
$$𝐗_N^c=𝒬_𝒩𝐗_N,$$
(3)
where $`𝐗_N^c=(𝐱_1^c,\mathrm{},𝐱_N^c)`$ and the matrix $`𝒬_N=\{q_{ij}^N\}`$ is orthogonal. One of these collective variables namely the center of mass of $`N`$-particle system plays a particularly important role:
$$𝐱_N^c=\frac{1}{\sqrt{N}}\underset{i=1}{\overset{N}{}}𝐱_i.$$
(4)
The choice of $`N1`$ remaining collective variables $`𝐗_{N1}^c=(𝐱_1^c,\mathrm{},𝐱_{N1}^c)`$ is not unique but this does not lead to any physical implications. In particular:
$$(𝐗_{N1}^c)^2=\underset{i=1}{\overset{N1}{}}(𝐱_i^c)^2=\underset{i=1}{\overset{N}{}}𝐱_i^2(𝐱_N^c)^2.$$
(5)
In the following we are going to use a similar notation for description of a subsystem of $`s`$-particles, $`s=1,\mathrm{},N`$.
The above-defined transformation brings the Hamiltonian to the diagonal form and its eigenenergies can be easily found. While determining a spectrum, however, one must take into account the proper symmetry of a total wave function. In the case of bosonic particles ($`N>2`$) the allowed energies are:
$$E=\left(\frac{3}{2}+m\right)+\left(\frac{3}{2}(N1)+n\right)\omega ,$$
(6)
where $`m=0,1,2\mathrm{}`$, $`n=0,2,3\mathrm{}`$ and $`\omega =\sqrt{1+\sigma N\mathrm{\Omega }^2}`$. The first term describes excitations of the center of mass, i.e., $`d`$-dimensional harmonic oscillator of frequency equal to one. The second term in the Eq. (6) corresponds to excitations of $`N1`$ relative degrees of freedom. The frequency $`\omega `$ characterizes some effective potential felt by an individual quasi-particle because it results from a combined effect of all particles of our system. Let us observe that $`\omega =1`$ corresponds to the noninteracting case, the repulsive interactions give $`0<\omega <1`$ while attractive forces lead to $`\omega >1`$. Moreover, very small values of $`\omega 0`$ signify very strong repulsion which almost destabilizes the whole system. It is very convenient to parameterize $`\omega `$ by an exponent $`\kappa `$ defined in the following way:
$$\omega =N^\kappa .$$
(7)
This exponent can be related to the actual strength of the interaction. In fact, for weakly interacting gas ($`\omega 1`$) we obtain very small values of this parameter: $`\kappa 0`$, while for strong interactions ($`\omega 0`$ – repulsion, $`\omega 1`$ – attraction) we have $`|\kappa |1`$. Moreover, $`\kappa `$ is positive in the case of attraction while it is negative for repulsion. Let us add at this point that in realistic situations of short-range interparticle interactions, large Bose-Einstein condensates can exist only for repulsive forces. In the case of attraction the size of the trapped condensate is limited to about 1500 atoms . In our oscillatory model the forces between particles are negligible at small distances, therefore the model leads to the condensation (in the thermodynamic limit) in both attractive and repulsive case.
The ground state of the system is the following:
$$\mathrm{\Psi }(𝐗_N)=\mathrm{\Phi }_0(\sqrt{\omega }𝐗_{N1}^c)\mathrm{\Phi }_0(𝐱_N^c),$$
(8)
where $`(𝐗_{N1}^c,𝐱_N^c)=𝒬_N𝐗_N`$ and the function $`\mathrm{\Phi }_0(\sqrt{\omega }𝐗_{N1}^c)`$ corresponds to the ground state of a system of $`N1`$ independent quasi-particles (in $`d`$ spatial dimensions) interacting with an external potential of the harmonic oscillator of frequency $`\omega `$:
$$\mathrm{\Phi }_0(\sqrt{\omega }𝐗_{N1}^c)=\left(\frac{\omega }{\pi }\right)^{d(N1)/4}\mathrm{exp}\left[\omega (𝐗_{N1}^c)^2/2\right],$$
(9)
and $`\mathrm{\Phi }_0(𝐱_N^c)`$ is the ground state of the single particle (center of mass) trapped into harmonic potential:
$$\mathrm{\Phi }_0(𝐱_N^c)=\left(\frac{1}{\pi }\right)^{d/4}\mathrm{exp}\left[(𝐱_N^c)^2/2\right].$$
(10)
Construction of excited eigenstates is difficult because it is not easy to impose the desired symmetry on the wave function. Such a procedure was describe in details in .
### B Order parameter and quantum depletion
If the energy of the system (or equivalently the temperature) is sufficiently small we expect that the system forms a Bose-Einstein condensate. The BEC of the ideal gas manifests itself by a macroscopic occupation of the single particle ground state. In the case of interacting system it is not obvious what is this particular state which is ‘macroscopically occupied’. The identification of the macroscopically occupied quantum state is equivalent to the definition of the order parameter – the single particle wave function which is inherently related to the Bose condensation. The condensate subsystem can be then quite accurately described by the $`N_0`$-fold product of the order parameter, where $`N_0𝒪(N)`$ is the occupation of the condensate. In the conventional approaches, for example in the Bogoliubov method, it is simply assumed that a mean value of the boson field operator is different than zero and this mean value is associated with the macroscopically occupied state. Then, consistently with the above assumption, the Bogoliubov equations give in fact the nonzero solution for the order parameter. However, because of the superselection rules (resulting from the conservation of the barionic charge) any $`N`$-particle system must be in the Fock state – the state with a well defined particle number. Therefore the mean value of the boson field operator must vanish in this state as the field operator changes the number of particles.
In the following we use our model to demonstrate how to define the order parameter, occupation of the condensate, and its fluctuations. At zero temperature the system is in the ground state and one might naively expect that it is totally Bose condensated. However, the ground state of the N-particle bosonic system is not equivalent to the Bose-Einstein condensate. Interactions can significantly deplete the condensate. We are going to show this effect in the most spectacular but also in relatively simple case of the zero temperature.
Let us now define the hierarchy of the reduced $`s`$-particle density matrices which can be conventionally obtained by averaging the density matrix of the total system of $`N`$ particles over the degrees of freedom of $`Ns`$ remaining particles. For a given $`N`$-particle quantum state $`\mathrm{\Psi }(𝐗_N)`$ the corresponding $`s`$-particle reduced density matrix $`\rho _s(𝐗_s;𝐘_s)`$ is defined by:
$$\rho _s(𝐗_s;𝐘_s)=d𝐑_{Ns}\mathrm{\Psi }^{}(𝐗_s,𝐑_{Ns})\mathrm{\Psi }(𝐘_s,𝐑_{Ns}).$$
(11)
We use previously defined shorthand notation for vectors in a configuration space of $`s`$-particles. The reduced density matrix describes the subsystem of $`s`$-particles and can be directly related to different measurement processes. For the statistical description of the system one should first of all define the statistical density matrix by averaging all $`N`$-particle density matrices with the appropriate statistical weights depending on the ensemble. In general it is quite a complicated task but at zero temperature there is only one quantum state of the system and no statistical averaging is necessary.
The total wave function (or density matrix) carries all the information about the system. In real experiments however one does not probe simultaneously all the particles. Typical detection scheme consists on the measurement of one or at most few particles at a given time. In other words a single measurement process is reduced to a subsystem of small number of particles. Such subsystems are described by reduced density matrices. In the considered here case of zero temperature the description of the interacting system the $`s`$-particle density matrix can be brought to the following form:
$$\rho _s(𝐗_s;𝐘_s)=\rho ^{CM}(𝐱_s^c,𝐲_s^c)\mathrm{\Phi }_0(\sqrt{\omega }𝐗_{s1}^c)\mathrm{\Phi }_0(\sqrt{\omega }𝐘_{s1}^c).$$
(12)
The functions $`\mathrm{\Phi }_0`$ describes the ground state of $`s1`$ quasi-particles (collective relative coordinates, see Eq. (9)) while $`\rho ^{CM}`$ corresponds to the density matrix of center of mass of the subsystem:
$`\rho ^{CM}(𝐱_s^c,𝐲_s^c)`$ $`=`$ $`\left({\displaystyle \frac{\omega _s}{\pi }}\right)^{d/2}\mathrm{exp}\left[{\displaystyle \frac{\delta _s}{2}}𝐱_s^c𝐲_s^c\right]`$ (14)
$`\mathrm{exp}\left[{\displaystyle \frac{1}{2}}\left(\omega _s+{\displaystyle \frac{\delta _s}{2}}\right)\left[(𝐱_s^c)^2+(𝐲_s^c)^2\right]\right].`$
The $`s`$-particles collective coordinates are defined in the familiar way: $`(𝐗_{s1}^c,𝐱_s^c)=𝒬_s𝐗_s`$, $`(𝐘_{s1}^c,𝐲_s^c)=𝒬_s𝐘_s`$ and frequencies $`\omega _s`$, $`\delta _s`$ as well as auxiliary parameter $`\gamma _s`$ are:
$`\gamma _s`$ $`=`$ $`1{\displaystyle \frac{s(1\omega )}{N}},`$ (16)
$`\omega _s`$ $`=`$ $`{\displaystyle \frac{\omega }{\gamma _s}},`$ (17)
$`\delta _s`$ $`=`$ $`\left({\displaystyle \frac{1\omega }{N}}\right)^2{\displaystyle \frac{s(Ns)}{\gamma _s}}.`$ (18)
Having defined the $`s`$-particle matrices we are ready now to analyze the nature of the Bose-Einstein condensation of the interacting system and to discuss the meaning of the order parameter. To this end we write the density matrix Eq. (12) in the diagonal form:
$$\rho _s(𝐗_s;𝐘_s)=\underset{𝐧}{}\lambda _𝐧^{(s)}\varphi _𝐧^{(s)}(𝐗_s)\varphi _𝐧^{(s)}(𝐘_s).$$
(19)
The function $`\varphi _𝐧^{(s)}(𝐗_s)`$ can be treated as the wave function of the $`s`$-particle subsystem:
$$\varphi _𝐧^{(s)}(𝐗_s)=\mathrm{\Phi }_0(\sqrt{\omega }𝐗_{s1}^c)\mathrm{\Phi }_𝐧(\sqrt{\alpha _s}𝐱_s^c),$$
(20)
where $`\mathrm{\Phi }_0(\sqrt{\omega }𝐗_{s1}^c)`$ is the ground state wave function of the relative degrees of freedom. This function corresponds to the ground state of $`s1`$ noninteracting quasi-particles (in $`d`$-spatial dimensions) subject to the external harmonic potential of frequency $`\omega `$. The second part of the Eq. (20) describes states of the center of mass of $`s`$-particles; $`\mathrm{\Phi }_𝐧`$ is simply the $`d`$-dimensional harmonic oscillator wave function corresponding to the effective center of mass frequency $`\alpha _s`$. Quantum numbers $`𝐧=(n_1,\mathrm{},n_d)`$ label different states of the center of mass while $`n=n_1+\mathrm{}+n_d`$ corresponds to the energy of the given state. The effective center of mass frequency $`\alpha _s`$ is:
$$\alpha _s=\left[\omega _s(\omega _s+\delta _s)\right]^{1/2}.$$
(21)
It is interesting to observe that all the frequencies of the relative motion of the $`s`$-particles subsystem are exactly the same as the frequencies of the relative motion of the whole system, i.e., equal to $`\omega `$. On the other hand the center of mass oscillation frequency of the subsystem is neither equal to $`\omega `$ nor to 1 (trap frequency). This collective degree of freedom couples to the center of mass of $`Ns`$ remaining particles what leads to some energy shift. Finally, the eigenvalues $`\lambda _𝐧^{(s)}`$ of $`\rho _s`$ are equal to the occupation probabilities of a given $`s`$-particle state:
$$\lambda _𝐧^{(s)}=\left(\frac{\omega _s}{\alpha _s}\right)^{d/2}\left(\frac{2\sqrt{\omega _s\alpha _s}}{\omega _s+\alpha _s}\right)^d\left(\frac{\alpha _s\omega _s}{\alpha _s+\omega _s}\right)^n.$$
(22)
It follows from the normalization condition for the density matrix that $`_𝐧\lambda _𝐧^{(s)}=1`$.
The spectral decomposition of the reduced single particle density matrix gives natural single-particles states $`\varphi _𝐧^{(1)}(𝐱)`$. These states are crucial for the definition of the condensate wave function (order parameter). It can be seen from Eq.(22) that if $`N`$ goes to infinity (thermodynamic limit) with fixed value of the interaction frequency $`\omega `$ the lowest eigenvalue $`\lambda _0^{(1)}`$ dominates the others:
$`\lambda _0^{(1)}`$ $``$ $`1,`$ (23)
$`\lambda _𝐧^{(1)}`$ $``$ $`\left({\displaystyle \frac{(1\omega )^2}{4N}}\right)^n,\mathrm{if}n0.`$ (24)
This behavior signifies nothing else but the onset of the Bose-Einstein condensation. The single particle density matrix becomes very close to the pure state because with quite good accuracy it can be approximated by $`\rho _1(𝐱,𝐲)\varphi _0^{(1)}(𝐱)\varphi _0^{(1)}(𝐲)`$. This particular single-particle ground state $`\varphi _0^{(1)}(𝐱)`$ is usually called the order parameter. The $`N`$-particle wave function can be quite accurately approximated by the $`N`$-fold product of the order parameter.
Above the critical temperature the situation is completely different, namely all eigenvalues of $`\rho _1`$ should be close to zero what means that the single particle reduced density matrix is ‘very far’ from the pure state and order parameter vanishes – there is no leading state in the spectral decomposition of the single-particle density matrix.
Our analytic formula allows to study quantitatively the role of interactions on the Bose-Einstein condensate. On the basis of the discussion it is obvious that the average occupation of the condensate becomes:
$$N_0=Nd𝐱d𝐲\varphi _0^{(1)}(𝐱)\rho _1(𝐱,𝐲)\varphi _0^{(1)}(𝐲)=N\lambda _0^{(1)}.$$
(25)
In the case of the ideal gas at zero temperature the above equation gives, of course, $`N_0=N`$; all particles occupy the single particle ground state. For a fixed number of particles, if the interaction strength $`|\mathrm{log}\omega |`$ grows, the occupation of the condensate decreases. This behavior is presented in Fig. 1 where we show the mean occupation of the condensate versus the exponent $`\kappa =\mathrm{log}\omega /\mathrm{log}N`$ for different values of particle number $`N`$ in three spatial dimensions ($`d=3`$). The values of $`\kappa `$ less than zero signify repulsive interactions while $`\kappa >0`$ corresponds to attraction. One can easily see that if the interaction becomes strong ($`|\kappa |1`$) the condensate is almost totally depleted. All curves presented in the figure tend to an universal curve if the number of particles increases. When $`N`$ increases to infinity with $`\kappa `$ being constant then our expression for the occupation of the condensate has the form:
$$\frac{N_0}{N}=\left(\frac{2}{1+\sqrt{N^{\kappa 1}+N^{(\kappa +1)}+1}}\right)^d.$$
(26)
The above formula, valid in the thermodynamic limit, gives an universal critical behavior. It exhibits no depletion ($`N_0=N`$) for $`|\kappa |<1`$ followed by an abrupt jump and total destruction of the condensate ($`N_0=0`$) for $`|\kappa |>1`$.
The effect of quantum depletion of the trapped atomic condensate with a short range interactions, for the realistic experimental parameters, has been estimated to be of the order of $`1\%`$ . This is opposite to the case of superfluid helium where this effect accounts for depletion as large as more than $`90\%`$ . Our model exhibits very interesting feature. It shows that in large $`N`$ limit the quantum effects are almost negligible or totally destroy the condensate depending on the value of the interaction strength. At this point it is not clear if this is an unique feature of our model or if it is a more general result.
We see that interactions play an important role. If they are strong, the condensate disappears although the $`N`$-particle system remains in its ground state. There is no coherence in the strongly interacting system, i.e., no wave function can be assign to a single particle subsystem. At this point we want to make a comment about the notion of the coherence of the Bose-Einstein condensate. Approximate methods assume explicitly that the mean value of the boson field operator is different from zero in the case of the Bose-Einstein condensate. Therefore, the folk wisdom associates the condensate with the coherent state – the analog of the coherent state of the electromagnetic field. This analogy is of limited value and in fact may be misleading because the condensate must be in a Fock state in which a mean value the field operator vanishes. However there is coherence in the condensate in the sense that majority of particles are described by the same wave function with the same phase. The expression for this wave function can be obtained rigorously only when one considers the single particle reduced density matrix.
The 2-particle reduced density matrix allows to find a joint probability of finding one particle in a given single particle state and simultaneously another particle in another given state. In particular we have:
$`N_0(N_01)`$ $`=`$ $`N(N1){\displaystyle d𝐗_2d𝐘_2\rho _2(𝐗_2,𝐘_2)}`$ (28)
$`\varphi _0^{(1)}(𝐱_1)\varphi _0^{(1)}(𝐱_2)\varphi _0^{(1)}(𝐲_2)\varphi _0^{(1)}(𝐲_1).`$
Simple integration gives:
$`N_0(N_01)`$ $`=`$ $`N(N1)\left({\displaystyle \frac{2\sqrt{\omega \alpha _1}}{\omega +\alpha _1}}\right)^d\left({\displaystyle \frac{2\sqrt{\omega _2\alpha _1}}{\omega _2+\alpha _1}}\right)^d`$ (30)
$`\times \left({\displaystyle \frac{\omega _2+\alpha _1}{\omega _2+\alpha _1+\delta _2}}\right)^{d/2}.`$
Now we are ready to analyze the fluctuations of the condensate defined as:
$$\delta ^2N_0=N_0^2N_0^2.$$
(31)
These fluctuations are shown in Fig.2. We see that as the interaction strength grows up (at fixed number of particles) the fluctuations start to grow from zero value for the ideal gas. However when the interactions become so strong that condensate practically disappears ($`|\kappa |1`$) fluctuations also decrease - as there is no condensate the fluctuations also die out. The fluctuations are maximal in a region of the critical destruction of the condensate by quantum effects.
### C Finite temperatures
In this subsection we estimate, within the exact model, some effects in finite-temperature behavior of the trapped gas. Rigorous description of the condensate requires a knowledge of the statistical density matrix of the $`N`$-particle system. Knowing this matrix one can apply the procedure described previously to define the finite temperature condensate, its occupation and fluctuations. However, because of a huge degeneracy of high energy states the statistical averaging procedure is difficult. Therefore, we limit our study to the case of weak interactions ($`|\kappa |<1`$) when we can neglect the quantum effects leading to a significant depletion of the condensate. In this case we can expect that the condensate wave function in a finite temperature is equal to the ground state of the harmonic oscillator with some effective frequency (characterizing a mean field experienced by a single particle) which at $`T=0`$ is equal to $`\alpha _1`$. In the case of a weak interaction this frequency can be approximated by $`\alpha _1\omega `$. We expect that, similarly to the noninteracting case, the main effect of the temperature is to deplete the condensate rather than modify the condensate wave function, i.e., the frequency $`\alpha _1`$.
As it has been shown in , the trace of the density matrix of the $`N`$-particle system, i.e., the microcanonical partition function $`\mathrm{\Gamma }(N,E)`$ is identical (in the thermodynamic limit) to the microcanonical partition function $`\mathrm{\Gamma }_0(N,E,\omega )`$ of the $`N`$ noninteracting bosonic particles (quasi particles) trapped by the harmonic potential of frequency $`\omega `$:
$$\mathrm{\Gamma }(N,E)\mathrm{\Gamma }_0(N,E,\omega ).$$
(32)
The effect of the center-of-mass excitations on the spectrum and on the statistical properties of the system is negligible since it is related to only one degree of freedom as compared to the $`N1`$ remaining collective degrees of freedom. This fact, together with our remarks about the condensate wave function, signifies that the system of interacting (via harmonic forces) particles is, in the thermodynamic limit, equivalent to the ideal gas in the oscillatory trap.
This observation allows to recall all results obtained for the ideal Bose gas . In particular, considered here interacting system undergoes the Bose-Einstein condensation at the temperature $`T_c`$ equal to:
$$T_c=\omega \left(\frac{N}{\zeta (3)}\right)^{1/3},$$
(33)
where the $`\zeta `$ is the Riemann function, $`\zeta (3)=1.2020569`$. This critical temperature should be compared to the critical temperature of the noninteracting system,
$$T_0=\left(\frac{N}{\zeta (3)}\right)^{1/3}.$$
(34)
The shift of the critical temperature $`\mathrm{\Delta }T=T_cT_0`$ is therefore equal to:
$$\frac{\mathrm{\Delta }T}{T_0}=\omega 1.$$
(35)
It is negative in the case of repulsive interaction and has the opposite sign for the attractive system what is in a qualitative agreement with results of Ref. for the system of trapped atoms. The critical temperature for the trapped gas with repulsive interactions is decreased as compared to the noninteracting case: due to interactions a mean separation between particles grows, therefore the quantum statistical effects become important at larger de Broglie wavelength.
It is worth to stress that the transition temperature for the system with short range interactions remains a controversial subject even in the uniform case . There is no consensus how the shift of the temperature should depend on the interaction strength, nor even the sign.
The fraction of condensate particles $`N_0`$ is:
$$N_0=N\left(\frac{T}{\omega }\right)^3\zeta (3).$$
(36)
It is known that the critical temperature and mean occupation of the ground state do not depend on the statistical ensemble used for the description of the system. The value of the fluctuations of the number of the particles in the condensate calculated for different statistical ensembles differ significantly . In the canonical ensemble the square of the fluctuations $`\delta ^2N_0_{\mathrm{CN}}=N_0^2_{\mathrm{CN}}N_0_{\mathrm{CN}}^2`$ is:
$$\delta ^2N_0_{\mathrm{CN}}=\left(\frac{T}{\omega }\right)^3\zeta (2),$$
(37)
where $`\zeta (2)=\pi ^2/6`$. In the case of perfectly isolated system (microcanonical ensemble) fluctuations of the ground state occupation are smaller than canonical ones :
$$\delta ^2N_0_{\mathrm{MC}}=\left(\frac{T}{\omega }\right)^3\zeta (2)\left(1\frac{3\zeta ^2(3)}{4\zeta (4)\zeta (2)}\right),$$
(38)
and $`\zeta (4)=\pi ^4/90`$.
## III Bogoliubov approximation
### A Ground state
In realistic cases the exact solution of the many-body Schrödinger equation is impossible. Instead, mean-field approaches are being developed. The basic idea of the mean-field theory was introduced by Bogoliubov. We will now formulate the Bogoliubov approximation in the case of inhomogeneous systems . Next we find the energy spectrum within this approximation for our exactly solvable model. In order to do this we rewrite the Hamiltonian (2) using the second quantization formalism. Thus we introduce the field operator $`\widehat{\psi }(𝐱)`$ which annihilates a particle at a point $`𝐱`$ and its conjugate $`\widehat{\psi }^{}(𝐱)`$ which creates a particle at a point $`𝐱`$. These operators fulfill standard bosonic commutation relations:
$$[\widehat{\psi }(𝐱),\widehat{\psi }^{}(𝐱^{})]=\delta (𝐱𝐱^{}).$$
(39)
In the second quantization formalism the Hamiltonian becomes:
$`\widehat{H}`$ $`=`$ $`{\displaystyle d𝐱\widehat{\psi }^{}(𝐱)H_0(𝐱)\widehat{\psi }(𝐱)}`$ (41)
$`+{\displaystyle \frac{1}{2}}{\displaystyle d𝐱𝑑𝐱^{}\widehat{\psi }^{}(𝐱)\widehat{\psi }^{}(𝐱^{})V(𝐱𝐱^{})\widehat{\psi }(𝐱)\widehat{\psi }(𝐱^{})},`$
where
$$H_0(𝐱)=\frac{1}{2}(𝐩^2+𝐱^2).$$
(42)
The Bogoliubov approximation is formulated in two steps. The first one is to express the field operator as a sum of its mean value $`\sqrt{N_0}\varphi _0(𝐱)`$ and an operator $`\widehat{\varphi }(𝐱)`$ responsible for the fluctuations around the mean value. Below the Bose-Einstein condensation temperature the occupation of the ground state is nonzero. It is therefore convenient to write
$$\widehat{\psi }(𝐱)=\sqrt{N_0}\varphi _0(𝐱)+\widehat{\varphi }(𝐱).$$
(43)
Note that now the mean value of the field operator $`\widehat{\psi }(𝐱)`$ is not equal to zero and is proportional to the square root of the number of condensed particles $`N_0`$. The spirit of the Bogoliubov approximation is based on the fact that the occupation of the condensate is of the order of total particles number $`N_0𝒪(N)`$. This, in principle, limits the Bogoliubov approach to low temperatures. In the following, consistently with the above assumption we will substitute in the Eq.(43) $`N_0`$ by $`N`$. For any physically realizable $`N`$-particle state, and in particular for our solutions of the system Eq.(2), the mean value of the field operator is zero. (Strictly, separation of the Eq.(43) should be done for operators conserving particle number. However, extracting a $`c`$–number part of the field operator has formally the same consequences and is easier to handle). If such a form is substituted into the total Hamiltonian Eq.(41) the number of particles is no longer conserved. To overcome this difficulty one considers the grand canonical Hamiltonian instead
$$\widehat{K}=\widehat{H}\mu \widehat{N},$$
(44)
where $`\widehat{N}`$ is the total particle number operator and $`\mu `$ is the chemical potential. It should be chosen in such a way that the mean particle number is equal to the desired value. Self consistent equation for the condensate wave function $`\varphi _0(𝐱)`$ follows from the assumption that the decomposition (43) gives the best self consistent function. We find:
$$\left\{H_0(𝐱)+V_{eff}[\varphi _0,𝐱]\right\}\varphi _0(𝐱)=\mu \varphi _0(𝐱),$$
(45)
where the effective potential is:
$$V_{eff}[\varphi _0,𝐱]=\frac{\sigma }{2}N\mathrm{\Omega }^2d𝐱^{}\varphi _0^{}(𝐱^{})(𝐱𝐱^{})^2\varphi _0(𝐱^{}).$$
(46)
This equation replaces the standard Gross-Pitaevskii equation for the condensate wave function. The effective potential in the Eq.(45) has different form than usuall nonlinear term appearing in the Gross-Pitaevskii equation because of long range forces assumed in our model as opposed to the more realistic zero range interactions.
The lowest energy state of the Hamiltonian from Eq.(45) is
$$\varphi _0(𝐱)=\left(\frac{\omega }{\pi }\right)^{\frac{3}{4}}\mathrm{exp}(\frac{1}{2}\omega 𝐱^2),$$
(47)
and the value of chemical potential is $`\mu =3/4(1+\omega )`$. The function Eq.(47) is the Bogoliubov approximation to the exact order parameter $`\varphi _0^{(1)}(\sqrt{\omega }𝐱)`$ found in the previous section:
$$\varphi _0^{(1)}(\sqrt{\omega }𝐱)=\left(\frac{\alpha _1}{\pi }\right)^{\frac{3}{4}}\mathrm{exp}(\frac{1}{2}\alpha _1𝐱^2),$$
(48)
where the effective frequency $`\alpha _1`$ in the limit of weak interactions can be approximated by:
$$\alpha _1\omega \left(1+\frac{1\omega }{N}\right).$$
(49)
Because the effective frequency $`\alpha _1`$ is very close to $`\omega `$ the Bogoliubov expression for the condensate wave function is quite accurate in the limit of weak interactions. However, if the interaction strength is large the Bogoliubov method fails to reproduce the condensate wave function. This is consistent with the basic assumption of the Bogoliubov method which requires the order parameter (multiplied by the mean occupation of the condensate) to be large. As we have shown in the previous section this is not the case for strongly interacting system.
### B Excitation spectra
The second step in the Bogoliubov method is to find the low energy excitation spectrum of the system by expanding the total Hamiltonian around the mean value of the field operator given by the solution Eq.(47). After substituting the field operator Eq.(43) into $`\widehat{K}`$ and retaining all terms up to $`𝒪(\widehat{\varphi }^2)`$, the operator $`\widehat{K}`$ can be diagonalized with the help of a canonical transformation:
$$\widehat{\varphi }(𝐱)=\underset{\lambda }{}\left(u_\lambda (𝐱)\beta _\lambda +v_\lambda ^{}(𝐱)\beta _\lambda ^{}\right).$$
(50)
where $`\beta _\lambda `$ and $`\beta _\lambda ^{}`$ are bosonic annihilation and creation operators. The diagonal form of the operator $`\widehat{K}`$ is
$$\widehat{K}=\underset{\lambda }{}\mathrm{\Delta }_\lambda d𝐱v_\lambda ^{}(𝐱)v_\lambda (𝐱)+\underset{\lambda }{}\mathrm{\Delta }_\lambda \beta _\lambda ^{}\beta _\lambda ,$$
(51)
provided that functions $`U_\lambda (𝐱)`$ and $`V_\lambda (𝐱)`$ defined as:
$`U_\lambda (𝐱)`$ $`=`$ $`u_\lambda (𝐱)+v_\lambda (𝐱),`$ (52)
$`V_\lambda (𝐱)`$ $`=`$ $`u_\lambda (𝐱)v_\lambda (𝐱),`$ (53)
satisfy the normal–mode equations:
$`H_\omega (𝐱)U_\lambda (𝐱)`$ $`+`$ $`{\displaystyle d𝐱^{}G(𝐱,𝐱^{})U_\lambda (𝐱^{})}=\mathrm{\Delta }_\lambda V_\lambda (𝐱),`$ (54)
$`H_\omega (𝐱)V_\lambda (𝐱)`$ $`=`$ $`\mathrm{\Delta }_\lambda U_\lambda (𝐱),`$ (55)
with the following normalization condition:
$$d𝐱\left(u_\lambda ^{}(𝐱)u_\lambda ^{}(𝐱)v_\lambda ^{}(𝐱)v_\lambda ^{}(𝐱)\right)=\delta _{\lambda ,\lambda ^{}}.$$
(56)
In these formulas $`\mathrm{\Delta }_\lambda `$ has a meaning of an eigenvalue, $`H_\omega (𝐱)=1/2(𝐩^2+\omega ^2𝐱^2)3\omega /2`$, and the integral kernel $`G(𝐱,𝐱^{})`$ is:
$$G(𝐱,𝐱^{})=2N\varphi _0(𝐱)V(𝐱𝐱^{})\varphi _0(𝐱^{}).$$
(57)
The above equations can be easily solved if we expand functions $`U_\lambda (𝐱)`$ and $`V_\lambda (𝐱)`$ in the basis of eigenfunctions $`\psi _𝐧(𝐱)=\psi _{n_xn_yn_z}(𝐱)`$ of the Hamiltonian $`H_\omega (𝐱)`$:
$`U_\lambda (𝐱)`$ $`=`$ $`{\displaystyle \underset{𝐧}{}}a_𝐧^\lambda \psi _𝐧(\sqrt{\omega }𝐱),`$ (58)
$`V_\lambda (𝐱)`$ $`=`$ $`{\displaystyle \underset{𝐧}{}}b_𝐧^\lambda \psi _𝐧(\sqrt{\omega }𝐱),`$ (59)
where components of the vector $`𝐧=(n_x,n_y,n_z)`$ are the standard quantum numbers of the harmonic oscillator eigenfunction of energy equal to $`n\omega `$, where $`n=n_x+n_y+n_z`$.
The coefficients $`a_𝐧^\lambda `$ and $`b_𝐧^\lambda `$ have to be determined from the Eqs. (54) and (55). Let us remind that the solution of the Eq. (45) for the order parameter $`\varphi _0(𝐱)`$ is the first function of the chosen basis set, $`\varphi _0(𝐱)=\psi _{000}(\sqrt{\omega }𝐱)`$. For this reason and also due to the oscillatory form of the inter-particle interactions, the integral kernel in Eqs. (54) and (55) couples only these basis functions $`\psi _𝐧`$ which correspond to three lowest eigenstates ($`n=0,1`$ and 2) of the Hamiltonian $`H_\omega (𝐱)`$. For larger $`n`$ quasi-particles excitation energies are those of the harmonic spectrum:
$$\mathrm{\Delta }_n=n\omega ,\mathrm{if}n>2.$$
(60)
These eigenvalues are degenerated. In general, there is a close link between the number of eigenmodes $`u_\lambda (𝐱)`$ and $`v_\lambda (𝐱)`$ corresponding to the eigenvalue $`n\omega `$ and the number of the oscillatory states of the same energy. Therefore, in order to classify independent solutions, it is convenient to use oscillatory quantum numbers instead the parameter $`\lambda `$ which simply enumerates quasi-particles eigenmodes. With this notational modification the solutions of the Eq.(54,55) corresponding to energies $`\mathrm{\Delta }_n`$ with $`n>2`$ are the following:
$`u_𝐧(𝐱)`$ $`=`$ $`\psi _𝐧(\sqrt{\omega }𝐱),`$ (61)
$`v_𝐧(𝐱)`$ $`=`$ $`0.`$ (62)
The low laying interacting states involve coupling of the bare oscillatory eigenfunctions. For the lowest excitation energy we get:
$$\mathrm{\Delta }_1=1.$$
(63)
There are three different modes of that energy corresponding to the excitation of one of the $`x`$, $`y`$, or $`z`$ degree of freedom. Below we present only one pair of the eigenmodes as the remaining two can be obtained by the permutation of the indices only. The $`x`$-direction eigenmodes are:
$`u_{100}(𝐱)`$ $`=`$ $`{\displaystyle \frac{(1+\omega )}{\sqrt{4\omega }}}\psi _{100}(\sqrt{\omega }𝐱),`$ (64)
$`v_{100}(𝐱)`$ $`=`$ $`{\displaystyle \frac{(1\omega )}{\sqrt{4\omega }}}\psi _{100}(\sqrt{\omega }𝐱).`$ (65)
Let us notice that energy of this mode of excitations is equal to the single excitation quantum of the trap mode.
The second excitation energy is equal to one of the excitation energy of the Hamiltonian $`H_\omega (𝐱)`$, namely:
$$\mathrm{\Delta }_2=2\omega .$$
(66)
There are six different eigenmodes of the above energy which is exactly the degeneracy of the second state of the 3D oscillator. The first three pairs of them are related to the double excitation along one of the axis of the coordinate system and are of the form:
$`u_{200}(𝐱)`$ $`=`$ $`\psi _{200}(\sqrt{\omega }𝐱)+{\displaystyle \frac{\omega ^21}{2\sqrt{2}\omega }}\psi _{000}(\sqrt{\omega }𝐱),`$ (67)
$`v_{200}(𝐱)`$ $`=`$ $`{\displaystyle \frac{\omega ^21}{2\sqrt{2}\omega }}\psi _{000}(\sqrt{\omega }𝐱),`$ (68)
(the other two pairs of eigenstates can obtained by the permutation of the oscillatory quantum numbers, as previously). The remaining three pairs of eigenmodes correspond to two single quanta of excitations along two different principal axis of the coordinate system. For example, one such pair is:
$`u_{110}(𝐱)`$ $`=`$ $`\psi _{110}(\sqrt{\omega }𝐱),`$ (69)
$`v_{110}(𝐱)`$ $`=`$ $`0,`$ (70)
and two others can be obtained by permutations of indices.
The shift in the ground state energy is given by
$$\underset{𝐧}{}\mathrm{\Delta }_ndxv_\lambda ^2(x)=\frac{(1\omega )^2}{4\omega }(2+2\omega +\omega ^2).$$
(71)
The excitation energies $`\mathrm{\Delta }_n`$ obtained within the Bogoliubov approach are the same as exact eigenenergies of the interacting system. The degeneracies of the eigenenergy state when we compare with Ref. are also the same. The Bogoliubov method is very well suited for the description of the excitation spectrum of the quantum degenerate gas. This result is somewhat surprising. One might rather expect that Bogoliubov approach works well only for the short range interaction. Our calculation shows that it works also in a rather exotic case when the interaction strength grows quadratically with the distance between particles. There are some differences between wave functions obtained in the Bogoliubov approximation and exact solutions of the $`N`$-particle Hamiltonian . They come from the fact that the exact solution cannot be written as a symmetrized product of any single particle functions.
### C Condensate fraction and fluctuations
In this subsection we consider Bose gas at finite temperatures. We will study the impact of interactions on the occupation of the condensate and its fluctuations using Bogoliubov method . We will compare the obtained results with those inferred from the exactly soluble model.
The statistical density matrix is $`\rho =Z^1\mathrm{exp}(\widehat{K}/T)`$, where $`\widehat{K}`$ is the Hamiltonian given in Eq.(44) and $`Z`$ is the statistical sum. This density matrix describes the excited subsystem only (quasi-particles). Number of quasi-particles is not conserved and the condensate is assumed to act as a reservoir of quasi-particles. In fact all this assumptions are in the spirit of the Maxwell’s demon ensemble introduced for the description of the ideal gas below the condensation temperature . Imposing the constraint on the total number of particles implies that occupation and fluctuations of the condensate can be directly related to the mean number and fluctuations of quasi-particles.
The mean number of excitations (quasi particles) above the condensed phase $`N_e=NN_0`$ is defined as:
$$N_e=d𝐱\widehat{\phi }^{}(𝐱)\widehat{\phi }(𝐱),$$
(72)
what leads to the following expression:
$$N_e=\underset{𝐧0}{}d𝐱\left\{\left[u_𝐧^2(𝐱)+v_𝐧^2(𝐱)\right]f_𝐧+v_𝐧^2(𝐱)\right\},$$
(73)
where $`f_𝐧=[\mathrm{exp}(\mathrm{\Delta }_𝐧/T)1]^1`$ plays the role of the mean quasi-particle occupation of the given energy state. The functions $`u_𝐧`$ and $`v_𝐧`$ are closely related to the wave functions of harmonic oscillator. The integration can be easily performed but the summation over all eigenstates might be quite difficult because of huge degeneracy of the energy levels. Fortunately, in our case almost all functions $`v_𝐧`$ vanish and their contribution to the final result is negligible at finite temperatures. Therefore the problem can be easily reduced to the calculation of the canonical occupation of the condensate trapped in the harmonic trap of frequency $`\omega `$. Again, similarly as in the exact solution we will treat separately the two regimes: (i) zero temperature limit where only quantum effects described by the last term of the Eq. (73) affect the condensate population, (ii) finite temperature case where the above mentioned effect is negligible. The occupation of the interacting condensate at zero temperature is:
$$N_0=N\frac{3}{4\omega }(1\omega )^2\frac{3}{8\omega ^2}(1\omega ^2)^2.$$
(74)
The first term in Eq. (74) corresponds to the weak interaction limit of the exact result, which is
$$N_0N\frac{3}{4\omega }(1\omega )^2.$$
(75)
The second term in Eq. (74) is new. This additional term overestimates the depletion of the condensate as compared to the rigorous treatment. Therefore the Bogoliubov treatment does not describe quantitatively an occupation of the zero temperature condensate.
On the other hand, in the thermodynamic limit, we recover the familiar expression:
$$N_0=N\left(\frac{T}{\omega }\right)^3\zeta (3).$$
(76)
This is exactly the same result which we obtained in the exact treatment. Similarly, the critical temperature which can be defined by setting $`N_0`$ to zero in the Eq. (76):
$$T_c=\omega \left(\frac{N}{\zeta (3)}\right)^{1/3},$$
(77)
is identical to the exact result, Eq. (33). It might be somewhat surprising but the Bogoliubov method works in our case pretty well up to the critical temperature, i.e., in the region where, in principle, its assumptions are not valid.
In a similar way we can obtain the fluctuations of the condensate population. Because fluctuations, contrary to the mean occupation, depend on the statistical ensemble, we use a notation which explicitly indicates the kind of performed averages (i.e., the canonical one). As we imposed constraint on the total number of particles, these fluctuations are equal to the fluctuations of the above-condensate part, $`\delta N_0^2_{\mathrm{CN}}=N_0^2_{\mathrm{CN}}N_0_{\mathrm{CN}}^2`$:
$`\delta N_0^2_{\mathrm{CN}}`$ $`=`$ $`{\displaystyle \underset{𝐧0}{}}\{{\displaystyle }\mathrm{d}𝐱[u_𝐧^2(𝐱)+v_𝐧^2(𝐱)]^2(f_𝐧^2+f_𝐧)`$ (79)
$`+4\left[{\displaystyle }\mathrm{d}𝐱u_𝐧(𝐱)v_𝐧(𝐱)\right]^2(f_𝐧^2+f_𝐧+1)\},`$
where the last term is responsible for the quantum fluctuations which do not vanish at zero temperature. Again the problem of calculating the condensate fluctuation can be reduced to finding the fluctuations of the ideal trapped Bose gas. In the case of zero temperature we have:
$$\delta N_0^2_{\mathrm{CN}}=\frac{3}{2\omega ^2}(1\omega ^2)^2+\frac{3}{(4\omega ^2)^2}\left(1\omega ^2\right)^4.$$
(80)
This formula overestimates (quite significantly) the condensate fluctuations due to the quantum effects as compared to the exact results in the limit of weak interaction, Eq. (31).
The fluctuations (in the thermodynamic limit and after neglecting the small quantum fluctuations obtained above) at finite temperatures are:
$$\delta N_0^2_{\mathrm{CN}}=\left(\frac{T}{\omega }\right)^3\zeta (2).$$
(81)
This result gives the correct value of the thermal fluctuations of the condensate. The Bogoliubov method works very well in predicting the thermal fluctuations of the condensate while it fails to reproduce the correct value of the quantum fluctuation. It might be surprising as in fact the approach should work in the low temperature region. In our opinion, it is the substitution of the operator annihilating the lowest energy state by a $`c`$-number that is responsible for inaccurate treatment of some quantum effects, particularly important at zero temperature.
## IV Conclusions
In our paper we used the exactly soluble many-particle model to illustrate the rigorous procedure of defining the condensate phase at zero temperature. By diagonalizing one-particle reduced density matrix we were able to study in details the role of interactions on the condensate at zero temperature. If the interaction strength becomes large $`|\kappa |>1`$ condensate disappears even when the system is in its ground state. This total depletion of the condensate has, in the thermodynamic limit, a character of critical phenomena. The destruction of the condensate signifies a breakdown of the mean-field theory. Due to strong quantum correlations the system cannot be viewed as being composed of independent quasi-particles moving in some effective potential resulting from interactions with a rest of the system. Instead the quasi-particles become strongly correlated and simple single-particle picture is not longer valid.
We have also carefully compared the exact quantum solutions of the oscillatory model with the approximate solutions obtained with the help of the Gross-Pitaevskii equation and Bogoliubov approximation. We have found that many of the characteristics of the exact solutions like the excitation spectrum, occupation of the condensate, and its thermal fluctuations are indeed reproduced with the help of the approximate methods. The Bogoliubov approach fails in the case of very strong interactions when the condensate is almost destroyed. This is however consistent with the basic assumption of the Bogoliubov method which explicitly assumes small condensate depletion and relays on the validity of a mean-field description. Surprisingly, in the studied model, the method works quite well even at temperatures close to the critical one. On the other hand the zero temperature (quantum) depletion and fluctuations of the condensate are not given correctly by the Bogoliubov method. Substitution of the destruction operator of the particle in the lowest energy state by a classical field results in inaccurate description of some quantum effects. Fortunately for the weakly interacting system, these effects are small and can be neglected.
###### Acknowledgements.
This work was supported by the KBN grant 2 P03B 130 15.
|
warning/0004/hep-ph0004099.html
|
ar5iv
|
text
|
# Unity of Fundamental Interactions
## I Introduction
The quantum mechanics of extended objects and its infinite dimensional generalization, namely, the quantum field theory of extended objects, in particular $`\varphi ^6`$ scalar field theory , quantum electrodynamics with the Pauli term, and quantum gravitodynamics have been presented by the author. In quantum gravitodynamics, the author develops an approach to understanding the response of of a lepton to a weak (linearized) gravitational field by making use of the vector representation of the linearized gravitational field . In the covariant perturbation theory approach, the quantum theory of gravity is rendered finite by making use of a Euclidean, retarded, graviton propagator given by:
$$\left(\delta \mu \rho \delta \nu \sigma +\delta \mu \sigma \delta \nu \rho \delta \mu \nu \delta \rho \sigma \right)\frac{e^{k^2/m^2}}{k^2}$$
(1)
where $`\frac{1}{m}`$ is the graviton Compton wavelength given by $`6.7\times 10^4R`$ where $`R=c/H`$ is the “Hubble radius” of the universe and $`H`$ is the Hubble constant. The graviton propagator is defined in the linear approximation since the notion of mass and spin of a field requires the presence of a flat background metric $`\eta _{\mu \nu }`$ which one does not have in the full theory. The full theory of general relativity may then be viewed as that of a graviton field which undergoes a nonlinear self-interaction. The propagator in Eq. (1) will render such a full theory finite to all orders. It is the discovery of this propagator which motivates us to study the possibility of unifying the graviton field with the existing electroweak theory. It is known that linearized gravity predicts that the motion of masses produces magnetic gravitational effects very similar to electromagnetism . The effective interaction between the electron and the graviton field can be understood in the vector representation where we make use of a propagator with the functional dependence given in Eq. (1) but with suitable vector indices. The author has calculated the order $`\alpha `$ correction to the magnetic gravitational moment by using such a propagator . Therefore, we are motivated to propose a gauge group $`SU(3)\times SU(2)\times U(1)\times U(1)`$ for the unified field theory which incorporates the strong force, the weak and electromagnetic interaction, and the graviton field. In this paper, we focus on the $`SU(2)\times U(1)\times U(1)`$ sector of the gauge theory. The feasibility of such a gauge structure and its implications for the existence of massive vector bosons, the $`Y^\pm `$ and the $`X^0`$, and the determination of their mass spectrum are studied in this paper. We also predict a modification to the fine structure constant under unified field conditions. Furthermore, the consequences of the extended object formulation for the gauge hierarchy problem are examined.
## II $`SU(2)\times U(1)\times U(1)`$
Let us consider the electronic-type lepton fields which consist of only the left- and right-handed parts of the electron field e:
$$e_L=\frac{1}{2}(1+\gamma _5)e,e_R=\frac{1}{2}(1\gamma _5)e$$
(2)
and a purely left-handed electron-neutrino field $`\nu _{eL}`$:
$$\gamma _5\nu _{eL}=\nu _{eL}.$$
(3)
In any representation of the gauge group, the fields must all have the same Lorentz transformation properties, so the representations of the gauge group must divide into a left-handed doublet $`(\nu _{eL},e_L)`$ and a right handed singlet $`e_R`$. Thus, the largest possible gauge group is then
$$SU(2)\times U(1)\times U(1)$$
(4)
under which the fields transform as
$$\delta \left(\begin{array}{c}\nu _e\\ e\end{array}\right)=i\left[\stackrel{}{ϵ}\stackrel{}{t}+ϵ_Lt_L+ϵ_Rt_R\right]\left(\begin{array}{c}\nu _e\\ e\end{array}\right)$$
(5)
where the generators are
$`\stackrel{}{t}`$ $`=`$ $`{\displaystyle \frac{g}{4}}(1+\gamma _5)\{\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right),\left(\begin{array}{cc}0& i\\ i& 0\end{array}\right),\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)\},`$ (6)
$`t_L`$ $``$ $`(1+\gamma _5)\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right),`$ (7)
$`t_R`$ $``$ $`(1\gamma _5)`$ (8)
with $`g`$ an unspecified constant. It will be convenient instead of $`t_L`$ and $`t_R`$ to consider the generators
$$y=g^{}\left[\frac{(1+\gamma _5)}{4}\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)+\frac{(1\gamma _5)}{2}\right]$$
(10)
and
$$n_e=g^{\prime \prime }\left[\frac{(1+\gamma _5)}{2}\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)+\frac{(1\gamma _5)}{2}\right],$$
(11)
where $`g^{}`$ and $`g^{\prime \prime }`$ are unspecified constants like $`g`$. The generator $`y`$ (the hypercharge) appears along with $`t_3`$ (the isospin operator) in a linear combination to define the charge $`q`$ of the pair $`(\nu _{eL},e_L)`$:
$$q=e\left(\begin{array}{cc}0& 0\\ 0& 1\end{array}\right)=\frac{e}{g}t_3\frac{e}{g^{}}y.$$
(12)
Also, $`n_e`$ is the electron-type lepton number and it defines the mass of the left-handed pair $`(\nu _{eL},e_L)`$ and the right-handed singlet $`e_R`$:
$$m_{ab}=m\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)=\frac{m}{g^{\prime \prime }}n_e$$
(13)
where $`m`$ is the electron mass. Thus, the charge couples to the electromagnetic field and the mass (in geometrized units) couples to the weak gravitational field. We want to include charge changing weak interactions (like beta decay), electromagnetism, and the graviton field in our theory, so we will assume there are gauge fields $`\stackrel{}{A}^\mu `$, $`B^\mu `$, and $`C^\mu `$ coupled to $`\stackrel{}{t}`$, $`y`$, and $`n_e`$ respectively. Before we include the graviton field in our theory we must ensure that it satisfies the stringent limits on long range forces that would be produced by a massless gauge field coupled to $`n_e`$ . Since the gravitational interaction is much weaker than the weak or electromagnetic interactions we are free to include a gauge field $`C^\mu `$ with strength $`g^{\prime \prime }`$ coupled to $`n_e`$. The gauge group is then
$$G=SU(2)_L\times U(1)\times U(1)$$
(14)
where the generators $`\stackrel{}{t}`$, $`y`$, and $`n_e`$ are given by Eq. (6), Eq. (10), and Eq. (11) respectively. The most general gauge- invariant and renormalizable Lagrangian that involves gauge-fields and electronic leptons is
$`_{YM}+_{LG}+_e={\displaystyle \frac{1}{4}}\left(_\mu \stackrel{}{A_\nu }_\nu \stackrel{}{A_\mu }+g\stackrel{}{A_\mu }\times \stackrel{}{A_\nu }\right)^2`$ (15)
$`{\displaystyle \frac{1}{4}}\left(_\mu \stackrel{}{B_\nu }_\nu \stackrel{}{B_\mu }\right)^2{\displaystyle \frac{1}{4}}\left(_\mu \stackrel{}{C_\nu }_\nu \stackrel{}{C_\mu }\right)^2`$ (16)
$`\overline{l}(/i\stackrel{}{/A}\stackrel{}{t}i/Byi/Cn_e)l.`$ (17)
The coupling constants $`g`$ and $`g^{}`$ are to be adjusted so that the gauge fields $`\stackrel{}{A}^\mu `$, $`B^\mu `$, and $`C^\mu `$ coupled to these generators are canonically normalized. Now, of these five gauge fields coupled to $`\stackrel{}{t}`$, $`y`$, and $`n_e`$, only two linear combinations, the electromagnetic field $`A_\mu `$, and the graviton field (vector representation) $`A_\mu ^G`$ are actually massless. We therefore must assume that $`SU(2)_L\times U(1)\times U(1)`$ is spontaneously broken into $`U(1)_{em}\times U(1)_{gravity}`$ with generators given by the hypercharge $`y`$ and the electron-type lepton number $`n_e`$. The details of the symmetry-breaking mechanism will be considered a little later. However, whatever this mechanism may be, we know that the canonically normalized vector fields corresponding to particles of spin one and definite mass consist of one field of charge $`+e`$ with mass $`m_Y`$
$$Y^\mu =\frac{1}{\sqrt{2}}\left(A_1^\mu +iA_2^\mu \right)$$
(18)
and another of charge $`e`$ and the same mass
$$Y^\mu ^{}=\frac{1}{\sqrt{2}}\left(A_1^\mu iA_2^\mu \right)$$
(19)
and three electrically neutral fields of mass $`m_X`$, zero, and zero respectively given by orthonormal linear combinations of $`A_3^\mu `$, $`B^\mu `$, and $`C^\mu `$:
$`X^\mu `$ $`=`$ $`\mathrm{cos}\varphi A_3^\mu +\mathrm{sin}\varphi B^\mu `$ (20)
$`A^\mu `$ $`=`$ $`\mathrm{cos}\theta \mathrm{sin}\varphi A_3^\mu +\mathrm{cos}\theta \mathrm{cos}\varphi B^\mu +\mathrm{sin}\theta C^\mu `$ (21)
$`A_G^\mu `$ $`=`$ $`\mathrm{sin}\theta \mathrm{sin}\varphi A_3^\mu \mathrm{sin}\theta \mathrm{cos}\varphi B^\mu +\mathrm{cos}\theta C^\mu `$ (22)
where $`\varphi `$ is the electroweak mixing angle (the Weinberg angle) and $`\theta `$ is the gravitoweak mixing angle. These linear combinations employ the Euler angles for a transformation from space axes to body coordinates with the third rotation set to zero. In this theory, the third rotation is set to zero because both the electromagnetic and graviton fields are massless $`U(1)`$ gauge fields and two $`U(1)`$’s are independent of each other. Hence, the mixing angle between the electromagnetic and graviton fields (the electrogravity angle) is zero. The electromagnetic field mixes with the weak interaction via the Weinberg angle and the weak interaction in turn mixes with the graviton field via the gravitoweak mixing angle. By making use of the inverse transformation back to space axes we have:
$`A_3^\mu `$ $`=`$ $`\mathrm{cos}\varphi X^\mu \mathrm{cos}\theta \mathrm{sin}\varphi A^\mu +\mathrm{sin}\theta \mathrm{sin}\varphi A_G^\mu `$ (23)
$`B^\mu `$ $`=`$ $`\mathrm{sin}\varphi X^\mu +\mathrm{cos}\theta \mathrm{cos}\varphi A^\mu \mathrm{sin}\theta \mathrm{cos}\varphi A_G^\mu `$ (24)
$`C^\mu `$ $`=`$ $`\mathrm{sin}\theta A^\mu +\mathrm{cos}\theta A_G^\mu `$ (25)
In the limit as the gravitoweak angle $`\theta `$ goes to zero, we recover the linear combinations necessary to generate the electroweak mass spectrum . In the above linear combinations we observe that the massive fields $`Y_\pm ^\mu `$ and $`X^\mu `$ are specified entirely in terms of the gauge fields $`\stackrel{}{A}^\mu `$ and $`B^\mu `$. Since the electrogravity mixing angle is zero, spontaneous symmetry breaking, which generates the vector meson term, occurs only in the electroweak sector of the theory. However, the coupling constants $`g`$ and $`g^{}`$ of the electroweak sector are specified in terms of the coupling constant $`g^{\prime \prime }`$ of the gravity sector as shown below. Thus, the spontaneous symmetry breaking of $`SU(2)_L\times U(1)\times U(1)`$ into $`U(1)\times U(1)`$ will generate two massless particles, namely, the photon and the graviton. Now, the generators of the unbroken symmetries, which are here electromagnetic and gravitodynamic gauge invariance are given by a linear combination of generators in which the coefficients are the same as the coefficients of the canonically normalized gauge fields coupled to these generators . Inspecting Eqs.(23) shows that
$`q`$ $`=`$ $`\mathrm{cos}\theta \mathrm{sin}\varphi t_3+\mathrm{cos}\theta \mathrm{cos}\varphi y,`$ (26)
$`m_{ab}`$ $`=`$ $`\mathrm{cos}\theta n_e.`$ (27)
Comparing this with Eq.(12) and Eq.(13) gives then
$$g=\frac{e}{\mathrm{cos}\theta \mathrm{sin}\varphi },g^{}=\frac{e}{\mathrm{cos}\theta \mathrm{cos}\varphi },g^{\prime \prime }=\frac{m}{\mathrm{cos}\theta }.$$
(28)
To complete the theory, we must now make some assumption about the mechanism of symmetry breaking. This mechanism must give masses not only to the $`Y^\pm `$ and $`X^0`$, but to the electron as well. Thus, we assume a ‘Yukawa’ coupling
$$_\varphi =G_e\overline{\left(\begin{array}{c}\nu _e\\ e\end{array}\right)}=\left(\begin{array}{c}\varphi ^+\\ \varphi ^0\end{array}\right)e_R+H.c.,$$
(29)
where $`(\varphi ^+,\varphi ^0)`$ is a doublet on which the $`SU(2)_L\times U(1)`$ generators are represented by the matrices:
$`\stackrel{}{t}^{(\varphi )}`$ $`=`$ $`{\displaystyle \frac{g}{2}}\{\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right),\left(\begin{array}{cc}0& i\\ i& 0\end{array}\right),\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)\},`$ (30)
$`y^{(\varphi )}`$ $`=`$ $`{\displaystyle \frac{g^{}}{2}}\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)`$ (31)
so that the charge matrix is
$$q=e\left(\begin{array}{cc}1& 0\\ 0& 0\end{array}\right)=\frac{e}{g}t_3^{(\varphi )}\frac{e}{g^{}}y^{(\varphi )}.$$
(32)
The most general form of the gauge-invariant term involving scalar and gauge fields consistent with the $`SU(2)_L\times U(1)`$ sector of the theory is:
$$_\varphi =\frac{1}{2}\left|\left(_\mu i\stackrel{}{A}\stackrel{}{t}^{(\varphi )}iB_\mu y^{(\varphi )}\right)\varphi \right|^2\frac{\mu ^2}{2}\varphi ^{}\varphi \frac{\lambda }{4}(\varphi ^{}\varphi )^2$$
(33)
where $`\lambda >0`$ and
$$\varphi =\left(\begin{array}{c}\varphi ^+\\ \varphi ^0\end{array}\right).$$
(34)
For $`\mu ^2<0`$, there is a tree-approximation vacuum expectation value at the stationary point of the Lagrangian
$$\varphi \varphi ^{}=v^2=|\mu ^2|/\lambda $$
(35)
In unitarity gauge the vacuum expectation values of the components of $`\varphi `$ are
$$\varphi ^+=0,\varphi ^0=v>0.$$
(36)
The scalar Lagrangian Eq. (33) then yields a vector meson mass term of the form
$$\frac{v^2g^2}{4}Y_\mu ^{}Y_\mu \frac{v^2}{8}\left(g^2+g^2\right)X_\mu X^\mu ,$$
(37)
where
$`{\displaystyle \frac{g}{g^{\prime \prime }}}`$ $`=`$ $`{\displaystyle \frac{e/m}{\mathrm{sin}\varphi }},`$ (38)
$`{\displaystyle \frac{g^{}}{g^{\prime \prime }}}`$ $`=`$ $`{\displaystyle \frac{e/m}{\mathrm{cos}\varphi }},`$ (39)
$`g^{\prime \prime }`$ $`=`$ $`{\displaystyle \frac{m}{\mathrm{cos}\theta }}.`$ (40)
Here, $`\varphi `$ is the Weinberg angle and $`\theta `$ is the gravitoweak mixing angle. We see that the photon mass is zero corresponding to an unbroken gauge symmetry $`U(1)_{em}`$ and the graviton mass is also zero corresponding to another unbroken gauge symmetry $`U(1)_{gravity}`$ while the $`Y^\pm `$ and $`X^0`$ have the masses
$$m_Y=\frac{v|g|}{2},m_X=\frac{\sqrt{g^2+g^2}}{2}.$$
(41)
Now, consider the relation
$$g^2/m_Y^2=4\sqrt{2}G_F$$
(42)
where $`G_F=1.16639(2)\times 10^5\text{GeV}^2`$ is the Fermi constant. This relation is obtained by comparing the effective interaction between low energy, $`e`$-type and $`\mu `$-type leptons with the effective ’V-A’ theory which is known to give a good description of muon decay. This allows an immediate determination of the vacuum expectation value as
$$v=\frac{2m_Y}{g}=247\text{GeV}.$$
(43)
By making use of the known value of the electroweak mixing angle given by $`\mathrm{sin}^2\varphi =0.23`$, we can determine the masses of $`Y^\pm `$ and $`X^0`$ in terms of the gravitoweak mixing angle as:
$`m_Y`$ $`=`$ $`{\displaystyle \frac{e_\mu v}{2|\mathrm{cos}\theta ||\mathrm{sin}\varphi |}}={\displaystyle \frac{80.2\text{GeV}}{|\mathrm{cos}\theta |}}`$ (44)
$`m_X`$ $`=`$ $`{\displaystyle \frac{e_\mu v}{2|\mathrm{cos}\theta ||\mathrm{sin}2\varphi |}}={\displaystyle \frac{91.3\text{GeV}}{|\mathrm{cos}\theta |}}`$ (45)
where $`e_\mu `$ is the electric charge defined at a sliding scale $`\mu `$ comparable to the energies of interest. We observe that as $`\theta 0`$ we regain the $`W`$ and $`Z`$ boson masses which is a result we expect. Thus, a unified field theory predicts the existence of massive vector bosons $`Y^\pm `$ and $`X^0`$ with the mass spectrum given in Eqs.(44)-(45). If we express the covariant derivative in Eq. (33) in terms of the mass eigenstate fields $`Y_\mu ^\pm `$, $`X^\mu `$, and $`A_\mu `$ we find that the coefficient of the electromagnetic interaction is not the electron charge $`e`$, but rather the effective electron charge $`e^{}`$
$$e^{}=\frac{e}{|\mathrm{cos}\theta |}=\frac{gg^{}}{\sqrt{g^2+g^2}}.$$
(46)
We observe that $`e^{}e`$ with equality being achieved when the gravitoweak mixing angle $`\theta `$ is zero. The mixing between the weak interaction and the graviton field causes an increase in the electromagnetic coupling strength. This is because the electromagnetic coupling is a function of the gauge couplings $`g`$ and $`g^{}`$ which have a dependence on $`\theta `$. If $`\alpha _G`$ is the fine structure constant of an electron in the unified field, then we have:
$$\frac{\alpha _G}{\alpha }=\frac{1}{\mathrm{cos}^2\theta }$$
(47)
implying that the fine structure constant suffers a modification. This would mean that if we were to measure the Lamb shift under unification conditions, the correction to the $`g`$-factor of the electron would be
$$a_e=\frac{\alpha _G}{2\pi }=\frac{0.0011597}{\mathrm{cos}^2\theta }.$$
(48)
## III The Gauge Hierarchy Problem
We begin with the reasonable observation that if $`SU(2)\times U(1)`$ is broken by the vacuum expectation value of an elementary scalar field, then that scalar field should be part of the grand unification. In order to produce a vacuum expectation value of the right size to give the observed $`W`$ and $`Z`$ boson masses, the Higgs scalar field must obtain a negative mass term of the size
$$\mu ^2(100GeV)^2.$$
(49)
Now, the mass term can be expressed in terms of the vacuum expectation value $`v`$ as
$$|\mu ^2|=\lambda v^2$$
(50)
where $`\lambda `$ is the renormalizable coupling in $`(\varphi ^{}\varphi )^2`$ charged scalar field theory. Therefore, the $`(mass)^2`$ receives additive renormalizations. In a theory with a cutoff scale $`\mathrm{\Lambda }`$, $`\mu ^2`$ can be much smaller than $`\mathrm{\Lambda }^2`$ only if the bare mass of the scalar field is of the order $`\mathrm{\Lambda }^2`$ and this value is canceled down to $`\mu ^2`$ by radiative corrections. If our theory of nature contains very large scales of grand unification, then the appropriate value for $`\mathrm{\Lambda }`$ is $`10^{16}`$ GeV or larger and it would require bizarre cancellations in the renormalized value of $`\mu ^2`$. Thus, the Higgs boson mass is very small compared to the grand unification scale. It is a mystery as to why the $`(mass)^2`$ of the Higgs boson has a value $`28`$ orders of magnitude or more below its natural value and this question is referred to as the gauge hierarchy problem. However, at grand unification energy scales the contributions of hitherto nonrenormalizable terms such as the Pauli term become significant . The description of quantum electrodynamics with the Pauli term necessitates the introduction of the quantum field theory of extended objects in which the finite extent of a particle defined via its Compton wavelength is incorporated into the field structure and leads to a finite interaction. Since hitherto nonrenormalizable terms become important at grand unification scales, it would be more correct if we consider $`SU(2)\times U(1)`$ to be broken by the vacuum expectation value of a hitherto nonrenormalizable $`(\varphi ^{}\varphi )^3`$ scalar field which can be rendered finite in the extended object formulation . The coupling $`\lambda `$ now becomes a finite coupling and the $`(mass)^2`$ does not receive additive renormalizations. Consider the potential
$$V(\varphi )=\mu ^2(\varphi ^{}\varphi )+\lambda (\varphi ^{}\varphi )^3$$
(51)
which has a tree-approximation vacuum expectation value at
$$\varphi \varphi ^{}=\left(|\mu ^2|/\lambda \right)^{\frac{1}{2}}$$
(52)
implying that
$$|\mu ^2|=\lambda v^4$$
(53)
where $`\lambda `$ is now a finite coupling. Therefore, we can now expect the Higgs boson mass to be of the order of $`100`$ GeV without any conceptual difficulty.
## IV Conclusion
The general gauge group $`SU(3)\times SU(2)\times U(1)\times U(1)`$ appears to describe the four known interactions in a consistent fashion. We are able to predict the existence of gauge bosons $`Y^\pm `$ and $`X^0`$ for the $`SU(2)\times U(1)\times U(1)`$ sector of this unified theory and determine mass spectrum of the gauge bosons. We have also shown that the fine structure constant is modified under unified field conditions. In addition, a possible resolution of the gauge hierarchy problem has been discussed. The results of this paper need to be subjected to experimental tests.
|
warning/0004/hep-ph0004024.html
|
ar5iv
|
text
|
# Inflation After Preheating
## 1 Introduction
The theory of cosmological phase transitions is usually associated with symmetry restoration due to high temperature effects and the subsequent symmetry breaking which occurs as the temperature decreases in an expanding universe . A particularly important version of this theory is the theory of first order cosmological phase transitions developed in . It served as a basis for the first versions of inflationary cosmology , as well as for the theory of electroweak baryogenesis .
Recently it was pointed out that preheating after inflation may rapidly produce a large number of particles that for a long time remain in a state out of thermal equilibrium. These particles may lead to specific nonthermal cosmological phase transitions . In some cases these phase transitions are first order ; they occur by the formation of bubbles of the phase with spontaneously broken symmetry inside the metastable symmetric phase. If the lifetime of the metastable state is large enough for the energy density of fluctuations to be diluted, one may encounter a short secondary stage of inflation after preheating . Such a secondary inflation stage, if it occurs late enough, could be important in solving the moduli and gravitino problems. In this respect secondary “nonthermal” inflation due to preheating may be an alternative to the “thermal inflation” , suggested for solving these problems.
In this paper we will briefly present the theory of such phase transitions and then give the results of numerical lattice simulations that directly demonstrated the possibility of such brief inflation. We will also discuss possible implications of our results for the theory of formation of topological defects during nonthermal phase transitions. A detailed description of numerical methods used in our work will be given in the appendix.
## 2 Theory of the phase transition
Consider a set of scalar fields with the potential
$$V(\varphi ,\chi )=\frac{\lambda }{4}(\varphi ^2v^2)^2+\frac{g^2}{2}\varphi ^2\chi ^2.$$
(1)
The inflaton field $`\varphi `$ has a double-well potential and interacts with an $`N`$-component scalar field $`\chi `$; $`\chi ^2_{i=1}^N\chi _i^2`$. For simplicity, the field $`\chi `$ is taken to be massless and without self-interaction. The fields couple minimally to gravity in a FRW universe with a scale factor $`a(t)`$.
The potential $`V(\varphi ,\chi )`$ has minima at $`\varphi =\pm v`$, $`\chi =0`$ and a local maximum in the $`\varphi `$ direction at $`\varphi =\chi =0`$ with curvature $`V_{,\varphi \varphi }=\lambda v^2`$. The effective potential acquires corrections due to quantum and/or thermal fluctuations of the scalar fields ,
$$\mathrm{\Delta }V=\frac{3}{2}\lambda \varphi ^2\varphi ^2+\frac{g^2}{2}\chi ^2\varphi ^2+\frac{g^2}{2}\varphi ^2\chi ^2+\mathrm{},$$
(2)
where we have written only the leading terms depending on $`\varphi `$ and $`\chi `$. The effective mass squared of the field $`\varphi `$ is given by
$$m_\varphi ^2=m^2+3\lambda \varphi ^2+3\lambda \varphi ^2+g^2\chi ^2,$$
(3)
where $`m^2=\lambda v^2`$. Symmetry is restored, i.e. $`\varphi =0`$ becomes a stable equilibrium point, when the fluctuations $`\varphi ^2,\chi ^2`$ become sufficiently large to make the effective mass squared positive at $`\varphi =0`$.
For example, one may consider matter in thermal equilibrium. Then, in the large temperature limit, one has $`\varphi ^2=\chi _i^2=\frac{T^2}{12}.`$ The effective mass squared of the field $`\varphi `$
$$m_{\varphi ,eff}^2=m^2+3\lambda \varphi ^2+3\lambda \delta \varphi ^2+g^2\chi ^2$$
(4)
is positive and symmetry is restored (i.e. $`\varphi =0`$ is the stable equilibrium point) for $`T>T_c`$, where $`T_c^2=\frac{12m^2}{3\lambda +Ng^2}m^2`$. At this temperature the energy density of the gas of ultrarelativistic particles is given by
$$\rho =𝒩(T_c)\frac{\pi ^2}{30}T_c^4=\frac{24m^4𝒩(T_c)\pi ^2}{5(3\lambda +Ng^2)^2}.$$
(5)
Here $`𝒩(T)`$ is the effective number of degrees of freedom at large temperature, which in realistic situations may vary from $`10^2`$ to $`10^3`$. We will assume that $`Ng^2\lambda `$, see below. For $`g^4<\frac{96𝒩(T_c)\pi ^2}{5N^2}\lambda `$ the thermal energy at the moment of the phase transition is greater than the vacuum energy density $`V(0)=\frac{m^4}{4\lambda }`$, which means that the phase transition does not involve a stage of inflation.
In fact, the phase transition with symmetry breaking occurs not at $`T>T_c`$, but somewhat earlier . To understand this effect let us compare the temperature $`T_cm/(\sqrt{N}g)`$ and the mass $`m_\chi =g\varphi `$ of the $`\chi `$ particles in the minimum of the zero-temperature effective potential at $`\varphi =v=m/\sqrt{\lambda }`$. One can easily see that $`m_\chi T_c`$ for $`Ng^4\lambda `$. This means that for $`Ng^4\lambda `$ the temperature $`T_c`$ is insufficient to excite perturbations of the fields $`\chi _i`$ at $`\varphi =v`$. As a result, these perturbations do not change the shape of the effective potential $`\varphi =v`$. Thus the potential at $`T`$ slightly above $`T_c`$ has its old zero-temperature minimum at $`\varphi =v`$, as well as the temperature-induced minimum at $`\varphi =0`$. Symmetry breaking occurs as a first-order phase transition due to formation of bubbles of the phase with $`\varphi v`$ at some temperature above $`T_c`$ when the minimum at $`\varphi =v`$ becomes deeper than the minimum at $`\varphi =0`$, and the probability of bubble formation becomes sufficiently large. A more detailed investigation in the case $`N=1`$ shows that the phase transition is first order under a weaker condition $`g^3\lambda `$ .
In the case $`Ng^4>10^2\lambda `$ the phase transition occurs after a secondary stage of inflation. In this regime radiative corrections are important. They lead to the creation of a local minimum of $`V(\varphi ,\chi )`$ at $`\varphi =0`$ even at zero temperature, and the phase transition occurs from a strongly supercooled state . That is why the first models of inflation required supercooling at the moment of the phase transition .
In supersymmetric theories one may have $`Ng^410^2\lambda `$ and still have a potential which is flat near the origin due to cancellation of quantum corrections of bosons and fermions . In such cases the thermal energy becomes smaller than the vacuum energy at $`T<T_0`$, where $`T_0^4=\frac{15}{2𝒩\pi ^2}m^2v^2`$. Then one may have a short stage of inflation which begins at $`TT_0`$ and ends at $`T=T_c`$. During this time the universe may inflate by the factor
$$\frac{a_c}{a_0}=\frac{T_0}{T_c}10^1(\frac{g^4}{\lambda })^{1/4}.$$
(6)
Similar phase transitions may occur much more efficiently prior to thermalization, due to the anomalously large fluctuations $`\varphi ^2`$ and $`\chi ^2`$ produced during preheating . These fluctuations can change the shape of the effective potential and lead to symmetry restoration. Afterwards, the universe expands, the values of $`\varphi ^2`$ and $`\chi ^2`$ drop down, and the phase transition with symmetry breaking occurs.
An interesting feature of nonthermal phase transitions is that they may occur even in theories where the usual thermal phase transitions do not happen. The main reason can be understood as follows. Suppose reheating occurs due to the decay of a scalar field with energy density $`\rho `$. If this energy is instantly thermalized, then one obtains relativistic particles with energy density $`O(T^4)`$ which in the first approximation can be represented as $`\rho E^2(\varphi ^2+\chi ^2)`$. Here $`ET\rho ^{1/4}`$ is a typical energy of a particle in thermal equilibrium. After preheating, however, one has particles $`\varphi `$ and $`\chi `$ with much smaller energy but large occupation numbers. As a result, the same energy release may create much greater values of $`\varphi ^2`$ and $`\chi ^2`$ than in the case of instant thermalization. This may lead to symmetry restoration after preheating even if the symmetry breaking occurs on the GUT scale, $`v10^{16}`$ GeV .
The main conclusions of have been confirmed by detailed investigation using lattice simulations in . One of the main results obtained in was that for sufficiently large $`g^2`$ nonthermal phase transitions are first order. They occur from a metastable vacuum at $`\varphi =0`$ due to the creation of bubbles with $`\varphi 0`$. This result is very similar to the analogous result in the theory of thermal phase transitions . According to , the necessary conditions for this transition to occur and to be of the first order can be formulated as follows:
(i) At the time of the phase transition, the point $`\varphi =0`$ should be a local minimum of the effective potential. From (3), we see that this means that $`Ng^2\chi _i^2>\lambda v^2`$.
(ii) At the same time, the typical momentum $`p_{}`$ of $`\chi _i`$ particles should be smaller than $`gv`$. This is the condition of the existence of a potential barrier. Particles with momenta $`p<gv`$ cannot penetrate the state with $`|\varphi |v`$, so they cannot change the shape of the effective potential at $`|\varphi |v`$. Therefore, if both conditions (i) and (ii) are satisfied, the effective potential has a local minimum at $`\varphi =0`$ and two degenerate minima at $`\varphi \pm v`$.
(iii) Before the minima at $`\varphi \pm v`$ become deeper than the minimum at $`\varphi =0`$, the inflaton’s zero mode should decay significantly, so that it performs small oscillations near $`\varphi =0`$. Then, after the minimum at $`|\varphi |v`$ becomes deeper than the minimum at $`\varphi =0`$, fluctuations of $`\varphi `$ drive the system over the potential barrier, creating an expanding bubble.
The investigation performed in confirmed that for sufficiently large $`g^2`$ and $`N`$ these conditions are indeed satisfied and the phase transition is first order. One may wonder whether for $`g^2\lambda `$ one may have a stage of inflation in the metastable vacuum $`\varphi =0`$.
Analytical estimates of Ref. suggested that this is indeed the case, and the degree of this inflation for $`N=1`$ is expected to be
$$\frac{a_c}{a_0}(\frac{g^2}{\lambda })^{1/4},$$
(7)
which is much greater than the number $`10^1(\frac{g^4}{\lambda })^{1/4}`$ in the thermal inflation scenario. One could also expect that the duration of inflation, just like the strength of the phase transition, increases if one considers N fields $`\chi _i`$ with $`N1`$.
However, the theory of preheating is extremely complicated, and there are some factors which could not be adequately taken into account in the simple estimates of . The most important factor is the effect of rescattering of particles produced during preheating . This effect tends to shut down the resonant production of particles and thus shorten or prevent entirely the occurrence of a secondary stage of inflation. Thus the estimates above reflect the maximum degree of inflation possible for a given set of parameter values, but in practice the expansion factor will be somewhat smaller than these predictions. The only way to fully account for all the effects of backreaction and expansion is through numerical lattice simulations. In our paper we used a generalized version of the method of lattice simulations developed in . In the next section we will describe the basic features of our method and describe our main results. A detailed description of the lattice simulations will be given in the appendix.
## 3 Simulation Results and Their Interpretation
In our paper we will take $`\lambda 10^{13}`$, which gives the proper magnitude of inflationary perturbations of density . We assume that $`g^2\lambda `$, and consider $`v10^{16}`$ GeV, which corresponds to the GUT scale. A numerical investigation of preheating in the model (1) was first performed in . The authors found a strongly first order phase transition. The strength of the phase transition increased with an increase of $`g^2/\lambda `$ and of the number $`N`$ of the fields $`\chi _i`$. However, for the parameters of the model studied in ($`g^2/\lambda 200`$) there was no inflation during symmetry restoration. This is not unexpected because the estimates discussed above indicated that the expansion of the universe during the short stage of nonthermal inflation cannot be greater than $`\left(\frac{g^2}{\lambda }\right)^{1/4}`$.
Keeping in mind that $`\lambda `$ in this model is extremely small, one would expect that in realistic versions of this model one may have $`g^2/\lambda `$ as large as $`10^{10}`$, which could lead to a relatively long stage of inflation. However, for very large $`g^2/\lambda `$ our analytical estimates are unreliable, and lattice simulations become extremely difficult: One needs to have enormously large lattices to keep both infrared and ultraviolet effects under control.
To mimic the effects of large $`g^2`$, we considered a large number of the fields $`\chi _i`$. We have performed simulations for $`g^2/\lambda =800`$ and $`N=19`$. With these parameters the strength of the phase transition became much greater, and there was a short stage of inflation prior to the phase transition. The details of our calculation and an explanation of our methods are given in the appendix. Here we only present the main results.
The simulation showed that the oscillations of the inflaton field decreased until the field was trapped near zero. It remained there until the moment of the phase transition when it rapidly jumped to its symmetry breaking value, as shown in Fig. 1.
The trapping of the field occurred because of the corrections to the effective potential induced by the particles $`\varphi `$ and $`\chi `$ produced during preheating, just like in the theory of high-temperature phase transitions. In our case, however, this effect has some unusual features.
To first order in $`g^2`$, the leading contribution to the equation of motion $`\ddot{\varphi }=V^{}`$ is given by $`g^2\varphi \chi ^2`$, where
$$\chi ^2\frac{N}{2\pi ^2}\underset{0}{\overset{\mathrm{}}{}}\frac{n_kk^2dk}{\omega _k(\varphi )}.$$
(8)
Here $`\omega _k=\sqrt{k^2+g^2(\varphi ^2+\varphi ^2)}`$ is the energy of $`\chi _i`$ particles with momentum $`k`$ and $`n_k`$ is their occupation number; $`\varphi `$ is the homogeneous component of the field. For $`\varphi \sqrt{\varphi ^2}`$, one has
$$\chi ^2_{_{\varphi =0}}\frac{N}{2\pi ^2}\underset{0}{\overset{\mathrm{}}{}}\frac{n_kk^2dk}{\sqrt{k^2+g^2\varphi ^2}}.$$
(9)
This quantity does not depend on $`\varphi `$; it can be evaluated using our lattice simulations when the field $`\varphi `$ oscillates near $`\varphi =0`$. It leads to the usual quadratic correction to the effective potential, see Eq. (2). This correction adequately describes the change of the shape of the effective potential for $`\varphi `$ smaller than the amplitude of the oscillations of this field, because most of the time prior to the moment of the phase transition this amplitude is much smaller than $`\sqrt{\varphi ^2}`$.
However, if we want to evaluate the effective potential at all values of $`|\varphi |`$ from $`0`$ to $`v`$, rather than for $`\varphi `$ similar to the amplitude of the oscillations, then one should take into account that for sufficiently large $`|\varphi |`$ the term $`g|\varphi |`$ becomes greater than $`g\sqrt{\varphi ^2}`$ and than the typical momentum $`k`$ of particles $`\chi _i`$. In this case the main contribution to $`\chi ^2`$ is given by nonrelativistic particles with $`\omega _k+g|\varphi |`$, and one has
$$\chi ^2\frac{N}{2\pi ^2}\underset{0}{\overset{\mathrm{}}{}}\frac{n_kk^2dk}{g|\varphi |}=\frac{n_\chi }{g|\varphi |},$$
(10)
where $`n_\chi `$ is the total density of all types of $`\chi _i`$ particles. This implies that at large $`|\varphi |`$ the effective potential acquires a correction
$$\delta Vg|\varphi |n_\chi .$$
(11)
Thus, instead of being quadratic or cubic in $`|\varphi |`$, as one could expect from the analogy with the high-temperature theory , the corrections to the effective potential at large $`|\varphi |`$ are proportional to $`|\varphi |`$ .
The combination of these two types of corrections to the effective potential (quadratic at small $`|\varphi |`$ and linear at large $`|\varphi |`$) leads to the symmetry restoration that we have found in our lattice simulations.
It is instructive to look in a more detailed way at the small region near the time of the phase transition. The first of the graphs in Fig. 2 shows the oscillations of the field $`\varphi `$ soon before the phase transition, whereas the second one shows these oscillations soon afterwards.
First of all, one can see that just before the phase transition the field oscillates with an amplitude three orders of magnitude smaller than $`v`$, which is a clear sign of symmetry restoration. Another interesting feature is that the frequency of oscillations does not vanish as we approach the phase transition, but remains nearly constant. Moreover, this frequency is only about two times smaller than the frequency of oscillations after the phase transition, which is equal to $`\sqrt{2}m`$. Note that the frequency of the oscillations is determined by the effective mass of the scalar field, which is given by the curvature of the effective potential: $`m_\varphi ^2=V^{\prime \prime }`$. This means that at the moment of the phase transition the effective potential has a deep minimum at $`\varphi =0`$ with curvature $`V^{\prime \prime }+m^2`$, i.e. the phase transition is strongly first order. Such phase transitions should occur due to the formation of bubbles containing nonvanishing field $`\varphi `$.
Indeed, we have found that this transition occurred in a nearly spherical region of the lattice that quickly grew to encompass the entire space. The growth of this region of the new phase is shown in Fig. 3. The nearly perfect sphericity of this region is an additional indication that the transition was strongly first-order. In comparison, the bubble observed in the lattice simulations of for $`g^2/\lambda 200`$ was not exactly spherically symmetric.
The first order phase transition and bubble formation seen in our simulations can be understood as a result of gradual accumulation of classical fluctuations $`\delta \varphi (t,\stackrel{}{x})`$. These fluctuations stochastically climb up from $`\varphi =0`$ towards the local maximum $`\varphi _{}`$ of the effective potential. Consider the regions in which $`\delta \varphi (t,\stackrel{}{x})>\varphi _{}`$. If the probability of formation of such regions is small because they correspond to high peaks of the random field $`\varphi `$, then these regions will have a nearly spherical shape and can be represented by spherical surfaces of radius $`R_{}`$ (bubbles). If the radius $`R_{}`$ is small, gradient terms will prevent the field $`\varphi `$ inside the region from rolling down towards the global minimum at $`\varphi =v`$ (subcritical bubble). If $`R_{}`$ is large enough, the gradient terms cannot push the field back to the metastable state $`\varphi =0`$ and the field inside the bubble rolls towards the global minimum, forming a bubble of ever increasing radius. This process can be described within the stochastic approach to tunneling proposed in . Typically, the gradient terms cannot win over the potential energy terms if $`R_{}>O(|m_\varphi ^1|)`$, where $`m_\varphi `$ corresponds to the effective mass of the scalar field in the interior of the bubble. This provides an estimate for the initial size of the bubble $`R_{}O(|m_\varphi ^1|)`$.
The phase transition occurs from a state with energy density dominated by the vacuum energy density $`V(0)`$. Figure 4 shows the scale factor $`a`$ as a function of time. The curvature becomes slightly positive at the time before the phase transition, which indicates a short stage of exponential growth of the universe. Because the curvature is hard to see in figure 4 we have also plotted the second derivative $`\ddot{a}`$ in figure 5. While the inflaton is trapped in the false vacuum state, $`\ddot{a}`$ becomes positive, indicating a brief stage of inflation.
Another signature of inflation is an equation of state with negative pressure. Figure 6 shows the parameter $`\alpha =p/\rho `$, which becomes negative during the metastable phase. At the moment of the phase transition the universe becomes matter dominated and the pressure jumps to nearly $`0`$.
From the beginning of this inflationary stage (roughly when the pressure becomes negative) to the moment of the phase transition the total expansion factor is 2.1. As expected this is of the same order but somewhat lower than the predicted maximum, $`\left(\frac{g^2}{\lambda }\right)^{1/4}5.3`$. We can thus conclude that it is possible to achieve inflation for parameters for which this would not have been possible in thermal equilibrium ($`g^4\lambda `$). In our simulation we have shown the occurrence of a very brief stage of inflation. This stage may be much longer for larger (realistic) values of $`g^2/\lambda `$. However, to check whether this is indeed the case one would need to perform a more detailed investigation on a lattice of a much greater size.
## 4 Nonthermal phase transitions and production of topological defects
In this section we would like to discuss possible implications of our investigation for the theory of production of topological defects after preheating .
The bubbles that appear after the phase transition can contain either positive or negative field, $`\varphi =\pm v`$. If bubbles of either type are formed with comparable probability, then after the phase transition the universe becomes divided into nearly equal numbers of domains with $`\varphi =\pm v`$, separated by domain walls. Such domain walls would lead to disastrous cosmological consequences, which would rule out the models where this may happen .
In general, the number of bubbles with $`\varphi =+v`$ may be much greater (or much smaller) than the number of bubbles with $`\varphi =v`$. Then the domain wall problem does not appear because the bubbles with $`\varphi =+v`$ would rapidly eat all their competitors with $`\varphi =v`$ (or vice versa). This may happen, for example, if the moment of the bubble production is determined by the coherently oscillating scalar field $`\varphi `$. In such a case, after oscillating a bit near the top of the effective potential, the field $`\varphi `$ may wind up in the same minimum of the effective potential everywhere in the universe.
To investigate the domain wall problem in our model one would need to repeat the calculation many times with slightly different initial conditions or to make them in a box of a much greater size that would allow one to see many bubbles simultaneously. Fortunately, the results obtained in our study may be sufficient to give an answer to this question without extremely large simulations.
First of all, according to the stochastic approach to the theory of tunneling with bubble formation , the bubbles of the field $`\varphi `$ are created as a result of the accumulation of long-wavelength fluctuations of the scalar field with momenta $`k`$ smaller than the typical mass scale $`m_\varphi `$ associated with this field, see Section 3. In our case this mass scale is related to the frequency of oscillations of the scalar field at the moment of the phase transition. At that moment the leading contribution to the fluctuations $`\varphi ^2`$ is given by fluctuations with momenta much smaller than $`m_\varphi `$. We calculated the value of the long-wavelength component of $`\sqrt{\varphi ^2}`$ and found that it is (approximately) of the same order as the amplitude of oscillations of the field $`\varphi `$ at the moment of the phase transition. The existence of a first-order phase transition suggests that the probability of bubble formation must have been exponentially suppressed during the metastable stage. Such suppression would only occur only if the amplitude of fluctuations required to form a bubble of the new phase was much larger than $`\sqrt{\varphi ^2}`$ , which would in turn mean the required amplitude was much greater than the amplitude of oscillations of $`\varphi `$. This suggests that the probability of the bubble formation is almost entirely determined by the incoherent fluctuations of the field $`\varphi `$ rather than by the small coherent oscillations of this field. Consequently, the probability of formation of bubbles containing $`\varphi =+v`$ in the first approximation must be equal to the probability of formation of bubbles containing $`\varphi =v`$.
To make this statement more reliable one would need to estimate the amplitude of the long-wavelength fluctuations of the field $`\varphi `$ in a more precise way, which would involve using lattices of a greater size. However, there is additional evidence suggesting that the number of bubbles with positive and negative $`\varphi `$ must be approximately equal to each other.
Indeed, as we have seen, the curvature of the effective potential remained approximately constant during dozens of oscillations of the field $`\varphi `$ prior to the moment of the phase transition. This suggests that the shape of the effective potential and, consequently, the probability of the tunneling, did not change much during a single oscillation. Therefore one may expect that, within a single oscillation, the probability of a bubble forming when the oscillating field $`\varphi `$ was negative was approximately the same as the probability of the bubble forming when it was positive.
If the number of bubbles with positive and negative $`\varphi `$ is approximately equal to each other, the phase transition leads to the formation of dangerous domain walls, which rules out our model . If correct, this is a rather important conclusion which shows that the investigation of nonthermal phase transition may rule out certain classes of inflationary models which otherwise would seem quite legitimate .
But this conclusion does not imply that all theories where the nonthermal phase transition is strongly first order are ruled out. For example, one may consider a model (1) with $`\varphi `$ being not a real but a complex field, $`\varphi =\frac{1}{\sqrt{2}}(\varphi _1+i\varphi _2)`$, $`|\varphi |^2=\frac{1}{2}(\varphi _1^2+\varphi _2^2)`$. Since the main contribution to the effective potential of the field $`\varphi `$ in the theory (1) is given not by the field(s) $`\varphi `$ but by the fields $`\chi _i`$, we expect that this generalization will not lead to a qualitative modification of our results. In particular, we expect that for sufficiently large $`N`$ and $`g^2/\lambda `$ the phase transition will be strongly first order and there will be a short stage of inflation after preheating. However, in the new model we will have strings instead of domain walls.
A similar model in the absence of interaction of the fields $`\varphi `$ with the fields $`\chi `$ was studied in . It was argued that even in this case infinite strings may be formed. The theory of galaxy formation due to cosmic strings is currently out of favor, but it is certainly true that cosmic strings produced after inflation may add new interesting features to the standard theory of formation of the large-scale structure of the universe
The possibility of strongly first order phase transitions induced by preheating in models with $`g^2\lambda `$ adds new evidence that infinite strings can be produced after nonthermal phase transitions. Indeed, infinite strings may not be produced if the direction in which the field $`\varphi `$ falls from the point $`\varphi =0`$ at the moment of the phase transition is determined by the oscillations of the field $`\varphi `$. If, just as in the case discussed above, the amplitude of these oscillations are much smaller than $`\sqrt{\varphi ^2}`$ at the moment of the phase transition, then infinite strings are indeed formed.
## 5 Conclusions
The results of our lattice simulation confirm our expectations that preheating may lead to nonthermal phase transitions even in those theories where spontaneous symmetry breaking occurs at the GUT scale, $`v10^{16}`$ GeV. Some time ago this question was intensely debated in the literature. Some authors claimed that nonthermal phase transitions induced by preheating are impossible, and the notion of the effective potential after preheating is useless. In our opinion, Figs. 1, 2 and 3 give a clear answer to this question. In particular, Fig. 1 shows that 90% of the time from the end of inflation to the moment of symmetry breaking the field $`\varphi `$ oscillates about $`\varphi =0`$ with an amplitude much smaller than $`v`$. This could happen only because the corrections to the effective potential induced by particles $`\varphi `$ and $`\chi `$ change the shape of $`V(\varphi )`$ near $`\varphi =0`$, turning its maximum into a deep local minimum.
In some theories, this effect may lead to production of superheavy strings, which may have important cosmological implications for the theory of formation of the large scale structure of the universe. In some other theories, these phase transitions may lead to excessive production of monopoles and domain walls. This may rule out a broad class of otherwise acceptable inflationary models.
In this paper we have shown that under certain conditions a nonthermal phase transition may lead to a short secondary stage of inflation. It would be interesting to study the possibility that a secondary stage of inflation induced by preheating could help solve the moduli and gravitino problems. The answer to this question will be strongly model-dependent because gravitinos can be produced by the oscillating scalar field even after the secondary inflation . Independently of all practical implications, the possibility of a secondary stage of inflation induced by preheating seems very interesting because it clearly demonstrates the potential importance of nonperturbative effects in post-inflationary cosmology.
The authors are grateful to Julian Borrill for very useful discussions. This work was supported by NSERC and CIAR and by NSF grant AST95-29-225. The work of G.F. and A.L. was also supported by NSF grant PHY-9870115, and the work of L.K. and A.L. by NATO Linkage Grant 975389.
## 6 Appendix: The Numerical Calculations
Our lattice program solves the classical equations of motion for a set of scalar fields in a Friedman-Robertson-Walker universe. These fields include an inflaton $`\varphi `$ coupled to a set of matter fields $`\chi _i`$. The scale factor $`a`$ is also solved for self-consistently. In this appendix we describe the exact equations being solved and the method used to solve them.
Unless otherwise specified, all variables names refer to their bare physical values measured in Planck units. The quantities as they appear in the program have been rescaled in several ways, and these values will be indicated with pr, as in $`\varphi _{pr}`$. The relations between the rescaled variables and their bare values are given below. Also, when we want to refer to a general property of the fields $`\varphi `$ and $`\chi _i`$ we will use $`\sigma `$ to indicate a generic scalar field.
The simulations we discuss here used a grid of $`128^3`$ points. The spacing of the points was chosen so as to ensure that the ultraviolet cutoff imposed by the grid was higher than all physically relevant momenta in the problem. To check this we monitored the power spectra $`|\varphi _k|^2`$ and $`|\chi _k|^2`$ throughout the run. On a log plot you can see that there is a momentum above which the slope of the power spectrum decreases sharply (i.e. becomes more negative), and we set our grid spacing to ensure that this cutoff was well below the cutoff of the grid. A sample spectrum for $`\chi `$, taken from the late stages of preheating, is shown in figure 7. The kink that can be seen at $`kk_{max}/2`$ persisted from preheating through the phase transition, and a similar feature can also be seen in the $`\varphi `$ spectrum.
The grid spacing we used was
$$dx=0.133\frac{1}{\sqrt{\lambda }\varphi _0}1.26\times 10^6M_P^1$$
(12)
measured in comoving coordinates with $`a=1`$ at the beginning of the simulation. The total box size, $`L=128dx`$, was equal to slightly more than $`10`$ Hubble radii at the end of inflation.
As a further check that the ultraviolet grid cutoff was not affecting the physics we did some smaller runs (i.e. with fewer fields) using these same parameters and another run using a grid of $`64^3`$ points with twice as large a grid spacing. Both the $`64^3`$ and $`128^3`$ runs had the same total box size and they showed essentially identical behavior for the fields, suggesting that the $`128^3`$ grid has more than enough modes in the ultraviolet.
### 6.1 The Field Equations
We work in a FRW universe with metric $`g_{\mu \nu }=diag\{1,a^2,a^2,a^2\}`$. The equations of motion for the scalar fields $`\varphi `$ and $`\chi _i`$ derive from the potential
$$V=\frac{\lambda }{4}\left(\varphi ^2v^2\right)^2+\frac{1}{2}g^2\varphi ^2\chi _i^2$$
(13)
where $`v`$, $`\lambda `$, and $`g`$ are constant parameters and $`\chi _i^2`$ is understood to include a summation over $`i`$. The results reported here used $`v=7\times 10^4`$, $`\lambda =9\times 10^{14}`$, and $`g^2=800\lambda `$, and took $`\chi `$ to be a 19-component field. This potential gives rise to the equations of motion
$$\ddot{\varphi }+3\frac{\dot{a}}{a}\dot{\varphi }\frac{1}{a^2}^2\varphi +\left(\lambda \left(\varphi ^2v^2\right)+g^2\chi _i^2\right)\varphi =0$$
(14)
$$\ddot{\chi _i}+3\frac{\dot{a}}{a}\dot{\chi _i}\frac{1}{a^2}^2\chi _i+g^2\varphi ^2\chi _i=0$$
(15)
These equations can be simplified by the following variable redefinitions
$$\varphi _{pr}=\frac{a}{\varphi _0}\varphi ;\chi _{i,pr}=\frac{a}{\varphi _0}\chi _i;v_{pr}=\frac{1}{\varphi _0}v;\stackrel{}{x}_{pr}=\sqrt{\lambda }\varphi _0\stackrel{}{x};\tau _{pr}=\sqrt{\lambda }\varphi _0\frac{dt}{a};ge\frac{g^2}{\lambda }$$
(16)
where $`\varphi _0`$ is the initial (bare) value of the field $`\varphi `$ and all other parameters come from the original equations of motion. These then give the rescaled equations of motion
$$\varphi _{pr}^{\prime \prime }_{pr}^2\varphi _{pr}+\left(\varphi _{pr}^2a^2v_{pr}^2+ge\chi _{i,pr}^2\frac{a^{\prime \prime }}{a}\right)\varphi _{pr}=0$$
(17)
$$\chi _{i,pr}^{\prime \prime }_{pr}^2\chi _{i,pr}+\left(ge\varphi _{pr}^2\frac{a^{\prime \prime }}{a}\right)\chi _{i,pr}=0$$
(18)
where derivatives are with respect to the rescaled time and distance variables defined above. Note that all first derivative terms have been eliminated, and the dependence on the coupling constants $`\lambda `$ and $`g`$ is now only through the ratio $`g^2/\lambda `$, denoted here by $`ge`$. The value of $`\lambda `$ itself appears only in setting the initial conditions (described below). The rescalings involving $`\varphi _0`$ do not affect the equations of motion.
### 6.2 The Scale Factor Equation
Our simulations calculate a single scale factor at each time, neglecting metric fluctuations. The equation for the scale factor $`a`$ is derived ¿from the following two equations
$$\ddot{a}=\frac{4\pi }{3}(\rho +3p)a$$
(19)
$$\left(\frac{\dot{a}}{a}\right)^2=\frac{8\pi }{3}\rho $$
(20)
where $`\rho `$ and $`p`$ are the total energy density and pressure, respectively. For simplicity we will first solve these for a single scalar field $`\sigma `$. The energy density and pressure can be derived ¿from the energy momentum tensor
$$T_{\mu \nu }=\sigma _{,\mu }\sigma _{,\nu }\frac{1}{2}g_{\mu \nu }g^{\alpha \beta }\sigma _{,\alpha }\sigma _{,\beta }+g_{\mu \nu }V(\sigma ).$$
(21)
Assuming the fields are isotropic this equation can be solved and compared to the usual energy momentum tensor of matter
$$T_\nu ^\mu =diag\{\rho ,p,p,p\}$$
(22)
to give
$$\rho =\frac{1}{2}\dot{\sigma }^2+\frac{1}{2a^2}|\sigma |^2+V(\sigma )$$
(23)
$$p=\frac{1}{2}\dot{\sigma }^2\frac{1}{6a^2}|\sigma |^2V(\sigma ).$$
(24)
Plugging these expressions into Eq. (19) and using Eq. (20) to eliminate the $`\dot{\sigma }`$ term gives
$$\ddot{a}=2\frac{\dot{a}^2}{a}+8\pi a\left(\frac{1}{3a^2}|\sigma |^2+V(\sigma )\right).$$
(25)
Switching now to the fields $`\varphi `$ and $`\chi _i`$ and to the potential in Eq. (13),
$$\ddot{a}=2\frac{\dot{a}^2}{a}+8\pi a\left(\frac{1}{3a^2}\left(|\varphi |^2+|\chi _i|^2\right)+\frac{\lambda }{4}\left(\varphi ^2v^2\right)^2+\frac{1}{2}g^2\varphi ^2\chi _i^2\right).$$
(26)
Finally we rescale all variables according to Eq. (16) and take a spatial average (denoted by “$`<>`$”) over the grid
$$a^{\prime \prime }=\frac{a^2}{a}+\frac{8\pi \varphi _0^2}{a}\frac{1}{3}\left(|_{pr}\varphi _{pr}|^2+|_{pr}\chi _{i,pr}|^2\right)+\frac{1}{4}\left(\varphi _{pr}^2a^2v_{pr}^2\right)^2+\frac{1}{2}ge\varphi _{pr}^2\chi _{i,pr}^2.$$
(27)
### 6.3 Initial Conditions
Although the field equations are solved in configuration space with each lattice point representing a position in space, the initial conditions are set in momentum space and then Fourier transformed to give the initial values of the fields and their derivatives at each grid point. It is assumed that no significant particle production has occurred before the beginning of the program, so Minkowski space quantum fluctuations are used for setting the initial values of the modes.<sup>1</sup><sup>1</sup>1This approximation is usually very good for fluctuations with momenta greater than $`H`$. For smaller momenta one can find a better approximation after performing the field quantization in curved space, see e.g. , but in our case the corresponding corrections only slightly modify the final result.
The calculation of these modes is given in detail in . The result is that each mode $`\sigma _k`$ has a probability distribution given by
$$P(\sigma _k,\sigma _k^{})e^{2\omega _k\sigma _k\sigma _k^{}}$$
(28)
where $`\omega _k\sqrt{k^2+m^2}`$. Separating $`\sigma _k`$ into a magnitude and a phase, the phase has a uniform probability distribution and the magnitude has a Rayleigh distribution
$$P(|\sigma _k|)|\sigma _k|e^{2\omega _k|\sigma _k|^2}.$$
(29)
The momentum $`k`$ is simply the Fourier transform variable and the mass is given by
$$m_\sigma ^2=\frac{^2V}{\sigma ^2}\{\begin{array}{cc}3\lambda \varphi _0^2\hfill & ,\varphi \hfill \\ g^2\varphi _0^2\hfill & ,\chi _i\hfill \end{array}$$
(30)
Taking into account the finite size of the box, $`L`$, and the discretization of the spatial points with spacing $`dx`$ and converting to the rescaled variables defined above yields for the initial magnitudes
$$|\varphi _{k,pr}|=\frac{\sqrt{\lambda }L_{pr}^{3/2}}{\sqrt{2}dx_{pr}^3\left(k_{pr}^2+3\right)^{1/4}}$$
(31)
$$|\chi _{ik,pr}|=\frac{\sqrt{\lambda }L_{pr}^{3/2}}{\sqrt{2}dx_{pr}^3\left(k_{pr}^2+ge\right)^{1/4}}$$
(32)
times a Rayleigh distributed random number with standard deviation $`1`$. Note that the program values of $`L`$, $`dx`$, and $`k`$ are defined by the rescaling of $`x`$, and recall that $`geg^2/\lambda `$. Finally the zero mode, which appears as a value uniformly added to all grid points at the beginning of the calculation, is set to $`0`$ for $`\chi _i`$ and $`1`$ for $`\varphi `$ (since $`\varphi _{pr}=\varphi /\varphi _0`$).
In order to set the field derivatives it is necessary to know the time dependence of the vacuum fluctuations being considered. In Minkowski space this time dependence is given simply by the term $`e^{i\omega t}`$ which suggests
$$\dot{\sigma }_k=i\omega _k\sigma _k.$$
(33)
Converting to rescaled values of time, mass, and momentum gives
$$\varphi _{k,pr}^{}=i\varphi _{k,pr}\sqrt{k_{pr}^2+3}$$
(34)
$$\chi _{ik,pr}^{}=i\chi _{ik,pr}\sqrt{k_{pr}^2+ge}$$
(35)
The initial derivatives of the zero modes of $`\varphi `$ and $`\chi _i`$ are set to $`0`$.
Note that having been eliminated from the equations of motion, the only place $`\lambda `$ shows up in the calculations at all aside from the ratio $`g^2/\lambda `$ is in the $`\sqrt{\lambda }`$ term in the magnitude of the initial fluctuations. The coupling constant $`g`$ appears nowhere except in $`g^2/\lambda `$. Meanwhile the initial value of the field $`\varphi `$, i.e. $`\varphi _0`$, is used in the rescaling of the field and spacetime variables but it appears neither in the equations of motion for the fields nor in the amplitude of their initial fluctuations. In fact it appears in two places in the calculation. The first is in the evolution equation for the scale factor, Eq. (27). The second way the initial value of $`\varphi `$ enters the calculations is subtler. We said that the zero mode of $`\varphi _{pr}^{}`$ is initially set to $`0`$. At first this may seem like a poor choice since the field $`\varphi `$ initially must be rolling towards $`0`$. In fact the beginning of the program is supposed to represent the end of inflation when the slow roll approximation is no longer valid, so why should $`\varphi _{pr}^{}`$ be set to $`0`$? The answer comes from the use of the conformal field $`\varphi _{pr}a\varphi `$. As time goes on $`\varphi `$ is decreasing but $`a`$ is increasing, and there is a moment when these two balance and $`\varphi _{pr}^{}`$ is momentarily $`0`$. By setting $`\varphi _{pr}^{}=0`$ the program automatically sets the beginning of the calculation at this moment. For the potential described here this moment occurs when $`\varphi .35M_p`$, so the initial conditions implicitly use the value of $`\varphi _0`$. In the one place in the program where it does appear (in the scale factor evolution) this parameter is set to $`.35`$. This value is also useful in converting the output of the program to physically meaningful units.
### 6.4 The Calculational Method: Staggered Leapfrog
The differential equations derived above are solved using a staggered leapfrog algorithm in which the field values and their time derivatives are calculated at alternating times separated by a half time-step. The time step is kept constant throughout the calculation. (Since the program uses conformal time the physical time elapsed during each time step changes as the program progresses.) This method is stable for second order equations involving no first time derivatives such as the field equations we use, see e.g. . However, extra care must be taken in solving the equation for the scale factor, Eq. (27), since this does contain a first derivative term $`a^{}`$. A naive calculation using the leapfrog algorithm as described above would mean that $`a^{\prime \prime }`$ would be calculated at time $`\tau `$ as a function of $`a(\tau )`$ and $`a^{}(\tau d\tau /2)`$. There is a solution for this, although in practice the evolution of $`a`$ is so slow and smooth that this problem makes no practical difference. We avoid this problem, though, by using the two following equations<sup>2</sup><sup>2</sup>2G.F. would like to thank Julian Borrill, who suggested this solution.
$$a^{}(\tau +d\tau /2)a^{}(\tau d\tau /2)+d\tau a^{\prime \prime }(\tau )$$
(36)
$$a^{}(\tau )\frac{1}{2}\left(a^{}(\tau +d\tau /2)+a^{}(\tau d\tau /2)\right).$$
(37)
Solving these simultaneously, using Eq. (27) for $`a^{\prime \prime }`$, gives
$$a^{}(t+d\tau /2)a^{}(\tau d\tau /2)+\frac{2a}{d\tau }\left(1+\sqrt{1+2\frac{a^{}(\tau d\tau /2)}{a}d\tau +\frac{f}{a}d\tau ^2}\right)$$
(38)
where
$$f(a(\tau ))=\frac{8\pi \varphi _0^2}{a}\frac{1}{3}\left(|\varphi _{pr}|^2+|\chi _{i,pr}|^2\right)+\frac{1}{4}\left(\varphi _{pr}^2a^2v_{pr}^2\right)^2+\frac{1}{2}ge\varphi _{pr}^2\chi _{i,pr}^2.$$
(39)
Since Eq. (27) needs to be solved only once per time step this correction involves virtually no added computational time.
|
warning/0004/quant-ph0004046.html
|
ar5iv
|
text
|
# Classical and quantum dynamics of a spin-1/2
## Abstract
We reply to a comment on ‘Semiclassical dynamics of a spin-$`\frac{1}{2}`$ in an arbitrary magnetic field’.
In a recent Comment Kochetov argues that our results on the coherent state path integral for a spin-$`\frac{1}{2}`$ in an arbitrary magnetic field are based on a ‘classical spin action inconsistent with the necessary boundary conditions’. Contrary from what is insinuated by the Comment our article is not concerned with the quantization of a classical spin but solely with the representation of a quantum spin-$`\frac{1}{2}`$ in terms of a spin coherent state path integral. Hence, the Comment by Kochetov arguing primarily on a classical level is only vaguely relevant to our work and chiefly reconsiders the author’s earlier work in the light of results in . Indeed, the analysis in allows for some conclusions of relevance to Kochetov’s work as discussed below.
Before addressing the Comment more specifically let us briefly reformulate the approach in . We start with the two-dimensional Hilbert space, represented in the basis of spin coherent states $`|\mathrm{\Psi }_g>=𝒟^{1/2}(g)|>`$, where $`gSU(2)`$ . Since $`|\mathrm{\Psi }_h>=\mathrm{exp}(\mathrm{i}\alpha )|>`$ for a $`h`$ in the maximal torus, $`|\mathrm{\Psi }_g>`$ and $`|\mathrm{\Psi }_g^{}>`$ describe the same physical state if there exists a $`hU(1)`$ such that $`g^{}=gh`$. Therefore the group $`SU(2)`$ can be viewed as fibre bundle over the base manifold $`SU(2)/U(1)S^2`$ with fibre $`U(1)`$ and the space of distinct spin coherent states is canonically isomorphic to these left cosets. Parametrizing any $`gSU(2)`$ with Euler angles $`(\vartheta ,\phi ,\chi )`$, we get
$$|\mathrm{\Omega }>=\mathrm{e}^{\frac{\mathrm{i}}{2}\chi }\mathrm{e}^{\mathrm{i}\phi S_z}\mathrm{e}^{\mathrm{i}\vartheta S_y}|>.$$
(1)
Here, the first factor on the righthand side is just a phase factor and the rest determines the physical state. These states are not orthogonal but form an overcomplete basis in the Hilbert space. The overlap is readily evaluated and the identity may be represented as
$$I=\frac{1}{2\pi }\mathrm{sin}(\vartheta )d\vartheta d\phi |\mathrm{\Omega }\mathrm{\Omega }|.$$
(2)
Employing a Trotter decomposition, the propagator may be written as
$`<\mathrm{\Omega }^{\prime \prime }|U(t)|\mathrm{\Omega }^{}>`$ $`=`$ $`\underset{ϵ0}{lim}N{\displaystyle \underset{k=1}{\overset{n}{}}\sqrt{det\omega _{ij}}\mathrm{d}\vartheta _k\mathrm{d}\phi _k}`$ (4)
$`\times \mathrm{exp}\{{\displaystyle \underset{k=0}{\overset{n}{}}}[\mathrm{log}<\mathrm{\Omega }_{k+1}.|\mathrm{\Omega }_k>+{\displaystyle \frac{\mathrm{i}}{2}}(\chi _{k+1}\chi _k)\mathrm{i}ϵ{\displaystyle \frac{<\mathrm{\Omega }_{k+1}|H(kϵ)|\mathrm{\Omega }_k>}{<\mathrm{\Omega }_{k+1}.|\mathrm{\Omega }_k>}}]\}.`$
where $`ϵ=t/n`$, $`(\mathrm{\Omega }_0,\chi _0)=(\mathrm{\Omega }^{},\chi ^{})`$, $`(\mathrm{\Omega }_{n+1},\chi _{n+1})=(\mathrm{\Omega }^{\prime \prime },\chi ^{\prime \prime })`$. We are allowed to pass to the continuum limit, if the paths stay continuous for $`ϵ0`$ which is not guaranteed if no Wiener measure occurs. Therefore care must be taken in calculating the path integral -. To ensure an integration over continuous Brownian motion paths, we introduce a regularization by the spherical Wiener measure and are then allowed to write
$`<\mathrm{\Omega }^{\prime \prime }|U(t)|\mathrm{\Omega }^{}>`$ $`=`$ $`\underset{\nu \mathrm{}}{lim}N{\displaystyle \underset{s=0}{\overset{t}{}}\sqrt{det\omega _{ij}}\mathrm{d}\vartheta (s)\mathrm{d}\phi (s)}`$ (6)
$`\times \mathrm{exp}\left\{\mathrm{i}{\displaystyle _0^t}ds\left[{\displaystyle \frac{\mathrm{i}}{\nu }}(g_{\vartheta \vartheta }\dot{\vartheta }^2+g_{\phi \phi }\dot{\phi }^2)+\theta _\vartheta \dot{\vartheta }+\theta _\phi \dot{\phi }H(\vartheta ,\phi ,s)\right]\right\}.`$
Here, $`N=lim_n\mathrm{}_{k=1}^n\frac{1}{\pi }`$ is a normalization factor, $`g=\frac{1}{4}(\mathrm{d}\vartheta \mathrm{d}\vartheta +\mathrm{sin}(\vartheta )^2\mathrm{d}\phi \mathrm{d}\phi )`$ the metrical tensor, $`\omega =\frac{1}{2}\mathrm{sin}(\vartheta )\mathrm{d}\vartheta \mathrm{d}\phi `$ the symplectic two-form of $`SU(2)/U(1)`$ and $`\theta =\frac{1}{2}(\mathrm{cos}(\vartheta )\mathrm{d}\phi +\mathrm{d}\chi )`$ its corresponding symplectic potential ($`\omega =\mathrm{d}\theta `$).
Choosing in every left coset one special representant, i.e. fixing $`\chi `$ for every coherent state, one defines a section of the $`SU(2)`$ bundle. In particular, the choice $`\chi =0`$ was adopted in . It is important to note that once $`\chi `$ has been fixed the symplectic potential is fixed as well and manipulations of the form suggested by Kochetov in equation (3) are no longer allowed. The very same reasoning applies in the parametrization used by Kochetov. Within the Gaussian decomposition of the elements of $`SU(2)`$ by $`g=z_{}hz_+`$ for $`z_{}Z_{}`$, $`hU(1)`$, $`z_+Z_+`$ or equivalently $`g=z_{}b_+`$ with $`b_+B_+`$, we recognize that $`𝒟^{1/2}(g)|>=𝒟^{1/2}(z_{}h)|>`$. Parametrizing $`z_{}`$ by the complex number $`\zeta `$, the space of distinct spin coherent states is now isomorphic to elements in $`SL(2,𝒞)/B_+SU(2)/U(1)`$. If the isomorphismus is defined explicitly by the spherical projection from the south pole of the sphere onto the complex plane, one has $`\zeta =\mathrm{tan}\left(\frac{\vartheta }{2}\right)\mathrm{e}^{\mathrm{i}\phi }`$, and makes use of a different section of the $`SU(2)`$ bundle by setting $`\chi =\phi `$. Therefore a corresponding phase factor appears
$$|\zeta >=\frac{1}{\sqrt{1+|\zeta |^2}}\mathrm{e}^{\zeta S_{}}|>=\mathrm{e}^{\frac{\mathrm{i}}{2}\phi }|\mathrm{\Omega }>.$$
(7)
Again with the choice $`\chi =\phi `$ there is no room for additional manipulations of the form (3) in . While Kochetov’s theory starts from a classical spin and employs geometric quantization to obtain a quantum propagator after an ad hoc modification of the symplectic potential, no such ambiguities arise if the representation of the quantum propagator in terms of a path integral is considered.
A main point in the critique by Kochetov is the claim that the approach in disagrees with boundary conditions in the classical limit. Since the physical states form a symplectic two-dimensional differential manifold with the closed two-form $`\omega `$, the classical dynamics is determined by the Hamiltonian vector field $`\omega (X_H,)=\mathrm{d}H`$ which leads immediately to the classical equations of motion (23) in . Note that in general there is no classical path connecting arbitrary but real boundary conditions $`\overline{\mathrm{\Omega }}(0)=\overline{\mathrm{\Omega }}^{}`$ and $`\overline{\mathrm{\Omega }}(t)=\overline{\mathrm{\Omega }}^{\prime \prime }`$. This is known as the ‘overspecification problem’. Kochetov modifies the action to allow always for a ‘classical’ path, which is usually complex. If one applied the same rules to a simple harmonic oscillator there would be a ‘classical’ path connecting any initial phase space point $`(q^{},p^{})`$ with any endpoint $`(q^{\prime \prime },p^{\prime \prime })`$. This is clearly not what is usually meant by classical. Hence, the overspecification problem should not be removed in the classical limit. Yet, in the quantum problem, there is indeed a semiclassical path for any pair of real boundary conditions \[see equations (23) and (24),(25) in \].
Finally, Kochetov believes that the exactness of the semiclassical propagator is ‘obvious’ and ‘self-evident’. Replacing Kochetov’s qualitative arguments by a more accurate treatment , one finds the necessary condition $`\theta (X_H)=H`$ that $`\theta `$ is $`SU(2)`$ invariant and the stationary phase approximation becomes exact. For a Hamilton operator which is a linear combination of all three generators of the $`su(2)`$ algebra we get three conditions which cannot be satisfied generally by fixing the phase $`\chi `$ appropriately. Therefore there are no $`SU(2)`$ invariant potentials on $`S^2`$ and although $`SU(2)`$ is the group of isometric canonical transformations on the two-sphere , it does not preserve the sections. Hence, for magnetic fields of arbitrary time-dependence, the exactness of dominant stationary phase approximation (DSPA) is not “self-evident”, and prior to our work it was rather expected that the DSPA does not provide a correct result .
|
warning/0004/cond-mat0004394.html
|
ar5iv
|
text
|
# Effect of quantum interference in the nonlinear conductance of microconstrictions
## I Introduction.
The Scanning Tunneling Microscopy (STM) and the Mechanically Controllable Break-junction (MCB) techniques offer an opportunity to study the conductance of metallic contacts consisting of only a few atoms (quantum contacts). The electrical conductance of such contacts, at small bias voltage is proportional to the number of propagating electron modes, $`N`$, each one contributing an amount of $`G_0=2e^2/h`$. With increasing the diameter of the contact the energies of modes continuously decreases, but the number of modes increases whenever a new mode fits into the constriction cross section. This number, $`N,`$ is limited by the requirement that the kinetic energy for the transverse motion is smaller than the Fermi energy $`\epsilon _F`$. When a new mode is occupied, a new quantum channel is opened. The conductance then undergoes a jump of $`G_0`$. Such quantization of conductance has been observed in both two and three dimensional contacts with diameters comparable to Fermi wave-length $`\lambda _F=h/p_F`$ ($`p_F`$ is the Fermi momentum) . Jumps in conductance are also expected to occur, at the constant contact diameter while bias voltage is varied. If the bias $`eV`$ is larger than the distances between the energy levels of quantum modes, it is possible to change the number of opened modes by changing the voltage $`V.`$ At a certain threshold voltage a channel is opened or closed for one direction of the electron wave vector along the constriction and consequently conductance suffers a $`G_0/2`$ stepwise change .
Quantum interference effects have been studied in different mesoscopic systems . In ballistic metallic microconstrictions it manifests itself as fluctuations in conductance when a magnetic field or an electrical voltage is applied . Now experimental efforts have been done using MCB techniques, in order to measure conductance as a function of voltage in atomic-size point contacts . A prominent feature of these measurements, is the existence of small random voltage-dependent fluctuations in conductance, far from steps. The measurements clearly indicate suppression of the fluctuations for conductance values near the integer multiples of the conductance quantum. Similar results have been reported by using a STM to show the strong voltage dependence of conductance of one-atom contacts at different temperatures . It is generally believed that the observed oscillations in conductance are due to the quantum interference effects . Ludoph and co-authors , propose the following interpretation: The electron wave transmitted through the contact is backscattered to the contact by an impurity and then partially reflected at the contact. These waves interfere and change the total conductivity. The energy and thus the wave number of an injected electron into the channel, depends on the voltage. Consequently the interference pattern in conductance oscillates as electron wave number varies with the voltage.
Although the theory developed by Ludoph et al. can explain the general feature of fluctuations but here we try to examine a different mechanism. Impurities (or defects) are assumed to be located inside the constriction, and the interference is effectively between waves scattered from the impurities. The existence of a few defects or impurities inside the constriction is rather natural considering the way the contact is formed. Using the model of a long microconstriction we can find the conductivity analytically. We discuss the theory of nonlinear electron transport through a mesoscopic microconstriction with a few impurities. We show that the nonlinear dependence of the quantum conductance on the voltage is obtained from this model. The form of this dependence is affected not only by the distances between impurities, but also by their positions inside the constriction.
In Sec. II the model Hamiltonian is discussed and is used to obtain a general expression for the nonlinear conductance. In Sec. III a $`\delta `$-function potential is assumed for the interaction of electrons with impurities and a simplified equation for the conductance is obtained. Within the framework of perturbation theory, a general analytical equation is also derived for conductance of the system, for arbitrary number of quantum modes and arbitrary number of impurities located in arbitrary positions. These analytical results are illustrated by numerical calculations for the contact in the form of a long cylindrical contact.A brief discussion of result is given in Sec. IV.
## II General equation for the nonlinear conductance of the long quantum microconstriction.
Let us consider a long narrow constriction, which connects two bulk metals, assuming $`eV\epsilon _F`$. The geometry is shown in Fig. 1. We assume that the contact shape is smooth on the scale of the wavelength $`\lambda _F.`$ This condition assures that different transverse modes pass through the ballistic contact independently (adiabatic approximation ). We also assume that the contact length is much larger than its diameter and we can neglect the constriction end effects. Under these approximations, the electrical field inside the contact far from the ends is negligible and the energy $`\epsilon `$ of ballistic electrons depends only on the sign of velocity along the contact axis . The Hamiltonian $`H`$ of the electrons contains the following terms:
$$H=H_0+H_1+H_{int},$$
(1)
where
$$H_0=\underset{n}{}\epsilon _\alpha c_\alpha ^{}c_\alpha $$
(2)
is Hamiltonian of free electrons, and
$$H_1=\frac{eV}{2}\underset{\alpha }{}signv_zc_\alpha ^{}c_\alpha $$
(3)
describes the influence of applied bias voltage. $`H_{int}`$ denotes interaction of electrons with impurities, and depends on the positions of impurities $`𝐫_i`$ in the constriction; the operator $`c_\alpha ^+\left(c_\alpha \right)`$ creates (annihilates) a conduction electron with the wave function $`\phi _\alpha ,`$ and energy $`\epsilon _\alpha .`$ The electron wave functions and eigenvalues are
$`\phi _\alpha (𝐫)`$ $`=`$ $`\psi _\beta (𝐑)\mathrm{exp}\left({\displaystyle \frac{i}{\mathrm{}}}p_zz\right);`$ (4)
$`\epsilon _\alpha `$ $`=\epsilon _\beta +{\displaystyle \frac{p_z^2}{2m_e}};`$ (5)
where $`\alpha =(\beta ,p_z),`$ $`\beta `$ is the set of two transverse quantum numbers; $`p_z`$ is the momentum of an electron along the contact axis. $`𝐫=(𝐑,z),`$ $`𝐑`$ is a coordinate in the plain, perpendicular to the $`z`$ axis $`m_e`$ is the electron mass.
In zero approximation in $`H_{int}`$ the current $`J_0`$ through the contact area $`S_c`$ is
$$J_0=eS_cTr\left(v_z\rho _0\right);$$
(6)
where
$$\rho _0=f_F\left(H_0+H_1\right);$$
(7)
$`v_z=p_z/m`$ is the electron velocity; $`f_F`$ is the Fermi function. Using the Eqs.(6),(7), and wavefunctions (4), we find the equation for the ballistic conductance:
$$G_1=\frac{1}{2}G_0\underset{\beta }{}\left[f_F\left(\epsilon _\beta +\frac{eV}{2}\right)+f_F\left(\epsilon _\beta \frac{eV}{2}\right)\right]$$
(8)
At zero temperature and $`V0`$, this formula describes the well known $`G_0`$ steps of quantum conductance and in the quasiclassical case it turns into the Sharvin conductance .
In order to investigate the influence of single impurities on the nonlinear quantum conductance of the point contact, we use the method, which was developed Kulik and others . The change in the electrical current $`\mathrm{\Delta }J`$ is related to the rate of energy dissipation by the relation:
$$\mathrm{\Delta }JV=\frac{dE}{dt}=\frac{dH_1}{dt};$$
(9)
Differential of $`H_1`$ with respect to time $`t`$ is we obtained from Heisenberg equation. The change $`\mathrm{\Delta }J`$ of the current due to interactions of electrons with impurities; would then be
$$\mathrm{\Delta }JV=\frac{1}{i\mathrm{}}[H_1\left(t\right),H_{int}\left(t\right)],$$
(10)
where
$$\mathrm{}=Tr\left(\rho \left(t\right)\mathrm{}\right).$$
(11)
All operators are in the interaction representation.
The statistical operator $`\rho \left(t\right)`$ satisfies the equation
$$i\mathrm{}\frac{\rho }{t}=[H_{int}\left(t\right),\rho \left(t\right)],$$
(12)
which can be solved using perturbation theory in $`H_{int}`$ (but for the arbitrary $`H_1`$):
$$\rho \left(t\right)=\rho _0+\frac{1}{i\mathrm{}}\underset{\mathrm{}}{\overset{t}{}}𝑑t^{}[H_{int}\left(t^{}\right),\rho _0]+\mathrm{}$$
(13)
We would then have
$$\mathrm{\Delta }J=\frac{1}{\mathrm{}^2V}\underset{\mathrm{}}{\overset{t}{}}𝑑t^{}Tr\left(\rho _0[[H_1,H_{int}(t)],H_{int}(t^{})]\right).$$
(14)
The decrease in total conductance $`\mathrm{\Delta }G=G_2,`$ results the quantum interference is defined as
$$G_2=\frac{d\mathrm{\Delta }J}{dV}.$$
(15)
If the applied bias $`eV`$ is much smaller than the differences between the energies $`\epsilon _\beta `$ of modes, the Eq. (15) describes the dependence of total conductance on the voltage $`V.`$
## III The conductance oscillations.
Now using the general Eqs. (14), (15), we investigate the behaviour of $`G_2`$ for the case of $`\delta `$ function scattering potential. The Hamiltonian $`H_{int}`$ can be written as
$$H_{int}(𝐫_j)=g\underset{\alpha \alpha ^\mathrm{`}}{}\phi _\alpha ^{}\left(𝐫_j\right)\phi _\alpha ^{}\left(𝐫_j\right)c_\alpha ^{}c_\alpha ^{}.$$
(16)
Here $`g`$ is the coupling constant of the interaction of an electron with an impurity located in the point $`𝐫_j`$.
Substituting Eqs.(7),(16) into Eq.(14), after some simple but cumbersome calculations we find
$`\mathrm{\Delta }J={\displaystyle \frac{e\pi }{2\mathrm{}}}g^2{\displaystyle \underset{\alpha ,\alpha ^{}}{}}{\displaystyle \underset{i,j}{}}(signv_{z\alpha }signv_{z\alpha ^{}})\times `$ (17)
$`\phi _\alpha ^{}^{}(𝐫_j)\phi _\alpha ^{}(𝐫_i)\phi _\alpha ^{}(𝐫_i)\phi _\alpha (𝐫_j)(f_\alpha ^{}f_\alpha )\delta (ϵ_\alpha ^{}ϵ_\alpha ),`$ (18)
where $`f_\alpha =f_F\left(\epsilon +\frac{eV}{2}signv_{z\alpha }\right)`$. At zero temperature $`f_F=\mathrm{\Theta }\left(\epsilon _F\epsilon \right),`$ the Eq.(17) can be further simplified. Using the wave functions (4), we obtain for nonlinear part of conductance the following equation:
$`G_2`$ $`=G_0{\displaystyle \frac{\pi m_eg^2}{2}}{\displaystyle \underset{\beta ,\beta ^{},i,j}{}}\{\mathrm{cos}\left[{\displaystyle \frac{1}{\mathrm{}}}(p_\beta ^{(+)}+p_\beta ^{}^{(+)})\mathrm{\Delta }z_{ij}\right]{\displaystyle \frac{1}{p_\beta ^{(+)}p_\beta ^{}^{(+)}}}\mathrm{\Theta }(\epsilon _F\epsilon _\beta +{\displaystyle \frac{eV}{2}})\times `$ (19)
$`\mathrm{\Theta }(\epsilon _F\epsilon _\beta ^{}+{\displaystyle \frac{eV}{2}})+\mathrm{cos}\left[{\displaystyle \frac{1}{\mathrm{}}}(p_\beta ^{()}+p_\beta ^{}^{()})\mathrm{\Delta }z_{ij}\right]{\displaystyle \frac{1}{p_\beta ^{()}p_\beta ^{}^{()}}}\mathrm{\Theta }(\epsilon _F\epsilon _\beta {\displaystyle \frac{eV}{2}})\times `$ (20)
$`\mathrm{\Theta }(\epsilon _F\epsilon _{\beta n^{}}{\displaystyle \frac{eV}{2}})\}\psi _\beta ^{}^{}(𝐑_j)\psi _\beta ^{}(𝐑_i)\psi _\beta ^{}(𝐑_i)\psi _\beta (𝐑_j).`$ (21)
Where
$$p_\beta ^{(\pm )}=\sqrt{2m_e(\epsilon _F\pm \frac{eV}{2}\epsilon _\beta ).}$$
(22)
The cosine terms in the Eq. (18) describe the conductance oscillations due to the interference of electrons waves scattered by impurities. The transverse parts $`\psi _\beta (𝐑_j)`$ of wave functions contain the mesoscopic effect of impurity positions inside the constriction. The equation (18) diverges at $`p_\beta ^{(\pm )}=0.`$ Physically it means that in the Born approximation the slowly moving electron is repeatedly scattered on the impurity. In this case the perturbation theory (Born approximation) is not valid any more, and we must take into account the interference of partial waves under the electron scattering by impurity. We assume that energy levels are not very close to the boundary energies $`\epsilon _F\pm \frac{eV}{2}`$ and the quantity $`G_2`$ added to the ballistic conductance $`G_1`$ (8) is small.
For the numerical calculations we have used the free electron model of point contact consisting of two infinite half-spaces connected by a long ballistic cylinder of a radius $`R`$ and a length $`L`$ (Fig.1).
In the limit $`L\mathrm{}`$ the electron wave functions $`\phi _\alpha ^{}\left(𝐫\right)`$ and energies $`\epsilon _\alpha `$ can be written as
$$\phi _\alpha ^{}\left(𝐫\right)=\frac{1}{\sqrt{\mathrm{\Omega }}J_{m+1}\left(\gamma _{mn}\right)}J_m\left(\gamma _{mn}\frac{\rho }{R}\right)\mathrm{exp}\left(im\phi +\frac{i}{\mathrm{}}p_zz\right);$$
(23)
where
$$\epsilon _\alpha =\epsilon _{mn}+\frac{p_z^2}{2m_e};\epsilon _{mn}=\frac{\mathrm{}^2}{2m_eR^2}\gamma _{mn}^2$$
(24)
We have used cylindrical coordinates $`𝐫=(\rho ,\phi ,z)`$ with $`z`$ along the axis of cylinder. Here $`a=(n,m,p_z)`$ are the quantum numbers, $`\mathrm{\Omega }=\pi R^2L`$ is the volume of the channel, $`\gamma _{mn}`$ are the n-th zero of the Bessel function $`J_m.`$ Since the electron energy has degeneracy for azimuthal quantum number $`m`$ ( as a result of the symmetry of the model), quantum modes with $`\pm m`$ have the same contribution in conductance. In this model, Sharvin conductance has not only steps $`G_0,`$ but also steps $`2G_0`$ . In Fig.2 the dependence of the interference pattern on the number of impurities inside a constriction with constant radius is shown. It shows that as a result of the interference of electron waves, which were scattered by different impurities, the interference maxima in $`G_2\left(V\right)`$ dependence, may both be depressed and increased. The interference oscillation of the conductance depends strongly on the number of opened quantum modes which follows from the dependence of its maximum value of longitudinal electron momentum (see, Eq. 3) on the contact size. The voltage dependence of $`G_2`$ for different contact sizes are shown in Fig.3. Fig. 4. illustrates how the changing in the nonlinear dependence $`G_2\left(V\right)`$ changes with the contact size. It corresponds to the case of two impurities in the contact. The position of impurities and the number of opened quantum modes are kept constant. The difference in the interference oscillations is a result of the changing in the relative positions of nodes and maxima of the electron wave function from the points, in which the impurities are situated.
## IV Conclusion
The dependence of quantum conductance of metallic ultrasmall contacts containing impurities on bias voltage has been theoretically studied. We have shown that impurities situated inside the quantum microconstriction produce a nonlinear dependence of the conductance on the applied voltage, which is the result of the interference of electron waves reflected by impurities. The transmission probability of the electron through constriction depends on the relation between the electron wave length $`\lambda `$ and $`\mathrm{\Delta }z_{ij},`$ the projection of distances between impurities along the channel. It is maximum when the condition $`\mathrm{\Delta }z_{ij}=\frac{n\lambda }{2}`$ ( $`n`$ is integer) is satisfied. Since the electron momentum depends on the applied bias, one can change the transmission by changing the voltage. Our numerical calculations show that the resulting nonmonotonic dependence of the conductance, is similar in shape to the ones observed in experiment . The amplitude of interference pattern is sensitive to the transversal position of impurities inside the constriction. If the impurity is located near the point, where the electron wave function corresponding to the $`n`$th quantum mode vanishes, then the decreasing of transmission of that mode would be negligible.
We acknowledge fruitful discussion with M.R.H. Khajehpour and I.K.Yanson.
|
warning/0004/hep-th0004052.html
|
ar5iv
|
text
|
# Acknowledgements
## Acknowledgements
This research is supported in part by PPARC. DB and GVK thank the Theory Division at CERN for hospitality received during the completion of this work.
|
warning/0004/astro-ph0004298.html
|
ar5iv
|
text
|
# Optical and Near Infrared Study of the Cepheus E outflow, a very low excitation object.
## 1 Introduction
The Cepheus E (Cep E) outflow was first detected in the <sup>12</sup>CO J=1-0 transition in some of the early radio studies of star formation in molecular clouds . In his catalog of molecular outflows, Fukui (1989) first pointed out the presence of a bipolar, high velocity outflow in this region. The bipolar nature of Cep E became clear with the K image of the molecular outflow obtained by Hodapp (1994), which included the stronger NIR H<sub>2</sub> lines, and revealed a relatively compact system (with a size of $``$ 1.5). Subsequent studies in the near/mid infrared have shown that the outflow is quite bright in the H<sub>2</sub> (1,0) S(1) 2.12 $`\mu `$m line, consistent with models of shock excited H<sub>2</sub> gas . Suttner et al. (1997) have used three-dimensional hydrodynamic simulations of highly collimated molecular outflows in order to model the morphology of Cep E. These simulations, however, required a very high density in the jet ($`10^5`$ cm<sup>-3</sup>), which are not consistent with the hot and dense CO bullets recently found in the flow , with densities of $`10^4`$ cm<sup>-3</sup>.
The IRAS 23011+6126 source was originally identified as the main candidate for the outflow source. However, the presence of multiple outflows in near infrared and radio wavelengths indicates the existence of at least two sources, which have recently been confirmed by OVRO observations at 1.3 and 2.6mm . The sources are embedded and invisible at optical and near infrared wavelengths, and are likely to be Class I or Class 0 protostellar objects . In addition, Noriega-Crespo et al. (1998) detect one source at 6.9 $`\mu `$m using ISOCAM, which is well detected in all IRAS bands.
The present study has been motivated by the detection of emission at optical wavelengths in a small section of the southern lobe of the Cepheus E outflow by Noriega-Crespo (1997) and Devine et al. (1997). Noriega-Crespo (1997) mentions that H$`\alpha `$ and \[SII\] 6717/31 images reveal a compact knot which is the optical counterpart of the southern bowshock observed at 2 $`\mu `$m by Eislöffel et al. (1996). This optical knot has been named HH 377 .
In this study, we explore the link between the physical properties of the outflow as determined from optical imaging and spectroscopy, and compare these results with those obtained from observations in the near infrared. Our goal is to understand the development of very young stellar outflows (we notice that Cep E has a dynamical age of $`3\times 10^3`$ years (Noriega-Crespo et al. 1998)) and the relationship between the mechanisms that produce the infrared and optical emission.
The paper is organized as follows. In Section 2 we describe the different observations obtained for this work, and comment on the reduction and calibration techniques. In Section 3, we present the results obtained from our infrared and optical observations. Finally, in Section 4 we compare the physical properties of Cepheus E deduced from the optical and the NIR observations with other Herbig Haro objects.
## 2 Observations
The optical and infrared observations were carried out at three different observatories. The log of the observations is presented in Table 1, and they are described in detail below.
### 2.1 Near Infrared Imaging and Spectroscopy
The NIR images of Cep E were obtained at two observatories. A set of images was obtained at the Apache Point Observatory 3.5m telescope (APO 3.5m) with a $`256\times 256`$ array at f/5 and a 0.482<sup>′′</sup> per pixel scale. The central wavelengths (and bandpass) of these images were at 2.12$`\mu `$m (1$`\%`$), for the H<sub>2</sub> (1,0) S(1) line, and 2.22$`\mu `$m(4$`\%`$) for the nearby continuum.
Another set of images was obtained at the Observatorio Astronómico Nacional at San Pedro Mártir 2.1m telescope (OAN SPM 2.1m) with the IR camera/spectrometer CAMILA , which has a $`256\times 256`$ array providing a 0.85 <sup>′′</sup>/pixel scale at f/4.5. The central wavelengths (and bandpass) of the filters used to obtain these images were 2.12$`\mu `$m (1$`\%`$), 2.25$`\mu `$m (1$`\%`$) for the H<sub>2</sub> (2,1) line, cK (2$`\%`$) and Br$`\gamma `$ 2.16$`\mu `$m.
The frames were flattened with a combination of low and high illumination sky flats obtained at sunset. The data were processed by subtracting a median-filtered image of nearby sky frames taken with the same integration time and with offsets between 30 and 100<sup>′′</sup>. Bad pixels were removed with standard techniques.
The processing of our frames was done with Image Reduction and Analysis Facility (IRAF)<sup>1</sup><sup>1</sup>1IRAF is distributed by the National Optical Astronomy Observatory (NOAO), which is operated by the Association of Universities for Research in Astronomy (AURA), Inc. under cooperative agreement with the National Science Foundation based programs. For each filter, 10 (APO data) and 9 (OAN SPM data) overlapping frames were taken (see Table 1 for the integration times). The frames were aligned using several field stars, and median combined into the final images. The image scale and orientation were calculated using 7 common stars between the IR frames and an optical image from the Digitized Sky Survey, with a resultant mean uncertainty of less than one arcsecond in the positions.
The NIR spectra were obtained with the Multiple Mirror Telescope Observatory (MMTO). The set of H and K band spectra of the North and South lobes of Cep E were taken with the Rieke FSPEC IR spectrometer. The slit was aligned in the E-W direction in both lobes (see Figure 1) and alternate exposures were chopped between the source and the local sky. The atmospheric absorption and sensitivity variation along the dispersion were corrected by observing a bright late-F type star at nearly the same airmass as the target data. The spectra were obtained with a 1.0<sup>′′</sup> wide slit, and the extraction was done over an area of 4.0 $`(^{\prime \prime })^2`$ for each outflow lobe.
### 2.2 Optical Imaging and Spectroscopy
As mentioned above, the south lobe of Cep E is detected at optical wavelengths, in the H$`\alpha `$ and \[S II\] 6717/31 lines . Our narrow band images in H$`\alpha `$ and \[S II\] were obtained with the 1.2m telescope at Fred Lawrence Whipple Observatory (FLWO 1.2m). A thinned, back-side illuminated, AR coated Loral $`2048\times 2048`$ CCD was used with a plate scale of 0.315 <sup>′′</sup>/pix. The central wavelength and FWHM of the narrow band H$`\alpha `$, \[S II\] and continuum filters are, respectively, $`\lambda `$6563, 25 Å, $`\lambda `$6724, 50 Å, and $`\lambda `$6950, 400 Å. The CCD images were binned two by two giving a 0.63 <sup>′′</sup>/pix scale, and were processed using IRAF, in the standard way. For each filter, the final image corresponds to the median-filter of three 600 seconds frames.
A low resolution, long-slit spectrum over a $`45007000`$ Å wavelength range with $`2`$ Å/pix was obtained at the MMTO telescope. The spectrum was reduced using IRAF and flux calibrated using the standard star HR8687.
Two long-slit spectra were also obtained at the FLWO 1.5m telescope with the FAST spectrometer, using a Loral $`512\times 2688`$ coated CCD with 15$`\mu `$m pixels. Two different spectral resolutions (1.49 Å/pix and 0.75 Å/pix) were used to cover the wavelength ranges of 5500-7000 Å and 6200-6800 Å. The slit width was 1.1<sup>′′</sup> and the slit was oriented N-S. The spectra were flux calibrated with the standard stars BD284211 and G191B2B.
## 3 Results
### 3.1 Infrared excitation
The complex structure shown by Cepheus E in vibrationally excited molecular hydrogen lines has been described in previous papers . Figure Optical and Near Infrared Study of the Cepheus E outflow, a very low excitation object. shows a section of our H<sub>2</sub> (1,0) S(1) continuum subtracted image, in which we plot the slit positions used to obtain the infrared spectra.
Using the H$`{}_{2}{}^{}\lambda `$2.121$`\mu `$m and $`\lambda `$2.248$`\mu `$m images we construct a (1,0)/(2,1) S(1) line ratio image. The line ratio is nearly constant throughout the outflow with a mean value of $`10\pm 5`$. This result is completely consistent with previous results on this object .
In our Bracket $`\gamma `$ \+ continuum image we have detected emission extending over both lobes. The distribution of this emission is very similar to the one of the continuum cK frame. Because of this similarity we suspect that the flux detected in our Br$`\gamma `$ frame is on the whole continuum emission. This interpretation is consistent with the absence of Br$`\gamma `$ line in the spectra obtained for the brightest knots of both lobes (see Figure Optical and Near Infrared Study of the Cepheus E outflow, a very low excitation object.).
The spatially integrated spectra in the H and K bands of the northern and southern lobes of Cepheus E are shown in Figure Optical and Near Infrared Study of the Cepheus E outflow, a very low excitation object.. These spectra were constructed by integrating along the slit over the width of the lobes ($`4`$<sup>′′</sup>) for the slit positions indicated in Figure Optical and Near Infrared Study of the Cepheus E outflow, a very low excitation object.. The identified and measured H<sub>2</sub> lines in these bands are indicated on the plots at the expected wavelengths. Note that no Br$`\gamma `$ ($`\lambda 2.166\mu `$m) line emission is detected in these slit positions, which correspond to the brightest knots of each lobe. Table 2 lists the H<sub>2</sub> transitions (column 1), their wavelengths (column 2; see Black & van Dishoeck, 1987); the energy $`E(v,J)`$ of the upper level (column 3, Dabrowski 1984), and the measured fluxes of the identified H<sub>2</sub> lines (columns 4 and 5 for the N and S lobes, respectively). From these spectra we determine a $`\mathrm{\Delta }V_r`$ 90$`\pm 30`$ km s<sup>-1</sup> between the northern and southern lobes, and the south lobe shows a blueshift. Comparing the fluxes for (1,0) S(1) and (2,1) S(1) lines from our spectra with previous values obtained from surface photometry for these transitions , we find that our (1,0) S(1) flux values are lower with respect of the Eislöffel et al. values, by a factor $`2`$. Whereas in the case of (2,1) S(1) fluxes the present and previous measurements are completely consistent.
We have estimated the column densities $`N(v,J)`$ assuming that the lines are optically thin, in order to determine the excitation temperature, $`T_{exc}`$, of the warm molecular gas in each lobe. Under conditions of local thermodynamic equilibrium (LTE), the relationship between column density and energy of the upper level $`E(v,J)`$, is $`ln[N(v,J)/g_J]=E(v,J)/kT_{exc}+C`$; where $`g_J`$ is the degeneracy of the corresponding level, $`k`$ is the Boltzmann constant, and $`C`$ is a constant.
Figure Optical and Near Infrared Study of the Cepheus E outflow, a very low excitation object. shows the excitation diagrams for Cep E, South and North lobes. We plot $`N(v,J)/g_J`$ versus energy $`E(v,J)`$ of the level for each lobe using the fluxes listed in Table 2<sup>2</sup><sup>2</sup>2The excitation diagrams do not include the very weak H<sub>2</sub> (3,2) S(5) line presented in Table 2. This transition shows a bigger column density than expected, which causes great dispersion in the estimation of the excitation temperature, so we have excluded it from our analysis.. In each panel, the transitions between the (1,0) vibrational levels are plotted with solid squares, the ones between the (2,1) levels with triangles and the (3,2) levels with solid circles. The excitation temperatures, T<sub>exc</sub> (calculated from linear fits), are $`2340\pm 100`$ for the South lobe and $`2260\pm 110`$ for the North lobe.
In the case of a purely thermal population caused by a shock, it is expected that a single smooth curve should fit all of the transitions in the excitation diagrams , meaning that a single temperature describes the excitation. On the other hand, in the case of fluorescent H<sub>2</sub> emission, the points within vibrational levels are expected to be aligned on separate curves , resulting in different temperatures for different vibrational levels. Intermediate cases in which there is a combination of both excitation mechanisms are also possible (Fernandes & Brand 1995; Fernandes, Brand & Burton 1995).
The derived excitation temperatures, T<sub>exc</sub>, for both lobes of Cepheus E are consistent with the T<sub>exc</sub> measured in HH objects (T$`{}_{exc}{}^{}2000`$ K, Gredel 1994), whereas the excitation temperatures observed in collisionally plus fluorescent excited objects have T$`{}_{exc}{}^{}3000`$ values (Fernandes, Brand & Burton 1997). However, in Figure Optical and Near Infrared Study of the Cepheus E outflow, a very low excitation object. there are real dispersions around the dashed lines drawn in each panel. If we fit the points from different vibrational levels separately, there are slight differences between these fits and the fit obtained considering all of the vibrational levels together.
We have computed the rotational excitation temperatures using the transitions within different vibrational levels. The resulting temperatures are listed in Table 3. In order to explore the excitation mechanisms occurring in the Cep E outflow, we have computed some interesting line ratios (with the fluxes presented in Table 2), which are also listed in Table 3. This table gives the vibrational levels (column 1), the rotational levels of the transitions used for the ratios (column 2), the rotational temperature T<sub>rot</sub> derived from these transitions (column 3), the empirical Ortho/Para ratio derived for each vibrational level (column 4), and the (1,0)/(2,1) S(1) ratio derived for each spectrum (column 5).
As we can see, for both lobes of Cepheus E the rotational excitation temperatures (Table 3) are lower than the calculated vibrational temperature T<sub>exc</sub> (see Figure Optical and Near Infrared Study of the Cepheus E outflow, a very low excitation object.). Though different from T<sub>exc</sub>, the rotational temperatures are still consistent with the temperatures observed in collisionally excited objects. We should note that the vibrational level (2,1) shows the lowest rotational temperature.
We find that the values obtained for the Ortho/Para ratio are similar for the two lobes of Cepheus E. For the Northern lobe there is a higher dispersion for this ratio between the different vibrational levels. This is a result of the dispersion in the points corresponding to the (2,1) and (3,2) levels (see top panel of Figure Optical and Near Infrared Study of the Cepheus E outflow, a very low excitation object.). This dispersion is probably a result of the fact that the spectrum of the Northern lobe is fainter. We conclude that the Ortho/Para ratios estimated for Cep E are completely consistent with the LTE value of 3, as expected for collisional excitation.
We analyzed the behavior of the (1,0)/(2,1) S(1) line ratio. Column (5) of Table 3 gives the $`8.50\pm 1.10`$ and $`9.40\pm 1.30`$ values obtained for the Northern and Southern lobes, respectively. These values are consistent with the result obtained from the (1,0)/(2,1) S(1) ratio image (see above), and slightly lower than the ones measured from (1,0)/(2,1) S(1) frames by Eislöffel et al. 1996. Also, the obtained (1,0)/(2,1) S(1) line ratio is completely consistent with the value of 10 expected for collisional excited H<sub>2</sub> lines, and does not approach the $``$ 1.7 value expected for fluorescent excitation .
A detailed analysis on the K-band spectra of Cep E has already been performed by Ladd & Hoddap (1997). They found that a C-type shock with 35 km/s can explain the K-band H<sub>2</sub> fluxes and ratios. This is consistent with the ISO Long Wavelength Spectrometer observations, which show several emission lines from H<sub>2</sub>O transitions , as it is expected from molecular C-type shocks .
In Cep E we find a similar situation as in other Herbig-Haro objects (e.g. HH 1-2; Davis et al. 1994; Noriega-Crespo & Garnavich 1994) where J-type (inferred from the optical spectra) and C-type (inferred from the IR spectral) seem to coexist spatially. The case of Cep E is particularly interesting since it appears to be a very young outflow, based on its compact size and the age of its exciting source. If so, a better interpretation of its spectral properties may require molecular time dependent shock models, which are currently under development .
### 3.2 The optical morphology
In contrast with the complex structure of the H<sub>2</sub> outflow, the optical emission consists of only one compact knot (HH 377), which approximately coincides with the brighter H<sub>2</sub> knot of the southern lobe . In Figure 4 we present an H<sub>2</sub> 1-0 S(1) contour map of Cepheus E, overlaid on grey-scale and contour representations of our optical images. Panel $`a)`$ shows grey-scale representations of the H$`\alpha `$ image and panel $`b)`$ of the \[S II\]6717,6731 image. Panels $`c)`$ and $`d)`$ show the H$`\alpha `$ and \[S II\]6717,6731 images (respectively) as contour maps, over a smaller angular area. These maps clearly show the coincidence between the optical knot and the brightest region of the southern lobe of the IR flow.
In Figure 4, we see that a star still remains in the H$`\alpha `$ continuum subtracted image (panel $`a`$), to the west of HH 377. This stellar emission could indicate either that the star has intrinsic H$`\alpha `$ emission, or that the stellar continuum has a substantially different slope than the continua of the stars used to obtain the relative scaling between the H$`\alpha `$ and the continuum frames.
The H$`\alpha `$ and \[S II\] morphologies of HH 377 are different, the knot is resolved in both images, but it is clearly brighter and more compact in the \[S II\] image. In our \[S II\] image, HH 377 has an angular size of 5<sup>′′</sup>, resulting in a diameter of 0.02 pc (at a distance of 0.75 kpc, Hodapp 1994) for the knot. We can also see that the morphology of the southern lobe of Cep E is qualitatively consistent with a bow-shock model, in which the wings have a stronger contribution of H<sub>2</sub> emission, while the head is dominated by the atomic/ionic emission.
Finally, in Figure 4 (panel $`b`$), we show the two slit positions used to obtain the spectra described in the following section.
### 3.3 Optical spectroscopic characteristics
We have obtained two spectra (intermediate and low resolution) with the N-S slit (see Figure 4), and a single (low resolution) spectrum with the E-W slit (see the discussion in section 2.2). Our spectrophotometric data gives us the optical excitation conditions of HH 377. Figure 5 shows the flux-calibrated spectrum, extracted from a 4 $`({}_{}{}^{\prime \prime })^2`$ in the E-W oriented slit (see panel $`b`$ of Figure 4). The results obtained for the relative fluxes with and without reddening correction (for our three spectra) are listed in Table 4. Column 1 gives the identification of the detected lines, column 2 lists the observed ($`F`$) and reddening-corrected ($`F_0`$) relative fluxes for the E-W slit, column 3 and 4 list the fluxes from the N-S slit at low and intermediate resolutions, respectively. The determination of the value of $`E(BV)`$ that we have used is discussed in the following section. In addition, the radial velocity estimate from these spectra is $`V_r=`$-70$`\pm 10`$ km s<sup>-1</sup>.
From the spectrum with larger wavelength coverage (see Table 4 and Figure 4), we can see that \[NI\]5200 is greater than H$`\beta `$ and that the \[SII\]6717/31 and \[OI\]6300/64 fluxes are stronger than H$`\alpha `$. No \[OIII\]5007 nor \[NII\]6548,83 emission is detected. These observed characteristics are typical of low excitation Herbig-Haro objects.
Finally, we use the \[SII\] 6731/6717 line ratio to compute an electron density $`n_e=4100cm^3`$ (assuming an electron temperature $`T_e=10^4`$ K). As discussed in the following section, this result is particularly interesting.
## 4 Discussion: Cep E as a low excitation Herbig-Haro object
As the limited wavelength coverage of our spectra does not allow us to use Miller’s (1968) method, we have determined the reddening by assuming that the intrinsic (dereddened) Balmer decrement has a H$`\alpha `$/H$`\beta `$=3, (i.e., a recombination cascade value). Using the measured H$`\alpha `$/H$`\beta `$=7.75 ratio and the standard, $`R_V=3.1`$ ISM reddening curve of Mathis (1990), we obtain $`E(BV)=0.88\pm 0.12`$ (corresponding to an $`A_V=2.72\pm 0.38`$ optical extinction). This method is of course uncertain for the case of Cepheus E (and other HH objects) since it is expected that in the case of low velocity shocks the Balmer decrement could differ substantially from the recombination value. We have used the optical total-to-selective ratio $`R_V=A(V)/E(BV)=3.1`$ which is frequently used in this cases and appears to be the most appropriate for HH objects .
In Table 5 we present line ratios observed for Cep E from the fluxes given in column 2 of Table 4. The optical spectroscopic characteristics of HH 377 (Cep E) identify it as a low excitation object. Indeed, we find that the excitation of this object appears to be anomalously low, compared to other low excitation HH objects. For example, most of the observed line ratios obtained for HH 377 are consistent with the ones observed in other low excitation objects, except for the \[SII\](6717+6731)/H$`\alpha `$ ratio. For HH 377, this ratio is a factor of $`3`$ higher than the corresponding values for the low excitation HH objects (HH 7, HH 11, HH34(jet), HH 47A, HH 111 D-J, HH 125 I and HH 128), this fact is clear in second panel top to bottom, Figure 6. The \[SII\] 6731/6717 ratio is also the highest one observed (with a value of $`0.59\pm 0.09`$). This line ratio implies an electron density $`n_e`$4100 cm<sup>-3</sup>, which is the largest $`n_e`$ measured for any previously detected, low excitation HH object (see Figure 6, bottom panel). The electron density is relatively high and comparable to the densities measured in high excitation HH objects. The low excitation in Cep E suggests, however, a low ionization fraction and therefore a higher gas density than usual.
Following Raga et al. (1996), we compare the line ratios of Cep E (HH 377) with those obtained for other HH objects. We refer the line ratios to the \[OI\]6300/H$`\alpha `$ ratio, which is different from Raga et al.(1996), who plot all of the line ratios versus the \[N I\](5198+5200)/H$`\beta `$ line ratio. We have done this because the \[N I\] lines have a very low critical density ($``$ 2900 cm<sup>-3</sup>), so that collisional quenching will occur in Cep E (while not in the other low excitation HH objects).
In Figure 6 we plot the \[NI\](5198+5200)/H$`\beta `$, \[SII\](6717+6731)/H$`\alpha `$, and \[SII\]6731/6717 line ratios (and also, the electron density $`n_e`$) as a function of the \[OI\]6300/H$`\alpha `$ ratio, indicating the values that correspond to high, intermediate and low excitation HH objects. The line ratios for Cep E are indicated with an asterisk. In this Figure we can see the anomalous line ratios, specially the \[SII\](6717+6731)/H$`\alpha `$ ratio and electron density ($`n_e`$) obtained for Cep E, in contrast to other low excitation HH objects.
Let us now compare the line ratios of HH 377 (Cep E) with shock model predictions. The plane shock models of Hartigan et al. (1994) do show high \[SII\](6717+6731)/H$`\alpha `$ ratios, in better agreement with the observations of HH 377. From this work, we list the range of line ratios obtained for models of shock velocities in the v$`{}_{s}{}^{}=20`$-30 km s<sup>-1</sup> range (labeled J4.20-30 in Table 5), and models with v$`{}_{s}{}^{}=30`$-40 km s<sup>-1</sup> (labeled J4.30-40), in both cases with a high pre-shock density n$`{}_{0}{}^{}=10^4`$ cm<sup>-3</sup> and magnetic fields between 30 and 300 $`\mu `$G. We have also listed the line ratios for models with v$`{}_{s}{}^{}=3040`$ km s<sup>-1</sup> and lower pre-shock density (n$`{}_{0}{}^{}=10^2`$ cm<sup>-3</sup>) and the same magnetic field range than the models described above.
We can see that the J4.20-30 models are the most successful at reproducing the line ratios observed in Cep E. In particular, these shock models do reproduce the observed \[SII\](6717+6731)/H$`\alpha `$ and \[S II\]6731/6717 ratios. Interestingly, these models also predict an H$`\alpha `$/H$`\beta 6`$ Balmer decrement.
This result seems to favor shock models with a very low shock velocity and high pre-shock density in order to reproduce the observed optical line ratios. One has to keep in mind that J-type shocks in molecular gas are restricted to velocities $`3050`$ km s<sup>-1</sup> (Hollenbach & McKee 1989) in order to not dissociate H2, and so at first order the simple J-type atomic/ionic plane parallel shock models are consistent with the rich H2 spectra observed in Cep E.
We have estimated a new value for the extinction to Cep E using the H$`\alpha `$/H$`\beta 6`$ predicted by these models, obtaining $`E(BV)0.24`$ (corresponding to an $`A_V0.77`$ optical extinction). However, it is important to note that for Cep E, only the brightest region of the southern lobe of the H<sub>2</sub> outflow reveals optical emission. This fact suggests that a high extinction close the outflow region could explain the observed differences between the optical and IR morphologies.
Lefloch et al. (1996) determined values for the optical extinction using observations of the continuum emission of Cepheus E outflow at 1.25 mm. They reported $`A_V=3.2,3.4`$ for the northern and southern H<sub>2</sub> lobes, respectively (on the same bright infrared knots which we analyze in this paper). These values are similar to our first $`A_V`$ estimation from the recombination cascade Balmer decrement for HH377, in the beginning of this section. The result implies a similar extinction for both infrared lobes. On the other hand, Noriega-Crespo et al. 1998 find a constant $`v`$ = 00 S(3)/0-0 S(5) ratio that also indicates the lack of a steep extinction gradient between the north and south lobes and that the extinction is mostly important around IRAS 23011+6126. This is interesting because we know that the south lobe is visible at optical wavelengths (Noriega-Crespo 1997), while the north lobe is not. It would be interesting to obtain extinction values for HH 377 through Miller’s method. This would give a realistic value for the fraction of the extinction which arises in the vicinity of the object, and would help to discriminate between and/or to constrain the different shock models.
### 4.1 The ionization fraction in HH 377
In order to estimate the ionization fraction in the southern lobe of Cep E (HH 377) we use the \[OI\]6300/H$`\alpha `$ ratio, which for recombination H$`\alpha `$ is given by
$$\frac{\mathrm{I}([\mathrm{O}\mathrm{I}]6300)}{\mathrm{I}(\mathrm{H}\alpha )}=X(\mathrm{O})\frac{\mathrm{n}(\mathrm{O}\mathrm{I})/\mathrm{n}(\mathrm{O})}{\mathrm{n}(\mathrm{H}{}_{}{}^{+})/\mathrm{n}(\mathrm{H})}\frac{q_{\mathrm{𝑒𝑥}}(6300)}{\alpha (\mathrm{H}\alpha )},$$
(1)
where $`X(\mathrm{O})`$ is the oxygen abundance, $`q_{ex}(6300)`$ is the excitation rate coefficient for \[O I\] 6300, and $`\alpha (\mathrm{H}\alpha )`$ is the effective recombination rate coefficient for H$`\alpha `$.
For low excitation HH objects we expect electron temperatures lower than 10<sup>4</sup> K (Bacciotti & Eislöffel, 1999). If we assume that T$`{}_{e}{}^{}`$ 5000 K, we obtain $`q_{ex}(6300)4\times 10^{10}`$ cm<sup>3</sup> s<sup>-1</sup>, and $`\alpha (\mathrm{H}\alpha )2\times 10^{13}`$ cm<sup>3</sup> s<sup>-1</sup>. For an oxygen abundance of $`X(\mathrm{O})=8\times 10^4`$, and the $`\mathrm{I}([\mathrm{O}\mathrm{I}]6300)/\mathrm{I}(\mathrm{H}\alpha )`$ = 2.5 ratio observed in HH 377, from eq. (1) we obtain $`[\mathrm{n}(\mathrm{O}\mathrm{I})/\mathrm{n}(\mathrm{O})]/[\mathrm{n}(\mathrm{H}{}_{}{}^{+})/\mathrm{n}(\mathrm{H})]=`$ 1.6; which implies that the gas is roughly 50$`\%`$ ionized.
On the other hand, we can also estimate the ionization fraction,$`x_e`$, from the nitrogen lines using the ratio
$$\frac{\mathrm{I}([\mathrm{N}\mathrm{I}]5200)}{\mathrm{I}([\mathrm{N}\mathrm{II}]6584)}0.13\frac{n(NI)}{n(NII)}.$$
(2)
Using the $`\mathrm{I}([\mathrm{N}\mathrm{I}]5200)/\mathrm{I}([\mathrm{N}\mathrm{II}]6584)>10`$ limit deduced from the spectrum of HH 377 (shown in Figure 5), we then obtain $`\mathrm{n}(\mathrm{N}\mathrm{I})/\mathrm{n}(\mathrm{N}\mathrm{II})<0.01`$, which implies that the ionization fraction is less than 1%. A similar value is obtained for the \[SII\]-weighted ionization fraction using the different line ratios in the diagrams shown in Hartigan et al. (1994, Figures 3 to 5), for the models J4.20-30 (see above).
From the discrepancy between this low ionization fraction and the much higher one deduced from eq. (1), we conclude that the H$`\alpha `$ line has to be collisionally excited. We therefore conclude that the low reddening for HH 377 obtained through the shock models (in which the Balmer decrement is about 6), is probably correct. If we consider the compression produced by the shock (which is approximately equal to the square of the Mach number) and the ionization fraction deduced above we can estimate the pre-shock density. The shock velocity estimated for HH 377 (see section 4.0) implicates a Mach number of $`M^210`$. Using an ionization fraction $`x_e1\%`$ and considering that $`N_e10^4`$ cm<sup>-3</sup>, we then obtain a pre-shock density $``$ 10<sup>5</sup> cm<sup>-3</sup>. Note that this total particle density results very high at the distance from the outflow source, and it is very unusual for HH objects with any excitation level.
### 4.2 The absolute fluxes in HH 377.
From the Hartigan et al. (1994) shock models with $`v=`$ 20 km s<sup>-1</sup> and pre-shock density of 10<sup>4</sup> cm<sup>-3</sup>, one obtains an H$`\alpha `$/H$`\beta 9`$ and fluxes of H$`\beta =1.0\times 10^5`$ and H$`\alpha =9.0\times 10^5`$ ergs cm<sup>-2</sup> s<sup>-1</sup>, out of the front of the shock. For a pre-shock density of 10<sup>5</sup> cm<sup>-3</sup> one would have fluxes larger by about an order of magnitude. This gives H$`\beta =1.6\times 10^5`$ and H$`\alpha =1.5\times 10^4`$ ergs cm<sup>-2</sup> s<sup>-1</sup> steradian<sup>-1</sup>, when applying the solid angle. Using an area of 1.5” x 4” (see column 2 of Table 4), we obtain that the predicted fluxes for H$`\beta `$ and H$`\alpha `$ are $`2.3\times 10^{15}`$ and $`2.1\times 10^{14}`$ergs cm<sup>-2</sup> s<sup>-1</sup>, respectively. These fluxes are about 6 times larger than the observed fluxes (see Table 4). However, the Balmer line fluxes of the Hartigan et al. (1994) models are extremely steep functions of shock speed, so a shock speed between 15 and 20 km s<sup>-1</sup> would give the correct total fluxes.
Finally, we must ask how the optical emission and the molecular hydrogen emission are related in this dense, low excitation HH object. Hartigan et al. (1996) discuss the possible morphologies. One option is a bow shock which has a J-shock nature at its tip (producing optical emission) with C-shocks or turbulent entrainment producing the $`\mathrm{H}_2`$ emission from the bow shock wings. Another is a C-shock in the ambient cloud accompanied by a J-shock at the Mach disk in the jet material. A third possibility is a J-shock with an MHD precursor which produces both optical and IR emission at the bow shock tip. Unfortunately, the high density of Cep E, the emitting regions extremely thin, so that it is difficult to use morphology to distinguish among the possibilities.
The various combinations of C- and J-shock models have many free parameters, so that it is also difficult to use spectra to distinguish among them. One interesting comparison, however, is the ratio between IR and optical luminosities. If the optical emission arises from the Mach disk and the $`\mathrm{H}_2`$ emission from the bow shock, and if the jet and ambient densities are not too drastically different, the luminosities should be comparable (e.g Hartigan 1989). The C-shock luminosity emerges mostly in the near IR lines we observe, while according to the slow J-shock models, the luminosity is around 1000 times the H$`\beta `$ luminosity, with most of the energy emerging in Ly$`\alpha `$ (cf. the 15 and 20 $`\mathrm{km}\mathrm{s}^1`$ models of Hartigan et al. 1994). Thus
$$\mathrm{F}_{\mathrm{IR}}7\times 10^{12}\mathrm{ergs}\mathrm{cm}^2\mathrm{s}^1,$$
(3)
calculated from data in Table 2. And $`\mathrm{F}(\mathrm{H}\alpha )3\times 10^{14}\mathrm{ergs}\mathrm{cm}^2\mathrm{s}^1`$ for a smaller reddening than the 0.88 (Table 3), then we can estimated
$$\mathrm{F}(\mathrm{Ly}\alpha )1000\mathrm{F}(\mathrm{H}\beta )150\mathrm{F}(\mathrm{H}\alpha )4.5\times 10^{12}\mathrm{ergs}\mathrm{cm}^1\mathrm{s}^1.$$
(4)
Thus it seems as though the J-shock is within an order of magnitude of the C-shock luminosity, so the possibility that the Mach disk makes the optical emission and the bow shock makes the IR emission is reasonable.
## 5 Conclusions
We have carried out a spectroscopic and imaging study of the molecular hydrogen and optical atomic/ionized emission in the Cepheus E outflow. Our main results are:
\- For deriving the excitation state of molecular gas in Cep E we use line ratios from our H-band and K-band spectra. We find that T<sub>exc</sub>= 2260$`\pm `$110 and 2340$`\pm `$100 K for the northern and southern lobes, respectively, which are consistent with the T<sub>exc</sub> measured in other HH objects. The (1,0)/(2,1) S(1) ratios (8.50 and 9.40), and the Ortho/Para ratios (values $`3`$) in both lobes are also consistent with the values observed in collisionally excited objects.
\- Contrasting with the complex structure of the H<sub>2</sub> outflow, the optical emission is a compact, well resolved knot (HH 377), that nearly coincides with the southern NIR lobe. The \[S II\] emission of HH 377 is clearly brighter and more compact than the H$`\alpha `$ emission and its angular size is about 0.02 pc (at a distance of 0.75 kpc). The \[S II\] and H$`\alpha `$ peak emission spatially coincide and appear offset a few arcseconds upstream from the H<sub>2</sub> peak emission.
\- Our spectroscopic optical analysis reveals that HH 377 has characteristics typical of low excitation Herbig-Haro objects. This is confirmed when comparing the relative fluxes of HH 377 with those of other HH objects using line ratio diagrams. However, HH 377 presents anomalous \[SII\](6717+6731)/H$`\alpha `$ line ratio, larger than those obtained for objects classified as low excitation HH objects. The electron density, $`n_e`$= 4100 cm<sup>-3</sup>, determined for this object from \[SII\] lines would be the highest density measured in low excitation HH objects. This value is similar to the electronic densities measured in high excitation HH objects.
\- We estimate an ionization fraction $`x_e1\%`$ for HH 377. Together with the observed \[OI\]6300/H$`\alpha `$ ratio, this result implies that the observed H$`\alpha `$ line has to be collisionally excited. This result supports the low reddening obtained for HH 377 obtained through the shock models. Using this ionization fraction, a post-shock electron density $`N_e10^4`$ cm <sup>-3</sup> and a compression of $`10`$, we obtain a pre-shock density $``$10<sup>5</sup> cm<sup>-3</sup> for HH 377. This exceptionally high pre-shock density is very unusual for HH objects.
\- From the shock model predictions and the H$`\beta `$ and H$`\alpha `$ observed fluxes, we find that a shock speed between 15 and 20 km s<sup>-1</sup> gives the correct total fluxes for HH 377. This velocity appears to be somewhat lower than the one deduced from the observed line ratios.
\- From a comparison between optical and infrared luminosities in HH 377 we find the possibility that the Mach disk produces the optical emission and the bow shock produces the IR emission.
\- We have determined a visual extinction $`A_V=2.72`$ ($`E(BV)`$=0.88) assuming a recombination cascade H$`\alpha `$/H$`\beta `$=3 Balmer decrement. If we use the H$`\alpha `$/H$`\beta 6`$ decrement predicted by the preferred shock wave models we obtain an $`A_V0.77`$ ($`E(BV)0.24`$) extinction. Interestingly, Lefloch et al. (1996) have obtained an $`A_V=3.4`$ from the mm continuum of the southern lobe of the Cep E outflow, in qualitatively good agreement with the extinction obtained assuming H$`\alpha `$/H$`\beta `$=3 (see above). This confusing situation involving the optical extinction towards HH 377 and the Balmer decrement predicted from low velocity shock models could be clarified with future observations of the blue and IR \[SII\] lines (or, alternatively, the IR \[Fe II\] lines) of this object, in order to have a model-independent determination of the extinction. The overall extinction along the Cep E flow varies drastically. One could speculate different reasons (e.g. a large inclination with respect to the plane of the sky, a dense molecular gas core surrounding the source with a rapidly decreasing density profile and/or a non-homogeneous ISM). We can not answer this question with our present data.
\- Comparing the optical line ratios observed in south lobe of Cep E to atomic/ionic plane-parallel shock models (J-type) presented in the literature (Hartigan et al. 1994), we find that low velocity (v$`{}_{s}{}^{}=20`$-30) shocks are the closest to reproduce the observations. These models have a high pre-shock density (n$`{}_{0}{}^{}=10^4`$ cm<sup>-3</sup>), and magnetic fields between 30 and 300 $`\mu `$G. These conditions (high density, low ionization fraction and strong magnetic field) are quite appropriate for the development of m͡olecular C-type shock as well. Previous comparisons of the H<sub>2</sub> near-ir spectra with molecular shocks (Ladd & Hodapp, 1997), indeed indicated a preference for C-type shocks with $`35`$ km $`\mathrm{s}^1`$. Preliminary results from ISO Long Wavelength Spectrometer (50 - 200$`\mu `$m) support also this view, given Cep E rich H<sub>2</sub>O spectra (Noriega-Crespo 2000), a characteristic signature for C-type shocks (Kaufman & Neufeld 1996ab; Noriega-Crespo et al. 2000)
\- Finally, our analysis seems to confirm that Cep E corresponds to an outflow in its earliest developing phases. Its short dynamical age of few 10<sup>3</sup> yrs, the high gas density estimated (at least 10<sup>5</sup> $`\mathrm{cm}^3`$) and the inhomogeneous nature of its extinction, suggest that the outflow is breaking through its placental molecular core.
This work was supported by DGAPA grant IN109297 and CONACYT grant 26833-E and 27546-E. S.A. & S.C. would like to thank César Briseño for obtaining some of the optical images. A.N.-C. & P.M.G thank Nichole King for her help on the observations. The FSPEC IR spectra were obtained with the help of Marcia Rieke & George Rieke at the MMTO, A.N.-C. & P.M.G are greatful to both. K.H.B. has been supported by NSF grant AST 9729096. Last but not least, we thank the referee David Devine for his comments and careful reading of our manuscript.
|
warning/0004/math-ph0004008.html
|
ar5iv
|
text
|
# Theorem 1
Krichever I.M.<sup>1</sup><sup>1</sup>1Krichever I.M., Colunbia University, New York NY 10027, USA and Landau Institute for Theoretical Physics, Kosygina str 2, 117940 Moscow Russia, e-mail krichev@math.columbia.edu. Research supported in part by the NSF Grant DMS-9800257., Novikov S.P. <sup>2</sup><sup>2</sup>2 Novikov S.P., University of Maryland, College Park MD 20742-2431 and Landau Institute for Theoretical Physics, Kosygina str 2, 117940, Moscow Russia, e-mail novikov@ipst.umd.edu. Research supported in part by the NSF Grant DMS-9704613
Holomorphic Bundles and Difference Scalar Operators:
One-point constructions
Let us consider a nonsingular algebraic curve $`\mathrm{\Gamma }`$ with marked point $`P_0=\mathrm{}`$ and local coordinate $`z=k^1,z(P_0)=0`$ like in . Let $`A=A(\mathrm{\Gamma },P_0)`$ be a ring of algebraic functions with only pole at the point $`P_0`$. Our inverse spectral data consist of the set of points $`\gamma _s,s=1,\mathrm{},lg`$, where $`l`$ is ”rank” of our construction (see below) and $`g`$–genus of the curve $`\mathrm{\Gamma }`$, parameters $`\alpha _{sj},s=1,\mathrm{},lg,j=1,\mathrm{},l1`$, and matrix-valued $`l\times l`$–functions $`\chi _n^{(0)}(k)`$ with only nonzero elements $`\chi _n^{(0)p,p+1}=1,pl1,\chi _n^{(0)lq}=a_{qn},q=1,\mathrm{},l`$. Here $`a_{qn}`$ are polinomials in the variable $`k`$ depending on $`n`$.
###### Theorem 1
For any fixed vector $`\eta _0`$, and a generic data there exists and unique the ”Baker-Akhiezer” $`l`$-vector-function $`\psi _n(P),P\mathrm{\Gamma }`$ that is meromorphic outside the point $`P_0`$ with first order poles at the points $`\gamma _s`$ such that
$$res_{\gamma _s}\psi ^{q+1}=\alpha _{sq}res_{\gamma _s}\psi ^1,s=1.\mathrm{},lg$$
In the neighborhood of the point $`P_0`$ this function $`\psi `$ has an asymptotic behavior
$$\psi =[\eta _0+\underset{p1}{}\eta _{pn}k^p]\mathrm{\Psi }_n^{(0)},$$
(1)
where $`\mathrm{\Psi }_n^{(0)}`$ is defined by the equation $`\mathrm{\Psi }_{n+1}^{(0)}=\chi _n^{(0)}\mathrm{\Psi }_n^{(0)}`$.
###### Theorem 2
Let matrix $`\chi _n^{(0)}`$ be such that $`a_{jn}=k+v_{jn}^{(0)}`$, all other elements $`a_q=v_{qn}^{(0)},qj`$ do not depend on $`k`$, and let $`\eta _0`$ be a vector with the cooordinates $`\eta _0^i=\delta ^{ij}`$. Then for every function $`fA(\mathrm{\Gamma },P_0)`$ with pole of order $`\tau `$ there exists a unique difference operator $`L_f`$
$$L_f=\underset{M}{\overset{+N}{}}u_{pn}T^p$$
where $`T\psi _n=\psi _{n+1},M=\tau (j1),N=\tau (lj+1)`$ such that the Baker-Akhiezer vector function defined by our data satisfies the equation
$$L_f\psi =f\psi .$$
Remark. These statements are descrete analogs of the results obtained in which give a construction of higher rank commuting differential operators and correspodning solutions of the KP equation. Let us remind that $`(\alpha ,\gamma )`$ is exactly the set of ”Tyurin Parameters” characterizing the stable framed holomorphic $`l`$–bundles $``$ over $`\mathrm{\Gamma }`$ where $`c_1(det)=lg`$. All previously known constructions of the commuting difference operators were of the rank 1. They always required exactly two infinite marked points $`\mathrm{}_\pm \mathrm{\Gamma }`$. Let us point out also that in our one-point rank $`l>1`$ construction presented here, the symmetric operators $`M=N`$ can be obtained for the even rank $`l=2(j1)`$ only.
Following the idea of multiparametric Baker-Akhiezer vector-function can be defined through the same inverse data $`\mathrm{\Gamma },P_0,z=k^1,\gamma ,\alpha ,\chi ^{(0)}`$, but for every new time variable $`t_p`$ we have to use corresponding matrix $`M_n^{(0p)}(t)`$, where $`t=(t_1,t_2,\mathrm{}),p=1,2,\mathrm{}`$. The initial matrix $`\mathrm{\Psi }^{(0)}`$ can be constructed using the integrable system (hierarhy):
$$\mathrm{\Psi }_{n+1}^{(0)}=\chi _n^{(0)}\mathrm{\Psi }_n^{(0)},\mathrm{\Psi }_{t_p}^{(0)}=M_n^{(0p)}\mathrm{\Psi }^{(0)}$$
For example let us take matrix functions $`M_n^{0+}`$ and $`M_n^0`$ of the form
$$M_n^{0,1+}=\left(\begin{array}{ccccc}w_{1n}& 1& 0& ..& ..\\ 0& w_{2n}& 1& ..& ..\\ ..& ..& ..& ..& ..\\ 0& 0& \mathrm{}& w_{l1,n}& 1\\ a_{1n}& a_{2n}& \mathrm{}& a_{l1,n}& a_{ln}+w_{ln}\end{array}\right)$$
$$M_n^{0,1}=\left(\begin{array}{ccccc}c_{1n}b_{2n}& c_{1n}b_{3n}& ..& c_{1n}b_{ln}& c_{1n}\\ c_{2n}& 0& 0& ..& ..\\ ..& ..& ..& ..& ..\\ 0& 0& c_{l1,n}& 0& 0\\ 0& 0& \mathrm{}& c_{ln}& 0\end{array}\right)$$
where $`c_{qn},w_{qn}`$ do not depend on $`k`$, the functions $`a_{qn}`$ are the same as in theorem 2, and $`b_{qn}=a_{1,n1}^1a_{qn}`$.
###### Theorem 3
For any $`l2`$ we can choose matrices $`M^{(0p)},p1`$, such that the Baker-Akhiezer vector-function $`\psi _n(t),t=(t_1^+,t_1^{},t_2,\mathrm{})`$ corresponding to any generic data $`\mathrm{\Gamma },P_0,z=k^1,\gamma ,\alpha ,\chi ^{(0)}`$ defines with the help of the formula
$$\varphi _n(t)=y_n+\mathrm{ln}(1+\eta _{1n}^{j1})$$
a solution of 2D Toda lattice hierarchy. Here $`y_n`$ are such that $`c_n=\mathrm{exp}(y_ny_{n1})`$, and $`\eta _{1n}^{j1}`$ is the $`(j1)`$-th component of the vector $`\eta _{1n}`$ in the expansion (1).
Example. Let $`g=1,l=2,a_1=c_{n+1}^{(0)},a_2=kv_{n+1}^{(0)}`$, the data $`(\mathrm{\Gamma },k^1=z,P_0=0,\gamma _1,\gamma _2,\alpha _1,\alpha _2)`$ and the Weierstrass function $`\mathrm{}(z)k^2`$ are given. For the Baker-Akhiezer matrix $`\widehat{\mathrm{\Psi }}_n`$ whose rows are $`\psi _n,\psi _{n+1}`$ we have $`\widehat{\mathrm{\Psi }}_{n+1}=\chi _n\widehat{\mathrm{\Psi }}_n`$ where the rows of $`\chi `$ are following:
$$\chi _n^1=(0,1),\chi _n^2=(c_{n+1},kv_{n+1})+O(k^1)$$
Poles of the matrix-function $`\chi _n`$ are located in the points $`\gamma _{sn}`$ where $`\gamma _{s0}=\gamma _s,s=1,2`$. Zeroes of the function $`det\chi _n`$ are located at the points $`\gamma _{s,n+1}`$. We have following relations
$$\alpha _{sn}res_{\gamma _{sn}}\chi ^{i1}=res_{\gamma _{sn}}\chi ^{i2},i=1,2$$
and $`\alpha _{s,n+1}=\chi ^{22}(\gamma _{s,n+1})`$. Our curve $`\mathrm{\Gamma }`$ is realized as a complex plane with the coordinate $`z`$ factorized by the lattice of periods. The sum $`\gamma _{2n}+\gamma _{1n}=c`$ does not depend on $`n`$. For every function $`fA(\mathrm{\Gamma },0)`$ we can compute the commuting operators $`L_f`$ given by the theorem 2. For the Weierstrass function $`f=\lambda =\mathrm{}(z)`$ and $`c=0`$ we have symmetrizable fourth order operator $`L_\lambda `$ given by the formulas:
$$\lambda \psi _n=L_\lambda \psi _n=[(L_2)^2+u_n]\psi _n,L_2\psi _n=\psi _{n+1}+v_n\psi _n+c_n\psi _{n1}$$
$$u_n=[\mathrm{}(\gamma _{n1})+\mathrm{}(\gamma _{n2})]+b_{n1}+b_{n2},\mathrm{}(z)=\zeta ^{}(z)$$
$$b_n=2\mathrm{}^{}(\gamma _n)[\mathrm{}((\gamma _{n+1}+\gamma _n)\mathrm{}(\gamma _{n+1}\gamma _n)][\mathrm{}^{}(\gamma _{n+1}+\gamma _n)\mathrm{}^{}(\gamma _{n+1}\gamma _n)]^1$$
$$c_n=(\alpha _{1n}\alpha _{2n})^1[\zeta (\gamma _{n+1}\gamma _n)\zeta (\gamma _{n+1}+\gamma _n)+2\zeta (\gamma _n)]$$
Here we have $`\gamma _n=\gamma _{1n}`$ by definition, $`\gamma _n`$ and $`v_n`$ are the arbitrary functions. The functions $`\alpha _{in},i=1,2`$ can be found from the formulas:
$$\alpha _{i,n+1}=v_{n+1}+(1)^{i1}\zeta (\gamma _{n+1})+(1)^{i1}\alpha _{1n}(\alpha _{1n}\alpha _{2n})^1\zeta (\gamma _{n+1}\gamma _n)+$$
$$+(1)^{i1}\alpha _{2n}(\alpha _{1n}\alpha _{2n})^1\zeta (\gamma _{n+1}+\gamma _n)$$
Consider now time dynamics in the variable $`t=t_1^+`$ such that $`M_n^{(0,1+)}(t)=\chi _n^{(0)}(t)+diag(v_n^{(0)},v_{n+1}^{(0)})`$. For the Baker-Akhiezer matrix $`\widehat{\mathrm{\Psi }}_n`$ we have
$$_t\widehat{\mathrm{\Psi }}_n=M_n\widehat{\mathrm{\Psi }}_n,\widehat{\mathrm{\Psi }}_{n+1}=\chi _n\widehat{\mathrm{\Psi }}_n$$
$$M_n=\chi _n+diag(v_n,v_{n+1})+O(k^1)$$
¿From the compatibility conditions in the variables $`n,t`$ we are coming to the nonlinear system for $`c_n(t),v_n(t))`$:
$$_tc_{n+1}=c_{n+1}(v_{n+1}v_m),_tv_{n+1}=c_{n+2}c_{n+1}+\kappa _{n+1}\kappa _n$$
Here we have $`\chi _n^{22}=kv_n+\kappa _nk^1+O(k^2)`$. This system is a difference analog of the so-called ”Krichever-Novikov” system . It parametrizes some family of the ”rank 2” solutions to the 2D Toda Lattice instead of KP. No problem to compute effectively the coefficient $`\kappa `$ using dynamics of the Tyurin parameters.
Remark. In our next note we present two-(and more)–point constructions of the rank $`l2`$ containing some very interesting new phenomena.
|
warning/0004/astro-ph0004177.html
|
ar5iv
|
text
|
# The Chandra X-Ray Observatory’s Radiation Environment and the AP-8/AE-8 Model
## 1 INTRODUCTION
Just past midnight on July 23, 1999, the space shuttle Columbia lifted-off from Cape Canaveral, Florida. In its payload bay lay the Chandra X-ray Observatory (CXO), the primary cargo of the STS-93 mission. Just under 8 hours after launch, Chandra was deployed from the space shuttle. However, it would be nearly two weeks later, after an Inertial Upper Stage booster and several “burns” by its own propulsion system, that Chandra would reach its final orbit. The CXO is now the third of NASA’s “great observatories” in space.
The CXO’s operational orbit has an apogee of approximately 140,000 km and a perigee of nearly 10,000 km, with a $`28.5^{}`$ initial inclination. The CXO’s highly elliptical orbit, with an orbital period of approximately 2.65 days, results in high efficiency for observing. Moreover, the fraction of the sky occulted by the Earth is small over most of the orbital period, as is the fraction of time when the detector backgrounds are high as the flight system dips into the Earth’s radiation belts. Consequently, approximately 85% of Chandra’s orbit is available for observing. In fact, uninterrupted observations lasting as long as 2.3 days are possible.
The CXO carries two focal plane science instruments: the High Resolution Camera (HRC) and the Advanced CCD Imaging Spectrometer (ACIS). The Observatory also possesses two objective transmission gratings: a Low Energy Transmission Grating (LETG) that is to be primarily used with the HRC, and the High Energy Transmission Grating (HETG) that is to be primarily used with the ACIS. In addition to these instruments, Chandra also carries a radiation monitor – the Electron, Proton, Helium Instrument (EPHIN) particle detector.
In order to attain this high level of observing efficiency, a robust radiation environment model is required so that times of high radiation are known a priori when producing a weekly sequence of CXO observations. The CXO orbit encounters higher radiation levels as the spacecraft approaches perigee. Routine science observing will cease whenever the scheduling system has indicated that the maximum radiation levels are excessive.
These times of high radiation are required not only so that observations do not take place but also so as to not cause significant radiation damage to the CXO’s focal plane instruments over the expected on-orbit design life. To that end, the Chandra X-ray Center (CXO) currently uses the National Space Science Data Center’s (NSSDC) “near Earth” AP-8/AE-8 radiation belt model to predict the start and end times of passage through the radiation belts.
In this paper, we provide a brief synopsis of the CXO and its instruments in Section 2. In Section 3, we present an overview of the Earth’s magnetosphere and the EPHIN detector. The AP-8/AE-8 model is described in Section 4. The following section, Section 5, presents our evidence that our implementation of the AP-8/AE-8 model does not always give sufficiently accurate predictions of the start and end times of transit of the Earth’s Van Allen belts. In that same section, we present an explanation for why the dipole implementation of the AP-8/AE-8 model gives inaccurate start and end times for radiation belt transit 75% of the time (for low energy electrons). Lastly, in Section 6, we provide a brief summary of our current operating procedure and the on-going work being done by the CXC and NASA’s Marshall Space Flight Center’s Radiation Environment working group in identifying a new radiation model to be used for scheduling purposes.
## 2 CHANDRA X-RAY OBSERVATORY’S FOCAL PLANE INSTRUMENTS
The observatory consists of a spacecraft system and a telescope/science-instrument payload. The spacecraft system provides mechanical controls, thermal control, electric power, communication/command/data management, and pointing and aspect determination. This section, however, only briefly describes the two focal plane instruments on-board the CXO, the HRC and the ACIS, since the main emphasis in this paper is the Observatory’s radiation environment and not the instruments per se. Nevertheless, the AXAF Observatory Guide and the AXAF Science Instrument Notebook contain a wealth of information about the CXO and its instruments. More in depth discussions of the Chandra mission, spacecraft, other instruments and subsystems are presented elsewhere. <sup>,</sup> <sup>,</sup> <sup>,</sup>
### 2.1 High Resolution Camera (HRC)
The High Resolution Camera, HRC, is a microchannel plate (MCP) instrument. It is comprised of two detector elements, a $``$ 100 mm square optimized for imaging (HRC-I) and a $``$ 20 x 300 mm rectangular device optimized for the Low Energy Transmission Grating (LETG) Spectrometer readout (HRC-S).
The HRC has the highest spatial resolution imaging on Chandra – $``$ 0.5 arcsec (FWHM) – matching the High Resolution Mirror Assembly (HRMA) point spread function most closely. The HRC energy range extends to low energies, where the HRMA effective area is the greatest. HRC-I has a large field of view (31 arcmin on a side) and is useful for imaging extended objects such as galaxies, supernova remnants, and clusters of galaxies as well as resolving sources in a crowded field. The HRC has good time resolution (16 $`\mu `$sec), valuable for the analysis of bursts, pulsars, and other time-variable phenomena and limited energy discrimination, E/$`\mathrm{\Delta }`$E $``$ 1 ($`<`$ 1 keV). The HRC-S is used primarily for readout of the low-energy grating, LETG, for which its large format with many pixels gives high spectral resolution ($`>`$ 1000, 40-60 $`\AA `$) and wide spectral coverage (3 - 160 $`\AA `$).
### 2.2 Advanced CCD Imaging Spectrometer (ACIS)
ACIS is the Advanced CCD Imaging Spectrometer. It is comprised of two arrays of CCDs, one optimized for imaging wide fields (2x2 chip array; ACIS-I), the other optimized for grating spectroscopy and for imaging narrow fields (1x6 chip array; ACIS-S). Each array is shaped to follow the relevant focal surface. In conjunction with the HRMA, the ACIS imaging array provides simultaneous time-resolved imaging and spectroscopy in the energy range $``$ 0.5 - 10.0 keV. When used in conjunction with the High Energy Transmission Gratings (HETG), the ACIS spectroscopic array will acquire high resolution (up to E/$`\mathrm{\Delta }`$E = 1000) spectra of point sources. The CCDs have an intrinsic energy resolution (E/$`\mathrm{\Delta }`$E) which varies from $``$5 to $``$50 across the energy range.
ACIS employs two varieties of CCD chips. Most of the chips are “front-side” (or FI) illuminated. That is, the front-side gate structures are facing the incident X-ray beam from the HRMA. Two of the 10 chips (S1 and S3) have had treatments applied to the back-sides of the chips, removing the insensitive, undepleted, bulk silicon material and leaving only the photo-sensitive depletion region exposed. These “back-side”, or BI chips, are deployed with the back side facing the HRMA. BI chips have a substantial improvement in low-energy quantum efficiency as compared to the FI chips because no X-rays are lost to the insensitive gate structures but suffer from poorer charge transfer inefficiency, poorer spectral resolution, and poorer calibration accuracy. In addition, early analysis from on-orbit data indicate that the BIs are more susceptible to “background flares”, which may compromise a measurement, than are FIs. These background flares (i.e. rapid increases in detector background) are thought to be a consequence of Chandra’s radiation environment.
## 3 THE EARTH’S MAGNETOSPHERE AND THE EPHIN DETECTOR
Before describing the EPHIN detector, we will first provide a brief overview of the Earth’s magnetosphere.
### 3.1 The Earth’s Magnetosphere
The solar wind, a streaming ionized plasma, flows at all times and in all directions from the Sun. When it encounters a magnetized obstacle, the Earth for example, a magnetosphere is formed (see Figure 1). The magnetosphere of a planet is defined as the region where the particle motion is determined by the magnetic field of the planet. The magnetopause is the boundary layer which separates this region from the solar wind plasma. At the bow shock, the solar wind, when sensing the obstacle prior to reaching the magnetopause, undergoes an abrupt transition from supersonic flow to subsonic flow. This shock allows the wind to be slowed, heated, and deflected around the planet in the magnetosheath. The polar cusps of clefts are singular points at northern and southern polar latitudes where the magnetic field is zero. Only here can solar wind particles directly reach the top of the atmosphere. Field lines in the neighbourhood are either closed toward the dayside or open toward the nightside, where the magnetotail is formed. The dayside extension of the magnetosphere varies between 4.5 and 20 $`R_{Earth}`$ depending on the solar wind pressure, whereas the nightside extension reaches several hundred $`R_{Earth}`$, much farther than the orbit of the Moon. The plasmasphere is a region of high density ($``$ $`10^3`$ $`cm^3`$), cold ($``$ 1 eV) plasma, an extension of the ionosphere to altitudes up to 3-4 $`R_{Earth}`$.
The region in which science observations must not occur is in the Van Allen belts. The radiation belts, or the Van Allen belts, consist of particles in orbits that circle the Earth from about 1,000 km above the surface to a geocentric distance of $``$ 6 $`R_{Earth}`$. Contrary to particles in the plasmasphere, these particles are trapped at high energies. These particles enter the radiation belts through a variety of means, including radial diffusion from more distant regions with accompanying acceleration, and the decay of neutrons from the sputtering of the atmosphere by galactic cosmic rays. Whereas energetic protons form a single belt, as illustrated in the top panel of Figure 2, electrons form two belts – an inner and an outer belt (as illustrated in the bottom panel). A co-ordinate system with L (L designates the magnetic-drift shell and is equal to the distance in $`R_{Earth}`$ from the center of the Earth to the point where the field line crosses the equator) and B, the magnetic field, organizes the radiation belt data quite well.
### 3.2 The EPHIN Detector
The natural space environment can cause a range of problems for a spacecraft and may even compromise its mission. Environmental factors include the radiation belts, solar energetic particles, cosmic rays, plasma, gases, and micrometeorites. In this paper, we will only address the radiation belts.
The CXO radiation environment is monitored by the EPHIN detector (although the HRC anti-coincidence shield may also be used to some degree). The EPHIN was designed and manufactured as a scientific instrument for the Solar and Heliospheric Observatory and the Chandra X-ray Observatory by the University of Kiel. A detailed description of the EPHIN instrument can be found in Müller-Mellin et al. (1995) and Sierks (1997).
Briefly, the EPHIN consists of an array of 5 silicon detectors with anti-coincidence to measure the energy of electrons in the range 150 keV - 10 MeV, and hydrogen/helium isotopes in the energy range 5 - 53 MeV/nucleon. Please see Table 1 for a more precise overview of EPHIN’s scientific data channels. As Figure 3 illustrates, the field of view of EPHIN is $``$ 83 degrees. The stack of 5 silicon detectors operates in a multi-dE/dx vs. E mode and it is surrounded by an anti-coincidence shield (Figure 3). Two passivated ion-implanted silicon detectors (PIPS) A and B define the 83 degree field of view with a geometric factor of 5.1 $`cm^2`$ sr. The silicon detectors are divided into 6 segments. This coarse position sensing permits sufficient correction for path length variations needed to resolve isotopes of hydrogen and helium. The directional information cannot be used for anisotropy measurements as particles from different directions are summed when they have the same degree of oblique incidence. Another important advantage of this sector paradigm is the capability to implement a commandable or self-adaptive geometric factor . This permits measurements of fluxes as high as 2 x $`10^5`$ counts/($`cm^2`$ s sr) without significant dead time losses. A third benefit is the reduction in capacitance at the input to the charge sensitive pre-amplifiers, resulting in low-noise performance.
Lastly, the lithium-drifted silicon detectors (Si(Li)) C, D, and E stop electrons up to 10 MeV and hydrogen and helium nuclei up to 53 MeV/n. The ion-implanted detector F will allow particles stopping in the telescope to be distinguished from penetrating particles. The fast plastic scintillation detector G, viewed by a 1-inch photomultiplier and used in anti-coincidence, is indispensible for accurate electron measurements. This whole stack of detectors is mounted in an aluminum housing, the aperature being covered by a 2 $`\mu `$m thin titanium foil for light tightness. A second foil in the viewing cone is made of 76 $`\mu `$m aluminized Kapton (original specifications in Figure 3 called for 8 $`\mu `$m) for thermal control.
## 4 AP/AE TRAPPED PARTICLE RADIATION MODEL
The CXO’s orbital parameters were designed to minimise the time spent in the Van Allen radiation belts. Energetic particles contribute to the background rate and can damage the CCD detectors over time. Hence, the apogee is $``$ 140,000 km. Now, the time spent above 60,000 km is a good indicator of the time spent outside the radiation belts, however in order to maximize orbital efficiency, a robust radiation environment model is required; the AP-8/AE-8 radiation belt model is the one the CXC has implemented to accomplish this task. The primary outputs from the AE-8/AP-8 model are the spatial fluxes of trapped electrons and protons in the near Earth environment.
These maps contain omnidirectional, integral electron (AE maps) and proton (AP maps) fluxes in the energy range 0.04 MeV to 7 MeV for electrons and 0.1 MeV to 400 MeV for protons in the Earth’s radiation belt (L = 1.2 to 11 for electrons, L = 1.17 to 7 for protons). The fluxes are stored as functions of energy, L-value, and B/$`B_0`$ (magnetic field strength normalized to its equatorial value on the field line). These maps are based on data from more than 20 satellites that operated from the early sixties to the mid-seventies. Indeed, the data contained within these flux maps span 14 years (1966 to 1980). The AE-8/AP-8 model is the latest edition in a series of updates starting with AE-1 and AP-1 in 1966.
The different previous electron models can be characterised as inner (L = 1.2-3) or outer (L = 3-11) zone models and as models for solar cycle maximum or minimum conditions. However, AE-8 is the first model that covers the whole L range and both solar cycle extrema. The AP maps differ in energy range and solar cycle phase. AP-8 is the first model for the whole energy range and both solar cycle extrema.
However, one significant shortcoming of the AP-8/AE-8 model is that none of the flux maps consider time variations beyond the solar cycle minimum/maximum distinction. A study carried out by the NSSDC found that for the inner zone electrons, the two dominant time effects are caused by magnetic storms and the solar cycle. In particular, magnetic storms strongly affect electrons with energies higher than 0.7 MeV at higher L-shells, whereas the solar cycle effect is most significant for electrons with energies below 0.7 MeV. It is important to note that magnetic storm effects are still not yet included in any of the AE maps. It is thought that the largest errors occur where steep gradients in spatial and spectral distribution exist and where time variations are not well understood. Therefore, it is clear that one has to be careful in extrapolating these models to later epochs.
The electron AE-8 and proton AP-8 flux maps for solar maximum and minimum are available through the NSSDC as part of its “RADBELT 1988” software program (see http://nssdc.gsfc.nasa.gov/). The software provides omnidirectional, integral (or differential) electron or proton fluxes in the Earth’s radiation belt by using an interpolation procedure that uses the AE-8 and AP-8 trapped particle flux maps. It is the output from this software program that we compare against EPHIN data in the next section.
## 5 AP-8/AE-8 MODEL VS. EPHIN DATA
In order to make a meaningful comparison of the AP-8/AE-8 model with the EPHIN data, the correct information must be provided to the radiation belt software program. Since EPHIN possesses 4 electron channels and 4 proton channels (see Table 1), the NSSDC software routine was queried to find the start times when various specified thresholds would be exceeded, and the end times when flux predictions would be below the same thresholds for each EPHIN channel. These time spans are then, effectively, a proxy for radiation belt transit (as measured by the EPHIN).
Now, the offline scheduling system (OFLS), which incorporates the radiation belt software that the CXC uses to produce transit predictions, is not designed to search for specific ranges of energies for flux excursions. That is, it can only do single energies with multiple flux values; a single time is generated at each entry and exit where the flux and energy conditions are met. Below is a sample query that was used to determine start and end times of radiation belt transit.
FOR EPHIN ELECTRONS:
E150: Flux = 1 cts/s/$`cm^2`$/sr Energy $``$ 0.25 MeV
E300: Flux = 1 cts/s/$`cm^2`$/sr Energy $``$ 0.67 MeV
E1300: Flux = 1 cts/s/$`cm^2`$/sr Energy $``$ 2.64 MeV
E3000: Flux = 1 cts/s/$`cm^2`$/sr Energy $``$ 4.80 MeV
FOR EPHIN PROTONS:
P4: Flux = 4.93 cts/s/$`cm^2`$/sr Energy $``$ 5.0 MeV
P8: Flux = 1 cts/s/$`cm^2`$/sr Energy $``$ 8.3 MeV
P25: Flux = 1 cts/s/$`cm^2`$/sr Energy $``$ 25.0 MeV
P41: Flux = 1 cts/s/$`cm^2`$/sr Energy $``$ 41.0 MeV
The time spans (i.e., entries and exits) for which the above conditions were met have been determined since Day 200 of 1999 to the middle of February, 2000. The time spans, which corresponds to an EPHIN proton or electron channel, have been superimposed on the EPHIN data. Generally, the agreement between the radiation belt model and the EPHIN data is quite poor. For instance, of the 73 radiation belt transits seen in the EPHIN E150 channel, 55 were poorly timed ($``$ 75%). That is, of the 73 radiation belt transits investigated, nearly 75% showed EPHIN exceedances of the flux criteria specified above before the expected exceedance from the AE-8/AP-8 model. Similar numbers for the E300, E1300, and E3000 channels are 48 (66%), 20 (27%) and 20 (27%), respectively.
Three of the most extreme examples of these radiation belt mis-timing are presented in Figures 4, 5, and 6. Although similar data exists for the EPHIN proton channels, they are not included in this paper since it appears that in a high electron flux environment, electrons can mimic a proton signal in the low energy EPHIN proton channels. More work is needed to fully understand and solve this problem before that data is presented.
These plots also demonstrate the significant variability in radiation belt transits, however, one has to be very careful in drawing conclusions about temporal variability during radiation belt transits based on these plots alone since the EPHIN detector saturates during perigee transit. The effect of the EPHIN anti-coincidence counter, when in saturation, will distort the electron fluxes and time profile in the innermost part of the radiation belts, approximately when E150 rates exceed $`10^5counts/s/cm^2/sr`$. The indicated intensities in Figures 4, 5, and 6 during these intervals, should be regarded as lower limits only. An internal CXC memo, however, used EPHIN’s leakage currents as a probe to understand Chandra’s radiation environment and also found significant variablity in radiation belt intensity and duration. In the following table, we note the average value of the discrepancy (between model and data) for each EPHIN channel for the three perigee transits presented in Figures 4, 5, and 6.
Hence, we observe that the predictive value of the AP-8/AE-8 model is adequate, though certainly not optimal. In particular, the mis-timing of the softer species is seen to be worse than that for their harder counterparts. Since the degradation of the ACIS FI CCDs is believed to be due to protons in the range $``$ 100 kev - $``$ 400 keV, it is important for us to accurately predict the location of these relatively low-energy species. Due to their low energy, however, these species are also very dynamic, and can rapidly change their spatial extent subject to the day-to-day solar conditions. In contrast, the implementation of AP-8/AE-8 model used is a static one and can only take into account solar MAX vs. MIN conditions, and we believe this is the major source of the scheduling error of the radiation belts we witness, especially for low-energy species. Other models, such as a 3-D version of the Magnetospheric Specification and Forecast Model currently under development will be better able to take into account solar variations on a day-to-day (or week-to-week) basis and will thus increase Chandra’s observing efficiency by reducing the necessary “padding” of the scheduled radiation belts which is currently implemented (discussed in Section 6) as a CXC mission planning policy.
## 6 CXC OPERATING PROCEDURE AND FUTURE WORK
Although the ACIS hardware is fairly robust with respect to radiation, repeated exposure to high levels of radiation will gradually degrade the hardware (see ACIS Flight Software User’s Guide). In fact, shortly after opening the telescope door to celestial sources for the first time, it was found that the FI ACIS CCDs had suffered radiation damage (http://chandra.harvard.edu). This radiation damage is believed to have been caused by forward scattering of protons with energies between 100 keV and 400 keV, associated with the radiation belts, from Chandra’s mirrors during passage through the Van Allen belts. Since this discovery, a number of policy changes have been implemented that will hopefully prevent further radiation damage from occuring.
Because of this mis-timing of the radiation belts by our scheduling software, we have had to be more conservative in identifying times we shut down the ACIS for the radiation belt transit and times we turn on the ACIS post-perigee transit. From Table 2, it can be seen that if a 10-11 ks “pad” was added to both sides of the predicted radiation belt times, one would get better agreement with the EPHIN data. Indeed, a more thorough analysis of Chandra’s radiation history to date yielded the same result. Therefore, it was decided to pad the AE-8 predictions of radiation belt entry and exit by 13 ks. This is because the AE-8 radiation belt transit time span is longer than the AP-8 time span since the proton belt lies below the outer electron belt (see Figure 2). Thus, this choice leads to the more conservative approach.
Additionally, since it is believed to be protons focused on the ACIS as it sat in the focal plane during a series of radiation belt transits early in the mission that has caused most of the radiation damage (and not this mis-timing of the “wings” of the radiation belt itself), it was also decided to place the HRC in the focal plane (i.e., keep ACIS out from the focal plane) during subsequent radiation belt transits and to have it partially close its door to protect its instrumentation. This translation of the science instruments occurs at the “modified” start and end times of radiation belt transit.
To protect ACIS during the science operations portion of the orbit, new threshold levels were uplinked to the CXO. Now, EPHIN data is monitored by Chandra’s on-board computer (OBC) which will activate commands to safe the focal plane instruments during periods of high radiation (e.g. a solar flare or coronal mass ejection). All of these new policy changes and our monitoring scheme will be discussed in more depth in a forthcoming paper.
However, clearly all of the above policies are “short-term” solutions. What is really needed is a more robust, higher fidelity radiation belt model that provides reliable start and end times of radiation belt transit. To that end, work is currently in progress to produce an empirical model of the Earth’s radiation belt that will provide the CXC with reliable perigee transit information.
Lastly, because of the present radiation damage that has been sustained by our FI ACIS CCDs, our radiation monitor, EPHIN, has become of paramount importance in preserving the health and safety of the CXO. Thus, work has been on-going to further analyse its data, to better understand the saturation issues with perigee transit, and to develop a procedure that is able to account for electron contamination of the proton data. Clearly, more work is needed so that the outstanding science being returned by the Chandra X-ray Observatory can continue unattenuated for many more years to come.
## ACKNOWLEDGMENTS
We are grateful to many people for their support, encouragement, fruitful discussions, suggestions, and data analysis. In particular, we would like to single out Stephen O’Dell and the entire CXC/Marshall Space Flight Center Radiation Environment team, Robert Cameron, Michael Juda, Richard Edgar, Dan Shropshire, and Dan Schwartz. The authors acknowledge support for this research from NASA contract NAS8-39073.
|
warning/0004/cond-mat0004146.html
|
ar5iv
|
text
|
# Where is the Luttinger liquid in one dimensional semiconductor quantum wire structures?
\[
## Abstract
We present the theoretical basis for analyzing resonant Raman scattering experiments in one-dimensional systems described by the Luttinger liquid fixed point. We make experimentally testable predictions for distinguishing Luttinger liquids from Fermi liquid and argue that presently available quantum wire systems are not in the regime where Luttinger liquid effects are important.
\]
It is theoretically well-established \[1-3\] that a one dimensional interacting electron system (1DES), is not a Fermi liquid (FL). Unlike a Fermi liquid, the interacting 1DES has neither sharp fermionic quasiparticle excitations nor a discontinuity in the electron momentum distribution function. The elementary excitations are charge $`e`$, spin zero bosons and spin 1/2 charge 0 ’semions’ (fractional statistics objects), and the fermion is a composite of these. Interacting 1DES have been generically termed Luttinger Liquids (LL) and have been the subject of extensive theoretical study over the last 40 years and particularly over the last decade. Despite the intense theoretical interest, there have been few convincing experimental demonstrations of the predicted LL behavior in real 1DES. The power-law density of states observed in tunneling into edges of quantized hall systems have been interpreted in terms of the theoretically expected ’chiral Luttinger liquid’ behavior of edge states. The origin of the differences between the observed and expected exponents is presently an area of active inquiry. Photoemission experiments on Mott insulating oxides have been interpreted in terms of the ’holon’ and ’spinon’ excitations of a charged Luttinger liquid .
A 1DES which is of particular interest both for fundamental physics and for technology is the system formed in GaAs-based semiconductor quantum wire (QWR) structures. Modern materials growth and fabrication techniques have produced nearly ideal 1DES in which the electron may move freely only along the length of the wire. The transverse motion is quantized with the quantum 1D subbands separated by several meV. It is possible to have low enough carrier densities so that at low temperatures only the lowest 1D subband is occupied by electrons. Such GaAs QWR based 1DES should be ideal systems for the study of interacting electrons in one dimension because they are free from complications arising from band structure, lattice effects, and crossovers to three dimensional behavior which often make interpretations of experimental data difficult in more traditional 1DES based on organic compounds.
It is surprising, therefore, that no definitive LL behavior has been reported in GaAs QWR systems, and in fact the 1D Fermi gas/liquid model seems to ”work” operationally very well in describing and explaining the observed 1DES experimental properties in GaAs QWR . Part of the reason for the apparent absence of the expected LL behavior is undoubtedly the fact that in weakly interacting 1DES, at finite temperatures and in the presence of impurity scattering, the actual quantitative difference between a LL and a FL is not large , although the qualitative conceptual difference between the two is huge. A more fundamental issue is that the differences between a Luttinger Liquid and a Fermi liquid are most obvious in the one-electron spectrum, while the experimental probes which may most conveniently be applied to the QWR structures produce ’particle-hole pairs’. The differences in particle-hole pair properties between Luttinger Liquid and Fermi Liquid systems is much less pronounced than are the differences in the one electron spectrum. This perhaps accounts for the fact that one of the most important probes of QWR structures, resonant inelastic light scattering or Raman scattering spectroscopy (RRS) , has not yet observed any definitive indications of LL behavior in these systems.
In RRS experiments, light is absorbed at one frequency and re-emitted at another, creating one or more particle-hole pairs. In the so-called polarized geometry with the incident and outgoing photons having the same polarization (so that no spin is transferred to the QWR), RRS experiments in GaAs QWRs consistently show two peaks which indeed look qualitatively very similar to the spectra for the corresponding 2D and 3D systems. In these higher dimensional systems, the two peaks have a clear and generally accepted Fermi liquid interpretation . The higher energy peak is associated with the plasmon or charge density excitation (CDE), a collective density excitation of the electron gas, and the lower energy spectral peak is associated with incoherent particle-hole pair excitations (SPE). In the QWR materials, the lower energy peak occurs at an (approximate) excitation energy of $`qv_F`$, where $`q`$ is the excitation momentum and $`v_F`$ is the 1D Fermi velocity obtained from the band structure of the QWR. An interpretation of the lower peak as an SPE contribution seems therefore natural . However, there is a strong theoretical objection to this interpretation: in a one dimensional system there is spin-charge separation: the only charge excitations live at the plasmon frequency, and cannot contribute to excitations at the SPE energy. The signal observed in this $`q,\nu `$ range must be due to the chargeless spin excitations of the LL; in particular it is possible to combine two $`S=1/2`$ excitations into a $`S=0`$ object, creation of which is allowed by the Raman selection rules.
Sassetti and Kramer (S-K) presented a qualitative theory of this effect . They showed that although the leading contribution to the RRS matrix element corresponds to coupling the light to the electron density operator, there is a sub-leading term (which becomes more important under resonance conditions) which may be interpreted as a coupling of light to the energy density fluctuations of the electrons in the QWR. The energy density fluctuations have a contribution from the spin excitations, which qualitatively explains the data, but the S-K theory did not calculate the spectral weights of the RRS peaks. Too close to resonance, the S-K theory breaks down. The S-K work also does not show how to distinguish a LL from a FL in the RRS experiment. The most important theoretical problem is that the S-K calculation is logically inconsistent, because it uses an expression for the RRS matrix element which is correct only if the conduction band is a FL not an LL. Thus S-K uses FL matrix elements but LL excitations. In our paper the correct LL matrix element is used, leading to expressions different from those derived by S-K.
In this paper we present an essentially complete treatment of RRS in a one dimensional electron gas. We obtain a precise expression for the energy transferred to the QWR in a RRS experiment, valid at all values of the difference of the energy from resonance, and evaluate it quantitatively in several experimentally relevant limits. We show which features of the data contain information about the LL exponents, obtain expressions for the relative amplitudes of the SPE and CDE peaks, determine lineshapes and discuss qualitatively the crossover from LL to FL behavior.
Resonant Raman scattering is a two-photon process in which a photon is absorbed, transferring an electron from the valence $`(V)`$ band to the conduction $`(c)`$ band and a photon is emitted, transferring an electron from the conduction band back to the valence band. We assume that the valence band is initially filled, and assume there is no excitonic interaction between conduction and valence band states. The excited valence hole is then described by a single-particle Hamiltonian, which we write as $`H_V`$, while the conduction band is described by some interacting Hamiltonian which we denote $`H_{LL}`$. We denote the photon absorption and emission by $`P_{1,2}`$ respectively. The RRS process is described by the following Hamiltonian:
$$H=H_V+H_{LL}+\widehat{P}_1+\widehat{P}_2$$
(1)
where the photon-in $`(P_1)`$ and photon-out $`(P_2)`$ terms are
$`\widehat{P}_1`$ $`=`$ $`e^{i(\mathrm{\Omega }+\nu /2)t}{\displaystyle \underset{p,s}{}}c_{p+q/2,s}^{}(t)v_{p,s}(t)+\mathrm{h}.\mathrm{c}.`$ (2)
$`\widehat{P}_2`$ $`=`$ $`e^{i(\mathrm{\Omega }\nu /2)t}{\displaystyle \underset{p,s}{}}v_{p,s}^{}(t)c_{pq/2,s}(t)+\mathrm{h}.\mathrm{c}.`$ (3)
with $`c`$ and $`v`$ the annihilation operators for electrons in conduction and valence band states respectively. Note that the operator $`v_{p,\sigma }^{}`$ creates an eigenstate of $`H_V`$ with energy $`E_p^V`$ while the $`c_{p,\sigma }^{}`$ operators does not create eigenstates of $`H_{LL}`$. The absorbed(emitted) photon energy and momentum are set $`\mathrm{\Omega }\pm \nu /2`$ and $`\pm q/2`$ respectively.
We now use the standard methods of time-dependent perturbation theory to calculate the amplitude, $`a_n(t_0)`$, for the system at some time $`t_0`$ to be in some excited state $`|n`$ of QWR, but with no holes in the valence band. We assume the system is in its ground state at $`t=0`$. Our neglect of any excitonic interaction between conduction and valence band simplifies the calculation and we obtain
$$a_n(t_0)=\frac{1}{L}\underset{r,s}{}𝑑Re^{iqR}_0^{t_0}𝑑Te^{i\nu T}n|\widehat{O}_{rs}(R,T)|0$$
(4)
with
$`\widehat{O}_{rs}(R,T)={\displaystyle }dx{\displaystyle _0^T}dt\varphi (x,t)\times `$ (5)
$`\psi _{rs}(R+x/2,T+t/2)\psi _{rs}^{}(Rx/2,Tt/2),`$ (6)
where $`r`$ and $`s`$ are band and spin indices ($`pm1`$), and
$$\varphi (x,t)=e^{i\mathrm{\Omega }t}\underset{p}{}e^{i(E_p^Vtpx)}.$$
(7)
Eqs. (4) and (5) are our fundamental new results: they show that the RRS process acts to create a particle-hole pair at a spatial separation $`x`$ and temporal separation $`t`$. These are determined by the average photon frequency $`\mathrm{\Omega }`$ and the valence-band properties encoded in $`E_p^V`$. Further, if interactions are present in the conduction band, the states created by $`\psi ^{}`$ and by $`\psi `$ are not eigenstates of $`H_{LL}`$ and therefore the matrix element is itself modified by interactions.
We note that Eqs. (4) and (5) maybe substantially simplified in the limit of greatest physical interest. We linearize the valence band energy about the conduction band Fermi momentum, writing $`E_F^V=\mathrm{\Delta }v_F^V(rpp_F)`$ for branch $`r`$ and define $`\omega _R=\mathrm{\Omega }\mathrm{\Delta }`$ as the photon frequency with respect to the resonance energy, $`\mathrm{\Delta }`$. The $`p`$integral gives $`\delta (x+v_F^Vt)`$. Finally we write the conduction band operators in terms of the bosons which create eigenstates of $`H_{LL}`$, and normal-order in the boson basis, obtaining
$`\widehat{O}_{rs}(R,T)`$ $`=`$ $`L{\displaystyle _0^T}𝑑te^{i\omega _Rt}G_{rs}^c(rv_F^Vt,t)`$ (9)
$`:e^{i\mathrm{\Phi }_{rs,\rho }(R,rv_F^Vt;T,t)}::e^{i\mathrm{\Phi }_{rs,\sigma }(R,rv_F^Vt;T,t)}:,`$
where
$`\mathrm{\Phi }_{rs,\rho }(R,x;T,t)=2{\displaystyle \underset{p>0}{}}e^{\alpha p/2}\sqrt{{\displaystyle \frac{\pi }{pL}}}\times `$ (10)
$`\{\mathrm{sinh}\theta _\rho \mathrm{sin}[p(rx+v_\rho t)/2][b_{rp}^{}e^{ip(rR+v_\rho T)}+\mathrm{h}.\mathrm{c}.]`$ (11)
$`+\mathrm{cosh}\theta _\rho \mathrm{sin}[p(rxv_\rho t)/2][b_{rp}^{}e^{ip(rRv_\rho T)}+\mathrm{h}.\mathrm{c}.]\},`$ (12)
(13)
$`\mathrm{\Phi }_{rs,\sigma }(R,x;T,t)=2s{\displaystyle \underset{p>0}{}}e^{\alpha p/2}\sqrt{{\displaystyle \frac{\pi }{pL}}}\times `$ (14)
$`\left\{\mathrm{sin}[p(rxv_F^ct)/2][\sigma _{rp}^{}e^{ip(rRv_F^cT)}+\mathrm{h}.\mathrm{c}.]\right\}.`$ (15)
Here $`b^+`$ and $`\sigma ^+`$ create charge and spin excitations respectively and $`v_\rho =v_F^ce^{2\theta _\rho }`$ is the plasmon velocity, where the exponent $`e^{2\theta _\rho }=\sqrt{1+2g/\pi v_F^c}`$ is defined for the short-ranged interaction, $`g`$. $`G^c`$ is the exact conduction band Green’s function at spatial separation $`rv_F^Vt`$, and temporal separation $`t`$. We have assumed that the interactions are negligible in the spin sector and therefore the spin excitation velocity is just the Fermi velocity. As long as $`v_F^V`$, the valence band velocity at the conduction band $`p_F`$ is different from the spin and charge velocities of Luttinger liquid, $`G^c`$ is a decaying function of $`t`$. In the noninteracting case, $`G^c1/t`$; interaction leads to a faster decay: $`G^c1/t^{1+\alpha }`$ with the LL exponent $`\alpha =\mathrm{sinh}^2\theta _\rho >0`$ (not the same one as we use in Eqs. (8-9) for infinitely small convergent factor) for short-ranged interactions; $`G^c`$ decays faster with the physically relevant long-ranged interactions. This faster decay of $`G^c`$ is the mathematical expression of the renormalization of the RRS vertex by the interactions, which produce the Luttinger liquid behavior. As we will now show, it has important consequences for various aspects of the RRS spectra; and in particular for the dependence of the CDE and SPE energies on the difference of the average photon energy from the resonance.
We defer to a subsequent paper a full evaluation of the RRS correlation function, which is computationally demanding and not very illuminating, and present here the results of expanding Eq. (7) in terms of boson operators. The essential point is that if the combination of $`e^{i\omega _Rt}G^c(rv_F^vt,t)`$ decays rapidly as $`t`$ increases (large $`\omega _R`$ as off-resonance or large $`\alpha `$ as strong interaction), then the $`t`$integral is dominated by small times and an expression in power of bosons is rapidly convergent. We will show below that the first order term, one-boson result, gives the main contribution to CDE spectrum and dominates the whole spectrum as off-resonance and the second order term, two-boson (spinon) result, gives the peak at ”SPE” energy as near resonance, but it still has relatively small weights as compared to the first order CDE.
Expanding the exponentials, keeping only the one-boson term and integrating explicitly, gives the one-boson transition rate as a delta function at $`\nu =qv_\rho `$ with the spectral weight ($`\alpha <1`$)
$$W_1=\frac{2L\mathrm{\Gamma }^2(\alpha )}{qv_\rho ^2}\left|\left(\frac{\omega _R\omega _q}{E_0}\right)^\alpha \left(\frac{\omega _R+\omega _q}{E_0}\right)^\alpha \right|^2,$$
(16)
where $`\omega _qqv_\rho /2`$, neglecting $`v_F^V`$ for simplicity. $`E_0`$ is the energy scale depending on the interaction range and roughly of the order of Fermi energy, $`E_F^c`$. For $`\omega _q|\omega _R|`$, $`W_1|\omega _R|^{2\alpha 2}`$, while for $`\omega _R=0`$, $`W_1\mathrm{sin}^2(\pi \alpha /2)`$. Thus LL effects enter the CDE portion (one-boson) of the spectrum in two ways (for short-ranged interaction): first, far from resonance, it changes the frequency dependence of spectral weights from $`\omega _R^2`$, the noninteracting result, to $`\omega _R^{2+2\alpha }`$ (note that all other higher order bosonic contribution decays much faster, this confirms the validity of the bosonic expansion we mentioned above). and secondly as on resonance ($`\omega _R=0`$) it changes the value to be nonzero due to finite interaction strength.
To second order, two new effects appear. In the density spectrum, branch mixing process appear. These lead to a continuum absorption beginning at the CDE threshold, $`\omega =qv_\rho `$. In addition, an $`S=0`$ combination of spin excitations may be excited via the two spinon, $`\sigma \sigma ,\sigma \sigma `$ (note that there is no first order contribution in spin channel due to the selection rule of RRS in the polarized spectroscopy), and gives the so-called ”SPE” mode at $`\nu =qv_F^c`$.
In Fig. 1, we show the spectrum from one and two bosons in different resonance energy. One can find that (i) the overall spectral weights decays very fast off resonance, and (ii) the ”SPE” peak is generated at $`\omega 0.2E_F^c`$ by the two-boson contribution near resonance. But as compared with the CDE peak at plasmon energy (about 0.57 $`E_F^c`$), the ”SPE” peak is still very small compared with CDE. This striking result arises from the fact that the contribution of one spin-boson in the first order is forbidden by the specific selection rule of polarization in depolarized RRS spectroscopy. (iii) At higher energy side above CDE peak, there is some continuum structure which is not shown in the range of Fig. 1. This continuum is from the interaction between different branches of charge bosons due to finite $`g_2`$ interaction. We are not interested in their structure because it goes to zero near the plasmon energy and their higher energy behavior is off the experimentally measurable region, and become unphysical due to the failure of the linear dispersion assumption. (iv) When including three or higher order boson contribution (not shown in this paper), we will see the mixture of charge boson and spin boson in a form like $`\sigma \sigma \rho ,\rho \sigma \sigma `$, which will come into the energy between $`qv_F^c`$ and $`qv_\rho `$, plasmon energy, as a continuum structure. A detailed analysis shows that these higher order contribution is relatively small and no special structure compared to the first two order result we present here. While Fig. 1 is for a specific value of $`\alpha `$ ($`=0.3`$) we show in Fig. 2 the calculated charge boson and spin boson RRS spectral weights at resonance and away from resonance. In general, the LL theory predicts much smaller spectral weight for the lower energy ”SPE” mode than the FL theory at resonance. This is particularly true since our best estimate for the Luttinger exponent of the experimental system (obtained from the CDE energy dispersion) is $`\alpha 0.4`$.
As compared with the experimental result, which shows possible comparable spectral weight of ”SPE” with CDE , we find that the LL theory result induced by resonance effect does not explain the experimental results quantitatively, even though we could recover the SPE peak through the coupling of two spinon in LL, not as the SPE in FL theory. This inconsistency cannot be resolved even evaluating the full bosonic contribution without expansion as what we have in this paper. Our future work shows that the spectral weight of the ”SPE” peak enhanced by spinon coupling in polarized RRS spectra is always relatively small compared to that of CDE. Therefore, unlike the conclusion of previous work based on the incorrect matrix element, we claim that the whole problem cannot be simply understood by the correct LL theory. We believe that the existing experimental results are in the high energy crossover regime where in fact a FL description maybe more appropriate for the RRS data than the LL description which is an asymptotic low energy description. This explains the spectacular quantitative success of the FL RRS theory developed in ref. .
In conclusion, we provide the correct LL theory for the RRS spectra calculation, and obtain some meaningful and interesting results to study the possible origin of LL features in the RRS spectra of 1D QWR systems. We also develop an useful bosonic expansion method to study the two-particle correlation function. Finally, we find that the LL theory cannot quantitatively explain the experimental data most likely because the RRS experiments are not in the asymptotic low energy LL regime.
|
warning/0004/hep-th0004131.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Like any other strong/weak duality which cannot be proven directly, the $`AdS`$/CFT duality was tested using BPS configurations. Such configurations are protected by supersymmetry and can be traced while interpolating from weak to strong coupling. Non-BPS configurations are not protected and in general any result obtained using the duality is considered to be a prediction rather then a test.
In this paper we study some non-BPS states of gauge theories at weak and strong coupling. The configurations we discuss are unstable classical solutions which sit at the top of non-contractible loops in configuration space (sphalerons) .
Let us remind the reader what a sphaleron is. Say there exists a one parameter family of field configurations that form a non-contractible loop. One should think of all homotopically equivalent loops and find the point with maximal energy along each loop. Now consider the minimum of all those energies, since the loops are not contractible, that energy has to be greater than zero, and the corresponding field configuration is a saddle point—the sphaleron. In practice, once one understands the topology, it is usually easy to find the loop going through the sphaleron. A schematic picture is given in fig. 1.
If there is a d-dimensional topologically charged object in the theory, then in general there would be a d+1-dimensional sphaleron. A simple example is a theory which has an instanton. Then consider the one parameter family of static field configurations where the extra parameter replaces the Euclidean time. This family of field configurations has the same topological charge as the instanton. By varying the parameter, one starts and ends at the vacuum, and at the middle point there will be an unstable solution to the equations of motion. It sits at the top of a non-contractible loop in the space of field configurations. This is the sphaleron.
It was recently argued by Harvey, Hořava and Kraus that unstable D-branes of string theory are sphalerons. For example the type IIB D0-brane can decay to the vacuum, but its existence is dictated by the same topology as the D-instanton, whose charge is classified by K-theory . One can construct a one parameter family of static configurations whose topology is that of the D-instanton. The D0-brane sits at the top of the loop.
This will serve as our first example. We consider the configuration of a D0-brane at the center of $`AdS`$. This is a massive, non-BPS object in the large $`N`$ and large coupling classical limit of the theory. In global $`AdS`$ geometry, where the topology of the boundary is $`S^3\times \text{}`$, this is a static, spherically symmetric, configuration.
A similar configuration exists at weak ’t Hooft coupling. It is explained in detail in Section 2, let us just say now that it is a “half pure gauge” configuration. If one considers the $`SU(2)`$ instanton , this is the configuration half-way through the tunneling process, which is at the top of the potential. That is why it is a solution of the equations of motion with one unstable mode. This gauge theory sphaleron has many properties similar to the D0-brane in $`AdS`$. It is static, spherically symmetric and has a single tachyonic mode. We will argue that it is dual to the D0-brane in $`AdS`$. We also find duals of the configuration with $`k`$ coincident D0-branes, which have $`k^2`$ unstable modes, in string theory and in the gauge theory.
It is rather perplexing at first that we are able to find a dual description for a non-BPS object. But there is, in fact a good reason for that. The D0-brane sits in the middle of a non-contractible loop with the same topology as the D-instanton, while the gauge theory solution is at the middle of a loop with the topology of the gauge theory instanton which is dual to the D-instantons.
Put differently, the instanton describes a tunneling process under a potential barrier, and the sphaleron sits at the top of the potential. The mass of the sphaleron is the maximum hight of the potential. In the dual theory, the D-instanton also describes a tunneling event, and the sphaleron is again at the top of the potential barrier. The mass of the D0-brane is the hight of the potential. Since the YM instanton and D-instanton are dual, they describe the same tunneling process in the dual pictures. The shape of the potential is altered by quantum corrections, but there is always an unstable point in the middle.
It is very simple to calculate the potential through which the instanton tunnels, it is given by a quartic of the field. The potential of string theory is much more complicated, understanding this potential is crucial to proving the brane anti-brane annihilation procedure, which is in the heart of Sen’s construction, and the classification of D-brane charges by K-theory. This issue was addressed recently by using level truncation in string field theory with impressive results. Our dual description fits neatly with Sen’s conjecture.
One should contrast this with other strong-weak dualities. It is more typical for the topological excitations of one theory to become the elementary excitations of the dual theory. For example the kink of the Sine-Gordon model become the fermions in the dual Thirring model. The same is true in S-duality of $`𝒩=4`$ Yang-Mills (and type IIB), where the topologically charged monopole goes over to the W-boson which is the elementary excitation. Here we find that one topologically charged object goes to another topologically charged object, and therefore there are sphalerons associated to those topologies. Roughly speaking, the AdS/CFT duality is special since it is a strong/weak duality with respect to the ’t Hooft coupling, while the solitons’ masses are of the order of $`1/g_{YM}^2`$.
These “half pure gauge” configurations were considered in the past on $`\text{}^4`$. They are singular at the origin and at infinity, but the singularities can be smoothed out. Those objects were named merons . The singularity at the origin and at infinity are replaced with half an instanton, interpolating between the vacuum and the meron.
This has an exact analog in Euclidean $`AdS`$, where a D0-brane appearing out of the vacuum, propagating and annihilating is dual to the meron. The D0-brane follows a geodesic in $`AdS`$, and it’s action depends logarithmically on the separation of the two end points. The same logarithmic behavior (up to a coefficient which depends on the ’t Hooft coupling) shows up on the gauge theory side. Because of the entropy of those configurations, they might dominate the path integral for large $`g_{YM}`$.
We will also argue that each of the two end points of the D0-brane carries half a unit of D-instanton charge. The D0-brane serves as a flux tube carrying half a unit of flux from one end to the other, thus preserving the Dirac quantization condition of D-instanton charge. A similar story applies to higher dimensional branes, so the unstable D-branes can be regarded as D-merons. Unlike $`AdS`$, where the action of the D0-brane is logarithmic, in flat space it’s linear, therefore it would not be dynamically favorable for D-branes to break by this mechanism.
The paper is organized as follows. We describe the details of the sphaleron on $`S^3\times \text{}`$ and the D0-brane in Lorentzian $`AdS`$ in section 2. In Section 3 we describe the meron configurations. We review the old construction in the gauge theory, and then we describe its dual. We interpret the unstable branes as D-merons in Section 4. In Section 5 we consider another example of a duality between unstable classical solutions. We show that gauge theories in the Coulomb phase admit unstable string solutions which do not carry gauge invariant magnetic or electric fluxes. We describe the $`AdS`$ dual of this solution. The unstable string can also serve as a meron, and we explain how a monopole can be separated into two halves as long as they are connected by one of those strings.
## 2 Sphaleron particle
In this section we consider sphaleron particles in four dimensional $`U(N)`$ Yang-Mills theory, and their $`AdS`$ duals. Since Yang-Mills theory is a conformal theory there are no static finite energy (stable or unstable) solutions on $`\text{}^4`$ simply because there is no scale to fix the mass of the solution. However, there is a sphaleron particle if we consider the gauge theory on $`S^3\times \text{}`$. In that case the size of the sphere, $`R`$, is the only scale in the theory and so the mass of any static solution is $`1/R`$.
We consider first the perturbative YM description, and then the $`AdS`$ dual. While the duality is true only for the theory with the $`𝒩=4`$ matter content, in perturbation theory the particle exists already in the pure gauge theory.
### 2.1 Gauge theory description
The topology that supports a stable particle in four dimensions is the map from the $`S^2`$ at spatial infinity to the fields. For $`U(N)`$ pure gauge theory the only relevant topology is $`\pi _2(U(N))=0`$. Hence this theory does not admit any topologically charged stable particles (on either $`\text{}^4`$ or $`\text{}\times S^3`$). However, since
$$\pi _{2l+1}(U(N))=\text{},\text{for}l<N,$$
(2.1)
there are unstable solutions to YM theory. These solutions, which we describe below, sit at the top of a non-contractible $`S^{2l1}`$ in configuration space.
We start by considering the simplest case of $`l=1`$. In that case we have a non-contractible loop in the configuration space of $`SU(2)`$ gauge theory which we embed in $`SU(N)`$. The topology of the non-contractible loop is the same as the instanton topology. It is useful to recall the instanton solution, it is given by the ansatz
$$A_\mu =if(r)_\mu UU^{},U=\frac{x^\mu \sigma _\mu }{r}=\frac{x_0+ix_i\sigma _i}{r},r^2=x_0^2+x_i^2,$$
(2.2)
where $`\sigma _i`$ are the Pauli matrices and $`x_0,x_i`$ the four Euclidean directions. The Yang-Mills action now yields
$$S=\frac{1}{4g_{YM}^2}_0^{\mathrm{}}𝑑r\mathrm{\hspace{0.17em}96}\pi ^2\left(\frac{r}{2}f^2+\frac{2}{r}f^2(1f)^2\right).$$
(2.3)
The equations of motions have three constant solutions $`f=0`$, $`f=1`$ and $`f=1/2`$. $`f=0,1`$ are stable solutions which correspond to two vacua. The instanton solution, $`f(r)=r^2/(a^2+r^2)`$, interpolates between $`f=0`$ at the origin and $`f=1`$ at infinity. The configuration with $`f=1/2`$ is an unstable solution, it solves the second order equation of motion, but unlike the two vacua and the instanton solution, does not solve the first order BPS equation.
On $`\text{}^4`$ we see from (2.2) that $`f=1/2`$ is a non-static singular solution. It was first discussed in and was studied further in . Those are the merons which we will discuss in the next section. On $`S^3\times \text{}`$ however, the solution is static, regular and completely delocalized<sup>1</sup><sup>1</sup>1 Since the solution is smeared over the entire $`S^3`$, it could be considered a tachyonic vacuum, rather than an unstable particle. Since the space is compact, it is hard to distinguish between the two notions. on $`S^3`$. To see this, note that the conformal transformation that takes $`\text{}^4`$ with metric $`ds^2=dr^2+r^2d\mathrm{\Omega }_3^2`$ to $`S^3\times \text{}`$ with metric $`ds^2=dt^2+R^2d\mathrm{\Omega }_3^2`$ is
$$r=\mathrm{exp}(t/R).$$
(2.4)
Therefore the action of the sphaleron on $`S^3\times \text{}`$ is
$$S=_0^{\mathrm{}}𝑑r\frac{3\pi ^2}{g_{YM}^2r}=\frac{3\pi ^2}{g_{YM}^2R}_{\mathrm{}}^{\mathrm{}}𝑑t.$$
(2.5)
We see that the action does not depend on $`t`$ and that the sphaleron mass is
$$M_{Sp}=\frac{3\pi ^2}{g_{YM}^2R}.$$
(2.6)
A non-contractible loop of static field configurations going between the two vacua and through the sphaleron is given by (2.2) with
$$f(r)=\alpha ,0\alpha 1.$$
(2.7)
Equation (2.2) implies that for constant $`f`$ we get $`A_r=0`$ (on $`\text{}^4`$) and hence $`A_t=0`$ (on $`S^3\times \text{}`$) and that $`A_\theta `$ does not depend on $`t`$. Therefore, $`F_{t\theta }=0`$ (where $`\theta `$ represents the $`S^3`$ coordinates). This has important implications for the non-contractible loop. First, the field configurations along the entire non-contractible loop (2.7) do not depend on $`t`$, and can be described in terms of the three dimensional theory on $`S^3`$. Second, even though the conformal map with Lorentzian signature (see e.g. ) is different from the Euclidean conformal map (2.4), Wick rotation to Lorentzian signature (on $`S^3\times \text{}`$) is trivial along the entire non-contractible loop. This is not the case for the instanton solution, which depends on $`r`$. Finally,
$$\mathrm{Tr}F\stackrel{~}{F}=0,\text{while}\mathrm{Tr}F^2=\frac{6}{R^4}0.$$
(2.8)
These features will prove to be important for the dual description, as we shall see in the next section.
Next we turn to the cases when $`l>1`$. In those cases the solution exists only for $`SU(N)`$ with $`N>2`$. Finding all sphaleron solutions for $`SU(N)`$ gauge theory is beyond the scope of the paper. However, there is a very simple construction which yields sphalerons related to arbitrarily high homotopy groups. Those are dual to the coincident D0-branes in $`AdS`$.
We can generalize the spherically symmetric ansatz (2.2) to larger gauge groups by replacing the Pauli matrices and the identity by
$$A_\mu =if(r)_\mu UU^{},U=\frac{x^\mu \gamma _\mu }{r},$$
(2.9)
where the $`\gamma `$’s satisfy the algebra $`\gamma _\mu \gamma _\nu ^{}+\gamma _\nu \gamma _\mu ^{}=2\delta _{\mu \nu }`$. We use the simple choice
$$\gamma _\mu =\sigma _\mu I_k=\left(\begin{array}{cccc}\sigma _\mu & 0& \mathrm{}& 0\\ 0& \sigma _\mu & \mathrm{}& 0\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ 0& 0& \mathrm{}& \sigma _\mu \end{array}\right),$$
(2.10)
where $`I_k`$ is the identity matrix of rank $`k`$. It is easy to see that this is still a solution of the equations of motion if $`f=1/2`$. The action simply scales as the rank, $`2k`$, of the matrices $`\gamma _\mu `$. Therefore the mass of the k-sphaleron is
$$M_k=kM_{Sp}.$$
(2.11)
This sphaleron solution has $`k^2`$ unstable modes, which correspond to each of the $`2\times 2`$ entries in the matrix in (2.10). The number of unstable modes alone does not fix the topology of the non-contractible loops associated with the sphaleron. For example, the fact that we have $`k^2`$ unstable modes does not mean that the sphaleron sits at the top of $`S^{k^2}`$. This would be inconsistent with $`\pi _{k^2}(U(N))=0`$ for even $`k`$. In fact the topology is exactly that of $`U(k)`$. The sphaleron sits at the point $`I_k`$ in the group, which is opposite to the identity<sup>2</sup><sup>2</sup>2 We described the sphaleron as the point in the algebra of $`U(k)`$ with $`f=\frac{1}{2}I_k`$. In the group that corresponds to $`\mathrm{exp}(2\pi if)=I_k`$.. The $`k^2`$ unstable modes are the tangent vectors in the algebra of $`U(k)`$. Since the group $`U(k)`$ has non-contractible $`S^{2l1}`$ for all $`0<lk`$, there are such loops going through the sphaleron. So we can choose to classify the tangent directions by those spheres. All together there are indeed $`1+3+\mathrm{}+2k1=k^2`$ unstable directions. The sphaleron sits, therefore, at the top of $`S^1,S^3,\mathrm{},S^{2k1}`$. In the next section we shall see that this fit neatly with the results of .
Let us show this explicitly for $`k=2`$. Consider
$$A_\mu =i_\mu UU^{}H,$$
(2.12)
where $`U`$ is of rank two, as defined in (2.2), and $`H`$ any $`2\times 2`$ Hermitian matrix. We can parameterize
$$H=\frac{1}{2}(1+\alpha )I_2+\frac{1}{2}\beta _i\sigma _i.$$
(2.13)
The sphaleron is at $`\alpha =\beta _i=0`$, which has $`H=\frac{1}{2}I_2`$. Two vacua are given by $`\alpha =\pm 1`$, $`\beta _i=0`$, so that $`H=0,1`$. There is another family of vacua, at $`\alpha =0`$, $`|\beta |=1`$, those are parameterized by an $`S^2`$, the direction of $`\beta _i`$. Those vacua give $`H`$ with one eigenvalue equal to zero and the other equal to one.
Identifying the two vacua at the end of the interval $`1\alpha 1`$ gives the non-contractible $`S^1`$. The parameters $`\beta _i`$ (with $`|\beta |<1`$) take values in the ball $`B^3`$. Identifying all the boundary points gives a non-contractible $`S^3`$.
The parameter $`\alpha `$ in (2.7) gives a one-dimensional family of configurations in $`SU(2)`$. In the previous paragraphs $`\alpha `$ and $`\beta _i`$ gave a one and a three dimensional family of configurations in $`SU(4)`$. Those are actually related to the non-trivial $`\pi _3`$ of $`SU(2)`$ and to the non-trivial $`\pi _3`$ and $`\pi _5`$ of $`SU(4)`$. This is true in general. To see this we have to include the spatial manifold $`S^3`$.
The parameters $`\alpha `$, $`\beta _i`$ and the higher dimensional ones live in $`B^{2l1}`$. At every point there is a static field configuration on $`S^3`$. So we have an $`S^3`$ for every point in $`B^{2l1}`$. At the boundary of the ball the field configuration is the vacuum, which is trivial on the $`S^3`$, so we can take to sphere to shrink to a point. This fibration of $`S^3`$ over $`B^{2l1}`$ gives $`S^{2l+2}`$. Now recall the well known fact that if the gauge group has a non trivial $`\pi _{2l+1}`$ then there is a non-trivial gauge bundle over $`S^{2l+2}`$ (the map from $`S^{2l+1}`$ to the group is the transition function on the equator of $`S^{2l+2}`$).
In the simplest case, adding the parameter $`\alpha `$ to $`S^3`$ allows us to build an $`S^4`$, on which there are configurations with the topology of the instanton.
### 2.2 Supergravity side—Unstable D0-branes in $`AdS_5\times S^5`$
The $`AdS`$/CFT duality is a strong/weak duality and as such it takes classical configuration of one description into a quantum excitation of the other description. Therefore, it is very hard to trace a generic (non-BPS) classical solution of weakly coupled SYM to the $`AdS`$ description. A sphaleron is a non-supersymmetric solution sitting at the top of a non-contractible loop in the classical configuration space. Therefore, it is natural to suspect that the quantum corrections will blur the non-contractible loop. And that by the time the ’t Hooft coupling is large there will be no trace of the non-contractible loop and the sphaleron.
However, as we saw, the non-contractible loop associated with the sphaleron of the previous subsection is described by the topology of the instanton. The dual of the instanton is a D-instanton in $`AdS`$, which carries a charge in K-theory. And so we should look for a non-contractible loop with the topology of the D-instanton. Such non-contractible loops in flat space-time were constructed in . There it was argued that the sphaleron at the top of the loop is the type IIB D0-brane. We claim, therefore, that the dual of the solution of the previous section are the unstable D0-branes located at the origin of $`AdS`$. This is illustrated in fig. 2.
Let us mention a few properties of the unstable D0-branes and how they fit into the claim that they are dual to the field theory sphalerons.
* A D0-brane (or $`k`$ coincident D0-branes) which are located at the center of $`AdS`$ are static objects with respect to the global time. Therefore they correspond to static objects in the gauge theory. The center of $`AdS`$ corresponds to the extreme infra-red of the gauge theory, so the energy is uniformly distributed over $`S^3`$.
* From the closed string theory point of view the low energy supergravity fields which are excited by the D0-branes are the NS-NS graviton and dilaton. The RR-fields are not excited. Using the dictionary of that would correspond to $`\mathrm{Tr}F\stackrel{~}{F}=0`$ and to $`\mathrm{Tr}F^20`$, in agreement with the field theory results (2.8). Note that the mass of the D0-brane (and $`\mathrm{Tr}F^2`$) do receive quantum corrections for they are not protected by supersymmetry,<sup>3</sup><sup>3</sup>3The origin of the $`\sqrt{2}`$ is the fact that the open strings living on an unstable brane carry two Chan Paton factors $`I`$ and $`\sigma _1`$ .
$$M_{D0}=\frac{\sqrt{2}}{g_s\sqrt{\alpha ^{}}}=\frac{4\sqrt{2}\pi \lambda ^{1/4}}{g_{YM}^2R}.$$
(2.14)
* In it was shown that the type IIB D0-branes are sphalerons of string theory. That is, in flat space-time they sit at the top of a non-contractible loop in the configuration space of string theory. Since for large ’t Hooft coupling the “center” of $`AdS`$ can be approximated by flat space-time, one can simply embed the construction of in $`AdS`$. There is also a global way to construct the D0-branes in $`AdS`$. Starting with a system of D1-brane anti D1-brane stretching all the way to the boundary of $`AdS`$, just like in flat space-time this system contains a complex tachyon mode which can support an unstable D0-brane.
* It was further argued in that $`k`$ coincident D0-branes, which have $`k^2`$ tachyonic modes correspond to sphalerons at the top of $`S^1,S^3,\mathrm{},S^{2k1}`$ in $`U(k)`$. This is exactly what we found from the field theory side. It is worth while to note that in both descriptions the mass is proportional to $`k`$.
* The NS sector of the excitations living on the D0-branes contains a real scalar tachyonic mode. According to Sen’s conjecture at the bottom of the tachyon potential the negative energy cancels the tension of the brane and we are left with the vacuum. This was tested, to a good accuracy, via level truncation method in string field theory . On the field theory side we see that indeed the bottom of the potential ($`f=0,1`$ in (2.3)) is the vacuum. While calculating the tachyon potential in string theory is complicated, in the field theory it’s just a quartic (2.3).
* Since the tachyon is real, the potential can support a stable lower dimensional brane. A D-instantons in our case. Again, the energy of such a configuration was calculated in string field theory with impressive agreement with expectations . On the field theory side the instanton indeed interpolates between the two minima of the potential.
* Of all the instanton solutions on $`\text{}^4`$, the one of radius $`R`$ centered around the origin is special when translating to $`S^3\times \text{}`$. It goes over to a spherically symmetric solution on $`S^3`$. In that theory, this instanton can be described as a quantum mechanical tunneling process between the two minima of the quartic potential in (2.3). The gauge theory sphaleron sits at the middle of the potential. The width of the potential is $`R`$ and the hight, which is the mass of the sphaleron, is proportional to $`1/g_{YM}^2R`$. The action of the instanton is the area under the potential. In string theory the same is true, only that $`R`$ is replaced by $`l_s`$. The hight of the potential $`\lambda ^{1/4}/g_{YM}^2R=1/g_{YM}^2l_s`$, and the width is of order $`l_s`$. Since the action of the D-instanton is the same as the gauge theory instanton, the area is the same, but the shape is altered.
We see therefore, that indeed the field theory sphaleron is dual to the unstable D0-branes in $`AdS`$. It is important to emphasize that the D0-branes are not sphalerons of the low energy supergravity. That is, there is no supergravity solution associated with the non-BPS D-branes which sits at the top of a non-contractible loop of field configurations of the classical supergravity. The unstable branes are sphalerons of the full string theory including all the quantum corrections to the sigma model. Since the full string theory on $`AdS`$ contains all the information about the dual SYM theory it is not surprising that in principle the field theory sphalerons can be described by string theory on $`AdS`$. What is remarkable is that the description is so simple.
A natural question that arises is whether the dual weakly coupled description sheds new light on the diagonal U(1) problem associated with the unstable D0-branes. Unfortunately, even though we can trace the D0-branes to the weakly coupled region, we cannot trace the gauge theory living on them to the weakly coupled description. Thus, as far as we can tell, the dual description does not lead to any new insight on the U(1) problem. It is worth mentioning that this problem of tracing the gauge theory living on the brane to the weakly coupled description is not special to D0-branes. For example, we know that the dual of a D1-brane stretched all the way to the boundary is the BPS monopole. But in weakly coupled field theory there are no fields living on the monopole, while there is a $`1+1`$ gauge theory living on D1-branes in $`AdS`$. The reason is that the size of the D1-brane is larger than the string scale only for large ’t Hooft coupling and so for small coupling the excitations which were supposed to live on the monopole cannot be separated from the other excitations.
It is interesting to note that when we have $`k`$ D0-branes the full topology of the non-contractible loop, $`U(k)`$, with its non-contractible $`S^1,S^3,\mathrm{},S^{2k1}`$, can be interpolated from the weakly to the strongly coupled region. The $`S^1`$ is “protected” by the instanton which is BPS. It should be interesting to understand why the other spheres are “protected” as well.
We would like to end this section with a comment on finite $`N`$. Our construction of the field theory solution which is dual to k coincident D0-branes is valid for $`kN/2`$. Equation (2.1) implies that a dual solution should be found at up to $`k=N`$. Presumably, a more complicated ansatz will indeed yield the right solution. It should be interesting to see if the mass is still linear with $`k`$. Another question is what happens when $`k>N`$. In the field theory side we get out of the stable regime. Is there any stringy exclusion principle associated with that? Recall that the global construction of $`k`$ D0-branes in $`AdS`$ involves $`k`$ D1-branes and anti-D1-branes stretched all the way to the boundary (this is a simple generalization of the discussion in ). Now, when $`k=N`$ the D1-branes can end on a NS-brane which wraps $`S^5`$ . So it seems that the existence of a baryon vertex in $`AdS`$ is the underlying mechanism which bounds the number of coincident D0-branes in $`AdS`$ to $`N`$. Clearly, it would be nice to understand this better.
## 3 Merons in gauge theories and in $`AdS`$
In Section 2.1 we studied the field configuration of “half pure gauge” on $`S^3\times \text{}`$, and interpreted it as a sphaleron. As we mentioned, those same configurations can be considered in the Euclidean theory on $`\text{}^4`$, they are still classical solutions, but there is a singularity at the origin and at infinity. By smoothing out the singularities one gets a configuration that solves the equations of motion almost everywhere and has finite action. Those are the merons .
We give a brief review of the merons in gauge theories and then will find analogous configurations in string theory on $`AdS`$.
### 3.1 Short review of merons
Let us write again the instanton ansatz (2.2)
$$A_\mu =if(r)_\mu UU^{},U=\frac{x^\mu \sigma _\mu }{r}=\frac{x_0+ix_i\sigma _i}{r},r^2=x_0^2+x_i^2.$$
(3.1)
$`f=0,1`$ are vacuum solutions, and $`f=\frac{1}{2}`$, the meron, is an unstable solution which is singular at $`r=0,\mathrm{}`$. The action (2.5) is logarithmically divergent
$$S=\frac{3\pi ^2}{g_{YM}^2}_0^{\mathrm{}}\frac{dr}{r}.$$
(3.2)
To regularize this divergence consider the following configuration
$$f(r)=\{\begin{array}{ccc}\frac{r^2}{r^2+R_1^2},& & r<R_1\\ \\ \frac{1}{2},& & R_1<r<R_2\\ \\ \frac{r^2}{r^2+R_2^2},& & R_2<r.\end{array}$$
(3.3)
This is the meron for $`R_1<r<R_2`$, glued to half an instanton at the origin and half at infinity. This carries the same topological charge as the instanton, but it is broken in two parts. If one takes $`R_1=R_2`$, the instanton solution is recovered. For $`R_1R_2`$ this is a solution of the equations of motion everywhere but at the spheres which separate the three regions.
This is illustrated in fig. 3.
a. Region I and III are the half instantons near the origin and infinity. Region II is the meron which connects the two. The action can be easily calculated, and is equal to
$$S=\frac{8\pi ^2}{g_{YM}^2}+\frac{3\pi ^2}{g_{YM}^2}\mathrm{ln}\frac{R_2}{R_1}.$$
(3.4)
Since classical YM is conformally invariant, we can use a large gauge transformation to map region III to a sphere at finite distance. The new configuration is shown in fig 3. b. Region I and III each carry half the topological charge of the instanton, so at infinity this configuration is pure gauge.
One can, of course, replace the meron with an anti-meron, where instead of half an instanton there is half an anti-instanton. The meron anti-meron pair will have zero topological charge and two anti-merons $`1`$ topological charge. The interaction between a meron and and anti-meron is the same as that between two merons.
The action of a meron grows with the distance. Thus a first guess is that the contribution of merons to the partition function is negligible. However, the action grows only logarithmically so it can be compensated by a large entropic factor.<sup>4</sup><sup>4</sup>4 In thermodynamics this is, of course, common. At finite temperature one has to minimize the free energy, $`F=EST`$ rather then the energy. Thus a phase transition between minimizing $`E`$ and maximizing the entropy can take place. Here the coupling constant plays the role of the temperature. The entropy contribution to the partition function goes like $`L^4`$, hence the partition function associated with a meron is
$$ZL^4\mathrm{exp}\left(\frac{1}{g_{YM}^2}\mathrm{ln}L\right)=L^{(41/g_{YM}^2)}.$$
(3.5)
This suggest a phase transition at $`g_{YM}^2\frac{1}{4}`$, wherein the meron charges that made up the instanton dipole are liberated. In the non-supersymmetric theories it was suggested that the appearance of this new phase at large coupling, or large scale size, is closely related to confinement, where the merons play the role of the three dimensional instantons in Polyakov’s mechanism for confinement . However, the full story is much more complicated for one has to consider a gas of merons and their interactions. This, as well as the fact the coupling runs, made it very hard to estimate the relevance of merons to confinement.
Even though the coupling does not run for $`𝒩=4`$, the main problem of understanding the interactions among the merons is still very complicated. In fact, in the $`𝒩=4`$ theory, because of the fermions and scalars and the fact that a meron breaks all supersymmetry, it is probably even more complicated. We however cannot resist the temptation of speculating that meron physics might be a clue for understanding $`𝒩=4`$ theory at the self-dual point ($`g_{YM}^2=2\pi `$).
### 3.2 Merons in $`AdS`$
We would now like to describe merons in the strong coupling limit of the field theory, using string theory on $`AdS`$. We saw in Section 2 that the sphaleron solution of the gauge theory on $`S^3\times \text{}`$ is described in the dual theory by an unstable D0-brane. Since the meron is the same field configuration as the sphaleron, only on $`\text{}^4`$, it is also described by a D0-brane in Euclidean $`AdS`$. Here we use the metric
$$\frac{ds^2}{\alpha ^{}}=\frac{\sqrt{\lambda }}{U^2}dU^2+\frac{U^2}{\sqrt{\lambda }}dx^2.$$
(3.6)
Consider a D0-brane which is created at some point $`U_1`$, propagates till $`U_2`$ (and the same point in $`\text{}^4`$) and annihilates. This is the $`AdS`$ dual of the configuration (3.3) which was illustrated in fig. 3a. By the UV/IR relation, for $`U_1>U_2`$, the internal circle has a radius $`R_1=\sqrt{\lambda }/U_1`$ and the external circle $`R_2=\sqrt{\lambda }/U_2`$.
The action of this configuration is
$$S=S_{\mathrm{D}(1)}+T_{\mathrm{D0}}𝑑s=\frac{2\pi }{g_s}+\frac{\sqrt{2}\lambda ^{1/4}}{g_s}\mathrm{ln}\frac{U_1}{U_2},$$
(3.7)
where the first term $`2\pi /g_s=8\pi ^2/g_{YM}^2`$ is equal to the instanton action and is related to the creation of the brane and its annihilation, like in the gauge theory. This contribution will be justified in the next section. Comparing this to the gauge theory result (3.4), the constant part of the action is unchanged, but the coefficient of the log is renormalized by a factor proportional to $`\lambda ^{1/4}`$, like the sphaleron mass (2.14). Again, one should not be surprised, since this is a non-BPS configuration.
Just as was explained in the previous section a conformal transformation will take this geodesic into a D0-brane which is created and annihilated at the same value of $`U`$, but at a distance $`L`$ on $`\text{}^4`$, this is the $`AdS`$ dual of the configuration in fig. 3b. The size of the two half instantons is simply $`R=\sqrt{\lambda }/U`$ Those two configurations are shown in fig. 4.
It is not surprising, therefore, that the corresponding action is
$$S=\frac{8\pi ^2}{g_{YM}^2}+\frac{4\pi \sqrt{2}\lambda ^{1/4}}{g_{YM}^2}\mathrm{ln}(L/R).$$
(3.8)
The fact that the logarithmic term is now proportional to $`\lambda ^{1/4}/g_{YM}^2`$, rather then just $`1/g_{YM}^2`$ as in the weakly coupled theory seems to imply that the entropy contribution cannot compete with the energy in strong coupling. That is,
$$ZL^4\mathrm{exp}\left(\frac{\lambda ^{1/4}}{g_s}\mathrm{ln}(L)\right).$$
(3.9)
So a phase transition at $`g_s1`$ is very unlikely for large $`\lambda `$.
## 4 Unstable branes as D-merons
In the previous section we studied D0-branes in Euclidean $`AdS`$. Since they are unstable they can appear out of the vacuum, propagate some distance and disappear again. This was dual to the meron in the gauge theory which connects two regions where there are half instantons. Since the $`AdS`$ dual of the instanton is the D-instanton, it is natural to suspect that at each end of the D0-brane sits half a D-instanton.
We reached that conclusion by studying D0-branes in $`AdS`$, but this is true in any string theory background, and the argument does not have to rely on the $`AdS`$/CFT correspondence. After all, the D0-brane is a sphaleron at the top of a non-contractible loop with the same topology of the D-instanton. Therefore the entire event of a D0-brane creation, propagation and annihilation can carry a unit of D-instanton charge. In fact, it can carry 1, 0, or $`1`$ units of D-instanton charge.
The creation or annihilation of a D0-brane is an event that carries half (or minus a half) of D-instanton charge. This might seem to contradict the charge quantization condition. The product of the charge of a single D7-brane and the charge of a single D-instanton is $`2\pi `$, so how can a D-instanton break in two? The answer is that the two halves of the D-instanton are connected by a D0-brane, which must carry half a unit of D-instanton flux.
This is analogous to a bar magnet, or a solenoid in electro-magnetism. Outside the magnet the magnetic field looks like that of two separated, oppositely charged, monopoles. But the monopole charge need not satisfy the Dirac quantization condition, as the magnet (or solenoid), carries the flux from one to the other.
It is amusing to push this analogy further. Just as the magnetic field in a magnet is created by the angular momentum of the electric charges, the D0-brane can be regarded as a very thin solenoid in which a current of D7-brane charge produces a dual flux, connecting the one-half D(-1) charges. It would be interesting to pursue this analogy further.
Since the unstable D0-branes connect pairs of $`1/2`$ D-instantons, they could be called D-merons.
Thus far we considered only D0-branes, but the same is true for higher dimensional branes as well. A D1-brane can break into two halves with an unstable D2-brane in the middle. That is the same as saying that the boundary of a Euclidean D2-brane could carry half-D1-brane charge. Likewise in type IIA, a D0-brane can break in two with an unstable D1-brane in the middle, and so on. A D2-brane ending on two half D1-branes is shown in fig. 5.
In $`AdS`$ the action of the D0-brane is logarithmic, however in flat space it will be linear. Therefore half D-instantons are clearly confined in flat space. The same is true for the higher dimensional half-branes.
## 5 Unstable strings in the Coulomb phase
In previous sections we discussed how the existence of the instanton implies that there is a point like sphaleron solution. By the same logic, the ’t Hooft-Polyakov monopole implies the existence of a string like sphaleron solution in gauge theories in the Coulomb phase. We discuss the field theory construction of the string and its supergravity dual.
### 5.1 Field theory description
We first study the unstable string in the $`SU(2)`$ gauge theory broken to $`U(1)`$ by an adjoint Higgs. The details of the construction, the relevant non-contractible loop in configuration space and the unstable string sitting at the top of the loop can be found in . Those papers considered the theory in three dimensions, where the monopole is an instanton and the sphaleron is a particle. We are interested in uplifting this to four dimensions. We shall not review the explicit construction but rather deduce the relevant properties from general arguments.
The monopole solution yields a radial $`U(1)`$ magnetic field,<sup>5</sup><sup>5</sup>5 We remind the reader that the $`U(1)`$ components of the $`SU(2)`$ is defined with respect to the Higgs field, $`F_{\mu \nu }=F_{\mu \nu }^aW^a`$.
$$F_{ij}=\frac{1}{er^3}ϵ_{ijk}x_k,.$$
(5.1)
To construct the non-contractible loop associated with this solution we have to consider configurations which are invariant under translation in one direction, say $`x_3`$. Then we replace the coordinate with a parameter in configuration space $`x_3\mathrm{tan}\alpha `$. This is pictured in fig. 6.
Note that to get configurations which are independent of the $`x_3`$ coordinate one has to perform an $`\alpha `$ dependent gauge transformation. This does not change the topology of the loop, but it does change the action. Therefore one cannot simply replace $`x_3`$ with $`\mathrm{tan}\alpha `$ in the solution.
After the gauge transformation, the sphaleron string is given by
$$A_a=f(x)ϵ_{ab}x_b\sigma _3,\mathrm{\Phi }=g(x)x_a\sigma _a,$$
(5.2)
with $`a,b=1,2`$. For more details see .
For $`\alpha =0`$ we see (from fig. 6, (5.1) or ) that there is a solution localized in the $`x^1,x^2`$ plane with no magnetic flux in the plane. Thus we have an unstable string solution (stretched along the $`x_3`$ direction). The string does not carry gauge invariant $`U(1)`$ flux, but it does carry $`SU(2)`$ magnetic flux in the $`x^3`$ direction. Dimensional analysis implies that the tension of such a string is
$$T\frac{W^2}{g_{YM}^2},$$
(5.3)
where $`W`$ is the Higgs expectation value. For $`\alpha 0`$ there is a $`U(1)`$ magnetic field and the full non-contractible loop $`\frac{\pi }{2}\alpha \frac{\pi }{2}`$ describes a transition which changes the total magnetic flux of the vacuum by one unit. Note that in the Coulomb phase this does not cost any energy as the flux expands to infinity and we are still in the vacuum.
Put differently, as one starts from the vacuum, $`\alpha =\frac{\pi }{2}`$ and goes around the non-contractible loop through the sphaleron, $`\alpha =0`$ back to the vacuum $`\alpha =\frac{\pi }{2}`$, one unit of magnetic flux is added in the $`x_3`$ direction. Thus the non-contractible loop goes between vacua with different Chern numbers.
### 5.2 Supergravity description
The $`AdS`$/CFT correspondence is not useful to describe $`SU(2)`$ broken to $`U(1)`$. Instead, we describe $`SU(2N)`$ gauge symmetry broken to $`(U(N)\times U(N))/U(1)`$ by the Higgs mechanism. The relevant supergravity background is
$$\frac{ds^2}{\alpha ^{^{}}}=\frac{1}{\sqrt{4\pi gN\left(\frac{1}{\stackrel{}{U}^4}+\frac{1}{|\stackrel{}{U}\stackrel{}{W}|^4}\right)}}dx_{||}^2+\sqrt{4\pi gN\left(\frac{1}{\stackrel{}{U}^4}+\frac{1}{|\stackrel{}{U}\stackrel{}{W}|^4}\right)}d\stackrel{}{U}^2,$$
(5.4)
where $`\stackrel{}{W}`$ is the vector that represents the Higgs expectation value.
Since the dual of the monopole is a D1-brane in the $`U`$ direction and since the sphaleron associated with the D1-brane charge is the unstable D2-brane it is natural to suspect that the dual of the unstable string is a D2-brane along the $`x_0,x_3`$ and $`U`$ directions. However, unlike in $`\text{}^{10}`$, where the boundary conditions are set at infinity, there is nothing holding the D2-brane to the horizon. One can easily see that such a D2-brane will not solve the equations of motion with free boundary conditions. Therefore, the unstable D2-brane cannot be the dual of the unstable gauge theory string.
To resolve this puzzle we should find another object. From the discussion in Section 4, the D2-brane can carry half a unit of D1-brane charge at each end. Another configuration with the same charge is a D1-brane (in the $`x_0,x_3`$ directions). To preserve the symmetry between $`\stackrel{}{U}=0`$ and $`\stackrel{}{U}=\stackrel{}{W}`$, the D1-brane should sit precisely at the center $`\stackrel{}{U}=\stackrel{}{W}/2`$. This is shown in fig. 7.
<sup>6</sup><sup>6</sup>6Other strings in this geometry were considered in .
Indeed, suppose that we place a D1-brane along the $`x_3`$ direction at some value of $`\stackrel{}{U}`$ (we could compactify the $`x^3`$ direction to get a finite mass object). The field theory tension of a such a string is calculated with respect to the field theory coordinates and is therefore
$$T_{D1}=\frac{\sqrt{g_{00}g_{11}}}{2\pi \alpha ^{^{}}g_s},$$
(5.5)
From (5.4) we see that the tension vanishes on the branes ($`\stackrel{}{U}=0`$ and $`\stackrel{}{U}=\stackrel{}{W}`$) and that the string would like to fall towards one of the branes.
There is one exception, the string located precisely in the middle $`\stackrel{}{U}=\stackrel{}{W}/2`$. It solves the classical equation of motion, however it is unstable. Any perturbation along such a string will eventually lead to either $`\stackrel{}{U}=0`$ or $`\stackrel{}{U}=\stackrel{}{W}`$. This is the instability of the string in the $`AdS`$ description.
The tension of such a D1-brane is
$$T\frac{W^2}{g_{YM}^2\sqrt{\lambda }}.$$
(5.6)
Again, we see that because this is not a BPS configuration, the tension is not protected as one interpolates from the weakly coupled region (5.3).
Such a D1 string carries magnetic flux in the diagonal $`U(1)`$ (which decouples from the bulk degrees of freedom), but not in the relative $`U(1)`$. If it falls towards one of the collections of branes, a flux is turned on in the relative gauge group. We see that if we start with a string at $`\stackrel{}{U}=0`$ and move it to $`\stackrel{}{U}=\stackrel{}{W}`$ we go back to the vacuum, but we changed the flux in the relative gauge group by one. This is the topological structure of the non-contractible loop and the configuration at the middle is a sphaleron.
One can, of course, consider the configuration with a fundamental string along the $`x_0,x_3`$ direction. Such a string carries an electric flux and has the same instability. However, on the field theory side there is no dual electric unstable string. This is an example of the case where a sphaleron of the strongly coupled theory does not have a weakly coupled analog. The reason is that the BPS configuration which is supposed to guarantee its existence is the W-boson. But unlike the monopole, the W-boson is an elementary excitation in the weakly coupled theory, and not a classical solution, and there is no related topological charge. This is related to the fact that the fundamental string does not carry a charge in K-theory.
### 5.3 $`1/2`$ Monopole configuration
In Sections 3 and 4 we showed that the unstable D0-brane is a meron connecting two half D-instantons. In this subsection we generalize the construction of merons to the ’t Hooft-Polyakov monopole. Consider a D1-brane in the double center $`AdS`$ solution (5.4) which follows one of the trajectories indicated in fig. 8.
such a brane will solve the equations of motion everywhere along the trajectory except for the two turning points. From the field theory side this corresponds to a monopole broken into two half monopoles.
Notice that, unlike the meron case, the energy of the configuration is linear with the distance between the 1/2 monopoles and hence it will not contribute to the partition function for any value of the coupling constant.
Half a monopole seems to contradict the Dirac quantization condition. Again there is a magnet connecting the two half monopoles. One might wonder how this works, since we argued that this string does not carry any $`U(1)`$ flux. The resolution of the puzzle is simple. Recall that the thickness of the string is $`1/W`$, so the string is in the region of unbroken gauge symmetry. The flux is carried, therefore, in $`SU(2)`$. See fig. 9.
## Acknowledgments
We are grateful to E. Gimon, P. Hořava and E. Witten, for stimulating discussions. The work on N.D. and D.J.G. is supported by the NSF under grant No. PHY94-07194. The work of N.I. is supported in part by the NSF under grant No. PHY97-22022.
|
warning/0004/hep-th0004007.html
|
ar5iv
|
text
|
# FIELD-DEPENDENT BRS TRANFORMATIONS AND CORRECT PRESCRIPTION FOR 1/(𝜂⋅𝑘)^𝑝-TYPE SINGULARITIES IN AXIAL GAUGES
## Acknowledgement
I wish to thank my collaborator Prof.S.D.Joglekar (IIT Kanpur, India) for introducing me to FFBRS transformations and sharing his insights with me. I also wish to thank the Moriond committee for inviting me to the conference.
## References
|
warning/0004/hep-lat0004021.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
To formulate a theory of quantum gravity in four dimensions, many approaches have been tried. One of them is a numerical approach with the method of dynamical triangulation. In two dimensions, the quantum gravity can be quantized for a central charge $`c<1`$. The method of dynamical triangulation has generally been considered to be a correct discretized model, and has given consistent results with the analytical approach: for example, the MINBU analysis and the loop length distributions. Two-dimensional quantum gravity has generally been regarded as being a toy model of four-dimensional quantum gravity. Recently, a numerical approach with the method of dynamical triangulation in analogy with a two-dimensional model has been studied. In the four-dimensional case, it has been expected that the phase transition point between the strong coupling phase and the weak coupling phase is statistically the second phase transition point. Moreover, this point is recognized as the ultraviolet (UV) fixed point of the quantum theory of gravity.
From numerical results, in four-dimensional pure gravity, it is known that there are two distinct phases. For small values of the bare gravitational coupling constant, the system is in the so-called elongated phase, which has the characteristics of a branched polymer phase. For large values of the bare gravitational coupling constant, it is in the so-called crumpled phase. Numerically, the phase transition between the two phases has actually been shown to be of first order. Therefore, it is difficult to construct a continuum theory. In other words, It is difficult to define the quantum theory of gravity on a four-dimensional triangle (4-simplex) as a simple application of the two-dimensional lattice model.
Our next step is to investigate the possibilities to extend the model of four-dimensional quantum gravity. We have three motivations: (1) a modified model in three-dimensional case, which suppresses the vertex order concentration (VOC) as the singular sub-simplex, and which can be changed of the phase structures, (2) the property of quantum field theory with background metric independence, that the manifold may be made stable by adding matter fields and (3) the balls-in-boxes model, that gives a scenario for a phase transition in four-dimensional simplicial quantum gravity, and that shows possibilities to change the phase structure with some modifications of the model. For the possibility of a continuum theory, at least, we have to find a new phase structure that has a second order phase transition point. One modification is to introduce gauge matter fields.
Recently, the phase structure with vector fields has been studied numerically. In the case of a model with vector fields, the phase structure is changed drastically and the intermediate phase, the so-called smooth phase, has been observed between the crumpled phase and the elongated phase. In this region, the string susceptibility exponents ($`\gamma _{st}`$) have negative values. They show that this region has a fractal property, and may be smooth compared with the branched polymer region. We thus expect the possibility of a continuum limit at the critical point between the crumpled phase and the smooth phase. In order to investigate the nature of the phase transition, we measure the critical exponent as the finite size scaling, and also study the scaling property of the mother boundary in analogy with the the two-dimensional case, and we expect that the scaling structure also appears in the boundary in the four-dimensional case.
This paper is organized as follows. In Section 2, we discuss the model of four-dimensional dynamical triangulation with some vector fields. In Section 3, we show our numerical results concerning measurements of the string susceptibility exponents ($`\gamma _{st}`$) and a schematic phase diagram. We thus discuss the statistical property of the phase transition between the crumpled phase and the smooth phase in the case of four-dimensional simplicial quantum gravity coupled to one gauge field ($`N_V=1`$). Furthermore, in section 4, we discuss the scaling property near to the critical point. In section 5, we discuss a scenario for the phase structure and the phase transition in four-dimensional simplicial quantum gravity coupled to matter fields. Finally, we discuss the possibility of a continuum limit of four-dimensional simplicial quantum gravity in this article.
## 2 Model
It is still not known how to provide a constructive definition of four-dimensional quantum gravity. We have considered a discretized random closed manifold in analogy with the two-dimensional case. We numerically evaluated the Euclidean path integral with the technique of dynamical triangulation, which gives a discrete summing over all possible connections of lattices that may replace the integral over diffeomorphism inequivalent metrics. Then, we naturally considered the Euclidean Einstein-Hilbert action coupled to $`N_V`$ copies of $`U(1)`$ vector fields and its discretized model with 4-simplices. The total action is $`S=S_{EH}+S_{pl}`$. We use the Einstein-Hilbert term for gravity,
$$S_{EH}[\mathrm{\Lambda },G]=d^4x\sqrt{g}\left(\mathrm{\Lambda }\frac{1}{G}R\right),$$
(1)
where $`\mathrm{\Lambda }`$ is the cosmological constant and $`G`$ is Newton’s constant. We use the discretized action for gravity,
$`S_{EH}[\kappa _2,\kappa _4]`$ $`=`$ $`\kappa _4N_4\kappa _2N_2`$ (2)
$`=`$ $`{\displaystyle \frac{2\pi }{G}}N_2+\left(\mathrm{\Lambda }^{}+{\displaystyle \frac{10}{G}}\mathrm{cos}^1\left({\displaystyle \frac{1}{4}}\right)\right),`$
where $`\kappa _21/G`$, $`\kappa _4`$ is related to $`\mathrm{\Lambda }^{}=c\mathrm{\Lambda }`$ (c is the unit volume) and $`N_i`$ is the number of $`i`$-simplices. We use the plaquette action for $`U(1)`$ gauge fields,
$$S_{pl}=\underset{t_{ijk}}{}o(t_{ijk})[A(l_{ij})+A(l_{jk})+A(l_{ki})]^2,$$
(3)
where $`l_{ij}`$ denotes a link between vertices $`i`$ and $`j`$, $`t_{ijk}`$ denotes a triangle with vertices $`i`$, $`j`$ and $`k`$ and $`o(t_{ijk})`$ denotes the number of 4-simplices sharing triangle $`t_{ijk}`$. We consider the $`U(1)`$ gauge field $`A(l_{ij})=A(l_{ji})`$ on a link $`l_{ij}`$.
We consider that a partition function of gravity with $`N_V`$ copies of $`U(1)`$ gauge fields is
$$Z(\kappa _2,\kappa _4,N_V)=\underset{N_4}{}e^{\kappa _4N_4}\underset{t(2D)T(4D)}{}e^{\kappa _2N_2}\underset{N_V}{}\underset{lt(2D)}{}dA(l)e^{S_{pl}}.$$
(4)
We sum over all four-dimensional simplicial triangulation $`T(4D)`$ in order to carry out a path integral over the metric, where we fix the topology with $`S^4`$. As is well known, we must add a small correction term ($`\delta S`$) to the lattice for fluctuations of volume,
$$\delta S=\delta \kappa _2(N_4N_4^{target})^2,$$
(5)
where $`\delta \kappa _2`$ is adjusted with an appropriate choice; we use $`\delta \kappa _2=0.00025`$.
Near to the critical point, it is expected that the partition function (eq.(4)) behaves as
$`Z(k_2k_2^c)`$ $``$ $`{\displaystyle \underset{T(4D)}{}}N_4^{\gamma _{st}3}e^{\mu N_2}`$ (6)
$``$ $`1/|N_4N_4^{target}|^{3\gamma _{st}},`$
where $`\gamma _{st}`$ is the string susceptibility exponent related to the entropy of the manifold. In the situation; $`\gamma _{st}>0`$, the manifold grows to the spiky configuration. Conversely, in $`\gamma _{st}<0`$, the surfaces grow to smooth structures.
Using Monte-Carlo simulations, we evaluated the partition function (eq.(4)). We followed the way of updating the configuration after ref.. This is that the gauge fields renew the sequence of the $`(p,q)`$-move according to the weight of the Bolzman factor. The action was checked by the Metro police methods. As we held the same conditions with Bilke et al., we performed a geometry update, $`(p,q)`$-move, updating the gauge fields by the heat-bath sweep and the over-relaxation sweep. An over-relaxation sweep was introduced for the convergence,
$$A_{ij}A_{ij}2\overline{A}_{ij}.$$
(7)
The measurement chance $`(N_4=N_4^{target})`$ came at intervals of about 100 sweeps, where we counted “1 sweep” as the flow of “heat-bath sweep $``$ $`(p,q)`$-move $``$ over-relaxation sweep”. The measurement processes and updating processes are the same as the pure gravity case. However, since the case of adding vector fields costs more CPU time, we arranged for suitable compute time.
## 3 Phase diagram with vector fields
In order to investigate the phase structure of simplicial quantum gravity, first we calculated the string susceptibility exponent ($`\gamma _{st}`$). This exponent is defined by the asymptotic form of the partition function (eq.(6)). It is known that the case of $`\gamma _{st}`$ corresponds to the dominance of the branched polymer. We measure the string susceptibility exponent ($`\gamma _{st}`$) with the method of the MINBU (Minimum Necked Baby Universe) analysis. It is a powerful exponent for probing the property of quantum geometry, and is easily measured. We can count a baby universe which is connected to the mother universe via a minimal neck. The distribution function for the MINBU analysis with the size B can written as
$`n_A`$ $`=`$ $`{\displaystyle \frac{3(AB+1)(B+1)Z[AB+1]Z[B+1]}{Z[A]}}`$ (8)
$``$ $`c\{(B+1)(1{\displaystyle \frac{B1}{A}}\left)\right\}^{\gamma _{st}2},`$
where we use the asymptotic form of the partition function, $`Z[A]A^{\gamma _{st}3}e^{\mu A}`$, where $`A`$ denotes the volume of the mother universe.
In the pure gravity case, a measurement of the string susceptibility exponent is discussed in ref.. From the numerical results, the string susceptibility exponent supports the idea that the pure case of simplicial quantum gravity has two distinct phases. What is important in the $`N_V=1`$ case is that the usual phase transition point ($`\kappa _2^c`$) is different from another transition point ($`\kappa _2^o`$), which separates the $`\gamma _{st}<0`$ region from the $`\gamma _{st}>0`$ region, and $`\gamma _{st}`$ becomes negative on the phase transition point ($`\kappa _2^c`$). This fact leads to the definition of a new smooth phase. This phase is defined by an intermediate region between these two transition points ($`\kappa _2^c`$ and $`\kappa _2^o`$). In the pure gravity case, it is clear that $`\kappa _2^c\kappa _2^o`$, and thus there is no evidence for the existence of a new smooth phase. On the other hand, in the case of $`N_V=1`$ with $`N_4=16K`$, we observe the $`\gamma _{st}<0`$ region beyond the usual phase transition point ($`\kappa _2^c`$). We also observe a very obscure transition from $`\gamma _{st}<0`$ to $`\gamma _{st}>0`$ at $`\kappa _2^o`$.
We give the numerical results at Table.1.
In Fig.2, we plot $`\gamma _{st}`$ for various numbers of gauge fields versus $`\kappa _2`$ with volume $`N_4=16K`$. In the case of adding vector fields, we can find that the usual phase-transition point ($`\kappa _2^c`$) is different from another transition point ($`\kappa _2^o`$) which separates the $`\gamma _{st}<0`$ region from the $`\gamma _{st}>0`$ region and $`\gamma _{st}`$ becomes negative at the phase-transition point ($`\kappa _2^c`$). This fact leads to the existence of a new phase. We thus call this intermediate region the smooth phase between these two transition points ($`\kappa _2^c`$ and $`\kappa _2^o`$).
We can consider that simplicial quantum gravity coupled to vector fields has three phases. In Fig.2, we show a schematic phase diagram.
We have three phases in this parameter space: a crumpled phase, a smooth phase (shaded portion) and a branched polymer phase. Furthermore, the smooth phase expands with adding more vector fields. We show the discontinuous phase transition as the thin line and the smooth phase transition as the thick line. In the case of adding the vector fields, there are two separate phase transition lines: the usual phase transition line ($`\kappa _2^c`$) and an obscure phase transition line ($`\kappa _2^o`$). The obscure transition at $`\kappa _2^o`$ has been shown to be third order or cross over, which is very similar to the $`c=1`$ barrier in two-dimensional quantum gravity. In two dimensions the $`c=1`$ barrier is well-known as an obscure transition from the fractal phase ($`c1`$ and $`\gamma _{st}^{2D}<0`$) to the branched polymer phase ($`c>1`$ and $`\gamma _{st}^{2D}>0`$). We consider that the obscure phase transition point may be a threshold point, like the $`c=1`$ case in two dimensions. From reports of Antoniadis et al., the quantum field theory of gravity with conformal invariance has a central charge $`Q^2`$, which has a threshold value, $`Q^28`$. We expect that the obscure phase transition point that is defined by the threshold value, $`\gamma _{st}0`$, may be the same situation as the $`Q^28`$ case. Another phase transition point is the usual phase transition point between the strong coupling region (the crumpled phase) and the intermediate region (the smooth phase). We expect that the phase transition at $`\kappa _2^c`$ is continuous. This leads to the continuum limit of four-dimensional quantum gravity.
Now, let us take a look at the transition at $`\kappa _2^c`$ in the case of $`N_V=1`$. In order to investigate the phase transition, we observe at the exponents of the node susceptibility. The node susceptibility is defined in ref. as follows:
$$\chi =\frac{1}{N_4}\left(<N_0^2><N_0>^2\right).$$
(9)
In Fig.3, we plot the node susceptibility($`\chi `$) as a function of $`\kappa _2`$ with the volume $`N_4=16K,24K`$ and $`32K`$, respectively.
The node susceptibility ($`\chi `$) has a peak value at the critical point ($`\kappa _2^c`$). We find the peak value in each size. The height and the width of the susceptibility peaks give the finite size scaling exponents of the phase transition. The peak value ($`\chi _{max}`$) and the width of peak ($`\delta \kappa _2`$) grow as $`N_4`$ in power. The susceptibility exponents ($`\mathrm{\Delta }`$ and $`\mathrm{\Gamma }`$) are defined by :
$$\chi _{max}N_4^\mathrm{\Delta },(\delta \kappa _2N_4^\mathrm{\Gamma }).$$
(10)
From the numerical results (Fig.3), we obtain the susceptibility exponents: $`\mathrm{\Delta }=0.4(1)`$, $`\mathrm{\Gamma }0.5(3)`$. These values are apparently smaller than 1, though they are 1 in the pure gravity case. This numerical results show that simplicial quantum gravity coupled to a vector field has the different type of phase transition from simplicial quantum gravity in the pure gravity case.
In Fig.4, we show a histogram of $`N_0`$ for a size of $`N_4=32K`$ near to the critical point ($`\kappa _2^c=1.37147(1)`$).
In the pure gravity case, previously, a double peak structure has been found. This structure relates to the latent energy. The fact that the phase transition is first order is shown. However, by adding the vector field, the double peak structure disappears. We thus conclude that the phase transition between the crumpled phase and the smooth phase may be continuous, not first order.
## 4 Scaling property of four-dimensional <br>simplicial quantum gravity
Recent numerical results obtained by dynamical triangulation in four space-time dimensions suggest the existence of the scaling behavior, for example, the MINBU distribution, correlation functions and the boundary volume distributions. If we assume the existence of a correct continuum limit, the scaling property of quantum geometry gives important information about the continuum theory, because some scaling properties are related to the universality of the theory. One of the interesting observes is the fractal dimension (Hausdorff dimension) ($`d_H`$),
$$d_H=\frac{d\mathrm{ln}V^{(4)}(D)}{d\mathrm{ln}D},(V^{(4)}D^{d_H}).$$
(11)
This is based on studying the behavior of the volume $`V^{(4)}(D)`$, the number of simplices, within a geodesic distance $`D`$. The geodesic distance $`D`$ is defined as the shortest distance between two simplices through the center of the simplices. In the pure gravity case, the Hausdorff dimension shows a different behavior in each phase. In the crumpled phase, the Hausdorff dimension diverges ($`d_H\mathrm{}`$). The Hausdorff dimension rises very steeply to a large value below the transition, while it falls very rapidly to a branched polymer, $`d_H=2`$, above the transition. However, in the case of gravity coupled to vector fields, each of three phases shows a different behavior for the Hausdorff dimension. In the crumpled phase and the branched polymer phase, the behavior for the Hausdorff dimension is similar to that of pure gravity. Furthermore, we find that the smooth phase is the intermediate region with $`2<d_H4`$. For the smooth phase, the value of the Hausdorff dimension ($`d_H`$) is changed to smooth, as compared with the pure gravity case. This fact supports the results of a finite size analysis. Especially, at the critical point, we observe the Hausdorff dimension, $`d_H=4.6(2)`$, with $`N_4=32K`$.
Next, let us discuss the scaling structure of four-dimensional DT mfd, at the focusing on the scaling structure of the boundaries in four-dimensional Euclidean space-time using the concept of geodesic distances. We consider that the scaling structure of the boundaries has more informations than the Hausdorff dimension, and that it is the way of directly searching for the structure of quantum geometry.
In the pure gravity case, the scaling property of the boundary volume is discussed in ref.. As in the previous analysis, we assume that the boundary volume distribution ($`\rho (x,D)`$) is a function of a scaling variable in analogy with the loop length distribution function in the two-dimensional model. Fortunately, in the two-dimensional model, the loop length distribution function has been calculated analytically . The loop length distribution function, $`\rho (x=L/D^2,D)`$, which gives the probability of the boundary loop with the loop length ($`L`$) within a geodesic distance ($`D`$), is given as a function of the scaling variable $`x=L/D^2`$:
$$\rho (x=L/D^2,D)=\frac{3}{7\sqrt{\pi }}\frac{1}{D^2}(x^{3/2}+\frac{1}{2}x^{5/2}+\frac{14}{3}x^{1/2})e^x.$$
(12)
This distribution function consists of two different types of distributions. The first two term of eq.(12) represent the baby loops that the universe has a small boundary volume; the last term represents the mother loop that the universe has a large boundary volume. The distribution function, $`\rho (x=L/D^2,D)`$ of eq.(12), satisfies the scaling relation under rescaling ($`DD^{}=\sqrt{\lambda }D,LL^{}=\lambda D`$):
$$\rho (L,D)=\lambda ^1\rho (L^{},D^{}).$$
(13)
In the four-dimensional model, unfortunately, a similar scaling relation about the boundary volume is not yet known. We thus assume the distribution function, $`\rho (x=V/D^\alpha )`$, of the volume of the boundary ($`V`$) within the geodesic distance ($`D`$) in four-dimensional dynamical triangulated manifold. This is a function of the scaling variable, $`x=V/D^\alpha `$, with scaling parameter ($`\alpha `$) in analogy with the two-dimensional model. In Fig.5, we show a schematic picture of our boundary analysis.
First, we consider the scaling structures of these three phases: a crumpled phase, a smooth phase and a branched polymer phase. Actually, in the smooth phase we observe that the distribution ($`\rho `$) becomes fractal in the sense that the sections of the manifold at different distances from a given $`4`$-simplex look exactly the same after a proper rescaling of the boundary volume. Furthermore, the shape of this scaling function is very similar to that of the two-dimensional case. The best account for this excellent agreement in the four-dimensional case can be found in the dominance of a conformal mode and a fractal property. We have also investigated the boundary volume distribution in the crumpled phase and the branched polymer phase. It seems reasonable to suppose that this new smooth phase has a similar fractal structure to that of the two-dimensional fractal surface, and that there is a possibility of taking a continuum limit in the phase. We have also investigated the boundary volume distribution in both the crumpled phase and the branched polymer phase. In the crumpled phase we find that one mother universe is dominant, while in the branched polymer phase there is no evidence for the existence of a mother universe.
Next, let us discuss the relation between the scaling parameter ($`\alpha `$) and the Hausdorff dimension ($`d_H`$). The expectation value of the boundary three-dimensional volume appearing at distance $`D`$ has been introduced in ref.:
$$<V^{(3)}>=\frac{1}{N}_{v_0}^{\mathrm{}}𝑑VV\rho (x=V/D^\alpha ,D),$$
(14)
where $`v_0`$ denotes the UV cut-off of the boundary volume and $`N`$ is the normalization factor. If the boundary volume has the scaling property with the universal distribution ($`\rho (x,D)`$) and $`v_00`$,
$$<V^{(3)}>D^\alpha .$$
(15)
Then, we should obtain a finite fractal dimension,
$$d_f=\alpha +1,$$
(16)
with the fractal dimension $`d_f`$. We measure the volume of the mother boundary as a function of $`D`$. The mother boundary is defined by the boundary having the largest tip volume.
In Fig.6, we plot the mother boundary volume $`<V^{(3)}>`$ with the size of $`N_4=32K`$ at the critical point. As a result, the mother boundary volume shows a scaling, and we obtain the scaling parameter ($`\alpha =3.7(5)`$). Then, we can estimate the fractal dimension ($`d_f=4.7(5)`$). On the other hand, we measured the Hausdorff dimension, which results in $`d_H=4.6(2)`$. Both results are consistent ($`d_fd_H`$). We thus expect that the boundary volume has a scaling property in the sense that the manifold at different distances from a given $`4`$-simplex looks exactly the same after a proper rescaling of the boundary volume.
## 5 Modified balls-in-boxes model
In this section, we discuss a scenario for the phase structure and the phase transition in four-dimensional simplicial quantum gravity coupled to matter fields. In the pure gravity case, the first order phase transition is described by the balls-in-boxes model. This model is considered to be a simple mean field model about simplicial quantum gravity. It describes a fixed number $`N`$ balls distributed into a variable $`M`$ of boxes. The partition function is given by
$$Z_{M,N}=\underset{q_1,\mathrm{},q_M}{}p(q_1)\mathrm{}p(q_M)\delta _{q_1+\mathrm{}+q_M,N},$$
(17)
where $`q_i`$ is the number of balls in box $`i`$, and $`p(q)`$ denotes a weight which depends on only the number of balls (q) in the box. The original balls-in-boxes model has the following correspondence:
Vertex $``$ Box and Simplex $``$ Ball.
Therefore, $`q_i`$ represents the vertex order. The original model has been analyzed in the case of power like weights: $`p(q)=q^\beta `$ (when $`\beta `$ is a parameter for a weight). The system has two distinct phases: a crumpled phase and an elongated phase, and it has a discontinuous phase transition for $`\beta >2`$. It describes the phase structure about four-dimensional pure simplicial quantum gravity. However, this model describes the continuous phase transition at the $`\beta <2`$ . We have noticed this fact, which gives one of the motivations for the modified model of simplicial quantum gravity. In the model for simplicial quantum gravity, we can consider that parameter $`\beta `$ is related to the measure term. If we consider the conformal matter fields (as the conformal charge), parameter $`\beta `$ has the same origin as the conformal anomaly, because they are related to the measure term. The gauge field gives rise to a modified measure factor,
$$M_t=\underset{t}{}o(t)^\beta .$$
(18)
This model shows three distinct phases: a crumpled phase, a crinkled phase (correspond to a smooth phase) and a branched polymer phase. It is consistent with our results. The natural correspondence between our phase structure and that in Ref. is as follows: the transition between the crumpled phase and the branched polymer phase is discontinuous (probably, first-order phase transition), the transition between the crumpled phase and the crinkled phase is continuous, and the transition between the crinkled phase and the branched polymer phase is obscure.
Thus, in order to investigate the phase structure of adding matter fields, we exchange the relation of “vertices-simplices” into that of “triangle-simplices”:
Triangle $``$ Box and Simplex $``$ Ball.
We thus introduce the triangle order ($`o(t)`$) instead of the vertex order ($`q`$). This modification causes the partition function to give rise a constant shift for the constraint:
$$\underset{i:vertex}{}q_i=5N_4,$$
(19)
$$\underset{t:triangle}{}o(t)=10N_4.$$
(20)
We analyzed the modified balls-in-boxes model with using the same method of the both original models.
We thus considered the case of simplicial quantum gravity coupled to matter fields. The partition function is
$`Z_N(\kappa ,\beta ,N_{matter})`$ $`=`$ $`{\displaystyle \underset{m}{\overset{M_{max}}{}}}e^{\kappa m}`$ (21)
$`{\displaystyle \underset{q_1,\mathrm{}q_m}{}}p(q_1)\mathrm{}p(q_m)\delta _{q_1+\mathrm{}+q_m,N}(Z_{mattar}(m))^{N_{mattar}}.`$
Now we discuss the following two cases: (a) the case of coupling to $`N_B`$ copies of the boson fields and (b) the case of coupling to $`N_V`$ copies of the vector fields. (a) In the first case, the matter action is
$$S_{boson}=\underset{(ij)}{}(\varphi _i\varphi _j)^2.$$
(22)
We discuss the effect from the matter. From the perturbation, the Gaussian boson matter gives
$$Z_{boson}=\underset{triangles}{}Constant.$$
(23)
We can estimate the effect to the weight of one-boxes:
$$p(q)Cp(q).$$
(24)
The Gaussian bosons give the effect of multiplication by a constant. From a mean fields analysis, the phase diagram will be changed. However, it is known that the phase diagram is not changed by only a few Gaussian boson fields, according to numerical analysis. We consider that the multiplication by a constant is not enough to change the phase diagram for “real” simplicial quantum gravity, because of the too small effect of multiplication by a constant. (b) In the second case, we consider adding the Gaussian $`U(1)`$ gauge matter. If we use the plaquette action for the gauge fields, the partition function is written as
$$Z_{gauge}=\underset{triangle}{}f(o(t)),$$
(25)
where $`f`$ is a function of the triangle order ($`o(t)`$). From the perturbation about the gauge interaction, $`f(o(t))=C_0+C_1o(t)`$. Then, the partition function is replaced, as follows:
$$Z_{gauge}\underset{t}{}o(t).$$
(26)
This is just the modified measure. We estimate the effect from the gauge matter fields as
$$\beta (\beta 1).$$
(27)
That is to say, the Gaussian gauge matter fields lower $`\beta `$.
## 6 Summary and Discussion
We have investigated the phase structure and the phase transition with a model of four-dimensional simplicial quantum gravity coupled to $`U(1)`$ gauge fields.
The results of our study are summarized by the schematic phase diagram in Fig.2. We checked this phase diagram in the case of $`N_V=1,2,3`$ at a volume of $`N_4=16K`$. We found three phases in this parameter space: a crumpled phase ($`\gamma _{st}\mathrm{}`$, $`d_f\mathrm{}`$), a smooth phase ($`\gamma _{st}<0`$, $`2<d_f4`$, $`N_0/N_4<0.25`$) and a branched polymer phase ($`\gamma _{st}>0`$, $`d_f=2`$, $`N_0/N_4=0.25`$). The thin line denotes a discontinuous phase-transition line which is known in pure gravity; moreover, the thick line denotes a smooth phase-transition line. As contrasted with the phase diagram of pure gravity, the phase diagram means richer structures. In the crumpled phase one can find singular sub-simplices, for example, vertex order concentration and link order concentration. The smooth phase is defined as a region between the critical point ($`\kappa _2^c`$) and the obscure phase transition point ($`\kappa _2^o`$). We observe the negative string susceptibility in this region. Then, with $`N_4=16K`$ ($`\kappa _2=1.7`$) and $`N_V=1`$, we obtained $`\gamma _{st}=0.38(5)`$, $`d_f=2.8(5)`$ and a good scaling relation of the boundary volume distributions. We consider that the scaling structure of this smooth phase is similar to that of a two-dimensional random (fractal) surface. This smooth phase is slowly broken to a branched polymer phase that has no mother structure at the obscure phase transition ($`\kappa _2^o`$). We obtained an obscure-transition line (a broken line in Fig.2); therefore, we suggest that the obscure-transition corresponds to the $`c=1`$ barrier in two-dimensional quantum gravity.
As for the phase transition at the critical point ($`\kappa _2=\kappa _2^c`$), we showed the finite size scaling at the critical point and the histogram of the node. We calculated some finite size scaling exponents, and showed that the value of these exponents is smaller than 1. It has been discussed that the value of 1 is expected to be a first order phase transition for the results of pure gravity. However, in the case of adding one gauge matter field ($`N_V=1`$), the numerical results show that the phase transition is smooth, in the contrast to pure gravity, and then that a double peak structure of the node histogram disappears.
Furthermore, in order to investigate the property of quantum geometry and to discuss the universality of the manifold, we considered the scaling property of the boundary volume. In the smooth phase with $`N_4=16K`$ ($`\kappa _2=1.7`$) and $`N_V=1`$, we obtained $`d_f=2.8(5)`$ and a good scaling relation of the boundary volume distributions. The scaling structure of this smooth phase is similar to that of a two-dimensional random (fractal) surface. This suggests the existence of a new smooth phase in four-dimensional simplicial gravity. The other two phases have a similar scaling property to that of pure gravity. We have shown the scaling property of the mother boundary, where the scaling parameter is consistent with the Hausdorff dimension. Then, the boundary volume has a scaling property with the scaling variable ($`x=V/D^{d_f1}`$). We expect that the boundaries have a fractal structure and universality of the scaling relations. We also discussed the modification of the balls-in-boxes model. The role of a vertex is exchanged into a triangle. This clarifies the relation between the measure factor of the numerical model and that of the analytical mean field model.
We expect the existence of genuine four-dimensional quantum gravity on the critical point ($`\kappa _2^c`$) with the vector fields. Our recent investigations will give further evidence for the existence of an ultraviolet fixed point of the quantum field theory of gravity.
## 7 Acknowledgments
We would like to thank H.Kawai, N.Ishibashi, K.Hamada and F.Sugino for fruitful discussions. We are also grateful to members of the KEK-IPNS theory group. The numerical calculations were performed using the originally designed cluster computer for the study about quantum gravity and strings; CCGS-01 ATROPOS(Tokai University), CCGS-02 EST and CCGS-03 LACHESIS(KEK).
|
warning/0004/astro-ph0004257.html
|
ar5iv
|
text
|
# On the pair-electromagnetic pulse from an electromagnetic Black Hole surrounded by a Baryonic Remnant
## 1 Introduction
That vacuum polarization process à la Heisenberg-Euler-Schwinger (Heisenberg & Euler (1931), Schwinger (1951)) can occur in the field of a Kerr Newmann EMBH and that they naturally lead to a model for gamma-ray bursts was pointed out in Damour & Ruffini (1975). The basic energetics of the process, governed by the Christodoulou-Ruffini mass formula, for an EMBH gives as shown in Christodoulou & Ruffini (1971),
$`E^2`$ $`=`$ $`M^2c^4=\left(M_{\mathrm{ir}}c^2+{\displaystyle \frac{Q^2}{2\rho _+}}\right)^2+{\displaystyle \frac{L^2c^2}{\rho _+^2}},`$ (1)
$`S`$ $`=`$ $`4\pi \rho _+^2=4\pi (r_+^2+{\displaystyle \frac{L^2}{c^2M^2}})=16\pi \left({\displaystyle \frac{G^2}{c^4}}\right)M_{\mathrm{ir}}^2,`$ (2)
with
$$\frac{1}{\rho _+^4}\left(\frac{G^2}{c^8}\right)\left(Q^4+4L^2c^2\right)1,$$
(3)
where $`M,L`$ and $`Q`$ are respectively the mass energy, the angular momentum and the charge of the EMBH and $`M_{\mathrm{ir}}`$ is the irreducible mass, $`r_+`$ is the horizon radius, $`\rho _+`$ is the quasi-spheroidal cylindrical coordinate of the horizon evaluated at the equatorial plane and $`S`$ is the horizon surface area. Extreme black holes satisfy the equality in Eq.(3). The vacuum polarization process being “reversible” transformations in the sense of Christodoulou & Ruffini (1971) can extract an energy up to 29% of the mass-energy of an extremal rotating black hole and up to 50% of the mass-energy of an extremely magnetized and charged black hole.
Although in general such a process is endowed with axial symmetry, in order to clarify the pure interplay of the gravitational and electrodynamical phenomena and also for simplicity, we have examined (Ruffini (1998) and Preparata et al. 1998a,b ) the limiting cases of such a process in the field of a Reissner-Nordstrom geometry whose spherically symmetric metric is given by
$$d^2s=g_{tt}(r)d^2t+g_{rr}(r)d^2r+r^2d^2\theta +r^2\mathrm{sin}^2\theta d^2\varphi ,$$
(4)
where $`g_{tt}(r)=\left[1\frac{2GM}{c^2r}+\frac{Q^2G}{c^4r^2}\right]\alpha ^2(r)`$ and $`g_{rr}(r)=\alpha ^2(r)`$. The dyadosphere, defined (see Fig. 1 of Preparata et al. 1998a,b ) as the region where the electric field exceeds the critical value, $`_\mathrm{c}=\frac{m^2c^3}{\mathrm{}e}`$ (Heisenberg & Euler (1931), Schwinger (1951)), where $`m`$ and $`e`$ are the mass and charge of the electron, extends between the horizon radius
$`r_+`$ $`=`$ $`1.4710^5\mu (1+\sqrt{1\xi ^2})\mathrm{cm}.`$ (5)
where we have introduced the dimensionless mass and charge parameters $`\mu =\frac{M}{M_{}}`$, $`\xi =\frac{Q}{(M\sqrt{G})}1`$, and the outer radius
$`r_{\mathrm{ds}}`$ $`=`$ $`1.1210^8\sqrt{\mu \xi }\mathrm{cm}.`$ (6)
Consequently the local number density of electron and positron pairs created in the dyadosphere as a function of radius
$$n_{e^+e^{}}(r)=\frac{Q}{4\pi r^2\left(\frac{\mathrm{}}{mc}\right)e}\left[1\left(\frac{r}{r_{\mathrm{ds}}}\right)^2\right],$$
(7)
and their energy density is given by
$$ϵ(r)=\frac{Q^2}{8\pi r^4}\left(1\left(\frac{r}{r_{\mathrm{ds}}}\right)^4\right),$$
(8)
(see Figs. & of Preparata et al. 1998a,b ). The total energy of pairs converted from the static electric energy and deposited within the dyadosphere is then
$$E_{\mathrm{dya}}=\frac{1}{2}\frac{Q^2}{r_+}(1\frac{r_+}{r_{\mathrm{ds}}})\left[1\left(\frac{r_+}{r_{\mathrm{ds}}}\right)^2\right],$$
(9)
and the total number of electron and position pairs in the dyadosphere is
$$N_{e^+e^{}}^{}\frac{QQ_c}{e}\left[1+\frac{(r_{\mathrm{ds}}r_+)}{\frac{\mathrm{}}{mc}}\right],$$
(10)
where $`Q_c=_\mathrm{c}r_+^2`$ (see Preparata et al. 1998a,b ). In the limit of $`\frac{r_+}{r_{\mathrm{ds}}}0`$, Eq.(9) leads to $`E_{\mathrm{dya}}\frac{1}{2}\frac{Q^2}{r_+}`$, which coincides with the energy extractable from EMBHs by reversible processes ($`M_{\mathrm{ir}}=\mathrm{const}.`$), namely $`EM_{\mathrm{ir}}=\frac{1}{2}\frac{Q^2}{r_+}`$(see Fig. 4 of Preparata et al. 1998a,b ).
Due to the very large pair density given by Eq.(7) and to the sizes of the cross-sections for the process $`e^+e^{}\gamma +\gamma `$, the system is expected to thermalize to a plasma configuration for which
$$N_{e^+}=N_e^{}N_\gamma N_{e^+e^{}}^{},$$
(11)
where $`N_{e^+e^{}}^{}`$ is the number of $`e^+e^{}`$-pairs created in the dyadosphere(see Preparata et al. 1998a,b ). These are the initial conditions for the evolution of the dyadosphere. In a previous paper (Ruffini et al. (1999)) we presented the temporal evolution of the dyadosphere in vacuum giving origin to an extremely sharply defined and extremely relativistic expanding pulse of pair and electromagnetical radiation of a constant length in the laboratory frame: the PEM pulse. In this paper we present results of the collision of the PEM pulse with a remnant of baryonic matter surrounding the just formed black hole. We assume the PEM pulse to collide with a shell of baryonic matter of constant density and at a radius of the order of 100 times the radius of the dyadosphere $`r_{\mathrm{ds}}`$ Eq.(6). The shells have this thickness of the order of 10 times $`r_{\mathrm{ds}}`$. The mass-energies of the shells are taken to be $`10^810^2`$ of the total energy of the dyadosphere.
The work contains the following sections: Sect. 2 presents discussions of the hydrodynamical equations of a PEM pulse interacting with the baryonic shell; Sect. 3 defines the parameters of the baryonic shells, the behaviour of the PEM pulse colliding with a baryon shell as well as before and after the collision are presented; in Sect. 4 the results of a numerical calculation solving the hydrodynamical equations of Sect. 2 are compared to the results of the analytical model of slab approximation for selected parameters.
## 2 General relativistic hydrodynamical equations for PEM pulse in presence of baryonic matter
The evolution of the PEM pulse in vacuum was treated in a previous work (Ruffini et al. (1999)). We here generalize that treatment to the case in which baryonic matter is present and we outline the relevant relativistic hydrodynamic equations. As in the previous treatment (Ruffini et al. (1999)), we assume the plasma fluid of $`e^+e^{}`$-pairs, photons and baryonic matter to be described by the stress-energy tensor
$$T^{\mu \nu }=pg^{\mu \nu }+(p+\rho )U^\mu U^\nu $$
(12)
where $`\rho `$ and $`p`$ are respectively the total proper energy density and pressure in the comoving frame of the plasma fluid. The $`U^\mu `$ is the four-velocity of the plasma fluid. We have
$$g_{tt}(U^t)^2+g_{rr}(U^r)^2=1,$$
(13)
where $`U^r`$ and $`U^t`$ are the radial and temporal contravariant components of the 4-velocity. The conservation law of baryon number can be expressed as a function of the proper baryon number density $`n_B`$
$`(n_BU^\mu )_{;\mu }`$ $`=`$ $`g^{\frac{1}{2}}(g^{\frac{1}{2}}n_BU^\nu )_{,\nu }`$ (14)
$`=`$ $`(n_BU^t)_{,t}+{\displaystyle \frac{1}{r^2}}(r^2n_BU^r)_{,r}=0.`$
The radial component of the energy-momentum conservation law of the plasma fluid reduces to
$`{\displaystyle \frac{p}{r}}+{\displaystyle \frac{}{t}}\left((p+\rho )U^tU_r\right)+{\displaystyle \frac{1}{r^2}}{\displaystyle \frac{}{r}}\left(r^2(p+\rho )U^rU_r\right)`$
$`{\displaystyle \frac{1}{2}}(p+\rho )\left[{\displaystyle \frac{g_{tt}}{r}}(U^t)^2+{\displaystyle \frac{g_{rr}}{r}}(U^r)^2\right]=0.`$ (15)
The component of the energy-momentum conservation law of the plasma fluid equation along a flow line is
$`U_\mu (T^{\mu \nu })_{;\nu }`$ $`=`$ $`(\rho U^\nu )_{;\nu }p(U^\nu )_{;\nu },`$ (16)
$`=`$ $`g^{\frac{1}{2}}(g^{\frac{1}{2}}\rho U^\nu )_{,\nu }pg^{\frac{1}{2}}(g^{\frac{1}{2}}U^\nu )_{,\nu }`$
$`=`$ $`(\rho U^t)_{,t}+{\displaystyle \frac{1}{r^2}}(r^2\rho U^r)_{,r}`$
$`+`$ $`p\left[(U^t)_{,t}+{\displaystyle \frac{1}{r^2}}(r^2U^r)_{,r}\right]=0.`$
Defining the total proper internal energy density $`ϵ`$ and the baryonic mass density $`\rho _B`$ in the comoving frame of the plasma fluid,
$$ϵ\rho \rho _B,\rho _Bn_Bmc^2,$$
(17)
and using the law of baryon-number conservation (14), from Eq. (16) we have
$$(ϵU^\nu )_{;\nu }+p(U^\nu )_{;\nu }=0.$$
(18)
Recalling that $`\frac{dV}{d\tau }=V(U^\mu )_{;\mu }`$, where $`V`$ is the comoving volume and $`\tau `$ is the proper time for the plasma fluid, we have along each flow line
$$\frac{d(Vϵ)}{d\tau }+p\frac{dV}{d\tau }=\frac{dE}{d\tau }+p\frac{dV}{d\tau }=0,$$
(19)
where $`E=Vϵ`$ is total proper internal energy of the plasma fluid. We represent the equation of state by the introduction of a thermal index $`\mathrm{\Gamma }(\rho ,T)`$
$$\mathrm{\Gamma }=1+\frac{p}{ϵ}.$$
(20)
We now turn to the analysis of $`e^+e^{}`$ pairs initially created in the dyadosphere and ionized electrons contained in the baryonic matter. Let $`n_e^{}`$ and $`n_{e^+}`$ be the proper number densities of electrons and positrons associated to pairs created, and $`n_e^{}^b`$ the proper number densities of ionized electrons, we clearly have
$$n_e^{}=n_{e^+}=n_{\mathrm{pair}},n_e^{}^b=\overline{Z}n_B,$$
(21)
where $`n_{\mathrm{pair}}`$ is the number of $`e^+e^{}`$ pairs and $`\overline{Z}`$ the average atomic number $`\frac{1}{2}<\overline{Z}<1`$ ($`\overline{Z}=1`$ for hydrogen atom and $`\overline{Z}=\frac{1}{2}`$ for general baryonic matter). The rate equation for electrons and positrons gives,
$`(n_{e^+}U^\mu )_{;\mu }`$ $`=`$ $`(n_{e^+}U^t)_{,t}+{\displaystyle \frac{1}{r^2}}(r^2n_{e^+}U^r)_{,r}`$ (22)
$`=`$ $`\overline{\sigma v}[(n_e^{}(T)+n_e^{}^b(T))n_{e^+}(T)`$
$``$ $`(n_e^{}+n_e^{}^b)n_{e^+}],`$
$`(n_e^{}U^\mu )_{;\mu }`$ $`=`$ $`(n_e^{}U^t)_{,t}+{\displaystyle \frac{1}{r^2}}(r^2n_e^{}U^r)_{,r}`$ (23)
$`=`$ $`\overline{\sigma v}\left[n_e^{}(T)n_{e^+}(T)n_e^{}n_{e^+}\right],`$
$`(n_e^{}^bU^\mu )_{;\mu }`$ $`=`$ $`(n_e^{}^bU^t)_{,t}+{\displaystyle \frac{1}{r^2}}(r^2n_e^{}^bU^r)_{,r}`$ (24)
$`=`$ $`\overline{\sigma v}\left[n_e^{}^b(T)n_{e^+}(T)n_e^{}^bn_{e^+}\right],`$
where $`\overline{\sigma v}`$ is the mean of the product of the annihilation cross-section and the thermal velocity of the electrons and positrons, $`n_{e^\pm }(T)`$ are the proper number densities of electrons and positrons associated to the pairs, given by appropriate Fermi integrals with zero chemical potential, and $`n_e^{}^b(T)`$ is the proper number density of ionized electrons, given by appropriate Fermi integrals with non-zero chemical potential $`\mu _e`$, at an appropriate equilibrium temperature $`T`$. These rate equations can be reduced to
$`(n_{e^\pm }U^\mu )_{;\mu }`$ $`=`$ $`(n_{e^\pm }U^t)_{,t}+{\displaystyle \frac{1}{r^2}}(r^2n_{e^\pm }U^r)_{,r}`$ (25)
$`=`$ $`\overline{\sigma v}\left[n_e^{}(T)n_{e^+}(T)n_e^{}n_{e^+}\right],`$
$`(n_e^{}^bU^\mu )_{;\mu }`$ $`=`$ $`(n_e^{}^bU^t)_{,t}+{\displaystyle \frac{1}{r^2}}(r^2n_e^{}^bU^r)_{,r}=0,`$ (26)
$`Frac`$ $``$ $`{\displaystyle \frac{n_{e^\pm }}{n_{e^\pm }(T)}}={\displaystyle \frac{n_e^{}^b(T)}{n_e^{}^b}}.`$ (27)
Equation (26) is just the baryon-number conservation law (14) and (27) is a relationship obeyed $`n_{e^\pm },n_{e^\pm }(T)`$ and $`n_e^{}^b,n_e^{}^b(T)`$.
The equilibrium temperature $`T`$ is determined by the thermalization processes occurring in the expanding plasma fluid with a total proper energy density $`\rho `$, governed by the hydrodynamical equations (14,15,16). We have
$$\rho =\rho _\gamma +\rho _{e^+}+\rho _e^{}+\rho _e^{}^b+\rho _B,$$
(28)
where $`\rho _\gamma `$ is the photon energy density, $`\rho _Bm_Bc^2n_B`$ is the baryonic mass density which is considered to be non relativistic in the range of temperature $`T`$ under consideration, and $`\rho _{e^\pm }`$ is the proper energy density of electrons and positrons pairs given by,
$$\rho _{e^\pm }=\frac{n_{e^\pm }}{n_{e^\pm }(T)}\rho _{e^\pm }(T),$$
(29)
where $`n_{e^\pm }`$ is obtained by integration of Eq.(25) and $`\rho _{e^\pm }(T)`$ is the proper energy density of electrons(positrons) obtained from zero chemical potential Fermi integrals at the equilibrium temperature $`T`$. Whereas, $`\rho _e^{}^b`$ is the energy density of the ionized electrons, obtained by the ionization of baryonic matter,
$$\rho _e^{}^b=\frac{n_e^{}^b}{n_e^{}^b(T)}\rho _e^{}^b(T),$$
(30)
where $`n_e^{}^b`$ is obtained by integration of Eq.(26) and $`\rho _e^{}(T)`$ is the proper energy density of ionized electrons obtained from an appropriate Fermi integral of non-zero chemical potential $`\mu _e`$ at the equilibrium temperature $`T`$.
Having intrinsically defined the equilibrium temperature $`T`$ in Eq.(28), we can also, analogously, evaluate the total pressure
$$p=p_\gamma +p_{e^+}+p_e^{}+p_e^{}^b+p_B,$$
(31)
where $`p_\gamma `$ is the photon pressure, $`p_{e^\pm }`$ and $`p_e^{}^b`$ given by,
$`p_{e^\pm }`$ $`=`$ $`{\displaystyle \frac{n_{e^\pm }}{n_{e^\pm }(T)}}p_{e^\pm }(T),`$ (32)
$`p_e^{}^b`$ $`=`$ $`{\displaystyle \frac{n_e^{}^b}{n_e^{}^b(T)}}p_e^{}^b(T),`$ (33)
where the pressures $`p_{e^\pm }(T)`$ are determined by zero chemical potential Fermi integrals, and $`p_e^{}^b(T)`$ is the pressure of the ionized electrons, evaluated by an appropriate Fermi integral of non-zero chemical potential $`\mu _e`$ at the equilibrium temperature $`T`$. In Eq.(31), the ions pressure $`p_B`$ is negligible by comparison with the pressures $`p_{\gamma ,e^\pm ,e^{}}(T)`$, since baryons and ions are expected to be non-relativistic in the range of temperature $`T`$ under consideration. Finally, using Eqs.(28,31), we compute the thermal factor $`\mathrm{\Gamma }`$ of the equation of state (20). The calculation is continued as the plasma fluid expands, cools and the $`e^+e^{}`$ pairs recombine, until it becomes optically thin:
$$_R𝑑r(n_{e^\pm }+\overline{Z}n_B)\sigma _TO(1),$$
(34)
where $`\sigma _T=0.66510^{24}\mathrm{cm}^2`$ is the Thomson cross-section and the integration is over the radial interval of the PEM-pulse in the comoving frame. At this point the energy is virtually entirely in the form of free-streaming photons and the calculation is stopped.
## 3 The quasi-analytical simplified model based on the constant-thickness approximation
The PEM pulse expansion in the absence of baryonic matter has been discussed in a previous paper (Ruffini et al. (1999)) where the quasi-analytical approach of an expanding shell of constant thickness in the laboratory frame was adopted and validated by comparison with the numerical integration of the general relativistic hydrodynamical equations. We here generalize these results by examining the collision of the PEM pulse with baryonic matter and adopting the constant-thickness approximation both for the description of the collision and the further expansion of the PEM pulse by a simplified approach to the system of equations outlined in the previous paragraph.
We first recall the main results of the PEM pulse expanding in vacuum: we indicate by $`U(r)=U_r=\mathrm{const}.`$ the four-velocity of the slab and by $`𝒟=r_{\mathrm{ds}}r_+`$ the constant width of the slab in the laboratory frame of the plasma fluid, the average bulk relativistic gamma-factor $`\overline{\gamma }`$ is,
$$\overline{\gamma }=\sqrt{1+U_r^2},V_r=\frac{U_r}{\overline{\gamma }}.$$
(35)
The evolution of the slab is governed by the total energy and entropy conservations, which are cast into the following equations as a function of the coordinate volume of the plasma fluid expanding from $`𝒱_{}`$ to $`𝒱`$,
$`{\displaystyle \frac{\overline{ϵ}_{}}{\overline{ϵ}}}`$ $`=`$ $`\left({\displaystyle \frac{V}{V_{}}}\right)^\mathrm{\Gamma }=\left({\displaystyle \frac{𝒱}{𝒱_{}}}\right)^\mathrm{\Gamma }\left({\displaystyle \frac{\overline{\gamma }}{\overline{\gamma }_{}}}\right)^\mathrm{\Gamma },`$ (36)
$`\overline{\gamma }`$ $`=`$ $`\overline{\gamma }_{}\sqrt{{\displaystyle \frac{\overline{ϵ}_{}𝒱_{}}{\overline{ϵ}𝒱}}},`$ (37)
$`{\displaystyle \frac{}{t}}(N_{e^\pm })`$ $`=`$ $`N_{e^\pm }{\displaystyle \frac{1}{𝒱}}{\displaystyle \frac{𝒱}{t}}+\overline{\sigma v}{\displaystyle \frac{1}{\overline{\gamma }^2}}(N_{e^\pm }^2(T)N_{e^\pm }^2),`$ (38)
where the proper volume $`V`$ of the plasma fluid $`V=\overline{\gamma }𝒱`$ and the thermal index $`\mathrm{\Gamma }`$ Eq.(20), a slowly-varying function of the state with values around 4/3, has been approximately assumed to be constant. The coordinate number density of $`e^\pm `$-pairs in equilibrium is $`N_{e^\pm }(T)\overline{\gamma }n_{e^\pm }(T)`$ and the coordinate number density of $`e^\pm `$-pairs $`N_{e^\pm }\overline{\gamma }n_{e^\pm }`$. These equations have already been numerically integrated (Ruffini et al. (1999)).
The baryonic matter remnant is assumed to be distributed well outside the dyadosphere in a shell of thickness $`\mathrm{\Delta }`$ between an inner radius $`r_{\mathrm{in}}`$ and an outer radius $`r_{\mathrm{out}}`$ at a distance from the EMBH at which the original PEM pulse expanding in vacuum has not yet reached transparency,
$$r_{\mathrm{in}}=100r_{\mathrm{ds}},\mathrm{\Delta }=10r_{\mathrm{ds}},r_{\mathrm{out}}=r_{\mathrm{in}}+\mathrm{\Delta }.$$
(39)
The total baryonic mass ($`M_B=N_Bm_pc^2`$) is assumed to be a fraction of the dyadosphere initial total energy $`(E_{\mathrm{dya}})`$. The total baryon-number ($`N_B`$) is then given by
$$N_B=B\frac{E_{\mathrm{dya}}}{m_pc^2}.$$
(40)
where $`B`$ is a parameter in the range $`10^810^2`$ and where $`m_p`$ is the proton mass. The baryon number density $`n_B^{}`$ is assumed to be a constant,
$$\overline{n}_B^{}=\frac{N_B}{V_B},\overline{\rho }_B^{}=m_p\overline{n}_B^{}.$$
(41)
As the PEM pulse reaches the region $`r_{\mathrm{in}}<r<r_{\mathrm{out}}`$, it interacts with the baryonic matter which is assumed to be at rest. In our simplified quasi-analytic model we make the following assumptions to describe this interaction:
* the PEM pulse does not change its geometry during the interaction;
* the collision between the PEM pulse and the baryonic matter is assumed to be inelastic,
* the baryonic matter reaches thermal equilibrium with the photons and pairs of the PEM pulse.
These assumptions are valid if: (i) the total energy of the PEM pulse is much larger than the total mass-energy of baryonic matter $`M_B`$, $`10^8<E_{\mathrm{dya}}/M_B<10^2`$, and (ii) the comoving number density ratio $`n_{e^+e^{}}/n_B^{}`$ of pairs and baryons at the moment of collision is extremely high (e.g., $`10^6<n_{e^+e^{}}/n_B^{}<10^{12}`$, and (iii) the PEM pulse has a large value of Lorentz factor ($`600<\overline{\gamma }<4000`$).
In the collision process between the PEM pulse and the baryonic matter at $`r_{\mathrm{out}}>r>r_{\mathrm{in}}`$ , we impose the total energy and momentum conservations. We consider the collision process between two radii $`r_2,r_1`$, satisfying $`r_{\mathrm{out}}>r_2>r_1>r_{\mathrm{in}}`$ and $`r_2r_1\mathrm{\Delta }`$. The amount of baryonic mass acquired by the PEM pulse is
$$\mathrm{\Delta }M=\frac{M_B}{V_B}\frac{4\pi }{3}(r_2^3r_1^3),$$
(42)
where $`M_B/V_B`$ is the mean-density of baryonic matter at rest. The conservation of total energy leads to the estimate of the corresponding quantities before (with “$``$”) and after such a collision
$$(\mathrm{\Gamma }\overline{ϵ}_{}+\overline{\rho }_B^{})\overline{\gamma }_{}^2𝒱_{}+\mathrm{\Delta }M=(\mathrm{\Gamma }\overline{ϵ}+\overline{\rho }_B+\frac{\mathrm{\Delta }M}{V}+\mathrm{\Gamma }\mathrm{\Delta }\overline{ϵ})\overline{\gamma }^2𝒱,$$
(43)
where $`\mathrm{\Delta }\overline{ϵ}`$ is the corresponding increase of internal energy due to the collision. Similarly the momentum-conservation gives
$$(\mathrm{\Gamma }\overline{ϵ}_{}+\overline{\rho }_B^{})\overline{\gamma }_{}U_r^{}𝒱_{}=(\mathrm{\Gamma }\overline{ϵ}+\overline{\rho }_B+\frac{\mathrm{\Delta }M}{V}+\mathrm{\Gamma }\mathrm{\Delta }\overline{ϵ})\overline{\gamma }U_r𝒱,$$
(44)
where radial component of the four-velocities of the PEM pulse $`U_r^{}=\sqrt{\overline{\gamma }_{}^21}`$ and $`\mathrm{\Gamma }`$ thermal index. We then find
$`\mathrm{\Delta }\overline{ϵ}`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{\Gamma }}}\left[(\mathrm{\Gamma }\overline{ϵ}_{}+\overline{\rho }_B^{}){\displaystyle \frac{\overline{\gamma }_{}U_r^{}𝒱_{}}{\overline{\gamma }U_r𝒱}}(\mathrm{\Gamma }\overline{ϵ}+\overline{\rho }_B+{\displaystyle \frac{\mathrm{\Delta }M}{V}})\right],`$ (45)
$`\overline{\gamma }`$ $`=`$ $`{\displaystyle \frac{a}{\sqrt{a^21}}},a{\displaystyle \frac{\overline{\gamma }_{}}{U_r^{}}}+{\displaystyle \frac{\mathrm{\Delta }M}{(\mathrm{\Gamma }\overline{ϵ}_{}+\overline{\rho }_B^{})\overline{\gamma }_{}U_r^{}𝒱_{}}}.`$ (46)
These equations determine the gamma-factor $`\overline{\gamma }`$ and the internal energy density $`\overline{ϵ}=\overline{ϵ}_{}+\mathrm{\Delta }\overline{ϵ}`$ in the capture process of baryonic matter by the PEM pulse. After collision ($`r>r_{\mathrm{out}}`$), the further adiabatic expansion of PEM pulse is described by the total baryon number, energy and entropy conservations, i.e., the following hydrodynamical equations which generalize those derived in our previous paper (Ruffini et al. (1999)) with $`\rho _B0`$,
$`{\displaystyle \frac{\overline{n}_B^{}}{\overline{n}_B}}`$ $`=`$ $`{\displaystyle \frac{V}{V_{}}}={\displaystyle \frac{𝒱\overline{\gamma }}{𝒱_{}\overline{\gamma }_{}}},`$ (47)
$`{\displaystyle \frac{\overline{ϵ}_{}}{\overline{ϵ}}}`$ $`=`$ $`\left({\displaystyle \frac{V}{V_{}}}\right)^\mathrm{\Gamma }=\left({\displaystyle \frac{𝒱}{𝒱_{}}}\right)^\mathrm{\Gamma }\left({\displaystyle \frac{\overline{\gamma }}{\overline{\gamma }_{}}}\right)^\mathrm{\Gamma },`$ (48)
$`\overline{\gamma }`$ $`=`$ $`\overline{\gamma }_{}\sqrt{{\displaystyle \frac{(\mathrm{\Gamma }\overline{ϵ}_{}+\rho _B^{})𝒱_{}}{(\mathrm{\Gamma }\overline{ϵ}+\overline{\rho }_B)𝒱}}},`$ (49)
$`{\displaystyle \frac{}{t}}(N_{e^\pm })`$ $`=`$ $`N_{e^\pm }{\displaystyle \frac{1}{𝒱}}{\displaystyle \frac{𝒱}{t}}+\overline{\sigma v}{\displaystyle \frac{1}{\overline{\gamma }^2}}(N_{e^\pm }^2(T)N_{e^\pm }^2).`$ (50)
In these equations ($`r>r_{\mathrm{out}}`$) the comoving baryonic mass- and number densities are $`\overline{\rho }_B=M_B/V`$ and $`\overline{n}_B=N_B/V`$, where $`V`$ is the comoving volume of the PEM pulse. The integration is continued untill the transparency condition in Eq.(34) is reached.
## 4 Validation of the constant-thickness approximation
The numerical integration of the general relativistic equations given in Sect. 2 have already been presented in a series of papers (see Wilson 1975, 1977, Wilson et al. (1996)). We can then proceed to compare and contrast the results obtained by the numerical integration of the Eqs.(14)-(16) and the simplified quasi-analytical approach given by Eq.(36)-(38), (43)-(44) and (47)-(50). We select the specific example of an EMBH of $`10^3M_{}`$ with a charge to mass ratio $`\xi =0.1`$. From the equations already published (Preparata et al. 1998a,b ), the total energy in the dyadosphere is $`3.0910^{54}`$ergs. The PEM pulse is assumed to collide with a baryonic shell of thickness $`\mathrm{\Delta }=10r_{\mathrm{ds}}`$ at rest at a radius of $`100r_{\mathrm{ds}}`$. We have considered five different cases: (1) a shell of baryonic mass $`2.2310^4M_{}`$ ($`410^{50}`$ ergs rest energy), from corresponding to a parameter $`B1.310^4`$; (2) $`6.710^4M_{}`$ ($`1.210^{51}`$ ergs, $`B3.810^4`$); (3) $`2.210^3M_{}`$ ($`410^{51}`$ ergs, $`B1.310^3`$); (4) $`6.710^3M_{}`$ ($`1.210^{52}`$ ergs, $`B3.810^3`$); (5) $`2.210^2M_{}`$ ($`410^{52}`$ ergs, $`B1.310^3`$). The collision between the expanding slab and the baryonic matter shell is treated as an inelastic collision in both calculations.
We first proceed to a qualitative analysis of the evolution of the PEM pulse in Eqs. (14)-(16) from Figs. 1,2 corresponding to $`B1.310^4`$ and Figs. 3,4,5 corresponding to $`B3.810^4,1.310^3,3.810^3`$, we see that the PEM pulse after collision with the baryonic matter shell continues to expand as a one dimensional slab. We see, however, from Figs. 6 corresponding to $`B1.310^2`$ that the expansion after collision becomes much more complex and the constant-thickness approximation ceases to be valid. From this qualitative analysis we now proceed to a quantitative analysis in Fig. 7. We compare and contrast the bulk gamma factor as computed from the constant-thickness approximation and the one from the full set of Eqs. (14)-(16). The computation refers to the case $`B1.310^4`$ where an excellent qualitative agreement with the one-dimensional-slab approximation has been found. The extremely good agreement validates the constant-thickness approximation. This agreement has been found up to values of $`B`$ no larger than $`10^2`$.
## 5 Considerations on the GRB structures descending from a constant-thickness approximation
We now proceed to some specific prediction of GRB features computed by using the constant-thickness approximation and the Eqs. (36)-(38), (43)-(44) and (47)-(50) in the range of validation of this approximation just defined in the previous paragraph. As an example for clearly showing the evolution of PEM pulses colliding with baryonic matter, we take the following black hole mass and charge, as well as the mass of baryon remnant as a typical case:
$$M_{\mathrm{BH}}=10^3M_{},\xi =0.1,M_B=10^2E_{\mathrm{dya}},$$
(51)
where the total energy of dyadosphere $`E_{\mathrm{dya}}=3.0910^{54}`$ergs, so the total number of $`e^+e^{}`$-pairs created in the dyadosphere (given by Eq.(10)) $`N_{e^+e^{}}^{}=1.910^{60}`$, and baryonic mass $`M_B=1.7310^2M_{}`$. This baryonic mass is close to the limit of validation of the slab model shown in Sect. 4. We have assumed the baryonic matter at a distance of $`r_{\mathrm{in}}=100r_{\mathrm{ds}}`$, very close to the transparency condition of the PEM pulse in vacuum (see Ruffini et al. (1999)).
In Fig. 8 we represent the Lorentz Factor of the PEM pulse as a function of the radius for collision with different amounts of baryonic matter, corresponding respectively to $`B=10^2`$, $`B=10^3`$ and $`B=10^4`$. The diagram extends to values of the radial coordinate at which the transparency condition given by Eq.(34) is reached. Also represented, for each diagram, is the “asymptotic” Lorentz Factor:
$$\overline{\gamma }_{\mathrm{asym}}\frac{E_{\mathrm{dya}}}{M_Bc^2}.$$
(52)
The closer the $`\overline{\gamma }`$ value approaches, at transparency, the “asymptotic” value (52), the smaller the intensity of the radiation emitted in the burst, and the larger the amount of kinetic energy left in the baryonic matter. This point is further clarified in Fig. 9, where are plotted the $`\overline{\gamma }`$-factor at transparency and the “asymptotic” one as functions of the baryonic matter. It is interesting that, for a given EMBH, there is a maximum value of the $`\overline{\gamma }`$-factor at transparency. After that maximum value, the energy available for the GRB is smaller in intensity, and at decreasing values of the energy, for increasing values of the baryonic mass.
The temperature in the laboratory frame $`\overline{\gamma }T`$ at the transparency point is plotted as a function of the baryonic mass in Fig. 10: it strongly decreases as the baryonic mass increase. The $`\overline{\gamma }T`$ is related to the observed energy-peak of the photon-number spectrum (see e.g., Ruffini et al. (1999)).
We plot in Fig. 11 the energy radiated in the burst and the final kinetic energy of baryonic matter. We find that, for small values of $`B`$ (around $`10^8`$), almost all total energy is radiated as GRB (see also our previous paper Ruffini et al. (1999)), and very little energy is left as kinetic energy of baryonic matter as afterglow. While for $`B10^2`$ roughly only $`10^2`$ of the total initial energy of the dyadosphere is radiated away as GRB, and almost all energy is restored as the kinetic energy of the baryonic matter. It is also clear that for $`B>10^2`$ the intensity of the Burst (see also Fig. 8) and the observed radiation frequency drifted to smaller and smaller values and are of little astrophysical interest from the point of view of GRBs. For such values, the energy of the dyadosphere is transferred practically totally to the bulk kinetic energy of the baryonic matter. $`B=10^2`$ is also the limit of the validation of our computations based on the analytical slab model, described in Sect. 3. For values of baryonic matter in between these two extremal cases ($`B=10^8`$ and $`B=10^2`$) the analytical slab model covers the whole range of the observed properties of Gamma Ray Bursts.
We turn now to the thermodynamic parameters relevant in the evolution of the PEM pulse. In Fig. 12 the temperature of PEM pulse, both in the comoving and in the laboratory frame, are given as a function of the radius for a typical case ($`M_{\mathrm{BH}}=10^3M_{},\xi =0.1`$ and $`B=10^2`$).
In Fig. 13, we plot the total energy of the non baryonic components of the PEM pulse, which includes both thermal and kinetic energy, and the kinetic energy of the baryonic matter as functions of the radius, for the typical case $`M_{\mathrm{BH}}=10^3M_{},\xi =0.1`$ and $`B=10^2`$. The total energy of the non baryonic components of the PEM pulse is equal to $`E_{\mathrm{dya}}`$ before the collision (see Ruffini et al. (1999)) and drops after the collision. While the kinetic energy of baryonic matter
$$E_k=\overline{\rho }_BV(\overline{\gamma }1)$$
(53)
increases as function of radius for $`(r>r_{\mathrm{in}})`$. The sum of both them is equal to the total energy $`E_{\mathrm{dya}}=3.0910^{54}`$ergs during the evolution of the PEM pulse. The value of the total energy of the non baryonic components of the PEM pulse at the transparency point, the ending point of the curve in Fig. 13, is the energy released in the burst. We have discussed this energy as function of baryonic masses in Fig. 11.
Before and after the collision, the condition of entropy conservation applies, and we have:
$$S_{\mathrm{before}}=\frac{V}{T}\left(\rho _\gamma +\rho _{e^\pm }+p_\gamma +p_{(e^\pm )}\right)$$
(54)
$`S_{\mathrm{after}}`$ $`=`$ $`{\displaystyle \frac{V}{T}}(\rho _\gamma +\rho _{e^\pm }+\rho _e^{}^b+ϵ_B`$ (55)
$`+`$ $`p_\gamma +p_{e^\pm }+p_B+p_e^{}^b),`$
where $`ϵ_B`$ is the thermal energy of baryonic matter, and we can neglect, in Eq. (55), the pressure of the baryonic matter. A sudden increase of the entropy occurs during the collision both for the addition of the baryonic matter, and for the thermal reheating due to the inelastic collapse of the PEM pulse with the baryonic matter at rest. From the energy and momentum conservations, we obtain for the values of the $`\overline{\gamma }`$ and the temperature during the collision: the proper internal energy, $`E_{\mathrm{int}}`$, of the plasma increases as
$$dE_{\mathrm{int}}=(\gamma 1)dM_B$$
(56)
and the slab is decelerated, in terms of Lorentz factor $`\gamma `$, as
$$d\gamma =\frac{\gamma ^21}{M_Bc^2+E_{\mathrm{int}}}dM_B,$$
(57)
as baryon mass, $`dM_B`$, is incrementally gained.
Before the collision the PEM pulse expands keeping its temperature in the laboratory frame constant, while its temperature in the comoving frame falls (see Ruffini et al. (1999)). During the collision, a heating of the plasma due to the energy and momentum conservation occurs (see also Fig. 14, where reheating process leads to an increment of the number density of $`e^+e^{}`$ pairs). As the system expands further, both the comoving temperature and the temperature in the laboratory frame decreases, since the total energy of the $`e^+e^{}`$ pairs and the photons before the collision is constant and equal to $`E_{\mathrm{dya}}`$, while after the collision
$$E_{\mathrm{dya}}=E_{\mathrm{Baryons}}+E_{e^+e^{}}+E_{\mathrm{photons}}$$
(58)
where $`E_{\mathrm{Baryons}}`$ is only the thermal energy of the baryons.
It is also interesting to monitor the change of temperature in the comoving frame before the collision and after the collision (see also Fig. 15). Before the collision, due to the entropy conservation, in the process of $`e^+e^{}`$ annihilation the factor $`\frac{T^3V}{T_{}^3V_{}}`$ (where the subscript “” means “initial value”) increases to a value near to $`\frac{11}{4}`$ (see Weinberg (1972); Ruffini et al. (1999)) since the collision occurs at $`r=100r_{\mathrm{ds}}`$, near the condition of transparency. The number of $`e^+e^{}`$ pairs has now been reduced drastically. The further jump in the value of the ratio $`\frac{T^3V}{T_{}^3V_{}}`$ is principally due to the energy and momentum conservation during the inelastic collision. After the collision, there is a small reheating due to the annihilation of the $`e^+e^{}`$ pairs created in the collision (see Fig. 14), which cannot be seen on the scale of our plot. Then, the ratio remains constant.
## 6 Conclusions
By the direct comparison with the numerical integration of the complete relativistic hydrodynamical equations, the use of the constant-thickness approximation has been validated for values of the parameter $`B10^2`$. For $`B10^2`$ the amount of energy released at transparency in the burst decreases, and its energy drifts toward low energy values, which are of little interest for the astrophysics of GRB. We conclude that the constant-thickness approximation is valid for all astrophysically relevant situations for the analysis of the GRB at transparency. Based on this validation we have studied the evolution of a PEM pulse in the presence of selected amounts of baryonic matter. We have studied for a typical case of $`M_{\mathrm{BH}}=10^3M_{},\xi =0.1`$ EMBH and $`B=10^2`$ baryonic shell, the thermal, the bulk Lorentz factor evolution of the PEM pulse as well as the kinetic energy left over in the baryonic matter after the decoupling of matter and radiation and the emission of the GRB. Additional results corresponding to a larger range of masses and charges of the EMBH and the correlations between the peak energy and the duration of GRBs to be expected in our model will be considered in a future paper, together with the results of analyzing the interaction of the kinetic energy left over in the baryonic matter, after the decoupling of matter and radiation, with the interstellar medium.
## 7 Acknowledgments
Work by JDS and JRW was performed under the auspices of the U.S. Department of Energy by University of California Lawrence Livermore National Laboratory under contract No. W-7405-Eng-48.
|
warning/0004/hep-ph0004073.html
|
ar5iv
|
text
|
# Charm production by cosmic muons
## Abstract
Narrow muon bundles in underground detectors permit to study muoproduction reactions that take place in the surrounding rock. We analyze the relevance of a QED+QCD reaction, muoproduction of “open charm”. The contribution to double muon events is estimated to be $`48`$% of the one due to QED “trident” process, for an ideal detector located under a rock depth of 3 km water equivalent, and an observation threshold of 1 GeV.
In recent years, there has been a certain experimental and theoretical interest on “narrow muon bundles” (multiple muons with a lateral separation less than a few meters) in underground detectors. These events have been observed as a peak at small lateral separation, and interpreted as an induced flux. In fact, the energetic muons that propagate underground in roughly $`1`$% of the cases interact and produce other muons. Thence, an analysis of these events requires to consider high energy muoproduction processes, in the rock surrounding the detector.
Up to now, the process considered was the muon “trident” reaction $`\mu Z\mu Z\mu ^+\mu ^{}`$, where a muon pair is formed in the field of the nucleus<sup>2</sup><sup>2</sup>2It is assumed that an effective rejection of muoproduced $`\pi ^\pm ,\gamma ,`$ e<sup>±</sup> … can be achieved.. For muons propagating in high $`Z`$ materials an amplification factor $`Z^2/A`$ ($`=5.5`$ for standard rock, $`A`$=22 and $`Z`$=11) is present, due to coherent character of the reaction. This interpretation has been pursued since the first evidences obtained in cosmic ray experiments . The trident reaction leads mostly to narrow bundles of three or two muons in an underground detector (one produced muon may stop before reaching the detector); a reference ratio in an ideal detector is of $`3`$ double muons per triple muon, for a threshold of observation $`E_h=1`$ GeV, and a depth $`h=3`$ km w.e. of standard rock. Recent studies , however, suggest that existing interpretations are insufficient to quantitatively account for the whole “narrow muon” data set.
In this work we analyze the role of another high energy process as source of prompt muons: production of charmed states due to cosmic (atmospheric) muons, whose relevant energies range from<sup>3</sup><sup>3</sup>3We neglect the energy loss in the rock of the charmed hadrons $`X_c,`$ for a 200 GeV $`D^\pm `$ meson travels on average just 3 cm in the rock before decaying. $`E100`$ GeV up to tenths of TeV‘s (for studies in laboratory, see ). More specifically, we are concerned with the “open charm” reaction of muoproduction: $`\mu N\mu c\overline{c}X`$ ($`X`$ denotes a byproduct which does not concern us). This process is stipulated by QED and QCD interactions, while weak interactions provide the instability of charmed states: $`cX_c\mu X.`$
In order to obtain a simple estimate of the flux of double muons due to this process, we adopted the collinear approximation, considering only how the initial muon with energy $`E`$ branches into those of the final muons ($`E^{}`$ and $`E^{\prime \prime }`$) and proceeded in the following way:
1) First, we calculated the cross section of muoproduction $`d\sigma _{\mu N\mu c\overline{c}X}/dE^{}dE_c`$ at leading order (LO) in $`\alpha _s,`$ double differential in the energies of the scattered muon, $`E^{}`$ and of the charm, $`E_c`$ (see appendix). This can be done with a limited effort by following the calculations documented in , where the cross section integrated over the hadronic phase space $`d\mathrm{\Phi }_{hadr}`$ was obtained: In fact, neglecting the gluon mass, the differential expression is simply $`d\mathrm{\Phi }_{hadr}=dE_c/(8\pi E_\gamma ),`$ where $`E_\gamma =EE^{}`$ is the energy of the virtual photon emitted by the muon<sup>4</sup><sup>4</sup>4Also, we found convenient to relate $`E_c`$ to the zenith angle and velocity of emission in the gluon-gamma c.m.s. as follows: $`E_c/E_\gamma =(1+\beta _c^{}\mathrm{cos}\theta _c^{})/2,`$ where $`\beta _c^{}=\sqrt{14m_c^2/(p+q)^2}`$ ($`p`$ and $`q`$ are the momenta of the gluon and of the virtual photon). In the actual calculation, that requires integrating over the photon virtuality $`Q^2`$ and the gluon momentum fraction $`x,`$ we use the GRV98 gluon distribution , and set: $`m_c=1.5,`$ 1.35 or 1.2 GeV. We multiplied the cross section by the factor $`K(E)=\sigma _{NLO}/\sigma _{LO}`$ (where $`\sigma `$ is the total cross section) to describe next-to leading order QCD effects . The differential cross section increases with $`E^{}`$ with a “1/$`v^2`$ behavior” and than rapidly decreases to zero in the range of energies of interest; instead, it is rather mildly distributed in $`E_c.`$ The total cross section $`\sigma `$ increases with $`E,`$ due to the smaller values of $`x`$ that are probed by the virtual photon, and to well known characteristics of the gluon distribution. Its value is $`4\times 10^{32}`$ cm<sup>2</sup> when $`E=1`$ TeV for $`m_c=1.35`$ GeV (almost equal to the trident cross section at the same energy); LO cross section increases by 50% if $`m_c`$ is lowered to 1.2 GeV, while decreases by 30% if $`m_c`$ is 1.5 GeV.
2) We estimate a “scaling” probability $`dP_{c\mu }/dwBR_{c\mu }\times \rho _{c\mu }(w)`$ that a charm yields a muon with a certain energy fraction $`w=E^{\prime \prime }/E_c,`$ by first hadronizing the charm into a $`D`$ meson (using the normalized distribution of with $`ϵ_D=0.135`$) and then letting it decay with a $`K_{\mu 3}`$ distribution (that is, retaining only the $`D`$ mass, and neglecting the $`Q^2`$ dependence of the form factors). The resulting normalized probability $`\rho _{c\mu }(w)`$ falls strongly with the energy fraction $`w;`$ the median of the distribution is in fact $`E^{\prime \prime }=0.15\times E_c.`$ We took as an effective branching ratio of charm into muons the value $`BR_{c\mu }=8`$% , and multiplied the result by two, to account for the fact that a charm or an anticharm can yield a muon<sup>5</sup><sup>5</sup>5Existing underground detectors do not distinguish between “same charge” and “opposite charge” double muon events.. Notice, incidentally, that the corresponding yield of triple muon is negligible, due to an a priori 4% suppression factor.
3) At this stage, we can calculate the cross section $`d\sigma _{\mu N\mu \mu }/dE^{}dE^{\prime \prime },`$ where $`E^{\prime \prime }`$ is the energy of the produced muon, and, with that, the cross section<sup>6</sup><sup>6</sup>6We consider only those events whose vertex is not contained in the detector. Those events profit of a large effective target mass, and correspond, in a sense, to the celebrated neutrino-induced single muon signal. $`\sigma _{\mu N\mu \mu }(E,f)`$ for production of two muons, each with a fraction of the initial muon energy greater than $`f.`$ Due to the behaviors of $`d\sigma _{\mu N\mu c\overline{c}X}`$ and $`dP_{c\mu }`$ with $`E^{}`$ and $`E^{\prime \prime }`$ mentioned above, this cross section diminishes dramatically with $`f;`$ when $`E=1`$ TeV, it drops down by one order of magnitude already when $`f0.07.`$ This cross section enters the elementary yield of double muons in the detector, which depends linearly on the infinitesimal depth crossed $`dh^{}`$ (in gr/cm<sup>2</sup>):
$$dY_{\mu \mu \mu }(E,h^{})=dh^{}\times N_A\times \sigma _{\mu N\mu \mu }(E,f)\text{ where }f=\frac{E_h^{}}{E}$$
(1)
$`N_A=6.023\times 10^{23}`$ is the number of nucleons in 1 mole (multiplying by $`dh^{},`$ we obtain the density of targets per cm<sup>2</sup>). The energy losses are evaluated in continuous energy loss approximation, $`E_h^{}=(E_h+ϵ)\mathrm{exp}[(hh^{})/h_0]ϵ,`$ where $`ϵ600`$ GeV and $`h_02.5`$ km w.e. are phenomenological parameters, and $`E_h=1`$ GeV is the (typical) threshold for detection. Multiplying this by the single muon flux differential in $`dE`$, $`dF_\mu ,`$ we get the differential double muon flux induced by “open charm” reaction at the depth $`h.`$ The total flux is then:
$$F_{\mu \mu }(h)=_0^h_{2E_h^{}}^{\mathrm{}}𝑑F_\mu (E,h^{})\times 𝑑Y_{\mu \mu \mu }(E,h^{})$$
(2)
where we integrated over the depth of production $`h^{},`$ and the energy $`E`$ of the primary muon at this depth (namely, where the reaction takes place); $`E`$ was related to the energy at the surface $`E_0`$ in the continuous energy loss approximation, which permits us to evaluate $`dF_\mu (E,h^{})/dE`$ by using the (approximate) expression for the flux at the surface given in (namely, $`dF_\mu /dE_0=0.14\times E_0^{2.7}\times \mathrm{}`$ /(cm<sup>2</sup> s sr GeV)). The results are shown in the figure, for muons arriving from the vertical direction.
The contribution of open charm reaction is not large; for instance, at a depth of 3 km w.e. it is just $`48`$% of the one due to the trident process. Equivalently, it can be compared with the flux of single muon: we get $`F_{\mu \mu }/F_\mu =0.7,1,1.4\times 10^5`$ in the case of open charm reaction (for $`m_c=1.5,1.35,1.2`$ GeV) while $`F_{\mu \mu }/F_\mu =1.8\times 10^4`$ in the case of trident reaction. For an ideal detector, it would require to accumulate several hundredth (trident narrow double muon) events, to become statistically interesting. The smallness of the result has to be attributed to the relatively small value of $`BR_{c\mu },`$ and to the effective leakage of energy of the virtual photon, during the conversion $`\gamma ^{}cD\mu `$ (while for tridents, $`\gamma ^{}`$ immediately materializes into muons). However, this contribution is not negligible if one aims at reaching the precision of $`510`$% in the predictions.
The following remarks illustrate other aspects of this result:
(a) going to shallower depths, the double muon flux increases, though less rapidly than the single muon flux: in fact, the effective target increases with the depth (but, of course, also the background increases);
(b) conversely, in deeper sites the relative contribution of the open charm process becomes more important, due to more energetic primaries–$`E`$ increases (but there are practical limitations, due to the time of data taking and area of the installation);
(c) keeping the depth fixed, and changing the angle of observation, there is an increase of $`F_{\mu \mu }`$ moving toward the horizon, directly related to the increase of $`F_\mu `$ (but the actual geometry of the rock in the underground site has an essential role in practical considerations);
(d) in water or ice, the trident curve would reduce by $`1.5`$ in comparison with the open charm one, due to the $`Z^2/A`$ factor. This would somehow emphasize the open charm contribution (but it should be reminded that no plan exists to have an underwater detector, with large area and capable to achieve a good discrimination at small lateral distances).
In conclusion, it seems to us rather difficult to account for a large fraction of narrow muon bundles on the basis of the open charm process. Thus, the chances of studying heavy quark physics with existing underground (or underwater) detectors are quite limited. This result, however, adds motivations for further search of unexpected sources of background (and, possibly, new sources of prompt muons) when we recall the difficulties to understand existing data on narrow muon bundles. For the future experimental perspectives, we consider interesting the possibility to achieve energy and charge identification of the muons in underground detectors, as a possible handle to separate various components in a narrow muon bundles data set (for instance, we found that the average energy of the parent muon–that forms two muons through the open charm process–is rather large, above 1 TeV). However, even if it will be possible to obtain sufficient statistics and control of the systematics, an attempt to proceed further and extract a signal of production of heavy quarks from studies of narrow muon bundles will need more refined calculations: To accurately describe the NLO effects , hadronization, and decay of charmed states ; but also those effect in the muon propagation, that are necessary to model the lateral distribution of the events in the underground detectors . In fact, the relatively large transverse momenta $`p_{}m_c`$ that result from charm production and decay lead to larger lateral separations than those due to the trident reaction, and this could be of interest to characterize the charm induced events.
Acknowledgments I would like to express my gratitude to several people: to V.S. Berezinsky for guide and support, and for informing me on the work , that the present study followed rather closely; to E. Scapparone for pointing my interest on the physics of “narrow muon bundles”; to G. Battistoni, S. Cecchini, R.P. Kokoulin, V.A. Kudryavtsev, P. Lipari and O.G. Ryazhskaya for helpful discussions on cosmic muons; A.V. Butkevich for explaining me how to include the nuclear effects; to B.W. Harris for help with NLO charm production cross section.
Appendix: Formulæ for LO cross sections The LO cross section, differential in $`y=E_\gamma /E`$ and $`z=E_c/E_\gamma `$ is:
$$\begin{array}{c}\frac{d\sigma _{\mu N\mu c\overline{c}X}}{dydz}=\frac{\alpha ^2}{9ME}d\mathrm{log}Q^2d\mathrm{log}x\alpha _s(\mu ^2)G(x,\mu ^2)\hfill \\ \left[\left(1\frac{2m_\mu ^2}{Q^2}+2g(y)\right)\frac{df_T}{dz}+\left(1\frac{2m_\mu ^2}{Q^2}+6g(y)\right)\frac{df_L}{dz}\right]\hfill \end{array}$$
(3)
where $`M=0.938`$ GeV, $`m_\mu =0.106`$ GeV and $`\alpha =1/137;`$ $`\alpha _s`$ is the strong coupling, $`G`$ the gluon distribution function, and $`\mu ^2`$ the factorization scale ($`\mu ^2=4m_c^2+Q^2`$ in present calculation). In order to describe nuclear effects in “standard rock” nuclei, we followed , and weighted the gluon distribution function by the density: $`r^{s.r.}(x)1.26\times x^{0.073}\times (10.3x);`$ see eq. 1 and fig. 2 in that work<sup>7</sup><sup>7</sup>7Notice that here $`x`$ is the momentum fraction of the gluon (the particle that feels the nuclear effects), as contrasted with $`x_FQ^2/(2Pq),`$ the variable introduced to parameterize the structure function $`F^{c\overline{c}}(x_F,Q^2).`$. The adimensional functions introduced in eq. 3 are $`g(y)=(1y)/y^2`$ and
$$\begin{array}{c}\frac{df_T}{dz}=g_1g_0+x_cg_1\frac{x_c^2}{4}(g_2+2g_1)2x_Q(g_1\frac{x_c}{4}g_2)+2x_Q^2g_1\hfill \\ \frac{df_L}{dz}=2x_Q(g_0\frac{x_c}{2}g_1)2x_Q^2g_0\hfill \end{array}$$
(4)
where $`g_n=1/z^n+1/(1z)^n`$ for $`n=0,1,2;`$ $`x_c=2m_c^2/pq`$ and $`x_Q=Q^2/(2pq)`$ are bound by $`x_{min}`$ to be lower than unity. The non-trivial limits are: $`(i)`$ $`x_{min}=(m_c^2g_1(z)+Q^2)/(2ME_\gamma ),`$ which results from the kinematics of the $`\gamma ^{}gc\overline{c}`$ process (setting the gluon mass to zero), and $`(ii)`$ $`Q_{min}^2m_\mu ^2/g(y)`$ which results from setting to zero the scattering angle in the laboratory frame; the limits on $`y`$ and $`z,`$ and $`Q_{max}^2`$ follow by consistence. Notice that the cross section can be easily integrated analytically over $`z,`$ which amounts to replace:
$$g_02\beta ^{},g_12\mathrm{log}\frac{1+\beta ^{}}{1\beta ^{}},g_2\frac{8\beta ^{}}{1(\beta ^{})^2}$$
(5)
and, also, $`g_1(z)g_1(1/2)=4`$ in the expression of $`x_{min};`$ the resulting expression for $`d\sigma _{\mu N\mu c\overline{c}X}/dy`$ is equivalent to the one shown in the appendix of . The cross section that enters the expression of the double muon yield (eq. 1) is obtained as:
$$\sigma _{\mu N\mu \mu }(E,f)=2\times BR_{c\mu }\times _f^{1f}𝑑y_{f/y}^1𝑑z\frac{d\sigma _{\mu N\mu c\overline{c}X}}{dydz}\times _{f/(yz)}^1𝑑w\rho _{c\mu }(w)$$
(6)
The last integral corresponds to an integral probability, and can be tabulated separately to simplify the calculation.
|
warning/0004/astro-ph0004376.html
|
ar5iv
|
text
|
# The cosmological lens equation in a universe with non zerocosmological constant
## 1 Introduction
Recent determination of distances to high redshift type Ia supernovae revealed that the universe is now expanding faster than expected. This accelerated expansion is linked with a non zero cosmological constant. Since gravitational lensing of cosmological objects involves also high redshift galaxies it is interesting to extend the cosmological lens equation to cosmological models with non zero cosmological constant.
Furthermore the mass distribution in the universe at the relevant range of redshifts is clearly non homogeneous. In our considerations we will take into account the fact that at the scale up to several hundred Mpc matter distribution is non homogeneous. Moreover, with the advent of the Next Generation Space Telescope (NGT), which will allow imaging of objects at $`z10`$ (\[Barkana & al.2000\]), it will be possible to check better and better the influence of non homogeneities .
Using the general equations describing propagation of light in a given spacetime we derive an equation for the angular diameter distance and we extend the notion of angular diameter distance to two objects at arbitrary redshifts. The general equation for the angular diameter distance, the so called Dyer-Roeder equation, can be exactly solved also in the case when the cosmological constant is different from zero. However, the exact solution is very complicated and therefore not useful for practical applications. We propose a simple analytic form of an approximate solution. The approximate form of the angular diameter distance depends on four arbitrary parameters. We fix the values of these parameters by fitting this form to the exact solution.
The paper is organized as follows: In Section $`2`$ we present the derivation of the angular diameter distance in an arbitrary Friedman- Robertson- Walker cosmological model. To take into account the non homogeneous distribution of matter, following the standard practice, we introduce a constant phenomenological parameter $`\stackrel{~}{\alpha }`$ and rewrite the final equation in a few different forms. In Section $`3`$ we present exact analytic solutions for the angular diameter distance, first in a flat universe and then in the general case of arbitrary spatial curvature. Section $`4`$ is devoted to the approximate solution and to discussions of the fitting procedure used to determine the values of arbitrary parameters. Discussion of our results is presented in Section $`5`$.
## 2 General considerations
Let us consider a beam of light emanating from a source S. The light rays propagate along a surface $`\mathrm{\Sigma }`$ which is determined by the eikonal equation
$$g^{\alpha \beta }\mathrm{\Sigma },_\alpha \mathrm{\Sigma },_\beta =0$$
(1)
A light ray is identified with a null geodesic on $`\mathrm{\Sigma }`$ with the tangent vector $`k_\alpha =\mathrm{\Sigma },_\alpha `$. The light rays in the beam can be described by $`x^\alpha =x^\alpha (v,y^a)`$ where $`v`$ is an affine parameter and $`y^a`$ (a=1, 2, 3) are three parameters specifying different rays. The tangent vector field to the light ray congruence, $`k^\alpha =\frac{dx^\alpha }{dv}=\mathrm{\Sigma },_\alpha `$, determines two optical scalars, the expansion $`\theta `$ and the shear $`\sigma `$:
$$\theta =\frac{1}{2}k_{}^{\alpha }{}_{;\alpha }{}^{}\sigma =k_{\alpha ;\beta }\overline{m}^\alpha \overline{m}^\beta ,$$
(2)
where $`\overline{m}^\alpha =\frac{1}{\sqrt{2}}(\xi ^\alpha i\eta ^\alpha )`$ is a complex vector spanning the spacelike 2-space (the screen space) orthogonal to $`k^\alpha `$ $`(k^\alpha \overline{m}_\alpha =0)`$ (actually the vorticity connected with the light beam is zero in all our considerations, therefore in our case these two scalars completely characterize the congruence). These two optical scalars describe the relative rate of change of an infinitesimal area A of the cross section of the beam of light rays and its distortion. In particular
$$\theta =\frac{1}{2}k_{}^{\alpha }{}_{;\alpha }{}^{}=\frac{1}{2}\frac{d\mathrm{ln}A}{dv}.$$
(3)
These two optical scalars satisfy the Sachs (\[Sachs & Kristian 1966\]) propagation equations
$$\dot{\theta }+\theta ^2+|\sigma |^2=\frac{1}{2}R_{\alpha \beta }k^\alpha k^\beta ,$$
(4)
and
$$\dot{\sigma }+2\theta \sigma =\frac{1}{2}C_{\alpha \beta \gamma \delta }\overline{m}^\alpha k^\beta \overline{m}^\gamma k^\delta ,$$
(5)
where the dot denotes the derivative with respect to $`v`$, $`R_{\alpha \beta }`$ is the Ricci tensor, and $`C_{\alpha \beta \gamma \delta }`$ is the Weyl tensor. Equations (4) and (5) follow from the Ricci identity. We will use these equations to study propagation of light in the Friedman-Robertson-Walker (FRW) spacetime. The FRW spacetime is conformally flat, so in such spacetimes $`C_{\alpha \beta \gamma \delta }=0`$. From equation (5) it follows that if initially the shear of the null ray congruence is equal to zero than it is always zero. Therefore assuming that the light beam emanating from the source S has vanishing shear we can disregard the shear parameter altogether. Using (3) we can rewrite equation (4) in the form
$$\ddot{\sqrt{A}}+\frac{1}{2}R_{\alpha \beta }k^\alpha k^\beta \sqrt{A}=0.$$
(6)
An observer moving with the 4-velocity vector $`u^\alpha `$ will be associated with the light ray circular frequency $`\omega =cu^\alpha k_\alpha `$. Different observers will assign different frequencies to the same light ray. The shift of frequencies as measured by an observer comoving with the sources and an arbitrary observer is related to the redshift by
$$1+z=\frac{\omega }{\omega _o}=\frac{c}{\omega _o}u^\alpha k_\alpha ,$$
(7)
where $`\omega _o`$ is the frequency measured by the distant observer. Differentiating this equation with respect to the affine parameter $`v`$ we obtain
$$\frac{dz}{dv}=\frac{c}{\omega _o}k^\alpha k^\beta u_{\alpha ;\beta }.$$
(8)
Since the angular diameter distance D is proportional to $`\sqrt{A}`$ we can rewrite equation (6) using D instead of $`\sqrt{A}`$ and at the same time we replace the affine parameter $`v`$ by the redshift $`z`$, we obtain
$$\left(\frac{dz}{dv}\right)^2\frac{d^2D}{dz^2}+\left(\frac{d^2z}{dv^2}\right)\frac{dD}{dz}+\frac{4\pi G}{c^4}T_{\alpha \beta }k^\alpha k^\beta D=0,$$
(9)
where we used the Einstein equations to replace the Ricci tensor by the energy-momentum tensor.
To relate a solution of (9) with the distance it has to satisfy the following initial conditions:
$`D(z)|_{z=0}=0,`$
(10)
$`{\displaystyle \frac{dD(z)}{dz}}|_{z=0}={\displaystyle \frac{c}{H_0}}.`$
To be able to use solutions of the Eq. (9) to describe gravitational lenses we have to introduce the distance between the source and the lens $`D(z_l,z_s)`$, where $`z_l`$ and $`z_s`$ denote correspondingly the redshift of the lens and the source. Let $`D(z_1,z_2)`$ denote the angular diameter distance between a fictitious observer at $`z_1`$ and a source at $`z_2`$, of course $`D(0,z)=D(z)`$. Suppose that we know the general solution of equation (9) for $`D(z)`$ which satisfies the initial conditions (2), then the function $`D(z_1,z)`$ defined by
$$D(z_1,z)=\frac{c}{H_0}(1+z_1)D(z_1)D(z)\left|_{z_1}^z\frac{dz^{}}{D^2(z^{})g(z^{})}\right|,$$
(11)
such that
$`D(z_1,z)|_{z=z_1}=0,`$
$`{\displaystyle \frac{d}{dz}}D(z_1,z)|_{z=z_1}=\mathrm{sign}(zz_1){\displaystyle \frac{1+z_1}{g(z_1)}},`$
satisfies equation (9), if the function $`g(z)`$ is a solution of
$`{\displaystyle \frac{d}{dz}}\mathrm{ln}g(z)={\displaystyle \frac{\frac{d^2z}{dv^2}}{\left(\frac{dz}{dv}\right)^2}},`$ (12)
so
$$g(z)=g_0\mathrm{exp}\frac{\frac{d^2z^{}}{dv^2}}{\left(\frac{dz^{}}{dv}\right)^2}𝑑z^{}.$$
(13)
$`g_0`$ is an arbitrary constant of integration, which, without restricting generality, we assume to be one. To obtain the equation (12) we inserted (11) into the equation (9) and demanded that it is satisfied. Let us note that the Etherington reciprocity relation (\[Etherington 1933\])
$$\frac{D(z_1,z_2)}{1+z_1}=\frac{D(z_2,z_1)}{1+z_2},$$
(14)
follows directly from (11).
We are interested in applying equation (9) to find the angular diameter distance to objects at high redshifts. Therefore let us consider the standard FRW spacetime described by the line element
$$ds^2=dt^2R^2(t)\left[\frac{dr^2}{1kr^2}+r^2(d\theta ^2+\mathrm{sin}^2\theta d\phi ^2)\right],$$
(15)
where k (-1, 0, +1) is the curvature constant, and $`R(t)`$ is the scale factor. If the matter content of the universe can be approximated by dust then, as shown by Friedman, the Einstein field equations for the metric (15) assume the form
$`{\displaystyle \frac{\dot{R}^2}{R^2}}=H^2={\displaystyle \frac{8\pi G}{3}}\varrho _m+{\displaystyle \frac{\mathrm{\Lambda }c^2}{3}}{\displaystyle \frac{kc^2}{R^2}},`$
(16)
$`{\displaystyle \frac{\ddot{R}}{R}}={\displaystyle \frac{4\pi G}{3}}\varrho _m+{\displaystyle \frac{\mathrm{\Lambda }c^2}{3}}`$
where here the dot denotes derivative with respect to $`t`$, $`H={\displaystyle \frac{\dot{R}}{R}}`$, $`\varrho _m`$ is the matter density, and $`\mathrm{\Lambda }`$ is the cosmological constant.
In the FRW spacetime the redshift $`z`$ is connected with the scale factor $`R(t)`$ by $`1+z={\displaystyle \frac{R_0}{R(t)}}`$. Differentiating this relation with respect to time we obtain
$`{\displaystyle \frac{dz}{dt}}=(1+z)H(z).`$
Let us now return to the description of propagation of a light beam. Assuming that the observer is a standard FRW observer, e.g. it is comoving with matter, from the equation (8), we obtain
$$\frac{c^2}{\omega _0}\frac{dz}{dv}=(1+z)^2H(z).$$
(17)
Introducing a new dimensionless affine parameter $`w=\frac{H_0\omega _0}{c^2}v`$ we transform equation (17) into
$$H_0\frac{dz}{dw}=(1+z)^2H(z)$$
(18)
The Friedman equation for $`H^2`$, Eq. (16), can be rewritten as
$$H^2(z)=H_0^2\left(\mathrm{\Omega }_m(1+z)^3+\mathrm{\Omega }_k(1+z)^2+\mathrm{\Omega }_\mathrm{\Lambda }\right)$$
(19)
where $`H_0`$ is the present value of the Hubble constant, and
$$\mathrm{\Omega }_m=\frac{8\pi G\varrho _0}{3H_0^2},\mathrm{\Omega }_k=\frac{kc^2}{R_0^2H_0^2},\mathrm{\Omega }_\mathrm{\Lambda }=\frac{\mathrm{\Lambda }c^2}{3H_0^2},$$
are the present density parameters of matter, curvature and cosmological constant respectively, and, in the case $`p=0`$ (no radiation), we have
$$\mathrm{\Omega }_m+\mathrm{\Omega }_k+\mathrm{\Omega }_\mathrm{\Lambda }=1.$$
(20)
Substituting (19) into (18) we get
$$\frac{dz}{dw}=(1+z)^2\sqrt{\mathrm{\Omega }_m(1+z)^3+\mathrm{\Omega }_k(1+z)^2+\mathrm{\Omega }_\mathrm{\Lambda }}.$$
(21)
To apply the angular diameter distance to the realistic distribution of galaxies it is necessary to take into account local non homogeneities. Unfortunately so far an acceptable averaging procedure for smoothing out local non homogeneities has not been developed (\[Krasinski 1997\]). Therefore following previous discussions we introduce the phenomenological parameter $`\stackrel{~}{\alpha }`$ which describes the influence of local non homogeneities on propagation of light (\[Dyer & Roeder 1972\]), (\[Tomita et al. 1999\]). With this alteration equation (9), in the FRW-dust case, can be rewritten in the form
$$\left(\frac{dz}{dw}\right)^2\frac{d^2D}{dz^2}+\left(\frac{d^2z}{dw^2}\right)\frac{dD}{dz}+\frac{3}{2}\stackrel{~}{\alpha }\mathrm{\Omega }_m(1+z)^5D=0.$$
(22)
It is customary to measure cosmological distances in units of $`\frac{c}{H_0}`$, introducing the dimensionless angular diameter distance (the Dyer-Roeder distance) $`r=\frac{DH_0}{c}`$, and using (22) we finally obtain
$`(1+z)\left[\mathrm{\Omega }_m(1+z)^3+\mathrm{\Omega }_k(1+z)^2+\mathrm{\Omega }_\mathrm{\Lambda }\right]{\displaystyle \frac{d^2r}{dz^2}}+`$
(23)
$`\left({\displaystyle \frac{7}{2}}\mathrm{\Omega }_m(1+z)^3+3\mathrm{\Omega }_k(1+z)^2+2\mathrm{\Omega }_\mathrm{\Lambda }\right){\displaystyle \frac{dr}{dz}}+{\displaystyle \frac{3}{2}}\stackrel{~}{\alpha }\mathrm{\Omega }_m(1+z)^2r=0.`$
with the initial conditions:
$`r(z)|_{z=0}=0,`$
(24)
$`{\displaystyle \frac{dr(z)}{dz}}|_{z=0}=1.`$
This equation can be cast into a different form by using $`a=\frac{R}{R_0}`$ as a parameter instead of $`z`$, we obtain
$$a^2(\mathrm{\Omega }_m+\mathrm{\Omega }_ka+\mathrm{\Omega }_\mathrm{\Lambda }a^3)\frac{d^2r}{da^2}a(\frac{3}{2}\mathrm{\Omega }_m+\mathrm{\Omega }_ka)\frac{dr}{da}+\frac{3}{2}\stackrel{~}{\alpha }\mathrm{\Omega }_mr=0,$$
(25)
or using the cosmic time $`t`$ as a parameter, the equation (22) assumes the form
$$\frac{d^2r}{dt^2}H(t)\frac{dr}{dt}+4\pi G\stackrel{~}{\alpha }\varrho _m(t)r=0.$$
(26)
Equation (26) needs some comments: this equation was first introduced by Dashevski & Zeldovich (\[Dashevski & Zeldovich 1965\]) (see also Dashevski & Slysh, \[Dashevski & Slysh 1966\]). More recently Kayser et al. (\[Kayser et al.1997\]) have used it to derive an equation similar to our (2). In Eq.(26) the clumpiness parameter $`\stackrel{~}{\alpha }`$ is usually considered as a constant. However in the papers by Dashevski & Zeldovich and by Kayser et al. $`\stackrel{~}{\alpha }`$ is allowed to vary with time but only the case $`\stackrel{~}{\alpha }=const.`$ is really considered. For a discussion of the case in which $`\stackrel{~}{\alpha }`$ depends on $`z`$ see, for example, the paper by Linder (\[Linder 1988\]).
As an example of our procedure let us consider the case when $`\mathrm{\Omega }_\mathrm{\Lambda }=0`$ and $`\mathrm{\Omega }_k=0`$, that is a flat Universe so the universe is flat. In this case it is easy to see that
$$r(z)=\frac{(1+z)^\beta (1+z)^\beta }{2\beta (1+z)^{5/4}},$$
(27)
where $`\beta =\frac{1}{4}\sqrt{2524\stackrel{~}{\alpha }}`$, is the general solution of equation (2) with appropriate initial conditions. This solution can be found in SEF (\[Schneider, Ehlers & Falco 1992\]), see their equation (4.56). The $`g`$ function can be easily deduced in the flat $`\mathrm{\Lambda }=0`$ Friedman-Robertson-Walker cosmological model and we obtain
$`g(z)=(1+z)^2\sqrt{\mathrm{\Omega }_{M_0}z+1}.`$
Substituting the expression for $`g(z)`$ we get the familiar solution $`D(z_l,z_s)`$ found in SEF.
## 3 Exact solutions
In recent papers Kantowski (\[Kantowski 1998\]),(\[Kantowski & Kao 2000\]) has found the general solution of the Dyer-Roeder equation written in the form (2). In what follow we use the DR equation in the form (25) and obtain the general solution using more direct approach.
### 3.1 The case $`\mathrm{\Omega }_k=0`$
When $`\mathrm{\Omega }_k=0`$, it is possible to divide the Eq.(25) by $`\mathrm{\Omega }_m`$ and it becomes :
$$a^2(1+\mu a^3)\frac{d^2\overline{r}}{da^2}\frac{3}{2}a\frac{d\overline{r}}{da}+\frac{3}{2}\stackrel{~}{\alpha }\overline{r}=0,$$
(28)
where $`\mu ={\displaystyle \frac{\mathrm{\Omega }_\mathrm{\Lambda }}{\mathrm{\Omega }_m}}`$.
To solve Eq.(28) we use the following strategy: First we solve this equation in the case when $`\mu =0`$ and we look for solutions in the power form $`a^s`$. Inserting this form into (28), we obtain:
$$2s^25s+3\stackrel{~}{\alpha }=0,$$
(29)
which has the solutions :
$`s_\pm ={\displaystyle \frac{5}{4}}\pm {\displaystyle \frac{1}{4}}\sqrt{2524\stackrel{~}{\alpha }}={\displaystyle \frac{5}{4}}\pm \beta ,`$
(30)
$`\beta ={\displaystyle \frac{1}{4}}\sqrt{2524\stackrel{~}{\alpha }}.`$
Writing the general solution is terms of $`z`$ instead of $`a`$ and imposing the initial conditions (2) we recover the solution (27).
In the general case when $`\mu 0`$ we look for solutions in the form $`\overline{r}_\mathrm{\Lambda }=a^sf(x)`$ where $`x=a^3`$. Inserting this form into (28) after some rearrangements we obtain
$$3x(1+\mu x)\frac{d^2f}{dx^2}+\left(2(s+1)(1+\mu x)\frac{3}{2}\right)\frac{df}{dx}+\frac{\mu }{2}(s\stackrel{~}{\alpha })f=0.$$
(31)
This equation can be reduced to the standard hypergeometric equation. The general solution of Eq. (28) can be written in the form :
$`\overline{r}_\mathrm{\Lambda }=A_1{\displaystyle \frac{(1+z)^\beta }{(1+z)^{5/4}}}f_{s_+}\left(\left({\displaystyle \frac{1}{1+z}}\right)^3\right)`$ $`+`$
$`A_2{\displaystyle \frac{(1+z)^\beta }{(1+z)^{5/4}}}f_s_{}\left(\left({\displaystyle \frac{1}{1+z}}\right)^3\right),`$
where $`A_1,A_2`$ are arbitrary constants and we denoted the solutions by $`\overline{r}_\mathrm{\Lambda }`$ to stress that it is the solution of the DR equation in a spacetime with $`\mathrm{\Lambda }0`$. Here $`f_s_{}`$ and $`f_{s_+}`$ are solutions of Eq.(28) with $`s=s_+`$ and $`s=s_{}`$ correspondingly. The constants $`A_1`$ and $`A_2`$ are determined from the initial conditions (2), or explicitly:
$`\overline{r}_\mathrm{\Lambda }(z)|_{z=0}`$ $`=`$ $`A_1f_{s_+}(1)+A_2f_s_{}(1)=0,`$
$`{\displaystyle \frac{d\overline{r}_\mathrm{\Lambda }}{dz}}|_{z=0}`$ $`=`$ $`s_+A_1f_{s_+}(1)+A_1{\displaystyle \frac{d}{dz}}f_{s_+}|_{z=0}s_{}A_2f_s_{}(1)+`$
$`A_2{\displaystyle \frac{d}{dz}}f_s_{}|_{z=0}=1.`$
To find the function $`\overline{r}_\mathrm{\Lambda }(z_1,z_2)`$ we have to solve the equation (12), which in this case is:
$$\frac{g_{\mathrm{\Lambda }}^{}{}_{}{}^{}}{g_\mathrm{\Lambda }}=\frac{3}{2a(1+\mu a^3)}$$
(34)
from which we get:
$$g_\mathrm{\Lambda }=\left[(1+z)^3+\mu \right]^{1/2}.$$
(35)
Then the general two points ($`z>z_1`$) solution of Eq.(28) is :
$`\overline{r}_\mathrm{\Lambda }(z_1,z)`$ $`=`$ $`\overline{r}_\mathrm{\Lambda }(z_1)(1+z_1)\overline{r}_\mathrm{\Lambda }(z)\times `$
$`{\displaystyle _{z_1}^z}{\displaystyle \frac{\sqrt{\mathrm{\Omega }_m}dz^{}}{\overline{r}_\mathrm{\Lambda }^2(z^{})(1+z^{})^2[(1+z^{})^3\mathrm{\Omega }_m+\mathrm{\Omega }_\mathrm{\Lambda }]^{1/2}}},`$
which satisfies the initial conditions:
$`\overline{r}_\mathrm{\Lambda }(z_1,z_1)=0,`$
(37)
$`{\displaystyle \frac{d\overline{r}_\mathrm{\Lambda }}{dz}}(z_1,z)|_{z=z_1}={\displaystyle \frac{\mathrm{sign}(zz_1)\sqrt{\mathrm{\Omega }_m}}{(1+z_1)\sqrt{(1+z)^3\mathrm{\Omega }_m+\mathrm{\Omega }_\mathrm{\Lambda }}}}.`$
(when $`\mathrm{\Omega }_\mathrm{\Lambda }=0`$, we get exactly the SEF solution (4.53) for $`\mathrm{\Omega }_m=1`$). It is worth to stress that using properties of the hypergeometric functions $`(f_{s_+},f_s_{})`$ it turns out that,
$$\overline{r}<\overline{r}_\mathrm{\Lambda },$$
(38)
which is also shown in the following figure.
### 3.2 The case $`\mathrm{\Omega }_k0`$
When $`\mathrm{\Omega }_k0`$ the Eq.(25) can be rewritten in the following form :
$`{\displaystyle \frac{d^2r}{da^2}}{\displaystyle \frac{\delta a+\frac{3}{2}}{a(aa_1)(aa_2)(a\overline{a}_2)}}{\displaystyle \frac{dr}{da}}+`$
(39)
$`+{\displaystyle \frac{3}{2}}\stackrel{~}{\alpha }{\displaystyle \frac{1}{a^2(aa_1)(aa_2)(a\overline{a}_2)}}r=0,`$
where $`a_1,a_2,a_2`$ are the roots of the equation $`\mathrm{\Omega }_m+\mathrm{\Omega }_ka+\mathrm{\Omega }_\mathrm{\Lambda }a^3=0`$, and $`\delta =\frac{\mathrm{\Omega }_k}{\mathrm{\Omega }_m}`$ (we are using the symbol $`r`$ for the Dyer-Roeder distance in this more general case). These roots are :
$$a_1=\frac{\left(\frac{2}{3}\right)^{1/3}\delta }{\sqrt{\mu }(9\sqrt{\mu }+\sqrt{3}\sqrt{4\delta ^3+27\mu })^{1/3}}+\frac{(9\sqrt{\mu }+\sqrt{3}\sqrt{4\delta ^3+27\mu })^{1/3}}{2^{1/3}3^{2/3}\sqrt{\mu }},$$
$$a_2=\frac{(1+i\sqrt{3})\delta }{2^{2/3}3^{1/3}\sqrt{\mu }(9\sqrt{\mu }+\sqrt{3}\sqrt{\delta ^3+27\mu })^{1/3}}$$
$$\frac{(1i\sqrt{3})(9\sqrt{\mu }+\sqrt{3}\sqrt{4\delta ^3+27\mu })^{1/3}}{22^{1/3}3^{2/3}\sqrt{\mu }},$$
$$\overline{a}_2=\frac{(1i\sqrt{3})\delta }{2^{2/3}3^{1/3}\sqrt{\mu }(9\sqrt{\mu }+\sqrt{3}\sqrt{4\delta ^3+27\mu })^{1/3}}$$
$$\frac{(1+i\sqrt{3})(9\sqrt{\mu }+\sqrt{3}\sqrt{4\delta ^3+27\mu })^{1/3}}{22^{1/3}3^{2/3}\sqrt{\mu }}.$$
where $`\mu =\mathrm{\Omega }_\mathrm{\Lambda }/\mathrm{\Omega }_n`$ as before. ($`a_2`$ and $`a_3`$ are complex conjugate and $`a_1<0`$, then Eq. (3.2) has non singular coefficients for real $`a>0`$). Equation (3.2) is of Fuchsian type (\[Ince 1956\],\[Tricomi 1961\]) with four regular singular points plus a regular singular point at infinity. This equation can be put in the following form
$`{\displaystyle \frac{d^2r}{da^2}}\left({\displaystyle \frac{\stackrel{~}{A}}{a}}+{\displaystyle \frac{\stackrel{~}{B}}{aa_1}}+{\displaystyle \frac{\stackrel{~}{C}}{aa_2}}+{\displaystyle \frac{\stackrel{~}{D}}{aa_3}}\right){\displaystyle \frac{dr}{da}}`$ $`+`$
$`{\displaystyle \frac{1}{a}}\left({\displaystyle \frac{A}{a}}+{\displaystyle \frac{B}{aa_1}}+{\displaystyle \frac{C}{aa_2}}+{\displaystyle \frac{D}{aa_3}}\right)r=0,`$
where the coefficient A, B, C, D, $`\stackrel{~}{A}`$, $`\stackrel{~}{B}`$, $`\stackrel{~}{C}`$, $`\stackrel{~}{D}`$ are easily found in terms of the cosmological parameters $`\mathrm{\Omega }_m`$, $`\mathrm{\Omega }_\mathrm{\Lambda }`$, $`\mathrm{\Omega }_k`$ and $`\stackrel{~}{\alpha }`$, and are given in Appendix A.
The solutions around each of the $`4+1`$ singularity points ( here ”$`1`$” denotes the singularity point relative to the infinity) are grouped together using the so called Riemann-P symbol.
$$𝒫\left(\begin{array}{cccccc}0& a_1& a_2& a_3& \mathrm{}& \\ \delta _{11}& \delta _{21}& \delta _{31}& \delta _{41}& \delta _1& a\\ \delta _{12}& \delta _{22}& \delta _{32}& \delta _{42}& \delta _2& \end{array}\right),$$
where the $`\delta _{ij}`$ and $`\delta _i`$ are given by
$`\delta _{11}={\displaystyle \frac{5}{2}}\sqrt{{\displaystyle \frac{25}{4}}{\displaystyle \frac{25\stackrel{~}{\alpha }}{4}}}(=2s_{}),`$
$`\delta _{12}={\displaystyle \frac{5}{2}}+\sqrt{{\displaystyle \frac{25}{4}}{\displaystyle \frac{25\stackrel{~}{\alpha }}{4}}}(=2s_+),`$
$`\delta _{21}=0,`$
$`\delta _{22}=1+{\displaystyle \frac{\frac{3}{2}\mathrm{\Omega }_m+\mathrm{\Omega }_ka_2}{a_1(a_2a_1)(a_2a_3)}},`$
$`\delta _{31}=0,`$
$`\delta _{32}=1+{\displaystyle \frac{\frac{3}{2}\mathrm{\Omega }_m+\mathrm{\Omega }_ka_1}{a_2(a_2a_1)(a_2a_3)}},`$
$`\delta _{41}=0,`$
$`\delta _{42}=1+{\displaystyle \frac{\frac{3}{2}\mathrm{\Omega }_m+\mathrm{\Omega }_ka_3}{a_3(a_3a_1)(a_3a_2)}},`$
$`\delta _1=0,`$
$`\delta _2=1,`$
and correspond, respectively, to the solutions of the indicial equation relative to the finite and infinite regular singular points of the equation. We do not discuss further properties of these solutions because unfortunately they cannot be given in an explicit analytical form. In the next section we will give an approximate analytical expression for the solution of equation (25).
## 4 The approximate lens equation
The exact solutions presented in the previous section are complicated and difficult to use in practical applications. Therefore we want to find an analytical approximate expression for $`r(z,\mathrm{\Omega }_m,\mathrm{\Omega }_\mathrm{\Lambda },\mathrm{\Omega }_k,\stackrel{~}{\alpha })`$ and for the function $`\chi `$ appearing in the cosmological lens equation. Following SEF let us briefly recall the basic equations used in the derivation of the cosmological lens equation and introduce appropriate notation. The time delay between different light rays reaching the observer is:
$$c\mathrm{\Delta }t=(1+z_d)\left\{\frac{D_dD_s}{D_{ds}}(\stackrel{}{\theta }\stackrel{}{\beta })\psi (\stackrel{}{\xi })\right\}+const,$$
(41)
where $`D_d`$, $`D_s`$ and $`D_{ds}`$ are correspondingly the angular diameter distances to deflector, source and the angular diameter distance between deflector and source.
The first term in the bracket represents the geometrical time delay and the second one is connected with the non homogeneous distribution of matter. To describe this effect we use a perturbed metric in the form :
$$ds^2=a^2(\eta )\left\{\left(1+\frac{2U}{c^2}\right)d\eta ^2\left(1\frac{2U}{c^2}\right)d\sigma ^2\right\},$$
where $`\eta `$ is the conformal time, $`U`$ is the gravitational potential of the deflector, and $`d\sigma ^2`$ is the spatial line element. The first term in Eq.(41) can be rewritten as :
$$(1+z_d)\frac{D_dD_s}{D_{ds}}=\frac{c}{H_0}[\chi (z_d)\chi (z_s)]^1,$$
(42)
where the function $`\chi `$ is given by :
$$\chi (z,\mathrm{\Omega }_m,\mathrm{\Omega }_\mathrm{\Lambda },\mathrm{\Omega }_k,\stackrel{~}{\alpha })=_z^{\mathrm{}}\frac{dz}{r^2g(z)}=_z^{\mathrm{}}\mathrm{exp}\left[\frac{\frac{d^2z}{d\lambda ^2}}{\left(\frac{dz}{d\lambda }\right)^2}\right]r^2𝑑z,$$
(43)
so it is connected with the general solution of Eq.(2) and the general expression for $`g(z)`$ (see Eq.25). Since the metric is conformally stationary, from the Fermat principle, we get the cosmological lens equation :
$$\stackrel{}{\beta }=\stackrel{}{\theta }\frac{2R_s}{cH_0}(1+z_d)[\chi (z_d)\chi (z_s)]\frac{\psi }{\stackrel{}{\theta }},$$
(44)
where $`R_s`$ is the Schwarzschild radius of the deflector. Denoting $`\stackrel{}{\xi }=D_d\stackrel{}{\theta }`$, $`\stackrel{}{\eta }=D_s\stackrel{}{\beta }`$, $`\stackrel{}{\alpha }(\stackrel{}{\xi })={\displaystyle \frac{2R_s}{D_d}}{\displaystyle \frac{\psi }{\stackrel{}{\theta }}}`$, we obtain
$$\stackrel{}{\eta }=\frac{D_s}{D_d}\stackrel{}{\xi }D_{ds}\stackrel{}{\alpha }(\stackrel{}{\xi }),$$
(45)
which is formally identical to the lens equation obtained for $`z<<1`$.
Let us first discuss the interesting range of the parameter space. From the available observations and theoretical considerations it follows that the radius of curvature of the universe is very large or equivalently that dimensionless curvature parameter $`\mathrm{\Omega }_k`$ is very small. In our considerations we assume that $`|\mathrm{\Omega }_k|0.05`$. Unfortunately no reliable estimates of $`\stackrel{~}{\alpha }`$ exist, but in the considered range of redshifts $`zϵ(0,100)`$, we consider the range of $`\stackrel{~}{\alpha }`$ $``$ $`[0.3,1]`$ (not completely clumpy Universe). We also allow different values of $`\mathrm{\Omega }_\mathrm{\Lambda }`$ and with special attention we treat the case $`\mathrm{\Omega }_\mathrm{\Lambda }=0.65`$ which seems to be singled out by observations (\[Kochanek 1996\], \[Kochanek 1996b\], \[Kochanek & al.1998\], \[Perlmutter 1997\],\[Tomita 1999\]). The density parameters satisfy the constraint $`\mathrm{\Omega }_m+\mathrm{\Omega }_\mathrm{\Lambda }+\mathrm{\Omega }_k=1`$.
In the considered range of redshifts and of the other cosmological parameters we propose to use the following form of the approximate function for $`r`$
$$r(z)=\frac{1}{\sqrt{d_1+d_2(d_3+d_4/z+z)^2}},$$
(46)
where the parameters $`d_1,d_2,d_3,d_4`$ depend on the cosmological parameters and they are determined by fitting Eq.(46) to the corresponding numerical solution. To optimize the fit we use a non linear regression procedure available in $`\mathrm{𝑀𝑎𝑡ℎ𝑒𝑚𝑎𝑡𝑖𝑐𝑎}\mathit{4.0}`$.
We use the same procedure to find the approximate function $`\chi (z)`$ which we take in the form
$$\chi (z)=\frac{1}{(e_1+e_2z+e_3z^2)},$$
(47)
where the parameters $`e_1,e_2,e_3`$ are determined by fitting $`\chi (z)`$ to numerically obtained exact values.
We are interested in finding an analytical approximation of the Dyer-Roeder distances for cosmological values of redshifts (say, for example, $`z0.05)`$; in this region the polynomial, which appears in the denominator of the approximate function $`\chi (z)`$ does not have zeros. In Fig.5 \- 6 we show the exact angular diameter distance and the fitted approximate relation for the same value of $`\stackrel{~}{\alpha }`$ and other cosmological parameters. (The calculations have been done for fixed $`\stackrel{~}{\alpha }`$, $`\mathrm{\Omega }_\mathrm{\Lambda }=0.65`$ and $`\mathrm{\Omega }_k=0`$.) In Fig.7 we see that the approximate angular diameter distance reproduces the exact curve with a very good accuracy, the error is not larger than $`1\%`$, for $`z`$ in the range (0, 50) and it is still good for larger $`z`$ in the range (50, 100). In Fig.8 we show exact numerically obtained function $`\chi (z)`$ and the fitted approximate relation for the same value of $`\stackrel{~}{\alpha }`$ and other cosmological parameters.
Finally we have analyzed how the typical maximum present in the angular diameter distance depends on the two cosmological parameters ($`\mathrm{\Omega }_\mathrm{\Lambda }`$, $`\stackrel{~}{\alpha }`$). Actually we perform this analysis both for the maximum of that distance $`r_{max}`$ and for the z-value for which that maximum occurs $`z_{max}`$. For $`\mathrm{\Omega }_k=0`$ and $`\mathrm{\Omega }_\mathrm{\Lambda }=const.`$, $`r_{max}`$ depends on $`\stackrel{~}{\alpha }`$ as:
$`r_{max}(\stackrel{~}{\alpha },\mathrm{\Omega }_\mathrm{\Lambda }=const.,\mathrm{\Omega }_k=0)=`$
$`=`$ $`\mathrm{exp}[\zeta _0+\zeta _1(\mathrm{\Omega }_\mathrm{\Lambda })+\zeta _2(\mathrm{\Omega }_\mathrm{\Lambda })^2+\zeta _3(\mathrm{\Omega }_\mathrm{\Lambda })^3]`$
If we fix $`\stackrel{~}{\alpha }=const.`$ and consider the dependence of $`r_{max}`$ on $`\mathrm{\Omega }_\mathrm{\Lambda }`$, we find:
$`r_{max}(\stackrel{~}{\alpha }=const.,\mathrm{\Omega }_\mathrm{\Lambda },\mathrm{\Omega }_k=0)=`$
$`=`$ $`\mathrm{exp}[\epsilon _0+\epsilon _1(\mathrm{\Omega }_\mathrm{\Lambda })+\epsilon _2(\mathrm{\Omega }_\mathrm{\Lambda })^2+\epsilon _3(\mathrm{\Omega }_\mathrm{\Lambda })^3],`$
where the coefficients $`\epsilon _i`$ depend on the clumpiness parameter $`\stackrel{~}{\alpha }`$ and on $`\mathrm{\Omega }_k`$. We can also study how $`z_{max}`$ depends on $`\stackrel{~}{\alpha }`$ and $`\mathrm{\Omega }_\mathrm{\Lambda }`$. For $`\stackrel{~}{\alpha }=const.`$, $`\mathrm{\Omega }_k=0`$, we have that
$`z_{max}(\stackrel{~}{\alpha }=const.,\mathrm{\Omega }_\mathrm{\Lambda },\mathrm{\Omega }_k=0)=`$
$`=`$ $`\mathrm{exp}[\tau _0+\tau _1(\mathrm{\Omega }_\mathrm{\Lambda })+\tau _2(\mathrm{\Omega }_\mathrm{\Lambda })^2+\tau _3(\mathrm{\Omega }_\mathrm{\Lambda })^3+\tau _4(\mathrm{\Omega }_\mathrm{\Lambda })^4+\tau _5(\mathrm{\Omega }_\mathrm{\Lambda })^5].`$
The coefficients $`\tau _i`$ depend on the clumpiness parameter $`\stackrel{~}{\alpha }`$ and on $`\mathrm{\Omega }_k`$. For $`\mathrm{\Omega }_\mathrm{\Lambda }=const.`$, $`\mathrm{\Omega }_k=0`$, we find
$`z_{max}(\stackrel{~}{\alpha },\mathrm{\Omega }_\mathrm{\Lambda }=const.,\mathrm{\Omega }_k=0)=`$
$`=`$ $`\mathrm{exp}[\gamma _0+\gamma _1\stackrel{~}{\alpha }+\gamma _2(\stackrel{~}{\alpha })^2+\gamma _3(\stackrel{~}{\alpha })^3+\gamma _4(\stackrel{~}{\alpha })^4+\gamma _5(\stackrel{~}{\alpha })^5].`$
where the $`\gamma _i`$ coefficients depend on and the parameters $`\mathrm{\Omega }_\mathrm{\Lambda }`$ and $`\mathrm{\Omega }_k`$. From Eq. (4) it follows that for $`\mathrm{\Omega }_\mathrm{\Lambda }=0.65`$ and $`\stackrel{~}{\alpha }=0.9`$, we get $`z_{max}=1.62`$. When $`\mathrm{\Omega }_\mathrm{\Lambda }=0`$, $`\mathrm{\Omega }_k=0`$, as it follows from Eq. (27) $`z_{max}=1.25`$. It is interesting to note that $`z_{max}(\mathrm{\Omega }_\mathrm{\Lambda }=0,\mathrm{\Omega }_k=0,\stackrel{~}{\alpha }=0.9)<z_{max}(\mathrm{\Omega }_\mathrm{\Lambda }=0.65,\mathrm{\Omega }_k=0,\stackrel{~}{\alpha }=0.9)`$, which shows how relevant is the role of the cosmological constant.
## 5 Conclusion
In this paper we have considered the cosmological lens equation in the case when the cosmological constant $`\mathrm{\Lambda }0`$ and the curvature of space is different for zero ($`k0`$). We have included the effects of non homogeneous distribution of matter which are described by a phenomenological parameter $`\stackrel{~}{\alpha }`$. Unfortunately at the moment there are no generally accepted models that describe the distribution of baryonic and dark matter at high redshifts and therefore the influence of non homogeneities of matter distribution can be included only at this approximate level. Following the standard procedure (\[Schneider, Ehlers & Falco 1992\]) we use the Dyer-Roeder distance to find the distance between two objects with redshifts $`z_1`$ and $`z_2`$.
To find the general solution we slightly enlarged the parameter space by considering the non zero cosmological constant, so the cosmological model is described by the following parameters $`\mathrm{\Omega }_\mathrm{\Lambda }0`$, $`\mathrm{\Omega }_m0`$, $`\stackrel{~}{\alpha }[0.3,1]`$ (not completely clumpy Universe), $`p=0`$ (no radiation). The solution that we have found is a functional combination of the solution given in SEF for the case when $`\mathrm{\Omega }_\mathrm{\Lambda }=0`$, $`\mathrm{\Omega }_k=0`$ and two hypergeometric functions. Fig. 1 shows that in the flat universe the Dyer-Roeder distance increases for increasing values of the cosmological constant. In the next step we allow the curvature of space to be different from zero. This further complicates the Dyer-Roeder equation which becomes of a Fuchsian type with 4 regular singular points and one regular singular point at infinity. The general form of this kind of ordinary differential equations is given in terms of the P-Riemann symbol.
The obtained exact solution is so complicated that it is practically useless in practical applications. Therefore we have looked for an approximate analytic solution simple enough to be used in many applications and at the same time sufficiently accurate at least in the interesting range of redshifts. The proposed form of the approximate solution of the Dyer-Roeder equation depends on four arbitrary parameters. To fix values of these parameters we fit the approximate solution to the exact one with the help of a non linear regression method (see Mathematica 4.0). Of course, the parameters depend on the density parameters, but their values for different cosmological parameters can be tabulated. Following SEF we have also found the function $`\chi `$ which appears in the expression for time delays as well as in the lens equation. We have also proposed approximate analytical form of the function $`\chi `$ which contains three arbitrary parameters. To find values of these parameters we use the same method as above.
We have also studied how the redshift corresponding to the maximal angular diameter distance depends on the basic parameters determining the cosmological model.
Let us now consider the following three combinations of the DR distance:
$``$ $`{\displaystyle \frac{D_{LS}}{D_{OS}}}={\displaystyle \frac{H_0}{c}}{\displaystyle \frac{r_{LS}}{r_{OS}}},`$
$``$ $`{\displaystyle \frac{D_{OL}D_{LS}}{D_{OS}}}={\displaystyle \frac{H_0}{c}}{\displaystyle \frac{r_{OL}r_{LS}}{r_{OS}}},`$
$``$ $`{\displaystyle \frac{D_{OL}D_{OS}}{D_{LS}}}={\displaystyle \frac{H_0}{c}}{\displaystyle \frac{r_{OL}r_{OS}}{r_{LS}}}.`$
We have selected these combinations because of the role they play, respectively, in bending of light, lensing statistics, and time delay. Asada (\[Asada 1997\], \[Asada 1998\], \[Asada 1998\]) have found that:
$``$ $`{\displaystyle \frac{D_{LS}}{D_{OS}}}(\alpha _1)<{\displaystyle \frac{D_{LS}}{D_{OS}}}(\alpha _2),\mathrm{for}\alpha _1<\alpha _2,`$
$``$ $`{\displaystyle \frac{D_{OL}D_{LS}}{D_{OS}}}(\alpha _1)<{\displaystyle \frac{D_{OL}D_{LS}}{D_{OS}}}(\alpha _2),\mathrm{for}\alpha _1<\alpha _2,`$
$``$ $`{\displaystyle \frac{D_{OL}D_{OS}}{D_{LS}}}(\alpha _1)>{\displaystyle \frac{D_{OL}D_{OS}}{D_{LS}}}(\alpha _2),\mathrm{for}\alpha _1<\alpha _2.`$
Using our approximate relations for $`r`$ and $`\chi `$ we obtain that:
$`{\displaystyle \frac{D_{LS}}{D_{OS}}}`$ $`=`$ $`(1+z_L)r(z_L)(\chi (z_L)\chi (z_S))=`$
$`=`$ $`(1+z_L){\displaystyle \frac{1}{\sqrt{d_1+d_2(d_3+d_4/z_L+z_L)^2)}}}({\displaystyle \frac{1}{(e_1+e_2z_L+e_3z_L^2)}}`$
$`{\displaystyle \frac{1}{(e_2+e_2z_S+e_3z_S^2)}}),`$
$`{\displaystyle \frac{D_{OL}D_{LS}}{D_{OS}}}`$ $`=`$ $`(1+z_L)r^2(z_L)(\chi (z_L)\chi (z_S))=`$
$`=`$ $`(1+z_L){\displaystyle \frac{1}{(d_1+d_2(d_3+d_4/z_L+z_L)^2)}}({\displaystyle \frac{1}{(e_1+e_2z_L+e_3z_L^2)}}`$
$`{\displaystyle \frac{1}{(e_1+e_2z_S+e_3z_S^2)}}),`$
$`{\displaystyle \frac{D_{OL}D_{OS}}{D_{LS}}}`$ $`=`$ $`{\displaystyle \frac{1}{(1+z_L)(\chi (z_L)\chi (z_S))}}=`$
$`=`$ $`{\displaystyle \frac{1}{(1+z_L)}}{\displaystyle \frac{1}{(\frac{1}{(e_1+e_2z_L+e_3z_L^2)}\frac{1}{(e_1+e_2z_S+e_3z^2)})}},`$
Using the fitted values for the parameters $`(d_1,d_2,d_3,d_4)`$ and $`(e_1,e_2,e_3)`$ we confirm results of Asada for $`\stackrel{~}{\alpha }`$ in the considered range $`1\stackrel{~}{\alpha }0.3`$.
Finally we like to stress that from our analysis it follows that variations in the angular diameter distance caused by the presence of cosmological constant are quite similar to variations due to changes in the value of $`\stackrel{~}{\alpha }`$.
Actually, in Fig.9, we consider two plots corresponding to the values $`\mathrm{\Omega }_\mathrm{\Lambda }=0`$ and $`\stackrel{~}{\alpha }0`$ and $`\mathrm{\Omega }_\mathrm{\Lambda }0`$ and $`\stackrel{~}{\alpha }=1`$ and we see that the $`\stackrel{~}{\alpha }`$ parameter could mimic the effect of a non zero cosmological constant. This is an important observation in view of the recent observational results concerning the non zero value of the cosmological constant (\[Kochanek 1996\], \[Kochanek & al.1998\], \[Kochanek 1996b\], \[Perlmutter 1997\]). In the future we would like to extend our work and include also the radiation density parameter.
Acknowledgments.
It is a pleasure to thank V.F. Cardone , G. Covone. C. Rubano and P. Scudellaro for discussions we had on the manuscript. RdR and AAM are financially sustained by the M.U.R.S.T. grant PRIN97 “SIN.TE.SI“ , MD by the grant 2-P03D-014-17 of the Polish State Committee for Scientific Research, and EP by the Social European Committee.
## 6 Appendix A
$`A`$ $`=`$ $`{\displaystyle \frac{3\stackrel{~}{\alpha }\mathrm{\Omega }_m}{2a_1a_2a_3}},`$
$`B`$ $`=`$ $`{\displaystyle \frac{3\stackrel{~}{\alpha }\mathrm{\Omega }_m}{2\alpha _9}}\times {\displaystyle \frac{\alpha _5(\alpha _3\alpha _4)+\alpha _7(\alpha _1\alpha _4)\alpha _8\times (\alpha _1\alpha _3)}{(\alpha _6(\alpha _3\alpha _4)+\alpha _7(\alpha _2\alpha _4)\alpha _8(\alpha _2\alpha _3))}},`$
$`C`$ $`=`$ $`(A+B+D),`$ (52)
$`D`$ $`=`$ $`{\displaystyle \frac{(A(\alpha _1\alpha _3)+B(\alpha _2\alpha _3))}{(\alpha _3\alpha _4)}}.`$
The coefficients $`\stackrel{~}{A}`$, $`\stackrel{~}{B}`$, $`\stackrel{~}{C}`$, $`\stackrel{~}{D}`$ are
$`\stackrel{~}{A}={\displaystyle \frac{3}{2}}{\displaystyle \frac{\mathrm{\Omega }_m}{a_1a_2a_3}},`$
$`\stackrel{~}{B}={\displaystyle \frac{\alpha _3\alpha _5\mathrm{\Omega }_m+3\alpha _4\alpha _5\mathrm{\Omega }_m}{2\alpha _9(\alpha _3\alpha _6\alpha _4\alpha _2\alpha _7+\alpha _2\alpha _8\alpha _2\alpha _8)}}+`$
$`+{\displaystyle \frac{3\alpha _1\alpha _7\mathrm{\Omega }_m3\alpha _4\alpha _7\mathrm{\Omega }_m+3\alpha _1\alpha _8\mathrm{\Omega }_m}{(2\alpha _9(\alpha _3\alpha _6\alpha _4\alpha _2\alpha _7+\alpha _2\alpha _8\alpha _2\alpha _8))}}+`$
$`+{\displaystyle \frac{2\alpha _7\alpha _9\mathrm{\Omega }_k2\alpha _8\alpha _9\mathrm{\Omega }_k}{(2\alpha _9(\alpha _3\alpha _6\alpha _4\alpha _2\alpha _7+\alpha _2\alpha _8\alpha _2\alpha _8))}},`$ (53)
$`\stackrel{~}{C}={\displaystyle \frac{\alpha _2\alpha _5\mathrm{\Omega }_m+3\alpha _4\alpha _5\mathrm{\Omega }_m}{2\alpha _9(\alpha _3\alpha _6\alpha _4\alpha _6\alpha _2\alpha _7+\alpha _4\alpha _7+\alpha _2\alpha _8\alpha _3\alpha _8)}}`$
$`{\displaystyle \frac{3\alpha _1\alpha _6\mathrm{\Omega }_m+3\alpha _2\alpha _6\mathrm{\Omega }_m}{(2\alpha _9(\alpha _3\alpha _6\alpha _4\alpha _6\alpha _2\alpha _7+\alpha _4\alpha _7+\alpha _2\alpha _8\alpha _3\alpha _8))}}+`$
$`+{\displaystyle \frac{3\alpha _1\alpha _8\mathrm{\Omega }_m3\alpha _2\alpha _8\mathrm{\Omega }_m2\alpha _6\alpha _9\mathrm{\Omega }_k+2\alpha _8\alpha _9\mathrm{\Omega }_m}{(2\alpha _9(\alpha _3\alpha _6\alpha _4\alpha _6\alpha _2\alpha _7+\alpha _4\alpha _7+\alpha _2\alpha _8\alpha _3\alpha _8))}},`$
$`\stackrel{~}{D}={\displaystyle \frac{3\alpha _2\alpha _5\mathrm{\Omega }_m3\alpha _3\alpha _5\mathrm{\Omega }_m+3\alpha _3\alpha _6\mathrm{\Omega }_m}{2\alpha _9(\alpha _3\alpha _6+\alpha _4\alpha _6+\alpha _2\alpha _7\alpha _4\alpha _7\alpha _2\alpha _8+\alpha _3\alpha _8)}}+`$
$`+{\displaystyle \frac{3\alpha _1\alpha _7\mathrm{\Omega }_m3\alpha _2\alpha _7\mathrm{\Omega }_m2\alpha _6\alpha _8\mathrm{\Omega }_m2\alpha _6\alpha _9\mathrm{\Omega }_k}{(2\alpha _9(\alpha _3\alpha _6+\alpha _4\alpha _6+\alpha _2\alpha _7\alpha _4\alpha _7\alpha _2\alpha _8+\alpha _3\alpha _8))}}+`$
$`+{\displaystyle \frac{2\alpha _8\alpha _9\mathrm{\Omega }_k+2\alpha _7\alpha _9\mathrm{\Omega }_m}{(2\alpha _9(\alpha _3\alpha _6+\alpha _4\alpha _6+\alpha _2\alpha _7\alpha _4\alpha _7\alpha _2\alpha _8+\alpha _3\alpha _8))}}.`$
The functions $`\alpha _i`$ being:
$`\alpha _1=a_1a_2+a_1a_3+a_2a_3,`$
$`\alpha _2=a_2a_3,`$
$`\alpha _3=a_1a_3,`$
$`\alpha _4=a_1a_2,`$
$`\alpha _5=a_1+a_2+a_3,`$
$`\alpha _6=a_2+a_3,`$
$`\alpha _7=a_1+a_3,`$
$`\alpha _8=a_1+a_2,`$
$`\alpha _9=a_1a_2a_3.`$
## 7 Appendix B
We give the form of the coefficient appearing in equations (46) and (47) , in function of the cosmological parameter. We show the dependence of the coefficient $`d_i`$ on the clumpiness parameters $`\stackrel{~}{\alpha }`$, fixing $`\mathrm{\Omega }_\mathrm{\Lambda }=0.65`$
$`d_1=\mathrm{exp}[a_{d1}+b_{d1}\stackrel{~}{\alpha }+c_{d1}\stackrel{~}{\alpha }^2+d_{d1}\stackrel{~}{\alpha }^3],`$
$`d_2=\mathrm{exp}[a_{d2}+b_{d2}\stackrel{~}{\alpha }+c_{d2}\stackrel{~}{\alpha }^2+d_{d2}\stackrel{~}{\alpha }^3],`$
$`d_3=\mathrm{exp}[a_{d3}+b_{d3}\stackrel{~}{\alpha }+c_{d3}\stackrel{~}{\alpha }^2+d_{d3}\stackrel{~}{\alpha }^3],`$
$`d_4=\mathrm{exp}[a_{d4}+b_{d4}\stackrel{~}{\alpha }+c_{d4}\stackrel{~}{\alpha }^2+d_{d4}\stackrel{~}{\alpha }^3].`$
Dependence of $`d_i`$ functions from $`\mathrm{\Omega }_\mathrm{\Lambda }`$ with $`\stackrel{~}{\alpha }=0.8`$
$`d_1=\mathrm{exp}[p_{d1}+q_{d1}\mathrm{\Omega }_\mathrm{\Lambda }+r_{d1}(\mathrm{\Omega }_\mathrm{\Lambda })^2+s_{d1}(\mathrm{\Omega }_\mathrm{\Lambda })^3],`$
$`d_2=\mathrm{exp}(p_{d2}+q_{d2}\mathrm{\Omega }_\mathrm{\Lambda }+r_{d2}(\mathrm{\Omega }_\mathrm{\Lambda })^2+s_{d2}(\mathrm{\Omega }_\mathrm{\Lambda })^3),`$
$`d_3=\mathrm{exp}(p_{d3}+q_{d3}\mathrm{\Omega }_\mathrm{\Lambda }+r_{d3}(\mathrm{\Omega }_\mathrm{\Lambda })^2+s_{d3}(\mathrm{\Omega }_\mathrm{\Lambda })^3),`$
$`d_4=\mathrm{exp}[p_{d4}+q_{d4}\mathrm{\Omega }_\mathrm{\Lambda }+r_{d4}(\mathrm{\Omega }_\mathrm{\Lambda })^2+s_{d4}(\mathrm{\Omega }_\mathrm{\Lambda })^3],`$
being $`𝖯_n(\mathrm{\Omega }_\mathrm{\Lambda })`$ a polynomial of n-degree in $`\mathrm{\Omega }_\mathrm{\Lambda }`$.
We see below the dependence of the functions $`e_i`$ from $`\stackrel{~}{\alpha }`$ with $`\mathrm{\Omega }_\mathrm{\Lambda }=0.65`$
$`e_1=m_{e1}+n_{e1}\stackrel{~}{\alpha }+o_{e1}\stackrel{~}{\alpha }^2+p_{e1}\stackrel{~}{\alpha }^3,`$
$`e_2=m_{e2}+n_{e2}\stackrel{~}{\alpha }+o_{e2}\stackrel{~}{\alpha }^2+p_{e2}\stackrel{~}{\alpha }^3,`$
$`e_3=m_{e3}+n_{e3}\stackrel{~}{\alpha }.`$
Here, we show the dependence of the functions $`e_i`$ from $`\mathrm{\Omega }_\mathrm{\Lambda }`$,with $`\stackrel{~}{\alpha }=0.9`$, and $`\mathrm{\Omega }_k=0`$:
$`e_1=\stackrel{~}{m_{e1}}+\stackrel{~}{n_{e1}}\mathrm{\Omega }_\mathrm{\Lambda }+\stackrel{~}{o_{e1}}\mathrm{\Omega }_\mathrm{\Lambda }^2+\stackrel{~}{p_{e1}}\mathrm{\Omega }_\mathrm{\Lambda }^3,`$
$`e_2=\stackrel{~}{m_{e2}}+\stackrel{~}{n_{e2}}\mathrm{\Omega }_\mathrm{\Lambda }+\stackrel{~}{o_{e2}}\mathrm{\Omega }_\mathrm{\Lambda }^2+\stackrel{~}{p_{e2}}\mathrm{\Omega }_\mathrm{\Lambda }^3,`$
$`e_3=𝖯_q(\mathrm{\Omega }_\mathrm{\Lambda })/(\mathrm{exp}[\stackrel{~}{m_{e3}}\mathrm{\Omega }_\mathrm{\Lambda }+\stackrel{~}{n_{e3}]}),`$
being $`𝖯_q(\mathrm{\Omega }_\mathrm{\Lambda })`$ a polynomial of q-degree in $`\mathrm{\Omega }_\mathrm{\Lambda }`$ depending on the value of $`k`$
|
warning/0004/math0004074.html
|
ar5iv
|
text
|
# Dickson Invariants in the image of the Steenrod Square
## 1. Introduction
A polynomial in $`𝔽_2[X_1,X_2,\mathrm{},X_n]`$ is hit if it is in the image of the summation of the Steenrod square: $`_{i1}\mathrm{Sq}^i`$. Let $`D_n`$ be the Dickson invariant algebra of $`n`$-variables. In this paper, we will prove the following,
###### Theorem 1.1.
When $`n>3`$, each polynomial in the Dickson invariant ring $`D_n`$ is hit.
In , Hung studies the Dickson invariants in the image of the Steenrod square. Since it is trivial that $`D_1`$ and $`D_2`$ are not hit, the problem starts interesting from $`n=3`$. In the same paper, Hung shows that each element in $`D_3`$ is hit and conjectured that it is true for $`D_{n>3}`$. So our result provides a positive answer to the conjecture, which supports to the positive answer of the conjecture on the spherical classes: there are no spherical classes in $`Q_0S^0`$, except the Hopf invariant one and Kervaire invariant one elements. We refer to and an excellent expository paper , p501 for more background regarding to this conjecture.
###### Remark 1.2.
Recently, K. F. Tan and the author has obtained an elementary proof of the case $`n=3`$.
## 2. Proof of Theorem 1.1
We first recall some basic properties regarding the Dickson algebra.
Write $`V_n`$ for the product
$$\underset{\alpha _i\{0,1\},i=1,..,n1}{}(\alpha _1x_1+\mathrm{}+\alpha _{n1}x_{n1}+x_n).$$
Then we have the following theorem.
###### Theorem 2.1 (Hung ).
$$\mathrm{Sq}^iV_n=\{\begin{array}{cc}V_n\text{ if }i=0\hfill & \\ V_nQ_{n1,s}\text{ if }i=2^{n1}2^s\text{}0sn1\hfill & \\ V_n^2\text{ if }i=2^{n1}\hfill & \\ 0\text{ otherwise.}\hfill & \end{array}$$
$$\mathrm{Sq}^iQ_{n,s}=\{\begin{array}{cc}Q_{n,r}\text{ if }i=2^s2^r\text{}rs\hfill & \\ Q_{n,r}Q_{n,t}\text{ if }i=2^n2^t+2^s2^r\text{}rs<t\hfill & \\ Q_{n,s}^2\text{ if }i=2^n2^s\text{ }\hfill & \\ 0\text{ otherwise.}\hfill & \end{array}$$
In the following, we will frequently use the above results without mentioning each time.
We use the induction on $`n`$ to prove Theorem 1.1. Suppose that the statement is true for $`n`$. Then we will prove that each polynomial in $`D_{n+1}`$ is hit.
Recall that
$$Q_{n+1,k}=Q_{n,k1}^2+V_{n+1}Q_{n,k}\text{ for }1kn.$$
So any monomial in $`𝔽_2[Q_{n+1,0},Q_{n+1,1},\mathrm{},Q_{n+1,n}]`$ can be written as the summation of the following form:
$$A:=V_{n+1}^aQ_{n,0}^{n_0}Q_{n,1}^{n_1}Q_{n,2}^{n_2}\mathrm{}Q_{n,n1}^{n_{n1}}.$$
Hence by the hypothesis of the induction, it is sufficient to show that $`A`$ is hit for any $`a>0`$. Notice that
(1)
$$V_{n+1}=\underset{s=1}{\overset{n}{}}\mathrm{Sq}^1(Q_{n,s}X_{n+1}^{2^s1}).$$
When $`n_1`$ is even, we have the hit polynomial
$$A=\mathrm{Sq}^1[\left(\underset{s=1}{\overset{n}{}}Q_{n,s}x_{n+1}^{2^s1}\right)V_{n+1}^{a1}Q_{n,0}^{n_0}Q_{n,1}^{n_1}Q_{n,2}^{n_2}\mathrm{}Q_{n,n1}^{n_{n1}}].$$
If $`n_1`$ is odd and $`n_2`$ is even, then $`A`$ can be written as the hit polynomial:
$$\begin{array}{c}\mathrm{Sq}^2[V_{n+1}^aQ_{n,0}^{n_0}Q_{n,1}^{n_11}Q_{n,2}^{n_2+1}\mathrm{}Q_{n,n1}^{n_{n1}}]\hfill \\ \hfill +\mathrm{Sq}^1[\left(\underset{s=1}{\overset{n}{}}Q_{n,s}x_{n+1}^{2^s1}\right)V_{n+1}^{a1}Q_{n,0}^{n_0}(\mathrm{Sq}^1Q_{n,1}^{\frac{n_11}{2}})^2Q_{n,2}^{n_2+1}\mathrm{}Q_{n,n1}^{n_{n1}}]\end{array}$$
In the following, we will always assume that $`n_1`$ and $`n_2`$ are both odd.
When $`n=3`$, $`n_0`$ is even and $`a`$ is odd, we have
$$\begin{array}{ccc}A\hfill & =\hfill & (V_4^{a1}\mathrm{Sq}^4V_4)Q_{3,0}^{n_0}Q_{3,1}^{n_1}Q_{3,2}^{n_21}\hfill \\ & \hfill & V_4\chi (\mathrm{Sq}^4)[V_4^{a1}Q_{3,0}^{n_0}Q_{3,1}^{n_1}Q_{3,2}^{n_21}]\text{ (modulo the hits)}\hfill \\ & \hfill & V_4^aQ_{3,1}\left(\mathrm{Sq}^2[Q_{3,0}^{\frac{n_0}{2}}Q_{3,1}^{\frac{n_11}{2}}Q_{3,2}^{\frac{n_21}{2}}]\right)^2\text{ (modulo the hits)}.\hfill \end{array}$$
Using the previous observation, the last polynomial is hit, since the order of $`Q_{3,2}`$ is even.
When $`n=3`$, $`n_0`$ is even and $`a`$ is even, notice that
$$Q_{3,0}^{n_0}Q_{3,1}^{n_1}Q_{3,2}^{n_2}=Q_{3,0}^{n_0}Q_{3,1}^{n_11}Q_{3,2}^{n_21}\mathrm{Sq}^4Q_{3,1}.$$
Then using the $`\chi `$-trick and doing some basic computation, we can see that the monomial $`Q_{3,0}^{n_0}Q_{3,1}^{n_1}Q_{3,2}^{n_2}`$ is in the image of $`_{i=1}^4\mathrm{Sq}^i`$. In fact,
$$\begin{array}{ccc}Q_{3,1}\chi (\mathrm{Sq}^4)[Q_{3,0}^{n_0}Q_{3,1}^{n_11}Q_{3,2}^{n_21}]\hfill & & \\ =[\mathrm{Sq}^2Q_{3,2}][\mathrm{Sq}^2(Q_{3,0}^{\frac{n_0}{2}}Q_{3,1}^{\frac{n_11}{2}}Q_{3,2}^{\frac{n_21}{2}})]^2\hfill & & \\ Q_{3,2}\chi (\mathrm{Sq}^2)[Q_{3,0}^{\frac{n_0}{2}}Q_{3,1}^{\frac{n_11}{2}}Q_{3,2}^{\frac{n_21}{2}}]^2\text{ (modulo the hits)}\hfill & & \\ =(Q_{2,1}^2+V_3)[\mathrm{Sq}^1\mathrm{Sq}^2(Q_{3,0}^{\frac{n_0}{2}}Q_{3,1}^{\frac{n_11}{2}}Q_{3,2}^{\frac{n_21}{2}})]\hfill & & \\ =\mathrm{Sq}^2\left(Q_{2,1}[\mathrm{Sq}^1\mathrm{Sq}^2(Q_{3,0}^{\frac{n_0}{2}}Q_{3,1}^{\frac{n_11}{2}}Q_{3,2}^{\frac{n_21}{2}})^2]\right)\hfill & & \\ +\mathrm{Sq}^1\left\{(Q_{2,1}X_3+X_3^3)[\mathrm{Sq}^1\mathrm{Sq}^2(Q_{3,0}^{\frac{n_0}{2}}Q_{3,1}^{\frac{n_11}{2}}Q_{3,2}^{\frac{n_21}{2}})]^2\right\},\hfill & & \end{array}$$
where we have used (1) in the last equality. On the other hand, $`\mathrm{Sq}^iV_4^a=0`$ for $`i=1,2,3`$ and $`4`$. Therefore using the $`\chi `$-trick , we know that $`A`$ is hit.
When $`n=3`$, $`n_0`$ is odd and $`a`$ is odd, the polynomial $`A`$ equals
$$V_4^{a1}(\mathrm{Sq}^7V_4)Q_{3,0}^{n_01}Q_{3,1}^{n_1}Q_{3,2}^{n_2}.$$
From the discussion above, we know that
$$Q_{3,0}^{n_01}Q_{3,1}^{n_1}Q_{3,2}^{n_2}$$
is in the image of $`_{i=1}^4\mathrm{Sq}^i`$. On the other hand, $`\chi (\mathrm{Sq}^i)(V_4^{a1}\mathrm{Sq}^7V_4)=0`$ for $`i=1,2,3`$ and $`4`$. So using the $`\chi `$-trick , we conclude that $`A`$ is hit.
When $`n=3`$, $`n_0`$ is odd and $`a`$ is even, let $`\nu `$ be the integer such that $`a=2^\nu b`$ where $`b`$ is odd. Then
$$V_4^a=\mathrm{Sq}^{4a}\mathrm{Sq}^{2a}\mathrm{}\mathrm{Sq}^{8b}V_4^b.$$
Hence
$$\begin{array}{ccc}A\hfill & =\hfill & (\mathrm{Sq}^{4a}\mathrm{Sq}^{2a}\mathrm{}\mathrm{Sq}^{8b}V_4^b)(Q_{3,0}^{n_0}Q_{3,1}^{n_1}Q_{3,2}^{n_2})\hfill \\ & \hfill & V_4^b\chi (\mathrm{Sq}^{8b})\mathrm{}\chi (\mathrm{Sq}^{2a})\chi (\mathrm{Sq}^{4a})(Q_{3,0}^{n_0}Q_{3,1}^{n_1}Q_{3,2}^{n_2})\text{ (modulo the hits)}.\hfill \end{array}$$
After expanding the last polynomial using Theorem 2.1, it is easy to see that each resulting term belongs to one of the previous cases. Therefore $`A`$ is hit.
When $`n4`$, the polynomial $`A`$ takes the following form,
(2)
$$V_{n+1}^a(\mathrm{Sq}^{2^n4}Q_{n,1})Q_{n,0}^{n_0}Q_{n,1}^{n_11}Q_{n,2}^{n_21}\mathrm{}Q_{n,n1}^{n_{n1}}.$$
Using a result of Don Davis, Theorem 2. of and the $`\chi `$-trick , we know that it is sufficient to show the polynomial:
$$Q_{n,1}\mathrm{Sq}^{2^{n1}}\mathrm{}\mathrm{Sq}^8\chi (\mathrm{Sq}^4)\left\{V_{n+1}^aQ_{n,0}^{n_0}Q_{n,1}^{n_11}Q_{n,2}^{n_21}\mathrm{}Q_{n,n1}^{n_{n1}}\right\}$$
is hit.
After expansion using the Steenrod operation, the above polynomial can be written as the summation of the form:
$$V_{n+1}^aQ_{n,0}^{k_0}Q_{n,1}^{k_1}Q_{n,2}^{k_2}\mathrm{}Q_{n,n1}^{k_{n1}}.$$
Using the previous discussion, we can conclude that all these polynomials are hit, except for those when $`k_1`$ and $`k_2`$ are both odd. But in this case, we can replace $`n_i`$ by $`k_i`$ for all $`i`$ in (2) and carry out the above process again. After using this process sufficiently many times with modulo the hits, we can conclude that the new $`k_0`$, $`k_1`$ and $`k_2`$ are independent of the process. To keep $`k_0`$, $`k_1`$ and $`k_2`$ unchanged with the process, we must require that
$$\mathrm{Sq}^{2^{n1}}\mathrm{}\mathrm{Sq}^8\chi (\mathrm{Sq}^4)\left\{V_{n+1}^aQ_{n,3}^{k_3}\mathrm{}Q_{n,n1}^{k_{n1}}\right\}\text{ (modulo the hits)}$$
contributes $`Q_{n,2}`$ after each process is done, since for $`j2^{n1}`$ and $`t<n`$ , $`\mathrm{Sq}^jQ_{n,0}=Q_{n,0}Q_{n,t}`$ only if $`j=t=n1(>2)`$. Finally because all $`k_i`$ ($`0i<n`$) are finite, we conclude that $`A`$ is hit after carrying on the process further for enough many times.
|
warning/0004/hep-th0004010.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Substantial progress in multiloop Feynman diagram calculations in recent years requires computation of scalar master integrals. Often the problem involving different mass scales can be reduced (e.g. by expanding) to integrals depending only on a single scale. Thus single-scale diagrams (e.g. bubbles with one non-zero mass, massless self-energy, massive on-shell self-energy integrals, etc.) form an important class of Feynman diagrams. Such integrals arise for example in renormalization group calculations. The structure of massless integrals is well understood now . In particular recently a correspondence between knot theory and massless diagrams has been found which can serve as a very useful guide to find the transcendental numbers which occur with rational coefficient in the counterterms. This relationship is known only for diagrams that are free of subdivergences.
The transcendental structure of massive single-scale diagrams is less investigated - (see also , ). In particular, we do not know whether there exist a theory to predict the transcendental numbers for these diagrams. Recently it was observed that all two-loop massive on-shell diagrams of propagator type without subdivergences can be written in following way
$$m^2𝐈_0|_{p^2=m^2}=r_1\zeta _3+r_2\pi \text{Ls}_2\left(\frac{\pi }{3}\right)+r_3i\pi \zeta _2+𝒪(\epsilon ),$$
(1)
where $`\zeta _a=\zeta (a)`$ is the Riemann $`\zeta `$-function, $`r_j`$ are rational coefficients and definition of $`\text{Ls}_n\left(z\right)`$ is given by (3). This observation suggests a conjecture that irrationalities occurring in these diagrams are defined by the topology of a diagram but not e.g. by the distribution of the masses on lines. In this paper we test this conjecture in the next order of the $`\epsilon `$-expansion.
Another problem under consideration is the test of the hypothesis about the connection between transcendental numbers occurring in the $`\epsilon `$-expansion of diagrams and the presence of certain massive-particles-cuts. This conjecture reads as follows: zero-, one- and three massive particle cuts give rise to appearance of structures $`\pi ^j\zeta _m(\mathrm{ln}2)^n\text{Li}_p\left(1/2\right)`$, where $`\mathrm{Li}_p(x)`$ is polylogarithm, or more complicated structures associated with Euler–Zagier sums (or multidimensional zeta/harmonic sums)
$`\zeta (a_1,\mathrm{},a_k)={\displaystyle \underset{n_i>n_{i+1}}{}}{\displaystyle \underset{j=1}{\overset{k}{}}}{\displaystyle \frac{(\mathrm{sign}a_j)^{n_i}}{n_i^{|a_j|}}},`$ (2)
whereas the two massive particle particle cuts bring other transcendental numbers connected with “sixth root of unity” : $`\left(\frac{\pi }{\sqrt{3}}\right)^k\zeta _m(\mathrm{ln}3)^n\text{Ls}_p\left(z_i\right)\text{Ls}_q^{(r)}\left(z_j\right)`$, where $`z_k=\{\frac{\pi }{3},\frac{2\pi }{3}\}`$ and $`\text{Ls}_n\left(z\right)`$ and $`\text{Ls}_n^{(m)}\left(z\right)`$ are so-called log-sine integrals defined by
$$\text{Ls}_n\left(\theta \right)=\underset{0}{\overset{\theta }{}}\mathrm{ln}^{n1}\left|2\mathrm{sin}\frac{\varphi }{2}\right|d\varphi ,\text{Ls}_n^{(m)}\left(\theta \right)=\underset{0}{\overset{\theta }{}}\varphi ^m\mathrm{ln}^{nm1}\left|2\mathrm{sin}\frac{\varphi }{2}\right|d\varphi .$$
(3)
We will show here that some of these irrationalities are related to sums -
$$\underset{n=1}{\overset{\mathrm{}}{}}\frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}\frac{1}{n^c}\underset{a,b,i,j}{}\left[\underset{m=1}{\overset{n1}{}}\frac{1}{m^a}\right]^i\left[\underset{k=1}{\overset{2n1}{}}\frac{1}{k^b}\right]^j$$
(4)
which we call multiple binomial sums.
The question about transcendental structures connected with four- and more massive particle cuts remains open.
## 2 Results
As examples we consider the diagrams shown in Fig.1. All diagrams posses different cuts in the external variable $`p^2`$ with following values of thresholds $`I_{125}=\{0,4m^2\}`$, $`I_{15}=\{0,1m^2,4m^2\}`$, and $`I_5=\{0,1m^2\}`$.
To evaluate these diagrams we use the semianalytic method developed in Ref. . This approach is based on a possibility to restore analytical results in terms of harmonic sums from several first coefficients of the small momentum expansion . In Ref. the $`𝒪(1)`$ parts of the diagrams shown in Fig.1 were found <sup>3</sup><sup>3</sup>3The finite part of $`𝐈_\mathrm{𝟓}`$ is given in , $`𝐈_{\mathrm{𝟏𝟐𝟓}}`$ in and exact result for $`𝐈_{\mathrm{𝟏𝟐𝟓}}`$ in terms of hypergeometric function presented in .. Here we extend these results calculating their $`\epsilon `$-parts. Omitting all technical details that can be found in the above paper we present the results of our calculation<sup>4</sup><sup>4</sup>4 We are working in Minkowski space-time with dimension $`N=42\epsilon `$. For each loop we assume a common normalization factor $`(m^2e^\gamma )^\epsilon /\pi ^{\frac{N}{2}}`$, where $`\gamma `$ is Euler constant. . We find ($`z=p^2/m^2`$)
$`𝐈_{\mathrm{𝟏𝟐𝟓}}={\displaystyle \frac{1}{p^2}}{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}z^n[{\displaystyle \frac{\mathrm{ln}(z)}{n^2}}+{\displaystyle \frac{3}{n^3}}+\epsilon ({\displaystyle \frac{\mathrm{ln}^2(z)}{2n^2}}\mathrm{ln}(z)\{{\displaystyle \frac{2}{n^2}}+{\displaystyle \frac{S_1(n1)}{n^2}}\}`$
$`{\displaystyle \frac{1}{n^4}}+{\displaystyle \frac{6}{n^3}}{\displaystyle \frac{\zeta _2}{n^2}}+{\displaystyle \frac{11}{n^3}}S_1(n1){\displaystyle \frac{4}{n^3}}S_1(2n1))+𝒪(\epsilon ^2)],`$ (5)
$`𝐈_{\mathrm{𝟏𝟓}}={\displaystyle \frac{1}{p^2}}{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}z^n[{\displaystyle \frac{\mathrm{ln}(z)}{n^2}}{\displaystyle \frac{\zeta _2}{n}}+{\displaystyle \frac{2}{n^3}}+{\displaystyle \frac{3}{n}}V_2(n1)+{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{4}{n^3}}`$
$`+\epsilon ({\displaystyle \frac{\mathrm{ln}^2(z)}{2n^2}}\mathrm{ln}(z)\{{\displaystyle \frac{1}{n^3}}+{\displaystyle \frac{2}{n^2}}+{\displaystyle \frac{2}{n^2}}S_1(n1)\}+{\displaystyle \frac{\zeta _3}{n}}{\displaystyle \frac{2}{n}}\zeta _2+{\displaystyle \frac{6}{n^3}}S_1(n1)`$
$`{\displaystyle \frac{3}{n^2}}V_2(n1)+{\displaystyle \frac{3}{n}}V_3(n1)+{\displaystyle \frac{15}{n}}V_{2,1}(n1){\displaystyle \frac{6}{n}}\stackrel{~}{V}_{2,1}(n1)+{\displaystyle \frac{4}{n^3}}+{\displaystyle \frac{6}{n}}V_2(n1)`$
$`+{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}\{{\displaystyle \frac{8}{n^3}}+{\displaystyle \frac{20}{n^3}}S_1(n1){\displaystyle \frac{8}{n^3}}S_1(2n1)\})+𝒪(\epsilon ^2)],`$ (6)
$`𝐈_\mathrm{𝟓}={\displaystyle \frac{1}{p^2}}{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}(z)^n[{\displaystyle \frac{\mathrm{ln}^2(z)}{n}}+{\displaystyle \frac{2}{n^2}}\mathrm{ln}(z){\displaystyle \frac{2}{n}}\zeta _2+{\displaystyle \frac{4}{n}}K_2(n1){\displaystyle \frac{2}{n^3}}2{\displaystyle \frac{()^n}{n^3}}`$
$`+\epsilon ({\displaystyle \frac{\mathrm{ln}^3(z)}{n}}{\displaystyle \frac{\mathrm{ln}^2(z)}{n}}\{{\displaystyle \frac{2}{n}}+{\displaystyle \frac{3}{n^2}}+{\displaystyle \frac{S_1(n1)}{n}}\}+\mathrm{ln}(z)\{{\displaystyle \frac{4}{n^2}}+{\displaystyle \frac{6}{n^3}}+{\displaystyle \frac{2}{n^2}}S_1(n1)`$
$`+{\displaystyle \frac{2}{n}}S_2(n1){\displaystyle \frac{2}{n}}\zeta _2\}+{\displaystyle \frac{2}{n}}\zeta _3+\zeta _2\{{\displaystyle \frac{2}{n^2}}{\displaystyle \frac{4}{n}}{\displaystyle \frac{2}{n}}S_1(n1)\}{\displaystyle \frac{4}{n^3}}{\displaystyle \frac{6}{n^4}}{\displaystyle \frac{2}{n^3}}S_1(n1)`$
$`{\displaystyle \frac{2}{n^2}}S_2(n1){\displaystyle \frac{2}{n}}S_3(n1)+{\displaystyle \frac{8}{n}}K_2(n1)+{\displaystyle \frac{12}{n}}K_{2,1}(n1)+{\displaystyle \frac{4}{n}}K_2(n1)S_1(n1)`$
$`+{\displaystyle \frac{2}{n}}K_3(n1)()^n\{{\displaystyle \frac{4}{n^3}}+{\displaystyle \frac{2}{n^4}}+{\displaystyle \frac{8}{n^3}}S_1(n1)\})+𝒪(\epsilon ^2)],`$ (7)
where we use the following notations for the finite sums elaborated in
$$S_a(n)=\underset{j=1}{\overset{n}{}}\frac{1}{j^a},K_a(n)=\underset{j=1}{\overset{n}{}}\frac{()^j}{j^a},K_{a,b}(n)=\underset{j=1}{\overset{n}{}}\frac{(1)^j}{j^a}S_b(j1),$$
$$V_a(n)=\underset{j=1}{\overset{n}{}}\left(\genfrac{}{}{0pt}{}{2j}{j}\right)^1\frac{1}{j^a},V_{a,b}(n)=\underset{j=1}{\overset{n}{}}\left(\genfrac{}{}{0pt}{}{2j}{j}\right)^1\frac{1}{j^a}S_b(j1),\stackrel{~}{V}_{a,b}(n)=\underset{j=1}{\overset{n}{}}\left(\genfrac{}{}{0pt}{}{2j}{j}\right)^1\frac{1}{j^a}S_b(2j1).$$
Sums $`K_a`$, $`K_{a,b}`$ were also used in . Sums $`V_a`$ and $`\stackrel{~}{V}_a`$ were predicted by differential equation method ,
For all infinite series occurring in (5)-(7) a one-fold integral representation for arbitrary $`z`$ can be written . Thus these series can be continued analytically in the whole complex $`z`$-plane. Some of these integrals can be rewritten in terms of polylogarithms. For example we note the following representation
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{z^n}{n^a}}`$ $`=`$ $`{\displaystyle \underset{0}{\overset{1}{}}}{\displaystyle \frac{ds}{s}}S_{a2,1}\left(zs(1s)\right)`$ (8)
$`=`$ $`{\displaystyle \frac{1}{(a2)!}}{\displaystyle \underset{0}{\overset{2\mathrm{arcsin}\frac{\sqrt{z}}{2}}{}}}\left[\mathrm{ln}z2\mathrm{ln}\left(2\mathrm{sin}{\displaystyle \frac{\theta }{2}}\right)\right]^{a2}\theta 𝑑\theta `$
$`=`$ $`{\displaystyle \underset{j=0}{\overset{a2}{}}}{\displaystyle \frac{(2)^j}{(a2j)!j!}}(\mathrm{ln}z)^{a2j}\text{Ls}_{j+2}^{(1)}\left(2\mathrm{arcsin}{\displaystyle \frac{\sqrt{z}}{2}}\right),`$
where $`a>1`$ and $`S_{a,b}(z)`$ are generalized Nielsen polylogarithms . The last two lines have been obtained in . Substituting $`z=1`$ into (8) (i.e. on-shell condition for Feynman diagram) we get
$$\underset{n=1}{\overset{\mathrm{}}{}}\frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}\frac{1}{n^a}V_a(\mathrm{})=\frac{(2)^{a2}}{(a2)!}\text{Ls}_a^{(1)}\left(\frac{\pi }{3}\right).$$
For $`a=1,\mathrm{},5`$ we can write explicitly
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n^{\{1,2,3,4,5\}}}}`$ $`=`$ $`\{{\displaystyle \frac{1}{3}}{\displaystyle \frac{\pi }{\sqrt{3}}},{\displaystyle \frac{1}{3}}\zeta _2,{\displaystyle \frac{2}{3}}\pi \text{Ls}_2\left({\displaystyle \frac{\pi }{3}}\right){\displaystyle \frac{4}{3}}\zeta _3,{\displaystyle \frac{17}{36}}\zeta _4,`$
$`{\displaystyle \frac{4}{9}}\pi \text{Ls}_4\left({\displaystyle \frac{\pi }{3}}\right){\displaystyle \frac{19}{3}}\zeta _5{\displaystyle \frac{2}{3}}\zeta _2\zeta _3\}.`$
However, the analytical results for arbitrary $`z`$ for other types of sums are not yet available. For example we may write the integral representation
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{z^n}{n^a}}S_1(n1)={\displaystyle \underset{0}{\overset{1}{}}}{\displaystyle \frac{ds}{s}}S_{a2,2}\left(zs(1s)\right)=`$
$`{\displaystyle \frac{2}{(a2)!}}{\displaystyle \underset{0}{\overset{2\mathrm{arcsin}\frac{\sqrt{z}}{2}}{}}}\left[\mathrm{ln}z2\mathrm{ln}\left(2\mathrm{sin}{\displaystyle \frac{\theta }{2}}\right)\right]^{a2}\left[\text{Ls}_2\left(\pi +\theta \right)+\theta \mathrm{ln}\left(2\mathrm{sin}{\displaystyle \frac{\pi +\theta }{2}}\right)\right]𝑑\theta ,`$ (9)
but we do not know how to evaluate these integrals explicitly for $`a>2`$ even at $`z=1`$. Nevertheless for each particular $`a=1,\mathrm{},5`$ we are able to obtain an analytical answer for (9) at $`z=1`$ using the PSLQ algorithm . This proceeds as follows. Each sum can be evaluated numerically with arbitrary accuracy. PSLQ expresses the obtained numerical value in terms of given transcendental numbers. The only problem is to define the full set of ”basis” elements. Such a basis for diagrams having two massive particle cut was elaborated in . The Ansatz for the construction of basis up to arbitrary order have been suggested and explicitly evaluated up to weight 5 that corresponds to the second order in $`\epsilon `$-expansion of two-loop propagator type diagrams. The important role in construction of this basis belongs to Broadhurst’s observations that sixth root of unity plays an important role in the calculation of the diagrams.
We have investigated all sums of type (4) up to weight 5. Not all of them are expressible in terms of our basis elements or the transcendental numbers given in . But it turns out that linear combinations of sums occurring in Feynman diagrams evaluated in the present paper are connected with basis given by the ”sixth root of unity”. All our results were obtained empirically by carefully compiling and examining a huge data base of high precision (several hundreds decimals) numerical calculations. Some details of this calculations are given in Appendix A. The results of multiple binomial sum’s elaboration up to weight 4 are collected in Appendix B. The results for sum of weight 5 are relatively lengthy and therefore will not be published here.
We note that all V-type sums (occurring e.g. in (6)) are reduced to the multiple binomial sums due to the identity
$$\underset{n=1}{\overset{\mathrm{}}{}}\frac{1}{n^a}\underset{j=1}{\overset{n1}{}}f(j)=\underset{n=1}{\overset{\mathrm{}}{}}f(n)\left[\zeta _aS_a(n1)\frac{1}{n^a}\right].$$
Finally we mention that all sums occurring in Eq.(5) are expressible in terms of Euler-Zagier sums (2).
## 3 Conclusion
Collecting the results of Appendix B we obtain the following values for on-shell integrals
$`m^2𝐈_{\mathrm{𝟏𝟐𝟓}}|_{p^2=m^2}=\left\{4\zeta _3+2\pi \text{Ls}_2\left({\displaystyle \frac{\pi }{3}}\right)+i\pi {\displaystyle \frac{\zeta _2}{3}}\right\}(1+2\epsilon )`$
$`+\epsilon \left({\displaystyle \frac{16}{3}}\right[\text{Ls}_2\left({\displaystyle \frac{\pi }{3}}\right)]^27\pi \text{Ls}_3\left({\displaystyle \frac{2\pi }{3}}\right){\displaystyle \frac{488}{9}}\zeta _4i\pi \{{\displaystyle \frac{4}{9}}\pi \text{Ls}_2\left({\displaystyle \frac{\pi }{3}}\right){\displaystyle \frac{11}{9}}\zeta _3\})+𝒪(\epsilon ^2),`$ (10)
$`m^2𝐈_{\mathrm{𝟏𝟓}}|_{p^2=m^2}=\left\{3\zeta _3+2\pi \text{Ls}_2\left({\displaystyle \frac{\pi }{3}}\right)+i\pi \zeta _2\right\}(1+2\epsilon )`$
$`+\epsilon \left(6\right[\text{Ls}_2\left({\displaystyle \frac{\pi }{3}}\right)]^29\pi \text{Ls}_3\left({\displaystyle \frac{2\pi }{3}}\right){\displaystyle \frac{2567}{36}}\zeta _4+i\pi {\displaystyle \frac{5}{4}}\zeta _3)+𝒪(\epsilon ^2),`$ (11)
$`m^2𝐈_\mathrm{𝟓}|_{p^2=m^2}=\left\{3\zeta _3+i\pi \zeta _2\right\}(1+2\epsilon )`$
$`+\epsilon \left(6\zeta _2\mathrm{ln}^22\mathrm{ln}^4224\text{Li}_4\left({\displaystyle \frac{1}{2}}\right){\displaystyle \frac{57}{4}}\zeta _4+i\pi \left\{{\displaystyle \frac{19}{4}}\zeta _39\zeta _2\mathrm{ln}2\right\}\right)+𝒪(\epsilon ^2).`$ (12)
It is convenient to multiply Eqs.(10)–(12) by $`(12\epsilon )`$. Then we can write (10) and (11) in the form
$`m^2(12\epsilon )𝐈|_{p^2=m^2}=r_1\zeta _3+r_2\pi \text{Ls}_2\left({\displaystyle \frac{\pi }{3}}\right)+r_3i\pi \zeta _2`$
$`+\epsilon \left(r_4\right[\text{Ls}_2\left({\displaystyle \frac{\pi }{3}}\right)]^2+r_5\pi \text{Ls}_3\left({\displaystyle \frac{2\pi }{3}}\right)+r_6\zeta _4+i\pi \{r_7\pi \text{Ls}_2\left({\displaystyle \frac{\pi }{3}}\right)+r_8\zeta _3\})+𝒪(\epsilon ^2),`$ (13)
with some rational numbers $`r_j`$. Both $`𝐈_{\mathrm{𝟏𝟐𝟓}}`$ and $`𝐈_{\mathrm{𝟏𝟓}}`$ have a threshold at $`4m^2`$ plus possible thresholds at $`0m^2`$ and $`1m^2`$. The above results suggest that all such diagrams have the form (13) where coefficients $`r_j`$ depend on the distribution of the masses on lines while the basis (13) is defined by the topology alone.
If a diagram has no threshold at $`4m^2`$ then it is expressible in terms of Euler–Zagier sums. The three particle massive cuts lead to the appearance of a new structures like $`\text{Ls}_4^{(1)}\left(\frac{2\pi }{3}\right)`$ and some others.
Let us return to the results of Appendix B. One can write down a representation in terms of a hypergeometric sum
$$\underset{n=1}{\overset{\mathrm{}}{}}\frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}\frac{z^n}{n^a}=\frac{z}{2}{}_{a+1}{}^{}F_{a}^{}\left(\begin{array}{c}\{1\}_{a+1};\\ \frac{3}{2},\{2\}_{a1};\end{array}\frac{z}{4}\right).$$
It is easy to see that multiple sums with nested harmonic summations $`S_a`$ can be obtained from the generating function
$${}_{p+1}{}^{}F_{p}^{}\left(\begin{array}{c}\{1+a_i\}_{p+1};\\ \frac{3}{2}+b;\{2+c_i\}_{p1};\end{array}\frac{1}{4}\right)$$
(14)
by expanding (14) in powers of $`a_i,c_j`$ and $`b`$.
There are certain sums (see Appendix B) which cannot be expressed (polynomially) in terms of a basis connected with ”sixth root of unity” or the one given in . We don’t have an explanation for this phenomenon. However all linear combinations arising in the Taylor expansion of (14) are expressible in terms of our basis.
Acknowledgments We are grateful to A. Davydychev, D. Broadhurst, F. Jegerlehner and A. Kotikov for useful discussions and carefully reading the manuscript. M.K’s research has been supported by the DFG project FL241/4-1 and in part by RFBR $`\mathrm{\#}`$98-02-16923.
## Appendix A Multiple precision calculation of $`\text{Ls}_n\left(\theta \right)`$ and $`\text{Ls}_n^{(1)}\left(\theta \right)`$
For the purposes of PSLQ we need to evaluate functions $`\mathrm{Ls}_n(\theta )`$ and $`\mathrm{Ls}_n^{(1)}(\theta )`$ to very high accuracy (several hundreds of decimals). It is clear that the definitions (3) are not suitable for such numerical calculations. For example, one needs several hours of running MAPLE to calculate $`\text{Ls}_6\left(\pi /3\right)`$ with accuracy about 200 decimals. As an alternative we found the following series for these functions which allows us obtain the results with needed accuracy in a few seconds. We have ($`z=4\mathrm{sin}^2(\theta /2)`$)
$`\mathrm{Ls}_n(\theta )`$ $`=`$ $`(n1)!\sqrt{z}{\displaystyle \underset{r=1}{\overset{n}{}}}{\displaystyle \frac{(\mathrm{log}\sqrt{z})^{nr}}{(nr)!}}{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\left(\genfrac{}{}{0pt}{}{2k}{k}\right)}{(2k+1)^r}}\left({\displaystyle \frac{z}{16}}\right)^k,`$
$`\mathrm{Ls}_n^{(1)}(\theta )`$ $`=`$ $`(1)^n{\displaystyle \frac{(n2)!}{2^{n2}}}{\displaystyle \underset{r=0}{\overset{n2}{}}}{\displaystyle \frac{(\mathrm{log}z)^r}{r!}}{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{z^k}{\left(\genfrac{}{}{0pt}{}{2k}{k}\right)k^{nr}}}.`$
## Appendix B Multiple binomial sums
In this section we present the results of our searching of relationships between multiple binomial sum<sup>5</sup><sup>5</sup>5 All multiple binomial sums (4) can be rewritten in terms of function $`\mathrm{\Psi }(n)=\mathrm{d}/\mathrm{d}n\mathrm{log}\mathrm{\Gamma }(n)`$ and its derivatives by means of the following relation
$$\mathrm{\Psi }^{(k1)}(j)=()^k(k1)!\left[\zeta _kS_k(j1)\right],k>1,$$
where $`\mathrm{\Psi }^{(k)}(z)`$ is the $`k`$-th derivative of the $`\mathrm{\Psi }`$-function. In particular, for $`k=1`$ we have $`\mathrm{\Psi }(j)=S_1(j1)\gamma `$. (4) up to weight 4 and the set of transcendental numbers given in . All sums were obtained numerically by using multiprecision FORTRAN with accuracy of about 300 decimals and analytical results were obtained by PSLQ. The sums of weight 3 can be extracted from results of . The sums of weigt 4 and depths $`k<3`$ are investigated in .
Below we omit argument of harmonic sums implying that $`S_aS_a(n1)`$ and $`\overline{S}_aS_a(2n1)`$.
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n}}S_1={\displaystyle \frac{1}{3}}{\displaystyle \frac{\pi }{\sqrt{3}}}\mathrm{ln}3+{\displaystyle \frac{4}{3}}{\displaystyle \frac{\text{Ls}_2\left(\frac{\pi }{3}\right)}{\sqrt{3}}},`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n^2}}S_1={\displaystyle \frac{4}{9}}\pi \text{Ls}_2\left({\displaystyle \frac{\pi }{3}}\right)+{\displaystyle \frac{11}{9}}\zeta _3,`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n^3}}S_1=\pi \text{Ls}_3\left({\displaystyle \frac{2\pi }{3}}\right)+{\displaystyle \frac{4}{3}}[\text{Ls}_2\left({\displaystyle \frac{\pi }{3}}\right)]^2{\displaystyle \frac{269}{36}}\zeta _4,`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n}}\overline{S}_1={\displaystyle \frac{1}{3}}{\displaystyle \frac{\pi }{\sqrt{3}}}\mathrm{ln}3+{\displaystyle \frac{7}{3}}{\displaystyle \frac{\text{Ls}_2\left(\frac{\pi }{3}\right)}{\sqrt{3}}},`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n^2}}\overline{S}_1={\displaystyle \frac{7}{9}}\pi \text{Ls}_2\left({\displaystyle \frac{\pi }{3}}\right)+{\displaystyle \frac{23}{9}}\zeta _3,`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n^3}}\overline{S}_1=\pi \text{Ls}_3\left({\displaystyle \frac{2\pi }{3}}\right)+{\displaystyle \frac{7}{3}}[\text{Ls}_2\left({\displaystyle \frac{\pi }{3}}\right)]^2{\displaystyle \frac{143}{18}}\zeta _4,`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n}}S_1^2={\displaystyle \frac{1}{3}}{\displaystyle \frac{\pi }{\sqrt{3}}}\mathrm{ln}^23{\displaystyle \frac{8}{3}}{\displaystyle \frac{\text{Ls}_2\left(\frac{\pi }{3}\right)}{\sqrt{3}}}\mathrm{ln}3+{\displaystyle \frac{55}{27}}{\displaystyle \frac{\pi }{\sqrt{3}}}\zeta _2+4{\displaystyle \frac{\text{Ls}_3\left(\frac{2\pi }{3}\right)}{\sqrt{3}}},`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n^2}}S_1^2={\displaystyle \frac{4}{3}}\pi \text{Ls}_3\left({\displaystyle \frac{2\pi }{3}}\right){\displaystyle \frac{16}{9}}[\text{Ls}_2\left({\displaystyle \frac{\pi }{3}}\right)]^2+{\displaystyle \frac{1085}{108}}\zeta _4,`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n}}S_1^3={\displaystyle \frac{1}{3}}{\displaystyle \frac{\pi }{\sqrt{3}}}\mathrm{ln}^33+4{\displaystyle \frac{\text{Ls}_2\left(\frac{\pi }{3}\right)}{\sqrt{3}}}\mathrm{ln}^23{\displaystyle \frac{55}{9}}{\displaystyle \frac{\pi }{\sqrt{3}}}\zeta _2\mathrm{ln}3`$
$`12{\displaystyle \frac{\text{Ls}_3\left(\frac{2\pi }{3}\right)}{\sqrt{3}}}\mathrm{ln}3+{\displaystyle \frac{2}{3}}\zeta _2{\displaystyle \frac{\text{Ls}_2\left(\frac{\pi }{3}\right)}{\sqrt{3}}}{\displaystyle \frac{179}{27}}{\displaystyle \frac{\pi }{\sqrt{3}}}\zeta _3{\displaystyle \frac{92}{27}}{\displaystyle \frac{\text{Ls}_4\left(\frac{\pi }{3}\right)}{\sqrt{3}}}+8{\displaystyle \frac{\text{Ls}_4\left(\frac{2\pi }{3}\right)}{\sqrt{3}}},`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n}}S_1\overline{S}_1={\displaystyle \frac{1}{3}}{\displaystyle \frac{\pi }{\sqrt{3}}}\mathrm{ln}^23{\displaystyle \frac{11}{3}}{\displaystyle \frac{\text{Ls}_2\left(\frac{\pi }{3}\right)}{\sqrt{3}}}\mathrm{ln}3+{\displaystyle \frac{157}{54}}{\displaystyle \frac{\pi }{\sqrt{3}}}\zeta _2+{\displaystyle \frac{11}{2}}{\displaystyle \frac{\text{Ls}_3\left(\frac{2\pi }{3}\right)}{\sqrt{3}}},`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n^2}}S_1\overline{S}_1={\displaystyle \frac{11}{6}}\pi \text{Ls}_3\left({\displaystyle \frac{2\pi }{3}}\right){\displaystyle \frac{28}{9}}[\text{Ls}_2\left({\displaystyle \frac{\pi }{3}}\right)]^2+{\displaystyle \frac{3125}{216}}\zeta _4,`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n}}S_1^2\overline{S}_1={\displaystyle \frac{1}{3}}{\displaystyle \frac{\pi }{\sqrt{3}}}\mathrm{ln}^33+5{\displaystyle \frac{\text{Ls}_2\left(\frac{\pi }{3}\right)}{\sqrt{3}}}\mathrm{ln}^23{\displaystyle \frac{212}{27}}{\displaystyle \frac{\pi }{\sqrt{3}}}\zeta _2\mathrm{ln}3`$
$`15{\displaystyle \frac{\text{Ls}_3\left(\frac{2\pi }{3}\right)}{\sqrt{3}}}\mathrm{ln}3+{\displaystyle \frac{8}{9}}\zeta _2{\displaystyle \frac{\text{Ls}_2\left(\frac{\pi }{3}\right)}{\sqrt{3}}}{\displaystyle \frac{727}{81}}{\displaystyle \frac{\pi }{\sqrt{3}}}\zeta _3{\displaystyle \frac{298}{81}}{\displaystyle \frac{\text{Ls}_4\left(\frac{\pi }{3}\right)}{\sqrt{3}}}+10{\displaystyle \frac{\text{Ls}_4\left(\frac{2\pi }{3}\right)}{\sqrt{3}}},`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n}}S_2={\displaystyle \frac{1}{27}}{\displaystyle \frac{\pi }{\sqrt{3}}}\zeta _2,`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n^2}}S_2={\displaystyle \frac{5}{108}}\zeta _4,`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n}}S_3={\displaystyle \frac{2}{3}}\zeta _2{\displaystyle \frac{\text{Ls}_2\left(\frac{\pi }{3}\right)}{\sqrt{3}}}{\displaystyle \frac{16}{9}}{\displaystyle \frac{\pi }{\sqrt{3}}}\zeta _3+{\displaystyle \frac{4}{3}}{\displaystyle \frac{\text{Ls}_4\left(\frac{\pi }{3}\right)}{\sqrt{3}}},`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n}}S_1S_2={\displaystyle \frac{1}{27}}{\displaystyle \frac{\pi }{\sqrt{3}}}\zeta _2\mathrm{ln}3{\displaystyle \frac{2}{9}}\zeta _2{\displaystyle \frac{\text{Ls}_2\left(\frac{\pi }{3}\right)}{\sqrt{3}}}+{\displaystyle \frac{49}{81}}{\displaystyle \frac{\pi }{\sqrt{3}}}\zeta _3{\displaystyle \frac{20}{81}}{\displaystyle \frac{\text{Ls}_4\left(\frac{\pi }{3}\right)}{\sqrt{3}}},`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n}}S_2\overline{S}_1={\displaystyle \frac{1}{27}}{\displaystyle \frac{\pi }{\sqrt{3}}}\zeta _2\mathrm{ln}3+{\displaystyle \frac{4}{9}}\zeta _2{\displaystyle \frac{\text{Ls}_2\left(\frac{\pi }{3}\right)}{\sqrt{3}}}+{\displaystyle \frac{229}{81}}{\displaystyle \frac{\pi }{\sqrt{3}}}\zeta _3{\displaystyle \frac{146}{81}}{\displaystyle \frac{\text{Ls}_4\left(\frac{\pi }{3}\right)}{\sqrt{3}}},`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n}}\left(\overline{S}_1^2+\overline{S}_2\right)={\displaystyle \frac{1}{3}}{\displaystyle \frac{\pi }{\sqrt{3}}}\mathrm{ln}^23{\displaystyle \frac{14}{3}}{\displaystyle \frac{\text{Ls}_2\left(\frac{\pi }{3}\right)}{\sqrt{3}}}\mathrm{ln}3+{\displaystyle \frac{113}{27}}{\displaystyle \frac{\pi }{\sqrt{3}}}\zeta _2+7{\displaystyle \frac{\text{Ls}_3\left(\frac{2\pi }{3}\right)}{\sqrt{3}}},`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n^2}}\left(\overline{S}_1^2+\overline{S}_2\right)={\displaystyle \frac{7}{3}}\pi \text{Ls}_3\left({\displaystyle \frac{2\pi }{3}}\right){\displaystyle \frac{49}{9}}\left[\text{Ls}_2\left({\displaystyle \frac{\pi }{3}}\right)\right]^2+{\displaystyle \frac{4505}{216}}\zeta _4,`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n}}S_1\left(\overline{S}_1^2+\overline{S}_2\right)={\displaystyle \frac{1}{3}}{\displaystyle \frac{\pi }{\sqrt{3}}}\mathrm{ln}^33+6{\displaystyle \frac{\text{Ls}_2\left(\frac{\pi }{3}\right)}{\sqrt{3}}}\mathrm{ln}^2310{\displaystyle \frac{\pi }{\sqrt{3}}}\zeta _2\mathrm{ln}3{\displaystyle \frac{112}{9}}{\displaystyle \frac{\pi }{\sqrt{3}}}\zeta _3`$
$`18{\displaystyle \frac{\text{Ls}_3\left(\frac{2\pi }{3}\right)}{\sqrt{3}}}\mathrm{ln}3+{\displaystyle \frac{2}{3}}\zeta _2{\displaystyle \frac{\text{Ls}_2\left(\frac{\pi }{3}\right)}{\sqrt{3}}}{\displaystyle \frac{8}{3}}{\displaystyle \frac{\text{Ls}_4\left(\frac{\pi }{3}\right)}{\sqrt{3}}}+12{\displaystyle \frac{\text{Ls}_4\left(\frac{2\pi }{3}\right)}{\sqrt{3}}},`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\genfrac{}{}{0pt}{}{2n}{n}\right)}}{\displaystyle \frac{1}{n}}\left(\overline{S}_1^3+3\overline{S}_1\overline{S}_2+2\overline{S}_3\right)={\displaystyle \frac{1}{3}}{\displaystyle \frac{\pi }{\sqrt{3}}}\mathrm{ln}^33+7{\displaystyle \frac{\text{Ls}_2\left(\frac{\pi }{3}\right)}{\sqrt{3}}}\mathrm{ln}^23{\displaystyle \frac{394}{27}}{\displaystyle \frac{\pi }{\sqrt{3}}}\zeta _3`$
$`{\displaystyle \frac{113}{9}}{\displaystyle \frac{\pi }{\sqrt{3}}}\zeta _2\mathrm{ln}321{\displaystyle \frac{\text{Ls}_3\left(\frac{2\pi }{3}\right)}{\sqrt{3}}}\mathrm{ln}3+{\displaystyle \frac{2}{3}}\zeta _2{\displaystyle \frac{\text{Ls}_2\left(\frac{\pi }{3}\right)}{\sqrt{3}}}{\displaystyle \frac{28}{27}}{\displaystyle \frac{\text{Ls}_4\left(\frac{\pi }{3}\right)}{\sqrt{3}}}+14{\displaystyle \frac{\text{Ls}_4\left(\frac{2\pi }{3}\right)}{\sqrt{3}}}.`$
|
warning/0004/astro-ph0004068.html
|
ar5iv
|
text
|
# A deep X-ray observation of NGC 4258 and its surrounding field
## 1. Introduction
The nearby low-luminosity active galactic nucleus (AGN) in NGC 4258 (M 106) has become crucially important in our understanding of accreting supermassive black holes. Exquisite position and velocity measurements of the H<sub>2</sub>O megamasers in NGC 4258 reveal that the masing material resides in a very thin and slightly warped disk that is in almost perfect Keplerian motion about a central black hole with a mass of $`3.5\times 10^7\mathrm{M}_{}`$ (Nakai, Inoue & Miyoshi 1993; Greenhill et al. 1995a; Miyoshi et al. 1995). Detailed studies of the maser proper motions and centripetal accelerations confirm this interpretation (Greenhill et al. 1995b) and allow us to measure the distance to this source independently of the Hubble constant ($`d=7.2\mathrm{Mpc}`$; Herrnstein et al. 1999).
Sensitive X-ray observations provide a powerful means of probing both large scale and small scale structures within NGC 4258. X-ray emission was first detected in a short Einstein observatory high resolution imager (HRI) observation (Fabbiano et al. 1992). More sensitive soft X-ray data collected by the ROSAT position sensitive proportional counters (PSPC) and HRI found an extended halo of hot ($`4\times 10^6\mathrm{K}`$) gas around NGC 4258 (Pietsch et al. 1994; Vogler & Pietch 1999) as well as X-ray emission associated with the well known helically twisted jets (the anomalous arms; Pietsch et al. 1994; Cecil, Wilson & De Pree 1995; Vogler & Pietch 1999). None of these soft X-ray observations penetrated the column of absorbing gas that obscures the AGN itself. This had to await ASCA observations (Makishima et al. 1994; hereafter M94). M94 found that the central X-ray source was well described by a power-law with photon index $`\mathrm{\Gamma }1.8`$, and was absorbed by a column density of $`N_\mathrm{H}1.5\times 10^{23}\mathrm{cm}^2`$. The soft X-ray spectrum was found to be complex with components arising from thermal plasma emission and, possibly, contributions from the underlying X-ray binary population in NGC 4258. A marginal detection of an iron line was claimed with an equivalent width of $`250\pm 100\mathrm{eV}`$.
Of great importance is the opportunity that NGC 4258 gives us to study accretion physics when the mass accretion rate is potentially very low. The significance of this issue is highlighted when it is realized that most of the (quiescent) supermassive black holes in the universe are inferred to accrete matter at rates which are comparable to, or less than, that in NGC 4258. It was realized by several authors that when the accretion rate is low (relative to the Eddington rate), an accretion disk may switch into a hot, radiatively-inefficient mode (Ichimaru 1977; Rees 1982; Narayan & Yi 1994; Narayan & Yi 1995). In essence, the plasma becomes so tenuous that the timescale for energy transfer from the protons to the electrons (via Coulomb interactions) becomes longer than the inflow timescale. The energy remains as thermal energy in the protons (which are very poor radiators) and gets advected through the event horizon of the black hole. These are the so-called Advection Dominated Accretion Flows (ADAFs). ADAFs are to be contrasted with ‘standard’ radiatively-efficient accretion disks in which the disk remains cool and geometrically thin all of the way down to the black hole (Shakura & Sunyaev 1973; Novikov & Thorne 1974).
X-ray observations of broad iron $`K\alpha `$ lines have shown that higher luminosity systems do indeed accrete in the radiatively-efficient mode (Tanaka et al. 1995; Fabian et al. 1995). However, the basic nature of the accretion disk when the mass accretion rate is low is still far from clear. The existence of the ADAF solution is at the mercy of poorly known physics such as the strength of the electron-ion coupling and the fraction of the viscous energy that is deposited into the electrons (Quataert & Gruzinov 1999). Also, Blandford & Begelman (1999) have suggested that the ADAF solutions discussed by Narayan & Yi (1994, 1995) are physically inconsistent and necessarily drive powerful outflows (producing the so-called Adiabatic Inflow-Outflow Solutions; ADIOS). Even if ADAF-type solutions exist, it remains an open question as to whether real disks can make a transition to this mode when the outer regions of the disk are cold and geometrically-thin (as they clearly are in systems such as NGC 4258). NGC 4258 provides a laboratory in which we can examine all of these issues.
This dichotomy in possible accretion disk physics has produced two models for the central regions of NGC 4258. By modeling the observed water maser emission in this system, Neufeld & Maloney (1995) conclude that the accretion disk in NGC 4258 has a high efficiency ($`10\%`$), and a low accretion rate ($`\dot{M}/\alpha 10^4\mathrm{M}_{}\mathrm{yr}^1`$, where $`\alpha `$ is the standard viscosity parameter of accretion disk theory). Furthermore, the observed maser emission in part traces out a warp in the disk. Modeling this warp as being driven by radiation pressure from the central X-ray source (Pringle 1996) also implies a radiative efficiency $`10\%`$ (Maloney, Begelman, & Pringle 1996). On the other hand, Lasota et al. (1996) use ADAF models of the 2-10 keV X-ray power law slope and continuum radio data to argue that the system has low efficiency ($`0.1\%`$) and high accretion rate ($`\dot{M}/\alpha 10^2\mathrm{M}_{}\mathrm{yr}^1`$, $`\alpha 0.3`$). While the original model of Lasota et al. (1996) postulated a large ADAF region ($`r10^5GM/c^2`$), recent radio data constrain the ADAF region to be smaller than $`r100GM/c^2`$ (Herrnstein et al. 1998).
In this paper, we report a deep ($`200`$ ksec of good exposure time) X-ray observation of NGC 4258 with ASCA (Ohasi et al. 1996; Makishima et al. 1996; Yamashita 1997). We also obtained simultaneous hard X-ray data with the RXTE satellite but, due to a gain change prior to our observation, there are currently no robust response matrices or background models for these particular RXTE data. A presentation of these RXTE data must await these developments in calibration. In section 2, we describe the data reduction before discussing the X-ray spectrum of NGC 4258 in Section 3. Section 4 focuses on the properties of the observed iron line and the implications of this line for the nature of the accretion disk. Section 5 presents a brief discussion of the other interesting objects in the field of view of the ASCA Gas Imaging Spectrometer (GIS). Our results are summarized in Section 6.
## 2. Data reduction and the X-ray image
The NGC 4258 field was observed by ASCA on 1999 May 15–20. The SIS data were cleaned in order to remove the effects of hot and flickering pixels and subjected to the following data-selection criteria : i) the satellite should not be in the South Atlantic Anomaly (SAA), ii) the object should be at least 7 degrees above the Earth’s limb, iii) the object should be at least 25 degrees above the day-time Earth limb and iv) the local geomagnetic cut-off rigidity (COR) should be greater than 6 GeV/c. We also applied a standard grade selection on SIS events in order to further minimize particle background. The GIS data were cleaned to remove the particle background and subjected to the following data-selection criteria : i) the satellite should not be in the SAA, ii) the object should be at least 7 degrees above the Earth’s limb and iii) the COR should be greater than 7 GeV/c. SIS and GIS data that satisfy these criteria shall be referred to as ‘good’ data.
After the above data selection, there are 170 ksec of good data per SIS detector and 185 ksec of good data per GIS detector. Images were then extracted from these good data for each of the four instruments (two SIS and two GIS).
Fig. 1 shows the GIS2 image for this observation. The nucleus of NGC 4258 is the brightest X-ray source in this field. Two other sources are also detected: (a) one point-like source $`7`$ arcmins to the west of NGC 4258, and (b) another slightly extended source $`17`$ arcmins to the north-west of NGC 4258. Both of these sources were clearly detected and studied with ROSAT by Pietsch et al. (1994). Source (a) can be readily identified as the $`z0.4`$ quasar Q1218+472 (Burbidge 1995; Burbidge & Burbidge 1997). The identification of source (b) is less secure, but Pietsch et al. argue that it is a background ($`z0.2`$) cluster of galaxies on the basis of a possible galaxy over-density on a deep optical plate.
We have extracted spectra and lightcurves for all three sources in our field. Unless otherwise stated, source counts were extracted from a circular region centered on the source with radii of 3 arcmin and 4 arcmin for the SIS and GIS respectively. Background spectra were obtained from blank regions of the same field (using the same chip in the case of the SIS). No temporal variability was observed in any of these sources. In order to facilitate $`\chi ^2`$ spectral fitting, all spectra were rebinned so as to contain at least 20 photons per spectral bin. In order to avoid poorly calibrated regions of the spectrum, the energy ranges considered were 1.0–10 keV for the SIS detectors, and 0.8–10 keV for the GIS detectors. Note that we use a lower-energy cutoff for the SIS that is considerably higher than the ‘standard’ 0.6 keV cutoff in order to avoid the effects of “residual dark current distribution”, or RDD, which is known to plague recent ASCA observations.
## 3. A detailed X-ray study of NGC 4258
We now discuss the X-ray properties of NGC 4258. The two serendipitous sources, Q1218+472 and the putative galaxy cluster, will be addressed in Section 5.
### 3.1. Fitting the soft X-ray spectrum
#### 3.1.1 The inclusion of ROSAT data
M94 showed that the soft X-ray spectrum of NGC 4258 is complex, with components arising from thermal plasma emission and, possibly, the integrated emission of the X-ray binary population. Since we do not consider ASCA data below 0.8 keV, spectrum models of this region will be poorly constrained with many different models able to explain the soft spectrum. To break these degeneracies, we included an archival ROSAT PSPC spectra in our analysis. We chose to use the longest single ROSAT PSPC integration of NGC 4258. This observation, which was performed on 1990-Jun-1, has about 25 ksec of good exposure time. This dataset was obtained from the heasarc archive at the NASA Goddard Space Flight Center and reduced using the ftools routine xselect v1.4b. ROSAT-PSPC data were used in the range 0.2–2 keV. Good agreement was obtained between the ASCA SIS/GIS and the ROSAT PSPC in the overlap band between 1–2 keV, thereby alleviating concerns that our analysis will be severely affected by poor ROSAT-ASCA cross-calibration.
#### 3.1.2 Characterizing thermal plasma emission
Table 1 details our spectral analysis of NGC 4258 (excluding our detailed analysis of the iron line which will be addressed in Section 4). As is evident from our spectrum and the previous work of M94, the spectrum has both hard and soft components. The simplest model that we attempted to fit consists of an absorbed power-law with an additional bremsstrahlung component, all absorbed by the Galactic column density of $`N_{\mathrm{Gal}}=1.2\times 10^{20}\mathrm{cm}^2`$. This was a dreadful fit to the spectrum, giving $`\chi ^2/\mathrm{dof}=3169/1507`$. Large residuals exist with the model under-predicting the data in the 0.7–1.5 keV range. Since this is the energy range in which powerful line emission can occur from a thermal plasma with $`kT1\mathrm{keV}`$, we next replaced the bremsstrahlung components with the thermal plasma model mekal (Mewe, Gronenschild & van den Oord 1985; Arnaud & Rothenflug 1985; Mewe, Lemen & van den Oord 1988; Kaastra 1992). Initially, we consider a thermal plasma model in which all metals are assumed to have the same fractional elemental abundance relative to the cosmic abundances of Anders and Grevesse (1989). The inclusion of the thermal plasma component (model-A in Table 1) leads to a dramatic improvement in the goodness of fit with $`\chi ^2/\mathrm{dof}=1925/1507`$. However, several line-like residuals in the X-ray spectrum (including one at iron K$`\alpha `$ energies), as well as a general curvature of the spectrum, prevent this model from being an adequate fit to the data. We now discuss extensions of this basic spectral model which can adequately explain the observed spectrum.
#### 3.1.3 An additional Bremsstrahlung-like component
The curvature in the spectrum requires us to consider an additional continuum component. M94 include an additional bremsstrahlung component in order to model the diffuse and off-nuclear emission in this object. Including a bremsstrahlung component (model B) accounts for this curvature and leads to a dramatic improvement in the goodness of fit (compare models A and B; $`\mathrm{\Delta }\chi ^2=298`$ for 2 additional degrees of freedom). This additional component is required even if we allow for abundance variations in the thermal plasma component (compare models B and C; also see below). For completeness, we also investigated the possibility that this additional continuum component is a power-law that does not suffer any intrinsic absorption (model D). Scattering of the AGN power-law emission around the absorbing material, or non-thermal spatially-extended emission from the galaxy or jet would be possible sources of such a component. This provides a marginally worse fit than the bremsstrahlung-based model, and implies large scattering fractions ($`f_{\mathrm{scat}}=0.16`$) or powerful distributed non-thermal emission ($`L_{\mathrm{X},\mathrm{dist}}10^{40}\mathrm{erg}\mathrm{s}^1`$ in the 0.5–10 keV range). Hence, we think this power-law alternative to be unlikely.
#### 3.1.4 Constraints on the plasma abundance
Whether this additional continuum is modeled as a bremsstrahlung or a power-law component has little influence on the best-fit parameters for the direct AGN component (i.e. it does not significantly affect the photon index or normalization of the AGN power-law emission, nor the inferred absorbing column through to the power-law source). However, it does effect the best-fit abundances of the soft thermal plasma component (compare models C and B in Table 1). In this section, we attempt to constrain the abundances of the thermal plasma under the assumption that this additional component has a bremsstrahlung form. We also make the assumption that the thermal plasma emission can be characterized by a single temperature. We make these assumptions here in order to be able to make progress with these data. Both of these assumptions may be invalid. We must await future high-resolution, high signal-to-noise data in order to test and relax these assumptions through the use of direct emission line diagnostics.
We investigated the plasma abundances in a two step process. Firstly, the metals were split into two classes with iron and nickel in one class, and all of the lighter metals in the other class. The relative abundances were fixed within each class, but the relative abundance of each class was allowed to vary independently. This had no effect on the goodness of fit, with the best-fit relative abundances of each class being very similar (compare models B and E). Since soft X-ray line-like residuals still persist in these fits, the relative abundances of all of the light elements were then allowed to vary independently. This leads to a further improvement in the goodness of fit (compare models E and F; $`\mathrm{\Delta }\chi ^2=70`$ for 10 additional degrees of freedom). After fitting model F (which includes an additional bremsstrahlung component; see below) it can be seen from Table 1 that the abundances of C, Ne, Na, Al, and Ar are poorly constrained with only weak upper limits possible. This is due to the lack of strong emission lines from these elements in the well-calibrated region of the ASCA-SIS/GIS. On the other hand, the abundances of Mg and Fe are well constrained (with $`Z_{\mathrm{Mg}}=0.36`$ and $`Z_{\mathrm{Fe}}=0.23`$) due to the detection of fairly strong emission line complexes associated with these elements. It is worth noting the apparent extreme overabundance of Calcium ($`Z_{\mathrm{Ca}}=5\pm 2`$). This result must be viewed with suspicion since the dominant Calcium emission lines emerge just below our usable ASCA band (which has a lower-energy cutoff at 0.8 keV for the GIS and 1.0 keV for the SIS). Hence, any slight ASCA/ROSAT cross-calibration problem might be manifested as an extreme Calcium abundance.
#### 3.1.5 Fluxes and luminosities
Using model F from Table 1, and assuming a distance of 7.2 Mpc, we can compute the absorption-corrected luminosities in each spectral component. The result is shown in Table 2. It can be seen that the power-law component dominates the energetics of the X-ray band by almost an order of magnitude. On energetic grounds, the thermal plasma emission can be powered via the absorption and re-emission of 10–25% of the hard power-law component (depending upon how low in energy the power-law component extends). Our data cannot probe the nature and origin of this thermal emission beyond pointing out the basic energetics. However, future observations with Chandra and XMM will be able to study the spatial distribution of the various spectral components that we have noted. One will then be able to address whether the thermal emission is located in the immediate vicinity of the AGN (as, for example, if it originated from an accretion disk wind) or distributed on scales of 100 pc or greater (as in the case of a galactic superwind).
### 3.2. Characterizing the iron K$`\alpha `$ emission line
Residuals due to the iron line are clearly visible when models A–E are fit to these data (see Fig. 2b). We choose to use model D of Table 1 as a base model for our iron line investigation. We do not use model E (i.e. the best fitting model from Table 1) because the excessive number of free parameters makes the iron line error analysis unnecessarily difficult. We have verified that our iron line results are not affected in any significant manner by the choice of using model D as a base model rather than model E. We initially fit this line by adding a Gaussian emission feature (model F of Table 1). The improvement in the goodness of fit is significant at more than the 90 per cent level ($`\mathrm{\Delta }\chi ^2=18`$ for 3 additional degrees of freedom). The full-width half maximum (FWHM) of the line is constrained to be less than $`22000\mathrm{km}\mathrm{s}^1`$, and is consistent with being zero (i.e. the line is unresolved). The inferred centroid energy of $`E=6.45_{0.07}^{+0.10}\mathrm{keV}`$ is consistent with the K$`\alpha `$ line energy from iron with an ionization state of less than Fexvii. The equivalent width of the line is $`W_{\mathrm{Fe}}=107_{37}^{+42}\mathrm{eV}`$.
We discuss the origins of this emission line and the possible implications for the nature of the accretion disk in Section 4.
### 3.3. The long term variability of NGC 4258
ASCA observed NGC 4258 on 4 previous occasions, thereby allowing us to examine the variability of this AGN on timescales of several years. We retrieved all of the available ASCA data on NGC 4258 from the heasarc public database located at the NASA Goddard Space Flight Center. These data were reduced in the same manner as described in Section 2, and then fitted with spectral model B from Table 1. Since these archival data possess significantly lower signal to noise, the temperatures of the thermal plasma and bremsstrahlung components were fixed to the best fit values from Table 1. The normalization of the bremsstrahlung component was also fixed. Thus, this investigation is probing changes in the AGN power-law and absorbing column density. Note that only GIS data were included for the 5-May-93 observation, since NGC 4258 falls very close to the SIS chip boundaries in this observation.
Table 3 summarized the results of this study. The photon index of the power-law emission is consistent with being constant over the 6 years covered by these data. However, there are indications of variations in both the absorbing column density (which seems to decrease by $`30\%`$ between the 1993 and 1996 observations) and the 5–10 keV flux (which almost doubles from the 1993 to the 1996 observations, and then drops back to the 1993 level by the time of the 1999 observation). The flux in the 5–10 keV range is dominated by the AGN power-law component and is little affected by the absorbing column.
## 4. Implications of the iron line in NGC 4258
### 4.1. Pure accretion disk models for the iron line
The interest in the iron line lies in its ability to diagnose the nature of the accretion disk. Suppose that the only significant source of iron K$`\alpha `$ line emission in NGC 4258 is the AGN accretion disk. We can then model the observed iron line as being from the surface of the accretion disk around a non-rotating (Schwarzschild) black hole by using the diskline model within the xspec software package. We will make the assumption that the disk is flat inside of the masing radii and so set the inclination of the inner disk to be $`i=85^{}`$. The outer radius of the iron line emitting region is set to $`r_{\mathrm{out}}=10^5GM/c^2`$ (the radius of the maser disk; note that the iron line fits are very insensitive to the actual value of $`r_{\mathrm{out}}`$ provided it is sufficiently large). The line energy was fixed at the value appropriate for K$`\alpha `$ emission from weakly ionized iron, $`E=6.40\mathrm{keV}`$. Free parameters in the fit are the inner radius of the line emitting region $`r_{\mathrm{br}}`$, the index describing line emission as a function of radius $`\beta `$ (where the surface emissivity $`ϵr^\beta `$), and the normalization of the line. Figure 3a reports the confidence contours that result when this accretion disk model is fit to the iron line. It can be seen that the narrowness of the line requires either flat emissivity as a function of radius ($`\beta >2`$) or an inner edge to the line emitting region at greater than a hundred gravitational radii.
If the X-ray emission traces the viscous dissipation in the disk, and the disk is geometrically-thin and radiatively-efficient so that it can be described by a Novikov & Thorne (1973) model, the emissivity index should tend to $`\beta =3`$ outside of the inner 20 gravitational radii or so. In this case, and given the assumption that the observed iron line is from the accretion disk, we are led to the conclusion (at the 90% confidence level) that the line emitting region has an inner edge at $`100GM/c^2`$. Such an inner edge may correspond to the point at which the disk surface becomes ionized, or where the disk undergoes a transition to a hot (possibly advection dominated) state. On the other hand, if the X-ray source is in the form of a geometrically-thick corona with size $`D`$, the emissivity index will be fairly flat for $`r<D`$, and will tend to $`\beta =3`$ for $`r>>D`$. In this case (and again, with the assumption that the observed iron line is from the accretion disk), we must conclude that the corona is large $`D100GM/c^2`$.
### 4.2. Hybrid disk/non-disk iron line models
An alternative that we must explore is one in which a some fraction of the observed narrow iron line is produced by distant material not directly related to the accretion disk. In this case, any broad emission line from the accretion disk would be blended with this narrow line and, possibly, buried in the noisy continuum spectrum. To investigate this possibility, we suppose that a narrow iron line from some non-disk origin contributes to the observed spectrum with an equivalent width of $`W_{K\alpha ,nar}`$. Fixing the line emissivity profile of the disk to the canonical $`\beta =3`$ case, Fig. 3b shows the confidence contours on the $`W_{K\alpha ,nar}r_{\mathrm{br}}`$ plane. It can be seen that the line emission needs an inner edge or break at $`r_{\mathrm{br}}50GM/c^2`$ unless the additional narrow line source contributes at the level $`W_{K\alpha ,nar}40\mathrm{eV}`$.
We cannot rule out, in any rigorous sense, the possibility that most (or all) of the observed iron line originates from matter that is not associated with the accretion disk. However, simple arguments lead us to disfavor such a scenario. Consider iron line emission in a geometrically-thick torus surrounding the accretion disk of NGC 4258. An upper limit to the column density of this structure along our line of sight to the AGN is given by the observed column density of $`N_\mathrm{H}10^{23}\mathrm{cm}^2`$. If we suppose that this torus is in the same plane as the accretion disk (so that we are also viewing it edge-on), it is plausible to assume that we are looking through the optically-thickest part of the torus. By considering the case in which the torus completely surrounds the AGN with uniform column density along all radii, an upper limit to the equivalent width of the iron line is given by:
$$W_{\mathrm{Fe},\mathrm{max}}=E_{\mathrm{line}}^2Y_{\mathrm{Fe}}N_\mathrm{H}Z_{\mathrm{abs},\mathrm{Fe}}_0^{\mathrm{}}\frac{\sigma (E)}{E^2}𝑑E,$$
(1)
where $`E_{\mathrm{line}}=6.4\mathrm{keV}`$ is the energy of the emission line, $`Y_{\mathrm{Fe}}=0.33`$ is the fluorescent yield of the transition, $`Z_{\mathrm{abs},\mathrm{Fe}}4\times 10^5`$ is the density of iron relative to hydrogen, and $`\sigma (E)`$ is the energy dependent photoionization cross section for ionization from the $`1s`$ shell. Here, we have approximated the photon index of the ionizing AGN power-law to $`\mathrm{\Gamma }=2`$ and assumed that the central X-ray source is isotropic. Using the photoionization cross-sections for neutral iron of Verner & Yakovlev (1994), this yields
$$W_{\mathrm{Fe},\mathrm{max}}65\mathrm{eV}.$$
(2)
However, there are several reasons why this upper limit would almost certainly not be achieved. Firstly, modeling of the accretion disk warp strongly suggests that our line of sight intercepts the disk and that the bulk of the column density which obscures the AGN originates in the disk (Herrnstein, priv. communication). Therefore, we might expect significantly smaller column densities along lines of sight that have smaller inclinations angles relative to the accretion disk. Secondly, the accretion disk may well occult half of this fluorescing cloud, thereby reducing this prediction further. Thus, the true iron line from surrounding non-disk material may well be reduced from our naive prediction by a factor of several.
A high column density torus that is misaligned with the accretion disk so as to leave the X-ray unobscured is still a viable source for the observed narrow iron line. Such a torus must exist on size scales significantly larger than the masing accretion disk, or else it would hydrodynamically disturb the disk and lead to strong deviations from the observed Keplerian rotation.
### 4.3. Equivalent width limits on a “Seyfert-like” broad iron line
Many higher luminosity Seyfert galaxies display iron lines which are so broad that the line emitting region is thought to extend down to near the radius of marginal stability ($`6GM/c^2`$ for a non-rotating black hole). The typical emissivity index $`\beta `$ lies between $`2`$ and $`3`$ (Nandra et al. 1997). As shown above, if $`W_{K\alpha ,nar}40\mathrm{eV}`$ then no such relativistically broad component is required by our fits to NGC 4258. However, even when all of the observed line is assumed to arise from some other structure, we can still obtain an upper limit to the equivalent width of any such “Seyfert-like” component by fixing $`r_{\mathrm{br}}=6GM/c^2`$ in the spectral fitting, and including an explicit narrow Gaussian component to fit the observed line. The resulting upper limit on the equivalent width of the relativistic iron line, as a function of the assumed line emissivity index, is shown in Fig. 4 (thick solid line).
If NGC 4258 possesses an iron line emissivity inner accretion disk similar to higher luminosity Seyfert galaxies, one might naively expect that limb-darkening due to absorption in the outer layers of the accretion disk will reduce the equivalent width of any such broad iron line to very small values (e.g. George & Fabian 1991). From the work of Ghisellini, Haardt & Matt (1994), the limb-darkening in the plane-parallel case is well described by the expression
$$W_{K\alpha }(\theta )=\frac{W_{K\alpha }(\theta =0)}{\mathrm{ln}2}\mathrm{cos}\theta \mathrm{ln}\left(1+\frac{1}{\mathrm{cos}\theta }\right)$$
(3)
where $`\theta `$ is the inclination angle. Nandra et al. (1997) shows that most Seyfert 1 galaxies have a broad iron line with $`W_{K\alpha }`$ in the range 200–400 eV. Taking these to be representative of the face-on values, an inclination of $`\theta =85^{}`$ and this limb-darkening law reduces the expected equivalent width to 66–132 eV. This is below our detection threshold.
Two important effects modify this estimate. Firstly, relativistic light bending means that significant portions of the innermost regions of the disk are viewed significantly more face-on than otherwise might be thought. The dotted and dashed lines on Fig. 4 show the effects of light bending on the predicted broad line equivalent width (assuming a face-on equivalent width, $`W_{K\alpha }(\theta =0)`$, of 200 eV and 400 eV respectively). These lines have been computed using the relativistic code presented in Reynolds et al. (1999) combined with the above limb-darkening law. Figure 4 shows that the broad line would not be detectable unless $`W_{K\alpha }(\theta =0)>300\mathrm{eV}`$ and $`\beta >3`$.
Secondly, the above limb-darkening law assumes that the $`\tau =1`$ surface of the accretion disk is strictly planar. It is very unlikely that this will be the case. MHD turbulence and violent plasma processes within the disk corona will inevitably corrugate this surface thereby reducing the effects of limb-darkening for highly inclined sources. In the extreme case, most of the matter in the outer layers of the accretion disk may be concentrated into dense filaments or clumps (e.g. see the MHD simulations for the ‘Channel solutions’ case presented by Miller & Hawley 2000), thereby removing orientation and limb-darkening effects almost entirely. The dot-dashed and the dot-dot-dot-dashed lines on Fig. 4 show the predicted equivalent width in the case of no limb-darkening, assuming $`W_{K\alpha }(\theta =0)=200\mathrm{eV}`$ and $`400\mathrm{eV}`$, respectively<sup>2</sup><sup>2</sup>2Note that, following the convention used in xspec, we have defined the equivalent width with respect to the continuum level at the energy where the line peaks in photon flux. For centrally concentrated disk illumination (i.e. very negative $`\beta `$), the iron line peaks at $`8\mathrm{keV}`$. Since the continuum flux is a strongly declining function of energy, the equivalent width of such a line can exceed the face-on value.. In this limiting case, we would expect to be sensitive to such a broad line for all Seyfert-like values of $`\beta `$ and $`W_{K\alpha }(\theta =0)`$.
To summarize, these data just begin to constrain the interesting regions of parameter space for any relativistic iron line component in NGC 4258. There are two possible reasons for our non-detection of a broad iron line. Firstly, such a line might be genuinely absent thereby setting NGC 4258 aside from its higher luminosity counterparts. Secondly, if NGC 4258 possesses a relativistically broad iron line with $`W_{K\alpha }(\theta =0)300\mathrm{eV}`$ and limb-darkening is important, then we would not expect to detect this line with these data. However, we can rule out the presence of a relativistically broad iron line with $`W_{K\alpha }(\theta =0)300\mathrm{eV}`$ and $`\beta >3`$. Neither do these data allow a Seyfert-like line in which limb-darkening is unimportant.
## 5. The serendipitous sources
### 5.1. Q1218+472
The point source 7 arcmins to the west of NGC 4258 is readily identified with the $`z=0.40`$ quasar Q1218+472. By choosing a slightly smaller extraction radius than normal (2.5 arcmins), we can isolate the GIS counts from this source with relatively little contamination from NGC 4258. The resulting GIS background subtracted count rate is $`4\times 10^3`$ cps per detector. No temporal variability was observed in the resulting light curve, although the low count rate results in weak limits on possible variability. This source is not within the field of view of the SIS.
The two GIS spectra were fitted with a model consisting of a power-law subjected to both Galactic absorption (with a column of $`N_{\mathrm{Gal}}=1.2\times 10^{20}\mathrm{cm}^2`$) and intrinsic absorption with the redshift of the quasar. The resulting best fit parameters are $`\mathrm{\Gamma }=1.40_{0.14}^{+0.24}`$ and $`N_\mathrm{H}<5\times 10^{21}\mathrm{cm}^2`$ (see Fig. 5 for the confidence contours that result from this fit). The (observer frame) 2–10 keV flux is $`F_{210}=4.0\times 10^{13}\mathrm{erg}\mathrm{cm}^2\mathrm{s}^1`$. Assuming a redshift of $`z=0.40`$, a Hubble parameter of $`H_0=65\mathrm{km}\mathrm{s}^1\mathrm{Mpc}^1`$, and an acceleration parameter of $`q_0=0.5`$, the rest-frame 2–10 keV luminosity of this source is $`L_{210}=1.6\times 10^{44}\mathrm{erg}\mathrm{s}^1`$. In computing the flux and luminosity, we have applied a correction factor of 1.5 to account for the smaller than usual extraction radius. This correction factor was estimated by extracting counts from other ASCA observations of bright Seyfert galaxies with various extraction radii.
While the flux and luminosity estimate are robust, the spectral results should be treated with caution since the GIS count rates for this source are below the threshold at which the ASCA Guest Observer Facility recommends spectroscopy. Combining our X-ray spectrum with the optical spectrum of Burbidge (1995), the spectral index between the (rest-frame) 2500Å and 2 keV emissions is $`\alpha _{\mathrm{ox}}1.0`$. Thus, Q1218+472 is relatively X-ray bright given its optical flux.
### 5.2. The north-west source: a putative galaxy cluster
The northwest source aligns well with the extended ROSAT-PSPC source found by Pietsch et al. (1994; hereafter P94). P94 note a concentration of galaxies on a deep optical plate. By noting that the brightest galaxy is approximately 19 magnitude in the visual band, they estimate the redshift of the cluster to be $`z0.2`$.
Using standard extraction radii, we have extracted the GIS spectra and light curves for this source (note that this source, too, fell outside the SIS field of view). The GIS count rate was $`7\times 10^3`$ cps per detector. Given the possible identification of this source with a cluster of galaxies, we fit the spectrum with a thermal plasma model (MEKAL<sub>1</sub> of Table 1) modified by Galactic absorption. We also allow the redshift of the cluster to be a free parameter. The best fit plasma temperature, relative abundance, and redshift are $`kT=5.1_{0.9}^{+1.3}\mathrm{keV}`$, $`Z=0.49_{0.32}^{+0.41}`$ and $`z=0.28\pm 0.05`$ respectively. Figure 6 shows the two GIS spectra along with this best fit model. It can be seen that the model is a good fit to these data (with $`\chi ^2`$/dof=202/217). This X-ray determination of the cluster redshift relies on fitting the cluster iron emission line to a bump in the GIS spectrum at $`5\mathrm{keV}`$. Since instrumental GIS features are possible at these energies, this must be taken as tentative.
The observer-frame 2–10 keV flux of this source is $`F_{210}=6\times 10^{13}\mathrm{erg}\mathrm{cm}^2\mathrm{s}^1`$. Assuming a cluster redshift of $`z=0.28`$, a Hubble parameter of $`H_0=65\mathrm{km}\mathrm{s}^1\mathrm{Mpc}^1`$, and an acceleration parameter of $`q_0=0.5`$, the rest-frame 2–10 keV luminosity of this source is $`L_{210}=1.4\times 10^{44}\mathrm{erg}\mathrm{s}^1`$. These results are completely consistent with the known relationship between cluster temperature and X-ray luminosity (e.g. Markevitch 1998). If, instead, this source is at the same distance as NGC 4258, the 2–10 keV luminosity is only $`L_{210}=4\times 10^{39}\mathrm{erg}\mathrm{s}^1`$.
## 6. Discussion and Conclusions
The X-ray spectrum of NGC 4258 shows 4 distinct components:
1. a power-law component with a 0.5–10 keV luminosity of $`7.9\times 10^{40}\mathrm{erg}\mathrm{s}^1`$ and photon index $`\mathrm{\Gamma }=1.79_{0.11}^{+0.31}`$. This component is readily identified as the primary X-ray emission from the AGN central engine itself. This emission is absorbed by an intrinsic column density of $`N_\mathrm{H}=8.2\pm 0.9\times 10^{22}\mathrm{cm}^2`$. Note that this absorbing column is a factor of 1.4 smaller than that obtained by M94. Also note that our value of $`N_\mathrm{H}`$ is insensitive to exactly which of the acceptable spectral models we use (cf. Gammie, Narayan & Blandford 1999; Chary et al. 2000).
2. thermal plasma emission from optically-thin gas at a temperature of $`kT0.5\mathrm{keV}`$. The total luminosity of this component is $`2\times 10^{40}\mathrm{erg}\mathrm{s}^1`$, approximately 10–25% of the power-law luminosity (depending on how low in energy the power-law extends).
3. a bremstrahlung component with $`kT>5\mathrm{keV}`$. This component may represent thermal emission from a very hot gaseous component in the anomalous arms and/or the integrated hard X-ray emission of the X-ray binary population.
4. a narrow fluroescent K$`\alpha `$ emission line of cold iron with an equivalent width of $`W_{K\alpha }=107_{37}^{+42}\mathrm{eV}`$.
We suggest that the bulk of the narrow iron emission line originates from the accretion disk. With this assumption, we have shown that the most of the line emission must originate at relatively large radii ($`r100GM/c^2`$) in order to produce the small line width. This, in turn, implies a large X-ray source (with a size $`D100GM/c^2`$), in contrast to the situation normally found in higher-luminosity Seyfert galaxies (e.g. Nandra et al. 1997). Note that a model in which the X-ray source is small ($`D10GM/c^2`$) but the inner region of the disk are too hot/ionized to produce line emission has difficulty producing the observed equivalent width of the line. Thus, under the assumption that the bulk of the observed iron line originates from the accretion disk, there appears to be a clear difference in the size/structure of the X-ray source between this low-luminosity source and higher-luminosity Seyfert galaxies.
However, we cannot rule out the possibility that the narrow iron line originates from some previously undetected distant matter not associated with the accretion disk. If the observed line does originate from a non-disk source, our data are consistent with but do not require the presence of a “Seyfert-like” relativistic broad iron line. Even in this case, interesting constraints can be placed on the parameter space of possible relativistic broad iron lines.
This AGN is ripe for study with the new generation of X-ray observatories. The high spatial resolution of Chandra will allow us to study large scale structure in this source directly. We will be able to pinpoint the location and nature of the thermal component seen in our ASCA data. We will also be able to spatially separate emission from the anomalous arms, the X-ray binary population, and the AGN. High signal-to-noise spectroscopy with XMM (and, in the longer term, Constellation-X) will allow the inner accretion disk to be probed.
We thank Mitch Begelman, Jim Chiang, Andy Fabian and Julian Krolik for useful discussions throughout the course of this work. CSR appreciates support from Hubble Fellowship grant HF-01113.01-98A. This grant was awarded by the Space Telescope Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., for NASA under contract NAS 5-26555. We also appreciate support from NASA under LTSA grant NAG5-6337, and the National Science Foundation under grants AST-9529170 and AST-9876887. PRM is supported by the NASA Astrophysical Theory Program under grant NAG5-4061 and by NSF under grant AST-9900871. MAN is supported by NASA under LTSA grant NAG5-3225.
|
warning/0004/math0004035.html
|
ar5iv
|
text
|
# Pre-Lagrangian Submanifolds In Contact ManifoldsProject 19871044 Supported by NSF
## 1 Introduction and Results
Let $`U`$ be a $`(2n1)`$-dimensional manifolds. A contact structure $`\xi `$ on $`U`$ is a completely nonintegrable codimension $`1`$ tangent distribution. It means that $`\xi `$ can be defined, at least locally, by a $`1`$form $`\lambda `$ with $`\lambda (d\lambda )^{n1}0`$. Note that if $`n`$ is odd then the contact distribution $`\xi `$ is automatically orientable. For an even $`n`$ the existence of a contact structure implies the orientability of the ambient manifold $`U`$. In both cases, the coorientability of $`\xi `$ implies that $`\xi `$ and $`U`$ are both orientable. We will asume from now on that $`\xi `$ is coorientable and fix its orientation. Then $`\xi `$ can be globally defined by a $`1`$form $`\lambda `$, which is determined up to a multiplication by a positive function. Let $`SU=U\times ]0,\mathrm{}[`$. We will still denote by $`\lambda `$ the pull back of $`\lambda `$ by the projection of $`SU=U\times R_+`$ on the first factor and denote by $`t`$ the projection on the second. Then the form $`\omega =d(a\lambda )`$ defines a symplectic structure on $`SU`$(indeed, $`(d(a\lambda ))^n=a^{n1}da\lambda d\lambda ^{n1}0)`$. The map $`(x,a)(x,a/f(x))`$ induces an isomorphism of forms $`d(a\lambda )`$ and $`d(a\mu )`$ for $`\mu =f\lambda `$. Therefore, the symplectic manifold $`(SU,\omega )`$ depends, up to a symplectomorphism, only on the contact manifolds $`(U,\xi )`$ and not on the choice of the $`1`$form $`\lambda `$. For an $`n`$dimensional manifolds $`M`$ let us denote by $`P_+T^{}(M)`$ the oriented projective cotangent bundle of $`M`$ with the contact structure $`\xi `$ defined by the form $`pdq`$. The manifold $`P_+T^{}M`$ can also be considered a space of cooriented $`(n1)`$-dimensional contact elements of $`M`$. With this interpretation the plane $`\xi _x`$ of $`\xi `$ at a point $`x=(p,q)`$, $`qM`$, $`pT_q^{}(M)`$, consists of infinitesimal deformations of $`\xi _x`$, which leaves fixed the point of contact $`qM`$. Then the symplectization $`Sympl(P_+T^{}(M),\xi )`$ is isomorphic to $`T^{}(M)M`$ with the standard symplectic structure $`\omega =d(pdq)`$, for more example, see\[1-6\].
According to , the following notion was suggested by D. Bennequin.
An $`n`$dimensional submanifolds $`L`$ of the $`(2n1)`$dimensional contact manifold $`(U,\xi )`$ is called $`preLagrangian`$ if it satisfies the following two conditions:
a. $`L`$ is transverse to $`\xi `$;
b. The distribution $`\xi T(L)`$ is integrable and can be defined by a closed $`1`$form.
For any pre-Lagrangian submanifold $`LU`$ there exists a Lagrangian submanifold $`\widehat{L}S_\xi U`$ such that $`\pi (\widehat{L})=L`$. The cohomology class $`\lambda H^1(L;R)`$, such that $`\pi ^{}\lambda =[\alpha _\xi |\widehat{L}]`$, is defined uniquely up to multiplication by a non-zero constant. Conversly, if $`LU`$ is the (embedded) image of a Lagrangian submanifold $`\widehat{L}S_\xi U`$ under the projection $`S_\xi UU`$ then $`L`$ is pre-Lagrangian(see). Thus with any pre-Lagrangian submanifold $`LU`$ one can canonically associate a projective class of the form $`\lambda `$. The main result of this paper is following:
###### Theorem 1.1
There does not exist any pre-Lagrangian submanifold $`LU`$ with the canonical projective class equal to zero, especially any simply connected manifold can not be embedded in $`(U,\xi )`$ as a pre-Lagrangian submanifold.
###### Theorem 1.2
Let $`(U,\lambda )`$ be a closed contact manifold and $`\phi :L(U,\lambda )`$ a closed Pre-Lagrangian embedding, then $`[\phi ^{}(\lambda )]0`$ in $`H^{}(L,R)`$, especially $`H^1(L)0`$.
Sketch of proofs: We will work in the framework proposed by Gromov in . In Section 2, we study the linear Cauchy-Riemann operator and sketch some basic properties. In section 3, we study the space $`𝒟(V,W)`$ of contractible disks in manifold $`V`$ with boundary in Lagrangian submanifold $`W`$ and construct a Fredholm section of tangent bundle of $`𝒟(V,W)`$. In Section 4, we use the Gromov’s trick in to estimate the energy of the solutions of the nonlinear Cauchy-Riemann equations. In the final section, we use Gromov’s nonlinear Fredholm trick to complete our proof as in .
## 2 Linear Fredholm Theory
For $`100<k<\mathrm{}`$ consider the Hilbert space $`V_k`$ consisting of all maps $`uH^{k,2}(D,C^n)`$, such that $`u(z)R^nC^n`$ for almost all $`zD`$. $`L_{k1}`$ denotes the usual Hilbert $`L_{k1}`$space $`H_{k1}(D,C^n)`$. We define an operator $`\overline{}:V_pL_p`$ by
$$\overline{}u=u_s+iu_t$$
(2.1)
where the coordinates on $`D`$ are $`(s,t)=s+it`$, $`D=\{z||z|1\}`$. The following result is well known(see).
###### Proposition 2.1
$`\overline{}:V_pL_p`$ is a surjective real linear Fredholm operator of index $`n`$. The kernel consists of the constant real valued maps.
Let $`(C^n,\sigma =Im(,))`$ be the standard symplectic space. We consider a real $`n`$dimensional plane $`R^nC^n`$. It is called Lagrangian if the skew-scalar product of any two vectors of $`R^n`$ equals zero. For example, the plane $`p=0`$ and $`q=0`$ are Lagrangian subspaces. The manifold of all (nonoriented) Lagrangian subspaces of $`R^{2n}`$ is called the Lagrangian-Grassmanian $`\mathrm{\Lambda }(n)`$. One can prove that the fundamental group of $`\mathrm{\Lambda }(n)`$ is free cyclic, i.e. $`\pi _1(\mathrm{\Lambda }(n))=Z`$. Next assume $`(\mathrm{\Gamma }(z))_{zD}`$ is a smooth map associating to a point $`zD`$ a Lagrangian subspace $`\mathrm{\Gamma }(z)`$ of $`C^n`$, i.e. $`(\mathrm{\Gamma }(z))_{zD}`$ defines a smooth curve $`\alpha `$ in the Lagrangian-Grassmanian manifold $`\mathrm{\Lambda }(n)`$. Since $`\pi _1(\mathrm{\Lambda }(n))=Z`$, one have $`[\alpha ]=ke`$, we call integer $`k`$ the Maslov index of curve $`\alpha `$ and denote it by $`m(\mathrm{\Gamma })`$, see().
Now let $`z:S^1R^nC^n`$ be a smooth curve. Then it defines a constant loop $`\alpha `$ in Lagrangian-Grassmanian manifold $`\mathrm{\Lambda }(n)`$. This loop defines the Maslov index $`m(\alpha )`$ of the map $`z`$ which is easily seen to be zero.
Now Let $`(V,\omega )`$ be a symplectic manifold and $`WV`$ a closed Lagrangian submanifold. Let $`u:D^2V`$ be a smooth map homotopic to constant map with boundary $`DW`$. Then $`u^{}TV`$ is a symplectic vector bundle and $`(u|_D)^{}TW`$ be a Lagrangian subbundle in $`u^{}TV`$. Since $`u`$ is contractible, we can take a trivialization of $`u^{}TV`$ as
$$\mathrm{\Phi }(u^{}TV)=D\times C^n$$
and
$$\mathrm{\Phi }(u|_D)^{}TW)S^1\times C^n$$
Let
$$\pi _2:D\times C^nC^n$$
then
$$\overline{u}:zS^1\pi _2\mathrm{\Phi }(u|_D)^{}TW(z)\mathrm{\Lambda }(n).$$
Write $`\overline{u}=u|_D`$.
###### Lemma 2.1
Let $`u:(D^2,D^2)(V,W)`$ be a $`C^k`$map $`(k1)`$ as above. Then,
$$m(u|_D)=0$$
Proof. Since $`u`$ is contractible in $`V`$ relative to $`W`$, we have a homotopy $`\mathrm{\Phi }_s`$ of trivializations such that
$$\mathrm{\Phi }_s(u^{}TV)=D\times C^n$$
and
$$\mathrm{\Phi }_s(u|_D)^{}TW)S^1\times C^n$$
Moreover
$$\mathrm{\Phi }_0(u|_D)^{}TW=S^1\times R^n$$
So, the homotopy induces a homotopy $`\overline{h}`$ in Lagrangian-Grassmanian manifold. Note that $`m(\overline{h}(0,))=0`$. By the homotopy invariance of Maslov index, we know that $`m(u|_D)=0`$.
Consider the partial differential equation
$`\overline{}u+A(z)u=0onD`$ (2.2)
$`u(z)\mathrm{\Gamma }(z)R^nforzD`$ (2.3)
$`\mathrm{\Gamma }(z)GL(2n,R)Sp(2n)`$ (2.4)
$`m(\mathrm{\Gamma })=0`$ (2.5)
For $`100<k<\mathrm{}`$ consider the Banach space $`\overline{V}_k`$ consisting of all maps $`uH^{k,2}(D,C^n)`$ such that $`u(z)\mathrm{\Gamma }(z)`$ for almost all $`zD`$. Let $`L_{k1}`$ the usual $`L_{k1}`$space $`H_{k1}(D,C^n)`$ and
$$L_{k1}(S^1)=\{uH^{k1}(S^1)|u(z)\mathrm{\Gamma }(z)R^nforzD\}$$
We define an operator $`P`$: $`\overline{V}_kL_{k1}\times L_{k1}(S^1)`$ by
$$P(u)=(\overline{}u+Au,u|_D)$$
(2.6)
where $`D`$ as in (2.1).
###### Proposition 2.2
$`\overline{}:\overline{V}_pL_p`$ is a real linear Fredholm operator of index n.
## 3 Nonlinear Fredholm Theory
3.1. Adapted metrics in symplectic manifold $`(M,\omega )`$. A Riemannian metric $`g`$ on $`M`$ is called adapted (to the symplectic form $`\omega `$) if $`g+\sqrt{1}\omega `$ is a Hermitian metric with respect to some almost complex structure $`J:T(M)T(M)`$ preserving $`g`$ and $`\omega `$. This is equivalent to the existence of a $`g`$orthonormal coframe $`x_i,y_i`$,$`i=1,\mathrm{},n=dimM/2`$, at each point in $`M`$ such that $`\omega `$ equals $`_1^nx_iy_i`$ at this point. Yet another equivalent definition reads
$$dH_g=grad_\omega H_g$$
for all smooth functions $`H`$ on $`M`$, where, recall, $`grad_\omega H`$ is the (Hamiltonian) vector field which is $`\omega `$dual to $`dH`$.
Let us show that a complete adapted metric always exists.
###### Lemma 3.1
(Eliashberg-Gromov). Every symplectic manifold $`M=(M,\omega )`$ admits a complete adapted metric $`g`$.
Proof(due to ). The required metric will be constructed starting with arbitrary adapted metric $`g_0`$ and applying a certain symplectic automorphism $`A`$ of $`T^{}(M)`$ to it. This $`A`$ is constructed with an exhaustion of $`M`$ by compact domains with smooth boundaries $`S_i`$ expands $`g_0`$ transversally to all $`S_i`$. Namely, we take small $`\epsilon _i`$neighbourhoods $`N_iM`$ of $`S_i`$, normally (with respect to $`g_0`$) decomposed as $`N_i=S_i\times [\epsilon _i,\epsilon _i]`$. We denote by $`\mathrm{\Sigma }_iT(N_i)`$ and $`\nu _iT(N_i)`$ the subbundles tangent and normal to the slices $`S_i\times t`$, $`t[\epsilon _i,\epsilon _i]`$, respectively and take some symplectic automorphisms $`A_i:T(N_i)T(N_i)`$ preserving the decomposition $`T(N_i)=\mathrm{\Sigma }_i\nu _i`$ and acting on $`\nu _i`$ by $`A_i(\nu )=2\nu `$. Then $`A`$ is taken equal to $`Id`$ outside all $`N_i`$ and $`A|T(N_i)=^{def}A^{\phi _i}`$ where $`\phi _i(s,t)=\phi _i(t)`$ is a suitable sequence of positive functions on $`[\epsilon _i,\epsilon _i]`$ such that $`\phi _i`$ vanish at ends $`\pm \epsilon _i`$ and are large and fast growing with $`i`$ on the subsegments $`[\epsilon _i/2,\epsilon _i/2]`$. Clearly $`g=Ag_0`$ is complete(as well as adapted) for suitable $`\phi _i`$.
3.2. Construction of Lagrangian submanifolds. Let $`(V^{},\omega ^{})`$($`\omega =d\alpha ^{}`$) be an exact symplectic manifold and $`W^{}V^{}`$ a closed submanifolds, we call $`W^{}`$ an exact Lagrangian submanifold if $`\alpha ^{}|W^{}`$ an exact form, i.e., $`\alpha ^{}|W^{}=df`$. Consider an isotopy of Lagrange submanifolds in $`V^{}`$ given by a $`C^{\mathrm{}}`$map $`F^{}:W^{}\times [0,1]V^{}`$ and let $`\stackrel{~}{\omega }^{}`$ be the pull-back of the form $`\omega ^{}`$ to $`W^{}\times [0,1]`$. The form $`\stackrel{~}{\omega }^{}`$ clearly is exact since $`\omega ^{}=d\alpha ^{}`$, $`\stackrel{~}{\omega }^{}=d\stackrel{~}{l}^{}`$, where the $`1`$form $`\stackrel{~}{l}^{}`$ is closed on $`W^{}\times t`$ for $`t[0,1]`$. Recall that $`F^{}`$ is called an $`exact`$ $`isotopy`$ if the class $`[\stackrel{~}{l}^{}|W^{}\times t]H^1(W^{}=W^{}\times t;R)`$ is constant in $`t[0,1]`$, for more detail see\[$`5,2.3B^{}`$\].
Let $`U`$ a contact manifold and $`LU`$ an exact pre-Lagrangian submanifold, as proved in , that one can choose a contact form $`\lambda `$ on $`U`$ such that $`(V^{},\omega ^{})=(U\times R_+,d(a\lambda ))`$ and $`d(a\lambda )|L\times \{1\}=0`$ and $`\lambda |L`$ is exact. So $`W^{}(=\{1\}\times L)V^{}`$ an exact Lagrangian submanifold in $`V^{}`$ and the manifold $`(SU,d(a\lambda ))`$ has a canonical diffeotopy $`v^{}sv^{}`$ for $`s[0,\mathrm{})`$. The induced isotopy on $`L`$ clearly is Lagrange; it is exact if and only if the form $`l^{}|W^{}`$ is exact. The isotopied manifolds $`W_s^{}=s(W^{})`$ are disjointed from $`W^{}`$ for any $`s`$. We choose a positive number $`s_0`$ small enough which will be determined in section 5 and define
$$F^{}:W^{}\times [0,1]SU$$
as
$$F^{}((w,1),t)=(w,1+ts_0)$$
Then one can easily check that $`F^{}`$ is an exact Lagrangian isotopy of $`W^{}`$ in $`SU`$.
Let $`(V,\omega )=(V^{}\times C,\omega ^{}\omega _0)`$. As in , we use figure eight trick to construct a Lagrangian submanifold in $`V`$ through the Lagrange isotopy $`F^{}`$ in $`V^{}`$. Fix a positive $`\delta <1`$ and take a $`C^{\mathrm{}}`$-map $`\rho :S^1[0,1]`$, where the circle $`S^1`$ ia parametrized by $`\mathrm{\Theta }[1,1]`$, such that the $`\delta `$neighborhood $`I_0`$ of $`0S^1`$ goes to $`0[0,1]`$ and $`\delta `$neighbourhood $`I_1`$ of $`\pm 1S^1`$ goes $`1[0,1]`$. Let
$`\stackrel{~}{l}`$ $`=`$ $`\psi (w^{},\rho (\mathrm{\Theta }))\rho ^{}(\mathrm{\Theta })d\mathrm{\Theta }`$ (3.1)
$`=`$ $`\mathrm{\Phi }d\mathrm{\Theta }`$ (3.2)
be the pull-back of the form $`\stackrel{~}{l}^{}=\psi (w^{},t)dt`$ to $`W^{}\times S^1`$ under the map $`(w^{},\mathrm{\Theta })(w^{},\rho (\mathrm{\Theta }))`$ and assume without loss of generality $`\mathrm{\Phi }`$ vanishes on $`W^{}\times (I_0I_1)`$.
Next, consider a map $`\alpha `$ of the annulus $`S^1\times [\mathrm{\Phi }_{},\mathrm{\Phi }_+]`$ into $`R^2`$, where $`\mathrm{\Phi }_{}`$ and $`\mathrm{\Phi }_+`$ are the lower and the upper bound of the fuction $`\mathrm{\Phi }`$ correspondingly, such that
$`(i)`$ The pull-back under $`\alpha `$ of the form $`dxdy`$ on $`R^2`$ equals $`d\mathrm{\Phi }d\mathrm{\Theta }`$.
$`(ii)`$ The map $`\alpha `$ is bijective on $`I\times [\mathrm{\Phi }_{},\mathrm{\Phi }_+]`$ where $`IS^1`$ is some closed subset, such that $`II_0I_1=S^1`$; furthermore, the origin $`0R^2`$ is a unique double point of the map $`\alpha `$ on $`S^1\times 0`$, that is
$$0=\alpha (0,0)=\alpha (\pm 1,0),$$
and $`\alpha `$ is injective on $`S^1=S^1\times 0`$ minus $`\{0,\pm 1\}`$.
$`(iii)`$ The curve $`S_0^1=\alpha (S^1\times 0)R^2`$ “bounds” zero area in $`R^2`$, that is $`_{S_0^1}x𝑑y=0`$, for the $`1`$form $`xdy`$ on $`R^2`$.
###### Proposition 3.1
Let $`V^{}`$, $`W^{}`$ and $`F^{}`$ as above. Then there exists an exact Lagrangian embedding $`F:W^{}\times S^1V^{}\times R^2`$ given by $`F(w^{},\mathrm{\Theta })=(F^{}(w^{},\rho (\mathrm{\Theta })),\alpha (\mathrm{\Theta },\mathrm{\Phi }))`$.
Proof. Similar to \[5,2.3$`B_3^{}`$\].
3.3. Formulation of Hilbert manifolds. Now let $`(U,\lambda )`$ be a contact manifold with contact form $`\lambda `$. Let $`SU=(U\times ]0,\mathrm{}[,d(a\lambda )`$ be its symplectization. By Lemma 3.1, one has
###### Proposition 3.2
There exists an adapted complete metric on the symplectization $`SU=(U\times ]0,\mathrm{}[,d(a\lambda ))`$ of contact manifolds $`(U,\lambda )`$.
In the following we denote by $`(V,\omega )=(SU\times R^2,d(a\lambda )dxdy))`$ with the adapted metric $`gg_0`$ and $`WV`$($`W=F(W^{}\times S^1)`$) the Lagrangian submanifold constructed in section 3.2.
Let $`k100`$ and
$$𝒟^k(V,W,p)=\{uH^k(D,V)|u(D)W,uhomotopictou_0=p,u(1)=p\}$$
###### Lemma 3.2
Let $`W`$ be a closed Lagrangian submanifold in $`V`$. Then, $`𝒟^k(V,W,p)`$ is a Hilbert manifold with the tangent bundle
$$T𝒟^k(V,W,p)=\underset{u𝒟^k(V,W,p)}{}\mathrm{\Lambda }^{k1}(u^{}TV,u|_D^{}TW,p)$$
(3.3)
here
$`\mathrm{\Lambda }^{k1}(u^{}TV,u|_D^{}TW,p)=`$ (3.4)
$`\{H^{k1}sectionsof(u^{}(TV),(u|_D)^{}TL)whichvanishesat1\}`$ (3.5)
Proof: See.
Now we construct a nonlinear Fredholm operator from $`𝒟^k(V,W,p)`$ to $`T𝒟^k(V,W,p)`$ follows in . Let $`\overline{}:𝒟^k(V,W,p)T𝒟^k(V,W,p)`$ be the Cauchy-Riemmann Section induced by the Cauchy-Riemann operator, locally,
$$\overline{}u=\frac{u}{s}+J\frac{u}{t}$$
(3.6)
for $`u𝒟^k(V,W,p)`$.
Since the space $`𝒟^k(V,W,p)`$ is Hilbert manifold, the tangent space $`T𝒟^k(V,W,p)`$ is trivial, i.e. there exists a bundle isomorphism
$$\mathrm{\Phi }:T𝒟^k(V,W,p)𝒟^k(V,W,p)\times E$$
where $`E`$ is a Hilbert Space. Then the Cauchy-Riemann section $`\overline{}`$ on $`T𝒟^k(V,W,p)`$ induces a nonlinear map
$$\mathrm{\Phi }\overline{}:𝒟^k(V,W,p)E$$
In the following, we still denote $`\mathrm{\Phi }\overline{}`$ by $`\overline{}`$ for convenience. Now we define
$`F:𝒟^k(V,W,p)E`$ (3.7)
$`F(u)=\mathrm{\Phi }(\overline{}u)`$ (3.8)
###### Theorem 3.1
The nonlinear operator $`F`$ defined in (3.6-3.7) is a nonlinear Fredholm operator of Index zero.
Proof. According to the definition of the nonlinear Fredholm operator, we need to prove that $`u𝒟^k(V,W,p)`$, the linearization $`DF(u)`$ of $`F`$ at $`u`$ is a linear Fredholm operator. Note that
$$DF(u)=D\overline{}_{[u]}$$
(3.9)
where
$$(D\overline{}_{[u]})v=\frac{v}{s}+J\frac{v}{t}+A(u)v$$
(3.10)
with
$$v|_D(u|_D)^{}TW$$
here $`A(u)`$ is $`2n\times 2n`$ matrix induced by the torsion of almost complex structure, see for the computation.
Observe that the linearization $`DF(u)`$ of $`F`$ at $`u`$ is equivalent to the following Lagrangian boundary value problem
$`{\displaystyle \frac{v}{s}}+J{\displaystyle \frac{v}{t}}+A(u)v=f,v\mathrm{\Lambda }^k(u^{}TV)`$ (3.11)
$`v(t)T_{u(t)}W,tD`$ (3.12)
One can check that (3.10-11) defines a linear Fredholm operator. In fact, by proposition 2.2 and Lemma 2.1, since the operator $`A(u)`$ is a compact, we know that the operator $`F`$ is a nonlinear Fredholm operator of the index zero.
###### Definition 3.1
A nonlinear Fredholm $`F:XY`$ operator is proper if any $`yY`$, $`F^1(y)`$ is finite or for any compact set $`KY`$, $`F^1(K)`$ is compact in $`X`$.
###### Definition 3.2
$`deg(F,y)=\mathrm{}\{F^1(y)\}mod2`$ is called the Fredholm degree of a nonlinear proper Fredholm operator(see).
###### Theorem 3.2
Assum that the nonlinear Fredholm operator $`F:𝒟^k(V,W,p)E`$ constructed in (3.6-7) is proper. Then,
$$deg(F,0)=1$$
Proof: We assume that $`u:DV`$ be a $`J`$holomorphic disk with boundary $`u(D)W`$. Since almost complex structure $`\stackrel{~}{J}`$ tamed by the symplectic form $`\omega `$, by stokes formula, we conclude $`u:D^2w`$ is a constant map. Because $`u(1)=p`$, We know that $`F^1(0)=p`$. Next we show that the linearizatioon $`DF(p)`$ of $`F`$ at $`p`$ is an isomorphism from $`T^p𝒟(V,W,p)`$ to $`E`$. This is equivalent to solve the equations
$`{\displaystyle \frac{v}{s}}+J{\displaystyle \frac{v}{t}}=f`$ (3.13)
$`v|_DT_pW`$ (3.14)
By Lemma 3.1, we know that $`DF(p)`$ is an isomorphism. Therefore $`deg(F,0)=1`$.
###### Corollary 3.1
$`deg(F,w)=1`$ for any $`wE`$.
Proof. Using the connectedness of $`E`$ and the homotopy invariance of $`deg`$.
## 4 Non-properness of Fredholm Operator
We shall prove in this section that the operator $`F:𝒟E`$ constructed in the above section is non proper along the line in .
4.1. Anti-holomorphic section. Let $`C=R^2`$ and $`(V^{},\omega ^{})`$, $`(V,\omega )=(V^{}\times C,\omega ^{}\omega _0)`$, and $`W`$ as in section 3 and $`J=J^{}i`$, $`g=g^{}g_0`$, $`g_0`$ the standard metric on $`C`$.
Now let $`cC`$(here $`C`$ the complex plane) be a non-zero vector. We consider the equations
$`v=(v^{},f):DV^{}\times C`$
$`\overline{}_J^{}v^{}=0,\overline{}f=c`$
$`v|_D:DW`$ (4.1)
here $`v`$ homotopic to constant map $`\{p\}`$ relative to $`W`$. Note that $`WV\times B_R(0)`$(here $`R`$ depends on the $`s_0`$ in section 3.2).
###### Lemma 4.1
Let $`v`$ be the solutions of (4.1), then one has the following estimates
$`E(v)=\{{\displaystyle _D}(g^{}({\displaystyle \frac{v^{}}{x}},J^{}{\displaystyle \frac{v^{}}{x}})+g^{}({\displaystyle \frac{v^{}}{y}},J^{}{\displaystyle \frac{v^{}}{y}})`$
$`+g_0({\displaystyle \frac{f}{x}},i{\displaystyle \frac{f}{x}})+g_0({\displaystyle \frac{f}{y}},i{\displaystyle \frac{f}{y}}))d\sigma \}4\pi R^2.`$ (4.2)
Proof: Since $`v(z)=(v^{}(z),f(z))`$ satisfy (4.1) and $`v(z)=(v^{}(z),f(z))V^{}\times C`$ is homotopic to constant map $`v_0:D\{p\}W`$ in $`(V,W)`$, by the Stokes formula
$$_Dv^{}(\omega ^{}\omega _0)=0$$
(4.3)
Note that the metric $`g`$ is adapted to the symplectic form $`\omega `$ and $`J`$, i.e.,
$$g=\omega (,J)$$
(4.4)
By the simple algebraic computation, we have
$$_Dv^{}\omega =\frac{1}{4}_{D^2}(|v|^2|\overline{}v|^2)=0$$
(4.5)
and
$$|v|=\frac{1}{2}(|v|^2+|\overline{}v|^2$$
(4.6)
Then
$`E(v)`$ $`=`$ $`{\displaystyle _D}|v|`$ (4.7)
$`=`$ $`{\displaystyle _D}\{{\displaystyle \frac{1}{2}}(|v|^2+|\overline{}v|^2)\}𝑑\sigma `$
$`=`$ $`\pi |c|_{g_0}^2`$
By the equations (4.1), one get
$$\overline{}f=conD$$
(4.8)
We have
$$f(z)=\frac{1}{2}c\overline{z}+h(z)$$
(4.9)
here $`h(z)`$ is a holomorphic function on $`D`$. Note that $`f(z)`$ is smooth up to the boundary $`D`$, then, by Cauchy integral formula
$`{\displaystyle _D}f(z)𝑑z`$ $`=`$ $`{\displaystyle \frac{1}{2}}c{\displaystyle _D}\overline{z}𝑑z+{\displaystyle _D}h(z)𝑑z`$ (4.10)
$`=`$ $`\pi ic`$ (4.11)
So, we have
$$|c|=\frac{1}{\pi }|_{D^2}f(z)𝑑z|$$
(4.12)
Therefore,
$`E(v)`$ $``$ $`\pi |c|^2{\displaystyle \frac{1}{\pi }}|{\displaystyle _D}f(z)𝑑z|^2`$ (4.13)
$``$ $`{\displaystyle \frac{1}{\pi }}|{\displaystyle _D}|f(z)dz^2`$ (4.14)
$``$ $`4\pi |diam(pr_2(W))^2`$ (4.15)
$``$ $`4\pi R^2.`$ (4.16)
This finishes the proof of Lemma.
###### Proposition 4.1
For $`|c|3R`$, then the equations (4.1) has no solutions.
Proof. By (4.11), we have
$`|c|`$ $``$ $`{\displaystyle \frac{1}{\pi }}{\displaystyle _D}|f(z)||dz|`$ (4.17)
$``$ $`{\displaystyle \frac{1}{\pi }}{\displaystyle _D}diam(pr_2(W))||dz|`$ (4.18)
$``$ $`2R`$ (4.19)
It follows that $`c=3R`$ can not be obtained by any solutions.
4.2. Modification of section $`c`$. Note that the section $`c`$ is not a section of the Hilbert bundle in section 3 since $`c`$ is not tangent to the Lagrangian submanifold $`W`$, we must modify it as follows:
Let $`c`$ as in section 4.1, we define
$`c_{\chi ,\delta }(z,v)=\{\begin{array}{cc}c\hfill & \text{if }|z|12\delta \text{,}\hfill \\ 0\hfill & \text{otherwise}\hfill \end{array}`$ (4.22)
Then by using the cut off function $`\phi _h(z)`$ and its convolution with section $`c_{\chi ,\delta }`$, we obtain a smooth section $`c_\delta `$ satisfying
$`c_\delta (z,v)=\{\begin{array}{cc}c\hfill & \text{if }|z|13\delta \text{,}\hfill \\ 0\hfill & \text{if }|z|1\delta \text{.}\hfill \end{array}`$ (4.25)
for $`h`$ small enough by well-known convolution theory.
Now let $`cC`$ be a non-zero vector and $`c_\delta `$ the induced anti-holomorphic section. We consider the equations
$`v=(v^{},f):DV^{}\times C`$
$`\overline{}_J^{}v^{}=0,\overline{}f=c_\delta `$
$`v|_D:DW`$ (4.26)
Note that $`WV\times B_R(0)`$ for $`2\pi R^2`$. Then by repeating the same argument as section 4.1., we obtain
###### Lemma 4.2
Let $`v`$ be the solutions of (4.16) and $`\delta `$ small enough, then one has the following estimates
$`E(v)4\pi R^2.`$ (4.27)
and
###### Proposition 4.2
For $`|c|3R`$, then the equations (4.16) has no solutions.
###### Theorem 4.1
The Fredholm operator $`F:𝒟^k(V,W,p)E`$ is not proper.
Proof. If $`F`$ is proper, taking a path $`\gamma (\mu )`$ connecting $`0`$ and $`c`$, then $`F^1(\gamma ())`$ is a compact set in $`𝒟^k(V,W,p)`$ for $`k100`$, then the gradients of map $`v`$ have a uniforms bounds, i.e.,
$$|v|c_1forvF^1(\gamma ())$$
(4.28)
Note that $`v(1)=p`$, the above bounds imply
$$v|_DW^{}(0)\times (K,K)$$
(4.29)
for $`K`$ large enough which only depends on $`c_1`$. Since $`W^{}(0)\times (k,+k)`$ is regular embedding in $`V`$, we know that $`v|_D`$ is compact in the submanifold $`W`$. Then Theorem 3.1 and 3.2, i.e., that the index of $`F`$ is zero and $`deg(F)=1`$ implies $`F`$ can take the value $`c`$ for $`c3R`$, This contradicts Proposition 4.1. So, $`F`$ is not proper.
## 5 Nonlinear Fredholm Alternative
In this section, we use the Sacks-Uhlenbeck-Gromov’s trick and Gromov’s nonlinear Fredholm alternative to prove the existence of $`J`$-holomorphic disk with boundary in $`W`$ if $`WSU\times C`$ is Lagrangian submanifold.
Proof of Theorem 1.1. If Theorem 1.1 does not hold, i.e., there exists a exact Pre-Lagrangian submanifold $`L`$ in a contact manifold $`U`$, we use the canonical isotopy in the symplectization to construct an very small Lagrangian isotopy of $`L`$ then by the Gromov’s figure eight construction in section 3.2 we obtain the exact Lagrangian submanifold $`W`$ in $`SU\times C`$. By choosing $`s_0`$ in section 3.2 small enough such that $`4\pi R^2`$ small enough we conclude that the solutions of (4.16) is bounded by using the monotone inequality of minimal surface since the boundary of solutions of (4.16) remain in the compact manifold $`W`$. Then for large vector $`cC`$ in equations (4.16) we know that the nonlinear Cauchy-Riemann equations has no solution, this implies that the operator $`F`$ constructed in section 3.3 is not proper or the solutions of equations (4.16) is non -compact. The non-properness of the operator implies
a. The existence of $`J`$holomorphic plane $`v:CV`$ with bounded energy $`E(v)E_0`$. Since $`v`$ has a bounded image then by Gromov’s removal singularity theorem we get a non constant map $`w:S^2V`$ which contradict the exactness of $`V`$.
b. The existence of $`J`$holomorphic half plane $`v:HV`$ with boundary $`H`$ in $`W`$. Since $`v`$ has a bounded image, then by the Gromov’s removal boundary singularity we get a $`J`$holomorphic disks $`w:DV`$ with boundary in $`W`$, this contradicts that $`W`$ is an exact Lagrangian submanifold.
This implies Theorem 1.1 holds.
|
warning/0004/hep-th0004066.html
|
ar5iv
|
text
|
# hep-th/0004066 A Semi-Infinite Construction of Unitary 𝑁=2 Modules
## 1. Introduction
We construct a new, semi-infinite, realization of unitary representations of the $`N=2`$ superconformal algebra, in which every state in the module is a “collective excitation” over a filled Dirac sea of fermionic operators satisfying a nontrivial exclusion principle. The $`N=2`$ superVirasoro algebra is spanned by Virasoro generators $`_n`$, Heisenberg generators $`_n`$, and the modes of two fermionic currents $`𝒬_n`$ and $`𝒢_n`$, $`n`$. The semi-infinite realization takes a considerably different input: a given unitary $`N=2`$ module is spanned by the action of a single fermionic current $`𝒢_n`$, $`n`$, subject to the conditions $`S_a^p=0`$, where, for a fixed positive integer $`p`$,
(1.1)
$$S_a^p=\underset{\begin{array}{c}i_0<\mathrm{}<i_{p2}\\ i_0+\mathrm{}+i_{p2}=a\end{array}}{}\left(\underset{m<n}{}(i_mi_n)\right)𝒢_{i_0}\mathrm{}𝒢_{i_{p2}},a.$$
Informally, the semi-infinite construction can be rephrased by saying that no other $`N=2`$ generators except $`𝒢_n`$ are needed to span a unitary $`N=2`$ representation. The apparent “mismatch in the number of the degrees of freedom” is resolved because the module is spanned by semi-infinite forms in $`(𝒢_n)_n`$. With the “exclusion principle” $`S_a^p=0`$ imposed in addition to the standard Pauli principle for fermions, it is a nontrivial result (a character identity) that there are precisely as many semi-infinite forms as there are states in a unitary $`N=2`$ module. Moreover, the semi-infinite forms carry a representation of the $`N=2`$ algebra with the central charge $`3(1\frac{2}{p})`$; although eventually related to the structure of the right-hand side of (1.1), this algebra action is highly non-obvious from the conditions $`S_a^p=0`$ imposed on the semi-infinite forms.
In physical terms, semi-infinite realizations of representations of infinite-dimensional algebras are collective effects of a “quasiparticle” type: the representation space is filled by excitations over a Dirac sea of operators satisfying an exclusion principle. The resulting nonstandard realizations of representations can be viewed from the utilitarian standpoint as a particular quasiparticle basis in the space of states. In general, it is in no way obvious from the construction that these quasiparticle states exactly fill an irreducible representation of any algebra. But in the cases where this is so, calculating the character in two ways (in accordance with the quasiparticle picture and in the standard basis) gives a nontrivial identity of the type of generalized Rogers–Ramanujan identities . These identities (often called the Rogers–Ramanujan–Gordon–Andrews identities), originally motivated by combinatorial correspondences,<sup>1</sup><sup>1</sup>1We recall the classical statement (see ): “The partitions of an integer $`n`$ for which the difference between any two parts is not smaller than two are equinumerous to the partitions of $`n`$ into parts $`1`$ or $`4(mod5)`$.” Relations between the Rogers–Ramanujan identities and conformal field theory characters go back to . were investigated by different methods, in particular with regard to their relation to conformal field theory . Semi-infinite constructions of representations thus give a representation-theory interpretation of a number of nontrivial combinatorial phenomena . The corresponding character formulas can be interpreted as a result of the simultaneous existence and interaction of particles of several types.<sup>2</sup><sup>2</sup>2The standard logic is as follows . Consider, for example, two types of bosons (of “colors” 1 and 2); without the interaction between the bosons, the partition function is $`1/(q)_{\mathrm{}}^2`$, which equals $`_{n_10}_{n_20}q^{n_1+n_2}/(q)_{n_1}(q)_{n_2}`$ by the Durfee formula (we use the standard notation in Eq. (1.7)). The “semi-infinite” character formulas have a similar form, however the exponent in the numerator acquires a quadratic form in $`n_1,n_2,\mathrm{}`$, which is interpreted as an interaction between the different types of quasiparticles. Formulas of this type for the partition function on the torus are related to the thermodynamic Bethe ansatz and were also investigated in , .
In semi-infinite constructions, the representation space (rigorously defined as an inductive limit) is generated by semi-infinite forms, or the products
(1.2)
$$V_{\alpha _1}^{(s_1)}V_{\alpha _2}^{(s_2)}\mathrm{}V_{\alpha _n}^{(s_n)}\mathrm{}.$$
The subscripts (“modes”) take an infinite number of values, for example $`\alpha _i`$, and it can be assumed as a rule that $`\alpha _1\alpha _2\mathrm{}`$ . The superscripts distinguish a finite number ($`s_1,s_2,\mathrm{}\{1,2,\mathrm{},M\}`$) of “types” of elements (all elements are the same type in the simplest case). It is assumed that starting with some number $`\iota `$, the sequence $`(s_i,\alpha _i)_{i\iota }`$ is periodic, i.e., the shift $`\alpha _i\alpha _i+\nu `$ maps the sequence into itself for some $`\nu `$. An essential point is that the semi-infinite forms are then considered modulo some identifications, whose role amounts to the possibility of expressing a semi-infinite form with a sequence $`(\alpha _i)`$ that is “too dense” through a linear combination of “thinned out” semi-infinite forms.
In the known constructions of the semi-infinite type, the elements $`V^{(s)}`$ are either some vertex operators for a given infinite-dimensional algebra or some currents taking values in the algebra. Thus, in the semi-infinite construction of for an affine algebra $`\widehat{𝔞}`$, the elements $`V^{(s)}`$ are a part of the currents with the values in the nilpotent subalgebra of $`𝔞`$; a description of the same space in terms of different operators is given in , where the semi-infinite construction is intermediate (in a certain sense) between the case where the operators $`V^{(s)}`$ are currents in the algebra and the case where they are vertex operators acting between different modules. Semi-infinite constructions for $`\widehat{s\mathrm{}}(2)`$ modules where $`V^{(s)}`$ are vertex operators arise from the Haldane–Shastry spin chains and the related Calogero model with spin . The corresponding quasiparticle basis is obtained as a conformal limit (taken in the neighborhood of the antiferromagnetic state) of the space of states in the Haldane–Shastry model; this limit produces a direct sum of the two integrable level-1 $`\widehat{s\mathrm{}}(2)`$ representations. A decomposition into irreducible representations of the Yangian then leads to a new basis in level-1 representations that can be written using spin-$`1/2`$ $`\widehat{s\mathrm{}}(2)`$ vertex operators, which in the context of the Haldane–Shastry model are interpreted as the creation operators for spinon excitations, quasiparticles with the spin $`1/2`$ and with a half-integer statistics , , . In the cases where $`V_\alpha ^{(s)}`$ are fermionic operators (for example, for the $`N=2`$ superconformal algebra), the semi-infinite construction can be viewed as a generalization to the “interacting” fermions of the infinite-wedge representation, which is a classical tool in investigating a number of problems in representation theory and beyond. The “interaction” here is understood not quite literally but in the sense that the semi-infinite forms are considered modulo identifications, i.e., satisfy a set of relations (as a result, in contrast to the infinite-wedge construction, semi-infinite constructions give irreducible representations or finite direct sums thereof).
Semi-infinite spaces can be investigated using filtrations by some subspaces that can be conveniently studied. For example, one of the filtrations involved in the semi-infinite construction in for the level-$`k`$ vacuum representation of the $`\widehat{s\mathrm{}}(2)`$ algebra consists of finite-dimensional spaces $`𝖬^+[\iota ]`$, which as vector spaces are isomorphic to the $`(\iota +1)`$-multiple product $`^{k+1}\mathrm{}^{k+1}`$ . This is similar to what is observed in the corner transfer matrix method, where the space of states is a semi-infinite tensor product of finite-dimensional spaces and its “approximations” have the form of the above tensor product. We recall that the results obtained in integrable statistical mechanics models indicate an intimate relation between the space of states of the corresponding lattice model and a representation of some infinite-dimensional algebra . Although the correspondence is not straightforward by far, it is very interesting to investigate the relations of semi-infinite constructions of representations to exactly integrable statistical mechanics models (this requires a $`q`$-deformation of the algebra action on the semi-infinite space).
For the $`N=2`$ super-Virasoro (superconformal) algebra, the possibility of constructing the semi-infinite realization of its unitary modules was pointed out in , and we now develop the recipe sketched there. We also note the interest in the $`N=2`$ supersymmetry precisely originating in investigations of the generalized Rogers–Ramanujan–Gordon–Andrews identities . As noted above, “reducing the number of generators” in the semi-infinite $`N=2`$ realization results in that only the $`𝒢_n`$ modes are needed for generating the entire module. This counterintuitive statement is summarized in Theorem 1.1 below (its proof starts with Theorem 3.2 and is finished in Theorem 5.5).
Let $`p3`$ be a fixed positive integer, and let $`𝖦(p)`$ be the algebra generated by anticommuting elements $`(𝒢_n)_n`$ and an invertible operator $`𝖴`$ such that $`𝖴𝒢_n𝖴^1=𝒢_{n+1}`$, with $`𝒢_n`$ satisfying the constraints $`S_a^p=0`$ $`(a)`$ with $`S_a^p`$ defined in (1.1).<sup>3</sup><sup>3</sup>3We consider only graded representations of $`𝖦(p)`$ of the form $`W=_{i0}W_i`$ with $`𝒢_nW_iW_{in}`$. Although the expression $`S_a^p`$ contains infinite sums, its action on such representations is well defined; following the standard abuse of terminology, we mean that the “algebra generated by $`𝒢_n`$” involves, in particular, infinite combinations with a well-defined action on graded spaces of the above form.
###### Theorem 1.1.
Let $`W`$ be the representation of $`𝖦(p)`$ induced from the trivial one-dimensional representation of the Grassmann algebra with the generators $`(𝒢_n)_{n0}`$ (on the vacuum vector $`|0`$). Let $`r`$ be a fixed integer such that $`1rp1`$ and let $`C_r`$ be the $`𝖦(p)`$-submodule in $`W`$ generated from the vector $`𝒢_r\mathrm{}𝒢_1|0`$ and $`V_{r,p}`$ the submodule generated from the set of vectors
(1.3)
$$𝒢_{\alpha p+1}\mathrm{}𝒢_{\alpha r1}𝒢_{\alpha r+1}\mathrm{}𝒢_{\alpha 1}|\alpha |\alpha p,\alpha ,$$
where $`|\alpha =𝖴^\alpha |0`$. The quotient space $`W/(V_{r,p}+C_r)`$ is a representation of the $`N=2`$ algebra with the central charge
(1.4)
$$c=3\left(1\frac{2}{p}\right)$$
and, moreover, is isomorphic to a direct sum of unitary $`N=2`$ representations,<sup>4</sup><sup>4</sup>4We recall (see Sec. 2.2) that unitary $`N=2`$ representations $`𝔎_{r,p;\theta }`$ with central charge (1.4) are labeled by a pair of integers $`(r,\theta )`$ such that $`0\theta r1`$ and $`1rp1`$. It can be assumed that $`0\theta p1`$ with each representation then labeled twice and the summation in (1.5) taken over the spectral flow orbit (see Sec. 2.1).
(1.5)
$$𝖶(r,p)\frac{W}{V_{r,p}+C_r}\underset{\theta =0}{\overset{p1}{}}𝔎_{r,p;\theta }.$$
The “semi-infiniteness” is here hidden in the relations $`|\alpha =𝒢_{\alpha +1}\mathrm{}𝒢_{\alpha +pr1}𝒢_{\alpha +pr+1}\mathrm{}𝒢_{\alpha +p1}|\alpha +p`$ imposed via taking the quotient; they can be applied recursively, leading to a representation of $`𝖶(r,p)`$ as the space of semi-infinite forms in $`(𝒢_n)_n`$ satisfying a set of relations (of which the most important ones follow from the conditions $`S_a^p=0`$). At the level of characters, we have a combinatorial corollary of Theorem 1.1: taking the characters of the representations whose isomorphism is established in Sec. 5.6, we obtain
###### Corollary 1.2.
For $`k`$, $`1rk+1`$, and $`0\theta k+1`$, there is the identity
(1.6)
$$\begin{array}{c}\underset{N_1\mathrm{}N_k}{}\frac{z^{\underset{m=1}{\overset{k}{}}N_m}q^{\frac{1}{2}\left(\underset{m=r}{\overset{k}{}}\underset{m=1}{\overset{r1}{}}\right)N_m\theta \underset{m=1}{\overset{k}{}}N_m+\frac{3}{2}\underset{m=1}{\overset{k}{}}N_m^2+\underset{1m<nk}{}N_mN_n}}{(q)_{N_1N_2}(q)_{N_2N_3}\mathrm{}(q)_{N_{k1}N_k}(q)_{\mathrm{}}}=\hfill \\ \hfill =\underset{\begin{array}{c}n1\\ n\theta ,r\theta mod(k+2)\end{array}}{}\frac{1+zq^n}{1q^n}\underset{\begin{array}{c}n1\\ n\theta ,\theta rmod(k+2)\end{array}}{}\frac{1+z^1q^n}{1q^n}.\end{array}$$
This involves the standard notation
(1.7)
$$(q)_n=(1q)\mathrm{}(1q^n).$$
It is useful to note that the integer parameter $`k`$ is $`p2`$ in terms of the parameter in Theorem 1.1 (regarding the quasiparticle interpretation of the left-hand side of (1.6), see footnote 2).
Although isomorphism (1.5) and therefore identity (1.6) are eventually derived from the structure of the right-hand side of (1.1), the relation between (1.1) and (1.5) is far from obvious. In this paper, we describe in detail the methods used in the construction and proofs. The problems solved in what follows can be put into a general perspective of semi-infinite constructions:
* choosing the elements whereby the space is generated and formulating a system of conditions (constraints) on these elements (for the $`N=2`$ algebra, these are the fermions $`𝒢_n`$ and conditions (1.1) respectively);
* counting the number of states remaining in each grade after imposing the conditions, i.e., finding the character (the main complication here consists in taking the constraints into account, i.e., in working with a space that is not freely generated);
* constructing an appropriate—monomial—basis in the semi-infinite space, typically by a certain “thinning out” procedure<sup>5</sup><sup>5</sup>5This point is in fact more fundamental than may seem at first glance: once established, the existence of any monomial basis, in itself nonobvious, becomes a powerful tool in analyzing spaces that are not freely generated, similarly to semi-infinite ones. The existence of a monomial basis depends on certain subtle properties of the imposed constraints and allows controlling the properties of mappings between different subspaces of the semi-infinite space. (we note that in the set of all semi-infinite forms, there are linear dependences resulting from the imposed constraints);
* constructing the algebra action on the semi-infinite forms built from the elements satisfying the chosen constraints (an obvious complication occurs in the case where, as with our construction, the semi-infinite forms constructed from the modes of a “small” number of currents must carry a representation of the entire algebra);
* decomposing the semi-infinite space into a (direct, if possible) sum of representations of the basic algebra;
* finding representations of some other algebraic structures on subspaces of the semi-infinite space.
Realizing this program (where we do not consider the last item) requires using a combination of different means. We now describe them in more detail for the $`N=2`$ superconformal algebra, essentially following the contents of this paper although in an order somewhat different from the order of sections (these methods are also interesting because the problems that we solve in constructing the semi-infinite realization are typical of a number of semi-infinite constructions beyond the present one (cf. the spinon basis construction in ).)
The semi-infinite space $`𝖶_{r,p;\theta }`$ that is eventually shown to be isomorphic to the unitary $`N=2`$ representation $`𝔎_{r,p;\theta }`$ is defined as the inductive limit of the spaces $`𝖶_{r,p;\theta }(\iota )`$ generated, as $`\iota `$ increases, from progressively more twisted highest-weight states, namely, from those states on which the progressively larger part of the generators $`𝒢_n`$ become creation operators as $`\iota \mathrm{}`$, with “only” the $`(𝒢_n)_{n\iota p+\theta }`$ generators remaining the annihilation operators.
A key role in studying the inductive limit is played by the dual space, which can be realized using “polynomials in an infinite number of variables” (more precisely, polynomial differential forms) . The dual to the quotient space in Eq. (1.5) is a subspace of polynomials satisfying certain conditions on $`(p1)`$-multiple diagonals $`x_1=\mathrm{}=x_{p1}`$ and at zero. Specifically, we investigate the space
(1.8)
$$𝖶_{r,p;0}(0)^{}[x]dxx_1,x_2dx_1dx_1x_1,x_2,x_3dx_1dx_2dx_3\mathrm{},$$
where $`x_1,\mathrm{},x_n`$ are antisymmetric polynomials in $`n`$ variables and the space $`𝖶_{r,p;0}(0)^{}`$ is singled out by the vanishing conditions on multiple diagonals and at zero. The character of this space can be evaluated by introducing a filtration on each $`x_1,\mathrm{},x_n`$ induced by the lexicographic order on partitions of $`n`$.
The sum over partitions gives a formula of the Rogers–Ramanujan–Gordon–Andrews type for the character of $`𝖶_{r,p;0}(0)`$, and after applying $`N=2`$ algebra automorphisms, also for all the spaces $`𝖶_{r,p;\theta }(\iota )`$, $`\iota `$, involved in the inductive limit. A remarkable property of these character formulas is that they admit the limit as $`\iota \mathrm{}`$. This limit is a candidate for the character of the semi-infinite space $`𝖶_{r,p;\theta }=\underset{}{\mathrm{lim}}_\iota \mathrm{}𝖶_{r,p;\theta }(\iota )`$, but proving that the limit is the character requires establishing that all the mappings $`𝖶_{r,p;\theta }(\iota )𝖶_{r,p;\theta }(\iota +1)`$ used in constructing the inductive limit are embeddings (and the limit of the characters is hence equal to the character of the inductive limit).
The above statement regarding the embeddings follows from the existence of a remarkable monomial basis consisting of the states in which the modes $`𝒢_n`$ are “thinned out” in accordance with the procedure described in Theorem 1.3 below. By the sequence of occupation numbers associated with a semi-infinite form $`𝒢_{i_1}\mathrm{}𝒢_{i_n}\mathrm{}`$ ($`i_1<\mathrm{}<i_n<\mathrm{}`$), we mean the sequence $`\left(\gamma (n)\right)`$ of elements labeled by integers $`ni_1`$ that are equal to zero except for $`\gamma (i_1)=\gamma (i_2)=\mathrm{}=1`$. The construction of the special monomial basis, called the thin basis, is given in the following theorem (in Sec. 3.2, we prove an equivalent statement in Lemma 3.3).
###### Theorem 1.3.
There exists a basis in $`𝖶_{r,p;\theta }`$ whose elements are semi-infinite forms $`𝒢_{i_1}𝒢_{i_2}`$ satisfying the following conditions:
1. for any $`n`$, $`i_{n+p2}i_np`$ (hence, any segment $`[n,n+p1]`$ of the length $`p`$ can contain at most $`p2`$ nonzero occupation numbers).
2. For $`n1`$, the sequence of occupation numbers $`\alpha (n)`$ is periodic with the period $`p`$ and exactly $`p2`$ occurrences of $`1`$ per period and is
(1.9)
$$\stackrel{p}{\stackrel{}{\underset{r1}{\underset{}{1,\mathrm{},1}},0,\underset{pr1}{\underset{}{1,\mathrm{},1}},0}},\mathrm{},\stackrel{p}{\stackrel{}{\underset{r1}{\underset{}{1,\mathrm{},1}},0,\underset{pr1}{\underset{}{1,\mathrm{},1}},0}},\mathrm{}.$$
Construction of the thin basis corresponds to a model statistical system on a semi-infinite one-dimensional lattice. Each semi-infinite form $`𝒢_{i_1}\mathrm{}𝒢_{i_n}\mathrm{}`$ can be represented as a configuration of crosses on the lattice, for example
(1.10)
$$\times \times \times \times \times \mathrm{}$$
(a site with a cross corresponds to the occupation number 1, others to 0). The “enumeration” of thin basis elements then becomes the problem of finding the partition function of all configurations of crosses with any $`p`$ consecutive lattice sites carrying at most $`p2`$ crosses. These partition functions can be analyzed using a version of the so-called Andrews–Schur method, which consists in establishing recursive relations and subsequently taking the limit as the “finitization parameter” goes to infinity. This allows us to show that the semi-infinite forms described in Theorem 1.3 indeed constitute a basis in the semi-infinite space.
It can be directly shown that the derived characters of the semi-infinite spaces $`𝖶_{r,p;\theta }`$ coincide with the characters of the corresponding unitary $`N=2`$ representations $`𝔎_{r,p;\theta }`$; from the existing mapping $`𝖶_{r,p;\theta }𝔎_{r,p;\theta }`$, it is then easy to deduce that $`𝖶_{r,p;\theta }`$ is in fact isomorphic to $`𝔎_{r,p;\theta }`$, which in turn implies the existence of the $`N=2`$ algebra action on $`𝖶_{r,p;\theta }`$. However, the explicit form of this action then remains unknown. We choose a more “conceptual” approach and directly construct the $`N=2`$ action on the semi-infinite space $`𝖶_{r,p;\theta }`$. The definition of the semi-infinite space does not suggest that this space carries an action of the generators $`𝒬_n`$, $`_n`$, and $`_n`$, with $`n`$ satisfying algebra (2.1); the existence of this action is extremely sensitive to the imposed constraints $`S_a^p=0`$. The methods for constructing the action developed here can also be useful in investigating other semi-infinite spaces where constraints of a different form can be compatible with the action of another algebra. These methods are as follows:
In constructing the $`N=2`$ algebra action on the semi-infinite space, we use a filtration of $`𝖶_{r,p;\theta }`$ by finite-dimensional spaces called the positive filtration. For affine algebras, similar subspaces are known as Demazure modules and have been studied from different standpoints (see and the bibliography therein).
On the above finite-dimensional subspaces, we construct the action of a part of the $`N=2`$ algebra generators via differential operators (in finitely many Grassmann variables $`𝒢_n`$, $`0nN`$). More precisely, differential operators a priori act on a freely generated space, and we must therefore show that their action is compatible with taking the quotient with respect to the relations induced by the basic conditions (1.1) on the subspace.
To verify compatibility of the differential operator action with relations (1.1), we again use the functional realization. The operators dual to the differential operators become the “homology-type” differentials on polynomials, which considerably simplifies the statements that must be proved.
The action of the $`N=2`$ generators on a vector of the semi-infinite space must not depend on the filtration term to which this vector is viewed to belong. We prove this independence as well as the independence of the constructed action from any arbitrariness involved in the construction. We actually construct the action of only a part of the $`N=2`$ generators that generate the entire algebra, and we then prove that these generate precisely the $`N=2`$ superconformal algebra.
As soon as it is established that the semi-infinite space $`𝖶_{r,p;\theta }`$ is a module over the $`N=2`$ algebra, it is easy to verify that the mapping $`𝖶_{r,p;\theta }𝔎_{r,p;\theta }`$ into the unitary representation is an isomorphism of $`N=2`$ modules, which completes the semi-infinite construction.
The semi-infinite construction of unitary $`N=2`$ modules is closely related to a similar construction of the unitary $`\widehat{s\mathrm{}}(2)`$ modules . In particular, the method used in constructing the $`N=2`$ algebra action on the semi-infinite space also allows us to define the $`\widehat{s\mathrm{}}(2)`$ algebra action on the corresponding semi-infinite space (Sec. 5.5). This involves the Demazure modules; the corresponding characters are related to the generalized Pascal triangles and supernomial coefficients , and in Sec. 6.1, we also consider a combinatorial construction of bases in the $`N=2`$ “Demazure” subspaces (cf. ). As another aspect of the relation between $`\widehat{s\mathrm{}}(2)`$ and $`N=2`$ structures, we consider the correspondence between the modular functors (Sec. 2.3) and use it in relating the (unitary) $`\widehat{s\mathrm{}}(2)`$ and $`N=2`$ fusion rules. The modular functor can be described somewhat more explicitly in the semi-infinite realization (see Sec. 6.2), because the representation is generated by only a part of the currents, which allows identifying the space dual to coinvariants with the space of meromorphic functions on products of punctured Riemann surfaces, with the functions required to possess a prescribed behavior on multiple diagonals (which is a counterpart of the relations $`S_a^p=0`$) and at the punctures. In some cases, the dimensions of these functional spaces can be evaluated directly; for the products of $`^1`$, on the other hand, these dimensions follow from the fusion rules for the $`N=2`$ algebra. The unitary $`N=2`$ fusion rules have been obtained from the Verlinde conjecture , but we give an independent derivation based on acting on the $`\widehat{s\mathrm{}}(2)`$ fusion rules with the functor that realizes the equivalence of categories ,<sup>6</sup><sup>6</sup>6This equivalence is “modulo” the spectral flows, and this is why the fusion algebras for the $`\widehat{s\mathrm{}}(2)`$ and $`N=2`$ theories are not identical; however, the equivalence statement is sufficiently powerful to derive the $`N=2`$ fusion algebra from the $`\widehat{s\mathrm{}}(2)`$ fusion. The correspondence between $`\widehat{s\mathrm{}}(2)`$ and $`N=2`$ fusion algebras can be considered in the framework of the correspondence between $`\widehat{s\mathrm{}}(2)`$ and $`N=2`$ modular functors in Sec. 2.3, which extends the equivalence of representation categories. see Sec. 6.3.
## 2. The $`N=2`$ algebra
In this section, we recall the main facts pertaining to the $`N=2`$ superconformal algebra and motivate the semi-infinite construction. The reader may wish to go directly to Sec. 2.4 and use Secs. 2.1, 2.2, and (partly) 2.3 for reference.
### 2.1. The $`N=2`$ algebra and the spectral flow
The $`N=2`$ superconformal algebra is generated by the bosonic operators $`_n`$ (Virasoro algebra generators) and $`_n`$ (Heisenberg algebra) and the fermionic operators $`𝒢_n`$ and $`𝒬_n`$. We consider the algebra in the basis where the nonvanishing commutation relations are given by
$`[_m,_n]=`$ $`(mn)_{m+n},`$ $`[_m,_n]=`$ $`\frac{𝖼}{3}m\delta _{m+n,0},`$
$`[_m,𝒢_n]=`$ $`(mn)𝒢_{m+n},`$ $`[_m,𝒢_n]=`$ $`𝒢_{m+n},`$
(2.1) $`[_m,𝒬_n]=`$ $`n𝒬_{m+n},`$ $`[_m,𝒬_n]=`$ $`𝒬_{m+n},`$
$`[_m,_n]=`$ $`n_{m+n}+\frac{𝖼}{6}(m^2+m)\delta _{m+n,0},`$
$`[𝒢_m,𝒬_n]=`$ $`2_{m+n}2n_{m+n}+\frac{𝖼}{3}(m^2+m)\delta _{m+n,0},`$
where $`m,n`$ and the bracket $`[,]`$ denotes the supercommutator. The central charge $`c3`$ can be conveniently parametrized as $`c=3(12/p)`$ with $`p\{0\}`$.
The algebra automorphisms include the group $``$ of automorphisms $`𝖴_\theta `$, $`\theta `$, called the spectral flow . In the basis chosen in (2.1), the spectral flow acts as
(2.2)
$$𝖴_\theta :\begin{array}{cccc}\hfill _n& _n+\theta _n+\frac{c}{6}(\theta ^2+\theta )\delta _{n,0},\hfill & \hfill _n& _n+\frac{c}{3}\theta \delta _{n,0},\hfill \\ \hfill 𝒬_n& 𝒬_{n\theta },\hfill & \hfill 𝒢_n& 𝒢_{n+\theta },\hfill \end{array}\theta .$$
The action of the spectral flow on an $`N=2`$ module gives a nonisomorphic representation in general. The modules transformed by the spectral flow are called the twisted modules. We use the notation $`𝔇_{;\theta }=𝖴_\theta 𝔇_{}`$, which is also applied to modules carrying other labels, for example, $`𝔎_{r,p;\theta }=𝖴_\theta 𝔎_{r,p}`$ for the unitary representations considered in what follows. The parameter $`\theta `$ is called the twist. We omit zero twist from the notation.
The character of an $`N=2`$ module $`𝔇`$ is defined as
(2.3)
$$\omega _𝔇(z,q)=Tr_𝔇^{}(z^_0q^_0),$$
where taking the trace over $`𝔇`$ involves a sesquilinear form that can be found in in our current notation. Under the spectral flow action, characters transform as
(2.4)
$$\omega _{𝖴_\theta 𝔇}(z,q)=z^{\frac{c}{3}\theta }q^{\frac{c}{6}(\theta ^2\theta )}\omega _𝔇(zq^\theta ,q).$$
We define the twisted highest-weight vector $`|h,p;\theta `$ as a state satisfying the annihilation conditions
(2.5) $`𝒬_{\theta +m}|h,p;\theta =𝒢_{\theta +m}|h,p;\theta =_{m+1}|h,p;\theta =_{m+1}|h,p;\theta =0,m_0`$
and also the conditions
(2.6) $`\left(_0+{\displaystyle \frac{c}{3}}\theta \right)|h,p;\theta =h|h,p;\theta ,`$
(2.7) $`\left(_0+\theta _0+{\displaystyle \frac{c}{6}}(\theta ^2+\theta )\right)|h,p;\theta =0`$
(where the second one follows from the annihilation conditions).
Anticipating some of what follows, we note that the semi-infinite construction also involves twisted states on which, however, only the action of the $`𝒢_n`$ operators is defined such that the same vanishing conditions as for the corresponding states in the unitary module are satisfied, but the action of the remaining generators must be reconstructed (such that all the relations that hold in the unitary module are satisfied).
### 2.2. Unitary representations of the $`N=2`$ algebra
. The unitary $`N=2`$ representations are periodic under the spectral flow with the period $`p`$; that is, the spectral flow $`𝖴_p`$ produces an isomorphic representation,
(2.8)
$$𝔎_{r,p;\theta +p}𝔎_{r,p;\theta }.$$
For unitary representations, the twist parameter $`\theta `$ can therefore be considered modulo $`p`$. Moreover, for a given $`p`$, there are only $`p(p1)/2`$ nonisomorphic unitary representations $`𝔎_{r,p;\theta }`$, which can be labeled by $`0\theta r1`$ and $`1rp1`$, because there are the isomorphisms of $`N=2`$ modules
(2.9)
$$𝔎_{r,p;\theta +r}𝔎_{pr,p;\theta },1rp1,\theta _p.$$
Representations with zero twist are denoted by $`𝔎_{r,p}𝔎_{r,p;0}`$ for brevity.
The characters $`\omega _{r,p;\theta }^𝔎\omega _{𝔎_{r,p;\theta }}`$ of the unitary $`N=2`$ representations $`𝔎_{r,p;\theta }`$ in the notation in (see also ) are given by
(2.10)
$$\begin{array}{c}\omega _{r,p;\theta }^𝔎(z,q)=z^{\frac{2\theta r+1}{p}1}q^{\frac{\theta r\theta ^2}{p}+\theta }\frac{\eta (q^p)^3}{\eta (q)^3}\frac{\vartheta _{1,0}(z,q)\vartheta _{1,1}(q^r,q^p)}{\vartheta _{1,0}(zq^\theta ,q^p)\vartheta _{1,0}(zq^{r\theta },q^p)},\hfill \\ \hfill 1rp1,0\theta r1,\end{array}$$
where we introduce the theta functions
(2.11) $`\vartheta _{1,1}(z,q)`$ $`=q^{\frac{1}{8}}{\displaystyle \underset{m}{}}(1)^mq^{\frac{1}{2}(m^2m)}z^m=q^{\frac{1}{8}}{\displaystyle \underset{m0}{}}(1z^1q^m){\displaystyle \underset{m1}{}}(1zq^m){\displaystyle \underset{m1}{}}(1q^m),`$
(2.12) $`\vartheta _{1,0}(z,q)`$ $`=q^{\frac{1}{8}}{\displaystyle \underset{m}{}}q^{\frac{1}{2}(m^2m)}z^m=q^{\frac{1}{8}}{\displaystyle \underset{m0}{}}(1+z^1q^m){\displaystyle \underset{m1}{}}(1+zq^m){\displaystyle \underset{m1}{}}(1q^m).`$
These characters (where we can set $`\theta =0`$ for simplicity, because the behavior of characters under the spectral flow transformations is determined by (2.4)) can also be expanded with respect to the theta functions $`\vartheta _{1,0}(,q^{p(p2)})`$ as
(2.13)
$$\begin{array}{c}\omega _{r,p}^𝔎(z,q)=z^{\frac{1+2\theta r}{p}\theta }q^{\frac{1}{2}(1\frac{2}{p})(\theta ^2\theta )\frac{1r}{p}\theta \frac{p(p2)}{8}}\times \hfill \\ \hfill \times \underset{a=0}{\overset{p3}{}}z^aq^{\frac{1}{2}(a^2(2\theta +1)a)}\vartheta _{1,0}(z^{p2}q^{pa(r1)(p2)\theta \frac{(p1)(p2)}{2}},q^{p(p2)})C_{r,p}^a(q),\end{array}$$
where the string functions $`C_{r,p}^a(q)`$ depend only on $`q`$.
### 2.3. Equivalence of categories and related issues
A linear combination of the unitary $`N=2`$ characters belonging to the same spectral flow orbit gives the characters $`\chi _{r,k}^𝔏`$ of the unitary $`\widehat{s\mathrm{}}(2)_k`$ representations $`𝔏_{r,k}`$,
(2.14)
$$\begin{array}{c}\chi _{r,p2}^𝔏(z,q)\vartheta _{1,0}(zy,q)=q^{\frac{r^21}{4p}\frac{p}{4}+\frac{1}{8}}y^{\frac{r1}{p}}z^{\frac{r1}{2}}\times \hfill \\ \hfill \times \underset{a=0}{\overset{p1}{}}\omega _{r,p}^𝔎(yq^a,q)y^az^aq^{\frac{a^2a}{2}\frac{r1}{p}a}\vartheta _{1,0}(z^py^2q^{p+r2a},q^{2p}).\end{array}$$
This follows from of the isomorphism of $`N=2`$ modules
(2.15)
$$𝔏_{r,k}\mathrm{\Omega }\underset{\theta =0}{\overset{k+1}{}}𝔎_{r,k+2;\theta }\left(\underset{\nu \sqrt{2(k+2)}}{}_{\sqrt{\frac{2}{k+2}}(\frac{r1}{2}\theta )\nu }^+\right),$$
where $`\mathrm{\Omega }`$ is the free-fermion module and $`_a^+`$ are Fock spaces (the sum over the lattice $`\sqrt{2(k+2)}`$ defines a vertex operator algebra). Equation (2.15) allows us to establish the functorial correspondence
(2.16)
$$𝔏_{r,k;}𝔎_{r,k+2;}$$
between the spectral flow orbits of each algebra. This is an equivalence between categories whose objects are spectral flow orbits ; in other words, this is the equivalence of representation categories of the algebras obtained by adding the spectral flow operator to the universal enveloping algebra (of the $`\widehat{s\mathrm{}}(2)`$ and the $`N=2`$ algebras respectively).
Expanding on the equivalence of categories, we now describe the correspondence between modular functors for the $`\widehat{s\mathrm{}}(2)`$ and $`N=2`$ theories. This correspondence is interesting, in particular, because the $`\widehat{s\mathrm{}}(2)`$ modular functor has the well-known geometric interpretation in terms of the moduli spaces of $`SL_2`$ vector bundles, whereas no a priori geometric interpretation is known for the $`N=2`$ modular functor.
For a given Riemann surface $``$, the modular functor $`\mathrm{𝖬𝖮𝖣}_{\widehat{s\mathrm{}}(2)}(k,)`$ of the level-$`k`$ $`\widehat{s\mathrm{}}(2)`$ Wess–Zumino–Witten theory and the modular functor $`\mathrm{𝖬𝖮𝖣}_{N=2}(p,)`$ of the theory based on the $`N=2`$ algebra with central charge (1.4) are related by
(2.17)
$$\mathrm{𝖬𝖮𝖣}_{N=2}(p,)=Coinv_{H^1(,_2)/V_+}^{}\left(Inv_{V_+}^{}\left(\mathrm{𝖬𝖮𝖣}_{\widehat{s\mathrm{}}(2)}(p2,)\mathrm{𝖬𝖮𝖣}_{\text{free}}(p,)\right)\right),$$
where $`\mathrm{𝖬𝖮𝖣}_{\text{free}}(p,)`$ is the modular functor of the free theory whose vertex operator algebra is associated with the lattice $`\sqrt{2p}`$ (see (2.15)). The $`\widehat{s\mathrm{}}(2)`$ modular functor $`\mathrm{𝖬𝖮𝖣}_{\widehat{s\mathrm{}}(2)}(k,)`$ carries a projective action $`𝒪()`$ of the group $`H^1(,_2)`$ such that $`𝒪(\alpha \beta )=𝒪(\alpha )𝒪(\beta )(1)^{k\alpha ,\beta }`$ for the cycles $`\alpha `$ and $`\beta `$ with the intersection form $`\alpha ,\beta `$ or, in other words, there is the action of the Heisenberg group $`\mathrm{\Gamma }`$ with the center $`_2`$ such that $`\mathrm{\Gamma }/_2=H^1(,_2)`$. The modular functor of the free theory, which is the space of level-$`2p`$ theta functions on the Jacobian of $``$, is an irreducible representation of the Heisenberg group obtained as the central extension of $`H^1(,_{2p})_{2p}^{2g}`$ with the help of $`_{2p}`$, where $`g`$ is the genus of $``$. It is therefore a projective representation of the subgroup $`H^1(,_2)H^1(,_{2p})`$ (with the embedding induced by $`_2_{2p}`$). Therefore, the group $`\mathrm{\Gamma }`$ acts on the tensor product $`\mathrm{𝖬𝖮𝖣}_{\widehat{s\mathrm{}}(2)}(p2,)\mathrm{𝖬𝖮𝖣}_{\text{free}}(p,)`$. Moreover, the center then acts trivially, and the tensor product becomes a representation of the $`H^1(,_2)`$ group. In this group, which is isomorphic to $`_2^{2g}`$, one must choose the maximal $`,`$-isotropic subspace $`V_+`$ and take invariants with respect to $`V_+`$ in the tensor product. The invariants carry the action of the quotient $`_2^{2g}/V_+`$, and one must then take coinvariants with respect to this action. This explains the notation used in (2.17).
### 2.4. Motivation for the semi-infinite construction
The semi-infinite construction of the unitary $`N=2`$ representations can be motivated by the following observations. In the vacuum representation, we consider the decoupling condition for the singular vector $`𝒬_{1p}𝒬_{2p}\mathrm{}𝒬_1|0,p;0`$. The corresponding field
(2.18)
$$^{p2}𝒬(z)\mathrm{}𝒬(z)𝒬(z)$$
(with $`=/z`$ and $`𝒬(z)=_n𝒬_nz^{n1}`$) has vanishing correlation functions with all the fields in the corresponding “minimal” model (cf. ). The identity $`^{p2}𝒬(z)\mathrm{}𝒬(z)𝒬(z)=0`$ therefore holds in unitary representations as an “operator” equality in the sense that it is satisfied when acting on any vector of any unitary representation. Similarly, the “symmetric” relation $`^{p2}𝒢(z)\mathrm{}𝒢(z)𝒢(z)=0`$ also holds for $`𝒢(z)=_n𝒢_nz^{n2}`$. Any of these relations can be chosen as the starting point of the semi-infinite construction. Remarkably, this suffices for reconstructing the entire representation. We choose the $`𝒢`$-relation (which is a matter of convention and/or application of $`N=2`$ algebra automorphisms). Therefore, we define
(2.19)
$$S^p(z)=^{p2}𝒢(z)\mathrm{}𝒢(z)𝒢(z),𝒢(z)=\underset{n}{}𝒢_nz^{n2},$$
and rewrite the condition $`S^p(z)=0`$ in terms of the modes $`𝒢_n`$ as $`S_a^p=0`$ (see (1.1)). For example, for $`p=3`$, the first several relations satisfied on the vacuum vector are given by
(2.20)
$$\begin{array}{c}𝒢_2𝒢_1=0,𝒢_3𝒢_1=0,3𝒢_4𝒢_1+𝒢_3𝒢_2=0,\hfill \\ \hfill 4𝒢_5𝒢_1+2𝒢_4𝒢_2=0,5𝒢_6𝒢_1+3𝒢_5𝒢_2+𝒢_4𝒢_3=0,\mathrm{}.\end{array}$$
In addition, the highest-weight vector $`|r,p;\theta _{\mathrm{irr}}`$ of a given unitary representation $`𝔎_{r,p;\theta }`$ satisfies another vanishing condition, $`𝒢_{\theta r}\mathrm{}𝒢_{\theta 1}|r,p;\theta _{\mathrm{irr}}=0`$, which is the decoupling condition for another singular vector. Next, the state $`𝒢_{\theta p+1}\mathrm{}𝒢_{\theta r1}𝒢_{\theta r+1}\mathrm{}𝒢_{\theta 1}|r,p;\theta _{\mathrm{irr}}`$ satisfies the twisted highest-weight conditions (2.5) with the twist $`\vartheta =\theta p`$. Acting on this state with the operators $`𝒢_{p+\theta p+1}\mathrm{}𝒢_{p+\theta r1}𝒢_{p+\theta r+1}\mathrm{}𝒢_{p+\theta 1}`$, we obtain a twisted highest-weight state with the twist $`\vartheta =\theta 2p`$, etc. Acting on $`|r,p;\theta _{\mathrm{irr}}`$ with the $`𝒬_n`$ modes, we similarly obtain twisted highest-weight states with the twists $`\theta +p`$, $`\theta +2p`$$`\mathrm{}`$ . We label these extremal states as $`|r,p;\theta |\iota `$, where the number $`\iota `$ determines the twist via $`\vartheta =\theta +\iota p`$. Explicitly, the charge–level coordinates of $`|r,p;\theta |\iota `$ are read off from
(2.21) $`_0|r,p;\theta |\iota =`$ $`(\theta {\displaystyle \frac{r1}{p}}+{\displaystyle \frac{1}{2}}{\displaystyle \frac{p2}{p}}(\theta ^2\theta )\iota {\displaystyle \frac{p}{2}}+\iota ^2{\displaystyle \frac{p(p2)}{2}}+\iota r+\iota (p2)\theta )|r,p;\theta |\iota ,`$
(2.22) $`_0|r,p;\theta |\iota =`$ $`({\displaystyle \frac{r1}{p}}{\displaystyle \frac{p2}{p}}\theta (p2)\iota )|r,p;\theta |\iota .`$
These states satisfy, in particular, the conditions
(2.23) $`𝒢_n|r,p;\theta |\iota `$ $`=0\text{for}n\iota p+\theta ,`$
$`𝒢_{\iota p+\theta r}\mathrm{}𝒢_{\iota p+\theta 1}|r,p;\theta |\iota `$ $`=0,`$
$`𝒢_{\iota p+\theta p+1}\mathrm{}𝒢_{\iota p+\theta r1}𝒢_{\iota p+\theta r+1}\mathrm{}𝒢_{\iota p+\theta 1}|r,p;\theta |\iota `$ $`=|r,p;\theta |\iota 1,`$
where $`\iota `$. In the charge–level coordinates on the plane, we represent the states $`\mathrm{}𝒢_3𝒢_2𝒢_1|r,p;0|0`$ with arrows (see Fig. 1) or somewhat more schematically, by replacing several consecutive arrows with sections of parabolas; states (2.23) are then the cusps where the sections of parabolas join (half of the cusps correspond to the mode $`𝒢_{\iota p+\theta r}`$ omitted in (2.23) and the other half are the states $`|r,p;\theta |\iota `$).
The idea behind the semi-infinite construction is to “generate the module from the state $`|r,p;\mathrm{}`$” located infinitely far in the bottom right in Fig. 1, using only the $`𝒢_n`$ modes; this amounts to introducing semi-infinite forms in the fermionic operators $`(𝒢_i)_i`$. We see in what follows that any unitary representation $`𝔎_{r,p;\theta }`$ can be realized via this semi-infinite construction.
## 3. The semi-infinite space $`𝖶_{r,p;\theta }`$
We fix an integer $`p3`$ and also $`r`$ and $`\theta `$ such that $`1rp1`$ and $`0\theta r1`$. We let $`𝔤(p)`$ denote the quotient of the Grassmann algebra of the operators $`𝒢_n`$, $`n`$, over the ideal $`𝒮^p`$ generated by $`(S_a^p)_a`$, see Eq. (1.1) (eventually, we identify $`𝒢_n`$ with the corresponding $`N=2`$ generators).
### 3.1. The inductive limit
The semi-infinite space $`𝖶_{r,p;\theta }`$ is spanned by polynomials in $`𝒢_n𝔤(p)`$, $`n`$, acting on the states
(3.1)
$$(|r,p;\theta |\iota _{\mathrm{}/2})_\iota ,$$
such that
(3.2) $`𝒢_n|r,p;\theta |\iota _{\mathrm{}/2}`$ $`=0,n\iota p+\theta ,`$
(3.3) $`𝒢_{\iota p+\theta r}\mathrm{}𝒢_{\iota p+\theta 1}|r,p;\theta |\iota _{\mathrm{}/2}`$ $`=0,`$
(3.4) $`𝒢_{\iota p+\theta p+1}\mathrm{}𝒢_{\iota p+\theta r1}𝒢_{\iota p+\theta r+1}\mathrm{}𝒢_{\iota p+\theta 1}|r,p;\theta |\iota _{\mathrm{}/2}`$ $`=|r,p;\theta |\iota 1_{\mathrm{}/2}.`$
Therefore, $`𝖶_{r,p;\theta }`$ is the linear span of states of the form $`𝒢_{i_1}𝒢_{i_2}\mathrm{}𝒢_{i_m}|r,p;\theta |\iota _{\mathrm{}/2}`$ considered modulo the ideal $`𝒮^p`$. We call the elements of $`𝖶_{r,p;\theta }`$ the semi-infinite forms; we also, somewhat loosely, use the term “semi-infinite form” for polynomials in $`𝒢_n`$ acting on $`|r,p;\theta |\iota _{\mathrm{}/2}`$, rather than for their images in the quotient. We write $`𝖶_{r,p}`$ instead of $`𝖶_{r,p;0}`$. Obviously, $`𝖶_{r,p;\theta }`$ is a module over $`𝔤(p)`$.
###### Remark 3.1.
In less formal terms, the state $`|r,p;\theta |\iota _{\mathrm{}/2}`$ can be viewed as the semi-infinite product
(3.5)
$$\begin{array}{c}𝒢_{\iota p+\theta +1}\mathrm{}𝒢_{\iota p+\theta +pr1}𝒢_{\iota p+\theta +pr+1}\mathrm{}𝒢_{\iota p+\theta +p1}\times \hfill \\ \hfill \times 𝒢_{\iota p+\theta +p+1}\mathrm{}𝒢_{\iota p+\theta +2pr1}𝒢_{\iota p+\theta +2pr+1}\mathrm{}𝒢_{\iota p+\theta +2p1}\mathrm{},\end{array}$$
where the associated sequence of occupation numbers is periodic with the period $`p`$. The elements of $`𝖶_{r,p;\theta }`$ are then those semi-infinite forms of $`(𝒢_i)_i`$ (modulo $`𝒮^p`$) for which the sequence of occupation numbers $`\alpha (i)`$ becomes periodic with the period $`p`$ for $`i1`$.
We let $`𝖶_{r,p;\theta }(\iota )`$ denote the space generated by $`(𝒢_n)𝔤(p)`$, $`n`$, from the state $`|r,p;\theta |\iota `$ satisfying the same annihilation conditions as those in (3.2) and (3.3). There are the mappings
(3.6)
$$\mathrm{}𝖶_{r,p;\theta }(2)𝖶_{r,p;\theta }(1)𝖶_{r,p;\theta }(0)𝖶_{r,p;\theta }(1)\mathrm{},$$
induced by mapping the vacuum vectors as
(3.7)
$$|r,p;\theta |\iota 1𝒢_{p\iota +\theta p+2}\mathrm{}𝒢_{p\iota +\theta r1}𝒢_{p\iota +\theta r+1}\mathrm{}𝒢_{p\iota +\theta 1}|r,p;\theta |\iota ,\iota .$$
Mappings (3.6) commute with the action of $`𝒢_n`$. The mappings $`𝖶_{r,p;\theta }(\iota )𝖶_{r,p;\theta }`$ induced by $`|r,p;\theta |\iota |r,p;\theta |\iota _{\mathrm{}/2}`$ make all the diagrams
(3.8)
$$\begin{array}{ccccc}\hfill 𝖶_{r,p;\theta }(\iota )& & & & 𝖶_{r,p;\theta }(\iota +1)\hfill \\ & & & & \\ & & 𝖶_{r,p;\theta }& & \end{array}$$
commutative. Therefore, the space $`𝖶_{r,p;\theta }`$ is the inductive limit of (3.6), $`𝖶_{r,p;\theta }=\underset{}{\mathrm{lim}}_\iota \mathrm{}𝖶_{r,p;\theta }(\iota )`$.
In addition to the spectral flow action $`𝖴_p`$ given by $`𝖴_p𝒢_n𝖴_p=𝒢_{n+p}`$, we define
(3.9)
$$𝖴_{\pm p}|r,p;\theta |\iota _{\mathrm{}/2}=|r,p;\theta |\iota \pm 1_{\mathrm{}/2}.$$
This makes the semi-infinite space into a module over $`𝖦^p(p)`$, the semi-direct product of $`𝔤(p)`$ and $`(𝖴_{pn})_n`$.
The formal definition given above does not tell us anything about the structure of the semi-infinite space, in particular, about whether mappings (3.6) are embeddings (and hence, whether the spaces $`𝖶_{r,p;\theta }(\iota )`$ give any “approximation” to the limit). For this to be the case, the vanishings occurring among the polynomials in $`𝒢_n`$ under mappings (3.7) must precisely agree with the vanishings due to taking the quotient over the ideal generated by $`(S_a^p)_a`$. The difficulty in showing this is that the ideal is not described explicitly. For example, Eqs. (2.20) imply additional relations $`𝒢_5𝒢_3𝒢_2=0`$ and $`𝒢_6𝒢_3𝒢_2=0`$; continuing the list in (2.20), we obtain more elements in the ideal. The situation becomes more complicated in the general case (1.1). Nevertheless, we show in what follows that the mappings involved in the inductive limit are embeddings. This implies that the results established for the individual spaces $`𝖶_{r,p;\theta }(\iota )`$ give an “approximation” to those for the entire semi-infinite space $`𝖶_{r,p;\theta }`$.
###### Theorem 3.2.
All the mappings in (3.8) are embeddings, i.e., the space $`𝖶_{r,p;\theta }`$ admits the filtration
(3.10) $`\mathrm{}𝖶_{r,p;\theta }(2)𝖶_{r,p;\theta }(1)𝖶_{r,p;\theta }(0)𝖶_{r,p;\theta }(1)\mathrm{}.`$
This follows from the existence of a remarkable monomial basis in $`𝖶_{r,p;\theta }(\iota )`$ that agrees with the mappings $`𝖶_{r,p;\theta }(\iota )𝖶_{r,p;\theta }(\iota +1)`$, as we discuss momentarily.
### 3.2. The thin basis
In each space $`𝖶_{r,p;\theta }(\iota )`$, we construct a monomial basis, which we call the thin basis because it consists of semi-infinite forms where the modes $`𝒢_n`$ are “thinned out” as explained in Theorem 1.3. We reformulate the desired result as follows (Theorem 1.3 is reproduced by writing $`|r,p;\theta |\iota _{\mathrm{}/2}`$ as the semi-infinite product (3.5)).
###### Lemma 3.3.
The set of states $`𝒢_{i_m}\mathrm{}𝒢_{i_2}𝒢_{i_1}|r,p;\theta |\iota _{\mathrm{}/2}`$, for which $`\iota p+\theta >i_1>i_2>\mathrm{}>i_m`$, $`i_ai_{a+p2}p`$ for any $`a`$, and $`\iota p+\theta i_r>r`$, constitute a basis in the space $`𝖶_{r,p;\theta }(\iota )`$.
We refer to the thin basis elements as thin monomials. Theorem 3.2 now follows from the observation that in terms of thin monomials, the mappings $`𝖶_{r,p;\theta }(\iota )𝖶_{r,p;\theta }(\iota +1)`$ are implemented by
(3.11)
$$\begin{array}{c}𝒢_{i_m}\mathrm{}𝒢_{i_2}𝒢_{i_1}|r,p;\theta |\iota _{\mathrm{}/2}𝒢_{i_m}\mathrm{}𝒢_{i_2}𝒢_{i_1}\times \hfill \\ \hfill \times 𝒢_{\iota p+\theta +1}\mathrm{}𝒢_{\iota p+\theta +pr1}𝒢_{\iota p+\theta +pr+1}\mathrm{}𝒢_{\iota p+\theta +p1}|r,p;\theta |\iota +1_{\mathrm{}/2}.\end{array}$$
Indeed, once the indices $`i_m,\mathrm{},i_1`$ satisfy the conditions of Lemma 3.3, the indices $`i_m,\mathrm{},i_1`$, $`\iota p+\theta +1`$, $`\mathrm{}`$, $`\iota p+\theta +pr1`$, $`\iota p+\theta +pr+1`$, $`\mathrm{}`$, $`\iota p+\theta +p1`$ also satisfy these conditions.
Proof of Lemma 3.3 consists of two parts, the first of which is simple, but the second requires certain effort. We briefly describe the first part and then focus on the second.
The first part of the proof amounts to asserting that any semi-infinite form can be rewritten as a linear combination of the states $`𝒢_{i_m}𝒢_{i_{m1}}\mathrm{}𝒢_{i_1}|r,p;\theta |\iota _{\mathrm{}/2}`$ satisfying the conditions of the Lemma. This follows from the fact that if any monomial in $`𝒢_n`$ contains a combination of modes violating the conditions of the lemma, then in accordance with the relations $`S_a^p=0`$, this monomial can be expressed through a linear combination of other monomials each of which is lexicographically smaller than the original one. The new monomials, obviously, can also involve combinations of modes violating the conditions of the lemma, however the argument regarding the lexicographic ordering allows developing an iteration procedure. It converges after a finite number of steps (and thus gives a linear combination of monomials satisfying the conditions of the Lemma) because
1. the space $`\overline{𝖶}_{r,p;\theta }(\iota )`$ is bigraded via $`𝒢_{i_m}𝒢_{i_{m1}}\mathrm{}𝒢_{i_1}|r,p;\theta |\iota _{\mathrm{}/2}(m,i_1+i_2+\mathrm{}+i_m)`$;
2. each graded component is a finite-dimensional space;
3. the procedure of expressing a monomial through a linear combination of others using the relations $`S_a^p=0`$ preserves the bigrading.
In the second part of the proof, it remains to show that the thin monomials are linearly independent. This follows by comparing the characters. The calculation of characters is rather involved, and we give it in Sec. 4. The proof is completed by the statement of Lemma 4.2 in Sec. 4.3.
#### Examples
We give several examples of the transformation to the thin basis.
#### A. $`p=3`$, $`r=1`$
Directly eliminating dense combinations of modes leads to $`𝒢_{15}𝒢_{14}𝒢_{12}𝒢_{13}𝒢_4𝒢_3|1,3;0|0=3393𝒢_{21}|1,3;0|5+4185𝒢_{20}𝒢_{15}|1,3;0|4+5697𝒢_{19}𝒢_{16}|1,3;0|4+1755𝒢_{19}𝒢_{15}𝒢_{12}|1,3;0|3+315𝒢_{18}𝒢_{15}𝒢_{12}𝒢_9|1,3;0|2`$.
#### B. $`p=4`$, $`r=1`$
As a simple example, we have $`𝒢_5𝒢_4𝒢_2|1,4;0|0=2𝒢_6|1,4;0|1`$. Similarly, $`𝒢_5𝒢_4𝒢_3|1,4;0|0=10𝒢_7|1,4;0|18𝒢_6𝒢_4𝒢_2|1,4;0|0`$.
#### C. $`p=5`$, $`r=1`$
Of the following two examples, the second is an extension of the first: $`𝒢_2𝒢_1𝒢_0𝒢_1|1,5;1|1=35𝒢_5|1,5;1|045𝒢_4𝒢_1𝒢_1𝒢_2|1,5;1|120𝒢_3𝒢_2𝒢_1𝒢_2|1,5;1|115𝒢_3𝒢_1𝒢_0𝒢_2|1,5;1|1`$ and $`𝒢_5𝒢_4𝒢_2𝒢_1𝒢_0𝒢_1|1,5;1|1=45𝒢_7𝒢_4𝒢_3|1,5;1|0(245/3)𝒢_6𝒢_5𝒢_3|1,5;1|075𝒢_6𝒢_4𝒢_3𝒢_1𝒢_1𝒢_2|1,5;1|1`$.
## 4. Functional realization and characters
We use the realization of the (graded-)dual space to the semi-infinite space in terms of polynomial differential forms. This functional realization is a powerful tool in studying properties of semi-infinite spaces.
### 4.1. The functional realization of $`𝖶_{r,p}(0)^{}`$
We first note that for any space generated from a vacuum $`|`$ by fermionic generators $`𝒢_1`$, the graded-dual space can be identified with differential forms in some variables $`x_1,x_2,\mathrm{}`$ as follows. In each graded component with a given charge (the number of $`𝒢_{}`$) $`n1`$, we arrange all the states into a generating function $`𝒢(x_1)\mathrm{}𝒢(x_n)|`$, where
(4.1)
$$𝒢(x)=\underset{m1}{}𝒢_mx^{m1}.$$
The states $`𝒢_{i_1}\mathrm{}𝒢_{i_n}|`$ are reproduced by taking the integrals
(4.2)
$$dx_1x_1^{i_1}\mathrm{}dx_nx_n^{i_n}𝒢(x_1)\mathrm{}𝒢(x_n)|.$$
Any functional $`\mathrm{}|`$ on the charge-$`n`$ subspace is determined by all of its values,<sup>7</sup><sup>7</sup>7Whether $`𝒢(x)`$ is viewed as a 1-differential or, for example, a 2-differential, is a matter of convention; however, this convention must agree with the choice of the functional spaces involved in the functional realization. With the choice made in the text, all functional spaces are polynomials $`[x_1,\mathrm{},x_n]`$, rather than $`(x_1\mathrm{}x_n)^\nu [x_1,\mathrm{},x_n]`$ with some positive or negative $`\nu `$.
(4.3)
$$\mathrm{}|𝒢(x_1)\mathrm{}𝒢(x_n)|=f_{\mathrm{}}(x_1,\mathrm{},x_n)dx_1\mathrm{}dx_n,$$
where $`f_{\mathrm{}}`$ is an antisymmetric polynomial in $`x_1,\mathrm{},x_n`$ (and the product of the differentials is symmetric). Therefore, the space dual to $`W(0)`$, the space freely generated by the modes $`𝒢_{n1}`$, can be identified with polynomial differential forms in $`x_1,x_2,\mathrm{}`$,
(4.4)
$$W(0)^{}=[x]dxx_1,x_2dx_1dx_1x_1,x_2,x_3dx_1dx_2dx_3\mathrm{},$$
where $`x_1,\mathrm{},x_n`$ are antisymmetric polynomials.
For quotient spaces, the functional realization of the corresponding dual space is given by a subspace in the space of polynomials. Let $`𝖶(0)_n`$ denote the charge $`n`$ subspace and $`𝖶(0)_n^{}`$ its dual, i.e., the subspace of polynomials in $`n`$ variables in (4.4). In the dual language, taking the quotient with respect to the ideal generated by $`S_a^p`$ (see (1.1)) corresponds to restricting to those antisymmetric polynomials $`f(x_1,\mathrm{},x_n)`$ for which
(4.5)
$$\frac{d^{p2}}{x_{i_1}^{p2}}\frac{^{p3}}{x_{i_2}^{p3}}\mathrm{}\frac{}{x_{i_{p2}}}f(x_1,x_2,\mathrm{})|_{x_{i_1}=x_{i_2}=\mathrm{}=x_{i_{p2}}=x_{i_{p1}}}=0.$$
Another condition is read off from Eq. (3.3) (for $`\theta =\iota =0`$) as
(4.6)
$$\frac{^{r1}}{x_{i_1}^{r1}}\frac{^{r2}}{x_{i_2}^{r2}}\mathrm{}\frac{}{x_{i_{r1}}}f(x_1,x_2,\mathrm{})|_{x_{i_1}=x_{i_2}=\mathrm{}=x_{i_{r1}}=x_{i_r}=0}=0.$$
Let $`𝖶_{r,p}(0)^{}`$ denote the space of antisymmetric polynomials $`f`$ satisfying these conditions.
Because any antisymmetric polynomial can be represented as
(4.7)
$$f(x_1,x_2,\mathrm{})=\mathrm{\Delta }(x_1,x_2,\mathrm{})\varphi (x_1,x_2,\mathrm{}),\mathrm{\Delta }(x_1,x_2,\mathrm{})=\underset{1i<j}{}(x_ix_j),$$
with a symmetric polynomial $`\varphi (x_1,x_2,\mathrm{})`$, conditions (4.6) and (4.5) become the following restrictions on symmetric polynomials:
1. $`\varphi (\underset{r}{\underset{}{0,\mathrm{},0}},x_{r+1},\mathrm{},x_n)=0`$;
2. $`\varphi (\underset{p1}{\underset{}{x,x,\mathrm{},x}},x_p,x_{p+1},\mathrm{},x_n)=0`$.
#### Example
For $`p=3`$, condition P$`\mathrm{𝟐}`$ can be easily solved as
(4.8)
$$f(x_1,\mathrm{},x_n)dx_1\mathrm{}dx_n=\underset{i<j}{}(x_ix_j)^3\phi (x_1,\mathrm{},x_n)dx_1\mathrm{}dx_n,$$
where $`\phi `$ is a symmetric polynomial. For $`r=1`$, condition P$`\mathrm{𝟏}`$ is immediately solved by $`\phi (x_1,\mathrm{},x_n)=x_1\mathrm{}x_n\overline{\phi }(x_1,x_2,\mathrm{},x_n)`$, whence it follows that the space $`𝖶_{1,3}(0)_n^{}`$ consists of the differential forms
(4.9) $`{\displaystyle \underset{i<j}{}}(x_ix_j)^3x_1\mathrm{}x_n\overline{\phi }(x_1,x_2,\mathrm{},x_n)dx_1\mathrm{}dx_n,`$
where $`\overline{\phi }`$ are arbitrary symmetric polynomials. Such an explicit description, however, is not available for $`p4`$.
### 4.2. Characters of $`𝖶_{r,p;\theta }(\iota )`$
The functional realization allows us to calculate the characters of $`𝖶_{r,p;\theta }(\iota )`$. The idea of consists in finding, for each $`n1`$, the character of the space $`𝖶_{r,p}(0)_n^{}=𝖶_{r,p}(0)^{}x_1,\mathrm{},x_n`$, which coincides with the character of $`𝖶_{r,p}(0)_n`$ (the subspace in $`𝖶_{r,p}(0)`$ generated by $`n`$ modes $`𝒢_i`$); then
(4.10)
$$char𝖶_{r,p}(0)(z,q)=\underset{n0}{}z^nchar𝖶_{r,p}(0)_n(q).$$
The dependence on both $`\iota `$ and $`\theta `$ can be reconstructed in accordance with the spectral flow.
#### 4.2.1. Example: $`p=3`$
The functional model of $`𝖶_{1,3}(0)_n^{}`$ is given by (4.9), and the space of symmetric polynomials in $`n`$ variables is algebraically generated by the elementary symmetric polynomials; the contribution of these polynomials to the character of $`𝖶_{1,3}(0)_n^{}`$ is therefore $`1/(q)_n`$. Next, the product $`_{i<j}(x_ix_j)^3`$ contributes $`q^{3n(n1)/2}`$ to the character, the differentials $`dx_1\mathrm{}dx_n`$ contribute $`q^n`$, and another $`q^n`$ comes from $`x_1\mathrm{}x_n`$. Therefore,
(4.11)
$$char𝖶_{1,3}(0)=char𝖶_{1,3}(0)^{}=\underset{n0}{}\frac{z^nq^{\frac{3n^2+n}{2}}}{(q)_n}.$$
Applying the spectral flow transform in accordance with (2.4), we find
(4.12)
$$\begin{array}{cc}\hfill char𝖶_{1,3}(\iota )(z,q)& =z^\iota q^{\frac{\iota }{2}(3\iota 1)}char𝖶_{1,3}(0)(zq^{3\iota },q)=\hfill \\ & =\underset{n0}{}\frac{z^{n\iota }q^{\frac{3(n\iota )^2+n\iota }{2}}}{(q)_n}=\underset{n\iota }{}\frac{z^nq^{\frac{3n^2+n}{2}}}{(q)_{n+\iota }}.\hfill \end{array}$$
This expression admits the limit
(4.13)
$$\underset{\iota \mathrm{}}{lim}char𝖶_{1,3}(\iota )(z,q)=\underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}\frac{z^nq^{\frac{3n^2+n}{2}}}{_{m1}(1q^m)}=q^{\frac{1}{3}}\frac{\vartheta _{1,0}(zq^1,q^3)}{\eta (q)},$$
which coincides with (2.10). Much more work is needed to show this remarkable coincidence in the general case.
#### 4.2.2. The general case: $`p4`$
###### Lemma 4.1.
For $`p3`$ and $`1rp1`$, the character of the space $`𝖶_{r,p}(0)`$ is given by
(4.14)
$$char𝖶_{r,p}(0)(z,q)=\underset{n0}{}\underset{\begin{array}{c}N_1\mathrm{}N_{p2}0\\ N_1+\mathrm{}+N_{p2}=n\end{array}}{}\frac{z^{n\frac{r1}{p}}q^{_{m=r}^{p2}N_m}q^{\frac{1}{2}n(n1)}q^{_{m=1}^{p2}N_m^2}}{(q)_{N_1N_2}(q)_{N_2N_3}\mathrm{}(q)_{N_{p3}N_{p2}}(q)_{N_{p2}}}.$$
Proof. In the functional realization of $`𝖶_{r,p}(0)_n^{}`$, we consider all the partitions of the set $`\{x_1,\mathrm{},x_n\}`$ and introduce the lexicographic ordering on the partition (i.e., we set $`(r_1,r_2,\mathrm{},r_k)(r_1^{},r_2^{},\mathrm{},r_k^{}^{})`$ if $`r_i<r_i^{}`$ for the first pair $`(r_i,r_i^{})`$ such that $`r_ir_i^{}`$).
We fix a partition of $`n`$ written as $`\{1,\mathrm{},n\}=M_1\mathrm{}M_{\mathrm{}}`$ and let $`|M_\alpha |=r_\alpha `$. We say that a polynomial $`\phi (x_1,\mathrm{},x_n)`$ vanishes on this partition if it vanishes whenever all the variables $`x_{i_\alpha }`$ for $`i_\alpha M_\alpha `$ take the same value, $`x_{i_\alpha }=a_\alpha `$ for all $`\alpha =1,\mathrm{},\mathrm{}`$. We write this as $`\phi (a_1;\mathrm{};a_{\mathrm{}})=0`$.
The character of $`𝖶_{r,p}(0)^{}`$ can be written as
(4.15)
$$char𝖶_{r,p}(0)(z,q)=z^{\frac{r1}{p}}\underset{n0}{}z^nq^{\frac{n^2n}{2}}Z_{r,p}^n(q),$$
where the factor $`q^{(n^2n)/2}`$ corresponds to $`\mathrm{\Delta }(x_1,\mathrm{},x_n)`$ in (4.7) and $`Z_{r,p}^n(q)`$ is the partition function of symmetric polynomials in $`n`$ variables that vanish on the partition $`\widehat{p}=(p1,1,1,\mathrm{},1)`$.
For a partition $`P`$, we consider the set of symmetric polynomials that vanish on every partition $`P^{}P`$. The lexicographic order on partitions then induces a filtration on symmetric polynomials in $`n`$ variables. If $`Z_{r,p}^{n,P}(q)`$ is the partition function of the associated graded factor $`𝖦𝗋_P`$, we have
(4.16)
$$Z_{r,p}^n(q)=\underset{\{PP\widehat{p}\}}{}Z_{r,p}^{n,P}(q).$$
To find $`Z_{r,p}^{n,P}(q)`$, we first consider the case where $`r=p1`$; condition P$`\mathrm{𝟏}`$ is then a consequence of P$`\mathrm{𝟐}`$. For a partition $`P`$ with the parts $`M_1`$, …, $`M_{\mathrm{}}`$ (and with $`|M_\alpha |=r_\alpha `$), the graded factor $`𝖦𝗋_P`$ is spanned by polynomials $`\phi (a)\phi (a_1;\mathrm{};a_{\mathrm{}})`$ satisfying the conditions
$`\phi (a)=0`$ if $`a_\alpha =a_\beta `$, with the multiplicity of the zero equal to $`2\mathrm{min}(r_\alpha ,r_\beta )`$,
$`\phi (\mathrm{};a_\alpha ;\mathrm{};a_\beta ;\mathrm{})=\phi (\mathrm{};a_\beta ;\mathrm{};a_\alpha ;\mathrm{})`$ whenever $`r_\alpha =r_\beta `$.
Recalling the differentials from (4.3), we can therefore see that the partition function $`Z_{p1,p}^{n,P}(q)`$ of the graded factor coincides with the partition function of the space spanned by
(4.17)
$$\underset{1\alpha \beta \mathrm{}}{}(a_\alpha a_\beta )^{2r_\beta }\overline{\phi }(a)\underset{\alpha =1}{\overset{\mathrm{}}{}}(da_\alpha )^{r_\alpha },$$
where a polynomial $`\overline{\phi }`$ satisfies symmetry requirement (sym). Let $`\nu _m`$ be the number of parts $`M_\alpha `$ such that $`r_\alpha =m`$. The contribution of the chosen partition $`P`$ to the partition function is then
(4.18)
$$\frac{q^{_{\alpha <\beta }2r_\beta +_\alpha r_\alpha }}{_{j1}(q)_{\nu _j}}=\frac{q^{_{\beta =1}^{\mathrm{}}(2\beta 1)r_\beta }}{_{j1}(q)_{\nu _j}}=\frac{q^{_{m=1}^{p2}N_m^2}}{(q)_{N_1N_2}(q)_{N_2N_3}\mathrm{}},$$
where $`N_m=\nu _m+\nu _{m+1}+\mathrm{}+\nu _{\mathrm{}}`$ are the elements of the partition $`P`$ transposed. Therefore,
(4.19)
$$Z_{p1,p}^n(q)=\underset{\begin{array}{c}N_1\mathrm{}N_{p2}0\\ N_1+\mathrm{}+N_{p2}=n\end{array}}{}\frac{q^{_{m=1}^{p2}N_m^2}}{(q)_{N_1N_2}(q)_{N_2N_3}\mathrm{}(q)_{N_{p2}}}.$$
If $`rp1`$, we must additionally account for condition P$`\mathrm{𝟏}`$, which amounts to taking symmetric polynomials $`\overline{\phi }(a)`$ satisfying $`\overline{\phi }(a)|_{a_1=a_2=\mathrm{}=a_r=0}=0`$. For this, we recall the filtration $`𝒫_1^n\mathrm{}𝒫_{r1}^n𝒫_r^n`$ on the space $`𝒫_r^n`$ of symmetric polynomials in $`n`$ variables satisfying condition P$`\mathrm{𝟏}`$,
(4.20)
$$𝒫_r^n=\sigma _n[\sigma _1,\mathrm{},\sigma _n]+\sigma _{n1}[\sigma _1,\mathrm{},\sigma _{n1}]+\mathrm{}+\sigma _{nr+1}[\sigma _1,\mathrm{},\sigma _{nr+1}],$$
where $`\sigma _1=x_1+\mathrm{}+x_n`$, $`\mathrm{}`$, $`\sigma _n=x_1\mathrm{}x_n`$ are the elementary symmetric polynomials in $`n`$ variables. In the corresponding graded factor, the explicit $`\sigma _i`$ factors result in additionally multiplying partition function (4.18) by $`q^{_{m=r}^{p2}N_m}`$. Thus,
(4.21)
$$Z_{r,p}^n(q)=\underset{\begin{array}{c}N_1\mathrm{}N_{p2}0\\ N_1+\mathrm{}+N_{p2}=n\end{array}}{}\frac{q^{_{m=r}^{p2}N_m}q^{_{m=1}^{p2}N_m^2}}{(q)_{N_1N_2}(q)_{N_2N_3}\mathrm{}(q)_{N_{p3}N_{p2}}(q)_{N_{p2}}}.$$
Inserting (4.21) in (4.15), we obtain (4.14).
Using relation (2.4) with $`\theta =p\iota `$, we find the characters of the spectral-flow transformed spaces as
(4.22)
$$\begin{array}{c}char𝖶_{r,p}(\iota )(z,q)=z^{(p2)\iota }q^{\frac{p2}{2}(p\iota ^2\iota )}char𝖶_{r,p}(0)(zq^{p\iota },q)\hfill \\ \hfill =q^{\frac{p2}{2}(p\iota ^2\iota )+(r1)\iota }\underset{n(p2)\iota }{}\underset{\begin{array}{c}N_1\mathrm{}N_{p2}0\\ N_1+\mathrm{}+N_{p2}=n+(p2)\iota \end{array}}{}q^{p\iota [n+(p2)\iota ]}z^{n\frac{r1}{p}}\times \\ \hfill \times q^{\frac{1}{2}_{m=r}^{p2}N_m\frac{1}{2}_{m=1}^{r1}N_m}\frac{q^{\frac{1}{2}(n^2+2n(p2)\iota +(p2)^2\iota ^2)}q^{_{m=1}^{p2}N_m^2}}{(q)_{N_1N_2}(q)_{N_2N_3}\mathrm{}(q)_{N_{p3}N_{p2}}(q)_{N_{p2}}}\\ \hfill =z^{\frac{1r}{p}}\underset{n(p2)\iota }{}\underset{\begin{array}{c}N_1\mathrm{}N_{p2}\iota \\ N_1+\mathrm{}+N_{p2}=n\end{array}}{}\frac{z^nq^{\frac{1}{2}n^2}q^{_{m=1}^{p2}N_m^2+_{m=r}^{p2}N_m+\frac{1}{2}_{m=r}^{p2}N_m\frac{1}{2}_{m=1}^{r1}N_m}}{(q)_{N_1N_2}(q)_{N_2N_3}\mathrm{}(q)_{N_{p3}N_{p2}}(q)_{N_{p2}+\iota }}.\end{array}$$
### 4.3. Linear independence of thin monomials
We now show that the thin basis defined in Sec. 3.2 is indeed a basis in $`𝖶_{r,p}(0)`$ (and hence, by the spectral flow, also a basis in $`𝖶_{r,p;\theta }(\iota )`$); in other words, we show that thin monomials defined in Lemma 3.3 are linearly independent in $`𝖶_{r,p}(0)`$. We actually prove that the partition function of thin monomials coincides with (4.14). For this, we represent each semi-infinite form in terms of the occupation numbers and hence as configurations of crosses on the semi-infinite one-dimensional lattice (see (1.10)). We are interested in the partition function of the configurations satisfying the basic condition that any $`p`$ consecutive lattice sites carry at most $`p2`$ crosses.
We first consider the same restriction on configurations on a lattice with a finite number of sites $`L`$, with the sites labeled by $`1iL`$. Each configuration of crosses at the sites $`(i_1,\mathrm{},i_m)`$ contributes $`z^mq^{i_1+\mathrm{}+i_m}`$ to the partition function. Let $`\omega _{L,r,p}(z,q)`$ be the partition function of all the configurations of crosses on $`L`$ sites satisfying the conditions that
1. there are no more than $`r1`$ crosses at the sites $`1`$, …, $`r`$;
2. on any $`p`$ consecutive sites, there are no more than $`p2`$ crosses.
For $`r=p1`$, the first condition follows from the second. The configurations of crosses encoding the thin basis elements are then recovered in the $`L\mathrm{}`$ limit. Remarkably, we have the following lemma.
###### Lemma 4.2.
The partition function of the configurations of crosses satisfying conditions C$`\mathrm{𝟏}`$ and C$`\mathrm{𝟐}`$ on a semi-infinite lattice is equal to the character $`char𝖶_{r,p}(0)(z,q)`$ defined in Lemma 4.1,
(4.23)
$$z^{\frac{r1}{p}}\underset{L\mathrm{}}{lim}\omega _{L,r,p}(z,q)=char𝖶_{r,p}(0)(z,q).$$
Proof. The idea of the proof is as follows. Expanding both sides of (4.23) in powers of $`z`$ gives two systems of functions, each of which is completely determined by a set of recursive relations and the appropriate “initial values.” These two sets of recursive relations and the initial values are identical, which implies (4.23). We now proceed with the details.
The partition function of the configurations of crosses satisfies the recursive relations
(4.24)
$$\omega _{L,r,p}(z,q)=\underset{j=0}{\overset{r1}{}}z^jq^{\frac{j(j+1)}{2}}\omega _{Lj1,pj1,p}(zq^{j+1},q).$$
Indeed, condition C$`\mathrm{𝟏}`$ selects those configurations that are the disjoint union $`_{j=0}^{r1}`$ of configurations with exactly $`j`$ occupied sites (and the site $`j+1`$ free). For each $`j`$, cutting off these occupied sites and the adjacent free site leaves configurations of crosses on $`Lj1`$ sites; these configurations satisfy a “boundary condition” that keeps track of the crosses at the sites $`1,\mathrm{},j`$ on the original lattice: by C$`\mathrm{𝟐}`$, there can be no more than $`pj2`$ crosses in the beginning of the sublattice. The overall factor $`z^jq^{j(j+1)/2}`$ in (4.24) is precisely the contribution of the crosses at $`1,\mathrm{},j`$, and the “spectral flow” transformation $`zzq^{j+1}`$ in the argument accounts for relabeling the sites on the sublattice. This shows (4.24).
The initial conditions for the recursive are set on lattices with $`Lp2`$, where condition C$`\mathrm{𝟐}`$ does not apply, and therefore,
(4.25)
$$\omega _{L,r,p}(z,q)=\chi _{L1,p}(zq,q)+\underset{\begin{array}{c}j=1\\ Lj10\end{array}}{\overset{r1}{}}z^jq^{\frac{j(j+1)}{2}}\chi _{Lj1,p}(zq^{j+1},q),$$
where $`Lp2`$ and
(4.26)
$$\chi _{\mathrm{},p}(z,q)=\underset{i=0}{\overset{\mathrm{}}{}}z^iq^{\frac{i(i+1)}{2}}\left[\begin{array}{c}\mathrm{}\\ i\end{array}\right]_q=(qz)_{\mathrm{}}\text{for}0<\mathrm{}p2$$
(and $`\chi _{0,p}(z,q)=1`$). With these initial conditions, recursive relations (4.24) completely determine $`\omega _{L,r,p}(z,q)`$.
In the limit as $`L\mathrm{}`$, we expand the partition function in powers of $`z`$ as
(4.27)
$$\underset{L\mathrm{}}{lim}\omega _{L,r,p}(z,q)=\underset{n0}{}z^nq^{\frac{n(n1)}{2}}B_{r,p}^n(q).$$
It then follows from (4.24) that
(4.28)
$$B_{r,p}^n(q)=\underset{j=0}{\overset{r1}{}}q^nB_{pj1,p}^{nj}(q)$$
(for $`n<r1`$, the summation on the right-hand side goes from 0 to $`n`$; it is convenient to set $`B_{r,p}^n(q)=0`$ for $`n<0`$, which allows Eq. (4.28) to be used in all cases). These can be rewritten as recursive relations expressing $`B_{r,p}^n(q)`$ through $`B_{r,p}^m(q)`$ with $`m<n`$. The initial values for these recursive relations are given by $`B_{r,p}^m`$ for $`m<p1`$, where condition C$`\mathrm{𝟐}`$ does not apply. We already saw in (4.25) and (4.26) how to evaluate $`\omega _{L,r,p}(z,q)`$ on a lattice where condition C$`\mathrm{𝟐}`$ is not imposed, and it only remains to take the $`L\mathrm{}`$ limit of (4.25). Let $`𝖯`$ denote the projector on the space of polynomials in $`z`$ of the order $`<p1`$. We then have
(4.29)
$$\begin{array}{c}\underset{m=0}{\overset{p2}{}}z^mq^{\frac{m(m1)}{2}}B_{r,p}^m(q)=𝖯\underset{j=0}{\overset{r1}{}}z^jq^{\frac{j(j+1)}{2}}(zq^{j+2})_{\mathrm{}}=\hfill \\ \hfill =𝖯\underset{j=0}{\overset{r1}{}}z^jq^{\frac{j(j+1)}{2}}\left(1+\underset{n1}{}\frac{(zq^{j+2})^nq^{\frac{n(n1)}{2}}}{(1q)\mathrm{}(1q^n)}\right)=\\ \hfill =𝖯\underset{n0}{}z^nq^{\frac{n(n1)}{2}}\underset{\begin{array}{c}j=0\\ jn\end{array}}{\overset{r1}{}}\frac{q^{2nj}}{(1q)\mathrm{}(1q^{nj})}.\end{array}$$
The inner sum in the right-hand side therefore defines the initial values $`B_{r,p}^n(q)`$ for $`np2`$.
Turning to the character of $`𝖶_{r,p}(0)(z,q)`$, we recall expansion (4.15), where, as we have seen,
(4.30)
$$Z_{r,p}^n(q)=\underset{\begin{array}{c}n_1,\mathrm{},n_{p2}0\\ N_1+\mathrm{}+N_{p2}=n\end{array}}{}\frac{q^{N_1^2+\mathrm{}+N_{p2}^2+N_r+\mathrm{}+N_{p2}}}{(q)_{n_1}\mathrm{}(q)_{n_{p2}}}$$
with $`N_i=n_i+\mathrm{}+n_{p2}`$; these expressions satisfy the relation
(4.31)
$$Z_{r,p}^n(q)=\underset{j=0}{\overset{r1}{}}q^nZ_{pj1,p}^{nj}(q).$$
Indeed, using the notation in , Eq. (4.30) becomes $`Z_{i,p}^n(q)=R_{p1,i}^n(q)`$, where the generating functions $`_{n0}x^nR_{\kappa ,i}^n(q)=R_{\kappa ,i}(x;q)=J_{\kappa ,i}(0;x;q)`$ are known to satisfy the identities
(4.32) $`R_{\kappa ,i}(x;q)R_{\kappa ,i1}(x;q)=(xq)^{i1}R_{\kappa ,\kappa i+1}(xq;q)\text{for}1i\kappa ,`$
(4.33) $`R_{\kappa ,0}(x;q)=0.`$
Summing these relations (with $`\kappa =p1`$) over $`i=1,\mathrm{},r1`$, we obtain (4.31).
To complete the proof, it remains to find the initial values for this recursive, namely, $`Z_{r,p}^m(q)`$ for $`m<\kappa `$. We have
(4.34)
$$R_{\kappa ,i}(x;q)=\underset{n0}{}\frac{x^{\kappa n}q^{\kappa n+\kappa n^2+nin}(1x^iq^{2ni+i})(1)^nq^{\frac{n(n1)}{2}}}{(q)_n(xq^{n+1})_{\mathrm{}}}.$$
The coefficients entering the expansion in powers of $`x^j`$ for $`j\kappa 1`$ follow only from the term with $`n=0`$ in this sum, and therefore,
(4.35)
$$\begin{array}{c}\underset{n=0}{\overset{\kappa 1}{}}x^nR_{\kappa ,i}^n(q)=𝖯\frac{1x^iq^i}{(xq)_{\mathrm{}}}=𝖯\frac{1x^iq^i}{(1xq)}\frac{1}{(1xq^2)(1xq^3)\mathrm{}}=\hfill \\ \hfill =𝖯\underset{m=0}{\overset{i1}{}}x^mq^m\left(1+\underset{n1}{}\frac{q^{2n}x^n}{(1q)\mathrm{}(1q^n)}\right)=\\ \hfill =𝖯\underset{n0}{}x^n\underset{\begin{array}{c}j=0\\ jn\end{array}}{\overset{i1}{}}\frac{q^{2nj}}{(1q)\mathrm{}(1q^{nj})},\end{array}$$
where the inner sum on the right-hand side determines $`R_{\kappa ,i}^nZ_{i,\kappa +1}^n`$ for $`n<\kappa `$. These are seen to be the same as $`B_{i,\kappa }^n`$ (obviously, there is no dependence on $`\kappa =p1`$ because it arises only in the higher $`B_{r,\kappa }^n`$ and $`Z_{r,\kappa }^n`$ due to conditions C$`\mathrm{𝟐}`$ and P$`\mathrm{𝟐}`$).
We can therefore see that $`Z_{r,p}^m(q)`$ satisfy $`Z_{r,p}^m(q)=B_{r,p}^m(q)`$ for $`m<p`$; by the recursive relations, this implies that $`Z_{r,p}^n(q)=B_{r,p}^n(q)`$ for all $`n`$ and hence Eq. (4.23).
The coincidence of the characters shows that there are no linear relations among thin monomials. This completes the proof of Lemma 3.3 and hence of Theorem 3.2.
### 4.4. The character of the semi-infinite space
Theorem 3.2 has an important corollary.
###### Theorem 4.3.
The character of the semi-infinite space $`𝖶_{r,p;\theta }`$ is given by
(4.36)
$$\begin{array}{c}char𝖶_{r,p;\theta }(z,q)=z^{\frac{1r+2\theta }{p}\theta }q^{(1\frac{2}{p})\frac{\theta ^2\theta }{2}\theta \frac{1r}{p}}\times \hfill \\ \hfill \times \underset{N_1\mathrm{}N_{p2}}{}\frac{z^{_{m=1}^{p2}N_m}}{(q)_{N_1N_2}(q)_{N_2N_3}\mathrm{}(q)_{N_{p3}N_{p2}}(q)_{\mathrm{}}}\times \\ \hfill \times q^{\frac{1}{2}\left(_{m=r}^{p2}_{m=1}^{r1}\right)N_m\theta _{m=1}^{p2}N_m+\frac{3}{2}_{m=1}^{p2}N_m^2+_{1m<m^{}p2}N_mN_m^{}}.\end{array}$$
Proof. It is a direct consequence of Theorem 3.2 that
(4.37)
$$\begin{array}{c}char𝖶_{r,p;\theta }(z,q)=\underset{\iota \mathrm{}}{lim}char𝖶_{r,p;\theta }(\iota )(z,q)=\hfill \\ \hfill =\underset{\iota \mathrm{}}{lim}z^{(p2)\iota }q^{\frac{p2}{2}(p\iota ^2\iota )}char𝖶_{r,p;\theta }(0)(zq^{p\iota },q),\end{array}$$
where we applied the spectral flow transform formula (2.4) in the last equality. Next, a remarkable property of Eq. (4.22) is that it has a well-defined limit as $`\iota \mathrm{}`$,
(4.38)
$$\begin{array}{cc}\hfill \underset{\iota \mathrm{}}{lim}char𝖶_{r,p;\theta }(\iota )(z,q)=& z^{\frac{1r+2\theta }{p}\theta }q^{(1\frac{2}{p})\frac{\theta ^2\theta }{2}\theta \frac{1r}{p}}\times \hfill \\ & \times \underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}\underset{\begin{array}{c}N_1\mathrm{}N_{p2}\\ N_1+\mathrm{}+N_{p2}=n\end{array}}{}\frac{z^nq^{\frac{1}{2}n^2\theta n+_{m=1}^{p2}N_m^2+\frac{1}{2}_{m=r}^{p2}N_m\frac{1}{2}_{m=1}^{r1}N_m}}{(q)_{N_1N_2}(q)_{N_2N_3}\mathrm{}(q)_{N_{p3}N_{p2}}(q)_{\mathrm{}}}\hfill \end{array}$$
(where the $`\theta `$ dependence is reconstructed in accordance with the spectral flow).
###### Remark 4.4.
The existence of the thin basis has the following implication for the construction of antisymmetric polynomials satisfying the basic conditions (4.5). Let $`x_1,\mathrm{},x_n^{(p)}`$ be polynomials in $`x_1,\mathrm{},x_n`$ satisfying (4.5). There is an associative multiplication on antisymmetric polynomials: for $`fx_1,\mathrm{},x_n`$ and $`gx_1,\mathrm{},x_n^{}`$, we define $`fgx_1,\mathrm{},x_{n+n^{}}`$ as
(4.39)
$$(fg)(x_1,\mathrm{},x_{n+n^{}})=\underset{x_1,\mathrm{},x_{n+n^{}}}{Alt}\left(f(x_1,\mathrm{},x_n)g(x_{n+1},\mathrm{},x_{n+n^{}})\underset{\begin{array}{c}1in\\ n+1i^{}n+n^{}\end{array}}{}(x_ix_i^{})\right),$$
where $`Alt`$ means alternation. For $`n=2`$ and $`n^{}=1`$, for example,
(4.40)
$$\begin{array}{c}(fg)(x_1,x_2,x_3)=f(x_1,x_2)g(x_3)(x_1x_3)(x_2x_3)+\hfill \\ \hfill +f(x_2,x_3)g(x_1)(x_2x_1)(x_3x_1)f(x_1,x_3)g(x_2)(x_1x_2)(x_3x_2).\end{array}$$
Now, if $`f_1x_1,\mathrm{},x_{n_1}^{(p_1)}`$ and $`f_2x_1,\mathrm{},x_{n_2}^{(p_2)}`$, it is easy to see that the polynomial $`f_1f_2`$ is in $`x_1,\mathrm{},x_{n_1+n_2}^{(p_1+p_22)}`$, i.e.,
(4.41)
$$:x_1,\mathrm{},x_{n_1}^{(p_1)}\times x_1,\mathrm{},x_{n_2}^{(p_2)}x_1,\mathrm{},x_{n_1+n_2}^{(p_1+p_22)}.$$
As a consequence of the thin-basis lemma, we have a stronger statement: the mapping
(4.42)
$$:\underset{\begin{array}{c}n_1,n_21\\ n_1+n_2=n\end{array}}{}x_1,\mathrm{},x_{n_1}^{(p_1)}\times x_1,\mathrm{},x_{n_2}^{(p_2)}x_1,\mathrm{},x_n^{(p_1+p_22)}$$
is an epimorphism. This construction is useful because for $`p=3`$, the polynomials satisfying (4.5) are known explicitly (see (4.9)).
### 4.5. Semi-infinite construction for the $`\widehat{s\mathrm{}}(2)`$ algebra
. The level-$`k`$ $`\widehat{s\mathrm{}}(2)`$ algebra
(4.43) $`[h_i,e_n]`$ $`=e_{n+i},`$ $`[h_i,f_n]`$ $`=f_{n+i},`$
$`[e_i,f_j]`$ $`=2h_{i+j}+ki\delta _{i+j,0},`$ $`[h_i,h_j]`$ $`={\displaystyle \frac{1}{2}}ki\delta _{i+j,0}`$
admits a semi-infinite realization of any unitary representation $`𝔏_{r,k}`$ (where $`1rk+1`$). The key elements of the construction are the semi-infinite forms in commuting elements $`(f_n)_n`$ satisfying the constraints following from the conditions
(4.44)
$$f(z)^{k+1}=0\text{for}f(z)=\underset{n}{}f_nz^{n1}.$$
The semi-infinite space is generated by $`(f_n)_n`$ from the vectors $`|r,k|\iota _{\widehat{s\mathrm{}}(2)}`$ such that $`f_{2\iota +i}|r,k|\iota _{\widehat{s\mathrm{}}(2)}=0`$, $`i1`$, and
(4.45) $`(f_{2\iota })^r|r,k|\iota _{\widehat{s\mathrm{}}(2)}=0,`$
(4.46) $`(f_{2\iota 1})^{kr+1}(f_{2\iota })^{r1}|r,k|\iota _{\widehat{s\mathrm{}}(2)}=|r,k|\iota 1_{\widehat{s\mathrm{}}(2)}.`$
In more formal terms, there is a theorem parallel to Theorem 1.1. Let $`𝖥(k)`$ be the algebra generated by the elements $`(f_n)_n`$ modulo the relations following (4.44) and by an invertible operator $`𝖴`$ such that $`𝖴f_n𝖴^1=f_{n+1}`$.
###### Theorem 4.5.
Let $`M`$ be the representation of the algebra $`𝖥(k)`$ induced from the trivial one-dimensional representation of the algebra of $`(f_n)_{n0}`$ (on the vacuum vector $`|0`$). Let $`C_r`$ $`(1rk+1)`$ be the $`𝖥(k)`$-submodule generated from the vector $`f_1^r|0`$, and $`N_{r,k}`$ the $`𝖥(k)`$-submodule generated from the set of vectors
(4.47)
$$f_{\alpha 2}^{kr+1}f_{\alpha 1}^{r1}|\alpha |\alpha 2,\alpha ,$$
where $`|\alpha =𝖴^\alpha |0`$. The quotient space $`M/(N_{r,k}+C_r)`$ is a representation of the $`\widehat{s\mathrm{}}(2)_k`$ algebra and, moreover, is isomorphic to a direct sum of unitary $`\widehat{s\mathrm{}}(2)_k`$ representations,
(4.48)
$$𝔐(r,k)\frac{M}{N_{r,k}+C_r}=\underset{\theta =0}{\overset{1}{}}𝔏_{r,k;\theta }=𝔏_{r,k}𝔏_{k+2r,k}.$$
In the semi-infinite $`\widehat{s\mathrm{}}(2)_k`$ space, there also exists a monomial basis consisting of “thin” monomials; the above proof applies with minimal modifications. The construction of the $`N=2`$ algebra action on the semi-infinite space can be easily carried over to the case of the $`\widehat{s\mathrm{}}(2)`$ algebra (see Sec. 5.5).
## 5. The $`N=2`$ algebra action on $`𝖶_{r,p;\theta }`$
A priori, the conditions imposed on the semi-infinite construction do not suggest that the space is a representation of any algebra; for the constraints (1.1), however, this representation can be found.
###### Theorem 5.1.
The semi-infinite space $`𝖶_{r,p;\theta }`$ is a module over the $`N=2`$ algebra.
The problem with constructing the $`N=2`$ action on $`𝖶_{r,p;\theta }`$ is nontrivial because we must define the action of $`𝒬_n`$, $`_n`$, and $`_n`$ on the states $`𝒢_{i_1}𝒢_{i_2}\mathrm{}𝒢_{i_m}|r,p;\theta |\iota _{\mathrm{}/2}`$ constructed only from $`𝒢_n`$ and show that this action can be pushed forward to the quotient with respect to the ideal $`𝒮^p`$ generated by $`S_a^p`$. The action of the $`N=2`$ algebra on $`𝖶_{r,p;\theta }`$ is defined in several steps. The main tool here is the positive filtration on $`𝖶_{r,p;\theta }`$ by finite-dimensional subspaces $`𝖶_{r,p;\theta }^+[\iota ]`$ (similar to the Demazure modules, see ) that allows rewriting any element of $`𝖶_{r,p;\theta }`$ as a linear combination of semi-infinite forms involving only nonnegative modes $`𝒢_{n0}`$. The next problem consists in taking the quotient, i.e., in verifying that the action is independent of the chosen representative of a state written in terms of nonnegative $`𝒢`$-modes. In Sec. 5.2, we define the action of a part of the $`N=2`$ generators using differential operators acting on finite-dimensional subspaces whose quotients are the subspaces in the positive filtration. To prove that these differential operators can be pushed forward to the quotient, we take the dual space and use the functional realization (Sec. 5.3). In Sec. 5.4, we then show that the action of the entire $`N=2`$ algebra on the entire semi-infinite space can be obtained by consistently gluing together the “partial” actions on the subspaces. In particular, this gives the action of $`𝒬_{n0}`$; together with $`𝒢_{n0}`$, which act on the semi-infinite space by definition, these generate the entire $`N=2`$ algebra, and it only remains to verify that their action on $`𝖶_{r,p;\theta }`$ is precisely the $`N=2`$ algebra action. We show that the necessary relations are satisfied on any vector from the semi-infinite space.
### 5.1. The positive filtration
Let $`𝖶_{r,p;\theta }^+[\iota ]𝖶_{r,p;\theta }`$ be the subspace generated from the extremal state $`|r,p;\theta |\iota _{\mathrm{}/2}`$ by $`𝒢_0𝔤(p)`$. Relations (3.4) with $`\iota 1`$ determine the sequence of embeddings
(5.1)
$$𝖶_{r,p;\theta }^+[0]\mathrm{}𝖶_{r,p;\theta }^+[\iota ]𝖶_{r,p;\theta }^+[\iota +1]\mathrm{}.$$
###### Lemma 5.2.
Sequence (5.1) is a filtration on the space $`𝖶_{r,p;\theta }`$.
Therefore, each state in $`𝖶_{r,p;\theta }`$ can be represented as a linear combination of monomials involving only nonnegative modes $`𝒢_{n0}`$.
Proof. Abusing the terminology, we say a “semi-infinite form $`𝒢_{i_m}\mathrm{}𝒢_{i_1}|r,p;\theta |\iota _{\mathrm{}/2}`$” meaning in fact its representative in the freely generated space whose quotient is $`𝖶_{r,p;\theta }`$. The statement of the lemma is that each semi-infinite form has a representative expressed only through nonnegative modes $`𝒢_n`$, i.e., a representative of each state $`𝒢_{i_m}\mathrm{}𝒢_{i_1}|r,p;\theta |\iota _{\mathrm{}/2}𝖶_{r,p;\theta }`$ can be chosen from some $`𝖶_{r,p;\theta }^+[\iota ^{}]`$. If $`\iota <0`$, Eq. (3.4) allows us to rewrite the state as $`𝒢_{i_m}\mathrm{}𝒢_{i_1}𝒢_{\iota p+\theta +1}\mathrm{}𝒢_{\iota p+\theta +pr1}\mathrm{}𝒢_{\iota ^{}p+\theta r+1}\mathrm{}𝒢_{\iota ^{}p+\theta 1}|r,p;\theta |\iota ^{}_{\mathrm{}/2}`$ with $`\iota ^{}0`$. We can therefore assume that $`\iota 0`$. We then choose the negative mode $`𝒢_{i_a}`$ that is nearest to $`|r,p;\theta |\iota _{\mathrm{}/2}`$ (i.e., $`i_a<0`$, but $`i_b0`$ for all $`a>b1`$) and consider the state $`𝒢_{i_a}\mathrm{}𝒢_{i_1}|r,p;\theta |\iota _{\mathrm{}/2}`$. Rewriting it (again using (3.4)) as
(5.2)
$$\begin{array}{c}𝒢_{i_a}\mathrm{}𝒢_{i_1}𝒢_{\iota p+\theta +1}\mathrm{}𝒢_{\iota p+\theta +pr1}𝒢_{\iota p+\theta +pr+1}\mathrm{}𝒢_{(\iota +1)p+\theta 1}|r,p;\theta |\iota +1_{\mathrm{}/2}=\hfill \\ \hfill =(1)^{a1}𝒢_{i_{a1}}\mathrm{}𝒢_{i_1}\underset{}{𝒢_{i_a}𝒢_{\iota p+\theta +1}\mathrm{}𝒢_{\iota p+\theta +pr1}𝒢_{\iota p+\theta +pr+1}\mathrm{}𝒢_{(\iota +1)p+\theta 1}}|r,p;\theta |\iota +1_{\mathrm{}/2},\end{array}$$
we apply the basic identity $`S_A^p=0`$ with $`A=(p2)(p\iota +\theta +p)p(p1)/2+r+i_a`$ to the selected modes. As the result, we obtain a linear combination of states in each of which the rightmost negative mode $`𝒢_{i_a^{}}`$ is such that $`i_a<i_a^{}`$ (this is so because in the typical term $`\mathrm{coeff}𝒢_{j_0}𝒢_{j_1}\mathrm{}𝒢_{j_{pr1}}𝒢_{j_{pr+1}}\mathrm{}𝒢_{j_{p1}}`$ resulting from applying condition $`S_A^p=0`$, we have $`j_b<\iota p+\theta +b`$ for $`1bp1`$, $`br`$, whereas the sum $`_{b0}j_b`$ is fixed; therefore, $`j_0>i_a`$). Repeating this procedure with each of the terms obtained, we obtain the product $`𝒢_{i_m}\mathrm{}𝒢_{i_{a+1}}`$ applied to a linear combination of states from some $`𝖶_{r,p;\theta }^+[\iota ^{}]`$. This procedure is then repeated for $`𝒢_{i_{a+1}}`$ etc. We finally obtain the states that belong to a finite sum $`_\iota ^{}𝖶_{r,p;\theta }^+[\iota ^{}]`$, and hence, to some space $`𝖶_{r,p;\theta }^+[\iota ^{\prime \prime }]`$ for a sufficiently large $`\iota ^{\prime \prime }`$.
###### Remark 5.3.
The choice of the modes $`𝒢_{n0}`$ is a matter of convention; for any $`\mu 0`$, there exists a similar filtration in terms of the spaces generated by $`𝒢_{n\mu }`$. We denote it as
(5.3)
$$𝖶_{r,p;\theta }^{(\mu )}[0]\mathrm{}𝖶_{r,p;\theta }^{(\mu )}[\iota ]𝖶_{r,p;\theta }^{(\mu )}[\iota +1]\mathrm{}.$$
In this notation, the above positive filtration corresponds to $`\mu =0`$, $`𝖶_{r,p;\theta }^+[\iota ]𝖶_{r,p;\theta }^{(0)}[\iota ]`$ (see Fig. 2). These filtrations are crucial for constructing the $`N=2`$ action.
We give several examples of rewriting semi-infinite forms in terms of positive modes. We recall that the $`|r,p;\theta |\iota `$ states are defined in (3.1)–(3.4). For $`p=3`$ and $`r=1`$, we have $`|1,3;0|1=(1/24)𝒢_1𝒢_2𝒢_3𝒢_4|1,3;0|3(3/40)𝒢_1𝒢_2𝒢_3𝒢_6𝒢_8|1,3;0|4`$. Complexity of the expressions involving only positive modes grows very fast, as, for example, we see from
(5.4)
$$\begin{array}{c}𝒢_4|1,3;0|0=\frac{9}{125}𝒢_1𝒢_2𝒢_3𝒢_4𝒢_8|1,3;0|4\frac{818}{17325}𝒢_1𝒢_2𝒢_3𝒢_5𝒢_7|1,3;0|4+\hfill \\ \hfill +\frac{4}{2475}𝒢_1𝒢_2𝒢_4𝒢_5𝒢_6|1,3;0|4\frac{66}{2275}𝒢_1𝒢_2𝒢_3𝒢_4𝒢_8𝒢_{12}|1,3;0|5+\frac{12498}{175175}𝒢_1𝒢_2𝒢_3𝒢_5𝒢_9𝒢_{11}|1,3;0|5+\\ \hfill +\frac{342}{2275}𝒢_1𝒢_2𝒢_3𝒢_6𝒢_8𝒢_{11}|1,3;0|5+\frac{24769}{175175}𝒢_1𝒢_2𝒢_3𝒢_6𝒢_9𝒢_{10}|1,3;0|5+\frac{21}{3575}𝒢_1𝒢_2𝒢_5𝒢_6𝒢_8𝒢_9|1,3;0|5\\ \hfill \frac{19}{3185}𝒢_1𝒢_3𝒢_4𝒢_6𝒢_8𝒢_9|1,3;0|5+\frac{4808}{175175}𝒢_1𝒢_3𝒢_5𝒢_6𝒢_7𝒢_9|1,3;0|5\frac{3}{3185}𝒢_2𝒢_3𝒢_4𝒢_5𝒢_8𝒢_9|1,3;0|5\\ \hfill \frac{6}{2275}𝒢_2𝒢_3𝒢_5𝒢_6𝒢_7𝒢_8|1,3;0|5+\frac{9}{280}𝒢_1𝒢_2𝒢_3𝒢_5𝒢_9𝒢_{12}𝒢_{15}|1,3;0|6\frac{171}{980}𝒢_1𝒢_2𝒢_3𝒢_6𝒢_9𝒢_{12}𝒢_{14}|1,3;0|6\\ \hfill \frac{171}{1960}𝒢_1𝒢_3𝒢_5𝒢_6𝒢_9𝒢_{11}𝒢_{12}|1,3;0|6+\frac{9}{980}𝒢_2𝒢_3𝒢_5𝒢_6𝒢_9𝒢_{10}𝒢_{12}|1,3;0|6.\end{array}$$
We thus obtain states belonging to $`𝖶_{1,3;0}^+[\iota =6]`$.
### 5.2. Differential operators for generators on subspaces
To prove Theorem 5.1, we first construct the action of a part of the $`N=2`$ generators on each space $`𝖶_{r,p;\theta }^{(\mu )}[\iota ]`$ involved in (5.3) (see Fig. 2 for $`\mu =0`$). In $`𝖶_{r,p;\theta }^{(\mu )}[\iota ]`$, we define the action of the operators $`𝒬_\mu ,𝒬_{\mu +1},\mathrm{},𝒬_{\iota p+\theta 1}`$, $`_1,_2,\mathrm{}`$, $`_1,_2,\mathrm{}`$, and $`𝒢_\mu ,𝒢_{\mu +1},\mathrm{},𝒢_{\iota p+\theta 1}`$ (the latter act tautologically), which eventually becomes a part of the $`N=2`$ algebra action on the entire semi-infinite space $`𝖶_{r,p;\theta }`$.
As the first step, we “standardize” the spaces by applying the spectral flow mapping each $`𝖶_{r,p;\theta }^{(\mu )}[\iota ]`$ into the space $`𝖵_{r,p}^N`$ generated by $`𝒢_1,\mathrm{},𝒢_N`$. Let $`𝖵(N)`$ (where $`N`$ is a positive integer) be the subspace in $`𝖶(0)`$ generated by $`𝒢_1,𝒢_2,\mathrm{},𝒢_N`$ from the corresponding vacuum vector $`|`$. As in Sec. 4.1, $`𝖶(0)`$ denotes the freely generated space, and $`𝖶_{r,p}(0)`$ is the quotient of $`𝖶(0)`$ with respect to the subspace $`_{r,p}|`$, where $`_{r,p}`$ is the ideal generated by the elements
(5.5)
$$\overline{S}_a^p=\underset{\begin{array}{c}n_0<\mathrm{}<n_{p2}1\\ n_0+\mathrm{}+n_{p2}=a\end{array}}{}\left(\underset{i<j}{}(n_in_j)\right)𝒢_{n_0}\mathrm{}𝒢_{n_{p2}},a=\frac{p(p1)}{2},\frac{p(p1)}{2}1,\mathrm{},$$
and by the element $`S_r=𝒢_r𝒢_{r+1}\mathrm{}𝒢_1`$. We define $`𝖵_{r,p}^N`$ as the subspace in $`𝖶_{r,p}(0)`$ generated by the modes $`𝒢_1`$, …, $`𝒢_N`$ from the vacuum vector $`||r,p;0|0`$. Therefore, $`𝖵_{r,p}^N=𝖵(N)/_{r,p}(N)`$, where $`_{r,p}(N)=𝖵(N)_{r,p}`$ (and we write $`_{r,p}_{r,p}|`$), or equivalently,
(5.6)
$$\begin{array}{ccccccccc}0& & _{r,p}& & 𝖶(0)& & 𝖶_{r,p}(0)& & 0\\ & & & & & & & & & & \\ 0& & _{r,p}(N)& & 𝖵(N)& & 𝖵_{r,p}^N& & 0.\end{array}$$
We let $`_n`$ denote the operator $`/𝒢_n`$ acting on the freely generated space $`𝖶(0)`$ and its subspaces. Although the operators $`_n`$ certainly do not act on the quotient space $`𝖵_{r,p}^N`$, there are differential operators constructed from $`_n`$ and $`𝒢_m`$ that do, and these are a part of the $`N=2`$ generators.
###### Lemma 5.4.
The differential operators
(5.7) $`𝒬_{\mathrm{}}^{(r,p,N)}=`$ $`𝗊^{(r,p,N)}\delta _{\mathrm{},N}_N+{\displaystyle \underset{\begin{array}{c}n,m=N\\ m+n+\mathrm{}1\end{array}}{\overset{1}{}}}(mn)𝒢_{\mathrm{}+n+m}_n_m,`$ $`\mathrm{}=N,N+1,\mathrm{},2N1,`$
(5.8) $`\begin{array}{cc}\hfill _n^{(r,p,N)}=& {\displaystyle \underset{m=N}{\overset{n1}{}}}(nm)𝒢_{n+m}_m+𝗅^{(r,p,N)}\delta _{n,0},\hfill \\ \hfill _n^{(r,p,N)}=& {\displaystyle \underset{m=N}{\overset{n1}{}}}𝒢_{n+m}_m+𝗁^{(r,p,N)}\delta _{n,0},\hfill \end{array}`$ $`n0,`$
where (for an arbitrary $`𝗁^{(r,p,N)}`$)
(5.9) $`𝗅^{(r,p,N)}={\displaystyle \frac{p2}{p}}N+N𝗁^{(r,p,N)}N{\displaystyle \frac{pr1}{p}},`$
(5.10) $`𝗊^{(r,p,N)}={\displaystyle \frac{p2}{p}}(N^2+N)2N{\displaystyle \frac{p1r}{p}}`$
have a well-defined action on the space $`𝖵_{r,p}^N`$. Together with $`𝒢_{\mathrm{}}`$ for $`\mathrm{}N`$, these operators satisfy the corresponding commutation relations in (2.1).
The spaces $`𝖶_{r,p;\theta }^+[\iota ]`$ involved in the positive filtration are related to $`𝖵_{r,p}^N`$ with $`N=\iota p+\theta `$ by spectral flow transformations, namely,
(5.11)
$$𝖴_{\iota p\theta }𝖶_{r,p;\theta }^+[\iota ]𝖵_{r,p}^{\iota p+\theta }.$$
These isomorphisms allow us to carry the operator action over to the above spaces $`𝖶_{r,p;\theta }^+[\iota ]`$.
The main complication in proving Lemma 5.4 is rooted in the fact that the ideal $`_{r,p}(N)𝖵(N)`$ is not described explicitly (cf. the remarks before Theorem 3.2). The idea of the (quite lengthy) proof of Lemma 5.4 is expressed in a more compact form in Sec. 5.5 for the $`\widehat{s\mathrm{}}(2)`$ algebra.
### 5.3. Duals to differential operators on functional spaces
To prove Lemma 5.4, we first note that $`𝗊^{(r,p,N)}`$ is fixed by the requirement that $`𝒬_N^{(r,p,N)}`$ preserve the relations $`\overline{S}_a^p=0`$ (which can be seen by directly applying the differential operators); the form of $`𝗅^{(r,p,N)}`$ is chosen such that the commutator $`[𝒬_N^{(r,p,N)},𝒢_N^{(r,p,N)}]`$ is the same as in the $`N=2`$ algebra. However, $`𝗁^{(r,p,N)}`$ is still arbitrary; moreover, the central term arising in the last commutator can be absorbed in $`_0^{(r,p,N)}`$ and $`_0^{(r,p,N)}`$, and therefore, the $`N=2`$ central charge is still undetermined. Now, the operators $`𝒢_{\mathrm{}}^{(r,p,N)}`$, $`𝒬_{\mathrm{}}^{(r,p,N)}`$, $`_{\mathrm{}}^{(r,p,N)}`$, and $`_{\mathrm{}}^{(r,p,N)}`$ preserve the subspace $`𝖵(N)`$, and it must be verified that they also preserve the ideal $`_{r,p}`$. We replace this with the dual statement and apply the functional realization in Sec. 4.1. Dualizing (5.6), we obtain
(5.12)
$$\begin{array}{ccccccccc}& & 0& & & & & & \\ & & & & & & & & & & \\ & & 𝖶_{r,p}(0)^{}𝒥(N)& & & & & & \\ & & & & & & & & & & \\ 0& & 𝖶_{r,p}(0)^{}& & 𝖶(0)^{}& & _{r,p}^{}& & 0\\ & & & & & & & & & & \\ 0& & 𝖵_{r,p}^{N}{}_{}{}^{}& & 𝖵(N)^{}& & _{r,p}(N)^{}& & 0\\ & & & & & & & & & & \\ & & 0& & & & & & \end{array}$$
The property of the operators to have a well-defined action on the quotient then reformulates as the condition for their duals to preserve the subspace $`𝖶_{r,p}(0)^{}`$ of antisymmetric polynomials satisfying conditions (4.5) and (4.6) modulo the subspace $`𝒥(N)`$ spanned by the monomials $`x_1^{i_1}\mathrm{}x_n^{i_n}`$ with $`i_jN`$ for at least one $`j`$ (more precisely, modulo the intersection of this subspace with the space of skew-symmetric polynomials).
As before, we let $`x_1,\mathrm{},x_M^{(p)}`$ denote polynomials in $`x_1,\mathrm{},x_M`$ satisfying condition (4.5). We now establish that for $`fx_1,\mathrm{},x_M^{(p)}`$,
(5.13)
$$𝒦_{n}^{(r,p,N)}{}_{}{}^{}fx_1,\mathrm{},x_M^{(p)},𝒦_n=_n\text{or}_n,$$
while $`𝒬_{\mathrm{}}`$ preserve the conditions modulo $`𝒥(N)`$,
(5.14)
$$𝒬_{\mathrm{}}^{(r,p,N)}{}_{}{}^{}fx_1,\mathrm{},x_{M+1}^{(p)}+𝒥(N),$$
and hence the expression
(5.15)
$$\frac{}{x_2}\frac{^2}{x_3^2}\mathrm{}\frac{^{p2}}{x_{p1}^{p2}}\left(𝒬_{\mathrm{}}^{(r,p,N)}{}_{}{}^{}f\right)(x_1,\mathrm{},x_{M+1})|_{x_1=x_2=\mathrm{}=x_{p1}}$$
vanishes modulo $`𝒥(Np+2)`$. For this, we derive a suitable form of the dual operators.
The operators dual to $`𝒢_i`$ act between the $`M`$-variable subspaces in $`𝖶(0)^{}`$ as
(5.16) $`x_1,\mathrm{},x_{M1}dx_1\mathrm{}dx_{M1}\stackrel{𝒢_i^{}}{}x_1,\mathrm{},x_Mdx_1\mathrm{}dx_M,`$
(5.17) $`\left(𝒢_{\mathrm{}}^{(r,p,N)}{}_{}{}^{}f\right)(x_1,\mathrm{},x_{M1})={\displaystyle \frac{1}{(\mathrm{}1)!}}{\displaystyle \frac{^\mathrm{}1}{x_M^\mathrm{}1}}f(x_1,\mathrm{},x_M)|_{x_M=0}.`$
The duals to $`_{\mathrm{}}=/𝒢_{\mathrm{}}`$ are the pairwise anticommuting differentials
(5.18)
$$x_1,\mathrm{},x_Mdx_1\mathrm{}dx_M\stackrel{_{\mathrm{}}^{}}{}x_1,\mathrm{},x_{M+1}dx_1\mathrm{}dx_{M+1},$$
acting as
(5.19)
$$(_{\mathrm{}}^{}f)(x_1,\mathrm{},x_{M+1})=\underset{j=1}{\overset{M+1}{}}(1)^{M+1j}f\left([𝒙]_j^{(M+1)}\right)x_j^\mathrm{}1,$$
where we use the notation
(5.20) $`[𝒙]^{(M)}=(x_1,\mathrm{},x_M),[𝒙]_i^{(M)}=(x_1,\mathrm{},x_{i1},x_{i+1},\mathrm{},x_M),`$
(5.21) $`[𝒙]_{ij}^{(M)}=(x_1,\mathrm{},x_{i1},x_{i+1},\mathrm{},x_{j1},x_{j+1},\mathrm{},x_M),1i<jM.`$
For $`fx_1,\mathrm{},x_M`$, we then have, for example,
(5.22)
$$(_m^{}_n^{}f)(x_1,\mathrm{},x_{M+2})=\underset{1i<jM+2}{}(1)^{i+j+1}f([𝒙]_{ij}^{(M+2)})(x_i^{n1}x_j^{m1}x_i^{m1}x_j^{n1}).$$
The dual action of $`_n`$ and $`_n`$ is given by
(5.23) $`_n^{}f`$ $`=\left(p_n+𝗁^{(r,p,N)}\delta _{n,0}\right)f,`$
(5.24) $`(_n^{}f)(x_1,\mathrm{},x_M)`$ $`=((2n+1)p_n(x_1,\mathrm{},x_M)+`$
$`+𝗅^{(r,p,N)}\delta _{n,0}+{\displaystyle \underset{j=1}{\overset{M}{}}}x_j^{n+1}{\displaystyle \frac{}{x_j}})f(x_1,\mathrm{},x_M),`$
where $`gh`$ denotes the “pointwise” multiplication (of a symmetric and an antisymmetric polynomial) and we introduce the symmetric polynomials
(5.25)
$$p_r(x_1,\mathrm{},x_M)=x_1^r+\mathrm{}+x_M^r.$$
It is obvious that the operators $`_n`$ with $`n0`$ map $`𝖶_{r,p}(0)^{}`$ into itself. For $`_n`$, this statement is verified as follows. We split the summation (5.24) as
(5.26)
$$\underset{j=1}{\overset{M}{}}\frac{}{x_j}f(x_1,\mathrm{},x_M)=\left(\underset{j=1}{\overset{p1}{}}+\underset{j=p}{\overset{M}{}}\right)\frac{}{x_j}f(x_1,\mathrm{},x_M).$$
Each term in the second sum in the right-hand side of (5.26) preserves the conditions
(5.27)
$$\frac{}{x_2}\frac{^2}{x_3^2}\mathrm{}\frac{^{p2}}{x_{p1}^{p2}}f|_{x_1=x_2=\mathrm{}=x_{p1}}=0.$$
To the first sum, we apply the operator
(5.28)
$$\frac{}{x_2}\frac{^2}{x_3^2}\mathrm{}\frac{^{p2}}{x_{p1}^{p2}},$$
commute it with
(5.29)
$$\underset{j=1}{\overset{p1}{}}x_j^{n+1}\frac{}{x_j}$$
and restrict the expression obtained to the $`(p1)`$-diagonal $`x_1=x_2=\mathrm{}=x_{p1}=x`$, after which the sum
(5.30)
$$\underset{j=1}{\overset{p1}{}}x_j^{n+1}\frac{}{x_j}=x^{n+1}\underset{j=1}{\overset{p1}{}}\frac{}{x_j}$$
becomes the derivative along the diagonal, and hence,
(5.31)
$$\underset{j=1}{\overset{p1}{}}\frac{}{x_j}\frac{}{x_2}\frac{^2}{x_3^2}\mathrm{}\frac{^{p2}}{x_{p1}^{p2}}f|_{x_1=x_2=\mathrm{}=x_{p1}}=0.$$
Property (5.13) is thus established for $`_n`$.
Some more work is required with the $`𝒬_{\mathrm{}}^{(r,p,N)}{}_{}{}^{}`$ operators. They act in the same direction as the differentials in (5.18); using (5.17) and (5.22), we find
(5.32)
$$\begin{array}{c}(𝒬_{\mathrm{}}^{(r,p,N)}{}_{}{}^{}f)(x_1,\mathrm{},x_{M+1})=𝗊^{(r,p,N)}\delta _{\mathrm{},N}\underset{j=1}{\overset{M+1}{}}(1)^{M+1j}f([𝒙]_j^{(M+1)})x_j^\mathrm{}1+\hfill \\ \hfill +\underset{1i<jM+1}{}(1)^{i+j}\underset{\begin{array}{c}m,n=1\\ m+n\mathrm{}1\end{array}}{\overset{N}{}}\left(x_j\frac{}{x_j}x_i\frac{}{x_i}\right)\frac{x_i^{n1}x_j^{m1}+x_i^{m1}x_j^{n1}}{(n+m\mathrm{}1)!}D_M^{n+m\mathrm{}1}f([𝒙]_{ij}^{(M+1)},0),\end{array}$$
where $`D_if(\xi _1,\mathrm{}\xi _M)=\frac{d}{dt}f(\xi _1,\mathrm{},\xi _{i1},\xi _i+t,\xi _{i+1},\mathrm{},\xi _M)|_{t=0}`$. Since we are only interested in terms modulo $`𝒥(N)`$, we can bring the last equation to a much more tractable form by adding suitable terms from $`𝒥(N)`$. For $`\mathrm{}N`$ as in the condition of the lemma, the Taylor series can be completed by terms from $`𝒥(N)`$, and $`𝒬_{\mathrm{}}^{(r,p,N)}{}_{}{}^{}`$ can therefore be redefined modulo $`𝒥(N)`$ to (omitting the <sup>(M+1)</sup> superscript for brevity)
$$\begin{array}{c}(𝒬_{\mathrm{}}^{(r,p,N)}{}_{}{}^{}f)(x_1,\mathrm{},x_{M+1})=𝗊^{(r,p,N)}\delta _{\mathrm{},N}\underset{j=1}{\overset{M+1}{}}(1)^{M+1j}f([𝒙]_j)x_j^\mathrm{}1+\hfill \\ \hfill +\underset{1i<jM+1}{}(1)^{i+j}\left(x_j\frac{}{x_j}x_i\frac{}{x_i}\right)\left(F_{\mathrm{}}(x_i,x_j)\left(f([𝒙]_{ij},x_i)+f([𝒙]_{ij},x_j)\right)\right),\end{array}$$
where
(5.33)
$$F_{\mathrm{}}(x,y)=\underset{n=0}{\overset{\mathrm{}1}{}}x^ny^{\mathrm{}1n}=\frac{x^{\mathrm{}}y^{\mathrm{}}}{xy}.$$
Note that for antisymmetric polynomials, $`f([𝒙]_{ij}^{(M+1)},x_i)=(1)^{Mi}f([𝒙]_j^{(M+1)})`$ and $`f([𝒙]_{ij}^{(M+1)},x_j)=(1)^{M+1j}f([𝒙]_i^{(M+1)})`$ for $`i<j`$. Further, when acting with the derivatives on the sum of two $`f`$ polynomials, we obtain
$$\begin{array}{c}F_{\mathrm{}}(x_i,x_j)\left(x_j\frac{}{x_j}x_i\frac{}{x_i}\right)\left(f([𝒙]_{ij},x_i)+f([𝒙]_{ij},x_j)\right)=\hfill \\ \hfill =(x_j^{\mathrm{}}x_i^{\mathrm{}})\underset{n1}{}\frac{(1)^{M+i}}{n!}(x_jx_i)^{n1}\frac{^ni[f]}{x_i^n}([𝒙]_j)𝒥(N),\end{array}$$
where $`i[f](y_1,\mathrm{},y_M)=y_iD_if(y_1,\mathrm{},y_M)`$. Thus, $`𝒬_{\mathrm{}}^{(r,p,N)}{}_{}{}^{}f`$ can be redefined modulo $`𝒥(N)`$ as
(5.34)
$$(𝒬_{\mathrm{}}^{(r,p,N)}{}_{}{}^{}f)(x_1,\mathrm{},x_{M+1})=\underset{j=1}{\overset{M+1}{}}(1)^{M+j}f([𝒙]_j)\left(\underset{i=1}{\overset{M+1}{}}A_{\mathrm{}}(x_j,x_i)𝗊^{(r,p,N)}\delta _{\mathrm{},N}x_j^\mathrm{}1\right)$$
with
(5.35)
$$\begin{array}{cc}\hfill A_{\mathrm{}}(x_j,x_i)=& \underset{m=1}{\overset{\mathrm{}}{}}(\mathrm{}+12m)x_j^\mathrm{}mx_i^{m1}=\frac{(\mathrm{}1)(x_j^{\mathrm{}+1}x_i^{\mathrm{}+1})(\mathrm{}+1)(x_j^{\mathrm{}}x_ix_i^{\mathrm{}}x_j)}{(x_jx_i)^2}\hfill \\ \hfill =& (x_jx_i)B_{\mathrm{}}(x_j,x_i),B_{\mathrm{}}(x_j,x_i)=\underset{m=1}{\overset{\mathrm{}1}{}}m(\mathrm{}m)x_j^{\mathrm{}m1}x_i^{m1}.\hfill \end{array}$$
This can be expressed in a more “invariant” form as follows. We define the differential $`d_m`$ to be $`_{m1}^{}`$ up to a sign,
(5.36)
$$(d_mf)(x_1,\mathrm{},x_{M+1})=\underset{j=1}{\overset{M+1}{}}(1)^{j+1}x_j^mf([x]_j),m0.$$
Up to a similar conventional sign factor $`(1)^{M+1}`$ (which is inessential for the vanishing property we want to show), we finally rewrite (5.34) as
(5.37)
$$𝒬_{\mathrm{}}^{(r,p,N)}{}_{}{}^{}f=\underset{m=1}{\overset{\mathrm{}}{}}(\mathrm{}+12m)p_{m1}d_\mathrm{}mf𝗊^{(r,p,N)}\delta _{\mathrm{},N}d_\mathrm{}1f.$$
Equation (5.14) can now be shown by induction on the number of variables $`M`$. Assuming that (5.14) holds for all $`fx_1,\mathrm{},x_{M1}^{(p)}`$, we take a polynomial $`fx_1,\mathrm{},x_M^{(p)}`$ and consider
(5.38)
$$\begin{array}{c}𝒢_m^{}𝒬_{\mathrm{}}^{(r,p,N)}{}_{}{}^{}f=𝒬_{\mathrm{}}^{(r,p,N)}{}_{}{}^{}(𝒢_m^{}f)+_{m+\mathrm{}}^{}f2\mathrm{}_{m+\mathrm{}}^{}f+\frac{p2}{p}(m^2+m)\delta _{m+\mathrm{},0}f,\hfill \\ \hfill m=1,\mathrm{},N,\end{array}$$
where the right-hand side follows from the explicit expressions (most easily, in the form of differential operators). In (5.38), we already know that $`_{m+\mathrm{}}^{}`$ and $`_{m+\mathrm{}}^{}`$ map $`x_1,\mathrm{},x_M^{(p)}`$ into itself. Moreover, we see from (5.16) that $`𝒢_m^{}f`$ is a polynomial in $`M1`$ variables and, moreover, it belongs to $`x_1,\mathrm{},x_{M1}^{(p)}`$. By the induction hypothesis, therefore, $`𝒬_{\mathrm{}}^{(r,p,N)}{}_{}{}^{}(𝒢_m^{}f)x_1,\mathrm{},x_M^{(p)}+𝒥(N)`$; thus, the same is true for $`𝒢_m^{}𝒬_{\mathrm{}}^{(r,p,N)}{}_{}{}^{}f`$. This in turn tells us much about $`𝒬_{\mathrm{}}^{(r,p,N)}{}_{}{}^{}f`$: the first $`N1`$ terms in its Taylor expansion in $`x_{M+1}`$ are in $`𝒥(N)+x_1,\mathrm{},x_M^{(p)}`$; the remaining terms do not interest us, however, because they certainly are in $`𝒥(N)`$. It only remains to prove the induction base, namely that $`𝒬_{\mathrm{}}^{(r,p,N)}{}_{}{}^{}fx_1,\mathrm{},x_p^{(p)}+𝒥(N)`$ for $`fx_1,\mathrm{},x_{p1}^{(p)}`$, which follows from the explicit form (5.37).
Finally, condition (4.6) on polynomials is easiest to consider in terms of the symmetric polynomial corresponding to a given antisymmetric one (see (4.7)); this condition states the vanishing of the polynomial at zero, which can be reformulated in terms of filtration (4.20). It is then easy to directly verify that the operators preserve these vanishing conditions.
### 5.4. The algebra action on $`𝖶_{r,p;\theta }`$ by gluing the pieces together
To complete the proof of Theorem 5.1, we construct the action of the $`N=2`$ generators on the entire semi-infinite space $`𝖶_{r,p;\theta }`$ by gluing together the actions constructed on each of the spaces involved in filtration (5.3). This is done in several steps.
Because the dependence on the spectral flow parameter $`\theta `$ does not affect the general structure of the results, we temporarily set $`\theta =0`$ to simplify the formulas. We therefore consider the semi-infinite spaces $`𝖶_{r,p}𝖶_{r,p;0}`$, with the dependence on $`\theta `$ to be reconstructed in accordance with the spectral flow. We write $`|r,p|\iota _{\mathrm{}/2}`$ for $`|r,p;0|\iota _{\mathrm{}/2}`$.
#### 5.4.1.
For a given semi-infinite form $`|x𝖶_{r,p}`$ with the charge–level bigrade $`(h,\mathrm{})`$, we choose $`\iota `$ and $`\mu <\iota p1`$ such that all the states in this bigrade lie in $`𝖶_{r,p}^{(\mu )}[\iota ]`$. Wishing to define the state $`𝒪_n|x`$ for $`𝒪=`$, $``$, or $`𝒬`$, we also require that the space $`𝖶_{r,p}^{(\mu )}[\iota ]`$ contain all the states in the corresponding bigrade (given by $`(h,\mathrm{}a)`$ for $`𝒪=`$ or $``$ and $`(h1,\mathrm{}a)`$ for $`𝒬`$). This can be always achieved by increasing $`\iota `$ or $`\mu `$.
Similarly to (5.11), the spectral flow transform can be used to map the state $`|r,p|\iota _{\mathrm{}/2}`$ into $`|r,p|0_{\mathrm{}/2}`$ and all the subspaces $`𝖶_{r,p}^{(\mu )}[\iota ]`$ into the corresponding spaces $`𝖵_{r,p}^N`$ in Sec. 5.2,
(5.39)
$$𝖴_{\iota p}𝖶_{r,p}^{(\mu )}[\iota ]𝖵_{r,p}^{\iota p\mu }.$$
In $`𝖵_{r,p}^{\iota p\mu }`$, we apply the operator $`𝖴_{\iota p}𝒪_n𝖴_{\iota p}`$ to $`𝖴_{\iota p}|x`$ using the differential operators in Lemma 5.4, and then use the spectral flow to transform the result back into $`𝖶_{r,p}^{(\mu )}[\iota ]`$. We thus set
(5.40)
$$𝒪_n|x=𝖴_{\iota p}\left(\left(𝖴_{\iota p}𝒪_n𝖴_{\iota p}\right)^{(r,p,\iota p\mu )}𝖴_{\iota p}|x\right),$$
where
(5.41)
$$𝒪_n=_n,n0;𝒪_n=_n,n0;𝒪_n=𝒬_n,n\mu ,$$
and the superscript in (5.40) means that the corresponding differential operator from Lemma 5.4 is applied.
The operators in Eqs. (5.7) and (5.8) depend on $`N=\iota p\mu `$ and involve an additional parameter $`𝗁^{(r,p,N)}`$. Another free parameter arises when the spectral flow is applied: while mapping semi-infinite forms involves only the relations $`𝖴_\theta 𝒢_n𝖴_\theta =𝒢_{n+\theta }`$ and $`𝖴_{\iota p}|r,p|\iota _{\mathrm{}/2}=|r,p|0_{\mathrm{}/2}`$, which are a part of the definition of the semi-infinite space, it is understood that the operator $`𝖴_{\iota p}𝒪_n𝖴_{\iota p}`$ is evaluated using the $`N=2`$ spectral flow formula (2.2). This gives rise to the parameter $`c`$, which is also free at this stage.
Remarkably, all the free parameters are uniquely fixed by the consistency requirements. Moreover, it can then be proved that the action defined above is independent of the chosen $`\iota `$ and $`\mu `$. This is shown in Secs. 5.4.2 and 5.4.3; in Secs. 5.4.4 and 5.4.5, we then establish that the action of the generators constructed defines precisely the $`N=2`$ algebra.
#### 5.4.2.
Using Eqs. (2.2) and (3.9), we compare the eigenvalues of $`_0`$ on the states $`|r,p|\iota _{\mathrm{}/2}`$ and $`|r,p|\iota 1_{\mathrm{}/2}`$. In view of Eqs. (3.4) (see also Fig. 2), these eigenvalues differ by $`p2`$. On the other hand,
(5.42)
$$𝖴_p_0|r,p|\iota 1_{\mathrm{}/2}=𝖴_p_0𝖴_p|r,p|\iota _{\mathrm{}/2}=(_0+\frac{c}{3}p)|r,p|\iota _{\mathrm{}/2},$$
which gives central charge (1.4) expressed through the parameter $`p`$ in the basic relation (1.1).
A similar argument applied to $`_0`$ gives
(5.43)
$$𝖴_p_0|r,p|\iota 1_{\mathrm{}/2}=𝖴_p_0𝖴_p^1|r,p|\iota _{\mathrm{}/2}=(_0+p_0+\frac{c}{6}(p^2+p))|r,p|\iota _{\mathrm{}/2},$$
where we already know $`c`$. On the other hand, (minus) the difference between the eigenvalues of $`_0`$ on the states $`|r,p|\iota 1_{\mathrm{}/2}`$ and $`|r,p|\iota _{\mathrm{}/2}`$ is given by (see (3.4) and Fig. 2)
(5.44)
$$\left([\iota p1]+\mathrm{}+\left[(\iota 1)p+1\right]\right)[\iota pr]=\iota p(p1)\frac{1}{2}p(p1)\iota p+r,$$
whence it follows that the parameter $`𝗁^{(r,p,N)}`$ in Lemma 5.4 is given by
(5.45)
$$𝗁^{(r,p,N)}=\frac{r1}{p}.$$
With this $`𝗁^{(r,p,N)}`$ in (5.9), it follows that $`𝗅^{(r,p,N)}=0`$.
Restoring the dependence on $`\theta `$, we now have the relation
(5.46)
$$_0|r,p;\theta |\iota _{\mathrm{}/2}=(\frac{r1}{p}\frac{p2}{p}\theta (p2)\iota )|r,p;\theta |\iota _{\mathrm{}/2}$$
holding in $`𝖶_{r,p;\theta }`$. This gives the same eigenvalues as in the unitary module (given by Eq. (2.22)), and the same is easily seen to be true for the eigenvalues of $`_0`$ on $`|r,p;\theta |\iota _{\mathrm{}/2}`$ (given by Eq. (2.21)).
#### 5.4.3.
In addition to $`_0`$ and $`_0`$, the action of all the operators $`_0`$ and $`_0`$ is carried over from $`𝖵_{r,p}^{\iota p\mu }`$ to $`𝖶_{r,p}`$. Similarly, in each $`𝖶_{r,p}^{(\mu )}[\iota ]`$, we obtain the action of the operators $`𝒬_n`$ for $`n\mu `$. A priori, they depend on the space $`𝖵_{r,p}^{\iota p\mu }`$ in which the differential operators are applied. Now, as noted above, Eq. (5.10) follows from the condition that the operator $`𝒬_N^{(r,p,N)}`$ preserve the relations $`\overline{S}_a^p=0`$. It is remarkable that for this $`𝗊^{(r,p,N)}`$ and with $`𝗁^{(r,p,N)}`$ chosen in (5.45), we can rewrite Eq. (5.7) as
(5.47)
$$𝒬_{\mathrm{}}^{(r,p,N)}=2\underset{n=N}{\overset{1}{}}\left(_{n+\mathrm{}}^{(r,p,N)}\mathrm{}_{n+\mathrm{}}^{(r,p,N)}+\frac{1}{2}\frac{p2}{p}(\mathrm{}^2\mathrm{})\delta _{\mathrm{}+n,0}\right)_n.$$
Moreover, for $`𝗁^{(r,p,N)}`$ of form (5.45) (and with $`𝗅^{(r,p,N)}=0`$), the right-hand side of (5.47) depends on $`N`$ only through the summation limits. This implies that the differential operators in Lemma 5.4 commute with the embeddings $`𝖵_{r,p}^{N1}𝖵_{r,p}^N`$. Therefore, for $`n0`$, the operators $`_n`$, $`_n`$, and $`𝒬_{\mu +n}`$ acting in $`𝖶_{r,p}`$ as defined in (5.40) are independent of $`\iota `$ and $`\mu `$.
#### 5.4.4.
We have seen that for any $`\mu `$, the operators $`_0`$, $`_0`$, $`𝒢_\mu `$, and $`𝒬_\mu `$ are well-defined on each state in $`𝖶_{r,p;\theta }`$; moreover, these operators satisfy the $`N=2`$ commutation relations. Hence, there is a family of subalgebras of the $`N=2`$ algebra acting on the semi-infinite space $`𝖶_{r,p;\theta }`$. It remains to define the action of $`_{<0}`$ and $`_{<0}`$ and show that all the $`N=2`$ commutation relations are satisfied for the operators $`_n`$, $`_n`$, $`𝒢_n`$, and $`𝒬_n`$ with $`n`$.
We still have not found the commutators $`[𝒬_m,𝒢_n]`$ with $`m+n1`$, which do not follow from the argument based on the spectral flow. We find them by “solving the Jacobi identities.” We first show that the operators $`𝒬_n`$ and $`𝒢_n`$, $`n`$, satisfy the relations
(5.48) $`[𝒢_m,[𝒢_n,𝒬_{\mathrm{}}]]=2(mn)𝒢_{m+n+\mathrm{}},`$
(5.49) $`[𝒬_{\mathrm{}},[𝒬_m,𝒢_n]]=2(\mathrm{}m)𝒬_{\mathrm{}+m+n}.`$
Starting with (5.49), we set
(5.50)
$$𝒳(\mathrm{},m,n)=[𝒬_{\mathrm{}},[𝒬_m,𝒢_n]]2(\mathrm{}m)𝒬_{\mathrm{}+m+n}$$
and show that this expression acts by zero on any state $`|\alpha 𝖶_{r,p;\theta }`$. It suffices to consider monomial states $`|\alpha `$. For given $`\mathrm{}`$, $`m`$, and $`n`$, there exists a positive integer $`\mu `$ such that $`n\mu `$ and $`\mathrm{},m\mu `$. We choose any such $`\mu `$ and use filtration (5.3). Let $`|\alpha 𝖶_{r,p;\theta }^{(\mu )}[\iota _0]`$. This means that the state is represented as $`|\alpha =𝒢_{a_1}\mathrm{}𝒢_{a_\nu }|r,p;\theta |\iota _0_{\mathrm{}/2}`$, where $`a_j\mu `$. We let $`a`$ denote one of $`a_j`$.
Because $`a+\mathrm{}0`$, $`a+m0`$, and $`a+\mathrm{}+m+n0`$, the commutation relations found so far allow us to evaluate
(5.51)
$$\begin{array}{c}[𝒢_a,𝒳(\mathrm{},m,n)]=2[_{a+\mathrm{}}\mathrm{}_{a+\mathrm{}},[𝒬_m,𝒢_n]]2[𝒬_{\mathrm{}},[_{a+m}m_{a+m},𝒢_n]]\hfill \\ \hfill 4(\mathrm{}m)\left(_{a+\mathrm{}+m+n}(\mathrm{}+m+n)_{a+\mathrm{}+m+n}\right).\end{array}$$
Using the already established commutation relations again, we then have
(5.52)
$$\begin{array}{c}[𝒢_a,𝒳(\mathrm{},m,n)]=2(\mathrm{}m)[𝒬_{a+\mathrm{}+m},𝒢_n]+2(an)[𝒬_m,𝒢_{a+\mathrm{}+n}]+\hfill \\ \hfill +2(an)[𝒬_{\mathrm{}},𝒢_{a+m+n}]4(\mathrm{}m)\left(_{a+\mathrm{}+m+n}(\mathrm{}+m+n)_{a+\mathrm{}+m+n}\right).\end{array}$$
We also obtain $`a+\mathrm{}+m\mu `$, $`a+\mathrm{}+n\mu `$, and $`a+m+n\mu `$, and once again applying the already known commutation relations, we see that the right-hand side of Eq. (5.52) vanishes.
The action of $`𝒳(\mathrm{},m,n)`$ on the state $`|\alpha =𝒢_{a_1}\mathrm{}𝒢_{a_\nu }|r,p;\theta |\iota _0_{\mathrm{}/2}`$ is thus given by $`(1)^\nu 𝒢_{a_1}\mathrm{}𝒢_{a_\nu }𝒳(\mathrm{},m,n)|r,p;\theta |\iota _0_{\mathrm{}/2}`$. But it can be assumed (possibly at the expense of increasing $`\iota _0`$) that $`𝒳(\mathrm{},m,n)|r,p;\theta |\iota _{\mathrm{}/2}=0`$, because the operator $`𝒳(\mathrm{},m,n)`$ maps each state $`|r,p;\theta |\iota _{\mathrm{}/2}`$ with $`\iota 0`$ into a bigrade containing no states. Therefore, $`𝒳(\mathrm{},m,n)=0`$ in $`𝖶_{r,p;\theta }`$.
To prove relations (5.48), we denote $`𝒴(m,n,\mathrm{})=[𝒢_m,[𝒢_n,𝒬_{\mathrm{}}]]2(mn)𝒢_{m+n+\mathrm{}}`$ and following (5.51) and (5.52) mutatis mutandis, directly verify that the commutators $`[𝒢_a,𝒴(m,n,\mathrm{})]`$ and $`[𝒬_a,𝒴(m,n,\mathrm{})]`$ vanish for all $`a`$ satisfying the conditions
(5.53)
$$a+\mathrm{}0,a+n0,a+m0,a+\mathrm{}+m+n0.$$
Next, examining the gradings shows that $`𝒴(m,n,\mathrm{})|r,p;\theta |\iota ^{}_{\mathrm{}/2}=0`$ for $`\iota ^{}1`$. However, this does not directly imply that $`𝒴(m,n,\mathrm{})`$ vanishes on any state from the semi-infinite space (compared to the case with $`𝒳(m,n,\mathrm{})`$, the problem is with the apparent asymmetry of the semi-infinite construction with respect to the $`𝒢_n`$ and $`𝒬_n`$ generators). It remains to be shown that $`𝒴(m,n,\mathrm{})|r,p;\theta |\iota _{\mathrm{}/2}=0`$ for $`\iota 1`$, because any state in the semi-infinite space can be generated by the modes $`𝒢_a`$ from $`|r,p;\theta |\iota _{\mathrm{}/2}`$ with a sufficiently large $`\iota `$.
It suffices to show that for any $`\iota 1`$, the state $`|r,p;\theta |\iota _{\mathrm{}/2}`$ can be obtained by acting with the modes $`𝒢_a`$ and $`𝒬_a`$ satisfying conditions (5.53) on the state $`|r,p;\theta |\iota ^{}_{\mathrm{}/2}`$ with some $`\iota ^{}`$ for which $`\mathrm{}+m+np\iota ^{}+\theta `$. Let $`M`$ denote the minimal integer $`a`$ satisfying (5.53). We fix $`\iota `$ such that $`p\iota M`$. We then have
(5.54)
$$𝒢_{\iota p+\theta +1}\mathrm{}𝒢_{\iota p+\theta +pr1}𝒬_{M+\mu }\mathrm{}𝒬_{M+1}𝒬_M|r,p;\theta |\iota ^{}_{\mathrm{}/2}=F|r,p;\theta |\iota _{\mathrm{}/2}$$
for
(5.55) $`\iota ^{}=(p1)\iota \theta Mp+r+1,`$
(5.56) $`\mu =(p2)(p\iota +\theta +M+pr1)+pr2,`$
which follows because both sides of (5.54) belong to the same eigenspace of $`_0`$ and $`_0`$ and this eigenspace is a $`1`$-dimensional subspace in $`𝖶_{r,p;\theta }`$. If we show that $`F0`$ in Eq. (5.54), we can conclude that $`𝒴(m,n,\mathrm{})|r,p;\theta |\iota _{\mathrm{}/2}=0`$ for $`\iota 1`$, because it has already been shown that $`𝒴(m,n,\mathrm{})`$ commutes with all $`𝒢_a`$ and $`𝒬_a`$ for $`aM`$ and $`𝒴(m,n,\mathrm{})|r,p;\theta |\iota ^{}_{\mathrm{}/2}=0`$.
To verify that the coefficient $`F`$ is nonvanishing, it is easiest to use the mapping $`𝖶_{r,p;\theta }𝔎_{r,p;\theta }`$ to the unitary $`N=2`$ module, under which each state (3.1) goes into the corresponding extremal state $`|r,p;\theta |\iota 𝔎_{r,p;\theta }`$ (see Sec. 2.4). In $`𝔎_{r,p;\theta }`$, the image of (5.54) is satisfied with a nonvanishing $`F`$. Therefore, the coefficient $`F`$ in (5.54) cannot vanish, which implies that $`[𝒢_m,[𝒢_n,𝒬_{\mathrm{}}]]=2(mn)𝒢_{m+n+\mathrm{}}`$ in the semi-infinite space.
#### 5.4.5.
Equations (5.48) and (5.49) imply all the commutation relations (2.1). Indeed, for $`a<0`$, we can define the operators
(5.57) $`_a^{m,n}`$ $`={\displaystyle \frac{1}{2(mn)}}\left(m[𝒢_{n+a},𝒬_n]n[𝒢_{m+a},𝒬_m]\right),mn,`$
(5.58) $`_a^{m,i,n}`$ $`={\displaystyle \frac{1}{2m}}\left([𝒢_{m+a},𝒬_m]2_a^{i,n}\right),m0.`$
Let $`A`$ be the algebra generated by $`𝒬_n`$ and $`𝒢_n`$ satisfying relations (5.48) and (5.49). It follows from (5.48) and (5.49) that for any fixed $`m,i,n`$, the operators
$`_a^{m,n},_a^{m,i,n},`$ $`a<0,`$
$`_j,_j,`$ $`j0,`$
$`𝒬_j,𝒢_j,`$ $`j,`$
satisfy commutation relations (2.1). In particular, the operators $`_a^{m,n}_a^{m^{},n^{}}`$ and $`_a^{m,i,n}_a^{m^{},i^{},n^{}}`$ commute with $`𝒬_j`$ and $`𝒢_{\mathrm{}}`$ and generate a commutative ideal $`Z`$ in $`A`$; moreover, $`Z`$ is in the center of $`A`$, and the quotient of $`A`$ with respect to $`Z`$ coincides with the $`N=2`$ algebra. Thus, $`A`$ is a central extension of the $`N=2`$ algebra of the form
(5.59)
$$[𝒢_m,𝒬_n]=2_{m+n}2n_{m+n}+\frac{c}{3}(m^2+m)\delta _{m+n,0}+f_{n,m},$$
where $`f_{n,m}0`$ only for $`n<0`$ and $`m<0`$. In view of the Jacobi identities, $`f_{n,m}=0`$, and therefore, $`Z=0`$ and the operators $`_a^m`$ and $`_a^{m,i,n}`$ are independent of $`m`$, $`i`$, and $`n`$.
This completes the proof of Theorem 5.1. This sufficiently strong result leads to the statement of Theorem 1.1. Before considering that theorem, however, we consider a similar construction on the $`\widehat{s\mathrm{}}(2)`$ semi-infinite space.
### 5.5. The $`\widehat{s\mathrm{}}(2)`$ action on the semi-infinite space
As for the $`N=2`$ algebra, we consider the filtration of the $`\widehat{s\mathrm{}}(2)`$ semi-infinite space by finite-dimensional subspaces
(5.60)
$$𝖬_{r,k}^+[0]𝖬_{r,k}^+[1]\mathrm{}𝖬_{r,k}^+[\iota ]\mathrm{},$$
where $`𝖬_{r,k}^+[\iota ]`$ is generated by $`f_0,f_1,\mathrm{},f_{2\iota }`$ from $`|r,k|\iota _{\widehat{s\mathrm{}}(2)}`$ (up to a spectral flow transform, $`𝖬_{r,k}^+[\iota ]`$ are the Demazure modules ). On each subspace $`𝖬_{r,k}^+[\iota ]`$, we can define a part of the $`\widehat{s\mathrm{}}(2)`$ generators as differential operators with respect to $`f_0,f_1,\mathrm{},f_{2\iota }`$; after the appropriate spectral flow transform, we obtain differential operators with respect to $`f_1,f_2,\mathrm{},f_{2\iota +1}`$. It must be shown that they preserve the ideal generated by the elements
(5.61)
$$\underset{\begin{array}{c}1i_1,\mathrm{},i_{k+1}2\iota +1\\ i_1+\mathrm{}+i_{k+1}=a\end{array}}{}f_{i_1}\mathrm{}f_{i_{k+1}}=0,a=k1,k2,\mathrm{}.$$
On each finite-dimensional space generated by $`f_1,f_2,\mathrm{},f_N`$ from the vacuum vector $`|0_{\widehat{s\mathrm{}}(2)}`$, the operators $`h_0`$ and $`e_N`$ are represented by the differential operators
(5.62) $`h_{\mathrm{}}={\displaystyle \underset{n=N}{\overset{\mathrm{}1}{}}}f_{\mathrm{}+n}_n,\mathrm{}0,`$
(5.63) $`e_{\mathrm{}}={\displaystyle \underset{\begin{array}{c}n,m=N\\ m+n+\mathrm{}1\end{array}}{\overset{1}{}}}f_{\mathrm{}+m+n}_n_m+kN\delta _{\mathrm{},N}_N,\mathrm{}N,`$
where $`_m=/f_m`$.
We now go over to the dual formulation, where the problem, similarly to Sec. 5.3, essentially reduces to finding the action of the generators on the space of symmetric polynomials. The dual generators are given by
(5.64) $`(f_{\mathrm{}}^{}\varphi )(x_1,x_2,\mathrm{},x_{M1})`$ $`={\displaystyle \frac{1}{(\mathrm{}1)!}}{\displaystyle \frac{^\mathrm{}1}{x_M^\mathrm{}1}}\varphi (x_1,x_2,\mathrm{},x_M)|_{x_M=0},`$
(5.65) $`(h_{\mathrm{}}^{}\varphi )(x_1,x_2,\mathrm{},x_M)`$ $`={\displaystyle \underset{i=1}{\overset{M}{}}}x_i^{\mathrm{}}\varphi (x_1,x_2,\mathrm{},x_M),`$
(5.66) $`(e_{\mathrm{}}^{}\varphi )(x_0,x_1,x_2,\mathrm{},x_M)`$ $`={\displaystyle \underset{0i<jM}{}}F_{\mathrm{}}(x_i,x_j)\left(\varphi \left([𝒙]_i\right)+\varphi \left([𝒙]_j\right)\right)+`$
$`+kN\delta _{\mathrm{},N}{\displaystyle \underset{0iM}{}}x_i^{N1}\varphi \left([𝒙]_i\right)`$
with $`F_{\mathrm{}}`$ defined in (5.33).
As for the $`N=2`$ algebra, the formula for $`e_{\mathrm{}}^{}`$ was obtained by adding some terms from the ideal $`𝒥(N)`$ generated by the monomials $`x_1^{i_1}\mathrm{}x_n^{i_n}`$ in which $`i_jN`$ for at least one $`j`$; in the algebra of symmetric polynomials, this ideal is generated by the polynomials $`p_j`$ with $`jN`$ (see (5.25)). It must be established that the action of $`e_{\mathrm{}}^{}`$ does not violate the vanishing conditions on $`(k+1)`$-diagonals,
(5.67)
$$\varphi (\underset{k+1}{\underset{}{x,x,\mathrm{},x}},x_{k+2},\mathrm{})=0,$$
or, more precisely, that the polynomial $`e_{\mathrm{}}^{}\varphi `$ satisfies these conditions modulo the ideal $`𝒥(N)`$. To calculate the right-hand side of (5.66) for $`x_0=x_1=\mathrm{}=x_k`$, we split the double sum in (5.66) as
(5.68)
$$\begin{array}{c}\underset{0i<jM}{}F_{\mathrm{}}(x_i,x_j)\left(\varphi \left([𝒙]_i\right)+\varphi \left([𝒙]_j\right)\right)=\hfill \\ \hfill =\left(\underset{0i<jk}{}+\underset{0ik<j}{}+\underset{k<i<jM}{}\right)F_{\mathrm{}}(x_i,x_j)\left(\varphi \left([𝒙]_i\right)+\varphi \left([𝒙]_j\right)\right).\end{array}$$
The third sum is readily seen to vanish on the diagonal in view of (5.67). Inserting $`x_0=x_1=\mathrm{}=x_k=x`$ into the second sum, we obtain terms of the form
(5.69)
$$\frac{x^{\mathrm{}}x_j^{\mathrm{}}}{xx_j}\varphi (\underset{k}{\underset{}{x,x,\mathrm{},x}},x_{k+1},\mathrm{}).$$
Again in view of (5.67), however, $`xx_j`$ is a divisor of the polynomial $`\varphi (\underset{k}{\underset{}{x,x,\mathrm{},x}},x_{k+1},\mathrm{})`$, and therefore each of these terms is in $`𝒥(N)`$ (we recall that $`\mathrm{}N`$).
It remains to consider the first sum in (5.68). In this case, $`F_{\mathrm{}}(x,x)=\mathrm{}x^\mathrm{}1`$. Each term in the sum lies in the ideal for $`\mathrm{}>N`$, but for $`\mathrm{}=N`$, we have the terms involving $`x^\mathrm{}1`$, which is not in the ideal. However, these terms can be summed up into the expression
(5.70)
$$2N\frac{k(k+1)}{2}x^{N1}\varphi (\underset{k}{\underset{}{x,x,\mathrm{},x}},x_{k+1},\mathrm{}),$$
which is precisely canceled by the term
(5.71)
$$+k(k+1)N\delta _{N,N}x^{N1}\varphi (\underset{k}{\underset{}{x,x,\mathrm{},x}},x_{k+1},\mathrm{}),$$
which arises from the second term in the formula for $`e_N^{}`$, Eq. (5.66). This finishes the proof that operators (5.64)–(5.66) preserve relations (5.67) and therefore have a well-defined action on $`𝖬_{r,k}^+[\iota ]^{}`$. Hence, the action is well defined on $`𝖬_{r,k}^+[\iota ]`$ in that it preserves the ideal generated by the left-hand sides of (5.61).
We note that the key role in the above cancellation is played by the coefficient $`k`$ in front of the second term in (5.66). The same coefficient becomes the level of the $`\widehat{s\mathrm{}}(2)`$ representation thereby constructed. A combination of these facts ensures the existence of the $`\widehat{s\mathrm{}}(2)`$ action.
The demonstration of the $`\widehat{s\mathrm{}}(2)`$ action on the semi-infinite space is now completed following the same strategy as in Sec. 5.4. One first verifies that the action of positive modes defined on different subspaces $`𝖬_{r,k}^+[\iota ]`$ agrees with the embeddings in (5.60). The spectral flow then allows one to define the action of negative modes (which again involves the argument that along with (5.60), there exist similar filtrations with $`𝖬_{r,k}^+[\iota ]`$ replaced by the spaces $`𝖬_{r,k}^{(\mu )}[\iota ]`$ generated by the modes $`f_n`$ with $`n\mu `$). To complete the proof, it only remains to verify the Serre relations. For this, one first establishes that the left-hand sides of the Serre relations commute with all the modes $`f_n`$ for $`n\mu `$, and because the modes $`f_n`$ generate the entire semi-infinite space, it then follows that the Serre relations are satisfied once they are satisfied on a particular extremal state, which can be verified directly. The Serre relations guarantee that the algebra action constructed is the action of the $`\widehat{s\mathrm{}}(2)`$ algebra.
### 5.6. The isomorphism with the representation $`𝔎_{r,p;\theta }`$
To continue with the $`N=2`$ algebra story, it remains to show that the semi-infinite space, which we now know to be an $`N=2`$ module, is isomorphic to a unitary $`N=2`$ representation.
There is a mapping
(5.72)
$$𝖶_{r,p;\theta }𝔎_{r,p;\theta },$$
obtained by identifying each state (3.1) with the corresponding extremal state in $`𝔎_{r,p;\theta }`$ defined in Sec. 2.4 (and obviously, identifying each $`𝒢_n`$ with the corresponding $`N=2`$ algebra generator acting in the unitary representation). Obviously, these mappings commute with the action of $`𝒢_n`$ and $`𝖴_{\pm p}`$.
###### Theorem 5.5.
Mapping (5.72) is an isomorphism of $`N=2`$ modules.
Proof. We have the $`N=2`$ module $`𝖶_{r,p;\theta }`$, in which relations (1.1) are satisfied. In this case, it is easiest to use the equivalence of the $`N=2`$ and $`\widehat{s\mathrm{}}(2)`$ representation categories . Applying the functor to $`𝖶_{r,p;\theta }`$, we obtain an $`\widehat{s\mathrm{}}(2)`$-module from the category $`𝒪`$ (more precisely, the spectral-flow orbit, whose length is equal to $`2`$ on integrable representations), in which conditions (4.44) are satisfied. Any such $`\widehat{s\mathrm{}}(2)`$-module is a direct sum of integrable (unitary) modules,<sup>8</sup><sup>8</sup>8We recall that an $`\widehat{s\mathrm{}}(2)`$ module is integrable if it decomposes into a sum of finite-dimensional representations with respect to any of its $`\widehat{s\mathrm{}}(2)`$ subalgebras generated by $`e_i`$ and $`f_i`$ (it actually suffices to establish this decomposition for two such subalgebras); it is this property that follows from (4.44.
(5.73)
$$𝖥(𝖶_{r,p;})=\underset{\alpha }{}𝔏_{(\alpha )}.$$
Applying the inverse functor, we obtain another sum of representations,
(5.74)
$$𝖶_{r,p;\theta }=\underset{\beta }{}𝔎_{(\beta )},$$
where each module in the right-hand side is necessarily a unitary $`N=2`$ representation, i.e., some $`𝔎_{r^{},p;\theta ^{}}`$, and the statement of the theorem follows by comparing the eigenvalues of $`_0`$ and $`_0`$.
It follows from Theorem 5.5 that the expression in (4.36) coincides with the corresponding unitary $`N=2`$ character. Comparing this with the known expression for the same character (see Sec. 2.2), we obtain the combinatorial identity in the corollary of Theorem 1.1 (see Sec. 1).
As another corollary, the “semi-infinite” expression for the unitary $`N=2`$ characters gives a formula for the string functions $`C_{r,p}^a(q)`$ read off from representing the $`N=2`$ characters as in (2.13). To obtain this representation, we rewrite Eq. (4.38) (where we can set $`\theta =0`$) by splitting the summation over $`n`$ as
(5.75)
$$\underset{n}{}f(n)=\underset{\mathrm{}}{}\underset{a=0}{\overset{p3}{}}f\left((p2)\mathrm{}+a\right).$$
We then shift each of the summation variables $`N_m`$ as $`N_mN_m+\mathrm{}`$, which does not change the factors $`(q)_{N_mN_{m+1}}`$. The summation over $`\mathrm{}`$ can therefore be performed, with the result that the “semi-infinite” formula for the character takes the form (2.13) with the string function
(5.76)
$$C_{r,p}^a(q)=\underset{\begin{array}{c}N_1\mathrm{}N_{p2}\\ N_1+\mathrm{}+N_{p2}=a\end{array}}{}\frac{q^{_{m=1}^{p2}N_m^2+_{m=r}^{p2}N_m}}{(q)_{N_1N_2}(q)_{N_2N_3}\mathrm{}(q)_{N_{p3}N_{p2}}(q)_{\mathrm{}}}.$$
## 6. Some related constructions
### 6.1. Positive bases: paths and the generalized Pascal triangles
Filtration (5.1) allows us to rewrite a representative of any state in the semi-infinite space $`𝖶_{r,p;\theta }`$ through only nonnegative modes $`𝒢_{n0}`$. The next interesting problem is to construct a basis in each term of this filtration. We construct such bases in the finite-dimensional subspaces $`𝖬_{r,k}^+[\iota ]`$ (see (5.60)) of the $`\widehat{s\mathrm{}}(2)`$ semi-infinite space using some kind of a “Demazure induction” (cf. ). The combination of bases in the corresponding $`𝖬_{r,k}^+[\iota ]`$ spaces gives a basis in the $`N=2`$ space $`𝖶_{r,p;\theta }^+[\iota ]`$.
#### 6.1.1. Positive bases in unitary $`\widehat{s\mathrm{}}(2)`$ modules
We consider the filtration by finite-dimensional subspaces in Eq. (5.60). Basis vectors in $`𝖬_{r,k}^+[\iota ]`$ are in a $`1:1`$ correspondence with paths on a rectangular lattice whose construction we now explain. We label the horizontal lines of the lattice by $`\mathrm{}_0,\mathrm{}_1,\mathrm{}`$ such that $`\mathrm{}_0`$ corresponds to the bottom. On each line $`\mathrm{}_m`$, we label the sites as $`\mathrm{}_{m,n}`$ with $`n0`$. We assign the site $`\mathrm{}_{0,0}`$ the symbol $`[\underset{\iota +1}{\underset{}{k+1,\mathrm{},k+1}},r]`$. A path ending at $`\mathrm{}_{m,n}`$ is a connected sequence of links between $`\mathrm{}_{0,0}`$ and $`\mathrm{}_{m,n}`$ (see Fig. 3), with each site along the path assigned an $`[\mathrm{}]`$ symbol as follows. If the site $`\mathrm{}_{m,n}`$ on the path is already assigned a symbol $`[a_1,\mathrm{},a_\nu ]`$, where $`a_1a_2\mathrm{}a_\nu `$, the path can be continued from this site via one of the following two steps whenever the corresponding conditions are satisfied:
1. Moving to the site $`\mathrm{}_{m+1,n}`$ and assigning it the symbol $`[a_2,\mathrm{},a_\nu ]`$ (provided $`\nu 2`$).
2. Moving to the site $`\mathrm{}_{m,n+1}`$ and assigning it the symbol $`[a_1,\mathrm{},a_{\lambda 1},a_\lambda 1,a_{\lambda +1},\mathrm{},a_\nu ]`$ if $`a_\lambda 2`$, where $`\lambda `$ is defined by the conditions $`a_1=\mathrm{}=a_\lambda >a_{\lambda +1}`$ and $`\lambda =\nu `$ if $`a_1=\mathrm{}=a_\nu 2`$.
We let $`\mathrm{}`$ and $``$ denote the link created by taking the respective steps 1 and 2. More precisely, we write $`_m`$ for each $``$ link between any two sites on $`\mathrm{}_m`$. We note that different paths assign different $`[\mathrm{}]`$ symbols to the same site.
A path ending at $`\mathrm{}_{m,n}`$ is said to be admissible if it cannot be continued by step 1 from $`\mathrm{}_{m,n}`$ (it may or may not be continued via step 2).
Basis vectors in $`𝖬_{r,k}^+[\iota ]`$ are in a $`1:1`$ correspondence with the admissible paths. The path consisting of only the $`\mathrm{}`$ links corresponds to the twisted highest-weight vector $`|r,k|\iota _{\widehat{s\mathrm{}}(2)}`$ (see (4.45) and (4.46)). For any other admissible path, let $`_{j_0},_{j_1},\mathrm{},_{j_K}`$ (where $`0j_0\mathrm{}j_K`$) be its $``$ links. The corresponding basis vector is given by
(6.1)
$$f_{j_0}f_{j_1}\mathrm{}f_{j_K}|r,k|\iota _{\widehat{s\mathrm{}}(2)}𝖬^+_{r,k}[\iota ].$$
This construction is based on the fact that
(6.2)
$$𝖬_{r,k}^+[\iota ]𝖦𝗋\left(\underset{\iota +1}{\underset{}{^{k+1}\mathrm{}^{k+1}}}^r\right)[k+1,\mathrm{},k+1,r],$$
where $`𝖦𝗋`$ means taking the graded object associated with a filtration existing on the tensor product. Traveling along the paths then corresponds to “traveling” through the tensor product factors. The last formula can also be viewed as an explanation of the square-bracket notation. For the representations with $`r=1`$, the last tensor factor and, correspondingly, ‘$`1`$’ in $`[k+1,\mathrm{},k+1,1]`$ can be dropped. Indeed, it is easy to see that replacing the starting symbol $`[\underset{\iota +1}{\underset{}{k+1,\mathrm{},k+1}},1]`$ with $`\left[\underset{\iota +1}{\underset{}{k+1,\mathrm{},k+1}}\right]`$ does not change the resulting vectors (6.1).
In Fig. 4, we consider the example where $`k=2`$, $`r=2`$, and $`\iota =1`$, and the origin of paths $`\mathrm{}_{0,0}`$ is therefore assigned $`[3,3,2]`$. The basis vectors read off from the admissible paths in accordance with (6.1) are given by
(6.3)
$$\begin{array}{ccccc}f_2& f_1f_2& f_1^3& f_0^2f_1^2& f_0^5\\ f_1& f_0f_2& f_0f_1^2& f_0^3f_1\\ f_0& f_1^2& f_0^2f_1& f_0^4\\ & f_0f_1& f_0^3\\ & f_0^2& f_0^2f_2\end{array}$$
acting on the twisted highest-weight state $`|2,2|\mathrm{\hspace{0.17em}1}_{\widehat{s\mathrm{}}(2)}`$.
#### 6.1.2. “Positive” characters
The space $`𝖬_{r,k}^+[\iota ]`$ is graded by the number of modes $`f_i`$ applied to the twisted highest-weight vector,
(6.4)
$$𝖬_{r,k}^+[\iota ]=\underset{j=0}{\overset{r+2k\iota }{}}𝖬_{r,k}^+[\iota ;j],$$
where each $`𝖬_{r,k}^+[\iota ;j]`$ is generated from $`|r,k|\iota _{\widehat{s\mathrm{}}(2)}`$ by precisely $`j`$ operators $`f_{}`$. The dimensions of $`𝖬_{r,k}^+[\iota ;j]`$ are arranged into a generalized Pascal triangle (cf. ).
In the generalized Pascal triangle labeled by $`k`$ and $`r`$, the top row consists of $`r`$ units, and each element in the $`i`$th row is the sum of $`k+1`$ elements of the $`(i1)`$th row: for even $`k`$, the sum includes the element above the chosen one and $`(k+2)/21`$ of its neighbours on each side; for odd $`k`$, it runs over $`(k+1)/2`$ elements “north-west” and $`(k+1)/2`$ elements “north-east” of the chosen one. For $`k=1`$, both cases $`r=1`$ and $`r=2`$ reduce to the standard Pascal triangles (starting with $`1`$ and $`1,1`$ in the top row respectively); for $`k=2`$, the triangles are shown in Fig. 5. The dimension of $`𝖬_{r,k}^+[\iota ;j]`$ is read of from the $`j`$th entry of the $`2\iota `$th row in the Pascal triangle with the parameters $`k`$ and $`r`$.
Each space $`𝖬_{r,k}^+[\iota ;j]`$ is graded by the sum of modes of $`f_i`$. The character of $`𝖬_{r,k}^+[\iota ]`$ can be written through the $`q`$-supernomial coefficients $`\left(𝐋_k\genfrac{}{}{0.0pt}{}{a}{}\right)_q`$, which are defined by the generating functions
(6.5)
$$\begin{array}{c}T_{𝐋_k}(z,q)=\underset{a=0}{\overset{\mathrm{}}{}}z^a\left(𝐋_k\genfrac{}{}{0.0pt}{}{a}{}\right)_q=\hfill \\ \hfill =\underset{L_1N_1,L_2+N_1N_2,\mathrm{},L_k+N_{k1}N_k0}{}z^{_{i=1}^kN_i}q^{_{i=1}^{k1}N_{i+1}((_{j=1}^iL_j)N_i)}\times \\ \hfill \times \left[\begin{array}{c}L_1\\ N_1\end{array}\right]_q\left[\begin{array}{c}L_2+N_1\\ N_2\end{array}\right]_q\left[\begin{array}{c}L_3+N_2\\ N_3\end{array}\right]_q\mathrm{}\left[\begin{array}{c}L_k+N_{k1}\\ N_k\end{array}\right]_q,\end{array}$$
where $`𝐋_k=(L_1,L_2,\mathrm{},L_k)`$ is a $`k`$-dimensional vector with nonnegative integer entries and we use the standard notation
(6.6)
$$\left[\begin{array}{c}n\\ m\end{array}\right]_q=\{\begin{array}{cc}\frac{(q)_n}{(q)_{nm}(q)_m},\hfill & nm0,\hfill \\ 0,\hfill & \text{otherwise}.\hfill \end{array}$$
The supernomial coefficients are generating functions of partitions admitting Durfee dissection with the defects $`(L_1,L_2,\mathrm{},L_k)`$ .
The character of $`𝖬_{r,k}^+[\iota ]`$ is given by
(6.7)
$$char𝖬_{r,k}^+[\iota ](z,q)=z^{k\iota r+1}q^{k\iota ^2+\iota (r1)}T_{𝐋_k}(z,q^1),$$
where the $`k`$-dimensional vector $`𝐋_k`$ is chosen as $`𝐋_k=(2\iota ,0,0,\mathrm{},0,\underset{r1}{\underset{}{1,0,\mathrm{},0}})`$. After some algebra, Eq. (6.7) can be rewritten through $`q`$-binomial coefficients as
(6.8)
$$\begin{array}{c}char𝖬_{r,k}^+[\iota ](z,q)=\underset{\begin{array}{c}N_1,N_2,\mathrm{},N_k\\ \iota N_1N_2\mathrm{}N_k\iota 1\\ N_i\iota \text{for}ik+1r\end{array}}{}z^{N_1+N_2+\mathrm{}+N_k}q^{_{m=1}^kN_m^2+_{m=k+2r}^kN_m}\times \hfill \\ \hfill \times \left[\begin{array}{c}2\iota \\ \iota +N_1\end{array}\right]_q\left[\begin{array}{c}\iota +N_1\\ \iota +N_2\end{array}\right]_q\mathrm{}\left[\begin{array}{c}\iota +N_{kr}\\ \iota +N_{k+1r}\end{array}\right]_q\times \\ \hfill \times \left[\begin{array}{c}\iota +N_{k+1r}+1\\ \iota +N_{k+2r}+1\end{array}\right]_q\left[\begin{array}{c}\iota +N_{k+2r}+1\\ \iota +N_{k+3r}+1\end{array}\right]_q\mathrm{}\left[\begin{array}{c}\iota +N_{k1}+1\\ \iota +N_k+1\end{array}\right]_q.\end{array}$$
The character in (6.8) is normalized such that the highest-weight vector $`|r,k|0_{\widehat{s\mathrm{}}(2)}`$ is in the bigrade $`(0,0)`$.
#### 6.1.3. Positive bases and characters of $`𝖶_{r,p;\theta }^+[\iota ]`$
For the $`N=2`$ spaces $`𝖶_{r,p;\theta }^+[\iota ]`$ in Eq. (5.1), we use the above construction to build a basis as follows. The space $`𝖶_{r,p;\theta }^+[\iota ]`$ is graded by the number of the $`𝒢_{}`$ operators,
(6.9)
$$𝖶_{r,p;\theta }^+[\iota ]=\underset{0j\frac{r1+(p2)(p\iota +\theta )}{p1}}{}𝖶_{r,p;\theta }^+[\iota ;j],$$
where $`𝖶_{r,p;\theta }^+[\iota ;j]`$ is the space obtained by applying $`j`$ operators $`𝒢_{}`$ to the extremal vector $`|r,p;\theta |\iota _{\mathrm{}/2}`$. There exists an isomorphism of vector spaces
(6.10)
$$𝖶_{r,p;\theta }^+[\iota ;j]𝖬_{r,p2}^+[p\iota +\theta j;j]$$
induced by mapping the basis elements of $`𝖬_{r,p2}^+[p\iota +\theta j;j]`$ as
(6.11)
$$\begin{array}{c}(f_0)^{i_0}(f_1)^{i_1}\mathrm{}(f_j)^{i_j}𝒢_0𝒢_1\mathrm{}𝒢_{i_01}𝒢_{i_0+1}𝒢_{i_0+2}\mathrm{}𝒢_{i_0+i_1}𝒢_{i_0+i_1+2}\mathrm{}\times \hfill \\ \hfill \times \mathrm{}𝒢_{i_0+i_1+\mathrm{}+i_{j1}+j}\mathrm{}𝒢_{i_0+i_1+\mathrm{}+i_{j1}+i_j+j1}.\end{array}$$
This gives a basis in $`𝖶_{r,p;\theta }^+[\iota ]`$.
In accordance with (6.10), the dimensions of $`𝖶_{r,p;\theta }^+[\iota ;j]`$ can also be read off from the generalized Pascal triangles (see an example in Fig. 6). These dimensions are given by the generalized Fibonacci numbers $`F_{p\iota +\theta }^{r,p}`$ satisfying the defining relations $`F_i^{r,p}=F_{i1}^{r,p}+F_{i2}^{r,p}+\mathrm{}+F_{i(p1)}^{r,p}`$.
We conjecture a character formula for $`𝖶_{r,p;\theta }^+[N]`$. We normalize the analogue of (6.8) such that the state $`|r,p;\theta |0_{\mathrm{}/2}`$ is in the grade $`(0,0)`$.
###### Conjecture 6.1.
The characters of the subspaces involved in the positive filtration of unitary $`N=2`$ representations are given by
(6.12)
$$char𝖶_{r,p;\theta }^+[\iota ](z,q)=z^{(p2)\iota r+1}q^{\frac{p(p2)\iota (\iota +1)}{2}\iota (\frac{p(p1)}{2}r\theta (p2))}S_{𝐋_p}(z,q^1),$$
where
(6.13)
$$S_{𝐋_p}(z,q)=\underset{a=0}{\overset{\mathrm{}}{}}z^aq^{\frac{a^2a}{2}}\left(\begin{array}{c}𝐋_p(a,0,\mathrm{},0)\\ a\end{array}\right)_q,$$
with $`𝐋_p=(p\iota +\theta ,0,0,\mathrm{},0,\underset{r1}{\underset{}{1,0,\mathrm{},0}})`$.
These characters can be rewritten as
(6.14)
$$\begin{array}{c}char𝖶_{r,p;\theta }^+[\iota ](z,q)=\underset{\begin{array}{c}N_1,N_2,\mathrm{},N_{p2}\\ N_1+n\iota +\theta r+1,N_1N_2\mathrm{}N_{p2}\iota 1\\ N_i\iota \text{for}ipr1\end{array}}{}z^nq^{\frac{n^2n}{2}\theta n+_{m=1}^{p2}N_m^2+_{m=r}^{p2}N_m}\times \hfill \\ \hfill \times \left[\begin{array}{c}2\iota +\theta r+1n\\ \iota +N_1\end{array}\right]_q\left[\begin{array}{c}\iota +N_1\\ \iota +N_2\end{array}\right]_q\mathrm{}\left[\begin{array}{c}\iota +N_{pr2}\\ \iota +N_{pr1}\end{array}\right]_q\times \\ \hfill \times \left[\begin{array}{c}\iota +N_{pr1}+1\\ \iota +N_{pr}+1\end{array}\right]_q\left[\begin{array}{c}\iota +N_{pr}+1\\ \iota +N_{pr+1}+1\end{array}\right]_q\mathrm{}\left[\begin{array}{c}\iota +N_{p3}+1\\ \iota +N_{p2}+1\end{array}\right]_q,\end{array}$$
where $`n=_{m=1}^{p2}N_m`$. It is easy to verify that this expression has the correct limit as $`\iota \mathrm{}`$,
(6.15)
$$\underset{\iota \mathrm{}}{lim}\left(z^{\frac{1r+2\theta }{p}\theta }q^{\frac{1}{2}(1\frac{2}{p})(\theta ^2\theta )+\theta \frac{r1}{p}}char𝖶_{r,p;\theta }^+[\iota ](z,q)\right)=char𝖶_{r,p;\theta }(z,q),$$
where the character $`char𝖶_{r,p;\theta }(z,q)`$ is given by (4.36).
### 6.2. The $`N=2`$ modular functor and functions on Riemann surfaces
The representation of the unitary modules via semi-infinite forms implies a relation between the $`N=2`$ modular functor and the spaces of skew-symmetric functions with prescribed singularities on Cartesian powers of a genus-$`g`$ Riemann surface.
#### 6.2.1. $`N=2`$ correlation functions in the semi-infinite picture
Let $`_n^g`$ be a genus-$`g`$ Riemann surface with $`n`$ marked points $`P_1,\mathrm{},P_n`$ and let $`(N=2)^{\text{out}}`$ denote the algebra generated by the part of the $`N=2`$ currents $`𝒢(z)`$, $`𝒬(z)`$$`(z)`$, and $`(z)`$ that is holomorphic outside the points $`P_1,\mathrm{},P_n`$. One then defines the space of coinvariants
(6.16)
$$=\frac{𝔎_{r_1,p;\theta _1}𝔎_{r_2,p;\theta _2}\mathrm{}𝔎_{r_n,p;\theta _n}}{(N=2)^{\text{out}}𝔎_{r_1,p;\theta _1}𝔎_{r_2,p;\theta _2}\mathrm{}𝔎_{r_n,p;\theta _n}},$$
where $`𝔎_{r_i,p;\theta _i}`$ are unitary representations and $`𝔎_{r_1,p;\theta _1}𝔎_{r_2,p;\theta _2}\mathrm{}𝔎_{r_n,p;\theta _n}`$ is a representation of the algebra $`(N=2)_{P_1}\mathrm{}(N=2)_{P_n}`$. Similarly to , this space of coinvariants can be shown to be isomorphic to the space of coinvariants with respect to the algebra generated by $`𝒢(z)`$ and $`_0`$,
(6.17)
$$=\left(\frac{𝔎_{r_1,p;\theta _1}𝔎_{r_2,p;\theta _2}\mathrm{}𝔎_{r_n,p;\theta _n}}{𝔤^{\text{out}}𝔎_{r_1,p;\theta _1}𝔎_{r_2,p;\theta _2}\mathrm{}𝔎_{r_n,p;\theta _n}}\right)^0,$$
where $`𝔤^{\text{out}}`$ is the algebra generated by the part of the $`𝒢(z)`$ current that is holomorphic outside $`P_1,\mathrm{},P_n`$ and $`()^0`$ denotes the restriction to the zero-charge component (the zero-charge restriction comes from taking the coinvariants with respect to $`_0`$).
For a given set of representations $`𝔎_{r_1,p;\theta _1},\mathrm{},𝔎_{r_n,p;\theta _n}`$ placed at the points $`P_1`$,…, $`P_n`$ on the Riemann surface and for fixed $`m`$ and $`\iota _1,\mathrm{},\iota _n`$, each linear functional $`F|:`$ defines the function of of $`(x_1,\mathrm{},x_m)_n^g\times \mathrm{}\times _n^g`$ given by
(6.18)
$$F|𝒢(x_1)𝒢(x_2)\mathrm{}𝒢(x_m)|\mathrm{\Phi }_{r_1,\theta _1}^{\iota _1}(P_1)\mathrm{\Phi }_{r_2,\theta _2}^{\iota _2}(P_2)\mathrm{}\mathrm{\Phi }_{r_n,\theta _n}^{\iota _n}(P_n)_p^g,$$
where $`\mathrm{\Phi }_{r,\theta }^\iota `$ are the operators corresponding to extremal vectors (3.1). In accordance with the chosen conformal dimension of $`𝒢`$, expression (6.18) is a 2-differential in each variable. We rewrite this expression in local coordinates by separating the correlation function,
(6.19)
$$\begin{array}{cc}& F_{r_1,r_2,\mathrm{},r_n;\theta _1,\theta _2,\mathrm{},\theta _n}^{p,g;\iota _1,\mathrm{},\iota _n}(x_1,x_2,\mathrm{},x_mP_1,P_2,\mathrm{},P_n)(dx_1)^2(dx_2)^2\mathrm{}(dx_m)^2=\hfill \\ & =F|𝒢(x_1)𝒢(x_2)\mathrm{}𝒢(x_m)\mathrm{\Phi }_{r_1,\theta _1}^{\iota _1}(P_1)\mathrm{\Phi }_{r_2,\theta _2}^{\iota _2}(P_2)\mathrm{}\mathrm{\Phi }_{r_n,\theta _n}^{\iota _n}(P_n)_p^g.\hfill \end{array}$$
The restriction to the zero charge component takes the form of the constraint
(6.20)
$$mp+n\left(r_1+(p2)(\theta _1+p\iota _1)\right)\mathrm{}\left(r_n+(p2)(\theta _n+p\iota _n)\right)=0.$$
It follows that the functions $`F_{r_1,r_2,\mathrm{},r_n;\theta _1,\theta _2,\mathrm{},\theta _n}^{p,g;\iota _1,\mathrm{},\iota _n}`$ defined in (6.19) are antisymmetric in $`x_1,\mathrm{},x_m`$, are regular on $`_n^g\times \mathrm{}\times _n^g`$ except at the points $`P_i`$, and possess the following properties.
1. On each $`(p1)`$-diagonal $`x_{i_1}=x_{i_2}=\mathrm{}=x_{i_{p1}}`$, one has
(6.21)
$$\frac{^{p2}}{x_{i_1}^{p2}}\frac{^{p3}}{x_{i_2}^{p3}}\mathrm{}\frac{}{x_{i_{p2}}}F_{r_1,r_2,\mathrm{},r_n;\theta _1,\theta _2,\mathrm{},\theta _n}^{p,g;\iota _1,\mathrm{},\iota _n}|_{x_{i_1}=x_{i_2}=\mathrm{}=x_{i_{p1}}}=0.$$
2. For each $`a`$ such that $`a<r_j`$, the function
(6.22)
$$\frac{^{a1}}{x_{i_a}^{a1}}\frac{^{a2}}{x_{i_{a1}}^{a2}}\mathrm{}\frac{}{x_{i_2}}F_{r_1,r_2,\mathrm{},r_n;\theta _1,\theta _2,\mathrm{},\theta _n}^{p,g;\iota _1,\mathrm{},\iota _n}$$
vanishes at $`x_{i_1}=x_{i_2}=\mathrm{}=x_{i_a}=xP_j`$ with an order not less than
(6.23)
$$a(\theta _j+\iota _jp)+\frac{a(a3)}{2}.$$
For each $`a`$ such that $`r_jap2`$, the function
(6.24)
$$\frac{^a}{x_{i_a}^a}\mathrm{}\frac{^{r+1}}{x_{i_{r+1}}^{r+1}}\frac{^r}{x_{i_r}^r}\frac{^{r2}}{x_{i_{r1}}^{r2}}\frac{^{r1}}{x_{i_{r2}}^{r1}}\mathrm{}\frac{}{x_{i_2}}F_{r_1,r_2,\mathrm{},r_n;\theta _1,\theta _2,\mathrm{},\theta _n}^{p,g;\iota _1,\mathrm{},\iota _n}$$
vanishes at $`x_{i_1}=x_{i_2}=\mathrm{}=x_{i_a}=xP_j`$ with the order not less than
(6.25)
$$a(\theta _j+\iota _jp)+\frac{a(a4r_j+3)}{2}(r_j+1)(r_j1)$$
(negative-order zeros are poles). Condition (6.21) follows from the vanishing of (1.1), and (6.23) and (6.25) from relations (3.2)–(3.4).
We let $`_{r_1,r_2,\mathrm{},r_n;\theta _1,\theta _2,\mathrm{},\theta _n}^{p,g;\iota _1,\mathrm{},\iota _n}(m)`$ denote the space of all functions satisfying these properties; with Eq. (6.20) assumed to be satisfied, the notation is somewhat redundant; however, it is useful to keep $`m`$ as a free parameter and determine some other label from (6.20).
###### Theorem 6.2.
For sufficiently large $`m`$, the assignment
(6.26)
$$F|F_{r_1,r_2,\mathrm{},r_n;\theta _1,\theta _2,\mathrm{},\theta _n}^{p,g;\iota _1,\mathrm{},\iota _n}(,\mathrm{},P_1,P_2,\mathrm{},P_n)$$
defined in (6.19) establishes an isomorphism between $`^{}`$ and $`_{r_1,r_2,\mathrm{},r_n;\theta _1,\theta _2,\mathrm{},\theta _n}^{p,g;\iota _1,\mathrm{},\iota _n}(m)`$. For these $`m`$, the dimensions $`d_{r_1,r_2,\mathrm{},r_n;\theta _1,\theta _2,\mathrm{},\theta _n}^{p,g}=dim_{r_1,r_2,\mathrm{},r_n;\theta _1,\theta _2,\mathrm{},\theta _n}^{p,g;\iota _1,\mathrm{},\iota _n}(m)`$ are independent of $`m`$ and $`\iota _1,\mathrm{},\iota _n`$.
The semi-infinite construction is essential in proving that all functions with the properties described above are the $`N=2`$ correlation functions. We do not give the proof here and consider only two examples for illustration. One example is a simple enumeration of functions on a single-punctured elliptic curve with the vacuum representation. The dimension of the space of these functions gives the number of primary fields in the corresponding minimal model. For $`p=3`$ and $`p=4`$, we explicitly construct a basis in the functional space. The second example is with a three-punctured Riemann sphere, where the dimensions of the corresponding functional spaces are related to the $`N=2`$ fusion algebra; we digress in Sec 6.3 to derive the unitary $`N=2`$ fusion algebra from the $`\widehat{s\mathrm{}}(2)`$ unitary fusion algebra.
#### 6.2.2. Tori with one marked point for small $`p`$
For $`p=3`$, we consider the vacuum module associated with a point on a torus (at the origin in the covering complex plane). Setting $`n=1`$ and $`r_1=1`$ in condition (6.20) then gives $`\iota =m`$. The independence from $`m`$ already occurs starting with $`m=1`$, and the space $`_{1;0}^{3,1}(1)`$ (omitting the $`\iota `$ superscript for brevity) consists of functions on the torus with a pole at zero of an order $`3`$. We then have $`d_{1;0}^{3,1}=3`$, and a basis in the space of such functions is
(6.27)
$$1,\mathrm{}(x),\mathrm{}^{}(x),$$
where
(6.28)
$$\mathrm{}(x)=\frac{1}{x^2}+\underset{\omega \mathrm{\Lambda }}{\overset{}{}}\left(\frac{1}{(x\omega )^2}\frac{1}{\omega ^2}\right)$$
is the Weierstrass function, $`\mathrm{\Lambda }=\{m\omega _1+n\omega _2m,n\}`$, $`\omega _1,\omega _2`$, and $`\mathrm{}(\omega _1/\omega _2)>0`$.
It is instructive to explicitly verify that the same dimension is also obtained for $`m=2`$, i.e., to verify that $`dim_{1;0}^{3,1}(2)=d_{1;0}^{3,1}=3`$, where the calculation is entirely different because the basic condition 1 (see (6.21)) applies in this case. The corresponding space $`_{1;0}^{3,1}(2)`$ consists of antisymmetric functions of two variables $`x_1`$ and $`x_2`$ with a pole of an order $`6`$ as $`x_1=x_20`$ and such that $`\left(f(x_1,x_2)/x_1\right)|_{x_1=x_2}=0`$. This space has a basis
$`f_1(x_1,x_2)=`$ $`\mathrm{}(x_1)^33\mathrm{}(x_1)^2\mathrm{}(x_2)+3\mathrm{}(x_1)\mathrm{}(x_2)^2\mathrm{}(x_2)^3,`$
$`f_2(x_1,x_2)=`$ $`g_2\mathrm{}^{}(x_1){\displaystyle \frac{1}{2}}g_1\mathrm{}(x_1)\mathrm{}^{}(x_1)+\mathrm{}(x_1)^3\mathrm{}^{}(x_1)`$
$`{\displaystyle \frac{3}{2}}g_1\mathrm{}(x_2)\mathrm{}^{}(x_1)+3\mathrm{}(x_2)^3\mathrm{}^{}(x_1)+g_2\mathrm{}^{}(x_2)+{\displaystyle \frac{3}{2}}g_1\mathrm{}(x_1)\mathrm{}^{}(x_2)`$
$`3\mathrm{}(x_1)^3\mathrm{}^{}(x_2)+{\displaystyle \frac{1}{2}}g_1\mathrm{}(x_2)\mathrm{}^{}(x_2)\mathrm{}(x_2)^3\mathrm{}^{}(x_2),`$
$`f_3(x_1,x_2)=`$ $`g_2\mathrm{}(x_1)+g_1\mathrm{}(x_1)^2g_2\mathrm{}(x_2)2\mathrm{}(x_1)^3\mathrm{}(x_2)g_1\mathrm{}(x_2)^2+`$
$`+2\mathrm{}(x_1)\mathrm{}(x_2)^3+\mathrm{}(x_1)\mathrm{}^{}(x_1)\mathrm{}^{}(x_2)\mathrm{}(x_2)\mathrm{}^{}(x_1)\mathrm{}^{}(x_2),`$
where $`g_1=30_{\omega \mathrm{\Lambda }}^{}\frac{1}{\omega ^4}`$ and $`g_2=140_{\omega \mathrm{\Lambda }}^{}\frac{1}{\omega ^6}`$. This recovers the dimension $`d_{1;0}^{3,1}=3`$.
For $`p=4`$, condition (6.20) becomes $`2\iota =m`$. The first “sufficiently large” value of $`m`$ is already $`m=2`$, with $`\iota =1`$. The corresponding space $`_{1;0}^{4,1}(2)`$ consists of antisymmetric functions of two variables with a pole of an order $`4`$ at zero and such that the function $`\left(^2f(x_1,x_2)/x_1^2\right)|_{x_1=x_2=x}`$ has a pole of the order $`9`$ as $`x0`$. A basis in the space of such functions can be chosen as
(6.29)
$$\begin{array}{c}f_1=\mathrm{}(x_1)\mathrm{}(x_2),f_2=\mathrm{}^{}(x_1)\mathrm{}^{}(x_2),f_3=\mathrm{}(x_1)^2\mathrm{}(x_2)^2,\hfill \\ \hfill f_4=\mathrm{}(x_1)\mathrm{}^{}(x_2)\mathrm{}(x_2)\mathrm{}^{}(x_1),f_5=\mathrm{}(x_1)^2\mathrm{}^{}(x_2)\mathrm{}(x_2)^2\mathrm{}^{}(x_1),\\ \hfill f_6=\mathrm{}(x_1)\mathrm{}(x_2)^2\mathrm{}(x_2)\mathrm{}(x_1)^2,\end{array}$$
and we have $`d_{1;0}^{4,1}=6`$.
These examples show that the dimensions of functional spaces coincide with the number of primary fields in the respective minimal model (i.e., with the dimensions of the modular functor for the torus).
### 6.3. The unitary $`N=2`$ fusion algebra
We consider the $`N=2`$ fusion algebra and then discuss its consequences for the functional spaces. The unitary $`N=2`$ fusion rules were derived in from the Verlinde hypothesis. We give an alternative derivation, which starts with the $`\widehat{s\mathrm{}}(2)`$ fusion algebra; this derivation does not rely on the Verlinde theorem statement for the $`N=2`$ algebra, and the coincidence with the result in may have an independent interest.
The fusion algebra $`𝔉_{\widehat{s\mathrm{}}(2)}(k)`$ of the unitary level-$`k`$ $`\widehat{s\mathrm{}}(2)`$ representations $`𝔏_{r,k}`$ is given by
(6.30)
$$𝔏_{r_1,k}\underset{s\mathrm{}\left(2\right)}{\overset{}{}}𝔏_{r_2,k}=\underset{\begin{array}{c}r_3=|r_1r_2|+1\\ \text{step}=2\end{array}}{\overset{k+1|r_1+r_2k2|}{}}𝔏_{r_3,k},1r_ik+1.$$
It is easy to apply the method of to (6.30): tensoring the $`\widehat{s\mathrm{}}(2)`$ modules with the module $`\mathrm{\Omega }`$ over free fermions and applying (2.15), we use $`\widehat{s\mathrm{}}(2)`$ fusion rules and then collect the terms on the right-hand side so as to identify some $`𝔎_{r^{},p;\theta ^{}}`$ representations. This gives the $`N=2`$ fusion algebra
(6.31)
$$𝔎_{r_1,p;\theta _1}\underset{N=2}{\overset{}{}}𝔎_{r_2,p;\theta _2}=\underset{\begin{array}{c}r_3=|r_1r_2|+1\\ \text{step}=2\end{array}}{\overset{p1|r_1+r_2p|}{}}𝔎_{r_3,p;\theta _1+\theta _2+\frac{1}{2}(1r_1r_2+r_3)},1r_1,r_2p1,\theta _1,\theta _2_p.$$
###### Remark 6.3 (the spectral flow).
In the fusion algebra, $`𝔎_{1;0}`$ is the identity and $`\mathrm{\Theta }=𝔎_{p1;0}`$ is the spectral flow operator acting on representations in accordance with (6.31) as
(6.32)
$$𝔎_{p1,p;0}\underset{N=2}{\overset{}{}}𝔎_{r_2,p;\theta }=𝔎_{pr_2,p;\theta +1r_2}𝔎_{r_2,p;\theta +1}.$$
We also have $`\underset{p}{\underset{}{𝔎_{p1,p;0}\underset{N=2}{\overset{}{}}\mathrm{}\underset{N=2}{\overset{}{}}𝔎_{p1,p;0}}}=𝔎_{1,p;0}1`$, as must be the case with the spectral flow on unitary representations.
The above fusion algebra is only for the Neveu–Schwarz sector. To extend the fusion to the Ramond sector, it suffices to add a single element $`𝔎_{1,p;\frac{1}{2}}`$ such that
(6.33)
$$𝔎_{1,p;\frac{1}{2}}\underset{N=2}{\overset{}{}}𝔎_{r,p;\theta }=𝔎_{r,p;\theta +\frac{1}{2}}.$$
This relation completely determines the fusion involving Ramond sector representations. Moreover, the fusion can be formally extended to any other “sector” with the fractional twists $`\theta =\beta /\alpha `$ by adding a single field $`𝔎_{1,p;\frac{1}{\alpha }}`$ such that
(6.34)
$$𝔎_{1,p;\frac{1}{\alpha }}\underset{N=2}{\overset{}{}}𝔎_{r,p;\theta }=𝔎_{r,p;\theta +\frac{1}{\alpha }},$$
which then determines all the fusion rules for $`𝔎_{r,p;\frac{2}{\alpha }+},\mathrm{},𝔎_{r,p;\frac{\alpha 1}{\alpha }+}`$.
###### Remark 6.4.
Fusion rules (6.31) mean that the three-point function is nonvanishing if and only if
(6.35)
$$\begin{array}{c}r+r^{}+r^{\prime \prime }2\theta 2\theta ^{}2\theta ^{\prime \prime }3=0,\hfill \\ \hfill |r^{}r^{\prime \prime }|<r<r^{}+r^{\prime \prime },r+r^{}+r^{\prime \prime }<2p,r+r^{}+r^{\prime \prime }1mod2.\end{array}$$
This agrees with the result in under the Neveu–Schwarz sector correspondence
(6.36)
$$k=r\theta \frac{1}{2},j=\theta +\frac{1}{2},1rp1,0\theta r1$$
between our parameterization of the unitary $`N=2`$ modules and the parameterization used in .
We now show how the correspondence between the $`\widehat{s\mathrm{}}(2)`$ and $`N=2`$ fusion algebras derived above fits into the general scheme between the modular functors expressed by (2.17). In (2.17), we take $``$ to be a torus and recall that there is a basis in the modular functor on the torus whose elements are given by unitary representation characters (such bases, more precisely, depend on a cycle on the torus). The fusion algebra is then an operation on the modular functor for the torus.
It turns out that the correspondence in Eq. (2.17) agrees with natural structures on the modular functors, in the present case, with the fusion algebra. With the modular functors for the torus identified with the respective fusion algebras $`𝔉_𝔞(\kappa )`$ for $`𝔞=\widehat{s\mathrm{}}(2)`$ and $`N=2`$, Eq. (2.17) becomes
(6.37)
$$𝔉_{N=2}(p)=Coinv_{R_1P_1}\left(Inv_{R_2P_2}\left(𝔉_{\widehat{s\mathrm{}}(2)}(p2)𝔉_{\text{free}}(p)\right)\right),$$
where the invariants and coinvariants are taken with respect to the subalgebras generated by the elements $`R_1`$, $`R_2`$, $`P_1`$, and $`P_2`$ explicitly constructed in what follows. The algebra $`𝔉_{\text{free}}(p)`$, which is the fusion algebra for the algebra of vertex operators associated with the lattice $`\sqrt{2p}`$ (see (2.15)), is isomorphic to the group algebra of the cyclic group $`_{2p}`$ (as a linear space, it is represented by theta-functions of the level $`2p=2(k+2)`$, as can be seen from Eq. (2.14) for the characters). It carries an action of the Heisenberg group associated with half-periods, and we let $`P_1`$ and $`P_2`$ denote the corresponding generators (such that $`P_1P_2=(1)^pP_2P_1`$): if $`\nu `$, with $`\nu ^{2p}=1`$, is a generator of $`_{2p}`$, we have
(6.38) $`P_1(\nu ^i)`$ $`=\nu ^{i+p},`$
(6.39) $`P_2(\nu ^i)`$ $`=(1)^i\nu ^i.`$
Using the chosen basis, the Heisenberg group action on the $`\widehat{s\mathrm{}}(2)`$ modular functor can be explicitly described as
(6.40) $`R_1(𝔏_{r,k})=𝔏_{kr+2,k},`$
(6.41) $`R_2(𝔏_{r,k})=(1)^{r1}𝔏_{r,k}`$
(we recall that basis elements are identified with unitary $`\widehat{s\mathrm{}}(2)`$ representations). We note that $`R_2`$ is an automorphism of the fusion algebra and $`R_1`$ is the spectral flow transform realized via fusion, $`R_1(𝔏_{r,k})=𝔏_{k+1,k}\underset{s\mathrm{}\left(2\right)}{\overset{}{}}𝔏_{r,k}`$. It has an important property that (omitting the fusion operation sign) $`(𝔏_{k+1,k}a)(𝔏_{k+1,k}b)=ab`$ for any $`a,b𝔉_{\widehat{s\mathrm{}}(2)}(k)`$. Taking the diagonal action of the Heisenberg group, we now see that $`R_1P_1`$ indeed commutes with $`R_2P_2`$ (i.e., the central element acts identically). In the tensor product $`𝔉_{\widehat{s\mathrm{}}(2)}(p2)𝔉_{\text{free}}(p)`$, we then take the invariants with respect to $`R_2P_2`$, i.e., restrict to the elements of the form $`𝔏_{r,k}\nu ^i`$ with even $`r+i1`$ (see Fig. 7). The construction is then completed by taking coinvariants with respect to the action of $`R_1P_1`$.
We now label the elements of the space constructed on the right-hand side of (6.37) as
(6.42)
$$𝔎_{r,p;\theta }=𝔏_{r,p2}\nu ^{2\theta r+1}modR_1P_1$$
and identify $`𝔎_{r,p;\theta }`$ with the corresponding generator of the $`N=2`$ fusion algebra. The identification under the action of $`R_1P_1`$ means that (2.9) is satisfied by construction; the periodicity under the spectral flow transform by $`p`$ is also obvious. Moreover, calculating
(6.43)
$$(𝔏_{r_1,k}\nu ^{\theta _1})\underset{N=2}{\overset{}{}}(𝔏_{r_2,k}\nu ^{\theta _2})=\left(𝔏_{r_1,p2}\underset{s\mathrm{}\left(2\right)}{\overset{}{}}𝔏_{r_2,p2}\right)\nu ^{\theta _1+\theta _2}$$
and using (6.30), we obtain the unitary $`N=2`$ fusion algebra (6.31). Therefore, the relation between the $`\widehat{s\mathrm{}}(2)`$ and $`N=2`$ fusion algebras can be considered as a particular case of the general relation (2.17) that extends the equivalence of categories to modular functors.
Returning to the functional spaces introduced in Sec. 6.2, we reformulate the $`N=2`$ fusion algebra as a statement on the dimensions of functional spaces. The structure constants of the fusion algebra coincide with the dimensions of the modular functor on the three-punctured $`^1`$. Recalling that this modular functor is related to the functional spaces via Theorem 6.2, we obtain (omitting the $`\iota `$ superscripts for brevity)
###### Proposition 6.5.
For $`mp1`$, the dimensions of the spaces $`_{r,r^{},r^{\prime \prime };\theta ,\theta ^{},\theta ^{\prime \prime }}^{p,0}(m)`$ and $`_{r,r^{};\theta ,\theta ^{}}^{p,0}(m)`$ are given by
(6.44)
$$d_{r,r^{},r^{\prime \prime };\theta ,\theta ^{},\theta ^{\prime \prime }}^{p,0}=dim_{r,r^{},r^{\prime \prime };\theta ,\theta ^{},\theta ^{\prime \prime }}^{p,0}(m)=\{\begin{array}{cc}1,\hfill & \text{conditions (}\text{6.35}\text{) are satisfied},\hfill \\ 0,\hfill & \text{otherwise},\hfill \end{array}$$
and
(6.45)
$$d_{r,r^{};\theta ,\theta ^{}}^{p,0}=dim_{r,r^{};\theta ,\theta ^{}}^{p,0}(m)=\{\begin{array}{cc}1,\hfill & r^{}=r\text{and}\theta ^{}=r\theta 1,\hfill \\ 0,\hfill & \text{otherwise},\hfill \end{array}$$
where $`1r,r^{},r^{\prime \prime }p1`$ and $`0\theta ,\theta ^{},\theta ^{\prime \prime }r1`$.
## 7. Concluding remarks
Possibly the most conspicuous feature of semi-infinite constructions is the asymmetry between the upper-triangular and lower-triangular generators, i.e., for the $`N=2`$ algebra, between the fermions $`𝒢`$ and $`𝒬`$: starting with a module generated by the modes $`𝒢_n`$ subject to the conditions $`^{p2}𝒢(z)\mathrm{}𝒢(z)𝒢(z)=0`$, we could then (with much effort) reconstruct the action of the other algebra generators, including $`𝒬_m`$. However, the relation $`^{p2}𝒬(z)\mathrm{}𝒬(z)𝒬(z)=0`$ is certainly satisfied for the current constructed from the $`𝒬_m`$ generators; it is interesting to investigate consequences of this relation for the corresponding functional spaces.
We expect that the methods developed above for the $`N=2`$ superconformal algebra can also be useful in other semi-infinite constructions. In similar constructs, depending on the chosen constraints (the analogue of the relations $`S_a^p=0`$) a module structure can be found on the semi-infinite space. Even in the case with only one (bosonic or fermionic) field satisfying the constraints, interesting representations of W algebras can thus be obtained. Nontrivial relations between $`\widehat{s\mathrm{}}(2|\mathrm{\hspace{0.17em}1})`$ and $`N=2`$ and $`\widehat{s\mathrm{}}(2)`$ representation theories allow us to expect interesting semi-infinite constructions for a class of $`\widehat{s\mathrm{}}(2|\mathrm{\hspace{0.17em}1})`$ representations.
Acknowledgments. The authors are grateful to J. Figueroa-O’Farrill, A. K. Pogrebkov, I. Yu. Segalov, V. A. Sirota, A. Taormina, and R. Weston for discussions and useful remarks. The work is partially supported by the RFBR Grants 98-01-01155 and 99-01-01169, by the Russian Federation President Grant 99-15-96037, and by INTAS-OPEN-97-1312.
|
warning/0004/math0004002.html
|
ar5iv
|
text
|
# Affine embeddings of homogeneous spacesThis work was supported by INTAS–OPEN–97–1570 and RFBR grant 98–01–00598.
## 1 Introduction.
Let $`G`$ be a connected reductive algebraic group over an algebraically closed field $`\mathrm{𝕜}`$ of characteristic zero and let $`H`$ be an algebraic subgroup of $`G`$. Let us recall that a pointed irreducible algebraic $`G`$-variety $`X`$ is said to be an *embedding* of the homogeneous space $`G/H`$ if the base point of $`X`$ has the dense orbit and stabilizer $`H`$. We shall denote this by $`G/HX`$.
Let $`B`$ be a Borel subgroup of $`G`$. By definition, the *complexity* $`c(X)`$ of a $`G`$-variety $`X`$ is the codimension of a generic $`B`$-orbit in $`X`$ for the restricted action $`B:X`$, see \[Vi1\] and \[LV\]. By Rosenlicht’s theorem, $`c(X)`$ is equal to the transcendence degree of the field $`\mathrm{𝕜}(X)^B`$ of rational $`B`$-invariant functions on $`X`$. A normal $`G`$-variety $`X`$ is called *spherical* if $`c(X)=0`$, or, equivalently, $`\mathrm{𝕜}(X)^B=\mathrm{𝕜}`$. A homogeneous space $`G/H`$ and a subgroup $`HG`$ are said to be spherical if $`G/H`$ is a spherical $`G`$-variety with respect to the natural $`G`$-action.
###### Theorem 1 (Servedio \[Ser\], Luna–Vust \[LV\], Akhiezer \[Akh1\]).
A homogeneous space $`G/H`$ is spherical if and only if each embedding of $`G/H`$ has finitely many $`G`$-orbits.
To be more precise, F. J. Servedio proved that any affine spherical variety contains finitely many $`G`$-orbits, D. Luna, Th. Vust and D. N. Akhiezer extended this result to an arbitrary spherical variety, and D. N. Akhiezer constructed a projective embedding with infinitely many $`G`$-orbits for any homogeneous space of positive complexity.
Let us say that an embedding $`G/HX`$ is *affine* if the variety $`X`$ is affine. In many problems of invariant theory, representation theory and other branches of mathematics, only affine embeddings of homogeneous spaces are considered. Hence for a homogeneous space $`G/H`$ it is natural to ask: does there exist an affine embedding $`G/HX`$ with infinitely many $`G`$-orbits?
Note that a given homogeneous space $`G/H`$ admits an affine embedding if and only if $`G/H`$ is quasiaffine (as an algebraic variety), see \[PV, Th. 1.6\]. In this situation, the subgroup $`H`$ is said to be *observable* in $`G`$. For a description of observable subgroups, see \[Su\], \[PV, Th. 4.18\]. By Matsushima’s criterion, $`G/H`$ is affine iff $`H`$ is reductive. (For a simple proof, see \[Lu1, §2\].) In particular, any reductive subgroup is observable. In the sequel, we suppose that $`H`$ is an observable subgroup of $`G`$. In this paper, we are concerned with the following problem: characterize all quasiaffine homogeneous spaces $`G/H`$ of a reductive group $`G`$ with the property:
(AF) For any affine embedding $`G/HX`$, the number of $`G`$-orbits in $`X`$ is finite.
###### Example 1.
For any spherical quasiaffine homogeneous space, property (AF) holds (Theorem 1).
###### Example 2 (\[Po\]).
Property (AF) holds for any homogeneous space of the group $`SL(2)`$. In fact, here $`dimX3`$, and only a one-parameter family of one-dimensional orbits can appear in $`X(G/H)`$. But $`SL(2)`$ contains no two-dimensional observable subgroups.
###### Example 3.
Let $`T`$ be a maximal torus in $`G`$ and let $`V`$ be a finite-dimensional $`G`$-module. Suppose that a vector $`vV`$ is $`T`$-fixed. Then the orbit $`Gv`$ is closed in $`V`$, see \[Kos\], \[Lu2\]. This shows that property (AF) holds for any subgroup $`H`$ such that $`TH`$.
###### Definition 1.
An affine homogeneous space $`G/H`$ is called *affinely closed* if it admits only one affine embedding $`X=G/H`$.
Homogeneous spaces $`G/H`$ of Example 3 are affinely closed. Denote by $`N_G(H)`$ the normalizer of $`H`$ in $`G`$. The following theorem generalizes Example 3:
###### Theorem 2 (Luna \[Lu2\]).
Let $`H`$ be a reductive subgroup of a reductive group $`G`$. The homogeneous space $`G/H`$ is affinely closed if and only if the group $`N_G(H)/H`$ is finite.
This theorem provides many examples of homogeneous spaces with property (AF). Let us note that the complexity of the space $`G/T`$ can be arbitrary large, whence property (AF) cannot be characterized only in terms of complexity.
In this paper, we show that the union of two conditions—the sphericity and the finiteness of $`N_G(H)/H`$—is very close to characterizing all affine homogeneous spaces of a reductive group $`G`$ with property (AF). Our main result is
###### Theorem 3.
For a reductive subgroup $`HG`$, (AF) holds if and only if either $`N_G(H)/H`$ is finite or any extension of $`H`$ by a one-dimensional torus in $`N_G(H)`$ is spherical in $`G`$.
###### Corollary 1.
For an affine homogeneous space $`G/H`$ of complexity $`>1`$, (AF) holds iff $`G/H`$ is affinely closed.
###### Corollary 2.
An affine homogeneous space $`G/H`$ of complexity $`1`$ has (AF) iff either $`N_G(H)/H`$ is finite or $`rkN_G(H)/H=1`$ and $`N_G(H)`$ is spherical.
The proofs of Theorem 3 and of its corollaries are given in Sections 2, 4.
For simple $`G`$, there is a list of all affine homogeneous spaces of complexity one \[Pan\]. We immediately deduce from this list and Corollary 2 that for simple $`G`$, there exists only one series of affine homogeneous spaces of complexity one that admit affine embeddings with infinitely many $`G`$-orbits. Namely, $`G=SL(n)`$, $`n>4`$, and $`H^0=SL(n2)\times \mathrm{𝕜}^{}`$, where $`SL(n2)`$ is embedded in $`SL(n)`$ as the stabilizer of the first two basis vectors $`e_1`$ and $`e_2`$ in the tautological representation of $`SL(n)`$, and $`\mathrm{𝕜}^{}`$ acts on $`e_1`$ and $`e_2`$ with weights $`\alpha _1`$ and $`\alpha _2`$ such that $`\alpha _1+\alpha _2=2n`$, $`\alpha _1\alpha _2`$, and acts on $`e_3,\mathrm{},e_n`$ by scalar multiplications.
In Section 5 we consider very symmetric affine embeddings $`G/HX`$, i.e. affine embeddings whose group of $`G`$-equivariant automorphisms $`Aut_G(X)`$ contains the identity component of $`Aut_G(G/H)N_G(H)/H`$. The criterion for finiteness of the number of $`G`$-orbits in any such embedding is given (Proposition 2).
The aim of Section 6 is to generalize Theorem 3 following ideas of \[Akh2\] and to find the maximal number of parameters in a continuous family of $`G`$-orbits over all affine embeddings of a given affine homogeneous space $`G/H`$. More precisely,
###### Definition 2.
Let $`F:X`$ be an algebraic group action. The integer
$$d_F(X)=\underset{xX}{\mathrm{min}}codim_XFx=\mathrm{tr}.\mathrm{deg}\mathrm{𝕜}(X)^F$$
is called the generic modality of the action. The modality of $`F:X`$ is the number $`mod_FX=\mathrm{max}_{YX}d_F(Y)`$, where $`Y`$ runs through $`F`$-stable irreducible subvarieties of $`X`$.
Note that $`c(X)=d_B(X)`$. It was proved by E. B. Vinberg \[Vi1\] that for any $`G`$-variety $`X`$ one has $`mod_B(X)=c(X)`$, which means that if we pass from $`X`$ to a $`B`$-stable irreducible subvariety $`YX`$, then the number of parameters for $`B`$-orbits does not increase.
Simple examples show that for $`G`$ itself the equality $`d_G(X)=mod_G(X)`$ is not true. This motivates the following
###### Definition 3.
With any $`G`$-variety $`X`$ we associate the integer
$$m_G(X)=\underset{X^{}}{\mathrm{max}}mod_G(X^{}),$$
where $`X^{}`$ runs through all $`G`$-varieties birationally $`G`$-isomorphic to $`X`$.
It is clear that for any subgroup $`HG`$ the inequality $`m_G(X)m_H(X)`$ holds. In particular, $`m_G(X)c(X)`$. The next theorem shows that $`m_G(X)=c(X)`$.
###### Theorem 4 (Akhiezer \[Akh2\]).
There exists a $`G`$-variety $`X^{}`$ birationally $`G`$-isomorphic to $`X`$ such that $`mod_G(X^{})=c(X)`$.
For a homogeneous space $`G/H`$ we have $`m_G(G/H)=\mathrm{max}_Xmod_G(X)`$, where $`X`$ runs through all embeddings of $`G/H`$. The affine version of this notion is the following
###### Definition 4.
With any quasiaffine homogeneous space $`G/H`$ we associate the integer
$$a_G(G/H)=\underset{X}{\mathrm{max}}mod_G(X),$$
where $`X`$ runs through all affine embeddings of $`G/H`$.
The following theorem is a direct generalization of Theorem 3.
###### Theorem 5.
Let $`G/H`$ be an affine homogeneous space.
(1) If the group $`N_G(H)/H`$ is finite then $`a_G(G/H)=0`$;
(2) If $`N_G(H)/H`$ is infinite then
$$a_G(G/H)=\underset{H_1}{\mathrm{max}}c(G/H_1),$$
where $`H_1`$ runs through all non-trivial extensions of $`H`$ by a one-dimensional subtorus of $`N_G(H)`$. In particular, $`a_G(G/H)=`$ $`c(G/H)`$ or $`c(G/H)1`$.
Applying this theorem to the case $`H=\{e\}`$, we obtain
###### Corollary 3.
$`a_G(G)=dimU1`$ if $`G`$ is semisimple, and $`a_G(G)=dimU`$ otherwise, where $`U`$ is a maximal unipotent subgroup of $`G`$.
### Acknowledgements
We are grateful to M. Brion, D. I. Panyushev and E. B. Vinberg for useful discussions.
This paper was partially written during our stay at the Erwin Schrödinger International Institute for Mathematical Physics (Wien, Austria). We wish to thank this institution for invitation and hospitality.
### Notation and conventions
$`G`$ is a connected reductive group;
$`H`$ is an observable subgroup of $`G`$;
$`TB`$ are a maximal torus and a Borel subgroup of $`G`$;
$`U`$ is the maximal unipotent subgroup of $`B`$;
$`N_G(H)`$ is the normalizer of $`H`$ in $`G`$;
$`W(H)`$ is the quotient group $`N_G(H)/H`$;
$`\gamma :N_G(H)W(H)`$ is the quotient homomorphism;
$`\mathrm{𝕜}^{}`$ is the multiplicative group of non-zero elements of the base field $`\mathrm{𝕜}`$;
$`L^0`$ is the identity component of an algebraic group $`L`$;
$`Z(L)`$ is the center of $`L`$, $`𝔷(L)`$ is its Lie algebra;
$`X^L`$ is the set of $`L`$-fixed points in an $`L`$-variety $`X`$;
$`L_x`$ is the isotropy subgroup of $`xX`$;
$`\mathrm{\Xi }(G)_+`$ is the semigroup of all dominant weights of $`G`$;
$`V_\mu `$ is an irreducible $`G`$-module with highest weight $`\mu `$;
$`\mathrm{𝕜}[X]`$ is the algebra of regular functions and $`\mathrm{𝕜}(X)`$ is the field of rational functions on an algebraic variety $`X`$;
$`SpecA`$ is the affine variety corresponding to a finitely generated algebra $`A`$ without nilpotent elements.
Algebraic groups are denoted by uppercase Latin letters and their Lie algebras by the respective lowercase Gothic letters.
## 2 Embeddings with infinitely many orbits.
###### Theorem 6.
Let $`H`$ be an observable subgroup in a reductive group $`G`$. Suppose that there is a non-trivial one-parameter subgroup $`\lambda :\mathrm{𝕜}^{}W(H)`$ such that the subgroup $`H_1=\gamma ^1(\lambda (\mathrm{𝕜}^{}))`$ is not spherical in $`G`$. Then there exists an affine embedding $`G/HX`$ with infinitely many $`G`$-orbits.
We shall prove this theorem in the next section. The idea of the proof is to apply Akhiezer’s construction for the non-spherical homogeneous space $`G/H_1`$ and to consider the affine cone over a projective embedding of $`G/H_1`$ with infinitely many $`G`$-orbits.
###### Proof of Corollary 1.
The assertion follows from Theorem 2 and Theorem 6, which is a part of Theorem 3 (for reductive $`H`$). Indeed, reductivity of $`H`$ implies reductivity of $`W(H)`$ \[Lu2\]. If $`W(H)`$ is not finite, then it contains a non-trivial one-parameter subgroup $`\lambda (\mathrm{𝕜}^{})`$. For $`H_1=\gamma ^1(\lambda (\mathrm{𝕜}^{}))`$, we have $`c(G/H_1)1`$ whenever $`c(G/H)>1`$. ∎
###### Corollary 4.
Let $`G`$ be a reductive group with infinite center $`Z(G)`$ and let $`H`$ be an observable subgroup in $`G`$ that does not contain $`Z(G)^0`$. Then property (AF) holds for $`G/H`$ if and only if $`H`$ is a spherical subgroup of $`G`$.
###### Proof.
As $`H`$ does not contain $`Z(G)^0`$, there exists a non-trivial one-parameter subgroup $`\lambda (\mathrm{𝕜}^{})`$ in $`Z(G)`$ with finite intersection with $`H`$. The corresponding extension $`H_1`$ is spherical iff $`H`$ is spherical in $`G`$. ∎
###### Corollary 5.
Let $`H`$ be a connected reductive subgroup in a reductive group $`G`$. Suppose that there exists a reductive non-spherical subgroup $`H_1`$ in $`G`$ such that $`HH_1`$ and $`dimH_1=dimH+1`$. Then property (AF) does not hold for $`G/H`$.
###### Proof.
Under these assumptions, there exists a non-trivial one-parameter subgroup of $`H_1`$ with finite intersection with $`H`$ which normalizes (and even centralizes) $`H`$. ∎
## 3 Proof of Theorem 6.
###### Lemma 1.
If property (AF) holds for a homogeneous space $`G/H`$, then it holds for any homogeneous space $`G/H^{}`$, where $`H^{}`$ is an overgroup of $`H`$ with $`(H^{})^0=H^0`$.
###### Proof.
Suppose that there exists an affine embedding $`G/H^{}X`$ with infinitely many $`G`$-orbits. Consider the morphism $`G/HG/H^{}`$. It determines an embedding $`\mathrm{𝕜}[G/H^{}]\mathrm{𝕜}[G/H]`$. Let $`A`$ be the integral closure of the subalgebra $`\mathrm{𝕜}[X]\mathrm{𝕜}[G/H^{}]`$ in the field of rational functions $`\mathrm{𝕜}(G/H)`$. We have the following commutative diagrams:
$$\begin{array}{ccccccccc}A& & \mathrm{𝕜}[G/H]& & \mathrm{𝕜}(G/H)& & SpecA& & G/H\\ & & & & & & & & \\ \mathrm{𝕜}[X]& & \mathrm{𝕜}[G/H^{}]& & \mathrm{𝕜}(G/H^{})& & X& & G/H^{}\end{array}$$
The affine variety $`SpecA`$ with a natural $`G`$-action can be considered as an affine embedding of $`G/H`$. The embedding $`\mathrm{𝕜}[X]A`$ defines a finite (surjective) morphism $`SpecAX`$ and therefore, $`SpecA`$ contains infinitely many $`G`$-orbits. This contradiction completes the proof. ∎
###### Remark 1.
The converse statement does not hold. Indeed, set $`G=SL(3)`$ and $`H=(t,t,t^2)TSL(3)`$. We can extend $`H`$ by a one-parameter subgroup $`(t,t^1,1)`$. Then $`H_1=T`$ is not a spherical subgroup in $`SL(3)`$ and, by Theorem 6, property (AF) does not hold here. On the other hand, one can extend $`H`$ to $`H^{}`$ by a finite noncyclical subgroup of $`W(H)PSL(2)`$. The group $`W(H^{})`$ is finite and, by Theorem 2, property (AF) holds for $`G/H^{}`$.
###### Lemma 2.
1. Let $`HG`$ be an observable subgroup and $`H_1`$ be the extension of $`H`$ by a one-dimensional torus $`\lambda (\mathrm{𝕜}^{})W(H)`$. Then there exists a finite-dimensional $`G`$-module $`V`$ and an $`H_1`$-eigenvector $`vV`$ such that
* the orbit $`Gv`$ of the line $`v`$ in the projective space $`(V)`$ is isomorphic to $`G/H_1`$;
* $`H`$ fixes $`v`$;
* $`H_1`$ acts transitively on $`\mathrm{𝕜}^{}v`$.
2. If $`H_1`$ is not spherical in $`G`$, then a couple $`(V,v)`$ in (a) can be chosen so that
* the closure of $`Gv`$ in $`(V)`$ contains infinitely many $`G`$-orbits.
3. If $`H`$ is reductive, then one can suppose that $`G_v=H`$.
###### Proof.
(a) By Chevalley’s theorem, there exists a $`G`$-module $`V^{}`$ and a vector $`v^{}V^{}`$ having property (1). Let us denote by $`\chi `$ the character of $`H`$ at $`v^{}`$. Since $`H`$ is observable in $`G`$, every finite-dimensional $`H`$-module can be embedded in a finite-dimensional $`G`$-module \[BBHM\]. In particular, there exists a finite-dimensional $`G`$-module $`V^{\prime \prime }`$ containing $`H`$-eigenvectors of character $`\chi `$. Choose among them a $`H_1`$-eigenvector $`v^{\prime \prime }`$ and put $`V=V^{}V^{\prime \prime }`$ and $`v=v^{}v^{\prime \prime }`$. Properties (1) and (2) are satisfied.
If condition (3) also holds, then we are done. Otherwise, consider any $`G`$-module $`W`$ having a vector with stabilizer $`H`$. Take an $`H_1`$-eigenvector $`wW^H`$ with nontrivial character, and replace $`V`$ by $`VW`$ and $`v`$ by $`vw`$. Now properties (1)(3) are satisfied.
(b) Since $`H_1`$ is not spherical in $`G`$, by a result due to Akhiezer \[Akh1\], we may choose $`(V^{},v^{})`$ in (a) so that properties (1) and (4) are satisfied. Then we proceed as in (a) to obtain the couple $`(V,v)`$. The closure $`\overline{Gv}(V)`$ is contained in the image of the Segre embedding
$$(V^{})\times (V^{\prime \prime })(V),\text{or}(V^{})\times (V^{\prime \prime })\times (W)(V),$$
and projects $`G`$-equivariantly onto $`\overline{Gv^{}}(V^{})`$. This implies (4) for $`(V,v)`$.
(c) Let $`\omega `$ be a fundamental weight of the group $`H_1/H`$. Suppose that $`H_1/H`$ acts at the vector $`v`$ constructed above by a character $`k\omega `$. Since $`H_1`$ is reductive (and, in particular, is observable), there exists a $`G`$-module $`W^{}`$ and an $`H_1`$-eigenvector $`w^{}W^H`$ with weight $`(1k)\omega `$ \[BBHM\]. It remains to replace $`V`$ by $`VW^{}`$ and $`v`$ by $`vw^{}`$. ∎
###### Remark 2.
For an arbitrary observable subgroup, statement (c) of Lemma 2 does not hold. For example, let $`G`$ be the group $`SL(3)`$ and $`H=U`$ be a maximal unipotent subgroup normalized by $`T`$. Consider the subtorus $`T^{}=diag(t^2,t,t^3)`$ in $`T`$ as a one-parameter subgroup $`\lambda (\mathrm{𝕜}^{})`$. Any $`H`$-stable vector in a finite-dimensional $`G`$-module is a sum of highest weight vectors. The restriction of any dominant weight to $`T^{}`$ has a non-trivial kernel and the stabilizer of such a vector contains $`H`$ as a proper subgroup.
###### Proof of Theorem 6.
Let $`(V,v)`$ be the couple from Lemma 2. Denote by $`H^{}`$ the stabilizer $`G_v`$ of the vector $`v`$. By (1)(3) and since $`H_1/H`$ is isomorphic to $`\mathrm{𝕜}^{}`$, $`H^{}`$ is an overgroup of $`H`$ with $`(H^{})^0=H^0`$. By (3), the closure of $`Gv`$ in $`V`$ is a cone, so by (4) the property (AF) does not hold for $`G/H^{}`$. Lemma 1 completes the proof. ∎
## 4 Proof of Theorem 3
Let $`H`$ be a reductive subgroup of $`G`$. If there exists a non-spherical extension of $`H`$ by a one-dimensional torus, then (AF) fails for $`G/H`$ by Theorem 6. To prove the converse, we begin with the following
###### Lemma 3 (\[Kn1, 7.3.1\]).
Let $`X`$ be an irreducible $`G`$-variety, and $`v`$ be a $`G`$-invariant valuation of $`\mathrm{𝕜}(X)/\mathrm{𝕜}`$ with residue field $`\mathrm{𝕜}(v)`$. Then $`\mathrm{𝕜}(v)^B`$ is the residue field of the restriction of $`v`$ to $`\mathrm{𝕜}(X)^B`$.
###### Proof.
For completeness, we give the proof in the case, where $`X`$ is affine (the only case we need below). It suffices to prove that any $`B`$-invariant element of $`\mathrm{𝕜}(v)`$ is the residue class of a $`B`$-invariant rational function on $`X`$.
For any $`f_1,f_2\mathrm{𝕜}(X)`$, we shall write $`f_1f_2`$ if $`v(f_1)=v(f_2)<v(f_1f_2)`$. Such “congruences” are $`G`$-stable and may be multiplied term by term, as usual numerical congruences.
Assume $`f=p/q`$, $`p,q\mathrm{𝕜}[X]`$, $`v(f)=0`$, and the residue class of $`f`$ belongs to $`\mathrm{𝕜}(v)^B`$. Then $`v(p)=v(q)=d`$, and $`bff`$, $`bB`$, i.e. $`bpqpbq`$.
Let $`M`$ be a complementary $`G`$-submodule to $`\{h\mathrm{𝕜}[X]v(h)>d\}`$ in $`\{h\mathrm{𝕜}[X]v(h)d\}`$, and $`p_0,q_0`$ be the projections of $`p,q`$ on $`M`$. Then $`bp_0qbpqpbqpbq_0`$, $`bB`$. By the Lie–Kolchin theorem, we may choose finitely many $`b_iB`$, $`\lambda _i\mathrm{𝕜}`$ so that $`q_1=\lambda _ib_iq_0`$ is a $`B`$-eigenfunction in $`M`$ of some weight $`\mu `$. Put $`p_1=\lambda _ib_ip_0`$. Then $`p_1qpq_1`$, whence $`p_1/q_1fbfbp_1/\mu (b)q_1`$, $`bB`$. It follows that $`bp_1\mu (b)p_1`$, hence $`bp_1=\mu (b)p_1`$, because $`p_1M`$. Thus $`p_1,q_1`$ are $`B`$-eigenfunctions of the same weight, and $`f_1=p_1/q_1\mathrm{𝕜}(X)^B`$ has the same residue class in $`\mathrm{𝕜}(v)`$ as $`f`$. ∎
###### Definition 5 (\[Kn2, §7\]).
Let $`X`$ be a normal $`G`$-variety. A discrete $``$-valued $`G`$-invariant valuation of $`\mathrm{𝕜}(X)`$ is called *central* if it vanishes on $`\mathrm{𝕜}(X)^B\{0\}`$. A *source* of $`X`$ is a non-empty $`G`$-stable subvariety $`YX`$ which is the center of a central valuation of $`\mathrm{𝕜}(X)`$.
For affine $`X`$, central valuations are described in a simple way. Consider the isotypic decomposition
$$\mathrm{𝕜}[X]=\underset{\mu \mathrm{\Xi }(X)_+}{}\mathrm{𝕜}[X]_\mu ,$$
where the *rank semigroup* $`\mathrm{\Xi }(X)_+\mathrm{\Xi }(G)_+`$ is the set of all dominant weights $`\mu `$ such that $`\mathrm{𝕜}[X]_\mu 0`$. For any $`\lambda ,\mu \mathrm{\Xi }(X)_+`$, we have
$$\mathrm{𝕜}[X]_\lambda \mathrm{𝕜}[X]_\mu \mathrm{𝕜}[X]_{\lambda +\mu }\underset{\alpha 𝒯_{\lambda ,\mu }(X)}{}\mathrm{𝕜}[X]_{\lambda +\mu \alpha },$$
$`()`$
where $`𝒯_{\lambda ,\mu }(X)`$ is a finite set of positive integral linear combinations of positive roots, and the inclusion fails for all proper subsets of $`𝒯_{\lambda ,\mu }(X)`$.
Let $`\mathrm{\Xi }(X)`$ be a sublattice spanned by $`\mathrm{\Xi }(X)_+`$ in the weight lattice of $`G`$, and $`\mathrm{\Xi }(X)_{}=\mathrm{\Xi }(X)`$. Define the “cone of tails” $`𝒯(X)`$ to be the convex cone in $`\mathrm{\Xi }(X)_{}`$ spanned by the union of all $`𝒯_{\lambda ,\mu }(X)`$.
A central valuation $`v`$ is constant on each $`\mathrm{𝕜}[X]_\mu `$ and defines a linear function $`\nu Hom(\mathrm{\Xi }(X),)=\mathrm{\Xi }(X)_{}^{}`$ so that $`\nu ,\mu =v(f)`$, $`f\mathrm{𝕜}[X]_\mu \{0\}`$. By definition of a valuation, we must have $`\nu ,\alpha 0`$ for $`\alpha 𝒯_{\lambda ,\mu }(X)`$, $`\lambda ,\mu \mathrm{\Xi }(X)_+`$. Conversely, each linear function $`\nu Hom(\mathrm{\Xi }(X),)`$ which is non-positive on $`𝒯(X)`$ defines a central valuation $`v`$ of $`\mathrm{𝕜}(X)`$ by the formula
$$v(f)=\mathrm{min}\{\nu ,\mu f_\mu 0\},$$
where $`f_\mu `$ is the projection of $`f\mathrm{𝕜}[X]`$ on $`\mathrm{𝕜}[X]_\mu `$.
The valuation $`v`$ has a center on $`X`$ iff $`\nu `$ is non-negative on $`\mathrm{\Xi }(X)_+`$, and the respective source $`YX`$ is determined by a $`G`$-stable ideal
$$I(Y)=\underset{\nu ,\mu >0}{}\mathrm{𝕜}[X]_\mu \mathrm{𝕜}[X]$$
Central valuations of $`\mathrm{𝕜}(X)`$, identified with respective linear functions on $`\mathrm{\Xi }(X)_{}`$, form a solid convex cone $`𝒵(X)\mathrm{\Xi }(X)_{}^{}`$, namely the dual cone to $`𝒯(X)`$. Knop proved \[Kn1, 9.2\], \[Kn2, 7.4\] that $`𝒵(X)`$ is a fundamental domain for a finite group $`W_XAut\mathrm{\Xi }(X)`$ (the *little Weyl group* of $`X`$) acting on $`\mathrm{\Xi }(X)_{}^{}`$ as a crystallographic reflection group.
The following lemma is an easy consequence of results of Knop \[Kn2\].
###### Lemma 4.
If $`X`$ is a normal affine $`G`$-variety containing a proper source, then there exists a one-dimensional torus $`SAut_G(X)`$ such that $`\mathrm{𝕜}(X)^B\mathrm{𝕜}(X)^S`$. (Here $`Aut_G(X)`$ denotes the group of $`G`$-equivariant automorphisms of $`X`$.)
###### Proof.
If $`X`$ is as above, then Knop has shown that the algebra $`\mathrm{𝕜}[X]`$ admits a non-trivial $`G`$-invariant grading, whose homogeneous components are sums of isotypic components of the $`G`$-module $`\mathrm{𝕜}[X]`$, see \[Kn2, 7.9\] and its proof. This grading is constructed as follows. Under the above assumptions, there is a central valuation $`v`$ of $`\mathrm{𝕜}(X)`$ such that the respective linear function $`\nu `$ on $`\mathrm{\Xi }(X)_{}`$ lies in $`𝒵(X)𝒵(X)`$, hence $`\nu `$ vanishes on $`𝒯(X)`$. In view of $`()`$, this $`\nu `$ defines a grading of $`\mathrm{𝕜}[X]`$ such that isotypic components $`\mathrm{𝕜}[X]_\mu `$ are homogeneous of degree $`\nu ,\mu `$.
Let $`SAut_G(X)`$ be the one-dimensional torus corresponding to this grading. Take any $`f\mathrm{𝕜}(X)^B`$, $`f=p/q`$, $`p,q\mathrm{𝕜}[X]`$. By the Lie–Kolchin theorem, we may choose finitely many $`b_iB`$, $`\lambda _i\mathrm{𝕜}`$ so that $`q_0=\lambda _ib_iq`$ is a $`B`$-eigenfunction of some weight $`\mu \mathrm{\Xi }(X)_+`$. Then $`p_0=\lambda _ib_ip`$ is a $`B`$-eigenfunction of the same weight, and $`f=p_0/q_0`$. Since $`p_0,q_0\mathrm{𝕜}[X]_\mu `$, the torus $`S`$ acts on them by the same weight $`\nu ,\mu `$, thence $`f\mathrm{𝕜}(X)^S`$. This shows the inclusion $`\mathrm{𝕜}(X)^B\mathrm{𝕜}(X)^S`$. ∎
###### Proof of Theorem 3.
It remains to prove that (AF) holds for $`G/H`$ whenever any extension of $`H`$ by a one-dimensional torus is spherical. As $`H`$ is reductive, $`W(H)`$ is reductive, too. If there exists no one-parameter extension of $`H`$ at all, then $`W(H)`$ is finite and $`G/H`$ is affinely closed by Theorem 2. Otherwise $`c(G/H)1`$. As the spherical case is clear, we may suppose $`c(G/H)=1`$.
Let $`X`$ be an affine embedding of $`G/H`$. In order to prove that $`X`$ has finitely many $`G`$-orbits, we may assume that $`X`$ is normal. If $`X`$ contains a proper source, then a one-dimensional torus $`SAut_G(X)Aut_G(G/H)=W(H)`$ provided by Lemma 4 yields a non-spherical extension of $`H`$. Indeed, if $`H_1`$ is the preimage of $`S`$ in $`N_G(H)`$, then $`\mathrm{𝕜}(G/H_1)^B=\mathrm{𝕜}(G/H)^{B\times S}=\mathrm{𝕜}(X)^{B\times S}=\mathrm{𝕜}(X)^B\mathrm{𝕜}`$, since $`c(X)=1`$. This implies $`c(G/H_1)=1`$, a contradiction.
If $`X`$ contains no proper source, then any proper $`G`$-stable subvariety $`YX`$ is the center of a non-central $`G`$-invariant valuation $`v`$. There is an inclusion of residue fields $`\mathrm{𝕜}(Y)\mathrm{𝕜}(v)\mathrm{𝕜}(Y)^B\mathrm{𝕜}(v)^B`$. By Lemma 3, $`\mathrm{𝕜}(v)^B`$ is the residue field of the restriction of $`v`$ to $`\mathrm{𝕜}(G/H)^B`$, which is the field of rational functions in one variable. As $`v`$ is non-central, $`\mathrm{𝕜}(Y)^B=\mathrm{𝕜}(v)^B=\mathrm{𝕜}`$, thence $`Y`$ is spherical. It follows that $`X`$ has finitely many orbits. (Otherwise, a one-parameter family of $`G`$-orbits provides a non-spherical $`G`$-subvariety.) ∎
###### Proof of Corollary 2.
The reductive group $`W(H)`$ acts on $`\mathrm{𝕜}(G/H)^B`$, which is the field of rational functions on a projective line. If the kernel of this action has positive dimension, then it contains a one-dimensional torus extending $`H`$ to a non-spherical subgroup.
Otherwise, either $`W(H)`$ is finite or $`rkW(H)=1`$ and each subtorus of $`W(H)`$ has a dense orbit on the projective line. The corollary follows. ∎
###### Remark 3.
In the proof of Theorem 3, we have used reductivity of $`H`$ only in the following assertion:
> If $`W(H)`$ contains no subtori, then it is finite, and $`G/H`$ is affinely closed.
In fact, we need this assertion only if $`c(G/H)>1`$. Theorem 3 holds for quasiaffine $`G/H`$ of complexity $`1`$.
One might hope that the situation described in the above assertion never occurs for non-reductive $`H`$, i.e. that $`W(H)`$ always contains a subtorus. Unfortunately, $`W(H)^0`$ may be a non-trivial unipotent group, as the following example shows.
###### Example 4.
Let $`e`$ be a regular nilpotent in the Lie algebra $`𝔰𝔩(3)`$, $`G=SL(3)\times SL(3)`$, and $`H`$ be the two-dimensional unipotent subgroup with the Lie algebra generated by $`(e,e^2)`$ and $`(e^2,e)`$. Then the Lie algebra of the normalizer of $`H`$ is the linear span of $`(e,0),(e^2,0),(0,e)`$ and $`(0,e^2)`$. Hence the group $`W(H)^0`$ is two-dimensional and unipotent.
(Another example was suggested by E.A.Tevelev.)
We are not able to characterize quasiaffine, but not affine, homogeneous spaces with the property (AF).
In this context we would like to formulate the following
###### Conjecture.
If $`HG`$ is observable, but not reductive, then $`W(H)`$ is infinite.
## 5 Very symmetric embeddings.
The group of $`G`$-equivariant automorphisms of a homogeneous space $`G/H`$ is isomorphic to $`W(H)`$. (The action $`W(H):G/H`$ is induced by the action $`N_G(H):G/H`$ by right multiplication.) Let $`G/HX`$ be an affine embedding. The group $`Aut_GX`$ of $`G`$-equivariant automorphisms of $`X`$ is a subgroup of $`W(H)`$.
###### Definition 6.
An embedding $`G/HX`$ is said to be *very symmetric* if $`W(H)^0Aut_GX`$.
Any spherical affine variety is very symmetric. In fact, for a spherical homogeneous space $`G/H`$, any isotypic component $`\mathrm{𝕜}[G/H]_\mu `$ of the $`G`$-algebra $`\mathrm{𝕜}[G/H]`$ is an irreducible $`G`$-module (see \[Ser\] or \[KV, Th.2\]), and $`W(H)`$ acts on $`\mathrm{𝕜}[G/H]_\mu `$ by scalar multiplications. This shows that any $`G`$-invariant subalgebra in $`\mathrm{𝕜}[G/H]`$ is $`W(H)`$-invariant, too.
In the case of affine $`SL(2)/\{e\}`$-embeddings, only the embedding $`X=SL(2)`$ is very symmetric; in all other cases the group $`Aut_{SL(2)}X`$ is isomorphic to a Borel subgroup in $`SL(2)`$, see \[Kr, III.4.8, Satz 1\]. More generally, if $`X`$ is an affine embedding of the homogeneous space $`G/\{e\}`$, then $`X`$ is very symmetric if and only if the action $`G:X`$ can be extended to an action of the group $`G\times G`$ with an open orbit isomorphic to $`(G\times G)/H`$, where $`H`$ is the diagonal in $`G\times G`$. Hence $`X`$ can be considered as an affine $`(G\times G)/H`$-embedding. Theorem 2 implies that if $`G`$ is a semisimple group, then $`X=(G\times G)/H`$, for other proofs see \[Wat\] and \[Vi2, Prop. 1\].
If $`G`$ is a reductive group, then the set of all very symmetric embeddings of the homogeneous space $`G/\{e\}`$ is exactly the set of all affine algebraic monoids with $`G`$ as the group of units \[Vi2\]. Thus very symmetric embeddings have a natural characterization in the variety of all affine $`G/\{e\}`$-embeddings. The classification of reductive algebraic monoids is obtained in \[Vi2\] and \[Rit\].
Put $`\widehat{G}=G\times W(H)^0`$, $`N=\gamma ^1(W(H)^0)`$, and $`\widehat{H}=\{(n,nH)nN\}`$ (the “diagonal” embedding of $`N`$). Any very symmetric affine embedding of $`G/H`$ may be considered as an embedding of $`\widehat{G}/\widehat{H}`$.
###### Proposition 1.
Under assumptions of Theorem 6, if $`\lambda (\mathrm{𝕜}^{})`$ is central in $`W(H)^0`$, then there exists a very symmetric affine embedding $`G/HX`$ with infinitely many $`G`$-orbits.
###### Proof.
We follow the proof of Theorem 6. Put $`\widehat{H}_1=\widehat{H}\lambda (\mathrm{𝕜}^{})`$; then $`\widehat{H}_1G=H_1`$. We modify the proof of Lemma 2(b) to obtain a $`\widehat{G}`$-module $`V`$ and an $`\widehat{H}_1`$-eigenvector $`vV`$ such that $`\widehat{G}_v=\widehat{H}_1`$, $`\widehat{G}_v`$ is a finite extension of $`\widehat{H}`$, and $`\overline{\widehat{G}v}(V)`$ contains infinitely many $`G`$\- (not $`\widehat{G}`$-) orbits. Arguing as in the proof of Theorem 6, we see that the closure $`X`$ of $`Gv=\widehat{G}vV`$ is $`\widehat{G}`$-stable and has infinitely many $`G`$-orbits, q.e.d. (Observe that $`\widehat{G}`$ may be not reductive, but Lemma 1, required in the proof, does not use the reductivity assumption.)
To construct such a couple $`(V,v)`$, it suffices, in the notation of Lemma 2, to construct a $`\widehat{G}`$-module $`V^{}`$ and a vector $`v^{}V^{}`$ such that $`Gv^{}=\widehat{G}v^{}\widehat{G}/\widehat{H}_1`$ and $`\overline{Gv^{}}`$ has infinitely many $`G`$-orbits. Then we proceed as in Lemma 2(a), replacing $`G`$ by $`\widehat{G}`$. (Note that the reductivity of $`G`$ is not essential in Lemma 2(a).) It remains to construct a couple $`(V^{},v^{})`$. For this purpose, we refine Akhiezer’s construction \[Akh1\].
By assumption, $`c(G/H_1)>0`$, hence there exists a character $`\xi :H_1\mathrm{𝕜}^{}`$ such that for the associated line bundle $`L_\xi `$ on $`G/H_1`$, the multiplicity of a certain simple $`G`$-module $`V_\mu `$ in $`H^0(G/H_1,L_\xi )`$ is greater than one \[KV, Th. 1\].
The group $`W(H)^0`$ acts on $`H^0(G/H_1,L_\xi )`$ and on the isotypic component $`E=H^0(G/H_1,L_\xi )_\mu `$ by $`G`$-module automorphisms.
Take a $`\widehat{G}`$-module $`M`$ and a vector $`mM`$ such that $`\widehat{G}_m=\widehat{H}_1`$. Let $`Y`$ be the closure of $`\widehat{G}m=Gm`$ in $`(M)`$. The natural rational map $`f:Y(E^{})`$ is $`\widehat{G}`$-equivariant.
Consider a decomposition $`E=E_0\mathrm{}E_k`$ into irreducible $`G`$-submodules and fix isomorphisms $`\psi _i:V_\mu E_i`$. Choose a basis $`\{\epsilon _0,\mathrm{},\epsilon _m\}`$ of $`T`$-eigenvectors with weights $`\mu _0=\mu ,\mu _1,\mathrm{},\mu _m`$ in $`V_\mu `$, and put $`\epsilon _j^{(i)}=\psi _i(\epsilon _j)`$. In projective coordinates,
$$f(gH_1)=[\epsilon _0^{(0)}(gH_1):\mathrm{}:\epsilon _0^{(k)}(gH_1):\mathrm{}:\epsilon _m^{(0)}(gH_1):\mathrm{}:\epsilon _m^{(k)}(gH_1)]$$
The closure $`Z`$ of the graph of $`f`$ in $`Y\times (E^{})`$ is $`\widehat{G}`$-stable. We claim that $`Z`$ contains infinitely many $`G`$-orbits. To prove it, take a strictly dominant one-parameter subgroup $`\delta :\mathrm{𝕜}^{}T`$. If all $`\epsilon _0^{(i)}(gH_1)0`$, then
$`f(\delta (t)gH_1)`$ $`=`$ $`[\mathrm{}:t^{\mu _j,\delta }\epsilon _j^{(i)}(gH_1):\mathrm{}]`$
$`=`$ $`[\mathrm{}:t^{\mu _0\mu _j,\delta }\epsilon _j^{(i)}(gH_1):\mathrm{}]`$
$``$ $`[\epsilon _0^{(0)}(gH_1):\mathrm{}:\epsilon _0^{(k)}(gH_1):\mathrm{}:0:\mathrm{}:0]`$
as $`t0`$, because $`\mu _0\mu _j`$ is a positive linear combination of positive roots for all $`j>0`$.
We may identify $`E^{}`$ with $`V_\mu ^{}\mathrm{𝕜}^{k+1}`$ and consider the Segre embedding $`(V_\mu ^{})\times ^k(E^{})`$. Then $`lim_{t0}\delta (t)f(gH_1)=(\epsilon _0^{},p)(V_\mu ^{})\times ^k`$, where $`\{\epsilon _j^{}\}`$ is the dual basis to $`\{\epsilon _j\}`$, and $`p=[\epsilon _0^{(0)}(gH_1):\mathrm{}:\epsilon _0^{(k)}(gH_1)]^k`$.
As the sections $`\epsilon _0^{(0)},\mathrm{},\epsilon _0^{(k)}`$ are linearly independent on $`G/H_1`$, $`\overline{f(Y)}`$ intersects infinitely many closed disjoint $`G`$-stable subvarieties $`(V_\mu ^{})\times \{p\}(E^{})`$, $`p^k`$. This proves the claim, because $`Z`$ projects $`G`$-equivariantly onto $`\overline{f(Y)}`$.
Finally, a $`\widehat{G}`$-module $`V^{}=ME^{}`$ and a vector $`v^{}=me`$ such that $`f(m)=e`$ are the desired, because $`\overline{Gv^{}}Z`$.
The proof is complete. ∎
Now we are interested in the following problem: when does any very symmetric affine embedding of a homogeneous space $`G/H`$ have finitely many $`G`$-orbits? The example of $`SL(3)/\{e\}`$-embeddings shows that the latter property is not equivalent to (AF).
###### Proposition 2.
Let $`H`$ be a reductive subgroup in a reductive group $`G`$. Every very symmetric affine embedding of $`G/H`$ has finitely many $`G`$-orbits iff either (AF) holds or $`W(H)^0`$ is semisimple. In the second case, there is only one very symmetric affine embedding, namely $`X=G/H`$.
###### Proof.
The Lie algebra of $`N_{\widehat{G}}(\widehat{H})`$ equals $`\widehat{𝔥}+\widehat{𝔷}`$, where $`\widehat{𝔷}`$ is the centralizer of $`\widehat{H}`$ in $`\widehat{𝔤}`$. We have $`\widehat{𝔷}=𝔷(N)𝔷(W(H)^0)`$, and $`𝔷(N)𝔷(H)𝔷(W(H)^0)`$.
If $`W(H)^0`$ is semisimple, then $`\widehat{𝔷}𝔥\widehat{𝔥}`$, and $`N_{\widehat{G}}(\widehat{H})`$ is finite. Theorem 2 implies the assertion for this case.
Now suppose that $`W(H)^0`$ is not semisimple. If there exists a non-spherical extension of $`H`$ by a one-dimensional torus $`SZ(N)`$, then by Proposition 1, there exists a very symmetric affine embedding of $`G/H`$ with infinitely many $`G`$-orbits.
Finally, suppose that any extension of $`H`$ by a one-dimensional torus in $`Z(N)`$ is spherical. Then $`c(G/H)1`$. As the spherical case is clear, we may assume that $`c(G/H)=1`$.
The connected kernel $`W_0`$ of the action $`W(H):\mathrm{𝕜}(G/H)^B`$ acts on isotypic components of $`\mathrm{𝕜}[G/H]`$ by scalar multiplications. Whence $`W_0`$ is diagonalizable and central in $`W(H)`$. By assumption, $`W_0=\{e\}`$. Hence $`W(H)^0`$ is a one-dimensional torus acting on $`\mathrm{𝕜}(G/H)^B`$ with finite kernel. By Corollary 2, (AF) holds for $`G/H`$. The proof is complete. ∎
## 6 Affine embeddings and modality
We begin this section with the generalization of Lemma 2.
###### Lemma 5.
Let $`HG`$ be an observable subgroup and $`H_1`$ be the extension of $`H`$ by a one-dimensional torus $`\lambda (\mathrm{𝕜}^{})W(H)`$. Then there exists a finite-dimensional $`G`$-module $`V`$ and an $`H_1`$-eigenvector $`vV`$ such that
* the orbit $`Gv`$ of the line $`v`$ in the projective space $`(V)`$ is isomorphic to $`G/H_1`$;
* $`H`$ fixes $`v`$;
* $`H_1`$ acts transitively on $`\mathrm{𝕜}^{}v`$;
* $`mod_G(\overline{Gv})=c(G/H_1)`$.
###### Proof.
We use exactly the same arguments as in the proof of Lemma 2 replacing an embedding of $`G/H_1`$ with infinitely many orbits from \[Akh1\] by an embedding of $`G`$-modality $`c(G/H_1)`$ constructed in \[Akh2\]. ∎
###### Lemma 6.
In the notation of Lemma 5,
$$c(G/H)a_G(G/H)c(G/H_1)c(G/H)1$$
In particular, $`a_G(G/H)=c(G/H)`$ or $`c(G/H)1`$.
###### Proof.
Clearly, $`a_G(G/H)c(G/H)`$. Taking an affine cone over the projective embedding constructed in Lemma 5, one obtains an affine embedding of $`G/H^{}`$ of modality $`c(G/H_1)`$, where $`H^{}=G_v`$ is a finite extension of $`H`$. Using the construction from the proof of Lemma 1, we get an affine embedding of $`G/H`$ of modality $`c(G/H_1)`$. The obvious inequality $`c(G/H_1)c(G/H)1`$ completes the proof. ∎
###### Proof of Theorem 5.
Statement (1) follows from Theorem 2. To prove (2), we can use Lemma 6. If there exists a one-dimensional torus in $`N_G(H)`$ such that the extension $`HH_1`$ is non-trivial and $`c(G/H)=c(G/H_1)`$, then there exists an affine embedding of $`G/H`$ of modality $`c(G/H)`$.
Conversely, suppose that $`G/HX`$ is an affine embedding of modality $`c(G/H)`$. We need to find a one-dimensional subtorus $`SW(H)`$ such that for the extended subgroup $`H_1`$ we will have $`c(G/H_1)=c(G/H)`$. By the definition of modality, there exists a proper $`G`$-invariant subvariety $`YX`$, such that the codimension of a generic $`G`$-orbit in $`Y`$ is $`c(G/H)`$. Therefore, $`c(Y)=c(G/H)`$.
Consider a $`G`$-invariant valuation $`v`$ of $`\mathrm{𝕜}(X)`$ with the center $`Y`$. For the residue field $`\mathrm{𝕜}(v)`$ we have $`\mathrm{tr}.\mathrm{deg}\mathrm{𝕜}(v)^B\mathrm{tr}.\mathrm{deg}\mathrm{𝕜}(Y)^B`$, hence $`\mathrm{tr}.\mathrm{deg}\mathrm{𝕜}(v)^B=\mathrm{tr}.\mathrm{deg}\mathrm{𝕜}(X)^B`$. If the restriction of $`v`$ to $`\mathrm{𝕜}(X)^B`$ is not trivial, then by Lemma 3, $`\mathrm{tr}.\mathrm{deg}\mathrm{𝕜}(v)^B<\mathrm{tr}.\mathrm{deg}\mathrm{𝕜}(X)^B`$, a contradiction. Thus $`v`$ is central, and $`Y`$ is a source of $`X`$. A one-dimensional subtorus $`SAut_G(X)Aut_G(G/H)=W(H)`$ provided by Lemma 4 yields the extension of $`H`$ of the same complexity. ∎
###### Proof of Corollary 3.
If $`G`$ is not semisimple, then for a central one-dimensional subtorus $`T_1`$ one has $`c(G/T_1)=c(G)=dimU`$. If $`G`$ is semisimple, then for any one-dimensional subtorus $`T_1G`$ there exists a Borel subgroup $`B`$ which does not contain $`T_1`$, and there is a $`B`$-orbit on $`G/T_1`$ of dimension $`dimB`$. This implies $`c(G/T_1)=c(G)1`$. ∎
|
warning/0004/nucl-th0004064.html
|
ar5iv
|
text
|
# Searching for the Next Yukawa Phase of QCD))footnote )Talk to be published in the Proc. of 14th Nishinomiya-Yukawa Memorial Symposium Nov. 1999, Japan; updated with a critique of the CERN press release Feb. 8, 2000
## 1 The Ubiquitous Yukawa
In 1935 H. Yukawa proposed a theory of nuclear forces based on the exchange of a massive boson
$$V_Y(r,m)=\alpha _{eff}\frac{e^{mr}}{r}$$
(1.1)
He estimated that $`m100`$ MeV to account for the short range $`2`$ fm, and in 1947 Powell discovered the pion and confired this theory. An important theoretical precursor was the Klein-Gordon equation , whose static Green’s function is (1.1). Elaboration of Yukawa’s meson theory since then (including spin, orbit, isospin vertex factors) forms the basis for the present effective theory of nuclear forces.
In the unrelated field of electrolytes, Debye and Hückle had also come across the Yukawa potential in another context. It emerged from solving the Poisson equation in a conducting medium. Assuming that local charge density fluctuations occur with a Boltzmann probability, $`\mathrm{exp}(q\varphi (x)/T)`$, the polarizability of the medium in the presence of an external charge density, leads to a non-linear self consistent equation
$$\varphi (x)=4\pi (\rho _{ex}(x)+\underset{q}{}qn_qe^{q\varphi (x)/T}).$$
(1.2)
In the linearized approximation, the solution for a point charge is again eq. (1.1), but in this case the effective mass is the Debye electric screening mass
$$\mu ^2=4\pi \underset{q}{}q^2|n_q|/T.$$
(1.3)
A dense conductive medium therefore transforms Coulomb into Yukawa.
In nuclear theory, the Yukawa meson mass results from the finite gap of the elementary excitations (pions, …) of the physical QCD vacuum. By analogy, it should be possible to modify the nuclear Yukawa potential by increasing the nuclear density or temperature. In this talk I discuss current efforts to try to manipulate nuclear matter in the laboratory to force a breakdown of Yukawa’s hadronic theory. As we will see, however, it seems difficult to escape from Yukawa. In the new deconfined, chirally symmetric phase of QCD at high temperatures, the Debye-Huckle mechanism takes over and the Yukawa potential between nucleons mutates into a color-electric screened Yukawa potential between partons of a quark-gluon plasma.
In QCD, the color potential between partons is approximately Coulombic at small distances due to the asymptotic freedom property of non-Abelian gauge theories. However, below a critical temperature, $`T_c150`$ MeV, the confinement property of the nonperturbative QCD vacuum allows only composite color singlet objects (hadrons) to “roam freely” in the laboratory. The effective potential between the colored partons has a long range linear confining term $`\kappa r`$ to prevent them to roaming more than 1 fm away from any color neutral blob (hadrons). The restraining force or ”string” tension, $`\kappa 1`$ GeV/fm, is huge. In this confining phase of QCD, the heavy $`q\overline{q}`$ potential is well parameterized by the Lüscher form
$$V_L(r,0)=\frac{\alpha _L}{r}+\kappa r,$$
(1.4)
(as long as dynamical quark pair production is ignored). The Coulombic part, with strength $`\alpha _L=\pi /12`$, arises from the zero point quantum fluctuations of the string (see p. 803 of Ref for an intuitive derivation). This confining potential between heavy quarks has been directly observed numerically using lattice QCD techniques. As the temperature increases, but remains below the deconfinement transition, $`T<T_c`$, the enhanced fluctuations due to thermal agitation of the string modifies the effective potential into the approximate Gao form
$`V_G(r,T)`$ $`=`$ $`\alpha _L[1{\displaystyle \frac{2}{\pi }}\mathrm{tan}^1(2rT)]{\displaystyle \frac{1}{r}}+[\kappa {\displaystyle \frac{\pi }{3}}T^2(1{\displaystyle \frac{2}{\pi }}\mathrm{tan}^1({\displaystyle \frac{1}{2rT}})]r+{\displaystyle \frac{T}{2}}\mathrm{ln}(1+(2rT)^2)`$
The decrease of the effective string tension, $`\kappa (T)`$, predicted above has been also observed in lattice QCD calculations as shown in Fig.(1). However, the “measured” string tension is found to decreases faster than predicted in eq.(LABEL:Gao) near the critical temperature. Note that in Fig.(1) $`\beta =2N_c/g^2=\beta _c+\frac{12}{11N_c2N_f}\mathrm{log}(T/T_c)`$, i.e., $`TT_c\mathrm{exp}(11/6(\beta \beta _c))`$ with $`\beta _c=4.0729`$ for this lattice calculation.
For temperatures above $`T_c`$, we see from Fig.1 that a new Yukawa phase of QCD is predicted, and that the heavy quark potential mutates into a short range generalized Yukawa form, which on the lattice is measured in the form
$$V_L(r,T,d)=\frac{\alpha (T)T}{(rT)^{d_L/2}}e^{\mu _L(T)r/2}.$$
(1.6)
Note that $`d_L=2,\mu _L(T)=2m_E(T)`$ correspond to a pure Yukawa interaction in terms of the lattice QCD fit parameters $`(d_L,\mu _L)`$. The perturbative thermal QCD chromo-electric Debye mass $`m_E=\mu (T)/2`$ is
$$m_E(T)=g(T)T\left(\frac{N_c}{3}+\frac{N_F}{6}\right)^{1/2}$$
(1.7)
for $`N_c`$ colors and $`N_F`$ flavors. For $`N_c=2,N_f=0`$ in Fig. (1), we expect $`\mu _L(T)=2m_E=1.6g(T)T`$ as shown by the solid line in Fig.(2). The fits to the lattice QCD measurements Fig.(1 b) from show that for $`T>2T_c`$ $`\mu _L2.5T`$ is not far from the pQCD estimate. However, the exponent $`d_L1.5`$ is significantly below the value 2 expected from pQCD. An even more striking nonperturbative deviation is seen near $`T_c`$, at which point $`d<1`$ and $`\mu T/2`$. This suggests a rather long range interaction that may be the precursor of the confinement transition.
This strong deviation from the perturbative QCD near $`T_c`$ was also reported previously in ref., where in addition it was found that the effective coupling is rather small $`\alpha 0.15`$ above $`T_c`$. It is important to keep in mind that the above numerical experiments do not include dynamical quarks and are limited to SU(2). Nevertheless, they provide strong evidence that QCD predicts a qualitatively new (nonperturbative) partonic Yukawa phase of matter that should exist at an energy density only an order of magnitude above that in ground state nuclei ($`ϵ>2`$ GeV/fm<sup>3</sup>).
The thermodynamic properties of the deconfined QCD phase are shown in Fig.3 from ref. for 2 flavor $`12^3\times 6`$ lQCD. A present limitation of all lattice results so far is that the pion is still too massive to make make contact with the “known” thermodynamic properties of ordinary hadronic/nuclear matter below $`T_c`$. Recent advances in implementing Domain Wall Fermions on the lattice and the availability of new TeraFlop scale computers at Columbia and the Riken Brookhaven Research Center and the CP-PACS project in Tsukuba, should enable much more precise calculations of the quark-gluon plasma equation of state in the near future.
Two striking features of Fig.(3) suggest two key observable signatures of this phase transition in nuclear collisions. First, the entropy density $`\sigma (T)=(ϵ+p)/T`$ increase very rapidly with $`T`$ in a narrow interval $`\mathrm{\Delta }T/T_c<0.1`$. Second, the plasma becomes extremely soft $`p/ϵ1`$ and $`c_s^21`$ near $`T_c`$. As we review below, the first feature can lead to time delay (the QGP stall) measurable via hadronic interferometry. The second feature can lead to interesting non-linear collective flow observables in nuclear collisions. The experimental verification of these fundamental predictions of QCD is the primary motivatiion for the heavy ion experimental program at Brookhaven and CERN. In the following sections, I review first how deep into the QGP phase RHIC may be able to reach, and then discuss several signatures used to test the QCD predictions in such experiments.
## 2 Initial conditions in A+A
In order to see the partonic Yukawa phase, we must first create an extended blob of matter at 100 times the density of nuclear matter. Head on collisions of heavy ions are used for that. There are many ways to describe how this dense matter is formed. One intuitive picture is given in terms of the McLerran-Venugopalan model. For highly boosted nuclei with $`E_{cm}100m_N`$, time dilation effectively freezes out the quantum chromo fluctuations inside the nuclei while the two pass through each other. Heavy Au beams can then be regarded as well collimated, ultra dense beams of partons. This (chromo Weizsacker-Williams) gluon cloud contains a very large number, $`G_A(x,p_0^2)A/x^{1+\delta }`$, of almost on-shell collinear gluons with longitudinal momentum fraction $`x=p_0/E_{cm}1`$. As the clouds pass through each other, partons scatter via chromo Rutherford and decohere into a mostly gluon plasma on a fast time scale $`1/p_{}1`$ fm/c. The number of gluons pairs (mini-jets) extracted from the nuclei by this mechanism at rapidities $`y_i`$ and transverse momentum $`\pm k_{}`$ can be calculated in pQCD as follows
$$\frac{dN_{ABggX}}{dy_1dy_2dk_{}^2}=Kx_1G_A(x_1,k_{}^2)x_2G_B(x_2,k_{}^2)\frac{d\sigma _{gggg}}{dk_{}^2}T_{AB}(\stackrel{}{b}),$$
(2.1)
where $`x_1=x_{}(\mathrm{exp}(y_1)+\mathrm{exp}(y_2))`$ and $`x_2=x_{}(\mathrm{exp}(y_1)+\mathrm{exp}(y_2))`$, with $`x_{}=k_{}/\sqrt{s}`$, and where the pQCD $`gggg`$ cross section for scattering with $`t=k_{}^2(1+\mathrm{exp}(y_2y_1))`$ and $`y_2y_1=y`$ is given by
$`{\displaystyle \frac{d\sigma ^{gg}}{dt}}`$ $`=`$ $`{\displaystyle \frac{9}{8}}{\displaystyle \frac{4\pi \alpha ^2}{k_{}^4}}{\displaystyle \frac{(1+e^y+e^y)^3}{(e^{y/2}+e^{y/2})^6}}.`$ (2.2)
The nuclear baryon number, $`A`$, only plays here the role of increasing the density of partons and providing the geometrical amplification, $`T_{AB}(\stackrel{}{b})\stackrel{<}{}30/`$mb, for the number of binary nucleon-nucleon collisions that occur per unit area. A factor $`K2`$ simulates next to leading order corrections.
For symmetric systems, $`A+A`$, with $`G_AAG`$. the inclusive gluon jet production cross section is obtained by integrating over $`y_2`$ with $`y_1=y`$ and $`k_{}`$ fixed. To about $`50\%`$ accuracy, the single inclusive gluon rapidity density in central collisions can be estimated by the following simple pocket formula
$$\frac{dN}{dydt}\frac{A^2}{\pi R^2}2N_g(x_{},t)x_{}G(x_{},t)\left(\frac{d\sigma _{gg}^{el}}{dt}\right)_RA^{4/3}$$
(2.3)
where $`N_g(x_{},t)=_x_{}^1𝑑xG(x,t)`$ is the total number of “hard” gluons coming down the beam pipe in an nuclear area $`\pi R^2`$ that can knock out unsuspecting gluons from the other nucleus. This copious mini-jet mechanism is believed to the dominant source of the gluon plasma that will be created when RHIC (finally) begins operation.
Recent upper bound estimates of the total gluon rapidity density as a function of the CM energy from are shown in Fig.(4).
The differential yields are integrated down to a transverse momentum scale $`p_012`$ GeV. This scale separates the “soft” nonperturbative beam jet fragmentation domain from the calculable perturbative one above. The curve marked saturation is an upper bound marking the point where the transverse gluon density of mini-jets becomes so high that the newly liberated gluons completely fill the nuclear area, i.e., $`dN/dyp_0^2R^2`$. At that point, higher order gluon absorption may limit the further increase of the gluon number. At RHIC energies these estimates yield up 1500 gluons per unit rapidity. My more conservative estimates together with X.N. Wang gives 500 gluons per unit rapidity when initial and final state radiation is also taken into account. This is obtained with a fixed $`p_0=2`$ GeV, that was found to be consistent with all available $`p\overline{p}`$ and low energy $`AA`$ data using the HIJING mini-jet event.
The energy dependence of the final charged particle radidity density from HIJING, including soft beam jet fragmentation, is shown in Fig.(5). Approximately one half of the height “Mt. RHIC” comes from soft beam jet fragmentation processes modeled in HIJING using Lund/Fritiof strings.
The initial energy density reached in such collisions can be estimated using the Bjorken formula
$$ϵ(\tau _0)\frac{1}{\pi R^2\tau _0}\frac{dE_T}{dy}$$
(2.4)
For $`p_012`$ GeV, $`dE_T/dy4002000`$ GeV, and so $`ϵ(\tau _00.5\mathrm{fm}/\mathrm{c})>10`$ GeV/fm<sup>3</sup> should be easily reached, well inside the the deconfinement phase of QCD. At SPS energies, on the other hand, my estimates indicate that nuclear collisions may just reach the transition region and depart from that region in a very short time.
Since the the time of this talk, provocative and somewhat overstated press releases have been issued from CERN stating that “the experiments on CERN’s Heavy Ion programme presented compelling evidence for the existence of a new state of matter in which quarks, instead of being bound up into more complex particles such as protons and neutrons, are liberated to roam freely. … We now have evidence of a new state of matter where quarks and gluons are not confined.”
As discussed below, I disagree with the above interpretation of that truly impressive body of data. What is compelling is that some form of matter, much denser than ever studied before, was created. Inferences about quark and gluon degrees of freedom are based on qualitative scenarios and schematic models. At the relatively low momentum scales accessible at SPS energies, the quarks and gluon degrees of freedom in the dense matter are mostly not resolvable. Even at the highest transverse momentum, the pion spectra were shown to be very sensitive to nonperturbative model assumptions such as to the magnitude of intrinsic momenta and soft initial state interactions. The dynamics of the non-perturbative beam jet fragmentation and hadronic final state interactions cannot be disentangled at SPS. It is useful to recall that in experiments on $`p\overline{p}`$ scattering, it was only possibly to see unambiguous evidence for the tell-tale Rutherford scattering of point-like partons when collider energies $`\sqrt{s}>2002000`$ GeV became available. At RHIC high $`p_{}>10`$ GeV probes become kinematically available and hence very small wavelength resolution of partonic degrees of freedom finally becomes kinematically possible. While there is an abundance of interesting signatures showing that dense matter was formed at the SPS (through the non-linear in dependence of several observables on multiplicity or $`A`$), the bottom line is that those data have said nothing about whether the QCD predictions in Figs 1-3 are correct or not. We simply need higher resolution. RHIC, with its factor of ten increase in C.M. energy reaches a factor of ten deeper into the new phase reaches. Coupled with the availability of ten times shorter wavelength probes, it should finally become possible to actually see direct evidence of “freely roaming quarks and gluons”.
### 2.1 Global Signatures of Collective Dynamics
The simplest but often ignored global barometer of collectivity in nuclear reactions is the A and energy dependence of the transverse energy and charged particle rapidity density. At RHIC energies, HIJING predicts that initial transverse energy density in central $`A+A`$ collisions scales nonlinearly with A
$$\frac{dE_{}}{dy}1\mathrm{GeV}A^{1.3}(1+\mathrm{log}\frac{\sqrt{s}}{200})$$
(2.5)
This leads to about $`1`$ TeV per unit rapidity. In , on the other hand, it was found that gluon saturation could limit the A dependence to approximately linear, $`E_{}A^{1.04}`$, but with a value several times that of HIJING. In Fig.(4), the initial gluon density actually grows less than linear $`A^{0.92}`$ in that model. The $`A^{1.3}`$ scaling of HIJING with its conservative $`p_0=2`$ GeV scale fixed by $`p\overline{p}`$ data is simply due to the number of binary interactions via (2.3). The initial energy density in HIJING varies approximately as
$$ϵ_00.6\mathrm{GeV}/\mathrm{fm}^3A^{0.63}(1+\mathrm{log}\frac{\sqrt{s}}{200}).$$
(2.6)
In its magnitude and scaling are predicted to go as $`ϵ(\tau =1/p_{sat})0.1A^{0.5}s^{0.38}`$.
If local equilibrium is achieved and maintained, then a very basic prediction of hydrodynamics is that longitudinal boost invariant expansion together with $`pdV`$ work done by pushing matter down the beam pipe will cool the plasma and convert some its random transverse energy into collective longitudinal kinetic energy. For an equation of state, $`p=c_s^2ϵ`$, this cooling and expansion causes the energy density to decrease with proper time as
$$ϵ(\tau )=ϵ(\tau _0)\left(\frac{\tau _0}{\tau }\right)^{1+c_s^2}$$
(2.7)
Entropy conservation leads to a conservation of $`\tau n(\tau )`$, where $`n`$ is the proper parton density. At least if the matter is initially deep in the plasma phase, then (as seen in Fig.3) longitudinal will be done with $`c_s^21/3`$. Consequently, the transverse energy per particle should decrease by a factor 2-3 before freeze-out as
$$e_{}(\tau )=\frac{dE_{}}{dN}=e_{}(\tau _0)\left(\frac{\tau _0}{\tau }\right)^{c_s^2}$$
(2.8)
However, dissipative effects due to finite mean free paths reduce the effective pressure in any system. For the Bjorken expansion, the relaxation time, $`\tau _c=1/(\sigma _Tn)\tau `$, then increases with time as $`n`$ decreases. Numerical solution of 3+1 D transport equations with pQCD cross sections, $`\sigma _T2`$ mb, indicate that dissipation reduces the transverse energy loss for HIJING initial conditions rather significantly (see detailed comparisons in ). Ideal hydrodynamics predicts that about one half of the initial produced transverse energy goes into longitudinal work
It is important to keep in mind that the commonly assumed freeze-out prescription with $`\tau _f`$ fixed on a fixed freeze-out energy density hypersurface, $`ϵ(\tau _f)=ϵ_f0.15`$ GeV/fm<sup>3</sup> is only a rough prescription that is never accurate. As the interaction size decreases ($`A`$ or multiplicity decreases), the effective speed of sound due to dissipation decreases and the system freezes out earlier. A detailed study of the A or multiplicity dependence of $`dE_{}/dy`$ and $`e_T`$ is needed to calibrate the interplay of pdV work and dissipation on these barometers.
One of the major experimental discoveries of WA98, NA49 and the other SPS experiments is that at those energies $`dE_{}/dy`$ as well as $`dN/dy`$ scale with $`A`$ or wounded nucleon number nearly linearly as shown in Fig.(6).
Both the $`E_{}`$ and the charge multiplicity increase as $`A^{1.07}`$. These findings differ from the VENUS model, which has considerable nonlinearly due to the assumed sea string contributions. In fact, simple Glauber wounded nucleon models reproduce very well the nearly linear correlation between $`E_{}`$ and the veto calorimeter (spectator) energy observed in all experiments at SPS. The implications of these data depend on the $`A`$ dependence of the initial conditions of course. One view is that the initial $`e_T`$ can scale arbitrarily with $`A`$, but because perfect local equilibrium is maintained up to a critical sharp freeze-out hypersurface, the final $`E_T`$ and $`e_T`$ always scale linearly with $`A`$. My view is that at the SPS, pQCD is mostly inapplicable to bulk phenomena and the linear dependence arises from additive nature of soft beam fragmentation together with the absence of $`pdV`$ work at early times. If the QGP transition region is just barely reached, as I believe without experimental proof, then the softness of the QCD equation of state with $`c_s^21/3`$ seen in Fig.(3) and dissipation can conspire to prevent the dense matter from performing longitudinal work. However, it is impossible to tell from the data whether the observed null effect is then due to a low pressure Hagedorn resonance gas of hadrons or to a low pressure lazy “plasma” with $`c_s1`$. One has to go to higher energies to the plasma a chance to work.
In spite of the above remarks, ideal Euler hydrodynamic calculations have in fact been able to fit the null effect in the data quite well. This is possible by assuming arbitrary initial conditions chosen such that that after hydro does its work, the final results just happens to reproduce the data. An illustration of this degree of freedom is shown in Fig.(7) from . By adjusting the initial four velocity field appropriately, the calculated final pion distributions can be made to reproduce the data starting from two completely different initial energy density profiles. The same good fit to data can be obtained also for ANY assumed equation of state of dense matter simply by readjusting the initial conditions appropriately. Therefore, it is not surprising that even hand calculator fireball models are able to fit much of the data without invoking any dynamical or thermodynamical assumptions about the properties of dense matter prior to freeze-out . It is clear that inferring the existence of freely roaming quarks and gluons based on such fits is impossible .
In order to use the global barometers to search for evidence of collective phenomena in dense matter at RHIC, it is first necessary to eliminate the freedom to choose arbitrary initial conditions. At collider energies, pQCD and non-Abelian field techniques provide the needed theoretical calibration tools to fix initial conditions At RHIC and LHC, the initial plasma is expected in any case to be so deep in the deconfined phase that the soft $`pϵ/3`$ transition region and dissipation cannot spoil the longitudinal work as is the case at SPS. From detailed covariant transport calculations the barometric evidence for longitudinal work should finally be observable in spite of finite mean free path effects. Deconvolution of the equation of state and dissipative corrections requires however a detailed study of the $`A`$ and multiplicity dependence of the global $`E_T`$ barometer.
This is the first important signatures to look forward to at RHIC. While it is not possible to converge on why no longitudinal work was observed at the SPS, a positive signature at RHIC would render such discussion mute. We can only be sure that a plasma was created if it does something collective! The global $`E_T`$ and multiplicity systematics as a function of centrality as well as $`A`$ therefore provide key handles in this search.
## 3 Transverse Flow
In contrast to global transverse energy barometer a completely different measure of barometric collectivity is afforded by the study of the triple differential distributions, $`dN/dyd^2\stackrel{}{p}_{}`$. Already at sub-luminal Bevalac energies ($`<1`$ AGeV), azimuthally asymmetric collective directed and elliptic flow were discovered long ago. For non central collisions, $`b0`$, the asymmetric transverse coordinate profile of the reaction region leads to different gradients of the pressure as a function of the azimuthal angle relative to beam axis. This leads to a “bounce” off of projectile and target fragments in the reaction plane and to azimuthally asymmetric transverse momentum dependence of particles with short mean free at mid rapidities. This phenomenon has now been observed at both AGS and SPS energies as well. It will certainly be there also at RHIC and LHC. In Fig.(8) the first two Fourier components of the azimuthal flow patterns are shown:
$$\frac{dN}{dyd^2\stackrel{}{p}_{}}=v_0(1+2v_1\mathrm{cos}(\varphi \varphi _R)+2v_2\mathrm{cos}(2(\varphi \varphi _R))+\mathrm{})$$
(3.1)
Azimuthal asymmetric collectivity is clearly observed at the SPS. The important question is how this type of barometer could serve to help search for evidence of the QCD transition. Unlike the global $`E_{}`$ barometer discussed in the previous section, transverse flow can develop at later times because the gradients are controlled by the transverse size of the nucleus, not the proper time interval relative to formation. In the idea was proposed that one could use $`v_2`$ for example to study the predicted softening of the QCD equation of state. Typically, hydrodynamics calculations lead to a factor of two smaller $`v_2`$ for an equation of state with a soft critical point as in Fig 3 versus one in which the speed of sound remains 1$`/\sqrt{3}`$. Searches for anomalous $`v_2`$ dependence as well as $`v_1`$ are underway. As with the global barometer, dissipation can of course also simulate a soft equation of state. In ref we studied the dependence of $`v_2`$ on the transport parton cross section. The results shown in Fig(8) for HIJING initial conditions indicate that there is indeed a significant reduction of $`v_2`$ relative to hydrodynamics, but that the impact parameter (multiplicity) dependence of that observable can again be used to disentangle the equation of state versus viscosity effects as in the case of the global $`E_T`$ barometer.
## 4 The QGP Stall and Time delay
Hadron interferometry has been developed into a fine art in heavy ion collisions to image the space-time region of the decoupling 4-volume. In it was proposed that a possible signature of the QGP phase transition would be a time delay associated with very slow hadronization. The plasma ”burns” into hadronic ashes along deflagration front that moves very slowly if the entropy drop across the transition is large. Fig.9 shows the evolution of a Bjorken cylinder with time and transverse coordinate from a hydrodynamic simulation with different equations of state from .
The main point to note is that time delay is a robust generic signature of a rapid cross over transition of the entropy density. In particular the ”stall” is expected even for a smooth cross-over transition as long as the width $`\mathrm{\Delta }T/T_c<0.1`$. However, its magnitude also depends on the entropy drop across that region. The figures are for an entropy drop by a factor 10 consistent with the lattice QCD results. Unfortunately, as noted before lattice QCD has not yet resolved the hadronic world below $`T_c`$ due to numerical problems. If the entropy jump is much smaller, then this signature would also disappear.
High statistics measurements of pion and kaon interferometry searches for time delay at AGS and SPS have come up empty handed thus far. No time delay has ever been observed in any nuclear reactions thus far. This could be due to (a) the absence of a large rapid entropy drop in real QCD, or (b) to unfavorable kinematic conditions at AGS and SPS energies. From the hydrodynamic calculations in , it was found that a large time delay signal requires that the initial energy density be deep within the plasma phase. The slow deflagration waves seen in Fig.9 do not arise if the initial energy density is close to the transition region because the initial longitudinal expansion cools the plasma too rapidly. The optimal conditions to see this effect was predicted to occur at RHIC energies. For much higher energies (LHC), on the other hand, the transverse expansion of the plasma has too much time to develop and that spoils the possibility of a slowly burning plasma stall. I note that further work has shown that high $`p_{}`$ kaon interferometry adds an especially sensitivity handle in the search for this time delay signature. If ever observed, the time delay signature would be smoking gun that a new form of matter with bulk collective properties was created.
## 5 The J/Psi puzzle
In 1986, Matsui and Satz proposed an intriguing direct measure of the transmutation of the $`q\overline{q}`$ forces in Fig. 1. The idea was that $`J/\psi `$ can form in the vacuum because the confining Luscher potential can bind a $`c\overline{c}`$ pair into that vector meson. If that pair were placed in a hot medium in which the chromo-Debye screening potential is short ranged, then above the temperature where the screening length is smaller than the $`J/\psi `$ radius, the $`c\overline{c}`$ would become unbound and and the charm quarks would emerge from the reaction region as an open charm $`D\overline{D}`$ pair. They thus predicted that $`J/\psi `$ suppression would be a smoking gun for the deconfinement transition.
$`J/\psi `$ suppression was first seen in 1987 in $`O+U`$ reactions by NA38. Since then this smoking gun has (unfortunately) never stopped smoking! $`J/\psi `$ suppression seems to be as ubiquitous as the Yukawa potential. It is now clearly observed in $`p+A`$ reactions as seen in Fig.(10). High mass Drell-Yan pairs, on the other hand, formed via $`q\overline{q}\mathrm{}\overline{\mathrm{}}`$ was observed to scale perfectly linearly with the number of binary collisions. This is because lepton pairs suffer no final state interactions and the quark initial state (Cronin) interactions are invisible in $`p_{}`$ integrated DY yields.
$`J/\psi `$ are suppressed on the other hand as $`(AB)^{0.9}`$ due to some nuclear effect that is independent of QGP production. In the recent Pb+Pb analysis an excess 25% suppression of $`J/\psi `$ was observed. NA50 claims that this enhanced suppression relative to the $`(AB)^{0.9}`$ trend from lighter projectile $`p,O,S+U`$ data is finally the real smoking gun.
From the suppression pattern in Pb as observed as a function of centrality (see Fig.(11)),
NA50 has now claimed that in fact “Together with the results previously established by the NA38 and NA50 collaborations, a rather clear picture emerges, indicating a step-wise pattern, with no visible saturation in the collisions generating the highest energy densities and temperatures. Our observations exclude the presently available models of $`J/\psi `$ suppression based on the absorption of the $`J/\psi `$ mesons by interactions with the surrounding hadronic (confined) matter. The first anomalous step can be understood as due to the disappearance of the $`\chi `$ mesons, responsible for a fraction of the observed $`J/\psi `$ yield through its (experimentally unidentified) radiative decay. In proton induced collisions this fraction is around 30-40%. The second drop signals the presence of energy densities high enough to also dissolve the more tightly bound $`J/\psi `$ charmonium state.”
While the deviation from the empirical $`(AB)^{0.9}`$ scaling is very clear, the dynamic origin of the effect is far from clear in my opinion. As shown by the several curves in Fig.(11), a $``$50% drop in the $`J/\psi `$ yield as a function of $`E_T`$ is consistent with final state co-moving hadronic absorption. While the details are not reproduced accurately, large theoretical uncertainties about several key dynamical ingredients preclude precise comparisons at this time. It is important to emphasize that the so called “conventional hadronic” models suffer from just as large theoretical uncertainties as the plasma scenario models. Key uncertain elements include (1) the dynamical treatment of cold nuclear absorption responsible for the $`AB^{0.9}`$ suppression even in $`p+A`$, (2) the unknown hadronic $`M+\psi D\overline{D}X`$ reaction rates, and (3) the actual density evolution, $`ϵ(\stackrel{}{x},t)`$ needed in order to make more precise calculations. Only a schematic “octet model” of pre-hadronization $`c\overline{c}`$ interactions in cold nuclei has been used in the above transport models to address the first effect. That this is highly uncertain is shown in the work of Ref.. The observed $`J/\psi `$ production cross section even without final state interactions is suppressed as follows:
$$\sigma _{AB\psi X}=𝑑\sigma _{ABc\overline{c}}F_{c\overline{c}\psi }(q^2)$$
(5.1)
where $`F`$ is the formation probability of the $`\psi `$ from a $`c\overline{c}`$ that emerges from the cold nuclear target with an invariant mass $`q^2<4M_D^2`$. If the pre-resonance $`c\overline{c}`$ pair multiply scatters in the nucleus random walk would increase $`q^2q^2+\delta q^2(\sigma \rho L)`$ linearly with nuclear thickness as shown by the curve G in Fig.(12). This leads to an approximate exponential suppression that can be fit well by an approximate Glauber nuclear absorption factor ansatz, $`\mathrm{exp}(\sigma _{eff}\rho L)`$ if a Gaussian assumption is made. This Gaussian model can thus account for the observed $`(AB)^{0.9}`$ scaling light projectiles. However, it was shown in that if the power law tails due to induced radiation in the medium is included (resulting from the multiple Rutherford rescattering of the color octet $`c\overline{c}`$) , then an additional nonlinear suppression in the nuclear thickness $`L`$ could result (curve P). This is because radiation provides another way to increase the invariant mass of the pre-resonance $`c\overline{c}`$ that can further reduce the probability for the pair to fit inside the $`\psi `$ wavefunction.
While this model is also schematic, I feel that it captures an important element of pre-formation transient dynamical effects in nuclei that can serve as additional sources of nonlinear suppression, without even considering the sought after suppression in the comoving dense matter. The transport model curves in Fig.11 co-moving dissociation possesses and pre-resonance nuclear dynamics as in Fig.12 should be combined in future analysis searching for the dynamical origin of beyond nuclear Glauber absorption in $`Pb+Pb`$.
Therefore, the claim of anomalous suppression cannot rest therefore merely on generic enhanced suppression in $`Pb`$. It must and has been made on the possible existence of singular “step-like” structure of the suppression pattern. The evidence for ”steps” is however the weakest link experimentally because a rigorous $`\chi ^2`$ test including the substantial systematic errors in the $`E_T`$ scale has yet to be performed. This very difficult experiment has fought valiantly for years to reduce the systematic errors associated with calibration of $`E_T`$, thick target multiple interactions, and calibrating the Drell-Yan and other backgrounds.
This is illustrated in Fig.(13) where plasma scenario fits to 1997 data are also shown. In Fig.(14) recent plasma scenario fits to the latest data show how much the data has evolved (see for details). It is also clear from Figs.(13,14) that the plasma scenario step-function models while capable of fitting the data cannot be be considered as prediction of QCD. Percolation models serve to motivate the steps.
Nevertheless, if the “step-like” suppression pattern survives further experimental scrutiny, it would certainly be the most dramatic nonlinearity observed at SPS. In this connection, it is also important to remark that the energy density scale in Fig.(11 b) is uncertain to at least a factor of two. There is no direct experimental measurement of the initial energy density. The energy scale there is infered from a RQMD model calculation. However, as shown in Fig.(7) a nice hydro fit to the data could be obtained with an initial energy density ten times higher than assumed in Fig.11 b.
Given the boldness of the NA50 claims, it is important to scrutinize both the experimental and theoretical foundations on which those claims are made. In science guilt is to assumed until innocence is proven. Based on the previous discussion, I remain skeptical.
An additional problem with the plasma scenario interpretations can be seen form the observed $`E_T`$ dependence of the $`J/\psi `$ $`p_{}`$ spectra in Fig.(LABEL:fig:psipt). Standard Glauber multiple collisions lead to a random walk in transverse momentum that are expected to lead to
$$p_{}^2_{AB}=p_{}^2_{pp}+\frac{L}{\lambda }\delta p_{}^2$$
(5.2)
his is found to hold in all reaction including $`Pb+Pb`$. In contrast, in the plasma scenario, only those $`\psi `$ are expected to survive that are near the surface where the nuclear depth $`L`$ is small. Thus the prediction as shown in Fig(LABEL:fig:psipt) was that the $`p_{}^2`$ should begin to DECREASE with increasing $`E_T`$. This was not observed.
Clearly, much more work is required to sort out the very interesting suppression pattern observed by NA38,NA50. Experimentally, the claims would carry considerable more punch if similar ”step-wise” patternswere observed in other systems ,e.g. $`Xe+Xe`$ suitably shifted in $`E_T`$ due to the expected smaller energy densities achieved. While it is not likely that such further measurements will get done, they will be at RHIC. A clear prediction by H. Satz at QM99, was that under RHIC conditions of higher energy density, the same step wise pattern as in Fig.(14) should be observable in $`Cu+Cu`$ interactions. PHENIX will provide a definitive test of this prediction soon.
## 6 The High $`p_{}`$ Frontier
One of the new areas that RHIC will open in the experimental search for the next Yukawa phase of QCD is high transverse momentum (short wavelength) jet probes. The rates of jet production and its fragmentation in the vacuum are well understood. The new physics here is the study of partonic interactions at extreme densities through the phenomenon of jet quenching. Final state interactions of a jet in a dense QGP are expected to induce a large radiative energy loss. In fact, it was discovered in by BDMPS that non-Abelian energy loss is in fact non-linear as a function of the thickness of the medium. Tests of this and other aspects of non-Abelian multiple collision dynamical will be possible at RHIC.
At SPS energies, this physics is out of reach given the dominance of nonperturbative effects as shown in Fig.(16) from ref. . HIJING accidently fits the WA98 data with or without jet quenching. At the SPS no clean separation of soft and hard dynamics is kinematically possible. However, at RHIC energies, the power law tails of the single inclusive distributions stick far enough above the confusing soft ”noise” to gain sensitivity to the form of the non-Abelian $`dE/dx`$ as seen in Fig.(16).
This problem is closely related also to the problem of pre-resonance $`J/\psi `$ absorption discussed previously. Understanding jet quenching is also prerequisite in testing the dynamical assumptions of recent covariant parton transport theories.
## 7 Summary
The next Yukawa phase is awaiting discovery. The SPS data have provided many intriguing indirect hints that new physics is operating in dense matter. Many puzzles, claims and counter claims remain because at SPS energies, both hadronic and partonic models have partial overlapping domains of validity. This “duality” is analogous to the problem of interpreting the R factor in $`e^+e^{}`$ collisions below $`\sqrt{s}<10`$ GeV. The ratio of hadronic to leptonic production cross sections only reaches the the magic 11/3 of pQCD above that threshold region where the $`s\overline{s},c\overline{c},b\overline{b}T`$ vector mesons dominate the nonperturbative hadronic physics. Similarly SPS is at the door-step where hardonic resonances begin to melt away and a pQCD continuum description start to become more relevant.
With RHIC the factor of ten increase in the initial energy density will be unambigously in the QGP continuum. The matter so formed will also have much more time to develop collective signatures. The factor of ten smaller wavelength probes will finally allow experimentalists to resolve (i.e. see) the quark and gluon degrees of freedom of that plasma. Direct observation of longitudinal work, transverse azimuthal collectivity, time delay, step-wise $`J/\psi `$ suppression in $`Cu+Cu`$, and jet quenching will offer direct signatures of the sought after new phase of QCD matter.
## Acknowledgements
I thank Profs. H. Horiuchi and T. Hatsuda for organizing this Yukawa symposium. I would especially acknowledge Mayor Baba and the city of Nishinomia their support of this symposium series. This work was supported by the Director, Office of Energy Research, Division of Nuclear Physics of the Office of High Energy and Nuclear Physics of the U.S. Department of Energy under Contract No. DE-FG-02-93ER-40764.
|
warning/0004/gr-qc0004075.html
|
ar5iv
|
text
|
# Gravitational wave bursts from cosmic strings
\[
## Abstract
Cusps of cosmic strings emit strong beams of high-frequency gravitational waves (GW). As a consequence of these beams, the stochastic ensemble of gravitational waves generated by a cosmological network of oscillating loops is strongly non Gaussian, and includes occasional sharp bursts that stand above the rms GW background. These bursts might be detectable by the planned GW detectors LIGO/VIRGO and LISA for string tensions as small as $`G\mu 10^{13}`$. The GW bursts discussed here might be accompanied by Gamma Ray Bursts.
04.30.Db, 95.85.Sz, 98.80.Cq
\]
Cosmic strings are linear topological defects that could be formed at a symmetry breaking phase transition in the early universe. Strings are predicted in a wide class of elementary particle models and can give rise to a variety of astrophysical phenomena . In particular, oscillating loops of string can generate a potentially observable gravitational wave (GW) background ranging over many decades in frequency. The spectrum of this stochastic background has been extensively discussed in the literature , but until now it has been tacitly assumed that the GW background is nearly Gaussian. In this paper, we show that the GW background from strings is strongly non-Gaussian and includes sharp GW bursts (GWB) emanating from cosmic string cusps . We shall estimate the amplitude, frequency spectrum and rate of the bursts, and discuss their detectability by the planned GW detectors LIGO/VIRGO and LISA.
We begin with a brief summary of the relevant string properties and evolution. A horizon-size volume at any cosmic time $`t`$ contains a few long strings stretching across the volume and a large number of small closed loops. The typical length and number density of loops formed at time $`t`$ are approximately given by
$$l\alpha t,n_l(t)\alpha ^1t^3.$$
(1)
The exact value of the parameter $`\alpha `$ in (1) is not known. We shall assume, following , that $`\alpha `$ is determined by the gravitational backreaction, so that $`\alpha \mathrm{\Gamma }G\mu `$, where $`\mathrm{\Gamma }50`$ is a numerical coefficient, $`G`$ is Newton’s constant, and $`\mu `$ is the string tension, i.e. the mass per unit length of the string. The coefficient $`\mathrm{\Gamma }`$ enters the total rate of energy loss by gravitational radiation $`d/dt\mathrm{\Gamma }G\mu ^2.`$ For a loop of invariant length $`l`$ , the oscillation period is $`T_l=l/2`$ and the lifetime is $`\tau _ll/\mathrm{\Gamma }G\mu t`$.
A substantial part of the radiated energy is emitted from near-cusp regions where, for a short period of time, the string reaches a speed very close to the speed of light . Cusps tend to be formed a few times during each oscillation period . Let us estimate the (trace-reversed) metric perturbation $`\overline{h}_{\mu \nu }=h_{\mu \nu }\frac{1}{2}h\eta _{\mu \nu }`$ emitted near a cusp. Let $`\kappa _{\mu \nu }=r_{\mathrm{phys}}\overline{h}_{\mu \nu }`$ denote the product of the metric perturbation by the distance away from the string, estimated in the local wave-zone of the oscillating string. $`\kappa _{\mu \nu }`$ is given by a Fourier series whose coefficients are proportional to the Fourier transform of the stress-energy tensor of the string:
$$T^{\mu \nu }(k^\lambda )=T_l^1_{T_l}d\tau d\sigma \mu (\dot{X}^\mu \dot{X}^\nu X^\mu X^\nu )\mathrm{exp}(ik.X).$$
(2)
Here, $`X^\mu (\tau ,\sigma )`$ represents the string worldsheet, parametrized by the conformal coordinates $`\tau `$ and $`\sigma `$. \[$`\dot{X}=_\tau X`$, $`X^{}=_\sigma X`$.\] In the direction of emission $`𝐧`$, $`k^\mu =(\omega ,𝐤)`$ runs over the discrete set of values $`4\pi l^1m(1,𝐧)`$, where $`m=1,2,\mathrm{}`$. Near a cusp (and only near a cusp) the Fourier series giving $`\kappa _{\mu \nu }`$ is dominated by large $`m`$ values, and can be approximated by a continuous Fourier integral. The continuous Fourier component (corresponding to an octave of frequency around $`f`$) $`\kappa (f)|f|\stackrel{~}{\kappa }(f)|f|𝑑t\mathrm{exp}(2\pi ift)\kappa (t)`$ is then given by
$$\kappa _{\mu \nu }(f)=2Gl|f|T_{\mu \nu }(k^\lambda ).$$
(3)
In a conformal gauge, the string motion is given by $`X^\mu =\frac{1}{2}(X_+^\mu (\sigma _+)+X_{}^\mu (\sigma _{}))`$,where $`\sigma _\pm =\tau \pm \sigma `$, and where $`\dot{X}_\pm ^\mu `$ is a null 4-vector. Further choosing a“time gauge” ($`\tau =X^0=t`$ ) ensures that the time components of these null vectors are equal to 1. A cusp is a point on the worldsheet where these two null vectors coincide, say $`\dot{X}_{c\pm }^\mu =l^\mu =(1,𝐧_𝐜)`$. One can estimate the waveform (3) emitted near the cusp ( localized, say, at $`\sigma _+=\sigma _{}=0`$) by replacing in the integral (2) $`X_\pm ^\mu `$ by their local Taylor expansions
$$X_\pm ^\mu (\sigma _\pm )=X_{c\pm }^\mu +l^\mu \sigma _\pm +\frac{1}{2}\ddot{X}_\pm ^\mu \sigma _\pm ^2+\frac{1}{6}\stackrel{\mathrm{}}{X}_\pm ^\mu \sigma _\pm ^3+\mathrm{}$$
(4)
For a given frequency $`fT_l^1`$, the integral (2) is significant only if the angle $`\theta `$ between the direction of emission $`𝐧`$ and the “3-velocity” of the cusp $`𝐧_𝐜)`$ is smaller than about $`\theta _m(T_l|f|)^{1/3}`$. When $`\theta \theta _m`$, the integral can be explicitly evaluated. After a suitable gauge transformation one finds
$$\kappa ^{\mu \nu }(f)=CG\mu (2\pi |f|)^{1/3}A_+^{(\mu }A_{}^{\nu )},$$
(5)
where $`C=4\pi (12)^{4/3}(3\mathrm{\Gamma }(1/3))^2`$ and where the linear polarization tensor is the symmetric tensor product of $`A_\pm ^\mu \ddot{X}_\pm ^\mu /|\ddot{X}_\pm |^{4/3}`$. The inverse Fourier transform of Eq.(5) gives a time-domain waveform proportional to $`|tt_c|^{1/3}`$ , where $`t_c`$ corresponds to the peak of the burst . The sharp spike at $`t=t_c`$ exists only if $`\theta =0`$ (i.e. if one observes it exactly in the direction defined by the cusp velocity). When $`0\theta 1`$ the spike is smoothed over $`|tt_c|\theta ^3T_l`$. In the Fourier domain this smoothing corresponds to an exponential decay for frequencies $`|f|1/(\theta ^3T_l)`$.
Eq.(5) gives the waveform in the local wave-zone of the oscillating loop: $`\overline{h}_{\mu \nu }=\kappa _{\mu \nu }/r_{\mathrm{phys}}`$. To take into account the subsequent propagation of this wave over cosmological distances, until it reaches us, one must introduce three modifications in this waveform: (i) replace $`r_{\mathrm{phys}}`$ by $`a_0r`$ where $`r`$ is the comoving radial coordinate in a Friedman universe (taken to be flat: $`ds^2=dt^2+a(t)^2(dr^2+r^2d\mathrm{\Omega }^2)`$) and $`a_0=a(t_0)`$ the present scale factor , (ii) express the locally emitted frequency in terms of the observed one $`f_{\mathrm{em}}=(1+z)f_{\mathrm{obs}}`$ where $`z`$ is the redshift of the source, and (iii) transport the polarization tensor of the wave by parallel propagation (pp) along the null geodesic followed by the GW:
$$\overline{h}_{\mu \nu }(f)=\kappa _{\mu \nu }^{\mathrm{pp}}((1+z)f)/(a_0r).$$
(6)
Here, and henceforth, $`f>0`$ denotes the observed frequency. In terms of the redshift we have $`a_0r=3t_0(1(1+z)^{1/2})`$, where $`t_0`$ is the present age of the universe (this relation holds during the matter era, and can be used for the present purpose in the radiation era because $`a_0r`$ has a finite limit for large $`z`$).
For our order-of-magnitude estimates we shall assume that $`|\ddot{X}_\pm |2\pi /l`$. The various numerical factors in the equations above nearly compensate each other to give the following simple estimate for the observed waveform in the frequency domain ($`h(f)|f|\stackrel{~}{h}(f)|f|𝑑t\mathrm{exp}(2\pi ift)h(t)`$)
$$h(f)\frac{G\mu l}{((1+z)fl)^{1/3}}\frac{1+z}{t_0z}.$$
(7)
Here the explicit redshift dependence is a convenient simplification of the exact one given above. This result holds only if, for a given observed frequency $`f`$, the angle $`\theta =\mathrm{cos}^1(𝐧𝐧_c)`$ satisfies
$$\theta \theta _m((1+z)fl/2)^{1/3}.$$
(8)
To know the full dependence of $`h(f)`$ on the redshift we need to express $`l\alpha t`$ in terms of $`z`$. We write
$$l\alpha t_0\phi _l(z);\phi _l(z)=(1+z)^{3/2}(1+z/z_{eq})^{1/2}.$$
(9)
Here $`z_{eq}2.4\times 10^4\mathrm{\Omega }_0h_0^210^{3.9}`$ is the redshift of equal matter and radiation densities, and we found it convenient to define the function $`\phi _l(z)`$ which interpolates between the different functional $`z`$-dependences of $`l`$ in the matter era, and the radiation era. \[We shall systematically introduce such interpolating functions of $`z`$, valid for all redshifts, in the following.\] Inserting Eq.(9) into Eq.(7) yields
$`h(f,z)G\mu \alpha ^{2/3}(ft_0)^{1/3}\phi _h(z),`$ (10)
$`\phi _h(z)=z^1(1+z)^{1/3}(1+z/z_{eq})^{1/3}.`$ (11)
We can estimate the rate of GWBs originating at cusps in the redshift interval $`dz`$, and observed around the frequency $`f`$, as $`d\dot{N}\frac{1}{4}\theta _m^2(1+z)^1\nu (z)dV(z)`$. Here, the first factor is the beaming fraction within the cone of maximal angle $`\theta _m(f,z)`$, Eq.(8), the second factor comes from the relation $`dt_{\mathrm{obs}}=(1+z)dt`$, $`\nu (t)f_cn_l(t)/T_l2\alpha ^2t^4`$ is the number of cusp events per unit spacetime volume, $`f_c1`$ is the average number of cusps per oscillation period of a loop, $`T_l\alpha t/2`$, and $`dV(z)`$ is the proper volume between redshifts $`z`$ and $`z+dz`$. In the matter era $`dV=54\pi t_0^3[(1+z)^{1/2}1]^2(1+z)^{11/2}dz`$, while in the radiation era $`dV=72\pi t_0^3(1+z_{eq})^{1/2}(1+z)^5dz`$. The function $`\dot{N}(f,z)d\dot{N}/d\mathrm{ln}z`$ can be approximately represented by the following interpolating function of $`z`$
$`\dot{N}(f,z)10^2t_0^1\alpha ^{8/3}(ft_0)^{2/3}\phi _n(z),`$ (12)
$`\phi _n(z)=z^3(1+z)^{7/6}(1+z/z_{eq})^{11/6}.`$ (13)
The observationally most relevant question is: what is the typical amplitude of bursts $`h_{\dot{N}}^{\mathrm{burst}}(f)`$that we can expect to detect at some given rate $`\dot{N}`$, say, one per year? Using $`\dot{N}=_0^{z_m}\dot{N}(f,z)d\mathrm{ln}z\dot{N}(f,z_m)`$, where $`z_m`$ is the largest redshift contributing to $`\dot{N}`$, one can estimate $`h_{\dot{N}}^{\mathrm{burst}}(f)`$ by solving for $`z`$ in Eq.(13) and substituting the result $`z=z_m(\dot{N},f)`$ in Eq.(11). The final answer has a different functional form depending on the magnitude of the quantity
$$y(\dot{N},f)10^2\dot{N}t_0\alpha ^{8/3}(ft_0)^{2/3}.$$
(14)
Indeed, if $`y<1`$ the dominant redshift will be $`z_m(y)<1`$; while, if $`1<y<z_{eq}^{}{}_{}{}^{11/6}`$, $`1<z_m(y)<z_{eq}`$, and if $`y>z_{eq}^{}{}_{}{}^{11/6}`$, $`z_m(y)>z_{eq}`$. We can again introduce a suitable interpolating function $`g(y)`$ to represent the final result as an explicit function of $`y`$:
$`h_{\dot{N}}^{\mathrm{burst}}(f)G\mu \alpha ^{2/3}(ft_0)^{1/3}g[y(\dot{N},f)],`$ (15)
$`g(y)=y^{1/3}(1+y)^{13/33}(1+y/(z_{eq})^{11/6})^{3/11}.`$ (16)
The prediction Eq.(16) for the amplitude of the GWBs generated at cusps of cosmic strings is the main new result of this work. To see whether or not these bursts can be distinguished from the stochastic gravitational wave background, we have to compare the burst amplitude (16) to the rms amplitude of the background, $`h_{\mathrm{rms}}`$, at the same frequency. We define $`h_{\mathrm{rms}}`$ as the “confusion” part of the ensemble of bursts Eq.(16), i.e. the superposition of all the “overlapping” bursts, those whose occurrence rate is higher than their typical frequency. This is estimated by multiplying the square of Eq.(11) by the number of overlapping bursts within a frequency octave $`n_z(f)f^1\dot{N}(f,z)`$ , and then integrating over all $`\mathrm{ln}z`$ such that $`n_z(f)>1`$ , and $`\theta _m(f,z)<1`$:
$$h_{\mathrm{rms}}^2(f)=h^2(f,z)n_z(f)d\mathrm{ln}zH(n_z1)H(1\theta _m),$$
(17)
where $`H`$ denotes Heaviside’s step function. Eq.(17) differs from previous estimates of the stochastic background (beyond the fact that we use the simplified loop density model Eq.(1)) in that the latter did not incorporate the restriction to $`n_z(f)>1`$, i.e. they included non-overlapping bursts in the average of the squared GW amplitude.
It is easily checked from Eq.(16) that $`h^{\mathrm{burst}}`$ is a monotonically decreasing function of both $`\dot{N}`$ and $`f`$. These decays can be described by (approximate) power laws , with an index which depends on the relevant range of dominant redshifts; e.g., as $`\dot{N}`$ increases, $`h^{\mathrm{burst}}`$ decreases first like $`\dot{N}^{1/3}`$ ( in the range $`z_m<1`$), then like $`\dot{N}^{8/11}`$ (when $`1<z_m<z_{eq}`$), and finally like $`\dot{N}^{5/11}`$ (when $`z_m>z_{eq}`$). For the frequency dependence of $`h^{\mathrm{burst}}`$ , the corresponding power-law indices are successively: $`5/9,9/11`$ and $`7/11`$. \[These slopes come from combining the basic $`f^{1/3}`$ dependence of the spectrum of each burst with the indirect dependence on $`f`$ of the dominant redshift $`z_m(\alpha ,\dot{N},f)`$.\] By contrast, when using our assumed link $`G\mu \alpha /50`$ between the string tension $`\mu `$ and the parameter $`\alpha `$, one finds that the index of the power-law dependence of $`h^{\mathrm{burst}}`$ upon $`\alpha `$ takes successively the values $`+7/9,3/11`$ and $`+5/11`$. Therefore, in a certain range of values of $`\alpha `$ (corresponding to $`1<z_m(\alpha ,\dot{N},f)<z_{eq}`$) the GWB amplitude (paradoxically) increases as one decreases $`\alpha `$ , i.e. $`G\mu `$.
In Fig. 1 we plot (as a solid line) the logarithm of the GW burst amplitude, $`\mathrm{log}_{10}(h^{\mathrm{burst}})`$, as a function of $`\mathrm{log}_{10}(\alpha )`$, for $`\dot{N}=1\mathrm{yr}^1`$, and for $`f=f_c=150`$ Hz. This central frequency is the optimal one for the detection of a $`f^{1/3}`$-spectrum burst by LIGO. We indicate on the same plot (as horizontal dashed lines) the (one sigma) noise levels $`h^{\mathrm{noise}}`$ of LIGO 1 (the initial detector), and LIGO 2 (its planned advanced configuration). The VIRGO detector has essentially the same noise level as LIGO 1 for the GW bursts considered here. These noise levels are defined so that the integrated optimal ( with a matched filter $`|f|^{1/3}`$) signal to noise ratio (SNR) for each detector is $`SNR=h^{\mathrm{burst}}(f_c)/h^{\mathrm{noise}}`$. The short-dashed line in the lower right corner is the rms GW amplitude, Eq.(17). One sees that the burst amplitude stands well above the stochastic background . Clearly the search by LIGO/VIRGO of the type of GW bursts discussed here is a sensitive probe of the existence of cosmic strings in a larger range of values of $`\alpha `$ than the usually considered search for a stochastic GW background.
From Fig. 1 we see that the discovery potential of ground-based GW interferometric detectors is richer than hitherto envisaged, as it could detect cosmic strings in the range $`\alpha 10^{10}`$, i.e. $`G\mu 10^{12}`$. Let us also note that the value of $`\alpha `$ suggested by the (superconducting-) cosmic-string Gamma Ray Burst (GRB) model of Ref., namely $`\alpha 10^8`$, nearly corresponds, in Fig. 1, to a local maximum of the GW burst amplitude. \[This local maximum corresponds to $`z_m1`$. The local minimum on its right corresponds to $`z_mz_{eq}`$.\] In view of the crudeness of our estimates, it is quite possible that LIGO 1/VIRGO might be sensitive enough to detect these GW bursts. Indeed, if one searches for GW bursts which are (nearly) coincident with (some) GRB the needed threshold for a convincing coincident detection is much closer to unity than in a blind search. \[ In a blind search, by two detectors, one probably needs SNRs $`4.4`$ to allow for the many possible arrival times. Note that the optimal filter, $`h^{\mathrm{template}}(f)=|f|^{1/3}`$, for our GWBs is parameter-free.\]
In Fig. 2 we plot $`\mathrm{log}_{10}(h^{\mathrm{burst}})`$ as a function of $`\mathrm{log}_{10}(\alpha )`$ for $`\dot{N}=1\mathrm{yr}^1`$, and for $`f=f_c=3.9\times 10^3`$ Hz. This frequency is the optimal one for the detection of a $`f^{1/3}`$ GWB by the planned spaceborne GW detector LISA. \[In determining the optimal SNR in LISA we combined the latest estimate of the instrumental noise with estimates of the galactic confusion noise .\] Fig. 2 compares $`h^{\mathrm{burst}}(f_c)`$ to both LISA’s (filtered) noise level $`h^{\mathrm{noise}}`$ and to the cosmic-string-generated stochastic background $`h^{\mathrm{rms}}`$, Eq.(17). The main differences with the previous plot are: (i) the signal strength, and the SNR, are typically much higher for LISA than for LIGO, (ii) though the GW burst signal still stands out well above the rms background, the latter is now higher than the (broad-band) detector noise in a wide range of values of $`\alpha `$. LISA is clearly a very sensitive probe of cosmic strings. It might detect GWBs for values of $`\alpha `$ as small as $`10^{11.6}`$. \[Again, a search in coincidence with GRBs would ease detection. Note, however, that, thanks to the lower frequency range, even a blind search by the (roughly) two independent arms of LISA would need a lower threshold, $`3`$, than LIGO.\]
We shall discuss elsewhere the consequences for the interpretation of the pulsar timing experiments of the GWB-induced non-Gaussianity of the stochastic GW background .
When comparing our results with observations, one should keep in mind that the model we used for cosmic strings involves a number of simplifying assumptions. (i) All loops at time $`t`$ were assumed to have length $`l\alpha t`$ with $`\alpha \mathrm{\Gamma }G\mu `$. It is possible, however, that the loops have a broad length distribution $`n(l,t)`$ and that the parameter $`\alpha `$ characterizing the typical loop length is in the range $`\mathrm{\Gamma }G\mu <\alpha 10^3`$. (Here, the upper bound comes from numerical simulations and the lower bound is due to the gravitational backreaction.) (ii) We also assumed that the loops are characterized by a single length scale $`l`$, with no wiggliness on smaller scales. Short-wavelength wiggles on scales $`\mathrm{\Gamma }G\mu t`$ are damped by gravitational backreaction, but some residual wiggliness may survive. As a result, the amplitude and the angular distribution of gravitational radiation from cusps may be modified. (iii) We assumed the simple, uniform estimate Eq.(1) for the space density of loops. This estimate may be accurate in the matter era but is probably too small by a factor $`10`$ in the radiation era . Taking into account this increase of $`n_l`$ would reinforce our conclusions in making easier (in some parameter range) the detection of GWBs.(iv) Finally, we disregarded the possibility of a nonzero cosmological constant which would introduce some quantitative changes in our estimates.
The work of A.V. was supported in part by the National Science Foundation.
|
warning/0004/hep-th0004122.html
|
ar5iv
|
text
|
# The torus and the Klein bottle amplitude of permutation orbifolds
## The torus and the Klein bottle amplitude of permutation orbifolds
Zoltan Kadar (zkadar@cs.elte.hu)
Eotvos University Budapest HUNGARY
Abstract
The torus and the Klein bottle amplitude coefficients are computed in permutation orbifolds of RCFT-s in terms of the same quantities in the original theory and the twist group. An explicit expression is presented for the number of self conjugate primaries in the orbifold as a polynomial of the total number of primaries and the number of self conjugate ones in the parent theory. The formulae in the $`Z_2`$ orbifold illustrate the general results.
Permutation orbifolds have been investigated in the last couple of years, as they are not only a special class of Rational Conformal Field Theories, but they are also closely related to second quantisation of strings (). The one loop amplitude is a starting point there, its form - that is, its dependence on the characters of the primary fields of the corresponding Conformal Field Theory - is determined from general principles. Finding the explicit dependence amounts to writing down the coefficients of the corresponding linear combination of the characters in the open, the sesquilinear combination in the closed case, respectively. This is possible in a permutation orbifold - and the topic of this paper - in terms of the same coefficients of the ”ascendant” CFT. It is useful for getting information about the structure of orbifolds and for providing explicit formulae which can be subject for resting conjectures about the further structure of the amplitude coefficients.
For any RCFT $`𝒞`$ and any permutation group $`\mathrm{\Omega }<S_n`$ a new CFT $`𝒞\mathrm{\Omega }`$ can be constructed by taking the n-fold tensor product of $`𝒞`$ and identifying states according to the orbits of the standard action of $`\mathrm{\Omega }`$. The new theory is called the permutation orbifold of $`𝒞`$ (, ) and every relevant quantity (e.g. conformal weights, genus one characters of the primaries, the matrix elements of the modular transformations, the partition function etc.) is completely determined in terms of the corresponding quantities of $`𝒞`$ and the twist group $`\mathrm{\Omega }`$. The general case (when the twist group is nonabelian) was discussed in .
In general, the torus amplitude of a CFT is always expressible as a sesquilinear combination of the characters of its primaries:
$$Z\left(\tau \right)=\underset{p,q}{}Z_{pq}\chi _p\left(\tau \right)\overline{\chi }_q\left(\tau \right)$$
(1)
where the matrix $`Z_{pq}`$ is invariant under the modular group and has nonnegative integer elements (see eg. ). On the other hand, based on covering surface considerations in , it was also shown that the partition function of the permutation orbifold $`𝒞\mathrm{\Omega }`$ of the RCFT $`𝒞`$ ($`\mathrm{\Omega }<S_n`$ is the twist group) reads
$$Z^\mathrm{\Omega }\left(\tau \right)=\frac{1}{\left|\mathrm{\Omega }\right|}\underset{xy=yx}{}\underset{\xi 𝒪(x,y)}{}Z\left(\tau _\xi \right).$$
(2)
$`𝒪(x,y)`$ is the set of orbits (on the set: $`\{1,2,\mathrm{},n\}`$) of the subgroup generated by the 2 elements $`x,y`$ of $`\mathrm{\Omega }`$, $`Z(\tau )`$ is the torus partition function of $`𝒞`$, and $`\tau _\xi `$ is the modular parameter of the covering torus corresponding to the orbit $`\xi `$. ($`\tau _\xi =\frac{\mu _\xi \tau +\kappa _\xi }{\lambda _\xi }`$ , where the three parameters corresponding to each orbit are:
$`\lambda _\xi `$ is the common length of the $`x`$ orbits in $`\xi `$
$`\mu _\xi `$ is the number of the $`x`$ orbits in $`\xi `$
$`\kappa _\xi `$ is the smallest nonnegative integer for which $`y^{\mu _\xi }=x^{\kappa _\xi }`$ holds in $`\xi `$.) Comparing (1) and (2) gives us way to express the matrix elements $`Z_{pq}`$ in the orbifold as a function of those of the original CFT. The primaries of the orbifold are characterized by pairs $`(P,\mathrm{\Phi })`$, where $`P`$ represents an orbit of $`\mathrm{\Omega }`$ acting on the n tuples $`p_1\mathrm{}p_n`$ of primaries of $`𝒞`$, and $`\mathrm{\Phi }`$ is an irreducible character of the double (, ) $`𝒟(\mathrm{\Omega }_P)`$ of the corresponding stabilizer (the subgroup of $`\mathrm{\Omega }`$ which leaves P invariant). Using the general formula of for the genus one character of the primary field $`(P,\mathrm{\Phi })`$
$$\chi _{(P,\mathrm{\Phi })}\left(\tau \right)=\frac{1}{\left|\mathrm{\Omega }_P\right|}\underset{x,y\mathrm{\Omega }_P}{}\overline{\mathrm{\Phi }}(x,y)\underset{\xi 𝒪(x,y)}{}\omega _{P\left(\xi \right)}^{\frac{\kappa _\xi }{\lambda _\xi }}\chi _{P\left(\xi \right)}\left(\tau _\xi \right)$$
(3)
we are lead to the following matrix
$$Z_{(P,\mathrm{\Phi })(Q,\mathrm{\Psi })}=\frac{1}{\left|\mathrm{\Omega }_P\right|\left|\mathrm{\Omega }_Q\right|}\underset{\genfrac{}{}{0pt}{}{z\mathrm{\Omega }}{x,y\mathrm{\Omega }_p\mathrm{\Omega }_{zQ}}}{}\mathrm{\Phi }(x,y)\overline{\mathrm{\Psi }}(x^z,y^z)\underset{\xi 𝒪(x,y)}{}Z_{P\left(\xi \right)\left(zQ\right)\left(\xi \right)}$$
(4)
where $`zP`$ denotes the action of the group element $`z`$ on the n-tuple $`p_1\mathrm{}p_n`$ corresponding to $`P`$ (one can always work with a chosen representative of the orbit P, then show that the resulting formula does not depend on the choice), $`P(\xi )`$ is the component of P associated to the orbit $`\xi `$, and
$$\omega _p=exp\left(2\pi i\left(\mathrm{\Delta }_p\frac{c}{24}\right)\right)$$
is the exponentiated conformal weight. The formula comes more or less directly from comparing the coefficients of the products of the original characters
$$\chi _{P\left(\xi 1\right)}\left(\tau _{\xi 1}\right)\mathrm{}\chi _{P\left(\xi k\right)}\left(\tau _{\xi k}\right)$$
of the same type in the two expressions. On one hand, in eqn (1) the characters of the orbifold contain these products, on the other hand, writing out the bilinear expressions for each $`Z(\tau _\xi )`$, eqn (4) can be read off taking into account the required symmetrization, so that there should be no dependence on the actual representative of the orbits $`P`$ and $`Q`$. Note that the phases appearing in the expressions of the characters in the orbifold indeed disappear (they are absent in (2)) as a consequence of $`[Z,T]=0`$ in $`𝒞`$, that is $`Z_{pq}`$ is nonzero only if $`\mathrm{\Delta }_p\mathrm{\Delta }_q`$ is an integer, so the corresponding phases cancel each other). The validity of the formula can be investigated via several checks.
1. It gives back (2) when writing down (1) for the orbifold and performing the summation for the pairs $`(P,\mathrm{\Phi })`$. This is a straightforward consequence of the method by which it was obtained. It is easily verified by using the second orthogonality relations for the irreducible characters of the doubles $`𝒟\left(\mathrm{\Omega }_P\right),𝒟\left(\mathrm{\Omega }_Q\right)`$ and interchanging the sum for $`P`$ (and $`Q`$) with the one for the pairs $`(x,y)`$.
2. The requirement of modular invariance. Since we know the explicit expression for the S and T matrices (), their commutator with the given $`Z^\mathrm{\Omega }`$ can be calculated explicitly. Using the modular invariance of Z in $`𝒞`$, the commutators give 0.
3. The important special case when $`Z_{pq}`$ is a permutation matrix. This is always the case whenever the primaries correspond to the full chiral algebra (). In this case the formula becomes simpler:
$$Z_{(P,\mathrm{\Phi })(Q,\mathrm{\Psi })}=\{\begin{array}{cc}\frac{1}{\left|\mathrm{\Omega }_P\right|}\underset{zN\left(\mathrm{\Omega }_P\right)}{}\delta _{\mathrm{\Phi },\mathrm{\Psi }^z}\underset{i𝒪\left(\mathrm{\Omega }_P\right)}{}Z_{P_i\left(zQ\right)_i}\hfill & \text{if }w\mathrm{\Omega }:\mathrm{\Omega }_P=\mathrm{\Omega }_{wQ}\hfill \\ 0\hfill & \text{otherwise}\hfill \end{array}$$
(5)
where $`\mathrm{\Psi }^z`$ is the irreducible character of $`𝒟\left(\mathrm{\Omega }_P\right)`$ defined by the relation: $`\mathrm{\Psi }^z(x,y)=\mathrm{\Psi }(x^z,y^z)`$, $`x^z`$ is an abbreviation for $`z^1xz`$, $`N\left(\mathrm{\Omega }_P\right)`$ is the normalizer of $`\mathrm{\Omega }_P`$ in $`\mathrm{\Omega }`$ (The subgroup $`\{x\mathrm{\Omega }:x\mathrm{\Omega }_P=\mathrm{\Omega }_Px\}`$) and by $`𝒪\left(G\right)`$ we mean the set of orbits of the permutation group G. (Note that we chose $`w^1Q`$ in the formula for identifying $`𝒟\left(\mathrm{\Omega }_P\right)`$ with $`𝒟\left(\mathrm{\Omega }_Q\right)`$, which is possible since (4) does not depend on the representatives of the orbits $`P`$ and $`Q`$). The simplification originates from the fact that any product of the form $`Z_{pq_1}Z_{pq_2}`$ is zero whenever $`q_1q_2`$; it a fundamental property of permutation matrices. The important property of this matrix is straightforward from the formula: the matrix $`Z_{(P,\mathrm{\Phi })(Q,\mathrm{\Psi })}`$ is again a permutation matrix. The map: $`Z_{pq}Z_{(P,\mathrm{\Phi })(Q,\mathrm{\Psi })}`$ is a homomorphism, that is $`\left(Z^{\left(1\right)}Z^{\left(2\right)}\right)^\mathrm{\Omega }=Z^{\left(1\right)\mathrm{\Omega }}Z^{\left(2\right)\mathrm{\Omega }}`$ holds for the general case when Z is not necessarily a permutation matrix. It would be desirable to know the trace of the matrix for several reasons. (Eg. for a permutation it gives the number of its fixed points.) After some calculation we get
$$TrZ^\mathrm{\Omega }=\frac{1}{\left|\mathrm{\Omega }\right|}\underset{x,y,z\mathrm{\Omega }^{\left(3\right)}}{}\underset{i}{}\left(TrZ^i\right)^{i_c}$$
(6)
where $`\mathrm{\Omega }^3`$ is the set of pairwise commuting triples form $`\mathrm{\Omega }`$ and $`i_c`$ is the number of cycles of length $`i`$ of the element $`z`$ on the set of orbits $`𝒪(x,y)`$.
Although we obtained a modular invariant which satisfies several consistency checks, it is not unique in general, due to the linear dependence of the Virasoro specialized characters. Indeed, whenever we have a modular invariant $`Z_{pq}`$ corresponding to a theory, $`Z_{\overline{p}q}=(S^2Z)_{pq}`$, is also a good one producing the same torus amplitude, since the Virasoro specialized character is the same for charge conjugate fields.
What makes (4) special is that it is unital, that is $`Z_{pq}=\delta _{pq}`$ implies $`Z_{(P,\mathrm{\Phi })(Q,\mathrm{\Psi })}=\delta _{(P,\mathrm{\Phi })(Q,\mathrm{\Psi })}`$. This property is easily seen by substituting $`Z_{pq}=\delta _{pq}`$ into (6). Using the notation $`s`$ for the number of primaries in $`𝒞`$ it gives
$$\frac{1}{\left|\mathrm{\Omega }\right|}\underset{x,y,z\mathrm{\Omega }^{\left(3\right)}}{}s^{\left|𝒪(x,y,z)\right|}$$
(7)
which is the number of primaries in $`𝒞\mathrm{\Omega }`$ (see eg. ).
One would like to know how $`Z_{\overline{p}q}`$ looks like in the orbifold. Having the expression for the $`S`$ matrix, this is not a difficult task. Introducing the notation $`x^z=z^1x^1z`$, we get for the charge conjugation:
$$S_{(P,\mathrm{\Phi })(Q,\mathrm{\Psi })}^2=\frac{1}{\left|\mathrm{\Omega }_P\right|\left|\mathrm{\Omega }_Q\right|}\underset{\genfrac{}{}{0pt}{}{z\mathrm{\Omega }}{x,y\mathrm{\Omega }_p\mathrm{\Omega }_{zQ}}}{}\mathrm{\Phi }(x,y)\mathrm{\Psi }(x^z,y^z)\underset{xi𝒪(x,y)}{}S_{P\left(\xi \right)\left(zQ\right)\left(\xi \right)}^2$$
(8)
either from the $`S`$ matrix or from the dimension of the space of genus 0 holomorphic blocks for the insertion of $`(P,\mathrm{\Phi })`$ and $`(Q,\mathrm{\Psi })`$ (). Multiplying it with (4) we find
$$Z_{\overline{(P,\mathrm{\Phi })}(Q,\mathrm{\Psi })}=\frac{1}{\left|\mathrm{\Omega }_P\right|\left|\mathrm{\Omega }_Q\right|}\underset{\genfrac{}{}{0pt}{}{z\mathrm{\Omega }}{x,y\mathrm{\Omega }_p\mathrm{\Omega }_{zQ}}}{}\mathrm{\Phi }(x,y)\overline{\mathrm{\Psi }}(x^z,y^z)\underset{\xi 𝒪(x,y)}{}Z_{\overline{P\left(\xi \right)}\left(zQ\right)\left(\xi \right)}$$
(9)
Its trace reads
$$Tr\left(S^2Z\right)^\mathrm{\Omega }=\frac{1}{\left|\mathrm{\Omega }\right|}\underset{x,y,z\mathrm{\Omega }}{}\delta _{x^y,x}\delta _{x^z,x^1}\delta _{y^z,y^1}\underset{i}{}\left(Tr\left(S^2Z\right)^i\right)^{i_c}$$
(10)
where the notations are as in (6), and finally, when $`Z_{pq}`$ is a permutation
$$Z_{\overline{(P,\mathrm{\Phi })}(Q,\mathrm{\Psi })}=\{\begin{array}{cc}\frac{1}{\left|\mathrm{\Omega }_P\right|}\underset{zN\left(\mathrm{\Omega }_P\right)}{}\delta _{\mathrm{\Phi },\mathrm{\Psi }^z}\underset{i𝒪\left(\mathrm{\Omega }_P\right)}{}Z_{\overline{P}_i\left(zQ\right)_i}\hfill & \text{if }w\mathrm{\Omega }:\mathrm{\Omega }_P=\mathrm{\Omega }_{wQ}\hfill \\ 0\hfill & \text{otherwise}\hfill \end{array}$$
(11)
($`\mathrm{\Psi }^z(x,y)=\mathrm{\Psi }((x^z,y^z)`$ is the defining relation for the irreducible character $`\mathrm{\Psi }^z`$.)
In the case of the Klein bottle amplitude, the same procedure as comparing eqn. (1) and (2) can be done using the result of for the amplitude in the orbifold:
$$K^\mathrm{\Omega }\left(t\right)=\frac{1}{\left|\mathrm{\Omega }\right|}\underset{x,y\mathrm{\Omega }}{}\delta _{x^y,x^1}\underset{\xi 𝒪_{}(x,y)}{}K\left(\frac{\lambda _\xi ^2t}{\left|\xi \right|}\right)\underset{\xi 𝒪_+(x,y)}{}Z\left(\frac{\left|\xi \right|}{2\lambda _\xi ^2it}+\frac{\kappa _\xi }{\lambda _\xi }\right)$$
(12)
according to the geometric picture. $`𝒪_{}(x,y)`$ (resp. $`𝒪_+(x,y)`$) is the set of orbits of the subgroup generated by $`x`$ and $`y`$ with odd (resp. even) number of $`x`$ orbits. ($`𝒪_{}(x,y)`$ contains those orbits on which the commutative subgroup generated by $`x`$ and $`y^2`$ acts transitively, $`𝒪_+(x,y)`$ contains those, which fall into two orbits: $`\xi +,\xi `$ under the above subgroup). This is to be compared with the general formula for the amplitude in a CFT ():
$$K\left(t\right)=\underset{p}{}\mathrm{\Gamma }_p\chi _p\left(\frac{1}{it}\right)$$
(13)
The comparison is easier than in the case of the torus, since $`K^\mathrm{\Omega }(t)`$ is a linear function of the characters of the orbifold. One only has to take into consideration the fact that the parameter $`\kappa _\xi `$ is zero for the doubly covering torus corresponding to the orbits in $`𝒪_{}(x,y)`$, and each element of $`𝒪_+(x,y)`$ is associated with 2 orbits in $`𝒪(x,y^2)`$ of equal length. All in all, we get for the coefficients in the orbifold:
$$\mathrm{\Gamma }_{(P,\mathrm{\Phi })}=\frac{1}{\left|\mathrm{\Omega }_P\right|}\underset{x,y^2\mathrm{\Omega }_P}{}\delta _{x^y,x^1}\mathrm{\Phi }(x,y^2)\underset{\xi 𝒪_{}(x,y)}{}\mathrm{\Gamma }_{P\left(\xi \right)}\underset{\xi 𝒪_+(x,y)}{}Z_{P\left(\xi +\right),P\left(\xi \right)}$$
(14)
(Note that the phases from the characters disappear again at $`𝒪_+(x,y)`$, due to the modular invariance of $`Z_{pq}`$, just like in the case of $`Z^\mathrm{\Omega }(\tau )`$.)
Equations (9) and (14) are worth of analysing a bit further. If we insert the charge conjugation modular invariant (that is, we substitute $`Z_{pq}`$ with $`\delta _{pq}`$ in (9)) we get the corresponding quantity in $`𝒞^\mathrm{\Omega }`$ (it is clear when comparing the resulting expression with (8)). But the trace of the charge conjugation matrix is nothing but the number of self-conjugate primaries t (or $`_p\nu _p^2`$, where $`\nu _p`$ is the 3-valued Frobenius-Schur indicator of the primary field p ). So we get
$$t^\mathrm{\Omega }=\frac{1}{\left|\mathrm{\Omega }\right|}\underset{x,y,z\mathrm{\Omega }}{}\delta _{x^y,x}\delta _{x^z,x^1}\delta _{y^z,y^1}t^{\left|𝒪_{}\right|}s^{\left|𝒪_+\right|}$$
(15)
which is a polynomial function of the corresponding quantity in $`𝒞`$, and the total number of primaries s in $`𝒞`$. $`𝒪_{}`$ is the set of orbits generated by $`x,y`$ and $`z`$ with odd number of $`<x,y>`$ orbits in it ($`<x,y>`$ is the subgroup of $`\mathrm{\Omega }`$ generated by $`x`$ and $`y`$), $`𝒪_{}`$ has the same definition changing the word ”odd” to ”even”.
Doing the same insertion in (14) and substituting $`\mathrm{\Gamma }_p`$ with $`\nu _p`$ we face $`\nu _{(P,\mathrm{\Phi })}`$ (see ), that is if the torus amplitude coefficients are the Frobenius-Schur indicators of primaries in a theory, they remain the same quantities in any orbifold of it. This is the Orbifold Covariance Principle presented in . This can be found in that article whithout the explicit formula for $`\mathrm{\Gamma }_{(P,\mathrm{\Phi })}`$, giving another strong argument for the Ansatz of (The case was already considered in , and ).
Let us see the explicit example of the group $`Z_2`$ (). There are 5 types of primary fields in the orbifold: one corresponds to the orbit $`p_1p_2`$ ($`p_1p_2`$), its stabilizer is trivial, so there is no choice for different irreducible characters but the trivial. The other four correspond to the orbit $`pp`$, its stabilizer is $`Z_2`$ itself, the double of which has four irreducible characters. Since one character is nonzero only if its first argument is in a specific conjugacy class of the group, let 1, 2 denote those two that are nonzero only if their first argument is the unit element and 3, 4 the other two. So summarizing the notation the primaries read: $`(p_1p_2)`$, $`(pp,1)`$, $`(pp,2)`$, $`(pp,3)`$, $`(pp,4)`$ Specifying eqn (4) for this case, we obtain
$$\begin{array}{cccc}Z_{\left(p_1p_2\right)\left(q_1q_2\right)}\hfill & =\hfill & Z_{p_1q_1}Z_{p_2q_2}+Z_{p_1q_2}Z_{p_2q_1}\hfill & \\ Z_{\left(p_1p_2\right)(qq,i)}\hfill & =\hfill & Z_{p_1q}Z_{p_2q}\hfill & i=1,2\hfill \\ Z_{(pp,i)(qq,i)}\hfill & =\hfill & \frac{1}{2}\left(Z_{pq}^2+Z_{pq}\right)\hfill & i=1,2\hfill \\ Z_{(pp,i)(qq,i)}\hfill & =\hfill & Z_{pq}\hfill & i=3,4\hfill \\ Z_{(pp,1)(qq,2)}\hfill & =\hfill & \frac{1}{2}\left(Z_{pq}^2Z_{pq}\right)\hfill & \end{array}$$
for the independent nonzero components of the symmetric matrix. For the trace we have
$$Tr\left(Z^\mathrm{\Omega }\right)=\frac{1}{2}Tr\left(Z\right)^2+\frac{1}{2}Tr\left(Z^2\right)+3Tr\left(Z\right)$$
(16)
which specializes to
$$s^\mathrm{\Omega }=\frac{1}{2}s^2+\frac{7}{2}s$$
(17)
when $`Z_{pq}=\delta _{pq}`$ ($`s`$ is the total number of primaries). There is no need to consider eqn (9) in the case of $`Z_2`$, since the inverse operation is the identity. However the trace specializes to the following polynomial when $`Z_{pq}`$ is the charge conjugation
$$t^\mathrm{\Omega }=\frac{1}{2}t^2+\frac{1}{2}s+3t$$
(18)
where t is the number of self conjugate primaries. This means for example that any $`Z_2`$ orbifold of an RCFT with no complex primary fields has only self-conjugate ones too. Should one find a closed formula for $`_p\nu _p`$ in the orbifold as well, one would have the number of real, pseudoreal and complex primaries in $`𝒞\mathrm{\Omega }`$ as polynomials of the same quantities in $`𝒞`$, which is a step in classifying RCFT-s.
Finally let us enumerate the results of the present paper. Two modular invariants are found for the partition function coefficients of a permutation orbifold expressed in terms of the original CFT. One keeps the structure $`Z(\tau )=_p|\chi _p|^2`$, its trace therefore concides with the number of primaries in the orbifold when $`Z_{pq}=\delta _{pq}`$, the other keeps the structure of $`Z(\tau )=_p\chi _p\overline{\chi }_{\overline{p}}`$, its trace therefore is the number of self conjugate primaries in the orbifold, depending on the same quantity and the total number of fields in the original theory. The Klein bottle coefficients are also computed and applied to illustrate how the OCP was used in , supporting the Ansatz of for the amplitude.
## Acknowledgements
I thank P. Bantay for discussions and helpful comments.
|
warning/0004/cond-mat0004269.html
|
ar5iv
|
text
|
# Acknowledgments
## Acknowledgments
The Sher Alam’s work is supported by the Japan Society for the Promotion of Science \[JST\] via the STA fellowship.
##
As is well-known that orthogonal rotation in N-dimension is specified by the $`SO(N)`$ groups. By definition they leave the norm squared of $`N`$ dimensional vector invariant. There is a distinction between when $`N`$ is odd and even, i.e. $`SO(2n)`$ and $`SO(2n+1)`$, in fact this is so in Cartan classification. We recall that in Cartan classification scheme groups are classified into the categories: $`A_n`$, $`B_n`$, $`C_n`$, $`D_n`$, $`G_2`$, $`F_4`$, $`E_6`$, $`E_7`$, and $`E_8`$. For example $`SU(n+1)`$ are in the category $`A_n`$, $`SO(2n)`$ are in $`D_n`$, and $`SO(2n+1)`$ are in $`C_n`$.
The maximal subalgebra of classical simple Lie algebra of $`SO(7)`$ reads:
$`SO(7)`$ $``$ $`SU(4),`$ (13)
$`SO(7)`$ $``$ $`SU(2)\times SU(2)\times SU(2),`$ (14)
$`SO(7)`$ $``$ $`Sp(4)\times U(1),`$ (15)
$`SO(7)`$ $``$ $`G(2).`$ (16)
We note that for $`Sp(4)`$ the maximal subalgebra reads:
$`Sp(4)`$ $``$ $`SU(2)\times SU(2),`$ (17)
$`Sp(4)`$ $``$ $`SU(2)\times U(1),`$ (18)
$`Sp(4)`$ $``$ $`SU(2).`$ (19)
$`Sp(4)`$ is isomorphic to $`SO(5)`$, and $`SO(4)SU(2)\times SU(2)`$. $`SU(2)U(1)`$ and $`Sp(2)`$, $`SO(3)`$, and $`SU(2)`$ are all isomorphic.
The maximal subalgebra of classical simple Lie algebra of $`SO(8)`$ is given by
$`SO(8)`$ $``$ $`SU(4)\times U(1),`$ (20)
$`SO(8)`$ $``$ $`SU(2)\times SU(2)\times SU(2)\times SU(2),`$ (21)
$`SO(8)`$ $``$ $`Sp(4)\times SU(2),`$ (22)
$`SO(8)`$ $``$ $`SU(3),`$ (23)
$`SO(8)`$ $``$ $`SO(7).`$ (24)
In a simple sense one may say that non-linear sigma model is like a Taylor expansion in field theory. Let us explain what this means. We can write the action of string \[1-dimensional\] propagating in a manifold with metric $`G_{\mu \nu }`$ as
$`LG_{\mu \nu }(X)_aX^\mu _bX^\nu g^{ab}+\mathrm{}.`$ (25)
$`g_{ab}`$ is the two-dimensional metric generated by the ‘motion’ of string in the background manifold $`G_{\mu \nu }(X)`$. A crucial observation is that $`X`$’s play a dual role of coordinate of the string in the background space and scalar field in the 2-dimensional space specifield by the metric $`g_{ab}`$ \[i.e. the space generated by the motion of the string\]. Different choices for the background metrics lead to different conformal field theories. Of interest to us is the choice that the string is propagating on a manifold specified by a Lie Group\[for e.g., SU(N), SO(N), etc\] in other words group manifold. We thus let $`g`$ be an element of the Lie group. From Eq. 25 we can guess that a string propagating on this group manifold has an action of form
$`Ltr(_ag^1^ag)`$ (26)
where $`g`$ is some function of the string field $`X`$. Simple differentiation gives
$`_ag=_aX_\mu f^{a\mu }`$ (27)
for some function $`f`$, thus the metric $`G`$ can be expressed in terms of $`f`$. The exact form of action is
$`S`$ $`=`$ $`{\displaystyle \frac{1}{4\lambda ^2}}{\displaystyle tr(_ag^1^ag)}+k\mathrm{\Gamma }(g)`$ (28)
$`\mathrm{\Gamma }(g)`$ $`=`$ $`{\displaystyle \frac{1}{24\pi }}{\displaystyle d^3Xϵ^{\alpha \beta \gamma }tr[(g^1^\alpha g)(g^1_\beta g)(g^1_\gamma g)]}`$ (29)
where $`\mathrm{\Gamma }(g)`$ is the Wess-Zumino term which is integrated over 3-dimensional disk whose boundary is two-dimensional space. For $`k=0`$ it reduces to ordinary sigma-model, which is not conformally invariant \[it is asymptotically free and massive\]. For special values of $`k=1,2,3,..`$ the theory becomes effectively massless and has an infrared-stable fixed point at the values of parameters $`\lambda `$ and $`k`$ related via
$`\lambda ^2=4\pi /k`$ (30)
Thus at these special values of $`k`$ we have a conformally invariant $`\sigma `$ model where the theory is defined on the group manifold. This theory is called Wess-Zumino-Witten \[WZW\] model. The symmetry generators $`J`$ satisfy a special case of Kac-Moody algebra, viz
$`[J_n^a,J_m^b]=f^{abc}J_{n+m}^c+{\displaystyle \frac{1}{2}}kn\delta ^{ab}\delta _{n+m,0}`$ (31)
In Eq. 31 we note the following the generators $`J`$ carry two indices, namely $`a,b,c\mathrm{}`$ which are the Lie group indices and $`n,m..`$ which are arise in the decomposition of the generator $`J=J(z)`$ in terms of its moments, viz,
$`J(z)={\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}J_nz^{n1}.`$ (32)
In some sense the Kac-Moody algebra smears the generators of ordinary Lie algebra around a circle or string.
Finally we recall that
$`qe^{2\pi i/(k+2)}`$ (33)
If we make the above correspondence it can be shown by examining various identities of WZW model that the braiding properties of WZW model at level $`k`$ are determined by the representation theory of quantum groups. As a trivial check if one sets $`q=1`$ in 33, which is the limit in which quantum group reduces to the ordinary classical group, then the right-hand side of 33 we must set $`k\mathrm{}`$, which is precisely the limit in which Kac-Moody algebra reduces to ordinary classical algebra. We recall that the symmetry generators of the WZW model obey a special case of Kac-Moody algebra.
Non-linear sigma models have been extensively used in particle theory to describe interactions phenomenologically between strongly interacting particle. For example the Lagarangian for non-linear sigma-model for the special case of $`SU(2)\times SU(2)`$ spontaneously broken to $`SU(2)`$
$`={\displaystyle \frac{1}{2}}{\displaystyle \frac{_\mu \stackrel{}{\pi }^\mu \stackrel{}{\pi }}{(1+\stackrel{}{\pi }^2/F^2)^2}}`$ (34)
where the factor $`1/F`$ acts as the coupling term that comes with the interaction of each additional pion. Expanding the expression in Eq. 34 and keeping only the first two terms we get
$`={\displaystyle \frac{1}{2}}_\mu \stackrel{}{\pi }^\mu \stackrel{}{\pi }+(1/F^2)\stackrel{}{\pi }^2_\mu \stackrel{}{\pi }^\mu \stackrel{}{\pi }+\mathrm{}.`$ (35)
The first term is the simple kinetic energy term, the second being the potential energy term of the form $`\pi _i\pi ^i_\mu \stackrel{}{\pi }_j^\mu \pi ^j`$ \[in words it is a velocity dependent potential term\]. We must also retain in general in our phenomenological modelling \[as in this paper\] of HTSC material keep such terms.
|
warning/0004/math0004055.html
|
ar5iv
|
text
|
# Généralisation de formules de type Waring
## 1 Introduction
L’un des problèmes fondamentaux dans l’étude de fonctions symétriques est le développement d’une fonction symétrique sur certaines bases linéaires de l’algèbre des fonctions symétriques. Un résultat classique de Waring explicite le développement des fonctions symétriques puissances $`p_n`$ dans la base linéaire des fonctions symétriques élémentaires $`(e_\lambda )`$.
Dans cet article nous généralisons la formule de Waring en développant les fonctions symétriques puissances $`p_n`$ évaluées sur l’alphabet $`Y=\{x_1/(1tx_1)`$, $`x_2/(1tx_2),\mathrm{}\}`$ dans la base linéaire des fonctions symétriques élémentaires $`(e_\lambda )`$ évaluées sur l’alphabet $`X=\{x_1,x_2,\mathrm{}\}`$. De même nous considérons le problème inverse, c’est-à-dire, le développement des fonctions symétriques $`h_n`$ et $`e_n`$ évaluées sur l’alphabet $`Y`$ dans la base des fonctions puissances $`p_\mu `$ évaluées sur l’alphabet $`X`$. Dans le dernier cas nous aurons besoin d’un coefficient binomial généralisé introduit par Lassalle . Nous en déduisons ensuite, comme applications, des développements intéressants, qui conduisent en particulier de nouvelles preuves des ex-conjectures de Lassalle . D’autres preuves de ces conjectures ont été tout récemment données par Lascoux et Lassalle dans le cadre des $`\lambda `$-anneaux. Notre approche repose essentiellement sur l’opérateur différentiel de l’algèbre des séries formelles. Il est remarquable que l’étude d’un problème si élémentaire puisse conduire à une preuve très simple de l’identité de Lascoux et Lassalle.
Nous terminons cette introduction par un rappel \[4, Chap.1\] des formules qui seront utilisées dans la suite. Observons d’abord que
$$\underset{n1}{}\left(\genfrac{}{}{0pt}{}{n1}{k1}\right)a_nt^{n1}=\frac{t^{k1}}{(k1)!}\frac{d^{k1}}{dt}\left(\underset{n1}{}a_nt^{n1}\right).$$
(1)
Comme les fonctions puissances $`p_n(X)=_{r1}x_r^n`$ satisfont $`_{n1}p_n(X)t^{n1}=_{r1}x_r/(1x_rt)`$, et pour tout $`k1`$
$$\frac{d^{k1}}{dt}\left(\frac{1}{1xt}\right)=(k1)!\frac{x^{k1}}{(1xt)^k},$$
nous en déduisons donc
$$\frac{d^{k1}}{dt}\left(\underset{n1}{}p_n(X)t^{n1}\right)=\frac{(k1)!}{t^k}p_k(\frac{tx_1}{1tx_1},\frac{tx_2}{1tx_2},\mathrm{}).$$
(2)
Pour toute partition d’entiers $`\mu `$ on pose $`z_\mu =_{i1}i^{m_i(\mu )}m_i(\mu )!,`$$`m_i(\mu )`$ est le nombre de parts dans $`\mu `$ égales à $`i1`$, et pour tout entier $`n`$ positif on définit le coefficient binomial généralisé $`\genfrac{}{}{0pt}{}{\mu }{n}`$ comme étant le nombre de façons de choisir $`n`$ éléments dans le diagramme de Ferrers de $`\lambda `$, dont au moins un par ligne.
Les fonctions symétriques $`h_n(X)`$ et $`e_n(X)`$ sont liées aux fonctions puissances $`p_\mu (X)=_{r1}p_{\mu _r}(X)`$ par la formule :
$`h_n(X)`$ $`=`$ $`{\displaystyle \underset{\mu n}{}}z_\mu ^1p_\mu (X),`$ (3)
$`e_n(X)`$ $`=`$ $`{\displaystyle \underset{\mu n}{}}(1)^{nl(\mu )}z_\mu ^1p_\mu (X).`$ (4)
L’inverse de la dernière est appelée *formule de Waring* :
$$p_n(X)=\underset{\lambda n}{}(1)^{nl(\lambda )}\frac{n(l(\lambda )1)!}{_im_i(\lambda )!}e_\lambda (X).$$
(5)
Par l’involution $`\omega `$ définie par $`\omega (e_n)=h_n`$ on a aussi \[4, p. 24\]
$$p_n(X)=\underset{\lambda n}{}(1)^{l(\lambda )1}\frac{n(l(\lambda )1)!}{_im_i(\lambda )!}h_\lambda (X).$$
(6)
On note $`m_\mu (X)`$ la fonction symétrique monomiale associée à la partition $`\mu `$.
Nous remercions Michel Lassalle pour ses remarques amicales sur une version antérieure de cet article.
## 2 Résultats principaux
Soit $`X=\{x_1,x_2,\mathrm{}\}`$ un ensemble fini ou infini d’indéterminées et $`\frac{X}{1tX}`$ l’alphabet $`\{\frac{x_1}{1tx_1},\frac{x_2}{1tx_2},\mathrm{}\}`$.
###### Théorème 1
Pour tout $`k1`$ on a
$`p_k\left({\displaystyle \frac{X}{1tX}}\right)`$ $`=`$ $`{\displaystyle \underset{|\mu |k}{}}t^{|\mu |k}\left({\displaystyle \genfrac{}{}{0pt}{}{|\mu |}{k}}\right)(1)^{|\mu |l(\mu )}{\displaystyle \frac{k(l(\mu )1)!}{_im_i(\mu )!}}e_\mu (X),`$ (7)
$`p_k\left({\displaystyle \frac{X}{1tX}}\right)`$ $`=`$ $`{\displaystyle \underset{|\mu |k}{}}t^{|\mu |k}\left({\displaystyle \genfrac{}{}{0pt}{}{|\mu |}{k}}\right)(1)^{l(\mu )1}{\displaystyle \frac{k(l(\mu )1)!}{_im_i(\mu )!}}h_\mu (X).`$ (8)
Démonstration. Les formules (1) et (2) impliquent directement
$$p_k\left(\frac{X}{1tX}\right)=\underset{jk}{}t^{jk}\left(\genfrac{}{}{0pt}{}{j1}{k1}\right)p_j(X).$$
On en déduit donc (7) et (8) respectivement de (5) et (6).
Par la même méthode nous obtenons le résultat suivant.
###### Théorème 2
Pour tout entier $`k1`$ on a
$`h_k\left({\displaystyle \frac{X}{1tX}}\right)`$ $`=`$ $`{\displaystyle \underset{|\mu |k}{}}t^{|\mu |k}{\displaystyle \frac{\genfrac{}{}{0pt}{}{\mu }{k}}{z_\mu }}p_\mu (X),`$ (9)
$`e_k\left({\displaystyle \frac{X}{1tX}}\right)`$ $`=`$ $`{\displaystyle \underset{|\mu |k}{}}t^{|\mu |k}(1)^{kl(\mu )}{\displaystyle \frac{\genfrac{}{}{0pt}{}{\mu }{k}}{z_\mu }}p_\mu (X).`$ (10)
Démonstration. Notons d’abord que
$$\genfrac{}{}{0pt}{}{\mu }{k}=\underset{\genfrac{}{}{0pt}{}{k_1+\mathrm{}+k_l=k}{k_1,\mathrm{},k_l1}}{}\underset{i=1}{\overset{l}{}}\left(\genfrac{}{}{0pt}{}{\mu _i}{k_i}\right),$$
$`l=l(\mu )`$. Comme chaque partition $`\mu `$ de $`j`$ correspond à $`l(\mu )!/_{i1}m_i(\mu )!`$ compositions $`(k_1,\mathrm{},k_l)`$ de $`j`$ telles que $`(k_1,\mathrm{},k_l)`$ soit une permutation des parts de $`\mu `$, nous avons, en tenant compte de (1) et (2),
$`{\displaystyle \underset{jk}{}}t^{jk}{\displaystyle \underset{\mu j}{}}{\displaystyle \frac{\alpha ^{kl(\mu )}}{z_\mu }}{\displaystyle \genfrac{}{}{0pt}{}{\mu }{k}}p_\mu (X)`$
$`={\displaystyle \underset{\genfrac{}{}{0pt}{}{k_1+\mathrm{}+k_l=k}{l1}}{}}{\displaystyle \frac{\alpha ^{kl}t^k}{l!k_1\mathrm{}k_l}}{\displaystyle \underset{r=1}{\overset{l}{}}}{\displaystyle \underset{\mu _r1}{}}\left({\displaystyle \genfrac{}{}{0pt}{}{\mu _r1}{k_r1}}\right)p_{\mu _r}(X)t^{\mu _r}`$
$`={\displaystyle \underset{\genfrac{}{}{0pt}{}{k_1+\mathrm{}+k_l=k}{l1}}{}}{\displaystyle \frac{\alpha ^{kl}}{l!k_1!\mathrm{}k_l!}}{\displaystyle \underset{r=1}{\overset{l}{}}}{\displaystyle \frac{d^{k_r1}}{dt}}\left({\displaystyle \underset{n1}{}}p_n(X)t^{n1}\right)`$
$`={\displaystyle \underset{\mu k}{}}{\displaystyle \frac{\alpha ^{kl(\mu )}}{z_\mu }}p_\mu ({\displaystyle \frac{x_1}{1tx_1}},{\displaystyle \frac{x_2}{1tx_2}},\mathrm{}).`$
En posant $`\alpha =1`$ (resp. $`1`$), nous en déduisons (9) (resp. (10)) en appliquant (5) (resp. (6)).
Remarque. 1) Lorsque $`t=0`$ on retrouve les formules classiques de type Waring.
2) Dans les théorèmes 1 et 2, $`t`$ n’est qu’un paramètre d’homogénéité, mais vu le rôle important qu’il joue dans notre démonstration, nous préférons garder cette forme.
Rappelons que $`h_n(X)`$ et $`e_n(X)`$ ont pour fonctions génératrices:
$`{\displaystyle \underset{n0}{}}h_n(X)t^n`$ $`=`$ $`{\displaystyle \underset{r1}{}}{\displaystyle \frac{1}{1x_rt}},`$ (11)
$`{\displaystyle \underset{n0}{}}e_n(X)t^n`$ $`=`$ $`{\displaystyle \underset{i1}{}}\left(1+x_it\right).`$ (12)
###### Théorème 3
Soit $`z`$ et $`X=\{x_1,x_2,\mathrm{}\}`$ des indéterminées indépendantes. Alors la série formelle
$$F(t,u)=(1+u)^z\underset{r1}{}\left(1+\frac{u}{1+u}\frac{tx_r}{1tx_r}\right)$$
admet les trois développements suivants
$`F(t,u)`$ $`=`$ $`{\displaystyle \underset{i,j0}{}}u^it^j{\displaystyle \underset{l(\mu )i,|\mu |=j}{}}\left({\displaystyle \genfrac{}{}{0pt}{}{zl(\mu )}{il(\mu )}}\right)m_\mu (X),`$ (13)
$`F(t,u)`$ $`=`$ $`{\displaystyle \underset{i,j0}{}}u^it^j{\displaystyle \underset{k=0}{\overset{\mathrm{min}(i,j)}{}}}\left({\displaystyle \genfrac{}{}{0pt}{}{zj}{ik}}\right){\displaystyle \underset{\mu j}{}}{\displaystyle \frac{\genfrac{}{}{0pt}{}{\mu }{k}}{z_\mu }}p_\mu (X),`$ (14)
$`F(t,u)`$ $`=`$ $`{\displaystyle \underset{i,j0}{}}u^it^j{\displaystyle \underset{k0}{}}\left({\displaystyle \genfrac{}{}{0pt}{}{zk}{ik}}\right){\displaystyle \underset{\mu j}{}}(1)^{kl(\mu )}{\displaystyle \frac{\genfrac{}{}{0pt}{}{\mu }{k}}{z_\mu }}p_\mu (X).`$ (15)
Démonstration. Tout d’abord, par définition nous avons
$`F(t,u)`$ $`=`$ $`{\displaystyle \underset{k0}{}}u^k(1+u)^{zk}{\displaystyle \underset{\genfrac{}{}{0pt}{}{1r_1<r_2<\mathrm{}<r_k}{m_1,\mathrm{},m_k1}}{}}(tx_{r_1})^{m_1}\mathrm{}(tx_{r_k})^{m_k}`$
$`=`$ $`{\displaystyle \underset{i,j0}{}}u^it^j{\displaystyle \underset{k0}{}}\left({\displaystyle \genfrac{}{}{0pt}{}{zk}{ik}}\right){\displaystyle \underset{l(\mu )=k,|\mu |=j}{}}m_\mu (X).`$
D’où (13). Ensuite, dans le membre de droite de (14) en remplaçant $`i`$ par $`i+k`$, nous obtenons en appliquant la formule du binôme
$$\underset{k0}{}u^k\underset{j0}{}(1+u)^{zj}t^j\underset{\mu j}{}\frac{\genfrac{}{}{0pt}{}{\mu }{k}}{z_\mu }p_\mu (X),$$
qui s’écrit, en posant $`s=t/(1+u)`$ et en appliquant (9) et (11),
$$(1+u)^z\underset{k0}{}u^kh_k(\frac{sx_1}{1sx_1},\frac{sx_2}{1sx_2},\mathrm{})=(1+u)^z\underset{r1}{}\left(1\frac{usx_r}{1sx_r}\right)^1.$$
Ceci est clairement égal à $`F(t,u)`$. Enfin nous déduisons (15) de façon analogue en appliquant (10) et (12).
###### Corollaire 4
Soit $`z`$ et $`X=\{x_1,x_2\mathrm{}\}`$ des indéterminées indépendantes. Pour tous entiers $`i,j1`$ on a
$`{\displaystyle \underset{l(\mu )i,|\mu |=j}{}}\left({\displaystyle \genfrac{}{}{0pt}{}{zl(\mu )}{il(\mu )}}\right)m_\mu (X)`$ $`=`$ $`{\displaystyle \underset{k=0}{\overset{\mathrm{min}(i,j)}{}}}\left({\displaystyle \genfrac{}{}{0pt}{}{zj}{ik}}\right){\displaystyle \underset{\mu j}{}}{\displaystyle \frac{\genfrac{}{}{0pt}{}{\mu }{k}}{z_\mu }}p_\mu (X)`$
$`=`$ $`{\displaystyle \underset{k=0}{\overset{\mathrm{min}(i,j)}{}}}\left({\displaystyle \genfrac{}{}{0pt}{}{zk}{ik}}\right){\displaystyle \underset{\mu j}{}}(1)^{kl(\mu )}{\displaystyle \frac{\genfrac{}{}{0pt}{}{\mu }{k}}{z_\mu }}p_\mu (X).`$
Comme $`(p_\mu )_\mu `$ forme une base linéaire de l’algèbre des fonctions symétriques, on déduit du corollaire 3 le résultat suivant.
###### Corollaire 5
Soit $`z`$ une variable. Pour des entiers $`i,j1`$ et toute partition $`\mu j`$ on a
$$\underset{k=0}{\overset{min(i,j)}{}}\left(\genfrac{}{}{0pt}{}{zj}{ik}\right)\genfrac{}{}{0pt}{}{\mu }{k}=\underset{k=0}{\overset{min(i,j)}{}}(1)^{kl(\mu )}\left(\genfrac{}{}{0pt}{}{zk}{ik}\right)\genfrac{}{}{0pt}{}{\mu }{k}.$$
Enfin le corollaire 3 implique aussi le résultat suivant, dû à Lascoux-Lassalle \[1, Lemme 2\].
###### Corollaire 6
Pour tous entiers $`k,j1`$ on a
$$\underset{l(\mu )=k,|\mu |=j}{}m_\mu (X)=\underset{\mu j}{}(1)^{kl(\mu )}\frac{\genfrac{}{}{0pt}{}{\mu }{k}}{z_\mu }p_\mu (X).$$
Remarque. On trouvera d’autres formules sur la somme $`_{l(\mu )=k,|\mu |=j}m_\mu (X)`$ dans Macdonald \[4, p. 33 et 68\].
## 3 Applications
On identifie chaque partition $`\lambda `$ avec son *diagramme de Ferrers* et on pose
$$(x)_\lambda =\underset{(i,j)\lambda }{}\left(x+j1(i1)/\alpha \right).$$
Lorsque $`\lambda =(n)`$ est une partition-ligne on retrouve la définition habituelle de factorielle montante $`(x)_n=x(x+1)\mathrm{}(x+n1)`$. Par un calcul direct et en posant $`Z=\left\{j1(i1)/\alpha \right\}`$ ($`(i,j)\lambda `$), nous obtenons :
$`{\displaystyle \frac{(yx)_\lambda }{(y)_\lambda }}`$ $`=`$ $`{\displaystyle \underset{zZ}{}}\left(1{\displaystyle \frac{x/y}{1+z/y}}\right)`$
$`=`$ $`{\displaystyle \underset{i=0}{\overset{|\lambda |}{}}}(x/y)^i{\displaystyle \underset{z_1,\mathrm{},z_iZ}{}}{\displaystyle \frac{1}{1+z_1/y}}\mathrm{}{\displaystyle \frac{1}{1+z_i/y}}`$
$`=`$ $`{\displaystyle \underset{i=0}{\overset{|\lambda |}{}}}{\displaystyle \underset{j=0}{\overset{\mathrm{}}{}}}(1)^{i+j}{\displaystyle \frac{x^i}{y^{i+j}}}{\displaystyle \underset{l(\mu )i,|\mu |=j}{}}\left({\displaystyle \genfrac{}{}{0pt}{}{|\lambda |l(\mu )}{il(\mu )}}\right)m_\mu (Z).`$
Nous déduisons donc du corollaire 3 une courte preuve d’un résultat de Lascoux-Lassalle \[1, Thm. 4\], qui fut conjecturé par Lassalle \[2, Conj. 2\].
###### Théorème 7
Soient $`x`$, $`y`$ deux indéterminées indépendantes. Pour toute partition $`\lambda `$ soit $`X=\left\{j1(i1)/\alpha \right\}`$, $`(i,j)\lambda `$, alors
$$\frac{(yx)_\lambda }{(y)_\lambda }=\underset{i=0}{\overset{|\lambda |}{}}\underset{j=0}{\overset{+\mathrm{}}{}}(1)^{i+j}\frac{x^i}{y^{i+j}}\underset{k=0}{\overset{\mathrm{min}(i,j)}{}}\left(\genfrac{}{}{0pt}{}{|\lambda |j}{ik}\right)\underset{\mu j}{}\frac{\genfrac{}{}{0pt}{}{\mu }{k}}{z_\mu }p_\mu (X).$$
En fait Lascoux et Lassalle ont déduit le théorème 5 d’un résultat plus général, qui fut aussi conjecturé par Lassalle . Nous en donnons aussi une nouvelle preuve.
###### Théorème 8
Soient $`z`$,$`u`$ et $`X=\{x_1,x_2\mathrm{}\}`$ des indéterminées indépendantes. Pour tous entiers $`n,r1`$ on a
$`{\displaystyle \underset{\mu n}{}}{\displaystyle \frac{(1)^{rl(\mu )}}{z_\mu }}{\displaystyle \genfrac{}{}{0pt}{}{\mu }{r}}{\displaystyle \underset{i1}{}}\left(z+{\displaystyle \underset{k1}{}}u^k{\displaystyle \frac{(i)_k}{k!}}x_k\right)^{m_i(\mu )}=`$
$`{\displaystyle \underset{j0}{}}u^j\left({\displaystyle \genfrac{}{}{0pt}{}{n+j1}{nr}}\right){\displaystyle \underset{k=0}{\overset{\mathrm{min}(r,j)}{}}}\left({\displaystyle \genfrac{}{}{0pt}{}{zj}{rk}}\right){\displaystyle \underset{\mu j}{}}{\displaystyle \frac{\genfrac{}{}{0pt}{}{\mu }{k}}{z_\mu }}{\displaystyle \underset{i1}{}}x_i^{m_i(\mu )}.`$
Démonstration. Soit $`Y=\{y_1,y_2,\mathrm{}\}`$ une famille infinie d’indéterminées. Comme les fonctions puissances $`p_i(Y)`$ sont algébriquement indépendantes dans ce cas, nous pouvons supposer $`x_i=p_i(Y)`$ pour $`i1`$. En multipliant le membre de gauche par $`t^nq^r`$ et sommant sur $`n,r1`$ nous pouvons écrire sa fonction génératrice comme suit (voir l’Appendice ci-après):
$$F(tu,Tq)=(1+Tq)^z\underset{j1}{}\left(1+\frac{Tq}{1+Tq}\frac{Tuz_j}{1Tuz_j}\right),$$
(16)
$`T=t/(1t)`$ et $`z_j=y_j/t`$ pour $`j1`$. Nous en déduisons par l’application du théorème 3 que
$`F(tu,Tq)`$ $`=`$ $`{\displaystyle \underset{r,j,k0}{}}T^{r+j}q^ru^j\left({\displaystyle \genfrac{}{}{0pt}{}{zj}{rk}}\right){\displaystyle \underset{\mu j}{}}{\displaystyle \frac{\genfrac{}{}{0pt}{}{\mu }{k}}{z_\mu }}p_\mu (Z)`$
$`=`$ $`{\displaystyle \underset{r,j,k0}{}}{\displaystyle \frac{t^rq^ru^j}{(1t)^{r+j}}}\left({\displaystyle \genfrac{}{}{0pt}{}{zj}{rk}}\right){\displaystyle \underset{\mu j}{}}{\displaystyle \frac{\genfrac{}{}{0pt}{}{\mu }{k}}{z_\mu }}p_\mu (Y).`$
En écrivant
$$\frac{t^r}{(1t)^{r+j}}=\underset{nr}{}\left(\genfrac{}{}{0pt}{}{n+j1}{nr}\right)t^n,$$
nous remarquons que l’expression plus haut est aussi la fonction génératrice du membre de droite.
## Appendice. Calcul de la fonction génératrice
Afin de rendre la lecture autonome nous incluons ici une preuve classique de (16). Remarquons d’abord que pour toute partition $`\mu `$
$$\underset{r1}{}\genfrac{}{}{0pt}{}{\mu }{r}q^r=\underset{i1}{}\left((1+q)^i1\right)^{m_i(\mu )},$$
et que la formule du binôme $`(1x)^\alpha =_{n0}x^n(\alpha )_n/n!`$ permet d’écrire
$$\underset{n1}{}u^n\frac{(i)_n}{n!}p_n(Y)=\underset{j1}{}\underset{n1}{}u^n\frac{(i)_n}{n!}y_j^n=\underset{j1}{}((1y_ju)^i1).$$
En multipliant le membre de gauche par $`t^nq^r`$ et sommant sur $`n,r1`$ nous obtenons sa fonction génératrice
$`{\displaystyle \underset{\mu }{}}{\displaystyle \frac{t^{|\mu |}}{z_\mu }}{\displaystyle \underset{i1}{}}\left[(1(1q)^i)\left(z+{\displaystyle \underset{j1}{}}u^j{\displaystyle \frac{(i)_j}{j!}}p_j(Y)\right)\right]^{m_i(\mu )}`$
$`=`$ $`{\displaystyle \underset{i1}{}}{\displaystyle \underset{m_i=0}{}}{\displaystyle \frac{t^{im_i}}{m_i!i^{m_i}}}\left[(1(1q)^i)(z+{\displaystyle \underset{j1}{}}((1y_ju)^i1))\right]^{m_i}.`$
Mais le dernier terme peut s’écrire
$`{\displaystyle \underset{i1}{}}\mathrm{exp}\left\{\left({\displaystyle \frac{t^i}{i}}{\displaystyle \frac{t^i}{i}}(1q)^i\right)\left(z+{\displaystyle \underset{j1}{}}((1y_ju)^i1)\right)\right\}`$
$`=`$ $`\left(1+{\displaystyle \frac{tq}{1t}}\right)^z{\displaystyle \underset{j1}{}}\mathrm{exp}{\displaystyle \underset{i1}{}}\left({\displaystyle \frac{t^i}{i}}{\displaystyle \frac{t^i}{i}}(1q)^i\right)\left({\displaystyle \frac{1}{(1y_ju)^i}}1\right)`$
$`=`$ $`\left(1+{\displaystyle \frac{tq}{1t}}\right)^z{\displaystyle \underset{j1}{}}\left(1+{\displaystyle \frac{tq}{1ty_ju}}\right)\left(1+{\displaystyle \frac{tq}{1t}}\right)^1.`$
On pourrait trouver des calculs similaires aux précédents dans ou dans en termes de $`\lambda `$-anneaux.
|
warning/0004/quant-ph0004108.html
|
ar5iv
|
text
|
# Exclusion Statistics in Classical Mechanics
## Abstract
We present a general method to derive the classical mechanics of a system of identical particles in a way that retains information about quantum statistics. The resulting statistical mechanics can be interpreted as a classical version of Haldane’s exclusion statistics.
Particle statistics enters quantum physics in two related but logically distinct way. The first one is related to the symmetry of the wave function, or more generally to phase factors associated with the exchange of identical particles. The other is related to entropy, i.e. to the counting of quantum states, and is expressed through the Pauli exclusion principle and the phenomenon of Bose condensation. Some years ago, Haldane pointed out that it is possible to have a certain kind of quantum statistics of the second type, so called exclusion statistics, without any reference to wave functions or exchange factors . While the exchange phase factors characterizing fermions, bosons or (in 2 space dimensions) anyons are intimately connected to quantum mechanics, there is no logical reason for not having effects of the second type of statistics even in classical systems. In fact, as is well known such effects must be put in by hand in order to avoid the Gibbs paradox in classical statistical mechanics.
In this letter we show that it is possible to formulate a classical mechanics that builds in the effects of quantum statistics at the Lagrangian level. The dynamics, i.e. the equations of motion, will not depend on the statistics parameter, but the counting of states, and thus the statistical mechanics, will. In the classical description the statistics is expressed through the occupation of phase space volume. Thus, each new particle introduced in the system will reduce the available phase space volume for the other particles, and the degree of reduction defines the classical statistics parameter.
We first sketch a general formulation of the problem, and then present some specific results for two cases, non-interacting charged particles in a strong magnetic field and vortices in the Landau-Ginzburg-Chern-Simons theory. Both these systems are of interest for the quantum Hall effect. In both examples, we show that the resulting classical statistical mechanics is a classical version of Haldane’s exclusion statistics. Our examples are two-dimensional, but just as in the case of exclusion statistics, there is no reason in principle for our construction not to work in an arbitrary dimension. Below we will only present the general ideas and some of the main results. A fuller account including calculational details can be found in ref. .
Consider a general quantum system and a subset of states $`|\psi _𝐳`$, which is labelled by a set of complex coordinates $`𝐳=\{z_1,z_2,\mathrm{},z_N\}`$. These may be the coordinates of a system of (identical) particles or the coordinates of an $`N`$ soliton configuration. We only assume that the wave function evolves smoothly with a change of these coordinates, and that it is symmetric under an interchange of any pair of the $`N`$ coordinates. To define the corresponding classical mechanics, consider the constrained system where the evolution of the full quantum system is projected to the manifold, $``$, defined by the (normalized) states $`|\psi _𝐳`$. The Schrödinger equation of the full system can be derived from the Lagrangian,
$`L=i\mathrm{}\psi |\dot{\psi }\psi |H|\psi ,`$ (1)
and the Lagrangian of the constrained system is obtained from this by restricting $`|\psi `$ to the subset of states $`|\psi _𝐳`$. Expressed in terms of the coordinates $`𝐳`$, it is of the generic form,
$`L(𝐳,\overline{𝐳})=A_{\overline{z}_i}\dot{\overline{z}_i}+A_{z_i}\dot{z_i}V(z_i,\overline{z}_i),`$ (2)
where $`A_z`$ is the Berry connection,
$`A_{z_i}=i\mathrm{}\psi _z|_{z_i}\psi _x,`$ (3)
$`V`$ is the expectation value of the Hamiltonian in the state $`|\psi _𝐳`$.
An important special case is when the state vectors which define $``$ are, up to normalization, analytic functions of $`z_i`$,
$`|\psi _𝐳=𝒩(\overline{𝐳},𝐳)|\varphi _𝐳,`$ (4)
where $`|\varphi _𝐳`$ denotes the analytic part of the state vector, and $`𝒩(\overline{z},z)`$ is the normalization factor. The vector potentials are then given by
$`A_{z_i}`$ $`=`$ $`i\mathrm{}_{z_i}\mathrm{ln}\overline{𝒩}(\overline{𝐳},𝐳),`$ (5)
and $``$ is a Kähler manifold where the Kähler potential is related in a simple way to the normalization factor,
$`K(\overline{𝐳},𝐳)=\mathrm{}\mathrm{ln}|𝒩(\overline{𝐳},𝐳)|^2.`$ (7)
The geometry of $``$ is described by the field strength
$`f_{\overline{z}_iz_j}`$ $`=`$ $`_{\overline{z}_i}A_j_{z_j}A_{\overline{i}}=i_{\overline{z}_i}_{z_j}K(z,\overline{z}),`$ (8)
in terms of which the symplectic form, $`\omega `$, and the metric, $`ds^2`$, takes the forms $`\omega =f_{\overline{z}_iz_j}d\overline{z}_idz_j`$ and $`ds^2=2if_{\overline{z}_iz_j}d\overline{z}_idz_j`$. The importance of the symplectic form $`\omega `$ is that it determines the Poisson brackets, and thus, together with the Hamiltonian, defines the classical mechanics . The Poisson bracket is
$`\{A,B\}=f_{z_i\overline{z}_j}^1_{z_i}A_{\overline{z}_j}B,`$ (9)
and the equation of motion can then be written as
$`\dot{z}_i=\{z_i,V\}.`$ (10)
To be more specific we now consider a system of charged particles moving in two dimensions in the presence of a strong magnetic field that restricts the available states to the lowest Landau level. In this example we can explicitly derive the metric and symplectic form, and show that they can be obtained from a Kähler potential. For calculations, it is convenient to consider particles moving on a sphere. On a unit sphere penetrated by $`2j`$ units of magnetic flux a particle with unit charge has a total angular momentum $`J=j+L`$, where $`L`$ is the orbital angular momentum, and the lowest Landau level corresponds to $`L=0`$ with a $`2j+1`$ degeneracy . We shall use the notation of ref. and define a coherent state by rotations of a minimum uncertainty reference state $`|0`$,
$`|z=D(z)|0=e^{zJ_+}e^{\eta J_0}e^{\overline{z}J_{}}|0.`$ (11)
Here the complex coordinate $`z`$ is defined via a stereographic projection, and the rotation operators, $`D(z)`$, form a unitary and irreducible representation of the rotation group, generated by $`J_m`$, $`\eta =\mathrm{ln}(1+\overline{z}z)`$, and $`|0`$ is annihilated by $`J_+`$.
Fully symmetrized and antisymmetrized states corresponding to fermions and bosons are given by
$`|𝐳,\pm =𝒩(𝐳,\overline{𝐳}){\displaystyle \frac{1}{\sqrt{N}!}}{\displaystyle \underset{P}{}}\eta _P^\pm e^{z_{i_P}J_+^i}|\mathrm{𝟎},`$ (12)
with $`\eta _P^\pm `$ the appropriate sign for the permutation $`P`$. The normalizations of these states are readily obtained from the properties of the $`D(z)`$:s,
$`|𝒩(𝐳,\overline{𝐳})|^2={\displaystyle \underset{P}{}}\eta _P{\displaystyle \underset{i}{}}\left(1+\overline{z}_{i_P}z_i\right)^{2j}.`$ (13)
For the case of N coinciding bosons, $`z_i=z`$, we immediately get the following Kähler potential
$`K(z,\overline{z})=N\mathrm{}2j\mathrm{ln}(1+\overline{z}z),`$ (14)
and the corresponding metric, $`ds^2=2N\mathrm{}/\left(1+\overline{z}z\right)^2dzd\overline{z}`$, is just $`N`$ times that of a sphere.
To assess the effect of statistics in the classical description we calculate the $`N`$-particle phase space volume. Following Manton , and Samols we can use (14) to obtain the $`N`$-particle volume from the volume of $`N`$ coinciding bosons, and the result is,
$`V_B={\displaystyle \frac{1}{N!}}(A)^N,`$ (15)
with $`A`$ as the volume of the single-particle space,
$`A=h{\displaystyle _{sph}}\omega =h2j={\displaystyle \frac{\mathrm{}4\pi R^2}{l^2}}=e\mathrm{\Phi },`$ (16)
which is $`h`$ times the number of flux quanta $`\varphi _0=h/e`$ that penetrate the sphere. Thus, for bosons the only effect of the indistinguishability of the particles is the factor $`1/N!`$, and there is no further reduction in phase space volume.
One should note that the classical phase space defined as above is everywhere a smooth manifold. This is different from the case when the N-particle space is defined as a product of single particle spaces with identification of equivalent configurations. In the latter case the points corresponding to a coincidence of two particle positions are singular. Also note that the factor $`1/N!`$ here appears naturally from the geometry, not through any additional assumption about identification of points.
For fermions a similar but technically more involved calculation can be done. The resulting phase space volume is,
$`V_F={\displaystyle \frac{1}{N!}}\left(A(N1)h\right)^N.`$ (17)
Compared with the Bose case there is an additional reduction of phase space. The available phase space for any particular fermion is reduced with an amount $`h`$ by each of the other particles present in the system. This can be understood as a classical version of the Pauli exclusion principle, and is consistent with the usual semi-classical interpretation of quantum mechanics, where each quantum state is associated with a phase space volume $`h`$. Note that there is a maximum number of particles allowed, $`N=2j+1`$, in which case the phase space volume (17) vanishes. This corresponds to all states with zero orbital angular momentum being filled, i.e. to a filled lowest Landau level, which is an incompressible state.
The calculation can be done also in the case of anyons although there is no explicit expression for the normalization constant corresponding to (13). For details we again refer to and only quote the result,
$`V_\nu ={\displaystyle \frac{1}{N!}}\left(A\nu (N1)h\right)^N,`$ (18)
where the exchange phase of the anyons is $`\nu \pi `$.
The expressions we have found above for the $`N`$-particle phase space volume demonstrates a classical fractional exclusion principle. Thus, each new particle introduced in the system will find the available volume reduced by $`\alpha =\nu h`$ relative to the previous one. The quantity $`\alpha `$, i.e. the reduction in phase space volume, can be taken as defining the classical statistics parameter of the particles. In the present case it is simply the (dimensionless) quantum statistics parameter $`\nu `$ multiplied with Planck’s constant $`h`$.
We next consider a classical field theory with soliton solutions, namely the Chern-Simons Ginzburg-Landau (CSGL) theory, originally introduced as a field theory for the quantum Hall effect ,
$`L={\displaystyle }d^2x[i\mathrm{}\varphi ^{}D_0\varphi `$ $``$ $`{\displaystyle \frac{\mathrm{}^2}{2m}}|\stackrel{}{D}\varphi |^2{\displaystyle \frac{\lambda }{4}}(|\varphi |^2\rho _0)^2`$ (19)
$`+`$ $`\mu \mathrm{}ϵ^{\mu \nu \rho }a_\mu _\nu a_\rho ],`$ (20)
where $`\varphi `$ is a complex matter field, $`a_\mu `$ a Chern-Simons field, $`m`$ a mass parameter, $`\lambda `$ the interaction strength, $`\rho _0`$ the preferred density of the system and $`\mu `$ a statistics parameter. For the original Laughlin quantum Hall states described by the model the statistics parameter takes the values $`\mu =1/[4\pi (2k+1)]`$. This theory has vortices (quasi-particles) as soliton solutions. In a certain approximation the dynamics can be described in terms of vortex coordinates alone, and a phase space description can be derived from the full theory. Again it is possible to calculate the phase space volume corresponding to a $`N`$-vortex solution.
The vortex configurations can be parameterized by a set of vortex coordinates $`𝐳=\{z_1,z_2,\mathrm{},z_n\}`$, just as the charged particles in a magnetic field discussed above. The precise form of the multi-vortex configurations for given coordinates is not known, but for critical coupling ($`\lambda =1`$) the existence of $`N`$-vortex configurations with arbitrary positions can be deduced . With the system constrained to the manifold of $`N`$-vortex configurations, a classical mechanics follows with a kinetic term for the $`N`$-vortex system corresponding to a phase space with Kähler metric . By use of earlier results obtained by Manton for the related relativistic abelian Higgs model we find for the phase space volume of $`N`$ vortices,
$`V_N={\displaystyle \frac{1}{N!}}(A4\pi \mu h(N1))^N,`$ (21)
where the classical statistics parameter, as determined by the reduction in available phase space due to the presence of other vortices, is $`\alpha =4\pi \mu hgh`$. We can interpret $`g`$, the classical parameter divided by $`h`$, as the dimensionless quantum statistics parameter. The value $`g=4\pi \mu `$ agrees with the value of the statistics parameter as determined from Berry phase calculations with Laughlin wave functions , or from the properties of vortices in the CSGL model .
We now discuss the statistical mechanics of the classical systems just derived. Both these systems have the special property that the energy does not depend on the state, but only on the number of particles. This means that the statistical mechanics is determined by the phase space volume $`V_N`$, which has been determined in the previous sections, and by the energy $`E_N`$. The classical partition function is simply the total number of states, $`V_N/h^N`$ multiplied with the Boltzmann factor, i.e.
$`Z_N={\displaystyle \frac{V_N}{h^N}}e^{\beta E_N},`$ (22)
Using standard thermodynamics we get for the entropy
$`S`$ $`=`$ $`N\mathrm{ln}(1\alpha \rho )+N\mathrm{ln}{\displaystyle \frac{A}{h}}N\mathrm{ln}N+N,`$ (23)
where $`\alpha =\nu h`$ or $`gh`$, and where we have introduced the classical phase space density $`\rho =N/A`$ and neglected the difference between $`N`$ and $`N1`$, which is irrelevant in the thermodynamic limit.
In the systems we have considered the real two-dimensional space where the particles or vortices move is proportional to the phase space. Defining the pressure as, $`P=\left(F/A\right)_T`$, where $`A=V_1`$ is the phase space volume for a single particle, we then get, by use of standard thermodynamic relations, the equation of state
$`\beta P`$ $`=`$ $`\rho /(1\alpha \rho ).`$ (24)
This expression shows that there is a maximum density $`\rho =1/\alpha `$ allowed by the system, which corresponds to an infinite pressure and therefore to an incompressible state. For the phase space volume this means $`V_N=0`$, i.e. there is no available phase space volume for any new particle added to the system. For the anyon system this situation corresponds to a completely filled Landau level. What is unusual about this is that the blocking, which can be interpreted as representing a generalized Pauli principle, shows up not only in the quantum, but also in the classical description of the system.
Finally we show that the thermodynamics just derived can be viewed as a classical limit of Haldane exclusion statistics .
The statistical mechanics of particles with exclusion statistics can be derived by assuming that the total energy can be written as a sum of single-particle energies and using the prescriptions for statistical weight given by Haldane separately for each single-particle energy level . The result for the entropy is
$`S`$ $`=`$ $`{\displaystyle \underset{k}{}}D_k\{[1+(1g)n_k]\mathrm{ln}[1+(1g)n_k]`$ (25)
$`+`$ $`(1gn_k)\mathrm{ln}(1gn_k)n_k\mathrm{ln}n_k\},`$ (26)
where the sum runs over single-particle energy states, and $`g`$ is the exclusion statistics parameter. $`D_k`$ is the degeneracy of the $`k`$-th level and the quantum distribution function $`n_k`$ is the average occupation number of the state $`k`$.
Since each quantum state occupies the phase space volume $`h^𝒟`$, with $`2𝒟`$ the dimension of the single-particle phase space, we can relate $`n`$ and $`\rho `$ in the semiclassical limit by $`n=\rho h^𝒟`$. In the Boltzmann limit, $`h0`$ and $`n0`$, all dependence on $`g`$ in (25) goes away. If we, however, define the classical physics by the limit $`h0`$, $`g\mathrm{}`$ and $`gh^𝒟\alpha `$, where $`\alpha `$ is interpreted as a classical statistics parameter (25) gets the nontrivial limit of
$`S={\displaystyle \underset{k}{}}D_kh^𝒟\left[\rho _k\mathrm{ln}(1\alpha \rho _k)\rho _k\mathrm{ln}(\rho _kh)+\rho _k\right].`$ (27)
If we further specialize to the case of fully degenerate states in a two-dimensional phase space, where the sum is simply replaced by the total number of available single-particle states, $`G=A/h`$, and where $`\rho _k`$ is replaced by $`N/A`$, we exactly regain (23). This demonstrates that the classical statistical mechanics discussed in the previous section can be regarded as a special limit of exclusion statistics, different from the Boltzmann limit. Starting from the the equation of state for exclusion statistics particles with the same energy, we can also derive (24) if we identify the the physical volume $`𝒱`$ with the one particle phase space volume $`A`$.
One can also obtain exact results when the particles move in an external harmonic potential. Again one finds equivalence between the statistical mechanics derived from the classical mechanics of identical particles with a statistics parameter and the statistical mechanics derived from exclusion statistics in the classical limit discussed above .
Although the phase space in the examples discussed in this paper are two-dimensional, there is no obvious reason for the construction not to work in higher dimensions. One point of particular interest to study further is how the “classical fermion” theory discussed here can be applied to systems of interacting fermions moving in two or three dimensions.
We thank A. Karlhede for discussions and comments on the manuscript. T. H. Hansson and U. Lindström were supported by the Swedish Natural Science Research Council. S. Isakov acknowledges the support received through a NATO Science Fellowship granted by the Norwegian Research Council, and also appreciates the warm hospitality of NORDITA during his stay there in the summer of 1999, where part of this work was done.
Present address: NumeriX Corporation, 546 Fifth Avenue, New York, NY 10036, USA
|
warning/0004/hep-th0004134.html
|
ar5iv
|
text
|
# Quantization of Four-form Fluxes and Dynamical Neutralization of the Cosmological Constant
## 1 Introduction
The cosmological constant problem is one of the central challenges in quantum gravity, and it continues to provoke novel and interesting ideas.<sup>1</sup><sup>1</sup>1For a classic survey of various approaches see Weinberg’s review . A brief survey of recent ideas is given in Ref. . In some approaches, the cosmological ‘constant’ becomes a dynamical variable: it is locally constant but has a continuous range of allowed values, with the effective value in our universe determined by some dynamical principle.
There are several mechanisms by which the cosmological constant can become a dynamical variable. One is through the existence of a four-form field strength . The equation of motion requires that such a field strength be constant, so it has no local dynamics but contributes a positive energy density, which can cancel a cosmological constant coming from other sourcs if the latter is negative. A second mechanism is fluctuations of the topology of spacetime (wormholes). Under a plausible interpretation of the path integral for quantum gravity these convert all constants of Nature into dynamical variables , though there is serious doubt as to whether this effect exists in a real theory of quantum gravity. A third mechanism is the existence of naked singularities in compactified dimensions , where the undetermined boundary conditions at the singularity become a variation of the effective four dimensional Lagrangian.
In this paper we will discuss certain aspects of the four-form idea, though at the end we will note that very similar considerations may apply to the naked singularity. Our first point is that the four-form field strength, although usually assumed to take continuous values, is in fact quantized. This quantization might be evaded in a purely four-dimensional theory, but is certainly necessary when gravity is embedded in a higher-dimensional theory such as M theory. The size of the quantum is fixed by microscopic physics, and so the spacing of energy densities is enormous compared to the actual value, or bound, on the cosmological constant. Therefore the four-form cannot play the assumed role of producing a small cosmological constant.
This is discouraging, but there is a variant of the four-form idea which has some interesting features. Typical M theory compactifications have extra four-form field strengths, arising from nontrivial three-cycles in the compactification. If there are several four-form field strengths, with incommensurate charges, then the allowed cosmological constants may form a closely spaced ‘discretuum’, with one or more values in the experimentally allowed range. The universe can reach such a value, starting from a larger density, through a series of domain wall nucleations. This resembles an idea of Brown and Teitelboim , but has some unique and attractive features. In particular, there is a plausible mechanism for heating the universe after nucleation produces a small cosmological constant.
A second complication is that in higher dimensional theories there are in general moduli, and the four-form does not produce a constant energy density but rather a potential for the moduli. The analysis of the four-form flux therefore cannot be separated from the consideration of the stability of the compactification. This is a difficult issue for a number of reasons, and aside from a brief discussion will sidestep it by working in an artificial, model where the charges are frozen in incommensurate ratios. Although this is rather optimistic, it may be that some features of the cosmology that we find will survive in more realistic circumstances.
In Sec. 2 we review the physics of four-form fluxes and explain how a generalized Dirac quantization condition constrains the value of the four-form flux. We then investigate the level spacing in a theory of many four-forms, as generally arise in M theory compactifications. We find that the discretuum is sufficiently dense if there is a membrane charge of order $`10^1`$ and a large number of fluxes, say 100. The large dimension scenario produces much smaller charges, and can lead to a sufficiently dense discretuum for as few as four fluxes.
In Sec. 3 we discuss the resulting cosmology. We review the Brown-Teitelboim scenario, in which a cosmological constant is neutralized by nucleation of membranes. We then extend this to multiple four-forms. If the flux density is initially large, so that the cosmological constant is positive, then one obtains a picture much like eternal inflation, where the cosmological constant takes different values in different expanding bubbles. De Sitter thermal effects provide a natural solution to one of the serious problems of the Brown-Teitelboim idea. The inflaton can be stabilized in the inflationary part of its potential until the nucleation reduces the cosmological constant to near zero, at which point it begins to roll. This is possible because with multiple four-forms the individual jumps in the cosmological constant can be quite large. In the end the observed cosmological constant is small for anthropic reasons, but in the weakest sense: we have a universe with different cosmological constants in different regions, and with galaxies only in regions of small cosmological constant. In many respects our picture resembles an idea of Banks . Another example of a discretuum is the irrational axion . While this work was being completed we learned that Feng, March-Russell, Sethi, and Wilczek are also considering extensions of the mechanism of Brown and Teitelboim.
## 2 Four-form quantization
### 2.1 Four-form energetics
We first review the basic physics of four-form field strengths. For antisymmetric tensor fields, the language of forms is used when convenient; this is indicated by bold face ($`𝐅_\mathrm{𝟒}`$). Normal fonts are used for index notation, or when index notation is implied, e.g. $`F_4^2=F_{\mu \nu \rho \sigma }F^{\mu \nu \rho \sigma }`$.
The action for gravity with a bare vacuum energy $`\lambda _{\mathrm{bare}}`$ plus four-form kinetic term is
$$S=d^4x\sqrt{g}\left(\frac{1}{2\kappa _4^2}R\lambda _{\mathrm{bare}}\frac{Z}{24!}F_4^2\right)+S_{\mathrm{branes}},$$
(2.1)
where $`𝐅_\mathrm{𝟒}=\mathrm{𝐝𝐀}_\mathrm{𝟑}`$. We include a general normalization constant $`Z`$ in the kinetic term for later convenience. Certain boundary terms must be added to this action. They do not affect the equations of motion and will not be prominent in the remainder of this paper. However, they are crucial for the correct evaluation of the on-shell action when physical quantities are measured on an equal time hypersurface $`\mathrm{\Sigma }`$. The usual Gibbons-Hawking term is given by
$$S_{\mathrm{GH}}=\frac{1}{\kappa _4^2}_\mathrm{\Sigma }d^3x\sqrt{h}K.$$
(2.2)
For the four-form field the following boundary term must be included to obtain stationary action under variations that leave $`F`$ fixed on the boundary :
$$S_{\mathrm{DJ}}=\frac{Z}{3!}d^4x_\mu \left(\sqrt{g}F^{\mu \nu \rho \lambda }A_{\nu \rho \lambda }\right).$$
(2.3)
On shell its value is negative twice the $`F^2`$ contribution in the volume term of the action. This removes the apparent discrepancy between the cosmological constant in the on-shell action and in the equations of motion.
Ignoring the brane sources (we will consider them shortly), the four-form equation of motion is $`_\mu \left(\sqrt{g}F^{\mu \nu \rho \sigma }\right)=0`$, with solution
$$F^{\mu \nu \rho \sigma }=cϵ^{\mu \nu \rho \sigma },$$
(2.4)
where $`ϵ^{\mu \nu \rho \sigma }`$ is the totally antisymmetric tensor and $`c`$ is any constant. Thus there is no local dynamics. One has $`F_4^2=24c^2`$, and so the on-shell effect of the four-form is indistinguishable from a cosmological constant term. The Hamiltonian density is given by
$$\lambda =\lambda _{\mathrm{bare}}\frac{Z}{48}F_4^2=\lambda _{\mathrm{bare}}+\frac{Zc^2}{2}.$$
(2.5)
Only $`\lambda `$ is observable: $`\lambda _{\mathrm{bare}}`$ and the four-form cannot be observed separately in the four-dimensional theory. Therefore, the bare cosmological constant can be quite large. For example, it might be on the Planck scale or on the supersymmetry breaking scale. In order to explain the observed value of the cosmological constant, $`\lambda _{\mathrm{bare}}`$ must be very nearly cancelled by the four-form contribution.
### 2.2 Four-form quantization
In the original work , and in many recent applications, it as assumed that the constant $`c`$ can take any real value, thus cancelling the bare cosmological constant to arbitrary accuracy. However, we are asserting that the value of $`c`$ is quantized. Since this is somewhat counterintuitive, let us first discuss two things that the reader might think we are saying, but are not.
First, if there is a gravitational instanton, a Euclidean four-manifold $`X`$, then it is natural to expect that the integral of the Euclidean four-form over $`X`$ is quantized,
$$_X𝐅_\mathrm{𝟒}=\frac{2\pi n}{e},n𝐙.$$
(2.6)
This is the generalized Dirac quantization condition . It arises from considering the quantum mechanics of membranes, which are the natural objects to couple to the potential $`𝐀_\mathrm{𝟑}`$,
$$S=e_W𝐀_\mathrm{𝟑}$$
(2.7)
with $`e`$ the membrane’s charge and $`W`$ its world-volume. The condition that membrane amplitudes be single-valued then implies the quantization (2.6). This is true, but we are asserting something in addition: that the actual value of $`𝐅_\mathrm{𝟒}`$ (or, more precisely, $`c`$) is quantized, in addition to the integral.
Of course, the inclusion of membranes means that $`c`$ is no longer globally constant, as the membranes are sources for $`𝐅_\mathrm{𝟒}`$. The value of $`c`$ jumps across a membrane,
$$\mathrm{\Delta }c=\frac{e}{Z}.$$
(2.8)
The total change in $`c`$ due to nucleation of any number of membranes is then a multiple of $`e/Z`$.
However, it is not this change that we are asserting is quantized, but the actual value:
$$c=\frac{en}{Z},n𝐙.$$
(2.9)
This may seem surprising, but in fact is quite natural. String theory has the satisfying property that for every gauge field there exist both electric and magnetic sources. This implies a quantization condition both for the field strength and its dual. The dual of a four-form is a zero-form,
$$𝐅_\mathrm{𝟒}=𝐅_\mathrm{𝟎}.$$
(2.10)
A zero-form is naturally integrated over a zero-dimensional manifold, which is to say that it is evaluated at a point. The generalized Dirac condition is that this be quantized, which is precisely Eq. (2.9):
$$𝐅_\mathrm{𝟎}=\frac{en}{Z},n𝐙.$$
(2.11)
The quantizations (2.6) and (2.11) are in just the usual relation for $`n`$-form and $`(dn)`$-form field strengths in $`d`$ spacetime dimensions.
Although natural, it is not clear that the quantization of $`𝐅_\mathrm{𝟎}`$ is necessary. The quantization of $`𝐅_\mathrm{𝟒}`$ arises from the consistency of the quantum mechanics of 2-branes, but that of $`𝐅_\mathrm{𝟎}`$ would come from the quantum mechanics of $`(2)`$-branes, and it is not clear what this should be. Further, there is the example of the Schwinger model, where the non-integer part of $`𝐅_\mathrm{𝟎}`$ is just the $`\theta `$-parameter, which can take any real value.
Nevertheless, the quantization condition (2.11) is necessary when the four-dimensional theory is embedded in string theory.<sup>2</sup><sup>2</sup>2This observation grew out of Ref. , where quantization of a top-form (or zero-form) field strength first appeared. Consider for example the compactification of M theory on a seven-manifold $`K`$. We begin with the eleven-dimensional action
$$S=2\pi M_{11}^9d^{11}X\sqrt{g_{11}}\left(R\frac{1}{24!}F_4^2\right)+S_{\mathrm{branes}},$$
(2.12)
where we omit the Chern-Simons and fermion terms, which will play no role. With this normalization the M2-brane tension and charge are $`2\pi M_{11}^3`$, and the M5-brane charge and tension are $`2\pi M_{11}^6`$.<sup>3</sup><sup>3</sup>3For a review see Ref. . The M5-brane couples to $`𝐀_\mathrm{𝟔}`$,
$$2\pi M_{11}^6_W𝐀_\mathrm{𝟔},$$
(2.13)
where $`W`$ is the M5 world-volume, and
$$\mathrm{𝐝𝐀}_\mathrm{𝟔}=𝐅_\mathrm{𝟕}=_{11}𝐅_\mathrm{𝟒},$$
(2.14)
where a subscript is used to distinguish the dual in eleven-dimensions from that in four dimensions. By the generalized Dirac quantization it follows that
$$2\pi M_{11}^6_K𝐅_\mathrm{𝟕}=2\pi n,n𝐙.$$
(2.15)
Now reduce to four dimensions. The eleven dimensional $`𝐅_\mathrm{𝟒}`$ reduces directly to a four dimensional $`𝐅_\mathrm{𝟒}`$, with action
$$S=V_72\pi M_{11}^9d^4x\sqrt{g}\left(R\frac{1}{24!}F_4^2\right)+S_{\mathrm{branes}},$$
(2.16)
where $`V_7`$ is the volume of $`K`$. Further, the condition (2.15) becomes
$$𝐅_\mathrm{𝟎}=\frac{n}{M_{11}^6V_7},n𝐙.$$
(2.17)
That is,
$$(2\kappa _4^2)^1=Z=2\pi M_{11}^9V_7.$$
(2.18)
The quantization (2.17) matches that found in Eq. (2.11) with $`e=2\pi M_{11}^3`$, which is just the M2-brane tension.
### 2.3 Discussion
The quantization that we have found rules out the precise cancellation of the cosmological constant that has been assumed in many discussions. Brown and Teitelboim considered the approximate neutralization of the cosmological constant by a field strength taking discrete values (see also Abbott for a closely related idea). In order that this be natural, the spacing between allowed values of $`\lambda `$ must be of order the observational bound. Since $`d\lambda /dn=2ne^2/Z`$ and $`n_{\mathrm{final}}\sqrt{|\lambda _{\mathrm{bare}}|Z}/e`$, the final value of $`\lambda `$ will lie within observational bounds only if
$$e|\lambda _{\mathrm{bare}}|^{1/2}Z^{1/2}<10^{120}\kappa _4^4.$$
(2.19)
Using the results above for $`e`$ and $`Z`$, the left-hand side (dropping $`2\pi `$’s) is
$$|\lambda _{\mathrm{bare}}|^{1/2}\kappa _4^{1/3}V_7^{1/3}|\lambda _{\mathrm{bare}}|^{1/2}\kappa _4M_{11}^3.$$
(2.20)
The step size is minimized in the low-energy string scenario , where $`\lambda _{\mathrm{bare}}`$ and $`M_{11}`$ are both TeV-scale, but even in this case it is far too large, $`10^{75}\kappa _4^4`$.
This is the ‘gap problem’: the Brown-Teitelboim mechanism requires an energy spacing which is infinitesimal compared to the scales of microphysics. In the next subsection we will consider compactification with multiple four-forms, which can reduce the step size to an acceptable value.
Because the compactification volume $`V_7`$ is a dynamical quantity and not a fixed parameter, the four-form energy density is not a constant but a potential for $`V_7`$. In a realistic compactification this must be stabilized, and the energetics of the four-form fluxes will enter into the stabilization.<sup>4</sup><sup>4</sup>4See for example the discussions . Thus the volume $`V_7`$ itself depends on $`n`$, and so the effective cosmological constant has additional $`n`$-dependence beyond that included above. For convenience we will in the rest of this paper ignore this effect, treating the geometry as fixed.
It should be noted that the allowed flux actually depends additively on the values of flat background gauge potentials<sup>5</sup><sup>5</sup>5We thank E. Witten for pointing this out. — these are just stringy generalizations of the Schwinger model $`\theta `$-parameter. As these backgrounds vary the flux can take arbitrary real values. This does not, however, restore the original continuously variable cosmological constant, because these background potentials are moduli and not parameters. As with the compactification geometry, these background moduli must eventually be stabilized and so the fluxes will in fact take discrete values.
### 2.4 Multiple four-forms
General compactifications actually give rise to several four-form fluxes, and this can solve the gap problem. Let there be $`J`$ such fluxes, with
$$\lambda =\lambda _{\mathrm{bare}}+\frac{1}{2}\underset{i=1}{\overset{J}{}}n_i^2q_i^2.$$
(2.21)
The question is whether there exists a set of $`n_i`$ such that
$$2|\lambda _{\mathrm{bare}}|<\underset{i=1}{\overset{J}{}}n_i^2q_i^2<2(|\lambda _{\mathrm{bare}}|+\mathrm{\Delta }\lambda ),$$
(2.22)
where $`\mathrm{\Delta }\lambda `$ corresponds to the observational bound, roughly $`10^{120}`$ in Planck units. This can be visualized in terms of a $`J`$-dimensional grid of points, spaced by $`q_i`$ and labeled by $`n_i`$ (see Fig. 1).
Consider a sphere of radius $`r=|2\lambda _{\mathrm{bare}}|^{1/2}`$ centered at $`n_i=0`$. If one of the points $`(n_1,n_2,\mathrm{},n_J)`$ is sufficiently close to the sphere, the field configuration corresponding to this point will lead to an acceptable value of the cosmological constant.
More precisely, one should think of a thin shell, whose width encodes the width of the observational range,
$$\mathrm{\Delta }r=|2\lambda _{\mathrm{bare}}|^{1/2}\mathrm{\Delta }\lambda .$$
(2.23)
We need at least one point to lie within the shell. As we will discuss, there may be large degeneracies — let the typical degeneracy be $`D`$. The volume per $`D`$ grid points must then be less than the volume of the shell, $`\omega _{J1}r^{J1}\mathrm{\Delta }r`$, where the area of a unit sphere is $`\omega _{J1}=2\pi ^{J/2}/\mathrm{\Gamma }(J/2)`$. Thus
$$\underset{i=1}{\overset{J}{}}q_i\frac{\omega _{J1}}{D}|2\lambda _{\mathrm{bare}}|^{\frac{J}{2}1}\mathrm{\Delta }\lambda ,$$
(2.24)
or
$$\frac{D}{\omega _{J1}}\underset{i=1}{\overset{J}{}}\frac{q_i}{|2\lambda _{\mathrm{bare}}|^{\frac{1}{2}}}\frac{\mathrm{\Delta }\lambda }{|2\lambda _{\mathrm{bare}}|}.$$
(2.25)
In other words, the typical spacing of the spectrum of the cosmological constant in a model with given $`J`$, $`e_i`$, and $`\lambda _{\mathrm{bare}}`$ will be given by
$$\mathrm{\Delta }\lambda _{\mathrm{min}}=\frac{D_{i=1}^Jq_i}{\omega _{J1}|2\lambda _{\mathrm{bare}}|^{\frac{J}{2}1}}.$$
(2.26)
An important feature of this result is that that the $`q_i`$ need not be exceedingly small if there are more than two four-form fields. In order to achieve a small $`\lambda `$, it is sufficient that there be a discrepancy between the magnitude of $`\lambda _{\mathrm{bare}}`$ and that of the charges. For fixed charges, the task of cancellation actually becomes easier, the larger the bare cosmological constant. This can be understood from Fig. 1. The larger the shell, the more points it will contain.<sup>6</sup><sup>6</sup>6Note, however, that the radius of the shell in Fig. 1 represents not $`|2\lambda _{\mathrm{bare}}|`$, but the square root of $`|2\lambda _{\mathrm{bare}}|`$. This is why one cannot recognize in Fig. 1 the need for the charges $`q_i`$ to be incommensurate, a fact that is immediately clear from Eq. (2.21). It is also the reason why increasing $`|\lambda _{\mathrm{bare}}|`$ has no beneficial effect in the case of $`J=2`$. For fixed $`\mathrm{\Delta }\lambda `$, the shell gets thinner as one increases its radius. If $`J=2`$, this precisely compensates for the increase of the shell radius, and the volume remains constant. The results (2.24) to (2.26) treat the $`n_i`$ as essentially continuous, and break down if any of the $`q_i`$ exceed $`J^{1/2}|2\lambda _{\mathrm{bare}}|^{1/2}`$. In this case the flux associated with $`q_i`$ should simply be ignored.
For illustration suppose that $`\lambda _{\mathrm{bare}}`$ is at the Planck scale, $`(\sqrt{2}\kappa _4)^4\lambda _{\mathrm{bare}}1`$, that the number of four-forms is 100, a number which is large but not unrealistic, and that $`D`$ is small. Then the inequality (2.25) implies that the typical charge must be of order $`10^{1.6}`$ in Planck units; note that $`q_i`$ is a mass-squared, so we should perhaps measure the smallness by the square root, $`10^{0.8}1/6`$. However, the assumption of no degeneracy is rather optimistic, as we will discuss in the next subsection.
### 2.5 M Theory Compactification
Consider the compactification of M theory on a general manifold $`K`$. The total number of fluxes is $`J=N_3+1`$, where $`N_3`$ is the number of nontrivial three-cycles of $`K`$. For each nontrivial three-cycle $`C_i`$ there is a harmonic three-form $`\omega _{\mathrm{𝟑},i}`$, and the seven-form field strength can be expanded
$$𝐅_\mathrm{𝟕}=\frac{1}{M_{11}^3}\underset{i=1}{\overset{N_3}{}}𝐅_{\mathrm{𝟒},i}(x)\omega _{\mathrm{𝟑},i}(y)+𝐅_{\mathrm{𝟒},N_3+1}(x)ϵ_\mathrm{𝟕}(y).$$
(2.27)
Here $`ϵ_\mathrm{𝟕}`$ is the volume form on $`K`$, so that $`𝐅_{\mathrm{𝟒},N_3+1}`$ is the flux discussed previously, obtained simply by reduction of the eleven-dimensional flux. Coordinates have been labeled as follows:
$$(X^0,\mathrm{},X^{11})=(x^0,\mathrm{},x^3,y^1,\mathrm{},y^7)(x^\mu ,y^m).$$
(2.28)
Associated to each flux $`𝐅_{\mathrm{𝟒},i}`$ is a four-dimensional domain wall (membrane), obtained by wrapping three legs of the M5-brane on $`C_i`$.
Let us illustrate this by a simple model, in which $`K`$ is simply a seven-torus with flat internal metric $`\delta _{mn}`$, and with $`y^m`$ identified with period $`2\pi r_m`$; then $`V_7=_{m=1}^7(2\pi r_m)`$. There is one three-cycle $`C_i`$ for each unordered triplet $`(m_i,m_i^{},m_i^{\prime \prime })`$, or $`(\genfrac{}{}{0pt}{}{7}{3})=35`$ in all. The volume and three-form associated with $`C_i`$ are
$$V_{3,i}=(2\pi )^3r_{m_i}r_{m_i^{}}r_{m_i^{\prime \prime }},\omega _{\mathrm{𝟑},i}=\frac{1}{V_{3,i}}\mathrm{𝐝𝐲}^{m_i}\mathrm{𝐝𝐲}^{m_i^{}}\mathrm{𝐝𝐲}^{m_i^{\prime \prime }}.$$
(2.29)
The four-dimensional action is
$$S=d^4x\sqrt{g}\left(\frac{1}{2\kappa _4^2}R\lambda _{\mathrm{bare}}\frac{1}{24!}\underset{i=1}{\overset{N_3+1}{}}Z_iF_{4,i}^2\right)+S_{\mathrm{branes}}.$$
(2.30)
Here
$$Z_i=\frac{2\pi M_{11}^3V_7}{V_{3,i}^2}(iN_3),\frac{1}{2\kappa _4^2}=Z_{N_3+1}=2\pi M_{11}^9V_7.$$
(2.31)
The bare cosmological constant has been added by hand in this model. In a real compactification, negative energy density can arise from positive scalar curvature or an orientifold plane, for example. The tension of a membrane wrapped on $`C_i`$ is
$$\tau _i=2\pi M_{11}^6V_{3,i}(iN_3),\tau _{N_3+1}=2\pi M_{11}^3.$$
(2.32)
Its coupling to the $`j`$’th three-form potential is
$$e_{i,j}=e\delta _{ij},e=2\pi M_{11}^3.$$
(2.33)
The quantization condition is
$$𝐅_{\mathrm{𝟎},i}=\frac{en_i}{Z_i},$$
(2.34)
and the effective cosmological constant is
$$\lambda =\lambda _{\mathrm{bare}}+\underset{i=1}{\overset{N_3+1}{}}\frac{e^2n_i^2}{2Z_i},$$
(2.35)
so that
$$q_i=eZ_i^{1/2}.$$
(2.36)
Thus,
$$q_i=\frac{(2\pi )^{1/2}M_{11}^{3/2}V_{3,i}}{V_7^{1/2}}(iN_3),q_{N_3+1}=\frac{(2\pi )^{1/2}}{M_{11}^{3/2}V_7^{1/2}}.$$
(2.37)
Note that $`q_i^2=2\kappa _4^2\tau _i^2`$ for all $`i`$.
If the radii are appropriately incommensurate then so are the charges. However, the degeneracy $`D`$ is still nontrivial, $`2^J`$, from $`n_in_i`$ for each $`i`$ (note that in the $`J=100`$ model this reduces $`q_i^{1/2}`$, but only by $`\sqrt{2}`$). This can be reduced to $`D=2`$ by skewing the torus, which couples the different $`n_i`$. However, if the stabilization respects the symmetries of the torus there will be an even larger degeneracy: permutations of the axes, obviously, and much more — the full $`E_{7(7)}`$ $`U`$-duality . The resulting $`D`$ could significantly change the density of levels. A less symmetric compactification will have a much smaller duality group, but we do not know how to estimate a reasonable degeneracy. This effect becomes less important with fewer fluxes.
### 2.6 Small charges from large dimensions
From Eq. (2.24) one finds that a Planck-size cosmological constant can be cancelled in a model with $`j=100`$ types of membranes with $`q_i^{1/2}`$ of order $`1/6`$ in Planck units. If only a few four-forms are present, much smaller charges will be required. For examples, with $`j=6`$ and $`D`$ small one would need $`q_i^{1/2}10^{10}`$ in Planck units. However, these are not small quantities compared to other numbers in elementary particle physics. Indeed, small membrane charges can be related to the gauge hierarchy, if the latter arises by confining the gauge fields to a three-brane living in eleven dimensions and taking some of the extra dimensions to be large, as proposed in Ref. .
The large internal dimensions will play a double role here. They are the origin of the gauge hierarchy. But in addition, they will lead to small membrane charges, if membranes arise by wrapping a five-brane as described in the previous subsection. This amounts to reducing the gauge hierarchy problem and the cosmological constant problem to the single problem of stabilizing large radii.
In such models the fundamental scale $`M_{11}`$ is assumed to be near a TeV. The reduction $`(2\kappa _4^2)^1=2\pi M_{11}^9V_7`$ then determines $`V_7`$ to be large in fundamental units. For illustration, consider again the seven-torus, with $`k`$ large dimensions of size
$$2\pi r_l=\frac{1}{M_{11}}(V_7M_{11}^7)^{1/k},l=1,\mathrm{},k,$$
(2.38)
and $`7k`$ dimensions of radius 1 in fundamental units:
$$2\pi r_l=\frac{1}{M_{11}},l=k+1,\mathrm{},7.$$
(2.39)
(In general, of course, the radii could have a range of different sizes. It is trivial to extend this discussion accordingly.) Recalling the charges
$$q_i=\frac{(2\pi )^{1/2}M_{11}^{3/2}V_{3,i}}{V_7^{1/2}}(i<J),q_J=\frac{(2\pi )^{1/2}}{M_{11}^{3/2}V_7^{1/2}},$$
(2.40)
it is most favorable to consider only $`q_J`$ plus those $`q_i`$ for which all the dimensions are small. For these, of which there are $`J_0=(\genfrac{}{}{0pt}{}{7k}{3})`$ the charges $`q_i=q_J`$.
We will consider a more general compactification with the same number and sizes of dimensions, but not be restricted by the $`J_0`$ attainable on the torus. The condition (2.24) that the charges $`q_i`$ allow for a sufficiently dense spectrum for $`\lambda `$ becomes
$$\frac{\omega _{J^{}1}}{D}|2\lambda _{\mathrm{bare}}|^{\frac{J^{}}{2}1}\mathrm{\Delta }\lambda \underset{i=1}{\overset{J}{}}q_i=(2\pi )^{J^{}/2}M_{11}^{3J^{}/2}V_7^{J^{}/2}=(2\pi )^{J^{}/2}M_{11}^{3J^{}}\kappa _4^J^{},$$
(2.41)
where $`J^{}J_0+1`$.
What is the bare cosmological constant in models with large extra dimensions? It receives contributions from the tension of the three-brane, $`\lambda _{\mathrm{brane}}`$, and from the bulk vacuum energy, $`\lambda _{\mathrm{bulk}}`$ (as usual, all contributions from quantum field theory are taken to be subsumed in these quantities) :
$$\lambda _{\mathrm{bare}}=\lambda _{\mathrm{brane}}+V_7\lambda _{\mathrm{bulk}}.$$
(2.42)
The most natural value for the brane tension is
$$\lambda _{\mathrm{brane}}2\pi M_{11}^4.$$
(2.43)
(This value does not follow uniquely from the fundamental theory. The factor of $`2\pi `$ has been included to mimic the form of the M2- and M5-brane tensions.) It is natural (but not necessary) to assume that $`\lambda _{\mathrm{bulk}}`$ is generated by supersymmetry breaking on the brane. This suppresses the vacuum energy by a factor of the compact volume: $`\lambda _{\mathrm{bulk}}2\pi M_{11}^4/V_7`$, so that both terms in Eq. (2.42) will be of order $`2\pi M_{11}^4`$. (Indeed, the cosmological constant problem in these models amounts to the assumption that the two terms cancel—an assumption that obviously will not be made here.) Supersymmetry breaking in the bulk could lead to a higher value for $`|\lambda _{\mathrm{bulk}}|`$; ultimately, the only constraint comes from bulk stability , which is weaker. Recall however that the cancellation mechanism becomes more accurate, the larger the magnitude of $`\lambda _{\mathrm{bare}}`$. We can therefore work with the value of Eq. (2.43).
The condition on the charges now becomes
$$(2^{1/2}\kappa _4M_{11})^{J^{}+4}10^{120}\frac{\omega _{J^{}1}}{\pi D}.$$
(2.44)
For the extreme low-dimension picture, where $`M_{11}`$ is of order 1 TeV, this allows the very modest value $`J^{}=4`$, independent of the number $`k`$ of large dimensions. This assumes that $`D`$ is not enormous, as is reasonable for a small value of $`J^{}`$. If we increase $`J^{}`$ to 5 then $`M_{11}`$ can increase to 30 TeV.
If we take the value $`\kappa _4M_{11}10^{1.5}`$ that is appropriate to the Witten GUT scenario , then we need a large number of fluxes, again of order 100 (the precise number is sensitive to uncertain numerical factors, for example in $`\lambda _{\mathrm{bare}}`$). Note also that this requires a cosmological constant of order the GUT scale; a weak-scale cosmological constant cannot be cancelled by our mechanism in this case.
## 3 Cosmology
In the previous section we showed that multiple four-form strengths arise in most M theory compactifications, and that these could lead to a spectrum of effective cosmological constants sufficiently finely spaced that some would lie in the observational range. We must now ask why the cosmological constant that we see actually takes such a small value.
### 3.1 The Brown-Teitelboim mechanism
There are two possible approaches. One could attempt to use the framework of quantum cosmology to argue that the universe was created with $`\lambda `$ equal to the smallest positive value in the spectrum . The other possibility is to identify a dynamical mechanism by which an appropriate value of $`\lambda `$ is obtained.
We will not follow the quantum cosmology approach. It has the disadvantage that the creation of a space-time from nothing (as opposed to the quantum creation of objects on a given background) is not well understood and possibly ill-defined. The wave-function of Hartle and Hawking would indeed be sharply peaked at the smallest possible value for the cosmological constant. But this would include the effective cosmological constant from any inflaton potential $`V(\varphi )`$, so that there would not be any period of inflation in generic models. The proposals of Linde and Vilenkin , on the other hand, would give preference to a large effective cosmological constant, which could come from any combination of contributions from the four-forms and the inflaton. To cancel the cosmological constant one would then need a dynamical effect anyway.
Thus we will employ a dynamical mechanism based on the creation of membranes. This is the approach followed by Brown and Teitelboim (BT) , who considered the first model discussed in Sec. 2, with a bare cosmological constant and a single four-form field strength. We will review the dynamics of this case before we generalize the mechanism to multiple four-forms.
BT take $`\lambda _{\mathrm{bare}}`$ to be negative and $`n`$ large and positive, so that $`\lambda >0`$. Thus the universe will initially be described by de Sitter space. On this background, membrane bubbles can nucleate spontaneously. They appear at a critical size and then expand. This is a non-perturbative quantum effect. Its semi-classical amplitude can be estimated from the Euclidean action of appropriate instanton solutions . Inside the membrane, the value of $`n`$ will be lower or higher by 1, and correspondingly the cosmological constant changes by $`(\pm n+1/2)q^2`$.
Increase of $`n`$ occurs though a dominantly gravitational instanton. It has no equivalent in non-compact spaces, as follows immediately from energy conservation. The instanton for a decreasing cosmological constant is similar to the Coleman instanton for false vacuum decay in flat space , with a small correction from gravity. Consequently, the amplitude for increasing the cosmological constant is vastly more suppressed than that for decrease, and one may neglect increase.
Starting from a generic, large value of $`\lambda `$, repeated membrane creation thus produces de Sitter regions with smaller and smaller cosmological constant. The nucleation rate decreases with $`\lambda `$. For a certain range of parameters, membrane creation becomes infinitely suppressed by gravitational effects once $`\lambda `$ is no longer positive.
The BT process is analogous to the neutralization of an electric field (an $`𝐅_\mathrm{𝟐}`$) wrapped around a circle in a (1+1)-dimensional world. The Schwinger pair creation of (zero-dimensional) charged particles decreases the field until it has too little energy to nucleate another pair.
BT identified two problems with this scenario. One is the ‘empty universe problem’, which we will address in Sec. 3.2. The other is the ‘gap problem’, that in order for some values of the cosmological constant to lie within the observational window one needs membrane charges that are enormously small compared to the ordinary scales of microphysics. We have shown that this problem may be absent in a theory with multiple four-forms.
Let us therefore extend the BT mechanism to the case of $`J`$ four-forms. In the simplest models one can take $`n_i0`$ without loss of generality. For the initial configuration $`(n_{1,\mathrm{initial}},\mathrm{},n_{J,\mathrm{initial}})`$ we only need to assume that the corresponding cosmological constant is positive:
$$\lambda _{\mathrm{bare}}+\frac{1}{2}\underset{i=1}{\overset{J}{}}n_{i,\mathrm{initial}}^2q_i^2>0.$$
(3.1)
This condition is generic, in the sense that it excludes only a finite number of configurations. In particular, if the unification scale is high and $`J100`$, the charges $`q_i`$ can be large ($`10^1`$) and the inequality will be satisfied with $`n_{i,\mathrm{initial}}1`$. If the unification scale is lower, the initial fluxes must be greater. But in such models large fluxes are often needed in any case to stabilize the internal dimensions.
It will be convenient to assume the slightly stronger initial condition that $`n_{i,\mathrm{initial}}n_{i,\mathrm{obs}}`$ for all $`i`$, where $`(n_{1,\mathrm{obs}},\mathrm{},n_{J,\mathrm{obs}})`$ denotes some configuration that lies in the observational window. This will permit us to neglect the strongly gravitational instantons responsible for increasing the $`n_i`$.
On the initial de Sitter background, $`J`$ different types of membranes can be nucleated through appropriate BT instantons. Inside a membrane of the $`i`$’th type, the flux $`n_i`$ is lowered by 1, and the cosmological constant is lowered by $`(n_i1/2)q_i^2`$. Although membranes expand at the speed of light, they typically never collide , because they are embedded in de Sitter space and cannot catch up with its expansion. Thus the ambient de Sitter space perdures eternally, harboring all types of membranes for which $`n_i>0`$. The same applies iteratively to the de Sitter regions with lower cosmological constant within each bubble. Thus, all combinations ($`n_1,\mathrm{},n_J`$) with $`n_in_{i,\mathrm{initial}}`$ and $`\lambda >0`$ are attained, including those with $`\lambda `$ in the observational range.
In the grid picture, Fig. 1, the initial configuration corresponds to a grid point some distance outside the sphere, in the $`n_i>0`$ quadrant. When a membrane of the $`i`$’th type is nucleated, the configuration in its interior corresponds to the neighboring grid point in the negative $`n_i`$ direction. Nested membranes correspond to a random walk in the grid. Since each membrane bubble harbors all other types of membranes (at least as long as $`\lambda >0`$), all such paths through the grid are realized in the universe. Overall, the membrane dynamics corresponds to diffusion through the grid. Every point is populated via many different paths.
### 3.2 The empty universe problem
#### 3.2.1 Inflation and reheating
The BT process involves spontaneous membrane nucleation in a prolonged de Sitter phase. One would expect this to lead to an empty universe. Particles are produced when a slow-roll field reaches a minimum of its potential and starts to oscillate. By this process, known as reheating, slow-roll inflation avoids the empty universe problem of old inflation. Because membrane nucleation is highly suppressed, it takes an exponentially long time to attain a suitable flux configuration. All fields will reach their vacua long before. Particles may be produced, but they will be wiped away by the remaining phase of the de Sitter expansion. In the end, it appears, we may have achieved too much of nothing: a (nearly) vanishing cosmological constant, but also vanishing entropy.
We will now discuss how this problem may be resolved. If any slow-roll field exists at all, one can argue that the problem does not occur in multiple four-form models with unification at the GUT scale or above, because the high temperature of de Sitter space before the final membrane nucleation kicks the inflaton out of its minimum. Moreover, if the inflaton potential contains a false vacuum, inflation and reheating can also occur in multiple four-form models with low unification scale.
With an inflaton field included, the effective cosmological constant is given by
$$\lambda _{\mathrm{eff}}(\varphi )=\lambda _{\mathrm{bare}}+\frac{1}{2}\underset{i=1}{\overset{J}{}}n_i^2q_i^2+V(\varphi ).$$
(3.2)
We take the inflaton potential $`V(\varphi )`$ to have a stable minimum at $`\varphi =0`$. By absorption into $`\lambda _{\mathrm{bare}}`$ we can arrange $`V(0)=0`$. The criterion for a suitable configuration $`(n_1,\mathrm{},n_J)`$ is that the cosmological constant be small for $`\varphi =0`$: $`\lambda _{\mathrm{eff}}(0)=\lambda 0`$. For $`\varphi 0`$ one obtains a positive effective cosmological constant, $`\lambda _{\mathrm{eff}}(\varphi )=V(\varphi )`$, at least temporarily.
Slow-roll inflation with $`\lambda =0`$ is described as follows. An inflaton field $`\varphi `$ rolls down in a potential $`V(\varphi )`$. During this time, the universe expands exponentially, like de Sitter space with an effective cosmological constant $`\lambda _{\mathrm{eff}}(\varphi )=V(\varphi )`$. Quantum fluctuations during this era freeze when they leave the horizon, forming seeds for density perturbations. When $`\varphi `$ reaches the bottom of the potential, it oscillates, inflation ends, and the universe is reheated.
The inflaton potential must be very flat. We will not address the difficult problem of how such potentials may arise from a fundamental theory; we note merely that they must exist if inflation is the correct explanation for homogeneity and density perturbations. For inflation to last long enough, one must require that the initial value of the inflaton field, $`\varphi _0`$, is sufficiently far from the minimum of $`V(\varphi )`$:
$$\varphi _0\varphi _{},$$
(3.3)
where $`\varphi _{}`$ corresponds to sixty $`e`$-foldings of de Sitter-like expansion. This is realized, for example, if the inflaton field is initially trapped in a false vacuum far enough from the true minimum at $`\varphi =0`$. More generically, suitable domains will exist if one assumes chaotic initial conditions in the early universe . In the scenario we have described, however, one must worry that any inflaton field will reach its minimum when $`\lambda `$ is still large, because membrane creation takes an exponentially long time. Consequently, $`\varphi `$ would no longer be available to perturb and reheat the universe when the flux configuration corresponding to $`\lambda 0`$ is reached.
#### 3.2.2 Kicking the inflaton
The evolution of $`\varphi `$ is actually a combination of classical slow-roll and Brownian motion . The latter can be understood as a random walk induced by the Gibbons-Hawking temperature of de Sitter space:
$$T(\varphi )=\frac{H(\varphi )}{2\pi },$$
(3.4)
where the Hubble parameter is given by<sup>7</sup><sup>7</sup>7We use the ‘reduced’ Planck mass, $`M_{\mathrm{Pl}}=(8\pi G_\mathrm{N})^{1/2}=\kappa _4^1=2.4310^{18}`$ GeV.
$$H(\varphi )^2=\frac{\lambda _{\mathrm{eff}}(\varphi )}{3M_{\mathrm{Pl}}^2}.$$
(3.5)
The characteristic time scale in de Sitter space is the Hubble time, $`\mathrm{\Delta }t=H^1`$. A typical quantum fluctuation of the field $`\varphi `$, during the time $`\mathrm{\Delta }t`$, is given by
$$|\delta \varphi |=\sqrt{2}T(\varphi )=\frac{H(\varphi )}{\sqrt{2}\pi }.$$
(3.6)
The classical decrease $`|\mathrm{\Delta }\varphi |`$ of the inflaton field can be estimated from the restoring force, $`V^{}(\varphi )`$:
$$|\mathrm{\Delta }\varphi |\frac{1}{2}V^{}(\varphi )(\mathrm{\Delta }t)^2=\frac{V^{}(\varphi )}{2H(\varphi )^2}.$$
(3.7)
A prime denotes differentiation with respect to $`\varphi `$. We neglect velocity effects because they are small during slow-roll and average to zero in Brownian motion. The random walk dominates over classical evolution if $`|\delta \varphi |>|\mathrm{\Delta }\varphi |`$, or
$$\sqrt{\frac{2}{3}}\frac{1}{3\pi }\lambda _{\mathrm{eff}}(\varphi )^{3/2}M_{\mathrm{Pl}}^3>V^{}(\varphi ).$$
(3.8)
We are interested in the temperature of the universe just before a final membrane is nucleated. Consider a flux configuration $`(n_1,\mathrm{},n_{j1},n_j+1,n_{j+1},\mathrm{},n_J)`$, where $`(n_1,\mathrm{},n_{j1},n_j,n_{j+1},\mathrm{},n_J)`$ corresponds to a cosmological constant in the observational window. If the charges $`q_i`$ are large, the penultimate cosmological constant,
$$\lambda _{\mathrm{pen}}=\left(n_j\frac{1}{2}\right)q_j^2,$$
(3.9)
will be large. Since $`\lambda _{\mathrm{eff},\mathrm{pen}}(\varphi )\lambda _{\mathrm{pen}}>0`$, Eq. (3.8) will be satisfied for a range of values of $`\varphi `$ including $`\varphi =0`$. Therefore the inflaton will take random values within a (finite or infinite) neighborhood of $`\varphi =0`$.
When the final membrane is nucleated, the temperature in its interior suddenly vanishes. Then the inflaton no longer experiences significant Brownian motion and rolls to its minimum. The question is whether this period of inflation is sufficient. (Less ambitiously, one could ask only whether $`\varphi `$ will be large enough to reheat the universe.) One would like the width of the random distribution of $`\varphi `$ to be a few times larger than $`\varphi _{}`$, the value required for 60 $`e`$-foldings. Then it will be likely that $`\varphi \varphi _{}`$ at the time of the final membrane nucleation.
From Eqs. (2.31), (2.39), and (2.40) one obtains
$$q_j^28\pi ^2\left(\frac{M_{11}}{M_{\mathrm{Pl}}}\right)^6M_{\mathrm{Pl}}^4.$$
(3.10)
The inequality, Eq. (3.8), will be satisfied if
$$(2n_j1)^{3/2}\sqrt{\frac{2}{3}}\frac{8\pi ^2}{3}\left(\frac{M_{11}}{M_{\mathrm{Pl}}}\right)^9M_{\mathrm{Pl}}^3>V^{}(\varphi ).$$
(3.11)
One may take as examples the polynomial potentials $`V(\varphi )=10^{12}M_{\mathrm{Pl}}^2\varphi ^2/2`$ and $`V(\varphi )=10^{14}\varphi ^4/4`$, and note that $`\varphi _{}15M_{\mathrm{Pl}}`$ in both cases. Then Eq. (3.8) must be satisfied with $`V^{}10^{10}M_{\mathrm{Pl}}^3`$ for sufficient inflation ($`\varphi >\varphi _{}`$) to be likely. The condition becomes
$$\frac{M_{11}}{M_{\mathrm{Pl}}}(2n_j1)^{1/6}10^{1.3}.$$
(3.12)
Thus the universe will undergo a normal period of inflation if the unification scale is $`10^{17}`$ GeV or higher.
The $`n_j`$-dependent factor does not contribute much since $`J`$ is large and the flux numbers will be of order one. Because of the $`(M_{11}/M_{\mathrm{Pl}})^9`$ suppression in Eq. (3.11), this mechanism rapidly becomes inefficient for lower unification scale.
#### 3.2.3 Trapping the inflaton
There is an alternative approach to the empty universe problem. It is less generic, but has the advantage that it can work in models with low unification scale, $`M_{11}1`$ TeV. Assume that the potential of the inflaton field (or of any other field with suitable coupling to the inflaton) contains a false vacuum. During the long de Sitter era before the $`\lambda 0`$ flux configuration is attained, this vacuum will of course decay by Coleman-De Lucia tunnelling . As we discussed earlier, Guth and Weinberg have shown that bubbles do not percolate in de Sitter space . Because the bubbles do not all collide, we need not fear that the entire universe will be converted to the true vacuum. In the ambient metastable space, the BT mechanism proceeds as before. Eventually, it produces regions where $`\lambda 0`$ while $`\lambda _{\mathrm{eff}}`$ is given by the energy density of the false vacuum. Only then will we be interested in the decay of the false vacuum. After tunnelling, the field emerges on the other side of the barrier. For a wide class of potentials, the field configuration at this point will still be far from the true vacuum. The fields can then roll to the minimum, thus inflating and reheating the universe.
The understanding of the cosmology of models with large internal dimensions is still being developed. This makes a detailed implementation of the generalized BT mechanism difficult. We should caution that there are important constraints on inflationary models that operate after radius stabilization .
In this subsection it has been assumed that the effective potential $`V(\varphi )`$ is the same before and after the final membrane nucleation. However, there can be important corrections from the high temperature before the final transition . They will typically make the potential steeper but also shift its minima. Thus, after the final nucleation, the inflaton field will not be in a local minimum and can roll down. Moreover, one would expect coupling constants of the effective field theory to depend on the fluxes. This also contributes to the flux-dependence of the effective potential and generically to a shift of its minima when a membrane is nucleated.<sup>8</sup><sup>8</sup>8We thank L. Susskind and S. Thomas for pointing this out to us.
### 3.3 Vacuum selection
In the BT scenario the universe generically develops a large number of different, exponentially large regions with every value of the cosmological constant in the discretuum, including large values. Why are we located in one of the regions with a small cosmological constant?
Most regions will not contain structure such as galaxies. Observers are necessarily located where structure does form, which restricts us to regions in the Weinberg window ,
$$10^{120}M_{\mathrm{Pl}}^4<\lambda <10^{118}M_{\mathrm{Pl}}^4.$$
(3.13)
The upper bound is about 100 times larger than the observed $`\lambda `$. It is obtained by demanding that the cosmological constant must not dominate the evolution of the universe before a redshift of about 4, so that gravitational clustering operates long enough for galaxies to form. The lower bound follows because the universe must not recollapse while stars and galaxies form. Its magnitude is comparable to the observed cosmological constant, but it has opposite sign. Much work has been devoted to strengthening these constraints by more careful astrophysical and statistical arguments (see, e.g., and references therein).
Such considerations may be distasteful to some but should not be viewed as an easy fix. They cannot be applied unless the fundamental theory satisfies a number of rather non-trivial conditions: it must admit different values of the cosmological constant; they must contain at least one value in the observational range; and there must be a dynamical mechanism that allows some regions to attain such a value. The aim of this paper has been to present evidence that all of these conditions may be satisfied in compactified 11D supergravity.
### 3.4 Stability
In order for our picture to be satisfactory we need the rate of bubble nucleation from a phase with small cosmological constant to be small on the scale of the age of our universe. The tunnelling amplitude is proportional to $`e^B`$, where $`B`$ is the normalized action of the corresponding instanton . A sufficient condition is $`B10^3`$.
We consider a single membrane, which changes the flux $`j`$ from $`n_j`$ to $`n_j1`$. The domain wall tension is $`\tau _j`$, given in Eq. (2.32), and the change in the cosmological constant is
$$\delta \lambda =\left(n_j\frac{1}{2}\right)q_j^2=2M_{\mathrm{Pl}}^2\left(n_j\frac{1}{2}\right)\tau _j^2.$$
(3.14)
For $`n_j1`$, gravity has negligible effects and the action is given by
$$B=\frac{27\pi ^2}{2\left(n_j\frac{1}{2}\right)^3(2M_{\mathrm{Pl}}^2q_j)^2}.$$
(3.15)
To estimate the $`n_i`$ we assume approximate equipartition of the energy among the fluxes so that
$$\frac{n_i^2q_i^2}{2}\frac{2\pi M_{11}^4}{J}.$$
(3.16)
For the nonzero fluxes in section 2.6, $`\tau _i=2\pi M_{11}^3`$, and $`q_i^2=8\pi ^2(M_{11}/M_{\mathrm{Pl}})^2M_{11}^4`$. We obtain
$$B\frac{27\pi ^{3/2}J^{3/2}}{16\sqrt{2}(M_{11}/M_{\mathrm{Pl}})^3}.$$
(3.17)
In the large dimension case $`B`$ is of order $`10^{46}`$ and so the tunneling is negligible. Even for the Witten GUT scenario, where $`J100`$, it is of order $`10^8`$ and again tunneling is negligible.
At higher unification scales, for which $`M_{11}/M_{\mathrm{Pl}}>10^{1.5}`$, one finds $`J>100`$. Then Eq. (3.16) yields $`n_i<1`$ and thus breaks down. Almost all relevant configurations will have $`n_i\{0,1\}`$ for all $`i\{1,\mathrm{},J\}`$. We can therefore assume $`n_j=1`$. In this case the additional suppression due to gravity is significant, although it is never total for our parameters. One finds
$$B=\frac{1728\pi ^2}{(2M_{\mathrm{Pl}}^2q_j)^2}=54(M_{11}/M_{\mathrm{Pl}})^6.$$
(3.18)
Tunnelling will be negligible for $`M_{11}/M_{\mathrm{Pl}}<0.6`$. Therefore vacuum stability is not a significant constraint on our mechanism. A stronger constraint on $`M_{11}/M_{\mathrm{Pl}}`$ is obtained from Eq. (2.41) by requiring a realistic number of three-cycles, say $`J<10^3`$.
## 4 Conclusions
Compactifications of M-theory generally give rise to multiple four-form field strengths. We showed that such theories have vacua with discrete but closely spaced values for the cosmological constant. In the Witten GUT scenario, the spectrum will contain values of $`\lambda `$ in the observable range if the number of four-forms is of order 100. (This requires that the cosmological constant to be cancelled is of GUT scale, not weak scale). In models with large internal dimensions, four or five four-forms suffice, and a weak-scale cosmological constant can be cancelled. By repeated membrane nucleation, flux configurations with $`\lambda 0`$ arise dynamically from generic initial conditions. We argued that entropy and density perturbations can be generated in such regions, and showed that the amplitude for the decay of the $`\lambda 0`$ vacuum is negligible.
An attractive feature of this proposal is that it simultaneously addresses two questions that are usually treated as distinct. The first question is: Why is the cosmological constant not huge? One would expect a vacuum density $`\lambda `$ of order $`M_{\mathrm{Pl}}^4`$, or at least TeV<sup>4</sup> with supersymmetry. Until recently this was the only cosmological constant problem. It appeared to require a symmetry ensuring the exact cancellation of all contributions to the cosmological constant. This is difficult because contributions are expected to come from many different scales. The second question is: Why is the cosmological constant not zero? Recent evidence<sup>9</sup><sup>9</sup>9A review of these observations can be found in Ref. . points to a flat universe with $`\mathrm{\Omega }_\mathrm{m}0.3`$ and $`\mathrm{\Omega }_\lambda 0.7`$. The favored value for the vacuum energy is $`\lambda 10^{120}M_{\mathrm{Pl}}^4(0.003\text{ eV})^4`$. In particular, a flat universe with vanishing vacuum energy has been ruled out. But if it is difficult to explain $`\lambda =0`$, a small non-zero cosmological constant seems to pose an even greater theoretical challenge. The mechanism we propose has limited accuracy because of flux quantization, so that a residual cosmological constant is inevitable.
Our proposal has certain features of the Brown-Teitelboim idea, and also certain features of eternal inflation . Previously, however, both of these ideas have been difficult to realize with a plausible microphysics. Our proposal allows both to be realized within string theory. For the Brown-Teitelboim idea, the main problem was the very small energy scale needed in the discretuum; we see that this can be obtained from a normal hierarchy with multiple fluxes. Eternal inflation with generic polynomial potentials requires scalar field expectation values strictly larger than the Planck scale. In string theory the scale of the field manifold is the string scale, which is no larger than the Planck scale. The manifold is actually noncompact, but the asymptotic regions generally correspond to decompactification of spacetime, and in this region the effective potential generally ceases to be flat. We have realized a version of eternal inflation that does not require such a large scalar, and uses elements already present in string theory.<sup>10</sup><sup>10</sup>10A precursor to the idea of four-form-driven eternal inflation was presented in Ref. . Moreover, if the membrane charges are large, the high temperature of de Sitter space before the final membrane nucleation induces Brownian motion of the inflaton field, thus preparing suitable initial conditions for chaotic inflation after the transition.
The main problem with realizing our picture is the stabilization of the compact dimensions, which is of course a ubiquitous problem in string theory. A positive bulk cosmological constant is a useful ingredient , but it is not clear that this can be realized in string theory.
It is interesting that the naked singularity proposal appears to lead in the end to a very similar picture. The free parameters that correspond to boundary conditions at a naked singularity in a compact space will become, in a four-dimensional effective Lagrangian, variable coupling constants. In the original proposal these were assumed to be continuous and constant in time, but in Ref. it was argued that they are discrete and can change across a domain wall, just as for the fluxes considered here. In the example there was a potentially large number of states, of order $`e^\sqrt{N}`$ where $`N`$ is at Ramond-Ramond charge of the singularity. Note, however, that a charge of order $`10^5`$ is needed to produce a discretuum sufficiently dense to account for the smallness of the cosmological constant. In Ref. the main focus was on supersymmetric states, which were all degenerate, but with supersymmetry breaking there will again be a spectrum for $`\lambda `$. Again, stabilization will be an issue.
The appearance of the anthropic principle, even in the weak form encountered here, is not entirely pleasant, but we would argue that it is necessary in any approach where the cosmological constant is a dynamical variable. That is, a small value for the present cosmological constant cannot be obtained by dynamical considerations alone. The point is that we can follow cosmology at least back to nucleosynthesis, when the present cosmological constant contributed only a fraction $`10^{30}`$ to the energy density of the universe, and so was dynamically irrelevant. At earlier times, including the point where the cosmological constant is to have been determined, the fraction would have been even smaller.<sup>11</sup><sup>11</sup>11One exception is the wormhole idea , where the value of the cosmological constant in our universe is determined by the presence of other, empty, universes. At least one of the authors retains a certain wary fondness for this possibility.
## Acknowledgments
We would like to thank N. Arkani-Hamed, S. Dimopoulos, S. Kachru, A. Linde, J. March-Russell, J. Rahmfeld, L. Susskind, S. Thomas, N. Toumbas, and E. Witten for useful discussions. We would also like to thank J. Feng, J. March-Russell, S. Sethi, and F. Wilczek for informing us about their forthcoming work. While completing this work we learned that J. Donoghue was also pursuing the idea of anthropic determination of the cosmological constant with nondynamical four-forms . The work of J.P. was supported by National Science Foundation grants PHY94-07194 and PHY97-22022. The work of R.B. was supported by a BASF fellowship of the German National Scholarship Foundation.
|
warning/0004/hep-ph0004045.html
|
ar5iv
|
text
|
# Enhanced antiproton production in Pb(160 AGeV)+Pb reactions: evidence for quark gluon matter?
## Acknowledgements
The authors would like to thank L. Gerland for fruitful discussions. M. Bleicher is supported by the A. v. Humboldt Foundation. S.A.B. acknowledges financial support from the U.S. National Science Foundation, grant PHY-9605207. M.Belkacem was supported by the U.S. Department of Energy under grant No. DE-FG02-87ER40328. This work is supported by the BMBF, GSI, DFG and the Graduiertenkolleg ’Theoretische und experimentelle Schwerionenphysik’. This research used resources of the National Energy Research Scientific Computing Center (NERSC).
|
warning/0004/hep-ph0004133.html
|
ar5iv
|
text
|
# Untitled Document
|
warning/0004/hep-th0004096.html
|
ar5iv
|
text
|
# hep-th/0004096 Bianchi Type I Cosmologies in Arbitrary Dimensional Dilaton Gravities
## I Introduction
In an attempt to address the potential, inherited from string theory, to eliminate the initial cosmological singularity, from which time and our Universe are supposed to have begun about 15 billion years ago, Gasperini and Veneziano initiated a program known as the pre-big bang scenario . The field equations of the pre-big bang cosmology are based on the low energy effective action resulting from string theory. In $`D`$-dimensions, the massless bosonic fields from the NS-NS sector are the dilaton, $`\varphi `$, the antisymmetric tensor, $`B_{\mu \nu }`$, and the metric tensor, $`\widehat{g}_{\mu \nu }`$, whose dynamics is described, in the “string frame”, by the following action:
$$\widehat{S}=d^Dx\sqrt{\widehat{g}}e^{2\varphi }\left\{\widehat{R}+\widehat{\kappa }(\widehat{}\varphi )^2\frac{1}{12}H_{[3]}^2\widehat{U}(\varphi )\right\},$$
(1)
where $`H_{[3]}=dB_{[2]}`$ and $`\widehat{\kappa }`$ is a generalized dilaton coupling constant ($`\widehat{\kappa }=4`$ for superstring theories). Moreover, we also allow for the existence of a potential $`\widehat{U}(\varphi )`$ of the dilaton field. From a physical point of view the most important candidate for the potential is a cosmological constant $`\mathrm{\Lambda }`$, which appears in the massive extension of type IIA supergravity and is restricted to be positive, $`\mathrm{\Lambda }>0`$ .
For simplicity in the following we shall assume that $`H_{[3]}`$ is vanishing. In this circumstance, via a conformal rescaling
$$g_{\mu \nu }=e^{\frac{4}{D2}\varphi }\widehat{g}_{\mu \nu },$$
(2)
the action (1) reduces to a $`D`$-dimensional dilaton gravity whose action, in the “Einstein frame”, has the form
$$S=d^Dx\sqrt{g}\left\{R\kappa (\varphi )^2U(\varphi )\right\},$$
(3)
with $`U(\varphi )=e^{\frac{4\varphi }{D2}}\widehat{U}(\varphi )`$ and $`\kappa =\frac{4(D1)}{D2}\widehat{\kappa }`$.
Pre-big bang inflationary cosmological models, based on the actions (1) or (3) have been recently intensively investigated in the physical literature . Gasperini and Ricci have obtained exact solutions to the four-dimensional low energy string effective action adopting a space-independent dilaton and vanishing Kalb-Ramond anti-symmetric tensor fields ansätz for the Bianchi type I, II, III, V, VI<sub>0</sub> and VI<sub>h</sub> geometries. They have shown that in such a context the initial curvature singularities can not be avoided. Brandenberger, Easther and Maia have found non-singular spatially homogeneous and isotropic solutions for dilaton gravity in the presence of a special combination of higher derivative terms in the gravitational action. Some of these solutions correspond to a spatially flat, bouncing Universe originating in a dilaton-dominated contracting phase and emerging as an expanding FRW Universe.
Very recently, the string cosmology equations with a dilaton potential have been examined, in the string frame, by Ellis et al. , who also give a generic algorithm for obtaining solutions with desired evolutionary properties. The presence of a dilaton potential leads to the violation of the pre-big bang symmetry $`a(t)1/a(t)`$. Moreover, Garcia de Andrade obtained several classes of solutions of the Einstein-Cartan dilatonic inflationary cosmology. In the cases where the dilatons are constrained by the presence of spin-torsion effects a repulsive gravity is found. The temperature fluctuation has also been computed from the nearly flat spectrum of the gravitational waves produced during inflation, with results agreeing with the COBE data.
Pre-big bang cosmological models, in which there is no need to introduce the inflation or to fine-tune potentials, have many attractive features . Inflation is natural, thanks to the duality symmetries of string cosmology, and the initial condition problem is decoupled from the singularity problem. Finally, quantum instability (pair creation) is able to heat up an initially cold Universe and generate a standard hot big bang with the additional features of homogeneity, flatness and isotropy.
It is the purpose of the present paper to study Bianchi type I cosmological models in the dilaton gravity (1) and (3). More specifically, we shall consider the effects of an exponential type potential, $`U(\varphi )=U_0e^{\lambda \varphi }`$, with arbitrary values of the constants $`U_0,\lambda `$, on the dynamics and evolution of an anisotropic space-time, in both the Einstein and string frames. In this case the general solution of the gravitational field equations can be expressed in an exact parametric form. The physical effects of the potential on the evolution of the anisotropic space-time are also considered in detail.
The present paper is organized as follows. The basic equations describing the dilatonic Bianchi type I cosmological model are obtained in Section 2. The general solution of the field equations for an exponential type dilaton potential is obtained in Section 3 (Einstein frame) and in Section 4 (string frame). In Section 5 we discuss our results and conclusions.
## II Einstein Frame Field Equations, Geometry and Consequences
In this paper, we shall consider the $`D`$-dimensional anisotropic generalization of the flat FRW geometry — the Bianchi type I space-time described by the line-element
$$ds^2=dt^2+\underset{i=1}{\overset{D1}{}}a_i^2(t)(dx^i)^2.$$
(4)
For this metric, it is convenient to introduce the following variables: volume scale factor $`V`$, directional Hubble factors $`H_i`$ and mean Hubble factor $`H`$ as
$`V`$ $`:=`$ $`{\displaystyle \underset{i=1}{\overset{D1}{}}}a_i,`$ (5)
$`H_i`$ $`:=`$ $`{\displaystyle \frac{\dot{a}_i}{a_i}},i=1,\mathrm{},D1,`$ (6)
$`H`$ $`:=`$ $`{\displaystyle \frac{1}{D1}}{\displaystyle \underset{i=1}{\overset{D1}{}}}H_i,`$ (7)
$`\mathrm{\Delta }H_i`$ $`:=`$ $`H_iH,i=1,\mathrm{},D1.`$ (8)
Then one can immediately check out the relation
$$H=\frac{1}{D1}\frac{\dot{V}}{V}.$$
(9)
In terms of variables (5)-(8) the Ricci tensor of the Bianchi type I geometry can be expressed as
$`R_{00}`$ $`=`$ $`{\displaystyle \frac{d}{dt}}{\displaystyle \underset{i=1}{\overset{D1}{}}}\left({\displaystyle \frac{\dot{a}_i}{a_i}}\right){\displaystyle \underset{i=1}{\overset{D1}{}}}\left({\displaystyle \frac{\dot{a}_i}{a_i}}\right)^2`$ (10)
$`=`$ $`(D1)\dot{H}{\displaystyle \underset{i=1}{\overset{D1}{}}}H_i^2,`$ (11)
$`R_{ii}`$ $`=`$ $`a_i^2\left[\dot{H}_i+(D1)HH_i\right],i=1,\mathrm{},D1.`$ (12)
On the other hand, the field equations of the action (3) can be achieved by variation with respect to the fields $`g^{\mu \nu }`$ and $`\varphi `$ giving
$`R_{\mu \nu }\kappa _\mu \varphi _\nu \varphi {\displaystyle \frac{U}{D2}}g_{\mu \nu }`$ $`=`$ $`0,`$ (13)
$`^2\varphi {\displaystyle \frac{1}{2\kappa }}{\displaystyle \frac{U}{\varphi }}`$ $`=`$ $`0,`$ (14)
where $``$ is the covariant derivative of $`g_{\mu \nu }`$. Thus, for the Bianchi type I space-time, the gravitational field equations in the Einstein frame reduce to
$$(D1)\dot{H}+\underset{i=1}{\overset{D1}{}}H_i^2+\kappa \dot{\varphi }^2\frac{1}{D2}U=0,$$
(15)
$`{\displaystyle \frac{1}{V}}{\displaystyle \frac{d}{dt}}(VH_i){\displaystyle \frac{1}{D2}}U`$ $`=`$ $`0,i=1,..,D1,`$ (16)
$`{\displaystyle \frac{1}{V}}{\displaystyle \frac{d}{dt}}(V\dot{\varphi })+{\displaystyle \frac{1}{2\kappa }}{\displaystyle \frac{U}{\varphi }}`$ $`=`$ $`0.`$ (17)
By summing equations (16) we obtain
$$\frac{1}{V}\frac{d}{dt}(VH)=\frac{1}{D2}U,$$
(18)
which, together with (16), leads to
$$H_i=H+\frac{K_i}{V},i=1,\mathrm{},D1.$$
(19)
In equations (19) $`K_i,i=1,\mathrm{},D1`$ are constants of integration, which satisfy the relation:
$$\underset{i=1}{\overset{D1}{}}K_i=0.$$
(20)
Substituting eqs.(19) into (15) and then combining with eq.(18) we obtain
$$\kappa \dot{\varphi }^2+(D2)\dot{H}+\frac{K^2}{V^2}=0,$$
(21)
where $`K^2:=_{i=1}^{D1}K_i^2`$. Consequently, the remaining task is to solve the equations (17), (18) and (21).
The physical quantities of interest in cosmology are the expansion scalar $`\theta `$, the mean anisotropy parameter $`A`$, the shear scalar $`\mathrm{\Sigma }^2`$ and the deceleration parameter $`q`$ defined according to:
$`\theta `$ $`:=`$ $`(D1)H={\displaystyle \frac{\dot{V}}{V}},`$ (22)
$`A`$ $`:=`$ $`{\displaystyle \frac{1}{D1}}{\displaystyle \underset{i=1}{\overset{D1}{}}}\left({\displaystyle \frac{\mathrm{\Delta }H_i}{H}}\right)^2={\displaystyle \frac{1}{D1}}{\displaystyle \frac{K^2}{V^2H^2}},`$ (23)
$`\mathrm{\Sigma }^2`$ $`:=`$ $`{\displaystyle \frac{1}{D2}}\left({\displaystyle \underset{i=1}{\overset{D1}{}}}H_i^2(D1)H^2\right)`$ (24)
$`=`$ $`{\displaystyle \frac{D1}{D2}}AH^2,`$ (25)
$`q`$ $`:=`$ $`{\displaystyle \frac{d}{dt}}H^11.`$ (26)
The sign of the deceleration parameter indicates whether the cosmological model inflates. The positive sign corresponds to standard decelerating models whereas the negative sign indicates inflationary behavior.
## III Exponential Potential in the Einstein Frame
The cosmological behavior of Universes filled with scalar field, $`\varphi `$, as well as a Liouville type exponential potential
$$U(\varphi )=U_0e^{\lambda \varphi },$$
(27)
with $`U_0`$ and $`\lambda `$ constants, has been extensively investigated in the physical literature for both homogeneous and inhomogeneous scalar fields -. An exponential potential arises in the four-dimensional effective Kaluza-Klein type theories from compactification of the higher-dimensional supergravity or superstring theories . A solution in the case of a flat space-time filled with a scalar field with an exponential potential but describing power-law inflationary behavior has been obtained by Barrow . Higher dimensional ($`D4`$) anisotropic cosmological models with a massless scalar field self-interacting through an exponential potential have been investigated in . A non-inflationary solution for an open FRW Universe exponential-potential pure scalar field filled space-time and with scalar field energy density decaying as $`\rho _\varphi t^2`$ has been recently found by Mubarak and Oezer . In the Einstein frame the exponential potential (27) is also generated by means of the conformal transformation (2) for $`\widehat{U}(\varphi )=\mathrm{\Lambda }`$, with $`\mathrm{\Lambda }`$ the central charge deficit.
For this type of potential, the combination of equations (16) and (17) leads to
$$\frac{d}{dt}\left(\frac{1}{D2}V\dot{\varphi }+\frac{\lambda }{2\kappa }VH\right)=0,$$
(28)
or, equivalently, to
$$\dot{\varphi }=\frac{(D2)C}{V}\frac{(D2)\lambda }{2\kappa }H,$$
(29)
with $`C`$ a constant of integration.
Substitution of Eq.(29) into Eq.(21) gives the “final” field equation
$$\frac{\ddot{V}}{V}+\alpha \frac{\dot{V}^2}{V^2}\beta \frac{\dot{V}}{V^2}+\gamma \frac{1}{V^2}=0,$$
(30)
where
$`\alpha `$ $`=`$ $`{\displaystyle \frac{(D2)\lambda ^2}{4(D1)\kappa }}1,`$ (31)
$`\beta `$ $`=`$ $`(D2)C\lambda ,`$ (32)
$`\gamma `$ $`=`$ $`(D1)(D2)C^2\kappa +{\displaystyle \frac{(D1)K^2}{D2}}.`$ (33)
By introducing a new variable $`u:=\dot{V}`$, equation (30) takes the form
$$\frac{udu}{\alpha u^2+\beta u\gamma }=\frac{dV}{V}.$$
(34)
Equation (34) has the general solution (with $`V_0`$ a constant of integration):
$$V=V_0\mathrm{exp}\left(\frac{udu}{\alpha u^2+\beta u\gamma }\right).$$
(35)
In the following we shall denote
$`\mathrm{\Delta }`$ $`=`$ $`\beta ^24\alpha \gamma ,`$ (36)
$`u_0`$ $`=`$ $`{\displaystyle \frac{\beta }{2\alpha }},`$ (37)
$`u_\pm `$ $`=`$ $`{\displaystyle \frac{\beta \pm \sqrt{\mathrm{\Delta }}}{2\alpha }},`$ (38)
$`m_\pm `$ $`=`$ $`{\displaystyle \frac{1}{2\alpha }}\left(1\pm {\displaystyle \frac{\beta }{\sqrt{\mathrm{\Delta }}}}\right).`$ (39)
Hence, taking $`u`$ as a parameter, we obtain three classes of solutions of the gravitational field equations describing a dilaton field filled Bianchi type I pre-big bang Universe. The explicit form of the solutions depends on the values of the parameters $`\alpha ,\beta `$ and $`\gamma `$. All the solutions are expressed in a closed parametric form and are given by:
### A $`\mathrm{\Delta }>0`$
$`t`$ $`=`$ $`t_0+V_0{\displaystyle (uu_+)^{m_+1}(uu_{})^{m_{}1}𝑑u},`$ (40)
$`V`$ $`=`$ $`V_0(uu_+)^{m_+}(uu_{})^m_{},`$ (41)
$`a_i`$ $`=`$ $`a_{i0}{\displaystyle \underset{ϵ=\pm }{}}u^{\frac{K_im_ϵ}{u_ϵ}}(uu_ϵ)^{(\frac{1}{D1}+\frac{K_i}{u_ϵ})m_ϵ},`$ (42)
$`q`$ $`=`$ $`(D2)+{\displaystyle \frac{(D1)\alpha }{u^2}}{\displaystyle \underset{ϵ=\pm }{}}(uu_ϵ),`$ (43)
$`U`$ $`=`$ $`{\displaystyle \frac{(D2)\alpha }{(D1)V_0^2}}{\displaystyle \underset{ϵ=\pm }{}}(uu_ϵ)^{12m_ϵ}.`$ (44)
### B $`\mathrm{\Delta }=0`$
$`t`$ $`=`$ $`t_0{\displaystyle \frac{V_0}{\alpha }}{\displaystyle (uu_0)^{\frac{1}{\alpha }2}\mathrm{exp}\left(\frac{u_0}{\alpha (uu_0)}\right)𝑑u},`$ (45)
$`V`$ $`=`$ $`V_0(uu_0)^{\frac{1}{\alpha }}\mathrm{exp}\left({\displaystyle \frac{u_0}{\alpha (uu_0)}}\right),`$ (46)
$`a_i`$ $`=`$ $`a_{i0}(uu_0)^{\frac{1}{(D1)\alpha }}\mathrm{exp}\left({\displaystyle \frac{u_0+K_i(D1)}{\alpha (D1)(uu_0)}}\right),`$ (47)
$`q`$ $`=`$ $`(D2)+{\displaystyle \frac{(D1)\alpha (uu_0)^2}{u^2}},`$ (48)
$`U`$ $`=`$ $`{\displaystyle \frac{(D2)\alpha }{(D1)V_0^2}}(uu_0)^{\frac{2}{\alpha }+2}\mathrm{exp}\left({\displaystyle \frac{2u_0}{\alpha (uu_0)}}\right).`$ (49)
### C $`\mathrm{\Delta }<0`$
$`t`$ $`=`$ $`t_0+V_0{\displaystyle (\alpha u^2+\beta u\gamma )^{\frac{1}{2\alpha }1}}`$ (51)
$`\times \mathrm{exp}\left({\displaystyle \frac{\beta }{\alpha \sqrt{\mathrm{\Delta }}}}\mathrm{arctan}{\displaystyle \frac{2\alpha u\beta }{\sqrt{\mathrm{\Delta }}}}\right)du,`$
$`V`$ $`=`$ $`V_0(\alpha u^2+\beta u\gamma )^{\frac{1}{2\alpha }}`$ (53)
$`\times \mathrm{exp}\left({\displaystyle \frac{\beta }{\alpha \sqrt{\mathrm{\Delta }}}}\mathrm{arctan}{\displaystyle \frac{2\alpha u\beta }{\sqrt{\mathrm{\Delta }}}}\right),`$
$`a_i`$ $`=`$ $`a_{i0}(\alpha u^2+\beta u\gamma )^{\frac{1}{2\alpha (D1)}}`$ (55)
$`\times \mathrm{exp}\left({\displaystyle \frac{\beta +2\alpha K_i(D1)}{\alpha (D1)\sqrt{\mathrm{\Delta }}}}\mathrm{arctan}{\displaystyle \frac{2\alpha u\beta }{\sqrt{\mathrm{\Delta }}}}\right),`$
$`q`$ $`=`$ $`(D2)(D1){\displaystyle \frac{\alpha u^2+\beta u\gamma }{u^2}},`$ (56)
$`U`$ $`=`$ $`{\displaystyle \frac{D2}{(D1)V_0^2}}(\alpha u^2+\beta u\gamma )^{\frac{1}{\alpha }+1}`$ (58)
$`\times \mathrm{exp}\left({\displaystyle \frac{2\beta }{\alpha \sqrt{\mathrm{\Delta }}}}\mathrm{arctan}{\displaystyle \frac{2\alpha u\beta }{\sqrt{\mathrm{\Delta }}}}\right).`$
For all three cases, the quantities $`\theta ,A`$ and $`\mathrm{\Sigma }^2`$ can be easily found from
$$\theta =\frac{u}{V},A=\frac{(D1)K^2}{u^2},\mathrm{\Sigma }^2=\frac{K^2}{(D2)V^2}.$$
(59)
## IV Exponential Potential in the String Frame
In the string frame the gravitational field equations and the dilaton equations are obtained by varying the action (1) and, under the assumption of vanishing $`H_{[3]}`$, are given by
$`\widehat{R}_{\mu \nu }{\displaystyle \frac{1}{2}}\widehat{g}_{\mu \nu }\widehat{R}+2\widehat{}_\mu \widehat{}_\nu \varphi +(\widehat{\kappa }4)\widehat{}_\mu \varphi \widehat{}_\nu \varphi `$ (60)
$`{\displaystyle \frac{1}{2}}\widehat{g}_{\mu \nu }\left\{4\widehat{}^2\varphi +(\widehat{\kappa }8)(\widehat{}\varphi )^2\widehat{U}\right\}`$ $`=`$ $`0,`$ (61)
$`\widehat{R}+\widehat{\kappa }\widehat{}^2\varphi \widehat{\kappa }(\widehat{}\varphi )^2\widehat{U}(\varphi )+{\displaystyle \frac{1}{2}}{\displaystyle \frac{\widehat{U}}{\varphi }}`$ $`=`$ $`0.`$ (62)
By eliminating $`\widehat{R}`$ between equations (61) and (62), the gravitational field and dilaton equations take the form
$`\widehat{R}_{\mu \nu }+2\widehat{}_\mu \widehat{}_\nu \varphi +(\widehat{\kappa }4)\widehat{}_\mu \varphi \widehat{}_\nu \varphi `$ (63)
$`{\displaystyle \frac{\widehat{g}_{\mu \nu }}{2}}\left\{(4\widehat{\kappa })\widehat{}^2\varphi +2(\widehat{\kappa }4)(\widehat{}\varphi )^2{\displaystyle \frac{1}{2}}{\displaystyle \frac{\widehat{U}}{\varphi }}\right\}`$ $`=`$ $`0,`$ (64)
$`\widehat{}^2\varphi 2(\widehat{}\varphi )^2+{\displaystyle \frac{4\widehat{U}+(D2)\frac{\widehat{U}}{\varphi }}{2[(D2)\widehat{\kappa }4(D1)]}}`$ $`=`$ $`0.`$ (65)
In the present section we shall consider the general solution of equations (61) and (62) for an exponential type potential, $`\widehat{U}(\varphi )=\widehat{U}_0\mathrm{exp}(\widehat{\lambda }\varphi )`$, with $`\widehat{\lambda }`$ an arbitrary constant. Since the metric tensors are connected via the conformal transformation (2), in the string frame the general solutions of the gravitational field equations can be obtained by applying the conformal transformation (2) to the solution obtained in the Einstein frame. In the string frame we shall also assume an anisotropic Bianchi type I geometry with line element
$$d\widehat{s}^2=d\widehat{t}^2+\underset{i=1}{\overset{D1}{}}\widehat{a}_i^2(\widehat{t})(dx^i)^2,$$
(66)
with the metric tensor components in the two frames connected by the conformal transformation (2) and with the time coordinate $`\widehat{t}`$ defined according to
$$\widehat{t}=\mathrm{exp}\left[\frac{2}{D2}\varphi (t)\right]𝑑t.$$
(67)
In the two frames the volume scale factor, the directional Hubble factors and the mean Hubble factor are related by means of the general relations:
$`\widehat{V}`$ $`=`$ $`Ve^{\frac{2(D1)}{D2}\varphi },`$ (68)
$`\widehat{H}_i`$ $`=`$ $`\left(H_i+{\displaystyle \frac{2}{D2}}\dot{\varphi }\right)e^{\frac{2}{D2}\varphi },i=1,\mathrm{},D1,`$ (69)
$`\widehat{H}`$ $`=`$ $`\left(H+{\displaystyle \frac{2}{D2}}\dot{\varphi }\right)e^{\frac{2}{D2}\varphi }.`$ (70)
To apply the conformal transformation, we need first to find the conformal transformation factor $`e^\varphi `$. From equation (18) it is easy to obtain that the potential $`U(\varphi )`$ can be expressed as
$$U(\varphi )=\frac{D2}{D1}\frac{u}{V^2}\left(\frac{d\mathrm{ln}V}{du}\right)^1,$$
(71)
leading to
$$e^\varphi =\left[\frac{D2}{(D1)U_0}\right]^{\frac{1}{\lambda }}\left(\frac{V^2}{u}\frac{d\mathrm{ln}V}{du}\right)^{\frac{1}{\lambda }}.$$
(72)
Therefore in the string frame the general solution of the gravitational field equation for a dilaton field filled Bianchi type I with an exponential potential of the form
$$\widehat{U}(\varphi )=U_0\mathrm{exp}\left[\left(\lambda \frac{4}{D2}\right)\varphi \right],$$
(73)
with $`\lambda `$ an arbitrary constant, can be expressed again in an exact closed parametric form, with $`u`$ taken as parameter, and is given by
$`\widehat{t}\widehat{t}_0`$ $`=`$ $`\left[{\displaystyle \frac{D2}{(D1)U_0}}\right]^{\frac{2}{(D2)\lambda }}`$ (75)
$`\times {\displaystyle }\left({\displaystyle \frac{V^2}{u}}{\displaystyle \frac{d\mathrm{ln}V}{du}}\right)^{1\frac{2}{(D2)\lambda }}{\displaystyle \frac{du}{V}},`$
$$\widehat{V}=V\left[\frac{D2}{(D1)U_0}\right]^{\frac{2(D1)}{(D2)\lambda }}\left(\frac{V^2}{u}\frac{d\mathrm{ln}V}{du}\right)^{\frac{2(D1)}{(D2)\lambda }},$$
(76)
$`\widehat{H}`$ $`=`$ $`\left[{\displaystyle \frac{D2}{(D1)U_0}}\right]^{\frac{2}{(D2)\lambda }}\left({\displaystyle \frac{V^2}{u}}{\displaystyle \frac{d\mathrm{ln}V}{du}}\right)^{\frac{2}{(D2)\lambda }}`$ (78)
$`\times \left[{\displaystyle \frac{2C}{V}}+{\displaystyle \frac{\kappa \lambda }{(D1)\kappa }}{\displaystyle \frac{u}{V}}\right],`$
$`\widehat{a}_i`$ $`=`$ $`a_{i0}\left[{\displaystyle \frac{D2}{(D1)U_0}}\right]^{\frac{2}{(D2)\lambda }}\left({\displaystyle \frac{V^2}{u}}{\displaystyle \frac{d\mathrm{ln}V}{du}}\right)^{\frac{2}{(D2)\lambda }}V^{\frac{1}{D1}}`$ (80)
$`\times \mathrm{exp}\left(K_i{\displaystyle }{\displaystyle \frac{1}{u}}{\displaystyle \frac{d\mathrm{ln}V}{du}}du\right),i=1,\mathrm{},D1,`$
$$\widehat{A}=(D1)\underset{i=1}{\overset{D1}{}}\left[\frac{\kappa K_i}{(\kappa \lambda )u+2\kappa C(D1)}\right]^2,$$
(81)
$$\widehat{q}=(D2)\frac{u}{D1}\left(\frac{d\mathrm{ln}V}{du}\right)^1\left[2C+\frac{\kappa \lambda }{(D1)\kappa }u\right]^2,$$
(82)
$$\widehat{U}=U_0\left[\frac{D2}{(D1)U_0}\right]^{1\frac{4}{(D2)\lambda }}\left(\frac{V^2}{u}\frac{d\mathrm{ln}V}{du}\right)^{\frac{4}{(D2)\lambda }1}.$$
(83)
In the string frame there are also three distinct classes of solutions, corresponding to $`\mathrm{\Delta }>0,\mathrm{\Delta }=0`$ and $`\mathrm{\Delta }<0`$ respectively. Substituting the values of $`V`$ obtained in the previous section in the formulae given above, we can find, via straightforward calculations, the explicit parametric representations, for each class of solutions, of the general solution of the gravitational field equations for a dilaton field filled Bianchi type I space-time, with an arbitrary exponential potential.
If in the solution given above we take $`\lambda =\frac{4}{D2}`$, we obtain the general solution of the gravitational field equations in the string frame corresponding to a constant potential, or equivalently, to a cosmological constant. In this case also there are three distinct classes of solutions, with all physical quantities represented as exact functions of time. For $`\widehat{U}(\varphi )\mathrm{\Lambda }=const.`$, Eq.(75) becomes
$$\widehat{t}\widehat{t}_0=\sqrt{\frac{D2}{(D1)\mathrm{\Lambda }}}\frac{du}{\sqrt{\alpha u^2+\beta u\gamma }}.$$
(84)
In order to obtain solutions defined for all values of the parameters we shall assume in the following that $`\alpha <0`$. Then Eq.(84) has the solutions
$`u`$ $`=`$ $`{\displaystyle \frac{\beta }{2|\alpha |}}+\delta _+\mathrm{cosh}{\displaystyle \frac{\widehat{t}\widehat{t}_0}{\widehat{\tau }_0}},\text{for}\mathrm{\Delta }>0,`$ (85)
$`u`$ $`=`$ $`{\displaystyle \frac{\beta }{2|\alpha |}}+\mathrm{exp}\left({\displaystyle \frac{\widehat{t}\widehat{t}_0}{\widehat{\tau }_0}}\right),\text{for}\mathrm{\Delta }=0,`$ (86)
$`u`$ $`=`$ $`{\displaystyle \frac{\beta }{2|\alpha |}}+\delta _{}\mathrm{sinh}{\displaystyle \frac{\widehat{t}\widehat{t}_0}{\widehat{\tau }_0}},\text{for}\mathrm{\Delta }<0,`$ (87)
where we denoted $`\delta _\pm :=\frac{\sqrt{\pm \mathrm{\Delta }}}{2|\alpha |}`$ and $`\widehat{\tau }_0:=\sqrt{\frac{D2}{(D1)\mathrm{\Lambda }|\alpha |}}`$.
In this way we can obtain the exact (non-parametric) solution for the anisotropic Bianchi type I geometry in the presence of a central charge deficit. We shall not present here the resulting formulae, due to their complicated (but elementary) mathematical form. As compared to the Einstein frame, the evolution of the Universe in the string frame in the presence of the cosmological constant can be quite complicated.
## V Discussions and Final Remarks
In order to consider the general effects of a dilaton field potential in the Einstein frame on the dynamics and evolution of an arbitrary dimensional Bianchi type I space-time, we shall also give the general solution of the gravitational field equations (15)-(17) corresponding to $`U(\varphi )0`$. In this case we easily obtain:
$`V`$ $`=`$ $`V_0t,`$ (88)
$`H`$ $`=`$ $`{\displaystyle \frac{1}{(D1)t}},`$ (89)
$`a_i`$ $`=`$ $`a_{i0}t^{p_i},i=1,\mathrm{},D1,`$ (90)
$`A`$ $`=`$ $`{\displaystyle \frac{(D1)K^2}{V_0^2}}=const.,`$ (91)
$`q`$ $`=`$ $`D2=const.,`$ (92)
$`\varphi `$ $`=`$ $`\varphi _0\mathrm{ln}t,`$ (93)
where $`\varphi _0=\varphi _0^{}\sqrt{\frac{1}{\kappa }\left(\frac{D2}{D1}\frac{K^2}{V_0^2}\right)}`$, with $`\varphi _0^{}`$ a constant of integration. The coefficients $`p_i:=\frac{1}{D1}+\frac{K_i}{V_0}`$ satisfy the relations $`_{i=1}^{D1}p_i=1`$ and $`_{i=1}^{D1}p_i^2=\frac{1}{D1}+\frac{K^2}{V_0^2}`$. Hence in the Einstein frame the geometry of the potential free dilaton field is of Kasner type, but with $`_{i=1}^{D1}p_i^21`$ (if we adopt the normalization $`_{i=1}^{D1}p_i^2=1`$ then we obtain the empty Bianchi type I Universe with $`\varphi 0`$). The anisotropic Bianchi type I dilaton field filled Universe does not isotropize (the mean anisotropy parameter is a constant for all times) and its evolution is non-inflationary with $`q>0`$ for all $`t`$.
In order to analyze the general effects of the dilaton field potential on the dilaton field filled Bianchi type I space-time in the Einstein frame, we shall obtain first the following anisotropy equation:
$$\frac{dA}{dt}=\frac{2A}{H}\left(\dot{H}+(D1)H^2\right),$$
(94)
which can also be written in the equivalent form
$$\frac{dA}{dt}=\frac{2AU(\varphi )}{(D2)H},$$
(95)
and integrated to give
$$A(t)=A_0\mathrm{exp}\left(\frac{2}{D2}_{t_0}^t\frac{U(\varphi )}{H}𝑑t\right).$$
(96)
In equation (96) we denoted by $`A_0`$ an arbitrary constant of integration. For $`U(\varphi )0`$ we always have $`AA_0=const.`$ If $`_{t_0}^t\frac{U(\varphi )}{H}𝑑t`$ is a monotonically increasing positive function of time then the presence of the dilaton field potential will lead to the fast isotropization of the Bianchi type I space-time.
In the presence of a potential the deceleration parameter $`q=(\dot{H}+H^2)/H^2`$ can be expressed as
$$q=(D2)\frac{U(\varphi )}{(D2)H^2}.$$
(97)
If $`U(\varphi )0`$ in the Einstein frame the evolution of the Universe is non-inflationary, but once the condition $`U(\varphi )>(D2)^2H^2`$ is fulfilled, the dynamics of the Bianchi type I space-times becomes inflationary.
In the present paper we have obtained the general solution of the gravitational field equations for a Bianchi type I space-time filled with a dilaton field with an exponential potential in both the Einstein and string frame. In the Einstein frame they describe generically an expanding Universe, with $`u=\dot{V}0`$ and with properties strongly dependent on the numerical values of the physical parameters describing the dilaton field and its potential. A contracting Universe with $`u=\dot{V}<0`$ generally does not satisfy the condition of reality of the scale factors. The solutions of the field equations can be classified into three classes, according to the sign of the quantity $`\mathrm{\Delta }`$. On the other hand for solutions B and C with $`\mathrm{\Delta }=0`$ and $`\mathrm{\Delta }<0`$ respectively, the condition $`\alpha <0`$ must also be imposed to ensure the positivity of the potential and well-defined physical quantities for all time. In the limit of large $`u`$, $`u\mathrm{}`$, all three solutions have a similar behavior. The mean anisotropy $`A`$ tends in all cases to zero, indicating that an exponential type potential leads to the isotropization of the Universe. In the large $`u`$ limit the deceleration parameter behaves as $`q=(D2)+\alpha (D1)`$. If the condition $`\alpha <\frac{D2}{D1}`$, or, equivalently, $`\lambda ^2<\frac{4\kappa }{D2}`$ is fulfilled, the Universe will enter in an inflationary phase. For values of $`\alpha `$ which do not satisfy this condition the evolution of the space-time will be generally non-inflationary. In the same limit of large $`u`$ the scalar field is given by $`\varphi \frac{\alpha +1}{\alpha }\mathrm{ln}u`$.
For class A solutions, the Bianchi type I dilaton field filled Universe starts in the Einstein frame from a singular state, corresponding to the values $`u=u_+`$ or $`u=u_{}`$ of the parameter. Hence for this model a singular state with zero values of the scale factors is unavoidable. But for class B of solutions the evolution of the Universe is non-singular for $`u_0<0`$. In this case the scale factors are finite for all finite values of the parameter $`u`$. Alternatively, class C models are non-singular for values of the constants $`\alpha `$ and $`\gamma `$ such that $`\alpha <0`$ and $`\gamma <0`$.
In Figs. 1-4 we have represented the variations in the Einstein frame of the volume scale factor, mean anisotropy, deceleration parameter and dilaton field for a four-dimensional ($`D=4`$) Bianchi type I space-time. The anisotropic Universe will always end in an isotropic state, but its dynamics can be either inflationary or non-inflationary. Generally the dilaton field $`\varphi `$ is a decreasing function of time.
We shall consider now the effects of the dilaton field and potential on the dynamics and evolution of a Bianchi type I space-time in the string frame. In the case in which there is no dilaton field potential, $`\widehat{U}(\varphi )0`$, the general solution of the gravitational field equations and of the dilaton equation can be obtained again by the conformal transformation (2) from (89-93). Hence in this case we obtain first the relation connecting the time coordinate in the string and Einstein frames in the form
$$t=\left(\frac{\widehat{t}}{\widehat{n}}\right)^{\widehat{n}},$$
(98)
where
$$\widehat{n}=\frac{D2}{D2+2\varphi _0},$$
(99)
In the string frame the general solution of the potential free dilaton field filled anisotropic Universe is given by
$`\widehat{V}`$ $`=`$ $`\widehat{V}_0\widehat{t}^{\widehat{h}},`$ (100)
$`\widehat{H}`$ $`=`$ $`{\displaystyle \frac{\widehat{h}}{(D1)\widehat{t}}},`$ (101)
$`\widehat{a}_i`$ $`=`$ $`\widehat{a}_{i0}\widehat{t}^{\widehat{p}_i},i=1,\mathrm{},D1,`$ (102)
and
$`\widehat{A}`$ $`:=`$ $`{\displaystyle \frac{1}{D1}}{\displaystyle \underset{i=1}{\overset{D1}{}}}\left({\displaystyle \frac{\mathrm{\Delta }\widehat{H}_i}{\widehat{H}}}\right)^2`$ (103)
$`=`$ $`{\displaystyle \frac{1}{D1}}{\displaystyle \underset{i=1}{\overset{D1}{}}}\left[1{\displaystyle \frac{(D1)\widehat{p}_i}{\widehat{h}}}\right]^2,`$ (104)
$`\widehat{q}`$ $`=`$ $`{\displaystyle \frac{D1}{\widehat{h}}}1,`$ (105)
where $`\widehat{V}_0`$ and $`\widehat{a}_{i0}`$ are arbitrary constants of integration. Here we also denoted
$`\widehat{h}`$ $`=`$ $`{\displaystyle \frac{D2+2(D1)\varphi _0}{D2+2\varphi _0}},`$ (106)
$`\widehat{p}_i`$ $`=`$ $`\widehat{n}\left(p_i+{\displaystyle \frac{2\varphi _0}{D2}}\right),i=1,\mathrm{},D1,`$ (107)
and the coefficients $`\widehat{p}_i`$ satisfy the relations
$`{\displaystyle \underset{i=1}{\overset{D1}{}}}\widehat{p}_i`$ $`=`$ $`\widehat{n}\left[1+{\displaystyle \frac{2(D1)}{D2}}\varphi _0\right],`$ (108)
$`{\displaystyle \underset{i=1}{\overset{D1}{}}}\widehat{p}_i^2`$ $`=`$ $`\widehat{n}^2\left[{\displaystyle \frac{1}{D1}}+{\displaystyle \frac{K^2}{V_0^2}}+{\displaystyle \frac{4\varphi _0}{D2}}\left(1+{\displaystyle \frac{D1}{D2}}\varphi _0\right)\right].`$ (109)
In the string frame the general physical behavior of the potential free dilatonic Bianchi type I Universe is quite similar to that in the Einstein frame. The geometry is of the Kasner type, with a power-law type time dependence of the scale factors. The mean anisotropy of the space-time is constant and the Universe will never isotropize. On the other hand if the condition $`\widehat{h}>D1`$ is fulfilled, the Universe experiences an eternal power law type inflationary anisotropic phase. Hence in the string frame a dilaton field filled Bianchi type I Universe provides an example of an inflating but never isotropizing cosmological type evolution.
In the string frame and in the presence of an exponential potential $`\widehat{U}(\varphi )=\widehat{U}_0\mathrm{exp}\left(\widehat{\lambda }\varphi \right)`$, with $`\widehat{\lambda }=\lambda \frac{4}{D2}`$ and $`\lambda `$ an arbitrary constant, the Bianchi type I Universe shows a very large variety of behaviors. In Figs.5-8 we represented the dynamics of the volume scale factor, anisotropy parameter, deceleration parameter and potential for different values of $`\lambda `$ (different exponential potential functions) but for fixed $`\alpha ,\beta `$ and $`\gamma `$. These solutions generically begin from a singular state, followed by an expansionary phase, with the volume scale factor and scale factors reaching a local maximum. Then the Universe re-collapse into a new singular phase. This type of evolution is associated with an initial rapid isotropization of the space-time, with the mean anisotropy parameter $`\widehat{A}`$ rapidly decreasing. Near the second singular state the evolution of the Universe is generally inflationary, with the string frame deceleration parameter $`\widehat{q}`$ smaller than zero, $`\widehat{q}<0`$. After this phase the effect of the dilaton becomes irrelevant to the dynamics of space-time.
In the limit of large values of the parameter $`u`$, the term $`\alpha u^2`$ dominates, $`\alpha u^2>>\beta u\gamma `$. Hence in the limit of large $`u`$ (and large time, $`\widehat{t}\mathrm{}`$, too), from equation (35) we obtain $`VV_0u^{\frac{1}{\alpha }}`$. Therefore from Eq.(75) it follows that $`\widehat{t}u^{(1+1/\alpha )(4/(D2)\lambda 1)}u^{1/\widehat{l}}`$, and, consequently,
$`\widehat{V}`$ $``$ $`\widehat{t}^{(D1)\widehat{l}^{}},`$ (110)
$`\widehat{H}`$ $``$ $`\widehat{t}^1,`$ (111)
$`\widehat{a}_i`$ $``$ $`\widehat{t}^{\widehat{l}^{}}\mathrm{exp}\left({\displaystyle \frac{V_0K_i}{\alpha }}\widehat{t}^{\widehat{l}}\right),i=1,\mathrm{},D1,`$ (112)
$`\widehat{A}`$ $``$ $`\widehat{t}^{2\widehat{l}},`$ (113)
$`\widehat{q}`$ $``$ $`(D2)+{\displaystyle \frac{(D1)\alpha \kappa ^2}{V_0(\kappa \lambda )^2}},`$ (114)
$`\widehat{U}`$ $``$ $`\widehat{t}^2,`$ (115)
where we denoted
$`\widehat{l}`$ $`=`$ $`{\displaystyle \frac{1}{\left(1+\frac{1}{\alpha }\right)\left[\frac{4}{(D2)\lambda }1\right]}},`$ (116)
$`\widehat{l}^{}`$ $`=`$ $`1+\widehat{l}\left[1+{\displaystyle \frac{D2}{(D1)\alpha }}\right].`$ (117)
In the long-time limit the behavior of the exponential potential dilaton field filled Universe is quite different to the behavior of the potential free dilatonic anisotropic Universe. The dependence of the coefficients $`\widehat{l},\widehat{l}^{}`$ on the two constants $`\alpha `$ and $`\lambda `$ leads to a larger spectrum of admissible final states, with isotropic inflationary or non-inflationary evolution or re-collapse into a singular state. For $`|\alpha |<1`$ generally $`\widehat{l}<0`$, and, if $`\widehat{l}<\left(1+\frac{D2}{(D1)\alpha }\right)^1`$, then the volume scale factor tends to zero in the string frame, $`\widehat{V}0`$.
It is well known that the action (1) with vanishing antisymmetric field strength $`H_{[3]}`$ is invariant with respect to scale factor duality transformations of the form $`G\overline{G}=G^1`$ and $`\varphi \overline{\varphi }\mathrm{ln}(detG)`$, where $`G`$ is a matrix build from the metric tensor components of the FRW, anisotropic or inhomogeneous metric . The inclusion of the potential breaks this duality, but leads, on the other hand, to the possibility of obtaining more general models allowing a better physical description of the very early evolution of our Universe.
## Acknowledgments
One of the authors (TH) would like to thank Dr. P. Blaga for useful suggestions. CMC thanks Prof. J.M. Nester for profitable discussions. The work of CMC was supported in part by the National Science Council (Taiwan) under grant NSC 89-2112-M-008-016.
|
warning/0004/hep-th0004175.html
|
ar5iv
|
text
|
# 1. Harmonic oscillator in Fock space
## 1. Harmonic oscillator in Fock space
The Hamiltonian of harmonic oscillator is defined by
(1)
$$=\frac{1}{2}\frac{^2}{x^2}+\frac{\omega ^2}{2}x^2,$$
where $`\omega `$ is the oscillator frequency. The configuration space of (1) is the whole real line, $`x(\mathrm{},\mathrm{})`$. The eigenfunctions and eigenvalues are given by
(2)
$$\mathrm{\Psi }_k(x)=H_k(\sqrt{\omega }x)e^{\omega \frac{x^2}{2}},E_k=\omega (k+1/2),k=0,1,\mathrm{}$$
where $`H_k(y)`$ is the $`k`$th Hermite polynomial in standard notation. Without a loss of generality we put all normalization constants in (2) equal to 1. The Hamiltonian (1) is $`Z_2`$-invariant, $`xx`$, which leads to two families of eigenstates: even and odd, or, in other words, symmetric and anti-symmetric with respect of reflection, correspondingly. This property is coded in parity of the Hermite polynomials and is revealed by the relation:
(3)
$$H_{2n+p}(\sqrt{\omega }x)=\omega ^{\frac{p}{2}}x^pL_n^{(p\frac{1}{2})}(\omega x^2),n=0,1,\mathrm{}$$
where $`L_n^{(\alpha )}(y)`$ is the $`n`$th associated Laguerre polynomial in standard notation, and the parameter $`p=0,1`$ determines parity, $`P=(1)^p`$. Hereafter we can call
(4)
$$\mathrm{\Psi }_0^{(p)}(x)=x^pe^{\omega \frac{x^2}{2}},$$
the ground state (the lowest energy state) of parity $`P`$. Thus, the formula (4) makes an unification of both possible values of parity and for the sake of simplicity we will call (4) the ground state eigenfunction without specifying parity.
Let us make a gauge rotation of the Hamiltonian (1) taking as a gauge factor the ground state eigenfunction (4) and changing the variable $`x`$ to $`y=\omega x^2`$, which incorporate the reflection symmetry. After dropping out non-essential constant term, $`(\frac{1}{2}+p)`$, which cause a change the reference point for energy only, we get an operator
(5)
$$h(y,_y)=\frac{1}{\omega }(\mathrm{\Psi }_0^{(p)}(x))^1\mathrm{\Psi }_0^{(p)}(x)_{y=\omega x^2}=2y_y^2+2(yp\frac{1}{2})_y$$
with the spectrum $`E_n=2n`$, where $`n=0,1,2\mathrm{}`$ <sup>2</sup><sup>2</sup>2It is necessary to emphasize that the spectrum is defined by action of the operator $`2y_y`$ in (5) on monomial: $`(2y_y)y^n=2ny^n`$. It is the only operator of zero-grading in (5). Now the eiegenvalue problem for the operator (5) is defined on the half-line $`y[0,\mathrm{})`$. The operator (5) simultaneously describes a family of eigenstates of positive parity if $`p=0`$ and a family of eigenstates of negative parity if $`p=1`$. We will call (5) the algebraic form of the Hamiltonian of the harmonic oscillator. The word ‘algebraic’ reflects the fact that the operator (5) has a form of linear differential operator with polynomial coefficients and furthermore possesses infinitely-many polynomial eigenfunctions. The latter implies that any eigenfunction can be found by algebraic means by solving a system of linear algebraic equations.
The algebraic form (5) admits a generalization of the original Hamiltonian (1) we started with. If we assume that the parameter $`p`$ can take any real value, $`p>1/2`$, one can make an inverse gauge transformation of the operator (5) back to the Hamiltonian form and we arrive at
$$\omega y^{p/2}e^{y/2}\left[2y_y^2+2(yp\frac{1}{2})_y\right]y^{p/2}e^{y/2}_{x=\sqrt{\frac{y}{\omega }}}$$
(6)
$$=\left[\frac{1}{2}_x^2+\frac{\omega ^2}{2}x^2+\frac{p(p1)}{2x^2}\right]_k,$$
which is known in literature as Kratzer Hamiltonian. It is worth to mention that this Hamiltonian coincides with 2-body Calogero Hamiltonian. Also this Hamiltonian appears as a Hamiltonian of the radial motion of multidimensional spherical-symmetrical harmonic oscillator. Hereafter we will continue to call the system characterized by the Hamiltonian (6) the harmonic oscillator.
The resulting Hamiltonian (6) is characterized by the eigenfunctions
(7)
$$\mathrm{\Psi }_k(x)=x^pL_n^{(p\frac{1}{2})}(\omega x^2)e^{\omega \frac{x^2}{2}},$$
which coincides with (2) at $`p=0,1`$. The spectrum (6) is still equidistant with energy gap $`\omega `$ and after appropriate shift of the reference point it coincides with the spectrum of the original harmonic oscillator (1). Thus, the deformation of (1) to (6) is isospectral which is, of course, well-known.
In order to move ahead let us introduce a notion of the Fock space. It will be a natural formalism to study canonical transformations. Take two operators $`a`$ and $`b`$ obeying the commutation relation
(8)
$$[a,b]abba=I,$$
with the identity operator $`I`$ on the r.h.s. – they span a three-dimensional Lie algebra which is called the Heisenberg algebra $`h_3`$. By definition the universal enveloping algebra of $`h_3`$ is the algebra of all normal-ordered polynomials in $`a,b`$: any monomial is taken to be of the form $`b^ka^m`$ <sup>3</sup><sup>3</sup>3Sometimes this is called the Heisenberg-Weyl algebra. If, besides the polynomials, all entire functions in $`a,b`$ are considered, then the extended universal enveloping algebra of the Heisenberg algebra appears or in other words, the extended Heisenberg-Weyl algebra. In the (extended) Heisenberg-Weyl algebra one can find the non-trivial embedding of the Heisenberg algebra: non-trivial elements obeying the commutation relations (8), whose can be treated as a certain type of quantum canonical transformations. We say that the (extended) Fock space, $``$ is determined if we take the (extended) universal enveloping algebra of the Heisenberg algebra and attach to it the vacuum state $`|0>`$ defined as
(9)
$$a|0>=0.$$
It is easy to check that the following statement holds. If the operators $`a,b`$ obey the commutation relation (8), then the operators
$$J_n^+=b^2anb,$$
(10)
$$J_n^0=ba\frac{n}{2},$$
$$J_n^{}=a,$$
span the $`sl_2`$-algebra with the commutation relations:
$$[J^0,J^\pm ]=\pm J^\pm ,[J^+,J^{}]=2J^0,$$
where $`n𝐂`$ <sup>4</sup><sup>4</sup>4For details and discussion see, for example, . For the realization (10) the quadratic Casimir operator is equal to
(11)
$$C_2\frac{1}{2}\{J_n^+,J_n^{}\}J_n^0J_n^0=\frac{n}{2}\left(\frac{n}{2}+1\right),$$
where $`\{,\}`$ denotes the anticommutator and is $`c`$number. If $`n_+`$, then (10) possesses a finite-dimensional, irreducible representation in Fock space leaving invariant the linear space of polynomials in $`b`$ acting on vacuum:
(12)
$$𝒫_n(b)=1,b,b^2,\mathrm{},b^n|0,$$
of dimension $`dim𝒫_n=(n+1)`$. It is evident that any operator which is a polynomial in generators $`J_n^{+,0,}`$ preserves the space $`𝒫_n(b)`$ (and converse is also correct ). Such an operator we call $`sl_2`$-quasi-exactly-solvable operator. Thus, the most general $`sl_2`$-quasi-exactly-solvable operator in the Fock space having a form of a polynomial in $`a`$ of degree not higher than two is given by
$$T_2=c_{++}J_n^+J_n^++c_{+0}J_n^+J_n^0+c_{00}J_n^0J_n^0+c_0J_n^0J_n^{}+c_{}J_n^{}J_n^{}+$$
(13)
$$c_+J_n^++c_0J_n^0+c_{}J_n^{}+c,$$
where $`c_{\alpha \beta },c_\alpha ,c`$ are arbitrary c-numbers, or after the substitution (10) in explicit form
(14)
$$T_2(b,a)=P_4(b)a^2+P_3(b)a+P_2(b),$$
where the $`P_j(b)`$ are polynomials of $`j`$th order with coefficients related to $`c_{\alpha \beta },c_\alpha ,c`$ and $`n`$.
The spaces $`𝒫_n`$ possess a property that $`𝒫_n𝒫_{n+1}`$ for each $`n_+`$ and form an infinite flag (filtration) and
$$\underset{n_+}{}𝒫_n=𝒫.$$
Hence it is evident that any operator which is a polynomial in generators $`J_n^{0,}`$ only preserves the flag of $`𝒫`$ . Such an operator we call $`sl_2`$-exactly-solvable operator. Thus, the most general $`sl_2`$-exactly-solvable operator in the Fock space having a form of a polynomial in $`a`$ of degree not higher than two is given by
(15)
$$E_2=c_{00}J^0J^0+c_0J^0J^{}+c_{}J^{}J^{}+c_0J^0+c_{}J^{}+c,$$
where $`J^{\pm ,0}J_0^{\pm ,0}`$ and $`c_{\alpha \beta },c_\alpha ,c`$ are arbitrary c-numbers. After the substitution (10) in explicit form
(16)
$$E_2(b,a)=Q_2(b)a^2+Q_1(b)a+Q_0,$$
where the $`Q_j(b)`$ are polynomials of $`j`$th order with coefficients related to $`c_{\alpha \beta },c_\alpha ,c`$.
Now we can introduce a notion of the spectral problem in the Fock space. Let $`L(b,a)`$ is an element of the Heisenberg-Weyl algebra. By definition, to solve the spectral problem for the operator $`L(b,a)`$ is to find a set of the elements $`\{\varphi (b)\}`$ in the Heisenberg-Weyl algebra and a corresponding set of parameters $`\{\lambda \}`$ for those the equation
$$L(b,a)\varphi (b)|0>=\lambda \varphi (b)|0>()$$
is fulfilled. We will call $`\{\varphi (b)\}`$ and $`\{\lambda \}`$ the eigenfunctions and eigenvalues, respectively. Attempting to study the spectral problem $`()`$ immediately leads to a delicate question about convergence of the operator series. In order to avoid possible difficulties we will restrict our consideration by the cases when the polynomial in $`b`$ eigenfunctions appear.
## 2. Translation-covariant discretization
Take as an example the $`sl_2`$-exactly-solvable operator of the following form
(17)
$$h^{(1)}(b,a)=2J^0J^{}+2J^02(p+\frac{1}{2})J^{}=2ba^2+2(bp\frac{1}{2})a,$$
where $`p`$ is a parameter and $`J^{\pm ,0}J_0^{\pm ,0}`$ (see (10)). Simple analysis leads to a statement that the eigenfunctions of $`h_f`$ are the associated Laguerre polynomials of the argument $`b`$, $`L_n^{(p\frac{1}{2})}(b)`$ and their eigenvalues, $`E_n=2n,n=0,1,2,\mathrm{}`$ <sup>5</sup><sup>5</sup>5As in (5) the eigenvalues are defined by action of the operator $`(2ba)`$ in (13) on monomials: $`(2ba)b^n=2nb^n+2b^{n+1}a`$. The second term disappears after action on the vacuum.
As the next step we consider two different realization of the Heisenberg algebra (8) in terms of differential and finite-difference operators. A traditional realization of (8) appearing in text-books is the so-called coordinate-momentum representation:
(18)
$$a=\frac{d}{dy}_y,b=y,$$
(see, for example, the book by Landau and Lifschitz ), where the operator $`b=y`$ stands for the multiplication operator : $`bf(y)=yf(y)`$. In this case the vacuum is a constant and without a loss of generality we put $`|0>=1`$. However, there exists another realization of (8) in terms of finite-difference operators. It is a finite-difference analogue of (18):
(19)
$$a=𝒟_+,b=(y+\alpha )(1\delta 𝒟_{}),$$
where
$$𝒟_\pm f(y)=\frac{f(y\pm \delta )f(y)}{\pm \delta },$$
(cf.) is the finite-difference operator. It represents what can be called a $`\delta `$discretization of the derivative: $`𝒟_\pm _y`$, if $`\delta 0`$ and, thus, which can be called $`\delta `$derivative <sup>6</sup><sup>6</sup>6This finite-difference operator is also known in mathematics literature as the Nörlund derivative (see, for example,) while we prefer to use a name $`\delta `$-derivative in order to distinguish from $`q`$-derivative (see below) – another type of discretization. In general, $`\delta ,\alpha `$ can be any complex numbers and $`𝒟_\pm (\delta )=𝒟_{}(\delta )`$. It is necessary to emphasize that from physical point of view the operator $`b`$ in (19) is nothing but the canonical-conjugate to a discrete momentum operator defined by $`\delta `$derivative. So, (19) represents two-parametric family of quantum canonical transformations. For $`\alpha =0`$ and $`\delta 0`$, the formulas (19) become (18). It allows to interpret (19) as the continuous deformation of (18).
A remarkable property of this realization is that the vacuum can be taken as a constant <sup>7</sup><sup>7</sup>7Any $`\delta `$periodic function can be chosen as a vacuum for (19) as well and thus, without loss of generality, it can be placed as $`|0>=1`$ for both cases (18)-(19). Realization (19) is translation-covariant: under a linear shift of variable $`yy+B`$ the functional form of (19) is preserved.
Substitution of (18) into (17) leads to the operator (5) – the algebraic form of the Hamiltonian of the harmonic oscillator. Thus, the operator (17) can be called the the Hamiltonian of the harmonic oscillator in the Fock space. It is evident that the procedure of realization of the Heisenberg generators $`a,b`$ by concrete operators (differential, finite-difference, discrete) provided that the vacuum remains unchanged leaves any polynomial operator in $`a,b`$ isospectral.
Now let us study another ‘face’ of harmonic oscillator by substituting the realization of the Heisenberg algebra by finite-difference operators (19) into (17). It results to
(20)
$$h_\delta ^{(1)}(y,D_\pm )=\frac{2}{\delta }[y+\alpha +\delta (p+\frac{1}{2})]D_++2(1+\frac{1}{\delta })(y+\alpha )D_{}.$$
In this realization the corresponding spectral problem $`()`$ can be written as
$$\frac{2}{\delta ^2}\left[y+\alpha +\delta (p+\frac{1}{2})\right]\varphi (y+\delta )+\frac{2}{\delta }\left[(1+\frac{2}{\delta })y+\alpha +p+\frac{1}{2}\right]\varphi (y)$$
(21)
$$\frac{2}{\delta }(1+\frac{1}{\delta })(y+\alpha )\varphi (y\delta )=E\varphi (y).$$
It is worth to note that the Askey condition that the sum of the coefficients in front of unknown functions should be zero is fulfilled at $`E=0`$. It is equivalent to a statement that the equation (21) possess an eigenfunction which is a constant.
In general, the operator $`h_\delta (y,D_\pm )`$ is a non-local, three-point, finite-difference and translation-covariant operator. This operator is canonically-equivalent to the harmonic oscillator. Pictorially it is illustrated by Fig.1.
Fig. 1. Graphical representation of the operator (20)
So, in the $`y`$space we have the uniform grid, linear lattice. However, from the point of view of the original configuration space, $`x`$space where the harmonic oscillator (1) is defined we have a square lattice.
In order to find the eigenfunctions of (20) a certain trick can be used. One can easily show that the following equality holds
$$[(y+\alpha )e^{\delta _y}]^nI=(y+\alpha )^{(n)}I,$$
where $`(y+\alpha )^{(n+1)}=(y+\alpha )(y+\alpha \delta )\mathrm{}(y+\alpha n\delta )`$ is a so-called quasi-monomial or generalized monomial and $`I`$ is the identity operator. Then making use this relation it is not difficult to prove that the eigenfunctions of (20) remain polynomials in $`y`$ and, furthermore, they can be found explicitly in rather elegant way
(22)
$$\varphi _n(y)=\widehat{L}_n^{(p\frac{1}{2})}(y,\delta )=\underset{\mathrm{}=0}{\overset{n}{}}a_{\mathrm{}}^{(p\frac{1}{2})}(y+\alpha )^{(\mathrm{})},$$
where $`a_{\mathrm{}}^{(p\frac{1}{2})}`$ are the coefficients in the expansion of the Laguerre polynomials, $`L_n^{(p\frac{1}{2})}(y)=_{\mathrm{}=0}^na_{\mathrm{}}^{(p\frac{1}{2})}y^{\mathrm{}}`$. We call these polynomials the modified associated Laguerre polynomials. Simultaneously, the eigenvalues in the equation (21) remain equal to $`(2n),n=0,1,2\mathrm{}`$ and they are the same as the eigenvalues of the harmonic oscillator problem (1) as well as (6) and (13). Thus, one can say that the operator (20) defines a finite-difference or $`\delta `$discrete algebraic form of harmonic oscillator Hamiltonian. Without loss of generality one can place $`\alpha =0`$ in (20), (21), (22).
There exists two non-trivial particular cases of (20). The first case corresponds to the spacing $`\delta =1`$. It leads to a disappearance of the term proportional to $`D_{}`$ and, thus, the operator (20) becomes two-point operator (!)
$$h_\delta ^{(1)}(y,D_\pm )=2[y+\alpha (p+\frac{1}{2})]D_+|_{\delta =1}$$
(23)
$$2[y+\alpha (p+\frac{1}{2})]D_{}|_{\delta =1}.$$
It is worth noting that the spectrum in this case is defined by the term $`(2yD_+|_{\delta =1})`$ stemming from the negative grading term $`(2ba^2)`$ in (17). Breaking a condition of performing the canonical transformations we re-insert the parameter $`\delta `$ in (23) in naive, straightforward manner: $`D_+|_{\delta =1}D_+(\delta )`$. It results to the operator
(24)
$$h_\delta ^{(1)}(y,D_\pm )=2[y+\alpha (p+\frac{1}{2})]D_+,$$
which is isospectral to (5), (17) for any $`\delta `$. In limit $`\delta 0`$ it leads to the first-order differential operator
(25)
$$h_\delta ^{(1)}(y,_y)=2[y+\alpha (p+\frac{1}{2})]_y.$$
Although, the operators (23), (24) are isospectral to various forms of the harmonic oscillator (1), (6), those operators are not related to each other by a quantum canonical transformation and therefore are not canonically equivalent. Above-mentioned isospectral transition from the second-order differential operator (5) to the first order differential operator (25) or from finite-difference one (20) to (24) reminds celebrated Bargmann transformation (see for example ). It is necessary to emphasized that the operator (24) has infinitely-many polynomial eigenfunctions, however, unlike the operators (13),(16), the operator (20) is not $`sl_2`$-exactly-solvable (!). Also it can not be rewritten in terms of the generators $`a,b`$ of the Heisenberg algebra (8) and does not belong to the Fock space (see discussion below).
Another important particular case occurs if the spacing $`\delta =2`$. In this case the operator (20) again becomes two-point one
(26)
$$\frac{1}{2}\left[y+\alpha 2(p+\frac{1}{2})\right]\varphi (y2)+\frac{1}{2}(y+\alpha )\varphi (y+2)=\stackrel{~}{E}\varphi (y),$$
where a new spectral parameter $`\stackrel{~}{E}=E+\alpha +p+\frac{1}{2}`$.
> Remark. The function $`\varphi (y)`$ in the r.h.s. of (21) can be replaced by $`\varphi (y+\delta )`$ or $`\varphi (y\delta )`$, or by a linear combination of $`\varphi (y\pm \delta ),\varphi (y)`$. It does not change the statement that (21) has infinitely-many polynomial eigenfunctions. This procedure preserves isospectral property. However, such changes of the r.h.s. lead to a replacement of the original standard spectral problem $`()`$ by a ‘generalized’ spectral problem. In this case the r.h.s. contains an operator other than the operator of multiplication on a function. In general, a physical relevance of such right-hand-sides is unclear. It is worth to note that it reminds the Sturm representation of the Coulomb problem when the energy is kept fixed but a set of discrete electric charges for which a corresponding excited state energy is equal to this energy is studied. In this formulation the r.h.s. of the Coulomb problem contains the Coulombic interaction potential as the weight factor.
A natural question can be posed about the most general second-order linear differential operator, which (i) is isospectral to the harmonic oscillator (1),(6); (ii) has infinitely-many polynomial eigenfunctions and (iii) related to (5) by a canonical transformation. Following the Theorem one can derive the most general operator with above properties
$$h_{\delta ,g}(y,_y)=2(AJ^0+BJ^{})J^{}+2J^02CJ^{}=$$
(27)
$$2(Ay+B)_y^2+2(yC)_y$$
where $`A,B,C`$ are arbitrary constants and the generators (10) are realized by differential operators (18). However, by a linear change of variable, $`y\alpha y+\beta `$ with appropriate $`\alpha 0,\beta `$ the operator (27) can be transformed into (5). It reflects a fact that the linear space of polynomials $`𝒫_n(x)`$ is invariant under a linear transformation: $`x\gamma _1x+\gamma _2,\gamma _10`$. Thus, without loss of generality we can put $`A=\alpha =1`$ and also $`C=p+1/2`$. The eigenfunctions of (27) remain the Laguerre polynomials but of a shifted argument, $`L_n^{(p\frac{1}{2})}(y+\beta )`$. It leads to a statement that among the second-order differential operators there exists no non-trivial isospectral deformation of the harmonic oscillator potential preserving polynomiality of the eigenfunctions.
The operator (27) can be rewritten in the Fock space formalism by using (18)
(28)
$$h_g(b,a)=2(b+B)a^2+2(bC)a.$$
It is evident that the operator (28) is the most general second order polynomial in $`a`$, which is isospectral to (17) and also preserves the space of polynomials (12). By substitution (19) into the operator (28) it becomes a finite-difference operator
(29)
$$h_{\delta ,g}(y,D_\pm )=2BD_+^2\frac{2}{\delta }[y+\alpha +\delta C]D_++2(1+\frac{1}{\delta })(y+\alpha )D_{}$$
(cf. (20)). Thus, in the realization by finite-difference operators the corresponding spectral problem (\*) has the form
$$\frac{2B}{\delta ^2}\varphi (y+2\delta )\frac{2}{\delta ^2}\left[y+\alpha 2B+\delta C\right]\varphi (y+\delta )$$
(30)
$$+\frac{2}{\delta }\left[(1+\frac{2}{\delta })(y+\alpha )\frac{B}{\delta }+C\right]\varphi (y)\frac{2}{\delta }(1+\frac{1}{\delta })(y+\alpha )\varphi (y\delta )=E\varphi (y),$$
and is characterized by existence of infinitely family of eigenfunctions given by polynomials in $`y`$.
The operator $`h_\delta (y,D_\pm )`$ now becomes the four-point finite-difference operator, see Fig.2 .
Fig. 2. Graphical representation of the operator (29).
It is quite surprising that a simple transformation like linear shift of variable $`y`$ in differential operator (20) (which leads to nothing non-trivial, see above) leads to occurrence of the extra point in the isospectral finite-difference counterpart changing a type of its non-locality. Nevertheless, these operators remain to be canonically equivalent being related each other by a canonical transformation. So, by a canonical transformation one can change a non-local nature of finite-difference operators.
It is quite interesting to abandon the condition of canonical equivalence and pose a question about the most general differential (finite-difference) operators isospectral to the harmonic oscillator and possessing the infinitely-many polynomial eigenfunctions. One can easily show that ignoring the condition of canonical equivalence leads to nothing new for differential operators. However, for the case of finite-difference operators much wider class of operators occurs than (20) or (29). As a natural constraint we have to impose a condition of maximal number of points in the finite-difference operators we search for. Let us consider three-point operators (see Fig.1). A simple analysis leads to a statement that the three-parametric operator
(31)
$$\stackrel{~}{h}_\delta (y,D_\pm )=2[(1A)y+C_+)]D_++2[Ay+C_{})D_{},$$
where $`A,C_\pm `$ are parameters, is the most general three-point finite-difference operator with infinitely-many polynomial eigenfunctions, which is isospectral to the harmonic oscillator. Of course, the operators (16) and (18) are particular cases of (31). The form (31) can be understood as a consequence of the Theorem which states that any finite-difference operator $`h_\delta (y,D_\pm )`$ preserving the infinite flag of polynomials
(32)
$$𝒫_n(y)=1,y,y^2,\mathrm{},y^n,$$
should have a representation in terms of the operators:
(33)
$$J_\pm ^{}=D_\pm ,J_\pm ^0=yD_\pm .$$
As a remark we should note that the generators (33) do not span an algebra closed with respect to commutators. However, if we consider a linear space spanned by (33), then the Heisenberg algebra (18) appears as its subspace.
It is worth to mention that the canonical discretization of the harmonic oscillator can be made directly in the configuration $`x`$space where the original harmonic oscillator (1) is defined. If we make a gauge transformation (5) with (4) at $`p=0`$ as the gauge factor but without the change of variable, then
(34)
$$h^{(2)}(x,_x)=\frac{1}{\omega }(\mathrm{\Psi }_0^{(0)}(x))^1\mathrm{\Psi }_0^{(0)}(x)=\frac{1}{2\omega }_x^2+x_x,$$
is another algebraic form of the harmonic oscillator (cf.(5)). The operator (34) has the Hermite polynomials, $`H_k(\sqrt{\omega }x)`$, as the eigenfunctions. Now we can make $`\delta `$discretization firstly rewriting (34) in the Fock space formalism
(35)
$$h^{(2)}(b,a)=\frac{1}{2\omega }a^2+ba=\frac{1}{2\omega }J^{}J^{}+J^0,$$
(cf.(17)) and then realizing $`a,b`$ by finite-difference operators (19)
(36)
$$h_\delta ^{(2)}(x,D_\pm )=\frac{1}{2\omega }D_+^2+(x+\alpha )D_{}.$$
This operator is canonically-equivalent to the harmonic oscillator, it is defined on the uniform linear lattice in $`x`$space and has the eigenvalues $`E_k=k,k=0,1,2\mathrm{}`$. However, it does not possess the symmetry $`xx`$ unlike the $`\delta `$ discretized, canonically-equivalent operator (20) in $`y`$space. The operator (36) is non-local, four-point operator with eigenfunctions
(37)
$$\varphi _k(x)=\underset{\mathrm{}=0}{\overset{k}{}}a_{\mathrm{}}\omega ^\frac{\mathrm{}}{2}(x+\alpha )^{(\mathrm{})}.$$
where $`z^{(\mathrm{})}`$ is quasi-monomial and $`a_{\mathrm{}}`$ are the coefficients in the expansion of the Hermite polynomials, $`H_k(x)=_{\mathrm{}=0}^ka_{\mathrm{}}x^{\mathrm{}}`$. These eigenfunctions are closely related to the polynomials which can be called the modified Hermite polynomials
(38)
$$\widehat{H}_k(x,\delta )=\underset{\mathrm{}=0}{\overset{k}{}}a_{\mathrm{}}x^{(\mathrm{})},$$
and, thus,
$$\varphi _k(x)=\widehat{H}_k(\sqrt{\omega }(x+\alpha ),\sqrt{\omega }\delta ).$$
Summarizing, in this Section we presented two $`\delta `$ discretized operators which are canonically-equivalent to the harmonic oscillator, (20) and (36), which are defined on the uniform grid in $`y`$ and $`x`$spaces, respectively. In spite of the fact that they are represented by different elements of the Heisenberg-Weyl algebra, they have infinitely-many polynomial eigenfunctions and are isospectral to the original harmonic oscillator (1).
## 3. Dilatation-covariant discretization
Apart from the translation-covariant discretization given by $`\delta `$-derivative (see (19)) there exists a dilatation-covariant discretization based on $`q`$-derivative, $`𝒟_q`$, which is also called the Jackson symbol
$$𝒟_qf(y)=\frac{f(qy)f(y)}{y(1q)},$$
where $`q`$ is a complex number. This Section will be devoted to a brief discussion of the dilatation-covariant discretization or $`q`$discretization. First of all by following the above-mentioned philosophy a natural question can be posed about an existence of a quantum canonical conjugate to $`q`$-derivative – the operator which obeys together with $`𝒟_q`$ the commutation relations (8). Up to our knowledge a definite answer is not found so far and very likely such an object does not exist in terms of well-defined operators. However, it is well known that the derivative $`𝒟_q`$ appears naturally in connection to a realization of quantum algebras in action on functions in one and several variables. Thus, it looks reasonable to explore a generalization of the underlying Heisenberg algebra described above to the case of the $`q`$deformed (quantum) Heisenberg algebra. Hence, we will study ‘deformed’ quantum systems possessing a $`q`$deformed (quantum) Heisenberg algebra as a hidden algebra instead of the standard Heisenberg algebra (8) (see discussion in Introduction).
In order to proceed let us ask first what would happen if in the expressions (17), (28) the operators $`a,b`$ are not the generators of the Heisenberg algebra (8) but the generators of the $`q`$deformed Heisenberg algebra
(39)
$$[a,b]_qabqba=1,$$
where $`q`$ is a parameter. Following the Theorem proved in , one can demonstrate that within the $`q`$deformed Fock space built on the $`q`$deformed Heisenberg algebra (39) there exists the flag $`𝒫`$ of linear spaces of polynomials in $`b`$ (see (12)), which is preserved by the operators (17), (28). By a simple calculation one can find the eigenvalues of the operators (17), (28) in the spectral problem $`()`$
(40)
$$E_n^{(q)}=2\{n\},n=0,1,\mathrm{},$$
where
$$\{n\}=\frac{1q^n}{1q},$$
is a so-called $`q`$number and $`\{n\}n`$, if $`q1`$. It is evident that if the parameter $`q`$ is a real number the spectra of (17), (28) are real.
The algebra (39) has a realization in terms of $`q`$derivative and the operator of multiplication (see, for example, )
(41)
$$a=𝒟_q,b=y,$$
with the same $`q`$ as in (39). This realization has a property that the vacuum remains the same as in the cases (18)-(19) and without loss of generality it can be set as $`|0>=1`$. Now we can substitute (41) in (17) and the following operator emerges
$$h_q^{(1)}(y,𝒟_q)=2\stackrel{~}{J}^0\stackrel{~}{J}^{}+2\stackrel{~}{J}^02(p+\frac{1}{2})\stackrel{~}{J}^{}=$$
(42)
$$2y𝒟_q^2+2(yp\frac{1}{2})𝒟_q,$$
where the generators $`\stackrel{~}{J}^0=ba,\stackrel{~}{J}^{}=a`$ have the same functional form as in (10) but obey the $`q`$deformed commutation relation
$$[\stackrel{~}{J}^0,\stackrel{~}{J}^{}]_{1/q}\stackrel{~}{J}^0\stackrel{~}{J}^{}\frac{1}{q}\stackrel{~}{J}^0\stackrel{~}{J}^{}=\stackrel{~}{J}^{},$$
forming the $`q`$deformed Borel subalgebra $`b(2)_q`$ of the $`q`$deformed algebra $`sl(2)_q`$ (for discussion see ). In this case the operators (17), (28) are the $`sl(2)_q`$-exactly-solvable operators. Moreover, the operator (17) (as well as (28)) can be called the $`q`$deformed harmonic oscillator Hamiltonian in Fock space possessing $`sl(2)_q`$ hidden algebra.
The operator $`h_q(y,D_q)`$ is a non-local, three-point, discrete, dilatation-covariant operator defined on exponential lattice. It is illustrated by Fig.3.
Fig. 3. Graphical representation of the operator (42)
The operator (42) can be called the algebraic form of the Hamiltonian of the $`q`$discretized harmonic oscillator.
The spectral problem for the operator (42) has a form
$$\frac{2}{yq(q1)^2}\varphi (q^2y)+\left[\frac{2+q+q^22pq(1q)}{q(q1)^2}\frac{1}{y}+\frac{2}{1q}\right]\varphi (qy)$$
(43)
$$\left[\frac{1+q2p(1q)}{y(q1)^2}+\frac{2}{1q}\right]\varphi (y)=E^{(q)}\varphi (y).$$
or, the r.h.s. can be taken as
(44)
$$=E^{(q)}\varphi (qy).$$
or as
(45)
$$=E^{(q)}\varphi (q^2y).$$
Usually, the spectral problem for $`q`$discrete operators is defined with (44) as the r.h.s. (see, for example, ). It assumes that the middle point in Fig.3 remains fixed under dilatation. Introducing the new variable $`\stackrel{~}{y}=qy`$, it can be seen explicitly.
If in the case (43) the eigenvalues are given by (40) while for (44), (45) the eigenvalues are equal to
(46)
$$E_n^{(q)}=2\{n\},n=0,1,\mathrm{}$$
(47)
$$E_n^{(q)}=2q^{2n}\{n\},n=0,1,\mathrm{}$$
correspondingly. In the limit $`q1`$ all three expressions coincide corresponding to the original harmonic oscillator spectrum. The spectral problems (43)–(45) can be considered as a possible definition of a $`q`$deformed harmonic oscillator. In the literature it is well-known many other definitions of the $`q`$deformed harmonic oscillator (see for example, , and references therein, ) <sup>8</sup><sup>8</sup>8Usually, these deformations are done by a direct discretization of the original Hamiltonian (1). Most of all are based on a discretization of the Infeld-Hall factorization representation of (1). Such a situation reflects an ambiguity of making a $`q`$deformation <sup>9</sup><sup>9</sup>9For example, any term in non-deformed expression can be modified by multipliers of the type $`q^a`$ and even some extra terms can be added with vanishing coefficients in the limit $`q1`$ like $`(1q)^b,b>0`$ as well as absence of clear physical criteria, which can remove or reduce this ambiguity. For instance, in the literature it is exploited three different types of the $`q`$Laguerre polynomials (see, for example, an excellent review ), but it is not clear why other possible $`q`$deformations of Laguerre polynomials are not studied.
Isospectrality of (17) and (28) is preserved by the $`q`$deformation. Substitution of (41) in (28) gives a slight modification of the expressions (42). Unlike translation-covariant case it does not lead to a change of the nature of non-locality changing the number of points in the operator (42) as it is happened for the operators (20) and (29). The $`q`$deformation of another algebraic form of the harmonic oscillator (34) in Fock space
(48)
$$h_q^{(2)}(b,a)=\frac{1}{2\omega }\stackrel{~}{J}^{}\stackrel{~}{J}^{}+\stackrel{~}{J}^0,$$
(cf.(35)) takes in terms of the $`q`$derivative the following form
(49)
$$h_q^{(2)}(x,𝒟_q)=\frac{1}{2\omega }𝒟_q^2+x𝒟_q,$$
(cf.(42)). It should be mentioned that the operators (42) and (49) are defined on the essentially different lattices, which are exponential in $`x`$ and $`y`$variables, respectively.
Similar to what was done for canonical transformations it seems natural to introduce a notion of $`q`$deformed canonical transformations, when two $`q`$deformed systems are $`q`$canonically equivalent if they can be connected through the $`q`$deformed canonical transformation (see below). However, we were unable to find well-defined, non-trivial realization of the algebra (39) other than (40) which, for instance, would be similar to the realization (20) for (8) (see discussion in ) .
## 4. Anharmonic oscillator (perturbation theory)
Anharmonic oscillator is one of the most important non-exactly-solvable problems of quantum mechanics. One of the simplest and the most popular examples is given by the Hamiltonian
(50)
$$^{(aho)}=\frac{1}{2}\frac{^2}{x^2}+\frac{\omega ^2}{2}x^2+gx^{2n},n=2,3,4,\mathrm{}$$
where $`g`$ is a coupling constant<sup>10</sup><sup>10</sup>10So far, the most comprehensive study of the anharmonic oscillator (50) was carried out by Bender-Wu in the classical paper (see also ) . Making first the gauge rotation of (50) with the gauge factor (4) we get
$$h^{(aho)}(y,_y)=\frac{1}{\omega }(\mathrm{\Psi }_0^{(p)}(x))^1^{(aho)}\mathrm{\Psi }_0^{(p)}(x)_{y=\omega x^2}$$
(51)
$$=2y_y^2+2(yp\frac{1}{2})_y+\lambda y^n,$$
(cf.(5)), where $`p=0,1`$ and has a meaning of parity and $`\lambda =g\omega ^{(n+1)}`$. By the reasons which will be clear later the operator (51) can be called algebraic form of the anharmonic oscillator Hamiltonian. Replacing in (51) the derivative and the coordinate by the elements of the Heisenberg algebra (8), $`_ya,yb`$ we will arrive at the element of the Heisenberg-Weyl algebra
(52)
$$h^{(aho)}(b,a)=2ba^2+2(bp\frac{1}{2})a+\lambda b^n,$$
which can be called the anharmonic oscillator Hamiltonian in the Fock space. Now we can fix $`n=2`$ and we will consider this particular case of the quartic anharmonic oscillator as the major example in further consideration.
Let us find a finite-difference operator which is canonically equivalent to (50). In order to do it we simply substitute the realization (19) of the Heisenberg algebra to the operator (52)
$$h_\delta ^{(aho)}(y,D_\pm )=\frac{2}{\delta }[y+\alpha +\delta (p+\frac{1}{2})]D_++2[(1+\frac{1}{\delta })(y+\alpha )$$
(53)
$$\frac{\delta \lambda }{2}(y+\alpha )^{(2)}]D_{}+\frac{\delta ^2\lambda }{2}(y+\alpha )^{(2)}D_{}D_{}+\frac{\lambda }{2}(y+\alpha )^{(2)}.$$
(see Fig.4). It is quite amazing that the perturbation $`\lambda b^2`$ leads solely to an addition of one more point (marked by $``$) to the harmonic oscillator operator. A model characterized by the operator (53) can be called a finite-difference anharmonic oscillator.
Fig. 4. Graphical representation of the problem (53)
It is non-local, four-point, finite-difference operator (for comparison see Fig.1 and Fig.2 with $`\delta \delta `$).
The corresponding spectral problem $`()`$ looks as follows
$$\frac{2}{\delta ^2}\left[y+\alpha +\delta (p+\frac{1}{2})\right]\varphi (y+\delta )+\frac{2}{\delta }\left[(1+\frac{2}{\delta })y+\alpha +p+\frac{1}{2}\right]\varphi (y)$$
(54)
$$\frac{2}{\delta }(1+\frac{1}{\delta })(y+\alpha )\varphi (y\delta )+\lambda (y+\alpha )^{(2)}\varphi (y2\delta )=E\varphi (y).$$
Their eigenvalues coincides with those of the anharmonic oscillator (50)–(52).
We have to emphasize that a presence of the anharmonic term changes the nature of non-locality of the harmonic oscillator leading to appearance an extra point. In particular case $`\delta =1`$ the number of points is reduced to three, however, the lattice becomes non-uniform.
In similar way one can construct a $`q`$deformed anharmonic oscillator taking the operator (52) as an element of the $`q`$Fock space. Substituting the realization (41) of the $`q`$deformed Heisenberg algebra into (52) we get
(55)
$$h_q^{(aho)}(y,𝒟_q)=2y𝒟_q^2+2(yp\frac{1}{2})𝒟_q+\lambda y^2,$$
(cf. (42)) which three-point, non-local operator (see, for example, Fig.3).
The spectral problem $`()`$ for the operator (55) has a form
$$2\frac{\varphi (q^2y)}{yq(q1)^2}+2\frac{1+q+(yp\frac{1}{2})q(1q)}{yq(q1)^2}\varphi (qy)$$
(56)
$$\frac{2(2p+1)(1q)+2y(1q)\lambda y^2(q1)^2}{y(q1)^2}\varphi (y)=E^{(q)}\varphi (y).$$
(cf.(42)). Certainly, the r.h.s. in (56) can vary being equal to either (44), or (45). It reflects a fact that in the case of the $`q`$Fock space the spectral problem $`()`$ can be modified in one way
$$L(b,a)\varphi (b)|0>=\lambda \varphi (qb)|0>,()$$
or another
$$L(b,a)\varphi (b)|0>=\lambda \varphi (q^2b)|0>.()$$
If we introduce the multiplication operator $`T_qf(x)=f(qx)`$ the problems $`()`$ or $`()`$ correspond to an appearance of non-trivial operator weight factors in r.h.s.
$$L(b,a)\varphi (b)|0>=\lambda T_q\varphi (b)|0>,$$
or
$$L(b,a)\varphi (b)|0>=\lambda T_q^2\varphi (b)|0>,$$
respectively.
It can be easily shown that the operator (52) describing the anharmonic oscillator does not belong to the class of the $`sl(2)`$exactly-solvable operators. Hence their eigenfunctions are not polynomials. However, it was proved in that in framework of some perturbative approach (see below) as a consequence of the fact that the perturbation $`b^2𝒫_2(b)`$ (see (12)) the perturbation theory in powers of the parameter $`\lambda `$ is algebraic one: any correction to any eigenfunction is a finite-order polynomial and hence can be found by algebraic means. In fact, such a perturbative approach provides a certain regular way to define an object in the Heisenberg-Weyl algebra which can be called the Hamiltonian of the quartic anharmonic oscillator. From practical point of view such a perturbation theory has very important feature: once being developed for the operator (52) it gives a unique possibility to construct simultaneously (!) the perturbation theory for the operators (50), (54), (56).
Let us approach to a construction of the above-mentioned perturbation theory. Following a standard prescription we take the spectral problem $`()`$ with
$$L(b,a)=h_0+\lambda h_1$$
and develop a perturbation theory in powers of $`\lambda `$ searching for corrections in a form
(57)
$$\varphi =\lambda ^n\varphi _n,E=\lambda ^nE_n.$$
Collecting the terms of the order $`\lambda ^n`$ it is easy to derive an equation for the $`n`$th correction
(58)
$$(h_0E_0)\varphi _n=\underset{i=1}{\overset{n}{}}E_i\varphi _{ni}h_1\varphi _{n1}.$$
A remarkable feature of this form of perturbation theory a possibility to study a single state separately, without touching other states as it was the case of the Rayleigh-Schroedinger form of perturbation theory.
Now we take as the unperturbed operator $`h_0`$ the Hamiltonian of the harmonic oscillator in the Fock space (17) and consider as the perturbation $`h_1=b^2`$. As it was mentioned already, in this case the $`n`$th correction $`\varphi _n`$ to eigenfunction should be polynomials in $`b`$. As an instructive example let us study the ground state of anharmonic oscillator. The ground state of the unperturbed problem (17) is characterized by
(59)
$$\varphi _0=1,E_0=1.$$
By solving the equation (58) after simple calculations we can get the explicit form for the first several corrections, for example,
$$\varphi _1=\frac{b^2}{2\{2\}}+\frac{(3+2p)}{4}b,E_1=\frac{(1+2p)(3+2p)}{4},$$
and
$$\varphi _2=\frac{1}{4}[\frac{1}{\{2\}\{4\}}b^4+\frac{2q^2+5q+6+2p(q+2)}{2\{2\}\{3\}}b^3+$$
$$\frac{4q^2+12q+15+8p(q+2)+4p^2}{4\{2\}}b^2+\frac{(3+2p)[q^2+3q+3+2p(q+1)]}{2}b],$$
(60)
$$E_2=\frac{(1+2p)(3+2p)}{8}\left[q^2+3q+3+2p(q+1)\right].$$
Without any difficulties one can find several next corrections. However, it becomes evident very quickly that the complexity of calculations is growing very fast with a number of correction. Using different realizations of the $`(q)`$Heisenberg algebra one can calculate perturbative corrections to the various forms of anharmonically-perturbed harmonic oscillator (differential, finite-difference, discrete).
* $`q=1,b=y`$.
This case corresponds to the coordinate-momentum realization of the Heisenberg algebra (18) and vacuum definition $`|0>=1`$. It leads to a standard anharmonic oscillator (50) and the corrections are:
$$\varphi _1=\frac{y^2}{4}+\frac{(3+2p)}{4}y,$$
and
$$\varphi _2=\frac{1}{4}[\frac{y^4}{8}+\frac{11+6p}{12}y^3+\frac{31+24p+4p^2}{8}y^2+$$
(61)
$$\frac{(3+2p)(7+4p)}{2}y],E_2=\frac{(1+2p)(3+2p)(7+4p)}{8},$$
where $`E_1`$ is the same as in (60).
Given form of the perturbation theory coincides to the so-called ‘$`F`$functions method’ developed by Dalgarno (see discussion in ), which in fact was realized for the case of the ahnarmonic oscillator (50) in . It is easy to check that the corrections (61) coincide to those calculated in text-books (see for example ).
* $`q=1,b=(y+\alpha )(1\delta 𝒟_{})`$ (see (19))
This case corresponds to the translation-covariant discretization and perturbed finite-difference harmonic oscillator (20) (see (53)):
$$\varphi _1=\frac{\stackrel{~}{y}^{(2)}}{4}+\frac{(3+2p)}{4}\stackrel{~}{y},$$
and
$$\varphi _2=\frac{1}{4}[\frac{\stackrel{~}{y}^{(4)}}{8}+\frac{11+6p}{12}\stackrel{~}{y}^{(3)}+\frac{31+24p+4p^2}{8}\stackrel{~}{y}^{(2)}+$$
(62)
$$\frac{(3+2p)(7+4p)}{2}\stackrel{~}{y}],$$
where $`E_1`$ is the same as in (60) and $`E_2`$ is the same as in (61). Here $`\stackrel{~}{y}^{(n+1)}=(y+\alpha )^{(n+1)}=(y+\alpha )(y+\alpha \delta )\mathrm{}(y+\alpha n\delta )`$ is quasi-monomial.
* $`q1,b=y`$.
This case corresponds to the perturbed $`q`$harmonic oscillator (55)-(56) and corrections are equal to
$$\varphi _1=\frac{y^2}{2\{2\}}+\frac{(3+2p)}{4}y,E_1=\frac{(1+2p)(3+2p)}{4}$$
and
$$\varphi _2=\frac{1}{4}[\frac{1}{\{2\}\{4\}}y^4+\frac{2q^2+5q+6+2p(q+2)}{2\{2\}\{3\}}y^3+$$
$$\frac{4q^2+12q+15+8p(q+2)+4p^2}{4\{2\}}y^2+\frac{(3+2p)[q^2+3q+3+2p(q+1)]}{2}y],$$
(63)
$$E_2=\frac{(1+2p)(3+2p)}{8}\left[q^2+3q+3+2p(q+1)\right].$$
It is quite interesting to see how above-mentioned results will be modified if instead of the spectral problem $`()`$ the spectral problem $`()`$ is considered. The equation (58) for the $`n`$th correction becomes
(64)
$$(h_0E_0T_q)\varphi _n=\underset{i=1}{\overset{n}{}}E_iT_q\varphi _{ni}h_1\varphi _{n1},$$
(cf. (58)). The ground state of the unperturbed problem (59) is unchanged. The first correction is also unchanged while the second correction now takes a modified form
$$\varphi _2=\frac{1}{4}[\frac{1}{\{2\}\{4\}}b^4+\frac{2q^2+5q+6+2p(q+2)}{2\{2\}\{3\}}b^3+$$
$$\frac{4q^2+12q+15+8p(q+2)+4p^2}{4\{2\}}b^2+\frac{(3+2p)[q^2+3q+3+2p(q+1)]}{2}b],$$
(65)
$$E_2=\frac{(1+2p)(3+2p)}{64}\left[659q^2+8p(73q^2)+12p^2(1q^2)\right].$$
## 5. Anharmonic oscillator (quasi-exactly-solvable model)
Among one-dimensional Schroedinger equations there is some class of problems possessing a certain outstanding property - first several eigenstates can be found explicitly, by algebraic means. Such problems are called quasi-exactly-solvable . Among ten known families of one-dimensional quasi-exactly-solvable potentials there exists one, which can be treated as an anharmonic oscillator and its Hamiltonian is
(66)
$$^{(qes)}=\frac{1}{2}\frac{^2}{x^2}+[\frac{\omega ^2}{2}(2n+\frac{3}{2}+p)g]x^2+g\omega x^4+\frac{g^2}{2}x^6.$$
Here $`g`$ is a coupling constant and $`x(\mathrm{},\mathrm{})`$. Parameter $`p=0,1`$ has meaning of parity. The first $`(n+1)`$ eigenfunctions of parity $`p`$ (but not others) are of the form
(67)
$$\mathrm{\Psi }_n^{(p)}(x)=x^pP_n(\omega x^2)e^{\omega x^2/2gx^4/4}$$
where $`P_n(y)`$ is a polynomial of degree $`n`$.
Making first the gauge rotation of (50) with the gauge factor (67) at $`n=0`$ we get
$$h^{(qes)}(y,_y)=\frac{1}{\omega }(\mathrm{\Psi }_0^{(p)}(x))^1^{(qes)}\mathrm{\Psi }_0^{(p)}(x)_{y=\omega x^2}$$
(68)
$$=2y_y^2+2(\lambda y^2+yp\frac{1}{2})_y2\lambda ny,$$
(cf. (5),(51)), where $`\lambda =g\omega ^2`$ and constant terms are dropped out. The spectral problem for (68) is defined on the half-line, $`y[0,\mathrm{})`$. The first $`(n+1)`$ eigenfunctions of (68) are some polynomials of the degree $`n`$, $`P_n(y)`$ (cf.(67)) possessing $`k=0,1,\mathrm{},n`$ real zeroes inside of the interval $`y[0,\mathrm{})`$.
Replacing in (68) the derivative and the coordinate by the elements of the Heisenberg algebra (8): $`_ya,yb`$ we arrive at the element of the Heisenberg-Weyl algebra
(69)
$$h^{(qes)}(b,a)=2ba^2+2(\lambda b^2+bp\frac{1}{2})a2\lambda nb,$$
(cf. (17),(28),(52)). As in previous consideration the operator $`h^{(qes)}(b,a)`$ can be treated as an element of the Fock space as well as an element of the $`q`$Fock space. This operator can be called the Fock space Hamiltonian of the anharmonic quasi-exactly-solvable oscillator. The operator $`h^{(qes)}(b,a)`$ in the Fock space is $`sl_2`$quasi-exactly-solvable operator. It can be rewritten in terms of the generators of $`sl_2`$algebra (10)
(70)
$$h^{(qes)}(b,a)=2J_n^0J_n^{}+2\lambda J_n^++2J_n^0(n+1+2p)J_n^{},$$
where as always the constant terms are dropped off.
However, it is easy to check that in the $`q`$Fock space the operator $`h^{(qes)}(b,a)`$ is not $`sl_{2q}`$quasi-exactly-solvable operator, since it can not be rewritten in terms of the generators of $`sl_{2q}`$algebra <sup>11</sup><sup>11</sup>11In the case of the $`sl_{2q}`$algebra the major modification of (10) comes for the positive-root generator
$$J_n^+=b^2anb\stackrel{~}{J}_n^+=b^2a\{n\}b,$$
while the Cartan generator
$$J_n^0=ba\frac{n}{2}\stackrel{~}{J}_n^0=ba\frac{\{n\}\{n+1\}}{\{2n+2\}},$$
where $`\{n\}`$ is the $`q`$number (see (31)) and $`J^{}`$ remains unchanged; for details, see . This operator needs a slight modification of the last term in order to become the $`sl_{2q}`$quasi-exactly-solvable:
(71)
$$h^{(qes)}(b,a)=2ba^2+2(\lambda b^2+bp\frac{1}{2})a2\lambda \{n\}b=$$
$$2\stackrel{~}{J}_n^0\stackrel{~}{J}_n^{}+2\lambda \stackrel{~}{J}_n^++2\stackrel{~}{J}_n^0\left(2\frac{\{n\}\{n+1\}}{\{2n+2\}}+1+2p\right)\stackrel{~}{J}_n^{},$$
(cf. (69)), where $`\{n\}`$ is the $`q`$number (see (31)) and the constant terms in the second expression are dropped off.
The first $`(n+1)`$ eigenfunctions of (71) are some polynomials in $`b`$ of degree $`n`$, $`P_n(b)`$. If in the case $`n=0`$ the first eigenfunctions for the spectral problems $`()`$ and $`()`$ coincide and equal to:
$$\varphi ^{(0)}=1,E^{(0)}=0,$$
while for the case of $`n=1`$ they are already different. Namely, for the spectral problem $`()`$,
$$\varphi _\pm ^{(1)}=b\frac{1\sqrt{1+4\lambda (1+2p)}}{4\lambda },$$
(72)
$$E_\pm ^{(1)}=\frac{1\sqrt{1+4\lambda (1+2p)}}{2},$$
where sub-indices $`(+)`$ and $`()`$ are assigned to the ground and the first excited state, respectively. As for the spectral problem $`()`$ the formulas (72) are modified
$$\varphi _\pm ^{(1)}=b\frac{1\sqrt{1+4\lambda q(1+2p)}}{4\lambda },$$
(73)
$$E_\pm ^{(1)}=\frac{1\sqrt{1+4\lambda q(1+2p)}}{2q}.$$
Of course, when $`q=1`$ the formulas (72) and (73) coincide.
Now we can find a finite-difference operator canonically equivalent to (68). In order to do it we put $`q=1`$ in (68) and substitute the realization (19) of the Heisenberg algebra into the operator (69)
$$h_\delta ^{(qes)}(y,D_\pm )=$$
$$\frac{2}{\delta }[y+\alpha +\delta (p+\frac{1}{2})]D_++2[(1+\frac{1}{\delta })(y+\alpha )+\lambda (y+\alpha )(y+\alpha +\delta (n1))]D_{}$$
(74)
$$2\lambda \delta (y+\alpha )^{(2)}D_{}D_{}2\lambda n(y+\alpha ),$$
(cf. (53)). It turns out that this operator is four-point finite-difference operator (see Fig.4). It is quite amazing that independently on $`n`$ the perturbation $`2\lambda b(ban)`$ to the harmonic oscillator operator (see Fig.1) leads solely to an appearance of one extra point (marked by $``$ in Fig.4) similarly to (53). A model characterized by the operator (74) can be called a finite-difference quasi-exactly-solvable anharmonic oscillator.
Without loss of generality we place $`\alpha =0`$ in (74) and the corresponding spectral problem $`()`$ for (74) looks like
$$\frac{2}{\delta }\left[\frac{y}{\delta }+p+\frac{1}{2}\right]\varphi (y+\delta )+\frac{2}{\delta }\left[(1+\frac{2}{\delta })y+p+\frac{1}{2}\right]\varphi (y)$$
$$\frac{2}{\delta }y\left[(1+\frac{1}{\delta })\lambda (y\delta (n+1))\right]\varphi (y\delta )$$
(75)
$$2\lambda \frac{y^{(2)}}{\delta }\varphi (y2\delta )=E\varphi (y).$$
It is natural to assume that the spectral problem (75) is defined in $`y[0,\mathrm{})`$. Their first $`(n+1)`$ eigenvalues coincide with those of the anharmonic oscillator Hamiltonian (66) as well as the operators (68),(69). Their eigenfunctions remain polynomials but modified – each monomial should be replaced by quasi-monomial, $`y^ky^{(k)}`$. For instance, the formulas (72)–(73) become
$$\varphi _\pm ^{(1)}=y\frac{1\sqrt{1+4\lambda (1+2p)}}{4\lambda },$$
(76)
$$E_\pm ^{(1)}=\frac{1\sqrt{1+4\lambda (1+2p)}}{2}.$$
It is worth to note that for physically relevant parameters $`g0`$ and $`q0`$ <sup>12</sup><sup>12</sup>12In this case $`q`$ has the meaning the parameter of dilatation the eigenfunctions (72), (76) in domain $`y0`$ have no nodes for ground state and have the only one node for the first excited state like an analogue of the oscillation (Sturm) theorem holds (see, for example, ). Likely, an agreement with Sturm theorem will hold for higher excited states but we were unable to prove it in full generality.
## 6. Conclusion
We introduced a simple-minded notion of canonical equivalence in quantum mechanics. Then we showed that for canonically-equivalent systems the spectra remain the same and constructed discrete systems, which are canonically-equivalent to the harmonic and a certain type of anharmonic oscillators – the quartic oscillator and the quasi-exactly-solvable oscillator. In general, these discrete systems are defined on non-linear lattices.
We restricted our studies to the operators in the algebraic form possessing the polynomial eigenfunctions. An important question to ask is what would happen if the operator in hand possesses an algebraic form but non-polynomial eigenfunctions. Perhaps, the most explicit instructive example is given by the original Hamiltonian (1) of the harmonic oscillator. Following our philosophy it can be rewritten as the element of the Fock space
(77)
$$=\frac{1}{2}a^2+\frac{\omega ^2}{2}b^2,$$
where its ground state eigenfunction can be written formally as
(78)
$$\mathrm{\Psi }_0(b)|0>=\underset{n=0}{\overset{\mathrm{}}{}}\frac{(\omega )^n}{2^nn!}b^{2n}|0>.$$
Substitution of (19) with $`|0>=1`$ into (78) leads to an infinite series of the form
(79)
$$\stackrel{~}{\mathrm{\Psi }}_0(x)=\underset{n=0}{\overset{\mathrm{}}{}}\frac{(\omega )^n}{2^nn!}x^{(2n)},$$
which has zero radius of convergence in $`x`$ for any $`\delta 0`$ <sup>13</sup><sup>13</sup>13This type of expansion on the basic set of quasi-monomials is known in literature as the Newton series. For discussion see, for example, . Although the Taylor expansion in powers of $`x`$ for the original eigenfunction (4) at $`p=0`$ had infinite radius of convergence. So, the radius of convergence has delta-function behaviour. Perhaps, one of possible ways to remedy this drawback is to gauge-rotate a Hamiltonian with non-polynomial eigenfunctions using a semi-classical or somehow modified semi-classical wavefunction as a gauge factor requiring an appearance of an algebraic form of the final operator.
Acknowledgement
I want to express my deep gratitude to M. Shifman, M. Shubin, E. Shuryak, Yu.F. Smirnov and N. Vasilevsky for their sincere interest to the work and valuable discussions, and especially to B. Julia for the conversation, which after some time turned out to be very inspiring.
|
warning/0004/quant-ph0004023.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The result presented here is the incompatibility, within the stochastic evolution scheme assumption, of the Markovian feature of a microscopic system and the quantum jump to which such systems are subjected (e.g. in a measurement process). These evolution schemes do not only show interest from a fundamental standpoint in quantum mechanics (cf., e.g., ), but also in quantum optics (cf. and references therein) where modern technology begins to offer the possibility of monitoring single systems. This reinforces the interest on the stochastic methods applied to Hilbert spaces. In the following lines, the concept of stochastic evolution scheme is briefly set forth, the notion of quantum jump is succintly discussed and the announced incompatibility theorem is proved. A short discussion upon the entailed physical situation and its more direct consequences is also included.
## 2 Quantum Stochastic Evolution Schemes
We shall understand as Stochastic Evolution Schemes (SES) those evolution models which are non-deterministic in the sense that, given the state of a system at any instant t, we cannot predict with absolute certainty its state at a time later than t, but at most the probability of evolution towards one or other state. It must be noticed that the mathematical form in which the state of physical system should be described has not still been specified. We shall center ourselves on quantum systems, hence the latter will be determined by vectors belonging to a Hilbert space. The conjuction of both a stochastic and a hilbertian structure is not a mathematically difficult task, provided the global definition of stochastic process is kept in mind. Thus, we define a Quantum Stochastic Evolution Scheme (QSES) as a measurable application from a probability space $`(\mathrm{\Omega },𝔄,P)`$ on the set of mappings from $`^+`$ (standing for time) on the Hilbert space $``$ (the state space of the system). Formally
$$\begin{array}{cccc}\mathrm{\Psi }:& (\mathrm{\Omega },𝔄,P)& & ^^+\\ & \omega & & \mathrm{\Psi }(\omega )=\psi _t(\omega )\end{array}$$
where $`\psi _t()`$ denotes a mapping from $`^+`$ onto $``$ (usually denoted $`\mathrm{\Psi }(t)`$ in orthodox quantum mechanics).
As a consequence of the metric structure of $``$, it is always possible to define a $`\sigma `$-algebra with respect to which $`\mathrm{\Psi }`$ is measurable. The usual concepts appearing in the ordinary theory of stochastic processes are still valid. In particular, we may carry on talking of the transition probabilities. Thus, we establish the following definition
###### Def 1
Let $`st^+`$ and $`\psi ,\varphi `$. We call *transition probability* asociated to the QSES to the quantity $`P(s,\varphi ,t,\psi )`$ defined by
$$P(s,\varphi ,t,\psi )P\left(\mathrm{\Psi }_t=\psi |\mathrm{\Psi }_s=\varphi \right)$$
This concept is analogous to the usual probability of Markov chains in continuous time. The time homogeneity condition (physically evident, on the other hand) is assumed from the beginning:
###### Def 2
A QSES is *homogeneous* if its transition probability is stationary, i.e.,
$$P(s+u,\varphi ,t+u,\psi )=P(s,\varphi ,t,\psi )$$
for all $`u`$ such that $`0s+ut+u`$.
This property enables us to speak of a transition probability in a time $`t`$, given by
$$P(t;\varphi ,\psi )=P(0,\varphi ,t,\psi )=P(u,\varphi ,t+u,\psi )$$
The quantity $`P(t;\varphi ,\psi )`$ will be our basic tool to obtain the desired result. In an obvious way, it satisfies the following relations:
$`P(t;\varphi ,\psi )0`$ $`\psi ,\varphi ,t[0,\mathrm{})`$ (1)
$`{\displaystyle _{}}P(t;\varphi ,\psi )\mu (d\psi )=1`$ $`\varphi `$ (2)
where $`\mu `$ denotes the Lebesgue measure on $``$. The conditions (1) and (2) briefly state that $`P(t;\varphi ,\psi )`$ is for each $`t^+`$ a stochastic matrix. The Markov property is equally stated in this formalism:
###### Def 3
A QSES is said to be Markovian if it satisfies
$$_{}P(t;\varphi ,\psi )P(s;\psi ,\phi )\mu (d\psi )=P(t+s;\varphi ,\phi )\varphi ,\phi ,t,s^+$$
This assumption will be the central objective of our analysis. The fundamental result we need is the following
###### Th 1
Let $`P(;\varphi ,\psi )`$ be a stochastic transition matrix corresponding to a Markovian QSES. Then $`P(;\varphi ,\psi )`$ is continuous in $`(0,\mathrm{})`$ for all $`\varphi ,\psi `$ if and only if the following limit exists
$$\underset{t0^+}{lim}P(t;\varphi ,\psi )=g(\varphi ,\psi )$$
where $`g(\varphi ,\psi )`$ satisfies
$$\begin{array}{cccc}\hfill g(\varphi ,\psi )& & 0\hfill & \varphi ,\psi \\ \hfill _{}g(\varphi ,\psi )\mu (d\psi )& & 1\hfill & \varphi \\ \hfill g(\varphi ,\psi )& =& _{}g(\varphi ,\phi )g(\phi ,\psi )\mu (d\phi )\hfill & \varphi ,\psi \end{array}$$
This theorem is but a translation of the corresponding known theorem for Markov chains in continuous time (cf. ). We shall outline its proof and relegate minor details to the appendix. Notice should be taken of the generality enabled by the function $`g(\varphi ,\psi )`$ whose expression has not been detailed. The utility of this result rests on the possibility of checking the continuity of a stochastic matrix at every point through the study of a simple limit at the origin. The only assumed hypothesis is the Markov property.
Proof.
$`()`$ Let us suppose that $`P(t;\varphi ,\psi )`$ is continuous in $`(0,\mathrm{})`$. By Bolzano-Weierstrass theorem it is possible to find sequences $`\{t_n\}`$ and $`\{t_n^{}\}`$ such that
$`P(t;\varphi ,\psi )`$ $`=`$ $`\underset{n\mathrm{}}{lim}P(t+t_n;\varphi ,\psi )`$
$`P(t^{};\varphi ,\psi )`$ $`=`$ $`\underset{n\mathrm{}}{lim}P(t^{}+t_n^{};\varphi ,\psi )`$
Let us call
$`u(\varphi ,\psi )`$ $``$ $`\underset{n\mathrm{}}{lim}P(t_n;\varphi ,\psi )`$
$`u^{}(\varphi ,\psi )`$ $``$ $`\underset{n\mathrm{}}{lim}P(t_n^{};\varphi ,\psi )`$
Our goal is to establish that $`u(\varphi ,\psi )=u^{}(\varphi ,\psi )`$ for all $`\varphi ,\psi `$. By making use of the continuity hypothesis, the Markov property and some measure-theoretic usual theorems, it can be easily shown that the following two relations are simultaneously fulfilled (cf. Appendix):
$`u(\varphi ,\psi )`$ $``$ $`{\displaystyle _{}}u^{}(\varphi ,\phi )u(\phi ,\psi )\mu (d\phi )`$ (3)
$`u^{}(\varphi ,\psi )`$ $`=`$ $`{\displaystyle _{}}u^{}(\varphi ,\phi )u(\phi ,\psi )\mu (d\phi )`$ (4)
whence by the symmetry and arbitrariness of $`\varphi `$ and $`\psi `$, the equality $`u(\varphi ,\psi )=u^{}(\varphi ,\psi )`$ for all $`\varphi ,\psi `$ is derived.
$`()`$ Let us now suppose that there exists a unique limit $`u(\varphi ,\psi )`$ when $`t0`$. Then
$`\underset{n\mathrm{}}{lim}P(t+t_n;\varphi ,\psi )`$ $`=`$ $`\underset{n\mathrm{}}{lim}{\displaystyle _{}}P(t;\varphi ,\phi )P(t_n;\phi ,\psi )\mu (d\phi )=`$
$`=`$ $`{\displaystyle _{}}P(t;\varphi ,\phi )u(\phi ,\psi )\mu (d\phi )=P(t;\varphi ,\psi )`$
which implies that $`P(t;\varphi ,\psi )`$ is right continuous. But it is elementary to show that a right continuous function has at most a denumerable set of discontinuities, hence it is measurable. Then we are left with the task of proving that a Markovian stochastic transition matrix which has a denumerable set of discontinuities is continuous, result which is established in the appendix.
## 3 The quantum jump
Undoubtedly the quantum jump is one of the most controversial aspects of orthodox quantum mechanics. We are, for the analysis we set forth here, interested in the following aspect of this phenomenon. According to Mittelstaedt’s schematic representation (cf. ), the measurement process can be divided into three stages, namely, preparation, premeasurement and objectification and reading. In particular we are interested in the fact that during premeasurement $`0tt^{}`$, the composite system system+apparatus evolves unitarily following quantum-mechanical laws. At the very instant $`t^{}`$, the objectification and reading stage begins. It is in this transition where the core of the measurement problem is rooted and where the origin of the reduction postulate is located. We will partially assume that such a reduction takes place, i.e., at $`t^{}`$ the state vector transforms instantaneously (*jumps*) into another state vector (autostate of the measured observable). However, we do not enter into considerations about the origin or the factors of that jump, not even we attempt to interpret it. We only assume as a hypothesis that there exist physical situations in which the state vector instantaneously jumps to another vector. It is even permitted that the difference between such vectors be of non-null norm. We must now translate these ideas into the language of QSES’s. We then say that a QSES reflects the quantum jump if its transition matrix satisfies
$$P(t;\varphi ,\psi )=\{\begin{array}{c}\{\begin{array}{cc}1\hfill & \text{ if }0tt^{}\text{ and }\psi =U(t)\varphi \hfill \\ 0\hfill & \text{ si }0tt^{}\text{ and }\psi U(t)\varphi \hfill \end{array}\hfill \\ \\ h(\varphi ,\psi )(0)\text{ if }t>t^{}\text{ and it is possible }\hfill \\ \psi \varphi >ϵ\text{ for some }ϵ>0\hfill \end{array}$$
where U(t) is the usual quantum-mechanical evolution operator.
Notice that $`h(\psi ,\varphi )=|(\varphi ,\psi )|^2`$ should be expected in order to reproduce the reduction postulate. We only claim that after an unitary evolution during the premeasurement and being arrived at the instant of objectification and reading, the system jumps with finite non-null probability. This hypothesis is clearly of physical nature and its relationship with the Markovianity of the QSES constitutes the central aim of this note. The more relevant mathematical property involved is the discontinuity of $`P(t;\varphi ,\psi )`$, which may be showed using usual Calculus techniques applied to the definition of a quantum jump in a QSES.
## 4 The quantum jump and the Markov condition
To confront the two previous hypothesis (Markovian QSES and quantum jump) is not an exceedingly complicated task provided we make use of the preceding results. We in no case adopt a priori attitudes, but only study the compatibility of both ideas, which we present in the form of the following
###### Th 2
Let S be a quantum system described by a time-homogeneous QSES. If S is subjected to quantum jumps, then its QSES is non-Markovian.
Proof. The proof, though elementary, requires the use of a physical hypothesis which lately we shall comment in greater detail. Let us suppose that the mentioned QSES is Markovian, then its stochastic matrix is continuous in $`^+`$ if and only if the limit $`lim_{t0}P(t;\varphi ,\psi )`$ exists for all $`\varphi ,\psi `$. This limit, due to physical assumptions, exists and indeed amounts to
$$\underset{t0}{lim}P(t;\varphi ,\psi )=\{\begin{array}{cc}0& \text{si }\varphi \psi \\ 1& \text{si }\varphi =\psi \end{array}$$
(5)
Then $`P(t;\varphi ,\psi )`$ is continuous for all $`t`$ and in particular at the instant $`t^{}`$ in which the jump takes place, contrary to the assumed existence of such a jump.
The core of the proof is undoubtedly the question about the existence of the limit at the origin. The validity of such an assumption, though apparently trivial, we believe, deserves careful discussion.
## 5 Discussion
Firstly we will consider the question about the existence of the limit at the origin, distinguishing between physical and mathematical aspects. We shall refer to the existence of such a limit with the value formerly assigned by eq. (5) as *standard condition* o *standarization*, in clear analogy with Markov chains. In orthodox quantum mechanics, in which the state vector evolves in a deterministc fashion, this question is positively solved by means, e.g., of the imposition of the usual initial condition on the evolution operator $`U(t_o,t_o)=I`$. The physical interpretation is immediate: the state vector of a quantum physical system does not change if time has hardly elapsed. Furthermore, standarization appears as a hypothesis *assumed on physical grounds* in the study of the quantum Zeno paradox (cf. ).
However, we should now discuss if standarization is kept when the mathematical character of the state is changed, i.e., instead of being represented by a vector in a Hilbert space, we let it be represented by a Hilbert-space-valued stochastic process. We think that there exist notably suggestive reasons to claim that the standard condition is satisfied. Let us consider, e.g., the open quantum system formalism. There the system evolution is given by an operator semigroup which satisfies, among other properties, standarization (cf. ). And this is so even despite the enviromental uncontrollable influence.
In connection with the previously established theorem, we should remember that the standard condition is not necessary, since, according to section 2, it is only required the existence of the limit at the origin independently of its value.
We believe equally important to draw our attention on the implications of the theorem of the foregone section. It does not establish the impossibility of the quantum jump, nor does it deny the Markovian character of the stochastic evolution of a quantum systm. We understand the result is: Assumed the description of a quantum system by means of a homogeneous QSES, if the system exhibits quantum jumps, then the QSES cannot be Markovian. What attitudes can be adopted before such a situation? Firstly, we can neglect the possibility of mathematically representing a quantum jump through an $``$-valued stochastic process. This solution is doubtlessly the sharpest, but in our opinion too restrictive. Secondly, it can be claimed that the quantum jump is not real, i.e., it does not take place and that the evolution of a quantum system, though stochastic, is continuous. This is the hypothesis adopted, e.g., in the CSL theory (cf. ). Nonetheless, it is also possible that the Markov condition not be satisfied even maintaining continuity in the evolution, as in, e.g., . Finally, the option is left of admitting every hypothesis in the theorem with the subsequent consequences. This alternative has not been studied profoundly yet.
## Acknowledgements
One of us (D.S.) must acknowledge the support of the CAM Education Council under grant BOCAM-20/8/99.
## 6 Appendix
To establish (3) we must apply the continuity hypothesis, the Markov property, Fatou lemma and the definition of $`u(\varphi ,\psi )`$ in the following way:
$`P(t;\varphi ,\psi )`$ $`=`$ $`\underset{n\mathrm{}}{lim\; inf}P(t_n^{}+t;\varphi ,\psi )=`$
$`=`$ $`\underset{n\mathrm{}}{lim\; inf}{\displaystyle _{}}P(t_n^{};\varphi ,\phi )P(t;\phi ,\psi )\mu (d\phi )`$
$``$ $`{\displaystyle _{}}u^{}(\varphi ,\phi )P(t;\phi ,\psi )\mu (d\phi )`$
Since this is fulfilled for all $`t(0,\mathrm{})`$, it will be in particular satisfied for each $`t_n`$, which enables us to write, making use again of Fatou lemma:
$$\underset{n\mathrm{}}{lim\; inf}P(t_n;\varphi ,\psi )=u(\varphi ,\psi )_{}u^{}(\varphi ,\phi )u(\phi ,\psi )\mu (d\phi )$$
which is the sought relation.
To settle eq. (4) we must show a few partial results:
Applying the continuity hypothesis, the Markov property and the dominated convergence theorem, we obtain for all $`t^+`$:
$`P(t;\varphi ,\psi )`$ $`=`$ $`\underset{n\mathrm{}}{lim}P(t+t_n;\varphi ,\psi )=`$
$`=`$ $`\underset{n\mathrm{}}{lim}{\displaystyle _{}}P(t;\varphi ,\phi )P(t_n;\phi ,\psi )\mu (d\phi )=`$
$`=`$ $`{\displaystyle _{}}P(t;\varphi ,\phi )u(\phi ,\psi )\mu (d\phi )`$
Making use of property (2) of stochastic matrices, of the just obtained result and of Fubini theorem, we may write:
$`1`$ $`=`$ $`{\displaystyle _{}}P(t;\varphi ,\psi )\mu (d\psi )=`$
$`=`$ $`{\displaystyle _{}}\left[{\displaystyle _{}}P(t;\varphi ,\phi )u(\phi ,\psi )\mu (d\phi )\right]\mu (d\psi )=`$
$`=`$ $`{\displaystyle _{}}P(t;\varphi ,\phi )\left[{\displaystyle _{}}u(\phi ,\psi )\mu (d\psi )\right]\mu (d\phi )`$
whence it is deduced that if $`_{}u(\phi ,\psi )\mu (d\psi )<1`$, then $`P(t;\varphi ,\phi )=0`$ a.e. for all $`\varphi `$ and for all $`t^+`$. In particular, it is satisfied for all $`t_n^{}`$, then
$$\text{If }_{}u(\phi ,\psi )\mu (d\psi )<1,\text{then }u^{}(\varphi ,\phi )=0\text{ a.e. }\varphi .$$
Again using the same techniques as before, we may write:
$`u^{}(\varphi ,\psi )`$ $`=`$ $`\underset{n\mathrm{}}{lim\; inf}P(t_n^{};\varphi ,\psi )=`$
$`=`$ $`\underset{n\mathrm{}}{lim\; inf}{\displaystyle _{}}P(t_n^{};\varphi ,\phi )u(\phi ,\psi )\mu (d\phi )`$
$``$ $`{\displaystyle _{}}u^{}(\varphi ,\phi )u(\phi ,\psi )\mu (d\phi )`$
With these and again the already used results, we arrive at
$`{\displaystyle _{}}u^{}(\varphi ,\psi )\mu (d\psi )`$ $``$ $`{\displaystyle _{}}\left[{\displaystyle _{}}u^{}(\varphi .\phi )u(\phi ,\psi )\mu (d\phi )\right]\mu (d\psi )=`$
$`=`$ $`{\displaystyle _{}}u^{}(\varphi ,\phi )\left[{\displaystyle _{}}u(\phi ,\psi )\mu (d\psi )\right]\mu (d\phi )=`$
$`=`$ $`{\displaystyle _{}}u^{}(\varphi ,\phi )\mu (d\phi )`$
whence straightforwardly it is obtained the second sought relation.
For the sufficiency, we present the deduction of continuity of $`P(t;\varphi ,\psi )`$ from its measurability in the form of a
###### Th 3
Let $`P(t;\varphi ,\psi )`$ be measurable for all $`\varphi ,\psi `$, then $`P(t;\varphi ,\psi )`$ is continuous in $`(0,\mathrm{})`$.
The proof will be established by parts. Firstly we shall show that the expression
$$_{}\left|P(t+h;\varphi ,\psi )P(t;\varphi ,\psi )\right|\mu (d\psi )$$
(6)
is a non-increasing function of $`t`$. Secondly, we will show that (6) converges uniformly to $`0`$ when $`h0`$, where $`t\delta >0`$. Thus it is elementarily deduced that $`P(t;\varphi ,\psi )`$ is uniformly continuous in $`[\delta ,\mathrm{})`$ for all $`\delta >0`$ or equivalently in $`(0,\mathrm{})`$. 1) Using Markov property and Fubini theorem it is showed that for $`0<s<t`$
$$_{}\left|P(t+h;\varphi ,\psi )P(t;\varphi ,\psi )\right|\mu (d\psi )=$$
$$=_{}\left|_{}\left[P(s+h;\varphi ,\phi )P(s;\varphi ,\phi )\right]P(ts;\phi ,\psi )\mu (d\phi )\right|\mu (d\psi )$$
$$_{}\left|P(s+h;\varphi ,\phi )P(s;\varphi ,\phi )\right|\mu (d\phi )_{}P(ts;\phi ,\psi )\mu (d\psi )=$$
$$_{}\left|P(s+h;\varphi ,\phi )P(s;\varphi ,\phi )\right|\mu (d\phi )$$
And the first part of the proof is established. 2) If now the measurability hypothesis of $`P(t;\varphi ,\psi )`$ is introduced, we can integrate between $`0`$ and $`\delta t`$ to obtain the bound
$$_{}\left|P(t+h;\varphi ,\psi )P(t;\varphi ,\psi )\right|\mu (d\psi )$$
$$_{}\frac{1}{\delta }\left[_0^\delta \left|P(s+h;\varphi ,\phi )P(s;\varphi ,\phi )\right|𝑑s\right]\mu (d\phi )$$
where if $`0h\delta `$ the second term is dominated by
$$_{}\frac{2}{\delta }\left[_0^{2\delta }P(s;\varphi ,\phi )𝑑s\right]\mu (d\phi )$$
hence uniform convergence is established. Now, it is elementary to show that for each $`\varphi ,\psi `$
$`\underset{h0}{lim}{\displaystyle _0^\delta }\left|P(s+h;\varphi ,\psi )P(s;\varphi ,\psi )\right|𝑑s`$ $`=`$ $`0`$
which together with the uniform convergence implies the second part of the proof.
The properties of $`g(\varphi ,\psi )`$ referred in the text are easily deduced from the several relations obtained throughout the proof.
|
warning/0004/cond-mat0004219.html
|
ar5iv
|
text
|
# Properties of the asymptotic 𝑛𝐴+𝑚𝐵→𝐶 reaction-diffusion fronts.
## 1 Introduction
Consider two reactants, initially separated, which are put in contact at time $`t=0`$ and start to mix one into each other by diffusion. A region where the reaction rate is high will develop at their interface. The mathematical function describing the variation, in space and time, of the amplitude of the reaction rate in this region is usually called a reaction front $`R(𝐱,t)`$.
Dynamical properties of reaction fronts in (purely or effectively) one-dimensional reaction-diffusion systems but that reduce to an effectively of a symmetry in the initial of numerous studies GalfiRacz ; ChopCorDroz ; Cornell1 ; Cornell2 ; Cornell3 ; Cornell4 ; Koza1 ; Koza2 ; Larralde ; usSemiperm ; Bazant1 ; Bazant2 . In general, it is observed that these fronts obey asymptotic scaling, characterized by a scaling function $`G`$ and scaling exponents $`\alpha ,\gamma `$ :
$$Rt^\gamma \mathrm{\Phi }\left(\frac{xx_f(t)}{t^\alpha }\right).$$
(1)
In the previous expression, $`x_f(t)`$ locates the position of the front (usually defined as its first moment), which generally obeys
$$x_f(t)\sqrt{t},$$
(2)
accounting for the diffusive origin of the front’s dynamics.
In the framework of a mean-field approximation (which is ours from now on), scaling hypothesis, together with balance considerations, can lead quite directly to the values of the dynamical scaling exponents (see GalfiRacz ; ChopCorDroz ). Accessing the structure (i.e. the shape) of the scaling function itself requires however to go one level deeper into the analysis of the process in consideration.
In addition to satisfying a purely theoretical curiosity, knowing better $`\mathrm{\Phi }`$ itself can provide a practical advantage. For example, in studying phenomena involving reaction-diffusion processes, like Liesegang patterns formation, it may be useful to dispose of an explicit, approximated analytical form for $`\mathrm{\Phi }`$. This allows, for example, to bypass the dynamical generation of this front in a numerical simulation and save computation time usmodelB . Finding such an analytical approximation requires evidently to gain sufficient information about the scaling function.
Both tasks (derivation of the scaling exponents and of the scaling function) were accomplished in the pioneering paper by Gálfi and Rácz GalfiRacz , where they studied the reaction front in the $`A+B\stackrel{k}{}C`$ process with initially segregated $`A`$-s and $`B`$-s in mean-field approximation.
In the present paper, we provide a generalization of their work to the case of arbitrary reaction-order kinetics, $`nA+mB\stackrel{k}{}C`$, and explain how to calculate the associated $`C`$ density profile in the asymptotic regime.
An important motivation for this generalization is the following : in the case of Liesegang patterns, the primary chemical reaction leading (through several complex coarsening processes) to the formation of precipitate turns out to be most often of the types $`A+2B`$ or $`2A+B`$, and not $`A+B`$, as usually considered for simplicity in theoretical models usMP ; usmodelB ; uswidth .
## 2 Scaling analysis.
### 2.1 Definition and notations.
The case of mean-field, general reaction-order kinetics in the initially segregated reactants case has already been addressed in ChopCorDroz . Using scaling analysis, the authors showed that the exponents controlling the asymptotic behaviour of the reaction front are given, in terms of the reaction-order constants, by :
$$\alpha (n,m)=\frac{n+m1}{2(n+m+1)},\gamma (n,m)=\frac{1}{n+m+1}$$
(3)
where $`\alpha ,\gamma `$ are the same as in (1), and $`(n,m)`$ are integers both $`1`$.
It is important to note the following properties of these exponents :
1. $`\alpha (n,m)=\alpha (n+m)`$, $`\gamma (n,m)=\gamma (n+m)`$.
2. $`\alpha (n,m)<1/2(n,m)`$, $`\alpha `$ increases monotonically from $`1/6`$ $`(n=m=1)`$ to $`1/2`$ $`(n+m\mathrm{})`$.
3. $`\alpha (n,m)+\gamma (n,m)=1/2`$.
We can start from the above results to formulate a general derivation that will lead us to the family of ordinary differential equations defining the asymptotic shape of the reaction fronts $`R_{(n,m)}`$ <sup>1</sup><sup>1</sup>1This derivation follows closely the steps and notation in GalfiRacz , and the reader should refer to it for further details and justifications.. To this goal, let us consider the following one dimensional initial-value problem, describing a reaction-diffusion process between initially separated $`A`$ and $`B`$ particles in the mean-field approximation :
$`_TA(X,T)`$ $`=`$ $`D_A_X^2A(X,T)kn(A^nB^m)(X,T),`$ (4a)
$`_TB(X,T)`$ $`=`$ $`D_B_X^2B(X,T)km(A^nB^m)(X,T),`$ (4b)
$`_TC(X,T)`$ $`=`$ $`kA^n(X,T)B^m(X,T),`$ (4c)
with
$`A(X,T=0)`$ $`=`$ $`a_0\theta (X),`$ (5a)
$`B(X,T=0)`$ $`=`$ $`b_0\theta (X),`$ (5b)
$`C(X,T=0)`$ $``$ $`0.`$ (5c)
In the above equations,
* $`\theta `$ denotes the Heaviside step function \[$`\theta (X<0)=0,\theta (X0)=1`$ \].
* $`A`$, $`B`$ and $`C`$ are concentrations with dimensions $`[A,B,C]=X^1`$.
* $`D_A`$ and $`D_B`$ are diffusion coefficients ($`[D_A,D_B]=X^2T^1`$).
* $`k`$ is the reaction rate ($`[k]=[X^{n+m1}T^1]`$).
In the following, we will only consider the case of equal diffusion coefficients, $`D_A=D_BD`$, since the method we are going to use requires this strong condition to be satisfied. The asymmetric case $`D_AD_B`$ reveals to be several orders of magnitude higher in difficulty. Interesting results have been obtained in the case $`n=m=1`$, in connection to the front’s dynamics Koza1 ; Koza2 , but a derivation of the shape of the scaling functions for arbitrary $`D_A/D_B`$ seems to be still out of reach for the moment.
The first step in our calculation is to render a-dimensional the problem we are dealing with. This can be done through the following change of variables :
$`x`$ $``$ $`\sqrt{{\displaystyle \frac{ka_0^{n+m1}}{D}}}X,`$ (6a)
$`t`$ $``$ $`ka_0^{n+m1}T,`$ (6b)
$`a,b,c`$ $``$ $`A/a_0,B/b_0,C/c_0.`$ (6c)
The equations then read :
$`_ta(x,t)`$ $`=`$ $`_x^2a(x,t)na^n(x,t)b^m(x,t),`$ (7a)
$`_tb(x,t)`$ $`=`$ $`_x^2b(x,t)ma^n(x,t)b^m(x,t),`$ (7b)
$`_tc(x,t)`$ $`=`$ $`a^n(x,t)b^m(x,t),`$ (7c)
with
$`a(x,t=0)`$ $`=`$ $`\theta (x),`$ (8a)
$`b(x,t=0)`$ $`=`$ $`{\displaystyle \frac{b_0}{a_0}}\theta (x),`$ (8b)
$`c(x,t=0)`$ $``$ $`0.`$ (8c)
### 2.2 Solution for $`a(n/m)b`$.
We define
$$u(x,t)\left(a\frac{n}{m}b\right)(x,t).$$
(9)
This function obeys the diffusion equation :
$`_tu(x,t)`$ $`=`$ $`_x^2u(x,t),`$ (10a)
$`u(x<0,t=0)`$ $`=`$ $`1,`$ (10b)
$`u(x>0,t=0)`$ $`=`$ $`{\displaystyle \frac{n}{m}}{\displaystyle \frac{b_0}{a_0}}{\displaystyle \frac{n}{m}}q,`$ (10c)
whose solution reads
$$u(x,t)=\frac{1}{2}\left((1\frac{n}{m}q)(1+\frac{n}{m}q)\mathrm{erf}\left(\frac{x}{2\sqrt{t}}\right)\right).$$
(11)
In the above equation, $`\mathrm{erf}`$ denotes the error function, $`\mathrm{erf}(x)(2/\sqrt{\pi })_0^x\mathrm{exp}(w^2)𝑑w`$.
Let $`x_f(t)`$ be such that $`u(x_f(t),t)=0`$. One can check that
$$x_f(t)=\sqrt{2D_ft},$$
(12)
with $`D_f=D_f(q)`$ given by
$$\mathrm{erf}\left(\sqrt{\frac{D_f}{2}}\right)=\frac{1\frac{n}{m}q}{1+\frac{n}{m}q}.$$
(13)
### 2.3 Equation for $`a`$ in the reaction zone.
We write now $`b=\frac{m}{n}(au)`$ and plug it into (7a), getting thereby an equation for $`a`$ involving only $`a`$ and the known function $`u`$ :
$$_ta(x,t)=_x^2a(x,t)n\left(\frac{m}{n}\right)^m\left[a^n(au)^m\right](x,t).$$
(14)
We are interested in the solution of this equation in the reactive region $`|xx_f|t^{\alpha (n,m)}`$. As the latter is believed to widen with a time exponent $`\alpha (n,m)<1/2`$ , this allows us to expand $`u`$ around $`x_f`$ to the lowest-order in $`x/\sqrt{t}`$, since the neglected terms will vanish as $`t\mathrm{}`$ :
$$u(x,t)K\frac{xx_f}{\sqrt{t}}|xx_f|t^{\alpha (n,m)},$$
(15)
with $`K`$ given by
$$K=\frac{1+\frac{n}{m}q}{2\sqrt{\pi }}\mathrm{exp}(D_f/2).$$
(16)
The boundary conditions that the solution to (14) must satisfy in the reactive region are :
$`a(x\mathrm{},t)`$ $`=`$ $`K{\displaystyle \frac{xx_f}{\sqrt{t}}},`$ (17a)
$`a(x+\mathrm{},t)`$ $`=`$ $`0.`$ (17b)
### 2.4 Scaling hypothesis.
We shall now assume that asymptotically (i.e. when $`t\mathrm{}`$), the solution to (14) adopts the following scaling form :
$$a(x,t)t^{\gamma (n,m)}G_{(n,m)}\left(\frac{xx_f(t)}{t^{\alpha (n,m)}}\right),$$
(18)
where $`\{G_{(n,m)}\}_{n,m1}`$ are a family of scaling functions remaining to be characterized. The scaling exponents are given by (3).
### 2.5 Differential equation for $`G_{(n,m)}`$.
Let’s define first the reaction zone coordinate $`z`$
$$z\frac{xx_f}{t^{\alpha (n,m)}}.$$
(19)
Inside the reaction zone, $`u`$ and $`b=\frac{m}{n}(au)`$ write
$$u(z)=Kt^{\alpha (n,m)1/2}z,$$
(20)
$$b(z)=\frac{m}{n}(t^{\gamma (n,m)}[G_{(n,m)}(z)+Kz]).$$
(21)
Using (18), eq. (14) becomes :
$$\begin{array}{cc}& t^{2\alpha (n,m)1}[\gamma (n,m)G_{(n,m)}\alpha (n,m)z_zG_{(n,m)}]\hfill \\ & \sqrt{\frac{D_f}{2}}t^{\alpha (n,m)1/2}_zG_{(n,m)}\hfill \\ & =_z^2G_{(n,m)}n\left(\frac{m}{n}\right)^mG_{(n,m)}^n[G_{(n,m)}+Kz]^m.\hfill \end{array}$$
(22)
We now take the asymptotic limit inside the reaction zone, i.e. we let $`t\mathrm{}`$, keeping $`z`$ fixed. The two terms on the left-hand side vanish (remember that $`\alpha _{(n,m)}<1/2`$ !) and we remain with the following ordinary, non-linear second-order differential equation for the scaling functions $`G_{(n,m)}`$ :
$$G_{(n,m)}^{\prime \prime }(z)=n\left(\frac{m}{n}\right)^mG_{(n,m)}^n(z)[G_{(n,m)}(z)+Kz]^m.$$
(23)
The boundary conditions (17a17b) imply the following asymptotics for $`G_{(n,m)}`$ GalfiRacz :
$`G_{(n,m)}(z)`$ $``$ $`Kz,z\mathrm{},`$ (24a)
$`G_{(n,m)}(z)`$ $``$ $`0,z\mathrm{}.`$ (24b)
We are now left with a boundary-value problem (23,24) that can be solved numerically.
### 2.6 Solving the equation for $`G_{(n,m)}`$.
We can make the problem K-independent by rescaling $`G`$ and $`z`$ :
$`G`$ $``$ $`K^{\mu (n,m)}\stackrel{~}{G},`$ (25a)
$`z`$ $``$ $`K^{\nu (n,m)}\stackrel{~}{z},`$ (25b)
and by using a suitable choice for $`\mu `$ and $`\nu `$. Inserting these scaled forms into (23) and imposing that K drops out leads to
$`\mu (n,m)`$ $`=`$ $`{\displaystyle \frac{2}{n+m+1}}=\mu (n+m),`$ (26a)
$`\nu (n,m)`$ $`=`$ $`{\displaystyle \frac{(n+m1)}{n+m+1}}=\nu (n+m).`$ (26b)
The problem we are left to treat is now :
$`\stackrel{~}{G}_{n,m}^{\prime \prime }(\stackrel{~}{z})`$ $`=`$ $`n\left({\displaystyle \frac{m}{n}}\right)^m\stackrel{~}{G}_{n,m}^n(\stackrel{~}{z})\left[\stackrel{~}{G}_{n,m}(\stackrel{~}{z})+\stackrel{~}{z}\right]^m,`$ (27a)
$`\stackrel{~}{G}_{n,m}(\stackrel{~}{z})`$ $``$ $`\stackrel{~}{z}\stackrel{~}{z}\mathrm{},`$ (27b)
$`\stackrel{~}{G}_{n,m}(\stackrel{~}{z})`$ $``$ $`0\stackrel{~}{z}\mathrm{}.`$ (27c)
The reader should keep in mind, from now on, that the “tilde” sign stands for quantities expressed in terms of the rescaled, $`K`$independent version of the scaling function $`G_{(n,m)}`$ and reactive coordinate $`z`$.
### 2.7 The dimensionless reaction front.
By definition :
$`R_{(n,m)}(x,t)`$ $`=`$ $`a^n(x,t)b^m(x,t)`$ (28)
$`=`$ $`\left({\displaystyle \frac{m}{n}}\right)^mt^{\frac{n+m}{n+m+1}}G_{(n,m)}^n[G_{(n,m)}+Kz]^m`$
$``$ $`t^{\beta (n,m)}F_{(n,m)}(z),`$
the last inequality defining both the reaction rate amplitude exponent $`\beta `$ and the asymptotic reaction front scaling function $`F_{(n,m)}`$.
Figure 1 shows the result of the numerical computation of the fronts $`\stackrel{~}{F}_{(n,m)}F(\stackrel{~}{G},\stackrel{~}{z})`$ for the cases $`2n+m4`$ (only such low values of $`n+m`$ are relevant in connection to experiments). The reader should not be surprised by the asymmetry between the $`(n,m)=(1,2)`$ and $`(n,m)=(2,1)`$ cases : it is due to the passage from the variables $`\{G_{(n,m)},z\}`$ to $`\{\stackrel{~}{G}_{(n,m)},\stackrel{~}{z}\}`$, since both quantities are rescaled by a $`K`$ dependent factor which is not $`(n,m)`$-symmetric !
These fronts are all essentially localized around $`z=0`$, as Fig. 1 suggests. In fact, one can check that in the simplest $`(n,m)=(1,1)`$ case, the dominant contribution to both tails at $`z\pm \mathrm{}`$ is proportional to $`\mathrm{exp}(z^{3/2})`$. In the cases $`n=1,m>1`$, the decay can be shown to be dominated by $`\mathrm{exp}(z^{1+m/2})`$ at $`z\mathrm{}`$ but only algebraically for $`z\mathrm{}`$ : $`R(z\mathrm{})z^{(2m+1)/(m1)}`$. Finally, when $`n>1,m>1`$, both tails are algebraic, with $`R(z\mathrm{})z^{(2m+n)/(n1)}`$.
It is also worth noting that in the asymmetric cases $`nm`$, $`x_f`$ does not coincide with the maximum of the front.
## 3 The $`C`$ concentration profile.
### 3.1 Derivation of the asymptotic profile.
We are interested now in estimating the (possibly x-dependent) density $`c_0^{(n,m)}(x)`$ of $`C`$ particles left behind by the fronts $`R_{(n,m)}`$ which travel diffusively through the system. This quantity is, for example, of great importance in the theories of Liesegang pattern formation usMP ; usmodelB .
In dimensionless units, $`c_0^{(n,m)}`$ is formally given by
$$c_0^{(n,m)}(x)\underset{0}{\overset{\mathrm{}}{}}R_{(n,m)}(x,t)𝑑t.$$
(29)
Due to the several timescales dependence of $`R_{(n,m)}`$, this integral is difficult to handle. The estimation of $`c_0^{(n,m)}(x)`$ turns out however to be possible by making use of
1. the precious algebraic relation $`\alpha +\gamma =1/2`$ between the scaling exponents, and
2. the particular structure of the solutions to (27).
Let’s consider first a narrow slice $`\delta F_{(n,m)}(z_0,\delta w)`$ of the scaling function $`F_{(n,m)}`$, centered on $`z=z_0`$, of width $`\delta w1`$. In the spirit of the Riemann integral, we can approximate the amplitude of $`\delta F`$ inside $`[z_0\delta w/2,z_0+\delta w/2]`$ by its value $`F_{(n,m)}(z_0)`$ at the center.
We can estimate the contribution $`\delta C^{(n,m)}(z_0,x)/\delta x`$ of this slice to the asymptotic local $`C`$ density inside $`[x,x+\delta x]`$ as follows. The fraction of the front we are considering will reach $`x`$ at a certain time $`t(z_0,x)`$. The quantity of $`C`$ particles deposited in the interval $`[x,x+\delta x]`$ will be proportional to the amplitude, the width and inversely proportional to the speed of the slice at $`t=t(z_0,x)`$ :
$`\delta C^{(n,m)}(z_0,x)`$ (30)
$`{\displaystyle \frac{t(z_0,x)^{\gamma (n,m)}F_{(n,m)}(z_0)t(z_0,x)^{\alpha (n,m)}\delta w}{\sqrt{D_f/(2t(z_0,x))}}}\delta x`$
$`=`$ $`\sqrt{2/D_f}F_{(n,m)}(z_0)\delta w\delta x.`$
In other words, the contribution of the slice to the density at $`x`$ is proportional to its ”mass” $`F_{(n,m)}(z_0)\delta w`$ but independent of $`t(z_0,x)`$, and hence of $`x`$. This indicates that the asymptotic density profile is flat. By superposition, our argument leads immediately to the result we are looking for :
$`c_0^{(n,m)}(x)`$ $``$ $`c_0^{(n,m)}=\mathrm{const}.`$ (31)
$``$ $`\sqrt{2/D_f}{\displaystyle _{\mathrm{}}^{\mathrm{}}}F_{(n,m)}(z)𝑑z.`$
Now our real fortune is that we are able to evaluate analytically $`F_{(n,m)}`$. Using (23) and (28), We have
$$\begin{array}{cc}\hfill _{\mathrm{}}^{\mathrm{}}F_{(n,m)}(z)dz=\frac{1}{n}[& _zG_{(n,m)}(z\mathrm{})\hfill \\ & _zG_{(n,m)}(z\mathrm{})].\hfill \end{array}$$
(32)
It is intuitively clear, from purely physical considerations, that the solution will converge to its values at $`\pm \mathrm{}`$ in such a way that
$$\underset{z\mathrm{}}{lim}G_{(n,m)}^{}(z)=K,\underset{z+\mathrm{}}{lim}G_{(n,m)}^{}(z)=0$$
(33)
\[see (27b)–(27c)\]. So we finally have
$$_{\mathrm{}}^{\mathrm{}}F_{(n,m)}(z)𝑑z=\frac{K}{n},$$
(34)
and we end up with the result
$$c_0^{(n,m)}\frac{K}{n}\sqrt{2/D_f}.$$
(35)
Going back to the dimensional variables $`A,B,C,X`$ and $`T`$, one can check that (35) writes :
$$c_0^{(n,m)}\sqrt{\frac{2D}{D_f}}\frac{Ka_0}{n}.$$
(36)
Fig. 2 shows the $`c_0^{(n,m)}`$ profiles obtained by numerical integration of the reaction-diffusion equations (4), together with the asymptotic values predicted by (36).
### 3.2 Low $`q`$ expansion.
In the context of Liesegang patterns-forming experiments, which are generally based on the penetration of a highly concentrated solution into a dissolved one, it is useful to dispose of an expansion of this result for low $`qb_0/a_0`$ values. We recall that $`D_f`$ is given by
$$\mathrm{Erf}(\sqrt{\frac{D_f}{2D}})=\frac{1\frac{n}{m}q}{1+\frac{n}{m}q}=12\frac{n}{m}q\stackrel{q1,n2}{}1.$$
(37)
The large $`x`$ asymptotics of $`\mathrm{Erf}(x)`$ is given by GradRyzh :
$$\mathrm{Erf}(x)1\frac{e^{x^2}}{\sqrt{\pi }x}\left[1\frac{1}{2x^2}+\mathrm{}\right],$$
(38)
so we obtain from (37) :
$$\begin{array}{cc}\hfill \sqrt{\frac{D}{D_f}}\mathrm{exp}(D_f/2D)& [1\frac{1}{2}\frac{D}{D_f}]^1\hfill \\ & \sqrt{2\pi }\frac{\frac{n}{m}q}{1+\frac{n}{m}q},\hfill \end{array}$$
(39)
which, together with the dimensional expression for $`K`$,
$$K=\frac{1+\frac{n}{m}q}{2\sqrt{\pi }}\mathrm{exp}(D_f/2D),$$
(40)
gives finally
$$c_0^{(n,m)}\frac{b_0}{m}\left[1+\frac{1}{2}\frac{D}{D_f}+𝒪\left((\frac{D}{D_f})^2\right)\right].$$
The physical meaning of this result is clear: if $`D_A=D_B=D`$ and $`q1`$, then $`D_fD`$ and the $`B`$ particles appear as nearly immobile for the invading $`A`$-s. As $`m`$ $`B`$-s are required to produce one $`C`$, the density equals $`b_0/m`$ to a very good approximation.
However, in the typical conditions of a Liesegang experiment ( where $`D_AD_B`$ and $`10^2q510^2`$ typically), the first order correction in $`D/D_f`$ to $`c_0^{(n,m)}`$ lies in the range $`(0.10.2)(b_0/m)`$, and should therefore, in principle, not be neglected, as can be seen of Fig. 2.
## 4 Summary.
We have derived the family of ordinary differential equations defining the asymptotic shape of the reaction fronts in the $`nA+mB\stackrel{k}{}C`$ reaction-diffusion process with initially separated reactants \[eqs. (23),(24)\]. The four lowest-order cases in $`n+m`$ have been solved numerically (Fig. 1). We have also shown, and confirmed by numerical simulations, that the density $`c_0^{(n,m)}`$ of $`C`$ particles deposited in the system by these traveling fronts is asymptotically constant (Fig. 2), and we have made explicit the dependence of this density on the reaction orders $`n,m`$, as well as on the material parameters $`D,a_0,k`$ and $`b_0`$ entering the problem \[eq. (36)\].
## 5 Conclusion.
To conclude this study, we would like to comment on the interesting phenomenon shown by Fig. 2 : when, for the two $`m=1`$ cases, no significant quantity of reaction product is created in the majority species subspace, the case where $`m=2`$ exhibits, on contrary, an important deposit of $`C`$ on the left hand side, up to far beyond the initial location of the interface. This fact must be evidently related to the details of the short-time dynamics of the reaction front. Some studies have already been carried on the early-time regime subject for the $`n=m=1`$ case Taitelbaum1 ; Taitelbaum2 in the past. They unveiled the existence of a surprisingly complex behaviour, including successive power-law regimes for the early front’s dynamics, and even the possibility of a change in its direction of motion. Such nontrivial behaviour has also been observed numerically in the higher-order kinetics cases we have addressed in the present paper, and a detailed study of the dependence on $`(n,m)`$ of the short-time dynamics should be the object of a forthcoming paper.
## 6 Acknowledgments
I thank Michel Droz for useful remarks.
|
warning/0004/hep-th0004128.html
|
ar5iv
|
text
|
# Untitled Document
NONLOCAL REGULARIZATION FOR NON-ABELIAN
GAUGE THEORIES
FOR ARBITRARY GAUGE PARAMETER
Anirban Basu
and
Satish D. Joglekar
Department of Physics
I.I.T. Kanpur
Kanpur 208016, INDIA
ABSTRACT
We study the nonlocal regularization for the non-abelian gauge theories for an arbitrary value of the gauge parameter $`\xi `$. We show that the procedure for the nonlocalization of field theories established earlier by the original authors, when applied in that form to the Faddeev-Popov effective action in a linear gauge cannot lead to a $`\xi `$-independent result for the observables. We then show that an alternate procedure which is simpler can be used and that it leads to the S-matrix elements (where they exist) independent of $`\xi `$.
1.INTRODUCTION
Local Quantum Field Theories are plagued with infinities and need regularization to make the process of renormalization mathematically well-defined. Many regularizations have been proposed over the last 50 years, dimensional regularization being one used widest due to its effectiveness. While dimensional regularization is useful in a wide class of Quantum Field Theories, it cannot be used directly in Supersymmetric Field Theories. A number of regularizations have been proposed over the last decade that can be used in Supersymmetric Field Theories. Nonlocal regularization is one of them.
Nonlocal regularization proposed by Evans et al has been extensively studied\[$`4,5,6`$\]. Renormalization procedure has been established upto two loop order in scalar theories. The scheme has found an elegant and neat formulation in reference 4 which has shown how nonlocally regularized field theories can be constructed from a local QFT in a systematic fashion. More importantly, it has been established that local/global symmetries can be preserved in their nonlocal form and the WT identities of local QFT’s derivable from local symmetries such as gauge invariance/BRS symmetry find their natural nonlocal extensions. This has been done for the abelian gauge theories to all orders and for nonabelian gauge theories in Feynman gauge upto one loop order \[limited only by the existence of measure beyond one loop\].
Nonlocally regularized theories have also found other equally useful
interpretations. Nonlocally regularized theories contain in them a large mass parameter $`\mathrm{\Lambda }`$. It has been shown (wherever the measure factor exists) that these theories are unitary even with a finite $`\mathrm{\Lambda }`$. Discussions of causality and renormalization group have also been carried out. Thus it has been suggested such nonlocally regularized theories with a finite $`\mathrm{\Lambda }`$ can themselves be looked upon as valid physical theories (rather than a regularization for which $`\mathrm{\Lambda }\mathrm{}`$ must be taken). The parameter $`\mathrm{\Lambda }`$ has been interpreted in two ways: (a) as a signal of an underlying space-time granularity; (b) as the mass scale beyond which the physical theory must be replaced by another, more fundamental theory. We may regard view (a) as a mathematically convenient way of embodying space-time granularity in QFT’s in a way that is physically consistent. In view (b), we may regard the nonlocal QFT as an effective field theory that may have been derived from a more fundamental theory beyond the scale $`\mathrm{\Lambda }`$. Thus, for example, we regard nonlocal standard model as the effective theory of fundamental processes at present energies, in which a signature of physics beyond standard model and the scale at which the SM should break down are both implicit in the scale $`\mathrm{\Lambda }`$. An attempt to put lower bound using (g-2) of the muon has been made in Ref. 8.
The setting of such nonlocal QFT’s has also been used to understand renormalization program in a mathematically rigorous way. A way to put an upper bound on $`\mathrm{\Lambda }`$ has also been suggested.
Nonlocal regularization has also found use in the discussion of higher loop anomalies in BV formulation.
In view of the above, it seems valuable to study these formulations further. One of the features of linear gauges in local gauge theories is the availability of a free parameter $`\xi `$ (gauge parameter) which helps in verifying the gauge independence of physical results. $`\xi `$-independence of physical results in spontaneously broken gauge theories has also been used to establish the cancellation of contributions from the unphysical poles to the cutting equations in SBGT.
It is therefore desirable that we have a formulation of non-local nonabelian gauge theories valid for an arbitrary $`\xi `$. Now, a well laid-out procedure for the nonlocalization of field theories has been presented in references 4 and 5. We found however that when we applied this procedure to the spontaneously broken theory (SM) in $`R_\xi `$ gauges and calculated the (g-2) for the muon we found a $`\xi `$-dependent result . This motivated us to look into the question of nonlocal formulation of unbroken nonabelian gauge theories and of spontaneously broken chiral abelian gauge theories. In the present work, we concern ourselves with the former.
We now discuss the plan of our work. In Section II, we summarize the results on the nonlocal quantum field theories of 2, 4, 5. In Section III, we adopt the procedure outlined in 4 for nonlocalization for arbitrary $`\xi `$ and evaluate $`\xi \frac{dW}{d\xi }|_{\xi =1}`$for this case. We obtain a term in $`\xi \frac{dW}{d\xi }|_{\xi =1}`$ that $`\mathrm{𝐜𝐚𝐧}`$ contribute to on-shell physical processes and which cannot be cancelled by a $`\xi `$ dependent measure. In Section IV, we suggest an alternate way of constructing nonlocal unbroken gauge theories for an arbitrary $`\xi `$ and establish the WT identity satisfied by the physical Green’s function $`\xi \frac{dW}{d\xi }|_{J=J_{phy}}`$ of eqn. (4.15). This equation is analogous to that in that in the local case; and should lead to the $`\xi `$-independence of physical quantities that are free of infra-red divergences.
II REVIEW OF KNOWN RESULTS
A Nonlocal Regularization
Let us briefly review the method of non-local regularization as proposed in . Let $`\varphi _i`$ stand for a generic, not necessarily scalar, field and let us assume that the local action can be written as a standard free part plus an interaction
S\[$`\varphi `$\] = F\[$`\varphi `$\] + I\[$`\varphi `$\] , (2.1)
where
F\[$`\varphi `$\] = $`\frac{1}{2}`$ $``$ $`d^Dx`$ $`\varphi _i`$$`(x)`$ $`_{ij}`$ $`\varphi _j`$$`(x)`$ (2.2)
Here, $``$ is the kinetic energy operator for the field $`\varphi `$, and I\[$`\varphi `$\] is the interaction term. For unbroken gauge theories, S\[$`\varphi `$\] would be the BRS gauge fixed action and $`\varphi _i`$ would include both the fields of the invariant action and the ghosts introduced in the process of fixing the gauge.
From the kinetic energy operator $``$, let us define a non-local smearing operator $``$ and a shadow kinetic operator $`𝒪^1`$ as<sup>1</sup><sup>1</sup>1In the case of the ghost Lagrangian, its overall sign (and therefore that of the quadratic form) is arbitrary. In such a case, we shall always take the sign of $``$ such that the ghost propagator is damped for large $``$$`k^2`$$``$ in Euclidean space.
$``$ = exp\[$`\frac{}{2\mathrm{\Lambda }^2}`$\] (2.3)
and
$`𝒪`$$`\frac{^2\mathrm{\hspace{0.17em}1}}{}`$ (2.4)
Further, the smeared field is defined as
$`\widehat{\varphi }`$ = $`^1`$ $`\varphi `$ (2.5)
For every field $`\varphi `$, an auxiliary field $`\psi `$ of the same type is introduced. Then, the auxiliary action is defined to be
$`𝒮`$\[$`\varphi ,`$$`\psi `$\] = F\[$`\widehat{\varphi }`$\] - A\[$`\psi `$\] \+ I\[$`\varphi +`$$`\psi `$\] (2.6)
where
A\[$`\psi `$\] = $`\frac{1}{2}`$$``$ $`d^Dx`$ $`\psi _i`$$`(x)`$ $`𝒪_{ij}^1`$ $`\psi _j`$$`(x)`$ (2.7)
The action for the nonlocalized theory $`\widehat{S}`$\[$`\varphi `$\] is defined to be
$`\widehat{S}`$\[$`\varphi `$\] =$`𝒮`$\[$`\varphi ,`$$`\psi `$\[$`\varphi `$\]\] (2.8)
where $`\psi `$\[$`\varphi `$\] is a solution of the classical shadow field equation
$`\frac{\delta 𝒮[\varphi ,\psi ]}{\delta \psi _i}`$ = 0 (2.9)
Quantization is carried out in the path integral formulation. The quantization rule is
$`<T^{}(O[\varphi ])>_{}`$ = $`[D\varphi ]`$$`\mu `$\[$`\varphi `$\]$`O[\widehat{\varphi }]`$$`e^{i\widehat{S}[\varphi ]}`$ (2.10)
Here O is any operator taken as a functional of fields. $`\mu `$\[$`\varphi `$\] is the measure factor defined such that $`[D\varphi ]`$$`\mu `$\[$`\varphi `$\] is invariant under the nonlocal generalisation of the local symmetry. For nonlocalized non-Abelian gauge theories, this measure factor can be non-trivial and has been evaluated upto one loop . For the abelian gauge theories, the measure factor is known to all orders.
The nonlocalized Feynman rules are simple extensions of the local ones. The vertices are unchanged but every leg can connect either to a smeared propagator
$`\frac{i^2}{+iϵ}`$ = -i$`_1^{\mathrm{}}\frac{d\tau }{\mathrm{\Lambda }^2}`$$`e^{\frac{\tau }{\mathrm{\Lambda }^2}}`$ (2.11)
or to a shadow propagator
$`\frac{i(1^2)}{}`$ = -i$`𝒪`$ = -i$`_0^1\frac{d\tau }{\mathrm{\Lambda }^2}`$$`e^{\frac{\tau }{\mathrm{\Lambda }^2}}`$ (2.12)
Diagramatically, they will be represented as
the smeared or “unbarred” propagator
the shadow or “barred” propagator
The shadow propagator lacks a pole and so carries no quanta. Thus, the following points are to be noted:
i) All external lines must be unbarred.
ii) The symmetry factor for any diagram is computed without distinguishing between barred and unbarred lines.
iii) The loop integrations are well defined in the Euclidean space because of the exponential damping factors coming from propagators within loops.
iv) Internal lines can be smeared or barred. However, loops containing only the shadow lines are forbidden.
v) Tree order Green’s functions are unchanged except for external line factors which are unity on shell. This follows because every internal line of a tree graph can be either barred or unbarred. Hence, it is the sum of both these that which enters, which gives the local propagator.
B Theorems Regarding Nonlocal Regularized Actions
Before discussing the nonlocal BRS symmetries, let us consider a few theorems concerning classical solutions of the Euler-Lagrange equations associated with the local action S\[$`\varphi `$\], the auxiliary action $`𝒮`$\[$`\varphi ,\psi `$\] and the nonlocalized $`\widehat{S}`$\[$`\varphi `$\] action:
Theorem A.1: The shadow fields can be expressed as follows:
$`\psi _i[\varphi ]`$ = -$`(\frac{^21}{^2})_{ij}`$$`\varphi _j`$ \+ $`O_{ij}`$$`\frac{\delta \widehat{S}[\varphi ]}{\delta \varphi _j}`$ (2.13)
Theorem A.2: If $`\varphi _i`$ and $`\psi _i`$ obey the Euler-Lagrange equations of $`𝒮`$\[$`\varphi ,\psi `$\] then $`\chi _i=`$ $`\varphi _i`$ \+ $`\psi _i`$ obeys the Euler-Lagrange equations of S\[$`\chi `$\].
Theorem A.3: If $`\chi _i`$ obeys the Euler-Lagrange equations of S\[$`\chi `$\], then the following fields
$`\varphi _i`$ = $`_{ij}^2`$$`\chi _j`$ (2.14)
$`\psi _i=`$ $`(1^2)_{ij}`$$`\chi _j`$
obey the Euler-Lagrange equations of $`𝒮`$\[$`\varphi ,\psi `$\].
Let us also consider another set of theorems concerning classical symmetries of S\[$`\chi `$\], $`𝒮`$\[$`\varphi ,\psi `$\] and $`\widehat{S}`$\[$`\varphi `$\]:
Theorem B.1: If S\[$`\varphi `$\] is invariant under the infinisitesimal transformation
$`\delta \varphi _i=`$ $`T_i[\varphi ]`$,
then the following transformation is a symmetry of $`𝒮`$\[$`\varphi ,\psi `$\]:
$`\mathrm{\Delta }\varphi _i=`$ $`_{ij}^2`$$`T_j[\varphi +\psi ]`$ (2.15)
$`\mathrm{\Delta }\psi _i`$ = $`(1^2)_{ij}T_j[\varphi +\psi ].`$
Theorem B.3<sup>2</sup><sup>2</sup>2We stick to the theorem numbering of .: If S\[$`\varphi `$\] is invariant under $`\delta \varphi _i=`$ $`T_i[\varphi ]`$, then $`\widehat{S}`$\[$`\varphi `$\] is invariant under
$`\widehat{\delta }\varphi _i`$ = $`_{ij}^2T_j[\varphi +\psi [\varphi ]]`$ (2.16)
TheoremB.4: The following transformation generates a symmetry of $`𝒮`$\[$`\varphi ,\psi `$\]:
$`\mathrm{\Delta }\varphi _i`$ = $`A_{ij}[\varphi ,\psi ]\{\frac{\delta 𝒮[\varphi ,\psi ]}{\delta \varphi _j}\frac{\delta 𝒮[\varphi ,\psi ]}{\delta \psi _j}\}`$ (2.17)
$`\mathrm{\Delta }\psi _i`$ = -$`A_{ij}[\varphi ,\psi ]\{\frac{\delta 𝒮[\varphi ,\psi ]}{\delta \varphi _j}\frac{\delta 𝒮[\varphi ,\psi ]}{\delta \psi _j}\}`$
provided $`A_{ij}[\varphi ,\psi ]`$ = -$`A_{ji}[\varphi ,\psi ]`$. This symmetry is a ”trivial symmetry” without a dynamical content.
$`\mathrm{\Delta }\varphi _i`$ can also be cast in the simple form
$`\mathrm{\Delta }\varphi _i`$ = $`A_{ij}[\varphi ,\psi ]`$$`_{jk}^2𝒪_{kl}^1[(1^2)_{lm}\varphi _m_{lm}^2\psi _m]`$ (2.18)
An important special case is given by the choice
$`A_{il}[\varphi ,\psi ]`$ = $`M_{ij}𝒪_{jk}_{kl}^2`$
where
$`[M_{ij},_{ij}]`$ = 0 and $`M_{ij}=`$ $`M_{ji}`$.
We next review the nonlocal BRS symmetries of the Yang-Mills theory using the above results.
C Nonlocal Regularization of Yang-Mills Theory in the Feynman Gauge
Finally, let us consider the nonlocal regularization of Yang-Mills theories. First, we will study the results obtained in the Feynman gauge.
The Feynman gauge local BRS Lagrangian is
$`_{BRS}`$=-$`\frac{1}{2}`$$`_\mu A_\nu ^a^\mu A^{a\nu }`$-$`_\mu `$$`\overline{\eta }^a^\mu \eta ^a`$+$`gf^{abc}_\mu \overline{\eta }^aA^{b\mu }\eta ^c`$+$`gf^{abc}_\mu A_\nu ^aA^{b\mu }A_{c\nu }`$
-$`\frac{g^2}{4}`$$`f^{abc}f^{cde}A_\mu ^aA_\nu ^bA^{d\mu }A^{e\nu }`$ (2.19)
Thus, the gluon and the ghost kinetic energy operators are $`_{ab}^{\mu \nu }=`$ $`\delta _{ab}\eta ^{\mu \nu }^2`$ and $`_{ab}`$ = $`\delta _{ab}^2`$ respectively<sup>3</sup><sup>3</sup>3See earlier footnote above eq.(2.3)..
Let us denote the auxiliary fields of $`A_\mu ^a`$ and $`\eta ^a`$ by $`B_\mu ^a`$ and $`\psi ^a`$ respectively.
Thus the non-localized BRS action is
$`\widehat{S}`$\[A,$`\eta ,\overline{\eta }`$\]=$``$$`d^4x`${-$`\frac{1}{2}`$$`_\mu \widehat{A}_\nu ^a^\mu \widehat{A}^{a\nu }`$-$`\frac{1}{2}`$$`B_\mu ^aO^1B^{a\mu }`$-$`_\mu \widehat{\overline{\eta }}^a^\mu \widehat{\eta }^a`$+$`\psi ^aO^1\psi ^a`$}
+I\[A+B,$`\eta +\psi `$,$`\overline{\eta }+\overline{\psi }`$\] (2.20)
The local BRS Yang-Mills action in the Feynman gauge has the following BRS symmetry transformations:
$`\delta A_\mu ^a`$ = $`(_\mu \eta ^agf^{abc}A_\mu ^b\eta ^c)\delta \varsigma `$
$`\delta \eta ^a`$ = -$`\frac{g}{2}`$$`f^{abc}\eta ^b\eta ^c\delta \varsigma `$ (2.21)
$`\delta \overline{\eta }^a=_\mu A^{a\mu }\delta \varsigma `$
where $`\delta \varsigma `$ is a constant anticommuting C number.
Given the local symmetry transformations for the fields, one can easily write down their non-local counterparts:
$`\widehat{\delta }A_\mu ^a`$ = $`^2`$$`[_\mu (\eta ^a+\psi ^a)gf^{abc}(A_\mu ^b+B_\mu ^b)(\eta ^c+\psi ^c)]\delta \varsigma `$
$`\widehat{\delta }\eta ^a`$ = -$`\frac{g}{2}`$$`f^{abc}^2(\eta ^b+\psi ^b)(\eta ^c+\psi ^c)\delta \varsigma `$ (2.22)
$`\widehat{\delta }\overline{\eta }^a=^2_\mu (A^{a\mu }+B^{a\mu })\delta \varsigma `$
where
$``$ = $`e^{\frac{^2}{2\mathrm{\Lambda }^2}}`$
In , it was found that it is convenient to construct a modified nonlocal BRS symmetry transformation by adding a ”trivial” symmetry transformation to the kind (2.18). This was so as noted in since it yielded a variation of $`\overline{c}`$ proportional to $``$.A. Put alternatively, we find that these new transformations of have two useful properties (i) $``$.$`\delta `$A is directly reducible in terms of the ghost action, (ii) WT identities so formulated allow an easy evaluation of $`\xi `$$`\frac{W}{\xi }`$. The measure factor is defined with respect to the latter transformations. They are
$`\widehat{\delta }A_\mu ^a`$ = $`[_\mu \eta ^agf^{abc}^2(A_\mu ^b+B_\mu ^b)(\eta ^c+\psi ^c)]\delta \varsigma `$
$`\widehat{\delta }\eta ^a`$ = -$`\frac{g}{2}`$$`f^{abc}^2(\eta ^b+\psi ^b)(\eta ^c+\psi ^c)\delta \varsigma `$ (2.23)
$`\widehat{\delta }\overline{\eta }^a=_\mu A^{a\mu }\delta \varsigma `$
The measure factor is
ln($`\mu [A,\eta ,\overline{\eta }]`$)=-$`\frac{g^2}{2}f_{acd}f_{bcd}d^DxA_{\mu a}`$$`A_{b}^{}{}_{}{}^{\mu }`$ \+ O($`g^3`$) (2.24)
where
$``$=$`\frac{1}{2^D\pi ^{\frac{D}{2}}}_0^1𝑑\tau \frac{\mathrm{\Lambda }^{D2}}{(\tau +1)^{\frac{D}{2}}}`$ exp$`(\frac{\tau }{\tau +1}\frac{^2}{\mathrm{\Lambda }^2})[\frac{2}{\tau +1}(D1)+2(D2)\frac{\tau }{\tau +1}]`$
III Difficulty with the method of non-local regularization for an arbitrary $`\xi `$:
The above method of regularization worrks correctly in the Feynman gauge $`\xi `$=1. The regulator operators are simple and calculations have been performed in this gauge with relative ease . When this procedure of taking the entire quadratic form $``$ which enters the regulator operators is used, the regulators are $`\xi `$-dependent and complicated. This, of course, is not a serious objection to the use of this procedure in , we find that the procedure in fact leads to WT identities which imply that the S-matrix elements, where they exist, are not $`\xi `$-independent even in one loop order. In this section, we wish to demonstrate it and then suggest, in the next section, an alternate way of regularization which is at once simpler and leads to a WT identity which formally implies the $`\xi `$-independence of the S-matrix.
An abelian special case of this has already been applied to QED.
Originally we derived the motivation for this work from the following observation in the context of the SM (where physical S-matrix elements generally exist). We had found $`\xi `$-dependence of the muon anomalous magnetic moment in the SM in when we follow the procedure of references as applied to the spontaneously broken (local) theory for the $`R_\xi `$ gauges. The discussion given here in this work has also been extended to the spontaneously broken U(1) chiral model where a similar $`\xi `$-dependence of a physical quantity has neen demonstrated. The procedure, formulated here in section IV has been applied there to this case; with formal $`\xi `$-independence established.
We now consider the non-local action for an arbitrary $`\xi `$. Here, we will generalize (following ), for an arbitrary $`\xi `$, the appropriate non-local action. We express
$`\widehat{S}_\xi `$=$`\widehat{S}+\mathrm{\Delta }\widehat{S}`$, (3.1)
where
$`\widehat{S}`$=$`d^4x(\frac{1}{2}_\mu \widehat{A}_\nu ^a^\mu \widehat{A}^{a\nu }_\mu \widehat{\overline{\eta }}^a^\mu \widehat{\eta }^a\frac{1}{2}B_\mu ^a𝒪_A^{1ab\mu \nu }B_\nu ^b+\overline{\psi }^a𝒪_\eta ^{1ab}\psi ^b`$
+Interaction terms) (3.2)
and
$`\mathrm{\Delta }\widehat{S}`$=$`\frac{1}{2}(1\frac{1}{\xi })d^4x(.\widehat{A}^a)^2`$ (3.3)
We note that as $`\mathrm{\Lambda }\mathrm{}`$, $`\widehat{S}`$ reduces to the local action of the Feynman gauge. Note, now, that the smeared gauge field $`\widehat{A}`$ has been constructed using the full ($`\xi `$ dependent) quadratic form $``$
$`\widehat{A}_\mu =_{A\mu \nu }^1A^\nu =(e^{\frac{}{2\mathrm{\Lambda }^2}})_{\mu \nu }A^\nu `$ (3.4)
and the ghost fields $`\widehat{\eta }`$ and $`\widehat{\overline{\eta }}`$ have been smeared using their respective quadratic forms
$`\widehat{\eta }=_\eta ^1\eta `$ $`\widehat{\overline{\eta }}=_\eta ^1`$$`\overline{\eta }`$ $`_\eta =e^{\frac{^2}{2\mathrm{\Lambda }^2}}`$ (3.5)
where it should be noted that $`_\eta `$ is independent of $`\xi `$.
We shall now proceed to evaluate $`\xi \frac{W}{\xi }|_{\xi =1}`$ for the above nonlocal theory. We note that the $`\xi `$ dependence of W comes from (i) the explicit $`\xi `$ dependence of $`\mathrm{\Delta }\widehat{S}`$ (ii) the implicit $`\xi `$ dependence of $`_A`$ in $`\widehat{A}`$ (iii) the explicit $`\xi `$ dependence of $`𝒪^1`$ in the B-field kinetic energy term and (iv) the implicit $`\xi `$ dependence of auxiliary fields B, $`\psi ,\overline{\psi }`$ and (v) finally from the measure $`\mu `$($`\xi `$). Of these, contribution (iv) vanishes since the auxiliary fields satisfy $`\frac{\delta S}{\delta \psi }|_{\psi =\psi [\varphi ]}`$=0.
The first contribution reads
(I)=$``$$``$$`\frac{i}{2\xi }d^4x(.\widehat{A}^a)^2`$$``$$``$=$``$$``$$`\frac{i}{2\xi }d^4x.A^a_R^2.A^a`$$``$$``$ (3.6)
\[with $`_R^2=e^{\frac{^2}{\mathrm{\Lambda }\xi }}`$\]
The implicit $`\xi `$ dependence of $`\widehat{A}`$ in $`\mathrm{\Delta }S`$ contributes
(IIA)=-$`\frac{i(1\xi )}{2}<<d^4x\frac{}{\xi }(.\widehat{A}^a)^2>>`$ (3.7)
=-$`\frac{i(1\xi )}{2\mathrm{\Lambda }^2\xi ^2}<<d^4x(.\widehat{A}^a)e^{\frac{^2}{2\mathrm{\Lambda }^2\xi }}^2(.A^a)>>`$ (3.8)
and it vanishes at $`\xi `$=1.
The contribution from the implicit dependence on $`\xi `$ of $`\widehat{A}`$ and $`𝒪^1`$ in the B-field kinetic terms can be computed straightforwardly. The result reads
(III)=-$`\frac{i}{2\mathrm{\Lambda }^2\xi }<<d^4x(.\widehat{A}^a)e^{\frac{^2}{2\mathrm{\Lambda }^2\xi }}^2(.A^a)>>`$\+
$`\frac{i}{2\xi }<<d^4x(.B^a)\frac{1}{1e^{\frac{^2}{\mathrm{\Lambda }^2\xi }}}(.B^a)>>`$\-
$`\frac{i}{2\mathrm{\Lambda }^2\xi ^2}<<d^4x(.B^a)e^{\frac{^2}{\mathrm{\Lambda }^2\xi }}^2(\frac{1}{1e^{\frac{^2}{\mathrm{\Lambda }^2\xi }}})^2(.B^a)>>`$ (3.9)
The $`(.A^a)^2`$ type terms in I and III combine to give
$`\frac{i}{2\xi }d^4x(.A^a)_R^2[1\frac{^2}{\mathrm{\Lambda }^2\xi }](.A^a)`$ (3.10)
At $`\xi `$=1, these can be simplified using the identity (A.5) in Appendix derived using the BRS WT identity. We note that as far as Green’s functions with external gauge fields are concerned, we can set terms $``$$``$$``$$`P(\xi )\overline{\eta }^a\frac{\delta S}{\delta \overline{\eta }^a}`$$``$$``$ to zero as shown in III of the appendix.We are then left with
$`\frac{i}{2}d^4x_R^2[1\frac{^2}{\mathrm{\Lambda }^2\xi }].A^a(x)`$$`\overline{\eta }^a(x)`$ $`d^4yJ^{b\mu }(_\mu \eta ^bgf^{bcd}^2(A_\mu ^c+B_\mu ^c)(\eta ^d+\psi ^d))`$ (3.11)
Next we simplify the $`(.B^a)^2`$ type terms using the relation
($`.B^a`$)=$`\frac{e^{\frac{^2}{\mathrm{\Lambda }^2\xi }}}{\mathrm{\Lambda }^2}[.\frac{\delta \widehat{S}_\xi }{\delta A^a}\frac{e^{\frac{^2}{\mathrm{\Lambda }^2\xi }}}{\xi }^2(.A^a)]`$ (3.12)
We note that these terms together simplify to yield
$`\xi \frac{W}{\xi }=\frac{i}{2\xi }d^4x<<.\frac{\delta \widehat{S}_\xi }{\delta A^a}\widehat{P}_1(\xi ).\frac{\delta \widehat{S}_\xi }{\delta A^a}>>`$
-$`\frac{i}{\xi }d^4x<<.\frac{\delta \widehat{S}_\xi }{\delta A^a}\widehat{P}_2(\xi ).A^a>>`$
+$`\frac{i}{2\xi }d^4x<<.A^a\widehat{P}_3(\xi ).A^a>>`$ (3.13)
where
$`\widehat{P}_1(\xi )=\widehat{P}(\xi )\frac{e^{\frac{2^2}{\mathrm{\Lambda }^2\xi }}}{\mathrm{\Lambda }^4}`$
$`\widehat{P}_2(\xi )=\widehat{P}(\xi )\frac{e^{\frac{3^2}{\mathrm{\Lambda }^2\xi }}}{\mathrm{\Lambda }^4}`$
$`\widehat{P}_3(\xi )=\widehat{P}(\xi )\frac{e^{\frac{4^2}{\mathrm{\Lambda }^2\xi }}}{\xi ^2}\frac{^2}{\mathrm{\Lambda }^2}\frac{^2}{\mathrm{\Lambda }^2}`$
$`\widehat{P}(\xi )`$=$`\frac{1}{1e^{\frac{^2}{\mathrm{\Lambda }^2\xi }}}`$ $`[1\frac{^2}{1e^{\frac{^2}{\mathrm{\Lambda }^2\xi }}}`$$`\frac{e^{\frac{^2}{\mathrm{\Lambda }^2\xi }}}{\mathrm{\Lambda }^2\xi }`$\]
The terms in (3.13) involving $`\widehat{P}_3(\xi )`$ can be simplified as those in (3.10) above and adds up to a term of the same form as (3.11). The second term on the right hand side in (3.13) can be simplified using (A.8). The residual term in (A.8) can be simplified using (A.5) as done earlier. (3.13) can be further simplified at $`\xi `$=1 as done in (A.10)-(A.12).
Combining all contributions together we find
$`\xi \frac{W}{\xi }|_{\xi =1,\chi =\chi =0}=<<\frac{i}{2}d^4x\{_R^2(1\frac{^2}{\mathrm{\Lambda }^2})+\widehat{P}_3(1)+2ig^2N\widehat{P}_2(1)\}.A^a(x)\overline{\eta }^a(x)`$
$`d^4yJ^{b\mu }(y)\{_\mu \eta ^bgf^{bcd}_A^2(A_\mu ^c+B_\mu ^c)(\eta ^d+\psi ^d)\}>>`$+$`F`$+measure contribution. (3.14)
where the Jacobian term $`F`$ reads
$`F`$=$`\frac{3g^2N}{4}`$\[$``$$`\frac{d^4k}{(2\pi )^4}`$f($`k^2)`$$`k^2`$\] $``$$`\frac{d^4p}{(2\pi )^4}`$ $`A_a`$(p).$`A_a`$(-p) (3.15)
where
$`f(k^2)`$=\[1+$`\frac{k^2e^{\frac{k^2}{\mathrm{\Lambda }^2}}}{\mathrm{\Lambda }^2(1e^{\frac{k^2}{\mathrm{\Lambda }^2}})}`$\] $`\frac{1}{1e^{\frac{k^2}{\mathrm{\Lambda }^2}}}`$$`\frac{e^{\frac{2k^2}{\mathrm{\Lambda }^2}}}{\mathrm{\Lambda }^4}`$ (3.16)
For arriving at the measure contribution, unlike other contributions, we need the form of the nonlocal BRS transformations for an arbitrary $`\xi `$. These are reproduced in Appendix B. We note that the nonlocal BRS and the trivial transformations do not anymore add up to a form where the following two desirable properties convenient for formulating WT identities hold:
(i) $`\delta `$$`A_\mu `$ involves the same combination that is involved in the ghost Lagrangian (ii) $`\delta `$$`\overline{\eta }`$ involves only $``$.A. If we define $`\mu (\xi )`$ with respect to either (A) nonlocal BRS of (B.1) or (B) resultant nonlocal BRS of (B.3), we have verified that the measure contribution to (3.14) cannot cancel the Jacobian contribution of (3.15). This cancellation has to be valid in the regularized theory (i.e. for any finite $`\mathrm{\Lambda }`$) and we, in particular, draw attention to the fact that $``$$``$$`\xi `$$`\frac{\mu }{\xi }`$$``$$``$ contains operators that have arbitrary order derivatives of A while $`F`$ does not.\[Note the form of $`\mu `$ in (2.24)for $`\xi `$=1\].
Finally, we wish to elaborate upon a shortcoming of the Feynman gauge treatment itself. As elaborated in 2C, the Lagrangian one starts with (of (2.19)) is actually for the case when the unrenormalized paprameter $`\xi _0`$=1. When one loop renormalization is carried out, the Lagrangian, when expressed in terms of renormalized fields, now does not retain its form of (2.19). So when we try to extend the treatment to two loops, we have the necessity for the treatment for arbitrary $`\xi `$ even in this case. This, as presented here, cannot be done along the lines of this section.
IV An Alternate way of regularization that preserves $`\xi `$ independence
In this section, we shall present a way of regularization that is at once simpler and leads to WT identities that would imply the $`\xi `$ independence of S-matrix elements (where they exist).We shall construct the relevent non-local BRS transformation that leads to the simpler form of the WT identity. A similar regularization has already been applied to QED.
We recall that the local action of the nonabelian gauge theory with an arbitrary $`\xi `$. It is expressed as
$`S_\xi =S_F+\mathrm{\Delta }S`$ (4.1)
where $`S_F`$ is the Feynman gauge local action.
We introduce the smeared field operators that depend only on the quadratic form in $`S_F`$. Hence, we have for an arbitrary $`\xi `$
$`\widehat{A^{}}_\mu =(_F^1)_{\mu \nu }A^\nu =e^{\frac{^2}{2\mathrm{\Lambda }^2}}A_\mu `$ (4.2a)
$`\widehat{\eta }^{^{}}=_F^1\eta =e^{\frac{^2}{2\mathrm{\Lambda }^2}}\eta `$ (4.2b)
We note $`_F=_\eta `$ here.
We write down the non-local action following the same rules as those in otherwise.
Explicitly,
$`S_\xi ^{^{}}=S_F^{^{}}+\mathrm{\Delta }S^{^{}}`$ (4.3)
with
$`S_F^{}`$\[A’,$`\eta ^{},\overline{\eta ^{}}`$\]=$``$$`d^4x`${-$`\frac{1}{2}`$$`_\mu A_\nu ^a^\mu A^{a\nu }`$-$`\frac{1}{2}`$$`B_\mu ^a𝒪^1B^{a\mu }`$-$`_\mu \overline{\eta ^{}}^a^\mu \eta _{}^{}{}_{}{}^{a}`$+$`\overline{\psi ^{}}^a𝒪^1\psi _{}^{}{}_{}{}^{a}`$}
+I\[A’+B’,$`\eta ^{}+\psi ^{}`$,$`\overline{\eta ^{}}+\overline{\psi ^{}}`$\] (4.4)
and
$`\mathrm{\Delta }S`$=$`\frac{1}{2}(1\frac{1}{\xi })d^4x(.A^a)^2`$ (4.5)
We note that in (4.4) the kinetic term for the auxiliary field B involves $`𝒪=\frac{_F^21}{_F}`$ that is $`\xi `$ independent. We further note that the form of the relations between auxiliary fields (B’,$`\psi ^{},\overline{\psi }^{}`$) and A’,$`\eta ^{},\overline{\eta }^{}`$ is same as in Feynman gauge as $`\mathrm{\Delta }\widehat{S}`$ does not contribute to these relations.<sup>4</sup><sup>4</sup>4We note that the present regularization does not preserve the properties of (2.13),(2.18). We also note that the shadow and the barred propagators in this regularization do not add up to the local one; but differ from it by ”gauge terms” (that vanish as $`\mathrm{\Lambda }\mathrm{}`$). This does not constitute a problem however. We can look upon this as a part of existing freedom in defining the local theory itself that exists on acount of gauge invariance.
Now, consider the change in the effective action $`S_\xi `$ under a field transformation
$`\delta \widehat{S}_\xi =\delta 𝒮[\varphi ,\psi [\varphi ]]`$=$`\delta 𝒮[\varphi ,\psi ]|_{\psi =\psi [\varphi ]}`$ (4.6)
Thus
$`\delta \widehat{S}_\xi =[\delta \varphi \frac{\delta 𝒮}{\delta \varphi }+\delta \psi \frac{\delta 𝒮}{\delta \psi }]_{\psi =\psi [\varphi ]}`$ (4.7)
The second term vanishes by the defining relation for $`\psi `$.
Thus
$`\delta \widehat{S}_\xi ^{^{}}=\delta \varphi \frac{\delta \widehat{S}_F^{^{}}}{\delta \varphi }+\delta \varphi \frac{\delta \mathrm{\Delta }\widehat{S}^{^{}}}{\delta \varphi }`$ (4.8)
Now consider the non-local BRS transformations of the Feynman gauge non-local action.
$`\widehat{\delta }A_\mu ^a=_F^2[_\mu (\eta +\psi )^agf^{abc}(A+B)_\mu ^b(\eta +\psi )^c]\delta \varsigma `$ (4.9a)
$`\widehat{\delta }\eta ^a=\frac{g}{2}f^{abc}_\eta ^2(\eta +\psi )^b(\eta +\psi )^c\delta \varsigma `$ (4.9b)
$`\widehat{\delta }\overline{\eta }^a=_\eta ^2(.A^a+.B^a)\delta \varsigma `$ (4.9c)
We know that since $`\widehat{S}_F[\varphi ,\psi [\varphi ]]`$ is exactly of the same form as in the Feynman gauge, it is invariant under the Feynman gauge non-local BRS transformations of (4.9). On the other hand we find by explicit calculation
$`\delta (\mathrm{\Delta }\widehat{S})=(1\frac{1}{\xi })d^4x[^2(.A^a)]\frac{\delta \widehat{S}_\xi }{\delta \eta ^a}\delta \varsigma `$ (4.10)
This change in $`\mathrm{\Delta }\widehat{S}`$ can be canceled by an additional change
$`\widehat{\delta ^{}}\overline{\eta }^a=(1\frac{1}{\xi })^2(.A^a)\delta \varsigma `$ (4.11)
In addition, we also note the dynamically irrelevent symmetries mentioned in the theorem (B.4). It reads
$`\widehat{\delta }^oA_\mu ^a=[(1^2)_\mu \eta ^a^2_\mu \psi ^a]\delta \varsigma `$ (4.12a)
$`\widehat{\delta }^o\eta ^a=0`$ (4.12b)
$`\widehat{\delta }^o\overline{\eta }^a=[^2(.B^a)\frac{1}{\xi }(1^2).A^a]\delta \varsigma `$ (4.12c)
This is an invariance of $`\widehat{S}_\xi `$ follows from theorem (B.4) and has been verified by explicit evaluation.
We add the transformations of (4.9),(4.11)and (4.12) to obtain the final non-local BRS symmetry of the non-localized action for an arbitrary $`\xi `$. It reads, for an arbitrary $`\xi `$:
$`\widehat{\delta }A_\mu ^a=[_\mu \eta ^agf^{abc}^2(A+B)_\mu ^b(\eta +\psi )^c]\delta \varsigma `$ (4.13a)
$`\widehat{\delta }\eta ^a=\frac{g}{2}f^{abc}^2(\eta +\psi )^b(\eta +\psi )^c\delta \varsigma `$ (4.13b)
$`\widehat{\delta }\overline{\eta }^a=\frac{1}{\xi }.A^a\delta \varsigma `$ (4.13c)
This now leads to the nonlocal BRS WT identity valid for an arbitrary $`\xi `$ of (A.4). We note here that as the Jacobian for the nonlocal trasformation (4.13) is $`\xi `$-independent by construction, $`\mu `$ can be taken to be independent of $`\xi `$.
We now obtain the value of $`\xi \frac{dW}{d\xi }|_{\chi =\overline{\chi }=0}`$, for an arbitrary $`\xi `$, in this formulation. We note the $`\xi `$ dependence now entirely comes from the explicit $`\xi `$ dependence of $`\mathrm{\Delta }\widehat{S}`$; since the regulators $``$, $`𝒪^1`$ are independent of $`\xi `$. We, thus, have
$`\xi \frac{W}{\xi }=\frac{i}{2\xi }<<d^4x.A^a_F^2.A^a>>`$
The above can be effectively simplified using (A.5) \[which now holds for an arbitrary $`\xi `$\], (A.6) and (A.9) to lead to
$`\xi \frac{W}{\xi }|_{J=J_{phy},\chi =\overline{\chi }=0}=<<\frac{i}{2}d^4x[_F^2.A^a(x)]\eta ^a(x)`$
$`d^4yJ^{b\mu }(y)\{_\mu \eta ^b(y)gf^{bcd}^2(A+B)_\mu ^c(\eta +\psi )^d\}>>`$ (4.15)
The above WT identity is the key to the $`\xi `$-independence of S-matrix elements (or quantities derived from them) wherever they exist. We note that in such cases, the discussion of $`\xi `$-independence should run entirely parallel to that in Ref.. One can adopt a limiting procedure of carrying out renormalization at an off-shell point $`p^2`$=-$`\mu ^2`$ and then take the limit $`\mu ^2`$$``$0 in the final result. We expect that when a similar regularization applied in the SBGT, the resulting WT identity similar to (4.15) will lead to the $`\xi `$-independence of the S-matrix elements that exist.
Appendix A
In this appendix, we shall derive the auxiliary equations needed in simplifying $`\xi `$$`\frac{W}{\xi }`$. We consider the field transformation, possibly nonlocal, \[$`ϵ`$ infinitesimal and $`\varphi `$ stands collectively for A, $`\eta `$ and $`\overline{\eta }`$\]
$`\varphi \varphi `$+$`ϵ`$F\[$`\varphi `$\] (A.1)
in the field variables in W\[J, $`\chi `$, $`\overline{\chi }]`$ to obtain the generalized equation of motion:
$``$$``$ $``$ $`d^4`$x{i$`_i`$$`F_i[\varphi ]`$ $`\frac{\delta \widehat{S}_\xi }{\delta \varphi _i}`$+$`_i`$$`J_i`$$`F_i`$\[$`\varphi `$\]+$``$+$`_i`$$`F_i`$$`\frac{\delta }{\delta \varphi _i}`$ln$`\mu `$$``$$``$=0. (A.2)
Here, $`_i`$$`J_i`$$`F_i`$\[$`\varphi `$\] collectively denotes the source terms and $`ϵ`$$``$ stands for the Jacobian (minus one) for the field transformation (A.1) viz
$``$=$``$ $`d^4`$x$`\frac{\delta F[\varphi ]}{\delta \varphi (y)}`$$`|_{x=y}`$ (A.3)
Note that the measure factor $`\mu `$ in Feynman gauge has been chosen so that the last two terms on the right hand side of (A.2) vanish for the (modified) nonlocal BRS transformations of (4.13). Thus, as a special case, we have the BRS nonlocal WT identity resulting from the surviving second term in (A.2)
$``$$``$$``$ $`d^4`$x\[$`J^{a\mu }`$$`\widehat{\delta }`$$`A_\mu ^a`$+$`\overline{\chi }^a`$ $`\widehat{\delta }`$$`\eta ^a`$+$`\widehat{\delta }`$$`\overline{\eta }^a`$ $`\chi ^a`$\]$``$$``$=0 (A.4)
Let P($`\xi `$) be any arbitrary differential operator that may depend on $`\xi `$ but not on the fields. We operate by $`_A^2`$P($`\xi `$)$`^\alpha `$$`\frac{\delta }{\delta J^{p\alpha }(y)}`$$`\frac{\delta }{\delta \chi ^c(y)}`$ on (A.4) and put $`\chi `$=0=$`\overline{\chi }`$ to obtain,
-i$`\frac{1}{\xi }`$$``$$``$\[P($`\xi `$)$`_A^2`$$`.`$$`A^p`$$`.`$$`A^c`$\]-P($`\xi `$)$`\overline{\eta }^c`$$`\frac{\delta S}{\delta \overline{\eta }^p}`$$``$$``$=
$`d^4zJ^{\mu b}(z)[_\mu \eta ^bgf^{bcd}_{A}^{}{}_{}{}^{2}[(A+B)_{\mu }^{}{}_{}{}^{c}(\eta +\xi )^d]](z)`$
$`_{A}^{}{}_{}{}^{2}P(\xi ).A^p(x)\overline{\eta }^c(x)`$ (A.5)
Further we note that the term of the form $``$$``$$``$$`d^4x`$ $`^\mu `$$`J_\mu (x)`$G\[$`\varphi `$\]$``$$``$ do not contribute to the Green’s function with external gauge boson lines with the physical polarization vectors since $`ϵ`$.k=0.
We express this by saying $``$$``$$``$$`d^4x`$ $`^\mu `$$`J_\mu (x)`$G\[$`\varphi `$\]$``$$``$$`|_{phy}`$=0. (A.6)
\[In using the subscript ’phy’ we do not necessarily imply mass shell limit however.\]
We shall need a set of results derivable from (A.2)
(I) We let F\[$`\varphi `$\] be linear in the fields. Then $``$ is field independent and can be dropped while evaluating $`\xi \frac{}{\xi }`$ of n-point Green’s functions.
(II) With $`F_\mu ^a`$\[$`\varphi `$\]=$`_\mu `$$`\widehat{P}_2(\xi )`$$`.A^a`$ (for $`\delta A_\mu `$) and $`F_\eta `$=$`F_{\overline{\eta }}`$=0, we then obtain
$``$$``$$``$$`d^4x`${i$`.`$$`A^a`$(x)$`\widehat{P}_2(\xi ).\frac{\delta \widehat{S}_\xi }{\delta A^a}+F_\mu ^a[A]\frac{\delta }{\delta A_\mu ^a}`$ln $`\mu `$+i$`J_\mu ^a(x)^\mu \widehat{P}_2(\xi ).A^a(x)\}`$$``$$``$=0 (A.7)
The measure factor at $`\xi =1`$ has been evaluated in \[$`K_1`$\] and is given by \[2.24\]. Using it, we obtain, at $`\xi =1`$ and with $`J`$=$`J_{phy}`$
$`d^4x[i.A^a\widehat{P}_2(\xi ).\frac{\delta \widehat{S}_\xi }{\delta A^a}g^2f^{acd}f^{bcd}F_{}^{a}{}_{\mu }{}^{}[A]`$$`A^{b\mu }(x)]`$$`|_{phy}`$=0 (A.8)
(III) We let $`F_A=0=F_\eta `$ and $`F_{\overline{\eta }}^{}{}_{}{}^{a}=P(\xi )\overline{\eta }^a`$ which is a linear transformation. Noting that $`\mu `$ does not depend on $`\overline{\eta }`$ and (I) above, we obtain that at $`\xi =1`$ and $`\chi =\overline{\chi }=0`$,
$``$$``$$`d^4x`$P($`\xi `$$`\overline{\eta }^a\frac{\delta \widehat{S}_\xi }{\delta \overline{\eta }^a(x)}`$$``$$``$=0 (A.9)
(IV) Finally, we let $`F_\mu ^a`$\[$`\varphi `$\]=$`_\mu `$$`\widehat{P}_1(\xi )`$ $`.\frac{\delta \widehat{S}_\xi }{\delta A^a(x)}`$. For physical sources, we find
i$`d^4x.\frac{\delta \widehat{S}_\xi }{\delta A^a}\widehat{P}_1(\xi ).\frac{\delta \widehat{S}_\xi }{\delta A^a}|_{J=J_{phy}}`$=-$`+\frac{\delta }{\delta A_\mu ^a}ln\mu F_\mu ^a`$ (A.10)
For $`\xi =1`$, we find that the term coming from the measure equals
$`d^4xg^2N.\frac{\delta \widehat{S}_\xi }{\delta A^a}\widehat{P}_1(\xi )`$.$`A^a`$$``$ +O($`g^3`$) (A.11)
This can be reduced further by using (A.8) to obtain
$`\frac{\delta ln\mu }{\delta A_\mu ^a}F_\mu ^a|_{\xi =1,J=J_{phy}}`$ =O ($`g^4`$) (A.12)
Appendix B
In this appendix, we shall write down the nonlocal BRS transformations for the case of an arbitrary $`\xi `$ following the general formalism of . We have, for the gauge fields,
$`_{\mu \nu }`$= $`\eta _{\mu \nu }`$$`^2`$-(1-$`\frac{1}{\xi }`$)$`_\mu _\nu `$
for the Euclidean formulation $`\eta _{\mu \nu }`$=diag(-1,-1,-1,-1).
We define
$`(_A)_{\mu \nu }`$=$`(e^{\frac{}{2\mathrm{\Lambda }^2}})_{\mu \nu }`$ (B.1)
We note $`^\mu `$($`_A)_{\mu \nu }`$=$`e^{\frac{^2}{2\mathrm{\Lambda }^2\xi }}_\nu `$$``$$`_{A}^{}{}_{}{}^{0}`$$`_\nu `$.
For the ghost case, we continue to define $`_\eta `$=$`e^{\frac{^2}{2\mathrm{\Lambda }^2}}`$$``$$`_{A}^{}{}_{}{}^{0}`$
Then the nonlocal BRS transformations read:
$`\widehat{\delta }A_\mu ^a`$ = $`(_{A}^{}{}_{}{}^{2})_{\mu }^{}{}_{}{}^{\nu }`$$`[_\nu (\eta ^a+\psi ^a)gf^{abc}(A_\nu ^b+B_\nu ^b)(\eta ^c+\psi ^c)]\delta \varsigma `$
$`\widehat{\delta }\eta ^a`$ = -$`\frac{g}{2}`$$`f^{abc}_{\eta }^{}{}_{}{}^{2}(\eta ^b+\psi ^b)(\eta ^c+\psi ^c)\delta \varsigma `$ (B.2)
$`\widehat{\delta }\overline{\eta }^a=_{\eta }^{}{}_{}{}^{2}_\mu (A^{a\mu }+B^{a\mu })\frac{\delta \varsigma }{\xi }`$
The trivial transformations, on the other hand, read
$`\widehat{\delta }^o`$$`A_{\mu }^{}{}_{}{}^{a}`$=$`\rho `$$`[(1_{\eta }^{}{}_{}{}^{2})`$$`_\mu \eta ^a`$-$`_{\eta }^{}{}_{}{}^{2}`$$`_\mu \psi ^a]`$$`\delta \varsigma `$
$`\widehat{\delta }^o\eta ^a=0`$ (B.3)
$`\widehat{\delta }^o\overline{\eta }^a`$=$`\rho `$$`[_{A}^{}{}_{}{}^{0}{}_{}{}^{2}(.B^a)(1_{A}^{}{}_{}{}^{0}{}_{}{}^{2}).A^a]\delta \varsigma `$
where $`\rho `$ is any constant. We note that in the first of (B.2) ‘$`_A`$’ appears in $`\widehat{\delta }A_{\mu }^{}{}_{}{}^{a}`$ , while is the first of (B.3) ‘$`_\eta `$ ’ appears in $`\widehat{\delta }^0A_{\mu }^{}{}_{}{}^{a}`$. In the case of $`\xi `$=1, $`_\eta `$=$`_A`$=$`_{A}^{}{}_{}{}^{0}`$ . Then with $`\rho =1,`$ the first of (B.2) and (B.3) added together lead to $`\widehat{\delta }`$’A which contains the same combination of terms present in the ghost Lagrangian and this leads to the simplification in the expression for $`\xi \frac{W}{\xi }`$. This no longer happens for $`\xi `$ 1 . Similarly, the last of (B.2) and (B.3) now contain different regulators ‘$`_{\eta }^{}{}_{}{}^{2}`$ ’ and ‘$`_{𝒜}^{}{}_{}{}^{\mathrm{0\hspace{0.17em}2}}`$’ respectively. So, even with $`\rho =\frac{1}{\xi }`$, they do not lead to the cancellation of $`.`$B terms. Moreover, note that the value of $`\rho `$ needed in the first of (B.3) needed for a ‘$`\mathrm{𝐧𝐞𝐚𝐫}`$’ cancellation of unwanted terms does not agree with the value of $`\rho `$ in the last of (B.3) for a ‘near’ cancellation of $`.`$B terms. As a result of this, for $`\xi `$ 1, we do not have the simplified treatment of \[$`4`$\] available in the standard treatment. This holds, even if we try to modify (B.2).
We could define measure $`\mu (\xi )`$ with respect to (B.2)+(B.3) with either $`\rho =1`$ or $`\rho =\frac{1}{\xi }`$ and, in either case, we find that $`\mu (\xi )`$ must contain terms that cannot cancel the term $`F`$ of (3.15).
References:
1: G. ’t Hooft and M. Veltman Nucl. Phys. B33, 189 (1972)
2: E. D. Evans et al, Phys Rev D43, 499 (1991)
3: See e.g, D. Z. Freedman et al, Nucl. Phys. B395, 454(1993)
4: G. Kleppe and R. P. Woodard, Nucl. Phys. B388, 81 (1992)
5: G. Kleppe and R. P. Woodard, Ann. Phys. (N.Y.) 221, 106 (1993)
6: See e.g. and references therein
7: M. A. Clayton Gauge Invariance in nonlocal regularized QED. (Toronto U.) UTPT-93-14, Jul 1993. \[hep-th/9307089\]
8: S. D. Joglekar and G. Saini, Z. Phys. C.76, 343-353 (1997)
9: S. D. Joglekar The condition 0 $``$ Z $``$ and an intrinsic mass scale in quantum field theory. Oct1999. 9pp. and hep-th/0003077
10: S. D. Joglekar Understanding of the renormalization programme in a mathematically rigorous framework and an intrinsic mass scale. Dec 1999. 18pp. and hep-th/0003104
11: J. Paris and W. Troost: hep-th/9607215
12: See e.g. E. Abers and B. W. Lee Phys. Rep. 9C(1973),1
13: S. D. Joglekar and G. Saini (Unpublished)
14: A. Basu and S. D. Joglekar \[in preparation\]
15: See e.g. B. W. Lee in “Methods in Field Theory” Les Houches 1975 Editor: R. Balian and J. Zinn-Justin.
|
warning/0004/cond-mat0004002.html
|
ar5iv
|
text
|
# Soft and non-soft structural transitions in disordered nematic networks
## I Introduction
Elastomers and gels are intrinsically disordered solids that retain the memory of their initial states. The non-equilibrium nature of their fabrication processes causes frozen heterogeneities in the network structure, which range in size from mesoscopic to macroscopic scales . The presence of the quenched disorder comes to the fore when we introduce some soft order in the system. For instance, density fluctuations of swollen gels near the critical point are strongly enhanced by the heterogeneities. Under stretching, they produce the so-called “abnormal butterfly” pattern in small-angle neutron scattering intensity . It illustrates how the elasticity of gels gives rise to a non-trivial effect unexpected in other random systems. Here we address another example of soft order in disordered elastic networks.
Nematic liquid-crystalline elastomers and gels constitute a unique class of solids characterized by a coupling between the orientational and translational degrees of freedom. Physical consequences of the strain-orientation coupling have been the subject of a considerable amount of studies, both theoretical and experimental. Notable theoretical advances in the past include: (i) De Gennes showed that a spontaneous elongation along the director is induced by the isotropic-nematic (I-N) transition; (ii) A molecular model of nematic networks was constructed by Warner et al. , extending the classical affine-deformation model of rubber elasticity; (iii) Uniformly oriented networks possess soft modes of strain and orientation fluctuations that do not accompany any change of rubber-elastic free energy. It was first predicted by Golubović and Lubensky on a phenomenological ground and later extended by the affine-deformation theory . Thus and in other ways, the behavior of homogeneous and clean nematic networks is now fairly well understood .
In practice, however, nematic networks in equilibrium quite often exhibit polydomain director textures, where the orientational correlation length is typically in the micron range. Under external strain, polydomain networks undergo a structural change into a macroscopically aligned monodomain state, where the director lies along the extensional direction. This change, called the polydomain-monodomain (P-M) transition, is characterized by a highly non-linear mechanical response . The strain-stress curve shows a small slope in the partially aligned (polydomain) state. Depending on the material and the method of synthesis, the slope is sometimes vanishingly small while it is sizable in other cases. The macroscopic stress as a function of strain shows a steep rise as the system turns into the monodomain state.
There have been a few theoretical attempts to describe polydomain networks and their mechanical responses. Ten Bosch and Varichon set up the first model, in which they attributed the origin of the equilibrium texture to a random anchoring field exerted by network crosslinks. An interesting analogy with random anisotropy magnets was pursued by Fridrikh and Terentjev . They proposed a mapping to the XY model under random and homogeneous magnetic fields, from analysis of which they predicted a discontinuous stress-orientation curve.
Nonetheless, the role of strain-orientation coupling in polydomain networks is still far from clear. There are two aspects to be considered. Firstly, the previous theories assume only local interactions between domains, for instance by arguing that the elastic free energy localizes in domain walls under strain . In general, however, inhomogeneities in an elastic material cause non-local or long-range interactions mediated by the strain field. Such elastic interactions control the physics of various systems, such as solids with dislocations or surface defects , phase separating alloys , gels undergoing swelling , and membranes with inclusions . Disordered nematic networks provide another intriguing example, and differ from any of the above materials in having a non-scalar order parameter. Secondly, the mechanical response should strongly depend on the crosslinking condition. Polydomain elastomers have been obtained by either of the following ways ; (i) to crosslink a polymer melt in the isotropic phase and then cool it into the nematic phase; (ii) to crosslink a nematic polymer melt containing polydomain textures. These two cases have not been theoretically well distinguished so far. We shall refer to them as the cases of isotropic and anisotropic crosslinkings, respectively.
Recently, we studied the elastic interaction in isotropically crosslinked networks , and found an almost completely soft P-M transition . The macroscopic stress due to the strain-orientation coupling was shown to be slightly negative and of $`O(\alpha ^2)`$ in the P-M transition regime, where $`\alpha `$ is the degree of chain anisotropy. This contrasts with the earlier prediction of a positive stress of $`O(\alpha )`$ . The elastic interaction also produces a “four-leaf clover” pattern in the depolarized light scattering intensity, which resembles the experimental observation by Clarke et al. .
In this article, we extend previous work and provide the details of our picture of the P-M transition. Here let us summarize the ideas and results which we have not emphasized in previous work. Firstly, we pursue the idea that random internal stresses destroy the long-range orientational order, which was suggested (but not proven) earlier in a broader context . This will be done by incorporating the notion of frozen heterogeneous strains into the extended affine-deformation theory . We argue that the random internal stresses act as stronger sources of disorder than the random molecular field due to crosslinks . Secondly, evolution of domain structure with and without external stretching is numerically simulated by a simplified dynamical model. The “four-leaf clover” scattering pattern has four peaks at finite wavenumbers, and the peak height is a non-monotonic function of the macroscopic strain, in qualitative agreement with experiment. We find a slow dynamical relaxation of the structure factor, and show that the peak wavenumber asymptotically goes to zero in the long-time limit. Thirdly, we study the case of anisotropic crosslinking. In this case, the initial director configuration of a macroscopic polydomain texture is memorized into the network. It provides a source of strong and correlated disorder, resulting in a non-soft P-M transition. The spatial distribution of elastic free energy in anisotropically crosslinked networks is strongly dehomogenized by strain, while that in isotropically crosslinked networks is unchanged during the P-M transition.
This paper is organized as follows. In Section II, we introduce a random stress model, derive an effective free energy, and discuss the mechanism of the soft mechanical response. Section III describes a numerical simulation of the polydomain state and the P-M transition. In Section IV, we analyze the effect of random stresses on director fluctuations in the monodomain state. We study networks prepared in the nematic phase in Section V. In Section VI, we summarize the results in comparison to existing experiments, and conclude with a proposal of future directions.
## II Model and Analysis
### A Random stresses in isotropic networks
It is known since long ago that the network structure of gels are often heterogeneous on many length-scales, which are considerably larger than the mesh size (see Fig.1). In the swollen state, these imperfections manifest themselves as density inhomogeneity and are observed through the so-called butterfly pattern in neutron scattering intensity or as speckles in light scattering experiment . Although less discussed in the literature, it is natural to expect that elastomers, often fabricated by drying gels, also contain the memory of heterogeneous network formation. The frozen heterogeneities reflect the non-equilibrium nature of the crosslinking processes, and produce random internal stresses in the material. While the roles of random stresses in gels and other amorphous solids have been discussed from a phenomenological point of view , much remains to be done to understand them on the basis of a molecular theory . In this subsection, we recapitulate the notion of random stresses using the classical affine-deformation theory of isotropic rubber networks , in order to prepare for modeling disordered nematic networks in the next subsection.
The basic object in the affine-deformation theory is the probability distribution of the chain’s end-to-end vector $`𝝆`$. The distribution function at thermal equilibrium is isotropic and Gaussian, and given by
$`P_{eq}(𝝆)=𝒩^1\mathrm{exp}\left({\displaystyle \frac{d}{2\mathrm{\Omega }}}\rho ^2\right),`$ (1)
where $`\mathrm{\Omega }`$ $`=\rho ^2_{eq}`$ is a constant, $`d`$ is the spatial dimension, and $`𝒩=𝑑𝝆P_{eq}(𝝆)`$ is the normalization factor. The macroscopic deformation of the network is described by the Cauchy deformation tensor,
$`\mathrm{\Lambda }_{ij}={\displaystyle \frac{r_i}{r_{0j}}},`$ (2)
where $`𝒓`$ and $`𝒓_0`$ are the positions of material points at observation and at the moment of crosslinking, respectively. The basic assumption of the theory is that each chain’s end-to-end vector affinely changes as $`𝝆\mathsf{\Lambda }𝝆`$ in response to the macroscopic deformation. The free energy per chain is given by
$`f_{chain}=k_BT{\displaystyle 𝑑𝝆P_0(𝝆)\mathrm{ln}P_{eq}(\mathsf{\Lambda }𝝆)},`$ (3)
where $`P_0(𝝆)`$ is the probability distribution function at the moment of crosslinking, which is not necessarily identical to the equilibrium distribution. We assume that the chains are distorted before crosslinking, and denote the deviation from the equilibrium conformation by a tensor $`𝖱`$, defined by
$`𝝆𝝆_0=\mathrm{\Omega }(𝖨+𝖱),`$ (4)
where $`𝖨`$ is the unit tensor. If the deviation is not very large and the chains are not stretched out, we may still approximate $`P_0`$ to be Gaussian, and put
$`P_0(𝝆)=𝒩_0^1\mathrm{exp}\left[{\displaystyle \frac{d}{2\mathrm{\Omega }}}𝝆(𝖨+𝖱)^1𝝆\right].`$ (5)
Substituting (1) and (5) into Eq.(3), we have
$`f_{chain}={\displaystyle \frac{k_BT}{2}}[\text{Tr}𝖦+𝖱:𝖦+\mathrm{ln}det(𝖨+𝖱)],`$ (6)
where $`G_{ij}=\mathrm{\Lambda }_{ki}\mathrm{\Lambda }_{kj}`$ is the metric tensor of deformation. The equation (6) is not new and essentially contained in the classical theory of Flory . Taking the spatial heterogeneity of $`𝖱`$ into account and neglecting terms independent of $`\mathsf{\Lambda }`$, the total elastic free energy is written as
$`F_{el}={\displaystyle \frac{k_BT\nu _0}{2}}{\displaystyle }d𝒓_0(\text{Tr}𝖦+𝖱:𝖦),`$ (7)
where $`\nu _0`$ is the number density of subchains. Inhomogeneous contribution of the form $`𝖱:𝖦`$ can be also derived from Cauchy’s theory of solids bound by a central force . We shall call $`R_{ij}`$ the quenched random stress , although it is more directly related to quenched random strain in the present model. For simplicity, we assume that the frozen heterogeneities have a single characteristic size $`\xi _R`$, which is substantially larger than the mesh size. After a coarse-graining on the scale $`\xi _R`$, we can regard $`R_{ij}`$ as a spatially uncorrelated Gaussian random variable satisfying
$`R_{ij}(𝒓_0)`$ $`=`$ $`0,`$ (8)
$`R_{ij}(𝒓_0)R_{kl}(𝒓_0^{})`$ $`=`$ $`\xi _R^d\delta (𝒓_0𝒓_0^{})`$ (10)
$`\left[\beta ^2\left(\delta _{ik}\delta _{jl}+\delta _{il}\delta _{jk}{\displaystyle \frac{2}{d}}\delta _{ij}\delta _{kl}\right)+\beta ^2\delta _{ij}\delta _{kl}\right].`$
The dimensionless constants $`\beta `$ and $`\beta ^{}`$ represent the magnitudes of shear and dilatational quenched strains, respectively.
### B Random stresses in nematic networks
Next we consider nematic elastomers and gels. Warner et al. constructed an affine-deformation theory of nematic networks, by generalizing the classical theory. Their basic observation is that nematic chains with low backbone rigidity are well characterized by an anisotropic Gaussian conformation, elongated along the director. The equilibrium distribution of the end-to-end vector can be written in the form,
$`P_{eq}(𝝆)`$ $`=`$ $`𝒩^1\mathrm{exp}\left[{\displaystyle \frac{d}{2\mathrm{\Omega }^{}}}𝝆(𝖨\alpha 𝖰)𝝆\right],`$ (11)
where $`\alpha `$ is the degree of chain anisotropy and
$`Q_{ij}=Q_0\left(\delta _{ij}{\displaystyle \frac{1}{d}}n_in_j\right)`$ (12)
is the orientational order parameter with $`𝒏`$ being the director. We consider a system deep in the nematic phase and put $`Q_0=1`$; the state of orientation is completely specified by the director. The coupling constant $`\alpha `$ is expressed in terms of the parameters used in as
$`\alpha ={\displaystyle \frac{\mathrm{}_{}\mathrm{}_{}}{(11/d)\mathrm{}_{}+(1/d)\mathrm{}_{}}}.`$ (13)
Note that $`\alpha `$ does not exceed $`d/(d1)`$, the value attained in the anisotropic limit $`\mathrm{}_{}/\mathrm{}_{}\mathrm{}`$. An advantage of the affine-deformation model is that it can describe arbitrary crosslinking conditions; the networks can be fabricated either in the isotropic or the nematic phase. First we consider networks originally crosslinked in the isotropic phase, and shall treat the case of anisotropic crosslinking in Section V. The random stresses are now readily incorporated into the original model. For the case of isotropic crosslinking, the initial chain conformation can be described by Eq.(5), with (8) and (10). Substituting (5) and (11) into Eq.(3), and dropping terms independent of $`\mathsf{\Lambda }`$ and/or $`𝖰`$, we arrive at the elastic free energy,
$`F_{el}={\displaystyle \frac{\mu }{2}}{\displaystyle 𝑑𝒓\text{Tr}\left[(𝖨+𝖱)\mathsf{\Lambda }^T(𝖨\alpha 𝖰)\mathsf{\Lambda }𝖨\right]}`$ (14)
with $`\mu =k_BT\nu _0(\mathrm{\Omega }/\mathrm{\Omega }^{})`$. Here we subtracted a constant so that $`F_{el}`$ vanishes when $`\mathsf{\Lambda }=𝖨`$ and $`\alpha =\beta =0`$. We also replaced $`𝑑𝒓_0`$ with $`𝑑𝒓`$, assuming an incompressible network and imposing the local constraint,
$`det\mathsf{\Lambda }=1.`$ (15)
We decompose the elastic free energy into proper and disorder contributions as $`F_{el}=F_{el}^P+F_{el}^D`$, where, by definition, the former is given by formally putting $`𝖱=0`$ in the right hand side of (14), and
$`F_{el}^D={\displaystyle \frac{\mu }{2}}{\displaystyle 𝑑𝒓\text{Tr}\left[𝖱\mathsf{\Lambda }^T(𝖨\alpha 𝖰)\mathsf{\Lambda }\right]}.`$ (16)
The total free energy of the system is written as $`F=F_{el}+F_F`$ where $`F_F`$ is the Frank free energy, for which we use the so-called one-constant approximation ,
$`F_F={\displaystyle \frac{K}{2}}{\displaystyle 𝑑𝒓(𝒏)^2}.`$ (17)
We assume that the average deformation $`\overline{\mathrm{\Lambda }}_{ij}`$ is a uniaxial strain along the $`x`$-axis, parameterized by the elongation ratio $`\lambda `$, as
$`\overline{\mathsf{\Lambda }}`$ $`=`$ $`\lambda 𝒆_x𝒆_x+\lambda ^{1/(d1)}(𝖨𝒆_x𝒆_x).`$ (18)
The internal displacement is defined as the deviation from the average deformation,
$`𝒖=𝒓\overline{\mathrm{\Lambda }}𝒓_0.`$ (19)
with which the deformation tensor is expressed as
$`\mathrm{\Lambda }_{ij}=\overline{\mathrm{\Lambda }}_{kj}(\delta _{ki}+_ku_i)`$ (20)
(here and hereafter, we imply summation over repeated indices $`i`$,$`j`$,$`k`$,$`l`$, and $`m`$).
In the absence of quenched disorder, the elastic free energy is minimized at $`\lambda =\lambda _m`$ and $`𝒖=0`$, where
$`\lambda _m=\left[{\displaystyle \frac{1+(1/d)\alpha }{1(11/d)\alpha }}\right]^{(d1)/2d}`$ (21)
is the ratio of the spontaneous elongation induced by the isotropic-nematic transition (see Fig.2). However, if the random stresses are enough strong, the long-range orientational order is destroyed and the ground state of the system is shifted to a polydomain state with $`\lambda =1`$, as we shall see. Hereafter and throughout the paper, we regard $`\lambda `$ as an externally controlled parameter.
### C Effective free energy
In this subsection, we derive the effective free energy in the mechanical equilibrium state under the constraint $`\lambda =1`$, and discuss its physical consequences. Substituting Eqs.(18), (19) and (20) into Eq.(14), expanding it with respect to $`𝒖`$ and retaining a bilinear form in $`𝒖`$, $`𝖱`$ and $`𝖰`$, we have
$`F_{el}|_{\lambda =1}`$ $`=`$ $`{\displaystyle \frac{\mu }{2}}{\displaystyle }d𝒓[(_iu_j)^2+2(R_{ij}\alpha Q_{ij})_iu_j`$ (23)
$`\alpha R_{ij}Q_{ij}]`$
From this we eliminate the displacement field using the mechanical equilibrium condition,
$`{\displaystyle \frac{\delta F_{el}}{\delta 𝒖}}=0,`$ (24)
and the incompressibility condition (25), which to the lowest order in $`𝒖`$ reads
$`𝒖=0.`$ (25)
After a straightforward calculation following the procedure described in , we obtain an effective elastic free energy which is correct to the quadratic order in $`\alpha `$, $`\beta `$, and $`\beta ^{}`$, as
$`\stackrel{~}{F}_{el}|_{\lambda =1}`$ $`=`$ $`{\displaystyle \frac{\mu }{2}}{\displaystyle _𝒒}[|(𝖨\widehat{𝒒}\widehat{𝒒})(\widehat{𝒒}𝖱(𝒒)\alpha \widehat{𝒒}𝖰(𝒒))|^2`$ (27)
$`+\alpha 𝖱(𝒒):𝖰(𝒒)],`$
where $`\widehat{𝒒}=𝒒/|𝒒|`$ and $`_𝒒=(2\pi )^d𝑑𝒒`$ (the tilde is to put to express the effective nature of the free energy). The proper contribution to the free energy is given by
$`\stackrel{~}{F}_{el}^P`$ $`=`$ $`{\displaystyle \frac{\mu \alpha ^2}{2}}{\displaystyle _𝒒}\left|(𝖨\widehat{𝒒}\widehat{𝒒})\left(\widehat{𝒒}𝖰(𝒒)\right)\right|^2.`$ (28)
In the real space, Eq.(28) is rewritten in the form of a two-body long-range interaction, as
$`\stackrel{~}{F}_{el}^P`$ $`=`$ $`{\displaystyle \frac{\mu \alpha ^2}{2}}{\displaystyle 𝑑𝒓𝑑𝒓^{}}`$ (31)
$`[Q_{ik}(𝒓)_i_jG_1(𝒓𝒓^{})Q_{jk}(𝒓^{})`$
$`+Q_{ij}(𝒓)_i_j_k_lG_2(𝒓𝒓^{})Q_{kl}(𝒓^{})],`$
where $`G_n(𝒓)(n=1,2)`$ are the Green functions defined by
$`(^2)^nG_n(𝒓)`$ $`=`$ $`\delta (𝒓),`$ (32)
$`G_n(r\mathrm{})`$ $`=`$ $`0.`$ (33)
In a similar manner, the disorder part of the free energy is written as
$`\stackrel{~}{F}_{el}^D`$ $`=`$ $`\mu \alpha {\displaystyle 𝑑𝒓𝑑𝒓^{}}`$ (36)
$`[R_{ik}(𝒓)({\displaystyle \frac{1}{2}}\delta _{ij}\delta (𝒓𝒓^{})+_i_jG_1(𝒓𝒓^{}))Q_{jk}(𝒓^{})`$
$`+R_{ij}(𝒓)_i_j_k_lG_2(𝒓𝒓^{})Q_{kl}(𝒓^{})]+\text{const.},`$
In a given direction of $`𝑹=𝒓𝒓^{}`$, the kernels $`_i_jG_1`$ and $`_i_j_k_lG_2`$ in (31) and (36) decay in proportion to $`R^d`$.
Let us discuss the physical meaning of the effective free energy. First we consider the disorder part, neglecting the proper elastic interaction for the moment. If we decompose the random stress into the dilatational part $`R_{kk}\delta _{ij}/d`$ and the shear (traceless) part $`R_{ij}R_{kk}\delta _{ij}/d`$, the former makes no contribution to the free energy (23) because of (25) and the tracelessness of $`𝖰`$. Thus, only the shear portion of the random stresses (whose strength is represented by $`\beta `$) is relevant, at least in the bilinear order. It is intuitively obvious that a mere volume change does not create any preferential director orientation, while anisotropic strain does. As seen from Eq.(36), the random stresses both locally and non-locally act on the director field. The classical scaling argument by Imry and Ma tells that arbitrary weak random stresses destroy the long-range orientation order in dimensions lower than four <sup>*</sup><sup>*</sup>* Although the original Imry-Ma argument assumes an uncorrelated random field, it is easy to see that it also holds in the present case, including the scaling law for the domain size. To see this, it is useful to rewrite the right hand side of Eq.(36) into the form $`\mu \alpha 𝑑𝐫𝖯(𝐫):𝖰(𝐫)`$, where the effective random field $`𝖯`$ has a long-range correlation schematically represented as $`P_{ij}(𝐫)P_{kl}(𝐫^{})=\xi _R^d\left[\mathrm{\Pi }_{ijkl}\delta (𝐫𝐫^{})+\mathrm{\Pi }_{ijkl}^{}|𝐫𝐫^{}|^d\right],`$ (37) where $`\mathsf{\Pi }^{}`$ depends on the direction of $`𝐫𝐫^{}`$ but not on its magnitude. Since both $`\mathsf{\Pi }`$ and $`\mathsf{\Pi }^{}`$ are dimensionless quantities, there appears no additional characteristic length that affects the Imry-Ma scaling of disorder free energy contained per domain. . The domain size or the orientational correlation length, which we denote by $`\xi `$, is determined by a balance between the effects of random stresses and Frank elasticity. The effective strength of disorder is expressed by the dimensionless parameter,
$`D={\displaystyle \frac{\mu \alpha \beta }{K}}\xi _R^2.`$ (38)
According to the Imry-Ma argument, the domain size scales as $`\xi /\xi _RD^{2/(d4)}`$ in the weak disorder regime $`D1`$. For a strong disorder $`D\stackrel{>}{}\mathrm{\hspace{0.33em}1}`$, we should have $`\xi \xi _R`$ and optimization of the director field will reduce the disorder free energy density roughly by $`\mu \alpha \beta `$.
Next we turn to the proper elastic interaction. It should play only a secondary role in selecting the domain size $`\xi `$ because of the invariance of Eq.(28) against a change of scale $`𝒒\text{const.}\times 𝒒`$. However, it creates a characteristic anisotropy in the the orientational correlation. We see this first in the two dimensional case. In 2D, the orientational configuration is specified by the director’s azimuthal angle $`\theta =\theta (𝒓)`$, as
$`𝒏=(\mathrm{cos}\theta ,\mathrm{sin}\theta ),`$ (39)
or, equivalently,
$`𝖰={\displaystyle \frac{1}{2}}\left[\begin{array}{cc}\mathrm{cos}2\theta & \hfill \mathrm{sin}2\theta \\ \mathrm{sin}2\theta & \hfill \mathrm{cos}2\theta \end{array}\right].`$ (42)
A straightforward calculation reduces Eq.(31) to
$`\stackrel{~}{F}_{el}^P`$ $`=`$ $`{\displaystyle \frac{\mu \alpha ^2}{16\pi }}{\displaystyle 𝑑𝒓𝑑𝒓^{}\frac{1}{R^2}}`$ (44)
$`\mathrm{cos}\left[2(\theta (𝒓)\psi )+2(\theta (𝒓^{})\psi )\right],`$
where $`\psi `$ is the azimuthal angle of $`𝑹=𝒓𝒓^{}=|𝑹|(\mathrm{cos}\psi ,\mathrm{sin}\psi )`$. From the angle-dependence of the integrand, we expect that the above free energy is minimized by a “checkered” domain configuration as depicted in Fig.3. Correlation in directions parallel and perpendicular to the local director is suppressed while those in oblique directions are enhanced. It has the following simple interpretation. Upon the isotropic-nematic transition, each part of the network tends to elongate along the local director. The domain in the center of the figure pushes the top neighbor upward, pulls the left neighbor rightward, and so on. To reduce the mechanical conflict without violating the global constraint $`\lambda =1`$, the top and left domains are reoriented perpendicular to the central one. This domain reconfiguration enables the I-N transition-induced elongation along the local director, despite of spatial inhomogeneity.
The same picture holds for orientational correlation in three dimensions. In 3D, Eq.(31) becomes
$`\stackrel{~}{F}_{el}^P`$ $`=`$ $`{\displaystyle \frac{\mu \alpha ^2}{16\pi }}{\displaystyle 𝑑𝒓𝑑𝒓^{}\frac{1}{R^3}g(𝒏,𝒏^{},\widehat{𝑹})},`$ (45)
$`g(𝒏,𝒏^{},\widehat{𝑹})`$ $`=`$ $`{\displaystyle \frac{5}{3}}+4(𝒏𝒏^{})^2+(𝒏\widehat{𝑹})^2+(𝒏^{}\widehat{𝑹})^2`$ (46)
$``$ $`18(𝒏𝒏^{})(𝒏\widehat{𝑹})(𝒏^{}\widehat{𝑹})`$ (47)
$`+`$ $`15(𝒏\widehat{𝑹})^2(𝒏^{}\widehat{𝑹})^2,`$ (48)
where $`𝒏=𝒏(𝒓)`$, $`𝒏^{}=𝒏(𝒓^{})`$ and $`\widehat{𝑹}=𝑹/|𝑹|`$. Correlation in the direction parallel to the director is suppressed as in the 2D case, which is known by observing that the function
$`g(𝒏,𝒏^{},𝒏)={\displaystyle \frac{2}{3}}+2(𝒏𝒏^{})^2`$ (49)
takes its minimum when $`𝒏𝒏^{}`$.
The domain reconfiguration due to the proper elastic interaction is suppressed by the Frank elasticity at wavelengths shorter than
$`\xi _c=\sqrt{{\displaystyle \frac{K}{\mu \alpha ^2}}}.`$ (50)
Thus we have three characteristic lengthscales, $`\xi `$, $`\xi _R`$, and $`\xi _c`$. The observed domain size $`\xi `$ is typically $`110^1\mu `$m, while we estimate $`\xi _c`$ to be $`10`$ nm for typical experimental values $`K=10^{11}`$ J/m, $`\mu =10^5`$ J/m<sup>3</sup>, and $`\alpha =1.0`$. There is a substantial gap between $`\xi `$ and $`\xi _c`$, where the proper elastic interaction plays a dominant role. The Frank free energy density $`f_F`$ (averaged over space) scales as $`f_Ff_{el}^P(\xi _c/\xi )^2f_{el}^P\mu \alpha ^2`$. The domain size $`\xi `$ can be cast into a scaling form,
$`{\displaystyle \frac{\xi }{\xi _R}}=\mathrm{\Xi }(D,{\displaystyle \frac{\xi _c}{\xi _R}}).`$ (51)
Although $`\mathrm{\Xi }`$ is a highly non-trivial function, it can be numerically obtainable unless $`D`$ is very small (or, unless $`\xi /\xi _R`$ is very large), as we see in Section III. We have a trial estimate $`D1`$ if we assume $`\beta 0.01`$ and $`\xi _R100`$ nm in addition to the above values of $`K`$, $`\mu `$, and $`\alpha `$. Of course, this estimate of $`D`$ is quite uncertain because the magnitudes of $`\beta `$ and $`\xi _R`$ should depend on the kinetics of the crosslinking process, quality of the solvent, etc. Our point here is that it is not unreasonable to have a moderately strong disorder in the presence of submicron-scale network heterogeneities, which is considered ubiquitous.
### D Mechanical response
Now we proceed to discuss the mechanical response during the polydomain-monodomain transition. To do so, it is useful to examine again the polydomain state at $`\lambda =1`$ and in 2D. The harmonic free energy (28) can be rewritten as
$`\stackrel{~}{F}_{el}^P`$ $`=`$ $`{\displaystyle \frac{\mu \alpha ^2}{2}}{\displaystyle _𝒒}\left|Q_1(𝒒)\right|^2,`$ (52)
$`Q_1(𝒒)`$ $`=`$ $`2\widehat{q}_x\widehat{q}_yQ_{xx}(𝒒)(\widehat{q}_x^2\widehat{q}_y^2)Q_{xy}(𝒒)`$ (53)
$`=`$ $`\mathrm{sin}2\phi Q_{xx}(𝒒)\mathrm{cos}2\phi Q_{xy}(𝒒),`$ (54)
where $`\phi `$ is the azimuthal angle of the wavevector, $`𝒒=|𝒒|(\mathrm{cos}\phi ,\mathrm{sin}\phi )`$. Complementary to $`Q_1(𝒒)`$ is the variable defined by
$`Q_2(𝒒)`$ $`=`$ $`(\widehat{q}_x^2\widehat{q}_y^2)Q_{xx}(𝒒)+2\widehat{q}_x\widehat{q}_yQ_{xy}(𝒒)`$ (55)
$`=`$ $`\mathrm{cos}2\phi Q_{xx}(𝒒)+\mathrm{sin}2\phi Q_{xy}(𝒒).`$ (56)
Note that $`Q_1(𝒒)`$ and $`Q_2(𝒒)`$ constitute a set of normal modes, and satisfy
$`\left|Q_1(𝒒)\right|^2+\left|Q_2(𝒒)\right|^2=\left|Q_{xx}(𝒒)\right|^2+\left|Q_{xy}(𝒒)\right|^2,`$ (57)
or
$`\overline{Q_1(𝒓)^2}+\overline{Q_2(𝒓)^2}=\overline{Q_{xx}(𝒓)^2}+\overline{Q_{xy}(𝒓)^2}={\displaystyle \frac{1}{4}},`$ (58)
where $`Q_a(𝒓)(a=1,2)`$ are the inverse Fourier transform of $`Q_a(𝒒)`$. To reduce the free energy (52), there arises an asymmetry $`Q_1(𝒓)^2>Q_2(𝒓)^2`$. In the limit where $`\mu \alpha ^2`$ is much larger than the disorder and Frank contributions to the free energy density, we expect from (58) to have
$`\overline{Q_1(𝒓)^2}={\displaystyle \frac{1}{4}},\overline{Q_2(𝒓)^2}=0,`$ (59)
which indeed is numerically confirmed . In this limit, the elastic free energy density is given by
$`f_{el}={\displaystyle \frac{\mu \alpha ^2}{8}},`$ (60)
as seen from (52). To the second order in $`\alpha `$, it is equal to the free energy in the monodomain state with $`\lambda =\lambda _m`$, as we can easily check by substituting (21) into (14) and expanding it with respect to $`\alpha `$. Thus we conclude that the elastic free energy change accompanied with the P-M transition is of $`O(\alpha ^3)`$, and the macroscopic stress averaged over the region $`1<\lambda <\lambda _m`$, or
$`{\displaystyle \frac{f_{el}(\lambda =\lambda _m)f_{el}(\lambda =1)}{\lambda _m1}},`$ (61)
is a quantity of $`O(\alpha ^2)`$.
To see the origin of the soft response, it is useful to look at the local elastic stress tensor, which is given in the harmonic approximation (23) as
$`\sigma _{ij}=\mu \left[{\displaystyle \frac{1}{2}}(_iu_j+_ju_i)\alpha Q_{ij}+R_{ij}\right].`$ (62)
Consider its variance $`\overline{\sigma _{ij}^2}`$. In the absence of random stresses, we have
$`{\displaystyle 𝑑𝒓\sigma _{ij}^2}`$ $`=`$ $`\mu F_{el}\mu ^2\alpha ^2{\displaystyle 𝑑𝒓Q_{ij}^2}`$ (63)
$`=`$ $`\mu ^2\alpha ^2{\displaystyle 𝑑𝒓\left(Q_1^2Q_{ij}^2\right)}`$ (64)
$`=`$ $`\mu ^2\alpha ^2{\displaystyle 𝑑𝒓Q_2^2},`$ (65)
which vanishes from (59). This means that each part of the system is stretched along the local director by $`1+\alpha /4+O(\alpha ^2)\lambda _m`$ times. This local elongation, realized by the checkered polydomain structure, reduces the free energy close to its absolute minimum.
## III Numerical Simulation
To further study non-linear mechanical response and effect of random disorder, we resort to numerical simulation by the continuum model. We utilize two different numerical schemes, one for the polydomain state in mechanical equilibrium at $`\lambda =1`$ and another for the P-M transition and dynamical effects. A two dimensional system is assumed for computational advantage. All the simulations below are performed on a $`N\times N`$ square lattice with $`N=128`$ unless otherwise stated. The grid spacing is chosen to be the unit of length. Periodic boundary conditions are imposed on $`𝒏(𝒓)`$ and $`𝒖(𝒓)`$, while the average strain $`\lambda `$ is externally controlled.
### A Polydomain state
First we study the the polydomain state in complete mechanical equilibrium and with no average strain ($`\lambda =1`$). To this end, we assume the harmonic free energy (23) and solve the linear equations (24) and (25) using fast Fourier transform. To minimize the free energy, we adapted a variant of the simulated annealing method . The orientational order parameter is evolved according to a Langevin equation,
$`{\displaystyle \frac{𝒏}{t}}`$ $`=`$ $`\mathrm{\Gamma }_n(𝖨𝒏𝒏)\left({\displaystyle \frac{F}{𝒏}}+𝜼\right),`$ (66)
where $`\mathrm{\Gamma }_n`$ is a constant and $`𝜼`$ is a “thermal” noise satisfying
$`𝜼(𝒓,t)𝜼(𝒓^{},t^{})=\eta _0^2𝖨\delta (𝒓𝒓^{})\delta (tt^{})`$ (67)
and Gaussian statistics. The noise strength $`\eta _0`$ is gradually reduced to zero until the end of each run. To be precise, we decrease $`\eta _0`$ to zero at a constant rate in the former half of a run, and set $`\eta _0=0`$ in the latter half. The initial noise strength and the annealing rate are chosen so that two different initial configurations, one with random and another with homogeneous director field, lead to indistinguishable results for the macroscopic quantities such as correlation function, average orientation, and free energy densities. As a standard set of static parameters we choose
$`\mu =400,\alpha =0.2,\beta =0.025,\xi _R=1,K=4,`$ (68)
for which $`\xi _c=0.5`$ and $`D=0.5`$. We integrated Eq.(66) using the Euler scheme with time increment $`\mathrm{\Delta }t=1`$ per step. A typical run consisted of $`5\times 10^4`$ time steps. Longer runs did not make an observable difference in the macroscopic quantities.
First we consider the orientational correlation function,
$`G(𝑹)`$ $`=`$ $`Q_{ij}(𝒓)Q_{ij}(𝒓+𝑹),`$ (69)
which is a function only of distance. We define the correlation length $`\xi `$ through
$`{\displaystyle \frac{G(\xi )}{G(0)}}={\displaystyle \frac{1}{2}}.`$ (70)
For each parameter set, we took statistical average over 20 samples. The data are shown in Fig.4. The decay of $`G(R)`$ is nearly exponential for strong disorder and faster than exponential for weak disorder. This qualitative tendency agrees with previous results for the 2D random-field XY model . The correlation length is a rapidly decreasing function of the effective disorder strength, $`D`$. The dependence is roughly exponential, also in agreement with previous results for the XY model . In the same figure we show the dependence of $`\xi `$ on $`\mu \alpha ^2`$, which is the measure of elastic interaction. Although the dependence is weak, the proper elastic interaction has an effect of increasing the correlation length. This is related to the enhancement of correlation in directions oblique to the local director, depicted in Fig.3. In order to quantify the director-relative correlation, we define the function
$`H(𝑹)`$ $`=`$ $`Q_{ij}(𝒓)Q_{ij}(𝒓+𝖴(𝒓)𝑹),`$ (71)
where $`𝖴(𝒓)`$ is a matrix of rotation that maps $`𝒏(𝒓)`$ to $`𝒆_x`$, or, explicitly,
$`𝖴`$ $`=`$ $`\left[\begin{array}{cc}\mathrm{cos}\theta & \hfill \mathrm{sin}\theta \\ \mathrm{sin}\theta & \hfill \mathrm{cos}\theta \end{array}\right].`$ (74)
By definition, $`H(x,0)`$ and $`H(0,y)`$ respectively describe the correlation in directions parallel and perpendicular to the local director. The data for the standard parameter are plotted in Fig.5. We see that the correlation is long-ranged in any specific direction, and the exponential-like decay in Fig.4 should be considered as a result of mutual cancellation of positive and negative correlation by taking the angular average.
A real space snapshot of the order parameter field $`Q_{xy}`$ is also given in Fig.5. As the grayscale shows, the contour $`Q_{xy}=0`$ preferentially lies in the horizontal ($`x`$-) and vertical ($`y`$-) directions. This corresponds to the checkered domain structure in Fig.3 (note that the grayscale is chosen so that the director is oblique to the horizontal axis in the brightest and darkest regions). More precisely, the checkered pattern is found on many different lengthscales, which is a natural consequence of the fact that the elastic interaction energy (27) is scale-independent.
An experimentally accessible way to characterize the anisotropic director correlation is the polarized light scattering. In a weakly inhomogeneous state, the depolarized (HV) light scattering intensity is given by
$`I(𝒒)=\left|Q_{xy}(𝒒)\right|^2,`$ (75)
except for a $`𝒒`$-independent prefactor. According to Ref. , the above formula holds even in a highly inhomogeneous state, if one assumes a two-dimensional configuration (see Eq.(2) in the reference). Our numerical data is shown in Fig.5. The intensity (75) is expressed in terms of $`Q_1`$ and $`Q_2`$ as $`I(q)=\mathrm{cos}\phi ^2|Q_1(𝒒)|^2+\mathrm{sin}\phi ^2|Q_2(𝒒)|^2`$, and the asymmetry $`Q_1>Q_2`$ explains the enhanced scattering on $`q_x`$\- and $`q_y`$\- axes .
Note that the peak is located at a small but finite wavenumber, in contrary to what is expected from the non-conserved nature of the orientational order parameter. In fact, we find it to be a finite size effect, and the peak wavenumber shrinks to zero as the system size $`N`$ is taken to infinity, leaving a singular minimum at the origin. To see this, we have computed the circularly averaged structure factor,
$`S(q)={\displaystyle _0^{2\pi }}𝑑\phi \left|Q_{ij}(𝒒)\right|^2,`$ (76)
for $`N=64`$,$`128`$, and $`256`$ systems, and found a peak in the region $`(2\pi /N)<q<2(2\pi /N)`$ in every case. The origin of the singular minimum at $`q=0`$ is explained as follows. Because of the periodic boundary condition on $`𝒖`$, the spatial average $`\overline{𝒖}`$ should complete vanish. This constraint suppresses formation of the checkered pattern with the check size larger than $`N/2`$.
### B P-M transition
Next we study the P-M transition using the non-linear elastic free energy (14). We found that complete minimization of the free energy takes very much computation time, and decided to take a more empirical approach : we utilize a simple dynamical model, and abandon to exclude non-equilibrium effects from the results. Fortunately, the stress-strain relation thus obtained is equilibrated to a good degree, because of fast relaxation of the rubber-elastic free energy. On the other hand, the domain structure exhibits a slow coarsening, which we study in the absence of external strain.
Our dynamical model consists of a set of equations that describe evolution of non-conserved order parameters in a simplest manner, namely,
$`{\displaystyle \frac{𝒏}{t}}`$ $`=`$ $`\mathrm{\Gamma }_n(𝖨𝒏𝒏){\displaystyle \frac{F}{𝒏}},`$ (77)
and
$`{\displaystyle \frac{𝒖}{t}}`$ $`=`$ $`\mathrm{\Gamma }_u{\displaystyle \frac{\delta F}{\delta 𝒖}}.`$ (78)
Instead of imposing the strict incompressibility condition (25), we penalized local volume change by adding an artificial potential $`F_v`$ to the free energy. By taking it in the form $`F_v=\frac{1}{2}𝑑𝒓\left[a_0(det\mathsf{\Lambda }1)^2+a_1(det\mathsf{\Lambda }1)^4\right]`$ and choosing appropriate values of the constants $`a_0`$ and $`a_1`$, we kept $`det\mathsf{\Lambda }`$ in the region $`[0.99,1.01]`$ throughout the runs.
We integrated (77) and (78) using the Euler scheme with $`\mathrm{\Gamma }_n=0.2`$, $`\mathrm{\Gamma }_u=0.02`$ and $`\mathrm{\Delta }t=1`$. A typical set of static parameters is same to that given by (68).
To prepare a polydomain state, we set site-wise random numbers to $`𝖰`$ and $`𝒖`$ as the initial condition, and integrated (77) and (78) for $`5\times 10^4`$ time steps with $`\lambda =1`$. Then we increased $`\lambda `$ at a constant rate $`d\lambda /dt=1\times 10^5`$ to induce the P-M transition. To check hysteresis, finally we decreased $`\lambda `$ back to unity at the rate $`d\lambda /dt=1\times 10^6`$.
Plotted in Fig.6 are the scaled macroscopic elastic stress $`\mu ^1\sigma _{macro}=\mu ^1(f_{el}/\lambda )`$ and the mean orientation $`S=\mathrm{cos}2\theta =2Q_{xx}`$ as functions of $`\lambda `$. We see from the figure that the elastic stress is vanishingly small and the orientation linearly increases in the polydomain region $`1<\lambda <\lambda _m(=1.05)`$. The stress shows a linear rise in the monodomain region $`\lambda >\lambda _m`$, where the orientation is nearly saturated to the maximum, $`S=1`$. While the strain-orientation curve has a small hysteresis, the strain-stress curve is almost completely reversible.
The smallness of the hysteresis manifests an important difference between the present system and random anisotropy magnets under magnetic field. In the latter, the macroscopic orientational order is broken solely by a random field. In contrast, in the present model, the monodomain state is unstable to an strain-mediated director buckling for $`\lambda <\lambda _m`$, even when there is no quenched disorder. This instability, which will be discussed in Section IV in detail, makes the P-M transition almost reversible.
The free energy densities are also shown in Fig.6. Both the proper and disorder parts of the rubber-elastic free energy change little in the region $`\lambda <\lambda _m`$. The latter curve has a slightly positive gradient. The situation is more subtle for the former. Its gradient is slightly positive in the figure, and turns to slightly negative for a smaller disorder strength. However, in the absence of random stresses and at $`\lambda =1`$, we had four domains whose sizes are limited by the system size, and the domain boundaries raise the elastic free energy. Because of this finite size effect, we cannot exactly tell the sign of the proper elastic stress in the macroscopic limit. We cannot exclude the possibility that the macroscopic stress completely vanishes in the limit of weak disorder.
The strain-stress and strain-orientation curves for larger values of $`\alpha `$ are given in Fig.7. Each curve shows a sharp crossover around $`\lambda =\lambda _m(\alpha )`$. The elastic stress in the polydomain region is vanishingly small even for large coupling. For any value of $`\alpha `$ studied, the changes of the proper and disorder elastic free energy densities, $`|f_{el}^P(\lambda =\lambda _m)f_{el}^P(\lambda =1)|`$ and $`|f_{el}^D(\lambda =\lambda _m)f_{el}^D(\lambda =1)|`$, were smaller than $`0.3`$ percent of $`\mu \alpha ^2`$. We find essentially no $`\alpha `$-dependence of the macroscopic stress in the polydomain region.
Shown in Fig.8 is the histogram of the elastic free energy contained in a lattice site. The distribution is fairly sharp and little changed by stretching for $`\lambda <\lambda _m`$, implying that the free energy is homogeneously minimized in the polydomain state. Real space snapshots of the domain morphology is shown in Fig.9. Pinned defects are observed just below the threshold $`\lambda =\lambda _m`$, while we find no defects remaining in the monodomain state.
The depolarized scattering intensity is shown in Fig.10. It has a minimum at $`q=0`$ and develops four peaks at finite wavenumbers. As we shall see in the next subsection, the peaks move toward the origin as the true equilibrium is approached and the domains coarsen. Here we concentrate on the effect of stretching. The peak intensity first increases and then decreases as a function of $`\lambda `$. Under stretching along the $`x`$-axis, the peaks on the $`q_x`$\- axis are more enhanced than those on the $`q_y`$\- axis. The shift of peak wavenumber by stretching is very small and difficult to estimate. By our choice of the stretching rate, the P-M transition completed in $`5\times 10^3`$ time steps, much before a significant coarsening can occur.
### C Slow structural relaxation
Now we turn to dynamical effects. First let us discuss the conditions under which Eqs.(77) and (78) are most reasonable as a model of dynamic evolution, not only as an artificial scheme of functional minimization. Firstly, Eq.(78) means that the velocity $`𝒖/t`$ is proportional to the force $`\delta F/\delta 𝒖`$. This applies to motion of a network in a viscous solvent , where we have a straightforward analogy to D’arcy’s law in porous media. On the other hand, in dry elastomers, there arises a viscous stress due to intra-network friction, which is proportional to $`(𝒖/t)`$. This is not accounted for in Eq.(78). Thus we consider that the dynamic model is more appropriate to swollen gels than to elastomers. Secondly, Eqs.(77) and (78) neglect dynamical coupling between the order parameters, i.e., the non-diagonal part of the Onsager coefficient matrix. This does not matter if the dynamics of the orientational order parameter is fast and slaved to that of the displacement field, which we expect to be the case. In fact, if the constituent polymer of the gel is not rigid, $`\mathrm{\Gamma }_n^1`$ is of the order of the viscosity of low-molecular weight fluids, $`\eta `$. On the other hand, the friction between the network and solvent renders $`\mathrm{\Gamma }_u^1`$ to be of the order of $`\eta /l^2`$, where $`l`$ is the mesh size of the network . Thus, the characteristic relaxation time of the strain at the scale of domain size is $`(\xi /l)^2(1)`$ times larger than that of $`𝖰`$.
Evolution of the structure factor $`S(q)`$ (as defined by Eq.(76)) is shown in Fig.11. The peak wavenumber decreases and the peak intensity increases as a function of time. Also shown in the figure is the structure factor at complete mechanical equilibrium, which is obtained by the numerical scheme used in Section III A. The correlation length $`\xi `$ and the inverse of the peak wavenumber $`q_0`$ are plotted in the middle of Fig.11. In the time region $`1\times 10^3<t<3\times 10^5`$, the former is well fitted by a power law $`\xi (t)t^ϵ`$ with $`ϵ=0.23\pm 0.02`$, and the latter grows almost in parallel to the former.
The Frank and rubber-elastic free energies are plotted as functions of time in Fig.11. While the elastic free energy changes little after an early stage of around $`t=10^3`$ time steps, the Frank energy density $`f_F`$ shows a slow and continuous decrease, which is approximately described by an power law $`f_Ft^ϵ^{}`$ in the region $`1\times 10^3<t<3\times 10^5`$, with $`ϵ^{}=0.22\pm 0.03`$.
Presently we have no explanation for the good fits of $`\xi (t)`$ and $`f_F(t)`$ by power laws. We keep ourselves to point out that the values of $`ϵ`$ and $`ϵ^{}`$ are much smaller than the corresponding exponents for the 2D non-conserved XY model without quenched disorder, which equal $`0.5`$ and $`1.0`$ from a simple scaling argument . The naive scaling relation $`ϵ^{}=2ϵ`$ is also broken here, which is not at all surprising if we consider the presence of quenched disorder . We should also stress that the final equilibrium values of $`\xi `$ and $`f_F`$ are finite. Preliminary study by a longer run without statistics finds a crossover from the power-law type kinetics to a slower one at $`t1\times 10^6`$ steps.
The above results show that the relaxation process can be decomposed into three characteristic stages : (i) The quench into the nematic phase from the isotropic phase produces microscopic textures, which coarsen to reduce both the rubber-elastic and Frank free energies. After the characteristic domain size reaches $`\xi _c`$, anisotropic domain reconfiguration on this scale follows. The rubber-elastic free energy is almost completely minimized at this early stage, because of the scale-independence of the proper elastic interaction (27). (ii) The “checkered” domain structure further coarsens to reduce the Frank free energy. The domain size $`\xi `$ and the peak wavelength $`2\pi /q_0`$ grow in parallel to each other. (iii) The domain size converges to a finite equilibrium value, while the anisotropic domain reconfiguration proceeds on larger scales ($`lim_t\mathrm{}q_0(t)=0`$).
## IV Fluctuation in the monodomain state
Recall that, if there is no quenched disorder, the ground state of the system is the macroscopically elongated state with $`\lambda =\lambda _m`$ and $`𝒖=0`$. In this state, there are so-called soft modes of director fluctuation, which do not accompany any change in the rubber elastic free energy . The presence of the soft modes implies that a homogeneous director configuration becomes unstable for $`\lambda <\lambda _m`$; when we compress the gel along the optical axis, the director “buckles” to partially cancel the rise of elastic free energy by compression. The result of the previous section means that this instability is almost completely soft even for large deformations. In this section, we look at the monodomain region $`\lambda \lambda _m`$ and analyze the director fluctuation modes in a harmonic level. It was suggested in Ref. that the soft fluctuations at the critical point $`\lambda =\lambda _m`$ is strongly enhanced by quenched disorder and satisfy $`|\delta 𝒏(𝒒)|^2q^4`$. Nonetheless, it was not fully confirmed because of a breakdown of the harmonic approximation at the critical point, $`\lambda =\lambda _m`$. Also, the model used in remains largely phenomenological, and a quantitative assessment of the prediction is necessary. Here we extend the analysis to arbitrary values of $`\lambda `$, and discuss the possibility to find disorder-enhanced fluctuations in practical situations.
The director is decomposed into a homogeneous part and a small deviation, as
$`𝒏(𝒓)=𝒆_x+\delta 𝒏(𝒓).`$ (79)
Expanding the basic free energy (14) with respect to $`\delta 𝒏`$ and $`𝒖`$, and then eliminating the elastic field using the mechanical equilibrium condition, we obtain an effective free energy in terms of $`\delta 𝒏`$. An outline of the calculation is given in Appendix A. For the three dimensional case, the result is
$`\stackrel{~}{F}_{el}`$ $`=`$ $`{\displaystyle \frac{\mu \alpha }{\lambda }}{\displaystyle _𝒒}[{\displaystyle \frac{1}{2}}A_1(\lambda ,\widehat{𝒒})\left|\delta 𝒏(𝒒)\right|^2+{\displaystyle \frac{1}{2}}A_2(\lambda ,\widehat{𝒒})|\widehat{𝒒}\delta 𝒏(𝒒)|^2`$ (84)
$`R_{ix}^{}(𝒒)\delta n_i(𝒒)`$
$`B_1(\lambda ,\widehat{𝒒})\widehat{q}_iR_{ix}^{}(𝒒)\left(\widehat{𝒒}\delta 𝒏(𝒒)\right)`$
$`B_2(\lambda ,\widehat{𝒒})\left(\widehat{𝒒}𝖱^{}(𝒒)\delta 𝒏(𝒒)\right)`$
$`B_3(\lambda ,\widehat{𝒒})(𝖱^{}(𝒒):\widehat{𝒒}\widehat{𝒒})(\widehat{𝒒}\delta 𝒏(𝒒))],`$
$`A_1(\lambda ,\widehat{𝒒})`$ $`=`$ $`\lambda ^31{\displaystyle \frac{3\alpha }{3+4\alpha }}{\displaystyle \frac{\lambda ^6\widehat{q}_x^2}{1+(\lambda ^31)\widehat{q}_x^2}},`$ (85)
$`A_2(\lambda ,\widehat{𝒒})`$ $`=`$ $`{\displaystyle \frac{\frac{3\alpha (3+4\alpha )}{3+\alpha }\frac{(\frac{3+\alpha }{32\alpha }+\lambda ^3)^2\widehat{q}_x^2}{3+\alpha (2+\widehat{q}_x^2)}\frac{3\alpha }{32\alpha }}{1+(\lambda ^31)\widehat{q}_x^2}},`$ (86)
$`B_1(\lambda ,\widehat{𝒒})`$ $`=`$ $`{\displaystyle \frac{1+\frac{3\alpha }{3+\alpha }(\frac{3+\alpha }{32\alpha }+\lambda ^3)\widehat{q}_x^2}{1+(\lambda ^31)\widehat{q}_x^2}},`$ (87)
$`B_2(\lambda ,\widehat{𝒒})`$ $`=`$ $`{\displaystyle \frac{\lambda ^3\widehat{q}_x}{1+(\lambda ^31)\widehat{q}_x^2}},`$ (88)
$`B_3(\lambda ,\widehat{𝒒})`$ $`=`$ $`{\displaystyle \frac{(\frac{3+\alpha }{32\alpha }+\lambda ^3)\widehat{q}_x}{1+(\lambda ^31)\widehat{q}_x^2}},`$ (89)
$`R_{ij}^{}(𝒒)`$ $`=`$ $`\overline{\mathrm{\Lambda }}_{ik}\overline{\mathrm{\Lambda }}_{jl}R_{kl}(𝒒),`$ (90)
which is correct to the bilinear order in $`\delta 𝒏`$ and $`𝖱`$ (we neglect terms independent of $`\delta 𝒏`$).
In the absence of quenched disorder, the integrand in (84) is of the form $`\frac{1}{2}(A_1𝖨+A_2\widehat{𝒒}\widehat{𝒒}):\delta 𝒏(𝒒)\delta 𝒏(𝒒)`$. It is convenient to introduce two unit vectors $`𝒆_1=𝒒\times 𝒆_x/|𝒒\times 𝒆_x|`$ and $`𝒆_2=𝒆_1\times 𝒆_x`$ , with which the integrand becomes
$`{\displaystyle \frac{1}{2}}A_1\left|𝒆_1\delta 𝒏(𝒒)\right|^2+{\displaystyle \frac{1}{2}}\left(A_1+A_2(1\widehat{q}_x^2)\right)\left|𝒆_2\delta 𝒏(𝒒)\right|^2.`$ (91)
At $`\lambda =\lambda _m`$, the coefficient $`A_1`$ takes its minimum value $`0`$ for $`𝒒𝒆_x`$, while $`A_1+A_2(1\widehat{q}_x^2)`$ is minimized and vanishes both on the line $`𝒒𝒆_x`$ and in the plane $`𝒒𝒆_x`$. These correspond to the soft modes. Similar results have been obtained by Olmsted for monodomain elastomers crosslinked in the nematic phase and without external strain.
The random stresses shift the ground state to an inhomogeneous state, $`\delta 𝒏=\delta 𝒏_R`$ and $`𝒖=𝒖_R`$. The frozen director deviation $`\delta 𝒏_R`$ is obtained by minimizing the total free energy $`\stackrel{~}{F}_{el}+F_F`$ with respect to $`\delta 𝒏`$, as
$`\delta n_{R,i}(𝒒)`$ $`=`$ $`{\displaystyle \frac{1}{A_1+K^{}q^2}}[R_{ix}^{}(𝒒)+B_2\widehat{q}_jR_{ij}^{}(𝒒)`$ (94)
$`+\left(B_1{\displaystyle \frac{A_2(1+B_1)}{A_1+A_2+K^{}q^2}}\right)\widehat{q}_i\widehat{q}_jR_{jx}^{}(𝒒)`$
$`+(B_3{\displaystyle \frac{A_2(B_2+B_3)}{A_1+A_2+K^{}q^2}})\widehat{q}_i(\widehat{𝒒}\widehat{𝒒}:𝖱^{}(𝒒))],`$
where we have introduced a scaled Frank constant,
$`K^{}`$ $`=`$ $`{\displaystyle \frac{K\lambda }{\mu \alpha }}.`$ (95)
At the critical point $`\lambda =\lambda _m`$, the quantity $`A_1+A_2`$ appearing in (94) vanishes for $`𝒒𝒆_x`$. Hence, in the long wavelength limit $`q0`$, we have $`\delta n_R(𝒒)q^2`$ in the soft directions $`𝒒𝒆_x`$ and $`𝒒𝒆_x`$. This means a divergence of the real space amplitude $`|𝒏_R(𝒓)|^2`$ and breakdown of the harmonic approximation, as pointed out in Ref.. Severer is the divergence of the frozen elastic field $`𝒖_R`$, which is related to $`\delta 𝒏_R`$ through the mechanical equilibrium condition (A3). At $`\lambda =\lambda _m`$, it behaves as $`|𝒖_R(𝒒)|^2q^6`$ in the soft directions, and $`|𝒖_R(𝒓)|^2`$ diverges. However, these divergences disappear for $`\lambda >\lambda _m`$, where the excess stretching acts as a stabilizing field. Now we consider this region. The condition for the harmonic approximation to be valid is $`|\delta 𝒏_R(𝒓)|^21`$, which implies $`|𝒖_R(𝒓)|^2\alpha ^2`$. To assess the condition by order estimate, we concentrate on the plane $`𝒒𝒆_x`$, from where arises the most significant contribution to the real space amplitude,
$`\left|\delta 𝒏_R(𝒓)\right|^2={\displaystyle \frac{1}{(\text{system’s volume})}}{\displaystyle _𝒒}\left|\delta 𝒏_R(𝒒)\right|^2.`$ (96)
On that plane, the strongest $`q`$-dependence of $`\delta 𝒏_R(𝒒)`$ comes from a factor $`(\mathrm{\Delta }+K^{}q^2)^1`$, where
$`\mathrm{\Delta }=\mathrm{\Delta }(\lambda )=\left[A_1(\lambda ,\widehat{𝒒})+A_2(\lambda ,\widehat{𝒒})\right]|_{\widehat{q}_x=0}.`$ (97)
is the measure of excess stretching. Note that $`\mathrm{\Delta }\lambda \lambda _m`$ for $`\lambda \lambda _m1`$. A saddle-point approximation around the plane yields
$`\left|\delta 𝒏_R(𝒓)\right|^2`$ $``$ $`{\displaystyle \frac{\beta ^2\xi _R^3q^2dq}{(\mathrm{\Delta }+K^{}q^2)^{3/2}}}`$ (101)
$`\{\begin{array}{cc}{\displaystyle \frac{\beta ^2\xi _R^3}{K^{3/2}}}\mathrm{ln}\left({\displaystyle \frac{K^{}q_{max}^2}{\mathrm{\Delta }}}\right)& (\mathrm{\Delta }K^{}q_{max}^2),\\ {\displaystyle \frac{\beta ^2\xi _R^3}{K^{3/2}}}\left({\displaystyle \frac{K^{}q_{max}^2}{\mathrm{\Delta }}}\right)^{3/2}& (\mathrm{\Delta }K^{}q_{max}^2),\end{array}`$
where $`q_{max}`$ is the upper cutoff wavenumber. We may roughly identify $`2\pi /q_{max}`$ with the network mesh size, $`l(k_BT/\mu )^{1/3}`$. Using typical experimental values $`\mu =10^5`$ J/m<sup>3</sup>, $`K=10^{11}`$ J/m, $`\alpha =1.0`$ and $`T=300`$ K, we have $`\sqrt{K^{}}=10`$ nm and $`K^{}q_{max}^21`$. For $`\mathrm{\Delta }\stackrel{<}{}K^{}q_{max}^2`$ and except in the close vicinity of the criticality, $`\mathrm{\Delta }=0`$, the amplitude only weakly depends on the elongation ratio. In this region, the harmonic approximation is valid if and only if
$`{\displaystyle \frac{\beta ^2\xi _R^3}{K^{3/2}}}1,`$ (102)
which is satisfied when we put $`\xi _R10^2`$ nm and $`\beta 0.01`$ as a trial.
The director exhibits a thermal fluctuation around the inhomogeneous ground state. Its amplitude is not affected by the quenched randomness, at least within the harmonic calculation. The total fluctuation amplitude is given by
$`P^{(a)}(𝒒)`$ $`=`$ $`\left|𝒆_a\delta 𝒏(𝒒)\right|^2=P_T^{(a)}(𝒒)+P_R^{(a)}(𝒒),`$ (103)
$`P_T^{(a)}(𝒒)`$ $`=`$ $`{\displaystyle \frac{k_BT\lambda }{\mu \alpha }}{\displaystyle \frac{1}{A_1+\delta _{a2}(1\widehat{q}_x^2)A_2+K^{}q^2}},`$ (104)
$`P_R^{(a)}(𝒒)`$ $`=`$ $`\left|𝒆_a\delta 𝒏_R(𝒒)\right|^2,`$ (105)
where $`a=1,2`$, and $`P_T^{(a)}`$ and $`P_R^{(a)}`$ are the thermal and frozen contributions, respectively. Let us compare the two contributions. To be explicit, we compare $`P_T^{(2)}(𝒒)`$ and $`P_R^{(2)}(𝒒)`$ on the plane $`𝒒𝒆_x`$. There the ratio $`P_R^{(2)}/P_T^{(2)}`$ is controlled by a factor of the form $`\mathrm{\Delta }_c/(\mathrm{\Delta }+K^{}q^2)`$, where
$`\mathrm{\Delta }_c={\displaystyle \frac{\mu \alpha \beta ^2\xi _R^3}{k_BT}}`$ (106)
defines a crossover point. In the region $`\mathrm{\Delta }\stackrel{<}{}\mathrm{\Delta }_c`$, the disorder part of the fluctuation dominates the thermal part at long wavelengths. We estimate $`\mathrm{\Delta }_c10^3`$ using the above mentioned values, for which the crossover length $`\sqrt{K^{}/\mathrm{\Delta }_c}10^210^3`$ nm is around or below the wavelength of visible light. The amplitudes for $`\alpha =0.2`$ and $`\lambda /\lambda _m=1.001`$ are plotted in Fig.12. We see that the anisotropy of the scattering pattern is not much affected by the quenched disorder.
The director fluctuation amplitudes are closely related to polarized light scattering intensity . By comparing experimental results to the above calculation, we may extract information on the network heterogeneity. In particular, if the macroscopic orientation is not saturated in the monodomain state, it means the presence of large-scale quenched strains that do not meet the condition (102). For instance, in an optical study of a swollen monodomain gel by Chang et al. , speckles on the few-micrometer scale are observed, which is attributed to heterogeneities as we consider here. We hope that the scattering intensity will be measured as a function of applied strain. Another origin of large scale heterogeneity will be discussed in the next section.
## V Effect of crosslinking condition
In this section, we consider the case of anisotropic crosslinking. Melts of nematic polymers often exhibit long-lived polydomain textures after a quench from the isotropic phase . The size of the domains is macroscopic and typically of micron order. When such a melt is crosslinked, its non-uniform orientation is imprinted into the network. We denote the initial configuration by $`𝖰_0(𝒓)`$. The extended affine-deformation theory prescribes the elastic free energy,
$`F_{el}={\displaystyle \frac{\mu }{2}}{\displaystyle 𝑑𝒓\text{Tr}\left[(𝖨+\alpha _0𝖰_0)\mathrm{\Lambda }^T(𝖨\alpha 𝖰)\mathrm{\Lambda }𝖨\right]},`$ (107)
where $`\alpha _0`$ is expressed in terms of the parameters used in as
$`\alpha _0={\displaystyle \frac{\mathrm{}_{}\mathrm{}_{}}{(1/d)\mathrm{}_{}+(11/d)\mathrm{}_{}}},`$ (108)
or equivalently, $`\alpha _0=\alpha /[1(12/d)\alpha ]`$. Note that the above free energy is obtained by just formally replacing $`𝖱`$ with $`\alpha _0𝖰_0`$ in the free energy (14). Thus, the initial texture field $`𝖰_0`$ provides a source of quenched disorder. An effective disorder strength that corresponds to (38) can be defined by
$`D={\displaystyle \frac{\mu \alpha \alpha _0}{K}}\xi _0^2.`$ (109)
where $`\xi _0`$ is the correlation length of the initial texture. If we set $`\xi _0=1\mu `$m and $`\alpha =0.1`$, we have a very large number $`D(\xi _0/\xi _c)^2=10^310^5`$. Since the orientational order is predominantly affected by this strong disorder, we do not take into account other mesoscopic sources of quenched stresses, which is legitimate as a first approximation.
We have simulated the P-M transition in the following way. To generate the initial configuration, we mimicked the phase ordering kinetics of nematic polymer solutions by numerically solving the equation,
$`{\displaystyle \frac{𝒏}{t}}`$ $`=`$ $`\mathrm{\Gamma }_n(𝖨𝒏𝒏){\displaystyle \frac{F_F}{𝒏}}.`$ (110)
Taking a sitewise-random director configuration as the initial condition, we integrated Eq.(110) over 50 time steps (with $`\mathrm{\Gamma }_n=0.2`$ and $`\mathrm{\Delta }t=1`$) to generate $`𝖰_0(𝒓)`$. After crosslinking the system by adding $`F_{el}`$ to the free energy and setting $`\lambda =1`$ and $`𝒖=0`$, we integrated Eqs.(77) and (78) for $`1\times 10^4`$ time steps to equilibrate the system. The mechanical response was studied in just the same way as described in Section III.
Fig.13 shows the strain-stress and strain-orientation curves for $`\alpha =0.2,0.4,0.8,`$ and $`1.2`$. The strain-stress curve bends at a value of $`\lambda `$ where the director is yet far from aligned. For $`\alpha 0.8`$, the gradient of the strain-stress curve has a non-monotonic dependence on the strain, and is smallest at an intermediate value of $`\lambda `$. The strain-orientation curve shows only a gradual crossover to the monodomain state, especially for larger values of $`\alpha `$. The slope of the scaled elastic stress $`\mu ^1\sigma _{macro}`$ is roughly independent of $`\alpha `$ in the vicinity of the point $`\lambda =1`$.
Evolution of the director texture during the P-M transition is shown in Fig.14, and the distribution of the elastic free energy in Fig.15. At $`\lambda =1`$ the director texture is almost same as that just before crosslinking, or $`𝖰(𝒓)=𝖰_0(𝒓)`$. This is reflected in the extremely homogeneous free energy distribution. External strain strongly dehomogenize the distribution, and the peak is continuously broadened as we increase $`\lambda `$ toward the monodomain region.
## VI Discussion and Summary
We have studied polydomain nematic networks from two aspects, namely, (i) breaking of long-range orientational order by frozen internal stresses and (ii) a non-local inter-domain interaction arising from strain-orientation coupling. The mechanical response is controlled by the latter if the quenched disorder is of moderate strength (or, if $`D\stackrel{<}{}\mathrm{\hspace{0.33em}1}`$). In this case, the proper elastic interaction reorganizes the polydomain structure so that local elongation along the director is achieved everywhere in the system. The resulting structure contains the checkered correlation pattern on various scales, which produces the “four-leaf clover” pattern in the depolarized scattering intensity. Upon stretching, the director and the local strain axis coincidently rotate toward the direction of macroscopic extension, and thus the elastic free energy keeps almost constant until a complete alignment is attained. The change in elastic free energy accompanying the P-M transition is analytically estimated to be of $`O(\alpha ^3)`$. This result does not depend on a specific model of non-linear elasticity. In fact, we obtained it by harmonic expansion of the elastic free energy, which is unique from symmetry . Numerical simulation reveals a more complete softness, and we find essentially no $`\alpha `$-dependence of the average macroscopic stress (61). We cannot exclude the theoretical possibility that the P-M transition is exactly soft in the weak disorder limit. We may say that there are mechanical quasi-Goldstone modes, which are distinguished from the genuine Goldstone modes of fluid nematic liquid crystals in that there is an anisotropic correlation even in the absence of external field and that the quenched disorder selects a characteristic lengthscale.
We have discussed two sources of quenched disorder. One is the random stress due to residual heterogeneous strains at the moment of crosslinking, considered ubiquitous in rubbery networks. The anisotropic (shear) part of random stresses act on the orientational order, both locally and non-locally. The macroscopic domain size observed in experiments can be explained if there are frozen heterogeneities of a reasonably small magnitude (e.g. $`1`$ percent in strain) and a size somewhat larger than that of individual network meshes (e.g. $`10^2`$ nm).
A different viewpoint is taken in previous theories , where a random molecular field operating at crosslinks is assumed to be the source of disorder. The random field hypothesized there has a small correlation length roughly equal to the distance $`l`$ between crosslinks. The ratio $`\xi /l`$ is a large number ($`10^3`$), which means a weak effective disorder.
Currently we know of no firm experimental indication of the disorder strength. A possible method of its estimate is to observe director fluctuation in the monodomain state. We have calculated thermal and disorder contributions to the fluctuation amplitude, which is proportional to the polarized light scattering intensity. The intensity diminishes as we stretch the network, and there is a region of macroscopic strain $`\lambda `$ where the disorder contribution dominates over the thermal one. The width of the region and the absolute value of the intensity should inform us the order of the disorder strength. The thermal and quenched contributions could be separately analyzed by use of dynamical light scattering (DLS). Indeed, DLS has been successfully used to decompose the two kinds of density fluctuation in gels . It is hoped that a similar method will be developed for orientation fluctuation in the present system.
By crosslinking the network in the nematic phase and in the course of phase ordering, we obtain another kind of quenched stresses. The polydomain texture of the liquid-crystalline polymer melt is almost completely frozen by crosslinking, if its characteristic size is larger than $`\xi _c`$. The memory of the initial macroscopic texture makes the mechanical response non-soft. Spatial distribution of the elastic free energy is strongly dehomogenized by applied strain, in contrast to the case of isotropic crosslinking.
The influence of crosslinking conditions has little been discussed in previous studies of the P-M transition, except for a few experimental papers . Küpfer and Finkelmann studied both isotropic and anisotropic crosslinkings under external stress of various magnitudes. Fig.8 in the reference shows that polydomain networks crosslinked in the nematic phase are harder than those prepared in the isotropic phase. Another example of soft and non-soft P-M transitions is given in Ref. , where it is stated that some of their samples were prepared above the isotropic-nematic transition temperature of the melt, while the others are crosslinked below it. Unfortunately, they do not explicitly state the crosslinking condition for each stress-strain curve. We wish further effort in this direction to be made in the future, especially to find more evidence of vanishing macroscopic stress.
A remark should be made in relation to this. We have assumed that the quenched heterogeneities have mesoscopic sizes in the case of isotropic crosslinking. However, if the network is crosslinked in poor solvents or near the spinodal line, the heterogeneities can be macroscopic and cause strong effective disorder. Therefore, the mechanical response should be discussed in terms of the size of the heterogeneity, not only on the phase where the gel is fabricated. Another problem in interpretation of strain-stress data arises from slowness of dynamical relaxation. A recent dynamic measurement by Clarke and Terentjev strongly suggests that the stress level will be substantially lowered in the final equilibrium state, which is not reachable on a practical timescale. It might be possible that a soft equilibrium P-M transition is masked behind a stress plateau of a sizable height, which is reported in earlier studies .
We have studied dynamical relaxation after a quench from the isotropic phase. The structure factor develops a peak at a finite wavenumber, which goes to zero as the true equilibrium is approached. Both the inverse peak wavenumber and the correlation length show a power-law type growth in an intermediate stage, while the elastic free energy is almost completely minimized in an early regime of the coarsening process.
Some of the experimentally observed features of the “four-leaf clover” scattering pattern have been reproduced in the present work. Firstly, we propose that the finiteness of the observed peak wavenumber is explained by the slow relaxation. The experimental peak wavenumber does not change during the P-M transition . Together with our simulation result, it suggests that the coarsening is very slow and does not occur in the timescale of observation. Further experimental study of structural relaxation in conjunction with stress relaxation would be informative to check this point. Secondly, the peak intensity increases and then decreases as we stretch the gel. Qualitatively the same monotonic behavior is reported in the experiment. The initial increase is due to a sharpening the peak, which is partially understood by the fact that the director fluctuation at $`\lambda =\lambda _m`$ is soft only on a plane and a line in the $`𝒒`$-space.
We close by listing some open questions. (i) We did not answer whether the long-range order is destroyed by an arbitrarily weak disorder under no external stress or, equivalently, when the average strain $`\lambda `$ is not externally constrained. A shift of the ground state from the monodomain ($`\lambda =\lambda _m`$) to polydomain ($`\lambda =1`$) states should occur, either gradually or abruptly, as we increase the disorder strength from zero. Probably this problem is not of practical importance because of a small but finite hysteresis and the slow dynamics. (ii) Stretching-induced-anisotropy of the depolarized scattering pattern as we numerically find is contrary to the experimental observation. We may suggest an effect of spatial dimensionality. In three dimensions there are three Frank constants, whose relative strengths may affect the anisotropy. Experimental investigation of 3D domain structure would be informative. (iii) Much remains to be done for understanding dynamical relaxation to the final equilibrium state. In theoretical part, the origin of the apparent power law is yet unknown. Dynamic equations for dry elastomers are to be constructed, taking the intra-network friction account. In numerical part, late stages of the relaxation process is left unexplored. Stress relaxation for strong quenched disorder and after stretching should be addressed to make a comparison to experiment. As these necessitate extensive computation, we leave them for future work.
###### Acknowledgements.
The author is grateful to Professor Akira Onuki for helpful comments and discussions. He also thanks Professor Ken Sekimoto, Dr. Alexandra ten Bosch, and Dr. Jun Yamamoto for valuable discussions.
## A Effective free energy in the monodomain state
Here we sketch the derivation of Eq.(84). Substituting Eqs.(18), (19) and (20) into Eq.(14), we have
$`F_{el}`$ $`=`$ $`{\displaystyle \frac{\mu }{2}}{\displaystyle }d𝒓[C_{ij}L_{ij}+2C_{ik}L_{jk}(_iu_j)`$ (A2)
$`+\overline{C}_{ij}\overline{L}_{kl}(_iu_k)(_ju_l)+\kappa (_iu_i)^2],`$
where $`C_{ij}=(\delta _{kl}+R_{kl})\overline{\mathrm{\Lambda }}_{ik}\overline{\mathrm{\Lambda }}_{jl}`$ and $`L_{ij}=\delta _{ij}\alpha Q_{ij}=(1+\alpha /3)\delta _{ij}\alpha n_in_j`$. In the third term of the integrand we have replaced $`C_{ij}`$ and $`L_{ij}`$ with their spatial averages as the deviations will contribute only to higher order terms in the effective free energy. The last term is added by hand to temporarily relax the incompressibility condition (25), which is recovered by taking the limit $`\kappa \mathrm{}`$ afterwards. The condition of mechanical equilibrium (24) can be written as
$`_i(C_{ik}L_{jl})+\overline{C}_{ij}\overline{L}_{kl}_i_ju_k+\kappa _i_ju_j=0`$ (A3)
Taking the incompressible limit $`\kappa \mathrm{}`$, we have
$`𝒖(𝒒)={\displaystyle \frac{1}{𝖢:𝒒𝒒}}\left[𝖫^1𝒈(𝒒){\displaystyle \frac{𝖫^1𝒒}{𝖫^1:𝒒𝒒}}\left(𝒒𝖫^1𝒈(𝒒)\right)\right],`$ (A4)
where $`𝒈`$ is an auxiliary variable defined by
$`g_i(𝒓)`$ $`=`$ $`_j(C_{jk}L_{ik}).`$ (A5)
Substituting (A4) into (A2), we obtain an effective free energy,
$`\stackrel{~}{F}_{el}`$ $`=`$ $`{\displaystyle \frac{\mu }{2}}{\displaystyle 𝑑𝒓𝖢}:𝖫`$ (A6)
$`+`$ $`{\displaystyle \frac{\mu }{2}}{\displaystyle _𝒒}{\displaystyle \frac{1}{\overline{𝖢}:\widehat{𝒒}\widehat{𝒒}}}[{\displaystyle \frac{1}{\overline{𝖫}^1:\widehat{𝒒}\widehat{𝒒}}}|\widehat{𝒒}\overline{𝖫}^1𝒈(𝒒)|^2`$ (A8)
$`𝒈(𝒒)\overline{𝖫}^1𝒈(𝒒)],`$
We arrive at Eq.(84) by putting (18) into $`C_{ij}`$, (79) into $`L_{ij}`$, and the resulting expressions into (A5) and (A8).
|
warning/0004/math0004027.html
|
ar5iv
|
text
|
# Untitled Document
Supported by the DFG-project HI 412/5-2 <sup>∗∗</sup>Supported by LEQSF grant (1996-99)-RD-A-12
The $`c`$-function for non-compactly causal symmetric spaces
Bernhard Krötz and Gestur Ólafsson<sup>∗∗</sup>
Introduction
In this paper we prove a product formula for the $`c`$-function associated to a non-compactly causal symmetric space $``$. Let us recall here the basic facts. Let $`G`$ be a connected semisimple Lie group, $`\tau :GG`$ be a non-trivial involution and $`H=G^\tau `$. Then $`:=G/H`$ is a semsimple symmetric space. The space $``$ is called non-compactly causal, if $`q:=\{Xg:\tau (X)=X\}`$ contains an open $`H`$-invariant hyperbolic cone $`CØ`$. In this case $`S:=H\mathrm{exp}(C)`$ is a open subsemigroup of $`G`$. A spherical function on $``$ is an $`H`$-biinvariant continuous function on $`S/H`$, which defines an eigendistribution of the algebra of $`H`$-invariant differential operators on $``$, see \[FHÓ94\], \[KNÓ98\], \[Ól97\]. There exists a maximal abelian hyperbolic subspace $`a`$ of $`q`$ such that $`C=Ad(H).(aC)`$. Let
$$\phi _\lambda (g.x_0)=_Ha_H(gh)^{\lambda \rho }d\mu _H(h)$$
be a spherical function given by a convergent integral similar to the expression for the spherical functions on the Riemannian symmetric spaces $`G/K`$. Here $`x_0`$ is the coset $`\{H\}`$ and $`a_H(g)A:=\mathrm{exp}(a)`$ is determined by $`gHa_H(g)N`$. The asypmtotic behaviour of $`\phi _\lambda (a.x_0)`$ along $`SA`$ is given by $`\phi _\lambda (a.x_0)c(\lambda )a^{\lambda \rho }`$, where $`\rho `$ is half the sum over the positive roots counted with multiplicities. The function $`c(\lambda )`$ is the $`c`$-function of the space $``$. It turns out that the $`c`$-function is a product of two $`c`$-function, $`c(\lambda )=c_\mathrm{\Omega }(\lambda )c_0(\lambda )`$ where $`c_0(\lambda )`$ is the Harish-Chandra $`c`$-function of a Riemannian subsymmetric space $`G(0)/K(0)`$ and $`c_\mathrm{\Omega }(\lambda )`$ is a function associated to the real bounded symmetric domain $`H/(HK)`$, where $`K`$ is a $`\tau `$-stable maximal compact subgroup of $`G`$. The $`c`$-function was first introduced by Oshima-Sekiguchi in \[OS80\], whereas $`c_\mathrm{\Omega }(\lambda )`$ was first introduced in \[FHÓ94\].
The $`c`$-function for a Riemannian symmetric space $`G/K`$ can be written as a product of $`c`$-functions of rank one symmetric spaces associated to each restricted root of $`g`$ (Gindikin–Karpelevic formula). For general non-Riemannian symmetric spaces $`G/H`$ one cannot expect this type of result. However, for non-compactly causal symmetric spaces we show in this paper (cf. Theorem III.5) that such a product formula holds. The case of Cayley type spaces has already been treated by J. Faraut in \[Fa95\] by the use of Jordan algebra methods and in \[Gr97\] the case $`Sl(n,R)/SO(p,q)`$ is dealed. The approach presented here is general, different and relies on new insights on the fine convex geometry of the real bounded symmetric domain $`\mathrm{\Omega }`$ (cf. Theorem II.5 and Theorem II.7.)
Our result has important applications. The $`c`$-function was the last unknown part in the formula for the formal degree of the spherical holomorphic discrete series representations representations (cf. \[Kr99\]). Further it gives us important information on the normalized spherical functions $`\stackrel{~}{\phi }_\lambda :=c_\mathrm{\Omega }(\lambda )^1\phi _\lambda `$. One knows that the function $`\lambda \stackrel{~}{\phi }_\lambda (s.x_0)`$ has a meromorphic continuation to $`a_C^{}`$ (cf. \[Ól97\]) and the product formula gives us important information on the poles. In particular, this allows more detailed analysis of the spherical Laplace transform, in particular Paley-Wiener type theorems.
I. Non-compactly causal symmetric spaces and Lie algebras
In this section we introduce notation and recall some facts concerning non-compactly causal symmetric Lie algebras and their associated symmetric spaces. Our source of reference is \[HiÓl96\].
Algebraic preliminaries
Let $`g`$ be a simple finite dimensional real Lie algebra. Let $`\tau :gg`$ be a non-trivial involution. Then $`(g,\tau )`$ is a symmetric Lie algebra. We write $`g=h+q`$ for the $`\tau `$-eigenspace decomposition of $`g`$ corresponding to the eigenvalues $`+1`$ and $`1`$. Let $`\theta `$ be a Cartan involution of $`g`$ which commutes with $`\tau `$ and let $`g=k+p`$ be the associated Cartan decomposition.
For $`a,bg`$ let $`z_a(b):=\{Xa:[X,Y]=0,Yb\}`$ be the centralizer of $`b`$ in $`a`$. We call $`(g,\tau )`$ non-compactly causal, or simply NCC, if $`z_{qp}(hk)\{0\}`$. We call $`(g,\tau )`$ non-compactly Riemannian (NCR) if $`\tau `$ is a Cartan involution. If not otherwise stated from now on $`(g,\tau )`$ denotes a NCC symmetric Lie algebra. Then $`z_{qp}(hk)=z(qp)=RX_0`$ is one dimensional. Let $`aqp`$ be a maximal abelian subspace and note that $`RX_0a`$ and that $`a`$ is maximal abelian in $`p`$. We write $`\mathrm{\Delta }=\mathrm{\Delta }(g,a)`$ for the root system of $`g`$ with respect to $`a`$ and
$$g=z_g(a)\underset{\alpha \mathrm{\Delta }}{}g^\alpha $$
for the corresponding root space decomposition. We write $`g(0):=hk+qp`$ and note that $`(g(0),\tau (0))`$, with $`\tau (0):=\tau _{g(0)}`$, is NCR. If $`\alpha \mathrm{\Delta }`$ then either $`g^\alpha g(0)`$ or $`g^\alpha qk+hp`$. A root $`\alpha \mathrm{\Delta }`$ is called compact if $`g^\alpha g(0)`$ and non-compact if $`g^\alpha qk+hp`$. We write $`\mathrm{\Delta }_k`$ and $`\mathrm{\Delta }_n`$ for the collection of compact and non-compact roots, respectively. Note that $`\mathrm{\Delta }=\mathrm{\Delta }_k\dot{}\mathrm{\Delta }_n`$.
We can and will normalize $`X_0`$ such that $`Spec(adX_0)=\{1,0,1\}`$. Then $`\mathrm{\Delta }_k=\{\alpha \mathrm{\Delta }:\alpha (X_0)=0\}`$ and we can choose a positive system $`\mathrm{\Delta }^+`$ of $`\mathrm{\Delta }`$ such that
$$\mathrm{\Delta }_n^+:=\mathrm{\Delta }_n\mathrm{\Delta }^+=\{\alpha \mathrm{\Delta }_n:\alpha (X_0)=1\}$$
and such that $`\mathrm{\Delta }_k^+:=\mathrm{\Delta }_k\mathrm{\Delta }^+`$ is a positive system in $`\mathrm{\Delta }_k`$. Let $`\mathrm{\Delta }^{}:=\mathrm{\Delta }^+`$, $`\mathrm{\Delta }_n^{}:=\mathrm{\Delta }_n^+`$ and $`\mathrm{\Delta }_k^{}:=\mathrm{\Delta }_k^+`$. We recall now few facts about the structure of the root system $`\mathrm{\Delta }`$. Two roots $`\alpha ,\beta \mathrm{\Delta }`$ are said to be strongly orthogonal if $`\alpha \pm \beta `$ is not a root. Let $`\mathrm{\Gamma }:=\{\gamma _1,\mathrm{},\gamma _r\}`$ be a system of strongly orthogonal roots in $`\mathrm{\Delta }_n^+`$ of maximal length, i.e., $`\mathrm{\Gamma }`$ consists of pairwise strongly orthogonal roots and has maximal number of elements with respect to this property. We set
$$𝒲:=N_{Inn(hk)}(a)/Z_{Inn(hk)}(a)$$
and call $`𝒲`$ the small Weyl group of $`\mathrm{\Delta }`$.
Proposition I.1.For the root system $`\mathrm{\Delta }=\mathrm{\Delta }(g,a)`$ of a non-compactly causal symmetric Lie algebra $`(g,\tau )`$ the following assertions hold: (i) The root system $`\mathrm{\Delta }`$ is reduced, i.e., if $`\alpha \mathrm{\Delta }`$ then $`2\alpha \mathrm{\Delta }`$. In particular, there exists at most two root lengths. (ii) All long roots in $`\mathrm{\Delta }_n^+`$ are conjugate under the small Weyl group $`𝒲`$. Moreover, all roots $`\gamma _i`$, $`1ir`$, are long. (iii) Write $`\mathrm{\Delta }_{n,s}^+`$ for the short roots in $`\mathrm{\Delta }_n^+`$. Then, if $`\mathrm{\Delta }_{n,s}^+Ø`$, one has
$$\mathrm{\Delta }_{n,s}^+=\{\frac{1}{2}(\gamma _i+\gamma _j):1i<jr\}$$
and all elements of $`\mathrm{\Delta }_{n,s}^+`$ are conjugate under $`𝒲`$.
Proof. (i) \[HiÓl96, Th. 3.2.4\] or \[NÓ99, Lemma 2.12\].
(ii) \[NÓ99, Lemma 2.26\].
(iii) \[NÓ99, Lemma 2.22, Lemma 2.24\].
For $`\alpha \mathrm{\Delta }`$ let $`H_\alpha \{[X,\tau (X)]:Xg^\alpha \}a`$ be such that $`\alpha (H_\alpha )=2`$. For each $`1ir`$ let $`H_i=H_{\gamma _i}`$. We set $`c:=span_R\{H_1,\mathrm{},H_r\}a`$ and write $`b`$ for the orthogonal complement of $`c`$ in $`a`$ with respect to the Cartan-Killing form; in particular $`a=cb`$.
Proposition I.2.The positive system $`\mathrm{\Delta }_k^+`$ can be chsosen such that for the restriction of $`\mathrm{\Delta }=\mathrm{\Delta }(g,a)`$ to $`c`$ the following assertions hold:
$$\mathrm{\Delta }_n^+_c=\{\frac{1}{2}(\gamma _i+\gamma _j):1i,jr\}\{\frac{1}{2}\gamma _i:1ir\},$$
$$\mathrm{\Delta }_k^+_c\backslash \{0\}=\{\frac{1}{2}(\gamma _i\gamma _j):1j<ir\}\{\frac{1}{2}\gamma _i:1ir\}.$$
Moreover, the second sets in the two unions from above may or may not occur simultaneously.
Proof. \[NÓ99, Th. 2.21\] or \[Kr99, Th. IV.4\].
Since we have free choice for $`\mathrm{\Delta }_k^+`$ we assume in the sequel that $`\mathrm{\Delta }_k^+_c\backslash \{0\}\{\frac{1}{2}(\gamma _i\gamma _j):1j<ir\}\{\frac{1}{2}\gamma _i:1ir\}`$.
Lemma I.3.Assume that $`\mathrm{\Delta }_{n,s}Ø`$ and let $`\mathrm{\Pi }_k`$ be the set of simple roots corresponding to $`\mathrm{\Delta }_k^+`$. Then there exits $`\beta _1,\mathrm{},\beta _mb^{}`$, $`\beta _j(X_0)=0`$, and $`\delta _1,\mathrm{}\delta _lb^{}`$, $`\delta _i(X_0)=\frac{1}{2}`$, such that
$$\mathrm{\Pi }_k=\{\frac{1}{2}(\gamma _{i+1}\gamma _i):1ir1\}\{\beta _1,\mathrm{},\beta _m\}\{\frac{1}{2}\gamma _r+\delta _i:1il\}.$$
Here the last set occurs if and only if there exits half roots in $`\mathrm{\Delta }_c`$.
Proof. For each $`\alpha \mathrm{\Delta }`$ let $`s_\alpha `$ denote the corresponding reflection. Then $`s_{\gamma _j}(\frac{1}{2}(\gamma _i+\gamma _j))=\frac{1}{2}(\gamma _i\gamma _j)`$, $`ij`$ together with Proposition I.1(iii) shows that $`\mathrm{\Delta }_k\{\frac{1}{2}(\gamma _i\gamma _j):1ijr\}`$. Thus Proposition I.2 yields that
$$\begin{array}{cc}\hfill \mathrm{\Delta }_k^+& \{\frac{1}{2}(\gamma _i\gamma _j):1j<ir\}+(b^{}X_0^{})\hfill \\ & \{\frac{1}{2}\gamma _i:1ir\}+\{\delta b^{}:\delta (X_0)=\frac{1}{2}\}b^{}.\hfill \end{array}$$
Now the assertion follows easily from Proposition I.2 and the fact that $`\mathrm{\Delta }`$ is a root system.
We define the maximal cone in $`a`$ is defined by
$$C_{\mathrm{max}}:=\{Xa:(\alpha \mathrm{\Delta }_n^+)\alpha (X)0\}.$$
Lemma I.4.Let $`X_0=X_0^b+X_0^c`$ with $`X_0^bb`$ and $`X_0^cc`$. Then we have $`X_0^b,X_0^cC_{\mathrm{max}}`$.
Proof. First note that $`X_0^c=\frac{1}{2}(H_1+\mathrm{}+H_r)`$ and so $`X_0^cC_{\mathrm{max}}`$ by Proposition I.2.
To show $`X_0^bC_{\mathrm{max}}`$ let $`\alpha \mathrm{\Delta }_n^+`$. Then Proposition I.2 shows that $`\alpha =\frac{1}{2}(\gamma _i+\gamma _j)+\beta `$ with $`\beta b^{}`$, $`\beta (X_0)=\beta (X_0^b)=0`$, or $`\alpha =\frac{1}{2}\gamma _i+\delta `$ with $`\delta b^{}`$ and $`\delta (X_0)=\delta (X_0^b)=\frac{1}{2}`$. In any case we have $`\alpha (X_0^b)0`$ concluding the proof of the lemma.
Finally we define subalgebras of $`g`$ by
$$n:=\underset{\alpha \mathrm{\Delta }^+}{}g^\alpha ,\overline{n}:=\underset{\alpha \mathrm{\Delta }^{}}{}g^\alpha ,n_k^\pm :=\underset{\alpha \mathrm{\Delta }_k^\pm }{}g^\alpha ,n_n^\pm :=\underset{\alpha \mathrm{\Delta }_n^\pm }{}g^\alpha $$
and note that $`n=n_n^+n_k^+`$ and $`\overline{n}=n_n^{}n_k^{}`$ are semidirect products.
Analytic preliminaries
Let $`G_C`$ be a simply connected Lie group with Lie algebra $`g_C`$ and let $`G`$ be the analytic subgroup of $`G_C`$ corresponding to $`g`$. Let $`H=G^\tau =\{XG:\tau (g)=g\}`$. We write $`A`$, $`K`$, $`N`$, $`\overline{N}`$, $`N_k^\pm `$, $`N_n^\pm `$ for the analytic subgroups of $`G`$ which correspond to $`a`$, $`g(0)`$, $`h`$, $`k`$, $`n`$, $`\overline{n}`$, $`n_k^\pm `$, $`n_n^\pm `$. Note that the groups $`A`$, $`N`$, $`\overline{N}`$, $`N_k^\pm `$, $`N_n^\pm `$ are all simply connected and that the corresponding exponential mappings $`\mathrm{exp}_A:aA`$, $`\mathrm{exp}_N:nN`$ etc. are all diffeomorphisms. Let $`G(0)=Z_G(X_0)=\{gG:Ad(g).X_0=X_0\}`$. Then $`H`$ and $`G(0)`$ are $`\tau `$ and $`\theta `$ invariant, $`H=(HK)\mathrm{exp}(hp)`$ and $`G(0)=(HK)\mathrm{exp}(qp)`$.
The Lie algebra $`g`$ decomposes as $`g=h+a+n`$ and the multiplication mapping
$$H\times A\times NG,(h,a,n)han$$
is an analytic diffeomorphism onto its open image $`HAN`$.
Note that $`\overline{N}=N_n^{}N_k^{}`$. We have
$$\overline{N}HAN=\mathrm{exp}(\mathrm{\Omega })N_k^{}=N_k^{}\mathrm{exp}(\mathrm{\Omega })$$
$`(1.1)`$
with $`\mathrm{\Omega }H/HK`$ a real bounded symmetric domain in $`n_n^{}`$.
II. The geometry of the real bounded symmetric domain $`\mathrm{\Omega }`$
We denote by $`\kappa `$ the Cartan-Killing form on $`g`$ and define an inner product on $`g`$ by $`X,Y:=\kappa (X,\theta (Y))`$ for $`X,Yg`$. Let $`X_ig^{\gamma _i}`$ be such that $`H_i=[X_i,X_i]`$, with $`X_i=\tau (X_i)`$. By \[HiÓl96\] and Herman’s Convexity Theorem we have
$$\begin{array}{ccc}\hfill \mathrm{\Omega }& =\{Xn_n^{}:ad(X+\tau (X))<1\}\hfill & (2.1)\hfill \\ & =Ad(HK).\{\underset{j=1}{\overset{r}{}}t_jX_j:1<t_j<1,j=1,\mathrm{},r\},\hfill & (2.2)\hfill \end{array}$$
where $``$ denotes the operator norm corresponding to the scalar product $`,`$ on $`g`$. Note that (2.1) implies that $`\mathrm{\Omega }`$ is a convex balanced subset of $`n_n^{}`$.
Remark II.1. Recall the definition of the maximal cone $`C_{rmmax}`$ in $`a`$. Then it is clear from the characterization (2.1) of $`\mathrm{\Omega }`$ that $`e^{adX}.\mathrm{\Omega }\mathrm{\Omega }`$ for all $`XC_{\mathrm{max}}`$. We also have a minimal cone in $`a`$ defined by
$$C_{\mathrm{min}}:=cone(\{[X,\tau (X)]:Xg^\alpha ,\alpha \mathrm{\Delta }^+\})=\overline{\underset{\alpha \mathrm{\Delta }_n^+}{}R^+H_\alpha }.$$
We note that $`C_{\mathrm{min}}C_{\mathrm{max}}`$ and in particular $`H_iC_{\mathrm{max}}`$ for each $`1ir`$.
The following concept turns out to be very useful for the investigation of the fine convex geomety of $`\mathrm{\Omega }`$.
Definition II.2. (Oshima-Sekiguchi) By a signature of $`\mathrm{\Delta }`$ we understand a map $`\epsilon :\mathrm{\Delta }\{1,1\}`$ with the following properties: (S1) $`\epsilon (\alpha )=\epsilon (\alpha )`$ for all $`\alpha \mathrm{\Delta }`$. (S2) $`\epsilon (\alpha +\beta )=\epsilon (\alpha )\epsilon (\beta )`$ for all $`\alpha ,\beta \mathrm{\Delta }`$ with $`\alpha +\beta \mathrm{\Delta }`$.
If $`\epsilon :\mathrm{\Delta }\{1,1\}`$ is a signature then $`\theta _\epsilon :gg`$ defined by $`\theta _\epsilon (X)=\epsilon (\alpha )\theta (X)`$, $`Xg^\alpha `$ and $`\theta _\epsilon |_{z_g(a)}=\theta |_{z_g(a)}`$ is an involution on $`g`$ that commutes with $`\theta `$ (see \[OS80, Def. 1.2\]). As $`\tau |_{z_g(a)}=\theta |_{z_g(a)}`$ and $`\tau |_{g^\alpha }=\pm \theta |_{g^\alpha }`$, with $`+`$ if $`\alpha `$ is compact and $`1`$ if $`\alpha `$ non-compact, it follows that $`\theta _\epsilon `$ also commutes with $`\tau `$.
Lemma II.3.Keep the notation of Definition II.2. (i) If $`\epsilon `$ is a signature of $`\mathrm{\Delta }`$, then the prescription
$$\sigma _\epsilon (X):=\{\begin{array}{cc}X\hfill & \text{for }Xz_g(a)\text{,}\hfill \\ \epsilon (\alpha )X\hfill & \text{for }Xg^\alpha \text{}\alpha \mathrm{\Delta }\hfill \end{array}$$
defines an involutive automorphism of $`g`$. The involution $`\sigma _\epsilon `$ commutes with both $`\tau `$ and $`\theta `$. (ii) Let $`\mathrm{\Pi }:=\{\alpha _1,\mathrm{},\alpha _n\}`$ be a basis of $`\mathrm{\Delta }`$. Then for any collection $`(\epsilon _1,\mathrm{},\epsilon _n)\{1,1\}^n`$ one can define a signature $`\epsilon `$ of $`\mathrm{\Delta }`$ by setting
$$\epsilon (\pm \underset{i=1}{\overset{n}{}}n_i\alpha _i):=\underset{i=1}{\overset{n}{}}\epsilon _i^{n_i}\text{for}\underset{i=1}{}n_i\alpha _i\mathrm{\Delta }.$$
(iii) Let the notation be as in (ii). Then $`\epsilon (\epsilon (\alpha _i))_{i=1}^n`$ defines a bijection between the set of signatures of $`\mathrm{\Delta }`$ and $`\{1,1\}^n`$.
Proof. (i) This follows by the Oshima-Sekiguchi construction because $`\sigma _\epsilon =\tau _\epsilon \theta `$. (ii) is clear and (iii) follows from (ii).
In the sequel we identify signatures with elements in $`\{1,1\}^n`$.
Lemma II.4.Let $`\epsilon `$ be a signature of $`\mathrm{\Delta }`$. Then $`\sigma _\epsilon (\mathrm{\Omega })=\mathrm{\Omega }`$.
Proof. Let $`X\mathrm{\Omega }`$. By (2.2) there is a $`kHK`$ and $`Y=_{j=1}^rt_jX_j`$, $`1<t_j<1`$ such that $`Ad(k).Y=X`$. As $`\sigma _\epsilon `$ commutes with $`\tau `$ and $`\theta `$ it follows that $`\sigma _\epsilon (k)KH`$. Hence $`\sigma _\epsilon (X)=Ad(\sigma _\epsilon (k))._{j=1}^r\epsilon (\gamma _j)t_jX_j\mathrm{\Omega }`$.
Recall that there is basis $`\mathrm{\Pi }\mathrm{\Delta }^+`$ having the form
$$\mathrm{\Pi }=\{\alpha _0,\alpha _1,\mathrm{},\alpha _n\}$$
with $`\alpha _0`$ long and non-compact and $`\alpha _i`$, $`1in`$ compact. Thus every non-compact negative root $`\gamma \mathrm{\Delta }_n^{}`$ can be written as $`\gamma =\alpha _0_{i=1}^nm_i\alpha _i`$, $`m_iN_0`$. By our choice of $`\mathrm{\Delta }_k^+`$ we have $`\alpha _0=\gamma _1`$.
Theorem II.5.For each $`\gamma \mathrm{\Delta }_n^{}`$ let $`p_\gamma :n_n^{}g^\gamma `$ be the orthogonal projection. Then
$$X\mathrm{\Omega }p_\gamma (X)\mathrm{\Omega }.$$
Proof. Let $`X=_{\gamma \mathrm{\Delta }_n^{}}X_\gamma \mathrm{\Omega }`$ with $`X_\gamma g^\gamma `$, $`\gamma \mathrm{\Delta }_n^{}`$. We have to show that $`X_\gamma \mathrm{\Omega }`$. Recall that there are at most two root length in $`\mathrm{\Delta }`$ (cf. Proposition I.1(i)).
Case 1: $`\gamma `$ is a long root.
By Proposition I.1(ii) there exists an element $`hN_{Inn(hk)}(a)`$ such that $`h.\gamma =\alpha _0`$. Thus we may assume that $`\gamma =\alpha _0=\gamma _1`$. Let $`H:=_{j=2}^rH_j`$. By Remark II.1 we have
$$X_1:=\underset{t+\mathrm{}}{lim}e^{tadH}.X\mathrm{\Omega }.$$
If we express $`X_1=_{\beta \mathrm{\Delta }_n^{}}X_\beta `$ as a sum of root vectors, then Proposition I.2 implies that $`\beta _c=\gamma _1`$ or $`\beta =\frac{1}{2}\gamma _1\delta `$ with $`\delta (X_0^b)=\frac{1}{2}`$. Since $`X_0^bC_{\mathrm{max}}`$ (cf. Lemma I.4), we now get
$$X_\gamma =\underset{t+\mathrm{}}{lim}e^{tadX_0^b}.X_1\mathrm{\Omega }.$$
Case 2: $`\gamma `$ is a short root.
By Proposition I.1(iii) we may assume that $`\gamma =\frac{1}{2}(\gamma _1+\gamma _2)`$ and by Lemma I.3 we may suppose $`\alpha _0=\gamma _1`$, $`\alpha _j=\frac{1}{2}(\gamma _{j+1}\gamma _j)`$ for $`1jr1`$. Write
$$X=\underset{m_i0}{}X_{m_1,\mathrm{},m_n},$$
where $`X_{m_1,\mathrm{},m_n}g^{(\alpha _0+_{i=1}^nm_i\alpha _i)}`$. Then we have to show that $`X_{1,0,\mathrm{},0}\mathrm{\Omega }`$. Set
$$X_{\mathrm{ev}}:=\underset{m_n0(2)}{}X_{m_1,\mathrm{},m_n}\text{and}X_{\mathrm{odd}}:=\underset{m_n1(2)}{}X_{m_1,\mathrm{},m_n}.$$
Then $`X=X_{\mathrm{ev}}+X_{\mathrm{odd}}`$ and we claim that $`X_{\mathrm{ev}}`$, $`X_{\mathrm{odd}}\mathrm{\Omega }`$. Let $`\epsilon =(1,1,1,\mathrm{},1)`$. Then by Lemma II.4 we get:
$$\sigma _\epsilon (X)=\sigma _\epsilon (X_{\mathrm{ev}}+X_{\mathrm{odd}})=X_{\mathrm{ev}}X_{\mathrm{odd}}\mathrm{\Omega }.$$
Since $`\mathrm{\Omega }`$ is balanced and convex we moreover have
$$X_{ev}=\frac{1}{2}(X+\sigma _\epsilon (X))\mathrm{\Omega }\text{and}X_{\mathrm{odd}}=\frac{1}{2}(X\sigma _\epsilon (X))\mathrm{\Omega }.$$
By repeating this argument we thus my assume that
$$X=\underset{\genfrac{}{}{0pt}{}{m_11(2)}{m_j0(2),j>1}}{}X_{m_1,\mathrm{},m_n}.$$
Now we apply the contraction semigroup generated by $`H=_{j=3}^rH_jC_{\mathrm{max}}`$ and obtain
$$X_1:=\underset{t+\mathrm{}}{lim}e^{tadH}.X\mathrm{\Omega }.$$
Thus we may assume $`X=X_1`$ and $`X=_{\beta \mathrm{\Delta }_n^{}}X_\beta `$ with $`\beta =\gamma ,\gamma _1,\gamma _2,\frac{1}{2}(\gamma _1+\gamma _2)+\beta ,\frac{1}{2}\gamma _1+\sigma _1,\frac{1}{2}\gamma _2+\sigma _2`$ and $`\beta ,\sigma _1,\sigma _2b^{}`$, $`\sigma _1(X_0)=\sigma _2(X_0)=\frac{1}{2}`$ (cf. Proposition I.2). Write $`\beta =\gamma _1_{j=1}m_j\alpha _j`$. The cases $`\beta =\gamma _1`$ and $`\beta =\gamma _2`$ are excluded, since we have $`m_1=0`$, resp. $`m_1=2`$, contradicting $`m_11(2)`$. Applying to $`X`$ the contraction semigroup generated by $`X_0^bC_{\mathrm{max}}`$ excludes the case $`\beta =\frac{1}{2}\gamma _1+\sigma _1`$ and $`\beta =\frac{1}{2}\gamma _2+\sigma _2`$. Let now $`Yb`$ such that $`\delta _j(Y)>0`$, $`1jl`$, and $`\beta _j(Y)>0`$, $`1jm`$ (cf. Lemma I.3). Then $`\mathrm{\Delta }_n^+N_0[\mathrm{\Pi }]`$ shows that $`YC_{\mathrm{max}}`$. But then
$$X_\gamma =\underset{t+\mathrm{}}{lim}e^{tadY}.X\mathrm{\Omega },$$
completing the proof Case 2 and hence of the theorem.
Subdomains of rank one
For $`\alpha \mathrm{\Delta }^+`$ we set
$$g(\alpha ):=\left(g^\alpha +g^\alpha +[g^\alpha ,g^\alpha ]\right)^{}$$
and $`\tau (\alpha ):=\tau _{g(\alpha )}`$. Then $`(g(\alpha ),\tau (\alpha ))`$ is a symmetric subalgebra of $`(g,\tau )`$ of real rank one, that is $`a(\alpha ):=ag(\alpha )`$ is one dimensional. Further we set $`h(\alpha ):=hg(\alpha )`$ etc. We denote by $`G(\alpha )`$, $`A(\alpha )`$ etc. the analytic subgroups of $`G`$ corresponding to $`g(\alpha )`$, $`a(\alpha )`$ etc. Let $`H(\alpha )=G(\tau )^{\tau (\alpha )}=G(\alpha )H`$.
Assume that $`\alpha \mathrm{\Delta }_n^+`$. Then $`(g(\alpha ),\tau (\alpha ))`$ is NCC and $`n(\alpha )=n_n^+(\alpha )=g^\alpha `$. Let $`\mathrm{\Omega }(\alpha )H(\alpha )/\left(K(\alpha )H(\alpha )\right)`$ be the real bounded symmetric domain in $`\overline{n}(\alpha )=n_n^{}(\alpha )`$.
Lemma II.6.Let $`\alpha \mathrm{\Delta }_n^+`$ and $`s_\alpha G(\alpha )`$ be a representaive of the one element big Weyl group $`N_{G(\alpha )}(a(\alpha ))/Z_{G(\alpha )}(a(\alpha ))`$ of $`g(\alpha )`$. Then
$$\left(\overline{N}(\alpha )H(\alpha )A(\alpha )N(\alpha )\right)\dot{}\left(\overline{N}(\alpha )H(\alpha )s_\alpha A(\alpha )N(\alpha )\right)$$
is open and dense in $`\overline{N}(\alpha )`$.
Proof. This follows by Matsukis Theorem (cf. \[Ma79, Theorem 3\]), if we can show that $`M(\alpha ):=Z_{K(\alpha )}(a(\alpha ))H(\alpha )`$ because $`s_\alpha M(\alpha )=M(\alpha )s_\alpha `$. Let $`F=\mathrm{exp}(ia(\alpha ))G(\alpha )`$. Then one has $`M(\alpha )=FZ_{H(\alpha )_o}(a(\alpha ))`$ by \[NÓ99, Lemma 5.7\]. But if $`fF`$ then $`\tau (\alpha )(f)=f^1=f`$, by the same lemma. Hence $`FH(\alpha )`$, which implies that $`M(\alpha )H(\alpha )`$.
Theorem II.7.Let $`\alpha \mathrm{\Delta }_n^+`$. Then $`\mathrm{\Omega }\overline{n}(\alpha )=\mathrm{\Omega }(\alpha )`$.
Proof. ”$``$”: This is clear.
$``$”: Note that $`\mathrm{\Omega }\overline{n}(\alpha )`$ is open and convex in $`\overline{n}(\alpha )`$. We have
$$\mathrm{exp}(\mathrm{\Omega })\left(H(\alpha )s_\alpha A(\alpha )N(\alpha )\right)=Ø,$$
$`(2.3)`$
since $`\mathrm{exp}(\mathrm{\Omega })HAN`$ and $`HANHs_\alpha AN=Ø`$ by Matsukis Theorem. In view of (2.3), Lemma II.7 implies that there exists an open dense subset $`\mathrm{\Omega }_\alpha `$ of $`\mathrm{\Omega }\overline{n}(\alpha )`$ such that $`\mathrm{\Omega }_\alpha \mathrm{\Omega }(\alpha )`$. Now the assertion follows from the fact that both $`\mathrm{\Omega }(\alpha )`$ and $`\mathrm{\Omega }\overline{n}(\alpha )`$ are open and convex.
III. The product formula for the $`c`$-function
Recall the $`HAN`$-decomposition in $`G`$ from Section I. For each $`\lambda a_C^{}`$ and $`g`$ in $`G`$ we set
$$a_H(g)^\lambda :=\{\begin{array}{cc}0\hfill & \text{if }gHAN\text{,}\hfill \\ e^{\lambda (\mathrm{log}a)}\hfill & \text{if }g=hanHAN\hfill \end{array}.$$
For a locally compact group $`G`$ we write $`\mu _G`$ for a left Haar measurwe on $`G`$.
Definition III.1. (The $`c`$-functions) For each $`\alpha a^{}`$ let $`m_\alpha :=dimg^\alpha `$ and put $`\rho :=\frac{1}{2}_{\alpha \mathrm{\Delta }^+}m_\alpha \alpha `$. For $`\lambda a_C^{}`$ we now set
$$c(\lambda ):=_{\overline{N}}a_H(\overline{n})^{(\lambda +\rho )}𝑑\mu _{\overline{N}}(\overline{n})=_{\overline{N}HAN}a_H(\overline{n})^{(\lambda +\rho )}𝑑\mu _{\overline{N}}(\overline{n}),$$
$$c_\mathrm{\Omega }(\lambda ):=_{N_n^{}}a_H(\overline{n})^{(\lambda +\rho )}𝑑\mu _{N_n^{}}(\overline{n})=_\mathrm{\Omega }a_H(\overline{n})^{(\lambda +\rho )}𝑑\mu _{N_n^{}}(\overline{n}),$$
and
$$c_0(\lambda ):=_{N_k^{}}a_H(\overline{n})^{(\lambda +\rho )}𝑑\mu _{N_k^{}}(\overline{n})$$
whenever the defining integral exist. We write $``$, $`_\mathrm{\Omega }`$ and $`_0`$ for the domain of definition of $`c`$, $`c_\mathrm{\Omega }`$ and $`c_0`$, respectively. We call $`c`$ the $`c`$-function of the non-compactly causal symmetric space $`G/H`$ and $`c_\mathrm{\Omega }`$ the $`c`$-function of the real bounded symmetric domain $`\mathrm{\Omega }`$, while $`c_0`$ is the usual $`c`$-function of the non-compact Riemannian symmetric space $`G(0)/K(0)`$.
Remark III.2. (a) The choice of the particular analytic realization $`G/H`$ of $`(g,\tau )`$ as a symmetric space is immaterial for the definition of the $`c`$-function.
(b) We have $`=_0_\mathrm{\Omega }`$ and for all $`\lambda `$ one has the splitting
$$c(\lambda )=c_0(\lambda )c_\mathrm{\Omega }(\lambda )$$
(cf. \[FHÓ94, Lemma 9.2\]).
(c) The $`c`$-functions can be written as Laplace transforms (cf. \[KNÓ98\]). Let us explain this for the $`c`$-function $`c`$. For $`c_0`$ and $`c_\mathrm{\Omega }`$ one has analogous statements.
There exists a positive Radon measure $`\mu `$ on $`a`$ such that
$$(\lambda )c(\lambda )=_\mu (\lambda ):=_ae^{\lambda (X)}𝑑\mu (X),$$
i.e., $`c`$ is the Laplace transform of $`\mu `$. In particular we see that the domain of definition $``$ is a tube domain over a convex set, i.e., one has
$$=ia^{}+_R$$
with $`_Ra^{}`$ a convex subset of $`a^{}`$. One knows that $`int`$ is non-empty. Moreover, the fact that $`c`$ is a Laplace transform implies that $`c`$ is holomorphic on $`int`$ and that $`c`$ has no holomorphic extension to a connected open tube domain strictly larger than $`int`$.
Now we are going to prove the product formula for the $`c`$-function $`c_\mathrm{\Omega }`$. Our srategy is a modified Gindikin-Karpelevic approach as presented in \[GaVa88, p. 175–177\] or \[Hel84, Ch. IV\].
For a positive system $`R\mathrm{\Delta }`$ we set $`\overline{n}_R:=_{\alpha (\mathrm{\Delta }^+R)}g^\alpha `$ and write $`\overline{N}_R`$ for the corresponding analytic subgroup of $`G`$. We define an auxiliary $`c`$-function by
$$c_R(\lambda ):=_{\overline{N}_R}a_H(\overline{n})^{(\lambda +\rho )}𝑑\mu _{\overline{N}_R}(\overline{n})$$
whenever the integral exists.
For a single root $`\alpha \mathrm{\Delta }^+`$ we set $`\rho _\alpha :=\frac{1}{2}m_\alpha \alpha `$ and write
$$c_\alpha (\lambda ):=_{\overline{N}(\alpha )}a_{H(\alpha )}(\overline{n})^{(\lambda +\rho _\alpha )}𝑑\mu _{\overline{N}(\alpha )}(\overline{n}).$$
We denote by $`_\alpha a_C^{}`$ the domain of definition of $`c_\alpha `$.
Proposition III.3.For any positive system $`R\mathrm{\Delta }`$ we have that
$$c_R(\lambda )=\underset{\alpha (R\mathrm{\Delta }^+)}{}c_\alpha (\lambda )$$
and $`c_R(\lambda )`$ is defined if and only if $`\lambda _{\alpha (R\mathrm{\Delta }^+)}_\alpha `$.
Proof. We proceed by induction on $`|R\mathrm{\Delta }^+|`$. If $`R\mathrm{\Delta }^+=Ø`$, then the assertion is clear.
Assume that $`R\mathrm{\Delta }^+Ø`$. Then we find an element $`\beta R\mathrm{\Delta }^+`$ which is simple in $`R`$. Set $`Q:=s_\beta .R`$. Then $`Q=R\backslash \{\beta \}\{\beta \}`$ since $`\mathrm{\Delta }`$ is reduced (cf. Proposition I.1(i)). Thus we have $`(Q\mathrm{\Delta }^+)\dot{}\{\beta \}=R\mathrm{\Delta }^+`$. We now have to distinguish to cases.
Case 1: $`\beta `$ is compact.
In this case, the $`HAN`$-decomposition of $`G(\beta )`$ coincides with the Iwasasa decomposition, i.e. $`G(\beta )=K(\beta )A(\beta )N(\beta )`$. Thus $`c_R(\lambda )=c_\beta (\lambda )c_Q(\lambda )`$ follow as in \[GaVa88, Prop. 4.7.6\].
Case 2: $`\beta `$ is non-compact.
Set $`\overline{N}_Q^k:=\overline{N}_QN_k^{}`$, $`\overline{N}_Q^n:=\overline{N}_QN_n^{}`$ and note that $`\overline{N}_Q=\overline{N}_Q^n\overline{N}_Q^k`$. Since $`\overline{N}_R=\overline{N}(\beta )\overline{N}_Q`$ we thus get
$$c_R(\lambda )=_{\overline{N}(\beta )}_{\overline{N}_Q^n}_{\overline{N}_Q^k}a_H(\overline{n}_\beta \overline{n}_n\overline{n}_k)^{(\lambda +\rho )}𝑑\mu _{\overline{N}(\beta )}(\overline{n}_\beta )𝑑\mu _{\overline{N}_Q^n}(\overline{n}_n)𝑑\mu _{\overline{N}_Q^k}(\overline{n}_k).$$
If $`\overline{n}_\beta \overline{n}_n\overline{n}_k\overline{N}HAN`$, then (1.1) implies that $`\overline{n}_\beta \overline{n}_n\mathrm{exp}(\mathrm{\Omega })`$. Since $`n_n^{}`$ is abelian, Theorem II.5 therefore implies that $`\overline{n}_\beta \mathrm{exp}(\mathrm{\Omega })`$ and so $`\overline{n}_\beta \mathrm{exp}(\mathrm{\Omega }(\beta ))`$ by Theorem II.7. Therefore we can write $`\overline{n}_\beta =h_\beta a_\beta n_\beta `$ with $`h_\beta H(\beta )`$, $`a_\beta A(\beta )`$ and $`n_\beta N(\beta )`$. Now one can proceed as in \[GaVa88, p. 175–177\] and one gets $`c_R(\lambda )=c_\beta (\lambda )c_Q(\lambda )`$.
Remark III.4. If we choose $`R=\mathrm{\Delta }_n^+\mathrm{\Delta }_k^+`$ (this is a positive system since $`\mathrm{\Delta }_n^+`$ is $`𝒲`$-invariant), then we have $`c_0=c_R`$ and Proposition III.3 results in the Gindikin-Karpelevic product formula
$$c_0(\lambda )=\underset{\alpha \mathrm{\Delta }_k^+}{}c_\alpha (\lambda )$$
of the $`c`$-function $`c_0`$ on $`G(0)/K(0)`$ (cf. \[GaVa88, Th. 4.7.5\] or \[Hel84, Ch. IV, Th. 6.13, 6.14\]).
Theorem III.5. (The product formula for $`c_\mathrm{\Omega }`$) For the $`c`$-function $`c_\mathrm{\Omega }`$ of the real bounded symmetric domain $`\mathrm{\Omega }`$ one has
$$_\mathrm{\Omega }=\{\lambda a_C^{}:(\alpha \mathrm{\Delta }_n^+)Re\lambda (H_\alpha )<2m_\alpha \}$$
and
$$c_\mathrm{\Omega }(\lambda )=\kappa \underset{\alpha \mathrm{\Delta }_n^+}{}B(\frac{m_\alpha }{2},\frac{\lambda (H_\alpha )}{2}\frac{m_\alpha }{2}+1)$$
where $`B`$ denotes the Beta function and $`\kappa `$ is a positive constant only depending on $`(g,\tau )`$.
Proof. Set $`_\mathrm{\Omega }^{}:=_{\alpha \mathrm{\Delta }_n^+}_\alpha `$. We want to apply Proposition III.3 to $`R=\mathrm{\Delta }^+`$. In view of Remark III.2(b) and Remark III.4, we thus get
$$(\lambda _\mathrm{\Omega }^{})c_\mathrm{\Omega }(\lambda )=\underset{\alpha \mathrm{\Delta }_n^+}{}c_\alpha (\lambda )=\underset{\alpha \mathrm{\Delta }_n^+}{}c_{\mathrm{\Omega }(\alpha )}(\lambda ).$$
$`(3.1)`$
By \[FHÓ94, (10.3)\] one has
$$c_{\mathrm{\Omega }(\alpha )}(\lambda )=2^{m_\alpha 1}B(\frac{m_\alpha }{2},\frac{\lambda (H_\alpha )}{2}\frac{m_\alpha }{2}+1)$$
$`(3.2)`$
and
$$_{\mathrm{\Omega }(\alpha )}=\{\lambda a_C^{}:Re\lambda (H_\alpha )<2m_\alpha \}.$$
$`(3.3)`$
It follows from (3.3) that
$$_\mathrm{\Omega }^{}=\{\lambda a_C^{}:(\alpha \mathrm{\Delta }_n^+)Re\lambda (H_\alpha )<2m_\alpha \}.$$
$`(3.4)`$
Besides $`_\mathrm{\Omega }=_\mathrm{\Omega }^{}`$ all assertions of the theorem now follow from (3.1)-(3.4). Finally, $`_\mathrm{\Omega }=_\mathrm{\Omega }^{}`$ follows from the fact that all $`c`$-functions involved are Laplace transforms (cf. Remark III.2(c)).
The following simple fact that shows that we can split off all the non-compact roots to get the $`c_\mathrm{\Omega }`$-function before we come to the compact roots.
Lemma III.6.Let $`R`$ be a any positive system of roots in $`\mathrm{\Delta }`$. If $`R\mathrm{\Delta }_n^+Ø`$, then $`R\mathrm{\Delta }_n^+`$ contains a a root that is simple in $`R`$.
Proof. Let $`\{\beta _0,\mathrm{},\beta _n\}`$ be the set of simple roots in $`R`$. Let $`\gamma R\mathrm{\Delta }_n^+`$. Then $`\gamma =_{i=0}^nn_i\beta _i`$ with $`n_iN_0`$. Thus $`1=\gamma (X_0)=_{i=0}^nn_i\beta _i(X_0)`$ which implies that $`\beta _i(X_0)>0`$ for at least one $`\beta _i`$. But then $`\beta _i\mathrm{\Delta }_n^+`$.
References
\[Fa95\] Faraut, J., Fonctions Sphériques sur un Espace Symétrique Ordonné de Type Cayley, Contemp. Math. 191 (1995), 41–55 .
\[FHÓ94\] Faraut, J., J. Hilgert, and G. Ólafsson, Spherical functions on ordered symmetric spaces, Ann. Inst. Fourier 44 (1994), 927–966 .
\[GaVa88\] Gangolli, R., and V.S. Varadarajan, “Harmonic Analysis of Spherical Functions on Real Reductive Groups,” Ergebniss der Mathematik 101, Springer, 1988 .
\[Gr97\] Graczyk, P., Function $`c`$ on an ordered symmetric space, Bull. Sci. math. 121 (1997), 561–572 .
\[Hel84\] Helgason, S., ”Groups and Geometric Analysis”, Acad. Press, London, 1984 .
\[HiÓl96\] Hilgert, J. and G. Ólafsson, “Causal Symmetric Spaces, Geometry and Harmonic Analysis,” Acad. Press, 1996 .
\[Kr99\] Krötz, B., Formal dimension of semisimple symmetric spaces, Compositio math., to appear .
\[KNÓ98\] Krötz, B., K.–H. Neeb, and G. Ólafsson, Spherical Functions on Mixed Symmetric Spaces, submitted .
\[Ma79\] Matsuki, T., The orbits of affine symmetric spaces under the action of minimal parabolic subgroups, J. Math. Soc. Jpn. 31, 331–357 (1979) .
\[NÓ99\] Neumann, A., and G. Ólafsson, Minimal and Maximal Semigroups Related to Causal Symmetric Spaces, Semigroup Forum, to appear .
\[Ól97\] Ólafsson, G., Spherical Functions and Spherical Laplace Transform on Ordered Symmetric Spaces, submitted .
\[OS80\] Oshima, S., Sekiguchi, J, Eigenspaces of Invariant Differential Operators on an Affine Symmetric Spaces, Invent. math. 57 (1980), 1–81 .
Bernhard Krötz Mathematical Institute TU Clausthal Erzstraße 1 D-38678 Clausthal–Zellerfeld Germany e-mail: mabk@math.tu-clausthal.de Gestur Ólafsson Department of Mathematics Louisiana State University Baton Rouge LA 70803 e-mail: olafsson@math.lsu.edu
|
warning/0004/hep-th0004164.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The complex Monge-Ampère equation plays a central role in physics and mathematics. On a complex manifold $``$ of dimension $`n`$ it is given by
$$\frac{1}{n!}(\overline{}u)^n=\kappa ^{}1,\kappa =\pm 1,0$$
(1)
where $``$ is the holomorphic exterior derivative, $`\overline{}`$ is its anti-holomorphic counter-part, the wedge product is understood and $`{}_{}{}^{}1`$ denotes the volume element. It will be referred to as $`CMA_n`$. Perhaps the more familiar form of this equation employs the Monge-Ampère determinant
$$\mu detu_{i\overline{k}}=\frac{1}{n!}^{}(\overline{}u)^n=\kappa $$
(2)
where
$$u_{i\overline{k}}\frac{^2u}{\zeta ^i\overline{\zeta }^k},$$
(3)
and is the Hodge star duality operator. Depending on $`\kappa =\pm 1`$ $`CMA_n`$ is elliptic, or hyperbolic respectively and it is called homogeneous for $`\kappa =0`$. Chern, Levine and Nirenberg have pointed out that the homogeneous $`CMA_n`$ is the fundamental equation in the theory of functions of many complex variables. The Laplace equation itself is $`CMA_1`$. In differential geometry elliptic $`CMA_n`$ is the equation governing the Kähler potential for metrics with Euclidean signature and vanishing first Chern class . This was noted by Calabi and Yau has given an existence proof. The Riemann curvature $`(1,1)`$-form is then self-dual which by the first Bianchi identity implies Ricci-flatness. These metrics are hyper-Kähler . Furthermore, since Ricci-flat metrics satisfy the vacuum Einstein field equation with Euclidean signature, $`CMA_2`$ is of vital interest for general relativistic instantons. The construction of the Riemannian metric for the gravitational instanton $`K3`$, also familiar as Kummer’s surface , still remains as the outstanding unsolved problem of elliptic $`CMA_2`$.
Recently we have found that real Monge-Ampère equations admit rich multi-symplectic structure. In this letter we shall present the symplectic structure of $`CMA_n`$ which, in spite of its radical differences with the real case, also admits interesting multi-symplectic structure. $`CMA_n`$ is invariant under arbitrary holomorphic changes of the independent variables. In the elliptic and hyperbolic cases we must add the proviso that the Jacobian must be unity . Clearly, the symplectic structure of $`CMA_n`$ must be covariant under this infinite group. The usual approach to symplectic structure starts with a choice of time parameter, thus breaking covariance from the outset. It is therefore not suitable for a discussion of $`CMA_n`$. Happily, Witten and Zuckerman have shown that there exists a covariant, closed and conserved $`2`$-form vector density, the time component of which gives the familiar symplectic $`2`$-form. We shall use the Witten-Zuckerman approach to symplectic structure, reformulating it in order to express everything in terms of holomorphic and anti-holomorphic differential forms.
Finally, we note that the usual discussion of the symplectic structure of $`CMA_n`$, cf , is based on the Kähler $`(1,1)`$-form
$$\omega _K=\frac{1}{2i}\overline{}u$$
(4)
but this is not the relevant object that emerges from our examination of symplectic structure. Our approach to symplectic structure will be in the framework of dynamical systems with infinitely many degrees of freedom . We shall start with action principles underlying $`CMA_n`$ and from the Lagrangian derive the corresponding Witten-Zuckerman symplectic $`2`$-form. The results we shall present for $`CMA_n`$ are quite different from the usual considerations using the Kähler $`(1,1)`$-form.
## 2 First symplectic structure of $`CMA_n`$
Everything that is of interest is derivable from a variational principle $`\delta I=0`$,
$$I=L$$
(5)
where $`L`$ is the Lagrangian volume form. For $`CMA_n`$ the Lagrangian is an $`(n,n)`$-form
$$L=\frac{1}{(n+1)!}u\overline{}u(\overline{}u)^{n1}+\kappa u^{}1$$
(6)
and it can be verified directly that first variation of the action (5) yields $`CMA_n`$. This Lagrangian has not been considered before and the easiest way of remembering it is through its determinantal structure
$${}_{}{}^{}L=det\left|\begin{array}{cc}0& u_i\\ u_{\overline{k}}& u_{i\overline{k}}\end{array}\right|+\kappa u$$
(7)
which should be compared to the Monge-Ampère determinant (2). The Lagrangian (6) is also a linear combination of the zeroth and second order differential invariants in the prolongation structure and group foliation of $`CMA_n`$ .
We shall use a reformulation of the Witten-Zuckerman theory in terms of holomorphic and anti-holomorphic differential forms to obtain the symplectic structure of $`CMA_n`$. The first variation of the Lagrangian (6) yields
$$\delta L=\alpha +(1)^n\overline{}\overline{\alpha }+(\kappa \mu )\delta u^{}1$$
(8)
where $`\alpha `$ is given by
$$\alpha =\frac{1}{2n!}\delta u\overline{}u(\overline{}u)^{n1}+\frac{n1}{2(n+1)!}\overline{}\delta uu\overline{}u(\overline{}u)^{n2}$$
(9)
which is an $`(n1,n)`$-form on $`\mathrm{\Lambda }^{n1,n}()`$ as well as a $`1`$-form on $`\mathrm{\Lambda }^1(())`$, the space of functions over $``$. We note that in Witten-Zuckerman theory of symplectic structure we require that the Jacobi equation
$$(\overline{}u)^{n1}\overline{}\delta u=0$$
(10)
must be satisfied in addition to $`CMA_n`$ itself. Then the Witten-Zuckerman symplectic $`2`$-form is obtained by applying the exterior-functional derivative to $`\alpha `$. We find that the symplectic $`2`$-form $`\omega =\delta \alpha `$ is given by
$`\omega `$ $`=`$ $`{\displaystyle \frac{1}{2n!}}\delta u(\overline{}u)^{n1}\overline{}\delta u`$
$`+{\displaystyle \frac{n1}{2n!}}\delta u\overline{}u(\overline{}u)^{n2}\overline{}\delta u`$
$`{\displaystyle \frac{n1}{2(n+1)!}}\overline{}\delta u\overline{}u(\overline{}u)^{n2}\delta u`$
$`+{\displaystyle \frac{n1}{2(n+1)!}}\overline{}\delta uu(\overline{}u)^{n2}\overline{}\delta u`$
$`+{\displaystyle \frac{(n1)(n2)}{2(n+1)!}}\overline{}\delta uu\overline{}u(\overline{}u)^{n3}\overline{}\delta u.`$
It can be directly verified that $`\omega `$ satisfies the closed
$$\delta \omega =0$$
(12)
and conserved
$$\omega +(1)^n\overline{}\overline{\omega }=0$$
(13)
properties of the Witten-Zuckerman symplectic structure on $`\mathrm{\Lambda }^2(())`$ $``$ $`[`$ $`\mathrm{\Lambda }^{n1,n}()`$ $``$ $`\mathrm{\Lambda }^{n,n1}()`$ $`]`$. The symmetry group of $`CMA_n`$ is the group of holomorphic changes of the independent variables with unit Jacobian . As in the case of its nearest relative, namely the group of diffeomorphisms, this is an infinite group. The symplectic $`2`$-form (2) should be expressible as a Lie-Poisson structure associated with this group. That is, it must come from the co-adjoint action of vector fields belonging to the Lie algebra of the group of holomorphic changes of the independent variables with unit Jacobian.
## 3 Hilbert’s variational principle
Kähler metrics with unit determinant, a requirement identical to $`CMA_n`$, are Ricci-flat. Since this condition is the same as the Euclideanized Einstein field equations for vacuum we are led to a second variational principle for $`CMA_n`$ which is simply Hilbert’s Lagrangian . First we recall that the Kähler metric is given by
$$g_{i\overline{k}}=u_{i\overline{k}}$$
(14)
through the definition (3) and we note that
$$\mu =\sqrt{g}$$
(15)
must be nonzero. Therefore we must exclude the homogeneous complex Monge-Ampère equation from this part of the discussion. Then it is a standard result in differential geometry that the Ricci tensor for Kähler metrics is given by
$$R_{i\overline{k}}=(\mathrm{ln}\mu )_{i\overline{k}}$$
(16)
which makes manifest the important role $`CMA_n`$ plays in Kähler geometry.
The Hilbert Lagrangian density is
$$_H=\sqrt{g}R$$
(17)
where $`R=g^{i\overline{k}}R_{i\overline{k}}`$ is the scalar of curvature formed out of the Riemann tensor. From eq.(16) and the definition of the contravariant metric, it can be verified that for Kähler metrics the Lagrangian (17) can be written as the $`(n,n)`$-form
$$L_H=\frac{1}{n!}(\overline{}u)^{n1}\overline{}\mathrm{ln}\mu $$
(18)
but this is a divergence. Another way of seeing this divergence property, which has not been generally remarked on, is through eq.(16) and the following remarkable identity:
Lemma
For Kähler metrics
$$(\sqrt{g}g^{i\overline{k}})_{\overline{k}}=0$$
(19)
is an identity. Proof is by direct calculation which is immediate through the observation that $`\sqrt{g}g^{i\overline{k}}`$ is given by the cofactors of $`u_{i\overline{k}}`$.
Hilbert’s Lagrangian $`(n,n)`$-form (18) can be written as
$$L_H=\overline{}Z$$
(20)
in several different ways
$$Z=\frac{1}{n!}\{\begin{array}{c}u(\overline{}u)^{n2}\overline{}\mathrm{ln}\mu ,\hfill \\ \mathrm{ln}\mu (\overline{}u)^{n1},\hfill \\ \overline{}u(\overline{}u)^{n1}\mathrm{ln}\mu ,\hfill \end{array}$$
(21)
which are alternative statements of the fact that for Kähler metrics the Hilbert Lagrangian is a topological invariant. Its first variation vanishes identically and we are left with only the boundary terms. Even though $`CMA_n`$ does not emerge as the Euler-Lagrange equations from (18) an examination of the divergence terms is interesting because it is precisely these boundary terms in the first variation of the action that are important in the Witten-Zuckerman construction of the symplectic $`2`$-form. For this purpose we write $`L_H`$ in the form
$`L_H`$ $`=`$ $`X+(1)^n\overline{}\overline{X}`$ (22)
$`X`$ $`=`$ $`{\displaystyle \frac{1}{2n!}}\overline{}u(\overline{}u)^{n2}\overline{}\mathrm{ln}\mu ,`$ (23)
skipping other alternatives manifest in eq.(21) because they will yield degenerate results for symplectic structure. From the boundary terms using $`\delta \mu =0`$ by eqs.(10) and (15), we find that
$$\omega _2=\overline{}\delta u\overline{}\delta u(\overline{}u)^{n3}\overline{}\mathrm{ln}\mu ,n>2$$
(24)
is the second Witten-Zuckerman symplectic $`2`$-form on $`\mathrm{\Lambda }^2(())`$ $``$ $`[`$ $`\mathrm{\Lambda }^{n1,n}`$ $`()`$ $``$ $`\mathrm{\Lambda }^{n,n1}()`$ $`]`$. Physically the most interesting case of complex Monge-Ampère equation is the case $`n=2`$ but then $`\omega _2`$ vanishes identically.
We have arrived at the symplectic $`2`$-form (24) in an unconventional way. However, the ultimate justification for any result is a direct check of its properties. Eq.(24) satisfies all the properties required of a symplectic $`2`$-form for $`CMA_{n>2}`$ excluding the homogeneous case. Namely the check of the closed and conserved property of $`\omega _2`$ given by eqs.(12), (13) is immediate. It is remarkable that for $`n=2`$, the physically interesting case, $`\omega _2`$ vanishes identically. For $`n>2`$ the symplectic $`2`$-form vanishes only on shell. This property defines a Lagrangian submanifold , . The symplectic $`2`$-form expressed in terms of action angle variables vanishes on shell for integrable dynamical systems and the phase space is reduced to a Lagrangian submanifold. In eq.(24) we have the infinite dimensional analogue of this situation.
## 4 Complex Monge-Ampère-Liouville equation
The results we have presented above can be immediately extended to the complex Monge-Ampère-Liouville equation, $`\mu =\kappa e^{\mathrm{\Lambda }u}`$, which will henceforth be referred to as $`CMAL_n`$
$$\frac{1}{n!}(\overline{}u)^n=\kappa e^{\mathrm{\Lambda }u}{}_{}{}^{}1$$
(25)
that governs Einstein-Kähler metrics satisfying $`R_{i\overline{k}}=\mathrm{\Lambda }g_{i\overline{k}}`$. The Lagrangian (6) is now modified to the form
$$L=\frac{1}{(n+1)!}u\overline{}u(\overline{}u)^{n1}+\frac{\kappa }{\mathrm{\Lambda }}\left(e^{\mathrm{\Lambda }u}1\right)^{}1$$
(26)
and it can be verified that $`\omega `$ in eq.(2) remains unchanged as the first Witten-Zuckerman symplectic $`2`$-form for $`CMAL_n`$.
The analysis of the second Witten-Zuckerman symplectic $`2`$-form for $`CMAL_n`$ starts with the appropriate modification of the Hilbert Lagrangian
$$L_\mathrm{\Lambda }=\frac{1}{n!}\left[(\overline{}u)^{n1}\overline{}\mathrm{ln}\mu \mathrm{\Lambda }(\overline{}u)^n\right]$$
(27)
and once again we find that it is a total divergence. As in the case of eq.(21) it can be written as a divergence in as many different ways. The result for the second Witten-Zuckerman symplectic $`2`$-form is given by
$$\omega _{2\mathrm{\Lambda }}=\overline{}\delta u\overline{}\delta u(\overline{}u)^{n3}\overline{}\mathrm{ln}\mu \mathrm{\Lambda }\overline{}\delta u\overline{}\delta u(\overline{}u)^{n2}$$
(28)
again with the proviso $`n>2`$. Just as in the case of $`\omega _2`$ we find that $`\omega _{2\mathrm{\Lambda }}`$ also vanishes when the Einstein-Kähler condition is satisfied. This is manifest when we put $`\mu =\kappa e^{\mathrm{\Lambda }u}`$ in eq.(28). Hence, as in the case of $`CMA_n`$, the second Witten-Zuckerman symplectic $`2`$-form (28) for $`CMAL_n`$ also defines a Lagrangian submanifold.
## 5 Conclusion
We have considered the covariant symplectic structure of complex Monge-Ampère and Monge-Ampère-Liouville equations in the covariant framework of the Witten-Zuckerman formalism adapted to holomorphic and anti-holomorphic differential forms. The Lagrangians (6) and (26) directly lead to the non-degenerate Witten-Zuckerman symplectic $`2`$-form (2) in arbitrary dimension and for all cases elliptic, hyperbolic and homogeneous.
In order to prove the complete integrability of $`CMA_n`$ we need two such symplectic $`2`$-forms and use the theorem of Magri . To this end we considered the Hilbert action principle (18) and (27) for Euclideanized vacuum Einstein field equations which are satisfied by virtue of $`CMA(L)_n`$ and arrived at the symplectic $`2`$-forms (24) and (28). Since Hilbert’s Lagrangian is a divergence for Kähler metrics, it serves as a topological invariant rather than a Lagrangian for the Euclideanized Einstein field equations. Nevertheless, we were able to obtain the symplectic $`2`$-forms (24) and (28) because only the boundary terms in the first variation of the action play a significant role in the Witten-Zuckerman construction. For $`n>2`$ they satisfy all the properties required of a symplectic $`2`$-form for the complex elliptic, or hyperbolic Monge-Ampère-(Liouville) equation: $`CMA(L)_{n>2}`$ admits bisymplectic structure.
## 6 Acknowledgement
I thank H. Gümral, A. S. Fokas and M. B. Sheftel’ for interesting conversations. I thank also the referees of this paper for remarks which made me clarify the meaning of some points and for raising the question about Lie-Poisson structure.
|
warning/0004/cond-mat0004183.html
|
ar5iv
|
text
|
# Morphology transitions in three-dimensional domain growth with Gaussian random fields
## I Introduction
Many physical processes involve motion of the interface between two domains through a disordered media. Classic examples are magnetic domain growth, fluid invasion of porous media and fluid segregation in gels. Previous studies of such systems have shown that changing the strength of disorder leads to critical transitions between different growth morphologies. Most theoretical work on the subject has been done on models with an underlying crystalline lattice of spins or pores. The crystalline anisotropy may lead to a faceted growth regime in the low disorder limit. As disorder increases, there may be a transition to compact growth with a self-affine interface. In the high disorder limit, growing domains are self-similar fractals, characteristic of percolation.
Studies of driven interfaces in the zero-temperature random-field (and random-bond) Ising model (RFIM) indicate that the dimensionality, coordination number, and distribution of random fields are all important in determining the sequence of morphological transitions. In two dimensions (2D), there is a transition from self-similar to faceted growth at a critical value of disorder, if the the distribution of random fields is bounded. The critical behavior is not universal, and can be related to the analytic form of the tails in the distribution. For an unbounded Gaussian distribution, percolative growth occurs for any finite strength of disorder. In the three dimensional (3D) RFIM with a bounded distribution of disorder all three types of growth morphology, faceted, self-affine, and percolative, are found as the strength of disorder increases. Analysis of growth probabilities suggests that the transition from percolative to self-affine growth might be universal, but that the self-affine to faceted transition is non-universal and could be suppressed by an unbounded distribution of fields.
The suggestion that faceted growth may be eliminated by unbounded distributions of disorder in 3D is supported by renormalization group studies of equilibrium interface conformations. Calculations for a Gaussian distribution of random fields show that faceted interfaces are suppressed by any amount of disorder in dimensions $`d3`$. Porous media and some magnetic systems do not have any underlying crystalline structure, and it is only introduced in the models for computational convenience. Thus suppression of faceted growth by unbounded distributions of disorder leads lattice models to provide a more accurate description of real systems.
In this paper we explore the effect of unbounded Gaussian disorder on growth morphology transitions in the 3D RFIM at zero temperature. We find that this unbounded distribution does indeed eliminate the faceted growth regime. There is a single transition from self-affine growth at low disorder to percolative growth at high disorder. For each value of the disorder we find the critical field $`H_c`$ needed to initiate steady growth. The critical behavior at the onset of growth is analyzed, and the critical exponents are consistent with previous results for bounded distributions of random fields.
The transition between self-similar and self-affine growth is shown to be a multi-critical point. We identify lengths that diverge as the strength of disorder is varied along the line of critical fields, and evaluate critical exponents. Our analysis reveals problems with previous work on bounded distributions of random fields. These studies identified a “fingerwidth” with the correlation length that diverges at the multi-critical point. However, our work with larger systems shows that this fingerwidth does not diverge, and its saturation leads to errors in the determination of critical properties. New work will be needed to determine whether the self-affine to percolative transitions for bounded and unbounded distributions are in the same universality class. However, we can compare our results to studies of magnetic hysteresis in the 3D RFIM with Gaussian random fields by Perković et al.. These studies use a different growth algorithm, and we find that this leads to changes in the critical disorder and critical exponents.
The paper is organized as follows. In Sec. II we describe the growth model. Studies of the critical field and critical exponents at the onset of motion are presented in Sec. III A. The transition in growth morphology is examined in Sec. III B, and Sec. IV contains our general summary and conclusions.
## II Growth model and algorithm
The energy of a system of Ising spins ($`s_i=\pm 1`$) at the sites $`i`$ of a simple cubic lattice is written as
$$=\underset{<i,j>}{}s_is_j\underset{i}{}(\eta _i+H)s_i.$$
(1)
The first term on the right-hand side of Eq. (1) represents the ferromagnetic coupling between nearest neighbor spins, with the exchange coupling taken as the energy unit. The second term gives the interaction of each spin with the uniform external magnetic field $`H`$ and with the random local field $`\eta _i`$. The local fields are uncorrelated, following a Gaussian distribution function whose width $`\mathrm{\Delta }`$ quantifies the degree of disorder:
$$P(\eta )=(2\pi \mathrm{\Delta }^2)^{1/2}e^{\eta ^2/(2\mathrm{\Delta }^2)}.$$
(2)
As in previous studies, the simulation cell is a cube of side $`L`$, with the lattice constant taken as the unit of length. Periodic boundary conditions are imposed along the $`x`$ and $`y`$ directions. All spins are initially anti-parallel to the external field, i.e. $`s_i=1`$, except those in the bottom layer, $`z=1`$. This layer constitutes the “seed” for growth of the +1 domain, which is driven by the external field. The orientation of this seed plane does not affect critical behavior in the self-affine and self-similar growth regimes of interest here, but does affect the faceted growth seen for weak, bounded distributions of disorder.
Growth proceeds through single spin flips at zero temperature. Whenever a spin flip from $`s_i=1`$ to $`s_i=+1`$ lowers the total energy, it is implemented. However, only spins on the growing interface are allowed to flip. This restriction is motivated by the fluid invasion problem, and differentiates our dynamics from the model considered by Perković et al. in studies of hysteresis in magnetic systems.
We have developed a new memory-efficient algorithm that allows us to consider cells with $`L`$ as large as 1152 in the present study. The simulation cell is subdivided into smaller cubic cells of side $`s`$. The $`s^3`$ spins of a sub-cell remain active in memory only while at least one of them is a “flippable” interface spin. Growth occurs at fixed driving field $`H`$. This allows the local random field to be encoded into a single byte that gives the minimum number of additional neighbors needed to flip the spin at that site. This number is decreased by one every time the spin acquires a new $`+1`$ neighbor. When enough neighbors are present, the spin is flipped. For reasons discussed in Sec. III B, we flipped all spins that were completely surrounded by $`+1`$ neighbors, regardless of their local field. Such spins can not affect growth at any other site, and being able to remove the cells containing them from memory allowed us to treat larger systems. Preliminary runs where these spins were not flipped gave equivalent results.
At each $`L`$ and $`\mathrm{\Delta }`$, domains were grown in an ensemble of samples with different configurations of $`\{\eta _i\}`$. In each sample, growth was stopped when all spins on the interface were stable, or when the interface first reached the top of the system ($`z=L`$). The morphology of the domain was then analyzed as described below.
## III Results
### A The critical field
The external field $`H`$ provides a driving force that causes the domain of $`+1`$ spins to grow. If $`H`$ is too small, the domain wall will remain pinned near the bottom of the system, while large $`H`$ will cause all spins to flip. Figure 1 shows the probability, $`P_{top}`$, for a domain to grow to the top of the cell as a function of $`H`$ at several different system sizes. Results for two values of $`\mathrm{\Delta }`$ corresponding to self-affine ($`\mathrm{\Delta }=2.1`$) and self-similar ($`\mathrm{\Delta }=3.6`$) growth are shown. For each $`\mathrm{\Delta }`$, the range of $`H`$ over which the probability rises from 0 to 1 becomes narrower as $`L`$ increases. All of the curves intersect at a critical field $`H_c(\mathrm{\Delta })`$ and probability $`P_c(\mathrm{\Delta })`$. If $`H>H_c`$, the probability approaches unity in the thermodynamic limit $`(L\mathrm{})`$, while the probability vanishes in this limit if $`H<H_c`$.
Previous studies with a uniform distribution of random fields have shown that there is a diverging correlation length $`\xi _H|HH_c|^{\nu _H}`$ as $`|HH_c|0`$. The critical exponent in the self-similar growth regime is consistent with the 3D percolation exponent, $`\nu _H=0.88\pm 0.02`$, while $`\nu _H=0.75\pm 0.02`$ for self-affine growth.
Near $`H_c`$, the probability for the domain to span the system should only depend on the ratio of system size to $`\xi _H`$. This suggests that when probability is plotted against $`(HH_c)L^{1/\nu _H}`$ the results for all system sizes should collapse onto a universal curve. Fig. 2 verifies that the data from Fig. 1 are consistent with this ansatz, and with the values of $`\nu _H`$ that were found for uniform distributions of random fields.
Examination of the spanning probability for other values of $`\mathrm{\Delta }`$ shows that $`P_c(\mathrm{\Delta })`$ is always near 2/3 for self-similar growth and decreases with $`\mathrm{\Delta }`$ in the self-affine regime. To determine $`H_c(\mathrm{\Delta })`$ we worked with the largest accessible system size ($`L=`$768 or 1152) and found the value of $`H`$ that gave a spanning probability between 0.5 and 0.7. This determines $`H_c`$ with an accuracy of about 0.0001. We found that even larger errors in the growth field did not change the morphology of spanning domains that is analyzed in the next section.
In Figure 3 we show the variation of $`H_c`$ with disorder for the range of interest in the present study. At small $`\mathrm{\Delta }`$ the value of $`H_c`$ increases monotonically. There is a maximum near $`\mathrm{\Delta }=2.5`$, and then $`H_c`$ drops monotonically, becoming negative for $`\mathrm{\Delta }7`$.
### B Transition in Growth Morphology
As expected, the faceted growth regime seen for bounded distributions of random fields was suppressed by Gaussian disorder. The only transition that we observed was from self-similar growth at high disorder to self-affine growth at low disorder.
Self-similar growth is isotropic, while self-affine growth has a well-defined direction at long-length scales. Previous work on other models shows that the transition between these two growth regimes is a multi-critical point at some $`H_c`$ and $`\mathrm{\Delta }_c`$. An order parameter can be defined in analogy to equilibrium magnetic transitions as the average of the unit vector normal to the interface. For $`\mathrm{\Delta }<\mathrm{\Delta }_c`$ this average is finite, while for $`\mathrm{\Delta }>\mathrm{\Delta }_c`$ the order parameter vanishes. A correlation length, $`\xi `$, that diverges at $`\mathrm{\Delta }_c`$ can also be defined. As $`\mathrm{\Delta }`$ approaches $`\mathrm{\Delta }_c`$ from above (self-similar regime), longer and longer segments of the domain wall advance in the same direction. In the self-affine regime, deviations from the mean direction occur over longer and longer length scales as $`\mathrm{\Delta }`$ increases to $`\mathrm{\Delta }_c`$. In the following subsections we examine the morphology of domains and use finite-size scaling to determine $`\mathrm{\Delta }_c`$ and the exponent $`\nu `$ that describes the divergence of $`\xi `$ as $`\mathrm{\Delta }\mathrm{\Delta }_c`$.
#### 1 Fingerwidths
Previous experimental and theoretical studies have used a simple measure of the range of correlations. A fingerwidth $`w`$ was calculated by examining lines of adjacent nearest-neighbor spins, and averaging the length of contiguous segments of $`+1`$ spins (or fluid-invaded regions). Results obtained from our simulations are given in Figure 4(a). As in previous work, $`w`$ is independent of $`L`$ at high disorder and proportional to $`L`$ at low disorder. If one assumes that $`w`$ diverges at $`\mathrm{\Delta }_c`$ in the thermodynamic limit, $`w(\mathrm{\Delta }\mathrm{\Delta }_c)^\nu `$, then one can determine $`\mathrm{\Delta }_c`$ and $`\nu `$ from finite-size scaling collapses of the fingerwidth data. While this assumption was used in previous studies, our results with larger system sizes indicate that it is not justified. Finite-size scaling collapses become worse and worse as the range of $`L`$ increases, and give estimates for $`\mathrm{\Delta }_c`$ that are clearly in the self-affine regime.
The reason that $`w`$ does not diverge at $`\mathrm{\Delta }_c`$ is quite simple. Even in the self-affine regime there are small clusters of unflippable spins (Fig. 5) due to the tails in the distribution of $`\eta _i`$. These clusters are left behind by the advancing interface, and do not affect the morphology at long-length scales. However they do lead to a finite value of $`w`$. The limiting value of $`w`$ can be estimated from the probability for single isolated spins, pairs of spins, etc.. The upper bound for $`w`$ from single spins and pairs is indicated by a dashed line in Fig. 4(a). This upper bound clearly inhibits divergence of the fingerwidth with $`L`$ at any finite $`\mathrm{\Delta }`$. It is only about 300 at the value of $`\mathrm{\Delta }_c2.5`$ determined below, and lower bounds would be obtained by considering larger clusters.
We attempted to define improved fingerwidths by eliminating all isolated spins, and then all pairs of isolated spins before determining the fingerwidth. However, this procedure was inefficient and did not converge rapidly. In the following we focus entirely on the morphology of the external interface, and thus eliminate surrounded unflipped regions of all sizes. The scaling behavior in the self-affine regime is only associated with the external interface: Once the surrounded regions are left behind, they become irrelevant. In the self-similar regime we know that the external interface of a percolation cluster has the same fractal dimension as the entire cluster. Thus the external interface should give nearly the same fingerwidth as the cluster.
The solid line in Figure 4(a) shows the fingerwidth calculated from the external interface, $`w_e`$, as a function of $`\mathrm{\Delta }`$ at $`L=768`$. As expected, the unflipped regions have no effect on fingerwidth for $`\mathrm{\Delta }3`$, where the interface forms a fractal percolation pattern with narrow fingers. However as $`\mathrm{\Delta }`$ decreases towards $`\mathrm{\Delta }_c2.5`$, the solid line rises sharply above the other data points. Figure 4(b) shows how the external fingerwidth changes with $`L`$. Finite-size scaling collapses of $`w_e`$ are discussed below.
#### 2 Interface roughness
Self-affine interfaces are characterized by the scaling properties of the interface roughness. Due to our periodic boundary conditions, the average direction of the external interface is normal to $`z`$. Its position is given by a height $`h(x,y)`$ that may be multivalued. The roughness $`\rho (\mathrm{})`$ over a square region of side $`\mathrm{}`$ in the $`xy`$ plane can be quantified by the root-mean-squared (rms) variation in $`h`$
$$\rho (\mathrm{})=\sqrt{(hh)^2_{\mathrm{}}},$$
(3)
where $`h`$ is the average over a given square and $`_{\mathrm{}}`$ indicates an average over all square regions of side $`\mathrm{}`$. For a self-affine interface, $`\rho (\mathrm{})\mathrm{}^\alpha `$ at large $`\mathrm{}`$. The roughness exponent $`\alpha <1`$ characterizes the degree of anisotropy, and would be unity for a self-similar fractal.
Figure 5 shows a cross-section through a system of size $`L=96`$ for $`\mathrm{\Delta }=2.1`$, which is well into the self-affine regime. The external interface (circles) is a multi-valued function. In some regions there are overhangs where the interface extends over itself. These overhangs are necessary if the interface is to grow around small clusters of unflippable spins (crosses). The size of overhangs and of clusters of unflipped spins increases as $`\mathrm{\Delta }`$ rises to $`\mathrm{\Delta }_c`$.
Previous work shows that overhangs can change the scaling of $`\rho (\mathrm{})`$ at small $`\mathrm{}`$. One way of highlighting their effect is to compare $`\rho (\mathrm{})`$ with the roughness $`\rho _t(\mathrm{})`$ of the single-valued interface, $`h_t(x,y)`$, obtained by taking the top (highest) point on the external interface at each $`\mathrm{}`$ (closed circles in Fig. 5). Figure 6 shows log-log plots of both quantities vs. $`\mathrm{}`$ at the indicated values of $`\mathrm{\Delta }`$. For $`\mathrm{\Delta }2.2`$, results for the external (solid symbols) and single-valued (open symbols) interfaces converge at large $`\mathrm{}`$. The slope of the curves in the converged region is consistent with the roughness exponent $`\alpha =2/3`$ that is predicted from scaling arguments and observed in previous simulations. At small $`\mathrm{}`$ the dashed and full lines separate due to overhangs. The value of $`\rho _t`$ goes to zero at $`\mathrm{}=1`$, while the value of $`\rho `$ goes to the rms variation in height above a single point in the $`(x,y)`$ plane. The growing separation between dashed and full lines as $`\mathrm{\Delta }`$ increases towards $`\mathrm{\Delta }_c=2.52`$ shows that the size of the overhangs increases. For $`\mathrm{\Delta }=2.5`$ there is no convergence of the lines even at the largest $`\mathrm{}`$ and $`L`$ we could study. The overhang size is one measure of a diverging length as $`\mathrm{\Delta }`$ approaches $`\mathrm{\Delta }_c`$ from below, and is analyzed in following sections.
#### 3 Determining $`\mathrm{\Delta }_c`$
The data in Figures 4 and 6 give clear evidence of a morphological transition, and rough bounds on the value of $`\mathrm{\Delta }_c`$. The saturation of the external fingerwidth with increasing $`L`$ for $`\mathrm{\Delta }2.65`$ gives an upper bound for $`\mathrm{\Delta }_c`$ (Fig. 4(b)), while the merging of $`\rho `$ and $`\rho _t`$ at large $`\mathrm{}`$ for $`\mathrm{\Delta }2.2`$ gives a lower bound (Fig. 6). In this subsection we investigate other morphological attributes of the external interface that provide more accurate bounds for $`\mathrm{\Delta }_c`$. In the next subsection we determine the exponent $`\nu `$ through finite size scaling analysis of these quantities.
A single-valued interface, $`h_b(x,y)`$, can also be defined by taking the bottom (lowest) value of the external interface $`h`$ for each $`(x,y)`$. Figure 7(a) shows the average heights of the top and bottom interfaces, $`h_t`$ and $`h_b`$, as a function of $`\mathrm{\Delta }`$ at various system sizes. In the self-similar regime, the fractal external interface extends throughout the entire height of the cell. The value of $`h_t`$ is a constant fraction of the system size and $`h_b`$ is of order of the fingerwidth. The two averages converge in the self-affine regime, where the difference between them, $`dhh_th_b`$, is a measure of the height and abundance of overhangs.
In Figure 7(b) we plot $`dh/L`$ vs. $`\mathrm{\Delta }`$. As implied by the above discussion, this ratio vanishes at small $`\mathrm{\Delta }`$ and rises to a constant fraction at large $`\mathrm{\Delta }`$. The increase becomes sharper with increasing $`L`$, and there is a clear crossing of all curves at $`\mathrm{\Delta }2.5`$. This means that for $`\mathrm{\Delta }`$ below the crossing point $`dh/L`$ decreases with $`L`$, while above the crossing point $`dh/L`$ increases with system size. We conclude that the crossing point must coincide with $`\mathrm{\Delta }_c`$.
We have examined a variety of other quantities to confirm that all give consistent values of $`\mathrm{\Delta }_c`$ and to minimize the error bars. Figure 8 shows results for two probabilities that are related to the global minimum, $`h_{}`$, of each external interface. Data points connected by solid lines give the probability that $`h_{}/L`$ is greater than $`1/3`$. This probability is unity in the self-affine limit and drops to zero in the self-similar regime where the fractal external interface extends all the way to the bottom of the system. The data points connected by dashed lines in Figure 8 give the probability that $`h_{}`$ remains at the height of the initial seed plane. This probability is unity in the self-similar regime and drops to zero in the self-affine regime. Both probabilities exhibit sharper transitions from one to zero as $`L`$ increases, and should become step-functions at $`\mathrm{\Delta }_c`$ in the limit $`L\mathrm{}`$. Crossing points for the two probabilities in Fig. 8, the interface width in Fig. 7(b), and all other quantities that we examined are consistent with $`\mathrm{\Delta }_c=2.52\pm 0.03`$.
It is interesting to note that Fig. 3 shows a maximum in $`H_c(\mathrm{\Delta })`$ at $`\mathrm{\Delta }_c`$. This is a reasonable result, given the difference in growth mechanisms for self-affine and self-similar regimes. In the self-affine regime, the interface must advance across the entire width of the system. Thus $`H_c`$ is sensitive to the regions that are hardest to flip, and rises with $`\mathrm{\Delta }`$. In the self-similar regime, the interface follows the path of least resistance. Since the number of spins that must be flipped (the percolation probability) is less than $`1/2`$, increasing $`\mathrm{\Delta }`$ makes it easier to flip enough spins to span the system, and decreases $`H_c`$. The rate of decrease in $`H_c`$ can be calculated exactly from the percolation probability in the large $`\mathrm{\Delta }`$ limit, where spins are decorrelated. The asymptotic slope is: $`dH_c/d\mathrm{\Delta }=0.4907`$.
#### 4 Finite-size scaling determination of $`\nu `$
As in Sec. III A, we use finite-size scaling to determine the exponent $`\nu `$ that describes the diverging correlation length $`\xi `$ at $`\mathrm{\Delta }_c`$. The deviation from $`\mathrm{\Delta }_c`$ is measured by $`\delta (\mathrm{\Delta }\mathrm{\Delta }_c)/\mathrm{\Delta }`$. We assume that $`\xi \delta ^\nu `$, and that close to the critical disorder the only relevant lengths are $`\xi `$ and the system size $`L`$. Then dimensionless quantities like those shown in Figs. 7(b) and 8 can only depend on $`L/\xi `$, or equivalently $`L^{1/\nu }\delta `$. When plotted against $`L^{1/\nu }\delta `$, results for all system sizes should collapse onto a universal scaling function.
Figure 9 shows a scaling collapse for the data of Fig. 7(b). The scaled interface widths $`dh/L`$ for system sizes $`L=48`$, 96, 192, 384 and 768 collapse well onto a universal curve near $`L^{1/\nu }\delta =0`$. As $`|L^{1/\nu }\delta |`$ increases, the curves for small $`L`$ begin to deviate from the others. These deviations reflect corrections to scaling. They appear first at small $`L`$ because these data points are for larger values of $`\delta `$ than their counterparts at large $`L`$.
Since the magnitude of corrections to scaling is not known, there is some uncertainty in determining the values of $`\mathrm{\Delta }_c`$ and $`\nu `$. We found acceptable collapses for $`dh/L`$ with $`\mathrm{\Delta }_c=2.52\pm 0.03`$ and $`\nu =2.5\pm 0.3`$. Scaling collapses of other quantities related to the interface width, including $`\rho (1)`$, the ratio of the lowest and highest points on the entire interface, and the probabilities shown in Figure 8, all gave consistent ranges of $`\mathrm{\Delta }_c`$ and $`\nu `$. We also considered the scaling variable $`\delta ^{}(\mathrm{\Delta }\mathrm{\Delta }_c)/\mathrm{\Delta }_c`$, which gives different, and often larger, corrections to scaling. This led to a narrower region of scaling, but the same range of values for $`\nu `$ and $`\mathrm{\Delta }_c`$.
The scaling behavior of the external fingerwidth is more complicated, because the distribution of fingerwidths becomes bimodal in the self-affine regime. Most of the fingerwidths are essentially equal to the system size $`L`$. However, there is a significant fraction of very small fingerwidths from the region of width $`dh`$ near the top of the interface (see Fig. 5). These make a disproportionate contribution to $`w_e`$ that does not obey the scaling ansatz.
The contribution of small fingerwidths decreases if one calculates higher moments of the fingerwidth. We define
$$w_{en}=\sqrt[n]{w_e^n,}$$
(4)
where $`n=1`$ gives the mean width, $`n=2`$ gives the rms width, etc.. We find a steady improvement in finite-size scaling collapses with increasing $`n`$. Figure 10 shows that results for $`w_{e4}/L`$ collapse onto a universal curve at large $`L`$ with the same $`\mathrm{\Delta }_c`$ and $`\nu `$ used in Fig. 9. Best fits for $`\nu `$ increased consistently from $`2.0\pm 0.2`$ at $`n=1`$ to $`2.2\pm 0.2`$ for $`n=4`$, and the quality of the collapse showed progressive improvement.
If $`w_e`$ is proportional to the diverging correlation length, then it must diverge with the same exponent. To check this, we examined the slope of plots of $`\mathrm{log}_{10}w_{en}`$ against $`\mathrm{log}_{10}\delta `$. The slopes were indeed consistent with values of $`\nu `$ from finite-size scaling, although the uncertainties were somewhat larger.
## IV Summary and Discussions
We have studied the zero-temperature phase diagram for interface growth in the 3D RFIM with a Gaussian distribution of random fields. There is a single transition from self-affine growth below $`\mathrm{\Delta }_c`$ to percolative growth at larger disorder. The tails of the Gaussian distribution eliminate the faceted regime observed in previous 3D studies of bounded field distributions. This means that the Gaussian RFIM provides a more realistic description of transitions in systems that do not have an underlying crystalline lattice, such as random porous media or amorphous magnets. The phase diagram is also quite different from that for the 2D RFIM where Gaussian randomness suppresses the self-affine regime.
The critical behavior at the onset of motion in the self-affine and self-similar regimes was analyzed using finite-size scaling. The critical exponent $`\nu _h`$ that describes the diverging length scale as $`HH_c`$ was found to be $`0.75\pm 0.02`$ for self-affine growth and $`0.88\pm 0.02`$ for self-similar growth. These values are consistent with results for bounded distributions of disorder. We also found the same roughness exponent $`\alpha =2/3`$ in the self-affine regime (see Fig. 6). These results indicate that changing the form of the distribution of random fields does not change the universality class of the self-affine and self-similar growth regimes.
The multi-critical point that separates self-affine and percolative growth was also analyzed. We found that the fingerwidth used in previous work does not diverge and can not be used to determine $`\mathrm{\Delta }_c`$. Examination of the external interface revealed two lengths that did diverge at $`\mathrm{\Delta }_c`$: The overhang size $`dh`$ diverges as $`\mathrm{\Delta }`$ increases to $`\mathrm{\Delta }_c`$ in the self-affine regime, and the external fingerwidth $`w_e`$ diverges as $`\mathrm{\Delta }`$ decreases to $`\mathrm{\Delta }_c`$ in the percolative regime. Finite-size scaling collapses of these and other quantities gave consistent values for $`\mathrm{\Delta }_c=2.52\pm 0.03`$ and $`\nu =2.4\pm 0.4`$. The error bars on these quantities are estimates of systematic uncertainties due to corrections to scaling.
The value of $`\nu `$ determined previously for a bounded distribution random fields, $`\nu =3.0\pm 0.5`$, is consistent with our result. However, this value was determined from the fingerwidth and is not reliable. Future work is needed to determine whether bounded and unbounded distributions are in the same universality class.
Perković et al have determined the critical behavior for the 3D Gaussian RFIM using a growth algorithm that appears to be in a different universality class. They analyzed the integrated avalanche size distribution occurring in one branch of a hysteresis loop ($`H`$ increasing from $`\mathrm{}`$ to $`\mathrm{}`$). From the divergence as $`\mathrm{\Delta }`$ decreased to $`\mathrm{\Delta }_c`$ they found numerical values for the critical disorder, $`\mathrm{\Delta }_c=2.16\pm 0.03`$, and correlation length exponent, $`\nu =1.43\pm 0.18`$.
The discrepancy between our results and those of Perković et al. seems to result from a crucial difference in our growth algorithms. They allowed any spin-flip that lowered the energy, while we only allowed spins on the interface to flip. The exchange coupling between neighbors dominates in the low disorder limit, and spins are very unlikely to flip unless they are on the interface. Nowak et al. found that the difference between the two algorithms was negligible in simulations of low disorder growth with uniform disorder. The two algorithms should also yield the same percolating cluster in the high disorder limit, where interactions between neighbors become irrelevant. The problem maps onto ordinary percolation, and the order in which spins are flipped becomes irrelevant. Near $`\mathrm{\Delta }_c`$ these arguments break down, and the algorithms may give different results. At intermediate disorder, the exchange coupling is weak enough to allow clusters to flip ahead of the interface, and correlations are important enough that these flipped clusters can aid the advance of the approaching interface. One expects that both $`H_c`$ and $`\mathrm{\Delta }_c`$ will be lowered by the advance clusters in Perković et al.’s model, and this is consistent with the numerical results.
###### Acknowledgements.
This work was partially supported by CNPq, CAPES, FAPERJ and FUJB (Brazil), by National Science Foundation Grant DMR 9634131, and by Intel Corporation through the donation of workstations that were used for our simulations. We thank G. Magnusson for assistance in implementing the growth algorithm, and R. Paredo and C. S. Nolle for useful conversations.
|
warning/0004/cond-mat0004148.html
|
ar5iv
|
text
|
# Untitled Document
Scaling and Persistence in the
Two-Dimensional Ising Model
S. Jain<sup>1</sup> and H. Flynn,
School of Mathematics and Computing,
University of Derby,
Kedleston Road,
Derby DE22 1GB,
U.K.
<sup>1</sup> Address after 17<sup>th</sup> April 2000:
School of Engineering and Applied Sciences,
Aston University,
Aston Triangle,
Birmingham B4 7ET,
U.K.
Classification Numbers:
05.20-y, 05.50+q, 05.70.Ln, 64.60.Cn, 75.10.Hk, 75.40.Mg
ABSTRACT
The spatial distribution of persistent spins at zero-temperature in the pure two-dimensional Ising model is investigated numerically. A persistence correlation length, $`\xi (t)t^Z`$ is identified such that for length scales $`r<<\xi (t)`$ the persistent spins form a fractal with dimension $`d_f`$; for length scales $`r>>\xi (t)`$ the distribution of persistent spins is homogeneous. The zero-temperature persistence exponent, $`\theta `$, is found to satisfy the scaling relation $`\theta =Z(2d_f)`$ with $`\theta =0.209\pm 0.002,Z=1/2`$ and $`d_f1.58`$.
The ‘persistence’ problem has attracted considerable interest in recent years \[1-9\]. In its most general form, it is concerned with the fraction of space which persists in its initial state up to some later time.
Hence, in the non-equilibrium dynamics of spin systems at zero-temperature we are interested in the fraction of spins, $`P(t)`$, that persist in the same state as at $`t=0`$ up to some later time $`t`$. For the pure ferromagnetic two-dimensional Ising model, $`P(t)`$ has been found to decay algebraically \[1-4\]
$$P(t)t^\theta $$
$`(1)`$
where $`\theta =0.209\pm 0.002`$ . Similar algebraic decay has been found in numerous other systems displaying persistence . Most of the recent theoretical effort has gone into obtaining the numerical value of $`\theta `$ for different models.
Very recently, Manoj and Ray have studied the spatial correlation of persistent sites in the $`1dA+A0`$ model. They found that the set of persistent sites in their $`1d`$ model forms a fractal over sufficiently small length scales.
In this letter we present the results of an extensive numerical study of the spatial distribution of persistent spins in the pure $`2d`$ Ising model at zero-temperature. As we will see, the $`2d`$ Ising model exhibits behaviour very similar to that found by Manoj and Ray in their simple $`1d`$ model.
The Hamiltonian for our model is given by
$$H=\underset{<ij>}{}S_iS_j$$
$`(2)`$
where $`S_i=\pm 1`$ are Ising spins situated on every site of a square lattice with periodic boundary conditions; the summation in Eqn. (2) runs over all nearest-neighbour pairs only.
The data presented in this work were obtained for a lattice with dimensions $`1000\times 1000(=N)`$.
Each simulation run begins at $`t=0`$ with a random ($`\pm 1`$) starting configuration of the spins and then we update the lattice via single spin flip zero-temperature Glauber dynamics . The rule we use is: always flip if the energy change is negative, never flip if the energy change is positive and flip at random if the energy change is zero.
For each spin $`S_i`$ we define
$$n_i(t)=(S_i(t)S_i(0)+1)/2.$$
$`(3)`$
Hence, if $`n_i(t)=1`$ for all $`t0`$ spin $`S_i`$ is persistent at time $`t`$; $`n_i(t)=0`$ otherwise.
The total number, $`n(t)`$, of spins which have never flipped until time $`t`$ is then given by $`n(t)=_in_i(t),`$ and the persistence probability by
$$P(t)=\underset{i}{}<n_i(t)>/N$$
$`(4)`$
where $`<\mathrm{}>`$ indicates averages over different initial conditions and histories. We averaged over at least 100 different initial conditions and histories for each run.
To investigate the spatial correlations in this model, we follow Manoj and Ray and study the 2-point correlator defined by
$$C(r,t)=<n_i(t)n_{i+r}(t)>/<n_i(t)>,$$
$`(5)`$
where $`<\mathrm{}>`$ now also includes the average over the lattice shown explicitly in Eqn (4). $`C(r,t)`$ is simply the probability that spin $`n_{i+r}(t)`$ is persistent given that $`n_i(t)`$ is persistent, averaged over the entire lattice. According to , the 2-point correlator satisfies the following dynamic scaling relation
$$C(r,t)=P(t)f(r/\xi (t))$$
$`(6)`$
where $`\xi (t)`$ is the persistence correlation length and $`f(x)`$ is a scaling function such that
$$f(x)\{\begin{array}{cc}x^\alpha ,\hfill & \text{for }x<<1\text{;}\hfill \\ 1,\hfill & \text{for }x>>1\text{.}\hfill \end{array}$$
$`(7)`$
As a consequence, the expected behaviour of $`C(r,t)`$ in the two limits is given by
$$C(r,t)\{\begin{array}{cc}r^\alpha \hfill & \text{for }r<<\xi (t)\text{;}\hfill \\ t^\theta \hfill & \text{for }r>>\xi (t)\text{.}\hfill \end{array}$$
$`(8)`$
Clearly, as $`P(t)t^\theta `$, we must also have $`\xi ^\alpha t^\theta `$ to satisfy Eqn (8) in the limit $`r<<\xi (t)`$. Assuming a power-law divergence for the persistence correlation length with $`t`$ i.e. $`\xi (t)t^Z`$ then leads to the scaling relation $`Z\alpha =\theta `$. As we are working with the pure $`2d`$ Ising model at zero-temperature, we expect $`Z=1/2`$; our results are completely consistent with this assumption.
To examine the correlated region ($`r<<\xi (t)`$) we study the average number of persistent spins, $`n(l,t)`$, in a square grid with dimensions $`l\times l`$. As
$$n(l,t)=_0^lC(r,t)r𝑑r$$
$`(9)`$
we have that
$$n(l,t)\{\begin{array}{cc}l^{2\alpha }\hfill & \text{for }l<<\xi (t)\text{;}\hfill \\ l^2P(t)\hfill & \text{for }l>>\xi (t)\text{.}\hfill \end{array}$$
$`(10)`$
Hence, we expect the persistent spins to form a fractal with dimension $`d_f=2\alpha `$ for length scales $`l<<\xi (t)`$; the distribution is homogeneous on longer length scales, namely for $`l>>\xi (t)`$. We expect the crossover to occur at $`l\xi (t)t^{1/2}`$. The scaling form for $`n(l,t)`$ is given by
$$n(l,t)=l^2P(t)g(l/\xi (t)),$$
$`(11)`$
where $`g(x)`$ is a scaling function satisfying
$$g(x)\{\begin{array}{cc}x^\alpha \hfill & \text{for }x<<1\text{;}\hfill \\ 1\hfill & \text{for }x>>1\text{.}\hfill \end{array}$$
$`(12)`$
We now discuss our results.
Figure 1 shows a plot of the scaling function $`f(x)(=C(r,t)/P(t))`$ against $`x=r/\xi (t)`$ for various different values of $`t`$. We have assumed that $`\xi (t)t^{1/2}`$. The data in Fig 1 ranges over almost three orders of magnitude and is clearly consistent with this assumption. The large $`x`$ behaviour of $`f(x)`$ clearly follows the expected behaviour given in Eqn (7).
To extract a value for $`\alpha `$ we re-plot the data shown in Fig 1 on a log-log scale in Fig 2. The algebraic behaviour for $`x<<1`$ of the scaling function is confirmed by the linear fit. The slope of the straight line implies a value of $`\alpha =0.428\pm 0.007`$. Hence, the scaling relation would suggest that $`\theta =Z\alpha =0.214\pm 0.004`$. This is, of course, consistent with value $`(0.209\pm 0.002)`$ quoted above for $`\theta `$ .
We investigate the correlated regions by obtaining a direct estimate of the fractal dimension $`d_f`$. This is undertaken by first partitioning the lattice into square grids of size $`l\times l`$ with $`l`$ ranging from 4 to 250. The average number of persistent spins in each $`l\times l`$ square is then obtained.
In Figure 3 we plot $`\mathrm{ln}n(l,t)`$ versus $`\mathrm{ln}l`$ for $`t=10^2,10^3,5\times 10^3`$ and $`10^4`$. We notice that for each of the values of $`t`$, the behaviour over sufficiently small (typically, $`l<<\sqrt{t}`$) length scales is consistent with a fractal dimension $`d_f=2\alpha 1.58`$; over longer length scales (typically, $`l>>\sqrt{t}`$) we retrieve homogeneous behaviour ($`d_f=d=2`$). Actual values of $`d_f`$ range from $`d_f(t=10^2)1.62`$ to $`d_f(t=10^4)1.58`$. The straight lines, with slopes $`1.58`$ and $`2.00`$, shown in Fig 3 are linear fits to the behaviour in the two respective regimes for $`t=10^4`$.
We obtain an independent estimate for the exponent $`\alpha `$ by re-plotting the data for $`t=5\times 10^3`$ and $`10^4`$ in scaling form. Figure 4 shows a log-log plot of the scaling function $`g(x)=n(l,t)/l^2P(t)`$ against $`x`$ where $`x=l/\sqrt{t}`$. We see that the data clearly fall onto a single scaling curve consistent with the expected behaviour given in Eqn (12). On fitting all of the data for $`\mathrm{ln}x<0.5`$ we get a value of $`\alpha 0.438`$. However, restricting the linear fit to $`\mathrm{ln}x<1`$, as indicated by the straight line in Fig 4, would imply a value of $`\alpha 0.50`$. Although this is slightly higher than the value we obtained from the analysis of the scaling behaviour of the 2-point correlator (see Eqn (8)), it is, nevertheless, consistent with our value of the fractal dimension in the correlated regime.
To conclude, we have investigated the spatial distribution of persistent spins at zero-temperature in the pure two-dimensional Ising model. We find that the persistent spins form a fractal with dimension $`d_f1.58`$ for length scales $`r<<\xi (t)`$, where $`\xi (t)t^Z`$ is the persistence correlation length. Furthermore, the persistence exponent satisfies the scaling relation $`\theta =Z(2d_f)`$ with $`Z=1/2`$.
Acknowledgement
The simulations were performed partly on the SGI Origin 2000 at the University of Manchester made available by the Engineering and Physical Sciences Research Council (EPSRC), Great Britain, and also on in-house workstations and a PC. HF would like to thank the University of Derby for a Research Studentship
FIGURE CAPTIONS
Fig. 1
A plot of the scaling function $`f(x)(=C(r,t)/P(t))`$ against $`x`$ where $`x=r/\sqrt{t}`$ for $`t`$ ranging over approximately three orders of magnitude.
Fig. 2
A re-plot of the data shown in Figure 1 on a log-log scale. The straight line implies a value of $`\alpha =0.428\pm 0.007`$.
Fig. 3
A log-log plot of $`n(l,t)`$ against $`l`$. Here, $`n(l,t)`$ is the average number of persistent spins in a square ($`l\times l`$) grid at time $`t`$. The data is shown for (top) $`t=10^2\mathrm{},10^3+,5\times 10^3\text{ }\text{ }\text{ }\text{ }\text{ }`$ and $`10^4\times `$ (bottom). There is a clear crossover at $`l\sqrt{t}`$ from a fractal distribution with dimension $`d_f1.58`$ to a homogeneous one with $`d_f=d=2`$. The two straight lines (with slopes 1.58 and 2.00) are fits of the data in the two extreme cases for $`t=10^4`$.
Fig. 4
A plot of $`\mathrm{ln}g(x)`$ against $`\mathrm{ln}x=\mathrm{ln}l/\xi (t)`$. Here the scaling function $`g(x)=n(l,t)/l^2P(t)`$. The straight line has slope $`=0.50`$ and implies a value of $`\alpha 0.50`$.
REFERENCES
B. Derrida, A. J. Bray and C. Godreche, J.Phys. A 27, L357 (1994).
A.J. Bray, B. Derrida and C. Godreche, Europhys. Lett. 27, 177 (1994).
D. Stauffer J.Phys.A 27, 5029 (1994).
B. Derrida, V. Hakim and V. Pasquier, Phys. Rev. Lett. 75, 751 (1995); J. Stat. Phys. 85, 763 (1996).
S. Jain, Phys. Rev. E59, R2493 (1999).
S.N. Majumdar, C. Sire, A.J. Bray and S.J. Cornell, Phys. Rev. Lett. 77, 2867 (1996).
B. Derrida, V. Hakim and R. Zeitak, Phys. Rev. Lett. 77 2971 (1996).
S.N. Majumdar and A.J. Bray, Phys. Rev. Lett. 81 2626 (1998).
S.N. Majumdar, Curr. Sci. 77 370 (1999)
G. Manoj and P. Ray, J.Phys. A33, L109 (2000)
J.D. Gunton, M. San Miguel, and P.S. Sahni, Phase Transitions and Critical Phenomena, edited by C. Domb and J.L. Lebowitz (Academic Press, New York, 1983), vol 8
|
warning/0004/astro-ph0004140.html
|
ar5iv
|
text
|
# The sign of temperature inhomogeneities deduced from time-distance helioseismology
## I Introduction
In recent years it has become possible to measure the travel times of acoustic waves travelling through the outer layers of the Sun through ‘time-distance helioseismology’. These travel times are used to infer information about the sub-surface structure of the Sun and have revealed inhomogeneities in the wave propagation conditions. (Duvall et al. 1993, Duvall 1995, D’Silva & Duvall 1995, Duvall 1997, Duvall et al. 1998). These inhomogeneities indicate shorter sound-wave travel times and have been associated with active regions.
Slight shifts in global p-mode frequencies in the course of a solar cycle have also been detected (Woodard & Libbrecht 1993, Dziembowski et al. 2000). As in the case of the time-distance measurements, they have been found to be closely related to magnetic activity (Dziembowski et al. 2000), and indicate shorter travel times in the outermost layers of the convection zone. A common interpretation for both effects is that the sound speed in active regions is somewhat higher because of an increase in temperature. However, it was shown by Goldreich et al. (1991) that, though intuitively appealing, such a temperature increase is unlikely to be the cause of the mode frequency shifts. Contrary to expectation, a higher temperature causes longer travel times, because a temperature increase causes a slight expansion of the envelope. The path length increase caused by this expansion dominates over the increase in sound speed since the expansion is proportional to the temperature $`T`$, while the sound speed increases only as $`T^{1/2}`$. Instead of a temperature increase, Goldreich et al. propose that most of the effect is due to the photospheric magnetic field. A magnetic field increases the stiffness of the gas as experienced by pressure waves.
The expansion argument would also affect the interpretation of travel-time anomalies discovered by time-distance helioseismology. The purpose of this paper is to verify to what extent the expansion of the envelope associated with a temperature rise increases the wave travel times. In Sec. 2 we show that in most cases a temperature enhancement does in fact lead to longer travel times. To explain shorter travel times one might then invoke magnetic fields. Alternatively, the subsurface temperatures could be slightly lower in the active regions, the opposite of what is concluded in most analyses of time-distance measurements. These possibilities are discussed briefly in Sec. 3.
## II Model
One would like to investigate the effects of temperature inhomogeneities without making too many assumptions about their origin. On the other hand, for actual wave propagation calculations a well defined equilibrium model is needed. A simple model is a geostrophic one, in which the pressure balance within the inhomogeneity is attained through Coriolis forces associated with the solar rotation. This balance is probably valid on large scales (of the order of an active region). On smaller scales, this is probably not a good model; see Sec.II C for a detailed discussion.
For definiteness in discussing the model, we regard a region of increased temperatures. Since the effects are nearly linear in $`\delta T`$, the results apply equally with opposite sign to regions of lower than average sub-surface temperatures.
### A Equilibrium model
In a plane-parallel envelope in hydrostatic equilibrium under constant gravitational acceleration $`g`$, the pressure, $`p`$, satisfies
$$\frac{dp}{dz}=g\rho ,$$
(1)
where $`z`$ is the depth beneath some reference layer. Assuming a polytropic relation between $`p`$ and density, $`\rho `$, i.e.
$$p=K\rho ^{1+1/n},$$
(2)
one finds that the pressure varies with depth as
$$p=p_0\left(\frac{z}{z_0}\right)^{n+1}$$
(3)
and similarly
$$\rho =\rho _0\left(\frac{z}{z_0}\right)^n,$$
(4)
where $`z_0`$ denotes the depth of some reference layer from where the polytropic layer extends downward. Above $`z_0`$ one may, for example, want to match the polytrope onto an isothermal atmosphere. Here we only consider rays that lie completely within the polytrope.
The sound speed is thus
$$c^2=\frac{\gamma g}{n+1}z,$$
(5)
where $`\gamma `$ is the ratio of the specific heats. The acoustic cut-off frequency is approximately given by
$$\omega _\mathrm{c}=\frac{c}{2H}=\left(\frac{\gamma g}{n+1}\right)^{1/2}\frac{n}{2}z^{1/2},$$
(6)
where $`H`$ is the density scale height defined as
$$H=\left(\frac{\mathrm{d}\mathrm{ln}\rho }{\mathrm{d}z}\right)^1=z/n.$$
(7)
This yields a depth of the upper turning point of
$$z_\mathrm{t}=\left(\frac{n}{2\omega }\right)^2\frac{\gamma g}{n+1}.$$
(8)
Assuming the gas to be ideal and the ionization to be constant, the temperature is given by
$$T=\frac{\mu p}{R\rho }=\frac{\mu g}{R(n+1)}z,$$
(9)
where $`\mu `$ is the mean molecular mass and $`R`$ the gas constant. Thus, the temperature profile is determined by the polytropic index $`n`$.
The entropy is given by
$$S\mathrm{ln}p/\rho ^\gamma =(1+1/n\gamma )\mathrm{ln}\rho +\mathrm{cst}.,$$
(10)
where an ideal gas equation of state has been assumed. Since the stratification of the convection zone is close to adiabatic, the values for the polytropic index and $`\gamma `$ are related by
$$\gamma 1+1/n.$$
(11)
### B Time-distance calculations
We now investigate the travel times of those waves that enter a column of hotter material, are then reflected near the surface and subsequently leave the hotter region again. We treat this problem in two dimensions (i.e. a slab geometry); we ignore the effect of the advection of the waves, assuming that to equal parts the waves travel in and out of the inhomogeneity and that therefore the net effect of advection vanishes to first order. The flows may cause second-order effects ($`v^2`$, and independent of the direction of the flow) on the travel times. This is discussed in section II C.
In the spirit of the JWKB approximation we will regard the sound waves as locally plane (see Gough 1993). In this approximation, which is commonly made in local helioseismology, the waves are assumed to follow rays that obey the laws of geometrical acoustics. For simplicity, we will regard the medium as plane-parallel.
We let the ray traverse a column in which the temperature profile is raised to $`T_1(z)`$. This will be achieved using two very simple models. In model 1 the temperature in the hotter column is raised by lowering the polytropic index to $`n_1<n_0`$ (subscript 0 shall denote the corresponding values of the ‘normal’ Sun). The temperature difference in this model is thus proportional to the depth $`z`$, and its depth is assumed infinite.
Model 2 mimics some proposed models for temperature enhancements associated with active regions (Kuhn and Stein 1996). In these models the source is assumed to be an entropy increase located at some depth $`D`$, the effect of which extends to the surface. This can be incorporated in a model with a higher, but still depth-independent entropy $`S`$. In a polytropic model, this corresponds to an increased value for the polytropic constant $`K`$. If this model is in lateral pressure balance at the source depth $`D`$, vertical equilibrium implies a vertical shift, such that the top of the polytrope is at some depth $`z_1<0`$. If $`K_1`$ is the polytropic constant of this model, one finds that
$$\left(\frac{K_1}{K_0}\right)^{\frac{n}{n+1}}=1+\frac{z_1}{D}.$$
(12)
For small entropy changes $`\delta K=K_1K_0`$ we have
$$\frac{z_1}{D}=\frac{n}{n+1}\frac{\delta K}{K_0}.$$
(13)
The vertical shift is thus proportional to the depth of the source. In the example shown in Fig. 2 we have chosen a depth of $`D=21`$ Mm and $`\frac{\delta K}{K_0}`$ was chosen to be $`610^3`$, so that $`z_1120`$ km.
In the hotter column the upper turning point is raised over the one in the adjacent colder medium. In case 1, $`n_1<n_0`$, which decreases the depth $`z_\mathrm{t}`$ of the upper turning point (see Eq.(8)) and in case 2 the upper turning point is shifted upward due to expansion.
The ray equations follow from the dispersion relation, which can be written as
$$k^2=k_\mathrm{v}^2+k_\mathrm{h}^2=\frac{\omega ^2\omega _\mathrm{c}^2}{c^2},$$
(14)
where $`k_\mathrm{v}`$ and $`k_\mathrm{h}`$ are the vertical and horizontal components of the wavevector $`𝐤`$.
The advantage of using a polytropic approximation (aside from the fact that the approximation is quite good for the upper layers of the Sun) is that the ray equations can be written down analytically. If $`x`$ is the horizontal coordinate in the plane of the wave and assuming that $`k_x^2+k_z^2=\omega ^2/c^2`$ (i.e. ignoring $`\omega _\mathrm{c}`$), one can write
$`x`$ $`=`$ $`{\displaystyle \frac{dx}{dz}𝑑z}={\displaystyle \frac{k_x}{k_z}𝑑z}={\displaystyle \frac{k_x}{\sqrt{\omega ^2/c^2k_x^2}}𝑑z}={\displaystyle \frac{dz}{\sqrt{a/z1}}}`$ (15)
$`=`$ $`a[\mathrm{sin}^1(z/a)^{1/2}(z/a)^{1/2}(1z/a)^{1/2}],`$ (16)
where $`a`$ is the depth of the lower turning point given by $`a=\omega ^2/c_0^2k_x^2`$. The time taken for this traverse is given by the integral along the ray
$`\tau `$ $`=`$ $`{\displaystyle \frac{k}{\omega }𝑑s}=c_0^1{\displaystyle z^{1/2}(1z/a)^{1/2}𝑑z}`$ (17)
$`=`$ $`c_0^1a^{1/2}\mathrm{sin}^1(z/a)^{1/2}.`$ (18)
When the cut-off frequency is included in the ray equations, the expressions become a bit more complicated, but the integrals can still be solved analytically.
At the vertical interface between the two regions of different temperatures, the rays are refracted. We are using the approximations of geometrical acoustics and calculate the angle of refraction by Snell’s law. Then we compare the travel times in the homogeneous Sun with the corresponding times in the scenario depicted in Fig. 1. In all cases, we consider rays which bounce (i.e. have their upper turning point) inside the inhomogeneity (see Fig. 1). In Fig. 2 we have plotted the relative travel time difference $`\delta \tau /\tau =(\tau _{\mathrm{inhom}}\tau _{\mathrm{hom}})/\tau _{\mathrm{hom}}`$ (between rays in a homogeneous and inhomogeneous Sun) for rays of different interskip distances (and depths) for a fixed width of the hotter region in models 1 and 2. In Fig. 3 $`\delta \tau /\tau `$ is shown for a ray of fixed depth but as a function of the width of the hotter region.
Fig. 2 shows that the travel-time can be positive, i.e. that the waves can take longer in the presence of the hot column. This implies that the hot column retards the waves, because the effect of the raised upper turning point outweighs the increased sound speed in the hotter column. However, $`\delta \tau /\tau `$ decreases with increasing depth of the ray, since the waves spend more time in the hotter (and therefore faster) region whereas the effect of the raised upper turning point depends very little on the depth of the rays (since the rays are almost vertical near the upper turning point).
In Fig. 3 one can note that the travel-time is positive as long as the width of the hot region is small compared to the length of the ray. But $`\delta \tau /\tau `$ decreases with increasing $`w`$ because the sound speed is higher in the hotter region, and this eventually dominates the effect of the lengthening of the ray through the raised upper turning point: the travel-time difference becomes negative.
### C Horizontal equilibrium of a temperature inhomogeneity
So far we have evaded the question about what restores the horizontal pressure equilibrium. The difference in temperature between the inhomogeneity and its surroundings causes a difference in gas pressure that is subject to adjustments on the short hydrodynamic time scale. Consider a temperature enhancement in a patch (an active region, say) at colatitude $`\theta `$, extending below the surface as a column of width $`L`$ (Fig. 4). In geostrophic balance, the pressure excess $`\delta p`$ is balanced by the Coriolis force acting on a flow $`𝐯`$ around the column, i.e.
$$(\delta p)_\mathrm{h}=2(\rho 𝐯\times \mathrm{\Omega })_\mathrm{h},$$
(19)
where the subscript $`\mathrm{h}`$ indicates the horizontal components.
The flow is concentrated at the boundary of the column, where the pressure gradients are greatest. Inside the column, the flow vanishes and the excess pressure $`\delta p`$ is just given by hydrostatic equilibrium,
$$p_r\delta p=g\delta \rho .$$
(20)
The column is assumed to extend down to some depth $`D`$ where $`\delta p=0`$. Below this depth, the temperature excess vanishes. The flow speed is then maximal at the surface, and vanishes at depth $`D`$.
In the preceding paragraph, we have proposed a geostrophic balance for the pressure changes. For relatively small inhomogeneities the effects of rotation are small, and a geostrophic balance is not realistic. Instead one could consider upwellings and downdrafts. If there is a source of heat at some depth below the surface, a circulation is set up, with upwelling above the heat source. This flow is driven by the higher gas pressure inside the rising column, and its velocity $`𝐯`$ is such that the dynamic pressure of the flow, $`\rho 𝐯𝐯\rho v^2/r_c`$, balances the pressure difference ($`r_c`$ is the gradient length scale of the velocity).
The flow advects the p-modes (to first order in $`v`$) but also has a second-order effect on their propagation speed. The second-order effect is proportional to, and of the same order of magnitude as the temperature increase and likely to give a positive contribution to the sound speed. This is because flows with vorticity speed up on compression. In the previous section we neglected the contribution to the sound speed which is provided by the compressibility of the flows. Its effect is to increase the sound speed, but the extent of the effect is difficult to model.
For a qualitative estimate, we assume the effect of the flow on the waves to be local and isotropic. In this case, the effect on wave propagation would be a mere increase of the propagation speed. Thus, we repeat the calculations presented above with an increased sound velocity inside the hotter region:
$$\delta c/c=\alpha \frac{1}{2}\delta T/T,$$
(21)
where $`\alpha `$ is a factor of order unity. In the absence of flows, we would have $`\alpha =1`$. Fig. 5 shows the same case as depicted in Fig. 2 only with $`\alpha =1.01`$. As expected, the relative time delay is smaller than in the absence of flows and even for a slight relative increase of the sound speed by 1 % the time delay quickly becomes negative. Therefore, one will have to know the properties of the turbulent medium very accurately before one can definitly predict its effect on the travel times.
In the above, the flows were presumed to have no effect on the position of the upper turning point. In reality, a ‘turbulent’ pressure increase due to small scale flows expands the region vertically. This would raise the upper turning point levels, and increase the travel times. In the case of granulation flows, this effect is the main contribution to the p-mode frequency anomalies associated with the outer envelope (Rosenthal, et al. 1999). By leaving it out, we are probably underestimating the travel time increase (decrease) in hotter (cooler) regions.
## III Summary and discussion
Time-distance helioseismology indicates the presence of subsurface inhomogeneities. In order to map these inhomogeneities quantitatively, a propagation model is needed. In most analyses it is assumed that shorter travel times correspond to higher propagation speeds. While this is correct if the inhomogeneities are due to a vertical magnetic field, the most obvious possibility, i.e. a change in temperature, requires careful treatment. This is because a temperature change has two side effects in addition to a change in propagation speed. Since hot gas is less dense, the resulting positive buoyancy causes the envelope to expand vertically in regions of higher temperature. Secondly, the resulting horizontal imbalance also sets up a circulation flow. Both have effects on acoustic wave propagation.
We have analysed here the effects of a temperature increase in a model in which horizontal imbalance is compensated geostrophically by a circulation (i.e. by Coriolis forces acting on a horizontal flow, as in high- and low-pressure systems in the Earth’s atmosphere). We find that if temperatures are increased, but vertical expansion is ignored, the changes in travel time as seen in time-distance measurements can be of either sign, depending on the wavenumber and the inter-skip distance. This is because the increased propagation speed is offset in part by the fact that the upper turning point of the waves is higher in a hotter model.
This effect is enhanced if vertical expansion of the perturbed model due to vertical pressure balance is taken into account. We have calculated this in a model in which the entropy has been increased locally (corresponding to an increase in the polytropic constant $`K`$). The results show that the travel times are increased by a temperature enhancement, as long as the horizontal extent of the inhomogeneity is not too large. We found it difficult to obtain time delays for inhomogeneities that have widths of more than about 10 Mm without making unrealistic assumptions. If one takes into account the effect of turbulence onto the sound speed inside the inhomogeneity, one finds that the sign of the time delay can change. This implies that predictions become sensitive to parametrizations of poorly known turbulent flows and therefore less robust (see Sec. II C).
The effects of temperature enhancements have been considered before by Kuhn and Stein (1996), whose conclusions differ from ours. By a numerical convection simulation, these authors calculate the effect of a temperature increase applied at a depth of 2.6 Mm on a part of the lower boundary. This increase causes both a circulation and a vertical expansion of the model above the hotter boundary region. The temperature change as a function of depth in their figure 2 mimics the vertical gradient of the unperturbed model. The vertical shift implied by the figure is about 10 km at the photosphere. This can be compared with a value of roughly 12 km expected from approximate vertical hydrostatic balance for an entropy change as applied by the authors. The authors then perform a ray tracing calculation similar to ours, and find reduced travel times.
The cause of the disagreement can be traced to the treatment by Kuhn and Stein of the layers below 2.6 Mm, i.e. at depths not covered by the numerical simulation. In their ray tracing calculations, they assume higher temperatures, where $`\delta T/T`$ declines linearly from 0.006 at 2.6 Mm to 0.003 at a depth of 50 Mm. The vertical expansion of the model, however, is based only on the 2.6 Mm layer included in the simulation. The expansion is approximately proportional to the depth over which the increased temperatures extend (see Eqn. 12). On the one hand, the convection simulation by Kuhn and Stein demonstrates the vertical expansion effect, but on the other hand their ray tracing calculation underestimates its effect on travel times by a large factor.
Kuhn and Stein do not specify the cause of their assumed temperature enhancement at a depth of 50 Mm, but suggest a source in those layers where the magnetic field of the solar cycle is produced, near the base of the convection zone ($`z200`$ Mm) (see also Kuhn, Libbrecht and Dicke 1988). If this were the case, for a given temperature increase at the surface, the vertical expansion effect would be four times larger than for an assumed depth of 50 Mm. The travel time increase by the vertical expansion would then certainly dominate over the reduction due to the higher sound speed for all time-distance measurements published so far (which reach depths of the order of 20 Mm).
### A Lower temperatures in active regions?
The shorter sound travel times in active regions found by time-distance seismology are consistent with the increase of mode frequencies with magnetic activity found by Woodard et al. (1993) and Dziembowski et al. (2000). As noted by Goldreich et al. (1991), the increased mode frequencies are not consistent with increased temperatures, for the same reason as in our time-distance calculations. Instead, Goldreich et al. suggest that the magnetic field of active regions causes the increase in propagation speeds. For the vertical magnetic fields seen near the surface, there would be no associated vertical expansion of the envelope. While this explanation is consistent with the data available then, it is no longer compatible with recent data. This is because a magnetic change in propagation speed is confined to a thin layer near the surface (unless very large magnetic fields are assumed at a depth of 10–20 Mm). The mode frequency changes measured by Woodard et al. and very accurately by Dziembowski et al. (2000) (with MDI) show that while most of the effect is concentrated near the surface, there are also significant changes at depths of 10 Mm. This would require field strengths of the order 20 kG covering large fractions of the surface.
Instead of getting involved in a discussion about the difficulties that such large unobserved magnetic fields would cause, we suggest here a more radical and simple solution. One could infer that the subsurface temperatures in active regions, in spite of the increased emission at the surface, are in fact reduced.
If the effect of the small scale magnetic fields were an increase in convective efficiency, or some other effect that increases the radiation losses at the surface, then the increased cooling would cause the intergranular downdrafts to be cooler than average (since by this assumption the granules would have lost more heat). These lower temperatures would be carried down with the downdrafts, and cause horizontal average temperatures to be reduced below active regions. The depth dependence of the effect would depend on the details of the rate of spreading of the downflows by entrainment.
An effect that would cause just such an increased cooling assciated with active region fields was proposed by Spruit (1977). There it was shown that the ‘dimples’ in the photosphere caused by the reduced opacity in the magnetic elements allow more radiation to escape. The magnetic elements are effectively small leaks through which more heat escapes than from the normal photosphere. This increased radiation at the same time implies a larger average cooling rate in active regions. The downflows inferred from time-distance helioseismology (Duvall et al. 1998) are consistent with this interpretation, but are hard to understand in models with increased sub-surface temperatures.
|
warning/0004/math-ph0004028.html
|
ar5iv
|
text
|
# On the use of Mellin transform to a class of q-difference-differential equations
(Mar 7, 2000)
## Abstract
We explore the possibility of using the method of classical integral transforms to solve a class of $`q`$-difference-differential equations. The Laplace and the Mellin transform of $`q`$-derivatives are derived. The results show that the Mellin transform of the $`q`$-derivative resembles most closely the corresponding expression in classical analysis, and it could therefore be useful in solving certain $`q`$-difference equations.
Revised version
1. The study of $`q`$-analysis is an old subject, which dates back to the end of the 19th century (-). It has found many applications in such areas as the theory of partitions, combinatorics, exactly solvable models in statistical mechanics, computer algebra, etc . Recent developments in the theory of quantum group has boosted further interests in this old subject .
The subject of $`q`$-analysis concerns mainly the properties of the so-called $`q`$-special functions, which are the extensions of the classical special functions based on a parameter, or the base, $`q`$. The relations among these functions, and the difference equations satisfied by them are among the topics most studied so far. The $`q`$-difference equations involve a new kind of difference operator, the $`q`$-derivative, which can be viewed as a sort of deformation of the ordinary derivative. Solutions of the $`q`$-difference equations in one variable have been well studied in terms of the $`q`$-hypergeometric series (also called the basic hypergeometric series). Partial $`q`$-difference equations and $`q`$-difference-differential equations with more than one variables are generally studied by means of the method of seperation of variables, or by the techniques of Lie symmetry in the literature (,-). The method of integral transforms, which is another powerful technique of solving differential equations in classical analysis, has not been, in our view, explored in $`q`$-analysis. The reason is not hard to understand. The main virtue of the classical integral transforms, particularly the Fourier and the Laplace transform, is to transform a differential equation into an algebraic equation, which can be solved easily. That this is possible is due to the fact that these transforms change the derivatives of a function to something proportional to the transform of the original function. As far as we know, integral tranforms or $`q`$-integral transforms which could transform $`q`$-difference equations into algebraic equations have not been found. It should be mentioned that in fact $`q`$-analogues of Fourier transform, based on the Jackson $`q`$-integral, have been proposed recently . However, in order for the $`q`$-Fourier transform of the $`q`$-derivative of a function $`f(x)`$ to be proportional to the $`q`$-Fourier transform of $`f(x)`$, the function $`f(x)`$ must satisfy very special conditions, such as $`f(q^1)=0=f(q^1)`$ . Hence, while these $`q`$-Fourier transforms may be useful in proving certain identities among the $`q`$-special functions, their use in solving $`q`$-difference equations seems limited.
In this paper we shall explore the possibility of using the method of classical integral transform to solve a class of $`q`$-difference-differential equations. We derive the Laplace and the Mellin transform of $`q`$-derivative, and argue that the Mellin transform, which is not generally employed in solving differential equations in classical analysis, may still be useful in solving certain $`q`$-difference equations.
2. Suppose we want to solve the following $`q`$-diffusion equation
$`D_t^qy(x,t)={\displaystyle \frac{^2}{x^2}}y(x,t)(\mathrm{}<x<\mathrm{},t>0)`$ (1)
subject to the initial condition
$`y(x,0)=f(x).`$ (2)
Here $`D_t^q`$ is the “forward” temporal $`q`$-derivative defined by
$`D_t^qh(t):={\displaystyle \frac{h(q^1t)h(t)}{(1q)t}}.`$ (3)
for any function $`h(x)`$. We assume $`0<q<1`$ in this paper. The function $`f(x)`$ is assumed to vanish as $`x\pm \mathrm{}`$. One may as well use the more common definition of $`q`$-derivative
$`𝒟_t^qh(t):={\displaystyle \frac{h(t)h(qt)}{(1q)t}}.`$ (4)
We shall not employ this definition of the $`q`$-derivative here for reason to be explained later. We note here that $`q`$-difference and $`q`$-difference-differential equations of the diffusion type such as eq.(1) have been considered before (-), but mostly from the point of view of Lie symmetry, or by seperation of variables.
We can remove the partial differential operator in $`x`$ in (1) by a Fourier transform. The question now is to choose an appropriate integral transform to remove the $`q`$-derivative. In view of the positivity of the time variable, the two most natural choices are the Laplace and the Mellin transform.
Let us first derive the expression of the Laplace transform of the $`q`$-derivative. The Laplace transform of a function $`h(t)`$ is defined as $`\overline{h}(s):=\{h(x),s\}=_0^{\mathrm{}}h(t)\mathrm{exp}(st)𝑑t`$. For the $`q`$-derivative of $`h(x)`$, the Laplace transform is
$`\{D_t^qh(t),s\}={\displaystyle \frac{1}{1q}}\left[{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{h(q^1t)}{t}}e^{st}𝑑t{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{h(t)}{t}}e^{st}𝑑t\right].`$ (5)
To proceed we have to use the following relation of the Laplace transform
$`{\displaystyle _s^{\mathrm{}}}\overline{h}(s^{})𝑑s^{}={\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{h(t)}{t}}e^{st}𝑑t,`$ (6)
provided the integral on the r.h.s. of (6) is well-defined. We may apply (6) to (5) directly if $`h(0)=0`$. However, if $`h(0)0`$, the r.h.s. of (6) is not well-defined, and direct application of (6) to (5) leads to incorrect result which does not reduce to the usual expression of the Laplace transform of derivative in the classical limit $`q1^{}`$. In order to recover the classical limit correctly, we find it necessary to regularise (5) in the form
$`{\displaystyle \frac{1}{1q}}\left[{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{h(q^1t)h(0)}{t}}e^{st}𝑑t{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{h(t)h(0)}{t}}e^{st}𝑑t\right].`$ (7)
We may now apply (6) to (7). Making use of
$`\{h(t)h(0),s\}=\overline{h}(s)s^1h(0)`$ (8)
we finally obtained
$`\{D_t^qh(t),s\}={\displaystyle \frac{1}{1q}}{\displaystyle _{sq}^s}\overline{h}(s^{})𝑑s^{}{\displaystyle \frac{\mathrm{ln}q^1}{1q}}h(0).`$ (9)
Eq.(9) reduces to the expression $`s\overline{h}(s)h(0)`$ for the Laplace transform of ordinary derivative as $`q1^{}`$.
If one uses instead the definition (4) for the $`q`$-derivative, the Laplace transform would be
$`\{𝒟_t^qh(t),s\}={\displaystyle \frac{1}{1q}}{\displaystyle _s^{\frac{s}{q}}}\overline{h}(s^{})𝑑s^{}{\displaystyle \frac{\mathrm{ln}q^1}{1q}}h(0).`$ (10)
It is now obvious that the Laplace transform is not useful in solving equations involving $`q`$-derivatives: it transforms such equations into integral equations!
3. We now consider the Mellin transform of a $`q`$-derivative. The Mellin transform is seldom being used in solving differential equations, because it generally transforms differential equations into difference equations instead of the much simpler algebraic equations. Now that the Fourier and the Laplace transform lose their virtues whenever $`q`$-derivatives are present, the Mellin transform is naturally the next one to be looked at. As we shall see below, the Mellin transform still transforms an equation containing $`q`$-derivatives into a difference equation of the transformed function, which is the best thing next to an algebraic equation one could get. Previously, the use of the Mellin transform in $`q`$-analysis is limited to proving various identities among the $`q`$-special functions .
The Mellin transform of a function $`h(t)`$ is defined as $`h^{}(s):=\{h(t),s\}=_0^{\mathrm{}}h(t)t^{s1}𝑑t`$. For $`q`$-derivative defined in (3), we have
$`\{D_t^qh(t),s\}=[s1]_qh^{}(s1).`$ (11)
Here $`[x]_q`$ is the $`q`$-number defined by
$`[x]_q:={\displaystyle \frac{1q^x}{1q}}.`$ (12)
Note that $`[x]_qx`$ as $`q1^{}`$. Hence (11) reduces to the expression $`(s1)h^{}(s1)`$ for the Mellin transform of the ordinary derivative as $`q1^{}`$. Repeated use of (11) leads to
$`\{(D_t^q)^nh(t),s\}=(1)^n[s1]_q[s2]_q\mathrm{}[sn]_qh^{}(sn),n1.`$ (13)
This is the $`q`$-analogue of the corresponding formula in the classical case .
For the definition (4), one has
$`\{𝒟_t^qh(t),s\}`$ $`=`$ $`[1s]_qh^{}(s1)`$ (14)
$`=`$ $`q^{1s}[s1]_qh^{}(s1).`$ (15)
Here an extra factor of $`q`$ appears compared with (11). In order to simplify our presentation, we therefore adopt the definition (3) in this paper. We must, however, mention that all the arguments given below apply equally well to the corresponding cases with $`q`$-derivatives replaced by the definition (4).
4. Let $`Y^{}(\xi ,s)`$ be the transformed function of $`y(x,t)`$ obtained by taking the Mellin transform in $`t`$ and a Fourier transform $`G(\xi ):=_{\mathrm{}}^{\mathrm{}}g(x)\mathrm{exp}(i\xi x)𝑑x`$ in $`x`$. Making these transforms to (1), one obtains
$`[s1]_qY^{}(\xi ,s1)=\xi ^2Y^{}(\xi ,s).`$ (16)
Fortunately solution to this equation can be readily found to be
$`Y^{}(\xi ,s)=A(\xi )\xi ^{2s}\mathrm{\Gamma }_q(s),`$ (17)
where $`A(\xi )`$ is some function of $`\xi `$ only, and $`\mathrm{\Gamma }_q(s)`$ is the q-gamma function defined by
$`\mathrm{\Gamma }_q(s):={\displaystyle \frac{(q;q)_{\mathrm{}}}{(q^s;q)_{\mathrm{}}}}\left(1q\right)^{1s},0<q<1.`$ (18)
$`(a;q)_{\mathrm{}}:={\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}(1aq^k).`$ (19)
$`\mathrm{\Gamma }_q(s)`$ satisfies
$`\underset{q1^{}}{lim}\mathrm{\Gamma }_q(s)`$ $`=`$ $`\mathrm{\Gamma }(s),`$ (20)
$`\mathrm{\Gamma }_q(s+1)`$ $`=`$ $`[s]_q\mathrm{\Gamma }(s),\mathrm{\Gamma }_q(1)=1.`$ (21)
Inverse-Mellin transform of $`\xi ^{2s}\mathrm{\Gamma }_q(s)`$ in (17) is
$`{\displaystyle \frac{1}{2\pi i}}{\displaystyle _i\mathrm{}^i\mathrm{}}\xi ^{2s}\mathrm{\Gamma }_q(s)t^s𝑑s.`$ (22)
The poles of $`\mathrm{\Gamma }_q(s)`$ are $`s=0,1,2,\mathrm{}`$. The residual of $`\mathrm{\Gamma }_q(s)`$ at pole $`s=n`$ ($`n0`$) is :
$`{\displaystyle \frac{(1q)^{n+1}}{(q^n;q)_n\mathrm{ln}q^1}}.`$ (23)
The symbol $`(a;q)_n`$ is the q-shifted factorial:
$`(a;q)_0:=1,n=0,`$ (24)
$`(a;q)_n`$ $`:=`$ $`(1a)(1aq)\mathrm{}(1aq^{n1}),n=1,2\mathrm{}`$ (25)
Hence (22) becomes
$`{\displaystyle \frac{1q}{\mathrm{ln}q^1}}{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\left[(1q)\xi ^2t\right]^n}{(q^n;q)_n}}.`$ (26)
In view of the identity
$`(q^n;q)_n=\left({\displaystyle \frac{1}{q}}\right)^nq^{n(n1)/2}(q;q)_n,`$ (27)
(26) can be expressed as
$`{\displaystyle \frac{1q}{\mathrm{ln}q^1}}{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{q^{n(n1)/2}}{(q;q)_n}}{\displaystyle \frac{\left[q(1q)\xi ^2t\right]^n}{}}`$ (28)
$`=`$ $`{\displaystyle \frac{1q}{\mathrm{ln}q^1}}E_q\left(q(1q)\xi ^2t\right).`$
The function $`E_q(z)`$ (for complex $`z`$) is the q-exponential function defined by
$`E_q(z):={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{q^{n(n1)/2}z^n}{(q;q)_n}}=(z;q)_{\mathrm{}}.`$ (29)
In the limit $`q1^{}`$, eq.(28) tends to the usual exponential function $`\mathrm{exp}(\xi ^2t)`$. Finally, performing an inverse Fourier transform we obtain the solution of the $`q`$-diffusion equation
$`y(x,t)={\displaystyle \frac{1}{\sqrt{2\pi }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}A(\xi )\left\{{\displaystyle \frac{1q}{\mathrm{ln}q^1}}E_q\left(q(1q)\xi ^2x\right)\right\}e^{i\xi x}𝑑\xi .`$ (30)
Setting $`t=0`$ in (30) shows that
$`{\displaystyle \frac{1q}{\mathrm{ln}q^1}}A(\xi )`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2\pi }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}y(x,0)e^{i\xi x}𝑑x`$ (31)
$`=`$ $`{\displaystyle \frac{1}{\sqrt{2\pi }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}f(x)e^{i\xi x}𝑑x`$
$``$ $`F(\xi )`$
is the Fourier transform of $`y(x,0)=f(x)`$. So the final solution of the initial problem is
$`y(x,t)={\displaystyle \frac{1}{\sqrt{2\pi }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}F(\xi )E_q\left(q(1q)\xi ^2t\right)e^{i\xi x}𝑑\xi .`$ (32)
This is the $`q`$-analogue of the solution given in for the corresponding classical case. One can easily check that (32) indeed satisfies (1) by using the following identity
$`D_t^qE_q(\lambda t)={\displaystyle \frac{\lambda }{q(1q)}}E_q(\lambda t).`$ (33)
Let us consider an example. Suppose the initial profile is $`f(x)=\mathrm{exp}(x^2/4b)/\sqrt{2b}`$, ($`b>0`$). Its Fourier transform is $`F(\xi )=\mathrm{exp}(b\xi ^2)`$. Then from (32) and (29), we get
$`y(x,t)=E_q\left(q(1q)t{\displaystyle \frac{d}{db}}\right)f(x).`$ (34)
In the limit $`q1^{}`$, eq.(34) gives the classical solution
$`y(x,t)`$ $`=`$ $`e^{t\frac{d}{db}}\left({\displaystyle \frac{1}{\sqrt{2b}}}e^{\frac{x^2}{4b}}\right)`$ (35)
$`=`$ $`{\displaystyle \frac{1}{\sqrt{2(t+b)}}}e^{\frac{x^2}{4(t+b)}}.`$
5. As another example, let us consider the following wave equation
$`\left(D_t^q\right)^2y(x,t)={\displaystyle \frac{^2}{x^2}}y(x,t)(\mathrm{}<x<\mathrm{},t>0)`$ (36)
with inital conditions
$`y(x,0)=f(x),D_t^qy(x,0)=g(x).`$ (37)
We assume that both $`f(x)`$ and $`g(x)`$ vanish as $`x\pm \mathrm{}`$. In this case the Fourier-Mellin transformed function $`Y^{}(\xi ,s)`$ obeys
$`[s1]_q[s2]_qY^{}(\xi ,s2)=\xi ^2Y^{}(\xi ,s).`$ (38)
The general solution is
$`Y^{}(\xi ,s)=\left[A(\xi )\left(i\xi \right)^s+B(\xi )\left(i\xi \right)^s\right]\mathrm{\Gamma }_q(s),`$ (39)
where $`A(\xi )`$ and $`B(\xi )`$ are some functions of $`\xi `$. Performing the inverse-Mellin transform, we get
$`Y(\xi ,t)={\displaystyle \frac{1q}{\mathrm{ln}q^1}}\left\{A(\xi )E_q\left(iq(1q)\xi t\right)+B(\xi )E_q\left(iq(1q)\xi t\right)\right\}.`$ (40)
Here $`Y(\xi ,t)`$ is the Fourier transform of $`y(x,t)`$ with respect to $`x`$. Now we rewrite (40) in terms of the q-Sine and the q-Cosine function which are defined by
$`\mathrm{Sin}_q(x)={\displaystyle \frac{E_q(ix)E_q(ix)}{2i}},`$ (41)
$`\mathrm{Cos}_q(x)={\displaystyle \frac{E_q(ix)+E_q(ix)}{2}}.`$ (42)
The result is
$`y(\xi ,t)={\displaystyle \frac{1q}{\mathrm{ln}q^1}}\left\{C(\xi )\mathrm{Cos}_q\left(q(1q)\xi t\right)+D(\xi )\mathrm{Sin}_q\left(q(1q)\xi t\right)\right\},`$ (43)
where the functions $`C(\xi )`$ and $`D(\xi )`$ are linear combinations of $`A(\xi )`$ and $`B(\xi )`$. The inverse-Fourier transform of (43) is
$`y(x,t)={\displaystyle \frac{1q}{\sqrt{2\pi }\mathrm{ln}q^1}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}\left\{C(\xi )\mathrm{Cos}_q\left(q(1q)\xi t\right)+D(\xi )\mathrm{Sin}_q\left(q(1q)\xi t\right)\right\}e^{i\xi x}𝑑\xi .`$ (44)
Letting $`t=0`$ in (44), one can check that the function $`C(\xi )`$ is related to the Fourier transform of $`f(x)`$ by
$`F(\xi )={\displaystyle \frac{1q}{\mathrm{ln}q^1}}C(\xi ).`$ (45)
Making use of the following relations, which can be obtained by means of (33):
$`D_t^q\mathrm{Sin}_q(\lambda t)`$ $`=`$ $`{\displaystyle \frac{\lambda }{q(1q)}}\mathrm{Cos}_q(\lambda t),`$ (46)
$`D_t^q\mathrm{Cos}_q(\lambda t)`$ $`=`$ $`{\displaystyle \frac{\lambda }{q(1q)}}\mathrm{Sin}_q(\lambda t),`$ (47)
we can relate $`D(\xi )`$ to the Fourier transform $`G(\xi )`$ of $`g(x)`$ as follows:
$`G(\xi )={\displaystyle \frac{1q}{\mathrm{ln}q^1}}D(\xi )\xi .`$ (48)
With these results, we finally obtain the solution to the initial problem of eq.(36):
$`y(x,t)={\displaystyle \frac{1}{\sqrt{2\pi }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}\left\{F(\xi )\mathrm{Cos}_q\left(q(1q)\xi t\right)+{\displaystyle \frac{G(\xi )}{\xi }}\mathrm{Sin}_q\left(q(1q)\xi t\right)\right\}e^{i\xi x}𝑑\xi .`$ (49)
This solution is the $`q`$-analogue of the solution to the corresponding classical case given in .
6. We now see how the above steps are generalised to the equation:
$`\left(D_t^q\right)^ny(x,t)={\displaystyle \frac{^2}{x^2}}y(x,t)(\mathrm{}<x<\mathrm{},t>0,n2)`$ (50)
with inital conditions
$`y(x,0)=f(x),\left(D_t^q\right)^ky(x,0)=g_k(x),k=1,\mathrm{},n1,`$ (51)
where the functions $`f(x)`$ and $`g_k(x)`$ are assumed to vanish as $`x\pm \mathrm{}`$. The Fourier-Mellin transformed function $`Y^{}(\xi ,s)`$ obeys
$`(1)^n[s1]_q[s2]_q\mathrm{}[sn]_qY^{}(\xi ,sn)=\xi ^2Y^{}(\xi ,s).`$ (52)
The general solution is
$`Y^{}(\xi ,s)=\mathrm{\Gamma }_q(s)\xi ^{\frac{2s}{n}}{\displaystyle \underset{m=0}{\overset{n1}{}}}A_m(\xi )\left[e^{\frac{(2m+1)}{n}\pi i}\right]^s.`$ (53)
where $`A_m(\xi )`$ are some functions of $`\xi `$. We can now perform the inverse Mellin and Fourier transforms to get the final solution, which is given formally as
$`y(x,t)={\displaystyle \frac{1q}{\sqrt{2\pi }\mathrm{ln}q^1}}{\displaystyle \underset{m=0}{\overset{n1}{}}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}A_m(\xi )E_q\left(q(1q)e^{\frac{(2m+1)}{n}\pi i}\xi ^{\frac{2}{n}}t\right)e^{i\xi x}𝑑\xi .`$ (54)
The functions $`A_m(\xi )`$ can then be related to the Fourier transforms of the functions $`f(x)`$ and $`g_k(x)`$ from the initial conditions.
7. To summarise, we show that the Mellin transform of the $`q`$-derivative resembles most closely the corresponding expression in classical analysis, whereas transforms such as the Fourier and the Laplace transform fail in this respect. As such the Mellin transform can be useful in solving certain $`q`$-difference equations. We illustrated this fact with a few examples. However, for the Mellin transform to be really useful, a more complete knowledge of the properties of the $`q`$-special functions under various integral transforms (Fourier, Laplace, Mellin, etc) and their inverses has yet to be attained. What is more desirable is to invent integral transforms or $`q`$-integral transforms that possess the virtue of the Fourier and the Laplace transform in the classical analysis mentioned in the introduction.
Acknowledgment
This work is supported in part by the Republic of China through Grant No. NSC-89-2112-M-032-004.
|
warning/0004/nucl-th0004059.html
|
ar5iv
|
text
|
# Role of deformation in the nonmesonic decay of light hypernuclei
## I Introduction
One of the main issues of nuclear physics is to understand the nucleon-nucleon (NN) interaction. The $`|\mathrm{\Delta }S|=1`$ NN interaction is particularly important in this respect, since the change of strangeness can be used as a signature to study both the parity-conserving (PC) and the parity-violating (PV) amplitudes. This is in clear contrast to the $`\mathrm{\Delta }S=0`$ weak NN interaction, where the weak PC signal is masked by the strong interaction.
Due to the lack of stable $`\mathrm{\Lambda }`$-particle beams, the weak decay of $`\mathrm{\Lambda }`$-hypernuclei has been the only source of information on the weak four-baryon $`|\mathrm{\Delta }S|=1`$ interaction. Single $`\mathrm{\Lambda }`$-hypernuclei are typically produced via either hadronic reactions, as ($`K_{stop}^{}`$,$`\pi ^0`$) or ($`\pi ^+`$,$`K^+`$), or electroproduction mechanisms, as (e,e’ $`K^+`$). These hypernuclei are typically produced in an excited state and reach their ground state by electromagnetic-$`\gamma `$ and/or particle emission. Once they are stable against strong decay, they decay via weak interaction mechanisms which are nonleptonic in nature and violate isospin, parity and strangeness. Since the mesonic decay mode, $`\mathrm{\Lambda }\pi N`$, is Pauli blocked in the nuclear medium, hypernuclei with $`A5`$ predominantly decay through the nonmesonic decay (NMD) mode, $`\mathrm{\Lambda }NNN`$.
In order to learn about the weak $`\mathrm{\Lambda }NNN`$ interaction from the theoretical side, one has to take into account different inputs as accurately as possible. These include the description of nuclear structure, the choice of the strong BB potential model , $`\mathrm{\Delta }I=1/2`$ violations and the importance of the 3N emission channel, $`\mathrm{\Lambda }npnnp`$ . In Refs., a one-meson-exchange (OME) model was applied to calculate the nonmesonic decay observables of the $`p`$-shell $`{}_{\mathrm{\Lambda }}{}^{}{}_{}{}^{11}`$B and $`{}_{\mathrm{\Lambda }}{}^{}{}_{}{}^{12}`$C and the $`s`$-shell $`{}_{\mathrm{\Lambda }}{}^{}{}_{}{}^{5}`$He and $`{}_{\mathrm{\Lambda }}{}^{}{}_{}{}^{3}`$H hypernuclei. We included the virtual exchange of the ground-state pseudoscalar and vector mesons $`\rho `$, $`\eta `$,$`\omega `$, $`K`$ and $`K^{}`$, in addition to the long-ranged pion. Except for the hypertriton, where the hypernuclear wave function was calculated exactly using the Faddeev formalism, the structure of the initial hypernucleus was described in a shell-model framework which assumed spherical configuration. In these calculations, the strong baryon-baryon (BB) interaction was accounted for using the Nijmegen BB potential model. Monopole form factors at each vertex were included in order to regularize the weak potential, while the weak baryon-baryon-meson coupling constants were derived based on SU<sub>w</sub>(6) and soft-meson theorems. The total NMD rate and the asymmetry in the distribution of emitted protons from the decay of polarized hypernuclei were in good agreement with the experimental data. However, the theoretical values for the neutron-to-proton ratio were found to be very small compared to the experimental data. Several attempts have been made to reconcile this discrepancy, but none of them has solved this problem yet.
Our aim in this paper is to investigate how much these observables depend on the deformation of hypernuclei. All previous calculations were performed using the spherical configuration, however, it is well known that many $`p`$-shell nuclei are deformed in the ground state. For instance, the quadrupole deformation parameter extracted from the experimental quadrupole moment is $`\beta _2=`$ 0.65 for <sup>10</sup>B and $`0.71`$ for <sup>11</sup>C. It may be important to take these deformation effects into account in order to describe quantitatively the nonmesonic decay of $`p`$-shell hypernuclei. Deformed hypernuclei can be described using several models such as the $`\alpha `$-cluster model or the deformed self-consistent Hartree-Fock method. In fact, one can also use realistic wave functions obtained by a diagonalization of a shell-model Hamiltonian for p-shell nuclei, as in Ref. . In the present paper, however, in order to perform a systematic study, we use instead the Nilsson model as a simplified Hartree-Fock method.
The paper is organized as follows. In Sec. II, we present the relevant formulae to evaluate the NMD observables in a OME model. In Sec. III, we briefly review the deformed shell model based on the Nilsson model. Sec. IV presents the deformation dependence of the nonmesonic observables for the decay of $`{}_{\mathrm{\Lambda }}{}^{}{}_{}{}^{9}`$Be, whose <sup>8</sup>Be core is known to be largely deformed. Although there is no experimental data for this hypernucleus at present, we choose this system as the simplest non-spherical p-shell hypernucleus and as a representative example of deformed p-shell nuclei. We compare our theoretical predictions with the typical experimental data for other p-shell hypernuclei. Sec. V summarizes the paper.
## II Nonmesonic weak decay in a one-meson-exchange model
Assuming that the initial hypernucleus is at rest, the NMD rate is
$$\mathrm{\Gamma }_{\mathrm{nm}}=\frac{d^3k_1}{(2\pi )^3}\frac{d^3k_2}{(2\pi )^3}\underset{_{\{1\}\{2\}}^{M_I\{R\}}}{}(2\pi )\delta (M_HE_RE_1E_2)\frac{1}{(2J+1)}_{fi}^2,$$
(1)
where $`_{fi}`$ is the hypernuclear transition amplitude. The quantities $`M_H`$, $`E_R`$, $`E_1`$ and $`E_2`$ are the mass of the hypernucleus, the energy of the residual $`(A2)`$-particle system, and the total asymptotic energies of the emitted nucleons, respectively. The integration variables $`\stackrel{}{k}_1`$ and $`\stackrel{}{k}_2`$ are the momenta of the two baryons in the final state. The momentum conserving delta function has been used to integrate over the momentum of the residual nucleus. The sum, together with the factor $`1/(2J+1)`$, indicates an average over the initial hypernucleus spin projections, $`M_I`$, and a sum over all quantum numbers of the residual $`(A2)`$-particle system, $`\{R\}`$, as well as the spin and isospin projections of the exiting particles, $`\{1\}`$ and $`\{2\}`$. In general, one can write the total nonmesonic decay rate as $`\mathrm{\Gamma }_{\mathrm{nm}}=\mathrm{\Gamma }^{\mathrm{\Lambda }NNN}=\mathrm{\Gamma }_\mathrm{n}+\mathrm{\Gamma }_\mathrm{p}`$, where $`\mathrm{\Gamma }_\mathrm{n}`$ ($`\mathrm{\Lambda }nnn`$) stands for the neutron-induced decay and $`\mathrm{\Gamma }_\mathrm{p}`$ ($`\mathrm{\Lambda }pnp`$) for the proton-induced one.
In addition to the total and partial decay rates, we also calculate the intrinsic $`\mathrm{\Lambda }`$ asymmetry parameter. When working with polarized hypernuclei and in combination with coincidence measurements of the decay particles, one can study the angular distribution of particles coming from the $`\mathrm{\Lambda }NNN`$ weak decay. Due to the interference between the PV and PC amplitudes, the distribution of the emitted protons in the weak decay displays an angular asymmetry with respect to the polarization axis. The asymmetry $`𝒜`$, defined by
$$𝒜=P_y\frac{3}{J+1}\frac{Tr(_{fi}S_y_{fi}^{}{}_{}{}^{})}{Tr(_{fi}_{fi}^{}{}_{}{}^{})},$$
(2)
is expressed in terms of the hypernuclear polarization created in the strong production reaction, $`P_y`$, the $`J`$-spin operator along the polarization axis, $`S_y`$, and the total spin of the initial hypernucleus, $`J`$. In Ref. it is shown that the asymmetry follows a simple $`\mathrm{cos}\chi `$ dependence, i.e., $`𝒜=P_yA_p\mathrm{cos}\chi `$, where $`\chi `$ stands for the angle between the direction of the proton and the polarization axis. The hypernuclear asymmetry parameter $`A_p`$ is characteristic of the hypernuclear weak decay process and depends on $`J`$ and the intensity of protons exiting along the quantization axis for the different spin projections of the hypernucleus. At $`\chi =0^{}`$, the asymmetry in the distribution of protons is thus determined by the product $`𝒜=P_yA_p`$. In the following, we assume a weak coupling scheme where the $`\mathrm{\Lambda }`$ hyperon is coupled only to the ground state of the $`(A1)`$-particle core. In this scheme, simple angular momentum algebra relates the hypernuclear polarization $`P_y`$ to the $`\mathrm{\Lambda }`$ polarization $`p_\mathrm{\Lambda }`$, and the hypernuclear asymmetry parameter $`A_p`$ to the intrinsic $`\mathrm{\Lambda }`$ asymetry parameter $`a_\mathrm{\Lambda }`$, such that $`𝒜=p_\mathrm{\Lambda }a_\mathrm{\Lambda }=P_yA_p`$ .
The nonmesonic decay of hypernuclei proceeds through a two-body mechanism. Therefore in order to evaluate the transition amplitude in Eq. (1), one has to decompose the (A-1)-core wave function into a set of states in which a nucleon couples to the residual (A-2)-particle state. This can be done using the Coefficients of Fractional Parentage (CFP), which are defined by
$$|JM,TT_z=\underset{J_R,T_R,j}{}JT\{|J_RT_R,jt[|J_RT_R|jt]_{JM,TT_z},$$
(3)
where $`J_R`$ and $`T_R`$ are the spin and isospin of the residual nucleus. The weak potential responsible for this transition can be obtained by making a nonrelativistic reduction of the free Feynman amplitude depicted in Fig. 1. In Table I we show the strong and weak vertices for pseudoscalar (PS) and vector (V) mesons. $`A,B,\alpha ,\beta `$ and $`ϵ`$ stand for the appropriate baryon-baryon-meson weak coupling constants, while $`g`$ ($`g^\mathrm{V},g^\mathrm{T}`$) represents the strong (vector, tensor) coupling. Details of the derivation of the transition potential can be found in Ref. and here only the final expression will be presented. For pseudoscalar mesons, the potential is
$$V_{ps}(\stackrel{}{q})=G_Fm_\pi ^2\frac{g}{2M}\left(\widehat{A}+\frac{\widehat{B}}{2\overline{M}}\stackrel{}{\sigma }_1\stackrel{}{q}\right)\frac{\stackrel{}{\sigma }_2\stackrel{}{q}}{\stackrel{}{q}^{\mathrm{\hspace{0.33em}2}}+\mu ^2},$$
(4)
where $`G_Fm_{\pi }^{}{}_{}{}^{2}=2.21\times 10^7`$ is the Fermi coupling constant, $`\stackrel{}{q}`$ is the momentum carried by the meson directed towards the strong vertex, $`\mu `$ the meson mass and $`M`$ ($`\overline{M}`$) is the average of the baryon masses at the strong (weak) vertex (the other way around for the exchange of strange mesons). For vector mesons, the potential is
$`V_v(\stackrel{}{q})`$ $`=`$ $`G_Fm_\pi ^2(g^\mathrm{V}\widehat{\alpha }{\displaystyle \frac{(\widehat{\alpha }+\widehat{\beta })(g^\mathrm{V}+g^\mathrm{T})}{4M\overline{M}}}(\stackrel{}{\sigma }_1\times \stackrel{}{q})(\stackrel{}{\sigma }_2\times \stackrel{}{q})`$ (6)
$`\mathrm{i}{\displaystyle \frac{\widehat{\epsilon }(g^\mathrm{V}+g^\mathrm{T})}{2M}}(\stackrel{}{\sigma }_1\times \stackrel{}{\sigma }_2)\stackrel{}{q}){\displaystyle \frac{1}{\stackrel{}{q}^{\mathrm{\hspace{0.33em}2}}+\mu ^2}}.`$
The values of the strong and weak couplings are listed in Table III of Ref. . In Eqs. (4) and (6) the operators $`\widehat{A}`$, $`\widehat{B}`$, $`\widehat{\alpha }`$, $`\widehat{\beta }`$ and $`\widehat{\epsilon }`$ contain, apart from the weak coupling constants, the specific isospin dependence of the potential, which is $`\stackrel{}{\tau }_1\stackrel{}{\tau }_2`$ for the isovector $`\pi `$ and $`\rho `$ mesons, $`\widehat{1}`$ for the isoscalar $`\eta `$ and $`\omega `$ mesons and a combination of both operators for the isodoublet $`K`$ and $`K^{}`$. In order to derive Eqs. (4) and (6) we assumed the validity of the $`\mathrm{\Delta }I=1/2`$ rule, which is known to experimentally dominate the decay of $`\mathrm{\Lambda }`$’s into pions. $`\mathrm{\Delta }I=3/2`$ transitions for vector mesons ($`\rho `$ and $`K^{}`$) are easily accomodated in the formalism and the results we present here account for such $`\mathrm{\Delta }I=1/2`$ violations.
We obtain a regularized potential by including a monopole form factor at each vertex, $`F(\stackrel{}{q}^{\mathrm{\hspace{0.17em}\hspace{0.17em}2}})={\displaystyle \frac{(\mathrm{\Lambda }^2\mu ^2)}{(\mathrm{\Lambda }^2+\stackrel{}{q}^{\mathrm{\hspace{0.17em}\hspace{0.17em}2}})}}`$, where the value of the cut-off, $`\mathrm{\Lambda }`$, different for each meson, is taken from the Jülich hyperon-nucleon interaction. To incorporate the effects of the strong NN interaction, we solve a T-matrix scattering equation in momentum space for the outgoing nucleons using the Nijmegen potential models. For the initial bound two-baryon system we use a spin independent parametrization of the type
$$f(r)=\left(1\mathrm{e}^{r^2/a^2}\right)^n+br^2\mathrm{e}^{r^2/c^2},$$
(7)
with $`a=0.5`$ fm, $`b=0.25`$ fm<sup>-2</sup>, $`c=1.28`$ fm, $`n=2`$. The results obtained with this parametrization lay in between of the ones obtained with a microscopic finite-nucleus $`G`$-matrix calculation using the soft-core and hard-core Nijmegen models.
## III Deformed shell model
As we mentioned in the previous section, we use a weak coupling scheme for the $`\mathrm{\Lambda }`$ hyperon in the initial hypernucleus. To this end, we must describe the ground state of the core nucleus which may be deformed. The Nilsson model provides a simple and convenient framework to describe deformed nuclei, and has been widely used in the literature . Its Hamiltonian consists of an anisotropic harmonic oscillator with the spin-orbit interaction as well as an angular momentum dependent term, which mocks up the deviation of the mean field potential from the harmonic oscillator
$`H`$ $`=`$ $`H_0{\displaystyle \frac{4}{3}}\sqrt{{\displaystyle \frac{\pi }{5}}}\delta m\omega _0^2Y_{20}(\theta ),`$ (8)
$`=`$ $`{\displaystyle \frac{\mathrm{}^2}{2m}}^2+{\displaystyle \frac{1}{2}}m\omega _0^2r^2+C\stackrel{}{l}\stackrel{}{s}+D(\stackrel{}{l}^2\stackrel{}{l}^2_N){\displaystyle \frac{4}{3}}\sqrt{{\displaystyle \frac{\pi }{5}}}\delta m\omega _0^2Y_{20}(\theta ).`$ (9)
Here $`\delta `$ is a deformation parameter, $`\stackrel{}{l}`$ and $`\stackrel{}{s}`$ are the single-particle orbital and the spin angular momenta, and $`C`$ and $`D`$ are adjustable parameters. $`\stackrel{}{l}^2_N=N(N+3)/2`$ is the expectation value of $`\stackrel{}{l}^2`$ averaged over one major shell with quantum number $`N`$. The relation between $`\delta `$ and $`\beta _2`$ is given by
$$\beta _2=\frac{4}{3}\sqrt{\frac{\pi }{5}}\frac{\delta }{12\delta /3}.$$
(10)
Since the Nilsson Hamiltonian (9) violates rotational invariance, the total angular momentum $`\stackrel{}{j}=\stackrel{}{l}+\stackrel{}{s}`$ is not a good quantum number. However, the projection of $`\stackrel{}{j}`$ on to the $`z`$-direction, $`k`$, is conserved and the single-particle levels are characterized by $`k`$ and other quantum numbers. We expand a Nilsson single-particle level, $`\psi _{k(q)}`$, in terms of the eigenfunctions of the spherical harmonic oscillator Hamiltonian $`H_0`$, $`\varphi _{nljk}`$, as
$$\psi _{k(q)}=\underset{nlj}{}x_{nljk}^{(q)}\varphi _{nljk},$$
(11)
where $`q`$ are quantum numbers other than $`k`$. We choose $`x_{nljk}^{(q)}=()^{jk}x_{nljk}^{(q)}`$ so that the eigenvalues of the Nilsson Hamiltonian do not depend on the sign of the projection of total angular momentum . We denote the creation operator of $`\psi _k`$ as $`a_k^{}`$ and that of $`\varphi _{jk}`$ as $`b_{jk}^{}`$. We explicity express only the $`j`$ and $`k`$ quantum numbers to simplify the notation. Intrinsic wave functions, i.e., eigenfunctions of the Nilsson Hamiltonian (9) are given by
$$|\chi _K=a_{k_1}^{}a_{k_2}^{}\mathrm{}a_{k_n}^{}|0=\underset{i=1}{\overset{n}{}}\left(\underset{j}{}x_{jk_i}b_{jk_i}^{}\right)|0,$$
(12)
where the $`K`$ quantum number is the sum over all $`k_i`$. The intrinsic wave function (12) is not an eigenstate of the total angular momentum $`\stackrel{}{J}`$, and thus has to be projected out to a good angular momentum state. This can be achieved by using the projector given by
$$\widehat{P}_{MK}^J=\frac{2J+1}{8\pi ^2}𝑑\mathrm{\Omega }D_{MK}^J(\mathrm{\Omega })\widehat{R}(\mathrm{\Omega }),$$
(13)
where $`\mathrm{\Omega }`$ are Euler angles, and $`D_{MK}^J(\mathrm{\Omega })`$ and $`\widehat{R}(\mathrm{\Omega })`$ are the Wigner $`D`$-function and the rotation operator, respectively.
For systems with a single-Nilsson level, such as <sup>8</sup>Be which we discuss in the next section, the CFP can be analitically obtained . Note that a single Nilsson level can accommodate up to four nucleons, i.e., two protons and two neutrons. For three particle systems ( 2 neutrons and 1 proton, for example), the wave function is given by
$`\mathrm{\Psi }_{JM}`$ $`=`$ $`[N(3)_J]^1\widehat{P}_{M,K=k}^Ja_{k\nu }^{}a_{k\nu }^{}a_{k\pi }^{}|0,`$ (14)
$`=`$ $`[N(3)_J]^1{\displaystyle \underset{j_1,j_2,j_3}{}}{\displaystyle \underset{J_{12}}{}}x_{j_1k}x_{j_2k}x_{j_3k}j_1kj_2k|J_{12}0J_{12}0j_3k|Jk`$ (16)
$`\times \left(\left[a_{j_1\nu }^{}a_{j_2\nu }^{}\right]_{J_{12}}a_{j_3\pi }^{}\right)_{JM}|0,`$
where $`\pi `$ stands for proton and $`\nu `$ for neutron. The normalization factor $`N(3)_J`$ is given by
$$[N(3)_J]^2=\underset{j,J_{12}}{}\delta _{J_{12},even}(x_{jk})^2U(J_{12},k)J_{12}0jk|Jk^2,$$
(17)
where
$$U(J,k)=2\underset{j_1,j_2}{}(x_{j_1k})^2(x_{j_2k})^2j_1kj_2k|J0.$$
(18)
The isospin of this system is 1/2. For four nucleon systems, the wave function reads
$`\mathrm{\Psi }_{JM}`$ $`=`$ $`[N(4)_J]^1\widehat{P}_{M,K=0}^Ja_{k\nu }^{}a_{k\nu }^{}a_{k\pi }^{}a_{k\pi }^{}|0,`$ (19)
$`=`$ $`[N(4)_J]^1{\displaystyle \underset{j_1,j_2}{}}{\displaystyle \underset{j_3,j_4}{}}{\displaystyle \underset{J_{12},J_R}{}}x_{j_1k}x_{j_2k}x_{j_3k}x_{j_4k}j_1kj_2k|J_{12}0J_{12}0j_3k|J_RkJ_Rkj_4k|J0`$ (21)
$`\times \left(\left[\left(a_{j_1\nu }^{}a_{j_2\nu }^{}\right)_{J_{12}}a_{j_3\pi }^{}\right]_{J_R}a_{j_4\pi }^{}\right)_{JM}|0,`$
with the normalization given by
$$[N(4)_J]^2=\underset{J_{12},J_{34}}{}\delta _{J_{12},even}\delta _{J_{34},even}U(J_{12},k)U(J_{34},k)J_{12}0J_{34}0|J0^2.$$
(22)
The isospin of this wave function is 0. Comparing Eqs. (16) and (21), the CFP for the four particle system reads
$$𝒞(j)=JT\{|J_RT_R,jt=\sqrt{2}x_{jk}J_Rkjk|J0\frac{N(3)_{J_R}}{N(4)_J}.$$
(23)
## IV nonmesonic decay of $`{}_{\mathrm{\Lambda }}{}^{}{}_{}{}^{9}`$Be
Let us now apply the deformed shell model of Sec. III to the nonmesonic decay of $`{}_{\mathrm{\Lambda }}{}^{}{}_{}{}^{9}`$Be. The quadrupole moment of the neighbour nucleus <sup>9</sup>Be was measured to be 5.86 $`e`$fm<sup>2</sup>, from which we extract the quadrupole deformation parameter $`\beta _2`$=1.00 using the radius parameter $`r_0`$=1.2 fm. Several theoretical calculations suggest that the core nucleus <sup>8</sup>Be and the $`{}_{\mathrm{\Lambda }}{}^{}{}_{}{}^{9}`$Be hypernucleus also have similar deformation parameters with the same sign . Our interest is to discuss such deformation effects on nonmesonic decay observables.
As is discussed in Sec. II, the use of CFP allows us to write the hypernuclear transition amplitude $`_{fi}`$ in terms of elementary two-body amplitudes. Therefore, our first task is to compute these coefficients for the core nucleus <sup>8</sup>Be. We assume the inert spherical <sup>4</sup>He core and explicitly work with only the four valence nucleons. Diagonalizing the Nilsson Hamiltonian (9), one finds that the lowest Nilsson level for the valence nucleons has $`k=1/2`$ for prolate deformation . We diagonalize the Nilsson Hamiltonian in the $`\mathrm{\Delta }N`$=0 states. Contributions from the $`\mathrm{\Delta }N=2`$ can be neglected unless the deformation is large. The $`k`$=1/2 state is thus
$$|\psi _{k=1/2}=x|\varphi _{p_{3/2,1/2}}+y|\varphi _{p_{1/2,1/2}},$$
(24)
where $`x`$ and $`y`$ are determined by diagonalizing the Nilsson Hamiltonian within this configuration space and depend upon the deformation of <sup>8</sup>Be. Using Eq. (23), the CFP’s are found to be
$$[𝒞(p_{3/2})]^2=\frac{3x^8+3x^6y^2+9x^4y^4}{3x^8+4x^6y^2+18x^4y^4+10y^8}$$
(25)
for the p<sub>3/2</sub> state, and
$$[𝒞(p_{1/2})]^2=\frac{x^6y^2+9x^4y^4+10y^8}{3x^8+4x^6y^2+18x^4y^4+10y^8}$$
(26)
for the p<sub>1/2</sub> state. Note that $`[𝒞(p_{3/2})]^2+[𝒞(p_{1/2})]^2=1`$. In the spherical limit, $`x`$=1 and $`y=0`$, so the CFP become $`[𝒞(p_{3/2})]^2=1`$ and $`[𝒞(p_{1/2})]^2=0`$. The CFP for the deeply bound 1s<sub>1/2</sub> state is just equal to 1 since <sup>4</sup>He is a spin-isospin saturated nucleus.
Our results for the nonmesonic decay rate, $`\mathrm{\Gamma }^{\mathrm{nm}}`$, in units of the free $`\mathrm{\Lambda }`$ decay rate, $`\mathrm{\Gamma }_\mathrm{\Lambda }=3.8\times 10^9s^1`$, the neutron-to-proton ratio, $`\mathrm{\Gamma }_\mathrm{n}/\mathrm{\Gamma }_\mathrm{p}`$, and the $`\mathrm{\Lambda }`$ asymmetry parameter, $`a_\mathrm{\Lambda }`$, are shown in Fig. 2 as a function of the deformation parameter $`\beta _2`$. We use an oscillator length $`b_N`$ of 1.65 fm for nucleons, so that the experimental root mean square radius of <sup>9</sup>Be is reproduced. Following Refs. , the parameters $`C`$ and $`D`$ in the Nilsson Hamiltonian (9) are taken to be $`0.16\mathrm{}\omega _0`$ and 0, respectively. As for the oscillator length $`b_\mathrm{\Lambda }`$ for the 1s<sub>1/2</sub> wave function of the $`\mathrm{\Lambda }`$ hyperon, we estimate it to be 1.5 fm in order to reproduce its binding energy in $`{}_{\mathrm{\Lambda }}{}^{}{}_{}{}^{9}`$Be ($`=6.71\pm 0.04`$ MeV). From the figure, we see that $`\mathrm{\Gamma }^{\mathrm{nm}}/\mathrm{\Gamma }_\mathrm{\Lambda }`$ is a decreasing function of $`\beta _2`$, while $`\mathrm{\Gamma }_\mathrm{n}/\mathrm{\Gamma }_\mathrm{p}`$ and $`a_\mathrm{\Lambda }`$ are increasing functions. As we have already mentioned, the deformation parameter of <sup>8</sup>Be is expected to be close to 1. We notice that the nonmesonic decay observables are altered by about 10% from the spherical limit at $`\beta _2=1`$.
An important question is whether this effect is significant when comparing to the experimental data. We note that the typical experimental uncertainties for nonmesonic decay of p-shell hypernuclei are: 7% – 17.5% for the total decay rate , 46.2% – 84.2% for the neutron-to-proton ratio, and 50% – 1000 % for the asymmetry . These experimental uncertainties are much larger than the theoretical one originating from the deformation effects. Thus we conclude that the spherical approximation gives a good estimate of the nonmesonic decay of p-shell nuclei, at least within the present experimental precision.
Before we close this section, we would like to stress that our conclusion is not altered qualitatively even if more realistic wave functions are used instead of the present schematic ones. For instance, using the shell-model C.F.P. of Cohen and Kurath for the decay of $`{}_{\mathrm{\Lambda }}{}^{}{}_{}{}^{12}`$C, we obtain $`\mathrm{\Gamma }^{\mathrm{nm}}/\mathrm{\Gamma }_\mathrm{\Lambda }=0.76`$ and $`\mathrm{\Gamma }_\mathrm{n}/\mathrm{\Gamma }_\mathrm{p}=0.22`$. Those numbers have to be compared to the spherical limit values of 0.74 and 0.25, respectively. As we see, the amount of deviation of those observables with respect to the spherical limit is of the same order of the one given here, although their behaviour is opposite due partly to the fact that the residual <sup>11</sup>C is oblate while <sup>8</sup>Be is prolate.
## V Summary
We have discussed the role of nuclear structure in the nonmesonic decay of p-shell hypernuclei, especially focusing on the effects of deformation. To this end, we have used the Nilsson model with explicit angular momentum projection. We have studied the nonmesonic decay of $`{}_{\mathrm{\Lambda }}{}^{}{}_{}{}^{9}`$Be as a typical example of deformed p-shell hypernuclei. We have shown that the deformation effects change the total NMD rate, the neutron-to-proton ratio and the $`\mathrm{\Lambda }`$ asymmetry parameter by about 10 % from the spherical limit. Although this value is not negligible, it still is smaller than the present typical experimental uncertainty and smaller than other theoretical uncertainties, e.g., the effects of SU(3) symmetry breaking or $`\mathrm{\Delta }I=1/2`$ violations. This indicates that the existing discrepancies between the experimental and theoretical values of hypernuclear weak decay observables cannot be attributed solely to deviations from the spherical configuration, and still remain an open question. New experiments are urged in order to reduce the large experimental error bars, which prevent any definite conclusion about the reliability of the theoretical models.
Our conclusions may not be the same for heavier hypernuclei such as $`{}_{\mathrm{\Lambda }}{}^{}{}_{}{}^{238}`$U. There are a lot of intruder states in such heavy deformed systems, unlike p-shell nuclei where there is only a few, or maybe zero, intruder states. Therefore, an interesting future work would be to discuss the nonmesonic decay of heavy hypernuclei including the deformation effects. For that purpose, the projected shell model developed in Refs. , which also uses the Nilsson model with angular momentum projection, would provide a powerful tool to describe the structure of deformed hypernuclei.
## Acknowledgements
We thank David Brown and Amour Margarian for useful and illuminating discussions. This work was supported by the U.S. Dept. of Energy under Grant No. DE-FG03-00-ER41132.
|
warning/0004/quant-ph0004022.html
|
ar5iv
|
text
|
# Discrete Moyal-type Representations for a Spin
## 1 Introduction
The idea to represent quantum mechanics of a particle in phase space $`\mathrm{\Gamma }`$ goes back to Wigner . He established a one-to-one correspondence between a quantum state $`|\psi `$ in the particle Hilbert-space $``$ and a real function
$$W_\psi (p,q)=\frac{2}{h}_\mathrm{\Gamma }𝑑x\psi ^{}(q+x)\psi (qx)\mathrm{exp}[2ipx/\mathrm{}].$$
(1)
Its properties suggest an interpretation as a quasi-probability in phase space, the only ‘drawback’ being due to the negative values it may take. A more general framework for phase-space representations of quantum states as well as operators $`\widehat{A}`$ is given by the relation
$$W_{\widehat{A}}(q,p)=\text{ Tr }\left[\widehat{\mathrm{\Delta }}(q,p)\widehat{A}\right],$$
(2)
with an operator kernel
$$\widehat{\mathrm{\Delta }}(q,p)=2\widehat{D}(q,p)\widehat{\mathrm{\Pi }}\widehat{D}^{}(q,p),(q,p)\mathrm{\Gamma }.$$
(3)
Here $`\widehat{\mathrm{\Pi }}:(\widehat{q},\widehat{p})(\widehat{q},\widehat{p})`$ is the unitary, involutive parity operator while $`\widehat{D}(q,p)`$ describes translations in phase space . If the operator $`\widehat{A}`$ is chosen as the density matrix of a pure state, $`\widehat{A}=\widehat{\rho }_\psi =|\psi \psi |`$, Eq. (2) reduces to (1). The kernel $`\widehat{\mathrm{\Delta }}(q,p)`$ can be derived from a set of conditions which a phase-space representation is required to satisfy (cf. below). It is intimately related to the behaviour of a function $`W_A(q,p)`$ under translations mapping the phase space $`\mathrm{\Gamma }`$ onto itself. The map (2) from operators to functions ($`\widehat{A}W_A`$) has an important feature: its inverse, mapping functions to operators ($`W_A\widehat{A}`$), is mediated by the same kernel —in other words, the kernel $`\widehat{\mathrm{\Delta }}(q,p)`$ is self-dual.
For a quantum spin, the symbol associated with an operator is a continuous function defined on a sphere $`𝒮^2`$, which is the phase space of classical spin. Now, instead of translations in planar phase space, it is the group $`SU(2)`$ of rotations which plays a dominant role when the Moyal formalism is set up. As for a particle, the set of Stratonovich-Weyl postulates characterizes the symbols in an elegant way. For clarity, the postulates are now displayed in their familiar form for the continuous symbols:
(S0)𝗅𝗂𝗇𝖾𝖺𝗋𝗂𝗍𝗒:A^WAisalinearonetoonemap,(S1)𝗋𝖾𝖺𝗅𝗂𝗍𝗒:WA(𝐧)=WA(𝐧),(S2)𝗌𝗍𝖺𝗇𝖽𝖺𝗋𝖽𝗂𝗌𝖺𝗍𝗂𝗈𝗇:
=+2s14πS2WA(n)dn Tr [^A],
(S3)𝗍𝗋𝖺𝖼𝗂𝖺𝗅𝗂𝗍𝗒:
=+2s14πS2WA(n)WB(n)dn Tr [^A^B],
(S4)𝖼𝗈𝗏𝖺𝗋𝗂𝖺𝗇𝖼𝖾:WgA=WAg,gSU(2).𝑆0:𝗅𝗂𝗇𝖾𝖺𝗋𝗂𝗍𝗒absentmaps-to^𝐴subscript𝑊𝐴isalinearonetoonemap𝑆1:𝗋𝖾𝖺𝗅𝗂𝗍𝗒absentWA(𝐧)=WA(𝐧),𝑆2:𝗌𝗍𝖺𝗇𝖽𝖺𝗋𝖽𝗂𝗌𝖺𝗍𝗂𝗈𝗇absent
=+2s14πS2WA(n)dn Tr [^A],
𝑆3:𝗍𝗋𝖺𝖼𝗂𝖺𝗅𝗂𝗍𝗒absent
=+2s14πS2WA(n)WB(n)dn Tr [^A^B],
𝑆4:𝖼𝗈𝗏𝖺𝗋𝗂𝖺𝗇𝖼𝖾absentWgA=WAg,gSU(2).\begin{array}[]{lll}{(S0)}&{\sf linearity}:&\widehat{A}\mapsto W_{A}~{}~{}~{}{\rm is~{}a~{}linear~{}one~{}to~{}one~{}map},\\
{(S1)}&{\sf reality}:&\begin{minipage}{199.16928pt} $W_{A^{\dagger}}({\bf n})=W_{A}^{*}({\bf n})\,,$ \end{minipage}\\
{(S2)}&{\sf standardisation}:&\begin{minipage}{28.45274pt} \vspace{0mm}
$${\frac{2s+1}{4\pi}\int_{{\cal S}^{2}}W_{A}({\bf n})d{\bf n}=\mbox{ Tr }[\widehat{A}]\hfill}\,,$$
\end{minipage}\\
{(S3)}&{\sf traciality}:&\begin{minipage}{28.45274pt}
$$\frac{2s+1}{4\pi}\int_{{\cal S}^{2}}W_{A}({\bf n})W_{B}({\bf n})d{\bf n}=\mbox{ Tr }[\widehat{A}\widehat{B}]\,,\hfill$$ \vspace{0mm}
\end{minipage}\\
{(S4)}&{\sf covariance}:&\begin{minipage}{199.16928pt}
$W_{g\cdot A}=W_{A}^{g}\,,\quad g\in SU(2)\,.$
\end{minipage}\end{array}
It is natural to have a linear relation between operators and symbols (S0), while (S1) implies that hermitean operators are represented by real functions. The third condition (S2) maps the identity operator to the constant function on phase space, and traciality (S3) ensures that the correspondence between operators and symbols is invertible. The covariant transformation of the symbols with respect to rotations $`gSU(2)`$ effectively introduces phase-space points as arguments of the symbols. The continuous Moyal representation for a spin compatible with these conditions can be based on a self-dual kernel $`\widehat{\mathrm{\Delta }}(𝐧)`$ (cf. Sect. 2) in analogy to (2).
In order to have a consistent and full-fledged classical formalism it is necessary to introduce a product between symbols which keeps track of the non-commutativity of the underlying operators. This Moyal product , or twisted product, for two operators $`\widehat{A}`$ and $`\widehat{B}`$ expresses the $`W_{AB}`$ of the operator product $`\widehat{A}\widehat{B}`$ in terms the symbols $`W_A`$ and $`W_B`$,
$$W_{AB}(𝐧)=W_A(𝐧)W_B(𝐧)=_{𝒮^2}_{𝒮^2}L(𝐧,𝐦,𝐤)W_A(𝐦)W_B(𝐤)𝑑𝐦𝑑𝐤,$$
(4)
with a function $`L(𝐧,𝐦,𝐤)`$ of three arguments given explicitly in , for example. The $``$ product is known to be associative.
Other continuous representations for a spin do exist, such as the Berezin symbols of spin operators which are the analog of the $`P`$\- and $`Q`$-symbols for a particle. Instead of a single self-dual kernel, the Berezin symbols require however, a pair of two different kernels, dual to each other: one of the kernels maps operators to functions while its dual is needed for the inverse procedure. It will become clear later on that the self-dual and the dual approach correspond to defining orthogonal and non-orthogonal bases, respectively, in the vector space $`𝒜_s`$ of operators acting on the Hilbert space of a spin $`s`$. When slightly modifying the postulates of Stratonovich, they are also compatible with kernels which are not self-dual.
A common feature of these representations is the redundancy of the continuous symbols. When represented by a $`(2s+1)\times (2s+1)`$ matrix, a hermitean operator is fixed by the values of $`(2s+1)^2`$ real parameters. Consequently, the values of the symbols, continuous functions on the sphere, cannot all be independent—in other words, the information contained in a symbol is redundant. The discrete version of $`P`$\- and $`Q`$-symbols for a spin $`s`$, introduced in as a means to reconstruct the quantum state of a spin, allows one to characterize a spin operator $`\widehat{A}`$ by using only the minimal number of parameters. In fact, a discrete symbol can be considered as living on a ‘discretized sphere,’ that is, as a function taking (real) values on a finite set of points on the sphere only. Such a formalism will be called a discrete Moyal-type formalism.
The purpose of the present paper is to develop the discrete Moyal formalism in analogy to the continuous one. In particular, the kernel and its dual defining the discrete symbols will be derived from a set of appropriate Stratonovich-type postulates. Subsequently, the properties of these symbols are studied in detail.
## 2 Continuous representations
#### Continuous self-dual kernel: Wigner symbols
The Stratonovich-Weyl correspondence for a spin $`s`$ is a rule associating with each operator $`\widehat{A}𝒜_s`$ on a Hilbert space $`_s`$ a function $`W_A`$ on the sphere $`𝒮^2`$, called its (Wigner-) symbol. Let us define it in analogy to Eq. (2), by means of a universal operator kernel $`\widehat{\mathrm{\Delta }}(𝐧)`$, which can also be thought of as a field of operators on the sphere. Then, the first requirement (S0) is already satisfied, and the postulates (S1) to (S4) turn into conditions on the kernel:
(C1)𝗋𝖾𝖺𝗅𝗂𝗍𝗒:Δ^(𝐧)=Δ^(𝐧),(C2)𝗌𝗍𝖺𝗇𝖽𝖺𝗋𝖽𝗂𝗌𝖺𝗍𝗂𝗈𝗇:
=+2s14πS2dn^Δ(n)^I,
(C3)𝗍𝗋𝖺𝖼𝗂𝖺𝗅𝗂𝗍𝗒:
=+2s14πS2dn Tr [^Δ(n)^Δ(m)]^Δ(n)^Δ(m),
(C4)𝖼𝗈𝗏𝖺𝗋𝗂𝖺𝗇𝖼𝖾:Δ^(g𝐧)=U^gΔ^(𝐧)U^g,gSU(2).𝐶1:𝗋𝖾𝖺𝗅𝗂𝗍𝗒absentΔ^(𝐧)=Δ^(𝐧),𝐶2:𝗌𝗍𝖺𝗇𝖽𝖺𝗋𝖽𝗂𝗌𝖺𝗍𝗂𝗈𝗇absent
=+2s14πS2dn^Δ(n)^I,
𝐶3:𝗍𝗋𝖺𝖼𝗂𝖺𝗅𝗂𝗍𝗒absent
=+2s14πS2dn Tr [^Δ(n)^Δ(m)]^Δ(n)^Δ(m),
𝐶4:𝖼𝗈𝗏𝖺𝗋𝗂𝖺𝗇𝖼𝖾absentΔ^(g𝐧)=U^gΔ^(𝐧)U^g,gSU(2).\begin{array}[]{lll}{(C1)}&{\sf reality}:&\begin{minipage}{199.16928pt}
${\widehat{\Delta}}^{\dagger}({\bf n})={\widehat{\Delta}}({\bf n})\,,$
\end{minipage}\\
{(C2)}&{\sf standardisation}:&\begin{minipage}{28.45274pt} \vspace{0mm}
$${\frac{2s+1}{4\pi}\int_{{\cal S}^{2}}d{\bf n}\,{\widehat{\Delta}}({\bf n})=\hat{I}\hfill}\,,$$
\end{minipage}\\
{(C3)}&{\sf traciality}:&\begin{minipage}{28.45274pt}
$$\frac{2s+1}{4\pi}\int_{{\cal S}^{2}}d{\bf n}\mbox{ Tr }\left[{\widehat{\Delta}}({\bf n}){\widehat{\Delta}}({\bf m})\right]{\widehat{\Delta}}({\bf n})={\widehat{\Delta}}({\bf m})\,,\hfill$$ \vspace{0mm}
\end{minipage}\\
{(C4)}&{\sf covariance}:&\begin{minipage}{199.16928pt}
${\widehat{\Delta}}(g\cdot{\bf n})={\widehat{U}}_{g}\,{\widehat{\Delta}}({\bf n})\,{\widehat{U}}_{g}^{\dagger}\,,\quad g\in SU(2)\,.$
\end{minipage}\end{array}
where the matrices $`\widehat{U}_g`$ are a unitary $`(2s+1)`$-dimensional irreducible representations of the group $`SU(2)`$.
The existence of a kernel $`\widehat{\mathrm{\Delta }}(𝐧)`$ satisfying (C1-4) has been proven in by explicit construction. The derivation in starts by expanding the kernel in a basis associated with the eigenstates of the operator $`\widehat{𝐬}𝐧_z`$,
$$\widehat{\mathrm{\Delta }}(𝐧)=\underset{m,m^{}=s}{\overset{s}{}}Z_{mm^{}}(𝐧)|m,𝐧_zm^{},𝐧_z|,$$
(5)
with unknown coefficients $`Z_{mm^{}}(𝐧)`$. It follows from (C1-4) that one must have
$$Z_{mm^{}}^s(𝐧)=\frac{\sqrt{4\pi }}{2s+1}\underset{l=0}{\overset{2s}{}}\epsilon _l\sqrt{2l+1}\begin{array}{ccc}s& l& s\\ m& m^{}m& m^{}\end{array}Y_{l,m^{}m}(𝐧),$$
(6)
where $`\epsilon _0=1`$ and $`\epsilon _l=\pm 1,l=1,\mathrm{},2s`$, and the definition of Clebsch-Gordan coefficient given in is used. Consequently, there are $`2^{2s}`$ different kernels which define a Stratonovich-Weyl correspondence rule.
A new and simple derivation of the kernel $`\widehat{\mathrm{\Delta }}(𝐧)`$, independent of the argument given in , is presented now which has two important advantages. On the one hand, it will provide a form of the kernel similar to that one of a particle (3), which is interesting from a conceptual point of view. On the other hand, it will be possible to transfer this approach to a large extent to the case of the discrete Moyal formalism.
Expand the kernel in the eigenbasis of the operator $`\widehat{𝐬}𝐧`$,
$$\widehat{\mathrm{\Delta }}(𝐧)=\underset{m,m^{}=s}{\overset{s}{}}\mathrm{\Delta }_{mm^{}}(𝐧)|m,𝐧m^{},𝐧|,$$
(7)
where the expansion coefficients $`\mathrm{\Delta }_{mm^{}}`$ are unknown so far. According to the reality condition (C1) they must satisfy $`\mathrm{\Delta }_{mm^{}}(𝐧)=\mathrm{\Delta }_{m^{}m}^{}(𝐧)`$. In a first step, the numbers $`\mathrm{\Delta }_{mm^{}}(𝐧)`$ are shown not to depend on the label $`𝐧`$. Consider the transformation of $`\widehat{\mathrm{\Delta }}(𝐧)`$ under a rotation $`g`$. According to (C4) one must have
$`{\displaystyle \underset{m,m^{}=s}{\overset{s}{}}}`$ $`\mathrm{\Delta }_{mm^{}}(g𝐧)|m,g𝐧m^{},g𝐧|=`$ (8)
$`=`$ $`\widehat{U}_g\widehat{\mathrm{\Delta }}(𝐧)\widehat{U}_g^{}={\displaystyle \underset{m,m^{}=s}{\overset{s}{}}}\mathrm{\Delta }_{mm^{}}(𝐧)|m,R_g𝐧m^{},R_g𝐧|,`$
where $`\widehat{U}_g|m,𝐧=|m,R_g𝐧=|m,g𝐧`$ with a rotation matrix $`R_gSO(3)`$ representing $`gSU(2)`$ in $`IR^3`$. Consequently, one must have
$$\mathrm{\Delta }_{mm^{}}(g𝐧)=\mathrm{\Delta }_{mm^{}}(𝐧),$$
(9)
which is only possible if $`\mathrm{\Delta }_{mm^{}}`$ does not depend on $`𝐧`$. Consider next a rotation $`g(𝐧)`$ about the axis $`𝐧`$ by an angle $`\phi [0,2\pi )`$, represented by the unitary $`\widehat{U}_{g(𝐧)}=\mathrm{exp}(i𝐧\widehat{𝐬}\phi )`$. The left-hand-side of (C4) is invariant under this transformation while the right-hand-side transforms:
$$\widehat{\mathrm{\Delta }}(R_{g(𝐧)}𝐧)=\widehat{\mathrm{\Delta }}(𝐧)=\underset{m,m^{}=s}{\overset{s}{}}\mathrm{\Delta }_{mm^{}}\mathrm{exp}[i(mm^{})\phi ]|m,𝐧m^{},𝐧|,$$
(10)
which is possible only if
$$\mathrm{\Delta }_{mm^{}}=\mathrm{\Delta }(m)\delta _{mm^{}}.$$
(11)
Therefore, covariance of the kernel under elements of $`SU(2)`$ requires it to be diagonal in the basis associated with the direction $`𝐧`$,
$$\widehat{\mathrm{\Delta }}(𝐧)=\underset{m=s}{\overset{s}{}}\mathrm{\Delta }(m)|m,𝐧m,𝐧|.$$
(12)
Next, the condition of traciality will be exploited. Upon rewriting (C3) in the form
$$_{𝒮^2}𝑑𝐧\delta _s(𝐦,𝐧)\widehat{\mathrm{\Delta }}(𝐧)=\widehat{\mathrm{\Delta }}(𝐦),$$
(13)
the function $`\delta _s(𝐦,𝐧)(2s+1)\text{ Tr }[\widehat{\mathrm{\Delta }}(𝐦)\widehat{\mathrm{\Delta }}(𝐧)]/(4\pi )`$ is seen to be the reproducing kernel for a certain subset of $`(2s+1)^2`$ functions on the sphere . In other words, $`\delta _s(𝐦,𝐧)`$ acts in this space as a delta-function with respect to integration over $`𝒮^2`$, and for spin $`s`$, it reads explicitly
$$\delta _s(𝐦,𝐧)=\underset{l=0}{\overset{2s}{}}\underset{m=l}{\overset{l}{}}Y_{lm}(𝐦)Y_{lm}^{}(𝐧)=\underset{l=0}{\overset{2s}{}}\frac{2l+1}{4\pi }P_l(𝐦𝐧).$$
(14)
Here the addition theorem for spherical harmonics $`Y_{lm}(𝐧),l=0,\mathrm{},2s,lml`$, has been used to express the sum over $`m`$ in terms of Legendre polynomials $`P_l(x),1x1`$. Upon choosing $`𝐦𝐧_z`$ and with $`𝐧_z𝐧=\mathrm{cos}\theta `$, the condition (C3) becomes
$$\text{ Tr }\left[\widehat{\mathrm{\Delta }}(𝐧_z)\widehat{\mathrm{\Delta }}(𝐧)\right]=\underset{l=0}{\overset{2s}{}}\frac{2l+1}{2s+1}P_l(\mathrm{cos}\theta ).$$
(15)
Use now the expansion (12) of the kernel on the left-hand-side as well as the identity
$$|m,𝐧_z|m^{},𝐧|^2=\underset{l=0}{\overset{2s}{}}\frac{2l+1}{2s+1}\begin{array}{ccc}s& l& s\\ m& 0& m\end{array}\begin{array}{ccc}s& l& s\\ m^{}& 0& m^{}\end{array}P_l(\mathrm{cos}\theta ),$$
leading to
$$\text{ Tr }\left[\widehat{\mathrm{\Delta }}(𝐧_z)\widehat{\mathrm{\Delta }}(𝐧)\right]=\underset{l=0}{\overset{2s}{}}\left(\underset{m=s}{\overset{s}{}}\mathrm{\Delta }(m)\begin{array}{ccc}s& l& s\\ m& 0& m\end{array}\right)^2\frac{2l+1}{2s+1}P_l(\mathrm{cos}\theta ).$$
Compare now the coefficients of the Legendre polynomials $`P_l(\mathrm{cos}\theta )`$ with those in Eq. (15). This leads to $`(2s+1)`$ conditions
$$\underset{m=s}{\overset{s}{}}\mathrm{\Delta }(m)\begin{array}{ccc}s& l& s\\ m& 0& m\end{array}=\epsilon _l,\epsilon _l=\pm 1,l=0,\mathrm{},2s.$$
(16)
These equations can be solved for $`\mathrm{\Delta }(m)`$ by means of an orthogonality relation for Clebsch-Gordan coefficients ,
$$\mathrm{\Delta }(m)=\underset{l=0}{\overset{2s}{}}\epsilon _l\frac{2l+1}{2s+1}\begin{array}{ccc}s& l& s\\ m& 0& m\end{array}.$$
(17)
Thus, the self-dual kernel for the continuous Moyal formalism is given by
$$\widehat{\mathrm{\Delta }}(𝐧)=\underset{m=s}{\overset{s}{}}\underset{l=0}{\overset{2s}{}}\epsilon _l\frac{2l+1}{2s+1}\begin{array}{ccc}s& l& s\\ m& 0& m\end{array}|m,𝐧m,𝐧|,$$
(18)
Out of these $`2^{2s+1}`$ distinct solutions only $`2^{2s}`$ are compatible with the condition of standardization (C2) which has not been used until now. This condition imposes
$$\underset{m=s}{\overset{s}{}}\mathrm{\Delta }(m)=1,$$
(19)
being satisfied if and only if $`\epsilon _0=+1`$.
The set of solutions (18) coincides indeed with those found in . The easiest way to see this is to calculate the matrix elements of the kernel $`\widehat{\mathrm{\Delta }}(𝐧)`$ in (18) with respect to the standard basis $`|m,𝐧_z`$. One reproduces the coefficients of the expansion (6): $`m,𝐧_z|\widehat{\mathrm{\Delta }}(𝐧)|m^{},𝐧_z=Z_{mm^{}}(𝐧)`$.
The expansion (18) is interesting from a conceptual point of view. It allows one to interpret physically the kernel $`\widehat{\mathrm{\Delta }}(𝐧)`$ in analogy with the kernel for a particle given in (3) by writing
$$\widehat{\mathrm{\Delta }}(𝐧)=\widehat{U}_𝐧\widehat{\mathrm{\Delta }}(𝐧_z)\widehat{U}_𝐧^{},$$
(20)
where $`\widehat{U}_𝐧`$ represents a rotation which maps the vector $`𝐧_z`$ on $`𝐧`$. Imagine now to contract the group $`SU(2)`$ to the Heisenberg-Weyl group. It is known this procedure turns rotations $`\widehat{U}_g`$ into translations $`\widehat{D}(q,p)`$ . As shown in , the operator $`\widehat{\mathrm{\Delta }}(𝐧_z)`$ contracts in the following way,
$$\widehat{\mathrm{\Delta }}(𝐧_z)2\widehat{\mathrm{\Pi }},$$
(21)
if $`\epsilon _l=+1,l=0,\mathrm{},2s`$. Therefore, the operator $`\widehat{\mathrm{\Delta }}(𝐧_z)`$ plays the role of parity for a spin which is by no means immediately obvious when looking at it.
Finally, we would like to point out that the integral kernel $`L`$, defining the $``$ product of symbols (4), has a simple expression in terms of Wigner kernels:
$$L(𝐧,𝐦,𝐤)=\left(\frac{2s+1}{4\pi }\right)^2\text{ Tr }\left[\widehat{\mathrm{\Delta }}(𝐧)\widehat{\mathrm{\Delta }}(𝐦)\widehat{\mathrm{\Delta }}(𝐤)\right].$$
(22)
### Continuous dual kernels: Berezin symbols
Wigner symbols of spin operators are calculated by means of a kernel which is its own dual. Other phase-space representations are known which do not exhibit this ‘symmetry’ between an operator and its symbol. $`P`$\- and $`Q`$-symbols for a particle are familiar examples which have their analog in the ‘Berezin’ symbols for a spin. It will be shown now that these symbolic representations also have a simple description in terms of kernels satisfying a modified set of Stratonovich-Weyl postulates. The conditions (C1-4) must be relaxed slightly in order to allow for a pair of dual kernels.
The required generalization is easily understood in terms of linear algebra. The ensemble of all operators, that is, the self-dual kernel is nothing but a an (overcomplete) set of vectors spanning the linear space $`𝒜_s`$ of operators on the Hilbert space of the spin $`s`$. As the traciality (C3) indicates, this family of vectors is ‘orthogonal’ with respect to integration over the sphere as a scalar product. Each operator $`\widehat{A}`$ can be written as a linear combination of the elements of the kernel with its Wigner symbol as expansion coefficients. More precisely, the expansion coefficients $`W_A(𝐧)`$ with respect to the basis $`\widehat{\mathrm{\Delta }}(𝐧)`$ are given by the ‘scalar product’ of $`\widehat{A}`$ with the same basis vector as shown, for example, in Eq. (2). The essential point now is, that there are also non-orthogonal bases of the same space. Given a non-orthogonal basis, denoted by $`\widehat{\mathrm{\Delta }}_𝐧`$, its dual basis $`\widehat{\mathrm{\Delta }}^𝐧`$ is uniquely determined through the scalar product. Furthermore, the dual basis also spans the original space which implies that now there will be two different expansions of one operator $`\widehat{A}`$ defining a symbol $`A_𝐧`$ and its dual $`A^𝐧`$. Consequently, both kernels and symbols now come in pairs. The familiar $`P`$\- and $`Q`$-symbols—or Berezin symbols —will turn out to be related in this precisely way.
Non-orthogonal bases are allowed in the present framework if, first of all, traciality (C3) is relaxed to
$$(C^{}3)\mathrm{𝗍𝗋𝖺𝖼𝗂𝖺𝗅𝗂𝗍𝗒}:\frac{2s+1}{4\pi }_{𝒮^2}d𝐧\text{ Tr }\left[\widehat{\mathrm{\Delta }}_𝐦\widehat{\mathrm{\Delta }}^𝐧\right]\widehat{\mathrm{\Delta }}_𝐧=\widehat{\mathrm{\Delta }}_𝐦.$$
(23)
The kernel and its dual are both real in analogy to (C1). Explicitly, the symbols and their duals are given by
$$A_𝐧=\text{ Tr }\left[\widehat{A}\widehat{\mathrm{\Delta }}_𝐧\right],A^𝐧=\text{ Tr }\left[\widehat{A}\widehat{\mathrm{\Delta }}^𝐧\right].$$
(24)
Furthermore, one is free to normalize one of the kernels, $`\widehat{\mathrm{\Delta }}_𝐧`$, say, in analogy to (C2), and one requires it to transform covariantly (C4).
It is possible as before to derive the explicit form of the kernels by a reasoning in analogy to above. The general ansatz for both $`\widehat{\mathrm{\Delta }}_𝐧`$ and $`\widehat{\mathrm{\Delta }}^𝐧`$ in the basis referring to the axis $`𝐧`$ as in (7) is again reduced to diagonal form by exploiting their behaviour under rotations:
$$\widehat{\mathrm{\Delta }}_𝐧=\underset{m=s}{\overset{s}{}}\mathrm{\Delta }_m|m,𝐧m,𝐧|,\widehat{\mathrm{\Delta }}^𝐧=\underset{m=s}{\overset{s}{}}\mathrm{\Delta }^m|m,𝐧m,𝐧|,$$
(25)
with two sets of numbers $`\mathrm{\Delta }_m`$ and $`\mathrm{\Delta }^m`$, which do not depend on $`𝐧`$. It is necessary that the trace of these two operators with labels $`𝐦𝐧_z`$ and $`𝐧`$, say, equals the reproducing kernel with respect to integration over the sphere, that is, instead of (15) one needs to have
$$\text{ Tr }\left[\widehat{\mathrm{\Delta }}_{𝐧_z}\widehat{\mathrm{\Delta }}^𝐧\right]=\underset{l=0}{\overset{2s}{}}\frac{2l+1}{2s+1}P_l(\mathrm{cos}\theta ),$$
(26)
where $`\mathrm{cos}\theta 𝐧_z𝐧`$. This leads to the conditions
$$\left[\underset{m=s}{\overset{s}{}}\mathrm{\Delta }_m\begin{array}{ccc}s& l& s\\ m& 0& m\end{array}\right]\left[\underset{m=s}{\overset{s}{}}\mathrm{\Delta }^m\begin{array}{ccc}s& l& s\\ m& 0& m\end{array}\right]=1,$$
(27)
with $`l=0,\mathrm{},2s`$. The ensemble of solutions is parameterized by $`(2s+1)`$ non-zero real numbers $`\gamma _l`$:
$$\underset{m=s}{\overset{s}{}}\mathrm{\Delta }_m\begin{array}{ccc}s& l& s\\ m& 0& m\end{array}=\left[\underset{m=s}{\overset{s}{}}\mathrm{\Delta }^m\begin{array}{ccc}s& l& s\\ m& 0& m\end{array}\right]^1=\gamma _l,$$
(28)
Solving for the expansion coefficients, one obtains
$`\mathrm{\Delta }_m`$ $`=`$ $`{\displaystyle \underset{l=0}{\overset{2s}{}}}\gamma _l{\displaystyle \frac{2l+1}{2s+1}}\begin{array}{ccc}s& l& s\\ m& 0& m\end{array},`$ (31)
$`\mathrm{\Delta }^m`$ $`=`$ $`{\displaystyle \underset{l=0}{\overset{2s}{}}}\gamma _l^1{\displaystyle \frac{2l+1}{2s+1}}\begin{array}{ccc}s& l& s\\ m& 0& m\end{array}.`$ (34)
As in the self-dual case, the standardisation implies that $`\gamma _0=+1`$. This class of solutions for kernels which are not their own dual has been obtained in by an entirely different approach. Self-dual kernels are a small subset: they require $`\mathrm{\Delta }_m\mathrm{\Delta }^m`$, which is $`\gamma _l=\gamma _l^1`$ or $`\gamma _l=\pm 1`$ in agreement with (18). Each set of numbers $`\gamma _l`$ defines a consistent phase-space representation of a quantum mechanical spin.
Consider the particular case $`\mathrm{\Delta }_m=\delta _{ms}`$ resulting from
$$\gamma _l=\begin{array}{ccc}s& l& s\\ s& 0& s\end{array}.$$
(35)
The associated kernels read
$`\widehat{\mathrm{\Delta }}_𝐧`$ $`=`$ $`|s,𝐧s,𝐧||𝐧𝐧|,`$ (36)
$`\widehat{\mathrm{\Delta }}^𝐧`$ $`=`$ $`{\displaystyle \underset{m=s}{\overset{s}{}}}{\displaystyle \underset{l=0}{\overset{2s}{}}}{\displaystyle \frac{2l+1}{2s+1}}\begin{array}{ccc}s& l& s\\ s& 0& s\end{array}^1\begin{array}{ccc}s& l& s\\ m& 0& m\end{array}|m,𝐧m,𝐧|.,`$ (41)
This choice has the advantage that one of the two symbols, reducing to the expectation value of an operator $`\widehat{A}`$ in coherent states , is particularly simple. It turns out to be just the $`Q`$-symbol, $`Q_A(𝐧)=𝐧|\widehat{A}|𝐧`$, that is, its expectation in a spin-coherent state. At the same time, one falls back on a familiar expression for the dual symbol which turns out to be the $`P`$-symbol for $`\widehat{A}`$, defined by an expansion in terms of a linear combination of operators projecting on coherent states,
$$\widehat{A}=\frac{(2s+1)}{4\pi }_{𝒮^2}P_A(𝐧)|𝐧𝐧|𝑑𝐧.$$
(42)
In the present notation one simply has in view of (24) that
$$Q_A(𝐧)A_𝐧=\text{ Tr }[\widehat{A}\widehat{\mathrm{\Delta }}_𝐧],P_A(𝐧)A^𝐧=\text{ Tr }[\widehat{A}\widehat{\mathrm{\Delta }}^𝐧],$$
(43)
so that (42) reads
$$\widehat{A}=\frac{(2s+1)}{4\pi }_{𝒮^2}𝑑𝐧\text{ Tr }[\widehat{A}\widehat{\mathrm{\Delta }}^𝐧]\widehat{\mathrm{\Delta }}_𝐧.$$
(44)
It is obvious now that one has (cf. )
$$\text{ Tr }[\widehat{A}\widehat{B}]=\frac{(2s+1)}{4\pi }_{𝒮^2}𝑑𝐧P_A(𝐧)Q_B(𝐧)=\frac{(2s+1)}{4\pi }_{𝒮^2}𝑑𝐧A^𝐧B_𝐧.$$
(45)
Finally, it is interesting to calculate the $`Q`$\- and $`P`$-symbols of the self-dual kernel $`\widehat{\mathrm{\Delta }}(𝐧)`$ as well as the pair of dual kernels $`\widehat{\mathrm{\Delta }}^𝐧`$ and $`\widehat{\mathrm{\Delta }}_𝐧`$ using the short-hand
$$𝖸_{lm}(𝐧,𝐦)=Y_{lm}^{}(𝐧)Y_{lm}(𝐦),$$
(46)
so that the reproducing kernel is given by
$$\delta _s(𝐧,𝐦)=\underset{ml}{}𝖸_{lm}(𝐧,𝐦).$$
(47)
| | $`\text{ Tr }[\widehat{\mathrm{\Delta }}_𝐦]`$ | $`\text{ Tr }[\widehat{\mathrm{\Delta }}^𝐦]`$ |
| --- | --- | --- |
| $`\widehat{\mathrm{\Delta }}_𝐧`$ | $`\frac{4\pi }{2s+1}_{ml}\begin{array}{ccc}s& l& s\\ m& 0& m\end{array}^2𝖸_{lm}(𝐧,𝐦)`$ | $`\delta _s(𝐧,𝐦)`$ |
| $`\widehat{\mathrm{\Delta }}^𝐧`$ | $`\delta _s(𝐧,𝐦)`$ | $`_{ml}\begin{array}{ccc}s& l& s\\ m& 0& m\end{array}^2𝖸_{lm}(𝐧,𝐦)`$ |
| $`\widehat{\mathrm{\Delta }}(𝐧)`$ | $`_{ml}\begin{array}{ccc}s& l& s\\ m& 0& m\end{array}𝖸_{lm}(𝐧,𝐦)`$ | $`_{ml}\begin{array}{ccc}s& l& s\\ m& 0& m\end{array}^1𝖸_{lm}(𝐧,𝐦)`$ |
Note that the entries of last row, the $`Q`$\- and $`P`$-symbols of the self-dual kernel $`\widehat{\mathrm{\Delta }}(𝐧)`$, do simultaneously provide the Wigner symbols of the dual kernels $`\widehat{\mathrm{\Delta }}_𝐦`$ and $`\widehat{\mathrm{\Delta }}^𝐦`$.
## 3 Discrete Moyal-type representations
A particular feature of the kernels discussed so far is their redundancy: the linear space of hermitean operators for a spin $`s`$ has dimension $`(2s+1)^2`$ while the kernels consist of a continuously labeled set of basis vectors. In other words, there are at most $`N_s=(2s+1)^2`$ linearly independent operators among all $`\widehat{\mathrm{\Delta }}(𝐧)`$, $`𝐧𝒮^2`$. In this section discrete kernels will be introduced, denoted by $`\widehat{\mathrm{\Delta }}_\nu `$, $`\nu =1,\mathrm{},N_s`$. No linear relations must exist between the operators $`\widehat{\mathrm{\Delta }}_\nu `$ which constitute the kernel, that is, they are a basis of $`𝒜_s`$ in the strict sense. It is natural to expect that a subset of precisely $`N_s`$ operators $`\widehat{\mathrm{\Delta }}(𝐧_\nu ),`$ $`\nu =1,\mathrm{},N_s`$ will give rise to a discrete kernel. Therefore, evaluating a continuous symbol of an operator $`\widehat{A}`$ at $`N_s`$ points $`𝐧_\nu `$ of the sphere $`𝒮^2`$, provides a promising candidate for a discrete symbol, i.e. the set $`A_\nu A_{𝐧_\nu }`$, $`\nu =1,\mathrm{},N_s`$. For brevity, $`N_s`$ points on $`𝒮^2`$ are called a constellation.
As before, one might expect orthogonal and non-orthogonal kernels to exist. It turns out, however, that an appropriately modified set of Stratonovich-Weyl postulates covering discrete kernels does not allow for orthogonal ones. Therefore we start immediately by deriving the discrete non-orthogonal kernels coming as before in combination with a dual.
### Discrete dual kernels
By analogy with the continuous representation of the preceding section, one modifies the Stratonovich-Weyl postulates in the following way (throughout the index $`\nu `$ takes all the values from $`1`$ to $`N_s`$):
(D0)𝗅𝗂𝗇𝖾𝖺𝗋𝗂𝗍𝗒:A^Aν is a linear map,(D1)𝗋𝖾𝖺𝗅𝗂𝗍𝗒:Δ^ν=Δ^ν,ν=1,,Ns,(D2)𝗌𝗍𝖺𝗇𝖽𝖺𝗋𝖽𝗂𝗌𝖺𝗍𝗂𝗈𝗇:
=1+2s1=ν1Ns^Δν^I,
(D3)𝗍𝗋𝖺𝖼𝗂𝖺𝗅𝗂𝗍𝗒:
=^Δν1+2s1=μ1Ns Tr [^Δν^Δμ]^Δμ
(D4)𝖼𝗈𝗏𝖺𝗋𝗂𝖺𝗇𝖼𝖾:Δ^gν=U^gΔ^νU^g,gSU(2).𝐷0:𝗅𝗂𝗇𝖾𝖺𝗋𝗂𝗍𝗒absentmaps-to^𝐴subscript𝐴𝜈 is a linear map𝐷1:𝗋𝖾𝖺𝗅𝗂𝗍𝗒absentΔ^ν=Δ^ν,ν=1,,Ns,𝐷2:𝗌𝗍𝖺𝗇𝖽𝖺𝗋𝖽𝗂𝗌𝖺𝗍𝗂𝗈𝗇absent
=1+2s1=ν1Ns^Δν^I,
𝐷3:𝗍𝗋𝖺𝖼𝗂𝖺𝗅𝗂𝗍𝗒absent
=^Δν1+2s1=μ1Ns Tr [^Δν^Δμ]^Δμ
𝐷4:𝖼𝗈𝗏𝖺𝗋𝗂𝖺𝗇𝖼𝖾absentΔ^gν=U^gΔ^νU^g,gSU(2).\begin{array}[]{lll}{(D0)}&{\sf linearity:}&\widehat{A}\mapsto A_{\nu}\qquad\mbox{ is a linear map}\,,\\
{(D1)}&{\sf reality:}&\begin{minipage}{199.16928pt}
$\widehat{\Delta}^{\dagger}_{\nu}=\widehat{\Delta}_{\nu},~{}~{}\nu=1,\ldots,N_{s}\,,$
\end{minipage}\\
{(D2)}&{\sf standardisation:}&\begin{minipage}{28.45274pt} \vspace{0mm}
$$\frac{1}{2s+1}\sum_{\nu=1}^{N_{s}}\widehat{\Delta}^{\nu}=\hat{I}\,,$$
\end{minipage}\\
{(D3)}&{\sf traciality:}&\begin{minipage}{28.45274pt}
$$\widehat{\Delta}_{\nu}=\frac{1}{2s+1}\sum_{\mu=1}^{N_{s}}\mbox{ Tr }\left[\widehat{\Delta}_{\nu}\widehat{\Delta}^{\mu}\right]\widehat{\Delta}_{\mu}$$
\vspace{0mm}
\end{minipage}\\
{(D4)}&{\sf covariance:}&\begin{minipage}{199.16928pt}
${\widehat{\Delta}}_{g\cdot\nu}={\widehat{U}}_{g}\,{\widehat{\Delta}}_{\nu}\,{\widehat{U}}_{g}^{\dagger}\,,\quad g\in SU(2)\,.$
\end{minipage}\end{array}
Let us briefly comment on these conditions. Linearity is automatically satisfied if discrete symbols are defined via kernels, that is, $`A_\nu =\text{ Tr }[\widehat{A}\widehat{\mathrm{\Delta }}_\nu ]`$. The second condition, reality, is obvious, and in (D2) the kernel $`\widehat{\mathrm{\Delta }}^\nu `$ dual to $`\widehat{\mathrm{\Delta }}_\nu `$ is standardized. The duality between $`\widehat{\mathrm{\Delta }}_\nu `$ and $`\widehat{\mathrm{\Delta }}^\nu `$ is made precise by the condition of traciality since (D3) only holds if one has
$$\frac{1}{2s+1}\text{ Tr }\left[\widehat{\mathrm{\Delta }}_\nu \widehat{\mathrm{\Delta }}^\mu \right]=\delta _\nu ^\mu ,\nu ,\mu =1,\mathrm{},N_s,$$
(48)
which, upon considering the trace as a scalar product, is precisely the condition defining the dual of a given basis. As a matter of fact, if $`\{\widehat{\mathrm{\Delta }}_\nu \}`$, $`\nu =1,\mathrm{},N_s`$, is a basis, its unique dual is guaranteed to exist. Finally, covariance under rotations $`gSU(2)`$ must be reinterpreted carefully. Under a transformation $`g`$, a constellation associated with $`N_s`$ points on the sphere will, in general, be mapped to a different constellation. In other words, the image $`\widehat{\mathrm{\Delta }}_{g\nu }=\widehat{\mathrm{\Delta }}(g𝐧_\nu )`$ is typically not one of the operators $`\widehat{\mathrm{\Delta }}_\nu `$. Nevertheless, condition (D4) is not empty: for appropriately chosen rotations $`g_{\nu \mu }`$ one can indeed map an operator defined at $`𝐧_\nu `$ to another one associated with the point $`𝐧_\mu `$, say. In this case, the consequences for the coefficients of the operators $`\widehat{\mathrm{\Delta }}_\nu `$ and $`\widehat{\mathrm{\Delta }}_\mu `$ are identical to those obtained in the continuous case. Similarly, invariance of the operator $`\widehat{\mathrm{\Delta }}_\nu `$ under a rotations about the axis $`𝐧_\nu `$ has the same impact as before. Thus the general ansatz for the discrete kernel (obtained from (7) by setting $`𝐧𝐧_\nu `$) is reduced by exploiting the postulates (D1-4) to the form
$$\widehat{\mathrm{\Delta }}_\nu \widehat{\mathrm{\Delta }}(𝐧_\nu )=\underset{m=s}{\overset{s}{}}\mathrm{\Delta }_m|m,𝐧_\nu m,𝐧_\nu |,\nu =1,\mathrm{},N_s.$$
(49)
Therefore, the discrete kernel $`\widehat{\mathrm{\Delta }}_\nu `$ can be thought of as a subset of $`N_s`$ operators $`\widehat{\mathrm{\Delta }}(𝐧_\nu )`$, each one associated with a point $`𝐧_\nu `$ of the sphere.
Let us mention an important difference between discrete and continuous kernels, $`\widehat{\mathrm{\Delta }}_\nu `$ and $`\widehat{\mathrm{\Delta }}(𝐧)`$, which arises in spite of their formal similarity. Once the coefficients $`\mathrm{\Delta }_m`$ are fixed a continuous kernel is determined completely. Discrete kernels, however, come in a much wider variety since they depend, in addition, on the selected constellation of points on the sphere. The discrete kernel does not enjoy the $`SU(2)`$ symmetry in the same way as does the continuous one. The discrete subgroups of $`SU(2)`$ being limited in type, the continuous symmetry will usually not be turn into a discrete one. Note, further, that the elements of the dual kernel depend, in general, on all the points of the constellation: $`\widehat{\mathrm{\Delta }}^\nu =\widehat{\mathrm{\Delta }}^\nu (𝐧_1\mathrm{},𝐧_{N_s})`$. This is easily seen from (48) since the variation of a single $`\widehat{\mathrm{\Delta }}_\nu `$ will have an effect on all $`\{\widehat{\mathrm{\Delta }}^\nu \}`$ in order to maintain orthogonality.
The additional freedom of selecting specific constellations is connected to a subtle point: actually, not all constellations of $`N_s`$ points give rise to a basis in the space $`𝒜_s`$. This remark is easily understood by considering $`IR^3`$ as an example of a linear space. The (continuous) collection of all unit vectors in three-space clearly spans it while not every subset of three vectors is a basis—they might lie in a plane. By analogy, one must ensure that the operators $`\widehat{\mathrm{\Delta }}_\nu ,\nu =1,\mathrm{},N_s`$, associated with a specific constellation, do indeed form a basis of $`𝒜_s`$. The operators are indeed linearly independent if the determinant of their (positive definite and symmetric) Gram matrix $`𝖦`$ satisfies
$$det𝖦>0,𝖦_{\nu \nu ^{}}=\text{ Tr }\left[\widehat{\mathrm{\Delta }}_\nu \widehat{\mathrm{\Delta }}_\nu ^{}\right],$$
(50)
a condition, which will be studied later in more detail.
Suppose now that the $`N_s`$ operators $`\widehat{\mathrm{\Delta }}_\nu `$ in (49) do form a basis. Then, the kernel dual to it, that is the set of operators $`\widehat{\mathrm{\Delta }}^\nu `$, is determined by the condition (48) instead of Eq. (2). Therefore, one cannot proceed as before to derive the conditions (27). In particular, it is no longer true that the elements of the dual kernel have an expansion analogous to (49). This follows immediately from the impossibility to satisfy (27) by an ansatz for $`\widehat{\mathrm{\Delta }}^\nu `$ of the form (49): Eq. (49) represent $`N_s`$ conditions but a dual of the form depends only on $`(2s+1)`$ free parameters $`\mathrm{\Delta }^m`$. Nevertheless, a dual kernel $`\widehat{\mathrm{\Delta }}^\nu `$ does exist and it is determined unambiguously—it simply cannot have the form (49). (A also ori, there is no self-dual kernel associated with the Stratonovich-Weyl postulates (D0-4)). Consequently, one expands any (self-adjoint) operator $`\widehat{A}`$ either in terms of a given kernel,
$$\widehat{A}=\frac{1}{2s+1}\underset{\nu =1}{\overset{N_s}{}}A^\nu \widehat{\mathrm{\Delta }}_\nu ,A^\nu =\text{ Tr }\left[\widehat{A}\widehat{\mathrm{\Delta }}^\nu \right],$$
(51)
or, equivalently, in terms of the dual kernel,
$$\widehat{A}=\frac{1}{2s+1}\underset{\nu =1}{\overset{N_s}{}}A_\nu \widehat{\mathrm{\Delta }}^\nu ,A_\nu =\text{ Tr }\left[\widehat{A}\widehat{\mathrm{\Delta }}_\nu \right].$$
(52)
The collection $`A(A_1,\mathrm{},A_{N_s})`$ of real coefficients in (52) now is defined as the discrete phase-space symbol of the operator $`\widehat{A}`$, and $`A^{dual}(A^1,\mathrm{},A^{N_s})`$ is the dual symbol.
The relation between the discrete symbol and its dual as well as between the pair of kernels is linear. It is easily implemented by means of the Gram matrix $`𝖦`$ and its inverse $`𝖦^1`$,
$$\left(𝖦^1\right)_{\nu \nu ^{}}𝖦^{\nu \nu ^{}}=\frac{1}{(2s+1)^2}\text{ Tr }\left[\widehat{\mathrm{\Delta }}^\nu \widehat{\mathrm{\Delta }}^\nu ^{}\right].$$
(53)
The matrix $`𝖦`$ thus plays the role of a metric,
$$\widehat{\mathrm{\Delta }}^\nu =(2s+1)\underset{\mu =1}{\overset{N_s}{}}𝖦^{\mu \nu }\widehat{\mathrm{\Delta }}_\mu ,$$
(54)
and the dual symbol is determined according to
$$A^\nu =(2s+1)\underset{\nu ^{}=1}{\overset{N_s}{}}𝖦^{\nu \nu ^{}}A_\nu ^{}.$$
(55)
The trace of two operators $`\widehat{A}`$ and $`\widehat{B}`$ is easily found to be expressible as a combination of a discrete symbol and a dual one,
$$\text{ Tr }\left[\widehat{A}\widehat{B}\right]=\underset{\nu =1}{\overset{N_s}{}}A^\nu B_\nu =\underset{\nu =1}{\overset{N_s}{}}A_\nu B^\nu ,$$
(56)
which is the discretized version of Eq. (45).
In order to have a discrete Moyal product, we seek to reproduce the multiplication of operators on the level of symbols. Using the definition of the symbols, it is straightforward to see that
$$(\widehat{A}\widehat{B})_\lambda =A_\lambda B_\lambda =\frac{1}{(2s+1)^2}\underset{\mu ,\nu =1}{\overset{N_s}{}}L_\lambda ^{\mu \nu }A_\mu B_\nu ,$$
(57)
with the trilinear kernel
$$L_\lambda ^{\mu \nu }=\text{ Tr }\left[\widehat{\mathrm{\Delta }}^\mu \widehat{\mathrm{\Delta }}^\nu \widehat{\mathrm{\Delta }}_\lambda \right],$$
(58)
in close analogy to Eq. (4).
### Discrete $`P`$\- and $`Q`$-symbols
A particularly interesting set of symbols emerges if, for a given allowed constellation, only one of the coefficients in the expansion (49) is different from zero, $`\mathrm{\Delta }_m=\delta _{ms}`$, say. Then, the kernel consists of $`N_s`$ operators projecting on coherent states,
$$\widehat{Q}_\nu =|𝐧_\nu 𝐧_\nu |.$$
(59)
This is obviously the non-redundant counterpart of Eq. (36) implying that a self-adjoint operator $`\widehat{A}`$ is determined by a symbol which consists of $`N_s`$ pure-state expectation values, the discrete $`Q`$-symbol,
$$A_\nu =\text{ Tr }[\widehat{A}\widehat{Q}_\nu ]=𝐧_\nu |\widehat{A}|𝐧_\nu .$$
(60)
Let us point out that the introduction of discrete symbols has actually been triggered by the search for a simple method to reconstruct the density matrix of a spin through expectation values . In fact, this problem is solved by Eq. (60) in the most economic way. If $`\widehat{A}`$ is chosen to be the density matrix $`\widehat{\rho }`$ of a spin $`s`$, then the $`\nu `$-th component of the $`Q`$-symbol equals the probability of measuring the eigenvalue $`s`$ in the direction $`𝐧_\nu `$,
$$p_s(𝐧_\nu )=𝐧_\nu |\widehat{\rho }|𝐧_\nu .$$
(61)
Knowledge of the $`N_s`$ measurable probabilities $`p_s(𝐧_\nu )`$ thus amounts to knowing the density matrix $`\widehat{\rho }`$.
If the $`Q`$-symbol (60) determines an operator $`\widehat{A}`$, the values of the continuous Q-symbol of $`\widehat{A}`$ at points different from those of the constellation must be functions of the numbers $`(A_1,\mathrm{},A_{N_s})`$. For a coherent state $`|𝐧_0|𝐧_\nu `$, not a member of the constellation, this dependence reads explicitly
$$𝐧_0|\widehat{A}|𝐧_0=\frac{1}{2s+1}\underset{\nu =1}{\overset{N_s}{}}A^\nu |𝐧_0|𝐧_\nu |^2.$$
(62)
Here the $`P`$-symbol $`A^{dual}`$ of $`\widehat{A}`$ is required, calculated from its $`Q`$-symbol by means of (55) once the matrix
$$𝖦𝖦_{\nu \nu ^{}}=\text{ Tr }\left[\widehat{Q}_\nu \widehat{Q}_\nu ^{}\right]=|𝐧_\nu |𝐧_\nu ^{}|^2,$$
(63)
has been inverted. Furthermore, knowledge of $`𝖦^1`$ provides immediately the dual kernel $`\widehat{Q}^\nu `$ via (54) but no explicit general expression such as (41) is known.
It will be shown now how to directly determine the matrix elements of the dual kernel without using the inverse of $`𝖦`$. The orthogonality of the kernel and its dual, Eq. (48) can be written as
$$\delta _\nu ^\nu ^{}=\frac{1}{2s+1}\text{ Tr }\left[\widehat{Q}_\nu \widehat{Q}^\nu ^{}\right]=\frac{1}{2s+1}\underset{m,m^{}=s}{\overset{s}{}}m^{}|\widehat{Q}_\nu |mm|\widehat{Q}^\nu ^{}|m^{},$$
(64)
using the completeness relation for the $`z`$ eigenstates $`|m,𝐧_z`$. Introduce an $`(N_s\times N_s)`$ matrix $`𝖰`$ with elements $`𝖰_{\nu ,mm^{}}=m|\widehat{Q}_\nu |m^{}`$, where the index $`(m,m^{})`$ of the columns runs through $`N_s`$ values according to
$$\{(2s,2s),(2s,2s1),\mathrm{},(2s,0),(2s1,2s),\mathrm{},(0,0)\}.$$
(65)
As is obvious from (64), the matrix elements of the dual kernel, $`𝖰_{mm^{}}^\nu =m|\widehat{Q}^\nu |m^{}`$ can be read off once the inverse of the matrix $`𝖰`$ has been found. The expansion coefficients of a coherent state $`|𝐧`$ in the $`z`$ basis $`|m,𝐧_z`$ are given by
$$m,𝐧_z|𝐧=\frac{1}{(1+|z|^2)^s}\left(\begin{array}{c}2s\\ sm\end{array}\right)^{1/2}z^{sm},$$
(66)
where the complex number $`z`$ is the stereographic image in the complex plane of the point $`𝐧`$ on the sphere. Therefore, one can write $`𝖰`$ as a product of three matrices two of which are diagonal: $`𝖰=𝖣_1\mathrm{𝖭𝖣}_2`$. The diagonal matrices
$`𝖣_1`$ $`=`$ $`\mathrm{diag}\left[(1+|z_\nu |^2)^{2s}\right],`$ (67)
$`𝖣_2`$ $`=`$ $`\mathrm{diag}\left[\left(\begin{array}{c}2s\\ 2sm\end{array}\right)^{1/2}\left(\begin{array}{c}2s\\ 2sm^{}\end{array}\right)^{1/2}\right],`$ (72)
with $`\nu =1,\mathrm{},N_s,`$ and $`m,m^{}=0,\mathrm{},2s`$, have inverses since all diagonal entries are different from zero. The hard part of the inversion is due to the matrix $`𝖭`$ with elements
$$𝖭_{\nu ,mm}=z_\nu ^{2sm}(z_\nu ^{})^{2sm^{}},$$
(73)
similar to but not identical with to the structure of a Vandermonde matrix. As discussed in the following chapter, particular constellations give rise to matrices $`𝖭`$ with inversion formulae simpler than the general one. Once $`𝖭`$ has been inverted, the matrix elements of the dual kernel are given by the rows of the $`(N_s\times N_s)`$ matrix
$$𝖰^1=𝖣_2^1𝖭^1𝖣_1^1.$$
(74)
For discrete $`Q`$-symbols, the kernel $`L`$ in (58), which implements the discrete $``$ product, has the form:
$$L_{\mu \nu \lambda }=\text{ Tr }\left[\widehat{Q}_\mu \widehat{Q}_\nu \widehat{Q}_\lambda \right]=𝐧_\lambda |𝐧_\mu 𝐧_\mu |𝐧_\nu 𝐧_\nu |𝐧_\lambda ,$$
(75)
which, by using results from , can be written as
$`L_{\mu \nu \lambda }`$ $`=`$ $`{\displaystyle \frac{1}{4^{2s}}}(1+𝐧_\mu 𝐧_\nu +𝐧_\nu 𝐧_\lambda +𝐧_\lambda 𝐧_\mu +i𝐧_\mu 𝐧_\nu 𝐧_\lambda )^{2s}`$ (76)
$`=`$ $`g_0(𝐧_\mu 𝐧_\nu )^sg_0(𝐧_\nu 𝐧_\lambda )^sg_0(𝐧_\lambda 𝐧_\mu )^se^{isA(\mu \nu \lambda )},`$ (77)
where $`g_0(𝐧_\mu 𝐧_\nu )=(1+𝐧_\mu 𝐧_\nu )/2`$, and, defining $`g(\mu \nu \lambda )`$ as the term in round brackets of (76), one has
$$A(\mu \nu \lambda )=\frac{1}{i}\mathrm{ln}\left(\frac{g(\mu \nu \lambda )}{g^{}(\mu \nu \lambda )}\right).$$
(78)
Therefore, the phase $`A`$ has a geometrical interpretation : it is the surface of the geodesic triangle given by the points $`𝐧_\mu ,𝐧_\nu ,𝐧_\lambda `$ .
## 4 Constellations
In this section examples of specific constellations are presented for which it is possible to prove at least that the Gram matrix has a determinant different from zero. Furthermore, in some cases relatively simple expressions for the dual kernel or, equivalently, for the inverse of the Gram matrix $`𝖦`$ are obtained. The kernel is supposed throughout to consist of $`N_s`$ projection operators $`\widehat{Q}_\nu `$ on coherent states as given in (59). In other words, the focus is on discrete $`Q`$-symbols and the $`P`$-symbols related to them. Note that, once a constellation has been shown to give rise to a basis in $`𝒜_s`$, the inversion of its Gram matrix is always possible but lengthy (already for a spin $`1/2`$): express the matrix elements of $`𝖦^1`$ in terms of the co-factors of $`𝖦`$. Four different types of constellations will be discussed involving randomly chosen points, points on nested cones, on free cones, and on spirals.
#### Random constellations
As shown in , almost any distribution of $`N_s`$ points on the sphere $`𝖲^2`$ gives rise to an allowed constellation. A random selection of directions leads with probability one to an invertible Gram matrix. This result shows that in an infinitesimal neighborhood of any forbidden constellation one can find an allowed one.
#### Nested cones
Historically, this family of constellations provided the first example of allowed constellations for both integer and half-integer spins . For an integer value of $`s`$, e.g., consider $`(2s+1)`$ cones about one axis in space, $`𝐞_z`$, say, all with different opening angles. Distribute $`(2s+1)`$ directions over each of these nested cones in such a way that the ensemble of directions on each cone is invariant under a rotation about $`𝐞_z`$ by an angle $`2\pi /(2s+1)`$. For specific opening angles of the cones, the inversion of the matrix $`𝖭`$ in (73) reduces after a Fourier transformation to the inversion of $`(2s+1)`$ Vandermonde matrices of size $`(2s+1)\times (2s+1)`$. For a half-integer spin the same construction is possible except that the directions on different cones must also lie on different meridians. There is, in fact, a slight generalization of this result: the same calculation with $`(2s+1)`$ arbitrary different opening angles leads to $`(2s+1)`$ generalized Vandermonde matrices with nonzero determinant defining thus also a allowed constellations.
Constellations on nested cones are useful also for numerical calculations because they allow one to distribute $`N_s`$ points in a homogeneous fashion on the surface of the sphere. If two points of a constellation approach each other, the determinant of the matrix $`𝖦`$ typically becomes very large, with a disastrous effect on numerical precision.
### Free cones
Here is another family of constellation involving $`(2s+1)`$ cones with directions located on them. However, now the cones may be oriented arbitrarily (no nesting), and the number of directions may vary from cone to cone. For example, the number of points on a cone can be chosen to equal the multiplicities of the spherical harmonics $`Y_{lm}`$ with $`l=2s`$. It is claimed that allowed constellations can be identified by taking into account the following properties (tested numerically for values up to $`s=6`$):
1. The determinant of $`𝖦`$ is zero if there are more than $`(4s+1)`$ directions on a single cone.
2. If there are $`(4s+1)`$ points on one cone, then another cone will contain at most $`(4s1)`$ points, allowing for no more than $`(4s3)`$ directions on the third cone, etc.
3. It is necessary to have directions located on at least $`(2s+1)`$ different cones.
For a spin $`1/2`$ these properties will be shown to hold in the next section. The first of these observations can be proved for arbitrary spin $`s`$ by using a particular decomposition of the matrix $`𝖦`$,
$$𝖦=𝗀^{}𝗀,$$
(79)
exploiting the fact that a positive definite matrix can always be written as the ‘square’ of its ‘root.’ A lengthy calculation involving properties of rotation matrices, Legendre polynomials and spherical harmonics leads to a factorization, $`𝗀=\mathrm{𝖽𝗒}`$, the first matrix being diagonal and having $`(2s+1)`$ different entries,
$$𝖽_{(l)}=\frac{2\sqrt{\pi }(2s)!}{\sqrt{(2s+1+l)!(2sl)!}},l=0,\mathrm{},2s,$$
(80)
each value occurring $`(2l+1)`$ times. The second matrix has columns given by the $`N_s`$ lowest spherical harmonics evaluated at one of the $`N_s`$ points of the constellation,
$$𝗒=\left(\begin{array}{cccc}Y_{00}(𝐧_1)& Y_{00}(𝐧_2)& \mathrm{}& Y_{00}(𝐧_{N_s})\\ Y_{11}(𝐧_1)& Y_{11}(𝐧_2)& \mathrm{}& Y_{11}(𝐧_{N_s})\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ Y_{2s2s}(𝐧_1)& Y_{2s2s}(𝐧_2)& \mathrm{}& Y_{2s2s}(𝐧_{N_s})\end{array}\right).$$
(81)
Consequently, the Gram matrix $`𝖦`$ is invertible if and only if $`det𝗒0`$. The matrix (81) can accomodate at most $`(4s+1)`$ directions on one cone, corresponding to one value of $`\vartheta `$ with respect to some fixed axis. The subsequent multiplicities $`(4s1),(4s3),\mathrm{}`$, are due to applying the same argument to the remaining subspaces with dimensions $`2(l1)+1,2(l2)+1,\mathrm{}`$
In physical terms, determinant of (81) is easily interpreted as a Slater determinant of a quantum system: it equals the (totally anti-symmetric) ground-state wave-function of $`N_s`$ non-interacting fermions restricted to move on a sphere. The node lines of this wave function correspond to forbidden constellations in which the corresponding operator kernel is degenerate, i.e., does not give rise to a basis in $`𝒜_s`$.
### Spirals
A particularly convenient constellation is defined in the following way: let the $`N_s`$ directions be defined by $`N_s`$ complex numbers points $`z_\nu `$ constructed out of a single point $`z_0`$ (neither of modulus one nor purely real),
$$z_\nu =z_0^{\nu 1},\nu =1,\mathrm{},N_s.$$
(82)
The points are thus located on a spiral in the complex plane.
The matrix $`𝖭`$ defined in (73) then turns into an $`(N_s\times N_s)`$ Vandermonde matrix, that is,
$$𝖵_{\nu \mu }=x_\nu ^{\mu 1},\nu ,\mu =1,\mathrm{},N_s.$$
(83)
Its inverse is known explicitly (84), given, for example, in , with elements
$$𝖵_{\nu \mu }^1=\frac{(1)^{\nu +1}}{_{\lambda \mu }(x_\lambda x_\mu )}S_{N_s\nu }(\{x_\lambda \}_{\lambda =1}^{N_s}x_\mu ),$$
(84)
where $`S_{N_s\nu }(\{x_\lambda \}_{\lambda =1}^{N_s}x_\mu )`$ is the symmetrical function constructed out of the $`N\nu `$ numbers $`x_\nu `$ with $`\nu \lambda `$. One has, for example, $`S_2(x_1,x_2,x_3)=x_1x_2+x_1x_3+x_2x_3`$.
## 5 Discrete Moyal representation for a spin $`1/2`$
In this section the discrete Moyal representation will be worked out in detail for a spin with quantum number $`s=1/2`$, allowing for explicit results throughout. For clarity, it is assumed from the outset that the kernel consists of four projection operators
$$\widehat{Q}_\nu =|𝐧_\nu 𝐧_\nu |,\nu =1,\mathrm{},N_s.$$
(85)
It is easy to generalize the results derived below to the case of four linear combinations of $`|\pm 𝐧_\nu \pm 𝐧_\nu |`$ compatible with Eq. (12).
Let us start with the determination of the dual kernel which can be found by the intermediate step of inverting the $`(4\times 4)`$ Gram matrix with elements
$$𝖦=|𝐧_\mu |𝐧_\nu |^2=\frac{1}{2}\left(1+𝐧_\mu 𝐧_\nu \right).$$
(86)
This matrix is easily factorized: $`𝖦=𝗀^{}𝗀/2`$, where
$$𝗀=\left(\begin{array}{cccc}1& 1& 1& 1\\ n_{1x}& n_{2x}& n_{3x}& n_{4x}\\ n_{1y}& n_{2y}& n_{3y}& n_{4y}\\ n_{1z}& n_{2z}& n_{3z}& n_{4z}\end{array}\right).$$
(87)
The absolute value of the determinant of $`𝗀`$ is proportional to the volume of the tetrahedron defined by the four points $`𝐧_\nu `$ on the surface of the sphere implying $`|det𝖦|=18V_{tetra}`$. Since a ‘flat’ tetrahedron has no volume, the entire set of forbidden constellations has a simple geometric description:
$$det𝖦=0\mathrm{the}\mathrm{four}\mathrm{points}𝐧_\nu \mathrm{are}\mathrm{located}\mathrm{on}\mathrm{a}\mathrm{circle}\mathrm{on}𝒮^2.$$
(88)
Consequently, allowed constellations are characterized by three vectors on a cone (any three points on a sphere define a circle), plus any fourth vector not on this cone. This agrees with the earlier statements about free-cones constellations.
Here is a simple way to invert the matrix $`𝗀`$ and subsequently $`𝖦`$. Consider a matrix
$$𝖿=\left(\begin{array}{cccc}1& f_x^1& f_y^1& f_z^1\\ 1& f_x^2& f_y^2& f_z^2\\ 1& f_x^3& f_y^3& f_z^3\\ 1& f_x^4& f_y^4& f_z^4\end{array}\right),$$
(89)
defined in terms of four vectors $`𝐟^\nu =(f_x^\nu ,f_y^\nu ,f_z^\nu )`$ not required to have length one. The matrix elements of the of product $`𝖿`$ and $`𝗀`$ are given by
$$\left(\mathrm{𝖿𝗀}\right)_\nu ^\mu =1+𝐟^\mu 𝐧_\nu .$$
(90)
This is a diagonal matrix if the scalar products $`𝐟^\mu 𝐧_\mu `$ equal $`1`$ whenever $`\mu \nu `$. Geometrically, such four vectors are constructed easily: the vector $`𝐟^1`$ points to the unique intersection of the three planes tangent to the sphere at the points $`𝐧_2`$, $`𝐧_3`$ and $`𝐧_4`$. Analytically, this vector reads
$$𝐟^1=\frac{𝐧_2𝐧_3+𝐧_3𝐧_4+𝐧_4𝐧_2}{(𝐧_2𝐧_3)𝐧_4},$$
(91)
and the three remaining vectors follow from cyclic permutation of the numbers $`1`$ to $`4`$. With this choice the inverse of the matrix $`𝗀`$ can be written as
$$𝗀^1=𝖽^1𝖿,$$
(92)
where $`𝖽`$ is the diagonal matrix in (90): $`𝖽_{\nu \nu }=1+𝐟^\nu 𝐧_\nu `$. Consequently, the inverse of the Gram matrix $`𝖦`$ for a general allowed constellation is given by
$$𝖦^1=2𝖽^1𝖿𝖿^{}𝖽^1,$$
(93)
having matrix elements
$$𝖦_{\mu \nu }^1𝖦^{\mu \nu }=2\frac{1+𝐟^\mu 𝐟^\nu }{(1+𝐧_\mu 𝐟^\mu )(1+𝐧_\nu 𝐟^\nu )}.$$
(94)
In general, the elements $`\widehat{Q}^\nu `$ of the dual kernel will thus be linear combinations of all four projection operators $`\widehat{Q}_\nu `$.
It is interesting to express the kernel and its dual in terms of the Pauli matrices $`\sigma \sigma =(\sigma _x,\sigma _y,\sigma _z)`$:
$$\widehat{Q}_\nu =\frac{1}{2}(𝖨+𝐧_\nu \sigma \sigma ),\widehat{Q}^\nu =2\frac{𝖨+𝐟^\nu \sigma \sigma }{1+𝐟^\nu 𝐧_\nu },$$
(95)
allowing one to show easily that they satisfy the required duality.
For reference, we give the $`Q`$\- and $`P`$-symbols of the spin operator
$$𝐬_\nu =\frac{1}{2}𝐧_\nu ,𝐬^\nu =\frac{2𝐟^\nu }{1+𝐟^\nu 𝐧_\nu },$$
(96)
and of the identity,
$$I_\nu =1,I^\nu =\frac{4}{1+𝐟^\nu 𝐧_\nu },$$
(97)
and the symbols of arbitrary operators for a spin $`1/2`$ follow from linear combinations.
## 6 Discussion
Operator kernels have been used for a systematic study of phase-space representations of a quantum spin $`s`$. The kernels have been derived from appropriate Stratonovich-Weyl postulates taking slightly different forms for continuous and discrete representations, respectively. Emphasis is on the discrete Moyal formalism which allows one to describe hermitean operators, including density matrices, by a minimal number of probabilities easily measured by a Stern-Gerlach apparatus. As a useful by-product a natural and most economic method of state reconstruction emerges when a quantum spin is described in terms of discrete symbols. Further, Schrödinger’s equation for a spin $`s`$ turns into a set of coupled linear differential equations for $`(2s+1)^2`$ probabilities .
In addition, a new form of the kernel defining continuous Wigner functions for a spin has been obtained (20): it has been expressed as an ensemble of operators obtained from all possible rotations of one fixed operator. This is entirely analogous to an elegant expression of the kernel for particle-Wigner functions as the ensemble of all possible phase-space translations of the parity operator. Therefore, continuous phase-space representations for both spin and particle systems now are seen to derive from structurally equivalent operator kernels.
The discrete symbolic calculus is an interesting ‘hybrid’ between the classical and quantal descriptions of a spin. On the one hand, this representation is equivalent to standard quantum mechanics of a spin. On the other, the independent variables carry phase-space coordinates as labels (51,52). However, only a finite subset of points in phase space (corresponding to an allowed constellation) are involved reflecting thus the discretization characteristic of quantum mechanics.
The $`N_s`$ projections operators associated with a constellation of points define a non-orthogonal basis for hermitean operators acting on the Hilbert space of the spin. Each projection is a positive operator, and, altogether, they give rise to a resolution of unity. One might suspect that they define a positive operator-valued measure or POVM, for short. However, this is not the case since the closure relation does not involve just the bare projections but they are multiplied with factors some of which necessarily take negative values. Such an obstruction through ‘negative probabilities’ is not surprising; other phase-space representations are based on quantum mechanical ‘quasi-probabilities,’ known to have this property, too.
Let us close with a synopsis of the fundamental Moyal-type representations for particle and spin systems known so far.
| | self-dual kernel | dual pairs |
| --- | --- | --- |
| particle | Wigner functions \[ unknown \] | $`P`$-, $`Q`$-symbols \[ unknown \] |
| spin | Stratonovich/Varilly \[ impossible \] | Berezin symbols \[ $`P^\nu `$-, $`Q_\nu `$-symbols\] |
The table provides both a summary and points at open questions. The individual entries give the names of the familiar continuous phase-space representations while the corresponding quantities for the discrete formalism are in square brackets. Future work will focus on developing a discrete Moyal-type formalism for a quantum particle. To do this, one must exhibit, for example, a pair of dual kernels one of which would consist of a countable set of projection operators on coherent states. This set is required to be a basis in the linear space of (bounded?) operators on the particle Hilbert-space. It is not obvious in which way the associate discrete $`P`$-symbol would reflect the subtleties of its continuous counterpart which may be singular. Similarly, the existence of a self-dual discrete kernel for a quantum particle is an open question.
### Acknowledgements
St. W. acknowledges financial support by the Schweizerische Nationalfonds.
|
warning/0004/math0004119.html
|
ar5iv
|
text
|
# On subgroups of minimal topological groups
## 1. Introduction
This paper was motivated by the following questions:
###### Question 1.1 (V. Pestov, A. Arhangelskii, 1980’s).
What are subgroups of minimal topological groups?
###### Question 1.2 (W. Roelcke, 1990).
What are subgroups of lower precompact topological groups?
We now explain and discuss the notions of a minimal group and of a lower precompact group.
Compact spaces $`X`$ can be characterized among all Tikhonov spaces by each of the following two properties: (1) $`X`$ is minimal, in the sense that $`X`$ admits no strictly coarser Tikhonov (or Hausdorff) topology; (2) $`X`$ is absolutely closed, which means that $`X`$ is closed in any Tikhonov space $`Y`$ containing $`X`$ as a subspace. One can consider the notions of minimality and absolute closedness also for other classes of spaces. For example, for the class of Hausdorff spaces one gets the notions of $`H`$-minimal and $`H`$-closed spaces which are no longer equivalent to each other or to compactness but are closely related: a space is $`H`$-minimal iff it is $`H`$-closed and semiregular, and a space is compact iff it is $`H`$-minimal and satisfies the Urysohn separation axiom. See the survey for a discussion of these notions.
Let us now consider the case of topological groups. All topological groups are assumed to be Hausdorff, unless otherwise explicitly stated. A topological group is minimal if it does not admit a strictly coarser Hausdorff group topology<sup>1</sup><sup>1</sup>1The survey on minimal groups contains a lot of information and more than a hundred references.. A topological group is absolutely closed if it is closed in every topological group containing it as a topological subgroup. A topological group $`G`$ is absolutely closed if and only if it is Rajkov-complete, or upper complete, that is complete with respect to the upper uniformity which is defined as the least upper bound $``$ of the left and the right uniformities on $`G`$. Recall that the sets $`\{(x,y):x^1yU\}`$, where $`U`$ runs over a base at unity of $`G`$, constitute a base of entourages for the left uniformity $``$ on $`G`$. In the case of the right uniformity $``$, the condition $`x^1yU`$ is replaced by $`yx^1U`$. We shall call Rajkov-complete groups simply complete. The Rajkov completion $`\widehat{G}`$ of a topological group $`G`$ is the completion of $`G`$ with respect to the upper uniformity $``$. For every topological group $`G`$ the space $`\widehat{G}`$ has a natural structure of a topological group. The group $`\widehat{G}`$ can be defined as a unique (up to an isomorphism) complete group containing $`G`$ as a dense subgroup. A group is Weil-complete if it is complete with respect to the left uniformity $``$ (or, equivalently, with respect to the right uniformity $``$). Every Weil-complete group is complete, but not vice versa.
Unlike the category of Hausdorff spaces, where “minimal” implies “absolutely closed”, minimal groups need not be absolutely closed (that is, complete). If $`G`$ is a minimal group, then its Rajkov completion $`\widehat{G}`$ also is minimal. On the other hand, if $`G`$ is a dense subgroup of a minimal group $`H`$, then $`G`$ is minimal if and only if for every closed normal subgroup $`N\{1\}`$ of $`H`$ we have $`GN\{1\}`$ (; see historical remarks in \[7, Section 2.1\]). Thus the study of minimal groups can be reduced to the study of complete minimal groups: a group $`G`$ is minimal if and only if its Rajkov completion $`\widehat{G}`$ is minimal, and for every closed normal subgroup $`N\{1\}`$ of $`\widehat{G}`$ we have $`GN\{1\}`$. Compact groups are complete minimal, and in the Abelian case the converse is also true, according to a deep theorem of Prodanov and Stoyanov : every complete minimal Abelian group is compact. In the non-Abelian case, the class of complete minimal groups properly contains the class of compact groups. There exist non-compact minimal Lie groups , and actually a discrete infinite group can be minimal . It is natural to ask how big the difference is between the class of compact groups and the class of complete minimal groups. For example, one can ask if the class of complete minimal groups is closed under infinite products (this question, to the best of my knowledge, is still open; the answer is positive for groups with a trivial center ), or if the relations between cardinal invariants of compact groups remain valid for complete minimal groups, etc.
If $`G`$ is a topological group, we denote by $`𝒩(G)`$ the filter of neighbourhoods of the neutral element. Besides the left, right, and upper uniformities (denoted by $``$, $``$, and $``$, respectively), every topological group has yet another compatible uniformity $``$, the greatest lower bound of $``$ and $``$. (Note that in general the greatest lower bound of two compatible uniformities on a topological space need not be compatible with the topology.) If $`U𝒩(G)`$, the cover $`\{UxU:xG\}`$ is $``$-uniform, and every $``$-uniform cover of $`G`$ has a refinement of this form. The uniformity $``$ is called the lower uniformity in ; we shall call it the Roelcke uniformity, in honour of Walter Roelcke who was the first to introduce and investigate this notion.
A uniform space $`X`$ is precompact if its completion is compact or, equivalently, if for every entourage $`U`$ the space $`X`$ can be covered by finitely many $`U`$-small sets. A topological group $`G`$ is precompact if one of the following equivalent conditions holds<sup>2</sup><sup>2</sup>2Answering a question of Walter Roelcke, I proved that these conditions are also equivalent to: (5) for every $`U𝒩(G)`$ there exists a finite set $`FG`$ such that $`FUF=G`$. This was later rediscovered by S. Solecki and other authors. A short proof can be found in \[4, Proposition 4.3\].: (1) $`(G,)`$ is precompact; (2) $`(G,)`$ is precompact; (3) $`(G,)`$ is precompact; (4) for every $`U𝒩(G)`$ there exists a finite set $`FG`$ such that $`FU=UF=G`$. Every Tikhonov space is a subspace of a compact space, but not every topological group is a subgroup of a compact group: the subgroups of compact groups are precisely precompact groups. Let us say that a topological group $`G`$ is Roelcke-precompact if it is precompact with respect to the Roelcke uniformity $``$. Thus $`G`$ is Roelcke-precompact iff for every $`U𝒩(G)`$ there exists a finite set $`FG`$ such that $`UFU=G`$. The Roelcke completion of a topological group $`G`$ is the completion of $`G`$ with respect to the Roelcke uniformity $``$. If $`G`$ is Roelcke-precompact, the Roelcke completion $`R(G)`$ of $`G`$ will be also called the Roelcke compactification.
Precompact groups are Roelcke-precompact, but not vice versa . For example, the unitary group of a Hilbert space or the group $`\mathrm{Sym}(E)`$ of all permutations of a discrete set $`E`$, both considered with the pointwise convergence topology, are Roelcke-precompact but not precompact. While the left, right and upper uniformities of a subgroup of a topological group are induced by the corresponding uniformities of the group, this is not so for the Roelcke uniformity, and a subgroup of a Roelcke-precompact group need not be Roelcke-precompact. This justifies Question 1.2.
The aim of the present paper is to provide a complete answer to Questions 1.1 and 1.2. Let us say that a group is $`G`$ topologically simple if $`G`$ has no closed normal subgroups besides $`G`$ and $`\{1\}`$.
###### Main Theorem 1.3.
Every topological group $`G`$ is isomorphic to a subgroup of a complete minimal group which is Roelcke-precompact, topologically simple and has the same weight as $`G`$.
“Isomorphic” means here “isomorphic as a topological group”. The weight of a topological space $`X`$ is the cardinal $`w(X)=\mathrm{min}\{||:\text{ is a base for }X\}`$. A group $`G`$ is totally minimal if all Hausdorff quotient groups of $`G`$ are minimal. Since minimal topologically simple groups are totally minimal, we could write “totally minimal” instead of “minimal” in our Main Theorem.
Let $`Q=[0,1]^\omega `$ be the Hilbert cube, and let $`\mathrm{Homeo}(Q)`$ be the topological group of all self-homeomorphisms of $`Q`$. The group $`H=\mathrm{Homeo}(Q)`$ is universal , \[27, Theorem 2.2.6\], in the sense that every topological group $`G`$ with a countable base is isomorphic to a topological subgroup of $`H`$. Therefore, for groups with a countable base a natural way to prove Theorem 1.3 would be to prove that the group $`\mathrm{Homeo}(Q)`$, which is known to be simple, is Roelcke-precompact and minimal. I do not know if $`\mathrm{Homeo}(Q)`$ indeed has these properties:
###### Problem 1.4.
Is the group $`\mathrm{Homeo}(Q)`$ Roelcke-precompact or minimal?
There is another universal topological group with a countable base, namely the group $`\mathrm{Iso}(𝕌)`$ of all self-isometries of the Urysohn universal metric space $`𝕌`$ , \[27, Theorem 2.3.1\], \[50, Theorem 6.1\]. The group $`\mathrm{Iso}(𝕌)`$ is not Roelcke-precompact \[50, p. 344\]; I do not know whether it is minimal or not.
###### Problem 1.5.
Is the group $`\mathrm{Iso}(𝕌)`$ minimal?
We consider the “bounded version” $`𝕌_1`$ of the space $`𝕌`$ and show that the group $`\mathrm{Iso}(𝕌_1)`$ is Roelcke-precompact, topologically simple and minimal. This proves Theorem 1.3 for groups with a countable base. For groups of uncountable weight the argument is similar, but we must consider non-separable analogues of the space $`𝕌_1`$.
Recall some definitions. A bijection between metric spaces is an isometry if it is distance-preserving. For a metric space $`M`$ we denote by $`\mathrm{Iso}(M)`$ the topological group of all isometries of $`M`$ onto itself, equipped with the topology of pointwise convergence (which coincides in this case with the compact-open topology). Let $`d`$ be the metric on $`M`$. The sets of the form $`U_{F,ϵ}=\{g\mathrm{Iso}(M):d(g(x),x)<ϵ\text{ for all }xF\}`$, where $`F`$ is a finite subset of $`M`$ and $`ϵ>0`$, constitute a base at the unity of $`\mathrm{Iso}(M)`$.
A metric space $`M`$ is $`\omega `$-homogeneous if every isometry $`f:AB`$ between finite subsets $`A,B`$ of $`M`$ can be extended to an isometry of $`M`$ onto itself. The Urysohn universal metric space $`𝕌`$ is the unique (up to an isometry) complete separable metric space which has either of the following two properties: (1) $`𝕌`$ is $`\omega `$-homogeneous and contains an isometric copy of every separable metric space; (2) $`𝕌`$ is finitely injective: if $`L`$ is a finite metric space, $`KL`$ and $`f:K𝕌`$ is an isometric embedding, then $`f`$ can be extended to an isometric embedding of $`L`$ into $`𝕌`$. For the equivalence of the conditions (1) and (2), see Proposition 1.6 below (we consider there the bounded version $`𝕌_1`$ of $`𝕌`$, but the proof for $`𝕌`$ is the same). Actually $`𝕌`$ is compactly injective as well: in the definition of finite injectivity, one can replace finite metric spaces $`KL`$ by arbitrary compact metric spaces , see also \[32, Lemma 5.1.19 and Proposition 5.1.20\].
We now introduce the bounded version of the space $`𝕌`$. The diameter of a metric space $`(M,d)`$ is the number $`sup\{d(x,y):x,yM\}`$. Let us say that a metric space $`M`$ is Urysohn if its diameter is equal to $`1`$ and it is finitely injective with respect to spaces of diameter $`1`$, that is, the following holds: if $`L`$ is a finite metric space of diameter $`1`$, $`KL`$ and $`f:KM`$ is an isometric embedding, then $`f`$ can be extended to an isometric embedding of $`L`$ into $`M`$. It suffices if this property holds for $`L=K\{p\}`$. Thus a metric space $`M`$ of diameter $`1`$ is Urysohn iff for any finite sequence $`x_1,\mathrm{},x_n`$ of points of $`M`$ and any sequence $`a_1,\mathrm{},a_n`$ of positive numbers $`1`$ such that $`|a_ia_j|d(x_i,x_j)a_i+a_j(i,j=1,\mathrm{},n)`$ there exists $`yM`$ such that $`d(y,x_i)=a_i(i=1,\mathrm{},n)`$. Using the notion of a Katětov function that will be introduced later in Section 3, we can reformulate this condition as follows: for every finite $`XM`$ and every Katětov function $`f:X[0,1]`$ there exists $`yM`$ such that $`d(x,y)=f(x)`$ for every $`xX`$.
Remark. A notation like Urysohn<sub>≤1</sub> might have been more appropriate for what we have called Urysohn (note that the unbounded space $`𝕌`$ is not Urysohn according to our definition!). However, we shall use the shorter term, in hope that no confusion will arise. Let us again bring to the reader’s attention that all Urysohn spaces have diameter $`1`$.
###### Proposition 1.6.
Let $`M`$ be a metric space of diameter $`1`$.
1. if $`M`$ is Urysohn, then $`M`$ contains an isometric copy of every countable metric space of diameter $`1`$. If $`M`$ moreover is complete, then it contains an isometric copy of every separable metric space of diameter $`1`$;
2. if $`M`$ is $`\omega `$-homogeneous and contains an isometric copy of every finite metric space of diameter $`1`$, then $`M`$ is Urysohn;
3. if $`M_1`$ and $`M_2`$ are complete separable Urysohn spaces, then every isometry between finite subsets $`AM_1`$ and $`BM_2`$ extends to an isometry between $`M_1`$ and $`M_2`$;
4. a complete separable metric space of diameter 1 is Urysohn if and only if it is $`\omega `$-homogeneous and contains an isometric copy of every finite metric space of diameter $`1`$;
5. there exists a unique (up to an isometry) complete separable Urysohn space $`𝕌_1`$. The space $`𝕌_1`$ is the unique complete separable metric space of diameter $`1`$ which is $`\omega `$-homogeneous and contains an isometric copy of every separable metric space of diameter $`1`$;
6. there exists a non-complete separable $`\omega `$-homogeneous Urysohn space which contains an isometric copy of every separable metric space of diameter $`1`$.
This is essentially due to Urysohn . The last item was added by Katětov , who answered a question of Urysohn that had remained open for more than 60 years.
###### Proof.
(1) is obvious (use induction). To prove (2), suppose that $`KL`$ are finite metric spaces, $`\mathrm{diam}(L)1`$, and let $`f:KM`$ be an isometric embedding. Pick an isometric embedding $`g:LM`$, and use $`\omega `$-homogeneity of $`M`$ to find an isometry $`h`$ of $`M`$ such that $`h`$ extends the isometry $`f(g_{|K})^1:g(K)f(K)`$. Then $`hg:LM`$ is an isometric embedding that extends $`f`$. For (3), enumerate dense countable subsets in $`M_1`$ and $`M_2`$ and use the “back-and-forth” (or “shuttle”) method to extend the given isometry between $`A`$ and $`B`$ to an isometry between dense subsets of $`M_1`$ and $`M_2`$. Then use completeness to obtain an isometry between $`M_1`$ and $`M_2`$. Applying (3) in the case when $`M_1=M_2`$, we see that every complete separable Urysohn space is $`\omega `$-homogeneous. Thus (4) and uniqueness in (5) follow from (1)–(3). The existence of $`𝕌_1`$ is a special case of Theorem 3.2 that we shall prove later; the idea of the proof is due to Katětov. The existence of a non-complete Urysohn space easily follows from Katětov’s methods presented in this paper; we refer the reader to for details. ∎
For the history of invention of the universal Urysohn space, see , , . According to P.S. Alexandrov , P.S. Urysohn was thinking about the universal space in the very last days of his life, and, after finishing another project on 14 August 1924, was going to work on two further papers: on metrization of normal spaces with a countable base and on the universal space. He wrote just the first page of the first of these papers. It was dated 17 August 1924, the day of his death.
For more on the Urysohn space, see , and papers in this volume. We mention the striking result of Vershik: for a generic point $`d`$ of the Polish space of metrics on a countable set $`X`$ the completion of $`(X,d)`$ is isometric to the Urysohn space $`𝕌`$ . Similarly, for a generic shift-invariant metric $`d`$ on $`𝐙`$ (= the group of integers) the completion of the metric group $`(𝐙,d)`$ is isometric to $`𝕌`$ .
The proof of Theorem 1.3 consists of two parts. We first prove that every topological group can be embedded in the group $`\mathrm{Iso}(M)`$ of isometries of a complete $`\omega `$-homogeneous Urysohn space $`M`$, and then prove that such groups of isometries are minimal, Roelcke-precompact and topologically simple.
###### Theorem 1.7.
For every topological group $`G`$ there exists a complete $`\omega `$-homogeneous Urysohn metric space $`M`$ of the same weight as $`G`$ such that $`G`$ is isomorphic to a subgroup of $`\mathrm{Iso}(M)`$.
###### Theorem 1.8.
If $`M`$ is a complete $`\omega `$-homogeneous Urysohn metric space, then the group $`\mathrm{Iso}(M)`$ is complete, Roelcke-precompact, minimal and topologically simple. The weight of $`\mathrm{Iso}(M)`$ is equal to the weight of $`M`$.
Theorem 1.3 follows from Theorems 1.7 and 1.8.
The proof of Theorem 1.7 depends on Katětov’s construction that leads to a canonical embedding of any metric space $`M`$ into a finitely injective space. In the non-separable case this construction must be complemented by a construction of a natural embedding of a metric space into an $`\omega `$-homogeneous space. We use Graev metrics on free groups for this.
The proof of Theorem 1.8 is based on the study of the Roelcke compactifications of groups of isometries. The Roelcke compactifications of some topological groups of importance admit an explicit description and are equipped with additional structures. For example, for the unitary group $`U_s(H)`$, where $`H`$ is a Hilbert space and the subscript $`s`$ indicates the strong operator topology (= the topology inherited from the Tikhonov product $`H^H`$), the Roelcke compactification can be identified with the unit ball $`\mathrm{\Theta }`$ in the algebra of bounded linear operators on $`H`$ . The ball $`\mathrm{\Theta }`$ is equipped with the weak operator topology. This is the topology inherited from $`H^H`$, where each factor $`H`$ carries the weak topology. Another case when the Roelcke compactification can be explicitly described is the following. Let $`K`$ be a zero-dimensional compact space such that all non-empty clopen subspaces of $`K`$ are homeomorphic to $`K`$. For example, $`K`$ might be the Cantor set. Let $`G=\mathrm{Homeo}(K)`$, equipped with the compact-open topology. Then $`R(G)`$ is the set of all closed relations $`R`$ on $`K`$ (= closed subsets of $`K^2`$) such that the domain and the range of $`R`$ is equal to $`K`$ . Yet another example of a topological group $`G`$ for which $`R(G)`$ is known is the group $`G=\mathrm{Homeo}_+[0,1]`$ of all orientation-preserving self-homeomorphisms of the closed interval $`I=[0,1]`$. In that case $`R(G)`$ can be identified with the closure of the set of graphs of elements of $`G`$ in the space of closed subsets of the square $`I^2`$, see the picture in \[27, Example 2.5.4\].
The proof of Theorem 1.8 leans on the study of the Roelcke compactification $`R(G)`$ for $`G=\mathrm{Iso}(M)`$, where $`M`$ is a complete $`\omega `$-homogeneous Urysohn metric space. In this case $`R(G)`$ can be identified with the space of metric spaces covered by two isometric copies of $`M`$, see Sections 6 and 7 below. Equivalently, $`\mathrm{\Theta }=R(G)`$ can be identified with a certain subset of $`I^{M\times M}`$ that we now are going to specify.
A semigroup is a set with an associative binary operation. Let $`S`$ be a semigroup with the multiplication $`(x,y)xy`$. An element $`xS`$ is an idempotent if $`x^2=x`$. We say that a self-map $`xx^{}`$ of $`S`$ is an involution if $`x^{}=x`$ and $`(xy)^{}=y^{}x^{}`$ for all $`x,yS`$. An element $`xS`$ is symmetrical if $`x^{}=x`$, and a subset $`AS`$ is symmetrical if $`A^{}=A`$. An ordered semigroup is a semigroup with a partial order $``$ such that the conditions $`xx^{}`$ and $`yy^{}`$ imply $`xyx^{}y^{}`$.
Denote by $`I`$ the closed unit interval $`[0,1]`$. Let $``$ be the associative operation on $`I`$ defined by $`xy=\mathrm{min}(x+y,\mathrm{\hspace{0.17em}1})`$. Let $`X`$ be a set, and let $`S=I^{X\times X}`$ be the set of all functions $`f:X^2I`$. We make $`S`$ into an ordered semigroup with an involution. Define an operation $`(f,g)fg`$ on $`S`$ by
$$fg(x,y)=inf\{f(x,z)g(z,y):zX\}(x,yX).$$
This operation is associative, since for $`f,g,hS`$ and $`x,yX`$ both $`(fg)h(x,y)`$ $`f(gh)(x,y)`$ are equal to
$$inf\{f(x,z)g(z,u)h(u,y):z,uX\}.$$
Define an involution $`ff^{}`$ on $`S`$ by $`f^{}(x,y)=f(y,x)`$.
Let $`(M,d)`$ be a complete $`\omega `$-homogeneous Urysohn metric space, and let $`G=\mathrm{Iso}(M)`$. The Roelcke compactification $`\mathrm{\Theta }`$ of $`G`$ can be identified with a closed subsemigroup of $`I^{M\times M}`$ and has a natural structure of an ordered semigroup with an involution. Namely, $`\mathrm{\Theta }`$ can be viewed as the set of all functions $`fI^{M\times M}`$ which are bi-Katětov in the sense of Definition 6.1. Such functions can be described in terms of the structure of an ordered semigroup with an involution on $`I^{M\times M}`$: a function $`fI^{M\times M}`$ is bi-Katětov if and only if
$$fd=df=f,f^{}fd,ff^{}d.$$
The metric $`d`$ is the unity of $`\mathrm{\Theta }`$, and the constant 1 is a zero element of $`\mathrm{\Theta }`$, in the sense that $`f1=1f=1`$ for every $`f\mathrm{\Theta }`$ (in fact, for every $`fI^{M\times M}`$).
Note that $`\mathrm{\Theta }`$ is a compact space and a semigroup, but it might be misleading to call it a “compact semigroup”, since the semigroup operation on $`\mathrm{\Theta }`$ is not (even separately) continuous. However, both the topology and the algebraic structure on $`\mathrm{\Theta }`$ will play an important role in our proofs.
The Roelcke compactification $`\mathrm{\Theta }`$ of $`G=\mathrm{Iso}(M)`$ is used to prove Theorem 1.8 in the following way. Let $`f:GG^{}`$ be a surjective morphism of topological groups. To prove that $`G`$ is minimal and topologically simple, we must prove that either $`f`$ is a homeomorphism or $`|G^{}|=1`$. Extend $`f`$ to a map $`F:\mathrm{\Theta }\mathrm{\Theta }^{}`$, where $`\mathrm{\Theta }^{}`$ is the Roelcke compactification of $`G^{}`$. Let $`S=F^1(e^{})`$, where $`e^{}`$ is the unity of $`G^{}`$. Then $`S`$ is a closed subsemigroup of $`\mathrm{\Theta }`$ which is invariant under inner automorphisms. To every closed subsemigroup of $`\mathrm{\Theta }`$ an idempotent can be assigned in a canonical way. Let $`p`$ be the idempotent corresponding to $`S`$. Since $`S`$ is invariant under inner automorphisms, so is $`p`$. We show that certain idempotents in $`\mathrm{\Theta }`$ are in a one-to-one correspondence with closed subsets of $`M`$ (Proposition 6.4). Since there are no non-trivial $`G`$-invariant closed subsets of $`M`$, it follows that $`p`$ is trivial: it is either the unity of $`\mathrm{\Theta }`$ or the constant 1. Accordingly, either $`f`$ is a homeomorphism or $`G^{}=\{e^{}\}`$.
The same method was used in and to give alternative proofs of Stoyanov’s theorem that the unitary group of a Hilbert space is totally minimal and of Gamarnik’s theorem that the group of homeomorphisms of the Cantor set is minimal, see Remarks 2 and 3 in Section 9 below.
Under the conditions of Theorem 1.8, the group $`\mathrm{Iso}(M)`$ has the fixed point on compacta (f.p.c.) property. This deep result is due to V. Pestov <sup>3</sup><sup>3</sup>3 The setting considered in these papers and books is not exactly the same as in Theorem 1.8 (detailed proofs are given either for the separable case or for unbounded metrics), but, as noted in , the same argument works for bounded metrics verbatim.. A topological group $`G`$ has the f.p.c. property, or is extremely amenable, if every compact $`G`$-space has a $`G`$-fixed point. As pointed out by Pestov, his theorem, combined with Theorem 1.7 of the present paper, implies that every topological group is a subgroup of an extremely amenable group.
We prove Theorem 1.7 in Section 5 and Theorem 1.8 in Section 8.
Another version of Question 1.1 is the following (see \[2, Problem VI.6\], \[26, Problem 519\]): is every topological group a quotient of a minimal topological group? I have earlier announced that the answer is positive. Moreover, I claimed that every topological group is a group retract of a minimal topological group. In other words, for every topological group $`G`$ there exist a minimal topological group $`G^{}G`$ and a morphism $`r:G^{}G`$ such that $`r^2=r`$ (it follows that $`G`$ is a quotient of $`G^{}`$). My announcement appears as Theorem 3.3F.2 in . However, my announcement was premature, and my “proof” contained a gap. A complete proof has been recently found by M. Megrelishvili .
Megrelishvili’s construction shows that every complete group is a group retract of a complete minimal group. This result, combined with the fact that every topological group is a quotient of a Weil-complete group <sup>4</sup><sup>4</sup>4It was proved in that the free topological group of any stratifiable space is Weil-complete. Since every topological space is the image of a stratifiable space under a quotient (even open) map , it follows that every topological group is a quotient of a Weil-complete group., implies that every topological group is a quotient of a complete minimal group. Indeed, given any topological group $`G`$, represent $`G`$ as a quotient of a complete group $`G^{}`$, and then, using Megrelishvili’s theorem, represent $`G^{}`$ as a group retract (and hence as a quotient) of a complete minimal group.
## 2. Invariant pseudometrics on groups
A pseudometric $`d`$ on a group $`G`$ is left-invariant if $`d(xy,xz)=d(y,z)`$ for all $`x,y,zG`$. Right-invariant pseudometrics are defined similarly. A pseudometric is two-sided invariant if it is left-invariant and right-invariant. Let $`e`$ be the unity of $`G`$. A non-negative real function $`p`$ on $`G`$ is a seminorm if it satisfies the following conditions: (1) $`p(e)=0`$; (2) $`p(xy)p(x)+p(y)`$ for all $`x,yG`$; (3) $`p(x^1)=p(x)`$ for all $`xG`$. If $`p`$ is a seminorm on $`G`$, define a left-invariant pseudometric $`d`$ by $`d(x,y)=p(x^1y)`$. We thus get a one-to-one correspondence between seminorms and left-invariant pseudometrics. Given a left-invariant pseudometric $`d`$, the corresponding seminorm $`p`$ is defined by $`p(x)=d(x,e)`$. A seminorm $`p`$ is invariant if it is invariant under inner automorphisms, that is $`p(yxy^1)=p(x)`$ for every $`x,yG`$. Invariant seminorms correspond to two-sided invariant pseudometrics.
Now let $`G`$ be a topological group. Then the topology of $`G`$ is determined by the collection of all continuous left-invariant pseudometrics \[15, Theorem 8.2\]. Equivalently, for every neighbourhood $`U`$ of unity there exists a continuous seminorm $`p`$ on $`G`$ such that the set $`\{xG:p(x)<1\}`$ is contained in $`U`$.
###### Theorem 2.1.
For every topological group $`G`$ there exists a metric space $`M`$ such that $`w(G)=w(M)`$ and $`G`$ is isomorphic (as a topological group) to a subgroup of $`\mathrm{Iso}(M)`$.
This theorem has been rediscovered many times by various authors, see historical remarks in .
###### 1st proof.
There exists a family $`D=\{d_\alpha :\alpha A\}`$ of continuous left-invariant pseudometrics on $`G`$ which determines the topology of $`G`$ and has the cardinality $`|A|=w(G)`$. Replacing, if necessary, each $`dD`$ by $`inf(d,1)`$, we may assume that all pseudometrics in $`D`$ are bounded by 1. For every $`\alpha A`$ let $`M_\alpha `$ be the metric space associated with the pseudometric space $`(G,d_\alpha )`$, and let $`M=_{\alpha A}M_\alpha `$ be the disjoint sum of the spaces $`M_\alpha `$. There is an obvious metric on $`M`$ which extends the metric of each $`M_\alpha `$: if two points of $`M`$ are in distinct pieces $`M_\alpha `$ and $`M_\beta `$, define the distance between them to be 1. The left action of $`G`$ on itself yields for every $`\alpha A`$ a natural continuous homomorphism $`G\mathrm{Iso}(M_\alpha )`$. The homomorphism $`G_{\alpha A}\mathrm{Iso}(M_\alpha )`$ thus obtained is a homeomorphic embedding. It remains to note that the group $`_{\alpha A}\mathrm{Iso}(M_\alpha )`$ can be identified with a topological subgroup of $`\mathrm{Iso}(M)`$. ∎
###### 2nd proof.
Let $`B`$ be the Banach space of all bounded real functions on $`G`$ which are uniformly continuous with respect to the right uniformity. The natural left action of $`G`$ on $`B`$, defined by the formula $`gf(h)=f(g^1h)`$ $`(g,hG,fB)`$, yields an isomorphic embedding of $`G`$ into $`\mathrm{Iso}(B)`$. The weight of $`B`$ may exceed the weight of $`G`$, but it is easy to find a $`G`$-invariant subspace $`B^{}`$ of $`B`$ such that $`B^{}`$ determines the topology of $`G`$ and $`w(B^{})=w(G)`$. Then the natural homomorphism $`G\mathrm{Iso}(B^{})`$ still is a homeomorphic embedding. ∎
Let us discuss invariant seminorms on free groups. For a set $`X`$ we denote by $`S(X)`$ the set of all words of the form $`x_1^{ϵ_1}\mathrm{}x_n^{ϵ_n}`$, where $`n0`$, $`x_iX`$ and $`ϵ_i=\pm 1`$, $`1in`$. In other words, $`S(X)`$ is the free monoid<sup>5</sup><sup>5</sup>5A monoid is a semigroup with a neutral element. We require that monoid morphisms should preserve the neutral element. on the set $`XX^1`$, where $`X^1`$ is a disjoint copy of $`X`$. A word $`wS(X)`$ is irreducible if it does not contain subwords of the form $`x^ϵx^ϵ`$. We consider the free group $`F(X)`$ on a set $`X`$ as the set of all irreducible words in $`S(X)`$. Every word $`wS(X)`$ represents a uniquely defined element $`w^{}F(X)`$ which can be obtained from $`w`$ by consecutive deletion of subwords of the form $`x^ϵx^ϵ`$. In this situation we say that the words $`w`$ and $`w^{}`$ are equivalent. For $`u,vS(X)`$ we denote by $`u|v`$ the product of $`u`$ and $`v`$ in the semigroup $`S(X)`$, that is the word obtained by writing $`v`$ after $`u`$ (without cancelations). If $`u`$ and $`v`$ are irreducible, we denote by $`uv`$ their product in the group $`F(X)`$, that is the irreducible word equivalent to $`u|v`$.
Let $`(X,d)`$ be a metric space. A real function $`f`$ on $`X`$ is non-expanding, or $`1`$-Lipschitz, if $`|f(x)f(y)|d(x,y)`$ for every $`x,yX`$. Let $`k`$ be a non-negative non-expanding function on $`X`$. We shall describe a two-sided invariant pseudometric $`Gr(d,k)`$ on the free group $`F(X)`$ which is called the Graev pseudometric . The corresponding invariant seminorm $`p`$ is characterized by the following property: $`p`$ is the greatest invariant seminorm on $`F(X)`$ such that $`p(x)=k(x)`$ and $`p(x^1y)d(x,y)`$ for every $`x,yX`$. We shall need later the following explicit construction of the seminorm $`p`$.
It will be convenient to define the function $`p`$ on the entire set $`S(X)`$. Given a word $`w=x_1^{ϵ_1}\mathrm{}x_n^{ϵ_n}S(X)`$, we define a $`w`$-pairing to be a collection $`E`$ of pairwise disjoint two-element subsets of the set $`J=\{1,\mathrm{},n\}`$ such that: (1) if $`\{a,b\}E`$ and $`\{i,j\}E`$, where $`a<b`$ and $`i<j`$, then the intervals $`[a,b]`$ and $`[i,j]`$ are either disjoint or one of them is contained in the other (this means that the cases $`a<i<b<j`$ and $`i<a<j<b`$ are excluded); (2) if $`\{i,j\}E`$, then $`ϵ_i=ϵ_j`$. To put it less formally, some pairs of letters of the word $`w`$ are connected by arcs, each letter is connected with at most one other letter, each arc connects a pair of letters of the form $`x`$ and $`y^1`$ ($`x,yX`$), and the arcs do not intersect each other. Given a $`w`$-pairing $`E`$, define the Graev sum $`s_E=s_E(w)`$ by
$$s_E=\{d(x_i,x_j):\{i,j\}E,i<j\}+\{k(x_i):iJE\},$$
and let $`p(w)`$ be the minimum of the numbers $`s_E`$, taken over the finite set of all $`w`$-pairings $`E`$.
We claim that $`p(w)=p(w^{})`$ if the words $`w,w^{}S(X)`$ are equivalent. It suffices to consider the case when $`w=u|v`$ and $`w^{}=u|x^ϵx^ϵ|v`$. We show that for every $`w^{}`$-pairing $`E^{}`$ there exists a $`w`$-pairing $`E`$ such that $`s_Es_E^{}`$, and vice versa. In one direction this is obvious: given a $`w`$-pairing $`E`$, which we consider as a system of arcs connecting the letters of the word $`w`$, add one more arc which connects the letters $`x^ϵ`$ and $`x^ϵ`$ of the word $`w^{}`$. We get a $`w^{}`$-pairing $`E^{}`$ for which $`s_E=s_E^{}`$. Conversely, let a $`w^{}`$-pairing $`E^{}`$ be given. We must construct a $`w`$-pairing $`E`$ for which $`s_Es_E^{}`$. As above, we consider $`E^{}`$ as a system of arcs. The word $`w`$ is obtained from $`w^{}`$ by deleting the subword $`x^ϵx^ϵ`$. To get $`E`$, we replace the arcs which go from the letters $`x^ϵ`$ and $`x^ϵ`$ and leave the other arcs unchanged. Consider the following cases.
Case 1. There is an arc in $`E^{}`$ connecting the letters $`x^ϵ`$ and $`x^ϵ`$. Then just delete this arc to get $`E`$. We have $`s_E=s_E^{}`$.
Case 2. The letters $`x^ϵ`$ and $`x^ϵ`$ are connected in $`E^{}`$, but not with each other. Let $`x^ϵ`$ be connected with $`y^ϵ`$ and $`x^ϵ`$ be connected with $`z^ϵ`$. Replace these two connections by one connection between $`y^ϵ`$ and $`z^ϵ`$. The sums $`s_E`$ and $`s_E^{}`$ differ by the terms $`d(y,z)`$ and $`d(y,x)+d(x,z)`$, hence the triangle inequality implies that $`s_Es_E^{}`$.
Case 3. One of the letters $`x^ϵ`$ and $`x^ϵ`$, say $`x^ϵ`$, is connected in $`E^{}`$ and the other is unpaired. Let $`x^ϵ`$ be connected with $`y^ϵ`$. Delete this connection and leave the letter $`y^ϵ`$ unpaired in $`E`$. The sums $`s_E`$ and $`s_E^{}`$ differ by the terms $`k(y)`$ and $`d(x,y)+k(x)`$. Since the function $`k`$ is non-expanding, we have $`k(y)d(x,y)+k(x)`$ and hence $`s_Es_E^{}`$.
Case 4. Both $`x^ϵ`$ and $`x^ϵ`$ are unpaired in $`E^{}`$. Then all arcs are left without change. The sum $`s_E`$ is obtained from $`s_E^{}`$ by omitting the non-negative term $`2k(x)`$, hence $`s_Es_E^{}`$.
We have thus proved the claim that $`p(w)=p(w^{})`$ for equivalent words $`w,w^{}S(X)`$. It follows that the restriction of $`p`$ to $`F(X)`$ is indeed a seminorm: if $`u,vF(X)`$, then $`p(uv)=p(u|v)p(u)+p(v)`$. It is easy to see that $`p(u)=p(u^1)`$ for every $`uF(X)`$. We show that $`p`$ is invariant under inner automorphisms. If $`uS(X)`$, $`xX`$, $`ϵ=\pm 1`$ and $`w=x^ϵ|u|x^ϵ`$, then $`p(w)p(u)`$, since every $`u`$-pairing can be extended in an obvious way to a $`w`$-pairing with the same Graev sum. It follows that for every $`u,vF(X)`$ we have $`p(uvu^1)=p(u|v|u^1)p(v)`$, and by symmetry of the relation of being conjugate in $`F(X)`$ also the opposite inequality holds. Thus $`p(uvu^1)=p(v)`$, which means that the seminorm $`p`$ is invariant.
Let $`Y`$ be a pseudometric space, and let $`\mathrm{Iso}(Y)`$ be the group of all distance-preserving permutations of $`Y`$, equipped with the topology of pointwise convergence. Then $`\mathrm{Iso}(Y)`$ is a topological group, not necessarily Hausdorff. For later use we note here the following:
###### Lemma 2.2.
Let $`(X,d)`$ be a metric space, and let $`k`$ be a non-expanding function on $`X`$. Let $`D=Gr(d,k)`$ be the Graev pseudometric on the free group $`G=F(X)`$. Let $`H_1\mathrm{Iso}(X)`$ be the topological group of all isometries of $`X`$ which preserve the function $`k`$, and let $`H_2\mathrm{Iso}(G)`$ be the topological group (not necessarily Hausdorff) of all automorphisms of $`G`$ which preserve the pseudometric $`D`$. Then the natural homomorphism $`\phi \phi ^{}`$ from $`H_1`$ to $`H_2`$ is continuous.
###### Proof.
It suffices to show that for every $`wG`$ the map $`\phi \phi ^{}(w)`$ from $`H_1`$ to $`(G,D)`$ is continuous at the unity. If $`w=x_1^{ϵ_1}\mathrm{}x_n^{ϵ_n}`$, then $`\phi ^{}(w)=\phi (x_1)^{ϵ_1}\mathrm{}\phi (x_n)^{ϵ_n}`$, and we have $`D(\phi ^{}(w),w)_{i=1}^nd(\phi (x_i),x_i)`$. Let $`ϵ>0`$ be given. If $`\phi H_1`$ is close enough to the unity, then $`d(\phi (x_i),x_i)<ϵ/n`$, $`1in`$, and therefore $`D(\phi ^{}(w),w)<ϵ`$. ∎
## 3. Katětov’s construction of Urysohn extensions
###### Definition 3.1.
Let $`M`$ be a subspace of a metric space $`L`$. We say that $`M`$ is $`g`$-embedded in $`L`$ if there exists a continuous homomorphism $`e:\mathrm{Iso}(M)\mathrm{Iso}(L)`$ such that for every $`\phi \mathrm{Iso}(M)`$ the isometry $`e(\phi ):LL`$ is an extension of $`\phi `$.
Let $`M`$ be a $`g`$-embedded subspace of a metric space $`L`$. A homomorphism $`e:\mathrm{Iso}(M)\mathrm{Iso}(L)`$ satisfying the condition of Definition 3.1 is a homeomorphic embedding, since the inverse map $`e(\phi )\phi =e(\phi )|M`$ is continuous. It follows that $`\mathrm{Iso}(M)`$ is isomorphic to a topological subgroup of $`\mathrm{Iso}(L)`$.
In this section we prove the following theorem:
###### Theorem 3.2.
Let $`M`$ be a metric space of diameter $`1`$. There exists a complete Urysohn metric space $`L`$ containing $`M`$ as a subspace such that $`w(L)=w(M)`$ and $`M`$ is $`g`$-embedded in $`L`$.
It follows that for every topological group $`G`$ there exists a complete Urysohn metric space $`M`$ of the same weight as $`G`$ such that $`G`$ is isomorphic to a subgroup of $`\mathrm{Iso}(M)`$. This is weaker than Theorem 1.7, since in the non-separable case the metric space $`M`$ need not be $`\omega `$-homogeneous. In the next section we shall prove that every metric space $`M`$ can be $`g`$-embedded into an $`\omega `$-homogeneous metric space $`L`$. Using this fact, we show that the Urysohn space $`L`$ in Theorem 3.2 can be additionally assumed $`\omega `$-homogeneous (Theorem 5.1). This yields Theorem 1.7, see Section 5.
The arguments of show that Theorem 3.2 essentially follows from Katětov’s construction of Urysohn extensions . For the reader’s convenience we give a detailed proof.
Let $`(X,d)`$ be a metric space of diameter $`1`$. We say that a function $`f:X[0,1]`$ is Katětov if $`|f(x)f(y)|d(x,y)f(x)+f(y)`$ for all $`x,yX`$. A function $`f`$ is Katětov if and only if there exists a metric space $`X^{}=X\{p\}`$ of diameter $`1`$ containing $`X`$ as a subspace such that for every $`xX`$ $`f(x)`$ is equal to the distance between $`x`$ and $`p`$. Let $`E(X)`$ be the set of all Katětov functions on $`X`$, equipped with the sup-metric $`d_X^E`$ defined by $`d_X^E(f,g)=sup\{|f(x)g(x)|:xX\}`$. If $`Y`$ is a non-empty subset of $`X`$ and $`fE(Y)`$, define $`g=\kappa _Y(f)E(X)`$ by
$$g(x)=inf(\{d(x,y)+f(y):yY\}\{1\})=inf\{d(x,y)f(y):yY\}$$
for every $`xX`$. It is easy to check that $`g`$ is indeed a Katětov function on $`X`$ and that $`g`$ extends $`f`$; one can define $`g`$ as the largest 1-Lipschitz function $`X[0,1]`$ that extends $`f`$. The map $`\kappa _Y:E(Y)E(X)`$ is an isometric embedding. Let
$$E(X,\omega )=\{\kappa _Y(E(Y)):YX,Y\text{ is finite and non-empty}\}E(X).$$
For every $`xX`$ let $`h_xE(X)`$ be the function on $`X`$ defined by $`h_x(y)=d(x,y)`$. Note that $`h_x=\kappa _{\{x\}}(0)`$ and hence $`h_xE(X,\omega )`$. The map $`xh_x`$ is an isometric embedding of $`X`$ into $`E(X,\omega )`$. Thus we can identify $`X`$ with a subspace of $`E(X,\omega )`$.
###### Lemma 3.3.
If $`xX`$ and $`fE(X)`$, then $`d_X^E(f,h_x)=f(x)`$.
###### Proof.
Since $`f`$ is a Katětov function, for every $`yY`$ we have $`f(y)d(x,y)f(x)`$ and $`d(x,y)f(y)f(x)`$. Hence $`d_X^E(f,h_x)=sup\{|f(y)d(x,y)|:yX\}f(x)`$, and at $`y=x`$ the equality is attained. ∎
###### Lemma 3.4.
Let $`Z=Y\{p\}`$ be a finite metric space of diameter $`1`$. Every isometric embedding $`j:YX`$ extends to an isometric embedding of $`Z`$ into $`E(X,\omega )`$.
###### Proof.
We may assume that $`Y`$ is a subspace of $`X`$ and that $`j(y)=y`$ for every $`yY`$. Let $`fE(Y)`$ be the Katětov function defined by $`f(y)=\nu (y,p)`$ for every $`yY`$, where $`\nu `$ is the metric on $`Z`$. Let $`g=\kappa _Y(f)E(X,\omega )`$. We claim that the extension of $`j`$ over $`Z`$ which maps $`p`$ to $`g`$ is an isometric embedding. It suffices to check that $`d_X^E(h_y,g)=\nu (y,p)`$ for every $`yY`$. Fix $`yY`$. Let $`h_y^{}E(Y)`$ be the restriction of $`h_y`$ to $`Y`$. According to Lemma 3.3 we have $`d_Y^E(h_y^{},f)=f(y)`$. Since $`h_y=\kappa _Y(h_y^{})`$, $`g=\kappa _Y(f)`$ and the map $`\kappa _Y:E(Y)E(X)`$ is distance-preserving, it follows that $`d_X^E(h_y,g)=d_Y^E(h_y^{},f)=f(y)=\nu (y,p)`$, as claimed. ∎
###### Lemma 3.5.
Any metric space $`X`$ of diameter $`1`$ is $`g`$-embedded in $`E(X,\omega )`$.
###### Proof.
It is clear that every isometry $`\phi :YZ`$ between any two metric spaces can be extended to an isometry $`\phi ^{}:E(Y,\omega )E(Z,\omega )`$. Such an extension is unique, since every point in $`E(Y,\omega )`$ (or, more generally, in $`E(Y)`$) is uniquely determined by its distances from the points of $`Y`$ (Lemma 3.3), and similarly for $`Z`$. In particular, every isometry $`\phi \mathrm{Iso}(X)`$ uniquely extends to an isometry $`\phi ^{}\mathrm{Iso}(E(X,\omega ))`$. The map $`\phi \phi ^{}`$ is a homomorphism of groups. We show that this homomorphism is continuous. Fix $`fE(X,\omega )`$ and $`ϵ>0`$. Pick a finite subset $`Y`$ of $`X`$ and $`gE(Y)`$ so that $`f=\kappa _Y(g)`$. Let $`U`$ be the set of all $`\phi \mathrm{Iso}(X)`$ such that $`d(\phi (y),y)<ϵ`$ for every $`yY`$. Then $`U`$ is a neighbourhood of unity in $`\mathrm{Iso}(X)`$. It suffices to show that $`d_X^E(\phi ^{}(f),f)<ϵ`$ for every $`\phi U`$. Fix $`\phi U`$. Let $`g_\phi =g\phi ^1E(\phi (Y))`$. Then $`\phi ^{}(f)=\kappa _{\phi (Y)}(g_\phi )`$. Thus for every $`xX`$ we have
$$\phi ^{}(f)(x)=inf\{d(x,z)g_\phi (z):z\phi (Y)\}=inf\{d(x,\phi (y))g(y):yY\}.$$
Since
$$f(x)=inf\{d(x,y)g(y):yY\},$$
it follows that
$$|\phi ^{}(f)(x)f(x)|sup\{|d(x,\phi (y))d(x,y)|:yY\}\mathrm{max}\{d(y,\phi (y)):yY\}<ϵ,$$
whence $`d_X^E(\phi ^{}(f),f)<ϵ`$. ∎
Let $`\alpha `$ be an ordinal, and let $`=\{M_\beta :\beta <\alpha \}`$ be a family of metric spaces such that $`M_\beta `$ is a subspace of $`M_\gamma `$ for all $`\beta <\gamma <\alpha `$. We say that the family $``$ is continuous if $`M_\beta =_{\gamma <\beta }M_\gamma `$ for every limit ordinal $`\beta <\alpha `$, $`\beta >0`$.
###### Proposition 3.6.
Let $`\{M_\beta :\beta \alpha \}`$ be an increasing continuous chain of metric spaces. If $`M_\beta `$ is $`g`$-embedded in $`M_{\beta +1}`$ for every $`\beta <\alpha `$, then $`M_0`$ is $`g`$-embedded in $`M_\alpha `$.
###### Proof.
For every $`\beta <\alpha `$ pick a continuous homomorphism $`e_\beta :\mathrm{Iso}(M_\beta )\mathrm{Iso}(M_{\beta +1})`$ such that $`e_\beta (\phi )`$ extends $`\phi `$ for every $`\phi \mathrm{Iso}(M_\beta )`$. By transfinite recursion on $`\beta \alpha `$ define a homomorphism $`\mathrm{\Lambda }_\beta :\mathrm{Iso}(M_0)\mathrm{Iso}(M_\beta )`$ such that $`\mathrm{\Lambda }_\beta (\phi )`$ extends $`\mathrm{\Lambda }_\gamma (\phi )`$ for every $`\phi \mathrm{Iso}(M_0)`$ and $`\gamma <\beta \alpha `$. Let $`\mathrm{\Lambda }_0`$ be the identity map of $`\mathrm{Iso}(M_0)`$. If $`\beta =\gamma +1`$, put $`\mathrm{\Lambda }_\beta =e_\gamma \mathrm{\Lambda }_\gamma `$. If $`\beta `$ is a limit ordinal, let $`\mathrm{\Lambda }_\beta (\phi )`$ be the isometry of $`M_\beta `$ such that for every $`\gamma <\beta `$ its restriction to $`M_\gamma `$ is equal to $`\mathrm{\Lambda }_\gamma (\phi )`$. We prove by induction on $`\beta `$ that each homomorphism $`\mathrm{\Lambda }_\beta `$ is continuous. This is obvious for non-limit ordinals. Assume that $`\beta `$ is limit. To prove that $`\mathrm{\Lambda }_\beta :\mathrm{Iso}(M_0)\mathrm{Iso}(M_\beta )`$ is continuous, it suffices to show that for every $`xM_\beta `$ the map $`\phi \mathrm{\Lambda }_\beta (\phi )(x)`$ from $`\mathrm{Iso}(M_0)`$ to $`M_\beta `$ is continuous. Fix $`xM_\beta `$. Pick $`\gamma <\beta `$ so that $`xM_\gamma `$. Then $`\mathrm{\Lambda }_\beta (\phi )(x)=\mathrm{\Lambda }_\gamma (\phi )(x)`$ for every $`\phi \mathrm{Iso}(M_0)`$. The map $`\mathrm{\Lambda }_\gamma `$ is continuous by the assumption of induction, hence the map $`\phi \mathrm{\Lambda }_\beta (\phi )(x)=\mathrm{\Lambda }_\gamma (\phi )(x)`$ also is continuous. Thus $`\mathrm{\Lambda }_\alpha :\mathrm{Iso}(M_0)\mathrm{Iso}(M_\alpha )`$ is a continuous homomorphism such that $`\mathrm{\Lambda }_\alpha (\phi )`$ extends $`\phi `$ for every $`\phi \mathrm{Iso}(M_0)`$. This means that $`M_0`$ is $`g`$-embedded in $`M_\alpha `$. ∎
Put $`X_0=X`$, $`X_{n+1}=E(X_n,\omega )`$. We consider each $`X_n`$ as a subspace of $`X_{n+1}`$, so we get an increasing sequence $`X_0X_1\mathrm{}`$ of metric spaces. Let $`X_\omega =\{X_n:n\omega \}`$.
###### Proposition 3.7.
The space $`X_\omega `$ is Urysohn, and $`X`$ is $`g`$-embedded in $`X_\omega `$.
###### Proof.
Let $`Z=Y\{p\}`$ be a finite metric space of diameter $`1`$, and let $`j:YX_\omega `$ be an isometric embedding. Pick $`n\omega `$ so that $`j(Y)X_n`$. In virtue of Lemma 3.4, there exists an isometric embedding of $`Z`$ into $`X_{n+1}X_\omega `$ which extends $`j`$. This means that $`X_\omega `$ is Urysohn. The second assertion of the proposition follows from Lemma 3.5 and Proposition 3.6. ∎
###### Proposition 3.8 ().
The weight of $`X_\omega `$ is equal to the weight of $`X`$.
###### Proof.
It suffices to show that for every metric space $`X`$ the weight of $`E(X,\omega )`$ is equal to the weight of $`X`$. Let $`w(X)=\tau `$, and let $`A`$ be a dense subset of $`X`$ of cardinality $`\tau `$. Let $`\gamma =\{\kappa _Y(E(Y)):YA,Y\text{ is finite}\}`$. Then $`\gamma `$ is a family of separable subspaces of $`E(X,\omega )`$, $`|\gamma |=\tau `$ and $`\gamma `$ is dense in $`E(X,\omega )`$ (see the proof of Lemma 1.8 in ). Hence $`E(X,\omega )`$ has a dense subspace of cardinality $`\tau `$. ∎
###### Proposition 3.9.
Every metric space is $`g`$-embedded in its completion.
###### Proof.
Let $`M`$ be a metric space, $`\overline{M}`$ be its completion. Every isometry $`\phi \mathrm{Iso}(M)`$ uniquely extends to an isometry $`\phi ^{}\mathrm{Iso}(\overline{M})`$. We show that the homomorphism $`\phi \phi ^{}`$ is continuous. Let $`d`$ be the metric on $`\overline{M}`$. Fix $`x\overline{M}`$ and $`ϵ>0`$. Pick $`yM`$ so that $`d(x,y)<ϵ`$. Let $`U=\{\phi \mathrm{Iso}(M):d(\phi (y),y)<ϵ\}`$. Then $`U`$ is a neighbourhood of unity in $`\mathrm{Iso}(M)`$. If $`\phi U`$, then $`d(\phi ^{}(x),x)d(\phi ^{}(x),\phi ^{}(y))+d(\phi ^{}(y),y)+d(y,x)<3ϵ`$. This implies the continuity of the homomorphism $`\phi \phi ^{}`$. ∎
###### Proposition 3.10 (, \[31, Lemma 3.4.10\], \[32, Lemma 5.1.17\], \[14, Section 3.11$`\frac{2}{3}_+`$\]).
The completion of any Urysohn metric space is Urysohn.
###### Proof.
Let $`(M,d)`$ be a complete metric space containing a dense Urysohn subspace $`A`$. We must prove that $`M`$ is Urysohn.
Let $`Y`$ be a finite subset of $`M`$, and let $`fE(Y)`$ be a Katětov function. It suffices to prove that there exists $`zM`$ such that $`d(y,z)=f(y)`$ for every $`yY`$. Pick a sequence $`\{a_n:n\omega \}A`$ such that:
* if $`A_n=\{a_k:kn\}`$ and $`r_n=d(a_{n+1},A_n)`$, $`n=0,1,\mathrm{}`$, then the series $`r_n`$ converges;
* every $`yY`$ is a cluster point of the sequence $`\{a_n:n\omega \}`$.
To construct such a sequence, enumerate $`Y`$ as $`Y=\{y_1,\mathrm{},y_s\}`$, and for every $`k`$ and $`j`$ ($`k\omega `$, $`1js`$) pick a point $`x_k^jA`$ such that $`d(x_k^j,y_j)<2^k`$. Then $`d(x_{k+1}^j,x_k^j)<2^{1k}`$ for every $`k`$ and $`j`$, and the sequence
$$x_0^1,x_0^2,\mathrm{},x_0^s,x_1^1,\mathrm{},x_1^s,x_2^1,\mathrm{}$$
has the required properties.
Let $`g=\kappa _Y(f)E(M)`$. We construct by induction a sequence $`\{z_n:n\omega \}`$ of points of $`A`$ such that:
1. if $`kn`$, then $`d(z_n,a_k)=g(a_k)`$;
2. $`d(z_{n+1},z_n)2r_n`$ for every $`n\omega `$.
Pick $`z_0A`$ so that $`d(z_0,a_0)=g(a_0)`$. This is possible since $`A`$ is Urysohn. Suppose that the points $`z_0,\mathrm{},z_n`$ have been constructed so that the conditions 1 and 2 are satisfied. Consider two Katětov functions $`f_n`$ and $`g_n`$ on the set $`A_{n+1}=\{a_k:kn+1\}`$: let $`f_n(x)=d(z_n,x)`$ for every $`xA_{n+1}`$, and let $`g_n=g|_{A_{n+1}}`$. By (1), the functions $`f_n`$ and $`g_n`$ coincide on $`A_n`$, hence the distance between them in the space $`E(A_{n+1})`$ is equal to
$$\begin{array}{cc}& |f_n(a_{n+1})g_n(a_{n+1})|=sup\{|f_n(a_{n+1})f_n(x)g_n(a_{n+1})+g_n(x)|:xA_n\}\hfill \\ & sup\{|f_n(a_{n+1})f_n(x)|:xA_n\}+sup\{|g_n(a_{n+1})g_n(x)|:xA_n\}\hfill \\ & 2d(a_{n+1},A_n)=2r_n.\hfill \end{array}$$
Let $`X_n`$ be the metric space $`A_{n+1}\{f_n\}`$, considered as a subspace of $`E(A_{n+1})`$. In virtue of Lemma 3.3, the map of $`X_n`$ to $`A`$ which leaves each point of $`A_{n+1}`$ fixed and sends $`f_n`$ to $`z_n`$ is an isometric embedding. Since $`A`$ is Urysohn, this map can be extended to an isometric embedding of $`X_n\{g_n\}`$ to $`A`$. Let $`z_{n+1}`$ be the image of $`g_n`$. Then $`d(z_{n+1},z_n)=d_{A_{n+1}}^E(g_n,f_n)2r_n`$. In virtue of Lemma 3.3, for every $`kn+1`$ we have $`d(z_{n+1},a_k)=g_n(a_k)=g(a_k)`$. Thus the conditions 1 and 2 are satisfied, and the construction is complete.
Since the series $`r_n`$ converges, it follows from (2) that the sequence $`\{z_n:n\omega \}`$ is Cauchy and hence has a limit in the complete space $`M`$. Let $`z=limz_n`$. By (1), we have $`d(z,a_k)=g(a_k)`$ for every $`k\omega `$. Since $`Y`$ is contained in the closure of the set $`\{a_n:n\omega \}`$, it follows that $`d(z,y)=g(y)=f(y)`$ for every $`yY`$. ∎
###### Proof of Theorem 3.2.
Let $`M`$ be a metric space of diameter $`1`$, and let $`M_\omega `$ be the Urysohn extension of $`M`$ constructed above. Consider the completion $`L`$ of $`M_\omega `$. Proposition 3.10 implies that $`L`$ is Urysohn. Proposition 3.8 shows that $`w(L)=w(M)`$. Finally, $`M`$ is $`g`$-embedded in $`M_\omega `$ (Proposition 3.7) and $`M_\omega `$ is $`g`$-embedded in $`L`$ (Proposition 3.9), so $`M`$ is $`g`$-embedded in $`L`$. Thus $`L`$ has the properties required by Theorem 3.2. ∎
## 4. Graev metrics and $`\omega `$-homogeneous extensions
In this section we prove the following:
###### Theorem 4.1.
Every metric space can be $`g`$-embedded into an $`\omega `$-homogeneous metric space of the same weight and the same diameter.
The proof is based on the construction of Graev metrics described in Section 2. We apply this construction to metric spaces of relations. A relation on a set $`X`$ is a subset of $`X^2`$. If $`R`$ and $`S`$ are relations on $`X`$, then the composition $`RS`$ (or simply $`RS`$) is defined by $`RS=\{(x,y):z((x,z)S\text{ and }(z,y)R)\}`$. The inverse relation $`R^1`$ is defined by $`R^1=\{(x,y):(y,x)R\}`$. The set of all relations on a set $`X`$ is a semigroup with an involution: the multiplication is given by the composition, and the involution is given by the map $`RR^1`$. The unity of this semigroup is the diagonal $`\mathrm{\Delta }`$ of $`X^2`$.
We use the notation of Section 2. In particular, if $`k`$ is a non-expanding function $`0`$ on a metric space $`(X,d)`$, then $`Gr(d,k)`$ is the Graev pseudometric on the free group $`F(X)`$. We consider the group $`F(X)`$ as a subset of the free monoid $`S(X)`$ on the set $`XX^1`$.
###### Proof of Theorem 4.1.
Let $`(M,d)`$ be a metric space. We first construct a $`g`$-embedding of $`M`$ into a metric space $`M^{}`$ such that $`w(M^{})=w(M)`$ and every isometry between finite subsets of $`M`$ extends to an isometry of $`M^{}`$.
For every isometry $`f:AB`$ between finite non-empty subsets of $`M`$ consider the graph $`R=\{(a,f(a)):aA\}`$ of $`f`$, and let $`\mathrm{\Gamma }`$ be the set of all such graphs. Thus a non-empty finite subset $`RM^2`$ is an element of $`\mathrm{\Gamma }`$ iff for any two pairs $`(x_1,y_1),(x_2,y_2)R`$ we have $`d(x_1,x_2)=d(y_1,y_2)`$. Equip $`M^2`$ with the metric $`d_2`$ defined by $`d_2((x_1,y_1),(x_2,y_2))=d(x_1,x_2)+d(y_1,y_2)`$, and let $`d_H`$ be the corresponding Hausdorff metric on the set of finite subsets of $`M^2`$. If $`R`$ and $`S`$ are two non-empty finite subsets of $`M^2`$ and $`a0`$, then $`d_H(R,S)a`$ iff for every $`pR`$ there exists $`qS`$ such that $`d_2(p,q)a`$, and for every $`pS`$ there exists $`qR`$ such that $`d_2(p,q)a`$.
Let $`k`$ be the non-expanding function on $`(\mathrm{\Gamma },d_H)`$ defined by $`k(R)=\mathrm{max}\{d(x,y):(x,y)R\}`$. Let $`G`$ be the free group on $`\mathrm{\Gamma }`$, equipped with the Graev pseudometric $`D=Gr(d_H,k)`$. To avoid confusion of multiplication in $`G`$ with composition of relations, we assign to each $`R\mathrm{\Gamma }`$ a symbol $`t_R`$, and consider elements of $`G`$ as irreducible words of the form $`x_1^{ϵ_1}\mathrm{}x_n^{ϵ_n}`$, where $`x_i=t_{R_i}`$. Similarly, we consider elements of the semigroup $`S(\mathrm{\Gamma })`$ as words of the same form. Let $`\mathrm{\Delta }=\{(x,x):xM\}`$ be the diagonal of $`M^2`$. The set $`\mathrm{\Gamma }^{}=\mathrm{\Gamma }\{\mathrm{}\}\{\mathrm{\Delta }\}`$ is a symmetrical subsemigroup of the semigroup of all relations on $`M`$. Let $`\mathrm{\Phi }:G\mathrm{\Gamma }^{}`$ be the map defined by the following rule: if $`w=t_{R_1}^{ϵ_1}\mathrm{}t_{R_n}^{ϵ_n}G`$ is a non-empty irreducible word, then $`\mathrm{\Phi }(w)=R_1^{ϵ_1}\mathrm{}R_n^{ϵ_n}`$. If $`a,bM`$, then $`(a,b)\mathrm{\Phi }(w)`$ iff there exists a chain $`c_0=b,c_1,\mathrm{},c_n=a`$ of points of $`M`$ such that for every $`i=1,\mathrm{},n`$ we have either $`ϵ_i=1`$ and $`(c_i,c_{i1})R_i`$ or $`ϵ_i=1`$ and $`(c_{i1},c_i)R_i`$. For the empty word $`e_GG`$ we put $`\mathrm{\Phi }(e_G)=\mathrm{\Delta }`$.
Note that the definition of $`\mathrm{\Phi }(w)`$ makes sense also without the assumption that the word $`w`$ be irreducible, so we can assume that $`\mathrm{\Phi }`$ is defined on the set $`S(\mathrm{\Gamma })`$ of all words of the form $`t_{R_1}^{ϵ_1}\mathrm{}t_{R_n}^{ϵ_n}`$. Recall that $`w_1|w_2`$ denotes the word obtained by writing $`w_2`$ after $`w_1`$ (without cancelations). We have $`\mathrm{\Phi }(w_1|w_2)=\mathrm{\Phi }(w_1)\mathrm{\Phi }(w_2)`$.
###### Lemma 4.2.
If $`wS(\mathrm{\Gamma })`$ and $`u`$ is the irreducible word equivalent to $`w`$, then $`\mathrm{\Phi }(u)\mathrm{\Phi }(w)`$.
###### Proof.
It suffices to prove that $`\mathrm{\Phi }(w^{})\mathrm{\Phi }(w)`$ if $`w^{}`$ is obtained from $`w`$ by canceling one pair of letters. Let $`w=u|t_R^ϵt_R^ϵ|v`$ and $`w^{}=u|v`$. Since $`R^ϵ`$ is a functional relation, we have $`R^ϵR^ϵ\mathrm{\Delta }`$ and hence $`\mathrm{\Phi }(w^{})=\mathrm{\Phi }(u)\mathrm{\Phi }(v)=\mathrm{\Phi }(u)\mathrm{\Delta }\mathrm{\Phi }(v)\mathrm{\Phi }(u)R^ϵR^ϵ\mathrm{\Phi }(v)=\mathrm{\Phi }(w)`$. ∎
For every $`wG`$ we have $`\mathrm{\Phi }(w^1)=\mathrm{\Phi }(w)^1`$. We claim that $`\mathrm{\Phi }(w_1w_2)\mathrm{\Phi }(w_1)\mathrm{\Phi }(w_2)`$ for every $`w_1,w_2G`$. Indeed, the product $`w_1w_2G`$ is the irreducible word equivalent to $`w_1|w_2`$, therefore $`\mathrm{\Phi }(w_1w_2)\mathrm{\Phi }(w_1|w_2)=\mathrm{\Phi }(w_1)\mathrm{\Phi }(w_2)`$ by Lemma 4.2.
For every $`a,bM`$ let $`H_{a,b}G`$ be the set of all $`wG`$ such that $`(a,b)\mathrm{\Phi }(w)`$. We claim that $`H_{a,b}^1=H_{b,a}`$ and $`H_{b,c}H_{a,b}H_{a,c}`$ for every $`a,b,cM`$. This follows from the properties of $`\mathrm{\Phi }`$ established in the preceding paragraph. Indeed, pick $`w_1H_{b,c}`$ and $`w_2H_{a,b}`$. Then $`(a,b)\mathrm{\Phi }(w_2)`$ and $`(b,c)\mathrm{\Phi }(w_1)`$, hence $`(a,c)\mathrm{\Phi }(w_1)\mathrm{\Phi }(w_2)\mathrm{\Phi }(w_1w_2)`$ and $`w_1w_2H_{a,c}`$. This proves the inclusion $`H_{b,c}H_{a,b}H_{a,c}`$. The equality $`H_{a,b}^1=H_{b,a}`$ is proved similarly.
Note that $`t_RH_{a,b}`$ if and only if $`(a,b)R`$, since $`\mathrm{\Phi }(t_R)=R`$. Note also that $`e_GH_{a,b}`$ if and only if $`a=b`$, since $`\mathrm{\Phi }(e_G)=\mathrm{\Delta }`$.
Consider the following equivalence relation $``$ on $`G\times M`$: a pair $`(g,a)`$ is equivalent to a pair $`(h,b)`$ iff $`h^1gH_{a,b}`$. Since $`e_GH_{a,a}`$, $`H_{a,b}^1=H_{b,a}`$ and $`H_{b,c}H_{a,b}H_{a,c}`$ for all $`a,b,cM`$, the relation $``$ is reflexive, symmetric and transitive and thus is indeed an equivalence relation. Let $`L`$ be the quotient set $`G\times M/`$. The group $`G`$ acts on $`G\times M`$ by the rule $`g(h,a)=(gh,a)`$. The relation $``$ is invariant under this action, so there is a uniquely defined left action of $`G`$ on $`L`$ which makes the canonical map $`G\times ML`$ into a morphism of $`G`$-sets. Let $`i:ML`$ be the map which sends each point $`aM`$ to the class of the pair $`(1,a)`$. If $`ab`$, then the pairs $`(1,a)`$ and $`(1,b)`$ are not equivalent, since $`e_GH_{a,b}`$. The map $`i`$ is therefore injective, and we can consider $`M`$ as a subspace of $`L`$, identifying $`M`$ with $`i(M)`$. Every $`xL`$ can be written in the form $`x=ga`$ (or simply $`x=ga`$), where $`gG`$ and $`aM`$.
Let $`a,bM`$. The set of all $`gG`$ such that $`ga=b`$ is equal to $`H_{a,b}`$. If $`R\mathrm{\Gamma }`$ is a relation containing the pair $`(a,b)`$, then $`t_RH_{a,b}`$ and hence $`t_Ra=b`$. It follows that the action of $`G`$ on $`L`$ is transitive. Moreover, for every isometry $`f:AB`$ between finite subsets of $`M`$ there exists $`gG`$ such that the self-map $`xgx`$ of $`L`$ extends $`f`$. Indeed, if $`R\mathrm{\Gamma }`$ is the graph of $`f`$, then $`t_RH_{a,f(a)}`$ and hence $`t_Ra=f(a)`$ for every $`aA`$. Thus $`g=t_R`$ has the required property.
We now define a $`G`$-invariant pseudometric $`\nu `$ on $`L`$ which extends the metric $`d`$ on $`M`$. Let $`p`$ be the Graev seminorm on $`G`$ corresponding to the pseudometric $`D=Gr(d_H,k)`$. We have $`p(w)=D(w,e_G)`$ for every $`wG`$. For every $`x,yL`$ let
$$\nu (x,y)=inf\{p(g):gG,gx=y\}.$$
Then $`\nu `$ is a pseudometric on $`L`$. Since the seminorm $`p`$ is invariant under inner automorphisms, the pseudometric $`\nu `$ is $`G`$-invariant. Indeed, for $`x,yL`$ and $`hG`$ we have $`\nu (hx,hy)=inf\{p(g):ghx=hy\}=inf\{p(h^1gh):h^1ghx=y\}=inf\{p(g^{}):g^{}x=y\}=\nu (x,y)`$. We claim that $`\nu `$ extends the metric $`d`$ on $`M`$: $`d(a,b)=\nu (a,b)`$ for every $`a,bM`$. Since for $`wG`$ the condition $`wa=b`$ is equivalent to $`wH_{a,b}`$, we have $`\nu (a,b)=inf\{p(w):wH_{a,b}\}`$. If $`R=\{(a,b)\}`$, then $`t_RH_{a,b}`$ and $`p(t_R)=k(R)=d(a,b)`$. It follows that $`\nu (a,b)d(a,b)`$. It remains to prove the opposite inequality, which is equivalent to the following assertion:
###### Lemma 4.3.
If $`a,bM`$ and $`wH_{a,b}`$, then $`p(w)d(a,b)`$.
###### Proof.
Let $`w=t_{R_1}^{ϵ_1}\mathrm{}t_{R_n}^{ϵ_n}`$. We argue by induction on $`n`$, the length of $`w`$. If $`n=0`$, then $`w=e_G`$, and we noted that $`e_GH_{a,b}`$ implies $`a=b`$. If $`n=1`$, then $`w=t_R^ϵ`$ and $`p(w)=k(R)`$. Since $`wH_{a,b}`$, the relation $`R`$ contains either $`(a,b)`$ or $`(b,a)`$ and hence $`p(w)=k(R)d(a,b)`$. Assume that $`n>1`$. It suffices to show that there exists $`uH_{a,b}`$ of length $`<n`$ such that $`p(u)p(w)`$.
We use the construction of the Graev seminorm $`p`$ described in Section 2. Let $`E`$ be a $`w`$-pairing for which $`p(w)`$ is attained. In other words, $`E`$ is a disjoint system of two-element subsets of the set $`J=\{1,\mathrm{},n\}`$ such that for the Graev sum
$$s_E=\{d_H(R_i,R_j):\{i,j\}E,i<j\}+\{k(R_i):iJE\}$$
we have $`p(w)=s_E`$. Considering the pair $`(i,j)E`$ with the least possible value of $`|ij|`$ (“the shortest arc”), we see that at least one of the following three cases must occur:
1. there exists an $`i`$ such that $`\{i,i+1\}E`$;
2. there exists an $`i`$ such that $`\{i,i+2\}E`$ and $`i+1JE`$;
3. there exists an $`i`$ such that $`i,i+1JE`$.
In cases (1) or (3) we replace the subword $`t_{R_i}^{ϵ_i}t_{R_{i+1}}^{ϵ_{i+1}}`$ of $`w`$ by the letter $`t_S`$, where $`S=R_i^{ϵ_i}R_{i+1}^{ϵ_{i+1}}`$. In case (2) we replace the subword $`t_{R_i}^{ϵ_i}t_{R_{i+1}}^{ϵ_{i+1}}t_{R_{i+2}}^{ϵ_{i+2}}`$ of $`w`$ by the letter $`t_S`$, where $`S=R_i^{ϵ_i}R_{i+1}^{ϵ_{i+1}}R_{i+2}^{ϵ_{i+2}}`$. In all cases we get a word $`w^{}`$ of length $`<n`$. To justify the usage of the symbol $`t_S`$, we must show that $`S\mathrm{\Gamma }`$, which reduces to the fact that $`S\mathrm{}`$. Had $`S`$ been empty, the same would have been true for $`\mathrm{\Phi }(w)=R_1^{ϵ_1}\mathrm{}R_n^{ϵ_n}`$. On the other hand, since $`wH_{a,b}`$, we have $`(a,b)\mathrm{\Phi }(w)\mathrm{}`$.
Let $`uG`$ be the irreducible word equivalent to $`w^{}`$. The length of $`u`$ is less than $`n`$. We show that $`uH_{a,b}`$ and $`p(u)p(w)`$.
By Lemma 4.2 we have $`\mathrm{\Phi }(w^{})\mathrm{\Phi }(u)`$. Plainly $`\mathrm{\Phi }(w)=R_1^{ϵ_1}\mathrm{}R_n^{ϵ_n}=\mathrm{\Phi }(w^{})`$. Since $`wH_{a,b}`$, we have $`(a,b)\mathrm{\Phi }(w)`$. Thus $`(a,b)\mathrm{\Phi }(w)=\mathrm{\Phi }(w^{})\mathrm{\Phi }(u)`$ and $`uH_{a,b}`$, as required.
We prove that $`p(u)p(w)`$. As in Section 2, we define $`p(w^{})`$ even if the word $`w^{}`$ is reducible, and we have $`p(u)=p(w^{})=infs_E^{}`$, where $`E^{}`$ runs over the set of all $`w^{}`$-pairings. The $`w`$-pairing $`E`$ in an obvious way yields a $`w^{}`$-pairing $`E^{}`$, which coincides with $`E`$ outside the changed part of $`w`$ and leaves the new letter $`t_S`$ unpaired. The Graev sums $`s_E`$ and $`s_E^{}`$ differ only by the term $`k(S)`$ in the sum $`s_E^{}`$ and the terms $`d_H(R_i,R_{i+1})`$ (case 1) or $`d_H(R_i,R_{i+2})+k(R_{i+1})`$ (case 2) or $`k(R_i)+k(R_{i+1})`$ (case 3) in the sum $`s_E`$. According to Lemma 4.4 below, we have $`s_E^{}s_E`$. Thus $`p(u)=p(w^{})s_E^{}s_E=p(w)`$. ∎
###### Lemma 4.4.
Let $`ϵ,\delta \{1,1\}`$.
1. If $`R_1,R_2\mathrm{\Gamma }`$ and $`S=R_1^ϵR_2^ϵ`$ is non-empty, then $`k(S)d_H(R_1,R_2)`$;
2. if $`R_1,R_2,R_3\mathrm{\Gamma }`$ and $`S=R_1^ϵR_2^\delta R_3^ϵ`$ is non-empty, then $`k(S)d_H(R_1,R_3)+k(R_2)`$;
3. if $`R_1,R_2\mathrm{\Gamma }`$ and $`S=R_1^ϵR_2^\delta `$ is non-empty, then $`k(S)k(R_1)+k(R_2)`$.
###### Proof.
Since $`k(R)=k(R^1)`$ and $`d_H(R,T)=d_H(R^1,T^1)`$ for every $`R,T\mathrm{\Gamma }`$, we may assume that $`ϵ=\delta =1`$. Pick $`(a,b)S`$ so that $`k(S)=d(a,b)`$. Case (1) follows from (2) (take for $`R_2`$ in (2) a sufficiently large finite part of $`\mathrm{\Delta }`$), so let us consider case (2). There exist $`x,yM`$ such that $`(a,x)R_3^1`$, $`(x,y)R_2`$ and $`(y,b)R_1`$. Since $`(x,a)R_3`$, there exists a pair $`(u,v)R_1`$ such that $`d(a,v)+d(u,x)d_H(R_1,R_3)`$. The relation $`R_1`$, being an element of $`\mathrm{\Gamma }`$, is the graph of a partial isometry, so from $`(y,b)R_1`$ and $`(u,v)R_1`$ it follows that $`d(v,b)=d(u,y)`$. Note that $`d(x,y)k(R_2)`$. Thus we have $`k(S)=d(a,b)d(a,v)+d(v,b)=d(a,v)+d(u,y)d(a,v)+d(u,x)+d(x,y)d_H(R_1,R_3)+k(R_2)`$, as required. Case (3) is easy: there exists a point $`cM`$ such that $`(a,c)R_2`$ and $`(c,b)R_1`$, hence $`k(S)=d(a,b)d(a,c)+d(c,b)k(R_2)+k(R_1)`$. ∎
We have thus proved that the pseudometric $`\nu `$ on $`L`$ extends the metric $`d`$ on $`M`$. Let $`(M^{},d^{})`$ be the metric space associated with the pseudometric space $`(L,\nu )`$. The metric space $`(M,d)`$ can be naturally identified with a subspace of $`(M^{},d^{})`$. We show that $`M`$ is $`g`$-embedded in $`M^{}`$.
In virtue of the functorial nature of the construction of $`M^{}`$, every isometry $`\phi `$ of $`M`$ naturally extends to an isometry $`\phi ^{}`$ of $`M^{}`$. The map $`\phi \phi ^{}`$ from $`\mathrm{Iso}(M)`$ to $`\mathrm{Iso}(M^{})`$ is a homomorphism of groups. We claim that this homomorphism is continuous. This follows from the fact that at each step of our construction new spaces are obtained from the old ones via functors “with finite support”: every element of $`\mathrm{\Gamma }`$ is a finite relation on $`M`$, and every word $`wG`$ involves only finitely many elements of $`\mathrm{\Gamma }`$. Given an isometry $`\phi \mathrm{Iso}(M)`$, the isometry $`\phi ^{}\mathrm{Iso}(M^{})`$ can be obtained step by step in the following way. First we consider the isometry $`\phi _1`$ of the metric space $`(\mathrm{\Gamma },d_H)`$ corresponding to $`\phi `$; the isometry $`\phi _1`$ preserves the function $`k`$ on $`\mathrm{\Gamma }`$ and gives rise to the automorphism $`\phi _2`$ of the group $`G=F(\mathrm{\Gamma })`$ which preserves the Graev pseudometric $`D`$; then we get the isometry $`\phi _3`$ of $`L`$ which maps the class of each pair $`(g,x)`$ ($`gG`$, $`xM`$) to the class of the pair $`(\phi _2(g),\phi (x))`$; finally we get the isometry $`\phi _4=\phi ^{}`$ of $`M^{}`$. We show step by step that $`\phi _i`$ depends continuously on $`\phi `$. For $`i=1`$ this is straightforward: use the fact that $`\mathrm{\Gamma }`$ consists of finite subsets of $`M^2`$. For $`i=2`$ apply Lemma 2.2 with $`X=\mathrm{\Gamma }`$. Let us consider the case $`i=3`$. Pick a point $`x=gaL`$ ($`gG`$, $`aM`$). It suffices to check that $`\nu (\phi _3(x),x)`$ is small if $`\phi `$ is close to the identity. We have $`\nu (\phi _3(x),x)=\nu (\phi _2(g)\phi (a),ga)\nu (\phi _2(g)\phi (a),g\phi (a))+\nu (g\phi (a),ga)=\nu (g^1\phi _2(g)\phi (a),\phi (a))+\nu (\phi (a),a)`$. By the definition of $`\nu `$, the first term of the last sum does not exceed $`p(g^1\phi _2(g))=D(\phi _2(g),g)`$ and hence is arbitrarily small if $`\phi `$ is close enough to the identity. The same is true for second term, and we are done. Finally, $`\phi _4`$ is the image of $`\phi _3`$ under the natural morphism $`\mathrm{Iso}(L)\mathrm{Iso}(M^{})`$, and the case $`i=4`$ follows.
We have thus proved that $`M`$ is $`g`$-embedded in $`M^{}`$. We saw that each isometry between finite subsets of $`M`$ extends to an isometry of $`L`$ and hence also to an isometry of $`M^{}`$. It is easy to see that $`w(M^{})=w(M)`$. If the diameter $`C`$ of $`M`$ is finite, replace the metric $`d^{}`$ of $`M^{}`$ by $`inf(d^{},C)`$. This operation can make the group $`\mathrm{Iso}(M^{})`$ only larger, and the diameter of $`M^{}`$ becomes equal to that of $`M`$.
To finish the proof of Theorem 4.1, iterate the construction of $`M^{}`$. We get an increasing chain $`M_0=MM_1=M^{}M_2=M_1^{}\mathrm{}`$ of metric spaces such that each $`M_n`$ is $`g`$-embedded in $`M_{n+1}`$, every isometry between finite subsets of $`M_n`$ extends to an isometry of $`M_{n+1}`$, $`w(M_n)=w(M)`$ and $`\mathrm{diam}M_n=\mathrm{diam}M`$, $`n=0,1,\mathrm{}`$. Consider the space $`M_\omega =_{n\omega }M_n`$. We have $`w(M_\omega )=w(M)`$ and $`\mathrm{diam}M_\omega =\mathrm{diam}M`$. In virtue of Proposition 3.6, each $`M_n`$ is $`g`$-embedded in $`M_\omega `$. Since every finite subset of $`M_\omega `$ is contained in some $`M_n`$, it is clear that $`M_\omega `$ is $`\omega `$-homogeneous. ∎
Remarks.
1. If $`a,bM`$ are distinct and $`S=\{(b,b)\}`$, the pairs $`(1,a)`$ and $`(t_S,a)`$ represent distinct points of $`L`$ that have the same image in $`M^{}`$. Early versions of this paper contained the false statement that $`\nu `$ itself is a metric and $`M^{}=L`$. I am indebted to the referee for catching this error.
2. The referee raised the question whether the methods of this section could be used to prove the following result by S. Solecki and A.M. Vershik that extends an earlier result by Hrushovski: for every finite metric space $`A`$ there exists another finite metric space $`A^{}`$ containing $`A`$ such that all partial isometries<sup>6</sup><sup>6</sup>6A partial isometry of $`A`$ is an isometry between two subsets of $`A`$. of $`A`$ extend to isometries of $`A^{}`$. I do not know the answer. A partial answer is provided by Pestov’s paper where the Hrushovski–Solecki–Vershik theorem is proved with the aid of pseudometrics on groups, and the notion of a residually finite group is used to construct isometric embeddings of finite metric spaces into finite metric groups. A similar technique was used in .
## 5. Proof of Theorem 1.7
In this section we prove Theorem 1.7.
###### Theorem 5.1.
Let $`M`$ be a metric space of diameter $`1`$. There exists a complete $`\omega `$-homogeneous Urysohn metric space $`L`$ containing $`M`$ as a subspace such that $`w(L)=w(M)`$ and $`M`$ is $`g`$-embedded in $`L`$.
###### Proof.
Consider two cases.
Case 1: $`M`$ is separable. According to Theorem 3.2, there exists a complete separable Urysohn space $`L`$ such that $`M`$ is a $`g`$-embedded subspace of $`L`$. According to Proposition 1.6, $`L=𝕌_1`$ is $`\omega `$-homogeneous.
Case 2: $`M`$ is not separable. Let $`\tau =w(M)`$. Applying in turn Theorem 3.2 and Theorem 4.1, construct an increasing continuous chain $`\{M_\alpha :\alpha \omega _1\}`$ of metric spaces of weight $`\tau `$ and diameter $`1`$ such that $`M_0=M`$, each $`M_\alpha `$ is $`g`$-embedded in $`M_{\alpha +1}`$ ($`\alpha <\omega _1`$), and $`M_{\alpha +1}`$ is complete Urysohn for $`\alpha `$ even and $`\omega `$-homogeneous for $`\alpha `$ odd. Let $`L=M_{\omega _1}=_{\alpha <\omega _1}M_\alpha `$. Proposition 3.6 implies that each $`M_\alpha `$ is $`g`$-embedded in $`L`$. The space $`L`$ is Urysohn, being the union of the increasing chain $`\{M_{2\alpha +1}:\alpha <\omega _1\}`$ of Urysohn spaces. For similar reasons the space $`L`$ is $`\omega `$-homogeneous. Finally, since every countable subset of $`L`$ is contained in some $`M_\alpha `$, $`\alpha <\omega _1`$, and all spaces $`M_{2\alpha +1}`$ are complete, every Cauchy sequence in $`L`$ converges, which means that $`L`$ is complete. Thus $`L`$ has the properties required by Theorem 5.1. ∎
###### Proof of Theorem 1.7.
Let $`G`$ be a topological group. According to Theorem 2.1, there exists a metric space $`(M,d)`$ such that $`w(M)=w(G)`$ and $`G`$ is isomorphic to a subgroup of $`\mathrm{Iso}(M)`$. We may assume that $`M`$ has diameter $`1`$: otherwise replace the metric $`d`$ by $`inf(d,1)`$. Theorem 5.1 implies that there exists a complete $`\omega `$-homogeneous Urysohn metric space $`L`$ such that $`w(L)=w(M)`$ and $`\mathrm{Iso}(M)`$ is isomorphic to a subgroup of $`\mathrm{Iso}(L)`$. Then $`w(L)=w(G)`$ and $`G`$ is isomorphic to a subgroup of $`\mathrm{Iso}(L)`$, as required. ∎
## 6. Semigroups of bi-Katětov functions
Let $`(M,d)`$ be a complete metric space of diameter $`1`$.
###### Definition 6.1.
A function $`f:M\times MI=[0,1]`$ is bi-Katětov if for every $`xM`$ the functions $`f(x,)`$ and $`f(,x)`$ on $`M`$ are Katětov (see Section 3).
Thus a function $`f:M^2I`$ is bi-Katětov if and only if for every $`x,y,zM`$ we have
$$|f(x,y)f(x,z)|d(y,z)f(x,y)+f(x,z),$$
$$|f(y,x)f(z,x)|d(y,z)f(y,x)+f(z,x).$$
Let $`\mathrm{\Theta }`$ be the compact space of all bi-Katětov functions on $`M^2`$, equipped with the topology of pointwise convergence. In the next section we shall prove that the Roelcke completion of the group $`\mathrm{Iso}(M)`$ can be identified with $`\mathrm{\Theta }`$, provided that the complete metric space $`M`$ is Urysohn and $`\omega `$-homogeneous. In the present section we study the structure of an ordered semigroup with an involution on $`\mathrm{\Theta }`$.
Recall that we defined in Section 1 an associative operation $``$ on the set $`S=I^{M\times M}`$. If $`f,gS`$ and $`x,yM`$, then
$$fg(x,y)=inf\{f(x,z)g(z,y):zM\}.$$
The involution $`ff^{}`$ on $`S`$ is defined by $`f^{}(x,y)=f(y,x)`$. Every idempotent in $`S`$ satisfies the triangle inequality. If $`fS`$ is zero on the diagonal of $`M^2`$, then $`f`$ is an idempotent in $`S`$ if and only if $`f`$ satisfies the triangle inequality. A function $`fS`$ is a pseudometric on $`X`$ if and only if $`f`$ is zero on the diagonal and $`f`$ is a symmetrical idempotent. In particular, we have $`d=d^{}=dd`$.
The semigroup $`S`$ has a natural partial order: for $`p,qS`$ the inequality $`pq`$ means that $`p(x,y)q(x,y)`$ for all $`x,yM`$. This partial order is compatible with the semigroup structure: if $`p_1p_2`$ and $`q_1q_2`$, then $`p_1q_1p_2q_2`$.
It is clear that the set $`\mathrm{\Theta }`$ of all bi-Katětov functions is closed under the involution. It is easy to verify that $`\mathrm{\Theta }`$ also is closed under the operation $``$. This fact also can be deduced from the following proposition:
###### Proposition 6.2.
A function $`f:M^2I`$ is bi-Katětov if and only if
$$fd=df=f,f^{}fd,ff^{}d,$$
where $`d`$ is the metric on $`M`$.
###### Proof.
The condition $`fd=f`$ (respectively, $`df=f`$) holds if and only if the function $`f(x,)`$ (respectively, $`f(,x)`$) is non-expanding for every $`xX`$. The condition $`f^{}fd`$ (respectively, $`ff^{}d`$) holds if and only if $`d(y,z)f(x,y)+f(x,z)`$ (respectively, $`d(y,z)f(y,x)+f(z,x)`$) for all $`x,y,zX`$. ∎
Let $`S`$ be any ordered semigroup with an involution, and let $`dS`$ be a symmetrical idempotent. The set $`S_d`$ of all $`xS`$ such that
$$xd=dx=x,x^{}xd,xx^{}d$$
is closed under the multiplication and under the involution and can be considered as a semigroup with the unity $`d`$. Indeed, we have $`dS_d`$ since $`d=d^{}=d^2`$, and it is clear that $`d`$ is the unity of $`S_d`$. If $`x,yS_d`$, then $`xyd=xy=dxy`$ and $`(xy)^{}xy=y^{}x^{}xyy^{}dy=y^{}yd`$; similarly, $`xy(xy)^{}d`$ and hence $`xyS_d`$. Thus $`S_d`$ is a semigroup. If $`xS_d`$, then $`x^{}d=x^{}d^{}=(dx)^{}=x^{}`$ and similarly $`dx^{}=x^{}`$. It follows that $`S_d`$ is symmetrical.
The arguments of the preceding paragraph and Proposition 6.2 show that $`\mathrm{\Theta }`$ is a semigroup with the unity $`d`$. In general, the operation $`(f,g)fg`$ need not be continuous (not even continuous on the left or on the right).
###### Proposition 6.3.
Let $`S`$ be a closed subsemigroup of $`\mathrm{\Theta }`$, and let $`T`$ be the set of all $`fS`$ such that $`fd`$. If $`T\mathrm{}`$, then $`T`$ has a greatest element $`p`$, and $`p`$ is an idempotent.
###### Proof.
We claim that every non-empty closed subset of $`\mathrm{\Theta }`$ has a maximal element. Indeed, if $`C`$ is a non-empty linearly ordered subset of $`\mathrm{\Theta }`$, then $`C`$ has a least upper bound $`b`$ in $`\mathrm{\Theta }`$, and $`b`$ belongs to the closure of $`C`$. Thus our claim follows from Zorn’s lemma.
The set $`T`$ is a closed subsemigroup of $`\mathrm{\Theta }`$. Let $`p`$ be a maximal element of $`T`$. For every $`qT`$ we have $`pqpd=p`$, whence $`pq=p`$. It follows that $`p`$ is idempotent and that $`p=pqdq=q`$. Thus $`p`$ is the greatest element of $`T`$. ∎
We now describe all idempotents in $`\mathrm{\Theta }`$ which are $`d`$. For every closed non-empty subset $`F`$ of $`M`$ let $`b_F\mathrm{\Theta }`$ be the bi-Katětov function defined by $`b_F(x,y)=inf\{d(x,z)d(z,y):zF\}`$. If $`F=\mathrm{}`$, let $`b_F=1`$, that is the function on $`M^2`$ which is identically equal to 1. (Note that 1 is not the unity of $`\mathrm{\Theta }`$; on the contrary, $`f1=1f=1`$ for every $`fI^{M\times M}`$, so 1 might be called a zero element of $`\mathrm{\Theta }`$.)
###### Proposition 6.4.
If $`F`$ is a closed subset of $`M`$, then $`b_F`$ is an idempotent $`d`$ in $`\mathrm{\Theta }`$, and every idempotent $`d`$ in $`\mathrm{\Theta }`$ is equal to $`b_F`$ for some closed $`FM`$.
###### Proof.
Let $`F`$ be a closed subset of $`M`$. It is clear that $`b_Fd`$. If $`F\mathrm{}`$, then $`b_Fb_F(x,y)=inf\{d(x,z_1)d(z_1,u)d(u,z_2)d(z_2,y):uM,z_1,z_2F\}=inf\{d(x,z)d(z,y):zF\}=b_F(x,y)`$ for every $`x,yM`$. Thus $`b_F`$ is an idempotent. The same is obviously true if $`F=\mathrm{}`$.
Conversely, let $`p`$ be an idempotent in $`\mathrm{\Theta }`$ such that $`pd`$. Let $`F=\{xM:p(x,x)=0\}`$. The function $`p:M^2I`$, being non-expanding in each argument, is continuous, hence $`F`$ is closed in $`M`$. We claim that $`p=b_F`$.
We first show that $`pb_F`$. This is evident if $`F=\mathrm{}`$, so assume that $`F\mathrm{}`$. For every $`x,y,zM`$ we have $`p(x,y)d(x,z)+p(z,y)d(x,z)+d(z,y)+p(z,z)`$, since the functions $`p(,y)`$ and $`p(z,)`$ are non-expanding. It follows that $`p(x,y)inf(\{d(x,z)+d(z,y)+p(z,z):zF\}\{1\})=b_F(x,y)`$.
We prove that $`b_Fp`$. Fix $`x,yM`$. We must show that $`b_F(x,y)p(x,y)`$. This is evident if $`p(x,y)=1`$, so assume that $`p(x,y)<1`$. Fix $`ϵ>0`$ so that $`p(x,y)+ϵ<1`$. Since $`pp=p`$, for every $`u,vM`$ we have $`p(u,v)=inf(\{p(u,z)+p(z,v):zM\}\{1\})`$. Hence we can construct by induction a sequence of points $`z_1,z_2,\mathrm{}`$ in $`M`$ such that
$`p(x,z_1)+p(z_1,y)`$ $`<p(x,y)+ϵ/2;`$
$`p(z_1,z_2)+p(z_2,y)`$ $`<p(z_1,y)+ϵ/4;`$
$`p(z_2,z_3)+p(z_3,y)`$ $`<p(z_2,y)+ϵ/8;`$
$`\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}`$ $`\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}..`$
$`p(z_n,z_{n+1})+p(z_{n+1},y)`$ $`<p(z_n,y)+ϵ/2^{n+1}`$
$`\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}`$ $`\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}..`$
Adding the first $`n`$ inequalities, we get
$`(\mathrm{I})`$
$$p(x,z_1)+\underset{i=1}{\overset{n1}{}}p(z_i,z_{i+1})+p(z_n,y)<p(x,y)+ϵ.$$
It follows that the series $`_{i=1}^{\mathrm{}}p(z_i,z_{i+1})`$ converges. Since $`dp`$, the series $`_{i=1}^{\mathrm{}}d(z_i,z_{i+1})`$ also converges. This implies that the sequence $`z_1,z_2,\mathrm{}`$ is Cauchy and hence has a limit in $`M`$. Let $`z=limz_i`$. Since the series $`_{i=1}^{\mathrm{}}p(z_i,z_{i+1})`$ converges, we have $`limp(z_i,z_{i+1})=0`$ and hence $`p(z,z)=0`$. Thus $`zF`$.
Since $`p`$ is an idempotent, it satisfies the triangle inequality: $`p(u,v)p(u,w)+p(w,v)`$ for all $`u,v,wM`$. The inequality (I) therefore implies that $`p(x,z_n)+p(z_n,y)<p(x,y)+ϵ`$ for every $`n`$. Passing to the limit, we get $`p(x,z)+p(z,y)p(x,y)+ϵ`$. Thus $`b_F(x,y)d(x,z)+d(z,y)p(x,z)+p(z,y)p(x,y)+ϵ`$. Since $`ϵ`$ was arbitrary, it follows that $`b_F(x,y)p(x,y)`$. ∎
Remark. We shall see later in this section that the elements of $`\mathrm{\Theta }`$ (= bi-Katětov functions on $`M^2`$) admit a geometric interpretation: they correspond to metric spaces covered by two isometric copies of $`M`$. If $`F`$ is a closed subset of $`M`$, the function $`b_F`$ considered above corresponds to the amalgam of two copies of $`M`$ with the copies of $`F`$ amalgamated. This description, together with the geometric description of the operation $``$ on $`\mathrm{\Theta }`$ provided in the last paragraph of this section, makes it obvious that each $`b_F`$ is an idempotent.
Let $`G=\mathrm{Iso}(M)`$. For every isometry $`\phi G`$ let $`i(\phi )\mathrm{\Theta }`$ be the bi-Katětov function defined by $`i(\phi )(x,y)=d(x,\phi (y))`$. It is easy to check that the map $`i:G\mathrm{\Theta }`$ is a homeomorphic embedding. We claim that the embedding $`i:G\mathrm{\Theta }`$ is a morphism of monoids with an involution. This means that $`i(e_G)=d`$, $`i(\phi ^1)=i(\phi )^{}`$ and $`i(\phi \psi )=i(\phi )i(\psi )`$ for all $`\phi ,\psi G`$. The first equality is obvious. For the second, note that $`i(\phi ^1)(x,y)=d(x,\phi ^1(y))=d(y,\phi (x))=i(\phi )(y,x)=i(\phi )^{}(x,y)`$. For the third, note that $`i(\phi \psi )(x,y)=d(x,\phi \psi (y))=inf\{d(x,\phi (z))+d(\phi (z),\phi \psi (y)):zM\}=inf\{d(x,\phi (z))+d(z,\psi (y)):zM\}=inf\{i(\phi )(x,z)+i(\psi )(z,y):zM\}=i(\phi )i(\psi )(x,y)`$.
Thus we can identify $`G`$ with a subgroup of $`\mathrm{\Theta }`$. There are natural left and right actions of $`G`$ on $`\mathrm{\Theta }`$, defined by $`(g,p)gp`$ and $`(g,p)pg`$ ($`gG,p\mathrm{\Theta }`$), respectively.
###### Proposition 6.5.
The maps $`(g,p)gp`$ and $`(g,p)pg`$ from $`G\times \mathrm{\Theta }`$ to $`\mathrm{\Theta }`$ are continuous. If $`p\mathrm{\Theta }`$ and $`x,yM`$, then $`gp(x,y)=p(g^1(x),y)`$ and $`pg(x,y)=p(x,g(y))`$.
###### Proof.
We have $`gp(x,y)=inf\{d(x,g(z))p(z,y):zM\}`$. Taking $`z=g^1(x)`$, we see that the right side is $`p(g^1(x),y)`$. On the other hand, for every $`zM`$ we have $`d(x,g(z))+p(z,y)=d(g^1(x),z)+p(z,y)p(g^1(x),y)`$, whence the opposite inequality. The continuity of the left action easily follows from the explicit formula that we have just proved. The argument for the right action is similar. ∎
Let us show that all invertible elements of $`\mathrm{\Theta }`$ are in $`i(G)`$.
It will be useful to establish a one-to-one correspondence between elements of $`\mathrm{\Theta }`$ and other objects which we call $`M`$-triples. Let $`s=(h_1,h_2,L)`$ be a triple such that $`L`$ is a metric space of diameter $`1`$, $`h_i:ML`$ is an isometric embedding ($`i=1,2`$) and $`L=h_1(M)h_2(M)`$. We say that $`s`$ is an $`M`$-triple. Two $`M`$-triples $`(h_1,h_2,L)`$ and $`(h_1^{},h_2^{},L^{})`$ are isomorphic if there exists an isometry $`g:LL^{}`$ such that $`h_i^{}=gh_i`$, $`i=1,2`$.
Given an $`M`$-triple $`s=(h_1,h_2,L)`$, let $`f_s\mathrm{\Theta }`$ be the bi-Katětov function defined by $`f_s(x,y)=\rho _L(h_1(x),h_2(y))`$, where $`\rho _L`$ is the metric on $`L`$. It is easy to verify that we get in this way a one-to-one correspondence between $`\mathrm{\Theta }`$ and the set of classes of isomorphic $`M`$-triples. The subset $`i(G)`$ of $`\mathrm{\Theta }`$ corresponds to the set of classes of triples $`s=(h_1,h_2,L)`$ such that $`h_1(M)=h_2(M)=L`$. Indeed, if $`\phi G`$, then for the $`M`$-triple $`s=(\text{id}_M,\phi ,M)`$ we have $`f_s=i(\phi )`$. Conversely, every $`M`$-triple $`s=(h_1,h_2,L)`$ such that $`h_1(M)=h_2(M)=L`$ is isomorphic to the triple $`(\text{id}_M,\phi ,M)`$, where $`\phi =h_1^1h_2`$ is an isometry of $`M`$. Thus $`s`$ corresponds to $`\phi G`$.
###### Proposition 6.6.
The set of invertible elements of $`\mathrm{\Theta }`$ coincides with $`i(G)`$.
###### Proof.
Let $`f\mathrm{\Theta }`$ be invertible. Let $`s=(h_1,h_2,L)`$ be an $`M`$-triple corresponding to $`f`$. This means that $`(L,\rho )`$ is a metric space, $`h_1`$ and $`h_2`$ are distance-preserving maps from $`M`$ to $`L`$, $`L=h_1(M)h_2(M)`$ and $`f(x,y)=\rho (h_1(x),h_2(y))`$ for all $`x,yM`$. We saw that elements of $`G`$ correspond to triples $`s`$ satisfying the condition $`h_1(M)=h_2(M)=L`$. Thus we must verify this condition.
Let $`g`$ be the inverse of $`f`$. Then $`fg=gf=d`$. For every $`xM`$ we have $`inf\{f(x,y)+g(y,x):yM\}=fg(x,x)=d(x,x)=0`$ and hence $`\rho (h_1(x),h_2(M))=inf\{f(x,y):yM\}=0`$. This means that $`h_1(x)`$ belongs to the closure of $`h_2(M)`$ in $`L`$. Since $`M`$ is complete and $`h_2`$ is an isometric embedding, $`h_2(M)`$ is closed in $`L`$. It follows that $`h_1(x)h_2(M)`$. Since $`xM`$ was arbitrary, we have $`h_1(M)h_2(M)`$. Similarly, $`h_2(M)h_1(M)`$ and therefore $`h_1(M)=h_2(M)=L`$. ∎
The operation $``$ has the following description in terms of $`M`$-triples. Let $`p,q\mathrm{\Theta }`$. There exists a quadruple $`s=(h_1,h_2,h_3,L)`$ such that $`(L,\rho )`$ is a metric space of diameter $`1`$, $`L=L_1L_2L_3`$, $`h_i:ML_i`$ is an isometry ($`i=1,2,3`$), $`(h_1,h_2,L_1L_2)`$ is an $`M`$-triple corresponding to $`p`$ and $`(h_2,h_3,L_2L_3)`$ is an $`M`$-triple corresponding to $`q`$. The bi-Katětov function $`f`$ corresponding to the $`M`$-triple $`(h_1,h_3,L_1L_3)`$ depends on $`s`$, and the largest function $`f`$ over all quadruples $`s`$ such as above is equal to $`pq`$. Indeed, we have $`f(x,y)=\rho (h_1(x),h_3(y))inf\{\rho (h_1(x),h_2(z))\rho (h_2(z),h_3(y)):zM\}=inf\{p(x,z)q(z,y):zM\}=pq(x,y)`$. To see that the function $`pq`$ can be attained, consider two disjoint copies $`M^{}`$ and $`M^{\prime \prime }`$ of $`M`$. For $`xM`$ denote by $`x^{}`$ the copy of $`x`$ in $`M^{}`$, and use similar notation for $`M^{\prime \prime }`$. Let $`\rho `$ be the pseudometric on $`X=MM^{}M^{\prime \prime }`$ defined by $`\rho (x,y)=\rho (x^{},y^{})=\rho (x^{\prime \prime },y^{\prime \prime })=d(x,y)`$, $`\rho (x,y^{})=p(x,y)`$, $`\rho (x^{},y^{\prime \prime })=q(x,y)`$ and $`\rho (x,y^{\prime \prime })=pq(x,y)`$. The triangle inequality for $`\rho `$ is easily verified. (The space $`X`$ is the amalgam (in the class of spaces of diameter $`1`$) of the spaces $`MM^{}`$ and $`M^{}M^{\prime \prime }`$ with the subspace $`M^{}`$ amalgamated, see for a definition.) Let $`L`$ be the metric space associated with the pseudometric space $`(X,\rho )`$. Let $`L_1,L_2,L_3`$ be the images of $`M`$, $`M^{}`$, $`M^{\prime \prime }`$ in $`L`$, respectively. Let $`h_i:ML_i`$ be the obvious isometry, $`i=1,2,3`$. The quadruple $`s=(h_1,h_2,h_3,L)`$ has the properties considered above, and the bi-Katětov function corresponding to the $`M`$-triple $`(h_1,h_3,L_1L_3)`$ is equal to $`pq`$.
## 7. The Roelcke compactification of groups of isometries
Let $`(M,d)`$ be a complete $`\omega `$-homogeneous Urysohn metric space, and let $`G=\mathrm{Iso}(M)`$. In the next section we shall prove that $`G`$ is minimal and topologically simple. The idea of the proof is to explicitly describe the Roelcke compactification of $`G`$. It turns out that the Roelcke completion of $`G`$ can be identified with the compact space $`\mathrm{\Theta }`$ of all bi-Katětov functions on $`M^2`$.
In the preceding section we defined the embedding $`i:G\mathrm{\Theta }`$ by $`i(\phi )(x,y)=d(x,\phi (y))`$. The space $`\mathrm{\Theta }`$, being compact, has a unique compatible uniformity. Let $`𝒰`$ be the coarsest uniformity on $`G`$ which makes the map $`i:G\mathrm{\Theta }`$ uniformly continuous. We say that $`𝒰`$ is the uniformity induced by $`i`$. The uniform space $`(G,𝒰)`$ is isomorphic to $`i(G)`$, considered as a uniform subspace of $`\mathrm{\Theta }`$. We are going to prove that $`𝒰`$ is the Roelcke uniformity on $`G`$ (Theorem 7.3).
Let us explain the idea of the proof. Let $`\phi ,\phi ^{}G`$. We want to prove that $`\phi `$ and $`\phi ^{}`$ are “sufficiently close” in $`\mathrm{\Theta }`$ if and only if $`\phi ^{}U\phi U`$, where $`U`$ is a “small” neighbourhood of the unity. Thus we are led to the following question: under what conditions does the equation $`\phi ^{}=\psi _1\phi \psi _2`$ have a solution with “small” $`\psi _1`$ and $`\psi _2`$? Here “small” means that points of a given finite subset $`AM`$ are moved by less than $`ϵ`$. Observe that similar questions for the equations $`\phi ^{}=\phi \psi `$ or $`\phi ^{}=\psi \phi `$ have an obvious answer: $`\phi ^{}\phi U`$ iff $`\phi `$ and $`\phi ^{}`$ move points of $`A`$ “almost in the same way”, that is, $`d(\phi (x),\phi ^{}(x))<ϵ`$ for every $`xA`$; similarly, $`\phi ^{}U\phi `$ iff the inverse maps $`\phi ^1`$ and $`\phi _{}^{}{}_{}{}^{1}`$ move points of $`A`$ “almost in the same way”. The equation $`\phi ^{}=\psi _1\phi \psi _2`$ with two unknowns $`\psi _1`$ and $`\psi _2`$ looks more complicated. However, the answer to the above question is easy also in this case: the condition $`\phi ^{}U\phi U`$ means that the finite metric spaces $`A\phi (A)`$ and $`A\phi ^{}(A)`$ are close to each other in the Gromov–Hausdorff metric.
We shall need the notion of the Gromov–Hausdorff metric only for finite metric spaces with a given enumeration (it differs from the usual notion dealing with non-enumerated spaces). Let $`X=\{x_1,\mathrm{},x_n\}`$ and $`Y=\{y_1,\mathrm{},y_n\}`$ be two such spaces. The Gromov–Hausdorff distance for enumerated spaces between $`X`$ and $`Y`$, denoted by $`d_{GH}^{en}(X,Y)`$, is the infimum of the numbers $`\mathrm{max}\{D(x_i,y_i):i=1,\mathrm{},n\}`$, taken over all pseudometrics $`D`$ on $`XY`$ (we assume that $`X`$ and $`Y`$ are disjoint) such that $`D`$ induces the given metrics on $`X`$ and $`Y`$. If $`X`$ and $`Y`$ have diameter $`1`$, we may assume that the same is true for $`(XY,D)`$, otherwise replace $`D`$ by $`D1`$. Since the Urysohn space $`(M,d)`$ contains an isometric copy of every finite metric space of diameter $`1`$ (Proposition 1.6), it follows that $`d_{GH}^{en}(X,Y)`$ is the infimum of the numbers $`\mathrm{max}\{d(a_i,b_i):i=1,\mathrm{},n\}`$, where $`a_i,b_iM(1in)`$ are such that the correspondences $`x_ia_i`$ and $`y_ib_i`$ are isometric embeddings of $`X`$ and $`Y`$ into $`M`$, respectively.
###### Proposition 7.1.
Let $`(X,d_X)`$ and $`(Y,d_Y)`$ be two enumerated finite metric spaces, $`X=\{x_1,\mathrm{},x_n\}`$, $`Y=\{y_1,\mathrm{},y_n\}`$. Let
$$ϵ=\mathrm{max}\{|d_X(x_i,x_j)d_Y(y_i,y_j)|:i,j=1,\mathrm{},n\}.$$
Then $`d_{GH}^{en}(X,Y)=ϵ/2`$
###### Proof.
The inequality $``$ is obvious: if $`D`$ is a pseudometric on $`XY`$ extending $`d_X`$ and $`d_Y`$ and $`ϵ=|d_X(x_i,x_j)d_Y(y_i,y_j)|`$, then at least one of the numbers $`D(x_i,y_i)`$ and $`D(x_j,y_j)`$ must be $`ϵ/2`$. To prove the reverse inequality, we construct a pseudometric $`D`$ on $`Z=XY`$ extending $`d_X`$ and $`d_Y`$ such that
$$D(x_i,y_i)=ϵ/2,i=1,\mathrm{},n.$$
The function $`D`$ is defined by these requirements on $`X^2`$, $`Y^2`$, and the set $`\{(x_i,y_i):i=1,\mathrm{},n\}`$. To see that $`D`$ can be extended to a pseudometric on $`Z`$, it suffice to verify that for any sequence $`z_1,\mathrm{},z_s`$ of points of $`Z`$ such that all the expressions $`D(z_i,z_{i+1})`$ ($`1i<s`$) and $`D(z_1,z_s)`$ are defined the inequality
$`(\mathrm{A})`$
$$D(z_1,z_s)\underset{i=1}{\overset{s1}{}}D(z_i,z_{i+1})$$
holds. Then the required extension is given by the formula
$$D(z,z^{})=inf\underset{i=1}{\overset{s1}{}}D(z_i,z_{i+1}),$$
where the infimum is taken over all chains $`z_1=z,z_2,\mathrm{},z_s=z^{}`$ such that all the terms $`D(z_i,z_{i+1})`$ are defined. An easy argument using induction shows that (A) follows from its special case: for any “quadrangle” in $`Z`$ of the form $`x_i,y_i,y_j,x_j`$ each of the four numbers $`d_X(x_i,x_j)`$, $`D(x_i,y_i)`$, $`d_Y(y_i,y_j)`$, and $`D(x_j,y_j)`$ does not exceed the sum of the three others. This case is obvious: for example, since $`d_X(x_i,x_j)d_Y(y_i,y_j)ϵ`$, we have
$$d_X(x_i,x_j)d_Y(y_i,y_j)+ϵ=D(x_i,y_i)+d_Y(y_i,y_j)+D(x_j,y_j).$$
###### Corollary 7.2.
Let $`(X,d)`$ be an Urysohn metric space. Let $`a_1,\mathrm{},a_n,b_1,\mathrm{},b_nX`$, and suppose that
$$|d(a_i,a_j)d(b_i,b_j)|2ϵ$$
for all $`i,j=1,\mathrm{},n`$. Then there exist points $`c_1,\mathrm{},c_nX`$ such that $`d(c_i,c_j)=d(b_i,b_j)`$ and $`d(a_i,c_i)ϵ`$ for all $`i,j=1,\mathrm{},n`$. ∎
We now are in a position to prove the main result of this section. Recall that $`(M,d)`$ is a complete $`\omega `$-homogeneous Urysohn metric space, $`G=\mathrm{Iso}(M)`$, and $`\mathrm{\Theta }`$ is the space of bi-Katětov functions on $`M^2`$ considered in the previous section.
###### Theorem 7.3.
The range of the embedding $`i:G\mathrm{\Theta }`$ is dense in $`\mathrm{\Theta }`$. The uniformity $`𝒰`$ on $`G`$ induced by the embedding $`i`$ coincides with the Roelcke uniformity $``$. Therefore, $`G`$ is Roelcke-precompact, and the Roelcke compactification of $`G`$ can be identified with $`\mathrm{\Theta }`$.
###### Proof.
If $`A`$ is a finite subset of $`M`$ and $`ϵ>0`$, let $`U_{A,ϵ}=\{\psi G:d(\psi (x),x)<ϵ\text{ for every }xA\}𝒩(G)`$. Let $`W_{A,ϵ}`$ be the set of all pairs $`(f,g)\mathrm{\Theta }^2`$ such that $`|f(x,y)g(x,y)|<ϵ`$ for all $`x,yA`$. The sets of the form $`W_{A,ϵ}`$ constitute a base of entourages of the uniformity on $`\mathrm{\Theta }`$. If $`(f,g)W=W_{A,ϵ}`$, we say that $`f`$ and $`g`$ are $`W`$-close. Our proof proceeds in three parts.
(a) We prove that $`i(G)`$ is dense in $`\mathrm{\Theta }`$. Let $`f\mathrm{\Theta }`$, and let $`Of`$ be a neighbourhood of $`f`$ in $`\mathrm{\Theta }`$. We must prove that $`i(\phi )Of`$ for some $`\phi G`$.
We may assume that $`Of`$ is the set of all $`g\mathrm{\Theta }`$ such that $`g`$ is $`W_{A,ϵ}`$-close to $`f`$:
$$Of=\{g\mathrm{\Theta }:|g(x,y)f(x,y)|<ϵ\text{ for all }x,yA\},$$
where $`A`$ is a finite subset of $`M`$ and $`ϵ>0`$. Let $`A=\{a_1,\mathrm{},a_n\}`$. We claim that there exist points $`b_1,\mathrm{},b_nM`$ such that $`d(b_i,b_j)=d(a_i,a_j)`$ and $`d(a_i,b_j)=f(a_i,a_j)`$, $`1i,jn`$. Indeed, since $`f`$ is bi-Katětov, the formulas above define a pseudometric on the set $`F=\{a_1,\mathrm{},a_n,b_1,\mathrm{},b_n\}`$, where $`b_1,\mathrm{},b_n`$ are new points. Since $`M`$ is Urysohn, the embedding of $`A`$ into $`M`$ extends to a distance-preserving map from $`F`$ to $`M`$.
Since $`M`$ is $`\omega `$-homogeneous, there exists an isometry $`\phi `$ of $`M`$ such that $`\phi (a_i)=b_i`$, $`1in`$. Let $`g=i(\phi )`$. For every $`i,j[1,n]`$ we have $`g(a_i,a_j)=d(a_i,\phi (a_j))=d(a_i,b_j)=f(a_i,a_j)`$. Thus $`gOf`$. This proves that $`i(G)`$ is dense in $`\mathrm{\Theta }`$.
(b) We prove that the uniformity $`𝒰`$ is coarser than $``$.
Whenever a topological group $`H`$ acts continuously on a compact space $`X`$ (on the left), for every $`xX`$ the orbit map $`hhx`$ from $`H`$ to $`X`$ is right-uniformly continuous. We saw that $`G`$ acts continuously on $`\mathrm{\Theta }`$ (Proposition 6.5). The embedding $`i:G\mathrm{\Theta }`$ can be viewed as the orbit map corresponding to $`d`$, the neutral element of $`\mathrm{\Theta }`$. It follows that $`i`$ is $``$-uniformly continuous. Similarly, $`i`$ is $``$-uniformly continuous (use the right action of $`G`$ on $`\mathrm{\Theta }`$, or, alternatively, use the involution on $`\mathrm{\Theta }`$ to deduce $``$-uniform continuity of $`i`$ from its $``$-uniform continuity). Therefore, the uniformity $`𝒰`$ is coarser than both $``$ and $``$ and hence coarser than $``$.
(c) We prove that $`𝒰`$ is finer than $``$. It suffices to show that for every $`U𝒩(G)`$ there exists an entourage $`W`$ of the uniformity on $`\mathrm{\Theta }`$ (in other words, a neighbourhood of the diagonal of $`\mathrm{\Theta }^2`$) with the following property: if $`\phi ,\phi ^{}G`$ are such that $`i(\phi )`$ and $`i(\phi ^{})`$ are $`W`$-close, then $`\phi ^{}U\phi U`$. Assume that $`U=U_{A,ϵ}`$. We claim that $`W=W_{A,2ϵ}`$ has the required property.
Let $`\phi ,\phi ^{}G`$ be such that $`i(\phi )`$ and $`i(\phi ^{})`$ are $`W_{A,2ϵ}`$-close. This means that
$$\delta =\mathrm{max}\{|d(x,\phi (y))d(x,\phi ^{}(y))|:x,yA\}<2ϵ.$$
Let $`A=\{a_1,\mathrm{},a_n\}`$, $`b_i=\phi (a_i)`$ and $`c_i=\phi ^{}(a_i)`$, $`i=1,\mathrm{},n`$. We have $`d(b_i,b_j)=d(a_i,a_j)=d(c_i,c_j)`$ and $`|d(a_i,b_j)d(a_i,c_j|\delta `$ for all $`i`$ and $`j`$. In virtue of Corollary 7.2, there exist points $`a_1^{},\mathrm{},a_n^{},b_1^{},\mathrm{},b_n^{}M`$ such that the correspondence $`a_ia_i^{}`$, $`b_ib_i^{}`$ is distance-preserving and $`d(a_i^{},a_i)\delta /2<ϵ`$, $`d(b_i^{},c_i)\delta /2<ϵ`$.
Since $`M`$ is $`\omega `$-homogeneous, there exists an isometry $`\psi _1`$ of $`M`$ such that $`\psi _1(a_i)=a_i^{}`$ and $`\psi _1(b_i)=b_i^{}`$, $`i=1,\mathrm{},n`$. We have $`\psi _1U`$, since each $`a_i`$ is moved by less than $`ϵ`$. Put $`\psi _2=\phi ^1\psi _1^1\phi ^{}`$. For every $`i=1,\mathrm{},n`$ we have $`d(\psi _2(a_i),a_i)=d(\phi ^{}(a_i),\psi _1\phi (a_i))=d(c_i,b_i^{})<ϵ`$, hence $`\psi _2U=U_{A,ϵ}`$. Thus $`\phi ^{}=\psi _1\phi \psi _2U\phi U`$, as required. ∎
Recall that a non-empty collection $``$ of non-empty subsets of a set $`X`$ is a filter base on $`X`$ if for every $`A,B`$ there is $`C`$ such that $`CAB`$. If $`X`$ is a topological space, $``$ is a filter base on $`X`$ and $`xX`$, then $`x`$ is a cluster point of $``$ if every neighbourhood of $`x`$ meets every member of $``$, and $``$ converges to $`x`$ if every neighbourhood of $`x`$ contains a member of $``$. If $``$ and $`𝒢`$ are two filter bases on $`G`$, let $`𝒢=\{AB:A,B𝒢\}`$.
For every $`p\mathrm{\Theta }`$ let $`_p=\{GV:V\text{ is a neighbourhood of }p\text{ in }\mathrm{\Theta }\}`$. In other words, $`_p`$ is the trace on $`G`$ of the filter of neighbourhoods of $`p`$ in $`\mathrm{\Theta }`$. If $`p,q\mathrm{\Theta }`$, it is not true in general that $`_p_q`$ converges to $`pq`$. However, we have the following result, which will be used in the proof of Theorem 1.8:
###### Proposition 7.4.
If $`p,q\mathrm{\Theta }`$, then $`pq`$ is a cluster point of the filter base $`_p_q`$.
###### Proof.
Let $`U_1`$, $`U_2`$, $`U_3`$ be neighbourhoods of $`p`$, $`q`$ and $`pq`$, respectively. We must show that $`U_3`$ meets the set $`(U_1i(G))(U_2i(G))`$.
We may assume that for some finite set $`A=\{a_1,\mathrm{},a_n\}M`$ and $`ϵ>0`$ we have
$$U_1=\{f\mathrm{\Theta }:|f(x,y)p(x,y)|<ϵ\text{ for all }x,yA\};$$
$$U_2=\{f\mathrm{\Theta }:|f(x,y)q(x,y)|<ϵ\text{ for all }x,yA\};$$
$$U_3=\{f\mathrm{\Theta }:|f(x,y)pq(x,y)|<ϵ\text{ for all }x,yA\}.$$
We saw in the last paragraph of the preceding section that there exist a metric space $`(L,\rho )`$ and isometric embeddings $`h_i:ML`$ ($`i=1,2,3`$) such that $`p(x,y)=\rho (h_1(x),h_2(y))`$, $`q(x,y)=\rho (h_2(x),h_3(y))`$ and $`pq(x,y)=\rho (h_1(x),h_3(y))`$ for all $`x,yM`$. Let $`X=h_1(A)h_2(A)h_3(A)`$. Since $`M`$ is Urysohn, there exists an isometric embedding of $`X`$ into $`M`$ which extends the isometry $`h_1^1:h_1(A)A`$. It follows that there exist points $`b_1,\mathrm{},b_n,c_1\mathrm{},c_nM`$ such that $`d(b_i,b_j)=d(c_i,c_j)=d(a_i,a_j)`$, $`d(a_i,b_j)=p(a_i,a_j)`$, $`d(b_i,c_j)=q(a_i,a_j)`$ and $`d(a_i,c_j)=pq(a_i,a_j)`$ for all $`i,j`$. Since $`M`$ is $`\omega `$-homogeneous, there exists an isometry $`\phi G`$ such that $`\phi (a_i)=b_i`$, $`1in`$. Let $`x_i=\phi ^1(c_i)`$. Using again the $`\omega `$-homogeneity of $`M`$, we find an isometry $`\psi G`$ such that $`\psi (a_i)=x_i`$, $`1in`$. Note that $`\phi \psi (a_i)=c_i`$ and $`d(a_i,x_j)=d(\phi (a_i),\phi (x_j))=d(b_i,c_j)=q(a_i,a_j)`$ for all $`i,j`$. We claim that $`i(\phi )U_1`$, $`i(\psi )U_2`$ and $`i(\phi \psi )U_3`$. Indeed, we have $`i(\phi )(x,y)=d(x,\phi (y))=p(x,y)`$ for all $`x,yA`$ and hence $`i(\phi )U_1`$. The other two cases are considered similarly. Thus $`i(\phi \psi )\left((U_1i(G))(U_2i(G))\right)U_3\mathrm{}`$. ∎
If $`H`$ is a group and $`gH`$, we denote by $`l_g`$ (respectively, $`r_g`$) the left shift of $`H`$ defined by $`l_g(h)=gh`$ (respectively, the right shift defined by $`r_g(h)=hg`$).
###### Proposition 7.5.
Let $`H`$ be a topological group, and let $`K`$ be the Roelcke completion of $`H`$. Let $`gH`$. Each of the following self-maps of $`H`$ extends to a self-homeomorphism of $`K`$: (1) the left shift $`l_g`$; (2) the right shift $`r_g`$; (3) the inversion $`gg^1`$; (4) the inner automorphism $`hghg^1`$.
###### Proof.
Let $``$ and $``$ be the left and the right uniformity on $`H`$, respectively. In each of the cases (1)–(4) the map $`f:HH`$ under consideration is an automorphism of the uniform space $`(H,)`$. This is obvious for the cases (3) and (4). For the cases (1) and (2), observe that the uniformities $``$ and $``$ are invariant under left and right shifts, hence the same is true for their greatest lower bound $``$. It follows that in all cases $`f`$ extends to an automorphism of the completion $`K`$ of the uniform space $`(H,)`$. ∎
For the group $`G`$ and its Roelcke completion $`\mathrm{\Theta }`$ the validity of Proposition 7.5 can be seen directly. Recall that the embedding $`i:G\mathrm{\Theta }`$ is a morphism of monoids with an involution (see the two paragraphs before Proposition 6.5). The involution $`ff^{}`$ on $`\mathrm{\Theta }`$ is continuous and hence coincides with the extension of the inversion on $`G`$ given by Proposition 7.5. For every $`gG`$ let $`L_g`$, $`R_g`$ and $`\mathrm{Inn}_g`$ be the self-maps of $`\mathrm{\Theta }`$ defined by $`L_g(p)=gp`$, $`R_g(p)=pg`$ and $`\mathrm{Inn}_g(p)=gpg^1`$. These maps are extensions over $`\mathrm{\Theta }`$ of the left shift $`l_g`$ of $`G`$, the right shift $`r_g`$, and the inner automorphism $`l_gr_{g^1}`$, respectively. In virtue of Proposition 6.5, the maps $`L_g`$ and $`R_g`$ are continuous, and the same is true for $`\mathrm{Inn}_g=L_gR_{g^1}`$.
An inner automorphism of $`\mathrm{\Theta }`$ is a map of the form $`\mathrm{Inn}_g`$, $`gG`$. Proposition 6.5 shows that $`\mathrm{Inn}_g(p)(x,y)=p(g^1(x),g^1(y))`$ for all $`p\mathrm{\Theta }`$ and $`x,yM`$. It follows that for every closed $`FM`$ we have $`\mathrm{Inn}_g(b_F)=b_{g(F)}`$, where $`b_F`$ is the idempotent corresponding to $`F`$ (see Proposition 6.4).
###### Proposition 7.6.
There are precisely two idempotents in $`\mathrm{\Theta }`$ which are $`d`$ and are invariant under all inner automorphisms: the unity $`d`$ and the constant 1.
###### Proof.
According to Proposition 6.4, every idempotent $`d`$ is of the form $`b_F`$ for some closed $`FM`$. If $`b_F`$ is invariant under inner automophisms, then $`b_{g(F)}=\mathrm{Inn}_g(b_F)=b_F`$ and hence $`g(F)=F`$ for every $`gG`$. Since the action of $`G`$ on $`M`$ is transitive, no proper non-empty subset of $`M`$ is $`G`$-invariant. Thus either $`F=M`$ or $`F=\mathrm{}`$. Accordingly, either $`b_F=d`$ or $`b_F=1`$. ∎
## 8. Proof of Theorem 1.8
We preserve the notation of the preceding section: $`M`$ is a complete $`\omega `$-homogeneous Urysohn metric space, $`G=\mathrm{Iso}(M)`$, $`\mathrm{\Theta }`$ is the set of all bi-Katětov functions on $`M^2`$. We saw that $`G`$ is Roelcke-precompact and that $`\mathrm{\Theta }`$ can be identified with the Roelcke compactification of $`G`$ (Theorem 7.3). In this section we prove that $`G`$ is minimal and topologically simple.
###### Proposition 8.1.
For every topological group $`H`$ the following conditions are equivalent:
1. $`H`$ is minimal and topologically simple;
2. if $`f:HH^{}`$ is a continuous onto homomorphism of topological groups, then either $`f`$ is a homeomorphism or $`|H^{}|=1`$.∎
###### Proposition 8.2.
The group $`G`$ has no compact normal subgroups other than $`\{e\}`$.
We shall prove later that actually $`G`$ has no non-trivial closed normal subgroups.
###### Proof.
Let $`H\{e\}`$ be a normal subgroup of $`G`$. We show that $`H`$ is not compact.
Fix $`aM`$ and $`fH`$ such that $`f(a)a`$. Let $`r=d(f(a),a)`$, and let $`S=\{xM:d(x,a)=r\}`$ be the sphere of radius $`r`$ centered at $`a`$. We claim that the orbit $`Ha`$ contains $`S`$. Fix $`xS`$. Since $`M`$ is $`\omega `$-homogeneous, there exists an isometry $`gG`$ which leaves the point $`a`$ fixed and maps $`f(a)`$ to $`x`$. Let $`h=gfg^1`$. Since $`H`$ is normal, we have $`hH`$ and hence $`x=h(a)Ha`$. Thus $`SHa`$, as claimed.
Since $`M`$ is Urysohn, we can construct by induction an infinite sequence $`x_1,x_2,\mathrm{}`$ of points in $`S`$ such that all the pairwise distances between distinct members of this sequence are equal to $`r`$. Since $`SHa`$, it follows that $`Ha`$ is not compact. Hence $`H`$ is not compact. ∎
Let $`(L,\rho )`$ be a metric space. A self-map $`f:LL`$ is non-expanding if $`\rho (f(x),f(y))\rho (x,y)`$ for all $`x,yL`$.
###### Lemma 8.3.
Let $`(L,\rho )`$ be a metric space, and let $`F`$ be the semigroup of all non-expanding self-maps of $`L`$, equipped with the topology of pointwise convergence. Then the map $`(f,g)fg`$ from $`F^2`$ to $`F`$ is continuous. Thus $`F`$ is a topological semigroup.
This lemma and Proposition 8.4 below are well known. We include a proof for the reader’s convenience.
###### Proof.
It suffices to show that for every $`xL`$ the map $`(f,g)f(g(x))`$ from $`F^2`$ to $`L`$ is continuous. Fix $`f_0,g_0F`$, $`xL`$ and $`ϵ>0`$. Let $`y=g_0(x)`$, $`Of_0=\{fF:\rho (f(y),f_0(y))<ϵ\}`$ and $`Og_0=\{gF:\rho (g(x),y)<ϵ\}`$. If $`fOf_0`$ and $`gOg_0`$, then $`\rho (f(g(x)),f_0(g_0(x)))\rho (f(g(x)),f(y))+\rho (f(y),f_0(y))<\rho (g(x),y)+ϵ<2ϵ`$. ∎
###### Proposition 8.4.
If $`L`$ is a complete metric space, then the group $`\mathrm{Iso}(L)`$ is complete.
Recall that we call a topological group complete if it is complete with respect to the upper uniformity.
###### Proof.
Let $`X=L^L`$ be the set of all self-maps of $`L`$, equipped with the product uniformity. The group $`H=\mathrm{Iso}(L)`$ can be considered as a subset of $`X`$. The uniformity $`𝒰`$ on $`H`$ induced by the product uniformity on $`X`$ coincides with the left uniformity $``$. Indeed, a basic entourage for $`𝒰`$ has the form $`W_{A,ϵ}=\{(f,g)H^2:\rho (f(x),g(x))<ϵ\text{ for all }xA\}`$, where $`\rho `$ is the metric on $`L`$, $`A`$ is a finite subset of $`L`$ and $`ϵ>0`$. Let $`U_{A,ϵ}=\{fH:\rho (f(x),x)<ϵ\text{ for all }xA\}`$. Then $`U_{A,ϵ}`$ is a basic neighbourhood of unity in $`H`$, and $`W_{A,ϵ}=\{(f,g)H^2:g^1fU_{A,ϵ}\}`$ is a basic entourage for $``$. Thus $`𝒰=`$. It follows that the map $`gg^1`$ from $`H`$ to $`X`$ induces the right uniformity on $`H`$, and the map $`j:HX^2`$ defined by $`j(g)=(g,g^1)`$ induces the upper uniformity $``$. Since $`X^2`$ is complete, to prove that $`H`$ is complete it suffices to show that $`j(H)`$ is closed in $`X^2`$. Let $`F`$ be the set of all non-expanding self-maps of $`L`$. Then $`F`$ is closed in $`X`$. The map $`(f,g)fg`$ from $`F^2`$ to $`F`$ is continuous (Lemma 8.3). Since $`j(G)=\{(f,g)F^2:fg=gf=\text{id}_L\}`$, it follows that $`j(G)`$ is closed in $`F^2`$ and hence in $`X^2`$. ∎
We say that a metric space $`L`$ is homogeneous if every point of $`L`$ can be mapped to every other point by an isometry of $`L`$ onto itself.
###### Lemma 8.5.
If $`L`$ is a homogeneous metric space, then $`w(\mathrm{Iso}(L))=w(L)`$.
###### Proof.
For every metric space $`X`$ we have $`w(\mathrm{Iso}(X))w(X)`$. If $`X`$ is homogeneous, then for every $`aX`$ the map $`ff(a)`$ from $`\mathrm{Iso}(X)`$ to $`X`$ is onto, whence $`w(X)w(\mathrm{Iso}(X))`$. ∎
We are now ready to prove Theorem 1.8:
If $`M`$ is a complete $`\omega `$-homogeneous Urysohn metric space, then the group $`G=\mathrm{Iso}(M)`$ is complete, Roelcke-precompact, minimal and topologically simple. The weight of $`G`$ is equal to the weight of $`M`$.
###### Proof.
We saw that $`G`$ is Roelcke-precompact (Theorem 7.3). Proposition 8.4 shows that $`G`$ is complete, and Lemma 8.5 shows that $`w(G)=w(M)`$. Let $`f:GG^{}`$ be a continuous onto homomorphism. According to Proposition 8.1, to prove that $`G`$ is minimal and topologically simple, it suffices to prove that either $`f`$ is a homeomorphism or $`|G^{}|=1`$.
Since $`G`$ is Roelcke-precompact, so is $`G^{}`$. Let $`\mathrm{\Theta }^{}`$ be the Roelcke compactification of $`G^{}`$. The homomorphism $`f`$ extends to a continuous map $`F:\mathrm{\Theta }\mathrm{\Theta }^{}`$. Let $`e^{}`$ be the unity of $`G^{}`$, and let $`S=F^1(e^{})\mathrm{\Theta }`$.
Claim 1. $`S`$ is a subsemigroup of $`\mathrm{\Theta }`$.
Let $`p,qS`$. In virtue of Proposition 7.4, there exist filter bases $`_p`$ and $`_q`$ on $`G`$ such that $`_p`$ converges to $`p`$ (in $`\mathrm{\Theta }`$), $`_q`$ converges to $`q`$ and $`pq`$ is a cluster point of the filter base $`_p_q`$. The filter bases $`_p^{}=F(_p)`$ and $`_q^{}=F(_q)`$ on $`G^{}`$ converge to $`F(p)=F(q)=e^{}`$, hence the same is true for the filter base $`_p^{}_q^{}=F(_p_q)`$. Since $`pq`$ is a cluster point of $`_p_q`$, $`F(pq)`$ is a cluster point of the convergent filter base $`F(_p_q)`$. A convergent filter on a Hausdorff space has only one cluster point, namely the limit. Thus $`F(pq)=e^{}`$ and hence $`pqS`$.
Claim 2. The semigroup $`S`$ is closed under involution.
In virtue of Proposition 7.5, the inversion on $`G^{}`$ extends to an involution $`xx^{}`$ of $`\mathrm{\Theta }^{}`$. Since $`F(p^{})=F(p)^{}`$ for every $`pG`$, the same holds for every $`p\mathrm{\Theta }`$. Let $`pS`$. Then $`F(p^{})=F(p)^{}=e^{}`$ and hence $`p^{}S`$.
Claim 3. If $`gG`$ and $`g^{}=f(g)`$, then $`F^1(g^{})=gS=Sg`$.
We saw that the left shift $`hgh`$ of $`G`$ extends to a continuous self-map $`L=L_g`$ of $`\mathrm{\Theta }`$ defined by $`l(p)=gp`$ (Proposition 6.5). According to Proposition 7.5, the self-map $`xg^{}x`$ of $`G^{}`$ extends to a self-homeomorphism $`L^{}`$ of $`\mathrm{\Theta }^{}`$. The maps $`FL`$ and $`L^{}F`$ from $`\mathrm{\Theta }`$ to $`\mathrm{\Theta }^{}`$ coincide on $`G`$ and hence everywhere. Replacing $`g`$ by $`g^1`$, we see that $`FL^1=(L^{})^1F`$. Thus $`F^1(g^{})=F^1L^{}(e^{})=LF^1(e^{})=gS`$. Using right shifts, we similarly conclude that $`F^1(g^{})=Sg`$.
Claim 4. $`S`$ is invariant under inner automorphisms of $`\mathrm{\Theta }`$.
We have just seen that $`gS=Sg`$ for every $`gG`$, hence $`gSg^1=S`$.
Let $`T=\{fS:fd\}`$. Note that $`i(e_G)=dT\mathrm{}`$. According to Proposition 6.3, there is a greatest element $`p`$ in $`T`$, and $`p`$ is idempotent. Since inner automorphisms of $`\mathrm{\Theta }`$ preserve the order on $`\mathrm{\Theta }`$ and the unity $`d`$, Claim 4 implies that $`p`$ is invariant under inner automorphisms. In virtue of Proposition 7.6, either $`p=d`$ or $`p=1`$. We shall show that either $`f`$ is a homeomorphism or $`|G^{}|=1`$, according to which of the cases $`p=d`$ or $`p=1`$ holds.
Consider first the case $`p=d`$.
Claim 5. If $`p=d`$, then all elements of $`S`$ are invertible in $`\mathrm{\Theta }`$.
Let $`fS`$. Then $`f^{}fS`$ and $`ff^{}S`$, since $`S`$ is a symmetrical semigroup. According to Proposition 6.2, we have $`f^{}fd`$ and $`ff^{}d`$. Since $`p=d`$, there are no elements $`>d`$ in $`S`$. Thus the inequalities $`f^{}fd`$ and $`ff^{}d`$ are actually equalities. It follows that $`f^{}`$ is the inverse of $`f`$.
Claim 6. If $`p=d`$, then $`S=\{e\}`$.
Claim 5 and Proposition 6.6 imply that $`S`$ is a subgroup of $`G`$. This subgroup is normal (Claim 4) and compact, since $`S`$ is closed in $`\mathrm{\Theta }`$. Proposition 8.2 implies that $`S=\{e\}`$.
Claim 7. If $`p=d`$, then $`f:GG^{}`$ is a homeomorphism.
Claims 6 and 3 imply that $`G=F^1(G^{})`$ and that the map $`f:GG^{}`$ is bijective. Since $`F`$ is a map between compact spaces, it is perfect, and hence so is the map $`f:G=F^1(G^{})G^{}`$. Thus $`f`$, being a perfect bijection, is a homeomorphism.
Now consider the case $`p=1`$.
Claim 8. If $`1S`$, then $`G^{}=\{e^{}\}`$.
Let $`gG`$ and $`g^{}=f(g)`$. We have $`g1=1S`$. On the other hand, Claim 3 implies that $`g1gS=F^1(g^{})`$. Thus $`g^{}=F(g1)=F(1)=e^{}`$. ∎
## 9. Remarks
1. Let $`M`$ be a complete $`\omega `$-homogeneous Urysohn metric space, and let $`G=\mathrm{Iso}(M)`$. In Section 7 we identified the Roelcke completion of $`G`$ with the set $`\mathrm{\Theta }`$ of all bi-Katětov functions on $`M^2`$. The set $`\mathrm{\Theta }`$ was equipped with structures of three kinds: topology, order, semigroup structure. The proof of Theorem 1.8 was based on the interplay between these three structures. We now establish a natural one-to-one correspondence between $`\mathrm{\Theta }`$ and a set of closed relations on a compact space. This correspondence will be an isomorphism for all three structures on $`\mathrm{\Theta }`$.
Let $`K`$ be a compact space. A closed relation on $`K`$ is a closed subset of $`K^2`$. Let $`E(K)`$ be the compact space of all closed relations on $`K`$, equipped with the Vietoris topology. The set $`E(K)`$ has a natural partial order. If $`R,SE(K)`$, then the composition $`RS`$ is a closed relation, since $`RS`$ is the image of the closed subset $`\{(x,z,y):(x,z)S,(z,y)R\}`$ of $`K^3`$ under the projection $`K^3K^2`$ which is a closed map. Thus $`E(K)`$ is a semigroup with involution. In general the map $`(R,S)RS`$ from $`E(K)^2`$ to $`E(K)`$ is not separately continuous (neither left nor right continuous).
We denote by $`\mathrm{Homeo}(K)`$ the group of all self-homeomorphisms of $`K`$, equipped with the compact-open topology. For every $`h\mathrm{Homeo}(K)`$ let $`\mathrm{\Gamma }(h)=\{(x,h(x)):xK\}`$ be the graph of $`h`$. The map $`h\mathrm{\Gamma }(h)`$ from $`\mathrm{Homeo}(K)`$ to $`E(K)`$ is a homeomorphic embedding and a morphism of monoids with an involution. The uniformity induced on $`\mathrm{Homeo}(K)`$ by this embbedding is coarser than the Roelcke uniformity.
Now let $`K`$ be the compact space of all non-expanding functions $`f:MI=[0,1]`$, considered as a subspace of the product $`I^M`$. There is a natural left action of $`G`$ on $`K`$, defined by $`gf(x)=f(g^1(x))(gG,fK,xM)`$. This action gives rise to a morphism $`G\mathrm{Homeo}(K)`$ of topological groups which is easily seen to be a homeomorphic embedding. Let $`j:GE(K)`$ be the composition of this embedding with the map $`h\mathrm{\Gamma }(h)`$ from $`\mathrm{Homeo}(K)`$ to $`E(K)`$. If $`gG`$, then $`j(g)`$ is the relation $`\{(f,gf):fK\}`$. Let $`\mathrm{\Phi }`$ be the closure of $`j(G)`$ in $`E(K)`$. Let $`\mathrm{\Theta }`$ and $`i:G\mathrm{\Theta }`$ be the same as in Sections 6 and 7.
###### Theorem 9.1.
The uniformity on $`G`$ induced by the embedding $`j:GE(K)`$ coincides with the Roelcke uniformity, hence $`\mathrm{\Phi }`$ can be identified with the Roelcke compactification of $`G`$. The set $`\mathrm{\Phi }`$ is a subsemigroup of $`E(K)`$. There exists a unique homeomorphism $`H:\mathrm{\Phi }\mathrm{\Theta }`$ such that $`i=Hj`$. The map $`H`$ is an isomorphism of ordered semigroups.
We omit the detailed proof and confine ourselves by a description of the isomorphism $`H`$. If $`R\mathrm{\Phi }`$, let $`H(R)`$ be the bi-Katětov function on $`M^2`$ defined by $`H(R)(x,y)=sup\{|q(x)p(y)|:(p,q)R\}`$, $`x,yM`$. If $`f\mathrm{\Theta }`$, the relation $`H^1(f)`$ is defined by $`H^1(f)=\{(p,q)K^2:|q(x)p(y)|f(x,y)\text{ for all }x,yM^2\}`$.
Let us see what some of the results about $`\mathrm{\Theta }`$ obtained in Section 6 mean in terms of relations on $`K`$. Functions $`p\mathrm{\Theta }`$ which are $`d`$ correspond (via the isomorphism $`H`$) to relations $`R\mathrm{\Phi }`$ which contain the diagonal of $`K^2`$ or, in other words, are reflexive. Thus Proposition 6.2 implies that for every $`R\mathrm{\Phi }`$ the relations $`RR^1`$ and $`R^1R`$ are reflexive. This is equivalent to the fact that for every $`R\mathrm{\Phi }`$ the domain and the range of $`R`$ is equal to $`K`$.
According to Proposition 6.4, each idempotent $`d`$ in $`\mathrm{\Theta }`$ has the form $`b_F`$ for some closed $`FM`$. Note that each $`b_F`$ is symmetrical. Symmetrical idempotents $`d`$ in $`\mathrm{\Theta }`$ correspond to relations in $`\mathrm{\Phi }`$ which are reflexive, symmetrical and transitive or, in other words, are equivalence relations. For every closed $`FM`$ let $`R_F=H(b_F)`$ be the equivalence relation corresponding to the idempotent $`b_F`$. Two non-expanding functions $`f,gK`$ are $`R_F`$-equivalent if and only if $`f|F=g|F`$. Proposition 6.4 implies that an equivalence relation $`R`$ on $`K`$ belongs to $`\mathrm{\Phi }`$ if and only if $`R=R_F`$ for some closed $`FM`$.
2. Let $`H`$ be a Hilbert space, and let $`G=U(H)`$ be the group of all unitary operators on $`H`$, equipped with the pointwise convergence topology. L. Stoyanov proved that $`G`$ is totally minimal . The methods of the present paper yield an alternative proof of this theorem. Let $`(H)`$ be the algebra of all bounded linear operators on $`H`$. The weak operator topology on $`(H)`$ is the coarsest topology such that for every $`x,yH`$ the function $`A(Ax,y)`$ on $`(H)`$ is continuous. Let $`T=\{A:A1\}`$ be the unit ball in $`(H)`$, equipped with the weak operator topology. The Roelcke compactification of the group $`G`$ can be identified with $`T`$ . The set $`T`$ is a semigroup, and the idempotents in $`T`$ are the orthogonal projectors. The proof of the fact that $`G`$ is totally minimal proceeds similarly to the proof of Theorem 1.8. Let us indicate the main steps. Let $`f:GG^{}`$ be a surjective morphism of topological groups. To prove that $`G`$ is totally minimal, it suffices to prove that $`f`$ is a quotient map. Extend $`f`$ to a map $`F:TT^{}`$, where $`T^{}`$ is the Roelcke compactification of $`G^{}`$. Let $`e^{}`$ be the unity of $`G^{}`$, and let $`S=F^1(e^{})`$ be the kernel of $`F`$. Then $`S`$ is a closed subsemigroup of $`T`$. It turns out that every closed subsemigroup of $`T`$ contains a least idempotent. Let $`p`$ be the least idempotent in $`S`$. Since $`S`$ is invariant under inner automophisms of $`\mathrm{\Theta }`$, so is $`p`$. It follows that either $`p=1`$ or $`p=0`$. If $`p=1`$, then $`SG`$, $`G=F^1(G^{})`$ and the map $`f`$ is perfect. If $`p=0`$, then $`G^{}=\{e^{}\}`$. See for more details.
3. Our method of proving minimality, based on the consideration of the Roelcke compactifications, can be applied to some groups of homeomorphisms. A zero-dimensional compact space $`X`$ is $`h`$-homogeneous if all non-empty clopen subsets of $`X`$ are homeomorphic to each other. Let $`K`$ be a zero-dimensional $`h`$-homogeneous compact space, and let $`G=\mathrm{Homeo}(K)`$. Then $`G`$ is minimal and topologically simple . Let us sketch a proof of this fact which closely follows the proof of Theorem 1.8. In the special case when $`K=2^\omega `$ is the Cantor set, the minimality of $`\mathrm{Homeo}(K)`$ was proved by Gamarnik .
Let $`T`$ be the compact space of all closed relations $`R`$ on $`K`$ such that the domain and the range of $`R`$ are equal to $`K`$. The map $`h\mathrm{\Gamma }(h)`$ from $`G`$ to $`T`$ induces the Roelcke uniformity on $`G`$, and the range $`\mathrm{\Gamma }(G)`$ of this map is dense in $`T`$. Thus the Roelcke compactification of $`G`$ can be identified with $`T`$. We noted that the set $`E(K)`$ of all closed relations on $`K`$ is an ordered semigroup with an involution. The set $`T`$ is a closed symmetrical subsemigroup of $`E(K)`$. Let $`\mathrm{\Delta }`$ be the diagonal in $`K^2`$. A relation $`RE(K)`$ is an equivalence relation if and only if $`R`$ is a symmetrical idempotent and $`R\mathrm{\Delta }`$. Let $`S`$ be a closed subsemigroup of $`E(K)`$, and let $`S_1`$ be the set of all $`RS`$ such that $`R\mathrm{\Delta }`$. The proof of Proposition 6.3 shows that the set $`S_1`$, if it is non-empty, has a largest element $`P`$, and $`P`$ is an idempotent. If $`S`$ is symmetrical, then so is $`P`$, hence $`P`$ is an equivalence relation.
Now let $`f:GG^{}`$ be a surjective morphism of topological groups. We show that either $`f`$ is a homeomorphism or $`|G^{}|=1`$. According to Proposition 8.1, this means that $`G`$ is minimal and topologically simple. Extend $`f`$ to a map $`F:TT^{}`$, where $`T^{}`$ is the Roelcke compactification of $`G^{}`$. Let $`e^{}`$ be the unity of $`G^{}`$, and let $`S=F^1(e^{})`$. Then $`S`$ is a closed symmetrical subsemigroup of $`T`$. Let $`P`$ be the largest element in the set $`S_1=\{RS:\mathrm{\Delta }R\}`$. Then $`P`$ is an equivalence relation on $`K`$. Since $`S`$ is $`G`$-invariant, so is $`P`$. But there are only two $`G`$-invariant closed equivalence relations on $`K`$, namely $`\mathrm{\Delta }`$ and $`K^2`$. If $`P=\mathrm{\Delta }`$, then $`SG`$, $`G=F^1(G^{})`$ and $`f`$ is perfect. Since $`G`$ has no non-trivial compact normal subgroups, we conclude that $`f`$ is a homeomorphism. If $`P=K^2`$, then $`S=T`$ and $`G^{}=\{e^{}\}`$.
It is not clear if a similar argument can be used when $`K`$ is a Hilbert cube and $`G=\mathrm{Homeo}(K)`$, see Problem 1.4.
## 10. Acknowledgement
I am much obliged to the referee for careful reading of the paper and for suggesting quite a few improvements.
|
warning/0004/quant-ph0004100.html
|
ar5iv
|
text
|
# On Rounds in Quantum Communication
## 1 Introduction
Quantum mechanical computing and communication has been studied extensively during the last decade. Communication has to be a physical process, so an investigation of the properties of physically allowed communication is desirable, and the fundamental theory of physics available to us is quantum mechanics.
The theory of communication complexity deals with the question how efficient communication problems can be solved and has various applications to lower bound proofs (introduction to (classical) communication complexity can be found in ). The communication complexity approach to lower bounds consists of reducing a lower bound proof for some computational model to a communication complexity lower bound, where several techniques for such proofs are available, see for many examples.
In a quantum protocol (as defined in ) two players Alice and Bob each receive an input, and have to compute some function defined on the pair of inputs cooperatively. To this end they exchange messages consisting of qubits, until the result can be produced by some measurement done by one of the players (for overviews on quantum communication complexity see and ).
A slightly different scenario proposed in and allows the players to start the protocol possessing some (input independent) qubits that are entangled with those of the other player. Due to the “superdense coding” technique of in this model 2 classical bits can be communicated by transmitting one qubit (and using up one EPR pair). See for some examples of lower bounds via communication complexity in the quantum setting. Unfortunately so far only few “applicable” lower bound methods for quantum protocols are known: the rank lower bound is known to hold for exact (i.e., errorless) quantum communication , the (usually weak) discrepancy lower bound for bounded error protocols .
One breakthrough result of the field of quantum computing is Grover’s search algorithm that retrieves an item from an unordered list within $`O(\sqrt{n})`$ questions , outperforming every classical algorithm for the problem. By an application of this search algorithm to communication complexity in an upper bound of $`O(\sqrt{n}\mathrm{log}n)`$ is shown for the bounded error quantum communication complexity of the disjointness problem $`DISJ_n`$ (both players receive an incidence vector of a subset of $`\{1,\mathrm{},n\}`$ and have to decide whether the sets are not disjoint: $`(x_iy_i)`$), one of the most important communication problems. This yields the largest gap between quantum and classical communication complexity known so far for a total function, the probabilistic communication complexity of disjointness is $`\mathrm{\Omega }(n)`$ . Currently no superlogarithmic lower bound on the bounded error quantum communication complexity of the disjointness problem is known.
Unfortunately the protocol for disjointness using Grover search takes $`\mathrm{\Theta }(\sqrt{n})`$ rounds, an unbounded increase in interaction compared to the trivial protocol communicating $`n`$ bits in 1 round. Similar phenomena show up in the polynomial gaps between quantum Las Vegas and probabilistic bounded error communication complexity for total functions, see .<sup>1</sup><sup>1</sup>1Exponential gaps between quantum communication complexity and classical probabilistic communication complexity are known only for partial functions, and are possible without interaction .
We are interested in the question how efficient total communication problems can be solved in the quantum model when the number of rounds is restricted. The most severe restriction is one-way communication, where only a monologue is transmitted from one player to the other, who then decides the function value. This has been investigated e.g. in , where a lower bound method based on the so called VC-dimension is proved, which allows to prove an exponential advantage for 2 round classical compared to 1 round quantum communication complexity. Kremer exhibits a complete problem for the class of problems with polylogarithmic quantum one-way communication complexity (in the case of bounded error).
In a series of papers (see , , , , ) more general round hierarchies of the following form are given for classical protocols: A function $`f_k`$ (usually the so called pointer jumping function) has $`k`$ round communication complexity $`k\mathrm{log}n`$ if Alice starts the communication, but has a much larger $`k`$ round communication complexity when Bob starts.
Our main result is that $`k`$-round quantum protocols need communication $`n/2^{2^{O(k)}}k\mathrm{log}n`$ to compute the pointer jumping function when Bob starts. So changing the starting player (or reducing the number of rounds by 1) may result in drastically increased communication also in the quantum case.
Nayak et. al. have proved a lower bound of $`\mathrm{\Omega }(n^{1/k})`$ for the quantum communication complexity of a certain subproblem of pointer jumping, in the situation when B starts and $`k`$ rounds are allowed.
We begin our consideration of the complexity of pointer jumping (in section 5) with the description of a classical randomized protocol for pointer jumping using communication $`O(n/k(\mathrm{log}^{(k/2)}n+\mathrm{log}k)+k\mathrm{log}n)`$ in the situation where Bob starts and only $`k`$ rounds are allowed. This upper bound is close to the known lower bounds for classical protocols .
The general strategy of our new lower bound for pointer jumping is to bound the value of a certain measure of information between the qubits of one player and the “next” pointer in terms of the analogous quantity for the previous pointer plus the average information on pointers in possession of the other player.
The mentioned protocol makes clear why the usual measure of information does not work in this approach. So (after defining the main notions of quantum computing in section 2 and the model of communication complexity in section 3) we introduce a new measure of quantum information in section 4. This measure is tied to the usual, von Neumann measure of quantum information by a theorem, which connects the trace distance between density matrices to their relative von Neumann entropy.
The lower bound on pointer jumping implies via reductions lower bounds for the $`k`$ round bounded error quantum communication complexity of the disjointness problem of the order $`\mathrm{\Omega }(n^{1/k})`$ for all constant $`k`$, see section 6.
We conclude from our result that quantum communication is dependent on interaction, as one should expect for a “realistic” mode of communication. We also conclude that good speedups by quantum protocols imply the use of nontrivial interaction in the case of total functions: for an asymptotic speedup by quantum Las Vegas protocols always more than one round is necessary . By the results in this paper (and similar results in ) rounds are also crucial in quantum speedups for the disjointness problem.
The lower bound for a subproblem of pointer jumping given in and the lower bound in this paper use at the basis of the proofs similar techniques. The main ingredient of the proof in , the “average encoding theorem”, follows directly from our theorem 1, which states a connection between a new measure of information (based on the trace distance) and the relative von Neumann entropy. We make use of a fact from , namely the “local transition theorem”. A combined version of both papers appears in .
Our results also hold in the model, in which prior entanglement is available.
The paper is organized as follows: In the next section we give some background on quantum mechanics. Then we define the communication model in section 3. In section 4 we consider measures of information and entropy. In section 5 we prove our results on the complexity of pointer jumping. Section 6 contains the lower bound for the disjointness problem.
## 2 Quantum States and Transformations
Quantum mechanics is a theory of reality in terms of states and transformations of states. See for general information on this topic with an orientation on quantum computing.
In quantum mechanics pure states are unit norm vectors in a Hilbert space, usually $`\text{ }\mathrm{C}^k`$. We use the Dirac notation for pure states. So a pure state is denoted $`|\varphi `$ or $`_{x\{0,\mathrm{},k1\}}\alpha _x|x`$ with $`_{x\{0,\mathrm{},k1\}}|\alpha _x|^2=1`$ and with $`\{|x|x\{0,\mathrm{},k1\}\}`$ being an orthonormal basis of $`\text{ }\mathrm{C}^k`$.
Inner products in the Hilbert space are denoted $`\varphi |\psi `$, outer (matrix valued) products $`|\varphi \psi |`$.
If $`k=2^l`$ then the basis is also denoted $`\{|x|x\{0,1\}^l\}`$. In this case the space $`\text{ }\mathrm{C}^{2^l}`$ is the $`l`$-wise tensor product of the space $`\text{ }\mathrm{C}^2`$. The latter space is called a qubit, the former space consists of $`l`$ qubits.
Usually also mixed states are considered.
###### Definition 1
Let $`\{(p_i,|\varphi _i)|i=1,\mathrm{},k\}`$ with $`_ip_i=1`$ and $`p_i[0,1]`$ be an ensemble of pure states of a quantum system, also called a mixed state.
$`\rho _i=|\varphi _i\varphi _i|`$ is the density matrix of the pure state $`|\varphi _i`$. $`_ip_i\rho _i`$ is the density matrix of the mixed state.
For a bipartite system with density matrix $`\rho _{AB}`$ denote $`\rho _A=trace_B\rho _{AB}`$.
As usual measurements of certain observables and unitary transformations are considered as basic operations on states, see for definitions.
For all possible measurements on a mixed state the results are determined by its density matrix. In quantum mechanics the density matrix plays an analogous role to the density function of a random variable in probability theory. Note that a density matrix is Hermitian, positive semidefinite and has trace 1. Thus it has only real, nonnegative eigenvalues that sum to 1.
Linear transformations on density matrices are called superoperators.
###### Definition 2
A superoperator is a linear map on density matrices. A superoperator is positive, if it sends positive semidefinite matrices to positive semidefinite matrices. A superoperator is called completely positive, if its tensor product with the identity superoperator is positive on density matrices over each finite dimensional extension of the underlying Hilbert space.
Trace-preserving completely positive superoperators map density matrices to density matrices and capture all physically allowed transformations. These include unitary transformations, tracing out subsystems, forming a tensor product with some constant qubits, and general measurements.
The following important fact characterizes the allowed superoperators in terms of unitary transformations and tracing out (see ). This fact is known as the Kraus representation theorem.
###### Fact 1
The following statements are equivalent:
1. A superoperator $`T`$ sending density matrices over $`H_1`$ to density matrices over $`H_2`$ is trace preserving and completely positive.
2. There is a Hilbert space $`H_3`$ with $`dim(H_3)dim(H_1)`$ and a unitary transformation $`U`$, such that for all density matrices $`\rho `$ over $`H_1`$ the following holds:
$$T\rho =trace_{H_1H_3}[U(\rho |0_{H_3H_2}0_{H_3H_2}|)U^{}].$$
So allowed superoperators can be simulated by adding some blank qubits, applying a unitary transformation and tracing out, i.e., “dropping some qubits”.
###### Definition 3
A purification of a mixed state with density matrix $`\rho `$ over some Hilbert space $`H`$ is any pure state $`|\varphi `$ over some space $`HK`$ such that $`trace_K|\varphi \varphi |=\rho `$.
## 3 The Communication Model
In this section we provide definitions of the computational models considered in the paper. We begin with the model of classical communication complexity.
###### Definition 4
Let $`f:\{0,1\}^n\times \{0,1\}^n\{0,1\}`$ be a function. In a communication protocol player Alice and Bob receive $`x`$ and $`y`$ and compute $`f(x,y)`$. The players exchange binary encoded messages. The communication complexity of a protocol is the worst case number of bits exchanged for any input. The deterministic communication complexity $`D(f)`$ of $`f`$ is the complexity of an optimal protocol for $`f`$.
In a randomized protocol both players have access to public random bits. The output is required to be correct with probability $`1ϵ`$ for some constant $`ϵ`$. The randomized communication complexity of a function $`R_ϵ(f)`$ is then defined analogously to the deterministic communication complexity. We define $`R(f)=R_{1/3}(f)`$.
A protocol has $`k`$ rounds, if the players exchange $`k`$ messages with Alice and Bob alternating as speakers. In message $`k+1`$ one player announces the result. Alice protocol is called one-way if only one players sends a message, and then the other player announces the result. The complexity notations are superscripted with the number of allowed rounds and eventually with the player starting, like $`D^{k,B}`$ or $`D^k`$ (usually Alice starts).
Now we define quantum communication protocols. For general information on quantum computation see and .
###### Definition 5
In a quantum protocol both players have a private set of qubits. Some of the qubits are initialized to the input before the start of the protocol, the other qubits are in state $`|0`$. In a communication round one of the players performs some unitary transformation on the qubits in his possession and then sends some of his qubits to the other player (the latter step does not change the global state but rather the possession of individual qubits). The choice of qubits to be sent and of unitary operations is fixed in advance by the protocol.
At the end of the protocol the state of some qubits belonging to one player is measured and the result is taken as the output. The communication complexity of the protocol is the number of qubits exchanged.
In a (bounded error) quantum protocol the correct answer must be given with probability $`1ϵ`$ for some $`1/2>ϵ>0`$. The (bounded error) quantum complexity of a function, called $`Q_ϵ(f)`$, is the complexity of an optimal protocol for $`f`$. $`Q(f)=Q_{1/3}(f)`$.
In fact we will consider a more general model of communication complexity, in which the players can apply all physically allowed superoperators to their private qubits. But due to the Kraus representation theorem (see fact 1) this model can be simulated by the above model without increasing communication (with the help of additional private qubits).
We have to note that in the defined model no intermediate measurements are allowed to control the choice of qubits to be sent or the time of the final measurement. Thus for all inputs the same amount of communication and rounds is used. As a generalization one could allow intermediate measurements, whose results could be used to choose the qubits to be sent and possibly when to stop the communication protocol. One would have to make sure that the receiving player knows when a message ends.
A protocol with $`k`$ rounds in this more general model can be simulated while loosing a factor of at most $`k`$ in the communication: for each measurement the operations given by the Kraus representation theorem are used. The measurement’s result is then stored in some ancilla qubits. Now the global state is a superposition over the results and a superposition of the appropriate communications can be used as a communication. This superposition uses as many qubits as the worst case message of that round. This may be at most the complexity of the whole protocol, so the overall complexity increases by at most a factor of $`k`$. While this simulation may not be satisfactory in general, it suffices to keep our lower bound valid in the more general model.
In and a different model of quantum communication (the communication model with entanglement) is proposed. Alice and Bob may possess an arbitrary input-independent set of (entangled) qubits in the beginning. Then they communicate according to an ordinary quantum protocol. This model can be simulated by allowing first an arbitrary input-independent communication with no cost followed by a usual quantum communication protocol in which the cost is measured. The superdense coding technique of allows to transmit $`n`$ bits of classical information with $`n/2`$ qubits in this model.
###### Definition 6
The quantum bounded error communication complexity with entanglement and error $`ϵ`$ is denoted $`Q_ϵ^{pub}(f)`$. Let $`Q^{pub}(f)=Q_{1/3}^{pub}(f)`$.
For surveys on quantum communication complexity see and .
## 4 Quantum Information Theory
Our main result in the next section uses information theory arguments. First we define the classical notions of entropy and information.
###### Definition 7
Let $`X:\mathrm{\Omega }S`$ be a random variable on finite sets $`\mathrm{\Omega },S`$ (as usual the argument of $`X`$ is dropped). The density function (or distribution) of $`X`$ is $`p_X:S[0,1]`$, where $`p_X(x)`$ is the probability of the event $`X=x`$.
The entropy of $`X`$ is $`H(X)=_{xS}p_X(x)\mathrm{log}p_X(x)`$.
Let $`X,Y`$ be random variables over $`\mathrm{\Omega }`$. The joint density function of $`XY`$ is $`p_{XY}(x,y)`$. The information between $`X`$ and $`Y`$ is $`H(X:Y)=H(X)+H(Y)H(XY)`$.
We use the convention $`0\mathrm{log}0=0`$.
Now we define the quantum mechanical notions of entropy and information.
###### Definition 8
The von Neumann entropy of a density matrix $`\rho _X`$ is defined by $`S(X)=S(\rho _X)=trace(\rho _X\mathrm{log}\rho _X)`$. The relative von Neumann entropy between two density matrices $`\rho ,\sigma `$ of the same size is $`S(\rho ||\sigma )=trace(\rho (\mathrm{log}\rho \mathrm{log}\sigma ))`$. This value may be infinite.
The von Neumann information is $`S(X:Y)=S(X)+S(Y)S(XY)`$ (see also ). Here $`S(X)`$ is the von Neumann entropy of the reduced density matrix $`\rho _X=trace_Y\rho _{XY}`$.
The conditional von Neumann information is $`S(X:Y|Z)=S(XZ)+S(YZ)S(Z)S(XYZ)`$.
Note that the von Neumann entropy depends only on the eigenvalues of a matrix and is thus invariant under unitary transformations. If the underlying Hilbert space has dimension $`d`$ then the von Neumann entropy of the density matrix is bounded by $`\mathrm{log}d`$.
Not all relations in classical information theory hold for von Neumann entropy. The following fact contains the so-called Araki-Lieb inequality (\*) and its consequences, which describes a notable difference to classical entropy (see ).
###### Fact 2
For all bipartite states $`\rho _{XY}`$:
$`S(X)+S(Y)S(XY)\stackrel{}{}|S(X)S(Y)|`$,
$`S(X:Y)2S(X)`$.
Then also $`S(X:Y|Z)S(X:YZ)2S(X)`$ holds.
The following is an important property of the von Neumann entropy, see . This property is known as the Lindblad-Uhlmann monotonicity of the von Neumann entropy.
###### Fact 3
For all trace-preserving, completely positive superoperators $`F`$ and all density matrixes $`\rho ,\sigma `$:
$$S(\rho ||\sigma )S(F\rho ||F\sigma ).$$
We are going to introduce another measure of information based on the distinguishability between a bipartite state and the state described by the tensor product of its two reduced density matrices. Now we consider measures of distinguishability. One such measure is the relative entropy. For probability distributions the total variational distance is another useful measure.
###### Definition 9
If $`p,q`$ are probability distributions on $`\{1,\mathrm{},n\}`$, then their distance is defined
$$pq=\underset{i=1}{\overset{n}{}}|p(i)q(i)|.$$
The following norm on linear operators is considered in .
###### Definition 10
Let $`\rho `$ be the matrix of a linear operator. Then the trace norm of $`\rho `$, denoted $`\rho _1`$, is the sum of the absolute values of the elements of the multiset of all eigenvalues of $`\rho `$. In particular $`\rho _1=Tr(\sqrt{\rho ^{}\rho })`$.
Note that the distance $`\rho \sigma _1`$ is a real value for Hermitian matrices $`\rho ,\sigma `$. The trace norm has a close relation to the measurable distance between states as shown in .
###### Fact 4
For an observable $`O`$ and a density matrix $`\rho `$ denote $`p_\rho ^O`$ the distribution on the outcomes of a measurement as induced by $`O`$ on the state $`\rho `$.
$$\rho \sigma _1=\underset{O}{\mathrm{max}}\{p_\rho ^Op_\sigma ^O\}.$$
So two density matrices that are close in the trace distance cannot be distinguished well by any measurement.
The next lemma is related to fact 4 and follows from fact 1.
###### Lemma 1
For each Hermitian matrix $`\rho `$ and each trace-preserving completely positive superoperator $`F`$:
$$\rho _1F(\rho )_1.$$
We employ the following theorem to bound the trace distance in terms of relative entropy. A classical analogue of the theorem can be found in and has been used e.g. in .
###### Theorem 1
For density matrices $`\rho ,\sigma `$ of the same size:
$$S(\rho ||\sigma )\frac{1}{2\mathrm{ln}2}||\rho \sigma ||_1^2.$$
Proof: Since both the norm and the relative entropy are invariant under unitary transformations we assume that the basis of the density matrices diagonalizes $`\rho \sigma `$. Note that in general neither $`\rho `$ nor $`\sigma `$ are diagonal now. Let $`S`$ be the multiset of all nonnegative eigenvalues of $`\rho \sigma `$ and $`R`$ the multiset of all its negative eigenvalues. All eigenvalues are real since $`\rho \sigma `$ is Hermitian. Now if the dimension of the space $`H_S`$ spanned by the eigenvectors belonging to $`S`$ has dimensions $`k`$ and the space $`H_R`$ spanned by the eigenvectors belonging to $`R`$ has dimensions $`nk`$, then increase the size of the underlying Hilbert space so that both spaces have the same dimension $`n^{}=\mathrm{max}\{k,nk\}`$. The density matrices have zero entries at the corresponding positions. Now we view the density matrices as density matrices over a product space $`H_2H_n^{}`$, where the $`H_2`$ space “indicates” the space $`H_S`$ or $`H_R`$.
We trace out the space $`H_n^{}`$ in $`\rho ,\sigma ,\rho \sigma `$. The obtained $`2\times 2`$ matrices are $`\stackrel{~}{\rho },\stackrel{~}{\sigma },\stackrel{~}{\rho \sigma }`$. Note that the matrix $`\stackrel{~}{\rho \sigma }`$ is diagonalized and contains the sum of all nonnegative eigenvalues, and the sum of all negative eigenvalues on its diagonal. Furthermore $`\stackrel{~}{\rho \sigma }=\stackrel{~}{\rho }\stackrel{~}{\sigma }`$.
Due to Lindblad-Uhlmann monotonicity of the relative von Neumann entropy we get $`S(\rho ||\sigma )S(\stackrel{~}{\rho }||\stackrel{~}{\sigma })`$. We will bound the latter by
$$1/(2\mathrm{ln}2)\stackrel{~}{\rho }\stackrel{~}{\sigma }_1^2=1/(2\mathrm{ln}2)\stackrel{~}{\rho \sigma }_1^2,$$
and then conclude the theorem since the trace norm of $`\stackrel{~}{\rho \sigma }`$ is the sum of absolute values of its eigenvalues which is the sum of absolute values of eigenvalues of $`\rho \sigma `$ by construction, i.e., $`\stackrel{~}{\rho }\stackrel{~}{\sigma }_1=\rho \sigma _1`$.
So we have to prove the theorem only for $`2\times 2`$ density matrices. Assume that the basis is chosen so that $`\sigma `$ is diagonal. Then
$$\rho =\left(\begin{array}{cc}a\hfill & b\hfill \\ b^{}\hfill & 1a\hfill \end{array}\right)\text{ and }\sigma =\left(\begin{array}{cc}c\hfill & 0\hfill \\ 0\hfill & 1c\hfill \end{array}\right).$$
The relative von Neumann entropy $`S(\rho ||\sigma )=S(\rho )trace[\rho \mathrm{log}\sigma ]`$. The second term is $`a\mathrm{log}c(1a)\mathrm{log}(1c)`$.
The first term is minus the entropy of the distribution induced by the eigenvalues of $`\rho `$. So we compute the eigenvalues.
The eigenvalues of $`\rho `$ are the zeroes of its characteristic polynomial $`t^2t+a(1a)bb^{}`$. These are $`1/2\pm \sqrt{1/4a(1a)+bb^{}}`$. Let $`x=1/2+\sqrt{1/4a(1a)+bb^{}}`$. Then $`S(\rho )=H(x).`$
The squared norm of $`\rho \sigma `$ is the squared sum of absolute values of the eigenvalues of $`\rho \sigma `$. That matrix has the characteristic polynomial $`t^2(a(1a)+a(1c)+(1a)cc(1c)+bb^{})`$. Thus its eigenvalues are $`\pm \sqrt{a(1a)+a(1c)+(1a)cc(1c)+bb^{}}`$. The squared norm as squared sum of absolute values of the eigenvalues is
$$4(a^2+c^22ac+bb^{}).$$
First we consider the case $`a=c`$. To prove this case we have to show that $`H(a)H(x)2\mathrm{log}(e)bb^{}=2\mathrm{log}(e)[(x1/2)^21/4+a(1a)]=2\mathrm{log}(e)[(x^2x)(a^2a)]`$.
Considering the function $`H(y)/\mathrm{log}(e)+2y^22y`$, we find that it is nonnegative and monotone decreasing for $`y`$ between $`1/2`$ and 1. Thus the inequality holds, when $`1/2a`$ and $`ax`$. The first condition can be assumed w.l.o.g., and the second condition follows from the fact that $`x1/2`$ is an eigenvalue and $`a1/2`$ is a diagonal element.
Now we look at the case $`ca`$. If $`c<a`$, we can use the same argument for $`1c`$ and $`1a`$ instead. We want to show that
$$f(c)=S(\rho ||\sigma )/\mathrm{log}(e)\frac{1}{2}||\rho \sigma ||_1^20.$$
We know this is true for $`a=c`$, so we show that increasing $`c`$ cannot decrease the difference. This holds since:
$`f^{}(c)`$ $`=`$ $`a/c+(1a)/(1c)2(2c2a)`$
$`=`$ $`{\displaystyle \frac{(1a)ca(1c)}{c(1c)}}4(ca)`$
$``$ $`4(ca)4(ca)0.`$
This yields the theorem for the $`2\times 2`$ case and thus in general by the previous considerations. $`\mathrm{}`$
Note that for a bipartite state $`\rho _{AB}`$ the following holds:
$$S(A:B)=S(\rho _{AB}||\rho _A\rho _B)\frac{1}{2\mathrm{ln}2}||\rho _{AB}\rho _A\rho _B||_1^2.$$
Thus the measurable distance between the tensor product state and the “real” bipartite state can be bounded in terms of the information. We will call the value $`D(A:B)=||\rho _{AB}\rho _A\rho _B||_1`$ the informational distance.
The next lemma collects a few properties of informational distance that follow easily from the previous discussion.
###### Lemma 2
For all states $`\rho _{ABC}`$ the following holds:
1. $`D(A:B)=D(B:A)`$.
2. $`D(AB:C)D(A:C)`$.
3. $`0D(A:B)2`$.
4. $`D(A:B)||F(\rho _{AB})F(\rho _A\rho _B)||_1`$ for all completely positive and trace-preserving superoperators $`F`$.
5. $`D(A:B)\sqrt{2S(A:B)}`$.
Note that lemma 2.5 implies one of the main ingredients of the round hierarchy discovered in (the “average encoding theorem”).
Consider some density matrix $`\rho _{AB}`$ that is block diagonal (with classical $`\rho _A`$) in the basis composed as the tensor product of the standard basis for $`A`$ and some other basis for $`B`$. Then denote $`\rho _B^{(a)}`$ the density matrix obtained by fixing $`A`$ to some classical value $`a`$ and normalizing. $`Pr(a)`$ is the probability of $`a`$.
The next properties of informational distance will be used later.
###### Lemma 3
1. Let $`\rho _{AB}`$ be the density matrix of a state, where $`\rho _B`$ corresponds to the density function of a classical random variable $`B`$ on $`|0`$ and $`|1`$ with $`Pr(B=1)=Pr(B=0)=1/2`$. Let there be a measurement acting on the $`A`$ system only and yielding a Boolean random variable $`X`$ with $`Pr(X=B)1ϵ`$ and $`Pr(XB)ϵ`$ (while the same measurement applied to $`\rho _A\rho _B`$ yields a distribution with $`Pr(X=B)=Pr(XB)=1/2`$). Then $`D(A:B)12ϵ`$.
2. For all block diagonal $`\rho _{AB}`$, where $`\rho _A`$ corresponds to a classical distribution $`p_A`$ on the standard basis vectors for $`A`$, the following holds:
$$D(A:B)=E_a||\rho _B^{(a)}\rho _B||_1.$$
Proof: For the first item observe that $`D(A:B)D(X:B)12ϵ`$.
The second item is a consequence of $`D(A:B)=||\rho _{AB}p_A\rho _B||_1.`$ $`\mathrm{}`$
## 5 Rounds in Quantum Communication
It is well known that for deterministic, probabilistic, (and even limited nondeterministic) communication complexity there are functions which can be computed much more efficiently in $`k`$ rounds than in $`k1`$ rounds (see , , , , ). In most of these results the pointer jumping function is considered.
###### Definition 11
Let $`V_A`$ and $`V_B`$ be disjoint sets of $`n`$ vertices each.
Let $`F_A=\{f_A|f_A:V_AV_B\}`$, and $`F_B=\{f_B|f_B:V_BV_A\}`$.
$`f(v)=f_{f_A,f_B}(v)=\{\begin{array}{cc}f_A(v)\hfill & \text{ if }vV_A,\hfill \\ f_B(v)\hfill & \text{ if }vV_B.\hfill \end{array}`$
Define $`f^{(0)}(v)=v`$ and $`f^{(k)}(v)=f(f^{(k1)}(v))`$.
Then $`g_k:F_A\times F_B(V_AV_B)`$ is defined by $`g_k(f_A,f_B)=f_{f_A,f_B}^{(k+1)}(v_1)`$, where $`v_1V_A`$ is fix. The function $`f_k:F_A\times F_B\{0,1\}`$ is the XOR of all bits in the binary code of the output of $`g_k`$.
Nisan and Wigderson proved in that $`f_k`$ has a randomized $`k`$ round communication complexity of $`\mathrm{\Omega }(n/k^2k\mathrm{log}n)`$ if B starts communicating and a deterministic $`k`$ round communication complexity of $`k\mathrm{log}n`$ if Alice starts. The lower bound can also be improved to $`\mathrm{\Omega }(n/k+k)`$, see . Nisan and Wigderson also describe a randomized protocol computing $`g_k`$ with communication $`O((n/k)\mathrm{log}n+k\mathrm{log}n)`$ in the situation, where Bob starts and $`k`$ rounds are allowed. Ponzio et. al. show that the deterministic communication complexity of $`f_k`$ is $`O(n)`$ then, if $`k=O(1)`$ .
With techniques similar to the ones in this paper we can also show a lower bound of $`\frac{(12ϵ)^2n}{2k^2}k\mathrm{log}n`$ for the randomized $`k`$ round complexity of $`f_k`$ when B starts, which is better than the above lower bounds for small constant values of $`k`$.
Interaction in quantum communication complexity has also been investigated by Nayak, Ta-Shma, and Zuckerman . For the pointer jumping function their results imply the following:
###### Fact 5
$`Q^{B,k}(f_k)=\mathrm{\Omega }(n^{1/k}/k^4)`$.
First we give a new upper bound. The next result combines ideas from and .
###### Theorem 2
$`R_ϵ^{k,B}(g_k)O(\frac{n}{kϵ}(\mathrm{log}^{(k/2)}n+\mathrm{log}k)+k\mathrm{log}n)`$.
Proof: First Bob guesses with public random bits $`(4/ϵ)(n/k)`$ vertices. For each chosen vertex $`v`$ Bob communicates the first $`\mathrm{log}^{(k/2)}n+3\mathrm{log}k`$ bits of $`f_B(v)`$.
In round $`t`$ the active player communicates the pointer value $`v_t=f^{(t1)}(v_1)`$. If it’s Alice’s turn, then she checks, whether $`v_t`$ is in Bob’s list of the first round. Then Alice knows $`\mathrm{log}^{(k/2)}n+3\mathrm{log}k`$ bits of $`f_B(v_t)`$. Note that this happens with probability $`1ϵ`$ during the first $`k/2`$ rounds. In the following assume that this happened in round $`ik/2`$, otherwise the protocol errs.
Beginning from the round $`i`$, when Alice gets to know the $`\mathrm{log}^{(k/2)}n+3\mathrm{log}k`$ bits of $`f(v_i)`$ the players communicate in round $`i+t`$ for all possible values of $`f(v_{i+t})`$ the most significant $`\mathrm{log}^{(k/2t)}n+3\mathrm{log}k`$ bits. Since there are at most $`n/(\mathrm{log}^{(k/2t)}nk^3)`$ such values $`O(n/k^2)`$ bits communication suffices. In the last round $`v_{k+2}`$ is found. Overall the communication is at most
$$k\mathrm{log}n+O((1/ϵ)(n/k)(\mathrm{log}^{(k/2)}n+3\mathrm{log}k))+kO(n/k^2).\text{ }\mathrm{}$$
###### Corollary 1
If $`k2\mathrm{log}^{}(n)`$ then $`R^{k,B}(g_k)O((\frac{n}{k}+k)\mathrm{log}k)`$.
We can replace $`k\mathrm{log}n`$ by $`k\mathrm{log}k`$ in the above expression, because that term dominates only if $`\mathrm{log}k=\mathrm{\Theta }(\mathrm{log}n)`$.
Next we are going to prove a lower bound on the quantum communication complexity of the pointer jumping function $`f_k`$, for the situation that $`k`$ rounds are allowed and Bob sends the first message. We will consider a quantity $`d_t`$ capturing the information the active player has in round $`t`$ on vertex $`t+1`$ of the path. This quantity will be the informational distance between the active player’s qubits and vertex $`t+1`$. Our goal will be to bound $`d_t`$ in terms of $`d_{t1}`$ plus a term related to the average information on pointers in the other player’s input. This leads to a recursion imposing a lower bound on the communication complexity, since in the end the protocol must have reasonably large information to produce the output, and in the beginning the respective information is 0.
The informational distance $`d_t`$ measures the distance between the state of, say, Alice’s qubits together with the vertex $`t+1`$ of the path, and the tensor product of the states of Alice’s qubits and vertex $`t+1`$. In the product state Alice has no information on the vertex, so if the two states are close, Alice’s powers to say something about the vertex are very limited. We will use the triangle inequality to bound $`d_t`$ by the sum of three intermediate distances. In the first step we move from the state given by the protocol to a state in which the $`t`$-th vertex is replaced by a uniformly random vertex, independent of previous communications. The distance to such a state can be bounded in terms of $`d_{t1}`$, because that quantity puts a bound on Bob’s ability to detect such a replacement. We use the local transition theorem from to conceal Alice’s ability to detect such a replacement. Once the $`t`$-th vertex is random, we can move in the next step to a state in which also vertex $`t+1`$ is random. The cost of this step corresponds to the average information a player has on a random pointer in the other player’s input. The last step is similar to the first and reverses the first one’s effect, i.e., replaces the “randomized” $`t`$-th vertex by its real value again.We arrive at the desired product state.
###### Theorem 3
$`Q^{k,B,pub}(f_k)n/2^{2^{O(k)}}k\mathrm{log}n`$.
Proof: Fix a quantum protocol, of the following form. The protocol computes $`f_k`$ with error $`\frac{1}{3}`$, $`k`$ rounds, Bob starting. At any time in the protocol Alice has access to qubits containing her input, some “work” qubits and some of the qubits used in messages so far, the same holds for Bob. We assume that the players never change their inputs. Usually a protocol gets some classical $`f_A`$ and $`f_B`$ as inputs, but we will investigate what happens if the protocol is started on a superposition over all inputs, in which all inputs have the same amplitude, i.e., on
$$\underset{f_A_A,f_B_B}{}\frac{1}{n^n}|f_A|f_B.$$
Note that $`|_A|=|_B|=n^n.`$ The superposition over all inputs is measured after the protocol has finished, so that a uniformly random input and the result of the protocol on that input are produced.
The density matrix of the global state of the protocol is $`\rho _{M_{A,t}M_{B,t}F_AF_B}`$. Here $`F_A,F_B`$ are the qubits holding the inputs of Alice and Bob and $`M_{A,t}`$ resp. $`M_{B,t}`$ are the other qubits in the possession of Alice and Bob before the communication of round $`t`$. The state of the latter two systems of qubits may be entangled. In the beginning these qubits are independent of the input.
We also require that in round $`t`$ the vertex $`v_t=f^{(t1)}(v_1)`$ is communicated by a classical message and stored by the receiving player. This increases the communication by an additive $`k\mathrm{log}n`$ term. We demand that before round $`t`$ the $`t`$’th vertex of the path is measured (remember that we are in a super-position over $`f_A,f_B`$). This vertex is stored in some qubits $`V_t`$. $`V_1`$ has the fixed value $`v_1`$. In general, before the beginning of round $`t`$, we have a mixture and in each pure state in the mixture the first $`t1`$ vertices are fixed and $`V_t`$ is either $`F_A(v_{t1})`$ or $`F_B(v_{t1})`$. We then measure $`V_t`$ in the standard basis. The measurements do not affect the correctness of the protocol.
Let us denote $`d_t=D(M_{B,t}F_B:F_A(V_t))`$ when $`t`$ is odd, and $`d_t=D(M_{A,t}F_A:F_B(V_t))`$ when $`t`$ is even. In this definition, we assume that the register $`F_A(V_t)`$ (or $`F_B(V_t)`$) has been measured, although this measurement is not part of the protocol. Note that for $`t>1`$, $`V_t`$ is uniformly random, so (for odd $`t`$) the distance $`d_t`$ is taken as the average over $`v`$, of the informational distance between the state of $`M_{B,t}F_B`$ restricted to $`V_t`$ being equal to $`v`$, and $`F_A(v)`$, and similarly for even $`t`$.
We assume that the communication complexity of the protocol is $`\delta n`$ and prove a lower bound $`\delta 2^{2^{O(k)}}`$. The general strategy of the proof is induction over the rounds, to successively bound $`d_1,d_2,\mathrm{},d_{k+1}`$.
Bob sends the first message. As Bob has seen no message yet, we have that $`I(M_{B,1}F_B:F_A(V_1))=0`$, and hence $`d_1=0`$. We show that
###### Lemma 4
$`d_{t+1}4\sqrt{d_t}+\sqrt{4\delta }`$.
We see that $`d_{t+1}3^t\delta ^{1/2^t}`$ for all $`t0`$. After round $`k`$ one player, say Alice, announces the result which is supposed to be the parity of $`F_B(V_{k+1})`$ and included in $`M_{A,k+1}`$. On the one hand $`d_{k+1}=D(M_{A,k+1}:F_B(V_{k+1}))3^k\delta ^{1/2^k}`$. On the other hand, by Lemma 3.1 $`D(M_{A,k+1}:F_B(V_{k+1}))1\frac{2}{3}=\frac{1}{3}`$. Together, $`\frac{1}{3}3^k\delta ^{1/2^k}`$, so $`\delta 2^{2^{O(k)}}`$.
We now turn to proving Lemma 4.
W.l.o.g. let Alice be active in round $`t+1`$. Let $`M_A=M_{A,t+1}`$ and $`M_B=M_{B,t+1}`$. Before the $`t+1`$ round $`V_{t+1}=F_A(V_t)`$ is measured. The resulting state is a probabilistic ensemble over the possibilities to fix $`V_1,\mathrm{},V_{t+1}`$, which are then classically distributed. Alice’s reduced state is block diagonal with respect to the possible values of the vertices $`V_1,\mathrm{},V_{t+1}`$. For any value $`v`$ of $`V_{t+1}`$ let $`\rho _{M_AM_BF_AF_B}^v=\rho _{M_AM_BF_AF_B}^{V_{t+1}=v}`$ denote the pure state with vertex $`V_{t+1}`$ fixed to $`v`$. We are interested in the value
$`d_{t+1}=D(M_AF_A:F_B(V_{t+1}))`$ $`=`$ $`𝐄_v\rho _{M_AF_AF_B(v)}^v\rho _{M_AF_A}^v\rho _{F_B(v)}_1,`$
where the distribution on vertices $`v`$ (induced by the state of the system $`F_A,F_B`$) is uniform. Recall that in the definition of $`d_{t+1}`$, $`F_B(v)`$ is assumed to be uniformly random (i.e., measured). The above quantity measures how much Alice knows about the value $`F_B`$ gives to the current vertex $`V_{t+1}`$.
We define
$`\gamma _v`$ $`\stackrel{\mathrm{def}}{=}`$ $`\rho _{M_BF_B}^v\rho _{M_BF_B}_1.`$ (1)
I.e., $`\gamma _v`$ is the distance between the state of Bob ($`\rho _{M_BF_B}^v`$) before he receives the message in round $`t+1`$, and the state $`\rho _{M_BF_B}`$, which is his state averaged over $`v=F_A(V_t)`$. We show below that these two are almost always close to each other (this reflects the fact that Bob does not know much about $`F_A`$).
For the purposes of the proof, we also consider a run of the protocol on the uniform superposition over inputs, where the qubits $`V_1,V_2,\mathrm{}`$ are not measured during the course of the protocol. Let $`\stackrel{~}{\rho }_{M_AM_BF_AF_B}`$ be the state before the communication in round $`t+1`$ in this run of the protocol. For any $`vV_B`$, we define:
$`\beta _v`$ $`\stackrel{\mathrm{def}}{=}`$ $`\stackrel{~}{\rho }_{M_AF_AF_B(v)}\stackrel{~}{\rho }_{M_AF_A}\rho _{F_B(v)}_1,`$ (2)
where $`F_B(v)`$ is assumed to have been measured. Note that $`\rho _{F_B(v)}=\stackrel{~}{\rho }_{F_B(v)}`$ with this measurement; both are randomly distributed over $`V_A`$.
Let $`\rho _{M_AM_BF_AF_BR}`$ (respectively, $`\rho _{M_AM_BF_AF_BR}^v`$) be a purification of $`\rho _{M_AM_BF_AF_B}`$ (resp. $`\rho _{M_AM_BF_AF_B}^v`$), where $`R`$ is some additional space used to purify the random path $`V_1,\mathrm{},V_{t+1}`$ (resp. $`V_1,\mathrm{},V_t`$).
We employ the following fact from (the “local transition theorem”). The fact is a variation of the impossibility result for unconditionally secure quantum bit commitment due to Mayers and Lo and Chau .
###### Fact 6
Let $`\rho _1,\rho _2`$ be two density matrices with support in a Hilbert space $`H`$, $`K`$ a Hilbert space of dimension at least $`dimH`$, and $`|\varphi _1,|\varphi _2,`$ any purifications of $`\rho _1`$ resp. $`\rho _2`$ in $`HK`$. Then there is a purification $`|\varphi _2^{}`$ of $`\rho _2`$ in $`HK`$, that is obtained by applying a unitary transformation $`IU`$ to $`|\varphi _2`$, where $`U`$ is acting on $`K`$ and $`I`$ is the identity operator on $`H`$.
$`|\varphi _2^{}`$ has the property
$$|\varphi _1\varphi _1||\varphi _2^{}\varphi _2^{}|_12\sqrt{\rho _1\rho _2_1}.$$
Now, due the above fact there is a local unitary transformation $`U_v`$ acting only on $`F_AM_AR`$ such that
$$\sigma _{M_AM_BF_AF_BR}^v\stackrel{\mathrm{def}}{=}U_v\rho _{M_AM_BF_AF_BR}U_v^{},$$
and $`\rho _{M_AM_BF_AF_BR}^v`$ are close to each other. Moreover,
###### Lemma 5
For all vertices $`vV_B`$,
$`\rho _{M_AF_A}^v\sigma _{M_AF_A}^v_1\rho _{M_AF_AF_B(v)}^v\sigma _{M_AF_AF_B(v)}^v_1`$ $``$ $`2\sqrt{\gamma _v},`$ (3)
$`\sigma _{M_AF_AF_B(v)}^v\sigma _{M_AF_A}^v\rho _{F_B(v)}_1`$ $``$ $`\beta _v.`$ (4)
We will also prove:
###### Lemma 6
For the uniform distribution on vertices $`v`$ (induced by the state of the system $`F_A,F_B`$),
$`𝐄_v\gamma _v`$ $``$ $`d_t,\text{ and}`$ (5)
$`𝐄_v\beta _v`$ $``$ $`\sqrt{4\delta }.`$ (6)
Thus, for all $`v`$:
$`\rho _{M_AF_AF_B(v)}^v\rho _{M_AF_A}^v\rho _{F_B(v)}_1`$
$`\begin{array}{ccc}& \rho _{M_AF_AF_B(v)}^v\sigma _{M_AF_AF_B(v)}^v_1\hfill & \\ & +\sigma _{M_AF_AF_B(v)}^v\sigma _{M_AF_A}^v\rho _{F_B(v)}_1\hfill & \\ & +\sigma _{M_AF_A}^v\rho _{F_B(v)}\rho _{M_AF_A}^v\rho _{F_B(v)}_1\hfill & \\ & 4\sqrt{\gamma _v}+\sigma _{M_AF_AF_B(v)}^v\sigma _{M_AF_A}^v\rho _{F_B(v)}_1\hfill & \text{From equation (}\text{3}\text{)}\hfill \\ & 4\sqrt{\gamma _v}+\beta _v\hfill & \text{From equation (}\text{4}\text{)}.\hfill \end{array}`$
Finally,
$`D(M_AF_A:F_B(V_{t+1}))`$
$`\begin{array}{ccc}=& 𝐄_v\rho _{M_AF_AF_B(v)}^v\rho _{M_AF_A}^v\rho _{F_B(v)}_1\hfill & \\ & 𝐄_v[4\sqrt{\gamma _v}+\beta _v]\hfill & \\ & 4\sqrt{𝐄_v\gamma _v}+𝐄_v\beta _v\hfill & \text{By Jensen’s inequality}\hfill \\ & 4\sqrt{d_t}+\sqrt{4\delta }\hfill & \text{By Lemma }\text{6}.\hfill \end{array}`$
This completes the proof of Lemma 4.
We finish the proof of Theorem 3 by proving Lemmas 5 and 6.
Proof of Lemma 5:
For equation (3) notice that
$`\rho _{M_AF_A}^v\sigma _{M_AF_A}^v_1`$ $``$ $`\rho _{M_AF_AF_B(v)}^v\sigma _{M_AF_AF_B(v)}^v_1`$
$``$ $`\rho _{M_AF_AM_BF_BR}^v\sigma _{M_AF_AM_BF_BR}^v_1,`$
and by fact 6 this is at most $`2\sqrt{\gamma _v}`$.
For equation (4),
$`\sigma _{M_AF_AF_B(v)}^v\sigma _{M_AF_A}^v\rho _{F_B(v)}_1`$
$`\begin{array}{ccc}& \sigma _{M_AF_ARF_B(v)}^v\sigma _{M_AF_AR}^v\rho _{F_B(v)}_1\hfill & \\ =& \rho _{M_AF_ARF_B(v)}\rho _{M_AF_AR}\rho _{F_B(v)}_1\hfill & \text{ By unitarity}\hfill \\ =& \stackrel{~}{\rho }_{M_AF_AF_B(v)}\stackrel{~}{\rho }_{M_AF_A}\rho _{F_B(v)}_1\hfill & ()\hfill \\ =& \beta _v\hfill & \text{ By definition (}\text{2}\text{)}.\hfill \end{array}`$
For ($``$) notice that $`R`$ holds the path $`V_1,\mathrm{},V_{t+1}`$, which is determined by $`M_AF_A`$. We can apply a unitary transformation that “erases” this, to give us the state $`\stackrel{~}{\rho }_{M_AF_A}`$. The lemma is proved.
Proof of lemma 6:
For equation (5), we have
$`𝐄_v\gamma _v`$ $`=`$ $`𝐄_u\rho _{M_BF_BF_A(u)}^{V_t=u}\rho _{M_BF_B}^{V_t=u}\rho _{F_A(u)}_1`$
$`=`$ $`D(M_BF_B:F_A(V_t))`$
$``$ $`D(M_{B,t}F_B:F_A(V_t))=d_t.`$
The last step follows from the fact that Bob sends the $`t`$’th message, and this only decreases the informational distance: $`D(M_{B,t+1}F_B:F_A(V_t))D(M_{B,t}F_B:F_A(V_t))`$.
To derive equation (6), we first bound the information Alice has on Bob’s input.
###### Lemma 7
For all $`t`$, $`I(M_{A,t}F_A:F_B)2\delta n`$, irrespective of whether some registers have been measured or not.
In the beginning Alice has no information about $`F_B`$, i.e., $`I(M_{A,1}F_A:F_B)=0`$. Recall that at most $`\delta n`$ qubits are communicated in the protocol. Any qubit sent from Alice to Bob does not increase her information on Bob’s input. Any local unitary transformation does not increase her information. Now assume Bob sends a qubit $`Q`$. Then $`I(M_AQF_A:F_B)=I(M_AF_A:F_B)+I(Q:F_B|M_AF_A)I(M_AF_A:F_B)+2`$ due to the Araki-Lieb inequality (fact 2) So each qubit sent from Bob to Alice increases her information on his input by at most 2. We thus get $`I(M_{A,t}F_A:F_B)2\delta n`$ at all times $`t`$. Since measurements only decrease mutual information, the bound also holds when certain registers are measured and others are not, during the course of the protocol. The lemma is proved.
Now consider the situation that $`F_B`$ is distributed uniformly instead of being in the uniform superposition (in other words, when $`F_B`$ has been measured). Then $`𝐄_vI(M_AF_A:F_B(v))2\delta `$ (where $`v`$ is uniformly random), using that the $`F_B(v)`$ are mutually independent. Now,
$`𝐄_v\beta _v`$ $`=`$ $`𝐄_v\stackrel{~}{\rho }_{M_AF_AF_B(v)}\stackrel{~}{\rho }_{M_AF_A}\rho _{F_B(v)}_1`$
$`=`$ $`𝐄_vD(M_AF_A:F_B(v)),`$
where $`F_AM_AM_BF_B`$ are as in the protocol without measurements. Further,
$`𝐄_vD(M_AF_A:F_B(v))`$ $``$ $`𝐄_v\sqrt{2I(M_AF_A:F_B(v))}`$
$``$ $`\sqrt{2𝐄_vI(M_AF_A:F_B(v))}`$
$``$ $`\sqrt{4\delta },`$
by Lemma 2 and Jensen’s inequality. $`\mathrm{}`$
## 6 The Disjointness Problem
We now investigate the bounded round complexity of the disjointness problem. Here Alice and Bob each receive the incidence vector of a subset of a size $`n`$ universe. They reject iff the sets are disjoint. It is known the $`Q_ϵ^1(DISJ_n)(1H(ϵ))n`$ . Furthermore $`Q(DISJ_n)=O(\sqrt{n}\mathrm{log}n)`$ by an application of Grover search . In this protocol $`\mathrm{\Theta }(\sqrt{n})`$ rounds are used. By a simple reduction (see ) we get the following result.
###### Theorem 4
$`Q^{k,pub}(DISJ_n)=\mathrm{\Omega }(n^{1/k})`$ for $`k=O(1)`$.
Proof: Suppose we are given a $`k`$ round quantum protocol for the disjointness problem having error $`1/3`$ and using communication $`c`$. W.l.o.g. we can assume Bob starts the communication, because the problem is symmetrical. We reduce the pointer jumping function $`f_k`$ to disjointness.
In a bipartite graph with $`2n`$ vertices and outdegree 1 there are at most $`n^k`$ possible paths of length $`k`$ starting at vertex $`v_1`$. For each such path we use an element in our universe for the disjointness problem. Given the left resp. right side of a specific graph Alice and Bob construct an instance of $`DISJ_{n^k}`$. Alice checks for each possible path of length $`k`$ from $`v_1`$ whether the path is consistent with her input and whether the paths leads to a vertex $`v_{k+2}`$ with odd number (if the $`k+1`$st vertex is on the left side). In this case she takes the corresponding element of the universe into her subset. Bob does the analogous with his input. Now, if the two subsets intersect, then the element in the intersection witnesses a length $`k+1`$ path leading to a vertex with odd number. If the subsets do not intersect, then the length $`k+1`$ path from $`v_1`$ leads to a vertex with even number.
So we obtain a $`k`$ round protocol for $`f_k`$ in which Bob starts. The communication is $`c=\mathrm{\Omega }(n)`$ for any constant $`k`$, the input length for the constructed instance of disjointness is $`N=n^k`$ and we get $`Q^{k,pub}(DISJ_N)=\mathrm{\Omega }(N^{1/k})`$ for $`k=O(1)`$. $`\mathrm{}`$
Acknowledgement
The author wishes to thank Ashwin Nayak and Amnon Ta-Shma for improving the presentation of theorem 3 and for helpful comments, as well as Pranab Sen for pointing out a mistake in an earlier version of the proof of theorem 1.
|
warning/0004/cond-mat0004379.html
|
ar5iv
|
text
|
# THE SPIN-1/2 𝑋𝑋𝑍 CHAIN AT FINITE MAGNETIC FIELD: CROSSOVER PHENOMENA DRIVEN BY TEMPERATURE
## 1 Introduction
In this contribution we report on a new crossover phenomenon observed in the longitudinal correlation function at finite magnetic field $`h`$ and finite temperature $`T`$ of the $`XXZ`$ spin-chain with anisotropy parameter $`0<\mathrm{\Delta }1`$.
As is well known any thermodynamic quantity derived from the free energy of the one-dimensional $`XXZ`$ model is an analytic function of finite $`h`$ and $`T`$. Phase transitions and associated mathematical singularities may and do occur in the ground state, i.e. at $`T=0`$. However, quantities not obtained from the free energy, but characterising the asymptotics of correlation functions may show “their own” non-analyticities at finite temperatures. Indeed the $`XXZ`$ chain in an external magnetic field does show well defined non-analyticities in the correlation lengths of the longitudinal spin-spin correlation functions. We want to point out that these crossover phenomena are a result of strong correlations and finite temperature. The crossover does not take place for the case $`\mathrm{\Delta }=0`$ corresponding to free fermions.
For the attractive regime $`1<\mathrm{\Delta }0`$ of the $`XXZ`$ chain the investigations are not yet carried out for finite field $`h`$ and finite $`T`$. Here the physical phenomena are expected to be much richer. Even for the case of vanishing magnetic field $`h=0`$ a sequence of crossovers (commensurate-incommensurate-commensurate) for an increase of $`T`$ from 0 to $`\mathrm{}`$ was found $`^\mathrm{?}`$.
The method we apply (as that used in $`^\mathrm{?}`$) is based on a mapping of the $`XXZ`$ chain at finite temperatures onto a classical model in two dimensions and the formulation of a suitable quantum transfer matrix (QTM) describing the transfer along the chain. The spectrum of the QTM succumbs to a Bethe ansatz (BA) treatment and the corresponding BA equations can be cast into the form of non-linear integral equations. Crossover phenomena of correlation lengths manifest themselves as level crossings of next-leading eigenvalues of the QTM. At finite field $`h>0`$ and sufficiently high temperature the relevant eigenvalues for the longitudinal correlation functions are 1-string and 2-string solutions (both solutions belong to the $`S=0`$ sector of the model). The truely next-leading eigenvalue is unique and given by the 1-string solution to the QTM taking real and negative values thus resulting into exponentially decaying correlations with antiferromagnetic oscillations. In some sense at sufficiently high temperature the properties of the system are determined by the longitudinal (“classical”) terms of the Hamiltonian, i.e. the $`S^zS^z`$ coupling and the field in $`z`$ direction, which dominate over the transversal exchange (“quantum mechanical”) terms. At sufficiently low temperature a different behavior is expected on grounds of predictions by conformal field theory (CFT)$`^{\mathrm{?},\mathrm{?}}`$. In particular, correlations with incommensurate $`2k_F`$ oscillations are expected$`^{\mathrm{?},\mathrm{?}}`$. As a consequence of this, the QTM has to develop complex conjugate pairs of eigenvalues at sufficiently low temperatures. This scenario has not been investigated before. The purpose of this contribution is to present the first “Bethe ansatz” study of this crossover phenomenon and to provide reasonably accurate values for the crossover temperature $`T_c`$. As pointed out already, for the free fermion case $`T_c=\mathrm{}`$, i.e. the crossover does not take place. In physical terms this may be understood in the way that due to the absence of longitudinal couplings the longitudinal terms never dominate over the transversal terms.
This report is organized as follows. In Sec. 2 we introduce some basic definitions of the $`XXZ`$ chain and present its properties at low temperature as obtained within conformal field theory (CFT). In Sec. 3 the approach to thermodynamic properties by use of a lattice path integral formulation and the quantum transfer matrix (QTM) is reviewed. In particular the set of non-linear integral equations for the two auxiliary functions $`𝔞(x)`$ and $`\overline{𝔞}(x)`$ corresponding to the energy density functions of spinons with spin $`\pm 1/2`$ are given. In Sec. 4 numerical results are given for the correlation length and Fermi momentum of the longitudinal spin-spin correlation function. Finally, the particle-hole picture resulting at low temperatures is discussed. A complete exposition of how this is related to the dressed charge formulation of CFT will appear in $`^\mathrm{?}`$.
## 2 Anisotropic Heisenberg model
The $`XXZ`$ model is defined by the Hamiltonian
$$H=J\underset{<i,j>}{}\left[S_i^xS_j^x+S_i^yS_j^y+\mathrm{\Delta }S_i^zS_j^z\right]h\underset{j}{}S_j^z,$$
(1)
where $`J>0`$, $`h>0`$ and $`S^{x,y,z}`$ are spin-1/2 operators. For $`T=0`$ the system is in one of three phases. For anisotropy parameter $`|\mathrm{\Delta }|>1`$ we have two ordered phases and for $`|\mathrm{\Delta }|1`$ a critical phase. In the latter case the anisotropy parameter is conveniently parameterized by $`\mathrm{\Delta }=\mathrm{cos}\gamma `$. In the following we take $`0<\gamma <\pi /2`$ (repulsive regime) for simplicity. The excitations are gapless spinons with spin $`1/2`$
$$ϵ_F(k)=v\mathrm{sin}k,0k\pi ,$$
(2)
$$v=\frac{\mathrm{sin}\gamma }{\gamma }\pi J.$$
(3)
At zero temperature the spin correlations decay algebraically$`^{\mathrm{?},\mathrm{?},\mathrm{?},\mathrm{?},\mathrm{?}}`$, with exponents depending on the magnetic field in a nontrivial way $`^\mathrm{?}`$. At finite temperature the correlations decay exponentially $`\mathrm{e}^{r/\xi }`$ with in general different correlation lengths $`\xi `$ for different correlation functions. At low temperatures the correlation lengths can be related by conformal mappings to the scaling dimensions $`x`$ of the fields at $`T=0`$
$$\xi =\frac{v}{2\pi xT}\text{ for}T1$$
The scaling dimensions $`x`$ are calculated from finite size corrections to the energy levels of the Hamiltonian and scaling predictions by CFT
$$E_xE_0=\frac{2\pi }{L}v(x+N^++N^{})+o\left(\frac{1}{L}\right),$$
(4)
where $`N^+`$ and $`N^{}`$ are integers labelling the conformal tower. The critical exponents of the spin-1/2 $`XXZ`$ model are those of a $`c=1`$ Gaussian theory. At zero magnetic field the exponents are given in terms of the anisotropy parameter $`\gamma `$
$$x=\frac{1\gamma /\pi }{2}S^2+\frac{1}{2(1\gamma /\pi )}m^2,$$
(5)
where $`S`$ and $`m`$ are integers corresponding to the $`S^z`$ component of the state and the number of excitations from the left to the right Fermi point, respectively. The longitudinal correlation is
$$S_0^zS_r^zC_1\frac{\mathrm{cos}(2k_Fr)}{r^{1/(1\gamma /\pi )}}\frac{C_2}{r^2},$$
(6)
where $`C_1`$ and $`C_2`$ do not depend on the distance $`r`$. For zero magnetic field the Fermi momentum $`k_F`$ is equal to $`\pi /2`$. For finite magnetic field and temperature we expect a deviation of $`x`$ and $`k_F`$ from the above quoted values.
## 3 Finite Temperatures
The properties of the quantum system at finite temperature are determined within a path integral formulation. The partition function of the 1d quantum system at finite temperature is mapped to that of a 2d classical system pictorially represented by
For the classical system we can choose a suitable transfer approach. The quantum transfer matrix (QTM), i.e. the column-to-column transfer matrix, possesses a unique largest eigenvalue and a spectral gap to the next-leading eigenvalues which persists even in the limit of infinite Trotter number as long as the temperature is finite$`^{\mathrm{?},\mathrm{?},\mathrm{?},\mathrm{?},\mathrm{?},\mathrm{?}}`$ . Hence the free-energy and correlation lengths are simply determined from the largest $`\mathrm{\Lambda }_0`$ and next largest eigenvalues $`\mathrm{\Lambda }_j`$ of the QTM
$`f`$ $`=`$ $`{\displaystyle \frac{1}{\beta }}lim\mathrm{ln}\mathrm{\Lambda }_0`$
$`C(r)`$ $`=`$ $`A_1\left({\displaystyle \frac{\mathrm{\Lambda }_1}{\mathrm{\Lambda }_0}}\right)^r+A_2\left({\displaystyle \frac{\mathrm{\Lambda }_2}{\mathrm{\Lambda }_0}}\right)^r+\mathrm{}.`$ (7)
with coefficients $`A_1`$, $`A_2`$ given by matrix elements. The diagonalization of the QTM can be performed by means of a Bethe ansatz resulting into eigenvalue equations of the form of Baxter’s $`\mathrm{\Lambda }q`$ equation. For finite Trotter number $`N`$ we find
$$\mathrm{\Lambda }(x)q(x)=\mathrm{e}^{\beta h/2}\mathrm{\Phi }(xi\gamma /2)q(x+i\gamma )+\mathrm{e}^{+\beta h/2}\mathrm{\Phi }(x+i\gamma /2)q(xi\gamma )$$
(8)
where
$`\mathrm{\Phi }(x)`$ $`=`$ $`[\mathrm{sinh}(xix_0)\mathrm{sinh}(x+ix_0)]^{N/2},x_0:={\displaystyle \frac{\gamma }{2}}{\displaystyle \frac{\beta }{N}}`$ (9)
$`q(x)`$ $`=`$ $`{\displaystyle \underset{j}{}}\mathrm{sinh}(xx_j)`$ (10)
The roots $`x_j`$ are determined from the Bethe ansatz equation
$$p(x_j)=1,\text{where}p(x):=\mathrm{e}^{\beta h}\frac{\mathrm{\Phi }(xi\gamma /2)q(x+i\gamma )}{\mathrm{\Phi }(x+i\gamma /2)q(xi\gamma )}$$
(11)
thus rendering $`\mathrm{\Lambda }(x)`$ analytic. We like to note that in general there are more solutions to the Bethe ansatz equation $`p(x)=1`$ than roots $`x_j`$. The additional solutions are called holes. For an illustration of a typical case see Fig 1.
The corresponding distributions of roots and holes remain discrete even in the limit of infinite Trotter number $`N\mathrm{}`$. The method of studying these equations was developed in $`^\mathrm{?}`$ and results into two equations for two auxiliary functions $`𝔞(x):=1/p(xi\gamma /2)`$ and $`\overline{𝔞}(x):=p(x+i\gamma /2)`$ (corresponding to $`S=1/2`$ spinon and antispinon dressed energy functions)
$$\mathrm{ln}𝔞(x)=\frac{v\beta }{\mathrm{cosh}\frac{\pi }{\gamma }x}+\frac{\pi \beta h}{2(\pi \gamma )}+[\kappa _1\mathrm{ln}(1+𝔞)](x)[\kappa _2\mathrm{ln}(1+\overline{𝔞})](xi\gamma +iϵ).$$
(12)
An analogous integral equation for $`\overline{𝔞}`$ can be obtained from the above one by use of the obvious identity $`\mathrm{ln}\overline{𝔞}(x)=\mathrm{ln}𝔞(x+i\gamma )`$ thus completing the non-linear integral equations.
The integral kernel $`\kappa (x)`$ is defined by a Fourier integral
$$\kappa (x):=\frac{1}{2\pi }_{\mathrm{}}^{\mathrm{}}\frac{\mathrm{sinh}(\frac{\pi }{2}\gamma )k\mathrm{cos}(kx)}{2\mathrm{cosh}\frac{\gamma }{2}k\mathrm{sinh}\frac{\pi \gamma }{2}k}\text{d}k.$$
(13)
The symbol $`_{}`$ denotes convolution
$$fg(x)=_{}f(xy)g(y)𝑑y$$
(14)
with a suitably defined integration contour $``$. In (12) the subscripts $`_1`$ ($`_2`$) refer to integration paths $`_1`$ and $`_2`$ that extend from $`\mathrm{}`$ to $`+\mathrm{}`$ and lie below the distribution of numbers $`x_j+i\gamma /2`$ and above $`x_ji\gamma /2`$. An additional and last requirement for these paths is that all roots $`x_j`$ be situated between $`_1i\gamma /2`$ and $`_2+i\gamma /2`$, but no hole solution to the Bethe ansatz equation.
For the largest eigenvalue the distribution of roots and holes is depicted in Fig. 1. Here the paths $`_1i\gamma /2`$ and $`_2+i\gamma /2`$ are just straight lines with imaginary parts $`\gamma /2`$. Hence, $`_{1,2}`$ coincide with the real axis.
The cases of next-leading excitations which are of 1-string type (root $`y_0`$ in the lower half plane) and 2-string type (upper and lower roots $`y_+`$ and $`y_{}`$ separated by approximately $`i\gamma `$) are shown in Figs. 2, 3. For the excited states the contours of integration are deformed. Only the path $`_1i\gamma /2`$ is shown explicitly, $`_2+i\gamma /2`$ is simpler as it is mostly following a straight line with imaginary part $`+\gamma /2`$ with loop in counter clockwise manner around $`y_0+i\gamma `$ and $`y_{}+i\gamma `$, respectively.
The contours may be straightened and by use of Cauchy’s theorem we pick up residues that lead to additive contributions to the driving terms in the non-linear integral equations $`^\mathrm{?}`$. This form of the NLIE is particularly useful for numerical treatments$`^\mathrm{?}`$.
Finally, the subsidiary conditions $`𝔞(y_0+i\gamma /2)=1`$ and $`𝔞(y_\pm +i\gamma /2)=1`$ with $`y_0`$, $`y_+`$ and $`y_{}`$ denoting the root of the 1-string and the upper and lower constituents of the 2-string, respectively, yield the information on the positions of the string parameters.
The eigenvalues of the QTM are expressed in terms of $`𝔞(x)`$, $`\overline{𝔞}(x)`$
$$\mathrm{ln}\mathrm{\Lambda }=\beta e_0+\frac{1}{2\gamma }\frac{\mathrm{ln}[(1+𝔞(x))(1+\overline{𝔞}(x))]}{\mathrm{cosh}\frac{\pi }{\gamma }x}𝑑x,$$
(15)
where $`e_0`$ is the groundstate energy for zero magnetic field and the integration contours are $`_{1,2}`$. However, due to certain analyticity properties of the integrand these contours can be simplified to just one contour along the real axis, surrounding the numbers $`\theta +i\gamma /2`$ in clockwise manner where $`\theta `$ is any of the hole solutions to the Bethe ansatz equation close to the real axis.
The range of validity (convergence) of the integral equation (12) for $`𝔞(x)`$ is the strip Im$`(x)[0,\gamma ]`$. Sometimes it is necessary to extend the above equation to the strip Im$`(x)[(\pi \gamma ),0]`$, e.g. for calculating $`𝔞(y_0+i\gamma /2)=1`$ with $`y_0`$ in the lower half plane, see Figs. 2. One version of such an expression for the 1-string pattern is
$`\mathrm{ln}𝔞(x)`$ $`=`$ $`{\displaystyle \frac{\pi \beta h}{\pi \gamma }}[r\mathrm{ln}(1+𝔞)](x)+[r\mathrm{ln}(1+\overline{𝔞})](xi\gamma )`$ (17)
$`+\mathrm{log}\left[{\displaystyle \frac{\mathrm{sinh}\frac{\pi }{\pi \gamma }(xy_0i3\gamma /2)}{\mathrm{sinh}\frac{\pi }{\pi \gamma }(xy_0+i\gamma /2)}}{\displaystyle \underset{j=1,2}{}}{\displaystyle \frac{\mathrm{sinh}\frac{\pi }{\pi \gamma }(x\theta _j+i\gamma /2)}{\mathrm{sinh}\frac{\pi }{\pi \gamma }(x\theta _ji\gamma /2)}}\right].`$
where
$$r(x)=\frac{i}{2(\pi \gamma )}\left(\mathrm{coth}\frac{\pi }{\pi \gamma }x\mathrm{coth}\frac{\pi }{\pi \gamma }(x+i\gamma )\right).$$
(18)
Note that the above expression for $`𝔞(x)`$ is valid for $`x`$ below the real axis, however $`𝔞`$, $`\overline{𝔞}`$ on the right hand side are evaluated for strictly real arguments, i.e. $``$ denotes standard convolution with the integration contour along the real axis. In particular, there are no deformations of the contour as discussed above. They have been removed by use of Cauchy’s theorem. In turn, the rapidities $`\theta _1`$, $`\theta _2`$ and the 1-string parameter $`y_0`$ show up explicitly. Note that the corresponding expression for the 2-string pattern is obtained by simply exchanging $`y_0`$ with $`y_{}`$ (the lower part of the 2-string). In this case, the upper part $`y_+`$ does not appear explicitly. The details will be presented in $`^\mathrm{?}`$.
## 4 Crossover scenario and the particle-hole picture
From the next-largest eigenvalue(s) $`\mathrm{\Lambda }_1`$ of the QTM the asymptotic behaviour of the correlation functions is determined. If $`\mathrm{\Lambda }_1`$ is real and positive (negative), the decay is purely exponential without (with) sublattice oscillations. In general $`\mathrm{\Lambda }_1`$ may be complex and the asymptotics are characterized by a finite correlation length $`\xi `$ and Fermi-momentum $`k_F`$ defined by
$$\frac{\mathrm{\Lambda }_1}{\mathrm{\Lambda }_0}=\mathrm{e}^{1/\xi \pm i2k_F},$$
We have numerically solved the nonlinear integral equations for the next-leading eigenvalue(s) and plot the results in Figs. 4, 5.
Summarizing our results we see that the previously drawn picture of excitations of distinctly 1-string and 2-string type is valid only for $`h=0`$ or, if $`h>0`$ for sufficiently high $`T`$. For this case the 1-string solution dominates and the eigenvalue is real because of the left-right symmetry of the Bethe ansatz pattern. The 2-string solution is subdominant. Both strings lie on the imaginary axis where for $`T=\mathrm{}`$ the 1-string is located at $`i\pi /2`$ and the 2-string is symmetric with respect to the real axis. For decreasing temperature we observe a characteristic motion of the 1-string upwards and the 2-string downwards along the imaginary axis, cf. Fig. 4. This motion continues until the root of the 1-string ($`y_0`$) and the lower root of the 2-string ($`y_{}`$) take identical values. The corresponding temperature defines $`T_c`$. For lower temperatures a horizontal motion sets in. The considered root previously on the imaginary axis develops a non-vanishing real part which may be positive or negative, see Fig. 4. The corresponding Bethe ansatz patterns are related by reflection at the imaginary axis, the corresponding eigenvalues are complex conjugate. There is no longer any qualitative distinction like for temperatures higher than $`T_c`$.
The reason for this crossover may be understood qualitatively in the following way. The subsidiary condition for the 1-string $`y_0`$ (or for the 2-string with $`y_0`$ replaced by $`y_{}`$) is $`𝔞(y_0+i\gamma /2)=1`$ which yields due to (17)
$`{\displaystyle \frac{\pi \beta h}{\pi \gamma }}+\mathrm{log}\left[{\displaystyle \frac{\mathrm{sinh}\frac{\pi }{\pi \gamma }(y_0\theta _1+i\gamma )}{\mathrm{sinh}\frac{\pi }{\pi \gamma }(y_0\theta _1)}}{\displaystyle \frac{\mathrm{sinh}\frac{\pi }{\pi \gamma }(y_0\theta _2+i\gamma )}{\mathrm{sinh}\frac{\pi }{\pi \gamma }(y_0\theta _2)}}\right]`$ (19)
$`=\text{“integral expressions”}.`$ (20)
Unfortunately, the right hand side has to be evaluated numerically. It turns out to be of order $`O(\beta h)`$, but smaller than the first term on the left hand side. In any case, the equation (20) for $`y_0`$ has two inequivalent solutions. These solutions are purely imaginary and different for small $`\beta h`$ ($``$ distinct 1- and 2-strings); they have same imaginary part but non-vanishing real parts with opposite signs for large $`\beta h`$ ($``$ degenerate 1- and 2-strings). The crossover is typically associated with square root singularities. The precise value of $`T_c`$ can only be calculated numerically.
The results for the correlation length and Fermi momentum are shown in Fig. 5. Most strikingly we see a non-analytic temperature dependence for non-vanishing external magnetic fields at the well-defined crossover temperature $`T_c`$! The singularity at $`T_c`$ is of square root type. Most significant is the non-analytic behaviour of the Fermi momentum. At low temperature the oscillations are incommensurate and at zero temperature the Fermi momentum $`k_F`$ and magnetization $`m`$ are strictly related by $`k_F=(1/2m)\pi `$, a relation which ceases to hold at elevated temperatures. At sufficiently high temperatures the oscillations are commensurate with $`k_F=\pi /2`$. The loss of the left-right symmetry in the Bethe ansatz patterns at low temperatures is the reason for the incommensurability of $`k_F`$ and the non-analytic behavior of the correlation length.
We note that temperature and magnetic field act in roughly opposite ways which can be inferred in part from their appearance in the combination $`h/T`$ in the NLIE. At sufficiently low temperatures the distinguishing characters of the 1-string and 2-string disappear completely and a root-hole picture emerges as shown in Fig 6. Here the relevant modifications of the excited state of the QTM in comparison to the ground state (see Fig. 1) can be characterized as a rearrangement of the roots and holes on an ellipsoidal curve. Clearly, the same constructions and terminology as used for the excitations of Fermi systems applies. In this finite field case ($`h>0`$), the analytic treatment of the low-temperature asymptotics not only confirms the predictions of CFT, but also recovers completely the dressed charge formalism as will be shown in$`^\mathrm{?}`$.
## Acknowledgments
The authors like to acknowledge valuable discussions with F. Eßler, H. Frahm, F. Göhmann, M. Inoue, K. Sakai, and J. Suzuki. A.K. acknowledges financial support by the Deutsche Forschungsgemeinschaft under grant No. Kl 645/3-2 and support by the research program of the Sonderforschungsbereich 341, Köln-Aachen-Jülich.
## References
|
warning/0004/hep-lat0004015.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The Overlap formalism provides a potential solution to the important problem of constructing chiral gauge theories nonperturbatively on the lattice. Gauge anomalies are a central issue in this context: One would like to show that a gauge-invariant formulation of chiral gauge theories on the lattice is possible when the usual (continuum) anomaly cancellation conditions are satisfied. Conversely, when these conditions are not satisfied, one would like to see the continuum anomalies emerge in the lattice formulation. In particular, an interesting test for a lattice formulation of chiral gauge theory is whether it can capture the global obstructions to gauge-invariance of the continuum theory, which reflect the topological structure of the determinant line bundle over the gauge orbit space .
In this paper we show that the overlap reproduces a basic class of such obstructions, described in the continuum by Alvarez-Gaumé and Ginsparg .<sup>1</sup><sup>1</sup>1 The possibility that this could happen had been previously mentioned in . This is a further demonstration of the ability of the overlap to reproduce topological features of the continuum theory.<sup>2</sup><sup>2</sup>2 The parity-invariant overlap formulation of vector gauge theory in odd dimensions reproduces the global gauge anomaly of the continuum theory , while for chiral gauge theory in even dimensions it has been shown to reproduce Witten’s global anomaly . As a consequence we will see that the usual (i.e. continuum) local anomaly cancellation condition ((1.4) below) is a necessary condition for local anomaly-free chiral gauge theory on the lattice in the overlap formulation.
Global obstructions to the vanishing of local gauge anomalies in the overlap formulation were previously considered by H. Neuberger in . An analogue of the the geometric perspective on the continuum chiral determinant as a section in a determinant line bundle was described for the overlap, and a class of global obstructions (which are naturally described in this setting) was explicitly constructed for the abelian theory. These were seen to vanish precisely when the fermion content of the theory satisfies
$`{\displaystyle \underset{\alpha }{}}e_\alpha ^3=0`$ (1.1)
where the $`e_\alpha `$’s label the irreducible U(1) representations of the fermion species. This is precisely the condition for cancellation of local gauge anomalies in the continuum theory. Thus, in the abelian case, the continuum anomaly cancellation condition (1.1) is a necessary condition for gauge-invariance of the lattice chiral determinant in the overlap formulation.<sup>3</sup><sup>3</sup>3 Herbert Neuberger has pointed out to me that these obstructions also arise for nonabelian gauge groups with U(1) subgroups, and that in four dimensions there are no additional restrictions in the nonabelian case, i.e. if all U(1) subgroups are free from anomalies then so is the nonabelian group. Our result is a nonabelian variant of this.<sup>4</sup><sup>4</sup>4 Although the class of obstructions that we consider is different: the ones in involve a torus in the orbit space of lattice gauge fields whereas ours involve a 2-sphere.
In the continuum theory, the global obstructions of Alvarez-Gaumé and Ginsparg arise as follows. Take spacetime to be the Euclidean 4-torus $`T^4`$ (the choice of dimension 4 is for concreteness; everything generalises to the $`T^{2n}`$ case for arbitrary $`n`$), gauge group $`SU(N)`$, and consider a family $`\varphi _\theta `$ of gauge transformations parameterised by $`\theta S^1`$. If the fermion is in the fundamental representation then each $`\varphi _\theta `$ is a map from $`T^4`$ to $`SU(N)`$, and the family of these corresponds to a map $`\mathrm{\Phi }:T^5SU(N)`$, $`\mathrm{\Phi }(\theta ,x)=\varphi _\theta (x)`$. The action of $`\varphi _\theta `$ on a gauge field $`A`$ determines a family $`\{A^\theta \}_{\theta S^1}`$. The winding number of the phase of the chiral determinant around this circle-family of gauge fields is an obstruction to gauge-invariance of the chiral determinant (since if the determinant is gauge-invariant then it is constant around the family $`\{A^\theta \}`$ and the winding number vanishes). In it was shown that this winding number equals the degree of the map $`\mathrm{\Phi }`$. Thus the obstruction is non-vanishing precisely when there exist maps $`\mathrm{\Phi }:T^5SU(N)`$ with non-vanishing degree (which happens, e.g., when $`N=3`$).<sup>5</sup><sup>5</sup>5In the spacetime was $`S^4`$ rather than $`T^4`$, and a condition $`\varphi _01`$ was imposed, which allows $`\mathrm{\Phi }`$ to be viewed as a map from $`S^5`$ to $`SU(N)`$. There is no essential difference with the present case though, since there is an isomorphism between the homotopy equivalence classes of $`\text{Map}(S^k,SU(N))`$ and $`\text{Map}(T^k,SU(N))`$.
In the general case where the fermion content is specified by some arbitrary (typically reducible) representation $`R`$ of $`SU(N)`$, the preceding generalises as follows. Instead of the degree of $`\mathrm{\Phi }`$, which is given by an expression of the form $`d^{abc}h_{abc}`$ where
$`d^{abc}=2\text{tr}((T^aT^b+T^bT^a)T^c)`$ (1.2)
and the $`T^a`$’s are the generators of $`SU(N)`$, the obstruction is given by $`d_R^{abc}h_{abc}`$ where $`d_R^{abc}`$ is given by (1.2) with $`T^a`$ replaced by $`R(T^a)`$ etc. Using the well-known fact that there is a relation of the form<sup>6</sup><sup>6</sup>6 The existence of a relation of this form can be seen as follows. Since the representation ring of $`SU(N)`$ is generated by the fundamental representation and its complex conjugate, it suffices to show (1.3) in the case where $`R`$ is a tensor product of copies of the fundamental representation. Then $`R(T^a)=T^a1\mathrm{}1+1T^a1\mathrm{}1+\mathrm{}+1\mathrm{}1T^a`$ etc, and it follows that $`(R(T^a)R(T^b)+R(T^b)R(T^a))R(T^c)=(T^aT^b+T^bT^a)T^c1\mathrm{}1+\mathrm{}+1\mathrm{}1(T^aT^b+T^bT^a)T^c+`$ terms which have a single $`T^a`$, $`T^b`$, or $`T^c`$ in one of the tensor slots. Since $`\text{tr}(T^a)=0`$ etc, it follows that the trace of these latter terms vanishes and we get (1.3).
$`d_R^{abc}=c(R)d^{abc}`$ (1.3)
we see that the obstruction in the general case is $`c(R)`$ times the degree of $`\mathrm{\Phi }`$. Thus in the case where $`\text{Map}(T^5,SU(N))`$ contains maps with non-vanishing degree a necessary condition for gauge-invariance of the chiral determinant is $`c(R)=0`$, or
$`d_R^{abc}=0`$ (1.4)
Of course, this is just the usual (necessary and sufficient) condition for anomaly cancellation in the continuum theory (the non-abelian analogue of (1.1)).
In this paper we consider a lattice version of the preceding obstructions in the overlap formulation, and show that they reduce to the continuum obstructions in the classical continuum limit. Since the lattice and continuum obstructions are both specified by integers, it follows that the lattice obstruction is exactly equal to the continuum one at small non-zero lattice spacing (i.e. close to the classical continuum limit). Our approach is similar to the recent analytic work of O. Bär and I. Campos on the lattice version of Witten’s global anomaly . When combined with the preceding observation (1.3), our result implies that (1.4) is a necessary condition for gauge-invariance of the lattice chiral determinant in the overlap formulation, at least in the case where $`\text{Map}(T^5,SU(N))`$ contains maps with non-vanishing degree.<sup>7</sup><sup>7</sup>7In higher dimensions there are cases where the maps all have vanishing degree yet the anomaly coefficient $`d^{a_1\mathrm{}a_n}`$ (the symmetrised trace of $`T^{a_1},\mathrm{},T^{a_n}`$) is non-vanishing. E.g. in dimension $`2n=6`$ with gauge group SU(3) one has $`\pi _7SU(3)=0`$ and $`d^{a_1a_2a_3a_4}0`$, cf. p.472 of .
It should be emphasised that global obstructions, and hence the results of this paper, are independent of the choice of phase in the overlap chiral determinant. In contrast, the consistent gauge anomaly for the overlap chiral determinant does depend on the phase choice. The consistent anomaly in the overlap formulation, and its classical continuum limit, has been previously studied in a number of works , although these have all involved some form of approximation (e.g. linearisation of the overlap) and/or assumptions (e.g. weak field, slowly varying field). No such approximations or assumptions are made in this paper.
The key question which these results lead on to (but which we do not pursue in this paper) is whether (1.4) is a sufficient condition for existence of a local anomaly-free lattice chiral gauge theory at non-zero lattice spacing. This is currently a topic of major interest and activity . To proceed with this in practice, a specific phase choice must be made to begin with. (The standard choice is the so-called Brillouin–Wigner phase .) One can then try to “improve” the starting phase choice in various ways to get a local anomaly-free overlap when the cancellation condition holds. One practical approach is to average along the gauge orbits (i.e. the FNN mechanism); see and the ref.’s therein.<sup>8</sup><sup>8</sup>8 The viability of this approach has been a topic of debate in the literature , although there is a body of evidence which is supportive of it —see, e.g., . Another, more theoretical, approach is to reduce the problem of going from an arbitrary starting phase choice to one for which the (local) anomalies vanish to that of solving a system of finite-difference equations on the lattice .<sup>9</sup><sup>9</sup>9 The formulation of ref.’s (a functional integral formulation based on a lattice Dirac operator satisfying the Ginsparg-Wilson relation , which had been rediscovered outside of the overlap setting in the work of Hasenfratz and collaborators ) is structurally identical to the overlap formulation after identifying the chiral fermion measures in the functional integral with the many-body groundstates in the overlap. More on this in §2, where the many-body groundstates are the “unit volume elements” in our terminology. In fact the integrability of these equations has been proved in the abelian case .<sup>10</sup><sup>10</sup>10 The argument in relied on a result on the structure of the abelian axial anomaly , which has been further elucidated in . There are strong indications that the same can be done in the nonabelian case , although a complete proof of this has not yet been given. Quite recently, a practically-oriented analytic prescription for constructing anomaly-free non-compact chiral U(1) gauge theory on the lattice has been given, starting from a adiabatic phase choice for the overlap .
In §2 we review the overlap construction of the chiral determinant. In §3 the lattice version of the global obstruction of Alvarez-Gaumé and Ginsparg is described, and is shown to reduce to the continuum obstruction in the classical continuum limit. The derivation of a key formula used to establish this is given separately in §4. This formula ((3.18) below) is due to Lüscher , and our detailed explicit derivation in §4 is intended to complement the rather brief argument in . In §5 we make some concluding remarks. Some details of our calculations are given in an appendix.
## 2 Overlap construction of the chiral determinant on the lattice
The spacetime is taken to be the Euclidean 4-torus $`T^4`$ with fixed edge length $`L`$. (Again, the choice of dimension 4 is for concreteness and simplicity; the arguments and results in the following generalise straightforwardly to the $`T^{2n}`$ case for arbitrary $`n`$.) We consider hyper-cubic lattices on $`T^4`$ with $`2N`$ sites along each edge and lattice spacing $`a=L/2N`$. <sup>11</sup><sup>11</sup>11This $`N`$ is of course not related to the $`N`$ in $`SU(N)`$. Given such a lattice, the space of lattice spinor fields $`\psi (x)`$ is denoted $`𝒞`$, and the space of lattice gauge fields $`U_\mu (x)`$ is denoted $`𝒰`$. The space $`𝒞`$ is finite-dimensional and comes equipped with an inner product:
$`\psi _1,\psi _2=a^4{\displaystyle \underset{x}{}}\psi _1(x)^{}\psi _2(x)`$ (2.1)
With suitable boundary conditions, the covariant forward and backward finite difference operators $`\frac{1}{a}_\mu ^\pm `$ act on $`𝒞`$ by
$`_\mu ^+\psi (x)`$ $`=`$ $`U_\mu (x)\psi (x+ae_\mu )\psi (x)`$ (2.2)
$`_\mu ^{}\psi (x)`$ $`=`$ $`\psi (x)U_\mu (xae_\mu )^1\psi (xae_\mu )`$ (2.3)
$`e_\mu `$ denotes the unit vector in the positive $`\mu `$-direction. We restrict to the case where $`U_\mu (x)`$ and $`\psi (x)`$ are periodic. This is the relevant case for considering the classical continuum limit with topologically trivial gauge fields. Since the chiral determinant vanishes in the topologically non-trivial case, this suffices for our purposes. Set $`_\mu =\frac{1}{2}(_\mu ^++_\mu ^{})`$; this operator is anti-hermitian with respect to the inner product (2.1) since $`(_\mu ^\pm )^{}=_\mu ^{}`$. The Wilson–Dirac operator is now given by
$`D_{Wilson}={\displaystyle \frac{1}{a}}\text{}+{\displaystyle \frac{r}{2}}a({\displaystyle \frac{1}{a^2}}\mathrm{\Delta })`$ (2.4)
where $`\text{}=_\mu \gamma ^\mu _\mu `$ (the $`\gamma ^\mu `$’s are taken to be hermitian so $`\text{}`$ is anti-hermitian), $`\mathrm{\Delta }=_\mu _\mu ^{}+_\mu ^+=_\mu (_\mu ^+)^{}_\mu ^+=_\mu (_\mu ^{})^{}_\mu ^{}`$ (hermitian, positive) and $`r>0`$ is the Wilson parameter. The hermitian operator
$`H(m)=\gamma _5(aD_{Wilson}rm)=\gamma _5(\text{}+r(\frac{1}{2}\mathrm{\Delta }m))`$ (2.5)
determines an orthogonal decomposition
$`𝒞=𝒞_+^{(m,U)}𝒞_{}^{(m,U)}`$ (2.6)
where $`𝒞_+^{(m,U)}`$ and $`𝒞_{}^{(m,U)}`$ are the subspaces spanned by the eigenvectors of $`H(m)`$ with positive and negative eigenvalues respectively. (We are restricting to the $`m,U`$ for which $`H(m)`$ has no zero-modes.) These subspaces are characterised by $`ϵ(m)=\pm 1`$ on $`𝒞_\pm ^{(m,U)}`$ where
$`ϵ(m)={\displaystyle \frac{H(m)}{\sqrt{H(m)^2}}}`$ (2.7)
(the dependence on $`U`$ has been suppressed). Noting that
$`ϵ(m)={\displaystyle \frac{\frac{1}{|rm|}H(0)\frac{m}{|m|}\gamma _5}{\sqrt{(\frac{1}{|rm|}H(0)\frac{m}{|m|}\gamma _5)^2}}}\gamma _5\text{for}m\mathrm{}`$ (2.8)
we see that in the $`m\mathrm{}`$ limit (2.6) reduces to the usual chiral decomposition
$`𝒞=𝒞_+𝒞_{}`$ (2.9)
independent of $`U`$. Set $`m=1`$ (the canonical value; $`0<m<2`$ would suffice) and let $`v_\pm `$ and $`w_\pm (U)`$ be unit volume elements<sup>12</sup><sup>12</sup>12A vectorspace $`V`$ determines vectorspaces $`\mathrm{\Lambda }^pV`$ ($`p=1,\mathrm{},dimV`$): the exterior algebra (=Grassmann algebra) of $`V`$ of degree $`p`$. An inner product in $`V`$ induces an inner product in each $`\mathrm{\Lambda }^pV`$. A “unit volume element on $`V`$” is an element $`v\mathrm{\Lambda }^dV`$ ($`d=dimV`$) with $`|v|=1`$. E.g. if $`v_1,\mathrm{},v_d`$ is an orthonormal basis for $`V`$ then $`v_1\mathrm{}v_d`$ is a unit volume element. Since $`\mathrm{\Lambda }^dV`$ is 1-dimensional, a unit volume element is unique up to $`\pm `$ if $`V`$ is real, or up to a phase if $`V`$ is complex. on $`𝒞_\pm `$ and $`𝒞_\pm ^{(1,U)}`$ respectively; these are unique up to phase factors. Then the lattice versions of the right- and left-handed chiral determinants in the overlap construction are, respectively,
$`v_+,w_+(U)`$ (right-handed) (2.10)
$`v_{},w_{}(U)`$ (left-handed) (2.11)
(see for background and motivation). The $`w_\pm (U)`$ are required to depend smoothly on $`U`$; then the overlaps (2.10)–(2.11) are smooth in $`U`$. Note that a condition for non-vanishing overlaps is $`dim𝒞_\pm ^{(1,U)}=dim𝒞_\pm d`$. The overlaps are unique up to a phase factor, and their norms are gauge-invariant (an easy consequence of the gauge-covariance of $`H(m)`$, $`ϵ(m)`$).
Remark 2.1. The construction of the overlaps (2.10)–(2.11) requires that $`H(1)`$ has no zero-modes. This can be guaranteed by imposing the condition
$`1U(p)\mathrm{\hspace{0.33em}0.04}p`$ (2.12)
on the lattice gauge field $`U`$, where $`U(p)`$ is the product of the link variables around a plaquette $`p`$. This condition is automatically satisfied in the classical continuum limit since $`1U(p_{x;\mu ,\nu })=a^2F_{\mu \nu }(x)+O(a^3)`$.
We henceforth restrict $`𝒰`$ to be the space of lattice gauge fields satisfying (2.12).
Remark 2.2. The overlaps (2.10)–(2.11) are determined (up to a phase) solely by $`ϵ=ϵ(1)`$. The construction could be carried through given any hermitian operator $`ϵ`$ with the property $`ϵ^2=1`$. The norms of the resulting overlaps would be gauge-invariant provided $`ϵ`$ is gauge-covariant.
Remark 2.3. The overlaps (2.10)–(2.11) can be written as<sup>13</sup><sup>13</sup>13 We are using the fact that a linear operator $`D:WV`$ induces linear operators $`\widehat{D}:\mathrm{\Lambda }^pW\mathrm{\Lambda }^pV`$ for all $`p`$, defined by $`\widehat{D}(w_1\mathrm{}w_p)=Dw_1\mathrm{}Dw_p`$. Note that if $`W=V`$ and $`d=dimV`$ then $`\widehat{D}(w_1\mathrm{}w_d)=detDw_1\mathrm{}w_d`$.
$`(\frac{2}{a})^dv_+,w_+`$ $`=`$ $`v_+,\widehat{D}w_+detD_L`$ (2.13)
$`(\frac{2}{a})^dv_{},w_{}`$ $`=`$ $`v_{},\widehat{D}w_{}detD_R`$ (2.14)
where
$`D={\displaystyle \frac{1}{a}}(\mathrm{\hspace{0.17em}1}+\gamma _5ϵ)`$ (2.15)
and $`ϵ=ϵ(1)`$ is given by (2.7). This follows easily from the facts that $`(1+\gamma _5ϵ)w=(1\pm \gamma _5)w`$ for $`w𝒞_\pm ^{(1,U)}`$ and $`(1\pm \gamma _5)v=2v`$ for $`v𝒞_\pm `$. The relations (2.13)–(2.14) show how the overlaps can be viewed as chiral determinants in an analogous way to the continuum setting: Set $`\widehat{\gamma _5}=\gamma _5(1aD)=ϵ`$, then $`\widehat{\gamma _5}^2=1`$ and $`D\widehat{\gamma _5}=\gamma _5D`$, which implies that $`D`$ maps $`\widehat{C}_{}:=𝒞_\pm ^{(1,U)}`$ to $`𝒞_\pm `$. Thus, modulo the factors $`(\frac{2}{a})^d`$, the right-handed overlap can be viewed as a left-handed chiral determinant, and vice-versa, as indicated in (2.13)–(2.14).<sup>14</sup><sup>14</sup>14 A careful consideration of the overlap prescription shows that the overlaps $`v_\pm ,w_\pm `$ really should be multiplied by a factor $`(\frac{2}{a})^d`$ as in (2.13)–(2.14). These factors are physically irrelevant though: they appear both in the numerator and denominator in expressions for physical expectation values, and hence cancel out, and they do not affect anomalies since these only have to do with the phase of the overlaps. Nevertheless, they are relevant if one considers the chiral determinant on its own and wishes to use the lattice regularisation as an alternative to, e.g., zeta-regularisation. In this form the overlap arises as the chiral determinant in Lüscher’s formulation after identifying the unit volume elements $`v_\pm `$ and $`w_\pm `$ (the many-body groundstates in the overlap) with the chiral fermion measures. These observations have been pointed out previously in . As mentioned there, it is easy to see that an operator $`D`$ is of the form (2.15), with $`ϵ`$ hermitian and $`ϵ^2=1`$, if and only if it satisfies the the following two conditions:
$`D\gamma _5+\gamma _5D`$ $`=`$ $`aD\gamma _5D\text{(Ginsparg–Wilson relation }\text{[21]}\text{)}`$ (2.16)
$`D^{}`$ $`=`$ $`\gamma _5D\gamma _5\text{(}\gamma _5\text{–hermiticity)}`$ (2.17)
Also, clearly $`D`$ is gauge-covariant if and only if $`ϵ`$ is gauge-covariant. The operator (2.15), with $`ϵ=ϵ(1)`$ given by (2.7), is the Overlap Dirac operator introduced by Neuberger in . It should also be mentioned that the Ginsparg–Wilson relation was rediscovered outside of the overlap setting in the work of P. Hasenfratz and collaborators —they considered a different solution, the so-called perfect Dirac operator .
The nullspace of $`D`$ is invariant under $`\gamma _5`$ (this follows from the GW relation (2.16): $`D\psi =0D(\gamma _5\psi )=(aD\gamma _5D\gamma _5D)\psi =0`$) so $`\text{index}D\text{Tr}(\gamma _5|_{\mathrm{ker}D})`$ is well-defined, as was first noted in . We only need to consider the lattice gauge fields $`U`$ for which $`dim𝒞_\pm ^{(1,U)}=dim𝒞_\pm d`$, since the overlaps vanish otherwise. As noted in , this corresponds to having $`\text{index}D=0`$. Therefore, we henceforth take $`𝒰`$ to be the space of lattice gauge fields satisfying (2.12) and $`\text{index}D=0`$.
## 3 Global obstructions to gauge-invariance of the Overlap
From now on we consider only the right-handed overlap $`v_+,w_+(U)`$ (the situation for the left-handed overlap is analogous). A lattice version of the obstructions considered by Alvarez-Gaumé and Ginsparg is as follows. Let $`\varphi _\theta `$ be a family of lattice gauge transformations parameterised by $`\theta S^1`$. We can assume that the fermion content is specified by the fundamental representation of $`SU(N)`$; it will be clear from what follows that the general case is related to this case in the same way as in the continuum setting discussed in the introduction. If $`U𝒰`$ is a lattice gauge field for which the overlap $`v_+,w_+(U)`$ is non-vanishing<sup>15</sup><sup>15</sup>15 It can be seen from (2.13) that $`v_+,w_+(U)`$ vanishes at the $`U`$ for which $`D`$ has zero-modes. Generically, these are isolated points in $`𝒰`$. then the action of $`\varphi _\theta `$ on $`U`$ determines a map
$`S^1𝐂\{0\},\theta v_+,w_+(\varphi _\theta U)`$ (3.1)
Since $`|v_+,w_+(U)|`$ is gauge-invariant, we have $`v_+,w_+(\varphi _\theta U)=e^{i\alpha (\theta )}v_+,w_+(U)`$ for some phase $`\alpha (\theta )`$, and the map (3.1) has integer winding number $`W(\mathrm{\Phi },U)=\frac{1}{2\pi }(\alpha (1)\alpha (0))`$. Obviously, if the winding number is non-vanishing $`v_+,w_+(U)`$ cannot be gauge-invariant. To see that this is a genuine obstruction to gauge-invariance we note that it is independent of the choice of $`w_+(U)`$: If $`\stackrel{~}{w}_+(U)`$ is another unit volume element on $`𝒞_+^{(1,U)}`$, smoothly varying with $`U`$, then $`\stackrel{~}{w}_+(U)=e^{i\beta (U)}w_+(U)`$ where the phase factor $`e^{i\beta (U)}`$ is smooth in $`U`$, and we have
$`v_+,\stackrel{~}{w}_+(\varphi _\theta U)=e^{i\alpha (\theta )+i\beta (\varphi _\theta U)}v_+,w_+(U)`$
Assuming that $`\{\varphi _\theta U\}_{\theta S^1}`$ is a contractible circle in $`𝒰`$ (which is certainly true close to the classical continuum limit), it follows that this has the same winding number as (3.1) since since $`e^{i\beta (U)}`$ is a smooth, non-vanishing, globally defined function of $`U`$. Hence the winding number $`W(\mathrm{\Phi },U)`$ is an obstruction to gauge-invariance of the overlap, independent of the choice of $`w_+`$.
Our main result is that this obstruction coincides with the continuum one at small non-zero lattice spacing close to the classical continuum limit:
Theorem. If $`\varphi _\theta `$ is the restriction to the lattice of a family of continuum gauge transformations (also denoted $`\varphi _\theta `$) and $`U`$ is the lattice transcript of a topologically trivial continuum gauge field, then there is an $`a_0>0`$ (depending on the $`\varphi _\theta `$’s and $`U`$) such that
$`W(\mathrm{\Phi },U)=\mathrm{deg}(\mathrm{\Phi })\text{for all}a<a_0`$ (3.2)
where $`\mathrm{deg}(\mathrm{\Phi })`$ is the degree of the continuum map $`\mathrm{\Phi }:T^5SU(N)`$ given by $`\mathrm{\Phi }(\theta ,x)=\varphi _\theta (x)`$.
In light of the discussion in the introduction we conclude from this that, in the general case where the fermion content is specified by a general representation $`R`$ of $`SU(N)`$, a necessary condition for existence of a gauge-invariant construction of the overlap is
$`d_R^{abc}=0`$ (3.3)
The remainder of the paper is concerned with the proof of the above theorem. We start by expressing the obstruction as
$`W(\mathrm{\Phi },U)`$ $`=`$ $`{\displaystyle \frac{1}{2\pi i}}{\displaystyle _0^1}𝑑\theta \frac{d}{d\theta }\mathrm{log}v_+,w_+(\varphi _\theta U)`$ (3.4)
$`=`$ $`{\displaystyle \frac{1}{2\pi i}}{\displaystyle _{𝒮^1}}d\mathrm{log}v_+,w_+`$
where $`𝒮^1`$ denotes the circle $`\{\varphi _\theta U\}_{\theta S^1}`$ in $`𝒰`$ and $`d`$ denotes the exterior derivative on functions (or more generally, differential forms) on $`𝒰`$. After noting that
$`dw_+=w_+,dw_+w_++(dw_+)_{}`$ (3.5)
where $`(dw_+)_{}`$ denotes the projection of $`dw_+`$ onto the orthogonal complement of $`w_+`$ in $`\mathrm{\Lambda }^d𝒞`$, one finds
$`d\mathrm{log}v_+,w_+={\displaystyle \frac{v_+,(dw_+)_{}}{v_+,w_+}}+w_+,dw_+`$ (3.6)
The first term on the right-hand side of (3.6) can be re-expressed as
$`{\displaystyle \frac{v_+,(dw_+)_{}}{v_+,w_+}}=\text{Tr}(dDD^1P_+)`$ (3.7)
where $`P_+=\frac{1}{2}(1+\gamma _5)`$ is the projection onto $`𝒞_+`$. This is a straightforward consequence of (2.13) and relations noted in ; for completeness we provide a derivation in the appendix. Then
$`d\mathrm{log}w_+,w_+=\text{Tr}(dDD^1P_+)+w_+,dw_+.`$ (3.8)
Set $`w_+^\theta =w_+(\varphi _\theta U)`$ and let $`D_\theta `$ denote $`D`$ with lattice gauge field $`\varphi _\theta U`$. The gauge-covariance of $`D`$ gives $`\frac{d}{d\theta }D_\theta =[\frac{d\varphi _\theta }{d\theta }\varphi _\theta ^1,D_\theta ]`$, leading to
$`\text{Tr}(\frac{d}{d\theta }D_\theta D_\theta ^1P_+)`$ $`=`$ $`\text{Tr}\left({\displaystyle \frac{d\varphi _\theta }{d\theta }}\varphi _\theta ^1P_+\right)\text{Tr}\left(D_\theta {\displaystyle \frac{d\varphi _\theta }{d\theta }}\varphi _\theta ^1D_\theta ^1P_+\right)`$ (3.9)
$`=`$ $`\text{Tr}\left({\displaystyle \frac{d\varphi _\theta }{d\theta }}\varphi _\theta ^1(P_+\widehat{P}_{})\right)`$
$`=`$ $`\frac{1}{2}a\text{Tr}\left({\displaystyle \frac{d\varphi _\theta }{d\theta }}\varphi _\theta ^1\gamma _5D_\theta \right)=\frac{1}{2}a\text{Tr}\left(\varphi _\theta ^1{\displaystyle \frac{d\varphi _\theta }{d\theta }}\gamma _5D_1\right)`$
where we have used the fact that $`P_+D=D\widehat{P}_{}`$ where $`\widehat{P}_{}=\frac{1}{2}(1\widehat{\gamma _5})`$, $`\widehat{\gamma _5}=\gamma _5(1aD)`$. Substituting (3.8) into (3.4) and using (3.9) we get
$`2\pi iW(\mathrm{\Phi },U)`$ $`=`$ $`{\displaystyle _{𝒮^1}}\text{Tr}(dDD^1P_+)+w_+,dw_+`$ (3.10)
$`=`$ $`{\displaystyle _0^1}𝑑\theta \left(\frac{1}{2}a\text{Tr}\left(\varphi _\theta ^1{\displaystyle \frac{d\varphi _\theta }{d\theta }}\gamma _5D_1\right)+w_+^\theta ,{\displaystyle \frac{dw_+^\theta }{d\theta }}\right)`$ (3.11)
We have derived this relation under the assumption that the overlap is non-vanishing for $`U`$, or equivalently, that the Overlap Dirac operator $`D`$ with lattice gauge field $`U`$ has no zero-modes. By construction $`W(\mathrm{\Phi },U)`$ is clearly smooth, and therefore locally constant, in such $`U`$. But it is ill-defined at the (generically isolated) points in $`𝒰`$ where the overlap vanishes. One such point is the trivial field $`U=1`$ (in this case the zero-momentum spinors with definite chirality are zero-modes for $`D`$). However, the right-hand side of (3.11) is clearly smooth in $`U`$ for all $`U𝒰`$ (since $`D`$ is smooth in the lattice gauge field when (2.12) is satisfied ), and must therefore be a locally constant function of $`U`$ for all $`U𝒰`$. In the continuum, any topologically trivial gauge field can be continuously deformed to the trivial field. It follows that when the lattice spacing $`a`$ is sufficiently small, the lattice transcript $`U`$ can be continuously deformed to the trivial lattice gauge field (using the lattice transcript of the continuum path). Therefore, to prove the theorem it suffices to show that there is an $`a_0>0`$ such that
$`W(\mathrm{\Phi })=\mathrm{deg}(\mathrm{\Phi })\text{for all}a<a_0`$ (3.12)
where $`W(\mathrm{\Phi })`$ denotes the right-hand side of (3.11) with trivial field $`U=1`$. In this case $`D_1`$ acts trivially in colour space, and since $`\varphi _\theta ^1\frac{d\varphi _\theta }{d\theta }`$ is in the Lie algebra of $`SU(N)`$ the trace over colour indices in the first term in (3.11) vanishes, resulting in
$`2\pi iW(\mathrm{\Phi })`$ $`=`$ $`{\displaystyle _0^1}𝑑\theta w_+^\theta ,\frac{d}{d\theta }w_+^\theta `$ (3.13)
$`=`$ $`{\displaystyle _{𝒮^1}}w_+,dw_+`$
A calculation gives
$`dw_+,dw_+`$ $`=`$ $`dw_+,dw_+`$ (3.14)
$`=`$ $`\text{Tr}(PdPdP)`$
where
$`P\widehat{P}_{}=\frac{1}{2}(1\widehat{\gamma _5})=\frac{1}{2}(1+ϵ)`$ (3.15)
with $`ϵ=ϵ(1)`$ given by (2.7). The last equality in (3.14) is derived in the appendix. A simpler version of it (originating in ) was used in ; the same relation in a different guise was subsequently noted in , and in more detail in . Now, by Stokes theorem, if $`^2`$ is a disc in $`𝒰`$ with boundary $`𝒮^1`$ it follows that
$`2\pi iW(\mathrm{\Phi })`$ $`=`$ $`{\displaystyle _^2}\text{Tr}(PdPdP)`$ (3.16)
$`=`$ $`{\displaystyle _^2}\text{Tr}(P[_\theta P,_tP])𝑑\theta 𝑑t`$
where $`(\theta ,t)`$ are taken to be polar coordinates on a unit disc $`B^2`$ parameterising $`^2`$. We take $`^2`$ to be the lattice transcript of a disc-family of continuum gauge fields, also denoted $`^2`$, given by
$`A^{(\theta ,t)}=f(t)\varphi _\theta d_x\varphi _\theta ^1(\theta ,t)B^2`$ (3.17)
where $`f(t)`$ is an arbitrary smooth function equal to 1 in a neighbourhood of $`t=1`$ and vanishing in a neighbourhood of $`t=0`$. The lattice transcript $`U^{(\theta ,t)}`$ has the property $`U^{(\theta ,1)}=\varphi _\theta 1`$ so the boundary of $`^2`$ is $`𝒮^1`$ as required in (3.16). Note that (3.16) is manifestly independent of the choice of $`w_+(U)`$, i.e. independent of the choice of phase in the overlap.
A general formula for the classical continuum limit of the integrand in (3.16) has been given by Lüscher in : If $`U^{(s,t)}`$ is the lattice transcript of a family $`A^{(s,t)}`$ of continuum gauge fields, and $`P=P^{(s,t)}`$ is the corresponding family of projection operators (given by (3.15)), then
$`\underset{a0}{lim}\text{Tr}(P[_sP,_tP])={\displaystyle \frac{1}{32\pi ^2}}{\displaystyle _{T^4}}d^4xϵ_{\mu \nu \rho \sigma }d^{abc}_sA_\mu ^a(x)_tA_\nu ^b(x)F_{\rho \sigma }^c(x).`$ (3.18)
Using this, (3.12) follows easily from (3.16) as we will see below, thereby proving the theorem. In the locality, smoothness and symmetry properties of $`P`$ were used to show that the limit in (3.18) exists and is given by the integral over $`T^4`$ of a polynomial in the gauge fields and its derivatives. However, the explicit form of this polynomial (i.e. the integrand on the right-hand side of (3.18)) was not obvious, at least to the present author, from the brief argument in . Since (3.18) is an important formula in this context (it was also a key ingredient in the arguments of ref.’s and ) we will give a detailed, explicit derivation of it in §4. This is intended to complement the brief argument in .
By (3.18), the classical continuum limit of (3.16) is
$`\underset{a0}{lim}2\pi iW(\mathrm{\Phi })={\displaystyle \frac{1}{32\pi ^2}}{\displaystyle _{^2\times T^4}}𝑑\theta 𝑑td^4xϵ_{\mu \nu \rho \sigma }d^{abc}_\theta A_\mu ^a(x)_tA_\nu ^b(x)F_{\rho \sigma }^c(x).`$ (3.19)
with $`A=A^{(\theta ,t)}`$ given by (3.17). It remains to show that this is equal to $`2\pi i\mathrm{deg}(\mathrm{\Phi })`$. Then, since $`W(\mathrm{\Phi })`$ is integer, we can conclude that $`W(\mathrm{\Phi })=\mathrm{deg}(\mathrm{\Phi })`$ for all lattice spacings $`a`$ smaller than some $`a_0>0`$ and the theorem is proved. The right-hand side of (3.18) can be shown to equal $`2\pi i\mathrm{deg}(\mathrm{\Phi })`$ by a direct calculation , but it is easier and perhaps more illuminating to proceed indirectly as follows. We view the family $`A^{(\theta ,t)}`$ as a gauge field on $`B^2\times T^4`$:
$`𝒜(x,\theta ,t)=A_\mu ^{(\theta ,t)}(x)dx^\mu =f(t)\varphi _\theta (x)_\mu \varphi _\theta ^1(x)dx^\mu `$ (3.20)
and define another gauge field on $`\stackrel{~}{B}^2\times T^4`$ by
$`\stackrel{~}{𝒜}(x,\theta ,s)=f(s)_\theta \varphi _\theta ^1(x)\varphi _\theta (x)d\theta `$ (3.21)
where $`\stackrel{~}{B}^2`$ is another copy of the unit disc, with polar coordinates $`(\theta ,s)`$. $`B^2\times T^4`$ and $`\stackrel{~}{B}^2\times T^4`$ can be glued together along their common boundary $`S^1\times T^4`$ to get the closed manifold $`S^2\times T^4`$. On the common boundary $`S^1\times T^4`$ the fields $`𝒜`$ and $`\stackrel{~}{𝒜}`$ are related by a gauge transformation:
$`\stackrel{~}{𝒜}=\mathrm{\Phi }^1𝒜=\mathrm{\Phi }^1𝒜\mathrm{\Phi }+\mathrm{\Phi }^1(d_\theta +d_x)\mathrm{\Phi }`$
where $`\mathrm{\Phi }:𝒮^1\times T^4SU(N)`$ is given by $`\mathrm{\Phi }(\theta ,x)=\varphi _\theta (x)`$. Therefore $`𝒜`$ and $`\stackrel{~}{𝒜}`$ constitute a gauge field $`\widehat{𝒜}`$ on an $`SU(N)`$ bundle over $`S^2\times T^4`$ with topological charge $`\mathrm{deg}(\mathrm{\Phi })`$. The topological charge is also given by the integral of the Chern character over $`S^2\times T^4`$, thus
$`\mathrm{deg}(\mathrm{\Phi })`$ $`=`$ $`{\displaystyle \frac{i^3}{(2\pi )^33!}}{\displaystyle _{S^2\times T^4}}\text{tr}\widehat{}^3`$ (3.22)
$`=`$ $`{\displaystyle \frac{i^3}{(2\pi )^33!}}\left[{\displaystyle _{B^2\times T^4}}\text{tr}^3{\displaystyle _{\stackrel{~}{B}^2\times T^4}}\text{tr}\stackrel{~}{}^3\right]`$
The second term vanishes: $`\stackrel{~}{}^3=0`$ since $`\stackrel{~}{𝒜}`$ only involves the 1-form $`d\theta `$. Regarding the first term, from $`=(d_x+d_\theta +d_t)𝒜+𝒜𝒜`$ we get
$`\text{tr}^3`$ $`=`$ $`\text{tr}(d_\theta Ad_tAF+d_tAd_\theta AF+d_\theta AFd_tA+d_tAFd_\theta A+Fd_\theta Ad_tA+Fd_tAd_\theta A)`$
$`=`$ $`3\left({\displaystyle \frac{d^{abc}}{2}}\right)d_\theta A^ad_tA^bF^c`$
After substituting this for $`\text{tr}^3`$ in (3.22) we see from (3.19) that $`lim_{a0}W(\mathrm{\Phi })=\mathrm{deg}(\mathrm{\Phi })`$ as required.
## 4 A detailed derivation of Lüscher’s formula
In this section we give a detailed, explicit derivation of the formula (3.18). We begin by noting that
$`\text{Tr}(P[_sP,_tP])`$ $`=`$ $`{\displaystyle \frac{1}{8}}\text{Tr}(ϵ[_sϵ,_tϵ])`$ (4.1)
$`=`$ $`{\displaystyle \frac{1}{4}}\text{Tr}(ϵ_sϵ_tϵ)`$
Setting $`H=H(1)`$ we have
$`ϵ={\displaystyle \frac{H}{\sqrt{H^2}}}={\displaystyle \frac{\gamma _5X}{\sqrt{X^{}X}}},X=a\gamma _5H=\text{}+r(\frac{1}{2}\mathrm{\Delta }1)`$ (4.2)
and a calculation using (4.1) gives
$`\text{Tr}(P[_sP,_tP])`$
$`={\displaystyle \frac{1}{4}}\text{Tr}\left({\displaystyle \frac{\gamma _5X}{\sqrt{X^{}X}}}_s(\gamma _5X){\displaystyle \frac{1}{\sqrt{X^{}X}}}_t(\gamma _5X){\displaystyle \frac{1}{\sqrt{X^{}X}}}\right)`$ (4.3)
$`+{\displaystyle \frac{1}{4}}\text{Tr}\left(_s\left({\displaystyle \frac{1}{\sqrt{X^{}X}}}\right)_t(\gamma _5X)\right)+{\displaystyle \frac{1}{4}}\text{Tr}\left({\displaystyle \frac{1}{X^{}X}}_s(\gamma _5X)_t\sqrt{X^{}X}\right)`$ (4.4)
$`{\displaystyle \frac{1}{4}}\text{Tr}\left(_s\left({\displaystyle \frac{1}{\sqrt{X^{}X}}}\right)\sqrt{X^{}X}\gamma _5X_t\left({\displaystyle \frac{1}{\sqrt{X^{}X}}}\right)\right)`$ (4.5)
We will see below that $`(\text{4.4})O(a)`$ and $`(\text{4.5})=Symm+O(a)`$ where $`Symm`$ is symmetric under interchange of $`_s`$ and $`_t`$. Since $`\text{Tr}(P[_sP,_tP])`$ is antisymmetric under this interchange it follows that
$`\text{Tr}(P[_sP,_tP])`$ $`=`$ $`{\displaystyle \frac{1}{8}}[\text{Tr}\left({\displaystyle \frac{\gamma _5X}{\sqrt{X^{}X}}}_s(\gamma _5X){\displaystyle \frac{1}{\sqrt{X^{}X}}}_t(\gamma _5X){\displaystyle \frac{1}{\sqrt{X^{}X}}}\right)`$ (4.6)
$`\text{Tr}\left({\displaystyle \frac{\gamma _5X}{\sqrt{X^{}X}}}_t(\gamma _5X){\displaystyle \frac{1}{\sqrt{X^{}X}}}_s(\gamma _5X){\displaystyle \frac{1}{\sqrt{X^{}X}}}\right)]+O(a)`$
To prove these statements, and evaluate the $`a0`$ limit of (4.6), we use the fact that the lattice transcript<sup>16</sup><sup>16</sup>16 Here $`A=A^{(s,t)}`$ and $`U=U^{(s,t)}`$ depend smoothly on the two parameters $`(s,t)`$.
$`U_\mu (x)=T\mathrm{exp}\left({\displaystyle _0^1}aA_\mu (x+(1\tau )ae_\mu )𝑑\tau \right)`$ (4.7)
can be expanded in powers of $`a`$ as
$`U_\mu (x)={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}a^n{\displaystyle _{0<\tau _1<\mathrm{}<\tau _n<1}}𝑑\tau _n\mathrm{}𝑑\tau _1A_\mu (x,\tau _n)\mathrm{}A_\mu (x,\tau _1)`$ (4.8)
where $`A_\mu (x,\tau )=A_\mu (x+(1\tau )ae_\mu )`$. Since $`A_\mu (x)`$ is a smooth function on the closed manifold $`T^4`$ there is a finite $`K`$ such that $`|A_\mu (x)|<K`$ for all $`x,\mu `$. Then the norm of the integral in the $`n`$’th term of (4.8) is bounded by $`\frac{1}{n!}K^n`$, so (4.8) is norm-convergent for all $`a`$. The inverse $`U_\mu (x)^1`$ also has an expansion in powers of $`a`$: using the fact that $`U_\mu (x)^1`$ is the parallel transport from $`x`$ to $`x+ae_\mu `$ specified by $`A`$ we see that $`U_\mu (x)^1`$ is given by the right-hand side of (4.8) with $`A_\mu (x,\tau )=A_\mu (x+\tau ae_\mu )`$.
Substituting these expansions in (2.2)–(2.3) gives an expansion $`_\mu ^\pm =_{n=0}^{\mathrm{}}a^n(_\mu ^\pm )_n`$. This in turn gives an expansion $`X=_{n=0}^{\mathrm{}}a^nX_n`$. It is not difficult to show that the $`(_\mu ^\pm )_n`$’s and the $`X_n`$’s have a finite bound $`K^{}`$ independent of $`a`$ and $`n`$, so the expansions are norm-convergent when $`a`$ is sufficiently small.
To expand $`1/\sqrt{X^{}X}`$ we note that
$`X^{}X=L+V`$ (4.9)
where
$`L`$ $`=`$ $`^2+r^2(\frac{1}{2}\mathrm{\Delta }m)^2`$ (4.10)
$`V`$ $`=`$ $`V^{(1)}+V^{(2)}`$ (4.11)
$`V^{(1)}`$ $`=`$ $`\frac{1}{2}r\gamma ^\mu [_\mu ,{\displaystyle \underset{\nu }{}}_\nu ^+_\nu ^{}],V^{(2)}=\frac{1}{4}[\gamma ^\mu ,\gamma ^\nu ][_\mu ,_\nu ]`$ (4.12)
Just as for $`X`$, the expansion (4.8) leads to expansions $`X^{}X=_{n=0}^{\mathrm{}}a^n(X^{}X)_n`$, $`L=_{n=0}^{\mathrm{}}a^nL_n`$, $`V=_{n=0}^{\mathrm{}}a^nV_n`$ where the $`(X^{}X)_n`$’s, $`L_n`$’s and $`V_n`$’s again have a finite bound independent of $`a`$ and $`n`$. Furthermore, explicit calculations show that
$`[_\mu ^+,_\nu ^+]\psi (x)`$ $`=`$ $`(a^2F_{\mu \nu }(x)+O(a^3))\psi (x+ae_\mu +ae_\nu )`$ (4.13)
$`[_\mu ^+,_\nu ^{}]\psi (x)`$ $`=`$ $`(a^2F_{\mu \nu }(x)+O(a^3))\psi (x+ae_\mu ae_\nu )`$ (4.14)
$`[_\mu ^{},_\nu ^+]\psi (x)`$ $`=`$ $`(a^2F_{\mu \nu }(x)+O(a^3))\psi (xae_\mu +ae_\nu )`$ (4.15)
$`[_\mu ^{},_\nu ^{}]\psi (x)`$ $`=`$ $`(a^2F_{\mu \nu }(x)+O(a^3))\psi (xae_\mu ae_\nu )`$ (4.16)
It follows that $`V_0=V_1=0`$, i.e. the expansion of $`V`$ starts with the $`a^2`$ term, hence $`VO(a^2)`$. The leading term $`a^2V_2`$ is explicitly given (mod $`O(a^3)`$) by substituting (4.13)–(4.16) into (4.11)–(4.12). We note from this that $`V_2=V_2^bT^b`$ where the $`T^b`$’s are the generators of the Lie algebra of $`SU(N)`$ and the $`V_2^b`$’s are trivial in colour space.
From (4.8) we also get expansions of $`_sU_\mu (x)`$ and $`_sU_\mu (x)^1=U_\mu (x)^1_sU_\mu (x)U_\mu (x)^1`$ in powers of $`a`$, leading to an expansion $`_sX=_{n=0}^{\mathrm{}}a^n(_sX)_n`$ and expansions of $`_s(X^{}X)`$, $`_sL`$, and $`_sV`$. Note that these begin with the order $`a`$ term, i.e. $`(_sX)_0=(_s(X^{}X))_0=(_sL)_0=(_sV)_0=0`$. For later use we also note the following: (1) The lowest order term in the expansion of $`_sU_\mu (x)`$ (or $`_sU_\mu (x)^1`$) is $`aA_\mu ^b(x)T^b`$ (or $`aA_\mu ^b(x)T^b`$). (2) Applying $`_s`$ to (4.13)–(4.16) results in $`F_{\mu \nu }(x)_sF_{\mu \nu }(x)`$, so $`(_sV)_0=(_sV)_1=0`$, $`_sVO(a^2)`$, and $`(_sV)_2=(_sV)_2^bT^b`$ where the $`(_sV)_2^b`$’s are trivial in colour space.
Note that the $`\gamma ^\mu `$’s in (4.9) are all contained in $`V`$. The hermitian positive operator $`L`$ is trivial in Dirac indices and the lowest order term $`L_0`$ in its expansion is diagonal with respect to the plane wave basis $`\{e^{iakx}\}`$; the diagonal elements are
$`L_0(ak)={\displaystyle \underset{\mu }{}}\mathrm{sin}^2(ak_\mu )+r^2\left(1+{\displaystyle \underset{\mu }{}}1\mathrm{cos}(ak_\mu )\right)^2`$ (4.17)
From this we see that there is a $`b>0`$ independent of $`a`$ such that $`L_0>2b`$. Then, by taking $`a`$ to be sufficiently small, we can achieve $`L>b`$ and $`V<\frac{1}{2}b`$, in which case $`1/\sqrt{X^{}X}`$ can be expanded as follows:
$`{\displaystyle \frac{1}{\sqrt{X^{}X}}}`$ $`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\sigma }{\pi }}{\displaystyle \frac{1}{X^{}X+\sigma ^2}}`$ (4.18)
$`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\sigma }{\pi }}\left({\displaystyle \frac{1}{1+(L+\sigma ^2)^1V}}\right)\left({\displaystyle \frac{1}{L+\sigma ^2}}\right)`$
$`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\sigma }{\pi }}{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}(1)^k((L+\sigma ^2)^1V)^k{\displaystyle \frac{1}{L+\sigma ^2}}.`$
For all $`p`$ we have
$`{\displaystyle \underset{k=0}{\overset{p}{}}}(1)^k((L+\sigma ^2)^1V)^k{\displaystyle \frac{1}{L+\sigma ^2}}<{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}({\displaystyle \frac{1}{b}}(\frac{1}{2}b))^k{\displaystyle \frac{1}{b+\sigma ^2}}={\displaystyle \frac{2}{b+\sigma ^2}}`$
Since the integral of this over $`(\mathrm{},\mathrm{})`$ is finite, the integral and sum in (4.18) can be interchanged, resulting in a norm-convergent expansion in powers of $`V`$:
$`(X^{}X)^{1/2}={\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\sigma }{\pi }}(1)^k((L+\sigma ^2)^1V)^k(L+\sigma ^2)^1`$ (4.19)
Since $`VO(a^2)`$ the $`k`$’th term in the sum is $`O(a^{2k})`$ and we conclude that
$`(X^{}X)^{1/2}={\displaystyle \underset{k=0}{\overset{p}{}}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\sigma }{\pi }}(1)^k((L+\sigma ^2)^1V)^k(L+\sigma ^2)^1+R_{p+1}`$ (4.20)
where $`\frac{1}{a^{2p}}R_{p+1}0`$ for $`a0`$. Similarly, we find
$`_s(X^{}X)^{1/2}={\displaystyle \underset{k=0}{\overset{p}{}}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\sigma }{\pi }}(1)^k_s\left(((L+\sigma ^2)^1V)^k(L+\sigma ^2)^1\right)+_sR_{p+1}`$ (4.21)
where $`\frac{1}{a^{2p}}_sR_{p+1}0`$ for $`a0`$.
The bound $`_\mu ^\pm 2`$ and triangle inequalities lead to an $`a`$independent upper bound $`L<b_1`$. Using this, the operator $`(L+\sigma ^2)^1`$ in (4.20) can be expanded as
$`{\displaystyle \frac{1}{L+\sigma ^2}}`$ $`=`$ $`\left({\displaystyle \frac{1}{b_1+\sigma ^2}}\right)\left({\displaystyle \frac{1}{1\frac{b_1L}{b_1+\sigma ^2}}}\right)`$ (4.22)
$`=`$ $`(b_1+\sigma ^2)^1{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}(b_1+\sigma ^2)^m(b_1L)^m`$
Substituting the expansion $`L=_{n=0}^{\mathrm{}}a^nL_n`$ in (4.22), and then substituting in (4.20) the resulting expansion of $`(L+\sigma ^2)^1`$, along with the expansion $`V=_{n=2}^{\mathrm{}}a^nV_n`$, we get an expansion $`(X^{}X)^{1/2}=_{n=0}^{\mathrm{}}a^n(X^{}X)_n^{1/2}`$. Similarly, after applying $`_s`$ to (4.22) and substituting the resulting expansion in (4.21), we get an expansion $`_s(X^{}X)^{1/2}=_{n=1}^{\mathrm{}}a^n(_s(X^{}X)^{1/2})_n`$ (note $`(_s(X^{}X)^{1/2})_0=0`$). These, together with the expansions of $`X`$ and $`X`$, lead to expansions $`𝒪=_{n=0}^{\mathrm{}}a^n𝒪_n`$ of the operators in (4.3)–(4.5). It can be shown that there is a finite $`\stackrel{~}{K}`$ independent of $`a`$ and $`n`$ such that $`𝒪_n<\stackrel{~}{K}`$ for these operators. This technical result will be presented elsewhere <sup>17</sup><sup>17</sup>17 It relies on the fact that $`A_\mu (x)`$ is periodic. The general (i.e. topologically non-trivial) case is more complicated.. This implies that the aforementioned operator expansions are all norm-convergent when $`a`$ is sufficiently small. An immediate consequence is the following: In the resulting expressions $`_{n=0}^{\mathrm{}}a^n\text{Tr}(𝒪_n)`$ for (4.3)–(4.5) the part $`_{n=5}^{\mathrm{}}a^n\text{Tr}(𝒪_n)`$ vanishes in the $`a0`$ limit. To see this, let $`\{\psi _j\}_{j=1,\mathrm{},𝒩}`$ be an arbitrary orthonormal basis for $`𝒞`$; then
$`\left|{\displaystyle \underset{n5}{}}a^n\text{Tr}(𝒪_n)\right|`$ $``$ $`{\displaystyle \underset{n5,j}{}}a^n|\psi _j,𝒪_n\psi _j|`$
$``$ $`{\displaystyle \underset{n5,j}{}}a^n\stackrel{~}{K}=a^4𝒩\stackrel{~}{K}{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}a^n=4L^4\stackrel{~}{K}{\displaystyle \frac{a}{1a}}0\text{for}a0`$
where we have used the fact that $`𝒩dim𝒞=4(L/a)^4`$. This shows that we only need to consider the $`a^n`$ terms with $`n4`$ in the expansions $`_{n=0}^{\mathrm{}}a^n\text{Tr}(𝒪_n)`$ of (4.3)–(4.5).
To proceed with the determination of (4.3)–(4.5) in the $`a0`$ limit, we note the following: (i) Due to the presence of an odd number of $`\gamma _5`$’s, terms in the expansions involving a product of less that 4 $`\gamma ^\mu `$’s vanish. (ii) $`L`$, $`V^{(1)}`$ and $`V^{(2)}`$ are are of order 0, 1 and 2 respectively in the $`\gamma ^\mu `$-matrices, c.f. (4.12). (iii) In the expansion $`𝒪=_{n=0}^{\mathrm{}}a^n𝒪_n`$ of any operator $`𝒪`$ constructed from the $`_\mu ^\pm `$’s, the term $`𝒪_0`$ is independent of the gauge field, so $`(_s𝒪)_0=0`$. Hence non-vanishing terms in such expansions are at least $`O(a)`$ for $`a0`$. (iv) As we have seen in (LABEL:q17), terms in the operator expansions which are of order $`5`$ in $`a`$ give vanishing contributions in the $`a0`$ limit.
At this point we can derive the postulated formula (4.6). Consider the first trace in (4.4): After substituting the expansion (4.19) for $`(X^{}X)^{1/2}`$ only the terms with at least two $`V`$’s are non-vanishing after taking the trace over spinor indices. Since $`VO(a^2)`$, such terms are all of order $`4`$ in $`a`$. Since $`_t(\gamma _5X)O(a)`$ it follows that the terms in the expansion of $`_s(X^{}X)^{1/2}_t(\gamma _5X)`$ which are non-vanishing after taking the trace are all of order $`5`$ in the $`a0`$ limit, so (4.4)$`O(a)`$ as claimed.
We now consider the trace (4.5). First, note from (4.20)–(4.22) that the lowest order term involving $`V`$ (or containing $`\gamma ^\mu `$’s) in the expansion of $`(X^{}X)^{1/2}`$ is
$`{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\sigma }{\pi }}(L_0+\sigma ^2)^1a^2V_2(L_0+\sigma ^2)^1.`$ (4.24)
Using (4.22) and the fact that $`[L_0,V]O(a)`$ we find that, modulo an $`O(a)`$ term, this is
$`a^2{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\sigma }{\pi }}(L_0+\sigma ^2)^2V_2=a^2L_0^{3/2}V_2{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\sigma }{\pi }}{\displaystyle \frac{1}{(1+\sigma ^2)^2}}={\displaystyle \frac{1}{2}}a^2L_0^{3/2}V_2`$ (4.25)
Similarly, the lowest order term containing $`\gamma ^\mu `$’s in the expansion $`_s(X^{}X)^{1/2}`$ is $`\frac{1}{2}a^2L_0^{3/2}(_sV)_2`$. To simplify the notation in the following, we write $`_s𝒪_n`$ for $`(_s𝒪)_n`$. Now, using (i)–(iv) above, we find
$`\text{Tr}\left(_s(X^{}X)^{1/2}(X^{}X)^{1/2}\gamma _5X_t(X^{}X)^{1/2}\right)`$
$`=a^4\text{Tr}\left(_s(X^{}X)_2^{1/2}(X^{}X)_0^{1/2}\gamma _5X_0_t(X^{}X)_2^{1/2}\right)+O(a)`$
$`={\displaystyle \frac{a^4}{4}}\text{Tr}\left(L_0^{3/2}_sV_2L_0^{1/2}\gamma _5X_0L_0^{3/2}_tV_2\right)+O(a)`$ (4.26)
where we have also used $`(X^{}X)_0=L_0`$. We now supplement (i)–(iv) above with the following observation (v): $`V_2`$ and $`(V)_2`$ commute with $`L_0`$ modulo an $`O(a)`$ term, and commute with $`\gamma _5X_0`$ modulo an $`O(a)`$ term and a term of order 1 in the $`\gamma ^\mu `$’s. It follows that, modulo an $`O(a)`$ term, (4.26) is symmetric under interchange of $`_s`$ and $`_t`$ as claimed. This proves the previously stated symmetry property of (4.5), thereby completing the derivation of (4.6).
Turning now to the traces in (4.6), similar arguments to the preceding give
$`\text{Tr}\left(\gamma _5X(X^{}X)^{1/2}_s(\gamma _5X)(X^{}X)^{1/2}_t(\gamma _5X)(X^{}X)^{1/2}\right)`$
$`=a^4\text{Tr}\left(\gamma _5X_0(X^{}X)_2^{1/2}_s(\gamma _5X)_1(X^{}X)_0^{1/2}_t(\gamma _5X)_1(X^{}X)_0^{1/2}\right)`$
$`+a^4\text{Tr}\left(\gamma _5X_0(X^{}X)_0^{1/2}_s(\gamma _5X)_1(X^{}X)_2^{1/2}_t(\gamma _5X)_1(X^{}X)_0^{1/2}\right)`$
$`+a^4\text{Tr}\left(\gamma _5X_0(X^{}X)_0^{1/2}_s(\gamma _5X)_1(X^{}X)_0^{1/2}_t(\gamma _5X)_1(X^{}X)_2^{1/2}\right)+O(a)`$
$`={\displaystyle \frac{1}{2}}a^4(t^{cab}+t^{acb}+t^{abc})\text{Tr}\left(L_0^{5/2}\gamma _5X_0_s(\gamma _5X)_1^a_t(\gamma _5X)_1^bV_2^c\right)+O(a)`$ (4.27)
where we have set $`t^{abc}=\text{tr}(T^aT^bT^c)`$. Modulo an $`O(a)`$ term, (4.27) is antisymmetric under interchange of $`_s(\gamma _5X)_1^a`$ and $`_t(\gamma _5X)_1^b`$. To see this, note that $`_s(X^{}X)_1^a=\gamma _5X_0_s(\gamma _5X)_1^a+_s(\gamma _5X)_1^a\gamma _5X_0`$. Since $`(X^{}X)_1=L_1`$ does not contain $`\gamma ^\mu `$’s, it follows that $`\gamma _5X_0_s(\gamma _5X)_1^a`$ can be replaced by $`_s(\gamma _5X)_1^a\gamma _5X_0`$ in (4.27). The claimed antisymmetry then follows from the cyclicity of the trace after using (v) above. Taking this into account in (4.6), and noting $`\frac{1}{2}d^{abc}=t^{cab}+t^{cba}=t^{acb}+t^{bca}=t^{abc}+t^{bac}`$, we get
$`\text{Tr}(P[_sP,_tP])={\displaystyle \frac{3}{32}}a^4d^{abc}\text{Tr}\left(L_0^{5/2}\gamma _5X_0_s(\gamma _5X)_1^a_t(\gamma _5X)_1^bV_2^c\right)+O(a)`$ (4.28)
We calculate
$`_s(\gamma _5X)_1^a_t(\gamma _5X)_1^b`$ $`=`$ $`\left(\gamma ^\mu (_s_\mu )_1^a+{\displaystyle \frac{1}{2}}r(_s\mathrm{\Delta })_1^a\right)\left(\gamma ^\nu (_t_\nu )_1^b+{\displaystyle \frac{1}{2}}r(_t\mathrm{\Delta })_1^b\right)`$ (4.29)
$`=`$ $`\stackrel{~}{V}^{ab}+\text{a term not involving }\gamma ^\mu \text{’s}`$
where
$`\stackrel{~}{V}^{ab}`$ $`=`$ $`\stackrel{~}{V}^{(1)ab}+\stackrel{~}{V}^{(2)ab}`$ (4.30)
$`\stackrel{~}{V}^{(1)ab}`$ $`=`$ $`{\displaystyle \frac{1}{2}}r\gamma ^\mu \left((_s_\mu )_1^a(_t\mathrm{\Delta })_1^b(_s\mathrm{\Delta })_1^a(_t_\mu )_1^b\right)`$ (4.31)
$`\stackrel{~}{V}^{(2)ab}`$ $`=`$ $`{\displaystyle \frac{1}{2}}[\gamma ^\mu ,\gamma ^\nu ](_s_\mu )_1^a(_t_\nu )_1^b`$ (4.32)
It follows from (4.28) and (4.29)–(4.32) that
$`\text{Tr}(P[_sP,_tP])={\displaystyle \frac{3}{32}}a^4d^{abc}\text{Tr}(L_0^{5/2}\gamma _5X_0\stackrel{~}{V}^{ab}V_2^c)+O(a)`$
$`={\displaystyle \frac{3}{32}}a^4d^{abc}\text{Tr}\left(L_0^{5/2}\gamma _5\left(\gamma ^\mu _\mu (\stackrel{~}{V}^{(1)ab}V_2^{(2)c}+\stackrel{~}{V}^{(2)ab}V_2^{(1)c})+r{\displaystyle \frac{1}{2}}\mathrm{\Delta }\stackrel{~}{V}^{(2)ab}V_2^{(2)c}\right)\right)`$
$`+O(a)`$ (4.33)
$`V^{(1)ab}`$ and $`V^{(2)ab}`$ can be determined as follows. Recalling $`_sU_\mu (x)=a_sA_\mu (x)+O(a^2)`$ we get
$`(_s_\mu ^\pm )_1\psi (x)=(_sA_\mu (x)+O(a))\psi (x\pm ae_\mu )`$ (4.34)
and calculations give
$`(_s_\mu )_1^a(_t_\nu )_1^b\psi (x)`$ $`=`$ $`{\displaystyle \frac{1}{4}}(_sA_\mu ^a(x)_tA_\nu ^b(x)+O(a))`$ (4.35)
$`\times \left[{\displaystyle \genfrac{}{}{0pt}{}{\psi (x+ae_\mu +ae_\nu )+\psi (x+ae_\mu ae_\nu )}{+\psi (xae_\mu +ae_\nu )+\psi (xae_\mu ae_\nu )}}\right]`$
$`((_s_\mu )_1^a(_t\mathrm{\Delta })_1^b`$ $``$ $`(_s\mathrm{\Delta })_1^a(_t_\mu )_1^b)\psi (x)`$ (4.36)
$`=`$ $`{\displaystyle \underset{\nu }{}}{\displaystyle \frac{1}{2}}\left(_sA_\mu ^a(x)_tA_\nu ^b(x)_sA_\nu ^a(x)_tA_\mu ^b(x)+O(a)\right)`$
$`\times \left[{\displaystyle \genfrac{}{}{0pt}{}{\psi (x+ae_\mu +ae_\nu )+\psi (x+ae_\mu ae_\nu )}{+\psi (xae_\mu +ae_\nu )+\psi (xae_\mu ae_\nu )}}\right]`$
which determine $`\stackrel{~}{V}^{(1)ab}`$ and $`\stackrel{~}{V}^{(2)ab}`$ in (4.31)–(4.32). On the other hand, from (4.11)–(4.12) and (4.13)–(4.16) we find that $`V_2^{(1)c}`$ and $`V_2^{(2)c}`$ coincide up to $`O(a)`$ with $`\stackrel{~}{V}^{(1)ab}`$ and $`\stackrel{~}{V}^{(2)ab}`$, respectively, after replacing $`_sA_\mu ^a(x)_tA_\nu ^b(x)`$ with $`\frac{1}{2}F_{\mu \nu }^c(x)`$.
Having determined the operators in (4.33) we can now evaluate the trace by first tracing over spinor indices and then evaluating the remaining trace in the plane wave orthonormal basis $`\{\varphi _k(x)\}`$ for periodic scalar fields, given by
$`\varphi _k(x)={\displaystyle \frac{1}{𝒩}}e^{ikx}`$ (4.37)
$`𝒩`$ $`=(2N)^4,k_\mu {\displaystyle \frac{\pi }{aN}}\{N,N+1,\mathrm{},N1\}`$ (4.38)
The result is
$`\text{Tr}(P[_sP,_tP])={\displaystyle \frac{}{32\pi ^2}}d^{abc}a^4{\displaystyle \underset{x}{}}ϵ_{\mu \nu \rho \sigma }_sA_\mu ^a(x)_tA_\nu ^b(x)F_{\rho \sigma }^c(x)`$ (4.39)
where
$``$ $`=`$ $`{\displaystyle \underset{k}{}}a^4\mathrm{\Delta }^4k_r(ak)`$ (4.41)
$`\mathrm{\Delta }^4k{\displaystyle \frac{(2\pi )^4}{a^4𝒩}}=\text{the “volume per }k\text{” in (}\text{4.38}\text{)},`$
$`_r(k)={\displaystyle \frac{3r}{8\pi ^2}}{\displaystyle \frac{_{\nu =1}^4\mathrm{cos}k_\nu \left[1+_\mu (1\mathrm{cos}k_\mu )_\mu \frac{\mathrm{sin}^2k_\mu }{\mathrm{cos}k_\mu }\right]}{\left[_\mu \mathrm{sin}^2k_\mu +r^2\left(1+_\mu (1\mathrm{cos}k_\mu )\right)^2\right]^{5/2}}}`$
(The denominator in this expression comes from $`L_0^{5/2}`$ in (4.33); we have used (4.17).) Changing variables from $`k`$ to $`ak`$ in (4.41) leads to
$`\underset{a0}{lim}={\displaystyle _\pi ^\pi }d^4k_r(k).`$ (4.42)
This integral was encountered in in connection with the axial anomaly for fermions with Overlap Dirac operator, and was found to equal 1 (independent of $`r`$). We can now take the $`a0`$ limit in (4.39) and get the desired result:
$`\underset{a0}{lim}\text{Tr}(P[_sP,_tP])={\displaystyle \frac{1}{32\pi ^2}}d^{abc}{\displaystyle _{T^4}}d^4xϵ_{\mu \nu \rho \sigma }_sA_\mu ^a(x)_tA_\nu ^b(x)F_{\rho \sigma }^c(x)`$ (4.43)
In deriving this formula we have followed the approach used in for calculating the axial anomaly for the overlap Dirac operator. Presumably the other approaches of ref.’s and could also be used to derive this formula.
Remark. The preceding also shows that
$`\underset{a0}{lim}{\displaystyle \frac{1}{8}}\text{tr}\{[ϵ[_sϵ,_tϵ]](x,x)\}`$ $`=`$ $`{\displaystyle \frac{1}{32\pi ^2}}d^{abc}ϵ_{\mu \nu \rho \sigma }_sA_\mu ^a(x)_tA_\nu ^b(x)F_{\rho \sigma }^c(x)={\displaystyle \frac{1}{24}}\text{tr}((x,s,t))^3`$
Here $`[\mathrm{}](x,y)`$ denotes the kernel of the operator $`[\mathrm{}]`$, $`(x,s,t)`$ is the curvature of the gauge field $`𝒜(x,s,t)=A_\mu ^{(s,t)}(x)dx^\mu `$ in 6 dimensions, and the last coefficient arises as $`2\pi i(\frac{1}{(2\pi i)^33!})=\frac{1}{24\pi ^2}`$. This result for the “topological field” $`q(x,s,t)=\frac{1}{8}\text{tr}\{[ϵ[_sϵ,_tϵ]](x,x)\}`$ in 6 dimensions was used in <sup>18</sup><sup>18</sup>18 In a more general $`q(x,s,t)`$ was considered; however, to obtain the mentioned result it suffices to consider the present $`q(x,s,t)`$. to show the existence of a $`w_+(U)`$ (or equivalently, the existence of a local gauge-invariant current $`j_\mu (x)`$ satisfying certain conditions) to all orders in an expansion in the lattice spacing $`a`$ such that the corresponding overlap $`v_+,w_+`$ is gauge-invariant when $`d_R^{abc}=0`$. The calculations in this section leading to (LABEL:q38) above could therefore be useful as a starting point for finding explicit expressions for the terms in $`j_\mu (x)`$.
## 5 Concluding remarks
The main result of this paper is that the overlap formulation of chiral gauge theory on the lattice reproduces the global obstructions to gauge-invariance discussed in the continuum by Alvarez-Gaumé and Ginsparg . We showed that the obstruction on the lattice reduces to the continuum obstruction in the classical continuum limit. This, together with the fact that the lattice obstruction is also an integer (winding number), implies that the lattice obstruction coincides exactly with the continuum one for small non-zero lattice spacing (i.e. close to the classical continuum limit). Thus the overlap formulation is seen to exactly capture topological structure of the continuum theory in the nonabelian case, just as it does in the abelian case considered previously in . We mention again that, while we have taken the spacetime to be the 4-dimensional, our arguments and results generalise straightforwardly to Euclidean spacetime $`T^{2n}`$ for arbitrary $`n`$.
It might be instructive to compare this with the situation for chiral Wilson fermions on the lattice (where gauge-invariance is explicitly broken due to the Wilson term, and only restored in the $`a0`$ limit). In this case the consistent local anomaly $`\frac{d}{d\theta }\mathrm{log}det(D_{Wilson}^{\varphi _\theta U})_+`$ has been shown to converge to the continuum anomaly in the classical continuum limit , so the integral
$`W_{Wilson}(\mathrm{\Phi })={\displaystyle \frac{1}{2\pi i}}{\displaystyle _0^1}𝑑\theta {\displaystyle \frac{d}{d\theta }}\mathrm{log}det(D_{Wilson}^{\varphi _\theta U})_+`$
converges to the continuum global obstruction $`\mathrm{deg}(\mathrm{\Phi })`$ in this limit. However, for non-zero lattice spacing the integral $`W_{Wilson}(\mathrm{\Phi })`$ is non-integer in general –it does not have a winding number interpretation since $`|det(D_{Wilson}^U)_+|`$ is not gauge-invariant. This is in contrast to the overlap case where
$`W(\mathrm{\Phi })={\displaystyle \frac{1}{2\pi i}}{\displaystyle _0^1}𝑑\theta {\displaystyle \frac{d}{d\theta }}\mathrm{log}v_+,w_+(\varphi _\theta U)`$
is an integer (winding number) since $`|v_+,w_+(U)|`$ is gauge-invariant and $`v_+,w_+(\varphi _\theta U)`$ differs from $`v,w_+(U)`$ only by a phase factor. Thus, in contrast to the overlap case, the non-integer $`W_{Wilson}(\mathrm{\Phi })`$ is in general not equal to the integer-valued continuum obstruction $`\mathrm{deg}(\mathrm{\Phi })`$ at any non-zero lattice spacing; equality only occurs in the limit $`a0`$.
As a consequence of our main result, we found that the continuum anomaly cancellation condition $`d_R^{abc}=0`$ is a necessary condition for anomaly cancellation in the overlap on the lattice (at least when $`Map(T^5,SU(N))`$ contains maps with non-trivial degree). While this is no surprise, our derivation is robust compared to other approaches: Firstly, it is independent of the choice of phase in the overlap (in contrast to the consistent local anomaly which does depend on the phase choice), and secondly, no approximations, or assumptions on the gauge field, have been used.
In the continuum argument of ref. , the winding number obstruction $`W(\mathrm{\Phi })`$ is shown to equal the index of a Dirac operator $`𝒟`$ in $`6=2n+2`$ dimensions. The index theorem is then used to get $`\text{index}𝒟=\mathrm{deg}\mathrm{\Phi }`$. In this paper we have followed a different route: a lattice version of the determinant line bundle approach of ref. . In fact, it is also possible to give a lattice version of the original argument of Alvarez-Gaumé and Ginsparg in the overlap setting, using a certain lattice Dirac operator in $`2n+2`$ dimensions (with the extra 2 dimensions being continuous) <sup>19</sup><sup>19</sup>19In this case, the index density of the lattice $`𝒟`$ turns out to be the aforementioned topological field $`q(x,s,t)`$ in $`6=2n+2`$ dimensions which appeared in ref. .
The obstructions of Alvarez-Gaumé and Ginsparg are but one type of obstruction to gauge-invariance of the chiral determinant. In general, the obstructions are manifestations of non-trivial topological structure of the determinant line bundle over the orbit space of gauge fields. This topic has been studied in detail in the continuum; see, e.g., . The results of and the present paper suggest that in general the continuum topological structure of the determinant line bundle can be reproduced on the lattice in the overlap formulation (at least when the spacetime manifold is a $`2n`$-dimensional torus). The determinant line bundle comes equipped with a canonical $`U(1)`$ connection, and the difference $`Im\mathrm{\Gamma }(A^{(1)})Im\mathrm{\Gamma }(A^{(0)})`$ for the effective action $`\mathrm{\Gamma }(A)=\mathrm{log}det(D_+^A)`$ can be expressed in terms of the parallel transport of this connection along a path joining $`A^{(0)}`$ to $`A^{(1)}`$ . This can in turn be expressed in terms of a spectral flow ($`\eta `$-invariant) of a Dirac operator and a Chern-Simons term, both in $`5=2n+1`$ dimensions. Lattice versions of these relations in the overlap setting have already been found .
Finally, it is interesting to note that the quantity $`\text{Tr}(PdPdP)=\frac{1}{8}\text{Tr}(ϵdϵdϵ)`$, which in the present setting appears as the curvature of the overlap determinant line bundle (or the ’Berry curvature’ in the terminology of ref. ), also arises as the curvature of a determinant line bundle in canonical quantisation of the continuum theory.<sup>20</sup><sup>20</sup>20 I thank Prof. J. Mickelsson for pointing this out to me. See and the ref.’s therein. In that setting one considers a certain infinite-dimensional Grassmannian manifold consisting of splittings $`V_+V_{}`$ of the Hilbert space of 1-particle states; each splitting corresponds to an $`ϵ=P_{V_+}P_V_{}`$. There is a canonical determinant line bundle on this manifold, and its curvature turns out to be a renormalised version of $`\frac{1}{8}\text{Tr}(ϵdϵdϵ)`$.<sup>21</sup><sup>21</sup>21 In fact, a lattice regularisation of chiral gauge theory (different to the overlap) in which this quantity also appears has been presented in ref. . Gauge-invariance appears to be problematic in this approach though: the notion of gauge symmetry needs to be modified on the lattice in a way that involves non-local operators. It could be interesting to explore the apparent analogy between this continuum formulation and the lattice overlap formulation. Recently, an obstruction to canonical quantisation of the continuum theory on odd-dimensional spacetimes was described in . Instead of the ’Berry curvature’ 2-cocycle, the obstruction there is given in terms of a 3-cocycle known as the Dixmier–Douadly class. It could be interesting to see if there is something analogous to this in the lattice overlap formulation.
Acknowledgements. This work was presented at the mini-workshop “New developments in lattice gauge theory”, CSSM, Adelaide, 4-5 April’00, and I thank Prof.’s Ting-Wai Chiu, Kazuo Fujikawa, Urs Heller, Jouko Mickelsson and Tony Williams for interesting comments and discussions at that time. I also thank Prof. Herbert Neuberger for correspondence on the Overlap and, together with Prof.’s Martin Lüscher and Jun Nishimura, for feedback on earlier versions of this paper. The author is supported by an ARC postdoctoral fellowship.
## Appendix
## Appendix A Derivation of (3.7): $`\frac{v_+,(dw_+)_{}}{v_+,w_+}=\text{Tr}(dDD^1P_+)`$.
Let $`U(t)`$ be a smooth curve in $`𝒰`$. Using (2.13) we calculate
$`{\displaystyle \frac{d}{dt}}\mathrm{log}v_+,w_+`$ $`=`$ $`{\displaystyle \frac{d}{dt}}\mathrm{log}v_+,\widehat{D}w_+`$ (A.1)
$`=`$ $`{\displaystyle \frac{1}{v_+,\widehat{D}w_+}}\left(v_+,\frac{d}{dt}\widehat{D}w_++v_+,\widehat{D}\frac{d}{dt}w_+\right)`$
At $`t=0`$ (and with our assumption that $`D`$ has no zero-modes) we have
$`v_+,\frac{d}{dt}\widehat{D}w_+`$ $`=`$ $`v_+,\frac{d}{dt}\widehat{D}\widehat{D}^1\widehat{D}w_+=v_+,\widehat{D}w_+v_+,\frac{d}{dt}\widehat{D}\widehat{D}^1v_+`$
$`=`$ $`v_+,\widehat{D}w_+{\displaystyle \frac{d}{dt}}|_{t=0}det\left(D(t)D(0)^1|_{𝒞_+}\right)`$
$`=`$ $`v_+,\widehat{D}w_+\text{Tr}(\frac{d}{dt}DD^1P_+)`$
so the first term on the right-hand side of (A.1) is $`\text{Tr}(\frac{d}{dt}DD^1P_+)`$. Comparing (A.1) with (3.6) we see that (3.7) holds iff
$`v_+,\widehat{D}\frac{d}{dt}w_+=v_+,\widehat{D}w_+w_+,\frac{d}{dt}w_+`$ (A.2)
Choose an orthonormal basis $`w_1(t),\mathrm{},w_d(t)`$ for $`𝒞_+^{(1,U(t))}`$ such that $`w_+=w_1\mathrm{}w_d`$. Then
$`v_+,\widehat{D}\frac{d}{dt}w_+={\displaystyle \underset{j=1}{\overset{d}{}}}v_+,\widehat{D}(w_1\mathrm{}\frac{d}{dt}w_j\mathrm{}w_d)`$ (A.3)
Substitute $`\frac{d}{dt}w_j=_{k=1}^dw_k,\frac{d}{dt}w_jw_k+(\frac{d}{dt}w_j)_{}`$ in (A.3), where $`(\frac{d}{dt}w_j)_{}𝒞_{}^{(1,U(t))}`$. Since $`D`$ maps $`\widehat{𝒞}_\pm =𝒞_{}^{(1,U)}`$ to $`𝒞_{}`$, the terms involving $`(\frac{d}{dt}w_j)_{}`$ give vanishing contribution, and we get
$`v_+,\widehat{D}\frac{d}{dt}w_+`$ $`=`$ $`{\displaystyle \underset{j=1}{\overset{d}{}}}v_+,\widehat{D}w_+w_j,\frac{d}{dt}w_j.`$
This equals the right-hand side of (A.2) as required, since
$`w_+,\frac{d}{dt}w_+`$ $`=`$ $`{\displaystyle \underset{j=1}{\overset{d}{}}}w_+,w_1\mathrm{}\frac{d}{dt}w_j\mathrm{}w_d`$
$`=`$ $`{\displaystyle \underset{j=1}{\overset{d}{}}}w_+,w_+w_j,\frac{d}{dt}w_j={\displaystyle \underset{j=1}{\overset{d}{}}}w_j,\frac{d}{dt}w_j.`$
## Appendix B Derivation of (3.14)
It suffices to restrict to a surface in $`𝒰`$ with coordinates $`(s,t)`$ and show
$`_sw_+,_tw_+_tw_+,_sw_+=\text{Tr}(P_sP_tP)\text{Tr}(P_tP_sP)`$ (B.1)
Let $`w_1(s,t),\mathrm{},w_d(s,t)`$ be an orthonormal basis for $`𝒞_+^{(1,U(s,t))}`$ such that $`w_+=w_1\mathrm{}w_d`$. It is convenient to use bra-ket notation:
$`P={\displaystyle \underset{k=1}{\overset{d}{}}}|w_kw_k|,_sP={\displaystyle \underset{k=1}{\overset{d}{}}}|_sw_kw_k|+|w_k_sw_k|`$
Then
$`\text{Tr}(P_sP_tP)`$ $`=`$ $`{\displaystyle \underset{j=1}{\overset{d}{}}}w_j|_sP_tP|w_j`$ (B.2)
$`=`$ $`{\displaystyle \underset{j,k,l=1}{\overset{d}{}}}w_j|\left(|_sw_kw_k|+|w_k_sw_k|\right)\left(|_tw_lw_l|+|w_l_tw_l|\right)|w_j`$
$`=`$ $`{\displaystyle \underset{j,k}{}}w_j|_sw_kw_k|_tw_j+{\displaystyle \underset{j,k}{}}w_j|_sw_k_tw_k|w_j`$
$`+{\displaystyle \underset{j}{}}_sw_j|_tw_j+{\displaystyle \underset{k,l}{}}_sw_j|w_l_tw_l|w_j`$
The first and fourth sums are clearly symmetric under $`_s_t`$. The second sum is likewise symmetric under $`_s_t`$ as is easily seen using $`\delta w_j|w_k+w_j|\delta w_k=\delta w_j|w_k=0`$. It follows that
$`\text{Tr}(P_sP_tP)\text{Tr}(P_tP_sP)={\displaystyle \underset{j}{}}_sw_j|_tw_j_tw_j|_sw_j`$
and this is equal to the left-hand side of (B.1) as required.
|
warning/0004/hep-ph0004079.html
|
ar5iv
|
text
|
# Proposal to observe the strong Van der Waals force in the low energy neutron-Pb scattering11footnote 1NUP-A-2000-7
## 1 Introduction
By the neutron transmission experiments of $`Pb^{208}`$ in the energy range of $`0.05keV<T_{lab}<40keV`$, Schmiedmayer et.al. successfully determined the electric polarizability of the neutron $`\alpha _n`$. In the determination of $`\alpha _n`$, they fitted the n-Pb<sup>208</sup> cross section as
$$\sigma (\nu )=11.508(5)+0.69(9)\sqrt{\nu }448(3)\nu +9500(400)\nu ^2$$
(1)
where $`\sigma `$ and $`\nu `$ are measured in the units of barn and fm<sup>-2</sup> respectively. The $`\sqrt{\nu }`$ term arises from the long range potential of the electric polarization of the neutron. From the coefficient of $`\sqrt{\nu }`$, they obtained the value $`\alpha _n=(1.20\pm 0.15\pm 0.20)\times 10^3fm^3`$. On the other hand, when all the forces are short range, $`\sigma (\nu )`$ must be a polynomial of $`\nu `$. However the smallness of $`\alpha _n`$ obtained from this precise experiment reveals an embarrassing situation. Namely the n-Pb data in the MeV. region have the behavior characteristic of the long range force, and if we attribute it to the same electric polarization of the neutron, the value of $`\alpha _n`$ becomes around two orders of magnitude greater compared to what is obtained in the very low energy region. Therefore we need another long range force which does not alter the coefficient of $`\sqrt{\nu }`$. An example of the required long range force is the strong Van der Waals force, because it gives rise to the singular term $`\nu ^{3/2}`$ or $`\nu ^2\mathrm{log}\nu `$ in Eq.(1) as we shall see in the next section.
Since in the hadron physics, interactions are believed for a long time to be short range arising from the exchanges of a pion, a set of pions or heavier particles, the appearance of the strong Van der Waals force may sound strange. However in 1960’s our view of hadrons changed from elementary particles to composite particles. In most of the composite model of hadron, the fundamental force is strong or super-strong Coulombic force, and it is responsible for the formation of the ”neutral” bound states, and which are identified with the hadrons. Here ”neutral” means total ”charge” zero, in which ”charge” corresponds to the fundamental Coulombic force. As in the case of the ordinary (electric) atoms, the quantum fluctuation gives rise to the Van der Waals interaction between the ”neutral” composite particles, namely between hadrons. When the radius of the hadron and the ”fine structure constant” $`{}_{}{}^{}e_{}^{2}`$ are known, we can estimate the strength $`C`$ of the Van der Waals potential , at least in the order of magnitude. It turns out that when the fundamental Coulombic force is super-strong such as in the case of the dyon model of Schwinger , in which $`{}_{}{}^{}e_{}^{2}=137.04/4`$, $`C`$ becomes large enough to compete with the potential of the one-pion exchange. Therefore it is desirable to search for the strong Van der Waals interaction whenever sufficiently precise data are available. The neutron-Pb scattering has an advantage in the search, because for large $`r`$ the strength $`C`$ of the strong Van der Waals potential of the nuclear force is magnified by a factor $`A`$ in the neutron-nucleus potential, where $`A`$ is the mass number of the nucleus and we shall set $`A=208`$ in this paper.
From the analysis of the S-wave phase shift data of the low energy proton-proton scattering, we have determined the parameters of the long range tail of nuclear potential $`v(r)=C/r^\alpha +\mathrm{}`$ in which $`\alpha =6.08`$ and $`C=0.196`$ in the unit of the Compton wave length of the neutral pion. Once the tail of the nuclear potential is known, it is not difficult to compute the asymptotic behavior of the n-Pb potential and to determine the singular behavior of the n-Pb scattering amplitude. Since the singular term has the form $`(t)^\gamma `$, where $`\gamma =(\alpha 3)/2`$, we can obseve the anomalous behavior of the n-Pb amplitude in two places. One is the anomalous behavior $`(1z)^\gamma `$ of the amplitude at $`z=1`$ for fixed $`\nu `$, and the other is the anomolous term $`\nu ^\gamma `$ in the partial wave amplitudes $`a_{\mathrm{}}(\nu )`$. The aim of the present paper is to predict the shapes and the magnitudes of these singular behaviors of the n-Pb amplitude by using the parameters of the long range force determined from the p-p scattering.
In section 2, the relations between the singular behavior of the amplitude at $`t=0`$ and the asymptotic behavior of the potential at large $`r`$ are described. In section 3, the anomaly of the angular distributions of the amplitudes are shown, in which the incident energies are fixed at $`T_{lab}=`$ 0.25, 0.5, 0.75 and 1.0 MeV. respectively. In section 4, the once subtracted P-wave amplitude $`a_1(\nu )/\nu `$ is computed, and it is shown that it has a characteristic cusp $`\nu ^{\gamma 1}`$ at $`\nu =0`$. Section 5 is used for comments and discussions. In Appendix, the long range components of the nuclear potential obtained from the S-wave amplitude of the p-p scattering are summarized along with the brief explanation of the analysis of the p-p data.
## 2 Asymptotic form of the potential and the <br>singular behavior of the amplitude
When we want to confirm the existence of the long range force unambiguously, it is desirable to use the difference of the analytical structure of the scattering amplitude $`A(s,t)`$. This is because the amplitude $`A(s,t)`$ is regular in the neighborhood of $`t=0`$ if all the forces are short range, on the other hand when the long range force is acting an extra singularity appears at $`t=0`$. It is fortunate that $`t=0`$ is the end point of the physical region $`4\nu t0`$, and we can in principle settle the ploblem whether the strong interaction involves the long range force, when the sufficiently accurate data are given.
Since the potential $`v(r)`$ and the spectral function $`A_t(s,t)`$ of the scattering amplitude are connected by
$$v(r)=\frac{1}{\pi m^2}\frac{1}{r}_0^{\mathrm{}}𝑑tA_t(s,t)e^{r\sqrt{t}},$$
(2)
the parameters of the threshold behavior of $`A_t(s,t)`$ and those of the asymptotic behavior of the potential, which are defined by
$$A_t(s,t)=\pi C^{}t^\gamma +\mathrm{}andv(r)C\frac{1}{r^\alpha }+\mathrm{}$$
(3)
respectively, relate to each other by
$$\alpha =2\gamma +3andC=\frac{2C^{}}{m^2}\mathrm{\Gamma }(2\gamma +2).$$
(4)
From the threshold behavior of $`A_t(s,t)`$, the singular behavior of the amplitude $`A(s,t)`$ at $`t=0`$ is determined :
$$A(s,t)=\frac{\pi }{\mathrm{sin}\pi \gamma }C^{}(t)^\gamma +\text{ (polynomial of }t\text{ ) .}$$
(5)
In particular when $`\gamma `$ is an integer $`n`$, we must take the limit $`\gamma n`$ of Eq.(5), and the singular term becomes $`(1)^{n+1}C^{}(t)^n\mathrm{log}(t)`$. For the case of the strong Van der Waals force, the power $`\alpha `$ of the asymptotic behavior of the potentials are $`\alpha =6`$ and $`\alpha =7`$ for the London type and the Casimir-Polder type respectively, and which correspond to $`\gamma =1.5`$ and $`\gamma =2`$ respectively.
Merits to use the large nucleus as the target of the neutron scattering are twofold. When we construct the neutron-nucleus potential by making the convolution of the nuclear potential and the form factor of the nucleus, the strength of the long range tail of the potential becomes $`A`$ times large compared to that of the nuclear potential between nucleons, where $`A`$ is the mass number of the nucleus and in our case $`A=208`$. On the other hand, since the one-pion exchange term of the nuclear potential involves the factor $`(\stackrel{}{\sigma }_n\stackrel{}{\sigma }_j)`$ or $`(\stackrel{}{\sigma }_n\widehat{r})(\stackrel{}{\sigma }_j\widehat{r})`$, sum of the contributions from all the constituent nucleons cancels out. Therefore we may expect to get very clear view of the singularity of the long range force at $`t=0`$, because the spectrum of the two-pion exchange starts at $`t=4`$ and gives rise to an almost constant back ground in the small neighborhood of $`t=0`$. Throughout of this paper, we shall use the neutral pion mass and its Compton wave length as the units of the energy and the length respectively. However because of the large radius $`r_1`$ of the nucleus, eg. for $`Pb^{208}`$ $`r_1`$ is $`4.42`$, we cannot neglect the back ground ploynomial function of $`t`$, when the domain of $`t`$ becomes wide and lies out side of $`\sqrt{t}<1/r_1`$. In order to know the necessary degree of the back ground polynomials, in section 3 and 4 we shall also compute the n-Pb amplitude arising from short range potential, the potential of the $`\sigma `$-meson exchange in particular.
Since $`t=2\nu (1z)`$, the singular term $`A^{sing}(s,t)`$ becomes
$$A^{sing}(s,t)=\frac{\pi }{\mathrm{sin}\pi \gamma }C^{}(2\nu )^\gamma (1z)^\gamma ,$$
(6)
and for fixed $`\nu `$, we must observe the singular behavior $`(1z)^\gamma `$ in the neighborhood of $`z=1`$. Another interesting property of the long range force is the anomaly of the threshold behavior of the partial wave amplitude $`a_{\mathrm{}}(\nu )`$. For the short range force, it is well-known the amplitude $`a_{\mathrm{}}(\nu )`$ is proportional to $`\nu ^{\mathrm{}}`$ at the threshold. This is because in the partial wave projection of the polynomial function of $`t`$, $`z^{\mathrm{}}`$ appears first in the term of $`t^{\mathrm{}}`$, thus the factor $`\nu ^\gamma `$ always appears in $`a_{\mathrm{}}(\nu )`$. On the other hand, the result of the partial wave projection of the singular term is
$$a_{\mathrm{}}^{sing}(\nu )=\frac{1}{2}_1^1𝑑zP_{\mathrm{}}(z)A^{sing}(s,t)=\frac{\pi }{\mathrm{sin}\pi \gamma }C^{}(2\nu )^\gamma I_{\mathrm{}}(\gamma ),$$
(7)
where the patial wave projection of $`(1z)^\gamma `$ is
$$I_{\mathrm{}}(\gamma )=2^\gamma \frac{(\gamma )(1\gamma )\mathrm{}(\mathrm{}1\gamma )}{(1+\gamma )(2+\gamma )\mathrm{}(\mathrm{}+1+\gamma )}$$
(8)
for $`\mathrm{}>0`$ and for $`\mathrm{}=0`$, $`I_0(\gamma )=2^\gamma /(1+\gamma )`$. Therefore all the partial wave amplitudes have a term whose threshold behavior is proportional to $`\nu ^\gamma `$, if the long range force is acting. For example for the case of the Van der Waals force of the London type ($`\gamma =1.5`$), all the partial waves other than S and P waves have the threshold behavior $`\nu ^{1.5}`$. Moreover although the threshold behavior of the P-wave is normal and we can consider $`a_1(\nu )/\nu `$, it involves a term proportional to $`\sqrt{\nu }`$ arising from the singular term $`A^{sing}(s,t)`$. Since Eqs.(7) and (8) indicate that the term of $`\sqrt{\nu }`$ has the negative sign, when the asymptotic force is attractive ($`C^{}>0`$), the cusp of $`a_1(\nu )/\nu `$ at $`\nu =0`$ must points upward. The shape of the P-wave in the low energy region is one of the ideal place to observe the effect of the long range force, because for the higher partial waves, we cannot obtain precise data in the low energy region. On the other hand, although the cusp of $`\sqrt{\nu }`$ with opposite sign appears in the once subtracted S-wave amplitude $`(a_0(\nu )a_0(0))/\nu `$, extremely precise value of the scattering length $`a_0(0)`$ is necessary to observe such a cusp. Therefore in the search of the long range force, we shall concentrate on the singular behavior of the angular distribution in the forward region $`z=1`$ and on the shape of $`a_1(\nu )/\nu `$ of the P-wave in the low energy region.
## 3 Anomaly of the angular distribution of the <br>neutron-Pb amplitude
Since the parameters of the long range component of the nuclear force are already known by the analysis of the S-wave amplitude of the proton-proton scattering, we can determine the potential of the neutron-nucleus scattering, at least in the region of the long range tail. It must be emphasized that the singular behavior of the amplitude at $`t=0`$ is determined solely by the asymptotic behavior of the long range potential, and therefore the change of the potential for finite $`r`$, for example inside of the nucleus, does not have any effect on the singularity at $`t=0`$. All the information on the long range component of the nuclear force is contained in the spectral function
$$A_t^{extra}(s,t)=\pi C^{}t^\gamma e^{\beta t}$$
(9)
with
$$\gamma =1.54,C^{}=0.175and\beta =0.0626$$
(10)
in the unit of the neutral pion mass. The value of $`\gamma `$ is close to that of the Van der Waals force of the London type( $`\gamma =1.5`$) and the sign of $`C^{}`$ indicates that the force is attractive. In this section, we shall compute the amplitude of the neutron-Pb<sup>208</sup> scattering using the parameters $`\gamma `$ and $`C^{}`$, and predict the shape of the singular behavior at $`t=0`$. Such a prediction will be helpful in designing the neutron-Pb<sup>208</sup> experiments of high precision and the data obtaind in the experiments will in turn improve the values of the parameters $`\gamma `$ and $`C^{}`$.
In computing the convolution of the nuclear potential and the form factor $`\rho (r)`$ of the nucleus, for reason of simplicity, we shall choose $`\rho (r)`$ as the box form of radius $`r_1`$, namely $`\rho (r)`$ is zero and non-zero constant in $`r>r_1`$ and in $`r<r_1`$ respectively. This is permissible because we are interested in the singularity and it is not affected by the change of the potential at finite $`r`$. Among three parameters appeared in the spectral function $`A_t^{extra}(s,t)`$ of Eq.(9), $`\beta `$ does not relate to the singularity, but it controls the depth of the n-Pb potential. Contrary to $`\gamma `$ and $`C^{}`$, $`\beta `$ will be regarded as a free parameter and whose value will be fixed by fitting to the scattering length of the S-wave amplitude of the n-Pb scattering. There is another free parameter $`r_1`$, size of the nucleus. We shall consider two amplitudes, in which $`r_1=4.4`$ and $`r_1=3.5`$ respectively. The former is the standard size of the nucleus of $`A=208`$, however $`r_1=4.4`$ gives the slope of $`\sqrt{\nu }\mathrm{cot}\delta _0(\nu )`$ around 10 % higher. If we move $`r_1`$ to $`3.5`$ it gives right value $`d(\sqrt{\nu }\mathrm{cot}\delta _0(\nu ))/d\nu =1.9`$, which was obtained in the transmission experiment of the n-Pb<sup>208</sup> scattering. In the following, we shall designate these two amplitudes starting from the strong Van der Waals potential \[vdw44\] and \[vdw35\] respectively. In order to compare the cases of the long range forces with that of the short range force, the third amplitude of n-Pb will be considered starting from the nuclear potential of the sigma meson exchange with $`m_\sigma =4`$. In the calculation we shall choose $`r_1=4.4`$ and the coupling constant $`g_\sigma ^2`$ is fixed by fitting to the scattering length of the S-wave. This amplitude will be designated \[sigma\]. In figure 1, the potentials $`V(r)`$ of the n-Pb scatterings are shown.
In this paper, we shall consider the n-Pb scattering in $`T_{lab}1`$ MeV.. Since $`\nu `$ and $`T_{lab}`$ is proportional and $`T_{lab}=1`$ MeV. corresponds to $`\nu =0.102123`$, the domain of $`t`$ necessary to make the patial wave projection is $`0t<0.41`$, whereas the nearest singularity on the $`t`$-plane arising from the exchange of the sigma meson is $`t=16`$. Therefore we expect that the partial wave amplitudes $`a_{\mathrm{}}(\nu )`$ of \[sigma\] decreases very rapidly with $`\mathrm{}`$. On the other hand, for the case of the Van der Waals interaction in which all the threshold behavior is proportional to $`\nu ^\gamma `$ for $`\mathrm{}2`$, $`a_{\mathrm{}}(\nu )`$ does not decrease so rapidly even if $`\nu `$ is small. However we can see that the difference of the partial wave amplitudes from those of the Born term, namely $`(a_{\mathrm{}}(\nu )a_{\mathrm{}}^{(Born)}(\nu ))`$, decreases very rapidly with $`\mathrm{}`$ for small $`\nu `$. Therefore the amplitudes of \[vdw44\] and \[vdw35\] are
$$F(\nu ,t)=F^{Born}(\nu ,t)+\underset{\mathrm{}=0}{\overset{L}{}}(2\mathrm{}+1)(a_{\mathrm{}}(\nu )a_{\mathrm{}}^{(Born)}(\nu ))P_{\mathrm{}}(z),$$
(11)
and we shall set $`L=4`$ because $`|a_5(\nu )a_5^{(Born)}(\nu )|<10^4`$. Since the n-Pb potential $`V(R)`$ is the convolution of $`\rho (r)`$ and $`v(r)`$, the Born term of the n-Pb scattering amplitude $`F^{Born}(\nu ,t)`$ is the product of the Fourier transformations of these functions.
The first one is
$$A\widehat{\rho }(t)d^3re^{i\stackrel{}{q}\stackrel{}{r}}\rho (r)=A\frac{3(\mathrm{sin}xx\mathrm{cos}x)}{x^3}|_{x=qr_1},$$
(12)
where $`q`$ is the momentum transfer and relates to $`t`$ by $`t=q^2`$. Since $`\widehat{\rho }(t)`$ is an even function of $`q`$, it is a regular function of $`t`$, and it is normalized as $`\widehat{\rho }(0)=1`$. The second one is the Born term of the neutron-nucleon scattering:
$$A^{extra}(t)\frac{1}{\pi }𝑑t^{}\frac{A_t^{extra}(s,t^{})}{t^{}t}=\mathrm{\Gamma }(\gamma +1)C^{}(t)^\gamma e^{\beta t}\mathrm{\Gamma }(\gamma ,\beta t,\mathrm{}),$$
(13)
in which the incomplete gamma function is used and whose definition is
$$\mathrm{\Gamma }(g,a,b)=_a^b𝑑xx^{g1}e^x$$
(14)
$`\mathrm{\Gamma }(g,p,\mathrm{})`$ can be divided into two terms: $`(\mathrm{\Gamma }(g)\mathrm{\Gamma }(g,0,p))`$. If we use the separation, $`A^{extra}(t)`$ is written as the sum of the singular and the regular terms of $`t`$ :
$$A^{extra}(t)=\frac{\pi C^{}}{\mathrm{sin}\pi \gamma }(t)^\gamma e^{\beta t}\mathrm{\Gamma }(\gamma +1)\frac{C^{}}{\beta ^\gamma }(\beta t)^\gamma e^{\beta t}\mathrm{\Gamma }(\gamma ,0,\beta t).$$
(15)
From Legendre’s formula
$$p^ge^p\mathrm{\Gamma }(g,0,p)=\underset{n=0}{\overset{\mathrm{}}{}}\frac{p^n}{g(g+1)(g+2)\mathrm{}(g+n)},$$
(16)
we see that the second term of the r.h.s. of Eq.(15) is a regular function of $`\beta t`$. By multiplying $`A\widehat{\rho }(t)`$, we finally obtain the Born term of the scattering amplitude $`F^{Born}(\nu ,t)`$
$$F^{Born}(\nu ,t)=F_s(\nu ,t)(t)^\gamma +F_r(\nu ,t),$$
(17)
where $`F_s(\nu ,t)`$ and $`F_r(\nu ,t)`$ are regular functions of $`t`$, and are defind by
$$F_s(\nu ,t)=\frac{\pi }{m\mathrm{sin}\pi \gamma }AC^{}e^{\beta t}\widehat{\rho }(t),$$
(18)
$$F_r(\nu ,t)=\frac{AC^{}}{m\beta ^\gamma }\mathrm{\Gamma }(\gamma +1)\{(\beta t)^\gamma e^{\beta t}\mathrm{\Gamma }(\gamma ,0,\beta t)\}\widehat{\rho }(t)$$
(19)
respectively.
Contrary to the case of the short range force, the amplitudes with the long range force cannot be fitted by a few terms of the polynomials of $`z`$. In fact curves \[vdw35\] and \[sigma\] in the figures show such property, where \[vdw35\] and \[sigma\] are the scattering amplitudes $`F(\nu ,t)`$ minus their S, P and D waves for the Van der Waals force and for the $`\sigma `$-meson exchange force respectively. Smallness of \[sigma\] means that the amplitude arising from the $`\sigma `$-meson exchange is fitted well by the quadratic function of $`z`$. In order to confirm that the singularity in \[vdw35\] is actually $`F_s(\nu ,t)(t)^\gamma `$, we must show the smallness of $`\{F(\nu ,t)F_s(\nu ,t)(t)^\gamma \}`$ minus the S, P and D waves. In the figure, the dotted curve is such a regular part of \[vdw35\]. Figure 2 and 3 are the graphs of $`T_{lab}=`$ 0.25 and 0.5MeV. respectively, in which the full line is the \[vdw35\], the dashed line is the \[sigma\] multiplied by 10 and the dotted line is the regular part of \[vdw35\] multiplied by factor 100.
Figure 4 and 5 are the graphs of $`T_{lab}=`$ 0.75 and 1.0MeV. respectively, in which the contents are the same as fig.2 and fig.3 except the multiplicative factor, namely in these figures only the dotted curves are multiplied by factor 10. As the incident energy increases, even the amplitude arising from the short range potential becomes more and more difficult to be fitted by the quadratic function of $`z`$. Therefore we must try the fit by the cubic function for $`T_{lab}=`$ 0.75 and 1.0MeV.. Figure 6 and 7 are the graphs of $`T_{lab}=`$ 0.75 and 1.0MeV. respectively, in which the contents are the same as fig.4 and fig.5 except the subtracted functions are the cubic function, namely the S, P, D and F waves, rather than the quadratic function of $`z`$. And as in fig.4 and fig.5, only the regular part of \[vdw35\] (dotted curve) is multiplied by factor 10. When we draw curves, values of parameters of \[vdw35\] are
$$\gamma =1.54,\beta =0.2517,C^{}=0.175andr_1=3.5$$
(20)
, whereas the nuclear potential of the $`\sigma `$-meson exchange is chosen as $`v(r)=0.59e^{4r}/r`$ and $`r_1=4.4`$ for \[sigma\].
When the precise n-Pb experimants are carried out, we shall obtain similar curves as \[vdw35\]. The content of our prediction is that the singular behaviors of the experimental curves will be removed by subtracting $`F_s(\nu ,t)(t)^\gamma `$ with the parameters given in Eq.(20), which are obtained from the data of the low energy p-p scattering. If the sufficiently accurate n-Pb data are available, we may expect to obtain the improved values of the long range parameters $`\gamma `$ and $`C^{}`$.
## 4 The cusp of $`a_1(\nu )/\nu `$ at $`\nu =0`$
Since the singular term $`(t)^\gamma `$ is equal to $`2^\gamma \nu ^\gamma (1z)^\gamma `$, the singular behaviors appear in two places. One is the singular behavior $`(1z)^\gamma `$ of the angular distribution at $`z=1`$ for fixed $`\nu `$, and the other is the singular term of $`\nu ^\gamma `$ in the partial wave amplitude. In the previous section we studied the former, and the latter will be studied in this section. When $`\gamma =1.54`$, $`\nu ^\gamma `$ is the leading threshold behavior of the D and the higher partial waves rather than the standard threshold behavior $`\nu ^{\mathrm{}}`$. For the P-wave $`a_1(\nu )`$, the threshold behavior is proportional to $`\nu `$, and therefore we can use the once subtracted form $`a_1(\nu )/\nu `$. In $`a_1(\nu )/\nu `$ the singular term becomes $`\nu ^{\gamma 1}`$, and which is a cusp at $`\nu =0`$ because its slope is infinity. When the long range tail of the nuclear potential is attractive, from Eqs.(7) and (8) the coefficient of $`\nu ^{\gamma 1}`$ is negative and therefore the cusp must point upward. On the other hand, the curve of \[sigma\] must be regular at $`\nu =0`$. In figure 8, such curves of \[vdw35\], \[vdw44\] and \[sigma\] are shown. In order to make the display compact, since we do not have interest in the back ground regular functions, we subtract linear functions, which are chosen in such ways that curves come to $`\nu `$-axis as close as possible in $`0.02<\nu <0.03`$.
If we speak precisely, the graphs in fig.8 are not the real parts of $`a_1(\nu )/\nu `$ but the once subtracted Kantor amplitude defined by
$$\frac{K_1(\nu )}{\nu }=\mathrm{Re}\frac{a_1(\nu )}{\nu }\frac{P}{\pi }_0^{\mathrm{}}𝑑\nu ^{}\frac{\mathrm{Im}a_1(\nu ^{})}{\nu ^{}(\nu ^{}\nu )},$$
(21)
where $`P`$ stands for Cauchy’s pricipal value integration. It is important that the unitarity cut in $`0\nu <\mathrm{}`$ of $`a_1(\nu )/\nu `$ is removed in $`K_1(\nu )/\nu `$. Therefore $`K_1(\nu )/\nu `$ has a wide domain of analyticity, and which is necessary to observe the delicate difference of behaviors arising from the spectrum on the left hand cut $`\mathrm{}<\nu m_1^2/4`$ of the short range force, where $`m_1`$ is the lightest mass exchanged, and from the spectrum of the long range force in $`\mathrm{}<\nu 0`$. It is fortunate that, for the P and higher partial waves, the integrations of Eq.(21) are small and in their estimation, very precise phase shift data are not required. This is because the threshold behavior of $`\mathrm{Im}a_1(\nu ^{})/\nu ^{}`$ is $`\nu ^{3/2}`$ even for the P-wave. On the other hand for the S-wave, the threshold behavior of $`\mathrm{Im}a_0(\nu ^{})/\nu ^{}`$ is $`1/\sqrt{\nu }`$ and the estimation of the principal value integration requires very precise data of the phase shift $`\delta _0(\nu )`$. Another difficulty to use the S-wave amplitude is that we need to know extremely precise value of the scaatering length $`a_0(0)`$, since in cmputing the once subtracted Kantor amplitude, we must know $`(a_0(\nu )a_0(0))/\nu `$. Therefore the S-wave is not the suitable place to observe the singularity $`\nu ^{\gamma 1}`$, unless we can overcome these difficulties. If we consider that we do not have sufficient data of the D and higher partial waves in the low energy region, the P wave is the most suitable place to observe the singularity $`\nu ^\gamma `$ of the partial waves.
## 5 Remarks and Comments
Since the Van der Waals force is universal , its existence in the nucleon-nucleon scattering implies the Van der Waals interaction also in other processes such as in the pion-pion and in the pion-nucleon scatterings. The possibility to find such an extra force depends on the precision of the data and the possibility to prepare the wide domain of analyticity. The low energy data of the p-p scattering is prominent in their accuracy, whereas the pion-pion process is prominent in its possibility to prepare the wide analytic domain, because the two-pion exchange spectrum can be constructed from the pi-pi amplitudes themselvs. In the Appendix, we see a cusp at $`\nu =0`$ in the once subtracted Kantor amplitude of the S-wave of the p-p scattering, in which the one-pion exchange spectrum is removed. Similarly the characteristic cusp at $`\nu =0`$ is observed in $`a_1(\nu )/\nu `$ of the pion-pion scattering, in which the unitarity cut and the cut of the two-pion exchange are removed.
The third place easy to observe the strong Van der Waals force is the low energy neutron-nucleus amplitude, because the strength $`C`$ of the strong Van der Waals potential of the nuclear force is magnified by factor $`A`$, the mass number. Although the effect of the Van der Waals force decreases as $`\nu ^\gamma `$ for small $`\nu `$, in the neuton-nucleus scattering the large value of the strength $`AC`$ allows us to observe it even at the smaller energy. It is advantageous to use the lower energy data in the determinations of the paremeters $`\gamma `$ and $`C^{}`$ of the threshold behavior appeared in the extra spectrum $`A_t^{extra}(s,t)=\pi C^{}t^\gamma e^{\beta t}`$, because they are determined without being disturbed by the parameter $`\beta `$, which specifies the deviation from the threshold behavior.
Therefore our proposal in this paper is to measure the angular distribution of the cross section of the neutron-nucleus scattering such as n-Pb<sup>208</sup> precisely for fixed energy around $`T_{lab}=`$1MeV., and to determine the parameters of the asymptotic tail of the long range component of the nuclear potential. Although the values of such parameters are known from the low energy proton-proton data, the independent determination will serve to confirm the actual existence of the strong Van der Waals interaction in the nuclear force.
Appendix
When we extract the information of the long range force from the partial wave amplitude $`h_{\mathrm{}}(\nu )`$, first of all we must remove the unitarity cut, and then the known near-by singularities. In this way we can prepare a function with the wide domain of analyticity in the neighborhood of $`\nu =0`$, and which is helpful to observe the extra singularity at $`\nu =0`$ arising from the long range interaction, if it exists. Since the phase shift data of the S-wave of the low energy proton-proton scattering is most accurate in the hadron physics, we shall consider the once subtracted S-wave amplitude $`(h_0(\nu )h_0(0))/\nu `$. If we make the Kantor amplitude by
$$K_0^{once}(\nu )=\mathrm{Re}(h_0(\nu )h_0(0))/\nu \frac{1}{\pi }_0^{\mathrm{}}\frac{\mathrm{Im}h_0(\nu ^{})}{\nu ^{}(\nu ^{}\nu )}𝑑\nu ^{}$$
(22)
, the unitarity cut is removed. Since the one-pion exchange (OPE) contribution is
$$h_0^{1\pi }(\nu )=\frac{1}{4}\frac{g^2}{4\pi }\frac{1}{4\nu }\mathrm{log}(1+4\nu )$$
(23)
, which has a left hand cut starting from $`\nu =1/4`$, the OPE cut in the Kantor amplitude is removed in
$$\stackrel{~}{K}_0^{once}(\nu )=K_0^{once}(\nu )\frac{(h_0^{1\pi }(\nu )h_0^{1\pi }(0))}{\nu }.$$
(24)
When we have precise phase shift data, we can evaluate $`\stackrel{~}{K}_0^{once}(\nu )`$, and it is shown in figure 9 and 10. Although the spectrum of the two-pion exchange starts at $`\nu =1`$, in the threshold region the spectrum is very small, because it arises from the partial wave projection of the continuous spectrum $`A_t^{2\pi }(s,t)`$. If we replace $`A_t^{2\pi }(s,t)`$ by the delta function of the $`\sigma `$-meson exchange, the left hand cut of $`\stackrel{~}{K}_0^{once}(\nu )`$ starts at $`\nu =4`$. Therefore if the nuclear force is short range due to the exchanges of a set of pions, $`\stackrel{~}{K}_0^{once}(\nu )`$ must be almost constant with very small slope and extremely small curvature in the neighborhood of $`\nu =0`$.
However figures 9 and 10 indicate this is not the case, but there is a cusp at $`\nu =0`$ (closed circles). This cusp is fitted by an extra spectrum of the long range interaction of the form
$$A_t^{extra}(4m^2,t)=\pi C^{}t^\gamma e^{\beta t}.$$
(25)
More explicitely, it is fitted by
$$\frac{h_0^{extra}(\nu )h_0^{extra}(0)}{\nu }=2C^{}_0^{\mathrm{}}𝑑tt^\gamma e^{\beta t}\{\frac{1}{2\nu }Q_0(1+\frac{t}{2\nu })\frac{1}{t}\}\frac{1}{\nu }$$
(26)
, in which the extra factor 2 in front of the integration appears because the contributions from the spectrum $`A_u`$ as well as from $`A_t`$ must be included. The integration of Eq.(26) can be written in terms of the generalized hypergeometric function $`{}_{p}{}^{}F_{q}^{}(x)`$ and of the confluent hypergeometric function $`F(a,b,x)`$.
$$\frac{h_0^{extra}(\nu )h_0^{extra}(0)}{\nu }=2(\nu ^{\gamma 1}\xi _s(\nu )+\xi _r(\nu ))$$
(27)
, where $`\xi _s(\nu )`$ and $`\xi _r(\nu )`$ are regular functions of $`\nu `$ and are difined by
$$\xi _s(\nu )=C^{}\frac{4^\gamma }{\gamma +1}\frac{\pi }{\mathrm{sin}\pi \gamma }F(1+\gamma ,2+\gamma ,4\beta \nu )$$
(28)
and
$`\xi _r(\nu )`$ $`=`$ $`C^{}{\displaystyle \frac{\mathrm{\Gamma }(\gamma )}{\beta ^\gamma \nu }}(_{\{1,1\}}F_{\{2,1\gamma \}}(4\beta \nu )1)`$ (29)
$`=`$ $`C^{}{\displaystyle \frac{\mathrm{\Gamma }(\gamma )}{\beta ^\gamma \nu }}({\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{(4\beta \nu )^n}{(1\gamma )\mathrm{}(n\gamma )}}{\displaystyle \frac{1}{n+1}})`$
respectively. Results of the chi-square fits are
$$\gamma =1.543,\beta =0.06264,C^{}=0.1762and\chi =0.441$$
(30)
in the unit of the neutral pion mass. Among three parameters, $`\gamma `$ and $`C^{}`$ are the threshold parameters of the long range force, whereas $`\beta `$ is necessary to make the integration convergent, and $`\sqrt{\beta }`$ must be the order of magnitude of the nucleon radius. In constructing the neutron-Pb potential, although we use the values of $`\gamma `$ and $`C^{}`$ given in Eq.(30), $`\beta `$ is left as a free parameter and later fixed by fitting to the scattering length of the n-Pb amplitude. Up to this point we have not considerd the ordinary Coulomb interaction and the vacuum polarization. Therefore the procedure explained above is applicable only to the neutron-neutron data. When we include the electromagnetic interaction, some modifications are necessary in the constructions of the singularity free functions $`K_0(\nu )`$ from the phase shift functions. An explicit construction of the Kantor amplitude from the phase shift $`\delta _0^E(\nu )`$ are explained in other paper.
Finally it will be helpful to write the normalization of the amplitudes explicitly:
$$A(s,t)=\underset{\mathrm{}=0}{\overset{\mathrm{}}{}}(2\mathrm{}+1)h_{\mathrm{}}(\nu )P_{\mathrm{}}(z)withh_{\mathrm{}}(\nu )=\frac{\sqrt{m^2+\nu }}{\sqrt{\nu }}e^{i\delta _{\mathrm{}}(\nu )}\mathrm{sin}\delta _{\mathrm{}}(\nu )$$
(31)
and
$$f(\nu ,t)=\underset{\mathrm{}=0}{\overset{\mathrm{}}{}}(2\mathrm{}+1)a_{\mathrm{}}(\nu )P_{\mathrm{}}(z)witha_{\mathrm{}}(\nu )=\frac{1}{\sqrt{\nu }}e^{i\delta _{\mathrm{}}(\nu )}\mathrm{sin}\delta _{\mathrm{}}(\nu ).$$
(32)
Therefore $`f(\nu ,t)=A(s,t)/m`$ in the approximation $`\nu m^2`$.
|
warning/0004/cond-mat0004418.html
|
ar5iv
|
text
|
# Stripes ordering in self-stratification experiments of binary and ternary granular mixtures
## Abstract
The self-stratification of binary and ternary granular mixtures has been experimentally investigated. Ternary mixtures lead to a particular ordering of the strates which was not accounted for in former explanations. Bouncing grains are found to have an important effect on strate formation. A complementary mechanism for self-stratification of binary and ternary granular mixtures is proposed.
45.70.Mg, 81.05.Rm, 64.75.+g
A recent experiment which has received much attention from the scientific community is the following. When a binary mixture of grains differing in size is poured between two vertical slabs, there is a global tendency for large and small grains to segregate in different regions of the pile . Additionally, a self-stratification of the mixture in alternating layers of small and large grains is observed if large grains have a larger angle of repose than the small ones. It has been also shown that the phase segregation takes also place when grains of same size but with different shapes are mixed together. Strates might also appear in those conditions .
For describing the self-stratification phenomenon, both continuous and discrete models have been proposed up to now for binary mixtures. These models are able to reproduce the alternating layers of the different granular species. The more elaborated models consider phenomenological continuous equations which account mainly for the angles of repose of the various species, the percolation of grains in the rolling phase, and the kink formation on the pile surface. A challenge for physicists is the generalization of both experiments and models to the case of a continuous distribution of grain sizes.
The first motivation of the present work was the experimental investigation of ternary mixtures which did not receive much attention in earlier works. As described below, the structures of the phase segregations in ternary mixtures raise new questions about the self-stratification mechanisms.
A vertical Hele-Shaw (HS) cell was specially built for our purpose. The distance $`e`$ separating the vertical planes of the HS could be continuously adjusted. Various granular species have been used: sand, wheat semolina, poppy seeds, Fe filings. We controlled the granulometry of each species by a preliminary sifting. Important properties of each type of grains are given in Table I. Granular mixtures contained an equal volume of each species. They were poured in the HS cell with a funnel. A CCD camera with a lens 12/120mm F5.6 was placed perpendicular to the HS cell for examining the physical processes involved at the grain scale. Top views have also been taken. They allowed us to observe the dynamics and composition of the upper surface during the flow. Precise video imaging is necessary for studying the evolution of granular piles. Our results are reproducible.
First, let us consider the case of ternary mixtures being composed of small ($`S`$), medium ($`M`$) and large ($`L`$) grains. Fig.1 presents three pictures of ternary piles in the HS cell. Two different mixtures were used. In each case, size segregation is clearly observed. Indeed, small grains have a global tendency to locate in the center of the pile while large grains are located in the tail. The medium-sized grains are dispersed in between the center and the tail. In addition, the self-stratification phenomenon is present. Three different patterns are obtained. In the first pile, a $`\mathrm{}SMLSMLSML\mathrm{}`$ ordering is displayed as found in . The second pile is characterised by a $`\mathrm{}SMLMSMLM\mathrm{}`$ ordering. The third pile is quite similar to the first, as far as wide stripes are concerned. However, the ordering seems to be more complex, especially in the tail: a thin stripe of small grains is visible in the center of wide stripes made of large grains. These different orderings require special attention and the presence of the thin stripe cannot be explained with conventional models for self-stratification. Another important observation is that for some mixtures, a difference in inclinaison angle appears between the different stripes (Fig.1$`b`$). This difference of stripe angles, the particular ordering and the occurrence of a thin additional stripe indicates that quite different mechanisms are present during the heap formation.
In the case of binary mixtures, Koeppe et al. have reported some anomalous stripe pattern that they called “pairing”. Indeed, stripes of the same grains seemed to be formed by pairs in particular experimental conditions. We made careful experiments with binary mixtures and we observed again the above-mentioned thin stripes (see Fig.2). We suggest that thin stripes can be at the origin of the previously reported “pairing”. Similarily to ternary mixtures, the successive stripes may have different angles.
Self-stratification for binary mixtures of grains differing in size has been explained by the combination of two main processes. They are illustrated in Fig.3. The first process is the size segregation mechanism. There is a global tendency of large grains to roll farther downhill than the small ones due to their larger mass/inertia. This effect is strongly increased by the percolation of small grains through the gap between large grains in the flow. Large grains will be on the top of small grains, in the upper part of the moving layer. Small grains form a relatively smooth surface, on which large grains roll easily. Thus the moving layer is characterized by a strong velocity gradient such that the upper large grains are the fastest. As a result, large grains also locate in the front of the avalanche while small grains are confined in the lower part of the tail of the moving layer. Large grains will travel farther than small grains. The second process coming into play is the so-called kink mechanism. When the head of an avalanche reaches the end of the pile, grains will stop moving due to the horizontal slope of the basis. If the flux is not too important down the slope, grains in the tail of the avalanche will come to rest on the wall just formed by preceding grains. This wall is the kink and it moves up the slope as more grains arrive on it. This process is similar to the formation and backing of traffic jams. A pair of layers is formed through the kink mechanism with the small grains in the lower layer and the large grains in the top layer, starting at the basis of the pile. Since small grains were located in the tail of the avalanche, the wide stripe of small grains does not extend to the basis of the pile. The resulting surface involves an efficient capture for small grains during the next avalanche.
This explanation can be extended to ternary mixtures. It accounts for patterns in Fig.1$`ab`$. The $`\mathrm{}SMLMSMLM\mathrm{}`$ ordering reflects the dynamics and composition of the lower layer of the avalanche: large grains in the head, followed by medium sized grains, themselves followed by small grains. So when an avalanche flow down on top of the surface formed by the preceeding kink, after the head of large grains has passed, medium grains will pass and some will be captured before small grains pass. A very close examination of the heap displayed in Fig.1$`a`$ reveals that there are indeed some medium sized grains caught below the stripes of small grains in the lower part of the heap. On the other hand, in the upper part of the heap, the $`\mathrm{}SMLSMLSML\mathrm{}`$ ordering is visible because there was not enough time for medium grains to move in front of small grains in the lower part of the moving layer. Thus, small grains were the first to be captured during the avalanche in the upper part of the pile. These dynamical processes depend strongly on the incoming flux of grains and the relative velocities of different granular species. However, the above mechanism for self-stratification cannot explain the thin layers of small grains observed in Fig.1$`c`$ and in Fig.2.
The complementary mechanism we propose is illustrated in Fig.3. Considering again a binary mixture, the first steps are identical to previous explanation. The kink, however, does not catch all moving grains. Indeed, in any granular flow, various transport mechanisms take place. From bottom to top in the moving layer, one encounters: combined movement of grains, motion of individual grains with friction and collisions, and above the rolling-slipping phase, a layer with boucing grains. Typically, the thickness of the rolling-slipping phase is about five to ten large grain diameters in our mixtures. The bouncing grains layer is characterised by a higher velocity and a much smaller density. Indeed grains does not touch each other permanently like in the rolling-slipping phase. We found that bouncing grains can jump as far as 2 cm and then rebounce or be captured. We stress that most bouncing grains found at the bottom of the pile have bounce all their way down. Both kinds of grains can bounce. The efficiency of this mechanism depends on resilience of the grains. We observed that a non-negligible amount of those bouncing grains can flow down the slope over the kink (see Fig.4). We stress that bouncing grains are those involved in the formation of the additional thin stripe observed in the center of the wide stripe made of large grains.
Fig.5 shows also the evolution of a binary pile. The first picture was taken as the kink was moving up. The thickness of the kink is about the same that the thickness of the rolling layer. As a consequence, the kink moves fast upward. The second picture shows the pile just after the kink has passed. The third picture was taken just before the next avalanche comes down, and the last picture shows the pile as the avalanche is flowing.
It is clearly visible that the thin stripe of small grains appears mainly during the time interval separating pictures $`b`$ and $`d`$, that is between the passage of the kink and the next avalanche. In addition, a close examination shows that this thin stripe is composed of both kinds of grains, in opposition to wider stripes where the granular species are quasi-pure. In the case of pile 1$`c`$, close look reveals that small, medium and large sized grains are involved in the thin stripes.
The reason these strates are thiner is that there are much less grains involved – only those that are not immediatelly stopped by the kink. When a small boucing grain is captured by the pile, it will fall through the gaps left by the large grains forming the static surface. This is why the stripes are formed of both kinds of grains. This also explains the fact that the thin stripe is observed to form below the static surface, about one large grain deep. Thin stripes have the same inclinaison angle as the surface left by the kink. The difference in angle for thin stripes and wide stripes is observed in Figure 2$`b`$ ($`\mathrm{\Delta }\theta 3^{}`$). It indicates that the kink builds at a greater angle than the angle of the avalanche (avalanches erode mainly the upper part of the pile). This can be related to another observation we made when pouring granular materials through the funnel. We noticed that the jamming limit diameter was lower in the case of mixture made of large and small grains than in the case of large grains only. Thus the presence of small grains seems to reduce the arching properties of large grains. We do not know if this is a general feature.
Several causes can inhibe or hide the phenomena of thin stripe formation. In the case of Fig.1 $`a`$ and $`c`$, the only difference is the funnel used for pouring the grains. As a consequence, not only the flux of grains was lower in the case of pile $`a`$ but also, and this is related, the initial velocity of grains seemed to be lower in $`a`$. The main effect is a lowering of the number of bouncing grains. The kink may become perfectly efficient, i.e. it may completely stop all moving grains it encounters. Then the thin layer will not form. Indeed we observed no boucing grains flowing over the kink during heap formation of $`a`$. Next, if the avalanches are important, they can erase by erosion the very thin layer that forms the top of the pile.
All the mixtures we used are characterised by a large aspect ratio ($`>2`$, up to 10) and by a larger density for small grains. Koeppe et al. used a mixture of sand and sugar that corresponds to these characteristics when they obtained their “pairing”. The difference in size causes percolation to be efficient and might favorise an important velocity gradient in the moving layer. These make self-stratification particularly pronounced. High velocity in the top of the rolling-slipping layer favorise the transport of bouncing grains. The large density of small grains allows them to bounce as far as large grains do. Indeed, in the case of grains all of the same density, small grains are much lighter than large grains. They would have much less kinetic energy and so they would travel less far than large grains.
The number of grains involved in the thin stripe, depending on the position along the slope, could be computed from the time interval between the passage of the kink and next avalanche, and the number of bouncing grains at that point, given that one knows the probability of capture by unit lenght for those grains by the pile. We observed that the thin stripe of small grains do not always presents the same profile, in particular, for some mixtures with narrower density difference between small and large grains, thin stripes do not extend to the basis of the pile. Taking the problem backward, this could provide an efficient way to study the transport properties of bouncing mechanism.
In summary, we have investigated both ternary and binary granular mixtures in vertical HS cells and have found particular patterns. We observed the formation of an additional thin stripe which we relate to the previously reported “pairing” . This cannot be fully explained by former self-stratification models. We have proposed a complementary mechanism which is compatible with observations of binary/ternary patterns. The mechanism is based on the presence of bouncing grains on top of the rolling-slipping layer. Size ratio and denstity difference between small and large grains seems to be decisive properties of mixtures for thin stripe forming. In particular, denser small grains can bounce as far as large grains. Two important experimental parameters were incoming flux and initial velocity of grains. More work on the subject is needed.
NV thanks the FNRS (Brussels, Belgium) for financial support. This work is also supported by the Belgian Royal Academy of Sciences through the Ochs-Lefebvre prize. Valuable discussion with M.Ausloos, J.Rajchenbach, E.Clément, S.Galam are acknowledged.
|
warning/0004/hep-ex0004002.html
|
ar5iv
|
text
|
# Resonance structure of 𝝉⁻→𝑲⁻𝝅⁺𝝅⁻𝝂_𝝉 decays
## I Introduction
Decays of the $`\tau `$ lepton into three pseudoscalars have been actively studied over the last several years. Lately, a number of relatively precise measurements of the branching fractions for $`\tau ^{}K^{}\pi ^+\pi ^{}\nu _\tau `$Charge conjugate states are implied throughout the paper. have become available from the ALEPH, CLEO and OPAL collaborations. However, the resonance substructure of these decays has not yet been measured with high precision.
The decay $`\tau ^{}K^{}\pi ^+\pi ^{}\nu _\tau `$, with its simple and well-understood initial state provides information on low-$`Q^2`$ QCD. The effects of $`SU(3)_f`$ symmetry breaking can be observed, the decay constants of the $`K_1`$ resonances can be measured, and the hadronic resonance substructure can be studied from an analysis of the final state invariant mass spectra . Other interesting topics include resonance parameters (such as the widths of the $`K_1`$ states), tests of isospin relations, and measurements of the Wess-Zumino anomaly. Current models of this decay are based on Chiral Perturbation Theory (ChPT) calculations. The question of the $`K_1`$ widths is of special interest because the theoretical models based on ChPT provide predictions for the $`\tau ^{}K^{}h^+h^{}\nu _\tau `$ branching fractions that are significantly larger than current experimental values. This discrepancy can be resolved if the $`K_1`$ resonances in $`\tau `$ decays are much wider than presently measured values. In the non-strange sector, it has long been realized that the $`a_1`$ width is considerably larger as measured in $`\tau ^{}a_1^{}\nu _\tau `$ compared to hadronic production of the $`a_1`$. The primary goal of this analysis is to measure the relative amplitudes of the $`K_1(1270)`$ and $`K_1(1400)`$ resonances that are believed to dominate $`\tau ^{}K^{}\pi ^+\pi ^{}\nu _\tau `$ decays and to determine the parameters of the $`K_1`$ resonances.
## II Theoretical aspects of $`\tau ^{}K^{}\pi ^+\pi ^{}\nu _\tau `$ decays
In the Standard Model, the general form for the semileptonic $`\tau `$-decay matrix element can be written as
$$=\frac{G}{\sqrt{2}}\overline{u}(p_\nu )\gamma ^\mu (1\gamma _5)u(p_\tau )J_\mu ,$$
(1)
where $`J_\mu K\pi \pi |V_\mu A_\mu |0`$ is the hadronic current and $`p_\nu `$ and $`p_\tau `$ are the four-momenta of the $`\tau `$ neutrino and the $`\tau `$ lepton, respectively.
General considerations based on Lorentz invariance and conservation of energy and momentum lead to the conclusion that only four independent form-factors are needed to describe the hadronic current in $`\tau ^{}K^{}\pi ^+\pi ^{}\nu _\tau `$. One parameterization describes this process as
$`J^\mu =`$ $`[F_1(s_1,s_2,Q^2)(p_2p_3)_\nu +F_2(s_1,s_2,Q^2)(p_1p_3)_\nu ]T^{\mu \nu }`$ (2)
$`+F_a(s_1,s_2,Q^2)ϵ^{\mu \nu \rho \sigma }p_{1\nu }p_{2\rho }p_{3\sigma }+F_s(s_1,s_2,Q^2)Q^\mu ,`$ (3)
where $`F_i`$ are form-factors, $`Q^\mu `$ is the $`K\pi \pi `$ 4-vector, $`s_1`$ is expressed in terms of the final state hadrons’ momenta $`p_i`$ ($`i`$=1 for the $`K^{}`$, $`i`$=2 for $`\pi ^+`$ and $`i`$=3 for $`\pi ^{}`$) as $`s_1=(p_2+p_3)^2`$, $`s_2=(p_1+p_3)^2`$ and $`T^{\mu \nu }=(g^{\mu \nu }Q^\mu Q^\nu /Q^2)`$. Here, there are two axial vector form-factors $`F_1`$ and $`F_2`$, an anomalous vector form-factor $`F_a`$, and a scalar form-factor $`F_s`$.
To derive specific expressions for the form-factors, some assumptions have to be made. It is believed that this decay is dominated by the lowest-mass resonances. There are two axial vector resonances which can produce the $`K^{}\pi ^+\pi ^{}`$ final state. These are the weak eigenstates $`{}_{}{}^{3}P_{1}^{}`$ and $`{}_{}{}^{1}P_{1}^{}`$ $`(u\overline{s})`$, called $`K_a`$ and $`K_b`$. The $`K_b`$ couples to the $`W`$ analagous to a “second class” current, violating $`SU(3)_f`$ symmetry. These two weak eigenstates mix with mixing angle $`\theta _K`$ to form the observable mass eigenstates, $`K_1(1270)`$ and $`K_1(1400)`$ . The $`K_1(1270)`$ subsequently decays into $`K^{}\pi `$, $`K\rho `$ or $`K\rho ^{}`$, while the $`K_1(1400)`$ decays almost entirely to $`K^{}\pi `$.
Within the context of ChPT, the form-factors can be written as
$$F_1=\frac{\sqrt{2}}{3}BW_{K_1(1270)}\frac{BW_\rho +\xi BW_\rho ^{}}{1+\xi }$$
(4)
$$F_2=\frac{\sqrt{2}}{3}\frac{\eta BW_{K_1(1270)}+BW_{K_1(1400)}}{(1+\eta )}BW_K^{},$$
(5)
where $`BW`$ denotes a Breit-Wigner mass distribution. The parameter $`\eta `$ is estimated to be 0.33 in . The coefficients preceding the Breit-Wigner expressions are fixed by ChPT. In the first form-factor, the coefficient $`\xi `$ is taken to be -0.145 based on application of the Conserved Vector Current (CVC) to $`e^+e^{}\pi ^+\pi ^{}`$ data .
In the chiral limit, the scalar form-factor $`F_s`$ is zero. The vector form-factor $`F_a`$ is expected to be numerically small compared to $`F_1`$ and $`F_2`$; it is only non-zero due to the Wess-Zumino anomaly. The vector contribution is approximately $`5.5\%`$ as calculated using the decay amplitudes found in . For this analysis, we will assume that the vector contribution is zero, and include our uncertainty in this term as a systematic error.
ChPT has found widespread application in tau decays. A model similar to this is used for our analysis of the invariant mass distributions in $`\tau ^{}K^{}\pi ^+\pi ^{}\nu _\tau `$, as will be described in the following sections.
## III Data sample and event selection
Our data sample contains approximately 4.3 million $`\tau `$-pairs produced in $`e^+e^{}`$ collisions, corresponding to an integrated luminosity of 4.7 fb<sup>-1</sup>. The data were collected with the CLEO II detector at the Cornell Electron Storage Ring, operating at a center-of-mass energy approximately 10.58 GeV.
The CLEO II detector is a general-purpose solenoidal magnet spectrometer and calorimeter. The detector was designed for efficient triggering and reconstruction of two-photon, tau-pair, and hadronic events. Measurements of charged particle momenta are made with three nested coaxial drift chambers consisting of 6, 10, and 51 layers, respectively. These chambers fill the volume from $`r`$=3 cm to $`r`$=1 m, with $`r`$ the radial coordinate relative to the beam ($`\widehat{z}`$) axis. This system is very efficient ($`ϵ`$98%) for detecting tracks that have transverse momenta ($`p_T`$) relative to the beam axis greater than 200 MeV/$`c`$, and that are contained within the good fiducial volume of the drift chamber ($`|\mathrm{cos}\theta |<`$0.94, with $`\theta `$ defined as the polar angle relative to the beam axis).In this analysis we use charged tracks with momentum above 300 MeV/$`c`$. This system achieves a momentum resolution of $`(\delta p/p)^2=(0.0015p)^2+(0.005)^2`$ ($`p`$ is the momentum, measured in GeV/$`c`$). Pulse-height measurements in the main drift chamber provide specific ionization ($`dE/dx`$) resolution of 5.5% for Bhabha events, giving good $`K/\pi `$ separation for tracks with momenta up to 700 MeV/$`c`$ and nearly 2$`\sigma `$ separation in the relativistic rise region above 2 GeV/$`c`$. Outside the central tracking chambers are plastic scintillation counters, which are used as a fast element in the trigger system and also provide particle identification information from time-of-flight measurements.
Beyond the time-of-flight system is the electromagnetic calorimeter, consisting of 7800 thallium-doped CsI crystals. The central “barrel” region of the calorimeter covers about 75% of the solid angle and has an energy resolution which is empirically found to follow:
$$\frac{\sigma _\mathrm{E}}{E}(\%)=\frac{0.35}{E^{0.75}}+1.90.1E;$$
(6)
$`E`$ is the shower energy in GeV. This parameterization includes effects such as noise, and translates to an energy resolution of about 4% at 100 MeV and 1.2% at 5 GeV. Two end-cap regions of the crystal calorimeter extend solid angle coverage to about 95% of $`4\pi `$, although energy resolution is not as good as that of the barrel region. The tracking system, time of flight counters, and calorimeter are all contained within a superconducting coil operated at 1.5 Tesla. Flux return and tracking chambers used for muon detection are located immediately outside the coil and in the two end-cap regions.
We select $`e^+e^{}\tau ^+\tau ^{}`$ events having a 1-prong vs. 3-prong topology in which one $`\tau `$ lepton decays into one charged particle (plus possible neutrals), and the other $`\tau `$ lepton decays into 3 charged hadrons (plus possible neutrals). An event is separated into two hemispheres based on the measured event thrust axis.<sup>\**</sup><sup>\**</sup>\**The thrust axis of an event is chosen so that the sum of longitudinal (relative to this axis) momenta of all charged tracks has a maximum value. Loose cuts on ionization measured in the drift chamber, energy deposited in the calorimeter and the maximum penetration depth into the muon detector system are applied to charged tracks in the signal (3-prong) hemisphere to reject leptons. Backgrounds from non-signal $`\tau `$ decays and hadronic events with $`K_S^0`$ are suppressed by requirements on the impact parameters of charged tracks. To reduce the background from two-photon collisions ($`e^+e^{}e^+e^{}\gamma \gamma `$ with $`\gamma \gamma `$hadrons or $`\gamma \gamma l^+l^{}`$), cuts on visible energy ($`E_{vis}`$) and total event transverse momentum ($`P_t`$) are applied: $`2.5`$ GeV $`<E_{vis}<10`$ GeV, and $`P_t>`$0.3 GeV/$`c`$. We also require the invariant mass of the tracks and showers in the 3-prong hemisphere, calculated under the $`\pi ^{}\pi ^+\pi ^{}`$ hypothesis, to be less than 1.7 GeV/$`c`$. Events are accepted for which the tag hemisphere (1-prong side) is consistent with one of the following four decays: $`\tau ^+e^+\nu _e\overline{\nu }_\tau `$, $`\tau ^+\mu ^+\nu _\mu \overline{\nu }_\tau `$, $`\tau ^+\pi ^+\overline{\nu }_\tau `$, or $`\tau ^+\rho ^+\overline{\nu }_\tau `$.
Candidate events are distinguished from background $`\tau `$ decays with $`\pi ^0`$’s and continuum hadronic background ($`e^+e^{}q\overline{q}`$) by the characteristics of showers in the electromagnetic calorimeter. A photon candidate is defined as a shower in the barrel region of the electromagnetic calorimeter with energy above 100 MeV having an energy deposition pattern consistent with true photons. It must be separated from the closest charged track by at least 30 cm. $`\tau ^{}K^{}\pi ^+\pi ^{}\nu _\tau `$ candidates are defined as those events having zero photon candidates in the 3-prong hemisphere.
The event selection described above provides a sample of events that contains $`\tau ^{}\pi ^{}\pi ^+\pi ^{}\nu _\tau `$, $`\tau ^{}K^{}\pi ^+\pi ^{}\nu _\tau `$ and $`\tau ^{}K^{}K^+\pi ^{}\nu _\tau `$. In this analysis we neglect possible contributions from the decays $`\tau ^{}\pi ^{}K^+\pi ^{}\nu _\tau `$ and $`\tau ^{}K^{}\pi ^+K^{}\nu _\tau `$ because they are unphysical in the Standard Model and have not been experimentally observed. We also neglect the $`\tau ^{}K^{}K^+K^{}\nu _\tau `$ final state. This rate is expected to be $`1\%`$ relative to that for $`\tau ^{}K^{}\pi ^+\pi ^{}\nu _\tau `$ due to the limited phase space and the low probability of forming an $`(s\overline{s})`$ pair from the vacuum.
## IV Reconstruction of invariant mass spectra
Due to the very small fraction of kaons in $`\tau ^{}h^{}h^+h^{}\nu _\tau `$ events<sup>††</sup><sup>††</sup>††Here and later $`h`$ designates either a kaon or pion. and the limited particle identification capabilities of the CLEO II detector, it is difficult to identify individual $`\tau ^{}K^{}\pi ^+\pi ^{}\nu _\tau `$ decays. In this analysis, a statistical approach is used in which the number of $`\tau ^{}K^{}h^+\pi ^{}\nu _\tau `$ events in any given sample is determined using the $`dE/dx`$ information of the two same-sign tracks in the signal hemisphere. The $`dE/dx`$ analysis is described in detail in .
For each $`h^{}h^+h^{}`$ candidate the invariant mass of the three hadrons is calculated under two hypotheses for the first and the third tracks, corresponding to the $`K^{}\pi ^+\pi ^{}`$ and $`\pi ^{}\pi ^+K^{}`$ mass assignments. Each of these two sub-samples is divided into bins of invariant mass. The bins are 100 MeV/$`c^2`$ wide, spanning the region $`0.81.7`$ GeV/$`c^2`$. After binning in mass, the sub-samples that correspond to the same mass bin are combined.
The $`dE/dx`$ analysis provides the number of kaons in each mass bin, which is equal to the number of $`\tau ^{}K^{}h^+\pi ^{}\nu _\tau `$ events in that mass interval. The invariant mass spectrum of the $`K^{}\pi ^+\pi ^{}`$ system is thereby reconstructed. This distribution contains a contribution from $`\tau ^{}K^{}K^+\pi ^{}\nu _\tau `$ decays which must be subtracted, as will be discussed in Sec.V.
In a similar way the invariant mass spectra of $`K^{}\pi ^+`$ and $`\pi ^+\pi ^{}`$ are reconstructed in ten bins over the range $`0.5`$ GeV/$`c^2<M_{K^{}\pi ^+}<1.5`$ GeV/$`c^2`$ and $`0.2`$ GeV/$`c^2<M_{\pi ^+\pi ^{}}<1.2`$ GeV/$`c^2`$, respectively. The reconstructed mass spectra are shown in Fig. 1.
## V Background and efficiency
There are two main types of background: continuum hadronic events ($`e^+e^{}q\overline{q}hadrons`$) and non-signal $`\tau `$ decays. Hadronic background is estimated from a continuum hadronic Monte Carlo sample (using the JETSET v7.3 event generator and GEANT detector simulation code). This background is subtracted as described in . The level of hadronic background is approximately 3%.
$`\tau `$-related background comes primarily from $`\tau ^{}K^{}K^+\pi ^{}\nu _\tau `$ decays. These events comprise approximately $`27\%`$ of the events in our reconstructed invariant mass distributions. Smaller background contributions arise from $`\tau ^{}K^{}h^+\pi ^{}\pi ^0\nu _\tau `$ with incomplete $`\pi ^0`$ reconstruction and also tau decays to the $`K^{}\pi ^+\pi ^{}(\pi ^0)\nu _\tau `$ final state through an intermediate $`K_S^0`$. These two backgrounds comprise 5% and 3% of the events in the $`K^{}\pi ^+\pi ^{}`$ invariant mass spectra, respectively. The invariant mass distributions for backgrounds are found using Monte Carlo simulations to obtain the shape; the normalization is set by the measured branching fractions . Background predictions are shown in Fig. 1. The invariant mass distributions for all backgrounds are subtracted from the corresponding invariant mass spectra reconstructed from data.
The efficiency of event reconstruction depends slightly on the invariant mass. Therefore, it is necessary to introduce a mass-dependent efficiency correction. This correction is calculated from $`\tau `$ Monte Carlo using the KORALB event generator . The maximum variation in efficiency across the mass interval of interest is of order 10%.
## VI Fitting method
The hadronic structure of the $`K^{}\pi ^+\pi ^{}`$ system is investigated by simultaneously fitting three invariant mass distributions: $`M_{K^{}\pi ^+\pi ^{}}`$, $`M_{K^{}\pi ^+}`$ and $`M_{\pi ^+\pi ^{}}`$. The fitting function is based upon a model similar to the one described in Sec. I, and now outlined in greater detail.
### A Parameterization of form-factors
In this analysis, we write the following expression for the axial vector form-factors $`F_1`$ and $`F_2`$:
$$F_1=\frac{1}{\sqrt{3}}(ABW_{1270}+BBW_{1400})\frac{BW_\rho +\xi BW_\rho ^{}}{1+\xi }$$
(7)
$$F_2=(\frac{2}{3})(CBW_{1270}+DBW_{1400})BW_K^{}$$
(8)
that contain the four real parameters $`AD`$. Of these, 3 are independent; the fourth is fixed by the normalization requirement that the squared sum of the $`\tau ^{}K_1^{}(1270)\nu _\tau `$ and $`\tau ^{}K_1^{}(1400)\nu _\tau `$ amplitudes must saturate the total $`\tau ^{}K^{}\pi ^+\pi ^{}\nu _\tau `$ rate. The coefficients $`A`$ and $`B`$ correspond to production of the $`K\rho `$ final state through either $`K_1`$(1270) ($`A`$) or $`K_1`$(1400) ($`B`$), modulo a factor which includes the appropriate phase space weighting for various final states (denoted as “$`R_A`$”, or “$`R_B`$”). In our analysis, we fix $`B`$ to be zero, consistent with current measurements . Similarly, $`C`$ and $`D`$ designate production of the $`K^{}\pi `$ final state through the $`K_1(1270)`$ and $`K_1(1400)`$ resonances. The decay amplitude parameters in Eqs. (7)-(8) therefore correspond to the possible decay chains as
$`\tau K_1(1270)\nu K\rho \nu K^{}\pi ^+\pi ^{}\nu _\tau :`$ $`\mathrm{`}\mathrm{`}A^{\prime \prime }`$ (9)
$`\tau K_1(1270)\nu K^{}\pi \nu K^{}\pi ^+\pi ^{}\nu _\tau :`$ $`\mathrm{`}\mathrm{`}C^{\prime \prime }=`$ $`A\sqrt{{\displaystyle \frac{16}{42}}}\sqrt{{\displaystyle \frac{R_A}{R_C}}}`$ (10)
$`\tau K_1(1400)\nu K^{}\pi \nu K^{}\pi ^+\pi ^{}\nu _\tau :`$ $`\mathrm{`}\mathrm{`}D^{\prime \prime }=`$ $`\sqrt{1A^2C^2}`$ (11)
In Eqs. (9)-(10) we have imposed constraints that follow from the tabulated branching fractions of the $`K_1`$ resonances : $`(K_1(1270)K^{}\pi )=(16\pm 5)`$% and $`(K_1(1270)K\rho )=(42\pm 6)`$%.
Thus, in our parameterization of the matrix element one unknown parameter $`A`$ defines all four amplitudes. In addition, the masses and widths of the $`K_1`$ resonances $`\mathrm{\Gamma }_{K_1(1270)}`$, $`\mathrm{\Gamma }_{K_1(1400)}`$, $`M_{K_1(1270)}`$, $`M_{K_1(1400)}`$ are considered unknown and left as free parameters in the fit. The Breit-Wigner distributions for the $`K_1`$ resonances are defined following the approach of as
$$BW(s,m_{K_1},\mathrm{\Gamma }_{K_1})=\frac{m_{K_1}^2im_{K_1}\mathrm{\Gamma }_{K_1}}{m_{K_1}^2sim_{K_1}\mathrm{\Gamma }_{K_1}}.$$
(12)
The Breit-Wigner distributions for the $`K^{}`$ and $`\rho `$ resonances contain mass-dependent widths:
$$BW(s,m,\mathrm{\Gamma })=\frac{m^2}{m^2si\sqrt{s}\mathrm{\Gamma }(s)},$$
(13)
where the mass-dependence is defined by Eq. (22) in :
$$\mathrm{\Gamma }(s)=\mathrm{\Gamma }_0\frac{m_0^2}{s}\left(\frac{p(s)}{p(m_0^2)}\right)^{\frac{3}{2}}.$$
(14)
Here, $`m_0`$ and $`\mathrm{\Gamma }_0`$ are the nominal mass and width of a particle, $`p`$ is the momentum of the particle, and the variable $`s`$ is the three- or two-body invariant mass, as appropriate.
The constants $`R_X`$ (where $`X=A,\mathrm{},D`$) in Eqs. (9), (10), and (11) depend on the masses and widths of the 2- and 3-body resonances in this decay and are calculated by numerical integration of the appropriate matrix element. Since we are interested in the ratios of quantities (e.g., branching fractions), the overall normalization of the $`R_X`$ parameters is arbitrary.
In this analysis, we have determined the numerical coefficients for the Breit-Wigner terms \[$`\sqrt{\frac{1}{3}}`$ and $`\frac{2}{3}`$ in Eqs. (7)-(8)\] using isospin relations rather than taking the chiral limit as in . The Wess-Zumino anomaly term is set to zero in our model; this term is numerically small enough that it can be neglected at our level of accuracy. Appropriate systematic errors are assigned to reflect the possible magnitude of this contribution. Note, however, that if the Wess-Zumino anomaly is much larger than expected, the $`K^{^{}}`$ may contribute events to the region of $`M_{K^{}\pi ^+\pi ^{}}`$ invariant mass close to 1.4 GeV/$`c^2`$, affecting our measurement of $`AD`$. Note also that we explicitly assume all $`\tau ^{}K^{}\pi ^+\pi ^{}\nu _\tau `$ decays proceed through either $`K_1(1270)`$ or $`K_1(1400)`$.
In principle, there may be a phase shift between the terms in the form-factors $`F_1`$ and $`F_2`$. Such phase differences may appear between various decay chains producing the final state $`K^{}\pi ^+\pi ^{}`$. This may cause additional constructive or destructive interference and, for example, enhance or suppress the $`\rho `$ peak in the distribution of $`\pi \pi `$ invariant mass. In the most general approach, one would introduce three independent phase angles $`\theta _1`$, $`\theta _2`$ and $`\theta _3`$, corresponding to the possible interfering decay chains. However, due to limited statistics, we have neglected such possible interference effects, and take into account only the inherent phase of the Breit-Wigner distributions (as described above).
### B Calculation of the observables
The interesting observables that we would like to measure are the relative branching fractions to the different $`K_1`$ resonances and the amounts of $`K^{}`$ and $`\rho `$ in this decay. The decay rate for any individual decay chain is proportional to $`X^2R_X`$. For example, the decay rate for the chain in Eq. 9 is proportional to $`A^2R_A`$. In calculating the ratios we choose to normalize to the sum of the separate contributions (not including interference effects). With this convention, the fractions of different contributions add to 100%.
With the above definitions we write
$$f_{1270}\frac{(\tau K_1(1270)\nu )}{(\tau K_1(1270)\nu )+(\tau K_1(1400)\nu )}=\frac{A^2R_A+C^2R_C}{A^2R_A+C^2R_C+D^2R_D}$$
(15)
and
$$f_\rho \frac{(\tau K\rho \nu )}{(\tau K\pi \pi \nu )}=\frac{A^2R_A}{A^2R_A+C^2R_C+D^2R_D}.$$
(16)
### C Fitting function
We use a Monte Carlo based fitting procedure, in which a large number (200,000) of simulated events are used to simultaneously fit the data distributions for $`M_{K\pi \pi }`$, $`M_{K\pi }`$ and $`M_{\pi \pi }`$. From a binned $`\chi ^2`$ fit, we determine the input values of $`A`$, $`\mathrm{\Gamma }_{K_1(1270)}`$, $`\mathrm{\Gamma }_{K_1(1400)}`$, $`M_{K_1(1270)}`$ and $`M_{K_1(1400)}`$ which give the best simultaneous match to these mass spectra. As outlined above, our event generator is identical to KORALB except that our form-factors \[Eqs.(7)-(8)\] are used and the Wess-Zumino form-factor is set to zero. To take into account finite resolution effects we introduce Gaussian smearing of the calculated mass equal to the smearing found from the full GEANT-based simulation of the detector. This smearing is typically 5-10 MeV/$`c^2`$.
## VII Fit results
The result of our fit is shown in Figure 2; the best values for the fit parameters are tabulated in Table I. In the same table the values of $`R_X`$ obtained from numerical integration and the derived values for $`f_{1270}`$ and $`f_\rho `$ are also given. The first error in the Table is statistical and the second is systematic (discussed in Sec. VIII). The statistical errors on $`A`$, the masses and widths are calculated using HESSE in MINUIT and take into account correlations between the fit parameters. The asymmetric statistical errors, where appropriate, are also evaluated from the fit.
The fit results showing contours of constant $`\chi ^2`$ in the $`M_{1270}`$ vs. $`M_{1400}`$ and $`\mathrm{\Gamma }_{1270}`$ vs. $`\mathrm{\Gamma }_{1400}`$ planes are shown in Figures 3 and 4. (Note that the constraint $`M_{1400}>M_{1270}`$ is introduced in obtaining Figures 3-4; we therefore fit only over the region above the dashed line.) In these plots, the curves represent 1$`\sigma `$, 2$`\sigma `$, 3$`\sigma `$ and 4$`\sigma `$ standard deviation error contours around the best fit point (indicated by a cross).
We perform a secondary fit, in which the masses and widths of the $`K_1`$ resonances are fixed to world average values . We obtain $`f_{1270}=0.40\pm 0.07`$, and $`f_\rho =0.29\pm 0.05`$ from this second fit (statistical errors only), to be contrasted with the significantly larger values extracted from our full fit, in which the $`K_1`$ masses and widths are allowed to float as free parameters. It is not surprising that the $`f`$ values are different in this second fit compared to the original fit, given the high degree of correlation between the $`K_1`$ widths and the relative branching fractions. The $`\chi ^2`$ for this second fit is considerably poorer (30.5/22 degrees of freedom) than the primary fit (12.6/18 degrees of freedom).
## VIII Systematic uncertainties
Systematic errors are summarized in Table II. The dominant errors are due to the uncertainty in the $`K_1`$ branching fractions and the uncertainty in the background. The errors associated with the uncertainty in the branching fractions of the $`K_1`$ resonances to $`K^{}\pi `$ and $`K\rho `$ are estimated by changing the values for these branching fractions in the form-factors $`F_1`$ and $`F_2`$ by one standard deviation of the PDG values. The resulting spread in the fit values is taken as the corresponding systematic error.
The uncertainty in the background is large because of the uncertainty in the branching fraction of its largest component, $`\tau ^{}K^{}K^+\pi ^{}\nu _\tau `$. During the background subtraction the level of all $`\tau `$-related backgrounds are varied by amounts corresponding to the errors on the branching fractions of these decays . The hadronic background is similarly varied by 100% to determine the systematic error due to our uncertainty in the $`q\overline{q}`$ background contribution.
Another large error comes from the choice of models in our Monte Carlo simulation. This includes the uncertainty in the shape of the kaon momentum spectrum ($`P_K`$) used to obtain the total number of events with kaons , and the uncertainty in the shape of the invariant mass distribution for the $`\tau ^{}K^{}K^+\pi ^{}\nu _\tau `$ background. These errors are estimated by using several different models to extract the invariant mass spectra. For $`\tau ^{}K^{}\pi ^+\pi ^{}\nu _\tau `$, we consider $`\tau ^{}K_1(1270)^{}\nu _\tau `$ and $`\tau ^{}K_1(1400)^{}\nu _\tau `$ and the model described in ; for $`\tau ^{}K^{}K^+\pi ^{}\nu _\tau `$, we consider $`\tau ^{}a_1^{}\nu _\tau K^0K^{}\nu _\tau `$, $`\tau ^{}\rho (1690)^{}\nu _\tau K^0K^{}\nu _\tau `$ and the model described in .
There are several fitting function uncertainties. The first is the contribution from the $`\rho ^{}`$ which may be different in this decay from that observed in $`e^+e^{}\pi ^+\pi ^{}`$ data due to the phase space suppression of $`\rho ^{}`$ in our case. Second, the model implemented in our fitting function contains no contribution from the vector current. The corresponding fitting function errors from these two sources are estimated by varying the level of the Wess-Zumino term and the $`\rho ^{}`$ amplitudes from zero to the predictions of and . Another possible source of systematic errors is a phase shift among the interfering decay chains. In this analysis, the parameters $`AD`$ in Eqns. (7)-(8) are real. We have done a study of interference effects with additional phase shifts and found that the possible imaginary part of $`AD`$ is consistent with zero at our level of sensitivity. Because the fitting function is based upon Monte Carlo, the Monte Carlo statistical error is also included here.
The bias associated with the procedure of fitting the invariant mass distribution is studied using 60 samples of signal Monte Carlo with a full detector simulation. The results of this study show no systematic shift of the fitted parameter values relative to the input values. Additionally, the errors we obtain from analyzing this Monte Carlo sample are fully consistent with statistical expectations.
In this analysis only the shape of the background-subtracted invariant mass distribution is of interest; possible systematic effects that affect the overall normalization of the reconstructed spectra are ignored. Among such effects are trigger and tracking efficiencies, and the photon veto. Non-$`\tau `$ backgrounds (2-photon events, beam-gas interactions, QED background, e.g.) have been determined to be negligible for the mass spectrum analysis.
## IX Discussion of Resonance Structure
### A Masses and Widths
As mentioned previously, theoretical predictions for $`(\tau Kh^+h^{}\nu _\tau )`$ based on ChPT are substantially larger than data. However, if the $`K_1`$ resonances are substantially broader than the PDG values, this discrepancy is resolved. In fact, our data suggest larger $`K_1`$ widths than previous world averages (this is evident from Fig. 2). As indicated in Table I, we extract the masses and widths of the $`K_1`$ resonances from this fit: $`\mathrm{\Gamma }_{1270}=0.26_{0.07}^{+0.09}`$ GeV, $`\mathrm{\Gamma }_{1400}=0.30_{0.11}^{+0.37}`$ GeV, $`M_{1270}=1.254\pm 0.033`$ GeV/$`c^2`$, and $`M_{1400}=1.463\pm 0.064`$ GeV/$`c^2`$.
In Table III, our result for the widths is compared to the data from ALEPH and DELPHI in their analyses of $`\tau `$ decays. One observes that all experimental data from $`\tau K\pi \pi \nu _\tau `$ for the $`K_1`$ widths are above the current world averages although the errors remain large. The masses of $`K_1(1270)`$ and $`K_1(1400)`$ measured in our analysis are in acceptable agreement with the current world averages .
### B Values of $`f_{1270}`$ and $`f_\rho `$
As calculated in section VII, our data indicate that there is slightly more $`K_1(1270)`$ than $`K_1(1400)`$ in the axial vector current of the $`\tau ^{}K^{}\pi ^+\pi ^{}\nu _\tau `$ decay ($`f_{1270}=0.66\pm 0.19\pm 0.13`$). Other experiments have also investigated the relative contributions of the two $`K_1`$ resonances to $`\tau s\overline{u}\nu _\tau `$. One of the first measurements of the $`\tau K_1\nu _\tau `$ branching fractions was performed by the $`TPC/2\gamma `$ collaboration in 1994 . Their results are $`(\tau K_1(1270)\nu )=0.41_{0.35}^{+0.41}`$% and $`(\tau K_1(1400)\nu )=0.76_{0.33}^{+0.40}`$%, giving the fraction $`f_{1270}=0.35_{0.35}^{+0.73}`$. The results of the $`TPC/2\gamma `$ experiment suggest that the decay proceeds mostly through $`K_1(1400)`$ although their errors are too large to draw firm conclusions. The latest branching fraction measurements by CLEO and ALEPH as well as this analysis suggest $`K_1(1270)`$ dominance. An analysis of the $`K^{}\pi ^+`$ and $`\pi ^{}\pi ^+`$ substructure in $`\tau ^{}K^{}\pi ^+\pi ^{}\nu _\tau `$ allowed ALEPH to determine $`f_{1270}=0.41\pm 0.19\pm 0.15`$, based on the known branching fractions of the $`K_1`$ resonances to $`K^{}\pi `$ and $`K\rho `$. Recent measurements therefore favor $`K_1`$(1270) dominance in $`\tau ^{}K^{}\pi ^+\pi ^{}\nu _\tau `$.
Calculating the amount of $`\rho `$ in $`K\pi \pi `$ from the fit parameters we find $`f_\rho =0.48\pm 0.14`$, close to the measurement by ALEPH of $`f_\rho =0.35\pm 0.11`$. This number also agrees with the measurement by ALEPH of another related decay channel, $`\tau ^{}K_S^0\pi ^{}\pi ^0\nu _\tau `$ where the component $`\overline{K}^0\rho ^{}`$ in the intermediate state is found to be $`(64\pm 9\pm 10)\%`$, approximately twice that of $`K^{}\rho ^0`$ as expected by isospin symmetries .
### C $`K_aK_b`$ mixing
From our result for the ratio of $`\tau K_1\nu _\tau `$ decay amplitudes, information about the mixing of the $`K_a`$ and $`K_b`$ eigenstates can be derived. The mixing between $`K_a`$ and $`K_b`$ is traditionally parameterized in the following way :
$`K_1(1400)=`$ $`K_a\mathrm{cos}\theta _KK_b\mathrm{sin}\theta _K`$ (17)
$`K_1(1270)=`$ $`K_a\mathrm{sin}\theta _K+K_b\mathrm{cos}\theta _K`$ (18)
In the case of exact $`SU(3)_f`$ symmetry, the second-class current $`\tau K_b\nu _\tau `$ is forbidden and only $`K_a`$ is produced. However, due to the difference between the masses of the up and strange quarks we may expect symmetry breaking effects of order $`|\delta |=(m_sm_u)/\sqrt{2}(m_s+m_u)0.18`$. Then, instead of pure $`K_a`$ a linear combination $`|K_a\delta |K_b`$ is produced and the ratio of decay rates of the $`K_1`$ resonances can be written as :
$$\frac{(\tau K_1(1270)\nu )}{(\tau K_1(1400)\nu )}=\left|\frac{\mathrm{sin}\theta _K\delta \mathrm{cos}\theta _K}{\mathrm{cos}\theta _K+\delta \mathrm{sin}\theta _K}\right|^2\times \mathrm{\Phi }^2$$
(19)
In this expression, $`\mathrm{\Phi }`$ is the ratio of appropriate kinematical and phase space terms and is calculated by numerical integration. With the parameters measured in this analysis the ratio of branching fractions (Eq. 19) is written as
$$\frac{(\tau K_1(1270)\nu _\tau )}{(\tau K_1(1400)\nu _\tau )}=\frac{A^2R_A+C^2R_C}{D^2R_D}$$
(20)
From Eqs. (19)-(20), solutions for $`\theta _K`$ can easily be found:
$`\text{(a) }\theta _K`$ $`=(69\pm 16\pm 19)^{}\text{ for }\delta =0.18,`$ (21)
$`\text{(b) }\theta _K`$ $`=(49\pm 16\pm 19)^{}\text{ for }\delta =0.18.`$ (22)
There is a second pair of solutions that has opposite sign and the same magnitude.
One can also calculate $`\theta _K`$ using the current experimental information on the masses and branching fractions of $`K_1(1270)`$ and $`K_1(1400)`$, independent of their production in $`\tau `$-decays. There are two possible solutions, $`\theta _K33^{}`$ and $`\theta _K57^{}`$ . Our result has the same two-fold ambiguity and is consistent with this calculation.
## X Summary and Conclusions
In this analysis we have measured the relative fractions and parameters of the $`K_1`$ resonances in $`\tau ^{}K^{}\pi ^+\pi ^{}\nu _\tau `$ decays. These measurements are made within the framework of the model described in Sec. VI. Briefly, we assume that $`K_1`$(1270) and $`K_1`$(1400) saturate the $`K^{}\pi ^+\pi ^{}`$ spectrum, and consider only the interference inherent in the Breit-Wigner mass distributions in calculating the relative $`\tau ^{}K_1(1270)\nu _\tau `$ and $`\tau ^{}K_1(1400)\nu _\tau `$ branching fractions. Our parameterization of the axial vector form-factors is different from in two respects – our form-factors are motivated by isospin relations, and we assume the Wess-Zumino anomaly to be negligible, as described in Sec. VI. We find $`f_{1270}=0.66\pm 0.19\pm 0.13`$ and $`f_\rho =0.48\pm 0.14\pm 0.10`$, with $`f_{1270}`$ and $`f_\rho `$ defined as the $`K_1(1270)`$ and $`K\rho `$ fractions in $`\tau K^{}\pi ^+\pi ^{}\nu _\tau `$.
These measurements agree well with the recent results from CLEO and ALEPH (see Sec VII). Our data slightly favor $`K_1(1270)`$ dominance in production of the $`K^{}\pi ^+\pi ^{}`$ final state. The widths that we extract for the $`K_1`$ resonances are considerably larger than previously tabulated values. We also calculate the $`K_aK_b`$ mixing angle, finding $`\theta _K`$ to be consistent with theoretical expectations.
###### Acknowledgements.
We gratefully acknowledge the effort of the CESR staff in providing us with excellent luminosity and running conditions. J.R. Patterson and I.P.J. Shipsey thank the NYI program of the NSF, M. Selen thanks the PFF program of the NSF, M. Selen and H. Yamamoto thank the OJI program of DOE, J.R. Patterson, K. Honscheid, M. Selen and V. Sharma thank the A.P. Sloan Foundation, M. Selen and V. Sharma thank Research Corporation, S. von Dombrowski thanks the Swiss National Science Foundation, and H. Schwarthoff thanks the Alexander von Humboldt Stiftung for support. This work was supported by the National Science Foundation, the U.S. Department of Energy, and the Natural Sciences and Engineering Research Council of Canada.
|
warning/0004/quant-ph0004064.html
|
ar5iv
|
text
|
# Theory of Decoherence-Free Fault-Tolerant Universal Quantum Computation
## I Introduction
The discovery that information encoded over quantum systems can exhibit strange and wonderful computational and information theoretic properties has led to an explosion of interest in understanding and exploiting the “quantumness” of nature. For the use of quantum information to progress beyond mere theoretical constructs into the realm of testable and useful implementations and experiments, it is essential to develop techniques for preserving quantum coherences. In particular, the coupling of a quantum system to its environment leads to a process known as decoherence, in which encoded quantum information is lost to the environment. In order to remedy this problem active quantum error correction codes (QECCs) have been developed, by analogy with classical error correction. These codes encode quantum information over an entangled set of codewords, the structure of which serves to preserve the quantum information, when used in conjunction with a frequently recurring error correcting procedure. It has been shown that when the rate of decoherence is below a certain threshold, fault tolerant quantum information manipulation is possible. Since it is believed that there are no systems for which the decoherence mechanism entirely vanishes, QECCs will be essential if quantum information manipulation is to become practical.
An alternative approach has been proposed and developed recently, in which the central motivation is the desire to reduce the effect of a specific decoherence mechanism. This is the decoherence-free subspace (DFS) approach (also referred to as “error avoiding”, or “noiseless” quantum codes) . In contrast to the active mode of QECC, DFS theory can be viewed as providing a passive approach, where a specific symmetry of the system-bath coupling is employed in order to seek out a quiet corner of the system’s Hilbert space which does not experience decoherence. Information encoded here over a subspace of (usually entangled) system states is robust against a specific form of decoherence. We shall refer to this as the “DFS supporting decoherence mechanism”. When this is the dominant form of decoherence in the physical system, there are major gains to be had by operating in the DFS. Previous work has shown that collective decoherence of the type experienced in condensed phase systems at low temperatures can be successfully eliminated in this way . Further research showed that DFSs are robust to perturbing error processes , and are thus ideally suited for concatenation in a QECC .
The motivating goal behind the DFS approach is to use symmetry first. Thus, one first identifies a DFS for the major sources of decoherence, via the symmetry of the interaction with the environment. One then proceeds to use the DFS states as a basis for a QECC which can deal with additional perturbing error processes. In order for this scheme to be credible, DFSs must support the ability to perform universal quantum computation on the encoded states. Towards this end, certain existential results have been derived showing that in principle universal quantum computation can be performed on any DFS. Constructive results for a set of universal quantum gates on a particular class DFSs were subsequently constructed in using known QECC constructions. However, these gates were constructed in such a way that during the operation of the gate, states within a DFS are taken outside of this subspace. Thus these gates would necessarily need to operate on a timescale faster than the DFS supporting decoherence mechanism, in order to be applied efficiently to a concatenated DFS-QECC scheme.<sup>*</sup><sup>*</sup>*Note that QECC fault-tolerant gates are also required to operate faster than the decoherence time of the main error process. Similarly, a universal computation result on DFSs for atoms in cavities was recently presented by Beige et al. in . It assumes that the interaction driving a system out of the DFS is much weaker than the coupling of non DF-states to the environment. It is then possible to make use of an environment-induced quantum Zeno effect. In order to make use of the robustness condition without resorting to gates which can be made faster than the main DFS supporting decoherence mechanism, one would prefer to explicitly construct a set of Hamiltonians which can be used to perform universal quantum computation, but which never allow states in the DFS to leak out of the DFS. Imperfections in these gates may be dealt with by the concatenation technique of (see also ).
In addition, one would, from a practical standpoint, like to use Hamiltonians which involve at most two-body interactions (under the assumption that any three-body interactions will be weak and not useful for operations which must compete with the decoherence rate). In such Hamiltonians were used for the important decoherence mechanism known as “collective decoherence”, on a system of $`4`$ physical qubits. In collective decoherence the bath cannot distinguish between individual system qubits, and thus couples in a collective manner to the qubits. The corresponding two-body Hamiltonians used to implement universal quantum computation are those that preserve the collective symmetry: the exchange interaction between pairs of qubits. The first and main purpose of this paper is therefore to extend the constructive results obtained in to other forms of collective decoherence and to larger DFSs. Two different forms of collective decoherence are considered here, and constructive results are obtained for these on DFSs of arbitrary numbers of qubits. These results have implications that extend far beyond the problem of dealing with collective decoherence. Since they imply that the exchange interaction by itself is sufficient to implement universal quantum computation on a subspace, it follows that using encoded (rather than physical) qubits can be advantageous when resources for physical operations are limited. After all, the standard results for universal quantum computation employ either arbitrary single-qubit operations in addition to a non-trivial two-qubit gate (e.g., a controlled-NOT), or at least two non-commuting two-qubit Hamiltonians . These issues will be explored in a separate publication.
Previous work established that DFSs correspond to the degenerate component of a QECC . A second purpose of this work is to present new results on a recently discovered generalization of DFSs, which has been termed “noiseless subsystems”, and arises from a theory of QECC for general decoherence mechanisms . In line with our previously established terminology we will refer to these as “decoherence-free subsystems”, where we take the term “decoherence” to mean both dephasing ($`T_2`$) and dissipation ($`T_1`$). Essentially, the generalization corresponds to allowing for information to be encoded into states transforming according to arbitrary-dimensional irreducible representations (irreps) of the decoherence-operators’ algebra, instead of just one-dimensional irreps as in the decoherence-free subspace case (we will present precise definitions later in this paper). These results all arise from a basic theorem on algebras that are closed under the Hermitian conjugation operation (“$``$-closed algebras”), and thereby unify the role of symmetry in both decoherence-free subspaces and quantum error correction. In this paper we extend the decoherence-free subsystem concept to situations governed by essentially non-$``$-closed evolution. Such situations arise from non-Hermitian terms in the system-bath interaction, which may occur, e.g., in generalized master equation and conditional Hamiltonian representations of open quantum dynamics . In particular, we derive an if and only if (iff) condition for the existence of decoherence-free subsystems with dynamics governed by a semigroup master equation. This is important because it is well known in decoherence-free subspace theory that such non-$``$-closed evolution can support different DFSs than in the $``$-closed case. A similar result is now shown here to hold for the decoherence-free subsystems.
Existential results for universal quantum computation on decoherence-free subsystems also exist . The universal quantum computation results we obtain in this paper extend beyond decoherence-free subspaces: we show how to achieve constructive universal quantum computation on the decoherence-free subsystems supported under collective decoherence. This most significant achievement of our paper settles the question of universal quantum computation under collective decoherence using realistic Hamiltonians.
Another aim of this paper is to elucidate the close link between DFS and QECC. In it was shown that DFSs are in fact maximally degenerate QECCs. This result was derived from the general condition for a code to be a QECC . A very fruitful approach towards QECC has been the stabilizer formalism developed in which led to the theory of universal fault-tolerant computation on QECCs . In we considered DFSs as abelian stabilizer codes. Here we generalize the stabilizer-framework to non-abelian stabilizers, and show that in general DFSs are stabilizer-codes that protect against errors in the stabilizer itself. This perspective allows in return to view QECCs as DFSs against a certain kind of errors, and establishes a kind of duality of QECCs and DFSs.
The paper is structured as follows: In Section II we review decoherence-free subsystems and place them into the context of the Markovian master equation. For decoherence-free subspaces this has been done in . These earlier results are therefore generalized here to subsystems. In Section III we introduce a generalized stabilizer-formalism for DFS, and connect to the theory of stabilizers on QECC developed in . This allows us to treat DFS and QECC within the same framework. It also sheds some light on the duality between DFS and QECC, in particular on the performance of a DFS viewed as a QECC and vice versa. In Section IV we deal with universal computation on DFS within both the stabilizer-framework and the representation-theoretic approach. We derive fault-tolerance properties of the universal operations. In particular, we show how to obtain operations that keep the states within a DFS during the entire switching-time of a gate. Further we define the allowed compositions of operations and review results on the length of gate sequences in terms of the desired accuracy of the target gates. In Section V we introduce the model of collective decoherence. Section VI explicitly deals with the abelian case of weak collective decoherence in which system-bath interaction coupling involves only a single system operator. Stabilizer and error-correcting properties are developed for this case, and it is shown how universal computation can be achieved. The same is done for the non-abelian and more general case of strong collective decoherence in Section VII. For both weak and strong collective decoherence we show how to fault-tolerantly encode into and read out of the respective DFSs. Finally, we analyze in Section VIII how to concatenate DFSs and QECCs to make them more robust against perturbing errors (as proposed in ) and show how the universality results can be applied to achieve fault-tolerant universal computation on these powerful concatenated codes. We conclude in Section IX. Derivations and proofs of a more technical nature are presented in the Appendix.
## II Overview of Decoherence-Free Subspaces and Subsystems
### A Decoherence-Free Subspaces
Consider the dynamics of a system $`S`$ (the quantum computer) coupled to a bath $`B`$ via the Hamiltonian
$$𝐇=𝐇_S𝐈_B+𝐈_S𝐇_B+𝐇_I,$$
(1)
where $`𝐇_S`$ ($`𝐇_B`$) \[the system (bath) Hamiltonian\] acts on the system (bath) Hilbert space $`_S`$ ($`_B`$), $`𝐈_S`$ ($`𝐈_B`$) is the identity operator on the system (bath) Hilbert space, and $`𝐇_I`$, which acts on both the system and bath Hilbert spaces $`_S_B`$, is the interaction Hamiltonian containing all the nontrivial couplings between system and bath. In general $`𝐇_I`$ can be written as a sum of operators which act separately on the system ($`𝐒_\alpha `$’s) and on the bath ($`𝐁_\alpha `$’s):
$$𝐇_I=\underset{\alpha }{}𝐒_\alpha 𝐁_\alpha .$$
(2)
In the absence of an interaction Hamiltonian ($`𝐇_I=0`$), the evolution of the system and the bath are separately unitary: $`𝐔(t)=\mathrm{exp}[i𝐇t]=\mathrm{exp}[i𝐇_St]\mathrm{exp}[i𝐇_Bt]`$ (we set $`\mathrm{}=1`$ throughout). Information that has been encoded (mapped) into states of the system Hilbert space remains encoded in the system Hilbert space if $`𝐇_I=0`$. However in the case when the interaction Hamiltonian contains nontrivial couplings between the system and the bath, information that has been encoded over the system Hilbert space does not remain encoded over solely the system Hilbert space but spreads out instead into the combined system and bath Hilbert space as the time evolution proceeds. Such leakage of quantum information from the system to the bath is the origin of the decoherence process in quantum mechanics.
Let $`\stackrel{~}{}_S`$ be a subspace of the system Hilbert space with a basis $`|\stackrel{~}{ı}`$. The evolution of such a subspace will be unitary if and only if (i)
$$𝐒_\alpha |\stackrel{~}{ı}=c_\alpha |\stackrel{~}{ı},c_\alpha \text{ }\mathrm{C}$$
(3)
for all $`|\stackrel{~}{ı}\stackrel{~}{}_S`$ and for all $`𝐒_\alpha `$, (ii) $`𝐇_S`$ does not mix states within the subspace with states that are outside of the subspace ($`j^{}|𝐇_S|\stackrel{~}{ı}=0`$ for all $`|\stackrel{~}{ı}`$ in the subspace and all $`|j^{}`$ outside of the subspace: $`𝐇_S=\stackrel{~}{𝐇}_S𝐇_S^{}`$) and (iii) system and bath are initially decoupled $`\rho (0)=\rho _S(0)\rho _B(0)`$. We call a subspace of the system’s Hilbert space which fulfills these requirements a decoherence-free subspace (DFS).
The above formulation of DFSs in terms of a larger closed system is exact. It is extremely useful for finding DFSs, providing often the most direct route via simple examination of the system components of the interaction Hamiltonian. In practical situations, however, the closed-system formulation of DFSs is often too strict. This is because the closed-system formulation incorporates the possibility that information which is put into the bath will back-react on the system and cause a recurrence. Such interactions will always occur in the closed-system formulation (due to the the Hamiltonian being Hermitian). However, in many practical situations the likelihood of such an event is extremely small. Thus, for example, an excited atom which is in a “cold” bath will radiate a photon and decohere but the bath will not in turn excite the atom back to its excited state, except via the (extremely long) recurrence time of the emission process. In these situations a more appropriate way to describe the evolution of the system is via a quantum dynamical semigroup master equation . By assuming that (i) the evolution of system density matrix is a one-parameter semigroup, (ii) the system density matrix retains the properties of a density matrix including “complete positivity”, and (iii) the system and bath density matrices are initially decoupled, Lindblad has shown that the most general evolution of the system density matrix $`\rho _S(t)`$ is governed by the master equation
$`{\displaystyle \frac{d\rho _S(t)}{dt}}`$ $`=`$ $`i[𝐇_S,\rho _S(t)]+𝙻_D[\rho _S(t)]`$ (4)
$`𝙻_D[\rho _S(t)]`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{\alpha ,\beta =1}{\overset{M}{}}}a_{\alpha \beta }\left([𝐅_\alpha ,\rho _S(t)𝐅_\beta ^{}]+[𝐅_\alpha \rho _S(t),𝐅_\beta ^{}]\right)`$ (5)
where $`𝐇_S`$ is the system Hamiltonian, the operators $`𝐅_\alpha `$ constitute a basis for the $`M`$-dimensional space of all bounded operators acting on $`_S`$, and $`a_{\alpha \beta }`$ are the elements of a positive semi-definite Hermitian matrix. As above, let $`\stackrel{~}{}_S`$ be a subspace of the system Hilbert space $`_S`$ with a basis $`|\stackrel{~}{ı}`$. The evolution over such a subspace is then unitary iff
$$𝐅_\alpha |\stackrel{~}{ı}=c_\alpha |\stackrel{~}{ı},c_\alpha \text{ }\mathrm{C}$$
(6)
for all $`|\stackrel{~}{ı}`$ and for all $`𝐅_\alpha `$. While this condition appears to be identical to Eq (3), there is an important difference between the $`𝐒_\alpha `$’s and the $`𝐅_\alpha `$’s which makes these two decoherence-freeness conditions different. In the Hamiltonian formulation of DFSs, the Hamiltonian is Hermitian. Thus the expansion for the interaction Hamiltonian Eq. (2) can always be written such that the $`𝐒_\alpha `$ are also Hermitian. On the other hand, the $`𝐅_\alpha `$’s in the master equation, Eq. (5), need only be bounded operators acting on $`_S`$ and thus the $`𝐅_\alpha `$’s need not be Hermitian. Because of this difference, Eq. (6) allows for a broader range of subspaces than Eq. (3). For example consider the situation where there are only two nonzero terms in a master equation for a two-level system, corresponding to $`𝐅_1=\sigma _{}`$ and $`𝐅_2=\sigma _z`$ where $`\sigma _{}=|01|`$ and $`\sigma _z=|00||11|`$ (e.g., cooling with phase damping). In this case there is a DFS corresponding to the single state $`|0`$. In the Hamiltonian formulation, inclusion of $`𝐒_1=\sigma _{}`$ in the interaction Hamiltonian expansion Eq. (2) would necessitate a second term in the Hamiltonian with $`𝐒_2=\sigma _{}^{}`$, along with the $`𝐒_z=\sigma _z`$ as above. For this set of operators, however, Eq. (3) allows for no DFS.
### B Decoherence-Free Subsystems
If one desires to encode quantum information over a subspace and requires that this information remains decoherence-free, then Eqs. (3),(6) provide necessary and sufficient conditions for the existence of such DFSs. The notion of a subspace which remains decoherence-free throughout the evolution of a system is not, however, the most general method for providing decoherence-free encoding of information in a quantum system. Recently, Knill, Laflamme, and Viola have discovered a method for decoherence-free coding into subsystems instead of into subspaces.
Decoherence-free subsystems are most easily presented in the Hamiltonian formulation of decoherence. Let $`𝒜`$ denote the associative algebra formed by the system Hamiltonian $`𝐇_S`$ and the system components of the interaction Hamiltonian, the $`𝐒_\alpha `$’s. To simplify our discussion we will assume that the system Hamiltonian vanishes. (It is easy to incorporate the system Hamiltonian into the $`𝐒_\alpha `$’s when one desires that the system evolution preserves the decoherence-free subsystem.) We also assume that the identity operator is included as $`𝐒_0=𝐈_S`$ and $`𝐁_0=𝐈_B`$. This will have no observable consequence but allows for the use of an important representation theorem. $`𝒜`$ consists of linear combinations of products of the $`𝐒_\alpha `$’s. Because the Hamiltonian is Hermitian the $`𝐒_\alpha `$’s must be closed under Hermitian conjugation: $`𝒜`$ is a $``$-closed operator algebra. A basic theorem of such operator algebras which include the identity operator states that, in general, $`𝒜`$ will be a reducible subalgebra of the full algebra of operators on $`_S`$ . This means that the algebra is isomorphic to a direct sum of $`d_J\times d_J`$ complex matrix algebras, each with multiplicity $`n_J`$:
$$𝒜\underset{J𝒥}{}𝐈_{n_J}(d_J,\text{ }\mathrm{C}).$$
(7)
Here $`𝒥`$ is a finite set labeling the irreducible components of $`𝒜`$, and $`(d_J,\text{ }\mathrm{C})`$ denotes a $`d_J\times d_J`$ complex matrix algebra. It is also useful at this point to introduce the commutant $`𝒜^{}`$ of $`𝒜`$. This is the set of operators which commutes with the algebra $`𝒜`$, $`𝒜^{}=\{𝐗:[𝐗,𝐀]=0,𝐀𝒜\}`$. They also form a $``$-closed algebra, which is reducible to
$$𝒜^{}=\underset{J𝒥}{}(n_J,\text{ }\mathrm{C})𝐈_{d_J}$$
(8)
over the same basis as $`𝒜`$ in Eq. (7).
The structure implied by Eq. (7) is illustrated schematically as follows, for some system operator $`𝐒_\alpha `$:
𝐒α=[
=J1
=J2
J=|𝒥|]subscript𝐒𝛼delimited-[]fragments
=J1
fragments
=J2
missing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionmissing-subexpressionfragmentsJ|J|{\bf S}_{\alpha}=\left[\begin{tabular}[]{cccc}\cline{1-1}\cr\vrule\lx@intercol\hfil$J=1$\hfil\lx@intercol &\vrule\lx@intercol\hfil\hfil\lx@intercol &&\\
\cline{1-2}\cr&\vrule\lx@intercol\hfil$J=2$\hfil\lx@intercol &\vrule\lx@intercol\hfil\hfil\lx@intercol &\\
\cline{2-2}\cr&&$\ddots$&\\
\cline{4-4}\cr&&&\vrule\lx@intercol\hfil$J=|{\cal J}|$\hfil\lx@intercol\vrule\lx@intercol \\
\cline{4-4}\cr\end{tabular}\right] (9)
where a typical block with given $`J`$ has the structure:
$$J\text{ given}\text{}\left[\begin{array}{ccccccccccc}& M_\alpha & & & & & & & & & \lambda =0\\ & & & & & & \mu & & & & \\ & & & & & & 0& & & & \\ & & & & M_\alpha & & \mathrm{}& & & & \lambda =1\\ & & & & & & d_J1& & & & \\ & & \mu ^{}:& 0& \mathrm{}& d_J1& \mathrm{}& & & & \\ & & & & & & & & M_\alpha & & \lambda =n_J1\end{array}\right]$$
(10)
Associated with this decomposition of the algebra $`𝒜`$ is the decomposition over the system Hilbert space:
$$_S=\underset{J𝒥}{}\text{ }\mathrm{C}^{n_J}\text{ }\mathrm{C}^{d_J}.$$
(11)
Decoherence-free subsystems are defined as the situation in which information is encoded in a single subsystem space $`\text{ }\mathrm{C}^{n_J}`$ of Eq. (11) (thus the dimension of the decoherence-free subsystem is $`n_J`$). The decomposition in Eq. (7) reveals that information encoded in such a subsystem will always be affected as identity on the subsystem space $`\text{ }\mathrm{C}^{n_J}`$, and thus this information will not decohere. It should be noted that the tensor product nature which gives rise to the name subsystem in Eq. (7) is a tensor product over a direct sum, and therefore will not in general correspond to the natural tensor product of qubits. Further, it should be noted that the subsystem nature of the decoherence implies that the information should be encoded in a separable way. Over the tensor structure of Eq. (11) the density matrix should split into two valid density matrices: $`\rho _S(0)=\rho \gamma `$ where $`\rho `$ is the decoherence-free subsystem and $`\gamma `$ is the corresponding component of the density matrix which does decohere. Finally it should be pointed out that not all of the subsystems in the different irreducible representations can be simultaneously used: (phase) decoherence will occur between the different irreducible components of the Hilbert space labeled by $`J𝒥`$. For this reason, from now on we restrict our attention to the subspace defined by a given $`J`$.
Decoherence-free subspaces are now easily connected to decoherence-free subsystems. Decoherence-free subspaces correspond to decoherence-free subsystems possessing one-dimensional irreducible matrix algebras: $`(1,\text{ }\mathrm{C})`$. The multiplicity of these one-dimensional irreducible algebras is the dimension of the decoherence-free subspaces. In fact it is easy to see how the decoherence-free subsystems arise out of a non-commuting generalization of the decoherence-free subspace conditions. Let $`\{|\lambda _\mu \}`$, $`1\lambda n_J`$ and $`1\mu d_J`$, denote a subspace of $`_S`$ with given $`J`$. Then the condition for the existence of an irreducible decomposition as in Eq. (7) is
$$𝐒_\alpha |\lambda _\mu =\underset{\mu ^{}=1}{\overset{d_J}{}}M_{\mu \mu ^{},\alpha }|\lambda _\mu ^{},$$
(12)
for all $`𝐒_\alpha `$, $`\lambda `$ and $`\mu `$. Notice that $`M_{\mu \mu ^{},\alpha }`$ is not dependent on $`\lambda `$, in the same way that $`c_\alpha `$ in Eq. (3) is not the same for all $`|\stackrel{~}{ı}`$ (there $`\mu =1`$ and fixed). Thus for a fixed $`\lambda `$, the subspace spanned by $`|\lambda _\mu `$ is acted upon in some nontrivial way. However, because $`M_{\mu \mu ^{},\alpha }`$ is not dependent on $`\lambda `$, each subspace defined by a fixed $`\mu `$ and running over $`\lambda `$ is acted upon in an identical manner by the decoherence process.
At this point it should be noted that the generalization of the Lindblad master equation Eq. (5) with a decoherence-free subspace to the corresponding master equation for a decoherence-free system is not trivial. This is because, as above, the $`𝐅_\alpha `$ operators in Eq. (5) are (for all practical purposes) not required to be closed under conjugation. The representation theorem Eq. (7) is hence not directly applicable. We will show, however, that the master equation analog of Eq. (12)
$$𝐅_\alpha |\lambda _\mu =\underset{\mu ^{}=1}{\overset{d_J}{}}M_{\mu \mu ^{},\alpha }|\lambda _\mu ^{}$$
(13)
provides a necessary and sufficient condition for the preservation of decoherence-free subsystems.
As above, we consider a subspace of the system Hilbert space spanned by $`|\lambda _\mu `$, with $`1\lambda n_J`$ and $`1\mu d_J`$. Our notation will be significantly simpler if we explicitly write out the formal tensor product over this subspace: $`|\lambda _\mu =|\lambda |\mu `$. In the subsystem notation, we claim that the decoherence-free subsystem condition is
$$𝐅_\alpha |\lambda |\mu =|\lambda 𝐌_\alpha |\mu .$$
(14)
A proper decomposition of the system Hilbert space requires, as noted above, that the system density matrix is a tensor product of two valid (Hermitian, positive) density matrices:
$$\rho _S(0)=\underset{\lambda \lambda ^{},\mu \mu ^{}}{}\rho _{\lambda \lambda ^{}}(0)\gamma _{\mu \mu ^{}}(0)|\lambda _\mu \lambda _\mu ^{}^{}|=\rho (0)\gamma (0),$$
(15)
where $`\rho (0)`$ contains the information which will remain decoherence-free, and $`\gamma (0)`$ is an arbitrary but valid density matrix.
In general the operators $`𝐅_\alpha `$ will not be decomposable as a single tensor product corresponding to $`\rho (0)\gamma (0)`$. Rather, they will be a sum over such tensor products, corresponding to an expansion over an operator basis: $`𝐅_\alpha =_p𝐍_\alpha ^p𝐌_\alpha ^p`$. The decohering generator of evolution (5) thus becomes
$`𝙻_D[\rho _S(0)]`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{\alpha \beta }{}}a_{\alpha \beta }{\displaystyle \underset{pq}{}}(2𝐍_\alpha ^p\rho (0)𝐍_\beta ^q𝐌_\alpha ^p\gamma (0)𝐌_\beta ^q𝐍_\beta ^q𝐍_\alpha ^p\rho (0)𝐌_\beta ^q𝐌_\alpha ^p\gamma (0)`$ (16)
$``$ $`\rho (0)𝐍_\beta ^q𝐍_\alpha ^p\gamma (0)𝐌_\beta ^q𝐌_\alpha ^p).`$ (17)
Tracing over the $`\gamma `$ component, and using the cyclic nature of the trace allows one to factor out a common $`m_{\alpha \beta }^{pq}\mathrm{Tr}_\gamma (𝐌_\alpha ^p\gamma (0)𝐌_\beta ^q)`$, yielding:
$`\mathrm{Tr}_\gamma \{𝙻_D[\rho _S(0)\}={\displaystyle \frac{1}{2}}{\displaystyle \underset{\alpha \beta ,pq}{}}a_{\alpha \beta }m_{\alpha \beta }^{pq}(2𝐍_\alpha ^p\rho (0)𝐍_\beta ^q𝐍_\beta ^q𝐍_\alpha ^p\rho (0)\rho (0)𝐍_\beta ^q𝐍_\alpha ^p)].`$
The evolution of the $`\rho `$ component of the density matrix thus satisfies the standard master equation (5), for which it is known that the evolution is decoherence-free if and only if
$$𝐍_\alpha ^q|\lambda =c_{\alpha ,q}|\lambda \alpha .$$
(18)
This implies that the necessary and sufficient condition for a decoherence-free subsystem is
$$𝐅_\alpha =\underset{q}{}c_{\alpha ,q}𝐈𝐌_\alpha ^q=𝐈\underset{q}{}c_{\alpha ,q}𝐌_\alpha ^q=𝐈𝐌_\alpha ,$$
(19)
which is the claimed generalization of the Hamiltonian condition of decoherence-free subsystems, Eq. (13).
We will use the acronym DFS to denote both decoherence-free subsystems and their restriction, decoherence-free subspaces, whenever no confusion can arise. When we refer to DF subspaces we will be specifically referring to the one-dimensional version of the DF subsystems.
## III The Stabilizer Formalism and Error-Correction
In the theory of quantum error correcting codes (QECCs) it proved fruitful to study properties of a code by considering its stabilizer $`𝒮`$. This is the group formed by those system operators which leave the codewords unchanged, i.e., they ”stabilize” the code. Properties of stabilizer codes and the theory of quantum computation on these stabilizer codes have been developed in . In the framework of QECCs, the stabilizer allows on the one hand to identify the errors the code can detect and correct. On the other hand it also permits one to find a set of universal, fault-tolerant gates by analyzing the centralizer of $`𝒮`$, defined as the set of operations that commute with all elements in $`𝒮`$ (equal to the normalizer - the set of operations that preserve $`𝒮`$ under conjugation – in the case of the Pauli group). In the context of QECCs, the stabilizer $`𝒮`$ is restricted to elements in the Pauli-group, i.e., the group of tensor products of $`𝐈,𝐗,𝐘,𝐙`$, and is a finite abelian group.
The extension of stabilizer theory yields much insight into DFSs. We do this here by defining a non-abelian, and in certain cases infinite stabilizer group. The observation that DFSs are highly degenerate QECCs will appear naturally from this formalism. Such a generalized stabilizer has already been defined in previous work dealing with decoherence-free subspaces , and its normalizer shown there to lead to identification of local gates for universal computation. A key consequence of this approach was the observation that the resulting gates do not take the system out of the DFS during the entire switching time of the gate.
We now review and extend the results in to analyze the error detection and correction properties of DFSs and QECCs. We shall incorporate DFSs and QECCs into a unified framework, similarly to the representation-theoretic approach of . The question of performing quantum computation on a specific DFS will be addressed in the next section.
### A The Stabilizer - General Theory
An operator $`𝐒`$ is said to stabilize a code $`𝒞`$ if
$$|\mathrm{\Psi }𝒞\mathrm{iff}𝐒|\mathrm{\Psi }=|\mathrm{\Psi }𝐒𝒮.$$
(20)
The set of operators $`\{𝐒\}`$ form a group $`𝒮`$, known as the stabilizer of the code . Clearly, $`𝒮`$ is closed under multiplication. In the theory of QECC the stabilizers that have been studied are subgroups of the Pauli-group (tensor products of $`𝐈,𝐗,𝐘,𝐙`$ ). Since any two elements of the Pauli group either commute or anticommute, a subgroup (and in particular the stabilizer), in this case is always abelian . The code is thus the common eigenspace of the stabilizer elements with eigenvalue $`1`$.
In general an error-process can be described by the Kraus operator-sum formalism : $`\rho _\mu 𝐀_\mu \rho 𝐀_\mu ^{}`$. The Kraus-operators $`𝐀_\mu `$ can be expanded in a basis $`\{𝐄_\alpha \}`$ of “errors”. The standard QECC error-model assumes that errors affect single qubits independently. Therefore the theory of QECCs has focused on searching for codes that make quantum information robust against 1, 2,… or more erroneous qubits. Detection and correction procedures must then be implemented at a rate higher than the intrinsic error rate. In the QECC error-model, the independent errors are spanned by single-qubit elements ($`𝐗,𝐘,𝐙`$). An analysis of the error-correction properties can then be restricted to correction of combinations of these basic errors (which are also members of the Pauli group) acting on a certain number of qubits simultaneously.
The distance $`d`$ of a QECC is the number of single-qubit errors that have to occur in order to transform one codeword in $`𝒞`$ to another codeword in $`𝒞`$. An error $`E`$ is detectable if it takes a codeword to a subspace of the Hilbert space that is orthogonal to the space spanned by $`𝒞`$ (this can be observed by a non-perturbing orthogonal von Neuman measurement). A distance $`d`$ code can detect up to $`d1`$ errors. In order to be able to correct an error on a certain codeword the error (up to a degenerate action of different errors) also needs to be identified, so that it can be undone. Hence errors on different codewords have to take the codewords to different orthogonal subspaces. The above translates to the QECC-condition :
A QECC $`𝒞`$ can correct errors $`=\{𝐄_\alpha \}`$ if and only if
$$\mathrm{\Psi }_j|𝐄_\beta ^{}𝐄_\alpha |\mathrm{\Psi }_i=c_{\alpha \beta }\delta _{ij}𝐄_\alpha ,𝐄_\beta .$$
(21)
The stabilizer of a QECC allows identification of the errors which the code can detect and correct . Two types of errors can be dealt with by stabilizer codes: (i) errors $`𝐄_\alpha ^{}𝐄_\beta 𝐈`$ that anticommute with an $`𝐒𝒮`$, and (ii) errors that are part of the stabilizer ($`𝐄_\alpha 𝒮`$ ). It is straightforward to see that both (i) and (ii) imply the QECC condition Eq. (21). For case (i), if $`𝐄_i^{}𝐄_j𝐒=\mathrm{𝐒𝐄}_i^{}𝐄_j`$, then $`\mathrm{\Psi }_j|𝐄_\beta ^{}𝐄_\alpha |\mathrm{\Psi }_i=\mathrm{\Psi }_j|𝐄_\beta ^{}𝐄_\alpha 𝐒|\mathrm{\Psi }_i=\mathrm{\Psi }_j|\mathrm{𝐒𝐄}_\beta ^{}𝐄_\alpha |\mathrm{\Psi }_i=\mathrm{\Psi }_j|𝐄_\beta ^{}𝐄_\alpha |\mathrm{\Psi }_i`$. Hence $`\mathrm{\Psi }_j|𝐄_\beta ^{}𝐄_\alpha |\mathrm{\Psi }_i=0`$ and $`c_{\alpha \beta }=\delta _{\alpha \beta }`$. Errors of type (ii), $`𝐄_\alpha 𝒮`$, leave the codewords unchanged and therefore trivially lead to Eq. (21). The first class, (i), are errors that require active correction. The second class, (ii), are “degenerate” errors that do not affect the code at all. QECCs can be regarded as (passive) DFSs for the errors in their stabilizer . Conversely, being passive, highly degenerate codes , DFSs can be viewed as a class of stabilizer codes that protect against type (ii) errors \[i.e., where the $`𝐀_\mu `$ are linear combinations of elements generated (under multiplication) by the stabilizer\], and against the (usually small) set of errors that anticommute with the DFS stabilizer. The stabilizer thus provides a unified tool to identify the errors that a given code can deal with, as a DFS and as a QECC. An analysis of the properties of DFSs with a stabilizer in the Pauli-group has been carried out in .
### B DFS-Stabilizer
Most of the DFSs stemming from physical error-models will not have a stabilizer in the Pauli-group, i.e., they are non-additive codes. The stabilizer may even be infinite. In particular, the codes obtained from a noise model where errors arise from a symmetric coupling of the system to the bath and that form the focus of this paper, are of this type.
As discussed in the previous section, a DFS is completely specified by the condition:
$$𝐒_\alpha |\mu |\lambda =|\mu 𝐌_\alpha |\lambda ,$$
(22)
arising from the splitting of the algebra generated by the $`𝐒_\alpha `$’s: $`𝒜=_{J𝒥}𝐈_{n_J}(d_J,\text{ }\mathrm{C})`$. This splitting of the algebra has allowed both DFSs and QECCs to be put into a similar framework . We will now show that the DFS condition on the algebra $`𝒜`$ generated by the $`𝐒_\alpha `$ can be converted into a stabilizer condition on the complex Lie algebra generated by the $`𝐒_\alpha `$’s. We define the continuous DFS-stabilizer $`𝐃(\stackrel{}{v})`$ as
$$𝐃(v_1,v_2,\mathrm{}v_N)=\mathrm{exp}\left[\underset{\alpha }{}v_\alpha \left(𝐒_\alpha 𝐈𝐌_\alpha \right)\right],v_\alpha \text{ }\mathrm{C}.$$
(23)
Clearly, if the DFS condition Eq. (22) is fulfilled for a set of states $`|\mu |\lambda `$, then
$$𝐃(\stackrel{}{v})|\mu |\lambda =|\mu |\lambda v_\alpha \stackrel{}{v}.$$
(24)
Thus the DFS condition implies that the $`𝐃(\stackrel{}{v})`$ stabilize the DFS. Further if Eq. (24) holds then in particular it must hold for a $`\stackrel{}{v}`$ which has only one non-vanishing component $`v_\beta `$. Thus Eq. (24) implies that $`𝐃(0,\mathrm{},0,v_\beta ,0\mathrm{}0)|\mu |\lambda =|\mu |\lambda `$. Recalling that $`\mathrm{exp}[]`$ is a one-to-one mapping from a neighborhood of the zero matrix to a neighborhood of the identity matrix, it follows that there must exist a small enough $`v_\alpha `$ such that Eq. (24) implies the DFS condition Eq. (22). Thus we see that we can convert the DFS condition into a condition on the stabilizer of the complex Lie algebra generated by the $`𝐒_\alpha 𝐈𝐌_\alpha `$’s:
$$|\mathrm{\Psi }\mathrm{DFS}\mathrm{iff}𝐃(\stackrel{}{v})|\mathrm{\Psi }=|\mathrm{\Psi }\stackrel{}{v}\text{ }\mathrm{C}^N$$
(25)
In some cases we will be able to pick a finite subgroup from elements of $`𝐃(\stackrel{}{v})`$ which constitutes a stabilizer. We will mention these instances in the following sections. However, apart from the conceptual framework, our main motivation to introduce the stabilizer for a DFS is to be able to analyze the errors which a DFS (i) detects/corrects (as a QECC), and (ii) those which it avoids (passive error correction). The continuous stabilizer provided in Eq. (23) will be sufficient to study these errors.
As in the previous paragraph, errors $`𝐄_\alpha `$ (i) that anticommute with an element in the stabilizer will take codewords to subspaces that are orthogonal to the code. These errors will be detectable (and correctable if $`𝐄_\beta ^{}𝐄_\alpha `$ anticommutes with a stabilizer element) . In order to identify the QECC-properties of a DFS, it will be convenient to look for elements of the Pauli group among the $`𝐃(\stackrel{}{v})`$.
A code $`𝒞`$ with stabilizer $`𝐃(\stackrel{}{v})`$ will avoid errors of type (ii) in its stabilizer in the sense that, if all of the Kraus operators of a given decoherence process can be expanded over stabilizer elements $`𝐀_i(t)=_{\text{ }\mathrm{C}^N}e_i(\stackrel{}{v},t)𝐃(\stackrel{}{v})𝑑\stackrel{}{v}`$, then
$$\rho (t)=\underset{i}{}𝐀_i(t)\rho (0)𝐀_i^{}(t)=\underset{i}{}\left|_{\text{ }\mathrm{C}^N}e_i(\stackrel{}{v},t)𝑑\stackrel{}{v}\right|^2\rho (0).$$
(26)
The normalization condition $`_i𝐀_i(t)^{}𝐀_i(t)=𝐈`$ then implies that $`_i\left|\underset{\text{ }\mathrm{C}^N}{}e_i(\stackrel{}{v},t)𝑑\stackrel{}{v}\right|^2=1`$. Consequently, as expected, the DFS does not evolve. Hence we see that the stabilizer provides an efficient method for identifying the errors which a code avoids. In later sections we analyze the concrete form of the stabilizer Eq. (23) for the error models studied in this paper.
## IV The Commutant and Universal Quantum Computation on a DFS
A DFS is a promising way to store quantum information in a robust fashion . From the perspective of quantum computation however, it is even more important to be able to controllably transform states in a DFS, if it is to be truly useful for quantum information processing. More specifically, to perform quantum algorithms on a DFS one has to be able to perform universal quantum computation using decoherence-free states. The notion of universal computation is the following: with a restricted set of operations or interactions at hand, one wishes to implement any unitary transformation on the given Hilbert space, to an arbitrary degree of accuracy. From a physical implementation perspective it seems clear that the operations used (gates) should be limited to at most two-body interactions. In particular we wish to identify a finite set of such gates that is universal on a DFS.
Since we do not wish to implement active QECC, we impose a very stringent requirement on the operations we shall allow for computation using DFSs: we do not allow gates that ever take the decoherence free states outside the DFS, where the states would decohere under the noise- process considered.We shall lift this requirement in Sec. VIII. As a first step towards this goal we thus need to be able to identify the physical operations which perform transformations entirely within the DFS.
### A Operations that Preserve the DFS
There are essentially two equivalent approaches to identify the “encoded” operations that preserve a DFS. One is via the normalizer of the stabilizer of a code ; the second is via the commutant of the $``$-closed algebra generated by the error operators . Both will be briefly reviewed here.
Computation on a stabilizer DFS: The stabilizer formalism is very useful for identifying allowed gates that take codewords to codewords . An operation $`𝐔`$ keeps code-words $`|\mathrm{\Psi }`$ inside the code-space, if and only if the transformed state $`𝐔|\mathrm{\Psi }`$ is an element of the code $`𝒞`$. Thus, using the stabilizer condition (20) for codes with stabilizer $`𝒮`$ and $`𝒞=\{|\mathrm{\Psi }:𝐒|\mathrm{\Psi }=|\mathrm{\Psi }𝐒𝒮\}`$, we have
$$𝐔|\mathrm{\Psi }𝒞\mathrm{iff}\mathrm{𝐒𝐔}|\mathrm{\Psi }=𝐔|\mathrm{\Psi }𝐒𝒮.$$
(27)
This implies $`𝐔^\mathrm{𝟏}\mathrm{𝐒𝐔}|\mathrm{\Psi }=|\mathrm{\Psi }`$ and so $`𝐔^\mathrm{𝟏}\mathrm{𝐒𝐔}𝒮`$: Allowed operations $`𝐔`$ transform stabilizer elements $`𝐒`$ by conjugation into stabilizer elements; $`𝐔`$ is in the normalizer of $`𝒮`$ (if $`𝒮`$ is a group). If we restrict the allowed operations to gates in the Pauli-group (as is done in ), then the allowed gates $`𝐔`$ will fix the stabilizer pointwise (element by element). In the case of DFS with a continuous stabilizer $`𝐃(\stackrel{}{v})`$, the above translates to the following condition
$$\mathrm{𝐔𝐃}(\stackrel{}{v})𝐔^{}=𝐃(\stackrel{}{v}^{}(\stackrel{}{v})),$$
(28)
together with the requirement that $`𝐃(\stackrel{}{v}^{}(\stackrel{}{v}))`$ must cover $`𝒮`$. To satisfy the covering condition, it is sufficient to have $`\stackrel{}{v}^{}(\stackrel{}{v})`$ be a one-to-one mapping.
Eq. (28), derived by generalizing concepts from the theory of stabilizers in the Pauli group, is a condition that allows one to identify gates $`𝐔`$ that transform codewords to codewords. In a physical implementation these gates will be realized by turning on Hamiltonians $`𝐇`$ between physical qubits for a certain time $`t`$: $`𝐔(t)=e^{it𝐇}`$. So far we only required that the action of the gate preserve the subspace at the conclusion of the gate operation, but not that the subspace be preserved throughout the entire duration of the gate operation. The stabilizer approach allows us to further identify the more restrictive set of Hamiltonians that keep the states within the DFS throughout the entire switching time of the gate. As a result, in the limit of ideal gates, the entire system is free from noise at all times. This is different from QECC, since there errors continuously take the codewords outside of the code-space , and hence error correction needs to be applied frequently even in the limit of perfect gate operations. Imperfections in gate operations can be dealt with in the DFS approach by concatenation with a QECC , as shown explicitly for the exchange interaction in .
By rewriting condition (28) as $`𝐔(t)𝐃(\stackrel{}{v})=𝐃(\stackrel{}{v}^{}(\stackrel{}{v},t))𝐔(t)`$, taking the derivative with respect to $`t`$ and evaluating at $`t=0`$ we obtain $`\mathrm{𝐇𝐃}(\stackrel{}{v})=𝐃(\stackrel{}{v}^{}(\stackrel{}{v},0))𝐇+i\frac{𝐃}{\stackrel{}{v}^{}}\frac{d\stackrel{}{v}^{}}{dt}|_{t=0}`$, so that:
Theorem 1 A sufficient condition for the generating Hamiltonian to keep a state at all times entirely within the DFS is $`\mathrm{𝐇𝐃}(\stackrel{}{v})=𝐃(\stackrel{}{v}^{}(\stackrel{}{v}))𝐇`$ where $`\stackrel{}{v}^{}(\stackrel{}{v})`$ is one-to-one and time-independent.
For most applications we will only need gates that commute with all stabilizer elements. The condition for the generating Hamiltonian then simplifies to $`\mathrm{𝐇𝐃}(\stackrel{}{v})=𝐃(\stackrel{}{v})𝐇`$.
Computation on irreducible subspaces: We can derive conditions to identify allowed gates on a DFS by using the representation theoretic approach developed in , and section II. Recall that the decomposition of the algebra $`𝒜_{J𝒥}𝐈_{n_J}(d_J,\text{ }\mathrm{C})`$ generated by the errors $`\{𝐒_\alpha \}`$ induces a splitting of the Hilbert space $`_S=_{J𝒥}\text{ }\mathrm{C}^{n_J}\text{ }\mathrm{C}^{d_J}`$ into subspaces possessing a tensor product structure suitable to isolate decoherence-free subsystems. The set of operators in the commutant of $`𝒜`$, $`𝒜^{}=\{𝐗:[𝐗,𝐀]=0,𝐀𝒜\}=𝒜^{}=_{J𝒥}(n_J,\text{ }\mathrm{C})𝐈_{d_J}`$, obviously generate transformations that affect the codespace only. In particular, they take states in a DFS to other states in that same DFS. $`𝒜^{}`$ is generated by operators which commute with the $`𝐒_\alpha `$. Again, our goal is to find gates that act within a DFS during the entire switching time, and to this end we need to identify Hermitian operators H in $`𝒜^{}`$ to generate an evolution $`𝐔(t)=\mathrm{exp}[it𝐇]`$ on the DFS.
Theorem 2— A sufficient condition for a Hamiltonian $`𝐇`$ to generate dynamics $`𝐔(t)=\mathrm{exp}[it𝐇]`$ which preserves a DFS is that $`𝐇`$ be in the commutant of the algebra $`𝒜`$.
However, because we can only use one particular DFS (corresponding to a specific $`K𝒥`$) to store quantum information (the coherences between superpositions of different DFSs are not protected), the operators which commute with the $`𝐒_\alpha `$’s are not the only operators which perform non-trivial operations on a specific DFS. The operations in $`𝒜^{}`$ preserve all DFSs in parallel. However, if we restrict our system to only one such DFS, we do not need any constraints on the evolution of the other subspaces. It is then possible to construct a necessary and sufficient condition for a Hamiltonian by modifying the commutant to:
$$𝒯\left((n_K,\text{ }\mathrm{C})𝐈_{d_K}\right)(dd_Kn_K,\text{ }\mathrm{C})$$
(29)
where $`d_K=\mathrm{dim}(_S)`$ and just leaves the specific DFS ($`K`$) invariant.
Theorem 3— A necessary and sufficient condition for a Hamiltonian $`𝐇`$ to generate dynamics which preserves a DFS corresponding to the irreducible representation $`K𝒥`$, is $`𝐇𝒯`$.
We will use both the stabilizer and the commutant approaches, to find a set of universal gates for decoherence processes of physical relevance. In the cases discussed in this paper, any one of the two approaches is clearly sufficient and we do not need all theorems in full generality. However we provide here a general framework and the tools required to analyze DFS and QECC stemming from any error model.
Finally we should point out again that from a practical perspective, it is crucial to look for the Hermitian operations which perform nontrivial operations on the DFS and which correspond to only one or two-body physical interactions. Without this requirement, it is clear that one can always construct a set of Hamiltonians (satisfying the conditions of Theorem 2) which span the allowed operations on a DFS. A primary goal of this paper is therefore to construct such one and two-body Hamiltonians for specific decoherence mechanisms, in order to achieve true universal computation on the corresponding DFSs.
### B Universality and Composition of Allowed Operations
Using the tools developed in the previous subsection, we can now find local one-and-two qubit gates that represent encoded operations on DFSs. However, in general, a discrete set of gates applied in alternation is not sufficient to generate a universal set of gates. Nor is it sufficient to obtain every encoded unitary operation exactly. Furthermore, for analysis of the complexity of computations performed with a given universal set of gates, it is essential to keep under control the number of operations needed to achieve a certain gate within a desired accuracy. In the theory of universality (e.g., ) the composition laws of operations have been analyzed extensively. We will review the essential results relevant for our purposes here.
Let us assume that we have a set of (up to two-body) Hamiltonians $`𝖧=\{𝐇_i:i=1,\mathrm{},M\}`$ that take DFS states to DFS states. We will construct gates using the following composition laws:
1. Arbitrary phases: Any interaction can be switched on for an arbitrary time. Thus any gate of the form $`𝐔(t)=\mathrm{exp}\left(it𝐇_i\right)`$ can be implemented.
2. Trotter formula: Gates performing sums of Hamiltonians are implemented by using the short-time approximation to the Trotter formula $`\mathrm{exp}\left[i(t_1𝐇_i+t_2𝐇_j)\right]`$$`=lim_n\mathrm{}\left[\mathrm{exp}\left(i\frac{t_1}{n}𝐇_i\right)\mathrm{exp}\left(i\frac{t_2}{n}𝐇_j\right)\right]^n`$:
$$e^{i(t_1𝐇_i+t_2𝐇_j)/n}=e^{it_1𝐇_i/n}e^{it_2𝐇_j/n}+O(\frac{1}{n^2}).$$
(30)
This is achieved by quickly turning on and off the two interactions $`𝐇_i,𝐇_j`$ with appropriate ratios of duration times. An alternative, direct, way of implementing this gate is to switch on the two interactions simultaneously for the appropriate time-intervals.
3. Commutator: It is possible to implement the commutator of operations that are already achievable. This is a consequence of the Lie product formula
$`\mathrm{exp}[𝐇_i,𝐇_j]=\underset{n\mathrm{}}{lim}\left[\mathrm{exp}\left(i𝐇_i/\sqrt{n}\right)\mathrm{exp}\left(i𝐇_j/\sqrt{n}\right)\mathrm{exp}\left(i𝐇_i/\sqrt{n}\right)\mathrm{exp}\left(i𝐇_j/\sqrt{n}\right)\right]^n,`$
which has the short-time approximation
$$e^{t[𝐇_i,𝐇_j]/n}=e^{it𝐇_i/\sqrt{n}}e^{it𝐇_j/\sqrt{n}}e^{it𝐇_i/\sqrt{n}}e^{it𝐇_j/\sqrt{n}}+O(\frac{1}{n\sqrt{n}}).$$
(31)
Again, the gate $`e^{it(i[𝐇_i,𝐇_j])}`$ can be implemented to high precision by alternately switching on and off the appropriate two interactions with a specific duration ratio.Note that in order to implement $`e^{it𝐀}`$ we would use $`e^{i\vartheta 𝐀}=I`$ and implement $`e^{i(\vartheta t)𝐀}`$ instead. This depends on $`𝐀`$ having rationally related eigenvalues, which will always be the case for the Hamiltonians of interest to us.
4. Conjugation by unitary evolution: Another useful action in constructing universal sets of gates comes from the observation that if a specific gate $`𝐔`$ and its inverse $`𝐔^{}`$ can be implemented, then any Hamiltonian $`𝐇`$ which can be implemented can be modified by performing $`𝐔`$ before, and $`𝐔^{}`$ after the gate $`\mathrm{exp}(it𝐇)`$. This gives rise to the transformed Hamiltonian
$$𝐔\mathrm{exp}(it𝐇)𝐔^{}=\mathrm{exp}(it\mathrm{𝐔𝐇𝐔}^{})=\mathrm{exp}(it𝐇_{\mathrm{eff}}).$$
(32)
Note that the laws (1-3) correspond to closing the set of allowed Hamiltonians as a Lie-algebra (scalar multiplication, addition and Lie-commutators can be obtained out of the given Hamiltonians).
If (a subset of) the composition laws (1-4) acting on the set $`𝖧`$ give rise to a set of gates that is dense in the group $`SU(d_K)`$ (via successive application of these gates), where $`d_K`$ is the dimension of the DFS, then we shall refer to $`𝖧`$ as a universal set of generators. Equivalently, this means that $`𝖧`$ generates the Lie-algebra $`su(d_K)`$ (traceless matrices) via scalar multiplication, addition, Lie-commutators, and conjugation by unitaries. The generators of this algebra can be obtained from $`𝖧`$ by these operations.
For all practical applications and implementations of algorithms, we will only be interested to approximate a certain gate sequence with a given accuracy. Note that the composition laws (2) and (3) use only repeated applications of (1) in order to approximate a certain gate. We can replace the requirement to perform an arbitrary phase, (1), by noting that $`e^{i\stackrel{~}{t}𝐇_i}`$ is dense in the group $`\{\mathrm{exp}(it𝐇_i):t[0,2\pi )\}`$ if $`\stackrel{~}{t}`$ is an irrational multiple of $`\pi `$. Repeated application of that gate can then approximate an arbitrary phase to any desired accuracy. Thus we can in principle restrict our available gates to $`\{\mathrm{exp}(i\stackrel{~}{t}_i𝐇_i)\}`$, together with fixed irrational switching times $`\stackrel{~}{t}_i`$. Repeated application of these gates can then be used to approximate any operation in $`SU(d_K)`$ to arbitrary accuracy.
In order to prove that a set $`𝖧`$ generates a universal set of Hamiltonians, we use the fact that a large group of universal sets have already been identified . It suffices to show that $`𝖧`$ generates one of these sets, in order to prove that $`𝖧`$ is a universal set of generators. We will use the fact that the set of one qubit operations $`SU(2)`$ is generated by any two arbitrary rotations with irrational phase, around two non-parallel axes. Alternatively, if we are given these two rotations with any phase, then an Euler-angle construction can be used to yield any gate in $`SU(2)`$ by application of a small number of rotations (three if the axes are orthogonal). In addition we will use (and prove) a lemma (Enlarging Lemma, Appendix C) that allows extension to $`su(n+1)`$ of a given $`su(n)`$ acting on an $`n`$-dimensional subspace of a Hilbert space of dimension $`n+1`$, with the help of an additional $`su(2)`$.
In order to use this approach to universality, it is crucial to have bounds on the length of the gate sequences approximating a certain gate in terms of the desired accuracy. This is all the more important if one universal set is to be replaced by any other with only polynomial overhead in the number of gates applied, for otherwise the complexity classes would not be robust under the exchange of one set for another. The whole notion of universality would then by questionable. The following key theorem proved independently by Solovay and Kitaev (see ) establishes the equivalence of universal sets, and provides bounds on the length of gate sequences for a desired accuracy of approximation. In order to quantify the accuracy of an approximation, we need to define a distance on matrices. Since our matrices act in a space of given (finite) dimension $`d_K`$, any metric is as good as any other. For example, we can use the trace-norm $`d(𝐔,𝐕)=\sqrt{1\frac{1}{d_K}\mathrm{Re}\left[\mathrm{Tr}(𝐔^{}𝐕)\right]}`$ . A matrix $`𝐕`$ is then said to approximate a transformation $`𝐔`$ to accuracy $`ϵ`$ if $`d(𝐔,𝐕)ϵ`$.
Theorem (Solovay-Kitaev) — Given a set of gates that is dense in $`SU(2^k)`$ and closed under Hermitian conjugation, any gate $`𝐔`$ in $`SU(2^k)`$ can be approximated to an accuracy $`ϵ`$ with a sequence of $`\mathrm{poly}\left[\mathrm{log}(1/ϵ)\right]`$ gates from the set.
DFS-Corollary to the Solovay-Kitaev Theorem— Assume that the DFS encodes a $`d_K`$-dimensional system into $`n`$ physical qubits. Given that one can exactly implement the gate set $`\{e^{i\stackrel{~}{t}_i𝐇_𝐢}:𝐇_i𝖧\}`$, \[$`\stackrel{~}{t}_i`$ are (fixed) irrational multiples of $`\pi `$, and $`𝖧`$ is a universal generating set\] it is possible to approximate any gate in $`SU(d_K)`$ (any encoded operation) using $`m=\mathrm{poly}\left[\mathrm{log}(1/ϵ)\right]`$ gates.
Furthermore, if we can only implement the given gates approximately, say to an accuracy $`\delta `$, we will still be able to approximate the target gate: It is known that a sequence of $`m`$ $`\delta `$-imprecise unitary matrices is (in some norm) at most distance $`m\delta `$ far from the desired gate. If a sequence of exactly implemented gates $`𝐔_1,\mathrm{},𝐔_m`$ approximates a target gate $`𝐔`$ up to $`ϵ`$, and instead of $`𝐔_1,\mathrm{},𝐔_m`$, we use gates that are at most some distance $`\delta `$ apart, then the total sequence will be at most $`ϵ+m\delta =ϵ+\mathrm{poly}\left[\mathrm{log}(1/ϵ)\right]\delta `$ apart from $`𝐔`$. If we make sure that $`\delta <ϵ\mathrm{poly}\left[\mathrm{log}(1/ϵ)\right]`$ then the $`\delta `$-faulty sequence will still approximate $`𝐔`$ to a precision $`2ϵ`$.
If we further assume that the physical interaction that we switch on and off is given by the device and is unlikely to change its form, then the imprecision of the gate comes entirely through the coupling strength and the interaction time, i.e. a faulty gate is of the form $`𝐔_f=e^{i(\varphi +\mathrm{\Delta }\varphi )𝐇}`$, where $`𝐔=e^{i\varphi 𝐇}`$ is the unperturbed gate. The distance
$`d(𝐔,𝐔_f)`$ $`=`$ $`\sqrt{1{\displaystyle \frac{1}{d_K}}Re\left[\mathrm{Tr}(e^{i\mathrm{\Delta }\varphi 𝐇})\right]}=\sqrt{1{\displaystyle \frac{1}{d_K}}Re\left[\mathrm{Tr}(\mathrm{cos}\mathrm{\Delta }\varphi 1+i\mathrm{sin}\mathrm{\Delta }\varphi 𝐇)\right]}`$ (33)
$`=`$ $`\sqrt{1\mathrm{cos}\mathrm{\Delta }\varphi }=\sqrt{2}\mathrm{sin}(\mathrm{\Delta }\varphi /2){\displaystyle \frac{\mathrm{\Delta }\varphi }{\sqrt{2}}}\delta ,`$ (34)
is proportional to the error $`\mathrm{\Delta }\varphi `$ of the product of coupling-strength and interaction time. This translates to (nearly) linear behavior in the desired final accuracy $`ϵ`$.
## V Collective Decoherence
We now focus on a particularly interesting and useful model of a DFS. This is the case of collective decoherence on $`n`$ qubits. We distinguish between two forms of collective decoherence. The first, and simpler, type of collective decoherence is weak collective decoherence (WCD). We define the collective operators as
$$𝐒_\alpha \underset{j=1}{\overset{n}{}}\sigma _\alpha ^j,$$
(35)
where $`\sigma _\alpha ^j`$ denotes a tensor product of the $`\alpha ^{\mathrm{th}}`$ Pauli matrix, $`\alpha =x,y,z`$,
$$\sigma _x=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)\sigma _y=\left(\begin{array}{cc}0& i\\ i& 0\end{array}\right)\sigma _z=\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)$$
(36)
(in the basis spanned by $`\sigma _z`$ eigenstates $`|0`$ and $`|1`$) operating on the $`j^{\mathrm{th}}`$ qubit, and the identity on all of the other qubits. WCD is the situation in which only one collective operator $`𝐒_\alpha `$ is involved in the coupling to the bath, i.e., $`𝐇_I=𝐒_\alpha 𝐁`$.
The second, more general type of collective decoherence is strong collective decoherence (SCD). We define SCD as the general situation in which the interaction Hamiltonian is given by $`𝐇_I=_{\alpha =1}^3𝐒_\alpha 𝐁_\alpha `$. The $`𝐒_\alpha `$ provide a representation of the Lie algebra $`su(2)`$:
$$[𝐒_\alpha ,𝐒_\beta ]=2iϵ_{\alpha \beta \gamma }𝐒_\gamma $$
(37)
The $`𝐁_\alpha `$’s are not required to be linearly independent.
Both of these collective decoherence mechanisms are expected to arise from the physical condition that the bath cannot distinguish the system qubits . If there are $`n`$ qubits interacting with a bath, the most general interaction Hamiltonian linear in the $`\sigma _\alpha ^i`$ is given by
$$𝐇_I=\underset{i=1}{\overset{n}{}}\underset{\alpha =x,y,z}{}\sigma _\alpha ^i𝐁_{i,\alpha }$$
(38)
where the $`𝐁_{i,\alpha }`$ are bath operators. If the bath cannot distinguish between the system qubits, then $`𝐁_{i,\alpha }`$ should not depend on $`i`$ and the Hamiltonian becomes $`𝐇_I=_{\alpha =x,y,z}𝐒_\alpha 𝐁_\alpha `$, i.e., strong collective decoherence.
As a concrete example of such collective decoherence, consider the situation in which the bath is the electromagnetic field, and the wavelength of the transition between the states of the qubits is larger than the spacing between the qubits. The electromagnetic field will interact with each of these qubits in an identical manner, because the field strength over a single wavelength will not vary substantially. This gives rise to the well- known phenomena of Dicke super- and sub-radiance . Whenever the bath is a field whose energy is dependent on its wavelength and this wavelength is much greater than the spacing between the qubits, one should expect collective decoherence to be the dominant decoherence mechanism. It is natural to expect this to be the case for condensed-phase high-purity materials at low temperatures. However, to the best of our knowledge at present a rigorous study quantifying the relevant parameter ranges for this interesting condition to hold in specific materials is still lacking (see Refs. for an application to quantum dots, though).
## VI The Abelian Case: Weak Collective Decoherence
For a decoherence mechanism with only one operator $`𝐒_\alpha `$ coupling to the bath, the implementation and discussion of universal computation with local interactions is simpler than in the general case, because we can work in the basis that diagonalizes $`𝐒_\alpha `$ ($`𝐒_\alpha `$ is necessarily Hermitian in the Hamiltonian model we consider here). The algebra generated by $`𝐒_\alpha `$ is abelian and reduces to one-dimensional (irreducible) subalgebras corresponding to the eigenvalues of $`𝐒_\alpha `$. More specifically, $`𝒜_1=_{\lambda _J}𝐈_{n_J}(\lambda _J)`$, where $`\lambda _J`$ is the $`J^{\mathrm{th}}`$ eigenvalue with degeneracy $`n_J`$, and $`(\lambda _J)`$ is the algebra generated by $`\lambda _J`$. $`(\lambda _J)`$ acts by multiplying the corresponding vector by $`\lambda _J`$. In this situation the DF subsystems are only of the DF subspace type. This simpler case of weak collective decoherence allows us to present a treatment with examples, that will make the general case of strong collective decoherence (SCD) more intuitive.
In the following we will, without loss of generality, focus on the case $`𝐒_\alpha 𝐒_z=_{k=1}^n\sigma _z^k`$.<sup>§</sup><sup>§</sup>§ The cases $`\alpha =x`$ ($`y`$) follow by applying a bitwise Hadamard (Hadamard+phase) transform to the code. This operator is already diagonal in the computational basis (the eigenstates are bitstrings of qubits in either $`|0`$ or $`|1`$). Since $`\sigma _z^k`$ acting on the $`k^{\mathrm{th}}`$ qubit contributes $`1`$ if the qubit is $`|0`$, and $`1`$ if the qubit is $`|1`$, the eigenvalue of a bitstring is $`\mathrm{\#}0\mathrm{\#}1`$ (the number of zeroes minus the number of ones), and the eigenvalues of $`𝐒_z`$ are $`\{n,n2,\mathrm{},n+2,n\}`$.
The degeneracy $`n_J`$ of the eigenspace corresponding to an eigenvalue
$$\lambda _J=n2J$$
(39)
is
$$n_J=\left(\genfrac{}{}{0pt}{}{n}{J}\right)$$
(40)
(the number of different bitstrings with $`nJ`$ zeroes and $`J`$ ones). The abelian algebra generated by $`𝐒_z`$ thus splits into one-dimensional subalgebras with degeneracy $`n_J`$. The largest decoherence-free subspaces in this situation correspond to the space spanned by bitstring-vectors where the number of zeroes and the number of ones are either the same ($`n`$ even), or differ by one ($`n`$ odd).
### A The Stabilizer and Error Correction Properties
Following the formalism developed in Section (III) we find, using Eq. (23) with $`v=i\theta `$ ($`\theta `$ can be complex), the stabilizer for the weak case corresponding to a DFS with eigenvalue $`\lambda _J`$ to be
$$𝐙_J^n(\theta )=\mathrm{exp}[i\theta (𝐒_z\lambda _J𝐈)]=\underset{k=1}{\overset{n}{}}e^{i\lambda _J\theta }(𝐈\mathrm{cos}\theta +\sigma _z^ki\mathrm{sin}\theta )=e^{i\lambda _J\theta }𝐏(\theta )^n$$
(41)
where
$$𝐏(\theta )=\left(\begin{array}{cc}e^{i\theta }& 0\\ 0& e^{i\theta }\end{array}\right).$$
(42)
For strictly real $`\theta `$ some of the errors which are protected against are simply collective rotations about the $`\sigma _z`$ axis (and an irrelevant global phase). For strictly imaginary $`\theta `$ we find that the errors which are protected against are contracting collective errors of the form $`\mathrm{diag}(e^\theta ,e^\theta )`$, i.e., they result in loss of norm of the wave function. Any physical process with Kraus operators that are linear combinations of these errors will therefore not affect the DFS.
This is the right framework in which to present another form of the stabilizer. We note that in the case of weak collective decoherence, we can find a stabilizer group with a finite number of elements. Define
$$Z_{\frac{1}{n}}=\mathrm{exp}\left(\frac{2\pi i}{n}\sigma _z\right)=\left(\begin{array}{cc}\mathrm{exp}(\frac{2\pi i}{n})& 0\\ 0& \mathrm{exp}(\frac{2\pi i}{n})\end{array}\right).$$
(43)
Then the $`n`$-element group $`𝒵_n`$ generated by $`\mathrm{exp}\left(i2\pi \lambda _J/n\right)Z_{1/n}^n`$ is a stabilizer for the DFS corresponding to the eigenvalue $`\lambda _J`$. To see that
$$\mathrm{exp}\left(\frac{2\pi i\lambda _J}{n}\right)Z_{\frac{1}{n}}^n|\mathrm{\Psi }=|\mathrm{\Psi }\mathrm{iff}|\mathrm{\Psi }\mathrm{DFS}(\lambda _J),$$
(44)
note that a $`Z_{1/n}`$ acting on a $`|0`$ contributes $`\mathrm{exp}(2\pi i/n)`$ to the total phase, whereas $`Z_{1/n}`$ acting on a $`|1`$ contributes $`\mathrm{exp}(2\pi i/n)`$. So $`Z_{1/n}^n`$ gives a total phase of $`\mathrm{exp}\left(2\pi i(\mathrm{\#}|0\mathrm{\#}|1)/n\right)=\mathrm{exp}(2\pi i\lambda _J/n)`$ when acting on a bitstring. This stabilizer and Eq. (44) provide a simple criterion to check whether a state is in a DFS or not.
Let us now briefly comment on the error-correction and detection properties of the code in the WCD case. The stabilizer elements are all diagonal, and equal to a tensor product of identical 1-qubit operators. The element $`𝐙^n`$ is in the stabilizer and anticommutes with odd-number $`𝐗`$ and $`𝐘`$ errors. So odd-number qubit bit-flips are detectable errors. However the code is not able to detect any form of error involving $`𝐙`$’s and even-number $`𝐗`$’s and $`𝐘`$’s, since any such error commutes with all elements in the stabilizer.
### B Nontrivial Operations
Observe that the algebra $`𝒜`$ in the WCD case is generated entirely by $`𝐒_z`$. Hence, by Theorem 2, the DFS-preserving operations are those that are in the commutant of $`𝐒_z`$. For single-body Hamiltonians it is easy to see that the only non-trivial such set is formed by interactions proportional to $`\sigma _z^i`$ operators. As for two-qubit Hamiltonians, it is simpler to use Theorem 1 and the expression (41) for the stabilizer. We are then looking for $`4\times 4`$ Hermitian matrices that commute with $`𝐏(\theta )^2`$; these are of the form
$$𝐓_{ij}(z_1,z_2,z_3,z_4,h)=\left(\begin{array}{cccc}z_1& 0& 0& 0\\ 0& z_2& h& 0\\ 0& h^{}& z_3& 0\\ 0& 0& 0& z_4\end{array}\right)$$
(45)
where $`𝐓_{ij}`$ acts on qubits $`i`$ and $`j`$ only. Here $`z_i`$ is real, $`h`$ is complex, and the row space is spanned by the $`i^{\mathrm{th}}`$ and $`j^{\mathrm{th}}`$ qubit basis $`\{|00,|01,|10,|11\}`$. We note that systems with an internal Hamiltonian of the Heisenberg type,
$$𝐇_{\mathrm{Heis}}=\underset{j=1}{\overset{n}{}}ϵ_j\sigma _z^j+\frac{1}{2}\underset{i,j=1}{\overset{n}{}}J_{ij}\stackrel{}{\sigma }_i\stackrel{}{\sigma }_j,$$
(46)
have exactly the correct form for any pair of spins $`i`$,$`j`$. Indeed, it is not hard to see that $`[𝐇_{\mathrm{Heis}},𝐒_z]=0`$ . The Heisenberg Hamiltonian is ubiquitous, and appears, e.g., in NMR. This means that the natural evolution of NMR systems under WCD preserves the DFS, and implements a non-trivial computation.
The specific case
$$𝐄_{ij}𝐓_{ij}(1,0,0,1,1)=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& 0& 1& 0\\ 0& 1& 0& 0\\ 0& 0& 0& 1\end{array}\right),$$
(47)
which flips the two states $`|01`$ and $`|10`$ of qubits $`i`$ and $`j`$ and leaves the other two states invariant, is especially important: it is the exchange interaction. The other interactions we employ are
$`𝐓_{ij}^P`$ $`𝐓_{ij}(1,0,0,0,0)=\mathrm{diag}(1,0,0,0)`$ (48)
$`𝐓_{ij}^Q`$ $`𝐓_{ij}(0,0,0,1,0)=\mathrm{diag}(0,0,0,1),`$ (49)
which introduce a phase on the state $`|00`$ ($`P`$) and $`|11`$ ($`Q`$) of qubits $`i`$ and $`j`$; and
$$\overline{𝐙}_{12}𝐓_{12}(0,0,1,0,0)\}=\mathrm{diag}(0,0,1,0).$$
(50)
In the following we show that these special interactions are sufficient to obtain a universal generating set operating entirely within a weak-collective DFS.
### C Universal Quantum Computation inside the Weak-Collective DFS
Let DFS<sub>n</sub>($`K`$) denote the DFS on $`n`$ physical qubits with eigenvalue $`K`$. We show here that
$$𝖧=\{𝐄_{i,i+1},𝐓_{i,i+1}^P,𝐓_{i,i+1}^Q:i=1,\mathrm{},n1,\overline{𝐙}_{12}\}$$
(51)
is a universal generating set for any of the DFSs occurring in a system of $`n`$ physical qubits. It is convenient to work directly with the Hamiltonians, and to show that $`𝖧`$ gives rise to the Lie-algebra $`su(d_K)`$ on each DFS<sub>n</sub>($`K`$) \[via scalar multiplication, addition, and Lie-commutator; see the allowed compositions of operations (1-3) in Section IV B\]. Exponentiation then gives the group $`SU(d_K)`$ on the DFS. We will proceed by induction on $`n`$, the number of physical qubits, building the DFS-states of $`n`$ qubits out of DFS-states for $`n1`$ qubits. A graphical representation of this construction is useful (and will also generalize to the strong case presented in the following section VII): see Fig. (1).
We have seen that in the WCD case the DFS states are simply bitstrings of $`n`$ qubits in either $`|0`$ or $`|1`$. The different $`n`$-qubit DFSs are labeled by their eigenvalue
$$\lambda _J=\mathrm{\#}0\mathrm{\#}1J_n.$$
(52)
To obtain a DFS-state of $`n`$ qubits out of a DFS-state of $`n1`$ qubits corresponding to $`J_{n1}`$ we can either add the $`n^{\mathrm{th}}`$ qubit as $`|0`$ ($`J_n=J_{n1}+1`$) or as $`|1`$ ($`J_n=J_{n1}1`$). Each DFS-state can be built sequentially from the first qubit onward by adding successively $`|0`$ or $`|1`$, and is uniquely defined by a sequence $`J_1,\mathrm{},J_n`$ of eigenvalues. In the graphical representation of Fig. (1) the horizontal axis marks $`n`$, the number of qubits up to which the state is already built, and the vertical axis shows $`J_n`$, the difference $`\mathrm{\#}0\mathrm{\#}1`$ up to the $`n^{\mathrm{th}}`$ qubit. Adding a $`|0`$ at the $`n+1^{\mathrm{th}}`$ step will correspond to a line pointing upwards, adding a $`|1`$ to a line pointing down. Each DFS-state of $`n`$qubits with eigenvalue $`\lambda _J=J_n`$ is thus in one-to-one correspondence with a path on the lattice from the origin to $`(n,J_n)`$.
Consider the first non-trivial case, $`n=2`$, which gives rise to one DFS-qubit: DFS<sub>2</sub>($`0`$). This corresponds to the two states $`|0_L=|01`$ \[path 2 in Fig. (1)\] and $`|1_L=|10`$ (path 3) with $`J_2=0`$. The remaining Hilbert space is spanned by the one-dimensional DFS<sub>2</sub>($`2`$) $`|00`$ (path 1) corresponding to $`J_2=2`$, and DFS<sub>2</sub>($`2`$) $`|11`$ (path 4) corresponding to $`J_2=2`$. The exchange $`𝐄_{12}`$ flips $`|0_L`$ and $`|1_L`$ (path 2 and 3), and leaves the other two paths unchanged. The interaction $`𝐀_{12}=\mathrm{diag}(0,0,1,0)`$ induces a phase on $`|1_L=|10`$ (path 3). Their commutator forms an encoded $`\sigma _y`$ acting entirely within the DFS<sub>2</sub>($`0`$) subspace. Its commutator with $`𝐄_{12}`$ in turn forms an encoded $`\sigma _z`$ with the same property. Together they form the (encoded) Lie algebra $`su(2)`$ acting entirely within this DFS. The Lie algebra is completed by forming the commutator between these $`\overline{𝐘}`$ and $`\overline{𝐙}`$ operations. To summarize:
$`\overline{𝐘}_{12}`$ $`=`$ $`i[\overline{𝐀},𝐄_{12}]=\left(\begin{array}{cccc}0& 0& 0& 0\\ 0& 0& i& 0\\ 0& i& 0& 0\\ 0& 0& 0& 0\end{array}\right)`$ (57)
$`\overline{𝐙}_{12}`$ $``$ $`i[𝐄_{12},\overline{𝐘}_{12}]`$ (58)
$`\overline{𝐗}_{12}`$ $``$ $`i[\overline{𝐘}_{12},\overline{𝐙}_{12}]`$ (59)
We call the property of acting entirely within the specified DFS independence, meaning that the corresponding Hamiltonian has zero entries in the rows and columns corresponding to the other DFSs \[DFS<sub>2</sub>($`2`$)=$`|00`$ and DFS<sub>2</sub>($`2`$)=$`|11`$ in this case\]. When the Hamiltonian is exponentiated, the corresponding gate will act as identity on all DFSs except DFS<sub>2</sub>(0).
To summarize these considerations, the Lie-algebra formed by $`𝖧_0^2=\{\overline{𝐗},\overline{𝐙}\}`$ is $`su(2)`$, and generates $`SU(2)`$ on DFS<sub>2</sub>(0) by exponentiation. In addition, this is an independent $`SU(2)`$, namely, these operations act as identity on the other DFSs: when written as matrices over the basis of DFS-states, their generators in $`𝖧_0^2`$ have zeroes in the rows and columns corresponding to all other DFSs.
In the following we show how this construction generalizes to $`n>2`$ qubits, by proving the following theorem:
Theorem 4— For any $`n2`$ qubits undergoing weak collective decoherence, there exist sets of Hamiltonians $`𝖧_{J_n}^n`$ \[obtained from $`𝖧`$ of Eq. (51) via scalar multiplication, addition, and Lie-commutator\] acting as $`su(d_{J_n})`$ on the DFS corresponding to the eigenvalue $`J_n`$. Furthermore each set acts independently on this DFS only (i.e., with zeroes in the matrix representation corresponding to their action on the other DFSs).
Before proving this theorem, we first explain in detail the steps taken in order to go from the $`n=2`$ to the $`n=3`$ case, so as to make the general induction procedure more transparent.
The structure of the DFSs for $`n=2`$ and $`3`$ qubits is:
$`\mathrm{DFS}_2(2)`$ $`=`$ $`\{|00\},\mathrm{DFS}_2(0)=\{\begin{array}{c}|01\\ |10\end{array},\mathrm{DFS}_2(2)=\{|11\}`$ (62)
$`\mathrm{DFS}_3(3)`$ $`=`$ $`\{|000\},\mathrm{DFS}_3(1)=\{\begin{array}{c}|001\\ |010\\ |100\end{array},\mathrm{DFS}_3(1)=\{\begin{array}{c}|011\\ |101\\ |110\end{array},\mathrm{DFS}_3(3)=\{|111\}.`$ (69)
DFS<sub>3</sub>($`3`$) is obtained by appending a $`|0`$ to DFS<sub>2</sub>($`2`$). Similarly DFS<sub>3</sub>($`3`$) is obtained by appending a $`|1`$ to DFS<sub>2</sub>($`2`$). Graphically, this corresponds to moving along the only allowed pathway from DFS<sub>2</sub>($`2`$) \[DFS<sub>2</sub>($`2`$)\] to DFS<sub>3</sub>($`3`$) \[DFS<sub>3</sub>($`3`$)\], as shown in Fig. (1). The lowest and highest $`\lambda _J`$ for $`n`$ qubits will always be made up of the single pathway connecting the lowest and highest $`\lambda _J`$ for $`n1`$ qubits. The structure of DFS<sub>3</sub>($`\pm 1`$) is only slightly more complicated. DFS<sub>3</sub>($`1`$) is made up of one state, $`|001`$, which comes from appending a $`|1`$ (moving down) to DFS<sub>2</sub>($`2`$). We call $`|001`$ a “Top-state” in DFS<sub>3</sub>($`1`$). The two other states, $`|010`$ and $`|100`$, come from appending $`|0`$ (moving up) to DFS<sub>2</sub>($`0`$). Similarly, we call $`|010`$ and $`|100`$ “Bottom-states” in DFS<sub>3</sub>($`1`$). DFS<sub>3</sub>($`1`$) is constructed in an analogous manner (Fig. 1).
We showed above that it is possible to perform independent $`su(2)`$ operations on DFS<sub>2</sub>($`0`$). DFS<sub>2</sub>($`\pm 2`$) are also both acted upon independently, but because they are one-dimensional subspaces, independence implies that $`su(2)`$ operations annihilate them. Since the states $`\{|010,|100\}`$ DFS<sub>3</sub>($`1`$) and the states $`\{|011,|101\}`$ DFS<sub>3</sub>($`1`$) both have $`\{|01,|10\}`$ DFS<sub>2</sub>($`0`$) as their first two qubits, one immediate consequence of the independent action on DFS<sub>2</sub>($`0`$) is that one can simultaneously perform $`su(2)`$ operations on the corresponding daughter subspaces created by expanding DFS<sub>2</sub>($`0`$) into DFS<sub>3</sub>($`\pm 1`$). The first step in the general inductive proof is to eliminate this simultaneous action, and to act independently on each of these subspaces (the “independence step”). To see how this is achieved, it is convenient to represent the operators acting on the $`8`$-dimensional Hilbert space of $`3`$ qubits in the basis of the $`4`$ DFSs:
| $`000`$ | $`001`$ | $`010`$ | $`100`$ | $`011`$ | $`101`$ | $`110`$ | $`111`$ |
| --- | --- | --- | --- | --- | --- | --- | --- |
| $`M_3`$ | | | | | | | |
| | | $`M_1`$ | | | | | |
| | | | | | $`M_1`$ | | |
| | | | | | | | $`M_3`$ |
The simultaneous action on DFS<sub>3</sub>($`\pm 1`$) can now be visualized in terms of both $`M_{\pm 1}`$ being non-zero. Let us show how to obtain an action where, say, just $`M_1`$ is non-zero. This can be achieved by applying the commutator of two operators with the property that their intersection has non-vanishing action just on $`M_1`$. This is true for the $`𝐓_{23}^P`$ and $`\overline{𝐗}_{12}`$ Hamiltonians: $`𝐓_{23}^P`$ annihilates every state except those that are $`|00`$ over qubits $`2`$ and $`3`$, namely $`|100`$ DFS<sub>3</sub>($`1`$) and $`|000`$DFS<sub>3</sub>($`3`$). This implies that the only non-zero blocks in its matrix are
$$M_3(𝐓_{23}^P)=1,M_1(𝐓_{23}^P)=\left(\begin{array}{ccc}0& 0& \\ 0& 0& \\ & & 1\end{array}\right).$$
(70)
On the other hand, $`\overline{𝐗}_{12}`$ is non-zero only on those states that are $`|01`$ or $`|10`$ on qubits $`1`$ and $`2`$. Therefore it will be non-zero on all $`3`$-qubit states that have $`|01`$ or $`|10`$ as “parents”. This means that in its matrix representation $`M_{\pm 3}=0`$ and
$$M_1(\overline{𝐗}_{12})=\left(\begin{array}{ccc}0& & \\ & 0& 1\\ & 1& 0\end{array}\right),M_1(\overline{𝐗}_{12})=\left(\begin{array}{ccc}0& 1& \\ 1& 0& \\ & & 0\end{array}\right).$$
(71)
Clearly, taking the product of $`𝐓_{23}^P`$ and $`\overline{𝐗}_{12}`$ leaves non-zero just the lower $`2\times 2`$ block of $`M_1`$, and this is the crucial point: it shows that an independent action on DFS<sub>3</sub>($`1`$) can be obtained by forming their commutator. Specifically, since the lower $`2\times 2`$ block of $`M_1(𝐓_{23}^P)`$ is just $`\frac{1}{2}\left(𝐈\sigma _z\right)`$:
$$i[𝐓_{23}^P,\overline{𝐗}_{12}]=\overline{𝐘}_{\{|100,|010\}},$$
(72)
i.e., this commutator acts as an encoded $`\sigma _y`$ inside the $`\{|100,|010\}`$ subspace of DFS<sub>3</sub>($`1`$). Similarly, $`\overline{𝐙}_{\{|100,|010\}}=\frac{i}{2}[\overline{𝐘}_{\{|100,|010\}},\overline{𝐗}_{12}]`$. Together $`\{\overline{𝐘}_{\{|100,|010\}},\overline{𝐙}_{\{|100,|010\}}\}`$ generate $`su(2)`$ acting independently on the $`\{|100,|010\}`$ subspace of DFS<sub>3</sub>($`1`$), which we achieved by subtracting out the action on DFS<sub>3</sub>($`1`$).
In an analogous manner, an independent $`su(2)`$ can be produced on the $`\{|011,|101\}`$ subspace of DFS<sub>3</sub>($`1`$) by using the Hamiltonians acting on DFS<sub>2</sub>($`0`$) in conjunction with $`𝐓_{23}^Q`$ to subtract out the $`su(2)`$ action on DFS<sub>3</sub>($`1`$). Since $`𝐓_{23}^Q`$ annihilates every state except those that are $`|11`$ over qubits $`2`$ and $`3`$, namely $`|011`$ DFS<sub>3</sub>($`1`$) and $`|111`$DFS<sub>3</sub>($`3`$), the only non-zero blocks in its matrix are
$`M_3(𝐓_{23}^Q)=1,M_1(𝐓_{23}^Q)=\left(\begin{array}{ccc}1& & \\ & 0& 0\\ & 0& 0\end{array}\right).`$
Thus we can obtain independent action for each of the daughters of DFS<sub>2</sub>($`0`$), i.e., separate actions on the subspace spanned by $`\{|010,|100\}`$ and $`\{|011,|101\}`$.
Having established independent action on the two subspaces of DFS<sub>3</sub>($`1`$) and DFS<sub>3</sub>($`1`$) arising from DFS<sub>2</sub>($`0`$), we need only show that we can obtain the full action on DFS<sub>3</sub>($`1`$) and DFS<sub>3</sub>($`1`$). For DFS<sub>3</sub>($`1`$) we need to mix the subspace $`\{|010,|100\}`$ over which we can already perform independent $`su(2)`$, with the $`|001`$ state. To do so, note that the effect of the exchange operation $`𝐄_{23}`$ is to flip $`|001`$ and $`|010`$, and leave $`|100`$ invariant. I.e., the matrix representation of $`𝐄_{23}`$ is
$$M_1(𝐄_{23})=\left(\begin{array}{ccc}0& 1& \\ 1& 0& \\ & & 1\end{array}\right).$$
(73)
Unfortunately, $`𝐄_{23}`$ has a simultaneous action on DFS<sub>3</sub>($`1`$). This, however, is not a problem, since we have already constructed an independent $`su(2)`$ on DFS<sub>3</sub>($`1`$) elements. Thus we can eliminate the simultaneous action by simply forming commutators with these $`su(2)`$ elements. The Lie algebra generated by these commutators will act independently on all of DFS<sub>3</sub>($`1`$). In fact we claim this Lie algebra to be all of $`su(3)`$ (see Appendix B for a general proof). In other words, the Lie algebra spanned by the $`su(2)`$ elements $`\{\sigma _x,\sigma _y,\sigma _z\}`$ acting on the subspace $`\{|100,|010\}`$, together with the exchange operation $`𝐄_{23}`$, generate all of $`su(3)`$ independently on DFS($`1`$). A similar argument holds for DFS<sub>3</sub>($`1`$). This construction illustrates the induction step: we have shown that it is possible to perform independent $`su(d_K)`$ actions on all four of the DFS<sub>3</sub>($`K`$) ($`K=\pm 3,\pm 1`$), given that we can perform independent action on the three DFS<sub>2</sub>($`K`$) ($`K=\pm 1,0`$). In Fig. (2) we have further illustrated these considerations by depicting the action of exchange on two the $`4`$-qubit DFSs. Let us now proceed to the general proof.
Proof— By induction.
The case $`n=2`$ already treated above will serve to initialize the induction. Assume now that the theorem is true for $`n1`$ qubits and let us show that it is then true for $`n`$ qubits as well.
First note that each DFS<sub>n</sub>($`K`$) is constructed either from the DFS<sub>n-1</sub>($`K1`$) (to its lower left) by adding a $`|0`$ for the $`n^{\mathrm{th}}`$ qubit, or from DFS<sub>n-1</sub>($`K+1`$) (to its upper left) by adding a $`|1`$: the states in DFS<sub>n</sub>($`K`$) correspond to all paths ending in $`(n,K)`$ that either come from below (B) or from the top (T). See Fig. (3).
If we apply a certain gate $`𝐔=\mathrm{exp}(i𝐇t)`$ to DFS<sub>n-1</sub>($`K+1`$), then this operation will induce the same $`𝐔`$ on DFS<sub>n</sub>($`K`$), by acting on all paths (states) entering DFS<sub>n</sub>($`K`$) from above. At the same time $`𝐔`$ is induced on DFS<sub>n</sub>($`K+2`$) by acting on all paths entering this DFS from below. So, $`𝐔`$ affects two DFSs simultaneously. In other words, the set of valid Hamiltonians $`𝖧_{K+1}^{n1}`$ \[acting on $`n1`$ qubits and generating $`su(d_{K+1})`$\] on DFS<sub>n-1</sub>($`K+1`$), that we are given by the induction hypothesis, induces a simultaneous action of $`su(d_{K+1})`$ on DFS<sub>n</sub>($`K`$) (on the paths coming from above only) and DFS<sub>n</sub>($`K+2`$) (on the paths coming from below only). Additionally, it does not affect any other $`n`$-qubit DFS, since we assumed that the action on DFS<sub>n-1</sub>($`K+1`$) was independent, and the only $`n`$-qubit DFSs built from DFS<sub>n-1</sub>($`K+1`$) are DFS<sub>n</sub>($`K`$) and DFS<sub>n</sub>($`K+2`$). These considerations are depicted schematically in Fig. (3).
We now show how to annihilate, for a given non-trivial (i.e., dimension $`>1`$) DFS<sub>n</sub>($`K`$), the unwanted simultaneous action on other DFSs (the “independence step”). Then we proceed to obtain the full $`su(d_K)`$, by using the $`su(d_{K\pm 1})`$ on DFS<sub>n-1</sub>($`K\pm 1`$) that are given by the induction hypothesis (the “mixing step”).
#### 1 Independence
Let us call all the $`t_K`$ paths converging on DFS<sub>n</sub>($`K`$) from above “Top-states”, or T-states for short, and the $`b_K`$ paths converging from below “Bottom- (or B) states” (recall that there is a 1-to-1 correspondence between paths and states). The total number of paths converging on a given DFS is exactly its dimension, so $`d_K=t_K+b_K`$. By using the induction hypothesis on DFS<sub>n-1</sub>($`K+1`$) we can obtain $`su(t_K)`$ (generated by $`𝖧_{K+1}^{n1}`$) on the T-states of DFS<sub>n</sub>($`K`$), which will simultaneously affect the B-states in the higher lying DFS$`{}_{n}{}^{}(K+2)`$ as $`su(b_{K+2}`$) (note that $`t_K=b_{K+2}`$). The set $`𝖧_{K+1}^{n1}`$ is non-empty only if $`n3K+1(n3)`$ \[because the “highest” and “lowest” DFS are always one-dimensional and $`su(1)=0`$\]. If this holds then DFS<sub>n</sub>($`K+2`$) “above” DFS<sub>n</sub>($`K`$) is non-trivial (dimension $`>1`$), and there are paths in DFS<sub>n</sub>($`K`$) ending in $`|11`$ (“down, down”). This is exactly the situation in which we can use $`𝐓_{n1,n}^Q`$ to wipe out the unwanted action on DFS<sub>n</sub>($`K+2`$): recall that $`𝐓_{n1,n}^Q`$ annihilates all states except those ending in $`|11`$, and therefore affects non-trivially only these special T-states in each DFS. Since the operations in $`𝖧_{K+1}^{n1}`$ affect only B-states on DFS<sub>n</sub>($`K+2`$), $`𝐓_{n1,n}^Q`$ commutes with $`𝖧_{K+1}^{n1}`$ on DFS<sub>n</sub>($`K+2`$). Therefore the commutator of $`𝐓_{n1,n}^Q`$ with elements in $`𝖧_{K+1}^{n1}`$ annihilates all states not in DFS<sub>n</sub>($`K`$). The argument thus far closely parallels the discussion above showing how to generate an independent $`su(2)`$ on the $`\{|011,|101\}`$ subspace of DFS<sub>3</sub>($`1`$), starting from the $`su(2)`$ on DFS<sub>2</sub>($`0`$) and $`𝐓_{23}^Q`$. To show that commuting $`𝐓_{n1,n}^Q`$ with $`𝖧_{K+1}^{n1}`$ generates $`su(t_K)`$ on the T-states of DFS<sub>n</sub>($`K`$) we need the following lemma, which shows how to form $`su(d)`$ from an overlapping $`su(d1)`$ and $`su(2)`$:
Enlarging Lemma— Let $``$ be a Hilbert space of dimension $`d`$ and let $`|i`$. Assume we are given a set of Hamiltonians $`𝖧_1`$ that generates $`su(d1)`$ on the subspace of $``$ that does not contain $`|i`$ and another set $`𝖧_2`$ that generates $`su(2)`$ on the subspace of $``$ spanned by $`\{|i,|j\}`$, where $`|j`$ is another state in $``$. Then $`[𝖧_1,𝖧_2]`$ (all commutators) generates $`su(d)`$ on $``$ under closure as a Lie-algebra (i.e., via scalar multiplication, addition and Lie-commutator).
Proof— See Appendix C
Now consider two states $`|i,|j`$DFS<sub>n</sub>($`K`$) such that $`|i`$ ends in $`|11`$ and $`|j`$ is a T-state, but does not end in $`|11.`$ Then we can generate $`su(2)`$ on the subspace spanned by $`\{|i,|j\}`$ as follows: (i) We use the exchange interaction $`\overline{𝐗}_{ij}=|i^{}j^{}|+|j^{}i^{}|`$ \[a prime indicates the bitstring with the last bit (a $`1`$ in this case) dropped\] in $`su(t_K)𝖧_{K+1}^{n1}`$ to generate a simultaneous action on DFS<sub>n</sub>($`K`$) and DFS<sub>n</sub>($`K+2`$). This interaction is represented by a $`2\times 2`$ $`\sigma _x`$-matrix in the subspace spanned by $`\{|i,|j\}`$. (ii) $`𝐓_{n1,n}^Q`$ is represented by the $`2\times 2`$ matrix $`\mathrm{diag}(1,0)=\frac{1}{2}\left(𝐈+\sigma _z\right)`$ in the same subspace, and commutes with $`\overline{𝐗}_{ij}`$ on DFS<sub>n</sub>($`K+2`$) (since $`\overline{𝐗}_{ij}`$ affects only B-states in DFS<sub>n</sub>($`K+2`$), and $`𝐓_{n1,n}^Q`$ is non-zero only on states ending in $`|11`$). Thus we can use it to create an independent action on DFS<sub>n</sub>($`K`$) alone: $`\overline{𝐘}_{ij}=i[𝐓_{n1,n}^Q,\overline{𝐗}_{ij}]`$, $`\overline{𝐙}_{ij}=\frac{i}{2}[\overline{𝐘}_{ij},\overline{𝐗}_{ij}]`$.
Together $`\{\overline{𝐘}_{ij},\overline{𝐙}_{ij}\}`$ generate $`su(2)`$ independently on $`\{|i,|j\}`$ DFS<sub>n</sub>($`K`$). Since these operators vanish everywhere except on DFS<sub>n</sub>($`K`$), their commutators with elements in $`𝖧_{K+1}^{n1}`$ \[acting as $`su(t_K)`$\] will annihilate all other DFSs. Therefore, using the Enlarging Lemma, in this way all operations in $`su(t_K)`$ acting on DFS<sub>n</sub>($`K`$) only can be generated.
So far we have shown how to obtain an independent $`su(t_K)`$ on the T-states of DFS<sub>n</sub>($`K`$) using $`𝖧_{K+1}^{n1}`$ (for $`Kn4`$). To obtain an independent $`su(b_K)`$ on the B-states of DFS<sub>n</sub>($`K)`$ we use Hamiltonians in $`𝖧_{K1}^{n1}`$ (acting on DFS<sub>n-1</sub>($`K1)`$ – the DFS from below). This will generate a simultaneous $`su(b_K)`$ in DFS<sub>n</sub>($`K`$) and $`su(t_{K2})`$ in DFS<sub>n</sub>($`K2`$). To eliminate the unwanted action on DFS<sub>n</sub>($`K2`$) we apply the previous arguments almost identically, except that now we use $`𝐓_{n1,n}^P`$ to wipe out the action on all states except those ending in $`|00`$. We thus get an independent $`su(b_K)`$ on DFS<sub>n</sub>($`K`$). Together, the “above” and “below” constructions respectively provide independent $`su(t_K)`$ and $`su(b_K)`$ on DFS<sub>n</sub>($`K`$). Finally, note that we did not really need both $`𝐓_{ij}^P`$ and $`𝐓_{ij}^Q`$, since once we established independent action on the T-states, we could have just subtracted out this action when considering the B-states. Also, the specific choice of $`𝐓_{ij}^{P,Q}`$ was rather arbitrary (though convenient): in fact almost any other diagonal interaction would do just as well.
#### 2 Mixing
In order to induce operations between the two sets of paths (from “above” and from “below”) that make up DFS<sub>n</sub>($`K`$) consider the effect of $`𝐄_{n1,n}`$. This gate does not affect any paths that “ascend” two steps to $`(n,K)`$ (corresponding to bitstrings ending in $`|00`$) and paths that “descend” two steps (ending in $`|11`$), but it flips the paths that pass from $`(n2,K)`$ via $`(n1,K+1)`$ with the paths from $`(n2,K)`$ via $`(n1,K1)`$ \[see Fig. (3)\]. It does this for all DFSs simultaneously.
In order to get a full $`su(d_K)`$ on DFS<sub>n</sub>($`K`$) we need to “mix” $`su(t_K)`$ (on the T-states) and $`su(b_K)`$ (on the B-states) which we already have. We show how to obtain an independent $`su(2)`$ between a T-state and a B-state. By the Enlarging Lemma this generates $`su(d_K)`$.
Since $`n3`$ DFS<sub>n</sub>($`K`$) contains states terminating in $`|00`$ and/or $`|11`$. Let us assume, w.l.o.g., that states terminating in $`|00`$ are present, and let $`|i`$ be such a state (B-state). Let $`|j`$ be a B-state not terminating in $`|00`$, and let $`|k=𝐄_{n1,n}|j`$ ($`|k`$ is a T-state). Let $`\overline{𝐙}_{ij}=|ii||jj|su(b_K)`$, and recall that we have independent $`su(b_K)`$. Then as is easily checked, $`i[𝐄_{n1,n},\overline{𝐙}_{ij}]\overline{𝐘}_{jk}`$ yields $`\sigma _y`$ between $`|j`$ and $`|k`$ only.<sup>\**</sup><sup>\**</sup>\** Since $`𝐄_{n1,n}=|ii|+|kj|+|jk|+O`$, where $`O`$ is some action on an orthogonal subspace. In addition, $`\overline{𝐙}_{jk}\frac{i}{2}[𝐄_{n1,n},\overline{𝐘}_{jk}]`$ gives $`\sigma _z`$ between $`|j`$ and $`|k`$, thus completing a generating set for $`su(2)`$ on the B-state $`|j`$ and the T-state $`|k`$, that affects these two states only and annihilates all other states. This completes the proof.
To summarize, we have shown constructively that it is possible to generate the entire Lie algebra $`su(d_K)`$ on a given weak collective-decoherence DFS<sub>n</sub>($`K`$) of dimension $`d_K`$, from the elementary composition of the operations of scalar multiplication, addition, Lie-commutators (conjugation by unitaries was not necessary in the WCD case). Moreover, this $`su(d_K)`$ can be generated independently on each DFS, implying that universal quantum computation can be performed inside each DFS<sub>n</sub>($`K`$). Naturally, one would like to do this on the largest DFS. Since given the number of qubits $`n`$ the dimensions of the DFSs are $`d_K=\left(\genfrac{}{}{0pt}{}{n}{K}\right)`$, the largest DFS is the decoherence-free subspace $`K=0`$. In principle it is possible, by virtue of the independence result, to universally quantum compute in parallel on all DFSs.
### D State Preparation and Measurement on the Weak Collective Decoherence DFS
To make use of a DFS for encoding information in a quantum computer, in addition to the universal quantum computation described above, it must also be possible to initially prepare encoded states and to decode the quantum information on the DFS at the end of a computation. Encoding requires that the density matrix of the prepared states should have a large overlap with the DFS. Note that it is not necessary to prepare states that have support exclusively within the DFS. This follows from the fact that in our construction, while a computation is performed there is no mixing of states inside and outside of the DFS. If an initially prepared state is “contaminated” (has some support outside the DFS we want to compute on), then the result of the computation will have the same amount of contamination, i.e., the initial error does not spread.
For example, suppose we can prepare the state $`\rho =(1p)|\psi \psi |+p|\psi _{}\psi _{}|`$ where $`|\psi `$ is a state of a particular DFS and $`|\psi _{}`$ is a state outside of this DFS. Then the computation will proceed independently on the DFS and the states outside of the DFS. Readout will then obtain the result of the computation with probability $`1p`$. Repeated application of the quantum computation will give the desired result to arbitrary confidence level.
There are many choices for the initial states of a computation and the decision as to which states to prepare should be guided by the available gates and measurements and the accuracy that is achievable. For efficient computation one should try to maximize the overlap of the prepared state with the desired initial DFS state.
For the WCD case preparation of initial pure states is very simple. Suppose we are concerned with the $`𝐒_z`$ error WCD-DFS. Pure state preparation into such a DFS then corresponds to the ability to prepare a state which has support over states with a specific number of $`|0`$ and $`|1`$ (eigenstates of the $`\sigma _z`$ operator). This is particularly simple if measurements in the $`\sigma _z`$ basis ($`|0`$ , $`|1`$) as well as $`\sigma _x`$ gates (to “flip” the bits) are available.
The second crucial ingredient for computation on a DFS (in addition to preparation) is the decoding or readout of quantum information resulting from a computation. Once again, there are many options for how this can be performed. For example, in the WCD case one can make a measurement which distinguishes all of the DFSs and all of the states within this DFS by simply making a measurement in the $`\sigma _z`$ basis on every qubit. Further, all measurements with a given number of distinct eigenvalues can be performed by first rotating the observable into one corresponding to a measurement in the computational basis (which, in turn, corresponds to a unitary operation on the DFS) and then performing the given measurement in the $`\sigma _z`$ basis, and finally rotating back. There are other situations where one would like to, say, make a measurement of an observable over the DFS which has only two different eigenvalues. This type of measurement can be most easily performed by a concatenated measurement . In this scheme, one attaches another DFS to the original DFS, forming a single larger DFS. Then, assuming universal quantum computation over this larger DFS one can always perform operations which allow a measurement of the first DFS by entangling it with the second DFS, and reading out (destructively as described for the WCD above) the second DFS. For example, suppose the first DFS encodes two bits of quantum information, $`|k,l_L`$, $`k,l=\{0,1\}`$, and the second DFS encodes a single bit of quantum information $`\{|0_L`$, $`|1_L\}`$. Then one can make a measurement of the observable $`\sigma _z𝐈`$ on the first DFS by performing an encoded controlled-NOT operation between the first and the second DFS, and reading out the second DFS in the encoded $`\sigma _z`$ basis. For the WCD case the ability to make this destructive measurement on the ancilla (not on the code) simply corresponds to the ability to measure single $`\sigma _z`$ operations.
Finally, we note that for a WCD-DFS there is a destructive measurement which distinguishes between different DFSs (corresponding to a measurement of the number of $`|1`$’s). One can fault-tolerantly prepare a WCD-DFS state by repeatedly performing such a measurement to guarantee that the state is in the proper DFS. The concatenated measurement procedures described above for any DFS are naturally fault-tolerant in the sense that they can be repeated and are non-destructive . Thus fault-tolerant preparation and decoding is available for the WCD-DFS.
## VII Strong Collective Decoherence
Strong collective decoherence on $`n`$ qubits is characterized by the three system operators $`𝐒_x`$, $`𝐒_y`$ and $`𝐒_z`$. These operators form a representation of the semisimple Lie algebra $`su(2)`$. The algebra $`𝒜`$ generated by these operators can be decomposed as<sup>††</sup><sup>††</sup>†† Note that as a complex algebra $`\{𝐒_x,𝐒_y,𝐒_z\}`$ span all of $`gl(2)`$, not just $`su(2)`$.
$$𝒜\underset{J=0(1/2)}{\overset{n/2}{}}𝐈_{n_J}gl(2J+1,\text{ }\mathrm{C})$$
(74)
where $`J`$ labels the total angular momentum of the corresponding Hilbert space decomposition (and hence the $`0`$ or $`1/2`$ depending on whether $`n`$ is even or odd respectively) and $`gl(2J+1,\text{ }\mathrm{C})`$ is the general linear algebra acting on a space of size $`2J+1`$. The resulting decomposition of the system Hilbert space
$$_S\underset{J=0(1/2)}{\overset{n/2}{}}\text{ }\mathrm{C}_{n_J}\text{ }\mathrm{C}_{2J+1}$$
(75)
is exactly the reduction of the Hilbert states into different Dicke states . The degeneracy for each $`J`$ is given by :
$$n_J=\frac{(2J+1)n!}{(n/2+J+1)!(n/2J)!}.$$
(76)
Eq. (74) shows that given $`J`$, a state $`|J,\lambda ,\mu `$ is acted upon as identity on its $`\lambda `$ component. Thus a DFS is defined by fixing $`J`$ and $`\mu `$. As we will show later, $`\lambda `$ corresponds to the paths leading to a given point $`(n,J)`$ on the diagram of Fig. (4).
The DFSs corresponding to the different $`J`$ values for a given $`n`$ can be computed using standard methods for the addition of angular momentum. We use the convention that $`|1`$ represents a $`|j=1/2,m_j=1/2`$ particle and $`|0`$ represents a $`|j=1/2,m_j=1/2`$ particle in this decomposition although, of course, one should be careful to treat this labeling as strictly symbolic and not related to the physical angular momentum of the particles.
The smallest $`n`$ which supports a DFS and encodes at least a qubit of information is $`n=3`$ . For $`n=3`$ there are two possible values of the total angular momentum: $`J=3/2`$ or $`J=1/2`$. The four $`J=3/2`$ states $`|J,\lambda ,\mu =|3/2,0,\mu `$ ($`\mu =m_J=\pm 3/2,\pm 1/2`$) are singly degenerate; the $`J=1/2`$ states have degeneracy $`2`$. They can be constructed by either adding a $`J_{12}=1`$ (triplet) or a $`J_{12}=0`$ (singlet) state to a $`J_3=1/2`$ state. These two possible methods of adding the angular momentum to obtain a $`J=1/2`$ state are exactly the degeneracy of the algebra. The four $`J=1/2`$ states are:
$`|0_L`$ $`=`$ $`\{\begin{array}{c}|\frac{1}{2},0,0=|0,0|\frac{1}{2},\frac{1}{2}=\frac{1}{\sqrt{2}}\left(|010|100\right)\hfill \\ |\frac{1}{2},0,1=|0,0|\frac{1}{2},\frac{1}{2}=\frac{1}{\sqrt{2}}\left(|011|101\right)\hfill \end{array}`$ (79)
$`|1_L`$ $`=`$ $`\{\begin{array}{c}|\frac{1}{2},1,0=\frac{1}{\sqrt{3}}\left(\sqrt{2}|1,1|\frac{1}{2},\frac{1}{2}+|0,0|\frac{1}{2},\frac{1}{2}\right)=\frac{1}{\sqrt{6}}\left(2|001+|010+|100\right)\hfill \\ |\frac{1}{2},1,1=\frac{1}{\sqrt{3}}\left(\sqrt{2}|1,1|\frac{1}{2},\frac{1}{2}|1,0|\frac{1}{2},\frac{1}{2}\right)=\frac{1}{\sqrt{6}}\left(2|110|101|011\right)\hfill \end{array}`$ (82)
where in the first column we indicated the grouping forming a logical qubit; in the second we used the $`|J,\lambda ,\mu `$ notation; in the third we used tensor products of the form $`|J_{12},m_{J_{12}}|J_3,m_{J_3}`$; and in the fourth the states are expanded in terms the single-particle $`|j=1/2,m_j=\pm 1/2`$ basis using Clebsch-Gordan coefficients. These states form a decoherence-free subsystem: the decomposition of Eqs. (74),(75) ensures that the states $`\{|\frac{1}{2},0,0,|\frac{1}{2},0,1\}`$ are acted upon identically, and so are the states $`\{|\frac{1}{2},1,0,|\frac{1}{2},1,1\}`$. Thus information of a qubit $`\alpha |0_L+\beta |1`$ should be encoded into these states as
$$\rho =\underset{J}{\underset{}{\left[|\frac{1}{2}\frac{1}{2}|\right]}}\underset{\lambda }{\underset{}{\left[(\alpha ^{}|0_L+\beta ^{}|1_L)(\alpha 0_L|+\beta 1_L|)\right]}}\underset{\mu }{\underset{}{\left[\gamma _{00}|00|+\gamma _{01}|01|\gamma _{01}+\gamma _{10}|10|+\gamma _{11}|11|\right]}}.$$
(83)
where $`\gamma _{ij}`$ form the components of a valid density matrix (unity trace and positive). Using Eq. (74) It follows that each of the $`𝐒_\alpha `$’s act on $`\rho `$ in such a manner that only the $`\lambda `$ component is changed. Indeed, the $`𝐒_\alpha `$’s act like a corresponding $`\sigma _\alpha `$ in the $`\mu `$-basis because this basis is two-dimensional, and $`\sigma _\alpha `$ are the two dimensional irreducible representations of $`su(2)`$. These considerations are illustrated in detail for the exchange interaction in Sec. VII C.
The smallest decoherence-free subspace (as opposed to subsystem) supporting a full encoded qubit comes about for $`n=4`$. Subspaces for the SCD mechanism correspond to the degeneracy of the zero total angular momentum eigenstates (there are also two decoherence-free subsystems with degeneracy $`1`$ and $`3`$). This subspace is spanned by the states:
$`|0_L`$ $`=`$ $`|0,0,0=|0,0|0,0={\displaystyle \frac{1}{2}}|(|01|10)(|01|10)`$ (84)
$`|1_L`$ $`=`$ $`|0,1,0={\displaystyle \frac{1}{\sqrt{3}}}(|1,1|1,1|1,0|1,0+|1,1|1,1)`$ (85)
$`=`$ $`{\displaystyle \frac{1}{\sqrt{12}}}(2|0011+2|1100|0101|1010|0110|1001).`$ (86)
The notation is the same as in Eq. (82), except that in the second column we used the notation $`|J_{12},m_{J_{12}}|J_{34},m_{J_{34}}`$ which makes it easy to see how the angular momentum is added.
As seen from Eqs. (82) and (86), there is a variety of useful bases which one can choose for the SCD-DFSs. We now show how the generic basis $`|J,\lambda ,\mu `$ can be given both a graphical and an angular momentum interpretation. Consider the addition of angular momentum as more particles are included, similar to the construction we used in the WCD case. To construct the $`n`$ qubit SCD-DFS for a specific $`J`$, denoted in this section as DFS<sub>n</sub>($`J`$), one takes DFS<sub>n-1</sub>($`J1/2`$) and DFS<sub>n-1</sub>($`J+1/2`$) and uses the angular momentum addition rules to add another qubit ($`j=1/2`$). Table (I) presents the degeneracy of the $`J^{\mathrm{th}}`$ irreducible representation for $`n`$ qubits. The entries are obtained just as in Pascal’s triangle, except that half of the triangle \[the bottom according to the scheme of Table (I)\] is missing.
Table (I) demonstrates how the degeneracies of the $`(n1)`$-qubit $`J\pm 1/2`$ irreducible representations (irreps), i.e., the dimensions of DFS$`{}_{n1}{}^{}(J\pm 1/2)`$, add to determine the dimension of DFS$`{}_{n}{}^{}(J)`$. This method of addition of the angular momentum leads to a natural interpretation of the $`|J,\lambda ,\mu `$ basis for the SCD-DFSs which we now present.
Define the partial collective operators
$$𝐒_\alpha ^k𝐒_\alpha ^{(1,2,\mathrm{},k)}=\underset{i=1}{\overset{k}{}}\sigma _\alpha ^i.$$
(87)
This can be used to find a set of mutually commuting operators for the SCD-DFSs: the partial total angular momentum operators
$$(𝐒^k)^2=\underset{\alpha =x,y,z}{}\left(𝐒_\alpha ^k\right)^2.$$
(88)
As shown in Appendix A:
$$[(𝐒^k)^2,(𝐒^l)^2]=0k,l.$$
(89)
Thus the $`\{(𝐒^k)^2\}`$ can be used to label the SCD-DFSs by their eigenvalues $`J_k`$.
In order to make the connection between the addition of angular momentum and the Dicke states one should, however, use
$$𝐬_\alpha ^k\underset{i=1}{\overset{k}{}}\frac{1}{2}\sigma _\alpha ^i=\frac{1}{2}𝐒_\alpha ^k.$$
(90)
With this definition $`(𝐬^k)^2=_\alpha (𝐬_\alpha ^k)^2`$ is just the operator whose eigenvalue for the $`J^{\mathrm{th}}`$ irrep of the $`k`$ qubit case is $`J_k(J_k+1)`$. We label the basis determined by the eigenvalues of $`(𝐬^k)^2`$ by
$$|J_1,J_2,J_3,\mathrm{},J_{n1},J;m_J,$$
(91)
where
$$(𝐬^k)^2|J_1,J_2,J_3,\mathrm{},J_{n1},J;m_J=J_k(J_k+1)|J_1,J_2,J_3,\mathrm{},J_{n1},J;m_J,$$
(92)
and where for consistency with the $`|J,\lambda ,\mu `$ notation we use $`J`$ for $`J_n`$. As in the WCD case, the degeneracy which leads to the SCD-DFS can be put into a one-to-one correspondence with a graphical representation of the addition of angular momentum. Here, however, each step does not simply correspond to adding a $`|0`$ or $`|1`$ state but instead corresponds to combining the previous spin $`J`$ particle with a spin $`1/2`$ particle to create a $`J+1/2`$ or $`|J1/2|`$ particle (note the absolute value so that the total spin is positive). In the graphical representation of Fig. (4) the horizontal axis counts qubits, and the vertical axis corresponds to the total angular momentum $`J_i`$ up to the $`i^{\mathrm{th}}`$ qubit \[note the similarity to Table (I)\]. Each SCD-DFS state then corresponds to a path constructed by successively moving up or down $`1/2`$ unit of angular momentum, starting from a single qubit with $`J_1=1/2`$ . For example, the two DFS<sub>3</sub>($`1/2`$) states are $`\{|1/2,0,1/2;\pm 1/2,|1/2,1,1/2;\pm 1/2\}`$ (corresponding, respectively, to the paths “up,down,up” and “up,up,down” and $`m_{J_3=1/2}=\pm 1/2`$), and the two DFS<sub>4</sub>($`0`$) states are $`\{|1/2,0,1/2,0;0,|1/2,1,1/2,0;0\}`$. Clearly, the set of paths $`𝐉_n\{J_1,J_2,J_3,\mathrm{},J_{n1},J_n\}`$ with fixed $`J_n`$ counts the degeneracy of DFS$`{}_{n}{}^{}(J_n)`$. Therefore we can identify the general degeneracy index $`\lambda `$ (of $`|J,\lambda ,\mu `$) with $`𝐉_n`$. Similarly, the dimensionality index $`\mu `$ can now be identified with $`m_{J_n}`$. Finally, as claimed above $`J`$ is just the final $`J_n`$.
### A The Stabilizer and Error Correction Properties
Note from Eq. (87) that the system operators $`𝐒_\alpha =𝐒_\alpha ^n`$. Therefore they can only affect the last component $`|J_n;m_{J_n}`$ of the DFS states. By the identification of the degeneracy index $`\lambda `$ with the paths $`\{J_1,\mathrm{},J_{n1},J\}`$, and from the general expression (74) for the action of the $`𝐒_\alpha `$ , we know that $`𝐒_\alpha `$ acts only on the dimensionality component:
$$𝐒_\alpha |J_1,\mathrm{},J_{n1},J;m_J=|J_1,\mathrm{},J_{n1},J(𝐏_\alpha |m_J),$$
(93)
where the $`𝐏_\alpha `$ are a $`2J+1`$ dimensional representation of $`su(2)`$ acting directly on the $`|m_J`$ components of the DFS. The corresponding DFS stabilizer is
$$𝐃(\stackrel{}{v})=𝐃(v_x,v_y,v_z)=\mathrm{exp}\left[\underset{\alpha =x,y,z}{}v_\alpha (𝐒_\alpha 𝐈𝐏_\alpha )\right].$$
(94)
For the $`J=0`$ DFSs this reduces to all collective rotations+contractions :
$`𝐃(\stackrel{}{v})=\mathrm{exp}\left[{\displaystyle \underset{\alpha =x,y,z}{}}v_\alpha 𝐒_\alpha \right]={\displaystyle \underset{i=1}{\overset{n}{}}}\mathrm{exp}\left[\stackrel{}{v}\stackrel{}{\sigma }_i\right]=\left[𝐈\mathrm{cos}\stackrel{}{v}+{\displaystyle \frac{\stackrel{}{\sigma }\stackrel{}{v}}{\stackrel{}{v}}}\mathrm{sin}\stackrel{}{v}\right]^n,`$
where $`v(_\alpha v_\alpha ^2)^{1/2}`$ may be complex. Thus DFS$`{}_{n}{}^{}(0)`$ protects against all processes described by Kraus operators that are linear combinations of collective rotations+contractions $`\mathrm{exp}\left[\stackrel{}{v}\stackrel{}{\sigma }\right]`$. The situation for $`J0`$ is more complicated to calculate analytically.
Let us now comment briefly on the error-correction and detection properties of DFS$`{}_{n}{}^{}(0)`$: The stabilizer elements are tensor products of identical 1-qubit operators, including the following elements of the Pauli group: $`𝐗^n`$, $`𝐘^n`$ and $`𝐙^n`$. Thus, for any odd–multiple $`2k1<n`$ of single qubit errors $`𝐗`$, $`𝐘`$ and $`𝐙`$ there is an element in the stabilizer that anticommutes with it: The code can detect any such error. The $`J=0`$ SCD-DFS is an error correcting code of distance $`2`$.
### B Nontrivial Operations
Are there any single-qubit operators which preserve a SCD-DFS (and thus allow for nontrivial operations on the DFS)? There are no nontrivial single-qubit operators that commute with all $`𝐒_\alpha `$ operators, since
$$[𝐒_\alpha ,\sigma _\beta ^j]=\underset{i}{}[\sigma _\alpha ^i,\sigma _\beta ^j]=i\underset{i}{}\delta _{ij}\epsilon _{\alpha \beta \gamma }\sigma _\gamma ^i$$
(95)
which vanishes iff $`\alpha =\beta `$. Therefore there are no single-qubit operators which preserve all SCD-DFSs simultaneously.
As for two-qubit operators, the only such Hermitian operators which commute with the $`𝐒_\alpha `$ are those that are proportional to the exchange interaction \[Eq. (47)\]: $`𝐄_{ij}|k_i|l_j=|l_i|k_j`$, where $`i,j`$ label the qubits acted upon . In both the single- and two-qubit cases, there could be additional operators in the generalized commutant $`𝒯`$ (e.g., for $`n=4`$ qubits there is an operator which mixes the different $`J`$’s and preserves DFS<sub>4</sub>($`0`$): $`𝐓=|J=1,\lambda _1,\mu _1J=2,\lambda _1,\mu _1|+H.c`$.). We will not be concerned with such operations as they are not needed in order to demonstrate universality, and since we will show that the exchange operator is sufficient for any SCD-DFS. Our task is thus to show that exchange interactions alone suffice to generate the entire $`SU(N)`$ group on each $`N`$-dimensional DFS, in the SCD case.
### C Quantum Computation on the $`n=3`$ and $`n=4`$ qubit SCD-DFS
We begin our discussion of universal quantum computation on SCD-DFSs by examining the simplest SCD-DFS which supports encoding of quantum information: the $`n=3`$ decoherence-free subsystem. We label these states as in Eq. (82) by $`|J,\lambda ,\mu `$. Recall that the $`J=3/2`$ irrep is not degenerate and the $`J=1/2`$ irrep has degeneracy $`2`$. The $`J=3/2`$ states can be written as $`|\frac{3}{2},0,\mu `$, with $`\mu =m_J=\pm 3/2,\pm 1/2`$. Since the action of exchange does not depend on $`\mu `$ (recall that it affects paths, i.e., the $`\lambda `$ component only) it suffices to consider the action on the representative $`\mu =3/2`$ only: $`|111`$. Let us then explicitly calculate the action of exchanging the first two physical qubits on this state and the four $`J=1/2`$ states. Using Eq. (82):
$`𝐄_{12}|{\displaystyle \frac{3}{2}},0,{\displaystyle \frac{3}{2}}`$ $`=`$ $`𝐄_{12}|111=|{\displaystyle \frac{3}{2}},0,{\displaystyle \frac{3}{2}}`$ (96)
$`𝐄_{12}|{\displaystyle \frac{1}{2}},0,0`$ $`=`$ $`𝐄_{12}{\displaystyle \frac{1}{\sqrt{2}}}\left(|010|100\right)={\displaystyle \frac{1}{\sqrt{2}}}\left(|100|010\right)=|{\displaystyle \frac{1}{2}},0,0`$ (97)
$`𝐄_{12}|{\displaystyle \frac{1}{2}},0,1`$ $`=`$ $`𝐄_{12}{\displaystyle \frac{1}{\sqrt{2}}}\left(|011|101\right)={\displaystyle \frac{1}{\sqrt{2}}}\left(|101|011\right)=|{\displaystyle \frac{1}{2}},0,1`$ (98)
$`𝐄_{12}|{\displaystyle \frac{1}{2}},1,0`$ $`=`$ $`𝐄_{12}{\displaystyle \frac{1}{\sqrt{6}}}\left(2|001+|010+|100\right)=|{\displaystyle \frac{1}{2}},1,0`$ (99)
$`𝐄_{12}|{\displaystyle \frac{1}{2}},1,1`$ $`=`$ $`𝐄_{12}{\displaystyle \frac{1}{\sqrt{6}}}\left(2|110|101|011\right)=|{\displaystyle \frac{1}{2}},1,1.`$ (100)
Focusing just on the $`J=1/2`$ states, the exchange action on $`|\lambda |\mu `$ can thus be written as:
$$𝐄_{12}=\sigma _z𝐈.$$
(101)
Since the action of the $`𝐒_\alpha `$ operators on the $`J=1/2`$ states is $`𝐈_{n_{1/2}}gl(2)`$ according to Eq. (74), this explicit form for $`𝐄_{12}`$ confirms that is has the expected structure of operators in the commutant of the algebra spanned by the $`𝐒_\alpha `$. It can also be seen that quantum information should be encoded in the $`|\lambda `$ component, as discussed before Eq. (83).
Using similar algebra it is straightforward to verify that the effect of the three possible exchanges on the $`n=3`$ DFS states is given by:
$$𝐄_{12}=\left(\begin{array}{ccc}1& 0& 0\\ 0& 1& 0\\ 0& 0& 1\end{array}\right)𝐄_{23}=\left(\begin{array}{ccc}1& 0& 0\\ 0& \frac{1}{2}& \frac{\sqrt{3}}{2}\\ 0& \frac{\sqrt{3}}{2}& \frac{1}{2}\end{array}\right)𝐄_{13}=\left(\begin{array}{ccc}1& 0& 0\\ 0& \frac{1}{2}& \frac{\sqrt{3}}{2}\\ 0& \frac{\sqrt{3}}{2}& \frac{1}{2}\end{array}\right),$$
(102)
where the rows and columns of these matrices are labelled by the basis elements $`\{|J=3/2,\lambda =0,|J=1/2,\lambda =0,|J=1/2,\lambda =1\}`$. As expected from general properties of the commutant, the exchange operators do not mix the different $`J`$ irreps. Now,
$`{\displaystyle \frac{1}{3}}(𝐄_{12}+𝐄_{13}+𝐄_{23})`$ $`=`$ $`\left(\begin{array}{ccc}1& 0& 0\\ 0& 0& 0\\ 0& 0& 0\end{array}\right)`$ (106)
$`{\displaystyle \frac{1}{2}}(𝐄_{12}+𝐄_{13}+𝐄_{23})`$ $`=`$ $`\left(\begin{array}{ccc}0& 0& 0\\ 0& 1& 0\\ 0& 0& 1\end{array}\right)`$ (110)
$`{\displaystyle \frac{1}{\sqrt{3}}}(𝐄_{13}𝐄_{23})`$ $`=`$ $`\left(\begin{array}{ccc}0& 0& 0\\ 0& 0& 1\\ 0& 1& 0\end{array}\right),`$ (114)
showing that the last two linear combinations of exchanges look like the Pauli $`\sigma _z`$ and $`\sigma _x`$ on DFS$`{}_{3}{}^{}(1/2)`$. Using a standard Euler angle construction it is thus possible to perform any $`SU(2)`$ gate on this DFS. Moreover, it is possible to act independently on DFS$`{}_{3}{}^{}(3/2)`$ and DFS$`{}_{3}{}^{}(1/2)`$. In other words, we can perform $`U(1)`$ on DFS$`{}_{3}{}^{}(3/2)`$ alone, and $`SU(2)`$ on DFS$`{}_{3}{}^{}(1/2)`$ alone. Note, however, that at this point we cannot yet claim universal quantum computation on a register composed of clusters of DFS$`{}_{3}{}^{}(J)`$’s ($`J`$ constant) because we have not shown how to couple such clusters.
For $`n=4`$ the Hilbert space splits up into one $`J=2`$-irrep \[DFS$`{}_{4}{}^{}(2)`$\], three $`J=1`$-irreps \[DFS$`{}_{4}{}^{}(1)`$\], and two $`J=0`$-irreps \[DFS$`{}_{4}{}^{}(0)`$\] – see Table (I). Direct calculation of the effect of exchange on these DFSs shows that we can independently perform $`su(1)`$ (i.e. zero), $`su(3)`$, and $`su(2)`$. In particular, we find that :
$$𝐗=\frac{1}{\sqrt{3}}(𝐄_{23}𝐄_{13})𝐘=\frac{i}{2\sqrt{3}}[𝐄_{23}𝐄_{13},𝐄_{34}]𝐙=\frac{i}{2}[𝐘,𝐗]=𝐄_{12}$$
(115)
act as the corresponding $`su(2)`$ Pauli operators on DFS$`{}_{4}{}^{}(0)`$ only. Further, the following operators act independently on the $`J=1`$-irreps (rows and columns are labelled by $`\lambda =0,1,2`$. The action occurs simultaneously on all three $`\mu `$ components corresponding to a given $`\lambda `$):
$`𝐘_{13}`$ $`=`$ $`{\displaystyle \frac{3i}{2\sqrt{2}}}[𝐄_{12},𝐄_{34}]=\left(\begin{array}{ccc}0& 0& i\\ 0& 0& 0\\ i& 0& 0\end{array}\right),𝐗_{13}={\displaystyle \frac{i}{2}}[𝐄_{12},𝐘_{13}]=\left(\begin{array}{ccc}0& 0& 1\\ 0& 0& 0\\ 1& 0& 0\end{array}\right),`$ (122)
$`𝐙_{13}`$ $`=`$ $`{\displaystyle \frac{i}{2}}[𝐘_{13},𝐗_{13}]=\left(\begin{array}{ccc}1& 0& 0\\ 0& 0& 0\\ 0& 0& 1\end{array}\right),𝐘_{23}={\displaystyle \frac{2i}{\sqrt{3}}}[𝐄_{23},𝐙_{13}]=\left(\begin{array}{ccc}0& 0& 0\\ 0& 0& i\\ 0& i& 0\end{array}\right).`$ (129)
These operators clearly generate $`su(3)`$, and hence we have an independent $`SU(3)`$ action on DFS$`{}_{4}{}^{}(1)`$.
### D Universal Quantum Computation on the $`n5`$ qubit SCD-DFSs
We are now ready to prove our central result: that using only the two-body exchange Hamiltonians every unitary operation can be performed on a SCD-DFS. More specifically:
Theorem 5— For any $`n2`$ qubits undergoing strong collective decoherence, there exist sets of Hamiltonians $`𝖧_J^n`$ obtained from exchange interactions only via scalar multiplication, addition, Lie-commutator and unitary conjugation, acting as $`su(d_J)`$ on the DFS corresponding to the eigenvalue $`J`$. Furthermore each set acts independently on this DFS only (i.e., with zeroes in the matrix representation corresponding to their action on the other DFSs).
In preparation for the proof of this result let us note several useful facts:
(i) The exchange operators do not change the value of $`m_J`$, because they are in the commutant of $`𝒜=\{S_\alpha \}`$ \[and recall Eq. (93)\]. Therefore in order to evaluate the action of the exchange operators on the different DFS$`{}_{n}{}^{}(J)`$ ($`n`$ given) it is convenient to fix $`m_J`$, and in particular to work in the basis given by the maximal $`m_J`$ value ($`m_J=J`$). Expressions for these “maximal” states in terms of $`|J_1,J_2,\mathrm{},J_{n2};m_J`$ and the single qubit states of the last two qubits are given in Appendix B.
(ii) Every $`(𝐬^k)^2`$ can be written as a sum of exchange operators and the identity operation . This follows from Eq. (A1) and noting that the exchange operator can be expanded as
$$𝐄_{ij}=\frac{1}{2}\left(𝐈+\sigma _x^i\sigma _x^j+\sigma _y^i\sigma _y^j+\sigma _z^i\sigma _z^j\right),$$
(130)
so that:
$$(𝐬^k)^2=k\left(1\frac{k}{4}\right)𝐈+\frac{1}{2}\underset{ij=1}{\overset{k}{}}𝐄_{ij}.$$
(131)
Thus $`(𝐬^k)^2`$ is a Hamiltonian which is at our disposal.
We are now ready to present our proof by induction. Recall the DFS-dimensionality formula for $`n_J`$, Eq. (76). We assume that it is possible to perform $`su(n_J)`$ on each of the different DFS$`{}_{n1}{}^{}(J)`$ independently using only exchange operators and the identity Hamiltonian. Our construction above proves that this is true for $`3`$ and $`4`$ qubits. The assumption that the actions we can perform can be performed independently translates into the ability to construct Hamiltonians which annihilate all of the DFSs except a desired one on which they act as $`su(n_J)`$.
As in the WCD case a specific DFS$`{}_{n}{}^{}(J)`$ of dimension $`n_J`$ splits into states which are constructed by the subtraction of angular momentum from DFS$`{}_{n1}{}^{}(J+1/2)`$ (T-states), or by the addition of angular momentum to DFS$`{}_{n1}{}^{}(J1/2)`$ (B-states) \[see Fig. (5)\]. Performing $`su(n_{J+1/2})`$ on DFS$`{}_{n1}{}^{}(J+1/2)`$ will simultaneously act on DFS$`{}_{n}{}^{}(J)`$ and DFS$`{}_{n}{}^{}(J+1)`$. In other words, $`su(n_{J+1/2})`$ on DFS$`{}_{n1}{}^{}(J+1/2)`$ acts on both the B-states of DFS$`{}_{n}{}^{}(J+1)`$ and on the T-states of DFS$`{}_{n}{}^{}(J)`$. We split the proof into three steps. In the first step we obtain an $`su(2)`$ set of operators which acts only on DFS$`{}_{n}{}^{}(J)`$ and mixes particular B- and T-states. In the second step we expand the set of operators which mix B- and T-states to cover all possible $`su(2)`$ algebras between any two B- and T-states. Finally, in the third step we apply a Mixing Lemma which shows that we can obtain the full $`su(n_J)`$ (i.e., also mix B-states and mix T-states).
#### 1 T- and B-Mixing
There are two simple instances where there is no need to show independent action in our proof: (i) The (upper) $`J=n/2`$ -irrep is always $`1`$-dimensional, so the action on it is always trivial (i.e., the Hamiltonian vanishes and hence the action is independent by definition); (ii) For odd $`n`$ the “lowest” DFS$`{}_{n}{}^{}(1/2)`$ is acted upon independently by the $`su(n_0)`$ from DFS$`{}_{n1}{}^{}(0)`$ \[i.e., $`su(n_0)`$ cannot act “downward”\]. In order to facilitate our construction we extend the notion of T and B-states one step further in the construction of the DFS. TB-states are those states which are constructed from T-states on ($`n1`$)-qubits and from the B-states on $`n`$-qubit states \[see Fig. (5)\]. Similarly we can define the BT, TT, and BB-states:
$`|\mathrm{TT}`$ $``$ $`|J_1,\mathrm{},J_{n3},J_n+1,J_n+{\displaystyle \frac{1}{2}},J_n;m_J=J_n=\begin{array}{cc}\hfill & \\ & \hfill \end{array}`$ (134)
$`|\mathrm{BT}`$ $``$ $`|J_1,\mathrm{},J_{n3},J_n,J_n+{\displaystyle \frac{1}{2}},J_n;m_J=J_n=`$ (135)
$`|\mathrm{TB}`$ $``$ $`|J_1,\mathrm{},J_{n3},J_n,J_n{\displaystyle \frac{1}{2}},J_n;m_J=J_n=`$ (136)
$`|\mathrm{BB}`$ $``$ $`|J_1,\mathrm{},J_{n3},J_n1,J_n{\displaystyle \frac{1}{2}},J_n;m_J=J_n=\begin{array}{cc}& \hfill \\ \hfill & \end{array}.`$ (139)
Every DFS$`{}_{n}{}^{}(J)`$ can be broken down into a direct sum of TT, BT, TB, and BB-states; e.g., as seen in Fig. (4), in DFS$`{}_{6}{}^{}(1)`$ there are 1 TT, 3 TB, 3 BT and 2 BB states. Note that for $`J=n/21`$ there are no TT-states, for $`J=0`$ there are no BB and BT-states, for $`J=1/2`$ there are no BB-states, and otherwise there are as many TB as there are BT states,
At this point it is useful to explicitly give the action of exchange on the last two qubits of a SCD-DFS. Using Eq. (B16) we find (assuming the existence of the given states, i.e., $`n`$ large enough and $`J`$ not too large) the representation
$$𝐄_{n,n1}=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& \mathrm{cos}(\theta _{J+1})& \mathrm{sin}(\theta _{J+1})& 0\\ 0& \mathrm{sin}(\theta _{J+1})& \mathrm{cos}(\theta _{J+1})& 0\\ 0& 0& 0& 1\end{array}\right)\begin{array}{c}\mathrm{TT}\\ \mathrm{BT}\\ \mathrm{TB}\\ \mathrm{BB}\end{array}$$
(140)
where $`\mathrm{tan}(\theta _J)=2\sqrt{J(J+1)}`$. Thus exchange acts to transform the BT and TB states entering a given DFS into linear combinations of one another, while leaving invariant the BB and TT states.
Let us now consider the action of $`su(n_{J1/2})`$ from DFS$`{}_{n1}{}^{}(J1/2)`$ \[see Fig. (5)\]. It acts on DFS$`{}_{n}{}^{}(J1)`$ and DFS$`{}_{n}{}^{}(J)`$ simultaneously. However, since the T-states of DFS$`{}_{n}{}^{}(J1)`$ and the B-states of DFS$`{}_{n}{}^{}(J)`$ share the same set of quantum numbers $`\{J_1,\mathrm{},J_{n1}\}`$, the action of the $`su(n_{J1/2})`$ operators is identical on these two sets of states.
We first deal with the case where the number of BT-states of DFS$`{}_{n}{}^{}(J)`$ is greater than 1. As can be inferred from Fig. (4), this condition corresponds to $`J<n/21`$ and $`n>4`$. We will separately deal with the $`J=n/21`$ case at the end of the proof. Let $`|a`$ and $`|b`$ be any two orthogonal BT-states of DFS$`{}_{n}{}^{}(J)`$ (i.e., states differing only by the paths on the first $`n2`$ qubits). Corresponding to these are $`\{|a^{},|b^{}\}`$: a pair of orthogonal BT-states of DFS$`{}_{n}{}^{}(J)`$. One of the elements in $`su(n_{J1/2})`$ is the traceless operator $`𝐂=|aa||bb|`$, which we have at our disposal by the induction hypothesis. Consider $`i[𝐄_{n,n1},𝐂]`$: since $`𝐄_{n,n1}`$ acts as identity on BB states, even though $`𝐂`$ has an action on DFS$`{}_{n}{}^{}(J1)`$ the commutator acting on the BB states of DFS$`{}_{n}{}^{}(J1)`$ vanishes. The action of $`i[𝐄_{n,n1},𝐂]`$ on the BT and TB states can be calculated by observing, using Eq. (140), that the matrix representations of $`𝐂`$ and $`𝐄_{n,n1}`$ are, in the ordered $`\{|a^{},|b^{},|a,|b\}`$ basis:
$`𝐂`$ $`=`$ $`\mathrm{diag}(0,0,1,1)={\displaystyle \frac{1}{2}}\left(𝐈\sigma _z\sigma _z\sigma _z\right)`$ (141)
$`𝐄_{n,n1}`$ $`=`$ $`\left(\begin{array}{cccc}\mathrm{cos}(\theta _J)& 0& \mathrm{sin}(\theta _J)& 0\\ 0& \mathrm{cos}(\theta _J)& 0& \mathrm{sin}(\theta _J)\\ \mathrm{sin}(\theta _J)& 0& \mathrm{cos}(\theta _J)& 0\\ 0& \mathrm{sin}(\theta _J)& 0& \mathrm{cos}(\theta _J)\end{array}\right)=\mathrm{cos}(\theta _J)\sigma _z𝐈+\mathrm{sin}(\theta _J)\sigma _x𝐈.`$ (146)
This yields:
$$i[𝐄_{n,n1},𝐂]=\mathrm{sin}(\theta _J)\sigma _y\sigma _z=i\mathrm{sin}(\theta _J)\left(|aa^{}|+|a^{}a|+|bb^{}||b^{}b|\right).$$
(147)
Now let $`|c`$ be a TT-state of DFS$`{}_{n}{}^{}(J)`$. Such a state always exists unless $`J=n/21`$, which is covered at the end of the proof. Then there is an operator $`𝐃=|a^{}a^{}||cc|`$ in $`su(n_{J+1/2})`$.<sup>‡‡</sup><sup>‡‡</sup>‡‡ We need to subtract $`|cc|`$ in order to obtain a traceless operator. It follows that:
$$𝐗_{aa^{}}\frac{1}{\mathrm{sin}(\theta _J)}i[i[𝐄_{n,n1},𝐂],𝐃]=|aa^{}|+|a^{}a|,$$
(148)
acts like an encoded $`\sigma _x`$ on $`|a`$ and $`|a^{}`$ and annihilates all other states. Further, one can implement the commutator
$$𝐘_{aa^{}}=i[𝐗_{aa^{}},𝐃]=i\left(|aa^{}||a^{}a|\right),$$
(149)
which acts like an encoded $`\sigma _y`$ on $`|a`$ and $`|a^{}`$. Finally, one can construct $`𝐙_{aa^{}}=i[𝐗_{aa^{}},𝐘_{aa^{}}]=|aa||a^{}a^{}|`$. Thus we have shown that for $`J<n/21`$ we can validly (using only exchange Hamiltonians) perform $`su(2)`$ operations between $`|a`$, a specific B-state and $`|a^{}`$, its corresponding T-state, on DFS$`{}_{n}{}^{}(J)`$ only.
#### 2 Extending the $`su(2)`$’s
We now show that by using the operation of conjugation by a unitary we can construct $`su(2)`$ between any two B and T-states. To see this recall Eq. (32), which allows one to take a Hamiltonian $`𝐇`$ and turn it via conjugation by a unitary gate into the new Hamiltonian $`𝐇_{\mathrm{eff}}=\mathrm{𝐔𝐇𝐔}^{}`$. By the induction hypothesis we have at our disposal every $`SU`$ gate which acts on the T-states of DFS$`{}_{n}{}^{}(J)`$ \[and simultaneously acts on the B-states of DFS$`{}_{n}{}^{}(J+1)`$\] and also every $`SU`$ gate which acts on the B-states of DFS$`{}_{n}{}^{}(J)`$ \[and simultaneously acts on the T-states of DFS$`{}_{n}{}^{}(J1)`$\]. Above we have shown how to construct $`𝐗`$, $`𝐘`$, and $`𝐙`$ operators between specific T- and B-states: $`|a^{}`$ and $`|a`$. Let $`|i^{}`$ and $`|i`$ be some other T- and B-states of DFS$`{}_{n}{}^{}(J)`$, respectively. Then we have at our disposal the gate $`𝐏_{i^{}i}=|a^{}i^{}|+|i^{}a^{}|+|ai|+|ia|+𝐎`$ where $`𝐎`$ is an operator which acts on a DFS other than DFS$`{}_{n}{}^{}(J)`$ (included to make $`𝐏_{i^{}i}`$ an $`SU`$ operator). It is simple to verify that
$$𝐗_{i^{}i}=𝐏_{i^{}i}𝐗_{aa^{}}𝐏_{i^{}i}^{}=|i^{}i|+|ii^{}|,$$
(150)
which acts as an encoded $`\sigma _x`$ between $`|i^{}`$ and $`|i`$. Note that because $`𝐗_{aa^{}}`$ only acts on DFS$`{}_{n}{}^{}(J)`$, $`𝐗_{i^{}i}`$ will also only act on the same DFS. Similarly one can construct $`𝐘_{i^{}i}=𝐏_{i^{}i}𝐘_{aa^{}}𝐏_{i^{}i}^{}`$ and $`𝐙_{i^{}i}=𝐏_{i^{}i}𝐙_{aa^{}}𝐏_{i^{}i}^{}`$ which act, respectively, as encoded $`\sigma _y`$ and $`\sigma _z`$ on $`|i^{}`$ and $`|i`$. Thus we have shown that one can implement every $`su(2)`$ between any two T- and B-states in DFS$`{}_{n}{}^{}(J)`$. Each of these $`su(2)`$ operations is performed independently on DFS$`{}_{n}{}^{}(J)`$.
#### 3 Mixing T- and B-States
Next we use a Lemma proved in Appendix C:
Mixing Lemma: Given is a Hilbert space $``$ $`=_1_2`$ where $`dim_j=`$ $`n_j`$. Let $`\{|i_1\}`$ and $`\{|i_2\}`$ be orthonormal bases for $`_1`$ and $`_2`$ respectively. If one can implement the operators $`𝐗_{i_1i_2}=|i_1i_2|+|i_2i_1|`$, $`𝐘_{i_1i_2}=i|i_1i_2|i|i_2i_1|`$, and $`𝐙_{i_1i_2}=|i_1i_1||i_2i_2|`$, then one can implement $`su(n_1+n_2)`$ on $``$.
Above we have explicitly shown that we can obtain every $`𝐗_{i_1i_2}`$, $`𝐘_{i_1i_2}`$, and $`𝐙_{i_1i_2}`$ acting independently on DFS$`{}_{n}{}^{}(J)`$. Thus direct application of the Mixing Lemma tells us that we can perform $`su(n_J)`$ independently on this DFS.
Special case of $`J=n/21`$: We have neglected DFS$`{}_{n}{}^{}(n/21)`$ because it did not contain two different BT-states (nor a TT) state. The dimension of this DFS is $`n1`$. We now show how to perform $`su(n1)`$ on this DFS using the fact that we have already established $`su(n_{J=n/22})`$ on DFS$`{}_{n}{}^{}(n/22`$). First, note that by the induction hypothesis we can perform $`su(n_{J=n/23/2})`$ independently on DFS$`{}_{n1}{}^{}(n/23/2)`$. As above, this action simultaneously affects DFS$`{}_{n}{}^{}(n/21)`$ and DFS$`{}_{n}{}^{}(n/22)`$. However, since we can perform $`su(n_{J=n/22})`$ on DFS$`{}_{n}{}^{}(n/22)`$, we can subtract out the action of $`su(n_{J=n/23/2})`$ on DFS$`{}_{n}{}^{}(n/22)`$. Thus we can obtain $`su(n_{J=n/23/2})`$ on all of the B-states of DFS$`{}_{n}{}^{}(n/21)`$. But the exchange operator $`𝐄_{n,n1}`$ acts to mix the B-states with the single T-state of DFS$`{}_{n}{}^{}(n/21)`$. Thus we can construct an $`su(2)`$ algebra between that single-T state and a single B-state in a manner directly analogous to the above proof for $`J<n/21`$. Finally, by the Enlarging Lemma it follows that we can obtain $`su(n1)`$ on DFS$`{}_{n}{}^{}(n/21)`$.
This concludes the proof that the exchange interaction is universal independently on each of the different strong-collective-decoherence DFSs.
### E State Preparation and Measurement on the Strong Collective Decoherence DFS
At first glance it might seem difficult to prepare pure states of a SCD-DFS, because these states are nontrivially entangled. However, it is easy to see that every DF subspace contains a state which is a tensor product of singlet states:
$$|0_D=\left(\frac{1}{\sqrt{2}}\right)^{n/2}_{j=1}^{n/2}(|01|10),$$
(151)
because these states have zero total angular momentum. Thus a supply of singlet states is sufficient to prepare DF subspace states. Further, DF subsystems always contain a state which is a tensor product of a DF subspace and a pure state of the form $`|1\mathrm{}|1`$. This can be seen from Fig. (4), where the lowest path leading to a specific DFS$`{}_{n}{}^{}(J)`$ is composed of a segment passing through a DF subspace (and is thus of the form $`|0_D`$), and a segment going straight up from there to DFS$`{}_{n}{}^{}(J)`$. The corresponding state is equivalent to adding a spin-$`0`$ (DF subspace) and a spin-$`J`$ DF subsystem (the $`|J,m_J=J`$ state of the latter is seen to be made up entirely of $`|1\mathrm{}|1`$). In general, addition of a spin-$`0`$ DFS and a spin-$`J`$ DFS simply corresponds to tensoring the two states. Note, however, that addition of two arbitrary DF subsystems into a larger DFS is not nearly as simple: concatenation of two $`J0`$ DFSs does not correspond to tensoring.
Pure state preparation for a SCD-DFS can thus be as simple as the ability to produce singlet states and $`|1`$ states (it is also possible to use the $`|J,m_J=J=|0\mathrm{}|0`$ or any of the other $`|J,m_J`$ states plus singlets). Other, more complicated pure state preparation procedures are also conceivable, and the decision as to which procedure to use is clearly determined by the available resources to manipulate quantum states. The pure state preparation of singlets and computational basis states has the distinct advantage that verification of these states should be experimentally achievable. Such verification is necessary for fault-tolerant preparation .
Measurements on the SCD-DFS can be performed by using the concatenated measurement scheme detailed in the WCD-DFS discussion \[Sec. (VI D)\]. In particular, by attaching a SCD-DF subspace ancilla via concatenation, one can construct any concatenated measurement scenario. All that remains to be shown is how to perform a destructive measurement on such an ancilla. In such schemes are presented for the $`n=4`$ SCD-DF subspace which encodes a single qubit of information. We will not repeat the details of these schemes here, but note that they involve measurements of single physical-qubit observables and thus are experimentally very reasonable. Further, we note that the ability to perform a concatenated measurement scenario by concatenating an ancilla DFS composed of a single encoded-qubit, can be used to perform any possible concatenated DFS measurement scenario. As mentioned in the WCD case, the concatenated measurement procedures are fault-tolerant. Thus we have shown how to perform fault-tolerant preparation and decoding on the SCD-DFS.
## VIII Universal Fault-Tolerant Computation on Concatenated Codes
So far we have shown how to implement universal computation with local Hamiltonians on a DFS corresponding to a single block of qubits. This construction assumes that the only errors are collective. This is a very stringent symmetry requirement, which obviously becomes less realistic the larger the number of particles $`n`$ is. It is thus desirable to be able to deal with perturbations that break the collective-decoherence (permutation) symmetry. To this end we have previously studied the effect of symmetry-breaking perturbations on decoherence-free subspaces , and have proposed a concatenation method to make DFSs robust in the presence of such perturbations. This method embeds DFS blocks of four particles (each block constituting a single encoded qubit) into a QECC . The QECC in the outer layer then takes care of any single encoded-qubit errors on each of its constituent DFS-blocks; in fact the code can correct for any “leakage” error taking a state outside of the DFS, by transforming this into a single encoded qubit error on the QECC. By choosing an appropriate QECC it is thus possible to deal with any type of non-collective error on the encoded DFS-qubits. In particular, by using the “perfect” 5-qubit code it is possible to correct all independent errors between blocks of four particles.
The problem with this construction so far was, that in order to correct on the outer QECC, it is necessary to perform encoded operations on the constituent DFSs in a fault-tolerant way, using (realistic) local interactions. Specifically, it is necessary to be able to implement all single encoded-qubit operations on the DFS-qubits of the outer QECC, as well as operations between two DFS-blocks (see for details). Given that one can perform single qubit (or “qupit” for higher-dimensional DFSs) operations on each DFS-block, the only additional gate necessary to implement error-correction and universal quantum computation on a concatenated QECC-DFS, is any non-separable two-encoded-qubit gate $`𝐊`$ between any four states in the two DFS-blocks. For instance, a controlled-phase operation, which gives a phase of $`1`$ to $`|0_L0_L`$ and leaves all other states unchanged. In fact, it is sufficient to be able to perform this gate $`𝐊`$ between neighboring blocks only .
The above results give us the tools to perform single DFS-qubit (or qupit) operations on a block. To construct an encoded $`𝐊`$ between two neighboring blocks, we assume that the corresponding physical qubits are spatially close together during the switching time of the gate. Since the symmetry of collective decoherence arises from the spatial correlation of the decoherence process, we can further assume that during this switching time, both DFS-blocks couple to the same bath mode. This assumption is physically motivated by the expectation that collective decoherence occurs in the analog of the Dicke limit of quantum optics, where the qubits have small spatial separations relative to the bath correlation length . Then the two DFS-clusters temporarily form a bigger DFS and we can use the universal operations we have constructed previously on this big DFS to implement the desired gate $`𝐊`$.
This makes the concatenated QECC-DFS fully workable as a code supporting universal fault-tolerant quantum computation.
## IX Summary and Conclusions
In this paper we have settled the issue of quantum computation with realistic (few-body) means on both decoherence-free subspaces and decoherence-free (noiseless) subsystems (DFSs) for two important forms of decoherence: collective phase damping (“weak collective decoherence”), and collective phase damping plus collective dissipation (“strong collective decoherence”). This resolves an outstanding question as to whether universal computation on these physically relevant DFSs by using just 1- and 2-body Hamiltonians is possible.
The implications of this result for the usefulness of DFSs are drastic. They put the theory of DFSs on an equal footing with the theory of quantum error correction, in that the full repertoire of universal fault tolerant quantum computation is now available on DFSs for collective decoherence: the most important pertinent decoherence process. Moreover, the strict assumption of collective decoherence can be lifted by allowing for perturbing independent qubit errors. As we proposed earlier it is possible to stabilize DFSs against such errors by concatenation with a quantum error correcting code (QECC). However, to be able to implement error-correction and fault-tolerant universal computation on these concatenated codes a crucial (and so far missing) ingredient was the ability to perform encoded operations on the DFS-blocks fault-tolerantly. This paper settles that matter, showing constructively that DFSs can be made robust.
Furthermore, this paper reports on a general framework incorporating both DFSs and QECCs, and generalizes the theory of stabilizer codes to the (non-abelian) DFS-case. This framework enabled us to identify the allowed operations on a DFS and to show that these operations can be performed while maintaining a very strong form of fault-tolerance: the states remain within the DFS during the entire switching time of the gate. Our formalism should be readily applicable for other non-additive codes.
There is an interesting duality between QECCs which are designed to correct single (or greater) qubit errors and DFSs. In QECC the errors are all single body interactions. The QECC condition therefore implies that any one or two-body Hamiltonian must take codewords outside of the code space because these interactions themselves look like errors. QECCs must leave their codespace in order to perform quantum computation on the encoded operations. This means that QECCs must have gates which act much faster than the decoherence mechanism so that a perturbative treatment can be carried out. QECC can correct small errors but the price paid for this is that gates must be executed quickly (not to mention that fault-tolerant gates must also be used). DFSs on the other hand, do not have the requirement of correcting single qubit errors and we have found that a single two-body interaction (exchange) is sufficient to generate universal quantum computation fault-tolerantly. DFSs have larger errors but this allows for an economy of Hamiltonians.
As corollaries to our results on weak and strong collective decoherence two additional properties of the corresponding DFS encodings appear:
* One can work on all DFSs in parallel: Since we are able to implement $`SU(d_n)`$ on each DFS<sub>n</sub> ($`n`$=number of particles) independently, we can in principle work on all DFSs in parallel. This means that we can encode quantum information into each of the DFSs and perform calculations (possibly different) on all of them at once.
* For the strong collective decoherence case the exchange gate is asymptotically universal: It is well known that the encoding efficiency of the singlet space of the strong collective decoherence-DFS for large $`n`$ approaches unity . More precisely, let $`k`$ be the number of encoded qubits in the singlet ($`J=0`$) sector of a Hilbert space of $`n`$ qubits, then
$$\underset{n\mathrm{}}{lim}\frac{k}{n}=1\frac{3}{2}\frac{\mathrm{log}_2n}{n}.$$
(152)
We have established that the exchange gate alone (with an irrational phase) implements universal computation on each DFS and on the singlet space in particular. Thus, we find that, for large $`n`$, in order to achieve universal computation with nearly perfect efficiency, all we need to be able to perform is the exchange interaction. This result is very promising from an experimental point of view, since the exchange interaction is prevalent whenever there is a Heisenberg coupling between systems . We emphasize that regardless of the decoherence mechanism, this implies that universal quantum computation can be achieved “asymptotically” using a single gate . We conjecture that there are many more such two-body interactions which similarly provide such “asymptotic universality” on their own.
## Acknowledgments
This material is based upon work supported by the U.S. Army Research Office under contract/grant number DAAG55-98-1-0371 and NSF DMS-9971169. It is a pleasure to acknowledge helpful discussions with Drs. Dorit Aharonov and Alexei Kitaev.
## A The Partial Collective Angular Momentum Operators are a Set of Commuting Observables
We prove here that the partial collective operators $`𝐒_\alpha ^k𝐒_\alpha ^{(1,2,\mathrm{},k)}=_{i=1}^k\sigma _\alpha ^i`$ form a commuting set and hence, a good operator basis. Note first that
$$(𝐒^k)^2=\underset{i,j=1}{\overset{k}{}}\underset{\alpha =x,y,z}{}\sigma _\alpha ^i\sigma _\alpha ^j.$$
(A1)
Thus
$$[(𝐒^k)^2,(𝐒^l)^2]=[\underset{i,j=1}{\overset{k}{}}\underset{\alpha =x,y,z}{}\sigma _\alpha ^i\sigma _\alpha ^j,\underset{m,n=1}{\overset{l}{}}\underset{\beta =x,y,z}{}\sigma _\beta ^m\sigma _\beta ^n].$$
(A2)
Terms with $`\alpha =\beta `$ obviously commute. Further, terms with ($`m=i,n=j`$), ($`m=j,n=i`$), or ($`im,n,jm,n`$), commute, so we need only consider ($`i=m,jn`$), ($`i=n,jm`$) or ($`im,j=n`$), ($`in,j=m`$). In addition, assuming w.l.o.g. that $`lk`$, terms with $`m,n>k`$ also commute. Thus we are left with
$$[(𝐒^k)^2,(𝐒^l)^2]=2\underset{i,j=1}{\overset{k}{}}\underset{n\left(j\right)=1}{\overset{k}{}}\underset{\beta \alpha =x,y,z}{}[\sigma _\alpha ^i\sigma _\alpha ^j,\sigma _\beta ^i\sigma _\beta ^n]+2\underset{i,j=1}{\overset{k}{}}\underset{m\left(i\right)=1}{\overset{k}{}}\underset{\beta \alpha =x,y,z}{}[\sigma _\alpha ^i\sigma _\alpha ^j,\sigma _\beta ^m\sigma _\beta ^j].$$
(A3)
Using the fact that $`[\sigma _\alpha ^i\sigma _\alpha ^j,\sigma _\beta ^i\sigma _\beta ^n]=i_\gamma \epsilon _{\alpha \beta \gamma }\sigma _\gamma ^i\sigma _\alpha ^j\sigma _\beta ^n`$ and $`[\sigma _\alpha ^i\sigma _\alpha ^j,\sigma _\beta ^m\sigma _\beta ^j]=i_\gamma \epsilon _{\alpha \beta \gamma }\sigma _\alpha ^i\sigma _\beta ^m\sigma _\gamma ^j`$:
$`[(𝐒^k)^2,(𝐒^l)^2]=2{\displaystyle \underset{i,j=1}{\overset{k}{}}}{\displaystyle \underset{n\left(j\right)=1}{\overset{k}{}}}{\displaystyle \underset{\alpha ,\beta ,\gamma =\{x,y,z\}}{}}\epsilon _{\alpha \beta \gamma }\sigma _\gamma ^i\sigma _\alpha ^j\sigma _\beta ^n+2{\displaystyle \underset{i,j=1}{\overset{k}{}}}{\displaystyle \underset{m\left(i\right)=1}{\overset{k}{}}}{\displaystyle \underset{\alpha ,\beta ,\gamma =\{x,y,z\}}{}}\epsilon _{\alpha \beta \gamma }\sigma _\alpha ^i\sigma _\beta ^m\sigma _\gamma ^j,`$
and both sums vanish due to the antisymmetric property of $`\epsilon _{\alpha \beta \gamma }`$.
## B Maximal-$`m_J`$ States of the Strong Collective Decoherence DFS
We show how to recursively express the $`n`$-particle total spin-$`J`$ states in terms of ($`n1`$)-particle states. Let us focus on DFS$`{}_{n}{}^{}(J)`$ and in particular on the maximal-$`m_J`$ state in it:
$$|\psi =|J_1,\mathrm{},J_{n1},J;m_J=J.$$
(B1)
In general ($`J0,n/2`$) there are two kinds of states: bottom ($`|\psi _\mathrm{B}`$) and top ($`|\psi _\mathrm{T}`$) ones. The angular momentum addition rule that must be satisfied for adding a single spin-$`\frac{1}{2}`$ particle is that
$`m_{J_{n1}}\pm {\displaystyle \frac{1}{2}}=m_J.`$
The B-state comes from adding a particle to the maximal $`m_J`$ state in DFS$`{}_{n1}{}^{}(J1/2)`$, which is:
$$|\mathrm{B}=|J_1,\mathrm{},J_{n2},J\frac{1}{2};m_{J_{n1}}=J\frac{1}{2}.$$
(B2)
There is only one way to go from $`|\mathrm{B}`$ to $`|\psi _\mathrm{B}`$, namely to add $`1/2`$ to $`m_{J_{n1}}=J\frac{1}{2}`$ in order to obtain $`m_J=J`$. Thus
$$|\psi _\mathrm{B}=|\mathrm{B}|\frac{1}{2},\frac{1}{2},$$
(B3)
where $`|\frac{1}{2},\frac{1}{2}`$ is the single-particle spin-up state. The situation is different for the T-state, which is constructed by adding a particle to
$$|\mathrm{T}_\pm =|J_1,\mathrm{},J_{n2},J+\frac{1}{2};m_{J_{n1}}=J\pm \frac{1}{2}.$$
(B4)
These two possibilities give:
$$|\psi _\mathrm{T}=\alpha |\mathrm{T}_+|\frac{1}{2},\frac{1}{2}+\beta |\mathrm{T}_{}|\frac{1}{2},\frac{1}{2}.$$
(B5)
To find the coefficients $`\alpha `$ and $`\beta `$ we use the collective raising operator $`𝐬_+=𝐬_x+i𝐬_y`$, where we recall that $`𝐬_\alpha ^{(k)}=\frac{1}{2}_{i=1}^k\sigma _\alpha ^i`$. Since $`|\psi `$ is a maximal-$`m_J`$ state it is annihilated by $`𝐬_+𝐬_\alpha ^{(n)}`$. Similarly, $`|\mathrm{T}_+`$ is annihilated by $`𝐬_+^{(n1)}`$. Therefore, since $`𝐬_+=𝐬_+^{(n1)}+\frac{1}{2}\sigma _+^n`$:
$`𝐬_+|\mathrm{T}_+|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}`$ $`=`$ $`|\mathrm{T}_+|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}`$
$`𝐬_+|\mathrm{T}_{}|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}`$ $`=`$ $`\sqrt{2J+1}|\mathrm{T}_+|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}},`$
where in the second line we used the elementary raising operator formula $`𝐉_+|j,m=\left[j(j+1)m(m+1)\right]^{1/2}|j,m+1`$ with $`j=J+\frac{1}{2}`$ and $`m=J\frac{1}{2}`$. Application of $`𝐬_+`$ to Eq. (B5) thus yields:
$$\alpha +\sqrt{2J+1}\beta =0$$
(B6)
Hence, up to an arbitrary phase choice, we find that
$$\alpha =\sqrt{\frac{2J+1}{2J+2}}\beta =\frac{1}{\sqrt{2J+2}}.$$
(B7)
The special cases of $`J=0,n/2`$ differ only in that the corresponding DFSs support just T- and B-states, respectively. The calculation of the coefficients, therefore, remains the same.
In a similar manner one can carry the calculation one particle deeper. Doing this we find for the maximal-$`m_J`$ states (provided they exist):
$`|\mathrm{TT}|J_1,`$ $`\mathrm{},J_{n3},J+1,J+{\displaystyle \frac{1}{2}},J;m_J=J=\sqrt{{\displaystyle \frac{2J+1}{2J+3}}}|J_1,\mathrm{},J_{n3},J+1;m_{J_{n2}}=J+1|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}`$ (B10)
$`\sqrt{{\displaystyle \frac{2J+1}{(2J+2)(2J+3)}}}|J_1,\mathrm{},J_{n3},J+1;m_{J_{n2}}=J\left(|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}+|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}\right)`$
$`+\sqrt{{\displaystyle \frac{2}{(2J+2)(2J+3)}}}|J_1,\mathrm{},J_{n3},J+1;m_{J_{n2}}=J1|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}`$
$`|\mathrm{BT}|J_1,`$ $`\mathrm{},J_{n3},J,J+{\displaystyle \frac{1}{2}},J;m_J=J=\sqrt{{\displaystyle \frac{2J+1}{2J+2}}}|J_1,\mathrm{},J_{n3},J;m_{J_{n2}}=J|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}`$ (B13)
$`+{\displaystyle \frac{1}{\sqrt{(2J+2)(2J+1)}}}|J_1,\mathrm{},J_{n3},J;m_{J_{n2}}=J|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}`$
$`+\sqrt{{\displaystyle \frac{2J}{(2J+1)(2J+2)}}}|J_1,\mathrm{},J_{n3},J;m_{J_{n2}}=J1|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}`$
$`|\mathrm{TB}|J_1,`$ $`\mathrm{},J_{n3},J,J{\displaystyle \frac{1}{2}},J;m_J=J=\sqrt{{\displaystyle \frac{2J}{2J+1}}}|J_1,\mathrm{},J_{n3},J;m_{J_{n2}}=J|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}`$ (B15)
$`+{\displaystyle \frac{1}{\sqrt{2J+1}}}|J_1,\mathrm{},J_{n3},J;m_{J_{n2}}=J1|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}`$
$`|\mathrm{BB}|J_1,`$ $`\mathrm{},J_{n3},J1,J{\displaystyle \frac{1}{2}},J;m_J=J=|J_1,\mathrm{},J_{n3},J1;m_{J_{n2}}=J1|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}|{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}}.`$ (B16)
Caution must be exercised in using these expressions near the boundary of Table (I), where some of the states may not exist.
## C Proofs of the Lemmas
Enlarging Lemma— Let $``$ be a Hilbert space of dimension $`d`$ and let $`|i`$. Assume we are given a set of Hamiltonians $`𝖧_1`$ that generates $`su(d1)`$ on the subspace of $``$ that does not contain $`|i`$, and another set $`𝖧_2`$ that generates $`su(2)`$ on the subspace of $``$ spanned by $`\{|i,|j\}`$, where $`|j`$ is another state in $``$. Then $`[𝖧_1,𝖧_2]`$ (all commutators) generates $`su(d)`$ on $``$ under closure as a Lie-algebra.
Proof. We explicitly construct the Lie-algebra $`su(d)`$ with the given Hamiltonians. Let $`\stackrel{~}{}`$ be the $`d1`$ dimensional subspace $`𝖧_1`$ acts on. Let us show that we can generate $`su(2)`$ between $`|k\stackrel{~}{}`$ and $`|i`$.
Let $`𝐗_{ij}|ij|+|ji|𝖧_2`$ and $`𝐗_{jk}|jk|+|kj|𝖧_1`$. Then
$$𝐘_{ik}i[𝐗_{jk},𝐗_{ij}]=i|ik|+i|ki|$$
(C1)
acts as $`\sigma _y`$ on the states $`|i,|k`$. Similarly
$$𝐗_{ik}i[𝐘_{ij},𝐗_{jk}]=|ik|+|ki|$$
(C2)
yields $`\sigma _x`$ on the space spanned by $`|i,|k`$. These two operations generate $`su(2)`$ on $`|i,|k`$ for all $`|k`$ in the subspace of $``$ that does not contain $`|i`$. Now the Mixing Lemma gives the desired result together with the observation that there we only use elements in $`[𝖧_1,𝖧_2]`$.
Mixing Lemma— Consider the division of an $`n`$ dimensional Hilbert space $``$ into a direct sum of two subspaces $`_1_2`$ of dimensions $`n_1`$ and $`n_2`$ respectively. Suppose that $`|i_n`$ is an orthonormal basis for $`_n`$. Then the Lie algebras generated by $`𝐗_{i_1,i_2}=|i_1i_2|+|i_2i_1|`$, $`𝐘_{i_1,i_2}=i|i_1i_2|i|i_2i_1|`$, and $`𝐙_{i_1,i_2}=|i_1i_1||i_2i_2|`$ generate $`su(n)`$.
Proof. We explicitly construct the elements of $`su(n)`$. Consider $`i[𝐗_{i_1,i_2},𝐘_{j_1,j_2}]`$. Clearly, if $`i_1i_2j_1j_2`$ this equals zero and if $`i_1=j_1`$ and $`i_2=j_2`$ then this commutator is $`𝐙_{i_1,i_2}`$. If, however, $`i_1=j_1`$ and $`i_2j_2`$ this becomes
$$i[𝐗_{i_1,i_2},𝐘_{i_1,j_2}]=|i_2j_2||j_2i_2|.$$
(C3)
Similarly:
$$i[𝐗_{i_1,i_2},𝐘_{j_1,i_2}]=|i_1j_1|+|j_1i_1|.$$
(C4)
Thus every $`|i_kj_l|+|j_li_k|`$ is in the Lie algebra. Similarly, $`i[𝐗_{i_1,i_2},𝐗_{j_1,j_2}]`$ yields
$`i[𝐗_{i_1,i_2},𝐗_{i_1,j_2}]`$ $`=`$ $`i|i_2j_2|i|j_2i_2|`$ (C5)
$`i[𝐗_{i_1,i_2},𝐗_{j_1,i_2}]`$ $`=`$ $`i|i_1j_1|i|j_1i_1|`$ (C6)
Thus every $`i|i_kj_l|i|j_li_k|`$ is in the Lie algebra. Taking the commutator of these with the $`|i_kj_l|+|j_li_k|`$ operators finally yields every $`|i_kj_l||j_li_k|`$. Since $`su(n)`$ can be decomposed into a sum of overlapping $`su(2)`$’s , the Lie algebra is the entire $`su(n)`$, as claimed.
|
warning/0004/hep-ex0004009.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The key question of calorimetry generally and hadronic calorimetry in particular is the energy reconstruction. This question is especially important when a hadronic calorimeter have a complex structure being a combined calorimeter. Such is the combined calorimeter with the electromagnetic and hadronic compartments of the ATLAS detector . In this paper we describe the non-parametrical method of the energy reconstruction for a combined calorimeter, which called the $`e/h`$ method, and demonstrate its performance on the basis of the test beam data of the ATLAS combined prototype calorimeter. For the energy reconstruction and description of the longitudinal development of a hadronic shower it is necessary to know the $`e/h`$ ratios, the degree of non-compensation, of these calorimeters. As to the ATLAS Tile barrel calorimeter there is the detailed information about the $`e/h`$ ratio presented in . But as to the liquid argon electromagnetic calorimeter such information is practically absent. The aim of the present work is also to develop the method and to determine the value of the $`e/h`$ ratio of the electromagnetic compartment.
One of the important questions of hadron calorimetry is the question of the longitudinal development of hadronic showers. This question is especially important for a combined calorimeter. This work is also devoted to the study of the longitudinal hadronic shower development in the ATLAS barrel combined prototype calorimeter.
This work has been performed on the basis of the 1996 combined test beam data . The data have been taken in the H8 beam line of the CERN SPS using pions of 10, 20, 40, 50, 80, 100, 150 and 300 GeV.
## 2 Combined Calorimeter
The combined calorimeter prototype setup has been made consisting of the LAr electromagnetic calorimeter prototype inside the cryostat and downstream the Tile calorimeter prototype as shown in Fig. 1. The two calorimeters have been placed with their central axes at an angle to the beam of $`12^{}`$. At this angle the two calorimeters have an active thickness of 10.3 $`\lambda _I`$. Beam quality and geometry were monitored with a set of beam wire chambers BC1, BC2, BC3 and trigger hodoscopes placed upstream of the LAr cryostat. To detect punchthrough particles and to measure the effect of longitudinal leakage a “muon wall” consisting of 10 scintillator counters (each 2 cm thick) was located behind the calorimeters at a distance of about 1 metre.
### 2.1 Electromagnetic Calorimeter
The electromagnetic LAr calorimeter prototype consists of a stack of three azimuthal modules, each one spanning $`9^{}`$ in azimuth and extending over 2 m along the Z direction. The calorimeter structure is defined by 2.2 mm thick steel-plated lead absorbers folded to an accordion shape and separated by 3.8 mm gaps filled with liquid argon. The signals are collected by Kapton electrodes located in the gaps. The calorimeter extends from an inner radius of 131.5 cm to an outer radius of 182.6 cm, representing (at $`\eta =0`$) a total of 25 radiation lengths ($`X_0`$), or 1.22 interaction lengths ($`\lambda _I`$) for protons. The calorimeter is longitudinally segmented into three compartments of $`9X_0`$, $`9X_0`$ and $`7X_0`$, respectively. More details about this prototype can be found in .
The cryostat has a cylindrical form with 2 m internal diameter, filled with liquid argon, and is made out of a 8 mm thick inner stainless-steel vessel, isolated by 30 cm of low-density foam (Rohacell), itself protected by a 1.2 mm thick aluminum outer wall.
### 2.2 Hadronic Calorimeter
The hadronic Tile calorimeter is a sampling device using steel as the absorber and scintillating tiles as the active material . The innovative feature of the design is the orientation of the tiles which are placed in planes perpendicular to the Z direction . For a better sampling homogeneity the 3 mm thick scintillators are staggered in the radial direction. The tiles are separated along Z by 14 mm of steel, giving a steel/scintillator volume ratio of 4.7. Wavelength shifting fibres (WLS) running radially collect light from the tiles at both of their open edges. The hadron calorimeter prototype consists of an azimuthal stack of five modules. Each module covers $`2\pi /64`$ in azimuth and extends 1 m along the Z direction, such that the front face covers $`100\times 20`$ cm<sup>2</sup>. The radial depth, from an inner radius of 200 cm to an outer radius of 380 cm, accounts for 8.9 $`\lambda `$ at $`\eta =0`$ (80.5 $`X_0`$). Read-out cells are defined by grouping together a bundle of fibres into one photomultiplier (PMT). Each of the 100 cells is read out by two PMTs and is fully projective in azimuth (with $`\mathrm{\Delta }\varphi =2\pi /640.1`$), while the segmentation along the Z axis is made by grouping fibres into read-out cells spanning $`\mathrm{\Delta }Z=20`$ cm ($`\mathrm{\Delta }\eta 0.1`$) and is therefore not projective. Each module is read out in four longitudinal segments (corresponding to about 1.5, 2, 2.5 and 3 $`\lambda _I`$ at $`\eta =0`$). More details of this prototype can be found in .
### 2.3 Data Selection
The data have been taken in the H8 beam line of the CERN SPS using pions of 10, 20, 40, 50, 80, 100, 150 and 300 GeV. We applied some similar to cuts to eliminate the non-single track pion events, the beam halo, the events with an interaction before the LAr calorimeter, the electron and muon events. The set of cuts is the following: the single-track pion events were selected by requiring the pulse height of the beam scintillation counters and the energy released in the presampler of the electromagnetic calorimeter to be compatible with that for a single particle; the beam halo events were removed with appropriate cuts on the horizontal and vertical positions of the incoming track impact point and the space angle with respect to the beam axis as measured with the beam chambers; a cut on the total energy rejects incoming muon.
## 3 $`e/h`$ Method of Energy Reconstruction
The response, $`R`$, of a calorimeter to a hadronic shower is the sum of the contributions from the electromagnetic, $`E_e`$, and hadronic, $`E_h`$, parts of the incident energy $`E`$ :
$$R=eE_e+hE_h=eE(f_{\pi ^0}+(h/e)(1f_{\pi ^0})),$$
(1)
$$E=E_e+E_h,$$
(2)
where $`e`$ ($`h`$) is the energy independent coefficient of transformation of the electromagnetic (pure hadronic, low-energy hadronic activity) energy to response, $`f_{\pi ^0}=E_e/E`$ is the fraction of electromagnetic energy. From this
$$E=\left(\frac{e}{\pi }\right)\frac{R}{e},$$
(3)
and
$$\frac{e}{\pi }=\frac{e/h}{1+(e/h1)f_{\pi ^0}}.$$
(4)
For a combined calorimeter the incident energy deposits into the LAr compartment, $`E_{LAr}`$, the Tile calorimeter compartment, $`E_{Tile}`$, and into the passive material between the LAr and Tile calorimeters, $`E_{dm}`$,
$$E=E_{LAr}+E_{dm}+E_{Tile}.$$
(5)
Using the expressions (3) – (5) the following equation for the energy reconstruction has been derived:
$$E=\underset{i}{}c_i\left(\frac{e}{\pi }\right)_iR_i=c_{LAr}\left(\frac{e}{\pi }\right)_{LAr}R_{LAr}+E_{dm}+c_{Tile}\left(\frac{e}{\pi }\right)_{Tile}R_{Tile},$$
(6)
where $`i=LAr,dm,Tile`$; $`c_i=1/e_i`$; $`(e/\pi )_i`$ are from equation (4) and
$$f_{\pi ^0,LAr}=k\mathrm{ln}(E),$$
(7)
$$f_{\pi ^0,Tile}=k\mathrm{ln}(E_{Tile}),$$
(8)
where $`E_{Tile}=c_{Tile}(e/\pi )_{Tile}R_{Tile}`$ and $`k=0.11`$. Note, that for $`70\%`$ of events an energy in Tile calorimeter is approximately equal to beam energy because a hadronic shower began in the hadron calorimeter.
The term, which accounts for the energy loss in the dead material between the LAr and Tile calorimeters, $`E_{dm}`$, is taken to be proportional to the geometrical mean of the energy released in the third depth of the electromagnetic compartment ($`E_{LAr,3}=c_{LAr}(e/\pi )_{LAr}R_{LAr,3}`$) and the first depth of the hadronic compartment ($`E_{Tile,1}=c_{Tile}(e/\pi )_{Tile}R_{Tile,1}`$)
$$E_{dm}=c_{dm}\sqrt{E_{LAr,3}E_{Tile,1}}$$
(9)
similar to . The validity of this approximation has been tested by the Monte Carlo simulation and by the study of the correlation between the energy released in the midsampler and the cryostat energy deposition . The value of $`c_{dm}=0.31`$ obtained on the basis of the results of the Monte Carlo simulation is used.
In order to use the equation (6) it is necessary to know the values of the following constants: $`c_{LAr}`$, $`c_{dm}`$, $`c_{Tile}`$, $`(e/h)_{LAr}`$, $`(e/h)_{Tile}`$, some of which are $`c_{LAr}=1/e_{LAr}=1.1`$ , $`(e/h)_{Tile}=1.3\pm 0.03`$ . The determination of the other constants ($`c_{dm}`$, $`c_{Tile}`$, $`(e/h)_{LAr}`$) is given below.
### 3.1 $`c_{Tile}`$ Constant
For the determining of the $`c_{Tile}`$ constant the following procedure was applied. We selected the events which start to shower only in the hadronic calorimeter. To select these events the energies deposited in each sampling of the LAr calorimeter and in the midsampler are required to be compatible with that of a single minimum ionization particle.
The following expression for the normalized hadronic response have been used :
$$\frac{R_{Tile}^c}{E_{beam}}=\frac{1+((e/h)_{Tile}1)f_{\pi ^0,Tile}}{c_{Tile}(e/h)_{Tile}},$$
(10)
where
$$R_{Tile}^c=R_{Tile}+\frac{c_{LAr}}{c_{Tile}}R_{LAr}$$
(11)
is the Tile calorimeter response corrected on the energy loss in the LAr calorimeter, $`f_{\pi ^0,Tile}`$ is determined by the formula (8). The value of $`c_{Tile}`$ obtained by fitting is equal to $`0.145\pm 0.002`$.
### 3.2 $`e/h`$ ratio of Electromagnetic Compartment
Using the expression (6) the value of the $`(e/\pi )_{LAr}`$ ratio can be obtained
$$\left(\frac{e}{\pi }\right)_{LAr}=\frac{E_{beam}E_{dm}E_{Tile}}{c_{LAr}R_{LAr}}.$$
(12)
The $`(e/h)_{LAr}`$ ratio and the function $`f_{\pi ^0,LAr}`$ (7) can be inferred from the energy dependent $`(e/\pi )_{LAr}`$ ratios.
For this case we select the events with the well developed hadronic showers in the electromagnetic calorimeter. Than mean that energy depositions were required to be more than 10% of the beam energy in the electromagnetic calorimeter and less than 70% in the hadronic calorimeter.
Fig. 2 shows the distributions of the $`(e/\pi )_{LAr}`$ ratio derived by formula (12) for different energies. The mean values of these distributions are shown in Fig. 3 as a function of the beam energy. The fit of this distribution by the expression (4) for LAr calorimeter yields $`(e/h)_{LAr}=1.74\pm 0.04`$ and $`k=0.108\pm 0.004`$. The quoted errors are the statistical ones obtained from the fit. The systematic error on the $`(e/h)_{LAr}`$ ratio, which is a consequence of the uncertainties in the input constants used in the equation (12), is estimated to be $`\pm 0.04`$.
Wigmans showed that the $`e/h`$ ratio for non-uranium calorimeters with high-Z absorber material is satisfactorily described by the formula:
$$\frac{e}{h}=\frac{e/mip}{0.41+0.12n/mip}$$
(13)
in which $`e/mip`$ and $`n/mip`$ represent the calorimeter response to e.m. showers and to MeV-type neutrons, respectively. These responses are normalized to the one for minimum ionizing particles. The Monte Carlo calculated $`e/mip`$ and $`n/mip`$ values for the R&D3 Pb-LAr electromagnetic calorimeter are $`e/mip=0.78`$ and $`n/mip<0.5`$ leading to $`(e/h)_{LAr}>1.66`$. Our measured value of the $`(e/h)_{LAr}`$ ratio agrees with this prediction.
### 3.3 Iteration Procedure
For the energy reconstruction by the formula (6) it is necessary to know the $`(e/\pi )_{Tile}`$ ratio and the reconstructed energy itself. Therefore, the iteration procedure has been developed. Two iteration cycles were made: the first one is devoted to the determination of the $`(e/\pi )_{Tile}`$ ratio and the second one is the energy reconstruction itself.
The expression (4) for the $`(e/\pi )_{Tile}`$ ratio can be written as
$$\left(\frac{e}{\pi }\right)_{Tile}=\frac{(e/h)_{Tile}}{1+((e/h)_{Tile}1)k\mathrm{ln}(c_{Tile}(e/\pi )_{Tile}R_{Tile})}.$$
(14)
As the first approximation, the value of $`(e/\pi )_{Tile}`$ is calculated using the equation (14) where in the right side of this equation we used $`(e/\pi )_{Tile}=1.13`$ corresponding to $`f_{\pi ^0,Tile}=0.5=0.11\mathrm{ln}(100GeV)`$. The iteration process is stopped when the convergence criterion $`(e/\pi )_{Tile}^{\nu +1}(e/\pi )_{Tile}^\nu /(e/\pi )_{Tile}^\nu <ϵ`$, where ($`\nu =0,1,\mathrm{}`$), is satisfied.
As the first approximation in the iteration cycle for the energy reconstruction, the value of $`E`$ is calculated using the equation (6) with the $`(e/\pi )_{Tile}`$ ratio obtained in the first iteration cycle and $`(e/\pi )_{LAr}`$ from equation (4) where in the right side of this equation we used $`(e/\pi )_{LAr}=1.27`$ corresponding to $`f_{\pi ^0,LAr}=0.5=0.11\mathrm{ln}(100GeV)`$. The convergence criterion is $`E^{\nu +1}E^\nu /E^\nu <ϵ`$.
For both these case the iteration values are under the logarithmic function that mean that iteration procedure will be very fast.
The average numbers of iterations $`<N_{it}>`$ for the various beam energies needed to receive the given value of accuracy $`ϵ`$ have been investigated. It turned out, it is sufficiently only the first approximation for achievement, on average, of convergence with an accuracy of $`ϵ=1\%`$ for energies 80 – 150 $`GeV`$ and it is necessary to perform only one iteration for the energies at 10 – 50 $`GeV`$ and 300 $`GeV`$.
We specially investigated the accuracy of the first approximation of energy. Fig. 4 shows the comparison between the energy linearities, the mean values of $`E/E_{beam}`$, obtained using the iteration procedure with $`ϵ=0.1\%`$ (black circles) and the first approximation of energy (open circles). Fig. 5 shows the comparison between the energy resolutions obtained using these two approaches. As can be seen, the compared values are consistent within errors.
The suggested algorithm of the energy reconstruction can be used for the fast energy reconstruction in the first level trigger.
### 3.4 Energy Spectra
Fig. 6 shows the pion energy spectra reconstructed with the $`e/h`$ method ($`ϵ=0.1\%`$). The mean and $`\sigma `$ values of these distributions are extracted with Gaussian fits over $`\pm 2\sigma `$ range. The obtained mean values $`E`$, the energy resolutions $`\sigma `$, and the fractional energy resolutions $`\sigma /E`$ are listed in Table 1 for the various beam energies.
### 3.5 Energy Linearity
Fig. 7 demonstrates the correctness of the mean energy reconstruction. The mean value of $`E/E_{beam}`$ is equal to $`(99.5\pm 0.3)\%`$ and the spread is $`\pm 1\%`$ except for the point at 10 $`GeV`$. But, as noted in , at this point the result is strongly dependent on the effective capability to remove events with interactions in the dead material upstream and to deconvolve the real pion contribution from the muon contamination. Fig. 7 also shows the comparison of the linearity, $`E/E_{beam}`$, as a function of the beam energy for the $`e/h`$ method and for the cells weighting method . As can be seen, the comparable quality of the linearity is observed for these two methods.
### 3.6 Energy Resolutions
Fig. 8 shows the fractional energy resolutions ($`\sigma /E`$) as a function of $`1/\sqrt{E_{beam}}`$ obtained by three methods: the $`e/h`$ method (black circles), the benchmark method (crosses), and the cells weighting method (open circles). As can be seen, the energy resolutions for the $`e/h`$ method are comparable with the benchmark method and only of $`30\%`$ worse than for the cells weighting method. A fit to the data points gives the fractional energy resolution for the $`e/h`$ method obtained using the iteration procedure with $`ϵ=0.1\%`$: $`\sigma /E=[(58\pm 3)\%\sqrt{GeV}/\sqrt{E}+(2.5\pm 0.3)\%](1.7\pm 0.2)GeV/E`$, for the $`e/h`$ method using the first approximation: $`\sigma /E=[(56\pm 3)\%\sqrt{GeV}/\sqrt{E}+(2.7\pm 0.3)\%](1.8\pm 0.2)GeV/E`$, for the benchmark method of $`\sigma /E=[(60\pm 3)\%\sqrt{GeV}/\sqrt{E}+(1.8\pm 0.2)\%](2.0\pm 0.1)GeV/E`$, for the cells weighting method of $`\sigma /E=[(42\pm 2)\%\sqrt{GeV}/\sqrt{E}+(1.8\pm 0.1)\%](1.8\pm 0.1)GeV/E`$, where the symbol $``$ indicates a sum in quadrature. As can be seen, the sampling term is consistent within errors for the $`e/h`$ method and the benchmark method and is smaller by 1.5 times for the cells weighting method. The constant term is the same for the benchmark method and the cells weighting method and is larger by $`(0.7\pm 0.3)\%`$ for the $`e/h`$ method. The noise term of about $`1.8GeV`$ is the same for these three methods that reflects its origin as the electronic noise. As to the two approaches for the $`e/h`$ method, the fitted parameters coincide within errors.
## 4 Hadronic Shower Development
We used this energy reconstruction method and obtained the energy depositions, $`E_i`$, in each longitudinal sampling with the thickness of $`\mathrm{\Delta }x_i`$ in units $`\lambda _\pi `$. Table 2 lists and Fig. 9 shows the differential energy depositions $`(\mathrm{\Delta }E/\mathrm{\Delta }x)_i=E_i/\mathrm{\Delta }x_i`$ as a function of the longitudinal coordinate $`x`$ for 10 – 300 GeV.
### 4.1 Longitudinal Hadronic Shower Parameterization
There is the well known parameterization of the longitudinal hadronic shower development from the shower origin suggested in
$$\frac{dE_s(x)}{dx}=N\left\{w\left(\frac{x}{X_0}\right)^{a1}e^{b\frac{x}{X_0}}+(1w)\left(\frac{x}{\lambda _I}\right)^{a1}e^{d\frac{x}{\lambda _I}}\right\},$$
(15)
where $`X_0`$ is the radiation length, $`\lambda _I`$ is the interaction length, $`N`$ is the normalization factor, $`a,b,d,w`$ are parameters: $`a=0.6165+0.3193lnE`$, $`b=0.2198`$, $`d=0.90990.0237lnE`$, $`\omega =0.4634`$.
This parameterization is from the shower origin. But our data are from the calorimeter face and due to the unsufficient longitudinal segmentation can not be transformed to the shower origin. Therefore, we used the analytical representation of the hadronic shower longitudinal development from the calorimeter face :
$`{\displaystyle \frac{dE(x)}{dx}}`$ $`=`$ $`N\{{\displaystyle \frac{wX_0}{a}}\left({\displaystyle \frac{x}{X_0}}\right)^ae^{b\frac{x}{X_0}}{}_{1}{}^{}F_{1}^{}(1,a+1,(b{\displaystyle \frac{X_0}{\lambda _I}}){\displaystyle \frac{x}{X_0}})`$ (16)
$`+{\displaystyle \frac{(1w)\lambda _I}{a}}\left({\displaystyle \frac{x}{\lambda _I}}\right)^ae^{d\frac{x}{\lambda _I}}{}_{1}{}^{}F_{1}^{}(1,a+1,(d1){\displaystyle \frac{x}{\lambda _I}})\},`$
here $`{}_{1}{}^{}F_{1}^{}(\alpha ,\beta ,z)`$ is the confluent hypergeometric function.
Note that the formula (16) is given for a calorimeter characterizing by the certain $`X_0`$ and $`\lambda _I`$ values. At the same time, the values of $`X_0`$, $`\lambda _I`$ and the $`e/h`$ ratios are different for electromagnetic and hadronic compartments of a combined calorimeter. So, it is impossible straightforward use of the formula (16) for the description of a hadronic shower longitudinal profiles in combined calorimetry.
In suggested the following algorithm of combination of the electromagnetic calorimeter ($`em`$) and hadronic calorimeter ($`had`$) curves of the differential longitudinal energy deposition $`dE/dx`$. At first, a hadronic shower develops in the electromagnetic calorimeter to the boundary value $`x_{em}`$ which corresponds to certain integrated measured energy $`E_{em}(x_{em})`$. Then, using the corresponding integrated hadronic curve, $`E(x)=`$ $`_0^x(dE/dx)𝑑x`$, the point $`x_{had}`$ is found from equation $`E_{had}(x_{had})=E_{em}(x_{em})`$ $`+E_{dm}`$. From this point a shower continues to develop in the hadronic calorimeter. In principle, instead of the measured value of $`E_{em}`$ one can use the calculated value of $`E_{em}=_0^{x_{em}}(dE/dx)𝑑x`$ obtained from the integrated electromagnetic curve. In this way, the combined curves have been obtained.
Fig. 9 shows the differential energy depositions $`(\mathrm{\Delta }E/\mathrm{\Delta }x)_i=E_i/\mathrm{\Delta }x_i`$ as a function of the longitudinal coordinate $`x`$ in units $`\lambda _\pi `$ for the 10 – 300 GeV and comparison with the combined curves for the longitudinal hadronic shower profiles (the dashed lines). It can be seen that there is a significant disagreement between the experimental data and the combined curves in the region of the LAr calorimeter and especially at low energies.
### 4.2 Modification of Shower Parameterization
We attempted to improve the description and to include such essential feature of a calorimeter as the $`e/h`$ ratio. Several modifications and adjustments of some parameters of this parameterization have been tried. It turned out that the changes of two parameters $`b`$ and $`w`$ in the formula (16) in such a way that $`b=0.22(e/h)_{cal}/(e/h)_{cal}^{}`$ and $`w=0.6(e/\pi )_{cal}/(e/\pi )_{cal}^{}`$ made it possible to obtain the reasonable description of the experimental data. Here the values of the $`(e/h)_{cal}^{}`$ ratios are $`(e/h)_{em}^{}1.1`$ and $`(e/h)_{had}^{}1.3`$ which correspond to the data used for the Bock et al. parameterization . The $`(e/\pi )_{cal}^{}`$ are calculated using formula (4).
In Fig. 9 the experimental differential longitudinal energy depositions and the results of the description by the modified parameterization (the solid lines) are compared. There is a reasonable agreement (probability of description is more than $`5\%`$) between the experimental data and the curves taking into account uncertainties in the parametrization function . In such case the Bock et al. parameterization is the private case for some fixed the $`e/h`$ ratio.
### 4.3 Energy Deposition in Compartments
The obtained parameterization has some additional applications. For example, this formula may be used for an estimate of the energy deposition in various parts of a combined calorimeter. This demonstrates in Fig. 10 in which the measured and calculated relative values of the energy deposition in the LAr and Tile calorimeters are presented. The relative energy deposition in the LAr calorimeter decreases from about 50% at 10 $`GeV`$ to 30% at 300 $`GeV`$. On the contrary, the one in Tile calorimeter increases with the energy increasing.
## 5 Conclusions
Hadron energy reconstruction for the ATLAS barrel prototype combined calorimeter, consisting of the lead-liquid argon electromagnetic part and the iron-scintillator hadronic part, in the framework of the non-parametrical method has been fulfilled. The non-parametrical method of the energy reconstruction for a combined calorimeter uses only the known $`e/h`$ ratios and the electron calibration constants, does not require the determination of any parameters by a minimization technique and can be used for the fast energy reconstruction in the first level trigger. The correctness of the reconstruction of the mean values of energies (for energy biger than 10 GeV) within $`\pm 1\%`$ has been demonstrated. The obtained fractional energy resolution is $`[(58\pm 3)\%\sqrt{GeV}/\sqrt{E}+(2.5\pm 0.3)\%](1.7\pm 0.2)GeV/E`$. The obtained value of the $`e/h`$ ratio for electromagnetic compartment of the combined calorimeter is $`1.74\pm 0.04`$ and agrees with the prediction that $`e/h>1.7`$ for this electromagnetic calorimeter. The results of the study of the longitudinal hadronic shower development are presented. The data have been taken on the H8 beam of the CERN SPS, with the pion beams of 10 – 300 GeV.
## 6 Acknowledgments
This work is the result of the efforts of many people from the ATLAS Collaboration. The authors are greatly indebted to all Collaboration for their test beam setup and data taking. Authors are grateful Peter Jenni for fruitful discussion.
|
warning/0004/astro-ph0004410.html
|
ar5iv
|
text
|
# Dynamics of superconducting strings with chiral currents
## I Introduction
Cosmic strings are linear topological defects that may have been created in the early Universe. They have been extensively studied in connection with several problems in cosmology (for reviews see). Assuming that the radius of curvature of a string is always much larger than the string core thickness, the string dynamics can be described by the Nambu-Goto action. In this approximation the equations of motion are simple, and can be easily solved.
In 1985, Witten showed that strings could behave as superconducting wires in certain particle physics models. This new internal degree of freedom opened up a variety of interesting effects. In particular, superconducting strings may have stable configurations, vortons and springs, which could contribute to the dark matter in the universe, or put constraints on the particle physics models that give rise to those strings. Superconducting cosmic strings have also been considered as sources for structure formation, gamma ray bursts, and ultra-high energy cosmic rays.
The general problem of a superconducting string coupled to the electromagnetic field cannot be solved analytically. However, if the charge carriers are not coupled to any long-range field (so-called neutral superconducting strings) the situation is significantly simpler. Such strings are of interest in their own right, and may also be considered as approximations to the full theory including electromagnetic coupling.
If in addition, we consider the case that the charge and current are equal in magnitude, then it was shown by Carter and Peter that the equations of motion can be solved exactly. In this case, the charge-current 4-vector is lightlike, and the current is said to be chiral or null. It consists of charge carriers which are all moving in the same direction. Such currents arise automatically in certain supersymmetric theories in which a zero mode can travel only in one direction along the string . They could also result from evolution of a loop with an arbitrary distribution of charge and current. Left- and right-moving charge carriers can scatter off the string, and if the numbers of left- and right-movers are not equal, the string can be driven towards the chiral limit. This will happen particularly near a cusp, where the string is contracted by a large factor, resulting in a much higher density of charge carriers than elsewhere along the string and in a great enhancement of the scattering rate. One can expect therefore that the current will be very nearly chiral in the vicinity of cusps. In the case of electrically charged currents, the electromagnetic radiation from oscillating loops is dominated by powerful bursts emitted from near-cusp regions. One can therefore use the solution for strings with chiral currents to estimate the radiation power from such loops.
To avoid confusion, we note that cusps were originally defined as points of infinite contraction, where the string momentarily reaches the speed of light. Strictly speaking, such cusps can be formed only on idealized Nambu-Goto strings. For realistic strings, the cusp development is truncated either by the annihilation of overlapping string segments at the tip of the cusp or for superconducting strings, by the back reaction of charge carriers or of the electromagnetic radiation. However, unless the string current is very large, so that the energy of the charge carriers is comparable to that of the string itself, the truncation occurs at a very large Lorentz factor and the string exhibits cusp-like behavior. Below we shall use the word “cusps” to refer to such ultra-relativistic string segments.
Carter and Peter derived their solution using Carter’s formalism which is not familiar to most cosmologists. In view of the importance of this result, in the present paper we shall give an alternative, elementary derivation. We shall also give an explicit solution for the initial value problem and discuss several physical implications of the result.
We begin in Section II by reviewing the well-known solution of string equations of motion in the case of non-superconducting Nambu-Goto strings. Our derivation of the solution for chiral superconducting strings is presented in Section III. In Section IV, we show how the motion of such strings can be found from given initial conditions. Some physical implications of the solution are discussed in Section V, where we show that the motion of string loops with chiral currents is strictly periodic and calculate the maximum Lorentz factor reached at the cusp. In Section VI we argue that the chiral string solution can be used as an approximate description of strings with small non-chiral currents. In Section VII we use this approximate description to estimate the electromagnetic radiation from oscillating loops with charged currents.
After this paper was submitted we learned about independent work by Davis et al. which gives a derivation of the chiral string solution similar to ours.
## II Nambu-Goto strings
We first review the equations of motion and their solution in the case of non-superconducting strings. We will use similar techniques in the next section to solve the equations of motion for superconducting strings, in the case that the current is chiral.
For an infinitely thin non-superconducting relativistic string, the flat spacetime equations of motion are
$$_a\left[\sqrt{\gamma }\gamma ^{ab}x_{,b}^\nu \right]=0$$
(1)
where $`a`$ and $`b`$ take the values 0 and 1 (denoting the worldsheet coordinates $`\tau `$ and $`\sigma `$ respectively), $`x^\nu (\sigma ,\tau )`$ is the position of the string, and $`\gamma _{ab}`$ is the induced metric on the worldsheet,
$$\gamma _{ab}=x_{,a}^\mu x_{\mu ,b}$$
(2)
and $`\gamma =det(\gamma _{ab})`$. We are free to choose a particular parameterization of the worldsheet, i.e. gauge condition. In this case, it is convenient to use the conformal gauge, namely
$$\gamma _{ab}=\mathrm{\Omega }(\sigma ,\tau )\eta _{ab}$$
(3)
where $`\eta _{ab}`$ is the two-dimensional Minkowski metric. Then $`\sqrt{\gamma }=\mathrm{\Omega }`$ and thus $`\sqrt{\gamma }\gamma ^{ab}=\eta ^{ab}`$. In this gauge the equation of motion for the string is just the two dimensional wave equation,
$$x^{\prime \prime \nu }\ddot{x}^\nu =0,$$
(4)
where $`\dot{x}`$ denotes $`x/\tau `$ and $`x^{}`$ denotes $`x/\sigma `$.
It can be seen that the constraints written above in Eq. (3), do not fix completely our gauge, and we can choose the condition
$$x^0=\tau $$
(5)
so the equations of motion become
$$𝐱^{\prime \prime }\ddot{𝐱}=0.$$
(6)
The general solution has the form
$$𝐱=\frac{1}{2}[𝐚(\sigma \tau )+𝐛(\sigma +\tau )].$$
(7)
In order for the metric to have the form of Eq. (3), we must impose the conditions
$$|𝐚^{}|^2=|𝐛^{}|^2=1.$$
(8)
## III Chiral superconducting strings
We consider a superconducting string with a neutral current (i.e., one not coupled to the electromagnetic field). We can describe the current via an auxiliary scalar field $`\varphi `$, in terms of which the conserved worldsheet current is
$$J^a=\frac{1}{\sqrt{\gamma }}ϵ^{ab}\varphi _{,b}.$$
(9)
The current is chiral if $`\varphi _{,a}`$ is a null worldsheet vector, i.e.,
$$\gamma ^{ab}\varphi _{,b}\varphi _{,a}=0,$$
(10)
in which case $`J_aJ^a=0`$. In this case, the equations of motion can be written
$`_a\left(𝒯^{ab}x_{,b}^\nu \right)`$ $`=`$ $`0`$ (12)
$`_a\left(\sqrt{\gamma }\gamma ^{ab}\varphi _{,b}\right)`$ $`=`$ $`0`$ (13)
where
$$𝒯^{ab}=\sqrt{\gamma }\left(\mu \gamma ^{ab}+\theta ^{ab}\right),$$
(14)
$`\mu `$ is the energy per unit length of the string, and $`\theta ^{ab}`$ is the worldsheet energy-momentum tensor of the charge carriers,
$$\theta ^{ab}=\gamma ^{ac}\gamma ^{bd}\varphi _{,c}\varphi _{,d}$$
(15)
in the chiral case.
As above, we would like to have a gauge in which the $`𝒯^{ab}`$ has the form
$$𝒯^{ab}=\mu \eta ^{ab}.$$
(16)
If we can accomplish this, the equation of motion will be the wave equation, Eq. (4), we can choose $`x^0=\tau `$, and the general solution will be given by Eq. (7), as before. However, we note that since $`𝒯`$ is a $`2\times 2`$ matrix,
$`det𝒯`$ $`=`$ $`(\gamma )det\left(\mu \gamma ^{ab}+\theta ^{ab}\right)`$ (17)
$`=`$ $`det\left[\gamma _{ab}\left(\mu \gamma ^{bc}+\theta ^{bc}\right)\right]=det\left(\mu \delta _a^c+\theta _a^c\right)`$ (18)
which is gauge-invariant. Since the matrices are $`2\times 2`$, the determinant is easily expanded,
$$det𝒯=\mu ^2\mu Tr\theta _a^cdet\theta _a^c.$$
(19)
Since $`\theta `$ is traceless, $`det𝒯`$ can only be $`\mu ^2`$ as required by Eq. (16), if $`det\theta _a^c=0`$. But for (and only for) a chiral current, from Eq. (15),
$$\theta _a^c=\varphi ^{,c}\varphi _{,a},$$
(20)
which is the outer product of two vectors, and thus has vanishing determinant.
Thus for a chiral current there is the possibility that Eq. (16) can be satisfied. In that case, from Eqs. (16) and (14) we see that
$$\sqrt{\gamma }\gamma ^{ab}\varphi _{,b}=\frac{1}{\mu }\left[𝒯^{ab}\sqrt{\gamma }\theta ^{ab}\right]\varphi _{,b}=\eta ^{ab}\varphi _{,b}$$
(21)
and so Eq. (13) becomes the wave equation,
$$\ddot{\varphi }\varphi ^{\prime \prime }=0.$$
(22)
We now contract Eq. (16) with $`\varphi _{,a}\varphi _{,b}`$. Since the current is chiral, $`\gamma ^{ab}\varphi _{,a}\varphi _{,b}=0`$, and using Eq. (15), $`\theta ^{ab}\varphi _{,a}\varphi _{,b}=0`$. Thus to satisfy Eq. (16), we must have $`\eta ^{ab}\varphi _{,a}\varphi _{,b}=0`$, or $`\dot{\varphi }^2=\varphi ^2`$. Without loss of generality we take the solution of the form
$$\varphi (\sigma ,\tau )=F(\sigma +\tau ),$$
(23)
which also satisfies Eq. (22).
Using Eq. (23), the condition for chirality, Eq. (10), becomes
$$\gamma ^{00}+\gamma ^{11}+2\gamma ^{01}=0.$$
(24)
Since $`\gamma ^{ab}`$ is the inverse of the $`2\times 2`$ matrix $`\gamma _{ab}`$ we have
$$\gamma ^{ab}=\frac{1}{\gamma }\left(\begin{array}{cc}\gamma _{11}& \gamma _{01}\\ \gamma _{01}& \gamma _{00}\end{array}\right),$$
(25)
so Eq. (24) implies
$$\gamma _{00}+\gamma _{11}2\gamma _{01}=0.$$
(26)
Using Eq. (2), this becomes
$$0=\dot{x}^\mu \dot{x}_\mu +x^\mu x_\mu ^{}2\dot{x}^\mu x_\mu ^{}=(\dot{x}^\mu x^\mu )(\dot{x}_\mu x_\mu ^{})$$
(27)
which means that $`\dot{x}^\mu x^\mu =(1,𝐚^{})`$ is a null 4-vector, or that
$$|𝐚^{}|=1.$$
(28)
To solve the rest of the problem, we define a matrix
$$𝒮_{ab}=\frac{1}{\sqrt{\gamma }}\left(\mu \gamma _{ab}\theta _{ab}\right).$$
(29)
From Eqs. (10) and (15), $`\theta _{ab}\theta ^{bc}=0`$, and since $`\gamma ^{ab}`$ and $`\theta ^{ab}`$ are symmetrical,
$$𝒮_{ab}𝒯^{bc}=\mu ^2\delta _a^c$$
(30)
and consequently if $`𝒯`$ has the form of Eq. (16) we have that
$$𝒮_{ab}=\mu \eta _{ab}.$$
(31)
Now, comparing the $`01`$ component of Eqs. (29) and (31) gives
$$\mu \gamma _{01}=\theta _{01}=F^2.$$
(32)
The metric component is
$$\gamma _{01}=\dot{x}^\mu x_\mu ^{}=\frac{1}{4}\left(|𝐚^{}|^2|𝐛^{}|^2\right).$$
(33)
From Eqs. (33) and (28), we find
$$1|𝐛^{}|^2=\frac{4F^2}{\mu }$$
(34)
Since the determinant of $`𝒯`$ and thus of $`𝒮`$ is fixed, it remains only to show that one more component of Eq. (31) is satisfied. For example, a sufficient condition is that
$$\mu \gamma _{00}\theta _{00}=\mu \sqrt{\gamma }.$$
(35)
Using Eq. (26) it is easy to show that
$$\sqrt{\gamma }=\frac{1}{2}(\gamma _{00}\gamma _{11}),$$
(36)
and Eq. (35) becomes
$$\mu \gamma _{00}F^2=\frac{\mu }{2}(\gamma _{00}\gamma _{11})$$
(37)
so
$$\frac{\mu }{2}(\gamma _{00}+\gamma _{11})=F^2$$
(38)
which is satisfied using Eq. (26) and Eq. (32).
Thus, Eq. (7) with the constraints given by Eqs. (28) and (34) are a general solution to the equations of motion for a superconducting string with a chiral current. This solution is the same as that found by Carter and Peter, except they did not explicitly specify the relation (34) between $`𝐛^{}`$ and $`F^{}`$, but gave instead the inequality $`|𝐛^{}|^2<1`$.
Note that in this gauge $`\sigma `$ parameterizes the total energy on the string. The energy in a region is
$$E=d^3xT_0^0,$$
(39)
where $`T_\nu ^\mu `$ is given by
$$T_\nu ^\mu (x)=𝑑\sigma 𝑑\tau 𝒯^{ab}x_{,a}^\mu x_{\nu ,b}\delta ^4[xx(\sigma ,\tau )].$$
(40)
With $`x^0=\tau `$, and using Eq. (16), the energy is
$$E=\mu 𝑑\sigma =\mu 𝑑l\frac{1}{|𝐱^{}|}$$
(41)
which means that with this parameterization the energy on the string is $`\mu \mathrm{\Delta }\sigma `$.
## IV Finding the solution from initial conditions.
We would now like to find the evolution of a string with a chiral current from given initial conditions. We suppose that we are given the position of the string at some time $`t_0`$, as a function $`𝐱(l)`$ parameterized by arc length in the laboratory frame, with $`l`$ increasing in the opposite direction to the current flow. We also need the initial charge and current as the values of the auxiliary scalar field $`\varphi (l)`$, and the perpendicular component of the string motion, $`\dot{𝐱}_{}(l)`$. Motion parallel to the string direction is dependent on the choice of parameter and has no physical meaning. From these conditions, we want to find the functions $`𝐚`$ and $`𝐛`$.
The first step is to reparameterize everything in terms of $`\sigma `$. For a stationary string, the linear energy density of the string itself is just $`\mu `$, and the energy due to the current is $`(d\varphi /dl)^2`$. Boosting the string in a transverse direction just gives the Lorentz factor $`\mathrm{\Gamma }=1/\sqrt{1|\dot{𝐱}_{}|^2}`$, so
$$\frac{dE}{dl}=\mathrm{\Gamma }\left[\mu +\left(\frac{d\varphi }{dl}\right)^2\right]$$
(42)
and thus
$$\frac{d\sigma }{dl}=\mathrm{\Gamma }\left[1+\frac{1}{\mu }\left(\frac{d\varphi }{dl}\right)^2\right].$$
(43)
Using Eq. (43) we can change parameters from $`l`$ to $`\sigma `$.
Now we need to determine the full form of $`\dot{𝐱}`$. We observe that $`\dot{𝐱}𝐱^{}=(|𝐛^{}|^2|𝐚^{}|^2)/4=\varphi ^2/\mu `$, so we can write
$$\dot{𝐱}=\dot{𝐱}_{}\frac{\varphi ^2𝐱^{}}{\mu |𝐱^{}|^2}.$$
(44)
Then
$`𝐚^{}`$ $`=`$ $`𝐱^{}\dot{𝐱}`$ (46)
$`𝐛^{}`$ $`=`$ $`𝐱^{}+\dot{𝐱}`$ (47)
and Eq. (7) gives a complete solution for the future evolution of the string.
## V Chiral String Dynamics
We have obtained the analytic general solution for superconducting strings with chiral currents. For a string with a current determined from,
$$\varphi (\sigma ,\tau )=F(\sigma +\tau ),$$
(48)
the general solution for the string is given by
$`x^0`$ $`=`$ $`\tau `$ (50)
$`𝐱`$ $`=`$ $`{\displaystyle \frac{1}{2}}[𝐚(\sigma \tau )+𝐛(\sigma +\tau )],`$ (51)
with the following constraints for the otherwise arbitrary functions $`𝐚^{}`$ and $`𝐛^{}`$,
$`|𝐚^{}|^2`$ $`=`$ $`1`$ (53)
$`|𝐛^{}|^2`$ $`=`$ $`1{\displaystyle \frac{4F^2}{\mu }}.`$ (54)
Note that $`F^{}=d\varphi /d\sigma `$ is in general not the same as the physical current, which goes as $`d\varphi /dl`$. Using Eq. (43) we can write
$$\frac{d\varphi }{dl}=\mathrm{\Gamma }\left[1+\frac{1}{\mu }\left(\frac{d\varphi }{dl}\right)^2\right]F^{}$$
(55)
so
$$F^{}=\frac{\mu (d\varphi /dl)}{[\mu +(d\varphi /dl)^2]\mathrm{\Gamma }}.$$
(56)
Thus $`F^{}`$ increases with $`d\varphi /dl`$ for small currents, but it reaches a maximum, $`F^{}=\sqrt{\mu }/(2\mathrm{\Gamma })`$ when $`d\varphi /dl=\sqrt{\mu }`$. For larger values of $`d\varphi /dl`$, $`F^{}`$ decreases again. This happens because $`F^{}`$ measures the winding number per unit energy, and the current contribution to the energy density goes as $`(d\varphi /dl)^2`$.
We now show several interesting consequences that we can extract from the result. First of all, we see that there are arbitrarily shaped static solutions for the case in which the current satisfies $`4F_{}^{}{}_{}{}^{2}/\mu =1`$. In this case $`|𝐛^{}|=0`$, so the position of the string, up to a constant vector, is given by
$$𝐱=\frac{1}{2}[𝐚(\sigma \tau )]$$
(57)
so the set of points traced by the string does not depend on time. These are vortons of any shape, which could have important cosmological consequences.
Another somewhat unexpected consequence of the exact solution is that the motion of a loop with a chiral current is strictly periodic in its rest frame. The period is $`T=E/(2\mu )`$, where E is the total energy of the loop in that frame.
We also see that chiral strings do not have true cusps. In fact, we can easily calculate the Lorentz factor that these strings can reach,
$$\mathrm{\Gamma }=\frac{1}{\sqrt{1\dot{𝐱}_{}^{}{}_{}{}^{2}}}=\sqrt{1+\left(\frac{|𝐛^{}|\mathrm{sin}\theta }{1+|𝐛^{}|\mathrm{cos}\theta }\right)^2},$$
(58)
where $`\theta `$ is the angle between $`𝐚^{}`$ and $`𝐛^{}`$. This expression has its maximum at $`\mathrm{cos}\theta =|𝐛^{}|`$, and the maximum Lorentz factor is
$$\mathrm{\Gamma }_{\mathrm{max}}=\frac{\sqrt{\mu }}{2|F^{}|}.$$
(59)
This result can be seen directly from Eq. (56). At this point, the energy density in the string itself and that in the charge carriers (see Eq. (42)), are equal.
However, this is not the maximum concentration of energy per unit length that can be achieved. Using Eq. (41) we see that the maximum energy density corresponds to the minimum of $`|𝐱^{}|`$, i.e. $`\theta =\pi `$. There,
$$|𝐱^{}|_{\mathrm{min}}=\frac{1}{2}\left(1|𝐛^{}|\right)$$
(60)
and so
$$\frac{dE}{dl}|_{\mathrm{max}}=\frac{2\mu }{1|𝐛^{}|}=\frac{2\mu }{1\sqrt{14F^2/\mu }}.$$
(61)
At this point $`𝐚^{}`$ and $`𝐛^{}`$ are antiparallel, and thus so are $`\dot{𝐱}`$ and $`𝐱^{}`$. Thus $`\dot{𝐱}_{}=0`$ and the string is not moving, so this is also the point of maximum energy density in the local rest frame.
## VI Strings with a small non-chiral current
The Carter-Peter solution for chiral strings, Eqs. (48)–(V) can also be used as an approximate description of strings having both right- and left-moving charge carriers in the case when the current is sufficiently small,
$$\dot{\varphi }^2,\varphi _{}^{}{}_{}{}^{2}\mu .$$
(62)
The contribution of charge carriers to the worldsheet energy momentum tensor is then suppressed compared to that of the string itself by a factor $`\mathrm{\Gamma }^2\varphi ^2`$, where $`\mathrm{\Gamma }`$ is the Lorentz factor of the string. (We assume for simplicity that $`\dot{\varphi }^2\varphi _{}^{}{}_{}{}^{2}`$.) The effect of charge carriers on the string dynamics will therefore be negligible, except in the vicinity of cusps where $`\mathrm{\Gamma }`$ can be very large.
The string current is also greatly enhanced in near-cusp regions. For strings with bosonic superconductivity, large currents can be unstable with respect to quenching and to quantum tunneling. A large charge density can also destabilize the condensate, resulting in ejection of charge carriers. These effects disappear for a chiral current and tend to drive the string towards the chiral limit . For strings with fermionic superconductivity, the growth of the current leads to enhanced scattering of left- and right-moving charge carriers off the string. The left- and right-moving currents are typically not equal, so the scatterings suppress the minority charge carriers, leaving the string with nearly chiral current in the near-cusp region.
We thus have a situation where away from the cusps the current is non-chiral but small and its effect on the string dynamics is negligible, while near the cusps the current is large and nearly chiral. Hence, both near cusps and away from cusps the string dynamics can be approximately described by the chiral string solution, Eqs. (48)–(V).
We shall now derive a quantitative criterion for this approximate description to be accurate. Near a cusp, the string gets contracted by a large factor, its rest energy being turned into kinetic energy. The density of charge carriers and the current are enhanced by the same factor. The contraction factor increases as we approach the tip of the cusp. The invariant length of string which attains Lorentz factor at least $`\mathrm{\Gamma }`$ is $`L/\mathrm{\Gamma }`$. Since this Lorentz factor is obtained by compressing the string, the physical length of this region of string is
$$\mathrm{\Delta }L_\mathrm{\Gamma }L/\mathrm{\Gamma }^2,$$
(63)
and since the physical string motion is perpendicular to the string, this is also the length of string in its rest frame. Let $`J=\sqrt{|J_\mu J^\mu |}`$ and let $`J_0`$ be the value of $`J`$ far from the cusp. Since the string has been compressed by factor $`\mathrm{\Gamma }`$, the current is increased by the same factor, so it becomes
$$J_\mathrm{\Gamma }\mathrm{\Gamma }J_0.$$
(64)
These values are sustained for a time interval (again in the rest frame)
$$\mathrm{\Delta }t_\mathrm{\Gamma }\mathrm{\Delta }L_\mathrm{\Gamma }L/\mathrm{\Gamma }^2.$$
(65)
We now have to check whether or not this time interval is sufficient to suppress the minority charge carriers. The answer to this depends on the scattering rate of left- and right-movers and is therefore model-dependent. As an illustration, let us suppose that we have fermionic charge carriers with large masses off the string but which can scatter inelastically by exchange of a GUT-scale gauge boson into light particles not bound to the string. Models of this sort have been studied in detail by Barr and Matheson . Their analysis indicates that the time it takes for the current to become nearly chiral is (in the rest frame of the string)
$$\tau M_X^4/J_\mathrm{\Gamma }^5,$$
(66)
where, $`M_X`$ is the GUT scale. The current will be nearly reduced to one chiral component during a single cusp occurrence, provided that this time is shorter than $`\mathrm{\Delta }t_\mathrm{\Gamma }`$. Using Eqs. (64), (65) and (66) we find that this condition is satisfied when the Lorentz factor exceeds a certain value $`\mathrm{\Gamma }_{}`$,
$$\mathrm{\Gamma }\mathrm{\Gamma }_{}\left(\frac{M_X^4}{LJ_0^5}\right)^{1/3}.$$
(67)
On the other hand, charge carriers have a substantial effect on the string dynamics when the Lorentz factor becomes comparable to $`\mathrm{\Gamma }_{\mathrm{max}}`$ from Eq. (59). Thus, we expect the approximate chiral string description to be accurate when $`\mathrm{\Gamma }_{}\mathrm{\Gamma }_{\mathrm{max}}`$, or
$$\frac{J_0}{\sqrt{\mu }}\left(\frac{M_X^2}{\mu }\right)^{5/4}(M_XL)^{1/2}.$$
(68)
This condition is extremely weak for macroscopic strings. For example, for superheavy strings with $`\mu M_X^2`$, $`\sqrt{M_XL}`$ is the maximum Lorentz factor which can be reached before overlap of the string cores causes the cusp to be truncated . Thus if Eq. (68) is violated, then neither $`\mathrm{\Gamma }_{}`$ nor $`\mathrm{\Gamma }_{\mathrm{max}}`$ can be reached, and the current has negligible effect on motion of the string.
## VII Electromagnetic radiation power from oscillating loops
If the string current is coupled to electromagnetism, then oscillating current-carrying loops emit electromagnetic radiation. Calculations disregarding the effect of charge carriers and of the electromagnetic back-reaction on the motion of the loop give an infinite radiation power. The divergence can be attributed to the infinite Lorentz factor reached at the cusp of a Nambu-Goto string. If the cusp is truncated at a maximum Lorentz factor $`\mathrm{\Gamma }_{\mathrm{max}}`$, the power can be estimated as
$$P30j^2\mathrm{\Gamma }_{\mathrm{max}}.$$
(69)
Here, $`jqJ_0`$ is the electric current away from the cusp, $`q0.1`$ is the effective charge of the charge-carrying field, and the coefficient comes from numerical calculations.
Later in this Section, we shall argue that the electromagnetic back-reaction in the near-cusp region is subdominant compared to the effect of the charge carriers and can therefore be neglected.
As discussed in the preceding Section, for sufficiently small currents the charge-carrier back-reaction is negligible away from the cusps, while near the cusps the current tends to be chiral and the solution Eqs. (48)–(V) can be used. Using this solution we can determine $`\mathrm{\Gamma }_{\mathrm{max}}`$ and therefore we can give an estimate for the electromagnetic power. Substituting Eq. (59) in Eq. (69) we obtain
$$Pqj\sqrt{\mu }.$$
(70)
This result is confirmed by more accurate calculations in .
Our neglect of the electromagnetic back-reaction can be justified as follows. The energy emitted in a single cusp event is $`\mathrm{\Delta }E_{em}PL`$, where $`L`$ is the length of the loop. The total energy of the string segment in which the maximum Lorentz factor is reached (and which is responsible for most of the radiation) is $`\mathrm{\Delta }E_s\mu L/\mathrm{\Gamma }_{\mathrm{max}}`$, and the energy of the charge carriers in that region is $`\mathrm{\Delta }E_J\mathrm{\Delta }E_s`$. Using Eqs. (59, 70), we find $`\mathrm{\Delta }E_{em}/\mathrm{\Delta }E_Jq^21`$. This suggests that the effect of electromagnetic back-reaction on the motion of the string is much smaller than that of the charge carriers.
## VIII Acknowledgments
We would like to thank Anne Davis and Xavier Siemens for helpful conversations. This work was supported in part by the National Science Foundation.
|
warning/0004/hep-ph0004119.html
|
ar5iv
|
text
|
# Fermilab-Pub-00/082-T A new method for extracting the bottom quark Yukawa coupling at the CERN Large Hadron Collider
## Abstract
We propose a new method for measuring the $`Hb\overline{b}`$ rate at the CERN Large Hadron Collider in a manner which would allow extraction of the $`b`$ quark Yukawa coupling. Higgs boson production in purely electroweak $`WHjj`$ events is calculated. The Standard Model signal rate including decays $`W\mathrm{}\nu `$ and $`Hb\overline{b}`$ is 11 fb for $`M_H=120`$ GeV. It is possible to suppress the principal backgrounds, $`Wb\overline{b}jj`$ and $`t\overline{t}jj`$, to approximately the level of the signal. As the top quark Yukawa coupling does not appear in this process, it promises a reliable extraction of $`g_{Hbb}`$ in the context of the Standard Model or some extensions, such as the MSSM.
preprint: Fermilab
The observation of electroweak symmetry breaking (ESB) in nature with the discovery of massive weak bosons is both a triumph and a tribulation of the Standard Model (SM) of particle interactions. While the SM has been incredibly successful in explaining the data from various experiments, it has also made very powerful predictions for other behavior that was later observed. To date there are no glaring contradictions with the SM, nor even any moderately uncomfortable ones. Despite this success, direct observation of the mechanism believed to be responsible for ESB, the Higgs boson, has not been made.
All quantum numbers of the SM Higgs boson are known from theory except for its mass, which is a free parameter. While fits to electroweak (EW) data suggest that the Higgs boson is considerably lighter than its theoretical upper limit of 1 TeV, probably on the order of 100 GeV , the current direct search exclusion limit of $`M_H114`$ GeV has pushed the SM Higgs boson mass above the EW fits central value. One may interpret this as a statement that we are on the verge of discovering the Higgs boson in present or near-future experiments.
If a new Higgs–like resonance is discovered at either the CERN Large Electron Positron collider, Fermilab’s Tevatron machine or the soon to be built CERN Large Hadron Collider (LHC), it will be up to the LHC to determine the quantum numbers of the resonance and state with conviction whether it is the SM Higgs boson, a Higgs boson of an extension to the SM, or something else entirely. Of primary importance are the spin and couplings, which must be $`SU(2)`$ gauge couplings to the weak bosons and Yukawa couplings to fermions, proportional to the mass of the fermion. At the LHC, it will be fairly straightforward to determine the spin as well as to extract the gauge couplings and total width of a SM-like Higgs boson . Measuring the Yukawa couplings is more difficult. Higgs boson decays to fermions are substantial only for $`M_H150`$ GeV (except for $`Ht\overline{t}`$, which is significant for $`M_H350`$ GeV); heavier Higgs bosons decay dominantly to $`W,Z`$. However, the LHC will be able to measure the $`\tau `$ Yukawa coupling quite easily for $`M_H150`$ GeV . <sup>*</sup><sup>*</sup>*Earlier studies proposed using $`b\overline{b}h,hb\overline{b}`$ events to to measure $`g_{Hbb}`$ at the LHC, but only for very large $`\mathrm{tan}(\beta )`$ in the context of the MSSM . However, detector simulations make questionable the feasibility of an effective use of this channel in practice .
This is made possible by the cleanliness and ease of observability of Higgs boson production in weak boson fusion (WBF), $`ppqqVV+XqqH+X`$, where the final state quarks appear as large invariant mass, high-pT tagging jets at far forward and backward rapidities in the detector, providing a unique signature with which to suppress the background rates. This technique would allow observation of decays to the final states $`\gamma \gamma `$ and $`\tau ^+\tau ^{}`$ for $`100<M_H150`$ GeV, and $`W^+W^{}`$ for $`M_H120`$ GeV , with signal to background (S/B) rate ratios typically much better than 1/1. This is the only technique in which the decay $`H\tau ^+\tau ^{}`$ could be observed in the SM. As such, it is the only method with which to extract a Higgs-fermion-fermion coupling. The technique of Ref. , which we rely on here, involves combining information from several Higgs boson channels to measure both the width to $`W`$ pairs, $`\mathrm{\Gamma }_W`$, which contains the $`SU(2)`$ gauge coupling of the Higgs boson to the weak bosons; and the total Higgs boson width, $`\mathrm{\Gamma }_{tot}`$. The ratio of the $`H\tau ^+\tau ^{}`$ rate to the $`HW^+W^{}`$ rate in WBF is then proportional to $`\mathrm{\Gamma }_\tau /\mathrm{\Gamma }_W`$, which allows one to determine the $`\tau `$ Yukawa coupling.
Measurement of additional Higgs Yukawa couplings is highly desirable. For example, in the MSSM there are five physical Higgs bosons, two of which have the same quantum numbers as the SM Higgs bosons, but can vary in mass, subject to the constraint that the lighter of the two states must have a mass $`M_h135`$ GeV. As such, this state will typically have substantial rate for decays to fermions. In the MSSM however, Yukawa couplings of up-type and down-type fermions can be altered relative to each other already at tree level. This characteristic of the MSSM affects also the rare decay modes (e.g. $`H\gamma \gamma `$). In addition, large radiative corrections to the Yukawa couplings can modify the tree level decay rates considerably , e.g. causing “misalignment” of the couplings to $`b`$ quarks and $`\tau `$ leptons. Thus, direct observation of more than one decay mode can provide a constraint on the model. Observation of $`Hc\overline{c}`$ or other light fermions is not likely to be possible at the LHC. $`Hb\overline{b}`$ will be extremely difficult, but possible in $`t\overline{t}H`$ associated production for Higgs boson masses below $`120`$ GeV . However, this process suffers from the complication that both an up-type and down-type Yukawa coupling are convoluted, thus leaving the model largely undetermined.
Measurement of the $`Hb\overline{b}`$ rate in WBF, $`qqVVqqHb\overline{b}jj`$, would be extremely difficult if not impossible, despite the rate being almost an order of magnitude larger than for the decay to tau pairs, as it would suffer from enormous QCD backgrounds. There is also the issue of triggering on $`b\overline{b}jj`$ events: they do not typically contain a high-$`p_T`$ lepton, so may not pass a $`E_T`$ trigger with great efficiency, thus the events would simply not be recorded. (An alternative is to design a dual tagging jet trigger.) Thus, another measurement of $`Hb\overline{b}`$ involving a measured production coupling would be highly desirable.
We seek a process that provides a high-$`p_T`$ lepton in addition to the far forward/backward tagging jets and the Higgs boson. The ideal choice is $`WHjj`$ production. Four classes of Feynman diagrams contribute to $`ppWHjj+X`$: WBF $`H`$ production with $`W`$ bremsstrahlung off a quark leg; WBF $`H`$ production with additional $`W`$ emission off the $`t`$-channel weak boson pair; and WBF $`W`$ production and $`W`$ bremsstrahlung where the Higgs boson is radiated off the $`W`$. Note that the $`WHjj`$ events we consider are not QCD corrections to $`WH`$ associated production, but are pure EW processes; QCD corrections to $`WH`$ events will ultimately constitute an enhancement of the signal, but will typically not survive the tagging jet cuts or minijet veto and so are neglected here, a conservative approximation.
We calculate the cross section for $`ppWHjj+X`$ at the LHC, $`\sqrt{s}=14`$ TeV, using full tree level matrix elements for all EW subprocesses, including finite width and off-shell effects for $`W\mathrm{}\nu `$ ($`\mathrm{}=e,\mu `$), and finite width effects for $`Hb\overline{b}`$. The matrix elements were generated by madgraph . The Higgs boson NLO decay and total widths are corrected via input from hdecay . CTEQ4L structure functions are employed with a choice of factorization scale $`\mu _{f_i}=p_{T_i}`$ of the outgoing tagging jets. To provide realistic resolution of the $`b\overline{b}`$ invariant mass, gaussian smearing of final state particle four-momenta is employed according to ATLAS expectations . We do not decay the bottom quarks explicitly, but do include a parameterized energy loss distribution to make a more realistic simulation of observed final state momenta, overall missing momentum and $`b\overline{b}`$ invariant mass. As some Feynman diagrams with a t-channel photon contribute, the total cross section, shown as a function of $`M_H`$ in line 1 of Table I, is calculated with an explicit initial-final state quark pair $`Q_{ij}^2>100`$ GeV<sup>2</sup> to avoid the singularity from the photon propagator. This cut introduces a small uncertainty for the total rate without cuts, $`\pm 15\%`$ for varying the Q<sup>2</sup> cut by a factor of 2 (1/2). It does not, however, affect the cross sections with cuts.
The total signal rate appears to be large enough to obtain a significant data sample. However, to determine whether this measurement is realistic, we calculate the cross section for the main background which can mimic the signal. The largest resonant backgrounds are QCD and EW $`WZjj;Zb\overline{b}`$ production, but in these cases the $`Z`$ pole is well-separated from the Higgs boson resonance so the overlap should be minimal given the superior detector jet resolutions. Thus, we ignore these backgrounds for the present viability check and instead concentrate on the largest backgrounds to this signal: nonresonant QCD $`Wb\overline{b}jj`$ production, and $`t\overline{t}+jets`$ events, where both $`W`$’s from the top quarks decay leptonically ($`e`$ or $`\mu `$) and one of the leptons is too low in $`p_T`$ to be observed; we take this cut to be $`p_T(l,min)<10`$ GeV for the simple check here. The latter events consist of QCD corrections to $`t\overline{t}`$ production, but are completely perturbative in the phase space region of interest, as the QCD radiation can appear in the detector as far forward/backward tagging jets. In addition, there are tree level processes that do not correspond to initial or final state gluon radiation. Other potential backgrounds are primarily irreducible, or fake signatures, which are naturally expected to be subdominant to continuum production.
We calculate the $`Wb\overline{b}jj`$ and $`t\overline{t}jj`$ rates using exact tree-level matrix elements, constructed using madgraph for the former, and the latter from Ref. . We include top quark and $`W`$ leptonic decays to $`e,\mu `$ in the matrix elements. CTEQ4L structure functions are employed throughout. We take the factorization scale for the $`Wb\overline{b}jj`$ background the same as the signal, and the renormalization scale $`\mu _r=\sqrt{\mu _{f,1}\mu _{f,2}}`$. For the $`t\overline{t}jj`$ background, $`\mu _f=min(E_T)`$ of the jets/top quarks, and renormalization scale $`\mu _r=E_T(jet/top)`$, with one factor of $`\alpha _s`$ taken from each of the outgoing jets/top quarks. In all cases, $`\alpha _s(M_Z)=0.118`$ with 1-loop running.
The basic WBF signature requires the two tagging jets to be at high rapidity and in opposite hemispheres of the detector, and the $`H,W`$ decay products to be central and in between the tagging jets. The kinematic requirements for the “rapidity gap” level of cuts are as follows:
$`p_{T_j}30\mathrm{GeV},|\eta _j|5.0,\mathrm{}R_{jj}0.6,`$ (1)
$`p_{T_b}15\mathrm{GeV},|\eta _b|2.5,\mathrm{}R_{jb}0.6,`$ (2)
$`p_T_{\mathrm{}}20\mathrm{GeV},|\eta _{\mathrm{}}|2.5,\mathrm{}R_{j\mathrm{},b\mathrm{}}0.6,`$ (3)
$`\eta _{j,min}+0.7<\eta _{b,\mathrm{}}<\eta _{j,max}0.7,`$ (4)
$`\eta _{j_1}\eta _{j_2}<0,\mathrm{}\eta _{tags}=|\eta _{j_1}\eta _{j_2}|4.4.`$ (5)
The results for $`M_H=120`$ GeV are shown in the first column of Table II. At this level the backgrounds are already somewhat manageable, but we observe that the QCD $`Wb\overline{b}jj`$ background is dominated by low invariant masses of the tagging jet pair and low-$`p_T`$ $`b`$ jets, so we impose a minimum tagging dijet mass and an additional staggered $`p_T(b)`$ cut to reduce this contribution:
$$m_{jj}>600\mathrm{GeV},p_T(b_1,b_2)>50,20\mathrm{GeV}.$$
(6)
The $`m_{jj}`$ cut is also somewhat effective against $`t\overline{t}jj`$ events, as shown in the second column of Table II. Furthermore, there are two strikingly different characteristics of the signal v. the $`t\overline{t}jj`$ background: the latter events have significantly higher $`/p_T`$ on average; and they do not exhibit a Jacobian peak in the $`m_T(\mathrm{},/p_T)`$ distribution, a characteristic of $`W`$ decays. Both features are due to the fact that by suppressing observation of the second charged lepton, the neutrino from that $`W`$’s decay has significantly enhanced transverse momentum, which is unobserved, and greatly distorts the $`m_T(\mathrm{},/p_T)`$ distribution. We choose maximum cutoff values for both observables as follows:
$$/p_T<100\mathrm{GeV},m_T(\mathrm{},/p_T)<100\mathrm{GeV}.$$
(7)
The result is shown in the third column of Table II.
A final cut that may be utilized is to reject any candidate event if it contains additional central QCD (jet) activity of moderate $`p_T`$: a minijet . Studies of the minijet rate for WBF, EW and QCD events can be found in Refs. and references therein. Here, we simply apply the results from those studies. The probability of a signal event surviving a minijet veto, $`p_T^{veto}(j)>20`$ GeV, is estimated to be $`75\%`$, slightly lower than that typical of WBF Higgs boson events, because the $`W`$ bremsstrahlung components of $`WHjj`$ events can slightly enhance the minijet activity. The probability of a $`t\overline{t}jj`$ background event surviving a minijet veto is much lower, only about $`30\%`$; previous studies have indicated it may be even better than this for $`t\overline{t}+jets`$ events, but we choose to remain conservative for our proof of concept estimates here. This allows us to achieve a S/B rate of 1/2 for $`M_H=120`$ GeV, considering only the two major backgrounds for this demonstration.
At this stage, shown in the fourth column of Table II, the situation appears quite good. However, in reality only about $`25\%`$ of these events will be captured in the data sample due to detector efficiencies. We take the expected values for ATLAS and CMS to be $`86\%`$ for each tagging jet, $`95\%`$ for the charged lepton, and $`60\%`$ for each $`b`$ quark tag. We assume 100 fb<sup>-1</sup> of data for each of two experiments, as in Ref. . Our result at this point is comparable to that expected for the $`t\overline{t}H;Hb\overline{b}`$ channel with $`M_H=120`$ GeV: ATLAS expects the significance to be $`3.6\sigma `$, with S/B $`1/3`$ . However, the $`b\overline{b}`$ mass peak in $`t\overline{t}H`$ events does not clearly stand out, and again, the value $`g_{Hbb}`$ is not easily extractable due to convolution with the top quark coupling.
To illustrate that our method would not be simply a counting experiment, rather a distinct resonance could be observed, Fig. 1 shows the invariant mass spectrum for the tagged $`b`$ quark pair. The $`Wb\overline{b}jj`$ and $`t\overline{t}jj`$ background distributions combined are monotonically decreasing above $`m_{b\overline{b}}=80`$ GeV, allowing the signal to present a clear peak in the spectrum for an intermediate mass Higgs boson.
We can make a preliminary rough estimate of the uncertainty in the measurement by calculating $`\mathrm{}\sigma /\sigma \sqrt{N_S+N_B}/N_S20\%`$, including a $`20\%`$ systematic uncertainty in the backgrounds. This yields about a $`35\%`$ overall uncertainty in the cross section measurement for $`M_H=120`$ GeV. As this error will dominate the extraction of $`\mathrm{\Gamma }(Hb\overline{b})`$, we may expect the overall measurement error for the partial width to $`b`$ quarks to be $`𝒪(50\%)`$ for a Higgs boson of this mass. More precise estimates must wait for a more complete consideration of the signal and backgrounds .
We have demonstrated the feasibility of a measurement of the $`Hb\overline{b}`$ decay rate at the LHC in a moderate background environment with reasonable luminosity, in a way that would allow extraction of the bottom quark Yukawa coupling. The technique shows promise for Higgs boson masses in the range of applicability to the MSSM, but would also be accessible in the SM or other SM-like extensions. More importantly, it does this independently of up-type Yukawa couplings (e.g. top quark), thus reducing model dependence. Due to the accuracy with which $`g_{HWW}`$, $`g_{H\tau \tau }`$ and the total Higgs boson width will be measured at the LHC , $`g_{Hbb}`$ can be determined by taking either the ratio $`\mathrm{\Gamma }_b/\mathrm{\Gamma }_W`$ or $`\mathrm{\Gamma }_b/\mathrm{\Gamma }_\tau `$, depending on the Higgs boson mass and how SM-like the observed state is. To be sure, this channel will not provide a precision measurement, but all that is needed to constrain models other than the SM, or to rule out regions of the MSSM parameter space, is a nonzero width measurement. A more detailed study of this process including other backgrounds is underway . Additional statistical significance may be added by using $`\gamma Hjj`$ production, which is also being studied. Early indications are that the data sample would approximately double using both production modes.
###### Acknowledgements.
We thank U. Baur, S. Dawson and D. Zeppenfeld for constructive comments and U. Baur, S. Parke and T. Plehn for a critical review of the manuscript. Fermilab is operated by URA under DOE contract No. DE-AC02-76CH03000.
|
warning/0004/cond-mat0004388.html
|
ar5iv
|
text
|
# Phonon-Induced Renormalization and Interaction: An Improvement on Frohlich Transformation
## Abstract
Starting from the Flohlich electron-phonon model, which has a history of 50 years, a new unitary transformation is proposed to implement the perturbation treatment. Our main results are: (1)The phonon-induced interaction shows a crossover from the BCS-like potential when the phonon frequency $`\omega _p`$ is much smaller than the Fermi energy $`E_F`$ to that of the small polarons when $`\omega _p/E_F1`$. (2)The jump of momentum distribution of electron number $`n_𝐤`$ at the Fermi surface goes to zero when the dimensionless coupling constant $`\lambda `$ increases to the critical value $`\lambda _c1`$, which means a possible broken down of the Fermi-liquid description.
PACS numbers: 71.38.+i; 74.20.-z
It is a well-known fact that within the Migdal-Eliashberg (ME) description of electrons and phonons coupled by the linear electron-phonon interaction there is no instability (where the Fermi-liquid description may break down) at any value of the dimensionless coupling constant $`\lambda `$. People tried to study the polaronic collapse of the electron band starting from the Lang-Firsov (LF) transformation followed by the small polaron approximation. Note that the LF transformation together with the small polaron approximation cannot lead to those results which can be obtained via the ME approach. For example, Bardeen, Cooper, and Schrieffer (BCS) proposed a square-well potential for electrons in momentum space, that is, only those electrons within a layer of width $`\omega _p`$ (the characteristic phonon frequency) near the Fermi surface can attract with each other. This potential is localized in momentum space, so it is extended in real space and leads to large coherence length of Cooper pairs. But the LF transformation results in an attractive potential for all electrons in Fermi sea, which is a localized potential in real space. As far as we know, there is no theories which can describe successfully the crossover between the two pictures. In this letter we propose a new approach which can (1)describe the crossover and (2)lead to a possible broken down of the Fermi-liquid description.
The Frohlich Hamiltonian of electron-phonon coupling system is
$`H={\displaystyle \underset{𝐤,\sigma }{}}(ϵ_𝐤\mu _0)d_{𝐤,\sigma }^{}d_{𝐤,\sigma }+{\displaystyle \underset{𝐪}{}}\omega _𝐪b_𝐪^{}b_𝐪+{\displaystyle \frac{1}{\sqrt{N}}}{\displaystyle \underset{𝐪}{}}{\displaystyle \underset{𝐤,\sigma }{}}g_𝐪d_{𝐤+𝐪,\sigma }^{}d_{𝐤,\sigma }(b_𝐪^{}+b_𝐪).`$ (1)
$`N`$ is the number of sites, $`ϵ_𝐤`$ is the bare band function. $`b_𝐪`$ and $`d_{𝐤,\sigma }`$ are usual notations for phonon or electron operators, respectively. $`\mu _0`$ is the chemical potential, $`\omega _𝐪`$ the phonon frequency and $`g_𝐪`$ the electron-phonon coupling. We set $`\mathrm{}=1`$ and $`k_B=1`$.
Frohlich used a unitary transformation to treat $`H`$, $`H^{}=\mathrm{exp}(S)H\mathrm{exp}(S)`$. The transformation can proceed order by order,
$$H^{}=H_0+H_1+[S,H_0]+[S,H_1]+\frac{1}{2}[S,[S,H_0]]+O(g_𝐪^3),$$
where $`H=H_0+H_1`$ and $`H_0`$ contains the first two terms and $`H_1`$ the last one. Frohlich let $`H_1+[S,H_0]=0`$ to get the generator $`S`$. The transformation leads to a phonon-induced interaction with the potential
$`V_F(𝐤+𝐪,𝐤)={\displaystyle \frac{g_𝐪^2\omega _𝐪}{|ϵ_{𝐤+𝐪}ϵ_𝐤|^2\omega _𝐪^2}},`$ (2)
which can be attarctive or repulsive with a singularity at the energy shell: $`|ϵ_{𝐤+𝐪}ϵ_𝐤|=\omega _𝐪`$. Frohlich noted that, although the transformation can eliminate the first order terms completely, because of the singularity one has to set a constraint $`||ϵ_{𝐤+𝐪}ϵ_𝐤|\omega _𝐪|>ϵ`$ ($`ϵ>0`$ is a constant) in the transformation. Because of the constraint the elimination of the first-order terms is not conplete.
The BCS theory simplifies the Frohlich potential as
$`V_{BCS}(𝐤+𝐪,𝐤)=\{\begin{array}{cc}g^2/\omega _p,\hfill & \text{for}|ϵ_{𝐤+𝐪}ϵ_𝐤|<\omega _p,\hfill \\ 0,\hfill & \text{otherwise}.\hfill \end{array}`$ (5)
We propose to improve the Frohlich transformation by a new generator
$`S={\displaystyle \frac{1}{\sqrt{N}}}{\displaystyle \underset{𝐪}{}}{\displaystyle \underset{𝐤,\sigma }{}}{\displaystyle \frac{g_𝐪}{\omega _𝐪}}d_{𝐤+𝐪,\sigma }^{}d_{𝐤,\sigma }(b_𝐪^{}b_𝐪)\delta (𝐤+𝐪,𝐤),`$ (6)
where $`\delta (𝐤+𝐪,𝐤)`$ is a function of the energies of incoming and outgoing electrons in the electron-phonon scattering process,
$`\delta (𝐤+𝐪,𝐤)=\delta (ϵ_{𝐤+𝐪},ϵ_𝐤;\omega _𝐪)=\left(1+{\displaystyle \frac{|ϵ_{𝐤+𝐪}ϵ_𝐤|}{\omega _𝐪}}\right)^1.`$ (7)
The reason of choosing this generator will become clear lator. After transformation the first order terms in $`H^{}`$ are
$`H_{I1}=H_1+[S,H_0]={\displaystyle \frac{1}{\sqrt{N}}}{\displaystyle \underset{𝐪}{}}{\displaystyle \underset{𝐤,\sigma }{}}{\displaystyle \frac{g_𝐪}{\omega _𝐪+|ϵ_{𝐤𝐪}ϵ_𝐤|}}d_{𝐤+𝐪,\sigma }^{}d_{𝐤,\sigma }`$
$`\times \left\{[|ϵ_{𝐤𝐪}ϵ_𝐤|(ϵ_{𝐤𝐪}ϵ_𝐤)]b_𝐪^{}+[|ϵ_{𝐤𝐪}ϵ_𝐤|+(ϵ_{𝐤𝐪}ϵ_𝐤)]b_𝐪\right\}.`$ (8)
The second order terms in $`H^{}`$ are:
$`H_{I2}=[S,H_1]+{\displaystyle \frac{1}{2}}[S,[S,H_0]]`$
$`={\displaystyle \frac{1}{2N}}{\displaystyle \underset{𝐤,𝐪}{}}{\displaystyle \underset{\sigma }{}}{\displaystyle \frac{g_𝐪^2}{\omega _𝐪}}(b_𝐪^{}b_𝐪+b_𝐪b_𝐪^{})\delta ^2(𝐤+𝐪,𝐤){\displaystyle \frac{ϵ_𝐤ϵ_{𝐤+𝐪}}{\omega _𝐪}}\left(d_{𝐤+𝐪,\sigma }^{}d_{𝐤+𝐪,\sigma }d_{𝐤,\sigma }^{}d_{𝐤,\sigma }\right)`$
$`{\displaystyle \frac{1}{N}}{\displaystyle \underset{𝐤,𝐤^{},𝐪}{}}{\displaystyle \underset{\sigma ,\sigma ^{}}{}}{\displaystyle \frac{g_𝐪^2}{\omega _𝐪}}\delta (𝐤+𝐪,𝐤)[2\delta (𝐤^{}𝐪,𝐤^{})]d_{𝐤+𝐪,\sigma }^{}d_{𝐤,\sigma }d_{𝐤^{}𝐪,\sigma ^{}}^{}d_{𝐤^{},\sigma ^{}},`$ (9)
where the non-diagonal phonon terms are neglected. $`H_{I2}`$ contains a phonon-induced interaction with potential
$`V(𝐤+𝐪,𝐤)={\displaystyle \frac{g_𝐪^2}{\omega _𝐪}}\delta (𝐤+𝐪,𝐤)[2\delta (𝐤,𝐤+𝐪)].`$ (10)
Fig.1 shows $`V(𝐤^{},𝐤)`$ as functions of $`ϵ_𝐤^{}ϵ_𝐤`$ for different ratios $`\omega _p/E_F`$ where $`E_F`$ is the Fermi energy. For comparison we also show $`V_{BCS}(𝐤^{},𝐤)`$, which is a narrow square-well in the middle, and the potential for small polarons which can be obtained via the LF transformation: $`V_{LF}=g^2/\omega _p`$ for all electrons.
The purpose of our transformation is to find a better way to divide the Hamiltonian into the unperturbed part and the perturbation. After transformation $`H^{}=H_0^{}+H_{I1}+H_{I2}^{}+O(g_𝐪^3)`$. The unperturbed part is
$`H_0^{}={\displaystyle \underset{𝐪}{}}\omega _𝐪b_𝐪^{}b_𝐪+{\displaystyle \underset{𝐤,\sigma }{}}(E_𝐤\mu _0)d_{𝐤,\sigma }^{}d_{𝐤,\sigma }{\displaystyle \underset{𝐤,\sigma }{}}\mathrm{\Delta }(𝐤)\left(d_{𝐤,}^{}d_{𝐤,}^{}+d_{𝐤,}d_{𝐤,}\right),`$ (11)
$`E_𝐤=ϵ_𝐤+{\displaystyle \frac{1}{N}}{\displaystyle \underset{𝐪}{}}{\displaystyle \frac{g_𝐪^2}{\omega _𝐪}}\mathrm{coth}({\displaystyle \frac{\omega _𝐪}{2T}}){\displaystyle \frac{ϵ_{𝐤+𝐪}ϵ_𝐤}{\omega _𝐪}}\delta ^2(𝐤+𝐪,𝐤)`$
$`{\displaystyle \frac{1}{N}}{\displaystyle \underset{𝐪}{}}{\displaystyle \frac{g_𝐪^2}{\omega _𝐪}}\delta (𝐤+𝐪,𝐤)[2\delta (𝐤+𝐪,𝐤)]{\displaystyle \frac{E_{𝐤+𝐪}\mu _0}{\xi _{𝐤+𝐪}}}\mathrm{tanh}{\displaystyle \frac{\xi _{𝐤+𝐪}}{2T}},`$ (12)
$`\mathrm{\Delta }(𝐤)={\displaystyle \frac{1}{N}}{\displaystyle \underset{𝐪}{}}{\displaystyle \frac{g_𝐪^2}{\omega _𝐪}}\delta (𝐤+𝐪,𝐤)[2\delta (𝐤+𝐪,𝐤)]{\displaystyle \frac{\mathrm{\Delta }(𝐤+𝐪)}{\xi _{𝐤+𝐪}}}\mathrm{tanh}{\displaystyle \frac{\xi _{𝐤+𝐪}}{2T}},`$ (13)
$`\xi _𝐤=\sqrt{(E_𝐤\mu _0)^2+\mathrm{\Delta }^2(𝐤)},`$ (14)
which can be solved exactly. The perturbation is $`H_{I1}+H_{I2}^{}`$ where $`H_{I2}^{}=H_{I2}(H_0^{}H_0)`$. $`E_𝐤`$ and $`\mathrm{\Delta }(𝐤)`$ have been determined by the condition that the lowest order contribution of $`H_{I2}^{}`$ to the self-energy is zero.
A renormalized chemical potential $`\mu `$ and a band renormalization factor $`\rho (ϵ_𝐤)`$ can be introduced, $`E_𝐤\mu _0=\rho (ϵ_𝐤)(ϵ_𝐤\mu )`$. The equation to determine $`\mu `$ is
$`1n={\displaystyle \frac{1}{N}}{\displaystyle \underset{𝐤}{}}{\displaystyle \frac{\rho (ϵ_𝐤)(ϵ_𝐤\mu )}{\xi _𝐤}}\mathrm{tanh}{\displaystyle \frac{\xi _𝐤}{2T}},`$
where $`n`$ is the electron number density. The equation to determine $`\rho (ϵ_𝐤)`$ will be given later.
The Green’s function of $`H_0^{}`$ is
$`G_0(𝐤,\omega )=(\omega +E_𝐤\mu _0)/(\omega ^2\xi ^2(𝐤)),\text{ }F_0^{}(𝐤,\omega )=\mathrm{\Delta }(𝐤)/(\omega ^2\xi ^2(𝐤)).`$
The contribution of $`H_{I1}`$ to the self-energy (to the second order of $`g_𝐪`$) iscitemah
$`\mathrm{\Sigma }(𝐤,\omega )={\displaystyle \frac{1}{N}}{\displaystyle \underset{𝐪}{}}{\displaystyle \frac{g_𝐪^2}{(\omega _𝐪+|ϵ_{𝐤𝐪}ϵ_𝐤|)^2}}{\displaystyle \frac{1}{\beta }}{\displaystyle \underset{n}{}}G_0(𝐤𝐪,\omega i\omega _n)`$
$`\times \left\{{\displaystyle \frac{[|ϵ_{𝐤𝐪}ϵ_𝐤|(ϵ_{𝐤𝐪}ϵ_𝐤)]^2}{i\omega _n\omega _𝐪}}{\displaystyle \frac{[|ϵ_{𝐤𝐪}ϵ_𝐤|+(ϵ_{𝐤𝐪}ϵ_𝐤)]^2}{i\omega _n+\omega _𝐪}}\right\},`$ (15)
$`W(𝐤,\omega )=0.`$ (16)
The abnormal self-energy $`W(𝐤,\omega )=0`$ because it contains a factor $`|ϵ_{𝐤𝐪}ϵ_𝐤|^2(ϵ_{𝐤𝐪}ϵ_𝐤)^2`$ in the $`𝐪`$-summation. Generally, the normal self-energy $`\mathrm{\Sigma }(𝐤,\omega )0`$. But for the normal state ($`\mathrm{\Delta }(𝐤)=0`$) and when $`T=0`$, we have
$`\mathrm{\Sigma }_n(ϵ_𝐤=\mu ,\omega )=0,`$ (17)
that is, the normal self-energy is zero at the Fermi surface. As the spectrum of elementary excitations in the normal state is $`\omega =\rho (ϵ_{𝐤𝐪})(ϵ_{𝐤𝐪}\mu )+\mathrm{\Sigma }_n(𝐤,\omega )`$, the mass renormalization at Fermi surface is
$`{\displaystyle \frac{m}{m^{}}}=\left[\rho (ϵ_𝐤)+{\displaystyle \frac{}{ϵ_𝐤}}\mathrm{\Sigma }_n(𝐤,\omega )\right]/\left[1{\displaystyle \frac{}{\omega }}\mathrm{\Sigma }_n(𝐤,\omega )\right]|_{ϵ_𝐤=\mu }=\rho (ϵ_𝐤=\mu ).`$ (18)
Eqs.(14), (15), and (16) are main reasons for the choice of the functional form of $`\delta (𝐤+𝐪,𝐤)`$ in (5). Now one can see clearly the purpose of our unitary transformation: The transformed $`H^{}`$ is divided into the unperturbed part $`H_0^{}`$, which contains the main physics of the problem, and the perturbation $`H_{I1}+H_{I2}^{}`$, which is small as shown by Eqs.(14), (15), and (16).
Gap equation (11) can be rewritten as an integral equation by introducing the Eliashberg function $`\alpha ^2F`$:
$`\mathrm{\Delta }(ϵ)={\displaystyle _D^D}𝑑ϵ^{}{\displaystyle _0^{\mathrm{}}}𝑑\mathrm{\Omega }{\displaystyle \frac{N(ϵ^{})}{N(\mu )}}\alpha ^2F(\mathrm{\Omega }){\displaystyle \frac{\mathrm{\Omega }+2|ϵ^{}ϵ|}{(\mathrm{\Omega }+|ϵ^{}ϵ|)^2}}{\displaystyle \frac{\mathrm{\Delta }(ϵ^{})}{\xi (ϵ^{})}}\mathrm{tanh}{\displaystyle \frac{\xi (ϵ^{})}{2T}}.`$ (19)
$`N(ϵ)`$ is the density of states (DOS). The integration over $`ϵ^{}`$ is from the bottom ($`D`$) of the band to the top ($`D`$). The equation for $`\rho (ϵ_𝐤)`$ can be rewritten in the same way,
$`\rho (ϵ)=1+{\displaystyle _D^D}𝑑ϵ^{}{\displaystyle _0^{\mathrm{}}}𝑑\mathrm{\Omega }{\displaystyle \frac{N(ϵ^{})}{N(\mu )}}\alpha ^2F(\mathrm{\Omega })\mathrm{coth}({\displaystyle \frac{\mathrm{\Omega }}{2T}})`$
$`\times \left\{{\displaystyle \frac{ϵ^{}ϵ}{(\mathrm{\Omega }+|ϵ^{}ϵ|)^2}}{\displaystyle \frac{ϵ^{}\mu }{(\mathrm{\Omega }+|ϵ^{}\mu |)^2}}\right\}/(ϵ_𝐤\mu )`$
$`{\displaystyle _D^D}dϵ^{}{\displaystyle _0^{\mathrm{}}}d\mathrm{\Omega }{\displaystyle \frac{N(ϵ^{})}{N(\mu )}}\alpha ^2F(\mathrm{\Omega })\{{\displaystyle \frac{\mathrm{\Omega }+2|ϵ^{}ϵ|}{(\mathrm{\Omega }+|ϵ^{}ϵ|)^2}}`$
$`{\displaystyle \frac{\mathrm{\Omega }+2|ϵ^{}\mu |}{(\mathrm{\Omega }+|ϵ^{}\mu |)^2}}\}{\displaystyle \frac{\rho (ϵ^{})(ϵ^{}\mu )}{(ϵ\mu )\xi (ϵ^{})}}\mathrm{tanh}{\displaystyle \frac{\xi (ϵ^{})}{2T}}.`$ (20)
These two equations are very similar to the Eliashberg equations.
For calculating the physical quantities we must calculate the thermodynamical potential and the average of electron or phonon operators. The thermodynamical potential is
$`\mathrm{\Omega }={\displaystyle \frac{1}{\beta }}\mathrm{ln}\text{Tr}\mathrm{exp}[\beta H]={\displaystyle \frac{1}{\beta }}\mathrm{ln}\text{Tr}\mathrm{exp}[\beta H^{}]{\displaystyle \frac{1}{\beta }}\mathrm{ln}\text{Tr}\mathrm{exp}[\beta H_0^{}].`$ (21)
The last ”$``$” is because of Eqs.(14) and (15), that is, to the order $`O(g_𝐪^2)`$ the contribution of $`H_{I1}+H_{I2}^{}`$ is very small in the lowest temperature region. Hence, for the normal state
$`\mathrm{\Omega }{\displaystyle \frac{2}{\beta N}}{\displaystyle \underset{𝐤}{}}\mathrm{ln}\left\{1+\mathrm{exp}[\beta \rho (ϵ_𝐤)(ϵ_𝐤\mu )]\right\}+\text{phonon part}.`$ (22)
The heat capacity can be calculated as $`C=T^2\mathrm{\Omega }/T^2=C_0/\rho (ϵ_𝐤=\mu )`$, where $`C_0`$ is the heat capacity for free electrons and
$`\rho (ϵ_𝐤=\mu )=1\lambda +\lambda {\displaystyle \frac{\omega _p(\omega _p+D)}{(\omega _p+D)^2\mu ^2}}.`$ (23)
In calculating we assume a constant DOS and $`\lambda =2𝑑\mathrm{\Omega }\alpha ^2F(\mathrm{\Omega })/\mathrm{\Omega }`$. The enhancement factor at $`\omega _p/D0`$ is $`1/(1\lambda )`$ which should be compared with the same factor in ME theory $`1+\lambda `$.
We have to take into account the effect of the unitary transformation when calculating the thermodynamical average of electron or phonon operators. The Green’s function for the original Hamiltonian $`H`$
$`\stackrel{~}{G}(𝐤,\tau )=\text{Tr}\left\{T_\tau \mathrm{exp}[\beta (H\mathrm{\Omega })]d_{𝐤,\sigma }(\tau )d_{𝐤,\sigma }^{}\right\}`$
$`=\text{Tr}\left\{T_\tau \mathrm{exp}[\beta (H^{}\mathrm{\Omega })]d_{𝐤,\sigma }^{}(\tau )d_{𝐤,\sigma }^{}\right\},`$ (24)
where $`T_\tau `$ means $`\tau `$ ordering. The transformation of a single fermion operator can proceed as
$$d_{𝐤,\sigma }^{}=e^Sd_{𝐤,\sigma }e^S=d_{𝐤,\sigma }+[S,d_{𝐤,\sigma }]+\frac{1}{2}[S,[S,d_{𝐤,\sigma }]]+O(g_𝐪^3).$$
$`\stackrel{~}{G}(𝐤,\omega )={\displaystyle _0^\beta }𝑑\tau \stackrel{~}{G}(𝐤,\tau )e^{\omega \tau }`$
$`=G(𝐤,\omega ){\displaystyle \frac{1}{N}}{\displaystyle \underset{𝐪}{}}{\displaystyle \frac{g_𝐪^2}{\omega _𝐪^2}}\delta ^2(𝐤,𝐤𝐪)G_0(𝐤,\omega )`$
$`{\displaystyle \frac{1}{N}}{\displaystyle \underset{𝐪}{}}{\displaystyle \frac{g_𝐪^2}{\omega _𝐪^2}}\delta ^2(𝐤,𝐤𝐪){\displaystyle \frac{1}{\beta }}{\displaystyle \underset{n}{}}{\displaystyle \frac{2\omega _𝐪}{(i\omega _n)^2\omega _𝐪^2}}G_0(𝐤𝐪,\omega i\omega _n),`$
where $`G(𝐤,\omega )`$ is the Green’s function of $`H^{}`$. Starting from the Green’s function one can calculate various physical quantities. As an example, we calculate the number of electrons in a momentum state $`𝐤`$ for the normal state when $`T=0`$,
$`n_𝐤={\displaystyle \frac{1}{\pi }}{\displaystyle _{\mathrm{}}^0}𝑑\omega \text{Im}\stackrel{~}{G}(𝐤,\omega +i0^+)`$
$`=\left(1{\displaystyle \frac{1}{N}}{\displaystyle \underset{𝐪}{}}{\displaystyle \frac{g_𝐪^2}{\omega _𝐪^2}}\delta ^2(𝐤,𝐤𝐪)\right)\theta (\mu ϵ_𝐤)+{\displaystyle \frac{1}{N}}{\displaystyle \underset{𝐪}{}}{\displaystyle \frac{g_𝐪^2}{\omega _𝐪^2}}\delta ^2(𝐤,𝐤𝐪)\theta (\mu ϵ_{𝐤𝐪}).`$
Fig.2 shows $`n_𝐤`$ as functions of $`ϵ_𝐤`$ around the Fermi surface. For smaller coupling (dashed line) or the larger frequency (dash-dotted line), there is a finite jump of $`n_𝐤`$ at Fermi surface. But for smaller frequency (solid line) and $`lambda1`$, the Fermi surface is smeared by the electron-phonon coupling. The jump of $`n_𝐤`$ at Fermi surface $`ϵ_𝐤=\mu `$ is $`\rho (ϵ_𝐤=\mu )`$ (Eq.(25)) and it predicts an instability of the Fermi liquid state at
$`\lambda _c=\left(1{\displaystyle \frac{\omega _p(\omega _p+D)}{(\omega _p+D)^2\mu ^2}}\right)^1.`$ (25)
$`\lambda _c1`$ when $`\omega _p0`$. For comparison, ME theory predicts a jump $`1/(1+\lambda )`$ and there is no instability.
At the end, as the transformation is truncated after the second order of $`g_𝐪`$, we justify the cutoff by showing the small expansion parameter. Roughly speaking, in three-dimension it is $`\lambda \omega _p/E_F`$ when $`\omega _p/E_F<1`$ but $`\lambda E_F/\omega _p`$ when $`\omega _p/E_F>1`$citezh. Note that they are the same as that of ME theory or LF transformation.
Figure Captions
Fig.1 The phonon-induced interaction.
Fig.2 The phonon-induced renormalization of the momentum distribution $`n_𝐤`$ on and near the Fermi surface.
|
warning/0004/hep-th0004068.html
|
ar5iv
|
text
|
# 1 Origins of the Enhançon
## 1 Origins of the Enhançon
Consider compactifying ten dimensional type II string theory on the four dimensional K3 surface of volume $`V`$. This gives a six dimensional theory with $`𝒩=2`$ supersymmetry; in other words, sixteen supercharges. Consider further wrapping $`N`$ D$`(p+4)`$–branes on the $`K3`$. Then there is an effective $`p`$–dimensional extended object in the six dimensions. It is in fact a BPS solution, and there are eight supercharges preserved by this situation. Furthermore, there is an $`SU(N)`$ pure gauge theory with eight supercharges on the $`(p+1)`$–dimensional world volume of the BPS soliton. This soliton has a description as a bound state of the wrapped brane and a negatively charged D$`p`$–brane. We shall refer to this as the D$`(p+4)`$–D$`p^{}`$ system, where the asterisk () is to remind us that this is not an ordinary D$`p`$ brane, since that would be an instanton.
Let us focus on $`p=2`$, hence studying type IIA. The supergravity theory contains twenty–four $`U(1)`$’s coming from the various R-R potentials in the theory. Of these, twenty–two come from wrapping the two–form on the 19+3 two–cycles of $`K3`$. The remaining two are special $`U(1)`$’s for our purposes: One of them arises from wrapping the five–form entirely on $`K3`$, while the final one is simply the plain one–form already present in the uncompactified theory.
In fact, the BPS soliton is actually a monopole of one of the six dimensional $`U(1)`$’s. It is obvious which $`U(1)`$ this is; the diagonal combination of the two special ones we mentioned above. Actually, we can simply ignore the 2 spatial directions in which the soliton is extended and see that the monopole sector (recall that it is also coupled to gravity) is nothing more than the usual problem of monopoles in a 3+1 dimensional gauge theory with an adjoint Higgs. The first order “Bogomolnyi” equations are:
$`B_i{\displaystyle \frac{1}{2}}ϵ_{ijk}F_{jk}=D_i\mathrm{H},\mathrm{with}`$
$`F_{ij}=_iA_j_jA_i+[A_i,A_j];D_i\mathrm{H}=_i\mathrm{H}+[A_i,\mathrm{H}],`$ (1.1)
with gauge invariance $`(g(𝐱)SU(2))`$:
$$A_ig^1A_ig+g^1_ig;\mathrm{H}g^1\mathrm{H}g.$$
(1.2)
Static, finite energy monopole solutions satisfy
$$\mathrm{H}(𝐱)\frac{1}{2}\mathrm{Tr}\left[\mathrm{H}^{}\mathrm{H}\right]H\mathrm{as}r\mathrm{},$$
(1.3)
where $`𝐱=(x_1,x_2,x_3)`$ and $`r^2=x_1^2+x_2^2+x_3^2`$. The $`SU(2)`$ is spontaneously broken to $`U(1)`$, by the Higgs vacuum expectation value (“vev”) $`H`$, whose magnetic charge the monopoles carry. For orientation, and for later use, the explicitly known fields of the one monopole solution is:
$`\mathrm{H}(r)={\displaystyle \frac{1}{r}}\left(\mathrm{coth}r{\displaystyle \frac{1}{r}}\right)\mathrm{i}\sigma _ix_i;A_i(r)={\displaystyle \frac{1}{r}}\left({\displaystyle \frac{1}{\mathrm{sinh}r}}{\displaystyle \frac{1}{r}}\right)\mathrm{i}ϵ_{ijk}\sigma _jx_k,`$ (1.4)
where $`𝐱=(0,0,r)`$ and
$$\sigma _1=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right);\sigma _2=\left(\begin{array}{cc}0& \mathrm{i}\\ \mathrm{i}& 0\end{array}\right);\sigma _3=\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right).$$
(1.5)
Note that it is spherically symmetric and has been normalized (for later use) such that $`H1r^1+\mathrm{}`$, as $`r\mathrm{}`$, with unit magnetic charge.
Where did the $`SU(2)`$ come from? When the $`K3`$’s volume reaches the value $`V_{}(2\pi \sqrt{\alpha ^{}})^4`$, our $`U(1)`$ is enhanced to $`SU(2)`$. This is a stringy phenomenon which has no description in supergravity, since (for example) the W–bosons for this $`SU(2)`$ are made of wrapped D4–branes.
This is an interesting system to use to study the large $`N`$ limit of the $`𝒩=2`$ $`SU(N)`$ gauge theory along the lines of recent ideas such as those in ref.. A large number of these D6–D2\* objects give a supergravity solution, which might be expected to encode (at least) some of the large $`N`$ physics. Interestingly, the supergravity solution which one naively writes down suffers from a naked singularity known as a “repulson” which is unphysical, and incompatible with the physics of the gauge theory. One expects that there should be a sensible supergravity solution, valid for $`g_sN`$ large, where $`g_s`$ is the string coupling.
In fact the repulson is not present, since it represents supergravity’s best attempt to construct a solution with the correct asymptotic charges. In the solution (not displayed here since we will not need it; see ref.), the volume of the $`K3`$, set to $`V`$ asymptotically, actually decreases as one approaches the core of the configuration. At the centre, the $`K3`$ radius is zero, and this is the singularity.
This ignores rather interesting physics, however. At a finite distance from the putative singularity, the volume of the $`K3`$ gets to $`V=V_{}`$, so the stringy phenomena —including new massless fields— giving the enhanced $`SU(2)`$ should have played a role. So the aspects of the supergravity solution near and inside the special radius, called the “enhançon radius”, should not be taken seriously at all, since it ignored this stringy physics.
To a first approximation, the supergravity solution should only be taken as physical down to the enhançon radius $`r_\mathrm{e}`$. That locus of points, a two–sphere $`S^2`$, is itself called an “enhançon”. It deserves a name, and to be considered as an object in its own right, since D6–D2\* objects probing this geometry seem to spread out or smear onto it as they approach it, losing their identity (see ref. and later in this paper). In this way, we see that the enhançon is a hypersurface apparently made of branes which have puffed up into a sphere. There is a natural generalization of this all to situations involving branes of different dimensions, (with the enhançon a sphere of different dimensionality), and including orientifolds. This pertains to $`SU(N)`$, $`SO(2N)`$, $`SO(2N+1)`$ and $`USp(2N)`$ gauge theories with eight supercharges in various dimensions.
### 1.1 Overview of This Paper
In the rest of this paper we will uncover many new properties of the enhançon pertaining to the 2+1 dimensional $`SU(N)`$ gauge theory. Many of the generic features will have meaning in other dimensions and for other gauge groups. We will obtain detailed information because we can exploit the connection to the classical physics of monopoles. The relevant properties of the enhançon already alluded to so far are reviewed in the next section, and the details that we will need are emphasized, including the perturbative expression for the metric on the spacetime geometry as seen by a probe brane. Section 3 shows that a number of (metric) geometrical details of the enhançon can be learned from the observation that the full non–perturbative spacetime geometry (as seen by the probe in the decoupling limit $`\alpha ^{}0`$) can be deduced from the Atiyah–Hitchin manifold, and appropriate generalisations thereof, which we conjecture to exist as an ADE family. Section 4 focuses on the description of the system of $`N`$ monopoles via Nahm data. The point is simply that the since the description of the $`N`$ coincident D6–D2\* branes carrying the $`SU(N)`$ is as classical monopoles, we ought to learn more about them by studying the well–established technology for describing multi–monopoles. In this way, we see that there is some essential non–commutativity in the description, and we exploit this in section 5 to show that the enhançon is actually a “fuzzy” or non–commutative sphere. This makes contact with the “dielectric brane” construction of Myers, and we discuss the similarities and differences between the two cases.
Figure 1 is a summary of some of the properties of the geometry.
Finally, in section 6 we make a tentative but contentful conjecture about a possible stringy dual description of the 2+1 dimensional $`SU(N)`$ gauge theory at large $`N`$. It is motivated by the fact that the part of the supergravity geometry which can be written, in the decoupling limit, entirely in terms of gauge theory quantities (after referral to the “frame” of a monopole probe) is a part of monopole moduli space. It is suggested that the Nahm data (a family of $`N\times N`$ matrices depending upon a single coordinate) might be used at large $`N`$ to describe a matrix string theory with many of the properties needed for a stringy dual. The construction of the string is likely quite analogous to the more familiar matrix strings, but this new string theory inherits rich properties of the monopole physics, since it is built out of Nahm data. Further work is needed on this proposal. We close with an added note on other work.
## 2 $`SU(N)`$ Gauge theory and BPS Monopoles
In fact, the phenomenology of the enhançon in supergravity is consistent with the monopole physics and with the physics of the 2+1 dimensional $`SU(N)`$ gauge theory at large $`N`$. The moduli space of supersymmetric vacua of the theory is parameterized by the vevs of the three adjoint scalars $`\mathrm{\Phi }_i`$ ($`i=3,4,5`$) in the vector–multiplet. This is $`3(N1)`$ dimensional, since they generically live in the Cartan subalgebra of the gauge group when satisfying this condition. At a generic point on this space, the gauge symmetry is therefore $`U(1)^{N1}`$, (hence the name “Coulomb branch”) and these Abelian gauge fields may be dualized to give $`N1`$ more scalars. The moduli space is therefore $`4(N1)`$ dimensional. The complete, quantum corrected moduli space is a smooth hyperkähler manifold, given that there are eight supercharges. In fact, the moduli space is that of $`N`$ BPS monopoles.
In the probe computation of ref., a single D6–D2\* object was used to probe all of the others, in order to investigate how the geometry looks from its point of view. As moving the probe slowly in the background of its siblings is a BPS process, there should be no potential in the effective Lagrangian for this procedure, and only kinetic terms. From these terms may be read the metric of spacetime as seen by the probe. The result of the computation is:
$$ds^2=F(r)\left(dr^2+r^2d\mathrm{\Omega }^2\right)+F(r)^1\left(ds/2\mu _2C_\varphi d\varphi /2\right)^2,$$
(2.6)
where
$$F(r)=\frac{Z_6}{2g_s}\left(\mu _6V(r)\mu _2\right),$$
(2.7)
and $`d\mathrm{\Omega }^2=d\theta ^2+\mathrm{sin}^2\theta d\varphi ^2`$ and $`C_\varphi =(r_6/g_s)\mathrm{cos}\theta `$. The volume of $`K3`$ is $`V(r)=VZ_2(r)/Z_6(r)`$, with
$`Z_2`$ $`=`$ $`1+{\displaystyle \frac{r_2}{r}},r_2={\displaystyle \frac{(2\pi )^4g_sN\alpha ^{5/2}}{2V}},`$
$`Z_6`$ $`=`$ $`1+{\displaystyle \frac{r_6}{r}},r_6={\displaystyle \frac{g_sN\alpha ^{1/2}}{2}},`$ (2.8)
the harmonic functions appearing in the supergravity solution, which we do not display here. The basic D6– and D2–brane charges are $`\mu _6=(2\pi )^6\alpha ^{7/2}`$ and $`\mu _2=(2\pi )^2\alpha ^{3/2}`$, respectively.
Notice that the metric (2.6) is singular where the monopole’s mass per unit volume, $`\tau =(\mu _6V(r)\mu _2)/g_s`$ vanishes, which is at
$$V(r)=\mu _2/\mu _6=(2\pi \sqrt{\alpha ^{}})^4V_{}.$$
(2.9)
This happens at the “enhançon” radius
$$r_\mathrm{e}=\frac{2V}{VV_{}}|r_2|.$$
(2.10)
This is consistent with the fact that a monopole’s mass is set by the value of the Higgs, while its size is inversely proportional to it. So as $`\mu `$ approaches zero at $`r_\mathrm{e}`$, a monopole probe becomes smeared out as it merges into all the other monopoles at the core. The departure from a sharp description as a heavy point–like object —the smearing— is signaled in the kinetic energy’s divergence.
Since the monopoles cannot go to $`r<r_\mathrm{e}`$ in the supergravity geometry in a supersymmetric way consistent with the gauge dynamics and common sense, it is sensible to conclude that there is simply new geometry and physics in that region, as anticipated in the supergravity discussion of the previous section. Perhaps we can learn more about the enhançon by a closer study of the gauge theory, and hence the monopole physics.
The coupling of the $`SU(N)`$ gauge theory is given by
$$g_{\mathrm{YM}}^2=(2\pi )^4g_s\alpha ^{3/2}V^1$$
(2.11)
To isolate the gauge theory, it is prudent to focus on the limit where we attempt a decoupling limit by sending $`\alpha ^{}0`$, holding the coupling and $`U=r/\alpha ^{}`$ finite. In this case, the metric becomes
$`ds^2=f(U)\left(dU^2+U^2d\mathrm{\Omega }^2\right)+f(U)^1\left(d\sigma {\displaystyle \frac{N}{8\pi ^2}}A_\varphi d\varphi \right)^2,`$
$`\mathrm{where}`$ (2.12)
$`f(U)={\displaystyle \frac{1}{8\pi ^2g_{\mathrm{YM}}^2}}\left(1{\displaystyle \frac{\lambda }{U}}\right),`$
the $`U(1)`$ monopole potential is $`A_\varphi =\pm 1\mathrm{cos}\theta ,`$ and $`\sigma =s\alpha ^{}/2`$. This metric is meaningful only for $`U>\lambda `$. It is the Euclidean Taub–NUT metric, with a negative mass. It is a hyperKähler manifold, because $`f=\times A`$, where $`A=(N/8\pi ^2)A_\varphi d\varphi `$.
It is intriguing to note (and we shall try to exploit this more fully later) that only the gauge theory quantities $`U`$ (a characteristic energy scale) and $`\lambda g_{\mathrm{YM}}^2N`$ (the ’t Hooft coupling) survive the limit, while all other details of the supergravity have disappeared. The enhançon is at $`U=\lambda `$. A crucial point which can be read off from this geometry is that the enhançon appears as the one–loop correction to the gauge coupling, representing the Landau pole. There are instanton corrections to this (and hence to the manifold), smoothing out the singular nature at the enhançon. The expectation was expressed in ref. that this manifold is would be thereby corrected to an Atiyah–Hitchin–like manifold, and we shall see that this is true presently.
From the point of view of the monopole description, this manifold should be related to the metric on the moduli space of monopoles. It is clearly a submanifold of the full $`4N4`$ dimensional metric on what is known as the “strongly centered” moduli space of $`N`$ BPS monopoles<sup>1</sup><sup>1</sup>1“Strongly centred” means that we have the relative moduli space, where the overall center of mass and overall phase of the monopoles are not included..
## 3 The Role of the Atiyah–Hitchin Manifold
Precisely which submanifold we have here should be of interest to us. First observe that we can change variables in our probe metric (2.12) by absorbing a factor of $`\lambda /2=g_{\mathrm{YM}}^2N/2`$ into the radial variable $`U`$, defining $`\rho =2U/\lambda `$. Further absorb $`\psi =\sigma 8\pi ^2/N`$ and gauge transform to $`A_\varphi =\mathrm{cos}\theta `$. Then we get:
$`ds^2={\displaystyle \frac{g_{\mathrm{YM}}^2N^2}{32\pi ^2}}ds_{\mathrm{TN}}^2,\mathrm{with}`$ (3.13)
$`ds_{\mathrm{TN}}^2=\left(1{\displaystyle \frac{2}{\rho }}\right)\left(d\rho ^2+\rho ^2d\mathrm{\Omega }^2\right)+4\left(1{\displaystyle \frac{2}{\rho }}\right)^1\left(d\psi +\mathrm{cos}\theta d\varphi \right)^2.`$
The latter is precisely the form of the Taub–NUT metric that one gets by expanding the Atiyah–Hitchin metric in large $`\rho `$ and neglecting the exponential corrections.
For the case of $`N=2`$, the Atiyah–Hitchin manifold is the full non–perturbative result for the moduli space of the $`SU(2)`$ gauge theory. In fact for this manifold, written in these coordinates, the periodicity of $`\psi `$ is $`2\pi `$ (see later) and so the $`SU(2)`$ isometry of the Taub–NUT manifold is broken to $`SO(3)SU(2)/𝐙_2`$ by the exponential corrections. The quotientied sphere $`S^3/𝐙_2`$ at infinity is an orbit under this.
In unscled coordinate of the $`U(1)`$ probe gauge theory, this means that the quantity $`4\pi ^2\sigma =\stackrel{~}{\sigma }`$ is the $`2\pi `$ periodic dual scalar to the photon. This periodicity is independent of $`N`$. Therefore, for arbitrary $`N`$, the periodicity of $`\psi `$ is $`4\pi /N`$. This allows us to characterise the manifolds we need for all $`N`$:
They have only local $`SO(3)`$ action, and globally the isometry is broken and we have only the action of $`SU(2)/𝐙_N`$. The manifolds are therefore asymptotically the negative mass Taub–NUT, and the space at infinity is an $`S^3/𝐙_N`$. We will characterise the form of the exponential corrections for arbitrary $`N`$ in the next section.
### 3.1 The Non–Perturbative Corrections
So this is our first set of information that we learn by studying the monopole physics: The exponential corrections to our manifold (as seen by the probe) —and hence information about the neighbourhood and interior of the enhançon geometry— are of the same form as those for the Atiyah–Hitchin manifold (with a generalisation we shall characterise shortly). This is remarkably fortuitous, and will teach us more presently. For definiteness, let us display the full Atiyah–Hitchin manifold:
$`ds_{\mathrm{AH}}^2=f^2d\rho ^2+a^2\sigma _1^2+b^2\sigma _2^2+c^2\sigma _3^2,\mathrm{where}`$
$`\sigma _1=\mathrm{sin}\psi d\theta +\mathrm{cos}\psi \mathrm{sin}\theta d\varphi ;`$
$`\sigma _2=\mathrm{cos}\psi d\theta +\mathrm{sin}\psi \mathrm{sin}\theta d\varphi ;`$
$`\sigma _3=d\psi +\mathrm{cos}\theta d\varphi ;`$
$`{\displaystyle \frac{2bc}{f}}{\displaystyle \frac{da}{d\rho }}`$ $`=`$ $`(bc)^2a^2,\text{ and cyclic perms.;}\rho =2K\left(\mathrm{sin}{\displaystyle \frac{\beta }{2}}\right),`$ (3.14)
and $`K(k)`$ is the elliptic integral of the first kind:
$$K(k)=_0^{\frac{\pi }{2}}(1k^2\mathrm{sin}^2\tau )^{\frac{1}{2}}𝑑\tau .$$
(3.15)
Also, $`k=\mathrm{sin}(\beta /2)`$, the “modulus”, runs from $`0`$ to $`1`$, so $`\pi \rho \mathrm{}`$.
The difference between this and negative mass Taub–NUT (3.13) at large $`\rho `$ is exponential, i.e., of the form $`e^\rho `$. In the case of $`SU(2)`$ gauge theory ($`N=2`$), this translates (using the formulae above (3.13) into precisely the right form to be instanton corrections $`e^{U/g_{\mathrm{YM}}^2}`$, and this has been proven to be the correct interpretation from a number of points of view.
For $`SU(N)`$, we expect instantons in the field theory to have essentially the same action, and so this translates into a set of exponential corrections of the form $`e^{N\rho /2}`$. So for large $`N`$ therefore, the corrections to the Taub–NUT manifold are quite small, but the instantons smooth it out on a small enough scale nonetheless. (This smallness of the intanton corrections to $`𝒩=2`$ $`SU(N)`$ gauge theory moduli space at large $`N`$ has been noticed in other contexts, e.g. in ref..)
So in short, our fully corrected moduli space, which also contains information about the spacetime geometry, is given by a family of manifolds naturally generalising the Atiyah–Hitchin manifold, after rescaling $`\rho `$, and $`\psi `$. It would be interesting to characterise these manifolds further. One expects them to be smooth, or at least to contain $`4(N2)`$ parameters which allow them to be deformed to a neighbouring smooth manifold for the simple reason that the gauge theory moduli space, not having a Higgs branch to connect to, is expected to be smooth. There is a concern that this is only true for the full $`4(N1)`$ dimensional manifold representing the moduli space, and that a restriction to a 4 dimensional submanifold can introduce singularities. This is where the $`4(N2)`$ deformation parameters come in, as they represent the frozen moduli of the other monopoles/vacua, now entering as parameters in the reduced theory; intuitively, one expects the process of a single monopole merging with $`N`$ others to be smooth, or at least smoothable by moving the others around.<sup>2</sup><sup>2</sup>2I am grateful to Gary Gibbons, Juan Maldacena, Robert Myers, and Edward Witten for suggestions and comments on this issue.
### 3.2 The Case of Two Monopoles
It is now worth reminding ourselves about the physics of the Atiyah–Hitchin manifold, using it as a prototype for our case involving general $`N`$. The Atiyah–Hitchin manifold is the metric on the strongly centred moduli space of two BPS monopoles. In fact, the two monopole solution itself (i.e., the gauge and Higgs fields) is not spherically symmetric<sup>3</sup><sup>3</sup>3This is generally true for the $`N`$ monopole solution, as we shall discuss.. At best, it is axisymmetric, and this is when the two monopoles are coincident. The coordinate $`\rho `$ represents the asymptotic separation of the monopoles. It really only has this meaning when the monopoles are separated quite far apart, and then the metric reduces to $`ds_{\mathrm{TN}}^2`$. Closer than this, the monopoles cease to be distinct. Actually, the singularity at $`\rho =2`$ in the Taub–NUT metric is completely meaningless, as it is well outside the range of validity of the large $`\rho `$ expansion used to get that metric. Further proof of this comes from the fact that the monopoles are coincident at $`\rho =\pi `$. This special (axisymmetric) solution has monopole charge 2, and really has no sensible description in terms of individual monopoles at all. Generically, all we can say (this can be confirmed by a study of the location of the zeros of the Higgs field part of the monopole solution) is that for any $`\rho `$ the monopoles are spaced symmetrically along an axis, which we can choose to be the $`x_3`$ axis. The Higgs field has zeros even when the monopoles cease to have any sensible meaning (since they grow large and diffuse), and are often used as a guide to the “location” of the monopoles, despite their finite coresize. When $`\rho =\pi `$, the two Higgs zeros are both at the origin, and this is the coincident case.
Note that despite the fact that the solution is axisymmetric, far away from it, in spacetime ($`r\mathrm{}`$), the Higgs field is
$$\frac{1}{2}\mathrm{Tr}[\mathrm{H}(r)^{}\mathrm{H}(r)]=H\frac{Ne_\mathrm{m}}{r}+\mathrm{}$$
(3.16)
for (here) $`N=2`$, and this form of the Higgs is generally true for all $`N`$. Here $`e_\mathrm{m}=2\pi /e`$ is the basic unit of magnetic charge, Dirac fixed in terms of the electric charge $`e`$.
Actually, the metric components in the neighbourhood of $`\rho =\pi `$ are:
$`a=2(\rho \pi )+O\left((\rho \pi )^2\right)+\mathrm{},`$
$`b=\pi +O\left((\rho \pi )^2\right)+\mathrm{},c=\pi +O\left((\rho \pi )^2\right)+\mathrm{}`$ (3.17)
and so the metric appears to be singular there, given that $`a0`$. In fact, the $`S^3`$ of $`(\psi ,\theta ,\varphi )`$ collapses to a two–sphere, $`S^2`$, there, but this point is actually a coordinate “bolt” singularity. It is the removal of this bolt which requires $`\psi `$ to be $`2\pi `$ periodic, a fact which featured in the previous two subsections.
It should be noted that we have only described a simple cover of the moduli space of two monopoles. There is an addition $`𝐙_2`$ which identifies configurations which correspond to each other after an exchange of the (identical) monopoles. In this way the bolt becomes an $`\mathrm{RP}_2`$ instead of an $`S^2`$. We will not have such a symmetry here, so our bolt (or generalisation thereof), for $`N>2`$ will always be an $`S^2`$.
### 3.3 Large $`N`$ and Spacetime Physics
How are we to make sense of the appearance of the Atiyah–Hitchin–like manifolds in our case, and what can we learn about spacetime physics? Well, we have a four dimensional submanifold of the full metric on moduli space, and so most of the parameters ($`4N8`$ of them) have been fixed. In ref. all of the branes were placed at the origin $`r=0`$ in the supergravity discussion. So all of the parameters were frozen except the single probe brane’s position and phase. Although in supergravity they are naively at $`r=0`$, this is not the case, and they are smeared out into a sphere of radius $`r_\mathrm{e}N`$.
Now, the Atiyah–Hitchin coordinate $`\rho `$ should have an interpretation as a separation from the center of mass. For the 2–monopole case that would tell us very little about the spacetime geometry, but since we have $`N`$ monopoles, and $`N`$ is large, $`U=g_{\mathrm{YM}}^2\rho N/2`$ is a good radial coordinate for spacetime, as the center of mass is still close to $`U=0`$ when only one probe is separated off. This is why at large $`N`$ our scaled Atiyah–Hitchin–like manifold has a dual meaning as a relative moduli space for monopoles as well as the spacetime geometry seen by the probe. For small $`N`$, the coordinate $`U`$ is not as good a guide to the spacetime geometry.
Note that for any $`N`$, the spacetime Higgs field will asymptotically behave as in equation (3.16). This is why the supergravity solution can be spherically symmetric, as its asymptotically spherically symmetric geometry matches on to this behaviour, while the deviation from spherical symmetry is a detail only visible in terms subleading in large $`N`$. Taking the expression for the volume $`V(r)`$ and expanding gives:
$$\frac{V(r)}{V_{}}1=\left(\frac{V}{V_{}}1\right)\left(\frac{V}{V_{}}+1\right)\frac{g_s\alpha ^{1/2}N}{r}+\mathrm{},$$
(3.18)
confirming the earlier statement about the relation between the volume and the Higgs field, and fixing $`H=(V/V_{}1)`$ and $`e_\mathrm{m}=(1+V/V_{})g_s\alpha ^{1/2}/2`$. (Later, we will set $`H=1`$ and hence $`V=2V_{}`$, since we are free to choose these parameters at our convenience. Note that the explicit one–monopole solution displayed in eqn.(1.4) is so normalized, and has $`e_\mathrm{m}`$ set to 1.)
Taking seriously the other lessons learned from the two monopole case, the analogue of the bolt sphere at $`\rho =\pi `$ is where the probe merges into all the other branes. For small $`N`$, the correction from $`\rho =2`$ to the bolt radius is significant, while for large $`N`$, as we have seen, the instanton corrections are small. Inside the bolt radius there is really no meaning to the coordinate $`\rho `$ as having anything to do with distinct monopoles. In fact, this is very robust: Scattering two identical monopoles using the Atiyah–Hitchin manifold shows that $`\rho =\pi `$ is truly the distance of closest approach: a head–on collision results in a $`90^\mathrm{o}`$ scattering angle at $`\rho =\pi `$. The finite core size of the monopoles takes over. We inherit this qualitative behaviour here for arbitrary $`N`$.
So in fact we learn that there is indeed a sharp meaning to the sphere of closest approach for the monopoles. It is also where they become massless, and also become indistinct. It is precisely where there occurs a bolt coordinate singularity in the smooth Atiyah–Hitchin manifold. The radii $`\rho <2`$ (or $`U<g_{\mathrm{YM}}^2N`$) do not have any meaning for the individual monopole probes.
## 4 The Multi–Monopole from Nahm Data
In brane/supergravity language, we naively placed the branes all at the same place ($`r=0`$), at the point of $`SU(N)`$ symmetry, the origin of the Coulomb branch. As we know from other examples, quantum corrections in the gauge theory alter the structure of that point. In fact, the enhançon is precisely a manifestation of this, since the monopoles are really not at the origin, but smeared into a sphere.
One of the crucial points of the present investigation is that the entire physics of the moduli space of the gauge theory is given in terms of the classical BPS monopoles, and so we should look no further than that system in order to learn more about the enhançon, and what it means. So how does one describe a clump of $`N`$ monopoles?
Charge $`N`$ multi–monopoles generalising the charge 1 BPS solution were constructed by a number of elegant techniques. On the one hand, algebraic techniques were employed by Ward, with generalisations; on the other hand, a connection to Bäcklund transformations and inverse scattering techniques was employed in refs.. Since the Bogomolnyi equations (1.1) are related by dimensional reduction to the self–dual equations in four Euclidean dimensions, there is another elegant description via an extension of the ADHM construction. The basic (“covariant” Nahm) equations are:
$$\frac{d\mathrm{\Phi }^i}{d\sigma }+[\mathrm{\Phi }_0,\mathrm{\Phi }_i]=\frac{1}{2}ϵ_{ijk}[\mathrm{\Phi }^j,\mathrm{\Phi }^k],$$
(4.19)
where $`i,j,k`$ run over $`1,2,3`$. Here, $`\mathrm{\Phi }_0(\sigma )`$ and $`\mathrm{\Phi }_i(\sigma )`$ are $`N\times N`$ anti–Hermitian $`SU(N)`$ matrices, with $`\mathrm{\Phi }_i^{}(\sigma )=\mathrm{\Phi }_i(\sigma )`$ and $`\mathrm{\Phi }_i(\sigma )=\overline{\mathrm{\Phi }}_i(\sigma )`$. The coordinate $`\sigma `$ has range $`H\sigma H`$, where $`H`$ is the asymptotic value of the Higgs field, which we shall presently set to 1 in much of the rest of this paper. The data appropriate to monopoles arise as solutions to this equation which are regular in the interior of $`[H,H]`$, with appropriate boundary conditions at $`\sigma =\pm H`$. Those boundary conditions require that $`\mathrm{\Phi }_i`$ have simple poles there, and that the residues of those poles (which are of course $`N\times N`$ matrices) are irreducible representations of $`SU(2)`$. There is an $`SU(N)`$ gauge invariance,
$$\mathrm{\Phi }_0G\mathrm{\Phi }_0G^1\frac{dG}{d\sigma }G^1,\mathrm{\Phi }_iG\mathrm{\Phi }_iG^1,$$
(4.20)
where the $`G(\sigma )SU(N)`$, and are the identity at the ends of the interval. There is a specific construction (also following the ADHM techniques) for converting the solutions of these equations —the “Nahm data”— into expressions for the spacetime fields $`A_i(𝐱)`$, $`\mathrm{H}(𝐱)`$, which we shall not reproduce here since for $`N>1`$, closed forms are not known. Hitchin has shown using algebraic methods that this method constructs all of the monopole solutions and indeed that it is equivalent to the aforementioned monopole constructions based on those of Ward.
The gauge transformations (4.20) can be used to set $`\mathrm{\Phi }_0`$ to zero, giving the standard Nahm equations, but should be left unfixed in order to perform the full hyperKähler quotient which constructs the metric on the moduli space of Nahm data. Nakajima has shown that the metric thus computed is indeed the monopole moduli space, and it is smooth.
This Nahm system arises naturally in the brane description as the condition on the brane fields for supersymmetric vacua, resulting in a hyperkähler quotient. The $`\mathrm{\Phi }`$ are adjoint scalars in a gauge theory on the brane. The most natural brane system where this arises is probably that of $`N`$ D1–branes stretched perpendicularly between two D3–branes separated by a distance $`2H`$. The coordinate $`\sigma `$ is that along the D1–branes, and the $`\mathrm{\Phi }`$ are the positions of the D1–branes inside the D3–branes. The boundary conditions arise by considering the 1+1 dimensional theory on the world–volume as a theory with “impurities” located at $`\sigma =\pm H`$, which is natural, since the massless 1–3 strings are localized there. These Nahm equations can also be derived in the brane wrapped on $`K3`$ system we started with here<sup>4</sup><sup>4</sup>4This follows since these systems are dual to one another. The details will appear in ref.. The asymptotic value of $`K3`$, $`V`$, sets the parameter $`H`$ via $`H=(V/V_{}1)`$. As the supergravity parameter $`r`$ runs from $`\mathrm{}`$ to $`r_\mathrm{e}`$ (or more properly as $`U`$ runs from $`\mathrm{}`$ to $`\lambda `$) the coordinate $`\sigma `$ runs from $`H`$ to zero. Let us set $`H=1`$ henceforth.
## 5 The Enhançon as a Fuzzy Sphere
It is easy to see that the enhançon is itself a fuzzy sphere in spacetime as follows. The D6–D2\* system is dual to a system of $`N`$ D3–branes stretched between a pair of NS5–branes in type IIB string theory. The $`SU(N)`$ gauge theory is on the flat part of the D3–branes. The Nahm equations above (4.19) have the following meaning: The $`\mathrm{\Phi }_i(\sigma )`$, multiplied by $`2\pi \sqrt{\alpha ^{}}`$, are coordinates in the $`\mathrm{I}\mathrm{R}^3`$ part of the NS5–branes where the D3–branes end. The coordinate $`\sigma `$ is the coordinate between the NS5–brane. The 5+1 dimensional theory on the NS5–branes is the spontaneously broken 3+1 dimensional $`SU(2)`$ theory if we ignore the two spatial directions common to both the branes. Translating further, there is a factor of $`1/g_s`$ in front of the commutator in the Nahm equation. (This is instead of $`g_s`$, appropriate to the case of D1–branes ending on D3–branes.)
There is a “double trumpet” shape describing the $`N`$ stretched D3–branes pulling on the NS5–branes, as depicted in ref. and reproduced in figure 2.
At the centre of the shape, there is a two–sphere where the fivebranes touch, restoring the $`SU(2)`$. Crucially, this is only a two–sphere for $`N`$ large enough, since the radius of the sphere is proportional to $`N`$, and only for spheres large enough are we far enough away from the details of the interior of the multi–monopole configuration to see an approximately spherically symmetric situation. (Recall that the multi–monopole is not spherically symmetric.)
We can make this a bit more precise as follows: First note that in the one–monopole case, the solution is spherically symmetric, and the Nahm data is simply the one dimensional representation of $`SU(2)`$, i.e., all the $`f_i(\sigma )`$ are equal. Assume that in our case, we are far away enough from the core that we can borrow this behaviour, making the symmetric choice $`\mathrm{\Phi }_i(\sigma )=\mathrm{i}f(\sigma )\mathrm{\Sigma }_i`$. In doing this, we connect to the discussion of ref.. There, this was shown to correspond to an infinite trumpet shape representing $`N`$ D1–branes merging into an orthogonal D3–brane. The required poles at the ends of the Nahm interval correspond to the flaring of the trumpet as it expands into the perpendicular shape. The Nahm equations become:
$$\frac{df}{d\sigma }=\frac{f^2}{g_s},$$
(5.21)
and the solution we seek here is made by gluing together two copies of the trumpet end to end at the centre of the interval:
$$f(\sigma )=\frac{g_s}{\sigma 1},\text{ for }\sigma [0,\pm 1].$$
(5.22)
This gives a shape like that in figure 2, but is only an approximation. The full solution should connect smoothly through the interior of the interval. A cross section at some value of $`\sigma `$ is a non–commutative, or “fuzzy” sphere of radius given by (remembering to put in the factor of $`2\pi \sqrt{\alpha ^{}}`$ for dimensions)
$$R^2=4\pi ^2\alpha ^{}\underset{i}{}\mathrm{Tr}(\mathrm{\Phi }_i^2)=4\pi ^2\alpha ^{}(N^21)f^2(\sigma ),$$
(5.23)
There is a minimum value, $`f_\mathrm{e}g_s`$ where the NS5–branes touch at $`\sigma =0`$. There, the radius is
$$R_N=2\pi \sqrt{\alpha ^{}}g_s\sqrt{N^21}2\pi \sqrt{\alpha ^{}}g_sN,$$
(5.24)
which compares well with the supergravity expression (2.10) for the enhançon radius, which is
$$r_\mathrm{e}=g_sN\sqrt{\alpha ^{}}\left(\frac{V}{V_{}}1\right)^1\frac{N}{He}.$$
(5.25)
This particular fuzzy sphere is the enhançon. It is a sensible smooth sphere of non–zero radius at large $`N`$. Notice also that it has roughly the correct behaviour for the $`N`$ monopole size in terms of the Higgs vev $`H`$ and the electric charge $`e`$.
For small $`N`$ ($`1`$) it is very non–spherical, while it collapses to zero size in the case $`N=1`$: The minimum value $`f_\mathrm{e}`$ is zero for a single monopole, and the double trumpet profile pinches off, as can be deduced (see ref.) from a study of the Higgs field (1.4) for the explicitly known one–monopole solution. We plot this in figure 3.
Returning to large $`N`$, it is in this sense that we see the connection between the dielectric brane construction of ref. (see also ref.) (where branes puff up into a sphere in the presence of a background R–R field) and the enhançon<sup>5</sup><sup>5</sup>5Non–commutativity in the enhançon geometry was suspected in ref., and a relation to the dielectric branes was suspected by many.. Both phenomena can be used as examples of a new mechanism for excising undesirable spacetime singularities, but the dielectric mechanism is adapted to $`𝒩=1`$ supersymmetry preserving vacua, while here we have $`𝒩=2`$ (counting in four dimensional units). The connection between the two is simply that there are non–zero commutators for the adjoint scalars (forming $`N`$–dimensional irreducible representations of $`SU(2)`$) which have been shown in one case to induce multipole couplings to higher rank R–R fields. This is equivalent to the growth of extra dimensions on the brane.
In the present case, the Nahm equations are the means by which non–zero commutators arise in an $`𝒩=2`$ supersymmetry preserving way. The higher dimensional aspect of the branes is realized in terms of their description as a finite sized multi–monopole configuration. The non–commutative sphere which is the enhançon is a preferred slice through this geometry, forming the effective shell around the $`N`$–monopole core.
## 6 A Candidate Dual?
One of the goals of a study of the supergravity solution mentioned in the introduction was to see if there is a large $`N`$ dual supergravity solution. As pointed out in ref., even with the improved understanding of the geometry by recognizing the role of the enhançon, the decoupling/scaling limit $`\alpha ^{}0`$ (with $`U=r/\alpha ^{}`$ finite) gives a ten dimensional supergravity solution which was not appropriate as a truly decoupled dual theory. One sign of this (among others) is the simple fact that the resulting geometry contained parameters which recalled the original data of the type IIA compactification. In other words, it did not assemble into an expression referring purely to gauge theory quantities, as happens in simpler cases where there is a genuine supergravity dual.
It would certainly be an excellent situation if there was a large $`N`$ dual theory all the same, and a persistent open question is whether there exists such a theory, and whether it is a useful dual, in the sense of being weakly coupled (or at least tractable) when the gauge theory is strongly coupled.
To this end, let us note again that in the decoupling limit, the part of the supergravity geometry transverse to the branes assembles into purely gauge theory quantities, when it is referred to from the “frame” of a brane probe. We ought to regard this as a clue. As the geometry is (a subspace of) the moduli space of multi–monopoles, this leads one to speculate that there may be something to be gained by focusing one’s attention there.
On general grounds, one might expect that this large $`N`$ dual theory might be a string theory whose world sheet genus expansion is isomorphic to the $`1/N`$ expansion in the planar diagrams of the field theory, in the usual manner. Besides this, the putative string theory should encode the structure of (and allow access to) the $`4N4`$ dimensional moduli space of the Coulomb branch of vacua. The $`SU(2)_R`$ symmetry of the gauge theory should be present, as should the phenomenon of the enhançon, etc.
Here is a proposal for such a string theory. Simply take the large $`N`$ limit of the Nahm data, but looked at in a different way! The $`N\times N`$ matrices $`\mathrm{\Phi }_i(\sigma )`$, for $`(i=0,1,2,3)`$, which satisfy the Nahm equations, may be thought of at large $`N`$ as giving the coordinates of a string in a four dimensional transverse space. It is a matrix string constructed by letting the matrices explore the full $`4N4`$ dimensional moduli space (or possibly a cover of it), which is imposed by the Nahm equations (4.19) and the accompanying gauge invariance (4.20).
This proposal is more easily motivated by analogy with a simpler matrix string construction, that of the ten dimensional case of type IIA. There, we have eight $`N\times N`$ matrices $`X^i(\sigma )`$. While they may be thought of the collective coordinates of $`N`$ D1–branes, and hence parameterising $`(\mathrm{I}\mathrm{R}^8)^N/S_N`$, the now–standard route reinterprets them at large $`N`$ as light cone “string fields” giving a description the shape of a single IIA string in eight transverse dimensions $`\mathrm{I}\mathrm{R}^8`$. The “second–quantized” description of the string theory is simply the (1+1)–dimensional $`U(N)`$ gauge theory of the $`X^i`$.
In fact, the very same structures are present here, but yielding a (with respect) potentially much more interesting string, at least for our purposes, since it contains all of the ingredients to make a dual string for the (2+1)–dimensional $`SU(N)`$ gauge theory. The naive interpretation of the $`\mathrm{\Phi }_i(\sigma )`$’s is as the non–commutative collective coordinates of $`N`$ D–strings stretched along a finite interval and acting as BPS monopoles. (We are ignoring the two directions common to both types of brane in the setup of section 5.) The proposal here is that at large $`N`$, the Nahm data $`\mathrm{\Phi }_i(\sigma )`$ can be thought of as string fields for a single string having a 4 dimensional transverse target space. To obtain the “long string” sectors required to enable a stringy limit, two modifications to our setup can be considered: The first is that the whole system of NS5–branes described in section 5 be placed on a circle, by compactifying the direction in which the branes are separated. The second is that the Donaldson–Nahm $`U(N)`$ gauge transformations (4.20), which reduce to the identity at $`\sigma =\pm 1`$ , be allowed to include permutations at the ends of the $`\sigma `$ interval in this periodic system. In this way, we include configurations in the theory corresponding to a D–string winding around a large number of times before terminating on an NS5–brane.
A periodic version of the 1+1 dimensional “impurity” gauges theories of the sort described in refs. should provide the dynamics. To get to strong coupling for the gauge theory, one must further tune the length of the interval (and hence the Higgs vev) and the size of the circle to be small, holding the ratio fixed.
By construction, the string thus defined has many of the properties which we seek for our dual string. It refers correctly to the moduli space of vacua of the gauge theory by using the monopole moduli space in an essential way. Different vacua of the gauge theory correspond to different backgrounds for the string theory.
A tantalizing consequence of this conjecture (which clearly needs more work) for a dual string is that it may provide a dictionary between many of the elegant results about monopole moduli space (such as the scattering of slowly moving monopoles, geodesics representing bound states, etc.), and properties of the dual gauge theory. It will certainly be interesting to pursue this further.
## Added Note on Other Work
There is other recent work on the large $`N`$ limit of monopoles and Nahm data, with connections to brane configurations, and non–commutativity<sup>6</sup><sup>6</sup>6The author is grateful to Micha Berkooz and Kimyeoug Lee for pointing out refs. after this manuscript first appeared. which we ought to mention here. While they appear to be separate avenues of investigation, it should be interesting to learn if there is anything in those approaches which might help with some of the issues discussed here.
Existing proposals by Fairlie and collaborators concerning a matrix/M–theory interpretation of Nahm equations (in diverse dimensions) at large $`N`$ emphasize a connection to the Moyal bracket(see also ref.), to non–commutative geometry, and the physics of membranes. In ref., Lee studies the case of an infinite sheet of BPS monopoles, highlighting its non–commutative description, and a relation to D1–strings stretching between D3–branes.
It is interesting to note that both of these sets of work make connections at large $`N`$ to a natural two dimensional non–commutative surface, while in this paper, we have pointed out that the enhançon itself is a non–commutative sphere. Perhaps a firmer connection might be made between these approaches which might lead to a description of the dynamics of the enhançon as a non–commutative membrane via those techniques.
Also, the work of Berkooz studies a 1+1 dimensional impurity model (related to the one we have in mind in this paper) as a DLCQ realization of the 4+1 dimensional $`SU(N)`$ Yang–Mills theory found on D4–branes. Features such as the crossover between open and closed string effective descriptions of the gauge theory at large $`N`$ are highlighted.
## Acknowledgements
I am grateful to Peter Forgacs for pointing out my unintentional omission (in release 1 of this paper) of the excellent independent early work on multi–monopoles. I would like to thank Alex Buchel, David Fairlie, Gary Gibbons, Juan Maldacena, Rob Myers, Amanda Peet, Joe Polchinski, Eric Weinberg and Edward Witten for useful comments, and Samantha Butler for her patience.
|
warning/0004/gr-qc0004073.html
|
ar5iv
|
text
|
# EQUILIBRIUM CONFIGURATION OF BLACK HOLES AND THE INVERSE SCATTERING METHOD11footnote 1journal reference:Theor. Math. Phys. v.111, p.667(1997)
## 1 Introduction
In the present paper, we study solutions to the Einstein equations in a vacuum that are stationary, axially symmetric, and asymptotically flat. The main purpose is to elaborate a description in which two classes of ideas, black hole theory and the theory of completely integrable equations, are unified.
In black hole physics, the well-known result claims that ”black holes have no hair”. In other words, each black hole is uniquely determined by its mass and angular momentum and is nothing but the Kerr solution. Many authors contributed to this result and we do not have room to give the exhaustive references. For our purpose, only the final stage of the proof is important, namely, Carter’s classification of all axially symmetric solutions having connected horizon . Carter demonstrated that these solutions must solve the boundary problem for a system of elliptic nonlinear equations, the Kerr solution being one of the possible solution to this boundary problem. The uniqueness of this solution was proven in .
Can one do without requiring the connectedness of the horizon? In other words, do solutions exists that correspond to an equilibrium configuration of black holes? To the best of the author knowledge, these questions still have no exhaustive answers (though some results can be found in ). Also, these questions can be formulated in terms of the boundary problem from because, there, the connectedness of the horizon played no role. This boundary problem is formulated in the following section.
It is well known that the Einstein equations with two commuting symmetries belong to a wide class of systems that can be integrated by the methods of the inverse scattering problem. This was shown in by Belinskii and Zakharov, who also investigated the axially symmetric case for which the $`2N`$-soliton solution was constructed . This solution was interpreted as the solution corresponding to the $`N`$ Kerr-NUT black holes. As we show in the present paper, the $`2N`$-soliton solution of Belinskii-Zakharov indeed contains all possible solutions (if any exist) corresponding to an equilibrium configuration of rotating black holes. Therefore, we reduce the question of the existence of solutions with disconnected horizon to the investigation of some subclass of the soliton solutions. This subclass is parameterized by the distances between black holes, the angular momenta, and the masses of the black holes. However, in general case, solutions from this have a conical singularity on the symmetry axis, which hinders the existence of solutions with disconnected horizon. It is still unclear whether it possible to choose the parameters in a way that removes this conical singularity. Most likely, the answer is negative .
The presence of the conical singularity does not make these solutions physically meaningless. On the contrary, the conical singularity itself has the physical sense of the interaction force between the black holes. This interpretation was proposed by Weyl; for details see .
## 2 Boundary conditions
In this section, we present some basic facts about the stationary axially symmetric solutions to the Einstein equations and formulate the boundary problem corresponding to an equilibrium configuration for black holes. The results of are crucial for us and we refer the reader there for details.
Recall that Lorentz manifold is stationary and axially symmetric if it possesses two commuting one-parameter isometry groups that are isomorphic to $`R`$ and $`SO(2)`$ respectively. In other words, there exist two Killing vector fields $`k^a`$ and $`m^a`$, which commute with each other, such that the vector $`k^a`$ is time-like and the vector $`m^a`$ is space-like. For the stationary and axially symmetric vacuum (or electrovacuum) space-time one can always choose the coordinate system in which the metric acquires the following form:
$$ds^2=Vdt^2+2Wdtd\varphi +Xd\varphi ^2+\frac{X}{\rho ^2}e^\beta (d\rho ^2+dz^2).$$
Here the metric coefficients depend only on $`\rho `$ and $`z`$, and $`k^a/x^a=/t`$, $`m^a/x^a=/\varphi `$. Then $`V,W,X,\rho `$ have a geometric meaning,
$$V=k^ak_a,X=m^am_a,W=k^am_a,$$
$`(2.1a)`$
$$VX+W^2=\rho ^2=\rho _{ab}\rho ^{ab}(\rho _{ab}=2k_{[a}m_{b]}).$$
$`(2.1b)`$
The multitudes of the Killing vectors are normalized as follows: $`V1`$ at infinity and $`\frac{X^{,a}X_{,a}}{4X}1`$ on the symmetry axis. The coordinates $`\varphi ,\rho `$ and $`z`$ become cylindrical (Weyl) coordinates.
The set of points with $`\rho =0,X=0`$ is the symmetry axis, while the set of points with $`\rho =0,X>0`$ is the event horizon. Let the event horizon have $`N`$ connected components and $`l^a`$ be an isotropic vector that is orthogonal to the event horizon. Then, in each connected component of the horizon, we can choose such a normalization of $`l^a`$ that
$$l^a=k^a+\mathrm{\Omega }_im^a,$$
where $`\mathrm{\Omega }_i`$ is some constant whose physical meaning is the angular velocity of the black hole.Let $`z_1,z_2,\mathrm{},z_{2N}`$ be the points of intersection of the horizon and the symmetry axis enumerated in increasing order.
We pass to a new coordinate system,
$$\rho ^2=(\lambda ^2m_i^2)(1\mu ^2),m_i=\frac{z_{2i}z_{2i1}}{2},$$
$$z\frac{z_{2i}+z_{2i1}}{2}=\lambda \mu .$$
In this coordinate system, the necessary conditions of regularity of the symmetry axis and the horizon are formulated as follows:
$$X(\lambda ,\mu )=(1\mu ^2)\widehat{X}(\lambda ,\mu )$$
$$W^{}(\lambda ,\mu )=(\lambda ^2m_i^2)(1\mu ^2)\widehat{W}(\lambda ,\mu )$$
$`(2.2)`$
$$V^{}(\lambda ,\mu )=(\lambda ^2m_i^2)\widehat{V}(\lambda ,\mu )$$
Here $`\widehat{X},\widehat{W},\widehat{V}`$ are smooth functions nowhere equal to zero and
$$V^{}=l^al_a,W^{}=l^am_a.$$
From (2.2) one can easily obtain the boundary conditions for the first group of Einstein equations,
$$d\rho dgg^1=0,g=\left(\begin{array}{cc}V& W\\ W& X\end{array}\right),$$
$`(2.3)`$
where $``$ is the Hodge operator $`d\rho =dz,dz=d\rho `$. Indeed using (2.2), one can easily prove that
$$\rho g_{,\rho }g^1=\left(\begin{array}{cc}0& O(1)\\ 0& 2\end{array}\right),\rho 0,z\mathrm{\Gamma },$$
$`(2.4a)`$
$$\widehat{\mathrm{\Omega }}_i\rho g_{,\rho }g^1\widehat{\mathrm{\Omega }}_i^1=\left(\begin{array}{cc}2& 0\\ O(1)& 0\end{array}\right),\rho 0zI_i,\widehat{\mathrm{\Omega }}_i=\left(\begin{array}{cc}1& \mathrm{\Omega }_i\\ 0& 1\end{array}\right).$$
$`(2.4b)`$
$$\rho g_{,z}g^1=O(1),\rho 0,zR.$$
$`(2.4c)`$
Here $`\mathrm{\Gamma }`$ is the symmetry axis consisting of $`N+1`$ connected components,
$$\mathrm{\Gamma }=\mathrm{\Gamma }_j=RI_i,j=1,\mathrm{},N+1,i=1,\mathrm{},N,I_i=(z_{2i1},z_{2i}).$$
The symbol $`O(1)`$ denotes a uniformly bounded function on corresponding interval. It follows from (2.2) that (2.4c) tends to zero almost everywhere except the points $`z_k`$; however for our purposes, a uniform boundness suffices. The function $`g(z,\rho )`$ is taken to be smooth at all points except the points $`(z_k,0)`$.
As we demonstrated below, Eqs. (2.4) completely determine the solution to Eqs. (2.3). At the same time $`\mathrm{\Omega }_i`$ and $`z_i`$ are the independent parameters of the boundary problem; these parameters can be choose arbitrary. The behavior of $`g`$ at infinity is discussed at the end of this section.
An alternative approach, in which the main parameters are angular momenta rather than angular velocities, exists . To show this let us introduce the Ernst potentials
$$\rho dgg^1=\left(\begin{array}{cc}dY^{12}& d\stackrel{~}{Y}\\ dY& dY^{21}\end{array}\right),d(Y^{21}Y^{12})=2dz.$$
$`(2.5)`$
Here $`Y`$ is the Ernst potential that is determined by the space-like Killing vector field, while $`\stackrel{~}{Y}`$ is the Ernst potential determined by the time-like Killing vector field. Then system (2.3) can be rewritten in equivalent form,
$$d(\frac{\rho dX}{X})\frac{\rho }{X^2}dYdY=0,$$
$$d(\frac{\rho dY}{X})+\frac{\rho }{X^2}dXdY=0,$$
$`(2.6a)`$
$$d\mathrm{\Omega }=\frac{\rho dY}{X^2},W=\mathrm{\Omega }X.$$
$`(2.6b)`$
From (2.4a) or (2.2) we can see that
$$\rho _\rho \mathrm{ln}X2,\rho 0,z\mathrm{\Gamma };Y|_{\mathrm{\Gamma }_i}=c_i,$$
$`(2.7)`$
where $`c_i`$ are some constants that are independent parameters. In , it was proved that (2.6) has a unique solution satisfying (2.7) and some condition at infinity for all $`z_i`$ and $`c_i`$. In fact, the latter condition is equivalent to the asymptotic flatness of the metric. Note that (2.4b) follows from (2.6b) provided that $`\mathrm{\Omega }_i`$ is defined as follows:
$$\mathrm{\Omega }|_{I_i}=\mathrm{\Omega }_i,$$
$`(2.8)`$
Thus, $`\mathrm{\Omega }_i`$ is a function of $`z_i`$ and $`c_i`$.
The constants $`c_i`$ unambiguously determine the angular momentum of all black holes. Indeed, let us define the angular momenta of a black hole using the Komar form:
$$L_i=\frac{1}{16\pi }_{S_i}m^{a;b}𝑑S_{ab},$$
where $`S_i`$ is a two-surface surrounding the black hole. Choosing $`S_i`$ to be the surface of revolution of the curve $`C_i`$ connecting the components of the axis $`\mathrm{\Gamma }_{i+1}`$ and $`\mathrm{\Gamma }_i`$, we obtain
$$L_i=\frac{1}{8}_{C_i}\frac{1}{\rho }(XdWWdX)=\frac{1}{8}_{C_i}𝑑Y=\frac{1}{8}(c_{i+1}c_i).$$
$`(2.9)`$
The mass of the black hole can be defined as the following integral :
$$M_i=\frac{1}{8\pi }_{S_i}k^{a;b}𝑑S_{ab}.$$
Proceeding as for finding the angular momentum, we obtain that
$$M_i=\frac{1}{4}_{C_i}\frac{1}{\rho }(XdV+WdW)=\frac{1}{4}_{C_i}𝑑Y^{12}.$$
Contracting the contour $`C_i`$ to the horizon and using (2.4b), we find that
$$M_i=\frac{1}{4}_{I_i}(2+\mathrm{\Omega }_iY_{,z})𝑑z=m_i+2\mathrm{\Omega }_iL_i,$$
$`(2.10)`$
where $`m_i=(z_{2j}z_{2j1})/2`$.
At infinity, we impose the following conditions:
$$W=\rho ^2O(\frac{1}{r^3}),X=\rho ^2(1+O(\frac{1}{r})).$$
$`(2.11)`$
where $`r=\sqrt{\rho ^2+z^2}`$. Asymptotic formulas (2.11) are assumed to be differentiable at least twice. Formulas (2.11) mean that the metric tensor $`g`$ tends to the Minkowski tensor in cylindrical coordinates. From (2.11) we obtain
$$g_{,z}g^1=\left(\begin{array}{cc}O(1/r^2)& O(1/r^4)\\ \rho ^2O(1/r^4)& O(1/r^2)\end{array}\right),V=1+O(1/r),$$
$`(2.12a)`$
$$\rho g_{,\rho }g^1\left(\begin{array}{cc}0& 0\\ 0& 2\end{array}\right)=\left(\begin{array}{cc}O(1/r)& O(1/r^3)\\ \rho ^2O(1/r^3)& O(1/r)\end{array}\right).$$
$`(2.12b)`$
When determining the angular velocities from (2.8), we should normalize $`\mathrm{\Omega }`$ in accordance with (2.11), i.e. $`\mathrm{\Omega }=O(1/r^3)`$.
The second group of Einstein equations allows one to determine the coefficient $`e^\beta `$ from the matrix $`g`$. Using (2.4) or (2.7), one can show that $`_z\beta =0`$ for$`\rho =0`$ and $`z\mathrm{\Gamma }`$, i.e. $`\beta |_{\mathrm{\Gamma }_i}=b_i`$, where $`b_i`$ are some constant. The conical singularity on the symmetry axis is absent iff $`b_i=0`$. However, $`b_i`$ cannot be treated as independent parameters since they are functions of $`z_k`$ and $`c_k`$. In the preset paper, we restrict ourselves to the study of boundary problem (2.4), (2.12) and do not discuss the properties of $`b_i`$.
## 3 Auxiliary linear problem
System of equations (2.3) is the compatibility condition for the following pair of matrix linear differential equations :
$$D_1\psi =\frac{\rho ^2g_{,z}g^1\omega \rho g_{,\rho }g^1}{\omega ^2+\rho ^2}\psi ,D_2\psi =\frac{\rho ^2g_{,\rho }g^1+\omega \rho g_{,z}g^1}{\omega ^2+\rho ^2}\psi .$$
$`(3.1)`$
Here $`D_1`$ and $`D_2`$ are the commuting differential operators:
$$D_1=_z\frac{2\omega ^2}{\omega ^2+\rho ^2}_\omega ,D_2=_\rho +\frac{2\omega \rho }{\omega ^2+\rho ^2}_\omega ,$$
and $`\omega `$ is a complex parameter that does not depend on the coordinates. We also use the $`UV`$ pair representation in which $`\omega `$ is a dependent parameter. To be more precise, let $`\omega `$ be a root of the equation
$$\omega ^22\omega (kz)\rho ^2=0,$$
$`(3.2)`$
where $`k`$, in turn, is independent spectral parameter. Using (3.2), one can easily check that
$$_z\omega =\frac{2\omega ^2}{\omega ^2+\rho ^2},_\rho \omega =\frac{2\omega \rho }{\omega ^2+\rho ^2}.$$
$`(3.3)`$
Passing from $`\psi (\omega )`$ to $`\psi ^{}(k)=\psi (\omega (k))`$, we obtain from (3.1) that
$$_z\psi (k)=A(z,\rho ,k)\psi (k),A=\frac{\rho ^2g_{,z}g^1\omega \rho g_{,\rho }g^1}{\omega ^2+\rho ^2},$$
$`(3.4a)`$
$$_\rho \psi (k)=B(z,\rho ,k)\psi (k),B=\frac{\rho ^2g_{,\rho }g^1+\omega \rho g_{,z}g^1}{\omega ^2+\rho ^2}$$
$`(3.4b)`$
Hereafter, we omit the prime for brevity. It is worth mentioning that Eqs. (3.4) are equivalent to Eqs. (3.1) only if we take $`\omega `$ to be the multivalued function in (3.4). Fixing the branch of the root in (3.4), we find the solution to system (3.1) only in the analyticity domain of $`\omega (k)`$.
In the present paper, we follow the general scheme for investigating integrable equations .
Since Eq.(3.2) is invariant with respect to the transformation $`\omega \rho ^2/\omega `$, we can fix the branch of the multivalued function $`\omega (k)`$, stipulating that the inequality $`|\omega |>\rho `$ holds. Then, from (3.2), we obtain
$$\omega 2(kz),\rho 0;\omega 2(kz),z\mathrm{};\omega 2(kz),k\mathrm{}$$
$`(3.5)`$
After choosing the branch of the root, we can introduce the monodromy matrix $`T(z,y)`$, which, by definition, is a solution to (3.4a) such that $`T(y,y)=I.`$ Note that for $`\rho 0`$,
$$A(z,\rho ,k)\frac{1}{2}\frac{1}{zk}\left(\begin{array}{cc}0& _z\stackrel{~}{Y}\\ 0& 2\end{array}\right),z\mathrm{\Gamma },$$
$$A(z,\rho ,k)\frac{1}{2}\frac{1}{zk}\widehat{\mathrm{\Omega }}_i^1\left(\begin{array}{cc}2& 0\\ _zY& 0\end{array}\right)\widehat{\mathrm{\Omega }}_i,zI_i.$$
Here we took into account (3.5) and boundary condition (2.4). Hence Eq. (3.4a) can be easily integrated at $`\rho =0`$. As a result, the explicit formulas for the monodromy matrix are
$$T(z,y)=\left(\begin{array}{cc}1& \frac{\stackrel{~}{Y}(z)\stackrel{~}{Y}(y)}{2(ky)}\\ 0& \frac{kz}{ky}\end{array}\right),z,y\mathrm{\Gamma }_k,$$
$`(3.7)`$
where $`\mathrm{\Gamma }_k`$ is the connected component of the symmetry axis and
$$T(z,y)=\widehat{\mathrm{\Omega }}_i^1\left(\begin{array}{cc}\frac{kz}{ky}& 0\\ \frac{Y(z)Y(y)}{2(ky)}& 1\end{array}\right)\widehat{\mathrm{\Omega }}_i,z,yI_i.$$
$`(3.8)`$
Let $`e(z,\rho ,k)`$ be the solution to (3.4) with the Minkowski tensor in cylindrical coordinates ($`V=1,W=0,X=\rho ^2`$),
$$e(z,k)=\left(\begin{array}{cc}1& 0\\ 0& \omega (z,k)\end{array}\right),_ze=A_0e,A_0=\left(\begin{array}{cc}0& 0\\ 0& \frac{2\omega }{\omega ^2+\rho ^2}\end{array}\right).$$
$`(3.9)`$
Let us define the Jost functions and reduced monodromy matrix,
$$\mathrm{\Psi }^\pm (z,k)=\underset{y\pm \mathrm{}}{lim}T(z,y)e(y),$$
$`(3,10)`$
$$T(k)=\underset{y\mathrm{},z+\mathrm{}}{lim}e^1(z)T(z,y)e(y).$$
$`(3.11)`$
We do not reproduce the explicit dependence on $`\rho `$. The functions $`\mathrm{\Psi }^\pm `$ satisfy the following integral equations:
$$\mathrm{\Psi }^{}(z)=e(z)+_{\mathrm{}}^ze(z)e^1(x)A^{}(x)\mathrm{\Psi }^{}(x)𝑑x,$$
$`(3.12)`$
$$\mathrm{\Psi }^+(z)=e(z)_z^{\mathrm{}}e(z)e^1(x)A^{}(x)\mathrm{\Psi }^+(x)𝑑x,$$
$`(3.13)`$
where
$$A^{}(z)=A(z)A_0(z)=\frac{\rho ^2g_{,z}g^1}{\omega ^2+\rho ^2}\frac{\omega }{\omega ^2+\rho ^2}\left(\rho g_{,\rho }g^1\left(\begin{array}{cc}0& 0\\ 0& 2\end{array}\right)\right).$$
$`(3.14)`$
Limit (3.11) exists at least for $`|`$Im$`k|>\rho `$. Further, using (3.4b), one can show that $`T(k)`$ does not depend on $`\rho `$. The basic property of the monodromy matrix reads
$$T(z,y)=T(z,z_{2N})T(z_{2N},z_{2N1})\mathrm{}T(z_1,y),z\mathrm{\Gamma }_{N+1},y\mathrm{\Gamma }_1,$$
$`(3.15)`$
where $`\mathrm{\Gamma }_{N+1}`$ and $`\mathrm{\Gamma }_1`$ are the extreme components of the symmetry axis. Then, from (3.15), (3.11), (3.8), (3.7) and (3.5) we get that
$$T(k)=\widehat{D}_{N+1}\underset{j=1,\mathrm{},N}{}T_j\widehat{D}_j.$$
$`(3.16)`$
Here
$$T_j=\left(\begin{array}{cc}1\frac{2M_j}{kz_{2j1}}& 4M_j\mathrm{\Omega }_j\\ \frac{2L_j}{(kz_{2j})(kz_{2j1})}& 1+\frac{2M_j}{kz_{2j}}\end{array}\right),\widehat{D}_j=\left(\begin{array}{cc}1& D_j\\ 0& 1\end{array}\right),$$
where the constant $`D_j`$ are defined as follows:
$$D_{N+1}=(\stackrel{~}{Y}(\mathrm{})\stackrel{~}{Y}(z_{2N})),D_j=(\stackrel{~}{Y}(z_{2j1})\stackrel{~}{Y}(z_{2j2})),$$
$$D_1=(\stackrel{~}{Y}(z_1)\stackrel{~}{Y}(\mathrm{}))$$
In (3.16), we took into account identities (2.9) and (2.10). Notice also that $`detT(k)=1`$.
The cut of $`\omega (k,z,\rho )`$ is the segment that connect the points $`z+i\rho `$ and $`zi\rho `$. Therefore, $`\mathrm{\Psi }^\pm (k)`$ are analytic functions in $`k`$ as $`|`$Im$`k|>\rho `$ and
$$\mathrm{\Psi }^{}(k)=\mathrm{\Psi }^+(k)T(k).$$
$`(3.17)`$
The function $`\mathrm{\Psi }^+(k)`$ ($`\mathrm{\Psi }^{}(k)`$) can be analytically continued in the domains Re$`k<z`$ (Re$`k>z`$). Using (3.5), we can see that for $`k\mathrm{}`$,
$$\mathrm{\Psi }^\pm (k)e^1(k)I$$
$`(3.18)`$
Substituting $`k=z+(\omega ^2\rho ^2)/2\omega `$ for $`k`$, we pass from $`\mathrm{\Psi }^\pm (k)`$ to $`\mathrm{\Psi }^\pm (\omega )`$. Then the function $`\mathrm{\Psi }^\pm (\omega )`$ become solutions to Eqs. (3.1) in the domain $`|\omega |>\rho `$, while $`\mathrm{\Psi }^+(\omega )`$ is analytic in $`\omega `$ as Re$`\omega <0,|\omega |>\rho `$ and $`\mathrm{\Psi }^{}(\omega )`$ is analytic as Re$`\omega >0,|\omega |>\rho `$. Furthermore, it follows from (3.18) that
$$\mathrm{\Psi }^\pm (\omega )e^1(\omega )I$$
$`(3.19)`$
as $`\omega \mathrm{}`$. Though $`\mathrm{\Psi }^\pm (\omega )`$ are determined for $`|\omega |<\rho `$ as well, they do not satisfy system (3.1) in this domain. Therefore, our next aim is to continue $`\mathrm{\Psi }^\pm (\omega )`$ into the domain $`|\omega |<\rho `$ in a manner that preserves Eqs.(3.1).
Let $`\omega _1(k),\omega _2(k)`$ be the roots of Eq. (3.2) and let Re$`\omega _1(k)<0`$ and Re$`\omega _2(k)>0`$. Then the cuts of $`\omega _{1,2}(k)`$ are half-lines going from points $`z+i\rho `$, $`zi\rho `$ to infinity in a direction that is perpendicular to the real axis. Hence, the functions $`\omega _{1,2}(k)`$ are analytic for $`|`$Im$`k|<\rho `$. Note that $`\omega _1(k)=\omega (k)`$ for Re$`k<z`$ and $`\omega _2(k)=\omega (k)`$ for Re$`k>z`$, whence the functions $`\mathrm{\Psi }^\pm (k)`$ are continued analytically into the strip $`|`$Im$`k|<\rho `$. Further, let $`\mathrm{\Psi }_1^{}(k)`$ and $`\mathrm{\Psi }_2^+(k)`$ are the solutions to Eqs. (3.12) (with $`\omega _1`$ substituted for $`\omega `$) and (3.13) (with $`\omega _2`$ substituted for $`\omega `$), respectively .
For $`|`$Im$`k|<\rho `$, the functions $`\mathrm{\Psi }^+(k)`$ and $`\mathrm{\Psi }_1^{}(k)`$ ($`\mathrm{\Psi }^{}(k)`$ and $`\mathrm{\Psi }_2^+(k)`$) are the solutions of the same differential equation (3.4a). Hence,
$$\mathrm{\Psi }_1^{}(k)=\mathrm{\Psi }^+(k)T_1(k),\mathrm{\Psi }^{}(k)=\mathrm{\Psi }_2^+(k)T_2(k),$$
$`(3.20)`$
where the matrices $`T_{1,2}(k)`$ do not depend on $`z`$. Since the solutions to the integral equations (3.12) and (3.13) automatically satisfy (3.4b) (this follows from boundary conditions (2.12) and the fact that $`e(z,\rho ,k)`$ is a common solution of Eq.(3.4) with the Minkowski tensor), we conclude that $`T_{1,2}(k)`$ do not depends on $`\rho `$ as well. Moreover, at $`\rho \mathrm{}`$,
$$\omega _1(k)\rho +(kz)+O(\frac{1}{\rho }),\omega _2(k)\rho +(kz)+O(\frac{1}{\rho }).$$
Therefore, accounting for Eq.(2.12), we derive that
$$\underset{\rho \mathrm{}}{lim}e_1^1(k)\mathrm{\Psi }_1^{}(k)=\underset{\rho \mathrm{}}{lim}e_1^1(k)\mathrm{\Psi }^+(k)=I,$$
$$\underset{\rho \mathrm{}}{lim}e_2^1(k)\mathrm{\Psi }^{}(k)=\underset{\rho \mathrm{}}{lim}e_2^1(k)\mathrm{\Psi }_2^+(k)=I,$$
$`(3.21)`$
and, hence, $`T_1(k)=T_2(k)=I`$. Here $`e_1(k)=e(\omega _1(k))`$ and $`e_2(k)=e(\omega _2(k)).`$ In other words, for $`|`$Im$`k|<\rho `$, we obtain
$$\mathrm{\Psi }_1^{}(k)=\mathrm{\Psi }^+(k),\mathrm{\Psi }^{}(k)=\mathrm{\Psi }_2^+(k).$$
$`(3.22)`$
The function $`\mathrm{\Psi }_1^{}(k)`$ and $`\mathrm{\Psi }_2^+(k)`$ continued analytically into the domains Re$`k>z`$ and Re$`k<z`$ respectively. Hence, the function $`\mathrm{\Psi }_1^{}(\omega )`$ is analytic for $`|\omega |<\rho `$, Re$`\omega <0`$, and the function $`\mathrm{\Psi }_2^+(\omega )`$ is analytic for $`|\omega |<\rho `$, Re$`\omega >0`$. As these functions are the solutions to Eq.(3.1) in the domain Re$`\omega <0`$ ($`\mathrm{\Psi }_1^{}`$) or in the domain Re$`\omega >0`$ ($`\mathrm{\Psi }_2^+)`$, it follows from (3.22) that $`\mathrm{\Psi }^+(\omega )`$ can be analytically continued into the half-plane Re$`\omega <0`$ and remains a solution of (3.1), and $`\mathrm{\Psi }^{}(\omega )`$ can be analytically continued into the half-plane Re$`\omega >0`$, and also remains a solution of (3.1).
System (3.1) is invariant w. r. t. the transformation
$$\mathrm{\Psi }(z,\rho ,\omega )g\stackrel{~}{\mathrm{\Psi }}^1(z,\rho ,\frac{\rho ^2}{\omega }),$$
$`(3.23)`$
which is valid because the matrix $`g`$ is symmetric (the tilde in (3.23) denotes transposition). Reduction (3.23) means that $`\mathrm{\Psi }^{}(k)`$ and $`g[\stackrel{~}{\mathrm{\Psi }}_1^{}(k)]^1`$ are the solutions of the compatible pair of equations (3.4) with $`\omega =\omega _1`$. However, at $`\rho \mathrm{}`$,
$$e_2^1(k)ge_1^1(k)I.$$
$`(3.24)`$
Then from (3.21) and (3.24) we have that $`\mathrm{\Psi }^{}(k)=g[\stackrel{~}{\mathrm{\Psi }}_1^{}(k)]^1`$ ,or, equivalently,
$$\mathrm{\Psi }^{}(\omega )=g[\stackrel{~}{\mathrm{\Psi }}^+(\frac{\rho ^2}{\omega })]^1.$$
$`(3.25)`$
The functions $`\mathrm{\Psi }^+(\omega )`$ and $`\mathrm{\Psi }^{}(\omega )`$ satisfy the compatible system of equations (3.1). Therefore, the combination $`[\mathrm{\Psi }^+(\omega )]^1\mathrm{\Psi }^{}(\omega )`$ depends only on $`k=z+(\omega ^2\rho ^2)/2\omega `$ ($`D_1k=D_2k=0`$). Then, by virtue of identity (3.17),
$$\mathrm{\Psi }^{}(\omega )=\mathrm{\Psi }^+(\omega )T(k).$$
$`(3.26)`$
Equations (3.26) and (3.25) show that
$$T(k)=\stackrel{~}{T}(k).$$
$`(3.27)`$
The monodromy matrix, $`T(k)`$ depends on $`3N+1`$ parameters except $`z_k`$. We treat equality (3.27) as the system of $`2N+1`$ nonlinear algebraic equations for the constants $`D_j,L_j`$ and $`\mathrm{\Omega }_j`$:
$$\underset{j=1}{\overset{N+1}{}}D_j+\underset{j=1}{\overset{N}{}}4\mathrm{\Omega }_jM_j=0,\text{Res}_{z_k}T_{12}(k)=\text{Res}_{z_k}T_{21}(k).$$
$`(3.28)`$
Assume $`D_j`$ can be excluded from (3.28); then the remaining $`N`$ equations give us the connection between the angular velocities and angular momenta. For instance, for the case of a single black hole, it follows from (3.28) that
$$D_1=D_2=2M_1\mathrm{\Omega }_1,\mathrm{\Omega }_1=\frac{L_1}{2M_1^2(m_1+M_1)}.$$
$`(3.29)`$
The expression for $`\mathrm{\Omega }_1`$ in (3.29) is known; it establishes the connection between the angular velocity and angular momentum of the Kerr black hole. The relationship is expressed by
$$\mathrm{\Omega }_1=\frac{a}{(M_1+m_1)^2+a^2},a=\frac{L_1}{M_1},M_1^2=m_1^2+a^2.$$
As in (3.29), equality (2.10) is also taken into account here. Then the monodromy matrix reads
$$T(k)=\left(\begin{array}{cc}1\frac{2M_1}{kz_1}+\frac{2M_1(M_1m_1)}{(kz_1)(kz_2)}& \frac{2L_1}{(kz_1)(kz_2)}\\ \frac{2L_1}{(kz_1)(kz_2)}& 1+\frac{2M_1}{kz_2}+\frac{2M_1(M_1m_1)}{(kz_1)(kz_2)}\end{array}\right)$$
Let us summarize the results of this section. Let boundary problem (2.4), (2.12) have a solution. Then there exists a piecewize analytic matrix $`\mathrm{\Psi }(\omega )`$ ($`\mathrm{\Psi }(\omega )=\mathrm{\Psi }^+(\omega ),`$ Re$`\omega <0`$ and $`\mathrm{\Psi }(\omega )=\mathrm{\Psi }^{}(\omega ),`$ Re$`\omega >0`$) that satisfies the compatible system of linear equations (3.1), the conjugation condition on the imaginary axis,
$$\mathrm{\Psi }_{}(\omega )=\mathrm{\Psi }_+(\omega )T(k),\mathrm{Re}\omega =0$$
$`(3.30)`$
and the normalization condition at infinity,
$$\mathrm{\Psi }(\omega )\left(\begin{array}{cc}1& 0\\ 0& \frac{1}{\omega }\end{array}\right)I,\omega \mathrm{}.$$
$`(3.31)`$
In (3.30) $`\mathrm{\Psi }_\pm (\omega )=lim_{ϵ0}\mathrm{\Psi }(\omega ϵ)`$ $`(ϵ>0)`$.
The only singularities of the matrix $`T(k)`$ as a function of $`\omega `$ are simple poles at the points $`\omega _i^\pm =(z_iz)\pm \sqrt{(z_iz)^2+\rho ^2}.`$ Since the matrix
$$\left(\begin{array}{cc}1& 0\\ 0& \omega \end{array}\right)T(k)\left(\begin{array}{cc}1& 0\\ 0& 1/\omega \end{array}\right)$$
is regular at $`\omega =0`$ and tends to $`I`$ as $`\omega \mathrm{}`$ (which follows from explicit form of $`T(k)`$ and from Eqs. (3.27), (3.28)), we conclude that
$$\mathrm{\Psi }^\pm (\omega )=\left(I+\underset{j=1}{\overset{2N}{}}\frac{A_j^\pm }{\omega \omega _j^\pm }\right)\left(\begin{array}{cc}1& 0\\ 0& \omega \end{array}\right),$$
$`(3.32)`$
where $`A_j^\pm `$ do not depend on $`\omega `$.
Formula (3.32) demonstrate that a rational-in-$`\omega `$ solution to the auxiliary linear problem corresponds to each solution with a disconnected horizon and, hence, each such solution should be contained in the Belinskii-Zakharov class of solutions .
Assume that for any $`\mathrm{\Omega }_i,z_i`$ (or $`L_i,z_i`$) the system of nonlinear equations (3.28) has a unique solution. Then a unique symmetric matrix $`T(k)`$ corresponds to each solution of the boundary problem (2.3), (2.4), and (2.12). However, since the solution to the Riemann problem (3.30), (3.31) is unique and $`g`$ is unambiguously reconstructed by $`\mathrm{\Psi }(\omega )`$ (see 3.25), we conclude that if a solution to the problem (2.3), (2.4), (2.12) exists, then it is unique as well. In particular, for the case where only one black hole is present, we obtain a new proof of the uniqueness of the Kerr solution.
The present paper was supported by the RFBR grant No. 96-01-00548.
|
warning/0004/hep-th0004191.html
|
ar5iv
|
text
|
# References
Soldering and Mass Generation in Four Dimensions
Rabin Banerjee<sup>b</sup> and Clovis Wotzasek<sup>a,b</sup>
<sup>a</sup>Instituto de Física
Universidade Federal do Rio de Janeiro
21945, Rio de Janeiro, Brazil
<sup>b</sup>S. N. Bose National Centre for Basic Sciences,
Block JD, Sector III, Salt Lake, Calcutta 700091, India.
## Abstract
We propose bosonised expressions for the chiral Schwinger models in four dimensions. Then, in complete analogy with the two dimensional case, we show the soldering of two bosonised chiral Schwinger models with opposite chiralities to yield the bosonised Schwinger model in four dimensions. The implications of the Schwinger model or its chiral version, as known for two dimensions, thereby get extended to four dimensions.
The Schwinger mechanism of generating a photon mass in two space time dimensions is crucial for our understanding of several aspects of different models in such dimensions. The particular features of the Schwinger model to produce a massive vectorial mode, while at same time avoiding the Goldstone bosons were, at some point, atributed to be a consequence of two-dimensional kinematics. As so these results could not be immediately generalized to higher dimensions. However, by following the soldering mechanism proposed by Stone , the massive mode in the Schwinger model was shown by us to be an interference effect between the massles modes of the chiral Schwinger models . Since the soldering formalism was very general and did not depend on any specific properties of two dimensions, it was felt that the analysis should be extended to higher dimensions. This would also provide a scenario for mass generation in these dimensions. In order to proceed with the analysis it is therefore essential to have some bosonised expression characterising the Schwinger model in four dimensions. This was achieved by Aurilia, Takahashi and Townsend. The model had important phenomenological consequences. In particular, it was found to simulate the effective theory representing $`QCD`$ in the large $`N`$ limit. Just as the bosonised Schwinger model in two dimensions was thought to be the effective theory for $`QED_2`$, it was argued that the appropriate generalisation of $`QED_2`$ to four dimensions done in was a prototype of $`QCD`$ in the large N limit . More recently, applications in other contexts like the study of skyrmions or Josephson arrays have also been mentioned . Now the Schwinger model in four dimensions was constructed by a direct lift of the well known bosonised version of the model in two dimensions. A natural question that arises in this context and which lacked an answer would be the construction of the chiral Schwinger model in four dimensions. We could then interpret this theory to characterise chiral $`QCD`$ in the large N limit. Of course we would like to have the result in the bosonised form since it is in this form the Schwinger model was written in four dimensions. It may be recalled that the chiral Schwinger model in two dimensions is iteslf an interesting model providing an alternative mechanism for mass generation, which is basically through the existence of an anomaly. One might be tempted to repeat the same logic of directly lifting the bosonised version of the chiral Schwinger model in two dimensions, which is familiar, to four dimensions. Such an approach, alas, fails. The reason is that in two dimensions the vector and axial currents are related by an identity so that effectively there is no difference between them. This is not so in four dimensions. Consequently, a simple lift of the chiral version of the Schwinger model from two to four dimensions is not possible.
Recently we have developed a method by which it is possible to combine two distinct models into an efective model. This is called the soldering technique. The idea is to exploit the dual aspects of the symmetry of the constituent models to construct a new model which effectively hides the symmetries. In the example at hand it is the chiral symmetry. Thus we expect that by soldering two chiral Schwinger models of opposite chiralities, the normal Schwinger model is reproduced. This was shown by us earlier in the usual case of two dimensions. By extending those notions we will construct here the explicit bosonised forms of the chiral Schwinger models in four dimensions. However this construction is fundamentally different from the two dimensional case. It necessitates the introduction of an internal space quite similar to what is done for describing electric-magnetic duality symmetric lagrangians and must be defined in the Fourier space. By soldering these chiral models we correctly reproduce the Schwinger model lagrangian in four dimensions as given in . Comparing with the two dimensional analysis we conclude that exactly the same mechanism for mass generation occurs also in four dimensions.
To illustrate the concepts in a simple setting, as well as for reasons of comparison, the soldering of two chiral models to yield a vector model in two dimensions is shown. The first order lagrangians for the gauged chiral bosons are given by ,
$$_R=\dot{\varphi }\varphi ^{}\varphi ^22e\varphi ^{}(A_0+A_1)\frac{e^2}{2}(A_0+A_1)^2+\frac{ae^2}{2}A_\mu A^\mu ,$$
(1)
$$_L=\dot{\rho }\rho ^{}\rho ^2+2e\rho ^{}(A_0A_1)\frac{e^2}{2}(A_0A_1)^2+\frac{be^2}{2}A_\mu A^\mu ,$$
(2)
These lagrangians correspond to the coupling with the right (R) handed and left (L) handed pieces, respectively. They were obtained by imposing the chirality constraints on the standard vector lagrangian so that one chirality is killed while the other is retained. The parameters $`a`$ and $`b`$ reflect the bosonisation or regularisation ambiguity of the chiral determinants. Imposing Bose symmetry , which means that both determinants are accounted by the same sort of ambiguity, we can effectively set $`a=b`$. Each of these lagrangians therefore corresponds to half a degree of freedom. Furthermore, taking the opposite chiralities into account we can count $`\varphi `$ to have half a degree of freedom while $`\rho `$ would carry minus half a degree of freedom.
Next, consider the variation of the lagrangians under the following transformations,
$$\delta \varphi =\delta \rho =\alpha ;\delta A_\mu =0.$$
(3)
We find,
$$\delta _R=2J_R\alpha ^{},\delta _L=2J_L\alpha ^{},$$
(4)
where the currents are given by,
$$J_R=(\dot{\varphi }+\varphi ^{}+e(A_0+A_1)),J_L=(\dot{\rho }\rho ^{}+e(A_0A_1)).$$
(5)
Introducing the soldering field $`B`$, transforming as,
$$\delta B=2\alpha ^{}$$
(6)
it is found that the soldered Lagrangian,
$$=_R_L=_R+_L+B(J_R+J_L)\frac{1}{2}B^2,$$
(7)
remains invariant under the combined transformations (3) and (6). Eliminating $`B`$ in favour of the other variables yields the final effective lagrangian,
$$=\frac{1}{2}_\mu \theta ^\mu \theta +2eϵ_{\mu \nu }A^\mu ^\nu \theta +(a1)e^2A_\mu A^\mu ;ϵ_{01}=1$$
(8)
where the new field $`\theta `$ is defined as,
$$\theta =\varphi \rho $$
(9)
The familiar gauge invariant vector lagrangian is obtained by setting $`a=1`$. Inclusion of the Maxwell term right from the beginning corresponds to the soldering of two chiral Schwinger models to yield the normal Schwinger model. All results, of course, are to be understood in the bosonised version. It might be observed that the $`a=b=1`$ parametrisation corresponds to the massless modes in the chiral Schwinger models. Thus the massive mode of the Schwinger model is a consequence of the interference of two massless modes in the chiral models.
Regarding the degree of freedom count we find the soldering of two half degree of freedom to yield a single degree of freeeom. This is also contained in the algebraic relation (9) which, for the degree of freedom count, can be written as $`1=\frac{1}{2}()\frac{1}{2}`$.
We shall now discuss the construction of the bosonised version of the chiral Schwinger models in four dimensions. The difficulties of a direct lift from its two dimensional counterpart have been already mentioned. One might think of an alternative possibility; namely of applying the chirality constraint on the four dimensional Schwinger model. This also fails since the peculiar form that the chiral constraints assume in the two dimensional case does not admit an unambiguous local dimensional extension. It is our main goal in this paper to propose such a dimensional lift of the chiral Schwinger model and to show that the interference effects provided by the soldering mechanism are able to produce the gauge invariant massive mode of the model in .
In order to motivate the analysis in four dimensions the results in two dimensions are reconsidered by omitting the gauge couplings. This implies the soldering of two free chiral bosons having opposite chiralities to yield the free scalar theory. The equation of motion for this Klein-Gordon field may be expressed in a factorised form,
$$\mathrm{}\theta =(_0+_1)(_0_1)\theta =0$$
(10)
where $`\mathrm{}`$ is the usual Klein Gordon operator. The factored forms are recognised to represent the equations of motion for the chiral bosons, following from (1) and (2),
$$\dot{\theta }\pm \theta ^{}=0$$
(11)
Indeed if we naively multiply the expressions appearing in the l.h.s. of the above equation we just recover the standard free scalar lagrangian. The main thrust would therefore be to find the four dimensional lagrangians leading to a set of equations mimicing (11). With these observations the ensuing analysis in four dimensions becomes more transparent.
It is now useful to go over to the $`k`$-space. To this end let us introduce an internal two-dimensional space spanned by the basis, $`\left\{\widehat{e}_a(𝐤,𝐱),a=1,2\right\}`$, with $`(𝐤,𝐱)`$ being conjugate variables and the orthonormalization condition given as,
$$d^3𝐱\widehat{e}_a(𝐤,𝐱)\widehat{e}_b(𝐤^{},𝐱)=\delta _{ab}\delta (𝐤𝐤^{})$$
(12)
We choose the vectors in the basis to be eigenvectors of the Laplacian, $`^2=_m_m`$,
$$^2\widehat{e}_a(𝐤,𝐱)=\omega ^2(𝐤)\widehat{e}_a(𝐤,𝐱)$$
(13)
and the action of $`_m`$ over the $`\widehat{e}_a(𝐤,𝐱)`$ basis to be
$$_m\widehat{e}_a(𝐤,𝐱)=k_mϵ_{ab}\widehat{e}_b(𝐤,𝐱);ϵ_{12}=1,$$
(14)
so that the dispersion relation $`k_\mu k^\mu =\omega ^2k_m^2=0`$ holds. Finally, we define the basic chiral fields carrying the opposite $`(\pm )`$ chiralities in the momentum space to be the expansion coefficients of a chiral scalar field in the function space as,
$$\varphi ^{(\pm )}(𝐱,t)=d^3𝐤\phi _a^{(\pm )}(t,𝐤)e_a(𝐤,𝐱).$$
(15)
Notice that the internal index disappears upon Fourier transforming the fields back in the coordinate space.
The expected equations of motion displaying the chiral properties would therefore be given by,
$$\dot{\phi }_a(k)\omega ϵ_{ab}\phi _b(k)=0$$
(16)
This is the exact analogue of the equation of motion for the free chiral bosons in two dimensions given in (11). Proceeding as before if we multiply the factors occuring in the l.h.s. of the above equation, we obtain, after making a Fourier transform to the coordinate space,
$$d^4k(\dot{\phi }_a+\omega ϵ_{ab}\phi _b)(\dot{\phi }_a\omega ϵ_{ac}\phi _c)=d^4x(_\mu \varphi ^\mu \varphi )$$
(17)
which is just the action for the free Klein Gordon field in four dimensions.
We next construct a first order action that would yield the equation of motion (16). This is given by,
$$𝒮_\pm ^{(0)}=d^4k\left[\pm \omega (𝐤)\dot{\phi }_a^{(\pm )}ϵ_{ab}\phi _b^{(\pm )}\omega ^2(𝐤)\phi _a^{(\pm )}\phi _a^{(\pm )}\right].$$
(18)
The basic brackets among the fields are easily read off from the symplectic structure,
$$\{\phi _a(k),\phi _b(k^{})\}=\pm \frac{1}{2\omega (𝐤)}ϵ_{ab}\delta (kk^{})$$
(19)
for either chirality. Since this is a first order action the hamiltonian is also simply read off,
$$H=d^3𝐤\omega ^2(𝐤)\phi _a^{(\pm )}\phi _a^{(\pm )}$$
(20)
From (19) and (20) the desired result (16) follows. It is in fact possible to explicitly carry out the soldering of these four dimensional chiral boson lagrangians to get the usual Klein Gordon lagrangian. We bypass this to discuss the interacting theory. Introduce the chiral combinations of gauge fields $`𝒜_\pm ^a`$ as follows,
$$𝒜_\pm ^a=ϵ^{ab}\left(\omega ^{}A_0^b\pm k_m^{}A_m^b\right)$$
(21)
where $`^{}A_\mu ^b`$ is the Hodge-dual of a pair of three form gauge fields $`A_{\mu \nu \lambda }^a`$, defined as,
$$^{}A_a^\mu =\frac{1}{2}ϵ^{\mu \nu \lambda \rho }A_{\nu \lambda \rho }^a;ϵ^{0123}=1$$
(22)
We now propose the four dimensional lift of the bosonised version of the interacting theory by a simple extension of the free theory (18),
$$𝒮_\pm =d^4k\left[\pm \omega (𝐤)\dot{\phi }_a^{(\pm )}ϵ^{ab}\phi _b^{(\pm )}\omega ^2(𝐤)\phi _a^{(\pm )}\phi _a^{(\pm )}\mathrm{\hspace{0.17em}2}g𝒜_\pm ^a\phi _a^{(\pm )}\frac{g^2}{2\omega ^2(𝐤)}𝒜_\pm ^a\right]+\frac{ag^2}{2}𝒮_{reg},$$
(23)
where $`𝒮_{reg}`$ manifests the bosonisation ambiguity, quite similarly to the two dimensional case, and is given by,
$`𝒮_{reg}`$ $`=`$ $`{\displaystyle d^4k\left[\left(^{}A_a^0\right)^2\frac{1}{\omega ^2}\left(k_m^{}A_a^m\right)^2\right]}`$ (24)
$`=`$ $`{\displaystyle d^4x\left[^{}A^m\left(\delta _{mn}\frac{_m_n}{^2}\right)^{}A^n+^{}A^\mu ^{}A_\mu \right]}`$
Notice that if we dimensionally reduce this expression to D=2 the first piece in the second line vanishes and we get back the Jackiw-Rajaraman regularisation ambiguity term. Indeed by including the Maxwell term in (23) the corresponding actions represent the bosonised versions of the chiral Schwinger models with opposite chiralities.
The soldering of the two pieces in (23) to reproduce the usual gauge invariant theory in four dimensions will now be demonstrated. Consider the soldering transformations
$$\phi _a^{(\pm )}=\eta _a;\delta ^{}A_a^\mu =0$$
(25)
and compute the Noether charges,
$$J_a^{(\pm )}=2\left(\pm \omega \phi _b^{(\pm )}ϵ_{ba}\omega ^2\phi _a^{(\pm )}g𝒜_a^\pm \right)$$
(26)
from the variations
$$\delta 𝒮_\pm =J_a^{(\pm )}\eta _a$$
(27)
The effective soldered action in momentum space that is invariant under (25), will be,
$$𝒮_{eff}[\varphi _a]=𝒮_+^{(0)}[\phi _a^{(+)}]+𝒮_{}^{(0)}[\phi _a^{()}]+d^4k\frac{1}{8\omega ^2}\left\{J_a^{()}[\phi _a^{(+)}]+J_a^{(+)}[\phi _a^{()}]\right\}^2$$
(28)
where $`\varphi _a=\phi _a^{(+)}\phi _a^{()}`$. Next we perform the inverse Fourier transform to obtain the coordinate space efecctive action as,
$$𝒮_{eff}[\varphi ]=d^4x\left\{\frac{1}{2}\left(_\mu \varphi \right)^2g_\mu \varphi ϵ^{\mu \nu \lambda \rho }A_{\nu \lambda \rho }\right\}+(a1)𝒮_{reg}.$$
(29)
Exactly as happened in the two dimensional case, the effective action will be gauge invariant provided,
$$a=1$$
(30)
By including the appropriate Maxwell term to impart dynamics to the gauge field, this is the precise form of the four dimensional Schwinger model as given in . The analogy with the two dimensional analysis is now complete.
Just as the chiral Schwinger models in two dimensions were soldered to yield the usual Schwinger model, the same has been feasible in four dimensions. The generation of the massive mode also follows along the same lines. The observation that not only the different models but the connection between them is very similar in two and four dimensions perhaps serves to demystify the special role that is usually assigned to the former. Indeed our analysis is completely general applicable to any 2n-dimensions. In other words the Schwinger model obtained by a straightforward lift can always be thought as an interference effect among the chiral components analogous to the forms given here. A natural outcome of the analysis has been the obtention of lagrangians representing chiral bosons in any even dimensions. Regarding further prospects it would be nice to actually show the connection between the effective theories proposed here and chiral $`QCD`$ in the large N limit, which is expected from plausibility arguments that relate the bosonised schwinger model in four dimensions with $`QCD`$ in the large N limit.
|
warning/0004/math-ph0004018.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
In the theory of random matrices , the n-point correlation functions of the eigenvalues are known to be expressible as the determinant of a two-point kernel . The expressions for those kernels depend on the various classes of universality : it is a simple sine-kernel within the bulk of unitary invariant ensembles, an Airy kernel at the edge of the spectrum or a Bessel kernel for other invariance properties of the measure. The level spacing probability $`p(s)`$, has also been computed recently for those different kernels .
Another interesting object is given by the average moments of the characteristic polynomial of the random matrix. These characteristic polynomials have been first investigated in for a uniform probability measure on unitary matrices, in connection with the moments of the Riemann zeta-function. These results have been generalized to random hermitian $`N\times N`$ matrices $`X`$ with a unitary invariant probability measure
$$P(X)=\frac{1}{Z}\mathrm{exp}N\mathrm{TrV}(\mathrm{X}).$$
(1)
Explicit formulae for the $`2K`$-point functions
$$F_{2K}(\lambda _1,\mathrm{},\lambda _{2K})=<\underset{1}{\overset{2K}{}}det(\lambda _lX)>$$
(2)
have been derived, which show that these functions are universal in the Dyson limit, in which the size N of the matrices goes to infinity, the distances between the $`\lambda `$’s go to zero, and the products $`N(\lambda _i\lambda _j)`$ remain finite. In particular the moments
$$F_{2K}(\lambda ,\mathrm{},\lambda )=<[det(\lambda M)]^{2K}>$$
(3)
of the distribution of the characteristic polynomials were given in the large N limit by
$$\mathrm{exp}(NKV(\lambda ))F_{2K}(\lambda ,\mathrm{},\lambda )=(2\pi N\rho (\lambda ))^{K^2}e^{NK}\gamma _K,$$
(4)
with
$$\gamma _K=\underset{0}{\overset{K1}{}}\frac{l!}{(K+l)!},$$
(5)
provided $`\lambda `$ belongs to the bulk of the support of the distribution of the eigenvalues, i.e. provided $`\rho (\lambda )`$ does not vanish. Then one sees explicitely that the only dependence upon the probability measure is through the average density of eigenvalues $`\rho (\lambda )`$, and even the coefficient $`\gamma _K`$ is a universal number.
However the result does take different forms for different universality classes. Our previous investigations for the three classical Lie groups, U(N), Sp(N) and O(N), are extended here to the Bessel kernel and Airy kernel, for which the density of states $`\rho (\lambda )`$ presents a singularity at the edge of the spectrum. Furthermore we have considered a Gaussian case in which an external matrix source is present in the probability distribution of the matrix
$$P(X)=\frac{1}{Z}\mathrm{exp}(N\mathrm{Tr}\frac{1}{2}\mathrm{X}^2+\mathrm{NTrAX}).$$
(6)
Explicit and simple formulae will be derived here again for the correlation functions and the moments of the characteristic polynomials of the matrix $`X`$, which depend on the spectrum of the matrix $`A`$. By tuning the spectrum of $`A`$ appropriately, one can generate a number of different situations. For instance, we have investigated in the past the case in which the average spectrum of $`X`$ presents a gap in the presence of $`A`$, and by tuning $`A`$ one can study the critical point at which this gap vanishes. This creates again a new class of universality, and a new kernel . Other cases, such as 2D gravity in the double scaling limit, or the Penner model, would certainly be of interest as well.
## 2 Sine-kernel
For completeness, and for later use, we begin with the bulk unitary case, governed by the sine-kernel, but with a derivation which differs from our previous one . An interesting geometric interpretation of this problem will also be provided. The kernel, from which all the correlation functions may be obtained, is given in terms of orthogonal polynomials for finite N, but reduces in the Dyson large N-limit to the sine-kernel
$$K(x,y)=\frac{\mathrm{sin}(xy)}{xy}$$
(7)
in which $`x`$ and $`y`$ are the eigenvalues measured in the scale of the average spacing $`\left(2\pi \rho (\lambda )N\right)^1`$. Then, one obtains the normalized moments
$$I_K=e^{NKV(\lambda )}\frac{<[det(\lambda X)]^{2K}>}{(2\pi N\rho (\lambda ))^{K^2}}=\underset{\lambda _i0}{lim}\frac{detK(\lambda _i,\lambda _j)}{\mathrm{\Delta }^2(\mathrm{\Lambda })}$$
(8)
where $`\mathrm{\Delta }(\mathrm{\Lambda })=_{i<j}(\lambda _i\lambda _j)`$ and $`i,j=1,\mathrm{},K.`$ The r.h.s. may be expressed as a contour integral, following eq.(52) of ,
$$I_K=\underset{i}{\overset{K}{}}\frac{du_idv_i}{(2\pi i)^2}\frac{\mathrm{\Delta }(U)\mathrm{\Delta }(V)}{_{i=1}^Ku_i^K_{i=1}^Kv_i^K}\underset{i=1}{\overset{K}{}}\frac{\mathrm{sin}(u_iv_i)}{u_iv_i}$$
(9)
This may be further reduced to
$$I_K=det(a_{nm})$$
(10)
$$a_{nm}=\frac{1}{n!m!}\frac{^n}{u^n}\frac{^m}{v^m}\frac{\mathrm{sin}(uv)}{uv}|_{u=v=0}$$
(11)
where $`n,m=0,1,\mathrm{},K1`$. The explicit evaluation of the determinant of $`a_{n,m}`$ gives
$$det(a_{nm})=2^{K^2K}\underset{l=0}{\overset{K1}{}}\frac{l!}{(K+l)!}$$
(12)
We do recover in this way the factor $`\gamma _K`$ (5)(up to a factor $`2^{K^2K}`$ due to a different normalization normalization of the kernel).
It is quite remarkable that this universal normalizing factor $`\gamma _K`$ has a geometric interpretation as a Fredholm determinant of the Dirac Laplacian on the two dimensional sphere $`S^2`$. The determinant of the Laplacian has been discussed in the connection to string theory , and the relation of $`\gamma _K`$ to this Fredholm determinant of the Laplacian has been noticed in . Indeed let us show that
$$\gamma _K=\frac{e^{K^2(1+\gamma )}}{\mathrm{\Delta }^+(K)}$$
(13)
where $`\gamma `$ is Euler’s constant and $`\mathrm{\Delta }^+(z)`$ the determinant of a Dirac operator, defined below. The derivation goes as follows : let us introduce a function $`G(z)`$ which satisfies the functional relation
$$G(z+1)=\mathrm{\Gamma }(z)G(z).$$
(14)
It is then straightforward to verify that
$$\gamma _K=\underset{l=0}{\overset{K1}{}}\frac{l!}{(K+l)!}=2^{K2K^2}\frac{\pi ^{K+\frac{1}{2}}}{\mathrm{\Gamma }(K+\frac{1}{2})}\left[\frac{G(\frac{1}{2})}{G(K+\frac{1}{2})}\right]^2.$$
(15)
A function $`G`$, satisfying the functional relation (14), is known in the literature as a Barnes function (or as the inverse of a di-gamma function). It is defined by
$$G(z+1)=\frac{1}{\mathrm{\Gamma }_2(z+1)}=(2\pi )^{z/2}e^{\frac{1}{2}\left[z+(1+\gamma )z^2\right]}\underset{1}{\overset{\mathrm{}}{}}\left[(1+\frac{z}{n})^ne^{z+z^2/(2n)}\right].$$
(16)
It has been noticed earlier ()that this Barnes function is related to the Fredholm determinant of the Laplacian on $`S^2`$ . Indeed this Fredholm determinant is the (regularized) product
$$\mathrm{\Delta }(z)=\underset{l}{}(1\frac{z}{\lambda _l})^{g_l}$$
(17)
where the $`\lambda _l`$ are the eigenvalues of the Laplacian, and $`g_l`$ their degeneracy, i.e. $`\lambda _l=l(l+1)`$ with multiplicity $`g_l=2l+1`$, $`l=0,1,2,\mathrm{}`$. It is convenient to shift $`z`$ by $`1/4`$, since this yields the the spectrum of the Dirac operator
$$\sqrt{\lambda _l+\frac{1}{4}}=l+\frac{1}{2}$$
(18)
Then the regularized (shifted) Fredholm determinant
$$\mathrm{\Delta }(z)=\underset{l=0}{\overset{\mathrm{}}{}}[(1\frac{z}{(l+\frac{1}{2})^2})e^{\frac{z}{(l+1/2)^2}}]^{2l+1},$$
(19)
factorizes as
$$\mathrm{\Delta }(y^2)=\mathrm{\Delta }^+(iy)\mathrm{\Delta }^+(iy)$$
(20)
with the determinant of the Dirac operator $`\mathrm{\Delta }^+(z)`$ given by
$$\mathrm{\Delta }^+(z)=\underset{l=0}{\overset{\mathrm{}}{}}[(1\frac{z}{l+\frac{1}{2}})e^{\frac{z}{l+1/2}+\frac{z^2}{2(l+1/2)^2}}]^{2l+1}.$$
(21)
Then this Dirac determinant $`\mathrm{\Delta }^+`$ is related to the Barnes function by
$$\mathrm{\Delta }^+(z)=\pi ^{\frac{1}{2}}(2\pi )^ze^{(1+\gamma +2\mathrm{log}2)z^2}\frac{\mathrm{\Gamma }(\frac{1}{2}z)G(\frac{1}{2}z)^2}{G(\frac{1}{2})^2}.$$
(22)
We thereby recover the expression relating the moment $`\gamma _K`$ to the determinant (13).
This relation between the moments of the distribution and the determinant of the Dirac operator on $`S^2`$ is in fact general. For instance in the simplest case of a single Gaussian random variable, the moments are
$$_{\mathrm{}}^{\mathrm{}}x^{2K}e^{x^2}𝑑x=\mathrm{\Gamma }(K+\frac{1}{2});$$
(23)
$`\mathrm{\Gamma }(K+1/2)`$ is thus the equivalent of $`\gamma _K`$ for this trivial problem. If we consider the ”Laplacian”, i.e. the harmonic oscillator whose eigenvalues are $`\lambda _n=n`$, then the Fredholm determinant $`\mathrm{\Delta }(\lambda )`$ is
$`\mathrm{\Delta }(\lambda )`$ $`=`$ $`\lambda {\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}(1{\displaystyle \frac{\lambda }{n}})e^{\frac{\lambda }{n}}`$ (24)
$`=`$ $`{\displaystyle \frac{e^{\gamma \lambda }}{\mathrm{\Gamma }(\lambda )}}`$
Hence, we have
$$<x^{2K}>\frac{e^{\gamma \lambda }}{\mathrm{\Delta }(\lambda )}|_{\lambda =\left(K+\frac{1}{2}\right)}$$
(25)
The expression (13) is a multi-variable version of this Gaussian integral.
An additional point of interest is that the Fredholm determinant of this Laplacian on $`S^2`$ may be factorized further into a product of two factors ; it turns out that each factor enters into the corresponding expression for the symplectic and orthogonal cases respectively. This will be seen below when we examine the moments related to the Bessel kernel.
## 3 Bessel kernel
We have discussed in our previous work the ensembles invariant under the unitary symplectic and unitary orthogonal Lie groups . The kernels for those ensembles are
$$K(x,y)=\frac{1}{2\pi }(\frac{\mathrm{sin}(xy)}{xy}\frac{\mathrm{sin}(x+y)}{x+y})$$
(26)
where the minus sign corresponds to the $`Sp`$ and the plus sign to the $`O`$ ensemble. It is convenient to introduce the Bessel kernel defined by
$$K_\alpha (x,y)=\frac{J_\alpha (x)J_\alpha ^{}(y)J_\alpha ^{}(x)J_\alpha (y)}{xy}$$
(27)
Since $`J_{1/2}(x)=\sqrt{2/\pi x}\mathrm{sin}x`$, $`J_{1/2}(x)=\sqrt{2/\pi x}\mathrm{cos}x`$, the kernels for the $`Sp`$ and $`O`$ ensembles are both related to this Bessel kernel
$$K_\pm (x,y)=\sqrt{xy}K_{\pm 1/2}(x^2,y^2)$$
(28)
namely, $`\alpha =1/2`$ and $`\alpha =1/2`$ represent respectively the $`Sp`$ and the $`O`$ ensemble. We consider from now on an arbitary $`\alpha `$. The 2K-th moment at the origin $`(\lambda =0)`$ is expressed as
$$I_K=\frac{dudv}{(2\pi )^2}\frac{\mathrm{\Delta }(u^2)\mathrm{\Delta }(v^2)}{_{i=1}^Ku_i^{2K}v_i^{2K}}\underset{i=1}{\overset{K}{}}(u_iv_i)^\alpha K_\alpha (u_i,v_i)$$
(29)
We define now the two functions $`\varphi (z)`$ and $`\psi (z)`$ by
$$J_\alpha (\sqrt{z})=(\frac{\sqrt{z}}{2})^\alpha \frac{1}{\mathrm{\Gamma }(\alpha +1)}\varphi (z)$$
(30)
and
$$\sqrt{z}J_\alpha ^{}(\sqrt{z})=\frac{z^{\alpha /2}}{2^\alpha \mathrm{\Gamma }(\alpha )}\psi (z).$$
(31)
Their expansions in powers of $`x`$ are given by
$$\varphi (x)=\underset{n=0}{\overset{\mathrm{}}{}}\frac{(1)^nx^n}{4^nn!_{l=1}^n(\alpha +l)}$$
(32)
$$\psi (x)=\underset{n=0}{\overset{\mathrm{}}{}}\frac{(1)^nx^n(\alpha +2n)}{4^nn!_{l=0}^n(\alpha +l)}$$
(33)
Keeping aside trivial factors we are then led to the kernel $`\stackrel{~}{K}_\alpha (x,y)`$ defined as
$$\stackrel{~}{K}_\alpha (x,y)=\frac{1}{2(xy)}[\varphi (x)\psi (y)\psi (x)\varphi (y)]$$
(34)
As before, we have
$$I_K=det(a_{nm})$$
(35)
with
$$a_{nm}=\frac{1}{n!m!}\frac{^n}{u^n}\frac{^m}{v^m}\stackrel{~}{K}_\alpha (u,v)|_{u=v=0}$$
(36)
This determinant may be computed explicitly, and it is given by
$$I_K=4^{K^2\alpha K}\underset{l=0}{\overset{2K1}{}}\frac{1}{(\alpha +l)!}$$
(37)
(We have $`I_1=\frac{1}{4},\frac{1}{3\pi },\frac{1}{\pi }`$ for $`\alpha =0,\frac{1}{2}`$ and $`\frac{1}{2}`$, respectively.)
It is interesting to relate the three determinants that we have introduced hereabove for the sine-kernel and for the $`Sp`$ and $`O`$ cases. The determinant for the sine-kernel (11) is
$`I_U=det\left(\begin{array}{ccccc}1& 0& \frac{1}{6}& 0& \mathrm{}\\ 0& \frac{1}{3}& 0& \frac{1}{30}& \mathrm{}\\ \frac{1}{6}& 0& \frac{1}{20}& 0& \mathrm{}\\ 0& \frac{1}{30}& 0& \frac{20}{7!}& \mathrm{}\end{array}\right).`$ (38)
In the symplectic case, $`\alpha =\frac{1}{2}`$, we have
$`I_{Sp}=det\left(\begin{array}{ccc}\frac{1}{3}& \frac{1}{30}& \mathrm{}\\ \frac{1}{30}& \frac{20}{7!}& \mathrm{}\\ \mathrm{}& \mathrm{}& \mathrm{}\end{array}\right).`$ (39)
In the orthogonal case, the determinant becomes for $`\alpha =\frac{1}{2}`$,
$`I_O=det\left(\begin{array}{ccc}1& \frac{1}{6}& \mathrm{}\\ \frac{1}{6}& \frac{1}{20}& \mathrm{}\\ \mathrm{}& \mathrm{}& \mathrm{}\end{array}\right).`$ (40)
Thus, we find the factorization of (38) as the product of (39) and (40), up to a trivial numerical factor due to the normalizations,
$$I_U=I_{Sp}\times I_O$$
(41)
The factors $`\gamma _K`$ for the unitary, symplectic and orthogonal case are related as $`2^{K^21}\gamma _K^{(U)}=\gamma _K^{(Sp)}\gamma _K^{(O)}`$, and $`\gamma _K^{(U)}=(_{l=1}^{K1}l!)^2/(_{l=1}^{2K1}l!),\gamma _K^{(Sp)}=2^{K(K+1)/2}_{l=1}^Kl!/_{l=1}^K(2l)!`$ and $`\gamma _K^{(O)}2^{K(K+1)/21}_{l=1}^{K1}l!/_{l=1}^{K1}(2l)!`$. It is again remarkable that, for arbitrary $`\alpha `$, $`\gamma _K`$ may still be expressed as the Fredholm determinant of the Laplacian, in which the eigenvalues are shifted by the amount $`\alpha `$ .
## 4 Airy kernel
When the eigenvalues lie near an edge $`\lambda _c`$ of the support of the asymptotic density of states (an edge of Wigner’s semi-circle in the Gaussian case), in a neighbourhood of size $`N^{2/3}`$ of that edge, there is a cross-over from the sine-kernel to the Airy kernel. In terms of the Airy function $`A_i(x)`$, defined by
$$A_i(x)=\frac{1}{2\pi }_{\mathrm{}}^{\mathrm{}}𝑑ze^{\frac{i}{3}z^3+izx},$$
(42)
which satisfies the differential equation
$$A_i^{\prime \prime }(x)=xA_i(x),$$
(43)
one has
$$K(x,y)=\frac{A_i(x)A_i^{}(y)A_i^{}(x)A_i(y)}{xy}.$$
(44)
In (44) we have used the scaling variables $`x`$ and $`y`$ proportional to $`N^{2/3}(\lambda \lambda _c)`$ .
There are two ways to obtain the moments under consideration. The first one is to write as before
$$I_K=<[det(\lambda _cX)]^{2K}>=\frac{du}{2\pi i}\frac{\mathrm{\Delta }(u)\mathrm{\Delta }(v)}{_{i=1}^Ku_i^Kv_i^K}\underset{i=3D1}{\overset{K}{}}K(u_i,v_i)$$
(45)
but in this case, there are three periodic structure due to three valleys of Airy functions, and the result is more complicated. It does not seem to be expressible as simple products of gamma-functions. However we can use a direct method starting with the expression
$$I_K=<[det(\lambda _cX)]^{2K}>=\frac{1}{(2\pi )^{2K}}_{\mathrm{}}^{\mathrm{}}𝑑z\mathrm{\Delta }^2(z)e^{\frac{i}{3}_{i=1}^{2K}z_i^3}$$
(46)
This representation is the edge limit $`\lambda _l0`$ of
$$F_{2K}=_{\mathrm{}}^{\mathrm{}}dz_l\frac{du_i}{2\pi i}\frac{\mathrm{\Delta }(z)\mathrm{\Delta }(u)}{_i_l(u_i\lambda _c+\lambda _l)}e^{N{\displaystyle \underset{1}{\overset{2K}{}}}\left({\displaystyle \frac{i}{3}}z_l^3+iz_lu_l\right)}$$
(47)
The sums and products over $`l`$ run from $`l=1`$ to $`l=2K`$. The dependence of $`F_{2K}`$ on $`N`$ is of order $`N^{\frac{2}{3}K^2K}`$.
We may then use the standard orthogonal polynomial method. To the complex measure
$$d\mu =dze^{\frac{i}{3}z^3}$$
(48)
we associate the orthogonal polynomials $`p_n`$ defined as
$$p_n(x)=x^n+\text{lower degree},$$
(49)
and
$$𝑑\mu p_n(x)p_m(x)=h_n\delta _{n,m}$$
(50)
The integral of (42) is then simply
$$I=K!h_0h_1\mathrm{}h_{K1}$$
(51)
Note that this looks similar to the partition function of a matrix model, but here it is the partition function of a $`K\times K`$ matrix model, instead of $`N\times N`$ ($`K`$ is finite, since it is the order of the moment that we are considering, whereas $`N`$ goes to infinity). Therefore this is for any K a completely explicit expression of the moments at the edge. Those coefficients $`h_n`$ are expressible in terms of ratios of determinants constructed with the moments of the measure :
$$h_n=\frac{d_n}{d_{n1}}$$
(52)
with
$`d_n=det\left(\begin{array}{cccc}m_0& m_1& \mathrm{}& m_n\\ m_1& m_2& \mathrm{}& m_{n+1}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ m_n& m_{n+1}& \mathrm{}& m_{2n}\end{array}\right)`$ (53)
in which the $`m_n`$ are the moments of the measure. Those determinants are constants along anti-diagonal lines (Hankel determinants). Then the product $`h_0h_1\mathrm{}h_{K1}`$ is reduced to a single determinant. For example, we have for $`K=4`$
$`h_0h_1h_2h_3=det\left(\begin{array}{cccc}C_1& iC_2& 0& iC_1\\ iC_2& 0& iC_1& 2C_2\\ 0& iC_1& 2C_2& 0\\ iC_1& 2C_2& 0& 4C_1\end{array}\right).`$ (54)
with $`C_1=A_i(0)=3^{2/3}/\mathrm{\Gamma }(2/3)`$, $`C_2=A_i^{}(0)=3^{1/3}/\mathrm{\Gamma }(1/3)`$, since all the moments up to $`m_6`$ are easily expressible in terms of $`m_0`$ and $`m_1`$ alone.
More generally we have
$$m_n=z^n𝑑\mu =(i)^n(n2)(n5)(n8)\mathrm{}\stackrel{~}{A}_n$$
(55)
where $`\stackrel{~}{A}_n=C_1`$ for n=0 (modulo 3), and $`\stackrel{~}{A}_n=C_2`$ for n=1 (modulo 3) and $`\stackrel{~}{A}_n=0`$ for n=2 (modulo 3). The last parenthesis of the product in the r.h.s. of (55) is the rest of the division of $`n2`$ by 3. Then, $`d_n`$ is the determinant of a Hankel matrix, whose matrix elements in the first row are $`[<z^0>,<z>,<z^2>,\mathrm{}]=[C_1,iC_2,0,iC_1,2C_2,0,4C_1,10iC_2,0,28iC_1,80C_2,0,\mathrm{}]`$, and all the others are given by the Hankel rule. In this way we obtain sucessively,
$`h_0`$ $`=`$ $`C_1=0.355028053`$
$`h_0h_1`$ $`=`$ $`C_2^2=0.066987483`$
$`{\displaystyle \underset{0}{\overset{2}{}}}h_l`$ $`=`$ $`2C_2^3+C_1^3=0.010074161`$
$`{\displaystyle \underset{0}{\overset{3}{}}}h_l`$ $`=`$ $`8C_1C_2^33C_1^4=0.001580882`$
$`{\displaystyle \underset{0}{\overset{4}{}}}h_l`$ $`=`$ $`72C_2^5+28C_1^3C_2^2=0.000313095517`$
$`{\displaystyle \underset{0}{\overset{5}{}}}h_l`$ $`=`$ $`2160C_2^61952C_1^3C_2^3432C_1^6=0.000090756324`$ (56)
Therefore for the edge problem we have found moments given by $`\gamma _K`$ ’s which are more complicated since $`\gamma _K=_0^{2K1}h_l`$. The result is explicit for any finite K, but we have not succeeded to continue it to non-integer K. The numerical values indicate a smooth curve in a logarithmic plot.
## 5 Finite N results
We have derived in our previous paper the correlation functions of the characteristic polynomials under the form of a determinant.
$`F_K(\lambda _1,\mathrm{},\lambda _K)={\displaystyle \underset{1}{\overset{K}{}}}det(\lambda _lX)`$
$`=`$ $`{\displaystyle \frac{1}{\mathrm{\Delta }(\lambda _1,\mathrm{},\lambda _K)}}det\left|\begin{array}{cccc}p_M(\lambda _1)& p_{M+1}(\lambda _1)\hfill & \mathrm{}& \hfill p_{M+K1}(\lambda _1)\\ p_M(\lambda _2)& p_{M+1}(\lambda _2)\hfill & \mathrm{}& \hfill p_{M+K1}(\lambda _2)\\ \mathrm{}& & & \\ p_M(\lambda _K)& p_{M+1}(\lambda _K)\hfill & \mathrm{}& \hfill p_{M+K1}(\lambda _K)\end{array}\right|,`$ (61)
in which $`X`$ is an $`M\times M`$ random matrix.
The polynomial $`p_n(x)`$ are the (monic) orthogonal polynomials, whose coefficients of highest degree are equal to unity
$$p_n(x)=x^n+\mathrm{lowerdegree}.$$
(62)
If we are concerned simply with the moments of the distribution of a single characteristic polynomial, we obtain from (5)
$`\mu _K(\lambda )=`$ $`F_K(\lambda ,\mathrm{},\lambda )=[det(\lambda X)]^K`$
$`=`$ $`{\displaystyle \frac{(1)^{K(K1)/2}}{_{l=0}^{K1}(l!)}}det\left|\begin{array}{cccc}p_M(\lambda )& p_{M+1}(\lambda )\hfill & \mathrm{}& \hfill p_{M+K1}(\lambda )\\ p_M^{}(\lambda )& p_{M+1}^{}(\lambda )\hfill & \mathrm{}& \hfill p_{M+K1}^{}(\lambda )\\ \mathrm{}& & & \\ p_M^{(K1)}(\lambda )& p_{M+1}^{(K1)}(\lambda )\hfill & \mathrm{}& \hfill p_{M+K1}^{(K1)}(\lambda )\end{array}\right|.`$ (67)
For the Gaussian distribution,
$$P(X)=\frac{1}{Z_M}\mathrm{exp}\frac{N}{2}\mathrm{TrX}^2,$$
(68)
with
$$M=NK,$$
(69)
the polynomial $`p_n(x)`$ are the Hermite polynomials $`H_n(x)`$, defined with our normalization as
$$H_n(x)=\frac{(1)^n}{N^n}e^{Nx^2/2}(\frac{d}{dx})^ne^{Nx^2/2}=x^n+\mathrm{l}.\mathrm{d}..$$
(70)
The integral representation
$$H_n(x)=\frac{(1)^nn!}{N^n}\frac{dz}{2i\pi }\frac{e^{N(z^2/2+xz)}}{z^{(n+1)}}$$
(71)
over a contour which circles around the origin in the z-plane, turns out to be well adapted.
Note that all these expressions are all valid for finite N. We may thereby recover readily several results that we have discussed in the previous sections. For instance, let us assume that $`M`$ is an even number, and consider the center value $`\lambda =0`$ (since the dependence in $`\lambda `$ is known to be contained entirely in the overall factor $`[\rho (\lambda )]^{K^2}`$, as far as the coefficient $`\gamma _K`$ is concerned, it is sufficient to put simply $`\lambda =0`$).
The Hermite polynomials $`H_n(x)`$ vanish for odd n at $`x=0`$. Similarly the odd derivatives of $`H_n(x)`$ for even n, also vanish at $`x=0`$. Hence, the elements of the determinant (5) are alternatively non-zero then zero. Thus the determinant is decomposed into a product of two determinants ; this is the exact phenomenon for N finite of the factorization of the symplectic and orthogonal determinants that we have seen earlier for large N. Since the matrix elements of (5) at $`\lambda =0`$ are all expressed as derivatives of Hermite polynomials at the origin, it is possible to compute this determinant exactly for finite and arbitrary M and K . For the even M case,
$`F_{2K}(0)`$ $`=`$ $`{\displaystyle \frac{(1)^{K(2K1)}}{_{l=3D0}^{2K1}(l!)}}det\left|\begin{array}{cccc}H_M(0)& H_{M+2}(0)\hfill & \mathrm{}& \\ H_M^{\prime \prime }(0)& H_{M+2}^{\prime \prime }(0)\hfill & \mathrm{}& \\ \mathrm{}& & & \\ H_M^{(2K2)}(0)& H_{M+2}^{(2K2)}(0)\hfill & \mathrm{}& \end{array}\right|`$ (76)
$`\times `$ $`det\left|\begin{array}{cccc}H_{M+1}^{}(0)& H_{M+3}^{}(0)\hfill & \mathrm{}& \\ H_{M+1}^{\prime \prime \prime }(0)& H_{M+3}^{\prime \prime \prime }(0)\hfill & \mathrm{}& \\ \mathrm{}& & & \\ H_{M+1}^{(2K1)}(0)& H_{M+3}^{(2K1)}(0)\hfill & \mathrm{}& \end{array}\right|`$ (81)
We denote each determinant as $`I^{(1)}/N^{KM/2}`$ and $`I^{(2)}/N^{KM/2}`$ respectively, and
$$F_{2K}(0)=I^{(1)}I^{(2)}\frac{1}{N^{KM}}\frac{1}{_{l=0}^{2K1}l!}.$$
(82)
The above two determinants are easily computed through the explicit expressions for the $`H_n(x)`$’s,
$$H_{2n}(x)=\frac{1}{n}(1)^n(2n1)!!\underset{m=0}{\overset{\mathrm{}}{}}\frac{(n)(n+1)\mathrm{}(n+m1)}{(\frac{1}{2})(\frac{1}{2}+1)\mathrm{}(\frac{1}{2}+m1)}\frac{1}{m!}(\frac{Nx^2}{2})^m,$$
(83)
$$H_{2n+1}(x)=\frac{1}{N^n}(1)^n(2n+1)!!x\underset{m=0}{\overset{\mathrm{}}{}}\frac{(n)(n+1)\mathrm{}(n+m1)}{(\frac{3}{2})(\frac{3}{2}+1)\mathrm{}(\frac{3}{2}+m1)}\frac{1}{m!}(\frac{Nx^2}{2})^m.$$
(84)
The two determinants contain overall products of factors of the form $`(2n1)!!`$ ; once they are extracted one finds
$`I^{(1)}`$ $`=`$ $`C(M+2K3)!!(M+2K5)!!\mathrm{}(M1)!!`$ (85)
$`=`$ $`C{\displaystyle \frac{1}{2^{\frac{(K(M+K3)}{2}}}}{\displaystyle \underset{l=1}{\overset{K}{}}}[{\displaystyle \frac{\mathrm{\Gamma }(M+2l2)}{\mathrm{\Gamma }(\frac{M}{2}+l1)}}]`$
$`I^{(2)}`$ $`=`$ $`C(M+2K1)!!(M+2K3)!!\mathrm{}(M+1)!!`$ (86)
$`=`$ $`C{\displaystyle \frac{1}{2^{\frac{K(M+K1)}{2}}}}{\displaystyle \underset{l=1}{\overset{K}{}}}[{\displaystyle \frac{\mathrm{\Gamma }(M+2l)}{\mathrm{\Gamma }(\frac{M}{2}+l)}}]`$
with
$$C=2^{\frac{K(K1)}{2}}\underset{l=0}{\overset{K1}{}}l!$$
(87)
which is independent of $`M`$. In the large M limit, from the Stirling formula, we have
$$I^{(1)}CM^{\frac{MK+K(K1)}{2}}e^{\frac{MK}{2}}2^{\frac{K}{2}}$$
(88)
$$I^{(2)}CM^{\frac{MK+K(K+1)}{2}}e^{\frac{MK}{2}}2^{\frac{K}{2}}$$
(89)
It is remarkable that, even for finite M (M is the size of the random matrix), $`F_{2K}(0)`$ for this Gaussian distribution, already exhibits the factor $`\gamma _K=_{l=0}^{K1}l!/((K+l)!=[_{l=0}^{K1}l!]^2/_{l=0}^{2K1}l!`$, which is known to be universal in the large M-limit. It is indeed obtained from the product of the factor $`C`$ and $`1/(_{l=0}^{2K1}l!)`$ in (82). This means that at each order of the $`1/N`$ expansion, we keep this universal factor for $`F_{2K}(0)`$. In the large N limit (M = N $``$ K), $`F_{2K}(0)`$ becomes
$$F_{2K}(0)(2N)^{K^2}e^{NK}\underset{0}{\overset{K1}{}}\frac{l!}{(K+l)!}.$$
(90)
In the previous paper, we have derived $`F_{2K}(\lambda )`$, in the large N limit, as
$$F_{2K}(\lambda ,\mathrm{},\lambda )(2\pi \rho (\lambda )N)^{K^2}e^{NK}\underset{0}{\overset{K1}{}}\frac{l!}{(K+l)!}$$
(91)
At the band center, $`\lambda =0`$, the density of state is $`\rho (0)=\frac{1}{\pi }`$ for the Gaussian distribution. Therefore, (90) is indeed consistent with (91).
It may be interesting to note that one of the factors of (91), namely $`_{l=0}^{2K1}(l!)`$, appears in $`F_{2K}(\lambda )`$ in (5) . This factor, a product of gamma-functions, remains for any set of orthogonal polynomials, since it stands in the front of the determinant of (5).
The factor $`e^{NK}`$ is cancelled by the normalization . For $`\lambda 0`$, we have evaluated $`F_{2K}(\lambda )`$. We have here considered the finite N case to see the universal factor $`\gamma _K`$.
One can recover again the Airy limit by the use of eq.(5). We use once more the properties of the Hermite polynomials such as
$$H_n^{}(x)=nH(x)$$
(92)
and their explicit integral representation
$$H_n(x)=\frac{1}{\sqrt{2\pi }}N^{\frac{1}{2}}e^{\frac{N}{2}x^2}_{\mathrm{}}^{\mathrm{}}𝑑ss^ne^{\frac{N}{2}s^2iNxs}.$$
(93)
We set $`n=\delta +N`$, and after exponentiation, we have
$$H_n(x)=\frac{1}{\sqrt{2\pi }}N^{\frac{1}{2}}e^{\frac{N}{2}x^2}_{\mathrm{}}^{\mathrm{}}𝑑ss^\delta e^{Nf(s)}$$
(94)
where $`f(s)=\frac{1}{2}s^2+isx\mathrm{log}s`$. The saddle points are degenerate at the edge $`x=2`$. The vicinity of this point is blown out through a change of variables, with a scaling ansatz,
$$x=2+N^\alpha y$$
(95)
and
$$s=i+N^\beta z$$
(96)
If one expands $`f(s)`$ up to order $`z^3`$, one sees that in the proper scaling choice $`\alpha =2/3`$ and $`\beta =1/3`$, one recovers the Airy limit which governs the properties of the system in a neighbourhood of size $`N^{2/3}`$ of the edge of Wigner’s semi-circle. Then, the integral becomes
$$I=(i)^\delta N^{\frac{1}{3}}_{\mathrm{}}^{\mathrm{}}𝑑ye^{\frac{i}{3}y^3+izy}$$
(97)
This is indeed the Airy function $`A_i(z)`$ of (42).
$$H_{N+\delta }(x)=\sqrt{2\pi N}e^{2N}(i)^\delta A_i((x2)N^{\frac{2}{3}})$$
(98)
We now consider all the $`\lambda _i=2`$, and the determinant (5) becomes in the large N limit a determinant of Airy functions. If we replace $`H_{M+2K1}`$ at the right-up corner of the determinant by the Airy function $`A_i(0)`$, the other matrix elements become derivatives of the Airy function, since there is a the recursion relation (92). For example,in the $`K=1`$ case, we have
$$det\left|\begin{array}{cccc}H_M(2)& H_{M+1}(2)\hfill & & \\ H_M^{}(2)& H_{M+1}^{}(2)\hfill & & \end{array}\right|det\left|\begin{array}{cccc}\frac{N^{\frac{2}{3}}}{M+1}A_i^{}(0)& A_i(0)\hfill & & \\ \frac{N^{\frac{4}{3}}}{M+1}A_i^{\prime \prime }(0)& N^{\frac{2}{3}}A_i^{}(0)\hfill & & \end{array}\right|$$
(99)
Then, we find in the large N limit, with $`N=MK`$,
$$F_{2K}(2)=\frac{N^{\frac{2}{3}K(K+1)}}{_{l=0}^{2K1}l!}det\left|\begin{array}{cccc}\mathrm{}& A_i^{}(0)\hfill & A_i(0)& \\ \mathrm{}& A_i^{\prime \prime }(0)\hfill & A_i^{}(0)& \\ \mathrm{}& \mathrm{}\hfill & \mathrm{}& \end{array}\right|$$
(100)
The above determinant was discussed earlier. Note the factor $`1/_{l=0}^{2K1}l!`$ in front.
## 6 Derivative moments
The same techniques may also be used if one is interested in the moments of the D-th derivatives (D = 1,2,…) of the characteristic polynomials. Let us consider for instance
$$F_{2K}^{(D)}(\lambda _1,\mathrm{},\lambda _{2K})=<\underset{l=3D1}{\overset{2K}{}}\frac{^D}{\lambda _i^D}det(\lambda _iX)>$$
(101)
From (5), one sees immediately that it has also the form of a determinant :
$`F_{2K}^{(D)}(\lambda _1,\mathrm{},\lambda _{2K})={\displaystyle \frac{1}{\mathrm{\Delta }(\lambda _1,\mathrm{},\lambda _{2K})}}det\left|\begin{array}{cccc}p_M^{(D)}(\lambda _1)& p_{M+1}^{(D)}(\lambda _1)\hfill & \mathrm{}& \hfill p_{M+2K1}^{(D)}(\lambda _1)\\ p_M(\lambda _2)^{(D)}& p_{M+1}^{(D)}(\lambda _2)\hfill & \mathrm{}& \hfill p_{M+2K1}^{(D)}(\lambda _2)\\ \mathrm{}& & & \\ p_M^{(D)}(\lambda _{2K})& p_{M+1}^{(D)}(\lambda _{2K})\hfill & \mathrm{}& \hfill p_{M+2K1}^{(D)}(\lambda _{2K})\end{array}\right|.`$ (106)
When all the $`\lambda _i`$’s are equal, we have
$`F_{2K}^{(D)}(\lambda ,\mathrm{},\lambda )`$ $`=`$ $`[{\displaystyle \frac{d^D}{d\lambda ^D}}det(\lambda X)]^{2K}`$ (112)
$`=`$ $`{\displaystyle \frac{(1)^{K(2K1)}}{_{l=0}^{2K1}(l!)}}det\left|\begin{array}{cccc}p_M^{(D)}(\lambda )& p_{M+1}^{(D)}(\lambda )\hfill & \mathrm{}& \hfill p_{M+2K1}^{(D)}(\lambda )\\ p_M^{(D+1)}(\lambda )& p_{M+1}^{(D+1)}(\lambda )\hfill & \mathrm{}& \hfill p_{M+2K1}^{(D+1)}(\lambda )\\ \mathrm{}& & & \\ p_M^{(D+2K1)}(\lambda )& p_{M+1}^{(D+2K1)}(\lambda )\hfill & \mathrm{}& \hfill p_{M+2K1}^{(D+2K1)}(\lambda )\end{array}\right|.`$
If we set $`\lambda =0`$ it may be again decomposed into a product of two determinants . Let us assume for definiteness that both $`M`$ and $`D`$ are even. Then, we have
$$I^{(1)}=det\left|\begin{array}{cccc}H_M^{(D)}(0)& H_{M+2}^{(D)}(0)\hfill & \mathrm{}& \\ H_M^{(D+2)}(0)& H_{M+2}^{(D+2)}(0)\hfill & \mathrm{}& \\ \mathrm{}& & & \\ H_M^{(D+2K2)}(0)& H_{M+2}^{(D+2K2)}(0)\hfill & \mathrm{}& \end{array}\right|$$
(114)
$$I^{(2)}=det\left|\begin{array}{cccc}H_{M+1}^{(D+1)}(0)& H_{M+3}^{(D+1)}(0)\hfill & \mathrm{}& \\ H_{M+1}^{(D+3)}(0)& H_{M+3}^{(D+3)}(0)\hfill & \mathrm{}& \\ \mathrm{}& & & \\ H_{M+1}^{(D+2K1)}(0)& H_{M+3}^{(D+2K1)}(0)\hfill & \mathrm{}& \end{array}\right|$$
(115)
Using the explicit expressions for the Hermite polynomials , we can compute these determinants. We find for arbitrary M,D and K,
$$F_{2K}^{(D)}(0)=\frac{1}{N^{K(MD)}}I^{(1)}I^{(2)}\frac{1}{_{l=0}^{2K1}l!}$$
(116)
$`I^{(1)}`$ $`=`$ $`(M+2K3)!!(M+2K5)!!\mathrm{}(M1)!!`$ (117)
$`\times `$ $`{\displaystyle \underset{l=0}{\overset{K1}{}}}[({\displaystyle \frac{M}{2}}+l)({\displaystyle \frac{M}{2}}+l1)\mathrm{}({\displaystyle \frac{M}{2}}{\displaystyle \frac{D}{2}}+l+1)]`$
$`\times `$ $`2^{\frac{DK+K(K1)}{2}}{\displaystyle \underset{l=0}{\overset{K1}{}}}l!`$
$`I^{(2)}`$ $`=`$ $`(M+2K1)!!(M+2K3)!!\mathrm{}(M+1)!!`$ (118)
$`\times `$ $`{\displaystyle \underset{l=0}{\overset{K1}{}}}[({\displaystyle \frac{M}{2}}+l+1)({\displaystyle \frac{M}{2}}+l)\mathrm{}({\displaystyle \frac{M}{2}}{\displaystyle \frac{D}{2}}+l+2)]`$
$`\times `$ $`2^{\frac{DK+K(K1)}{2}}{\displaystyle \underset{l=0}{\overset{K1}{}}}l!`$
(One may easily check these results for D = M, since the matrix elements below the diagonal vanish, i.e. the determinants are then simply given by the product of the diagonal elements, $`_{l=0}^K(M+2l)!`$ which agrees with (117). When $`D=0`$, it reduces to the previous expression (85). $`I^{(2)}`$ is obtained from $`I^{(1)}`$ by the shift $`MM+2`$.)
In the large N limit, we have
$$F_{2K}^{(D)}(0)(2N)^{K^2+2KD}e^{KN}\frac{1}{2^{2KD}}\underset{l=0}{\overset{K1}{}}\frac{l!}{(K+l)!}$$
(119)
Hence, for this derivative moments at finite M, again the universal factor $`\gamma _K`$ is present, and it persists of course in the large N limit.
These results lead to the conjecture that the average of the moment of derivatives of the Riemann zeta-function along the critical line
$$I=\frac{1}{T}_0^T𝑑t|\frac{d^D}{dt^D}\zeta (\frac{1}{2}+it)|^{2K},$$
(120)
also have this universal factor $`\gamma _K`$.
## 7 External source
We now consider the case in which the external source matrix $`A`$ is coupled to the random matrix $`X`$. The measure of the random matrix $`X`$ is
$$d\mu (X)=\frac{1}{Z}e^{\frac{N}{2}\mathrm{TrX}^2+\mathrm{NTrXA}}d^{N^2}X$$
(121)
The eigenvalues of the matrix $`A`$ are denoted by $`a_i`$, $`i=1,\mathrm{},N`$. In such cases, the standard orthogonal polynomial method cannot be used. However, the n-point correlation functions $`\rho (\lambda _1,\mathrm{},\lambda _n)`$ have been found to be described again by the determinant of a kernel ; from there the level spacing probability $`p(s)`$ has been also investigated . If we specialize to a source which has two opposite eigenvalues, namely $`a_i=+a`$ for $`i=1,\mathrm{},N/2`$ and $`a_i=a`$ for $`i=N/2+1,\mathrm{},N`$, one finds a support for the eigenvalues made of two disconnected segments for $`a>1`$. If one tunes the external source so that $`a=1`$, i.e. $`a_i=\pm 1`$, the gap between the two segments closes and the spectrum consists of a single segment for $`a<1`$. We want to investigate here the critical point $`a=1`$ which gives rise to yet another class of universality.
The moments $`F_{2K}(\lambda ,\mathrm{},\lambda )`$ at $`\lambda =0`$ at this closing gap point may turn out to have interesting applications.
Since $`X`$ and $`A`$ are Hermitian matrices, we write
$$\mathrm{TrXA}=\mathrm{TrU}^1\mathrm{X}_0\mathrm{UA}_0$$
(122)
where $`X_0=diag(x_1,\mathrm{},x_M),A_0=diag(a_1,\mathrm{},a_M)`$, and $`U`$ belongs to the unitary group. The integration over this unitary group $`U`$ is well known from the work of Harish-Chandra, Itzykson-Zuber , and this is the starting point of the formulae found in . For instance the n-point correlation functions are given by the determinant of the $`n\times n`$ matrices made with the kernel $`K(\lambda _i,\lambda _j)`$ with
$$K(\lambda ,\mu )=_{\mathrm{}}^{\mathrm{}}\frac{dt}{2\pi }\frac{du}{2\pi i}\underset{l=1}{\overset{N}{}}\frac{a_lit}{ua_l}\frac{1}{uit}e^{\frac{N}{2}t^2+Nit\lambda \frac{N}{2}u^2+Nu\mu +\frac{N}{4}\lambda ^2\frac{N}{4}\mu ^2}$$
(123)
where the contour encloses all the $`a_l`$’s.
However we may proceed without that here and compute the correlation functions of the characteristic polynomials directly. Indeed
$`F_K(\lambda _1,\mathrm{},\lambda _K)`$ $`=`$ $`<{\displaystyle \underset{\alpha =1}{\overset{K}{}}}det(\lambda _\alpha X)>`$ (124)
$`=`$ $`{\displaystyle \frac{1}{Z}}{\displaystyle 𝑑X\underset{\alpha =1}{\overset{K}{}}det(\lambda _\alpha X)e^{\frac{N}{2}\mathrm{TrX}^2+\mathrm{NTrXA}}}`$
In the above equation, the random matrix $`X`$ is assumed to be an $`M\times M`$ matrix, with $`M=NK`$, as before. When $`K=1`$, this gives a polynomial, which was investigated before .
The explicit integration over the unitary group , leads to
$$F_K(\lambda _1,\mathrm{},\lambda _K)=\underset{i=1}{\overset{M}{}}dx_i\frac{\mathrm{\Delta }(x_1,\mathrm{},x_M;\lambda _1,\mathrm{},\lambda _K)}{\mathrm{\Delta }(a)\mathrm{\Delta }(\lambda )}e^{\frac{N}{2}_{i=1}^Mx_i^2+N_{i=1}^Mx_ia_i}$$
(125)
where $`\mathrm{\Delta }(x_1,\mathrm{},x_M;\lambda _1,\mathrm{},\lambda _K)`$ is the Van der Monde determinant $`(M+K)\times (M+K)`$ made with the $`x`$’s and the $`\lambda `$’s. This determinant may be replaced by a determinant of (monic) polynomials , and we choose the Hermite polynomials defined in (70). It is then straightforward to verify that
$$_{\mathrm{}}^{\mathrm{}}H_n(x)e^{\frac{N}{2}x^2+Nax}𝑑x=a^ne^{\frac{N}{2}a^2}\sqrt{\frac{2\pi }{N}}$$
(126)
Therefore we can explicitely integrate over the M variables $`x_i`$’s in (125) and one obtains
$`F_K(\lambda _1,\mathrm{},\lambda _K)={\displaystyle \frac{1}{\mathrm{\Delta }(a)\mathrm{\Delta }(\lambda )}}`$
$`\times det\left|\begin{array}{cccccc}1& \mathrm{}& 1& H_0(\lambda _1)& \mathrm{}& H_0(\lambda _K)\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ a_1^{M+K1}& \mathrm{}& a_M^{M+K1}& H_{M+K1}(\lambda _1)& \mathrm{}& H_{M+K1}(\lambda _K)\end{array}\right|.`$ (127)
Let us first check that in the limit of a vanishing source in which all the $`a_i0`$, we do recover the previous formula (5). The column which depends upon $`a_i`$ is expanded in Taylor series around $`a_1`$, and subtracting the successive columns, we obtain, after factoring the Van der Monde determinant $`\mathrm{\Delta }(a)`$ which cancels the denominator, a vanishing upper triangle (up to the M-th column) , ones on the diagonal and powers of the $`a_i`$’s below the diagonal. We can now let the $`a_i`$’s go to zero and we are left with the $`K\times K`$ of (5). (In we gave a different derivation of this same formula).
If we return to an arbitrary non vanishing external source, we may proceed by returning to (7) and define $`G_K(b_1,\mathrm{},b_K)`$,
$`G_K(b_1,\mathrm{},b_K)`$ $`=`$ $`{\displaystyle F_K(\lambda _1,\mathrm{},\lambda _K)\mathrm{\Delta }(\lambda )e^{\frac{N}{2}{\scriptscriptstyle \lambda _l^2}+N{\scriptscriptstyle \lambda _lb_l}}d\lambda _i}`$ (128)
$`=`$ $`{\displaystyle \frac{\mathrm{\Delta }(a;b)}{\mathrm{\Delta }(a)}}e^{\frac{N}{2}{\scriptscriptstyle b_l^2}}`$
We may now recover $`F_K`$ by taking the Fourier transform of $`G_K(ib_1,\mathrm{},ib_K)`$,
$$G_K(ib_1,\mathrm{},ib_K)e^{iN{\scriptscriptstyle \lambda _ib_i}}\underset{i=1}{\overset{K}{}}\frac{db_i}{2\pi }=(\frac{1}{N})^K\mathrm{\Delta }(\lambda )F_K(\lambda _1,\mathrm{},\lambda _K)e^{\frac{N}{2}{\scriptscriptstyle \lambda _l^2}}$$
(129)
Therefore, we obtain the following explicit formula,
$`F_K(\lambda _1,\mathrm{},\lambda _K)`$ $`=`$ $`{\displaystyle \frac{N^K}{\mathrm{\Delta }(\lambda )}}e^{\frac{N}{2}{\scriptscriptstyle \lambda _l^2}}{\displaystyle \frac{1}{K!}}`$
$`\times `$ $`{\displaystyle \underset{i=1}{\overset{K}{}}\frac{db_i}{2\pi }\underset{j=1}{\overset{M}{}}(ib_la_j)\underset{l<l^{}}{\overset{K}{}}(ib_lib_l^{})e^{\frac{N}{2}{\scriptscriptstyle b_l^2}}det(e^{iN\lambda _lb_l^{}})}`$
Note that we could replace in the integrand of (7) $`det(e^{iN\lambda _lb_l^{}})`$ by the diagonal term $`e^{iN_1^K\lambda _lb_l}`$ and cancel the $`K!`$ in the denominator. Again we can examine the limit of this formula when the external source goes to zero, and putting all $`\lambda _i=\lambda `$, we obtain
$$F_{2K}(\lambda )=\frac{N^{K(2K+1)}}{_{l=0}^{2K1}l!}\frac{1}{(2K)!}e^{KN\lambda ^2}\underset{l=1}{\overset{2K}{}}b_l^M\mathrm{\Delta }^2(b)e^{\frac{N}{2}{\scriptscriptstyle b_l^2}iN\lambda {\scriptscriptstyle b_l}}\underset{l=1}{\overset{2K}{}}\frac{db_l}{2\pi },$$
(131)
(we have considered $`F_{2K}`$ instead of $`F_K`$ in order to compare with our previous results). In the large N limit, we exponentiate $`b_l^M`$, ($`M=NK`$), and look for the saddle points which are the roots of the equation $`b^2+i\lambda b1=0`$ ; let us call the two roots $`b^+`$ and $`b^{}`$. The difference $`|b^+b^{}|=2\pi \rho (\lambda )`$. The leading saddle-point for the $`b_l`$’s,$`(l=1,\mathrm{},2K),`$ is obtained by choosing half of them equal to $`b^+`$, and $`b^{}`$ for the another half. The following Gaussian integral with a Van der Monde determinant,
$$\frac{1}{K!}\underset{i=1}{\overset{K}{}}db_ie^{\frac{N}{2}f^{^{\prime \prime }}b^2}\underset{i<j}{\overset{K}{}}(b_ib_j)^2=(\frac{2\pi }{Nf^{^{\prime \prime }}})^{\frac{K}{2}}\frac{_{l=0}^{K1}l!}{(Nf^{^{\prime \prime }})^{\frac{K(K1)}{2}}},$$
(132)
where $`f^{^{\prime \prime }}`$ is the second derivative of $`f`$ at the saddle-point, allows us to complete the calculation. Integrating then around the saddle-points $`b^+`$ and $`b^{}`$, and keeping in mind the combinatorial factor $`\frac{(2K)!}{K!K!}`$, which is the number of choices of K $`b^+`$ and K $`b^{}`$ among the $`2K`$ $`b_l`$’s, we recover precisely our previous result,
$$e^{NKV(\lambda )}F_{2K}(\lambda )=(2\pi N\rho (\lambda ))^{K^2}e^{NK}\gamma _K$$
(133)
where $`\gamma _K=(_{l=0}^{K1}l!)^2/(_{l=0}^{2K1}l!)=(_{l=0}^{K1}l!)/_{l=0}^{K1}(K+l)!`$, and $`V(\lambda )=\frac{\lambda ^2}{2}`$.
From the expression (7), it is also easy to obtain the moments at the critical point corresponding to the closure of the gap :
$$F_K(0)=\frac{N^K}{K!}e^{\frac{NM}{2}}\underset{l=1}{\overset{K}{}}\frac{db_l}{2\pi }(1+b_l^2)^{\frac{M}{2}}\mathrm{\Delta }^2(b)e^{\frac{N}{2}_{l=1}^Kb_l^2}$$
(134)
Note that this expression is exact for finite N. In the large N limit, we exponentiate the logarithmic term and expand the exponent about $`b_l`$ up to order $`b_l^4`$ term. The critical point, is precisely the point at which the coeffeicient of the quadratic term $`b_l^2`$ vanishes. We then have
$$F_K(0)=\frac{N^K}{K!}e^{\frac{NM}{2}}\frac{db_l}{2\pi }e^{\frac{N}{2}_{l=1}^Kb_l^4}\mathrm{\Delta }^2(b)$$
(135)
As in all the cases which appeared in the previous sections, this integral is expressed by a Hankel determinant, in which the matrix elements are $`\mathrm{\Gamma }(\frac{2n1}{4})`$. The determinant is
$$I=det\left|\begin{array}{ccccc}\mathrm{\Gamma }(\frac{1}{4})& 0& \mathrm{\Gamma }(\frac{3}{4})& 0& \mathrm{}\\ 0& \mathrm{\Gamma }(\frac{3}{4})& 0& \mathrm{\Gamma }(\frac{5}{4})& \mathrm{}\\ \mathrm{\Gamma }(\frac{3}{4})& 0& \mathrm{\Gamma }(\frac{5}{4})& 0& \mathrm{}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\end{array}\right|.$$
(136)
Note that we have considered the case $`a_l=\pm 1`$ case, but the formulae are explicit for any spectrum of the source and they could be easily used to study for instance multi-critical situations which were discussed in .
Acknowledgement
We thank Zeev Rudnick and John Keating for useful discussions. This work has been supported by the CREST programme of JST.
|
warning/0004/astro-ph0004406.html
|
ar5iv
|
text
|
# Cataclysmic Variables as Sources of Gravitational Waves
## 1 Introduction
Detection of gravitational radiation from astrophysical sources will mark a breakthrough in the history of astronomy (see, e.g., Thorne thor87 (1987) and Schutz schu96 (1996)). Experimental efforts to search for these space-time wrinkles have been under development for the past twenty years (Thorne thor95 (1995, 1996)). With the advent of technological improvements in several crucial aspects of the detection process we will soon be ready to turn them a physical reality (Schutz schu96 (1996), Thorne thor95 (1995), Finn & Chernoff finn93 (1993)).
In particular, the Laser Interferometric Space Antenna (LISA) is designed to detect low frequency gravitational waves in the frequency range $`10^41`$ Hz, which are not possible to detect on the Earth because of seismic noise. There is a lot of very interesting astrophysical phenomena which are believed to generate GWs in the frequency band detectable by LISA, namely: formation of supermassive black holes (SMBHs), SMBH-SMBH binary coalescence, compact stars orbiting around SMBHs (in, e.g., galactic nuclei), a wide variety of binaries, such as pairs of close white dwarfs (WDs), pairs of neutron stars, neutron star and black hole binaries, pairs of contacting normal stars, normal stars and white dwarfs (cataclysmic) binaries, and pairs of stellar black holes.
Due to the fact the GWs are produced by a large variety of astrophysical sources and cosmological phenomena it is quite probable that the Universe is pervaded by a background of such waves. Binary stars of a variety of stars (ordinary, compact or combinations of them), Population III stars, phase transitions in the early Universe, cosmic strings are examples of sources able to produce a background of GWs.
As the GWs possess a very weak interaction with matter passing through it unharmed, relic radiation (spectral properties for example) once detected can provide information on the physical conditions from the era in which they were produced. In principle it will be possible, for example, to get information from the epoch when the galaxies and stars started to form and evolve.
Concerning our galaxy, it presents a large number of binary systems, which produce a GW background named binary confusion noise (see Hils, Bender & Webbink hils90 (1990), Bender & Hils bend97 (1997)). Some of the galactic binary sources are: close white dwarfs binaries (CWDBs), neutron star binaries (NSBs), unevolved binaries, WUMs binaries and cataclysmic binaries.
The binary systems are the most understood of all sources of GWs (see, e.g., Thorne 1987). Knowing the masses of the stars, the orbital parameters and their estimated distances, one can calculate the details of the GW produced.
The LISA’s sensitivity as well as the binary confusion noise will determine in the end if one is able to discriminate the signal of a particular astrophysical source.
The first papers concerning the gravitational radiation from binaries systems was written by Mironowskii (1966), who studied in particular the W UMa stars, and by Forward & Berman (1967), approximately 30 years ago. After that many other studies concerning the evaluation of GWs background produced by various types of binary stars in the Galaxy followed (see, e.g., Douglass & Braginsky 1979, Lipunov & Postnov 1987, Lipunov, Postnov & Prokhorov 1987, Evans, Iben & Smarr 1987, Hils, Bender & Webbink 1990, Bender & Hils 1997, Webbink & Han 1998, Hils 1998)
Here we are particularly interested in the cataclysmic variable binaries as sources of GWs, such a system is formed by a white dwarf and a low mass secondary star. The total number of such a kind of binary is estimated to amount $`10^6`$ in the Galaxy (see, e.g. Hils, Bender & Webbink 1990). These systems produce low frequency GWs, namely, $`f_{gw}<10^3`$, which could be detected by LISA.
We are not concerned here with the calculation of a confusion noise produced by such binaries, our aim is similar to the study by Douglass & Braginsky (1979) who evaluate the dimensionless amplitude h for a series of specific low frequency GW binaries. Based mainly on the 6th edition of the catalogue of cataclysmic binaries, low mass X-ray binaries and related objects (Ritter & Kolb 1998) we have catalogued almost 160 CV systems for which it is possible to evaluate the GW amplitude. We have catalogued firstly those CVs with known distances, orbital period and masses, quantities necessary to evaluate the GW amplitude produced by such objects; secondly we have catalogued those systems for which the distances and the orbital periods are known, the masses being obtained from a mass-period relationships.
The remainder of the paper is as follows: Section 2 deals with the cataclysmic variables. Section 3 addresses the gravitational waves from cataclysmic variables. The discussion and conclusions are summarized in Section 4.
## 2 The Cataclysmic Variables
A Cataclysmic Variable (CV) is a semi-detached binary system of low mass and very short orbital period. The primary star is an accreting degenerate white dwarf and the secondary one is usually, but not always, a late-type star that fills its critical Roche lobe and transfers matter to the companion. There are 1020 cataclysmic variables classified (Downes, Webbink & Shara, 1997) and more than 300 of them have known periods (Ritter & Kolb, 1998, hereafter RK98). From a period histogram Patterson (1998, his Figure 1) shows, with data taken from RK98 (see also Kolb, King and Ritter, 1998, figure 4, to orbital periods below 5 hours), that the majority of these systems have periods ranging from 1.2 to 15.0 h.
We have catalogued, in Tables 1 and 2, 156 CV systems. In Table 1 we have catalogued 68 CVs, where in column 1 we present their names, in column 2 the distances in parsecs, in column 3 the periods in days, in column 4 the primary mass in solar masses, in column 5 the secondary mass in solar masses, in column 6 the gravitational wave amplitude $`h`$ (see section 3 for its calculation), and finally in column 7 we present the references used to obtain the data of each CV system. In Table 2 we have catalogued 88 CVs for which only the distances and periods are known; the label of the columns are the same as in Table 1.
For the systems with orbital periods of up to 10 hours it is possible to make use of a mass function to compute the masses of the secondary stars. We have computed the mass of the secondary star using an equation obtained by Smith and Dhillon (1998, hereafter SD98). Their mass-period relationship has the following best fit (equation 9 of SD98):
$$M_2/M_{}=0.126P0.11,\mathrm{with}\mathrm{period}\mathrm{in}\mathrm{hours}.$$
(1)
To calculate the mass of the primary star we have used the unweighted average for all systems (see, Table 4a of SD98):
$`M_1=0.69M_{}\mathrm{below}\mathrm{period}\mathrm{gap}`$
$`M_1=0.80M_{}\mathrm{above}\mathrm{period}\mathrm{gap}`$ (2)
The period gap, namely, $`2<P<3`$ hours, a failure in the distribution of cataclysmic variables, has been discussed in the literature recently by, for example, Clemens et al.(1998) and Kolb, King and Ritter (1998). For stars in the gap period we have considered the mean value $`M_1=0.74M_{}`$ .
It is worth noting that Equations 1 and 2 (SD98) were obtained from 14 reliable CV mass determination. In our sample there are 68 CVs with known masses, whose values were obtained by various methods. A fit with 62 CVs gives a relationship consistent with SD98. Five do not fit the $`M_2`$ $`\times `$ orbital period distribution, namely: AE Aqr, OY Car, BV Cen, GK Per, V Sge. Our fit is given by:
$`M_2/M_{}=(0.121\pm 0.004)P0.070\pm 0.020,`$ (3)
$`\mathrm{with}\mathrm{period}\mathrm{in}\mathrm{hours}.`$
In our catalogue 9 systems have periods above 9 hours, namely: QU Car, V394 CrA, V841 Oph, TY PsC, VV Pup, U Sco, MR Ser, NA UMa and SU UMa. From RK98 we have obtained the spectral type only for VV Pup (M4-5), U Sco (F6-G0-5) and MR Ser (M5-6/5). The secondary mass is then obtained from the spectral type versus $`M_2`$ diagram of Kolb & Baraffe (1999). For VV Pup RK98 give a mass ratio $`M_1/M_2=5.5`$, giving in this way $`M_2=0.2M_{}`$ and $`M_1=1.1M_{}`$; for U Sco $`1.0<M_2<1.3M_{}`$; and for MR Ser $`M_2<0.1M_{}`$. For all these systems with the exception of VV Pup, we have also to make use of the mass function. We have considered these values as upper limits to the secondary mass.
## 3 Gravitational Waves from Cataclysmic Variables
We proceed now calculating the gravitational wave amplitude ($`h`$) and frequency ($`f_{gw}`$) for the CVs presented in our catalogue. As already mentioned the binary systems are the most understood of all sources of GWs (see, e.g., Thorne 1987). Knowing the masses of the stars, the orbital parameters and their estimated distances, one can calculate the details of the GW produced. In our catalogue for 68 of the CVs the necessary parameters for the calculation of $`h`$ and $`f_{gw}`$ are known, for the other 88 CVs we needed to obtain their masses through the equations 1 and 2, as discussed in preceding section.
The CVs emit GWs at twice the orbital frequency and harmonics thereof (see, e.g., Thorne 1987). For eccentricity $`ϵ<0.2`$ the line at $`f_{gw}=2f_{orb}`$ is the dominant; for $`ϵ0.5`$ the lines at $`f_{gw}/f_{orb}2`$ through 8 are all strong; for $`ϵ0.7`$ the lines at $`f_{gw}/f_{orb}4`$ through 20 are all strong (see, e.g., Thorne 1987). Following Thorne (1987), the characteristic amplitude, in the low eccentricity case with $`f_{gw}=2f_{orb}`$, is given by
$`h=\mathrm{\hspace{0.17em}8.7}\times \mathrm{\hspace{0.17em}10}^{21}\times `$ (4)
$`\times \left({\displaystyle \frac{\mu }{M_{}}}\right)\left({\displaystyle \frac{M}{M_{}}}\right)^{2/3}\left({\displaystyle \frac{100pc}{r}}\right)\left({\displaystyle \frac{f}{10^3Hz}}\right)^{2/3}`$
The above equation takes into account both polarizations, $`h_+`$ and $`h_\times `$, and it is averaged over the orientation angles of the source (Thorne 1987). The amplitude given by this equation is thus a factor of $`2`$ smaller than the maximum amplitude.
We are also considering that all CVs of our catalogue have low eccentricity, and therefore equation 4 can be applied to them.
The LISA curves, as discussed by the LISA Study Team (1998) are calculated realistically, and in some sense somewhat conservative, due to the fact that the sensitivity could in principle be improved in many aspects.
The LISA mission is planned to last 2 years, but it could last up to 10 years, as a result: a) its sensitivity to long-lived sources is improved; b) the noise, the threshold curves and the GW noise from white-dwarf binaries would lower, as a result it would be possible to resolve more sources and remove them from the binary confusion noise background.
Although the three LISA arms are not independent, LISA could in some sense act as two interferometers, improving its capability of detection and sensitivity. A third arm allows LISA to detect two different GW observable, which can be thought of as being formed from the signals of two different interferometers, with one arm common to both. As a result, besides an improvement in sensitivity, LISA’s ability to measure, for example, the polarization of the GWs is improved. It is worth mentioning that the LISA curves usually presented elsewhere only consider a single interferometer.
In Figure 1 we show the dimensionless amplitude h (using equation 3), for all the CVs presented in Tables 1 and 2, as a function of the GW frequency; also plotted are the curves for the LISA instrumental threshold and the binary confusion noise threshold estimate curves for 1 year of observations and S/N=1. The values for h for all CVs of our catalogue, calculated via equation 4, are also presented in Tables 1 and 2.
In Figure 2 we zoom Figure 1 for the frequency band $`15\times 10^4`$ Hz, and also plot the curves labeled L1 (L5) and CN1 (CN5) which are the LISA instrumental threshold and the binary confusion noise threshold estimate curves for 1 year of observations and S/N=1 (S/N=5), respectively.
Among the CVs presented in our catalogue no one has $`S/N>5`$, and therefore at this signal-to-noise ratio it is not possible to detected them.
It is worth mentioning at this point that even the CV named WZ Sge, which is usually considered to be one of the most promising CVs capable of being detected by LISA, cannot be detected at $`S/N>5`$. We have used here new data presented mainly in the 6th edition of the catalogue of cataclysmic binaries, low mass X-ray binaries and related objects (Ritter & Kolb 1998), and in particular for the WZ Sge the masses presented are smaller than thought before (see, e.g., Douglass & Braginsky 1979). This explain why WZ Sge appears here in our study with a dimensionless amplitude $`h`$ much smaller than presented by the LISA Study Team (1998).
The parameters for WZ Sge used by Douglass & Braginsky (1979) were obtained from Warner (1976), namely, $`M_1=1.5M_{}`$ and $`M_2=0.12M_{}`$ (the masses), and from Kraft (1962), namely, $`d=75pc`$ (the distance). Barret (1996), on the other hand, obtained a distance of 194 pc (this is the distance that appears in Table 1) with the linear polarimetric technique. Smack (1993), instead, obtained the masses $`M_1=0.45M_{}`$, $`M_2=0.058M_{}`$ and a distance of $`d=48pc`$, from visual and ultraviolet observations. Even considering a distance of $`d=48pc`$, WZ Sge would appear below $`S/R=5`$ curves. For comparison we plot WZ Sge for a distance of $`d=48pc`$ (see, Figure 2).
As usual in astrophysics the distance plays a key role here. The case for WZ Sge is an example that we have addressed to call attention to an issue that could occur with almost all other CVs of our catalogue. As a result this uncertainty in the distance could move the points plotted in Figures 1 and 2 upwards or downwards.
From our sample we note that 37 CVs have $`h`$ values greater than the S/N =1 LISA curve and also appear above the binary confusion noise curve, such CVs, therefore, could in principle be detected at this signal-to-noise ratio; Of these 37 CVs, 33 are below the period gap ($`1.25<P<2.16`$ hours). We also note that the maximum distance of these CVs to the Earth is approximately 300 pc. Patterson (1998) estimates that the space density of active CVs is $`d=10^5pc^3`$, with 75% of them below the period gap. So, the expected number of active CVs up to a radius of 300 pc would amount to approximately 850 systems with periods below the gap. We have therefore only a part of them in our catalogue. This means that the prospect of detection of CVs is improved.
It is worth mentioning that even if the sources are not detectable after 1 year of observation they can be detected after an additional integration time t, namely
$$h_{\mathrm{CV}}>(f_{gw}t)^{1/2}h_{\mathrm{confusion}\mathrm{noise}}$$
(5)
(see, e.g., Thorne 1987). It is important to have in mind, however, that below 1 mHz there could exist many binaries per frequency bin that could be hard to resolve individual sources (see, e.g., Hils 1998), unless we know their position in sky like those in Tables 1 and 2.
It is worth mentioning also that due to the fact that the LISA curves presented here are only for single interferometers, and that LISA could work as two independent interferometers, this improves the possibilities of detection of CVs by LISA, since LISA curves as well as the binary confusion noise curves go down.
## 4 Discussion and conclusions
The CVs produce GWs which could in principle be detected by the LISA antenna, since CVs produce low frequency GWs in the frequency band where LISA is sensitive. Due to the fact that a positive detection of a CV by the LISA antenna might be improved once we know the sources beforehand we compile in the present study a catalogue of CVs, for which we know at least their orbital periods and distances.
We argue that the present study is of interest since in the literature one has not found a systematic identification of possible detectable GW CVs, since an early study made by Douglass & Braginsky (1979) twenty years ago, and also a preliminary study by Aguiar et al. (1998). We have been able to catalogue approximately 160 CVs, from which a reasonable part of them could be detected once the LISA antenna become operative.
We argue that it would be of interest whether other groups performed a similar study for the other binary systems which produce low frequency GWs in the frequency band where the LISA antenna is sensitive.
It is worth mentioning that a positive detection of a binary system through its gravitational emission, with some help of electromagnetic data observations, could lead one to know all the parameters related to the binary system, namely, the masses of the stars, their distances to the earth, the period of the system and their orientation angles.
###### Acknowledgements.
MTM and JCNA thank FAPESP (Brazil) for financial support (grants 97/13415-0, 98/07641-0 and 97/06024-4, 97/13720-7, respectively). ODA thanks CNPq (Brazil) for financial support (grant 300619/92-8). We would like to thank Dr. Robin Stebbins and Prof. Peter Bender for kindly providing us with the LISA sensitivity curve.
|
warning/0004/cond-mat0004023.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The writing of this review is provoked by the progress in the study of fluctuation phenomena in high-$`T_\mathrm{c}`$ superconductors.<sup>?</sup> The small coherence lengths in the layered cuprates $`\xi _c\xi _a\xi _b`$ give rise to a very high density of fluctuation degrees of freedom $`1/(\xi _{ab}^2(0)\xi _c(0))`$ which makes the fluctuation effects easier to be observe in the high-$`T_\mathrm{c}`$\- rather than in the conventional superconductors. An intriguing feature of the fluctuation effects to be pointed out is that they can be observed even in the case when the interaction between fluctuations is vanishing or can be treated in a self-consistent manner. In such a case, for high quality crystals, the fluctuations are of Gaussian nature and their theory is very simple. A number of good experimental studies have already been performed in the Gaussian regime thus initiating the Gaussian fluctuation spectroscopy for high-$`T_\mathrm{c}`$ materials.
By spectroscopy here we imply only those experiments with trivial theory where every measurement provides an immediate information for some parameter(s) important for the material science or fundamental physics of these interesting materials. Half a century ago Landau used to speak about himself as being the greatest trivializator in the theoretical physics. At present, the Ginzburg-Landau (GL) theory (called by Ginzburg also $`\mathrm{\Psi }`$-theory) is the adequate tool to describe the fluctuation phenomena in the superconductors. The parameters of the $`\mathrm{\Psi }`$-theory, such as coherence lengths, relaxation time $`\tau _{0,\mathrm{\Psi }}`$ of the GL order parameter $`\mathrm{\Psi },`$ the GL parameter $`\kappa _{_{\mathrm{GL}}}=\lambda _{ab}(0)/\xi _{ab}(0)`$ are also ”meeting point” between the theory and the experiment.
From one hand, these parameters are necessary for the description of the experimental data and from another hand they can be derived form the microscopic theory using the methods of the statistical mechanics. That is why the determination of the GL parameters is an important part of the investigations of every superconductor and the Gaussian fluctuation spectroscopy is an indispensable tool in these comprehensive investigations.
The purpose of this review is to systematize the known classical results for the GL Gaussian fluctuations, to derive new ones when needed, and to finally give suitable for coding formulae necessary for the further development of the Gaussian spectroscopy. The derivation of all results is described in detail and trivialized to the level of the Landau-Lifshitz encyclopedia on theoretical physics,<sup>?</sup> the textbooks by Abrikosov<sup>?</sup> and Tinkham<sup>?</sup> or the well-known reviews by Cyrot<sup>?</sup> on the GL theory, by Bulaevskii<sup>?</sup> concerning the layered superconductors with Josephson coupling, and by Skocpol and Tinkham<sup>?</sup> on the fluctuation phenomena in superconductors. The present work is intended as a review on the theoretical results which can be used by the Gaussian spectroscopy of fluctuations but no historical survey of the experimental research is attempted. Therefore we do explicitly refer to only a limited number of experimental studies in this field. Instead, the reader is referred to the citations-reach conference proceedings,<sup>?</sup> but even therein a number of good works are probably not included. We do not cite directly even the epoch-creating paper by Bednorz and Müller but its spirit can be traced to every contemporary paper on high-$`T_\mathrm{c}`$ superconductivity. Even to focus on the theoretical results related to fluctuation phenomena is a very difficult problem by itself and therefore, when referring to any result one should imply ”to the best of our knowledge…”. One of our goals was also to fill the gap between the textbooks and experimentalists’ needs for a compilation of theoretical formulae written in common notations, appropriate for direct use.
Of course, there is a great number of interesting physical situations especially related to vortices where the fluctuations are definitely non-Gaussian. Those problems fall beyond the scope of the present review and we include only some references from this broad field in the physics of superconductivity.<sup>?</sup>
The review is organized as follows: in Sec. 22 the case of weak magnetic fields is considered and the thermodynamic variables are expanded in power series in the dimensionless magnetic field $`h=B_z/B_{c2}(0).`$ The standard notations for the thermodynamic variables in a layered superconductor are then introduced in Subsec. 2.12.1, and Subsec. 2.22.2 is dedicated to the Euler-MacLaurin summation formula in the form appropriate for the analysis of the GL results for the free energy and its ultraviolet (UV) regularization. A systematic procedure to derive the results for a layered superconductor from the results for a two-dimensional (2D) superconductor is developed in Subsec. 2.3 and the action of the introduced ”layering” operator $`\widehat{𝖫}`$ is illustrated on the example of the formulae for the paraconductivity. Further we consider the static paraconductivity in case of perpendicular magnetic field as well as the high frequency conductivity in zero magnetic field. The power series for the nonlinear magnetic susceptibility and the magnetic moment in the Lawrence-Doniach (LD) model are derived in Subsec. 2.4 and the $`\epsilon `$-method for summation of such divergent series is described in Subsec. 2.5. For practical purposes a simple fortran90 program is given in the Appendix.
Further in Subsec. 2.6 we present the power series for the differential susceptibility and general weak-magnetic-field expansion formulae for the magnetization. Section 3 is dedicated to the study of the strong magnetic fields limit. Firstly, in Subsec. 3.1 the general formula for the Gibbs free energy in perpendicular to the layers magnetic field is analyzed. The fluctuation part of the thermodynamic variables is found then by differentiation in Subsec 3.2. Subsec. 3.3 is devoted to the self-consistent mean-field treatment of the fluctuation interactions in the LD model. The important limit case of an anisotropic 3D GL model is considered in Subsec. 3.4 where we derive the Gibbs free energy and the fluctuation magnetic moment. In Sec. 4 an account is given of the fitting procedure for the GL parameters which rests on theoretical results and some recommendations for the most appropriate formulae are also given for determination of the cutoff energy $`\epsilon _{_{\text{}}}`$ in Subsec. 4.1, the in-plane coherence length $`\xi _{ab}(0)`$ in Subsec. 4.2, the Cooper pair life-time constant $`\tau _0`$ in Subsec 4.3, and the 2D Ginzburg number in Subsec. 4.4. All new results derived throughout this review are summarized in Sec. 5 and some perspectives for the Gaussian spectroscopy are discussed as well.
## 2 Weak magnetic fields
### 2.1 Formalism
Before embarking on a detailed analysis we shall briefly introduce all entities entering the basic for our further considerations quantity—the GL functional $`G`$ for the Gibbs free energy in external magnetic field $`𝐇^{(\mathrm{ext})}`$. For compliance with the previous works we follow the standard notations in which $`G`$ reads
$`G[\mathrm{\Psi }_{j,n}`$ $`(x,y),𝐀(𝐫)]`$
$`=`$ $`{\displaystyle \underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}}{\displaystyle \underset{j=1}{\overset{N}{}}}{\displaystyle }dxdy\left\{{\displaystyle \underset{l=x,y}{}}{\displaystyle \frac{1}{2m_{ab}}}\right|({\displaystyle \frac{\mathrm{}}{i}}{\displaystyle \frac{}{x_l}}e^{}A_l)\mathrm{\Psi }_{jn}|^2`$
$`+a_0ϵ|\mathrm{\Psi }_{j,n}|^2+{\displaystyle \frac{1}{2}}\stackrel{~}{b}|\mathrm{\Psi }_{j,n}|^4+a_0\gamma _j|\mathrm{\Psi }_{j+1,n}\mathrm{\Psi }_{j,n}\mathrm{exp}\left({\displaystyle \frac{ie^{}}{\mathrm{}}}{\displaystyle _{z_{j,n}}^{z_{j+1,n}}}A_zdz\right)|^2\}`$
$`+{\displaystyle \frac{1}{2\mu _0}\left(\times 𝐀\mu _0𝐇^{(\mathrm{ext})}\right)^2𝑑x𝑑y𝑑z},`$ (1)
with A being the vector potential of the magnetic field $`B=\times \text{A}.`$
The material parameters in this sizable expression are illustrated in Fig. 2.1, thus we only need to note that the GL potential $`a(ϵ)=a_0ϵ`$ is parameterized by $`a_0=\mathrm{}^2/2m_{ab}\xi _{ab}^2(0)`$, and $`ϵ\mathrm{ln}(T/T_\mathrm{c})(TT_\mathrm{c})/T_\mathrm{c}`$ is the reduced (dimensionless) temperature. If not otherwise stated we shall make use of the SI units, thus the magnetic permeability of vacuum $`\mu _0=4\pi \times 10^7.`$
Here we will restrict ourselves to the study of fluctuations in the Gaussian regime in the normal phase not too close to the critical line $`H_{c2}(T).`$ In this case the nonlinear term in $`G[\mathrm{\Psi },\text{A}]`$ is negligible, $`\stackrel{~}{b}|\mathrm{\Psi }|^4/20.`$ For the normal phase the magnetization is also very small and with high accuracy $`\mu _0𝐇^{(\mathrm{ext})}𝐁=\mu _0(𝐇+𝐌)\mu _0𝐇.`$
To begin with, consider the simplest case of zero external magnetic field $`𝐇^{(\mathrm{ext})}=0.`$ Given the above assumption for $`\mathrm{\Psi },`$ the GL functional is a quadratic form and one needs to sum over all eigenvalues of the energy spectrum
$$\epsilon _j(𝐩,p_z)=\frac{𝐩^2}{2m_{ab}}+\epsilon _{cj}(p_z),\epsilon _{cj}(p_z)=a_0\omega _j^{(N)}(\theta ),$$
(2)
where $`\epsilon _{cj}(p_z)`$ are the tight-binding energy bands describing the motion of Cooper pair in $`z`$ ($`c`$) direction, $`𝐩=(p_x,p_y)`$ is the in-plane ($`ab`$-plane) momentum of the fluctuating Cooper pairs and $`\theta =p_zs/\mathrm{}(0,2\pi )`$ is the Josephson phase. For a single layered material, $`N=1`$, this corresponds to the well known Lawrence-Doniach model,<sup>?</sup>
$$\omega _1^{(\mathrm{LD})}(\theta )=2\gamma _1(1\mathrm{cos}\theta )$$
(3)
while the case $`N=2`$ is the Maki-Thompson (MT) model,<sup>?</sup> proposed independently by Hikami and Larkin<sup>?</sup> as well,
$$\omega _j^{(\mathrm{MT})}(\theta )=\gamma _1+\gamma _2+(1)^j\sqrt{\gamma _1^2+\gamma _2^2+2\gamma _1\gamma _2\mathrm{cos}\theta },j=1,2.$$
(4)
Thus, the sum over the energy spectrum gives
$$G[\mathrm{\Psi }]=\underset{𝐩,p_z,j}{}\left(\epsilon _j(𝐩,p_z)+a\right)\left|\mathrm{\Psi }_{𝐩,p_z,j}\right|^2,$$
(5)
where $`\mathrm{\Psi }_{𝐩,p_z,j}`$ is the wave function of the superconducting condensate in momentum space. We use standard periodic boundary conditions for a bulk domain of volume $`V=L_x\times L_y\times L_z`$ which give
$$\underset{𝐩}{}=L_xL_y\frac{dp_xdp_y}{(2\pi \mathrm{})^2},\underset{p_z}{}=L_z\frac{d\theta }{2\pi s}.$$
(6)
In order to calculate the fluctuation part of the Gibbs free energy $`G(T)`$ at zero magnetic field, one usually solves for every point in the momentum space $`𝐩,p_z,j`$ the Gaussian integral
$`\mathrm{exp}\left[{\displaystyle \frac{G}{k__\mathrm{B}T}}\right]`$ $`={\displaystyle \frac{d\mathrm{\Psi }^{}d\mathrm{\Psi }^{\prime \prime }}{2\pi }\mathrm{exp}\left\{\frac{\epsilon +a}{k__\mathrm{B}T}\left[\left(\mathrm{\Psi }^{}\right)^2+\left(\mathrm{\Psi }^{\prime \prime }\right)^2\right]\right\}}`$
$`={\displaystyle \frac{1}{2}}{\displaystyle _0^{\mathrm{}}}\mathrm{exp}\left[{\displaystyle \frac{\epsilon +a}{k__\mathrm{B}T}}\rho \right]𝑑\rho ={\displaystyle \frac{k__\mathrm{B}T/2}{\epsilon +a}},`$ (7)
where $`\mathrm{\Psi }=\sqrt{\rho }e^{i\phi }\mathrm{\Psi }^{}+i\mathrm{\Psi }^{\prime \prime },`$ and $`\phi (0,2\pi ).`$ Making use of this auxiliary result the calculation of the fluctuation part of the Gibbs free energy reduces to summation over the spectrum of an effective Hamiltonian, i.e.
$$G=k__\mathrm{B}T\underset{𝐩,p_z,j}{}\mathrm{ln}\frac{k__\mathrm{B}T/2}{\epsilon _j(𝐩,p_z)+a},$$
(8)
or, taking into account Eqs. (2) and (6),
$$\frac{G}{V}=k__\mathrm{B}T\frac{d\left(\pi 𝐩^2\right)}{(2\pi \mathrm{})^2}\frac{1}{N}\underset{j=1}{\overset{N}{}}\frac{d\theta }{2\pi s}\mathrm{ln}\left[\frac{(𝐩^2/2m_{ab})+a_0\omega _j^{(N)}(\theta )+a_0ϵ}{a_0}\frac{a_0}{\frac{1}{2}k__\mathrm{B}T}\right].$$
(9)
In view of the further calculations it is also useful to introduce a dimensionless in-plane kinetic energy
$$\stackrel{~}{x}=\frac{𝐩^2}{2m_{ab}a_0}=\left(\frac{\xi _{ab}(0)𝐩}{\mathrm{}}\right)^2(0,c),$$
(10)
bound by a dimensionless cutoff parameter $`c`$ which we consider to be an important parameter of the GL theory when applied to copper oxide superconductors. Later in Sec 4.1 we demonstrate how the value of the dimensional cutoff energy $`\epsilon _{_{\text{}}}`$,
$$\frac{𝐩^2}{2m_{ab}}<\epsilon _{_{\text{}}}=ca_0=\frac{p_c^2}{2m_{ab}},$$
(11)
can be determined by fitting to the experimental data. An immediate simplification to Eq. (9) can be achieved by dropping the $`\frac{1}{2}k__\mathrm{B}T/a_0\mathrm{const}`$ multiplier in the argument of the logarithm as it is irrelevant for the critical behavior of the material. Furthermore, since fluctuational observables are related to non analytical dependence of the Gibbs free energy on the reduced temperature, we can substitute $`T=T_\mathrm{c}(1+ϵ)T_\mathrm{c}`$ and the free energy per unit volume $`F(ϵ)`$ is cast in more elegant form,
$$F(ϵ)\frac{G}{L_xL_yL_z}=\frac{k__\mathrm{B}T_\mathrm{c}}{4\pi \xi _{ab}^2(0)}\frac{N}{s}_0^c𝑑\stackrel{~}{x}\frac{1}{N}\underset{j=1}{\overset{N}{}}\frac{d\theta }{2\pi }\mathrm{ln}\left(\stackrel{~}{x}+\omega _j^{(N)}(\theta )+ϵ\right),$$
(12)
that could easily include the $`(1+ϵ)`$-factor in all cases when necessity appears. The physical meaning of this important for our further considerations expression is fairly transparent: one has to integrate with respect to the Josephson phase $`\theta ,`$ which describes the motion of Cooper pairs in $`c`$-direction, and to take into account as many different Cooper pair energy bands as are there the different superconducting layers per unit cell. Finally, integration with respect to the in-plane Cooper pair kinetic energy is to be carried out.
Consider now the important case of an external magnetic field applied parallel to the $`c`$-direction, i.e. perpendicular to the CuO<sub>2</sub> planes, $`𝐁=(0,0,B).`$ In this case, the in-plane kinetic energy of the Cooper pairs acquires oscillator spectrum,<sup>?</sup> corresponding to the quantum mechanical problem of an electron in an external magnetic field,<sup>?</sup>
$$\frac{𝐩^2}{2m_{ab}}\mathrm{}\omega _c\left(n+\frac{1}{2}\right),$$
(13)
where $`n=0,1,2,3,\mathrm{}`$ is a non-negative integer and $`\omega _c=\left|e^{}\right|B/m_{ab}`$ is the cyclotron frequency. The integration over the momentum space is thus reduced to summation over oscillator energy levels
$$_{\left|𝐩\right|<p_c}\frac{d^2𝐩}{(2\pi \mathrm{})^2}\frac{B}{\mathrm{\Phi }_0}\underset{n=0}{\overset{n_c1}{}},$$
(14)
where $`n_cc/2h`$ and $`\mathrm{\Phi }_0=2\pi \mathrm{}/\left|e^{}\right|=2.07`$ $`\mathrm{fT}\mathrm{m}^2`$ is the flux quantum. The energy cutoff is to be applied now to the oscillator levels,<sup>?</sup> $`\mathrm{}\omega _c(n_c+\frac{1}{2})=ca_0.`$ Let us recall that the equation for the upper critical field $`H_{c2}(T)`$ within the GL theory is nothing but the equation for annulment of the lowest energy level, $`\frac{1}{2}\mathrm{}\omega _c+a(ϵ)=0.`$ Thereby introducing the upper critical field linearly extrapolated to zero temperature,
$$\mu _0H_{c2}(0)=B_{c2}(0)T_\mathrm{c}\frac{dB_{c2}(T)}{dT}|_{T_\mathrm{c}}=\frac{\mathrm{\Phi }_0}{2\pi \xi _{ab}^2(0)},$$
(15)
and the dimensionless reduced magnetic field,
$$h\frac{B}{B_{c2}(0)}=\frac{H}{H_{c2}(0)},$$
(16)
we obtain a linear approximation for the critical line about $`T_\mathrm{c},`$
$$h_{c2}(ϵ)=\frac{H_{c2}(T)}{H_{c2}(0)}ϵ1.$$
(17)
With the help of the dimensionless variables introduced so far it is easily worked out that the influence of the external magnetic field is reduced to discretization of the dimensionless in-plane kinetic energy,
$$\stackrel{~}{x}h(2n+1)$$
(18)
and the integrals of an arbitrary function $`f`$ with respect to $`\stackrel{~}{x}`$ are converted to sums,
$$_0^cf(\stackrel{~}{x})𝑑\stackrel{~}{x}2h\underset{n=0}{\overset{n_c1}{}}f(h(2n+1)).$$
(19)
In fact, Max Planck discovered the quantum statistics of the black-body radiation using the same replacement. Applying this procedure to the previously derived free energy at zero magnetic field, Eq. (12), we obtain
$$F(ϵ)F(ϵ,h)=\frac{\mathrm{\Delta }G}{L_xL_yL_z},$$
(20)
$$F(ϵ,h)=\frac{k__\mathrm{B}T_\mathrm{c}}{4\pi \xi _{ab}(0)^2}\frac{N}{s}2h\underset{n=0}{\overset{n_c1}{}}\frac{1}{N}\underset{j=1}{\overset{N}{}}\frac{d\theta }{2\pi }\mathrm{ln}\left[h(2n+1)+\omega _j^{(N)}(\theta )+ϵ\right].$$
(21)
This expression represents the starting point for all further considerations. As a first step we address in the next section the Euler-MacLaurin method and its application to the sum over the Landau levels which appears in Eq. (21).
### 2.2 Euler-MacLaurin summation for the free energy
Near the critical temperature, when $`ϵc,`$ one can consider formally $`c\mathrm{}`$, and $`n_c(h)c/2h\mathrm{}.`$ Within such a local approximation the previous finite sums are transformed into infinite ones,
$$2h\underset{n=0}{\overset{\mathrm{}}{}}f(ϵ+h(2n+1))=\widehat{\mathsf{\Sigma }}_{\mathrm{𝖤𝖬}}\underset{ϵ}{\overset{\mathrm{}}{}}f(\stackrel{~}{x})𝑑\stackrel{~}{x},$$
(22)
where
$`\widehat{\mathsf{\Sigma }}_{\mathrm{𝖤𝖬}}{\displaystyle \frac{h\frac{}{ϵ}}{\mathrm{sinh}\left(h\frac{}{ϵ}\right)}}`$ $`={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}(1)^n{\displaystyle \frac{2}{\pi ^{2n}}}\left(1{\displaystyle \frac{1}{2^{2n1}}}\right)\zeta (2n)\left(h{\displaystyle \frac{}{ϵ}}\right)^{2n}`$
$`=1{\displaystyle \frac{1}{6}}h^2{\displaystyle \frac{^2}{ϵ^2}}+{\displaystyle \frac{7}{360}}h^4{\displaystyle \frac{^4}{ϵ^4}}{\displaystyle \frac{31}{15120}}h^6{\displaystyle \frac{^6}{ϵ^6}}+\mathrm{}`$ (23)
is the Euler-MacLaurin operator for summation of series, in which we employ the Riemann and Hurwitz zeta functions, respectively,
$$\zeta (\nu )=1+\frac{1}{2^\nu }+\frac{1}{3^\nu }+\mathrm{}=\zeta (\nu ,1),\zeta (\nu ,z)=\underset{n=0}{\overset{\mathrm{}}{}}\frac{1}{\left(n+z\right)^\nu },$$
(24)
instead of Bernoulli numbers. The $`\widehat{\mathsf{\Sigma }}_{\mathrm{𝖤𝖬}}`$ operator can be easily obtained exploiting the exponential representation of the standard translation operator $`\widehat{𝖳}`$, whose action is defined as follows
$$f(b+ϵ)=\widehat{𝖳}_z(b)f(z)|_{z=ϵ}=\mathrm{exp}\left(b\frac{}{z}\right)f(z)|_{z=ϵ}.$$
(25)
If summed up from zero to infinity the above expression would give an infinite geometric progression,
$$\underset{n=0}{\overset{\mathrm{}}{}}\left[\widehat{𝖳}_z(b)\right]^n=\underset{n=0}{\overset{\mathrm{}}{}}\left[\mathrm{exp}\left(b\frac{}{z}\right)\right]^n=\frac{1}{1\mathrm{exp}\left(b\frac{}{z}\right)}.$$
(26)
Let us introduce now the fluctuational part of the heat capacity,
$$C(ϵ)=\frac{1}{T_\mathrm{c}}\frac{^2}{ϵ^2}F(ϵ).$$
(27)
Using this physical observable and Eq. (21), one can extract the magnetic field dependent part of the free energy,
$`F(ϵ,h)F(ϵ)`$ $`={\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}(1)^{n1}{\displaystyle \frac{2}{\pi ^{2n}}}\left(1{\displaystyle \frac{1}{2^{2n1}}}\right)\zeta (2n)h^{2n}{\displaystyle \frac{^{2(n1)}}{ϵ^{2(n1)}}}T_\mathrm{c}C(ϵ)`$
$`=\left[{\displaystyle \frac{1}{6}}h^2{\displaystyle \frac{7}{360}}h^4{\displaystyle \frac{^2}{ϵ^2}}+{\displaystyle \frac{31}{15120}}h^6{\displaystyle \frac{^4}{ϵ^4}}\mathrm{}\right]T_\mathrm{c}C(ϵ).`$ (28)
It is then straightforward to calculate the magnetization $`M`$ and the nonlinear susceptibility defined as $`\chi (ϵ,h)M/H,`$ i.e.
$$M=\frac{[F(ϵ,h)F(ϵ)]}{B}=\chi (ϵ,h)H.$$
(29)
For the Meissner-Ochsenfeld (MO) state, for example, $`\chi ^{(\mathrm{MO})}=1.`$ Importantly, Eq. (29) incorporates the regularized free energy which, by virtue of Eq. (28), does not contain zero magnetic field part,
$$F_{\mathrm{reg}}(ϵ,h)F(ϵ,h)F(ϵ):=\widehat{\mathrm{𝖱𝖾𝗀}}_{\mathrm{𝖤𝖬}}F(ϵ)$$
(30)
where
$$\widehat{\mathrm{𝖱𝖾𝗀}}_{\mathrm{𝖤𝖬}}=\widehat{\mathsf{\Sigma }}_{\mathrm{𝖤𝖬}}\widehat{\mathrm{𝟣}}=2h\underset{n=0}{\overset{n_c}{}}_0^c𝑑\stackrel{~}{x}$$
(31)
is the Euler-MacLaurin regularization operator. This method was applied<sup>?,?,?</sup> for calculation of zero-field limit of the magnetic susceptibility, cf. Ref. 2. Hence, inserting the $`h^2`$-term of Eq. (28) into Eq. (29) one finds<sup>?</sup> for $`H0`$
$$\chi (ϵ)=\frac{\mu _0T_\mathrm{c}C(ϵ)}{3B_{c2}^2(0)}=\frac{4\pi ^2\mu _0}{3\mathrm{\Phi }_0^2}\xi _{ab}^4(0)T_\mathrm{c}C(ϵ).$$
(32)
For illustration, let us analyze how this relation between susceptibility and heat capacity can be applied to the LD model. For arbitrary multilayered structure we can calculate the curvature of the lowest dimensionless energy band in $`c`$-direction $`\omega _1(\theta )=\epsilon _{c1}(p_z)/a_0,`$
$$r2\frac{^2}{\theta ^2}\omega _1(\theta )|_{\theta =0}.$$
(33)
According to this definition $`r`$ parameterizes the effective mass in $`c`$-direction $`m_c`$ for an anisotropic GL model, $`|p_z|\pi \mathrm{}/s`$,
$$\epsilon _{c1}a_0\frac{r}{4}\theta ^2=\frac{p_z^2}{2m_c}.$$
(34)
It is now easily realized that the identity holds true,
$$r=\left(\frac{2N\xi _c(0)}{s}\right)^2,$$
(35)
which for $`N=1`$ is the LD parameter $`r`$ that determines the effective dimensionality of the superconductor, cf. the review by Varlamov et al.<sup>?</sup> In many other studies, e.g. Ref. 7, the wave vector $`𝐤=𝐩/\mathrm{}`$ has been used as well. In terms of the latter, for the dimensionless kinetic energy in the long-wavelength approximation we have
$$\frac{\left[\epsilon _{ab}(\mathrm{}𝐤)+\epsilon _{c1}(\mathrm{}k_z)\right]}{a_0}\xi _{ab}^2(0)𝐤^2+\xi _c^2(0)k_z^2=\xi _{ab}^2(0)𝐤^2+r\theta ^2,$$
(36)
where $`\epsilon _{ab}(\mathrm{}𝐤)`$ represents the in-plane part of the kinetic energy. Let us mention that the LD model is not only applicable to single layered cuprates with $`N=1,`$ but is it also to bi-layered cuprates ($`N=2`$) in the limit cases $`\gamma _1\gamma _2`$ as well as in the case $`\gamma _1\gamma _2`$ when formally $`N=1`$. That is why we use in our formulae an effective periodicity of the LD-model
$$s_{\mathrm{eff}}=\frac{s}{N}.$$
(37)
For completeness we list below without deriving some of the well-known results within the LD-model. The single energy band has the form
$`\epsilon _c(p_z)`$ $`=a_0\omega _1\left({\displaystyle \frac{p_zs}{2\pi \mathrm{}}}\right)={\displaystyle \frac{\mathrm{}^2}{m_c(s/N)^2}}\left(1\mathrm{cos}\theta \right),\text{ where}`$ (38)
$`\omega _1(\theta )`$ $`={\displaystyle \frac{1}{2}}r\left(1\mathrm{cos}\theta \right)=r\mathrm{sin}^2{\displaystyle \frac{\theta }{2}},`$ (39)
being parameterized by the Josephson coupling energy
$$J_1=a_0\gamma _1=\frac{\mathrm{}^2}{m_c(s/N)^2},r=4\gamma _1=\left(2N\xi _c(0)/s\right)^2.$$
(40)
For the heat capacity one has
$$C^{(\mathrm{LD})}(ϵ)=\frac{k__\mathrm{B}}{4\pi \xi _{ab}^2(0)}\frac{N}{s}\frac{1}{\sqrt{ϵ(ϵ+r)}},$$
(41)
and the magnetic susceptibility for a weak magnetic field applied in $`c`$-direction, according to Tsuzuki<sup>?</sup> and Yamayi,<sup>?</sup> reads as
$$\chi ^{(\mathrm{LD})}(ϵ)=\frac{\pi }{3}\mu _0\frac{k__\mathrm{B}T_\mathrm{c}}{\mathrm{\Phi }_0^2}\xi _{ab}^2(0)\frac{N}{s}\frac{1}{\sqrt{ϵ}}\frac{1}{\sqrt{ϵ+r}}=\frac{1}{6}\frac{M_0}{H_{c2}(0)}\frac{1}{\sqrt{ϵ(ϵ+r)}},$$
(42)
where
$$M_0\frac{k__\mathrm{B}T_\mathrm{c}}{\mathrm{\Phi }_0}\frac{N}{s}.$$
(43)
Before proceeding we feel it appealing to make some technical remarks concerning the representation of the general formulae for fluctuations in arbitrary layered superconductor. To be specific, we shall demonstrate how the expressions for the magnetic susceptibility, Eq. (42), and heat capacity, Eq. (41), within the LD model can be obtained as special cases of a general procedure described in the next subsection.
### 2.3 Layering operator $`\widehat{𝖫}`$ illustrated on the example of paraconductivity
In the general formula for the density of the free energy, Eq. (12), the energies related to motion in $`c`$-direction $`\epsilon _{cj}(p_z)`$ enter the final result solely via the fragment $`ϵ+\omega _j(\theta ).`$ Thus, in all such cases one can first solve the corresponding 2D problem and then for a layered superconductor the result can be derived by merely averaging the 2D result with respect to the motion of the fluctuation Cooper pairs in perpendicular to the layers direction. Formally, this method reduces to introducing a layering operator $`\widehat{𝖫}`$ acting on functions of $`ϵ`$; e.g. for the conductivity one would have the relation
$$\sigma (ϵ)=\widehat{𝖫}\sigma ^{(2\mathrm{D})}(ϵ)\frac{1}{N}\underset{j=1}{\overset{N}{}}\frac{d\theta }{2\pi }\sigma ^{(2\mathrm{D})}\left(ϵ+\omega _j^{(N)}(\theta )\right).$$
(44)
In terms of the so introduced operator $`\widehat{𝖫}`$ the expression for the free energy, Eq. (21), takes the form
$$F(ϵ)=F_0_0^c𝑑\stackrel{~}{x}\underset{j=1}{\overset{N}{}}\widehat{𝖫}\mathrm{ln}\left(\stackrel{~}{x}+\omega _j^{(N)}(\theta )+ϵ\right),F_0\frac{k__\mathrm{B}T_\mathrm{c}}{4\pi \xi _{ab}^2(0)}\frac{N}{s}.$$
(45)
Besides for thermodynamic variables this operator works also for the fluctuation in-plane conductivity. Conforming with the work of Hikami and Larkin,<sup>?</sup> for the conductivity within the LD model we have to integrate the 2D conductivity with respect to the Josephson phase,
$$\sigma (ϵ)=\widehat{𝖫}^{(\mathrm{LD})}\sigma ^{(2\mathrm{D})}(ϵ)\frac{d\theta }{2\pi }\sigma ^{(2\mathrm{D})}\left(ϵ+\frac{1}{2}r(1\mathrm{cos}\theta )\right).$$
(46)
Given a system with independent 2D layers, having density in $`c`$-direction $`N/s`$, for zero magnetic field we have to average the well-known Aslamazov-Larkin expression for the static (zero-frequency) conductivity,
$$\sigma _{\mathrm{AL}}(ϵ)=\frac{e^2}{16\mathrm{}}\frac{N}{s}\frac{1}{ϵ}=\frac{\pi }{8}R_{\mathrm{QHE}}^1\frac{N}{s}\frac{1}{ϵ},$$
(47)
where $`R_{\mathrm{QHE}}2\pi \mathrm{}/e^2=25.813`$ k$`\mathrm{\Omega }.`$ A simple integration gives
$$f_{\mathrm{LD}}(ϵ;r)\widehat{𝖫}^{(\mathrm{LD})}\frac{1}{ϵ}=\frac{d\theta }{2\pi }\frac{1}{ϵ+\frac{1}{2}r\left(1\mathrm{cos}\theta \right)}=\frac{1}{\sqrt{ϵ(ϵ+r)}}.$$
(48)
We note that this integral determines both the heat capacity and magnetic susceptibility for the LD model and is widely used for fitting to experimental data. Another important integral is
$`\widehat{𝖫}^{(\mathrm{LD})}{\displaystyle _ϵ^c}\mathrm{ln}\stackrel{~}{ϵ}d\stackrel{~}{ϵ}`$ $`={\displaystyle _ϵ^c}2\mathrm{ln}\left({\displaystyle \frac{\sqrt{\stackrel{~}{ϵ}}+\sqrt{\stackrel{~}{ϵ}+r}}{2}}\right)𝑑\stackrel{~}{ϵ}`$
$`=\left[(2\stackrel{~}{ϵ}+r)\mathrm{ln}\left(\sqrt{\stackrel{~}{ϵ}}+\sqrt{\stackrel{~}{ϵ}+r}\right)\sqrt{\stackrel{~}{ϵ}(\stackrel{~}{ϵ}+r)}\mathrm{ln}(4)\stackrel{~}{ϵ}\right]|_ϵ^c,`$ (49)
which is used in representing the free energy at zero magnetic field, Eq. (12), cf. also Eqs. (129) and (130) below. Further, the $`ϵ`$-derivative of this equation,
$$\widehat{𝖫}^{(\mathrm{LD})}\mathrm{ln}ϵ=\frac{d\theta }{2\pi }\mathrm{ln}\left(ϵ+\frac{1}{2}r\left(1\mathrm{cos}\theta \right)\right)=2\mathrm{ln}\left(\frac{\sqrt{ϵ}+\sqrt{ϵ+r}}{2}\right),$$
(50)
is important for the calculation of the fluctuation part of the entropy and the density of fluctuation Cooper pairs. We provide also two other integrals employed in calculating the magnetoconductivity<sup>?</sup>
$`\widehat{𝖫}^{(\mathrm{LD})}{\displaystyle \frac{1}{ϵ^2}}`$ $`=\widehat{𝖫}^{(\mathrm{LD})}{\displaystyle \frac{}{ϵ}}{\displaystyle \frac{1}{ϵ}}={\displaystyle \frac{ϵ+\frac{1}{2}r}{[ϵ(ϵ+r)]^{3/2}}},`$ (51)
$`\widehat{𝖫}^{(\mathrm{LD})}{\displaystyle \frac{1}{ϵ^3}}`$ $`={\displaystyle \frac{1}{2}}{\displaystyle \frac{^2}{ϵ^2}}\widehat{𝖫}^{(\mathrm{LD})}{\displaystyle \frac{1}{ϵ}}={\displaystyle \frac{(ϵ+r)ϵ+\frac{3}{8}r^2}{[ϵ(ϵ+r)]^{5/2}}}.`$ (52)
Analogously, in the Maki-Thompson bi-layered model<sup>?</sup> ($`N=2`$) one summation should precede the integration,<sup>?</sup>
$$\sigma (ϵ)=\widehat{𝖫}^{(\mathrm{MT})}\sigma ^{(2\mathrm{D})}(ϵ)=\frac{1}{2}\frac{d\theta }{2\pi }\left[\sigma ^{(2\mathrm{D})}\left(ϵ+\omega _1(\theta )\right)+\sigma ^{(2\mathrm{D})}\left(ϵ+\omega _2(\theta )\right)\right],$$
(53)
i.e. in order to calculate the conductivity<sup>?</sup> and susceptibility<sup>?</sup> we have to add the terms
$$\frac{1}{ϵ+\omega _1}+\frac{1}{ϵ+\omega _2}=\frac{2ϵ+(\omega _1+\omega _2)}{ϵ^2+(\omega _1+\omega _2)ϵ+\omega _1\omega _2}.$$
(54)
In this expression both $`\omega _1+\omega _2`$ and $`\omega _1\omega _2`$ are rational, cf. Eq. (4), and the integral (53) is reduced to the integral (48),
$`\widehat{𝖫}^{(\mathrm{MT})}{\displaystyle \frac{1}{ϵ}}`$ $`={\displaystyle \frac{ϵ+\gamma _1+\gamma _2}{\sqrt{ϵ\left[ϵ+2(\gamma _1+\gamma _2)\right]\left(ϵ+2\gamma _1\right)\left(ϵ+2\gamma _2\right)}}}`$
$`={\displaystyle \frac{ϵ+\frac{1}{2}rw}{\sqrt{\left(ϵ^2+rwϵ\right)\left(ϵ^2+rwϵ+\frac{1}{4}r^2w\right)}}}f_{\mathrm{MT}}(ϵ,h;r,w),`$ (55)
where
$$r4\frac{2}{\frac{1}{\gamma _1}+\frac{1}{\gamma _2}},u\frac{J_{\mathrm{max}}}{J_{\mathrm{min}}}=\frac{\gamma _{\mathrm{max}}}{\gamma _{\mathrm{min}}},w\frac{1}{4}\left(2+\frac{\gamma _1}{\gamma _2}+\frac{\gamma _2}{\gamma _1}\right)=\frac{1}{4}\left(2+u+\frac{1}{u}\right).$$
Such a form, involving the $`w`$ parameter, is convenient for fitting the experimental data since for both $`w=1`$ and $`w1`$ cases the LD approximation holds true, which is often found to give a satisfactory explanation of the experimental observations. For more detailed discussion the reader is referred to Ref. 21. The inverse relations for the above introduced parameters read as
$$u=\left(2w1\right)+2\sqrt{w(w1)},\gamma _{\mathrm{min}}=\frac{1}{2}\left(1+\frac{1}{u}\right)\frac{r}{4},\gamma _{\mathrm{max}}=\frac{1}{2}\left(1+u\right)\frac{r}{4},$$
(56)
and
$$J_{\mathrm{max}}=a_0\gamma _{\mathrm{max}}=\frac{\mathrm{}^2}{2m_{ab}\xi _{ab}^2(0)}\frac{1}{2}\left(1+u\right)\frac{r}{4}.$$
(57)
Importantly, at known effective mass $`m_{ab}`$ the last equation gives the possibility one to determine the Josephson coupling energy between double CuO<sub>2</sub> planes.
Let us now illustrate in more details the action of the $`\widehat{𝖫}`$ operator. Towards this end we consider the famous Aslamazov-Larkin<sup>?</sup> formula for the 2D conductivity $`\sigma _{ab}^{(2\mathrm{D})}(ϵ)`$, the results for the susceptibility $`\chi _{ab}^{(2\mathrm{D})}(ϵ)`$ due to V. Schmidt,<sup>?</sup> A. Schmid<sup>?</sup> and H. Schmidt,<sup>?</sup> and Ferrell<sup>?</sup> and Thouless<sup>?</sup> fluctuation part of the heat capacity $`C^{(2\mathrm{D})}(ϵ)`$. With the help of the $`\widehat{𝖫}`$ operator for arbitrary layered superconductor, cf. Refs. 25 and 26, these three quantities can be generally written as
$`\sigma _{ab}(ϵ)`$ $`={\displaystyle \frac{1}{R_{\mathrm{QHE}}}}\left({\displaystyle \frac{2\tau _0k__\mathrm{B}T_\mathrm{c}}{\mathrm{}}}\right){\displaystyle \frac{N}{s}}\widehat{𝖫}{\displaystyle \frac{1}{ϵ}},`$ (58)
$`\chi _{ab}(ϵ)`$ $`={\displaystyle \frac{\pi }{3}}\mu _0{\displaystyle \frac{k__\mathrm{B}T}{\mathrm{\Phi }_0^2}}\xi _{ab}^2(0){\displaystyle \frac{N}{s}}\widehat{𝖫}{\displaystyle \frac{1}{ϵ}},`$ (59)
$`C(ϵ)`$ $`={\displaystyle \frac{k__\mathrm{B}}{4\pi \xi _{ab}^2(0)}}{\displaystyle \frac{N}{s}}\widehat{𝖫}{\displaystyle \frac{1}{ϵ}},`$ (60)
where the “$`ab`$”-subscript in Eq. (58) indicates that the conductivity is in the $`ab`$-planes, while in Eq. (59) it indicates that the vanishing magnetic field is perpendicular to the same planes. It is immediately apparent that the common for all these expressions function $`\widehat{𝖫}ϵ^1`$ cancels when calculating the $`\chi /C,`$ $`\sigma /C,`$ and $`\sigma /\chi `$ quotients.<sup>?</sup> In particular, the temperature independent ratio
$$\tau _0=\frac{\mu _0}{3}\xi _{ab}^2(0)\frac{\sigma _{ab}(ϵ)}{\chi _{ab}(ϵ)}=\text{const}$$
(61)
provides the best method for probing the time constant $`\tau _0`$ parameterizing the life time of the fluctuation Cooper pairs with zero momentum,
$$\tau (ϵ)=\frac{\tau _0}{ϵ},\left|\psi _{𝐩=0}(t)\right|^2\mathrm{exp}\left(\frac{t}{\tau (ϵ)}\right).$$
(62)
The $`\tau _0`$ constant participates in time-dependent GL (TDGL) theory; see for example the reviews by Cyrot,<sup>?</sup> Skocpol and Tinkham,<sup>?</sup> and the textbooks by Abrikosov<sup>?</sup> and Tinkham.<sup>?</sup> Within the weak coupling BCS theory in the case of negligible depairing mechanisms<sup>?,?,?,?,?,?,?,?,?,?</sup> $`\tau _0`$ satisfies the relation
$$\frac{\tau _{0,\mathrm{\Psi }}^{(\mathrm{BCS})}k__\mathrm{B}T_\mathrm{c}}{\mathrm{}}=\frac{\pi }{8},\tau _0^{(\mathrm{BCS})}=\frac{\pi }{16}\frac{\mathrm{}}{k__\mathrm{B}T_\mathrm{c}},\tau ^{(\mathrm{BCS})}(ϵ)=\frac{\tau _0^{(\mathrm{BCS})}}{ϵ}=\frac{\pi }{16}\frac{\mathrm{}}{k__\mathrm{B}T_\mathrm{c}}\frac{1}{ϵ},$$
(63)
where $`\tau _{0,\mathrm{\Psi }}^{(\mathrm{BCS})}2\tau _0^{(\mathrm{BCS})}`$ is the relaxation time constant for the order parameter being two times larger.<sup>?</sup> At the present experimental accuracy this BCS value agrees well with the experimental data for the layered cuprates. Thus, the above observation led us to propose the dimensionless ratio<sup>?</sup>
$$\stackrel{~}{\tau }_{\mathrm{rel}}\frac{\tau _0}{\tau _0^{(\mathrm{BCS})}}=32\frac{k__\mathrm{B}T_\mathrm{c}\tau _0}{(2\pi \mathrm{})}=\frac{8k__\mathrm{B}T_\mathrm{c}\tau _{0,\mathrm{\Psi }}}{\pi \mathrm{}}=\frac{16\mu _0}{3\pi \mathrm{}}\xi _{ab}^2(0)\frac{k__\mathrm{B}T\sigma _{ab}(ϵ)}{\chi _{ab}(ϵ)}=\text{const},$$
(64)
to be used for more reliable experimental data processing; any deviation of $`\stackrel{~}{\tau }_{\mathrm{rel}}`$ from unity should be interpreted as a hint towards unconventional behavior and presence of depairing mechanisms. Notice also that the BCS value $`\pi /8=0.393`$ in Eqs. (47), (63), and (64) is extremely robust, being originally derived for dirty 3D superconductors, and the $`\tau _0T_\mathrm{c}`$ product remains the same<sup>?,?</sup> for clean 2D superconductors and is not affected by the multilaminarity. The general formula for the fluctuation conductivity of a layered superconductor in perpendicular magnetic field can be also rewritten via the layering operator and relative life-time employing the 2D results by Redi,<sup>?</sup> and Abrahams, Prange and Stefen<sup>?</sup> (APS), cf. also Ref. 16,
$$\sigma _{ab}(ϵ,h)=\stackrel{~}{\tau }_{\mathrm{rel}}\frac{e^2}{16\mathrm{}}\frac{N}{s}\widehat{𝖫}f_{\mathrm{APS}}(ϵ,h),$$
(65)
where, for $`ϵ+h>0`$,
$$f_{\mathrm{APS}}(ϵ,h)\frac{1}{ϵ}2\left(\frac{ϵ}{h}\right)^2\left[\psi \left(\frac{1}{2}+\frac{ϵ}{2h}\right)\psi \left(1+\frac{ϵ}{2h}\right)+\frac{h}{ϵ}\right],$$
(66)
is an universal dimensionless function of dimensionless reduced temperature $`ϵ`$ and dimensionless magnetic field $`h.`$ The functions
$$\mathrm{\Gamma }(z)_0^{\mathrm{}}e^tt^{z1}𝑑t,\psi (z)\frac{d}{dz}\mathrm{ln}\mathrm{\Gamma }(z),\psi ^{(1)}(z)\frac{d}{dz}\psi (z)=\zeta (2,z)$$
(67)
are respectively the Euler gamma, digamma, and trigamma functions. This general formula is often utilized to process the experimental data for the paraconductivity. We provide also several useful asymptotics of $`f_{\mathrm{APS}}(ϵ,h)`$ in different physical conditions,
$$f_{\mathrm{APS}}(ϵ,h)\{\begin{array}{cc}\frac{2}{h}\left[1\frac{ϵ}{2h}\mathrm{ln}2\right],\hfill & h|ϵ|\hfill \\ \frac{4}{ϵ+h}=4\frac{T_\mathrm{c}}{TT_{c2}(H)},\hfill & ϵ+hh\hfill \\ \left[1\frac{1}{2}\left(\frac{h}{ϵ}\right)^2\right]\frac{1}{ϵ}=\left[1\frac{h^2}{4}\frac{^2}{ϵ^2}\right]\frac{1}{ϵ},\hfill & hϵ\hfill \end{array}$$
(68)
For the LD model, for example, the ($`hϵ`$)-asymptotics gives,<sup>?,?,?</sup> according to Eq. (52),
$$\sigma _{ab}(ϵ,h)\stackrel{~}{\tau }_{\mathrm{rel}}\frac{e^2}{16\mathrm{}}\frac{N}{s}\left[\frac{1}{\sqrt{ϵ(ϵ+r)}}\frac{h^2}{2}\frac{ϵ(ϵ+r)+\frac{3}{8}r^2}{[ϵ(ϵ+r)]^{5/2}}\right].$$
(69)
Note that the classical Aslamazov-Larkin result, Eq. (47), is recovered for $`r=0,`$ $`h=0,`$ and $`\stackrel{~}{\tau }_{\mathrm{rel}}=1.`$ In the practical application to layered cuprates, however, we need to take into account the nonlocality effects. In $`\epsilon _{_{\text{}}}`$-approximation to the GL theory we have to subtract the part of the corresponding cutoff area in the 2D momentum space. Thereby, the fluctuation conductivity is given by the difference
$$\sigma _{ab}(ϵ,h;c)=\widehat{𝖢}\sigma _{ab}(ϵ,h)\sigma _{ab}(ϵ,h)\sigma _{ab}(c+ϵ,h)\sigma _{ab}(ϵ,h)\sigma _{ab}(c,h),$$
(70)
where a cutoff operator $`\widehat{𝖢}`$ is introduced, and the approximation is valid for $`ϵc.`$ Similarly, for the magnetization we have the same “cutoff” expression which appears when calculating the truncated sums over the Landau levels, $`_0^{n_c1}=_0^{\mathrm{}}_{n_c}^{\mathrm{}},`$ or integral with respect to the dimensionless in-plane kinetic energy,
$$_0^c𝑑\stackrel{~}{x}=_0^{\mathrm{}}𝑑\stackrel{~}{x}_c^{\mathrm{}}𝑑\stackrel{~}{x}.$$
(71)
As a rule the GL theory allows for ultraviolet (UV) regularization—every expression can be easily regularized in the local ($`c\mathrm{}`$)-approximation. Therefore the energy cutoff parameter $`c`$ is not viewed as a tool for UV regularization, it is simply an important and immanent parameter of the GL theory, being of the order $`c1.`$ The cutoff procedure has been essentially introduced from the beginning in the GL theory.<sup>?</sup> Unfortunately, for many superconductors systematic studies for determination of the energy cutoff parameter are still missing. Here we suggest only the simplest possible interpolation formula within the LD model for $`ϵc`$,
$$\sigma _{ab}(ϵ,h)=\frac{\pi }{8}\frac{\stackrel{~}{\tau }_{\mathrm{rel}}}{R_{\mathrm{QHE}}}\frac{N}{s}\widehat{𝖢}f_{\mathrm{LD}}(ϵ,r)\stackrel{~}{\tau }_{\mathrm{rel}}\frac{e^2}{16\mathrm{}}\frac{N}{s}\left[\frac{1}{\sqrt{ϵ(ϵ+r)}}\frac{1}{\sqrt{c(c+r)}}\right],$$
(72)
which takes into account only the fist nonlocal correction. This simple expression fits very well<sup>?</sup> the experimental data for YBa<sub>2</sub>Cu<sub>3</sub>O<sub>7-δ</sub> and Bi<sub>2</sub>Sr<sub>2</sub>CaCu<sub>2</sub>O$`_8.`$
As a last example of the action of the layering operator $`\widehat{𝖫}`$ we consider the 2D frequency-dependent paraconductivity at zero magnetic field. Taking the general expression for $`D`$-dimensional GL model<sup>?</sup> and performing carefully the limit $`D2`$ (note, that there is an omitted term in the expression for the 2D conductivity in Ref. 41) we get for the in-plane complex conductivity
$$\sigma _{ab}^{}(\omega \tau (ϵ))=\sigma _{ab}^{}\left(\frac{\omega \tau _0}{ϵ}\right)+i\sigma _{ab}^{\prime \prime }\left(\frac{\omega \tau _0}{ϵ}\right)=\stackrel{~}{\tau }_{\mathrm{rel}}\frac{e^2}{16\mathrm{}}\frac{N}{s}\widehat{𝖫}\left[\frac{1}{ϵ}\varsigma _1\left(\frac{\omega \tau _0}{ϵ}\right)+\frac{i}{ϵ}\varsigma _2\left(\frac{\omega \tau _0}{ϵ}\right)\right],$$
(73)
or in expanded notation for singe layered superconductor,
$$\sigma _{ab}^{}(\omega )=\frac{2\tau _0k__\mathrm{B}T_\mathrm{c}/\mathrm{}}{s_{\mathrm{eff}}R_{\mathrm{QHE}}}_0^{\pi /2}\frac{d\varphi }{\pi /2}\frac{\varsigma _1\left(\frac{\omega \tau _0}{ϵ+r\mathrm{sin}^2\varphi }\right)+i\varsigma _2\left(\frac{\omega \tau _0}{ϵ+r\mathrm{sin}^2\varphi }\right)}{ϵ+r\mathrm{sin}^2\varphi },$$
(74)
where we have for the dimensionless real and imaginary conductivity, $`\varsigma _1(0)=1,`$
$`\varsigma _1(z)`$ $`{\displaystyle \frac{2}{z^2}}\left[z\mathrm{arctan}(z){\displaystyle \frac{1}{2}}\mathrm{ln}\left(1+z^2\right)\right]={\displaystyle \frac{2}{\pi }}𝒫{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{y\varsigma _2(y)}{y^2z^2}}𝑑y,`$ (75)
$`\varsigma _2(z)`$ $`{\displaystyle \frac{2}{z^2}}\left[\mathrm{arctan}(z)z+z{\displaystyle \frac{1}{2}}\mathrm{ln}\left(1+z^2\right)\right]={\displaystyle \frac{2z}{\pi }}𝒫{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{\varsigma _1(y)}{y^2z^2}}𝑑y.`$ (76)
As usual in the above Kramers-Kronig integrations $`𝒫`$ indicates that the principal value of the integral is taken. For computer implementation of the $`\widehat{𝖫}`$ operator we have to verify that, for $`r1,`$ $`\widehat{𝖫}`$ is simply equivalent to an incremental operator for the spatial dimensionality,
$$\sigma ^{\text{(D+1)}}(ϵ)\widehat{𝖫}\sigma ^{(\mathrm{D})}(ϵ).$$
(77)
In many cases the GL results for integer dimensionality are well-known and we can derive a generalization for a layered system. For both the MT and LD models the integration in Eq. (73) can be easily programmed, so we have a useful formula for fitting of the ultra high frequency measurements of $`\sigma _{ab}^{}(\omega ).`$ The original explicit expressions derived from retarded electromagnetic operator by Aslamazov and Varlamov<sup>?</sup> are too cumbersome to be used by experimentalists. Hence, one may realize that the GL theory is not some phenomenological alternative to the microscopic BCS theory (this scorn, dating back to the beginning of fifties, is still living even nowadays among students). The GL theory is a tool for applying the theory of superconductivity for the important for applications, let us say “hydrodynamic”, case of low frequencies and small wave-vectors. For $`ϵr`$ the frequency dependent conductivity $`\sigma _{ab}^{}\left(\omega \right)`$, having dimension $`(\mathrm{\Omega }\mathrm{cm})^1`$, from Eq. (74) displays 3D behavior, while in the opposite case of $`ϵr`$ it shows 2D character. For thin films of layered superconductors with thickness $`d_{\mathrm{film}}`$ we have to calculate the 2D conductivity $`\sigma ^{(2\mathrm{D})}=d_{\mathrm{film}}\sigma ,`$ while for single layered films of conventional superconductors, for example, we have to substitute in Eq. (74) formally $`s_{\mathrm{eff}}=d_{\mathrm{film}},`$ and certainly $`r=0.`$
Having analyzed in detail the action of the $`\widehat{𝖫}`$ operator, we developed practically all technical tools necessary to proceed our investigation of the thermodynamics of Gaussian fluctuations and fluctuation magnetization.
### 2.4 Power series for the magnetic moment within the LD model
We will calculate in this subsection the nonlinear susceptibility by substituting first into the free energy, Eq. (28), the heat capacity, expressed via the susceptibility from Eq. (32). Then, the formula for the magnetization, Eq. (29), gives
$$\chi (ϵ,h)=6\underset{n=1}{\overset{\mathrm{}}{}}(1)^{n1}\frac{2n}{\pi ^{2n}}\left(1\frac{1}{2^{2n1}}\right)\zeta (2n)h^{2(n1)}\frac{^{2(n1)}}{ϵ^{2(n1)}}\chi (ϵ).$$
(78)
Taking the LD expression for the susceptibility at zero field Eq. (42), calculating the derivatives with respect to $`ϵ`$ by means of the relation
$$\frac{^m}{ϵ^m}\frac{1}{\sqrt{ϵ}}=\frac{(2m1)!!}{2^mϵ^m}\frac{1}{\sqrt{ϵ}},$$
(79)
and defining the relative susceptibility as
$$\stackrel{~}{\chi }_{\mathrm{rel}}(ϵ,h)\frac{\chi (ϵ,h)}{\chi (ϵ)}$$
(80)
we obtain
$`\stackrel{~}{\chi }_{\mathrm{rel}}(ϵ,h;r)=`$ $`\mathrm{\hspace{0.17em}12}{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{(1)^n}{2n+1}}\left(1{\displaystyle \frac{1}{2^{2n+1}}}\right){\displaystyle \frac{(2n+2)!}{2^{2n+1}}}{\displaystyle \frac{\zeta (2n+2)}{\pi ^{2n+2}}}`$
$`\times \left({\displaystyle \frac{h^2}{ϵ^2}}\right)^n{\displaystyle \underset{m=0}{\overset{2n}{}}}{\displaystyle \frac{(2m1)!!(4n2m1)!!}{m!(2nm)!(1+r/ϵ)^{2nm}}}`$
$`=`$ $`\mathrm{\hspace{0.17em}1}{\displaystyle \frac{7}{15}}{\displaystyle \frac{ϵ^2+rϵ+3r^2/8}{(ϵ+r)^2}}\left({\displaystyle \frac{h}{ϵ}}\right)^2+\mathrm{}.`$ (81)
Although these series is found to be a solution to the problem of calculating the fluctuational magnetization,
$$M(ϵ,h)=\frac{k__\mathrm{B}T_\mathrm{c}}{\mathrm{\Phi }_0s_{\mathrm{eff}}}\left\{\frac{\stackrel{~}{\chi }_{\text{rel}}(ϵ,h;r)}{6\sqrt{ϵ(r+ϵ)}}\frac{\stackrel{~}{\chi }_{\text{rel}}(c+ϵ,h;r)}{6\sqrt{(c+ϵ)(c+r+ϵ)}}\right\},$$
(82)
for the physical conditions of interest, i.e. an observable effect of magnetic field on the susceptibility, one needs to extend the series summation onto arguments $`h^2/ϵ^2`$ beyond the radius of convergence. Analogous series has been already reported for the 3D paraconductivity.<sup>?</sup> One of the best devices for extending the convergence of series and also for calculating slowly convergent series is the $`\epsilon `$-algorithm<sup>?,?</sup> based on Padé approximants.<sup>?</sup> In the next section we describe a simplified version of this algorithm suitable for computer implementation.
### 2.5 The epsilon algorithm
The epsilon algorithm is a method for finding the limit $`L`$ of infinite series
$$L=\underset{N\mathrm{}}{lim}S_N\underset{N\mathrm{}}{lim}\underset{i=0}{\overset{N}{}}a_i,$$
(83)
in case where only the first $`N+1`$ terms $`a_i`$, $`i=0,\mathrm{}N,`$ are known. The algorithm operates by employing two rows. The first one, called here auxiliary $`A`$-row, is initially set to zero, i.e.
$$A_0^{[0]}=0,A_1^{[0]}=0,A_2^{[0]}=0,\mathrm{}A_N^{[0]}=0.$$
(84)
The second one is sequential $`S`$-row loaded in zero-order approximation with the partial sums of the series
$$S_0^{[0]}=a_0,S_1^{[0]}=a_0+a_1,\mathrm{}S_N^{[0]}=a_0+a_1+a_2+\mathrm{}+a_N.$$
(85)
The above assignments, as indicated by Eqs. (84) and (85), constitute the initialization phase of the $`\epsilon `$-algorithm. The essence of the latter consists of filling in the so called $`\epsilon `$-table
$`\left(\begin{array}{ccccc}A_0^{[0]}& A_1^{[0]}& A_2^{[0]}& A_3^{[0]}& \mathrm{}\\ S_0^{[0]}& S_1^{[0]}& S_2^{[0]}& S_3^{[0]}& \mathrm{}\\ A_0^{[1]}& A_1^{[1]}& A_2^{[1]}& A_3^{[1]}& \mathrm{}\\ S_0^{[1]}& S_1^{[1]}& S_2^{[1]}& S_3^{[1]}& \mathrm{}\\ 5\end{array}\right)`$ $`=\left(\begin{array}{cccc}\epsilon _0^{[0]}& \epsilon _1^{[0]}& \epsilon _2^{[0]}& \mathrm{}\\ \epsilon _0^{[1]}& \epsilon _1^{[1]}& \epsilon _2^{[1]}& \mathrm{}\\ \epsilon _0^{[2]}& \epsilon _1^{[2]}& \epsilon _2^{[2]}& \mathrm{}\\ \epsilon _0^{[3]}& \epsilon _1^{[3]}& \epsilon _2^{[3]}& \mathrm{}\\ 4\end{array}\right)`$
$`=\left(\begin{array}{c}4\\ \left[0/0\right]& [1/0]& [2/0]& \mathrm{}\\ 4\\ \left[1/1\right]& [2/1]& [3/1]& \mathrm{}\\ 4\end{array}\right),`$ (86)
where according to the standard notations $`[j/k]=P_j(z)/P_k(z)|_{z=1}`$ designates a Padé approximant having power $`j`$ in the nominator and, respectively, $`k`$ in the denominator.<sup>?</sup>
Starting from the $`A^{[0]}`$\- and $`S^{[0]}`$ rows every subsequent row is derived by applying the cross rule (known also as the missing identity of Frobenius). To be specific, for calculation of the $`k`$th $`A`$-row we have to solve the cross rule equation
$$\left(\begin{array}{cc}\mathrm{}& A_{i+1}^{[k1]}\\ S_i^{[k1]}& S_{i+1}^{[k1]}\\ A_i^{[k]}& \mathrm{}\\ 2\end{array}\right)=\left(\begin{array}{cc}\mathrm{}& \mathrm{North}\\ \mathrm{West}& \mathrm{East}\\ \mathrm{South}& \mathrm{}\\ 2\end{array}\right),$$
$$\left(\mathrm{South}\mathrm{North}\right)\left(\mathrm{East}\mathrm{West}\right)=1.$$
(87)
Likewise, for calculating the $`k`$-th $`S`$ row we have to apply the same cross rule
$$\left(\begin{array}{c}2\\ \mathrm{}& S_{i+1}^{[k1]}\\ A_i^{[k]}& A_{i+1}^{[k]}\\ S_i^{[k]}& \mathrm{}\end{array}\right)=\left(\begin{array}{c}2\\ \mathrm{}& \mathrm{North}\\ \mathrm{West}& \mathrm{East}\\ \mathrm{South}& \mathrm{}\end{array}\right),$$
$$\mathrm{South}=\mathrm{North}+\left(\mathrm{East}\mathrm{West}\right)^1.$$
(88)
Having applied the algorithm we get in the $`S`$-rows of the $`\epsilon `$-table, Eq. (86), a set of different Padé approximants to the limit $`L.`$ The $`i`$th term of the $`k`$th $`A`$-row can be easily obtained by
$$A_i^{[k]}=A_{i+1}^{[k1]}+\left(S_{i+1}^{[k1]}S_i^{[k1]}\right)^1,\text{ for }i=0,1,\mathrm{},N2k+1,$$
(89)
but for practical implementation of the algorithm, we can omit the index of the approximation and to use only one auxiliary row, updating it each time,
$$A_i:=A_{i+1}+\left(S_{i+1}S_i\right)^1,\text{ for }i=0,1,\mathrm{},N2k+1.$$
(90)
For the $`k`$-th $`S`$-row, the $`i`$-th term reads as
$$S_i^{[k]}=S_{i+1}^{[k1]}+\left(A_{i+1}^{[k]}A_i^{[k]}\right)^1,\text{ for }i=0,1,\mathrm{},N2k,$$
(91)
and can be updated in the same manner as described for the $`A`$-row,
$$S_i:=S_{i+1}+\left(A_{i+1}A_i\right)^1\text{ for }i=0,1,\mathrm{},N2k.$$
(92)
In order to find an estimate for the limit $`L`$ of the infinite series, two different empirical criteria can be implemented. In the first one, the $`\epsilon `$-table is scanned for a minimal difference $`|S_{i+1}^{[k1]}S_i^{[k1]}|.`$ The limit $`L`$ is then given by
$$\underset{i,k}{\mathrm{min}}\left|S_{i+1}^{[k1]}S_i^{[k1]}\right|LS_i^{[k1]}.$$
(93)
This minimal difference gives also an estimate for the empirical error of the method. In the second criterion the $`\epsilon `$-table is scanned for the maximum of the East $`A`$-row element, cf Eqs. (88) and (91),
$$\underset{i,k}{\mathrm{max}}\left|A_{i+1}^{[k]}\right|LS_i^{[k]}.$$
(94)
The reciprocal of the maximum auxiliary value gives in this case the estimate for the empirical error of the method. It is the second criterion that we have used in the fortran90 implementation of the $`\epsilon `$-algorithm given in Appendix A. Therein we have also made use of pseudo-inverse numbers in order to ensure provisions against division by zero in Eqs. (90) and (92),
$$z^1:=\{\begin{array}{cc}0,\hfill & \text{ for }z=0\hfill \\ 1/z,\hfill & \text{ for }z0\hfill \end{array}.$$
(95)
For an illustration, consider the first approximation. In the beginning we have for the first $`A`$-row according to Eq. (89)
$$A_0^{[1]}=\left[\left(a_0+a_1\right)(a_0)\right]^1=\frac{1}{a_1},A_1^{[1]}=\frac{1}{a_2},\mathrm{},A_{N1}^{[1]}=\frac{1}{a_N}.$$
(96)
The first $`S`$-row then reads
$$S_0^{[1]}=a_0+a_1+\frac{1}{1/a_21/a_1},S_1^{[1]}=a_0+a_1+a_2+\frac{1}{1/a_31/a_2},\mathrm{},$$
(97)
and for the last element of the $`S^{[1]}`$-row we have
$$S_{N2}^{[1]}=a_0+a_1+a_2+\mathrm{}+a_{N2}+a_{N1}+(1/a_N1/a_{N1}).$$
(98)
The above approximation $`S_{N2}^{[1]}`$ to the limit $`L`$ is nothing but the well-known Aitken’s $`\mathrm{\Delta }^2`$-method, which gives an exact result for the geometric progression
$$S_N^{[0]}=1+q+q^2+\mathrm{}+q^N,S_0^{[1]}=S_1^{[1]}=S_2^{[1]}=\mathrm{}=S_{N2}^{[1]}=\frac{1}{1q},$$
(99)
for an arbitrary $`q1.`$ This fact can rationalize the success of the $`\epsilon `$-algorithm when applied to weak magnetic field series expansion of susceptibility. In the Euler-MacLaurin summation, Eqs. (22) and (26), we have a hidden geometric progression of translation operators.
As a rule divergent series do not exist in physics; 99% of the divergent series born by real physical problems can be summed up by some combination of the Euler-MacLaurin method and the $`\epsilon `$-algorithm and the reason is lies in the analytical dependence of the coefficients on the index. In the Gaussian spectroscopy of superconductors, for example, it is necessary series related to asymptotic expansion of Euler polygamma and Hurwitz zeta functions to be summed up, but the same methods could be applied to many other physical problems. The solution often can be derived by less efforts than required to verify that a series is divergent accordingly some strict mathematical criterion. Nowadays the mathematical education in the physical departments is conquered by scholastic mathematicians. Alas, none of the students of physics knows what really happens when we press the sin key of a calculator. On the other hand this is a commercial secret of the manufacturer. The physicists do not even lightly touch the brilliant achievements of mathematics indispensable not only for the theoretical physics but for experimentalist to fit their data as well. This is the motivation why we, following the spirit of the century of enlightenment, present in Appendix A a simple fortran90 program illustrating the operation of the $`\epsilon `$-algorithm. Certainly, fysics is phun,<sup>?</sup> being in part art cosa mentale,<sup>?</sup> and every new software cannot be foolproof, but there are methods which must be taken into account in every complicated calculation.
### 2.6 Power series for differential susceptibility
Having calculated the relative dimensionless susceptibility by employing the $`\epsilon `$-algorithm we can recover the usual susceptibility from the dimensionless one,
$$\chi (ϵ,h)=\stackrel{~}{\chi }_{\mathrm{rel}}(ϵ,h)\chi (ϵ).$$
(100)
In order to take into account the effects of nonlocality the cutoff area in the momentum space should subtracted out from the susceptibility
$$\chi _{_{\text{}}}(ϵ,h)=\widehat{𝖢}\chi (ϵ,h)=\stackrel{~}{\chi }_{\mathrm{rel}}(ϵ,h)\chi (ϵ)\stackrel{~}{\chi }_{\mathrm{rel}}(c+ϵ,h)\chi (c+ϵ).$$
(101)
Then we can easily find the magnetization
$$M(H,T)=\chi _{_{\text{}}}(ϵ,h)H.$$
(102)
The calculation of the differential susceptibility
$$\chi ^{(\mathrm{dif})}(ϵ,h)=\left(\frac{M}{H}\right)_T,$$
(103)
where $`H=H_{c2}(0)h,`$ gives an alternative method to determine the magnetization. Next we define a dimensionless relative differential susceptibility
$$\stackrel{~}{\kappa }(ϵ,h,r)\chi ^{(\mathrm{dif})}(ϵ,h)/\chi (ϵ).$$
(104)
For this variable, using Eq. (81), we have the series
$`\stackrel{~}{\kappa }`$ $`=12{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\left(1{\displaystyle \frac{1}{2^{2n+1}}}\right){\displaystyle \frac{(2n+2)!}{2^{2n+1}}}{\displaystyle \frac{\zeta (2n+2)}{\pi ^{2n+2}}}\left({\displaystyle \frac{h^2}{ϵ^2}}\right)^n`$
$`\times {\displaystyle \underset{m=0}{\overset{2n}{}}}{\displaystyle \frac{(2m1)!!(4n2m1)!!}{m!(2nm)!(1+r/ϵ)^{2nm}}}`$
$`=1{\displaystyle \frac{7}{5}}{\displaystyle \frac{ϵ^2+rϵ+3r^2/8}{(ϵ+r)^2}}\left({\displaystyle \frac{h}{ϵ}}\right)^2+\mathrm{},`$ (105)
which, just as done in deriving Eq. (81), can be summed up by means of the $`\epsilon `$-algorithm. For instance, in the local GL limit we have for the magnetization
$`M={\displaystyle _0^H}\chi ^{(\mathrm{dif})}(T,H)𝑑H`$ $`=\chi (ϵ)H_{c2}(0){\displaystyle _0^h}\stackrel{~}{\kappa }(ϵ,h^{})𝑑h^{}`$
$`=\chi (ϵ)H_{c2}(0)\stackrel{~}{\chi }_{\mathrm{rel}}(ϵ,h)h.`$ (106)
For the further analysis, however, it is more suitable to introduce a dimensionless magnetization
$$\stackrel{~}{m}\frac{M}{M_0}=\frac{\mathrm{\Phi }_0}{k__\mathrm{B}T_\mathrm{c}}\frac{s}{N}M,M_0\frac{k__\mathrm{B}T_\mathrm{c}}{\mathrm{\Phi }_0}\frac{N}{s}.$$
(107)
Then, using the relation
$$\stackrel{~}{\chi }_{\mathrm{rel}}(ϵ,h)=\frac{1}{h}_0^h\stackrel{~}{\kappa }(ϵ,h^{})𝑑h^{}$$
(108)
the result for the dimensionless fluctuation magnetization takes the form
$$\stackrel{~}{m}(ϵ,h)=\frac{1}{6}\frac{1}{\sqrt{ϵ(ϵ+r)}}\stackrel{~}{\chi }_{\mathrm{rel}}(ϵ,h)h=\frac{1}{6}\frac{1}{\sqrt{ϵ(ϵ+r)}}_0^h\stackrel{~}{\kappa }(ϵ,h^{})𝑑h^{}.$$
(109)
In this section we have calculated the magnetization by means of power series in the magnetic field assuming in the beginning $`h/ϵ1.`$ In the next section we develop another method for calculating the fluctuation magnetic moment which is appropriate for strong magnetic fields and allows for studying the high magnetic field asymptotics for large enough values of the reduced magnetic field, $`h/ϵ1.`$ The overlap between these expansions about $`h/ϵ1`$ would be a test for the accuracy of the calculations.
## 3 Strong magnetic fields
### 3.1 General formula for the free energy
In order to derive a general formula for the Gibbs free energy for arbitrary non-vanishing magnetic field we will start again by representing the free energy density as a sum over the energy spectrum, Eq. (21),
$$F(ϵ,h)=F_0\mathrm{\hspace{0.33em}2}h\widehat{𝖫}\underset{n=0}{\overset{n_c1}{}}\left[\mathrm{ln}\left(n+\frac{1}{2}+\frac{ϵ}{2h}\right)+\mathrm{ln}(2h)\right],$$
(110)
where
$$F_0\frac{k__\mathrm{B}T_\mathrm{c}}{4\pi \xi _{ab}^2(0)}\frac{N}{s}=\frac{1}{2}M_0B_{c2}(0).$$
(111)
The first way to go in deriving convenient for programming formula is to calculate the action of the $`\widehat{𝖫}`$ operator on the integrand, cf. Ref. 48. In this case we write down the free energy as a finite sum over the Landau levels
$$F(ϵ,h)=\frac{k__\mathrm{B}T_\mathrm{c}}{4\pi \xi _{ab}^2(0)}\frac{N}{s}\mathrm{\hspace{0.33em}2}h\underset{n=0}{\overset{n_c1}{}}\widehat{𝖫}\mathrm{ln}\left[ϵ+h(2n+1)\right],$$
(112)
where, according to Eq. (50),
$$\widehat{𝖫}^{(\mathrm{LD})}\mathrm{ln}\left[ϵ+h(2n+1)\right]=2\mathrm{ln}\frac{\sqrt{ϵ+h(2n+1)}+\sqrt{ϵ+h(2n+1)+r}}{2}.$$
(113)
This formula is useful especially in the case of strong magnetic fields when the finite series are not too long. However, in order to have a good working expression, applicable to all cases, it is much better to solve the problem analytically. Towards this end consider the last term in the integrand of Eq. (110). The summation of this constant term and simply yields the cutoff parameter $`c`$
$$2h\underset{n=0}{\overset{n_c1}{}}1=(2h)n_c=c.$$
(114)
Next we introduce a dimensionless function
$$x(ϵ,h)\frac{1}{2}+\frac{ϵ}{2h}=\frac{ϵ+h}{2h}=\frac{1}{2H}\left(TT_{c2}(H)\right)\left(\frac{H_{c2}(T)}{T}\right)|_{T=T_\mathrm{c}0},$$
(115)
which is the argument of some of the analytical functions we use in the following. Further, we have to present the sum in Eq. (110) as a difference of two appropriately regularized infinite series
$$\underset{n=0}{\overset{n_c1}{}}\mathrm{ln}(n+x)=\widehat{\mathrm{𝖱𝖾𝗀}}_\zeta \underset{n=0}{\overset{\mathrm{}}{}}\mathrm{ln}(n+x)\widehat{\mathrm{𝖱𝖾𝗀}}_\zeta \underset{n_c}{\overset{\mathrm{}}{}}\mathrm{ln}(n+x).$$
(116)
In fact, one does not have any other possibility except the $`\zeta `$-regularization
$$\widehat{\mathrm{𝖱𝖾𝗀}}_\zeta \underset{n=0}{\overset{\mathrm{}}{}}\mathrm{ln}(n+z)=\frac{}{\nu }\zeta (\nu ,z)|_{\nu =0}=\mathrm{ln}\frac{\mathrm{\Gamma }(z)}{\sqrt{2\pi }},$$
(117)
based on one relation between the Euler $`\mathrm{\Gamma }`$-function and the Hurwitz $`\zeta `$-function, and the definition of the logarithmic function
$$\mathrm{ln}z=\underset{\nu 0}{lim}\frac{z^\nu 1}{\nu }=\frac{}{\nu }z^\nu |_{\nu =0}.$$
(118)
According to the famous results by Riemann, the analytical continuations of the $`\zeta `$-function and the factorial $`n!`$ are unique. Therefore the UV regularization of the partition function in the GL model in an external magnetic field is practically included in the second, Gauss definition of the $`\mathrm{\Gamma }`$-function as a infinite product, see e.g. Ref. 49,
$$_0^{\mathrm{}}t^{z1}\mathrm{e}^t𝑑t\mathrm{\Gamma }(z)\underset{n_c\mathrm{}}{lim}\frac{n_c!n_c^{z1}}{z(z+1)(z+2)\mathrm{}(z+n_c1)}.$$
(119)
Let us recall some particular values,
$$\mathrm{\Gamma }(n+1)=n!,\mathrm{\Gamma }(1)=0!=1,\mathrm{\Gamma }\left(\frac{1}{2}\right)=\sqrt{\pi },$$
(120)
and the Stirling’s approximation for $`n_c1,`$ derived by Gaussian saddle point approximation applied to the first, Euler definition of the $`\mathrm{\Gamma }`$-function, Eq. (119),
$$n_c!\left(\frac{n_c}{\mathrm{e}}\right)^{n_c}\sqrt{2\pi n_c},\mathrm{ln}\left(n_c!\right)\left(n_c+\frac{1}{2}\right)\mathrm{ln}n_cn_c+\mathrm{ln}\sqrt{2\pi }.$$
(121)
For the local limit or for the case of weak magnetic fields we shall also make use of the asymptotic formulae for $`z1`$
$`\mathrm{ln}\mathrm{\Gamma }(z)`$ $`\left(z{\displaystyle \frac{1}{2}}\right)\mathrm{ln}zz+{\displaystyle \frac{1}{2}}\mathrm{ln}(2\pi )+{\displaystyle \frac{1}{12z}},`$ (122)
$`\psi ^{(1)}(z)`$ $`\left(z{\displaystyle \frac{1}{2}}\right)\mathrm{ln}zz+{\displaystyle \frac{1}{12z}},`$ (123)
$`\psi (z)`$ $`\mathrm{ln}z{\displaystyle \frac{1}{2z}}{\displaystyle \frac{1}{12z^2}},\psi ^{(1)}(z)\zeta (2,z){\displaystyle \frac{1}{z}}+{\displaystyle \frac{1}{2z^2}}+{\displaystyle \frac{1}{6z^3}}.`$ (124)
Substituting the Stirling asymptotics in the second Gauss definition, Eq. (119), and taking a logarithm we arrive at the function $`\psi ^{(1)}(z)`$, generating the polygamma functions
$`\psi ^{(1)}(z)`$ $`\underset{n_c\mathrm{}}{lim}\left\{{\displaystyle \underset{n=0}{\overset{n_c1}{}}}\mathrm{ln}(n+z)+\left(n_c{\displaystyle \frac{1}{2}}+z\right)\mathrm{ln}\left(n_c\right)n_c\right\}`$
$`=\mathrm{ln}{\displaystyle \frac{\mathrm{\Gamma }(z)}{\sqrt{2\pi }}}.`$ (125)
As a result, the above Gauss definition for $`\mathrm{ln}\mathrm{\Gamma }(z)`$ solves the problem for UV regularization of the infinite sum of logarithms, Eq. (117). The first derivative of this equation gives the well-known definition of the digamma function $`\psi (z)\psi ^{(0)}(z),`$
$$\psi ^{(0)}(z)\frac{d}{dz}\psi ^{(1)}(z)=\widehat{\mathrm{𝖱𝖾𝗀}}_\zeta \underset{n=0}{\overset{\mathrm{}}{}}\frac{1}{n+z}=\underset{n_c\mathrm{}}{lim}\left\{\underset{n=0}{\overset{n_c1}{}}\frac{1}{n+z}+\mathrm{ln}\left(n_c\right)\right\}.$$
(126)
In particular,
$$\psi (1)=C_{\mathrm{Euler}}=\underset{n_c\mathrm{}}{lim}\left\{\underset{n=1}{\overset{n_c1}{}}\frac{1}{n}\mathrm{ln}\left(n_c\right)\right\}=0.577216\mathrm{}.$$
(127)
All other polygamma functions are actually Hurwitz $`\zeta `$-functions with integer first argument $`2`$ and the sums are trivially convergent,
$$\psi ^{(N)}(z)=\frac{d^N}{dz^N}\psi (z)=(1)^NN!\zeta (N+1,z).$$
(128)
To summarize, we have applied the well known $`\zeta `$-technique<sup>?</sup> for UV regularization of the partition function and revealed that the archetype of this powerful method comes from the century of enlightenment and finally we can bring the free energy, Eq. (110), to the form
$`F(ϵ,h)`$ $`={\displaystyle \frac{T}{T_\mathrm{c}}}F_0\left\{2h\widehat{𝖫}\left[\mathrm{ln}\mathrm{\Gamma }\left({\displaystyle \frac{1}{2}}+{\displaystyle \frac{ϵ}{2h}}\right)+\mathrm{ln}\mathrm{\Gamma }\left({\displaystyle \frac{1}{2}}+{\displaystyle \frac{ϵ+c}{2h}}\right)\right]+c\mathrm{ln}(2h)\right\}`$
$`={\displaystyle \frac{k__\mathrm{B}T}{4\pi \xi _{ab}^2(0)}}{\displaystyle \frac{N}{s}}\widehat{𝖫}\widehat{𝖢}\left[(2h)\mathrm{ln}{\displaystyle \frac{\mathrm{\Gamma }\left(\frac{ϵ+h}{2h}\right)}{\sqrt{2\pi }}}ϵ\mathrm{ln}(2h)\right].`$ (129)
For weak magnetic field, $`hϵ,`$ cf. Eqs. (49), (50), (123), and (143),
$$(2h)\mathrm{ln}\frac{\mathrm{\Gamma }\left(\frac{ϵ+h}{2h}\right)}{\sqrt{2\pi }}ϵ\mathrm{ln}(2h)ϵ\left[\mathrm{ln}(ϵ)1\right]+\frac{1}{6}\frac{h^2}{ϵ}.$$
(130)
This is our main analytical result and all thermodynamic properties now can be obtained via derivatives. However, having this analytical result it is trivially to check that it can be derived by finite sums. The latter do not require UV regularization and the Euler $`\mathrm{\Gamma }`$-function is commonly available in many textbooks on mathematical analysis.
### 3.2 Fluctuation part of thermodynamic variables
Having an analytical result for the free energy we can easily find other thermodynamic variables by differentiating. The magnetization, for example, is given by the derivative
$$M=\left(\frac{F}{B}\right)_T=\frac{1}{B_{c2}(0)}\left(\frac{F}{h}\right)_ϵ=M_0\stackrel{~}{m},$$
(131)
where a dimensionless diamagnetic moment is introduced
$`\stackrel{~}{m}(ϵ,h){\displaystyle \frac{M}{M_0}}`$ $`={\displaystyle \frac{c}{2h}}\widehat{𝖫}\left[\mathrm{ln}\mathrm{\Gamma }\left({\displaystyle \frac{ϵ+h}{2h}}\right)\mathrm{ln}\mathrm{\Gamma }\left({\displaystyle \frac{ϵ+c+h}{2h}}\right)\right]`$
$`+\widehat{𝖫}\left[{\displaystyle \frac{ϵ}{2h}}\psi \left({\displaystyle \frac{ϵ+h}{2h}}\right){\displaystyle \frac{ϵ+c}{2h}}\psi \left({\displaystyle \frac{ϵ+h+c}{2h}}\right)\right].`$ (132)
In expanded notations within the LD model this formula, according to Eqs. (35), (37), (39), (43), and (46), reads as
$`M^{(\mathrm{LD})}(ϵ,h)=`$ $`{\displaystyle \frac{k__\mathrm{B}T_\mathrm{c}}{\mathrm{\Phi }_0s_{\mathrm{eff}}}}({\displaystyle \frac{c}{2h}}+{\displaystyle \frac{2}{\pi }}{\displaystyle _0^{\frac{\pi }{2}}}d\varphi \{[\mathrm{ln}\mathrm{\Gamma }\left({\displaystyle \frac{ϵ+r\mathrm{sin}^2\varphi +h}{2h}}\right)`$
$`\mathrm{ln}\mathrm{\Gamma }\left({\displaystyle \frac{c+ϵ+r\mathrm{sin}^2\varphi +h}{2h}}\right)]+[{\displaystyle \frac{ϵ+r\mathrm{sin}^2\varphi }{2h}}\psi \left({\displaystyle \frac{ϵ+r\mathrm{sin}^2\varphi +h}{2h}}\right)`$
$`{\displaystyle \frac{c+ϵ+r\mathrm{sin}^2\varphi }{2h}}\psi \left({\displaystyle \frac{c+ϵ+r\mathrm{sin}^2\varphi +h}{2h}}\right)]\left\}\right),`$ (133)
where $`\varphi =\frac{1}{2}\theta .`$ For $`|ϵ|,hr,c`$ this general expression recovers the local 3D result, Eq. (180), analyzed later in Sec. 3.4, while in the opposite case of extremely high anisotropy $`r<|ϵ|,hc`$ we get the local 2D result, Eq. (142). Here we want to emphasize the existence to mention a universal magnetization law at $`T=T_\mathrm{c},`$ or $`ϵ=0,`$ which can be observed for many high-$`T_\mathrm{c}`$ materials at strong magnetic fields $`hr`$
$$M(T_\mathrm{c},B)\frac{\mathrm{\Phi }_0s_{\mathrm{eff}}}{k__\mathrm{B}T_\mathrm{c}}=\stackrel{~}{m}=\frac{1}{2}\mathrm{ln}2U_M\left(\frac{2}{c}\frac{B}{B_{c2}(0)}\right),$$
(134)
where the universal function of the nonlocal magnetization
$$U_M(y)\frac{2}{\mathrm{ln}2}\left\{\mathrm{ln}\mathrm{\Gamma }\left(\frac{1}{y}+\frac{1}{2}\right)\frac{1}{2}\mathrm{ln}\pi +\frac{1}{y}\left[1\psi \left(\frac{1}{y}+\frac{1}{2}\right)\right]\right\}$$
(135)
is normalized so that $`U_M(0)=1,`$ $`U_M(\mathrm{})=0,`$ $`y2h/c.`$ For conventional bulk superconductors the nonlocality effects on magnetization are well understood, see for example Refs. 51, 52, 53, 54, 55, 56. To the best of our knowledge, the first observation of fluctuation-induced diamagnetism for a cuprate superconductor well inside the finite-magnetic-field regime was reported by Carretta et al.<sup>?</sup> for YBa<sub>2</sub>Cu<sub>3</sub>O$`_{6+x}.`$ Soon after, analogous measurement was reported for La<sub>1.9</sub>Sr<sub>0.1</sub>CuO<sub>4</sub> by Carballeira et al.<sup>?</sup> Being familiar with the preliminary version of the present review (cf. Ref. 59) Carballeira et al. have entirely based their interpretation and theoretical analysis on Eq. (133) and Eq. (142) below. Alas, we find it very disappointing and impolite that the authors of Ref. 58 do not give any credits (e.g. in the author list, acknowledgments, or references section) to the author (the first author of the present review, T. M.) of the theory they have used. We will not discuss in any detail their attitude and would instead refer to the Comment.<sup>?</sup>
Returning now to the general expression for the magnetization, Eq. (132), we derive another expression for the relative differential susceptibility based on Eq. (109)
$`\kappa (ϵ,h)=`$ $`\mathrm{\hspace{0.17em}6}\sqrt{ϵ(ϵ+r)}\left({\displaystyle \frac{\stackrel{~}{m}}{h}}\right)`$
$`=`$ $`\mathrm{\hspace{0.17em}6}\sqrt{ϵ(ϵ+r)}\widehat{𝖫}[{\displaystyle \frac{c}{2h^2}}{\displaystyle \frac{ϵ^2}{4h^3}}\psi ^{(1)}\left({\displaystyle \frac{ϵ+h}{2h}}\right)`$
$`+{\displaystyle \frac{\left(ϵ+c\right)^2}{4h^3}}\psi ^{(1)}\left({\displaystyle \frac{ϵ+h+c}{2h}}\right)].`$ (136)
The comparison of this result with Eq. (105) is one of the best methods to check the accuracy of the programmed formulae. Analogously, differentiating the free energy with respect to the temperature $`T=(1+ϵ)T_\mathrm{c}`$ we derive the general formula for the most singular part of the entropy (neglecting the derivative of the $`T`$-prefactor in Eq. (129)),
$$S\frac{1}{T_\mathrm{c}}\frac{F}{ϵ}=\frac{k__\mathrm{B}}{4\pi \xi _{ab}^2(0)}\frac{N}{s}\widehat{𝖫}\left[\psi \left(\frac{ϵ+h}{2h}\right)\psi \left(\frac{ϵ+h+c}{2h}\right)\right],$$
(137)
and the most singular part of the heat capacity
$`C(ϵ,h)`$ $`={\displaystyle \frac{S}{ϵ}}={\displaystyle \frac{1}{T_\mathrm{c}}}{\displaystyle \frac{^2F}{ϵ^2}}`$
$`={\displaystyle \frac{k__\mathrm{B}}{4\pi \xi _{ab}^2(0)}}{\displaystyle \frac{N}{s}}{\displaystyle \frac{1}{2h}}\widehat{𝖫}\left[\psi ^{(1)}\left({\displaystyle \frac{ϵ+h}{2h}}\right)\psi ^{(1)}\left({\displaystyle \frac{ϵ+h+c}{2h}}\right)\right].`$ (138)
This expression for $`C`$ can be directly derived from the starting formulae (21) and (110). The sums for the heat capacity are convergent, cf. Ref. 40, and do not require any regularization. The simplest way to reproduce the analytical result for the free energy density, Eq. (129), is to integrate two times the result for its second derivative, i.e. that for the heat capacity, cf. Ref. 66. In general finite sums from $`0`$ to $`n_c1`$ for logarithms and powers can be found in many textbooks on mathematics and all our results can thus be easily checked even by experimentalists.
The fluctuation part of the entropy $`S`$ is proportional to the mean square of the order parameter $`\mathrm{\Psi }`$, i.e. the volume density of fluctuation Cooper pairs. The thermally averaged density deserves a special attention because it is the main ingredient of the self-consistent treatment of the interaction of order parameter fluctuations. This Hartree type approximation due to Ullah and Dorsey<sup>?</sup> will be briefly described in the next subsection.
In the following, for completeness, we will derive the local 2D asymptotics applicable for $`|ϵ|,hc.`$ The substitution of the first term from Eq. (123) into the general formula for the free energy, Eq. (129), gives
$$\stackrel{~}{f}_{2\mathrm{D}}\frac{F(ϵ,h)}{F_0}=(2h)\psi ^{(1)}\left(\frac{ϵ+h}{2h}\right)ϵ\mathrm{ln}(2h)+A(c)ϵ+B(c)+O(1/c),$$
(139)
where for $`ϵc`$, cf. Eq. (12),
$$f_c(ϵ)A(c)ϵ+B(c)_0^{c+ϵ}\mathrm{ln}\stackrel{~}{x}d\stackrel{~}{x}ϵ\mathrm{ln}c+c(\mathrm{ln}c1).$$
(140)
This irrelevant for the fluctuation phenomena linear function of $`ϵ`$ gives constant additions to the free energy $`F_c=F_0B(c),`$ and entropy $`S_c=F_0A(c)/T_\mathrm{c}`$ and can be omitted hereafter. The subtraction of $`F_0f_c`$ from the free energy, Eq. (129), can be considered as a cutoff procedure for UV regularization,
$$\widehat{\mathrm{𝖱𝖾𝗀}}_{\text{}}F(ϵ,h)=F(ϵ,h)(F_cT_\mathrm{c}S_cϵ),$$
(141)
which, when applied, allows the analysis of the local GL approximation to be carried out simply as $`(c\mathrm{})`$-limit. Now a trivial differentiation gives for the dimensionless magnetization, being a positive quantity,
$`\stackrel{~}{m}_{2\mathrm{D}}(ϵ,h)={\displaystyle \frac{M(ϵ,h)}{M_0}}`$ $`={\displaystyle \frac{ϵ}{2h}}\left[\psi \left({\displaystyle \frac{ϵ}{2h}}+{\displaystyle \frac{1}{2}}\right)1\right]\psi ^{(1)}\left({\displaystyle \frac{ϵ+h}{2h}}\right)`$
$`={\displaystyle \frac{1}{2}}{\displaystyle \frac{\stackrel{~}{f}_{2\mathrm{D}}}{h}}`$ (142)
This result is also a local $`(c\mathrm{})`$-asymptotic of Eq. (132), which for $`hϵ`$ yields
$$\stackrel{~}{m}_{2\mathrm{D}}h/6ϵ.$$
(143)
In the general case the local approximation gives $`\stackrel{~}{m}=\widehat{𝖫}\stackrel{~}{m}_{2\mathrm{D}},`$ or for the LD model
$`\stackrel{~}{m}(ϵ,h;r)={\displaystyle \frac{M}{M_0}}=`$ $`{\displaystyle _0^{\pi /2}}{\displaystyle \frac{d\varphi }{\pi /2}}\{{\displaystyle \frac{ϵ+r\mathrm{sin}^2\varphi }{2h}}[\psi ({\displaystyle \frac{ϵ+r\mathrm{sin}^2\varphi }{2h}}+{\displaystyle \frac{1}{2}})1]`$
$`\mathrm{ln}\mathrm{\Gamma }({\displaystyle \frac{ϵ+r\mathrm{sin}^2\varphi }{2h}}+{\displaystyle \frac{1}{2}})+{\displaystyle \frac{1}{2}}\mathrm{ln}(2\pi )\}.`$ (144)
The next differentiation with respect to the magnetic field, using Eqs. (42), (103), gives the relative dimensionless susceptibility
$$\stackrel{~}{\kappa }_{2\mathrm{D}}(ϵ,h)=6ϵ\left(\frac{\stackrel{~}{m}}{h}\right)=12\left(\frac{ϵ}{2h}\right)^2\left[1\frac{ϵ}{2h}\psi ^{(1)}\left(\frac{ϵ}{2h}+\frac{1}{2}\right)\right]=\frac{\chi ^{(\mathrm{dif})}(ϵ,h)}{\chi (ϵ)},$$
(145)
which is also a local $`ch,|ϵ|`$ asymptotic of Eq. (136). For the LD model after averaging with respect to the Josephson phase, according to Eq. (39), we obtain
$$\kappa (ϵ,h;r)=12_0^{\pi /2}\frac{d\varphi }{\pi /2}\left(\frac{ϵ+r\mathrm{sin}^2\varphi }{2h}\right)^2\left[1\frac{ϵ+r\mathrm{sin}^2\varphi }{2h}\zeta (2,\frac{ϵ+r\mathrm{sin}^2\varphi }{2h})\right].$$
(146)
This final result can be directly compared to low field series expansion Eq. (105). Similar differentiations of the free energy, Eq. (139), with respect to the temperatures gives the most singular part of the entropy
$$\stackrel{~}{s}_{2\mathrm{D}}\frac{}{ϵ}\stackrel{~}{f}_{2\mathrm{D}}=\left[\psi \left(\frac{ϵ+h}{2h}\right)+\mathrm{ln}(2h)\right]=T_\mathrm{c}S(ϵ,h)/F_0,$$
(147)
and of the heat capacity
$$\widehat{𝖫}\stackrel{~}{c}_{2\mathrm{D}}\widehat{𝖫}\frac{^2}{ϵ^2}\stackrel{~}{f}_{2\mathrm{D}}(ϵ,h)=\frac{1}{2h}\psi ^{(1)}\left(\frac{ϵ+h}{2h}\right)=T_\mathrm{c}C(ϵ,h)/F_0.$$
(148)
Restoring the $`T`$ prefactor instead of $`T_\mathrm{c}`$ in Eq. (12), as was done in Eq. (129), we arrive at a slightly different expression for the fluctuation part of the free energy $`F=F_0(1+ϵ)\stackrel{~}{f}_{2\mathrm{D}}(ϵ,h)`$ and the heat capacity
$`\widehat{𝖫}\stackrel{~}{c}_{2\mathrm{D}}`$ $`=(1+ϵ){\displaystyle \frac{^2}{ϵ^2}}(1+ϵ)\widehat{𝖫}\stackrel{~}{f}_{2\mathrm{D}}(ϵ,h)`$
$`=(1+ϵ)\left[(1+ϵ)\widehat{𝖫}\stackrel{~}{c}_{2\mathrm{D}}+2\widehat{𝖫}\stackrel{~}{s}_{2\mathrm{D}}\right]={\displaystyle \frac{T_\mathrm{c}C(ϵ,h)}{F_0}},`$ (149)
which gives
$$C(ϵ,h)=\frac{(1+ϵ)k__\mathrm{B}}{4\pi \xi _{ab}^2(0)}\frac{N}{s}\widehat{𝖫}\left[\frac{1+ϵ}{2h}\psi ^{(1)}\left(\frac{ϵ+h}{2h}\right)+2\psi ^{(0)}\left(\frac{ϵ+h}{2h}\right)+2\mathrm{ln}(2h)\right].$$
(150)
For zero magnetic field we have
$$C(ϵ,h=0)=\frac{(1+ϵ)k__\mathrm{B}}{4\pi \xi _{ab}^2(0)}\frac{N}{s}\left[(1+ϵ)\widehat{𝖫}\frac{1}{ϵ}2ϵ\widehat{𝖫}\mathrm{ln}\frac{1}{ϵ}\right],$$
(151)
which in the LD model takes the form
$$C(ϵ,r)=\frac{(1+ϵ)k__\mathrm{B}}{4\pi \xi _{ab}^2(0)}\frac{N}{s}\left[\frac{(1+ϵ)}{\sqrt{ϵ(r+ϵ)}}2ϵ\mathrm{\hspace{0.33em}2}\mathrm{ln}\frac{2}{\sqrt{ϵ}+\sqrt{r+ϵ}}\right].$$
(152)
These expression differs from Eqs. (41), and (60). However, the $`(1+ϵ)^21+2ϵ`$ correction and the less singular part of the heat capacity $`2(1+ϵ)F_0\widehat{𝖫}\stackrel{~}{s}_{2\mathrm{D}}/T_\mathrm{c},`$ which appears due to differentiation of $`T`$ in the numerator of Eq. (12) and Eq. (129), are difficult to be identified experimentally.
For the superconducting phase below the critical temperature, $`0<ϵ1,`$ one has to take into account more or less space homogeneous order parameter $`\mathrm{\Psi }_ϵ`$ which minimizes the nongradient part of the free energy density $`F=a(ϵ)n_ϵ+\frac{1}{2}n_ϵ^2,`$
$$\mathrm{\Psi }_ϵ=\sqrt{a_0(ϵ)/b},n_ϵ=\mathrm{\Psi }_ϵ^2=a_0(ϵ)/b.$$
(153)
The fluctuations around this minimum
$$\mathrm{\Psi }=\mathrm{\Psi }_ϵ+\mathrm{\Psi }^{}+i\mathrm{\Psi }^{\prime \prime },n=\mathrm{\Psi }^2=n_ϵ+2\mathrm{\Psi }_ϵ\mathrm{\Psi }^{}+\left(\mathrm{\Psi }^{}\right)^2+\left(\mathrm{\Psi }^{\prime \prime }\right)^2$$
(154)
should be considered as a small perturbation, thus only the quadratic term in the free energy is taken into account,
$$F(ϵ<0)=a(ϵ)+\frac{1}{2}bn^2\frac{1}{2b}a_0^2ϵ^2+a_0(2ϵ)\left[1\left(\mathrm{\Psi }^{}\right)^2+0\left(\mathrm{\Psi }^{\prime \prime }\right)^2\right].$$
(155)
The first term in this equation corresponds to the jump in the heat capacity $`\mathrm{\Delta }C=a_0^2/bT_\mathrm{c}`$ at $`T_\mathrm{c}.`$ The linear term $`\mathrm{\Psi }^{}`$ simply cancels. The phase fluctuations $`\left(\mathrm{\Psi }^{\prime \prime }\right)^2`$ are coupled to the plasmons and vortexes but they are irrelevant for the thermodynamic fluctuations significantly below $`T_\mathrm{c}.`$ In this way mainly fluctuations related to the modulus of the order parameter are essential for the heat capacity below $`T_\mathrm{c}.`$ Finally, the comparison of the second term $`\mathrm{\Psi }^{}`$ in Eq. (155) with the corresponding expression above $`T_\mathrm{c}`$
$$F(ϵ>0)=a(ϵ)+\frac{1}{2}bn^2a_0(ϵ)\left[\left(\mathrm{\Psi }^{}\right)^2+\left(\mathrm{\Psi }^{\prime \prime }\right)^2\right],$$
(156)
provides a prescription to derive the fluctuation part below $`T_\mathrm{c}`$ from the fluctuation expression for the normal phase above $`T_\mathrm{c}`$
$$\frac{1}{2}\widehat{𝖫}\frac{1}{(2ϵ)}\widehat{𝖫}\frac{1}{ϵ}.$$
(157)
Applying this prescription to Eq. (152) results in the following expression
$$C(ϵ<0,r)=\frac{(1+ϵ)k__\mathrm{B}}{4\pi \xi _{ab}^2(0)}\frac{N}{s}\frac{1}{2}\left[\frac{(1+ϵ)}{\sqrt{(2ϵ)(r2ϵ)}}2ϵ\mathrm{\hspace{0.33em}2}\mathrm{ln}\frac{2}{\sqrt{(2ϵ)}+\sqrt{r2ϵ}}\right].$$
(158)
These fluctuation part as well as the the phonon heat capacity should be subtracted from the experimental data in order to extract the jump $`\mathrm{\Delta }C`$ and related to it penetration depth $`\lambda _{ab}(0).`$ Such a procedure, in fact, gives a purely thermodynamic method to determine the latter quantity.
The dimensionless functions Eqs. (139), (142), (145), and (148)) derived with the local approximation are just as important for the thermodynamics of the layered superconductors as is the APS function for the paraconductivity, Eq. (66). The operator $`\widehat{𝖫}`$ gives the possibility to extend the 2D analytical result for layered or even isotropic 3D superconductor. Additionally the $`\widehat{𝖢}`$ operator gives the energy cutoff approximation for the nonlocality effects in the conducting CuO<sub>2</sub> planes. Therefore the analytical 2D result plays a key role for the fluctuation phenomena in layered superconductors.
We will finish the analysis of the local $`c|ϵ|,h`$ 2D approximation $`hr,`$ i.e.
$$\mu _0HrB_{c2}(0)=\left(\frac{2\xi _c(0)N}{s}\right)^2\frac{\mathrm{\Phi }_0}{2\pi \xi _{ab}^2(0)},$$
(159)
with the important case of strong magnetic field $`h|ϵ|.`$ Under these conditions ($`|ϵ|,rhc`$) the layered superconductors display a magnetization corresponding to the local 2D one in strong magnetic fields. The substitution of $`ϵ=0`$ in Eq. (142), using Eq. (120), recovers the result by Klemm, Beasley, and Luther<sup>?</sup>
$$\stackrel{~}{m}(hr,ϵh)0.3465735902799726\mathrm{},M\frac{\mathrm{ln}2}{2}\frac{k__\mathrm{B}T_\mathrm{c}}{\mathrm{\Phi }_0}\frac{N}{s}.$$
(160)
In the concise review by Koshelev<sup>?</sup> on the properties of 2D GL model the calculation of $`\frac{1}{2}\mathrm{ln}20.346`$ by infinite series with three decimal digits accuracy is described in great details.
### 3.3 Self-consistent approximation for the LD model
The bulk (3D) density of the fluctuation Cooper pairs $`n(ϵ,h)`$ can be calculated from the general expression for the Gibbs free energy Eqs. (8) and (12). The differentiation with respect of the ”chemical potential” of Cooper pairs $`\mu _{\mathrm{CP}}=a_0ϵ,`$ according to the relation $`N_{\mathrm{CP}}=(G/\mu _{\mathrm{CP}})_{T,H},`$ gives
$$n(ϵ,h)=\frac{N}{s}\left|\mathrm{\Psi }_n\right|^2=\frac{1}{a_0}\frac{}{ϵ}F(ϵ,h)=\frac{k__\mathrm{B}T_\mathrm{c}}{a_0}S(ϵ,h;c).$$
(161)
This formula can be alternatively derived by summation of the Rayleigh-Jeans asymptotics of the energy distribution of the fluctuation Cooper pairs
$$n(ϵ,h)=\frac{1}{V}\underset{𝐩,p_z,j}{}\frac{k__\mathrm{B}T}{\epsilon _j(\text{p},p_z)+a}=\frac{N}{s}\widehat{𝖫}^{(\mathrm{LD})}\underset{|𝐩|<p_c}{}\frac{d(\pi p^2)}{(2\pi \mathrm{})^2}\frac{k__\mathrm{B}T_\mathrm{c}}{p^2/2m_{ab}+a_0ϵ},$$
(162)
see for example the monograph by Patashinskii and Pokrovsky.<sup>?</sup>
Let us give an illustration for zero magnetic field. In this case for the density of fluctuation Cooper pairs, using Eq. (111) and Eq. (50), we obtain
$$n(ϵ,0)=\frac{F_0}{a_0}\mathrm{\hspace{0.33em}2}\mathrm{ln}\frac{\sqrt{c+ϵ}+\sqrt{c+r+ϵ}}{\sqrt{ϵ}+\sqrt{ϵ+r}}.$$
(163)
This formula sets the stage for the self-consistent treatment of the order parameter fluctuations in the LD model in which the nonlinear term is replaced by its average. The idea has its origin in the Maxwell consideration of the ring of Saturn; probably it is the first work on collective phenomena in physics. Having no possibility to consider motion of all particles in detail we must search for some approximation. Within a self-consistent picture, the motion of every particle creates an average potential in which the others are moving. From the dust of the ring of Saturn to the Cooper pairs in cuprates the idea is the same, only the mechanics slightly changes. In the self-consistent approximation the nonlinear term in GL equations gives an addendum to the linear one
$$a_{\mathrm{ren}}(ϵ,h)=a_0ϵ+bn(\frac{a_{\mathrm{ren}}}{a_0},h),$$
(164)
where the coefficient $`b=\stackrel{~}{b}N/s`$ can be expressed via the jump of the heat capacity $`\mathrm{\Delta }C`$ at the phase transition or, which is more convenient for the high-$`T_\mathrm{c}`$ cuprates, via the extrapolated to zero temperature penetration depth $`1/\lambda _{ab}^2(T)=\mu _0n(T)e^2/m_{ab},`$ $`n(T)=a(T)/b,`$
$$b=\frac{a_0^2}{T_\mathrm{c}\mathrm{\Delta }C}=2\mu _0\left(\frac{\pi \mathrm{}^2\kappa _{_{\mathrm{GL}}}}{\mathrm{\Phi }_0m_{ab}}\right)^2,T_\mathrm{c}\mathrm{\Delta }C=\frac{1}{8\pi ^2\mu _0}\left(\frac{\mathrm{\Phi }_0}{\lambda _{ab}(0)\xi _{ab}(0)}\right)^2,$$
(165)
where $`\kappa _{_{\mathrm{GL}}}\lambda _{ab}(0)/\xi _{ab}(0)`$ is the GL parameter. One can easily check that Eq. (165) has the same form in Gaussian units, where $`\mu _0^{(\mathrm{Gauss})}=4\pi .`$ Introducing the renormalized reduced temperature $`ϵ_{\mathrm{ren}}>0`$ for the normal phase we have the self-consistent equation
$$ϵ_{\mathrm{ren}}=\mathrm{ln}\frac{T}{T_\mathrm{c}}+\frac{b}{a_0}n(ϵ_{\mathrm{ren}},h),$$
(166)
where $`n(ϵ,h)`$ is calculated by means of Gaussian saddle point approximation.<sup>?</sup> For the LD model this equation, by virtue of Eq. (163), takes the form
$$ϵ_{\mathrm{ren}}=\mathrm{ln}\frac{T}{T_\mathrm{c}}+ϵ_{_{\mathrm{Gi}}}\mathrm{\hspace{0.33em}2}\mathrm{ln}\frac{\sqrt{c+ϵ_{\mathrm{ren}}}+\sqrt{c+ϵ_{\mathrm{ren}}+r}}{\sqrt{ϵ_{\mathrm{ren}}}+\sqrt{ϵ_{\mathrm{ren}}+r}}=ϵ+ϵ_{_{\mathrm{Gi}}}\widehat{𝖫}^{(\mathrm{LD})}\mathrm{ln}\frac{c+ϵ_{\mathrm{ren}}}{ϵ_{\mathrm{ren}}},$$
(167)
where the dimensionless parameter
$$ϵ_{_{\mathrm{Gi}}}\frac{bF_0}{a_0^2}=2\pi \mu _0\frac{N}{s}\left(\frac{\lambda _{ab}(0)}{\mathrm{\Phi }_0}\right)^2k__\mathrm{B}T_\mathrm{c}=\frac{1}{4\pi \xi _{ab}^2(0)}\frac{N}{s}\frac{k__\mathrm{B}}{\mathrm{\Delta }C}$$
(168)
is closely related to the Ginzburg number; cf. Eq. (60) which now reads
$$\frac{C(ϵ)}{\mathrm{\Delta }C}=ϵ_{_{\mathrm{Gi}}}\widehat{𝖫}\widehat{𝖢}\frac{1}{ϵ},$$
(169)
and the review article by Varlamov et al.<sup>?</sup> At $`T_\mathrm{c},`$ for $`ϵ_{_{\mathrm{Gi}}}rc,`$ Eq. (167) gives
$$ϵ_{\mathrm{ren},c}ϵ_{_{\mathrm{Gi}}}\mathrm{ln}\frac{4c}{r}$$
(170)
and the effective heating $`\mathrm{\Delta }T=T_\mathrm{c}ϵ_{\mathrm{ren},c}`$ constrains the fluctuation variables at $`T_\mathrm{c}.`$ To provide an order estimate we take for illustration $`s_{\mathrm{eff}}=1`$ nm, $`\lambda _{ab}(0)=207`$ nm, $`T_\mathrm{c}=100`$ K, $`\xi _{ab}(0)=2.07`$ nm, $`\kappa _{_{\mathrm{GL}}}=100,`$ $`k__\mathrm{B}=1.381\times 10^{23}`$ J/K. The substitution of these values in Eq. (168) gives
$$ϵ_{_{\mathrm{Gi}}}=\frac{8\pi ^2\times 1.381}{1000}11\%,\frac{ϵ_{_{\mathrm{Gi}}}}{6\kappa _{_{\mathrm{GL}}}^2}2\times 10^6.$$
(171)
In the case of nonzero magnetic field the self-consistent equation for the renormalized reduced temperature, Eq. (167), according to Eqs. (137), (161), and (166), takes the form
$$ϵ_{\mathrm{ren}}=\mathrm{ln}\frac{T}{T_\mathrm{c}}+ϵ_{_{\mathrm{Gi}}}\widehat{𝖫}\left[\psi \left(\frac{ϵ_{\mathrm{ren}}+h}{2h}\right)+\psi \left(\frac{c+ϵ_{\mathrm{ren}}+h}{2h}\right)\right],$$
(172)
or, within the LD model,
$`ϵ_{\mathrm{ren}}=\mathrm{ln}{\displaystyle \frac{T}{T_\mathrm{c}}}+ϵ_{_{\mathrm{Gi}}}{\displaystyle _0^{\pi /2}}{\displaystyle \frac{d\varphi }{\pi /2}}[\psi \left({\displaystyle \frac{ϵ_{\mathrm{ren}}+h+r\mathrm{sin}^2\varphi }{2h}}\right)`$
$`+\psi \left({\displaystyle \frac{c+ϵ_{\mathrm{ren}}+h+r\mathrm{sin}^2\varphi }{2h}}\right)],`$ (173)
cf. also Ref. 65. For weak magnetic fields, $`hϵ,`$ using the asymptotic formula for the digamma function, Eq. (124), we recover Eq. (167). The formulae pointed out could be easily programmed for the self-consistent LD fit to the paraconductivity near to the critical temperature $`T_\mathrm{c}.`$ With the foregoing discussion we finish the analysis of the thermodynamics of layered superconductors. We only note that all final formulae can be used to fit the experimental data. Before proceeding however, for reliability sake, it is necessary to check if the formulae implementation correctly reproduces the 3D limit case $`r\mathrm{}.`$
### 3.4 3D test example
Every layered superconductor near the critical point $`|ϵ|,hr`$ displays 3D behavior. For high-$`T_\mathrm{c}`$ cuprates, however, $`r1`$ and 3D behavior can be observed only in crystals of extremely high quality. Due to fluctuation of the stoichiometry and of the $`T_\mathrm{c}`$ 3D regime of Gaussian fluctuations may not occur. However there are many conventional layered compounds with moderate anisotropy, $`r1,`$ to which the 3D behavior has broader applicability. The 3D case can be derived as $`(r\mathrm{})`$-asymptotics if the parabolic band approximation $`\omega _1(\theta )r\theta ^2/4,`$ Eq. (34), is substituted into the $`\widehat{𝖫}`$ operator, Eq. (46). Using the variable $`x(ϵ,h)`$ from Eq. (115) and a new dimensionless variable $`q`$, defined as
$$q=\sqrt{\frac{r}{8h}}\frac{sp_z}{\mathrm{}},q^2\frac{r}{8h}\theta ^2,d\theta =2\sqrt{\frac{2h}{r}}dq,$$
(174)
we get for the regularized sum of logarithms in Eq. (112) the local approximation
$$\widehat{𝖫}^{(\mathrm{LD})}\widehat{\mathrm{𝖱𝖾𝗀}}_\zeta \underset{n=0}{\overset{\mathrm{}}{}}\mathrm{ln}\left(n+\frac{1}{2}+\frac{ϵ}{2h}\right)2\sqrt{\frac{2h}{r}}\widehat{\mathrm{𝖱𝖾𝗀}}_\zeta \underset{n=0}{\overset{\mathrm{}}{}}_{\mathrm{}}^+\mathrm{}\frac{dq}{2\pi }\mathrm{ln}\left(n+x+q^2\right).$$
(175)
The UV regularization in this expression is carried out with the help of the equation
$$\widehat{\mathrm{𝖱𝖾𝗀}}_\zeta \underset{n=0}{\overset{\mathrm{}}{}}_{\mathrm{}}^+\mathrm{}\frac{dq}{2\pi }\mathrm{ln}\left(n+x+q^2\right)=\zeta (\frac{1}{2},x),$$
(176)
which can be easily proved using derivatives of $`\zeta `$-functions
$$\frac{d}{dx}\zeta (\nu ,x)=\nu \zeta (\nu +1,x).$$
(177)
The second derivative of Eq. (176) is trivially convergent; the essence of the $`\zeta `$-function regularization lies in the omission of an arbitrary linear function $`A(c)x+B(c)`$, being analytical with respect to $`ϵ`$ and therefore irrelevant to the critical behavior, cf. Eq. (141). In fact $`c1`$ but having dropped $`A(c)`$ and $`B(c)`$ we can obtain the local approximation, $`|ϵ|,hc,`$ as $`c\mathrm{}`$ even if $`A(\mathrm{})=\mathrm{}`$ and $`B(\mathrm{})=\mathrm{}.`$ The substitution of this UV regularization in Eq. (112), using Eq. (35), gives the result by Mishonov<sup>?</sup> for the fluctuation part of the Gibbs free energy
$$G(T,H)=VF(ϵ,h)=\frac{\sqrt{2}}{2\pi }k__\mathrm{B}T\frac{V}{\xi _a(0)\xi _b(0)\xi _c(0)}h^{3/2}\zeta (\frac{1}{2},\frac{1}{2}+\frac{ϵ}{2h}).$$
(178)
This result was confirmed by Baraduc et al.,<sup>?</sup> using the same notations, with the $`\zeta `$-function presented implicitly.
In order to bridge the 3D result with the notations introduced for layered systems we can rewrite the coefficient in Eq. (178) as
$$\frac{\sqrt{2}}{2\pi }\frac{k__\mathrm{B}T}{\xi _a(0)\xi _b(0)\xi _c(0)}=4\sqrt{\frac{2}{r}}\left(\frac{1}{2}M_0B_{c2}(0)\right).$$
(179)
Now differentiation $`F(ϵ,h)`$ with respect to the magnetic field we obtain for the dimensionless magnetization, in agreement with the result by Kurkijärvi, Ambegaokar and Eilenberger<sup>?</sup>
$$\stackrel{~}{m}(ϵ,h)=3\left(\frac{2}{r}\right)^{1/2}\sqrt{h}\left[\zeta (\frac{1}{2},\frac{1}{2}+\frac{ϵ}{2h})\frac{1}{3}\zeta (\frac{1}{2},\frac{1}{2}+\frac{ϵ}{2h})\frac{ϵ}{2h}\right].$$
(180)
The subsequent differentiation with respect to the magnetic field gives the differential susceptibility. In the particular case of strong magnetic fields, $`ϵh,`$ the local approximation to the GL model, Eq. (180), gives the well known-result by Prange<sup>?</sup> with an anisotropy correction multiplier<sup>?</sup> $`\xi _{ab}(0)/\xi _c(0)`$
$$\stackrel{~}{m}(0,h)=3\sqrt{2}\times 0.0608885\sqrt{\frac{h}{r}},M=3\pi ^{1/2}\zeta (\frac{1}{2},\frac{1}{2})\frac{k__\mathrm{B}T_\mathrm{c}}{\mathrm{\Phi }_0^{3/2}}\frac{\xi _{ab}(0)}{\xi _c(0)}\sqrt{\mu _0H},$$
(181)
where for the values of the $`\zeta `$-function we have
$`\zeta ({\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}})`$ $`=\left[1+{\displaystyle \frac{1}{\sqrt{2}}}\right]\zeta \left({\displaystyle \frac{1}{2}}\right)=\text{Zeta[-1/2,1/2]}=0.0608885\mathrm{},`$
$`\zeta \left({\displaystyle \frac{1}{2}}\right)`$ $`=\text{Zeta[-1/2]}=0.207886\mathrm{}.`$ (182)
The syntax Zeta\[…\] is used in the commercial software Mathematica$`^{\text{tiny{R}⃝}}`$.<sup>?</sup> We stress, however, that these are only test mathematical asymptotics for $`c\mathrm{}`$. For the magnetization, as well as for every quantity exhibiting UV divergences in the local limit, the nonlocal effects are strongly pronounced simply because the contribution of high momenta is significant. That is why the local approximation could be quantitatively fairly good for fitting to the data for fluctuation conductivity and heat capacity. For the magnetization in strong magnetic field regime we have to take into account the effect of nonlocality by fitting the energy cutoff parameter $`\epsilon _{_{\text{}}}`$. A systematic procedure for determination of the parameters of the GL theory is developed in the next section.
## 4 Some remarks on the fitting of the GL parameters
### 4.1 Determination of the cutoff energy $`\epsilon _{_{\text{}}}`$
Let we start with designing a general procedure to fit some parameters of the GL theory which employs only data for the in-plane paraconductivity. Later on we shall address the advantage of investigating several variables simultaneously. The first step is to extract the fluctuation part of the conductivity from the temperature dependence of the resistivity $`R(T)`$. For layered cuprates the resistivity of the normal phase is to within good accuracy a linear function of temperature, $`R_N(T)=A_R+B_RT,`$ and we can fit the coefficients $`A_R`$ and $`B_R`$ far enough from the critical temperature $`T_\mathrm{c},`$ e.g. in the temperature interval $`(1.5T_\mathrm{c},3T_\mathrm{c}).`$ After that we can determine the experimental data for the fluctuation conductivity
$$\sigma _i=R(T_i)^1(A_R+B_RT_i)^1$$
(183)
for all experimental points $`i=1,\mathrm{},N_{\mathrm{exp}}.`$ For bi-layered cuprates, such as YBa<sub>2</sub>Cu<sub>3</sub>O<sub>7-δ</sub> and Bi<sub>2</sub>Sr<sub>2</sub>CaCu<sub>2</sub>O<sub>8</sub>, one can attempt fitting the data with the formula for the bilayered model, Eq. (2.3), where an arbitrary life-time $`\stackrel{~}{\tau }_{\mathrm{rel}}`$ and cutoff parameter $`c`$ are included in the interpolation
$$\sigma (ϵ;\stackrel{~}{\tau }_{\mathrm{rel}},r,w,c)=\frac{e^2}{16\mathrm{}}\stackrel{~}{\tau }_{\mathrm{rel}}\frac{N}{s}\left[f_{\mathrm{MT}}(ϵ;r,w)f_{\mathrm{MT}}(c+ϵ;r,w)\right],$$
(184)
where
$$f_{\mathrm{MT}}(ϵ;r,w)\frac{ϵ+\frac{1}{2}rw}{\sqrt{\left(ϵ^2+rwϵ\right)\left(ϵ^2+rwϵ+\frac{1}{4}r^2w\right)}}=\widehat{𝖫}^{(\mathrm{MT})}f_{\mathrm{APS}}(ϵ,h=0)=\widehat{𝖫}^{(\mathrm{MT})}\frac{1}{ϵ}.$$
(185)
As a next step, if necessary, one may fit the data using logarithmic plot that generates the dimensionless deviations from the $`\mathrm{ln}(\sigma _{ab}(T))`$-values,
$$x_i(r,w,c)=\mathrm{ln}\left[f_{\mathrm{MT}}(ϵ_i;r,w)f_{\mathrm{MT}}(c+ϵ_i;r,w)\right]\mathrm{ln}\left(\frac{\sigma _i}{\frac{e^2}{16\mathrm{}}\frac{N}{s}}\right).$$
(186)
For the $`x_i`$ data we can calculate the mean value, the averaged square
$$x=\frac{1}{N_{\mathrm{exp}}}\underset{i=1}{\overset{N_{\mathrm{exp}}}{}}x_i,x^2=\frac{1}{N_{\mathrm{exp}}}\underset{i=1}{\overset{N_{\mathrm{exp}}}{}}x_i^2,$$
(187)
and the dispersion
$$S(r,w,c)=x^2x^2.$$
(188)
The fitting procedure is then reduced to numerically finding the minimum of the dispersion
$$S(r_0,w_0,c_0)S(r,w,c)$$
(189)
in the space of parameters $`(r,w,c).`$ We have to start from some acceptable set of parameters, for example, $`r_0=\frac{1}{7},`$ $`w_0=1,`$ and $`c_0=\frac{1}{2}`$, and to search for the minimal value in certain range, e.g. $`r_0=(0,\mathrm{\hspace{0.33em}1}),`$ $`w_0(1,30)`$ and $`c_0=\frac{1}{2}(0.2,2).`$ It is possible that the parameters of the normal resistivity be corrected by the same procedure for minimization of the dispersion $`S(r,w,c,A_R,B_R)`$. The contemporary methods of the mathematical statistics, such as the bootstrap and Jack-knife, can tell us how reliable is the set of the fitted parameters; the simplest possible realization is to decrement sequentially $`N_{\mathrm{exp}}`$ by one and to investigate the distribution of the fitted GL parameters at every step. For example, the $`w`$ parameter is almost inaccessible since for $`w=1`$ and $`w\mathrm{}`$ we have LD-type temperature dependence of the paraconductivity. On the other hand if we try to fit the paraconductivity far from the critical temperature, e.g. $`T(1.02T_\mathrm{c},1.15T_\mathrm{c})`$ we can easily find some estimate for the cutoff parameter $`c.`$ In any case a good fit would be useful because as a by-product we determine the life-time of the fluctuation Cooper pairs
$$\stackrel{~}{\tau }_{\mathrm{rel}}=\mathrm{exp}\left(x(r_0,w_0,c_0)\right).$$
(190)
The same procedure can be applied to the magnetic susceptibility at vanishing magnetic field which, according to Eq. (61), is proportional to the conductivity, or for the susceptibility in the LD model which, according to Eqs. (42), (72), and (168), reads
$$\chi _{_{\mathrm{LD}}}(ϵ)=\frac{1}{6}\frac{ϵ_{_{\mathrm{Gi}}}}{\kappa _{_{\mathrm{GL}}}^2}\left[\frac{1}{\sqrt{ϵ(ϵ+r)}}\frac{1}{\sqrt{(c+ϵ)(c+ϵ+r)}}\right],$$
(191)
where
$$\frac{ϵ_{_{\mathrm{Gi}}}}{\kappa _{_{\mathrm{GL}}}^2}=2\pi \mu _0\frac{k__\mathrm{B}T_\mathrm{c}}{\mathrm{\Phi }_0^2}\xi _{ab}^2(0)\frac{N}{s}=\frac{M_0}{H_{c2}(0)}.$$
(192)
The general formula for the conductivity, Eqs. (58), (70),
$$\sigma _{ab}(ϵ,h)=\frac{\pi }{8}\frac{\tau _{\mathrm{rel}}}{R_{\mathrm{QHE}}}\widehat{𝖫}^{(\mathrm{LD})}\widehat{𝖢}f_{\mathrm{APS}}(ϵ,h),$$
(193)
which for single layered superconductor reads, cf. Eq. (133),
$`\sigma _{ab}(ϵ,h;r,C)=`$ $`\stackrel{~}{\tau }_{\mathrm{rel}}{\displaystyle \frac{e^2}{16\mathrm{}s_{\mathrm{eff}}}}{\displaystyle \frac{2}{h^2}}{\displaystyle _0^{\pi /2}}{\displaystyle \frac{d\varphi }{\pi /2}}\{(ϵ+r\mathrm{sin}^2\varphi )[\psi ({\displaystyle \frac{1}{2}}+{\displaystyle \frac{ϵ+r\mathrm{sin}^2\varphi }{2h}})`$
$`\psi (1+{\displaystyle \frac{ϵ+r\mathrm{sin}^2\varphi }{2h}})+{\displaystyle \frac{h}{ϵ+r\mathrm{sin}^2\varphi }}]`$
$`(c+ϵ+r\mathrm{sin}^2\varphi )[\psi ({\displaystyle \frac{1}{2}}+{\displaystyle \frac{c+ϵ+r\mathrm{sin}^2\varphi }{2h}})`$
$`\psi (1+{\displaystyle \frac{c+ϵ+r\mathrm{sin}^2\varphi }{2h}})+{\displaystyle \frac{h}{c+ϵ+r\mathrm{sin}^2\varphi }}]\}`$ (194)
gives another possibility for determining the energy cutoff parameter $`c.`$ As most appropriate regime we recommend that the measurements of the conductivity as a function of the magnetic field to be carried out at the critical temperature $`T=T_\mathrm{c}.`$ In this case for strong magnetic field, $`hr,`$ the layered superconductors with strong anisotropy $`r1`$ show 2D behavior. The substitution $`ϵ=0`$ in Eq. (194) gives another universal law derived within the GL theory with energy cutoff
$`{\displaystyle \frac{1}{2}}{\displaystyle \frac{B}{B_{c2}(0)}}{\displaystyle \frac{\sigma _{ab}(ϵ=0,h)}{\stackrel{~}{\tau }_{\mathrm{rel}}(e^2/16\mathrm{}s_{\mathrm{eff}})}}`$ $`=\widehat{𝖢}{\displaystyle \frac{h}{2}}f_{\mathrm{APS}}(ϵ=0,h)`$
$`={\displaystyle \frac{\pi \mathrm{}}{e}}s_{\mathrm{eff}}{\displaystyle \frac{\xi _{ab}^2(0)}{\tau _0}}B\sigma _{ab}(T_\mathrm{c},B)=U_\sigma \left({\displaystyle \frac{2}{c}}{\displaystyle \frac{B}{B_{c2}(0)}}\right),`$ (195)
where $`y=2h/c,`$ cf. Eq. (135),
$$U_\sigma (y)=\frac{2}{y}\left[\psi \left(1+\frac{1}{y}\right)\psi \left(\frac{1}{2}+\frac{1}{y}\right)\right],$$
(196)
$`U_\sigma (0)=1,`$ and $`U_\sigma (\mathrm{})=0.`$ At best, the universal dimensionless conductivity $`U_\sigma B\sigma (B)`$ and magnetization $`U_MM`$ have to be fitted simultaneously using the data for the same crystal and common dimensionless argument $`B.`$ Similar universal scaling law for the heat capacity can be derived from Eq. (138)
$$\frac{2}{\zeta (2,\frac{1}{2})}\frac{B}{B_{c2}(0)}\frac{C(T_\mathrm{c},B)}{ϵ_{_{\mathrm{Gi}}}\mathrm{\Delta }C}=U_C\left(\frac{2}{c}\frac{B}{B_{c2}(0)}\right),$$
(197)
where
$$U_C(y)=1\frac{\zeta (2,\frac{1}{2}+\frac{1}{y})}{\zeta (2,\frac{1}{2})},$$
(198)
but the accuracy of thermal measurements is probably not high enough in order for this to be experimentally confirmed.
### 4.2 Determination of the coherence length $`\xi _{ab}(0)`$
The fit of every fluctuation variable as a function of the dimensional magnetic field $`h,`$ the conductivity
$$\sigma (ϵ,h)=\sigma (ϵ)+\mathrm{\Delta }\sigma (ϵ,h),$$
(199)
for example, provides a method for determination of $`B_{c2}(0)`$ and $`\xi _{ab}(0).`$ At weak magnetic fields, $`hϵ`$, the magnetoconductivity is proportional to the square of the magnetic field $`\mathrm{\Delta }\sigma (ϵ,h)\sigma (ϵ,h)\sigma (ϵ)B^2.`$ For this small negative quantity, $`0<\mathrm{\Delta }\sigma (ϵ,h)\sigma (ϵ),`$ the APS result, Eq. (68), reads<sup>?</sup>
$$\mathrm{\Delta }\sigma (ϵ,h)\frac{h^2}{4}\frac{^2}{ϵ^2}\sigma (ϵ),$$
(200)
where $`h=2\pi \xi _{ab}^2(0)B_z/\mathrm{\Phi }_0.`$ The common multiplier $`\tau _0`$ from Eq. (65) is obviously canceled in this relation because, roughly speaking, the transport takes time even in the presence of magnetic field. We note that a multiplier $`\stackrel{~}{\tau }_{\mathrm{rel}}`$ was misintroduced by M. V. Ramallo in Ref. 36 in the right-hand-side of the above equation (see Eq. (4) in Ref. 36). Thereby the old experimental data in Ref. 36 have been apparently processed by employing erroneous expression and therefore the discussion related to Fig. 2 in Ref. 36 is physically unsound. As a consequence, the life-time constant of metastable Cooper pairs in cuprates is still waiting for its first experimental determination. Nevertheless the novel theoretical result that the life-time constant $`\tau _0`$ and the diffusion coefficient of the fluctuation Cooper pairs $`\xi _{ab}^2(0)/\tau _0`$ can be determined from the $`\sigma /\chi `$-ratio remains unchanged.
Returning to Eq. (200) we note that after two-fold integration of the relation (200) in some temperature interval, e.g. $`(ϵ_a,ϵ_b)=(0.03,0.09),`$ the “noise” in the experimental data is already irrelevant and we can rewrite Eq. (200) as<sup>?</sup>
$$\xi _{ab}(0)=l_B\left[\frac{{\displaystyle _{ϵ_a}^{ϵ_b}}𝑑ϵ^{}{\displaystyle _ϵ^{}^{ϵ_b}}\left(\mathrm{\Delta }\sigma (ϵ^{\prime \prime },h)\right)𝑑ϵ^{\prime \prime }}{\sigma (ϵ_a)\sigma (ϵ_b)+(ϵ_bϵ_a){\displaystyle \frac{d\sigma }{dϵ}}(ϵ_b)}\right]^{1/4},$$
(201)
where $`l_B`$ is the magnetic length
$$l_B=\sqrt{\frac{\mathrm{\Phi }_0}{\pi B}}=\sqrt{\frac{\mathrm{}}{eB}}=\frac{25.6\mathrm{nm}}{\sqrt{B(T)}}.$$
(202)
For practical application we have to take into account that far from the critical temperature, even for $`TT_\mathrm{c}=15\%T_\mathrm{c}`$ the fluctuation conductivity is negligible $`\sigma (0.15)\sigma (\mathrm{})=0.`$ That is why in acceptable approximation we can take $`ϵ_b=0.15\mathrm{"}\mathrm{}\mathrm{"}.`$ For $`ϵ>ϵ_b`$ the temperature dependence of the magnetoconductivity in the numerator of Eq. (201) can be an extrapolated LD fit.
However, due to the strong critical behavior $`\mathrm{\Delta }\sigma h^2/ϵ^3`$ for $`ϵr`$ the influence of the interval $`(ϵ_b,\mathrm{})`$ can be neglected. In such a way, after a partial integration, we arrive at a simpler equation for determination of the in-plane coherence length, cf. Ref. 36,
$$\xi _{ab}(0)l_B\left[\frac{1}{\sigma (ϵ)}\left(_ϵ^{\mathrm{}}ϵ^{}\left(\mathrm{\Delta }\sigma (ϵ^{},h)\right)𝑑ϵ^{}ϵ_ϵ^{\mathrm{}}\left(\mathrm{\Delta }\sigma (ϵ^{\prime \prime },h)\right)𝑑ϵ^{\prime \prime }\right)\right]^{1/4}=\text{const},$$
(203)
where the integrations should be performed in the whole experimentally accessible temperature range above $`(1+ϵ)T_\mathrm{c}.`$ This result of the Gaussian fluctuation theory does not depend upon the $`\tau _0`$ parameter, effective mass of Cooper pairs $`m_{ab},`$ and the space dimensionality. We consider this procedure for determination of the coherence length $`\xi _{ab}(0)`$ as being the best one, as it is model-free and does not depend on the multilaminarity of the superconductor, i.e. on the dispersion of Cooper pairs in $`c`$-direction $`\epsilon _{c,j}(p_z)`$. Equation (203) has the same form for both strongly anisotropic high-$`T_\mathrm{c}`$ cuprates and bulk conventional dirty alloys. Of course, methods particularly based on the proximity to the critical line $`H_{c2}(T)`$ can be very useful in determining $`\xi _{ab}(0)`$ especially in the case of strong magnetic fields. For example, Eq. (68) gives another appropriate formula
$$\sigma _{ab}(ϵ,h)\stackrel{~}{\tau }_{\mathrm{rel}}\frac{e^2}{16\mathrm{}}\frac{N}{s}\frac{4}{\sqrt{(ϵ+h)(ϵ+h+r)}}$$
(204)
applicable for $`ϵ_{_{\mathrm{Gi}}}ϵ+hh.`$ Similar result, cf. also Eq. (142),
$$M=M_0\stackrel{~}{m}\frac{k__\mathrm{B}T_\mathrm{c}}{\mathrm{\Phi }_0}\frac{N}{s}\frac{h}{\sqrt{(ϵ+h)(ϵ+h+r)}}$$
(205)
can be derived under the same physical conditions from the formula for the fluctuation magnetic moment, Eq. (133), using the approximations for $`0<x1,`$
$$\mathrm{\Gamma }(x)\mathrm{ln}x\frac{1}{2}\mathrm{ln}(2\pi ),\psi (x)\frac{1}{x}.$$
(206)
The experimental investigation of the conductivity, Eq. (204), and magnetization, Eq. (205), is probably the best way to extract the upper critical field $`H_{c2}(T)`$ for high-$`T_\mathrm{c}`$ cuprates; the $`H\sigma /M`$ quotient near the critical line is $`2/3`$ of the $`\sigma /\chi `$ quotient for weak magnetic fields.
### 4.3 Determination of the Cooper pair life-time constant $`\tau _0`$
Having a reliable estimate for the coherence length, the life-time constant of the metastable Cooper pairs above $`T_\mathrm{c}`$ can be determined via the $`\sigma /\chi `$-quotient, Eq. (61). We believe that this method will become a standard procedure in the physics of high-$`T_\mathrm{c}`$ materials. Certainly the most transparent method is just the fit to the phase angle of high-frequency complex fluctuation conductivity
$$\varphi _\sigma (\omega \tau (ϵ_{\mathrm{ren}}))=\mathrm{arctan}\frac{\sigma ^{\prime \prime }(\omega )}{\sigma ^{}(\omega )}=\mathrm{arctan}\frac{\widehat{𝖫}\varsigma _2\left(\omega \tau _0/ϵ_{\mathrm{ren}}\right)}{\widehat{𝖫}\varsigma _1(\omega \tau _0/ϵ_{\mathrm{ren}})}.$$
(207)
The state-of-the-art electronics gives such a possibility, but unfortunately the first experiments of the type<sup>?,?</sup> was not performed in the Gaussian region. For the development of Gaussian spectroscopy which will give results relevant for the microscopic mechanisms of superconductivity we recommend the use of the conventional thin films and high-quality low temperature cuprate films, such as Bi<sub>2</sub>Sr<sub>2</sub>CaCu<sub>2</sub>O$`_8.`$
### 4.4 Determination of the Ginzburg number and penetration depth $`\lambda _{ab}(0)`$
The applicability of the self-consistent approximation in the theory of fluctuation phenomena in superconductors is strongly limited by the quality of the samples. The fluctuation of the critical temperature $`\mathrm{\Delta }T_\mathrm{c}`$, e.g. due to the oxygen stoichiometry in cuprates, should be small enough, $`\mathrm{\Delta }T_\mathrm{c}ϵ_{_{\mathrm{Gi}}}T_\mathrm{c},`$ and this has to be verified empirically. If the $`\sigma /\chi `$ ratio remains temperature independent for $`ϵ<3\%`$ and both $`\sigma (ϵ)`$ and $`\chi (ϵ)`$ demonstrate weak deviation from the LD fit obtained from the range $`ϵ(3\%,9\%),`$ this could be considered as a hint in favor of the self-consistent approximation. In this case $`ϵ_{_{\mathrm{Gi}}}`$ can be fitted by substituting the solution $`ϵ_{\mathrm{ren}}(ϵ)`$ of Eq. (167) into the LD fit to $`\sigma ^{(\mathrm{LD})}(ϵ_{\mathrm{ren}})`$ and $`\chi ^{(\mathrm{LD})}(ϵ_{\mathrm{ren}}).`$ We note that the reliability in fitting $`ϵ_{_{\mathrm{Gi}}}`$ is determined by the condition whether the self-consistent approach and the use of $`ϵ_{\mathrm{ren}}`$ significantly improve the accuracy of the fit to the experimental data near $`T_\mathrm{c}.`$ According to Eq. (168) we can parameterize $`ϵ_{_{\mathrm{Gi}}}`$ with the help of the penetration depth $`\lambda _{ab}(0).`$ In any case, an evaluation of such type should be a part of the complete set of GL parameters of the superconductor. Another possibility for the thermodynamic determination of the penetration depth $`\lambda _{ab}(0)`$ is provided by the jump in the specific heat at the critical temperature, Eq. (165). As a rule the accuracy of the determination of the penetration depth by the thermodynamic methods cannot be high, especially for high-$`T_\mathrm{c}`$ cuprates where the phonon part strongly dominates. An acceptable value of $`\mathrm{ln}\kappa _{_{\mathrm{GL}}}`$ derived from the heat capacity is necessary for the establishment of a coherent understanding of the superconductivity; there is no doubt that the direct investigation of the vortex phase of the superconductors or vortex-free high-frequency measurements constitute the best methods for determination of $`\lambda _{ab}(0).`$
## 5 Discussion and conclusions
In the attempts to systematize the available results we had to derive in parallel new ones too. We shall summarize the most important of them starting with remarks concerning the theory. As the ultimate result we consider the representation of the fluctuation part of the Gibbs free energy by the Euler $`\mathrm{\Gamma }`$-function Eq. (129) in Gaussian approximation. This result trivializes the derivation of all thermodynamic variables, such as fluctuation magnetization Eq. (133), or fluctuation heat capacity Eq. (150). To our knowledge this is a novel result, but we find it strange that it remained unobserved given the great attention which the fluctuations in high-$`T_\mathrm{c}`$ superconductors have attracted. The importance of fluctuations was mentioned even in the classical work by Bednorz and Müller. Fluctuations in superconductors were among the main topics in many scientific activities; the $`\mathrm{\Gamma }`$-function is well-known to all physicist; the mathematical physics behind the 2D statistical mechanics is well developed, polygamma functions can be found in a number of BCS papers, and finally the solution turns out to be on a textbook level. Just the same is the situation for the 3D GL model. In this case the solution for the free energy is given in terms of the Hurwitz $`\zeta `$-functions. Analogous result gave the name of one of the most powerful methods in the field theory — $`\zeta `$-function method for ultraviolet regularization, but this method was never applied to the most simple problem of a 3D GL model related to numerous experiments in the physics of superconductivity.
Another simple but useful detail is the layering operator $`\widehat{𝖫}`$, Eq. (44), which allows us to extend the 2D result onto layered superconductors and even to 3D superconductors. The method can be applied not only to the thermodynamic variables but to the fluctuation part of the kinetic coefficients as well. In this way we obtained useful formulae for the in-plane fluctuation conductivity in perpendicular magnetic field, Eq. (194), and for the high-frequency Aslamazov-Larkin conductivity in layered superconductors, Eq. (74). We proposed further convenient $`r`$-$`w`$ parameters, Eq. (2.3), for the bi-layered model which could be utilized for experimental data processing of the fluctuation phenomena in bi-layered cuprates, such as YBa<sub>2</sub>Cu<sub>3</sub>O<sub>7-δ</sub> for example.
The representation of the thermodynamic variables via polygamma functions is very helpful at strong magnetic fields but due to the presence of the magnetic field in denominator these results cannot be directly applied to zero-magnetic-field limit. For small magnetic fields, on the other hand, we have to use asymptotic formulae for polygamma and $`\zeta `$-functions with large arguments. This is the reason why the weak-magnetic-field expansion of the magnetization and the other thermodynamic variables has so bad convergence. In order to fit the experimental data for the magnetization in weak magnetic field using the new analytical result for the LD model, Eq. (82), we arrive at the problem for summation of divergent asymptotic series. At least for experimentalists this is a nontrivial problem which led us to give a prescription for usage of series from the theoretical papers. There is no doubt that the $`\epsilon `$-method is one of the brilliant achievements of the applied mathematics of XX century. However, it turns out that this method was not cast in an suitable form to be employed by users like experimentalists having no time to understand how the underlying mathematics can be derived. That is why we presented an oversimplified version of this algorithm illustrated by a simple fortran90 program. The latter can be also used for calculation of the differential nonlinear susceptibility at finite magnetic field, Eq. (105), which is another novel result in the present work.
Let us now address the simple final formulae that can be directly used for experimental data processing. First of all we advocate that the relation between fluctuation conductivity and magnetoconductivity, Eq. (203), provides the best method (shortly announced in Ref. 36) for determination of the in-plane coherence length $`\xi _{ab}(0)`$ in layered high-$`T_\mathrm{c}`$ cuprates and conventional superconductor superlattices and thin films. Having such a reliable method for determination of $`\xi _{ab}(0),`$ the Cooper pair life-time spectroscopy can be created<sup>?</sup> on the basis of determination of the life-time constant $`\tau _0`$ by the $`\sigma /\chi `$ quotient, Eq. (61).
Usually science starts with some simplicity, thus it is surprising that the temperature independence of the $`\sigma /\chi ,`$ $`\chi /C,`$ and $`\sigma /C`$ quotients has not attracted any attention in physics. The question of whether the high-$`T_\mathrm{c}`$ cuprates are BCS superconductors, or they have a non BCS behavior, consumed more paper and brought more information pollution than that about the sense of life, about the smile of Mona Lisa. Now we possess a perfect tool to check whether this sacramental $`\frac{\pi }{8}`$ BCS ratio, Eq. (63), still exists in the physics of high-$`T_\mathrm{c}`$ superconductivity. A careful study of the relative life-time constant $`\tau _0`$ by the $`\sigma /\chi `$ ratio, Eq. (64), will provide a unique information on the presence of depairing impurities in the superconducting cuprates. The doping dependence of this ratio will give important information for the limits of applicability of the self-consistent BCS approximation. In principle the same life-time spectroscopy can be applied to heavy fermions and other exotic superconductors.
The methods we proposed in this review can be initially tested by means of alternative methods for determination of $`\xi _{ab}(0)`$, e.g. from the slope of the upper critical field $`H_{c2}(T)`$ defined by the fluctuation magnetization of the normal phase near the critical line, Eq. (205), being another new result derived here, or from the fluctuation conductivity of the LD model, Eq. (204). Addressing the conductivity we consider that the fitting to high frequency experimental data with the help of formulae (73) and (207) will give a direct method for determination of the relaxation time of the superconducting order parameter. A good monocrystal of layered cuprate or high-quality thin film with as low as possible critical temperature could ensure the overlap of both the suggested methods for Cooper pair life-time spectroscopy. At present we only know<sup>?</sup> that for 93 K YBa<sub>2</sub>Cu<sub>3</sub>O<sub>7-δ</sub> $`\tau _{0,\mathrm{\Psi }}=2\tau _0=32\mathrm{fs}.`$ We hope, however, that several experimental methods for determination of $`\xi _{ab}(0)`$ and $`\tau _0`$ will be mutually verified in the nearest future. Thus, the investigation of the Gaussian fluctuations may become a routine procedure in the materials science of superconductors.
We also believe that the development of the Gaussian spectroscopy will lead to determination of the Ginzburg number $`ϵ_{_{\mathrm{Gi}}},`$ the energy cutoff, i.e. the maximal kinetic energy of the Cooper pairs $`\epsilon _{_{\text{}}}=c\mathrm{}^2/2m_{ab}\xi _{ab}^2(0).`$ Up to now these parameters of the GL theory are inaccessible. We hope that our derived self-consistent equation for the reduced temperature, Eq. (173), will stimulate experimentalists to reexamine the data for high-quality crystals in the region close to $`ϵ+h3\%`$ in order to extract $`ϵ_{_{\mathrm{Gi}}}.`$
Virtually all final results are presented by taking into account the energy cutoff parameter $`c.`$ The nonlocality corrections can be extracted from almost all fluctuation variables, if $`ϵ+h>10`$%, but we suggest special new experiments to be conducted for investigation of nonlocality effects in quasi 2D superconductors at $`T_\mathrm{c}.`$ Analogous investigations for fluctuation diamagnetism for classical bulk superconductors are already classics in physics of superconductivity; see for example Fig. 8.5 in the well-known textbook by Tinkham.<sup>?</sup> The universal scaling law for the heat capacity, Eq. (198), for the magnetization, Eq. (135), and conductivity, Eq. (196), versus the reduced magnetic field $`y=2h/c`$ are depicted in Fig. 5.
At least for conductivity the experimental confirmation for the quasi-2D superconductors ($`hr,ϵ_{_{\mathrm{Gi}}}`$) can be easily achieved. How different are the animals…? The biochemists considered that what is true for Escherichia coli holds true for the elephant. Analogously, we consider that $`B\sigma (T_\mathrm{c},B)`$ versus $`B`$ will be within 20% accuracy the same for conventional Pb layers and for strongly anisotropic underdoped Bi<sub>2</sub>Sr<sub>2</sub>CaCu<sub>2</sub>O<sub>8</sub> in spite of the bunch of sophisticated theories of high-$`T_\mathrm{c}`$ superconductivity. The GL theory gives the scaling law, the notions and notations, and in this sense the language for analysis of the fluctuation phenomena. The precisely measured deviations from the GL scaling low could give the basis for further microscopic consideration using the methods of the statistical mechanics. This is the last example how the development of the Gaussian fluctuation spectroscopy could be of importance not only for the materials science but for the fundamental physics of superconductivity as well.
Acknowledgments
Some parts of this review are based on the lectures on statistical mechanics, solid state theory, numerical methods and physics of superconductivity read by one of the authors (T.M.) at the University of Sofia. He would like to thank to many of his former students for the help and collaboration in the early stages of this work, especially to D. Damianov for the collaboration on the Boltzmann equation for the fluctuation Cooper pairs<sup>?</sup> and to N. Zahariev for implementing the $`\epsilon `$-algorithm in C. It is a pleasure to acknowledge the cooperation of C. Carballeira in implementing the $`\epsilon `$-algorithm in Mathematica$`^{\text{tiny{R}⃝}}`$ and checking the numerical equivalence for $`r=0`$ of Eqs. (82) and (133) using this software. The same author appreciates the discussions with Prof. Vidal on the experiments concerning $`\sigma /\chi `$ and other quotients. The completion of this review would be impossible without the warm hospitality of Prof. Indekeu, his support and interest to this work.
The early stages of this study were supported by the Bulgarian NSF ($`\mathrm{\Phi }`$627/1996) and the Visiting Professor Fellowship from the Spanish Ministry of Education. This paper was supported by the Belgian DWTC, the Flemish Government Programme VIS/97/01, the IUAP and the GOA.
Appendix A
```
!+ Test driver program for subroutine Limes
PROGRAM Test
IMPLICIT NONE
INTEGER, PARAMETER :: pr = SELECTED_REAL_KIND (30,150)
REAL (pr), PARAMETER :: zero = 0.0, one = 1.0
REAL (pr) :: S(0:137), C(0:137), x, xi, arg
REAL (pr) :: rLimes
REAL (pr) :: err
INTEGER :: N
INTEGER :: i
INTEGER :: i_Pade
INTEGER :: k_Pade
INTEGER :: is
WRITE (*, ’(15X,A)’) ’+---------------------------------------------+’
WRITE (*, ’(15X,A)’) ’| Test driver program for subroutine Limes |’
WRITE (*, ’(15X,A)’) ’| Calculate Ln[x] |’
WRITE (*, ’(15X,A)’) ’+---------------------------------------------+’
WRITE (*, ’(A)’) ’ ’
WRITE (*, ’(A)’, ADVANCE=’NO’) ’ Enter argument of Ln[x], x = ’
READ (*,*) arg
WRITE (*, ’(A)’, ADVANCE=’NO’) ’ Enter the number of known terms, N = ’
READ (*,*) N
IF (N > 137) N = 137 ! ... we like this number ;-)
x = arg - one
xi = one
is = 1
! Initialize S to store the first N+1 known partial sums
! S0, S1, S2,..., Sn-1, Sn
!
! Sn = x + x^2/2 - x^3/3! + x^4/4! - ... + (-1)^n x^n/n!
!
S(0) = zero ! *** lower bound of the subscript should start at 0 ! ***
DO i=1,N
xi = xi*x
C(i) = xi/i
S(i) = S(i-1) + is * C(i)
is = -is
END DO
WRITE (*, ’(A)’) ’ ’
WRITE (*, ’(A)’) ’ ===========================================&
&===========================================’
WRITE (*, ’(14X,A8,5(X,A12))’) ’S(0)’, ’S(1)’, ’S(2)’, ’S(3)’,&
’S(4)’, ’S(5)’
WRITE (*, ’(A)’) ’ -------------------------------------------&
&-------------------------------------------’
WRITE (*, ’(A)’, ADVANCE=’NO’) ’ before call: ’
WRITE (*, ’(F8.4,5(X,f12.4))’) S(0:5)
CALL Limes & ! call subroutine Limes to calculate Ln[x]
( N, & ! in
S(0:N), & ! inout
rLimes, & ! out
i_Pade, & ! out
k_Pade, & ! out
err ) ! out
! Formated output of results
!
WRITE (*, ’(A)’, ADVANCE=’NO’) ’ after call: ’
WRITE (*, ’(F8.4,5(X,f12.4))’) S(0:5)
WRITE (*, ’(A)’) ’ ===========================================&
&===========================================’
WRITE (*,*) ’ ’
WRITE (*, ’(X,A,X,F9.3,X,A,ES12.5)’) ’Ln[’, arg, ’] = ’, LOG (arg)
WRITE (*, ’(X,A,I3,A,I3,A,ES12.5)’) ’rLimes[’, i_Pade, ’,’, k_Pade, &
’] = ’, rLimes
WRITE (*, ’(13X, 1A, ES12.5)’) ’err = ’, err
WRITE (*, ’(6X, 1A, ES12.5)’) &
’true error = ’, ABS ( rLimes - LOG (one + x))
!-----------------------------------------------------------------------------
CONTAINS
!-----------------------------------------------------------------------------
!+ Finds the limit of a series
SUBROUTINE Limes &
( N, & ! in
S, & ! inout
rLimes, & ! out
i_Pade, & ! out
k_Pade, & ! out
err ) ! out
! Description:
! Finds the limit of a series in the case where only
! the first N+1 terms are known.
!
! Method:
! The subroutine operates by applying the epsilon-algorithm
! to the sequence of partial sums of a seris supplied on input.
! For desciption of the algorithm, please see:
!
! [1] T. Mishonov and E. Penev, Int. J. Mod. Phys. B 14, 3831 (2000)
!
! Owners: Todor Mishonov & Evgeni Penev
!
! History:
! Version Date Comment
! ======= ==== =======
! 1.0 01/04/2000 Original code. T. Mishonov & E. Penev
!
! Code Description:
! Language: Fortran 90.
! Software Standards: "European Standards for Writing and
! Documenting Exchangeable Fortran 90 Code".
!
! Declarations:
IMPLICIT NONE
!* Subroutine arguments
! Scalar arguments with intent(in):
INTEGER, INTENT (IN) :: N ! width of the epsilon-table
! Array arguments with intent(inout):
REAL (pr), INTENT (INOUT) :: S(0:) ! sequential row of the epsilon-table
! Scalar arguments with intent(out)
REAL (pr), INTENT (OUT) :: rLimes ! value of the series limes
INTEGER, INTENT (OUT) :: i_Pade ! power of the numerator
INTEGER, INTENT (OUT) :: k_Pade ! power of the denominator
REAL (pr), INTENT (OUT) :: err ! empirical error
!* End of Subroutine arguments
! Local parameters ! these two need no description ;-)
REAL (pr), PARAMETER :: zero = 0.0
REAL (pr), PARAMETER :: one = 1.0
! Local scalars
REAL (pr) :: A_max ! maximum element of A
INTEGER :: i ! index variable for columns
INTEGER :: k ! index variable for rows
! Local arrays
REAL (pr) :: A(0:N) ! auxiliary row of the epsilon-table
!- End of header --------------------------------------------------------------
! Parse input: the algorithm cannot employ more elements than supplied on
! input, i.e. N <= size(S)
!
IF ( N > SIZE (S(:)) ) THEN
WRITE (*, ’(A)’) ’*** Illegal input to Limes: N > size(S)’
STOP 1
END IF
! Algorithm not applicable for N < 2
!
IF ( N < 2 ) THEN
WRITE (*, ’(A)’) ’*** Illegal input to Limes: N < 2’
STOP 2
END IF
!-----------------------------------------------------------------------------
! I. Initialize with natural assignments
!-----------------------------------------------------------------------------
rLimes = S(N) ! the N-th partial sum
err = ABS ( S(N) - S(N-1) ) ! error -> |S(N) - S(N-1)|
i_Pade = N ! Pade approximant [N/0]
k_Pade = 0 !
A(:) = zero ! auxiliary row initially set to zero
A_max = zero ! max. element set to zero
k = 1 ! algorithm starts from the first row
!-----------------------------------------------------------------------------
! II. Main loop: fill in the epsilon table, check for convergence ...
! (provision against devision by zero employs pseudo-inverse numbers)
!-----------------------------------------------------------------------------
DO
IF ( N - 2 * k + 1 < 0 ) EXIT
! Update the auxiliary row A(i) of the epsilon-table
! by applying the "cross rule".
!
DO i=0, N - 2 * k + 1
IF ( S(i+1) /= S(i) ) THEN
A(i) = A(i+1) + one/(S(i+1) - S(i))
ELSE
A(i) = A(i+1)
END IF
END DO
IF ( N - 2 * k < 0 ) EXIT
! Update the sequential row S(i) of the epsilon-table
! by applying the "cross rule".
!
DO i=0, N - 2 * k
IF ( A(i+1) /= A(i) ) THEN
S(i) = S(i+1) + one/(A(i+1) - A(i))
ELSE
S(i) = S(i+1)
END IF
! Check for convergence, based on A_max; see Ref. [1]
!
IF ( ABS ( A(i) ) > A_max ) THEN
A_max = ABS ( A(i) )
rLimes = S(i)
k_Pade = k
i_Pade = i + k_Pade
err = one/A_max
IF ( S(i+1) == S(i) ) RETURN
END IF
END DO
k = k + 1 ! increment row index
END DO
END SUBROUTINE Limes
END PROGRAM Test
```
References
|
warning/0004/math0004103.html
|
ar5iv
|
text
|
# Lines on contact manifolds
## 1. Introduction
If $`X`$ is a complex projective manifold which carries a contact structure, then the results of \[Dem00\] and \[KPSW00\] show that $`X`$ is either isomorphic to a projectivized tangent bundle of a complex manifold, or that $`X`$ is Fano and $`b_2(X)=1`$. In this paper we study the latter case where $`X`$ is Fano. It is generally believed that these assumptions imply that $`X`$ is homogeneous —for an introduction, see the excellent survey in \[Bea99\].
It follows from our previous work \[Keb00\] that $`X`$ can always be covered by lines. Thus, it seems natural to consider the geometry of lines in greater detail. We will show that if $`xX`$ is a general point, then all lines through $`x`$ are smooth. If $`X\cong ̸_{2n+1}`$, then the tangent spaces to these lines generate the contact distribution at $`x`$. It follows that the contact structure on $`X`$ is unique, thus answering a question of C. LeBrun \[Leb97, question 11.3\]. The result was previously obtained by C. LeBrun \[Leb95\] if $`X`$ is a twistor space.
#### Acknowledgement
The paper was written while the author enjoyed the hospitality of RIMS in Kyoto. The author is grateful to Y. Miyaoka for the invitation and to the members of that institute for creating a stimulating atmosphere. He would like to thank S. Helmke, J.-M. Hwang, S. Kovács and Y. Miyaoka for a number of enlightening discussions on the subject. Finally, the author thanks the referee for criticism concerning the exposition of earlier versions of the paper.
## 2. Setup
### 2.1. Contact Manifolds
Throughout the present work, let $`X`$ be a complex projective contact manifold of dimension $`2n+1`$. By definition, the contact structure is given by a sequence of vector bundles
where $`L`$ is of rank 1 and the contact form $`\theta H^0(\mathrm{\Omega }_X^1L)`$ yields a nowhere vanishing section
$$\theta (d\theta )^nH^0(K_XL^{n+1})$$
—it is an elementary calculation to see that $`\theta (d\theta )^k`$ is a well-defined section of $`\mathrm{\Omega }_X^{2k+1}L^{k+1}`$ for all numbers $`nk1`$. In particular, we assume that $`K_X=(n+1)L`$.
The assumptions imply that the natural map $`[,]:FFL`$ derived from the Lie-bracket is non-degenerate. In view of the Frobenius theorem this means that if $`YX`$ is an $`F`$-integral submanifold, i.e. one where $`T_YF|_Y`$, then $`dimYn`$. If $`Y`$ is of maximal dimension $`dimY=n`$, then $`Y`$ is called “Legendrian”. Note that some authors (e.g. \[Hwa97\], \[KPSW00\]) prefer to use “Lagrangian” instead of “Legendrian”.
The usual Darboux theorem of real contact and symplectic geometry applies equally well in the complex case. Thus, for any point $`xX`$ we can find coordinates $`(z_i)_{i=1\mathrm{}2n+1}`$ centered about $`x`$ and a bundle coordinate for $`L`$ so that we can write
$$\theta =dz_{2n+1}+\underset{i=1\mathrm{}n}{}z_idz_{n+i}$$
In particular, we remark the following:
###### Remark 2.1.
If $`xX`$ is any point and $`\stackrel{}{v}T_{X,x}`$ any tangent vector, then there exists a Legendrian submanifold $`UX`$ which contains $`x`$ and is transversal to $`\stackrel{}{v}`$, i.e. $`\stackrel{}{v}T_U|x`$.
### 2.2. Parameter spaces
For the benefit of readers coming from differential geometry we will briefly recall some facts about the parameter spaces which we will use in the sequel. Our chief reference will be \[Kol96, chap. II\], and our notation will be compatible with this book. Mori’s paper \[Mor79\] on the Hartshorne conjecture is also recommended for these matters.
If $`V`$ is any projective manifold and $`C`$ a projective variety, then we will often parameterize those morphisms from $`C`$ to $`V`$ which are birational onto their images. In fact, there exists a scheme $`\mathrm{Hom}_{bir}(C,V)`$ whose geometric points correspond to these morphisms. Furthermore, there exists a “universal morphism”: $`\mu :\mathrm{Hom}_{bir}(C,V)\times CV`$. We refer to \[Kol96, chapt. II.1\] for an authoritative reference on this.
The tangent space to $`\mathrm{Hom}_{bir}(C,V)`$ is described as follows: If a birational morphism $`f:CV`$ is given, then the (Zariski-)tangent space of $`\mathrm{Hom}_{bir}(C,V)`$ at $`f`$ corresponds naturally to sections in $`H^0(C,f^{}(T_V))`$.
If $`cC`$ and $`vV`$ are (geometric) points, the subfamily of morphisms mapping $`c`$ to $`v`$ is usually denoted as $`\mathrm{Hom}_{bir}(C,V,cv)`$. The tangent space to $`\mathrm{Hom}_{bir}(C,V,cv)`$ at a point $`f`$ corresponds to $`H^0(C,f^{}(T_V)𝒥_c)`$, where $`𝒥_c`$ is the ideal sheaf of $`c`$.
In the special case that $`C_1`$, and $`f\mathrm{Hom}_{bir}(_1,V)`$ a morphism whose image contains a point $`vX`$, the Riemann-Roch theorem yields an estimate for the dimensions of the deformation spaces
(2.1)
$$\begin{array}{ccc}\hfill dim_{[f]}\mathrm{Hom}_{bir}(_1,V)& & K_V.\mathrm{}+dimV\hfill \\ \hfill dim_{[f]}\mathrm{Hom}_{bir}(_1,V,[0:1]v)& & K_X.\mathrm{}\hfill \end{array}$$
where $`\mathrm{}:=\text{Image}(f)`$. See \[Kol96, prop. II.1.13 and thm. II.1.7\] for an explanation and a proof.
The group $`SL_2`$ acts on the normalization $`\mathrm{Hom}_{bir}^n(_1,V)`$ of $`\mathrm{Hom}_{bir}(_1,V)`$, and the geometric quotient in the sense of Mumford \[FM82\] exists, see \[Mor79, lem. 9\]. More precisely, by \[Kol96, thm. II.2.15\] there exists a commutative diagram
(2.2)
where $`u`$ and $`U`$ are principal $`SL_2`$ bundles, $`\pi `$ is a $`_1`$-bundle and the restriction of $`\iota `$ to any fiber of $`\pi `$ is a morphism which is birational onto its image. We call the quotient space $`\mathrm{RatCurves}^n(V)`$ the “parameter space of rational curves on $`V`$”. The letter “$`n`$” in $`\mathrm{RatCurves}^n`$ may be a little confusing. It has nothing to do with the dimension of $`V`$, but serves as a reminder that the parameter space is isomorphic the normalization of a suitable quasiprojective subset of the Chow-variety.
### 2.3. Lines
Unless otherwise mentioned, throughout this work we will assume that $`X`$ is Fano and that $`b_2(X)=1`$. In this setup it follows from the classic work of Mori (\[Mor79\], but see also \[CKM88, lect. 1\]) that we can find an irreducible component $`H\mathrm{RatCurves}^n(X)`$ with the following properties:
1. the evaluation morphism $`\iota `$ is dominant
2. if $`xX`$ is a general point, then the subfamily $`H_x:=\pi (\iota ^1(x))H`$ (i.e. the subfamily which parameterizes curves containing $`x`$) is compact.
3. if $`\mathrm{}X`$ is a curve which is associated with a point of $`H`$, then $`1K_X.\mathrm{}dimX+1`$.
Since $`X`$ is a contact manifold, we have that $`K_X=(n+1)L`$, and it follows from point (3) that either $`L.\mathrm{}=2`$ or $`L.\mathrm{}=1`$.
If $`L.\mathrm{}=2`$, then the estimate (2.1) shows that
$$dim\mathrm{Hom}_{bir}(_1,X,[0:1]x)K_X.\mathrm{}=dimX+1.$$
Because there is a 2-dimensional group of automorphisms of $`_1`$ which fix $`[0:1]`$, the assumption that $`L.\mathrm{}=2`$ implies
$$dimH_xdimX1,$$
and it follows from the generalized Kobayashi-Ochiai theorem of \[Keb00, thm. 3.6\] or \[KS99, thm. 0.2\] that $`X_{2n+1}`$. For that reason we assume in the sequel that $`L.\mathrm{}=1`$. We call $`\mathrm{}`$ a “contact line”.
###### Remark 2.2.
The assumption that $`L.\mathrm{}=1`$ implies that the irreducible component $`H\mathrm{RatCurves}^n(X)`$ is compact. See \[Kol96, prop. II.2.14\].
###### Remark 2.3.
Let $`f:\stackrel{~}{\mathrm{}}\mathrm{}`$ be the normalization. Since $`\stackrel{~}{\mathrm{}}_1`$, and $`T_\stackrel{~}{\mathrm{}}𝒪__1(2)`$, it is clear that the map $`T_\stackrel{~}{\mathrm{}}f^{}(L)`$ must be trivial. It follows that $`T_{\mathrm{}}|_xF|_x`$ for all smooth points $`x\mathrm{}`$. We say that contact lines are $`F`$-integral where they are smooth.
## 3. Deformations of lines
We will show that lines passing through a general point are smooth. For this, we employ deformations of lines in order to obtain sections of $`L`$. The following local proposition shows that there are severe restrictions for such sections to exist.
###### Proposition 3.1.
Consider a family of morphisms $`\mathrm{\Phi }_t:\mathrm{\Delta }_CX`$ written as
$$\begin{array}{cccc}\mathrm{\Phi }:& \mathrm{\Delta }_{}\times \mathrm{\Delta }_C& & X\\ & (t,z)& & \mathrm{\Phi }_t(z)\end{array}$$
where $`\mathrm{\Delta }_{}`$ and $`\mathrm{\Delta }_C`$ are unit disks. Assume that for all $`t\mathrm{\Delta }_{}`$ the image $`\mathrm{\Phi }_t(\mathrm{\Delta }_C)`$ is $`F`$-integral, i.e. assume that $`\mathrm{\Phi }^{}(\theta )\left(\frac{}{z}\right)0`$. If
$$\sigma =\mathrm{\Phi }_0^{}(\theta )\left(\frac{}{t}\right)H^0(\mathrm{\Delta }_C,\mathrm{\Phi }_0^{}(L))$$
vanishes at $`0\mathrm{\Delta }_C`$ but does not vanish identically, then $`\sigma `$ vanishes with multiplicity at least two if and only if
(3.1)
$$\mathrm{\Phi }^{}(d\theta )(\frac{}{t}|_{(0,0)},\frac{}{z}|_{(0,0)})=0.$$
###### Remark 3.2.
The exterior derivative $`d\theta `$ which appears in equation (3.1) depends on a choice of bundle coordinates for $`L`$ and is therefore not well-defined. Note, however, that the requirement $`\sigma (0)=0`$ implies that $`T\mathrm{\Phi }\left(\frac{}{t}|_{(0,0)}\right)F|_{\mathrm{\Phi }(0,0)}`$, where $`T\mathrm{\Phi }`$ is the tangent map associated with $`\mathrm{\Phi }`$. In this setting an elementary calculation shows that the validity of equation (3.1) does in fact not depend on the choice of bundle coordinates.
###### Proof.
Recall the following formula: if $`\omega `$ is any 1-form on a manifold, and $`\stackrel{}{X}_0`$ and $`\stackrel{}{X}_1`$ are vector fields, then
(3.2)
$$d\omega (\stackrel{}{X}_0,\stackrel{}{X}_1)=\stackrel{}{X}_0(\omega (\stackrel{}{X}_1))\stackrel{}{X}_1(\omega (\stackrel{}{X}_2))\omega ([\stackrel{}{X}_0,\stackrel{}{X}_1]).$$
See e.g. \[War71, prop. 2.25(e) on p. 70\] for an explanation. We choose local bundle coordinates on $`\mathrm{\Phi }^{}(L)`$ and apply equation (3.2) with $`\omega =\mathrm{\Phi }^{}(\theta )`$, $`\stackrel{}{X}_0=\frac{}{z}`$ and $`\stackrel{}{X}_1=\frac{}{t}`$. Since $`\frac{}{t}`$ and $`\frac{}{z}`$ commute and $`\mathrm{\Phi }^{}(\theta )\left(\frac{}{z}\right)0`$, we obtain
$$\frac{}{z}\mathrm{\Phi }^{}(\theta )\left(\frac{}{t}\right)=d\mathrm{\Phi }^{}(\theta )(\frac{}{t},\frac{}{z}).$$
Note that $`d\mathrm{\Phi }^{}(\theta )=\mathrm{\Phi }^{}(d\theta )`$ and evaluate at $`t=0`$, $`z=0`$. ∎
Another argument which uses the deformation of lines shows that most lines are smooth.
###### Proposition 3.3.
If $`xX`$ is a general point and $`\mathrm{}`$ a contact line passing through $`x`$, then $`\mathrm{}`$ is smooth.
###### Proof.
Assume to the contrary, i.e. assume that for a general point $`xX`$ there exists a singular line $`\mathrm{}`$ passing through $`x`$. Recall that the rational curve $`\mathrm{}`$ can always be dominated by an integral singular plane cubic, i.e. by a rational curve with a single node or cusp. We will reach a contradiction by constructing a section of the pull-back of $`L`$ to the plane cubic which vanishes at a prescribed generically chosen point. This section will be constructed by a deformation of the singular curves.
Because $`x`$ was chosen to be general, there exists a singular (i.e. cuspidal or nodal) plane cubic $`C_2`$ and an irreducible component $`\mathrm{Hom}^{\mathrm{birat}}(C,X)`$ such that the universal morphism $`\mu :\times CX`$ is dominant and such that for all $`f`$ we have $`\mathrm{deg}f^{}(L)=1`$.
Fix a general morphism $`f`$ and note that there is an open set $`UC`$ such that for all $`cU`$, the tangent map of the restricted morphism $`\mu _c:=\mu |_{\times \{c\}}`$ has maximal rank at $`f`$:
$$\mathrm{rank}_{[f]}T\mu _c=dimX=2n+1.$$
Recall from \[Har77, II.6.10.2, II.6.11.4 and Ex. II.6.7\] that the smooth points of $`C`$ are in 1:1-correspondence with line bundles of degree one, and fix a point $`cU`$ such that $`𝒪_C(c)\cong ̸f^{}(L)`$.
Next, let $`U_XX`$ be a Legendrian submanifold of $`X`$ which contains $`x`$ and is transversal to $`f(C)`$ at $`x`$. By remark 2.1, these exist in abundance. Since $`\mu _c`$ has maximal rank, we can find a section $`U_{}`$ over $`U_X`$, i.e. a submanifold $`U_{}`$ such that $`\mu _c|_U_{}:U_{}U_X`$ is an isomorphism. By construction, $`\mu (U_{}\times C)`$ has dimension $`n+1`$ and cannot be Legendrian. It follows that there exists a unit disc $`\mathrm{\Delta }_{}U_{}`$ with coordinate $`t`$ centered about $`f=\{t=0\}`$ such that
$$\sigma :=f^{}(\theta )\left(\frac{}{t}|_{t=0}\right)H^0(C,f^{}(L))\{0\}.$$
For this, recall that the tangent vector $`\frac{}{t}|_{t=0}T_{}|_f`$ is canonically identified with an element in $`H^0(C,f^{}(T_X))`$. By choice of $`U_{}`$, we have $`\sigma (c)=0`$. By choice of $`c`$, this is impossible, a contradiction. ∎
It is an immediate corollary that a tangent morphism exists.
###### Corollary 3.4.
If $`xX`$ is a general point, then there exists a tangent morphism $`\tau _x:H_x(F^{}|_x)`$ which sends a line $`\mathrm{}`$ to its tangent space $`T_{\mathrm{}}|_xF|_x`$.
The morphism $`\tau _x`$ was already studied in \[Hwa97\]. It was, however, not all clear at that time that $`\tau _x`$ really is a morphism and not just a rational map. See \[Keb00\] for a weaker but more general result.
Finally, we remark that deformations of lines through a general point are unobstructed in a strong sense.
###### Lemma 3.5.
If $`xX`$ is a general point and $`\mathrm{}`$ any line through $`x`$, then
$$T_X|_{\mathrm{}}𝒪_{\mathrm{}}(2)𝒪_{\mathrm{}}(1)^{n1}𝒪_{\mathrm{}}^{n+1}.$$
###### Proof.
It follows from the definition of the contact structure that $`FF^{}L`$. Since $`L|_{\mathrm{}}𝒪_{\mathrm{}}(1)`$, and since vector bundles on $`_1`$ always decompose into sums of line bundles we may therefore write
$$F|_{\mathrm{}}\underset{i=1}{\overset{n}{}}\left(𝒪_{\mathrm{}}(a_i)𝒪_{\mathrm{}}(1a_i)\right)$$
where $`a_i>0`$. Thus, the splitting of $`F|_{\mathrm{}}`$ has exactly $`n`$ positive entries. It follows that the splitting of $`T_X|_{\mathrm{}}`$ has at most $`n`$ positive entries. By \[KMM92, prop. 1.1\], $`T_X|_{\mathrm{}}`$ is nef, and since $`c_1(T_X|_{\mathrm{}})=n+1`$, the claim follows. ∎
We will apply lemma 3.5 to study the locus of lines through a general point. For this, fix a general point $`xH`$, define the subfamily $`H_xH`$ as in section 2.3 and consider the restricted diagram associated to diagram 2.2 on page 2.2.
Here $`\stackrel{~}{H_x}`$ is the normalization of $`H_x`$, $`\stackrel{~}{U_x}`$ the normalization of the pull-back $`\mathrm{Univ}^{rc}(X)\times _{\mathrm{RatCurves}^n(X)}\stackrel{~}{H_x}`$ and $`\mathrm{locus}(H_x)=\iota (\pi ^1(H_x))`$. Recall from remark 2.2 that $`H`$ and therefore $`H_x`$ are compact. In particular, $`\mathrm{locus}(H_x)`$ is a proper subvariety of $`X`$.
It follows from \[Kol96, thms. II.3.11.5 and II.2.8\] that $`\stackrel{~}{H}_x`$ is smooth and $`\stackrel{~}{\pi }_x`$ a $`_1`$-bundle. As an immediate corollary to the preceding lemma, we obtain that both $`\stackrel{~}{\iota }_x`$ and $`\stackrel{~}{\tau }_x`$ are immersive.
###### Corollary 3.6.
If $`xX`$ is a general point, then
1. the universal morphism $`\stackrel{~}{\iota }_x:\stackrel{~}{U}_x\mathrm{locus}(H_x)X`$ is a birational immersion away from a section $`\sigma _{\mathrm{}}`$ which is contracted to $`x`$.
2. the tangent map $`\stackrel{~}{\tau }_x:\stackrel{~}{H}_x(F^{}|_x)`$ is also an immersion
###### Proof.
The fact that $`\stackrel{~}{\iota }_x`$ and $`\stackrel{~}{\tau }_x`$ are immersive follows from \[Kol96, props. II.3.4 and II.3.10\] and lemma 3.5. It follows from an argument of Miyaoka that $`\stackrel{~}{\iota }_x`$ is birational because all lines through $`x`$ are smooth. For this, see \[Kol96, prop. V.3.7.5\]. ∎
## 4. Lines through a fixed point
For a better understanding of contact Fano manifolds, the locus of lines through a given point is of greatest interest. The following proposition gives a first description. This result is contained implicitly in \[KPSW00, sect. 2\] and we could have used the results of that paper here, but we prefer to give a short and self-contained proof in our context.
###### Proposition 4.1.
If $`xX`$ is any point, then $`\mathrm{locus}(H_x)`$ has dimension $`n`$ and is $`F`$-integral where it is smooth.
###### Proof.
Since $`K_X.\mathrm{}=n+1`$, it follows from the estimate (2.1) for the dimension of the parameter space that
$$dim\mathrm{Hom}_{bir}(_1,X,[0:1]x)n+2.$$
By Mori’s bend-and-break argument \[Kol96, thm. II.5.4\], for a given point $`yX\{x\}`$, there are at most finitely many lines containing both $`x`$ and $`y`$. It follows that
$$dim\mathrm{locus}(H_x)=dim\mathrm{Hom}(_1,X,[0:1]x)dim\mathrm{Aut}(_1,[0:1])n.$$
### Claim
The subvariety $`\mathrm{locus}(H_x)`$ is $`F`$-integral where it is smooth.
### Application of the claim
It follows immediately from Frobenius’ theorem and from the non-degeneracy of the contact distribution that $`dim\mathrm{locus}(H_x)n`$, and we are done.
### Proof of the claim
Let $`y\mathrm{locus}(H_x)`$ be a general (smooth) point, $`\mathrm{}H_x`$ a curve which contains $`x`$ and $`y`$ and is smooth at $`y`$. By general choice of $`y`$, such a curve can always be found. Let $`f:_1\mathrm{}`$ be a birational morphism with $`f([0:1])=x`$ and $`f([1:1])=y`$. If
$$\mathrm{Hom}^{\mathrm{birat}}(_1,X,[0:1]x)$$
is an irreducible component of the reduced Hom-scheme which contains $`f`$, then we have that
(4.1)
$$T_{\mathrm{locus}(H_x)}|_yT_{\mathrm{}}|_y+\mathrm{Image}(T_{}|_fT_X|_y)$$
where $`T_{}|_f`$ is identified with $`H^0(_1,f^{}(T_X)𝒪__1([0:1]))`$ and the map $`T_{}|_fT_X|_y`$ is an application of the tangent map $`Tf`$ and evaluation at $`y`$. In other words, the tangent space $`T_{\mathrm{locus}(H_x)}|_y`$ at $`y`$ is spanned by the tangent space to the curve $`\mathrm{}`$ and by sections of $`f^{}(T_X)`$ which vanish at $`[0:1]`$. We refer to \[Kol96, prop. II.3.4\] for a proof of (4.1).
In remark 2.3 we have already seen that $`T_{\mathrm{}}F|_{\mathrm{}}`$ so that it suffices to prove that
$$\text{Image}(T_{}|_fT_X|_y)F|_y$$
In other words, we have to show that if $`\mathrm{\Delta }_{}`$ is any unit disk centered about $`f`$ with coordinate $`t`$, then the section $`\sigma H^0(_1,f^{}(T_X))`$ associated with $`\frac{}{t}|_{t=0}`$ is contained in $`H^0(_1,f^{}(F))`$. For this, note that proposition 3.1 asserts that the section $`f^{}(\theta )(\sigma )H^0(_1,f^{}(L))`$ has a zero at $`0_1`$ whose order is at least two. But since $`\mathrm{deg}f^{}(L)=1`$, this implies that $`f^{}(\theta )(\sigma )`$ vanishes identically. In particular, $`\sigma H^0(_1,f^{}(F))`$.
This proves that $`T_{\mathrm{locus}(H_x)}|_yF|_y`$. Since $`y`$ was chosen generically, $`\mathrm{locus}(H_x)`$ is $`F`$-integral where it is smooth, and we are done. ∎
Under the assumptions spelled out in section 2, if $`xX`$ is a general point, then the contact distribution $`F|_x`$ is generated by the tangent spaces to lines through $`x`$. Hence, it is canonically given. Before starting the proof, however, it is convenient to introduce the following notation first.
###### Notation 4.2.
Consider the incidence variety
$$V:=\{(x^{},x^{\prime \prime })X\times X|x^{\prime \prime }\mathrm{locus}(x^{})\}X\times X.$$
An elementary calculation shows that $`V`$ is a closed subvariety of $`X\times X`$. We call $`V`$ the “universal locus of lines through points”.
Let $`\pi _1,\pi _2:VX`$ are the projection morphisms. Then for every $`xX`$ we have that (set-theoretically) $`\pi _2(\pi _1^1(x))=\mathrm{locus}(H_x)`$. It may well happen that $`\pi _1^1(x)`$ is not reduced for special points $`xX`$.
If $`YX`$ is a subset, we shall write $`V|_Y`$ for $`\pi _1^1(Y)`$.
###### Lemma 4.3.
Let $`VX\times X`$ be the universal locus of lines through points which we defined above. Let $`\mathrm{\Delta }`$ be a unit disk with coordinate $`t`$ and $`\gamma :\mathrm{\Delta }X`$ an embedding. Then there exists an open set $`V^0V|_{\gamma (\mathrm{\Delta })}`$ such that $`\pi _2(V^0)`$ is a submanifold of dimension
$$dim\pi _2(V^0)=n+1.$$
In particular, by Frobenius’ theorem, $`\pi _2(V^0)`$ is not $`F`$-integral.
###### Proof.
We have already seen in proposition 4.1 that $`\pi _1|_V`$ is equidimensional of relative dimension $`n`$. Thus, $`V`$ is a well-defined family of algebraic cycles over $`X`$ in the sense of \[Kol96, I.3.10\] and the universal property of the Chow-variety yields a map $`\varphi :X\mathrm{Chow}(X)`$ such that $`V`$ is the pull-back of the universal family over $`\mathrm{Chow}(X)`$. Because $`dim\mathrm{locus}(H_x)=n<dimX`$, it is clear that the image of $`\varphi `$ is not a point. Use the assumption that $`b_2(X)=1`$ to obtain that $`\varphi `$ is actually a finite morphism. Because two reduced algebraic cycles are equal if and only if their supports agree, it follows that for a given point $`x_0X`$, there are at most finitely many points $`(x_i)_{i=1\mathrm{}k}X`$ such that
$$\mathrm{locus}(H_{x_0})=\mathrm{locus}(H_{x_i}).$$
In particular, if $`V^0V|_{\gamma (\mathrm{\Delta })}`$ is an open set such that $`\pi _2|_{V^0}`$ is an embedding, then $`\pi _2(V^0)`$ has dimension $`n+1`$. Hence the claim. ∎
With these preparations we can now start the proof of the main theorem of this work.
###### Theorem 4.4.
If $`xX`$ is a general point, then $`F|_x`$ is spanned by the image of the tangent map $`\tau _x`$.
###### Proof.
Our argument involves an analysis of the deformations of $`\mathrm{locus}(H_y)`$ which are obtained by varying the base point $`y`$. We shall argue by contradiction and assume that the assertion of the proposition is wrong. With this assumption we will construct a family of morphisms $`_1X`$ which contradicts proposition 3.1, and we are done.
We will now produce a map $`\gamma `$ to which lemma 4.3 can be applied. Assuming that the statement of the proposition is wrong, we can find an analytic open neighborhood $`U=U(x)X`$ and a subbundle $`F^{}F|_U`$ such that
1. For all $`yU`$, the vector spaces $`F^{}|_y`$ and $`\mathrm{Span}(\mathrm{Image}\tau _y)F|_y`$ are perpendicular with respect to the non-degenerate form $`FFL`$ which comes with the contact structure.
2. All lines through $`U`$ are smooth.
After shrinking $`U`$, if necessary, let $`\stackrel{}{v}H^0(U,F^{})`$ be a nowhere-vanishing vector field. Thus, if $`yU`$ is any point and $`\mathrm{}y`$ is any line through $`y`$, then
(4.2)
$$T_{\mathrm{}}|_y\stackrel{}{v}(y)^{},$$
where “$``$” again means: perpendicular with respect to the non-degenerate form on $`F`$. Let $`\mathrm{\Delta }`$ be a unit disc with coordinate $`t`$ and $`\gamma :\mathrm{\Delta }X`$ be an integral curve of $`\stackrel{}{v}`$ with $`\gamma (0)=x`$.
Now let $`\mathrm{Hom}_{bir}(_1,X)`$ be the family of morphisms parameterizing the curves associated with $`H`$. Set
$$_\mathrm{\Delta }:=\{f|f([0:1])\gamma (\mathrm{\Delta })\}.$$
If $`\mu _\mathrm{\Delta }:_\mathrm{\Delta }\times _1X`$ is the universal morphism, then it follows by construction that
$$\mu _\mathrm{\Delta }(_\mathrm{\Delta }\times _1)=\pi _2(V|_{\gamma (\mathrm{\Delta })})\pi _2(V^0).$$
In particular, since $`\pi _2(V^0)`$ is not $`F`$-integral, for a general point $`(f,p)_\mathrm{\Delta }\times _1`$ there exists a tangent vector $`\stackrel{}{w}T_{_\mathrm{\Delta }\times _1}|_{(f,p)}`$ such that the image of the tangent map is not in $`F`$:
$$T\mu _\mathrm{\Delta }(\stackrel{}{w})F$$
Decompose $`\stackrel{}{w}=\stackrel{}{w}^{}+\stackrel{}{w}^{\prime \prime }`$, where $`\stackrel{}{w}T__1|_p`$ and $`\stackrel{}{w}^{\prime \prime }T__\mathrm{\Delta }|_f`$. Then, since $`f(_1)`$ is $`F`$-integral, it follows that $`T\mu _\mathrm{\Delta }(\stackrel{}{w}^{})F`$ and therefore
(4.3)
$$T\mu _\mathrm{\Delta }(\stackrel{}{w}^{\prime \prime })F.$$
As a next step, choose an immersion
$$\begin{array}{cccc}\hfill \beta :& \hfill \mathrm{\Delta }& & _\mathrm{\Delta }\hfill \\ & \hfill t& & \beta _t\hfill \end{array}$$
such that $`\beta _0=f`$ and such that
$$T\beta \left(\frac{}{t}|_{t=0}\right)=\stackrel{}{w}^{\prime \prime }$$
In particular, if $`\sigma H^0(_1,f^{}(T_X))`$ is the section associated with $`\stackrel{}{w}^{\prime \prime }=T\beta (\frac{}{t}|_{t=0})`$, and $`\sigma ^{}:=f^{}(\theta )(\sigma )H^0(_1,f^{}(L))`$, then the following holds:
1. it follows from (4.3) and from \[Kol96, prop. II.3.4\] that $`\sigma ^{}`$ is not identically zero.
2. at $`[0:1]_1`$, the section $`\sigma `$ satisfies $`\sigma ([0:1])f^{}(T_{\gamma (\mathrm{\Delta })})f^{}(F)`$. In particular, $`\sigma ^{}([0:1])=0`$.
3. If $`z`$ is a local coordinate on $`_1`$ about $`[0:1]`$, then it follows from (4.2) that $`\frac{}{z}|_{[0:1]}f^{}(F)`$ and $`\sigma ([0:1])f^{}(F^{})`$ are perpendicular with respect the the non-degenerate form.
Items (2) and (3) ensure that we can apply proposition 3.1 to the family $`\beta _t`$. Since the section $`\sigma ^{}`$ does not vanish completely, the proposition states that $`\sigma ^{}`$ has a zero of order at least two at $`[0:1]`$. But $`\sigma ^{}`$ is an element of $`H^0(_1,f^{}(L))`$, and $`f^{}(L)`$ is a line bundle of degree one. We have thus reached a contradiction, and the proof of theorem 4.4 is therefore finished. ∎
It follows that there are only two types of contact manifolds whose structure is not unique.
###### Corollary 4.5.
If $`X`$ is any complex projective contact manifold with more than one contact structure, then either $`X_{2n+1}`$ or $`X(T_Y)`$ for a manifold $`Y`$ whose tangent bundle $`T_Y`$ has bundle automorphisms.
We refer to \[KPSW00, sect. 2.6\] for a study of the different contact structures on $`(T_Y)`$.
###### Proof.
By \[KPSW00, prop. 3.1\], the canonical bundle $`\omega _X`$ is not nef. But then it has already been shown in \[KPSW00, thm. 1.1\] that $`X`$ is automatically of type $`(T_Y)`$ if $`b_2(X)>1`$. We may therefore assume without loss of generality that $`X`$ is Fano and that $`b_2(X)=1`$.
Let $`xX`$ be a general point, and $`\mathrm{}x`$ a minimal rational curve through $`x`$. It follows from the classical argument of Mori that $`K_X.\mathrm{}dimX+1`$. Since $`X`$ is a contact manifold, $`K_X=(n+1)L`$ so that $`L.\mathrm{}\{1,2\}`$. If $`L.\mathrm{}=2`$, then $`X_{2n+1}`$. If $`L.\mathrm{}=1`$, then theorem 4.4 shows that the contact distribution $`F`$ is canonically defined, hence unique. ∎
|
warning/0004/hep-ph0004145.html
|
ar5iv
|
text
|
# DESY 00-056 April 2000 QUANTUM MECHANICS OF BARYOGENESIS
## Abstract
The cosmological baryon asymmetry can be explained as remnant of heavy Majorana neutrino decays in the early universe. We study this out-of-equilibrium process by means of Kadanoff-Baym equations which are solved in a perturbative expansion. To leading order the problem is reduced to solving a set of Boltzmann equations for distribution functions.
The generation of the cosmological matter-antimatter asymmetry in an expanding universe requires baryon number violation, $`C`$ and $`CP`$ violation, and a deviation from thermal equilibrium . The classic mechanism which realizes these conditions is the decay of weakly interacting massive particles in a thermal bath . Particularly successful is the leptogenesis scenario where the decaying particles are heavy Majorana neutrinos . The resulting baryon asymmetry is entirely determined by neutrino properties. The observed order of magnitude can be naturally explained without any fine tuning of parameters and in accord with present experimental indications for neutrino masses .
The generation of a baryon asymmetry is an out-of-equilibrium process which is generally treated by means of Boltzmann equations. A thorough discription of the basic ideas can be found in . Some subtleties have recently been discussed in . A shortcoming of this approach is that the Boltzmann equations are classical equations for the time evolution of phase space distribution functions. On the contrary, the involved collision terms are $`S`$-matrix elements which involve quantum interferences of different amplitudes in a crucial manner. Clearly, a full quantum mechanical treatment is highly desirable. It is also required in order to justify the use of Boltzmann equations and to determine the size of corrections.
All information about the time evolution of a system is contained in the time dependence of its Green functions , which can be determined by means of Dyson-Schwinger equations. Originally these techniques were developed for non-relativistic many-body problems. More recently, they have also been applied to transport phenomena in nuclear matter , the electroweak plasma and the QCD plasma . Alternatively, one may study the time evolution of density matrices . In the following we shall investigate non-equilibrium Green functions which are relevant for leptogenesis. We shall construct a perturbative solution of the corresponding Kadanoff-Baym equations which, to leading order, turn out to be equivalent to a set of Boltzmann equations. Higher-order corrections can then be systematically evaluated.
Consider now the standard model with three additional right-handed neutrinos whose interactions are described by the lagrangian,
$$=\overline{l}_L\stackrel{~}{\varphi }\lambda ^{}\nu _R\frac{1}{2}\overline{\nu }_R^cM\nu _R+h.c.$$
(1)
Here $`l_L`$ and $`\varphi `$ denote lepton and Higgs doublets, respectively. We shall restrict our discussion to the case of hierarchical Majorana neutrino masses, $`M_1M_2,M_3`$. The baryon asymmetry will then be determined by the $`CP`$ violating decays of the lightest Majorana neutrino $`N_1=\nu _{R1}+\nu _{R1}^cN`$,
$$\mathrm{\Gamma }(Nl\varphi )=\frac{1}{2}(1+ϵ)\mathrm{\Gamma },\mathrm{\Gamma }(N\overline{l}\overline{\varphi })=\frac{1}{2}(1ϵ)\mathrm{\Gamma }.$$
(2)
Here $`\mathrm{\Gamma }`$ is the total decay width and the parameter $`ϵ1`$ measures the amount of $`CP`$ violation. The generation of the baryon asymmetry takes place at a temperature $`TM_1MM_2,M_3`$. It is therefore convenient to describe the system by an effective lagrangian where the two heavier neutrinos have been integrated out,
$``$ $`=`$ $`\overline{l}_{Li}\stackrel{~}{\varphi }\lambda _{i1}^{}N+N^T\lambda _{i1}Cl_{Li}\varphi {\displaystyle \frac{1}{2}}MN^TCN`$ (3)
$`+{\displaystyle \frac{1}{2}}\eta _{ij}l_{Li}^T\varphi Cl_{Lj}\varphi +{\displaystyle \frac{1}{2}}\eta _{ij}^{}\overline{l}_{Li}\stackrel{~}{\varphi }C\overline{l}_{Lj}^T\stackrel{~}{\varphi },`$
with
$$\eta _{ij}=\underset{k=2}{\overset{3}{}}\lambda _{ik}\frac{1}{M_k}\lambda _{kj}^T.$$
(4)
For leptogenesis one has to consider the phase space distributions for heavy neutrinos ($`f_N`$), leptons ($`f_l`$), anti-leptons ($`f_{\overline{l}}`$), Higgs ($`f_\varphi `$) and anti-Higgs bosons ($`f_{\overline{\varphi }}`$). The generation of the lepton asymmetry is a process close to equilibrium. Hence one can linearize the Boltzmann equations in the deviations from the equilibrium distributions. Due to the interactions in (3) one has $`\delta f_l=\delta f_{\overline{l}}=\delta f_\varphi =\delta f_{\overline{\varphi }}`$. The Boltzmann equation for the Majorana neutrino reads
$`g_N{\displaystyle \frac{}{t}}\delta f_N(t,p)=`$ (5)
$`g_N{\displaystyle \frac{}{t}}f_N(p){\displaystyle \frac{1}{2E}}{\displaystyle 𝑑\mathrm{\Phi }_{\overline{1}\overline{2}}(p)\delta f_N(t,p)\left(|(Nl\varphi )|^2+|(N\overline{l}\overline{\varphi })|^2\right)}.`$
For the lepton doublets one obtains
$`2g_l{\displaystyle \frac{}{t}}\delta f_l(t,k)=`$
$`{\displaystyle \frac{1}{2k}}{\displaystyle 𝑑\mathrm{\Phi }_{\overline{1}2}(k)ϵ\delta f_N(t,p_1)\left(|(Nl\varphi )|^2+|(N\overline{l}\overline{\varphi })|^2\right)}`$
$`{\displaystyle \frac{1}{2k}}{\displaystyle 𝑑\mathrm{\Phi }_{1\overline{2}}(k)\left(\delta f_l(t,k)f_\varphi (p_1)+f_l(k)\delta f_\varphi (t,p_1)\right)}`$
$`\times \left(|(l\varphi N)|^2+|(\overline{l}\overline{\varphi }N)|^2\right)`$
$`{\displaystyle \frac{1}{2k}}{\displaystyle 𝑑\mathrm{\Phi }_{1\overline{2}\overline{3}}(k)\left(\delta f_l(t,k)f_\varphi (p_1)+f_l(k)\delta f_\varphi (t,p_1)\right)}`$
$`\times \left(|(l\varphi \overline{l}\overline{\varphi })|^2+|(\overline{l}\overline{\varphi }l\varphi )|^2\right)`$
$`{\displaystyle \frac{1}{2k}}{\displaystyle 𝑑\mathrm{\Phi }_{\overline{1}\overline{2}3}(k)\left(\delta f_l(t,p_1)f_\varphi (p_2)+f_l(p_1)\delta f_\varphi (t,p_2)\right)}`$
$`\times \left(|(l\varphi \overline{l}\overline{\varphi })|^2+|(\overline{l}\overline{\varphi }l\varphi )|^2\right)`$
$`{\displaystyle \frac{1}{4k}}{\displaystyle 𝑑\mathrm{\Phi }_{1\overline{2}\overline{3}}(k)\left(\delta f_l(t,k)f_l(p_1)+f_l(k)\delta f_l(t,p_1)\right)}`$
$`\times \left(|(ll\overline{\varphi }\overline{\varphi })|^2+|(\overline{l}\overline{l}\varphi \varphi )|^2\right)`$
$`{\displaystyle \frac{1}{2k}}{\displaystyle 𝑑\mathrm{\Phi }_{\overline{1}\overline{2}3}(k)\delta f_\varphi (t,p_1)f_\varphi (p_2)\left(|(\varphi \varphi \overline{l}\overline{l})|^2+|(\overline{\varphi }\overline{\varphi }ll)|^2\right)}.`$ (6)
Here
$$d\mathrm{\Phi }_{1\mathrm{}\overline{n}\mathrm{}}(p)=\frac{d^3p_1}{(2\pi )^32E_1}\mathrm{}\frac{d^3p_{\overline{n}}}{(2\pi )^32E_{\overline{n}}}\mathrm{}(2\pi )^4\delta ^4(p+p_1+\mathrm{}p_{\overline{n}}\mathrm{}),$$
(7)
$$f_i(p)=\mathrm{exp}(\beta E_i(p)),$$
(8)
denote phase space integrations and distribution functions, respectively. The temperature $`T=1/\beta `$, and $`(\mathrm{})`$ is the matrix element of the indicated process. $`g_N=2`$ and $`g_l=g_{\overline{l}}=6`$ are the number of ‘internal’ degrees of freedom for the Majorana neutrino and the lepton doublets for three generations, respectively. For simplicity we have assumed small number densities so that we can use Boltzmann distribution functions for bosons and fermions and also neglect distribution functions for particles in the final state. The effect of the Hubble expansion is included by introducing the ‘covariant’ derivative $`/t/tHp/p`$. Integration over momenta then yields the more familiar form of the Boltzmann equations for the number densities. Eq. (5) describes the decay of the heavy Majorana neutrinos. Note, that also the equilibrium distributions are time dependent since the temperature varies with time. For massless particles the distribution functions are constant with respect to the ‘covariant’ time derivative. The first term in eq. (6) drives the generation of a lepton asymmetry; the remaining terms tend to wash out an existing asymmetry. Eqs. (5) and (6) determine $`\delta f_N`$ and $`\delta f_l`$ as function of time. We have only kept the interactions given by the lagrangian (3). A complete discussion can be found in .
Green functions near thermal equilibrium
The time evolution of an arbitrary multi-particle lepton-Higgs system can be studied by means of the Green functions of lepton and Higgs fields. For the heavy Majorana neutrino one has
$$iG_{\alpha \beta }(x_1,x_2)=\text{Tr}\left(\rho TN_\alpha (x_1)N_\beta (x_2)\right),$$
(9)
where $`T`$ denotes the time ordering, $`\rho `$ is the density matrix of the system, the trace extends over all states, and the time coordinates $`t_1`$ and $`t_2`$ lie on an appropriately chosen contour $`C`$ in the complex plane . $`G(x_1,x_2)`$ can be written as a sum of two parts,
$$G(x_1,x_2)=\mathrm{\Theta }(t_1t_2)G^>(x_1,x_2)+\mathrm{\Theta }(t_2t_1)G^<(x_1,x_2),$$
(10)
where
$$iG^>(x_1,x_2)_{\alpha \beta }=\text{Tr}\left(\rho N_\alpha (x_1)N_\beta (x_2)\right),iG^<(x_1,x_2)_{\alpha \beta }=\text{Tr}\left(\rho N_\beta (x_2)N_\alpha (x_1)\right).$$
(11)
The ‘time ordering’ in eq. (10) is along the contour $`C`$.
For a system in thermal equilibrium at a temperature $`T=1/\beta `$ the density matrix is $`\rho =\mathrm{exp}(\beta H)`$, where $`H`$ is the Hamilton operator. In this case the Green function only depends on the difference of coordinates and it is convenient to introduce the Fourier transform,
$$G(p)=d^4xe^{ipx}G(x).$$
(12)
The contour $`C`$ can be chosen as a sum of two branches, $`C=C_1C_2`$, which lie above and below the real axis. The time coordinates are real and associated with one of the two branches. Correspondingly, the Green function becomes a $`2\times 2`$ matrix,
$`G(p)=\left(\begin{array}{cc}G^{11}(p)& G^{12}(p)\\ G^{21}(p)& G^{22}(p)\end{array}\right).`$ (15)
The off-diagonal terms are given by
$$G^{12}(p)=G^<(p),G^{21}(p)=G^>(p).$$
(16)
The diagonal terms of the matrix (15) are the familiar causal and anti-causal Green functions. The functions $`G^>(p)`$ and $`G^<(p)`$ satisfy the KMS-condition,
$$G^<(p)=e^{\beta p_0}G^>(p),$$
(17)
and the free Green functions are explicitly given by
$`iG^>(p)`$ $`=`$ $`\left(\mathrm{\Theta }(p_0)\mathrm{\Theta }(p_0)f_N(E)\mathrm{\Theta }(p_0)f_{\overline{N}}(E)\right)\rho _N(p),`$ (18)
$`iG^<(p)`$ $`=`$ $`\left(\mathrm{\Theta }(p_0)\mathrm{\Theta }(p_0)f_N(E)\mathrm{\Theta }(p_0)f_{\overline{N}}(E)\right)\rho _N(p),`$ (19)
with the spectral density
$$\rho _N(p)=2\pi (/p+M)C^1\delta (p^2M^2),$$
(20)
and the Fermi-Dirac distribution functions
$$f_N(E)=f_{\overline{N}}(E)=\frac{1}{e^{\beta E}+1},E=\sqrt{M^2+p^2}.$$
(21)
Since $`N(x)`$ is a Majorana field one has $`f_N=f_{\overline{N}}`$, and in the spectral density (20) the charge conjugation matrix $`C`$ occurs. In the following we shall also need the retarded and advanced Green functions,
$$G^\pm (x)=\pm \mathrm{\Theta }(\pm x^0)\left(G^>(x)G^<(x)\right),$$
(22)
which can be written as sum of an on-shell and an off-shell contribution,
$$G^\pm (p)=\pm \frac{1}{2}\left(G^>(p)G^<(p)\right)+\frac{1}{2\pi i}𝒫𝑑\omega ^{}\frac{G^>(x,\omega ^{},\stackrel{}{p})G^<(x,\omega ^{},\stackrel{}{p})}{\omega \omega ^{}}.$$
(23)
The Green functions for the lepton doublets and for the Higgs doublet,
$$iS(x_1,x_2)_{\alpha \beta }\delta _b^a=\text{Tr}\left(\rho Tl_\alpha ^a(x_1)\overline{l}_{b\beta }(x_2)\right),i\mathrm{\Delta }(x_1,x_2)\delta _b^a=\text{Tr}\left(\rho T\varphi ^a(x_1)\varphi _b^{}(x_2)\right),$$
(24)
have the same structure as $`G(x_1,x_2)`$. The corresponding equations for $`S(p)`$ are obtained from eqs. (10)-(21) by replacing the spectral density $`\rho _N(p)`$ by
$$\rho _l(p)=2\pi P_L/p\delta (p^2M^2),P_L=\frac{1\gamma _5}{2},$$
(25)
and the distribution functions $`f_N(E)`$ and $`f_{\overline{N}}(E)`$ by
$$f_l(E,\mu _l)=f_{\overline{l}}(E,\mu _l)=\frac{1}{e^{\beta (E\mu _l)}+1},E=|\stackrel{}{p}|,$$
(26)
where $`\mu _l`$ is the lepton chemical potential. For the Higgs field one has
$$\rho _\varphi (p)=2\pi \delta (p^2M^2),$$
(27)
$$f_\varphi (E,\mu _\varphi )=f_{\overline{\varphi }}(E,\mu _\varphi )=\frac{1}{e^{\beta (E\mu _\varphi )}1},E=|\stackrel{}{p}|.$$
(28)
We are considering a process close to equilibrium. This suggests that the corresponding deviations of the Green functions may be obtained from the equilibrium Green functions by a small change of the distribution functions,
$$i\delta G(x,p)=\delta f_N(x,p)\rho _N(p),$$
(29)
$$i\delta S(x,k)=ϵ(k_0)\delta f_l(x,k)\rho _l(k),i\delta \mathrm{\Delta }(x,q)=ϵ(q_0)\delta f_\varphi (x,q)\rho _\varphi (q).$$
(30)
Here we have used that due to the interactions given in (3) $`\delta f_l=\delta f_{\overline{l}}=\delta f_\varphi =\delta f_{\overline{\varphi }}`$.
Kadanoff-Baym equations
The Green functions for the heavy neutrino and the leptons satisfy Dyson-Schwinger equations,
$`C(i/_1M)G(x_1,x_2)`$ $`=`$ $`\delta (x_1x_2)+{\displaystyle _C}d^4x_3\mathrm{\Sigma }(x_1,x_3)G(x_3,x_2),`$ (31)
$`i/_1S(x_1,x_2)`$ $`=`$ $`\delta (x_1x_2)+{\displaystyle _C}d^4x_3\mathrm{\Pi }(x_1,x_3)S(x_3,x_2),`$ (32)
where $`\mathrm{\Sigma }`$ and $`\mathrm{\Pi }`$ are the corresponding self energies and the time integration is carried out along the contour $`C`$. Eqs. (31) and (32) can be turned into matrix equations with real time integration in the usual manner. For the off-diagonal elements $`G^>`$ and $`S^>`$ one then obtains
$`C(i/_1M)G^>(x_1,x_2)`$ $`=`$ $`{\displaystyle }d^4x_3(\mathrm{\Sigma }^>(x_1,x_3)G^{}(x_3,x_2)`$ (33)
$`+\mathrm{\Sigma }^+(x_1,x_3)G^>(x_3,x_2)),`$
$`i/_1S^>(x_1,x_2)`$ $`=`$ $`{\displaystyle }d^4x_3(\mathrm{\Pi }^>(x_1,x_3)S^{}(x_3,x_2)`$ (34)
$`+\mathrm{\Pi }^+(x_1,x_3)S^>(x_3,x_2)).`$
Here $`\mathrm{\Sigma }^>`$ and $`\mathrm{\Pi }^>`$ are off-diagonal matrix elements of $`\mathrm{\Sigma }`$ and $`\mathrm{\Pi }`$, which are defined analogous to $`G^>`$. The equations for $`G^<`$ and $`S^<`$ can be obtained from eqs. (33) and (34) by replacing the superscripts ‘$`>`$’ by ‘$`<`$’. Equations of the type (33), (34) have first been obtained by Kadanoff and Baym for non-relativistic many-body systems .
For processes where the overall time evolution is slow compared to relative motions the Kadanoff-Baym equations can be solved in a derivative expansion. One considers the Wigner transform for $`G(x_1,x_2)`$,
$$G(x,p)=d^4ye^{ipy}G(x+\frac{y}{2},x\frac{y}{2}),$$
(35)
and $`S(x,p)`$, $`\mathrm{\Delta }(x,p)`$, respectively. For the Wigner transform of a convolution one has in general,
$`{\displaystyle d^4ye^{ipy}d^4x_2A(x_1,x_2)B(x_2,x_3)}`$ (36)
$`=A(x,p)B(x,p){\displaystyle \frac{i}{2}}\left({\displaystyle \frac{}{x}}A(x,p){\displaystyle \frac{}{p}}B(x,p){\displaystyle \frac{}{x}}B(x,p){\displaystyle \frac{}{p}}A(x,p)\right)+\mathrm{},`$
where $`x=(x_1+x_3)/2`$ and $`y=(x_1x_3)/2`$.
Using the derivative expansion (36) the Kadanoff-Baym equations (33) and (34) become local in the space-time coordinate $`x`$. Keeping to zeroth order only the on-shell part of retarded and advanced Green functions and self-energies, which are given by expressions analogous to eq. (23), one obtains the equations
$`C({\displaystyle \frac{i}{2}}/+/pM)G^>(x,p)`$ $`=`$ $`C({\displaystyle \frac{i}{2}}/+/pM)G^<(x,p)`$ (37)
$`=`$ $`{\displaystyle \frac{1}{2}}\left(\mathrm{\Sigma }^>(x,p)G^<(x,p)\mathrm{\Sigma }^<(x,p)G^>(x,p)\right),`$
$`({\displaystyle \frac{i}{2}}/+/k)S^>(x,k)`$ $`=`$ $`({\displaystyle \frac{i}{2}}/+/k)S^<(x,k)`$ (38)
$`=`$ $`{\displaystyle \frac{1}{2}}\left(\mathrm{\Pi }^>(x,k)S^<(x,k)\mathrm{\Pi }^<(x,k)S^>(x,k)\right).`$
Solutions of these equations yield the first terms for the non-equilibrium Green functions $`G^>(x,p)`$$`S^<(x,k)`$ in an expansion involving off-shell effects and space-time variations, which include ‘memory effects’. As an example for the type of corrections we list the first derivative term on the right-hand side of eq. (37),
$`\mathrm{\Delta }_{}`$ $`=`$ $`{\displaystyle \frac{i}{4}}({\displaystyle \frac{}{x}}\mathrm{\Sigma }^>(x,p){\displaystyle \frac{}{p}}G^<(x,p){\displaystyle \frac{}{p}}\mathrm{\Sigma }^>(x,p){\displaystyle \frac{}{x}}G^<(x,p)`$ (39)
$`{\displaystyle \frac{}{x}}\mathrm{\Sigma }^<(x,p){\displaystyle \frac{}{p}}G^>(x,p)+{\displaystyle \frac{}{p}}\mathrm{\Sigma }^<(x,p){\displaystyle \frac{}{x}}G^>(x,p)).`$
Solutions of the Kadanoff-Baym equations can be studied once the self-energies $`\mathrm{\Sigma }`$ and $`\mathrm{\Pi }`$ are known. For weak coupling these can be determined in perturbation theory.
Self-energies for lepton fields
The one-loop contributions to the self-energy of the Majorana neutrino are shown in fig. (1). For vanishing chemical potential fig. (1a), for instance, yields the result
$$i\mathrm{\Sigma }_{eq}^{(a)>}(p)=2(\lambda ^{}\lambda )_{11}\frac{d^4p_1}{(2\pi )^4}\frac{d^4p_2}{(2\pi )^4}(2\pi )^4\delta (pp_1p_2)CS^>(p_1)\mathrm{\Delta }^>(p_2).$$
(40)
In the following we shall only consider the case $`MT`$, where the heavy neutrinos are non-relativistic. In this case the number density is small, $`f_N(E)1`$, and one finds for the difference of the self-energies ($`p_0>0`$),
$`i(\mathrm{\Sigma }_{eq}^>(p)\mathrm{\Sigma }_{eq}^<(p))`$ $`=`$ $`2(\lambda ^{}\lambda )_{11}{\displaystyle 𝑑\mathrm{\Phi }_{\overline{1}\overline{2}}(p)C/p_1}`$ (41)
$`\times \left((1f_l(p_1))(1+f_\varphi (p_2))+f_l(p_1)f_\varphi (p_2)\right)`$
$``$ $`\mathrm{\Gamma }C{\displaystyle \frac{/p}{M}}.`$ (42)
Here $`\mathrm{\Gamma }`$ is the vacuum decay rate of the Majorana neutrino.
The one- and two-loop contributions to the lepton self-energy are shown in fig. (2a)-(2d). In the following we only list the terms which are needed for the solution of the Kadanoff-Baym equations to leading order.
Particularly interesting are the terms fig. (2b) which drive the generation of an asymmetry. After some algebra one finds ($`k_0>0`$),
$$i\left(\delta \mathrm{\Pi }^{(b)<}(k)+\delta \mathrm{\Pi }^{(b)>}(k)\right)=\frac{3}{4\pi }\text{Im}(\lambda ^{}\eta \lambda ^{})_{11}M𝑑\mathrm{\Phi }_{1\overline{2}}(k)/p_2P_L\delta f_N(t,p_2).$$
(43)
Here we have only given the deviation from the equilibrium self-energy which is obtained by using for the Majorana neutrino propagator the deviation from the equilibrium propagator $`\delta G\delta f_N`$. Furthermore, we have again considered the case of small densities. The combination of Yukawa coulings is precisely the one occuring in the $`CP`$ asymmetry of the Majorana neutrino decay. Since the heavy neutrinos $`N_2`$ and $`N_3`$ have been integrated out, the contributions fig. (2b) involve both, self-energy and vertex corrections which, up to a numerical factor, are identical in this limit.
All graphs in fig. (2) contain washout processes. In the case of small densities one obtains from fig. (2a) ($`k_0>0`$),
$$i\left(\mathrm{\Pi }_{eq}^{(a)>}(k)\mathrm{\Pi }_{eq}^{(a)<}(k)\right)=2(\lambda ^{}\lambda )_{11}𝑑\mathrm{\Phi }_{1\overline{2}}(k)/p_2P_L(f_\varphi (p_1)+f_N(p_2)).$$
(44)
For $`TM`$ the dominant contribution to the washout processes is due to fig. (2c) where the Majorana neutrino propagator can be replaced by a local interaction. For small densities one finds ($`k_0>0`$),
$$i\left(\mathrm{\Pi }_{eq}^{(c)>}(k)\mathrm{\Pi }_{eq}^{(c)<}(k)\right)=6\frac{(\lambda ^{}\lambda )_{11}^2}{M^2}𝑑\mathrm{\Phi }_{1\overline{2}\overline{3}}(k)\left(/p_1f_l(p_1)+2/p_2f_\varphi (p_1)\right)P_L.$$
(45)
The complete expressions for the self-energies will be given elsewhere.
Kinetic equations
Given the lepton self-energies we can now look for solutions of the Kadanoff-Baym equations (37) and (38). A straightforward calculation shows that the right-hand side of these equations vanishes for equilibrium Green functions and self-energies. Since baryogenesis is a process close to thermal equilibrium we can search for solutions which are linear in the deviations,
$`\delta G(t,p)`$ $`=`$ $`G^>(t,p)G_{eq}^>(p)=G^<(t,p)G_{eq}^<(p),`$ (46)
$`\delta S(t,p)`$ $`=`$ $`S^>(t,p)S_{eq}^>(p)=S^<(t,p)S_{eq}^<(p).`$ (47)
One then obtains for the perturbations $`\delta G(t,p)`$ and $`\delta S(t,p)`$,
$`iC\gamma ^0{\displaystyle \frac{}{t}}\delta G(t,p)`$ $`=`$ $`iC\gamma ^0{\displaystyle \frac{}{t}}G_{eq}^>(p)+(\mathrm{\Sigma }_{eq}^>(p)\mathrm{\Sigma }_{eq}^<(p))\delta G(t,p),`$ (48)
$`i\gamma ^0{\displaystyle \frac{}{t}}\delta S(t,k)`$ $`=`$ $`(\mathrm{\Pi }_{eq}^>(k)\mathrm{\Pi }_{eq}^<(k))\delta S(t,k)`$ (49)
$`+\delta \mathrm{\Pi }^>(t,k)S_{eq}^<(k)\delta \mathrm{\Pi }^<(t,k)S_{eq}^>(k).`$
The Green functions depend on time explicitly, as well as implicitly through the time-dependence of the temperature. Once the ‘covariant’ time derivative is used, the later vanishes for equilibrium Green functions of massless fields. This is not the case for massive fields. Hence, the first term on the right-hand side of (48) drives the deviation from thermal equilibrium.
We can now insert the perturbative expressions for the self-energies into eqs. (48), (49) and check whether the ansatz (29), (30) for $`\delta G`$ and $`\delta S`$ yields a solution. After some algebra one finds that this is indeed the case provided the distribution functions $`\delta f_N`$ and $`\delta f_l`$ satisfy the following ordinary differential equations,
$`E{\displaystyle \frac{}{t}}\delta f_N(t,p)`$ $`=`$ $`E{\displaystyle \frac{}{t}}f_N(p)2(\lambda ^{}\lambda )_{11}{\displaystyle 𝑑\mathrm{\Phi }_{\overline{1}\overline{2}}(p)\delta f_N(t,p)pp_1},`$ (50)
$`g_lk{\displaystyle \frac{}{t}}\delta f_l(t,k)`$ $`=`$ $`{\displaystyle \frac{3}{8\pi }}\text{Im}(\lambda ^{}\eta \lambda ^{})_{11}M{\displaystyle 𝑑\mathrm{\Phi }_{\overline{1}2}(k)\delta f_N(t,p_1)kp_1}`$ (51)
$`2(\lambda ^{}\lambda )_{11}{\displaystyle 𝑑\mathrm{\Phi }_{1\overline{2}}(k)\left(\delta f_l(t,k)f_\varphi (p_1)+f_l(k)\delta f_\varphi (t,p_1)\right)kp_2}`$
$`6{\displaystyle \frac{(\lambda ^{}\lambda )_{11}^2}{M^2}}{\displaystyle }d\mathrm{\Phi }_{1\overline{2}\overline{3}}(k)(2(\delta f_l(t,k)f_\varphi (p_1)+f_l(k)\delta f_\varphi (t,p_1)`$
$`+\delta f_l(t,p_2)f_\varphi (p_3)+f_l(p_2)\delta f_\varphi (t,p_3))kp_2`$
$`+(\delta f_l(t,k)f_l(p_1)+f_l(k)\delta f_l(t,p_1)`$
$`+2\delta f_\varphi (t,p_2)f_\varphi (p_3))kp_1).`$
Comparing these equations with the Boltzmann equations (5) and (6) one finds that the two sets of equations are identical to leading order in the coupling where matrix elements and $`CP`$ asymmetry are given by
$`|(N(p)l(p_1)\varphi (p_2))|^2`$ $`=`$ $`4(\lambda ^{}\lambda )_{11}pp_1,`$ (52)
$`|(l(k)\varphi (p_1)\overline{l}(p_2)\overline{\varphi }(p_3))|^2`$ $`=`$ $`24{\displaystyle \frac{(\lambda ^{}\lambda )_{11}^2}{M^2}}kp_2,`$ (53)
$$ϵ=\frac{3}{16\pi }\frac{\text{Im}(\lambda ^{}\eta \lambda ^{})_{11}}{(\lambda ^{}\lambda )_{11}}M.$$
(54)
We conclude that for non-relativistic heavy neutrinos a solution of the Boltzmann equations generates a solution of the full Kadanoff-Baym equations to leading order in the expansion described above. For relativistic heavy neutrinos the matrix structure of the equations is more complicated and the time evolution of the different poles of the Majorana neutrino propagator are described by different equations.
Given a solution of the Kadanoff-Baym equations to leading order the various corrections can be systematically studied. Note, that the solutions of eqs. (50) and (51) are not of the form $`\delta f_i(t,p)=h_i(t)f_i(p)`$. Hence, the usual assumption of kinetic equilibrium does not appear to be justified. The size of ‘derivative terms’, which correspond to memory effects, and off-shell corrections can be determined by inserting the leading order solution into the various correction terms described above. Particularly interesting are relativistic corrections in the case that leptogenesis takes place at temperatures $`TM`$.
The analysis of the Kadanoff-Baym equations for leptogenesis can be used to obtain constraints on the parameters $`M`$, $`(\lambda ^{}\lambda )_{11}`$ and $`ϵ`$, which provides a quantitative relation between the cosmological baryon asymmetry and neutrino properties.
|
warning/0004/hep-ph0004048.html
|
ar5iv
|
text
|
# Lepton number violation interactions and their effects on neutrino oscillation experiments
## I Introduction
Experimentalists have reported three different kinds of “neutrino anomalies”, which seem to indicate that the standard model (SM) description of the neutrino is incorrect. Today, many physicists consider the recent SuperKamiokande high statistics result , which confirmed the long-standing atmospheric neutrino (AN) problem , as the strongest experimental evidence for New Physics (NP) beyond the SM. However, also increasingly convincing arguments, that the solar neutrino (SN) data can only be explained by extending the SM neutrino picture, have been established in recent years . Finally the LSND collaboration has found unexpected signals for neutrino flavor conversion in two appearance experiments . So far none of the other short baseline experiments , has been able to confirm these results, but major experimental efforts are underway to search for neutrino oscillations both at short and long baseline facilities.
The favorite explanation for the existing neutrino anomalies is to allow for massive neutrinos that mix and therefore undergo flavor oscillations while propagating. Neutrino oscillations provide convincing solutions to each of the above mentioned neutrino problems. However, the SN, the AN and the LSND observations imply three separated scales for the mass-squared differences $`\mathrm{\Delta }_{ij}=m_i^2m_j^2`$
$`\mathrm{\Delta }_{SN}`$ $`<`$$``$ $`10^5\text{ eV}^2,`$ (1)
$`\mathrm{\Delta }_{AN}`$ $``$ $`10^3\text{ eV}^2,`$ (2)
$`\mathrm{\Delta }_{LSND}`$ $`>`$$``$ $`10^1\text{ eV}^2,`$ (3)
which cannot be accommodated simultaneously in a three neutrino framework . Consequently, unless one ignores one of the three anomalies or allows for a forth non-sequential light neutrino , already the present neutrino data indicate that neutrino masses and mixing alone might not be the complete picture of the New Physics in the neutrino sector. It is important to note that many extensions of the SM that could provide massive neutrinos also predict non-standard neutrino interactions. In fact, in some cases new interactions induce neutrino masses in loop processes and one can relate the two aspects of New Physics quantitatively.
Solutions of the various neutrino anomalies in terms of new interactions with and without neutrino masses and mixing have been studied in Ref. . While for the SN problem this is indeed a viable possibility it has been shown that new flavor changing neutrino interactions that conserve total lepton number are constrained by the high precision data that confirm the SM predictions to be too small to affect the atmospheric and the LSND anomalies.
In this work we study another class of NP interactions, which so far has only received little attention , namely new neutrino interactions that violate the total lepton number $`L`$. Such interactions can arise naturally in models where there is mixing between bosons that transform differently under the SM gauge group, but identically under its unbroken subgroup. As an example consider the anomalous muon decays that produce two antineutrinos
$$\mu ^+e^+\overline{\nu }_e\overline{\nu }_{\mathrm{}}(\mathrm{}=e,\mu ,\tau ).$$
(4)
Such decays violate $`L`$ by two units. In principle the reaction in (4) could produce the $`\overline{\nu }_e`$’s that are observed at LSND in the decay at rest (DAR) channel and which are usually accounted for by $`\overline{\nu }_\mu \overline{\nu }_e`$ flavor oscillations.
Unlike for the $`L`$-conserving interactions, replacing the antineutrinos by their (positively) charged $`SU(2)_L`$ partners gives rise to interactions that violate $`U(1)_{EM}`$. Thus, it follows that the effective couplings of the above decays (4) must vanish in the $`SU(2)_L`$ symmetric limit and be proportional to $`SU(2)_L`$ breaking effects. This breaking is not proportional to a mass splitting within a given multiplet as for the $`L`$-conserving interactions , but it shows up as mixing between (heavy) bosons of different $`SU(2)_L`$ representations.
In Section II we present the general framework that expresses the effective strength of the lepton number violating interactions (as well as those that conserve total lepton number) in terms of the boson masses, their mixing angle and the relevant trilinear couplings. In Section III we discuss supersymmetry without $`R`$-parity (SUSY $`\overline{)}R_p`$) as a prominent example for such a scenario, where the mixing between left-handed and right-handed sfermions that couple to the SM fermions via $`R_p`$ violating interactions, induces lepton number violating interactions. In Section IV we establish relations between lepton number violating interactions and those that conserve total lepton number. We use these relations to derive constraints on the new interactions. Additional bounds on these interactions arise from the limit on universality violation in pion decays, the data on neutrinoless double beta decay and from loop-induced neutrino masses. In Section V we investigate whether the lepton number violating interactions could be relevant for any of the three anomalies as well as for the up-coming terrestrial neutrino oscillation experiments. Also implications for neutrinos from a supernova are discussed. We conclude in Section VI.
## II Formalism
Consider a generic extension of the standard model with two bosonic fields $`\varphi `$ and $`\chi `$ that transform differently under $`SU(2)_L`$. In general, after $`SU(2)_L`$ breaking, the $`SU(2)_L`$ components of these fields $`\varphi _q`$, $`\chi _q`$ which transform identically under the unbroken SM gauge group $`SU(3)_C\times U(1)_{EM}`$ can mix with each other giving rise to a hermitian mass-matrix
$$𝑴^2=\left(\begin{array}{cc}M_{11}^2& M_{12}^2\\ M_{21}^2& M_{22}^2\end{array}\right).$$
(5)
Diagonalizing $`𝑴^2`$ yields the eigenvalues
$$M_{1,2}^2=\frac{1}{2}\left(\mathrm{\Sigma }\sqrt{\delta ^2+4|M_{12}^2|^2}\right),$$
(6)
with $`\mathrm{\Sigma }=M_{11}^2+M_{22}^2`$ and $`\delta ^2=(M_{11}^2M_{22}^2)^2`$. The mass-eigenstates are linear combinations of $`\varphi _q`$ and $`\chi _q`$, i.e.
$$|i>=V_{i1}|\varphi _q>+V_{i2}|\chi _q>,(i=1,2).$$
(7)
Assuming that $`M_{12}=M_{21}`$ is real, the mixing matrix can be parameterized as
$$𝑽=\left(\begin{array}{cc}\mathrm{cos}\theta & \mathrm{sin}\theta \\ \mathrm{sin}\theta & \mathrm{cos}\theta \end{array}\right),$$
(8)
with
$$\mathrm{sin}2\theta =\frac{2M_{12}^2}{\sqrt{\delta ^2+4M_{12}^4}}.$$
(9)
Let us add now renormalizable interactions that couple the bosonic fields $`\varphi `$ and $`\chi `$ to bilinears $`A`$, $`B`$ which are built out of two SM fermions
$$_{A,B}=\lambda _A(\varphi A)+\lambda _B(\chi B)+\mathrm{h}.\mathrm{c}.,$$
(10)
where $`\lambda _A`$ and $`\lambda _B`$ denote the elementary trilinear couplings. These couplings are induced by New Physics that may be present at or above the weak scale. Any such theory will include the SM gauge symmetry, implying that $`_{A,B}`$ is invariant under $`𝒢_{\mathrm{SM}}=SU(3)_C\times SU(2)_L\times U(1)_Y`$. If the bosons are vector fields the couplings in (10) may be gauge interactions. We will focus on scalar bosons that couple to fermions via a priori arbitrary Yukawa couplings $`\lambda _{A,B}`$. The fermion bilinears may be composed of quarks, leptons or both, as well as the respective antiparticles, which all belong to either $`SU(2)_L`$ singlets or doublets. Then gauge invariance implies that the bosonic fields may be singlets (s), doublets (d) or triplets (t) of $`SU(2)_L`$. Since we require $`\varphi `$ and $`\chi `$ to have different transformation properties under $`SU(2)_L`$ also $`A`$ and $`B`$ will transform differently. On the other hand, since $`\varphi `$ and $`\chi `$ transform identically under $`SU(3)_C`$ this also applies to $`A`$ and $`B`$.
Given the elementary couplings in (10) one can construct four-fermion interactions that are mediated by the bosonic fields. Each bilinear $`A`$ and $`B`$ can be either coupled to itself or there can be a coupling between $`A`$ and $`B`$. Since $`\varphi _q`$ and $`\chi _q`$ are required to have the same electric charge $`q`$ there is only a coupling between the $`SU(2)_L`$ components $`A_q`$ and $`B_q`$ that have the same charge such that the resulting four-fermion operator $`A_q^{}B_q`$ conserves $`U(1)_{EM}`$. Similarly, since $`A`$ and $`B`$ transform identically under $`SU(3)_C`$ it follows that $`A_q^{}B_q`$ is a color singlet.
We stress that the coupling between fermion bilinears that have different $`SU(2)_L`$ transformations requires the mixing between $`\varphi _q`$ and $`\chi _q`$. As a unique consequence the coupling between a bilinear that transforms as an $`SU(2)_L`$ doublet to a bilinear that is a singlet or a triplet of $`SU(2)_L`$ can induce effective four-fermion operators that do not conserve the total lepton number $`L`$. In addition such operators may also violate the individual lepton number $`L_{\mathrm{}}`$. For example, if the $`SU(2)_L`$ doublet $`A=\overline{L}_eE_\mu `$ ($`E_{\mathrm{}}`$ and $`L_{\mathrm{}}`$ denote a lepton singlet and doublet field, respectively, of flavor $`\mathrm{}`$) couples to an $`SU(2)_L`$ doublet $`\varphi `$ and if the $`SU(2)_L`$ singlet $`B_1=(L_\mu L_e)_s`$ couples to an $`SU(2)_L`$ singlet $`\chi ^+`$, then the mixing between the $`q=1`$ doublet-component $`\varphi ^+`$ and $`\chi ^+`$ gives rise to the operator $`A_1^{}B_1=(\overline{\mu _R}\nu _e)(\nu _\mu e_L\mu _L\nu _e)`$, which induces $`\mu _L^+e_R^+\overline{\nu }_\mu \overline{\nu }_e`$ and $`\mu _L^+\nu _e\mu _R^+\overline{\nu }_e`$. Both processes violate $`L_e,L_\mu `$ and $`L`$ by two units.
Note that if one scalar field couples to two different bilinears (which, consequently must have the same $`SU(2)_L`$ transformation) then these two bilinears can be coupled to each other. However, the four fermion operators that arise from this mechanism (see e.g. ) may only violate $`L_{\mathrm{}}`$, but not $`L`$.
The effective four-fermion operators $`A_q^{}A_q`$, $`B_q^{}B_q`$ and $`A_q^{}B_q`$ at energies well below the masses of the scalar fields \[i.e. the eigenvalues of $`𝑴^2`$ given in (6)\] are obtained by integrating out the bosonic degrees of freedom. Assuming weak trilinear couplings, $`\lambda _{A,B}<1`$, the tree-level diagrams result into the effective couplings
$$G_N^{A^{}A}=\frac{|\lambda _A|^2}{4\sqrt{2}M_A^2},G_N^{B^{}B}=\frac{|\lambda _B|^2}{4\sqrt{2}M_B^2},G_N^{A^{}B}=\frac{\lambda _A^{}\lambda _B}{4\sqrt{2}M_{AB}^2},$$
(11)
where the respective low-energy propagators are given by
$`M_A^2`$ $``$ $`{\displaystyle \underset{i}{}}{\displaystyle \frac{<\varphi |i><i|\varphi >}{M_i^2}}={\displaystyle \frac{\mathrm{cos}^2\theta }{M_1^2}}+{\displaystyle \frac{\mathrm{sin}^2\theta }{M_2^2}},`$ (12)
$`M_B^2`$ $``$ $`{\displaystyle \underset{i}{}}{\displaystyle \frac{<\chi |i><i|\chi >}{M_i^2}}={\displaystyle \frac{\mathrm{cos}^2\theta }{M_2^2}}+{\displaystyle \frac{\mathrm{sin}^2\theta }{M_1^2}},`$ (13)
$`M_{AB}^2`$ $``$ $`{\displaystyle \underset{i}{}}{\displaystyle \frac{<\varphi |i><i|\chi >}{M_i^2}}={\displaystyle \frac{\mathrm{sin}2\theta }{2}}\left({\displaystyle \frac{1}{M_1^2}}{\displaystyle \frac{1}{M_2^2}}\right).`$ (14)
From (14) it is obvious that $`A_q^{}`$ and $`B_q`$ can only couple to each other when there is a non-vanishing mixing ($`\mathrm{sin}2\theta 0`$) and when the physical masses are not degenerate ($`M_1M_2`$). We note that for maximal mixing ($`\mathrm{sin}2\theta =1`$) the propagator $`M_A^2`$ equals to $`M_B^2`$, and $`M_{AB}^2`$ is maximal. Moreover, we remark that using (9) it follows that the propagator in (14) is simply
$$M_{AB}^2=\frac{M_{12}^2}{M_1^2M_2^2}.$$
(15)
## III Lepton number violation in SUSY without $`R`$-parity
In this section we present an explicit example for the general mechanism developed in Section II by discussing scalar mixing in supersymmetric extensions of the standard model without $`R`$-parity . In particular, we show how the model-specific parameters of this theory translate into those we introduced in Section II.
The $`R`$-parity violating couplings $`\lambda _{ıȷ\kappa }L_ıL_ȷE_\kappa ^c`$ and $`\lambda _{ıȷ\kappa }^{}L_ıQ_ȷD_\kappa ^c`$, where $`L_\kappa ,Q_\kappa ,E_\kappa `$ and $`D_\kappa `$ denote the chiral superfields containing, respectively, the left-handed lepton and quark doublets and the right-handed charged-lepton and $`d`$-quark singlets of generation $`\kappa =1,2,3`$, introduce a variety of couplings between fermion bilinears and sfermions:
$`_\lambda `$ $`=`$ $`\lambda _{ıȷ\kappa }\left[\stackrel{~}{\nu }_L^ı\overline{\mathrm{}_R^\kappa }\mathrm{}_L^ȷ+\stackrel{~}{\mathrm{}}_L^ȷ\overline{\mathrm{}_R^\kappa }\nu _L^ı+\stackrel{~}{\mathrm{}}_R^\kappa \overline{\nu _L^ı}^c\mathrm{}_L^ȷ(ıȷ)\right]+\mathrm{h}.\mathrm{c}.,`$ (16)
$`_\lambda ^{}`$ $`=`$ $`\lambda _{ıȷ\kappa }^{}\left[\stackrel{~}{\nu }_L^ı\overline{d_R^\kappa }d_L^ȷ+\stackrel{~}{d}_L^ȷ\overline{d_R^\kappa }\nu _L^ı+\stackrel{~}{d}_R^\kappa \overline{\nu _L^ı}^cd_L^ȷ\stackrel{~}{e}_L^ı\overline{d_R^\kappa }u_L^ȷ\stackrel{~}{u}_L^ȷ\overline{d_R^\kappa }e_L^ı\stackrel{~}{d}_R^\kappa \overline{e_L^ı}^cu_L^ȷ\right]+\mathrm{h}.\mathrm{c}..`$ (17)
Due to charge conservation only sfermions of the same type can mix. In principle mixing is allowed between the left-handed and the right-handed components of the (charged) sfermions, as well as between sfermions of different generation, but for simplicity we will assume that the latter is negligible.
The leptonic couplings in (16) can induce $`L`$-violating interactions like in (4) . For example, identifying the scalar fields $`\varphi ^+=\stackrel{~}{\tau }_R^+`$ and $`\chi ^+=\stackrel{~}{\tau }_L^+`$ and the couplings $`\lambda _A=\lambda _{132}^{}`$ and $`\lambda _B=\lambda _{123}`$ reproduces exactly the example in Section II that gave rise to $`\mu _L^+e_R^+\overline{\nu }_\mu \overline{\nu }_e`$. Note that SUSY without $`R_p`$ not only provides the required scalar fields and their couplings, but also an explicit expression for the mass-matrix in (5), i.e.
$$𝑴_{\stackrel{~}{f}}^{}{}_{}{}^{2}=\left(\begin{array}{cc}M_{\stackrel{~}{L}}^2+m_f^2+M_Z^2(T_3^f/2q_f\mathrm{sin}^2\theta _W)& m_f(A_f\mu \mathrm{cot}^{T_3^f}\beta )\\ m_f(A_f\mu \mathrm{cot}^{T_3^f}\beta )& M_{\stackrel{~}{R}}^2+m_f^2+M_Z^2q_f\mathrm{sin}^2\theta _W\end{array}\right),$$
(18)
where $`m_f`$ and $`q_f`$ denote the mass and the charge of the fermion $`f`$, $`T_3^f=1(1)`$ for $`f=u_\kappa (d_\kappa ,e_\kappa )`$, $`M_{\stackrel{~}{L}}^2`$ $`(M_{\stackrel{~}{R}}^2)`$ is the soft supersymmetric breaking mass-squared term for the left- (right-)handed sfermion, and $`A_{\mathrm{}}`$, $`\mu `$ and $`\mathrm{tan}\beta `$ are the familiar SUSY parameters .
We note that in the absence of right-handed (s)neutrinos $`\stackrel{~}{f}_L\stackrel{~}{f}_R`$ mixing can occur only for charged sfermions. This implies that the mass-matrix (18) has to be positive definite in order to avoid the spontaneous breaking of $`U(1)_{EM}`$.
## IV Experimental Constraints
In this section we discuss constraints on the effective couplings for lepton number violating neutrino interactions. As we mentioned already, the corresponding four-fermion operators cannot be related to the ones where the neutrinos are rotated into their charged lepton partners, since such an $`SU(2)_L`$ rotation violates $`U(1)_{EM}`$. Hence, while in many cases such a rotation can provide stringent bounds on the product of trilinear couplings for interactions that only violate $`L_{\mathrm{}}`$ (see ), it does not help for $`L`$-violating neutrino interactions. Instead one can use the constraints on each of the trilinear couplings which arise from the interactions induced by the self-couplings of a specific fermion bilinear relevant for the lepton number violating neutrino interaction. Alternatively, in some cases, there are direct constraints on the $`L`$-violating interactions. Operators that induce lepton number violating pion decays can be constrained using the limit on universality violation. Upper bounds on certain operators containing the electron neutrino follow from the data on neutrinoless double beta decay. Finally, in case the $`L`$-violating operator involves two neutrinos, one can connect the two external charged fermions in order to generate neutrino masses and use their upper bounds.
### A Constraints from the trilinear couplings
Any non-vanishing trilinear coupling $`\lambda _A`$ between a fermion bilinear $`A`$ and a boson $`\varphi `$ can be used to create the effective interaction
$$\frac{|\lambda _A|^2A^{}A}{4\sqrt{2}M_A^2}.$$
(19)
If the intermediate boson does not mix, the low-energy propagator is simply $`M_A^2=M_\varphi ^2`$, but if there is mixing the correct expression is the one in (12).
This is important, since the constraints on various trilinear couplings in the literature are derived assuming that the respective intermediate boson is a mass eigenstates which has a definite mass $`M`$. Therefore, if we denote the upper bound on any trilinear coupling derived under such an assumption by $`\widehat{\lambda }_A`$, then this implies for the parameter $`\lambda _A`$, which describes the coupling between $`A`$ and a boson $`\varphi `$ that is not a mass eigenstate (but which mixes with a different boson $`\chi `$ as discussed in Section II), that
$$|\lambda _A|<\widehat{\lambda }_A\times \frac{M_A}{M}.$$
(20)
This rescaling corrects for the fact that if the effective propagator $`M_A^2`$ is smaller (larger) than $`M^2`$, then the constraint on $`\lambda _A`$ will be weaker (stronger).
Consequently the upper bound on any $`L`$-violating operator $`(G_N^{A^{}B}/\sqrt{2})A^{}B`$ which is induced by $`\varphi \chi `$ mixing is constrained by
$$G_N^{A^{}B}=\frac{\lambda _A^{}\lambda _B}{4\sqrt{2}M_{AB}^2}<\frac{\widehat{\lambda }_A^{}\widehat{\lambda }_B}{4\sqrt{2}M^2}\times \frac{M_AM_B}{M_{AB}^2}.$$
(21)
The upper bound on the right-hand side of (21) factorizes into
$$\widehat{G}_N^{A^{}B}\frac{\widehat{\lambda }_A^{}\widehat{\lambda }_B}{4\sqrt{2}M^2},$$
(22)
which only depends on the upper bounds $`\widehat{\lambda }_A`$ and $`\widehat{\lambda }_B`$ derived from experimental observations (under the assumption that the intermediate particle has mass $`M`$) and the ratio
$$\varrho \frac{M_AM_B}{M_{AB}^2}$$
(23)
which is a function of the mixing angle $`\theta `$ and the mass eigenvalues $`M_1`$ and $`M_2`$ only. Note that $`\varrho (\mathrm{sin}\theta ,M_1,M_2)1`$ and that $`\varrho `$ is maximal at $`\mathrm{sin}\theta =\mathrm{cos}\theta =1/\sqrt{2}`$, where it takes the value $`\widehat{\varrho }=(M_2^2M_1^2)/(M_2^2+M_1^2)`$, which is small when the masses are almost degenerate, but it quickly approaches unity when the degeneracy is lifted. We show $`\varrho (\mathrm{sin}^2\theta )`$ for various values of $`M_2/M_1`$ in Fig. 1.
Since the interactions induced by the self-couplings of any fermion bilinear $`A`$ do not violate $`L_{\mathrm{}}`$ and $`L`$, the corresponding NP operator only induces additional contributions to reactions that are already present in the standard model. Therefore any non-zero NP effective coupling $`G_N^{A^{}A}`$ modifies the SM predictions for the relevant processes and precision measurements can be used to put upper bounds on $`G_N^{A^{}A}`$.
It is conventional to assume that only one trilinear coupling $`\lambda _A`$ is non-zero for each bound and that the intermediate boson has a mass of $`M=100`$ GeV. Then the constraint is expressed in terms the dimensionless real number $`\widehat{\lambda }_A`$ as
$$|\lambda _A|<\widehat{\lambda }_A\times \left(\frac{M}{100\mathrm{GeV}}\right),$$
(24)
which translates into
$$G_N^{A^{}A}<\widehat{G}_N^{A^{}A}\frac{\widehat{\lambda }_A^2}{4\sqrt{2}(100\mathrm{GeV})^2}=1.52\widehat{\lambda }_A^2G_F.$$
(25)
for the effective coupling. If the process that was used to derive the bound and the one which one wants to constrain are mediated by bosons which are different members of the same $`SU(2)_L`$ multiplet, then one has to correct for differences in the propagators
$$G_N^q^{}=\frac{M_q^2}{M_q^{}^2}G_N^q<1.52\widehat{\lambda }^2\frac{M_q^2}{M_q^{}^2}G_F,$$
(26)
where $`q,q^{}`$ refer to the charge of the intermediate boson. In Ref. it has been shown that electroweak precision data imply that $`M_q/M_q^{}`$ is of order unity. Even for masses close to the weak scale this ratio is at most $`2.6`$ unless one allows for some fine-tuned cancellations.
In Tab. 1 we list all bilinears that couple to scalar weak singlets or doublets that appear in SUSY without $`R`$-parity. We also show in Tab. 1 the upper bounds (at $`2\sigma `$) for both the trilinear couplings ($`\widehat{\lambda }`$) and the effective couplings ($`\widehat{G}_N^{A^{}A}`$). We assume here that $`M_q/M_q^{}=1`$, bearing in mind that for scalar doublets the maximal correction from $`SU(2)_L`$ breaking effects could be a factor a few. For the bounds we use the results obtained within the framework of SUSY $`\overline{)}R_p`$. All limits are at $`2\sigma `$, except for $`\lambda _{1\kappa 1}`$ which is at $`3\sigma `$. The most stringent constraints relevant to our discussion arise from charged current universality ($`V_{ud}`$), lepton universality \[$`R_\tau =\mathrm{\Gamma }(\tau e\nu \overline{\nu })/\mathrm{\Gamma }(\tau \mu \nu \overline{\nu })`$, $`R_\pi =\mathrm{\Gamma }(\pi e\nu )/\mathrm{\Gamma }(\pi \mu \nu )`$ and $`R_{\tau \pi }=\mathrm{\Gamma }(\tau \pi \nu _\tau )/\mathrm{\Gamma }(\pi \mu \nu _\mu )`$\], forward-backward asymmetries in $`e^+e^{}`$ collisions at the $`Z`$ peak ($`A_{FB}`$), atomic parity violation (APV), $`\nu _\mu `$ deep inelastic scattering ($`\nu _\mu `$ DIS) and constraints on the compositeness scale \[$`\mathrm{\Lambda }(qqqq)`$\]. We note that the listed bounds apply to any theory that contains the respective trilinear coupling, since we consider only the constraints that are derived directly from a specific coupling, that is we allow only one term in the $`R_p`$-violating couplings (16) and (17), to be non-zero at a time. If one relaxes this assumption and takes all the $`R_p`$-violating couplings together as they appear in (16) and (17), i.e. one evokes supersymmetry, then in some cases additional constraints (given is square brackets) can be found from processes which have different intermediate scalars, but rely on the same trilinear coupling. We note that for the coupling $`\lambda _{31}`$ of $`L_3\overline{D}_1`$ to a scalar doublet to the best of our knowledge there is no model-independent bound. Demanding that the theory remains perturbative at large energies implies that $`\lambda _{31}<1`$. In SUSY $`\overline{)}R_p`$ $`\lambda _{31}=\lambda _{3\kappa 1}^{}<0.52`$, which is due to the upper bound on $`\lambda _{321}^{}`$ from $`D_s`$ decays.
Tab. 1: Experimental constraints on fermion bilinear self-couplings
| $`\lambda _A`$ | $`A`$ | $`\varphi `$ | $`\widehat{\lambda }_A`$ \[SUSY $`\overline{)}R_p`$\] | $`\widehat{G}_N^{A^{}A}/G_F`$ \[SUSY $`\overline{)}R_p`$\] | from |
| --- | --- | --- | --- | --- | --- |
| $`\lambda _{12\kappa }`$ | $`L_1L_2`$ | $`\stackrel{~}{e}_R^\kappa `$ | 0.05 | 0.0038 | $`V_{ud}`$ |
| $`\lambda _{13\kappa }`$ | $`L_1L_3`$ | $`\stackrel{~}{e}_R^\kappa `$ | 0.06 | 0.0055 | $`R_\tau `$ |
| $`\lambda _{23\kappa }`$ | $`L_2L_3`$ | $`\stackrel{~}{e}_R^\kappa `$ | 0.06 | 0.0055 | $`R_\tau `$ |
| $`\lambda _{1\kappa 1}`$ | $`L_1\overline{E}_1`$ | $`\stackrel{~}{\nu }_L^\kappa `$ | 0.37 \[0.06\] | 0.21 \[0.0055\] | $`A_{FB}`$ |
| $`\lambda _{2\kappa 1}`$ | $`L_2\overline{E}_1`$ | $`\stackrel{~}{\nu }_L^\kappa `$ | 0.25 \[0.07\] | 0.095 \[0.0074\] | $`A_{FB}`$ |
| $`\lambda _{3\kappa 1}`$ | $`L_3\overline{E}_1`$ | $`\stackrel{~}{\nu }_L^\kappa `$ | 0.11 \[0.07\] | 0.018 \[0.0074\] | $`A_{FB}`$ |
| $`\lambda _{1\kappa 2}`$ | $`L_1\overline{E}_2`$ | $`\stackrel{~}{\nu }_L^\kappa `$ | 0.25 \[0.06\] | 0.095 \[0.0055\] | $`A_{FB}`$ |
| $`\lambda _{2\kappa 2}`$ | $`L_2\overline{E}_2`$ | $`\stackrel{~}{\nu }_L^\kappa `$ | 0.25 \[0.06\] | 0.095 \[0.0055\] | $`A_{FB}`$ |
| $`\lambda _{3\kappa 2}`$ | $`L_3\overline{E}_2`$ | $`\stackrel{~}{\nu }_L^\kappa `$ | 0.25 \[0.06\] | 0.095 \[0.0055\] | $`A_{FB}`$ |
| $`\lambda _{11\kappa }^{}`$ | $`L_1Q_1`$ | $`\stackrel{~}{d}_R^\kappa `$ | 0.02 | 0.0006 | $`V_{ud}`$ |
| $`\lambda _{21\kappa }^{}`$ | $`L_2Q_1`$ | $`\stackrel{~}{d}_R^\kappa `$ | 0.06 | 0.0055 | $`R_\pi `$ |
| $`\lambda _{31\kappa }^{}`$ | $`L_3Q_1`$ | $`\stackrel{~}{d}_R^\kappa `$ | 0.11 | 0.018 | $`R_{\tau \pi }`$ |
| $`\lambda _{1\kappa 1}^{}`$ | $`L_1\overline{D}_1`$ | $`\stackrel{~}{q}_L^\kappa `$ | 0.02 | 0.0006 | APV |
| $`\lambda _{2\kappa 1}^{}`$ | $`L_2\overline{D}_1`$ | $`\stackrel{~}{q}_L^\kappa `$ | 0.22 | 0.07 | $`\nu _\mu `$ DIS |
| $`\lambda _{3\kappa 1}^{}`$ | $`L_3\overline{D}_1`$ | $`\stackrel{~}{q}_L^\kappa `$ | $`\widehat{\lambda }_{31}`$ \[0.52\] | 1.52 $`\widehat{\lambda }_{31}^2`$ \[0.41\] | $`\nu _\tau `$ DIS” |
| $`\lambda _{\kappa 11}^{}`$ | $`Q_1\overline{D}_1`$ | $`\stackrel{~}{\mathrm{}}_L^\kappa `$ | 0.3 \[0.11\] | 0.14 \[0.018\] | $`\mathrm{\Lambda }(qqqq)`$ |
In general there could also be trilinear couplings involving the up-type quark singlet as well as couplings to scalar triplets, which we do not discuss explicitly. We remark that replacing the scalar weak singlets by triplets of the same charge, while keeping the flavor structure, only changes the sign in the doublet-doublet contraction and yields the same effective interactions. However, a neutral triplet may also couple to $`\nu \nu `$ inducing additional effective couplings. Moreover a triplet can have flavor diagonal coupling to $`LL`$, while for scalars $`\lambda `$ has to be antisymmetric in flavor space. The $`\mathrm{\Delta }_L`$ in left-right symmetric models is an example for a scalar triplet with flavor diagonal couplings. We do not consider here the possibility of intermediate vector bosons, which will couple to different bilinears than the scalar fields, and moreover produce a different spin structure for the four-fermion operator.
From the definition of $`\widehat{G}_N^{A^{}B}`$ and $`\widehat{G}_N^{A^{}A}`$ it follows that
$$\widehat{G}_N^{A^{}B}=\sqrt{\widehat{G}_N^{A^{}A}\widehat{G}_N^{B^{}B}}.$$
(27)
In Tab. 2 we show $`\widehat{G}_N^{A^{}B}`$ (based on this relation and the constraints tabulated in Tab. 1) for the various $`L`$-violating effective couplings that are relevant for neutrino oscillation experiments.
Tab. 2: Experimental constraints on $`L`$-violating couplings
| $`A^{}`$ | $`B`$ | $`\widehat{G}_N^{A^{}B}/G_F`$ \[SUSY $`\overline{)}R_p`$\] | reaction | relevant for |
| --- | --- | --- | --- | --- |
| $`L_1\overline{E}_2`$ | $`L_1L_2`$ | 0.019 \[0.0046\] | $`\mu _L^+e_R^+\overline{\nu }_e\overline{\nu }_\mu `$ | LSND: DAR |
| $`L_1\overline{E}_2`$ | $`L_1L_3`$ | 0.023 \[0.0055\] | $`\mu _L^+e_R^+\overline{\nu }_e\overline{\nu }_\tau `$ | LSND: DAR |
| $`L_1\overline{D}_1`$ | $`L_1Q_1`$ | 0.0006 | $`\nu _eu_Ld_Re_R^+`$ | LSND: “fake” $`\overline{\nu }_e`$ |
| $`L_2\overline{D}_1`$ | $`L_1Q_1`$ | 0.0067 | $`\nu _\mu u_Ld_Re_R^+`$ | LSND: “fake” $`\overline{\nu }_e`$ |
| $`Q_1\overline{D}_1`$ | $`L_1L_2`$ | 0.023 \[0.0083\] | $`\nu _\mu u_Ld_Re_R^+`$ | LSND: “fake” $`\overline{\nu }_e`$ |
| $`L_2\overline{D}_1`$ | $`L_3Q_1`$ | 0.037 | $`\nu _\mu u_Ld_R\tau _R^+`$ | NOMAD/CHORUS: “fake” $`\overline{\nu }_\tau `$ |
| $`Q_1\overline{D}_1`$ | $`L_2L_3`$ | 0.028 \[0.010\] | $`\nu _\mu u_Ld_R\tau _R^+`$ | NOMAD/CHORUS: “fake” $`\overline{\nu }_\tau `$ |
| $`L_1\overline{E}_1`$ | $`L_1L_2`$ | 0.028 \[0.0046\] | $`\nu _ee_L^{}\overline{\nu }_\mu e_R^{}`$ | SN: $`\nu _e\overline{\nu }_\mu `$ |
| $`L_1\overline{E}_1`$ | $`L_1L_3`$ | 0.034 \[0.0055\] | $`\nu _ee_L^{}\overline{\nu }_\tau e_R^{}`$ | SN: $`\nu _e\overline{\nu }_\tau `$ |
| $`L_1\overline{D}_1`$ | $`L_1Q_1`$ | 0.0006 | $`\nu _ed_L\overline{\nu }_ed_R`$ | SN: $`\nu _e\overline{\nu }_e`$ |
| $`L_1\overline{D}_1`$ | $`L_2Q_1`$ | 0.0018 | $`\nu _ed_L\overline{\nu }_\mu d_R`$ | SN: $`\nu _e\overline{\nu }_\mu `$ |
| $`L_1\overline{D}_1`$ | $`L_3Q_1`$ | 0.0033 | $`\nu _ed_L\overline{\nu }_\tau d_R`$ | SN: $`\nu _e\overline{\nu }_\tau `$ |
| $`L_2\overline{D}_1`$ | $`L_1Q_1`$ | 0.0067 | $`\nu _ed_L\overline{\nu }_\mu d_R`$ | SN: $`\nu _e\overline{\nu }_\mu `$ |
| $`L_3\overline{D}_1`$ | $`L_1Q_1`$ | 0.030 $`\widehat{\lambda }_{31}`$ \[0.016\] | $`\nu _ed_L\overline{\nu }_\tau d_R`$ | SN: $`\nu _e\overline{\nu }_\tau `$ |
| $`L_2\overline{E}_1`$ | $`L_1L_3`$ | 0.023 \[0.0064\] | $`\nu _\mu e_L^{}\overline{\nu }_\tau e_R^{}`$ | AN: $`\nu _\mu \overline{\nu }_\tau `$ |
| $`L_3\overline{E}_1`$ | $`L_1L_2`$ | 0.0083 \[0.0053\] | $`\nu _\mu e_L^{}\overline{\nu }_\tau e_R^{}`$ | AN: $`\nu _\mu \overline{\nu }_\tau `$ |
| $`L_2\overline{D}_1`$ | $`L_3Q_1`$ | 0.037 | $`\nu _\mu d_L\overline{\nu }_\tau d_R`$ | AN: $`\nu _\mu \overline{\nu }_\tau `$ |
| $`L_3\overline{D}_1`$ | $`L_2Q_1`$ | 0.091 $`\widehat{\lambda }_{31}`$ \[0.047\] | $`\nu _\mu d_L\overline{\nu }_\tau d_R`$ | AN: $`\nu _\mu \overline{\nu }_\tau `$ |
From Tab. 2 one can see that model-independently almost all effective couplings for the lepton number violating operators are constrained to be at most a few percent of $`G_F`$. The weakest constraints are those involving $`\lambda _{31}`$, but even allowing $`\lambda _{31}`$ to be of order unity implies that $`G_N^{A^{}B}<0.1G_F`$. Imposing SUSY we find that all of the effective couplings are constrained to be less than one percent of $`G_F`$, except those involving $`\lambda _{3\kappa 1}^{}`$ which could be at most a few percent of $`G_F`$.
### B Direct constraints
We turn now to a discussion of additional constraints on the effective four-fermion operators that violate total lepton number. Unlike the bounds derived in the previous section these bounds do not depend upon the constraints on the trilinear couplings, but apply to the $`L`$-violating operator itself.
#### 1 Pion decays
Consider the ratio between the decay rates of $`\pi ^+e^+\nu `$ and $`\pi ^+\mu ^+\nu `$ ,
$$R_\pi =\frac{\mathrm{\Gamma }(\pi ^+e^+\nu )}{\mathrm{\Gamma }(\pi ^+\mu ^+\nu )}.$$
(28)
The measured value of this ratio ,
$$R_\pi (expt)=(1.235\pm 0.004)\times 10^4,$$
(29)
is in good agreement with the value predicted by the standard model, including radiative corrections ,
$$R_\pi (SM)=(1.230\pm 0.008)\times 10^4.$$
(30)
Consequently any non-standard contribution to either $`\pi ^+e^+\nu `$ or $`\pi ^+\mu ^+\nu `$ is constrained to be small . Since the final neutrino is not detected this applies to $`\pi ^+`$ decays with both final neutrinos and antineutrinos. The latter case is particularly interesting, because it allows to constrain lepton number violating operators that induce pion decays. Note that for these operators (unlike for lepton number conserving operators) the decay amplitude is not suppressed by the charged lepton mass $`m_{\mathrm{}}`$ ($`\mathrm{}=e,\mu `$), but there is an enhancement by
$$f_{\mathrm{}}\frac{m_\pi ^2}{m_{\mathrm{}}(m_u+m_d)}$$
(31)
with respect to the standard model currents. Therefore, when adding NP interactions to those of the SM, to leading order in the effective couplings $`G_N^{\mathrm{}}G_F`$ of the lepton number violating operators that induce $`\pi ^+\mathrm{}\overline{\nu }`$, the ratio in (28) is
$$\frac{R_\pi (SM+NP)}{R_\pi (SM)}=1+\frac{(f_eG_N^e)^2(f_\mu G_N^\mu )^2}{(V_{ud}G_F)^2},$$
(32)
where $`V_{ud}`$ is the CKM matrix element relevant for the SM pion decay. Then, assuming that there are no fine-tuned cancellations in (32), it follows from (29) and (30) that
$`G_N^e`$ $`<`$$``$ $`3\times 10^5G_F,`$ (33)
$`G_N^\mu `$ $`<`$$``$ $`4\times 10^3G_F.`$ (34)
We conclude that the effective couplings of all the operators in second section of Tab. 2 that induce $`\nu _{\mathrm{}}ude^+`$ must be severely suppressed due to the bound on $`G_N^e`$, since they also give rise to the lepton number violating pion decays. In particular the model-independent bound for $`(Q_1\overline{D}_1)(L_1L_2)`$ is improved significantly. Moreover, also the operator $`(Q_1\overline{D}_1)(L_2L_3)`$ can be constrained by the limit on $`G_N^\mu `$, because the structure of the singlet bilinear $`(L_2L_3)_s=\nu _\mu \tau \mu \nu _\tau ,`$ implies that the operator obtained by exchanging the flavors of the neutrino and the charged lepton must have the same effective coupling. A similar argument applies to the operators $`(L_\alpha \overline{D})(L_{\mathrm{}}Q)`$ ($`\alpha =e,\mu ,\tau `$ and $`\mathrm{}=e,\mu `$) appearing in the third and forth section of Tab. 2 . Since $`(L_{\mathrm{}}Q)_s=\mathrm{}u_L\nu _{\mathrm{}}d_L`$ they induce both $`\nu _{\mathrm{}}d_L\nu _\alpha d_R`$ and $`\pi ^+\mathrm{}^+\overline{\nu }_\alpha `$. However the upper bounds on $`G_N^{\mathrm{}}`$ in (33) and (34) are not useful to constrain any of the purely leptonic operators or those involving $`(QL_3)_s`$.
#### 2 Neutrinoless double beta decay
The combination of the SM operator for beta decay with a new physics operator that mediates the $`L`$-violating process
$$\overline{\nu }_ene^{}p$$
(35)
gives rise to neutrinoless double beta decay ($`0\nu \beta \beta `$ due to the exchange of a virtual neutrino. The crucial point is that if the leptonic current of the (if necessary Fierz transformed) NP operator contains a right-handed neutrino then the contribution from the neutrino propagator is
$$P_L\frac{q^\mu \gamma _\mu +m_\nu }{q^2m_\nu ^2}P_R=\frac{q^\mu \gamma _\mu }{q^2m_\nu ^2}.$$
(36)
Therefore the $`0\nu \beta \beta `$ amplitude is proportional to $`G_NG_Fq`$, where the neutrino momentum $`q`$ is typically given by the nuclear Fermi momentum $`p_F100`$ MeV . The present half-life limit of the Heidelberg-Moscow experiment, $`T_{1/2}^{0\nu \beta \beta }>1.610^{25}y`$ then implies severe limits on any lepton-number violating operator $`(\overline{u}d\overline{e}\nu _e^c)`$. For scalar couplings one finds
$$G_N[(\overline{u}d\overline{e}\nu _e^c)]<10^8G_F.$$
(37)
For tensor couplings the constraints are even stronger. The above argument only applies to the operator $`(L_1\overline{D}_1)(L_1Q_1)`$ that appears in Tab. 2, since the remaining operators also contain leptons of the second or third generation.
Note that one cannot combine two identical NP operators that contain one neutrino to derive a constraint on its coupling, since the resulting $`0\nu \beta \beta `$ amplitude would be proportional to the neutrino mass, which has no lower bound.
#### 3 Neutrino masses
Lepton number violating operators that include two neutrinos (or two antineutrinos) give rise to neutrino Majorana masses when closing the external charged fermion lines by one or two loops. If the fermions in the loop have identical flavor a contribution to the neutrino mass is generated at one loop.
Assume that one neutrino $`\nu _i`$ couples to a charged fermion $`f_k`$ with a mass $`m_k^f`$ via a scalar singlet (s) or triplet (t) with coupling $`\lambda _{ik}^{s,t}`$, while the second neutrino $`\nu _j`$ couples to another charged fermion $`f_l`$ via a scalar doublet (d) with coupling $`\lambda _{jl}^d`$. The mixing of the equal charge components of the two scalar fields gives rise to a $`L`$-violating operator as shown in Section II. Let us consider the case where the two charged fermions are identical, i.e. $`l=k`$. Then the lowest order contribution to the neutrino mass arises at one loop (see Fig. 2). Since the charged fermion propagating in the loop has to flip chirality a mass insertion is needed and the neutrino mass is proportional to $`m_k^f`$. The (momentum dependent) propagator for the scalar fields in the loop follows from (14) by replacing $`M_{1,2}^2M_{1,2}^2p^2`$, where $`p`$ is the loop momentum. Then, the neutrino mass matrix is given by
$`m_{ij}^\nu `$ $`=`$ $`iN_c{\displaystyle \underset{k}{}}(\lambda _{ik}^{s,t}\lambda _{jk}^d+\lambda _{jk}^{s,t}\lambda _{ik}^d){\displaystyle \frac{d^4p}{(2\pi )^2}\frac{m_k^f}{(m_k^f)^2p^2}\frac{\mathrm{sin}2\varphi }{2}\left(\frac{1}{M_1^2p^2}\frac{1}{M_2^2p^2}\right)}`$ (38)
$``$ $`N_c{\displaystyle \underset{k}{}}{\displaystyle \frac{\lambda _{ik}^{s,t}\lambda _{jk}^d+\lambda _{jk}^{s,t}\lambda _{ik}^d}{32\pi ^2}}m_k^f\mathrm{sin}2\varphi \mathrm{ln}\left({\displaystyle \frac{M_2^2}{M_1^2}}\right),`$ (39)
where $`N_c=3(1)`$ for intermediate quarks (leptons) and the approximation in (39) is valid for $`m_k^fM_{1,2}`$.
How this result can be used to derive constraints on the effective coupling of lepton number violating operators? Connecting two SM beta decays with an intermediate neutrino, one induces $`0\nu \beta \beta `$ with an amplitude proportional to $`G_F^2m_{11}^\nu `$ and the lower bound on $`T_{1/2}^{0\nu \beta \beta }`$ translates into a constraint on the $`m_{11}^\nu `$ entry of Majorana mass-matrix. Moreover, assuming that two of the three mentioned neutrino problems are explained by neutrino oscillations (implying $`\mathrm{\Delta }_{ij}<1`$eV<sup>2</sup>), it follows from unitarity that all of the entries of the Majorana mass-matrix can be at most of the order of the upper bound from the Troitsk tritium beta-decay experiment on the lightest neutrino mass eigenstate, $`m_1<2.5`$ eV. So we have
$$m_{ij}^\nu <\{\begin{array}{cc}0.36\text{ eV}& i=j=1\\ 3\text{ eV}& \mathrm{else}\end{array}.$$
(40)
Note that if only one of the three neutrino anomalies is explained by neutrino oscillations or if one introduces additional light (sterile) neutrinos the above argument for $`i,j1`$ does not hold. However, since our main phenomenological motivation for introducing $`L`$-violating interactions is to see whether in such a framework all the three neutrino anomalies could be explained simultaneously with three light neutrinos, we shall use the bounds in (40) in the following.
Assuming that there are no significant cancellations between the various terms that contribute to the neutrino mass in (39) it follows from (40) that
$$\lambda _{ik}^{s,t}\lambda _{jk}^d\mathrm{sin}2\varphi \mathrm{ln}\left(\frac{M_2^2}{M_1^2}\right)<\left(\frac{\mathrm{MeV}}{N_cm_k^f}\right)\{\begin{array}{cc}6.3\times 10^5& i=j=1\\ 9.5\times 10^4& \mathrm{else}\end{array}.$$
(41)
This implies that the effective coupling of the $`L`$-violating operator satisfies:
$`G_N[(\overline{f_{kL}}\nu _i)(f_{kR}\nu _j^c)]`$ $`=`$ $`{\displaystyle \frac{\lambda _{ik}^{s,t}\lambda _{jk}^d}{8\sqrt{2}}}\mathrm{sin}2\varphi \left({\displaystyle \frac{1}{M_1^2}}{\displaystyle \frac{1}{M_2^2}}\right)`$ (42)
$`<`$$``$ $`f(M_2/M_1)\left({\displaystyle \frac{\mathrm{MeV}}{N_cm_k^f}}\right)\{\begin{array}{cc}1.9\times 10^4G_F& i=j=1\\ 2.9\times 10^3G_F& \mathrm{else}\end{array},`$ (43)
where
$$f(x)\frac{1x^2}{\mathrm{ln}x^2}<1\mathrm{for}x>1,$$
(44)
and we have set the lower mass $`M_1=50`$ GeV to its minimal value. We learn that the above constraints on $`G_N`$ in many cases are stronger than those listed in the third and fourth section of Tab. 2 . In particular, all the effective couplings for lepton number violating scattering off electrons and quarks are at most of the order $`6\times 10^3G_F`$ and a few$`\times 10^4G_F`$, respectively.
## V Lepton number violating interactions and neutrino oscillation experiments
Having introduced and motivated the $`L`$-violating interactions induced by scalar mixing, we turn now to a systematic survey of those interactions that are relevant to the terrestrial, solar and atmospheric neutrino experiments.
### A Terrestrial neutrino experiments
The LSND collaboration has reported a positive signal in two different appearance channels. The first analysis uses $`\overline{\nu }_\mu `$’s from muon decay at rest (DAR) and searches for $`\overline{\nu }_e`$’s via inverse beta decay. The observed excess of $`\overline{\nu }_e`$ events corresponds to an average transition probability of
$$P(\overline{\nu }_\mu \overline{\nu }_e)=(3.1_{1.0}^{+1.1}\pm 0.5)\times 10^3.$$
(45)
Explaining this result in terms neutrino oscillations, requires $`\mathrm{\Delta }m^2`$ and $`\mathrm{sin}^22\theta `$ in the range indicated in Fig. 3 of Ref. . Taking into account the restrictions from the null results of other experiments, the preferred values of the neutrino parameters are $`\mathrm{\Delta }m^22\text{ eV}^2`$ and $`\mathrm{sin}^22\theta 2\times 10^3`$ and the lower limit on $`\mathrm{\Delta }m^2`$ for the neutrino oscillation solution is given by
$$\mathrm{\Delta }m^2>0.3\text{ eV}^2.$$
(46)
The second analysis uses $`\nu _\mu `$’s from pion decay in flight (DIF) and searches for $`\nu _e`$’s via the $`\nu _eCe^{}X`$ inclusive reaction. Again a positive signal with a transition probability $`P(\nu _\mu \nu _e)`$ similar to the one in (45), but with less statistical significance, has been reported.
Besides the orthodox neutrino oscillation hypothesis it has been proposed that the LSND signals could be due to non-standard neutrino interactions . Assuming that the result in (45) is due to New Physics interactions of strength $`G_N^\nu `$ (while there is no significant contribution from neutrino oscillations) the appearance probability is given by
$$P(e^+)=\left|\frac{G_N^\nu }{G_F}\right|^2.$$
(47)
From eqs. (45) and (47) we learn that, in order to explain the LSND result, the effective NP coupling should satisfy
$$G_N^\nu >4.0\times 10^2G_F$$
(48)
at the 90% confidence level (CL). In Ref. it has been shown that lepton flavor violating neutrino interactions cannot satisfy the condition (48) even if one allows for maximal $`SU(2)_L`$ breaking effects.
Here we investigate whether lepton number violating interactions could be large enough to be relevant for the LSND results. As we already mentioned in the Introduction the anomalous $`L`$-violating decays (4) could in principle be a possible source for the $`\overline{\nu }_e`$’s in the DAR channel of LSND. From the first section of Tab. 2 it follows that these decays can be mediated by a scalar doublet that couples to $`A^{}=L_1\overline{E}_2`$ whose charged component mixes with a scalar singlet that couples to $`B=L_1L_{\mathrm{}}`$. (Here $`\mathrm{}=\mu ,\tau `$, but for a scalar triplet also $`\mathrm{}=e`$ is possible.) Comparing the model-independent bounds for the effective couplings $`G_N^{A^{}B}<0.02G_F`$ with the required effective coupling strength in (48) we conclude that in the $`SU(2)_L`$ symmetric case $`G_N^{A^{}B}`$ is too small to explain the LSND DAR result. However, while $`SU(2)_L`$ breaking effects cannot be large, an enhancement by a factor of two, which is required to satisfy (48), is indeed conceivable. Thus we cannot rule out in a model independent way that the lepton number violating decays in (4) are the source of the LSND anomaly. However, moving to the explicit framework of SUSY $`\overline{)}R_p`$ the constraints on $`G_N^{A^{}B}`$ are stronger by a factor of four implying that even with maximal $`SU(2)_L`$ breaking one cannot fulfill (48), unless one allows for some fine-tuned cancellations.
It is interesting to ask whether the $`L`$-violating interactions
$$\nu _{\mathrm{}}pe^+n.$$
(49)
could provide an alternative explanation for the DAR signal. As one can see from the second section of Tab. 2 the processes in (49) can be induced by scalar mixing if either $`A^{}=L_{\mathrm{}}\overline{D}_1`$ to $`B=Q_1L_1`$ or $`A_{}^{}{}_{}{}^{}=Q_1\overline{D}_1`$ to $`B^{}=L_1L_{\mathrm{}}`$. While the scalar fields coupling to $`A`$ and $`A^{}`$ have to be doublets, those coupling to $`B`$ and $`B^{}`$ could be either singlets or triplets of $`SU(2)_L`$. (Note that if $`L_1L_{\mathrm{}}`$ couples to a singlet then this excludes $`\mathrm{}=e`$ due to the antisymmetry of the singlet contraction.) However, as we have noted in Section IV B 1, any operator that induces the reaction in (49) also necessarily gives rise to lepton number violating pion decays. Thus, the stringent constraints in (33) and (34) apply, unless one is willing to allow for a fine tuned cancellation in (32) by setting $`G_N^e/m_e=G_N^\mu /m_\mu `$. But even in this case according to the bound in Tab. 2 the effective coupling $`G_N^{A^{}B}`$ of the operator $`A^{}B`$ is much too small to satisfy (48). Using only the bounds from the trilinear couplings in Tab. 2 $`G_N^{A_{}^{}{}_{}{}^{}B^{}}`$ could be consistent with (48) provided that there is an enhancement from $`SU(2)_L`$ breaking effects by a factor of two. The corresponding bound within SUSY $`\overline{)}R_p`$ is only stronger by a factor of three. Although it is rather unlikely, we cannot rule out completely that the lepton number violating reactions in (49) play a role for the LSND DAR result.
We note that lepton number violating pion decays $`\pi ^+\mathrm{}^+\overline{\nu }_e`$ cannot be responsible for the $`\overline{\nu }_e`$’s observed by LSND, even though they are not helicity suppressed. First, according to (29) the BR for $`\mathrm{}=e`$ is measured to be too small. Second, for $`\mathrm{}=\mu `$ the kinetic energy of the final $`\overline{\nu }_e`$ is at most 34 MeV, which is below the threshold energy of the LSND DAR analysis.
As concerns the LSND DIF channel an interpretation of this anomaly in terms of $`L`$-violating interaction is less attractive for the following reason: The presence of additional $`L`$-violating pion decays of the form $`\pi ^+\mu ^+\overline{\nu }_{\mathrm{}}`$ cannot produce the observed $`\nu _e`$’s. Likewise $`L`$-violating interactions in the detection process could only imply that neutrons capture antineutrinos which are absent in the SM pion DIF. Hence the (generically suppressed) $`L`$-violation processes would be required for both the neutrino production and detection, ruling out this scenario as an explanation for the LSND DIF signal.
We note that the KARMEN experiment , which uses the same detection processes as LSND has found no evidence for neutrino flavor transitions ruling out a transition probability as in (45) at 90% CL. In general this situation somewhat favors an explanation of the LSND anomaly in terms of “standard” neutrino oscillations, since due to the different baselines there is still as small region in the $`\mathrm{\Delta }m^2\mathrm{sin}^22\theta `$ plane consistent with both experiments, while for new physics reactions as the source of the LSND anomalies KARMEN should observe the same transition probabilities. Still, the bound from KARMEN gains from the fact that less events were observed than expected from the background, so for conclusive evidence we will have to wait for the upcoming MiniBooNE experiment (see also Section V D).
A different search for neutrino oscillations has been performed by the CHORUS and NOMAD experiments at CERN looking for $`\nu _\mu \nu _\tau `$ oscillations transitions. In the absence of neutrino flavor transitions the relative flavor composition of the neutrino beam is predicted to be $`\nu _\mu :\overline{\nu }_\mu :\nu _e:\overline{\nu }_e=1.00:0.061:0.0094:0.0024`$ with a negligible ($`10^7`$) contamination of tau neutrinos. The search for $`\nu _\tau `$ is based on charged current tau production with subsequent detection of the various tau decay modes. Both experiments have found no indication for $`\nu _\mu \nu _\tau `$ oscillation. The upper bound from NOMAD on the transition probability is
$$P(\nu _\mu \nu _\tau )<0.6\times 10^3(90\%CL).$$
(50)
Since also the observed $`\tau ^+`$ events are in agreement with the estimated background a similar bound as in (50) applies to $`P(\nu _\mu \overline{\nu }_\tau )`$. The production of $`\tau ^+`$’s could also be induced by the lepton number violating reaction
$$\nu _\mu p\tau ^+n,$$
(51)
which would result from the operators $`(L_2\overline{D}_1)(L_3Q_1)`$ or $`(Q_1\overline{D}_1)(L_2L_3)`$ that appear in Tab. 2 . It is interesting to note that the upper bounds we obtained for the effective coupling of these operators in Section IV (see Tab. 2) are of the same order \[$`(\widehat{G}_N/G_F)^210^3`$\] as the experimental constraint from NOMAD (and CHORUS). Unfortunately the proposed TOSCA experiment that would have been sensitive to a transition probability as small as $`10^5`$ has been rejected.
### B Solar neutrino experiments
The long standing solar neutrino puzzle is now confirmed by five experiments using three different experimental techniques and thus probing different neutrino energy ranges. All these experiments observe a solar neutrino flux that is smaller than expected. The most plausible solution is that the neutrinos are massive and there is mixing in the lepton sector. Then neutrino oscillations can explain the deficit of observed neutrinos with respect to the Standard Solar Model. In the case of matter-enhanced neutrino oscillations, the famous MSW effect provides an elegant solution to the solar neutrino problem with $`\mathrm{\Delta }_{SN}`$ as given in (1).
Several authors have studied alternative solutions to the solar neutrino problem with and without neutrino masses . In the scenario with massive neutrinos $`\mathrm{\Delta }_{SN}`$ is still required to be of the same order as in (1). However, the vacuum mixing can be vanishingly small , when the effective mixing is dominantly induced by the flavor changing neutrino scattering
$$\nu _ef\nu _{\mathrm{}}f,$$
(52)
where $`\mathrm{}=\mu ,\tau `$ and $`f=e,u,d`$.
For the scenario without neutrino masses additional non-universal flavor diagonal interactions
$$\nu _{\mathrm{}}f\nu _{\mathrm{}}f,$$
(53)
are required. For both scenarios the effective couplings in (52) $`G_e\mathrm{}^f`$ have to be of the order of a few percent, while for the second scenario the difference between the effective couplings in (53) $`G_{\mathrm{}\mathrm{}}^fG_{ee}^f`$ has to be in the narrow interval $`[0.50G_F,0.77G_F]`$ ($`[0.40G_F,0.46G_F]`$) for $`f=d(u)`$ to allow for a resonant neutrino conversion. This requires rather large non-universal flavor diagonal couplings, which is ruled out for $`\mathrm{}=\mu `$ .
It is interesting to ask whether also the $`L`$-violating neutrino scattering
$$\nu _ef\overline{\nu }_{\mathrm{}}f,$$
(54)
where $`\mathrm{}=\mu ,\tau `$ and $`f=e,u,d`$ in combination with either massive neutrinos (but negligible mixing) or additional non-universal flavor diagonal interactions of the type
$$\overline{\nu }_{\mathrm{}}f\overline{\nu }_{\mathrm{}}f,$$
(55)
could give rise to matter-induced $`\nu _e\overline{\nu }_{\mathrm{}}`$ neutrino oscillation that provide an alternative solution to the solar neutrino problem. Note that the $`L_{\mathrm{}}`$ and $`L`$ conserving interactions in (53) and (55) are related by crossing symmetry. So their effective couplings are subject to the same bound.
As we have seen in Section II lepton number violating reactions as in (54) require mixing between intermediate bosons with different $`SU(2)_L`$ transformations. The third section of Tab. 2 contains various combinations of bilinears that when coupled to each other by scalars that mix can induce the $`L`$-violating neutrino scattering off a fermion as in (54). The total lepton number will only be violated if the two bilinears $`A_{q_f}=\overline{\nu }f_A`$ and $`B_{q_f}=\nu f_B`$ contain charged fermions $`f_A`$ and $`f_B`$ that belong to different presentations of $`SU(2)_L`$. Consequently, independent of the details of the model, the reordered four-fermion operator that induces the effective neutrino potential
$$_{\mathrm{int}}=\frac{G_F}{\sqrt{2}}\underset{a=S,P,T}{}(\overline{\nu ^c}\mathrm{\Gamma }^a\nu )\left[\overline{\psi }_f\mathrm{\Gamma }_a(g_a+g_a^{}\gamma ^5)\psi _f\right]+\mathrm{h}.\mathrm{c}.,$$
(56)
can only contain scalar ($`\mathrm{\Gamma }^S=I`$), pseudo-scalar ($`\mathrm{\Gamma }^S=\gamma _5`$) or tensor ($`\mathrm{\Gamma }^T=\sigma ^{\mu \nu }`$) couplings. (Axial)vector couplings are not possible, since they couple between fermions of the opposite chirality.
To be explicit consider the example within SUSY $`\overline{)}R_p`$ where the $`q=1/3`$ component of $`A^{}=L_3\overline{D}_1`$ may couple to the weak singlet $`B=L_1Q_1`$ if there is $`\stackrel{~}{b}_L\stackrel{~}{b}_R`$ mixing. The resulting four-fermion operator is:
$$_{\mathrm{int}}=\frac{\lambda _{331}^{}\lambda _{313}^{}}{M_{AB}^2}(\overline{d_R}\nu _\tau )(\overline{\nu _e^c}d_L)=\frac{\lambda _{331}^{}\lambda _{313}^{}}{M_{AB}^2}\left[\frac{1}{2}(\overline{\nu _e^c}\nu _\tau )(\overline{d_R}d_L)+\frac{1}{8}(\overline{\nu _e^c}\sigma _{\mu \nu }\nu _\tau )(\overline{d_R}\sigma ^{\mu \nu }d_L)\right].$$
(57)
The question is then whether scalar and tensor couplings can affect the neutrino propagation in dense matter significantly. The bounds from the neutrino masses in (43) indicate that the relevant effective couplings $`G_N`$ are less than $`10^{2(3)}G_F`$ for neutrino scattering of electrons (quarks). This constraint could be evaded if we only accept one measurement for mass squared difference $`\mathrm{\Delta }m^2`$ and allow for one mass eigenstate much heavier than $`2.5`$ eV. However, we still have the bounds from the trilinear couplings (see Tab. 2), which imply that $`G_N`$ is at most at the few percent level. Then the lepton number violating interactions could only affect the standard MSW oscillations if the averaged matrix element of the background fermion current in (56) is of similar order as the one from the SM weak current . In Ref. it has been shown that for (pseudo)scalar interactions the effective neutrino potential that is induced by (56) is proportional to the ratio of the neutrino mass and the characteristic fermion energy. Thus (pseudo)scalar couplings in (56) are not relevant for matter induced neutrino oscillations in the Sun. Moreover, it has been pointed out that transverse tensor couplings are not suppressed by the neutrino mass . However, in this case the effective neutrino potential is proportional to the average (transverse) polarization of the background matter. Since the polarization in the solar interior due to the magnetic field is expected to be tiny , we conclude that the $`L`$ violating neutrino scattering in (54) is not relevant for the solar neutrino problem.
### C Atmospheric neutrino experiments
Several experiments have observed an anomalous ratio between the atmospheric muon neutrino and electron neutrino fluxes . This atmospheric neutrino problem has recently been confirmed by the Super-Kamiokande high statistics data . Explaining this result in terms of “standard” neutrino oscillations requires a mass squared difference $`\mathrm{\Delta }_{AN}`$ as shown in (2).
Recently an alternative solution to the atmospheric neutrino anomaly based on new neutrino interactions was proposed . The suggested scenario is similar to the one we discussed previously for the solar neutrino, but for $`\nu _\mu \nu _\tau `$ oscillations. Even if neutrino masses are negligible the effective mixing between the flavor eigenstates could in principle be induced by the flavor changing neutrino scattering
$$n_\mu fn_\tau f,$$
(58)
where $`n=\nu ,\overline{\nu }`$ and $`f=e,u,d`$, in combination with non-universal flavor diagonal interactions
$$n_{\mathrm{}}fn_{\mathrm{}}f,$$
(59)
with $`\mathrm{}=\mu ,\tau `$. According to the effective couplings $`G_{\mu \tau }^f`$ for (58) and the difference between the effective couplings for (59), $`G_{\tau \tau }^fG_{\mu \mu }^f`$, have to be both of order $`0.1G_F`$. As has been shown in Ref. $`G_{\mu \tau }^f`$ is constrained by electroweak precision data to be at most at the few percent level ruling out such an explanation, unless one allows for some fine-tuned cancellations.
However, in view of the bounds in the fourth section of Tab. 2 on the effective couplings for the lepton number violating neutrino scattering
$$\nu _\mu f\overline{\nu }_\tau f\text{ and }\overline{\nu }_\mu f\nu _\tau f,$$
(60)
one might wonder whether such interactions could offer an alternative mechanism to solve the atmospheric neutrino anomaly. Although the effective coupling for $`(L_3\overline{D}_1)(L_2Q_1)`$ could be of order $`0.1G_F`$, such an explanation faces the same problem that we encountered in the discussion of solar neutrinos. Namely, the inherent change of the chirality of the background fermions restricts the couplings to be of scalar or tensor type. Consequently the effective neutrino potential is suppressed by either the neutrino mass or the average polarization. Thus we conclude that lepton number violating interactions do not effect atmospheric neutrinos that propagate through earth matter.
### D Future terrestrial neutrino oscillation experiments
The fundamental difference between neutrino transitions induced by new interactions and “standard” vacuum neutrino oscillations due to non-vanishing neutrino masses and mixing is that only the latter have a non-trivial $`L/E`$ (distance over energy) dependence if the neutrinos propagate in vacuum. Flavor changing neutrino interactions that conserve total lepton number in principle can induce matter-induced neutrino oscillations that are distance dependent. Among laboratory neutrinos matter effects are only relevant for long baseline experiments, where the neutrinos propagate through the earth mantle . However, for $`\nu _\mu \nu _\tau `$ transitions the flavor changing parameter $`ϵ`$ has be of order unity to be relevant for the K2K and MINOS long baseline neutrino experiments, which is inconsistent with the model-independent bounds presented in Refs. . Also for $`\nu _e\nu _\mu (\nu _\tau )`$ transitions $`ϵ`$ is at most of order $`10^5(10^2)`$ implying that earth effects will not probe the flavor changing interactions in the upcoming long baseline detectors.
As we pointed out in our discussion of solar neutrinos in Section V B matter effects on the neutrino propagation due to lepton number violating interactions are even less significant due to the suppression from the neutrino mass or the background polarization. Therefore the dominant impact on terrestrial neutrino experiments from new interactions comes from the modification of the relevant detection and/or production processes, like the reaction in (4) that we discussed for LSND. This fact allows us to distinguish the proposed solution of LSND in terms of lepton number violating interactions without any theoretical assumptions, just on the basis of the experimental observations. In the future a number of terrestrial neutrino experiments with different baselines will try to clarify the nature of neutrino oscillations. Should a certain neutrino transition channel maintain a distance independent contribution (beyond the trivial decrease of the flux inverse to the distance squared) this would signal non-standard neutrino interaction. In the following we discuss briefly some of the upcoming experiments and their potential to observe lepton number violating interactions.
The MiniBooNE experiment at Fermilab is designed to confirm (or refute) the $`\nu _e\nu _\mu `$ oscillations signals observed at LSND by searching for $`\nu _\mu \nu _e`$ transitions with an expected sensitivity to $`P(\nu _\mu \nu _e)>210^4`$ at 90% CL. Also a search for $`\overline{\nu }_\mu \overline{\nu }_e`$ seems feasible even though $`\overline{\nu }_\mu `$’s are produced less copiously (by a factor $`0.2`$). The important feature of MiniBooNE is its small background of $`\nu _e`$ and $`\overline{\nu }_e`$, which allows to search also for the lepton number violating transitions $`\nu _\mu \overline{\nu }_e`$ and $`\overline{\nu }_\mu \nu _e`$ with a sensitivity of a few times $`10^4`$. Therefore it might be possible to probe the operator $`(Q_1\overline{D}_1)(L_1L_2)`$ at the level of its model-independent bound obtained in Section IV (c.f. Tab. 2).
Several long baseline experiments will search for neutrino transitions in particular aiming at the atmospheric region of evidence for $`\nu _\mu \mu _\tau `$ oscillations. While the long baseline allows these experiments to explore small mass-squared differences $`\mathrm{\Delta }m^2>10^3\text{ eV}^2`$, the neutrino flux at the detector of the upcoming experiments is rather small yielding at most a few hundred events per year. Therefore we do not expect that these experiments could probe lepton number violating interactions anywhere close to the bounds obtained in Section IV (c.f. Tab. 2). However it might be possible to obtain interesting information from a second detector very close to the neutrino source, which is supposed to study the initial neutrino beam.
### E Supernova neutrinos
While the effects from lepton number violating neutrino scattering as in (54) and (60) are negligible for solar and atmospheric neutrinos, these reactions could be relevant for the neutrinos emerging from a supernova explosion. Here tensor interactions could affect the neutrino propagation provided that there is a very large magnetic field $`B10^{16}`$G which induces a large polarization $`\lambda _f10^210^1`$ .
## VI Conclusions
We have presented a comprehensive analysis of lepton number violating neutrino interactions. $`L`$-violating four-fermion operators involving one or two neutrinos are induced by heavy boson exchange, if there is mixing between the equal charge components of a doublet and a singlet (or a triplet) of $`SU(2)_L`$.
As an example we have discussed SUSY $`\overline{)}R_p`$, where such operators are induced by the mixing of “right-handed” sfermions that are $`SU(2)_L`$ singlets with the “left-handed” sfermions that are $`SU(2)_L`$ doublets.
We have studied four approaches to constrain the $`L`$-violating operators in a model-independent framework:
1. Constraints from the trilinear couplings,
2. Universality in pion decays,
3. Neutrinoless double beta decay,
4. Loop-induced neutrino masses.
Any non-vanishing coupling between any fermionic bilinear and a scalar field induces an effective four-fermion operator containing the bilinear and its hermitian conjugate with an effective coupling proportional to the square of the trilinear coupling over the scalar mass. Combining the upper bounds on such effective operators one can derive constraints on the lepton number violating operators that consist of two different bilinears.
The measured ratio between the BR for pion decays with final electrons and those with final muons is in good agreement with the value predicted by the SM. This implies severe constraints on NP contribution to these pion decays. Since the final neutrinos are not observed this includes lepton number violating pion decays. Moreover, since these decays are not helicity suppressed, there is a significant enhancement from the hadronic matrix element, which gives rise to stringent limits on the relevant effective couplings.
Any $`L`$-violating four-fermion operator that can be combined with the SM operator responsible for beta decay in order to induce neutrinoless double beta decay is severely constrained by the experimental limit on this process. Unlike for the “standard” neutrinoless double beta decay, where the lepton number violation is induced by the neutrino Majorana mass, in the New Physics case, $`L`$ is broken by the operator and the intermediate neutrino can contribute by its momentum rather than its mass.
Lepton number violating operators containing two neutrinos and two charged fermions of identical flavor (but opposite chirality) induce neutrino Majorana masses at one loop when the external fermions lines are connected by a fermion propagator (see Fig. 2). Using the upper bound on the lightest neutrino mass-eigenstate from the Troitsk tritium experiment and assuming a three neutrino framework with mass-splittings not larger than $`1\text{ eV}`$, we derive stringent constraints on the relevant lepton number violating operators.
Our constraints lead to the following conclusions:
* Lepton number violating neutrino scattering $`\nu _\alpha f\overline{\nu }_\beta f`$ off matter fermions $`f=e,u,d`$ are severely suppressed by the bounds from neutrino masses and do not play a role for the present solar and atmospheric neutrino anomalies. Since any effect due to such interactions would require also a polarized background, it could – at best – be relevant for supernova neutrino oscillations.
* Model-independent considerations show that only the operators $`(L_1\overline{E}_2)(L_1L_{2,3})`$ (inducing the anomalous muon decay $`\mu _L^+e_R^+\overline{\nu }_e\overline{\nu }_{\mu ,\tau }`$) could have an effective coupling at the few percent level (of $`G_F`$) and thus might be significant for the LSND anomaly. Lepton number violating neutrino capture by protons is severely constrained by the data on pion decays, and not relevant for LSND, unless one is willing to accept some fine-tuned cancellations. Within SUSY $`\overline{)}R_p`$ the relevant constraints are stronger by a factor of four, and an explanation of the LSND DAR data via lepton number violating interactions is inconsistent with the upper bound on the maximal $`SU(2)_L`$ breaking.
A solution of LSND in terms of New Physics is attractive since this way the solar and the atmospheric neutrino anomalies could be explained via standard neutrino oscillations avoiding the introduction of a sterile neutrino. However, this interpretation seems to be somewhat disfavored by the confirmation of the LSND anomaly in the DIF data and by the null signal of KARMEN.
Future terrestrial neutrino oscillation experiments that are sensitive to the “neutrino transition” probability at the level of $`10^4`$ could observe lepton number violating neutrino interactions. Any signal that does not depend on the baseline is a potential candidate for new neutrino interactions. However, the present solar and atmospheric neutrino anomalies are not (significantly) “contaminated” by such interactions.
###### Acknowledgements.
We thank Y. Grossman, St. Kolb, W.C. Louis and Y. Nir for useful discussions and comments. One of us (SB) would like to thank the Max–Planck–Institut für Kernphysik in Heidelberg, where part of this work was done, for the hospitality.
|
warning/0004/hep-ex0004008.html
|
ar5iv
|
text
|
# Track Fit Hypothesis Testing and Kink Selection using Sequential Correlations
## 1 Introduction
Trajectory fitting aims to determine a set of fitted parameters and to test the validity of the trajectory hypothesis. Both of these questions are usually addressed by minimizing a $`\chi ^2`$ constructed from the squares of the deviations between the measurements and the parameterized trajectory. The value of $`\chi ^2`$ at the minimum is used to test the adequacy of the fitted hypothesis.Cowan The $`\chi ^2`$ test, however, explicitly ignores correlations amongst the residuals<sup>1</sup><sup>1</sup>1The term residual in this paper refers to the signed distance of closest approach between a measurement and the fitted trajectory divided by the uncertainty assigned to the measurement. from the fit. Such correlations arise naturally in trajectory fitting, where adjacent measurements are in causal order. Deviations from the expected trajectory due to either a discrete change at some point (e.g. a scattering or decay in flight) or to a continuous parameter change (e.g. dE/dx energy loss or magnetic field anomalies) introduce correlated shifts in the positions of all subsequent measurements. The $`\chi ^2`$ test is not very sensitive to these trajectory deviations when they are small on the scale of the measurement errors, since it considers only the squares of the residuals.
This paper introduces new quantities for hypothesis testing that are applicable to any fits for which an ordering variable (e.g. time) can be identified.<sup>2</sup><sup>2</sup>2Not all fits are in this category: e.g. one cannot in general define a useful ordering variable for vertex fits. The mean correlation of ordered sets of residuals (nearest neighbor, next nearest neighbor, etc.) are used for this purpose. These correlations are essentially independent of $`\chi ^2`$, and test the assumption that the measurements are mutually independent (or, more generally, that the correlations amongst the measurements are properly accounted for in the fit). It should be noted that the presumption of the independence of the measurements once known sources of correlation (e.g. multiple Coulomb scattering) are taken into account is also present in sequential fitting methods such as the Kalman FilterKalman , Fruhwirth , and the correlation test developed here can be applied to the output of such a fit.
The power of these correlation statistics for hypothesis testing is studied as a function of several trajectory deviations that arise naturally in charged particle tracking. For simplicity, the trajectories studied are circular arcs, corresponding to the projection of charged particle trajectories onto a plane transverse to a uniform axial magnetic field. The following sections describe the correlation variables, the simulation used to measure their effectiveness, and the improvement in discrimination between true and false hypotheses beyond what can be achieved using $`\chi ^2`$ alone.
## 2 Description of correlation variables
The degree to which nearby measurements are correlated can be gauged by considering the mean correlation as a function of the distance between measurements. Using the standard correlation estimatorCowan to form an average correlation
$$\mathrm{r}_\mathrm{k}=\underset{\mathrm{i}}{}\frac{\delta _\mathrm{i}\delta _{\mathrm{i}+\mathrm{k}}}{\sigma _\mathrm{i}\sigma _{\mathrm{i}+\mathrm{k}}}/\sqrt{\left(\underset{\mathrm{i}}{}\frac{\delta _\mathrm{i}^2}{\sigma _\mathrm{i}^2}\right)\left(\underset{\mathrm{i}}{}\frac{\delta _{\mathrm{i}+\mathrm{k}}^2}{\sigma _{\mathrm{i}+\mathrm{k}}^2}\right)}$$
(1)
as a function of the distance between measurements gives fairly good discrimination between true and false trajectory hypotheses. The sums run from 1 to $`\mathrm{N}\mathrm{k}`$, $`\mathrm{N}`$ is the number of measurements on the trajectory, $`\delta _\mathrm{i}`$ is the signed distance to the fitted trajectory for measurement $`\mathrm{i}`$, $`\sigma _\mathrm{i}`$ is the estimated uncertainty of measurement $`\mathrm{i}`$, and $`\mathrm{k}`$ is the correlation distance: $`\mathrm{k}[1,\mathrm{N}1]`$. However, the following combination,
$$\rho _\mathrm{k}=\underset{\mathrm{i}}{}\mathrm{w}_\mathrm{i}\frac{2\delta _\mathrm{i}\delta _{\mathrm{i}+\mathrm{k}}}{\delta _\mathrm{i}^2+\delta _{\mathrm{i}+\mathrm{k}}^2}/\underset{\mathrm{i}}{}\mathrm{w}_\mathrm{i};\mathrm{w}_\mathrm{i}=\frac{\delta _\mathrm{i}^2+\delta _{\mathrm{i}+\mathrm{k}}^2}{\sigma _\mathrm{i}^2+\sigma _{\mathrm{i}+\mathrm{k}}^2}$$
(2)
motivated by considering the measure $`(\delta _\mathrm{i}\delta _{\mathrm{i}+\mathrm{k}})^2/(\sigma _\mathrm{i}^2+\sigma _{\mathrm{i}+\mathrm{k}}^2)`$, gives slightly better discrimination. This is because the correlation sought is in the actual distances $`\delta `$ from the fitted trajectory, not in the residuals $`\delta /\sigma `$. The weight factor emphasizes those pairs of measurements with significant deviations. These correlation measures satisfy $`\left|\rho _\mathrm{k}\right|1`$ for all $`\mathrm{k}`$. For uncorrelated measurements the expectation values of the $`\rho _\mathrm{k}`$ are close to zero. Negative correlations are introduced by the trajectory fit; they are small provided the number of fitted parameters is much smaller than the number of measurements.
Fig. 1(a-c) shows the expectation value<sup>3</sup><sup>3</sup>3Calculated numerically using the simulation described below. of each $`\rho _\mathrm{k}`$ as a function of the correlation distance k (for $`\mathrm{k}`$ up to 19) for true circle trajectories and for three common trajectory deviations: a discrete angular kink, an uncorrected continuous energy loss and the decay in flight of a pion to a muon. The locations of the angular kink and decay in flight were uniformly distributed along the trajectory. The fitted trajectory in each case was a circular arc. The magnitudes of the $`\rho _\mathrm{k}`$ for the incorrect hypotheses depend on the particular choice of parameters in the simulation. The number and location of the crossing points between $`\rho _\mathrm{k}`$ for the incorrect trajectory hypotheses and $`\rho _\mathrm{k}`$ for the correct hypothesis are very similar for the trajectory deviations studied here. Fig. 1(d) compares a sample with a discrete angular kink located near the center of the measurement region with a sample where the kink is located closer to the edge of the measurement region. The kink location clearly has an impact on the behavior of $`\rho _\mathrm{k}`$, most notably for large $`\mathrm{k}/\mathrm{N}`$.
Using all of the $`\rho _\mathrm{k}`$ values gives the best discrimination between the correct hypothesis and a particular type of incorrect hypothesis, but does not provide optimal discrimination for all types of incorrect hypothesis as can be seen by considering the different shapes (note in particular the locations of the extrema) of the trajectory hypotheses shown in Fig. 1. Furthermore, most of the discrimination power is concentrated at small $`\mathrm{k}`$, since the r.m.s. of the $`\rho _\mathrm{k}`$ distributions expected for the correct hypothesis are smallest there.<sup>4</sup><sup>4</sup>4The number of correlation measures summed for a given $`\mathrm{k}`$ is $`\mathrm{N}\mathrm{k}`$, giving a statistical error proportional to $`1/\sqrt{\mathrm{N}\mathrm{k}}`$. These considerations, along with the essentially linear behavior of the difference $`\rho _\mathrm{k}_{\mathrm{false}}\rho _\mathrm{k}_{\mathrm{true}}`$ as a function of the correlation distance $`\mathrm{k}`$ for small $`\mathrm{k}/\mathrm{N}`$, lead us to the following test statistic:
$$\lambda =\underset{\mathrm{k}=1}{\overset{\mathrm{L}}{}}\mathrm{C}_\mathrm{k}\rho _\mathrm{k}\mathrm{with}\mathrm{C}_\mathrm{k}=\frac{2}{\mathrm{L}(\mathrm{L}1)}(\mathrm{L}\mathrm{k})\mathrm{and}\underset{\mathrm{k}=1}{\overset{\mathrm{L}}{}}\mathrm{C}_\mathrm{k}=1.$$
(3)
Choosing $`\mathrm{L}`$ as the nearest integer to $`\mathrm{N}/8`$ was found to give good sensitivity to the trajectory deviations studied.
## 3 Description of simulation
The sensitivity of $`\lambda `$ to trajectory deviations was studied using a simple simulation. Charged particles were generated and tracked through a uniform axial magnetic field and measured points were generated in the plane orthogonal to the magnetic field direction. The measurement uncertainty was taken as Gaussian. Trajectories were generated with parameters typical of charged particle tracking detectorsbabartdr :
* average hit resolution: 150 $`\mu \mathrm{m}`$;
* variation in resolution: factor of 5 between the best and worst measured points;
* radial difference between first and last measurement layer: 54 cm;
* number of measurements: varied from 20 to 160;
* magnetic field: axial field of magnitude 1.5 Tesla;
* initial particle momentum between 0.5 GeV/c and 5 GeV/c.
The measurements were uniformly distributed along the trajectories, the hit efficiency was unity and no noise hits were generated. The location of discrete trajectory deviations (angular kink or particle decay) was randomly distributed along the trajectory. The generated points were fitted to a circle assuming perfect pattern recognition, i.e. all generated measurements were used in the fit.
## 4 Results
The generated data were used to study the discrimination power of $`\lambda `$ and of $`\chi ^2`$ as a function of specific trajectory deviations. Fig. 2 shows the fraction of trajectories surviving a cut on $`\lambda `$ or on the $`\chi ^2`$ probability $`\mathrm{P}(\chi ^2)`$ as a function of the position of a discrete angular kink, where the kink angle was uniformly distributed between $`\pm 0.02`$ radians. The cuts were set to give 95% efficiency for true circular trajectories. There is little discrimination power for kinks occurring at either end of the measurement region. Based on this, only those trajectory deviations occurring within a fiducial region consisting of the central 80% of the measurements were selected for subsequent study.
The fraction of these selected trajectories surviving a cut that gives 95% efficiency for true circular trajectories is shown as a function of the size of the angular deviation in Fig. 3. The correlation variable $`\lambda `$ is more powerful than $`\mathrm{P}(\chi ^2)`$ provided $`\mathrm{N}>20`$ and becomes relatively more powerful as the number of measurements increases. This is to be expected, since the physical correlation length sampled by $`\lambda `$ is of the order of $`\frac{1}{8}`$ of the track length, and increasing the density of measurements allows a more precise determination of the correlation. Similar behavior is observed for different choices for the efficiency for true circular trajectories. The power of $`\lambda `$ and $`\mathrm{P}(\chi ^2)`$ to discriminate $`\pi \mu `$ decays from true circular trajectories is shown in Fig. 4. Note that the decay angle between the
$`\pi `$ and $`\mu `$ does not in general lie in the measurement plane, and that the momentum of the muon will be smaller than that of the pion. The correlation variable $`\lambda `$ is again a significantly better discriminant than is $`\mathrm{P}(\chi ^2)`$.
The extent to which the discrimination afforded by $`\lambda `$ is optimal was studied. The measurements along trajectories generated with a discrete angular kink were fitted using both the correct hypothesis, i.e. two circular arcs of constant curvature with an angular deviation at a point, and using the nominal (incorrect) hypothesis of a single circular arc. The ratio of the resulting $`\chi ^2`$ probabilities $`\mathrm{P}_{\mathrm{circle}}(\chi ^2)/\mathrm{P}_{\mathrm{correct}}(\chi ^2)`$ is an optimal discriminant variable in this case.<sup>5</sup><sup>5</sup>5In practice, trajectory deviations of several different types may be present in a sample, so fitting for one particular type of deviation will not be optimal. The position of the angular kink was taken as either the true position of the generated deviation (denoted “fixed $`\mathrm{R}_{\mathrm{kink}}`$” in Fig. 5) or as the best fit value after considering all
potential kink positions (denoted “fitted $`\mathrm{R}_{\mathrm{kink}}`$” in Fig. 5). The kink position will in general be unknown, so the “fitted $`\mathrm{R}_{\mathrm{kink}}`$” curve is in practice the best one can achieve. As is seen in Fig. 5, the correlation variable $`\lambda `$ gives discrimination that approaches the optimal value as the number of measurements increases.
## 5 Discussion
The correlation variables $`\rho _\mathrm{k}`$ introduced here test the assumption that a set of measurements are mutually independent once known sources of correlation have been taken into account. Use of the simple combination of $`\rho _\mathrm{k}`$ introduced here as $`\lambda `$ leads to an improvement in the sensitivity for detecting small trajectory deviations relative to that achievable using $`\mathrm{P}(\chi ^2)`$. This additional goodness-of-fit test is independent of the $`\chi ^2`$ test, and can be applied to the residuals from both traditional least squares fits and from Kalman filter fits. The improved sensitivity to small deviations can be qualitatively understood by recognizing that $`\chi ^2`$ has a quadratic dependence on individual deviations, while $`\lambda `$ has a linear dependence. For larger deviations (not shown here) $`\mathrm{P}(\chi ^2)`$ becomes a more powerful discriminant than $`\lambda `$, as expected. For the deviations studied, the gain from combining the $`\lambda `$ and $`\mathrm{P}(\chi ^2)`$ tests was negligible.
The $`\lambda `$ and $`\chi ^2`$ tests have different dependencies on the input to the fit. The $`\chi ^2`$ test is sensitive to the scale of the assigned measurement errors $`\sigma _\mathrm{i}`$ and can be compromised by mis-estimates of and non-Gaussian contributions to the measurement errors. The $`\lambda `$ test is insensitive to the scale of the $`\sigma _\mathrm{i}`$. It is, however, sensitive to correlations introduced in calibration procedures. The extent to which this is a practical problem in using the $`\lambda `$ test is a function of detector design and calibration. In particular we expect the $`\lambda `$ test to be most useful in devices where the effect of calibrations is randomized over the measurements on a trajectory (e.g. in small cell drift chambers, where the drift direction changes layer by layer), and to be less effective in devices where coherent effects dominate (e.g. in detectors employing a jet cell design).
Potential uses for the $`\lambda `$ test in charged particle tracking involve selecting decays in flight and enabling high quality track samples (with reduced non-Gaussian tails on the track parameter resolutions due to trajectory deviations) to be selected. Given the ease with which it can be calculated, a test on $`\lambda `$ might serve as a filter for selecting tracks on which more computationally intensive testsFruhwirth , Cousins will be performed. The correlations measured by the $`\rho _\mathrm{k}`$ parameters may find use in a broader range of applications.
## 6 Acknowledgments
The authors would like to thank Dr. Michael Roney for useful discussions and to acknowledge Louis Desroches, whose work with one of us (Kowalewski) on run test variables was a precursor to the present work.
|
warning/0004/cond-mat0004210.html
|
ar5iv
|
text
|
# Effect of Substitutional Impurities on the Electronic States and Conductivity of Crystals with Half–filled Band
## I Introduction
The investigation of effects of weak localization in low–dimensional electronic systems still preserves its popularity today, since experimental studies of these systems give occasionally surprising results like metal–insulator phase transitions in the two–dimensional (2D) electron gas at zero magnetic field and unusual behavior of dephasing time in quasi one–dimensional metallic wires at low temperatures. Impurity effects in one–dimensional (1D) disordered systems have been studied accurately due to the existence of methods which give exact results for the cases of both weak and strong disorder. Also perturbative approaches give good results for three–dimensional (3D) weakly disordered systems.
Weak localization corrections to the kinetic coefficients of 2D disordered systems are logarithmically divergent corrections and it is hard to sum perturbatively the leading divergent contributions. The dimension two is the marginal dimension for the localization problem and a small external perturbation can change the character of localization in these systems.
According to the scaling theory of Abrahams et. al. all states of 1D and 2D electronic gases moving in the field of randomly distributed impurities are completely localized irrespective of the degree of randomness, and in 3D the Anderson metal–insulator phase transition occurs with increasing impurity concentration. Notice that the result of the scaling theory for 1D disordered systems is in agreement with exact results. The scaling theory, and also the diagrammatic approach to the problem for 2D weakly disordered systems has revealed logarithmic quantum corrections to the conductivity, which tend towards localization. The logarithmic correction to the conductivity found in \[\] can be expressed as
$$\sigma \sigma _0=\frac{e^2}{2\pi ^2}\mathrm{ln}\frac{\tau ^{}}{\tau }$$
(1)
where $`\sigma _0`$ is the Drude conductivity, $`\tau ^{}=\mathrm{min}\{\tau _\phi ,\frac{L}{v_F},\frac{1}{\omega }\}`$ with $`\tau `$ and $`\tau _\phi `$ being the elastic and inelastic scattering times, respectively; $`L`$ is the linear size of a 2D system and $`\omega `$ is the external frequency.
Notice that weak disorder gives no singular contribution to the density of states (DOS) $`\rho (ϵ)`$ of electron gas models when the condition $`p_Fl1`$ or $`ϵ_F\tau 1`$ is satisfied ($`p_F`$ and $`ϵ_F`$ are the Fermi momentum and the Fermi energy; $`l`$ is the mean free path). According to the Einstein relation $`\sigma =e^2\rho D`$ the quantum correction Eq.(1) to $`\sigma `$ is due to changes in the diffusion coefficient $`D`$.
Even short–range and weak correlations in weakly disordered metals have been shown to result in nontrivial quantum corrections to the DOS of low–dimensional conductors near the Fermi level. Electron–electron interactions in disordered systems also give rise to quantum corrections to the conductivity, which tend to localize the electronic states. Further, the DOS of strongly doped semiconductors in the presence of long range Coulomb interaction vanishes at the Fermi surface, which is commonly referred to as Coulomb gap.
Disordered metals in most of the papers on localization problems are modeled as a free electron gas moving in the random field of impurities. However, doped crystals with low concentrations of impurities usually preserve their periodical structures and the impurity atoms in most cases substitute the host atoms of the lattice. Substitutional impurities appear to have strong influence on the physical properties of low–dimensional lattices near the commensurate values of the electron wave length $`\lambda `$ with the lattice constant $`a`$ \[\].
In this paper we shall study the effects of substitutional impurities with small concentration on the density of states and the conductivity of a 2D square lattice. The effects of commensurability on the DOS for the 3D simple cubic lattice will also be considered for the sake of completeness.
Bragg reflections of the electronic wave on the Brillouin zone boundary in the process of impurity scattering become essential as the middle of the band is approached. This process introduces a new relaxation time for umklapp scattering to the problem, which increases for a deviation from half–filling due to the distortion of the commensurability condition for the Bragg reflection.
The lattices under consideration have square structure for 2D and simple cubic structure for 3D crystals. Our calculations show that umklapp scattering strongly changes the energy dependence of the DOS close to half–filling, causing it to decrease. For the 2D square lattice the DOS of the pure system has a logarithmic van Hove singularity at the middle of the band. This peak is shown to be preserved for the lattice with substitutional impurities. However its energy dependence is changed strongly, and the peak becomes narrower. For the 3D lattice, the van Hove singularities lie far from the center of the band and umklapp scattering forms a shallow dip on the Fermi surface, the depth of which increases with the impurity concentration.
Periodicity causes the appearance of a new class of diagrams which give a contribution to the conductivity $`\sigma (\omega )`$. These corrections to the conductivity can be separated into two classes: corrections due to the diffusion coefficient and due to the DOS. Although quantum corrections to the diffusion coefficient slightly increase $`\sigma (\omega )`$, the total corrections sum up to a decrease of $`\sigma (\omega )`$ due to DOS corrections in the vicinity of half–filling.
The paper is structured in the following form. In Sec.II the method is described, where new diffuson and cooperon blocks due to umklapp scattering are introduced and calculated. Sec.III is devoted to the calculation of corrections to the DOS near half–filling for 2D and 3D lattices. In Sec.IV, the calculation of the conductivity of the 2D square lattice with substitutional impurities is presented. In Sec.V, a conclusion and a discussion of the results are given.
## II Description of the method
The Hamiltonian of a $`d`$-dimensional simple cubic lattice with substitutional impurities is given by
$$\widehat{H}=\widehat{H}_0+V(𝐫)$$
(2)
where $`\widehat{H}_0`$ is the tight–binding Hamiltonian of noninteracting electrons in a regular lattice with lattice constant $`a`$ and nearest–neighbor hopping; $`V(𝐫)=_iU(𝐫𝐑_i)`$ is the impurity potential with $`𝐑_i`$ being the positional vector of an impurity randomly located on the $`i`$-th lattice site.
By introducing the second quantization operators $`c_{𝐩,\sigma }^{}`$ and $`c_{𝐩,\sigma }`$ for an electron with momentum $`𝐩`$ and spin $`\sigma `$, Eq.(2) can be rewritten as
$$\widehat{H}=\underset{𝐩,\sigma }{}ϵ(𝐩)c_{𝐩,\sigma }^{}c_{𝐩,\sigma }+\underset{𝐩,𝐪,𝐆,\sigma }{}\rho _{\mathrm{imp}}(𝐪)U(𝐪)c_{𝐩,\sigma }^{}c_{𝐩+𝐪+𝐆,\sigma }$$
(3)
where $`\rho _{\mathrm{imp}}(𝐪)=L^d_i\mathrm{exp}(i\mathrm{𝐪𝐑}_i)`$ and $`U(𝐪)`$ is the Fourier transform of a single impurity potential. Throughout this paper, we work in units of $`\mathrm{}=1`$. The momenta $`𝐩`$ and $`𝐪`$ vary in the first Brillouin zone and $`𝐆`$ is a reciprocal lattice vector. $`ϵ(𝐩)`$ in Eq.(3) is the energy spectrum of an electron on a $`d`$-dimensional square lattice,
$`ϵ(𝐩)`$ $`=t[2\mathrm{cos}(p_xa)\mathrm{cos}(p_ya)]`$ $`\text{for}d=2`$ (4)
$`ϵ(𝐩)`$ $`=t[3\mathrm{cos}(p_xa)\mathrm{cos}(p_ya)\mathrm{cos}(p_za)]`$ $`\text{for}d=3`$ (5)
where $`p_{x,y,z}=\frac{2\pi }{aN_{x,y,z}}n_{x,y,z}`$ with $`\frac{N_{x,y,z}}{2}<n_{x,y,z}\frac{N_{x,y,z}}{2}`$ and $`t`$ is the tunneling integral for nearest–neighbor sites.
The electronic bandwidth is $`W=4t`$ for $`d=2`$ and $`W=6t`$ for 3D systems. As half–filling is approached, which corresponds to $`ϵ_F=2t`$ for 2D and $`ϵ_F=3t`$ for 3D system, the Fermi surface becomes nested, i.e. it exhibits surface elements which can be mapped onto each other through one nesting vector. In our case, the nesting condition is satisfied for all parts of the surface, and for 2D square lattices, the Fermi surface at half–filling is flat (see Fig.1).
The second part of the Hamiltonian $`\widehat{H}`$ in Eq.(3), which describes an electron scattering on randomly distributed impurities, contains both normal (for $`𝐆=0`$) and umklapp (for $`𝐆0`$) processes. The impurity concentration is assumed to be small, so that the Born approximation is appropriate to estimate the scattering process on the impurities. The impurity potential in this case is chosen to be a $`\delta `$-correlated Gaussian potential with zero average value.
Conventional diagrammatic techniques are applied to calculate the effects of umklapp scattering on the DOS and the conductivity. The bare Green’s functions $`G_{R,A}^0(ϵ,𝐩)`$ at zero temperature are given as
$$G_{R,A}^0(ϵ,𝐩)=\frac{1}{ϵ(ϵ(𝐩)ϵ_F)+\frac{i}{2\tau }\mathrm{sign}ϵ}$$
(6)
where $`ϵ>0`$ and $`ϵ<0`$ correspond to retarded $`G_R^0`$ and advanced $`G_A^0`$ Green’s functions, respectively. The energy spectrum $`ϵ(𝐩)`$ in Eq.(6) is expressed by Eqs.(4) and (5) for 2D and 3D cases, respectively. For momenta lying close to the Fermi surface (in the first Brillouin zone) the energy spectrum can be linearized around the Fermi surface and $`ϵ(𝐩)ϵ_F𝐯_F(𝐩𝐩_F)=v_F(|𝐩|p_F)\mathrm{cos}\alpha `$, where $`𝐯(𝐩)=\frac{ϵ}{𝐩}=at\{\mathrm{sin}p_xa,\mathrm{sin}p_ya\}`$ is the group velocity of the electron wave packet. In contrast to the electron gas model the linearized energy spectrum contains the additional parameter $`\mathrm{cos}\alpha `$ due to non–collinearity of the momentum and velocity vectors. For small band filling, the Fermi surface is very similar to a sphere, so this factor can be set equal to unity. Approaching half–filling, this parameter becomes weakly varied, e.g. $`\frac{1}{\sqrt{2}}\mathrm{cos}\alpha 1`$ for a 2D system at half–filling, where the range of variation of $`\mathrm{cos}\alpha `$ is maximal. As an approximation, we can always set $`\mathrm{cos}\alpha =1`$. In the vicinity of commensurate points, especially at half–filling, electron scattering on impurities with large momentum transfer involves umklapp processes. The procedure of expansion of the energy spectrum around the Fermi surface can in this case be performed after the separation of the large momentum transfer $`𝐐_0`$.
In weak localization theory the maximally crossed diagrams are responsible for the low temperature quantum corrections to the kinetic properties of low–dimensional disordered systems, modeled as an electron gas moving in the field of randomly distributed impurities . These diagrams can be redrawn as ladder diagrams in a particle–particle channel (Fig.2a). The cooperon block has a pole for small total momentum of particles, $`|𝐪|l1`$, and for small energy difference $`|\omega |\tau 1`$. The cooperon block turns out to have a diffusion pole also for large momentum transfer when the total momenta of the particles are $`(𝐐_0+𝐪)`$ with $`|𝐪|l1`$ and when the total energy is small. In this case each act of scattering on an impurities involves a large momentum transfer close to $`𝐐_0`$. This process of course takes place mainly around half–filling when $`𝐐_0=2𝐩_F`$, and it consists of simultaneous Bragg reflection of the electron in the process of scattering on the impurity, with relaxation time $`\tau _\pi `$. This scattering process will be further referred to as $`\pi `$–scattering.
Two different scattering processes and corresponding relaxation times can be separated: Normal scattering with $`\tau _N`$ (see Fig.3a), when the scattering on an impurity maintains the electron’s momentum inside the first Brillouin zone after scattering and $`\pi `$–scattering, with a relaxation time $`\tau _\pi `$ (Fig.3b). Here, momentum conservation is violated and this scattering process corresponds to an umklapp process. The total relaxation time $`\tau `$ can then be expressed as $`\frac{1}{\tau }=\frac{1}{\tau _N}+\frac{1}{\tau _\pi }`$. Notice here that the expression for the normal relaxation time $`\tau _N`$ according to Fig.3a for the 2D case can be given as
$$\frac{1}{\tau _N}=\frac{C_{\mathrm{imp}}}{(2\pi )^2}_S\frac{\mathrm{d}𝐒}{|𝐯_𝐤|}|U(𝐩_F,𝐒)|^2$$
(7)
We assume $`\tau _N`$ to be constant and independent of the band filling. This means that the singularities of the integrand due to vanishing velocity at the saddle points are compensated by appropriate zeros of the impurity potential.
To illustrate diagrammatically the impurity scattering including umklapp processes we represent the Green’s function of an electron with large momentum by a dashed line. The ladder series for the cooperon block with large momentum transfer, $`C_\pi (𝐪,ϵ)`$, is diagrammatically shown in Fig.2b. Notice here that the vertices c) and d) in Fig.3 are irrelevant for the cooperon blocks $`C_\pi (𝐪,ϵ)`$ and $`C_N(𝐪,ϵ)`$ (Fig. 2a,b). Indeed, the dashing of the line alone has no physical meaning, it has only meaning for an electron scattering with large momentum transfer. In ladders, the vertices c) and d) of Fig.3 can always be transformed into normal vertices or those of Fig.3b by proper redefinition of the internal momenta.
In the process of propagation in the particle–particle channel the momenta of the particles for the normal cooperon block $`C_N(𝐪,\omega )`$ lie on opposite sites of the Fermi surface during the diffusion process. In contrast to this, the momenta for the two particle propagator $`C_\pi (𝐪,ϵ)`$ with umklapp process lie on the same section of the Fermi surface and each act of impurity scattering coherently relocates the particles to the opposite section of the Fermi surface due to Bragg reflection. We will refer to this two–particle propagator $`C_\pi (𝐪,ϵ)`$ as $`\pi `$-cooperon.
The same situation occurs for the $`\pi `$–diffusion block $`D_\pi (𝐤,ϵ,\omega )`$ (Fig.2) with a large difference ($`𝐐_0+𝐤`$) of electron and hole momenta, with $`|𝐤|l1`$. The comparison of the normal diffuson block $`D_N(𝐤,ϵ,\omega )`$ with the $`\pi `$–diffuson $`D_\pi (𝐤,ϵ,\omega )`$ reveals again strong differences between these two processes. In the process of propagation in the $`N`$–diffusion channel , electron and hole lie close to each other on the Fermi surface, with momenta $`𝐩`$ and $`𝐩+𝐤`$, and $`|𝐤|l1`$. They diffuse on the Fermi surface and keep the proximity of the momenta after each act of scattering. In contrast to this, the momenta of electron and hole for the $`\pi `$–diffuson block $`D_\pi `$ are located on opposite sides of the Fermi surface. Each scattering on the impurities interchanges the positions of electron and hole on the Fermi surface. Now let us sum the series shown in Fig.2 for $`C_N`$, $`C_\pi `$, and $`D_\pi `$.
Summing up the ladder series in Fig.2a, the following expression for $`C_N(𝐪,\omega )`$ is obtained:
$$C_N(𝐪,\omega )=C_{\mathrm{imp}}|U(0)|^2\left\{\frac{\sqrt{(1i\tau |\omega |)^2+(ql)^2}}{\sqrt{(1i\tau |\omega |)^2+(ql)^2}\frac{\tau }{\tau _N}}\mathrm{\Theta }(ϵ(ϵ+\omega ))+\mathrm{\Theta }(ϵ(ϵ+\omega ))\right\}$$
(8)
Far from half–filled electronic band, $`\tau _\pi \tau _N`$ and $`\tau \tau _N`$. Therefore, $`C_N(𝐪,\omega )`$ has a diffusion pole for $`|\omega |\tau 1`$ and $`|𝐪|l1`$. As half–filling is approached, umklapp scattering is intensified and $`\tau _\pi \tau _N`$ and $`\tau \tau _\pi `$. As a result, the diffusion pole of $`C_N(𝐪,ϵ)`$ disappears at half–filling. In this regime, the Fermi surface of the square or simple cubic crystal becomes nested (see Fig.1 for the 2D case) with a nesting vector $`𝐐_0=\{\pm \frac{\pi }{a},\pm \frac{\pi }{a}\}`$ for the 2D lattice and $`𝐐_0=\{\pm \frac{\pi }{a},\pm \frac{\pi }{a},\pm \frac{\pi }{a}\}`$ for the 3D lattice. In this case, the following particle–hole symmetry of the electron dispersion with respect to the vector $`𝐐_0`$ holds for the half–filled band:
$$ϵ(𝐩+𝐐_0)ϵ_F=[ϵ(𝐩)ϵ_F]$$
(9)
The main contribution to the low temperature properties now gives the $`\pi `$–cooperon $`C_\pi (𝐪,ϵ,\omega )`$ in Fig.2b. One gets the following expression for $`C_\pi (𝐪,ϵ,\omega )`$ by summing the ladder series in Fig.2:
$$C_\pi (𝐪,ϵ,\omega )=C_{\mathrm{imp}}|U|^2\frac{J_\pi ^2}{1J_\pi ^2}$$
(10)
where $`J_\pi `$ is the expression for an elementary “bubble” in the ladder series and
$$J_\pi =C_{\mathrm{imp}}\frac{\mathrm{d}^dp}{(2\pi )^d}|U|^2G(𝐩,ϵ),G(𝐩+𝐪+𝐐_0,ϵ+\omega )$$
(11)
To calculate $`J_\pi `$, the large momentum $`𝐐_0`$ is removed using the electron–hole symmetry relation Eq.(9), and after this the energy spectrum is linearized around the Fermi surface. As a result we get the following expression for $`C_\pi (𝐪,ϵ,\omega )`$:
$$C_\pi (𝐪,ϵ,\omega )=C_{\mathrm{imp}}|U(2p_F)|^2\left\{\frac{(\tau /\tau _\pi )^2}{(1i\tau |2ϵ+\omega |)^2+(ql)^2(\tau /\tau _\pi )^2}\mathrm{\Theta }(ϵ(ϵ+\omega ))+\mathrm{\Theta }(ϵ(ϵ+\omega ))\right\}$$
(12)
So the $`\pi `$–cooperon has a diffusion pole for total momenta close to $`𝐐_0`$ ($`|𝐤|l1`$) and small total energy $`|2ϵ+\omega |\tau 1`$ of the particles. The diffusion block $`D_\pi (𝐤,ϵ,\omega )`$, which differs from $`C_\pi (𝐪,ϵ,\omega )`$ through time reversal of one electron line, has a pole for large ($`𝐐_0+𝐤`$) momenta difference with $`|𝐤|l1`$ and small total energy of electron and hole (see Fig.2c). The calculation of $`D_\pi (𝐤,ϵ,\omega )`$ is similar to that for $`C_\pi (𝐪,ϵ,\omega )`$ and we obtain
$$D_\pi (𝐤,ϵ,\omega )=C_{\mathrm{imp}}|U(2p_F)|^2\left\{\frac{(\tau /\tau _\pi )^2}{(1i\tau |2ϵ+\omega |)^2+(kl)^2(\tau /\tau _\pi )^2}\mathrm{\Theta }(ϵ(ϵ+\omega ))+\mathrm{\Theta }(ϵ(ϵ+\omega ))\right\}$$
(13)
After this, we can calculate the corrections to the DOS and the conductivity due to umklapp processes.
## III Density of States at Half–filling
The one–particle DOS of the regular $`d`$–dimensional lattice is expressed as
$$\rho _0^{(d)}=\frac{2}{(2\pi )^d}_S\frac{\mathrm{d}𝐒}{|ϵ(𝐤)|}$$
(14)
where $`\mathrm{d}𝐒`$ is an element of an isoenergetic surface in $`d`$–dimensional space. $`\rho _0^{(d)}`$ has a van Hove singularity at the points where the group velocity of the electron wave packet $`𝐯_k=ϵ(𝐤)`$ vanishes.
The DOS of a clean 1D lattice with constant spacing $`a`$ is easily calculated to be $`\rho _0^{(1)}=\frac{1}{\pi a\sqrt{ϵ(2tϵ)}}`$. This expression shows that $`\rho _0^{(1)}`$ is a regular function of the energy near half–filling when $`ϵϵ_F=t`$ and it has a power–like singularity when $`ϵ`$ approaches the band edges $`ϵ0,2t`$. The bare DOS of a 2D crystal with an energy spectrum given by Eq.(4) has a logarithmic van Hove singularity in the middle of the band:
$$\rho _0^{(2)}=\frac{1}{\pi ^2a^2t}K\left(\sqrt{1(\frac{ϵ}{2t})^2}\right)=\{\begin{array}{cc}\frac{2}{\pi a^2\sqrt{4t^2ϵ^2}}\mathrm{ln}\frac{4t^2ϵ^2}{ϵ^2}\hfill & \text{for }|ϵ|2t\hfill \\ \frac{1}{\pi a^2|ϵ|}\hfill & \text{for }|ϵ|2t\hfill \end{array}$$
(15)
Here and below, the electron energy $`ϵ`$ is measured from the middle of the band, i.e. $`ϵ=0`$ corresponds to half–filling. The van Hove singularity of simple 3D crystals with nearest–neighbor hopping exhibits cusps near $`|ϵ|=t`$:
$$\rho _0^{(3)}=\mathrm{const}\frac{2}{\pi ^2t^{3/2}a^3}\sqrt{ϵt}$$
(16)
In this section we will study effects of substitutional impurities on the DOS of 2D and 3D crystals. Randomly distributed impurities in $`d`$–dimensional electron gases have no effect on the DOS.
The DOS of 2D and 3D simple (cubic) crystals with substitutional impurities turns out to have a quantum correction near the middle of the band even for noninteracting electrons. Effects of commensurability on the DOS and on the kinetics of 1D disordered crystals near the middle of the band have been calculated by many authors. Dyson first pointed out that the DOS of phonons of a 1D disordered chain has a singularity as $`\rho ^{(1)}|ϵ|^1\mathrm{ln}^3|ϵ|`$ near the middle of the band. Notice that the singular increasing of the DOS in a 1D system at half–filling is not connected with the van Hove singularity, since the latter is located at the band edge for the 1D system. Instead it is mediated by the interference of impurity scattering and Bragg reflections at half–filling. Later an analogous singularity has been found in the electronic DOS of many 1D models. Since the 1D scattering problem is characterized by forward and backward scattering processes, 1D models with off–diagonal disorder display a singularity at the center of the band only if the forward scattering amplitude turns to zero. Small forward scattering with Bragg reflection seems to enhance localization and as a result the DOS divergence in the band center is blurred.
Effects of substitutional impurities on the DOS of 2D (square, honeycomb and triangular) and 3D (simple cubic) lattices have been studied computationally for cases with diagonal and off–diagonal disorder. The existence of a van Hove singularity in the DOS of a two sublattice model for $`d2`$ has been shown in \[\] on the basis of an $`\frac{1}{n}`$ expansion in disordered systems with $`n`$ orbitals per site for energies approaching the band center. In all cases diagonal disorder has been shown to suppress the van Hove singularity, whereas it is preserved for the 2D simple square lattice with off-diagonal disorder.
We start from the following expression for the DOS
$$\rho ^{(d)}(ϵ)=\frac{2}{\pi }\mathrm{Im}\frac{\mathrm{d}^dp}{(2\pi )^d}G_R(𝐩,ϵ)$$
(17)
which expresses $`\rho (ϵ)`$ by means of the retarded Green’s function $`G_R(𝐩,ϵ)`$. The new class of diagrams which give the dominating contribution to the self–energy of $`G_R(𝐩,ϵ)`$ is drawn in Fig.4.
These diagrams represent the contributions to the self–energy in first order in cooperon and diffuson blocks ($`C_\pi (𝐪,ϵ)`$ and $`D_\pi (𝐤,ϵ)`$). Since $`C_\pi (𝐪,ϵ)`$ and $`D_\pi (𝐤,ϵ)`$ do not carry an external frequency ($`\omega =0`$), their expressions are obtained from Eqs.(12) and (13) with $`\omega =0`$:
$`C_\pi (𝐪,ϵ)`$ $`=C_{\mathrm{imp}}|U(2p_F)|^2\left({\displaystyle \frac{\tau }{\tau _\pi }}\right)^2{\displaystyle \frac{1}{(12i\tau |ϵ|)^2+(ql)^2(\tau /\tau _\pi )^2}}`$ (18)
$`D_\pi (𝐤,ϵ)`$ $`=C_{\mathrm{imp}}|U(2p_F)|^2\left({\displaystyle \frac{\tau }{\tau _\pi }}\right)^2{\displaystyle \frac{1}{(12i\tau |ϵ|)^2+(kl)^2(\tau /\tau _\pi )^2}}`$ (19)
As $`\tau _\pi \tau `$, both blocks have a diffusion pole for $`|ϵ|\tau 1`$ and $`ql,kl1`$. The cooperon and the diffuson blocks give rise to logarithmic corrections to the physical parameters for $`d=2`$ and $`\sqrt{|ϵ|\tau }`$ corrections for $`d=3`$. Therefore, diagrams of higher order in the cooperon and diffuson blocks for the self–energy parts are not necessary for 3D systems for $`|ϵ|\tau 1`$. Nevertheless the logarithmic divergency existing in the first order contribution to the self–energy for 2D systems requires to examine higher order logarithmic corrections. The structure of the diagrams which contain all possible combinations of diffuson and cooperon blocks becomes complicated with increasing number of inserted blocks.
However, as it has been shown by various theoretical methods, higher logarithmic corrections to the conductance cancel each other and the leading correction to the conductance is only the first order logarithmic term.
High order corrections to the self–energy seem to cancel each other also in our case. The second order diagrams which give logarithm–squared contributions to the self–energy are drawn in Fig.5. Straightforward calculations show that the sum of these diagrams gives zero.
The retarded Green’s function $`G_R(ϵ,𝐩)`$ is expressed by the sum of the first–order self–energy parts $`\mathrm{\Sigma }(ϵ,𝐩)`$ in Fig.4 according to the Dyson equation as
$$G_K(ϵ,𝐩)=\frac{1}{(G_R^0(ϵ,𝐩))^1\mathrm{\Sigma }(ϵ,𝐩)}=\underset{n=0}{\overset{\mathrm{}}{}}\left(G_R^0(ϵ,𝐩)\right)^{n+1}\left(\mathrm{\Sigma }(ϵ,𝐩)\right)^n$$
(20)
By summing the diagrams in Fig.4 one gets the expression for $`\mathrm{\Sigma }(ϵ,𝐩)`$:
$$\mathrm{\Sigma }(ϵ,𝐩)=\frac{\mathrm{d}^dk}{(2\pi )^d}\left\{C_\pi (ϵ,𝐤)\frac{2\tau }{\tau _\pi }\left(1\frac{\tau }{\tau _\pi }\right)D_\pi (ϵ,𝐤)\right\}G_R^0(𝐩+𝐤+𝐐_0,ϵ)$$
(21)
By substituting Eqs.(20) and (21) into Eq.(17) one obtains the following expression for $`\rho (ϵ)`$:
$$\rho (ϵ)=\rho _0^{(d)}\frac{2}{\pi }\mathrm{Im}\underset{n=1}{\overset{\mathrm{}}{}}A_n\alpha _d^n(ϵ)$$
(22)
with
$$\alpha _d(ϵ)=4\tau ^2\frac{\mathrm{d}^dk}{(2\pi )^d}\left\{C_\pi (ϵ,𝐤)\frac{2\tau }{\tau _\pi }\left(1\frac{\tau }{\tau _\pi }\right)D_\pi (ϵ,𝐤)\right\}$$
(23)
and
$$A_n=\frac{1}{(2\tau )^{2n}}\frac{\mathrm{d}^dk}{(2\pi )^d}G_0^{n+1}(ϵ,𝐩)G_0^n(𝐩+𝐩_0,ϵ)=(1)^{n+1}2\pi i\rho _0^{(d)}\frac{n(2n1)!}{2^{2n}(n!)^2}$$
(24)
Here, $`\rho _0^{(d)}`$ is the DOS for a $`d`$–dimensional clean crystal and is given by Eqs.(15) and (16) for $`d=2`$ and $`d=3`$, respectively. The sum over $`n`$ in Eq.(22) is performed easily using Eq.(24):
$$\rho (ϵ)=\rho _0^{(d)}\left\{1\mathrm{Re}\frac{\alpha _d(ϵ)}{\sqrt{1+\alpha _d(ϵ)}(1+\sqrt{1+\alpha _d(ϵ)})}\right\}=\rho _0^{(d)}\mathrm{Re}\frac{1}{\sqrt{1+\alpha _d(ϵ)}}$$
(25)
where the energy dependence of $`\alpha _d(ϵ)`$ is obtained from Eqs.(23), (18) and (19):
$$\alpha _d(ϵ)=\{\begin{array}{cc}\frac{1}{2\pi ϵ_F\tau _\pi }\left(1\frac{2\tau }{\tau \pi }+\frac{2\tau ^2}{\tau _\pi ^2}\right)\mathrm{ln}\frac{2\frac{\tau ^2}{\tau _\pi ^2}}{1\frac{\tau ^2}{\tau _\pi ^2}4i|ϵ|\tau }\hfill & \text{for }d=2\hfill \\ \begin{array}{c}\hfill \frac{3\left(1\frac{2\tau }{\tau \pi }+\frac{2\tau ^2}{\tau _\pi ^2}\right)}{4\pi \tau \tau _\pi ϵ_F^2}\{1\frac{\pi \sqrt{3}}{4}[\sqrt{\sqrt{(1\frac{\tau ^2}{\tau _\pi ^2})^2+(4\tau |ϵ|)^2}1+\frac{\tau ^2}{\tau _\pi ^2}}\\ \hfill i\sqrt{\sqrt{(1\frac{\tau ^2}{\tau _\pi ^2})^2+(4\tau |ϵ|)^2}+1\frac{\tau ^2}{\tau _\pi ^2}}]\}\end{array}\hfill & \text{for }d=3\hfill \end{array}$$
(26)
It can be seen from Eqs.(25) and (26) that away from half–filling, the effects of Umklapp scattering are weakened and $`\tau _\pi \tau _N\tau `$, as a result of which the quantum corrections to the DOS disappear. In the vicinity of half–filling, $`\tau \tau _\pi <\tau _N`$ and impurity effects become essential.
The DOS for a 2D system with half–filled energy band can be expressed as
$$\rho ^{(2)}(ϵ)=\rho _0^{(2)}(ϵ)\frac{1}{\sqrt{1+\frac{1}{2\pi ϵ_F\tau _\pi }\mathrm{ln}\frac{1}{4|ϵ|\tau _\pi }}}$$
(27)
where $`\rho _0^{(2)}(ϵ)`$, given by Eq.(15), contains a logarithmic singularity in the middle of the band. Eq.(27) shows that the van Hove singularity in the DOS of a pure 2D square crystal is preserved in the presence of substitutional impurities. However, the central peak becomes narrower due to impurity scattering than that of the van Hove peak in clean systems. In the vicinity of the band center the energy dependence of $`\rho ^{(2)}(ϵ)`$ is changed from logarithmic dependence to the square root of the logarithm,
$$\rho ^{(2)}(ϵ)=\frac{2}{(\pi a)^2ϵ_F}(2\pi ϵ_F\tau _\pi )^{1/2}\mathrm{ln}^{1/2}(\frac{1}{4\tau _\pi |ϵ|})\text{as}|ϵ|0$$
(28)
In Fig.6a the dependence of $`\rho ^{(2)}(ϵ)`$ on the impurity potential strength is drawn, which displays a narrowing of the central peak with increasing disorder strength or decreasing $`\tau _\pi `$. The DOS of a square lattice with off–diagonal disorder computed in \[\] shows the same effect. Notice that the impurity potential studied in our problem corresponds to off–diagonal disorder, since the part of the Hamiltonian Eq.(3) corresponding to the impurity potential can be decoupled into diagonal ($`U(\mathrm{𝟎})`$) and off–diagonal ($`U(𝐪)`$ with $`𝐪0`$) parts. The diagonal part is chosen to be equal to zero for the white–noise potential in the averaging process.
Unlike 2D systems, substitutional impurities have small effect on the DOS of 3D simple cubic crystals. In the close vicinity of half–filling, Eq.(23) for $`\alpha _3(ϵ)`$ is simplified:
$$\alpha _3(ϵ)=\frac{3}{4\pi \tau _\pi ^2ϵ_F^2}\left\{1\frac{\pi \sqrt{3}}{2}\sqrt{|ϵ|\tau }(1i)\right\}\alpha _R+i\alpha _I$$
(29)
By using Eq.(29), on expresses $`\rho ^{(3)}(ϵ)`$ as
$$\rho ^{(3)}(ϵ)=\frac{\rho _0^{(3)}}{\sqrt{2}}\frac{\sqrt{1+\alpha _R+\sqrt{(1+\alpha _R)^2+\alpha _I^2}}}{\sqrt{(1+\alpha _R)^2+\alpha _I^2}}\frac{\rho _0^{(3)}}{\sqrt{1+\frac{3}{4\pi ϵ_F^2\tau _\pi ^2}\left(1\frac{\pi \sqrt{3|ϵ|\tau }}{2}\right)}}$$
(30)
where $`\alpha _R(ϵ)`$ and $`\alpha _I(ϵ)`$ are real and imaginary parts of $`\alpha _3(ϵ)`$, respectively.
Eq.(30) shows that a small dip of the DOS is formed at the middle of a half–filled band of a 3D simple cubic lattice due to substitutional impurities (see Fig.6b). The depth of this dip increases with disorder strength as
$$\rho _0^{(3)}(0)=\frac{\rho _0^{(3)}}{\sqrt{1+\frac{3}{4\pi ϵ_F^2\tau _\pi ^2}}}$$
(31)
The results obtained for the DOS of 2D systems, Eqs.(25)-(28), and 3D systems, Eqs.(29)-(31), show that in contrast to 1D systems substitutional impurities tend to reduce the DOS on the Fermi surface of 2D and 3D crystals at half filling. Indeed, the Dyson singularity, expressing the enhancement of the states density in the 1D lattice with off–diagonal disorder, is mediated by impurities, since the DOS of a regular 1D lattice is a smooth function of the energy of a half–filled band.
The mechanism of the relative reduction of the DOS in 2D \[Eq.(27)\] and 3D \[Eq.(30)\] systems is Umklapp scattering, the same that increases the 1D DOS. However, the nested Fermi surface of a 2D square crystal with nearest–neighbor hopping contains also saddle points at $`\{\pm \frac{\pi }{a},0\}`$ and $`\{0,\pm \frac{\pi }{a}\}`$ which cause the van Hove singularity in the DOS. Umklapp scattering in this case weakens the central van Hove peak in the DOS, nevertheless it could not damp the peak or reverse its sign. We understand the different effects of substitutional impurities on the DOS of 1D and higher dimensional systems in the following way: The main mechanism of localization in a 1D disordered system is backward scattering on impurities. In the process of scattering on an impurity, simultaneous Bragg reflection at half–filling reverses backward scattering to forward scattering and vice versa. Therefore for vanishingly small forward scattering amplitude Bragg reflection prevents localization due to impurity scattering. However, the situation is different for 2D and 3D systems. According to the intuitive discussion given by Bergmann, localization in 2D systems is due to interference of the electronic wave, returning to the starting point after multiple scattering on impurities with only a small change of the momentum at each act of scattering, with the wave on the time–reversed path (see also\[\]). In this case, the Bragg reflection accompanying the impurity scattering can not destroy the picture of interference and as a result can not completely delocalize all states.
The addition of next–nearest–neighbor hopping terms into the model with strength $`t^{}`$ splits saddle points from the nested Fermi surface. For the energy spectrum $`ϵ(𝐤)=t[\mathrm{cos}p_xa+\mathrm{cos}p_ya\frac{t^{}}{t}\mathrm{cos}p_xa\mathrm{cos}p_ya+\frac{\mu }{2}]`$ the saddle points again lie at $`\{\pm \frac{\pi }{a},0\}`$ and $`\{0,\pm \frac{\pi }{a}\}`$. However, the optimal nested Fermi surface is realized at $`t^{}/t=0.165`$ and $`\mu =0.56`$, with the new nesting vector $`𝐐_0^{}=0.91𝐐_0`$. Our method can be applied also for this dispersion band. The singular blocks are again calculated after separating the large momentum transfer $`𝐐_0^{}`$ and linearizing the energy spectrum around the Fermi surface. As a result the same expression (28)) for the 2D DOS is obtained. However, $`\rho _0^{(2)}`$ in this case has no singularity on the Fermi surface at half–filling. Therefore $`\rho ^{(2)}(ϵ)`$ decreases with the energy around the Fermi surface for half filling and vanishes on it.
## IV Conductivity
The DC-conductivity for the 2D Anderson model at zero temperature has been computed in \[\] and the behavior of $`\sigma `$ as a function of the Fermi energy and the disorder has been studied. However, these numerical results provide limited insight into the physical origin of the processes giving contributions to the conductivity. In this section, we study low temperature quantum contributions to the conductivity for the model under consideration. Maximally crossed or “fan” diagrams are responsible for the quantum interference corrections to the conductivity in the weak localization theory. Interference between electronic wave functions in the process of multiple scattering on randomly distributed impurities changes the mobility of an electronic system and “fan” diagrams give a contribution to the diffusion coefficient. The conductivity of simple crystalline systems with substitutional impurities can also be affected through changes in the DOS due to Bragg reflection in the scattering processes on the impurities which becomes essential for an electronic band close to half–filling. The diagrams which describe first order weak localization corrections to the conductivity and represent both the diffusion coefficient and the DOS contributions, are drawn in Fig.7.
Since interference effects are essential for low–dimensional systems, we here will calculate the quantum corrections to the conductivity only for a 2D square lattice with substitutional impurities. The localization problem for a 1D lattice with off-diagonal impurities has been calculated in \[\].
The quantum correction to the conductivity is calculated according to the Kubo expression
$$\sigma _{\alpha ,\beta }(\omega )=i\frac{Ne^2}{m\omega }\delta _{\alpha \beta }+\frac{2e^2}{\omega }\frac{\mathrm{d}ϵ}{2\pi }\frac{\mathrm{d}^2p}{(2\pi )^2}v_\alpha (𝐩)v_\beta (𝐩)G(𝐩,ϵ+\omega )G(𝐩,ϵ)$$
(32)
where $`G(𝐩,ϵ)`$ is the Green’s function and $`v_\alpha (𝐩)=\frac{ϵ}{p_\alpha }`$ with $`\alpha =(x,y)`$ is a component of the electron velocity; the bracket $`\mathrm{}`$ denotes averaging over the impurity realizations.
Far from the half–filled energy band, maximally crossed diagrams with normal scattering are responsible for logarithmic corrections to the conductivity of the 2D weakly disordered non–interacting electron gas. These diagrams can be redrawn as a ladder series in the particle–particle channel as shown in Fig.7a.
Bragg reflection is intensified as half–filling is approached. In this case, an act of electron scattering on an impurity may be accompanied by Bragg reflection on the Brillouin zone boundary. The new diagrams which give a contribution to the conductivity as half–filling is approached are drawn in Figs.7b-d’.
The contribution of the diagrams in Fig.6a to the conductivity, $`\delta \sigma _a(\omega )`$ can be expressed as follows according to the Kubo expression Eq.(32):
$$\delta \sigma _a(\omega )=\frac{2e^2}{\omega }\frac{\mathrm{d}ϵ}{2\pi }\frac{\mathrm{d}^2p}{(2\pi )^2}C_N(𝐪,ϵ,\omega )A_\alpha ^N(𝐪,ϵ,\omega )$$
(33)
where $`C_N((𝐪,ϵ,\omega )`$ is the cooperon block Eq.(8) for normal scattering and
$$A_\alpha ^N(𝐪,ϵ,\omega )=\frac{\mathrm{d}^2p}{(2\pi )^2}v_\alpha (𝐩)v_\alpha (𝐩+𝐪)G(𝐩,ϵ+\omega )G(𝐩+𝐪,ϵ+\omega )G(𝐩,ϵ)G(𝐩+𝐪,ϵ)$$
(34)
For a tight–binding model with nearest–neighbor hopping, the $`\alpha =(x,y)`$ component of the velocity $`v_\alpha (𝐩)`$ is given as $`v_\alpha (𝐩)=ta\mathrm{sin}(p_\alpha a)`$. The integration over $`𝐩`$ can be written as $`\frac{\mathrm{d}𝐒}{(2\pi )^2|𝐯_p|}d\xi `$, and then the integral over the energy variable $`\xi `$ is done easily. The van Hove singularity, arising at saddle points of the Fermi surface for the half–filled band when $`|𝐯_𝐩|=ta\sqrt{\mathrm{sin}^2p_xa+\mathrm{sin}^2p_ya}`$ vanishes, is removed due to the $`v_\alpha ^2(𝐩)`$ term under the integral over the Fermi surface $`\mathrm{d}𝐒`$. Therefore the value of $`A_\alpha (q,ϵ,\omega )`$ does not strongly differ from that for the free–electron gas model and we get
$$A_\alpha (𝐪,ϵ,\omega )=p_F\tau ^2l\mathrm{\Theta }(ϵ(ϵ+\omega ))$$
(35)
Using Eqs.(8) and (35) for $`C_N(𝐪,ϵ,\omega )`$ and $`A_\alpha (𝐪,ϵ,\omega )`$ in Eq.(33) one gets
$$\delta \sigma _a(\omega )=\frac{e^2\tau }{2\pi ^2\tau _N}\mathrm{ln}\left(\frac{1}{\frac{\tau }{\tau _\pi }i|\omega |\tau |}\right)$$
(36)
Far from half–filling and far from other commensurate points the Bragg reflection is weakened and $`\tau \tau _N\tau _\pi `$. Therefore $`\delta \sigma _a(\omega )`$ in this case represents the conventional weak localization correction to the conductivity. $`\delta \sigma _a(\omega )`$ is reduced close to half–filling when umklapp scatterings are enhanced and $`\tau \tau _\pi \tau _N`$.
The diagrams which give an essential contribution to the conductivity at half–filling are shown in Fig.7b-7d’. The appearance of these new diagrams is due to intensified scattering in the $`\pi `$–channel with large momentum transfer. The main quantum correction to the diffusion coefficient at half–filling comes from the “fan” diagrams in Fig.7b with the $`\pi `$–cooperon block. The expression corresponding to this diagram is
$$\delta \sigma _b(\omega )=\frac{2e^2}{\omega }\frac{\mathrm{d}ϵ}{2\pi }\frac{\mathrm{d}^2q}{(2\pi )^2}C_\pi (𝐪,ϵ,\omega )A_\pi (𝐪,ϵ,\omega )$$
(37)
where $`C_\pi (𝐪,ϵ,\omega )`$ is the $`\pi `$–cooperon given by Eq.(12) and
$$\begin{array}{cc}\hfill A_\pi (𝐪,ϵ,\omega )=& \frac{\mathrm{d}^2p}{(2\pi )^2}v_\alpha (𝐩)v_\alpha (𝐩+𝐪+𝐐_0)G(𝐩,ϵ+\omega )G(𝐩,ϵ)\hfill \\ & G(𝐩+𝐪+𝐐_0,ϵ)G(𝐩+𝐪+𝐐_0,ϵ+\omega )\hfill \end{array}$$
(38)
The calculation of Eq.(38) is similar to that for $`A_N(𝐪,ϵ,\omega )`$. Taking into account the condition that $`v_\alpha (𝐤+𝐐_0)=v_\alpha (𝐤)`$ we obtain the result for $`A_\pi (𝐪,ϵ,\omega )`$
$$A_\pi (𝐪,ϵ,\omega )=\frac{p_Fl\tau ^2}{(12i\tau |ϵ|)(12i\tau |ϵ+\omega |)}\left\{\frac{1}{1i\tau |2ϵ+\omega |}\mathrm{\Theta }(ϵ(ϵ+\omega ))+\mathrm{\Theta }(ϵ(ϵ+\omega ))\right\}$$
(39)
Substituting Eqs.(12) and (39) in Eq.(37), the expression for $`\delta \sigma _b(\omega )`$ can be reduced by some simple calculations to
$$\delta \sigma _b(\omega )=\frac{e^2\tau }{2\pi ^2\omega \tau _\pi }\left\{\left(\frac{\tau }{\tau _\pi }\right)^2I(\frac{\tau }{\tau _\pi },\omega \tau )+\frac{\omega }{2}\right\}$$
(40)
with
$$\begin{array}{cc}\hfill I& (\frac{\tau }{\tau _\pi },\overline{\omega }=\omega \tau )=\frac{i}{2\tau }_0^1dx_0^{\mathrm{}}dz\frac{1}{(z+i)^3}\frac{1}{(z+\overline{\omega }+i)^2x+(\frac{\tau }{\tau _\pi })^2}\hfill \\ \hfill =& \frac{i}{4\tau }\{2i\overline{\omega }[\frac{1}{\overline{\omega }^21+(\frac{\tau }{\tau _\pi })^2}\frac{1}{\overline{\omega }^2+(\frac{\tau }{\tau _\pi })^2}]+[1+\frac{1}{(\overline{\omega }\sqrt{1(\frac{\tau }{\tau _\pi })^2})^2}]\mathrm{ln}(1i\overline{\omega }+i\sqrt{1(\frac{\tau }{\tau _\pi })^2})\hfill \\ & +\left[1+\frac{1}{(\overline{\omega }+\sqrt{1(\frac{\tau }{\tau _\pi })^2})^2}\right]\mathrm{ln}\left(1i\overline{\omega }i\sqrt{1(\frac{\tau }{\tau _\pi })^2}\right)+\left[\frac{1}{(\frac{\tau }{\tau _\pi }i\overline{\omega })^2}1\right]\mathrm{ln}\left(1i\overline{\omega }+\frac{\tau }{\tau _\pi }\right)\hfill \\ & +[\frac{1}{(\frac{\tau }{\tau _\pi }+i\overline{\omega })^2}1]\mathrm{ln}(1i\overline{\omega }\frac{\tau }{\tau _\pi })\}\hfill \end{array}$$
(41)
The expression for $`\delta \sigma _b(\omega )`$ can be presented for two limiting cases, namely near the middle of the band when $`\tau \tau _\pi \tau _N`$ and far from half–filling when $`\tau \tau _N\tau _\pi `$ by using Eq.(41) in (40):
$$\delta \sigma _b(\omega )=\{\begin{array}{cc}\frac{e^2}{4\pi ^2}\mathrm{ln}(1\frac{\tau }{\tau _\pi }i\omega \tau )+i\frac{e^2}{8\pi ^2\omega \tau _\pi }\hfill & \text{for}\frac{\tau }{\tau _\pi }1\hfill \\ \mathrm{const}+i\frac{e^2}{4\pi ^2\omega \tau _\pi }\left(\frac{\tau }{\tau _\pi }\right)^2\mathrm{ln}2\hfill & \text{for}\frac{\tau }{\tau _\pi }\frac{\tau _N}{\tau _\pi }0\hfill \end{array}$$
(42)
So the quantum correction to the conductivity corresponding to the diagram in Fig.7b decreases $`\sigma (\omega )`$ logarithmically with the external frequency near half–filling. The contribution of Fig.7b vanishes through a small offset from the middle of the band.
Analyzing the effect of substitutional impurities on the conductivity by means of the diffusion coefficient, which is expressed by the diagrams a) and b) in Fig.7, it can be seen that the logarithmic correction to the localization correction to $`\sigma (\omega )`$ is preserved irrespectively of the band filling. However, the coefficient of the logarithm of the quantum correction to the conductivity is changed from $`\frac{e^2}{2\pi ^2}`$ in Eq.(36) far from half–filling to $`\frac{e^2}{4\pi ^2}`$ in Eq.(42) close to half–filling. On the other hand, the $`\pi `$–scattering mechanism gives a perturbative contribution to the dielectric constant $`ϵ^{}(\omega )\mathrm{Im}\sigma (\omega )`$. The contribution to $`ϵ^{}(\omega )`$ increases close to half–filling. This also means that there exist a few delocalized states at the center of the band.
The diagrams in Fig.7c-d’ represent the quantum correction to the conductivity due to changes in the DOS. The expression corresponding to the sum of these diagrams can be presented as
$$2(\delta \sigma _c(\omega )+\delta \sigma _d(\omega ))=\frac{2e^2}{\omega }\frac{\mathrm{d}ϵ}{2\pi }\frac{\mathrm{d}^2q}{(2\pi )^2}\alpha (𝐪,ϵ+\omega )B(𝐪,ϵ,\omega )$$
(43)
where
$$\alpha (𝐪,ϵ)=C_\pi (𝐪,ϵ)\frac{2\tau }{\tau _\pi }(1\frac{\tau }{\tau _\pi })D_\pi (𝐪,ϵ)$$
(44)
and
$$B(𝐪,ϵ,\omega )=\frac{p_Fl\tau ^2}{(12i\tau |ϵ+\omega |)^2}\left\{(\frac{\tau }{\tau _N}2)\mathrm{\Theta }(ϵ(ϵ+\omega ))+\frac{1}{1i\tau |ϵ+\omega |i\tau |ϵ|}\mathrm{\Theta }(ϵ(ϵ+\omega ))\right\}$$
(45)
The fact that the contributions from diagrams c’ and d’ in Fig.7 are equal to those from c and d is taken into account in Eq.(43). The cooperon and diffusion insertions into the Green’s function are given by Eqs.(18) and (19), respectively. After some routine calculations, Eq.(43) for the correction to the conductivity due to the DOS is reduced to the form
$$\begin{array}{cc}\hfill 2(\delta \sigma _c(\omega )+& \delta \sigma _d(\omega ))=\frac{e^2\tau ^3}{2\pi ^2\omega \tau _\pi ^3}[1\frac{2\tau }{\tau _\pi }(1\frac{\tau }{\tau _\pi })\left]\right\{I(\frac{\tau }{\tau _\pi },\overline{\omega }=\omega \tau )\hfill \\ & +\frac{\omega }{2}(\frac{\tau }{\tau _N}3)\mathrm{ln}\frac{1(\frac{\tau }{\tau _\pi }))^2+(12i\omega \tau )^2}{(12i\omega \tau )^2+(\frac{\tau }{\tau _\pi })^2}\}\hfill \end{array}$$
(46)
Where $`I(\frac{\tau }{\tau _\pi },\overline{\omega }=\omega \tau )`$ is given by Eq.(41). These contributions vanish as expected far from half–filled band, when $`\tau _\pi \tau \tau _N`$. However, they give a large contribution to the conductivity near half–filling, $`\tau \tau _\pi \tau _N`$, giving rise to a rapid increasing of the conductivity with the external frequency. So, the total contribution to the conductivity is obtained by summing Eqs.(36),(42), and (46). Far from half–filling, only the normal cooperon gives a contribution and, as a result $`\delta \sigma _a(\omega )`$ survives:
$$\delta \sigma (\omega )=\frac{e^2}{2\pi ^2}\mathrm{ln}(\frac{1}{i|\omega |\tau })$$
(47)
The quantum correction to the conductivity near half–filling is due to the $`\pi `$–cooperon and
$$\delta \sigma (\omega )=\frac{5e^2}{2\pi ^2}\mathrm{ln}(\frac{1}{i|\omega |\tau _\pi })$$
(48)
with $`\tau \tau _\pi `$. Although the conductivity decreases with the external frequency, the diffusion coefficient increases with approaching half–filling for given $`\omega `$, which means a partial lifting of localization at the center of the band due to a few delocalized states. The rapid decrease of the conductivity (Eq.(48)) near half–filling is the result of the impurity effect on the DOS.
## V Conclusion
The character of localization in the 2D disordered electronic gas has been a subject of debate for a long time. The question whether delocalized states exist in the center of the band of 2D disordered systems or not is also one of the crucial points for the integer quantum Hall effect.
Searches for delocalized states have been mainly concentrated around the 2D Anderson model. Expectations are connected with the existence of the van Hove singularity in the band center of the square lattice, which might give rise to a delocalization of states at half–filling. Various numerical approaches have been applied to compute the DOS, the localization length, and the conductivity of the 2D Anderson model with diagonal and off–diagonal disorder. The computations show that although even a small concentration of diagonal disorder suppresses the van Hove singularity, off–diagonal disorder preserves it, however with a modified shape.
The exponential localization of all states for diagonal disorder has been revealed by computational approaches whereas the existence of quasi–localized states has been predicted for the 2D Anderson model with off–diagonal disorder. In other words, the computation of the effects of off–diagonal disorder provides evidence for the transformation from exponential to power–like localization as half–filling is approached.
In this paper we tried to give a complete picture for the DOS and the conductivity, and also to give some physical insight into the processes occuring near half–filling in simple lattices with substitutional impurities.
The impurity concentration is considered to be small, so that the condition $`ϵ_F\tau 1`$ or $`p_Fl1`$ is satisfied and the weak localization approximation is applicable.
Apart from normal scattering with small momentum transfer, umklapp scattering with large momentum transfer is shown to be essential near half–filling. This new scattering involves coherent reflection of an electron on the boundary of the Brillouin zone for each act of scattering.
Our diagrammatical approach shows that the new singular blocks, namely the $`\pi `$–cooperon $`C_\pi (𝐪,ϵ,\omega )`$ and the $`\pi `$–diffuson $`D_\pi (𝐤,ϵ,\omega )`$, give a large contribution to the DOS and the conductivity at half–filling. The dependence of the DOS on the energy and the impurity strength near the middle of the band have been determined by summing the leading logarithmically divergent contributions. The results obtained \[Eqs.(27) and (30)\] for the DOS show that weak disorder due to substitutional impurities does not remove the van Hove singularity at the center of the 2D band. However its energy dependence is strongly changed. The impurity effect on the 3D DOS is a shallow dip on the smooth background of the bare DOS in the middle of the band.
The effect might be observable in the temperature dependence of the magnetic susceptibility according to
$$\chi _{(d)}(T)=2\mu _B^2\frac{\mathrm{d}ϵ}{4T}\mathrm{cosh}^2(\frac{ϵ}{2T})\rho ^{(d)}(ϵ)$$
(49)
Where $`\mu _B`$ is the Bohr magneton. Straightforward calculation using Eq.(27) for the 2D DOS gives
$$\chi _{2D}(T)\mathrm{ln}^{1/2}(\frac{1}{T\tau _\pi })$$
For $`d=3`$, the correction to the magnetic susceptibility can again be calculated according to Eqs.(49) and (30). However, this correction is negligibly small.
The first logarithmic corrections to the conductivity of the 2D square lattice with substitutional impurities come from both the diffusion coefficient and the DOS. The corrections due to the diffusion coefficient, $`\delta \sigma _a(\omega )+\delta \sigma _b(\omega )`$, given by Eqs.(36) and (42), decrease in magnitude as half–filling is approached while they remain logarithmically dependent on the external frequency. Such a partial lifting of localization may be due to a few delocalized states in the center of the band. The imaginary part of $`\sigma (\omega )`$, which corresponds to the dielectric constant, increases close to half–filling. This fact supports the picture of an increase of the electronic mobility at the center of the band.
Nevertheless, the contribution to $`\sigma (\omega )`$ due to the DOS suppresses the relative increase in the conductivity coming from the diffusion coefficient.
|
warning/0004/hep-ph0004154.html
|
ar5iv
|
text
|
# Phenomenology of Neutrino Oscillations *footnote **footnote *Plenary talk given at WHEPP-6, 3-16 January 2000, Chennai, India.
## I Introduction
In this talk, I shall try to give a bird’s eye view of the current status of neutrino oscillations. Also I shall highlight local work wherever possible since phenomenological contributions have been made from the Institute of Mathematical Sciences on various aspects of neutrino oscillations.
Solar neutrino physics is 30-years old. Starting from the pioneering $`Cl`$ experiments of Davis and collaborators , all the experiments have observed the depletion of the solar neutrinos as compared to the theoretically calculated flux from the standard solar model . Atmospheric neutrino physics is much younger. The observation of the reduction in the $`\nu _\mu /\nu _e`$ ratio is only 10-years old. Nevertheless, atmospheric neutrino physics has won over solar neutrino physics, since the evidence for neutrino oscillation from the former is derived from ratios and is hence independent of the absolute flux. On the whole, very good evidence for oscillation exists from both fronts. There are still loose ends ; for instance, the recoil energy spectrum of the solar neutrinos does not agree with the theoretically calculated one, even with oscillations.
Alternative ideas must still be explored and ruled out, before neutrino oscillations are finally established as a part of Physics. Nevertheless it is fair to state that oscillations are the natural explanation for the observed neutrino anomalies. For, oscillation is the most conservative explanation for the anomalies. It does not violate any known principle of physics and it does not invoke any exotic new physics. It only uses ordinary principles of quantum mechanics, especially the principle of superposition. In quantum mechanics, the neutrino states will mix and oscillate, if they have masses.
Hence, I think it is time to look at the next job. Attention must now shift from the oscillation itself, to determining the 6 fundamental parameters (2 mass differences, 3 mixing angles and one CP violating phase) describing the oscillation phenomena among the 3 flavours $`\nu _e,\nu _\mu `$ and $`\nu _\tau `$. Even after 30 years, none of these parameters is yet determined with any certainty. But, progress has recently been made in limiting their ranges or determining their orders of magnitude. One parameter, namely the mixing angle $`\varphi `$, has been bracketed rather closely.
It is important to stress the desirability of explaining neutrino anomalies and fitting the data using a three-neutrino framework. For, there exist three neutrinos. The high-quality data coming out from the great neutrino experiments of the present-day deserve to be treated by a realistic $`3\nu `$ analysis rather than the $`2\nu `$ toy model often used by the experimenters to give their results. This is the point of view with which we started our $`\nu `$-phenomenology work at IMSc . When we started, there were very few groups doing a $`3\nu `$ analysis.
Having said the above, I have also to point out that Nature has helped the (lazy) $`2\nu `$ people. First, it chose the mass hierarchy :
$$m_2^2m_1^2<<m_3^2m_2^2$$
(1)
so that solar $`\nu `$ could be explained by $`\delta m_{21}^2(m_2^2m_1^2)`$ and atmospheric $`\nu `$ could be explained by $`\delta m_{32}^2(m_3^2m_2^2)`$. Next, the CHOOZ reactor experiment showed that one of the mixing angles $`(\varphi )`$ is so small that the $`3\nu `$ problem decouples into two effective $`2\nu `$ problems.
Although neutrino oscillation is a direct consequence of quantum mechanics, it leads to a result of profound consequence for Physics and Astrophysics – the result that neutrinos have mass. That is the importance of the whole subject of neutrino oscillations. Neutrino mass is the only concrete evidence we have for physics beyond the Standard Model (SM) of high energy physics. Every other physics beyond the SM, that is being searched for, has remained speculative so far. Every other experimental signal for physics beyond the SM that appears now and then on the horizon has been disappearing in about 6 months to one year. Neutrino mass is the only evidence for physics beyond SM that has remained robust for the past 30 years.
## II Results
The neutrino flavour states $`|\nu _\alpha (\alpha =e,\mu ,\tau )`$ are linear superpositions of the neutrino mass eigenstates $`|\nu _i(i=1,2,3)`$ with masses $`m_i`$ : $`|\nu _\alpha =\mathrm{\Sigma }_iU_{\alpha i}|\nu _i`$ where $`U`$ is the $`3\times 3`$ unitary matrix :
$$U=\left(\begin{array}{ccc}c_\varphi c_\omega & c_\varphi s_\omega & s_\varphi \\ c_\psi s_\omega s_\psi s_\varphi c_\omega e^{i\delta }& c_\psi c_\omega s_\psi s_\varphi s_\omega e^{i\delta }& s_\psi c_\varphi e^{i\delta }\\ s_\psi s_\omega c_\psi s_\varphi c_\omega e^{i\delta }& s_\psi c_\omega c_\psi s_\varphi s_\omega e^{i\delta }& c_\psi c_\varphi e^{i\delta }\end{array}\right)$$
(2)
where $`c`$ and $`s`$ stand for sine and cosine of the angle appearing as subscript.
From the oscillation phenomena, one has to determine the 6 parameters $`\delta m_{21}^2,\delta m_{32}^2,\omega ,\psi ,\varphi `$ and $`\delta `$. Most of the presently studied oscillation phenomena are insensitive to the CP-violation parameter $`\delta `$. So let us ignore $`\delta `$ but we shall justify it at the end of the section.
Under the hiearchy assumption of Eq.(1), one can show that the solar neutrino problem depends only on $`\delta m_{21}^2,\omega `$ and $`\varphi `$ while the atmospheric neutrino problem depends only on $`\delta m_{32}^2,\psi `$ and $`\varphi `$. It is this simplification that allows us to analyse the two problems within a $`3\nu `$ framework under reasonable control and one gets a fairly broad and stable set of allowed regions in the 5-parameter space . If we temporarily put $`\varphi `$ as zero, then the two problems decouple and the results are the following :-
There are three solutions of the solar neutrino problem :
1. MSW-small angle : $`\delta m_{21}^210^5eV^2,sin^22\omega 10^3`$
2. MSW-large angle : $`\delta m_{21}^210^5eV^2,sin^22\omega 1`$
3. Vacuum oscillations : $`\delta m_{21}^210^{10}eV^2,sin^22\omega 1`$
Atmospheric neutrinos problem has the solution : $`\delta m_{32}^210^3eV^2,sin^22\psi 1`$
The nonobservation of any depletion of $`\overline{\nu }_e`$ in the reactor experiment at CHOOZ turns out to be a crucial result. Interpreted in the $`3\nu `$ framework , with the hieararchy assumption of Eq.1 it implies $`\varphi <9^o`$, which is a very powerful constraint and this result justifies the neglect of $`\varphi `$ in the analysis of solar and atmospheric $`\nu `$. This constraint is independent of CP violation.
One can show that CP violation always occurs in the combination $`\mathrm{sin}\varphi e^{\pm i\delta }`$. So, in view of the CHOOZ result, CP violation is suppressed in all neutrino oscillation phenomena.
In the next few sections, we shall briefly describe how these results were obtained.
## III Oscillations in vacuum and matter
The probability of a neutrino of flavour $`\alpha `$ to be observed with flavour $`\beta `$ after a distance of travel $`L`$ in vacuum is given by
$$P(\nu _\alpha \nu _\beta )=\delta _{\alpha \beta }4\underset{i}{}\underset{j>i}{}U_{\alpha i}U_{\beta i}U_{\alpha j}U_{\beta j}\mathrm{sin}^2\left[\frac{1.27\delta m_{ij}^2L}{E}\right]$$
(3)
where $`U`$ is defined in Eq.(2), $`\delta m_{ij}^2`$ is in $`eV^2`$, $`L`$ is in metres, neutrino energy $`E`$ is in $`MeV`$ and $`CP`$ violation is ignored.
In matter (especially of varying density), the above formulae are drastically changed because of the famous Mikheyev-Smirnov-Wolfenstein (MSW) effect. We consider the propagation of the neutrinos through solar matter. Let a neutrino of flavour $`\alpha `$ be produced at time $`t=t_o`$ in the solar core. Its state vector is $`|\mathrm{\Psi }_\alpha (t_o)=|\nu _\alpha =_iU_{\alpha i}^c|\nu _i^c`$ where $`|\nu _i^c`$ are the mass eigenstates with mass eigenvalues $`m_i^c`$ and mixing matrix elements $`U_{\alpha i}^c`$ in the core of the sun. The neutrino propagates in the sun adiabatically upto $`t_R`$ (the resonance point), makes nonadiabatic Landau-Zener transition $`ij`$ at $`t_R`$ with probability amplitude $`M_{ji}^{LZ}`$, propagates adiabatically upto $`t_1`$ (the edge of the sun) and propagates as a free particle upto $`t_2`$ when it reaches the earth.
The state vector at $`t_2`$ is
$$|\mathrm{\Psi }_\alpha (t_2)=\underset{i,j}{}|\nu _jM_{ji}^{LZ}U_{\alpha i}^cexp\left(i_{t_R}^{t_2}ϵ_j(t)𝑑ti_{t_o}^{t_R}ϵ_i(t)𝑑t\right)$$
(4)
where $`ϵ_i(t)\left\{=E+m_i^2(t)/2E\right\}`$ are the matter-dependent energy eigenvalues in the sun upto $`t_1`$ and vaccuum eigenvalues for $`t_1<t<t_2`$. The probability for detecting a neutrino of flavour $`\beta `$ on the earth is
$`|\nu _\beta |\mathrm{\Psi }_\alpha (t_2)|^2`$ $`=`$ $`{\displaystyle \underset{iji^{}j^{}}{}}U_{\beta j}^{}U_{\beta j^{}}M_{ji}^{LZ}M_{j^{}i^{}}^{LZ^{}}U_{\alpha i}^cU_{\alpha i^{}}^c^{}`$ (6)
$`exp\{i{\displaystyle _{t_R}^{t_2}}(ϵ_jϵ_j^{})dti{\displaystyle _{t_o}^{t_R}}(ϵ_iϵ_i^{}dt\}`$
Next comes the crucial step of averaging over $`t_0`$ and $`t_2`$ and the assumption that the oscillations are rapid enough so that the averaged exponential in this equation can be replaced by $`\delta _{ii^{}}\delta _{jj^{}}`$. Calling this averaged probability as $`P_{\alpha \beta }^D`$ (The probability for a $`\nu _\alpha `$ produced in the sun to be detected as a $`\nu _\beta `$ on the earth at daytime), we get
$$P_{\alpha \beta }^D=\underset{ij}{}|U_{\beta j}|^2|M_{ji}^{LZ}|^2|U_{\alpha i}^c|^2$$
(7)
## IV Solar, atmospheric and reactor neutrinos
Extensive literature exists on the solar and atmospheric neutrinos . So, we shall be brief.
The experimental results on the total solar neutrino flux from the three types of detectors ($`Cl,Gl`$ and $`H_2O`$) are the following : $`R_{Cl}=0.33\pm 0.028;R_{Ga}=0.56\pm 0.05;R_{SK}=0.475\pm 0.015`$ where we have given the ratios of the experimental rates to the theoretical rates without neutrino oscillations calculated in the Standard Solar Model (SSM). One can see that the statistical uncertainty is the least for the SuperKamioka (SK) water Cerenkov detector, which is thus presaging the era of precision neutrino physics. Since the three types of detectors are sensitive to different regions of the solar neutrino spectrum, the above three numbers already contain some spectral information. Ascribing the above observed depletion factors to oscillation and using the formulae in Eq.(3) or (7), the best fits for the neutrino parameters can be obtained, with the results already discussed in Sec.2. However, the recoil electron energy spectrum as measured by SK does not fit with any oscillation scenerio, at the higher energy end where the observed number is too large (but with large errors).
Since the neutral current (NC) week interaction is flavour-blind, the flux of solar $`\nu `$ detected through NC node would be the total $`(\nu _e+\nu _\mu +\nu _\tau )`$ flux and hence is independent of oscillation and would be a test of the standard solar model. Also, the ratio CC/NC, would be a test of neutrino oscillation independent of the uncertainties of the solar models. Therefore great expetations have been raised by the SNO detector which will soon give the first results. Another exciting avenue will open when Borexino starts its operation, since it is the first detector zeroing in on the monochromatic $`\nu `$ line from the $`Be^7`$ decay in the sun.
Cosmic rays impingent on the atmosphere produce hadrons (especially pions) that decay, resulting in a wide spectrum of $`\nu _\mu ,\overline{\nu }_\mu ,\nu _e,\overline{\nu }_e`$ ranging in energy upto about 100 GeV, with the calculated ratio $`R_{cal}=(N_{\nu _\mu }+N_{\overline{\nu }_\mu })/(N_{\nu _e}+N_{\overline{\nu }_e})`$ approximately equal to $`2`$, whereas the experimentally measured ratio is nearer to unity. More precisely, $`r`$ defined as $`R_{obs}/R_{cal}`$ is about 0.6. This is the atmospheric neutrino problem whose solution is the oscillation of $`\nu _\mu (\overline{\nu }_\mu )`$ into $`\nu _\tau (\overline{\nu }_\tau )`$.
Abundant data from superkamioka detector is now available on the zenith angle dependence as well as on the energy-distribution of the $`\nu _\mu `$ and $`\nu _e`$ events. All these are consistent with $`\nu _\mu `$ oscillating into $`\nu _\tau `$ over the distance scale of about 10,000 km ; so it is for the upward-going $`\nu _\mu `$ travelling through the earth that the effect is most dominant. The effect of earth matter in the atmospheric neutrino problem is not significant at the present level of accuracy and hence one can use Eq.(3). The resulting neutrino prameters were given in Sec 2.
Although the analysis is performed in terms of ratios such as $`\nu _\mu /\nu _e`$ or up/down and hence is relatively insensitive to the rather large uncertainties that exist in the primary cosmic ray flux and spectrum, further improvement in our knowledge of the latter will be essential for the complete understanding of all atmospheric neutrino data.
Also, it is worth pointing out that, if the above explanation of atmospheric neutrino anomaly is correct, then the upward-going atmospheric neutrino beam has the approximate composition of $`\nu _e:\nu _\mu :\nu _\tau =1:1:1`$. The direct detection of this $`\nu _\tau `$ through production of $`\tau `$ will be a crucial test.
Nuclear fission reactors are powerful sources of $`\overline{\nu }_e`$. The detector was placed about 1 km away from the CHOOZ power reactor and the observed $`\overline{\nu }_e`$ flux was compared with the calculated flux : $`\varphi _{obs}/\varphi _{cal}=0.98\pm 0.04\pm 0.04`$. If $`\delta m_{32}^2>>\delta m_{21}^2`$ and $`\delta m_{21}^2\stackrel{}{<}10^5eV^2`$, the $`3\nu `$ formula of Eq.(3) reduces to the 2-$`\nu `$ formula
$$P(\overline{\nu }_e\overline{\nu }_e)=1\mathrm{sin}^22\varphi \mathrm{sin}^2\frac{1.27\delta m_{32}^2L}{E}$$
(8)
Comparison with the CHOOZ result yields $`\varphi <12^o`$ for $`\delta m_{32}^2\stackrel{>}{}10^3eV^2`$. CHOOZ have now improved their limit to $`\varphi <9^o`$.
For the first time, a negative result on neutrino oscillations from laboratory experiment has given a constraint of significance in the context of solar and atmospheric neutrinos. That is the importance of the CHOOZ experiment. Independent confirmation of this result has come from the Palo Verde experiment, although with less statistics. Hopefully the dependence on the calcualted flux $`\varphi _{cal}`$ will be removed in the future by placing another detector near the reactor.
## V Neutrinos through the earth and the moon
Neutrino oscillation is a complex phenomenon depending on many unknown parameters (six parameters for three flavours $`\nu _e,\nu _\mu `$ and $`\nu _\tau `$) and a considerable amount of experimental work and ingenuity will be required before the neutrino problem is solved.
Whenever physicists are confronted with a beam of unknown properties, they pass it through different amounts of matter. Nature has fortunately provided us with such opportunities (See Fig.1) : (a) Neutrinos produced in the solar core pass through solar matter ; (b) solar neutrinos detected at night pass through earth ; (c) solar neutrinos detected during a solar eclipse pass through the moon ; (d) solar neutrinos detected at the far side of earth during a solar eclipse pass through the moon and earth ; (e) upward going atmospheric neutrinos pass through the earth. To these we may add two more experiments of the future : (f) Long-base-line experiments of accelerator and reactor produced neutrinos and (g) detection of geophysical neutrinos .
It is possible to treat these effects analytically. The analytical formula for (a) was already given in Eq.(7). We shall now derive the formulae for (b) the night effect , (c) the eclipse effect and (d) the double eclipse effect .
### A The night effect
Starting with $`|\mathrm{\Psi }_\alpha (t_2)`$ on the surface of the earth given by Eq.(4), we multiply the righthand side by $`_k|\nu _k^E\nu _k^E|(=1)`$ where $`|\nu _k^E(k=1,2,3)`$ is the complete set of matter dependent mass eigenstates just inside the earth. If the neutrino propagates adiabatically upto $`t_3`$ on the other side fo the earth (we shall soon correct for nonadiabatic jumps during the propagation), the state vector at $`t_3`$ is
$`|\mathrm{\Psi }_\alpha (t_3)`$ $`=`$ $`{\displaystyle \underset{i,j,k}{}}|\nu _k^E\nu _k^E|\nu _jM_{ji}^{LZ}U_{\alpha i}^c`$ (10)
$`exp\left\{i{\displaystyle _{t_2}^{t_3}}ϵ_k𝑑ti{\displaystyle _{t_R}^{t_2}}ϵ_j𝑑ti{\displaystyle _{t_0}^{t_R}}ϵ_i𝑑t\right\}`$
This expression automatically contains $`\nu _k^E|\nu _j`$ which is the probability amplitude for nonadiabatic transition $`jk`$ at the vacuum-earth boundary and we shall call it $`M_{kj}^E`$ :
$$M_{kj}^E=\nu _k^E|\nu _j=\underset{\sigma }{}\nu _k^E|\nu _\sigma \nu _\sigma |\nu _j=\underset{\sigma }{}U_{\sigma k}^EU_{\sigma j}^{}$$
(11)
where $`\nu _k^E`$ and $`U^E`$ are mass eigenstates and mixing matrix just inside the earth.
Averaging the probability $`|\nu _\beta |\mathrm{\Psi }_\alpha (t_3)|^2`$ over $`t_R`$ results in the desired incoherent mixture of mass eigenstates of neutrinos reaching the surface of the earth. Calling this average probability as $`P_{\alpha \beta }^N`$ (the probability for $`\nu _\alpha `$ produced in the sun to be detected as $`\nu _\beta `$ in the earth at night), we can write the result as
$$P_{\alpha \beta }^N=\underset{j}{}P_{\alpha j}^SP_{j\beta }^E$$
(12)
where $`P_{\alpha j}^S`$ is the probability of $`\nu _\alpha `$ produced in the sun being detected as $`\nu _j`$ (mass eigenstate) as it enters the earth and $`P_{j\beta }^E`$ is the probability of $`\nu _j`$ entering the earth to be detected as $`\nu _\beta `$ after it propagates through the earth. These are given by
$$P_{\alpha j}^S=\underset{i}{}|M_{ji}^{LZ}|^2|U_{\alpha i}^c|^2$$
(13)
$$P_{j\beta }^E=\underset{k,k^{}}{}U_{\beta k}^E^{}U_{\beta k^{}}^EM_{kj}^EM_{k^{}j}^Eexp(i_{t_2}^{t_3}(ϵ_kϵ_k^{})𝑑t)$$
(14)
It is important to note that the factorization of probabilities in Eq(12) (which has been derived here as a consequence of the averaging over $`t_R`$), is valid only for mass eigenstates in the intermediate state. An equivalent statement of this result is that the density matrix is diagonal only in the mass-eigenstate representation and not in the flavour representation.
During the day, put $`t_3=t_2`$ so that $`P_{j\beta }^E`$ becomes $`|U_{\beta j}|^2`$ and so Eq.(12) reduces to Eq.(7). One can justify the averaging over $`t_0`$ and $`t_2`$ by the facts that the neutrinos are produced over an extended region in the solar core and they are detected over an extended region or time since the detector is moving with the earth. While averaging over $`t_0`$ and $`t_2`$ is equivalent to averaging over $`t_R`$ as far as $`P_{\alpha \beta }^D`$ is concerned, it is not so for $`P_{\alpha \beta }^N`$, but we have adopted the latter method for $`P_{\alpha \beta }^N`$ because of its simplicity in giving us the factored probability expression in Eq(12).
However, two points have to be made : (i) For $`P_{\alpha \beta }^N`$, it is not justified to average over $`t_2`$ or $`t_3`$ since we would like to detect the neutrinos during a narrow time-bin in the night. (ii) Averaging over $`t_R`$ (as we have done) may be partially justified since the result may be effectively the same for energy-integrated rates. However, this argument does not apply for Borexino , where the monochromatic $`Be^7`$ neutrino line spectrum will be detected.
Next we show how to take into account nonadiabatic jumps during the propagation inside the earth. Consider $`\nu `$ propagation through a series of slabs of matter, density varying inside each slab smoothly but changing abruptly at the junction between adjacent slabs. The state vector of the neutrino at the end of the $`n^{th}`$ slab $`|n`$ is related to that at the end of $`(n1)`$th slab $`|n1`$ by $`|n=F^{(n)}M^{(n)}|n1`$ where $`M^{(n)}`$ describes the nonadiabatic jump occuring at the junction between $`(n1)`$th and $`n`$th slabs while $`F^{(n)}`$ describes the adiabatic propagation in the $`n`$th slab. They are given by
$$M_{ij}^{(n)}=\nu _i^{(n)}|\nu _j^{(n1)}=(U^{(n)^{}}U^{(n1)})_{ij}^{}$$
(15)
$$F_{ij}^{(n)}=\delta _{ij}exp\left(i_{t_{n1}}^{t_n}ϵ_i(t)𝑑t\right)$$
(16)
where the indices $`(n)`$ and $`(n1)`$ ocuring on $`\nu `$ and $`U`$ refer respectively to the $`n`$th and $`(n1)`$th slabs at the junction between these slabs. Also note that $`M^{(1)}`$ is the same as $`M^E`$ defined in Eq.(11). Defining the density matrix at the end of the $`n`$th slab as $`\rho ^{(n)}=|nn|`$, we have the recursion formula
$$\rho ^{(n)}=F^{(n)}M^{(n)}\rho ^{(n1)}M^{(n)^{}}F^{(n)^{}}$$
(17)
Starting with $`\rho ^{(0)}=|\nu _j\nu _j|`$ (i.e. $`\nu _j`$ entering the earth), we can calculate $`\rho ^{(N)}`$ at the end of the $`N`$th slab using Eq.(17). The probability of observing $`\nu _\beta `$ at the end of the $`N`$th slab is
$$P_{j\beta }^E=\nu _\beta |\rho ^{(N)}|\nu _\beta =(U^{(N)}\rho ^{(N)^{}}U^{(N)^{}})_{\beta \beta }$$
(18)
This formula (which reduces to Eq.(14) for $`N=1`$) can be used for the earth modeled as consisting of $`(N+1)/2`$ concentric shells, with the density varying gradually within each shell.
We have already referred to the detection of solar $`\nu `$ through the neutral current mode for bypassing the uncertainties of the solar models. Yet another way would be the detection of the night effect. An asymmetry between the night and day rates would be an unambiguous signal for neutrino oscillations independent of the details of the solar models. The recent results from SK for this asymmetry is at the level of $`0.06\pm 0.03`$ and is hence consistent with zero (at $`2\sigma `$). Even the absence of the effect contains important information since it helps to rule out certain regions of neutrino parameter space in an unambiguous manner.
The night effect is bound to exist at some level and the accummulated data will soon reveal its magnitude. Further, since the neutrino samples different amount of matter in the earth during a single night and also during the period of a year, the data accumulated in various bins at different times of the night contain an enormous amount of information on neutrino parameters. We have stressed the importance of analyzing this time-of-night variation and recent results from SK do suggest such a variation.
Many interesting physical effects are contained in the analytical formulae already presented. As an example, we shall mention what we may call ”vacuum oscillations in matter”. For $`\varphi 0`$, we get the following simple formula relating the survival probability in the night and day :
$$P_{ee}^N=P_{ee}^D+\left(12P_{ee}^D\right)\left(P_{2e}^E\mathrm{sin}^2\omega \right)\frac{1}{\mathrm{cos}2\omega }$$
(19)
where
$`P_{2e}^E`$ $`=`$ $`\mathrm{sin}^2\omega _E+\mathrm{sin}2\omega _E\mathrm{sin}2(\omega _E\omega )\mathrm{sin}^2{\displaystyle \frac{1}{4E}}{\displaystyle _{t_1}^{t_2}}\left[m_2^2(t)m_1^2(t)\right]𝑑t`$ (20)
$``$ $`\mathrm{sin}^2\omega +2(\omega _E\omega )\mathrm{sin}2\omega \mathrm{sin}^2{\displaystyle \frac{\delta m_{21}^2L}{4E}}`$ (21)
Here $`\omega _E`$ is the mixing angle just below the surface of the earth and $`L`$ is the distance the neutrino travels inside the earth. In arriving at the approximate expression for $`P_{2e}^E`$ given in Eq.(21), we have assumed that $`\delta m_{21}^2>>A(2\sqrt{2}G_FNE)`$, $`N`$ being the electron number density inside the earth. Under this approximation of small matter effect, $`P_{2e}^E`$ and hence $`P_{ee}^N`$ will exhibit vaccuum type oscillations as a function of the distance travelled within earth, but their amplitude will be controlled by matter density (since $`(\omega _E\omega )`$ is of order $`A/\delta m_{21}^2`$). Such regular oscillations were indicated in the earlier numerical calculations for appropriate choice of parameters and their interpretation is clear from our analytical formulae. This effect can perhaps be detected at the Borexino (however, see the remark made above concerning the average over $`t_R`$).
It is particularly important to see the effect of the core of the earth . A detector situated near the equator, such as one in South India can do this.
### B The eclipse effect
The above calculation can be extended to include the effect of the moon . We can consider both the case of the single eclipse when the neutrino passes through the moon only and the case of the “double eclipse” when it passes through the moon and the earth and gets detected on the night-side. We present the results only. We get , for the single eclipse,
$$P_{\alpha \beta }^M=\underset{j}{}P_{\alpha j}^SP_{j\beta }^M$$
(22)
where
$`P_{j\beta }^M`$ $`=`$ $`{\displaystyle \underset{\mathrm{}k\mathrm{}^{}k^{}}{}}U_\beta \mathrm{}^{}U_\beta \mathrm{}^{}M_{kj}^MM_{k^{}j}^M^{}M_k\mathrm{}^M^{}M_k^{}\mathrm{}^{}^M`$ (24)
$`exp\left\{i\left(ϵ_k^Mϵ_k^{}^M\right)d_Mi\left(ϵ_{\mathrm{}}ϵ_{\mathrm{}^{}}\right)r\right\}`$
and for the double eclipse
$$P_{\alpha \beta }^{ME}=\underset{j}{}P_{\alpha j}^SP_{j\beta }^{ME}$$
(25)
where
$`P_{j\beta }^{ME}`$ $`=`$ $`{\displaystyle \underset{\stackrel{\mathrm{}kp}{\mathrm{}^{}k^{}p^{}}}{}}U_{\beta p}^E^{}U_{\beta p^{}}^EM_p\mathrm{}^EM_p^{}\mathrm{}^{}^E^{}M_{kj}^MM_{k^{}j}^M^{}M_k\mathrm{}^M^{}M_k^{}\mathrm{}^{}^M`$ (27)
$`exp\left\{i(ϵ_k^Mϵ_k^{}^M)d_Mi(ϵ_{\mathrm{}}ϵ_{\mathrm{}^{}})ri(ϵ_p^Eϵ_p^{}^E)d_E\right\}`$
Here, the superscripts $`M`$ and $`E`$ refer to the moon and earth respectively and we have assumed constant densities for simplicity (but the expressions can be easily generalized to include variable densities and discrete jumps in densities) ; $`d_M`$, $`d_E`$ and $`r`$ denote the diameter of the moon, diameter of the earth and the earth-moon distance respectively ; $`M^M`$ and $`M^M^{}`$ are the non-adiabatic jump probability amplitudes at the vacuum-moon interface and the moon-vacuum interface respectively defined analogously to Eq.(11). We again note the convenient factorization in the results of Eqs(22) and (25).
Our calculations show considerable enhancements in the neutrino counting rate during the eclipse - even as high as 100%. However, since the counting rates are currently no more than about one per hour, the enhancement during the hour or two of the duration of the eclipse is hard to see at the present detectors. Perhaps we have to wait for the next generation of detectors.
## VI Neutrinos from supernovae
The observation of neutrinos from the supernova SN 1987A was an exciting event and it spurred much activity in this field. Since we now have some idea of the mass-differences and the mixing angles from the study of solar, atmospheric and reactor neutrinos, we may ask what effect do the oscillations have on the neutrinos from supernovae and whether such effects can be observed in the neutrino detectors during a supernova event in the future. Here we shall restrict ourselves to focussing attention on one such important signal discussed recently .
Consider the thermal or cooling phase of the supernova when all the three flavours of neutrinos and antineutrinos are emitted. We denote the flux of $`\nu _\alpha `$ and $`\overline{\nu }_\alpha `$ produced in the core by $`F_\alpha ^o`$ and $`F_{\overline{\alpha }}^o`$ respectively. For all practical purposes one can put
$$F_\mu ^o=F_{\overline{\mu }}^o=F_\tau ^o=F_{\overline{\tau }}^oF_x^o$$
(28)
If $`P_{\alpha \beta }`$ is the probability for $`\alpha `$ changing to flavour $`\beta `$ during propagation through the supernova, then, the fluxes of $`\nu _e`$ and $`\nu _\mu +\nu _\tau `$ coming out of the supernova are given by
$`F_e`$ $`=`$ $`F_e^oP_{ee}+F_\mu ^oP_{\mu e}+F_\tau ^oP_{\tau e}`$ (29)
$`=`$ $`F_e^o(1P_{ee})(F_e^oF_x^o)`$ (30)
$`2F_x`$ $`=`$ $`F_\mu +F_\tau =2F_x^o+(1P_{ee})(F_e^oF_x^o)`$ (31)
where we have used Eq.(28) and the constraint $`_\alpha P_{\alpha \beta }=1`$. A similar analysis for the antineutrinos gives
$`F_{\overline{e}}`$ $`=`$ $`F_{\overline{e}}^o(1P_{\overline{e}\overline{e}})(F_{\overline{e}}^oF_x^o)`$ (32)
$`2F_{\overline{x}}`$ $`=`$ $`2F_x^o+(1P_{\overline{e}\overline{e}})(F_{\overline{e}}^oF_x^o)`$ (33)
Let us now calculate $`P_{ee}`$ and $`P_{\overline{e}\overline{e}}`$. The variation of the three matter-dependent mass eigenvalues for the neutrinos and antineutrinos as functions of matter-density $`\rho `$ are schematically depicted in Fig(2). The $`\nu _e`$ has the decomposition in matter :
$$|\nu _e=\mathrm{cos}\varphi _m\mathrm{cos}\omega _m|\nu _1^m+\mathrm{cos}\varphi _m\mathrm{sin}\omega _m|\nu _2^m+\mathrm{sin}\varphi _m|\nu _3^m$$
(34)
where ‘m’ denotes matter. In the dense core of the supernova, $`\varphi _m\pi /2`$ and so the $`\nu _e`$ is emitted as $`\nu _3^m`$ in the fireball : $`|\nu _e=|\nu _3^m`$.
For the parameters relevant for supernovae, one can show that the Landau-Zener nonadiabatic jump probabilities at the two MSW resonances depicted in Fig(2) are vanishingly small as long as $`\mathrm{sin}\varphi 10^2`$. Hence, the neutrino state vector $`|\nu _3^m`$ evolves adiabatically and ends up as $`|\nu _3`$ as it emerges out of the supernova. Since $`\nu _e|\nu _3=\mathrm{sin}\varphi `$, we have $`P_{ee}=\mathrm{sin}^2\varphi 0`$ where we have used the result $`\varphi <9^o`$ from the CHOOZ reactor experiment.
For the antineutrinos, we start with
$$|\overline{\nu }_e=\mathrm{cos}\varphi _m\mathrm{cos}\overline{\omega }_m|\overline{\nu }_1^m+\mathrm{cos}\varphi _m\mathrm{sin}\overline{\omega }_m|\overline{\nu }_2^m+\mathrm{sin}\overline{\varphi }_m|\overline{\nu }_3^m$$
(35)
and since $`\overline{\varphi }_m0,\overline{\omega }_m0`$, at high densities, we see that, when produced, $`|\overline{\nu }_e=|\overline{\nu }_1^m`$ and $`\overline{\nu }_1^m`$ emerges from the supernova as $`\overline{\nu }_1`$. Using $`\overline{\nu }_e|\overline{\nu }_1=\mathrm{cos}\varphi \mathrm{cos}\omega `$, we therefore get $`P_{\overline{e}\overline{e}}=\mathrm{cos}^2\varphi \mathrm{cos}^2\omega 1`$ or $`\frac{1}{2}`$ for $`\varphi <9^o`$ and the small or large $`\omega `$ solar solution respectively.
Substituting these results into Eqs.(31) and (33), we get the changed neutrino fluxes due to oscillations:
$`F_e`$ $``$ $`F_x^0;2F_xF_e^0+F_x^0`$ (36)
$`F_{\overline{e}}`$ $``$ $`F_{\overline{x}}^0;2F_{\overline{x}}F_{\overline{e}}^0+F_{\overline{x}}^0\text{ (for small}\omega )`$ (37)
$`F_{\overline{e}}`$ $``$ $`{\displaystyle \frac{1}{2}}(F_{\overline{e}}^0+F_{\overline{x}}^0);2F_{\overline{x}}{\displaystyle \frac{1}{2}}(F_{\overline{e}}^0+3F_{\overline{x}}^0)\text{ (for large}\omega )`$ (38)
The changes in the neutrino detection rates arising from these have been calculated , but the important signal for oscillation is contained in Eq.(36) which states that $`\nu _x`$ (i.e $`\nu _\mu `$ or $`\nu _\tau `$) are converted into $`\nu _e`$. Since the original average energies of $`\nu _e`$ and $`\nu _x`$ in the supernova are 12 MeV and 24 MeV, respectively, the average energy of $`\nu _e`$ is shifted upwards by the oscillation. This can be detected by the charged current mode of <sup>16</sup>O in the H<sub>2</sub>O detector which has a threshold of 15.4 MeV. The rate for this mode can be enhanced by as much as two orders of magnitude as a consequence of oscillation . If one can construct a detector with <sup>16</sup>O or <sup>12</sup>C without protons, it will be ideal since otherwise the $`\overline{\nu }_ep`$ absorption reaction is dominant.
## VII Majorana neutrinos and global analysis
If neutrinos are Majorana fermions, then neutrinoless double beta decay is allowed. However the latter has not been seen yet and the experimental limits on it are getting stronger. The strongest upper limit so far comes from the Germanium experiment and it is
$$|\underset{j}{}m_jU_{ej}^2\eta _j|<0.2eV\text{(}at90\%CL)$$
(39)
where $`\eta _j(=\pm 1)`$ is the CP parity (apart from a factor $`i`$) of the Majorana neutrino $`\nu _j`$. The mixing matrix for Majorana neutrinos is
$$U=\left(\begin{array}{ccc}c_\omega c_\varphi & s_\omega c_\varphi e^{i\delta _1}& s_\varphi e^{i\delta _2}\\ s_\omega c_\psi e^{i\delta _1}c_\omega s_\psi s_\varphi e^{i(\delta _2+\delta _3)}& c_\omega c_\psi s_\omega s_\psi s_\varphi e^{i(\delta _3+\delta _2\delta _1)}& s_\psi c_\varphi e^{i\delta _3}\\ s_\omega s_\psi e^{i(\delta _1\delta _3)}c_\omega c_\psi s_\varphi e^{i\delta _2}& c_\omega s_\psi e^{i\delta _3}s_\omega c_\psi s_\varphi e^{i(\delta _2\delta _1)}& c_\psi c_\varphi \end{array}\right)$$
(40)
There are three CP-violating phases for Majorana neutrinos, in contrast to the case of Dirac neutrinos where there is only one phase (see Eq.2). However, for oscillation phenomena, only one combination of the three phases occurs and so oscillations cannot distinguish between Majorana and Dirac neutrinos.
Is it possible to combine the very important constraint on the neutrino masses and mixings provided by Eq.(39) with the information already derived from the solar, atmospheric and reactor neutrinos ? The answer is yes, provided we make some assumption about the neutrino mass scale. Until two years ago, cosmologists had claimed that their analysis of the data on the anisotropies of the cosmic microwave background radiation and the large scale structure of the universe require the presence of some hot component (presumably massive neutrinos) in the dark matter and their best fit was
$$\underset{j=1}{\overset{3}{}}m_j\mathrm{a}\mathrm{few}eV.$$
(41)
Many years ago , I had formulated the law that allowed only a one-way traffic between High Energy Physics and Cosmology : High Energy Physics $``$ Cosmology. We violated this law when we used the cosmological result of Eq.(41) in neutrino physics and punishment came in the form of the observation of high red shift supernovae and their interpretation in terms of a nonvanishing cosmological constant. The hot dark-matter component is no longer favoured by cosmologists. So, we now have to regard Eq.(41) merely as a cosmological assumption about the neutrino mass scale.
Since the oscillations of the solar and atmospheric neutrinos imply mass differences which are much smaller than the cosmological scale of Eq.(41), we can take all the three neutrinos as almost degenerate in mass: $`m_im_\nu 1eV\text{(}fori=1,2,3)`$ and so Eq.(39) becomes
$$|(\eta _1\mathrm{cos}^2\omega +\eta _2\mathrm{sin}^2\omega e^{2i\delta _1})\mathrm{cos}^2\varphi +\eta _3\mathrm{sin}^2\varphi e^{2i\delta _2}|<0.2$$
(42)
This constraint can be analysed for all possible choices of $`\delta _1,\delta _2`$ and $`\eta _i`$ and the allowed regions for the mixing angles $`\omega `$ and $`\varphi `$ can be mapped out . One fact can be immediately noted. Since $`\varphi <9^o`$ according to the reactor experiment, small values of $`\omega `$ cannot be consistent with Eq.(42). So we have the important conclusion : If the cosmological assumption of $`m_\nu `$ in the eV scale is correct and if the small-$`\omega `$ solution turns out to be the correct solution of the solar $`\nu `$ problem, then neutrinos cannot be Majorana fermions.
## VIII LSND and the fourth neutrino
Since all the results of the solar, atmospheric and reactor neutrino experiments could be consistently explained within the framework of three neutrinos, it seemed that all that was required was the resolution of the three-fold ambiguity of the solar neutrino solutions and more precision neutrino experiments to pin down the fundamental neutrino parameters. But a spanner was thrown into the works by the LSND experiments reporting positive results on $`\overline{\nu }_\mu \overline{\nu }_e`$ and $`\nu _\mu \nu _e`$. Since the base-line length of these experiments is as short as 29m, the implied $`\delta m^2`$ is in the rangle $`110eV^2`$. It is difficult to incorporate this result within the three-neutrino framework and so most theorists have decided to ignore the LSND result, citing the fact that it has not yet been confirmed by an independent experiment. The independent experiment KARMEN has not confirmed the LSND result, but KARMEN has not ruled out the full parameter space allowed by LSND either. A real confirmation or ruling out has to await the Mini BOONE experiment at FNAL, a long agonizing 4 years away.
If the LSND result is correct, we need a 4th neutrino, but since the known invisible width of $`Z`$ is completely exhausted by $`\nu _e,\nu _\mu `$ and $`\nu _\tau `$, the new neutrino has to be a singlet under $`SU(2)`$ and be sterile under known interactions. The natural mass hierarchy would be to place $`\nu _4`$ a few eV above the known three neutrinos, but this is contradicted by a combination of known experimental data, unless $`\nu _4`$ decays .
## IX Future
Is it possible to confirm or refute the results on neutrino oscillations claimed by the solar and atmospheric neutrino observations using laboratory experiments ? That would be one of the chief goals of the long-base-line neutrino experiments that are being planned (see Table 1.) One can see that sensitivities upto the level of $`\mathrm{\Delta }m^2`$ needed for solar and atmospheric neutrinos will be reached in these experiments. Even larger baselines can be contemplated.
Finally, there are prospects of constructing muon storage rings that will function as neutrino factories and these promise to take neutrino physics to a new era. Hopefully these as well as the long baseline experiments will lead to a determination of the neutrino parameters. Of course, entirely new phenomena could also be discovered.
|
warning/0004/astro-ph0004277.html
|
ar5iv
|
text
|
# The Population of Faint Transients in the Galactic Centre
## 1 Introduction
Over the past 3 yr the Wide Field Cameras on BeppoSAX have detected 9 soft X–ray transients in a 40x40 degree field around the Galactic Centre (GC) (see Heise et al. 1998 for a review). These outbursts are rather faint ($`L_X3\times 10^{37}`$ erg s<sup>-1</sup>) and have very short e–folding timescales $`\tau 26`$ d. Of the 9 detected systems, 7 have shown Type I X–ray bursts, indicating that the accretor is a neutron star. This statistic contrasts sharply with that for brighter transients in the Galaxy, which suggests that no more than one half contain neutron stars, while a large fraction are believed on the basis of dynamical mass measurements to contain black holes.
The faintness and short decay timescales of the GC transient outbursts imply a remarkably low accreted mass
$$\mathrm{\Delta }M1.8\times 10^{11}L_{37}\tau _4\mathrm{M}_{},$$
(1)
where $`L_{37},\tau _4`$ are $`L_x,\tau `$ in units of $`10^{37}`$ erg s<sup>-1</sup> and 4 d respectively, and we have assumed that each gram of accreted matter releases $`10^{20}`$ erg. (This mass estimate would rise if a significant fraction of the accreted matter radiated at low efficiency, i.e. was advected. This is only possible for black–hole systems.) The limits quoted above show that the maximum value of $`L_{37}\tau _4`$ is about 4.5, allowing $`\mathrm{\Delta }M`$ as large as $`8\times 10^{11}\mathrm{M}_{}`$.
Only two of the GC transients, SAX J1808.4-3658, and very recently SAX J1747.0-2853 (Markwardt et al. 2000) have been observed to recur. If all of the mass transferred from the companion is accreted efficiently by the compact primary, one can then estimate its mass transfer rate in SAX J1808.4-3658 as $`\dot{M}_{\mathrm{tr}}1\times 10^{11}\mathrm{M}_{}\mathrm{yr}^1`$ (Chakrabarty & Morgan, 1998). The non–recurrence of the other GC transients implies that their mass transfer rates must also be very low, i.e.
$$\dot{M}_{\mathrm{tr}}\mathrm{\Delta }M/(3\mathrm{yr})6\times 10^{12}L_{37}\tau _4\mathrm{M}_{}\mathrm{yr}^1.$$
(2)
We shall see in Section 2 below that explicit modelling of the accretion disc stability in these systems leads to a similar limit. (Note that the estimate (2) would no longer be an upper limit if, contrary to the assumption made above, a significant fraction of the transferred mass was lost from the system rather than accreted, cf Meyer & Meyer–Hofmeister, 1994).
There seem to be only two kinds of relatively abundant binary system which would have such low transfer rates:
1. wind–fed systems
2. Roche–lobe–filling systems in which the donor star has an extremely low mass $`M_2<0.1\mathrm{M}_{}`$, with mass transfer driven by gravitational radiation. This gives a rate (cf King, Kolb & Szuszkiewicz, 1997)
$$\dot{M}_{\mathrm{GR}}1\times 10^{11}m_1^{2/3}\left(\frac{m_2}{0.1}\right)^2\left(\frac{P}{2\mathrm{hr}}\right)^{8/3}\mathrm{M}_{}\mathrm{yr}^1,$$
(3)
where $`m_1,m_2`$ are the accretor and donor masses in $`\mathrm{M}_{}`$, and I have assumed $`m_1>>m_2`$.
Note that 1. need not automatically imply a high–mass donor, which would be in possible contradiction with the usual identification of soft X–ray transients as low–mass X–ray binaries (LMXBs). For the mass transfer rates (2) are so low that even the comparatively weak wind of a low–mass main sequence star may be sufficient to power them. Alternatively, radio pulsar irradiation could excite a wind from a low–mass companion star. Orbital angular momentum losses may then shrink the binary to the point where the neutron star begins to capture this wind material and turns on as an X–ray source. The source will be transient because the accretion rate through the disc is too low to keep it fully ionized (see equation 11 below).
A possible example of a higher–mass wind–fed transient may be CI Cam (e.g. Marshall et al., 1998, Clark et al., 2000), which has been variously classified as a symbiotic system, or a B\[e\] binary. Any such system must be wide enough that the UV radiation from the high–mass donor is unable to keep the accretion disc ionized. Whatever the mass of the donor, accretion by wind capture will tend to produce an accretion disc of small radius: if much of the disc mass is accreted in an outburst, the outer radius of the disc will be determined by the low specific angular momentum of the captured wind material, naturally producing a small disc. This in turn would account for the low accreted mass $`\mathrm{\Delta }M`$ and short decay times $`\tau `$ (see equations 5, 9 below). One might test for the presence of high–mass companions of this type in the GC transients through optical/IR identifications. There are currently only two reported identifications, both of which are inconsistent with high–mass stars. In SAX J1808.4-3658 (Roche et al., 1998) the 2–hour orbital period rules out any such companion. In SAX J1810.8-2609, Greiner et al. (1999) find $`R=19`$ fading to $`R>21.5`$ for the optical/IR source. This variation means that this source is presumably dominated by the accretion disc, which is probably irradiated by the central X–rays. This is inconsistent with a high–mass companion, which would dominate the optical/infrared as it would have a similar effective temperature to the disc, but be much larger.
These arguments fall short of proving that no GC transient is wind–fed, in either high–mass or low–mass versions. However the second possibility listed above seems inescapable. At the end of their lives as accreting binaries, LMXBs must reduce the donor mass $`M_2`$ to the point where this star begins to become degenerate (i.e. $`M_20.1\mathrm{M}_{}`$). If the orbital evolution of LMXBs is similar to the standard picture assumed for cataclysmic variables (CVs), where the accretor is a white dwarf rather than a neutron star or black hole, the binary period will be close to $`80`$ minutes at this point (see e.g. King, 1988; Kolb & Baraffe, 1999 for reviews). Subsequently this period begins to increase rather than decrease as orbital angular momentum is lost via gravitational radiation, as the donor expands in response to further mass loss. The mass transfer rate drops to values comparable with those of equation (2). Because of this, the mass transfer timescale $`M_2/\dot{M}_{\mathrm{tr}}`$ becomes very long, eventually approaching a Hubble time. The system’s evolution thus slows considerably; even a system ‘born’ at the minimum period of $`80`$ minutes requires a time of order the age of the Galaxy to reach $`P2`$ hr. Mass transfer continues, albeit at a very slow rate ($`(0.51)\times 10^{11}\mathrm{M}_{}\mathrm{yr}^1`$). Hence there should exist a population of such extremely faint LMXB systems, directly analogous to the post–minimum–period population of CVs. While no post–minimum CV has been certainly identified, their LMXB analogues are potentially much easier to observe since they are all likely to be soft X–ray transients (King, Kolb & Szuszkiewicz, 1997). As we shall see, their expected properties are very similar to those of the faint Galactic Centre transients. Note that there may be other more exotic ways of producing individual systems among the faint GC transients, as has for example been proposed for SAX J1808.4-3658 by Ergma & Antipova (1999). However post–minimum systems offer the most likely way of producing them in significant numbers.
The identification of the faint GC transients with post–minimum LMXBs has the desirable property of explaining why neutron stars are favoured among this population. Black–hole LMXBs are likely to be transient at all orbital periods (King, Kolb & Szuszkiewicz, 1997), while neutron–star LMXBs with main–sequence donors (i.e. before they reach the minimum period) are likely to be persistent. Accordingly black–hole systems predominate among bright transients, since a bright neutron–star transient requires an unusual (nuclear–evolved) donor. However the calculations of King, Kolb & Szuszkiewicz (1997) show that neutron–star LMXBs will become transient as they evolve beyond the the minimum period. Thus the ratio of neutron star to black hole systems must be much higher for post–minimum LMXBs than for bright transients. This naturally explains why the faint Galactic Centre transients seem mostly to contain neutron stars.
In the next Section I show that the simple irradiated–disc model of soft X–ray transient (SXT) outbursts proposed by King & Ritter (1998) predicts outburst masses $`\mathrm{\Delta }M`$ and decay times $`\tau `$ in excellent agreement with the observational estimates given above, provided that the outer disc radius is small, i.e. $`10^{10}`$ cm. Moreover the maximum average mass transfer rate consistent with irradiation not suppressing the outbursts is close to the limit (2). In Section 3 I discuss the two possible types of binary system considered above in the light of these disc properties. Section 4 is the Conclusion.
## 2 Accretion disc properties in the faint GC transients
Soft X–ray transient outbursts are thought to result from instabilities in LMXB accretion discs. Both the incidence and the nature of these outbursts are strongly affected by irradiation of the disc surfaces by the central X–rays. Irradiation can suppress outbursts in some LMXB discs by removing their hydrogen ionization zones (van Paradijs, 1996; King, Kolb & Burderi, 1996; King, Kolb & Szuszkiewicz, 1997), thus making them stable (persistent) at lower mass transfer rates than is true for the otherwise similar discs in CVs. The irradiation effect appears to be weaker if the accretor is a black hole rather than a neutron star, possibly because of the lack of a hard surface (King, Kolb & Szuszkiewicz, 1997). The result is that neutron–star LMXBs with main–sequence companions tend to be persistent, while similar black–hole binaries are largely transient.
If an LMXB disc goes into outburst, irradiation of the disc greatly prolongs the high state, and causes viscous rather than thermal effects to dominate the light–curve (King & Ritter, 1998). This often produces an exponential decay, particularly if the whole disc is efficiently irradiated. Several predictions of this simple irradiated–disc picture are confirmed by observation (cf Shahbaz, Charles & King, 1998). More detailed calculations with a full 1–D disc code (Dubus et al, 1999) give results very similar to those of King and Ritter (1998) so I shall use their simple analytic expressions in the following.
As the observed decays are approximately exponential I assume that the entire disc of a faint transient (of radius $`R`$) is irradiated during an outburst. Most of this mass is accreted. Immediately before the outburst the surface density in the disc must have been close to the value
$$\mathrm{\Sigma }_{\mathrm{max}}=11.4R_{10}^{1.05}m_1^{0.35}\alpha _c^{0.86}\mathrm{g}\mathrm{cm}^2$$
(4)
(Cannizzo, Shafter & Wheeler, 1988) triggering the thermal instability through local viscous dissipation, where $`R_{10}=R/10^{10}`$ cm, with $`R`$ the radial disc coordinate, $`m_1`$ is the central accreting mass in $`\mathrm{M}_{}`$, and $`\alpha _c`$ is the cold–state viscosity parameter. Thus by integrating over $`R`$ we predict the mass accreted in the outburst as
$$\mathrm{\Delta }M_{\mathrm{pr}}1.5\times 10^{11}m_1^{0.35}\alpha _{0.05}^{0.86}R_{10}^{3.05}\mathrm{M}_{},$$
(5)
where $`\alpha _{0.05}=\alpha _c/0.05`$. This relation can be compared with equation (8) of King & Ritter (1998), who used a simpler form of $`\mathrm{\Sigma }_{\mathrm{max}}`$. Equating this to the observational estimate (1) we find a disc radius
$$R1.2\times 10^{10}(L_{37}\tau _4)^{0.33}m_1^{0.11}\alpha _{0.05}^{0.28}\mathrm{cm}.$$
(6)
King & Ritter (1998) predict that the e–folding time for the decay will be
$$\tau _{\mathrm{pr}}=\frac{R^2}{3\nu },$$
(7)
where $`\nu =\alpha _hc_SH`$ is the hot–state kinematic viscosity at the disc edge. Here $`c_S,H`$ are the local sound speed and scale height, and we have
$$H=\frac{c_S}{\mathrm{\Omega }}$$
(8)
with $`\mathrm{\Omega }=(GM_1/R^3)^{1/2}`$ the Kepler frequency at disc radius $`R`$. Writing $`T_4`$ for the surface temperature at this point in units of $`10^4`$ K, we find the predicted timescale
$$\tau _{\mathrm{pr}}=3.9\alpha _h^1m_1^{0.5}R_{10}^{0.5}T_4^1\mathrm{d}.$$
(9)
The theoretical predictions (5, 9) are in excellent agreement with the observational values $`\mathrm{\Delta }M`$, $`\tau `$ and thus $`L_X`$ provided that the hot–state viscosity parameter takes a value $`\alpha _hT_41`$, as expected, and $`R10^{10}`$ cm. Evidently the distinctive feature of the discs in the faint GC transients is their very small size, which accounts for both their low peak luminosities and their rapid decays. We can now ask how low the mean mass transfer rate $`\dot{M}_{\mathrm{tr}}`$ must be if such a disc is to have ionization zones despite being so small. This requires that the surface temperature $`T_{\mathrm{irr}}`$ resulting from irradiation should be less than the ionization temperature $`T_\mathrm{H}6500`$ K at the disc edge. I take
$$T_{\mathrm{irr}}(R)^4=\frac{10^{20}\dot{M}_{\mathrm{tr}}}{4\pi \sigma R^2}\left(\frac{H}{R}\right)^n\left[\frac{\mathrm{d}\mathrm{ln}H}{\mathrm{d}\mathrm{ln}R}1\right],$$
(10)
(e.g. van Paradijs, 1996) where the factor in square brackets lies between 1/8 and 2/7 and the index $`n=1`$ or 2 for a neutron star or black hole respectively (cf King, Kolb & Szuszkiewicz, 1997). Requiring $`T_{\mathrm{irr}}<T_\mathrm{H}`$ and using the estimate (6) gives
$$\dot{M}_{\mathrm{tr}}1.3\times 10^{11}(L_{37}\tau _4)^{0.66}m_1^{0.22}\alpha _{0.05}^{0.56}\mathrm{M}_{}\mathrm{yr}^1$$
(11)
for a neutron star, and a limit about 10 times larger for a black hole. This is again in excellent agreement with the observational limit (2).
## 3 Binary models for the faint GC transients
The good agreement with simple irradiated–disc theory found above shows that the faint GC transient population must be binaries with two key properties
(a) disc radii $`R1.2\times 10^{10}`$ cm, and
(b) mean mass transfer rates $`\dot{M}_{\mathrm{tr}}(0.61)\times 10^{11}\mathrm{M}_{}\mathrm{yr}^1`$.
Both the binary types 1 (wind–fed) and 2 (low–mass donor) considered in the Introduction are able to reproduce these conditions. In the wind–fed case, the low specific angular momentum of matter captured from a wind will produce a small disc radius provided that most of the disc mass is accreted during an outburst. The low mass transfer rate is a natural consequence of inefficient wind capture. However, the specific values of $`R`$ and $`\dot{M}_{\mathrm{tr}}`$ depend in detail on the properties of the donor wind, and it is not obvious why these should cluster around the values producing (a) and (b) above. By contrast, the version of model 2 involving post–minimum LMXBs will naturally produce these values. If most of the disc mass is accreted in an outburst, as is implicit in the estimate (5), the disc radius $`R`$ will be close to the circularization radius
$$R_{\mathrm{circ}}(1+q)(0.70.227\mathrm{log}q)^4a,$$
(12)
(e.g. Frank et al, 1992) where $`q=M_2/M_1`$ is the mass ratio and $`a`$ is the binary separation, i.e.
$$a=3.53\times 10^{10}m_1^{1/3}P_{\mathrm{hr}}^{2/3}\mathrm{cm}.$$
(13)
With $`M_1=1.4\mathrm{M}_{},M_2=0.1\mathrm{M}_{}`$ we find an expected disc radius
$$RR_{\mathrm{circ}}=1.7\times 10^{10}\left(\frac{m_1}{1.4}\right)^{1/3}\left(\frac{P}{80\mathrm{min}}\right)^{2/3}\mathrm{cm}.$$
(14)
Thus binary periods in the typical range 80 min – 2 hr will produce discs of the right size (cf eqn. 6) for neutron–star masses $`m_1=1.4`$. Equating the two expressions (6, 14) gives the requirement
$$L_{37}\tau _4=5m_1^{2/3}\alpha _{0.05}^{0.84}\left(\frac{P}{2\mathrm{hr}}\right)^2,$$
(15)
which compares well with the observed range of this quantity given in the Introduction. The calculations of King, Kolb & Szuszkiewicz (1997) show that such systems have transfer rates at or below the limit (b), which also ensure that the systems are indeed transient there (see their Fig. 3).
For black–hole masses $`m_17`$ we see that the predicted disc radius $`R`$ tends to become rather larger than the estimate (6), leading to more prolonged outbursts (transferred mass $`\mathrm{\Delta }M_{\mathrm{pr}}10^{10}\mathrm{M}_{}`$, decay timescale $`\tau _{\mathrm{pr}}27`$ d) than are typical of the faint GC transients (note that eqs. 5, 14 imply that $`\mathrm{\Delta }M_{\mathrm{pr}}m_1`$, while eq. 9 implies $`\tau _{\mathrm{pr}}m_1`$). Thus such systems would probably not be classified as faint transients. Moreover black–hole systems will have longer recurrence times $`t_{\mathrm{rep}}\mathrm{\Delta }M/\dot{M}_{\mathrm{tr}}m_1/m_1^{2/3}m_1^{1/3}`$ and thus lower discovery probability ($`\dot{M}_{\mathrm{tr}}m_1^{2/3}`$ for gravitational radiation, cf eqn 3). This already tends to suggest agreement with the observation that at least 7 out of 9 GC transients contain neutron stars. A still stronger reason comes from the fact that Fig. 3 of King, Kolb & Szuszkiewicz (1997) shows that essentially all neutron–star LMXBs will be transient once they have evolved sufficiently far beyond the minimum period, and neutron–star systems are much more common than black–hole ones in general. Thus the ratio of neutron–star to black–hole systems among the faint GC transients should be
$`{\displaystyle \frac{N_{\mathrm{NS}\mathrm{GCtransients}}}{N_{\mathrm{BH}\mathrm{GCtransients}}}}{\displaystyle \frac{N_{\mathrm{NS}\mathrm{post}}}{N_{\mathrm{BH}\mathrm{post}}}}`$
$`={\displaystyle \frac{N_{\mathrm{NS}\mathrm{post}}}{N_{\mathrm{NS}\mathrm{pre}}}}.{\displaystyle \frac{N_{\mathrm{NS}\mathrm{pre}}}{N_{\mathrm{BH}\mathrm{pre}}}}.{\displaystyle \frac{N_{\mathrm{BH}\mathrm{pre}}}{N_{\mathrm{BH}\mathrm{post}}}}`$ , (16)
where the $`N`$ are space densities and ‘pre’ and ‘post’ refer to the minimum period. Now Fig. 3 of King, Kolb & Szuszkiewicz (1997) implies
$$\frac{N_{\mathrm{NS}\mathrm{post}}}{N_{\mathrm{NS}\mathrm{pre}}}\frac{N_{\mathrm{BH}\mathrm{post}}}{N_{\mathrm{BH}\mathrm{pre}}},$$
(17)
i.e. the slow–down of neutron–star LMXB evolution after passing the minimum period is more dramatic than that of black–hole systems. Using (17) in (16), the two outer factors combine to give a number $`1`$, so we get
$$\frac{N_{\mathrm{NS}\mathrm{GCtransients}}}{N_{\mathrm{BH}\mathrm{GCtransients}}}\frac{N_{\mathrm{NS}\mathrm{pre}}}{N_{\mathrm{BH}\mathrm{pre}}}>>1,$$
(18)
where the last inequality expresses the fact that there are far more neutron–star then black–hole binaries.
## 4 Conclusions
I have shown that the simple irradiated–disc picture gives a consistent fit to the properties of the faint GC transients, and that this population probably consists of post–minimum LMXBs. Neutron star systems far outweigh black–hole ones here, mainly because all post–minimum systems are transient, and neutron–star LMXBs are simply more common than black–hole ones in the Galaxy (outbursts of the latter would also probably be too long to be classified among the faint transients). By contrast among brighter transients, which are generally pre–minimum systems, the greater incidence of transient behaviour among black–hole systems makes them prominent. If the identification as post–minimum LMXBs is correct, the faint transients should be binaries with periods in the range $`80120`$ minutes, with extremely low–mass companions.
An interesting point emerges from the fact that only two faint GC transients have yet been observed to repeat. From (2) we see that if this state of affairs persists for a decade or so, the resulting upper limit on the mean mass transfer rate will become embarassingly low even for post–minimum LMXBs. On the other hand, a typical repetition time $`t_{\mathrm{rep}}`$ of order a few years would also severely limit the total number of such systems in the Galaxy: Heise et al. (1998) estimate from the sky and temporal coverage of the BeppoSAX WFC that there are about 18 faint transient outbursts in the Galaxy per year, leading to a total population $`N_{\mathrm{Gal}}18(t_{\mathrm{rep}}/\mathrm{yr})`$. This number ($`50N_{\mathrm{Gal}}180`$) is considerably smaller than the usual estimate of $`1000`$ LMXBs in the Galaxy, whereas one might expect it to be much larger, as the evolution of post–minimum systems is so slow. There are several possible reasons for this, of which two seem most likely.
(a) If most LMXBs first reach contact at initial periods $`P_i`$ greater than a few hours (as indeed suggested by theoretical studies of LMXB formation, e.g. Kalogera & Webbink, 1996; King & Kolb 1997), their lifetimes before becoming post–minimum transients may be comparable with the age of the Galaxy. This lifetime is spent mostly near the minimum period (cf Kolb & Baraffe, 1999) while according to King, Kolb & Szuszkiewicz (1997), neutron–star LMXBs have to evolve somewhat beyond this period in order to become transient. A large fraction of the neutron–star LMXBs ever formed in the Galaxy may still not have reached this stage.
(b) Isolated millisecond pulsars are thought to be neutron stars spun up in LMXBs, but which have evaporated their companions (see Bhattacharya & van den Heuvel, 1991, for a review). This suggests that many neutron–star LMXBs may not even reach the theoretical minimum period, and thus do not become faint transients at all. It is sometimes hypothesized (cf Bhattacharya & van den Heuvel, 1991) that most neutron–star LMXBs evaporate the companion star as the system attempts to cross the analogue of the CV period gap between 3 hr and 2 hr (see e.g. King, 1988 for a review of the latter). If so, this would limit post–minimum neutron–star systems to the rare examples first coming into contact at periods below the period gap, i.e. with $`P_i2`$ hr. Of course there can be no such effect for black–hole LMXBs.
At present we do not have any clear idea of the mean repetition time for the faint GC transients. The arguments above show that there are interesting consequences whatever this number turns out to be. Extensive X–ray monitoring of the Galactic Centre region clearly has much more to tell us about the stellar populations there.
## 5 Acknowledgment
I gratefully acknowledge the support of a PPARC Senior Fellowship. I thank John Heise, Uli Kolb, Hans Ritter and Klaus Schenker for useful discussions, and the referee for a perceptive and helpful report.
|
warning/0004/quant-ph0004020.html
|
ar5iv
|
text
|
# Multipartite pure-state entanglement and the generalized GHZ states
## I INTRODUCTION
Ever since it was first noted by Einstein-Podolsky-Rosen (EPR) and Schrödinger , entanglement has played an important role in quantum information theory. Quantum entanglement provides strong tests of quantum nonlocality , and it is also a useful resource for various kinds of quantum information processing, including teleportation , cryptographic key distribution , quantum error correction and quantum computation .
Now, one of the key open questions in quantum information theory is how many fundamentally different types of quantum entanglement there are. It was known that asymptotically there is only one kind of entanglement for bipartite pure states, any pure entangled state of two parties (Alice and Bob) may be reversibly transformed into EPR states by local quantum operations and classical communication (LOCC) asymptotically.
For multipartite pure states, it is more difficult to understand the types of entanglement. It was not known whether the EPR states are the only type of entanglement until the recent work of Bennett-Popescu-Rohrlich-Smolin-Thapliyal (BPRST) , which shows that the 4-party GHZ state can not be reversibly transformed into EPR states by LOCC asymptotically. Furthermore Linden-Popescu-Schumacher-Westmoreland (LPSW) had shown that the n-party GHZ state cannot be reversibly transformed into any combination of $`k`$-party entangled pure states for all $`k<n`$. This means that the generalized $`n`$-party GHZ state
$$|GHZ_n_{ABC\mathrm{}}\frac{1}{\sqrt{2}}(|0^n+|1^n)$$
(1)
represents a different type of entanglement with respect to the $`k`$-party GHZ state (for all $`kn`$) <sup>*</sup><sup>*</sup>*A particular $`n`$-party GHZ state is chosen to represent all the $`n`$-party GHZ states since they are related by local unitary transformations..
A natural question arises, are the generalized GHZ states the only types of entanglement? Thapliyal had shown that any multi-separable pure state is Schmidt decomposable, thus a $`m`$-party separable pure state ( a state contains no entanglement of $`k`$-party for all $`k<m`$) can be reversibly transformed into the $`m`$-party GHZ state, this result supports (but does not prove) the hypothesis that the generalized GHZ states are the only types of entanglement with the $`k`$-party GHZ state representing ”essential” $`k`$-party entanglement.
In this note, we show that the generalized GHZ states are not the only types of entanglement through an example of 4-party pure state. We also present some properties of the relative entropy of entanglement for those 3-party pure states that can be generated reversibly from 2- and 3-party GHZ states, and then we use these properties to analyze the additivity of the relative entropy of entanglement.
Before going to the results, we state some terminology more clearly. Two $`m`$-party pure states $`|\psi `$ and $`|\phi `$ are LOCCa equivalent if and only if
$`_{\delta >0,ϵ>0}_{n_1,n_2,n_3,n_4,L,L^{^{}}}\text{ so that}`$ (2)
$`|(n_1/n_2)1|`$ $`<`$ $`\delta \text{ ,}`$ (3)
$`|(n_3/n_4)1|<\delta \text{ , and}`$ (4)
$`F(L(|\psi ^{n_1}),|\phi ^{n_2})`$ $``$ $`1ϵ`$ (5)
$`F(L^{^{}}(|\psi ^{n_3}),|\phi ^{n_4})`$ $``$ $`1ϵ`$ (6)
where $`L`$ and $`L^{^{}}`$ are local quantum operations assisted by classical communication, and
$$F(|\mathrm{\Phi },|\mathrm{\Psi })\left|\mathrm{\Phi }|\mathrm{\Psi }\right|^2$$
(7)
is the fidelity of $`|\mathrm{\Psi }`$ relative to $`|\mathrm{\Phi }`$. Condition (4) means that, in the limit of large $`n`$, $`n`$ copies of $`|\psi `$ can be transformed into almost the same number of copies of $`|\phi `$ by LOCC with high fidelity, and vice versa. The LOCCa equivalence of the two $`m`$-party pure states $`|\psi `$ and $`|\phi `$ is denoted as
$$|\psi \stackrel{LOCCa}{}|\phi $$
(8)
We say that a $`m`$-party pure state $`|\psi `$ is GHZ reducible, if and only if the state $`|\psi `$ is LOCCa equivalent to a combination of $`2`$-, $`3`$-, $`\mathrm{}`$, $`m`$-party generalized GHZ states. For example, since any bipartite pure state $`|\psi _{AB}`$ is GHZ reducible , we can write
$$|\psi _{AB}\stackrel{LOCCa}{}|EPR_{AB}^{E(|\psi _{AB})}$$
(9)
where $`E(|\psi _{AB})`$ is the unique measure of entanglement for bipartite pure states, it is equal to the entropy of the reduced density matrix of either Alice or Bob, as well as the entanglement of formation , entanglement of distillation and the relative entropy of entanglement . A $`3`$-party GHZ reducible pure state $`|\psi _{ABC}`$ can be written as
$`|\psi _{ABC}\stackrel{LOCCa}{}|EPR_{AB}^{E_2(AB)}|EPR_{AC}^{E_2(AC)}`$ (10)
$`|EPR_{BC}^{E_2(BC)}|GHZ_{ABC}^{E_3(ABC)}`$ (11)
i.e., in the limit of large $`n`$, with high fidelity, $`n`$ copies of the state $`|\psi _{ABC}`$ can be transformed reversibly by LOCC into $`nE_2(AB)`$ copies of the state $`|EPR_{AB}`$ held by Alice and Bob, $`nE_2(AC)`$ copies of $`|EPR_{AC}`$ held by Alice and Claire, $`nE_2(BC)`$ copies of $`|EPR_{BC}`$ held by Bob and Claire, and $`nE_3(ABC)`$ copies of $`|GHZ_{ABC}`$ held by Alice, Bob and Claire. The GHZ reducible multipartite pure states can be written in similar forms.
Let Ç denotes a set of pure states, if each of the $`m`$-party pure states is LOCCa equivalent to a certain combination of the states in Ç, then we say that Ç is a reversible entanglement generating set (REGS) for $`m`$-party pure states. A minimal reversible entanglement generating set (MREGS) for $`m`$-party pure states is a REGS of minimal cardinality. It is obvious that the set Ç<sub>2</sub>={$`|EPR_{AB}`$} is a MREGS for bipartite pure states. The question that whether the set of $`2`$-, $`3`$-, $`\mathrm{}`$, $`m`$-party generalized GHZ states is a MREGS for $`m`$-party pure states is, in fact, equivalent to the question that whether all $`m`$-party pure states are GHZ reducible.
We now state BPRST’s lemma about the LOCC equivalence.
BPRST’s lemma: If two $`m`$-party quantum states $`|\mathrm{\Psi }`$ and $`|\mathrm{\Phi }`$ are LOCCa equivalent, then they must be isentropic, i.e.,
$$S_X\left(|\mathrm{\Psi }\right)=S_X\left(|\mathrm{\Phi }\right)$$
(12)
where $`S_X\left(|\mathrm{\Psi }\right)=tr\left\{\rho _X\left(|\mathrm{\Psi }\right)\mathrm{log}_2\rho _X\left(|\mathrm{\Psi }\right)\right\}`$ with $`\rho _X\left(|\mathrm{\Psi }\right)tr_{\overline{X}}\left(|\mathrm{\Psi }\mathrm{\Psi }|\right)`$, and $`X`$ denotes a nontrivial subset of the parties (say Alice, Bob, Claire, Daniel, et al.), $`\overline{X}`$ denotes the set of the remaining parties.
This lemma is a consequence of the fact that average partial entropy $`S_X`$ cannot increase under LOCC, details of proof can be found in ref. .
## II The set of generalized GHZ states is not a MREGS
Now we show that the generalized GHZ states are not the only types of entanglement by proving that the set of $`2`$-, $`3`$-, $`4`$-party GHZ states is not a MREGS for $`4`$-party pure states, or in another word, not all $`4`$-party pure states are GHZ reducible.
Proposition 1: The set of $`2`$-, $`3`$-, $`4`$-party GHZ states is not a MREGS for $`4`$-party pure states.
Before the proof, let us first give a property of all the GHZ reducible $`4`$-party pure states. Suppose the $`4`$-party pure state $`|\mathrm{\Psi }_{ABCD}`$ is GHZ reducible, i.e.,
$`|\mathrm{\Psi }_{ABCD}`$ $`\stackrel{LOCCa}{}`$ $`|EPR_{AB}^{E_2(AB)}|EPR_{AC}^{E_2(AC)}|EPR_{AD}^{E_2(AD)}|EPR_{BC}^{E_2(BC)}`$ (15)
$`|EPR_{BD}^{E_2(BD)}|EPR_{CD}^{E_2(CD)}|GHZ_{ABC}^{E_3(ABC)}|GHZ_{ABD}^{E_3(ABD)}`$
$`|GHZ_{ACD}^{E_3(ACD)}|GHZ_{BCD}^{E_3(BCD)}|GHZ_4_{ABCD}^{E_4(ABCD)}`$
From BPRST’s lemma and the additivity of the von Neumann entropy, we have
| $`S\left(\rho _A\right)=E_2\left(AB\right)+E_2\left(AC\right)+E_2\left(AD\right)+E_3\left(ABC\right)+E_3\left(ABD\right)+E_3\left(ACD\right)+E_4\left(ABCD\right)`$ |
| --- |
| $`S\left(\rho _B\right)=E_2\left(AB\right)+E_2\left(BC\right)+E_2\left(BD\right)+E_3\left(ABC\right)+E_3\left(ABD\right)+E_3\left(BCD\right)+E_4\left(ABCD\right)`$ |
| $`S\left(\rho _C\right)=E_2\left(AC\right)+E_2\left(BC\right)+E_2\left(CD\right)+E_3\left(ABC\right)+E_3\left(ACD\right)+E_3\left(BCD\right)+E_4\left(ABCD\right)`$ |
| $`S\left(\rho _D\right)=E_2\left(AD\right)+E_2\left(BD\right)+E_2\left(CD\right)+E_3\left(ABD\right)+E_3\left(ACD\right)+E_3\left(BCD\right)+E_4\left(ABCD\right)`$ |
(16)
and
$$\begin{array}{ccc}S\left(\rho _{AB}\right)\hfill & =\hfill & E_2\left(AC\right)+E_2\left(AD\right)+E_2\left(BC\right)+E_2\left(BD\right)+E_3\left(ABC\right)\hfill \\ & & +E_3\left(ABD\right)+E_3\left(ACD\right)+E_3\left(BCD\right)+E_4\left(ABCD\right)\hfill \\ S\left(\rho _{AC}\right)\hfill & =\hfill & E_2\left(AB\right)+E_2\left(AD\right)+E_2\left(BC\right)+E_2\left(CD\right)+E_3\left(ABC\right)\hfill \\ & & +E_3\left(ABD\right)+E_3\left(ACD\right)+E_3\left(BCD\right)+E_4\left(ABCD\right)\hfill \\ S\left(\rho _{AD}\right)\hfill & =\hfill & E_2\left(AB\right)+E_2\left(AC\right)+E_2\left(BD\right)+E_2\left(CD\right)+E_3\left(ABC\right)\hfill \\ & & +E_3\left(ABD\right)+E_3\left(ACD\right)+E_3\left(BCD\right)+E_4\left(ABCD\right)\hfill \end{array}$$
(17)
From Eqs. (16) it follows that
$$\underset{i\{A,B,C,D\}}{}S\left(\rho _i\right)=2E_{2t}+3E_{3t}+4E_4$$
(18)
with $`E_{2t}`$ ($`E_{3t}`$, $`E_4`$) representing the ”total” $`2`$\- ($`3`$-, $`4`$-) party entanglement, which is defined by
$$\begin{array}{ccc}E_{2t}\hfill & =\hfill & E_2\left(AB\right)+E_2\left(AC\right)+E_2\left(AD\right)\hfill \\ & & +E_2\left(BC\right)+E_2\left(BD\right)+E_2\left(CD\right)\hfill \\ E_{3t}\hfill & =\hfill & E_3\left(ABC\right)+E_3\left(ABD\right)\hfill \\ & & +E_3\left(ACD\right)+E_3\left(BCD\right)\hfill \\ E_4\hfill & =\hfill & E_4\left(ABCD\right)\hfill \end{array}$$
(19)
And from Eqs. (17), there is
$$S\left(\rho _{AB}\right)+S\left(\rho _{AC}\right)+S\left(\rho _{AD}\right)=2E_{2t}+3E_{3t}+3E_4$$
(20)
It follows from eq. (18) and (20) that
$$E_4=\underset{i\{A,B,C,D\}}{}S\left(\rho _i\right)\left\{S\left(\rho _{AB}\right)+S\left(\rho _{AC}\right)+S\left(\rho _{AD}\right)\right\}$$
(21)
This is the amount of ”essential” $`4`$-party entanglement contained in the state $`|\mathrm{\Psi }_{ABCD}`$, therefore it must be non-negative, i.e.,
$$\underset{i\{A,B,C,D\}}{}S\left(\rho _i\right)\left\{S\left(\rho _{AB}\right)+S\left(\rho _{AC}\right)+S\left(\rho _{AD}\right)\right\}0$$
(22)
Eq. (22) is a property of all the GHZ reducible $`4`$-party pure states. Similar results for $`m`$-party GHZ reducible pure states can follow from the same argument.
Now let us take the state
$$|\psi _{ABCD}=\frac{1}{2}\left\{|0000+|0110+|1001|1111\right\}$$
(23)
as an example. It’s obvious that
$`S\left(\rho _A\right)`$ $`=`$ $`S\left(\rho _B\right)=S\left(\rho _C\right)=S\left(\rho _D\right)=1`$ (24)
$`S\left(\rho _{AB}\right)`$ $`=`$ $`S\left(\rho _{AC}\right)=2`$ (25)
$`S\left(\rho _{AD}\right)`$ $`=`$ $`1`$ (26)
therefore
$$E_4=4\times 1\left(2+2+1\right)=1$$
(27)
This contradicts eq. (22). Thus we have shown that not all $`4`$-party pure states are GHZ reducible, so the set of $`2`$-, $`3`$-, $`4`$-party GHZ states is not a MREGS for $`4`$-party pure states. This completes the proof of proposition 1.
Proposition 1 shows that the set of $`11`$ generalized GHZ states in eq. (15) is not enough for a MREGS, i.e., the number of members in a MREGS for $`4`$-party pure states must be greater than 11.
## III GHZ REDUCIBLE TRIPARTITE PURE STATES
It was known that any bipartite pure state is GHZ reducible, and from proposition 1 we know that not all $`4`$-party pure states are GHZ reducible. It is natural to ask whether all tripartite pure states are GHZ reducible. The answer of this question is not found yet, however we give some properties of the GHZ reducible tripartite pure states.
Let us first recall the definitions of the relative entropy of entanglement and Rains’ bound of entanglement. Let the systems A and B be in a joint state $`\rho _{AB}`$, the relative entropy of entanglement $`E_r(A,B)`$ is defined by
$`E_r(A,B)`$ $``$ $`E_r\left(\rho _{AB}\right)`$ (28)
$``$ $`\underset{\sigma \stackrel{~}{D}}{\mathrm{min}}tr_{AB}\left\{\rho _{AB}\left(\mathrm{log}_2\rho _{AB}\mathrm{log}_2\sigma \right)\right\}`$ (29)
where $`\stackrel{~}{D}`$ is the set of all disentangled states of the two systems A and B. Let $`\stackrel{~}{P}`$ be the set of all bipartite states that have positive partial transposes (PPT), similarly Rains’ bound of entanglement $`B_\mathrm{\Gamma }`$ is defined by
$`B_\mathrm{\Gamma }\left(\rho _{AB}\right)`$ $`\underset{\sigma \stackrel{~}{P}}{\mathrm{min}}tr_{AB}\left\{\rho _{AB}\left(\mathrm{log}_2\rho _{AB}\mathrm{log}_2\sigma \right)\right\}`$ (30)
It is obvious that $`B_\mathrm{\Gamma }\left(\rho _{AB}\right)E_r\left(\rho _{AB}\right)`$ since any separable state is a PPT state . Now we give the following proposition about the relative entropy of entanglement for a special kind of tripartite pure states.
Proposition 2: For the $`3`$-party pure state
$$|\mathrm{\Phi }_{ABC}=|\psi _{AB}^m|\phi _{AC}^n|\varphi _{BC}^l|\mathrm{\Theta }_{ABC}^k$$
(31)
where the state $`|\mathrm{\Theta }_{ABC}`$ is Schmidt decomposable, (i.e., $`|\mathrm{\Theta }_{ABC}=_i\sqrt{p_i}|i_A|i_B|i_C`$) there is
$`E_r(A,B)`$ $`=`$ $`mE_r\left(|\psi _{AB}\right)`$ (32)
$`E_r(A,C)`$ $`=`$ $`nE_r\left(|\phi _{AC}\right)`$ (33)
$`E_r(B,C)`$ $`=`$ $`lE_r\left(|\varphi _{BC}\right)`$ (34)
If the relative entropy of entanglement is additive, proposition 2 is obviously true. However, the additivity of the relative entropy of entanglement has not been proved yet (maybe it is not provable at all), so this proposition should be proved. A proof of this proposition can be found in Appendix A, here we prove this proposition by proving the following lemma.
Lemma 1: For a bipartite pure state $`\rho `$ and a bipartite separable state $`\rho ^{}`$, there is $`E_r\left(\rho \rho ^{}\right)=E_r\left(\rho \right)`$.
Proof. On one hand, it is obvious that
$$E_r\left(\rho \rho ^{}\right)E_r\left(\rho \right).$$
(35)
On the other hand, as a property of $`B_\mathrm{\Gamma }`$, there is
$$B_\mathrm{\Gamma }\left(\rho \rho ^{}\right)=B_\mathrm{\Gamma }\left(\rho \right)=E_r\left(\rho \right)$$
(36)
i.e.,
$$E_r\left(\rho \rho ^{}\right)B_\mathrm{\Gamma }\left(\rho \rho ^{}\right)=E_r\left(\rho \right)$$
(37)
Thus lemma 1 follows from eqs. (35) and (37). From lemma 1, and the additivity of the relative entropy of entanglement for pure states, proposition 2 can easily be proved.
LPSW’s lemma: If two $`3`$-party (Alice,Bob,Claire) quantum states $`|\mathrm{\Psi }`$ and $`|\mathrm{\Phi }`$ are LOCCa equivalent, then each of the relative entropies of entanglement of $`|\mathrm{\Psi }`$ is equal to the corresponding one of $`|\mathrm{\Phi }`$, i.e.,
$`E_r^{|\mathrm{\Psi }}(A,B)`$ $`=`$ $`E_r^{|\mathrm{\Phi }}(A,B)`$ (38)
$`E_r^{|\mathrm{\Psi }}(A,C)`$ $`=`$ $`E_r^{|\mathrm{\Phi }}(A,C)`$ (39)
$`E_r^{|\mathrm{\Psi }}(B,C)`$ $`=`$ $`E_r^{|\mathrm{\Phi }}(B,C)`$ (40)
This lemma follows from LPSW’s inequality that for any LOCC protocol, the average increase in $`E_r(B,C)`$ is no greater than the average decrease in the entanglement between Alice and the joint Bob-Claire system. Detailed discussion can be found in ref. . By this lemma, LPSW had made quantitative statements about tripartite entanglement, they notice that there are relations between the one-party entropies and relative entropies. Here we look more carefully into this issue and extract the relations of the entropies.
Proposition 3: If tripartite pure state $`|\mathrm{\Psi }_{ABC}`$ is GHZ reducible, then there must be
$`S\left(\rho _A\right)+E_r(B,C)`$ $`=`$ $`S\left(\rho _B\right)+E_r(A,C)`$ (41)
$`=`$ $`S\left(\rho _C\right)+E_r(A,B)`$ (42)
and
$$\begin{array}{c}S\left(\rho _A\right)E_r(A,B)+E_r(A,C)\\ S\left(\rho _B\right)E_r(A,B)+E_r(B,C)\\ S\left(\rho _C\right)E_r(A,C)+E_r(B,C)\end{array}$$
(43)
where $`S\left(\rho _A\right)`$ is the von Neumann entropy of the reduced density matrix of system A, and $`E_r(A,B)`$ is the relative entropy of entanglement of the systems A+B.
Proof. Since $`|\mathrm{\Psi }_{ABC}`$ is GHZ reducible, i.e.,
$$\begin{array}{ccc}& & |\mathrm{\Psi }_{ABC}\stackrel{LOCCa}{}|\mathrm{\Phi }_{ABC}|EPR_{AB}^{E_2(AB)}\hfill \\ & & |EPR_{AC}^{E_2(AC)}|EPR_{BC}^{E_2(BC)}|GHZ_{ABC}^{E_3(ABC)}\hfill \end{array}$$
(44)
From LPSW’s lemma and proposition 2, we have
$$\begin{array}{c}E_2\left(AB\right)=E_r(A,B)\\ E_2\left(AC\right)=E_r(A,C)\\ E_2\left(BC\right)=E_r(B,C)\end{array}$$
(45)
From eq. (45) and the additivity of the von Neumann entropy, it follows that
$$\begin{array}{c}S\left(\rho _A\right)=E_r(A,B)+E_r(A,C)+E_3\left(ABC\right)\\ S\left(\rho _B\right)=E_r(A,B)+E_r(B,C)+E_3\left(ABC\right)\\ S\left(\rho _C\right)=E_r(A,C)+E_r(B,C)+E_3\left(ABC\right)\end{array}$$
(46)
Since $`E_3\left(ABC\right)0`$, proposition 3 follows from eq. (46).
Eqs. (45) and (46) are also obtained in ref. . If we suppose that the relative entropy of entanglement is additive, then eqs. (45), (46) and proposition 3 are obvious results, however, here we have given a proof of these results without the assumption of additivity.
We do not know whether conditions (42) and (43) are satisfied by all tripartite pure states, but it can be shown that eq. (42) is satisfied for the following case.
Proposition 4: For the tripartite pure state $`|\mathrm{\Psi }_{ABC}`$, there are 3 reduced density matrixes of two parties, $`\rho _{AB}`$, $`\rho _{AC}`$ and $`\rho _{BC}`$, if at least two of them are separable states, then eq. (42) is satisfied.
Proof of proposition 4 is left to Appendix B.
## IV REDUCIBILITY OF TRIPARTITE PURE STATES AND ADDITIVITY OF THE RELATIVE ENTROPY OF ENTANGLEMENT
Let Alice (Bob) hold systems 1 and 3 (2 and 4), $`\rho _{12}`$ ($`\rho _{34}`$) be the joint state of the systems 1 and 2 (3 and 4), and let the systems 1+2 be uncorrelated with the systems 3+4, i.e., the overall state of the systems 1+2+3+4 can be written as
$$\rho _{AB}=\rho _{12}\rho _{34}$$
(47)
We would like to have the additivity
$$E_r(A,B)=E_r\left(\rho _{AB}\right)=E_r\left(\rho _{12}\right)+E_r\left(\rho _{34}\right)$$
(48)
as an important property desired from a measure of entanglement . The additivity has been proved for the case that both $`\rho _{12}`$ and $`\rho _{34}`$ are pure states , for more general cases, it remains a conjecture.
Proposition 5: The relative entropy of entanglement is additive if each of the two uncorrelated states (i.e., the above states $`\rho _{12}`$ and $`\rho _{34}`$) can be purified into a GHZ reducible tripartite pure state.
Proposition 5 says that, if there are two GHZ reducible tripartite pure states $`|\psi _{125}`$ and $`|\phi _{346}`$ such that
$`\rho _{12}`$ $`=`$ $`tr_5\left\{|\psi _{125}\psi |\right\}`$ (49)
$`\rho _{34}`$ $`=`$ $`tr_6\left\{|\phi _{346}\phi |\right\}`$ (50)
then eq. (48) is satisfied. This proposition follows directly from proposition 2. And we give the following proposition as a corollary.
Proposition 6: If all the tripartite pure states are GHZ reducible, then the relative entropy of entanglement is generally additive.
In another word, if we can find a counter-example for the additivity of the relative entropy of entanglement, then we can make the statement that not all tripartite pure states are GHZ reducible.
## V Conclusions
In the above discussions, it is shown that the set of generalized GHZ states is not a minimal reversible entanglement generating set, a MREGS for $`m`$-party pure states ($`m4`$) generally includes states other than the generalized GHZ states, for $`4`$-party pure states, there must be at least 12 member states in a MREGS.
For the GHZ reducible tripartite pure states, there are strong relations among the relative entropies of entanglement. And the additivity of the relative entropy of entanglement is shown to be a necessary condition for all the tripartite pure states to be GHZ reducible.
## VI ACKNOWLEDGMENTS
We thank Prof. C. H. Bennett and V. Vedral for valuable communications, and we also thank Prof. Wu Qiang, Dr. Hou Guang, Mr. Zhou Jindong, Huang minxing, Luo Yifan, Ms. Chen Xuemei for helpful discussions.
## VII Appendix A: Another Proof of proposition 2
Before the proof, we first state another lemma.
Lemma 2: For bipartite quantum state
$`\rho `$ $`=`$ $`{\displaystyle \underset{n_1n_2}{}}a_{n_1n_2}|\varphi _{n_1}\psi _{n_1}\varphi _{n_2}\psi _{n_2}|`$ (51)
$``$ $`{\displaystyle \underset{n_1n_2}{}}a_{n_1n_2}|\varphi _{n_1}_A\varphi _{n_2}||\psi _{n_1}_B\psi _{n_2}|`$ (52)
the relative entropy of entanglement is given by
$$E_r\left(\rho \right)=\underset{n}{}a_{nn}\mathrm{log}_2a_{nn}S\left(\rho \right)$$
(53)
where $`|\varphi _n`$ ($`|\psi _n`$) is a set of orthogonal normalized states of system A (B), $`S\left(\rho \right)tr_{AB}\left(\rho \mathrm{log}_2\rho \right)`$ is the von Neumann entropy.
This lemma is a extension of Vedral and Plenio’s theorem (Theorem 3 in ref. ), the proof is similar to that in ref. , details can be found in ref. , this lemma can also follow directly from Rains’ theorem 9 in ref. .
Now we come to the proof of proposition 2. The following pure states are written in their Schmidt decomposition form,
$$\begin{array}{c}|\psi _{AB}^m=_i\sqrt{p_i^\psi }|i^{A_1}|i^{B_1}\\ |\phi _{AC}^n=_i\sqrt{p_i^\phi }|i^{A_2}|i^{C_1}\\ |\varphi _{BC}^l=_i\sqrt{p_i^\varphi }|i^{B_2}|i^{C_2}\\ |\mathrm{\Theta }_{ABC}^k=_i\sqrt{p_i^\mathrm{\Theta }}|i^{A_3}|i^{B_3}|i^{C_3}\end{array}$$
(54)
where $`p_i^\alpha `$ ($`\alpha =\psi `$,$`\phi `$,$`\varphi `$,$`\mathrm{\Theta }`$) satisfy the normalization condition $`_ip_i^\alpha =1`$, the systems $`A_k`$ ($`B_k`$,$`C_k`$)($`k=1,2,3`$) are held by Alice (Bob, Claire). Since for pure states the relative entropy of entanglement is additive , we have
$$E_r\left(|\psi _{AB}^m\right)=mE_r\left(|\psi _{AB}\right)=\underset{i}{}p_i^\psi \mathrm{log}_2p_i^\psi $$
(55)
Set $`|\mathrm{\Psi }_1_{ABC}=|\psi _{AB}^m|\mathrm{\Theta }_{ABC}^k`$, then
$`|\mathrm{\Psi }_1_{ABC}`$ $`=`$ $`{\displaystyle \underset{i}{}}\sqrt{p_i^\psi }|i^{A_1}|i^{B_1}`$ (57)
$`{\displaystyle \underset{j}{}}\sqrt{p_j^\mathrm{\Theta }}|j^{A_3}|j^{B_3}|j^{C_3}`$
$`=`$ $`{\displaystyle \underset{ij}{}}\sqrt{p_i^\psi p_j^\mathrm{\Theta }}|ij_A|ij_B|j_C`$ (58)
therefore
$`\rho _{AB}^{|\mathrm{\Psi }_1}`$ $``$ $`tr_C\left\{|\mathrm{\Psi }_1_{ABC}\mathrm{\Psi }_1|\right\}`$ (59)
$`=`$ $`{\displaystyle \underset{ii^{^{}}j}{}}\sqrt{p_i^\psi p_i^{^{}}^\psi }p_j^\mathrm{\Theta }|ij_Ai^{^{}}j||ij_Bi^{^{}}j|`$ (60)
From lemma 2, it follows that
$`E_r\left(\rho _{AB}^{|\mathrm{\Psi }_1}\right)`$ $`=`$ $`{\displaystyle \underset{ij}{}}p_i^\psi p_j^\mathrm{\Theta }\mathrm{log}_2\left(p_i^\psi p_j^\mathrm{\Theta }\right)S\left(\rho _{AB}^{|\mathrm{\Psi }_1}\right)`$ (61)
$`=`$ $`{\displaystyle \underset{i}{}}p_i^\psi \mathrm{log}_2p_i^\psi {\displaystyle \underset{j}{}}p_j^\mathrm{\Theta }\mathrm{log}_2p_j^\mathrm{\Theta }`$ (63)
$`+{\displaystyle \underset{j}{}}p_j^\mathrm{\Theta }\mathrm{log}_2p_j^\mathrm{\Theta }`$
$`=`$ $`{\displaystyle \underset{i}{}}p_i^\psi \mathrm{log}_2p_i^\psi `$ (64)
$`=`$ $`mE_r\left(|\psi _{AB}\right)`$ (65)
We now come to prove that
$$E_r\left(\rho _{A_1B_1}\right)=E_r\left(\rho _{A_1B_1}\rho _{A_2}\rho _{B_2}\right)$$
(66)
It is known that
$`E_r\left(\rho _{A_1B_1}\rho _{A_2}\rho _{B_2}\right)`$ $``$ $`E_r\left(\rho _{A_1B_1}\right)+E_r\left(\rho _{A_2}\rho _{B_2}\right)`$ (67)
$`=`$ $`E_r\left(\rho _{A_1B_1}\right)`$ (68)
On the other hand, Alice and Bob can perform local unitary transformations and measurements to transform the state $`\rho _{A_1B_1}\rho _{A_2}\rho _{B_2}`$ into the state $`\rho _{A_1B_1}|0_{A_2}0||0_{B_2}0|`$, as the relative entropy of entanglement does not increase under LOCC, there is
$`E_r\left(\rho _{A_1B_1}\rho _{A_2}\rho _{B_2}\right)`$ $``$ $`E_r(\rho _{A_1B_1}|0_{A_2}0|`$ (70)
$`|0_{B_2}0|)`$
$`=`$ $`E_r\left(\rho _{A_1B_1}\right)`$ (71)
The last equality is true since there is no limit on the dimension of the Hilbert space for the systems held by Alice and Bob. Therefore eq. (66) follows from eq. (68) and (70).
From eq. (63) and (66), we have
$$E_r(A,B)=E_r\left(\rho _{AB}^{|\mathrm{\Phi }}\right)=E_r\left(\rho _{AB}^{|\mathrm{\Psi }_1}\right)=mE_r\left(|\psi _{AB}\right)$$
(72)
The other two equalities in proposition 2 follow from the symmetry of the state $`|\mathrm{\Phi }_{ABC}`$. Thus the proof of proposition 2 is completed.
## VIII Appendix B: Proof of proposition 4
Let $`\rho _{AB}`$ and $`\rho _{BC}`$ be separable states, then
$$\rho _{BC}=\underset{i=1}{\overset{M}{}}p_i|\psi _i^B\psi _i^B||\varphi _i^C\varphi _i^C|$$
(73)
where $`\epsilon =\left\{p_i\text{}|\psi _i^B\varphi _i^C|i=1,2,\mathrm{},M\right\}`$ is an ensemble of $`\rho _{BC}`$ with the fewest members.
Let us first show that, the states $`|\varphi _i^C`$ in eq. (73) can always be chosen to be orthogonal.
Alice appends an ancilla and performs a local unitary transformation on $`|\mathrm{\Psi }_{ABC}`$, resulting in
$$|\stackrel{~}{\mathrm{\Psi }}_{ABC}=\underset{i=1}{\overset{M}{}}\sqrt{p_i}|i^A\psi _i^B\varphi _i^C$$
(74)
where $`|i^A`$ is a set of orthogonal normalized states of Alice’s enlarged system. The Hughston-Joza-Wootters result ensures that this is always possible. The reduced density matrix
$`\stackrel{~}{\rho _{AB}}`$ $`=`$ $`tr_C\left(|\stackrel{~}{\mathrm{\Psi }}_{ABC}\stackrel{~}{\mathrm{\Psi }}|\right)`$ (75)
$`=`$ $`{\displaystyle \underset{i,j=1}{\overset{M}{}}}\sqrt{p_ip_j}\varphi _j^C|\varphi _i^C|i^Aj^A||\psi _i^B\psi _j^B|`$ (76)
is also a separable state, since local unitary transformation by Alice does not change the entanglement of the two systems A and B, i.e., $`\stackrel{~}{\rho _{AB}}`$ can be written as
$$\stackrel{~}{\rho _{AB}}=\underset{k}{}p_k\rho _k^A\rho _k^B$$
(77)
Let $`P_A`$ be any projection acting on the Hilbert space of system A. It is obvious that the state
$$\rho _P=\left(P_AI_B\right)\stackrel{~}{\rho _{AB}}\left(P_AI_B\right)$$
(78)
is also a separable state (except for a normalization factor). Set $`P_A=|m^Am^A|+|n^An^A|`$, therefore
$`\rho _P`$ $`=`$ $`p_m|m^Am^A||\psi _m^B\psi _m^B|`$ (82)
$`+p_n|n^An^A||\psi _n^B\psi _n^B|`$
$`+\sqrt{p_mp_n}\varphi _m^C|\varphi _n^C|n^Am^A||\psi _n^B\psi _m^B|`$
$`+\sqrt{p_mp_n}\varphi _n^C|\varphi _m^C|m^An^A||\psi _m^B\psi _n^B|`$
Let $`|\psi _n=\alpha |\psi _m+\beta |\psi _m^{}`$, where $`\beta |\psi _m^{}\left(|\psi _n\psi _m|\psi _n|\psi _m\right)`$ is orthogonal to $`|\psi _m`$. Let the states $`|\psi _m`$ and $`|\psi _m^{}`$ be the basis vectors for the Hilbert space of system B, the partial transpose of $`\rho _P`$ is written as
$$\left(\rho _P\right)^{T_B}=\left(\begin{array}{cccc}p_m& 0& K\alpha & 0\\ 0& 0& K\beta & 0\\ K^{}\alpha ^{}& K^{}\beta ^{}& p_n|\alpha |^2& p_n\alpha \beta ^{}\\ 0& 0& p_n\alpha ^{}\beta & p_n|\beta |^2\end{array}\right)$$
(83)
where $`K\sqrt{p_np_m}\varphi _n^C|\varphi _m^C`$. The separability of $`\rho _P`$ ensures the positivity of its partial transpose $`\left(\rho _P\right)^{T_B}`$ , this positivity requires
$$\varphi _n^C|\varphi _m^C=0\text{ or }\beta =0$$
(84)
i.e., for all $`ij`$, there is
$$|\varphi _j^C|\varphi _i^C\text{ or }|\psi _j^B=|\psi _i^B$$
(85)
If $`|\psi _j^B|\psi _i^B`$, we have that $`|\varphi _j^C|\varphi _i^C`$. And if $`|\psi _j^B=|\psi _i^B|\psi _k^B`$, we can always write
$`p_i|\varphi _i^C\varphi _i^C\left|+p_j\right|\varphi _j^C\varphi _j^C|`$ (86)
$`=p_i^{^{}}|\varphi _i^{}_{}{}^{}C\varphi _i^{}_{}{}^{}C\left|+p_j^{^{}}\right|\varphi _j^{}_{}{}^{}C\varphi _j^{}_{}{}^{}C|`$ (87)
where $`p_i^{^{}}+p_j^{^{}}=p_i+p_j`$ and $`|\varphi _j^{}_{}{}^{}C|\varphi _i^{}_{}{}^{}C`$, each of the two states $`|\varphi _j^{}_{}{}^{}C`$ and $`|\varphi _i^{}_{}{}^{}C`$ is a linear addition of the two states $`|\varphi _j^C`$ and $`|\varphi _i^C`$, so, $`|\varphi _j^{}_{}{}^{}C`$ and $`|\varphi _i^{}_{}{}^{}C`$ are also orthogonal to $`|\psi _k^B`$. That is to say, we can rewrite $`\rho _{BC}`$ as
$$\rho _{BC}=\underset{i=1}{\overset{M}{}}p_i^{^{}}|\psi _i^B\psi _i^B||\varphi _i^{}_{}{}^{}C\varphi _i^{}_{}{}^{}C|$$
(88)
where $`|\varphi _i^{}_{}{}^{}C`$ is a set of orthogonal normalized states of system C. Thus we prove that, the states $`|\varphi _i^C`$ in eq. (73) can always be chosen to be orthogonal.
Then Alice can append an ancilla and perform a local unitary transformation on $`|\mathrm{\Psi }_{ABC}`$ , resulting in
$$|\stackrel{~}{\stackrel{~}{\mathrm{\Psi }}}_{ABC}=\underset{i=1}{\overset{m}{}}\sqrt{p_i^{^{}}}|i^A\psi _i^Bi^C$$
(89)
where $`_ip_i^{^{}}=1`$ and $`|i^A`$ ($`|i^C|\varphi _i^{}_{}{}^{}C`$) is a set of orthogonal normalized states of system A (C), while $`|\psi _i^B`$ is a set of normalized states of system B, not necessarily orthogonal. We have that
$`\rho _{AC}^{^{}}`$ $`=`$ $`tr_B\left(|\stackrel{~}{\stackrel{~}{\mathrm{\Psi }}}_{ABC}\stackrel{~}{\stackrel{~}{\mathrm{\Psi }}}|\right)`$ (90)
$`=`$ $`{\displaystyle \underset{ij}{}}\sqrt{p_i^{^{}}p_j^{^{}}}\psi _j^B|\psi _i^B|i^Ai^Cj^Aj^C|`$ (91)
Since local unitary transformation do not change the relative entropy of entanglement as well as the von Neumann entropies, from lemma 2, we have
$`E_r(A,C)`$ $`=`$ $`E_r\left(\rho _{AC}\right)=E_r\left(\rho _{AC}^{^{}}\right)`$ (92)
$`=`$ $`{\displaystyle \underset{i}{}}p_i^{^{}}\mathrm{log}_2p_i^{^{}}S\left(\rho _{AC}\right)`$ (93)
$`=`$ $`H\left\{p_i^{^{}}\right\}S\left(\rho _B\right)`$ (94)
where $`H\left\{p_i^{^{}}\right\}_ip_i^{^{}}\mathrm{log}_2p_i^{^{}}`$. Since $`\rho _{AB}`$, $`\rho _{BC}`$ are separable states, there is
$$E_r(A,B)=E_r(B,C)=0$$
(95)
And
$`\rho _A^{^{}}`$ $`=`$ $`tr_C\left(\rho _{AC}^{^{}}\right)={\displaystyle \underset{i}{}}p_i^{^{}}|i^Ai^A|`$ (96)
$`\rho _C^{^{}}`$ $`=`$ $`tr_A\left(\rho _{AC}^{^{}}\right)={\displaystyle \underset{i}{}}p_i^{^{}}|i^Ci^C|`$ (97)
$`S\left(\rho _A\right)`$ $`=`$ $`S\left(\rho _A^{^{}}\right)={\displaystyle \underset{i}{}}p_i^{^{}}\mathrm{log}_2p_i^{^{}}H\left\{p_i^{^{}}\right\}`$ (98)
$`S\left(\rho _A\right)`$ $`=`$ $`S\left(\rho _C\right)=H\left\{p_i^{^{}}\right\}`$ (99)
Finally we get the result
$`S\left(\rho _A\right)+E_r(B,C)=S\left(\rho _B\right)+E_r(A,C)`$ (100)
$`=S\left(\rho _C\right)+E_r(A,B)=H\left\{p_i^{^{}}\right\}`$ (101)
|
warning/0004/hep-th0004017.html
|
ar5iv
|
text
|
# Untitled Document
hep-th/0004017
Two-Black-Hole Bound States
Ruth Britto-Pacumio<sup>1</sup> britto@boltzmann.harvard.edu, Andrew Strominger<sup>2</sup> andy@planck.harvard.edu and Anastasia Volovich<sup>3</sup> nastya@physics.harvard.edu <sup>4</sup> On leave from L. D. Landau Institute for Theoretical Physics, Moscow, Russia
Department of Physics
Harvard University
Cambridge, MA 02138
Abstract
The quantum mechanics of $`N`$ slowly-moving BPS black holes in five dimensions is considered. A divergent continuum of states describing arbitrarily closely bound black holes with arbitrarily small excitation energies is found. A superconformal structure appears at low energies and can be used to define an index counting the weighted number of supersymmetric bound states. It is shown that the index is determined from the dimensions of certain cohomology classes on the symmetric product of $`N`$ copies of $`\mathrm{IR}^4`$. An explicit computation for the case of $`N=2`$ with no angular momentum yields a finite nonzero result.
1. Introduction
The low energy dynamics of $`N`$ near-coincident BPS black holes in five dimensions is described by superconformal quantum mechanics \[1--4\]. (See also \[5--10\].) The states of this theory describe black holes whose separations and excitation energies go to zero in the infrared scaling limit. Hence every state of this low energy theory describes a marginally bound state of the $`N`$ black holes.
The existence of infinitely many bound states is puzzling. In this paper we compute a supersymmetric index which counts a weighted number of bound states (roughly the difference between the numbers of hypermultiplets and vector multiplets). This index, in contrast, turns out to be finite and nonvanishing. After factoring out center-of-mass degrees of freedom, the index essentially reduces to the Witten index of the black hole quantum mechanics:
$$_{BH}(j_L)Tr()^{2J_R^3}|_{j_L}.$$
In this expression $`J_R`$ is the generator of $`SU(2)_R`$ spatial rotations and is a generator of the superconformal algebra. $`SU(2)_L`$ spatial rotations commute with the superconformal algebra, and the trace in (1.1) is at fixed total $`j_L`$. Ordinarily this would be computed as a trace over eigenstates of the hamiltonian $`H`$. This trace is ill-defined for superconformal quantum mechanics because of the infrared continuum. (The usual method of putting the system in a box does not work here because the continuum arises from near coincident black holes.) We define the index by tracing instead over eigenstates of $`L_0=\frac{1}{2}(H+K)`$, where $`K`$ is the generator of special conformal transformations. For the case of two black holes we find that $`_{BH}(0)=1`$. For general $`N`$ we relate the index to the counting of a certain type of noncompact cohomology class in the symmetric (for identical black holes) product of $`N`$ copies of $`\mathrm{IR}^4`$. At higher $`N`$ there appears to be a rich structure of supersymmetric bound states whose elucidation we defer to later work.
On the way to studying the $`N`$-black hole problem we describe some general properties of the Hilbert space of superconformal quantum mechanics. We begin with the simplest case with two supersymmetries and work up to black holes. In section 2 we relate the spectrum of an $`Osp(1|2)`$ sigma model to the eigenvalues of a certain Dirac operator on the target space. In section 3 we show that states of $`SU(1,1|1)`$ superconformal quantum mechanics are naturally viewed as $`(p,0)`$-forms, but with a non-canonical measure, and we derive the chiral primary condition. In section 4 we describe the $`D(2,1;0)`$ quantum mechanics. In section 5 we explicitly compute the index $`_{BH}(0)`$ for two black holes.
2. $`Osp(1|2)`$
In this section we relate the spectrum of an $`Osp(1|2)`$ sigma model to the eigenvalues of a certain Dirac operator with torsion on the target space.
2.1. Symmetries
The $`Osp(1|2)`$ superalgebra is generated by the three bosonic operators $`H`$, $`K`$ and $`D`$ and two fermionic operators $`Q`$ and $`S`$. The nonvanishing commutation relations are
$$\begin{array}{ccccccccccccc}[H,K]& =& iD,& & & [H,D]& =& 2iH,& & & [K,D]& =& 2iK,\\ \\ \{Q,Q\}& =& 2H,& & & [Q,D]& =& iQ,& & & [Q,K]& =& iS,\\ \\ \{S,S\}& =& 2K,& & & [S,D]& =& iS,& & & [S,H]& =& iQ,\\ \\ \{S,Q\}& =& D.\end{array}$$
Alternatively one may define (suppressing an arbitrary dimensionful constant)
$$\begin{array}{cc}\hfill G_{\pm {\scriptscriptstyle \frac{1}{2}}}& =\frac{1}{\sqrt{2}}(QiS),\hfill \\ \hfill L_0& =\frac{1}{2}(H+K),\hfill \\ \hfill L_{\pm 1}& =\frac{1}{2}(HKiD),\hfill \end{array}$$
in terms of which the algebra becomes
$$[L_m,G_r]=\frac{m2r}{2}G_{m+r},$$
$$\{G_r,G_s\}=2L_{r+s},$$
$$[L_m,L_n]=(mn)L_{m+n},$$
where $`r,s=\pm \frac{1}{2}`$ and $`m,n=0,\pm 1.`$
The $`Osp(1|2)`$ algebra can be realized with the supermultiplet $`(X^M,\lambda ^M)`$, where $`\lambda ^M=\lambda ^M`$. The nonvanishing commutation relations for these fields and their conjugate momenta are, in the notation of ,
$$\begin{array}{cc}\hfill \{\lambda ^M,\lambda ^N\}& =g^{MN},\hfill \\ \hfill [P_M,X^N]& =i\delta _M^N,\hfill \\ \hfill [P_M,\lambda ^N]& =i(\mathrm{\Gamma }_{MP}^N\omega _{MP}^N)\lambda ^P,\hfill \end{array}$$
with $`\omega `$ the spin connection and $`\mathrm{\Gamma }`$ the Christoffel connection. The last relation is necessitated by the fact that the commutator of $`\lambda `$ with itself depends on $`X`$, and implies that $`e_M^\alpha \lambda ^M`$, where $`e`$ is the vielbein, commutes with $`P`$. Explicit expressions for the supercharges are then
$$Q=\lambda ^MP_M+\frac{i}{6}(c_{MNP}3\omega _{MNP})\lambda ^M\lambda ^N\lambda ^P,$$
and
$$S=\lambda ^MD_M.$$
The algebra requires that the vector field $`D`$ is a so-called closed homothety obeying
$$\begin{array}{cc}\hfill _Dg_{MN}& =2g_{MN},\hfill \\ \hfill d(D_MdX^M)& =0,\hfill \end{array}$$
and the torsion $`c`$ obeys
$$\begin{array}{cc}\hfill D^Mc_{MNP}& =0,\hfill \\ \hfill _Dc_{MNP}& =2c_{MNP}.\hfill \end{array}$$
A derivation of these results can be found in .
2.2. Quantum States
States $`|\psi `$ in the Hilbert space form a representation of the algebra of fermions. From the commutation relations (2.1), we conclude that the states are target space spinors, with the fermions acting as
$$\lambda ^M|\psi =\frac{1}{\sqrt{2}}\gamma ^M|\psi ,$$
where $`\gamma ^M`$ are the usual $`SO(N)`$ gamma matrices.
The states can be organized into infinite-dimensional superconformal multiplets. At the bottom of every multiplet is a superprimary state obeying
$$G_{{\scriptscriptstyle \frac{1}{2}}}|\psi =0.$$
The remaining tower of states is generated by the action of $`G_{{\scriptscriptstyle \frac{1}{2}}}`$. Superprimary states are related by a similarity transformation to states $`|\psi ^{}=e^K|\psi `$ obeying
$$Q|\psi ^{}=0.$$
Using (2.1), (2.1) becomes the modified Dirac equation
$$i(\gamma ^M_M\frac{1}{12}\gamma _M^{NP}c_{NP}^M)|\psi ^{}=0.$$
In order to better understand (2.1), we introduce the conformally related metric
$$d\stackrel{~}{s}^2=K^1ds^2.$$
It follows from (2.1) that the vector field $`D`$ is covariantly constant in this metric. Hence coordinates can be chosen so that it takes the simple product form
$$d\stackrel{~}{s}^2=2(dX^0)^2+\widehat{g}_{IJ}dX^IdX^J,$$
where the $`(N1)`$-dimensional metric $`\widehat{g}`$ on the space $`M^{N1}`$ transverse to the orbits of $`D`$ is independent of $`X^0`$. In these coordinates, $`D^M_M=_0`$ and $`K=\frac{1}{2}e^{2X^0}`$. Equation (2.1) then reduces to
$$i(\gamma ^I\widehat{}_I\frac{1}{12}\gamma _I^{JK}c_{JK}^I)|\psi ^{}=i\gamma ^0(_0+\frac{N1}{2})|\psi ^{}.$$
The left hand side of (2.1) is a Dirac equation with torsion on the transverse space $`M^{N1}`$. Let $`\lambda _j`$ denote the (real) spectrum of this Dirac operator and $`\widehat{\psi }_j`$ the corresponding orthonormal eigenspinors. In an even-dimensional space, there is a basis in which
$$\gamma ^I=\left(\begin{array}{cc}0& \gamma ^I\\ \gamma ^I& 0\end{array}\right),\gamma ^0=\left(\begin{array}{cc}0& i\mathrm{𝟏}\\ i\mathrm{𝟏}& 0\end{array}\right),$$
and a solution of the Dirac equation can be written as
$$|\psi _i^{}=\left(\genfrac{}{}{0pt}{}{\psi _U^{}}{\psi _L^{}}\right)=\left(\genfrac{}{}{0pt}{}{\widehat{\psi }_ie^{(\frac{N1}{2}+\lambda _i)X^0}}{\widehat{\psi }_ie^{(\frac{N1}{2}\lambda _i)X^0}}\right).$$
Superprimaries obeying (2.1) are obtained by a similarity transformation as
$$|\psi _i=\left(\genfrac{}{}{0pt}{}{\psi _U}{\psi _L}\right)=\left(\genfrac{}{}{0pt}{}{\widehat{\psi }_ie^{\frac{1}{2}e^{2X^0}(\frac{N1}{2}+\lambda _i)X^0}}{\widehat{\psi }_ie^{\frac{1}{2}e^{2X^0}(\frac{N1}{2}\lambda _i)X^0}}\right).$$
Normalizability then requires convergence of the integral
$$\psi _j|\psi _k=d^NX\sqrt{g}\psi _j^{}\psi _k=\delta _{jk}2^{\frac{1N}{2}}𝑑X^0(2\mathrm{sinh}2\lambda _jX^0)e^{[X^0e^{2X^0}]},$$
which is equivalent to the condition that $`|\lambda _j|<\frac{1}{2}.`$ We conclude that there is one superprimary for every such normalizable solution of the Dirac equation with torsion on the transverse space $`M^{N1}`$.
3. $`SU(1,1|1)`$
In this section we show that states of $`SU(1,1|1)`$ superconformal quantum mechanics are naturally viewed as $`(p,0)`$-forms, but with a non-canonical measure. The chiral primary condition is derived and expressed as a condition on forms.
3.1. Symmetries
The $`SU(1,1|1)`$ algebra is
$$[L_m,L_n]=(mn)L_{m+n}$$
$$\{G_r,\overline{G}_s\}=2L_{r+s}+2J(rs)\delta _{r,s}$$
$$[J,G_r]=\frac{1}{2}G_r,[J,\overline{G}_r]=\frac{1}{2}\overline{G}_r$$
$$[L_m,G_r]=\frac{m2r}{2}G_{m+r},[L_m,\overline{G}_r]=\frac{m2r}{2}\overline{G}_{m+r}$$
where $`m,n=0,\pm 1`$ and $`r,s=\pm \frac{1}{2}`$. As shown in , the $`Osp(1|2)`$ model of the previous section has this larger symmetry if and only if there is a complex structure preserved by the action of $`D`$, the metric is hermitian, and the $`(1,2)`$ part of the torsion is given by
$$c_{\overline{b}c}^{\overline{a}}=\mathrm{\Gamma }_{\overline{b}c}^{\overline{a}}.$$
Indices in lower (upper) case denote complex (real) coordinates, so that $`a,b=1,2,\mathrm{},n`$, where $`n`$ is the complex dimension. In general $`c`$ may also have $`(0,3)`$ and $`(3,0)`$ parts unconstrained by (3.1). These are constrained to vanish for $`𝒩=4`$ supersymmetry, and for simplicity we set them to zero here. The relation (3.1) then implies that there is a $`U(n)`$ connection,
$$\mathrm{\Omega }_{MP}^N\mathrm{\Gamma }_{MP}^N+c_{MP}^N.$$
Let us first collect some formulae from describing the $`SU(1,1|1)`$ theories.<sup>5</sup> Our notation is the same as in with the following exceptions. We use capital indices $`M,N,P`$ for the $`2n`$-dimensional moduli space coordinates. The indices $`A,B`$ are used only for the black holes themselves. In this section, $`Q`$ and $`S`$ are the holomorphic supercharges that were called $`𝒬=\frac{1}{2}(Qi\stackrel{~}{Q})`$ and $`𝒮=\frac{1}{2}(Si\stackrel{~}{S})`$ in . In section 4, $`R`$ and $`J_R`$ are the angular momentum operators corresponding to $`R_+`$ and $`R_{}`$ in . It is convenient to define the shifted momentum
$$\mathrm{\Pi }_M=P_M\frac{i}{2}(\omega _{MNP}c_{MNP})\lambda ^N\lambda ^P,$$
which has the property
$$[\mathrm{\Pi }_M,\lambda ^N]=i\mathrm{\Omega }_{MP}^N\lambda ^P.$$
Defining $`Q`$, $`S`$ and their hermitian conjugates $`\overline{Q}`$, $`\overline{S}`$ by
$$\begin{array}{cc}\hfill G_{\pm {\scriptscriptstyle \frac{1}{2}}}& =\frac{1}{\sqrt{2}}(QiS),\hfill \\ \hfill \overline{G}_{\pm {\scriptscriptstyle \frac{1}{2}}}& =\frac{1}{\sqrt{2}}(\overline{Q}i\overline{S}),\hfill \end{array}$$
and
$$c_a=c_{}^{b}{}_{ba}{}^{},$$
one has
$$\begin{array}{cc}\hfill Q& =\lambda ^a(\mathrm{\Pi }_aic_{ab\overline{c}}\lambda ^b\lambda ^{\overline{c}}ic_a),\hfill \\ \hfill S& =\lambda ^aD_a.\hfill \end{array}$$
We note that
$$\begin{array}{cc}\hfill \{Q,\lambda ^a\}& =0,\hfill \\ \hfill \{Q,\lambda ^{\overline{a}}\}& =g^{\overline{a}b}(\mathrm{\Pi }_bic_b).\hfill \end{array}$$
3.2. The Ground State of $`H`$
In this subsection we construct a state $`|\eta `$ annihilated by the supercharges $`Q`$ and $`\overline{Q}`$ as well as the Hamiltonian $`H`$. We compute the norm of $`|\eta `$ and its $`U(1)`$ charge.
We begin by imposing the chirality conditions
$$\gamma ^{\overline{a}}|\eta =\gamma _a|\eta =0,$$
which states that $`|\eta `$ is a singlet under the $`SU(n)`$ subgroup of $`U(n)`$. It also implies that $`|\eta `$ is annihilated by $`\overline{Q}`$ (as well as $`\overline{S}`$). The action of the supercharge $`Q`$ on a general state $`|\psi `$ is
$$Q|\psi =i\gamma ^a(𝒟_a+c_{ab\overline{c}}\gamma ^b\gamma ^{\overline{c}}+c_a)|\psi ,$$
where $`𝒟`$ is the covariant derivative on spinors with the $`U(n)`$ spin connection associated to $`\mathrm{\Omega }`$. Given (3.1), the state $`|\eta `$ will be annihilated by $`Q`$ if the wavefunction $`\eta (X)`$ is a covariantly constant spinor obeying
$$(𝒟_a+c_a)\eta =0.$$
This has solutions because the integrability conditions $`[𝒟_a+c_a,𝒟_b+c_b]\eta =2c_{ab}^c(𝒟_c+c_c)\eta `$ are satisfied, as can be checked from the expression (3.1) for the connection. The solution is fixed up to transformations of the form
$$\eta f(X^{\overline{a}})\eta ,$$
where $`f`$ is an antiholomorphic function. We will fix this freedom shortly.
We now describe some properties of $`|\eta `$. Define the $`U(1)`$ connection $`A`$ by
$$A_N=\mathrm{\Omega }_{NP}^MI_M^P=i\mathrm{\Omega }_{Na}^ai\mathrm{\Omega }_{N\overline{a}}^{\overline{a}},$$
where $`I`$ is the complex structure. $`A`$ obeys
$$A_b=4ic_b+2i_b\varphi ,$$
with
$$\varphi =\frac{1}{4}\mathrm{ln}detg.$$
Under an infinitesimal complex coordinate transformation
$$X^aX^a+\zeta ^a(X^b),$$
one finds
$$\begin{array}{cc}\hfill \varphi & \varphi +\frac{i}{2}(ϵ\overline{ϵ})\hfill \\ \hfill A& Ad(ϵ+\overline{ϵ}),\hfill \end{array}$$
with complex holomorphic gauge parameter
$$ϵ=i_a\zeta ^a.$$
Using the fact that the spinorial wavefunction $`\eta `$ is an $`SU(n)`$ (but not a $`U(1)`$) singlet, the equation (3.1) can be written, using (3.1), as
$$(_a\frac{i}{4}A_a+c_a)\eta =(_a+\frac{1}{2}_a\varphi )\eta =0.$$
The solution of this equation is
$$\eta =e^{\frac{\varphi }{2}+\overline{\mathrm{\Theta }}}\eta _0,$$
where $`\overline{\mathrm{\Theta }}`$ is an arbitrary antiholomorphic function, and $`\eta _0`$ is the constant spinor obeying (3.1). The norm of $`\eta `$ is then
$$\eta ^{}\eta =e^{\varphi +\mathrm{\Theta }+\overline{\mathrm{\Theta }}}.$$
The expression (3.1) is coordinate invariant because $`\mathrm{\Theta }+\overline{\mathrm{\Theta }}`$ and $`\varphi `$ transform the same way under (3.1). In the following it will be convenient to choose coordinates so that $`\varphi `$ is nonsingular at smooth points in the geometry. (In the $`𝒩=4`$ case, such coordinates are singled out by the existence of a quaternionic structure.) The norm (3.1) is then in general singular, except if the freedom (3.1) is used to shift away $`\mathrm{\Theta }`$ altogether. We shall henceforth assume that this has been done, so that in nonsingular coordinates
$$\eta ^{}\eta =e^\varphi .$$
The $`U(1)`$ $`R`$-charge is measured by the operator
$$J=\frac{1}{2}\left(iD^a\mathrm{\Pi }_aiD^{\overline{a}}\mathrm{\Pi }_{\overline{a}}+g_{a\overline{b}}(\lambda ^a\lambda ^{\overline{b}}\lambda ^{\overline{b}}\lambda ^a)+D^ac_{aMN}\lambda ^M\lambda ^ND^{\overline{a}}c_{\overline{a}MN}\lambda ^M\lambda ^N\right).$$
Acting on $`|\eta `$, one finds
$$J|\eta =\frac{1}{2}(D^{\overline{a}}_{\overline{a}}\varphi n)|\eta .$$
In dilational gauge, defined by
$$D^M=\frac{2}{h}X^M,$$
for some constant $`h`$, one has
$$D^{\overline{a}}_{\overline{a}}\varphi =\frac{n(h2)}{2h},$$
and
$$J|\eta =n\frac{h+2}{4h}|\eta ,$$
where $`n`$ is the complex dimension of the target space.
3.3. Chiral Primaries and $`(p,0)`$-Forms
In the previous subsection we constructed a supersymmetric ground state $`|\eta `$ of the Hamiltonian $`H`$. In general this state may not be normalizable due to the noncompact regions of the target space. $`L_0`$ eigenstates will in some cases be normalizable. In this subsection we build such states by acting on $`|\eta `$ with bosonic and fermionic operators.
As in the case of $`Osp(1|2)`$, $`L_0`$ eigenstates lie in superconformal representations containing lowest weight states annihilated by $`G_{{\scriptscriptstyle \frac{1}{2}}}`$ and $`\overline{G}_{{\scriptscriptstyle \frac{1}{2}}}`$. For $`SU(1,1|1)`$ there are special representations whose lowest weight states are also annihilated by $`G_{{\scriptscriptstyle \frac{1}{2}}}`$ (or $`\overline{G}_{{\scriptscriptstyle \frac{1}{2}}}`$). These are chiral (or antichiral) primary states.
Consider a similarity transformation from $`|\eta `$ to the state
$$|0=e^K|\eta .$$
Using
$$G_{{\scriptscriptstyle \frac{1}{2}}}=\frac{1}{\sqrt{2}}e^KQe^K,$$
and (3.1), this state is seen to obey
$$G_{{\scriptscriptstyle \frac{1}{2}}}|0=\overline{G}_{\pm {\scriptscriptstyle \frac{1}{2}}}|0=0.$$
Hence it is a lowest weight antichiral primary. We shall see later that this state is not normalizable in the black hole case.
The most general state is of the form
$$|f_p=f_p|0,$$
with
$$f_p=\frac{1}{p!}f_{a_1a_2\mathrm{}a_p}\lambda ^{a_1}\lambda ^{a_2}\mathrm{}\lambda ^{a_p},$$
where $`f_{a_1a_2\mathrm{}a_p}`$ is totally antisymmetric. Hence the Hilbert space can be identified with the space of $`(p,0)`$-forms on the target space. The action of the supercharges on these states is<sup>6</sup> The use of a capital index in the second equation produces one extra term. For example, $`^Mf_M=g^{a\overline{b}}(_{\overline{b}}f_a2\mathrm{\Gamma }_{\overline{b}a}^cf_c)`$, while $`^af_a=g^{a\overline{b}}(_{\overline{b}}f_a\mathrm{\Gamma }_{\overline{b}a}^cf_c)`$.
$$(QiS)f_p|0=\frac{i}{p!}(_{a_1}f_{a_2\mathrm{}a_{p+1}})\lambda ^{a_1}\lambda ^{a_2}\mathrm{}\lambda ^{a_{p+1}}|0,$$
$$(\overline{Q}i\overline{S})f_p|0=\frac{i}{(p1)!}(e^\varphi ^Me^\varphi f_{Ma_2\mathrm{}a_p})\lambda ^{a_2}\mathrm{}\lambda ^{a_p}|0,$$
$$Sf_p|0=\frac{1}{p!}D_{a_1}f_{a_2\mathrm{}a_{p+1}}\lambda ^{a_1}\mathrm{}\lambda ^{a_{p+1}}|0,$$
$$\overline{S}f_p|0=\frac{1}{(p1)!}D^{a_1}f_{a_1a_2\mathrm{}a_p}\lambda ^{a_2}\mathrm{}\lambda ^{a_p}|0.$$
Regarding $`f_p`$ as a $`(p,0)`$-form, such states are chiral primary if and only if
$$f_p=Df_p=\overline{}e^\varphi f_p=0,$$
where $``$ is the Hodge dual and here $`D=D_adX^a`$ is a $`(1,0)`$-form.
Note that the inner product,
$$f_p^{}|f_p=\frac{1}{p!}d^{2n}x\sqrt{g}e^{\varphi 2K}\overline{f}^{a_1a_2\mathrm{}a_p}f_{a_1a_2\mathrm{}a_p},$$
contains extra factors in the integration measure.
3.4. $`Tr()^{2J}`$
In this subsection we consider the Witten index
$$\chi =Tr()^{2J}.$$
In order to compute the trace one must choose a basis for the Hilbert space and a regulator for the infinite sum. One might attempt to define the trace as the limit of a weighted sum over $`H`$ eigenstates
$$\chi =Tr_H()^{2J}e^{\beta H},\beta 0.$$
Of course the sum is actually independent of $`\beta `$, since states with nonzero $`H`$ come in bose-fermi pairs which cancel. The trace (3.1) is nevertheless difficult to evaluate because of the continuum of eigenstates extending down to zero energy. An alternate way to define the index is as a weighted sum over $`L_0`$ eigenstates<sup>7</sup> In 1+1 dimensions, (3.45) is exactly the expression obtained for the Witten index in the NS sector obtained by spectral flow from the R sector. In 0+1 there is no obvious analog of spectral flow.
$$\chi =Tr_{L_0}()^{2J}e^{\beta (L_0J)},\beta 0.$$
The sum is also independent of $`\beta `$. This follows from considering the operator
$$\stackrel{~}{Q}=\frac{1}{\sqrt{2}}(\overline{G}_{{\scriptscriptstyle \frac{1}{2}}}+G_{{\scriptscriptstyle \frac{1}{2}}})$$
with the properties
$$\stackrel{~}{Q}^2=L_0J,\{\stackrel{~}{Q},()^{2J}\}=0.$$
States with nonzero $`L_0J`$ come in bose-fermi pairs generated by $`\stackrel{~}{Q}`$ and cancel in the sum (3.1). Hence $`\chi `$ receives contributions only from states with $`L_0J=0`$. Such states are chiral primaries annihilated by $`\stackrel{~}{Q}`$, or equivalently, states annihilated by $`\overline{G}_{{\scriptscriptstyle \frac{1}{2}}}`$ and $`G_{\pm {\scriptscriptstyle \frac{1}{2}}}`$. Hence the Witten index can be computed as a weighted sum over superconformal chiral primaries.
A more general index can be defined when the theory contains an operator $`𝒪`$, usually associated with a symmetry, which commutes with the generators of the superalgebra. In that case the preceding argment may be repeated to show that
$$\chi _𝒪=Tr_{L_0}()^{2J}𝒪e^{\beta (L_0J)},\beta 0$$
also defines an index. Alternatively the index can be restricted to a sum over the eigenstates of $`𝒪`$ with eigenvalue $`\lambda `$
$$\chi (\lambda )=Tr_{L_0}()^{2J}e^{\beta (L_0J)}|_\lambda ,\beta 0$$
4. $`D(2,1;0)`$
In this section we describe $`D(2,1;0)`$ quantum mechanics in terms of its symmetries and chiral primaries. This algebra contains an $`SU(1,1|1)`$ subalgebra and will be used in section 5 to describe the theory of $`N`$ BPS black holes.
4.1. Symmetries
$`D(2,1;0)`$ is the semidirect product of $`SU(1,1|2)`$ and $`SU(2)_R`$. In $`4k`$ dimensions the target space geometry has a triplet of self-dual complex structures obeying
$$I^rI^s=\delta ^{rs}+ϵ^{rst}I^t,$$
for $`r,s=1,2,3`$. There are also isometries generated by
$$D^{rM}=D^NI_N^{rM},$$
whose associated charges generate the $`R`$-symmetry in $`SU(1,1|2)`$. This $`R`$-symmetry will be denoted by $`SU(2)_{\mathrm{Right}}`$. A large class of $`D(2,1;0)`$ theories are sigma models with target space metrics
$$g_{a\overline{b}}=\frac{1}{2}(_a_{\overline{b}}L+I_a^{\overline{c}}I_{\overline{b}}^{+d}_{\overline{c}}_dL).$$
In this expression, $`L`$ is a homogeneous function of degree $`2`$, the complex coordinates are adapted to $`I^3`$, and $`I^\pm =\frac{1}{2}(I^1\pm iI^2)`$. All three complex structures are constant self-dual matrices, and $`D^M=X^M`$.
It follows from (4.1) that
$$_a\varphi =2c_a.$$
Hence for $`D(2,1;0)`$ the $`U(1)`$ connection $`A`$ in (3.1) vanishes, and $`\mathrm{\Omega }`$ is an $`SU(n)`$ (rather than $`U(n)`$) connection. The holomorphic two-form $`I^{}`$ obeys the relations
$$I^{}=0,\overline{}e^\varphi I^{}=0.$$
There are 8 supercharges in $`SU(1,1|2)`$,
$$G_{\pm {\scriptscriptstyle \frac{1}{2}}}^{\pm \pm }=\frac{1}{\sqrt{2}}(Q^{\pm \pm }iS^{\pm \pm }).$$
Explicit expressions for four of these are
$$\begin{array}{cc}\hfill G_{\pm {\scriptscriptstyle \frac{1}{2}}}^+& =\lambda ^a(\mathrm{\Pi }_aic_{ab\overline{c}}\lambda ^b\lambda ^{\overline{c}}ic_aiD_a),\hfill \\ \hfill G_{\pm {\scriptscriptstyle \frac{1}{2}}}^+& =\lambda ^{\overline{a}}(\mathrm{\Pi }_{\overline{a}}ic_{\overline{a}b\overline{c}}\lambda ^b\lambda ^{\overline{c}}ic_{\overline{a}}iD_{\overline{a}}).\hfill \end{array}$$
These charges are $`SU(2)_{\mathrm{Right}}\times SU(2)_R`$ doublets. The $`SU(2)_{\mathrm{Right}}`$ ($`SU(2)_R`$) spin is indicated by the first (second) superscript. The other four supercharges can accordingly be expressed as
$$G_{\pm {\scriptscriptstyle \frac{1}{2}}}^{++}=i[J_R^+,G_{\pm {\scriptscriptstyle \frac{1}{2}}}^+]=i[R^+,G_{\pm {\scriptscriptstyle \frac{1}{2}}}^+],$$
$$G_{\pm {\scriptscriptstyle \frac{1}{2}}}^{}=i[J_R^{},G_{\pm {\scriptscriptstyle \frac{1}{2}}}^+]=i[R^{},G_{\pm {\scriptscriptstyle \frac{1}{2}}}^+].$$
Some important commutators are
$$\begin{array}{cc}\hfill \{G_p^{\alpha \alpha ^{}},G_q^{\beta \beta ^{}}\}& =2\delta ^{\alpha ,\beta }\delta ^{\alpha ^{},\beta ^{}}L_{p+q}+2(pq)J_R^{\alpha \beta }\delta ^{\alpha ^{},\beta ^{}},\hfill \\ \hfill [J_R^3,G_p^{\alpha \alpha ^{}}]& =\frac{\alpha }{2}G_p^{\alpha \alpha ^{}},\hfill \\ \hfill [R^3,G_p^{\alpha \alpha ^{}}]& =\frac{\alpha ^{}}{2}G_p^{\alpha \alpha ^{}},\hfill \end{array}$$
with $`p,q=\pm \frac{1}{2}`$, $`\alpha \beta =\pm `$ and $`J_R^+=J_R^+=J_R^3`$ etc. $`J_R`$ generates $`SU(2)_{\mathrm{Right}}`$ rotations of the bosons $`X^A`$,
$$[J_R^r,\lambda ^M]=0,[J_R^r,X^M]=\frac{i}{2}X^NI_N^{rM}.$$
The second $`SU(2)_R`$, which does not lie in $`SU(1,1|2)`$, generates $`R`$-symmetry transformations of the fermions.<sup>8</sup> $`(R,J_R)`$ are the operators $`(R_+,R_{})`$ of .
$$[R^r,\lambda ^M]=\frac{i}{2}\lambda ^NI_N^{rM},[R^r,X^M]=0.$$
The supercharges transform in the $`(2,2)`$ of $`SU(2)_{\mathrm{Right}}\times SU(2)_R`$.
4.2. Chiral Primaries
Chiral primaries in these theories are defined by embedding an $`SU(1,1|1)`$ subalgebra in $`D(2,1;0)`$. A natural embedding is defined by
$$G_{\pm {\scriptscriptstyle \frac{1}{2}}}=G_{\pm {\scriptscriptstyle \frac{1}{2}}}^+,\overline{G}_{\pm {\scriptscriptstyle \frac{1}{2}}}=G_{\pm {\scriptscriptstyle \frac{1}{2}}}^+,J=J_R^3.$$
With this embedding, the conditions for a chiral primary are given by (3.1), in the complex structure defined by $`I^3`$.
It follows from the algebra that normalizable chiral primaries annihilated by $`\overline{G}_{{\scriptscriptstyle \frac{1}{2}}}`$ and $`G_{\pm {\scriptscriptstyle \frac{1}{2}}}`$ as in (3.1) are annihilated by six of the eight supercharges
$$G_{\pm {\scriptscriptstyle \frac{1}{2}}}^{++},G_{\pm {\scriptscriptstyle \frac{1}{2}}}^+,G_{{\scriptscriptstyle \frac{1}{2}}}^+,G_{{\scriptscriptstyle \frac{1}{2}}}^{}.$$
Antichiral primaries are annihilated by the complementary set of negatively moded supercharges. Note that $`SU(2)_R`$ does not mix the operators in (4.1) with $`G_{{\scriptscriptstyle \frac{1}{2}}}^\pm `$. Therefore if a chiral primary is not an $`SU(2)_R`$ singlet, then the $`SU(2)_R`$ action gives new chiral primaries. The algebra further implies that all chiral primaries are highest weight $`SU(2)_{\mathrm{Right}}`$ states annihilated by $`J_R^+`$.
We note that the state $`|0`$ is annihilated by all the positively moded supercharges as well as $`G_{{\scriptscriptstyle \frac{1}{2}}}^{\pm +}`$. Hence it is neither a chiral nor an antichiral primary. This implies it cannot be a normalizable state, as indeed can be seen from explicit computation. Since $`h=2`$ for $`D(2,1;0)`$, it follows from (3.1) that $`J_R^3|0=0`$, while $`R^3|0=\frac{n}{4}|0`$.
5. Black Hole Quantum Mechanics
At low energies the quantum mechanics of $`N`$ black holes is described by the product of a free theory containing the center-of-mass coordinates and an interacting near-horizon superconformal theory with a $`4(N1)`$-dimensional target space. In this section we describe this theory and the index $`_{BH}`$ that counts the weighted number of supersymmetric bound states. For the case of $`N=2`$ we find all the $`j_L=0`$ chiral primaries and discover $`_{BH}(0)=1`$.
5.1. The Center-of-Mass Multiplet
Throughout this paper we have largely neglected the center-of-mass degrees of freedom. We give a brief description here for completeness. The center-of-mass theory contains two complex bosons $`X^k`$ in the (2,2) of $`SU(2)_{\mathrm{Left}}\times SU(2)_{\mathrm{Right}}`$ and two complex fermions $`\lambda ^k`$ in the (2,2) of $`SU(2)_{\mathrm{Left}}\times SU(2)_R`$. The ground state is therefore a spinor of $`SU(2)_{\mathrm{Left}}\times SU(2)_R`$. Explicitly, defining $`|\eta `$ as a zero momentum state obeying
$$\lambda ^{\overline{k}}|\eta =0,$$
there is an $`SU(2)_R`$ doublet of spacetime bosons
$$|\eta ,\lambda ^1\lambda ^2|\eta ,$$
and an $`SU(2)_{\mathrm{Left}}`$ doublet of spacetime fermions
$$\lambda ^1|\eta ,\lambda ^2|\eta .$$
This is exactly the content of a massive, positively charged, spacetime hypermultiplet. These states are all annihilated by the four supercharges $`Q^\alpha `$ because $`P_k=0`$.
5.2. The Superconformal Sector
The low energy interactions between $`N`$ five-dimensional BPS black holes with charges $`Q_A`$ are described by the $`4(N1)`$-dimensional $`D(2,1;0)`$ theory with metric (4.1) constructed from the potential
$$L=d^4X\left(\underset{A=1}{\overset{N}{}}\frac{Q_A}{|\stackrel{}{X}\stackrel{}{X}^A|^2}\right)^3.$$
In this context $`SU(2)_{\mathrm{Right}}`$ in $`SU(1,1|2)`$ generates right-handed spatial rotations , while $`SU(2)_R`$ generates the spacetime $`R`$-symmetry transformations.
The theory following from (5.1) has additional global symmetries corresponding to $`SU(2)_{\mathrm{Left}}`$ spatial rotations. These are generated by
$$J_L^r=\frac{1}{2}X^MK_M^{rN}(\mathrm{\Pi }_N\frac{i}{2}\mathrm{\Omega }_{NQ}^P\lambda ^Q\lambda _P)+\frac{i}{4}K_{MN}^r\lambda ^M\lambda ^N,$$
where $`K^r`$ are the constant anti-self dual complex structures on $`\mathrm{IR}^4`$. $`J_L`$ obeys
$$[J_L^r,\lambda ^M]=\frac{i}{2}\lambda ^NK_N^{rM},[J_L^r,X^M]=\frac{i}{2}X^NK_N^{rM}[J_L^r,J_R^s]=0,$$
and commutes with all the generators of the superconformal group. $`\lambda ^M`$ transforms in the $`(1,2,2)`$ of $`SU(2)_{\mathrm{Right}}\times SU(2)_R\times SU(2)_{\mathrm{Left}}`$, as appropriate for a goldstino. The full symmetry group of the system is $`SU(2)_{\mathrm{Left}}\times D(2,1;0)`$.
5.3. Validity of the Approximations
In this subsection we discuss potential corrections to the theory defined by (5.1), and in particular whether or not they could affect the conclusion that there is a divergent continuum of infrared states.
The superconformal theory defined by (5.1) was derived in from a more general (non-superconformal) quantum mechanics, in which a constant is added to the sum inside the parentheses. (5.1) then arises in an $`M_p\mathrm{}`$ limit \[7,,1\], with the rescaled separations $`|\stackrel{}{X}^A\stackrel{}{X}^B|M_p^{3/2}|\stackrel{}{X}^A\stackrel{}{X}^B|`$ (with dimensions $`\sqrt{mass}`$) held fixed. At the same time the energies are rescaled by a factor of $`M_p`$ so that, in the limit, all excitations of the superconformal theory have zero energy as measured with respect to the original time coordinate at spatial infinity. We wish to know whether all of these zero-energy states are really present or whether some of them might be removed by corrections which have been neglected so far.
Since $`M_p`$ has been taken to infinity, there can be no $`1/M_p`$ corrections to (5.1). As there are no dimensionful parameters in the infrared limit, corrections to (5.1) must be supressed by dimensionless quantities such as $`\dot{X}/X^3`$. Such terms can indeed be seen to arise for example as Born-Infeld type corrections. For the case of two black holes the corrected action is of the general form
$$\frac{1}{2}dt(\frac{|_t\stackrel{}{X}|^2}{|\stackrel{}{X}|^4}+\frac{|_t\stackrel{}{X}|^4}{|\stackrel{}{X}|^{10}}\mathrm{}.),$$
where $`\stackrel{}{X}`$ is the relative separation. In terms of the momentum defined from the leading term
$$\stackrel{}{P}=\frac{_t\stackrel{}{X}}{|\stackrel{}{X}|^4},$$
this becomes
$$\frac{1}{2}dt(|\stackrel{}{X}|^4|\stackrel{}{P}|^2+|\stackrel{}{X}|^6|\stackrel{}{P}|^4+\mathrm{}.).$$
Such correction terms can be neglected as long as
$$|\stackrel{}{P}|\frac{1}{|\stackrel{}{X}|}.$$
As the black holes approach one another, the relative momenta must be smaller and smaller in order to suppress corrections. Hence we do not expect all states to be reliably described by the superconformal theory.
The number of states which can be reliably described by the superconformal theory can be estimated by the volume $`\mathrm{\Omega }`$ of phase space in which (5.1) is obeyed. This is, with an infrared cutoff $`ϵ0`$,
$$\mathrm{\Omega }_{|\stackrel{}{X}|>ϵ}d^4X_{|\stackrel{}{P}|<1/|\stackrel{}{X}|}d^4P\mathrm{ln}ϵ.$$
Hence, according to this rather crude estimate, a logarithmic infrared divergence in the number of states appears to remain even when the untrustworthy regions of phase space are removed.
We can also compute the density of states as function of the energy. This is
$$d\mathrm{\Omega }d^4Xd^4P\delta (E|\stackrel{}{X}|^4|\stackrel{}{P}|^2)𝑑E.$$
This is divergent for any $`E`$ if one does not impose the restriction (5.1). Imposing (5.1) leads to the finite, scale invariant, result,
$$d\mathrm{\Omega }\frac{dE}{E}.$$
Since this density of (reliably present) states is finite, it is possible that the infrared divergences do not appear in physical processes involving scattering off of the collection of black holes. A similar mechanism was discussed in .
Another potential source of corrections comes from black hole fragmentation as in . The moduli space geometry (4.1), (5.1) was derived using the low-energy supergravity approximation. The validity of this requires that the spacetime curvature is small compared to the inverse Planck length, or equivalently $`Q_A1`$.<sup>9</sup> Perhaps surprisingly, if $`Q_A1`$, then the curvatures remain small even for $`\stackrel{}{X}^A\stackrel{}{X}^B0`$. In this paper we have ignored the possibility that the black holes might fragment into smaller pieces. In five dimensions we know of no way to suppress this energetically. One might try to avoid this by taking all the black holes to carry the minimum quantum of charge. However it is not clear whether the expression for the moduli space metric remains valid for small charges. The expression is highly constrained both by the symmetries and the known long-distance behavior. Whether or not corrections do appear at small charge is an open question which we shall not attempt to resolve here.
In four dimensions the situation is better. Let all the black holes carry the same charges, with large, nonzero coprime electric and magnetic charges. In that case the possiblity of fragmentation is eliminated energetically, as can be seen from the BPS mass formula. The moduli space geometry \[10,,4\] and the bound-state analysis are similar for this case.
5.4. $`_{BH}`$
In this section we relate a spacetime index counting weighted degeneracies of spacetime BPS multiplets to an index in the superconformal quantum mechanics of the type discussed in subsection 3.4.
Massive representations of $`𝒩=2`$ Poincaré supersymmetry in five dimensions are of two types. The generic representation is the long multiplet $`L_j`$, with $`SU(2)_{\mathrm{Left}}\times SU(2)_{\mathrm{Right}}`$ spin content
$$L_j:[j_L,j_R]([1/2,1/2]+2[1/2,0]+2[0,1/2]+4[0,0]).$$
This multiplet has $`8(2j_L+1)(2j_R+1)`$ bosons, $`8(2j_L+1)(2j_R+1)`$ fermions, and maximal spin $`(j_L+1/2,j_R+1/2)`$. There is also a short multiplet $`S_j`$ which is annihilated by half of the supercharges and has $`Mass=Charge`$ in appropriate units. This multiplet has spin content
$$S_j:[j_L,j_R]([1/2,0]+2[0,0]).$$
This multiplet has one quarter<sup>10</sup> We do not include the conjugate multiplet with negative charges. as many bosons and fermions, and maximal spin $`(j_L+1/2,j_R)`$. It is easy to see that for either multiplet
$$Tr()^{2J_L^3+2J_R^3}=0.$$
An alternative index,
$$Tr()^{2J_R^3}y^{2J_L^3},$$
vanishes for long multiplets
$$(L_j)=0,$$
but not for short ones:
$$(S_j)=()^{2j_R}(2j_R+1)\frac{(y^{{\scriptscriptstyle \frac{1}{2}}}+y^{{\scriptscriptstyle \frac{1}{2}}})^2}{yy^1}(y^{2j_L+1}y^{2j_L1}).$$
The value of this index traced over all the quantum states of $`N`$ black holes gives a measure of the weighted number of supersymmetric states.
In the supersymmetric quantum mechanics, $`j_L`$ and $`j_R`$ are the eigenvalues of the operators $`J_L^3`$ (equation (5.1)) and $`J_R^3`$ (equation (3.1) specialized to the black hole case),<sup>11</sup> We recall that $`J_R`$ is part of the superconformal algebra, while $`J_L`$ commutes with it. augmented by the corresponding operators for the center-of-mass multiplet. The trace over this latter multiplet gives a universal factor of $`(y^{{\scriptscriptstyle \frac{1}{2}}}+y^{{\scriptscriptstyle \frac{1}{2}}})^2`$. The total index $`_{BH}`$ counting the weighted number of supersymmetric black hole bound states is then defined by
$$_{BH}^{tot}=(y^{{\scriptscriptstyle \frac{1}{2}}}+y^{{\scriptscriptstyle \frac{1}{2}}})^2Tr_{SCQM}()^{2J_R^3}y^{2J_L^3},$$
where the trace is over the internal Hilbert space of the superconformal quantum mechanics, without the center-of-mass multiplet. We may also define a reduced index of the form (3.1) by factoring out the center of mass factor and restricting to the subspace transforming in the dimension $`2j_L+1`$ representation of $`SU(2)_L`$, as
$$_{BH}(j_L)=Tr_{SCQM}()^{2J_R^3}|_{j_L}.$$
In the next section we will evaluate this for two black holes and $`j_L=0`$ by counting chiral primaries.
5.5. $`N=2`$
For the case of two black holes, the $`D(2,1;0)`$ quantum mechanics governing their relative separation $`\stackrel{}{X}_{12}\stackrel{}{X}^1\stackrel{}{X}^2`$ is described by the metric
$$ds^2=12\pi ^2(Q_1^2Q_2+Q_1Q_2^2)\frac{d\stackrel{}{X}_{12}d\stackrel{}{X}_{12}}{|\stackrel{}{X}_{12}|^4},$$
as follows from (4.1) and (5.1). One finds
$$\stackrel{}{D}=\stackrel{}{X}_{12},K=\frac{6\pi ^2(Q_1^2Q_2+Q_2^2Q_1)}{|\stackrel{}{X}_{12}|^2},e^\varphi =\frac{|\stackrel{}{X}_{12}|^4}{6\pi ^2(Q_1^2Q_2+Q_2^2Q_1)}.$$
There are states constructed as in (3.1) corresponding to $`(p,0)`$-forms with $`p=0,1,2`$. For $`p=0`$ the chiral primary condition (3.1) reduces to $`f_0=0`$ and $`Df_0=0`$, which has no nontrivial solutions.
For $`p=1`$ the condition $`Df_1=0`$ implies that $`f_1`$ is proportional to $`D`$. The condition $`f_1=0`$ then implies that the proportionality factor is a function of $`K`$ times an antiholomorphic function. In complex coordinates $`(z^1,z^2)`$
$$f_1=c(\overline{z}^1,\overline{z}^2,K)D.$$
For the case considered here of $`j_L=0`$, $`f_1`$, and therefore $`c`$, must be invariant under $`SU(2)_L`$ rotations. This requires $`c`$ to be a function of $`K`$ only. Using the formula (3.1) for the special case $`n=2`$ and $`h=2`$, one finds
$$\overline{}e^\varphi D=0.$$
It follows that the last condition in (3.1) is satisfied only when the proportionality factor is a constant. Hence the state
$$|D=D_a\lambda ^a|0$$
is the unique chiral primary with $`p=1`$ for $`N=2`$. The norm is
$$D|D=2d^4X\sqrt{g}e^\varphi Ke^{2K}.$$
For a pair of black holes $`K`$ goes like $`\frac{1}{r^2}`$, and so (5.1) converges at both large and small $`r`$.<sup>12</sup> It is now seen explicitly that the norm $`0|0`$ of the $`L_0`$ ground state diverges logarithmically at large $`r`$. This norm is given by the integral in (5.1) without the factor of $`K`$. The state $`|D`$ is an $`SU(2)_R`$ singlet and the $`J_R^3=+\frac{1}{2}`$ element of an $`SU(2)_{\mathrm{Right}}`$ doublet.
At $`p=2`$, the first two chiral primary conditions are trivially satisfied. The general solution of the last condition (using the relations (4.1)) is the $`(2,0)`$-form
$$f(X)I_{ab}^{},$$
where $`f(X)`$ is holomorphic. (5.1) is singular unless $`f`$ is constant. In this case the norm becomes
$$I^{}|I^{}=2d^4X\sqrt{g}e^{\varphi 2K},$$
where we have used $`I^{+ab}I_{ab}^{}=2`$. The integral in (5.1) diverges logarithmically at $`x\mathrm{}`$. Hence there are no normalizable chiral primaries at $`p=2`$.
We conclude that the supersymmetric index $`_{BH}(j_L=0)`$ is $`1`$ for a pair of black holes.
Acknowledgements
We are grateful to J. Maldacena, A. Maloney, J. Michelson and M. Spradlin for useful conversations, and to M. Stern for pointing out an error in an earlier version of this work. This work is supported in part by DOE grant DE-FG02-91ER40654 and an NDSEG graduate fellowship. A. V. is also supported by INTAS-OPEN-97-1312.
References
relax J. Michelson and A. Strominger, Superconformal Multi-Black Hole Quantum Mechanics, JHEP 9909 (1999) 005, hep-th/9908044. relax R. Britto-Pacumio, J. Michelson, A. Strominger and A. Volovich, Lectures on Superconformal Quantum Mechanics and Multi-Black Hole Moduli Spaces, contributed to NATO Advanced Study Institute on Quantum Geometry, Akureyri, Iceland, hep-th/9911066. relax J. Gutowski and G. Papadopoulos, The Dynamics of Very Special Black Holes, Phys. Lett. B472 (2000) 45, hep-th/9910022. relax J. Gutowski and G. Papadopoulos, Moduli Spaces for Four-and Five-Dimensional Black Holes, Phys. Rev. D62 (2000) 064023, hep-th/0002242. relax G. W. Gibbons and P. K. Townsend, Black Holes and Calogero Models, Phys. Lett. B454 (1999) 187, hep-th/9812034. relax P. Claus, M. Derix, R. Kallosh, J. Kumar, P.K. Townsend and A. Van Proeyen, Black Holes and Superconformal Mechanics, Phys. Rev. Lett. 81 (1998) 4553, hep-th/9804177. relax J. Maldacena, J. Michelson and A. Strominger, Anti-de Sitter Fragmentation, JHEP 9902 (1999) 011, hep-th/9812073. relax S. Cacciatori, D. Klemm and D. Zanon, $`w_{\mathrm{}}`$ Algebras, Conformal Mechanics, and Black Holes, Class. Quant. Grav. 17 (2000) 1731, hep-th/9910065. relax G. Papadopoulos, Conformal and Superconformal Mechanics, Class. Quant. Grav. 17 (2000) 3715, hep-th/0002007. relax A. Maloney, M. Spradlin and A. Strominger, Superconformal Multi-Black Hole Moduli Spaces in Four Dimensions, hep-th/9911001. relax J. Michelson and A. Strominger, The Geometry of (Super) Conformal Quantum Mechanics, Commun. Math. Phys. 213 (2000) 1, hep-th/9907191. relax J.P. Gauntlett, R.C. Myers and P.K. Townsend, Black Holes of D=5 Supergravity, Class. Quant. Grav. 16 (1999) 1, hep-th/9810204. relax T. Banks, M. O’Loughlin and A. Strominger, Black Hole Remnants and the Information Puzzle, Phys. Rev. D47 (1993) 4476, hep-th/9211030.
|
warning/0004/hep-ph0004166.html
|
ar5iv
|
text
|
# Neutrinos and Gauge Unification
\[
## Abstract
The approximate unification of gauge couplings is the best indirect evidence for low-energy supersymmetry, although it is not perfect in its simplest realizations. Given the experimental evidence for small non-zero neutrino masses, it is plausible to extend the MSSM with three right-handed neutrino chiral multiplets, with large Majorana masses below the unification scale, so that a see-saw mechanism can be implemented. In this extended MSSM, the unification prediction for the strong gauge coupling constant at $`M_Z`$ can be lowered by up to $`5\%`$, bringing it closer to the experimental value at $`1\sigma `$, therefore improving significantly the accuracy of gauge coupling unification.
\]
Gauge coupling unification is an important test for candidates to fundamental theories beyond the Standard Model (SM). On the one hand, gauge unification is a prediction of most of such theories. In particular, GUT models predict the unification of the gauge couplings at a scale $`M_X`$, corresponding to the breakdown of the unifying group. Also, most of the string constructions predict unification of all the gauge couplings, including gravity, even in the absence of a GUT group. On the other hand, due to the present high-precision knowledge of the gauge couplings at low energy, one can extrapolate them to high energy through the corresponding renormalization group equations (RGEs) and verify whether unification takes place or not.
Of course the evolution of the couplings (as described by the RGEs) depends on the type of theory under study. In particular, unification can depend on whether it is supersymmetric or not, whether it has the minimal matter content or not, whether it has (large) extra dimensions or not, etc. Consequently, the requirement of gauge coupling unification offers the possibility of testing these theories at very high scales. In this respect, it has been known for a long time that non-supersymmetric gauge unification (with minimal matter content) does not work. In contrast, supersymmetric gauge unification takes place with remarkable accuracy at $`M_X2\times 10^{16}`$ GeV. This fact was confirmed in 1990 with the precise LEP determinations of the gauge couplings. The unification occurs in such a beautiful way that it is hard to believe that it is a mere coincidence, and it is currently considered as a strong hint in favor of supersymmetry. However, there are some problems with the current status of this issue. First, the conventional prediction of (weakly coupled heterotic) superstrings is that the gauge couplings (including gravity) are unified at the string scale $`M_{string}5\times 10^{17}`$ GeV, which is one order of magnitude too high compared with $`M_X`$. During the last years many new string (or string-inspired) scenarios have appeared addressing this problem: strongly coupled strings, large extra dimensions, etc. Another problem is that the ordinary supersymmetric unification, however beautiful, is not perfect in the simplest scenarios. A convenient way to express the discrepancy is to start with the $`SU(2)\times U(1)_Y`$ low-energy couplings $`\alpha _1(M_Z)`$ and $`\alpha _2(M_Z)`$ to obtain both the unification scale and gauge coupling, $`M_X`$ and $`\alpha (M_X)`$. Then, running from $`M_X`$ back to $`M_Z`$, one gets a prediction for $`\alpha _3(M_Z)`$, to be compared to the experimental value.
Suppose then that, following for example ref. one has a MSSM spectrum as obtained from a minimal supergravity model. Starting with the experimental input (in the $`\overline{\mathrm{MS}}`$ renormalization scheme, before conversion to the $`\overline{\mathrm{DR}}`$ scheme)
$`\widehat{\alpha }_1^1(M_Z)`$ $`=`$ $`58.98\pm 0.04,`$ (1)
$`\widehat{\alpha }_2^1(M_Z)`$ $`=`$ $`29.57\pm 0.03,`$ (2)
$`M_Z`$ $`=`$ $`91.197\pm 0.007,`$ (3)
one integrates numerically two-loop RGEs, including also the supersymmetric thresholds until one finally obtains a prediction $`\alpha _3(M_Z)0.13`$, to be compared to the experimental value
$$\widehat{\alpha }_3(M_Z)=0.119\pm 0.004,$$
(4)
which represents a $`3\sigma `$ discrepancy \[other evaluations give a smaller experimental uncertainty in (4), thus leading to a stronger discrepancy. We choose to be rather conservative at this point\]. Several comments are in order here. The effect of supersymmetric thresholds introduces a dependence on the type of MSSM spectrum considered. For instance, the larger the supersymmetric masses, the lower $`\alpha _3(M_Z)`$. Allowing the squark masses to be up to 1 TeV, one gets $`\alpha _3(M_Z)\stackrel{>}{_{}}0.127`$. On the other hand, working in the context of gauge-mediated SUSY breaking, which implies the presence of extra (messenger) fields and thus a departure from the minimal matter content, one can get slightly lower values, $`\alpha _3(M_Z)\stackrel{>}{_{}}0.125`$. This alleviates the disagreement, but it does not solve it. One would need a negative correction between$`3\%`$ and $`9\%`$ in $`\alpha _3(M_Z)`$ \[or a correction three times smaller in $`\alpha _3(M_X)`$\] in order to reconcile prediction and experiment. It has been argued that the origin of the discrepancy could be high-energy threshold corrections. These may be of stringy or GUT origin and are model-dependent. E.g. in a GUT scenario these corrections are due to the appearance at the GUT scale of incomplete $`SU(5)`$ multiplets, most notably the color triplet Higgs bosons. Demanding a threshold correction of the correct magnitude and sign severely restricts the models, e.g. minimal $`SU(5)`$ is incompatible with such a requirement .
In the present paper we explore the impact of massive neutrinos in the unification of gauge couplings. Observations of the flux of atmospheric neutrinos by SuperKamiokande provide strong evidence for neutrino oscillations, which in turn imply that (at least two species of) neutrinos must be massive. Preliminary results from the long-baseline experiment K2K tend to confirm this evidence . Additional support to non-zero neutrino masses is given by the need of neutrino oscillations to explain the solar neutrino flux deficit . The simplest and most beautiful mechanism to account for small neutrino masses in a natural way is probably the see-saw mechanism . Its supersymmetric version has superpotential
$`W=W_{MSSM}{\displaystyle \frac{1}{2}}\nu _R^c\nu _R^c+\nu _R^c𝐘_\nu LH_2,`$ (5)
where $`W_{MSSM}`$ is the superpotential of the MSSM. The extra terms involve, beside the usual three lepton doublets, $`L`$, three additional neutrino chiral fields $`\nu _R`$ singlets under the SM group (generation indices are suppressed). $`𝐘_\nu `$ is the matrix of neutrino Yukawa couplings and $`H_2`$ is the hypercharge $`+1/2`$ Higgs doublet. The Dirac mass matrix is $`𝐦_𝐃=𝐘_\nu v\mathrm{sin}\beta `$. Finally, $``$ is a $`3\times 3`$ Majorana mass matrix which does not break the SM gauge symmetry. It is natural to assume that the overall scale of $``$, which we will denote by $`M`$, is much larger than the electroweak scale or any soft mass. Below $`M`$ the heavy neutrino fields can be integrated out, giving rise to an effective mass term for the left-handed neutrinos, $`\frac{1}{2}\nu ^T_\nu \nu `$, with $`_\nu =𝐦_{𝐃}^{}{}_{}{}^{T}^1𝐦_𝐃`$, suppressed with respect to the typical fermion masses by the inverse power of the large scale $`M`$. In fact, large values of the neutrino Yukawa couplings are perfectly consistent with tiny neutrino masses for values of $`M`$ sufficiently close to $`M_X`$.
The influence of heavy right-handed neutrinos on the unification of gauge couplings is due to the fact that above $`M`$, the Yukawa couplings $`𝐘_\nu `$ of eq.(5) affect the RGEs of the gauge couplings at two-loop order, in a similar way as the top-Yukawa coupling does in the MSSM. Since $`𝐘_\nu `$ can be sizeable, one expects a non-trivial impact in gauge unification. The (two-loop) RGEs at scales $`Q`$ between $`M`$ and $`M_X`$, for $`g_i`$, $`𝐘_𝐭`$ and $`𝐘_\nu `$ read \[with $`g_1`$ normalized as in $`SU(5)`$\]
$`{\displaystyle \frac{dg_i^2}{dt}}`$ $`=`$ $`2\kappa \left[b_i+\kappa \left(b_{ij}g_j^2a_{i\alpha }\mathrm{tr}𝐇_\alpha \right)\right]`$ (6)
$`{\displaystyle \frac{d𝐘_𝐭}{dt}}`$ $`=`$ $`\kappa 𝐘_𝐭\left(c_ig_i^2+3𝐇_𝐭+\mathrm{T}\right)`$ (7)
$`+`$ $`\kappa ^2𝐘_𝐭[(c_ib_i+{\displaystyle \frac{c_i^2}{2}})g_i^4+{\displaystyle \frac{136}{45}}g_1^2g_3^2+8g_2^2g_3^2`$ (8)
$`+`$ $`g_1^2g_2^2+\left({\displaystyle \frac{2}{5}}g_1^2+6g_2^2\right)𝐇_𝐭+\left({\displaystyle \frac{4}{5}}g_1^2+16g_3^2\right)\mathrm{Tr}𝐇_𝐭`$ (9)
$``$ $`9\mathrm{T}\mathrm{r}𝐇_𝐭^\mathrm{𝟐}4𝐇_{𝐭}^{}{}_{}{}^{2}3𝐇_𝐭\mathrm{T}9\mathrm{T}\mathrm{r}𝐇_𝐭^\mathrm{𝟐}{\displaystyle \frac{}{}}]`$ (10)
$`{\displaystyle \frac{d𝐘_\nu }{dt}}`$ $`=`$ $`\kappa 𝐘_\nu \left(c_i^{}g_i^2+3𝐇_\nu +\mathrm{T}\right)`$ (11)
$`+`$ $`\kappa ^2𝐘_\nu [(c_i^{}b_i+{\displaystyle \frac{c_i^{}^2}{2}})g_i^4+{\displaystyle \frac{9}{5}}g_1^2g_2^2`$ (12)
$`+`$ $`6\left({\displaystyle \frac{1}{5}}g_1^2+g_2^2\right)𝐇_\nu +\left({\displaystyle \frac{4}{5}}g_1^2+16g_3^2\right)\mathrm{Tr}𝐇_𝐭`$ (13)
$``$ $`3\mathrm{T}\mathrm{r}𝐇_\nu ^24𝐇_\nu ^23𝐇_\nu \mathrm{T}9\mathrm{T}\mathrm{r}𝐇_𝐭^\mathrm{𝟐}{\displaystyle \frac{}{}}]`$ (14)
(summation over repeated indices is understood) with $`t=\mathrm{ln}Q`$, $`\kappa =1/(16\pi ^2)`$ and
$$𝐇_\alpha 𝐘_\alpha ^{}𝐘_\alpha ,\mathrm{T}\mathrm{Tr}(3𝐇_𝐭+𝐇_\nu ),$$
(15)
with $`\alpha =𝐭,\nu `$. Here $`𝐘_𝐭`$ represents the Yukawa matrix of the $`u`$-type quarks (dominated by the top Yukawa coupling) and the numerical coefficients are given by
$`b_i`$ $`=`$ $`(33/5,1,3),c_i=(13/15,3,16/3),`$ (16)
$`c_i^{}`$ $`=`$ $`(3/5,3,0),a_{i\alpha }=\left(\begin{array}{cc}26/5& 6/5\\ 6& 2\\ 4& 0\end{array}\right),`$ (20)
$`b_{ij}`$ $`=`$ $`\left(\begin{array}{ccc}199/25& 27/5& 88/5\\ 9/5& 25& 24\\ 11/5& 9& 14\end{array}\right).`$ (24)
It is clear from eq.(6) that the presence of neutrinos will only affect the RGEs of $`g_1`$ and $`g_2`$. The magnitude of the final correction will depend on the numerical values of $`𝐘_\nu `$ and $`M`$. These are not independent quantities, since the final neutrino masses $`_\nu =v^2\mathrm{sin}^2\beta 𝐘_{\nu }^{}{}_{}{}^{T}^1𝐘_\nu `$ (appropriately ran down to the electroweak scale ) must be consistent with observations. Atmospheric neutrino data only allow to determine one neutrino mass splitting, $`\mathrm{\Delta }m_\nu ^210^3\mathrm{eV}^2`$. On the other hand, ’standard’ explanations of solar neutrino flux deficits require neutrino oscillations with a much smaller mass splitting, corresponding to a different pair of neutrinos. In addition, there are upper bounds on neutrino masses, coming e.g. from the non-observation of neutrinoless double $`\beta `$-decay and other experiments. All this information implies that there are three possible types of neutrino spectrum: Hierarchical, $`m_1^2<m_2^2m_3^2`$; Intermediate, $`m_1^2m_2^2m_3^2`$; and Degenerate, $`m_1^2m_2^2m_3^2`$. In the hierarchical case, the larger neutrino mass should be $`𝒪(10^110^2)\mathrm{eV}`$, while in the degenerate case cosmological observations require $`m_\nu \stackrel{<}{_{}}𝒪(2)\mathrm{eV}`$. This means, that once a particular scenario for the neutrino spectrum is chosen, $`𝐘_\nu `$ and $`M`$ are not independent quantities any more.
The presence of right-handed neutrinos above $`M`$ induces a small change both on the scale of unification and the value of the unified gauge coupling. This in turn affects the prediction of $`\alpha _3(M_Z)`$ derived from the assumption of perfect unification. Given that the changes with respect to the MSSM case can be treated as a small perturbation, it is simple to make an analytical estimate from eqs.(6) and (14) to get:
$$\frac{\mathrm{\Delta }\alpha _3(M_Z)}{\alpha _3(M_Z)^2}\frac{9}{7\pi }\frac{1}{16\pi ^2}<\mathrm{Tr}𝐘_\nu ^{}𝐘_\nu >\mathrm{ln}\frac{M_X}{M}.$$
(25)
In this result, $`<\mathrm{Tr}𝐘_\nu ^{}𝐘_\nu >`$ is an average value in the interval from $`M`$ to $`M_X`$:
$$<\mathrm{Tr}𝐘_\nu ^{}𝐘_\nu >\mathrm{ln}\frac{M_X}{M}=_{Q=M}^{M_X}\mathrm{Tr}𝐘_\nu ^{}𝐘_\nu d\mathrm{ln}Q.$$
(26)
Two important implications of Eq. (25) are, first, that neutrino corrections always make the predicted $`\alpha _3(M_Z)`$ smaller, as required to improve the agreement with the experimental value, and second, that the effect will be important if the neutrino Yukawa couplings $`𝐘_\nu `$ are sizable, which is a natural assumption, as we saw.
We can now estimate the average in Eq. (26) considering that the evolution of $`𝐘_\nu (t)`$ is well described by its 1-loop RGE neglecting all couplings different from $`𝐘_\nu `$. This approximation is justified if $`𝐘_\nu `$ is large (the case of interest). The final result will depend on how many neutrino Yukawa couplings (or more precisely, eigenvalues of $`𝐘_{\nu }^{}{}_{}{}^{}𝐘_\nu `$) are large. With only one large neutrino coupling (this corresponds naturally to a hierarchical neutrino spectrum) it is straightforward to evaluate the integral in (26) in the approximation just described to get
$`\mathrm{\Delta }\alpha _3(M_Z){\displaystyle \frac{3N}{28\pi }}\alpha _3(M_Z)^2\mathrm{log}\left[{\displaystyle \frac{Y_\nu (M_X)^2}{Y_\nu (M)^2}}\right],`$ (27)
with $`N=1`$ counting the number of large neutrino Yukawa couplings. Of course the effect on $`\alpha _3(M_Z)`$ is larger if all three neutrino Yukawa couplings are large (this corresponds naturally to a degenerate neutrino spectrum, but not necessarily: a hierarchy in the spectrum can be induced also in this case by flavor-dependent Majorana masses). In this $`N=3`$ case, the result for $`\mathrm{\Delta }\alpha _3(M_Z)`$ slightly depends on the kind of texture for $`𝐘_\nu `$. A good estimate is given by simply taking $`𝐘_\nu =Y_\nu \mathrm{𝟏}`$ and in that case, Eq. (27) also holds, simply setting $`N=3`$.
The magnitude of the correction (27) depends on the values of $`Y_\nu `$ at $`M_X`$ and $`M`$, but the latter is fixed in order to fit the physical neutrino mass, as explained above. Hence, the correction is just a function of a unique parameter, $`Y_\nu (M_X)`$. \[It is also a function of the neutrino mass, $`m_\nu `$, but changes on $`m_\nu `$ just imply a corresponding modification of $`M`$ and since the evolution of $`Y_\nu `$ occurs mainly near $`M_X`$ the final correction on $`\alpha _3(M_Z)`$, given by eq.(27), is quite insensitive to $`m_\nu `$.\] The most favorable case occurs when $`Y_\nu (M_X)`$ is large, near a Landau pole. A representative case is $`Y_\nu (M_X)8`$. Then $`Y_\nu (M)2`$ and we get $`\mathrm{\Delta }\alpha _3(M_Z)1.4\%`$ for the case $`N=1`$ and $`\mathrm{\Delta }\alpha _3(M_Z)4.1\%`$ for $`N=3`$. It is clear that in the first case this correction is not sufficient to reconcile $`\alpha _3(M_Z)`$ with the experimental value, but it can be enough in the second case. These estimates agree well with our full numerical results, obtained by numerical integration of the RGEs (6-10). Now the two-loop contributions soften the RGE for $`𝐘_\nu `$, which implies that $`𝐘_\nu `$ decreases more slowly when the energy scale goes down. This increases the Yukawa coupling average in eq.(25), and thus the (negative) shift of the low-energy strong coupling, $`\mathrm{\Delta }\alpha _3(M_Z)`$.
An important question at this point is how large can $`Y_\nu (M_X)`$ be without jeopardizing the perturbative expansion, and hence the reliability of the results. This has been studied in great detail in ref. for the case of the top Yukawa coupling. Since, in the large coupling regime relevant for this question, the RGEs for the top and neutrino Yukawa couplings (in the $`N=3`$ case) are identical, the results of ref. are applicable also to the present case. The conclusions of the authors of ref. , using Padé-Borel resummations of the 4-loop beta-function, were that the qualitative behaviour of the running is well described by the one-loop approximation and can be further improved by Padé-Borel approximants which are reliable for values of the Yukawa coupling up to $`\stackrel{<}{_{}}8`$ (For the $`N=1`$ case this is a conservative estimate, since in that case the beta-function for $`Y_\nu `$ is softer than that of $`Y_t`$.). We have used a $`[1,1]`$ Padé-Borel resummation of the $`𝐘_\nu `$ beta-function to obtain our most reliable numerical results. As the beta-function in this approximation is smaller than at one-loop, the final result for the average in (25) turns out to be a bit larger than our analytical estimates \[for a fixed value of $`Y_\nu (M_X)`$\]. The negative shift on $`\alpha _3(M_Z)`$ goes then from 4% to about 5%. This fact, together with the uncertainties in $`\alpha _1(M_Z)`$, $`\alpha _2(M_Z)`$ and the top mass, imply that the usual MSSM scenarios, which yielded an unsuccessful prediction $`\alpha _3(M_Z)0.13`$, are now rescued within $`1\sigma `$ confidence level. \[The agreement can be perfect in other MSSM scenarios, which predict slightly lower values of $`\alpha _3(M_Z)`$, even if the error in (4) becomes smaller.\]
This is illustrated in fig. 1 for a typical MSSM scenario The first plot shows how the gauge couplings fail to unify in the ordinary MSSM in the absence of massive neutrinos (the requirement of unification would imply $`\alpha _3(M_Z)=0.13`$ in the case plotted). The second plot corresponds to the MSSM extended with neutrinos getting mass via a see-saw mechanism. More precisely, the plot corresponds to a scenario with degenerate neutrinos of mass $`m_\nu =2`$ eV and $`Y_\nu (M_X)=8`$. It is apparent how the $`\alpha _1`$ and $`\alpha _2`$ runnings are modified in a suitable way to get gauge unification.
Besides the logarithmic effect on unification we have described, the presence of heavy right-handed neutrinos affects the running of the gauge couplings also through finite two-loop threshold effects at the Majorana scale $`M`$. However, these will depend on the neutrino Yukawa coupling $`Y_\nu (M)`$, which in the case of interest is much smaller than $`Y_\nu (M_X)`$, so that it is safe to neglect these matching effects.
Finally, it is interesting to point out that effects similar to the ones we have found are expected in the next-to-minimal supersymmetric standard model (NMSSM), i.e. the MSSM extended with a singlet chiral multiplet $`S`$. The superpotential of the NMSSM does not contain a mass term for the Higgs multiplets (the $`\mu `$-term). However, an effective $`\mu `$ term of the correct order of magnitude is generated dynamically by $`\mathrm{\Delta }W=Y_sSH_1H_2`$, when $`S`$ takes a vacuum expectation value. This solves in fact the $`\mu `$ problem of the MSSM and is one of the main virtues of the NMSSM (although this model has its own drawbacks). The influence of the new Yukawa coupling $`Y_s`$ on the running of the gauge couplings is also a two-loop effect, of exactly the same form as in (6) with $`Y_s^2`$ instead of $`\mathrm{Tr}𝐇_\nu `$ (the coefficients $`a_{\alpha i}`$ are exactly the same for $`\alpha =\nu `$ and $`\alpha =s`$). The final impact on $`\alpha _3(M_Z)`$ is therefore given by a formula like (25) with $`\mathrm{Tr}𝐘_\nu ^{}𝐘_\nu `$$``$$`Y_s^2`$ and $`M`$ replaced by the mass $`M_S`$ of the singlet $`S`$, which is close to the electroweak scale (this makes the logarithm much larger). Numerically, we find $`\mathrm{\Delta }\alpha _3(M_Z)3\%`$ for $`M_S=1`$ TeV and $`Y_s(M_X)=8`$. Similar effects can occur in other extensions of the MSSM with additional Yukawa couplings, as has been shown for models with $`R`$-parity violating couplings in ref. .
In conclusion, we have examined the impact of heavy see-saw neutrinos (plausible in view of the growing experimental evidence in favor of non-zero neutrino masses) on the unification of gauge couplings, in particular as reflected in the unification prediction for the strong gauge coupling constant at the electroweak scale. We find that the effect is small, but is always of the right sign and can be of the right magnitude to bring the too high MSSM prediction for $`\alpha _3(M_Z)`$ down to values within $`1\sigma `$ of the experimental value. Given that adding three heavy right-handed neutrinos is not an ad-hoc extension of the MSSM but on the contrary is well motivated by experiment and theory alike, this result is welcome and noticeable. This effect should be taken into account, even in models with sizeable stringy or GUT high-energy threshold corrections. For example, models that have been discarded for not giving the appropriate threshold corrections (e.g. minimal $`SU(5)`$, ), can be now perfectly consistent.
Acknowledgments We thank H. Dreiner for useful correspondence. We also thank A. Delgado for very useful discussions. A. I. thanks the Comunidad de Madrid (Spain) for a pre-doctoral grant.
|
warning/0004/astro-ph0004223.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Gamma-ray bursts (GRBs) are short and intense flashes of cosmic high energy ($``$ 10 KeV$``$10 GeV) photons. The study of GRBs was revolutionized in 1997 when the Italian-Dutch $`X`$ray satellite BeppoSAX started providing positions of some GRBs with an accuracy of a few arcminutes within a few hours after the burst. This made the multi-wavelength observations of the long-lived emission, known as afterglow of GRB, at longer wavelengths as a routine. Such observations are of crucial importance for understanding and constraining the active emission mechanisms of GRBs as well as for the study of the nature, structure and composition of surroundings. For this, early light curves of GRB afterglows need to be densely sampled for long time intervals. A huge amount of observing time is therefore required on optical telescopes. Since the optical transient (OT) of a GRB has generally apparent $`R`$ magnitude between 18 to 22, if it is detected within a day or so after the burst, the 1-m class optical telescopes equipped with modern CCD detector are capable of observing them. Such telescopes are not only large in number throughout the world but also less in demand compared to 2-m class or larger size optical telescopes. The large amount of observing time is therefore available on them (cf. Sagar 2000 for detailed discussions). The 1-m class optical telescopes equipped with CCD detector, thus, can contribute to the world-class science in an emerging front-line research area of GRB. We at U.P. State Observatory (UPSO), Nainital, therefore, started the optical follow-up observations of the GRB afterglows in January 1999 under an international collaborative programme coordinated by one of us (AJCT). So far, successful photometric observations have been carried out for 4 GRB afterglows from UPSO, Nainital. The UPSO photometric observations for earlier 3 events namely GRB 990123, GRB 991208 and GRB 991216 have been presented by Sagar et al. (1999, 2000). Such observations for the GRB000301C are presented here. It is worth mentioning here that first earliest optical observations of GRB 000301C have been carried out by us. An introduction to the GRB 000301C studied here is given below.
Smith et al. (2000) reported All Sky Monitor (ASM) on the Rossi $`X`$ray Timing Explorer (RXTE), Ulysses and Near Earth Asteroid Rendezvous (NEAR) detection of a GRB on 2000 March 01 at 09:51:37 UT. GRB 000301C therefore joins the group of GRB 991208 (see Hurley et al. 2000) whose positions were determined only by the Interplanetary Network Localization alone without Compton Gamma-Ray observatory BATSE or BeppoSAX observations within 1.5 day of the event. The positions were of such accuracy ($``$50 arcmin<sup>2</sup> in this case) that it led to the successful identification of radio, millimeter, optical and near-IR afterglows, and eventually to the measurement of its redshift. Fig. 1 shows the light curve of the prompt $`\gamma `$ray emissions of the GRB 000301C detected by NEAR in the energy band 100 – 1000 KeV downloaded from the http://lheawww.gsfc.nasa.gov/. The burst profile is dominated by only one strong peak with no spike type structures generally observed with GRB events. The three ASM energy channels showed the strongest response in the 5 – 12 KeV band, reaching a peak flux of 3.7$`\pm `$0.7 Crab in 1 s time bin. Jensen et al. (2000) derived a peak flux of 6.3$`\times 10^7`$ erg cm<sup>-2</sup> in 0.25 s time bin in the 25 – 100 KeV energy range and the hardness ratio $`\frac{f100300}{f50100}=2.7\pm 0.6`$ for the burst. It has a sharp rise and a relatively slow decline. Duration (full width at half maximum) of the profile at trigger of the burst is only $``$ 2 s. The detection of the GRB 000301C OT was reported first by Fynbo et al. (2000a) in $`U,B,R`$ and $`I`$ passbands at $`\alpha _{2000}=16^h20^m18.^s6;\delta _{2000}=+29^{}26^{^{}}36^{^{\prime \prime }}`$ with an uncertainty of $`1^{^{\prime \prime }}`$. It was confirmed by Bernabei et al. (2000) on a $`R`$ image and by Stecklum et al. (2000) on a $`K^{^{}}`$ (2.1 $`\mu `$m) image. Coincident at the optical position, Bertoldi (2000) detected a flux of 1.9$`\pm `$0.3 mJy at 250 GHz (1.2 mm) on 2000 March 4.385 UT while Berger & Frail (2000) detected a flux of $``$ 300 $`\mu `$Jy at 8.46 GHz on 2000 March 5.67 UT. No obvious emission or absorption features are visible in the low resolution spectrum of the OT of GRB 000301C taken on 2000 March 4.41 UT in the wavelength range of 410 – 800 nm by Eracleous et al. (2000). An ultra-violet spectrum of the OT taken on 2000 March 6 with the Hubble Space Telescope (HST) by Smette et al. (2000) indicates a redshift of $`z=1.95\pm 0.1`$. It was precisely determined to a value of $`z=2.0335\pm 0.0003`$ by Castro et al. (2000) using a moderately high resolution spectra taken with the Keck-II 10-m telescope on 2000 March 4. This determination was also supported by the low resolution spectrum obtained by Feng et al. (2000) on 2000 March 3.47 UT. The value is not too different from 2.0404$`\pm `$0.0008 determined by Jensen et al. (2000) using absorption features in the spectrum obtained with very large telescope on 2000 March 5 and 6. Rhoads & Fruchter (2000), Masetti et al. (2000) and Jensen et al. (2000) present the near-infrared (IR) and $`U,B,V,R`$ and $`I`$ observations while Berger et al. (2000) provide broad-band (1.4 to 350 GHz) radio and millimeter wave observations. These data indicate that short term achromatic flux variability is superposed on the overall steepening of the light curve. The cause of the short term variability is not understood. However, it makes difficult the determination of break-time while fitting jet model in the light curve of the afterglow emission. Therefore, time scales determined in the published studies range from $`3.57.5`$ days. Present observations in combination with data published in the literature are used to study flux decay at optical and near-IR wavelengths and spectral index from ultra-violet to radio regions. These data have been used to determine precise parameters of the light curve. Details of present optical observations etc. are given in the next section while light curves and other results are presented in the remaining sections.
## 2 Optical observations, data reduction and calibrations
The optical observations of the GRB 000301C afterglow were carried out from 2000 March 2 to 9. We used a 2048 $`\times `$ 2048 pixel<sup>2</sup> CCD system attached at the f/13 Cassegrain focus of the 104-cm Sampurnanand telescope of UPSO, Nainital. All the observations were done in good photometric sky condition, expect for 2000 March 6. One pixel of the CCD chip corresponds to 0.$`^{^{\prime \prime }}`$38, and the entire chip covers a field of $`13^{^{}}\times 13^{^{}}`$ on the sky. Fig. 2 shows the location of the GRB 000301C afterglow on the CCD image taken from UPSO, Nainital. For comparison, image extracted from the Digital Palomar Observatory Sky Survey (DSS) is also shown where the absence of a GRB OT is clearly seen.
Several short exposures up to a maximum of 15 minutes were generally given. In order to improve the signal-to-noise ratio of the OT, the data have been binned in $`2\times 2`$ pixel<sup>2</sup> and also all images of a night have been stacked after correcting them for bias, non-uniformity in the pixels and cosmic ray events. Exposure times for the stacked images were 70, 50, 85, 35, 105 and 75 minutes in $`R`$ on 2000 March 2, 3, 5, 6, 7 and 8 respectively. Only one image in each $`V`$ and $`I`$ filters could be taken on 2000 March 3 with corresponding exposure times of 30 and 10 minutes respectively. As the OT is close to a bright star, DAOPHOT profile-fitting technique is used for the magnitude determination. In the field of GRB 000301C, stars (as identified in Fig. 2) are photometrically calibrated in $`R`$ passband by Garnavich et al. (2000a). The quoted uncertainty in the zero-point calibration is $`\pm `$0.05 mag. Henden (2000) provides the $`UBVRI`$ photometry for stars fainter than $`R=`$ 20 mag in the GRB 000301C field. The $`R`$ magnitudes determined by Garnavich et al. (2000a) agree with an independent measurement reported by Henden (2000). This indicates that photometric calibration used in this work is secure. Present photometric magnitudes are relative to comparison star A and D (see Fig. 2). These along with other photometric measurements of GRB 000301C afterglow published by the time of paper submission are given in Table 1. In order to avoid errors arising due to different photometric calibrations, we have used only those published photometric measurements whose magnitudes could be determined relative to determinations given by either Garnavich et al. (2000a) or Henden (2000). In $`JHK`$ filters, all published photometric measurements have been used.
Present $`R`$ images have also been independently processed, reduced and calibrated by Masetti et al. (2000). A comparison of their $`R`$ values with ours indicates good agreement. A small difference observed between the two sets of values is, perhaps, due to different data processing and calibration procedures.
## 3 Optical and near-IR photometric light curves
We have used the published data in combination with the present measurements to study the flux decay of GRB 000301C afterglow. Fig. 3 shows the plot of photometric measurements as a function of time. The X-axis is log ($`tt_0`$) where $`t`$ is the time of observation and $`t_0`$ is the time of GRB burst which is 2000 March 1.411 UT. All times are measured in unit of day. Before deriving the flux decay constants of the OT, it is mandatory to subtract the contributions from foreground/background galaxies, if there is any. Both ground based and the early HST images clearly show that any underlying galaxy would have to be fainter than $`R>`$ 25 mag (Fruchter et al. 2000a). In fact, the late-time HST images taken on 2000 April 3.9 UT by Fruchter et al. (2000b) showed that the $`R`$ magnitude of the host galaxy is 27.8$`\pm `$0.25. We have therefore not applied any correction upto $`R<`$ 23 mag of the OT for the contamination by host galaxy.
The flux decay of most of the earlier GRB afterglows is generally well characterized by a single power law $`F(t)(tt_0)^\alpha `$, where $`F(t)`$ is the flux of the afterglow at time $`t`$ and $`\alpha `$ is the decay constant. However, optical and near-IR light curves of GRB 000301C (Fig. 3) show erratic behaviour with an overall flux decay. Observers therefore took relatively long time to accept it as an OT of the GRB 000301C.
UPSO observation in $`R`$ filter on 2000 March 2.93 UT is the earliest optical observations published so far. Bhargavi & Cowsik (2000) measurements are just after us. Fig. 3 clearly indicates peculiar behaviour of the light curve and perhaps, even shows $``$ 0.5 mag brightening of the $`R`$ magnitude with $`\alpha =0.5\pm 1.0`$ during $`\mathrm{\Delta }t=1.51.8`$ day. This could be an indication of a rising phase similar to that seen in GRB 970228 (Guarnieri et al. 1997) and GRB 970508 (Castro-Tirado et al. 1998). Contrary to most of the earlier GRB afterglows, light curve of GRB 000301C can not be fitted by a single power-law (see also Masetti et al. 2000, Rhoads & Fruchter 2000, Berger et al. 2000; Jensen et al. 2000). Overall the OT flux decay seems to have broken power-law as expected in GRB afterglows having jet-like relativistic ejecta (Sari et al. 1999; Rhoads 1999). This appears to be superimposed with some shorter time flux variability especially during $`\mathrm{\Delta }t<8`$ day. Among equally well monitored GRB afterglows, GRB 000301C appears therefore peculiar. Correlated variability can be clearly noticed in $`B,R`$ and $`I`$ passbands. The lack of such apparent correlation in the light curves of other passbands is most probably due to non-strict simultaneity of the data points. Broken power-law in these light curves can be empirically fitted by functions of the form (see Rhoads & Fruchter 2000)
$`F(t)=2F_0/[(t/t_b)^{\alpha _1\beta }+(t/t_b)^{\alpha _2\beta }]^{1/\beta }`$,
where $`\alpha _1`$ and $`\alpha _2`$ are asymptotic power-law slopes at early and late times with $`\alpha _1<\alpha _2`$ and $`\beta >0`$. $`\beta `$ controls the sharpness of the break, with larger $`\beta `$ implying a sharper break. With $`\beta =1`$, this function becomes the same that Stanek et al. (1999) fit the optical light curve of GRB 990510 afterglow. $`F_0`$ is the flux at the cross-over time $`t_b`$. The function describes a light curve falling as $`t^{\alpha _1}`$ at $`t<<t_b`$ and $`t^{\alpha _2}`$ at $`t>>t_b`$. The function can be written in magnitudes as
$`m=m_b+\frac{2.5}{\beta }[log_{10}\{(t/t_b)^{\alpha _1\beta }+(t/t_b)^{\alpha _2\beta }\}log_{10}(2)]`$,
where $`m_b`$ is the magnitude at time $`t_b`$. In jet models, an achromatic break in the light curve is expected when the jet makes the transition to sideways expansion after the relativistic Lorentz factor drops below the inverse of the opening angle of the initial beam. Slightly later, the jet begins a lateral expansion which causes a further steepening of the light curve. Before fitting jet model to the light curve to derive accurate flux decay parameters of the afterglow, it is mandatory to deconvolve the short term variability component. Otherwise, it will confuse the determination of $`t_b`$. Perhaps it is the main reason for having a range of $`t_b`$ values in the literature. The short term variability component of the light curve is determined as described below.
In order to minimize the effects of short term variability on the determination of the parameters of the fireball model, Berger et al. (2000) fitted the entire data set from radio to optical simultaneously and derive $`t_b=7.5\pm 0.5`$ days, $`\alpha _1=1.28`$ for $`t<t_b`$ and $`\alpha _2=2.70`$ for $`t>t_b`$ as global parameters for the jet fireball model. The observed $`U,B,V,R,I,`$ and $`K^{^{}}`$ fluxes are divided by the values obtained from the jet model fit yields, as also is noticed by Berger et al. (2000) that the variability is simultaneous and of similar amplitude in all bands (see upper right corner box in Fig. 3). There is a sharp rise and decline centered on $`\mathrm{\Delta }t=4`$ day. Berger et al. (2000) also found similar variability at 250 GHz. All these indicate that variability is the result of a real physical process which produces simultaneously similar level of absolute variation over a large range in frequency. Berger et al. (2000) therefore explain this fluctuation in terms of non-uniform ambient density which varied by about a factor of 3.
We use the densely covered observations in $`B`$, $`R`$ and $`I`$ to determine the parameters of jet model using the above function. For this, the short term variability was deconvolved from the observed light curves. It has been noticed that the minimum value of $`\chi ^2`$ is achieved for $`\beta 5`$. This indicates that the observed break in the light curve is sharp and is unlike the smooth break observed in the optical light curve of GRB 990510 (cf. Stanek et al. 1999; Harrison et al. 1999). In order to avoid a fairly wide range of model parameters for a comparable $`\chi ^2`$ due to degeneracy between $`t_b,\alpha _2`$ and $`\beta `$, we have used fixed value of $`\beta =5`$ in our further analyses. The least square best fitted parameters $`t_b,m_b,\alpha _1,`$ and $`\alpha _2`$ have values 7.51$`\pm `$0.63, 22.15$`\pm `$0.15, 1.18$`\pm `$0.14 and 3.01$`\pm `$0.53 respectively in $`R`$. The corresponding values are 8.27$`\pm `$1.11, 23.20$`\pm `$0.24, 1.24$`\pm `$0.20 and 3.48$`\pm `$2.07 respectively in $`B`$ and 7.27$`\pm `$1.04, 21.64$`\pm `$0.26, 1.17$`\pm `$0.29 and 2.92$`\pm `$2.93 respectively in $`I`$. This indicates that average values of $`t_b,\alpha _1,`$ and $`\alpha _2`$ are 7.6$`\pm `$0.5 day, $`1.2\pm 0.1`$ and $`3.0\pm 0.5`$ respectively. The light curves derived with these averaged parameters using the jet model are shown by dotted curves in the $`U,V,J`$ and $`K^{^{}}`$ passbands. This clearly indicates the presence of simultaneous short term variability in all passbands. We therefore conclude in agreement with Masetti et al. (2000) and Rhoads & Fruchter (2000) that optical and near-IR flux decays of GRB 000301C afterglow are peculiar in comparison to other such well observed GRB afterglows.
### 3.1 Spectral index of the GRB 000301C afterglow
The flux distribution of the GRB 000301C afterglow has been studied using the broadband photometric measurements listed in Table 1 along with the published radio, millimeter and ultra-violet observations. We used the reddening map provided by Schlegel, Finkbeiner & Davis (1998) for estimating Galactic interstellar extinction towards the burst and found a small value of $`E(BV)=0.05`$ mag. We used the standard Galactic extinction reddening curve given by Mathis (1990) in converting apparent magnitudes into fluxes and used the effective wavelengths and normalisations by Bessell (1979) for $`B,V,R`$ and $`I`$ and by Bessell & Brett (1988) for $`J`$ and $`K^{^{}}`$. The fluxes thus derived are accurate to $``$ 10%. Fig. 4 shows the spectrum of GRB 000301C afterglow from optical to radio region. The fluxes closest to $`\mathrm{\Delta }t=4.8`$ day at 1.4 GHz, 4.86 GHz, 8.46 GHz, 15 GHz, 22.5 GHz, 100 GHz, 250 GHz and 350 GHz are taken from Berger et al. (2000). It is observed that as the frequency decreases the flux increases from optical to millimeter wavelengths and then it turns over. The spectrum thus can be described by a single power law in some frequency interval as $`F_\nu \nu ^\beta `$, where $`F_\nu `$ is the flux at frequency $`\nu `$ and $`\beta `$ is the spectral index. In the optical to millimeter region, the value of $`\beta `$ is $`0.73\pm `$0.06 at $`\mathrm{\Delta }t=`$ 4.8 day. The optical-near-IR spectrum has not changed significantly (see Table 2 and Fig. 4) and has average value around $`1.0`$. This is in agreement with a single value of $`\beta =1.1`$ derived from the low-resolution spectrum taken on 2000 March 3.47 UT by Feng et al. (2000) in the wavelength range of 0.3 to 0.6 $`\mu `$m. The HST observations taken around $`\mathrm{\Delta }t`$ = 33.5 day by Fruchter et al. (2000b) also indicate similar slope. All these, perhaps, indicate no change in the spectral slope of GRB 000301C at later times. There is thus no evidence for a cooling break passing through the optical band on these time scales. This is unlike GRB 980326 ( Bloom et al. 1999) and GRB 970228 (Fruchter et al. 1999; Galama et al. 2000) where spectral index changed to a value of $`\beta 3.0`$ after $`\mathrm{\Delta }t>`$ 20 days. The spectral slope at radio to millimeter frequencies is generally expected to be $`+1/3`$ at these early times. However, the observed slope is much larger with a value of $`+0.90\pm 0.08`$. The peak frequency seems to lie in millimeter region. This peak frequency is thus similar to that of GRB 970508 (cf. Galama et al. 1998) but different from that of GRB 990123 (Galama et al. 1999) where the peak is in radio region and that of GRB 971214 for which the peak is in optical/near-infrared waveband (Ramaprakash et al. 1998). From this, one may infer that the synchrotron peak frequency may span a large range in GRB afterglows.
## 4 Comparison with the synchrotron emisson model
It is generally believed that the observed afterglow results from slowing down of a relativistic shell on the external ISM and therefore is produced by external shocks. Recent afterglow observations of GRBs show that a relativistic blast wave, in which the highly relativistic electrons radiate via synchrotron mechanism, provides a generally good description of the observed properties. Here we will discuss briefly the implications of the observed flux decay exponent $`\alpha `$ and the spectral slope index $`\beta `$ in different wavelength range for such models. All models have the flux as $`F(\nu ,t)t^\alpha \nu ^\beta `$ for a range of frequencies and times that contain no spectral breaks. In each model $`\alpha `$ and $`\beta `$ are functions of $`p`$ only, the power-law exponent of the electron Lorentz factor. The measurement of either one of $`\alpha `$ or $`\beta `$ therefore fixes $`p`$ and predicts the other one.
In order to study the expected changes in the spectral indices with $`\alpha `$, we derive the value of $`\beta `$ in optical to near-IR region at different epochs. They are plotted inside a box in Fig. 4 and listed in Table 2. Where necessary, flux measurements were interpolated between adjacent data points at one wavelength in order to determine a contemporaneous flux with another wavelength using the measurements listed in Table 1. There is no evidence for statistically significant large ($`\mathrm{\Delta }\beta 0.5`$) variation in the spectral index on these time scales.
For comparison with model predictions, we assume that our observations are in the slow cooling regime and the $`\nu _m`$ has passed optical but not the cooling frequency, $`\nu _c`$ which most probably lies above optical region. Following Sari et al. (1999), values of $`\alpha `$ and $`p`$ are predicted using observed value of $`\beta `$ for the spherical model of the afterglow. They are listed in Table 2. The observed flux decay constant at early times agrees well with the predicted ones given in Table 2 while exactly opposite is the case at late times for spherical afterglow emission. But the value of flux decay constant $`\alpha `$ is expected to approach the electron energy distribution index $`p`$, when the evolution of the afterglow is dominated by the spreading of the jet. On the other hand, the value of $`\beta `$ is the same for both spherical and jet models. Since the observed values of $`\alpha `$ for late times agree with the predicted values of $`p`$ and hence to the values of $`\alpha `$ in jet model, we conclude that afterglow emission from GRB 000301C is of jet type and not spherical.
## 5 The energetics of the GRB 000301C
Redshift determination of $`z=2.0335\pm 0.0003`$ (Castro et al. 2000) for the GRB 000301C afterglow yields a minimum luminosity distances of 16.6 Gpc for a standard Friedmann cosmological model with Hubble constant $`H_0`$ = 65 km/s/Mpc, cosmological density parameter $`\mathrm{\Omega }_0`$ = 0.2 and cosmological constant $`\mathrm{\Lambda }_0`$ = 0 (if $`\mathrm{\Lambda }_0>0`$ then the inferred distances would increase). The GRB 000301C thus becomes the second farthest GRB after GRB 971214 (Kulkarni et al. 1998) amongst the GRBs with known redshift measurements so far.
As there is no published observed fluence in any energy range for this GRB, we estimate it indirectly assuming that present GRB event may also have the ratio between optical flux density and gamma-ray energy fluence similar to those observed so far which is $`10^{23}`$ (see Table 3 in Briggs et al. 1999). Taking $`R=20`$ mag at $`\mathrm{\Delta }t`$ 1 day, this ratio yields an energy fluence of at least $`10^5`$ ergs/cm<sup>2</sup> above 20 KeV. Considering isotropic energy emission and this observed fluence and using the inferred luminosity distances, we estimate the $`\gamma `$ray energy release to be at least $`3.4\times 10^{53}`$ ergs $`0.2M_{}c^2`$ for this GRB. Considering the different fluence energy ranges used, this is not too different from the values $`5.4\times 10^{52}`$ ergs and $`2.27\times 10^{52}`$ ergs derived by Breger et al. (2000) and Jensen et al. (2000) respectively. Theoretical models predict that brightness of the prompt optical flash can be as bright as 9 – 10 mag (Sari & Piran 1999); as was observed in the case of the GRB 990123, the only prompt optical emission detected so far. At the optical distance of GRB 000301C, this implies a peak optical luminosity of $`6.3\times 10^{16}`$ times the solar luminosity, if the prompt optical emission is of similar order. This is about a million times the luminosity of a normal galaxy and about a thousand times the luminosity of the brightest quasars known. The present energy and $`t_b`$ estimates imply a jet opening angle of $`0.15n^{1/8}`$ radian, where $`n`$ is the number density (in units of cm<sup>-3</sup>) of the ambient medium. This means that the actual energy released from the GRB 000301C is reduced by a factor of $`90`$ relative to the isotropic value and becomes $`3.8\times 10^{51}`$ ergs.
Of the over dozen GRBs with known redshifts, six with total fluence energies $`>`$ 20 keV in excess of 10<sup>53</sup> erg (assuming isotropic emission) are GRB 000301C (discussed here); GRB 991216 and GRB 991208 (Sagar et al. 2000); GRB 990510 (Harrison et al. 1999); GRB 990123 (Andersen et al. 1999; Galama et al. 1999) and GRB 971214 (Kulkarni et al. 1998). Recent observations suggest that GRBs are associated with stellar deaths, and not with quasars or the nuclei of galaxies as some GRBs are found off-set from their host galaxy. However, release of huge amount of isotropic energy of $`10^{53}`$ erg or more is essentially incompatible with the popular stellar death models (coalescence of neutron stars and death of massive stars). Recent observations seem to indicate non-isotropic emission as the most plausible way to reduce the enormous energy release. Indeed, almost all energetic sources in astrophysics such as pulsars, quasars and accreting stellar black holes display jet-like geometry and hence, non-isotropic emission. Beaming reduces the estimated energy by a factor of 10 - 300, depending upon the size of its opening angle (Sari et al. 1999). The $`\gamma `$ray energy released then becomes $`10^{52}`$ erg, a value within reach of current popular models for the origin of GRBs (see Piran 1999 and references therein).
## 6 Discussions and Conclusions
Prompt $`\gamma `$ray emission light curve of the GRB 000301C burst shows, unlike most of the GRB events, only one strong peak with a flux of 3.7$`\pm `$0.7 Crab in the 5 – 12 KeV energy range. Using optical and near-IR observations, we obtained the values of flux decay constants and spectral indices. Light curves of the GRB 000301C afterglow emissions are peculiar. The light curves show a steepening superposed by a short term flare type variability which could be detected mainly due to the dense observations in $`R`$ filter. A large fraction of these observations have been carried out using the 1-m class optical telescopes. This indicates that in future these telescopes, as large amount of observing time is available on them, will play an important role in understanding the origin of such short term variability in the light curves of GRBs during early times. The overall flux decay in observed light curves are well understood in terms of a jet model. The parameters of the jet model are derived by fitting least square non-linear fit to the light curves obtained after deconvolving the short term variability from the observed light curves. The flux decay constants at early and late times are 1.2$`\pm `$0.1 and 3.0$`\pm `$0.5 respectively. The value of jet break time is 7.6$`\pm `$0.5 day. Before deriving any further conclusions from the light curve of GRB 000301C afterglow, we compare it with other well studied GRBs. Except GRB 990123, GRB 990510 and GRB 991216, all exhibit, at both early and late times a single power-law decay, generally $``$ 1.2, a value reasonable for spherical expansion in the fireball synchrotron model. GRB 000301C thus becomes the fourth burst for which a strong break in the light curve is clearly observed. Such breaks were observed first in the optical light curves of the afterglow of GRB 990123 (Castro-Tirado et al. 1999; Kulkarni et al. 1999) and recently in that of GRB 990510 (Harrison et al. 1999, Stanek et al. 1999) and GRB 991216 (Halpern et al. 2000, Sagar et al. 2000). They have generally been considered as evidence for collimation of the jet-like relativistic GRB ejecta in accordance with the prediction by recent theoretical models (Mészáros & Rees 1999; Rhoads 1999; Sari et al. 1999).
The quasi-simultaneous spectral energy distributions determined in optical and near-IR regions for various epochs indicate that spectral index of the GRB 000301C afterglow has not changed significantly during a period of about 35 days after the burst. The value of $`\beta `$ is $`1.0`$. However, the early time flux decay constant has varied from 1.2$`\pm `$0.1 to 3.0$`\pm `$0.5. A steepening of flux decay constant with no corresponding change in spectral index is attributed to the presence of a jet in the GRB 000301C OT. The jet breaks around 7.6 days after the burst.
Redshift determination yields a minimum distance of 16.6 Gpc, if one assumes standard Friedmann cosmology with $`H_o=65`$ km/s/Mpc, $`\mathrm{\Omega }_0=0.2`$ and $`\mathrm{\Lambda }_0=0`$. GRB 000301C is thus at cosmological distance and becomes the second farthest amongst the GRBs with known distances so far. Considering isotropic energy emission, we estimate enormous amount of the $`\gamma `$ray energy release ($`10^{53}`$ erg) above 20 KeV. This high energy is reduced to $`<10^{52}`$ erg when effects of non-isotropic emission are considered due to the presence of a jet of an opening angle of 0.15 radian in the GRB 000301C.
The peculiarity in the light curves of GRB 000301C seems to be due to superposition of a short term achromatic variability over a large frequency range on the overall steepening in the flux of the GRB 000301C. In separating the two components of the observed light curves, dense as well as multi-wavelength observations during early times have played major role. Such observations of recent GRBs have started revealing features which require explanations other than generally accepted so far indicating that there may be yet new surprises in GRB afterglows.
Acknowledgements: Suggestions/comments given by anonymous referee improved the scientific content of the paper significantly. 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.
References:
* Andersen M. I. et al., 1999, Science, 283, 2075
* Bernabei S. et al., 2000, GCN Observational Report No. 571
* Bertoldi F., 2000, GCN Observational Report 580
* Bessell M.S., 1979, PASP, 91, 589
* Bessell M.S., Brett J.M., 1988, PASP, 100, 1134
* Bhargavi S.G., Cowsik, R., 2000, GCN Observational Report No. 630
* Bloom J.S. et al., 1999, Nature, 401, 453
* Berger E., Frail, D.A., 2000, GCN Observational Report No. 589
* Berger E. et al., 2000, Astro-ph/0005465
* Briggs M.S. et al., 1999, ApJ, 524, 82
* Castro S. M. et al., 2000, GCN Observational Report No. 605
* Castro-Tirado A. J. et al., 1998, Science, 279, 1012
* Castro-Tirado A. J. et al., 1999, Science, 283, 2069
* Eracleous, M., Shetrone M., Sigurdsson S., Meszaros P., Wheeler J.C., Wang L., 2000, GCN Observational Report 584
* Feng M. et al., 2000, GCN Observational Report No. 607
* Fruchter A.S. et al., 1999, ApJ, 516, 683
* Fruchter A.S. et al., 2000a, GCN Observational Report No. 602
* Fruchter A.S. et al., 2000b, GCN Observational Report No. 627
* Fynbo J.P.U., Jensen B.L., Hjorth J., Pedersen H., Gorosabel J., 2000a, GCN Observational Report No. 570
* Fynbo J.P.U., Jensen B.L., Hjorth J., Pedersen H., Gorosabel J., 2000b, GCN Observational Report No. 576
* Galama, T.J. et al., 1998, ApJ, 500, L97
* Galama T.J. et al., 1999, Nature, 398, 394
* Galama T.J. et al., 2000, ApJ Submitted
* Gal-Yam A., Ofek E., Maoz D., Leibowitz, E.M., 2000, GCN Observational Report No. 593
* Garnavich P., Barmby P., Jha S., Stanek K., 2000a, GCN Observational Report No. 573
* Garnavich P., Barmby P., Jha S., Stanek K., 2000b, GCN Observational Report No. 581
* Guarnieri et al., 1997, A&A, 328, L13
* Halpern J.P., Mirabal N., Lawrence, S., 2000a, GCN Observational Report No. 578
* Halpern J.P., Mirabal N., Lawrence, S., 2000b, GCN Observational Report No. 582
* Halpern J.P., Mirabal N., Lawrence, S., 2000c, GCN Observational Report No. 585
* Halpern J.P., Kemp J., 2000, GCN Observational Report No. 604
* Halpern J.P. et al., 2000, ApJ Submitted
* Harrison F.A. et al., 1999, ApJ, 523, L121
* Henden A., 2000, GCN Observational Report 583
* Hurley K., et al., 2000, ApJ Lett (accepted)
* Jensen B.L. et al., 2000, A&A Submitted/astro-ph/0005609
* Kobayashi N., Goto M., Terada H., Tokunaga A.T., 2000, GCN Observational Report No. 577
* Kulkarni S.R. et al., 1998, Nature, 393, 35
* Kulkarni S.R. et al., 1999, Nature, 398, 389
* Masetti N. et al., 2000, A&A Submitted/astro-ph/0004186
* Mathis J.S., 1990, ARAA, 28, 37
* Mészáros P., Rees M. J., 1999, MNRAS, 306, L39
* Mujica R., 2000, GCN Observational Report No. 597
* Piran T., 1999, Physics Reports 314, 575
* Ramaprakash, A.N. et al., 1998, Nature, 393, 43
* Rhoads J.E., 1999, ApJ, 525, 737
* Rhoads J.E., Fruchter A.S., 2000, ApJ Submitted/astro-ph/0004057
* Sagar R., 2000, Current Science 78, 1076
* Sagar R., Pandey A.K., Mohan V., Yadav R.K.S., Nilakshi, Bhattacharya D., Castro-Tirado A.J., 1999, BASI, 27, 3
* Sagar R., Mohan V., Pandey A.K., Pandey S.B., Castro-Tirado A.J., 2000, BASI, 28, 15
* Sari R., Piran T., 1999, ApJ, 517, L109
* Sari R., Piran T., Halpern J. P., 1999, ApJ, 519, L17
* Schlegel D.J., Finkbeiner D.P., Davis M., 1998, ApJ, 500, 525
* Smette A. et al., 2000, GCN Observational Report No. 603
* Smith D.A., Hurley K., Cline T., 2000, GCN observational Report No. 568
* Stanek K. Z. et al., 1999, ApJ 522, L39
* Stecklum B. et al., 2000, GCN Observational Report No. 572
* Veillet C., 2000a, GCN Observational Report No. 588
* Veillet C., 2000b, GCN Observational Report No. 598
* Veillet C., 2000c, GCN Observational Report No. 610
* Veillet C., 2000d, GCN Observational Report No. 611
|
warning/0004/hep-ph0004257.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The dominant weak decays of hadrons containing a heavy quark, $`c`$ or $`b`$, are caused by the decay of the heavy quark. In the limit of a very large mass $`m_Q`$ of a heavy quark $`Q`$ the parton picture of the hadron decay should set in, where the inclusive decay rates of hadrons, containing $`Q`$, mesons ($`Q\overline{q}`$) and baryons ($`Qqq`$), are all the same and equal to the inclusive decay rate $`\mathrm{\Gamma }_{parton}(Q)`$ of the heavy quark. Yet, the known inclusive decay rates are conspicuously different for different hadrons, especially for charmed hadrons, whose lifetimes span a range of more than one order of magnitude from the shortest $`\tau (\mathrm{\Omega }_c)=0.064\pm 0.020`$ ps to the longest $`\tau (D^+)=1.057\pm 0.015`$ ps, while the differences of lifetime among $`b`$ hadrons are substantially smaller. The relation between the relative lifetime differences for charmed and $`b`$ hadrons reflects the fact that the dependence of the inclusive decay rates on the light quark-gluon ‘environment’ in a particular hadron is a pre-asymptotic effect in the parameter $`m_Q`$, which effect vanishes as an inverse power of $`m_Q`$ at large mass.
A theoretical framework for systematic description of the leading at $`m_Q\mathrm{}`$ term in the inclusive decay rate $`\mathrm{\Gamma }_{parton}(Q)m_Q^5`$ as well as of the terms relatively suppressed by inverse powers of $`m_Q`$ is provided by the operator product expansion (OPE) in $`m_Q^1`$. Existing theoretical predictions for inclusive weak decay rates are in a reasonable agreement, within the expected range of uncertainty, with the data on lifetimes of charmed particles and with the so far available data on decays of $`B`$ mesons. The only outstanding piece of present experimental data is on the lifetime of the $`\mathrm{\Lambda }_b`$ baryon: $`\tau (\mathrm{\Lambda }_b)/\tau (B_d)0.8`$, for which ratio a theoretical prediction, given all the uncertainty involved, is unlikely to produce a number lower than 0.9. The number of available predictions for inclusive decay rates of charmed and $`b`$ hadrons is sufficiently large for future experimental studies to firmly establish the validity status of the OPE based theory of heavy hadron decays, and, in particular, to find out whether the present contradiction between the theory and the data on $`\tau (\mathrm{\Lambda }_b)/\tau (B_d)`$ is a temporary difficulty, or an evidence of fundamental flaws in theoretical understanding.
It is a matter of common knowledge that application of OPE to decays of charmed and $`b`$ hadrons has potentially two caveats. One is that the OPE is used in the Minkowsky kinematical domain, and therefore relies on the assumption of quark-hadron duality at the energies involved in the corresponding decays. In other words, it is assumed that sufficiently many exclusive hadronic channels contribute to the inclusive rate, so that the accidentals of the low-energy resonance structure do not affect the total rates of the inclusive processes. Theoretical attempts at understanding the onset of the quark-hadron duality are so far limited to model estimates , not yet suitable for direct quantitative evaluation of possible deviation from duality in charm and $`b`$ decays. This point presents the most fundamental uncertainty of the OPE based approach, and presently can only be clarified by confronting theoretical predictions with experimental data. The second possible caveat in applying the OPE technique to inclusive charm decays is that the mass of the charm quark, $`m_c`$, may be insufficiently large for significant suppression of higher terms of the expansion in $`m_c^1`$. The relative lightness of the charm quark, however, accounts for a qualitative, and even semi-quantitative, agreement of the OPE based predictions with the observed large spread of the lifetimes of charmed hadrons: the nonperturbative effects, formally suppressed by $`m_c^2`$ and $`m_c^3`$ are comparable with the ‘leading’ parton term and describe the hierarchy of the lifetimes.
Another uncertainty of a technical nature arises from poor knowledge of matrix elements of certain quark operators over hadron, arising as terms in OPE. These can be estimated within theoretical models, with inevitable ensuing model dependence, or, where possible, extracted from the experimental data. With these reservations spelled out, we discuss here the OPE based description of inclusive weak decays of charm and $`b`$ hadrons, with emphasis on specific experimentally testable predictions, and on the measurements, which would less rely on model dependence of the estimates of the matrix elements, thus allowing to probe the OPE predictions at a fundamental level.
## 2 OPE for inclusive weak decay rates
The optical theorem of the scattering theory relates the total decay rate $`\mathrm{\Gamma }_H`$ of a hadron $`H_Q`$ containing a heavy quark $`Q`$ to the imaginary part of the ‘forward scattering amplitude’. For the case of weak decays the latter amplitude is described by the following effective operator
$$L_{eff}=2\mathrm{Im}\left[id^4xe^{iqx}T\{L_W(x),L_W(0)\}\right],$$
(1)
in terms of which the total decay rate is given by<sup>1</sup><sup>1</sup>1We use here the non-relativistic normalization for the heavy quark states: $`Q|Q^{}Q|Q=1`$.
$$\mathrm{\Gamma }_H=H_Q|L_{eff}|H_Q.$$
(2)
The correlator in equation (1) in general is a non-local operator. However at $`q^2=m_Q^2`$ the dominating space-time intervals in the integral are of order $`m_Q^1`$ and one can expand the correlator in $`x`$, thus producing an expansion in inverse powers of $`m_Q`$. The leading term in this expansion describes the parton decay rate of the quark. For instance, the term in the non-leptonic weak Lagrangian $`\sqrt{2}G_FV(\overline{q}_{1L}\gamma _\mu Q_L)(\overline{q}_{2L}\gamma _\mu q_{3L})`$ with $`V`$ being the appropriate combination of the CKM mixing factors, generates through eq.(1) the leading term in the effective Lagrangian
$$L_{eff,nl}^{(0)}=|V|^2\frac{G_F^2m_Q^5}{64\pi ^3}\eta _{nl}\left(\overline{Q}Q\right),$$
(3)
where $`\eta _{nl}`$ is the perturbative QCD radiative correction factor. This expression reproduces the well known formula for the inclusive non-leptonic decay rate of a heavy quark, associated with the underlying process $`Qq_1q_2\overline{q}_3`$, due to the relation $`H_Q|\overline{Q}Q|H_QH_Q|Q^{}Q|H_Q=1`$, which is valid up to corrections of order $`m_Q^2`$. One also sees form this example, that in order to separate individual semi-inclusive decay channels, e.g. non-leptonic with specific flavor quantum numbers, or semi-leptonic, one should simply pick up the corresponding relevant part of the weak Lagrangian $`L_W`$, describing the underlying process, to include in the correlator (1).
The general expression for first three terms in the OPE for $`L_{eff}`$ has the form
$`L_{eff}=L_{eff}^{(0)}+L_{eff}^{(2)}+L_{eff}^{(3)}=`$
$`c^{(0)}{\displaystyle \frac{G_F^2m_Q^5}{64\pi ^3}}\left(\overline{Q}Q\right)+c^{(2)}{\displaystyle \frac{G_F^2m_Q^3}{64\pi ^3}}\left(\overline{Q}\sigma ^{\mu \nu }G_{\mu \nu }Q\right)+{\displaystyle \frac{G_F^2m_Q^2}{4\pi }}{\displaystyle \underset{i}{}}c_i^{(3)}(\overline{q}_i\mathrm{\Gamma }_iq_i)(\overline{Q}\mathrm{\Gamma }_i^{}Q),`$ (4)
where the superscripts denote the power of $`m_Q^1`$ in the relative suppression of the corresponding term in the expansion with respect to the leading one, $`G_{\mu \nu }`$ is the gluon field tensor, $`q_i`$ stand for light quarks, $`u,d,s`$, and, finally, $`\mathrm{\Gamma }_i`$, $`\mathrm{\Gamma }_i^{}`$ denote spin and color structures of the four-quark operators. The coefficients $`c^{(a)}`$ depend on the specific part of the weak interaction Lagrangian $`L_W`$, describing the relevant underlying quark process.
One can notice the absence in the expansion (4) of a term suppressed by just one power of $`m_Q^1`$, due to non-existence of operators of suitable dimension. Thus the decay rates receive no correction of relative order $`m_Q^1`$ in the limit of large $`m_Q`$, and the first pre-asymptotic corrections appear only in the order $`m_Q^2`$.
The mechanisms giving rise to the three discussed terms in OPE are shown in Figure 1. The first, leading term corresponds to the parton decay, and does not depend on the light guark and gluon ‘environment’ of the heavy quark in a hadron. The second term describes the effect on the decay rate of the gluon field that a heavy quark ‘sees’ in a hadron. This term in fact is sensitive only to the chromomagnetic part of the gluon field, and contains the operator of the interaction of heavy quark chromomagnetic moment with the chromomagnetic field. Thus this term depends on the spin of the heavy quark, but does not depend on the flavors of the light quarks or antiquarks. Therefore this effect does not split the inclusive decay rates within flavor SU(3) multiplets of heavy hadrons, but generally gives difference of the rates, say, between mesons and baryons. The dependence on the light quark flavor arises from the third term in the expansion (4) which explicitly contains light quark fields. Historically, this part is interpreted in terms of two mechanisms : the weak scattering (WS) and the Pauli interference (PI). The WS corresponds to a cross-channel of the underlying decay, generically $`Qq_1q_2\overline{q}_3`$, where either the quark $`q_3`$ is a spectator in a baryon and can undergo a weak scattering off the heavy quark: $`q_3Qq_1q_2`$, or an antiquark in meson, say $`\overline{q}_1`$, weak-scatters (annihilates) in the process $`\overline{q}_1Qq_2\overline{q}_3`$. The Pauli interference effect arises when one of the final (anti)quarks in the decay of $`Q`$ is identical to the spectator (anti)quark in the hadron, so that an interference of identical particles should be taken into account. The latter interference can be either constructive or destructive, depending on the relative spin-color arrangement of the (anti)quark produced in the decay and of the spectator one, thus the sign of the PI effect is found only as a result of specific dynamical calculation. In specific calculations, however, WS and PI arise from the same terms in OPE, depending on the hadron discussed, and technically there is no need to resort to the traditional terminology of WS and PI.
In what follows we discuss separately the effects of the three terms in the expansion (4) and their interpretation within the existing and future data.
## 3 The parton decay rate
The leading term in the OPE amounts to the perturbative expression for the decay rate of a heavy quark. In $`b`$ hadrons the contribution of the subsequent terms in OPE is at the level of few percent, so that the perturbative part can be confronted with the data in its own right. In particular, for the $`B_d`$ meson the higher terms in OPE contribute only about 1% of the total non-leptonic as well as of the semileptonic decay rate. Thus the data on these rates can be directly compared with the leading perturbative term in OPE.
The principal theoretical topic, associated with this term is the calculation of QCD radiative corrections, i.e. of the factor $`\eta _{nl}`$ in eq.(3) and of a similar factor, $`\eta _{nl}`$, for semileptonic decays. It should be noted, that even at this, perturbative, level there is a known long-standing problem between the existing data and the theory in that the current world average for the semileptonic branching ratio for the $`B`$ mesons, $`B_{sl}(B)=10.45\pm 0.21\%`$, is somewhat lower than the value $`B_{sl}(B)11.5`$ preferred from the present knowledge of theoretical QCD radiative corrections to the ratio of non-leptonic to semileptonic decay rates (see e.g. ). However, this apparent discrepancy may in fact be due to insufficient ‘depth’ of perturbative QCD calculation of the ratio $`\eta _{nl}/\eta _{sl}`$. In order to briefly elaborate on this point, we notice that the standard way of analyzing the perturbative radiative corrections in the nonleptonic decays is through the renormalization group (RG) summation of the leading log terms and the first next-to-leading terms in the parameter $`L\mathrm{ln}(m_W/m_b)`$. For the semileptonic decays the logarithmic dependence on $`m_W/m_b`$ is absent in all orders due to the weak current conservation at momenta larger than $`m_b`$, thus the correction is calculated by the standard perturbative technique, and a complete expression in the first order in $`\alpha _s`$ is available both for the total rate and for the lepton spectrum . In reality however the parameter $`L2.8`$ is not large, and non-logarithmic terms may well compete with the logarithmic ones. This behavior is already seen from the known expression for the logarithmic terms: when expanded up to the order $`\alpha _s^2`$ the result of Ref. for the rate of decays with single final charmed quark takes the form
$$\frac{\mathrm{\Gamma }(bc\overline{u}d)+\mathrm{\Gamma }(bc\overline{u}s)}{3\mathrm{\Gamma }(bce\overline{\nu })}=1+\frac{\alpha _s}{\pi }+\frac{\alpha _s^2}{\pi ^2}\left[4L^2+\left(\frac{7}{6}+\frac{2}{3}c(m_c^2/m_b^2)\right)L\right],$$
(5)
where, in terms of notation of Ref., $`c(a)=c_{22}(a)c_{12}(a)`$. The behavior of the function $`c(a)`$ is known explicitly and is quite weak: $`c(0)=19/2`$, $`c(1)=6`$, and $`c(m_c^2/m_b^2)9.0`$ for the realistic mass ratio $`m_c/m_b0.3`$. One can see that the term with the single logarithm $`L`$ contributes about two thirds of that with $`L^2`$ in the term quadratic in $`\alpha _s`$. Under such circumstances the RG summation of the terms with powers of $`L`$ does not look satisfactory for numerical estimates of the QCD effects, at least at the so far considered level of the first next-to-leading order terms, and the next-to-next-to-leading terms can be equally important as the two known ones, which would eliminate the existing impasse between the theory and the data on $`B_{sl}(B)`$. One can present some arguments that this is indeed the case for the $`b`$ quark decay, although a complete calculation of these corrections is still unavailable.
## 4 Chromomagnetic and time dilation effects in decay rates
The corrections suppressed by two powers of $`m_Q^1`$ to inclusive decay rates arise from two sources : the $`O(m_Q^2)`$ corrections to the matrix element of the leading operator, $`(\overline{Q}Q`$, and the second term in OPE (4) containing the chromomagnetic interaction. The expression for the matrix element of the leading operator with the correction included is written in the form
$$H_Q|\overline{Q}Q|H_Q=1\frac{\mu _\pi ^2(H_Q)\mu _g^2(H_Q)}{2m_Q^2}+\mathrm{},$$
(6)
where $`\mu _\pi ^2`$ and $`\mu _g^2`$ are defined as
$`\mu _\pi ^2=H_Q|\overline{Q}(i\stackrel{}{D})^2Q|H_Q,`$
$`\mu _g^2=H_Q|\overline{Q}{\displaystyle \frac{1}{2}}\sigma ^{\mu \nu }G_{\mu \nu }Q|H_Q,`$ (7)
with $`D`$ being the QCD covariant derivative. The correction in equation (6) in fact corresponds to the time dilation factor $`m_Q/E_Q`$, for the heavy quark decaying inside a hadron, where it has energy $`E_Q`$, which energy is contributed by the kinetic part ($`\mu _\pi ^2`$) and the chromomagnetic part ($`\mu _g^2`$). The second term in OPE describes the effect of the chromomagnetic interaction in the decay process, and is also expressed through $`\mu _g^2`$.
The explicit formulas for the decay rates, including the effects up to the order $`m_Q^2`$ are found in and for decays of the $`b`$ hadrons read as follows. For the semileptonic decay rate
$$\mathrm{\Gamma }_{sl}(H_b)=\frac{|V_{cb}|^2G_F^2m_b^5}{192\pi ^3}H_b|\overline{b}b|H_b\left[1+\frac{\mu _g^2}{m_b^2}\left(\frac{x}{2}\frac{d}{dx}2\right)\right]\eta _{sl}I_0(x,\mathrm{\hspace{0.17em}0},\mathrm{\hspace{0.17em}0}),$$
(8)
and for the non-leptonic decay rate
$$\mathrm{\Gamma }_{nl}(H_b)=\frac{|V_{cb}|^2G_F^2m_b^5}{64\pi ^3}H_b|\overline{b}b|H_b\left\{\left[1+\frac{\mu _g^2}{m_b^2}\left(\frac{x}{2}\frac{d}{dx}2\right)\right]\eta _{nl}I(x)8\eta _2\frac{\mu _g^2}{m_b^2}I_2(x)\right\}.$$
(9)
These formulas take into account only the dominant CKM mixing $`V_{cb}`$ and neglect the small one, $`V_{ub}`$. The following notation is also used: $`x=m_c/m_b`$, $`I_0(x,y,z)`$ stands for the kinematical suppression factor in a three-body weak decay due to masses of the final fermions. In particular,
$`I_0(x,0,0)=(1x^4)(18x^2+x^4)24x^4\mathrm{ln}x,`$ (10)
$`I_0(x,x,0)=(114x^22x^412x^6)\sqrt{14x^2}+24(1x^4)\mathrm{ln}{\displaystyle \frac{1+\sqrt{14x^2}}{1\sqrt{14x^2}}}.`$
Furthermore, $`I(x)=I_0(x,0,0)+I_0(x,x,0)`$, and
$$I_2(x)=(1x^2)^3+\left(1+\frac{1}{2}x^2+3x^4\right)\sqrt{14x^2}3x^2(12x^4)\mathrm{ln}\frac{1+\sqrt{14x^2}}{1\sqrt{14x^2}}.$$
Finally, the QCD radiative correction factor $`\eta _2`$ in eq.(9) is known in the leading logarithmic approximation and is expressed in terms of the well known coefficients $`C_+`$ and $`C_{}`$ in the renormalization of the non-leptonic weak interaction: $`\eta _2=(C_+^2(m_b)C_{}^2(m_b))/6`$ with
$$C_{}(\mu )=C_+^2(\mu )=\left[\frac{\alpha _s(\mu )}{\alpha _s(m_W)}\right]^{4/b},$$
(11)
and $`b`$ is the coefficient in the QCD beta function. The value of $`b`$ relevant to $`b`$ decays is $`b=23/3`$.
Numerically, for $`x0.3`$, the expressions for the decay rates can be written as
$`\mathrm{\Gamma }_{sl}(H_b)=\mathrm{\Gamma }_{sl}^{parton}\left(1{\displaystyle \frac{\mu _\pi ^2(H_b)\mu _g^2(H_b)}{2m_b^2}}2.6{\displaystyle \frac{\mu _g^2(H_b)}{m_b^2}}\right),`$
$`\mathrm{\Gamma }_{nl}(H_b)=\mathrm{\Gamma }_{nl}^{parton}\left(1{\displaystyle \frac{\mu _\pi ^2(H_b)\mu _g^2(H_b)}{2m_b^2}}1.0{\displaystyle \frac{\mu _g^2(H_b)}{m_b^2}}\right),`$ (12)
where $`\mathrm{\Gamma }^{parton}`$ is the perturbation theory value of the corresponding decay rate of $`b`$ quark.
The matrix elements $`\mu _\pi ^2`$ and $`\mu _g^2`$ are related to the spectroscopic formula for a heavy hadron mass $`M`$,
$$M(H_Q)=m_Q+\overline{\mathrm{\Lambda }}(H_Q)+\frac{\mu _\pi ^2(H_Q)\mu _g^2(H_Q)}{2m_Q}+\mathrm{}$$
(13)
Being combined with the spin counting for pseudoscalar and vector mesons, this formula allows to find the value of $`\mu _g^2`$ in pseudoscalar mesons from the mass splitting:
$$\mu _g^2(B)=\frac{3}{4}\left(M_B^{}^2M_B^2\right)0.36GeV^2.$$
(14)
The value of $`\mu _\pi ^2`$ for $`B`$ mesons is less certain. It is constrained by the inequality , $`\mu _\pi ^2(H_Q)\mu _g^2(H_Q)`$, and there are theoretical estimates from the QCD sum rules : $`\mu _\pi ^2(B)=0.54\pm 0.12GeV^2`$ and from an analysis of spectroscopy of heavy hadrons : $`\mu _\pi ^2(B)=0.3\pm 0.2GeV^2`$. In any event, the discussed corrections are rather small for $`b`$ hadrons, given that $`\mu _g^2/m_b^20.015`$. The largest, in relative terms, effect of these corrections in $`B`$ meson decays is on the semileptonic decay rate, where it amounts to 4 – 5 % suppression of the rate, which rate however is only a moderate fraction of the total width. In the dominant non-leptonic decay rate the effect is smaller, and, according to the formula (12) amounts to about 1.5 – 2 %.
The effect of the $`m_Q^2`$ corrections can be evaluated with a somewhat better certainty for the ratio of the decay rates of $`\mathrm{\Lambda }_b`$ and $`B`$ mesons. This is due to the fact that $`\mu _g^2(\mathrm{\Lambda }_b)=0`$, since there is no correlation of the spin of the heavy quark in $`\mathrm{\Lambda }_b`$ with the light component, having overall quantum numbers $`J^P=0^+`$. Then, applying the formula (12) to $`B`$ and $`\mathrm{\Lambda }_b`$, we find for the ratio of the (dominant) non-leptonic decay rates:
$$\frac{\mathrm{\Gamma }_{nl}(\mathrm{\Lambda }_b)}{\mathrm{\Gamma }_{nl}(B)}=1\frac{\mu _\pi ^2(\mathrm{\Lambda }_b)\mu _\pi ^2(B)}{2m_b^2}+0.5\frac{\mu _g^2(B)}{m_b^2}.$$
(15)
The difference of the kinetic terms, $`\mu _\pi ^2(\mathrm{\Lambda }_b)\mu _\pi ^2(B)`$, can be estimated from the mass formula:
$$\mu _\pi ^2(\mathrm{\Lambda }_b)\mu _\pi ^2(B)=\frac{2m_bm_c}{m_bm_c}\left[\overline{M}(B)\overline{M}(D)M(\mathrm{\Lambda }_b)+M(\mathrm{\Lambda }_c)\right]=0\pm 0.04GeV^2,$$
(16)
where $`\overline{M}`$ is the spin-averaged mass of the mesons, e.g. $`\overline{M}(B)=(M(B)+3M(B^{}))/4`$. The estimated difference of the kinetic terms is remarkably small. Thus the effect in the ratio of the decay rates essentially reduces to the chromomagnetic term, which is also rather small and accounts for less than 1% difference of the rates. For the ratio of the semileptonic decay rates the chromomagnetic term is approximately four times larger, but then the contribution of the semileptonic rates to the total width is rather small. Thus one concludes that the terms of order $`m_b^2`$ in the OPE expansion for the decay rates can account only for about 1% difference of the lifetimes of $`\mathrm{\Lambda }_b`$ and the $`B`$ mesons.
The significance of the $`m_Q^2`$ terms is substantially different for the decay rates of charmed hadrons, where these effects suppress the inclusive decays of the $`D`$ mesons by about 40% with respect to those of the charmed hyperons in a reasonable agreement with the observed pattern of the lifetimes.
It should be emphasized once again that the $`m_Q^2`$ effects do not depend on the flavors of the spectator quarks or antiquarks. Thus the explanation of the variety of the inclusive decay rates within the flavor SU(3) multiplets, observed for charmed hadrons and expected for the $`b`$ ones, has to be sought among the $`m_Q^3`$ terms.
## 5 $`L_{eff}^{(3)}`$. Coefficients and operators
Although the third term in the expansion (4) is formally suppressed by an extra power of $`m_Q^1`$, its effects are comparable to, or even larger than the effects of the second term. This is due to the fact that the diagrams determining the third term (see Fig. 1) contain a two-body phase space, while the first two terms involve a three-body phase space. This brings in a numerical enhancement factor, typically $`4\pi ^2`$. The enhanced numerical significance of the third term in OPE, generally, does not signal a poor convergence of the expansion in inverse heavy quark mass for decays of $`b`$, and even charmed, hadrons the numerical enhancement factor is a one time occurrence in the series, and there is no reason for similar ‘anomalous’ enhancement among the higher terms in the expansion.
Here we first present the expressions for the relevant parts of $`L_{eff}^{(3)}`$ for decays of $`b`$ and $`c`$ hadrons in the form of four-quark operators and then proceed to a discussion of hadronic matrix elements and the effects in specific inclusive decay rates. The consideration of the effects in decays of charmed hadrons is interesting in its own right, and leads to new predictions to be tested experimentally, and is also important for understanding the magnitude of the involved matrix elements using the existing data on charm decays.
We start with considering the term $`L_{eff}^{(3)}`$ in $`b`$ hadron non-leptonic decays, $`L_{eff,nl}^{(3,b)}`$, induced by the underlying processes $`bc\overline{u}d`$, $`bc\overline{c}s`$, $`bc\overline{u}s`$, and $`bc\overline{c}d`$. Unlike the case of three-body decay, the kinematical difference between the two-body states $`c\overline{c}`$ and $`c\overline{u}`$, involved in calculation of $`L_{eff,nl}^{(3,b)}`$ is of the order of $`m_c^2/m_b^20.1`$ and is rather small. At present level of accuracy in discussing this term in OPE, one can safely neglect the effect of finite charmed quark mass<sup>2</sup><sup>2</sup>2The full expression for a finite charmed quark mass can be found in . In this approximation the expression for $`L_{eff,nl}^{(3,b)}`$ reads as
$`L_{eff,nl}^{(3,b)}=|V_{bc}|^2{\displaystyle \frac{G_F^2m_b^2}{4\pi }}\{\stackrel{~}{C}_1(\overline{b}\mathrm{\Gamma }_\mu b)(\overline{u}\mathrm{\Gamma }_\mu u)+\stackrel{~}{C}_2(\overline{b}\mathrm{\Gamma }_\mu u)(\overline{u}\mathrm{\Gamma }_\mu b)+`$
$`\stackrel{~}{C}_5(\overline{b}\mathrm{\Gamma }_\mu b+{\displaystyle \frac{2}{3}}\overline{b}\gamma _\mu \gamma _5b)(\overline{q}\mathrm{\Gamma }_\mu q)+\stackrel{~}{C}_6(\overline{b}_i\mathrm{\Gamma }_\mu b_k+{\displaystyle \frac{2}{3}}\overline{b}_i\gamma _\mu \gamma _5b_k)(\overline{q}_k\mathrm{\Gamma }_\mu q_i)+`$ (17)
$`{\displaystyle \frac{1}{3}}\stackrel{~}{\kappa }^{1/2}(\stackrel{~}{\kappa }^{2/9}1)[2(\stackrel{~}{C}_+^2\stackrel{~}{C}_{}^2)(\overline{b}\mathrm{\Gamma }_\mu t^ab)j_\mu ^a`$
$`(5\stackrel{~}{C}_+^2+\stackrel{~}{C}_{}^26\stackrel{~}{C}_+\stackrel{~}{C}_{})(\overline{b}\mathrm{\Gamma }_\mu t^ab+{\displaystyle \frac{2}{3}}\overline{b}\gamma _\mu \gamma _5t^ab)j_\mu ^a]\},`$
where the notation $`(\overline{q}\mathrm{\Gamma }q)=(\overline{d}\mathrm{\Gamma }d)+(\overline{s}\mathrm{\Gamma }s)`$ is used, the indices $`i,k`$ are the color triplet ones, $`\mathrm{\Gamma }_\mu =\gamma _\mu (1\gamma _5)`$, and $`j_\mu ^a=\overline{u}\gamma _\mu t^au+\overline{d}\gamma _\mu t^ad+\overline{s}\gamma _\mu t^as`$ is the color current of the light quarks with $`t^a=\lambda ^a/2`$ being the generators of the color SU(3). The notation $`\stackrel{~}{C}_\pm `$, is used as shorthand for the short-distance renormalization coefficients $`C_\pm (\mu )`$ at $`\mu =m_b`$: $`\stackrel{~}{C}_\pm C_\pm (m_b)`$. The expression (5) is written in the leading logarithmic approximation for the QCD radiative effects in a low normalization point $`\mu `$ such that $`\mu m_b`$ (but still, at least formally, $`\mu \mathrm{\Lambda }_{QCD}`$). For such $`\mu `$ there arises so called ‘hybrid’ renormalization , depending on the factor $`\stackrel{~}{\kappa }=\alpha _s(\mu )/\alpha _s(m_b)`$. The coefficients $`\stackrel{~}{C}_A`$ with $`A=1,\mathrm{},6`$ in eq.(5) have the following explicit expressions in terms of $`\stackrel{~}{C}_\pm `$ and $`\stackrel{~}{\kappa }`$:
$`\stackrel{~}{C}_1=\stackrel{~}{C}_+^2+\stackrel{~}{C}_{}^2+{\displaystyle \frac{1}{3}}(1\kappa ^{1/2})(\stackrel{~}{C}_+^2\stackrel{~}{C}_{}^2),`$
$`\stackrel{~}{C}_2=\kappa ^{1/2}(\stackrel{~}{C}_+^2\stackrel{~}{C}_{}^2),`$
$`\stackrel{~}{C}_3={\displaystyle \frac{1}{4}}\left[(\stackrel{~}{C}_+\stackrel{~}{C}_{})^2+{\displaystyle \frac{1}{3}}(1\kappa ^{1/2})(5\stackrel{~}{C}_+^2+\stackrel{~}{C}_{}^2+6\stackrel{~}{C}_+\stackrel{~}{C}_{})\right],`$
$`\stackrel{~}{C}_4={\displaystyle \frac{1}{4}}\kappa ^{1/2}(5\stackrel{~}{C}_+^2+\stackrel{~}{C}_{}^2+6\stackrel{~}{C}_+\stackrel{~}{C}_{}),`$
$`\stackrel{~}{C}_5={\displaystyle \frac{1}{4}}\left[(\stackrel{~}{C}_++\stackrel{~}{C}_{})^2+{\displaystyle \frac{1}{3}}(1\kappa ^{1/2})(5\stackrel{~}{C}_+^2+\stackrel{~}{C}_{}^26\stackrel{~}{C}_+\stackrel{~}{C}_{})\right],`$
$`\stackrel{~}{C}_6={\displaystyle \frac{1}{4}}\kappa ^{1/2}(5\stackrel{~}{C}_+^2+\stackrel{~}{C}_{}^26\stackrel{~}{C}_+\stackrel{~}{C}_{}).`$ (18)
The expression for the CKM dominant semileptonic decays of $`b`$ hadrons, associated with the elementary process $`bc\mathrm{}\nu `$ does not look to be of an immediate interest. The reason is that this process is intrinsically symmetric under the flavor SU(3), and one expects no significant splitting of the semileptonic decay rates within SU(3) multiplets of the $`b`$ hadrons. The only possible effect of this term, arising through a penguin-like mechanism can be in a small overall shift of semileptonic decay rates between $`B`$ mesons and baryons. However, these effects are quite suppressed and are believed to be even smaller than the ones arising form the discussed $`m_b^2`$ terms.
For charm decays there is a larger, than for $`b`$ hadrons, variety of effects associated with $`L_{eff}^{(3)}`$, that can be studied experimentally, and we present here the relevant parts of the effective Lagrangian. For the CKM dominant non-leptonic decays of charm, originating from the quark process $`csu\overline{d}`$, the discussed term in OPE has the form
$`L_{eff,nl}^{(3,\mathrm{\Delta }C=\mathrm{\Delta }S)}=\mathrm{cos}^4\theta _c{\displaystyle \frac{G_F^2m_c^2}{4\pi }}\{C_1(\overline{c}\mathrm{\Gamma }_\mu c)(\overline{d}\mathrm{\Gamma }_\mu d)+C_2(\overline{c}\mathrm{\Gamma }_\mu d)(\overline{d}\mathrm{\Gamma }_\mu c)+`$
$`C_3(\overline{c}\mathrm{\Gamma }_\mu c+{\displaystyle \frac{2}{3}}\overline{c}\gamma _\mu \gamma _5c)(\overline{s}\mathrm{\Gamma }_\mu s)+C_4(\overline{c}_i\mathrm{\Gamma }_\mu c_k+{\displaystyle \frac{2}{3}}\overline{c}_i\gamma _\mu \gamma _5c_k)(\overline{s}_k\mathrm{\Gamma }_\mu s_i)+`$ (19)
$`C_5(\overline{c}\mathrm{\Gamma }_\mu c+{\displaystyle \frac{2}{3}}\overline{c}\gamma _\mu \gamma _5c)(\overline{u}\mathrm{\Gamma }_\mu u)+C_6(\overline{c}_i\mathrm{\Gamma }_\mu c_k+{\displaystyle \frac{2}{3}}\overline{c}_i\gamma _\mu \gamma _5c_k)(\overline{u}_k\mathrm{\Gamma }_\mu u_i)+`$
$`{\displaystyle \frac{1}{3}}\kappa ^{1/2}(\kappa ^{2/9}1)[2(C_+^2C_{}^2)(\overline{c}\mathrm{\Gamma }_\mu t^ac)j_\mu ^a(5C_+^2+C_{}^2)(\overline{c}\mathrm{\Gamma }_\mu t^ac+{\displaystyle \frac{2}{3}}\overline{c}\gamma _\mu \gamma _5t^ac)j_\mu ^a]\},`$
where, $`\theta _c`$ is the Cabibbo angle, and the coefficients without the tilde are given by the same expressions as above for the $`b`$ decays (i.e. those with tilde) with the replacement $`m_bm_c`$. The part of the notation in the superscript $`\mathrm{\Delta }C=\mathrm{\Delta }S`$ points to the selection rule for the dominant CKM unsuppressed non-leptonic decays. One can rather realistically envisage however a future study of inclusive rates for the once CKM suppressed decays of charmed hadrons<sup>3</sup><sup>3</sup>3Even if the inclusive rate of these decays is not to be separated experimentally, they contribute about 10% of the total decay rate, and it is worthwhile to include their contribution in the balance of the total width., satisfying the selection rule $`\mathrm{\Delta }S=0`$ and associated with the quark processes $`cdu\overline{s}`$ and $`cdu\overline{d}`$. The corresponding part of the effective Lagrangian for these processes reads as
$`L_{eff,nl}^{(3,\mathrm{\Delta }S=0)}=\mathrm{cos}^2\theta _c\mathrm{sin}^2\theta _c{\displaystyle \frac{G_F^2m_c^2}{4\pi }}\{C_1(\overline{c}\mathrm{\Gamma }_\mu c)(\overline{q}\mathrm{\Gamma }_\mu q)+C_2(\overline{c}_i\mathrm{\Gamma }_\mu c_k)(\overline{q}_k\mathrm{\Gamma }_\mu q_i)+`$
$`C_3(\overline{c}\mathrm{\Gamma }_\mu c+{\displaystyle \frac{2}{3}}\overline{c}\gamma _\mu \gamma _5c)(\overline{q}\mathrm{\Gamma }_\mu q)+C_4(\overline{c}_i\mathrm{\Gamma }_\mu c_k+{\displaystyle \frac{2}{3}}\overline{c}_i\gamma _\mu \gamma _5c_k)(\overline{q}_k\mathrm{\Gamma }_\mu q_i)+`$ (20)
$`2C_5(\overline{c}\mathrm{\Gamma }_\mu c+{\displaystyle \frac{2}{3}}\overline{c}\gamma _\mu \gamma _5c)(\overline{u}\mathrm{\Gamma }_\mu u)+2C_6(\overline{c}_i\mathrm{\Gamma }_\mu c_k+{\displaystyle \frac{2}{3}}\overline{c}_i\gamma _\mu \gamma _5c_k)(\overline{u}_k\mathrm{\Gamma }_\mu u_i)+`$
$`{\displaystyle \frac{2}{3}}\kappa ^{1/2}(\kappa ^{2/9}1)[2(C_+^2C_{}^2)(\overline{c}\mathrm{\Gamma }_\mu t^ac)j_\mu ^a(5C_+^2+C_{}^2)(\overline{c}\mathrm{\Gamma }_\mu t^ac+{\displaystyle \frac{2}{3}}\overline{c}\gamma _\mu \gamma _5t^ac)j_\mu ^a]\}`$
where again the notation $`(\overline{q}\mathrm{\Gamma }q)=(\overline{d}\mathrm{\Gamma }d)+(\overline{s}\mathrm{\Gamma }s)`$ is used.
The semileptonic decays of charm, the CKM dominant, associated with $`cs\mathrm{}\nu `$, and the CKM suppressed, originating from $`cs\mathrm{}\nu `$, contribute to the semileptonic decay rate, which certainly can be measured experimentally. The expression for the part of the effective Lagrangian, describing the $`m_Q^3`$ terms in these decays is
$`L_{eff,sl}^{(3)}=`$
$`{\displaystyle \frac{G_F^2m_c^2}{12\pi }}\{\mathrm{cos}^2\theta _c[L_1(\overline{c}\mathrm{\Gamma }_\mu c+{\displaystyle \frac{2}{3}}\overline{c}\gamma _\mu \gamma _5c)(\overline{s}\mathrm{\Gamma }_\mu s)+L_2(\overline{c}_i\mathrm{\Gamma }_\mu c_k+{\displaystyle \frac{2}{3}}\overline{c}_i\gamma _\mu \gamma _5c_k)(\overline{s}_k\mathrm{\Gamma }_\mu s_i)]+`$
$`\mathrm{sin}^2\theta _c\left[L_1(\overline{c}\mathrm{\Gamma }_\mu c+{\displaystyle \frac{2}{3}}\overline{c}\gamma _\mu \gamma _5c)(\overline{d}\mathrm{\Gamma }_\mu d)+L_2(\overline{c}_i\mathrm{\Gamma }_\mu c_k+{\displaystyle \frac{2}{3}}\overline{c}_i\gamma _\mu \gamma _5c_k)(\overline{d}_k\mathrm{\Gamma }_\mu d_i)\right]`$
$`2\kappa ^{1/2}(\kappa ^{2/9}1)(\overline{c}\mathrm{\Gamma }_\mu t^ac+{\displaystyle \frac{2}{3}}\overline{c}\gamma _\mu \gamma _5t^ac)j_\mu ^a\},`$ (21)
with the coefficients $`L_1`$ and $`L_2`$ found as
$$L_1=(\kappa ^{1/2}1),L_2=3\kappa ^{1/2}.$$
(22)
## 6 Effects of $`L_{eff}^{(3)}`$ in mesons
The expressions for the terms in $`L_{eff}^{(3)}`$ still leave us with the problem of evaluating the matrix elements of the four-quark operators over heavy hadrons in order to calculate the effects in the decay rates according to the formula (2). In doing so only few conclusions can be drawn in a reasonably model independent way, i.e. without resorting to evaluation of the matrix elements using specific ideas about the dynamics of quarks inside hadrons. The most straightforward prediction can in fact be found for $`b`$ hadrons. Namely, one can notice that the operator (5) is symmetric under the flavor U spin (an SU(2) subgroup of the flavor SU(3), which mixes $`s`$ and $`d`$ quarks). This is a direct consequence of neglecting the small kinematical effect of the charmed quark mass. However the usual (in)accuracy of the flavor SU(3) symmetry is likely to be a more limiting factor for the accuracy of this symmetry, than the corrections of order $`m_c^2/m_b^2`$. Modulo this reservation the immediate prediction from this symmetry is the degeneracy of inclusive decay rates within U spin doublets:
$$\mathrm{\Gamma }(B_d)=\mathrm{\Gamma }(B_s),\mathrm{\Gamma }(\mathrm{\Lambda }_b)=\mathrm{\Gamma }(\mathrm{\Xi }_b^0),$$
(23)
where $`\mathrm{\Gamma }(B_s)`$ stands for the average rate over the two eigenstates of the $`B_s\overline{B}_s`$ oscillations. The data on decay rates of the cascade hyperon $`\mathrm{\Xi }_b^0`$ are not yet available, while the currently measured lifetimes of $`B_d`$ and $`B_s`$ are within less than 2% from one another. Theoretically, the difference of the lifetimes, associated with possible violation of the SU(3) symmetry and with breaking of the U symmetry of the effective Lagrangian (5), is expected to not exceed about 1%.
For the non-vanishing matrix elements of four-quark operators over pseudoscalar mesons one traditionally starts with the factorization formula and parametrizes possible deviation from factorization in terms of ‘bag constants’. Within the normalization convention adopted here the relations used in this parametrization read as
$`P_{Q\overline{q}}|(\overline{Q}\mathrm{\Gamma }_\mu q)(\overline{q}\mathrm{\Gamma }_\mu Q)|P_{Q\overline{q}}={\displaystyle \frac{1}{2}}f_P^2M_PB,`$
$`P_{Q\overline{q}}|(\overline{Q}\mathrm{\Gamma }_\mu Q)(\overline{q}\mathrm{\Gamma }_\mu q)|P_{Q\overline{q}}={\displaystyle \frac{1}{6}}f_P^2M_P\stackrel{~}{B},`$ (24)
where $`P_{Q\overline{q}}`$ stands for pseudoscalar meson made of $`Q`$ and $`\overline{q}`$, $`f_P`$ is the annihilation constant for the meson, and $`B`$ and $`\stackrel{~}{B}`$ are bag constants. The parameters $`B`$ and $`\stackrel{~}{B}`$ generally depend on the normalization point $`\mu `$ for the operators, and this dependence is compensated by the $`\mu `$ dependence of the coefficients in $`L_{eff}^{(3)}`$, so that the results for the physical decay rate difference do not depend on $`\mu `$. If the normalization point $`\mu `$ is chosen at the heavy quark mass (i.e. $`\mu =m_b`$ for $`B`$ mesons, and $`\mu =m_c`$ for $`D`$ mesons) the predictions for the difference of total decay rates take a simple form in terms of the corresponding bag constants (generally different between $`B`$ and $`D`$):
$`\mathrm{\Gamma }(B^\pm )\mathrm{\Gamma }(B^0)`$ $`=`$ $`|V_{cb}|^2{\displaystyle \frac{G_F^2m_b^3f_B^2}{8\pi }}\left[(\stackrel{~}{C}_+^2\stackrel{~}{C}_{}^2)B(m_b)+{\displaystyle \frac{1}{3}}(\stackrel{~}{C}_+^2+\stackrel{~}{C}_{}^2)\stackrel{~}{B}(m_b)\right]`$ (25)
$``$ $`0.025\left({\displaystyle \frac{f_B}{200MeV}}\right)^2ps^1,`$
$`\mathrm{\Gamma }(D^\pm )\mathrm{\Gamma }(D^0)`$ $`=`$ $`\mathrm{cos}^4\theta _c{\displaystyle \frac{G_F^2m_c^3f_D^2}{8\pi }}\left[(C_+^2C_{}^2)B(m_c)+{\displaystyle \frac{1}{3}}(C_+^2+C_{}^2)\stackrel{~}{B}(m_c)\right]`$ (26)
$``$ $`0.8\left({\displaystyle \frac{f_D}{200MeV}}\right)^2ps^1,`$
where the numerical values are written in the approximation of exact factorization: $`B=1`$, and $`\stackrel{~}{B}=1`$. It is seen from the numerical estimates that, even given all the theoretical uncertainties, the presented approach is in reasonable agreement with the data on the lifetimes of $`D`$ and $`B`$ mesons. In particular, this approach describes, at least qualitatively, the strong suppression of the decay rate of $`D^\pm `$ mesons relative to $`D^0`$, the experimental observation of which has in fact triggered in early 80-s the theoretical study of preasymptotic in heavy quark mass effects in inclusive decay rates. For the $`B`$ mesons the estimate (25) is also in a reasonable agreement with the current data for the discussed difference ($`0.043\pm 0.017ps^1`$).
## 7 Effects of $`L_{eff}^{(3)}`$ in baryons
The weakly decaying heavy hyperons, containing either $`c`$ or $`b`$ quark are: $`\mathrm{\Lambda }_QQud,\mathrm{\Xi }_Q^{(u)}Qus,\mathrm{\Xi }_Q^{(d)}Qds`$, and $`\mathrm{\Omega }_QQss`$. The first three baryons form an SU(3) (anti)triplet. The light diquark in all three is in the state with quantum numbers $`J^P=0^+`$, so that there is no correlation of the spin of the heavy quark with the light component of the baryon. On the contrary, in $`\mathrm{\Omega }_Q`$ the two strange quarks form a $`J^P=1^+`$ state, and a correlation between the spins of heavy and light quarks is present. The absence of spin correlation for the heavy quark in the triplet of hyperons somewhat reduces the number of independent four-quark operators, having nonvanishing diagonal matrix elements over these baryons. Indeed, the operators entering $`L_{eff}^{(3)}`$ contain both vector and axial bilinear forms for the heavy quarks. However the axial part requires a correlation of the heavy quark spin with that of a light quark, and is thus vanishing for the hyperons in the triplet. Therefore only the structures with vector currents are relevant for these hyperons. These structures are of the type $`(\overline{c}\gamma _\mu c)(\overline{q}\gamma _\mu q)`$ and $`(\overline{c}_i\gamma _\mu c_k)(\overline{q}_k\gamma _\mu q_i)`$ with $`q`$ being $`d`$, $`s`$ or $`u`$. The flavor SU(3) symmetry then allows to express, for each of the two color combinations, the matrix elements of three different operators, corresponding to three flavors of $`q`$, over the baryons in the triplet in terms of only two combinations: flavor octet and flavor singlet. Thus all effects of $`L_{eff}^{(3)}`$ in the triplet of the baryons can be expressed in terms of four independent combinations of matrix elements. These can be chosen in the following way:
$`x`$ $`=`$ $`{\displaystyle \frac{1}{2}}(\overline{Q}\gamma _\mu Q)\left[(\overline{u}\gamma _\mu u)(\overline{s}\gamma _\mu s)\right]_{\mathrm{\Xi }_Q^{(d)}\mathrm{\Lambda }_Q}={\displaystyle \frac{1}{2}}(\overline{Q}\gamma _\mu Q)\left[(\overline{s}\gamma _\mu s)(\overline{d}\gamma _\mu d)\right]_{\mathrm{\Lambda }_Q\mathrm{\Xi }_Q^{(u)}},`$ (27)
$`y`$ $`=`$ $`{\displaystyle \frac{1}{2}}(\overline{Q}_i\gamma _\mu Q_k)\left[(\overline{u}_k\gamma _\mu u_i)(\overline{s}_k\gamma _\mu s_i)\right]_{\mathrm{\Xi }_Q^{(d)}\mathrm{\Lambda }_Q}={\displaystyle \frac{1}{2}}(\overline{Q}_i\gamma _\mu Q_k)\left[(\overline{s}_k\gamma _\mu s_i)(\overline{d}_k\gamma _\mu d_i)\right]_{\mathrm{\Lambda }_Q\mathrm{\Xi }_Q^{(u)}},`$
with the notation for the differences of the matrix elements: $`𝒪_{AB}=A|𝒪|AB|𝒪|B`$, for the flavor octet part and the matrix elements:
$`x_s={\displaystyle \frac{1}{3}}H_Q|(\overline{Q}\gamma _\mu Q)\left((\overline{u}\gamma _\mu u)+(\overline{d}\gamma _\mu d)+(\overline{s}\gamma _\mu s)\right)|H_Q`$
$`y_s={\displaystyle \frac{1}{3}}H_Q|(\overline{Q}_i\gamma _\mu Q_k)\left((\overline{u}_k\gamma _\mu u_i)+(\overline{d}_k\gamma _\mu d_i)+(\overline{s}_k\gamma _\mu s_i)\right)|H_Q`$ (28)
for the flavor singlet part, where $`H_Q`$ stands for any heavy hyperon in the (anti)triplet.
The initial, very approximate, theoretical estimates of the matrix elements were essentially based on a non-relativistic constituent quark model, where these matrix elements are proportional to the density of a light quark at the location of the heavy one, i.e. in terms of the wave function, proportional to $`|\psi (0)|^2`$. Using then the same picture for the matrix elements over pseudoscalar mesons, relating the quantity $`|\psi (0)|^2`$ to the annihilation constant $`f_P`$, and assuming that $`|\psi (0)|^2`$ is approximately the same in baryons as in mesons, one arrived at the estimate
$$y=x=x_s=y_s\frac{f_D^2M_D}{12}0.006GeV^2,$$
(29)
where the sign relation between $`x`$ and $`y`$ is inferred from the color antisymmetry of the constituent quark wave function for baryons. Since the constituent picture was believed to be valid at distances of the order of the hadron size, the estimate (29) was applied to the matrix elements in a low normalization point where $`\alpha _s(\mu )1`$. For the matrix elements of the operators, containing $`s`$ quarks over the $`\mathrm{\Omega }_Q`$ hyperon, this picture predicts an enhancement factor due to the spin correlation:
$$\mathrm{\Omega }_Q|(\overline{Q}\mathrm{\Gamma }_\mu Q)(\overline{s}\mathrm{\Gamma }_\mu s)|\mathrm{\Omega }_Q=\mathrm{\Omega }_Q|(\overline{Q}_i\mathrm{\Gamma }_\mu Q_k)(\overline{s}_k\mathrm{\Gamma }_\mu s_i)|\mathrm{\Omega }_Q=\frac{10}{3}y$$
(30)
Although these simple estimates allowed to correctly predict the hierarchy of lifetimes of charmed hadrons prior to establishing this hierarchy experimentally, they fail to quantitatively predict the differences of lifetimes of charmed baryons. We shall see that the available data indicate that the color antisymmetry relation is badly broken, and the absolute value of the matrix elements is larger, than the naive estimate (29), especially for the quantity $`x`$.
It should be emphasized that in the heavy quark limit the matrix elements (27) and (28) do not depend on the flavor of the heavy quark, provided that the same normalization point $`\mu `$ is used. Therefore, applying the OPE formulas to both charmed and $`b`$ baryons, one can extract the values for the matrix elements from available data on charmed hadrons, and then make predictions for $`b`$ baryons, as well as for other inclusive decay rates, e.g. semileptonic, for charmed hyperons.
The only data available so far, which would allow to extract the matrix elements, are on the lifetimes of charmed hyperons. Therefore, one has to take into account several essential types of inclusive decay, at least those that contribute to the total decay rate at the level of about 10%. Here we first concentrate on the differences of the decay rates within the SU(3) triplet of the hyperons, which will allow us to extract the non-singlet quantities $`x`$ and $`y`$, and then discuss the SU(3) singlet shifts of the rates.
The differences of the dominant Cabibbo unsuppressed non-leptonic decay rates are given by
$`\delta _1^{nl,\mathrm{\hspace{0.17em}0}}\mathrm{\Gamma }_{\mathrm{\Delta }S=\mathrm{\Delta }C}^{nl}(\mathrm{\Xi }_c^0)\mathrm{\Gamma }_{\mathrm{\Delta }S=\mathrm{\Delta }C}^{nl}(\mathrm{\Lambda }_c)=\mathrm{cos}^4\theta _c{\displaystyle \frac{G_F^2m_c^2}{4\pi }}\left[(C_5C_3)x+(C_6C_4)y\right],`$
$`\delta _2^{nl,\mathrm{\hspace{0.17em}0}}\mathrm{\Gamma }_{\mathrm{\Delta }S=\mathrm{\Delta }C}^{nl}(\mathrm{\Lambda }_c)\mathrm{\Gamma }_{\mathrm{\Delta }S=\mathrm{\Delta }C}^{nl}(\mathrm{\Xi }_c^+)=\mathrm{cos}^4\theta _c{\displaystyle \frac{G_F^2m_c^2}{4\pi }}\left[(C_3C_1)x+(C_4C_2)y\right].`$ (31)
The once Cabibbo suppressed decay rates of $`\mathrm{\Lambda }_c`$ and $`\mathrm{\Xi }_c^+`$ are equal, due to the $`\mathrm{\Delta }U=0`$ property of the corresponding effective Lagrangian $`L_{eff,nl}^{(3,1)}`$ (eq.(5)). Thus the only difference for these decays in the baryon triplet is
$$\delta ^{nl,1}\mathrm{\Gamma }_{\mathrm{\Delta }S=0}^{nl}(\mathrm{\Xi }_c^0)\mathrm{\Gamma }_{\mathrm{\Delta }S=0}^{nl}(\mathrm{\Lambda }_c)=\mathrm{cos}^2\theta _c\mathrm{sin}^2\theta _c\frac{G_F^2m_c^2}{4\pi }\left[(2C_5C_1C_3)x+(2C_6C_2C_4)y\right].$$
(32)
The dominant semileptonic decay rates are equal among the two $`\mathrm{\Xi }_c`$ baryons due to the isotopic spin property $`\mathrm{\Delta }I=0`$ of the corresponding interaction Lagrangian, thus there is only one non-trivial splitting for these decays:
$$\delta ^{sl,0}\mathrm{\Gamma }_{\mathrm{\Delta }S=1}^{sl}(\mathrm{\Xi }_c)\mathrm{\Gamma }_{\mathrm{\Delta }S=1}^{sl}(\mathrm{\Lambda }_c)=\mathrm{cos}^2\theta _c\frac{G_F^2m_c^2}{12\pi }\left[L_1x+L_2y\right].$$
(33)
Finally, the Cabibbo suppressed semileptonic decay rates are equal for $`\mathrm{\Lambda }_c`$ and $`\mathrm{\Xi }_c^0`$, due to the $`\mathrm{\Delta }V=0`$ property of the corresponding interaction. Thus the only difference for these is
$$\delta ^{sl,1}\mathrm{\Gamma }_{\mathrm{\Delta }S=0}^{sl}(\mathrm{\Lambda }_c)\mathrm{\Gamma }_{\mathrm{\Delta }S=0}^{sl}(\mathrm{\Xi }_c^+)=\mathrm{sin}^2\theta _c\frac{G_F^2m_c^2}{12\pi }\left[L_1x+L_2y\right].$$
(34)
Using the relations (31) - (34) on can find expressions for two differences of the measured total decay rates, $`\mathrm{\Delta }_1=\mathrm{\Gamma }(\mathrm{\Xi }_c^0)\mathrm{\Gamma }(\mathrm{\Lambda }_c)`$ and $`\mathrm{\Delta }_2=\mathrm{\Gamma }(\mathrm{\Lambda }_c)\mathrm{\Gamma }(\mathrm{\Xi }_c^+)`$, in terms of the quantities $`x`$ and $`y`$:
$`\mathrm{\Delta }_1=\delta _1^{nl,\mathrm{\hspace{0.17em}0}}+\delta ^{nl,1}+2\delta ^{sl,0}=`$
$`{\displaystyle \frac{G_F^2m_c^2}{4\pi }}\mathrm{cos}^2\theta \{x[\mathrm{cos}^2\theta (C_5C_3)+\mathrm{sin}^2\theta (2C_5C_1C_3){\displaystyle \frac{2}{3}}L_1]+`$
$`y[\mathrm{cos}^2\theta (C_6C_4)+\mathrm{sin}^2\theta (2C_6C_2C_4){\displaystyle \frac{2}{3}}L_2]\},`$ (35)
and
$`\mathrm{\Delta }_2=\delta _2^{nl,\mathrm{\hspace{0.17em}0}}2\delta ^{sl,0}+2\delta ^{sl,1}=`$
$`{\displaystyle \frac{G_F^2m_c^2}{4\pi }}\{x[\mathrm{cos}^4\theta (C_3C_1)+{\displaystyle \frac{2}{3}}(\mathrm{cos}^2\theta \mathrm{sin}^2\theta )L_1]+`$
$`y[cos^4\theta (C_4C_2)+{\displaystyle \frac{2}{3}}(\mathrm{cos}^2\theta \mathrm{sin}^2\theta )L_2]\}.`$ (36)
By comparing these relations with the data, one can extract the values of $`x`$ and $`y`$. Using the current data for the total decay rates: $`\mathrm{\Gamma }(\mathrm{\Lambda }_c)=4.85\pm 0.28ps^1`$, $`\mathrm{\Gamma }(\mathrm{\Xi }_c^0)=10.2\pm 2ps^1`$, and the updated value $`\mathrm{\Gamma }(\mathrm{\Xi }_c^+)=3.0\pm 0.45ps^1`$, we find for the $`\mu `$ independent matrix element $`x`$
$$x=(0.04\pm 0.01)GeV^3\left(\frac{1.4GeV}{m_c}\right)^2,$$
(37)
while the dependence of the thus extracted matrix element $`y`$ on the normalization point $`\mu `$ is shown in Fig. 2 <sup>4</sup><sup>4</sup>4It should be noted that the curves at large values of $`\kappa `$, $`\kappa >3`$, are shown only for illustrative purpose. The coefficients in the OPE, leading to the equations (35,36), are purely perturbative. Thus, formally, they correspond to $`\alpha _s(\mu )1`$, i.e. to $`\kappa 1/\alpha _s(m_c)(34)`$. .
Notably, the extracted values of $`x`$ and $`y`$ are in a drastic variance with the simplistic constituent model: the color antisymmetry relation, $`x=y`$, does not hold at any reasonable $`\mu `$, and the absolute value of $`x`$ is substantially enhanced<sup>5</sup><sup>5</sup>5A similar, although with a smaller enhancement, behavior of the matrix elements was observed in a recent preliminary lattice study .
Once the non-singlet matrix elements are determined, they can be used for predicting differences of other inclusive decay rates within the triplet of charmed hyperons as well as for the $`b`$ baryons. Due to correlation of errors in $`x`$ and $`y`$ it makes more sense to express the predictions directly in terms of the total decay rates of the charmed hyperons. The thus arising relations between the rates do not depend on the normalization parameter $`\mu `$. In this way one finds for the difference of the Cabibbo dominant semileptonic decay rates between either of the $`\mathrm{\Xi }_c`$ hyperons and $`\mathrm{\Lambda }_c`$:
$$\mathrm{\Gamma }_{sl}(\mathrm{\Xi }_c)\mathrm{\Gamma }_{sl}(\mathrm{\Lambda }_c)\delta ^{sl,0}=0.13\mathrm{\Delta }_10.065\mathrm{\Delta }_20.59\pm 0.32ps^1.$$
(38)
When compared with the data on the total semileptonic decay rate of $`\mathrm{\Lambda }_c`$, $`\mathrm{\Gamma }_{sl}(\mathrm{\Lambda }_c)=0.22\pm 0.08ps^1`$, this prediction implies that the semileptonic decay rate of the charmed cascade hyperons can be 2–3 times larger than that of $`\mathrm{\Lambda }_c`$.
The predictions found in a similar way for the inclusive Cabibbo suppressed decay rates are : for non-leptonic decays
$$\delta ^{nl,1}=0.082\mathrm{\Delta }_1+0.054\mathrm{\Delta }_20.55\pm 0.22ps^1$$
(39)
and for the semileptonic ones
$$\delta ^{sl,1}=\mathrm{tan}^2\theta _c\delta ^{sl,0}0.030\pm 0.016ps^1.$$
(40)
For the only difference of the inclusive rates in the triplet of $`b`$ baryons, $`\mathrm{\Gamma }(\mathrm{\Lambda }_b)\mathrm{\Gamma }(\mathrm{\Xi }_b^{})`$, one finds an expression in terms of $`x`$ and $`y`$, or alternatively, in terms of the differences $`\mathrm{\Delta }_1`$ and $`\mathrm{\Delta }_2`$ between the charmed hyperons,
$`\mathrm{\Gamma }(\mathrm{\Lambda }_b)\mathrm{\Gamma }(\mathrm{\Xi }_b^{})=\mathrm{cos}^2\theta _c|V_{bc}|^2{\displaystyle \frac{G_F^2m_b^2}{4\pi }}\left[(\stackrel{~}{C}_5\stackrel{~}{C}_1)x+(\stackrel{~}{C}_6\stackrel{~}{C}_2)y\right]`$
$`|V_{bc}|^2{\displaystyle \frac{m_b^2}{m_c^2}}(0.85\mathrm{\Delta }_1+0.91\mathrm{\Delta }_2)0.015\mathrm{\Delta }_1+0.016\mathrm{\Delta }_20.11\pm 0.03ps^1.`$ (41)
When compared with the data on the total decay rate of $`\mathrm{\Lambda }_b`$ this result predicts about 14% longer lifetime of $`\mathrm{\Xi }_b^{}`$ than that of $`\mathrm{\Lambda }_b`$.
The singlet matrix elements $`x_s`$ and $`y_s`$ (cf. eq.(28)) are related to the shift of the average decay rate of the hyperons in the triplet:
$$\overline{\mathrm{\Gamma }}_Q=\frac{1}{3}\left(\mathrm{\Gamma }(\mathrm{\Lambda }_Q)+\mathrm{\Gamma }(\mathrm{\Xi }_Q^1)+\mathrm{\Gamma }(\mathrm{\Xi }_Q^2)\right).$$
(42)
For the charmed baryons the shift of the dominant non-leptonic decay rate is given by
$$\delta _{nl}^{(3,0)}\overline{\mathrm{\Gamma }}_c=\mathrm{cos}^4\theta \frac{G_F^2m_c^2}{8\pi }(C_+^2+C_{}^2)\kappa ^{5/18}(x_s3y_s),$$
(43)
while for the $`b`$ baryons the corresponding expression reads as
$$\delta ^{(3)}\overline{\mathrm{\Gamma }}_b=|V_{bc}|^2\frac{G_F^2m_b^2}{8\pi }(\stackrel{~}{C}_+\stackrel{~}{C}_{})^2\stackrel{~}{\kappa }^{5/18}(x_s3y_s).$$
(44)
The combination $`x_s3y_s`$ of the SU(3) singlet matrix elements cancels in the ratio of the shifts for $`b`$ hyperons and the charmed ones:
$$\delta ^{(3)}\overline{\mathrm{\Gamma }}_b=\frac{|V_{bc}|^2}{\mathrm{cos}^4}\frac{m_b^2}{m_c^2}\frac{(\stackrel{~}{C}_+\stackrel{~}{C}_{})^2}{C_+^2+C_{}^2}\left[\frac{\alpha _s(m_c)}{\alpha _s(m_b)}\right]^{5/18}\delta _{nl}^{(3,0)}\overline{\mathrm{\Gamma }}_c0.0025\delta _{nl}^{(3,0)}\overline{\mathrm{\Gamma }}_c.$$
(45)
(One can observe, with satisfaction, that the dependence on the unphysical parameter $`\mu `$ cancels out, as it should.) This equation shows that relatively to the charmed baryons the shift of the decay rates in the $`b`$ baryon triplet is greatly suppressed by the ratio $`(\stackrel{~}{C}_+\stackrel{~}{C}_{})^2/(C_+^2+C_{}^2)`$, which parametrically is of the second order in $`\alpha _s`$, and numerically is only about 0.12.
An estimate of $`\delta ^{(3)}\overline{\mathrm{\Gamma }}_b`$ from eq.(45) in absolute terms depends on evaluating the average shift $`\delta _{nl}^{(3,0)}\overline{\mathrm{\Gamma }}_c`$ for charmed baryons. The latter shift can be conservatively bounded from above by the average total decay rate of those baryons: $`\delta _{nl}^{(3,0)}\overline{\mathrm{\Gamma }}_c<\overline{\mathrm{\Gamma }}_c=6.0\pm 0.7ps^1`$, which then yields, using eq.(45), an upper bound $`\delta ^{(3)}\overline{\mathrm{\Gamma }}_b<0.015\pm 0.002ps^1`$. More realistically, one should subtract from the total average width $`\overline{\mathrm{\Gamma }}_c`$ the contribution of the ‘parton’ term, which can be estimated from the decay rate of $`D_0`$ with account of the $`O(m_c^2)`$ effects, as amounting to about $`3ps^1`$. (One should also take into account the semileptonic contribution to the total decay rates, which however is quite small at this level of accuracy). Thus a realistic evaluation of $`\delta ^{(3)}\overline{\mathrm{\Gamma }}_b`$ does not exceed $`0.01ps^1`$, which constitutes only about 1% of the total decay rate of $`\mathrm{\Lambda }_b`$. Thus the shift of the total decay rate of $`\mathrm{\Lambda }_b`$ due to the effects of $`L_{eff}^{(3)}`$ is dominantly associated with the SU(3) non-singlet difference (41). The shift of the $`\mathrm{\Lambda }_b`$ decay rate with respect to the average width $`\overline{\mathrm{\Gamma }}_b`$ due to the non-singlet operators is one third of the splitting (41), i.e. about 5%. Adding to this the 1% shift of the average width and another 1% difference from the meson decays due to the suppression of the latter by the $`m_b^2`$ chromomagnetic effects, one concludes that at the present level of theoretical understanding it looks impossible to explain a more than 10% enhancement of the total decay rate of $`\mathrm{\Lambda }_b`$ relative to $`B_d`$, where an ample 3% margin is added for the uncertainties of higher order terms in OPE as well as for higher order QCD radiative effects in the discussed corrections. In other words, the expected pattern of the lifetimes of the $`b`$ hyperons in the triplet, relative to $`B_d`$, is
$$\tau (\mathrm{\Xi }_b^0)\tau (\mathrm{\Lambda }_b)<\tau (B_d)<\tau (\mathrm{\Xi }_b^{}),$$
(46)
with the “best” theoretical estimate of the differences to be about 7% for each step of the inequality.
For the double strange hyperons $`\mathrm{\Omega }_c`$ and $`\mathrm{\Omega }_b`$ there is presently no better approach to evaluating the four-quark matrix elements, than the use of simplistic relations, like (30) based on constituent quark model. Such relations imply that the effects of the strange quark, WS and PI, in the $`\mathrm{\Omega }_Q`$ baryons are significantly enhanced over the same effects in the cascade hyperons. In charmed baryons a presence of strange spectator quark enhances the decay through positive interference with the quark emerging from the $`cs`$ transition in the decay. For $`\mathrm{\Omega }_c`$ this implies a significant enhancement of the total decay rate , which is in perfect agreement with the data on the $`\mathrm{\Omega }_c`$ lifetime. Also a similar enhancement is expected for the semileptonic decay rate of $`\mathrm{\Omega }_c`$. In $`b`$ baryons, on the contrary, the interference effect for a spectator strange quark is negative. Thus the non-leptonic decay rate of $`\mathrm{\Omega }_b`$ is expected to be suppressed, leaving $`\mathrm{\Omega }_b`$ most probably the longest-living particle among the $`b`$ baryons.
## 8 Relation between spectator effects in baryons and the decays $`\mathrm{\Xi }_Q\mathrm{\Lambda }_Q\pi `$
Rather unexpectedly, the problem of four-quark matrix elements over heavy hyperons is related to decays of the type $`\mathrm{\Xi }_Q\mathrm{\Lambda }_Q\pi `$. The mass difference between the charmed cascade hyperons $`\mathrm{\Xi }_c`$ and $`\mathrm{\Lambda }_c`$ is about 180 MeV. The expected analogous mass splitting for the $`b`$ hyperons should be very close to this number, since in the heavy quark limit
$$M(\mathrm{\Xi }_b)M(\mathrm{\Lambda }_b)=M(\mathrm{\Xi }_c)M(\mathrm{\Lambda }_c)+O(m_c^2m_b^2).$$
(47)
Therefore in both cases are kinematically possible decays of the type $`\mathrm{\Xi }_Q\mathrm{\Lambda }_Q\pi `$, in which the heavy quark is not destroyed, and which are quite similar to decays of ordinary ‘light’ hyperons. Surprisingly, the rate of these decays for both $`\mathrm{\Xi }_c`$ and $`\mathrm{\Xi }_b`$ is not insignificantly small, but rather their branching fraction can reach a level of few per mill for $`\mathrm{\Xi }_c`$ and of one percent or more for $`\mathrm{\Xi }_b`$ .
The transitions $`\mathrm{\Xi }_Q\mathrm{\Lambda }_Q\pi `$ are induced by two underlying weak processes: the ‘spectator’ decay of a strange quark, $`su\overline{u}d`$, which does not involve the heavy quark, and the ‘non-spectator’ weak scattering (WS)
$$sccd$$
(48)
trough the weak interaction of the $`cd`$ and $`sc`$ currents. One can also readily see that the WS mechanism contributes only to the decays $`\mathrm{\Xi }_c\mathrm{\Lambda }_c\pi `$ and is not present in the decays of the $`b`$ cascade hyperons. An important starting point in considering these transitions is that in the heavy quark limit the spin of the heavy quark completely decouples from the spin of the light component of the baryon, and that the latter light component in both the initial an the final baryons forms a $`J^P=0^+`$ state with quantum numbers of a diquark. Since the momentum transfer in the considered decays is small in comparison with the mass of the heavy quark the spin of the amplitudes with spin flip of the heavy quark, and thus of the baryon, are suppressed by $`m_Q^1`$. In terms of the two possible partial waves in the decay $`\mathrm{\Xi }_Q\mathrm{\Lambda }_Q\pi `$, the $`S`$ and $`P`$, this implies that the $`P`$ wave is strongly suppressed and the decays are dominated by the $`S`$ wave.
According to the well known current algebra technique, the $`S`$ wave amplitudes of pion emission can be considered in the chiral limit at zero four-momentum of the pion, where they are described by the PCAC reduction formula (pole terms are absent in these processes):
$$\mathrm{\Lambda }_Q\pi _i(p=0)|H_W|\mathrm{\Xi }_Q=\frac{\sqrt{2}}{f_\pi }\mathrm{\Lambda }_Q|[Q_i^5,H_W]|\mathrm{\Xi }_Q,$$
(49)
where $`\pi _i`$ is the pion triplet in the Cartesian notation, and $`Q_i^5`$ is the corresponding isotopic triplet of axial charges. The constant $`f_\pi 130MeV`$, normalized by the charged pion decay, is used here, hence the coefficient $`\sqrt{2}`$ in eq.(49). The Hamiltonian $`H_W`$ in eq.(49) is the non-leptonic strangeness-changing hamiltonian:
$`H_W=`$ $`\sqrt{2}G_F\mathrm{cos}\theta _c\mathrm{sin}\theta _c\{(C_++C_{})[(\overline{u}_L\gamma _\mu s_L)(\overline{d}_L\gamma _\mu u_L)(\overline{c}_L\gamma _\mu s_L)(\overline{d}_L\gamma _\mu c_L)]+`$ (50)
$`(C_+C_{})[(\overline{d}_L\gamma _\mu s_L)(\overline{u}_L\gamma _\mu u_L)(\overline{d}_L\gamma _\mu s_L)(\overline{c}_L\gamma _\mu c_L)]\}.`$
In this formula the weak Hamiltonian is assumed to be normalized (in LLO) at $`\mu =m_c`$. The terms in the Hamiltonian (50) without the charmed quark fields describe the ‘spectator’ nonleptonic decay of the strange quark, while those with the $`c`$ quark correspond to the WS process (48).
It is straightforward to see from eq.(49) that in the PCAC limit the discussed decays should obey the $`\mathrm{\Delta }I=1/2`$ rule. Indeed, the commutator of the weak Hamiltonian with the axial charges transforms under the isotopic SU(2) in the same way as the Hamiltonian itself. In other words, the $`\mathrm{\Delta }I=1/2`$ part of $`H_W`$ after the commutation gives an $`\mathrm{\Delta }I=1/2`$ operator, while the $`\mathrm{\Delta }I=3/2`$ part after the commutation gives an $`\mathrm{\Delta }I=3/2`$ operator. The latter operator however cannot have a non vanishing matrix element between an isotopic singlet, $`\mathrm{\Lambda }_Q`$, and an isotopic doublet, $`\mathrm{\Xi }_Q`$. Thus the $`\mathrm{\Delta }I=3/2`$ part of $`H_W`$ gives no contribution to the $`S`$ wave amplitudes in the PCAC limit.
Once the isotopic properties of the decay amplitudes are fixed, one can concentrate on specific charge decay channels, e.g. $`\mathrm{\Xi }_b^{}\mathrm{\Lambda }_b\pi ^{}`$ and $`\mathrm{\Xi }_c^0\mathrm{\Lambda }_c\pi ^{}`$. An application of the PCAC relation (49) with the Hamiltonian from eq.(50) to these decays, gives the expressions for the amplitudes at $`p=0`$ in terms of baryonic matrix elements of four-quark operators:
$`\mathrm{\Lambda }_b\pi ^{}(p=0)|H_W|\mathrm{\Xi }_b^{}=`$
$`{\displaystyle \frac{\sqrt{2}}{f_\pi }}G_F\mathrm{cos}\theta _c\mathrm{sin}\theta _c\mathrm{\Lambda }_b|\left(C_++C_{}\right)\left[(\overline{u}_L\gamma _\mu s_L)(\overline{d}_L\gamma _\mu d_L)(\overline{u}_L\gamma _\mu s_L)(\overline{u}_L\gamma _\mu u_L)\right]+`$
$`\left(C_+C_{}\right)\left[(\overline{d}_L\gamma _\mu s_L)(\overline{u}_L\gamma _\mu d_L)(\overline{u}_L\gamma _\mu s_L)(\overline{u}_L\gamma _\mu u_L)\right]|\mathrm{\Xi }_b^{}=`$
$`{\displaystyle \frac{\sqrt{2}}{f_\pi }}G_F\mathrm{cos}\theta _c\mathrm{sin}\theta _c\mathrm{\Lambda }_b|C_{}\left[(\overline{u}_L\gamma _\mu s_L)(\overline{d}_L\gamma _\mu d_L)(\overline{d}_L\gamma _\mu s_L)(\overline{u}_L\gamma _\mu d_L)\right]`$
$`{\displaystyle \frac{C_+}{3}}\left[(\overline{u}_L\gamma _\mu s_L)(\overline{d}_L\gamma _\mu d_L)+(\overline{d}_L\gamma _\mu s_L)(\overline{u}_L\gamma _\mu d_L)+2(\overline{u}_L\gamma _\mu s_L)(\overline{u}_L\gamma _\mu u_L)\right]|\mathrm{\Xi }_b^{},`$ (51)
where in the last transition the operator structure with $`\mathrm{\Delta }I=3/2`$ giving a vanishing contribution is removed and only the structures with explicitly $`\mathrm{\Delta }I=1/2`$ are retained, and
$`\mathrm{\Lambda }_c\pi ^{}(p=0)|H_W|\mathrm{\Xi }_c^0=\mathrm{\Lambda }_b\pi ^{}(p=0)|H_W|\mathrm{\Xi }_b^{}+`$
$`{\displaystyle \frac{\sqrt{2}}{f_\pi }}G_F\mathrm{cos}\theta _c\mathrm{sin}\theta _c\mathrm{\Lambda }_c|\left(C_++C_{}\right)(\overline{c}_L\gamma _\mu s_L)(\overline{u}_L\gamma _\mu c_L)+`$
$`\left(C_+C_{}\right)(\overline{u}_L\gamma _\mu s_L)(\overline{c}_L\gamma _\mu c_L)|\mathrm{\Xi }_c^0.`$ (52)
In the latter formula the first term on the r.h.s. expresses the fact that in the heavy quark limit the ‘spectator’ amplitudes do not depend on the flavor or the mass of the heavy quark. The rest of the expression (52) describes the ‘non-spectator’ contribution to the amplitude of the charmed hyperon decay. Using the flavor SU(3) symmetry the latter contribution can be related to the non-singlet matrix elements (27) (normalized at $`\mu =m_c`$) as
$`\mathrm{\Delta }A\mathrm{\Lambda }_c\pi ^{}(p=0)|H_W|\mathrm{\Xi }_c^0\mathrm{\Lambda }_b\pi ^{}(p=0)|H_W|\mathrm{\Xi }_b^{}=`$
$`{\displaystyle \frac{G_F\mathrm{cos}\theta _c\mathrm{sin}\theta _c}{2\sqrt{2}f_\pi }}\left[\left(C_{}C_+\right)x\left(C_++C_{}\right)y\right].`$ (53)
Furthermore, with the help of the equations (35) and (36) relating the matrix elements $`x`$ and $`y`$ to the differences of the total decay widths within the triplet of charmed hyperons, one can eliminate $`x`$ and $`y`$ in favor of the measured width differences. The resulting expression has the form
$`\mathrm{\Delta }A{\displaystyle \frac{\sqrt{2}\pi \mathrm{cos}\theta _c\mathrm{sin}\theta _c}{G_Fm_c^2f_\pi }}\left[0.45\left(\mathrm{\Gamma }(\mathrm{\Xi }_c^0)\mathrm{\Gamma }(\mathrm{\Lambda }_c)\right)+0.04\left(\mathrm{\Gamma }(\mathrm{\Lambda }_c)\mathrm{\Gamma }(\mathrm{\Xi }_c^+)\right)\right]=`$
$`10^7\left[0.97\left(\mathrm{\Gamma }(\mathrm{\Xi }_c^0)\mathrm{\Gamma }(\mathrm{\Lambda }_c)\right)+0.09\left(\mathrm{\Gamma }(\mathrm{\Lambda }_c)\mathrm{\Gamma }(\mathrm{\Xi }_c^+)\right)\right]\left({\displaystyle \frac{1.4GeV}{m_c}}\right)^2ps,`$ (54)
where, clearly, in the latter form the widths are assumed to be expressed in $`ps^1`$, and $`m_c=1.4GeV`$ is used as a ‘reference’ value for the charmed quark mass. It is seen from eq.(54) that the evaluation of the difference of the amplitudes within the discussed approach is mostly sensitive to the difference of the decay rates of $`\mathrm{\Xi }_c^0`$ and $`\mathrm{\Lambda }_c`$, with only very little sensitivity to the total decay width of $`\mathrm{\Xi }_c^+`$. Using the current data the difference $`\mathrm{\Delta }A`$ is estimated as
$$\mathrm{\Delta }A=(5.4\pm 2)\times 10^7,$$
(55)
with the uncertainty being dominated by the experimental error in the lifetime of $`\mathrm{\Xi }_c^0`$. An amplitude $`A`$ of the magnitude, given by the central value in eq.(55) would produce a decay rate $`\mathrm{\Gamma }(\mathrm{\Xi }_Q\mathrm{\Lambda }_Q\pi )=|A|^2p_\pi /(2\pi )0.9\times 10^{10}s^1`$, which result can also be written in a form of triangle inequality
$$\sqrt{\mathrm{\Gamma }(\mathrm{\Xi }_b^{}\mathrm{\Lambda }_b\pi ^{})}+\sqrt{\mathrm{\Gamma }(\mathrm{\Xi }_c^0\mathrm{\Lambda }_c\pi ^{})}\sqrt{0.9\times 10^{10}s^1}.$$
(56)
Although at present it is not possible to evaluate in a reasonably model independent way the matrix element in eq.(51) for the ‘spectator’ decay amplitude, the inequality (56) shows that at least some of the discussed pion transitions should go at the level of $`0.01ps^1`$, similar to the rates of analogous decays of ‘light’ hyperons.
## 9 Summary
We summarize here specific predictions for the inclusive decay rates, which can be argued with a certain degree of theoretical reliability, and which can be possibly experimentally tested in the nearest future.
$`B`$ mesons:
$$\tau (B_d)/\tau (B_s)=1\pm 0.01.$$
Charmed hyperons:
$$\mathrm{\Gamma }_{sl}(\mathrm{\Xi }_c)=(23)\mathrm{\Gamma }_{sl}(\mathrm{\Lambda }_c)\mathrm{\Gamma }_{sl}(\mathrm{\Omega }_c)>\mathrm{\Gamma }_{sl}(\mathrm{\Xi }_c),$$
$$\mathrm{\Gamma }_{\mathrm{\Delta }S=1}^{nl}(\mathrm{\Xi }_c^+)\mathrm{\Gamma }_{\mathrm{\Delta }S=1}^{nl}(\mathrm{\Lambda }_c),$$
$$\mathrm{\Gamma }_{\mathrm{\Delta }S=1}^{nl}(\mathrm{\Xi }_c^0)\mathrm{\Gamma }_{\mathrm{\Delta }S=1}^{nl}(\mathrm{\Lambda }_c)0.55\pm 0.22ps^1.$$
$`b`$ hyperons:
$$\tau (\mathrm{\Xi }_b^0)\tau (\mathrm{\Lambda }_b)<\tau (B_d)<\tau (\mathrm{\Xi }_b^{})<\tau (\mathrm{\Omega }_b),$$
$$\mathrm{\Gamma }(\mathrm{\Lambda }_b)\mathrm{\Gamma }(\mathrm{\Xi }_b^{})0.11\pm 0.03ps^1,$$
$$0.9<\frac{\tau (\mathrm{\Lambda }_b)}{\tau (B_d)}<1.$$
Strangeness decays $`\mathrm{\Xi }_Q\mathrm{\Lambda }_Q\pi `$:
The $`\mathrm{\Delta }I=1/2`$ rule should hold in these decays, so that $`\mathrm{\Gamma }(\mathrm{\Xi }_Q^{(d)}\mathrm{\Lambda }_Q\pi ^{})=2\mathrm{\Gamma }(\mathrm{\Xi }_Q^{(u)}\mathrm{\Lambda }_Q\pi ^0)`$. The rates are constrained by the triangle inequality (56).
Acknowledgement
This work is supported in part by DOE under the grant number DE-FG02-94ER40823.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.